text
stringlengths
1
2.56M
id
stringlengths
40
40
metadata
dict
\section{Introduction} Let $G$ be a simple undirected graph with the \textit{vertex set} $V(G)$ and the \textit{edge set} $E(G)$. A vertex with degree one is called a \textit{pendant vertex}. The distance between the vertices $u$ and $v$ in graph $G$ is denoted by $d_G(u,v)$. A cycle $C$ is called \textit{chordless} if $C$ has no \textit{cycle chord} (that is an edge not in the edge set of $C$ whose endpoints lie on the vertices of $C$). The \textit{Induced subgraph} on vertex set $S$ is denoted by $\langle S\rangle$. A path that starts in $v$ and ends in $u$ is denoted by $\stackrel\frown{v u}$. A \textit{traceable} graph is a graph that possesses a Hamiltonian path. In a graph $G$, we say that a cycle $C$ is \textit{formed by the path} $Q$ if $ | E(C) \setminus E(Q) | = 1 $. So every vertex of $C$ belongs to $V(Q)$. In 2011 the following conjecture was proposed: \begin{conjecture}(Hoffmann-Ostenhof \cite{hoffman}) Let $G$ be a connected cubic graph. Then $G$ has a decomposition into a spanning tree, a matching and a family of cycles. \end{conjecture} Conjecture \theconjecture$\,$ also appears in Problem 516 \cite{cameron}. There are a few partial results known for Conjecture \theconjecture. Kostochka \cite{kostocha} noticed that the Petersen graph, the prisms over cycles, and many other graphs have a decomposition desired in Conjecture \theconjecture. Ozeki and Ye \cite{ozeki} proved that the conjecture holds for 3-connected cubic plane graphs. Furthermore, it was proved by Bachstein \cite{bachstein} that Conjecture \theconjecture$\,$ is true for every 3-connected cubic graph embedded in torus or Klein-bottle. Akbari, Jensen and Siggers \cite[Theorem 9]{akbari} showed that Conjecture \theconjecture$\,$ is true for Hamiltonian cubic graphs. In this paper, we show that Conjecture \theconjecture$\,$ holds for traceable cubic graphs. \section{Results} Before proving the main result, we need the following lemma. \begin{lemma} \label{lemma:1} Let $G$ be a cubic graph. Suppose that $V(G)$ can be partitioned into a tree $T$ and finitely many cycles such that there is no edge between any pair of cycles (not necessarily distinct cycles), and every pendant vertex of $T$ is adjacent to at least one vertex of a cycle. Then, Conjecture \theconjecture$\,$ holds for $G$. \end{lemma} \begin{proof} By assumption, every vertex of each cycle in the partition is adjacent to exactly one vertex of $T$. Call the set of all edges with one endpoint in a cycle and another endpoint in $T$ by $Q$. Clearly, the induced subgraph on $E(T) \cup Q$ is a spanning tree of $G$. We call it $T'$. Note that every edge between a pendant vertex of $T$ and the union of cycles in the partition is also contained in $T'$. Thus, every pendant vertex of $T'$ is contained in a cycle of the partition. Now, consider the graph $H = G \setminus E(T')$. For every $v \in V(T)$, $d_H(v) \leq 1$. So Conjecture \theconjecture$\,$ holds for $G$. \vspace{1em} \end{proof} \noindent\textbf{Remark 1.} \label{remark:1} Let $C$ be a cycle formed by the path $Q$. Then clearly there exists a chordless cycle formed by $Q$. Now, we are in a position to prove the main result. \begin{theorem} Conjecture \theconjecture$\,$ holds for traceable cubic graphs. \end{theorem} \begin{proof} Let $G$ be a traceable cubic graph and $P : v_1, \dots, v_n$ be a Hamiltonian path in $G$. By \cite[Theorem 9]{akbari}, Conjecture A holds for $v_1 v_n \in E(G)$. Thus we can assume that $v_1 v_n \notin E(G)$. Let $v_1 v_j, v_1 v_{j'}, v_i v_n, v_{i'} v_n \in E(G)\setminus E(P)$ and $j' < j < n$, $1 < i < i'$. Two cases can occur: \begin{enumerate}[leftmargin=0pt,label=] \item \textbf{Case 1.} Assume that $i < j$. Consider the following graph in Figure \ref{fig:overlapping} in which the thick edges denote the path $P$. Call the three paths between $v_j$ and $v_i$, from the left to the right, by $P_1$, $P_2$ and $P_3$, respectively (note that $P_1$ contains the edge $e'$ and $P_3$ contains the edge $e$). \begin{figure}[H] \begin{center} \includegraphics[width=40mm]{engImages/overlapping.pdf} \caption{Paths $P_1$, $P_2$ and $P_3$} \label{fig:overlapping} \end{center} \end{figure} If $P_2$ has order $2$, then $G$ is Hamiltonian and so by \cite[Theorem 9]{akbari} Conjecture \theconjecture$\,$ holds. Thus we can assume that $P_1$, $P_2$ and $P_3$ have order at least $3$. Now, consider the following subcases:\\ \begin{enumerate}[leftmargin=0pt,label=] \label{case:1} \item \textbf{Subcase 1.} There is no edge between $V(P_r)$ and $V(P_s)$ for $1 \leq r < s \leq 3$. Since every vertex of $P_i$ has degree 3 for every $i$, by \hyperref[remark:1]{Remark 1}$\,$ there are two chordless cycles $C_1$ and $C_2$ formed by $P_1$ and $P_2$, respectively. Define a tree $T$ with the edge set $$ E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big) \rangle\Big) \bigcap \big(\bigcup_{i=1}^3 E(P_i)\big).$$ Now, apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T, C_1, C_2\}$.\\ \item \textbf{Subcase 2.} \label{case:edge} There exists at least one edge between some $P_r$ and $P_s$, $r<s$. With no loss of generality, assume that $r=1$ and $s=2$. Suppose that $ab \in E(G)$, where $a \in V(P_1)$, $b \in V(P_2)$ and $d_{P_1}(v_j, a) + d_{P_2}(v_j, b)$ is minimum. \begin{figure}[H] \begin{center} \includegraphics[width=40mm]{engImages/ab.pdf} \caption{The edge $ab$ between $P_1$ and $P_2$} \label{fig:ab} \end{center} \end{figure} Three cases occur: \\ (a) There is no chordless cycle formed by either of the paths $\stackrel\frown{v_j a}$ or $\stackrel\frown{v_j b}$. Let $C$ be the chordless cycle $\stackrel\frown{v_j a}\stackrel\frown{ b v_j}$. Define $T$ with the edge set $$ E\Big(\langle V(G) \setminus V(C)\rangle\Big) \bigcap \big(\bigcup_{i=1}^3 E(P_i)\big).$$ Now, apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T,C\}$. \\ (b) There are two chordless cycles, say $C_1$ and $C_2$, respectively formed by the paths $\stackrel\frown{v_j a}$ and $\stackrel\frown{v_j b}$. Now, consider the partition $C_1$, $C_2$ and the tree induced on the following edges, $$E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big) \rangle\Big) \; \bigcap \; E\Big(\bigcup_{i=1}^3 P_i\Big),$$ and apply \hyperref[lemma:1]{Lemma 1}.\\ (c) With no loss of generality, there exists a chordless cycle formed by the path $\stackrel\frown{v_j a}$ and there is no chordless cycle formed by the path $\stackrel\frown{v_j b}$. First, suppose that for every chordless cycle $C_t$ on $\stackrel\frown{v_j a}$, at least one of the vertices of $C_t$ is adjacent to a vertex in $V(G) \setminus V(P_1)$. We call one of the edges with one end in $C_t$ and other endpoint in $V(G) \setminus V(P_1)$ by $e_t$. Let $v_j=w_0, w_1, \dots, w_l=a$ be all vertices of the path $\stackrel\frown{v_j a}$ in $P_1$. Choose the shortest path $w_0 w_{i_1} w_{i_2} \dots w_l$ such that $0 < i_1 < i_2 < \dots < l$. Define a tree $T$ whose edge set is the thin edges in Figure \ref{fig:deltaCycle}.\\ Call the cycle $w_0 w_{i_1} \dots w_l \stackrel\frown{b w_0}$ by $C'$. Now, by removing $C'$, $q$ vertex disjoint paths $Q_1, \dots, Q_q$ which are contained in $\stackrel\frown{v_j a}$ remain. Note that there exists a path of order $2$ in $C'$ which by adding this path to $Q_i$ we find a cycle $C_{t_i}$, for some $i$. Hence there exists an edge $e_{t_i}$ connecting $Q_i$ to $V(G) \setminus V(P_1)$. Now, we define a tree $T$ whose the edge set is, $$\quad\quad\quad \bigg( E\Big(\langle V(G) \setminus V(C') \rangle \Big)\; \bigcap \; \Big(\bigcup_{i=1}^3 E(P_i)\Big) \bigg) \bigcup \Big(\big\{e_{t_i} \mid 1 \leq i \leq q \big\} \Big).$$ Apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T,C'\}$.\\ \begin{figure}[H] \begin{center} \includegraphics[width=40mm]{engImages/deltaCycle.pdf} \caption{The cycle $C'$ and the tree $T$} \label{fig:deltaCycle} \end{center} \end{figure} Next, assume that there exists a cycle $C_1$ formed by $\stackrel\frown{v_j a}$ such that none of the vertices of $C_1$ is adjacent to $V(G) \setminus V(P_1)$. Choose the smallest cycle with this property. Obviously, this cycle is chordless. Now, three cases can be considered:\\ \begin{enumerate}[leftmargin=5pt,label=(\roman*)] \item There exists a cycle $C_2$ formed by $P_2$ or $P_3$. Define the partition $C_1$, $C_2$ and a tree with the following edge set, $$E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big)\rangle \Big) \bigcap \Big( \bigcup_{i=1}^3 E(P_i) \Big),$$ and apply \hyperref[lemma:1]{Lemma 1}.\\ \item There is no chordless cycle formed by $P_2$ and by $P_3$, and there is at least one edge between $V(P_2)$ and $V(P_3)$. Let $ab \in E(G)$, $a \in V(P_2)$ and $b \in V(P_3)$ and moreover $d_{P_2}(v_j, a) + d_{P_3}(v_j,b)$ is minimum. Notice that the cycle $\stackrel\frown{v_j a} \stackrel\frown{b v_j}$ is chordless. Let us call this cycle by $C_2$. Now, define the partition $C_2$ and a tree with the following edge set, $$E\Big(\langle V(G) \setminus V(C_2)\rangle \Big) \bigcap \Big( \bigcup_{i=1}^3 E(P_i) \Big),$$ and apply \hyperref[lemma:1]{Lemma 1}.\\ \item There is no chordless cycle formed by $P_2$ and by $P_3$, and there is no edge between $V(P_2)$ and $V(P_3)$. Let $C_2$ be the cycle consisting of two paths $P_2$ and $P_3$. Define the partition $C_2$ and a tree with the following edge set, $$E\Big(\langle V(G) \setminus V(C_2)\rangle \Big) \bigcap \Big( \bigcup_{i=1}^3 E(P_i) \Big),$$ and apply \hyperref[lemma:1]{Lemma 1}. \end{enumerate} \end{enumerate} \vspace{5mm} \item \textbf{Case 2.} \label{case:2} Assume that $j < i$ for all Hamiltonian paths. Among all Hamiltonian paths consider the path such that $i'-j'$ is maximum. Now, three cases can be considered:\\ \begin{enumerate}[leftmargin=0pt,label=] \item \textbf{Subcase 1.} There is no $s < j'$ and $t > i'$ such that $v_s v_t \in E(G)$. By \hyperref[remark:1]{Remark 1} $\,$ there are two chordless cycles $C_1$ and $C_2$, respectively formed by the paths $v_1 v_{j'}$ and $v_{i'} v_n$. By assumption there is no edge $xy$, where $x \in V(C_1)$ and $y \in V(C_2)$. Define a tree $T$ with the edge set: $$ E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big) \rangle \Big) \bigcap \Big( E(P) \cup \{v_{i'}v_n, v_{j'}v_1\} \Big).$$ Now, apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T, C_1, C_2\}$.\\ \item \textbf{Subcase 2.} \label{subcase:22} There are at least four indices $s, s' < j$ and $t, t' > i$ such that $v_s v_t, v_{s'} v_{t'} \in E(G)$. Choose four indices $g, h < j$ and $e, f > i$ such that $v_h v_e, v_g v_f \in E(G)$ and $|g-h| + |e-f|$ is minimum. \begin{figure}[H] \begin{center} \includegraphics[width=90mm]{engImages/case2-subcase2.pdf} \caption{Two edges $v_h v_e$ and $v_g v_f$} \label{fig:non-overlapping} \end{center} \end{figure} Three cases can be considered:\\ \begin{enumerate}[leftmargin=0pt,label=(\alph*)] \item There is no chordless cycle formed by $\stackrel\frown{v_g v_h}$ and by $\stackrel\frown{v_e v_f}$. Consider the cycle $\stackrel\frown{v_g v_h} \stackrel\frown{v_e v_f}v_g$ and call it $C$. Now, define a tree $T$ with the edge set, $$\,\,\,E\Big(\langle V(G) \setminus V(C)\rangle \Big) \bigcap \Big( E(P) \cup \{v_1v_{j}, v_{i}v_n\} \Big),$$ apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T, C\}$.\\ \item With no loss of generality, there exists a chordless cycle formed by $\stackrel\frown{v_e v_f}$ and there is no chordless cycle formed by the path $\stackrel\frown{v_g v_h}$. First suppose that there is a chordless cycle $C_1$ formed by $\stackrel\frown{v_e v_f}$ such that there is no edge between $V(C_1)$ and $\{v_1, \dots, v_j\}$. By \hyperref[remark:1]{Remark 1} $,$ there exists a chordless cycle $C_2$ formed by $\stackrel\frown{v_1 v_j}$. By assumption there is no edge between $V(C_1)$ and $V(C_2)$. Now, define a tree $T$ with the edge set, $$\quad\quad\quad\quad E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big)\rangle \Big) \bigcap \Big( E(P) \cup \{v_1v_{j}, v_{i}v_n\} \Big),$$ and apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T, C_1, C_2\}$. $\;$ Next assume that for every cycle $C_r$ formed by $\stackrel\frown{v_e v_f}$, there are two vertices $x_r \in V(C_r)$ and $y_r \in \{v_1, \dots, v_j\}$ such that $x_r y_r \in E(G)$. Let $v_e=w_0, w_1, \dots, w_l=v_f$ be all vertices of the path $\stackrel\frown{v_e v_f}$ in $P$. Choose the shortest path $w_0 w_{i_1} w_{i_2} \dots w_l$ such that $0 < i_1 < i_2 < \dots < l$. Consider the cycle $w_0 w_{i_1} \dots w_l \stackrel\frown{v_g v_h}$ and call it $C$. Now, by removing $C$, $q$ vertex disjoint paths $Q_1, \dots, Q_q$ which are contained in $\stackrel\frown{v_e v_f}$ remain. Note that there exists a path of order $2$ in $C$ which by adding this path to $Q_i$ we find a cycle $C_{r_i}$, for some $i$. Hence there exists an edge $x_{r_i} y_{r_i}$ connecting $Q_i$ to $V(G) \setminus V(\stackrel\frown{v_e v_f})$. We define a tree $T$ whose edge set is the edges, $$\quad\quad\quad\quad\quad\quad E\Big(\langle V(G) \setminus V(C)\rangle \Big) \bigcap \Big( E(P) \cup \{v_1v_{j}, v_{i}v_n\} \cup \big\{x_{r_i} y_{r_i} \mid 1 \leq i \leq q\big\} \Big),$$ then apply \hyperref[lemma:1]{Lemma 1} $\,$ on the partition $\{T, C\}$.\\ \begin{figure}[H] \begin{center} \includegraphics[width=90mm]{engImages/deltaNonOverlapping.pdf} \caption{The tree $T$ and the shortest path $w_0 w_{i_1}\dots w_l$} \label{fig:delta-non-overlapping} \end{center} \end{figure} \item There are at least two chordless cycles, say $C_1$ and $C_2$ formed by the paths $\stackrel\frown{v_g v_h}$ and $\stackrel\frown{v_e v_f}$, respectively. Since $|g-h| + |e-f|$ is minimum, there is no edge $xy \in E(G)$ with $x \in V(C_1)$ and $y \in V(C_2)$. Now, define a tree $T$ with the edge set, $$\quad\quad\quad\quad E\Big( \langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big) \rangle \Big) \bigcap \Big( E(P) \cup \{v_1 v_{j}, v_{i}v_n\} \Big),$$ and apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition $\{T, C_1, C_2\}$.\\ \end{enumerate} \item \textbf{Subcase 3.} There exist exactly two indices $s,t$, $s < j' < i' < t$ such that $v_s v_t \in E(G)$ and there are no two other indices $s', t'$ such that $s' < j < i < t'$ and $v_{s'} v_{t'} \in E(G)$. We can assume that there is no cycle formed by $\stackrel\frown{v_{s+1} v_j}$ or $\stackrel\frown{v_i v_{t-1}}$, to see this by symmetry consider a cycle $C$ formed by $\stackrel\frown{v_{s+1} v_j}$. By \hyperref[remark:1]{Remark 1} $\,$ there exist chordless cycles $C_1$ formed by $\stackrel\frown{v_{s+1} v_j}$ and $C_2$ formed by $\stackrel\frown{v_{i} v_n}$. By assumption $v_s v_t$ is the only edge such that $s < j$ and $t > i \;$. Therefore, there is no edge between $V(C_1)$ and $V(C_2)$. Now, let $T$ be a tree defined by the edge set, $$ E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big)\rangle \Big) \bigcap \Big( E(P) \cup \{v_1v_{j}, v_{i}v_n\} \Big),$$ and apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition \{$T$, $C_1$, $C_2$\}.\\ $\quad$Furthermore, we can also assume that either $s \neq j'-1$ or $t \neq i'+1$, otherwise we have the Hamiltonian cycle $\stackrel\frown{v_1 v_s} \stackrel\frown{v_t v_n} \stackrel\frown{v_{i'} v_{j'}} v_1$ and by \cite[Theorem 9]{akbari} Conjecture \theconjecture$\,$ holds. $\quad$By symmetry, suppose that $s \neq j'-1$. Let $v_k$ be the vertex adjacent to $v_{j'-1}$, and $k \notin \{j'-2, j'\}$. It can be shown that $k > j'-1$, since otherwise by considering the Hamiltonian path $P': \; \stackrel\frown{ v_{k+1} v_{j'-1}}\stackrel\frown{v_k v_1} \stackrel\frown{v_{j'} v_n}$, the new $i'-j'$ is greater than the old one and this contradicts our assumption about $P$ in the \hyperref[case:2]{Case 2}. $\quad$We know that $j' < k < i$. Moreover, the fact that $\stackrel\frown{v_{s+1} v_j}$ does not form a cycle contradicts the case that $j' < k \le j$. So $j < k < i$. Consider two cycles $C_1$ and $C_2$, respectively with the vertices $v_1 \stackrel\frown{v_{j'} v_{j}} v_1$ and $v_n \stackrel\frown{v_{i'} v_{i}} v_n$. The cycles $C_1$ and $C_2$ are chordless, otherwise there exist cycles formed by the paths $\stackrel\frown{v_{s+1} v_j}$ or $\stackrel\frown{v_i v_{t-1}}$. Now, define a tree $T$ with the edge set $$ E\Big(\langle V(G) \setminus \big(V(C_1) \cup V(C_2)\big)\rangle \Big) \bigcap \Big( E(P) \cup \{v_s v_t, v_k v_{j'-1}\} \Big),$$ and apply \hyperref[lemma:1]{Lemma 1} $\,$for the partition \{$T$, $C_1$, $C_2$\}. \end{enumerate} \end{enumerate} \end{proof} \noindent\textbf{Remark 2.} \label{remark:2} Indeed, in the proof of the previous theorem we showed a stronger result, that is, for every traceable cubic graph there is a decomposition with at most two cycles.
b7c40b41b7eedaa408f87d154284a1aba126589c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Principle of nano strain-amplifier} \begin{figure*}[t!] \centering \includegraphics[width=5.4in]{Fig1} \vspace{-0.5em} \caption{Schematic sketches of nanowire strain sensors. (a)(b) Conventional non-released and released NW structure; (c)(d) The proposed nano strain-amplifier and its simplified physical model.} \label{fig:fig1} \vspace{-1em} \end{figure*} Figure \ref{fig:fig1}(a) and 1(b) show the concept of the conventional structures of piezoresistive sensors. The piezoresistive elements are either released from, or kept on, the substrate. The sensitivity ($S$) of the sensors is defined based on the ratio of the relative resistance change ($\Delta R/R$) of the sensing element and the strain applied to the substrate ($\varepsilon_{sub}$): \begin{equation} S = (\Delta R/R)/\varepsilon_{sub} \label{eq:sensitivity} \end{equation} In addition, the relative resistance change $\Delta R/R$ can be calculated from the gauge factor ($GF$) of the material used to make the piezoresistive elements: $\Delta R/R = GF \varepsilon_{ind}$, where $\varepsilon_{ind}$ is the strain induced into the piezoresistor. In most of the conventional strain gauges as shown in Fig. \ref{fig:fig1} (a,b), the thickness of the sensing layer is typically below a few hundred nanometers, which is much smaller than that of the substrate. Therefore, the strain induced into the piezoresistive elements is approximately the same as that of the substrate ($\varepsilon_{ind} \approx \varepsilon_{sub}$). Consequently, to improve the sensitivity of strain sensors (e.g. enlarging $\Delta R/R$), electrical approaches which can enlarge the gauge factor ($GF$) are required. Nevertheless, as aforementioned, the existence of the large gauge factor in nanowires due to quantum confinement or surface state, is still considered as controversial. It is also evident from Eq. \ref{eq:sensitivity} that the sensitivity of strain sensors can also be improved using a mechanical approach, which enlarges the strain induced into the piezoresistive element. Figure \ref{fig:fig1}(c) shows our proposed nano strain-amplifier structure, in which the piezoresistive nanowires are locally fabricated at the centre of a released bridge. The key idea of this structure is that, under a certain strain applied to the substrate, a large strain will be concentrated at the locally fabricated SiC nanowires. The working principle of the nano strain-amplifier is similar to that of the well-known dogbone structure, which is widely used to characterize the tensile strength of materials \cite{dogbone1,dogbone2}. That is, when a stress is applied to the dogbone-shape of a certain material, a crack, if generated, will occur at the middle part of the dogbone. The large strain concentrated at the narrow area located at the centre part with respect to the wider areas located at outer region, causes the crack. Qualitative and quantitative explanations of the nano strain-amplifier are presented as follows. For the sake of simplicity, the released micro frame and nanowire (single wire or array) of the nano strain-amplifier can be considered as solid springs, Fig. \ref{fig:fig1}(d). The stiffness of these springs are proportional to their width ($w$) and inversely proportional to their length (l): $K \propto w/l$. Consequently, the model of the released nanowire and micro frames can be simplified as a series of springs, where the springs with higher stiffness correspond to the micro frame, and the single spring with lower stiffness corresponds to the nanowire. It is well-known in classical physics that, for serially connected springs, a larger strain will be concentrated in the low--stiffness string, while a smaller strain will be induced in the high--stiffness string \cite{Springbook}. The following analysis quantitatively explained the amplification of the strain. \begin{figure}[b!] \centering \includegraphics[width=3in]{Fig2} \vspace{-1em} \caption{Finite element analysis of the strain induced in to the nanowire array utilizing nano strain-amplifier.} \label{fig:fig2} \end{figure} When a tensile mechanical strain ($\varepsilon_{sub}$) is applied to the substrate, the released structure will also be elongated. Since the stiffness of the released frame is much smaller than that of the substrate, it is safe to assume that the released structure will follows the elongation of the substrate. The displacement of the released structure $\Delta L$ is: \begin{equation} \Delta L = \Delta L_m + \Delta L_n = L_m \varepsilon_m + L_n \varepsilon_n \label{eq:displacement} \end{equation} where $L_m$, $L_n$ are the length; $\Delta L_m$, $\Delta L_n$ are the displacement; and $\varepsilon_m$, $\varepsilon_n$ are the strains induced into the micro spring and nano spring, respectively. The subscripts m and n stand for the micro frames and nanowires, respectively. Furthermore, due to the equilibrium of the stressing force ($F$) along the series of springs, the following relationship is established: $F= K_m\Delta L_m = K_n \Delta L_n$, where $K_m$, $K_n$ are the stiffness of the released micro frames and nanowires, respectively. Consequently the relationship between the displacement of the micro frame (higher stiffness) and nanowires (lower stiffness) is: \begin{equation} \frac{\Delta L_m}{\Delta L_n}=\frac{K_n}{K_m}=\frac{L_mw_n}{L_nw_m} \label{eq:euili} \end{equation} Substituting Eqn. \ref{eq:euili} into Eqn. \ref{eq:displacement}, the strain induced into the locally fabricated nanowires is: \begin{equation} \varepsilon_n = \frac{\Delta L_n}{L_n} = \frac{1}{1-\frac{w_m-w_n}{w_m}\frac{L_m}{L}}\varepsilon_{sub} \label{eq:strainamp} \end{equation} Equation \ref{eq:strainamp} indicates that increasing the ratio of $w_m/w_n$ and $L_m/L_n$ significantly amplifies the strain induced into the nanowire from the strain applied to the substrate. This model is also applicable to the case of nanowire arrays, in which $w_n$ is the total width of all nanowires in the array. The theoretical model is then verified using the finite element analysis (FEA). In the FEA simulation, we compare the strain induced into (i) non released nanowires, (ii) the conventionally released nanowires, and (iii) our nano strain-amplifier structure, using COMSOL Multiphysics \texttrademark. In our nano strain amplifying structure, the width of the released frame was set to be 8 $\mu$m, while the width of each nanowire in the array (3 wires) was set to be 370 nm. The nanowires array structure was selected as it can enhance the electrical conductance of the SiC nanowires resistor which makes the subsequent experimental demonstration easier. The ratio between the length of nanowires and micro bridge was set to be 1: 20. With this geometrical dimensions, strain induced into nanowires array $\varepsilon_n$ was numerically calculated to be approximately 6 times larger than $\varepsilon_{sub}$, Eqn. \ref{eq:strainamp}. The simulation results show that for all structure, the elongation of non-released and released nanowires follow that of the substrate. In addition, strain was almost completely transferred into conventional released and non-released structures. Furthermore, the ratio of the strain induced in to the locally fabricated nanowires was estimated to be 5.9 times larger than that of the substrate, Fig. \ref{fig:fig2}. These results are in solid agreement with the theoretical analysis presented above. For a nanowire array with an average width of 470 nm, the amplified gain of strain was found to be 4.5. Based on the theoretical analysis, we conducted the following experiments to demonstrate the high sensitivity of SiC nanowire strain sensors using the nano strain-amplifier. A thin 3C-SiC film with its thickness of 300 nm was epitaxially grown on a 150 mm diameter Si wafer using low pressure chemical vapour deposition \cite{SiC_growth}. The film was \emph{in situ} doped using Al dopants. The carrier concentration of the p-type 3C-SiC was found to be $5 \times 10^{18}$ cm$^{-3}$, using a hot probe technique \cite{philip}. The details of the characteristics of the grown film can be found elsewhere \cite{Phan_JMC}. Subsequently, I-shape p-type SiC resistors with aluminum electrodes deposited on the surface were patterned using inductive coupled plasma (ICP) etching. As the piezoresistance of p-type 3C-SiC depends on crystallographic orientation, all SiC resistors of the present work were aligned along [110] direction to maximize the piezoresistive effect. Next, the micro scale SiC resistors were then released from the Si substrate using dry etching (XeF$_2$). Finally, SiC nanowire arrays were formed at the centre of the released bridge using focused ion beam (FIB). Two types of nanowire array were fabricated with three nanowires for each array. The average width of each nanowire in each type were 380 nm and 470 nm, respectively. Figure \ref{fig:fig3} shows the SEM images of the fabricated samples, including the conventional released structure, non-released nanowires, and the nano strain-amplifier. \begin{figure}[t!] \centering \includegraphics[width=3in]{Fig3} \caption{SEM image of SiC strain sensors. (a) Released SiC micro bridge used for the subsequent fabrication of the nano strain-amplifier; (b) SEM of a micro SiC resistor where the SiC nanowires array were formed using FIB; (c) SEM of non-released SiC nanowires; (d) SEM of locally fabricated SiC nanowires released from the Si substrate (nano strain-amplifier).} \label{fig:fig3} \vspace{-1em} \end{figure} The current voltage (I-V) curves of all fabricated samples were characterized using a HP 4145 \texttrademark ~parameter analyzer. The linear relationship between the applied voltage and measured current, indicated that Al made a good Ohmic contact with the highly doped SiC resistance, Fig. \ref{fig:IV}. Additionally, the electrical conductivity of both nanowires and micro frame estimated from the I-V curve and the dimensions of the resistors shows almost the same value. This indicated that the FIB process did not cause a significant surface damage to the fabricated nanowires. \begin{figure}[b!] \centering \includegraphics[width=3in]{Fig4} \vspace{-1.5em} \caption{Current voltage curves of the fabricated SiC resistors.} \label{fig:IV} \end{figure} The bending experiment was used to characterize the piezoresistive effect in micro size SiC resistors and locally fabricated SiC nanowire array. In this experiment one end of the Si cantilever (with a thickness of 625 $\mu$m, and a width of 7 mm) was fixed while the other end was deflected by applying different forces. The distance from the fabricated nanowires to the free end of the Si cantilever was approximately 45 mm. The strain induced into the Si substrate is $\varepsilon_\text{sub} = Mt/2EI$, where $M$ is the applied bending moment; and $t$, $E$ and $I$ are the thickness, Young's modulus and the moment of inertia of the Si cantilever, respectively. The response of the SiC resistance to applied strain was then measured using a multimeter (Agilent \texttrademark 34401 A). \begin{figure}[h!] \centering \includegraphics[width=3in]{Fig5.eps} \vspace{-1.5em} \caption{Experimental results. (a) A comparision between the relative resistance change in the nano strain-amplifiers, non released nanowires and released micro frames; (b) The repeatability of the SiC nanowires strain sensors utilizing the proposed structure.} \label{fig:DRR} \vspace{-1em} \end{figure} The relative resistance change ($\Delta R/R$) of the micro and nano SiC resistors was plotted against the strain induced into the Si substrate $\varepsilon_{sub}$, Fig. \ref{fig:DRR}(a). For all fabricated samples, the relative resistance change shows a good linear relationship with the applied strain ($\varepsilon_{sub}$). In addition, with the same applied strain to the Si substrate, the resistance change of the SiC nanowires using the nano strain-amplifier was much larger than that of the the SiC micro resistor and the conventional non-released SiC nanowires. In addition, reducing the width of the SiC nanowires also resulted in the increase of the sensitivity. The magnitude of the piezoresistive effect in the nano strain-amplifier as well as conventional structures were then quantitatively evaluated based on the effective gauge factor ($GF_{eff}$), which is defined as the ratio of the relative resistance change to the applied strain to the substrate: $GF_{eff} = (\Delta R/R)/\varepsilon_{sub}$. Accordingly, the effective gauge factor of the released micro SiC was found to be 28, while that of the non-released SiC nanowires was 35. From the data shown in Fig. \ref{fig:DRR}, the effective gauge factor of the 380 nm and 470 nm SiC nanowires in the nano strain-amplifier were calculated as 150 and 124, respectively. Thus for nanowire arrays with average widths of 380 nm and 470 nm, the sensitivity of the nano strain-amplifier was 5.4 times and 4.6 times larger than the bulk SiC, respectively. These results were consistent with analytical and numerical models presented above. The relative resistance change of the nano strain-amplifier also showed excellent linearity with the applied strain, with a linear regression of above 99\%. The resistance change of the nano strain-amplifier can also be converted into voltage signals using a Wheatstone bridge, Fig. \ref{fig:DRR}(b). The output voltage of the nano strain-amplifier increases with increasing tensile strains from 0 ppm to 180 ppm, and returned to the initial value when the strain was completely removed, confirming a good repeatability after several strain induced cycles. The linearity of the relative resistance change, and the repeatability indicate that the proposed structure is promising for strain sensing applications. In conclusion, this work presents a novel mechanical approach to obtain highly sensitive piezoresistance in nanowires based on a nano strain-amplifier. The key factor of the nano strain-amplifier lies on nanowires locally fabricated on a released micro structure. Experimental studies were conducted on SiC nanowires, confirming that by utilizing our nano strain-amplifier, the sensitivity of SiC nanowires was 5.4 times larger than that of conventional structures. This result indicated that the nano strain-amplifier is an excellent platform for ultra sensitive strain sensing applications.
1b77ae9f541b19668cc96624c7ec0f83945284e2
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{intro} Gas has a fundamental role in shaping the evolution of galaxies, through its accretion on to massive haloes, cooling and subsequent fuelling of star formation, to the triggering of extreme luminous activity around super massive black holes. Determining how the physical state of gas in galaxies changes as a function of redshift is therefore crucial to understanding how these processes evolve over cosmological time. The standard model of the gaseous interstellar medium (ISM) in galaxies comprises a thermally bistable medium (\citealt*{Field:1969}) of dense ($n \sim 100$\,cm$^{-3}$) cold neutral medium (CNM) structures, with kinetic temperatures of $T_{\rm k} \sim 100$\,K, embedded within a lower-density ($n \sim 1$\,cm$^{-3}$) warm neutral medium (WNM) with $T_{\rm k} \sim 10^{4}$\,K. The WNM shields the cold gas and is in turn ionized by background cosmic rays and soft X-rays (e.g. \citealt{Wolfire:1995, Wolfire:2003}). A further hot ($T_{\rm k} \sim 10^{6}$\,K) ionized component was introduced into the model by \cite{McKee:1977}, to account for heating by supernova-driven shocks within the inter-cloud medium. In the local Universe, this paradigm has successfully withstood decades of observational scrutiny, although there is some evidence (e.g. \citealt{Heiles:2003b}; \citealt*{Roy:2013b}; \citealt{Murray:2015}) that a significant fraction of the WNM may exist at temperatures lower than expected for global conditions of stability, requiring additional dynamical processes to maintain local thermodynamic equilibrium. Since atomic hydrogen (\mbox{H\,{\sc i}}) is one of the most abundant components of the neutral ISM and readily detectable through either the 21\,cm or Lyman $\alpha$ lines, it is often used as a tracer of the large-scale distribution and physical state of neutral gas in galaxies. The 21\,cm line has successfully been employed in surveying the neutral ISM in the Milky Way (e.g. \citealt{McClure-Griffiths:2009,Murray:2015}), the Local Group (e.g. \citealt{Kim:2003,Bruns:2005,Braun:2009,Gratier:2010}) and low-redshift Universe (see \citealt{Giovanelli:2016} for a review). However, beyond $z \sim 0.4$ (\citealt{Fernandez:2016}) \mbox{H\,{\sc i}} emission from individual galaxies becomes too faint to be detectable by current 21\,cm surveys and so we must rely on absorption against suitably bright background radio (21\,cm) or UV (Lyman-$\alpha$) continuum sources to probe the cosmological evolution of \mbox{H\,{\sc i}}. The bulk of neutral gas is contained in high-column-density damped Lyman-$\alpha$ absorbers (DLAs, $N_{\rm HI} \geq 2 \times 10^{20}$\,cm$^{-2}$; see \citealt*{Wolfe:2005} for a review), which at $z \gtrsim 1.7$ are detectable in the optical spectra of quasars. Studies of DLAs provide evidence that the atomic gas in the distant Universe appears to be consistent with a multi-phase neutral ISM similar to that seen in the Local Group (e.g. \citealt*{Lane:2000}; \citealt*{Kanekar:2001c}; \citealt*{Wolfe:2003b}). However, there is some variation in the cold and warm fractions measured throughout the DLA population (e.g. \citealt*{Howk:2005}; \citealt{Srianand:2005, Lehner:2008}; \citealt*{Jorgenson:2010}; \citealt{Carswell:2011, Carswell:2012, Kanekar:2014a}; \citealt*{Cooke:2015}; \citealt*{Neeleman:2015}). The 21-cm spin temperature affords us an important line-of-enquiry in unraveling the physical state of high-redshift atomic gas. This quantity is sensitive to the processes that excite the ground-state of \mbox{H\,{\sc i}} in the ISM (\citealt{Purcell:1956,Field:1958,Field:1959b,Bahcall:1969}) and therefore dictates the detectability of the 21\,cm line in absorption. In the CNM the spin temperature is governed by collisional excitation and so is driven to the kinetic temperature, while the lower densities in the WNM mean that the 21\,cm transition is not thermalized by collisions between the hydrogen atoms, and so photo-excitation by the background Ly $\alpha$ radiation field becomes important. Consequently the spin temperature in the WNM is lower than the kinetic temperature, in the range $\sim$1000 -- 5000\,K depending on the column density and number of multi-phase components (\citealt{Liszt:2001}). Importantly, the spin temperature measured from a single detection of extragalactic absorption is equal to the harmonic mean of the spin temperature in individual gas components, weighted by their column densities, thereby providing a method of inferring the CNM fraction in high-redshift systems. Surveys for 21\,cm absorption in known redshifted DLAs have been used to simultaneously measure the column density and spin temperature of \mbox{H\,{\sc i}} (see \citealt{Kanekar:2014a} and references therein). There is some evidence for an increase (at $4\,\sigma$ significance) in the spin temperature of DLAs at redshifts above $z = 2.4$, and a difference (at $6\,\sigma$ significance) between the distribution of spin temperatures in DLAs and the Milky Way (\citealt{Kanekar:2014a}). The implication that at least 90\,per\,cent of high-redshift DLAs may have CNM fractions significantly less than that measured for the Milky Way has important consequences for the heating and cooling of neutral gas in the early Universe and star formation (e.g. \citealt*{Wolfe:2003a}). However, these targeted observations rely on the limited availability of simultaneous 21\,cm and optical/UV data for the DLAs and assumes commonality between the column density probed by the optical and radio sight-lines. The first issue can be overcome by improving the sample statistics through larger 21\,cm line surveys of high-redshift DLAs, but the latter requires improvements to our methodology and understanding of the gas distribution in these systems. There are also concerns about the accuracy to which the fraction of the source structure subtended by the absorber can be measured in each system, which can only be resolved through spectroscopic very long baseline interferometry (VLBI). It has been suggested that the observed evolution in spin temperature could be biased by assumptions about the radio-source covering factor (\citealt{Curran:2005}) and its behaviour as a function of redshift (\citealt{Curran:2006b, Curran:2012b}). In this paper we consider an approach using the statistical constraint on the average spin temperature achievable with future large 21\,cm surveys using precursor telescopes to the Square Kilometre Array (SKA). This will enable independent verification of the evolution in spin temperature at high redshift and provide a method of studying the global properties of neutral gas below $z \approx 1.7$, where the Lyman\,$\alpha$ line is inaccessible using ground-based observatories. In an early attempt at a genuinely blind 21\,cm absorption survey, \cite{Darling:2011} used pilot data from the Arecibo Legacy Fast Arecibo L-band Feed Array (ALFALFA) survey to obtain upper limits on the column density frequency distribution from 21\,cm absorption at low redshift ($z \lesssim 0.06$). However, they also noted that the number of detections could be used to make inferences about the ratio of the spin temperature to covering factor. Building upon this work, \cite{Wu:2015} found that their upper limits on the frequency distribution function measured from the 40\,per\,cent ALFALFA survey ({$\alpha$}.40; \citealt{Haynes:2011}) could only be reconciled with measurements from other low-redshift 21\,cm surveys if the typical spin temperature to covering factor ratio was greater than 500\,K. At higher redshifts, \cite{Gupta:2009} found that the number density of 21\,cm absorbers in known \mbox{Mg\,{\sc ii}} absorbers appeared to decrease with redshift above $z \sim 1$, consistent with a reduction in the CNM fraction. We pursue this idea further by investigating whether future wide-field 21\,cm surveys can be used to measure the average spin temperature in distant galaxies that are rich in atomic gas. \section{The expected number of intervening \mbox{H\,{\sc i}} absorbers}\label{section:expected_number} We estimate the expected number of intervening \mbox{H\,{\sc i}} systems towards a sample of background radio sources by evaluating the following integral over all sight-lines \begin{equation}\label{equation:expected_number} \mu = \iint{f(N_{\rm HI},X)\,\mathrm{d}X\,\mathrm{d}N_{\rm HI}}, \end{equation} where $f(N_{\rm HI}, X)$ is the frequency distribution as a function of column density ($N_{\rm HI}$) and comoving path length ($X$). We use the results of recent surveys for 21\,cm emission in nearby galaxies (e.g. \citealt{Zwaan:2005}) and high-redshift Lyman-$\alpha$ absorption in the Sloan Digitial Sky Survey (SDSS; e.g. \citealt*{Prochaska:2005}; \citealt{Noterdaeme:2009}), which show that $f(N_{\rm HI}, X)$ can be parametrized by a gamma function of the form \begin{equation} f(N_{\rm HI}, X) = \left({f_{\ast} \over N_{\ast}}\right)\left({N_{\rm HI} \over N_{\ast}}\right)^{-\beta}\exp{\left(-{N_{\rm HI} \over N_{\ast}}\right)}\,\mathrm{cm}^{2}, \end{equation} where $f_{\ast} = 0.0193$, $\log_{10}(N_{\ast}) = 21.2$ and $\beta = 1.24$ at $z = 0$ (\citealt{Zwaan:2005}), and $f_{\ast} = 0.0324$, $\log_{10}(N_{\ast}) = 21.26$ and $\beta = 1.27$ at $z \approx 3$ (\citealt{Noterdaeme:2009}). While the observational data do not yet constrain models for evolution of the \mbox{H\,{\sc i}} distribution at intermediate redshifts between $z \sim 0.1$ and $3$\footnote{Measurements of $f(N_{\rm HI},X)$ at intermediate redshifts come from targeted ultra-violet surveys of DLAs using the \emph{Hubble Space Telescope} (\citealt*{Rao:2006}; \citealt{Neeleman:2016}). However, due to the limited sample sizes these are currently an order-of-magnitude less sensitive than the nearby 21-cm and high-redshift optical Lyman-$\alpha$ surveys.}, it is known to be much weaker than the significant decline seen in the global star-formation rate and molecular gas over the same epoch (e.g. \citealt{Lagos:2014}). We therefore carry out a simple linear interpolation between the low and high redshift epochs to estimate $f(N_{\rm HI},X)$ as a function of redshift. \begin{figure} \centering \includegraphics[width=0.475\textwidth]{width_dist.pdf} \caption{The distribution of 21\,cm line widths based on existing detections of intervening absorption at $z > 0.1$ (see the text for details of this sample). The sample size in each bin is denoted by the number above and errorbars denote the standard deviation. The solid red line is a log-normal fit to the data, from which we draw random samples for our analysis.}\label{figure:width_dist} \end{figure} The probability of detecting an absorbing system of given column density depends on the sensitivity of the survey, the flux density and structure of the background source and the fraction of \mbox{H\,{\sc i}} in the lower spin state, given by the spin temperature. We express the column density ($N_{\rm HI}$; in atoms\,cm$^{-2}$) in terms of the optical depth ($\tau$) and spin temperature ($T_{\rm spin}$; in K) by \begin{equation}\label{equation:column_density} N_{\rm HI} = 1.823\times10^{18}\,T_{\rm spin} \int{\tau(v)\mathrm{d}v}, \end{equation} where the integral is performed across the spectral line in the system rest-frame velocity $v$ (in km\,s$^{-1}$). We then express the optical depth in terms of the observables as \begin{equation} \tau = -\ln\left[1 + {\Delta{S}\over c_{\rm f}S_{\rm cont}}\right], \end{equation} where $\Delta{S}$ is the observed change in flux density due to absorption, $S_{\rm cont}$ is the background continuum flux density and $c_{\rm f}$ is the (often unknown) fraction of background flux density subtended by the intervening gas. \begin{figure} \centering \includegraphics[width=0.465\textwidth]{covfact_dist.pdf} \caption{The distribution of \mbox{H\,{\sc i}} covering factors from the main sample of \citet{Kanekar:2014a}, which were estimated using the fraction of total continuum flux density in the quasar core. The sample size in each bin is denoted by the number above and errorbars denote the standard deviation. The solid red line is the uniform distribution, from which we draw random samples for our analysis.}\label{figure:covfact_dist} \end{figure} We assume that a single intervening system can be described by a Gaussian velocity distribution of full width at half maximum (FWHM) dispersion ($\Delta{v_{\rm 50}}$) and peak optical depth ($\tau_{\rm peak}$), so that \autoref{equation:column_density} can be re-written as \begin{equation}\label{equation:column_density_gaussian} N_{\rm HI} = 1.941\times10^{18}\,T_{\rm spin}\,\tau_{\rm peak}\,\Delta{v_{\rm 50}}. \end{equation} If we further assume that the rms spectral noise is Gaussian, with a standard deviation $\sigma_{\rm chan}$ per independent channel $\Delta{v_{\rm chan}}$, then the 5$\sigma$ column density detection limit is given by \begin{equation} N_{5\sigma} \approx 1.941\times10^{18}\,T_{\rm spin}\,\tau_{\rm 5\sigma}\,\Delta{v_{\rm conv}}, \end{equation} where \begin{equation}\label{equation:optical_depth_limit} \tau_{5\sigma} \approx -\ln\left[1 - {5\,\sigma_{\rm chan}\over c_{\rm f}\,S_{\rm cont}}\sqrt{\Delta{v}_{\rm chan}\over \Delta{v_{\rm conv}}}\right], \end{equation} and $\Delta{v_{\rm conv}} \approx \sqrt{\Delta{v}_{\rm chan}^{2} + \Delta{v}_{50}^{2}}$, which is the observed width of the line, given by the convolution of the physical velocity distribution and the spectral resolution of the telescope. We now redefine $\mu$ as the expected number of intervening \mbox{H\,{\sc i}} detections in our survey as a function of the column density sensitivity along each sight-line where each comoving path element $\delta{X}(z)$\footnote{For the purposes of this work we adopt a flat $\Lambda$ cold dark matter cosmology with $H_{0}$ = 70\,km\,s$^{-1}$, $\Omega_\mathrm{M}$ = 0.3 and $\Omega_{\Lambda}$ = 0.7. } in the integral defined by \autoref{equation:expected_number} is given by \begin{equation} \delta{X}(z)= \begin{cases} {\delta{z}\,(1+z)^{2}\over \sqrt{(1+z)^{2}(1+z\Omega_{\rm M})-z(z+2)\Omega_{\rm \Lambda}}}, & \text{if}\ N_{\rm HI} \geq N_{5\sigma}, \\ 0, & \text{otherwise}. \end{cases} \end{equation} To calculate the column density sensitivity for each comoving element we draw random samples for $\Delta{v}_{50}$ and $c_{\rm f}$ from continuous prior distributions based on existing evidence. In the case of $\Delta{v}_{50}$ we use a log-normal distribution obtained from a simple least-squares fit to the sample distribution from previous 21-cm absorption surveys reported in the literature (see \autoref{figure:width_dist})\footnote{References for the literature sample of line widths shown in \autoref{figure:width_dist}: \citet*{Briggs:2001}; \citet*{Carilli:1993}; \citet*{Chengalur:1999}; \citet{Chengalur:2000, Curran:2007b, Davis:1978, Ellison:2012, Gupta:2009, Gupta:2012, Gupta:2013}; \citet{Kanekar:2001b,Kanekar:2003b}; \citet{Kanekar:2001c, Kanekar:2006, Kanekar:2009a, Kanekar:2013, Kanekar:2014a}; \citet{Kanekar:2003a}, \citet*{Kanekar:2007}; \citet*{Kanekar:2014b}; \citet{ Lane:2001, Lovell:1996, York:2007, Zwaan:2015}.}, assuming that this correctly describes the true distribution for the population of DLAs. However, direct measurement of the \mbox{H\,{\sc i}} covering factor is significantly more difficult and so for the purposes of this work we draw random samples assuming a uniform distribution between 0 and 1. In \autoref{figure:covfact_dist}, we show a comparison between this assumption and the sample distribution estimated by \cite{Kanekar:2014a} from their main sample of 37 quasars. Kanekar et al. used VLBI synthesis imaging to measure the fraction of total quasar flux density contained within the core, which was then used as a proxy for the covering factor. By carrying out a two-tailed Kolmogorov-Smirnov (KS) test of the hypothesis that the Kanekar et al. data are consistent with our assumed uniform distribution, we find that this hypothesis is rejected at the 0.05 level, but not at the 0.01 level (this outcome is dominated by the paucity of quasars in the sample with $c_{\rm f} \lesssim 0.2$). It is therefore possible that the population distribution of \mbox{H\,{\sc i}} covering factors may deviate somewhat from the uniform distribution assumed in this work. We discuss the implications of this further in \autoref{section:covering_factor}. \section{A 21\,cm absorption survey with ASKAP}\label{section:all_sky_survey} We use the Australian Square Kilometre Array Pathfinder (ASKAP; \citealt{Johnston:2007}) as a case study to demonstrate the expected results from planned wide-field surveys for 21\,cm absorption (e.g. the ASKAP First Large Absorption Survey in \mbox{H\,{\sc i}} -- Sadler et al., the MeerKAT Absorption Line Survey -- Gupta et al., and the Search for HI absorption with AperTIF -- Morganti et al.). ASKAP is currently undergoing commissioning. Proof-of-concept observations with the Boolardy Engineering Test Array (\citealt{Hotan:2014}) have already been used to successfully detect a new \mbox{H\,{\sc i}} absorber associated with a probable young radio galaxy at $z = 0.44$ (\citealt{Allison:2015a}). Here we predict the outcome of a future 2\,h-per-pointing survey of the entire southern sky ($\delta \leq +10\degr$) using the full 36-antenna ASKAP in a single 304\,MHz band between 711.5 and 1015.5\,MHz, equivalent to \mbox{H\,{\sc i}} redshifts between $z = 0.4$ and 1.0. \begin{figure} \centering \includegraphics[width=0.475\textwidth]{nsources_flux.pdf} \caption{The number of radio sources in our simulated southern sky survey ($\delta \leq +10\degr$) estimated from existing catalogues at $L$-band frequencies (see the text for details). The grey region encloses the expected number across the 711.5 - 1015.5\,MHz ASKAP frequency band, assuming a canonical spectral index of $\alpha = -0.7$.}\label{figure:nsources_flux} \end{figure} Our expectations of the ASKAP performance are based on preliminary measurements by \cite{Chippendale:2015} using the prototype Mark {\sc II} phase array feed. We estimate the noise per spectral channel using the radiometer equation \begin{equation} \sigma_{\rm chan} = {S_{\rm system} \over \sqrt{n_{\rm pol}\,n_{\rm ant}\,(n_{\rm ant} - 1)\,\Delta{t}_{\rm in}\,\Delta{\nu}_{\rm chan}}}, \end{equation} where $S_{\rm system}$ is the system equivalent flux density, $n_{\rm pol}$ is the number of polarizations, $n_{\rm ant}$ is the number of antennas, $\Delta{t}_{\rm in}$ is the on-source integration time and $\Delta{\nu}_{\rm chan}$ is the spectral resolution in frequency. The sensitivity of the telescope in the 711.5 - 1015.5\,MHz band is expected to vary between $S_{\rm system} \approx 3200$ and $2000$\,Jy, with the largest change in sensitivity between 700 and 800\,MHz. ASKAP has dual linear polarization feeds, 36 antennas and a fine filter bank that produces 16\,416 independent channels across the full 304\,MHz bandwidth, so the expected noise per 18.5\,kHz channel in a 2\,h observation is approximately 5.5 - 3.5\,mJy\,beam$^{-1}$ across the band. In the case of an actual survey, the true sensitivity will of course be recorded in the spectral data as a function of redshift (see e.g. \citealt{Allison:2015a}), but for the purposes of the simulated survey presented in this work we split the band into several frequency bins to capture the variation in sensitivity and velocity resolution (which is in the range 7.8\,km\,s$^{-1}$ at 711.5\,MHz to 5.5\,km\,s$^{-1}$ at 1015.5\,MHz). \begin{figure} \centering \includegraphics[width=0.475\textwidth]{zdist.pdf} \caption{The distribution of CENSORS sources (\citealt{Brookes:2008}) brighter than 10\,mJy beyond a given redshift $z$. The red line denotes the cumulative distribution calculated from the parametric model of \citet{deZotti:2010}.}\label{figure:zdist} \end{figure} In order to simulate a realistic survey of the southern sky we select all radio sources south of $\delta = +10\degr$ from catalogues of the National Radio Astronomy Observatory Very Large Array Sky Survey (NVSS, $\nu = 1.4$\,GHz, $S_{\rm src} \gtrsim 2.5$\,mJy; \citealt{Condon:1998}), the Sydney University Molonglo SkySurvey ($\nu = 843$\,MHz, $S_{\rm src} \gtrsim 10$\,mJy; \citealt{Mauch:2003}) and the second epoch Molonglo Galactic Plane Survey ($\nu = 843$\,MHz, $S_{\rm src} \gtrsim 10$\,mJy; \citealt{Murphy:2007}). The source flux densities, used to calculate the optical depth limit in \autoref{equation:optical_depth_limit}, are estimated at the centre of each frequency bin by extrapolating from the catalogue values and assuming a canonical spectral index of $\alpha = -0.7$. In \autoref{figure:nsources_flux}, we show the resulting cumulative distribution of radio sources in our sample as a function of flux density across the band. \begin{figure} \centering \includegraphics[width=0.475\textwidth]{nsources_tau.pdf} \caption{The number of sources in our simulated ASKAP survey with a 21\,cm opacity sensitivity greater than or equal to $\tau_{5\sigma}$, as defined by \autoref{equation:optical_depth_limit}. The grey region encloses opacity sensitivities for the 711.5 - 1011.5\,MHz band. Random samples for the line FWHM and covering factor were drawn from the distributions shown in \autoref{figure:width_dist} and \autoref{figure:covfact_dist}.}\label{figure:nsources_tau} \end{figure} For any given sight-line, the redshift interval over which absorption may be detected is dependent upon the distance to the continuum source. The lack of accurate spectroscopic redshift measurements for most radio sources over the sky necessitates the use of a statistical approach based on a model for the source redshift distribution. We therefore apply a statistical weighting to each comoving path element $\delta{X}(z)$ such that the expected number of absorber detections is now given by \begin{equation}\label{equation:weighted_sum_number} \mu = \iint{f(N_{\rm HI},X)\,\mathcal{F}_{\rm src}(z^{\prime} \geq z)\,\mathrm{d}X\,\mathrm{d}N_{\rm HI}}, \end{equation} where \begin{equation} \mathcal{F}_{\rm src}(z^{\prime} \geq z) = {\int_{z}^{\infty} \mathcal{N}_{\rm src}(z^{\prime})\mathrm{d}z^{\prime}\over\int_{0}^{\infty}\mathcal{N}_{\rm src} (z^{\prime})\mathrm{d}z^{\prime}}, \end{equation} and $\mathcal{N}_{\rm src}(z)$ is the number of radio sources as a function of redshift. To estimate $\mathcal{N}_{\rm src}(z)$ we use the Combined EIS-NVSS Survey Of Radio Sources (CENSORS; \citealt{Brookes:2008}), which forms a complete sample of radio sources brighter than 7.2\,mJy at 1.4\,GHz with spectroscopic redshifts out to cosmological distances. In \autoref{figure:zdist} we show the distribution of CENSORS sources brighter than 10\,mJy beyond a given redshift $z$, and the corresponding analytical function derived from the model fit of \citet{deZotti:2010}, given by \begin{equation} \mathcal{N}_{\rm src}(z) \approx 1.29 + 32.37z - 32.89z^{2} + 11.13z^{3} - 1.25z^{4}, \end{equation} which we use in our analysis. For the redshifts spanned by our simulated ASKAP survey, the fraction of background sources evolves from 87\,per\,cent at $z = 0.4$ to 53\,per\,cent at $z = 1.0$. We assume that this redshift distribution applies to any sight-line irrespective of the continuum flux density. However, this assumption is only true if the source population in the target sample evolves such that the effect of distance is nullified by an increase in luminosity. Given this criterion, and the sensitivity of our simulated survey, we limit our sample to sources with flux densities between 10 and 1000\,mJy, which are dominated by the rapidly evolving population of high-excitation radio galaxies and quasars (e.g. \citealt{Jackson:1999, Best:2012, Best:2014, Pracy:2016}) and for which the redshift distribution is known to be almost independent of flux density (e.g. \citealt{Condon:1984, Condon:1998}). In \autoref{figure:nsources_tau}, we show the number of sources from this sub-sample as a function of opacity sensitivity [as defined by \autoref{equation:optical_depth_limit}], drawing random samples of the line FWHM and covering factor from the distributions shown in \autoref{figure:width_dist} and \autoref{figure:covfact_dist}. There are approximately 190\,000 sightlines with sufficient sensitivity to detect absorption of optical depth greater than $\tau_{5\sigma} \approx 1.0$ and 25\,000 sensitive to optical depths greater than $\tau_{5\sigma} \approx 0.1$. Since this distribution converges at optical depth sensitivities greater than $\tau_{5\sigma} \approx 5$, the population of sources fainter than 10\,mJy, which are excluded from our simulated ASKAP survey, would not significantly contribute to further detections of absorption. Similarly, while sources brighter than 1\,Jy are good probes of low-column-density \mbox{H\,{\sc i}} gas, they do not constitute a sufficiently large enough population to significantly affect the total number of absorber detections expected in the survey and can also be safely excluded. Based on these assumptions, we can estimate the number of absorbers we would expect to detect in our survey with ASKAP as a function of spin temperature. In \autoref{figure:ndetections_nhi}, we show the expected detection yield as a cumulative function of column density. We show results for two scenarios where the spin temperature is fixed at a single value of either 100 or 1000\,K, and the line FWHM and covering factors are drawn from the random distributions shown in \autoref{figure:width_dist} and \autoref{figure:covfact_dist}. We find that for both these cases the expected number of detections is not sensitive to column densities below the DLA definition of $N_{\rm HI} = 2\times 10^{20}$\,cm$^{-2}$. We also show in \autoref{figure:ndetections_total} the expected total detection yield (integrated over all \mbox{H\,{\sc i}} column densities) as a function of a single spin temperature $T_\mathrm{spin}$ and line FWHM $\Delta{v}_\mathrm{50}$. We find that for typical spin temperatures of a few hundred kelvin (consistent with the typical fraction of CNM observed in the local Universe) and a line FWHM of approximately $20$\,km\,s$^{-1}$, a wide-field 21\,cm survey with ASKAP is expected to yield $\sim 1000$ detections. However, even moderate evolution to a higher spin temperature in the DLA population should see significant reduction in the detection yield from this survey. \section{Inferring the average spin temperature}\label{section:spin_temp} \begin{figure} \centering \includegraphics[width=0.475\textwidth]{ndetections_nhi.pdf} \caption{The expected number of absorber detections (as a cumulative function of column density) in our simulated ASKAP survey. We show two scenarios for a single spin temperature $T_{\rm spin} = 100$ and $1000$\,K, where we have drawn random samples for the line FWHM and covering factor from the distributions shown in \autoref{figure:width_dist} and \autoref{figure:covfact_dist}. In both cases we find that the expected number of detections is not sensitive to column densities below $N_{\rm HI} = 2 \times 10^{20}$\,cm$^{-2}$, indicating that such a survey will only be sensitive to DLA systems.}\label{figure:ndetections_nhi} \end{figure} We cannot directly measure the spin temperatures of individual systems without additional data from either 21\,cm emission or Lyman-$\alpha$ absorption. However, from \autoref{figure:ndetections_total} it is evident that the total number of absorbing systems expected to be detected with a reasonably large 21\,cm survey is strongly dependent on the assumed value for the spin temperature. Therefore, by comparing the actual survey yield with that expected from the known \mbox{H\,{\sc i}} distribution, we can infer the average spin temperature of the atomic gas within the DLA population for a given redshift interval. Assuming that the total number of detections follows a Poisson distribution, the probability of detecting $\mathcal{N}$ intervening absorbing systems is given by \begin{equation} p(\mathcal{N}|\overline{\mu}) = {\overline{\mu}^{\mathcal{N}} \over \mathcal{N}!} \mathrm{e}^{-\overline{\mu}}, \end{equation} where $\overline{\mu}$ is the expected total number of detections given by the integral \begin{equation} \overline{\mu} = \iiint{\mu(T_{\rm spin},\Delta{v}_{50},c_{\rm f})\rho(T_{\rm spin},\Delta{v}_{50},c_{\rm f})\mathrm{d}T_{\rm spin}\mathrm{d}\Delta{v}_{50}\mathrm{d}c_{\rm f}}, \end{equation} and $\rho$ is the distribution of systems as a function of spin temperature, line FWHM and covering factor. We assume that all three of these variables are independent\footnote{In the case where thermal broadening contributes significantly to the velocity dispersion, and the spin temperature is dominated by collisional excitation, the assumption that these are independent may no longer hold. However, given that collisional excitation dominates in the CNM, where $T_{\rm spin} \sim 100\,\mathrm{K}$, the velocity dispersion would have to satisfy $\Delta{v}_{50} \ll 10$\,km\,s$^{-1}$ (c.f. the distribution shown in \autoref{figure:width_dist}).} so that $\rho$ factorizes into functions of each. We then marginalise over the covering factor and line width distributions shown in \autoref{figure:width_dist} and \autoref{figure:covfact_dist} so that the expression for $\overline{\mu}$ reduces to \begin{equation}\label{equation:harmonic_spin_temp} \overline{\mu} = \int{\mu(T_{\rm spin})\rho(T_{\rm spin})\mathrm{d}T_{\rm spin}} = \mu(\overline{T}_{\rm spin}), \end{equation} where $\overline{T}_{\rm spin}$ is the harmonic mean of the unknown spin temperature distribution, weighted by column density. This is analogous to the spin temperature inferred from the detection of absorption in a single intervening galaxy averaged over several gaseous components at different temperatures (e.g. \citealt{Carilli:1996}). \begin{figure} \centering \includegraphics[width=0.475\textwidth]{ndetections_total.pdf} \caption{The expected total number of detections in our simulated ASKAP survey, as a function of a single spin temperature ($T_{\rm spin}$) and line FWHM ($\Delta{v}_{50}$). The vertical dotted lines enclose the velocity resolution across the observed frequency band. We draw random samples for the covering factor from a uniform distribution between 0 and 1 (as shown in \autoref{figure:covfact_dist}). The contours are truncated at $\mu < 1$ for clarity.}\label{figure:ndetections_total} \end{figure} In the event of the survey yielding $\mathcal{N}$ detections, we can calculate the posterior probability density of $\overline{T}_{\rm spin}$ using the following relationship between conditional probabilities \begin{equation} p(\overline{T}_{\rm spin}|\mathcal{N}) = {p(\mathcal{N}|\overline{T}_{\rm spin})p(\overline{T}_{\rm spin})\over p(\mathcal{N})}, \end{equation} where $p(\overline{T}_{\rm spin})$ is our prior probability density for $\overline{T}_{\rm spin}$ and $p(\mathcal{N})$ is the marginal probability of the number of detections, which can be treated as a normalizing constant. The minimally informative Jeffreys prior for the mean value $\mu$ of a Poisson distribution is $1/\sqrt{\mu}$ (\citealt{Jeffreys:1946})\footnote{A suitable alternative choice for the prior is the standard scale-invariant form $1/\mu$ (e.g. \citealt{Jeffreys:1961, Novick:1965, Villegas:1977}). While we find that our choice of non-informative prior has negligible effect on the spin temperature posterior for the full \mbox{H\,{\sc i}} absorption survey, as one would expect this choice becomes more important for smaller surveys. For the early-science 1000\,deg$^{2}$ survey discussed in \autoref{section:tspin_results} we find that the difference in these two priors produces a $\sim 2$ to 20\,per\,cent effect in the posterior. However, in all cases considered this change is smaller than the 68.3\,per\,cent credible interval spanned by the posterior.}. From \autoref{equation:harmonic_spin_temp} it therefore follows that a suitable form for the non-informative spin temperature prior is $p(\overline{T}_{\rm spin}) = 1/\sqrt{\overline{\mu}}$, so that \begin{equation}\label{equation:tspin_prob} p(\overline{T}_{\rm spin}|\mathcal{N}) = C^{-1}\,{\overline{\mu}^{(\mathcal{N}-1/2)} \over \mathcal{N}!} \mathrm{e}^{-\overline{\mu}}, \end{equation} where the distribution is normalised to unit total probability by evaluating the integral \begin{equation} C = \int{{\overline{\mu}^{(\mathcal{N}-1/2)} \over \mathcal{N}!} \mathrm{e}^{-\overline{\mu}}}\,\mathrm{d}\overline{T}_{\rm spin}. \end{equation} The probabilistic relationship given by \autoref{equation:tspin_prob} and the expected detection yield derived in \autoref{section:all_sky_survey} can be used as a frame-work for inferring the harmonic-mean spin temperature using the results of any homogeneous 21-cm survey. We have assumed that we can accurately distinguish intervening absorbing systems from those associated with the host galaxy of the radio source. However, any 21-cm survey will be accompanied by follow-up observations, at optical and sub-mm wavelengths, which will aid identification. Furthermore, future implementation of probabilistic techniques to either use photometric redshift information or distinguish between line profiles should provide further disambiguation. Of course we have not yet accounted for any error in our estimate of $\overline{\mu}$, which will increase our uncertainty in $\overline{T}_\mathrm{spin}$. In the following section we discuss these possible sources of error and their effect on the result. \section{Sources of error}\label{section:errors} Our estimate of the expected number of 21\,cm absorbers is dependent upon several distributions describing the properties of the foreground absorbing gas and the background source distribution. For future large-scale 21\,cm surveys, the accuracy to which we can infer the harmonic mean of the spin temperature distribution will eventually be limited by the accuracy to which we can measure these other distributions. In this section, we describe these errors and their propagation through to the estimate of $\overline{T}_\mathrm{spin}$, summarizing our results in \autoref{table:tspin_uncertainties}. \subsection{The covering factor}\label{section:covering_factor} \subsubsection{Deviation from a uniform distribution between 0 and 1} The fraction $c_{\rm f}$ by which the foreground gas subtends the background radiation source is difficult to measure directly and is thereby a significant source of error for 21\,cm absorption surveys. In this work, we have assumed a uniform distribution for $c_{\rm f}$, taking random values between 0 and 1. In \autoref{section:expected_number}, we tested this assumption by comparing it with the distribution of flux density core fractions in a sample of 37 quasars, used by \cite{Kanekar:2014a} as a proxy for the covering factor. By carrying out a two-tailed KS test, we found some evidence (at the 0.05 level) that this quasar sample was inconsistent with our assumption of a uniform distribution between 0 and 1. Noticeably there seems to be an under-representation of quasars in the Kanekar et al. sample with estimated $c_{\rm f} \lesssim 0.2$. In the low optical depth limit, the detection rate is dependent on the ratio of spin temperature to covering factor, in which case a fractional deviation in $c_{\rm f}$ will propagate as an equal fractional deviation in $\overline{T}_\mathrm{spin}$. Based on the difference seen in the covering factor distribution of the Kanekar et al. sample and the uniform distribution, we assume that the spin temperature can deviate by as much as $\pm$10\,per\,cent. \subsubsection{Evolution with redshift} We also consider that the covering factor distribution may evolve with redshift, which would mimic a perceived evolution in the average spin temperature. Such an effect was proposed by \cite{Curran:2006b} and \cite{Curran:2012b}, who claimed that the relative change in angular-scale behaviour of absorbers and radio sources between low- and high-redshift samples could explain the apparent evolution of the spin temperature found by \cite{Kanekar:2003b}. To test for this effect in their larger DLA sample, \cite{Kanekar:2014a} considered a sub-sample at redshifts greater than $z = 1$, for which the relative evolution of the absorber and source angular sizes should be minimal. While the significance of their result was reduced by removing the lower redshift absorbers from their sample, they still found a difference at $3.5\,\sigma$ significance between spin temperature distributions in the two DLA sub-samples separated by a median redshift of $z = 2.683$. Future surveys with ASKAP and the other SKA pathfinders will search for \mbox{H\,{\sc i}} absorption at intermediate redshifts ($z \sim 1$), where the relative evolution of the absorber and source angular sizes is expected to be more significant than for the higher redshift DLA sample considered by \cite{Kanekar:2014a}. We therefore consider the potential effect of this cosmological evolution on the inferred value of $\overline{T}_\mathrm{spin}$. We approximate the covering factor using the following model of \cite{Curran:2006b} \begin{equation}\label{equation:covering_factor} c_{f} \approx \begin{cases} \left({\theta_{\rm abs}\over \theta_{\rm src}}\right)^{2}, & \text{if}\ \theta_{\rm abs} < \theta_{\rm src}, \\ 1, & \text{otherwise}, \end{cases} \end{equation} where $\theta_{\rm abs}$ and $\theta_{\rm src}$ are the angular sizes of the absorber and background source, respectively. Under the small-angle approximation $\theta_{\rm abs} \approx {d_{\rm abs}/D_{\rm abs}}$ and $\theta_{\rm abs} \approx {d_{\rm src}/D_{\rm src}}$, where $d_{\rm abs}$ and $D_{\rm abs}$ are the linear size and angular diameter distance of the absorber, and likewise $d_{\rm src}$ and $D_{\rm src}$ are the linear size and angular diameter distance of the background source. Assuming that the ratio $d_{\rm abs}/d_{\rm src}$ is randomly distributed and independent of redshift, any evolution in the covering factor is therefore dominated by relative changes in the angular diameter distances. We calculate the expected angular diameter distance ratio at a redshift $z$ by \begin{equation} \left\langle{D_{\rm abs}\over D_{\rm src}}\right\rangle_{z} = D_{\rm abs}(z){\int_{z}^{\infty} \mathcal{N}_{\rm src}(z^{\prime})D_{\rm src}(z^{\prime})^{-1}\mathrm{d}z^{\prime}\over\int_{z}^{\infty}\mathcal{N}_{\rm src} (z^{\prime})\mathrm{d}z^{\prime}}, \end{equation} which, for the source redshift distribution model given by \cite{deZotti:2010}, evolves from 0.7 at $z = 0.4$ to 1.0 at $z = 1.0$ (see \autoref{figure:dang_ratio}). We note that this is consistent with the behaviour measured by \cite{Curran:2012b} for the total sample of DLAs observed at 21\,cm wavelengths. By applying this as a correction to the otherwise uniformly distributed covering factor (using \autoref{equation:covering_factor}), we find that the inferred value of $\overline{T}_{\rm spin}$ systematically increases by approximately 30 per\,cent. \begin{figure} \centering \includegraphics[width=0.475\textwidth]{dang_ratio.pdf} \caption{The expected redshift behaviour of $D_{\rm abs}/D_{\rm src}$ based on the \citet{deZotti:2010} model for the radio source redshift distribution.}\label{figure:dang_ratio} \end{figure} \subsection{The $\bmath{N_{\rm HI}}$ frequency distribution} \subsubsection{Uncertainty in the measurement of $f(N_{\rm HI},X)$} We assume that $f(N_{\rm HI}, X)$ is relatively well understood as a function of redshift by interpolating between model gamma functions fitted to the distributions at $z = 0$ and $3$. However, these distributions were measured from finite samples of galaxies, which of course have associated uncertainties that need to be considered. In the case of the data presented by \cite{Zwaan:2005} and \cite{Noterdaeme:2009}, both have typical measurement uncertainties in $f(N_{\rm HI}, X)$ of approximately 10\,per\,cent over the range of column densities for which our simulated ASKAP survey is sensitive (see \autoref{figure:ndetections_nhi}). This will propagate as a 10\,per\,cent fractional error in the expected number of absorber detections, and contribute a similar percentage uncertainty in the inferred average spin temperature. \subsubsection{Correcting for 21\,cm self-absorption} In the local Universe, \cite{Braun:2012} showed that self-absorption from opaque \mbox{H\,{\sc i}} clouds identified in high-resolution images of the Local Group galaxies M31, M33 and the Large Magellanic Cloud may necessitate a correction to the local atomic mass density of up to 30\,per\,cent. Although it is not yet clear whether this small sample of Local Group galaxies is representative of the low-redshift population, it is useful to understand how this effect might propagate through to our average spin temperature measurement. We therefore replace the gamma-function parametrization of the local $f(N_{\rm HI})$ given by \cite{Zwaan:2005} with the non-parametric values given in table\,2 of \cite{Braun:2012}, and recalculate $\overline{T}_{\rm spin}$. For an all-sky survey with the full 36-antenna ASKAP we find that $\overline{T}_{\rm spin}$ increases by $\sim$30 for 100 detections and $\sim$10\,per\,cent for 1000 detections. Note that the correction increases for low numbers of detections, which are dominated by the highest column density systems. \subsubsection{Dust obscuration bias in optically-selected DLAs} At higher redshifts, it is possible that the number density of optically-selected DLAs could be significantly underestimated as a result of dust obscuration of the background quasar (\citealt{Ostriker:1984}). This would cause a reduction in the $f(\mbox{H\,{\sc i}}, X)$ measured from optical surveys, thereby significantly underestimating the expected number of intervening 21\,cm absorbers at high redshifts. The issue is further compounded by the expectation that the highest column density DLAs ($N_{\rm HI} \gtrsim 10^{21}$\,cm$^{-2}$), for which future wide-field 21\,cm surveys are most sensitive (see \autoref{figure:ndetections_nhi}), may contain more dust than their less-dense counterparts. This conclusion was supported by early analyses of the existing quasar surveys at that time (e.g. \citealt{Fall:1993}), which indicated that up to 70\,per\,cent of quasars could be missing from optical surveys through the effect of dust obscuration, albeit with large uncertainties. However, subsequent optical and infrared observations of radio-selected quasars (e.g. \citealt{Ellison:2001}; \citealt*{Ellison:2005}; \citealt{Jorgenson:2006}), which are free of the potential selection biases associated with these optical surveys, found that the severity of this issue was substantially over-estimated and that there was minimal evidence in support of a correlation between the presence of DLAs and dust reddening. Furthermore, the \mbox{H\,{\sc i}} column density frequency distribution measured by \cite{Jorgenson:2006} was found to be consistent with the optically-determined gamma-function parametrization of \cite{Prochaska:2005}, with no evidence of DLA systems missing from the SDSS sample at a sensitivity of $N_{\rm HI} \lesssim 5 \times 10^{21}$\,cm$^{-2}$. Although radio-selected surveys of quasars are free of the selection biases associated with optical surveys, they do typically suffer from smaller sample sizes and are therefore less sensitive to the rarer DLAs with the highest column densities. Another approach is to directly test whether optically-selected quasars with intervening DLAs, selected from the SDSS sample, are systematically more dust reddened than a control sample of non-DLA quasars. Comparisons in the literature are based on several different colour indicators, which include the spectral index (e.g. \citealt{Murphy:2004,Murphy:2016}), spectral stacking (e.g. \citealt{Frank:2010, Khare:2012}) and direct photometry (e.g. \citealt*{Vladilo:2008}; \citealt{Fukugita:2015}). The current status of these efforts is summarized by \citet{Murphy:2016}, showing broad support for a missing DLA population at the level of $\sim$5\,per\,cent but highlighting that tension still exists between different dust measurements. No substantial evidence has yet been found to support a correlation between the dust reddening and \mbox{H\,{\sc i}} column density in these optically selected DLA surveys (e.g. \citealt{Vladilo:2008, Khare:2012, Murphy:2016}). In an attempt to reconcile the differences and myriad biases associated with these techniques, \cite{Pontzen:2009} carried out a statistically-robust meta-analysis of the available optical and radio data, using a Bayesian parameter estimation approach to model the dust as a function of column density and metallicity. They found that the expected fraction of DLAs missing from optical surveys is 7\,per\,cent, with fewer than 28\,per\,cent missing at 3\,$\sigma$ confidence. Based on this body of work we therefore assume that approximately 10\,per\,cent of DLAs are missing from the SDSS sample of \cite{Noterdaeme:2009} and consider the affect on our estimate of $\overline{T}_{\rm spin}$. We further assume that there is no dependance on column density, an assumption which is supported by the aforementioned observational data for the range of column densities to which our 21\,cm survey is sensitive. We find that increasing the high-redshift column density frequency distribution by 10\,per\,cent introduces a systematic increase of approximately 3\,per\,cent in the expected number of detections for the redshifts covered by our ASKAP surveys. We note that this error will increase significantly for 21\,cm surveys at higher redshifts where the optically derived $f(\mbox{H\,{\sc i}}, X)$ dominates the calculation of the expected detection rate. \subsection{The radio source background} As described in \autoref{section:all_sky_survey}, we weight the comoving path-length for each sight-line by a statistical redshift distribution in order to account for evolution in the radio source background. We use the parametric model of \cite{deZotti:2010}, which is derived from fitting the measured redshifts of \cite{Brookes:2008} for CENSORS sources brighter than 10\,mJy, and assume that this applies to all sources in the range 10 - 1000\,mJy. In \autoref{figure:zdist}, we show the cumulative distribution of sources located behind a given redshift and the associated measurement uncertainty given by the errorbars. For the intermediate redshifts covered by the ASKAP survey, the fractional uncertainty in this distribution increases from $\sigma_{\mathcal{F}_{\rm src}}/\mathcal{F}_{\rm src} \approx 3.5$ to 8\,per\,cent between $z = 0.4$ and 1.0, which propagates through to a similar fractional uncertainty in $\overline{T}_{\rm spin}$. However, for higher redshifts this fractional uncertainty increases rapidly at $z > 2$, to more than 50\,per\,cent at $z = 3$, reflecting the paucity of optical spectroscopic data for the high-redshift radio source population. Understanding how the radio source population is distributed at lower flux densities and at higher redshifts is therefore a concern for the future 21\,cm absorption surveys undertaken with the SKA mid- and low-frequency telescopes (see \citealt{Kanekar:2004} and \citealt*{Morganti:2015} for reviews). \begin{table} \begin{threeparttable} \caption{An account of errors in our estimate of $\overline{T}_{\rm spin}$ due to the accuracy to which we can determine the expected number of absorber detections.}\label{table:tspin_uncertainties} \begin{tabular}{l@{\hspace{0.05in}}l@{\hspace{0.05in}}l@{\hspace{0.05in}}l@{\hspace{0.05in}}l} \hline & Source of error & $\mathrm{err}(\overline{T}_{\rm spin})$ & Refs. \\ & & [per\,cent] & \\ \hline Covering factor & Distribution uncertainty & $\pm10$ & $a$ \\ Covering factor & Systematic evolution & +30 & $a$, $b$\\ $f(N_{\rm HI}, X)$ & Measurement uncertainty & $\pm10$ & $c, d$\\ Low-$z$ $f(N_{\rm HI}, X)$ & Systematic self-absorption & $+(10-30)$ & $e$ \\ High-$z$ $f(N_{\rm HI}, X)$ & Systematic dust-obscuration & $+3$ & $f$, $g$ \\ $\mathcal{F}_{\rm src}(z^{\prime} \geq z)$ & Measurement uncertainty & $\pm 5$ & $h$, $i$ \\ \hline \end{tabular} \begin{tablenotes} \item[] References: $^{a}${\citet{Kanekar:2014a}}, $^{b}${\citet{Curran:2012b}}, $^{c}${\citet{Zwaan:2005}}, $^{d}${\citet{Noterdaeme:2009}} , $^{e}${\citet{Braun:2012}}, $^{f}${\citet{Pontzen:2009}}, $^{g}${\citet{Murphy:2016}}, $^{h}${\citet{Brookes:2008}}, $^{i}${\citet{deZotti:2010}}. \end{tablenotes} \end{threeparttable} \end{table} \section{Expected results for future 21-cm absorption surveys}\label{section:tspin_results} \begin{figure} \centering \includegraphics[width=0.475\textwidth]{tspin_prob.pdf} \caption{The posterior probability density of the average spin temperature, as a function of absorber detection yield ($\mathcal{N}$). We show results for our simulated all-southern-sky survey with 2-h per pointing using the full 36-antenna ASKAP (top panel) and a smaller 1000\,deg$^{2}$ survey with 12-h per pointing and 12 antennas of ASKAP (bottom panel). The dashed curves show the cumulative effect of the systematic errors discussed in \autoref{section:errors}. $\overline{\mathcal{F}}_{\rm CNM}$ is the average CNM fraction assuming a simple two-phase neutral ISM with $T_{\rm spin,CNM} = 100$\,K and $T_{\rm spin,WNM} = 1800$\,K (\citealt{Liszt:2001}).}\label{figure:tspin_prob} \end{figure} In the top panel of \autoref{figure:tspin_prob} we show the results of applying our method for inferring $\overline{T}_{\rm spin}$ to the simulated all-southern-sky \mbox{H\,{\sc i}} absorption survey with ASKAP described in \autoref{section:all_sky_survey}. We account for the uncertainties in the expected detection rate $\overline{\mu}$, discussed in \autoref{section:errors}, by using a Monte Carlo approach and marginalizing over many realizations. A yield of 1000 absorbers from such a survey would imply an average spin temperature of $\overline{T}_\mathrm{spin} = 127^{+14}_{-14}\,(193^{+23}_{-23})$\,K\footnote{We give the 68.3\,per\,cent interval about the median value measured from the posterior distributions shown in \autoref{figure:tspin_prob}.}, where values in parentheses denote the alternative posterior probability resulting from the systematic errors discussed in \autoref{section:errors}. This scenario would indicate that a large fraction of the atomic gas in DLAs at these intermediate redshifts is in the classical stable CNM phase. Conversely, a yield of only 100 detections would imply that $\overline{T}_\mathrm{spin} = 679^{+64}_{-65}\,(1184^{+116}_{-120})$\,K, indicating that less than 10\,per\,cent of the atomic gas is in the CNM and that the bulk of the neutral gas in galaxies is significantly different at intermediate redshifts compared with the local Universe. We also consider the effect of reducing the sky area and array size, which is relevant for planned early science surveys with ASKAP and other SKA pathfinder telescopes. In the bottom panel of \autoref{figure:tspin_prob}, we show the spin temperatures inferred when observing a random 1000\,deg$^{2}$ field for 12\,h per pointing, between $z_{\rm HI} = 0.4$ and $1.0$, using a 12-antenna version of ASKAP. We find that detection yields of 30 and 3 from such a survey would give inferred spin temperatures of $\overline{T}_{\rm spin} =134^{+23}_{-27}\,(209^{+40}_{-47})$ and $848^{+270}_{-430}\,(1535^{+513}_{-837})$\,K, respectively. The significant reduction in telescope sensitivity and sky-area, compensated by the increase in integration time per pointing planned for early-science, results in a factor of 30 decrease in the expected number of detections and therefore an increase in the sample variance and uncertainty in $\overline{T}_{\rm spin}$. However, this result demonstrates that we expect to be able to distinguish between the limiting cases of CNM-rich or deficient DLA populations even during the early-science phases of the SKA pathfinders. For example 30 detections with the early ASKAP survey rules out an average spin temperature of 1000\,K at high probability. \section{Conclusions} We have demonstrated a statistical method for measuring the average spin temperature of the neutral ISM in distant galaxies, using the expected detection yields from future wide-field 21\,cm absorption surveys. The spin temperature is a crucial property of the ISM that can be used to determine the fraction of the cold ($T_{\rm k} \sim 100$\,K) and dense ($n \sim 100$\,cm$^{-2}$) atomic gas that provides sites for the future formation of cold molecular gas clouds and star formation. Recent 21\,cm surveys for \mbox{H\,{\sc i}} absorption in \mbox{Mg\,{\sc ii}} absorbers and DLAs towards distant quasars have yielded some evidence of an evolution in the average spin temperature that might reveal a decrease in the fraction of cold dense atomic gas at high redshift (e.g. \citealt{Gupta:2009, Kanekar:2014a}). By combining recent specifications for ASKAP, with available information for the population of background radio sources, we show that strong statistical constraints (approximately $\pm10$\,per\,cent) in the average spin temperature can be achieved by carrying out a shallow 2-h per pointing survey of the southern sky between redshifts of $z = 0.4$ and $1.0$. However, we find that the accuracy to which we can measure the average spin temperature is ultimately limited by the accuracy to which we can measure the distribution of the covering factor, the $N_{\rm HI}$ frequency distribution function and the evolution of the radio source population as a function of redshift. By improving our understanding of these distributions we will be able to leverage the order-of-magnitude increases in sensitivity and redshift coverage of the future SKA telescope, allowing us to measure the evolution of the average spin temperature to much higher redshifts. \section*{Acknowledgements} We thank Robert Allison, Elaine Sadler and Michael Pracy for useful discussions, and the anonymous referee for providing comments that helped improve this paper. JRA acknowledges support from a Bolton Fellowship. We have made use of \texttt{Astropy}, a community-developed core \texttt{Python} package for astronomy (\citealt{Astropy:2013}); NASA's Astrophysics Data System Bibliographic Services; and the VizieR catalogue access tool, operated at CDS, Strasbourg, France. \bibliographystyle{mnras}
467d37196f1b4f439bbd55a25dd550222fb466db
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Given $\rho>0$, we consider the problem \begin{equation}\label{eq:main_prob_U} \begin{cases} -\Delta U + \lambda U = |U|^{p-1}U & \text{in }\Omega,\smallskip\\ \int_\Omega U^2\,dx = \rho, \quad U=0 & \text{on }\partial\Omega, \end{cases} \end{equation} where $\Omega\subset{\mathbb{R}}^N$ is a Lipschitz, bounded domain, $1<p<2^*-1$, $\rho>0$ is a fixed parameter, and both $U\in H^1_0(\Omega)$ and $\lambda\in{\mathbb{R}}$ are unknown. More precisely, we investigate conditions on $p$ and $\rho$ (and also $\Omega$) for the solvability of the problem. The main interest in \eqref{eq:main_prob_U} relies on the investigation of standing wave solutions for the nonlinear Schr\"odinger equation \[ i\frac{\partial \Phi}{\partial t}+\Delta \Phi+ |\Phi|^{p-1}\Phi=0,\qquad (t,x)\in {\mathbb{R}}\times \Omega \] with Dirichlet boundary conditions on $\partial\Omega$. This equation appears in several different physical models, both in the case $\Omega={\mathbb{R}}^N$ \cite{MR2002047}, and on bounded domains \cite{MR1837207}. In particular, the latter case appears in nonlinear optics and in the theory of Bose-Einstein condensation, also as a limiting case of the equation on ${\mathbb{R}}^N$ with confining potential. When searching for solutions having the wave function $\Phi$ factorized as $\Phi(x,t)=e^{i\lambda t} U(x)$, one obtains that the real valued function $U$ must solve \begin{equation}\label{eq:NLS} -\Delta U + \lambda U = |U|^{p-1}U ,\qquad U\in H^1_0(\Omega), \end{equation} and two points of view are available. The first possibility is to assign the chemical potential $\lambda\in{\mathbb{R}}$, and search for solutions of \eqref{eq:NLS} as critical points of the related action functional. The literature concerning this approach is huge and we do not even make an attempt to summarize it here. On the contrary, we focus on the second possibility, which consists in considering $\lambda$ as part of the unknown and prescribing the mass (or charge) $\|U\|_{L^2(\Omega)}^2$ as a natural additional condition. Up to our knowledge, the only previous paper dealing with this case, in bounded domains, is \cite{MR3318740}, which we describe below. The problem of searching for normalized solutions in ${\mathbb{R}}^N$, with non-homogeneous nonlinearities, is more investigated \cite{MR3009665,MR1430506}, even though the methods used there can not be easily extended to bounded domains, where dilations are not allowed. Very recently, also the case of partial confinement has been considered \cite{BeBoJeVi_2016}. Solutions of \eqref{eq:main_prob_U} can be identified with critical points of the associated energy functional \[ \mathcal{E}(U) = \frac12\int_\Omega|\nabla U|^2\,dx - \frac{1}{p+1} \int_\Omega|U|^{p+1}\,dx \] restricted to the mass constraint \[ {\mathcal{M}}_\rho=\{U\in H_0^1(\Omega) : \|U\|_{L^2(\Omega)}=\rho\}, \] with $\lambda$ playing the role of a Lagrange multiplier. A cricial role in the discussion of the above problem is played by the Gagliardo-Nirenberg inequality: for any $\Omega$ and for any $v\in H^1_0(\Omega)$, \begin{equation} \label{sobest} \|v\|^{p+1}_{L^{p+1}(\Omega)} \leq C_{N,p} \| \nabla v \|_{L^2(\Omega)}^{N(p-1)/2} \| v \|_{L^2(\Omega)} ^{(p+1)-N(p-1)/2}, \end{equation} the equality holding only when $\Omega={\mathbb{R}}^N$ and $v=Z_{N,p}$, the positive solution of $-\Delta Z + Z = Z^{p}$ (which is unique up to translations \cite{MR969899}). Accordingly, the exponent $p$ can be classified in relation with the so called \emph{$L^2$-critical exponent} $1+4/N$ (throughout all the paper, $p$ will be always Sobolev-subcritical and its criticality will be understood in the $L^2$ sense). Indeed we have that ${\mathcal{E}}$ is bounded below and coercive on ${\mathcal{M}}_\rho$ if and only if either $p$ is subcritical, or it is critical and $\rho$ is sufficiently small. The recent paper \cite{MR3318740} deals with problem \eqref{eq:main_prob_U} in the case of the spherical domain $\Omega = B_1$, when searching for positive solutions $U$. In particular, it is shown that the solvability of \eqref{eq:main_prob_U} is strongly influenced by the exponent $p$, indeed: \begin{itemize} \item in the subcritical case $1<p<1+4/N$, \eqref{eq:main_prob_U} admits a unique positive solution for every $\rho>0$; \item if $p=1+4/N$ then \eqref{eq:main_prob_U} admits a unique positive solution for \[ 0<\rho<\rho^*=\left(\frac{p+1}{2C_{N,p}}\right)^{N/2}=\|Z_{N,p}\|^2_{L^2({\mathbb{R}}^N)}, \] and no positive solutions for $\rho\geq\rho^*$; \item finally, in the supercritical regime $1+4/N<p<2^*-1$, \eqref{eq:main_prob_U} admits positive solutions if and only if $0<\rho\leq\rho^*$ (the threshold $\rho^*$ depending on $p$), and such solutions are at least two for $\rho<\rho^*$. \end{itemize} In this paper we carry on such analysis, dealing with a general domain $\Omega$ and with solutions which are not necessarily positive. More precisely, let us recall that for any $U$ solving \eqref{eq:main_prob_U} for some $\lambda$, it is well-defined the Morse index \[ m(U) = \max\left\{k : \begin{array}{l} \exists V\subset H^1_0(\Omega),\,\dim(V)= k:\forall v\in V\setminus\{0\}\smallskip\\ \displaystyle\int_\Omega |\nabla v|^2 + \lambda v^2 - p|U|^{p-1}v^2\,dx<0 \end{array} \right\}\in{\mathbb{N}}. \] Then, if $\Omega=B_1$, it is well known that a solution $U$ of \eqref{eq:main_prob_U} is positive if and only if $m(U)=1$. Under this perspective, the results in \cite{MR3318740} can be read in terms of Morse index one--solutions, rather than positive ones: introducing the sets of admissible masses \[ {\mathfrak{A}}_k ={\mathfrak{A}}_k(p,\Omega) := \left\{\rho>0 : \begin{array}{l} \eqref{eq:main_prob_U} \text{ admits a solution $U$ (for some $\lambda$)}\\ \text{having Morse index }m(U)\leq k \end{array} \right\}, \] then \cite{MR3318740} implies that ${\mathfrak{A}}_1(p,B_1)$ is a bounded interval if and only if $p$ is critical or supercritical, while ${\mathfrak{A}}_1(p,B_1)={\mathbb{R}}^+$ in the subcritical case. On the contrary, when considering general domains and higher Morse index, the situation may become much more complicated. We collect some examples in the following remark. \begin{remark}\label{rem:specialdomains} In the case of a symmetric domain, one can use any solution as a building block to construct other solutions with a more complex behavior, obtaining the so-called necklace solitary waves. Such kind of solutions are constructed in \cite{MR3426917}, even though in such paper the focus is on stability, rather than on normalization conditions. For instance, by scaling argument, any Dirichlet solution of $-\Delta U + \lambda U = |U|^{p-1}U$ in a rectangle $R=\prod_{i=1}^N(a_i,b_i)$ can be scaled to a solution of $-\Delta U + k^2\lambda U = |U|^{p-1}U$ in $R/k$, $k\in{\mathbb{N}}_+$, and then $k^N$ copies of it can be juxtaposed, with alternating sign. In this way one obtains a new solution on $R$ having $k^{4/(p-1)}$ times the mass of the starting one, and eventually solutions in $R$ with arbitrarily high mass (but with higher Morse index) can be constructed even in the critical and supercritical case. An analogous construction can be performed in the disk, using solutions in circular sectors as building blocks, even though in this case explicit bounds on the mass obtained are more delicate. Also, instead of symmetric domains, singular perturbed ones can be considered, such as dumbbell domains \cite{MR949628}: for instance, using \cite[Theorem 3.5]{MR2997381}, one can show that for any $k$, there exists a domain $\Omega$, which is close in a suitable sense to the disjoint union of $k$ domains, such that \eqref{eq:main_prob_U} has a \emph{positive} solution on $\Omega$ with Morse index $k$ and $\rho=\rho_k\to+\infty$ as $k\to+\infty$. This kind of results justifies the choice of classifying the solutions in terms of their Morse index, rather than in terms of their nodal properties. \end{remark} Motivated by the previous remark, the first question we address in this paper concerns the boundedness of ${\mathfrak{A}}_k$. We provide the following complete classification. \begin{theorem}\label{thm:bbd_index} For every $\Omega\subset{\mathbb{R}}^N$ bounded $C^1$ domain, $k\ge1$, $1<p<2^*-1$, \[ \sup{\mathfrak{A}}_k(p,\Omega) < +\infty \qquad\iff\qquad p \ge 1+\frac{4}{N}. \] \end{theorem} The proof of such result, which is outlined in Section \ref{sec:blow-up}, is obtained by a detailed blow-up analysis of sequences of solutions with bounded Morse index, via suitable a priori pointwise estimates (see \cite{MR2063399}). In this respect, the regularity assumption on $\partial\Omega$ simplifies the treatment of possible concentration phenomena towards the boundary. The argument, which holds for solutions which possibly change sign, is inspired by \cite{MR2825606}, where the case of positive solutions is treated. Once Theorem \ref{thm:bbd_index} is established, in case $p\geq 1 + 4/N$ two questions arise, namely: \begin{enumerate} \item is it possible to provide lower bounds for $\sup{\mathfrak{A}}_k$? Is it true that $\sup{\mathfrak{A}}_k$ is strictly increasing in $k$, or, at least, that $\sup{\mathfrak{A}}_k > \sup{\mathfrak{A}}_1$ for some $k$? \item is \eqref{eq:main_prob_U} solvable for every $\rho\in(0,\sup{\mathfrak{A}}_k)$, or at least can we characterize some subinterval of solvability? \end{enumerate} It is clear that both issues can be addressed by characterizing values of $\rho$ for which existence (and multiplicity) of solutions with bounded Morse index can be guaranteed. To this aim, it can be useful to restate problem \eqref{eq:main_prob_U} as \begin{equation}\label{eq:main_prob_u} \begin{cases} -\Delta u + \lambda u = \mu|u|^{p-1}u & \text{in }\Omega,\\ \int_\Omega u^2\,dx = 1, \quad u=0 & \text{on }\partial\Omega, \end{cases} \qquad\text{where}\quad \begin{cases} U=\sqrt{\rho} u\\ \mu = \rho^{(p-1)/2}, \end{cases} \end{equation} where now $\mu>0$ is prescribed. Since \begin{equation} \label{Emu} \text{both } \mathcal{E}_{\mu}(u) := \frac{1}{2}\int_{\Omega}|\nabla u|^2- \frac{\mu}{p+1}\int_{\Omega}| u|^{p+1} \qquad \text{and }{\mathcal{M}}={\mathcal{M}}_1=\{u : \|u\|_{L^2(\Omega)}=1\} \end{equation} are invariant under the ${\mathbb{Z}}_2$-action of the involution $u\mapsto -u$, solutions of \eqref{eq:main_prob_u} can be found via min-max principles in the framework of index theories (see e.g. \cite[Ch. II.5]{St_2008}). Notice that in the supercritical case ${\mathcal{E}}_\mu$ is not bounded from below on ${\mathcal{M}}$. Following \cite{MR3318740}, it can be convenient to parameterize solutions to \eqref{eq:main_prob_u} with respect to the $H^1_0$-norm, therefore we introduce the sets \begin{equation}\label{eq:defBU} \mathcal{B}_\alpha:=\left\{u\in {\mathcal{M}}:\,\int_\Omega |\nabla u|^2\,dx<\alpha\right\},\quad\quad \mathcal{U}_\alpha:=\left\{u\in {\mathcal{M}}:\,\int_\Omega |\nabla u|^2\,dx=\alpha\right\}. \end{equation} Introducing the first Dirichlet eigenvalue of $-\Delta$ in $H^1_0(\Omega)$, $\lambda_1(\Omega)$, we have that the sets above are non-empty whenever $\alpha> \lambda_1(\Omega)$. Since we are interested in critical points having Morse index bounded from above, following \cite{MR968487,MR954951,MR991264} we introduce the following notion of genus. \begin{definition}\label{def:genus} Let $A\subset H^1_0(\Omega)$ be a closed set, symmetric with respect to the origin (i.e. $-A=A$). We define the \emph{genus} $\gamma$ of a $A$ as \[ \gamma(A) := \sup\{m : \exists h \in C({\mathbb{S}}^{m-1};A),\, h(-u)=-h(u)\}. \] Furthermore, we define \[ \Sigma_{\alpha}=\{A\subset \overline{\mathcal{B}}_\alpha: A\text{ is closed and }-A=A\}, \qquad \Sigma^{(k)}_{\alpha}=\{A\in \Sigma_{\alpha} : \gamma(A)\ge k\}, \] \end{definition} We remark that this notion of genus is different from the classical one of \emph{Krasnoselskii genus}, which is well suited for estimates of the Morse index from below, rather than above. Nonetheless, $\gamma$ shares with the Krasnoselskii genus most of the main properties of an index \cite{MR0163310,MR0065910}. In particular, by the Borsuk-Ulam Theorem, any set $A$ homeomorphic to the sphere ${\mathbb{S}}^{m-1} := \partial B_1 \subset {\mathbb{R}}^m$ has genus $\gamma(A) = m$. Furthermore, we show in Section \ref{sec:2const} that $\Sigma^{(k)}_{\alpha}$ is not empty, provided $\alpha>\lambda_k(\Omega)$ (the $k$-th Dirichlet eigenvalue of $-\Delta$ in $H^1_0(\Omega)$). Equipped with this notion of genus we provide two different variational principles for solutions of \eqref{eq:main_prob_u} (and thus of \eqref{eq:main_prob_U}). The first one is based on a variational problem with \emph{two constraints}, which was exploited as the main tool in proving the results in \cite{MR3318740}. \begin{theorem}\label{thm:genus_2constr} Let $k\geq1$ and $\alpha>\lambda_{k}(\Omega)$. Then \begin{equation} \label{maxmin} M_{\alpha,\,k}:= \sup_{A\in\Sigma^{(k)}_{\alpha}}\inf_{u\in A}\int_{\Omega}|u|^{p+1} \end{equation} is achieved on ${\mathcal{U}}_\alpha$, and there exists a critical point $u_\alpha\in {\mathcal{M}}$ such that, for some $\lambda_\alpha\in{\mathbb{R}}$ and $\mu_\alpha>0$, \begin{equation} \label{lagreq} \int_\Omega|\nabla u_\alpha|^2 = \alpha\qquad\text{and}\qquad -\Delta u_\alpha+\lambda_\alpha\,u_\alpha=\mu_\alpha |u_\alpha|^{p-1}u_\alpha\quad \text{in }\Omega. \end{equation} \end{theorem} As a matter of fact, the results in \cite{MR3318740} were obtained by a detailed analysis of the map $\alpha \mapsto \mu_\alpha$ in the case $k=1$, i.e. when dealing with \[ M_{\alpha,1} = \max\left\{\|u\|_{L^{p+1}}^{p+1} : \|u\|_{L^2}^2=1,\,\|\nabla u\|_{L^2}^2=\alpha \right\}. \] In the present paper we do not investigate the properties of the map $\alpha \mapsto \mu_\alpha$ for general $k$, but we rather prefer to exploit the characterization of $M_{\alpha,k}$ in connection with a second variational principle, which deals with only \emph{one constraint}. \begin{theorem}\label{thm:genus_1constr} Let $1+{N}/{4}\leq p<2^*-1$. There exists a sequence $(\hat \mu_k)_k$ (depending on $\Omega$ and $p$) such that, for every $k\geq 1$ and $0<\mu<\hat \mu_k$, the value \begin{equation} \label{infsuplev} c_k:= \inf_{A\in\Sigma^{(k)}_{\alpha}} \sup_{A}{\mathcal{E}}_\mu, \end{equation} is achieved in $\mathcal{B}_\alpha$, for a suitable $\alpha>\lambda_{k}(\Omega)$. Furthermore there exists a critical point $u_\mu\in {\mathcal{M}}$ such that, for some $\lambda_\mu\in{\mathbb{R}}$, \[ -\Delta u_\mu+\lambda_\mu\,u_\mu=\mu |u_\mu|^{p-1}u_\mu\quad \text{in }\Omega, \] $\|\nabla u\|_{L^2}^2<\alpha$, and $m(u_\mu)\le k$. \end{theorem} \begin{remark} Of course, if $p<1+4/N$, the above theorem holds with $\hat\mu_k=+\infty$ for every $k$. \end{remark} \begin{corollary} Let $\hat \rho_k := \hat \mu_k^{2/(p-1)}$. Then \[ (0,\hat\rho_k) \subset {\mathfrak{A}}_k. \] \end{corollary} The link between Theorem \ref{thm:genus_2constr} and Theorem \ref{thm:genus_1constr} is that we can provide explicit estimates of $\hat \mu_k$ (and hence of $\hat\rho_k$) in terms of the map $\alpha\mapsto M_{\alpha,k}$ (see Section \ref{sec:1const}). We stress that the above results hold for any Lipschitz $\Omega$. As a first consequence, this allows to extend the existence result in \cite{MR3318740} to non-radial domains. \begin{theorem}\label{thm:intro_GS} For every $0<\rho<\hat\rho_1=\hat\rho_1(\Omega,p)$ problem \eqref{eq:main_prob_U} admits a solution which is a local minimum of the energy ${\mathcal{E}}$ on ${\mathcal{M}}_\rho$. In particular, $U$ is positive, has Morse index one and the associated solitary wave is orbitally stable. Furthermore, for every Lipschitz $\Omega$, \begin{itemize} \item $\displaystyle 1<p<1+\frac{4}{N} \implies \hat\rho_1\left(\Omega,p\right) = +\infty$, \item $\displaystyle p=1+\frac{4}{N} \implies \hat\rho_1\left(\Omega,p\right) \geq \|Z_{N,p}\|^2_{L^2({\mathbb{R}}^N)}$, \item $\displaystyle 1+\frac{4}{N}<p<2^*-1 \implies \hat\rho_1\left(\Omega,p\right) \geq D_{N,p} \lambda_1(\Omega)^{\frac{2}{p-1}-\frac{N}{2}}$, \end{itemize} where the universal constant $D_{N,p}$ is explicitly written in terms of $N$ and $p$ in Section \ref{sec:1const}. \end{theorem} \begin{remark}\label{rem:introGS} Of course, in the subcritical and critical cases, $c_1$ is actually a global minimum. Furthermore, the lower bound for the supercritical case agrees with that of the critical one since, as shown in Section \ref{sec:1const}, $D_{N,1+4/N} = \|Z_{N,p}\|^2_{L^2({\mathbb{R}}^N)}$ (and $\lambda_1(\Omega)$ is raised to the $0^{\text{th}}$-power). Notice that the estimate for the supercritical case is new also in the case $\Omega=B_1$. \end{remark} We observe that the exponent of $\lambda_1(\Omega)$ in the supercritical threshold is negative, therefore such threshold decreases with the size of $\Omega$. Once the first thresholds have been estimated, we turn to the higher ones: by exploiting the relations between $M_{\alpha,k}$ and $c_k$, we can show that the thresholds obtained for Morse index one--solutions in Theorem \ref{thm:intro_GS} can be increased, by considering higher Morse index--solutions, at least for some exponent. \begin{proposition}\label{thm:intro_3>1} For every $\Omega$ and $1<p<2^*-1$, \[ \hat\rho_3\left(\Omega,p\right) \geq 2 \cdot D_{N,p} \lambda_3(\Omega)^{\frac{2}{p-1}-\frac{N}{2}}. \] \end{proposition} \begin{remark} In the critical case, the lower bound for $\hat\rho_3$ provided by Proposition \ref{thm:intro_3>1} is twice that for $\hat\rho_1$ obtained in Theorem \ref{thm:intro_GS}. By continuity, the estimate for $\hat\rho_3$ is larger than that for $\hat\rho_1$ also when $p$ is supercritical, but not too large. To quantify such assertion, we can use Yang's inequality \cite{MR1894540,MR2262780}, which implies that for every $\Omega$ it holds \[ \lambda_3(\Omega)\leq \left(1+\frac{N}{4}\right)2^{2/N} \lambda_1(\Omega). \] We deduce that $2 \cdot D_{N,p} \lambda_3(\Omega)^{\frac{2}{p-1}-\frac{N}{2}} \geq D_{N,p} \lambda_1(\Omega)^{\frac{2}{p-1}-\frac{N}{2}}$ whenever \[ p\leq 1+\frac{4}{N} + \frac{8}{N^2\log_2\left(1+\frac{4}{N}\right)}. \] In particular, the physically relevant case $N=3$, $p=3$ is covered. Furthermore, if $N\geq 7$, the above condition holds for every $p<2^*-1$. \end{remark} Beyond existence results for \eqref{eq:main_prob_U}, also multiplicity results can be achieved. A first general consideration, with this respect, is that Theorem \ref{thm:genus_1constr} holds true also when using the standard Krasnoselskii genus instead of $\gamma$; this allows to obtain critical points having Morse index bounded from below (see \cite{MR968487,MR954951,MR991264}), and therefore to obtain infinitely many solutions, at least when $\rho$ is less than some threshold. More specifically, we can also prove the existence of a second solution in the supercritical case, thus extending to any $\Omega$ the multiplicity result obtained in \cite{MR3318740} for the ball. Indeed, on the one hand, in the supercritical case ${\mathcal{E}}_\mu$ is unbounded from below; on the other hand the solution obtained in Theorem \ref{thm:genus_1constr}, for $k=1$, is a local minimum. Thus the Mountain Pass Theorem \cite{MR0370183} applies on $\mathcal{M}$, and a second solution can be found for $\mu<\hat\mu_1$, see Proposition \ref{mpcritlev} for further details (and also Remark \ref{rem:further_crit_lev} for an analogous construction for $k\ge2$). To conclude this introduction, let us mention that the explicit lower bounds obtained in Theorem \ref{thm:intro_GS} can be easily applied in order to gain much more information also in the case of special domains, as those considered in Remark \ref{rem:specialdomains}. For instance, we can prove then following. \begin{theorem}\label{pro:symm} Let $\Omega=B$ be a ball in ${\mathbb{R}}^N$. Then \[ p<1+\frac{4}{N-1} \quad\implies\quad \text{\eqref{eq:main_prob_U} admits a solution for every }\rho>0. \] An analogous result holds when $\Omega=R$ is a rectangle, without further restrictions on $p<2^*-1$. \end{theorem} Therefore our starting problem in $\Omega=B$ can be solved for any mass value also in the critical and supercritical regime, at least for $p$ smaller than this further critical exponent $1+4/(N-1) > 1+ 4/N$. Of course, higher masses require higher Morse index--solutions. In particular, since by \cite{MR3318740} we know that ${\mathfrak{A}}_1(B,1+4/N) = (0,\|Z_{N,p}\|_{L^2})$, we have that for larger masses, even though no positive solution exists, nodal solutions with higher Morse index can be obtained: in such cases \eqref{eq:main_prob_U} admits \emph{nodal ground states with higher Morse index}. The paper is structured as follows: in Section \ref{sec:blow-up} we perform a blow-up analysis of solutions with bounded Morse index, in order to prove Theorem \ref{thm:bbd_index}; Section \ref{sec:2const} is devoted to the analysis of the variational problem with two constraints \eqref{maxmin} and to the proof of Theorem \ref{thm:genus_2constr}; that of Theorems \ref{thm:genus_1constr}, \ref{thm:intro_GS} and Proposition \ref{thm:intro_3>1} is developed in Section \ref{sec:1const}, by means of the variational problem with one constraint \eqref{infsuplev}; finally, Section \ref{sec:symm} contains the proof of Theorem \ref{pro:symm}. \textbf{Notation.} We use the standard notation $\{\varphi_k\}_{k\geq1}$ for a basis of eigenfunctions of the Dirichlet laplacian in $\Omega$, orthogonal in $H^1_0(\Omega)$ and orthonormal in $L^2(\Omega)$. Such functions are ordered in such a way that the corresponding eigenvalues $\lambda_k(\Omega)$ satisfy \[ 0<\lambda_1(\Omega)<\lambda_2(\Omega)\leq\lambda_3(\Omega)\leq\dots, \] and $\varphi_1$ is chosen to be positive on $\Omega$. $C_{N,p}$ denotes the universal constant in the Gagliardo-Nirenberg inequality \eqref{sobest}, which is achieved (uniquely, up to translations and dilations) by the positive, radially symmetric function $Z_{N,p}\in H^1({\mathbb{R}}^N)$, with \[ \|Z_{N,p}\|^2_{L^2({\mathbb{R}}^N)}=\left(\frac{p+1}{2C_{N,p}}\right)^{N/2}. \] Finally, $C$ denotes every (positive) constant we need not to specify, whose value may change also within the same formula. \section{Blow-up analysis of solutions with bounded Morse index}\label{sec:blow-up} Throughout this section we will deal with a sequence $\{(u_n,\mu_n,\lambda_n)\}_n \subset H^1_0(\Omega)\times{\mathbb{R}}^+\times{\mathbb{R}}$ satisfying \begin{equation}\label{eq:auxiliary_n} -\Delta u_n+\lambda_n u_n=\mu_n |u_n|^{p-1}u_n,\qquad\int_\Omega u_n^2\, dx=1,\qquad \int_\Omega |\nabla u_n|^2\, dx=:\alpha_n. \end{equation} To start with, we recall the following result (actually, in \cite{MR3318740}, the result is stated for positive solution, but the proof does not require such assumption). \begin{lemma}[{\cite[Lemma 2.5]{MR3318740}}]\label{lemma:case_alpha_n_bounded} Take a sequence $\{(u_n,\mu_n,\lambda_n)\}_n$ as in \eqref{eq:auxiliary_n}. Then \[ \{\alpha_n\}_n \text{ bounded} \qquad\implies\qquad \{\lambda_n\}_n,\,\{\mu_n\}_n\text{ bounded}. \] \end{lemma} Next we turn to the study of sequences having arbitrarily large $H^1_0$-norm. In particular, we will focus on sequences of solutions having a common upper bound on the Morse index \[ m(u_n) = \max\left\{k : \begin{array}{l} \exists V\subset H^1_0(\Omega),\,\dim(V)= k:\forall v\in V\setminus\{0\}\smallskip\\ \displaystyle\int_\Omega |\nabla v|^2 + \lambda_n v^2 - p\mu_n|u_n|^{p-1}v^2\,dx<0 \end{array} \right\}. \] Throughout this section we will assume that \begin{equation}\label{eq:mainass_secMorse} \text{the sequence }\{(u_n,\mu_n,\lambda_n)\}_n\text{ satisfies \eqref{eq:auxiliary_n}, with }\alpha_n\to+\infty\text{ and }m(u_n)\leq \bar k, \end{equation} for some $\bar k\in{\mathbb{N}}$ not depending on $n$. \begin{lemma}\label{lem:lambda_bdd_below} Let \eqref{eq:mainass_secMorse} hold. Then \( \lambda_n \geq -\lambda_{\bar k}(\Omega). \) \end{lemma} \begin{proof} Assume, to the contrary, that for some $n$ it holds $\lambda_n < -\lambda_{\bar k}(\Omega)$. For any real $t_1,\dots t_{\bar k}$ we define \[ \phi := \sum_{h=1}^{\bar k} t_h \varphi_h. \] By denoting $J_{\lambda,\mu}(u)={\mathcal{E}}_\mu(u)+\frac{\lambda}{2}\|u\|_{L^2}^2$, so that Morse index properties can be written in terms of $J''_{\lambda,\mu}$, we have \[ \begin{split} J''_{\lambda_n,\mu_n}(u_n)[u_n,\phi] &= -(p-1)\mu_n\int_\Omega |u_n|^{p-1}u_n\phi,\\ J''_{\lambda_n,\mu_n}(u_n)[\phi,\phi] &= \sum_{h=1}^{\bar k} t_h^2 \int_\Omega \bigl (|\nabla \varphi_h| + \lambda_n \varphi_h^2\bigr )\,dx - p\mu_n\int_\Omega |u_n|^{p-1}\phi^2\,dx \\ &\leq \sum_{h=1}^{\bar k} t_h^2(\lambda_{h}(\Omega) + \lambda_n) - (p-1)\mu_n\int_\Omega |u_n|^{p-1}\phi^2\,dx \leq - (p-1)\mu_n\int_\Omega |u_n|^{p-1}\phi^2\,dx, \end{split} \] where equality holds if and only if $t_1=\dots=t_{\bar k}=0$. As a consequence \begin{multline*} J''_{\lambda_n,\mu_n}(u_n)[t_0 u_n+ \phi, t_0 u_n + \phi] \leq -t_0^2 (p-1)\mu_n\int_\Omega |u_n|^{p-1}u_n^2 \\ - 2t_0(p-1)\mu_n\int_\Omega |u_n|^{p-1}u_n\phi \,dx - (p-1)\mu_n\int_\Omega |u_n|^{p-1}\phi^2\,dx. \end{multline*} We deduce that $J''_{\lambda_n,\mu_n}(u_n)$ is negative definite on $\spann\{u_n, \varphi_1,\dots,\varphi_{\bar k}\}$, in contradiction with the bound on the Morse index (note that $u_n$ cannot be a linear combination of a finite number of eigenfunctions, otherwise using the equations we would obtain that such eigenfunctions are linearly dependent). \end{proof} \begin{lemma} \label{localblow} Let \eqref{eq:mainass_secMorse} hold. Then $\lambda_n\to +\infty$. \end{lemma} \begin{proof} By Lemma \ref{lem:lambda_bdd_below} we have that $\lambda_n$ is bounded below. As a consequence, we can use H\"{o}lder inequality with $\|u_n\|_{L^2}=1$ and \eqref{eq:auxiliary_n} to write \[ \mu_n\,\|u_n\|^{p-1}_{L^{\infty}}\ge \mu_n \,\|u_n\|^{p+1}_{L^{p+1}}=\alpha_n+\lambda_n \rightarrow +\infty. \] Let us define \begin{equation} \label{equn} U_n: =\mu_n^{\frac{1}{p-1}}\,u_n, \quad\text{ so that } -\Delta U_n+\lambda_n U_n=|U_n|^{p-1}U_n\quad \text{in}\,\,\Omega,\quad U|_{\partial\Omega}=0. \end{equation} Pick $P_n\in\Omega$ such that $|U_n(P_n)|=\|U_n\|_{L^{\infty}(\Omega)}$ and set \begin{equation} \label{tildepsn} \tilde\varepsilon_n: =|U_n(P_n)|^{-\frac{p-1}{2}}=\frac{1}{\sqrt{\mu_n\,\|u_n\|^{p-1}_{L^{\infty}}}}\longrightarrow 0 \end{equation} Hence, $|U_n(P_n)|\to+\infty$; moreover, as $P_n$ is a point of positive maximum or of negative minimum, we have \begin{equation} \nonumber 0\le \frac{-\Delta U_n(P_n)}{U_n(P_n)}=|U_n(P_n)|^{p-1}-\lambda_n\,. \end{equation} Thus $\lambda_n|U_n(P_n)|^{1-p}\le 1$, and since $\lambda_n$ is bounded from below, we conclude \begin{equation} \label{limtildelam} \frac{\lambda_n}{|U_n(P_n)|^{p-1}}\longrightarrow \tilde\lambda\in [0,1]. \end{equation} Now, we are left to prove that $\tilde\lambda>0$. Let us define \begin{equation} \label{tildeVn} \tilde V_n(y)=\tilde\varepsilon_n^{\frac{2}{p-1}}\, U_n(\tilde\varepsilon_n\,y+P_n),\quad\quad y\in \tilde\Omega_n :=\big (\Omega-P_n\big )/\tilde\varepsilon_n, \end{equation} and let $d_n := d(P_n,\partial\Omega)$; we have, up to subsequences, \[ \frac{\tilde\varepsilon_n}{d_n}\longrightarrow L\in [0,+\infty] \qquad\text{and}\qquad \tilde\Omega_n\rightarrow\left\{ \begin{array}{ll} {\mathbb{R}}^n, & \text{if $L=0$;} \\ H, & \text{if $L>0$,} \end{array} \right. \] where $H$ is a half-space such that $0\in \overline H$ and $d(0,\partial H)=1/L$. The function $\tilde V_n$ satisfies \begin{equation} \nonumber \left\{ \begin{array}{ll} -\Delta \tilde V_n+\lambda_n\,\tilde \varepsilon_n^2\,\tilde V_n=|\tilde V_n|^{p-1}\tilde V_n, & \hbox{in}\,\, \tilde\Omega_n;\\ |\tilde V_n|\le |\tilde V_n(0)|=1, & \hbox{in}\,\, \tilde\Omega_n;\\ \tilde V_n=0, & \hbox{on}\,\, \partial\tilde \Omega_n. \end{array} \right. \end{equation} From \eqref{tildepsn} and \eqref{limtildelam} we get $\tilde\varepsilon_n^2\,\lambda_n\rightarrow \tilde\lambda$; hence, by elliptic regularity and up to a further subsequence, $\tilde V_n\rightarrow \tilde V$ in $\mathcal{C}^1_{{\mathrm{loc}}}(\overline H)$ where $\tilde V$ solves \begin{equation} \label{limprob1} \left\{ \begin{array}{ll} -\Delta \tilde V+\tilde\lambda\,\tilde V=|\tilde V|^{p-1}\tilde V, & \hbox{in}\,\, H;\\ |\tilde V|\le |\tilde V(0)|=1, & \hbox{in}\,\, H;\\ \tilde V=0, & \hbox{on}\,\, \partial H. \end{array} \right. \end{equation} Since $\sup_n m(U_n)\leq \bar k$ (as a solution to \eqref{equn}), one can show as in Theorem $3.1$ of \cite{MR2825606} that $m(\tilde V)\leq \bar k$. In particular, $\tilde V$ is stable outside a compact set (see Definition $2.1$ in \cite{MR2825606}) so that, by Theorem $2.3$ and Remark $2.4$ of \cite{MR2825606}, we have $$\tilde V(x)\rightarrow 0 \quad\quad \text{as} \quad\quad |x|\rightarrow +\infty.$$ Moreover, since $\tilde V$ is not trivial, we also have that $\tilde\lambda>0$. For, if $\tilde\lambda=0$ the function $\tilde V$ would be a solution of the Lane-Emden equation $-\Delta u=|u|^{p-1}u$ either in ${\mathbb{R}}^n$ or in $H$. In both cases, $\tilde V$ would contradict Theorems $2$ and $9$ of \cite{MR2322150}, being non trivial and stable outside a compact set. Thus, $\tilde\lambda >0$ and by \eqref{limtildelam} we conclude $\lambda_n\rightarrow +\infty$. \end{proof} \begin{remark}\label{rem4} We stress that the scaling argument in Lemma \ref{localblow}, leading to the limit problem \eqref{limprob1} (with $\tilde\lambda>0$), can be repeated also near points of \emph{local} extremum. More precisely, let $Q_n$ be such that $|U_n(Q_n)|\to +\infty$ and $$ |U_n(Q_n)|=\max_{\Omega\cap B_{R_n\tilde\varepsilon_n}(Q_n)}U_n, $$ for some $R_n\to +\infty$. Then the above procedure can be repeated by replacing $P_n$ with $Q_n$ in definition \eqref{tildepsn}. \end{remark} The local description of the asymptotic behaviour of the solutions $U_n$ to \eqref{equn} with bounded Morse index can be carried out more conveniently by defining the sequence (see \cite[Theorem $3.1$]{MR2825606}) \begin{equation} \label{defVn} V_n(y)=\varepsilon_n^{\frac{2}{p-1}}\, U_n(\varepsilon_n\,y+P_n),\quad y\in \Omega_n :=\frac{\Omega-P_n}{\varepsilon_n}, \end{equation} where $P_n$ is defined before \eqref{tildepsn}, and $\varepsilon_n=\frac{1}{\sqrt{\lambda_n}}\to 0$. Then, $V_n$ satisfies \begin{equation} \nonumber \left\{ \begin{array}{ll} -\Delta V_n+ V_n=| V_n|^{p-1} V_n, & \hbox{in}\,\,\Omega_n;\\ |V_n|\le | V_n(0)|=\big ({\varepsilon_n/\tilde\varepsilon_n}\big )^{\frac{2}{p-1}}\rightarrow \tilde\lambda^{-\frac{1}{p-1}}, & \hbox{in}\,\, \Omega_n;\\ V_n=0, & \hbox{on}\,\, \partial\Omega_n. \end{array} \right. \end{equation} As before, we have (up to a subsequence) $V_n\rightarrow V$ in $\mathcal{C}^1_{\mathrm{loc}}(\overline H)$ where $H$ is either ${\mathbb{R}}^N$ or a half space and $V$ solves \begin{equation} \label{limprob2} \left\{ \begin{array}{ll} -\Delta V+ V=| V|^{p-1} V, & \hbox{in}\,\, H;\\ |V|\le | V(0)|=\tilde\lambda^{-\frac{1}{p-1}}, & \hbox{in}\,\, H;\\ V=0, & \hbox{on}\,\, \partial H. \end{array} \right. \end{equation} By recalling the discussion following \eqref{limprob1} we also have $m(V)<+\infty$. We collect some well known property of such a $V$ in the following result. \begin{theorem}[\cite{MR688279,MR2825606,MR2322150,MR2785899}]\label{thm:unif_est_Farina} Let $V$ be a classical solution to \eqref{limprob2} such that $m(V)\leq\bar k$. Then: \begin{enumerate} \item $H={\mathbb{R}}^N$; \item $V(x)\to 0$ as $|x|\rightarrow +\infty$, $V \in H^1({\mathbb{R}}^N)\cap L^{p+1}({\mathbb{R}}^N)$; \item there exist $C$ only depending on $\bar k$ (and not on $V$) such that \[ \|V\|_{L^{\infty}} + \|\nabla V\|_{L^{\infty}}<C. \] \end{enumerate} \end{theorem} \begin{proof} Claim 2 follows from Theorem 2.3 and Remark 2.4 of \cite{MR2825606}, see also \cite[Remark 1.4]{MR688279}. As a consequence, Theorem 1.1 of \cite[Remark 1.4]{MR688279} readily applies, providing claim 1 ($V$ is not trivial as $V(0)>0$). On the other hand, the $L^\infty$ estimates in claim 3. are proved in Theorem 1.9 of \cite{MR2785899}. \end{proof} \begin{corollary} \label{distpnbound} If the sequence $\{U_n\}$ of solutions to \eqref{equn} has uniformly bounded Morse index, and if $P_n\in\Omega$ is such that $|U_n(P_n)|=\|U_n\|_{L^{\infty}(\Omega)}\to+\infty$, then $$ \sqrt{\lambda_n}\,d(P_n,\partial\Omega)\rightarrow +\infty,\qquad\text{where }\frac{\lambda_n}{|U_n(P_n)|^{p-1}}\to\tilde\lambda\in(0,1]. $$ \end{corollary} \begin{remark} Recall that $Z_{N,p}$, the unique positive solution to $-\Delta u+ u=| u|^{p-1} u$ in ${\mathbb{R}}^N$, has Morse index $1$ \cite{MR969899}; then, if $V$ solves \eqref{limprob2} in ${\mathbb{R}}^N$ and $1<m(V)<+\infty$, then $V$ is necessarily sign-changing. \end{remark} Following the same pattern as in \cite{MR2825606}, we now analyze the global behaviour of a sequence $\{U_n\}$ of solutions to \eqref{equn} for $\lambda_n\to +\infty$, assuming that \( \lim_{n\to +\infty} m(U_n)\leq\bar k<\infty. \) By the previous discussion, if $P^1_n$ is a sequence of points such that $|U_n(P^1_n)|=\|U_n\|_{L^{\infty}(\Omega)}$, we have $|U_n(P^1_n)|\rightarrow +\infty$ and ${\lambda_n}\,d(P^1_n,\partial\Omega)^2\rightarrow +\infty$. We now look for other possible sequences of (local) extremum points $P^i_n$, $i=2,3,..$, along which $|U_n|$ goes to infinity. For any $R>0$, consider the quantity \begin{equation} \nonumber h_1(R)=\limsup_{n\to +\infty} \Bigl (\lambda_n^{-\frac{1}{p-1}}\max_{|x-P^1_n|\ge R\,\lambda_n^{-1/2}} |U_n(x)| \Bigr ). \end{equation} We will prove that if $h_1(R)$ is \emph{not vanishing} for large $R$, then there exists a 'blow-up' sequence $P^2_n$ for $u_n$, 'disjoint' from $P^1_n$. Indeed, let us suppose that \begin{equation} \nonumber \limsup_{R\to +\infty} h_1(R)=4\delta>0. \end{equation} Hence, up to a subsequence and for arbitrarily large $R$, we have \begin{equation} \label{ass1} \lambda_n^{-\frac{1}{p-1}}\max_{|x-P^1_n|\ge R\,\lambda_n^{-1/2}} |U_n(x)| \ge 2\delta. \end{equation} Since $U_n$ vanishes on $\partial\Omega$, there exists $P^2_n\in\Omega\backslash B_{R\,\lambda_n^{-1/2}}(P_n^1)$ such that \begin{equation} \label{pn2} |U_n(P_n^2)|=\max_{|x-P^1_n|\ge R\,\lambda_n^{-1/2}}|U_n(x)|. \end{equation} Clearly, assumption \eqref{ass1} implies that $|U_n(P_n^2)|\rightarrow +\infty$. We first prove that the sequences $P_n^1$ and $P_n^2$ are far away each other. \begin{lemma} \label{disj} Take $R$ such that \eqref{ass1} holds, and let $P_n^2$ be defined as in \eqref{pn2}; then \begin{equation} \label{limp1p2} \lambda_n^{1/2}|P_n^2-P^1_n|\rightarrow +\infty \end{equation} as $n\to \infty$. \end{lemma} \begin{proof} Assuming the contrary one would get, up to a subsequence \[ \lambda_n^{1/2}|P_n^2-P^1_n|\rightarrow R'\ge R. \] Let us now recall that by \eqref{defVn} and the subsequent discussion, we have: \begin{equation} \label{limblowseq} \lambda_n^{-\frac{1}{p-1}}\, U_n(\lambda_n^{-1/2}\,y+P^1_n) =: V^1_n(y)\rightarrow V(y)\quad \textrm{in} \,\, \mathcal{C}^1_{{\mathrm{loc}}}({\mathbb{R}}^N) \end{equation} as $n\to +\infty$. Then, up to subsequences, \begin{equation} \nonumber \lambda_n^{-\frac{1}{p-1}}\,|U_n(P_n^2)|=\big |V^1_n\bigr(\lambda_n^{1/2}(P_n^2-P^1_n)\bigl)\big| \rightarrow \big |V(y')\big |,\quad |y'|=R'\ge R. \end{equation} Since $V$ is vanishing for $|y|\to +\infty$, one can choose $R$ such that $|V(y)|\le\delta$ for every $ |y|\ge R$. But this contradicts \eqref{ass1}. \end{proof} Furthermore, we also have that the blow-up points stay far away from the boundary. \begin{lemma} \label{distbd} Assume \eqref{ass1} and let $P_n^2$ be defined as in \eqref{pn2}; then \begin{equation} \label{distp2nbound} \sqrt{\lambda_n}\,d(P^2_n,\partial\Omega)\rightarrow +\infty \end{equation} as $n\to \infty$. Moreover, \begin{equation} \label{maxp2nball} |U_n(P_n^2)|=\max_{\Omega\cap B_{R_n\lambda^{-1/2}_n}(P^2_n)}|U_n| \end{equation} for some $R_n\to +\infty$. \end{lemma} \begin{proof} Let us set \begin{equation} \nonumber \tilde\varepsilon^2_n: =|U_n(P^2_n)|^{-\frac{p-1}{2}}\quad \mathrm{and} \quad R_n^{(2)}:=\frac{1}{2}\,\frac{|P_n^2-P_n^1|}{\tilde\varepsilon^2_n}. \end{equation} Clearly, $\tilde\varepsilon^2_n\rightarrow 0$; moreover, by \eqref{ass1} and \eqref{pn2}, $\tilde\varepsilon^2_n\le (2\delta)^{-\frac{p-1}{2}}\lambda_n^{-1/2}$, so that $$R^{(2)}_n\ge \frac{(2\delta)^{\frac{p-1}{2}}}{2}\,\lambda_n^{1/2}\,{|P_n^2-P_n^1|} \rightarrow +\infty, $$ as $n\to +\infty$ by Lemma \ref{disj}. We claim that this implies \begin{equation} \label{maxp2n} |U_n(P_n^2)|=\max_{\Omega\cap B_{R^{(2)}_n\tilde\varepsilon^2_n}(P^2_n)}| U_n|. \end{equation} For, if $x\in B_{R^{(2)}_n\tilde\varepsilon^2_n}(P^2_n)$, by \eqref{limp1p2} we would have $$|x-P^1_n|\ge |P^2_n-P^1_n|-|x-P^2_n|\ge \frac{1}{2}\,|P^2_n-P^1_n|\ge R\,\lambda_n^{-1/2},$$ for arbitrarily large $R$. This means that $$\Omega\cap B_{R^{(2)}_n\tilde\varepsilon^2_n}(P^2_n)\subset \Omega\backslash B_{R\,\lambda_n^{-1/2}}(P^1_n).$$ Then, the claim follows. Now, by recalling Remark \ref{rem4}, we can apply to $U_n$ satisfying \eqref{maxp2n} the same scaling arguments as in the proof of Lemma \ref{localblow}, so that we conclude $$ 0< \lim_{n\to +\infty}\tilde\varepsilon_n^2\,\sqrt{\lambda_n}. $$ Hence, \eqref{maxp2nball} holds by defining $R_n=R^{(2)}_n\tilde\varepsilon_n^2\,\sqrt{\lambda_n}$, and \eqref{distp2nbound} follows by Corollary \ref{distpnbound}. \end{proof} We can now iterate the previous arguments: let us define, for $k\ge 1$, \begin{equation} \label{defhn} h_k(R)=\limsup_{n\to +\infty} \Bigl (\lambda_n^{-\frac{1}{p-1}}\max_{d_{n,k}(x)\ge R\,\lambda_n^{-1/2}} |U_n(x)| \Bigr ), \end{equation} where $$d_{n,k}(x): =\min\{|x-P^i_n|\,:\, i=1,...,k\}$$ and the sequences $P^i_n$ are such that \begin{equation} \nonumber \sqrt{\lambda_n}\,d(P^i_n,\partial\Omega)\rightarrow +\infty;\quad \lambda_n^{1/2}|P_n^i-P^j_n|\rightarrow +\infty,\quad\quad i,j=1,...,k,\quad i\neq j \end{equation} as $n\to +\infty$. Assume that $$\limsup_{n\to +\infty} h_k(R)=4\delta>0.$$ As before, up to a subsequence and for arbitrarily large $R$, we have \begin{equation} \label{assk} \lambda_n^{-\frac{1}{p-1}}\max_{d_{n,k}(x)\ge R\,\lambda_n^{-1/2}} |U_n(x)| \ge 2\delta \end{equation} and there exist $P^{k+1}_n$ so that \begin{equation} \nonumber |U_n(P_n^{k+1})|=\max_{d_{n,k}(x)\ge R\,\lambda_n^{-1/2}}|U_n(x)| \end{equation} with $\lim_{n\to +\infty}|U_n(P_n^{k+1})|=+\infty$. Moreover, as in Lemma \ref{disj} we deduce that, for every $i=1,...,k$ \begin{equation} \label{limblowseqi} \lambda_n^{-\frac{1}{p-1}}\,U_n(\lambda_n^{-1/2}\,y+P^i_n): = V^i_n(y)\rightarrow V^i(y)\quad \textrm{in} \,\, \mathcal{C}^1_{{\mathrm{loc}}}({\mathbb{R}}^N) \end{equation} as $n\to +\infty$; hence, by \eqref{assk} and again from the vanishing of $V$ at infinity, we conclude that \begin{equation} \lambda_n^{1/2}|P_n^{k+1}-P^i_n|\rightarrow +\infty \end{equation} as $n\to \infty$, for every $i=1,...,k$. Setting now \begin{equation} \nonumber \tilde\varepsilon^{k+1}_n: =|U_n(P^{k+1}_n)|^{-\frac{p-1}{2}}\quad \mathrm{and} \quad R_n^{(k+1)}:=\frac{1}{2}\,\frac{d_{n,k}(P^{k+1}_n)}{\tilde\varepsilon^{k+1}_n} \end{equation} we still have $\tilde\varepsilon^{k+1}_n\to 0$ and, by \eqref{assk}, $R_n^{(k+1)} \to +\infty$ as $n\to \infty$ (see Lemma \ref{distbd}). Then, by the same arguments as in Lemma \ref{distbd}, we get \begin{equation} \label{maxpkn} |U_n(P_n^{k+1})|=\max_{\Omega\cap B_{R^{(k+1)}_n\tilde\varepsilon^{k+1}_n}(P^{k+1}_n)} |u_n|\,, \end{equation} and furthermore $$ \lim_{n\to +\infty}\tilde\varepsilon_n^{k+1}\,\sqrt{\lambda_n}>0\,,$$ so that by defining $R_n=:R^{(k+1)}_n\tilde\varepsilon_n^{k+1}\,\sqrt{\lambda_n}\rightarrow +\infty$ we have \begin{equation} \label{maxpknball} |U_n(P_n^{k+1})|=\max_{\Omega\cap B_{R_n\lambda^{-1/2}_n}(P^{k+1}_n)}| U_n|. \end{equation} Now, by the same arguments as in \cite{MR2825606}, it turns out that the iterative procedure must stop after \emph{at most} $\bar k-1$ steps, where $\bar k =\lim_{n\to +\infty} m(u_n)$. Thus, we have proved: \begin{proposition} \label{glob1} Let $\{U_n\}_n$ be a solution sequence to \eqref{equn} such that $\lambda_n\to+\infty$ and $m(U_n)\leq\bar k$. Then, up to a subsequence, there exist $P_n^1,...,P_n^k$, with $k\le \bar k$ such that \begin{equation} \label{limpin} \sqrt{\lambda_n}\,d(P^i_n,\partial\Omega)\rightarrow +\infty;\quad \lambda_n^{1/2}|P_n^i-P^j_n|\rightarrow +\infty,\quad\quad i,j=1,...,k,\quad i\neq j \end{equation} as $n\to +\infty$ and \begin{equation} \nonumber |U_n(P_n^{i})|=\max_{\Omega\cap B_{R_n\lambda^{-1/2}_n}(P^{i}_n)}|U_n|,\quad i=1,...,k, \end{equation} for some $R_n\to +\infty$ as $n\to +\infty$. Finally, \begin{equation} \label{limhr0} \lim_{R\to +\infty} h_k(R)=0 \end{equation} where $h_k(R)$ is given by \eqref{defhn}. \end{proposition} We now show that the sequence $U_n$ decays exponentially away from the blow-up points. \begin{proposition} \label{glob2} Let $\{U_n\}_n$ satisfy the assumptions of Proposition \ref{glob1}. Then, there exist $P_n^1,...,P_n^k$ and positive constants $C$, $\gamma$, such that \begin{equation} \label{stimglob} |U_n(x)|\le C\lambda^{\frac{1}{p-1}}_n \sum_{i=1}^ke^{-\gamma\sqrt{\lambda_n}|x-P_n^i|}\,,\quad\quad \forall \,x\in\Omega,\quad n\in{\mathbb{N}}\,. \end{equation} \end{proposition} \begin{proof} By \eqref{limhr0}, for large $R>0$ and $n>n_0(R)$ it holds \begin{equation} \nonumber \lambda_n^{-\frac{1}{p-1}}\max_{d_{n,k}(x)\ge R\,\lambda_n^{-1/2}} |U_n(x)| \le \Bigr (\frac{1}{2p} \Bigl )^{\frac{1}{p-1} } \end{equation} Then, for $n>n_0(R)$ and for $x\in \{d_{n,k}(x)\ge R\,\lambda_n^{-1/2}\}$, we have \begin{equation} \nonumber a_n(x): = \lambda_n-p |U_n(x)|^{p-1}\ge \lambda_n-\frac{\lambda_n}{2}=\frac{\lambda_n}{2} \end{equation} We stress that the linear operator \begin{equation} \nonumber L_n: =-\Delta + a_n(x) \end{equation} comes from the linearization of equation \eqref{equn} at $U_n$; let us compute this operator on the functions $$\phi^i_n(x)=e^{-\gamma\sqrt{\lambda_n}\,|x-P^i_n|}\,,\quad\quad \gamma>0,\quad\quad i=1,...,k$$ in $\{d_{n,k}(x)\ge R\,\lambda_n^{-1/2}\}$. We obtain: $$L_n \phi^i_n(x)=\lambda_n\phi^i_n(x)\Bigr [-\gamma^2+(N-1)\frac{\gamma}{\sqrt{\lambda_n}\,|x-P^i_n|} +\frac{a_n(x)}{\lambda_n}\Bigl ]\ge \lambda_n\phi^i_n(x)\bigr [-\gamma^2+1/2\bigl ]\ge 0$$ for $n$ large, provided $0<\gamma\le 1/\sqrt 2$. Moreover, for $|x-P^i_n|=R\lambda_n^{-1/2}$, $ i=1,...,k,$ and $R$ large we have $$ e^{\gamma R}\phi^i_n(x)-\lambda_n^{-\frac{1}{p-1}}|U_n(x)|= 1-\lambda_n^{-\frac{1}{p-1}}|U_n(x)|>0 $$ as $n\to +\infty$, by \eqref{limblowseq}. Note further that $$\{x:d_{n,k}(x)= R\,\lambda_n^{-1/2}\} = \bigcup_{i=1}^k \partial B_{R\,\lambda_n^{-1/2}}(P_n^i) \subset \Omega$$ for large enough $n$. Then, by defining \begin{equation} \nonumber \phi_n: = e^{\gamma R}\lambda_n^{\frac{1}{p-1}}\sum_{i=1}^k \,\phi^i_n \end{equation} we have $$\phi_n(x)-|U_n(x)|\ge 0\quad\quad \mathrm{on}\quad\quad\{d_{n,k}(x)= R\,\lambda_n^{-1/2}\}\cup\partial\Omega$$ and \begin{equation} \nonumber L_n(\phi_n-|U_n|)\ge -L_n\,|U_n|=\Delta \,|U_n|-\lambda_n\,|U_n|+p|U_n|^p\ge (p-1)\,|U_n|^p\ge 0 \end{equation} in $\Omega\backslash \{d_{n,k}(x)\le R\,\lambda_n^{-1/2}\}$. Then (for $R$ large and $n\ge n_0(R)$) we obtain $|U_n|\le \phi_n$ in the same set, by the minimum principle. Moreover, since by \eqref{limtildelam} $$|U_n(x)|\le\|U_n\|_{L^{\infty}(\Omega)}=|U_n(P^1_n)|\le C \lambda_n^{\frac{1}{p-1}}$$ for some $C>0$, we also have, in $\{d_{n,k}(x)\le R\,\lambda_n^{-1/2}\}$, $$ |U_n(x)|\le\|U_n(x)\|_{L^{\infty}(\Omega)}=|U_n(P^1_n)|\le C e^{\gamma R}\lambda_n^{\frac{1}{p-1}}\sum_{i=1}^k e^{-\gamma\sqrt{\lambda_n}|x-P_n^i|}. $$ Then, possibly by choosing a larger $C$, estimate \eqref{stimglob} follows for every $n$. \end{proof} We now exploit the previous results to show that suitable rescalings of the solutions to \eqref{eq:auxiliary_n} converge (locally) to some bounded solution $V$ of \begin{equation} \label{eqV} -\Delta V+ V=| V|^{p-1} V \end{equation} in ${\mathbb{R}}^N$. \begin{lemma} \label{lemlim1} Let \eqref{eq:mainass_secMorse} hold. Then $|u_n|$ admits $k\le \bar k$ local maxima $P_n^1,...,P_n^k$ in $\Omega$ such that, defining \begin{equation} u_{i,n}(x)= \Bigl ( \frac{\mu_n}{\lambda_n}\Bigr )^{\frac{1}{p-1}}u_n \bigr (\frac{x} {\sqrt {\lambda_n}}+P_n^i\bigl ),\quad\quad x\in \Omega_{n,i}:=\sqrt{\lambda_n}\bigr (\Omega-P_n^i \bigl ), \end{equation} it results, up to a subsequence, \begin{equation} u_{i,n}(x)\rightarrow V_i\quad\quad \mathrm{in}\,\,\mathcal{C}^1_{{\mathrm{loc}}}({\mathbb{R}}^n)\quad \mathrm{as}\,\,n\to +\infty,\quad \forall\,\,i=1,2,...,k, \end{equation} where $V_i$ is a bounded solution of \eqref{eqV} with $m(V_i)\le \bar{k}$. \noindent As a consequence, for every $q\ge 1$, \begin{equation} \label{convlq} \Bigl ( \frac{\mu_n}{\lambda_n}\Bigr )^{\frac{q}{p-1}}\lambda_n^{N/2}\int_{\Omega} |u_n|^q \,dx\rightarrow \sum_{i=1}^{k}\int_{{\mathbb{R}}^n}|V_i|^q\,dx\quad\quad \mathrm{as}\,\, n\to +\infty. \end{equation} \end{lemma} \begin{proof} By Lemma \ref{localblow} we have $\lambda_n\to +\infty$; then, the first part of the lemma follows by definition \eqref{equn}, by \eqref{limblowseqi} and by Proposition \ref{glob1}; by the same proposition and by Proposition \ref{glob2} we also have that the local maxima $P^i_n$ satisfies \eqref{limpin} and that the pointwise estimate \begin{equation} \label{stimglobvn} |u_n(x)|\le C\Bigl (\frac{\lambda_n}{\mu_n}\Bigr )^{\frac{1}{p-1}} \sum_{i=1}^ke^{-\gamma\sqrt{\lambda_n}|x-P_n^i|}\,,\quad\quad \forall \,x\in\Omega,\quad n\in{\mathbb{N}}\,. \end{equation} holds. Let us fix $R>0$ and set $r_n=R/\sqrt{\lambda_n}$; for large enough $n$, \eqref{limpin} implies $$B_{r_n}(P^i_n)\subset \Omega,\quad\quad B_{r_n}(P^i_n)\cap B_{r_n}(P^j_n)=\emptyset, \quad i\neq j.$$ Then we obtain \[ \begin{array}{cl} &\left |\left ( \frac{\mu_n}{\lambda_n}\right)^{\frac{q}{p-1}}\lambda_n^{N/2}\int_{\Omega} |u_n|^q \,dx- \sum_{j=1}^{k}\int_{B_R(0)}|u_{j,n}|^q\,dx\,\,\right | \smallskip\\& =\left ( \frac{\mu_n}{\lambda_n}\right )^{\frac{q}{p-1}}\lambda_n^{N/2}\left |\int_{\Omega} |u_n|^q \,dx- \sum_{j=1}^{k}\int_{B_{r_n}(P^j_n)}|u_{n}|^q\,dx\,\,\right | \smallskip\\ &=\left ( \frac{\mu_n}{\lambda_n}\right )^{\frac{q}{p-1}}\lambda_n^{N/2} \int_{\Omega\backslash \bigcup_{j=1}^k\,B_{r_n}(P^j_n)} |u_n|^q \,dx \le C^q\lambda_n^{N/2} \int_{\Omega\backslash \bigcup_{j=1}^k\,B_{r_n}(P^j_n)} \left |\sum_{i=1}^ke^{-\gamma\sqrt{\lambda_n}|x-P_n^i|}\right |^q \,dx \smallskip\\ &\le C^qk^{q-1}\lambda_n^{N/2} \sum_{i=1}^k \int_{\Omega\backslash \bigcup_{j=1}^k\,B_{r_n}(P^j_n)} e^{-q\gamma\sqrt{\lambda_n}|x-P_n^i|} \,dx \smallskip\\ &\le C^qk^{q-1}\lambda_n^{N/2} \sum_{i=1}^k \int_{{\mathbb{R}}^N\backslash \,B_{r_n}(P^i_n)} e^{-q\gamma\sqrt{\lambda_n}|x-P_n^i|} \,dx \smallskip\\ &\le (Ck)^{q}\sum_{i=1}^k \int_{{\mathbb{R}}^N\backslash \,B_{R}(0)} e^{-q\gamma\,|y|} \,dy\le C_1 \,e^{-C_2 R}, \end{array} \] for some positive $C_1$, $C_2$. Letting $n\to +\infty$ we have, up to subsequences, \begin{multline*} \Bigg |\lim_{n\to +\infty}\Bigl ( \frac{\mu_n}{\lambda_n}\Bigr )^{\frac{q}{p-1}}\lambda_n^{N/2}\int_{\Omega} |u_n|^q \,dx- \sum_{i=1}^{k}\int_{B_R(0)}|V_i|^q\,dx\,\,\Bigg | \\ =\lim_{n\to +\infty}\Bigg |\Bigl ( \frac{\mu_n}{\lambda_n}\Bigr )^{\frac{q}{p-1}}\lambda_n^{N/2}\int_{\Omega} |u_n|^q \,dx- \sum_{i=1}^{k}\int_{B_R(0)}|u_{i,n}|^q\,dx\,\,\Bigg |\le C_1 \,e^{-C_2 R}. \end{multline*} Then, \eqref{convlq} follows by taking $R\to +\infty$. \end{proof} The previous lemma allows us to gain some information on the asymptotic behavior of the sequences $\lambda_n$, $\mu_n$ and $\|u_n\|_{L^{p+1}(\Omega)}$. We first provide some bounds for the solutions of the limit problem \eqref{eqV} which will be useful in the sequel. \begin{lemma} \label{boundbelow} Let $V_i$, $i=1,\dots,k$ be as in Lemma \ref{lemlim1} (so that $m(V_i)\leq\bar k$). There exists a constant $C$, only depending on the full sequence $\{u_n\}_n$ and not on $V_i$ (and on the particular associated subsequence), such that \[ \|V_i\|_{H^1}^2 = \|V_i\|_{L^{p+1}}^{p+1} \leq C. \] Furthermore, if also $m(V_i)\geq2$ (or, equivalently, if $V_i$ changes sign) the following estimates hold: \begin{equation} \label{uppstiml2} \|V_i\|^{p+1}_{L^{p+1}}> 2\,\|Z\|^{{p+1}}_{L^{{p+1}}},\qquad \|V_i\|^2_{L^2}> 2\,\|Z\|^{2}_{L^{2}}, \end{equation} where $Z\equiv Z_{N,p}$ is the unique positive solution to \eqref{eqV}. \end{lemma} \begin{proof} To prove the bounds from above we claim that there exists $\bar R>0$, not depending on $i$, such that $V_i$ is stable outside $\overline{B_{\bar R}}$. Then the desired estimate will follow, since \[ \|V_i\|^{p+1}_{L^{p+1}} = \int_{B_{\bar R}} |V_i|^{p+1} + \int_{{\mathbb{R}}^N\setminus B_{\bar R}} |V_i|^{p+1}, \] where the first term is uniformly bounded by Theorem \ref{thm:unif_est_Farina}, while the second one can be estimated in an uniform way by reasoning as in the proof of \cite[Theorem 2.3]{MR2825606}. To prove the claim, recalling \eqref{defhn} and \eqref{limhr0}, let $\bar R$ be such that \[ h_k(\bar R) \leq \left(\frac{1}{p}\right)^{1/(p-1)}. \] Then $|V_i(x)|^{p-1}\leq 1/p $ on ${\mathbb{R}}^N\setminus B_{\bar R}$ and thus, for any $\psi\in C^\infty_0({\mathbb{R}}^N)$, $\psi\equiv0$ in $B_{\bar R}$, it holds \[ \int_{{\mathbb{R}}^N} |\nabla \psi|^2 + \psi^2 - p|V_i|^{p-1}\psi^2\,dx \geq \left( 1 - p \|V_i\|^{p-1}_{L^{\infty}({\mathbb{R}}^N\setminus B_{\bar R})}\right)\int_{{\mathbb{R}}^N} \psi^2 \geq 0. \] Hence $V_i$ is stable outside $B_{\bar R}$, and the first part of the lemma follows. On the other hand, if $V_i$ is a sign-changing solution to \eqref{eqV}, the associated energy functional \begin{equation} \nonumber E(V_i)= \frac{1}{2}\|\nabla V_i\|^2_{L^2}+\frac{1}{2}\|V_i\|^2_{L^2}-\frac{1}{p+1}\|V_i\|^{p+1}_{L^{p+1}} \end{equation} satisfies the following \emph{energy doubling property} (see \cite{MR2263672}): $$E(V_i)>2\,E(Z)$$ On the other hand, by using the equation $E'(V_i)V_i=0$ and the Pohozaev identity one gets \begin{equation} \label{eulp} \|V_i\|^{p+1}_{L^{p+1}}= 2\,\frac{p+1}{p-1}\,E(V_i),\qquad \|V_i\|^2_{L^2}= \frac{N+2-p\,(N-2)}{p-1}\,E(V_i) \end{equation} Since the ground state solution $Z$ satisfies the same identities, the bounds \eqref{uppstiml2} are readily verified. \end{proof} \begin{proposition} Let \eqref{eq:mainass_secMorse} hold and the functions $V_i$ be defined as in Lemma \ref{lemlim1}. We have, as $n\to +\infty$, \begin{eqnarray} \label{convl2} {\mu_n}^{\frac{2}{p-1}}\,\lambda_n^{N/2-2/(p-1)}&\longrightarrow \sum_{i=1}^{k}\int_{{\mathbb{R}}^n}|V_i|^2\,dx \\ \label{convlp} {\mu_n}^{\frac{p+1}{p-1}}\,\lambda_n^{N/2-(p+1)/(p-1)}\int_{\Omega} |u_n|^{p+1} \,dx&\longrightarrow \sum_{i=1}^{k}\int_{{\mathbb{R}}^n}|V_i|^{p+1}\,dx \\ \label{convl2grad} \alpha_n\,{\mu_n}^{\frac{2}{p-1}}\,\lambda_n^{N/2-(p+1)/(p-1)}&\longrightarrow \sum_{i=1}^{k}\int_{{\mathbb{R}}^n}|\nabla V_i|^2\,dx. \end{eqnarray} \end{proposition} \begin{proof} The limits \eqref{convl2} and \eqref{convlp} follow respectively by choosing $q=2$ and $q=p+1$ in \eqref{convlq} (recall that $\|u_n\|_{L^{2}}=1$). Furthermore, from the equations for $u_n$ and $V_k$, we have \[ \alpha_n+\lambda_n=\mu_n\|u_n\|_{L^{p+1}}^{p+1},\qquad \int_{{\mathbb{R}}^n}|\nabla V_i|^2\,dx + \int_{{\mathbb{R}}^n}|V_i|^2\,dx = \int_{{\mathbb{R}}^n}|V_i|^{p+1}\,dx, \] and also \eqref{convl2grad} follows. \end{proof} \begin{corollary} \label{limmass} With the same assumptions as above, we have that \begin{enumerate} \item if $1<p<1+\frac{4}{N}$, then $\mu_n\to +\infty$ \item if $p=1+\frac{4}{N}$, then $\mu_n\to \big (\sum_{i=1}^{k}\|V_i\|_{L^2}^2\big )^{2/N}\ge k^{2/N} \|Z\|_{L^2}^{4/N}$ \item if $1+\frac{4}{N}<p<2^*-1$, then $\mu_n\to 0$. \end{enumerate} Furthermore \begin{equation} \label{limalphalam} \frac{\alpha_n}{\lambda_n}\longrightarrow \frac{N(p-1)}{N+2-p(N-2)}. \end{equation} \end{corollary} \begin{proof} The limits of $\mu_n$ follow by the previous proposition. To prove the lower bound in $2$, recall that either $V_i=Z$ or $V_i$ satisfies \eqref{uppstiml2}. Finally, taking the quotient between \eqref{convl2grad} and \eqref{convl2}, we have $$ \frac{\alpha_n}{\lambda_n}\longrightarrow \frac{\sum_{i=1}^{k}\int_{{\mathbb{R}}^n}|\nabla V_i|^2\,dx}{\sum_{i=1}^{k}\int_{{\mathbb{R}}^n}| V_i|^2\,dx} $$ On the other hand, for every $i=1,2,...,k$ it holds $$ \|\nabla V_i\|_{L^2}^2=\Bigg (\frac{\| V_i\|_{L^{p+1}}^{p+1}}{\|V_i\|_{L^2}^2} -1 \Bigg )\|V_i\|_{L^2}^2= \frac{N(p-1)}{N+2-p(N-2)}\, \|V_i\|_{L^2}^2 $$ where the last equality follows by \eqref{eulp}. By inserting this into the above limit, we get \eqref{limalphalam}. \end{proof} \begin{proof}[Proof of Theorem \ref{thm:bbd_index}] Let $(U_n,\lambda_n)$ solve \eqref{eq:main_prob_U}, with $\rho=\rho_n\to +\infty$ and $m(U_n)\leq k$. Changing variables as in \eqref{eq:main_prob_u}, we have that $u_n=\rho_n^{-1/2}U_n$ satisfies \eqref{eq:auxiliary_n} with $\mu_n = \rho_n^{(p-1)/2} \to +\infty$. As a consequence, Lemma \ref{lemma:case_alpha_n_bounded} guarantees that $\alpha_n\to+\infty$, and Corollary \ref{limmass} yields $p<1+4/N$. On the other hand, by direct minimization of the energy one can show that, if $p<1+4/N$, for every $\rho>0$ there exists a solution of \eqref{eq:main_prob_U} having Morse index one (see also Section \ref{sec:1const}). \end{proof} \begin{remark} \label{limGN} Reasoning as above we can also show that \begin{equation} \label{newcnp1} \frac{\int_{\Omega} |u_n|^{p+1} \,dx}{\alpha_n^{N(p-1)/4}}\longrightarrow C_{N,p}\,\frac{\|Z\|_{L^2}^{p-1}}{\big (\sum_{i=1}^{k}\| V_i\|_{L^2}^2\big )^{(p-1)/2}}. \end{equation} \end{remark} \section{Max-min principles with two constraints}\label{sec:2const} In this section we deal with the maximization problem with two constraints introduced in \cite{MR3318740}, aiming at considering more general max-min classes of critical points. Let ${\mathcal{M}}$ be defined in \eqref{Emu} and, for any fixed $\alpha>\lambda_1(\Omega)$, let $\mathcal{B}_\alpha$, $\mathcal{U}_\alpha$ be defined as in \eqref{eq:defBU}. We will look for critical points of the $\mathcal{C}^2$ functional \[ f(u)=\int_{\Omega}|u|^{p+1},\quad\quad\quad u\in {\mathcal{M}}, \] constrained to $\mathcal{U}_\alpha$. To start with, we notice that the topological properties of such set depend on $\alpha$. \begin{lemma}\label{lemma:tilde_U_manifold} Let $\alpha>\lambda_1(\Omega)$. Then the set \[ {\mathcal{U}}_{\alpha}\setminus\left\{ \varphi\in {\mathcal{U}}_\alpha : -\Delta\varphi = \alpha\varphi\right\} \] is a smooth submanifold of $H^1_0(\Omega)$ of codimension 2. In particular, this property holds true for ${\mathcal{U}}_\alpha$ itself, provided $\alpha\neq\lambda_k(\Omega)$, for every $k$. \end{lemma} \begin{proof} Let us set $F(u)=(\int_\Omega u^2\,dx-1, \ \int_\Omega|\nabla u|^2\,dx)$. For every $u\in{\mathcal{U}}_\alpha$, if the range of $F'(u)$ is ${\mathbb{R}}^2$ then ${\mathcal{U}}_\alpha$ is a smooth manifold at $u$. Since \[ F'(u)[v]=2\left(\int_\Omega uv\,dx, \ \int_\Omega\nabla u\cdot\nabla v\,dx\right), \qquad\text{for every }v\in H^1_0(\Omega), \] and $F'(u)[u]=2(1,\alpha)$, we have that $F'(u)$ is not surjective if and only if \[ \int_\Omega\nabla u\cdot\nabla v\,dx = \alpha \int_\Omega uv\,dx \qquad\text{for every }v\in H^1_0(\Omega). \qedhere \] \end{proof} \begin{remark} If $\varphi$ belongs to the eigenspace corresponding to $\lambda_k(\Omega)$, then $\varphi \in {\mathcal{U}}_{\lambda_k(\Omega)}$. As a consequence ${\mathcal{U}}_{\lambda_k(\Omega)}$ may not be smooth near $\varphi$. For instance, ${\mathcal{U}}_{\lambda_1(\Omega)}$ consists of two isolated points, $\pm\varphi_1$. \end{remark} Of course ${\mathcal{U}}_\alpha$ is closed and odd, for any $\alpha$. Recalling Definition \ref{def:genus} we deduce that its genus $\gamma({\mathcal{U}}_\alpha)$ is well defined. \begin{lemma} If $\alpha<\lambda_{k+1}(\Omega)$, for some $k$, then $\gamma({\mathcal{U}}_\alpha)\leq k$. \end{lemma} \begin{proof} Let $V_k:=\spann\{\varphi_1,\dots,\varphi_k\}$. Since \[ \min\left\{\int_\Omega |\nabla u|^2\,dx : u\in V_k^\perp,\, \int_\Omega u^2\,dx=1\right\}=\lambda_{k+1}(\Omega), \] we have that ${\mathcal{U}} \cap V_k^\perp = \emptyset$, thus the projection \[ g := \proj_{V_k} \colon {\mathcal{U}}_\alpha \to V_k\setminus\{0\} \] is a continuous odd map of ${\mathcal{U}}_\alpha$ into $V_k\setminus\{0\}$. Now, let $h\colon{\mathbb{S}}^{m}\to {\mathcal{U}}$ be continuous and odd. Then $g\circ h$ is continuous and odd from ${\mathbb{S}}^{m}$ to $V_k\setminus\{0\}$, and Borsuk-Ulam's Theorem forces $m\leq k-1$. \end{proof} \begin{lemma}\label{lemma:genusbigger} If $\alpha>\lambda_{k}(\Omega)$, for some $k$, then $\gamma({\mathcal{U}}_\alpha)\geq k$. \end{lemma} \begin{proof} To prove the lemma we will construct a continuous map $h\colon {\mathbb{S}}^{k-1} \to {\mathcal{U}}$. Let $\ell\in{\mathbb{N}}$ be such that $\lambda_{\ell+1}(\Omega)>\alpha$. For every $i=1,\dots,k$ we define the functions \[ u_i:=\left(\frac{\lambda_{\ell+i}(\Omega)-\alpha}{\lambda_{\ell+i}(\Omega)-\lambda_i(\Omega)}\right)^{1/2}\varphi_i +\left(\frac{\alpha-\lambda_{i}(\Omega)}{\lambda_{\ell+i}(\Omega)-\lambda_i(\Omega)}\right)^{1/2}\varphi_{\ell+i}. \] We obtain the following straightforward consequences: \begin{enumerate} \item as $\lambda_i(\Omega)<\alpha<\lambda_{\ell+i}(\Omega)$, for every $i$, $u_i$ is well defined; \item $\int_\Omega u_i^2\,dx=1$, $\int_\Omega |\nabla u_i|^2\,dx=\alpha$; \item for every $j\neq i$ it holds $\int_\Omega u_iu_j\,dx=\int_\Omega \nabla u_i\cdot\nabla u_j \,dx=0$. \end{enumerate} Therefore the map $h\colon {\mathbb{S}}^{k-1} \to {\mathcal{U}}$ defined as \[ h\colon x=(x_1,\dots,x_k) \mapsto \sum_{i=1}^k x_iu_i \] has the required properties. \end{proof} Now we turn to the properties of the functional $f$. To start with, it satisfies the Palais-Smale (P.S. for short) condition on $\overline{\mathcal{B}}_{\alpha}$; more precisely, the following holds. \begin{lemma} \label{psball} Every P.S. sequence $u_n$ for $f\big |_{\overline{\mathcal{B}}_{\alpha}}$ is a P.S. sequence for $f\big |_{\mathcal{U}_{\alpha}}$ and has a strongly convergent subsequence in $\mathcal{U}_{\alpha}$. \end{lemma} \begin{proof} We first show that there are no P.S. sequences in ${\mathcal{B}_{\alpha}}$. In fact, if $u_n$ is such a sequence, there is a sequence of real numbers $k_n$ such that \begin{equation} \label{ps} \int_{\Omega}|u_n|^{p-1}u_n\,v-k_n\int_{\Omega}u_n\,v=o(1)\,\|v\|_{H^1_0} \end{equation} for every $v\in H^1_0(\Omega)$. Since $u_n$ is bounded in $H^1_0(\Omega)$, there is a subsequence (still denoted by $u_n$) weakly convergent to $u\in H^1_0(\Omega)$; moreover, $u_n$ converges strongly in $L^{p+1}(\Omega)$ and in $L^2(\Omega)$ to the same limit. By choosing $v=u_n$, we see that $k_n$ is bounded, so that we can also assume that $k_n\rightarrow k$. By taking the limit of \eqref{ps} for $n\to\infty$ we get \begin{equation} \nonumber \int_{\Omega}|u|^{p-1}u\,v=k\int_{\Omega}u\,v \end{equation} for every $v\in H^1_0(\Omega)$. Hence $u$ is constant, but this contradicts $u\in {\mathcal{M}}$. Now, if $u_n$ is a P.S. sequence for $f$ on ${\mathcal{U}}_{\alpha}$, there are sequences of real numbers $k_n$, $l_n$ such that \begin{equation} \label{ps1} \int_{\Omega}|u_n|^{p-1}u_n\,v-k_n\int_{\Omega}u_n\,v-l_n\int_{\Omega}\nabla u_n\,\nabla v=o(1)\,\|v\|_{H^1_0}. \end{equation} It is readily seen that $l_n$ is bounded away from zero, otherwise \eqref{ps1} is equivalent to \eqref{ps} (for some subsequence) and we still reach a contradiction. Then, we can divide both sides by $l_n$ and find that there are sequences $\{\lambda_n\}_n$, $\{\mu_n\}_n$, with $\mu_n$ bounded, such that \begin{equation} \nonumber \int_{\Omega}\nabla u_n\,\nabla v+\lambda_n\int_{\Omega}u_n\,v-\mu_n\int_{\Omega}|u_n|^{p-1}u_n\,v=o(1)\,\|v\|_{H^1_0}. \end{equation} Now, by reasoning as before one finds that also the sequence $\{\lambda_n\}_n$ is bounded, so that by the relation $$-\Delta u_n+\lambda_n u_n-\mu_n |u_n|^{p-1}u_n=o(1)\quad \mathrm{in}\,\, H^{-1}(\Omega)$$ and by the compactness of the embedding $H^1_0(\Omega)\hookrightarrow L^{p+1}(\Omega)$, the P.S. condition holds for the functional $f\big |_{{\mathcal{U}}_{\alpha}}$. \end{proof} We can combine the previous lemmas to prove one of the main results stated in the introduction. \begin{proof}[Proof of Theorem \ref{thm:genus_2constr}] Lemma \ref{psball} allows to apply standard variational methods (see e.g. \cite[Thm. II.5.7]{St_2008}). We deduce that $M_{\alpha,\,k}$ is achieved at some critical point $u$ of $f\big |_{\mathcal{U}_{\alpha}}$. This amounts to say that $u$ satisfies \eqref{lagreq} for some real $\lambda$ and $\mu\neq 0$. We claim that there exists at least one $u\in f^{-1}(M_{\alpha,\,k})\cap \mathcal{U}_{\alpha}$ such that \eqref{lagreq} holds with $\mu>0$. Assume by contradiction that for \emph{every} critical point of $f\big |_{\mathcal{U}_{\alpha}}$ at level $M_{\alpha,k}$ it holds $\mu< 0$ in equation \eqref{lagreq}. Let us define the functional $T:\,H^1_0(\Omega)\to {\mathbb{R}}$ as $$T(u)=\frac{1}{2}\int_{\Omega}|\nabla u|^2.$$ By denoting with $D$ the Fr\'{e}chet derivative and by $<\,,\,>$ the pairing between $H_0^1$ and its dual $H^{-1}$, our assumption can be restated as follows: \noindent if there are $u\in f^{-1}(M_{\alpha,\,k})\cap\mathcal{U}_{\alpha}$ and $\mu\neq 0$ such that \begin{equation} \label{lagreq1} \langle DT(u),\phi\rangle=\mu\langle Df(u),\phi\rangle \end{equation} for every $\phi\in H^1_0(\Omega)$ satisfying $\int_{\Omega}\phi u=0$ (that is for every $\phi$ tangent to ${\mathcal{M}}$ at $u$) then $\mu<0$. We stress that both $DT(u)$ and $Df(u)$ in the above equation are bounded away from zero, since there are no Dirichlet eigenfunctions in $\mathcal{U}_{\alpha}$ nor critical points of $f$ on ${\mathcal{M}}$. Hence, by denoting with $\nabla_{T{\mathcal{M}}}$ the gradient of a functional (in $H^1_0$) in the direction tangent to ${\mathcal{M}}$, if $u\in f^{-1}(M_{\alpha,\,k})\cap \mathcal{U}_{\alpha}$ then $\nabla_{T{\mathcal{M}}}T(u)$ and $\nabla_{T{\mathcal{M}}}f(u)$ \emph{are either opposite or not parallel}. Moreover, the angle between these (non vanishing) vectors is \emph{bounded away from zero}; otherwise, we would find sequences $u_n\in \mathcal{U}_{\alpha}$, $\mu_n>0$ such that \begin{equation} \label{noparal} (\nabla_{T{\mathcal{M}}}T(u_n),v)_{H^1_0}-\mu_n(\nabla_{T{\mathcal{M}}}f(u_n),v)_{H^1_0}=o(1)\|v\|_{H^1_0} \end{equation} for every $v\in H^1_0(\Omega)$; but since $$(\nabla_{T{\mathcal{M}}}T(u_n),v)_{H^1_0}=\int_{\Omega}\nabla u_n\,\nabla v-\lambda_n^T\int_{\Omega}u_n\, v\,,$$ $$(\nabla_{T{\mathcal{M}}}f(u_n),v)_{H^1_0}=\int_{\Omega}|u_n|^{p-1}u_n\,v-\lambda_n^f\int_{\Omega}u_n\, v\,,$$ for suitable bounded sequences $\lambda_n^T$, $\lambda_n^f$, this is equivalent to saying that $u_n$ is a P.S. sequence for $f\big |_{\mathcal{U}_{\alpha}}$, so that, by Lemma \ref{psball}, we would get a constrained critical point with $\mu>0$. Then, by choosing suitable linear combinations of the above tangential components one can define a bounded $\mathcal{C}^1$ map $u\mapsto v(u)\in H_0^1(\Omega)$, with $v(u)$ tangent to ${\mathcal{M}}$ and satisfying the following property: there is $\delta>0$ such that \begin{equation} \label{diseqv} \int_{\Omega}\nabla u\,\nabla v(u)< -\delta\, ,\quad\quad \int_{\Omega}|u|^{p-1}u\,v(u)>\delta\,, \end{equation} for every $u\in f^{-1}(M_{\alpha,\,k})\cap \mathcal{U}_{\alpha}$. By continuity and possibly by decreasing $\delta$, inequalities \eqref{diseqv} extend to \begin{equation} \label{diseqv1} f^{-1}(M_{\alpha,\,k}-\bar\varepsilon, M_{\alpha,\,k}+\bar\varepsilon)\cap \big (\overline{\mathcal{B}}_{\alpha} \backslash \overline{\mathcal{B}}_{\alpha-\tau}\big ) \end{equation} for small enough, positive $\bar\varepsilon$ and $\tau$. Finally, since there are no critical points of $f$ in ${\mathcal{B}}_{\alpha}$ we can take that the \emph{second of \eqref{diseqv} holds on} \begin{equation} \label{diseqv2} f^{-1}(M_{\alpha,\,k}- \bar\varepsilon, M_{\alpha,\,k}+\bar\varepsilon)\cap \overline{\mathcal{B}}_{\alpha}. \end{equation} Let $\varphi$ be a $\mathcal{C}^1$ function on ${\mathbb{R}}$ such that: $$0\le\varphi\le 1, \quad\varphi\equiv 1\,\, \mathrm{in}\,\,(M_{\alpha,\,k}-\bar\varepsilon/2, M_{\alpha,\,k}+\bar\varepsilon/2),\quad \varphi\equiv 0\,\, \mathrm{in}\,\,{\mathbb{R}}\backslash (M_{\alpha,\,k}-\bar\varepsilon, M_{\alpha,\,k}+\bar\varepsilon),$$ and define \begin{equation} \label{vectfield} e(u)=\varphi(f(u))\,v(u). \end{equation} Clearly, $e$ is a $\mathcal{C}^1$ vector field on ${\mathcal{M}}$ and is uniformly bounded, so that there exists a global solution $\Phi(u,t)$ of the initial value problem $$\partial_t\Phi(u,t)=e\big (\Phi(u,t)),\quad\quad \Phi(u,0)=0.$$ By definition \eqref{vectfield} and by the first of \eqref{diseqv} (on \eqref{diseqv1}) we get $\Phi(u,t_0) \in \overline{\mathcal{B}}_{\alpha}$ for $t_0> 0$ and for any $u\in \overline{\mathcal{B}}_{\alpha}$; moreover, by the second inequality of \eqref{diseqv} (on \eqref{diseqv2}) there exists $\varepsilon\in (0,\bar\varepsilon)$ such that $$f(\Phi(u,t_0))>M_{\alpha,\,k}+\varepsilon$$ for every $u\in f^{-1}(M_{\alpha,\,k}-\varepsilon, +\infty)\cap \overline{\mathcal{B}}_{\alpha}$. Now, by \eqref{maxmin}, there is $A_{\varepsilon}\subset \overline{\mathcal{B}}_{\alpha}$ such that $\gamma(A_{\varepsilon})\ge k$ and $$\inf_{u\in A_{\varepsilon}} f(u)\ge M_{\alpha,\,k}-\varepsilon.$$ Hence, $\gamma\big (\Phi(A_{\varepsilon},t_0) \big )\ge k$ and $$\inf_{u\in \Phi(A_{\varepsilon},t_0)} f(u)\ge M_{\alpha,\,k}+\varepsilon$$ contradicting the definition of $M_{\alpha,\,k}$. \end{proof} \begin{remark} \label{rem1} If $\mu>0$, by testing \eqref{lagreq} with $u$ and by integration by parts we readily get $\lambda>-\alpha$. An alternative lower bound, independent of $\alpha$, could be obtained by adapting arguments from \cite{MR968487,MR954951,MR991264} in order to prove that the Morse index of $u$ (as a solution of \eqref{lagreq}) is less or equal than $k$. Then Lemma \ref{lem:lambda_bdd_below} would provide $\lambda\geq-\lambda_{k}$. \end{remark} \begin{remark}\label{rem:MvsCNp} By the Gagliardo-Nirenberg inequality \eqref{sobest} we readily obtain that, for every $k\geq1$, \[ M_{\alpha,k}\leq C_{N,p} \alpha^{N(p-1)/4}. \] Taking into account the previous remark, this agrees with Remark \ref{limGN}. \end{remark} We conclude this section with the following estimate. \begin{lemma} \label{lem:M3vsM1} Under the assumptions and notation of Theorem \ref{thm:genus_2constr}, \[ M_{\alpha,3} \leq 2^{-(p-1)/2} M_{\alpha,1}. \] \end{lemma} \begin{proof} Let $A\in \Sigma^{(3)}_{\alpha}$, according to Definition \ref{def:genus}. Notice that the map \[ A\ni u \mapsto \left(\int_\Omega |u|u , \int_\Omega |u|^p u\right)\in{\mathbb{R}}^2 \] is continuous and equivariant. By the Borsuk-Ulam Theorem, we deduce the existence of $u_a\in A$ such that \[ \int_\Omega |u^+_a|^2 = \int_\Omega |u^-_a|^2 = \frac12,\qquad \int_\Omega |u^+_a|^{p+1} = \int_\Omega |u^-_a|^{p+1} = \frac12 \int_\Omega |u_a|^{p+1}, \] while \[ \text{either }\int_\Omega |\nabla u^+_a|^2 \leq \frac{\alpha}{2} \qquad \text{or }\int_\Omega |\nabla u^-_a|^2 \leq \frac{\alpha}{2}. \] For concreteness let us assume that the first alternative holds; as a consequence, we obtain that $v:=\sqrt2 u_a^+$ belongs to $\overline{\mathcal{B}}_\alpha$. This yields \[ M_{\alpha,1}\geq \int_\Omega |v|^{p+1} = 2^{(p+1)/2} \int_\Omega |u_a^+|^{p+1} = \frac{2^{(p+1)/2}}{2} \int_\Omega |u_a|^{p+1} \geq 2^{(p-1)/2} \inf_{u\in A} \int_\Omega |u|^{p+1}, \] and since $A\in \Sigma^{(3)}_{\alpha}$ is arbitrary the proposition follows. \end{proof} \section{Min-max principles on the unit sphere in \texorpdfstring{$L^2$}{L\texttwosuperior}}\label{sec:1const} According to equation \eqref{Emu}, let ${\mathcal{M}}\subset H^1_0(\Omega)$ denote the unit sphere with respect to the $L^2$ norm and $\mathcal{E}_{\mu}$ the energy functional associated to \eqref{eq:main_prob_u}. In this section we are concerned with critical points of $\mathcal{E}_{\mu}$ on ${\mathcal{M}}$ (which, in turn, correspond to solutions of our starting problem \eqref{eq:main_prob_U}). By the Gagliardo-Nirenberg inequality \eqref{sobest}, setting $\|\nabla u\|^2_{L^2}=\alpha$, one obtains \begin{equation}\label{eq:boundonboundEmu} \frac12\,\alpha- \mu\frac{C_{N,p}}{p+1}\,\alpha^{N(p-1)/4} \leq \mathcal{E}_{\mu}(u)\le \frac12\alpha. \end{equation} In particular, $\mathcal{E}_{\mu}$ is bounded on any bounded subset of ${\mathcal{M}}$, and it is bounded from below (and coercive) on the entire ${\mathcal{M}}$ for {subcritical} $p<1+4/N$ and for {critical} $p=1+4/N$ whenever $\mu< \frac{p+1}{2}C_{N,p}^{-1}$ . In these cases, one can easily show that $\mathcal{E}_{\mu}$ satifies the P.S. condition and apply the classical {minimax principle for even functionals} on a closed symmetric submanifold (see e.g. \cite[Thm. II.5.7]{St_2008}). In the complementary case, when $p$ is either supercritical, i.e. $p>1+4/N$, or critical and $\mu$ is large, then $\mathcal{E}_{\mu}$ is not bounded from below (see e.g. \eqref{minusinfty} below). In order to provide a minimax principle suitable for this case, we recall the Definition \ref{def:genus} of genus and that of $\mathcal{B}_\alpha$ (see equation \eqref{eq:defBU}). Furthermore, we denote with $K_{c}$ the (closed and symmetric) set of critical points of ${\mathcal{E}}_\mu$ at level $c$ contained in $\mathcal{B}_\alpha$. The following theorem is an adaptation of well known arguments of previous critical point theorems relying on index theory. \begin{theorem} \label{infsupteo} Let $k\ge1$, $\alpha>\lambda_k(\Omega)$, $\mu>0$ and $\tau>0$ be fixed, and let $c_k$ be defined as in Theorem \ref{thm:genus_1constr}, equation \eqref{infsuplev}. If \begin{equation} \label{ass2} c_k < \hat c_k:= \inf_{\substack{A\in\Sigma^{(k)}_{\alpha}\\ A\setminus\mathcal{B}_{\alpha-\tau}\neq\emptyset }} \sup_{A\setminus\mathcal{B}_{\alpha-\tau}}{\mathcal{E}}_\mu, \end{equation} then $K_{c_k}\neq\emptyset$, and it contains a critical point of Morse index less or equal to $k$. \end{theorem} \begin{remark} In case assumption \eqref{ass2} holds for $k,k+1,\dots,k+r$, and $c=c_k=...c_{k+r}$, then it is standard to extend Theorem \ref{infsupteo} to obtain \begin{equation} \label{indexK} \gamma(K_c)\ge r+1, \end{equation} so that $K_c$ contains infinitely many critical points. \end{remark} \begin{proof}[Proof of Theorem \ref{infsupteo}] For any $a\in{\mathbb{R}}$ we denote by ${\mathcal{M}}_a$ the sublevel set $\{\mathcal{E}_\mu<a\}$. First of all we notice that both $c_k$ and $\hat c_k$ are well defined and finite, by Lemma \ref{lemma:genusbigger} and equation \eqref{eq:boundonboundEmu}. Suppose now by contradiction that $K_{c_k}=\emptyset$. By a suitably modified version of the Deformation Lemma (recall that ${\mathcal{E}}_\mu$ satisfies the P.S. condition on ${\mathcal{M}}$), there exist $\delta>0$ and an equivariant homeomorphism $\eta$ such that $\eta(u)=u$ outside $\mathcal{B}_{\alpha}\cap {\mathcal{M}}_{c_k+2\delta}$ and \begin{equation} \label{lowlev} \eta({{\mathcal{M}}_{c_k+\delta}\cap \mathcal{B}_{\alpha-\tau}})\subset {\mathcal{M}}_{c_k-\delta}\cap \mathcal{B}_{\alpha}. \end{equation} By definition of $c_{k}$ there exists $A\in \Sigma^{(k)}_{\alpha}$ such that $A\subset {\mathcal{M}}_{c_k+\delta}$; it follows by assumption \eqref{ass2} (and by decreasing $ \delta$ if necessary) that $A\subset {\mathcal{M}}_{c_k+\delta}\cap \mathcal{B}_{\alpha-\tau}$. Then, since $\eta$ is an odd homeomorphism, $\eta(A)\in \Sigma^{(k)}_{\alpha}$ and, by definition, $\sup_{\eta(A)}{\mathcal{E}}_\mu \ge c_k$, in contradiction with \eqref{lowlev}. Finally, the estimate of the Morse index is a direct consequence of the definition of genus we deal with: see \cite{MR968487}, Proposition on page 1030, or the discussion at the end of Section 2 in \cite{MR991264}. \end{proof} We now provide a sufficient condition to guarantee the validity of assumption \eqref{ass2}. \begin{lemma}\label{lem:ckMak} Let $k\ge1$, $\alpha>\lambda_k(\Omega)$ and $\mu>0$ satisfy \begin{equation} \label{muboundef} 0<\mu<\frac{p+1}{2}\,\frac{\alpha-\lambda_{k}(\Omega)}{M_{\alpha,k}-|\Omega|^{-\frac{p-1}{2}}} \end{equation} where $M_{\alpha,k}$ is defined in Theorem \ref{thm:genus_2constr}. Then, for $\tau>0$ sufficiently small, \eqref{ass2} holds true. \end{lemma} \begin{proof} We first estimate $c_k$ from above. To this aim, we construct a subset $\tilde A\in \Sigma^{(k)}_{\alpha-\tau}$ (for any $\tau$ sufficiently small) as \begin{equation} \label{Atilde} \tilde A= \left\{\sum_{i=1}^k x_i\varphi_i : x=(x_1,\dots,x_k)\in{\mathbb{S}}^{k-1}\right\}, \end{equation} where, as usual $\varphi_i$ denotes the Dirichlet eigenfunction associated to $\lambda_i(\Omega)$. Indeed $\gamma(\tilde A)=k$ (it is homeomorphic to ${\mathbb{S}}^{k-1}$), and $\max_{u\in\tilde A}\|u\|^2_{H^1_0}=\lambda_{k}(\Omega)<\alpha-\tau$ for $\tau$ small. Hence Holder inequality yields \begin{equation} \label{boundabove1} c_k \leq \sup_{\tilde A}\mathcal{E}_{\mu}\le \frac{1}{2}\lambda_{k}(\Omega) - \frac{\mu}{p+1}\,|\Omega|^{-\frac{p-1}{2}}. \end{equation} On the other hand, let $A\in\Sigma^{(k)}_{\alpha}$. Theorem \ref{thm:genus_2constr} implies \[ \inf_{u\in A} \int_\Omega |u|^{p+1} \leq M_{\alpha,k}. \] If moreover $A\setminus\mathcal{B}_{\alpha-\tau}\neq\emptyset$ we infer \[ \sup_{A\setminus\mathcal{B}_{\alpha-\tau}}\mathcal{E}_{\mu}\ge \frac{1}{2}(\alpha - \tau) - \frac{\mu}{p+1} M_{\alpha,k}, \] and taking the infimum an analogous inequality holds true for $\hat c_k$. Comparing with \eqref{boundabove1} the lemma follows. \end{proof} Exploiting the results above, we are ready to prove our main existence results. \begin{proof}[End of the proof of Theorem \ref{thm:genus_1constr}] By Theorem \ref{infsupteo} and Lemma \ref{lem:ckMak} the proof is completed by choosing \[ \hat\mu_k:=\sup_{\alpha>\lambda_k(\Omega)} \frac{p+1}{2}\,\frac{\alpha-\lambda_{k}(\Omega)}{M_{\alpha,k}-|\Omega|^{-\frac{p-1}{2}}}. \qedhere \] \end{proof} \begin{proof}[Proof of Theorem \ref{thm:intro_GS}] We write the proof in terms of ${\mathcal{E}}_\mu$, the theorem following by the relations in \eqref{eq:main_prob_u}. Recall that, for every $u\in \overline{\mathcal{B}}_\alpha$, $\gamma \left(\{u,-u\}\right)=1$. We deduce that $c_1$ is actually a local minimum for ${\mathcal{E}}_\mu$, achieved by some $u$ which solves \eqref{eq:main_prob_u} (for a suitable $\lambda$), and it can be chosen positive by symmetry. Since \[ \int_\Omega |\nabla u|^2 + \lambda u^2 - p\mu|u|^{p+1}\,dx=-(p-1)\int_\Omega \mu|u|^{p+1}\,dx<0, \] and $H^1_0(\Omega) = \spann\{u\}\oplus T_{u}\mathcal{M}$, we have that $u$ has Morse index $1$. In a standard way, the minimality property of $u$ implies also orbital stability of the associated solitary wave (see e.g. \cite{MR677997}). Turning to the estimates for $\hat\mu_1 = \hat\rho_1^{(p-1)/2}$, we can deduce it using Lemma \ref{lem:ckMak} and Remark \ref{rem:MvsCNp}, which yield \[ \hat\mu_1\left(\Omega,p\right):=\sup_{\alpha>\lambda_1(\Omega)} \frac{p+1}{2}\,\frac{\alpha-\lambda_{1}(\Omega)}{C_{N,p} \alpha^{\frac{N(p-1)}{4}}-|\Omega|^{-\frac{p-1}{2}}} \geq \frac{p+1}{2C_{N,p}}\,\sup_{\alpha>\lambda_1(\Omega)}\frac{\alpha-\lambda_{1} (\Omega)}{\alpha^{\beta}}, \] where $\beta:=N(p-1)/4$. Now, if $\beta\leq1$ we obtain the desired bound for the subcritical and critical cases. On the other hand, when $\beta>1$, elementary calculations show that \[ \hat\mu_1\left(\Omega,p\right)\geq \frac{p+1}{2C_{N,p}} \,\frac{(\beta-1)^{(\beta-1)}}{\beta^\beta}\, \lambda_1(\Omega)^{-(\beta-1)}, \] and finally \[ \hat\rho_1\left(\Omega,p\right)\geq \underbrace{\left[\frac{p+1}{2C_{N,p}} \,\frac{(\beta-1)^{(\beta-1)}}{\beta^\beta}\right]^{\frac{2}{p-1}}}_{D_{N,p}}\, \lambda_1(\Omega)^{\frac{2}{p-1}-\frac{N}{2}}.\qedhere \] \end{proof} \begin{proof}[Proof of Proposition \ref{thm:intro_3>1}] As usual, by \eqref{eq:main_prob_u}, we have to prove that \[ \hat\mu_3\left(\Omega,p\right)\geq 2^{(p-1)/2} D_{N,p}\lambda_3(\Omega)^{\frac{2}{p-1}-\frac{N}{2}}. \] By Lemmas \ref{lem:ckMak}, \ref{lem:M3vsM1}, and Remark \ref{rem:MvsCNp} we obtain \[ \begin{split} \hat\mu_3 &= \sup_{\alpha>\lambda_3(\Omega)} \frac{p+1}{2}\,\frac{\alpha-\lambda_{3}(\Omega)}{M_{\alpha,3}-|\Omega|^{-\frac{p-1}{2}}} \geq \sup_{\alpha>\lambda_3(\Omega)} \frac{p+1}{2}\,\frac{\alpha-\lambda_{3}(\Omega)}{ 2^{-(p-1)/2}M_{\alpha,1}-|\Omega|^{-\frac{p-1}{2}}}\\ &\geq 2^{(p-1)/2}\sup_{\alpha>\lambda_3(\Omega)} \frac{p+1}{2}\,\frac{\alpha-\lambda_{3} (\Omega)}{C_{N,p}\alpha^\beta-2^{(p-1)/2}|\Omega|^{-\frac{p-1}{2}}}, \end{split} \] where $\beta:=N(p-1)/4$, and the desired result follows by arguing as in the proof of Theorem \ref{thm:intro_GS}. \end{proof} To conclude this section we prove that in the supercritical case, if $\mu$ is not too large, in addition to $(c_k)_k$ there is a further sequence of critical levels $(\bar c_k)$ of $\mathcal{E}_{\mu}$ constrained to $\mathcal{M}$. For concreteness, let us first consider the case $k=1$: since in such case $c_1$ is a local minimum of ${\mathcal{E}}_\mu$ in $\mathcal{M}$, and ${\mathcal{E}}_\mu$ is unbounded from below in $\mathcal{M}$, the critical level $\bar c_1$ is of mountain pass type. \begin{proposition} \label{mpcritlev} Let $p>1+4/N$, $\mu<\hat\mu_1$, and $u_1$ denote the local minimum point of ${\mathcal{E}}_\mu$ in $\mathcal{M}$, according to Theorems \ref{infsupteo} and \ref{thm:intro_GS}. The value \[ \bar c_1 : =\inf_{\gamma\in \Gamma}\sup_{[0,1]}\mathcal{E}_{\mu}(\gamma(s)), \quad\text{where }\Gamma:=\left\{\gamma\in C([0,1];{\mathcal{M}}) : \gamma(0)=u_1,\,\gamma(1)<c_1-1\right\}, \] is a critical level for ${\mathcal{E}}_\mu$ in $\mathcal{M}$. \end{proposition} \begin{proof} Notice that, if $p>1+4/N$, then $\mathcal{E}_{\mu}\to -\infty$ along some sequence in ${\mathcal{M}}$. Indeed, by defining \begin{equation} \label{concfun} w_n(x): = \eta(x)Z_{N,p}\big ((x-x_0)/a_n\big )\quad \text{and}\quad \tilde{w_n}:=\frac{w_n}{\| w_n\|^2_{L^2(\Omega)}}\,\in\, {\mathcal{M}}, \end{equation} where $a_n\to 0^+$, $x_0\in \Omega$ and $\eta\in \mathcal{C}_0^{\infty}(\Omega)$, $\eta(x_0)=1$, we obtain \begin{equation}\label{minusinfty} \alpha_n :=\|\nabla \tilde w_n\|^2_{L^2(\Omega)}\to +\infty, \qquad \frac{\int_{\Omega} |\tilde w_n|^{p+1} \,dx}{\alpha_n^{N(p-1)/4}}\to C_{N,p}, \qquad \mathcal{E}_{\mu}(\tilde w_n)\to -\infty \end{equation} for $n\to +\infty$. Since $u_1$ is a local minimum, the functional $\mathcal{E}_{\mu}$ has a mountain pass structure on ${\mathcal{M}}$; by recalling that $\mathcal{E}_{\mu}$ satisfies the P.S. condition the proposition follows. \end{proof} \begin{remark}\label{rem:further_crit_lev} One can generalize Proposition \ref{mpcritlev} by constructing critical points via a saddle-point theorem in the following way: let us pick $k$ points $x_1, x_2,...,x_k$ in $\Omega$ and consider the corresponding function $\tilde w_i$; we may assume that $\mathrm{supp}\,\tilde w_i\cap \mathrm{supp}\,\tilde w_j=\emptyset$ for $i\neq j$, so that these functions are orthogonal. Let us now define the subspace $V_k=\mathrm{span}\{\varphi_1,\dots,\varphi_k;\tilde w_1,...,\tilde w_k\}$; note that dim $V_k=2k$. Let $R$ be an operator (in $L^2(\Omega)$) such that $R=I$ on $V_k^{\perp}$, $Ru_i=\tilde w_i$, $i=1,2,..,k$. Possibly after permutations, we can choose $R$ such that $R\big |_{V_k}\in SO(2k)$ (actually, there are infinitely many different choices of $R$). Now, since $SO(2k)$ is (arcwise) connected, there is a continuous path $\tilde{\gamma}:\,[0,1]\rightarrow SO(2k)$ such that $\gamma(0)=I$, $\gamma(1)=R\big |_{V_k}$. Then, we can define the following map \[ \gamma:\, [0,1] \times S^{k-1}\rightarrow {\mathcal{M}}, \quad\quad \gamma(s;t_1,....,t_k)=\sum_{i=1}^{k}t_i\tilde{\gamma}(s)u_i,\quad \] where $\sum_{i=1}^{k}t_i^2=1$. It is clear that $\gamma$ is continuous; moreover, \[ \gamma(0;t_1,....,t_k)\in \mathrm{span }\{\varphi_1,\dots,\varphi_k\}\cap {\mathcal{M}} \quad \mathrm{and}\quad \gamma(1;t_1,....,t_k)\in \mathrm{span }\{\tilde w_1,\dots,\tilde w_k\}\cap {\mathcal{M}}. \] Then, by denoting with $\Gamma_k$ the set of the above paths, if $\mu$ is sufficiently small we obtain the critical levels \[ \bar c_k : =\inf_{\gamma\in \Gamma_k}\sup_{[0,1]\times S^{k-1}}\mathcal{E}_{\mu}(\gamma(s;t_1,....,t_k)). \] \end{remark} \section{Results in symmetric domains}\label{sec:symm} This section is devoted to the proof of Theorem \ref{pro:symm}, therefore we assume $1+4/N \leq p < 2^*-1$. We perform the proof in the case of $\Omega=B$, but it will be clear that the main assumption on $\Omega$ is the following: \begin{itemize} \item[\textbf{(T)}] there is a tiling of $\Omega$, made by $h$ copies of a subdomain $D$, in such a way that from any solution $U_D$ of \eqref{eq:main_prob_U} on $D$ one can construct, using reflections, a solution $U_\Omega$ of \eqref{eq:main_prob_U} on $\Omega$. \end{itemize} Then $U_\Omega$ has $h$ times the mass of $U_D$, and recalling Theorem \ref{thm:intro_GS} we deduce that \eqref{eq:main_prob_U} on $\Omega$ is solvable for any $\rho< h \cdot D_{N,p} \lambda_1(D)^{\frac{2}{p-1}-\frac{N}{2}}$. At this point, for a sequence $(D_k,h_k)_k$ of tilings satisfying \textbf{(T)}, we obtain the solvability of \eqref{eq:main_prob_U} on $\Omega$ whenever \[ \rho< h_k \cdot D_{N,p} \lambda_1(D_k)^{\frac{2}{p-1}-\frac{N}{2}}, \] and if we can show that \begin{equation}\label{eq:finaltarget} \frac{ h_k }{ \lambda_1(D_k)^{\frac{N}{2}-\frac{2}{p-1}}} \to +\infty\qquad\text{as }k\to+\infty, \end{equation} we deduce the solvability of \eqref{eq:main_prob_U} on $\Omega$ for every mass. Having this scheme in mind, it is easy to prove analogous results on rectangles and also in other kind of domains. Then let $B\subset{\mathbb{R}}^N$ be the ball (w.l.o.g. of radius one), and let \[ D_k:=\left\{(r\cos\theta, r\sin\theta,x_3,\dots,x_N)\in B: - \frac{\pi}{k} < \theta < \frac{\pi}{k}\right\} \] Then $D_k$ satisfies \textbf{(T)}, with $h_k=k$. In order to estimate $\lambda_1(D_k)$ we observe that, by elementary trigonometry, \[ B'_k = B_{\frac{\sin(\pi/k)}{\sin(\pi/k)+1}}\left(\frac{1}{\sin(\pi/k)+1},0,0,\dots,0\right) \subset D_k, \] and therefore \[ \lambda_1(D_k) \le \lambda_1(B'_k) \le C k^2, \] for some dimensional constant $C=C(N)$ and $k$ large. Then \[ \frac{ h_k }{ \lambda_1(D_k)^{\frac{N}{2}-\frac{2}{p-1}}}\ge C \frac{k}{k^{{N}-\frac{4}{p-1}}} = C k^{1-{N}+\frac{4}{p-1}} = C k^{\frac{N-1}{p-1}\left[1+\frac{4}{N-1} - p\right]}, \] and finally \eqref{eq:finaltarget} holds true whenever $p< 1+\frac{4}{N-1}$, thus completing the proof of Theorem \ref{pro:symm}. \small \subsection*{Acknowledgments} We would like to thank Jacopo Bellazzini, who pointed out that the results in \cite{MR3318740}, in the supercritical case, can be read in terms of a local minimization. We would also like to thank Benedetta Noris, who read a preliminary version of this manuscript. This work is partially supported by the PRIN-2012-74FYK7 Grant: ``Variational and perturbative aspects of nonlinear differential problems'', by the ERC Advanced Grant 2013 n. 339958: ``Complex Patterns for Strongly Interacting Dynamical Systems - COMPAT'', and by the INDAM-GNAMPA group.
bdfca2090653522b89d9415dc698595c4bfce1ec
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} Despite the immense popularity and availability of online video content via outlets such as Youtube and Facebook, most work on object detection focuses on static images. Given the breakthroughs of deep convolutional neural networks for detecting objects in static images, the application of these methods to video might seem straightforward. However, motion blur and compression artifacts cause substantial frame-to-frame variability, even in videos that appear smooth to the eye. These attributes complicate prediction tasks like classification and localization. Object-detection models trained on images tend not to perform competitively on videos owing to domain shift factors \cite{KalogeitonFS15}. Moreover, object-level annotations in popular video data-sets can be extremely sparse, impeding the development of better video-based object detection models. Girshik \emph{et al}\bmvaOneDot \cite{RCNN_girshick14CVPR} demonstrate that even given scarce labeled training data, high-capacity convolutional neural networks can achieve state of the art detection performance if first pre-trained on a related task with abundant training data, such as 1000-way ImageNet classification. Followed the pretraining, the networks can be fine-tuned to a related but distinct domain. Also relevant to our work, the recently introduced models Faster R-CNN \cite{Faster_RCNN_RenHG015} and You Look Only Once (YOLO) \cite{YOLO_RedmonDGF15} unify the tasks of classification and localization. These methods, which are accurate and efficient, propose to solve both tasks through a single model, bypassing the separate object proposal methods used by R-CNN \cite{RCNN_girshick14CVPR}. In this paper, we introduce a method to extend unified object recognition and localization to the video domain. Our approach applies transfer learning from the image domain to video frames. Additionally, we present a novel recurrent neural network (RNN) method that refines predictions by exploiting contextual information in neighboring frames. In summary, we contribute the following: \begin{itemize} \item A new method for refining a video-based object detection consisting of two parts: (i) a \emph{pseudo-labeler}, which assigns provisional labels to all available video frames. (ii) A recurrent neural network, which reads in a sequence of provisionally labeled frames, using the contextual information to output refined predictions. \item An effective training strategy utilizing (i) category-level weak-supervision at every time-step, (ii) localization-level strong supervision at final time-step (iii) a penalty encouraging prediction smoothness at consecutive time-steps, and (iv) similarity constraints between \emph{pseudo-labels} and prediction output at every time-step. \item An extensive empirical investigation demonstrating that on the YouTube Objects \cite{youtube-Objects} dataset, our framework achieves mean average precision (mAP) of $68.73$ on test data, compared to a best published result of $37.41$ \cite{Tripathi_WACV16} and $61.66$ for a domain adapted YOLO network \cite{YOLO_RedmonDGF15}. \end{itemize} \section{Methods} \label{sec:method} In this work, we aim to refine object detection in video by utilizing contextual information from neighboring video frames. We accomplish this through a two-stage process. First, we train a \emph{pseudo-labeler}, that is, a domain-adapted convolutional neural network for object detection, trained individually on the labeled video frames. Specifically, we fine-tune the YOLO object detection network \cite{YOLO_RedmonDGF15}, which was originally trained for the 20-class PASCAL VOC \cite{PASCAL_VOC} dataset, to the Youtube-Video \cite{youtube-Objects} dataset. When fine-tuning to the 10 sub-categories present in the video dataset, our objective is to minimize the weighted squared detection loss (equation \ref{eqn:obj_det_loss}) as specified in YOLO \cite{YOLO_RedmonDGF15}. While fine-tuning, we learn only the parameters of the top-most fully-connected layers, keeping the $24$ convolutional layers and $4$ max-pooling layers unchanged. The training takes roughly 50 epochs to converge, using the RMSProp \cite{RMSProp} optimizer with momentum of $0.9$ and a mini-batch size of $128$. As with YOLO \cite{YOLO_RedmonDGF15}, our fine-tuned $pseudo-labeler$ takes $448 \times 448$ frames as input and regresses on category types and locations of possible objects at each one of $S \times S$ non-overlapping grid cells. For each grid cell, the model outputs class conditional probabilities as well as $B$ bounding boxes and their associated confidence scores. As in YOLO, we consider a \emph{responsible} bounding box for a grid cell to be the one among the $B$ boxes for which the predicted area and the ground truth area shares the maximum Intersection Over Union. During training, we simultaneously optimize classification and localization error (equation \ref{eqn:obj_det_loss}). For each grid cell, we minimize the localization error for the \emph{responsible} bounding box with respect to the ground truth only when an object appears in that cell. Next, we train a Recurrent Neural Network (RNN), with Gated Recurrent Units (GRUs) \cite{Cho14_GRU}. This net takes as input sequences of \emph{pseudo-labels}, optimizing an objective that encourages both accuracy on the target frame and consistency across consecutive frames. Given a series of \emph{pseudo-labels} $\mathbf{x}^{(1)}, ..., \mathbf{x}^{(T)}$, we train the RNN to generate improved predictions $\hat{\mathbf{y}}^{(1)}, ..., \hat{\mathbf{y}}^{(T)}$ with respect to the ground truth $\mathbf{y}^{(T)}$ available only at the final step in each sequence. Here, $t$ indexes sequence steps and $T$ denotes the length of the sequence. As output, we use a fully-connected layer with a linear activation function, as our problem is regression. In our final experiments, we use a $2$-layer GRU with $150$ nodes per layer, hyper-parameters determined on validation data. The following equations define the forward pass through a GRU layer, where $\mathbf{h}^{(t)}_l$ denotes the layer's output at the current time step, and $\mathbf{h}^{(t)}_{l-1}$ denotes the previous layer's output at the same sequence step: \begin{equation} \label{eqn:GRU} \begin{aligned} \mathbf{r}^{(t)}_l &= \sigma(\mathbf{h}^{(t)}_{l-1}W^{xr}_l + \mathbf{h}^{(t-1)}_lW^{hr}_l + \mathbf{b}^r_l)\\ \mathbf{u}^{(t)}_l &= \sigma(\mathbf{h}^{(t)}_{l-1}W^{xu}_l + \mathbf{h}^{(t-1)}_lW^{hu}_l + \mathbf{b}^u_l)\\ \mathbf{c}^{(t)}_l &= \sigma(\mathbf{h}^{(t)}_{l-1}W^{xc}_l + r_t \odot(\mathbf{h}^{(t-1)}_lW^{hc}_l) + \mathbf{b}^c_l)\\ \mathbf{h}^{(t)}_l &= (1-\mathbf{u}^{(t)}_l)\odot \mathbf{h}^{(t-1)}_l + \mathbf{u}^{(t)}_l\odot \mathbf{c}^{(t)}_l \end{aligned} \end{equation} Here, $\sigma$ denotes an element-wise logistic function and $\odot$ is the (element-wise) Hadamard product. The reset gate, update gate, and candidate hidden state are denoted by $\textbf{r}$, $\textbf{u}$, and $\textbf{c}$ respectively. For $S = 7$ and $B=2$, the pseudo-labels $\mathbf{x}^{(t)}$ and prediction $\hat{\mathbf{y}}^{(t)}$ both lie in $\mathbb{R}^{1470}$. \vspace{-2.5mm} \subsection{Training} We design an objective function (Equation \ref{eqn:objective}) that accounts for both accuracy at the target frame and consistency of predictions across adjacent time steps in the following ways: \begin{equation} \label{eqn:objective} \mbox{loss} = \mbox{d\_loss} + \alpha \cdot \mbox{s\_loss} + \beta \cdot \mbox{c\_loss} + \gamma \cdot \mbox{pc\_loss} \end{equation} Here, d\_loss, s\_loss, c\_loss and pc\_loss stand for detection\_loss, similarity\_loss, category\_loss and prediction\_consistency\_loss described in the following sections. The values of the hyper-parameters $\alpha=0.2$, $\beta=0.2$ and $\gamma=0.1$ are chosen based on the detection performance on the validation set. The training converges in 80 epochs for parameter updates using RMSProp \cite{RMSProp} and momentum $0.9$. During training we use a mini-batch size of $128$ and sequences of length $30$. \subsubsection{Strong Supervision at Target Frame} On the final output, for which the ground truth classification and localization is available, we apply a multi-part object detection loss as described in YOLO \cite{YOLO_RedmonDGF15}. \vspace{-2.5mm} \begin{equation} \label{eqn:obj_det_loss} \begin{aligned} \mbox{detection\_loss} &= \lambda_{coord}\sum^{S^2}_{i=0}\sum^{B}_{j=0}\mathbbm{1}^{obj}_{ij}\big(\mathit{x}^{(T)}_i - \hat{\mathit{x}}^{(T)}_i\big)^2 + \big(\mathit{y}^{(T)}_i - \hat{\mathit{y}}^{(T)}_i\big)^2 \\ & + \lambda_{coord}\sum^{S^2}_{i=0}\sum^{B}_{j=0}\mathbbm{1}^{obj}_{ij}\big(\sqrt{w_i}^{(T)} - \sqrt{\hat{w}^{(T)}_i}\big)^2 + \big (\sqrt{h_i}^{(T)} - \sqrt{\hat{h}^{(T)}_i} \big)^2 \\ & + \sum^{S^2}_{i=0}\sum^{B}_{j=0}\mathbbm{1}^{obj}_{ij}(\mathit{C}_i - \hat{\mathit{C}_i})^2 \\ & + \lambda_{noobj}\sum^{S^2}_{i=0}\sum^{B}_{j=0}\mathbbm{1}^{noobj}_{ij}\big(\mathit{C}^{(T)}_i - \hat{\mathit{C}}^{(T)}_i\big)^2 \\ & + \sum^{S^2}_{i=0}\mathbbm{1}^{obj}_{i}\sum_{c \in classes}\big(p_i^{(T)}(c) - \hat{p_i}^{(T)}(c)\big)^2 \end{aligned} \end{equation} where $\mathbbm{1}^{obj}_{i}$ denotes if the object appears in cell $i$ and $\mathbbm{1}^{obj}_{ij}$ denotes that $j$th bounding box predictor in cell $i$ is \emph{responsible} for that prediction. The loss function penalizes classification and localization error differently based on presence or absence of an object in that grid cell. $x_i, y_i, w_i, h_i$ corresponds to the ground truth bounding box center coordinates, width and height for objects in grid cell (if it exists) and $\hat{x_i}, \hat{y_i}, \hat{w_i}, \hat{h_i}$ stand for the corresponding predictions. $C_i$ and $\hat{C_i}$ denote confidence score of \emph{objectness} at grid cell $i$ for ground truth and prediction. $p_i(c)$ and $\hat{p_i}(c)$ stand for conditional probability for object class $c$ at cell index $i$ for ground truth and prediction respectively. We use similar settings for YOLO's object detection loss minimization and use values of $\lambda_{coord} = 5$ and $\lambda_{noobj} = 0.5$. \vspace{-2.5mm} \subsubsection{Similarity between \emph{Pseudo-labels} and Predictions} Our objective function also includes a regularizer that penalizes the dissimilarity between \emph{pseudo-labels} and the prediction at each time frame $t$. \vspace{-2.5mm} \begin{equation} \label{auto_enc_loss} \mbox{similarity\_loss} = \sum^T_{t=0}\sum^{S^2}_{i=0}\hat{C}^{(t)}_i\Big(\mathbf{x}^{(t)}_i - \hat{\mathbf{y}_i}^{(t)} \Big)^2 \end{equation} Here, $\mathbf{x}^{(t)}_i$ and $\hat{\mathbf{y}_i}^{(t)}$ denote the \emph{pseudo-labels} and predictions corresponding to the $i$-th grid cell at $t$-th time step respectively. We perform minimization of the square loss weighted by the predicted confidence score at the corresponding cell. \subsubsection{Object Category-level Weak-Supervision} Replication of the static target at each sequential step has been shown to be effective in \cite{LiptonKEW15, yue2015beyond, dai2015semi}. Of course, with video data, different objects may move in different directions and speeds. Yet, within a short time duration, we could expect all objects to be present. Thus we employ target replication for classification but not localization objectives. We minimize the square loss between the categories aggregated over all grid cells in the ground truth $\mathbf{y}^{(T)}$ at final time step $T$ and predictions $\hat{\mathbf{y}}^{(t)}$ at all time steps $t$. Aggregated category from the ground truth considers only the cell indices where an object is present. For predictions, contribution of cell $i$ is weighted by its predicted confidence score $\hat{C}^{(t)}_i$. Note that cell indices with positive detection are sparse. Thus, we consider the confidence score of each cell while minimizing the aggregated category loss. \vspace{-2.5mm} \begin{equation} \label{category_supervision} \mbox{category\_loss} = \sum^T_{t=0}\bigg(\sum_{c \in classes} \Big(\sum^{S^2}_{i=0} \hat{C}^{(t)}_i\big(\hat{p}^{(t)}_i(c)\big) - \sum^{S^2}_{i=0}\mathbbm{1}^{obj^{(T)}}_i \big(p_i^{(T)}(c)\big)\Big) \bigg)^2 \end{equation} \subsubsection{Consecutive Prediction Smoothness} Additionally, we regularize the model by encouraging smoothness of predictions across consecutive time-steps. This makes sense intuitively because we assume that objects rarely move rapidly from one frame to another. \vspace{-2.5mm} \begin{equation} \label{prediction_smoothness} \mbox{prediction\_consistency\_loss} = \sum^{T-1}_{t=0}\Big(\hat{\mathbf{y}_i}^{(t)} - \hat{\mathbf{y}_i}^{(t+1)} \Big)^2 \end{equation} \vspace{-2.5mm} \subsection{Inference} The recurrent neural network predicts output at every time-step. The network predicts $98$ bounding boxes per video frame and class probabilities for each of the $49$ grid cells. We note that for every cell, the net predicts class conditional probabilities for each one of the $C$ categories and $B$ bounding boxes. Each one of the $B$ predicted bounding boxes per cell has an associated \emph{objectness} confidence score. The predicted confidence score at that grid is the maximum among the boxes. The bounding box with the highest score becomes the \emph{responsible} prediction for that grid cell $i$. The product of class conditional probability $\hat{p}^{(t)}_i(c)$ for category type $c$ and \emph{objectness} confidence score $\hat{C}^{(t)}_i$ at grid cell $i$, if above a threshold, infers a detection. In order for an object of category type $c$ to be detected for $i$-th cell at time-step $t$, both the class conditional probability $\hat{p}^{(t)}_i(c)$ and \emph{objectness score} $\hat{C}^{(t)}_i$ must be reasonably high. Additionally, we employ Non-Maximum Suppression (NMS) to winnow multiple high scoring bounding boxes around an object instance and produce a single detection for an instance. By virtue of YOLO-style prediction, NMS is not critical. \section{Experimental Results} \label{sec:results} In this section, we empirically evaluate our model on the popular \textbf{Youtube-Objects} dataset, providing both quantitative results (as measured by mean Average Precision) and subjective evaluations of the model's performance, considering both successful predictions and failure cases. The \textbf{Youtube-Objects} dataset\cite{youtube-Objects} is composed of videos collected from Youtube by querying for the names of 10 object classes of the PASCAL VOC Challenge. It contains 155 videos in total and between 9 and 24 videos for each class. The duration of each video varies between 30 seconds and 3 minutes. However, only $6087$ frames are annotated with $6975$ bounding-box instances. The training and test split is provided. \subsection{Experimental Setup} We implement the domain-adaption of YOLO and the proposed RNN model using Theano \cite{Theano2016arXiv160502688short}. Our best performing RNN model uses two GRU layers of $150$ hidden units each and dropout of probability $0.5$ between layers, significantly outperforming domain-adapted YOLO alone. While we can only objectively evaluate prediction quality on the labeled frames, we present subjective evaluations on sequences. \subsection{Objective Evaluation} We compare our approach with other methods evaluated on the Youtube-Objects dataset. As shown in Table \ref{table:per_category_results} and Table \ref{table:final_mAP}, Deformable Parts Model (DPM) \cite{FelzenszwalbMR_CVPR_2008})-based detector reports \cite{KalogeitonFS15} mean average precision below $30$, with especially poor performance in some categories such as \emph{cat}. The method of Tripathi \emph{et al}\bmvaOneDot (VPO) \cite{Tripathi_WACV16} uses consistent video object proposals followed by a domain-adapted AlexNet classifier (5 convolutional layer, 3 fully connected) \cite{AlexNet12} in an R-CNN \cite{RCNN_girshick14CVPR}-like framework, achieving mAP of $37.41$. We also compare against YOLO ($24$ convolutional layers, 2 fully connected layers), which unifies the classification and localization tasks, and achieves mean Average Precision over $55$. In our method, we adapt YOLO to generate \emph{pseudo-labels} for all video frames, feeding them as inputs to the refinement RNN. We choose YOLO as the \emph{pseudo-labeler} because it is the most accurate among feasibly fast image-level detectors. The domain-adaptation improves YOLO's performance, achieving mAP of $61.66$. Our model with RNN-based prediction refinement, achieves superior aggregate mAP to all baselines. The RNN refinement model using both input-output similarity, category-level weak-supervision, and prediction smoothness performs best, achieving $\mbox{68.73}$ mAP. This amounts to a relative improvement of $\mbox{11.5\%}$ over the best baselines. Additionally, the RNN improves detection accuracy on most individual categories (Table \ref{table:per_category_results}). \begin{table} \label{table:per_category_results} \centering \footnotesize \begin{tabular}{lllllllllll} \multicolumn{11}{c}{\textbf{Average Precision on 10-categories}} \\ \midrule Methods & airplane & bird & boat & car & cat & cow & dog & horse & mbike & train \\ \midrule DPM\cite{FelzenszwalbMR_CVPR_2008} & 28.42 & 48.14 & 25.50 & 48.99 & 1.69 & 19.24 & 15.84 & 35.10 & 31.61 & 39.58 \\ VOP\cite{Tripathi_WACV16} & 29.77 & 28.82 & 35.34 & 41.00 & 33.7 & 57.56 & 34.42 & 54.52 & 29.77 & 29.23 \\ YOLO\cite{YOLO_RedmonDGF15} & 76.67 & 89.51 & 57.66 & 65.52 & 43.03 & 53.48 & 55.81 & 36.96 & 24.62 & 62.03 \\ DA YOLO & \textbf{83.89} & \textbf{91.98} & 59.91 & 81.95 & 46.67 & 56.78 & 53.49 & 42.53 & 32.31 & 67.09 \\ \midrule RNN-IOS & 82.78 & 89.51 & 68.02 & \textbf{82.67} & 47.88 & 70.33 & 52.33 & 61.52 & 27.69 & \textbf{67.72} \\ RNN-WS & 77.78 & 89.51 & \textbf{69.40} & 78.16 & 51.52 & \textbf{78.39} & 47.09 & 81.52 & 36.92 & 62.03 \\ RNN-PS & 76.11 & 87.65 & 62.16 & 80.69 & \textbf{62.42} & 78.02 & \textbf{58.72} & \textbf{81.77} & \textbf{41.54} & 58.23 \\ \bottomrule \end{tabular} \caption{Per-category object detection results for the Deformable Parts Model (DPM), Video Object Proposal based AlexNet (VOP), image-trained YOLO (YOLO), domain-adapted YOLO (DA-YOLO). RNN-IOS regularizes on input-output similarity, to which RNN-WS adds category-level weak-supervision, to which RNN-PS adds a regularizer encouraging prediction smoothness.} \end{table} \begin{table}[h] \label{table:final_mAP} \centering \begin{tabular}{llllllll} \multicolumn{8}{c}{\textbf{mean Average Precision on all categories}} \\ \midrule Methods & DPM & VOP & YOLO & DA YOLO & RNN-IOS & RNN-WS & RNN-PS\\ \midrule mAP & 29.41 & 37.41 & 56.53 & \textbf{61.66} & 65.04 & 67.23 & \textcolor{blue}{\textbf{68.73}}\\ \bottomrule \end{tabular} \caption{Overall detection results on Youtube-Objects dataset. Our best model (RNN-PS) provides $7\%$ improvements over DA-YOLO baseline.} \end{table} \vspace{-2.5mm} \vspace{-2.5mm} \subsection{Subjective Evaluation} We provide a subjective evaluation of the proposed RNN model in Figure \ref{fig:subjective1}. Top and bottom rows in every pair of sequences correspond to \emph{pseudo-labels} and results from our approach respectively. While only the last frame in each sequence has associated ground truth, we can observe that the RNN produces more accurate and more consistent predictions across time frames. The predictions are consistent with respect to classification, localization and confidence scores. In the first example, the RNN consistently detects the \emph{dog} throughout the sequence, even though the \emph{pseudo-labels} for the first two frames were wrong (\emph{bird}). In the second example, \emph{pseudo-labels} were \emph{motorbike}, \emph{person}, \emph{bicycle} and even \emph{none} at different time-steps. However, our approach consistently predicted \emph{motorbike}. The third example shows that the RNN consistently predicts both of the cars while the \emph{pseudo-labeler} detects only the smaller car in two frames within the sequence. The last two examples show how the RNN increases its confidence scores, bringing out the positive detection for \emph{cat} and \emph{car} respectively both of which fell below the detection threshold of the \emph{pseudo-labeler}. \begin{figure*} \begin{center} \includegraphics[scale=0.75]{result2_category_consistency-eps-converted-to.pdf} \includegraphics[scale=0.75]{result3_category_consistency-eps-converted-to.pdf} \includegraphics[scale=0.75]{result1_localizations-eps-converted-to.pdf} \includegraphics[scale=0.75]{result8_detection_through_consistency-eps-converted-to.pdf} \includegraphics[scale=0.75]{result16_detection_through_consistency-eps-converted-to.pdf} \end{center} \caption{ Object detection results from the final eight frames of five different test-set sequences. In each pair of rows, the top row shows the \emph{pseudo-labeler} and the bottom row shows the RNN. In the first two examples, the RNN consistently predicts correct categories \emph{dog} and \emph{motorbike}, in contrast to the inconsistent baseline. In the third sequence, the RNN correctly predicts multiple instances while the \emph{pseudo-labeler} misses one. For the last two sequences, the RNN increases the confidence score, detecting objects missed by the baseline. } \label{fig:subjective1} \end{figure*} \subsection{Areas For Improvement} The YOLO scheme for unifying classification and localization \cite{YOLO_RedmonDGF15} imposes strong spatial constraints on bounding box predictions since each grid cell can have only one class. This restricts the set of possible predictions, which may be undesirable in the case where many objects are in close proximity. Additionally, the rigidity of the YOLO model may present problems for the refinement RNN, which encourages smoothness of predictions across the sequence of frames. Consider, for example, an object which moves slightly but transits from one grid cell to another. Here smoothness of predictions seems undesirable. \begin{figure*} \begin{center} \includegraphics[scale=0.75]{failure_cases-eps-converted-to.pdf} \end{center} \caption{Failure cases for the proposed model. Left: the RNN cannot recover from incorrect \emph{pseudo-labels}. Right: RNN localization performs worse than \emph{pseudo-labels} possibly owing to multiple instances of the same object. } \label{fig:failure_cases} \vspace{-2.5mm} \end{figure*} Figure \ref{fig:failure_cases} shows some failure cases. In the first case, the \emph{pseudo-labeler} classifies the instances as \emph{dogs} and even as \emph{birds} in two frames whereas the ground truth instances are \emph{horses}. The RNN cannot recover from the incorrect pseudo-labels. Strangely, the model increases the confidence score marginally for a different wrong category \emph{cow}. In the second case, possibly owing to motion and close proximity of multiple instances of the same object category, the RNN predicts the correct category but fails on localization. These point to future work to make the framework robust to motion. The category-level weak supervision in the current scheme assumes the presence of all objects in nearby frames. While for short snippets of video this assumption generally holds, it may be violated in case of occlusions, or sudden arrival or departure of objects. In addition, our assumptions regarding the desirability of prediction smoothness can be violated in the case of rapidly moving objects. \vspace{-2.5mm} \section{Related Work} \label{sec:prior-art} Our work builds upon a rich literature in both image-level object detection,video analysis, and recurrent neural networks. Several papers propose ways of using deep convolutional networks for detecting objects \cite{RCNN_girshick14CVPR,fast_RCNN_15,Faster_RCNN_RenHG015, YOLO_RedmonDGF15, SzegedyREA14, Inside_Outside_Net_BellZBG15, DeepID-Net_2015_CVPR, Overfeat_SermanetEZMFL13, CRAFTCVPR16, Gidaris_2015_ICCV}. Some approaches classify the proposal regions \cite{RCNN_girshick14CVPR,fast_RCNN_15} into object categories and some other recent methods \cite{Faster_RCNN_RenHG015, YOLO_RedmonDGF15} unify the localization and classification stages. Kalogeiton \emph{et al}\bmvaOneDot \cite{KalogeitonFS15} identifies domain shift factors between still images and videos, necessitating video-specific object detectors. To deal with shift factors and sparse object-level annotations in video, researchers have proposed several strategies. Recently, \cite{Tripathi_WACV16} proposed both transfer learning from the image domain to video frames and optimizing for temporally consistent object proposals. Their approach is capable of detecting both moving and static objects. However, the object proposal generation step that precedes classification is slow. Prest \emph{et al}\bmvaOneDot \cite{Weak_obj_from_videoPrestLCSF12}, utilize weak supervision for object detection in videos via category-level annotations of frames, absent localization ground truth. This method assumes that the target object is moving, outputting a spatio-temporal tube that captures this most salient moving object. This paper, however, does not consider context within video for detecting multiple objects. A few recent papers \cite{DeepID-Net_2015_CVPR, Inside_Outside_Net_BellZBG15} identify the important role of context in visual recognition. For object detection in images, Bell \emph{et al}\bmvaOneDot \cite{Inside_Outside_Net_BellZBG15} use spatial RNNs to harness contextual information, showing large improvements on PASCAL VOC \cite{PASCAL_VOC} and Microsoft COCO \cite{COCOLinMBHPRDZ14} object detection datasets. Their approach adopts proposal generation followed by classification framework. This paper exploits spatial, but not temporal context. Recently, Kang \emph{et al}\bmvaOneDot \cite{KangCVPR16} introduced tubelets with convolutional neural networks (T-CNN) for detecting objects in video. T-CNN uses spatio-temporal tubelet proposal generation followed by the classification and re-scoring, incorporating temporal and contextual information from tubelets obtained in videos. T-CNN won the recently introduced ImageNet object-detection-from-video (VID) task with provided densely annotated video clips. Although the method is effective for densely annotated training data, it's behavior for sparsely labeled data is not evaluated. By modeling video as a time series, especially via GRU \cite{Cho14_GRU} or LSTM RNNs\cite{LSTM_Hochreiter_97}, several papers demonstrate improvement on visual tasks including video classification \cite{yue2015beyond}, activity recognition \cite{LongTermRecurrentDonahueHGRVSD14}, and human dynamics \cite{Fragkiadaki_2015_ICCV}. These models generally aggregate CNN features over tens of seconds, which forms the input to an RNN. They perform well for global description tasks such as classification \cite{yue2015beyond,LongTermRecurrentDonahueHGRVSD14} but require large annotated datasets. Yet, detecting multiple generic objects by explicitly modeling video as an ordered sequence remains less explored. Our work differs from the prior art in a few distinct ways. First, this work is the first, to our knowledge, to demonstrate the capacity of RNNs to improve localized object detection in videos. The approach may also be the first to refine the object predictions of frame-level models. Notably, our model produces significant improvements even on a small dataset with sparse annotations. \vspace{-2.5mm} \vspace{-2.5mm} \section{Conclusion} We introduce a framework for refining object detection in video. Our approach extracts contextual information from neighboring frames, generating predictions with state of the art accuracy that are also temporally consistent. Importantly, our model benefits from context frames even when they lack ground truth annotations. For the recurrent model, we demonstrate an efficient and effective training strategy that simultaneously employs localization-level strong supervision, category-level weak-supervision, and a penalty encouraging smoothness of predictions across adjacent frames. On a video dataset with sparse object-level annotation, our framework proves effective, as validated by extensive experiments. A subjective analysis of failure cases suggests that the current approach may struggle most on cases when multiple rapidly moving objects are in close proximity. Likely, the sequential smoothness penalty is not optimal for such complex dynamics. Our results point to several promising directions for future work. First, recent state of the art results for video classification show that longer sequences help in global inference. However, the use of longer sequences for localization remains unexplored. We also plan to explore methods to better model local motion information with the goal of improving localization of multiple objects in close proximity. Another promising direction, we would like to experiment with loss functions to incorporate specialized handling of classification and localization objectives.
0139d272d2b2af4606ec78c8e956da717b63bad8
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The wave-particle duality is an alternative statement of the complementarity principle, and it establishes the relation between \ankb{corpuscular and undulatory}{the corpuscular and the ondulatory} nature of quantum entities \cite{Bohr1928}. It can be illustrated in a two-way interferometer, where the apparatus can be set to observe \ankb{particle}{the particle} behavior when a single path is \ankb{taken}{taken,} or wave-like behavior, when the impossibility to define a path is shown by the interference. A modern approach to the wave-particle duality includes quantitative relations between quantities that represent the possible \textit{a priori} knowledge of the which-way information (\ankb{predictability}{predicability}) and the ``quality'' of the interference fringes (\ankb{Visibility}{visibility}). Several publications in the literature \cite{Bohr1928, Wootters1979, Summhammer1987, Greenberger1988, Mandel1991} contributed to the formulation of the quantitative analysis of the wave-particle duality. For a bipartite system\ankb{ entanglement, the quantum correlations between each part, can play a role. Such correlations can}{, entanglement can} give an extra which-way (path) information about the interferometric possibilities. The quantitative relations for systems composed by two particles were \ankb{}{extensively} studied in \cite{Jaeger1993, Jaeger1995, Englert1996, Englert2000, Scully1989, Scully1991, Mandel1995, Tessier2005, Jakob2010, Miatto2015,Bagan2016, Coles2016}. Therefore, \ankb{understand}{understanding} the behavior of such quantities, in various regimes and situations, is essential to answer fundamental and/or technological questions of the quantum theory \cite{Greenberger1999}. \alams{The Complementarity quantities can present interesting dynamical behaviors,}{Concerning the study of the dynamical behavior of complementarity quantities,} an example is the so-called \textit{quantum eraser}, \alams{where an increase or preservation of the visibility of an interferometer experiment is caused when the ``which-way'' information is erased.}{\ankb{ i.e. an increasing or preservation of the \ankb{Visibility}{visibility} in an interferometric scheme (or the ``erasure'' of the which-way information probably stored in the initial state).}{ where an increase or preservation of the visibility of an interferometer experiment is caused when the which-path information is erased.}} Since its proposal \cite{Scully1982}\ankb{ this phenomena}{, it} has \ankb{investigated it carefully,}{been investigated carefully} both theoretically and experimentally (see for example Refs. \cite{Englert2000, Scully1991, Mandel1995, Storey1994, Wiseman1995, Mir2007, Luis1998, Busch2006, Rossi2013, Walborn2002, Mir2007, Teklemariam2001, Teklemariam2002, Kim2000, Salles2008, Heuer2015}). In \ankb{a recent work }{Ref.~}\cite{Rossi2013}, the authors explore the quantum eraser problem in \ankb{multipartite}{a multipartite} model\ankb{, where two cavities ($q_A+q_B$), which will be taken as a two qubit system $A + B$, in an initial maximally entangled state (and therefore with zero \ankb{Visibility}{visibility}), couple through a Jaynes-Cummings Hamiltonian to $N$ two-level atoms (we will call the global system as $q_A + q_B + R$, where all the individual systems are qubits)}{. Initially a bipartite qubit system is prepared in a maximally entangled state and interacts with $N$ other qubits. This model can be implemented considering the qubits of interest the cavity modes of two cavities and the $N$ qubits as two-level atoms}. In this work \cite{Rossi2013}, an increase of visibility is achieved by performing appropriate projective measurements. An intrinsic relation between the complementarity quantities and the performed measurements is outlined: since \ankb{they}{the measurements} were made in order to obtain an \ankb{increasing}{increase} of the \ankb{Visibility}{visibility}, the remaining quantities (Entanglement as measured by the concurrence, and the predictability) must obey a ``complementary'' behavior. In that case, \ankb{Visibility}{visibility} and predictability increases, and entanglement decreases, since the measurements are made in order to \ankb{establishes}{establish} the quantum eraser. In Reference \cite{Rossi2013} only the maximization of the visibility was considered, in the present work we extend the analysis and consider maximization of predictability, visibility and concurrence. Also, in the previous work \cite{Rossi2013}, only one value of the coupling constant was considered. In this contribution we consider a second coupling regime that allows for the comparison between stronger and weaker interactions. Some questions may arise from the analysis presented in \cite{Rossi2013}: how is the behavior of the \ankb{Visibility}{visibility}, predictability and Entanglement, for different strengths of the coupling between the cavities and the $N$ atoms? \ankb{There are any difference in this behavior,}{Is there any difference in this behavior} if one measure the qubits in order to maximize another complementarity quantity? For finite $N$, could \alams{this behavior}{the behavior of entanglement} resemble the reservoir (dissipative) limit? Moreover, one can think about a three-part control scheme: initially parts $A$ and $B$ possesses a maximally entanglement state, constituted by two qubits $q_A$ and $q_B$, respectively. A third part $R$ may have, in principle, \emph{full} control of a group of $N$-qubits (each one we call as $q_i$); i.e. $R$ may control: (i) the initial state of each qubit $q_i$, (ii) the interaction strength between $q_i$ and $q_B$ and (iii) the measurement basis where each $q_i$ could be projected by $R$. Here we will focus in the control of item (iii), therefore the initial state of all $q_i$ and the coupling strenght will be fixed for each realization of the scheme. Thus, part $R$ is able to control which complementarity quantity of part $A$ they ($A$ and $R$) would like to maximize. For instance, if $A$ and $R$ desire that $q_A$ is in a superposition state, $R$ can choose which basis he/she will project each qubit in order to accomplish the task (quantum eraser task \cite{Rossi2013}). However, now $R$ and $A$ are able to choose another complementarity quantity: if they would like to obtain and/or maintain an Entangled state between $A$ and $B$, $R$ may project each $q_i$ in a basis chosen in order obtain an state nearly maximally Entangled (the same idea follows for the predictability). More than that, since part $R$ can adjust the strenght of the interaction between $q_B$ and $q_i$, he/she can study what is the best option of coupling to do each task (together with the freedom to choose the basis of projection). In that way, parts $A$, $B$ and $R$ are able to study in details the behavior of the complementarity quantities, for a variety of conditions. In the present work we answer the questions and provide a useful tool to implement the control scheme mentioned above, considering a similar model\ankb{: the interaction between}{ compound by} two entangled qubits, $q_{A}$ and $q_{B}$, and a third system ($R$) which is composed by $N$ qubits. \ankb{They interact, one at the time, with the qubit B. The $N$ qubits of $R$ can be measured after the interaction.}{Each qubit of $R$ interacts one at a time with only qubit $B$ and can be projectively measured afterwards.} It is well known that \ankb{when the interaction time $t\rightarrow 0$ and on the coupling strength $g\rightarrow \infty$ the subsystem}{in the limit that the interaction time $t\rightarrow 0$ and that the coupling strength $g\rightarrow \infty$, the system} $R$ will play the role of a reservoir \cite{Carmichael1999, Breuer2007, Jacobs1998}. As it is possible to measure each qubit of system $R$ after the interaction, we can control the evolution of $q_{A}$ and $q_{B}$, induced by the interaction with $R$, by selecting an adequate sequence of results of measurements performed in the qubits of $R$. Such control would allow us to make $q_{A}$ and $q_{B}$ approach a chosen asymptotic state. This scheme can be implemented in cavity-QED system, where $q_{A}$ and $q_{B}$ would be cavity modes, prepared in an entangled state with one excitation, and $N$ two level atoms, interacting with the cavities one at the time, would play the role of the qubits that compose the system $R$. We consider the complementarity quantities \cite{Jakob2010} concurrence, predictability and \ankb{Visibility}{visibility} to guide the manipulation over $R$ and to quantify the information present in each subsystem. \ankb{To manipulate the evolution we consider three kinds of sequences of experimental results: The first maximizes the Visibility, the second maximizes the predictability and the third maximizes the concurrence.}{Each quantity is maximized by a different set of projective measurements on $R$.} We also consider two regimes for the strength of the coupling constant $g$ between $q_B$ and each qubit of $R$, $g= \frac{1}{4}$ and $g= 4$. \ankb{We perform a numerical calculation in order to find the right sequence of measurement that lead to the maximization of a given quantity.}{} \alams{We show that for $g= \frac{1}{4}$ it is possible to manipulate the evolution to make the subsystem of interest $q_{A}+q_{B}$ to approach a chosen asymptotic state. For $g T = 2 \pi \times 4$, the subsystem $q_{A}+q_{B}$ always tends to a state with no excitation in $q_{B}$, independently of the chosen sequence of measurement results in $R$. Therefore, when the coupling constant increases, the result of the interaction between the system of interest and $R$ is similar to the interaction with a thermal reservoir. We consider also the case with no maximization, assuming that \ankb{the}{} all measurements are made in the same basis while we observe the complementarity quantities behavior. Finally we show how the information is distributed over the global system, and how this distribution changes after the measurements in $R$.}{\textcolor[rgb]{0.00,0.00,0.00}{We show that for $g = \frac{1}{4}$ it is possible to maximize the concurrence of the subsystem $q_A+q_B$, while for $g T= 2 \pi \times 4$ (i.e. $g = 4$) the concurrence decays quickly and the maximization is not possible. When the coupling constant increases, the behavior of concurrence is similarly to the one expected if the system $R$ had the properties of a thermal reservoir. However, the visibility shows a different behavior, its maximization is more efficient for $g = 4$. This behavior is caused by the different which-way information distribution produced by the interactions with $g = 4$ and $g=\frac{1}{4}$, as we shown in the last section. Numerical calculation shows that, for $g = 4$, the first two qubits of $R$ retain a large amount of which-way information, that was initially present in $q_B$. When measurements that maximizes the visibility are performed, the which-way information is erased, and the visibility of $q_A$ increases quickly. For $g=\frac{1}{4}$ the which-way information is distributed almost equally among the qubits of $R$, therefore less information is erased and consequently measurements that maximize the visibility are less efficient.}} The paper is organized as \ankb{follow}{follows}: in section \ref{model} we \ankb{present the model in details, including the complete dynamics of the global system and}{briefly review the model and the definition of} the principal quantities studied: \ankb{Visibility}{visibility}, concurrence and predictability. \ankb{We also study in this section}{Moreover, we analyze} the distinguishability between different parts of the global system\ankb{ ($q_A$ and $q_B$, $q_A$ and the i-esim qubit of $R$)}{}. In subsection \ref{digression} we briefly review the case where $q_A+q_B$ interacts in a dissipative reservoir, and how the complementarity quantities behave in this case. Section \ref{results} shows how we implement the projective measurements in $R$, and present our results and discussions for the behavior of the complementarity quantities and for the variation of distinguishability (after and before the measurements). In section \ref{conclusion} we conclude our work. \begin{figure}[h] \centering {\includegraphics[scale=0.32]{figure_1.pdf}} \caption{(Color online) A schematic figure of our proposal. The qubits $q_A$ and $q_B$ are initially Entangled (orange squares), and part $R$ (which contains $N$ qubits, represented by green circles) interacts with $q_B$, throught the Hamiltonian indicated in the text (represented by the pink curve). The projective measures follows after $R$ decides which complementarity quantity of $A$ and $B$ will be maximized.} \label{scheme} \end{figure} \section{Model and definitions}\label{model} Let us consider that initially qubits $q_A$ and $q_B$ were prepared in the entangled state $\ket{\psi(0)} = \frac{1}{\sqrt 2} (\ket{0_A 1_B} + \ket{1_A 0_B})$ and a third system $R$ composed by \ankb{$N$-qubits}{$N$ qubits}, prepared all in the ground state $\ket 0$. Each qubit of $R$ interacts one at a time with $q_B$ and after a sequence of these interactions, the information, initially stored in $q_A$ and $q_B$, will be distributed over the \ankb{$N$-qubits of $R$}{$N$ qubits of $R$} and qubits $q_A$ and $q_B$. As an example of our interaction model, consider the following dynamics governing the interaction of an atom (between a total of $N$ atoms) and a cavity ($q_B$) (see Figure \ref{scheme}). The Hamiltonian that gives the interaction between the $k$-th atom and $q_B$ is $\hat{H}^{(k)}= \omega \hat{b}^{\dagger }\hat{b}+ \frac{\omega}{2}\hat{\sigma}_{z}^{(k)}+ g (\hat{b}^{\dagger}\hat{\sigma}_{-}^{(k)}+\hat{b}\hat{\sigma}_{+}^{(k)}),$ where $\hat{b}^{\dagger }$ ($\hat{b}$) corresponds to the creation (annihilation) operator for $q_B$, $\omega $ their transition frequency, $\hat{\sigma}_{z}^{(k)}=|1^{(k)}\rangle\langle 1^{(k)}|-|0^{(k)}\rangle\langle 0^{(k)}|$, $\hat{\sigma}_{-}^{(k)}=|0^{(k)}\rangle\langle 1^{(k)}|$, $\hat{\sigma}_{+}^{(k)}=|1^{(k)}\rangle\langle 0^{(k)}|$, \ankb{}{and} $g$ the coupling constant for the interaction between the $k$-th qubit of $R$ and $q_B$. As the initial state has one excitation and the Hamiltonian preserves the excitation number, the states of mode $B$ can be written in the basis $\lbrace\vert 0 \rangle , \vert 1 \rangle\rbrace$. Although constant in each preparation, we let the parameter $g$ free in order to quantify the strength of the interaction, since we will analyze different coupling regimes and the correspondent behavior of the Complementarity quantities. In this model, $\left|0^{(k)} \right\rangle$ and $\left|1^{(k)}\right\rangle$ stand for the levels $0$ and $1$ of the $k$-th interacting atom, respectively. After $n$ qubits of $R$ have interacted with $q_B$ (note the difference between $N$, the total number of qubits that are able to interact, and $n$, the number of qubits that will interact at a given time) the global system is left in the state \cite{Rossi2013} \begin{eqnarray} \ket{\psi^{(n)}} &=& \frac{1}{\sqrt 2} \Big( a^n \ket{0_{A}} \ket{0_{res}} \ket{1_{B}} + \Gamma \ket{0_{A}} \ket{1_{res}} \ket{0_{B}} + \nonumber \\ && + \ket{1_{A}} \ket{0_{res}} \ket{0_{B}} \Big), \label{psin}\end{eqnarray} where $a = \cos\left( \frac{g T}{N} \right)$ and $b = - i \sin \left( \frac{g T}{N} \right)$, assuming the same interaction time between each qubit of $R$ and $q_{B}$, given by $\Delta t=\frac{T}{N}$. To simplify the notation we define a normalized state with one excitation in subsystem $R$: $\ket{1_{res}}= \frac{1}{\Gamma} \left( a^{n-1} b \ket{n} + \ldots + a b \ket{2} + b\ket{1} \right)$\ankb{ with}{, where} $\Gamma^{2} = 1 - a^{2n}$\ankb{ and}{,} $\ket{i}= \ket{0_1 0_2 \ldots 0_{i-1} ~1_i ~ 0_{i+1} \ldots 0_n}$ represents a state with an excitation in the $i$-th qubit and $\ket{0_{res}} = \ket{0_1 0_2 \ldots 0_n}$. For a general pure two qubit state \cite{Englert1996, Englert2000, Jakob2010}: $\ket{\Psi} = \gamma_1 \ket{00} + \gamma_2 \ket{01} + \gamma_3 \ket{10} + \gamma_4 \ket{11},$ the complementary quantities are defined in the following way. The concurrence, which is related to the quantum correlation between the parts, is given by $C(\ket{\Psi}) = 2 \abs{\gamma_1 \gamma_4 - \gamma_2 \gamma_3}$. The coherence between two orthogonal states gives the visibility, defined by $V = 2 \abs{\bra{\Psi}\sigma_{+}\ket{\Psi}} = 2 \abs{\gamma_1 \gamma_3^* + \gamma_2 \gamma_4^*}$. Besides, the predictability measures the knowledge if one of the parts is in state $\ket{0}$ or $\ket{1}$, $P= \abs{\bra{\Psi}\sigma_{z}\ket{\Psi}} = \abs{\abs{\gamma_3}^2+\abs{\gamma_4}^2- \left( \abs{\gamma_1}^2 + \abs{\gamma_2}^2 \right)}$. The distinguishability, or the ``measure of the possible which-path information that one can obtain'' in an interferometer setup \cite{Englert2000, Jakob2010}, is given by $D = \sqrt{C^2 + P^2}$. \alams{}{For completeness, we present here explicitly the global system state operator: \begin{widetext} \begin{eqnarray} \rho^{(n)} &=& \ket{\psi^{n}}\bra{\psi^{n}} = \frac{1}{2} \Big( a^{2n} \ket{0_{A} ~ 0_{res} ~ 1_{B}} \bra{0_{A} ~ 0_{res} ~ 1_{B}} + \Gamma^2 \ket{0_{A} ~ 1_{res} ~ 0_{B}} \bra{0_{A} ~ 1_{res} ~ 0_{B}} + \ket{1_{A} ~ 0_{res} ~ 0_{B}} \bra{1_{A} ~ 0_{res} ~ 0_{B}} + \nonumber \\ && + a^n \Gamma^* \ket{0_{A} ~ 0_{res} ~ 1_{B}} \bra{0_{A} ~ 1_{res} ~ 0_{B}} + a^n \ket{0_{A} ~ 0_{res} ~ 1_{B}} \bra{1_{A} ~ 0_{res} ~ 0_{B}} + \Gamma \ket{0_{A} ~ 1_{res} ~ 0_{B}} \bra{1_{A} ~ 0_{res} ~ 0_{B}} + h.c. \Big), \label{rhon} \end{eqnarray} where h.c. stands for the hermitian conjugate of the previous quantity. The distinguishability between $q_A$ and $q_B$ is obtained from the reduced state operator of subsystem $q_{A}+q_{B}$: \begin{eqnarray} \rho_{q_{A}, q_{B}}^{(n)} =\operatorname{tr}_{res} \rho^{(n)} = \frac{1}{2} \Big[ \ket{0_{A}} \bra{0_{A}} \Big( a^{2n} \ket{1_{B}} \bra{1_{B}} + \Gamma^2 \ket{0_{B}} \bra{0_{B}} \Big) + \ket{1_{A}} \bra{1_{A}} \ket{0_{B}} \bra{0_{B}} + a^n \ket{0_{A} 1_B} \bra{{1_{A}} 0_B} + h.c. \Big], \end{eqnarray} \end{widetext} as: \begin{eqnarray} D_{q_{A},q_{B}}^{(n)} &=& \operatorname{tr} \Big\{ \abs{\frac{1}{2} \Big( a^{2n} \ket{1_{B}} \bra{1_{B}} + \Gamma^2 \ket{0_{B}} \bra{0_{B}} \Big) - \frac{1}{2} \ket{0_{B}} \bra{0_{B}}} \Big\} \nonumber \\ &=& \frac{1}{2} (a^{2n} + \abs{\Gamma^2 - 1}) = a^{2n}. \label{DnqAqB}\end{eqnarray}} \subsection{Continuous Limit - A digression}\label{digression} Defining $k = g^2 \frac{T}{N},$ one can write \cite{Carmichael1999, Breuer2007, Jacobs1998} $a = \cos \left(\sqrt{\frac{k T}{N}} \right),$ where $T$ is the total time of interaction between $q_B$ and $R$ \ankb{(composed by $N$-qubits)}{}. We consider that the interaction time between each qubit of $R$ and $q_B$ is equal, given by $\Delta t=\frac{T}{N}$. We also assume that $0<\Delta t<\pi/2g.$ After $N$ interactions the reduced state in the subsystem $q_A$ is a statistical mixture $\rho_{A}=\frac{1}{2}\left(\vert 0_{A} \rangle \langle 0_{A} \vert + \vert 1_{A} \rangle \langle 1_{A} \vert\right)$, therefore $V_{q_A} = 0$ and $P_{q_A} = 0$. The concurrence can be calculated from the reduced state of the subsystem $q_A, q_B$ and is given by $C_{q_A, q_B} = a^N$. The limit $N \rightarrow \infty$ (and consequently $g \rightarrow \sqrt{\frac{N}{T}} \rightarrow \infty$; $\Delta t \rightarrow dt$) is well known in the Literature \cite{Carmichael1999, Breuer2007, Jacobs1998} and it gives the reservoir limit (at a given temperature implicitly defined in $k$) of a qubit interacting with a Markovian pure dissipative reservoir. The term $a^n$ in \nkb{}{Eq.}\eqref{psin} is\ankb{,}{} in this limit\nkb{:}{} $\lim_{N \rightarrow \infty} a^N = \lim_{N \rightarrow \infty} \cos^N \sqrt{\frac{k T}{N}} = \mathrm{e}^{-kT/2}$, and consequently the concurrence $C_{q_A, q_B} = \mathrm{e}^{\frac{-k T}{2}}$ decays exponentially with $T$. \section{Results}\label{results} \subsection{Complementarity quantities versus coupling intensity} Similar to what was done in Ref.~\cite{Rossi2013}, let us now consider that, after $n$ interactions, the $i$-th qubit of $R$ is projected in the state: $\ket{M_i} = \alpha_i \ket{0_i} + \beta_i \ket{1_i},$ where $\alpha_i = \cos \theta_i$ e $\beta_i = \mathrm{e}^{i \phi_i} \sin \theta_i$ (this measure can be done experimentally, see for example \cite{Haroche2006, Salles2008, Aguilar2014}). The vector state $\ket{M_i}$ is an eigenstate of the operator \ankb{$\hat{\sigma}_{i}=\vec{n}\cdot \vec{\sigma}_{i}$}{$\hat{\sigma}_{i}=\vec{n}\cdot \vec{\sigma}$} with $\vec{n}=\left(\sin2\theta_{i} ~ \cos2\phi_{i},~ \sin2\theta_{i}~\sin2\phi_{i},~\cos2\theta_{i}\right)$ and \ankb{$\vec{\sigma}=\left(\sigma_{x,i},~\sigma_{y,i},~\sigma_{z,i}\right)$}{$\vec{\sigma}=\left(\sigma_{x},~\sigma_{y},~\sigma_{z}\right)$}, the Pauli matrices. One can, in principle, choose in which base ($\theta_i$ and $\phi_i$) the global state will be measured. \ankb{Considering}{Let us consider} projective measurements performed on the state \eqref{psin}, the projector is given by $ \Pi = \Pi_1 \otimes \ldots \otimes \Pi_n, $ with $\Pi_i = \mathbb{I}_1 \otimes \ldots \otimes \mathbb{I}_{i-1} \otimes \ket{M_i}\bra{M_i} \otimes \ldots \mathbb{I}_n.$ Notice that the projective measure $\Pi$ acts only on the subsystem $R$. After $n$ projective measurements the normalized global state vector is given by: \begin{eqnarray} \ket \psi^{(n,M)} &=& \frac{1}{N}\Big( \gamma_1 \ket{0_A} \ket M \ket{0_B} + \gamma_2 \ket{0_A} \ket{M} \ket{1_B} \nonumber \\ &&+ \gamma_3 \ket{1_A} \ket M \ket{0_B} \Big), \label{rhoreducedM}\end{eqnarray} where $\ket M = \ket{M_1} \ldots \ket{M_n}$, $N = \sqrt{\abs{\gamma_1}^2 + \abs{\gamma_2}^2 + \abs{\gamma_3}^2},$ and \begin{eqnarray} \gamma_1 &=& \frac{1}{\sqrt 2} \left(b \sum_{i = 1}^{n} \left[ a^{i-1} \frac{\beta_i}{\alpha_i} \left( \prod_{j=1}^{n} \alpha_j \right) \right] \right),\nonumber \\ \gamma_2 &=& \frac{1}{\sqrt 2} \left(a^n \prod_{i=1}^{n} \alpha_i \right), \nonumber \\ \gamma_3 &=& \frac{1}{\sqrt 2} \left( \prod_{i=1}^{n} \alpha_i \right).\end{eqnarray} The information carried by the qubits of $R$ are now embodied in the measurement outcomes $\theta_i$ and $\phi_i$. The complementarity quantities after the measurements are given by: \begin{eqnarray} V^{(n,M)}_{q_{A}} &=& \frac{2 \abs{\gamma_1 \gamma_3^*}}{N^2} \nonumber \\ P^{(n,M)}_{q_{A}} &=& \frac{\abs{\abs{\gamma_3}^2 - \abs{\gamma_1}^2-\abs{\gamma_2}^2}}{N^2} \nonumber \\ C^{(n,M)}_{q_{A},q_{B}} &=& \frac{2 \abs{\gamma_2 \gamma_3}}{N^2}. \nonumber \label{complementarity} \end{eqnarray} Since the reduced state is pure \eqref{rhoreducedM}, the closure relation for complementarities \ankb{hold}{holds}: \begin{equation} \left(C^{(n,M)}_{q_{A},q_{B}}\right)^2 +\left(P^{(n,M)}_{q_{A}}\right)^2 +\left(V^{(n,M)}_{q_{A}}\right)^2 = 1. \label{complem1}\end{equation} \ankb{Those}{These} quantities depend explicit on the coefficients $\theta_i$ and $\phi_i$ of $\ket M_i$. \alams{If one \ankb{perform}{performs} a measurement $P_i$ on the \ankb{i-th}{$i$-th} qubit for instance, he/she can in principle choose $\theta_i$ and $\phi_i$ \alams{arbitrarily}{so that the outcome of $P_i$ return the required information about}.}{In principle $\theta_i$ and $\phi_i$ can be chosen such that by performing a measurement $\Pi_i$, the complementary quantities will change accordingly.} Concerning the Complementarity quantities, one can project the global state so that $V^{(n,M)}_{q_{A}}, P^{(n,M)}_{q_{A}}$ or $C^{(n,M)}_{q_{A}, q_B}$ acquire the maximum allowed values, after $n$ measurements on the qubits of $R$ have interacted with $q_B$. In \ankb{Reference \cite{Rossi2013}}{Ref.~\cite{Rossi2013}}, the authors studied a similar maximization procedure, \ankb{}{although} only for the \ankb{Visibility}{visibility} $V^{(n,M)}_{q_{A}}$\ankb{, in order}{. In order} to produce a multipartite quantum eraser\ankb{:}{,} the coefficients $\alpha_i$ and $\beta_i$ were chosen to obtain an \ankb{increasing}{increase} in the \ankb{Visibility}{visibility}, maintaining a standard value for the coupling parameter ($g T = 2 \pi$). Here we are interested in how each of the Complementarity quantities behaves, \ankb{if one change the coupling intensity $g$}{for different coupling intensities $g$'s} between $q_B$ and the $i$-th \ankb{qubits}{qubit} of $R$, while making projective measurements in each qubit of $R$. \alams{The measurement outcomes, i.e. the values of $\theta_i$ and $\phi_i$, are given by the following numerical maximization:}{The values of $\theta_i$ and $\phi_i$ were chosen by the following numerical simulation:} if the function to be maximized is the concurrence $C^{(n,M)}_{q_{A}, q_B}$, for example, the procedure gives the values of $\theta_i$ and $\phi_i$ that provides the maximum value of $C^{(n,M)}_{q_{A}, q_B}$, after $n$ qubits have interacted with $q_B$; then, in the possession of $\theta_i$ and $\phi_i$, one can evaluate $V^{(n,M)}_{q_{A}}$ and $P^{(n,M)}_{q_{A}}$. The same procedure is carried out in order to maximize the \ankb{Visibility}{visibility} or the predictability. Therefore, we have all the Complementarity quantities for each function to be maximized: $V^{(n,M)}_{q_{A}}, P^{(n,M)}_{q_{A}}$ or $C^{(n,M)}_{q_{A}, q_B}$. \begin{figure}[h] \centering {\includegraphics[scale=0.43]{comp_maxvG.pdf} \label{maxvisa}}\hspace{0.5cm} {\includegraphics[scale=0.43]{comp_maxvS.pdf} \label{maxvisb}} \caption{(Color online) Complementarity quantities -- $V^{(n,M)}_{q_A}$ (solid), $C^{(n,M)}_{q_A,q_B}$ (dashed) and $P^{(n,M)}_{q_A}$ (dotted) -- as a function of \emph{n}, for optimization procedure in order to maximize the \emph{visibility}. Parameters: (a) $g T = 2 \pi \times 4$ and (b) $g T = \frac{2 \pi}{4}$. Also $N = 20$ and the coefficients $\alpha_i$ and $\beta_i$ are given by the maximization procedure. In Figure (a), the solid red curve represents the limit $N \rightarrow \infty$, leading to $C_{q_A, q_B} = \mathrm{e}^{\frac{-k T}{2}}$ (subsection \ref{digression} with $k=3$). All quantities are dimensionless.} \label{maxvis} \end{figure} \begin{figure}[h] \centering { \includegraphics[scale=0.42]{comp_maxpG.pdf} \label{maxprea }\hspace{0.5cm} { \includegraphics[scale=0.42]{comp_maxpS.pdf} \label{maxpreb} } \caption{(Color online) Complementarity quantities -- $V^{(n,M)}_{q_A}$ (solid), $C^{(n,M)}_{q_A,q_B}$ (dashed) and $P^{(n,M)}_{q_A}$ (dotted) -- as a function of \emph{n}, for optimization procedure in order to maximize the \emph{predictability}. Parameters: (a) $g T = 2 \pi \times 4$ and (b) $g T = \frac{2 \pi}{4}$. Also $N = 20$ and the coefficients $\alpha_i$ and $\beta_i$ are given by the maximization procedure. In Figure (a), the solid red curve represents the limit $N \rightarrow \infty$, leading to $C_{q_A, q_B} = \mathrm{e}^{\frac{-k T}{2}}$ (subsection \ref{digression} with $k=3$). All quantities are dimensionless.} \label{maxpre} \end{figure} \begin{figure}[h] \centering {\includegraphics[scale=0.42]{comp_maxcG.pdf}\label{maxcona} }\hspace{0.5cm} { \includegraphics[scale=0.43]{comp_maxcS.pdf}\label{maxconb} } \caption{(Color online) Complementarity quantities -- $V^{(n,M)}_{q_A}$ (solid), $C^{(n,M)}_{q_A,q_B}$ (dashed) and $P^{(n,M)}_{q_A}$ (dotted) -- as a function of \emph{n}, for optimization procedure in order to maximize the \emph{concurrence}. Parameters: (a) $g T = 2 \pi \times 4$ and (b) $g T = \frac{2 \pi}{4}$. Also $N = 20$ and the coefficients $\alpha_i$ and $\beta_i$ are given by the maximization procedure. In Figure (a), the solid red curve represents the limit $N \rightarrow \infty$, leading to $C_{q_A, q_B} = \mathrm{e}^{\frac{-k T}{2}}$ (subsection \ref{digression} with $k=3$). All quantities are dimensionless.} \label{maxcon} \end{figure} In Figures \ref{maxvis}, \ref{maxpre} and \ref{maxcon} we show the Complementarity quantities, for the maximization of $V^{(n,M)}_{q_{A}}, P^{(n,M)}_{q_{A}}$ and $C^{(n,M)}_{q_{A}, q_B}$, respectively. We also present the behaviour for different couplings (each Figure (a) depicts $g = 4$ and each Figure (b) $g = \frac{1}{4}$). The \ankb{Black}{black} solid curve is related to the $\ankb{Visibility}{visibility}$; the same follows for the $concurrence$ (\ankb{Black}{black} dashed curve) and the $predictability$ (black dotted curve). Note in Figures \ref{maxvis}a and \ref{maxvis}b\ankb{, that for this coupling strength}{}, the \ankb{Visibility}{visibility} is \ankb{a}{an} increasing function of $n$, when the measurements are made to maximize $V^{(n,M)}_{q_{A}}$ \cite{Rossi2013}\ankb{, however}{. However,} if one \ankb{evaluates}{implements} measurements in order to maximize another Complementarity quantity, the \ankb{Visibility}{visibility} does not increase, and remains \ankb{in a value near}{close to} zero, as one can see in Figures \ref{maxpre} and \ref{maxcon}. \alams{Now, if the measurements are made in order to maximize the \emph{\ankb{Visibility}{visibility}}, for higher values of $g$, the behavior is completely different: the function goes to a maximum, and then decreases rapidly. In fact, since $(V^{(n,M)}_{q_{A}})^2+(C^{(n,M)}_{q_{A}, q_B})^2+(P^{(n,M)}_{q_{A}})^2 = 1$, if $V^{(n,M)}_{q_{A}}$ diminishes, other Complementarity quantity must increase.}{For smaller values of $g$, Figure \ref{maxvis}b, one can see that a perfect visibility is not reachable within our range of parameters, and this quantity reaches a maximum value $\sim 0.70$. This feature can be understood when we analyze the behavior of all the Complementarity quantities all together (Figures \ref{maxvis}b).} \textcolor[rgb]{0.00,0.00,0.00}{Differently from the visibility, always when the predictability P is maximized it achieves the maximum value for some finite value of $n$ (Figures \ref{maxpre}a and \ref{maxpre}b). Moreover, it achieves a maximum valeu also when other quantities are maximized (Figure \ref{maxcon}a). Using an interferometric analogy, predictability is the which-way information that is available in the interferometric system, while \ankb{Visibility}{visibility} gives the quality of the interference pattern and concurrence is a measure of entanglement between $q_a$ (main system) and $q_B$ (which-way detector). Therefore, in Figures \ref{maxpre}a, \ref{maxpre}b and \ref{maxcon}a, since the system loses entanglement, with no acquisition of visibility, the predictability must increase.} \alams{}{} \alams{In Figures \ref{maxpre}a and \ref{maxpre}b it is possible to observe that the predictability tends to increase after some interactions, apart the case where the visibility is maximized (for $g T = 2 \pi \times 4$).}{} This feature can be understood in our approach, since the projective measurements will inherently modify the global system. In Table \ref{table} we show the states after $n$ interactions ($n = 1, 2$ and $10$), and after performing the maximization procedures. For instance, performing a maximization of $P_{q_A}^{(n,M)}$, the state $\ket{\psi}^{(n,M)}$ for $g T = 2 \pi \times 4$ tends to the state $\ket{1_A 0_B}$. \alams{}{For $g T = \frac{2 \pi}{4}$, the predictability can possess high values, as seen in Figure \ref{maxpre}b, since the asymptotic state is $\ket{0_A 0_B}$.} \textcolor[rgb]{0.00,0.00,0.00}{The dashed curves in Figures \ref{maxvis}, \ref{maxpre} and \ref{maxcon} show the concurrence as a function of $n$. For $g T = \frac{2 \pi}{4}$, if the function to be maximized is the concurrence itself $C^{(n,M)}_{q_{A}, q_B}$ (Figure \ref{maxcon}b) \alams{or the predictability $P^{(n,M)}_{q_{A}}$}{}, it is possible to maintain the state almost maximally Entangled -- $C^{(n,M)}_{q_{A}, q_B} \sim 1$ -- by choosing the proper values of $\theta_i$ and $\phi_i$. \alams{}{Moreover, one can obtain Entanglement values near to $0.5$, performing measurements in order to maximize the visibility, Figure \ref{maxvis}b. This result is interesting, since we can see a clear complemental character between all quantities.} However, if the coupling is increased -- Figures \ref{maxvis}a, \ref{maxpre}a, or \ref{maxcon}a -- even if the projective measures were made to maximize the concurrence (Figure \ref{maxcon}a), Entanglement decreases to zero. This behavior is similar to two entangled qubits, where \alams{qubit}{one of them, say} $q_B$, is coupled to a thermal reservoir (red solid curves), but in our case we have a finite number of interacting qubits with $q_B$ (where the maximum number of interacting qubits is $N = 20$). \alams{This feature can be understood when we analyse}{This can be conprehended by analyzing} how the initial information (given by the distinguishability between $q_A$ and the $i$-th qubit) is distributed over the $N$ qubits \alams{}{(section \ref{distsec})}. An interesting aspect concerning Figure \ref{maxcon}b, for $g T = \frac{2 \pi}{4}$, is that one can see an approximately steady behavior of the concurrence $C^{(n,M)}_{q_{A}, q_B}$, near the initial value $C^{(0)}_{q_{A}, q_B} = 1$. It is possible, therefore, to maintain the system $q_A + q_B$ in an approximately maximal Entangled state, notwithstanding the qubits of $R$ became dynamically correlated with $q_A$ and $q_B$ (Equation \eqref{rhon}). One can see from Table \ref{table}, for $g T = \frac{2 \pi}{4}$ and $n=10$, the state resemble the initial maximally entangled state\alams{, for all maximization procedures,}{, if the maximization is over $C_{q_A, q_B}^{(n,M)}, $} corroborating Figure \ref{maxcon}b.} \begin{table*}[t] \caption{Approximate values of the state $\ket{\psi}^{(n,M)}$ after $n$ measurements, for different optimization procedures and coupling strengths.}\label{table} \begin{center} \begin{ruledtabular} \begin{tabular}{cc||c||cc} & & $\mathbf{gT = 2 \pi \times 4}$& $\mathbf{gT = 2 \pi / 4}$ & \\[5pt] \cline{1-4} \multicolumn{1}{ c }{\multirow{3}{*}{\textbf{Max. of $P_{q_A}^{(n,M)}$}} } & \multicolumn{1}{ c|| }{$\mathbf{n=1}$} & $\Big( (-0.05 + 0.99 i) \ket{0_A 0_B} \Big)\otimes \ket{M}$ & $\Big( (-0.73 + 0.68 i) \ket{0_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \multicolumn{1}{ c }{} & \multicolumn{1}{ c|| }{$\mathbf{n=2}$} & $\Big( 0.10 \ket{0_A 1_B} + 0.99 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( (-0.45 + 0.89 i) \ket{0_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \multicolumn{1}{ c }{} & \multicolumn{1}{ c|| }{$\mathbf{n=10}$} & $\Big( \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( (-0.15 + 0.98 i) \ket{0_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \cline{1-4} \multicolumn{1}{ c }{\multirow{3}{*}{\textbf{Max. of $V_{q_A}^{(n,M)}$}} } & \multicolumn{1}{ c|| }{$\mathbf{n=1}$} & $\Big( (0.23 - 0.66 i) \ket{0_A 0_B} + 0.21 \ket{0_A 1_B} + 0.67 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( (0.61 - 0.35 i) \ket{0_A 0_B} - 0.50 \ket{0_A 1_B} - 0.50 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \multicolumn{1}{ c }{} & \multicolumn{1}{ c|| }{$\mathbf{n=2}$} & $\Big( (0.17 - 0.68 i) \ket{0_A 0_B} + 0.06 \ket{0_A 1_B} + 0.70 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( (0.47 - 0.52 i) \ket{0_A 0_B} + 0.49 \ket{0_A 1_B} + 0.49 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \multicolumn{1}{ c }{} & \multicolumn{1}{ c|| }{$\mathbf{n=10}$} & $\Big( (-0.43 + 0.55 i) \ket{0_A 0_B} + 0.70 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( (0.40 - 0.58 i) \ket{0_A 0_B} - 0.49 \ket{0_A 1_B} - 0.50 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \cline{1-4} \multicolumn{1}{ c }{\multirow{3}{*}{\textbf{Max. of $C_{q_A,q_B}^{(n,M)}$}} } & \multicolumn{1}{ c|| }{$\mathbf{n=1}$} & $\Big( 0.29 \ket{0_A 1_B} + 0.95 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( 0.71 \ket{0_A 1_B} + 0.71 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \multicolumn{1}{ c }{} & \multicolumn{1}{ c|| }{$\mathbf{n=2}$} & $\Big( 0.10 \ket{0_A 1_B} + 0.99 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & $\Big( 0.70 \ket{0_A 1_B} + 0.71 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \multicolumn{1}{ c }{} & \multicolumn{1}{ c|| }{$\mathbf{n=10}$} & $ \Big(\ket{1_A 0_B} \Big)\otimes \ket{M}$ & $ \Big( 0.69 \ket{0_A 1_B} + 0.72 \ket{1_A 0_B} \Big)\otimes \ket{M}$ & \\[5pt] \end{tabular} \end{ruledtabular} \end{center} \end{table*} \subsection{Complementarity without maximization procedure}\label{noMaximization} \begin{figure*}[t] \centering \includegraphics[scale=0.3]{beta_alfa.pdf} \caption{(Color online) $C^{(n,M)}_{q_{A}, q_B}$, $V^{(n,M)}_{q_{A}}$ and $P^{(n,M)}_{q_{A}}$, respectively, as a function of $n$ and $\theta$. Parameters: $N = 20$, (a) $g T = 2 \pi \times 4$ and (b) $g T = \frac{2 \pi}{4}$. All quantities are dimensionless.}\label{alfaequal} \end{figure*} \alams{}{Here we develop a study without assuming any maximization, adding an evident picture of the complementarity between the quantities. }Suppose that instead of performing the measurements $\Pi_i$ in order to maximize a given quantity, one is able to project the state $\rho^{(n)}$ only in the same base for each qubit of $R$. In other words, lets consider the $i$-qubit of $R$ is projected in the state $\ket{M_i} = \alpha \ket{0_i} + \beta \ket{1_i}$, where $\alpha = \cos \theta$ and $\beta = \mathrm{e}^{i \phi} \sin \theta$ are \alams{equal}{the same, individually, } for all measurements. With this assumption, the coefficients for the state \eqref{rhoreducedM} are given by: \begin{eqnarray} \gamma_1 &=& \frac{(a^n -1) b \alpha^n \beta}{\sqrt{2} (a-1) \alpha}, \nonumber \\ \gamma_2 &=& \frac{\alpha^n a^n}{\sqrt{2}}, \nonumber \\ \gamma_3 &=& \frac{\alpha^n}{\sqrt{2}}, \end{eqnarray} where $a = \cos \left( \frac{g T}{N}\right)$ and $b = - i \sin \left( \frac{g T}{N}\right)$. The coefficients above allow us to calculate the Complementarity quantities $V^{(n,M)}_{q_{A}}, P^{(n,M)}_{q_{A}}$ and $C^{(n,M)}_{q_{A}, q_B}$, that we depict in Figure \ref{alfaequal} for different couplings strength. One can see clearly the Complementarity behavior between the quantities in all plots. Note in Figure \ref{alfaequal}a ($g T = 2 \pi \times 4$), when $\theta = \frac{m \pi}{2}$ ($m \in$ Integer) for large values of $n$, the predictability has values near unity, while the \ankb{Visibility}{visibility} is near zero (the concurrence is negligible in this regime, as pointed out previously). This can be seen through Equation \eqref{rhon}: if $\theta = \frac{m \pi}{2}$ for odd values of $m$ the state after $n$ measurements is approximately $\rho_{m} \thickapprox \ket{1_A 0_B}$, while for $m$ assuming even values the state will become $\rho_{m} \thickapprox \ket{0_A 0_B}$, explicitly the maximum values of the predictability in Figure \ref{alfaequal}a. The same argument can be used for the \ankb{Visibility}{visibility} in this regime, but in this case $\theta = \frac{(2m+1) \pi}{4}$ ($m$ Integer) and the final state tends to $\rho_{m} \thickapprox \frac{\ket{1_A 0_B} + \ket{0_A 1_B}}{\sqrt{2}}$. Figure \ref{alfaequal}b, on the other hand, shows the Complementarity quantities in the weak coupling regime. It is noticeable that one can sustain the state in the maximal entangled state by performing specific measurements ($\theta = m \pi$, with $m$ Integer). This can be achieved since for this coupling regime only little information about the initial state is captured by the interaction between $q_B$ and $q_i$ (Subsection \ref{distsec}). The large values for the predictability (and consequently small values of \ankb{Visibility}{visibility} and concurrence), when $\theta = \frac{m \pi}{2}$ (where $m$ is an odd Integer), can be understood in the same sense as the case of $g T = 2 \pi \times 4$ (Figure \ref{alfaequal}a), where the state is approximately $\rho_{m} \thickapprox \ket{1_A 0_B}$. \subsection{Information distribution - Distinguishability}\label{distsec} In this section we use the distinguishability to quantify the information stored by each subsystem. We show how this quantity varies as one evaluate the measurements. We are concerned in how the information stored in some parts of the global system behaves, given that $n$ measurements are performed in the subsystem $R$. The distinguishability is calculated before and after $n$ measurements, and we show the curves that illustrate the behavior of this quantity for the maximizations of $V$, $P$ and $C$. Figures \ref{dist}a and \ref{dist}b show how the distinguishability is distributed in the $N$ qubits, i.e. $D^{(n)}_{q_A,q_i}$: \begin{eqnarray} D_{q_{A},q_i}^{(n)} &=& \operatorname{tr}_{q_i} \Big\{ \Big| \frac{1}{2} a^{2n} \ket{0_i} \bra{0_i} + \abs{a^{i-1} b}^2 \ket{1_i} \bra{1_i} + \nonumber \\ && + (\abs{a^{n-1} b}^2 + \ldots + \abs{a^{i-2} b}^2 + \nonumber \\ &&+ \abs{a^i b} + \ldots + \abs{b}^2) \ket{0_i} \bra{0_i} - \ket{0_i} \bra{0_i} \Big| \Big\} = \nonumber \\ &=& \abs{a^{i-1} b}^2. \nonumber \\ \label{dist_eq}\end{eqnarray} This relation gives the amount of information the $i$-th qubit possess about the initial state. Both Figures are independent of the maximization procedure. Note that for $g T = \frac{2 \pi}{4}$, Figure \ref{dist}b, the information is almost equally distributed in all the qubits of $R$. For coupling $g T = 2 \pi \times 4$, the first qubits that interacted with $q_B$ retain a large amount of information from $q_A + q_B$ (Figure \ref{dist}a). Comparing the results of Figures \ref{maxcon}a and \ref{dist}a, we can argue that, if $g T = 2 \pi \times 4$ (for high coupling intensity between $R$ and $q_B$), the first qubits of $R$ that interact with $q_B$ ``extract'' sufficient information, that was initially stored only in $q_A + q_B$. Therefore, the global system ($R + q_A + q_B$) becomes strongly correlated, leading the concurrence between $q_A$ and $q_B$ diminishes rapidly, compared with the case $g T = \frac{2 \pi}{4}$ (Figures \ref{maxvis}a, \ref{maxpre}a and \ref{maxcon}a). \begin{figure}[h] \centering { \includegraphics[scale=0.4]{distingui_g.pdf}\label{dist_a } { \includegraphics[scale=0.4]{distingui_p.pdf}\label{dist_b} } \caption{Distinguishability between $q_A$ and the $i$-th qubit of $R$, as a function of \emph{n}, i.e. how much information about the initial state the qubit $i$ have. Parameters: $N = 20$, (a) $g T = 2 \pi \times 4$ and (b) $g T = \frac{2 \pi}{4}$. All quantities are dimensionless.} \label{dist} \end{figure} \begin{figure}[h] \centering { \includegraphics[scale=0.4]{delta_t_g.pdf}\label{maxdeltaTa} } { \includegraphics[scale=0.4]{delta_T_s.PDF}\label{maxdeltaTb} } \caption{$\Delta D_T$ as a function of \emph{n}, for optimization procedure in order to maximize the \ankb{Visibility}{visibility} (solid black), the concurrence (dashed black) and the predictability (dotted black). Parameters: (a) $g T = 2 \pi \times 4$ and (b) $g T = \frac{2 \pi}{4}$. Also, $N = 20$ and the coefficients $\alpha_i$ and $\beta_i$ are given by the maximization procedure. All quantities are dimensionless.} \label{maxdeltaT} \end{figure} The total Distinguishability, for the global system composed by $q_A + q_B + R$, is: \begin{eqnarray} D^{(n)}_{T} &=& \sum^{N}_{i\neq q_{A}}C^{2}_{q_{A},i} = a^{2n}+ \abs{a^{n-1} b}^2 + \ldots + \abs{a^i b}^{2} + \ldots + \abs{b}^2 \nonumber \\ &=& D_{q_{B}}^{(n)} + \sum_{i=1}^{n} D_{q_i}^{(n)} = 1. \end{eqnarray} After performing the projective measurements, the Distinguishability of the subsystem $q_{A} + q_{B}$ is: \begin{equation} D^{(n,M)}_{T}= \sqrt{\left(C^{(n,M)}_{q_{A},q_{B}}\right)^{2}+\left(P^{(n,M)}_{q_{A}}\right)^{2}}. \end{equation} Since the projected subsystem can now be decoupled from $q_A$ and $q_B$ \eqref{rhoreducedM}, the variation of the total Distinguishability is: \begin{eqnarray} \Delta D_{T} &=& D^{(n,\textbf{M})}_{T} - D^{(n)}_{T} =\sqrt{\left(C^{(n,\textbf{M})}_{q_{A},q_{B}}\right)^{2}+\left(P^{(n,\textbf{M})}_{q_{A}}\right)^{2}} - 1 \nonumber \\ &=& \sqrt{1-\left(V^{(n,\textbf{M})}_{q_{A}}\right)^{2}} - 1. \end{eqnarray} The variation of Distinguishability between subsystems $q_{A}$ and $q_{B}$ is given by \eqref{DnqAqB}: \begin{equation} \Delta D_{F} = D^{(n,\textbf{M})}_{T} - D^{(n)}_{q_A, q_B} = \sqrt{1-\left(V^{(n, M)}_{q_{A}}\right)^{2}}-a^{2n}.\end{equation} Figures \ref{maxdeltaT}a and \ref{maxdeltaT}b show $\Delta D_T$ for \ankb{$g T = 2 \pi \times 4$}{$g T = 2 \pi \times 4$} and \ankb{$g T = \frac{2 \pi}{4}$}{$g T = \frac{2 \pi}{4}$}, respectively. We can see that the measurements that maximizes the visibility, erases the information stored by the qubits of $R$. Notice that in \ref{maxdeltaT}a for $N=4$ the the information of the global system is approximately zero, this is in accord with Figure \ref{maxvis}a that shows visibility approximately $1$ for $N=4$. In figure \ref{maxdeltaT}b the stabilization of the curve shows that the information erased is limited, because it is equally distributed among the qubits of $R$, as it is shown in Figure \ref{dist}b, therefore it is erased by small amounts after each measurements. Measurements that maximizes predictability and concurrence show no variation of information before and after measurements of the qubits, therefore, they do not erase information form the system $R$. \section{Conclusion}\label{conclusion} In this work we have proposed and discussed in details a scheme to observe the behavior of Complementarity quantities (concurrence, \ankb{Visibility}{visibility} and predictability) of a two qubit system, initially maximally entangled, after two steps: one of the qubits ($q_B$) interacts with $R$ (composed by $N$ other qubits -- $q_i$); after the interaction, projective measurements are made in each qubit of $R$ in order to maximize a given quantity. We observe that, if the coupling strength between $q_B$ and $R$ is considerable, the concurrence behaves similarly as a system of two qubits coupled to a thermal reservoir, even though our subsystem $R$ is composed by a finite number of qubits $N$. On the other hand, if the coupling is ``small'' (compared to the previous one), the system entanglement may as well be preserved. \textcolor[rgb]{0.00,0.00,0.00}{The visibility however shows a different behavior, when the coupling is stronger its maximization is more effective.} To explicit these results, we show some intermediate states for different couplings and number of interactions. The differences of the behavior can be understood from the distribution of information over the global system, as measured by the Distinguishability between $q_A+q_i$. We also studied the variation of distinguishability (before and after the measurements) for the global system and for the initial qubits $q_A + q_B$, making a connection between the information stored in each part of the system and the corresponding behavior of the Complementarity quantities. Note that the presented model may be feasible experimentally. One can, in principle, call qubits $q_A$ and $q_B$ as the modes inside microwave cavities, which nowadays are constructed with a lifetime of approximately $0.1$ seconds \cite{Gleyzes2007, Kuhr2007}. The qubits of the subsystem $R$ could represent two-level atoms in a QED experiment, for instance. The interaction time between each atom ($q_i$) and the mode $q_B$ is of the order of $10^{-5} s$ \cite{Brune1996}. So, for the $10$ qubits of our model, the effect would be in fact visible and dissipation can be neglected. We showed how to fully control and prepare maximized states for different complementarity quantities. Since many parts are involved in this control scheme, this could have applications for protocols of quantum information, for example as protocols of bit commitment \cite{bitcomm1, bitcomm2}, where Alice wants to save safely the information of her bit for some time, but wants to reveal it later on to Bob. In this case, we could imagine that Alice would have access to part R and Bob, on the other hand, would have access to qubit A and B. \begin{acknowledgments} The authors would like to thank the support from the Brazilian agencies CNPq and CAPES (CAPES (6842/2014-03); CNPq (470131/2013-6)). L.A.M.S. also thanks the University of Nottingham for hospitality and support during part of this work preparation. The authors acknowledge useful discussions with P. Saldanha, M. P. Fran\c{c}a Santos and G. Murta. \end{acknowledgments}
ef6fecbf91bf23ce97668d9e0f68279d16e146af
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{#1}\setcounter{equation}{0}} \newcommand{\subsect}[1]{\subsection{#1}} \renewcommand{\theequation}{\arabic{section}.\arabic{equation}} \font\mbn=msbm10 scaled \magstep1 \font\mbs=msbm7 scaled \magstep1 \font\mbss=msbm5 scaled \magstep1 \newfam\mbff \textfont\mbff=\mbn \scriptfont\mbff=\mbs \scriptscriptfont\mbff=\mbss\def\fam\mbff{\fam\mbff} \def{\mbf T}{{\fam\mbff T}} \def{\mbf P}{{\fam\mbff P}} \def{\mbf F}{{\fam\mbff F}} \newcommand{{\mbf D}} {\mathbb{D}} \newcommand{{\mbf R}} { \mathbb{R}} \newcommand{\cH} {{\mathcal H}} \newcommand{\cP} {{\mathcal P}} \newcommand{{\mbf N}} { \mathbb{N}} \newcommand{{\mbf Z}} {\mathbb{Z} } \newcommand{\mbf C} {{\mathbb C}} \newcommand {\mbf Q} {{\mathbb Q}} \newtheorem{Th}{Theorem}[section] \newtheorem{Lm}[Th]{Lemma} \newtheorem{C}[Th]{Corollary} \newtheorem{D}[Th]{Definition} \newtheorem{Proposition}[Th]{Proposition} \newtheorem{R}[Th]{Remark} \newtheorem{Problem}[Th]{Problem} \newtheorem{E}[Th]{Example} \newtheorem*{P1}{Problem 1} \newtheorem*{P2}{Problem 2} \newtheorem*{P3}{Problem 3} \begin{document} \title[On Properties of Geometric Preduals of ${\mathbf C^{k,\omega}}$ Spaces]{On Properties of Geometric Preduals of ${\mathbf C^{k,\omega}}$ Spaces} \author{Alexander Brudnyi} \address{Department of Mathematics and Statistics\newline \hspace*{1em} University of Calgary\newline \hspace*{1em} Calgary, Alberta\newline \hspace*{1em} T2N 1N4} \email{[email protected]} \keywords{Predual space, Whitney problems, Finiteness Principle, linear extension operator, approximation property, dual space, Jackson operator, weak$^*$ topology, weak Markov set} \subjclass[2010]{Primary 46B20; Secondary 46E15} \thanks{Research supported in part by NSERC} \date{} \begin{abstract} Let $C_b^{k,\omega}({\mbf R}^n)$ be the Banach space of $C^k$ functions on ${\mbf R}^n$ bounded together with all derivatives of order $\le k$ and with derivatives of order $k$ having moduli of continuity majorated by $c\cdot\omega$, $c\in{\mbf R}_+$, for some $\omega\in C({\mbf R}_+)$. Let $C_b^{k,\omega}(S):=C_b^{k,\omega}({\mbf R}^n)|_S$ be the trace space to a closed subset $S\subset{\mbf R}^n$. The geometric predual $G_b^{k,\omega}(S)$ of $C_b^{k,\omega}(S)$ is the minimal closed subspace of the dual $\bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$ containing evaluation functionals of points in $S$. We study geometric properties of spaces $G_b^{k,\omega}(S)$ and their relations to the classical Whitney problems on the characterization of trace spaces of $C^k$ functions on ${\mbf R}^n$. \end{abstract} \maketitle \section{Formulation of Main Results} \subsection{Geometric Preduals of ${\mathbf C^{k,\omega}}$ Spaces} In what follows we use the standard notation of Differential Analysis. In particular, $\alpha=(\alpha_1,\dots,\alpha_n)\in \mathbb Z^n_+$ denotes a multi-index and $|\alpha|:=\sum^n_{i=1}\alpha_i$. Also, for $x=(x_1,\dots, x_n)\in\mathbb R^n$, \begin{equation}\label{eq1} x^\alpha:=\prod^n_{i=1}x^{\alpha_i}_i \ \ \text{ and} \ \ D^\alpha:=\prod^n_{i=1}D^{\alpha_i}_i,\quad {\rm where}\quad D_i:=\frac{\partial}{\partial x_i}. \end{equation} Let $\omega$ be a nonnegative function on $(0,\infty)$ (referred to as {\em modulus of continuity}) satisfying the following conditions: \begin{enumerate} \item[(i)] $\omega(t)$ and $\displaystyle \frac {t}{\omega( t)}$ are nondecreasing functions on $(0,\infty)$;\medskip \item[(ii)] $\displaystyle \lim_{t\rightarrow 0^+}\omega(t)=0$. \end{enumerate} \begin{D}\label{def1} $C^{k,\omega}_b(\mathbb R^n)$ is the Banach subspace of functions $f\in C^k(\mathbb R^n)$ with norm \begin{equation}\label{eq3} \|f\|_{C^{k,\omega}_b(\mathbb R^n)}:=\max\left(\|f\|_{C^k_b(\mathbb R^n)}, |f|_{C^{k,\omega}_b(\mathbb R^n)}\right) , \end{equation} where \begin{equation}\label{eq4} \|f\|_{C^k_b(\mathbb R^n)}:=\max_{|\alpha|\le k}\left\{\sup_{x\in\mathbb R^n}|D^\alpha f(x)|\right\} \end{equation} and \begin{equation}\label{eq5} |f|_{C^{k,\omega}_b(\mathbb R^n)}:=\max_{|\alpha|=k}\left\{\sup_{x,y\in\mathbb R^n,\, x\ne y}\frac{|D^\alpha f(x)-D^\alpha f(y)|}{\omega(\|x-y\|)}\right\}. \end{equation} Here $\|\cdot\|$ is the Euclidean norm of $\mathbb R^n$. \end{D} If $S\subset\mathbb R^n$ is a closed subset, then by $C_b^{k,\omega}(S)$ we denote the trace space of functions $g\in C_b^{k,\omega}(\mathbb R^n)|_{S}$ equipped with the quotient norm \begin{equation}\label{eq1.5} \|g\|_{C_b^{k,\omega}(S)}:=\inf\{\|\tilde g\|_{C_b^{k,\omega}(\mathbb R^n)}\, :\, \tilde g\in C_b^{k,\omega}({\mbf R}^n),\ \tilde g|_{S}=g\}. \end{equation} Let $\bigl(C_b^{k,\omega}(\mathbb R^n)\bigr)^*$ be the dual of $C_b^{k,\omega}(\mathbb R^n)$. Clearly, each evaluation functional $\delta_x^0$ at $x\in\mathbb R^n$ (i.e., $\delta_x^0(f):=f(x)$, $f\in C_b^{k,\omega}({\mbf R}^n)$) belongs to $\bigl(C_b^{k,\omega}(\mathbb R^n)\bigr)^*$ and has norm one. By $G^{k,\omega}_b(S)\subset \bigl(C_b^{k,\omega}(\mathbb R^n)\bigr)^*$ we denote the minimal closed subspace containing all $\delta_x^0$, $x\in S$. \begin{Th}\label{te1.2} The restriction map to the set $\{\delta_s^0\, :\, s\in S\}\subset G_b^{k,\omega}(S)$ determines an isometric isomorphism between the dual of $G^{k,\omega}_b(S)$ and $C^{k,\omega}_b(S)$. \end{Th} In what follows, $G_b^{k,\omega}(S)$ will be referred to as the {\em geometric predual} of $C^{k,\omega}_b(S)$. In the present paper we study some properties of these spaces. The subject is closely related to the classical Whitney problems, see \cite{W1, W2}, asking about the characterization of trace spaces of $C^k$ functions on ${\mbf R}^n$ (see survey \cite{F3} and book \cite{BB2} and references therein for recent developments in the area). Some of the main results of the theory can be reformulated in terms of certain geometric characteristics of spaces $G_b^{k,\omega}(S)$, see sections~1.2 and 1.3. \subsection{Finiteness Principle} For $C_b^{k,\omega}(S)$ this principle was conjectured by Yu.~Brudnyi and P.~Shvartsman in the 1980s in the following form (cf. \cite[p.\,210]{F3}).\smallskip \noindent \textsl{\large{Finiteness Principle}.} To decide whether a given $f:S\rightarrow{\mbf R}$, $S\subset{\mbf R}^n$, extends to a function $F\in C_b^{k,\omega}({\mbf R}^n)$, it is enough to look at all restrictions $f|_{S'}$, where $S'\subset S$ is an arbitrary $d$-element subset. Here $d$ is an integer constant depending only on $k$ and $n$.\smallskip More precisely, if $f|_{S'}$ extends to a function $F^{S'}\in C_b^{k,\omega}({\mbf R}^n)$ of norm at most $1$ (for each $S'\subset S$ with at most $d$ elements), then $f$ extends to a function $F\in C_b^{k,\omega}({\mbf R}^n)$, whose norm is bounded by a constant depending only on $k$ and $n$.\smallskip For $k=0$ (the Lipschitz case) the McShane extension theorem \cite{McS} implies the Finiteness Principle with the optimal constant $d=2$. Also, for $n=1$ the Finiteness Principle is valid with the optimal constant $d=k+2$. The result is essentially due to Merrien \cite{M}. In the multidimensional case the Finiteness Principle was proved by Yu.~Brudnyi and Shvartsman for $k=1$ with the optimal constant $d=3\cdot 2^{n-1}$, see \cite{BS1}. In the early 2000s the Finiteness Principle was proved by C.~Fefferman for all $k$ and $n$ for regular moduli of continuity $\omega$ (i.e., $\omega(1)=1$), see \cite{F1}. The upper bound for the constant $d$ in the Fefferman proof was reduced later to $d=2^{ k+n \choose k}$ by Bierstone and P. Milman \cite{BM} and independently and by a different method by Shvartsman \cite{S}. The obtained results (and the Finiteness Principle in general) admit the following reformulation in terms of geometric characteristics of closed unit balls $B_b^{k,\omega}(S)$ of $G_b^{k,\omega}(S)$. Specifically, let $B_b^{k,\omega}(S;m)\subset B_b^{k,\omega}(S)$, $m\in{\mbf N}$, be the balanced closed convex hull of the union of all finite-dimensional balls $B_b^{k,\omega}(S')\subset G_b^{k,\omega}(S')$, $S'\subset S$, ${\rm card}\, S'\le m$. \begin{Th}\label{teo1.5} There exist constants $d\in{\mbf N}$ and $c\in (1,\infty)$ such that \[ B_b^{k,\omega}(S;d)\subset B_b^{k,\omega}(S)\subset c\cdot B_b^{k,\omega}(S;d). \] Here for $k=0$, $d=2$ (-\,optimal) and $c=1$, for $n=1$, $d=k+2$ (-\,optimal) and $c$ depends on $k$ only, for $k=1$, $d=3\cdot 2^{n-1}$ (-\,optimal) and $c$ depends on $k$ and $n$ only, and for $k\ge 2$, $d=2^{ k+n \choose k}$ and $c=\frac{\tilde c}{\omega(1)}$, where $\tilde c$ depends on $k$ and $n$ only. \end{Th} \subsection{Complementability of Spaces ${\mathbf G_b^{k,\omega}(S)}$} We begin with a result describing bounded linear operators on $G_b^{k,\omega}({\mbf R}^n)$. To this end, for a Banach space $X$ by $C_b^{k,\omega}({\mbf R}^n;X)$ we denote the Banach space of $X$-valued $C^k$ functions on ${\mbf R}^n$ with norm defined similarly to that of Definition \ref{def1} with absolute values replaced by norms $\|\cdot\|_X$ in $X$. Let $\mathcal L\bigl(X_1;X_2\bigr)$ stand for the Banach space of bounded linear operators between Banach spaces $X_1$ and $X_2$ equipped with the operator norm. \begin{Th}\label{te1.6} The restriction map to the set $\{\delta_x^0\, :\, x\in {\mbf R}^n\}\subset G_b^{k,\omega}({\mbf R}^n)$ determines an isometric isomorphism between $\mathcal L\bigl(G_b^{k,\omega}({\mbf R}^n);X\bigr)$ and $C_b^{k,\omega}({\mbf R}^n;X)$. \end{Th} Let $q_S: C_b^{k,\omega}({\mbf R}^n)\rightarrow C_b^{k,\omega}(S)$ be the quotient map induced by the restriction of functions on ${\mbf R}^n$ to $S$. A right inverse $T\in {\mathcal L}(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$ for $q_S$ (i.e., $q_S\circ T={\rm id}$) is called a {\em linear extension operator}. The set of such operators is denoted by $Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$. \begin{D}\label{def1.5} An operator $T\in Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$ has depth $d\in{\mbf N}$ if for all $x\in{\mbf R}^n$ and $f\in C_b^{k,\omega}(S)$, \begin{equation}\label{equ1.6} (Tf)(x)=\sum_{i=1}^d \lambda_i^x\cdot f(y_i^x), \end{equation} where $y_i^x\in S$ and $\lambda_i^x$ depend only on $x$. \end{D} Linear extension operators of finite depth exist. For $k=0$ (the Lipschitz case) the Whitney-Glaeser linear extension operators $C_{b}^{0,\omega}(S)\rightarrow C_{b}^{0,\omega}({\mbf R}^n)$, see \cite{Gl}, have depth $d$ depending on $n$ only and norms bounded by a constant depending on $n$ only. In the 1990s bounded linear extension operators $C_{b}^{1,\omega}(S)\rightarrow C_{b}^{1,\omega}({\mbf R}^n)$ of depth $d$ depending on $n$ only with norms bounded by a constant depending on $n$ only were constructed by Yu.~Brudnyi and Shvartsman \cite{BS2}. Recently bounded linear extensions operators of depth $d$ depending on $k$ and $n$ only were constructed by Luli \cite{Lu} for all spaces $C_b^{k,\omega}(S)$; their norms are bounded by $\frac{C}{\omega(1)}$, where $C\in (1,\infty)$ is a constant depending on $k$ and $n$ only. (Earlier such extension operators were constructed for finite sets $S$ by C.~Fefferman \cite[Th.\,8]{F2}.) In the following result we identify $(G_b^{k,\omega}(S))^*$ with $C_b^{k,\omega}(S)$ by means of the isometric isomorphism of Theorem \ref{te1.2}. \begin{Th}\label{teo1.6} For each $T\in Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$ of finite depth there exists a bounded linear projection $P:G_b^{k,\omega}({\mbf R}^n)\rightarrow G_b^{k,\omega}(S)$ whose adjoint $P^*=T$. \end{Th} \begin{R}\label{rem1.7} {\rm It is easily seen that if $T$ has depth $d$ and is defined by \eqref{equ1.6}, then \[ p(x):=P(\delta_x^0)=\sum_{i=1}^d \lambda_i^x\cdot \delta_{y_i^x}^0 \quad {\rm for\ all}\quad x\in{\mbf R}^n. \] Moreover, $p\in C_b^{k,\omega}({\mbf R}^n; G_b^{k,\omega}(S))$ and has norm equal to $\|T\|$ by Theorem \ref{te1.6}. } \end{R} \subsection{Approximation Property} Recall that a Banach space $X$ is said to have the {\em approximation property}, if, for every compact set $K\subset X$ and every $\varepsilon > 0$, there exists an operator $T : X\to X$ of finite rank so that $\|Tx-x\|\le\varepsilon$ for every $x\in K$. Although it is strongly believed that the class of spaces with the approximation property includes practically all spaces which appear naturally in analysis, it is not known yet even for the space $H^\infty$ of bounded holomorphic functions on the open unit disk. The first example of a space which fails to have the approximation property was constructed by Enflo \cite{E}. Since Enflo's work several other examples of such spaces were constructed, for the references see, e.g., \cite{L}. A Banach space has the $\lambda$-{\em approximation property}, $1\le\lambda<\infty$, if it has the approximation property with the approximating finite rank operators of norm $\le\lambda$. A Banach space is said to have the {\em bounded approximation property}, if it has the $\lambda$-approximation property for some $\lambda$. If $\lambda=1$, then the space is said to have the {\em metric approximation property}. Every Banach spaces with a basis has the bounded approximation property. Also, it is known that the approximation property does not imply the bounded approximation property, see \cite{FJ}. It was established by Pe\l czy\'nski \cite{P} that a separable Banach space has the bounded approximation property if and only if it is isomorphic to a complemented subspace of a separable Banach space with a basis. Next, for Banach spaces $X,Y$ by ${\mathcal F}(X,Y)\subset {\mathcal L}(X,Y)$ we denote the subspace of linear bounded operators of finite rank $X\to Y$. Let us consider the trace mapping $V$ from the projective tensor product $Y^*\hat{\otimes}_\pi X\to {\mathcal F}(X,Y)^*$ defined by \[ (Vu)(T)={\rm trace}(Tu),\quad\text{where}\quad u\in Y^*\hat{\otimes}_\pi X,\ T\in {\mathcal F}(X,Y), \] that is, if $u=\sum_{n=1}^\infty y_n^*\otimes x_n$, then $(Vu)(T)=\sum_{n=1}^\infty y_n^*(Tx_n)$. It is easy to see that $\|Vu\|\le \|u\|_\pi$. The $\lambda$-bounded approximation property of $X$ is equivalent to the fact that $\|u\|_\pi\le\lambda\|Vu\|$ for all Banach spaces $Y$. This well-known result (see, e.g., \cite[page 193]{DF}) is essentially due to Grothendieck \cite{G}. Our result concerning spaces $G_b^{k,\omega}(S)$ reads as follows. \begin{Th}\label{te1.3} \begin{enumerate} \item Spaces $G_b^{k,\omega}({\mbf R}^n)$ have the $\lambda$-approximation property with \penalty-10000 $\displaystyle \lambda=\lambda(k,n,\omega):=1+C\cdot\lim_{t\rightarrow\infty}\,\mbox{$\frac{1}{\omega(t)}$}$, where $C$ depends on $k$ and $n$ only. \item All the other spaces $G_b^{k,\omega}(S)$ have the $\lambda$-approximation property with \penalty-10000 $\lambda= C'\cdot\lambda(1,n,\omega)$, where $C'$ is a constant depending on $n$ only, if $k=0,1$, and with $\lambda=\frac{C''\cdot \lambda(k,n,\omega)}{\omega(1)}$, where $C''$ is a constant depending on $k$ and $n$ only, if $k\ge 2$. \end{enumerate} \end{Th} If $\lim_{t\rightarrow\infty}\omega(t)=\infty$, then (1) implies that the corresponding space $G_b^{k,\omega}({\mbf R}^n)$ has the metric approximation property. In case $\lim_{t\rightarrow\infty}\omega(t)<\infty$, one can define the new modulus of continuity $\widetilde\omega$ (cf. properties (i) and (ii) in its definition) by the formula \[ \widetilde\omega(t)=\max\{\omega(t),t\},\quad t\in (0,\infty). \] It is easily seen that spaces $C_b^{k,\omega}({\mbf R}^n)$ and $C_b^{k,\widetilde\omega}({\mbf R}^n)$ are isomorphic. Thus $G_b^{k,\omega}({\mbf R}^n)$ is isomorphic to space $G_b^{k,\widetilde\omega}({\mbf R}^n)$ having the metric approximation property. However, the distortion of this isomorphism depends on $\omega$. So, in general, it is not clear whether $G_b^{k,\omega}({\mbf R}^n)$ itself has the metric approximation property. In fact, in some cases spaces $G_b^{k,\omega}(S)$ still have the metric approximation property. E.g., by the classical result of Grothendieck \cite[Ch.\,I]{G}, separable dual spaces with the approximation property have the metric approximation property. The class of such spaces $G_b^{k,\omega}(S)$ is studied in the next section. \begin{R}\label{k} {\rm It is not known, even for the case $k=0$, whether all spaces $C_b^{k,\omega}({\mbf R}^n)$ have the approximation property (for some results in this direction for $k=0$ see, e.g., \cite{K}). } \end{R} At the end of this section we formulate a result describing the structure of operators in ${\mathcal L}(G_b^{k,\omega}({\mbf R}^n);X)$, where $X$ is a separable Banach space with the $\lambda$-approximation property. In particular, it can be applied to $X=G_b^{k,\omega}(S)$ and $\lambda:=\lambda(S,k,n,\omega)$ the constant of the approximation property for $G_b^{k,\omega}(S)$ of Theorem \ref{te1.3}\,(2). \begin{Th}\label{teor1.10} There exists the family of norm one vectors $\{v_j\}_{j\in{\mbf N}}\subset X$ and given $H\in {\mathcal L}(G_b^{k,\omega}({\mbf R}^n);X)$ the family of functions $\{h_j\}_{j\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ of norms $\le 32\cdot\lambda^2\cdot\|H\|$ such that for all $x\in{\mbf R}^n$, $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, \begin{equation}\label{equa1.7} H(\delta_x^\alpha)=\sum_{j=1}^\infty D^\alpha h_j(x)\cdot v_j \end{equation} (convergence in $X$). \end{Th} \begin{R} {\rm If $X=G_b^{k,\omega}(S)$ and $H\in\mathcal L(G_b^{k,\omega}({\mbf R}^n);G_b^{k,\omega}(S))$ is a projection onto $G_b^{k,\omega}(S)$, then in addition to \eqref{equa1.7} we have \begin{equation}\label{equa1.8} \delta_x^0=\sum_{j=1}^\infty h_j(x)\cdot v_j\quad {\rm for\ all}\quad x\in S. \end{equation} In this case, the adjoint $H^*$ of $H$ belongs to $Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ and for all $x\in{\mbf R}^n$, $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, the extension $H^*f$ of $f\in C_b^{k,\omega}(S)$ satisfies \begin{equation}\label{equa1.9} D^\alpha (H^*f)(x):=\sum_{j=1}^\infty D^\alpha h_j(x)\cdot f(v_j). \end{equation} } \end{R} \subsection{Preduals of ${\mathbf G_b^{k,\omega}(S)}$ Spaces} Let $C_{0}^{k,\omega}({\mbf R}^n)$ be the subspace of functions $f\in C_b^{k,\omega}({\mbf R}^n)$ such that \begin{itemize} \item[(i)] for all $\alpha\in{\mbf Z}_+^n$, $0\le |\alpha|\le k$, \[ \lim_{\|x\|\rightarrow\infty}D^\alpha f(x)=0; \] \item[(ii)] for all $\alpha\in{\mbf Z}_+^n$, $|\alpha|=k$, \[ \lim_{\|x-y\|\rightarrow 0}\frac{D^\alpha f(x)-D^\alpha f(y)}{\omega(\|x-y\|)}=0. \] \end{itemize} It is easily seen that $C^{k,\omega}_0({\mbf R}^n)$ equipped with the norm induced from $C^{k,\omega}_b({\mbf R}^n)$ is a Banach space. By $C^{k,\omega}_0(S)$ we denote the trace of $C^{k,\omega}_0({\mbf R}^n)$ to a closed subset $S\subset{\mbf R}^n$ equipped with the trace norm. If $\lim_{t\rightarrow 0^+}\,\frac{t}{\omega(t)}> 0$ (see condition (i) for $\omega$ in section~1.1), then clearly, the corresponding space $C^{k,\omega}_0({\mbf R}^n)$ is trivial. Thus we may naturally assume that $\omega$ satisfies the condition \begin{equation}\label{omega2} \lim_{t\rightarrow 0^+}\,\frac{t}{\omega(t)}=0. \end{equation} In the sequel, the weak$^*$ topology of $C_b^{k,\omega}(S)$ is defined by means of functionals in $G_b^{k,\omega}(S)\subset \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$. Convergence in the weak$^*$ topology is described in section~4.2. \begin{Th}\label{te1.4} Suppose $\omega$ satisfies condition \eqref{omega2}. \begin{enumerate} \item Space $\bigl(C^{k,\omega}_0({\mbf R}^n)\bigr)^*$ is isomorphic to $G_b^{k,\omega}({\mbf R}^n)$, isometrically if $\displaystyle \lim_{t\rightarrow\infty}\omega(t)=\infty$. \item If there exists a weak$^*$ continuous operator $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ such that $T\bigl(C_0^{k,\omega}(S)\bigr)\subset C_0^{k,\omega}({\mbf R}^n)$, then $\bigl(C^{k,\omega}_0(S)\bigr)^*$ is isomorphic to $G_b^{k,\omega}(S)$. \end{enumerate} \end{Th} From the first part of the theorem we obtain (for $\omega$ satisfying \eqref{omega2}): \begin{C}\label{cor1.10} The space of $C^\infty$ functions with compact supports on ${\mbf R}^n$ is dense in $C^{k,\omega}_0({\mbf R}^n)$. In particular, all spaces $C^{k,\omega}_0(S)$ are separable. \end{C} It is not clear whether the condition of the second part of the theorem is valid for all spaces $C_b^{k,\omega}(S)$ with $\omega$ subject to \eqref{omega2}. Here we describe a class of sets $S$ satisfying this condition. As before, by $\mathcal P_{k,n}$ we denote the space of real polynomials on ${\mbf R}^n$ of degree $k$, and by $Q_r(x)\subset {\mbf R}^n$ the closed cube centered at $x$ of sidelength $2r$. \begin{D}\label{wm} A point $x$ of a subset $S\subset{\mbf R}^n$ is said to be weak $k$-Markov if \[ \varliminf_{r\rightarrow 0}\left\{\sup_{p\in\mathcal P_{k,n}\setminus 0}\left(\frac{\sup_{Q_r(x)}|p|}{\sup_{Q_r(x)\cap S}|p|} \right) \right\}<\infty . \] A closed set $S\subset{\mbf R}^n$ is said to be weak $k$-Markov if it contains a dense subset of weak $k$-Markov points. \end{D} The class of weak $k$-Markov sets, denoted by ${\rm Mar}^*_k({\mbf R}^n)$, was introduced and studied by Yu.~Brudnyi and the author, see \cite{BB1, B}. It contains, in particular, the closure of any open set, the Ahlfors $p$-regular compact subsets of ${\mbf R}^n$ with $p > n-1$, a wide class of fractals of arbitrary positive Hausdorff measure, direct products $\prod_{j=1}^l S_j$, where $S_j\in {\rm Mar}^*_k({\mbf R}^{n_j})$, $1\le j\le l$, $n=\sum_{j=1}^l n_j$, and closures of unions of any combination of such sets. Solutions of the Whitney problems (see sections 1.2 and 1.3 above) for sets in ${\rm Mar}^*_k({\mbf R}^n)$ are relatively simple, see \cite{BB1}. We prove the following result. \begin{Th}\label{te1.11} Let $S'\in {\rm Mar}^*_k({\mbf R}^n)$ and $\omega$ satisfy \eqref{omega2}. Suppose $H:{\mbf R}^n\rightarrow{\mbf R}^n$ is a differentiable map such that \begin{itemize} \item[(a)] the entries of its Jacobian matrix belong to $C_b^{k-1,\omega_o}({\mbf R}^n)$, where $\omega_o$ satisfies \begin{equation}\label{equ1.8} \lim_{t\rightarrow 0^+}\frac{\omega_o(t)}{\omega(t)}=0; \end{equation} \item[(b)] the map $H|_{S'}:S'\rightarrow S=:H(S')$ is a proper retraction.\footnote{I.e., $S\subset S'$ and $H|_{S'}(x)=x$ for all $x\in S$, and for each compact $K\subset S$ its preimage $(H|_{S'})^{-1}(K)$ is compact.} \end{itemize} Then the condition of Theorem \ref{te1.4} holds for $C_b^{k,\omega}(S)$. Thus $G_b^{k,\omega}(S)$ is isomorphic to $\bigl(C^{k,\omega}_0(S)\bigr)^*$ and so $G_b^{k,\omega}(S)$ and $C^{k,\omega}_0(S)$ have the metric approximation property. \end{Th} \begin{R} {\rm (1) In addition to weak $k$-Markov sets $S\subset{\mbf R}^n$, Theorem \ref{te1.11} is valid, e.g., for a compact subset $S$ of a $C^{k+1}$-manifold $M\subset{\mbf R}^n$ such that the base of the topology of $S$ consists of relatively open subsets of Hausdorff dimension $> {\rm dim}\,M - 1$. Indeed, in this case there exist tubular open neighbourhoods $U_M\subset V_M\subset{\mbf R}^n$ of $M$ such that ${\rm cl}(U_M)\subset V_M$ together with a $C^{k+1}$ retraction $r: U_M\rightarrow M$. Then, due to the hypothesis for $S$, the base of topology of $S':=r^{-1}(S)\cap {\rm cl}(U_M) $ consists of relatively open subsets of Hausdorff dimension $>n-1$ and so $S'\in {\rm Mar}_p^*({\mbf R}^n)$ for all $p\in{\mbf N}$, see, e.g., \cite[page\,536]{B}. Moreover, it is easily seen that $r|_{S'}$ is the restriction to $S'$ of a map $H\in C_b^{k+1}({\mbf R}^n; {\mbf R}^n)$. Decreasing $V_M$, if necessary, we may assume that $S'$ is compact, and so the triple $(H, S', S)$ satisfies the hypothesis of the theorem. \noindent (2) Under conditions of Theorem \ref{te1.11}, $C_b^{k,\omega}(S)$ is isomorphic to the second dual of $C^{k,\omega}_0(S)$.} \end{R} \section{Proof of Theorem \ref{te1.2}} By $\delta_x^\alpha$, $x\in{\mbf R}^n$, $\alpha\in\mathbb Z^n_+$, we denote the evaluation functional $D^\alpha|_{\{x\}}$. By definition each $\delta_x^\alpha$, $|\alpha|\le k$, belongs to $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ and has norm $\le 1$. Similarly, functionals $\frac{\delta_x^\alpha-\delta_y^\alpha}{\omega(\|x-y\|)}$, $|\alpha|=k$, $x,y\in{\mbf R}^n$, $x\ne y$, belong to $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ and have norm $\le 1$. \begin{Proposition}\label{p2.1} The closed unit ball $B$ of $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ is the balanced weak$\,^*$ closed convex hull of the set $V$ of all functionals $\delta_x^\alpha$, $|\alpha|\le k$, and $\frac{\delta_x^\alpha-\delta_y^\alpha}{\omega(\|x-y\|)}$, $|\alpha|=k$, $x,y\in{\mbf R}^n$, $x\ne y$. \end{Proposition} \begin{proof} Clearly, $V\subset B$ and therefore the required hull $\widehat V\subset B$ as well. Assume, on the contrary, that $\widehat V\ne B$. Then due to the Hahn-Banach theorem there exists an element $f\in C^{k,\omega}_b({\mbf R}^n)$ of norm one such that $\sup_{v\in\widehat V}|v(f)|\le c<1$. Since $V\subset \widehat V$, this implies \[ \|f\|_{C^{k,\omega}_b({\mbf R}^n)}\le c<1, \] a contradiction proving the result. \end{proof} Let $X$ be the minimal closed subspace of $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ containing $V$. \begin{Proposition}\label{p2.2} $X^*$ is isometrically isomorphic to $C^{k,\omega}_b({\mbf R}^n)$. \end{Proposition} \begin{proof} For $h\in X^*$ we set $H(x):=h(\delta^0_x)$, $x\in\mathbb R^n$. Let $e_1,\dots, e_n$ be the standard orthonormal basis in ${\mbf R}^n$. By the mean-value theorem for functions in $C^{k,\omega}_b({\mbf R}^n)$ we obtain, for all $\alpha\in\mathbb Z^n_+$, $|\alpha|<k$, $x\in{\mbf R}^n$, \begin{equation}\label{eq2.6} \lim_{t\rightarrow 0}\frac{\delta^{\alpha}_{x+t\cdot e_i}-\delta^\alpha_x}{t}=\delta^{\alpha+e_i}_x \end{equation} (convergence in $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*)$. From here by induction we deduce easily that $H\in C^k({\mbf R}^n)$ and for all $\alpha\in\mathbb Z^n_+$, $|\alpha|\le k$, $x\in{\mbf R}^n$, \[ h(\delta^\alpha_x)=D^\alpha H(x). \] This shows that $H\in C^{k,\omega}_b({\mbf R}^n)$ and $\|H\|_{C^{k,\omega}_b({\mbf R}^n)}\le \|h\|_{X^*}$. Considering $H$ as the bounded linear functional on $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ we obtain that $H|_V=h|_V$. Thus, by the definition of $X$, \[ H|_{X}=h. \] Since the unit ball of $X$ is $B\cap X$, \[ \|h\|_{X^*}\le \|H\|_{C^{k,\omega}_b({\mbf R}^n)}\, \bigl(\le \|h\|_{X^*}\bigr). \] Hence, the correspondence $h\mapsto H$ determines an isometry $I :X^*\rightarrow C_b^{k,\omega}({\mbf R}^n)$. Since the restriction of each $H\in C_b^{k,\omega}({\mbf R}^n)$, regarded as the bounded linear functional on $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$, to $X$ determines some $h\in X^*$, map $I$ is surjective. This completes the proof of the proposition. \end{proof} Note that equation \eqref{eq2.6} shows that the minimal closed subspace $G_b^{k,\omega}({\mbf R}^n)\subset\bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$ containing all $\delta_x^0$, $x\in{\mbf R}^n$, coincides with $X$. Thus, $\bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ is isometrically isomorphic to $C^{k,\omega}_b({\mbf R}^n)$; this proves Theorem \ref{te1.2} for $S={\mbf R}^n$. \begin{C}\label{cor2.3} The closed unit ball of $G^{k,\omega}_b({\mbf R}^n)$ is the balanced closed convex hull of the set $V$ of all functionals $\delta_x^\alpha$, $|\alpha|\le k$, and $\frac{\delta_x^\alpha-\delta_y^\alpha}{\omega(\|x-y\|)}$, $|\alpha|=k$, $x,y\in{\mbf R}^n$, $x\ne y$. \end{C} \begin{proof} The closed unit ball of $G^{k,\omega}_b({\mbf R}^n)$ is $B\cap G^{k,\omega}_b({\mbf R}^n)$. Since the weak$^*$ topology of $\bigl(C^{k,\omega}_b({\mbf R}^n)\bigr)^*$ induces the weak topology of $G^{k,\omega}_b({\mbf R}^n)$ and the weak closure of the balanced convex hull of $V$ coincides with the norm closure of this set, the result follows from Proposition \ref{p2.1}. \end{proof} Now, let us consider the case of general $S\subset\mathbb R^n$. Let $h\in \bigl(G_b^{k,\omega}(S)\bigr)^*$. We set $H(x):=h(\delta_x^0)$, $x\in S$. Due to the Hahn-Banach theorem, there exists $\tilde h\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $\tilde h|_{G_b^{k,\omega}(S)}=h$ and $\|\tilde h\|_{(G_b^{k,\omega}({\mbf R}^n))^*}=\|h\|_{(G_b^{k,\omega}(S))^*}$. Let us define $\widetilde H(x)=\tilde h(\delta_x^0)$, $x\in {\mbf R}^n$. According to Proposition \ref{p2.2}, $\widetilde H\in C^{k,\omega}_b({\mbf R}^n)$ and $\|\widetilde H\|_{C_b^{k,\omega}({\mbf R}^n)}=\|\tilde h\|_{(G_b^{k,\omega}({\mbf R}^n))^*}$. Moreover, $\widetilde H|_S=H$. This implies that $H\in C_b^{k,\omega}(S)$ and has norm $\le \|h\|_{(G_b^{k,\omega}(S))^*}$. Hence, the correspondence $h\mapsto H$ determines a bounded nonincreasing norm linear injection $I_S:\bigl(G_b^{k,\omega}(S)\bigr)^*\rightarrow C_b^{k,\omega}(S)$. Let us show that $I_S$ is a surjective isometry. Indeed, for $H\in C_b^{k,\omega}(S)$ there exists $\widetilde H\in C_b^{k,\omega}({\mbf R}^n)$ such that $\widetilde H|_S=H$ and $\|\widetilde H\|_{C_b^{k,\omega}({\mbf R}^n)}=\|H\|_{C_b^{k,\omega}(S)}$. In turn, due to Proposition \ref{p2.2} there exists $\tilde h\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $\widetilde H(x)=\tilde h(\delta_x^0)$, $x\in{\mbf R}^n$, and $\|\widetilde H\|_{C_b^{k,\omega}({\mbf R}^n)}=\|\tilde h\|_{(G_b^{k,\omega}({\mbf R}^n))^*}$. We set $h:=\tilde h|_{G_b^{k,\omega}(S)}$. Then $h\in \bigl(G_b^{k,\omega}(S)\bigr)^*$ and $H(x)=h(\delta_x^0)$, $x\in S$, i.e., $I_S(h)=H$ and \[ \bigl(\|h\|_{(G_b^{k,\omega}(S))^*}\ge\bigr)\, \|I_S(h)\|_{C_b^{k,\omega}(S)}\ge \|h\|_{(G_b^{k,\omega}(S))^*}. \] The proof of Theorem \ref{te1.2} is complete. \section{Proofs of Theorems \ref{teo1.5}, \ref{te1.6}, \ref{teo1.6}} \subsection{Proof of Theorem \ref{teo1.5}} \begin{proof} According to the Finiteness Principle there exist constants $d\in{\mbf N}$ and $c\in (1,\infty)$ such that for all $f\in C_b^{k,\omega}(S)$, \begin{equation}\label{e3.13} \sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\|f\|_{C_b^{k,\omega}(S')}\le\|f\|_{C_b^{k,\omega}(S)}\le c\cdot\left(\sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\|f\|_{C_b^{k,\omega}(S')}\right). \end{equation} Here for $k=0$, $d=2$ (-\,optimal) and $c=1$, see \cite{McS}, for $n=1$, $d=k+2$ (-\,optimal) and $c$ depends on $k$ only, see \cite{M}, for $k=1$, $d=3\cdot 2^{n-1}$ (-\,optimal) and $c$ depends on $k$ and $n$ only, see \cite{BS1}, and for $k\ge 2$, $d=2^{ k+n \choose k}$ and $c=\frac{\tilde c}{\omega(1)}$, where $\tilde c$ depends on $k$ and $n$ only, see \cite{F1} and \cite{BM}, \cite{S}. Considering $f$ as the bounded linear functional on $G_b^{k,\omega}(S)$, we get from \eqref{e3.13} the required implications \[ B_b^{k,\omega}(S;d)\subset B_b^{k,\omega}(S)\subset c\cdot B_b^{k,\omega}(S;d). \] Indeed, suppose, on the contrary, that there exists $v\in B_b^{k,\omega}(S)\setminus c\cdot B_b^{k,\omega}(S;d)$. Let $f\in C_b^{k,\omega}(S)$ be such that \[ \sup_{c\cdot B_b^{k,\omega}(S;d)}|f|<|f(v)|. \] By the definition of $B_b^{k,\omega}(S;d)$ the left-hand side of the previous inequality coincides with $c\cdot\bigl(\sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\,\|f\|_{C_b^{k,\omega}(S')}\bigr)$. Hence, \[ c\cdot\left(\sup_{S'\subset S\,;\, {\rm card}\,S'\le d}\|f\|_{C_b^{k,\omega}(S')}\right)<|f(v)|\le \|f\|_{C_b^{k,\omega}(S)}, \] a contradiction with \eqref{e3.13}. \end{proof} \subsection{Proof of Theorem \ref{te1.6}} \begin{proof} We set \begin{equation}\label{e4.14} r_{X}(F)(s):=F(\delta_s^0),\quad F\in \mathcal L(G_b^{k,\omega}({\mbf R}^n); X),\quad s\in {\mbf R}^n. \end{equation} Applying the arguments similar to those of Proposition \ref{p2.2} we obtain \[ r_{X}(F)\in C_b^{k,\omega}({\mbf R}^n;X)\quad {\rm and}\quad \|r_X(F)\|_{C_b^{k,\omega}({\mbf R}^n;X)}\le \|F\|_{\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)}. \] On the other hand, for each $\varphi\in X^*$, $\|\varphi\|_{X^*}=1$, function $r_{{\mbf R}}(\varphi\circ F)\in C_b^{k,\omega}({\mbf R}^n)$. So, since $r_{{\mbf R}}(\varphi\circ F)=\varphi (r_{X}(F))$, \[ \|\varphi\circ F\|_{(G_b^{k,\omega}({\mbf R}^n))^*}=\|r_{{\mbf R}}(\varphi\circ F)\|_{C_b^{k,\omega}({\mbf R}^n)}=\|\varphi (r_{{\mbf R}^n;X}(F))\|_{C_b^{k,\omega}({\mbf R}^n)}\le \|r_{X}(F)\|_{C_b^{k,\omega}({\mbf R}^n;X)}. \] Taking supremum over all such $\varphi$ we get \[ \| F\|_{\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)}\le \|r_{X}(F)\|_{C_b^{k,\omega}({\mbf R}^n;X)}\, \bigl(\le \| F\|_{\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)}\bigr). \] This shows that $r_{X}:\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)\rightarrow C_b^{k,\omega}({\mbf R}^n;X)$ is an isometry. Let us prove that it is surjective. Since every finite subset of ${\mbf R}^n$ is interpolating for $C_b^{k,\omega}({\mbf R}^n)$, the set of vectors $\delta_s^0\in G_b^{k,\omega}({\mbf R}^n)$, $s\in{\mbf R}^n$, is linearly independent. Hence, each $f\in C_b^{k,\omega}({\mbf R}^n;X)$ determines a linear map $\hat f:{\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}\rightarrow X$, \[ \hat f\left(\sum_{j}c_j\delta_{s_j}^0\right):=\sum_j c_j f(s_j),\quad \sum_{j}c_j\delta_{s_j}^0\in {\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}\, (\subset G_b^{k,\omega}({\mbf R}^n)). \] Next, for each $\varphi\in X^*$, $\|\varphi\|_{X^*}=1$, function $\varphi\circ f\in C_b^{k,\omega}({\mbf R}^n)$ and \[ \|\varphi\circ f\|_{C_b^{k,\omega}({\mbf R}^n)}\le\|f\|_{C_b^{k,\omega}({\mbf R}^n;X)}. \] Since $r_{{\mbf R}}:\bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*\rightarrow C_b^{k,\omega}({\mbf R}^n)$ is an isometric isomorphism, there exists $\ell_{\varphi\circ f}\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $r_{{\mbf R}}(\ell_{\varphi\circ f})=\varphi\circ f$. Clearly, $\ell_{\varphi\circ f}$ coincides with $\varphi\circ\hat f$ on ${\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}$ and for all $v\in G_b^{k,\omega}({\mbf R}^n)$, \[ \begin{array}{r} \displaystyle |\ell_{\varphi\circ f}(v)|\le \| \ell_{\varphi\circ f}\|_{(G_b^{k,\omega}({\mbf R}^n))^*}\cdot \|v\|_{G_b^{k,\omega}({\mbf R}^n)} =\|\varphi\circ f\|_{C_b^{k,\omega}({\mbf R}^n)}\cdot \|v\|_{G_b^{k,\omega}({\mbf R}^n)}\medskip\\ \displaystyle \le \|f\|_{C_b^{k,\omega}({\mbf R}^n;X)}\cdot \|v\|_{G_b^{k,\omega}({\mbf R}^n)}. \end{array} \] These imply that $\hat f:{\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\}\rightarrow X$ is a linear continuous operator of norm $\le \|f\|_{C_b^{k,\omega}({\mbf R}^n;X)}$. Hence, it extends to a bounded linear operator $F:{\rm cl}({\rm span}\{\delta_s^0\, :\, s\in {\mbf R}^n\})=:G_b^{k,\omega}({\mbf R}^n)\rightarrow X$ such that $r_{X}(F)=f$. Thus, $r_{X}(F):\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)\rightarrow C_b^{k,\omega}({\mbf R}^n;X)$ is an isometric isomorphism. The proof of the theorem is complete. \end{proof} \subsection{Proof of Theorem \ref{teo1.6}} \begin{proof} Without loss of generality we may assume that $T$ has depth $d$ and is defined by \eqref{equ1.6}. Let $T:\bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*\rightarrow \bigl(C_b^{k,\omega}(S)\bigr)^*$ be the adjoint of $T$ and $q_S^*:\bigl(C_b^{k,\omega}(S)\bigr)^*\rightarrow \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$ the adjoint of the quotient map $q_S: C_b^{k,\omega}({\mbf R}^n)\rightarrow C_b^{k,\omega}(S)$. Clearly, $q_S^*$ is an isometric embedding which maps the closed subspace of $\bigl(C_b^{k,\omega}(S)\bigr)^*$ generated by $\delta$-functionals of points in $S$ isometrically onto $G_b^{k,\omega}(S)\subset \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$. We define \begin{equation}\label{proj} P:=q_S^*\circ T^*. \end{equation} By the definition of $T_S$, for each $\delta_x^0\in G_b^{k,\omega}({\mbf R}^n)$, $x\in{\mbf R}^n\setminus S$, and $f\in C_b^{k,\omega}(S)$ we have, for some $y_i^x\in S$, \[ (P\delta_x^0)(f)=\delta_x^0(Tq_S f)=\sum_{i=1}^d \lambda_i^x\cdot f(y_i^x)=\left(\sum_{i=1}^d \lambda_i^x\cdot\delta_{y_i^x}^0\right)(f). \] Hence, \begin{equation}\label{equ3.11} P\delta_x^0=\sum_{i=1}^d \lambda_i^x\cdot\delta_{y_i^x}^0\quad {\rm for\ all}\quad x\in{\mbf R}^n\setminus S. \end{equation} Since $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$, \begin{equation}\label{equ3.12} P\delta_x^0=\delta_x^0\quad {\rm for\ all}\quad x\in S. \end{equation} Thus $P$ maps $G_b^{k,\omega}({\mbf R}^n)$ into $G_b^{k,\omega}(S)$ and is identity on $G_b^{k,\omega}(S)$. Hence, $P$ is a bounded linear projection of norm $\|P\|\le \|T q_S\|\le \|T\|$. Next, under the identification $\bigl(G_b^{k,\omega}(S)\bigr)^*=C_b^{k,\omega}(S)$ for all closed $S\subset{\mbf R}^n$ (see Theorem \ref{te1.2}), for all $x\in{\mbf R}^n\setminus S$, and $f\in C_b^{k,\omega}(S)$ we have by \eqref{equ3.11} \[ (P^*f)(\delta_x^0)=f(P\delta_x^0)=f\left(\sum_{i=1}^d \lambda_i^x\cdot\delta_{y_i^x}^0\right)=\sum_{i=1}^d \lambda_i^x\cdot f(y_i^x)=(Tf)(\delta_x^0). \] The same identity is valid for $x\in S$, cf. \eqref{equ3.12}. This implies that $P^*=T$ and completes the proof of the theorem. \end{proof} \section{Proofs of Theorems \ref{te1.3} and \ref{teor1.10}} Sections 4.1 and 4.2 contain auxiliary results used in the proof of Theorem \ref{te1.3}. \subsection{Jackson Theorem} Recall that the {\em Jackson kernel} $J_N$ is the trigonometric polynomial of degree $2\widetilde N$, where $\widetilde N:= \left\lfloor\frac N2\right\rfloor$, given by the formula \[ J_N(t)=\gamma_N\left(\frac{\sin\frac{\widetilde N t}{2}}{\sin \frac t2}\right)^4, \] where $\gamma_N$ is chosen so that $\displaystyle \int_{-\pi}^\pi J_N(t)\, dt =1$. For a $2\pi$-periodic real function $f\in C({\mbf R})$ we set \begin{equation}\label{jack1} (L_N f)(x):=\int_{-\pi}^\pi f(x-t)J_N(t)\, dt,\quad x\in\mathbb R. \end{equation} Then the classical {\em Jackson theorem} asserts (see, e.g., \cite[Ch.\,V]{T}): $L_N f$ is a real trigonometric polynomial of degree at most $N$ such that \begin{equation}\label{jack2} \sup_{x\in (-\pi,\pi)}|f(x)-(L_N f)(x)|\le c\,\omega\bigl(f, \mbox{$\frac 1N$}\bigr), \end{equation} for a numerical constant $c>0$; here $\omega(f,\cdot)$ is the modulus of continuity of $f$.\subsection{Convergence in the Weak$^*$ Topology of ${\mathbf C_b^{k,\omega}({\mbf R}^n)}$} In the proof of Theorem \ref{te1.3} we use the following result. As before, we equip $C_b^{k,\omega}({\mbf R}^n)$ with the weak$^*$ topology induced by means of functionals in $G_b^{k,\omega}({\mbf R}^n)\subset \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$. \begin{Proposition}\label{prop3.1} A sequence $\{f_i\}_{i\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ weak$\,^*$ converges to $f\in C_b^{k,\omega}({\mbf R}^n)$ if and only if \begin{itemize} \item[(a)] \[ \sup_{i\in{\mbf N}}\|f_i\|_{C_b^{k,\omega}({\mbf R}^n)}<\infty; \] \item[(b)] For all $\alpha\in\mathbb Z_+^n$, $0\le |\alpha|\le k$, $x\in{\mbf R}^n$ \[ \lim_{i\rightarrow\infty}D^\alpha f_i(x)=D^\alpha f(x). \] \end{itemize} \end{Proposition} \begin{proof} Without loss of generality we may assume that $f=0$. If $\{f_i\}_{i\in{\mbf N}}$ weak$^*$ converges to $0$, then (a) follows from the Banach-Steinhaus theorem and (b) from the fact that each $\delta_x^\alpha\in G_b^{k,\omega}({\mbf R}^n)$. Conversely, suppose that $\{f_i\}_{i\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ satisfies (a) and (b) with $f=0$. Let $g\in G_b^{k,\omega}({\mbf R}^n)$. According to Corollary \ref{cor2.3}, given $\varepsilon>0$ there exist $J\in{\mbf N}$ and families $c_{j\alpha}\in {\mbf R}$, $x_{j\alpha}\in{\mbf R}^n$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $0\le |\alpha|<k$, and $d_{j\alpha}\in {\mbf R}$, $x_{j\alpha}, y_{j\alpha}\in{\mbf R}^n$, $x_{j\alpha}\ne y_{j\alpha}$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $|\alpha|=k$, such that \[ g=\sum_{j,\alpha}c_{j\alpha}\delta_{x_{j\alpha}}^\alpha+\sum_{j,\alpha}d_{j\alpha}\frac{\delta_{x_{j\alpha}}^\alpha-\delta_{y_{j\alpha}}^\alpha}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)}+g', \] where \[ \sum_{j,\alpha}|c_{j\alpha}|+\sum_{j,\alpha}|d_{j\alpha}|\le \|g\|_{G_b^{k,\omega}({\mbf R}^n)}\quad {\rm and}\quad \|g'\|_{G_b^{k,\omega}({\mbf R}^n)}<\frac{\varepsilon}{2M},\quad M:=\sup_{i\in{\mbf N}}\|f_i\|_{C_b^{k,\omega}({\mbf R}^n)}. \] Further, due to condition (b), there exists $I\in{\mbf N}$ such that for all $i\ge I$, \[ \left|f_i\left(\sum_{j,\alpha}c_{j\alpha}\delta_{x_{j\alpha}}^\alpha+\sum_{j,\alpha}d_{j\alpha}\frac{\delta_{x_{j\alpha}}^\alpha-\delta_{y_{j\alpha}}^\alpha}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)} \right) \right|<\frac{\varepsilon}{2}. \] Also, for such $i$, \[ |f_i(g')|\le \|f_i\|_{C_b^{k,\omega}({\mbf R}^n)}\cdot \|g'\|_{G_b^{k,\omega}({\mbf R}^n)}< M\cdot \frac{\varepsilon}{2M}=\frac{\varepsilon}{2}. \] Combining these inequalities we obtain for all such $i$: \[ |f_i(g)|<\varepsilon. \] This shows that $\lim_{i\rightarrow\infty}f_i(g)=0$. Thus $\{f_i\}_{i\in{\mbf N}}$ weak$^*$ converges to $0$, as required. \end{proof} \subsection{Proof of Theorem \ref{te1.3}\,(1)} We set \[ \mathbb K_N^n:=\bigl\{x=(x_1,\dots, x_n)\in{\mbf R}^n\, :\, \max_{1\le i\le n}|x_i|\le N\bigr\}. \] Let $\rho:\mathbb R^n\rightarrow [0,1]$ be a fixed $C^\infty$ function with support in the cube $\mathbb K_2^n$, equals one on the unit cube $\mathbb K_1^n$. For a natural number $\ell$ we set $\rho_\ell(x):=\rho(x/\ell)$, $x\in\mathbb R^n$. Then there exist constants $c_{k,n}$ (depending on $k$ and $n$) such that \begin{equation}\label{rhol} \sup_{x\in\mathbb R^n}|D^\alpha\rho_\ell(x)|\le \frac{c_{k,n}}{\ell^{|\alpha|}}\quad {\rm for\ all}\quad \alpha\in\mathbb Z_+^n\quad {\rm with }\quad |\alpha|\le k+1. \end{equation} Let $f\in C_b^{k,\omega}(\mathbb R^n)$. We define a $8\ell\sqrt n\,$-periodic in each variable function $f_\ell$ on $\mathbb R^n$ by \begin{equation}\label{eq3.10} f_\ell(v+x)=\rho_\ell(x)\cdot f(x),\qquad v+x\in 8\ell\sqrt n\cdot\mathbb Z^n+\mathbb K_{4\ell\sqrt n}^n. \end{equation} Note that $f_\ell$ coincides with $f$ on the cube $\mathbb K_\ell^n$. \begin{Lm}\label{norm} There exists a constant $C_\ell=C(\ell,k,n,\omega)$ (i.e., depending on $\ell, k,n$ and $\omega$) such that \[ \lim_{\ell\rightarrow\infty}C_{\ell}=1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}\,\frac{1}{\omega(t)}, \] and \[ \|f_\ell\|_{C_b^{k,\omega}(\mathbb R^n)}\le C_\ell\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}. \] \end{Lm} \begin{proof} We use the standard multi-index notation. According to the general Leibniz rule, for $\alpha\in\mathbb Z_+^n$, $|\alpha|\le k$, \[ D^\alpha f_\ell=\sum_{\nu\, :\, \nu\le\alpha}\binom{\alpha}{\nu}(D^\nu\rho_\ell)\cdot (D^{\alpha-\nu}f)\quad {\rm on}\quad \mathbb K_{4\ell\sqrt n}^n . \] From here and \eqref{rhol} we get for $\alpha\in \mathbb Z_+^n$, $|\alpha|\ge 1$, \[ \begin{array}{l} \displaystyle \sup_{x\in\mathbb R^n}|D^\alpha f_\ell(x)|\le \|f\|_{C_b^k(\mathbb R^n)}\cdot\left(\sum_{\nu\, :\, 0< \nu\le\alpha}\binom{\alpha}{\nu}\cdot\frac{c_{k,n}}{\ell^{|\nu|}}+1\right)\medskip\\ \displaystyle \le \|f\|_{C_b^k(\mathbb R^n)}\cdot\left(c_{k,n}\cdot\left(\left(1+\frac 1\ell \right)^{|\alpha|}-1\right)+1\right)\le \|f\|_{C_b^k(\mathbb R^n)}\cdot\left( \left(1+\frac 1\ell\right)^{|\alpha|-1}\cdot \frac{c_{k,n}\cdot |\alpha|}{\ell}+1 \right). \end{array} \] Hence, \[ \|f_\ell\|_{C_b^k(\mathbb R^n)}\le\left( \left(1+\frac 1\ell\right)^{\max\{k-1,0\}}\cdot \frac{c_{k,n}\cdot k}{\ell}+1 \right)\cdot \|f\|_{C_b^k(\mathbb R^n)}=:C_1(\ell,k,n)\cdot \|f\|_{C_b^k(\mathbb R^n)}. \] Similarly, for $\alpha\in\mathbb Z_+^n$, $|\alpha|=k$, and $x,y \in \mathbb K_{2\ell}^n$ using properties of $\omega$ we obtain\smallskip \begin{equation}\label{eq2.9} \begin{array}{lr} \qquad\\ \displaystyle |D^\alpha f_\ell (x)-D^\alpha f_\ell (y)|\medskip\\ \displaystyle \le \sum_{\nu\, :\, \nu\le\alpha}\binom{\alpha}{\nu}\cdot \bigl(| D^\nu \rho_\ell(x)-D^\nu\rho_\ell(y)|\cdot |D^{\alpha-\nu}f(x) |+| D^\nu\rho_\ell(y)|\cdot |D^{\alpha-\nu}f(x)-D^{\alpha-\nu}f(y)|\bigr)\medskip\\ \displaystyle \le \|f\|_{C_b^{k,\omega}(\mathbb R^n)} \cdot\left(\sum_{\nu\, :\, 0<\nu\le\alpha}\binom{\alpha}{\nu}\cdot c_{k,n}\cdot\left(\frac{\|x-y\|}{\ell^{|\nu|+1}}+\frac{\|x-y\|}{\ell^{|\nu|}}\right)+\frac{c_{k,n}\cdot\|x-y\|}{\ell}+\omega(\|x-y\|)\right)\medskip\\ \displaystyle = \|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot\left(c_{k,n}\cdot\left(\left(1+\frac 1\ell\right)^{k+1}-1\right)\cdot \|x-y\|+\omega(\|x-y\|\right)\medskip\\ \displaystyle \le \|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|)\cdot\left( c_{k,n}\cdot \left(1+\frac 1\ell\right)^{k}\cdot\frac{4\ell\sqrt n \cdot (k+1)}{\ell\cdot\omega\bigl(4\ell \sqrt n\bigr)}+1\right)\medskip\\ \displaystyle =:\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|)\cdot C_2(\ell,k,n,\omega). \end{array} \end{equation} Observe that \begin{equation}\label{equ4.18} \lim_{\ell\rightarrow\infty}C_{2}(\ell,k,n,\omega)=1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}\,\frac{1}{\omega(t)}. \end{equation} Next, assume that $x,y\in {\rm supp}\, f_\ell$. Since the case $x,y\in \mathbb K_{2\ell}^n$ was considered above, without loss of generality we may assume that $x\in \mathbb K_{2\ell}^n$ and $y\in v+\mathbb K_{2\ell}^n$ for some $v\in (8\ell\sqrt n\cdot\mathbb Z^n)\setminus\{0\}$. Then for $y':=y-v\in \mathbb K_{2\ell}^n$ we have $D^\alpha f_\ell(y')=D^\alpha f_\ell(y)$. Also, \[ \|x-y\|=\|x-y'-v\|\ge \|v\|-\|x-y'\|\ge 8\ell\sqrt n- 4\ell\sqrt n=4\ell\sqrt n. \] Therefore for $\alpha\in\mathbb Z_+^n$, $|\alpha|=k$, \[ \begin{array}{l} \displaystyle |D^\alpha f_\ell (x)-D^\alpha f_\ell (y)|= |D^\alpha f_\ell (x)-D^\alpha f_\ell (y')|\le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y'\|)\medskip\\ \displaystyle \le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(4\ell\sqrt n)\le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|). \end{array} \] Finally, in the case $x\in {\rm supp}\, f_\ell \cap \mathbb K_{2\ell}^n$, $y\not\in {\rm supp}\, f_\ell$, there exists a point $y'\in \mathbb K_{2\ell}^n$ lying on the interval joining $x$ and $y$ such that $f_\ell(y')=0$. Since $\|x-y'\|\le \|x-y\|$ and inequality \eqref{eq2.9} is valid for $x$ and $y'$, a similar inequality is valid for $x$ and $y$. Hence, combining the considered cases we conclude that (cf. \eqref{eq5}) \[ |f_\ell|_{C_b^{k,\omega}(\mathbb R^n)}\le C_2(\ell,k,n,\omega)\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}. \] Therefore the inequality of the lemma is valid with \[ C(\ell,k,n,\omega):=\max\{C_1(\ell,k,n), C_2(\ell,k,n,\omega)\}, \] see \eqref{equ4.18}. \end{proof} We set \begin{equation}\label{eq2.10} \lambda:=\frac{4\ell\sqrt n}{\pi}. \end{equation} For a natural number $N$ and $x=(x_1,\dots, x_n)\in\mathbb R^n$ we define \begin{equation}\label{eq2.11} (E_Nf_\ell)(x)=\int_{-\pi}^\pi\cdots\int_{-\pi}^\pi f_\ell(x_1-\lambda t_1,\dots, x_n-\lambda t_n)J_N(t_1)\cdots J_N(t_n)\, dt_1\cdots dt_n. \end{equation} \begin{Lm}\label{lem2.2} Function $E_Nf_\ell$ satisfies the following properties: \begin{itemize} \item[(a)] $(E_Nf_\ell)(\lambda x)$, $x\in\mathbb R^n$, is the trigonometric polynomial of degree at most $N$ in each coordinate; \item[(b)] \[ \|E_Nf_\ell\|_{C_b^{k,\omega}(\mathbb R^n)}\le C_\ell\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}; \] \item[(c)] There is a constant $c_N$, $\lim_{N\rightarrow\infty} c_N=0$, such that \[ \|f_\ell-E_Nf_\ell\|_{C_b^{k}(\mathbb R^n)}\le c_N\cdot\|f\|_{C_b^{k,\omega}(\mathbb R^n)}. \] \end{itemize} \end{Lm} \begin{proof} (a) If we set $f_\ell^\lambda (x):=f_\ell(\lambda x)$, $x=(x_1,\dots, x_n)\in \mathbb R^n$, then \[ E_Nf_\ell(\lambda x)=(L_N^1\cdots L_N^nf_\ell^\lambda)(x), \] where $L_N^i$ is the Jackson operator \eqref{jack1} acting on univariate functions in variable $x_i$, $1\le i\le n$. This and the properties of $L_N^i$ give the required statement.\smallskip (b) According to definition \eqref{eq2.11}, for each $\alpha\in\mathbb Z_+^n$, $|\alpha|\le k$, \begin{equation}\label{eq2.12} \sup_{x\in\mathbb R^N}|D^\alpha(E_Nf_\ell)(x)|=\sup_{x\in\mathbb R^n}|(E_N D^\alpha f_\ell)(x)|\le \|f_\ell\|_{C_b^k(\mathbb R^n)}\le C_\ell\cdot \|f\|_{C_b^k(\mathbb R^n)}. \end{equation} In turn, if $|\alpha|=k$ and $x,y\in\mathbb R^n$, then \begin{equation}\label{eq2.13} \begin{array}{l} \displaystyle |D^\alpha (E_Nf_\ell)(x)-D^\alpha(E_Nf_\ell)(y)|\le |(E_N D^\alpha f_\ell)(x)- (E_ND^\alpha f_\ell)(y)|\medskip\\ \displaystyle \le |f_\ell|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|)\le C_\ell\cdot |f|_{C_b^{k,\omega}(\mathbb R^n)}\cdot \omega(\|x-y\|). \end{array} \end{equation} Thus \eqref{eq2.12} and \eqref{eq2.13} give the required statement.\smallskip (c) For each $\alpha\in\mathbb Z_+^n$, $|\alpha|\le k-1$, using \eqref{eq2.11} and due to \eqref{jack2} one obtains\smallskip \begin{equation}\label{eq2.14} \\ \begin{array}{l} \displaystyle \sup_{x\in\mathbb R^n}|D^\alpha\bigl( f_\ell -E_Nf_\ell\bigr)(x)|=\sup_{x\in\mathbb R^n}|D^\alpha f_\ell(x)-(E_ND^\alpha f_\ell)(x)|\medskip\\ \displaystyle \le \sup_{x\in\mathbb R^n}\bigl\{|D^\alpha f_\ell(\lambda x)-(L_N^1D^\alpha f_\ell)(\lambda x)|+|\bigl(L_N^1\bigl(D^\alpha f_\ell-L_N^2D^\alpha f_\ell\bigr)\bigr)(\lambda x)|+\,\cdots \medskip\\ \displaystyle + |\bigl(L_N^1\cdots L_N^{n-1}\bigl(D^\alpha f_\ell-L_N^nD^\alpha f_\ell\bigr)\bigr)(\lambda x)| \bigr\}\le \frac{c\cdot n}{N}\cdot\|f_\ell\|_{C_b^k(\mathbb R^n)}\le \frac{c\cdot n\cdot C_\ell}{N}\cdot\|f\|_{C_b^k(\mathbb R^n)}. \end{array} \end{equation} Similarly, for $|\alpha|=k$, we have\smallskip \begin{equation}\label{eq2.15} \sup_{x\in\mathbb R^n}|D^\alpha(f_\ell -E_Nf_\ell)(x)|\le c\cdot n\cdot C_\ell\cdot\omega(\mbox{$\frac 1N$})\cdot |f|_{C_b^{k,\omega}(\mathbb R^n)}. \end{equation} Equations \eqref{eq2.14} and \eqref{eq2.15} imply that the required statement is valid with \[ c_N:=c\cdot n\cdot C_\ell\cdot \max\left\{\mbox{$\frac 1N ,\, \omega(\frac 1N)$}\right\}. \] \end{proof} \begin{proof}[Proof of Theorem \ref{te1.3}\,(1)] For $f\in C_b^{k,\omega}({\mbf R}^n)$ we set \begin{equation}\label{eq3.18} L_{N,\ell}f:=E_Nf_\ell. \end{equation} According to Lemma \ref{lem2.2}, $L_{N,\ell}: C_b^{k,\omega}({\mbf R}^n)\rightarrow C_b^{k,\omega}({\mbf R}^n)$ is a finite rank bounded linear operator of norm $\le C_\ell$. \begin{Lm}\label{lem3.3} Operators $L_{N,\ell}$ are weak$\,^*$ continuous. \end{Lm} \begin{proof} Since $C_b^{k,\omega}({\mbf R}^n)=\bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*$ and $G_b^{k,\omega}({\mbf R}^n)$ is separable, $C_b^{k,\omega}({\mbf R}^n)$ equipped with the weak$^*$ topology is a Frechet space. Then $L_{N,\ell}$ is weak$^*$ continuous if and only if for each sequence $\{f_i\}_{i\in{\mbf N}}\subset C_b^{k,\omega}({\mbf R}^n)$ weak$^*$ converging to $0\in C_b^{k,\omega}({\mbf R}^n)$ the sequence $\{L_{N,\ell}f_i\}_{i\in{\mbf N}}$ weak$\,^*$ converges to $0$ as well. Note that such a sequence $\{f_i\}_{i\in{\mbf N}}$ is bounded in $C_b^{k,\omega}({\mbf R}^n)$ due to the Banach-Steinhaus theorem. Then $\{L_{N,\ell}f_i\}_{i\in{\mbf N}}$ is bounded as well and according to Proposition \ref{prop3.1} we must prove only that \begin{equation}\label{eq3.19} \lim_{i\rightarrow\infty}D^\alpha (L_{N,\ell}f_i)(x)=0\quad {\rm for\ all}\quad \alpha\in{\mbf Z}_+^n,\ 0\le |\alpha|\le k,\ x\in{\mbf R}^n. \end{equation} Further, since $D^\alpha (L_{N,\ell}f_i)(x)=(E_ND^\alpha (f_i)_\ell)(x)$ for such $\alpha$ and $x$, $\{D^\alpha (f_i)_\ell\}_{i\in{\mbf N}}$ is a bounded sequence of continuous functions and $E_N$ is the convolution operator with the absolutely integrable kernel, to establish \eqref{eq3.19} it suffices to prove (due to the Lebesgue dominated convergence theorem) that \[ \lim_{i\rightarrow\infty}D^\alpha (f_i)_\ell(x)=0 \quad {\rm for\ all}\quad \alpha\in{\mbf Z}_+^n,\ 0\le |\alpha|\le k,\ x\in{\mbf R}^n. \] The latter follows directly from the definition of $(f_i)_\ell$, see \eqref{eq3.10}, the general Leibniz rule and the fact that $D^\alpha f_i(x)\rightarrow 0$ as $i\rightarrow\infty$ for all the required $\alpha$ and $x$ (because $\{f_i\}_{i\in{\mbf N}}$ weak$^*$ converges to $0$). Thus we have proved that operators $L_{N,\ell}$ are weak$^*$ continuous. \end{proof} Lemma \ref{lem3.3} implies that there exists a bounded operator of finite rank $H_{N,\ell}$ on $G_b^{k,\omega}({\mbf R}^n)$ whose adjoint $H_{N,\ell}^*$ coincides with $L_{N,\ell}$. \begin{Lm}\label{lem3.5} The sequence of finite rank bounded operators $\{H_{N,N}\}_{N\in{\mbf N}}$ converges pointwise to the identity operator on $G_b^{k,\omega}({\mbf R}^n)$. \end{Lm} \begin{proof} Let $g\in G_b^{k,\omega}({\mbf R}^n)$. Due to Corollary \ref{cor2.3}, given $\varepsilon>0$ there exist $J\in{\mbf N}$ and families $c_{j\alpha}\in {\mbf R}$, $x_{j\alpha}\in{\mbf R}^n$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $0\le |\alpha|<k$, and $d_{j\alpha}\in {\mbf R}$, $x_{j\alpha}, y_{j\alpha}\in{\mbf R}^n$, $x_{j\alpha}\ne y_{j\alpha}$, $1\le j\le J$, $\alpha\in {\mbf Z}_+^n$, $|\alpha|=k$, such that \[ g=\sum_{j,\alpha}c_{j\alpha}\delta_{x_{j\alpha}}^\alpha+\sum_{j,\alpha}d_{j\alpha}\frac{\delta_{x_{j\alpha}}^\alpha-\delta_{y_{j\alpha}}^\alpha}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)}+g''=:g'+g'', \] where \[ \sum_{j,\alpha}|c_{j\alpha}|+\sum_{j,\alpha}|d_{j\alpha}|\le \|g\|_{G_b^{k,\omega}({\mbf R}^n)}\quad {\rm and}\quad \|g''\|_{G_b^{k,\omega}({\mbf R}^n)}<\frac{\varepsilon}{2(C_N+1)}, \] see Lemma \ref{norm} for the definition of $C_N$. Next, for each $f\in C_b^{k,\omega}({\mbf R}^n)$, $\|f\|_{C_b^{k,\omega}({\mbf R}^n)}=1$, we have by means of Lemma \ref{lem2.2},\smallskip \begin{equation}\label{eq3.20} \begin{array}{l} \ \ \ \vspace*{-2mm}\\ \displaystyle \bigl|f\bigl(H_{NN}\,g-g\bigr)\bigr|=\bigl|\bigl(L_{NN}f-f\bigr)(g)\bigr|\le \bigl|\bigl(L_{NN}f-f\bigr)(g')\bigr|+ \bigl|\bigl(L_{NN}f-f\bigr)(g'')\bigr|\medskip\\ \displaystyle < \bigl|\bigl(E_{N}f_N-f\bigr)(g')\bigr|+\|L_{NN}-{\rm id}\|\cdot\frac{\varepsilon}{2(C_{N}+1)}\le\bigl|\bigl(E_{N}f_N-f\bigr)(g')\bigr|+\frac{\varepsilon}{2}. \end{array} \end{equation} Let $N_0\in{\mbf N}$ be so large that all points $x_{j\alpha}, y_{j\alpha}$ as above belong to $\mathbb K_{N_0}^n$. Since $f_{N_0}=f$ on $\mathbb K_{N_0}^n$, for all $N\ge N_0$, \[ \bigl(E_{N}f_N-f\bigr)(g')=\bigl(E_{N}f_N-f_N\bigr)(g'). \] Hence, due to Lemma \ref{lem2.2}\,(c) for $z=x_{j\alpha}$ or $y_{j\alpha}$, \[ \bigl|\bigl(E_{N}f_N-f\bigr)(\delta_z^\alpha)\bigr|\le \|E_{N}f_N-f_N\|_{C_b^{k}({\mbf R}^n)} \le c_N. \] This implies that for all $N\ge N_0$ \[ \bigl|\bigl(E_{N}f_N-f\bigr)(g')\bigr|\le c_N\cdot\left(\sum_{j,\alpha}|c_{j\alpha}| +\left(\sum_{j,\alpha}2|d_{j\alpha}|\right)\cdot\max_{j,\alpha}\left\{\frac{1}{\omega(\|x_{j\alpha}-y_{j\alpha} \|)}\right\}\right). \] Choose $N_0'\ge N_0$ so large that for all $N\ge N_0'$ the right-hand side of the previous inequality is less than $\frac{\varepsilon}{2}$. Then combining this with \eqref{eq3.20} we get for all $N\ge N_0'$, \[ \|H_{NN}\,g-g\|_{G_b^{k,\omega}({\mbf R}^n)}<\varepsilon. \] This shows that for all $g\in G_b^{k,\omega}({\mbf R}^n)$ \[ \lim_{N\rightarrow\infty}H_{NN}\,g=g \] which completes the proof of the lemma. \end{proof} Let us finish the proof of the theorem for $S={\mbf R}^n$. We set \begin{equation}\label{equ4.28} T_{N}:=\left(1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}\,\frac{1}{\omega(t)}\right)\cdot\frac{H_{NN}}{C_N}, \end{equation} see Lemma \ref{norm} for the definition of $C_N$. Since $\{C_N\}_{N\in{\mbf N}}$ converges to the first factor in the definition of $T_N$, due to Lemma \ref{lem3.5} $\{T_N\}_{N\in{\mbf N}}$ is the sequence of operators of finite rank on $G_b^{k,\omega}({\mbf R}^n)$ of norm at most $\lambda:=1+c_{k,n}\cdot 4\sqrt n\cdot (k+1)\cdot\lim_{t\rightarrow\infty}(1/\omega(t))$ converging pointwise to the identity operator. In particular, this sequence converges uniformly to the identity operator on each compact subset of $G_b^{k,\omega}({\mbf R}^n)$. This shows that $G_b^{k,\omega}({\mbf R}^n)$ has the $\lambda$-approximation property with respect to the approximating sequence of operators $\{T_N\}_{N\in{\mbf N}}$. The proof of Theorem \ref{te1.3} for $S={\mbf R}^n$ is complete. \end{proof} \subsection{Proof of Theorem \ref{te1.3}\,(2)} \begin{proof} In the case of $G_b^{k,\omega}(S)$, the required sequence of finite rank linear operators approximating the identity map is $\bigl\{PT_N|_{G_b^{k,\omega}(S)}\bigr\}_{N\in{\mbf N}}$, where $T_N$ are linear operators defined by \eqref{equ4.28} and $P: G_b^{k,\omega}({\mbf R}^n)\rightarrow G_b^{k,\omega}(S)$ is the projection of Theorem \ref{teo1.6}. We have \[ \|PT_N|_{G_b^{k,\omega}(S)}\|\le \|P\|\cdot\|T_N\|=:\|P\|\cdot\lambda(k,n,\omega). \] Choosing here $P$ corresponding to the extension operators of papers \cite{Gl} ($k=0$), \cite{BS2} ($k=1$) and \cite{Lu} ($k\ge 2$) we obtain the required result. The proof of Theorem \ref{te1.3} is complete. \end{proof} \subsection{Proof of Theorem \ref{teor1.10}} \begin{proof} Due to the result of Pe\l czy\'nski \cite{P} there are a separable Banach space $Y$ with a norm one monotone basis $\{b_j\}_{j\in{\mbf N}}$, an isomorphic embedding $T:X\rightarrow Y$ with distortion $\|T\|\cdot\|T^{-1}\|\le 4\lambda$, and a linear projection $P:Y\rightarrow T(X)$ with $\|P\|\le 4\lambda$. For an operator $H\in\mathcal L(G_b^{k,\omega}({\mbf R}^n);X)$ we define \[ \widetilde H:=T\cdot H\in \mathcal L(G_b^{k,\omega}({\mbf R}^n);Y). \] Then for each $x\in{\mbf R}^n$, \[ \widetilde H(\delta_x^0)=\sum_{j=1}^\infty \tilde h_j(x)\cdot b_j \] for some $\tilde h_j(x)\in{\mbf R}$, $j\in{\mbf N}$. Further, consider the family of bounded linear functionals $\{b_j^*\}_{j\in{\mbf N}}\subset Y^*$ such that $b_j^*(b_i)=\delta_{ij}$ (- the Kronecker delta) for all $i, j\in{\mbf N}$. As the basis $\{b_j\}_{j\in{\mbf N}}$ is monotone, $\|b_j^*\|\le 2$ for all $j\in{\mbf N}$. Since $b_j^*\circ\widetilde H\in \bigl(G_b^{k,\omega}({\mbf R}^n)\bigr)^*=C_b^{k,\omega}({\mbf R}^n)$, the functions $\tilde h_j$, $\tilde h_j(x):=(b_j^*\circ \widetilde H)(\delta_x^0)$, $x\in{\mbf R}^n$, belong to $C_b^{k,\omega}({\mbf R}^n)$ and \begin{equation}\label{equ4.32a} \|\tilde h_j\|_{C_b^{k,\omega}({\mbf R}^n)}\le 2\cdot\|T\|\cdot\|H\|\quad {\rm for\ all}\quad j\in{\mbf N}. \end{equation} In particular, $(b_j^*\circ\widetilde H)(\delta_x^\alpha)=D^\alpha (b_j^*\circ\widetilde H)(\delta_x^0)=D^\alpha \tilde h_j(x)$ for all $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, $x\in{\mbf R}^n$, $j\in{\mbf N}$. This implies that for all such $\alpha$ and $x$, \begin{equation}\label{equ4.32} \widetilde H(\delta_x^\alpha)=\sum_{j=1}^\infty D^\alpha \tilde h_j(x)\cdot b_j. \end{equation} Next, since the range of $\widetilde H$ is the subset of $T(X)$, \[ H=T^{-1}\cdot P\cdot \widetilde H. \] From here and \eqref{equ4.32} we obtain, for all $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, $x\in{\mbf R}^n$, \begin{equation}\label{equ4.33} H(\delta_x^\alpha)=\sum_{j=1}^\infty D^\alpha \tilde h_j(x)\cdot (T^{-1}\cdot P)(b_j) \end{equation} (convergence in $X$). Finally, we set \begin{equation}\label{equ4.34} h_j:=\|(T^{-1}\cdot P)(b_j)\|\cdot\tilde h\quad {\rm and}\quad v_j:=\frac{(T^{-1}\cdot P)(b_j)}{\|(T^{-1}\cdot P)(b_j)\|},\quad j\in{\mbf N}. \end{equation} Then all $v_j\in X$ are of norm one. In turn, all $h_j\in C_b^{k,\omega}({\mbf R}^n)$ and due to \eqref{equ4.32a}, \eqref{equ4.34} and the properties of $T$ and $P$ for all $j\in{\mbf N}$, \[ \|h_j\|_{C_b^{k,\omega}({\mbf R}^n)}\le \|T^{-1}\|\cdot\|P\|\cdot\|\tilde h_j\|_{C_b^{k,\omega}({\mbf R}^n)}\le 2\cdot \|T\|\cdot\|T^{-1}\|\cdot\|P\|\cdot\|H\|\le 32\cdot\lambda^2\cdot\|H\|. \] Moreover, by \eqref{equ4.33}, for all $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, $x\in{\mbf R}^n$, \[ H(\delta_x^\alpha)=\sum_{j=1}^\infty D^\alpha h_j(x)\cdot v_j, \] as required. The proof of Theorem \ref{teor1.10} is complete. \end{proof} \section{Proofs of Theorem \ref{te1.4} and Corollary \ref{cor1.10}} \subsection{Proof of Theorem \ref{te1.4}\,(1)} \begin{proof} Let $\Lambda_{n,k}:=\{\alpha\in {\mbf Z}_+^n\, :\, |\alpha|\le k\}$. We set \begin{equation}\label{eq4.23} M_{n,k}:=\bigl(\Lambda_{n,k}\times{\mbf R}^n\bigr)\sqcup \bigl(\bigl(\Lambda_{n,k}\setminus\Lambda_{n,k-1}\bigr)\times\bigl(({\mbf R}^n\times{\mbf R}^n)\setminus\Delta_n\bigr)\bigr), \end{equation} where $\Delta_n:=\{(x,y)\in{\mbf R}^n\times{\mbf R}^n\, :\, x=y\}$. Space $M_{n,k}$ has the natural structure of a $C^\infty$ manifold, in particular, it is a locally compact Hausdorff space. By $C_b(M_{n,k})$ we denote the Banach space of bounded continuous functions on $M_{n,k}$ equipped with supremum norm. Let us define a linear map $\mathcal I: C_b^{k,\omega}({\mbf R}^n)\rightarrow C_b(M_{n,k})$ by the formula\medskip \begin{equation} \mathcal I(f)(m)=\left\{ \begin{array}{lll} \displaystyle D^\alpha f(x)&{\rm if}&m=(\alpha,x)\in\Lambda_{n,k}\times {\mbf R}^n\\ \\ \displaystyle \frac{D^\alpha f(x)-D^\alpha f(y)}{\omega(\|x-y\|)}&{\rm if}&m=(\alpha, (x,y))\in \bigl(\Lambda_{n,k}\setminus\Lambda_{n,k-1}\bigr)\times\bigl({\mbf R}^n\times{\mbf R}^n\setminus\Delta_n\bigr), \end{array} \right. \end{equation} \begin{Proposition}\label{prop5.1} $\mathcal I$ is a linear isometric embedding. \end{Proposition} \begin{proof} The statement follows straightforwardly from the definitions of the involved spaces. \end{proof} Since $\mathcal I\bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)$ is a closed subspace of $C_b(M_{n,k})$, the Hahn-Banach theorem implies that the adjoint map \begin{equation}\label{eq4.25} \mathcal I^*:\bigl(C_b(M_{n,k})\bigr)^*\rightarrow \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^* \end{equation} of $\mathcal I$ is surjective of norm one. Similarly, $\mathcal I$ maps $C_0^{k,\omega}({\mbf R}^n)$ isometrically into the Banach subspace $C_0(M_{n,k})\subset C_b(M_{n,k})$ of continuous functions on $M_{n,k}$ vanishing at infinity. Thus the adjoint of $\mathcal I_0:=\mathcal I |_{C_0^{k,\omega}({\mbf R}^n)}$ is the surjective map of norm one \begin{equation}\label{eq4.26} \mathcal I_0^*:\bigl(C_0(M_{n,k})\bigr)^*\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*. \end{equation} According to the Riesz representation theorem (see, e.g.,\cite{DS}), $\bigl(C_0(M_{n,k})\bigr)^*$ is isometrically isomorphic to the space of countably additive regular Borel measures on $M_{n,k}$ with the norm being the total variation of measure. In what follows we identify these two spaces. In the proof we use the following result. \begin{Proposition}\label{prop4.2} If $\omega$ satisfies condition \eqref{omega2}, then $C_0^{k,\omega}({\mbf R}^n)$ is weak$^*$ dense in $C_b^{k,\omega}({\mbf R}^n)$. \end{Proposition} \begin{proof} Let $\{L_{NN}\}_{N\in{\mbf N}}$ be finite rank bounded linear operators on $C_b^{k,\omega}({\mbf R}^n)$ defined by \eqref{eq3.18}. According to Lemma \ref{lem3.5} for each $f\in C_b^{k,\omega}({\mbf R}^n)$ the sequence $\{L_{NN}f\}_{N\in{\mbf N}}$ weak$^*$ converges to $f$. Moreover, each $L_{NN}f\in C^\infty({\mbf R}^n)$, cf. Lemma \ref{lem2.2}\,(a). We set \begin{equation}\label{e4.27} \hat f_N:=\rho_N\cdot L_{NN}f, \end{equation} see section~4.3. Then $\hat f_N$ is a $C^\infty$ function with compact support on ${\mbf R}^n$ satisfying, due to Lemmas \ref{norm} and \ref{lem2.2}\,(b), the inequality \begin{equation}\label{e4.28} \|\hat f_N\|_{C_b^{k,\omega}({\mbf R}^n)}\le C_N^2\|f\|_{C_b^{k,\omega}({\mbf R}^n)}, \end{equation} where $\lim_{N\rightarrow\infty}C_N=1+c_{k,n}\cdot 4\sqrt n \cdot (k+1)\cdot\lim_{t\rightarrow\infty}\frac{1}{\omega(t)}$. Clearly, sequence $\{D^\alpha \hat f_N\}_{N\in{\mbf N}}$ converges pointwise to $D^\alpha f$ for all $\alpha\in {\mbf Z}_+^n$, $|\alpha|\le k$. Also, due to condition \eqref{omega2} all $\hat f_N\in C_0^{k,\omega}({\mbf R}^n)$. Hence, according to Proposition \ref{prop3.1}, sequence $\{\hat f_N\}_{N\in{\mbf N}}$ weak$^*$ converges to $f$. This shows that $C_0^{k,\omega}({\mbf R}^n)$ is weak$^*$ dense in $C_b^{k,\omega}({\mbf R}^n)$. \end{proof} Next, let $i^*: \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ be the linear surjective map of norm one adjoint to the isometrical embedding $i: C_0^{k,\omega}({\mbf R}^n)\hookrightarrow C_b^{k,\omega}({\mbf R}^n)$. \begin{C}\label{cor4.3} Restriction of $i^*$ to $G_b^{k,\omega}({\mbf R}^n)$ is injective. \end{C} \begin{proof} Proposition \ref{prop4.2} implies that functions in $C_0^{k,\omega}({\mbf R}^n)$ regarded as linear functionals on $G_b^{k,\omega}({\mbf R}^n)$ separate the points of $G_b^{k,\omega}({\mbf R}^n)\, (\subset \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*)$. If $i^*(v)=0$ for some $v\in G_b^{k,\omega}({\mbf R}^n)$, then \[ 0=(i^*(v))(f)=f(v)\quad {\rm for\ all}\quad f\in C_0^{k,\omega}({\mbf R}^n). \] Hence, $v=0$. \end{proof} We set \[ \tilde\delta_x^\alpha:=i^*(\delta_x^\alpha),\quad |\alpha|\le k,\ x\in{\mbf R}^n. \] By definition, maps $\phi_\alpha: {\mbf R}^n\rightarrow \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*$, $x\mapsto\delta_x^\alpha$, $|\alpha|\le k$, are continuous and bounded and so are the maps $i^*\circ\phi_\alpha:{\mbf R}^n\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$. \begin{Proposition}\label{prop4.4} The range of $\mathcal I_0^*$ coincides with $i^*(G_b^{k,\omega}({\mbf R}^n))$. \end{Proposition} \begin{proof} Let $\mu\in \bigl(C_0(M_{n,k})\bigr)^*$ be a countably additive regular Borel measures on $M_{n,k}$. We set, for all admissible $\alpha$ and all Borel measurable sets $U\subset M_{n,k}$, \[ \mu_\alpha^1(U)=\mu\bigl(U\cap (\{\alpha\}\times{\mbf R}^n)\bigr)\quad {\rm and}\quad \mu_\alpha^2(U)=\mu\bigl(U\cap \bigl(\{\alpha\}\times \bigl(({\mbf R}^n\times{\mbf R}^n)\setminus\Delta_n\bigr)\bigr)\bigr). \] Then $\mu=\sum_{\alpha,j}\mu_\alpha^j$. Let us show that each $\mathcal I_0^*(\mu_\alpha^j)$ belongs to $i^*(G_b^{k,\omega}({\mbf R}^n))$. Indeed, for $j=1$ consider the Bochner integral \begin{equation}\label{eq4.27} J(\mu_\alpha^1):=\int_{x\in{\mbf R}^n}i^*(\phi_\alpha(x))\, d\mu_\alpha^1(x)=i^*\left(\int_{x\in{\mbf R}^n}\phi_\alpha(x)\, d\mu_\alpha^1(x)\right). \end{equation} Since $\phi_\alpha$ is continuous and bounded, the above integral is well-defined and its value is an element of $i^*(G_b^{k,\omega}({\mbf R}^n))$. By the definition of the Bochner integral, for each $f\in C_0^{k,\omega}({\mbf R}^n)$, \[ (J(\mu_\alpha^1))(f)=\int_{x\in{\mbf R}^n}\bigl(i^*(\phi_\alpha(x))\bigr)(f)\,d\mu_\alpha^1(x)=\int_{x\in{\mbf R}^n} D^\alpha f(x)\,d\mu_\alpha^1(x)=:\bigl(\mathcal I_0^*(\mu_\alpha^1)\bigr)(f). \] Hence, \[ \mathcal I_0^*(\mu_\alpha^1)=J(\mu_\alpha^1)\in i^*(G_b^{k,\omega}({\mbf R}^n)). \] Similarly, for $\alpha\in\Lambda_{n,k}\setminus\Lambda_{n,k-1}$ we define \begin{equation}\label{eq4.28} J(\mu_\alpha^2):=\int_{z=(x,y)\in ({\mbf R}^n\times{\mbf R}^n)\setminus\Delta_n}\frac{i^*(\phi_\alpha(x))-i^*(\phi_\alpha(y))}{\omega(\|x-y\|)} d\mu_\alpha^2(z). \end{equation} Then, as before, we obtain that \[ \mathcal I_0^*(\mu_\alpha^2)=J(\mu_\alpha^2)\in i^*(G_b^{k,\omega}({\mbf R}^n)). \] Thus we have established that the range of $\mathcal I_0^*$ is a subset of $i^*(G_b^{k,\omega}({\mbf R}^n))$. Since the map $\mathcal I_0^*$ is surjective and its range contains all $\tilde\delta_x^\alpha$, it must contain $i^*(G_b^{k,\omega}({\mbf R}^n))$ as well. This completes the proof of the proposition. \end{proof} In particular, we obtain that $\bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*=i^*(G_b^{k,\omega}({\mbf R}^n))$, i.e., by the inverse mapping theorem $i^*$ restricted to $G_b^{k,\omega}$ maps it isomorphically onto $\bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$. Let us show that if $\lim_{t\rightarrow\infty}\omega(t)=\infty$, then $i^*$ is an isometry. Assume, on the contrary, that for some $v\in G_b^{k,\omega}({\mbf R}^n)$, \begin{equation}\label{eq4.31} \|i^*(v)\|_{(C_0^{k,\omega}({\mbf R}^n))^*}<\|v\|_{G_b^{k,\omega}({\mbf R}^n)}. \end{equation} Let $f\in C_0^{k,\omega}({\mbf R}^n)$, $\|f\|_{C_b^{k,\omega}({\mbf R}^n)}=1$, be such that \[ v(f)=\|v\|_{G_b^{k,\omega}({\mbf R}^n)}. \] Let $\{f_N\}_{N\in{\mbf N}}\subset C_0^{k,\omega}({\mbf R}^n)$, $\|f_N\|_{C_b^{k,\omega}({\mbf R}^n)}\le C_N^2$, $N\in{\mbf N}$, be the sequence of Proposition \ref{prop4.2} weak$^*$ converging to $f$. Observe that $\lim_{N\rightarrow\infty} C_N=1$ due to the above condition for $\omega$. Then from \eqref{eq4.31} and \eqref{e4.28} we obtain \[ \begin{array}{l} \displaystyle \|v\|_{G_b^{k,\omega}({\mbf R}^n)}=v(f)=\lim_{N\rightarrow\infty} v(f_N)=\lim_{N\rightarrow\infty} \bigl(i^*(v)\bigr)(f_N)\medskip\\ \displaystyle \le \|i^*(v)\|_{(C_0^{k,\omega}({\mbf R}^n))^*}\cdot\varlimsup_{N\rightarrow\infty} \|f_N\|_{C_b^{k,\omega}({\mbf R}^n)}\le \|i^*(v)\|_{(C_0^{k,\omega}({\mbf R}^n))^*}<\|v\|_{G_b^{k,\omega}({\mbf R}^n)}, \end{array} \] a contradiction proving that $i^*$ is an isometry. The proof of Theorem \ref{te1.4}\,(1) is complete. \end{proof} \subsection{Proof of Theorem \ref{te1.4}\,(2)} \begin{proof} By the hypotheses of the theorem there exists a weak$^*$ continuous operator $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ such that $T(C_0^{k,\omega}(S))\subset C_0^{k,\omega}({\mbf R}^n)$. This implies that there is a bounded linear projection of the geometric preduals of the corresponding spaces $P: G_b^{k,\omega}({\mbf R}^n)\rightarrow G_b^{k,\omega}(S)$ such that $P^*=T$. Let $q_S: C_b^{k,\omega}({\mbf R}^n)\rightarrow C_b^{k,\omega}(S)$ and $q_{S0}: C_0^{k,\omega}({\mbf R}^n)\rightarrow C_0^{k,\omega}(S)$ denote the quotient maps induced by restrictions of functions on ${\mbf R}^n$ to $S$. Finally, let $i: C_0^{k,\omega}({\mbf R}^n)\rightarrow C_b^{k,\omega}({\mbf R}^n)$ and $i_S: C_0^{k,\omega}(S)\rightarrow C_b^{k,\omega}(S)$ be the bounded linear maps corresponding to inclusions of the spaces. Note that $i$ is an isometric embedding and $i_S$ is injective of norm $\le 1$. \begin{Lm}\label{lem5.5} $T_0:=T|_{C_0^{k,\omega}(S)}:C_0^{k,\omega}(S)\rightarrow C_0^{k,\omega}({\mbf R}^n)$ is a bounded linear map between Banach spaces. \end{Lm} \begin{proof} For $f\in C_0^{k,\omega}(S)$ we have \[ \|T_0f\|_{C_0^{k,\omega}({\mbf R}^n)}=\|(T\circ i_S)(f)\|_{C_b^{k,\omega}({\mbf R}^n)}\le\|T\|\cdot\|i_S(f)\|_{C_b^{k,\omega}(S)}\le \|T\|\cdot\|f\|_{C_0^{k,\omega}(S)}, \] as required. \end{proof} Now, we have the following two commutative diagrams of adjoints of the above bounded linear maps (one corresponding to upward arrows and another one to downward arrows): \begin{equation}\label{equ5.38} \begin{array}{cccc} \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*&\stackrel{i^*}{\longrightarrow}&\bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*\smallskip\\ _{\mbox{{\tiny $q_S^*$}}}\!\uparrow\ \ \ \downarrow\mbox{{\tiny $T^*$}}&&_{\mbox{{\tiny $q_{S0}^*$}}}\!\uparrow\ \ \ \downarrow\mbox{{\tiny $T_0^*$}} \\ \bigl(C_b^{k,\omega}(S)\bigr)^*&\stackrel{i_S^*}{\longrightarrow}&\bigl(C_0^{k,\omega}(S)\bigr)^*. \end{array} \end{equation} Here $T^*\circ q_S^*=(q_S\circ T)^*={\rm id}$ and $T_0^*\circ q_{S0}^*=(q_{S0}\circ T_0)^*={\rm id}$, maps $q_S^*$ and $q_{S0}^*$ are isometric embeddings and map $i^*$ is surjective. Note that $i^*|_{G_b^{k,\omega}({\mbf R}^n)}:G_b^{k,\omega}({\mbf R}^n)\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ is an isomorphism by the first part of the theorem. Also, by the definition of $P$ (see \eqref{proj} in section~3.3 above), \begin{equation}\label{equ5.39} q_S^*\circ (T^*|_{G_b^{k,\omega}({\mbf R}^n)})=P. \end{equation} Let us show that the map \[ I:=i_S^*\circ(T^*|_{G_b^{k,\omega}(S)}):G_b^{k,\omega}(S)\rightarrow \bigl(C_0^{k,\omega}(S)\bigr)^* \] is an isomorphism.\smallskip (a) Injectivity of $I$: If $I(v)=0$ for some $v\in G_b^{k,\omega}(S)$, then by the commutativity of \eqref{equ5.38} and by \eqref{equ5.39}, \[ 0=(q_{S0}^*\circ i_S^*)(T^*v)=(i^*\circ q_S^*)(T^*v)=i^*(Pv)=i^*(v). \] Since $i^*$ is injective, the latter implies that $v=0$, i.e., $I$ is an injection.\medskip (b) Surjectivity of $I$: Let $v\in \bigl(C_0^{k,\omega}(S)\bigr)^*$. Since $T_0^*$ is surjective, there exists $v_1\in \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $T_0^*(v_1)=v$. Further, since $i^*|_{G_b^{k,\omega}({\mbf R}^n)}:G_b^{k,\omega}({\mbf R}^n)\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ is an isomorphism, there exists $v_2\in G_b^{k,\omega}({\mbf R}^n)$ such that $i^*(v_2)=v_1$. Now, by the commutativity of \eqref{equ5.38}, \[ v=(T_0^*\circ i^*)(v_2)=(i_S^*\circ T^*)(v_2)=(i_S^*\circ (T^*\circ q_S^*)\circ T^*)(v_2)=(i_S^*\circ T^*)(Pv_2)=I(Pv_2), \] i.e., $I$ is a surjection. So $I$ is a bijection and therefore by the inverse mapping theorem it is an isomorphism. This completes the proof of the second part of Theorem \ref{te1.4}. \end{proof} \subsection{Proof of Corollary \ref{cor1.10}} \begin{proof} Let $X\subset C_0^{k,\omega}({\mbf R}^n)$ be the closure of the space of $C^\infty$ functions with compact supports on ${\mbf R}^n$. Assume, on the contrary, that there exists $f\in C_0^{k,\omega}({\mbf R}^n)\setminus X$. Then there exists a functional $\lambda\in \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ such that $\lambda|_X=0$ and $\lambda(f)=1$. Let $i^*: \bigl(C_b^{k,\omega}({\mbf R}^n)\bigr)^*\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ be the adjoint of the isometrical embedding $i:C_0^{k,\omega}({\mbf R}^n)\hookrightarrow C_b^{k,\omega}({\mbf R}^n)$. According to the arguments of the proof of Theorem \ref{te1.4}, $i^*|_{G_b^{k,\omega}({\mbf R}^n)}:G_b^{k,\omega}({\mbf R}^n)\rightarrow \bigl(C_0^{k,\omega}({\mbf R}^n)\bigr)^*$ is an isomorphism. Hence, for $\tilde\lambda:=(i^*|_{G_b^{k,\omega}({\mbf R}^n)})^{-1}(\lambda)$ we have $g(\tilde\lambda)=0$ for all $g\in X$ and $f(\tilde\lambda)=1$. Observe that $X$ is weak$^*$ dense in $C_0^{k,\omega}({\mbf R}^n)$ (see the proof of Proposition \ref{prop4.2}). Thus $X$ separates the points of $G_b^{k,\omega}({\mbf R}^n)$. Since $g(\tilde\lambda)=0$ for all $g\in X$, the latter implies that $\tilde\lambda=0$, a contradiction with $f(\tilde\lambda)=1$. This shows that $X=C_0^{k,\omega}({\mbf R}^n)$. Clearly, $X$ is separable (it contains, e.g., the dense countable set of functions of the form $\rho_N\cdot p$, $N\in{\mbf N}$, where $p$ are polynomials with rational coefficients and $\{\rho_N\}_{N\in{\mbf N}}$ is a fixed sequence of $C^\infty$ cut-off functions weak$^*$ converging in $C_b^{k,\omega}({\mbf R}^n)$ to the constant function $f= 1$). This completes the proof of the corollary. \end{proof} \section{Proof of Theorem \ref{te1.11}} \subsection{Proof of Theorem \ref{te1.11} for Weak $k$-Markov Sets} First, we recall some results proved in \cite{BB1,BB2, B}. (1) If $S\in {\rm Mar}_k^*({\mbf R}^n)$, then a function $f\in C_b^{k,\omega}(S)$ has derivatives of order $\le k$ at each weak $k$-Markov point $x\in S$, i.e., there exists a (unique) polynomial $T_x^k(f)\in\cP_{k,n}$ such that \[ \lim_{y\to x}\frac{|f(y)-T_x^k(f)(y)|}{\|y-x\|^k}=0. \] If $T_x^k(f)(z):=\sum_{|\alpha|\le k} \frac{c_\alpha}{\alpha !} (z-x)^\alpha$, $\alpha\in{\mbf Z}_+^n$, then $c_\alpha$ is called the partial derivative of order $|\alpha|$ at $x$ and is denoted as $D_S^\alpha f (x)$.\medskip (2) If $\tilde f\in C_b^{k,\omega}({\mbf R}^n)$ is such that $\tilde f|_S=f$, then the Taylor polynomial $T_x^k(\tilde f)$ of order $k$ of $\tilde f$ at $x$ coincides with $T_x^k(f)$.\medskip (3) The analog of the the classical Whitney-Glaeser theorem holds: A function $f\in C(S)$ belongs to $C_b^{k,\omega}(S)$ if and only if it has derivatives of order $\le k$ at each weak $k$-Markov point $x\in S$ and there exists a constant $\lambda>0$ such that for all weak $k$-Markov points $x,y\in S$, $z\in\{x,y\}$ \begin{equation}\label{equ6.40} \begin{array}{l} \displaystyle \max_{|\alpha|\le k}|D^\alpha_S f(x)|\le \lambda\quad\text{and}\\ \\ \displaystyle \max_{|\alpha|\le k}\frac{|D_S^\alpha \bigl(T_x^k(f)-T_y^k(f)\bigr)(z)|}{\|x-y\|^{k-|\alpha|}}\le\lambda\cdot\omega(\|x-y\|). \end{array} \end{equation} Moreover, \[ \|f\|_{C_b^{k,\omega}(S)}\approx\inf\lambda \] with constants of equivalence depending only on $k$ and $n$.\medskip (4) There exists a bounded linear extension operator $T: C_b^{k,\omega}(S)\to C_b^{k,\omega}({\mbf R}^n)$ of finite depths \[ (Tf)(x):= \left\{ \begin{array}{ccc} \displaystyle \sum_{i=1}^\infty \lambda_i(x)f(x_i)&{\rm if}&x\in{\mbf R}^n\setminus S \\ f(x)&{\rm if}& x\in S, \end{array} \right. \] where all $\lambda_i\in C^\infty({\mbf R}^n)$ and have compact supports in ${\mbf R}^n\setminus S$, all $x_i\in S$ and for each $x\in{\mbf R}^n$ the number of nonzero terms in the above sum is at most ${n+k\choose n}\cdot w$, where $w$ is the order of the Whitney cover of ${\mbf R}^n\setminus S$. The construction of $T$ repeats that of the Whitney-Glaeser extension operator \cite{Gl}, where instead of jets $T_x^k(f)$ of $f\in C_b^{k,\omega}(S)$ at weak $k$-Markov points $x\in S$ (forming a dense subset of $S$) one uses polynomials of degree $k$ interpolating $f$ on certain subsets of cardinality ${n+k\choose n}$ close to $x$. In particular, as in the case of the Whitney-Glaeser extension operator, we obtain that $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ for all moduli of continuity $\omega$. Also, by the construction, if $f\in C_b^{k,\omega}(S)$ is the restriction to $S$ of a $C^\infty$ function with compact support on ${\mbf R}^n$, then $Tf\in C_b^{k,\omega}({\mbf R}^n)$ has compact supports in all closed $\delta$-neihbourhoods of $S$ (i.e., sets $[S]_\delta:=\{x\in{\mbf R}^n\, :\, \inf_{y\in S}\|x-y\|\le\delta\}$, $\delta>0$). \begin{proof}[Proof of Theorem \ref{te1.11} for $S\in {\rm Mar}_k^*({\mbf R}^n)$] Let $\rho\in C^\infty({\mbf R}^n)$, $0\le \rho\le 1$, be such that $\rho|_{[S]_{1}}=1$, $\rho|_{{\mbf R}^n\setminus [S]_{3}}=0$ and for some $C_{k,n}\in{\mbf R}_+$ (depending on $k$ and $n$ only) \begin{equation}\label{rho1} \sup_{x\in\mathbb R^n}|D^\alpha\rho(x)|\le C_{k,n}\quad {\rm for\ all}\quad \alpha\in\mathbb Z_+^n. \end{equation} (E.g., such $\rho$ can be obtained by the convolution of the indicator function of $[S]_2$ with a fixed radial $C^\infty$ function with support in the unit Euclidean ball of ${\mbf R}^n$ and with $L_1({\mbf R}^n)$ norm one.) We define a new extension operator by the formula \begin{equation}\label{eq6.42} \widetilde Tf=\rho\cdot Tf,\qquad f\in C_b^{k,\omega}(S). \end{equation} \begin{Lm}\label{lem6.1} Operator $\widetilde T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ for all moduli of continuity $\omega$. \end{Lm} \begin{proof} We equip $C_b^{k,\omega}({\mbf R}^n)$ with equivalent norm \[ \|f\|_{C_b^{k,\omega}({\mbf R}^n)}':=\max\left\{\|f\|_{C_b^{k,\omega}({\mbf R}^n)},|f|_{C_b^{k,\omega}({\mbf R}^n)}'\right\}, \] where $|f|_{C_b^{k,\omega}({\mbf R}^n)}'$ is defined similarly to $|f|_{C_b^{k,\omega}({\mbf R}^n)}$ but with the supremum taken over all $x\ne y$ such that $\|x-y\|\le 1$, see \eqref{eq3}--\eqref{eq5}. (Note that the constants of equivalence between these two norms depend on $\omega$.) Now, using word-by-word the arguments of Lemma \ref{norm} with $\rho_\ell$ replaced by $\rho$, $\ell$ replaced by $1$, and $c_{k,n}$ replaced by $C_{k,n}$ we obtain for some constant $C=C(k,n,\omega)$ and all $h\in C_b^{k,\omega}({\mbf R}^n)$, \begin{equation}\label{eq6.43} \|\rho\cdot h\|_{C_b^{k,\omega}({\mbf R}^n)}'\le C\cdot\|h\|_{C_b^{k,\omega}({\mbf R}^n)}'. \end{equation} Since $T\in Ext(C_b^{k,\omega}(S);C_b^{k,\omega}({\mbf R}^n))$ for all moduli of continuity $\omega$, inequality \eqref{eq6.43} implies the required statement. \end{proof} Clearly, $\widetilde T$ is of finite depth. Moreover, if $f\in C_b^{k,\omega}(S)$ is the restriction to $S$ of a $C^\infty$ function with compact support on ${\mbf R}^n$, then $\widetilde Tf\in C_b^{k,\omega}({\mbf R}^n)$ and has compact support on ${\mbf R}^n$ due to the properties of operator $T$, (see part (4) above). Finally, since the set of all $C_b^{k,1}(S)$ functions (i.e., for this space $\omega(t):=t$, $t\in{\mbf R}_+$) with compact supports on $S$ is dense in $C_0^{k,\omega}(S)$ (because $\omega$ satisfies condition \eqref{omega2}, see Corollary \ref{cor1.10}), the preceding property of $\widetilde T$ and Lemma \ref{lem6.1} imply that $\widetilde T(C_0^{k,\omega}(S))\subset C_0^{k,\omega}({\mbf R}^n)$. Therefore $\widetilde T$ satisfies the hypotheses of Theorem \ref{te1.4}\,(2) (weak$^*$ continuity of $\widetilde T$ follows from Theorem \ref{teo1.6}). This implies the required statement: $\bigl(C^{k,\omega}_0(S)\bigr)^*$ is isomorphic to $G_b^{k,\omega}(S)$ for all $\omega$ satisfying \eqref{omega2} and all weak $k$-Markov sets $S$. Now, $G_b^{k,\omega}(S)$ has the metric approximation property due to the Grothendieck result \cite[Ch.\,I]{G} (formulated before Remark \ref{k} of section~1.4 above) because this space has the approximation property by Theorem \ref{te1.3}. Also, $C_0^{k,\omega}(S)$ has the metric approximation property because its dual has it, see, e.g., \cite[Th.\,3.10]{C}. The proof of the theorem for $S\in {\rm Mar}_k^*({\mbf R}^n)$ is complete. \end{proof} \subsection{Proof of Theorem \ref{te1.11} in the General Case} \begin{proof} We require some auxiliary results. Let $\widetilde\omega$ be the modulus of continuity satisfying \begin{equation}\label{eq6.46} \varlimsup_{t\rightarrow 0^+}\frac{\omega_o(t)}{\widetilde\omega(t)}<\infty. \end{equation} \begin{Lm}\label{lem6.2} The restriction of the pullback map $H^*:C_b^k({\mbf R}^n)\rightarrow C_b^k({\mbf R}^n)$, $H^*f:=f\circ H$, to $C_b^{k,\tilde\omega}({\mbf R}^n)$ belongs to $\mathcal L(C_b^{k,\tilde\omega}({\mbf R}^n);C_b^{k,\tilde\omega}({\mbf R}^n))$. \end{Lm} \begin{proof} We set $H=(h_1,\dots, h_n)$. Then by the hypothesis (a) of the theorem all $D^i h_j\in C_b^{k-1,\omega_o}({\mbf R}^n)$, where $\omega_o$ satisfies \eqref{equ1.8}. Let $f\in C_b^{k,\tilde\omega}({\mbf R}^n)$. Then for each $\alpha\in{\mbf Z}_+^n$, $|\alpha|\le k$, by the Fa\`{a} di Bruno formula, see, e.g., \cite{CS}, we obtain \begin{equation}\label{eq6.47} (D^\alpha (f\circ H))(x)=\sum_{0<|\lambda|\le |\alpha|} D^\lambda f(H(x))\cdot P_\lambda\left(\bigl[D^\beta H(x)\bigr]_{0<|\beta|\le |\alpha|}\right); \end{equation} here $P_\lambda\left(\bigl[D^\beta H(x)\bigr]_{0<|\beta|\le |\alpha|}\right)$ are polynomials of degrees $\le |\alpha|$ without constant terms with coefficients in $\mathbb Z_+$ bounded by a constant depending on $k$ and $n$ only in variables $D^\beta h_j$, $0<|\beta|\le |\alpha|$, $1\le j\le n$. Since clearly $C_b^{0,\tilde\omega}({\mbf R}^n)$ is a Banach algebra with respect to the pointwise multiplication of functions, to prove the lemma it suffices to check that all $D^\lambda f(H(\cdot))$ and $D^\beta h_j$ belong to $C_b^{0,\tilde\omega}({\mbf R}^n)$. For $|\beta|=k$ this is true because $D^\beta h_j\in C_b^{0,\omega_o}({\mbf R}^n)\subset C_b^{0,\widetilde\omega}({\mbf R}^n)$ by the definition of $H$ and by condition \eqref{eq6.46}, while for $1\le |\beta|\le k-1$ because $D^\beta h_j\in C_b^{0,1}({\mbf R}^n)$ which is continuously embedded into $C_b^{0,\widetilde\omega}({\mbf R}^n)$. Similarly, for $D^\lambda f(H(\cdot))$ with $1\le |\alpha|\le k-1$ this is true because of the continuous embedding $C_b^{0,1}({\mbf R}^n)\hookrightarrow C_b^{0,\widetilde\omega}({\mbf R}^n)$ and because $H$ is Lipschitz, while for $|\lambda|=k$ by the definition of $f$ and the fact that $H$ is Lipschitz. \end{proof} Using this lemma we prove the following result. \begin{Lm}\label{lem6.3} The operator $(H|_{S'})^*: C_b^{k,\widetilde\omega}(S)\rightarrow C_b^{k,\widetilde\omega}(S')$, $(H|_{S'})^*f:=f\circ H|_{S'}$, is well-defined and belongs to $\mathcal L(C_b^{k,\widetilde\omega}(S); C_b^{k,\widetilde\omega}(S'))$. Moreover, it is weak$^*$ continuous.\footnote{Here the weak$^*$ topologies are defined by means of functionals in $G_b^{k,\widetilde\omega}(\tilde S)$, where $\tilde S$ stands for $S'$ or $S$.} \end{Lm} \begin{proof} Let $\tilde f\in C_b^{k,\widetilde\omega}({\mbf R}^n)$ be such that $\tilde f|_{S}=f$ and $\|\tilde f\|_{C_b^{k,\widetilde\omega}({\mbf R}^n)}=\|f\|_{C_b^{k,\widetilde\omega}(S)}$. Then by Lemma \ref{lem6.2} we have \[ \begin{array}{l} \displaystyle (H|_{S'})^*f=f\circ H_{S'}=(\tilde f\circ H)|_{S'}=(H^*\tilde f)|_{S'}\in C_b^{k,\widetilde\omega}(S')\quad {\rm and}\\ \\ \displaystyle \|(H|_{S'})^*f\|_{C_b^{k,\widetilde\omega}(S')}\le \|H^*\|\cdot\|\tilde f\|_{C_b^{k,\widetilde\omega}({\mbf R}^n)}=\|H^*\|\cdot\|f\|_{C_b^{k,\widetilde\omega}(S)}. \end{array} \] This shows that the operator $(H|_{S'})^*: C_b^{k,\widetilde\omega}(S)\rightarrow C_b^{k,\widetilde\omega}(S')$ is well-defined and belongs to $\mathcal L(C_b^{k,\widetilde\omega}(S); C_b^{k,\widetilde\omega}(S'))$. Further, the fact that the operator $H^*:C_b^{k,\widetilde\omega}({\mbf R}^n)\rightarrow C_b^{k,\widetilde\omega}({\mbf R}^n)$ is weak$^*$ continuous follows straightforwardly from Proposition \ref{prop3.1}, Lemma \ref{lem6.2} and the the Fa\`{a} di Bruno formula \eqref{eq6.47}. Let $T\in Ext(C_b^{k,\widetilde\omega}(S);C_b^{k,\widetilde\omega}({\mbf R}^n))$ be the extension operator of finite depth (see section~1.3) and $q_{S'}: C_b^{k,\widetilde\omega}({\mbf R}^n)\rightarrow C_b^{k,\widetilde\omega}(S')$ be the quotient map induced by restrictions of functions on ${\mbf R}^n$ to $S'$. Then clearly, for all $f\in C_b^{k,\widetilde\omega}(S)$, \[ (H|_{S'})^*f=f\circ H|_{S'}=((Tf)\circ H)|_{S'}=(q_{S'}\circ H^*\circ T)f. \] Therefore $(H|_{S'})^*=q_{S'}\circ H^*\circ T$. Here the operator $T$ is weak$^*$ continuous by Theorem \ref{teo1.6} and the operator $q_{S'}$ is weak$^*$ continuous because it is adjoint of the isometric embedding $G_b^{k,\widetilde\omega}(S')\hookrightarrow G_b^{k,\widetilde\omega}({\mbf R}^n)$. This implies that the operator $(H|_{S'})^*$ is weak$^*$ continuous as well. \end{proof} We are ready to prove Theorem \ref{te1.11}. Let $\widetilde T\in Ext(C_b^{k,\omega}(S');C_b^{k,\omega}({\mbf R}^n))$ be the extension operator of the first part of Theorem \ref{te1.11}, see \eqref{eq6.42}. We set (for $\widetilde \omega:=\omega$) \begin{equation}\label{eq6.48} E:=\widetilde T\circ (H|_{S'})^*. \end{equation} \begin{Lm}\label{lem6.4} Operator $E\in Ext(C_b^{k,\omega}(S); C_b^{k,\omega}({\mbf R}^n))$, is weak$^*$ continuous and maps $C_0^{k,\omega}(S)$ in $C_0^{k,\omega}(S')$. \end{Lm} \begin{proof} The first two statements follow from the hypotheses of the theorem, Lemma \ref{lem6.3} and the fact that $\widetilde T$ is weak$^*$ continuous. So let us check the last statement. Let $f\in C_0^{k,\omega}(S)$ be the restriction of a $C^\infty$ function with compact support on ${\mbf R}^n$. Since $H|_{S'}:S'\rightarrow S$ is a proper map (by hypothesis (b) of the theorem), $(H|_{S'})^*f\in C_b^{k,\omega}(S')$ has compact support. Moreover, since $f\in C_b^{k,\omega_o}(S)$, Lemma \ref{lem6.3} applied to $\widetilde\omega=\omega_o$ implies that $(H|_{S'})^*f\in C_b^{k,\omega_o}(S')$. Finally, since $\widetilde T\in Ext(C_b^{k,\omega_o}(S');C_b^{k,\omega_o}({\mbf R}^n))$ as well, $Ef\in C_b^{k,\omega_o}({\mbf R}^n)$ and has compact support (because $(H|_{S'})^*f$ has it). Due to condition \eqref{equ1.8} for $\omega_o$ we obtain from here that $Ef\in C_0^{k,\omega}({\mbf R}^n)$. Since the set of such functions $f$ is dense in $C_0^{k,\omega}(S)$ (see Corollary \ref{cor1.10}), $E$ maps $C_0^{k,\omega}(S)$ in $C_0^{k,\omega}({\mbf R}^n)$, as required. \end{proof} Now the result of the theorem follows from Lemma \ref{lem6.4} and Theorem \ref{te1.4}\,(2); that is, $G_b^{k,\omega}(S)$ is isomorphic to $\bigl(C_0^{k,\omega}(S)\bigr)^*$ and so $G_b^{k,\omega}(S)$ and $C_0^{k,\omega}(S)$ have the metric approximation property (see the argument at the end of section~6.1 above). The proof of the theorem is complete. \end{proof}
12790d44df56e9367af1706e5d4b7a67a1df785f
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:introduction} In recent years there has been a resurgence of interest in the properties of metastable states, due mostly to the studies of the jammed states of hard sphere systems; see for reviews Refs. \onlinecite{charbonneau16, baule16}. There are many topics to study, including for example the spectrum of small perturbations around the metastable state, i.e. the phonon excitations and the existence of a boson peak, and whether the Edwards hypothesis works for these states. In this paper we shall study some of these topics in the context of classical Heisenberg spin glasses both in the presence and absence of a random magnetic field. Here the metastable states which we study are just the minima of the Hamiltonian, and so are well-defined outside the mean-field limit. It has been known for some time that there are strong connections between spin glasses and structural glasses ~\cite{tarzia2007glass,fullerton2013growing, moore06}. It has been argued in very recent work~\cite{baity2015soft} that the study of the excitations in classical Heisenberg spin glasses provides the opportunity to contrast with similar phenomenology in amorphous solids~\cite{wyart2005geometric, charbonneau15}. The minima and excitations about the minima in Heisenberg spin glasses have been studied for many years \cite{bm1981, yeo04, bm1982} but only in the absence of external fields. In Sec. \ref{sec:models} we define the models to be studied as special cases of the long-range one - dimensional $m$-component vector spin glass where the exchange interactions $J_{ij}$ decrease with the distance between the spins at sites $i$ and $j$ as $1/r_{ij}^{\sigma}$. The spin $\mathbf{S}_i$ is an $m$-component unit vector. $m=1$ corresponds to the Ising model, $m=2$ corresponds to the XY model and $m=3$ corresponds to the Heisenberg model. By tuning the parameter $\sigma$, one can have access to the Sherrington-Kirkpatrick (SK) model and on dilution to the Viana-Bray (VB) model, and indeed to a range of universality classes from mean-field-type to short-range type \cite{leuzzi2008dilute}, although in this paper only two special cases are studied; the SK model and the Viana-Bray model. We intend to study the cases which correspond to short-range models in a future publication. In Sec. \ref{sec:metastability} we have used numerical methods to learn about the metastable minima of the SK model and the Viana Bray model. Our main procedure for finding the minima is to start from a random configuration of spins and then align each spin with the local field produced by its neighbors and the external random field, if present. The process is continued until all spins are aligned with their local fields. This procedure finds local minima of the Hamiltonian. In the thermodynamic limit, the energy per spin $\varepsilon$ of these states reaches a characteristic value, which is the same for almost all realization of the bonds and random external fields, but slightly dependent on the dynamical algorithm used for selecting the spin to be flipped e.g. the ``polite'' or ``greedy'' or Glauber dynamics or the sequential algorithm used in the numerical work in this paper \cite{newman:99,parisi:95}. In the context of Ising spin glasses in zero random fields such states were first studied by Parisi \cite{parisi:95}. For Ising spins these dynamically generated states are an unrepresentative subset of the totality of the one-spin flip stable metastable states, which in general have a distribution of local fields $p(h)$ with $p(0)$ is finite \cite{roberts:81}, whereas those generated dynamically are marginally stable and have $p(h) \sim h$, just like that in the true ground state \cite{yan:15}. Furthermore these states have a trivial overlap with each other: $P(q)= \delta(q)$ \cite{parisi:95}; there is no sign of replica symmetry breaking amongst them. Presumably to generate states which show this feature one needs to start from initial spin configurations drawn from a realization of the system at a temperature where broken replica symmetry is already present before the quench. Because the initial state is random, one would also expect for vector spin glasses that the states reached after the quench from infinite temperature would have only a trivial overlap with each other \cite{newman:99} and this is indeed found to be the case in Sec. \ref{sec:overlap}. We have studied the energy which is reached in the quench for both the $m=2$ and $m=3$ SK models but for the case of zero applied random field and in both cases it is very close to the energy $E_c$ which marks the boundary above which the minima where spins are parallel to their local fields have trivial overlaps with each other, while below it the minima have overlaps with full broken replica symmetry features \cite{bm:81a, bm1981}. In Ref. \onlinecite{bm1981} the number $N_S(\varepsilon)$ of minima of energy $\varepsilon$ was calculated for the case of zero random field in the SK model and in fact it is only for this model and zero field that the value of $E_c$ is available. That is why in Sec. \ref{sec:marginal} only this case was studied numerically. The work in Sec. \ref{sec:SKanalytic} was the start of an attempt to have the same information in the presence of random vector fields. The number of minima $N_S(\varepsilon)$ is exponentially large so it is useful to study the complexity defined as $g(\varepsilon)= \ln N_S(\varepsilon)/N$, where $N$ is the number of spins in the system. Despite the fact that minima exist over a large range of values of $\varepsilon$ a quench by a particular algorithm seems to reach just the minima which have a characteristic value of $\varepsilon$. What is striking is that this characteristic value is close to the energy $E_c$ at which the minima would no longer have a trivial overlap with each other but would start to acquire replica symmetry breaking features, at least for the $m=2$ and $m=3$ SK models in zero field. The states reached in the quenches are usually described as being marginally stable \cite{muller:15}. The coincidence of the energy obtained in the numerical quenches with the analytically calculated $E_c$ suggests that long-range correlations normally associated with a continuous transition will also be found for the quenched minima since such features are present in the analytical work at $E_c$ \cite{bm:81a}. In the Ising case the field distribution $p(h)$ produced in the quench is very different from that assumed when determining $E_c$, and the quenched state energy at $\approx -0.73$ was so far below from the Ising value of $E_c=-0.672$ that the connection of its marginality to the onset of broken replica symmetry has been overlooked. We believe that the identification of the energy $E_c$ reached in the quench with the onset of replica symmetry breaking in the overlaps of the minima is the most important of our results. In Sec. \ref{sec:SKanalytic} we present our analytical work on the $m$-component SK model in the presence of an $m$-component random field. It has been shown that in the mean-field limit ~\cite{sharma2010almeida} that under the application of a random magnetic field, of variance $h_r^2$, there is a phase transition line in the $h_r - T$ plane, the so-called de Almeida-Thouless (AT) line, across which the critical exponents lie in the Ising AT universality class. Below this line, the ordered phase has full replica symmetry breaking. This ordered phase is similar to the Gardner phase expected in high-dimensional hard sphere systems \cite{charbonneau16}. In Sec. \ref{sec:SKanalytic} we study the minima of the Heisenberg Hamiltonian in the presence of a random vector field. In the presence of such a field the Hamiltonian no longer has any rotational invariances so one might expect there to be big changes in the excitations about the minimum as there will be no Goldstone modes in the system. We start Sec. \ref{sec:SKanalytic} by studying the number of local minima $N_S(\varepsilon)$ of the Hamiltonian which have energy per spin of $\varepsilon$. The calculation within the annealed approximation, where one calculates the field and bond averages of $N_S(\varepsilon)$ is just an extension of the earlier calculation of Bray and Moore for zero random field \cite{bm1981}. When the random field $h_r> h_{AT}$, where $h_{AT}$ is the field at which the AT transition occurs, the complexity is zero, but $g(\varepsilon)$ becomes non-zero for $h_r < h_{AT}$. When it is non-zero, it is thought better to average the complexity itself over the random fields and bonds so that one recovers results likely to apply to a typical sample. We have attempted to calculate the quenched complexity $g$ for the SK model in the presence of a random field. The presence of this random field greatly complicates the algebra and the calculations in Sec. \ref{sec:quenched} and the Appendix really just illustrate the problems that random fields pose when determining the quenched average but do not overcome the algebraic difficulties. The annealed approximation is much simpler and using it we have calculated the density of states $\rho(\lambda)$ of the Hessian matrix associated with the minimum for the SK model. When $h_r > h_{AT}$ there is a gap $\lambda_0$ in the spectrum below which there are no excitations. $\lambda_0$ tends to zero as $h_r \to h_{AT}$. For $m \ge 4$, $\rho(\lambda) \sim \sqrt{\lambda-\lambda_0}$ as $\lambda \to \lambda_0$. For $m =3$ the square root singularity did not occur, much to our surprise. For $h_r < h_{AT}$, the square-root singularity applies for all $m > 2$ with $\lambda_0=0$. Thus in the low-field phase, despite the fact that in the presence of the random fields there are no continuous symmetries in the system and hence no Goldstone modes, there are massless modes present. In Sec. \ref{sec:density} we present numerical work which shows that even for $h_r < h_{AT}$ when the annealed calculation of the density of states of the SK model cannot be exact, it nevertheless is in good agreement with our numerical data. We have also calculated in Sec. \ref{sec:spinglasssusceptibility} the zero temperature spin glass susceptibility $\chi_{SG}$ for $h_r > h_{AT}$ for the SK model and find that for all $m > 2$ it diverges to infinity as $h_r \to h_{AT}$ just as is found at finite temperatures \cite{sharma2010almeida}. For the SK model, because the complexity is zero for $h_r > h_{AT}$, the quench produces states sensitive to the existence of an AT field. The quench then goes to a state which is the ground state or at least one very like it. The AT field is a feature of the true equilibrium state of the system, which in our case is the state of lowest energy. In Sec. \ref{sec:spinglasssusceptibility} we have studied a ``spin glass susceptibility'' obtained from the minima obtained in our numerical quenches and only for the SK model is there evidence for a diverging spin glass susceptibility. For the VB model, there is no sign of any singularity in the spin glass susceptibility defined as an average over the states reached in our quench from infinite temperature, but we cannot make any statement concerning the existence of an AT singularity in the true ground state. This is the problem studied in Ref. \onlinecite{lupo:16}. Finally in Sec. \ref{sec:conclusions} we summarise our main results and make some suggestions for further research. \section{Models} \label{sec:models} The Hamiltonians studied in this paper are generically of the form \begin{equation} \mathcal{H} = -m\sum_{\langle i, j \rangle} J_{ij} \mathbf{S}_i \cdot \mathbf{S}_j - \sqrt{m}\sum_i \mathbf{h}_i \cdot \mathbf{S}_i \, , \label{Ham} \end{equation} where the $\mathbf{S}_i$, $i = 1, 2, \cdots, N$, are classical $m$-component vector spins of unit length. This form of writing the Hamiltonian allows for easy comparison against a Hamiltonian where the spins are normalized to have length $\sqrt{m}$. We are particularly interested in Heisenberg spins, for which $m=3$. The magnetic fields $h_i^\mu$, where $\mu$ denotes a Cartesian spin component, are chosen to be independent Gaussian random fields, uncorrelated between sites, with zero mean, which satisfy \begin{equation} [ h_i^\mu h_j^\nu]_{av} = h_r^2\, \delta_{ij}\, \delta_{\mu\nu} \, . \label{hs} \end{equation} The notation $[\cdots]_{av}$ indicates an average over the quenched disorder and the magnetic fields. We shall study two models, the Sherrington-Kirkpatrick (SK) model and the Viana-Bray (VB) model. Both are essentially mean-field models. In the Sherrington-Kirkpatrick model, the bonds $J_{ij}$ couple all pairs of sites and are drawn from a Gaussian distribution with zero mean and the variance $1/(N-1)$. The Viana-Bray model can be regarded as a special case of a diluted one-dimensional model where the sites are arranged around a ring. The procedure to determine the bonds $J_{ij}$ to get the diluted model is as specified in Refs. \onlinecite{leuzzi2008dilute,sharma2011phase,sharma2011almeida}. The probability of there being a non-zero interaction between sites $(i,j)$ on the ring falls off with distance as a power-law, and when an interaction does occur, its variance is independent of $r_{ij}$. The mean number of non-zero bonds from a site is fixed to be $z$. To generate the set of pairs $(i,j)$ that have an interaction with the desired probability the spin $i$ is chosen randomly, and then $j \ (\ne i)$ is chosen at distance $r_{ij}$ with probability \begin{equation} p_{ij} = \frac{r_{ij}^{-2\sigma}}{\sum_{j\, (j\neq i)}r_{ij}^{-2\sigma}} \, , \end{equation} where $r_{ij}=\frac{N}{\pi}\sin\left[\frac{\pi}{N}(i-j)\right]$ is the length of the chord between the sites $i,j$ when all the sites are put on a circle. If $i$ and $j$ are already connected, the process is repeated until a pair which has not been connected before is found. The sites $i$ and $j$ are then connected with an interaction picked from a Gaussian interaction whose mean is zero and whose standard deviation is set to $J \equiv 1$. This process is repeated precisely $N_b = z N / 2 $ times. This procedure automatically gives $J_{ii} = 0$. Our work concentrates on the case where the coordination number is fixed at $z=6$ to mimic the $3$-d cubic scenario. The SK limit ($z=N-1, \sigma = 0$) is a special case of this model, as is the VB model which also has $\sigma = 0$, but the coordination number $z$ has (in this paper) the value $6$. The advantage of the one-dimensional long-range model for numerical studies is that by simply tuning the value of $\sigma$ one can mimic the properties of finite dimensional systems~\cite{leuzzi2008dilute,sharma2011phase,sharma2011almeida} and we have already done some work using this device. However, in this paper we only report on our work on the SK and VB models. \section{Numerical studies of the minima obtained by quenching} \label{sec:metastability} In this section we present our numerical studies of the minima of the VB and SK models. We begin by describing how we found the minima numerically. They are basically just quenches from infinite temperature. In Sec. \ref{sec:overlap} we have studied the overlap between the minima and we find that the minima produced have only trivial overlaps with one another. In Sec \ref{sec:marginal} we describe our evidence that the minima of the SK model in zero field have marginal stability as they have an energy per spin close to the energy $E_c$ which marks the energy at which the minima starting to have overlaps showing replica symmetry breaking features. At zero temperature, the metastable states (minima) which we study are those obtained by aligning every spin along its local field direction, starting off from a random initial state. In the notation used for our numerical work based on Eq. (\ref{Ham}) we iterate the equations \begin{equation} \mathbf{S}^{n+1}_i= \frac{\mathbf{H}^{n}_i}{|\mathbf{H}_i^{n}|}, \label{eq:parn} \end{equation} where the local fields after the $n$th iteration, $\mathbf{H}_i^{n}$, are given by \begin{equation} \mathbf{H}_i^n= \sqrt{m} \mathbf{h}_i+m \sum_j J_{ij} \mathbf{S}_j^{n}. \label{eq:hdefn} \end{equation} For a given disorder sample, a random configuration of spins is first created which would be a possible spin configuration at infinite temperature. Starting from the first spin and scanning sequentially all the way up to the $N^{th}$ spin, every spin is aligned to its local field according to Eq.~(\ref{eq:parn}), this whole process constituting one sweep. The vector $(\Delta \mathbf{S}_{1},\Delta \mathbf{S}_{2},\cdots,\Delta \mathbf{S}_{N})$ is computed by subtracting the spin configuration before the sweep from the spin configuration generated after the sweep. The quantity $\eta = \frac{1}{Nm}\sum_{\mu=1}^{m}\sqrt{\sum_{j=1}^{N}(\Delta S_{j\mu})^{2}}$ is a measure of how close the configurations before and after the sweep are. The spin configurations are iterated over many sweeps until the value of $\eta$ falls below $0.00001$, when the system is deemed to have converged to the metastable state described by Eq.~(\ref{eq:par}), which will be a minimum of the energy at zero temperature. Differing starting configurations usually generate different minima, at least for large systems. \subsection{Overlap distribution} \label{sec:overlap} \begin{figure} \includegraphics[width=\columnwidth]{fig_VB.eps} \caption{(Color online) The overlap distribution $P(q)$ for the VB model ($\sigma =0, z =6$, $h_r=0.6$) for the minima generated by the prescription described in the text. $P(q)$ seems to be approaching a delta function as $N$ tends to infinity.} \label{fig0} \end{figure} It is informative to study the overlaps between the various minima. Consider the overlap between two minima $A$ and $B$ defined as \begin{equation} q \equiv \frac{1}{N}\sum_{i}\mathbf{S}_i^{A} \cdot \mathbf{S}_i^{B}. \end{equation} Numerically, the following procedure is adopted. A particular realization of the bonds and fields is chosen. Choosing a random initial spin configuration, the above algorithm is implemented and descends to a locally stable state. This generates a metastable spin state that is stored. One then chooses a second initial condition, and the algorithm is applied, which generates a second metastable spin state which is also stored. One repeats this $N_{min}$ times generating in total $N_{min}$ metastable states (some or all of which might be identical). One then overlaps all pairs of these states, so there are $N_{pairs} = N_{min}(N_{min}-1)/2$ overlaps which are all used to make a histogram. The whole process is averaged over $N_{samp}$ samples of disorder. Fig.~\ref{fig0} shows the overlap distribution of the metastable states obtained by the above prescription for the VB model. The figure suggests that in the thermodynamic limit, the distribution of overlaps, $P(q) = \delta(q-q_0(h_r))$. In zero field we have found that $q_0(h_r=0)=0$. Since we study only a finite system of $N$ spins, the delta function peak is broadened to a Gaussian centered around $q_0$ and of width $O(\frac{1}{\sqrt{N}})$. We studied also the SK model, for a range of values for the $h_r$ fields, and the data are consistent with $P(q)$ just having a single peak in the thermodynamic limit. This suggests that the metastable states generated by the procedure of repeatedly putting spins parallel to their local fields starting from a random state always produces minima which have a $P(q)$ of the same type as would be expected for the paramagnetic phase. Newman and Stein \cite{newman:99} showed that for Ising spins in zero field that when one starts off from an initial state, equivalent to being at infinite temperature, and quenches to zero temperature one always ends up in a state with a trivial $P(q)=\delta(q)$, in agreement, for example with the study of Parisi \cite{parisi:95}. Our results for vector spin glasses seem exactly analogous to the Ising results. \subsection{Marginal stability} \label{sec:marginal} In this subsection we shall focus on the Ising, XY ($m = 2$) and Heisenberg ($m=3$) SK models with zero random field. Parisi found for the Ising case that when starting a quench from infinite temperature, when the spins are just randomly up or down, and putting spins parallel to their local fields according to various algorithms, the final state had an energy per spin $\varepsilon=-0.73$ \cite{parisi:95}. In their studies of one-spin flip stable spin glasses in zero field, Bray and Moore \cite{bm1981,bm:81a} found that such states associated with a trivial $P(q)=\delta(q)$ should not exist below a critical energy $E_c$ and for the Ising case $E_c=-0.672$. States with an energy close to $-0.73$ would be expected to be have a $P(q)$ rather similar to those for full replica symmetry breaking, but those generated in the quench have a trivial $P(q)$. There is no paradox as the states generated in the quench have more than one-spin flip stability \cite{yan:15}. This results in a distribution of local fields behaving at small fields so that $p(h) \sim h$, very different from that expected from the study of the $p(h)$ of one-spin flip stable states \cite{roberts:81} for which $p(0)$ is finite, and instead similar to what is found in the true ground state -- the state which is stable against flipping an arbitrary number of spins. It is by that means that the theorem of Newman and Stein \cite{newman:99} that in a quench from a random initial state the final $P(q)$ should be trivial is realized, despite the quenched energy being in the region where one would expect the $P(q)$ of one spin flip stable states to be non-trivial. The change in the form of $p(h)$ means that the true $E_c$ is not at $-0.672$, but instead is at least closer to $-0.73$. For the vector SK spin glasses in zero field we have studied the energy reached in a quench from infinite temperature by putting the spins parallel to their local fields. In Figs. \ref{fig:SKXYEc} and \ref{fig:SKHeisEc} we have plotted our estimates of this energy as a function of $1/N^{2/3}$, the form commonly used for the energy size dependence of the SK model \cite{boettcher:03, billoire:08}. For $m=2$, the extrapolated energy per spin component is $\approx -0.870$, whereas its $E_c=-0.866$ according to the analysis in Ref. \onlinecite{bm1981}; for $m=3$ the extrapolated energy per spin component is $\approx -0.915$ whereas its $E_c =-0.914$ \cite{bm1981}. Minima whose energies lie below the critical energy $E_c$, are associated with non-trivial (i.e. RSB) form for their $P(q)$, calculated from the overlaps of the minima at the same energy \cite{bm1981, bm:81a}. We found just as for the Ising SK model that the energy reached in the quench varied little when the greedy algorithm was used instead of the sequential algorithm \cite{parisi:95}. \begin{figure} \includegraphics[width=\columnwidth]{fig_XY_SK_energy.eps} \caption{(Color online) The average energy per site and spin component for the XY SK spin glass model ($m =2$) with $h_r=0$ plotted against $1/N^{2/3}$ in order to estimate the infinite system value of the energy obtained from a quench from infinite temperature. For $m=2$, $E_c=-0.866$ \cite{bm1981}.} \label{fig:SKXYEc} \end{figure} \begin{figure} \includegraphics[width=\columnwidth]{fig_Heisenberg_SK_energy.eps} \caption{(Color online) The average energy per site and spin component for the Heisenberg SK spin glass ($m =3$) with $h_r=0$ plotted against $1/N^{2/3}$ in order to estimate the infinite system value of the energy obtained from a quench from infinite temperature. For $m =3$, $E_c=-0.914$ \cite{bm1981}. } \label{fig:SKHeisEc} \end{figure} As the energy of the quenched state is remarkably close to the critical energies calculated by Bray and Moore \cite{bm1981,bm:81a} for $m=2$ and $m=3$, this suggests that the state reached in the quench is well-described by the calculations in Ref. \onlinecite{bm1981}, whereas for the Ising case the equivalent calculation which enumerates the number of one-spin flip stable states does not give the resulting $p(h)$ of the quenched states with much accuracy and so does not produce an accurate estimate of $E_c$. One knows a lot about behavior at $E_c$ at least for Ising spins in zero random field \cite{bm:81a}. For states of energy per spin $\varepsilon > E_c$, the annealed and quenched averages agree with each other, but for energies $\varepsilon < E_c$, the two calculations differ. As $\varepsilon$ approaches $E_c$, behavior is as at a critical point, with growing length scales etc. and massless modes \cite{bm:81a}. For the Ising case the properties of these modes were discussed in Ref. \onlinecite{bm:81a}. We intend to return to this topic in a future publication for the case of vector spin glasses. When one sets an Ising spin parallel to its local field in the course of the quench, spin avalanches may be triggered. If the number of neighbors $z$ is of order $N$ then the avalanches can be on all size scales \cite{boettcher:08, andresen:13}. Thus the Ising SK model is an example of a system with marginal stability as discussed by M\"{u}ller and Wyart \cite{muller:15}. It was argued in Ref. \onlinecite{muller:15} that as the quench progresses the system will reach the marginal manifold which separates stable from unstable configurations. As this point is approached the dynamics slows and eventually freezes near the marginal manifold. The VB model with $z =6$ does not have large scale avalanches \cite{andresen:13} and does not have any marginal features; a first study of avalanches in the undiluted one-dimensional long-range models can be found in \cite{boettcher:08}. While the Ising VB model does not have large scale avalanches, there certainly will be an energy $E_c$ below which the minima will have non-trivial overlaps. What is not clear is whether it is the large avalanches which ensures that the states generated in a quench are close to this energy. We also do not know what difference the existence of a finite temperature phase might make to the properties of the quenched state. For example, are there features of the quenched states of one and two dimensional Ising spin glasses, where there is no finite temperature spin glass transition, which differ significantly from those of the three dimensional spin glass, where there is a finite temperature phase transition? We also do not know what features might arise if there is a phase transition to a state with full replica symmetry breaking, as opposed to a state with just replica symmetry. For systems for which the excitations are not discrete, such as in vector spin glasses, marginality takes a different form, and seems related to the development of negative eigenvalues in the Hessian \cite{muller:15,sharma2014avalanches}. Such eigenvalue instabilities might be triggered in a quench where one puts spins parallel to their local fields. On the other hand, one could imagine a steepest descent procedure starting from the initial spin orientation and smoothly proceeding to a minimum. Does that result in a final state whose properties differ from those generated by putting spins parallel to their local fields? There are many topics which should be studied! We believe that the proximity of the quenched energy to the calculated critical energy $E_c$, at least for the cases of $m =2$ and $m =3$ will provide valuable analytical insights concerning marginal stability. One of our motivations for the analytic work in the next section was to calculate $E_c(h_r)$ in the presence of a non-zero random vector field, but, as we shall see, algebraic difficulties prevented us from achieving this goal. But it would be good to know how general is the result that the energy obtained in a quench coincides with the energy at which the overlaps of the minima display replica symmetry breaking features. \section{Metastable states in the SK model in the presence of a random field} \label{sec:SKanalytic} In this section we follow the method of Ref. \onlinecite{bm1981} to study the complexity and Hessian properties of the minima for the SK model but in the presence of a random vector field. We begin by writing down the first steps in the formalism following Ref. \onlinecite{bm1981}. In subsection \ref{SKannealed} we show that within the annealed approximation, where one averages $N_S(\varepsilon)$ itself over the bonds $J_{ij}$ and the random fields $h_i^{ex}$ analytical progress is fairly straightforward. Fortunately the annealed approximation is also exact for fields $h_r > h_{AT}$. In subsection \ref{sec:quenched} we describe our attempts to solve the quenched case. We believe that our approach based on replica symmetry assumptions should be good down to its limit of stability which would be at $E_c(h_r)$, but algebraic difficulties prevented us from actually determining $E_c(h_r)$. We find it convenient to write the Hamiltonian for the $m$-vector spin glass in an $m$-component external field as \begin{equation} \mathcal{H}=-\frac{m}{2}\sum_{i,j} J_{ij}\bm{S}_i\cdot\bm{S}_j-m\sum_i\bm{h}^{\rm ex}_i\cdot\bm{S}_i, \label{Hamil} \end{equation} where the $m$-component spins $\bm{S}_i=\{S^\alpha_i\}$, ($\alpha=1,\cdots,m$, $i=1,\cdots,N$) have a unit length $S_i=1$. The interactions $J_{ij}$ are chosen from a Gaussian distribution with zero mean and the variance $1/N$. In this section, for convenience, we use the notation $\bm{h}^{\rm ex}_i=\bm{h}_i/\sqrt{m}$ for the random Gaussian external fields with zero mean and the variance \begin{equation} \langle h^{{\rm ex},\alpha}_i h^{{\rm ex},\beta}_j\rangle =\frac{h^2_r}{m}\delta_{ij}\delta^{\alpha\beta}. \end{equation} At zero temperature, the spins are aligned in the direction of the local internal field $\bm{H}_i$, i.e. \begin{equation} \bm{S}_i=\hat{\bm{H}}_i\equiv\frac{\bm{H}_i}{H_i}, \label{eq:par} \end{equation} where \begin{equation} \bm{H}_i=\sum_j J_{ij}\bm{S}_j +\bm{h}^{\rm ex}_i. \label{eq:hdef} \end{equation} In terms of the local fields, the ground state energy $E$ can be written as \begin{equation} E=-\frac{m}{2}\sum_i(H_i+\hat{\bm{H}}_i\cdot \bm{h}^{\rm ex}_i). \end{equation} The number of metastable states with energy $\varepsilon$ per site and per spin component is given by \begin{align} &N_S(\varepsilon)=\int\prod_{i,\alpha}dH_i^{\alpha}\int\prod_{i,\alpha}dS_i^{\alpha} \prod_{i,\alpha}\delta(S_i^{\alpha}-\hat{H}_i^{\alpha}) \nonumber \\ &~~~~~~\times \prod_{i,\alpha}\delta\left(H_i^{\alpha}-\sum_j J_{ij}S_j^{\alpha}-h^{{\rm ex},\alpha}_{i}\right) |\det\mathsf{M}\{J_{ij}\}| \nonumber \\ &~~~~~~\times \delta\left(Nm\varepsilon+\frac 1 2 m \sum_i (H_i + \hat{\bm{H}}_i\cdot\bm{h}^{\rm ex}_i ) \right), \label{NS} \end{align} where \begin{equation} M^{\alpha\beta}_{ij}=\frac{\partial}{\partial S^\beta_j}(S^\alpha_i-\hat{H}^\alpha_i) =\delta_{ij}\delta^{\alpha\beta}-J_{ij}\frac{P^{\alpha\beta}_i}{H_i} \end{equation} with $P^{\alpha\beta}_i\equiv \delta^{\alpha\beta}-\hat{H}^\alpha_i\hat{H}^\beta_i$ is the projection matrix. \subsection{Annealed Approximation} \label{SKannealed} We now calculate the average of $N_S(\varepsilon)$ over the random couplings and the random external fields. As we will see below, the direct evaluation of the quenched average $\langle \ln N_S(\varepsilon) \rangle$ is very complicated. Here we first present the annealed approximation, where we evaluate the annealed complexity $g_A(\varepsilon)=\ln \langle N_S(\varepsilon) \rangle/N$. The whole calculation is very similar to those in Appendix 2 of Ref.~\onlinecite{bm1981} except for the part involving the average over the random field. Below we sketch the calculation. The first delta functions in Eq.~(\ref{NS}) can be integrated away. We use the integral representations for the second and third delta functions using the variables $x_i^{\alpha}$ and $u$, respectively, along the imaginary axis. The average over the random couplings can be done in an exactly the same way as in Ref.~\onlinecite{bm1981}. We briefly summarize the results below. The random couplings appear in the factor \begin{align} & \left\langle \exp\Big[-\sum_{i<j}J_{ij}\sum_{i,\alpha} (x^\alpha_{i}\hat{H}^\alpha_{j} +x^\alpha_{j}\hat{H}^\alpha_{i})\Big] |\det\mathsf{M} \{J_{ij}\}| \right\rangle_J \nonumber \\ =&\exp \Big[ \frac 1 {2N} \sum_{i<j}\Big\{ \sum_{\alpha}(x^\alpha_{i}\hat{H}^\alpha_{j} +x^\alpha_{j}\hat{H}^\alpha_{i}) \Big\}^2\Big] \nonumber \\ &~~~~~~~~~~~~~\times \left\langle |\det\mathsf{M} \{J_{ij}-O(\frac 1 N)\}| \right\rangle_J . \end{align} After neglecting the $O(1/N)$ term, we evaluate the average of the determinant as \cite{bm1981} \begin{equation} \left\langle |\det\mathsf{M} \{J_{ij}\}| \right\rangle_J=\exp(\frac 1 2 Nm\bar{\chi})\prod_i \left(1-\frac{\bar{\chi}}{H_i}\right)^{m-1}, \label{det} \end{equation} where the susceptibility $\bar{\chi}$ satisfies the self-consistency equation \cite{bm1981} \begin{equation} \bar{\chi}=(1-\frac 1 m)\frac 1 N \sum_i \frac 1 {H_i-\bar{\chi}} \label{chi} \end{equation} with the condition $H_i\ge\bar{\chi}$. Using the rotational invariance and the Hubbard-Stratonovich transformation, we can rewrite the exponential factor in front of the determinant as \begin{align} &\exp[\frac 1{2m}\sum_{i,\alpha}(x^\alpha_{i})^2 ] \\ &~\times \int\frac{dv}{(2\pi/Nm)^{1/2}}\; \exp [ -\frac{Nm}{2} v^2+ v \sum_{i,\alpha}x^\alpha_{i}\hat{H}^\alpha_{i} ]. \nonumber \end{align} In the present case, we have to average over the random field. Collecting the relevant terms, we have \begin{align} & \left\langle \exp\Big[-\sum_{i,\alpha} (x^\alpha_{i}+\frac 1 2 u m \hat{H}^\alpha_{i})h^{{\rm ex},\alpha}_{i}\Big] \right\rangle_{\bm{h}^{\rm ex}} \\ =&\exp\Big[ \frac{h^2_r}{2m}\sum_{i,\alpha}(x^\alpha_{i})^2 +\frac{h^2_r}{2} u \sum_{i,\alpha}x^\alpha_{i}\hat{H}^\alpha_{i} +Nm\frac{h^2_r }{8}u^2 \Big] . \nonumber \end{align} All the site indices are now decoupled. We express the condition Eq.~(\ref{chi}) using the integral representation of the delta function with the variable $\lambda$ running along the imaginary axis. Putting all the terms together, we have \begin{align} &\langle [N_S(\varepsilon)] \rangle_{J,h^{\rm ex}}= \int \frac{du}{2\pi i} \int \frac{dv}{\sqrt{2\pi/Nm}} \int d\bar{\chi} \int \frac{d\lambda}{2\pi i} \nonumber \\ & \times \exp\Big[ Nm \lambda \bar{\chi} + \frac{Nm}{2} \bar{\chi}^2 -Nm\varepsilon u -\frac{Nm}{2} v^2 \nonumber \\ &~~~~~~~~~~~~~~~~~~~~~~~+Nm\frac{h^2_r}{8}u^2 +N\ln I^\prime \Big], \label{NS_ann} \end{align} where \begin{align} I^\prime=&\int_{H\ge\bar{\chi}}\prod_{\alpha}dH^\alpha \int\prod_{\alpha}\frac{dx^\alpha}{2\pi i} \left(1-\frac{\bar{\chi}}{H}\right)^{m-1} \nonumber \\ &\times \exp\Bigg[ \frac{1+h^2_r}{2m}\sum_{\alpha}(x^\alpha)^2 + (v+\frac{h^2_r}{2}u)\sum_\alpha x^\alpha\hat{H}^\alpha \nonumber \\ & +\sum_{\alpha}x^\alpha H^\alpha -(m-1) \lambda (H-\bar{\chi})^{-1} -\frac{m}{2} u H \Bigg] \end{align} The Gaussian integral over $x^\alpha$ can be done analytically. The integrals in Eq.~(\ref{NS_ann}) are evaluated via the saddle point method in the $N\to\infty$ limit. Following the procedure described in Ref.~\onlinecite{bm1981}, we introduce new variables $\bm{h}\equiv (H-\bar{\chi})\hat{\bm{H}}$ and $\Delta=-v-\bar{\chi}$ and use the saddle point condition for $\bar{\chi}$, which is \begin{equation} \lambda-\Delta-\frac u 2=0. \end{equation} We finally have an expression for the annealed complexity $g_A(\varepsilon)\equiv N^{-1}\ln\langle N_S(\varepsilon)\rangle$ as \begin{equation} g_A(\varepsilon) = m (-\frac{\Delta^2}{2} -\varepsilon u +\frac{h^2_r}{8}u^2) +\ln I, \end{equation} where \begin{align} &I= \left(\frac{m}{2\pi(1+h^2_r)}\right)^{m/2}S_m\int^\infty_0 dh\; h^{m-1} \\ &\times\exp\Big[ -\frac{m}{2(1+h^2_r)}(h-\Delta+\frac{h^2_r}{2}u)^2 \nonumber \\ &~~~~~~~~~~~~~~~ -\frac{(m-1)}{h}(\Delta+\frac u 2) -\frac{m}{2}uh\Big] \nonumber \end{align} with the surface area of the $m$-dimensional unit sphere $S_m=2\pi^{m/2}/\Gamma(m/2)$. The parameters $\Delta$ and $u$ are determined variationally as $\partial g_A/\partial \Delta =\partial g_A/\partial u =0$. We focus on the total number of metastable states, which are obtained by integrating $\exp(Ng_A(\varepsilon))$ over $\varepsilon$, or equivalently by setting $u=0$. Thus we are effectively focussing on the most numerous states, those at the top of the band where $g_A(\varepsilon)$ is largest. In this case, $g_A=-(m/2)\Delta^2+\ln I_0$, where \begin{align} I_0 =& S_m \left(\frac{m}{2\pi(1+h^2_r)}\right)^{m/2}\int_0^\infty dh\; h^{m-1} \nonumber \\ &\times \exp\left[ -(m-1)\frac{\Delta}{h}-\frac{m(h-\Delta)^2}{2(1+h^2_r)}\right] \label{I0}. \end{align} The parameter $\Delta$ is determined by the saddle point equation \begin{equation} \Delta=\frac{1}{2+h^2_r}\langle h \rangle - \left(1-\frac{1}{m}\right)\left(\frac{1+h^2_r}{2+h^2_r}\right) \left\langle\frac{1}{h}\right\rangle, \label{Delta} \end{equation} where the average is calculated with respect to the probability distribution for the internal field given by the integrand of $I_0$ in Eq.~(\ref{I0}). Using $\langle h \rangle =\Delta + \langle h-\Delta \rangle$, we can rewrite Eq.~(\ref{I0}) as \begin{equation} \Delta \left[ 1- \left(1-\frac 1 m \right)\left\langle \frac 1 {h^2} \right\rangle \right]=0. \label{Delta1} \end{equation} For various values of the external field $h_r$, we solve numerically Eq.~(\ref{Delta}). For $m=3$, we find that when $h_r>h_{AT}=1$ there is only a trivial solution, $\Delta=0$. (Note that the Almeida-Thouless field $h_{AT}$ at $T=0$ is $h_{AT}=1/\sqrt{m-2}$ ~\cite{sharma2010almeida}). From Eq.~(\ref{I0}), we see that in this case $I_0=1$ and the complexity $g$ vanishes above the AT field. For $h_r<h_{AT}$, a nontrivial solution, $\Delta\neq 0$ exists. We find that the values of $\Delta$ and $g_A$ increase as the external field $h_r$ decreases from $h_{AT}$, and approach the known values, 0.170 and 0.00839 at zero external field \cite{bm1981}. For $h_r$ smaller than but very close to $h_{AT}$, $\Delta$ is very small. We may obtain an analytic expression for $g_A$ in this case. By expanding everything in Eq.~(\ref{Delta1}) in powers of $\Delta$, we find for $m=3$ that \begin{equation} g_A=\frac 3 2 (h^2_{AT}-h^2_r)\tilde{\Delta}^2+8\sqrt{\frac{3}{2\pi}} \tilde{\Delta}^3\ln\tilde{\Delta}+O(\tilde{\Delta}^3), \label{gA0} \end{equation} where $\tilde{\Delta}=\Delta/\sqrt{1+h^2_r}$. The fact that $g_A$ must be stationary with respect to $\tilde{\Delta}$, enables one to determine how the complexity vanishes as $h_r \to h_{AT}$ and the value of $\tilde{\Delta}$ in this limit. Using the distribution for the internal field $H$ (or $h$), we first calculate the spin glass susceptibility $\chi_{SG}\equiv (Nm)^{-1}\mathrm{Tr} \mathbf{\chi}^2$ with the susceptibility matrix $\mathbf{\chi}=\chi_{ij}^{\alpha\beta}$ \cite{bm1981}. Note that the susceptibility in Eq.~(\ref{chi}) is just $\bar{\chi}=(Nm)^{-1}\mathrm{Tr}\mathbf{\chi}$. The spin glass susceptibility is given by \cite{bm1981} $\chi_{SG}=(1-\lambda_R)/\lambda_R$, where \begin{equation} \lambda_R=1-(1-\frac 1 m)\frac 1 N \sum_i \frac 1{(H_i-\bar{\chi})^2}. \label{lambda_R} \end{equation} This quantity is exactly the one in the square bracket in Eq.~(\ref{Delta1}). Therefore, since $\Delta\neq 0$ for $h_r<h_{AT}$, $\lambda_R$ vanishes and consequently $\chi_{SG}$ diverges. Above the AT field, there is only a trivial solution $\Delta=0$. In this case the integrals are just Gaussians and we can evaluate explicitly $\frac{1}{N} \sum_i \frac 1{(H_i-\bar{\chi})^2}$, with the result that $\lambda_R=(h_r^2-1/(m-2))/(1+h_r^2)$, so the spin glass susceptibility as a function of the external random field for $h_r>h_{AT}$ is given by \begin{equation} \chi_{SG}=\frac{1+h_{AT}^2}{h_r^2-h_{AT}^2}, \label{chisgexact} \end{equation} provided $h_r > h_{AT}$ and $m > 2$. The simple divergence of $\chi_{SG}$ as $h_r \to h_{AT}$ is a feature of the SK limit and is not found in the Viana-Bray model at least amongst the quenched states of our numerical studies, see Sec. \ref{sec:spinglasssusceptibility} We now calculate the eigenvalue spectrum of the Hessian matrix $\mathsf{A}$. The calculation closely follows the steps in Ref.~\onlinecite{bm1982} for the case of zero external field. We consider (transverse) fluctuations around the $T=0$ solution $\bm{S}^{0}_i\equiv\hat{\bm{H}}_i$ by writing $\bm{S}_i=\bm{S}^{0}_i+\bm{\epsilon}_i$, where $\bm{\epsilon}_i=\sum_\alpha \epsilon^\alpha_i \hat{\bm{e}}_\alpha(i)$ with the $(m-1)$ orthonormal vectors $\hat{e}_\alpha(i)$, $\alpha=1,\cdots,m-1$ satisfying $\bm{S}^{0}_i\cdot \hat{\bm{e}}_\alpha(i)=0$. Inserting this into Eq.~(\ref{Hamil}), we have the Hessian matrix as \begin{equation} A^{\alpha\beta}_{ij}\equiv\frac{\partial(\mathcal{H}/m)}{\partial\epsilon^\alpha_i \partial\epsilon^\beta_j} =H_i\delta_{ij}\delta^{\alpha\beta}-J_{ij} \hat{\bm{e}}_\alpha(i)\cdot\hat{\bm{e}}_\beta(j). \end{equation} The eigenvalue spectrum $\rho(\lambda)$ can be calculated from the resolvent $\mathsf{G}=(\lambda\mathsf{I}-\mathsf{A})^{-1}$ as \begin{equation} \rho(\lambda)=\frac{1}{N(m-1)\pi}\mathrm{Im} \; \mathrm{Tr} \mathsf{G}(\lambda-i\delta), \label{eqn:rho} \end{equation} where $\mathsf{I}$ is the $(m-1)N$-dimensional unit matrix and $\delta$ is an infinitesimal positive number. The locator expansion method \cite{bm1979} is used to evaluate $\rho(\lambda)$, which yields the following self-consistent equation for $\bar{G}(\lambda)\equiv ((m-1)N)^{-1} \mathrm{Tr} \mathsf{G}(\lambda)$: \begin{equation} \bar{G}(\lambda)=\left\langle \frac 1{\lambda-H-(1-\frac 1 m)\bar{G}(\lambda)}\right\rangle, \label{Gbar} \end{equation} where $\langle ~\rangle$ denotes the average over the distribution for $h$ given in the integrand in Eq.~(\ref{I0}). Note that $H=h+\bar{\chi}$ and $\bar{\chi}=(1-1/m)\langle 1/h\rangle$ from Eq.~(\ref{chi}). We first separate $\bar{G}=\bar{G}^\prime+i\bar{G}^{\prime\prime}$ into real and imaginary parts and solve Eq.~(\ref{Gbar}) numerically for $\bar{G}^\prime(\lambda)$ and $\bar{G}^{\prime\prime}(\lambda)$ as a function of $\lambda$. The eigenvalue spectrum is just $\rho(\lambda)=\pi^{-1}\bar{G}^{\prime\prime}(\lambda)$. \begin{figure} \includegraphics[width=\columnwidth]{rho_SK.eps} \caption{(Color online) The eigenvalue spectrum of the Hessian at zero temperature for the vector spin glass with $m=3$ in the SK limit.The various lines correspond to different values of $h_r$, the external random field.} \label{fig:rho} \end{figure} \vspace{0.2cm} \begin{figure} \includegraphics[width=\columnwidth]{rho_mag_SK.eps} \caption{(Color online) The magnified view of the same figure as Fig.~\ref{fig:rho} but for the small eigenvalues.} \label{fig:rho_mag} \end{figure} As we can see from Figs.~\ref{fig:rho} and \ref{fig:rho_mag}, $\rho(\lambda)$ does not change very much as we increase $h_r$ from zero up to $h_{AT}=1$. For the external field larger than the AT field, however, Fig.~\ref{fig:rho_mag}\ clearly shows that the eigenvalue spectrum develops a gap. The gap increases with the increasing external field. By directly working on Eq.~(\ref{Gbar}) in the small-$\lambda$ limit, we find that for small eigenvalues \begin{equation} \rho(\lambda)\simeq \frac{1}{\pi(1-1/m)}\frac{1}{\sqrt{s}}\sqrt{\lambda-\lambda_0}, \end{equation} where $s=(1-m^{-1})\langle 1/h^3 \rangle$ and $\lambda_0=\lambda^2_R/4s$ with $\lambda_R$ defined in Eq.~(\ref{lambda_R}). Our numerical solution of the equations for $G(\lambda)$ confirms that there is no gap below $h_{AT}$ which is consistent with the previous observation that $\lambda_R$ vanishes there. However, the integral by which $s$ is defined diverges for $h_r > h_{AT}$ when $m <3$ and we no longer see a square root singularity at the band-edge. In the case of $m =3$ our numerical solution shown in Fig. \ref{fig:rho_mag} suggests instead of the square root dependence there is a roughly linear dependence as $\lambda$ approaches the numerically determined band-edge $\lambda_0$, but unfortunately we have not been able to derive its form analytically. Fig. \ref{fig:rho} shows that away from $\lambda_0$ the density of states is rather as if it had the square root form. As $h_r \to h_{AT}$ this square root form works all the way to zero. \subsection{Quenched Average} \label{sec:quenched} In this subsection, we attempt to evaluate the quenched complexity $g(\varepsilon)=N^{-1}\langle\ln N_S(\varepsilon)\rangle$. The calculations are quite complicated and some of the details are sketched in the Appendix. In order to calculate $\langle \ln N_S(\varepsilon) \rangle$, we consider an average of the replicated quantity $\langle [N_S(\varepsilon)]^n \rangle_{J,h^{\rm ex}}$. We then have an expression similar to Eq.~(\ref{NS_ann}), where the integrals are now over replicated variables, $u^\eta$, $v^\eta$, $\bar{\chi}^\eta$ and $\lambda^\eta$ with the replica indices $\eta,\mu=1,\cdots,n$. In addition to these, the expression also involves the integrals over the variables carrying off-diagonal replica indices, which are denoted by $A_{\eta\nu}$, $A^*_{\eta\nu}$, $B_{\eta\nu}$ and $B^*_{\eta\nu}$ with $\eta<\nu$. In the absence of external field, it can be shown \cite{bm1981} that $A_{\eta\nu}=A^*_{\eta\nu} =B_{\eta\nu}=B^*_{\eta\nu}=0$ is always a solution to the saddle point equations. It is shown to be stable for $\varepsilon> E_c$ for the $E_c$, for which the quenched average coincides with the annealed one. For $h_r\neq 0$, however, we find that this is no longer the case. $A_{\eta\nu}=A^*_{\eta\nu}=B_{\eta\nu}=B^*_{\eta\nu}=0$ is not a solution to saddle point equations. The saddle point solutions involve nonvanishing off-diagonal variables in replica indices. We find that in general the saddle point equations are too complicated to allow explicit solutions. (See the Appendix for details.) The quenched average is different from the annealed one for a finite external field when $h_r< h_{AT}$. When $h_r> h_{AT}$ the annealed and quenched averages are identical in every way for the SK model, which has vanishing complexity in this region. We doubt whether the same statement is true for any model such as the Viana-Bray model which has non-zero complexity for $h_r > h_{AT}$. We also do not know for sure whether our replica symmetric solution for $A_{\eta \nu}$ etc. is stable. It is possible that even at $u=0$ there is a need to go to full replica symmetry breaking. Unfortunately algebraic complexities have prevented us from even finding a solution of the replica symmetric equations, so determining their stability looks very challenging. However, the results of the numerical work reported on the form of $P(q)$ in Sec. \ref{sec:metastability} for the Viana-Bray model in a field suggests that the states reached in the quench have replica symmetry. We look for the saddle points in the replica symmetric form, \begin{align} &A_{\eta\nu}=A,~~A^*_{\eta\nu}=A^*,~~B_{\eta\nu}=B^*_{\eta\nu}=B, \nonumber \\ &u^\eta=u,~~v^\eta=v,~~\bar{\chi}^\eta=\bar{\chi},~~\lambda^\eta=\lambda. \end{align} After a lengthy calculation (see Appendix), we arrive at the expression for the quenched complexity as follows. \begin{align} g(\varepsilon)=& m \Big\{ -\frac{\Delta^2}{2} -\varepsilon u - \frac{A}{2m} +\frac{1}{2} (AA^*+B^2)\Big\} \label{geps} \\ +& \int \frac{d^m \bm{w}}{(2\pi)^{m/2}} \int \frac{d^m \bm{y}}{(2\pi)^{m/2}} \int \frac{d^m \bm{z} d^m\bm{z}^*}{(2\pi)^m}\; \nonumber \\ &\times \exp[-\frac 1 2 \sum^m_\alpha (w^2_\alpha+y^2_\alpha+z_\alpha z^*_\alpha)] \; \ln K(\bm{w},\bm{y},\bm{z},\bm{z}^*), \nonumber \end{align} where \begin{align} K=& \int d^m\bm{h} \int^{i\infty}_{-i\infty} \frac{d^m \bm{x}}{2\pi i} \;\exp\Bigg[ \frac{1-mA^*}{2m}\bm{x}^2 \nonumber \\ &+ (h-\Delta-B)\bm{x}\cdot\hat{\bm{h}} -(m-1)\frac{\Delta + u/2}{h} -\frac{m}{2}uh \nonumber \\ &+ \sqrt{A^*+\frac{h^2_r}{m}} \;\bm{w}\cdot\bm{x} +\sqrt{A+\frac{mh^2_r}{4}u^2 } \; \bm{y}\cdot \hat{\bm{h}} \nonumber \\ &+ \sqrt{\frac{1}{2}(B+\frac{h^2_r}{2}u) } \left(\bm{z}\cdot\bm{x}+\bm{z}^* \cdot \hat{\bm{h}}\right) \Bigg]. \end{align} All the parameters, $\Delta$, $A$, $A^*$, $B$ and $u$ are to be determined in a variational way. We found, however, that it is very difficult to solve the saddle point equations and obtain the quenched complexity, even numerically. For the total number of metastable states, $u=0$, we can find a simple solution to saddle point equations at $\Delta=A=B=0$ and $A^*=1/m$. In this case, $K=1$ and the complexity $g$ vanishes. This solution must describe the case where $h_r>h_{AT}$ and it is identical to the annealed average. For the external field $h_r$ just below $h_{AT}$, $\Delta$, $A$, $B$ and $C\equiv 1/m-A^*$ are expected to be very small, and we may expand the integrals in Eq.~(\ref{geps}) in these variables. We find after a very lengthy calculation that \begin{equation} g\simeq \frac{m}{1+h^2_r} (h^2_{AT}-h^2_r)\Big[ \frac{\Delta^2}{2} + \frac{AC}{2} - \frac{B^2}{2} \Big] . \end{equation} Note that from Eq.~(\ref{saddle}), we expect $B$ is pure imaginary. In order to determine how these variables behave near $h_{AT}$, we need higher order terms. Unfortunately, the complicated nature of these equations, however, has prevented us from going beyond the quadratic orders. It seems natural to expect that the $\Delta$ sector is decoupled from the off-diagonal variables, and so will have the same $\Delta^3\ln\Delta$ behavior as in Eq.~(\ref{gA0}). But the effort to obtain a full solution is so large that we abandoned further work on it. \section{Hessian studies} \label{sec:hessian} In this section we write down the Hessian for the $m=3$ Heisenberg spin glass in a form which is convenient for numerical work. The Hessian is of interest as it describes the nature of the energy of the spin glass in the vicinity of the minima. It is also closely related to the matrices needed to describe the spin waves in the system \cite{bm1981}. We follow the approach used in the paper of Beton and Moore~\cite{beton1984electron} to find the elements of the Hessian matrix $T$ corresponding to directions transverse to each spin subject to the above metastability condition. We first define the site-dependent two-dimensional orthogonal unit vectors $\hat{e}_{x}(i)$ and $\hat{e}_{y}(i)$ such that \begin{align} \hat{e}_{m}(i)\cdot\mathbf{S}_{i}^{0} &= 0\\ \hat{e}_{m}(i)\cdot\hat{e}_{n}(i) &= \delta^{mn}, \end{align} where $m,n = x,y$ denotes the directions perpendicular to the spin at the $i$th site, which is deemed in the $\lq \lq z"$ direction. The linear combinations $\hat{e}_{i}^{\pm}=\frac{1}{\sqrt{2}}(\hat{e}_{x}(i)\pm i\hat{e}_{y}(i))$ turn out to be particularly useful. Expanding $\mathbf{S}_{i}$ about $\mathbf{S}_{i}^{0}$, subject to the condition that the length of the spins remains unchanged yields, upto second-order: \begin{align} \mathbf{S}_{i} = \mathbf{S}_{i}^{0}+\Gamma_{i}^{x} \hat{e}_{x}(i)+\Gamma_{i}^{y} \hat{e}_{y}(i)-\frac{1}{2}[(\Gamma_{i}^{x})^{2}+(\Gamma_{i}^{y})^{2}]\mathbf{S}_{i}^{0}. \end{align} Equivalently, \begin{align} \mathbf{S}_{i} = \mathbf{S}_{i}^{0}+Z_{i}^{-}\hat{e}_{i}^{+}+Z_{i}^{+}\hat{e}_{i}^{-}-Z_{i}^{-}Z_{i}^{+}\mathbf{S}_{i}^{0}, \end{align} where $Z_{i}^{\pm} = \frac{1}{\sqrt{2}}(\Gamma_{i}^{x}\pm i\Gamma_{i}^{y})$, and $(Z_{i}^{+})^{*}=Z_{i}^{-}$. Defining the $2N$-dimensional vector \begin{align} |Z\rangle = \begin{pmatrix} Z_{i}^{-}\\ Z_{i}^{+} \end{pmatrix}, \end{align} the change in energy per spin component degree of freedom $\frac{\delta E}{3}$ due to a change in spin orientations $|Z\rangle$, is given by: \begin{align} \frac{\delta E}{3} = \frac{1}{2}\langle Z|T|Z\rangle, \end{align} where $T$ is the $2N \times 2N$ Hessian matrix given by \begin{align*} \begin{aligned} T = \frac{1}{3} \begin{pmatrix} |\mathbf{H}_{i}|\delta_{ij}+A_{ij}^{*} & B_{ij}^{*}\\ B_{ij} & |\mathbf{H}_{i}|\delta_{ij}+A_{ij} \end{pmatrix} \end{aligned}, \end{align*} where the matrix elements are \begin{align*} A_{ij} = A_{ji}^{*} = -3J_{ij}\hat{e}_{i}^{+}\cdot\hat{e}_{j}^{-}\\ B_{ij} = B_{ji}^{*} = -3J_{ij}\hat{e}_{i}^{+}\cdot\hat{e}_{j}^{+}. \end{align*} Converting to spherical coordinates, the matrix elements are \begin{widetext} \begin{align} A_{ij}^{*} &= - \frac{3J_{ij}}{2}[(\cos(\theta_{i})\cos(\theta_{j})+1)\cos(\phi_{i}-\phi_{j})+i(\cos(\theta_{i})+\cos(\theta_{j}))\sin(\phi_{i}-\phi_{j})+\sin(\theta_{i})\sin(\theta_{j})]\nonumber\nonumber\\ B_{ij}^{*} &= - \frac{3J_{ij}}{2}[(\cos(\theta_{i})\cos(\theta_{j})-1)\cos(\phi_{i}-\phi_{j})-i(\cos(\theta_{i})-\cos(\theta_{j}))\sin(\phi_{i}-\phi_{j})+\sin(\theta_{i})\sin(\theta_{j})]\nonumber\\ B_{ij} &= - \frac{3J_{ij}}{2}[(\cos(\theta_{i})\cos(\theta_{j})-1)\cos(\phi_{i}-\phi_{j})+i(\cos(\theta_{i})-\cos(\theta_{j}))\sin(\phi_{i}-\phi_{j})+\sin(\theta_{i})\sin(\theta_{j})]\nonumber\\ A_{ij} &= - \frac{3J_{ij}}{2}[(\cos(\theta_{i})\cos(\theta_{j})+1)\cos(\phi_{i}-\phi_{j})-i(\cos(\theta_{i})+\cos(\theta_{j}))\sin(\phi_{i}-\phi_{j})+\sin(\theta_{i})\sin(\theta_{j})]\nonumber\\ \end{align}. \end{widetext} \begin{figure} \includegraphics[width=\columnwidth]{fig_SKchi.eps} \caption{(Color online) The inverse of the spin glass susceptibility $\chi_{SG}^{-1}$ versus $h_r^2$ for a range of system sizes of the Heisenberg SK model. The analytic curve is the result of Eq. (\ref{chisgexact}). For $h_r \le 1$, one expects that $\chi_{SG}^{-1} =0$, but finite size effects make it non-zero.} \label{fig1SK} \end{figure} \begin{figure} \includegraphics[width=\columnwidth]{fig_VBchi.eps} \caption{(Color online) The inverse of the spin glass susceptibility $\chi_{SG}^{-1}$ versus $h_r^2$ for a range of system sizes for the VB model with $z = 6$.} \label{fig1VB} \end{figure} In the next subsection we use the Hessian to numerically calculate the spin glass susceptibility of both the SK model and VB model in a range of random fields for the Heisenberg spin glass. \subsection{Spin Glass Susceptibility} \label{sec:spinglasssusceptibility} The spin glass susceptibility for the metastable states can be computed from the inverse of the Hessian matrix using the relation~\cite{bm1981} \begin{equation} \chi_{SG} = \frac{1}{N}\Tr{(T^{-1})^{2}}. \end{equation} For the SK model and $h_r> h_{AT}=1$, we have calculated $\chi_{SG}$ analytically and Fig. \ref{fig1SK} shows that our numerical work is approaching the analytical solution, but finite size effects are still very considerable at the sizes we can study. Notice that for the SK model there is (weak) numerical evidence that $\chi_{SG}$ diverges below the AT field. For the VB model, the plot of $\chi_{SG}$ in Fig. \ref{fig1VB} obtained from our metastable states which lie above the true ground state energy provides no evidence that an AT field has much relevance for these states. \begin{figure} \includegraphics[width=\columnwidth]{fig_dos.eps} \caption{(Color online) The averaged density of states of the Hessian matrix of the metastable states obtained after a quench to $T=0$ starting from spins with random orientations i.e. $T= \infty$ for the SK model ($\sigma =0, z =N-1$ of the diluted model). Data shown here for the special case of $h_{r} = 0.8$, for which the system is in the spin glass phase, just below $h_{AT} = 1$. The analytical curve is that calculated from Eqs. (\ref{eqn:rho}) and (\ref{Gbar})) for metastable states at the top of the band within the annealed approximation. The numerical results are strikingly similar to the analytical results, despite the fact that they refer to Hessians for quite different situations! } \label{fig2} \end{figure} \subsection{Density of States} \label{sec:density} The density of states of the eigenvalues of the Hessian matrix has been obtained numerically for the minima obtained in a quench from infinite temperature to zero temperature. The results have remarkable agreement with the analytical calculation performed on the Heisenberg SK model as shown in Fig.~\ref{fig2}. The analytical calculation itself is not for the same set of metastable states. It applies to the states corresponding to $u=0$ (i.e. those with the largest complexity within the annealed approximation). In Fig.~\ref{fig2}, data are shown for $h_r =0.8 h_{AT}$, where no gap is present. The agreement between the analytical curve which is obtained for the thermodynamic limit, and the data for a $N=1024$ size system from numerical simulations, is striking. Notice that the $\sqrt{\lambda}$ form predicted from the annealed study (see Sec. \ref{SKannealed}) seems to hold as $\lambda \to 0$, despite there being no Goldstone theorem in the presence of a random field to ensure the existence of massless modes. We have also studied the density of states and quantities like the inverse participation ratios for the quenched state minima in models like the VB model and the one-dimensional long range models. Basically the results seem similar to those reported in Refs. \onlinecite{baity2015soft} for the three dimensional Heisenberg spin glass model in a random field. But it requires large systems to get accurate results for the density of states at small values of $\lambda$ and we are leaving these issues to a future publication. \section{Conclusions} \label{sec:conclusions} We believe that the most interesting feature which has turned up in our studies is the discovery for the SK model in zero external fields that the quenched states reached for $m=2$ and $m =3$ are quite close to the critical energies $E_c$ at which the overlap of the states would acquire features associated with a $P(q)$ with broken replica symmetry. In the Ising SK model the local fields after the quench are so different from those used in the analytical calculations of $E_c$ that the connection of the quenched state to being just at the edge of the states with broken replica symmetry was not recognized. Thus in systems with marginal stability this means that features normally associated with continuous phase transitions, in particular diverging length scales, could be studied as in Ref. \onlinecite{bm:81a}. We have noticed too that the energy of the states reached from the quench have zero overlap with each other. This behavior was predicted for the Ising case in Ref.~\onlinecite{newman:99} by Newman and Stein who proved that after a quench from infinite temperature for Ising systems the states which are reached have a characteristic energy and a trivial $P(q)$. It would be good to extend their theorems to vector spin systems both in zero field and also in the presence of random fields. In Sec. \ref{sec:SKanalytic} we attempted to extend the old calculations of Bray and Moore \cite{bm1981} which were for zero random field to non-zero random fields. For fields $h_r >h_{AT}$ where the complexity is zero, the annealed approximation is exact and we were able to obtain the exact form for the behaviour of the density of states of the Hessian matrix. There was found to be a gap in the spectrum which went to zero in the limit $h_r \to h_{AT}$. When $h_r< h_{AT}$ one needs to study the quenched average in order to get results pertinent to typical minima, but we were not able to overcome the algebraic complexities (see Sec. \ref{sec:quenched} and the Appendix), although the only difficulty is that of solving the equations which we have obtained. If that could be done then one could investigate the limit of stability of the replica symmetric solution and determine $E_c(h_r)$. Then one could investigate whether a quench in a field $h_r$ takes one to the limit of stability towards full replica symmetry breaking i.e. $E_c(h_r)$, just as we found for $h_r=0$. The annealed approximation is tractable but alas it is only an approximation. Nevertheless the studies in Sec. \ref{sec:density} shows that it gives good results for the density of states of the Hessian for the SK model for $h_r < h_{AT}$. The VB model is a mean-field model and one could hope that it too could be understood analytically, but we do not know how this might be achieved. Our numerical studies of the density of states of its Hessian indicates that this is very different from that of the SK model. This is probably because for the SK model all the eigenstates are extended, whereas for the VB model, eigenvectors can also be localized. In fact our results for the VB model are quite similar to those reported for the three dimensional Heisenberg spin glass in a field \cite{baity2015soft}. There seems to be localized states lying in the gap region, all the way down to $\lambda=0$. But understanding the VB model analytically is very challenging. \acknowledgements We should like to thank the authors of Ref.~\onlinecite{lupo:16} for an advance copy of their paper and helpful discussions. One of us (MAM) would like to thank Dan Stein for discussions on quenches in Ising systems. AS acknowledges support from the DST-INSPIRE Faculty Award [DST/INSPIRE/04/2014/002461]. JY was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (2014R1A1A2053362). \begin{widetext}
3935b10ebd2ab44f820e6d8233731be14a55d558
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} The production of hadrons and jets at a future Electron Ion Collider (EIC) will play a central role in understanding the structure of the protons and nuclei which comprise the visible matter in the universe. Measurements of inclusive jet and hadron production with transversely polarized protons probe novel phenomena within the proton such as the Sivers function~\cite{Kang:2011jw}, and address fundamental questions concerning the validity of QCD factorization. Event shapes in jet production can give insight into the nuclear medium and its effect on particle propagation~\cite{Kang:2012zr}. The precision study of these processes at a future EIC will provide a much sharper image of proton and nucleus structure than is currently available. Progress is needed on both the experimental and theoretical fronts to achieve this goal. Currently, much of our knowledge of proton spin phenomena, such as the global fit to helicity-dependent structure functions~\cite{deFlorian:2008mr}, comes from comparison to predictions at the next-to-leading order (NLO) in the strong coupling constant. Theoretical predictions at the NLO level for jet and hadron production in DIS suffer from large theoretical uncertainties from uncalculated higher-order QCD corrections~\cite{Hinderer:2015hra} that will eventually hinder the precision determination of proton structure. In some cases even NLO is unknown, and an LO analysis fails to describe the available data~\cite{Gamberg:2014eia}. Given the high luminosity and expected precision possible with an EIC, it is desirable to extend the theoretical precision beyond what is currently available. For many observables, a prediction to next-to-next-to-leading order (NNLO) in the perturbative QCD expansion will ultimately be needed. An important step toward improving the achievable precision for jet production in electron-nucleon collisions was taken in Ref.~\cite{Hinderer:2015hra}, where the full NLO ${\cal O}(\alpha^2\alpha_s)$ corrections to unpolarized $lN \to jX$ and $lN \to hX$ scattering were obtained. Focusing on single-inclusive jet production for this discussion, it was pointed out that two distinct processes contribute: the deep-inelastic scattering (DIS) process $lN \to ljX$, where the final-state lepton is resolved, and $\gamma N \to jX$, where the initial photon is almost on-shell and the final-state lepton is emitted collinear to the initial-state beam direction. Both processes were found to contribute for expected EIC parameters, and the shift of the leading-order prediction was found to be both large and dependent on the final-state jet kinematics. Our goal in this manuscript is to present the full ${\cal O}(\alpha^2\alpha_s^2)$ NNLO contributions to single-inclusive jet production in electron-nucleon collisions, including all the relevant partonic processes discussed above. Achieving NNLO precision for jet and hadron production is a formidable task. The relevant Feynman diagrams which give rise to the NNLO corrections consist of two-loop virtual corrections, one-loop real-emission diagrams, and double-real emission contributions. Since these three pieces are separately infrared divergent, some way of regularizing and canceling these divergences must be found. However, theoretical techniques for achieving this cancellation in the presence of final-state jets have seen great recent progress. The introduction of the $N$-jettiness subtraction scheme for higher order QCD calculations~\cite{Boughezal:2015dva,Gaunt:2015pea} has lead to the first complete NNLO descriptions of jet production processes in hadronic collisions. During the past year several NNLO predictions for processes with final-state jets have become available due to this theoretical advance~\cite{Boughezal:2015dva,Boughezal:2015aha,Boughezal:2015ded,Boughezal:2016dtm,Boughezal:2016isb,Boughezal:2016yfp}. In some cases the NNLO corrections were critical in explaining the observed data~\cite{Boughezal:2016yfp}. We discuss here the application of the $N$-jettiness subtraction scheme to inclusive jet production in electron-proton collisions. Our result includes both the DIS and photon-initiated contributions, and allows arbitrary selection cuts to be imposed on the final state. Upon integration of the DIS terms over the final-state hadronic phase space we compare our result against the known NNLO prediction for the inclusive structure function, and we find complete agreement. We present phenomenological results for proposed EIC parameters. We find that all partonic channels, including new ones that first appear at this order, contribute in a non-trivial way to give the complete NNLO correction. We note that the NNLO corrections to similar processes, massive charm-quark production in deeply inelastic scattering and dijet production, were recently obtained~\cite{Berger:2016inr,Currie:2016ytq}. \section{Lower-order results}\label{sec:low} We begin by discussing our notation for the hadronic and partonic cross sections, and outlining the expressions for the LO and NLO cross sections. We will express the hadronic cross section in the following notation: \begin{equation} {\rm d}\sigma = {\rm d}\sigma_{\text{LO}}+{\rm d}\sigma_{\rm NLO}+{\rm d}\sigma_{\rm NNLO}+\ldots \,, \end{equation} where the ellipsis denotes neglected higher-order terms. The LO subscript refers to the ${\cal O}(\alpha^2)$ term, the NLO subscript denotes the ${\cal O}(\alpha^2\alpha_s)$ correction, while the NNLO subscript indicates the ${\cal O}(\alpha^2\alpha_s^2)$ contribution. For the partonic cross sections, we introduce superscripts that denote the powers of both $\alpha$ and $\alpha_s$ that appear. For example, the leading quark-lepton scattering process is expanded as \begin{equation} {\rm d}\hat{\sigma}_{ql}= {\rm d}\hat{\sigma}_{ql}^{(2,0)}+{\rm d}\hat{\sigma}_{ql}^{(2,1)}+{\rm d}\hat{\sigma}_{ql}^{(2,2)}+\ldots \,. \end{equation} Here, the ${\rm d}\hat{\sigma}_{ql}^{(2,0)}$ denotes the ${\cal O}(\alpha^2)$ correction, while ${\rm d}\hat{\sigma}_{ql}^{(2,1)}$ indicates the ${\cal O}(\alpha^2\alpha_s)$ term. The leading-order hadronic cross section can be written as a convolution of parton distribution functions with a partonic cross section, \begin{eqnarray} \label{eq:sigLO} {\rm d}\sigma_{\text{LO}} &=& \int \frac{{\rm d} \xi_1}{\xi_1} \frac{{\rm d} \xi_2}{\xi_2} \sum_q \left[ f_{q/H}(\xi_1) f_{l/l}(\xi_2) {\rm d}\hat{\sigma}_{ql}^{(2,0)} \right. \\ \nonumber &+& \left. f_{\bar{q}/H}(\xi_1) f_{l/l}(\xi_2) {\rm d}\hat{\sigma}_{\bar{q}l}^{(2,0)}\right]. \end{eqnarray} Here, $f_{q/H}(\xi_1)$ is the usual parton distribution function that describes the distributions of a quark $q$ in the hadron $H$ carrying a fraction $\xi_1$ of the hadron momentum. $f_{l/l}(\xi_2)$ is the distribution for finding a lepton with momentum fraction $\xi_2$ inside the original lepton. At leading order this is just $f_{l/l}(\xi_2)=\delta(1-\xi_2)$, but it is modified at higher orders in the electromagnetic coupling by photon emission. $d\hat{\sigma}_{ql}^{(2,0)}$ is the differential partonic cross section. At leading order only the partonic channel $q(p_1)+l(p_2) \to q(p_3)+l(p_4)$ and the same process with anti-quarks instead contribute. The relevant Feynman diagram is shown in Fig.~\ref{fig:LOdiag}. It is straightforward to obtain these terms. \begin{figure}[h] \centering \includegraphics[width=1.5in]{LO.pdf}% \caption{Feynman diagram for the leading-order process $q(p_1)+l(p_2) \to q_(p_3)+l(p_4)$. We have colored the photon line red, the lepton lines green and the quark lines black.} \label{fig:LOdiag} \end{figure} At the next-to-leading order level several new contributions first occur. The quark-lepton scattering channel that appears at LO receives both virtual and real-emission corrections that are separately infrared divergent. We use the antennae subtraction method~\cite{Kosower:1997zr} to regularize and cancel these divergences. Initial-state collinear divergences are handled as usual by absorbing them into the PDFs via mass factorization. A gluon-lepton scattering channel also contributes at this order. The collinear divergences that appear in these contributions are removed by mass factorization. Example Feynman diagrams for these processes are shown in Fig.~\ref{fig:NLOlepdiag}. \begin{figure}[h] \centering \includegraphics[width=3.0in]{NLOlep.pdf}% \caption{Representative Feynman diagrams contributing to the following perturbative QCD corrections at NLO: virtual corrections to the $q(p_1)+l(p_2) \to q(p_3)+l(p_4)$ process (left); real emission correction $q(p_1)+l(p_2) \to q(p_3)+l(p_4)+g(p_5)$ (middle); the process $g(p_1)+l(p_2) \to q_(p_3) +l(p_4)+\bar{q}(p_5)$ (right). We have colored the photon line red, the lepton lines green, the gluon lines blue and the quark lines black.} \label{fig:NLOlepdiag} \end{figure} The processes discussed above exhaust the possible NLO contributions when the final-state lepton is observed. However, for single-inclusive jet production a kinematic configuration is allowed where the $t$-channel photon is nearly on-shell, and the final-state lepton travels down the beam pipe. The transverse momentum of the leading jet is balanced by the additional jet present in these diagrams. This kinematic configuration leads to a QED collinear divergence for vanishing lepton mass, since the photon can become exactly on-shell in this limit. While it is regulated by the lepton mass, it is more convenient to obtain these corrections by introducing a photon distribution function in analogy with the usual parton distribution function. The collinear divergences that appear in the matrix elements computed with vanishing lepton mass can be absorbed into this distribution function, which can be calculated in perturbation theory. Representative diagrams for the photon-initiated processes are shown in Fig.~\ref{fig:NLOphotdiag}. \begin{figure}[h] \centering \includegraphics[width=3.0in]{NLOphot.pdf}% \caption{Representative Feynman diagrams contributing to the $q(p_1)+\gamma(p_2) \to q(p_3)+g(p_4)$ (left) and $g(p_1)+\gamma(p_2) \to q(p_3)+\bar{q}(p_4)$ scattering processes.} \label{fig:NLOphotdiag} \end{figure} The full expression for the NLO hadronic cross section then takes the form \begin{eqnarray}\label{eq:hxsecnlo} {\rm d}\sigma_{\rm NLO}&=&\int\frac{{\rm d}\xi_1}{\xi_1}\frac{{\rm d}\xi_2}{\xi_2}\bigg\{f_{g/H}^1f_{l/l}^2 {\rm d}\hat{\sigma}_{gl}^{(2,1)} + f_{g/H}^1f_{\gamma/l}^2{\rm d}\hat{\sigma}_{g\gamma}^{(1,1)} \phantom{\sum_q}\nonumber \\ && +\sum_q\bigg[f_{q/H}^1f_{l/l}^2{\rm d}\hat{\sigma}_{ql}^{(2,1)}+f_{\bar{q}/H}^1f_{l/l}^2{\rm d}\hat{\sigma}_{\bar{q}l}^{(2,1)} \nonumber \\ &&+f_{q/H}^1f_{\gamma/l}^2{\rm d}\hat{\sigma}_{q\gamma}^{(1,1)}+f_{\bar{q}/H}^1f_{\gamma/l}^2{\rm d}\hat{\sigma}_{\bar{q}\gamma}^{(2,1)}\bigg]\bigg\}, \end{eqnarray} where we have abbreviated $f_{i/j}^k=f_{i/j}(\xi_k)$. The contributions ${\rm d}\hat{\sigma}_{gl}^{(2,1)}$, ${\rm d}\hat{\sigma}_{ql}^{(2,1)}$ and ${\rm d}\hat{\sigma}_{\bar{q}l}^{(2,1)}$ denote the usual DIS partonic channels computed to NLO in QCD with zero lepton mass. The terms ${\rm d}\hat{\sigma}_{\bar{q}\gamma,\rm LO}$ and ${\rm d}\hat{\sigma}_{g\gamma,\rm LO}$ denote the new contributions arising when $Q^2 \approx 0$ and the virtual photon is nearly on-shell. The photon distribution can be expressed as \begin{equation} f_{\gamma/l}(\xi) = \frac{\alpha}{2\pi} P_{\gamma l}(\xi) \left[ \text{ln}\left(\frac{\mu^2}{\xi^2 m_l^2} \right)-1\right] +{\cal O}(\alpha^2), \end{equation} where the splitting function is given by \begin{equation} \label{eq:gammalsplit} P_{\gamma l}(\xi) = \frac{1+(1-\xi)^2}{\xi}. \end{equation} This is the well-known Weizs\"{a}cker-Williams (WW) distribution for the photon inside of a lepton~\cite{WW}. The appearance of the renormalization scale $\mu$ indicates that an $\overline{\text{MS}}$ subtraction of the QED collinear divergence is used in the calculation of the $gl$ and $ql$ scattering channels, and consequently in the derivation of the photon distribution function. \section{Calculation of the NNLO result} \label{sec:nnlo} The calculation of the full ${\cal O}(\alpha^2\alpha_s^2)$ corrections involves several distinct contributions. The quark-lepton and gluon-lepton scattering channels receive two-loop double virtual corrections, one-loop corrections to single real-emission diagrams, and double-real emission corrections. These contributions necessitate the use of a full-fledged NNLO subtraction scheme. We use the recently-developed $N$-jettiness subtraction scheme~\cite{Boughezal:2015dva,Gaunt:2015pea}. Its application to this process is discussed here in detail. In addition, the photon-initiated scattering channels receive virtual and single real-emission corrections. The calculation of these terms follows the standard application of the antennae subtraction scheme at NLO. There is in addition a new effect that appears at the NNLO level. The initial-state lepton can emit a photon which splits into a $q\bar{q}$ pair, all of which are collinear to the initial lepton direction. In the limit of vanishing fermion masses there is a collinear singularity associated with this contribution. This divergence appears in the quark-lepton, gluon-lepton, and photon-initiated scattering channels. It can be absorbed into a distribution function that describes the quark distribution inside a lepton. Treating the collinear singularity in this way leads to new scattering channels that first appear at NNLO: $q\bar{q} \to q\bar{q}$, $q\bar{q} \to q^{\prime}\bar{q}^{\prime}$, $q\bar{q} \to gg$, $qq^{\prime} \to qq^{\prime}$, and $qg \to qg$. For our numerical predictions for these channels we need the quark distribution in a lepton. We obtain this by solving the DGLAP equation, which to the order we are working can be written as \begin{eqnarray}\label{eq.dglap} &&\mu^2\frac{\partial f_{q/l}}{\partial \mu^2}(\xi,\mu^2)=e_q^2\frac{\alpha}{2\pi}\int_{\xi}^1\frac{{\rm d} z}{z}P_{q\gamma}^{(0)}(z)\,f_{\gamma/l}\left(\frac{\xi}{z},\mu^2\right) \nonumber \\ &&+ e_q^2\left(\frac{\alpha}{2\pi}\right)^2\int_{\xi}^1\frac{{\rm d} z}{z}P_{ql}^{(1)}(z)\,f_{l/l}\left(\frac{\xi}{z},\mu^2\right), \end{eqnarray} where the two needed splitting kernels are \begin{eqnarray} P_{q\gamma}^{(0)}(x)&=&x^2+(1-x)^2, \nonumber \\ P_{ql}^{(1)}(x) &=& -2+\frac{20}{9x}+6x-\frac{56x^2}{9}\nonumber \\ &&\hspace{-0.8cm}+\left(1+5x+\frac{8x^2}{3}\right) \log(x)-(1+x) \log^2(x). \end{eqnarray} This expression for the NLO splitting kernel can be obtained from upon replacement of the appropriate QCD couplings with electromagnetic ones. To derive the full quark-in-lepton distribution we use as an initial condition $ f_{q/l}(\xi,m_l^2)=0$. Solving Eq.~(\ref{eq.dglap}) with this initial condition gives \begin{eqnarray} \label{eq:qinl} &&f_{q/l}(\xi,\mu^2)=e_q^2\left(\frac{\alpha}{2\pi}\right)^2\bigg\{\bigg[\frac{1}{2}+\frac{2}{3\xi}-\frac{\xi}{2}-\frac{2\xi^2}{3} \nonumber \\ && +(1+\xi)\log\xi \bigg] \log^2\left(\frac{\mu^2}{m_l^2}\right) + \bigg[-3-\frac{2}{\xi}+7\xi-2\xi^2+\nonumber \\ &&\bigg( -5 -\frac{8}{3\xi}+\xi+\frac{8\xi^2}{3}\bigg)\log(\xi)-3(1+x)\log^2(\xi)\bigg] \nonumber \\ && \times \log\left(\frac{\mu^2}{m_l^2}\right). \end{eqnarray} \begin{figure}[h] \centering \includegraphics[width=3.0in]{NNLO2jet.pdf}% \caption{Representative Feynman diagrams contributing to the $q(p_1)+\bar{q}(p_2) \to g(p_3)+g(p_4)$ (left), $q(p_1)+q^{\prime}(p_2) \to q(p_3)+q^{\prime}(p_4)$ (middle), and $q(p_1)+g(p_2) \to q(p_3)+g(p_4)$ (right) scattering processes.} \label{fig:NNLO2jet} \end{figure} With this distribution function it is straightforward to obtain numerical predictions for these partonic channels. Representative Feynman diagrams for these processes are shown in Fig.~\ref{fig:NNLO2jet}. We can now write down the full result for the ${\cal O}(\alpha^2 \alpha_s^2)$ correction to the cross section: \begin{eqnarray} \label{eq:hxsecnnlo} {\rm d}\sigma_{\rm NNLO}&=&\int\frac{{\rm d}\xi_1 {\rm d}\xi_2}{\xi_1 \xi_2}\bigg\{f_{g/H}^1f_{l/l}^2\,{\rm d}\hat{\sigma}_{gl}^{(2,2)} +f_{g/H}^1f_{\gamma/l}^2\,{\rm d}\hat{\sigma}_{g\gamma}^{(1,2)} \phantom{\sum_q}\nonumber \\ &+&\sum_q\bigg[f_{g/H}^1f_{q/l}^2\,{\rm d}\hat{\sigma}_{gq}^{(0,2)}+f_{g/H}^1f_{\bar{q}/l}^2\,{\rm d}\hat{\sigma}_{g\bar{q}}^{(0,2)}\bigg] \nonumber\\ &+&\sum_q\bigg[f_{q/H}^1f_{l/l}^2\,{\rm d}\hat{\sigma}_{ql}^{(2,2)}+f_{\bar{q}/H}^1f_{l/l}^2\,{\rm d}\hat{\sigma}_{\bar{q}l}^{(2,2)} \nonumber \\ &+&f_{q/H}^1f_{\gamma/l}^2\,{\rm d}\hat{\sigma}_{q\gamma}^{(1,2)}+f_{\bar{q}/H}^1f_{\gamma/l}^2\,{\rm d}\hat{\sigma}_{\bar{q}\gamma}^{(1,2)} \phantom{\sum_q\bigg[} \nonumber \\ &+&f_{q/H}^1f_{\bar{q}/l}^2\,{\rm d}\hat{\sigma}_{q\bar{q}}^{(0,2)}+f_{\bar{q}/H}^1f_{q/l}^2\,{\rm d}\hat{\sigma}_{\bar{q}q}^{(0,2)} \bigg] \phantom{\sum_q} \nonumber \\ &+&\sum_{q,q'}\bigg[f_{q/H}^1f_{q'/l}^2\,{\rm d}\hat{\sigma}_{qq'}^{(0,2)}+f_{\bar{q}/H}^1f_{\bar{q}'/l}^2\,{\rm d}\hat{\sigma}_{\bar{q}\bar{q}'}^{(0,2)} \nonumber \\ &+&f_{q/H}^1f_{\bar{q}'/l}^2\,{\rm d}\hat{\sigma}_{q\bar{q}'}^{(0,2)}+f_{\bar{q}/H}^1f_{q'/l}^2\,{\rm d}\hat{\sigma}_{\bar{q}q'}^{(0,2)}\bigg]\bigg\}. \phantom{\sum_q} \end{eqnarray} The most difficult contribution is the quark-lepton scattering channel. It receives contributions from two-loop virtual corrections (double-virtual), one-loop corrections to single real emission terms (real-virtual), and double-real emission corrections. Sample Feynman diagrams for these corrections are shown in Fig.~\ref{fig:NNLOql}. These are separately infrared divergent, and require a full NNLO subtraction scheme to combine. We apply the $N$-jettiness subtraction scheme~\cite{Boughezal:2015dva,Gaunt:2015pea}. The starting point of this method is the $N$-jettiness event shape variable~\cite{Stewart:2010tn}, defined in the one-jet case of current interest as \begin{equation} \label{eq:taudef} {\cal T}_1 = \frac{2}{Q^2} \sum_i \text{min}\left\{p_B \cdot q_i, p_J \cdot q_i \right\}, \end{equation} with $Q^2 = -(p_2-p_4)^2$. Here, $p_B$ and $p_J$ are light-like four-vectors along the initial-state hadronic beam and final-state jet directions, respectively.\footnote{This choice of ${\cal T}_1$ corresponds to $\tau_1^a$ in Ref.~\cite{Kang:2013nha}. We note that this definition is dimensionless, unlike the choice in previous applications of $N$-jettiness subtraction~\cite{Boughezal:2015dva}.} The $q_i$ denote the four-momenta of all final-state partons. Values of ${\cal T}_1$ near zero indicate a final state containing a single narrow energy deposition, while larger values denote a final state containing two or more well-separated energy depositions. A measurement of ${\cal T}_1$ is itself of phenomenological interest. It has been proposed as a probe of nuclear properties in electron-ion collisions~\cite{Kang:2012zr,Kang:2013wca}, and has also been suggested as a way to precisely determine the strong coupling constant~\cite{Kang:2013nha}. \begin{figure}[h] \centering \includegraphics[width=3.0in]{NNLOql.pdf}% \caption{Representative Feynman diagrams contributing to the quark-lepton scattering channel at NNLO.} \label{fig:NNLOql} \end{figure} We will use ${\cal T}_1$ to establish the complete ${\cal O}(\alpha^2\alpha_s^2)$ calculation of the quark-lepton scattering channel. Our ability to do so relies on two key observations, as first discussed in Ref.~\cite{Boughezal:2015dva} for a general $N$-jet process. \begin{itemize} \item Restricting ${\cal T}_1 >0$ removes all doubly-unresolved limits of the quark-lepton matrix elements, for example when the two additional partons that appear in the double-real emission corrections are soft or collinear to the beam or the final-state jet. This can be seen from Eq.~(\ref{eq:taudef}); if ${\cal T}_1 >0$ then at least one $q_i$ must be resolved. Since all doubly-unresolved limits are removed, the ${\cal O}(\alpha^2\alpha_s^2)$ correction in this phase space region can be obtained from an NLO calculation of two-jet production in electron-nucleon collisions. \item When ${\cal T}_1$ is smaller than any other hard scale in the problem, it can be resummed to all orders in perturbation theory~\cite{Stewart:2009yx,Stewart:2010pd}. Expansion of this resummation formula to ${\cal O}(\alpha^2\alpha_s^2)$ gives the NNLO correction to the quark-lepton scattering channel for small ${\cal T}_1$. \end{itemize} The path to a full NNLO calculation is now clear. We partition the phase space for the real-virtual and double-real corrections into regions above and below a cutoff on ${\cal T}_1$, which we label ${\cal T}_1^{cut}$: \begin{equation} \label{eq:partition} \begin{split} {\rm d}\sigma_{ql}^{(2,2)} &= \int {\rm d}\Phi_{\text{VV}} \, |{\cal M}_{\text{VV}}|^2 +\int {\rm d}\Phi_{\text{RV}} \, |{\cal M}_{\text{RV}}|^2 \, \theta_1^{<} \\ &+\int {\rm d}\Phi_{\text{RR}} \, |{\cal M}_{\text{RR}}|^2 \, \theta_1^{<}+\int {\rm d}\Phi_{\text{RV}} \, |{\cal M}_{\text{RV}}|^2 \, \theta_1^{>} \\ &+\int {\rm d}\Phi_{\text{RR}} \, |{\cal M}_{\text{RR}}|^2 \, \theta_1^{>} \\ \equiv & {\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 < {\cal T}_1^{cut})+{\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 > {\cal T}_1^{cut}) \end{split} \end{equation} We have abbreviated $\theta_1^{<} = \theta({\cal T}_1^{cut}-{\cal T}_1)$ and $\theta_1^{>} = \theta({\cal T}_1-{\cal T}_1^{cut})$. The first three terms in this expression all have ${\cal T}_1<{\cal T}_1^{cut}$, and have been collectively denoted as ${\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 < {\cal T}_1^{cut})$. The remaining two terms have ${\cal T}_1>{\cal T}_1^{cut}$, and have been collectively denoted as ${\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 > {\cal T}_1^{cut})$. The double-virtual corrections necessarily have ${\cal T}_1=0$. We obtain ${\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 > {\cal T}_1^{cut})$ from a NLO calculation of two-jet production. This is possible since no genuine double-unresolved limit occurs in this phase-space region. We discuss the calculation of ${\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 < {\cal T}_1^{cut})$ using the all-orders resummation of this process in the following sub-section. We note that only the quark-lepton and gluon-lepton scattering channels have support for ${\cal T}_1=0$. For the other processes there are two final-state jets with non-zero transverse momentum. Such configurations necessarily have ${\cal T}_1 >0$, and therefore only receive contributions from the above-the-cut phase-space region. We only need these contributions to at most NLO in QCD perturbation theory, which can be obtained via standard techniques. We use antennae subtraction. The only non-standard aspect of this NLO calculation is the appearance of triple-collinear QED limits associated with the emission of a photon and a $q\bar{q}$ pair which require the use of integrated antennae found in Refs.~\cite{Daleo:2009yj,Boughezal:2010mc,GehrmannDeRidder:2012ja}. A powerful aspect of the $N$-jettiness subtraction method is its ability to upgrade existing NLO calculations to NNLO precision. Previous applications of $N$-jettiness subtraction~\cite{Boughezal:2015dva,Boughezal:2015aha,Boughezal:2015ded} have used the NLO dipole subtraction technique~\cite{Catani:1996vz} to facilitate the calculation of the above-cut phase-space region. This work demonstrates that it can also be used in conjunction with the NLO antennae subtraction scheme. \subsubsection{Below ${\cal T}_1^{cut}$} An all-orders resummation of the ${\cal T}_1$ event-shape variable in the DIS process for the limit ${\cal T}_1 \ll 1$ was constructed in Refs.~\cite{Kang:2013wca,Kang:2013nha}: \begin{eqnarray} \label{eq:SCETfac} \frac{{\rm d} \sigma}{{\rm d} {\cal T}_1} &=& \int {\rm d} \Phi_2(p_3,p_4;p_1,p_2) \int {\rm d} t_J {\rm d} t_B {\rm d} k_S \nonumber \\ &\times & \delta\left( {\cal T}_1-\frac{t_J}{Q^2}-\frac{t_B}{Q^2} -\frac{k_S}{Q} \right) \nonumber \\ &\times & \sum_q J_q(t_J,\mu) \,S(k_S,\mu) H_q(\Phi_2,\mu) B_q(t_B,x,\mu)+ \ldots \nonumber \\ \end{eqnarray} We have allowed the index $q$ to run over both quarks and anti-quarks. $x$ denotes the usual Bjorken scaling variable for DIS: \begin{equation} x = \frac{Q^2}{2 P\cdot (p_2-p_4)}, \end{equation} where $P$ is the initial-state nucleon four-momentum. $\Phi_2$ denotes the Born phase space, which consists of a quark and a lepton. The derivation of this result relies heavily on the machinery of soft-collinear effective theory (SCET)~\cite{Bauer:2000ew}. A summary of the SCET functions that appear in this expression and what they describe is given below. \begin{itemize} \item $H$ is the hard function that encodes the effect of hard virtual corrections. At leading order in its $\alpha_s$ expansion it reduces to the leading-order partonic cross section. At higher orders it also contains the finite contributions of the pure virtual corrections, renormalized using the $\overline{\text{MS}}$ scheme. It depends only on the Born-level kinematics and on the scale choice. \item $J_q$, the quark jet function, describes the effect of radiation collinear to the final-state jet (which for this process is initiated by a quark at LO). It depends on $t_J$, the contribution of final-state collinear radiation to ${\cal T}_1$. It possesses a perturbative expansion in $\alpha_s$. \item $S$ is the soft function that encodes the contributions of soft radiation. It depends on $k_S$, the contribution of soft radiation to ${\cal T}_1$, and has a perturbative expansion in $\alpha_s$. \item $B$ is the beam function that contains the effects of initial-state collinear radiation. It depends on $t_B$, the contribution of initial-state collinear radiation to ${\cal T}_1$. The beam function is non-perturbative; however, up to corrections suppressed by $\Lambda^2_{\text{QCD}}/t_B$, it can be written as a convolution of perturbative matching coefficients and the usual PDFs: \begin{equation} B_q(t_B,x,\mu) = \sum_i \int_x^1 \frac{{\rm d} \xi}{\xi} {\cal I}_{qi} (t_B,x/\xi,\mu) f_{i/H}(\xi), \end{equation} where we have suppressed the scale dependence of the PDF, and $i$ runs over all partons. \end{itemize} The delta function appearing in Eq.~(\ref{eq:SCETfac}) combines the contribution of each type of radiation to produce the measured value of ${\cal T}_1$. The ellipsis denotes power corrections that are small as long as we restrict ourselves to the phase-space region ${\cal T}_1 \ll 1$. The hard, jet, and soft functions as well as the beam-function matching coefficients all have perturbative expansions in $\alpha_s$ that can be obtained from the literature~\cite{Zijlstra:1992qd,Becher:2006qw,Gaunt:2014xga,Boughezal:2015eha}. Upon expansion to ${\cal O}(\alpha_s^2)$ and integration over the region ${\cal T}_1 < {\cal T}_1^{cut}$, Eq.~(\ref{eq:SCETfac}) will give exactly the cross section ${\rm d}\sigma_{ql}^{(2,2)}({\cal T}_1 < {\cal T}_1^{cut})$ that we require. We can match the beam function onto the PDFs and rewrite the cross section below the cut as \begin{eqnarray} {\rm d}\sigma &=& \int \frac{{\rm d} \xi_1}{\xi_1} \frac{{\rm d} \xi_2}{\xi_2} \sum_{q,i} f_{i/H}(\xi_1) f_{l/l}(\xi_2) \int {\rm d} \Phi_2(p_3,p_4;p_1,p_2) \nonumber \\ && \int_0^{{\cal T}_1^{cut}} {\rm d} {\cal T}_1 \int {\rm d} t_J {\rm d} t_B {\rm d} k_S \delta\left( {\cal T}_1-\frac{t_J}{Q^2}-\frac{t_B}{Q^2} -\frac{k_S}{Q} \right) \nonumber \\ &\times & J_q(t_J,\mu) \,S(k_S,\mu) H_q(\Phi_2,\mu) \, {\cal I}_{qi} (t_B,x/\xi,\mu) \nonumber \\ &\equiv& \sum_{q,i} \int {\rm d}\Phi^{i}_{\text{Born}} \, \left[ J_q \otimes S \otimes H_q \otimes {\cal I}_{qi} \right]. \end{eqnarray} We have introduced the schematic notation $J_q \otimes S \otimes H_q \otimes {\cal I}_{qi}$ for the integrations of the SCET functions over ${\cal T}_1$, $t_J$, $t_B$, and $k_S$; ${\rm d}\Phi^{i}_{\text{Born}}$ represents all other terms for the given index $i$: the parton distribution functions, the integral over the Born phase space, and any measurement function acting on the Born variables. We denote the expansion of these functions in $\alpha_s$ as \begin{equation} {\cal X} = {\cal X}^{(0)}+{\cal X}^{(1)}+{\cal X}^{(2)}+\ldots, \end{equation} where the superscript denotes the power of $\alpha_s$ appearing in each term. With this notation, we need the following contributions to obtain the ${\cal O}(\alpha_s^2)$ correction to the cross section below ${\cal T}_1^{cut}$: \begin{eqnarray} \label{eq:belowcutexp} {\rm d}\sigma_{\text{NNLO}} &=& \sum_{q,i} \int {\rm d}\Phi^i_{\text{Born}} \, \bigg\{J_q^{(2)} \otimes S^{(0)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(0)} \nonumber \\ &+&J_q^{(0)} \otimes S^{(2)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(0)} + J_q^{(0)} \otimes S^{(0)} \otimes H_q^{(2)} \nonumber \\ &\otimes & {\cal I}_{qi}^{(0)}+J_q^{(0)} \otimes S^{(0)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(2)} \nonumber \\ &+&J_q^{(1)} \otimes S^{(1)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(0)}+J_q^{(1)} \otimes S^{(0)} \otimes H_q^{(1)} \nonumber \\ &\otimes & {\cal I}_{qi}^{(0)} +J_q^{(1)} \otimes S^{(0)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(1)}+J_q^{(0)} \otimes S^{(1)} \nonumber \\ & \otimes &H_q^{(1)} \otimes {\cal I}_{qi}^{(0)} + J_q^{(0)} \otimes S^{(1)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(1)}\nonumber \\ &+& J_q^{(0)} \otimes S^{(0)} \otimes H_q^{(1)} \otimes {\cal I}_{qi}^{(1)} \bigg\}. \end{eqnarray} To simplify this expression, we first note that the hard function has no dependence on the hadronic variables ${\cal T}_1$, $t_J$, $t_B$, and $k_S$. It depends only on the Born phase space and is a multiplicative factor for the hadronic integrations. Next, we note that the leading-order expressions for the SCET functions are proportional to delta functions in their respective hadronic variable: \begin{equation} {\cal X}^{(0)} \propto \delta(t_{\cal X}) \end{equation} for ${\cal X}=J_q, S$, or ${\cal I}_{qi}$. This simplifies the integrals involving an ${\cal X}^{(2)}$. Using the $J_q^{(2)}$ term in Eq.~(\ref{eq:belowcutexp}) as an example, we have \begin{equation} J_q^{(2)} \otimes S^{(0)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(0)} = Q^2 H_q^{(0)} \delta_{qi}\, \int_0^{{\cal T}_1^{cut}} \hspace{-0.5cm}{\rm d} {\cal T}_1 J_q^{(2)} ({\cal T}_1 Q^2,\mu), \end{equation} where the $\delta_{qi}$ comes from the ${\cal I}_{qi}^{(0)}$ term. Using the fact that the jet function can be written in the form \begin{equation} J_q^{(2)} (t_J,\mu) = a_{-1} \delta(t_J)+\sum_{n=0}^3 a_n \, \frac{1}{\mu^2} \left[ \frac{\mu^2 \,\text{ln}^n(t_J/\mu^2)}{t_J}\right]_+, \end{equation} where the $a_i$ denote coefficients that can be found in the literature~\cite{Becher:2006qw}, we can immediately derive \begin{equation} \begin{split} J_q^{(2)} \otimes S^{(0)} \otimes H_q^{(0)} & \otimes {\cal I}_{qi}^{(0)} = H_q^{(0)} \,\delta_{qi} \left\{ a_{-1} \right. \\ & \left. +\sum_{n=0}^3 \frac{1}{n+1} a_{n+1} \,\text{ln}^{n+1} \left( \frac{{\cal T}_1^{cut} Q^2}{\mu^2}\right)\right\}. \end{split} \end{equation} We have used the standard definition of the plus distribution of a function: \begin{equation} \int_0^1 {\rm d} x \, \left[f(x)\right]_+ \, g(x) = \int_0^1 {\rm d} x \, f(x) \left[g(x)-g(0) \right]. \end{equation} This analysis shows how to analytically calculate any term containing one of the NNLO SCET functions. It remains only to calculate contributions containing two NLO SCET functions. We focus on $J_q^{(1)} \otimes S^{(1)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(0)} $ as an example. Using the LO expression for the beam function we can immediately derive the equation \begin{equation} \begin{split} J_q^{(1)} \otimes S^{(1)} \otimes H_q^{(0)} &\otimes {\cal I}_{qi}^{(0)} = Q \, H_q^{(0)} \,\delta_{qi}\, \int_0^{{\cal T}_1^{cut}} \hspace{-0.5cm}{\rm d} {\cal T}_1 \int_0^{{\cal T}_1 Q} dt_J \\ & \times J_q^{(1)} (t_J,\mu) \, S^{(1)}\left(Q{\cal T}_1-\frac{t_J}{Q} \right). \end{split} \end{equation} The NLO results for the jet and soft functions can be written as \begin{eqnarray} J_q^{(1)} (t_J,\mu) &=& b_{-1} \delta(t_J)+\sum_{n=0}^1 b_n \, \frac{1}{\mu^2} \left[ \frac{\mu^2 \,\text{ln}^n(t_J/\mu^2)}{t_J}\right]_+, \nonumber \\ S^{(1)} (k_S,\mu) &=& c_{-1} \delta(k_S)+\sum_{n=0}^1 c_n \, \frac{1}{\mu} \left[ \frac{\mu \,\text{ln}^n(k_S/\mu)}{k_S}\right]_+ . \end{eqnarray} Using these expressions it is straightforward to derive the result \begin{eqnarray} J_q^{(1)} &\otimes& S^{(1)} \otimes H_q^{(0)} \otimes {\cal I}_{qi}^{(0)} = H_q^{(0)} \,\delta_{qi} \bigg\{ b_{-1} c_{-1} \nonumber \\ &+& c_{-1} \, \sum_{n=0}^1 \frac{1}{n+1} \,b_{n+1}\, L_J^{n+1} + b_{-1} \, \sum_{n=0}^1 \frac{1}{n+1} \,c_{n+1}\, L_S^{n+1} \nonumber \\ &+&\left. \sum_{n,m=0}^1 b_m c_n \left[ L_J^{\alpha} \,L_S^{\beta} \,\frac{\Gamma(\alpha)\Gamma(\beta)}{\Gamma(1+\alpha+\beta)}\right] \bigg|_{\alpha^m,\beta^n}\right\}. \end{eqnarray} The vertical bar in the last term indicates that we should take the $\alpha^m \beta^n$ coefficient of the series expansion of the bracketed term. We have introduced the abbreviations \begin{equation} L_J = \text{ln} \left( \frac{{\cal T}_1^{cut} Q^2}{\mu^2}\right), \;\;\; L_S = \text{ln} \left( \frac{{\cal T}_1^{cut} Q}{\mu}\right). \end{equation} Using these results it is straightforward to analytically compute all of the necessary hadronic integrals in Eq.~(\ref{eq:belowcutexp}). The remaining integrals are over the Born phase space and parton distribution functions, and are simple to perform numerically. This completes the calculation of the ${\cal T}_1 < {\cal T}_1^{cut}$ phase space region. We note that the cross section below ${\cal T}_1^{cut}$ will contain terms of the form $\text{ln}^n ({\cal T}_1^{cut})$, where $n$ ranges from 0 to 4 at NNLO. An important check of this framework is the cancellation of these terms against the identical logarithms that appear for ${\cal T}_1 > {\cal T}_1^{cut}$. We must also choose ${\cal T}_1^{cut}$ small enough to avoid the power corrections in Eq.~(\ref{eq:SCETfac}) that go as ${\cal T}_1^{cut}/Q$. Both of these issues will be addressed in our later section on numerical results. \subsubsection{Above ${\cal T}_1^{cut}$} We now briefly outline the computation of the quark-lepton channel in the phase space region ${\cal T}_1 > {\cal T}_1^{cut}$. As discussed above this can be obtained from a NLO calculation of the DIS process with an additional jet. In addition to the virtual corrections to the $q\,l \to q\,l\,g$ partonic process, there are numerous radiation processes that also contribute: $q\,l \to q\,l\,g\,g$, $q\,l \to q\,l\,q'\,\bar{q}'$ and $q\,l \to q\,l\,q\,\bar{q}$. In addition to the usual ultraviolet renormalization of the strong coupling constant, the real and virtual corrections are separately infrared divergent. Remaining divergences after introducing an NLO subtraction scheme are associated with initial-state collinear singularities, and are handled via mass factorization. \section{Validation and numerical results} \label{sec:numerics} We have implemented the NNLO cross section of Eq.~(\ref{eq:hxsecnnlo}), as well as the LO and NLO results of Eqs.~(\ref{eq:sigLO}) and~(\ref{eq:hxsecnlo}) in a numerical code {\tt DISTRESS} that allows for arbitrary cuts to be imposed on the final-state lepton and jets. We describe below the checks we have performed on our calculation. The antennae subtraction method provides an analytic cancellation of the $1/\epsilon$ poles that appear in an NLO calculation. We are therefore able to check this cancellation of poles for all components computed in this way. This includes the entire $\sigma_{\rm NLO}$, as well as the following contributions to the NNLO hadronic cross section: ${\rm d}\hat{\sigma}_{g\gamma}^{(1,2)}$ and ${\rm d}\hat{\sigma}_{q\gamma}^{(1,2)}$. The various contributions ${\rm d}\hat{\sigma}_{ij}^{(0,2)}$ that occur in $\sigma_{\rm NNLO}$ are finite, and simple to obtain. Our NLO results for the transverse momentum distribution and pseudorapidity distribution of the leading jet are compared against the plots of Ref.~\cite{Hinderer:2015hra}. We find good agreement with these results. This leaves only the validation of the ${\rm d}\hat{\sigma}_{gl}^{(2,2)}$ and ${\rm d}\hat{\sigma}_{ql}^{(2,2)}$ contributions, which both utilize the $N$-jettiness subtraction technique. There are two primary checks that these pieces must satisfy. First, they must be independent of the parameter ${\cal T}_N^{cut}$. This checks the implementation of the beam, jet and soft functions, which have logarithmic dependence on this parameter. It also determines the range of ${\cal T}_N^{cut}$ for which the power corrections denoted by the ellipsis in Eq.~(\ref{eq:SCETfac}) are negligibly small. Second, upon integration over the final-state hadronic phase space we must reproduce the NNLO structure functions first determined in Ref.~\cite{Zijlstra:1992qd}. This is an extremely powerful check on our calculation, which essentially cannot be passed if any contribution is implemented incorrectly. We show in Fig.~\ref{fig:valid} the results of these checks for the $ql$ and $gl$ scattering channels. We have set the total center-of-mass energy of the lepton-proton collision to 100 GeV. For the purpose of this validation check only we have imposed the phase-space cut $Q^2>100$ GeV$^2$, and have integrated inclusively over the hadronic phase space. We have equated the renormalization and factorization scales to the common value $\mu=Q$, and have used the CT14 NNLO parton distribution functions~\cite{Dulat:2015mca}. ${\cal T}_1^{cut}$ has been varied from $5 \times 10^{-6}$ to $1 \times 10^{-4}$, and the ratio of the NNLO correction to the LO result for the cross section is shown. The solid lines show the prediction of the inclusive structure function. We first note that correction is extremely small, less than 1\% of the leading-order result. Nevertheless we have excellent numerical control over the NNLO coefficient, as indicated by the vertical error bars. Our numerical error on the NNLO coefficient is at the percent-level, sufficient for 0.01\% precision on the total cross section. The $N$-jettiness prediction for the $ql$ scattering channel is independent of ${\cal T}_1^{cut}$ over the studied range, while the $gl$ scattering channel is independent of ${\cal T}_1^{cut}$ for ${\cal T}_1^{cut} < 10^{-5}$. Both channels are in excellent agreement with the structure-function result. We have also checked bin-by-bin that the transverse momentum and pseudorapidity distributions of the jet have no dependence on ${\cal T}_1^{cut}$. \begin{figure}[h] \centering \includegraphics[width=3.6in]{valid.pdf}% \caption{Plot of the NNLO corrections normalized to the LO cross section for the quark-lepton and gluon-lepton scattering channels as a function of ${\cal T}_1^{cut}$. The points denote values obtained from $N$-jettiness subtraction, with the vertical error bars denoting the numerical errors, while the solid lines indicate the inclusive structure function result. } \label{fig:valid} \end{figure} Having established the validity of our calculation we present phenomenological results for proposed EIC run parameters. We set the collider energy to $\sqrt{s}=100$ GeV and study the inclusive-jet transverse momentum and pseudorapidity distributions in the range $p_T^{jet}>5$ GeV and $|\eta_{jet}|<2$. We use the CT14 parton distribution set~\cite{Dulat:2015mca} extracted to NNLO in QCD perturbation theory. We reconstruct jets using the anti-$k_t$ algorithm~\cite{Cacciari:2008gp} with radius parameter $R=0.5$. Our central scale choice for both the renormalization and factorization scales is $\mu=p_T^{jet}$. To estimate the theoretical errors from missing higher-order corrections we vary the scale around its central value by a factor of two. The transverse momentum distributions at LO, NLO and NNLO are shown in Fig.~\ref{fig:ptj}. The $K$-factors, defined as the ratios of higher order over lower order cross sections, are shown in the lower panel of this figure. The NNLO corrections are small in the region $p_T^{jet}>10$ GeV , changing the NLO result by no more than 10\% over the studied $p_T^{jet}$ region. The shift of the NLO cross section is slightly positive in the low transverse momentum region, and become less than unity at high-$p_T^{jet}$. Both the NNLO corrections and the scale dependence grow large at low-$p_T^{jet}$. The large scale dependence arises primarily from the partonic channels $qq$ and $gq$. These channels are effectively treated at leading order in our calculation, since they first appear at ${\cal O}(\alpha^2\alpha_s^2)$, and they are evaluated at the low scale $\mu=p_T^{jet}/2$ in our estimate of the theoretical uncertainty. It is therefore not surprising that their uncertainty dominates at low-$p_T^{jet}$ These channels do not contribute at NLO, and consequently the NLO scale uncertainty is smaller. This is an example of the potential pitfalls in using the scale uncertainty as an estimate of the theoretical uncertainty. Only an explicit calculation can reveal qualitatively new effects that occur at higher orders in perturbation theory. \begin{figure}[h] \centering \includegraphics[width=3.65in]{ptj.pdf}% \caption{Plot of the inclusive-jet transverse momentum distribution at LO, NLO and NNLO in QCD perturbation theory. The upper panel shows the distributions with scale uncertainties, while the lower panel shows the $K$-factors for the central scale choice.} \label{fig:ptj} \end{figure} We next show the pseudorapidity distribution in Fig.~\ref{fig:etaj}, with the restriction $p_T^{jet}>10$ GeV. There are a few surprising aspects present in the NNLO corrections. First, the scale dependence at NNLO in the region $\eta_{jet}<0$ is larger than the corresponding NLO scale variation. Although the corrections are near unity over most of the studied pseudorapidity range, they become sizable near $\eta_{jet} \approx 2$, reducing the NLO rate by nearly 50\%. To determine the origin of these effects we show in Fig.~\ref{fig:eta_breakdown_muv} the breakdown of the NNLO correction into its separate partonic channels. This reveals that the total NNLO correction comes from an intricate interplay between all contributing channels, with different ones dominating in different $\eta_{jet}$ regions. Only the gluon-lepton partonic process is negligible over all of phase space. For negative $\eta_{jet}$, the dominant contribution is given by the quark-quark process. As discussed before, this appears first at ${\cal O}(\alpha^2\alpha_s^2)$. It is therefore effectively treated at leading-order in our calculation, and consequently has a large scale dependence. We note that the quark-in-lepton distribution from Eq.~(\ref{eq:qinl}) is larger at high-$x$ than the corresponding photon-in-lepton one, leading to this channel being larger in the negative $\eta_{jet}$ region. At high $\eta_{jet}$, the distribution receives sizable contributions from the gluon-photon process. No single partonic channel furnishes a good approximation to the shape of the full NNLO correction. \begin{figure}[h] \centering \includegraphics[width=3.65in]{etaj.pdf}% \caption{Plot of the inclusive-jet pseudorapidity distribution at LO, NLO and NNLO in QCD perturbation theory. The upper panel shows the distributions with scale uncertainties, while the lower panel shows the $K$-factors for the central scale choice.} \label{fig:etaj} \end{figure} \begin{figure}[h] \centering \includegraphics[width=3.65in]{eta_breakdown_muv.pdf}% \caption{Breakdown of the NNLO correction to the $\eta_{jet}$ distribution into its constituent partonic channels, as a ratio to the full NLO cross section in the bin under consideration. Also shown is the total result obtained by summing all channels. The bands indicate the scale variation. } \label{fig:eta_breakdown_muv} \end{figure} \section{Conclusions} \label{sec:conc} We have presented in this paper the full calculation of the ${\cal O}(\alpha^2 \alpha_s^2)$ perturbative corrections to jet production in electron-nucleus collisions. To obtain this result we have utilized the $N$-jettiness subtraction scheme introduced to allow NNLO calculations in hadronic collisions. We have described the application of this method to inclusive jet production in detail, and have shown that upon integration over the final-state hadronic phase that we reproduce the known NNLO result for the inclusive structure functions. Our results have been implemented in a numerical program {\tt DISTRESS} that we plan to make publicly available for future phenomenological studies. We have shown numerical results for jet production at a proposed future EIC. Several new partonic channels appear at the ${\cal O}(\alpha^2 \alpha_s^2)$ level, which have an important effect on the kinematic distributions of the jet. No single partonic channel furnishes a good approximation to the full NNLO result. The magnitude of the corrections we find indicate that higher-order predictions will be an important part of achieving the precision understanding of proton structure desired at the EIC, and we expect that the methods described here will be an integral part of achieving this goal. \vspace{0.5cm} \noindent \mysection{Acknowledgements} G.~A. is supported by the NSF grant PHY-1520916. R.~B. is supported by the DOE contract DE-AC02-06CH11357. X.~L. is supported by the DOE grant DE-FG02-93ER-40762. F.~P. is supported by the DOE grants DE-FG02-91ER40684 and DE-AC02-06CH11357. This research used resources of the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231. It also used resources of the Argonne Leadership Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-06CH11357. This research was supported in part by the NSF under Grant No. NSF PHY11-25915 to the Kavli Institute of Theoretical Physics in Santa Barbara, which we thank for hospitality during the completion of this manuscript.
417ff865a9cbac3a94a3b3975ec2721852d90a63
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} Compact stars have a large number of pulsation modes that have been extensively studied since the seminal work of Chandrasekhar on radial oscillations \cite{APJ140:417:1964,PRL12:114:1964}. In general, these modes are very difficult to observe in the electromagnetic spectrum; therefore most efforts have concentrated on gravitational wave asteroseismology in order to characterise the frequency and damping times of the modes that emit gravitational radiation. In particular, various works focused on the oscillatory properties of pure hadronic stars, hybrid stars and strange quark stars trying to find signatures of the equation of state of high density neutron star matter (see \cite{AA325:217:1997,IJMPD07:29:1998,AA366:565:2001,APJ579:374:2002,PRD82:063006:2010,EL105:39001:2014} and references therein). More recently, compact star oscillations have attracted the attention in the context of Soft Gamma ray Repeaters (SGRs), which are persistent X-ray emitters that sporadically emit short bursts of soft $\gamma$-rays. In the quiescent state, SGRs have an X-ray luminosity of $\sim 10^{35}$ erg/s, while during the short $\gamma$-bursts they release up to $10^{42}$ erg/s in episodes of about 0.1 s. Exceptionally, some of them have emitted very energetic giant flares which commenced with brief $\gamma$-ray spikes of $\sim 0.2$ s, followed by tails lasting hundreds of seconds. Hard spectra (up to 1 MeV) were observed during the spike and the hard X-ray emission of the tail gradually faded modulated at the neutron star (NS) rotation period. The analysis of X-ray data of the tails of the giant flares of SGR 0526-66, SGR 1900+14 and SGR 1806-20 revealed the presence of quasi-periodic oscillations (QPOs) with frequencies ranging from $\sim$ 18 to 1840 Hz \cite{APJ628:L53:2005,APJ632:L111:2005,AA528:A45:2011}. There are also candidate QPOs at higher frequencies up to $\sim 4$ kHz in other bursts but with lower statistical significance \cite{ElMezeini2010}; in fact, according to a more recent analysis only one burst shows a marginally significant signal at a frequency of around 3706 Hz \cite{Huppenkothen2013}. Several characteristics of SGRs are usually explained in terms of the \textit{magnetar} model, assuming that the object is a neutron star with an unusually strong magnetic field ($B \sim 10^{15} $ G) \cite{Woods2006}. In particular, giant flares are associated to catastrophic rearrangements of the magnetic field. Such violent phenomena are expected to excite a variety of oscillation modes in the stellar crust and core. In fact, recent studies have accounted for magnetic coupling between the crust and the core, and associate QPOs to global magneto-elastic oscillations of highly magnetized neutron stars \cite{Levin2007,CerdaDuran2009,Colaiuda2012,Gabler2014}. There has also been interest in the possible excitation of low order $f$-modes because of their strong coupling to potentially detectable gravitational radiation \cite{Levin2011}. In the present paper we focus on radial oscillations of neutron stars permeated by ultra-strong magnetic fields. These modes might be relevant within the magnetar model because they could be excited during the violent events associated with gamma flares. Since they have higher frequencies than the already known QPOs, they cannot be directly linked to them at present. However, it is relevant to know all the variety of pulsation modes of strongly magnetized neutron stars because the number of observations is still small and new features could emerge in future flares' data. On the other hand, in the case of rotating objects we can expect some amount of gravitational radiation from even the lowest ($l = 0$) quasi-radial mode \cite{Stergioulas2003,Passamonti2006} making them potentially relevant for gravitational wave astronomy. \section{Equations of state} \label{sec2} \subsection{Hadronic phase under a magnetic field} \label{sec:A} In this section we present an overview of the hadronic equations of state (EOS) used in this work. We describe hadronic matter within the framework of the relativistic non-linear Walecka (NLW) model \cite{WALECKA1986}. In this model we employ a field-theoretical approach in which the baryons interact via the exchange of $\sigma-\omega-\rho$ mesons in the presence of a magnetic field $B$ along the $z-$axis. The total lagrangian density reads: \begin{equation}\label{lt} \mathcal{L}_{H}=\sum_{b}\mathcal{L}_{b}+\mathcal{L}_{m}+\sum_{l}\mathcal{L}_{l}+\mathcal{L}_B\,. \end{equation} where $\mathcal{L}_{b}$, $\mathcal{L}_{m}$, $\mathcal{L}_{l}$ and $\mathcal{L}_{B}$ are the baryons, mesons, leptons and electromagnetic field Lagrangians, respectively, and are given by \begin{eqnarray} \mathcal{L}_{b} &=& \overline{\psi}_{b}\left(i\gamma_{\mu}\partial^{\mu}-q_{b}\gamma_{\mu}A^{\mu}-m_{b}+g_{\sigma b}\sigma \right. \nonumber \\ && \left. -g_{\omega b}\gamma_{\mu}\omega^{\mu}-g_{\rho b}\tau_{3b}\gamma_{\mu}\rho^{\mu}\right)\psi_{b}\,, \\ \mathcal{L}_{m} &=& \tfrac{1}{2}(\partial_{\mu}\sigma\partial^{\mu}\sigma-m_{\sigma}^{2}\sigma^{2})-U(\sigma)+ \tfrac{1}{2}m_{\omega}^{2}\omega_{\mu}\omega^{\mu} \nonumber \\ && -\tfrac{1}{4}\Omega_{\mu\nu}\Omega^{\mu\nu}+ \tfrac{1}{2}m_{\rho}^{2}\vec{\rho}_{\mu}\cdot\vec{\rho}_{\mu}-\tfrac{1}{4}P^{\mu\nu}P_{\mu\nu} \,, \\ \mathcal{L}_{l} &=& \overline{\psi}_{l}\left(i\gamma_{\mu}\partial^{\mu}-q_{l}\gamma_{\mu}A^{\mu}-m_{l}\right)\psi_{l} \,, \\ \mathcal{L}_{B} &=& -\tfrac{1}{4}F^{\mu\nu}F_{\mu\nu} \,. \end{eqnarray} where he $b$-sum runs over the baryonic octet $b\equiv N~(p,~n),~\Lambda,~\Sigma^{\pm,0},~\Xi^{-,0}$, $\psi_{b}$ is the corresponding baryon Dirac field, whose interactions are mediated by the $\sigma$ scalar, $\omega_{\mu}$ isoscalar-vector and $\rho_{\mu}$ isovector-vector meson fields. The baryon charge, baryon mass and isospin projection are denoted by $q_{b}$, $m_{b}$ and $\tau_{3b}$, respectively, and the masses of the mesons are $ m_{\sigma}= 512~$ MeV, $m_{\omega}=783~$MeV and $m_{\rho}=770~$MeV. The strong interaction couplings of the nucleons with the meson fields are denoted by $g_{\sigma N}=8.910$, $g_{\omega N}=10.610$ and $g_{\rho N}=8.196$. We consider that the couplings of the hyperons with the meson fields are fractions of those of the nucleons, defining $g_{iH}=X_{iH}g_{iN}$, where the values of $X_{iH}$ are chosen as $X_{\sigma H}=0.700$ and $X_{\omega H}=X_{\rho H}=0.783$ \cite{GLENDENNING2000}. The term $U(\sigma)=\frac{1}{3}\,bm_{n}(g_{\sigma N}\sigma)^{3}-\frac{1}{4}\,c(g_{\sigma N}\sigma)^{4}$ denotes the scalar self-interactions \cite{NPA292:413:1977,PLB114:392:1982,AJ293:470:1985}, with $c=-0.001070$ and $b=0.002947$. The mesonic and electromagnetic field tensors are given by their usual expressions $\Omega_{\mu\nu}=\partial_{\mu}\omega_{\nu}-\partial_{\nu}\omega_{\mu}$, ${\bf P}_{\mu\nu}=\partial_{\mu}\vec{\rho}_{\nu}-\partial_{\nu}\vec{\rho}_{\mu}-g_{\rho b}(\vec{\rho}_{\mu}\times\vec{\rho}_{\nu})$ and $F_{\mu\nu}=\partial_{\mu}A_{\nu}-\partial_{\nu}A_{\mu}$. The $l$-sum runs over the two lightest leptons $l\equiv e,\mu$ and $\psi_{l}$ is the lepton Dirac field. The symmetric nuclear matter properties at saturation density adopted in this work are given by the GM1 parametrization \cite{PRL67:2414:1991}, with compressibility $K=300$ MeV, binding energy $B/A=-16.3$ MeV, symmetry energy $a_{sym}=32.5$ MeV, slope $L=94$ MeV, saturation density $\rho_{0}= 0.153$ fm$^{-3}$ and nucleon mass $m=938$ MeV. The following equations present the scalar and vector densities for the charged and uncharged baryons \cite{JPG36:115204:2009,PRC89:015805:2014}, respectively: \begin{eqnarray} &\label{densities1}\rho_{b}^{s}=\frac{|q_{b}|B \bar{m}_{b} }{2\pi^{2}}\sum_{\nu}^{\nu_{\mathrm{max}}}\sum_{s}\frac{\bar{m}_{b}^{c}}{\sqrt{ \bar{m}_{b}^2 + 2\nu |q_{b}|B}} \ln\bigg|\frac{k_{F,\nu,s}^{\,b}+E_{F}^{\,b}}{\bar{m}_{b}^{c}} \bigg|,\\ &\label{densities2}\rho_{b}^{v}=\frac{|q_{b}|B}{2\pi^{2}}\sum_{\nu}^{\nu_{\mathrm{max}}}\sum_{s}k_{F,\nu,s}^{\,b},\\ &\label{densities3}\rho_{b}^{s}=\frac{\bar{m}_{b}}{4\pi^{2}}\sum_{s}\bigg[E_{F}^{\,b}k_{F,s}^{\,b}-\bar{m}_{b}^{2}\ln\bigg|\frac{k_{F,s}^{\,b}+E_{F}^{\,b}}{\bar{m}_{b}}\bigg|\bigg] ,\\ &\label{densities4}\rho_{b}^{v}=\frac{1}{2\pi^{2}}\sum_{s}\bigg[\frac{1}{3}(k_{F,s}^{\,b})^{3}\bigg] , \end{eqnarray} % where $\bar{m}_{b}=m_{b}-g_{\sigma}\sigma$ and $\bar{m}_{b}^{c}=\sqrt{ \bar{m}_{b}^2 + 2\nu |q_{b}|B}$. $\nu=n+\frac{1}{2}-$sgn$(q_{b})\frac{s}{2}=0,1,2,...$ are the Landau levels for the fermions with electric charge $q_{b}$, $s$ is the spin and assumes values $+1$ for spin up and $-1$ for spin down cases. The energy spectra for the baryons are given by \cite{JPG35:125201:2008,APJ537:351:2000}: \begin{eqnarray}\label{energy} E_{\nu,s}^{\,b}=\sqrt{(k_{z}^{\,b})^{2}+\bar{m}_{b}^{2}+2\nu |q_{b}|B}+g_{\omega b}\omega^{0}+\tau_{3b}g_{\rho b}\rho^{0}\\ E_{s}^{\,b}=\sqrt{(k_{z}^{\,b})^{2}+\bar{m}_{b}^{2}+(k_{\perp}^{\,b})^{2}}+g_{\omega b}\omega^{0}+\tau_{3b}g_{\rho b}\rho^{0}, \end{eqnarray} where $k_{\perp}^{\,b}=k_{x}^{\,b}+k_{y}^{\,b}$. The Fermi momenta $k_{F,\nu,s}^{\,b}$ of the charged baryons and $k_{F,s}^{\,b}$ of the uncharged baryons and their relationship with the Fermi energies of the charged baryons $E_{F,\nu,s}^{\,b}$ and uncharged baryons $E_{F,s}^{\,b}$ can be written as: \begin{eqnarray}\label{Momentum} &&(k_{F,\nu,s}^{\,b})^{2}=(E_{F,\nu,s}^{\,b})^{2}-(\bar{m}_{b}^{c})^{2} \\ &&(k_{F,s}^{\,b})^{2}=(E_{F,s}^{\,b})^{2}-\bar{m}_{b}^{2}. \end{eqnarray} For the leptons, the vector density is given by: \begin{eqnarray}\label{vector_{density_{leptons}}} \rho_{l}^{v}=\frac{|q_{l}|B}{2\pi^{2}}\sum_{\nu}^{\nu_{\mathrm{max}}}\sum_{s}k_{F,\nu,s}^{\,l}, \end{eqnarray} where $k_{F,\nu,s}^{\,l}$ is the lepton Fermi momentum, which is related to the Fermi energy $E_{F,\nu,s}^{\,l}$ by: \begin{eqnarray}\label{Momentuml} (k_{F,\nu,s}^{\,l})^{2}=(E_{F,\nu,s}^{\,l})^{2}-\bar{m}_{l}^{2}\,, \qquad l=e,\mu, \end{eqnarray} \noindent with $\bar{m}_{l}=m_{l}^{2}+2\nu |q_{l}|B$. The summation over the Landau level runs until $\nu_{\mathrm{max}}$; this is the largest value of $\nu$ for which the square of Fermi momenta of the particle is still positive and corresponds to the closest integer, from below to: \begin{eqnarray} &&\nu_{\mathrm{max}}=\bigg[\frac{(E_{F}^{\,l})^{2}-m_{l}^{2}}{2|q_{l}|B}\bigg], \quad\mathrm{leptons}\label{ll1}\\ &&\nu_{\mathrm{max}}=\bigg[\frac{(E_{F}^{\,b})^{2}-\bar{m}_{b}^{2}}{2|q_{b}|B}\bigg], \quad\mathrm{charged~baryons}.\label{ll2} \end{eqnarray} The chemical potentials of baryons and leptons are: \begin{eqnarray}\label{chemicalp} &&\mu_{b}=E_{F}^{\,b}+g_{\omega b}\omega^{0}+\tau_{3b}g_{\rho b}\rho^{0},\\ &&\mu_{l}=E_{F}^{\,l}=\sqrt{(k_{F,\nu,s}^{\,l})^{2}+m_{l}^{2}+2\nu |q_{l}|B}\,. \end{eqnarray} From the Lagrangian density~(\ref{lt}), and mean-field approximation, the energy density is given by \begin{eqnarray}\label{energym} \varepsilon_{m}= & & \sum_{b}(\varepsilon_{b}^c + \varepsilon_{b}^n) +\tfrac{1}{2}m_{\sigma}\sigma_{0}^{2}\nonumber\\ && +U(\sigma)+\tfrac{1}{2}m_{\omega}\omega_{0}^{2}+\tfrac{1}{2}m_{\rho}\rho_{0}^{2}\,, \end{eqnarray} where the expressions for the energy densities of charged baryons $\varepsilon_{b}^{c}$ and neutral baryons $\varepsilon_{b}^{n}$ are, respectively, given by: \begin{eqnarray}\label{energy-densities-baryons} \varepsilon_{b}^{c}& = &\frac{|q_{b}|B}{4\pi^{2}}\sum_{\nu}^{\nu_{\mathrm{max}}}\sum_{s}\bigg[k_{F,\nu,s}^{\,b}E_{F}^{\,b} \nonumber \\ & & + (\bar{m}_{b}^{c})^{2}\ln\bigg|\frac{k_{F,\nu,s}^{\,b}+E_{F}^{\,b}}{\bar{m}_{b}^{c}}\bigg|\bigg],\label{ea1} \\ \varepsilon_{b}^{n}& = &\frac{1}{4\pi^{2}}\sum_{s}\bigg[\tfrac{1}{2} k_{F,\nu,s}^{\,b}(E_{F}^{\,b})^{3} - \tfrac{1}{4}\bar{m}_{b} \bigg(\bar{m}_{b}k_{F,\nu,s}^{\,b}E_{F}^{\,b} \nonumber \\ && + \bar{m}_{b}^{3}\ln\bigg|\frac{E_{F}^{\,b}+k_{F,\nu,s}^{\,b}}{\bar{m}_{b}}\bigg|\bigg)\bigg]\, . \label{ea2} \end{eqnarray} \noindent The expression for the energy density of leptons $\varepsilon_{l}$ reads \begin{equation} \varepsilon_{l}= \frac{|q_{l}|B}{4\pi^{2}}\sum_{l}\sum_{\nu}^{\nu_{\mathrm{max}}}\sum_{s}\bigg[k_{F,\nu,s}^{\,l}E_{F}^{\,l}+ \bar{m}_{l}^{2}\ln\bigg|\frac{k_{F,\nu,s}^{\,l}+E_{F}^{\,l}}{\bar{m}_{l}}\bigg|\bigg]\,. \label{ea3} \end{equation} The pressures of baryons and leptons are: \begin{eqnarray}\label{pressurem} P_{m}&=&\mu_{n}\sum_{b}\rho_{b}^{v}-\varepsilon_{m}, \\ P_{l}&=&\sum_{l}\mu_{l}\rho_{l}^{v}-\varepsilon_{l}, \end{eqnarray} where the expression of the vector densities $\rho_{b}^{v}$ and $\rho_{l}^{v}$ are given in~(\ref{densities2}) and (\ref{vector_{density_{leptons}}}), respectively. \subsection{Density-dependent magnetic field} \label{B-density-depndence} We assume that the magnetic field $B$ in the EOS depends on the density according to \cite{PRL79:2176:1997,ChJAA3:359:2003,PRC80:065805:2009,JPG36:115204:2009,BJP42:428:2012} \begin{equation}\label{cmdd} B\left(\frac{\rho}{\rho_{0}}\right)=B_{\mathrm{surf}}+B_{0}\left\{ 1-\mathrm{exp}\left[ -\beta\left( \frac{\rho}{\rho_{0}}\right)^{\gamma} \right] \right\}\,, \end{equation} where $\rho=\sum_{b}\rho_{b}^{v}$ is the baryon density, $\rho_{0}$ is the saturation density, $B_{\mathrm{surf}}$ is the magnetic field on the surface of a magnetar, taken equal to $10^{15}$~G in agreement with observational values, and $B_{0}$ is the magnetic field for larger densities. The parameters $\beta$ and $\gamma$ are chosen to reproduce two behaviors of the magnetic field: a fast decay with $\gamma=3.00$ and $\beta=0.02$ and a slow decay with $\gamma=2.00$ and $\beta=0.05$ \cite{PRC89:015805:2014}. According to the discussion in the previous subsection, we use two values for the magnetic field $B_{0}$, namely $10^{17}$~G and $3.1\times10^{18}$~G. \subsection{On the isotropy of the pressure} Notice that in the previous subsections we assumed that the matter pressure ($P_m + P_l$) is isotropic in spite of the high values of the magnetic field. As it has been shown in Ref. \cite{JPG41:015203:2014}, the anisotropic effects around $3.1\times10^{18}$~G are small, thus we restrict ourselves to magnetic fields below this value. However, the purely field-related pressure $P_B \sim B^2$ may become dominant in the core of the star. In such cases, the total pressure perpendicular to the magnetic field lines would be significantly larger than the pressure parallel to the field lines. Therefore, 3D or at least 2D stellar structure equations should be used in order to incorporate the effect of the pressure anisotropy. We must stress however, that the magnetic field geometry inside a neutron star can be extremely complex, and depending on its configuration the use of the spherically symmetric stellar structure equations can still be a good enough approximation, as we argue below. If we consider, for example, a purely dipole field (purely poloidal field), a 1D stellar structure calculation would be certainly inappropriate for some very high field objects. But such configuration is unstable and cannot be realised in Nature; in fact, any neutron star with a purely poloidal or a purely toroidal magnetic field is unstable in general relativity (see \cite{Ciolfi2011} and references therein). This strongly supports the idea that any long-lived magnetic field configuration in a NS has to consist of a mixture of poloidal and toroidal field components. On the other hand, while stellar shapes have long been considered to be oblate due to the effects of centrifugal and/or magnetic forces, this is not necessarily true. For example, it has been found recently that the shape of our Sun is perfectly round, against the common expectation of an oblate shape due to its rotation \cite{Kuhn2012}. In the specific case of neutron stars, the results found thus far are that purely poloidal magnetic fields make stars oblate (equatorial radius larger than polar radius), while purely toroidal magnetic fields lead stars to become prolate (polar radius larger than equatorial radius). Very recently, it has been shown that equilibrium states of magnetized stars with mixed poloidal$-$toroidal magnetic fields are possible; in particular, it was possible to build neutron stars with twisted-torus configurations in equilibria with any toroidal and poloidal field energy content \cite{Ciolfi2013}. Such poloidal and toroidal magnetic fields act as increasing and decreasing mechanisms for stellar oblateness, respectively \cite{Fujisawa2013}. An additional finding of Ref. \cite{Ciolfi2013} is that for a fixed polar magnetic field strength, a higher relative content of toroidal field energy ($> 10 \%$) implies in general a much higher total (poloidal and toroidal) magnetic energy inside the star. This means that a highly magnetized neutron star can harbor internal magnetic fields that are significantly stronger than commonly expected. In addition, a very complex field is expected from the formation process of magnetized neutron stars, since they are born hot, highly convective and differentially rotating \cite{Thompson1996,Bonanno2003}. During its early evolution, neutron star's magnetic fields can be significantly amplified and redistributed by several mechanisms including possibly dynamo action and shear instabilities. In this context, a distribution of magnetic energy with poloidal and toroidal fields close to equipartition appears as a very reasonable candidate for the internal magnetic field configuration. In such a case, since both the toroidal and poloidal components are of the same order, we may expect that oblateness and prolateness cancel out approximately, leading to stars close to the spherical symmetry. Since a study of stellar pulsations in 2D or 3D is numerically involving, we shall adopt spherical symmetry as a first approximation, and use the much more simple 1D equations. Within this approach, we must add to the matter pressure ($P_m + P_l$) given in Sec. \ref{sec:A}, an effective mean magnetic pressure $P_B$ representing the effective isotropic contribution arising from the combined effect of poloidal and toroidal field components of the same order. Thus, the total energy density and the total pressure of the system read: \begin{eqnarray} \label{et} \varepsilon & = & \varepsilon_{m}+\varepsilon_{l} + \tfrac{1}{2} \left[B\left(\frac{\rho}{\rho_{0}} \right)\right]^{2} , \\ P & = & P_{m}+P_{l} + P_B .\label{et222} \end{eqnarray} For finding the effective mean magnetic pressure $P_B$ we shall follow here an averaging procedure similar to the one employed in works that focus on the study of magnetic fluctuations in turbulent fluids \cite{Brandenburg2012,Rogachevskii2007}. The magnetic stress tensor $\sigma_{ij}$ is given by \begin{eqnarray} \sigma_{ij} = - \frac{\langle {\bf B}^2 \rangle}{2} \delta_{ij} + \langle B_i B_j \rangle , \label{stress_tensor} \end{eqnarray} where $ \delta_{ij} $ is the Kronecker tensor and the brackets $ \langle \cdots \rangle$ denote the averaging. For a completely isotropic distribution of the magnetic field we have $\langle B_i B_j \rangle = \delta_{ij} \, \langle {\bf B}^2 \rangle / 3 $, and the magnetic stress tensor reads \begin{eqnarray} \sigma_{ij}= - {\langle {\bf B}^2 \rangle \over 6} \, \delta_{ij} \label{I2} \end{eqnarray} The effective mean magnetic pressure $P_B$ is related to the magnetic stress tensor by $\sigma_{ij} = - P_B \, \delta_{ij}$. Therefore, we find: \begin{eqnarray} P_{B} = \tfrac{1}{6} \left[B\left(\frac{\rho}{\rho_{0}} \right)\right]^{2} . \label{WA1} \end{eqnarray} A similar expression has been used in \cite{Bednarek2003} for studying the influence of asymmetry on a magnetized proto-neutron star. \section{Radial Oscillations} In order to study the radial oscillations of a compact star, we must know first its equilibrium configuration. Such configuration is perturbed in such a way that the spherical symmetry of the body is not violated. The space-time and fluid perturbations are inserted into the Einstein equations and into the energy, momentum and baryon number conservation equations, assuming a sinusoidal time dependence $\exp{(i \omega t)}$ and retaining only the first-order terms. The result of this procedure is a second order ordinary differential equation for the perturbations \cite{APJ140:417:1964} or alternatively a set of two first order ordinary differential equations, as shown below. In the following we present the explicit form of the equilibrium and oscillation equations employed in the present work. \subsection{Equilibrium configuration} \label{sec3} We consider that the unperturbed compact star is totally composed of a perfect fluid. In this case the stress-energy momentum tensor can be expressed as \begin{equation} {T}_{\mu\nu} =(\epsilon + p) u_{\mu}u_{\nu} + p{g}_{\mu\nu}, \end{equation} where $\epsilon$ and $p$ are the energy density and pressure respectively. The background spacetime of the spherically symmetric star, can be expressed through the line element \begin{equation} \label{dsz_tov} ds^{2}=-e^{ \nu(r)} dt^{2} + e^{ \lambda(r)} dr^{2} + r^{2}(d\theta^{2}+\sin^{2}{\theta}d\phi^{2}), \end{equation} where $t,r,\theta,\phi$ are the set of Schwarzschild-like coordinates, and the metric potentials $\nu(r)$ and $\lambda(r)$ are functions of the radial coordinate $r$ only. The Einstein equations in such a spacetime lead to the following set of stellar structure equations (Tolman-Oppenheimer-Volkoff equations) \begin{eqnarray} \frac{dm}{dr} = 4 \pi r^2 \epsilon, \\ \frac{d\nu}{dr} = - \frac{2}{\epsilon} \frac{dp}{dr} \bigg(1 + \frac{p}{\epsilon}\bigg)^{-1}, \\ \frac{dp}{dr} = - \frac{\epsilon m}{r^2}\bigg(1 + \frac{p}{\epsilon}\bigg) \bigg(1 + \frac{4\pi p r^3}{m}\bigg)\bigg(1 - \frac{2m}{r}\bigg)^{-1}, \end{eqnarray} where $m$ is the gravitational mass inside the radius $r$. The metric function $\nu$ has the following boundary condition \begin{equation} \label{BoundaryConditionMetricFunction} \nu(r=R)= \ln \bigg( 1-\frac{2M}{R} \bigg), \end{equation} where $R$ is the radius of the star and $M$ its mass. With this condition the metric function $\nu$ will match smoothly to the Schwarzschild metric outside the star. The boundary conditions for $m$ and $p$ are $m(r=0) = 0$ and $p(r=R) = 0$. \begin{figure} \includegraphics[angle=0,scale=0.66]{fig1.eps} \includegraphics[angle=0,scale=0.66]{fig2.eps} \includegraphics[angle=0,scale=0.66]{fig3.eps} \caption{The period of the fundamental mode of hadronic stars as a function of (a) the mass $M$, (b) the gravitational redshift $Z$, and (c) the central energy density $\epsilon_c$. In all the figures we set three different values for the magnetic field: B$=0,10^{17}$ G and $3.1 \times 10^{18}$G. We also show the effect of slow and fast decay in the magnetic field profile. } \label{fig1} \end{figure} \begin{figure} \includegraphics[angle=0,scale=0.66]{fig4.eps} \caption{The frequency of the fundamental mode ($\omega_{n=0}=2\pi/ \tau_{n=0}$) as a function of the neutron star mass. The values for the magnetic field as well as the slow and fast decays in the magnetic field profile are the same as given in Fig. \ref{fig1}. } \label{fig2} \end{figure} \subsection{Oscillation equations} \label{sec4} Several forms of the oscillation equation have been presented in the literature (for more details the reader is referred to \cite{PRD82:063006:2010}). In this work, we use the set of equations of Gondek et al. \cite{AA325:217:1997} and adopt $G = c = 1$. The equations read \begin{eqnarray} \label{ecuacionparaXI} \frac{d\xi}{dr} &=& -\frac{1}{r}\bigg(3\xi+\frac{\Delta p}{\Gamma p}\bigg)-\frac{dp}{dr}\frac{\xi}{(p+\epsilon)}, \\ \label{ecuacionparaP} \frac{d\Delta p}{dr} &=& \xi \bigg\{\omega^{2}e^{\lambda-\nu}(p+\epsilon)r-4\frac{dp}{dr} \bigg \} \nonumber \\ && +\xi\bigg \{\bigg(\frac{dp}{dr}\bigg)^{2}\frac{r}{(p+\epsilon)}-8\pi e^{\lambda}(p+\epsilon)pr \bigg \} \nonumber \\ && +\Delta p\bigg \{\frac{dp}{dr}\frac{1}{(p+\epsilon)} - 4\pi(p+\epsilon)r e^{\lambda}\bigg\}, \end{eqnarray} where $\omega$ is the eigenfrequency and the quantities $\xi \equiv \Delta r / r$ and $\Delta p$ are assumed to have a harmonic time dependence $\varpropto e^{i\omega t}$. To solve equations (\ref{ecuacionparaXI}) and (\ref{ecuacionparaP}) one needs two boundary conditions. The condition of regularity at the centre ($r=0$) can be written as \cite{AA260:250:1992,AA325:217:1997,AA344:117:1999} \begin{equation}\label{DeltaP} (\Delta p)_{center}=-3(\xi \Gamma p)_{center}. \end{equation} where the eigenfunctions are normalized in order to have $\xi(0)=1$. The second boundary condition, expresses the fact that the Lagrangian perturbation in the pressure at the stellar surface is zero, thus: \begin{equation} \label{PenSuperficie} (\Delta p)_{surface}=0. \end{equation} To solve numerically the oscillation equations we employ a shooting method in order to fulfil the required boundary conditions. For more details on the method see Ref. \cite{PRD82:063006:2010}. \section{Results and conclusions} In this section we analyse the effect that a strong magnetic field could produce on the fundamental mode of the radial oscillations of hadronic stars. As mentioned before, we consider that the magnetic field decays with the density following the fast and slow profiles presented in Section \ref{B-density-depndence}. All the models for hadronic stars investigated in the present work have a maximum mass in agreement with the recent observation of the pulsars PSR J1614-2230 with $M = (1.97 \pm 0.04) M_{\odot}$ \cite{Demorest2010} and PSR J0348-0432 with $M = (2.01 \pm 0.04) M_{\odot}$ \cite{Antoniadis2013}. In Fig. \ref{fig1} we see that a magnetic field profile with $B_0 = 10^{17}$ G, produces very small changes on the oscillation period of the fundamental mode with respect to an unmagnetized star, for both slow and fast decays. This can be explained by the small effect that such magnetic field intensity has on the equation of state. In contrast, when $B_0 = 3.1 \times 10^{18}$ G is selected, there is a clear change in the oscillation period. As a function of the stellar mass, the curves fall below and to the right of the curves for weaker fields. Notice that for large mass objects the period changes because of the shift of the curves due to the increase of the maximum stellar mass. For smaller masses, the curves for $B_0 = 3.1 \times 10^{18}$ G are also significantly different with respect to the unmagnetized case; e.g. for a neutron star with $1.4 M_{\odot}$ the period is around $20 \%$ smaller, and the difference increases for less massive stars. For completeness we present also the oscillation period as a function of the gravitational redshift (see middle panel of Fig. \ref{fig1}) and as a function of the central mass-energy density (lower panel of Fig. \ref{fig1}). The effect of strong magnetic fields is more apparent in the frequency of the fundamental mode as can be seen in Fig. \ref{fig2}. The oscillation frequency for $B_0 = 10^{17}$ G is slightly above the one of an unmagnetized object of the same mass, and there is almost no difference between the fast and the slow decaying profiles of the magnetic field. However, if $B_0 = 3.1 \times 10^{18}$ G the oscillation frequencies are clearly larger than for an unmagnetized star of the same mass. For example, for a star with $1.4 M_{\odot}$ the frequency is around $20 \%$ larger and for a $1.7 M_{\odot}$ star it is around $10 \%$ larger. The difference between the fast and the slow decaying profiles of $B$ is very small. As stated before, purely radial modes do not emit gravitational waves and consequently they are essentially damped by the bulk viscosity, originated from the re-establishment of chemical equilibrium when a fluid element of the star is compressed and rarified during pulsations. Unfortunately, there is great uncertainty about the amount of viscosity inside neutron stars since it depends sensitively on the composition of matter which is uncertain beyond few times the nuclear saturation density \cite{Jones2001,Lindblom2002,Haensel2002,Chatterjee2007,Jha2010}. If the damping time due to viscous forces is long enough, radial pulsations in magnetars can leave an imprint in the microstucture of magnetar flare lightcurves opening a new window for the study of highly magnetized ultradense matter. \acknowledgments L. B. Castro thanks CNPq, Brazil, Grants No 455719/2014-4 (Universal) and No 304105/2014-7 (PQ) for partial support. C. V\'asquez Flores acknowledges the financial support of CAPES, Brazil. G. Lugones acknowledges FAPESP and CNPq for financial support. We acknowledge an unknown referee for valuable comments.
1c5ca046551cfa4aa819c11109c3debac9a8e52b
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Ultra-cold molecules can play important roles in studies of many-body quantum physics~\cite{pupillo2008cold}, quantum logic operations~\cite{demille2002quantum}, and ultra-cold chemistry~\cite{ni2010dipolar}. In our recent studies of LiRb, motivated largely by its large permanent dipole moment~\cite{aymar2005calculation}, we have explored the generation of these molecules in a dual species MOT~\cite{dutta2014formation,dutta2014photoassociation}. In particular, we have found that the rate of generation of stable singlet ground state molecules and first excited triple state molecules through photoassociation, followed by spontaneous emission decay, can be very large~\cite{v0paper,Adeel,lorenz2014formation}. There have been very few experimental studies of triplet states in LiRb~\cite{Adeel}, in part because they are difficult to access in thermally-distributed systems. Triplet states of bi-alkali molecules are important to study for two reasons: first, Feshbach molecules, which are triplet in nature, provide an important association gateway for the formation of stable molecules~\cite{marzok2009feshbach}; also, photoassociation (PA) of trapped colliding atoms is often strongest for triplet scattering states. Mixed singlet - triplet states are usually required to transfer these molecules to deeply bound singlet states. \begin{figure}[t!] \includegraphics[width=8.6cm]{PEC.png}\\ \caption{(Color on-line) Energy level diagram of the LiRb molecule, showing relevant PECs from Ref.~\protect\cite{Korek}. Vertical lines show the various optical transitions, including {\bf (a)} photoassociation of atoms to molecular states below the D$_1$ asymptote; {\bf (b)} spontaneous decay of excited state molecules leading to the $a \: ^3 \Sigma ^+$ state; {\bf (c)} RE2PI to ionize LiRb molecules, ($\nu_{c}$ used later in this paper is the frequency of this laser source); and {\bf (d)} state-selective excitation of the $a \: ^3 \Sigma ^+$ state for depletion of the RE2PI signal (with laser frequency $\nu_{d}$). The black dashed line represents our PA states. The inset shows an expanded view of the different $d \ ^3\Pi$ spin-orbit split states as well as the perturbing neighbor $D \ ^1\Pi$.} \label{fig:PEC} \end{figure} \begin{figure*} [t!] \includegraphics[width=\textwidth]{v11Progression.png}\\ \caption{(Color on-line) Subsection of the RE2PI spectra. The PA laser is tuned to the $2(0^-) \ v=-11 \ J=1$ resonance, from which spontaneous decay is primarily to the $a \ ^3\Sigma^+ \ v^{\prime \prime}=11$ state. Most of these lines are $d \ ^3\Pi_{\Omega} \ v^{\prime} \leftarrow a \ ^3\Sigma^+ \ v^{\prime \prime}=11$ transitions, where $v^\prime$ is labeled on individual lines. From top to bottom: black solid lines label transitions to $\Omega=2$, blue dashed lines label transitions to $\Omega=1$, green dot-dashed lines label transitions to $\Omega=0$. Also shown (red dotted lines) are three $ D \ ^1\Pi \ v^\prime \leftarrow a \ ^3\Sigma^+ \ v^{\prime \prime}=11$ transitions.} \label{fig:v11progression} \end{figure*} We show an abbreviated set of potential energy curves (PEC), as calculated in Ref.~\cite{Korek}, in Fig.~\ref{fig:PEC}. The d$^3 \Pi$ - D$^1 \Pi$ complex in LiRb, asymptotic to the Li 2p $^2P_{3/2, 1/2}$ + Rb 5s $^2S_{1/2}$ free atom state, has several features that can promote its utility in stimulated-Raman-adiabatic-passage (STIRAP) and photoassociation. First, the \textit{ab inito} calculations of Ref.~\cite{Korek} predict mixing between low vibrational levels of the $d \ ^3\Pi_1$ and the D$^1 \Pi$ states. Second, both legs of a STIRAP transfer process from loosely bound triplet-character Feshbach molecules to the rovibronic ground state can be driven with commercially-available diode lasers. And third, similar deeply bound $^3 \Pi$ resonances have been successfully used for short-range PA in RbCs ~\cite{RbCs1,RbCs2,RbCs3}. While an interesting discovery on its own, spontaneous decay of these states after PA can populate the $a \ ^3\Sigma^+ \ v^{\prime \prime}=0$ state; one RbCs team~\cite{RbCs2} found spontaneous decay of these states even populated the $X \ ^1\Sigma^+ \ v^{\prime \prime}=0$ state. In the present work, we study the $d \ ^3\Pi_{\Omega}$ states of LiRb, from the asymptote to the most bound vibrational level. We have found signatures of state mixing between low-lying vibrational levels of the d$^3 \Pi_1$ and D$^1 \Pi$ levels. We have determined the term energies of the different spin-orbit components of the $d \ ^3\Pi$ state, as well as their vibrational energies. We have also observed alternation of the intensities of the rotational lines, a possible indication of a $p$-wave resonance in the scattering state, and determined the rotational constants of the lowest vibrational levels. \section{Experiment} We have previously described the details of our experimental apparatus~\cite{DuttaALEC14b}, and provide here only a brief summary. We trap $\sim 5 \times 10^7$ Li atoms and $\sim 2 \times 10^8$ Rb atoms in a dual species magneto-optical trap (MOT), $\lesssim$1 mK in temperature and 1 mm in diameter~\cite{Adeel}. Our Rb MOT is a spatial dark spot MOT~\cite{Ketterle}. We photoassociate Li and Rb atoms to form LiRb molecules using either a 300 mW cw Ti:Sapphire laser or a 150 mW cw external cavity diode laser. After spontaneous decay to a distribution of vibrational levels of the $a \ ^3\Sigma^+$ state, we use two-color resonantly-enhanced-multi-photon-ionization (RE2PI) to ionize the molecules. The lasers that we use to drive the RE2PI process are a Nd:YAG-pumped, pulsed dye laser (PDL) for the first photon, tunable in the wavelength range between 667 nm - 750 nm (14950 - 13300 cm$^{-1}$ frequency range), and part of the 532 nm pump laser for the second photon. The repetition rate of this system is 10 Hz, and it delivers $\sim$1.5 mJ/pulse of dye energy in a 4 mm diameter beam and a larger $\sim$2 mJ/pulse 532 nm beam to the MOT region. When the frequency of the dye laser is resonant with a transition from an initial state (populated through spontaneous decay from the PA state) to an intermediate bound state (in this experiment the $d \ ^3\Pi$ state), absorption of a dye photon and a 532 nm photon ionizes the molecule. We detect the molecular LiRb$^+$ ions using a time-of-flight spectrometer and a microchannel plate detector. In this paper we will adopt the following notation: $v^{\prime \prime}$ and $J^{\prime \prime}$ denote the vibrational and rotational levels of the $a \: ^3 \Sigma ^+$ and $X \ ^1\Sigma^+$states, $v$ and $J$ (without a prime) denote the vibrational and rotational levels of the PA resonances (and for these vibrational numbers, we count down from the asymptote using negative integers), and $v^{\prime}$ and $J^{\prime}$ denote vibrational and rotational labeling of other excited electronic states. \begin{figure*} [t] \includegraphics[width=\textwidth]{v10Progression.png}\\ \caption{(Color on-line) Subsection of REMPI data with the PA laser tuned to the $4(1) \ v=-16 \ J=1$ resonance~\cite{v0paper}, while scanning the REMPI laser frequency, $\nu_c$. Transitions are labeled as $d \ ^3\Pi_{1} \ v^{\prime} \leftarrow X \ ^1\Sigma^+ \ v^{\prime \prime}=10$ (blue dashed) and $D \ ^1\Pi \ v^{\prime} \leftarrow X \ ^1\Sigma^+ \ v^{\prime \prime}=10$ (red dotted). The inset shows confirmation of the assignments with depletion spectroscopy. The orange curve in the inset is the REMPI data retaken in the presence of a depletion laser tuned to the $A \ ^1\Sigma^+ \ v^{\prime}=25 \ J^{\prime}=1 \leftarrow X \ ^1\Sigma^+ \ v^{\prime \prime}=10 \ J^{\prime \prime}=0$~\cite{A1Sigma+} transition; the reduction in peak height confirms the assignments. Because the depletion laser reduces the population available for REMPI in both peaks, they must have the same initial state, from which we conclude that we can access triplet REMPI resonances from singlet states.} \label{fig:v10progression} \end{figure*} We used two techniques, RE2PI and depletion spectroscopy, to measure the $d \ ^3\Pi$ bound states. In RE2PI spectroscopy, we tune the PA laser to either the $v=-11 \ J=1$ or $v=-8 \ J=1$ lines of the $2(0^-)$ long range state, from which spontaneous decay leads primarily to the vibrational levels of the $a \ ^3\Sigma^+$ state~\cite{Adeel}. We count the number of ions detected over the course of 100 laser pulses, and tune the laser frequency $\nu_c$ of the PDL in 0.35 cm$^{-1}$ increments. We record the number of ions detected, normalized by the number of laser pulses, as a function of the PDL frequency $\nu_c$. In order to reach the full range of vibrational levels of the $d \ ^3\Pi$ states, we used two different laser dyes in the PDL. An LDS 698 dye covered the 13950 - 14950 cm$^{-1}$ range, and an LDS 750 dye covered from 13300 to 13950 cm$^{-1}$. These dyes are difficult to work with because of short lifetimes and low power output. The LDS 750 dye in particular was very troublesome: it has a lifetime $\le$8 hours, produces low power ($\le$0.5 mJ/pulse for much of its range) and because it has a very broad pulse width (i.e. lots of spontaneous emission) the baseline noise of our RE2PI spectra is enhanced over what we have observed with other dyes. The $2(0^-) \ v=-11 \ J=1$ PA line at $\nu_a = 12516.89$ cm$^{-1}$ is relatively weak, but it decays almost exclusively to a single vibrational level ($v^{\prime \prime}=11$) of the $a \ ^3\Sigma^+$ state. This facilitates straight-forward identification of the vibrational levels of the intermediate state. Unfortunately, several vibrational levels of the $d \ ^3\Pi$ state do not appear in this spectrum, presumably due to poor Franck-Condon overlap with the $v^{\prime \prime}=11$ state. This problem was even more evident when using the LDS 750 dye in the PDL. For this reason, we collected several RE2PI spectra using the stronger $2(0^-) \ v=-8 \ J=1$ PA resonance $\nu_a = 12557.60$ cm$^{-1}$. This line decays to a wider spread of vibrational levels, giving more complete coverage of the vibrational lines of the $d \ ^3\Pi$ states, but also making our analysis more difficult, due to the increased congestion of the spectra and frequent overlap between individual lines. To explore the deeply-bound levels of the $d \ ^3\Pi$ states, we used a depletion spectroscopy technique. In these measurements, we used the 150 mW ECDL tuned to the $2(0^-) \ v=-5 \ J=1$ PA resonance at $\nu_a = 12575.05$ cm$^{-1}$~\cite{Adeel}. Spontaneous decay of this state populates the $a \ ^3\Sigma^+ \ v^{\prime \prime}=13$ state. We tune the PDL laser frequency to the $(3) \ ^3\Pi_{0} \ v^{\prime}=6 \leftarrow a \ ^3\Sigma^+ \ v^{\prime \prime}=13$ one-color resonantly-enhanced two-photon ionization (REMPI) transition at 17736.6 cm$^{-1}$~\cite{Adeel}. We then tune the frequency of the Ti:Sapphire laser into resonance with bound-to-bound transitions from the $a \ ^3\Sigma^+ \ v^{\prime \prime}=13$ state to ro-vibrational levels in the $d \ ^3\Pi$ state. Exciting these transitions depletes the population of the $a \ ^3\Sigma^+ \ v^{\prime \prime}=13$ state, causing the REMPI signal to decrease. \section{RE2PI Measurements} We show an example of a RE2PI spectrum in Fig.~\ref{fig:v11progression}. Transitions observed in this spectrum are $ d \ ^3\Pi_{\Omega} \ v^{\prime} \leftarrow a \ ^3\Sigma^+ \ v^{\prime \prime}=11$. We have marked the transitions to the $\Omega=2$ progression with black solid lines, $\Omega=1$ with blue dashed lines, and $\Omega=0$ with green dot-dashed lines. $\Omega$ is the total electronic angular momentum, orbital $L$ + spin $S$, projected onto the internuclear axis. The numerical label for each peak is the vibrational number $v^{\prime}$ of the $d \ ^3\Pi_{\Omega}$ state. We have also marked three lines in this spectrum corresponding to transitions to the $D \ ^1\Pi$ state with red dotted lines. From the spectrum of these $d \ ^3\Pi$ states, we observe the typical hierarchy of line spacings: the vibrational splitting is large (on the order of 100 cm$^{-1}$ for low vibrational quantum number $v^\prime$, and decreasing with increasing $v^\prime$), and the spin-orbit splitting between different $\Omega$ states is smaller (on the order of 30 cm$^{-1}$). The rotational splitting for low $J^\prime$ (on the order of 0.1 cm$^{-1}$) is too small to be resolved in these RE2PI spectra since the spectral resolution of the PDL is $\sim$0.5 cm$^{-1}$. The appearance of transitions belonging to vibrational levels of the $D \ ^1\Pi$ electronic state in Fig.~\ref{fig:v11progression} is evidence of mixing between the $D \ ^1\Pi$ and the $d \ ^3\Pi_{1}$ potentials near an avoided crossing between the two states. The energy of these $D \ ^1\Pi$ states is known from Refs.~\cite{Ivanova11,JohnThesis}. State mixing gives these states partial character of each electronic state, which in this case manifests itself through strong transitions from a triplet state (i.e. $a \ ^3\Sigma^+ \ v^{\prime \prime}=11$) to singlet states ($D \ ^1\Pi \ v^{\prime}$ = 6, 7, and 8). This state mixing also adds $D ^1\Pi$ character to the $d \ ^3\Pi_{1}$ states, so one should expect the nearby $d \ ^3\Pi_{1}$ states to appear in the singlet spectra. This expectation is borne out in the spectrum shown in Fig.~\ref{fig:v10progression}. This spectrum is a REMPI scan generated in our system after photoassociating ultracold LiRb molecules through a 2(1) - 4(1) mixed state at $\nu_a = 12574.85$ cm$^{-1}$~\cite{v0paper}, which spontaneously decays to vibrational levels of the $X \ ^1\Sigma^+$ ground electronic state. The spectrum in Fig.~\ref{fig:v10progression} primarily shows transitions to low-lying $D \ ^1\Pi$ vibrational levels from $X \ ^1\Sigma^+ \ v^{\prime \prime}=10$. We also observe in this spectrum $d \ ^3\Pi_{1} \ v^{\prime}=4$, 5, and 6 $\leftarrow X \ ^1\Sigma^+ \ v^{\prime \prime}=10$ transitions. We chose $X \ ^1\Sigma^+ \ v^{\prime \prime}=10$ because it is strongly populated by spontaneous decay of the 2(1) - 4(1) PA resonance and transitions to deeply bound $D \ ^1\Pi$ vibrational levels are clearly identified. We can estimate the degree of mixing based on the relative strength of the different REMPI peaks. The $D \ ^1\Pi \ v^\prime = 7 \leftarrow X \ ^1\Sigma^+ \ v^{\prime \prime}=10$ transition is twice as strong as the $d \ ^3\Pi_1 \ v^\prime = 5 \leftarrow X \ ^1\Sigma^+ \ v^{\prime \prime}=10$ transition so there is twice as much singlet character to $D \ ^1\Pi \ v^\prime = 7$ as $d \ ^3\Pi_1 \ v^\prime = 5$. Following the same procedure that we have used in the past~\cite{v0paper}, we can estimate the interaction strength to be about 7 cm$^{-1}$. Interestingly, this rough estimate is consistent with the following simple perturbative argument. The spin-orbit interaction responsible for the state mixing can be estimated as about one half the spin orbit mixing in atomic rubidium~\cite{WeidemullerReview}, or about 120 cm$^{-1}$. The Franck - Condon factor (FCF) between $D \ ^1\Pi \ v^\prime = 7$ and $d \ ^3\Pi_1 \ v^\prime = 5$, as calculated by LEVEL 8.0~\cite{LEVEL} using the PEC from~\cite{Korek}, is about 0.08 and thus the strength of interaction between these states should be approximately 10 cm$^{-1}$. We have applied this perturbative analysis to each of the vibrational levels of the $d \ ^3\Pi_{1} $ state (not including $v^{\prime}=5$), and find that each contains some small component of $D \ ^1\Pi$ perturber state, on the order of 10\% or smaller. This is too small to be seen in the spectra of Fig.~\ref{fig:v11progression}, but could be sufficient to be useful in a Raman or STIRAP transfer of population to low lying levels of the electronic ground state in the future. Many of the RE2PI spectra that we collected are less clear than that shown in Fig.~\ref{fig:v11progression}. In particular, the peaks in the RE2PI spectra near the Rb 5S + Li 2P asymptote are strong, but line congestion becomes significant, and clear identification of the lines in this region becomes difficult. These assignments could probably be improved using a spectroscopic technique that is capable of higher spectral resolution, such as photoassociative spectroscopy, but this was beyond the scope of the present work. Assigning peaks in the RE2PI spectra was equally difficult for deeply bound vibrational states (i.e. $v^{\prime}$ $\le$ 4). In fact we were unable to observe a clear cut off in our RE2PI data corresponding to $v^{\prime}=0$. We attribute this to difficulties with the LDS 750 dye, specifically the large spontaneous emission content in the pulse. To rectify this problem, we turned to a second set of measurements, based upon depletion spectroscopy. \section{Depletion Measurements} \begin{figure} [b] \includegraphics[width=8.6cm]{v13depletion.png}\\ \caption{Depletion spectra of an $a \ ^3\Sigma^+ \ v^{\prime \prime}=13$ REMPI line using the $d \ ^3\Pi \ v^{\prime}=0$ state. The PA laser is locked to the $2(0^-) \ v=-5, \ J=1$ line, the REMPI laser is tuned to the $(3) \ ^3\Pi_{0} \ v^{\prime}=6 \leftarrow v^{\prime \prime}=13$ transition at $\nu_c = 17736.6$ cm$^{-1}$. Panel (a) shows a global view of our depletion data. Panels (b) - (d) show the rotational structure of the depletion lines. All repeatable transitions are labeled $J^{\prime} \leftarrow K^{\prime \prime}$. For transitions $ J^\prime = 0$ or 1 $\leftarrow K^{\prime \prime} = 1$ (red lines), $J^{\prime \prime}=0$ or 1 are possible. For transitions $J^\prime = 2 \leftarrow K^{\prime \prime} = 1$ (green lines), $J^{\prime \prime}$ can only be 1. The abscissa of (b) - (d) is the depletion laser frequency offset by $E_{depletion} = T_{0, \ \Omega} - T_{13} - 2B_{13}$ where $T_{0, \ \Omega}$ is the rotationless energy of the $d \ ^3\Pi_{\Omega} \ v^{\prime}=0$ state, $T_{13}$ and $B_{13}$ are the binding energy and the rotational constant of the $a \ ^3\Sigma^+ \ v^{\prime \prime}=13$ state respectively.} \label{fig:depletion} \end{figure} \begin{table*}[t] \centering \begin{tabular}{ccccccccc} \hline \hline & \multicolumn{2}{c}{$d \ ^3\Pi_{0^+}$}& \multicolumn{2}{c}{$d \ ^3\Pi_{0^-}$} &\multicolumn{2}{c}{$d \ ^3\Pi_{1}$} &\multicolumn{2}{c}{$d \ ^3\Pi_{2}$} \\ $v^{\prime}$ & $T_v$ ($cm^{-1}$) & $B_{v}$ ($cm^{-1}$) & $T_v$ ($cm^{-1}$) & $B_{v}$ ($cm^{-1}$) & $T_v$ ($cm^{-1}$) & $B_{v}$ ($cm^{-1}$)& $T_v$ ($cm^{-1}$) & $B_{v}$ ($cm^{-1}$)\\ \midrule[1.pt] 0 & 13507.9 & 0.148 (4) & 13508.8 & & 13544.7 & 0.153 (6)& 13577.6 & \\ 1 & 13606.8 & 0.146 (4) & 13607.6 & & & & & \\ \hline \hline \end{tabular} \caption{Experimental assignments for $T_v$, the rotationless energy, and $B_v$, the rotational constant, of the vibrational levels of the $d \ ^3\Pi$ state based on our depletion data for $v^{\prime}=0$ and 1. Uncertainty for all $T_v$ is 0.5 cm$^{-1}$, and uncertainty in the last digit in $B_v$ is given in parentheses. Blank entries denote rotational constants or energies that we were not able to measure because of either the tuning range of our Ti:Sapphire laser or because of bound to bound selection rules limited by our PA state.} \label{tab:depletionAssignments} \end{table*} We used depletion spectroscopy to identify the lowest two vibrational levels of the $d \ ^3\Pi$ state. Compared to RE2PI, depletion spectra are sparse and have narrow peak widths, in this case $\sim$1 GHz, a typical linewidth for this type of measurement at this intensity~\cite{Deiglmayr}. Additionally, depletion spectra allow us to extract some rotational constants of the lines, a very useful tool for comparing experiment to theory. We show depletion spectra for $v^{\prime}=0$ in Fig.~\ref{fig:depletion}. To assign these data care must be taken with selection rules for radiative transitions in molecules. The two that apply here are: $\Delta J= 0, \ \pm1$ and $- \leftrightarrow +$, that is positive symmetry states (with respect to coordinate inversion) must transition to negative symmetry states and vise versa. The initial state in this depletion transition is a $^3\Sigma^+$ state, which is a strict Hund's case (b) state. As such, its rotational energy is determined by quantum number $K$, which designates the total angular momentum of the molecule apart from spin, rather than the total angular momentum (including spin) quantum number $J$. For this $a ^3\Sigma^+$state, the electronic spin is $S$=1, and levels with $J=K+1, K$ and $K-1$, are nearly degenerate for $K\ge 1$. Additionally $K$ determines the symmetry of the state. This is summarized in Fig.~\ref{fig:parity}, adapted and modified from Ref.~\cite{Herzberg}. \begin{figure} [t] \includegraphics[width=8.6cm]{parity.png}\\ \caption{Rotational structure and parity of $^3\Sigma^+$ and $^3\Pi$ vibrational states, the two states in red and green highlight the ground states we populate at the beginning of the depletion process. Transitions from the populated ground states to $^3\Pi_{0^{+}}$ and $^3\Pi_{0^{-}}$ are shown by arrows. Transitions to the positive symmetry levels of $^3\Pi_{1}$ and $^3\Pi_{2}$ do occur, but are omitted for clarity. The rotational splitting in $^3 \Sigma^+$ is determined by $E=B_vK^{\prime \prime}(K^{\prime \prime}+1)$ (which makes the $J$ manifold within each $K^{\prime \prime}$ state degenerate) and in $^3 \Pi$ by $E=B_v[J(J+1)-\Omega^2]$. Adapted and modified from Ref.~\cite{Herzberg}} \label{fig:parity} \end{figure} The logic that leads to our assignments in Fig.~\ref{fig:depletion} goes as follows. We start with the depletion data on transitions to $\Omega=1$ shown in Fig.~\ref{fig:depletion}(c), which is extensive enough to show that we do not populate $J^{\prime \prime}=2$ of the $a \ ^3\Sigma^+$ state; that is, we see no depletion signal corresponding to a transition to $J^{\prime} = 3$. Since we do see transitions to $J^{\prime} = 1$ and $2$, we know that we populate some mix of $J^{\prime \prime}=0$ and $1$ belonging to either $K^{\prime \prime}=0$, 1, or 2. $K^{\prime \prime}=0$ and $K^{\prime \prime}=2$ have the same symmetry and some of the same rotational numbers; this implies that any PA state that could decay to $K^{\prime \prime}=0$ or 2 could decay to the other as well. Because the energies of $K^{\prime \prime}=0$ and $K^{\prime \prime}=2$ differ from one another, but we see no additional structure in the depletion spectra, we infer that the transitions seen in Fig.~\ref{fig:depletion}(c) originate from a single state, that is $K^{\prime \prime}=1$. The spacing between the two peaks in this spectrum should be $4B_0$, where $B_0$ is the rotational constant of the $d \ ^3\Pi \ v^\prime = 0$ state, allowing us to determine $B_0 = 4.59$ GHz. This rotational constant agrees with the prediction from LEVEL 8.0 with PECs from Ref.~\cite{Korek}, which confirms our assignments and guides our interpretation of the $\Omega=0$ data shown in Fig.~\ref{fig:depletion}(d), in which we see three transitions. The first two are spaced by $6B_0$ (27.5 GHz), implying that these peaks are transitions to the positive symmetry levels of the $\Omega=0^+$ electronic state, $J^{\prime}=0 \leftarrow K^{\prime \prime}=1$ and $J^{\prime}=2 \leftarrow K^{\prime \prime}=1$. The absence of a peak for $J^{\prime}=1$ is consistent with the selection rule $- \not\leftrightarrow -$. The remaining transition in $\Omega=0$ must be the only allowed transition to $\Omega=0^-$, that is $J^{\prime}=1 \leftarrow K^{\prime \prime}=1$. There is only one transition in $\Omega=2$, which is trivial to identify as $J^{\prime}=2 \leftarrow K^{\prime \prime}=1$. This picture is consistent with what we know about our PA state. That is, we PA into the $J=1$ level of a $2(0^-)$ state, which has positive symmetry, and the allowed decay paths are to $K^{\prime \prime}=1$ (with $J^{\prime \prime}=0$, $1$, and $2$, each with negative symmetry, of which we populate $0$ and $1$). Of note, we chose to PA to $J=1$ because it is the strongest PA resonance in the $2(0^-) \ v=-5$ progression. Our depletion data confirms that the parity of this PA state is even, which is suggestive of a scattering state $p$-wave shape resonance~\cite{d3PiPA}. Similar shape resonances have been seen in other bi-alkali's~\cite{NaCs,Deiglmayr}. We used these data to determine the spin-orbit splitting between the different $\Omega$ progressions deep in the $d \ ^3\Pi$ well, and followed these progressions back to the asymptote in our RE2PI data. Most importantly, these data provide accurate locations of $v^{\prime}=0$ and $1$, $J^{\prime}$ for future work on short range PA~\cite{d3PiPA}. The depletion data is limited by the shot noise in our ion counting. For each data point we integrate 200 shots and usually count $\sim$350 ions. To get 2$\sigma$ resolution the smallest depletion feature that we can be confident in has to be a 10\% decrease. Here our on-resonance depletion signal drops by around 30\% which is 6$\sigma$ and statistically significant. Unfortunately, our Ti:Sapphire struggles to tune to wavelengths much shorter than 740 nm, which limited our depletion spectra to the lowest two vibrational levels only. Despite these short comings, depletion spectroscopy gives us accurate identifications of these bottom two vibrational levels of the $d \ ^3\Pi$ electronic state and unambiguously determines the spin orbit splitting. \begin{table*}[t] \centering \begin{tabular}{ccccccc} \hline \hline & \multicolumn{2}{c}{$d \ ^3\Pi_0$} &\multicolumn{2}{c}{$d \ ^3\Pi_1$} &\multicolumn{2}{c}{$d \ ^3\Pi_2$} \\ $v^{\prime}$ & $T_v$ ($cm^{-1}$) & $\Delta$E ($cm^{-1}$) & $T_v$ ($cm^{-1}$) & $\Delta$E ($cm^{-1}$) & $T_v$ ($cm^{-1}$) & $\Delta$E ($cm^{-1}$)\\ \midrule[1.5pt] 0 & 13508.2 & 98.8 & 13545.2 & 99.1 & 13578.1 & 99.3\\ 1 & 13607.0 & 94.9 & 13644.3 & 95.4 & 13677.4 & 92.0\\ 2 & 13701.9 & 94.8 & 13739.7 & 92.0 & 13769.4 & 96.0\\ 3 & 13796.7 & 87.3 & 13831.7 & 89.6 & 13865.4 & 88.1\\ 4 & 13883.9 & 83.8 & 13921.3 & 83.5 & 13953.5 & 86.1\\ 5 & 13967.7 & 80.5 & 14004.8 & 81.2 & 14039.6 & 80.9\\ 6 & 14048.2 & 79.0 & 14086.0 & 78.1 & 14120.5 & 77.5\\ 7 & 14127.2 & 74.8 & 14164.1 & 75.1 & 14198.0 & 75.2\\ 8 & 14202.0 & 71.2 & 14239.2 & 71.6 & 14273.2 & 71.5\\ 9 & 14273.2 & 71.5 & 14310.8 & 68.7 & 14344.7 & 68.0\\ 10 & 14344.7 & 66.3 & 14379.5 & 64.8 & 14412.5 & 64.1\\ 11 & 14411.0 & 64.2 & 14444.3 & 65.2 & 14476.6 & 63.0\\ 12 & 14475.2 & 62.1 & 14509.5 & 59.4 & 14539.6 & 58.8\\ 13 & 14537.3 & 59.4 & 14568.9 & 56.9 & 14598.4 & 55.8\\ 14 & 14596.7 & 55.1 & 14625.8 & 54.7 & 14654.2 & 51.4\\ 15 & 14651.8 & 51.4 & 14680.5 & 51.4 & 14705.6 & 46.2\\ 16 & 14703.2 & 47.6 & 14731.9 & 44.3 & 14751.4 & 43.4\\ 17 & 14750.8 & 47.0 & 14776.2 & 40.1 & 14795.2 & 37.9\\ 18 & 14797.8 & 45.9 & 14816.3 & 35.5 & 14833.1 & 32.5\\ 19 & 14843.7 & 25.2 & 14851.8 & 24.4 & 14865.6 & 23.9\\ 20 & 14868.9 & 21.3 & 14876.2 & 16.6 &14889.5 & 8.7\\ 21 & 14890.2 & 10.7 & 14892.8 & 6.9 & 14898.2 & \\ 22 & 14900.9 & & 14899.7 & & & \\ \hline \hline \end{tabular} \caption{Experimental assignments for the rotationless energy of the vibrational levels of the $d \ ^3\Pi$ state based on our RE2PI data, aided by our depletion data for $v^{\prime}=0$ and 1. Uncertainty for all assignments is 0.5 cm$^{-1}$. We have referenced the term energies, $T_v$, to the Rb 5S + Li 2S asymptote.} \label{tab:masterAssignments} \end{table*} \section{Discussion} \begin{figure} [b] \includegraphics[width=8.6cm]{vibrationalSplitting.png}\\ \caption{(Color on-line) Comparison of our extracted vibrational splitting to predicted vibrational splitting. The circles represent our data, while the triangles are predicted by \textit{ab initio} curves~\cite{Korek}. Green markers label $\Omega=0$ (compared to \textit{ab initio} $\Omega=0^+$) spacings. Blue markers are shifted by +20 cm$^{-1}$ and label $\Omega=1$ spacings. Black markers are shifted by +40 cm$^{-1}$ and label $\Omega=2$ spacings.} \label{fig:vibrationalSplitting} \end{figure} We determine the vibrational binding energies and rotational constants of the states seen in depletion spectroscopy, which we tabulate in Table.~\ref{tab:depletionAssignments}. We list in Table~\ref{tab:masterAssignments} the assignments and energy of each of the $d \ ^3\Pi$ states that we observe through RE2PI and depletion spectroscopy. We also include in this table the energy difference between adjacent states, which aids in the assignment of the lines. The theoretical vibrational levels and spin-orbit splittings that we used to guide our work and for comparison of results come from \textit{ab initio} calculations by Korek et al.~\cite{Korek} with aid from LEVEL 8.0~\cite{LEVEL}. We found good overall agreement with these \textit{ab initio} results. The spin-orbit splittings for $\Omega=0$ to $\Omega=1$, predicted to be 21 cm$^{-1}$, are measured here to be 37 cm$^{-1}$; for $\Omega=1$ to $\Omega=2$, they are predicted to be 38 cm$^{-1}$, and we found them to be 33 cm$^{-1}$. For the spin-orbit splitting between the $\Omega=0^{+}$ to $\Omega=0^{-}$ states, however, we observe 0.9 cm$^{-1}$, significantly less than the predicted 36 cm$^{-1}$. Our depletion data is unambiguous in establishing the $\Omega=0^{+}$ to $\Omega=0^{-}$ splitting, and a small $\Omega=0^{+}$ to $\Omega=0^{-}$ splitting is consistent with our observations in the $(3) \ ^3\Pi$ state~\cite{Adeel}. \begin{table*}[t] \centering \begin{tabular}{ccccccc} \hline \hline \multirow{2}{*}{} & \multicolumn{2}{c}{$d \ ^3\Pi_0$} & \multicolumn{2}{c}{$d \ ^3\Pi_1$} & \multicolumn{2}{c}{$d \ ^3\Pi_2$} \\ & Exp. & Th. & Exp. & Th. & Exp. & Th. \\ \midrule[1.5pt] $T_e$ (cm$^{-1}$) & 13459.2 (1.5) & 13300.7 & 13497.5 (2.0) & 13359.9 & 13528.0 (1.4) & 13398.2\\ $\omega_e$ (cm$^{-1}$) & 101.4 (0.7) & 103.6 & 100.4 (0.9) & 102.8 & 101.7 (0.6) & 102.4\\ $x_e$ ($10^{-3}$) & 16.7 (0.8) & 13.1 & 15.0 (1.0) & 12.7 & 15.8 (0.7) & 12.5\\ $y_e$ ($10^{-3}$) & 0.068 (0.028) & -0.06 & -0.038 (0.033) & -0.10 & -0.036 (0.026) & -0.12\\ \hline \hline \end{tabular} \caption{Molecular vibrational constants fitted to our data, $T(v) = T_e + \omega_e (v + 1/2) - \omega_e x_e (v + 1/2)^2 + \omega_e y_e (v + 1/2)^3$ where $T(v)$ is the rotationless energy of the $v^{th}$ vibrational level. The uncertainty is given in parentheses. The theory values are from fitting the bound states calculated by LEVEL 8.0 using PECs from Ref.~\cite{Korek}. When fitting the experimental data, we used only the $v^{\prime}$=0-19 to increase the accuracy.} \label{tab:molecularConstants} \end{table*} We show the vibrational spacing, $\Delta E = E_{v+1} - E_v$ vs $v$, of the different series in Fig.~\ref{fig:vibrationalSplitting}. These data are in reasonable agreement with the predicted vibrational splittings although there appears to be a nearly uniform difference of a few cm$^{-1}$. We found that the depth of the $d \ ^3\Pi$ potential (exp. value) is less than that predicted (th. value). We looked extensively for another vibrational level below our assigned $v^{\prime}=0$ level. We covered $\pm$ 10 cm$^{-1}$ around the expected vibrational location with our depletion spectra, but found no indication of a depletion resonance. Our extracted molecular constants are listed in Tab.~\ref{tab:molecularConstants}. These provide an easy estimation of the spectral structure of the $d \ ^3\Pi$ states as well as a quick comparison to theoretical predictions. As borne out in Fig.~\ref{fig:vibrationalSplitting}, there is good agreement between our fitted harmonic constant, $\omega_e$, and the predictions. However, there is considerably less agreement between our extracted term energy, $T_e$, and the predictions for reasons discussed previously. Additionally, it is important to note that when we fitted the experimental data to determine $T_e$, $\omega_e$, $x_e$ and $y_e$ we used only $v^{\prime}$=0-19. This increased the accuracy of the fit so that for these vibrational levels our molecular constants reproduce our data with a standard deviation of 2 cm$^{-1}$. We believe most of the deviations are caused by experimental uncertainties on determining the frequencies of the peaks, as well as small perturbations to state locations caused by spin-orbit mixing. \section{Conclusion} In this work we have found and identified the $v=0$ to $v=22$ states of the $d \ ^3\Pi_{\Omega}$ electronic state in LiRb. We explored singlet - triplet mixing to evaluate the possibility of using the $d \ ^3\Pi$ - $D \ ^1\Pi$ mixed states as the intermediate state in a STIRAP transfer from a weakly bound triplet state to deeply bound singlet states. We know from heat pipe spectra~\cite{Ivanova11} and other REMPI data~\cite{Lorenz14} that the transitions from deeply bound $D \ ^1\Pi \ v^{\prime}$ to deeply bound $X \ ^1\Sigma^+$ states are strong. Our depletion data shows that transitions from weakly bound triplet molecules to deeply bound $d \ ^3\Pi_1 \ v^\prime$ states are also strong. Finally, our RE2PI data demonstrate that there is about 10\% mixing between most of these singlet and triplet states. We suggest using the $d \ ^3\Pi_1 \ v^\prime = 0 \ J^\prime = 1$ state as the intermediate state for a STIRAP transfer in future work. The laser wavelengths for this transfer would be 740 nm, achievable with Ti:Sapphire or diode lasers, and 516 nm, which is accessible with green laser diodes. Using calculated FCFs and an estimate of the transition dipole moment of a few times $e a_0$, the `up' transition would be around $4 \times 10^{-2}$ $e a_0$ and the `down' transition would be around $10^{-2}$ $e a_0$ which is competitive with the transfer strength used by the KRb JILA team~\cite{Ni08}. Using the binding energy of the lowest several vibrational levels we have assessed their potential for short-range photoassociation. There are only six possible PA transitions we can observe with our current Ti:Sapphire laser, the same six we saw in depletion. However, our depletion data provides the perfect stepping stone for this type of work, providing very trusted locations of vibrational levels (to within 0.5 cm$^{-1}$, the uncertainty in the binding energy of $a \ ^3\Sigma^+ \ v^{\prime \prime}=13$). Of particular interest to us is the combination of short range PA and mixing between $d \ ^3\Pi$ - $D \ ^1\Pi$. This provides an avenue through simple spontaneous decay to the rovibronic ground state although we will need a different laser to access these PA resonances. We are happy to acknowledge useful conversations with Jes\'{u}s P\'{e}rez-R\'{\i}os, and university support of this work through the Purdue OVPR AMO incentive grant. And we would like to acknowledge the work done by S. Dutta and J. Lorenz in building the LiRb machine.
d95635bc08fd14ade0a89cf08420e98ffd0f6b8b
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{sec:introduction} The use of GPS for localizing sensor nodes in a sensor network is considered to be excessively expensive and wasteful, also in some cases intractable, \cite{bul:00,bis:06}. Instead many solutions for the localization problem tend to use inter-sensor distance or range measurements. In such a setting the localization problem is to find unknown locations of say $N$ sensors using existing noisy distance measurements among them and to sensors with known locations, also referred to as anchors. This problem is known to be NP hard \cite{mor:97}, and there have been many efforts to approximately solve this problem, \cite{kim:09,bis:04,wan:06,bis:06,gho:13,nar:14,cha:09,soa:15,sch:15,sri:08}. One of the major approaches for approximating the localization problem, has been through the use of convex relaxation techniques, namely semidefinite, second-order and disk relaxations, see e.g., \cite{kim:09,bis:06,bis:04,wan:06,sri:08,gho:13,soa:15}. Although the centralized algorithms based on the these approximations reduce the computational complexity of solving the localization problem, they are still not scalable for solving large problems. Also centralized algorithms are generally communication intensive and more importantly lack robustness to failures. Furthermore, the use of these algorithms can become impractical due to certain structural constraints resulting from, e.g., privacy constraints and physical separation. These constraints generally prevent us from forming the localization problem in a centralized manner. One of the approaches to evade such issues is through the use of scalable and/or distributed algorithms for solving large localization problems. These algorithms enable us to solve the problem through collaboration and communication of several computational agents, which could correspond to sensors, without the need for a centralized computational unit. The design of distributed localization algorithms is commonly done by first reformulating the problem by exploiting or imposing structure on the problem and then employing efficient optimization algorithms for solving the reformulated problem, see e.g., some recent papers \cite{sim:14,gho:13,soa:15,sri:08}. For instance, authors in \cite{sri:08} put forth a solution for the localization problem based on minimization the discrepancy of the squared distances and the range measurements. They then propose a second-order cone relaxation for this problem and apply a Gauss-Seidel scheme to the resulting problem. This enables them to solve the problem distributedly. The proposed algorithm does not provide a guaranteed convergence and at each iteration of this algorithm, each agent is required to solve a second-order cone program, SOCP, which can potentially be expensive. Furthermore, due to the considered formulation of the localization problem, the resulting algorithm is prone to amplify the measurement errors and is sensitive to outliers. In \cite{sim:14}, the authors consider an SDP relaxation of the maximum likelihood formulation of the localization problem. They further relax the problem to an edge-based formulation as suggested in \cite{wan:06}. This then allows them to devise a distributed algorithm for solving the reformulated problem using alternating direction method of multipliers (ADMM). Even though this algorithm has convergence guarantees, each agent is required to solve an SDP at every iteration of the algorithm. In order to alleviate this, authors in \cite{gho:13} and \cite{soa:15} consider a disk relaxation of the localization problem and which correspond to an under-estimator of the original problem. They then use projection-based methods and Nestrov's optimal gradient method, respectively, for devising distributed algorithms for solving the resulting problem. These algorithms rely on finding a solution that lies in the intersection of the disks or spheres defined by the range measurements. Consequently, the computational demand on each agent for these algorithms is far less than the aforementioned algorithms. These algorithms commonly work well when there are many range measurements available and their performance is adversely affected if the number of measurements are decreased. Moreover, for the case of low quality, particularly biased, measurements, the convergence of the algorithms can be interrupted as the intersection can be empty. The proposed algorithms in the aforementioned papers have been shown to be effective in analyzing large-scale localization problems. However, all these methods rely on first-order optimization algorithms and hence can require many iterations and communications to converge to an accurate enough solution. Furthermore, the number of iterations can vary significantly with different realizations of range measurements and changing topology of the sensor network. In this paper we show that in case it is possible to provide a tree representation of the inter-sensor range measurement graph of the sensor network (which is the case in many scenarios with few available range measurements), it is possible to alleviate these issues by devising far more efficient distributed localization algorithms that purely rely on second-order methods. \subsection*{Contributions} In this paper, we consider the localization problem for sensor networks where we have access to few range measurements among sensors. The availability of range measurements among $N$ sensors can be described using a graph with $N$ vertices or nodes and an edge between two nodes if there exists a range measurement between them. We refer to this graph as the inter-sensor measurement graph. For our purpose this graph is connected but sparse, i.e., it has few edges. For these sensors networks, it is commonly possible to represent the graph using a tree. We here propose a distributed localization algorithm based on the semidefinite relaxation of the localization problem \cite{sim:14,kim:09}. This algorithm relies on second-order methods, particularly state-of-the-art primal-dual interior-point methods, \cite{wri:97,tod:96,kho:15c,kho:15d}, and is obtained by distributing the computations of each iteration of the primal-dual method among several computational agents. This is done by first clustering the sensor nodes and providing a tree representation of inter-sensor measurement graph. The tree representation then allows us to use message-passing or dynamic programming over trees, \cite{kol:09,kho:15c,kho:15d,ber:73}, to compute the search directions at every iterations of the primal-dual methods distributedly by performing an upward-downward pass through the aforementioned tree. Consequently, and since primal-dual methods commonly converge within 20-50 iterations, our proposed algorithm in comparison to existing ones requires far fewer iterations and communications among agents to converge to a solution. Furthermore, the computational burden for each agent at each iteration only concerns factorizing a relatively small matrix, c.f., \cite{sim:14,sri:08}. \subsection*{Outline} In Section \ref{sec:MLL} we review a maximum-likelihood formulation of the localization problem. Section \ref{sec:scattered} provides a formal description of tree-structured scattered sensor networks and describes how the structure in the problem can be reflected in the localization optimization problem. Section \ref{sec:chordal} reviews how certain structure in nonlinear SDPs enable us to utilize domain-space decomposition to decompose them. This decomposition technique is then used in Section \ref{sec:decompositionloc} to decompose the localization optimization problem. In this section we also describe how the decomposed problem can be written as coupled SDP. We then put forth a generic description of primal-dual interior-point methods in Section \ref{sec:DPDIPM} and show how they, in combination with message-passing, can be used to devise efficient distributed solvers for the localization problems. In this section we also discuss the computational and communication complexity of the proposed distributed algorithm. The numerical experiments are presented in Section~\ref{sec:numerical}, and we conclude the paper with final remarks in Section~\ref{sec:conclusions}. \subsection*{Notations and Definitions}\label{sec:notation} We denote by $\mathbb R$ the set of real scalars and by $\mathbb R^{n\times m}$ the set of real $n\times m$ matrices. The set of $n \times n$ symmetric matrices are represented by $\mathbf S^n$. The transpose of a matrix $A$ is denoted by $A^T$ and the column and null space of this matrix is denoted by $\mathcal{C}(A)$ and $\mathcal N(A)$, respectively. We denote the set of positive integers $\{1,2,\ldots,p\}$ with $\mathbb{N}_p$. Given a set $J \subset \mathbb{N}_n$, the matrix $E_J \in \mathbb{R}^{|J|\times n}$ is the $0$-$1$ matrix that is obtained by deleting the rows indexed by $\mathbb{N}_n \setminus J$ from an identity matrix of order $n$, where $|J|$ denotes the number of elements in set $J$. This means that $E_Jx$ is a $|J|$- dimensional vector with the components of $x$ that correspond to the elements in $J$, and we denote this vector with $x_J$. Also $e_j$ denotes a 0--1 $n$-dimensional vector with only a nonzero element at the $j$th component. Similarly, given $J \subset \mathbb N_n$, $e_J$ denotes a 0--1 $n$-dimensional vector with ones at elements specified by $J$. With $x^{i,(k)}_l$ we denote the $l$th element of vector $x^i$ at the $k$th iteration. Also given vectors $x^i$ for $i= 1, \dots, N$, the column vector $(x^1, \dots, x^N)$ is all of the given vectors stacked. For a vector $x$, with $\diag(x)$ we denote a diagonal matrix with its diagonal elements given by $x$. Similarly, given matrices $X^i$ for $i = 1, \dots, N$, with $\blkdiag(X^1, \dots, X^N)$ we denote a block-diagonal matrix with diagonal blocks given by each of the given matrices. For a matrix $X \in \mathbb R^{m\times n}$, $\vectorize(X)$ is an $mn$-dimensional vector that is obtained by stacking all columns of $X$ on top of each other. Given a symmetric matrix $X \in \mathbf S^n$ \begin{multline*} \svec(X) := (X_{11}, \sqrt{2}X_{21}, \dots, \sqrt{2} X_{n1}, X_{22},\\ \sqrt{2} X_{32}, \dots, \sqrt{2} X_{n2}, \dots, X_{nn}). \end{multline*} Also for a square matrix $X \in \mathbb R^{n\times n}$ we denote with $\vectri(X)$ a column vector which includes all elements on the upper triangle of $X$ stacked. Given two matrices $X$ and $Y$ by $X\otimes Y$ we denote the standard Kronecker product. Given $X \in \mathbf S^n$, define $U$ as an $n(n + 1)/2 \times n^2$ matrix such that $U \vectorize(X) = \svec(X)$. Then for two matrices $X, Y \in \mathbb R^{n\times n}$, $\otimes_s$ denotes the symmetrized Kronecker product that is defined as \begin{align*} X \otimes_s Y := \frac{1}{2} U(X \otimes Y + Y \otimes X)U^T. \end{align*} For properties of the symmetrized Kronecker product refer to \cite{tod:96}. A graph is denoted by $Q(V,\mathcal E)$ where $V = \{1, \dots, n\}$ is its set of vertices or nodes and $\mathcal E \subseteq V\times V$ denotes its set of edges. Vertices $i, j \in V$ are adjacent if $(i, j) \in E$, and we denote the set of adjacent vertices of $i$ by $\Ne(i) = \{ j \in V | (i, j) \in\mathcal E \}$. A graph is said to be complete if all its vertices are adjacent. An induced graph by $V^\prime \subseteq V$ on $Q(V,\mathcal E)$, is a graph $Q_I(V^\prime,\mathcal E^\prime)$ where $\mathcal E^\prime = \mathcal E\cap V^\prime \times V^\prime$. A clique $C_i$ of $Q(V,\mathcal E)$ is a maximal subset of $V$ that induces a complete subgraph on $Q$, i.e., no clique is properly contained in another clique, \cite{blp:94}. Assume that all cycles of length at least four of $Q(V,\mathcal E)$ have a chord, where a chord is an edge between two non-consecutive vertices in a cycle. This graph is then called chordal \cite[Ch. 4]{gol:04}. It is possible to make graphs chordal by adding edges to the graph. The resulting graph is then referred to as a chordal embedding. Let $\mathbf C_Q = \{ C_1, \dots, C_q \}$ denote the set of its cliques, where $q$ is the number of cliques of the graph. Then there exists a tree defined on $\mathbf C_Q$ such that for every $C_i, C_j \in\mathbf C_Q$ where $i \neq j$, $C_i \cap C_j$ is contained in all the cliques in the path connecting the two cliques in the tree. This property is called the clique intersection property, \cite{blp:94}. Trees with this property are referred to as clique trees. \section{Maximum Likelihood Localization}\label{sec:MLL} In this paper we consider a localization problem for a network of $N$ sensors distributed in an area in presence of $m$ anchors. The exact locations of these sensors, $x^i_s$, are deemed to be unknown however we assume that the positions of the anchors, $x^i_a$, are given. Furthermore, the sensors are capable of performing computations and some can measure their distance to certain sensors and some of the anchors. We assume that if sensor $i$ can measure its distance to sensor $j$ so can sensor $j$ measure its distance to sensor $i$. This then allows us to describe the range measurement availability among sensors using an undirected graph $G_r(V_r, \mathcal E_r)$ with vertex set $V_r = \{ 1, \dots, N \}$ and edge set $\mathcal E_r$. An edge $(i,j) \in \mathcal E_r$ if and only if a range measurement between sensors $i$ and $j$ is available. We refer to this graph as inter-sensor measurement graph and assume that it is connected. Let us define the set of neighbors of each sensor $i$, $ \text{Ne}_r(i)$, as the set of sensors to which this sensor has an available range measurement. In a similar fashion let us denote the set of anchors to which sensor $i$ can measure its distance to by $\text{Ne}_a(i) \subseteq \{ 1, \dots, m\}$. Let us describe the inter-sensor range measurements for each sensor, $i \in \mathbb N_N$, as \begin{align}\label{eq:InterSensorMeasurement} \mathcal R_{ij} = \mathcal D_{ij} + E_{ij}, \quad j \in \text{Ne}_r(i), \end{align} where $\mathcal D_{ij} = \| x_s^i - x_s^j \|_2 $ defines the noise-free sensor distance, $E_{ij}$ is the inter-sensor measurement noise and $E_{ij} \sim P_{ij}^s(\mathcal D_{ij}|\mathcal R_{ij})$ with $P_{ij}^s(\cdot)$ being the so-called inter-sensor sensing probability density function (PDF). We here make the standard assumption that $\mathcal R_{ij} = \mathcal R_{ji}$, see e.g., \cite{soa:15,shi:10}. Similarly we can describe the anchor range measurements for each sensor $i$ as \begin{align}\label{eq:AnchorSensorMeasurement} \mathcal Y_{ij} = \mathcal Z_{ij} + V_{ij}, \quad j \in \text{Ne}_a(i), \end{align} where $\mathcal Z_{ij} = \| x_s^i - x_a^j \|_2$ defines the noise-free anchor-sensor distance, $V_{ij}$ is the anchor-sensor measurement noise and $V_{ij} \sim P_{ij}^a(\mathcal Z_{ij}|\mathcal Y_{ij})$ with $P_{ij}^a(\cdot)$ being the so-called anchor-sensor sensing PDF. Here we assume that the inter-sensor and anchor-sensor measurement noise PDFs, i.e., $P_{ij}^s(\cdot)$ and $P_{ij}^a(\cdot)$, respectively, are Gaussian. Particularly, we assume that the inter-sensor and anchor-sensor measurement noises are independent and that $E_{ij} \sim \mathcal N(0, \Sigma^r_{ij})$ and $V_{ij} \sim \mathcal N(0, \Sigma^a_{ij})$. Notice that this assumption can be relaxed to any distribution that is a log-concave function of distances $\mathcal D_{ij}$ and $\mathcal Z_{ij}$, however, for the sake of brevity we limit ourselves to the case of Gaussian distributions. Having defined the setup of the sensor network, we can write the localization problem in a maximum likelihood setting as \small \begin{multline}\label{eq:MLOriginal} X_{\text{ML}}^\ast = \argmin_{X} \Bigg\{ \sum_{i = 1}^N\Bigg( \sum_{\tiny\begin{split}j \in &\text{Ne}_r(i)\\ i &< j\end{split}\normalsize} \frac{1}{\Sigma^r_{ij}}\left( \mathcal D_{ij}(x^i_s, x^j_s) - \mathcal R_{ij}\right)^2 \\ + \sum_{j \in \text{Ne}_a(i)} \frac{1}{\Sigma^a_{ij}}\left(\mathcal Z_{ij}(x^i_s, x^j_a) - \mathcal Y_{ij}\right)^2 \Bigg) \Bigg\}, \end{multline} \normalsize where $X = \begin{bmatrix} x^1_s & \dots & x^N_s \end{bmatrix}\in \mathbb R^{d\times N}$ with $d = 2$ or $d = 3$. This problem can be formulated as a constrained optimization problem, as was described in \cite{sim:14}, which is discussed next. First let us define the function \small \begin{multline} f(\Lambda, \Xi, D, Z) : = \sum_{i = 1}^N \Bigg(\sum_{\tiny\begin{split}j \in &\text{Ne}_r(i)\\ i &< j\end{split}\normalsize}\frac{1}{\Sigma^r_{ij}} (\Lambda_{ij} - 2D_{ij}\mathcal R_{ij} +\mathcal R_{ij}^2) \\ + \sum_{j \in \text{Ne}_a(i)}\frac{1}{\Sigma^a_{ij}} (\Xi_{ij} - 2Z_{ij}\mathcal Y_{ij} +\mathcal Y_{ij}^2) \Bigg). \end{multline} \normalsize Then the problem in \eqref{eq:MLOriginal} can be equivalently rewritten as the following constrained optimization problem \small \begin{subequations}\label{eq:MLConstrained} \begin{align} \minimize_{X, S, \Lambda, \Xi, D, Z}& \quad f(\Lambda, \Xi, D, Z)\\ \subject & \notag \\ & \begin{rcases*} S_{ii} + S_{jj} -2S_{ij} = \Lambda_{ij} \\ \Lambda_{ij} = D_{ij}^2, \quad D_{ij} \geq 0, \ j\in \text{Ne}_r(i),i<j\end{rcases*}, \ i \in \mathbb N_N\label{eq:MLConstrained-b}\\ &\begin{rcases*} S_{ii} - 2(x_s^i)^Tx_a^j + \| x_a^j \|^2_2 = \Xi_{ij} \\ \Xi_{ij} = Z_{ij}^2, \quad Z_{ij} \geq 0, \ \ j \in \text{Ne}_a(i) \end{rcases*}, \ \ i \in \mathbb N_N \label{eq:MLConstrained-c}\\ & S = X^T X.\label{eq:MLConstrained-d} \end{align} \end{subequations} \normalsize So far we have reviewed a way to formulate the localization problem over general sensor networks as a constrained optimization problem. In this paper, however, we are particularly interested in localization problem pertaining to sensor networks with an inherent tree structure which relies on the assumption that the graph $G_r(V_r, \mathcal E_r)$ can be represented using a tree. We describe the localization problem of such networks in the next section. \section{Localization of Tree-structured Scattered Sensor Networks}\label{sec:scattered} Let the graph $G_r(V_r, \mathcal E_r)$ be connected with few edges. Also assume that a chordal embedding $\bar G_r(V_r, \bar{\mathcal E}_r)$ of this graph can be achieved by adding only a few edges. This graph can then be represented using its clique tree. Furthermore, given the set of its cliques $\mathbf C_{\bar G_r} = \{ C_1, \dots, C_q \}$, we have $| C_i | \ll N$. We refer to such sensor networks as tree-structured scattered. The localization problem of these sensor networks can also be formulated as a constrained optimization problem using the approach discussed in Section \ref{sec:notation}. However, the formulation of the problem in~\eqref{eq:MLConstrained} is not fully representative of the structure in the problem. In order to exploit the structure in our localization problem we modify \eqref{eq:MLConstrained}, and equivalently rewrite it as \small \begin{subequations}\label{eq:MLConstrainedScattered} \begin{align} \minimize_{X, S, \Lambda, \Xi, D, Z}& \quad f(\Lambda, \Xi, D, Z) \\ \subject & \notag \\ & \ \begin{rcases*} S_{ii} + S_{jj} -2S_{ij} = \Lambda_{ij} \\ \Lambda_{ij} = D_{ij}^2, \quad D_{ij} \geq 0, \ j\in \text{Ne}_r(i), i<j\end{rcases*}, \ i \in \mathbb N_N\label{eq:MLConstrainedScattered-b}\\ &\ \begin{rcases*} S_{ii} - 2(x_s^i)^Tx_a^j + \| x_a^j \|^2_2 = \Xi_{ij} \\ \Xi_{ij} = Z_{ij}^2, \quad Z_{ij} \geq 0, \ \ j \in \text{Ne}_a(i) \end{rcases*}, \ i \in \mathbb N_N \label{eq:MLConstrainedScattered-c}\\ & \ S \succeq 0, \ \ S_{ij} = (x_s^i)^Tx_s^j, \notag \\ &\hspace{25mm} \forall \ (i,j) \in \mathcal E_r \cup \{ (i, i) \ | \ i \in V_r \}\label{eq:MLConstrainedScattered-d}. \end{align} \end{subequations} \normalsize Note that, here, we have modified the constraint in \eqref{eq:MLConstrained-d} so that the structure in the problem is more explicit. This modification is based on the observation that not all the elements of $S$ are used in \eqref{eq:MLConstrained-b} and \eqref{eq:MLConstrained-c}, and hence we only have to specify the ones that are needed and can leave the rest free. In \cite{kim:09}, \cite{wan:06}, the authors first conduct a semidefinite relaxation on \eqref{eq:MLConstrained}. They then exploit the structure as we did in \eqref{eq:MLConstrainedScattered} and use the ideas in \cite{fukuda_exploitingsparsity} to devise efficient centralized solvers for the localization problem. Here, however, we stick to the formulation in \eqref{eq:MLConstrainedScattered} which is a nonlinear SDP, and use scheme in \cite{kim+koj+mev+yam10} to decompose this problem directly. We then perform a semidefinite relaxation on the resulting problem and rewrite the problem as a coupled SDP. This in turn facilitates the use of efficient scalable or distributed solvers. The use of the so-called domain-space decomposition presented in \cite{kim+koj+mev+yam10} is at the heart of this reformulation approach. We review this decomposition scheme next, for the sake of completeness. \begin{rem} Notice that the added edges for computing a chordal embedding for the inter-sensor measurement graph does not affect the problem description in \eqref{eq:MLConstrainedScattered}, and only facilitates the clustering of the sensor nodes. \end{rem} \section{Chordal Sparsity in Semidefinite Programs}\label{sec:chordal} In this section we first briefly review some of important properties of sparse semidefinite matrices and then discuss how these can be used for reformulating semidefinite programs with chordal sparsity suitable to be solved distributedly. \subsection{Chordal Sparsity}\label{sec:sparsity} Graphs can be used to characterize partial symmetric matrices. Partial symmetric matrices correspond to symmetric matrices where only a subset of their elements are specified and the rest are free. We denote the set of all $n \times n$ partially symmetric matrices on a graph $Q(V,\mathcal E)$ by $\mathbf S_Q^n$, where only elements with indices belonging to $\mathbf I_s = \mathcal E \cup \{ (i,i) \ | \ i \in \mathbb N_n\}$ are specified. Now consider a matrix $X \in \mathbf S_Q^n$. Then $X$ is positive semidefinite completable if by manipulating its free elements, i.e., elements with indices belonging to $\mathbf I_f = (V \times V) \setminus \mathbf I_s $, we can generate a positive semidefinite matrix. The following theorem states a fundamental result on positive semidefinite completion. \begin{theorem}(\cite[Thm. 7]{gro:84})\label{thm:NSDC} Let $Q(V,\mathcal E)$ be a chordal graph with cliques $C_1, \dots, C_q$ such that clique intersection property holds. Then $X \in \mathbf S_Q^n$ is positive semidefinite completable, if and only if \begin{align} X_{C_iC_i} \succeq 0, \quad \ i \in \mathbb N_q, \end{align} where $X_{C_iC_i} = E_{C_i} X E_{C_i}^T$. \end{theorem} Note that the matrices $X_{C_iC_i}$ for $i \in \mathbb N_q$, are the fully specified principle submatrices of $X$. Hence, Theorem~\ref{thm:NSDC} states that a chordal matrix $X \in \mathbf S_{Q}^n$ is positive semidefinite completable if and only if all its fully specified principle submatices are positive semidefinite. As we will see next this property can be used for decomposing SDPs with this structure. \subsection{Domain-space Decomposition}\label{sec:decomposition} Consider a chordal graph $Q(V,\mathcal E)$, with $\{ C_1, \dots, C_q \}$ the set of cliques such that the clique intersection property holds. Let us define sets $J_i \subset \mathbb N_n$ such that the sparsity pattern graph for $\sum_{i=1}^N e_{J_i}e_{J_i}^T $ is $Q(V,\mathcal E)$. Then for the following nonlinear SDP \begin{subequations}\label{eq:SDP} \begin{align} \minimize_{z^1, \dots, z^N, X} &\quad \sum_{i=1}^N f^i(z^i,\svec(E_{J_i}XE_{J_i}^T)) \label{eq:SDP1} \\ \subject & \quad g^i(z^i, \svec(E_{ J_i}XE_{J_i}^T)) \in \Omega_i, \hspace{3mm} i \in \mathbb N_N, \label{eq:SDP2} \\ & \quad X \succeq 0, \label{eq:SDP3} \end{align} \end{subequations} the only elements of $X$ that affect the cost function in \eqref{eq:SDP1} and the constraint in \eqref{eq:SDP2} are elements specified by indices in $\mathbf I_s$. Using Theorem~\ref{thm:NSDC}, the optimization problem in~\eqref{eq:SDP} can then be equivalently rewritten as \begin{subequations}\label{eq:MSDP1} \begin{align} \minimize_{z^1, \dots, z^N, X} & \quad \sum_{i=1}^N f^i(z^i,\svec(E_{J_i}XE_{J_i}^T)) \label{eq:MSDP1-1} \\ \subject & \quad g^i(z^i, \svec(E_{ J_i}XE_{J_i}^T)), \hspace{3mm} i \in \mathbb N_N, \label{eq:MSDP1-2} \\ & \quad X_{C_iC_i} \succeq 0, \hspace{3mm} i \in \mathbb N_q, \label{eq:MSDP1-3} \end{align} \end{subequations} where notice that the constraints in \eqref{eq:MSDP1-3} are coupled semidefinite constraints, \cite{fukuda_exploitingsparsity, kim+koj+mev+yam10}. It is possible to explicitly describe the coupling using consistency constraints and rewrite \eqref{eq:MSDP1} as \begin{subequations}\label{eq:MSDP} \begin{align} \minimize_{z^1, \dots, z^N, X^1, \dots, X^q, X} & \quad \sum_{i=1}^N f^i(z^i,\svec(E_{J_i}XE_{J_i}^T)) \label{eq:MSDP-1} \\ \subject & \quad g^i(z^i, \svec(E_{ J_i}XE_{ J_i}^T)), \hspace{3mm} i \in \mathbb N_N, \label{eq:MSDP-2} \\ & \quad X^{i} \succeq 0, \hspace{3mm} i \in \mathbb N_q, \label{eq:MSDP-3}\\ & \quad X^i = E_{C_i}XE_{C_i}^T, \quad i \in \mathbb N_q, \label{eq:MSDP-5} \end{align} \end{subequations} where $X^i \in \mathbf S^{|C_i|}$. This method of reformulating \eqref{eq:SDP} as \eqref{eq:MSDP} is referred to as the domain-space decomposition, \cite{kim+koj+mev+yam10,and:11}. The structure in the localization of tree-structured scattered sensor networks enable us to use this technique for reformulating the problem in such a way that would better facilitate the use of efficient distributed solvers. This is discussed in the next section. \section{Decomposition and Convex Formulation of Localization of Tree-structured Scattered Sensor Networks}\label{sec:decompositionloc} Consider the inter-sensor measurement graph $G_r(V_r, \mathcal E_r)$, and assume that it is chordal. In case this graph is not chordal the upcoming discussions hold for any of its chordal embeddings. Let $\mathbf C_{G_r} = \{ C_1, \dots, C_q \}$ and $T(V_t,\mathcal E_t)$ be a clique tree. Based on the discussion in Section~\ref{sec:decomposition}, then for the problem in \eqref{eq:MLConstrainedScattered} we have $S \in \mathbf S^N_{G_r}$. Hence, we can rewrite~\eqref{eq:MLConstrainedScattered} as \small \begin{subequations}\label{eq:MLConstrainedScattered1} \begin{align} \minimize_{X, S_{C_iC_i}, \Lambda, \Xi, D, Z}& \quad f(\Lambda, \Xi, D, Z)\label{eq:MLConstrainedScattered1-a} \\ \subject & \notag \\ & \begin{rcases*} S_{ii} + S_{jj} -2S_{ij} = \Lambda_{ij} \\ \Lambda_{ij} = D_{ij}^2, \quad D_{ij} \geq 0, j\in \text{Ne}_r(i), i<j\end{rcases*}, i \in \mathbb N_N,\label{eq:MLConstrainedScattered1-b}\\ & \begin{rcases*} S_{ii} - 2(x_s^i)^Tx_a^j + \| x_a^j \|^2_2 = \Xi_{ij} \\ \Xi_{ij} = Z_{ij}^2, \quad Z_{ij} \geq 0, \ \ j \in \text{Ne}_a(i) \end{rcases*}, \ \ i \in \mathbb N_N, \label{eq:MLConstrainedScattered1-c}\\ & S_{C_iC_i} \succeq 0, \quad S_{C_iC_i} = E_{C_i} X^TXE_{C_i}^T, \ \ i \in \mathbb N_q \label{eq:MLConstrainedScattered1-d}, \end{align} \end{subequations} \normalsize Notice that even though the cost function for this problem is convex, the constraints in \eqref{eq:MLConstrainedScattered1-b}--\eqref{eq:MLConstrainedScattered1-d} are non-convex and hence the problem is non-convex. Consequently, we next address the localization problem by considering a convex relaxation of this problem. This allows us to solve the localization problem approximately. One of the ways to provide a convex approximation of the problem in \eqref{eq:MLConstrainedScattered1} is to relax the quadratic equality constraints in \eqref{eq:MLConstrainedScattered1-b}--\eqref{eq:MLConstrainedScattered1-d} using Schur complements, which results in \small \begin{subequations}\label{eq:MLConstrainedScattered2} \begin{align} &\minimize_{\tiny\begin{matrix}X, S_{C_iC_i}, \Lambda_{ij}, \Xi_{ij}, D_{ij},\\ Z_{ij}, T^i, \Gamma^{ij}, \Phi^{ij} \end{matrix}\normalsize} \quad \sum_{i = 1}^N \left(\sum_{\tiny\begin{matrix}j \in \text{Ne}_r(i)\\ i < j\end{matrix}\normalsize}f_{ij}(\Lambda_{ij}, D_{ij}) +\right. \notag\\& \left. \hspace{50mm} \sum_{j \in \text{Ne}_a(i)}g_{ij}(\Xi_{ij}, Z_{ij}) \right) \label{eq:MLConstrainedScattered2-a} \\ &\subject \notag\\ & \quad\quad \quad(S_{ii}, S_{jj}, S_{ij}, \Lambda_{ij}, D_{ij}, \Gamma^{ij}) \in \Omega_{ij}, \ (i,j) \in \mathcal E_r,\ i<j, \label{eq:MLConstrainedScattered2-b}\\ &\quad\quad\quad (S_{ii}, x_s^i, \Xi_{ij} , Z_{ij},\Phi^{ij}) \in \Theta_{ij}, \ \ j \in \text{Ne}_a(i), \ \ i \in \mathbb N_N, \label{eq:MLConstrainedScattered2-c}\\ & \quad\quad\quad \begin{bmatrix} I & XE_{C_i}^T \\ E_{C_i}X^T & S_{C_iC_i} \end{bmatrix} = T^i, \quad T^i \succeq 0, \ \ i \in \mathbb N_q \label{eq:MLConstrainedScattered2-d}, \end{align} \end{subequations} \normalsize where \small \begin{align*} f_{ij}(\Lambda_{ij}, D_{ij}) & = \frac{1}{\sigma_{ij}^2} (\Lambda_{ij} - 2D_{ij}R_{ij} + R_{ij}^2),\\ g_{ij}(\Xi_{ij}, Z_{ij}) & = \frac{1}{\delta_{ij}^2} (\Xi_{ij} - 2Z_{ij}Y_{ij} + Y_{ij}^2), \end{align*} \normalsize and \small \begin{align*} \Omega_{ij} &= \Bigg \{ (S_{ii}, S_{jj}, S_{ij}, \Lambda_{ij}, D_{ij}, \Gamma^{ij}) \ \Bigg | \ S_{ii} + S_{jj} -2S_{ij} = \Lambda_{ij}, \\ & \quad \quad \quad \quad \quad \quad \quad \quad \quad \ \begin{bmatrix} 1 & D_{ij} \\ D_{ij} & \Lambda_{ij} \end{bmatrix} = \Gamma^{ij}, \ \Gamma^{ij} \succeq 0, \ D_{ij} \geq 0 \Bigg\},\\ \Theta_{ij} &= \Bigg \{ (S_{ii}, x_s^i, \Xi_{ij} , Z_{ij}, \Phi^{ij}) \ \Bigg | \ S_{ii} - 2(x_s^i)^Tx_a^j + \| x_a^j \|^2_2 = \Xi_{ij}, \\ & \quad \quad \quad \ \ \ \begin{bmatrix} 1 & Z_{ij} \\ Z_{ij} & \Xi_{ij} \end{bmatrix} = \Phi^{ij}, \ \Phi^{ij} \succeq 0 , Z_{ij} \geq 0, \ \ j \in \text{Ne}_a(i) \Bigg\}, \end{align*} \normalsize with the variables $\Gamma^{ij}$, $\Phi^{ij}$ and $T^i$ as slack variables. The addition of the slack variables enable us to make the description of the semidefinite constraints simpler. This problem is a coupled SDP and can be solved distributedly using $q$ computational agents. In order to see this with more ease, let us introduce a grouping of the cost function terms and constraints in~\eqref{eq:MLConstrainedScattered2-a}--\eqref{eq:MLConstrainedScattered2-c}. To this end we first describe a set of assignment rules. It is possible to assign \begin{enumerate} \item the constraint $(S_{ii}, S_{jj}, S_{ij}, \Lambda_{ij}, D_{ij}, \Gamma^{ij}) \in \Omega_{ij}$ and the cost function term $f_{ij}$ to agent $k$ if $(i,j) \in C_k \times C_k$; \item the set of constraints $(S_{ii}, x_s^i, \Xi_{ij} , Z_{ij}, \Phi^{ij}) \in \Theta_{ij}, \ \ j \in \text{Ne}_a(i)$ and the cost function terms $g_{ij}, \ \ j \in \text{Ne}_a(i)$ to agent $k$ if $i \in C_k$. \end{enumerate} We denote the indices of the constraints and cost function terms assigned to agent $k$ through Rule 1 above as $\phi_{k}$, and similarly we denote the set of constraints and cost function terms that are assigned to agent $k$ through Rule 2 by $\bar \phi_k$. Using the mentioned rules and the defined notations, we can now group the constraints and the cost function terms and rewrite the problem in \eqref{eq:MLConstrainedScattered2} as \small \begin{subequations}\label{eq:MLConstrainedScattered3} \begin{align} &\minimize_{\tiny\begin{matrix}X, S_{C_iC_i}, \Lambda_{ij}, \Xi_{ij}, D_{ij},\\ Z_{ij}, T^i, \Gamma^{ij}, \Phi^{ij} \end{matrix}\normalsize} \quad \sum_{k = 1}^q \left[ \sum_{(i,j) \in \phi_k} f_{ij}(\Lambda_{ij}, D_{ij}) + \right. \notag \\ &\left. \hspace{50mm} \sum_{i \in \bar \phi_k} \sum_{j \in \text{Ne}_a(i)} g_{ij}(\Xi_{ij}, Z_{ij}) \right] \label{eq:MLConstrainedScattered3-a} \\ &\subject \notag \\ & \quad \quad \begin{rcases*} (S_{ii}, S_{jj}, S_{ij}, \Lambda_{ij}, D_{ij}, \Gamma^{ij}) \in \Omega_{ij}, \ \ (i,j) \in \phi_k \\ (S_{ii}, x_s^i, \Xi_{ij} , Z_{ij},\Phi^{ij}) \in \Theta_{ij}, \ \ j \in \text{Ne}_a(i) \ \ i \in \bar \phi_k \\ \begin{bmatrix} I & XE_{C_k}^T \\ E_{C_k}X^T & S_{C_kC_k} \end{bmatrix} = T^k, \ T^k \succeq 0 \end{rcases*}, k \in \mathbb N_q \label{eq:MLConstrainedScattered3-b} \end{align} \end{subequations} \normalsize Notice that this problem can now be seen as a combination of $q$ coupled subproblems, each defined by a term in the cost function together with its corresponding set of constraints in~\eqref{eq:MLConstrainedScattered3-b}. It is possible to decompose this problem by introducing additional local variables and consistency constraints and use any proximal point splitting method, e.g., ADMM, to solve this problem distributedly. However, there are major disadvantages for the resulting distributed solution, such as \begin{itemize} \item the local subproblems that needs to be solved by each agent is a semidefinite program that are computationally expensive to solve; \item inexact solutions for semidefinite programs can be far away from the optimal solution; \item the algorithm generally requires many iterations to converge to an accurate solution that particularly satisfies the consistency constraints; \item the number of consistency constraints are generally big for such problems which can even further adversely affect the convergence and numerical properties of such algorithms. \end{itemize} In order to evade the aforementioned issues, we next put forth an alternative distributed algorithm based on primal-dual interior-point methods that fully takes advantage of the structure in the problem and yields an accurate solution within much lower number of iterations and with far less computational demands from each agent. \begin{rem} The accuracy of the estimates obtained from solving \eqref{eq:MLConstrainedScattered3} can be improved by pushing the rank of matrices $\Gamma^{ij}$ and $\Phi^{ij}$ to 1 and the rank of matrices $T^i$ to $d$, see e.g., \cite{wan:06}. One way to achieve this is through the use of nuclear norm regularization by adding \small \begin{align} \sum_{k = 1}^q \left[ \alpha^k \| T^k \|_* + \sum_{(i,j) \in \phi_k} \rho^{ij} \| \Gamma^{ij} \|_* + \sum_{i \in \bar \phi_k} \sum_{j \in \text{Ne}_a(i)} \mu^{ij} \| \Phi^{ij} \|_* \right], \end{align} \normalsize to the cost function of \eqref{eq:MLConstrainedScattered3}, see \cite{rec:10}, where $\| \cdot \|_*$ denotes the nuclear norm of a matrix and $\alpha^k>0$, $\rho^{ij}>0$ and $\mu^{ij}>0$ are the so-called regularization parameters. Since all the aforementioned matrices are restricted to be positive semidefinite this will be equivalent to \small \begin{align} \sum_{k = 1}^q \left[ \alpha^k \trace( T^k ) + \sum_{(i,j) \in \phi_k} \rho^{ij} \trace( \Gamma^{ij} ) + \sum_{i \in \bar \phi_k} \sum_{j \in \text{Ne}_a(i)} \mu^{ij} \trace( \Phi^{ij} ) \right]. \end{align} \normalsize Notice that by increasing the regularization parameters the rank of these matrices are further pushed towards lower values. Furthermore, this does not affect the coupling structure in the problem since the added terms to the cost function concern the local matrix variables. Here, for the sake of brevity and notational simplicity, we do not consider the use of regularization. The coming discussion in Section \ref{sec:DPDIPM} can be extended to the regularized problem with little effort. \end{rem} \subsection{A Simple Assignment Strategy} Before we continue, let us first put forth an assignment strategy that is simple and satisfies the assignment rules discussed above. Recall that in order to form the problem in \eqref{eq:MLConstrainedScattered3}, we first need to cluster the sensor nodes. Based on this clustering, we use the assignment strategy described in Algorithm \ref{alg:Ass}. \begin{algorithm}[tb] \caption{A Simple Assignment Strategy}\label{alg:Ass} \begin{algorithmic}[1] \small \State{Given the inter-sensor measurement graph $G_r(V_r \mathcal E_r)$ and $C_{G_r} = \{ C_1, \dots, C_q \}$} \For {$k = 1, \dots, q$} \For {$i \in C_k$} \For {$j \in \text{Ne}_r(i)$ and $i<j$} \If{$\Omega_{ij}$ is not assigned and $j\in C_k$} \State{Assign it to agent $k$} \EndIf \If{$f_{ij}$ is not assigned and $j\in C_k$} \State{Assign it to agent $k$} \EndIf \EndFor \For {$j \in \text{Ne}_a(i)$} \If{$\Theta_{ij}$ is not assigned} \State{Assign it to agent $k$} \EndIf \If{$g_{ij}$ is not assigned} \State{Assign it to agent $k$} \EndIf \EndFor \EndFor \EndFor \normalsize \end{algorithmic} \end{algorithm} Notice that the resulting assignment heavily relies on the ordering of the cliques or clusters of sensors. Consequently, different ordering of the cliques may result in different assignments of constraints and terms in the objective function. Furthermore, even though this assignment algorithm is simple, it may lead to unbalanced distribution of constraints and cost function terms. This means that some agents maybe assigned a disproportionate number of variables, constraints and objective function terms. One can avoid such a situation by modifying the \emph{if} statements in steps 5, 8, 13 and 16 of the algorithm, by adding watchdogs that prevent unbalanced assignments. For the sake of brevity and so as to not clutter the presentation, we do not discuss this any further. \begin{rem} Notice that each pair $\Omega_{ij}$ and $f_{ij}$ corresponds to the range measurement between sensors $i$ and $j$ and each pair $\Theta_{ij}$ and $g_{ij}$ corresponds to a range measurement between sensor $i$ and anchor $j$. Based on this, using the assignment rules, we essentially assign different range measurements to each sensor cluster or computational agent. \end{rem} \section{Distributed Primal-dual Interior-point Method for Coupled SDPs}\label{sec:DPDIPM} The problem in \eqref{eq:MLConstrainedScattered3} can be written in the following standard form \small \begin{subequations} \begin{align} \minimize &\quad \sum_{i = 1}^q (c^i)^T y \\ \subject &\notag \\ & \begin{rcases*} Q^i_j \svec(X_j^i) + W_j^i y = b^i_j, \quad j = 1, \dots, m_i \\ A^iy = \bar b^i \\ D^iy \leq g^i \\ X^i_j \succeq 0, \quad j = 1, \dots, m_i \end{rcases*}, \ i \in \mathbb N_q \end{align} \end{subequations} \normalsize where the variables $X^i_j$ and $y$ are matrix and linear variables, respectively. This problem can be written more compactly as \begin{subequations}\label{eq:GeneralSDP} \begin{align} \minimize &\quad \sum_{i = 1}^q (c^i)^T y \\ \subject &\notag \\ & \quad \begin{rcases*} Q^i x^i + W^i y = b^i \\ A^iy = \bar b^i \\ D^iy \leq g^i \\ X^i \succeq 0 \end{rcases*}, \ i \in \mathbb N_q \end{align} \end{subequations} with $Q^i = \blkdiag(Q^i_1, \dots, Q^i_{m_i})$, $W^i = \begin{bmatrix} (W^i_1)^T & \dots & (W^i_{m_i})^T\end{bmatrix}^T$, $b^i = (b^i_1, \dots, b^i_{m_i})$, $x^i = \left(\svec(X^i_1), \dots, \svec(X^i_{m_i})\right)$ and $X^i = \blkdiag(X^i_1, \dots, X^i_{m_i})$. It is possible to solve this problem using a primal-dual interior-point method, \cite{wri:97}, \cite{tod:96}. Next we briefly discuss the main stages of such a method. The Karush-Kuhn-Tucker, KKT, optimality conditions for this problem are given as \begin{subequations}\label{eq:KKT} \begin{align} \sum_{i = 1}^q \left((W^i)^Tv^i + (A^i)^T\bar v^i + (D^i)^T\lambda^i \right) &= -\sum_{i = 1}^q c^i, \\ (Q^i)^T v^i - z^i &= 0, \quad i \in \mathbb N_q,\\ X^iZ^i &= 0, \quad i \in \mathbb N_q, \label{eq:KKT-c}\\ \diag(\lambda^i)\left( D^iy-g^i \right) &= 0, \quad i \in \mathbb N_q,\label{eq:KKT-d}\\ Q^i x^i + W^i y &= b^i, \quad i \in \mathbb N_q, \\ A^i y &= \bar b^i, \quad i \in \mathbb N_q, \end{align} \end{subequations} together with $D^i x \leq g^i$ and $X^i \succeq 0$, where $Z^i = \blkdiag(Z^i_1, \dots, Z^i_{m_i})$ and $z^i = \left(\svec(Z^i_1), \dots, \svec(Z^i_{m_i})\right)$. Any solution to this set of nonlinear equations is optimal for~\eqref{eq:GeneralSDP}. Within a primal-dual interior-point method, we set out to compute a solution to \eqref{eq:GeneralSDP}, by considering a sequence of perturbed KKT conditions where \eqref{eq:KKT-c} and \eqref{eq:KKT-d} are modified~as \begin{align*} X^iZ^i &= \delta I, \quad i \in \mathbb N_q,\\ \diag(\lambda^i)\left( D^iy-g^i \right) &= -\delta \mathbf 1, \quad i \in \mathbb N_q. \end{align*} where $\delta >0$ is the perturbation parameter. Particularly at each iteration, given feasible iterates $\lambda^i > 0$, $y$ so that $D^i y > g^i$ and $X^i \succ 0$ for $i = 1, \dots, q$, the primal-dual search directions are computed by solving a linearized version of the perturbed KKT conditions, given as \small \begin{subequations}\label{eq:KKTLin} \begin{align} \sum_{i=1}^q \left((W^i)^T \Delta v^i + (A^i)^T \Delta \bar v^i + (D^i)^T \Delta \lambda^i\right) &= r_{\textrm{d,lin}},\\ (Q^i)^T \Delta v^i - \Delta z^i &= r_{\textrm{d}}^i, \ i \in \mathbb N_q,\\ U^i \Delta x^i + F^i \Delta z^i &= r_{\textrm{c}}^i, \ i \in \mathbb N_q, \\ \diag(\Delta \lambda^i) (D^i y - g^i) + \diag(\lambda^i) D^i \Delta y &= r_{\textrm{c,lin}}^i, \ i \in \mathbb N_q,\\ Q^i \Delta x^i + W^i \Delta y &= r_{\textrm{p}}^i, \quad i \in \mathbb N_q, \\ A^i \Delta y &= r_{\textrm{p,lin}}^i, \quad i \in \mathbb N_q, \end{align} \end{subequations} \normalsize with $U^i = \blkdiag(U^i_1, \dots, U^i_{m_i})$, $F^i = \blkdiag(F^i_1, \dots, F^i_{m_i})$, where given \begin{equation}\label{eq:scaling} \begin{split} W^i_j :&= (X^i_j)^{\frac{1}{2}} \left( (X^i_j)^{\frac{1}{2}} Z^i_j (X^i_j)^{\frac{1}{2}} \right)^{-\frac{1}{2}} (X^i_j)^{\frac{1}{2}}\\ & = (Z^i_j)^{-\frac{1}{2}} \left( (Z^i_j)^{\frac{1}{2}} X^i_j (Z^i_j)^{\frac{1}{2}} \right)^{\frac{1}{2}} (Z^i_j)^{-\frac{1}{2}}, \end{split} \end{equation} $W^i_j =: G^i_j(G^i_j)^T$ and $D^i_j = (G^i_j)^{-1}$, we have $U^i_j = D^i_j \otimes_s (D^i_j)^{-T}Z^i_j$ and $F^i_j = D^i_jX^i_j \otimes_s (D^i_j)^{-T}$. Furthermore, the residuals are given as \begin{subequations} \begin{align} r_{\textrm{d,lin}} &= \sum_{i = 1}^q \underbrace{-c^i - (Q^i)^Tv^i - (A^i)^T\bar v^i - (D^i)^T\lambda^i}_{r_{d,lin}^i}\\ r_{\textrm{d}}^i &= z^i - (Q^i)^Tv^i, \quad i \in \mathbb N_q,\\ r_{\textrm{c}}^i &= \svec(\delta I - H_{D^i_j}(X^i_jZ^i_j)), \quad i \in \mathbb N_q,\\ r_{\textrm{c,lin}}^i &= -\delta \mathbf 1 - \diag(\lambda^i)(D^iy-g^i), \quad i \in \mathbb N_q,\\ r_{\textrm{p}}^i &= b^i - W^ix^i -Q^i y , \quad i \in \mathbb N_q,\\ r_{\textrm{p,lin}}^i &= \bar b^i - A^i y , \quad i \in \mathbb N_q, \end{align} \end{subequations} where $H_D(M) = 1/2(DMD^{-1} + D^{-T}MD^{T})$. Having computed the search directions, suitable primal and dual step sizes, i.e., $t_d$ and $t_p$, are calculated so as to guarantee feasibility of the iterates with respect to inequality constraints and persistent reduction of residual norms, see e.g., \cite{tod:96} and references therein, which then allows us to update the iterates. This process is then repeated until certain stopping criteria are satisfied, which commonly depend on the residual norms and the size of the perturbation parameter. A generic description of a primal-dual interior-point method is given in Algorithm \ref{alg:PDIPM}. \begin{algorithm}[tb] \caption{Primal-dual Interior-point Method}\label{alg:PDIPM} \begin{algorithmic}[1] \small \State{Given feasible iterates with respect to inequality constraints} \Repeat \State{Compute the primal-dual search directions} \State{Compute primal and dual step sizes} \State{Update primal and dual iterates} \State{Update the perturbation parameter} \Until{stopping criteria is satisfied} \normalsize \end{algorithmic} \end{algorithm} The most computationally demanding step at every iteration of a primal-dual interior-point method, concerns the computation of the search directions. This requires solving the linear system of equations in \eqref{eq:KKTLin}, which can be written more compactly as \begin{multline}\label{eq:KKTCompact} \begin{bmatrix} W^T & A^T & & & & D^T \\ Q^T & & & & -I & \\ & & U & & F & \\ & & & \Lambda D & & E \\ & & Q & W & & \\& & & A & & \end{bmatrix} \begin{bmatrix} \Delta v \\ \Delta \bar v \\ \Delta x \\ \Delta y \\ \Delta z \\ \Delta \lambda \end{bmatrix} = \begin{bmatrix} r_{\textrm{d,lin}}\\ r_{\textrm{d}} \\ r_{\textrm{c}} \\ r_{\textrm{c,lin}}\\ r_{\textrm{p}} \\ r_{\textrm{p,lin}} \end{bmatrix} \end{multline} where \begin{align*} W &= \begin{bmatrix} (W^1)^T & \dots & (W^q)^T\end{bmatrix}^T,\\ A &= \begin{bmatrix} (A^1)^T & \dots & (A^q)^T\end{bmatrix}^T,\\ Q &= \blkdiag(Q^1, \dots, Q^q),\\ U &= \blkdiag(U^1, \dots, U^q),\\ F &= \blkdiag(F^1, \dots, F^q),\\ D &= \begin{bmatrix} (D^1)^T & \dots & (D^q)^T\end{bmatrix}^T,\\ \Lambda &= \blkdiag(\Lambda^1, \dots, \Lambda^q), \quad \Lambda^i = \diag(\lambda^i),\\ E &= \blkdiag(E^1, \dots, E^q), \quad E^i = \diag(D^iy - g^i), \end{align*} and the variables and the right hand side terms correspond to all variables and residuals stacked. One way to solve this system of equations is by first eliminating the third and fourth row equations as \begin{subequations} \begin{align} \Delta z &= F^{-1}\left( r_c - U \Delta x \right), \\ \Delta \lambda &= E^{-1}\left( r_{\textrm{c,lin}} - \Lambda D \Delta y \right), \end{align} \end{subequations} which is possible since $F$ and $E$ are both invertible, see e.g., \cite{tod:96, wri:97}. This then allows us to rewrite \eqref{eq:KKTCompact} as \begin{multline} \begin{bmatrix} -D^TE^{-1}D & & W^T & A^T \\ & F^{-1}U & Q^T & \\ W & Q & & \\A & & & \end{bmatrix} \begin{bmatrix} \Delta y \\ \Delta x \\ \Delta v \\ \Delta \bar v \end{bmatrix} = \begin{bmatrix} r_{\textrm{lin}}\\ r \\ r_{\textrm{p}} \\ r_{\textrm{p,lin}} \end{bmatrix} \end{multline} where $r_{\textrm{lin}} = r_{\textrm{d,lin}} - D^T E^{-1} r_{\textrm{c,lin}}$ and $r = r_{\textrm{d}} - F^{-1} r_{\textrm{c}}$. Notice that this set of linear equations also defines the optimality conditions for the convex quadratic program (QP) \small \begin{subequations}\label{eq:KKTQP} \begin{align} \minimize \quad& \begin{bmatrix}\Delta y \\ \Delta x \end{bmatrix}^T \begin{bmatrix} -D^TE^{-1}D & \\ & F^{-1}U \end{bmatrix} \begin{bmatrix} \Delta y \\ \Delta x \end{bmatrix} -\notag\\ &\hspace{50mm} \begin{bmatrix} r_{\textrm{lin}}\\ r \end{bmatrix}^T \begin{bmatrix} \Delta y \\ \Delta x \end{bmatrix}\\ \subject \quad& W\Delta y + Q \Delta x = r_{\textrm{p}} \\ & A \Delta y = r_{\textrm{p,lin}} \end{align} \end{subequations} \normalsize For the localization problem, this QP has a particular structure which enables us to solve it distributedly and efficiently, using message-passing. Next we briefly discuss this algorithm for the sake of completeness and to provide a better understanding of the presented material. \subsection{Solving Coupled Optimization Problems Using Message-passing} Consider the following coupled optimization problem \begin{align}\label{eq:UncostrainedProblem} \minimize & \quad F_1(x) + F_2(x) + \dots + F_q(x), \end{align} where $x \in \mathbb R^n$ and and the functions $F_i \ : \ \mathbb R^n \rightarrow \mathbb R$ for $i \in \mathbb N_q$ are convex. Also we assume that each term in the objective function (each subproblem) only depends on a few variables. Let us denote the indices of the variables that appear in the $i$th term, $F_i$, by $J_i$. This definition allows us to rewrite the problem in \eqref{eq:UncostrainedProblem} as \begin{align}\label{eq:UncostrainedProblemReform} \minimize & \quad \bar F_1(x_{_{J_1}}) +\bar F_2(x_{_{J_2}}) + \dots +\bar F_q(x_{_{J_q}}), \end{align} where $x_{_{J_i}} = E_{J_i} x$. The functions $\bar F_i \ : \ \mathbb R^{|J_i|} \rightarrow \mathbb R$ are lower dimensional descriptions of $F_i$s such that $F_i(x) = \bar F_i(E_{J_i}x)$ for all $x \in \mathbb R^n$ and $i \in \mathbb N_N$. We also define $\mathcal I_j$ as the set of indices of terms in the cost function that depend on $x_j$, i.e., $\{ i \ | \ j \in J_i \}$. The sets $J_i$ for $i \in \mathbb N_q$ and $\mathcal I_j$ for $j \in \mathbb N_n$ provide a clear mathematical description of the coupling structure in the problem. It is also possible to describe the coupling structure in the problem graphically, using graphs. For this purpose, we introduce the sparsity graph. The \emph{sparsity graph} $G_s(V_s, \mathcal E_s)$ of a coupled problem is an undirected graph with the vertex set $V_s = \left\{ 1, \dots, n\right\}$ and the edge set $\mathcal E_s = \left\{ (i, j) \ | \ i, j \in V_s, \ \mathcal I_i \cap \mathcal I_j \neq \emptyset \right\}$. As an example consider the following problem \begin{multline}\label{eq:Example} \minimize_x \quad \bar F_1(x_1, x_3, x_4) + \bar F_2(x_1, x_2, x_4) + \bar F_3(x_4, x_5) +\\ \bar F_4(x_3, x_6, x_7) + \bar F_5(x_3, x_8). \end{multline} The sparsity graph for this problem are illustrated in Figure~\ref{fig:Example}. \begin{figure}[t] \begin{center} \includegraphics[width=3cm]{Example.pdf} \caption{ The sparsity graph for the problem in~\eqref{eq:Example}.} \label{fig:Example} \vspace{-5mm} \end{center} \end{figure} It is possible to devise scalable or distributed algorithms for solving the problem in \eqref{eq:UncostrainedProblem}. In this paper we focus on message-passing. Consider the problem in \eqref{eq:UncostrainedProblemReform}, and assume that its sparsity graph is chordal. Let its set of cliques be given as $\mathbf C_{G_s} = \{ J_1, \dots, J_q\}$ and $T_s(V_t, \mathcal E_t)$ be a clique tree over the cliques. It is possible to solve the problem in \eqref{eq:UncostrainedProblemReform} distributedly, using an algorithm with the clique tree as its computational graph. That means each node in the tree corresponds to a computational agent and they communicate/collaborate with one another if there is an edge between them. Recall that each node in the clique tree is assigned a clique of the sparsity graph, i.e., $J_i$. In such a setting, we also assign each term in the objective function (each subproblem), i.e., $\bar F_i$, to each agent $i$. We can now describe how the problem in \eqref{eq:UncostrainedProblemReform} can be solved using message-passing by performing an upward-downward pass through the clique tree. The message-passing algorithm starts from the agents at the leaves of the tree, i.e., all $i \in \leaves(T)$, where every such agent computes the following message \begin{align}\label{eq:mijLeaves} m_{i\parent(i)}(x_{_{S_{i\parent(i)}}}) = \minimum_{x_{_{R_{i\parent(i)}}}} \left\{ \bar F_i(x_{_{J_i}})\right\}, \end{align} with $S_{i\parent(i)} := J_i \cap J_{\parent(i)}$ and $R_{i\parent(i)} := J_i \setminus S_{i\parent(i)}$ are the so-called separators and residuals, respectively, and communicates it to its corresponding parent, denoted by $\parent(i)$. Notice that this message is a functional and not a scalar value, and hence agent $i$ needs to communicate the functional form. Then every parent $j$ that has received these messages from its children, denoted by $\children(j)$, computes its corresponding message to its parent as \small \begin{align}\label{eq:mij} m_{j\parent(j)}(x_{_{S_{j\parent(j)}}}) = \minimum_{x_{_{R_{j\parent(j)}}}} \left\{ \bar F_j(x_{_{J_j}}) + \sum_{k \in \children(j)} m_{kj}(x_{_{S_{kj}}}) \right\}. \end{align} \normalsize This procedure is then continued until we arrive at the agent at the root. At this point, the agent at the root, indexed $r$, having received all messages from its children can compute the optimal solution for its corresponding variables specified by $J_r$ as \begin{align}\label{eq:RLocalProblem} x^\ast_{_{J_r}} = \argmin_{x_{_{J_r}}} \left\{ \bar F_k(x_{_{J_r}}) + \sum_{k \in \children(r)} m_{kr}(x_{_{S_{rk}}}) \right\}. \end{align} This agent then having computed its optimal solution, communicates this solution to its children, at which point every such agent $i \in \children(r)$ computes its optimal solution as \begin{multline}\label{eq:LocalProblempar} x^\ast_{_{J_i}} = \argmin_{x_{_{J_i}}} \left\{ \bar F_i(x_{_{J_i}}) + \sum_{k \in \children(i)} m_{ki}(x_{_{S_{ik}}}) + \right. \\ \left. \frac{1}{2} \left\| x_{_{S_{\parent(i)i}}} - \left( x_{_{S_{\parent(i)i}}}^\ast \right)^{\parent(i)} \right\|^2 \right\}, \end{multline} where $\left( x_{_{S_{\parent(i)i}}}^\ast \right)^{\parent(i)}$ is the the computed optimal solution by the parent $\parent(i)$. This procedure is continued until we reach the agents at the leaves. At this point all agents have computed their corresponding optimal solution and the algorithm can be terminated, and hence, we have convergence after one upward-downward pass through the tree, \cite{kho:15c}, \cite{kol:09}. Let us now illustrate this procedure using an example. Consider the example given in \eqref{eq:Example}. The sparsity graph of this problem is chordal and its cliques are marked in Figure \ref{fig:ExampleClique} on the left. A clique tree for this graph is illustrated in the same figure on the right, where also a valid subproblem assignment is presented. \begin{figure}[t] \begin{center} \includegraphics[width=9cm]{ExampleClique.pdf} \caption{ The coupling and sparsity graphs for the problem in~\eqref{eq:Example}, illustrated on the right and left figures, respectively.} \label{fig:ExampleClique} \vspace{-5mm} \end{center} \end{figure} As was discussed above we start the message-passing from the leaves of the tree, particularly agents 3, 4 and 5. These agents compute and communicate their messages to their corresponding parents as \begin{align*} m_{32}(x_4) &= \minimum_{x_5}\left\{ \bar F_3(x_4, x_5)\right\}\\ m_{41}(x_3) & = \minimum_{x_6, x_7}\left\{\bar F_4(x_3, x_6, x_7)\right\} \\ m_{51}(x_3) & = \minimum_{x_8}\left\{ \bar F_5(x_3, x_8)\right\}. \end{align*} At this point agent 2 has received all messages from its children and can in turn compute and communicate its message to its parent as \begin{align*} m_{12}(x_1, x_4) &= \minimum_{x_3}\left\{ m_{41}(x_3) + m_{51}(x_3) + \bar F_1(x_1, x_3, x_4)\right\}. \end{align*} This completes the upward pass and now the agent at the root, i.e., agent 2, can compute its optimal solution as \begin{multline*} (x_1^*, x_2^*, x_4^*) = \argmin_{x_1, x_2, x_4} \\\quad\left\{ m_{12}(x_1, x_4) + m_{32}(x_4) + \bar F_2(x_1, x_2, x_4)\right\}, \end{multline*} which initiates the downward pass. Agent 2 will then communicate $x_1^*, x_4^*$ and $x_4^*$ to agents 2 and 3 respectively, where they compute their corresponding optimal solution for the remainder of their variables as \begin{align*} x_3^* &= \argmin_{x_3} \quad\left\{ m_{41}(x_3) + m_{51}(x_3) + \bar F_1(x_1^*, x_3, x_4^*)\right\}\\ x_5^* &= \argmin_{x_5} \quad\left\{ \bar F_3(x_4^*, x_5)\right\}. \end{align*} The last step of the downward pass is then accomplished by agent 2 communicating $x_3^*$ to agents 4 and 5, and these agents computing their optimal solution as \begin{align*} (x_6^*, x_7^*) &= \argmin_{x_6, x_7} \quad\left\{ \bar F_4( x_3^*, x_6, x_7)\right\}\\ x_8^* &= \argmin_{x_8} \quad\left\{ \bar F_5(x_3^*, x_8)\right\}, \end{align*} which finishes the algorithm. Notice that the message-passing algorithm described in this section can be viewed as dynamic programming over trees. Next we discuss how message-passing can be used within the primal-dual method. \subsection{Distributed Computations In Primal-dual methods} The problem in \eqref{eq:KKTQP} can be written as \small \begin{subequations}\label{eq:KKTQPSum} \begin{align} \minimize \quad& \sum_{i = 1}^q \begin{bmatrix}\Delta y \\ \Delta x^i \end{bmatrix}^T \begin{bmatrix} H^i & \\ & (F^i)^{-1}U^i \end{bmatrix} \times \notag\\ & \hspace{35mm} \begin{bmatrix} \Delta y \\ \Delta x^i \end{bmatrix} - \begin{bmatrix} r^i_{\textrm{lin}}\\ r^i \end{bmatrix}^T \begin{bmatrix} \Delta y \\ \Delta x^i \end{bmatrix}\\ \subject &\notag \\\quad& \begin{rcases*} W^i\Delta y + Q^i \Delta x^i = r^i_{\textrm{p}}, \\ A^i \Delta y = r^i_{\textrm{p,lin}}, \end{rcases*}\quad i \in \mathbf N_q \end{align} \end{subequations} \normalsize where $H^i = -(D^i)^T(E^i)^{-1}D^i$ and $r^i_{\textrm{lin}} = r^i_{\textrm{d,lin}} - (D^i)^T (E^i)^{-1} r^i_{\textrm{c,lin}}$ and $r^i = r^i_{\textrm{d}} - (F^i)^{-1} r^i_{\textrm{c}}$. This problem can be viewed as a combination of $q$ subproblems, where each of which is defined by a term in the objective function and its corresponding equality constraints. Notice that the coupling among the subproblems does not stem from the matrix variables and on the surface all subproblems seem to be coupled to one another through the linear variables directions $\Delta y$. However, for the localization problem in \eqref{eq:MLConstrainedScattered3}, each subproblem only relies on a certain elements of $\Delta y$. This can be seen by first noticing that the linear variables for each subproblem $k$ is given by $\vectri(S_{C_kC_k})$, $\Lambda_{ij}, D_{ij}$ for $(i,j) \in \phi_k$ and $x^i_s, \Xi_{ij} , Z_{ij}$ for $j \in \textrm{Ne}_a(i)$ and $i \in \bar \phi_k$. Let us assume that the indices of elements of $\Delta y$ that correspond to these variables be given by set $J_k$. We can then rewrite the problem in \eqref{eq:KKTQPSum} as \small \begin{subequations}\label{eq:KKTQPSum} \begin{align} \minimize \quad& \sum_{i = 1}^q \begin{bmatrix}\Delta y_{_{J_i}} \\ \Delta x^i \end{bmatrix}^T \begin{bmatrix} \bar H^i & \\ & (F^i)^{-1}U^i \end{bmatrix} \times \notag\\ & \hspace{35mm} \begin{bmatrix} \Delta y_{_{J_i}} \\ \Delta x^i \end{bmatrix} - \begin{bmatrix}\bar r^i_{\textrm{lin}}\\ r^i \end{bmatrix}^T \begin{bmatrix} \Delta y_{_{J_i}} \\ \Delta x^i \end{bmatrix}\\ \subject \quad& \bar W^i\Delta y_{_{J_i}} + Q^i \Delta x^i = r^i_{\textrm{p}}, \quad i = 1, \dots, q \\ & \bar A^i \Delta y_{_{J_i}} = r^i_{\textrm{p,lin}}, \quad i = 1, \dots, q \end{align} \end{subequations} \normalsize where $\bar H^i = E_{J_i}H^iE_{J_i}^T$, $\bar r^i_{\textrm{lin}} = E_{J_i}r^i_{\textrm{lin}}$, $\bar A^i = A^iE_{J_i}^T$ and $\bar W^i = W^i E_{J_i}^T$. Through the use of indicator functions, this problem can be written as \small \begin{align} \minimize \quad& \sum_{i = 1}^q \begin{bmatrix}\Delta y_{_{J_i}} \\ \Delta x^i \end{bmatrix}^T \begin{bmatrix} \bar H^i & \\ & (F^i)^{-1}U^i \end{bmatrix} \begin{bmatrix} \Delta y_{_{J_i}} \\ \Delta x^i \end{bmatrix} - \notag\\ & \hspace{25mm}\begin{bmatrix}\bar r^i_{\textrm{lin}}\\ r^i \end{bmatrix}^T \begin{bmatrix} \Delta y_{_{J_i}} \\ \Delta x^i \end{bmatrix} + \mathcal I_{\mathcal C_i}(\Delta y_{_{J_i}}, \Delta x^i) \end{align} \normalsize where $\mathcal C_i = \{ (\Delta y_{_{J_i}},\Delta x^i)\ | \ \bar W^i\Delta y_{_{J_i}} + Q^i \Delta x^i = r^i_{\textrm{p}}, \ \bar A^i \Delta y_{_{J_i}} = r^i_{\textrm{p,lin}}\}$ and \begin{align*} \mathcal I_{\mathcal C_i}\left(x\right) = \begin{cases} 0 \hspace{6mm} x \in \mathcal C_i \\ \infty \hspace{4mm} \text{Otherwise} \end{cases} \end{align*} This problem is in the same format as \eqref{eq:UncostrainedProblemReform}. It is now possible to see that the coupling comes from the fact that for some $C_i$ and $C_j$, $C_i \cap C_j \neq \emptyset$. Recall that one way to describe the intersection among the cliques of the inter-sensor measurement graph can be described using its clique tree, $T(V_t, \mathcal E_t)$. The sparsity graph of this problem is in fact chordal with cliques defined by the variables that appear in each subproblem. Furthermore, the clique tree for the sparsity graph of this problem has the same structure as that of the inter-sensor measurement graph. This is the case since the ordering defined by this tree defines perfect elimination ordering for the sparsity graph, see \cite{gol:04} for more details. Consequently, this problem can be solved distributedly using message-passing as discussed above. As a result, we can compute the primal-dual search directions for the problem in \eqref{eq:MLConstrainedScattered3} distributedly, by an upward-downward pass through the clique tree. Notice that the messages for solving this problem are quadratic functions, and hence the hessian and linear term that describes this function need to be communicated. The remaining stages of a primal-dual interior-point method can also be done distributedly over the clique tree. For the sake of brevity, we here do not discuss the details any further, for more info see, \cite{kho:15c} and \cite{kho:15d}. A summary of our proposed distributed localization method is given in Algorithm \ref{alg:Local}. \begin{algorithm}[tb] \caption{Distributed Primal-dual Localization Algorithm, DPDLA}\label{alg:Local} \begin{algorithmic}[1] \small \State{Given the inter-sensor measurement graph $G_r(V_r, \mathcal E_r)$, its cliques set $C_{G_r} = \{ C_1, \dots, C_q \}$ and a clique tree over its cliques $T(V_t, \mathcal E_t)$ with $V_t = \{ 1, \dots, q\}$} \State{Conduct assignments such that the assignment rules in Section \ref{sec:decompositionloc} are satisfied, for instance using Algorithm \ref{alg:Ass}} \State{Each agent $i \in \mathbb N_q$ forms its corresponding subproblem} \State{Given feasible initial primal and dual iterates with respect to inequality constraints} \Repeat \State{Compute the primal-dual search directions distributedly using message-passing over $T(V_t, \mathcal E_t)$} \State{Compute primal and dual step sizes distributedly (this can be done by performing an upward-downward pass through $T(V_t, \mathcal E_t)$, see \cite[Sec. 6.4]{kho:15c},\cite[Sec. V-B]{kho:15d})} \State{Update primal and dual iterates} \State{Update the perturbation parameter and the compute the stopping criteria distributedly (this can be done by performing an upward-downward pass through $T(V_t, \mathcal E_t)$, see \cite[Sec. 6.4]{kho:15c},\cite[Sec. V-B]{kho:15d})} \Until{stopping criteria is satisfied} \normalsize \end{algorithmic} \end{algorithm} \subsection{Computational and Communication Complexity} At each iteration of the primal-dual method, we need to conduct three upward-downward passes, namely one for computing the primal-dual directions, one for computing the primal and dual step sizes and one for updating the perturbation parameter and checking the termination condition. This means that if the primal-dual method converges within $p$ iterations, the algorithm converges within $3\times2\times p\times h$ steps where $h$ is the height of the considered clique tree. Furthermore, during the execution of the algorithm, each agent is required to communicate twice with its neighbors during each upward-downward pass. Once with its parent during the upward pass and once with its children during the downward pass. Consequently, the total number of times each agent needs to communicate with its neighbors is given by $3\times 2\times p$. Among the upward and downward passes, the upward pass for computing the search directions, is the most computationally demanding and communication intensive one. Particularly, during this upward pass each agent needs to compute a factorization of a relatively small matrix to compute its message to the parent, see \cite[Sec. 6.2]{kho:15c}. This needs to be done once at every primal-dual iteration, which means that in total each agent is required to compute $p$ factorizations during the run of Algorithm \ref{alg:Local}. Also recall that during these upward passes, each agent needs to communicate a quadratic functional to its parent. This entails sending the data matrices that define the quadratic function. Depending on the number of variables shared between each agent and its parent, the information that needs to be communicated can be considerable. Notice that the computational burden of the other upward and downward passes are comparatively trivial. Moreover, the information that needs to be communicated during these upward and downward passes is limited to a few scalars. Due to this, in the remainder of this section, we discuss the computational and communication burden for each agent during the upward pass for computing the search directions. Firstly, recall that each subproblem $k$ in \eqref{eq:KKTQPSum}, depends on variables \begin{align*} &\vectri(S_{C_kC_k}),\\ &\Lambda_{ij}, D_{ij} \quad \textrm{for} \ (i,j) \in \phi_k, \\ &x^i_s, \Xi_{ij} , Z_{ij} \quad \textrm{for} \ j \in \textrm{Ne}_a(i), \ i \in \bar \phi_k, \\ & T^k, \\ & \Gamma^{ij} \quad \textrm{for} \ (i,j) \in \phi_k, \\ &\Phi^{ij} \quad \textrm{for} \ j \in \textrm{Ne}_a(i), \ i \in \bar \phi_k. \end{align*} Let us assume that each agent $k$ is assigned $b_k$ and $a_k$ inter-sensor and anchor-sensor range measurements, respectively. The number of variables that appear in each subproblem $k$ is then given as \small \begin{multline} n_k = \frac{|C_k|(|C_k|+1)}{2} + 2b_k + 2|C_k| + 2a_k + \\ \frac{(|C_k|+2)(|C_k|+3)}{2} + 3b_k + 3a_k. \end{multline} \normalsize Notice that the number of equality constraints defined by each range measurement is equal to four, see \eqref{eq:MLConstrainedScattered2-b} and \eqref{eq:MLConstrainedScattered2-c}. Consequently the number of equality constraints for each subproblem $k$ is given as \small \begin{align} e_k = 4b_k + 4a_k + \frac{(|C_k|+2)(|C_k|+3)}{2}. \end{align} \normalsize Let us define $U_k = C_k \cap C_{\parent(k)}$. The variables that are shared between agent $k$ and its parent are given as $x^i_s$ for $i \in U_k$ and $\vectri(S_{U_kU_k})$. The number of these variables is then $s_k = 2|U_k| + |U_k|(|U_k|+1)/2$. The number of variables that agent $k$ does not share with its parent is then $r_k = n_k - s_k$. Each agent in order to compute the message to its parent, needs to factorize a symmetric indefinite matrix, see \cite[Sec. 6.2]{kho:15c}. The size of this matrix depends on the number of equality constraints for its subproblem and the variables it does not share with its parent. Hence, the size of this matrix is given by $r_k + e_k$. Moreover recall that the messages are quadratic functions of the variables that are shared between two agents. Consequently, each agent in order to communicate this functional to its parent would need to send $s_k(s_k+1)/2+s_k$ scalars to its parent. We can now summarize the dominant computational and communication burden for each agent with the following items. \begin{itemize} \item The size of the matrix that needs to be factorized by each agent $k$ grows quadratically with the number of sensors assigned to the agent and linearly with the number of range measurements assigned to it. This number is also reduced quadratically with the number of variables that this agent shares with its parent. \item The size of the information that each agent needs to communicate to its parent grows quadratically with the number of variables it shares with the parent. \end{itemize} \begin{rem} Notice that these summarizing items also provide guidelines on how to devise heuristics to perform a better clustering of sensors. They also enable us to propose improvements to the measurement assignment strategy, in order to distribute the computations among agents in a more balanced manner. Despite this, for the sake of brevity and simplicity, such heuristics are not considered in this study. \end{rem} Next we investigate the performance of our proposed algorithm, using two sets of numerical experiments. \section{Numerical Experiments}\label{sec:numerical} In this section we compare the performance of our proposed distributed algorithm with that of presented in \cite{soa:15}. We refer to this algorithm as distributed disk relaxation algorithm (DDRA). To this end, we conduct two sets of experiments, one that relies on simulated data and one that is based on real data from \cite{pat:03}. Notice that we do not conduct a comparison with other algorithms, since a thorough comparison with DDRA has been conducted in \cite{soa:15}, which illustrated the superiority of their proposed algorithm to high performance algorithms in \cite{gho:13} and \cite{sim:14} both in accuracy and number of communications among agents. \subsection{Experiments Using Simulated Data}\label{sec:numerical-A} Our experiments based on simulated data concern networks of sensors with connected inter-sensor measurement graphs. In all experiments there are 9 anchors in the network which are uniformly distributed in the area. The experiments in this section are divided into two setups. In both setups, we consider a network of several sensors which are placed in a two-dimensional area, with their locations randomly generated using a uniform distribution. The noisy range measurements are generated as \begin{align*} \mathcal R_{ij} &= \left|\| (x^*_s)^i - (x^*_s)^j \|_2 + E_{ij}\right|, \quad j \in \text{Ne}_r(i),\\ \mathcal Y_{ij} &= \left|\| (x^*_s)^i - (x_a)^j \|_2 + V_{ij}\right|, \quad j \in \text{Ne}_a(i), \end{align*} where $(x^*_s)^i$ denotes the true location of the $i$th sensor. Furthermore we assume that all noises are gaussian and mutually independent, see also \cite{soa:15}. In the first setup we conduct experiments using a network 50 sensors in a $0.8 \times 0.8$ area. We consider four different measurement noise standard deviations, namely $0.01, 0.05$, $0.1$ and $0.3$, and for each noise level we generate 50 problem instances. In order to ensure that the generated inter-sensor measurement graph is loosely connected, we assume there exist a measurement between two sensors or between a sensor and an anchor if the distance between them is less than the communication range $r_c =0.2$. \begin{figure}[t] \begin{center} \includegraphics[width=6.5cm]{Network.pdf} \caption{\footnotesize The sensor network considered for our experiment. Each red cross depicts one of the 50 sensors in the network and each green circle marks one of the 9 anchors. An edge between two nodes, implies existence of a range measurement between the two nodes.\normalsize} \label{fig:network} \vspace{-5mm} \end{center} \end{figure} The resulting sensor network is depicted in Figure \ref{fig:network}. In this figure, the sensor nodes are marked with red crosses and the anchors are marked with green circles. As can be seen from the figure the inter-sensor measurement graph is connected. The performance of distributed algorithms are quantified using three measures. Namely (i) their accuracy based on the root mean squared error (RMSE) defined as \begin{align} \textrm{RMSE} = \sqrt{\frac{1}{MN}\sum_{i=1}^M\sum_{j=1}^N \| (x^*_s)^j - x_s^j(m) \|^2} \end{align} where $M$ is the number of experiments and the argument $m$ marks the computed estimate for the $m$th experiment, (ii) number of required iterations and communications to converge to a solution with a given accuracy and (iii) the computational time. Notice that both algorithms are run in a centralized manner. The algorithm in \cite{soa:15} is terminated if the norm of the gradient of its considered cost function is below $10^{-6}$. This threshold was chosen based on the authors experience, so as to guarantee DDRA generates accurate enough solutions. Figures~\ref{fig:ex11}--\ref{fig:ex13} illustrate the achieved results. In these figures and the ones to come the $*$-marked curves illustrate the results from DPDLA, whereas the o-marked curves show the results from DDRA. \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex1-RMSEL-shaded.pdf} \caption{\footnotesize The RMSE results from the considered algorithms when applied to a network of 50 sensors, depicted in Figure \ref{fig:network}, with four different measurement noise standard deviation, namely $0.01, 0.05$, $0.1$ and $0.3$.\normalsize} \label{fig:ex11} \vspace{-5mm} \end{center} \end{figure} As can be seen from Figure \ref{fig:ex11}, DPDLA outperforms or provides comparable accuracy with respect to DDRA for different levels of measurement noise. This shows the superiority of semidefinite relaxation to disk relaxation. The number of communications that each agent is required to conduct for each algorithm to converge to a solution is depicted in Figure~\ref{fig:ex12}. For these experiments, the considered clique tree for the inter-sensor measurement graph in Figure~\ref{fig:network}, has height~8, and the primal-dual method converged within around 10 iterations. As can be seen from the figure, DPDLA requires roughly two orders of magnitude less number of communications for computing a solution. The shaded areas depict the maximum and minimum values within the 50 instances for each of these quantities. Notice that this area for the results corresponding to DPDLA is not even visible. We can hence deduce that in comparison DDRA, the number of communications for DPDLA seems to be much less sensitive to the noise level and also to data realizations. \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex1-Iter-shaded.pdf} \caption{\footnotesize The number of communications that each agent needs to conduct for each of the algorithms to converge to a solution. The sensor network consists of 50 sensors, depicted in Figure \ref{fig:network}, with three different measurement noise standard deviation, namely $0.01, 0.05$, $0.1$ and~$0.3$. \normalsize} \label{fig:ex12} \vspace{-5mm} \end{center} \end{figure} \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex1-Time-shaded.pdf} \caption{\footnotesize The required time to converge for each of the considered algorithms when applied to a network of 50 sensors, depicted in Figure \ref{fig:network}, with four different measurement noise standard deviation, namely $0.01, 0.05$, $0.1$ and~$0.3$. \normalsize} \label{fig:ex13} \vspace{-5mm} \end{center} \end{figure} The computational time for the considered algorithms are presented in Figure \ref{fig:ex13}. As can be seen from the figure DDRA is at least twice as fast as DPDLA, owing to very simple computations required from each agent at every iteration. This is the case if both algorithms are executed in a centralized manner and if we neglect the communication cost or delay. Based on the presented results, our proposed algorithm provides more accurate estimates, and even though slower when implemented in a centralized manner, it provides a better distributed algorithm as it requires far less amount of communications. \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex2-RMSE-shaded.pdf} \caption{\footnotesize The RMSE results from the considered algorithms when applied to networks of varying number of sensors, with measurement noise standard deviation of 0.01.\normalsize} \label{fig:ex21} \vspace{-7mm} \end{center} \end{figure} In the second simulation setup, we test the performance of the considered algorithms, when applied to networks with varying number of sensors, namely, 10, 20, 30, 40 and 50. In this setup we assume that the measurement noise standard deviation is 0.01, and we consider 50 instances for each network size. Furthermore the size of the considered area and the communication range, $r_c$, for each network size are chosen such that the resulting inter-sensor measurement graphs are connected but loosely. Figure \ref{fig:ex21} illustrates the RMSE results for this experiment. As before, as can be seen from the figure, DPDLA provides more accurate estimates for all network sizes. \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex2-Iter-shaded.pdf} \caption{\footnotesize The number of communications that each agent needs to conduct for each of the algorithms to converge to a solution when applied to networks of varying number of sensors, with measurement noise standard deviation of 0.01. \normalsize} \label{fig:ex22} \vspace{-8mm} \end{center} \end{figure} \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex2-Time-shaded.pdf} \caption{\footnotesize The required time for each of the considered algorithms to converge when applied to networks of varying number of sensors, with measurement noise standard deviation of 0.01. \normalsize} \label{fig:ex23} \vspace{-7mm} \end{center} \end{figure} Also as can be seen from Figure \ref{fig:ex22}, the estimates are computed using far fewer communications among agents. The primal-dual method converged within around 11 iterations and the heights of the clique trees for the different sensor networks ware between 3 to 8. As can be seen from the figure, the number of required communications for the DDRA to converge grows much faster with network size than that of DPDLA which seems to be far less sensitive to this change. Figure \ref{fig:ex23} illustrates the total computational time of both algorithms when implemented in a centralized manner. As can be seen from this figure, our proposed algorithm requires similar or less amount of time to converge to a solution for networks of up to 30 sensors. Consequently, for networks with less than 30 sensors, our proposed algorithm outperforms DDRA in all the performance criteria. It is also worth mentioning that, the performance of our algorithm can be improved considerably, if the clustering of the sensors and generation of a clique tree are done using more sophisticated and tailored approaches. However, since we did not discuss such approaches, we abstained from any manipulation of the cliques and the clique tree and simply relied on standard and simple heuristics for this purpose, see e.g., \cite{kho:15c} and references~therein. \subsection{Experiments Using Real Data} In this section, we present the results from conducted experiments based on real data. This data was taken from \cite{pat:03}, that includes time of arrival (TOA) measurements among 44 sensors, 4 of which are deemed to be anchors. The sensors are spread out in a $14 \times 13$ area. We extract the range measurements from the available TOA measurements. This provides us with biased range measurements with a standard deviation of 1.82 meters, see \cite{pat:03}. We here study the performance of DDRA and DPDRA for different levels of connectivity of the inter-sensor graph. To this end, we gradually change the communication range from 4 to 6.5 meters. Figures \ref{fig:ex31} and \ref{fig:ex33} illustrate the results. \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex3-RMSE.pdf} \caption{\footnotesize The RMSE results from the considered algorithms when applied to a localization problem based on real data, with a varying communication range. The $*$-marked line illustrates the RMSE results from DPDLA, whereas the o-marked line shows the RMSE results from DDRA.\normalsize} \label{fig:ex31} \vspace{-3mm} \end{center} \end{figure} Notice that due to biasedness and quality of the measurements, the intersection of the range measurement disks can be empty and hence DDRA fails to converge. This is because the gradient of the cost function of the disk relaxation problem does not vanish. Consequently, this algorithm has been terminated after 5000 iterations. Figure \ref{fig:ex31} illustrates the RMSE results from the experiment, which clearly depicts that DPDLA outperforms DDRA. \begin{figure}[t] \begin{center} \includegraphics[width=7.5cm]{Ex3-Time.pdf} \caption{\footnotesize The required time for each of the considered algorithms to converge when applied to a localization problem based on real data, with a varying communication range.\normalsize} \label{fig:ex33} \vspace{-5mm} \end{center} \end{figure} Furthermore, DPDLA required each agent was required to communicate with its neighbors around 100 times which seemed to be robust with respect to the level of connectivity of the inter-sensor range measurement graph. The primal-dual method for all these instances converged within roughly 17 iterations and the height of the clique tree varied between 7 to 9. Figure~\ref{fig:ex33} illustrates the computational time for DDRA and DPDLA. As was also observed from the experiments in Section \ref{sec:numerical-A}, DDRA clearly outperforms DPDLA when implemented in a centralized manner. \section{Conclusions}\label{sec:conclusions} In this paper we proposed a distributed localization algorithm for tree-structured scattered sensor networks founded on semidefinite relaxation of the localization problem. This algorithm is based on state-of-the-art primal-dual interior-point methods and relies on message-passing or dynamic programming over trees to distribute the computations. Due to this, the resulting algorithm requires far fewer steps and even fewer communications among computational agents to converge to an accurate solution, and it achieves this by putting a moderate computational burden on the agents. Furthermore, the proposed distributed algorithm is robust to biases in the measurements, or in general bad quality of the measurements. This stems from the power of semidefinite relaxation for localization problems. Despite these advantages, the proposed algorithm is much more complicated than algorithms that rely on first-order methods. This is largely due to the fact that generally second-order methods are far more complicated than their first-order counter parts. The choice of clustering of the sensors and the strategy for assigning the available measurements to computational agents can have a significant effect on the performance of our proposed algorithm. Also smart clustering of the sensors, may even enable us to use the computational infrastructure at the anchors and utilize them as computational agents. In this paper, we briefly discussed the importance of this and provided some suggestions on how the used heuristic strategies for this purpose can be improved. We did not investigate this topic in detail, however, we believe that further exploration of this matter can result in interesting results. Furthermore, distributed approaches for computing cliques and clique trees of the inter-sensor measurement were not covered in this paper, although, complementing the proposed algorithm with such methods can enhance the practicality of the algorithm. \vspace{-10pt} \ifCLASSOPTIONcaptionsoff \newpage \fi \bibliographystyle{plain}
d8a5e808b3b90b0513cf0c72dcfd2e3df48814d2
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Quantum theory of gravity is one of the most sought-after goals in physics. Despite continuous efforts to tackle this important problem, resulting in interesting proposals such as superstring theory and loop quantum gravity, there is still no clear sign of the theory of quantum gravity \cite{Giulini,Rovelli,Kiefer}. However, when the back-action of matter on the gravitation field is neglected, one can write down a theory of quantum fields in a background curved spacetime by extending quantum field theory in Minkowski metric to a general metric \cite{Birrell, Fulling, Mukhanov, Parker}. This is analogous to treating external fields as c-numbers and predicts interesting new phenomena that are valid in appropriate regimes. Similarly to the prediction of pair creation in external electric fields \cite{Sauter, HeisenbergEuler1936, Schwinger}, external gravitational fields (as described by a background spacetime) induce particle creation \cite{Parker1971, Hawking1975, Unruh1976}. In the latter, particle creation is caused by the change in the vacuum state itself under quite generic conditions. In this work, we propose a classical optical simulation of particle creation in binary waveguide arrays. There have been many proposals and experimental demonstrations of a plethora of interesting physics in coupled waveguide arrays \cite{Peschel, Morandotti, Lahini, Longhi11, Longhi12, Crespi, Keil, Rodriguez13, Rodriguez14, LeeAngelakis14, Marini14, RaiAngelakis15, Keil15}. In particular, optical simulation of the 1+1 dimensional Dirac equation in binary waveguide arrays has been proposed \cite{Longhi10a,Longhi10b} and experimentally demonstrated \cite{Dreisow10,Dreisow12}. We show that the setup can be generalised to also simulate the Dirac equation in 2 dimensional curved spacetime. Particle creation is by definition a multi-particle phenomenon and the full simulation of the result requires quantum fields as a main ingredient (for example, see \cite{Boada} for a simulation of the Dirac equation in curved spacetime with cold atoms on optical lattices). However, light propagation in a waveguide array is an intrinsically classical phenomena, so how can we simulate particle creation in a binary waveguide array? The short answer is that we will be looking at a single-particle analog of particle creation. As we will see, the fundamental reason behind particle creation is the difference in the vacuum state, which in turn is captured by different mode-expansions of quantum fields. We can thus concentrate on a single-mode at a time and simulate the effect. In fact, the well-known Klein paradox shows that the single-particle Dirac equation contains subtle hints of multi-particle effects, and the phenomenon of pair production in strong electric fields has been studied within the single particle picture \cite{Ruf}. We study the time evolution of spinor wave packets and demonstrate that an analog of particle creation can be {\it visualised} in the light evolution in a binary waveguide array. Here, we stress that by using the phenomenon of `zitterbewegung' (the jittering motion of a Dirac particle), one can bypass the quantitative checks in `proving' the simulation of particle creation. This article is organised as follows. In section II, we provide a pedagogical introduction to the Dirac equation in curved spacetime, assuming familiarity with the conventional Dirac equation. In section III, we specialise to the 1+1 dimensions and provide a few examples of spacetime metrics. Particle creation in curved spacetime is explained, using scalar fields for simplicity, and the single-particle analogs for the Dirac spinors is discussed. Section IV shows the wave packet evolution both in flat and curved spacetimes, using time-dependent gravitational fields as an example. A single-particle analog of particle creation in a particular case is explicitly demonstrated. Section V explains the optical simulation of the Dirac equation in a binary waveguide array and proposes a generalisation to curved spacetimes. We conclude in section VI. We have tried to be as pedagogical and self-contained as we could. We have tried to collect and present essential ideas to understand quantum fields in curved spacetime and how it predicts particle creation. It is our hope that the reader will find this helpful in understanding the essential ideas quickly and develop further interesting analogies. \section{Dirac equation in curved spacetime} Let us start with the derivation of the Dirac equation in curved spacetime. This requires the notion of spin-connection, which will be discussed at a pedagogical level. We closely follow ref.~\cite{Lawrie}. \subsection{General covariance} Special theory of relativity has taught us that time and space are observer-dependent concepts in that observers moving relative to each other have different notions of time and space intervals. This comes about because the laws of physics are covariant under a Lorentz transformation, meaning that all observers agree on the form of physical laws in their own coordinate frames. General theory of relativity takes this one step further and states that the laws should be covariant under general coordinate transformations. Equations of motion are written in terms of tensors--quantities that are independent of local coordinate systems used to describe them. To write down the equations of motion one also requires the notion of a covariant derivative, which basically is the correct way of differentiating tensors to yield another tensor (of higher rank). To understand this, consider a vector in Minkowski spacetime $V^\mu$, which transforms under the Lorentz transformation as $V^{\mu'} = \Lambda \indices{^{\mu' }_{\mu}} V^\mu$. We use the Einstein convention where the repeated indices are summed over. Now let's see how the partial derivative of a vector transforms in flat spacetime: \begin{align} \partial_{\nu '}V^{\mu '} = \Lambda \indices{_{\nu '}^{\nu}}\partial_{\nu} \left( \Lambda \indices{^{\mu '}_{\mu}} V^\mu \right) = \Lambda \indices{_{\nu '}^{\nu}}\Lambda \indices{^{\mu '}_{\mu}} \partial_{\nu} V^\mu. \end{align} In tensorial language, one says that $\partial_{\nu}V^{\mu}$ transforms as a tensor of rank (1,1), where $(n,m)$ denote $n$ indices on the top and $m$ indices on the bottom. In curved spacetime, this is no longer true because $ \Lambda \indices{_{\nu '}^{\nu}} \rightarrow \tfrac{\partial x^{\nu '}}{\partial x^{\nu}}$ is generically a spacetime dependent quantity. The partial differential operator must therefore be generalised (to a covariant derivative), so that the `differentiated' object is also a tensor. This requires the notion of parallel transport. \subsection{Parallel transport and affine connection} Mathematically, one has to be careful when taking a derivative of a vector because the definition of a vector along a curve defined by $\lambda$ \begin{align} \frac{d V^\mu}{d\lambda} = {\rm lim}_{\delta\lambda \rightarrow 0} \frac{V^\mu(Q) - V^\mu(P)}{\delta\lambda}, \end{align} where $Q$ and $P$ are spacetime points at $\lambda + \delta\lambda$ and $\lambda$ respectively, require comparison between two vectors in different tangent spaces. This is okay in flat spaces, but in curved spaces there is no intrinsic way to do it. What we need is a concept of parallel transport that moves a vector--or more generally a tensor--along a curve while keeping it `constant'. $V^\mu(P)$ can be parallel transported to $Q$ with the help of the `affine connection' $\Gamma$ such that \begin{align} \label{paralleltransport} V^\mu(P\rightarrow Q) = V^\mu(P) - \delta\Lambda \Gamma^\mu_{\nu\sigma}(P) V^\nu(P)\frac{dx^\sigma}{d\lambda}. \end{align} The corresponding covariant derivative is written as \begin{align} \nabla_\sigma V^\mu = \frac{V^\mu(Q) - V^\mu(P\rightarrow Q)}{\delta x^\sigma} = \partial_\sigma V^\mu + \Gamma^\mu_{\nu\sigma}V^\nu, \end{align} with which we can define the parallel transport condition as \begin{align} \frac{dx^\sigma}{d\lambda}\nabla_\sigma V^\mu = 0. \end{align} Note that this is a generalisation of the condition in flat space $\tfrac{\partial V^\mu}{\partial x^\sigma} = 0$. Covariant derivative of a one-form, i.e.~a (0,1) tensor $\omega_\mu$ is found to be \begin{align} \nabla_\sigma \omega_{\nu} = \partial_\sigma \omega_\mu - \Gamma^\lambda_{\sigma\nu}\omega_\lambda. \end{align} The transformation properties of a connection can be found by demanding that $\nabla_\sigma V^\mu$ transforms as a tensor, i.e., \begin{align} \nabla_{\sigma '} V^{\mu '} = \frac{\partial x^\sigma}{\partial x^{\sigma '}} \frac{\partial x^{\mu '}}{\partial x^{\mu}} \nabla_\sigma V^\mu. \end{align} This yields \begin{align} \Gamma^{\mu'}_{\nu'\sigma'} = \frac{\partial x^{\nu}}{\partial x^{\nu'}}\frac{\partial x^{\sigma}}{\partial x^{\sigma'}}\frac{\partial x^{\mu'}}{\partial x^{\mu}}\Gamma^{\mu}_{\nu\sigma} - \frac{\partial x^{\nu}}{\partial x^{\nu'}}\frac{\partial x^{\sigma}}{\partial x^{\sigma'}}\frac{\partial}{\partial x^\sigma}\left( \frac{\partial x^{\mu'}}{\partial x^{\nu}}\right), \end{align} showing that $\Gamma$ is not a tensor. \subsection{The metric connection} The above consideration shows that the concept of parallel transport requires an affine connection to be defined. This is a very general property irrespective of the detailed structure of the manifold, allowing many distinct definitions of parallel transport. In general relativity, however, it is possible define a unique connection compatible with the metric $g_{\mu\nu}$ (remember, $ds^2 = g_{\mu\nu}dx^\mu dx^\nu$) as follows. First, the connection is assumed to be torsion-free, meaning that $\Gamma^\lambda_{\mu\nu} = \Gamma^\lambda_{\nu\mu}$. Second, the metric is assumed to obey the parallel transport condition: $\nabla_\sigma g_{\mu\nu} = 0$. The latter guarantees that the scalar product of two parallel-transported vectors is constant. That is, if $\tfrac{dx^\sigma}{d\lambda}\nabla_\sigma U^\mu = \tfrac{dx^\sigma}{d\lambda}\nabla_\sigma V^\nu = 0$, then $\tfrac{dx^\sigma}{d\lambda}\nabla_\sigma(g_{\mu\nu}U^\mu V^\nu) = 0$. From the second assumption, we have the following three relations \begin{align} \nabla_\rho g_{\mu\nu} = \partial_\rho g_{\mu\nu} - \Gamma^{\lambda}_{\rho\mu}g_{\lambda\nu}- \Gamma^{\lambda}_{\rho\nu}g_{\mu\lambda} = 0, \nonumber \\ \nabla_\mu g_{\nu\rho} = \partial_\mu g_{\nu\rho} - \Gamma^{\lambda}_{\mu\nu}g_{\lambda\rho}- \Gamma^{\lambda}_{\mu\rho}g_{\nu\lambda} = 0, \nonumber \\ \nabla_\nu g_{\rho\mu} = \partial_\nu g_{\rho\mu} - \Gamma^{\lambda}_{\nu\rho}g_{\lambda\mu}- \Gamma^{\lambda}_{\nu\mu}g_{\rho\lambda} = 0. \end{align} Subtracting the second and third lines from the first and using the first assumption, we obtain \begin{align} \partial_\rho g_{\mu\nu} - \partial_\mu g_{\nu\rho} - \partial_\nu g_{\rho\mu} + 2\Gamma^{\lambda}_{\mu\nu}g_{\lambda\rho} = 0, \end{align} which after some rearranging yields \begin{align} \Gamma^\sigma_{\mu\nu} = \frac{1}{2}g^{\sigma\rho}\left( \partial_\mu g_{\nu\rho} + \partial_\nu g_{\rho\mu} - \partial_\rho g_{\mu\nu} \right). \end{align} This is the metric connection, called the Christoffel symbol. \subsection{Spin connection} So far, we have seen how to take covariant derivatives of tensors. However, this is not enough to write down the Dirac equation in a curved spacetime. We also need to know how to take the covariant derivative of a spinor. For this purpose, we will use the fact that locally it is always possible (due to the equivalence principle) to find an inertial coordinate system in which the metric becomes Minkowskian. Suppose that $y^a$ are such local coordinates at point $x^\mu = X^\mu$ (we use the convention that latin indices $a,b,...$ are used to label local inertial coordinates and greek indices $\mu,\nu,...$ for general coordinates). Then defining \begin{align} e\indices{^\mu_a}(X) &= \frac{\partial x^\mu}{\partial y^a} \Bigg |_{x^\mu = X^\mu} \;\;\;\;\; {\rm and} \nonumber \\ e\indices{^a_\mu}(X) &= \frac{\partial y^a}{\partial x^\mu} \Bigg |_{x^\mu = X^\mu} \end{align} at each point in spacetime, we get the {\it vielbein} (`many-legs') $e^\mu_a(x)$ that diagonalises the metric, as well as its inverse \begin{align} e\indices{^\mu_a}(x)e\indices{^\nu_b}(x)g_{\mu\nu}(x) &= \eta_{ab}, \nonumber \\ e\indices{^a_\mu}(x)e\indices{^b_\nu}(x)\eta_{ab}(x) &= g_{\mu\nu}. \end{align} What we want is to know what the parallel transport equation (\ref{paralleltransport}) looks like in the local inertial coordinate system. Basically, we want \begin{align} \label{spintransport} V^a(x \rightarrow x + dx) = V^a(x) - \omega\indices{^a_b_\nu}V^b(x)dx^\nu, \end{align} where $\omega\indices{^a_b_\nu}$ is a generalisation of the affine connection called the spin connection. Noting that \begin{align} V^\mu(x) &= e\indices{^\mu_a}(x)V^a(x), \nonumber \\ V^\mu(x\rightarrow x+dx) &= e\indices{^\mu_a}(x+dx)V^a(x\rightarrow x+dx), \end{align} and $e\indices{^\mu_a}(x+dx) \approx e\indices{^\mu_a}(x) + \partial_\nu (e\indices{^\mu_a}(x))dx^\nu$, we can obtain the spin connection in terms of the affine connection \begin{align} \label{spinconnection} \omega\indices{^a_b_\nu} = e\indices{^a_\mu} \partial_\nu (e\indices{^\mu_b}) + e\indices{^a_\mu}e\indices{^\sigma_b}\Gamma^\mu_{\sigma\nu}. \end{align} This allows one to take the covariant derivative of a tensor with mixed indices. For example \begin{align} \nabla_\nu e\indices{^\mu_a} = \partial_\nu e\indices{^\mu_a} + \Gamma^\mu_{\sigma\nu}e\indices{^\sigma_a} - \omega \indices{^b_a_\nu} e\indices{^\mu_b} = 0, \end{align} where the last equality follows from (\ref{spinconnection}). The spin connection is antisymmetric in the first two indices, i.e., $\omega_{ab\nu} = -\omega_{ba\nu}$, so that the magnitude of the Lorentz vector remains constant upon parallel transport. Using the spin connection we can derive the covariant derivative operator for spinors. The latter is defined through the parallel transport equation \begin{align} \psi(x\rightarrow x + dx) = \psi(x) - \Omega_\nu(x)\psi(x)dx^\nu, \end{align} where $\Omega_\nu$ is an n-by-n matrix for each index $\nu$ with n=2 for 2 and 3 spacetime dimensions and 4 for 4-dimensional spacetime. To determine $\Omega_\nu$, we use the fact that $S(x) = \bar{\psi}(x)\psi(x)$ and $V^a(x) = \bar{\psi}(x)\gamma^a\psi(x)$ transform as a scalar and a vector, respectively. Firstly, \begin{align} S(x\rightarrow & x + dx) - S(x) \nonumber \\ &= \bar{\psi}(x)\left[ \gamma^0\Omega^\dagger_\nu(x)\gamma^0 + \Omega_\nu(x)\right]\psi(x) dx^\nu, \end{align} yielding \begin{align} \label{cond1} \gamma^0\Omega^\dagger_\nu(x)\gamma^0 =- \Omega_\nu(x). \end{align} Secondly, \begin{align} V^a(x\rightarrow & x + dx) - V^a(x) \nonumber \\ &= - \bar{\psi}(x)\left[ \gamma^a\Omega_\nu +\gamma^0\Omega^\dagger_\nu\gamma^0\gamma^a\right]\psi(x) dx^\nu, \nonumber \\ &= - \bar{\psi}(x)\left[ \gamma^a\Omega_\nu - \Omega_\nu\gamma^a\right]\psi(x) dx^\nu, \nonumber \\ &= - \omega\indices{^a_b_\nu}\gamma^bdx^\nu. \end{align} The second equality results from (\ref{cond1}), while the third equality results from the definition of the spin connection. From the second and third lines we conclude that \begin{align} \label{cond2} [\gamma^a,\Omega_\nu] = \omega\indices{^a_b_\nu}\gamma^b. \end{align} Noting that \begin{align} [\gamma^a,\sigma^{bc}] = 2i(\eta^{ab}\gamma^c - \eta^{ac}\gamma^b), \end{align} where $\sigma^{bc} = i[\gamma^b,\gamma^c]/2$, one can verify by direct substitution that \begin{align} \Omega_\nu(x) = -\frac{i}{4}\omega_{ab\nu}(x)\sigma^{ab} \end{align} satisfies (\ref{cond2}), while (\ref{cond1}) can be verified using the relationship $\gamma^0\gamma^{a\dagger}\gamma^0 = -\gamma^a$. Using the covariant derivative $\nabla_\nu = \partial_\nu + \Omega_\nu$, the flat spacetime Dirac equation (c=1) \begin{align} \left[i\gamma^a\partial_a\psi -m \right] \psi(x) = 0, \end{align} generalises to the curved spacetime Dirac equation \begin{align} \label{curveddiraceqn} \left[ i\gamma^\mu\nabla_\mu -m\right]\psi(x) = 0, \end{align} where the vielbein was used to transform the local $\gamma$ matrices: $\gamma^\mu = e\indices{^\mu_a}(x)\gamma^a$. Note that $\{ \gamma^\mu(x),\gamma^\nu(x) \} = 2g^{\mu\nu}(x)$. From here on, we will use the notation $\tilde{\gamma}^a \equiv \gamma^a$ to avoid confusion when numerical indices are substituted. \section{Dirac equation in 2 dimensional curved spacetime} A special feature of the 2 dimensional spacetime is that the metric can always be reduced to the conformally flat form \begin{equation} ds^2=\Omega^2(dt^2-dx^2) \end{equation} for some function $\Omega(x,t)$. To derive the Dirac equation in this metric we first need the Christoffel symbols which are readily calculated to be: $\Gamma^0_{00}=\Gamma^0_{11}=\Gamma^1_{10}=\Gamma^1_{01}=\dot{\Omega}/\Omega$ and $\Gamma^0_{01}=\Gamma^0_{10}=\Gamma^1_{00}=\Gamma^1_{11} =\Omega'/\Omega$. The dot denotes a derivative with respect to time and the prime a spatial derivative. Using these, the vielbein can be readily calculated: $e\indices{^0_0}= e\indices{^1_1}= 1/\Omega$. These lead to non-vanishing spin connections $\omega\indices{^0_1_0}=\omega\indices{^1_0_0}= \Omega'/\Omega$ and $\omega\indices{^0_1_1}=\omega\indices{^1_0_1}= \dot{\Omega}/\Omega$, which in turn lead to $\Omega_0= \Omega'/(4\Omega) [\tilde{\gamma}^0,\tilde{\gamma}^1]$ and $\Omega_1= \dot{\Omega}/(4\Omega) [\tilde{\gamma}^0,\tilde{\gamma}^1]$. That this spacetime is curved can be verified by calculating the Ricci curvature $R = g^{\mu\nu}R_{\mu\nu} = 2\left( (\dot{\Omega}/\Omega)^2 - \ddot{\Omega}/\Omega\right)/\Omega^2$, where $R_{\mu\nu} = \partial_\lambda \Gamma^{\lambda}_{\mu\nu} - \partial_\nu \Gamma^{\lambda}_{\mu\lambda} + \Gamma^{\lambda}_{\mu\nu}\Gamma^{\sigma}_{\lambda\sigma}- \Gamma^{\lambda}_{\mu\sigma}\Gamma^{\sigma}_{\nu\lambda}$ is the Ricci tensor. We can thus write down the Dirac equation in a general 1+1 dimensional spacetime. Inserting the above results into Eq.~(\ref{curveddiraceqn}), multiplying with $\Omega \tilde{\gamma}^0$ from the left, and rearranging, we get \begin{equation} \begin{split} i(\partial_t+\tilde{\gamma}^0\tilde{\gamma}^1\frac{[\tilde{\gamma}^0,\tilde{\gamma}^1]}{4}\frac{\dot{\Omega}}{\Omega})\psi &= i[\tilde{\gamma}^0\tilde{\gamma}^1\partial_x + \frac{[\tilde{\gamma}^0,\tilde{\gamma}^1]}{4} \frac{\Omega'}{\Omega} ]\psi \\ &\quad+\tilde{\gamma}^0\Omega m\psi. \end{split} \end{equation} Choosing $\tilde{\gamma}^0 = \sigma_z$ and $\tilde{\gamma}^1= i\sigma_y$, the Dirac equation becomes \begin{equation} i\left(\partial_t+\frac{\dot{\Omega}}{2\Omega}\right)\psi= \left[-i\sigma_x\left(\partial_x + \frac{\Omega'}{2\Omega} \right)+\sigma_z\Omega m\right]\psi(x). \end{equation} Which, upon defining $\sqrt{\Omega} \xi= \psi$, can be recast into our final form: \begin{equation} \label{curvedDirac} i\partial_t(\Omega\xi)= -i\sigma_x\partial_x(\Omega\xi)+\sigma_z \Omega m (\Omega\xi). \end{equation} $\Omega \xi$ thus solves the regular Dirac equation with an effective mass of $\Omega m$ which, as we show below, means that it can be simulated in binary waveguide arrays. Note that in flat spacetime $\Omega = 1$ and the equation reduces to \begin{align} \label{flateqn} i\partial_t \psi= -i\sigma_x\partial_x\psi+\sigma_z m \psi. \end{align} \subsection{Specific examples} Here we provide examples of specific spacetimes and corresponding Dirac equations. \subsubsection{Static spacetime} Static spacetimes are described by a metric that has a time-like variable $x_0$ such that $g_{0i} = 0$ and $\partial_0 g_{\mu \nu} = 0$. In 1+1 dimensions the line element for these spacetimes may be written as \begin{equation} ds^2=e^{2\Phi}dt^2-e^{2\Psi}dx^2, \end{equation} where $\Phi$ and $\Psi$ are independent of $t$. Non-vanishing Christoffel symbols for this metric are $\Gamma^0_{10}=\Gamma^0_{01}= \Phi'$, $\Gamma^1_{00}= \Phi'e^{2(\Phi-\Psi)}$, and $\Gamma^1_{11}= \Psi'$, whereas the vielbein is easily found to be $e\indices{^0_0}= e^{-\Phi}$, $e\indices{^1_1}= e^{-\Psi}$. Then the non-vanishing spin connections are $\omega\indices{^1_0_0}=\omega\indices{^0_1_0}= \Phi'e^{\Phi-\Psi}$, leaving us with the non-vanishing element of $\Omega_\nu$: $\Omega_0 = \frac14\Phi'e^{\Phi-\Psi}[\tilde{\gamma}^0,\tilde{\gamma}^1]$. Inserting the results and using the same gamma-matrices as above, we can write the Dirac equation as \begin{equation} i\partial_t \psi =-ie^{\Phi-\Psi}\sigma_x\left(\partial_x+\frac{\Phi'}{2} \right)\psi + e^{\Phi}\sigma_z m\psi. \end{equation} \subsubsection{FRW Metric} The FRW metric in 1+1 D reads \begin{equation} ds^2=dt^2-a^2(t) dx^2. \end{equation} The non-vanishing spin connections, Christoffel symbols, and $\Omega_{\nu}$ are $e\indices{^0_0}= 1$, $e\indices{^1_1}= 1/a$; $\Gamma^0_{11}= a\dot{a}$; $\Gamma^1_{10}=\Gamma^1_{01}= \dot{a}/{a}$; $\omega\indices{^0_1_1}=\omega\indices{^1_0_1}= \dot{a}$; $\Omega_1= \frac{\dot{a}}{4}[\tilde{\gamma}^0,\tilde{\gamma}^1]$. The Dirac equation reads \begin{equation} i(\tilde{\gamma}^0 \partial_0 \psi + \tilde{\gamma}^1\frac1a \partial_x \psi+\tilde{\gamma}^1\frac{\dot{a}}{4a}[\tilde{\gamma}^0,\tilde{\gamma}^1] \psi)-m\psi = 0. \end{equation} After plugging in $\tilde{\gamma}^a$, we obtain \begin{equation} i\partial_t \psi = -i\frac{\dot{a}}{2a} \psi -i\frac{\sigma_x}{a} \partial_x \psi+\sigma_z m\psi. \end{equation} Note that the FRW metric can be converted to the conformally flat form by setting $\eta(t) = \int^t \frac{dt}{a(t)}$. \subsubsection{Rindler spacetime} Rindler metric describes the dynamics of a uniformly accelerating observer in flat spacetime. Even though the spacetime is flat, Unruh showed that spontaneous particle creation occurs in the frame of the accelerating observer \cite{Unruh1976} analogously to the famous Hawking radiation \cite{Hawking1975}. The vielbein formalism described above proves useful for deriving the wave equation in this metric. By introducing the coordinate $t = u\sinh v$ and $x = u \cosh v$, the Minkowski metric is converted to \cite{Soffel1980} \begin{align} ds^2 = u^2dv^2 - du^2, \end{align} when $u$ is constrained to be positive. Because this is a static spacetime, we can use the formula derived above to obtain the Dirac equation in the Rindler spacetime: \begin{align} i\partial_v\psi = -iu\sigma_x\left( \partial_u +\frac{1}{2u} \right)\psi + mu\sigma_z\psi. \end{align} The metric can also be put in a conformally flat form by introducing new coordinates \begin{align} t = \frac{1}{a}e^{a\xi}\sinh (a\eta), \;\;\; x = \frac{1}{a}e^{a\xi}\cosh (a\eta), \end{align} in the region $x>|t|$. In these coordinates, the metric takes the form \begin{align} ds^2 = e^{2a\xi}\left( d\eta^2 - d\xi^2\right), \end{align} and the Dirac equation reads \begin{align} i\partial_\eta\psi = \left[ -i\sigma_x\left( \partial_\xi + \frac{1}{2}\right) + \sigma_z e^{a\xi}m \right]\psi. \end{align} \subsection{Particle creation in curved spacetime} \label{sect3b} It is well established that there is particle creation in curved spacetime in general \cite{Birrell, Fulling, Mukhanov, Parker}. The background metric acts in a similar manner to an external field such as, for example, the electromagnetic field, and something akin to the Schwinger effect (creation of electron-positron pair in strong electric fields) occur. Consider as an example an expanding-universe scenario where the expansion is asymptotically turned on. The field is initially in the vacuum state and as expansion is gradually turned on, electrons and positrons pop out in pairs. The essence of understanding this phenomena is that a vacuum state is not unique. It is defined as an eigen-state of the field operator $\psi$ with eigenvalue 0. To define the latter more precisely, one has to expand the field operator in terms of mode functions and assign creation and annihilation operators that create and annihilate these modes. The vacuum is then the state with eigenvalue 0 for all the mode-annihilation operators. Now, these mode functions depend on the background spacetime, which means that the vacuum state of one spacetime need not be the vacuum state of another. Below, we provide a more detailed and pedagogical explanation of this effect for scalar fields in FRW spacetime, before we come back to Dirac fermions. \subsubsection{Scalar field in FRW spacetime} A scalar field obeying the Klein-Gordon equation is one of the simplest quantum fields to deal with and a spatially-flat, time-dependent metric is one of the simplest examples of curved spacetime metrics. As an example of such a metric, we choose the FRW metric and explain creation of scalar fields in it. We follow closely the exposition by Mukhanov and Winitzki \cite{Mukhanov}, but work in 2 dimensional spacetime instead of the usual 4 dimensional one. Remember that a real scalar field in Minkowski (or flat) spacetime obeys the Klein-Gordon equation \begin{align} \partial_\mu\partial^\mu \phi + m^2\phi = 0. \end{align} In curved spacetime this becomes (in the minimal coupling scheme) \begin{align} g^{\mu\nu}\nabla_\mu\nabla_\nu \phi + m^2\phi = 0. \end{align} To work out the explicit form of this equation, it is helpful to convert it to the following form \begin{align} g^{\mu\nu}\partial_\mu\partial_\nu\phi + \frac{1}{\sqrt{-g}}\left( \partial_\nu\phi\right) \partial_\nu \left( g^{\mu\nu}\sqrt{-g}\right) + m^2\phi = 0, \end{align} where $g$ is the determinant of the metric. Note that $g$ depends on the spacetime dimension: for the case of conformal spacetime described by $ds^2 = \Omega^2(dt^2 - dx^2 \cdots )$, $g = \Omega^2d$, where $d$ is the dimension of the spacetime. Working with the conformal version of the FRW metric, the wave equation evaluates to \begin{align} \ddot{\phi} - \phi '' + \Omega(\eta)^2m^2\phi = 0 \end{align} in 1+1D and \begin{align} \ddot{\phi} + 2\frac{\dot{\Omega}}{\Omega}\dot{\phi}- \nabla\phi + \Omega(\eta)^2m^2\phi = 0 \end{align} in 3+1D. As before, the dot denotes derivative with respect to the conformal time $t$, whereas the prime denotes a spatial derivative. Going to the momentum space by defining \begin{align} \phi(x,t) = \frac{1}{\sqrt{2\pi}}\int dk \phi_k(t) e^{ikx}, \end{align} the wave equation reduces to \begin{align} \ddot{\phi}_k - k^2\phi_k + \Omega^2(t) m^2\phi_k \equiv \ddot{\phi}_k + \omega_k^2(t)\phi_k = 0. \end{align} The general solution of this time-dependent oscillator equation can be written in terms of complex mode functions $v_k(t)$ with the normalisation condition $-iW[v_k,v_k^*]/2 \equiv -i(\dot{v}_k v_k^* - v_k\dot{v}_k^*)./2= 1$, where $W$ is called the Wronskian. The field operator $\phi_k$ can be expanded as \begin{align} \phi_k(t) = \frac{1}{2}\left[ a_k v_k^*(t) + a^\dagger_k v_k(t) \right], \end{align} where $a_k$ and $a^\dagger_k$ are the bosonic annihilation and creation operators for mode $k$. It is interesting to note that the (bosonic) commutation relation is crucial for preserving the normalisation defined by the Wronskian. The anticommutation relation (for fermions) is not consistent with the latter. Once the mode expansion is defined, the vacuum state $|0\rangle$ is defined by the condition $a_k|0\rangle = 0$ for all $k$. Now consider two different mode functions $u_k(t)$ and $v_k(t)$. Since $u_k$ and $u_k^*$ form a basis, we can expand $v_k$ in terms of them, $v_k^*(t) = \alpha_k u_k^*(t) + \beta_k u_k(t)$, with $t$-independent coefficients $\alpha_k$ and $\beta_k$ that obey the normalisation condition $|\alpha_k|^2 - |\beta_k|^2 = 1$. If $b_k$ and $b_k^\dagger$ are the annihilation and creation operators corresponding to the mode functions $u_k$, the following relations hold \begin{align} b_k = \alpha_ka_k + \beta_k^* a_{-k}^\dagger, \;\;\;\;\; b_k^\dagger = \alpha_k^*a_k^\dagger + \beta_k a_{-k}. \end{align} This is called the Bogolyubov transformation in the literature. What is interesting is that the vacuum state for $a$ is not the vacuum state for $b$. Indeed a simple calculation reveals that the expectation value of the number of $b$-particles in $a$'s vacuum state is generically non-zero: \begin{align} \langle 0_a | b_k^\dagger b_k |0_a\rangle &= \langle 0_a | (\alpha_k^*a_k^\dagger + \beta_k a_{-k}) (\alpha_k a_k + \beta_k^*a_{-k}^\dagger) |0_a\rangle \nonumber \\ &= |\beta_k|^2 \langle 0_a | a_{-k}a_{-k}^\dagger |0_a\rangle = |\beta_k|^2. \end{align} To repeat, if $\beta_k$ is non-zero, the vacuum state defined with respect to the $a_k$ modes contains a finite density of $b_k$ particles. It is generally impossible to find a unique vacuum state in curved spacetime unlike in flat (Minkowski) spacetime. To discuss particle creation, it is therefore a good idea to work with an asymptotically flat spacetime that has the Minkowski metric both in far-enough past and far-enough future. As a simple example, we here consider a specially chosen FRW spacetime with $\Omega^2(t) = 1$ if $t<0$ or $t>t_0$ and $\Omega^2(t)=-1$ if $0<t<t_0$. The `in' and `out' vacuum modes at $t<0$ and $t>t_0$ are described by the standard Minkowski mode functions \begin{align} v_k^{(in)}(t) = \frac{1}{\sqrt{\omega_k}}e^{i\omega_kt},\;\;\; v_k^{(out)}(t) = \frac{1}{\sqrt{\omega_k}}e^{i\omega_k(t-t_0)}, \end{align} with $\omega_k = \sqrt{k^2 + m^2}$. To find the Bogolyubov coefficients, we need to find the relationship between the input and output modes. Solving the mode equations in all regions, one obtains the following relation for $t>t_0$ \begin{align} v_k^{(in)}(t) = \frac{1}{\sqrt{\omega_k}}\left[ \alpha_k^* e^{i\omega_k(t-t_0)}+\beta_k^* e^{-i\omega_k(t-t_0)}\right], \end{align} where \begin{align} \alpha_k &= \tfrac{e^{-i\Omega_k t_0}}{4}\left( \sqrt{\tfrac{\omega_k}{\Omega_k}}+\sqrt{\tfrac{\Omega_k}{\omega_k}}\right)^2 -\tfrac{e^{i\Omega_k t_0}}{4}\left( \sqrt{\tfrac{\omega_k}{\Omega_k}}-\sqrt{\tfrac{\Omega_k}{\omega_k}}\right)^2, \nonumber \\ \beta_k &= \frac{1}{2}\left(\tfrac{\Omega_k}{\omega_k}-\tfrac{\omega_k}{\Omega_k}\right)\sin(\Omega_kt_0), \end{align} with $\Omega = \sqrt{k^2-m^2}$. This tells us that the field mode that has evolved from $|0_{in}\rangle$ is different from the vacuum state at $t>t_0$ and has a finite particle number density $n_k = |\beta_k|^2 = \tfrac{m^4}{|k^4-m^4|}|\sin\left( t_0(k^2-m^2)\right)|^2$ \subsubsection{Dirac field in curved spacetime, a single particle analog of particle creation} The above procedure applies in exactly the same way for Dirac fields, except that one has to use the anticommutation relation instead of the commutation relation when quantizing the field operators. One other difference is that particles are produced in pairs so that the total charge is conserved. Because the mode expansion is a little more involved, and we are actually interested in the single-particle manifestation of pair creation, we will not go into the details here. Interested readers are referred to the original article by Parker \cite{Parker1971}. Instead let us discuss the single-particle manifestation of pair creation. In order to see how this is possible, consider the pair creation process due to an electric field (Schwinger effect) in the Dirac sea picture. Initially, the negative-energy sea is fully occupied. Then upon applying a strong electric field, a negative energy electron is kicked (or tunnels) to occupy a positive energy state, leaving a hole (positron) behind. Therefore, in terms of single-particle dynamics, what we will see is a conversion of a negative energy wave packet to a positive energy wave packet. In the next section we verify this claim, by studying the dynamics of a wave packet in curved spacetime. \section{Dynamics of spinor wave packet} \label{dynamics} In this section, we provide a basic background on the wave packet dynamics, both in flat and curved spacetime. In particular, we show a single-particle analog of particle creation in an FRW spacetime. The results from this section will be used in the next section, where we discuss the implementation of wave packet dynamics in binary waveguide arrays. \subsection{Flat spacetime} The Dirac equation in flat 2 dimensional space time was written in Eq.~(\ref{flateqn}), which we repeat here for the reader's convenience \begin{align} i\partial_t \psi= -i\sigma_x\partial_x\psi+\sigma_z m \psi. \end{align} Using the plane wave ansatz, the eigen-solutions can be easily found. The two solutions with positive and negative energies can be written as \begin{align} \label{eestates} \psi_+^k(x,t) &= \frac{1}{\sqrt{2 \pi}}\frac{\exp [ -iE_kt + ikx]}{\sqrt{2E_k(E_k + m)}}\begin{pmatrix} E_k+m \\ k \end{pmatrix}, \nonumber \\ \psi_-^k(x,t) &= \frac{1}{\sqrt{2 \pi}}\frac{\exp [ iE_kt + ikx]}{\sqrt{2E_k(E_k + m)}}\begin{pmatrix} -k \\ E_k + m \end{pmatrix}. \end{align} The time evolution of an initial spinor wave packet, $(\phi_1(k),\phi_2(k))$, is given by \begin{align} \psi(x,t) = \int dk \tfrac{\alpha (E_k+m)\phi_1(k) + \beta k\phi_2(k) }{2 \sqrt{2\pi} E_k(E_k + m)}\begin{pmatrix} E_k +m \\ k \end{pmatrix} e^{i(kx-E_kt)} \nonumber \\ +\int dk \tfrac{-\alpha k \phi_1(k) + \beta (E_k+m)\phi_2(k) }{2 \sqrt{2\pi} E_k(E_k + m)}\begin{pmatrix} -k \\ E_k +m \end{pmatrix} e^{i(kx+E_kt)}. \end{align} Let us first look at the dynamics of a Gaussian wave packet $\psi_0(x) \propto \exp[-x^2/18](1,1)^T$. Figure \ref{flat1}(a) shows the time evolution of the wave packet for $m=1$. Notice the zig-zag motion of the centre of mass as exemplified in Fig.~\ref{flat1}(b). This phenomenon is called zitterbewegung and occurs because of the interference between the positive- and negative-energy components of the spinor. Properties of zitterbewegung can be seen from the position operator in the Heisenberg picture. Noting that $v(t) = -i[x,H] = \sigma_x(t)$ and \begin{align} \frac{\partial\sigma_x}{\partial t} = -i\left[ \sigma_x,H\right] = -2i\left[p-H\sigma_x \right], \end{align} we obtain \begin{align} \sigma_x(t) = H^{-1}p + e^{2iHt}\left[ \sigma_x(0) - H^{-1}p\right]. \end{align} $x(t) = \int v(t)$ is then given by \begin{align} x(t) = x(0) + \frac{p}{H}t - \frac{i}{2H}\left( e^{2iHt} - 1\right) \left[ \sigma_x(0) - \frac{p}{H}\right]. \end{align} From the last term of this expression we notice several things. 1) the frequency of the oscillation is $2E$, 2) the amplitude is proportional to 1/(2E), and 3) zitterbewegung is non-zero only if there is a superposition between positive and negative energy states with equal momentum. The last observation follows because $\sigma_x - p/H$ anti-commutes with the Hamiltonian, which means that the matrix element of it is non-zero only between the eigenstates with equal momentum and opposite energies. Note that for a small initial momentum, the amplitude and frequency of ZB is thus 2m and 1/(2m) respectively. This is demonstrated in Fig.~\ref{ZB}. \begin{figure}[ht] \begin{center} \includegraphics[width=0.4\columnwidth]{flat1.pdf} \includegraphics[width=0.4\columnwidth]{flat1pos.pdf} \caption{Wave packet evolution of an initial Gaussian state, showing zitterbewegung, the trembling oscillation of the centre of mass.} \label{flat1} \end{center} \end{figure} \begin{figure}[ht] \begin{center} \includegraphics[width=0.4\columnwidth]{ZB1.pdf} \includegraphics[width=0.4\columnwidth]{ZB2.pdf} \includegraphics[width=0.4\columnwidth]{ZB3.pdf} \includegraphics[width=0.4\columnwidth]{ZB4.pdf} \caption{The mass dependence of zitterbewegung for Gaussian initial states $\propto \exp[-x^2/2\sigma^2](1,1)^T$ with the width $\sigma$ = 3 and mass $m$= 1 (a), 2 (b), 4 (c), and 8 (d). } \label{ZB} \end{center} \end{figure} Absence of zitterbewegung in a wave packet composed of positive-energy spinors only is shown in Fig.~\ref{flat2}. The positive energy spinor is constructed by superposing the positive-energy spinor in momentum space with a Gaussian weight $\exp[-k^2/2]$. \begin{figure}[ht] \begin{center} \includegraphics[width=0.4\columnwidth]{flat2.pdf} \includegraphics[width=0.4\columnwidth]{flat2pos.pdf} \caption{Wave packet evolution of a positive-energy state. No zitterbewegung is observed.} \label{flat2} \end{center} \end{figure} \subsection{Curved spacetime} \label{sect:curvedDirac} We have seen that in curved spacetime, the Dirac equation is given by Eq.~(\ref{curvedDirac}): \begin{equation*} i\partial_t(\Omega\xi)= -i\sigma_x\partial_x(\Omega\xi)+\sigma_z \Omega m (\Omega\xi), \end{equation*} which we write as \begin{equation} i\partial_t\psi= -i\sigma_x\partial_x\psi+\sigma_z m_{eff}(t)\psi, \end{equation} Here we concentrate on the FRW spacetime with a time-dependent conformal factor $a(t) \equiv \Omega(t)$. Neglecting the overall factor $\Omega(t)$, this equation is simply the Dirac equation with a time-dependent mass term. Analogously to a real scalar field in an FRW spacetime, Dirac fermions are produced (in pairs) in generic cases, which we will show by demonstrating conversion of a negative energy wavepacket to a positive energy wave packet. Quantitatively, the conversion can be proven by calculating the norm of the positive-energy-projected spinor wave packet, but we can do better: We can make use of the fact that ZB only occurs when positive-energy spinors are superposed with negative-energy spinors. So our aim is to show how ZB is induced in an asymptotically flat spacetime with an initial negative-energy wave packet. It turns out that physically interesting scenarios such as expanding spacetime (anti-de Sitter spacetime) produce only tiny effects that are unobservable, so instead we employ the `inverted square hat' profile of $m_{eff}(t)$ already introduced earlier in Sect.~\ref{sect3b}. Actually, we will use the smoothed version of this, by replacing the `inverted square hat' by an inverted Gaussian profile. Evidence of particle creation in this scenario is evident in Fig.~\ref{curved1}(a) where the induction of ZB by the excursion of $m_{eff}$ from it asymptotic value 1, is clearly visible. The initial state was constructed by superposing $\psi_-^k(x,0)$ from Eq.~(\ref{eestates}) with a Gaussian weight $\propto \exp[-4(k-0.1)^2]$. \begin{figure}[ht] \begin{center} \includegraphics[width=0.45\columnwidth]{curvedpos.pdf} \includegraphics[width=0.45\columnwidth]{mprofile.pdf} \caption{Left, mean position of an initial positive-energy state in curved spacetime. Right, time profile of the conformal factor.} \label{curved1} \end{center} \end{figure} \section{Simulation of the Dirac equation in binary waveguide arrays} In this section, we show how to simulate the 2 dimensional Dirac equation in binary waveguide arrays. We start off with a description of the coupled-mode theory of light propagation in a waveguide array and then demonstrate an equivalence between the discretised Dirac equation and the coupled-mode equations for a binary waveguide array. In particular, we show that the Dirac equation in 2 dimensional FRW spacetime can be straightforwardly implemented, paving the way towards experimental demonstration of the single particle analog of gravitational pair creation. \subsection{Coupled-mode theory of waveguide arrays} Propagation of an electromagnetic field $E$ in a medium is described by \begin{equation} \nabla^2 E - \frac{n^2}{c^2} \frac{\partial^2 E}{\partial t^2} = 0, \end{equation} where $n$ is the refractive index of the medium, spatially varying in general. For a monochromatic field $E$ propagating predominantly in the $z-$direction, i.e., \begin{equation} E= Re[E_0({\bf r})e^{i(kz-wt)}], \end{equation} one can make the so-called paraxial approximation which amounts to assuming $|\nabla^2E_0| \ll |k_z \partial_z E_0|$. In this case, the wave equation becomes the paraxial Helmholtz equation \begin{equation} k \frac{\partial E_0}{\partial z} + \frac{i}{2}\left( k^2 - \omega^2 \frac{n^2(r)}{c^2} \right) E_0 - \frac{i}{2} \nabla^2_{\perp} E_0 = 0, \end{equation} $\nabla^2_\perp$ denoting the Laplacian in $x$ and $y$ directions. Let us choose $\omega/k = v_0 \equiv c/n_0$, where $v_0$ is the mean velocity in the medium and $n_0$ the mean refractive index. Then the above equation becomes \begin{align} 2ik\partial_z E_0 + \nabla_\perp^2E_0-\frac{2k^2}{n_0}\Delta n E_0 = 0, \end{align} where we have assumed that $n_0 \approx n$ and $\Delta n\equiv n_0 - n$. Lastly, define the reduced wavelength $\lambdabar = n_0/k$ to rewrite the equation as \begin{align} i\lambdabar \partial_z E_0 = -\frac{\lambdabar^2}{2n_0}\nabla^2_\perp E_0 + \Delta n E_0. \end{align} Written this way, resemblance to the Schr\"odinger equation is obvious, the role of time being played by the spatial dimension $z$. Now imagine periodically modulating the index of refraction in the plane perpendicular to the z-direction. An EM field is attracted to regions of increased index of refraction and mainly stays in the vicinity of these regions during propagation. The field, however, is not completely confined to these regions but leaks into the area between the waveguides, with evanescent tails. From the similarity with the Schr\"odinger equation, it is clear that one can apply the tight-binding approximation to describe the propagation of the EM field in this case. In optics, this is called the coupled mode approximation (see \cite{Szameit10}, which we are closely following). In a one dimensional lattice we get for the amplitude $c_n$ in the $n$th waveguide: \begin{equation} i\frac{\partial c_n}{\partial z}= k_n \left( c_{n+1} + c_{n-1} \right), \end{equation} where $k_n$ is the coupling strength, determined by the overlap between the transverse components of the modes in adjacent guides. Introducing yet another modulation such that alternating lattice sites have deep and shallow `potentials', one obtains the following coupled mode equations \begin{equation} \label{cmeqn} i\frac{\partial c_n}{\partial z}= k_n \left( c_{n+1} + c_{n-1} \right) + (-1)^n \sigma c_n. \end{equation} In the next section we show that this equation is equivalent to the discretised Dirac equation. \subsection{Discretising the Dirac equation} \subsubsection{Flat spacetime} To simulate the Dirac equation in flat 2D spacetime, \begin{equation} i\frac{\partial \psi}{\partial t}=\left[-i\sigma_x{\partial_x}+m\sigma_z\right]\psi, \end{equation} in a waveguide array, we need to first change the differential equation to a difference equation by discretising the spatial coordinates. Choosing a discretisation length $d$, we define \begin{align} \psi(x,t) &\rightarrow \psi(nd,t) \equiv \tilde{\psi}(n,t), \nonumber \\ {\partial_x}\psi_1(x,t) &\rightarrow \frac{ \tilde{\psi}_1(n,t)-\tilde{\psi}_1(n-1,t)}{d}, \nonumber \\ {\partial_x}\psi_2(x,t) &\rightarrow \frac{ \tilde{\psi}_2(n+1,t)-\tilde{\psi}_2(n,t)}{d}, \end{align} where $n \in \mathbb{Z}$. To put this in the form of coupled mode equations, let us first define $\tilde{c}_{2n}(t) \equiv \tilde{\psi}_1(n,t)$ and $\tilde{c}_{2n-1} \equiv \tilde{\psi}_2(n,t)$. Then the Dirac equation becomes \begin{align} i\dot{\tilde{c}}_{2n}(t) = -\frac{i}{d}\left( \tilde{c}_{2n+1}(t) - \tilde{c}_{2n-1}(t) \right) + m\tilde{c}_{2n}(t), \nonumber \\ i\dot{\tilde{c}}_{2n-1}(t) = -\frac{i}{d}\left( \tilde{c}_{2n}(t) - \tilde{c}_{2n-2}(t) \right) - m\tilde{c}_{2n-1}(t). \end{align} To change the difference into the sum, we further define $c_{2n} = (-1)^n\tilde{c}_{2n} = (-1)^n\tilde{\psi}_1(n,t)$ and $c_{2n-1} = -i(-1)^n\tilde{c}_{2n-1} = -i(-1)^n\tilde{\psi}_2(n,t)$. The above equation changes to \begin{align} i\dot{c}_{2n}(t) = \frac{1}{d}\left( c_{2n+1}(t) + c_{2n-1}(t) \right) + mc_{2n}(t), \nonumber \\ i\dot{c}_{2n-1}(t) = \frac{1}{d}\left( c_{2n}(t) + c_{2n-2}(t) \right) - mc_{2n-1}(t), \end{align} which can be combined to \begin{equation} i\dot{c_l}=-\frac{1}{d}(c_{l-1}+c_{l+1})+(-1)^{l} m c_l. \end{equation} Figure \ref{assignspinorcomp} provides a schematic illustration of how the spinor components are assigned to waveguides. \begin{figure}[ht] \begin{center} \includegraphics[width=0.8\columnwidth]{AssignSpinorComp.png} \caption{Schematic illustration of how spinor components are assigned to waveguides. Different colours signify alternating refractive indices.} \label{assignspinorcomp} \end{center} \end{figure} The above equation is equivalent to Eq.~(\ref{cmeqn}) given that $k_n = 1/d$ and $\sigma = m$, as first noted by Longhi in his proposal to simulate zitterbewegung and Klein's paradox \cite{Longhi10a, Longhi10b}. Subsequently, these effects, which are predicted by the single-particle Dirac equation, have been observed in laser-written waveguide arrays \cite{Dreisow10,Dreisow12}. \subsubsection{Curved spacetime} To facilitate the simulation of the Dirac equation in 2 dimensional curved spacetime, we need to discretise Eq.~(\ref{curvedDirac}). In an optical simulation of single particle physics, the overall factor $\Omega$ is irrelevant, which means that we have at hand the Dirac equation with a spacetime-dependent mass as already noted in Sect.~\ref{sect:curvedDirac}. The generalisation of the coupled-mode equations Eq.~(\ref{cmeqn}) is then simple: \begin{align} i\frac{\partial c_n}{\partial z}= k_n \left( c_{n+1} + c_{n-1} \right) + (-1)^n \sigma_n (z) c_n. \end{align} \subsection{Optical simulation} Let us start with flat spacetime. Figure \ref{simflat1} shows the evolution of a Gaussian spinor wave packet $\propto \exp[-x^2/18](1,1)^T$ when $m=1$. Simulation results for $N=502$ (requiring $k \approx 6.2$) waveguides are displayed in Fig.~\ref{simflat1}(a) and (b). They are in excellent agreement with the results in Sect.~\ref{dynamics}. However 502 is quite a large number and experiments are usually implemented with a much smaller number. To show the effects of discretisation we depict analogous results for $N=50$ ($k \approx 0.63$) in Fig.~\ref{simflat1}(c) and (d). Apart from coarse graining effects in visualisation, the wave packet evolution is seen to be remarkably accurate as exemplified by the average position. \begin{figure}[!htbp] \begin{center} \includegraphics[width=0.4\columnwidth]{simflat1a.pdf} \includegraphics[width=0.4\columnwidth]{simflat1b.pdf} \includegraphics[width=0.4\columnwidth]{simflat1c.pdf} \includegraphics[width=0.4\columnwidth]{simflat1d.pdf} \caption{Optical simulation of the Dirac equation in flat 2 dimensional spacetime. a) Absolute value of the mode amplitude, and b) average position of the simulated wave function as a function of time, for 502 waveguides. c) and d), analogous quantities for 50 coupled waveguides showing feasibility of the simulation despite strong discretisation. The initial state is $\propto \exp[-x^2/2\sigma^2-ik_0x](1,1)^T$ with the width $\sigma$ = 3, $k_0 = 0.1$, and mass $m = 1$. } \label{simflat1} \end{center} \end{figure} Next, we simulate the conversion of a negative energy wave packet into a mixture of positive and negative energy wave packets. The latter is made by an arbitrary superposition of a negative energy eigen-spinor \begin{align} \phi_-^k = \frac{1}{\sqrt{2E_k(E_k + m)}}\begin{pmatrix} -k \\ E_k + m \end{pmatrix}. \end{align} In Fig.~\ref{simflat2}, we consider the evolution of a Gaussian-averaged spinor, $\propto \int dk \exp[-(k-k_0)^2/2\sigma_k^2] \phi_-^k$ with the width $\sigma_k = 1/(2\sqrt{2})$, $k_0 = 0.1$, and mass $m = 1$. There is a small amount of ZB as a result of discretisation (that goes away with the increasing number of waveguides), but the magnitude is quite small. \begin{figure}[!htbp] \begin{center} \vspace{0.5cm} \includegraphics[width=0.4\columnwidth]{simflatnega.pdf} \includegraphics[width=0.4\columnwidth]{simflatnegb.pdf} \caption{Evolution of a negative energy spinor in flat spacetime. a) Absolute value of the mode amplitude, and b) average position of the simulated wave function as a function of time, for 50 waveguides. The initial state is $\propto \int dk \exp[-(k-k_0)^2/2\sigma_k^2] \phi_-^k$ with the width $\sigma_k = 1/(2\sqrt{2})$, $k_0 = 0.1$, and mass $m = 1$. } \label{simflat2} \end{center} \end{figure} Finally, we show the evolution of a negative energy spinor in a FRW spacetime in Fig.~\ref{simcurved}, with the inverted Gaussian conformal factor as used in Fig.~\ref{curved1}. We see a good agreement with the exact numerical result shown in Fig.~\ref{curved1}, signifying the feasibility of optical simulation of particle creation. \begin{figure}[!htbp] \begin{center} \vspace{0.5cm} \includegraphics[width=0.4\columnwidth]{simcurveda.pdf} \includegraphics[width=0.4\columnwidth]{simcurvedb.pdf} \caption{Evolution of a negative energy spinor in FRW spacetime with inverted Gaussian conformal factor (cf Fig.~\ref{curved1}). a) Absolute value of the mode amplitude, and b) average position of the simulated wave function as a function of time, for 50 waveguides. The initial state is $\propto \int dk \exp[-(k-k_0)^2/2\sigma_k^2] \phi_-^k$ with the width $\sigma_k = 1/(2\sqrt{2})$, $k_0 = 0.1$, and mass $m = 1$. } \label{simcurved} \end{center} \end{figure} \section{Conclusion} We gave a pedagogical introduction to the Dirac equation in background curved spacetime and particle creation. Using the fact that the Dirac equation in the FRW metric is equivalent to the flat-spacetime Dirac equation with a time-dependent mass term, we demonstrated that a single-particle analog of particle creation can be observed in the dynamical evolution of spinor wave packets. In particular, we showed how a negative energy spinor gets mixed with a positive energy spinor when the conformal factor changes in time. Finally, we demonstrated that the Dirac equation in curved spacetime can be simulated in binary waveguide arrays, allowing direct experimental simulation of particle creation in curved spacetime. Although our example was for a time-dependent conformal factor, a general spacetime dependence can be easily simulated in waveguide arrays. {\it Acknowledgments.} D.G.A would like to acknowledge the financial support provided by the National Research Foundation and Ministry of Education Singapore (partly through the Tier 3 Grant ``Random numbers from quantum processes'' (MOE2012-T3-1-009)), and travel support by the EU IP-SIQS. \bibliographystyle{unsrt}
3a890d61b35aff2fdcf6bc715cfb6957a8a2cb8c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\part{Background} \bigskip \addcontentsline{toc}{section}{1. Preliminaries} \textbf{\S 1. Preliminaries} \stepcounter{section} \bigskip The prerequisites for reading this paper are a background of one year of graduate level study in set theory, a working knowledge of forcing, and some basic familiarity with proper forcing and generalized stationarity. For a regular uncountable cardinal $\lambda$ and a set $X$ with $\lambda \subseteq X$, we let $P_\lambda(X)$ denote the set $\{ a \subseteq X : |a| < \lambda \}$. A set $S \subseteq P_\lambda(X)$ is stationary iff for any function $F : X^{<\omega} \to X$, there exists $b \in S$ such that $b \cap \lambda \in \lambda$ and $b$ is closed under $F$. \bigskip In this paper, a \emph{forcing poset} is a pair $(\mathbb{P},\le_\mathbb{P})$, where $\mathbb{P}$ is a nonempty set and $\le_\mathbb{P}$ is a reflexive and transitive relation on $\mathbb{P}$. To simplify notation, we usually refer to $\mathbb{P}$ itself as a forcing poset, with the relation $\le_\mathbb{P}$ being implicit. If $\mathbb{Q}$ is a forcing poset, we will write $\dot G_\mathbb{Q}$ for the canonical $\mathbb{Q}$-name for a generic filter on $\mathbb{Q}$. Let $\mathbb{P}$ and $\mathbb{Q}$ be forcing posets. Then $\mathbb{P}$ is a \emph{suborder} of $\mathbb{Q}$ if $\mathbb{P} \subseteq \mathbb{Q}$ and $\le_\mathbb{P} \ = \ \le_\mathbb{Q} \cap \ (\mathbb{P} \times \mathbb{P})$. Let $\mathbb{P}$ be a suborder of $\mathbb{Q}$. We say that $\mathbb{P}$ is a \emph{regular suborder} of $\mathbb{Q}$ if: \begin{enumerate} \item whenever $p$ and $q$ are in $\mathbb{P}$ and are incompatible in $\mathbb{P}$, then $p$ and $q$ are incompatible in $\mathbb{Q}$; \item if $A$ is a maximal antichain of $\mathbb{P}$, then $A$ is predense in $\mathbb{Q}$. \end{enumerate} \begin{lemma} Suppose that $\mathbb{P}_0$ is a suborder of $\mathbb{P}_1$, and $\mathbb{P}_1$ is a suborder of $\mathbb{Q}$. Assume, moreover, that $\mathbb{P}_0$ and $\mathbb{P}_1$ are both regular suborders of $\mathbb{Q}$. Then $\mathbb{P}_0$ is a regular suborder of $\mathbb{P}_1$. \end{lemma} \begin{proof} Straightforward. \end{proof} Let $\mathbb{P}$ be a regular suborder of $\mathbb{Q}$, and assume that $G$ is a generic filter on $\mathbb{P}$. In $V[G]$, define the forcing poset $\mathbb{Q} / G$ to consist of conditions $q \in \mathbb{Q}$ such that for all $s \in G$, $q$ and $s$ are compatible in $\mathbb{Q}$, with the same ordering as $\mathbb{Q}$. Then $\mathbb{Q}$ is forcing equivalent to the two-step iteration $\mathbb{P} * (\mathbb{Q} / \dot G_\mathbb{P})$. Moreover: \begin{lemma} Let $\mathbb{P}$ be a regular suborder of $\mathbb{Q}$. \begin{enumerate} \item Suppose that $H$ is a $V$-generic filter on $\mathbb{Q}$. Then $H \cap \mathbb{P}$ is a $V$-generic filter on $\mathbb{P}$, and $H$ is a $V[H \cap \mathbb{P}]$-generic filter on $\mathbb{Q} / (H \cap \mathbb{P})$. \item Suppose that $G$ is a $V$-generic filter on $\mathbb{P}$ and $H$ is a $V[G]$-generic filter on $\mathbb{Q} / G$. Then $H$ is a $V$-generic filter on $\mathbb{Q}$, $G = H \cap \mathbb{P}$, and $V[G][H] = V[H]$. \end{enumerate} \end{lemma} \begin{proof} See \cite[Lemma 1.6]{jk26}. \end{proof} Let $\mathbb{P}$ and $\mathbb{Q}$ be forcing posets with maximum conditions. A function $\pi : \mathbb{Q} \to \mathbb{P}$ is said to be a \emph{projection mapping} if: \begin{enumerate} \item $\pi$ maps the maximum condition in $\mathbb{Q}$ to the maximum condition in $\mathbb{P}$; \item if $q \le p$ in $\mathbb{Q}$, then $\pi(q) \le \pi(p)$ in $\mathbb{P}$; \item if $v \le \pi(q)$ in $\mathbb{P}$, then there is $r \le q$ in $\mathbb{Q}$ such that $\pi(r) \le v$ in $\mathbb{P}$. \end{enumerate} If $\pi : \mathbb{Q} \to \mathbb{P}$ is a projection mapping, and $G$ is a generic filter on $\mathbb{Q}$, then the set $\{ s \in \mathbb{P} : \exists p \in G \ \pi(p) \le s \}$ is a generic filter on $\mathbb{P}$. \bigskip Let $\mathbb{Q}$ be a forcing poset. For a set $N$ and a condition $q \in \mathbb{Q}$, we say that $q$ is \emph{strongly $N$-generic} if whenever $D$ is a dense subset of $N \cap \mathbb{Q}$, then $D$ is predense below $q$ in $\mathbb{Q}$. We say that $\mathbb{Q}$ is \emph{strongly proper on a stationary set} if for all sufficiently large cardinals $\lambda$ with $\mathbb{Q} \subseteq H(\lambda)$, there are stationarily many $N$ in $P_{\omega_1}(H(\lambda))$ such that for all $p \in N \cap \mathbb{Q}$, there is $q \le p$ which is strongly $N$-generic. If $\mathbb{Q}$ is strongly proper on a stationary set, then $\mathbb{Q}$ preserves $\omega_1$, because being a strongly $N$-generic condition implies being an $N$-generic condition in the sense of proper forcing. Let $\lambda_\mathbb{Q}$ denote the smallest cardinal such that $\mathbb{Q} \subseteq H(\lambda_\mathbb{Q})$. Note that a condition $q$ is strongly $N$-generic iff $q$ is strongly $(N \cap H(\lambda_\mathbb{Q}))$-generic. Using this fact, standard arguments show that $\mathbb{Q}$ is strongly proper on a stationary set iff there are stationarily many $N$ in $P_{\omega_1}(H(\lambda_\mathbb{Q}))$ such that for all $p \in N \cap \mathbb{Q}$, there is $q \le p$ which is strongly $N$-generic. \bigskip Let $V \subseteq W$ be transitive class models of \textsf{ZFC}. A set $X \subseteq V$ is said to be \emph{countably approximated by $V$} if for any set $a \in V$ which is countable in $V$, $a \cap X \in V$. We say that the pair $(V,W)$ has the \emph{$\omega_1$-approximation property} if whenever $X$ is a subset of $V$ in $W$ which is countably approximated by $V$, then $X \in V$. A forcing poset $\mathbb{Q}$ is said to have the \emph{$\omega_1$-approximation property} if $\mathbb{Q}$ forces that the pair $(V,V^\mathbb{Q})$ has the $\omega_1$-approximation property. Note that the $\omega_1$-approximation property is equivalent to the definition in the previous paragraph, except replacing the assumption that $X$ is a subset of $V$ with the assumption that $X$ is a set of ordinals. Namely, if $X \subseteq V$, then for some $\alpha$, $X \subseteq V_\alpha$. And in $V$ we can fix a bijection $g : V_\alpha \to \mu$ for some ordinal $\mu$. Then $X$ is countably approximated by $V$ iff $g[X]$ is countably approximated by $V$, and $X \in V$ iff $g[X] \in V$. The next lemma shows that the $\omega_1$-approximation property defined above is equivalent to the version of the property used in \cite{mitchell}. \begin{lemma} A pair $(V,W)$ has the $\omega_1$-approximation property iff whenever $\mu$ is an ordinal, $k : \mu \to On$ is in $W$, and for any set $a$ in $V$ which is countable in $V$, $k \restriction a \in V$, then $k \in V$. \end{lemma} \begin{proof} Assume that the pair $(V,W)$ has the $\omega_1$-approximation property, and let $k : \mu \to On$ be a function in $W$ satisfying that for any countable set $a$ in $V$, $k \restriction a \in V$. We will show that $k \in V$. It suffices to show that whenever $x \in V$ is countable in $V$, then $k \cap x \in V$. Suppose that $x$ is a countable set in $V$, and we will show that $k \cap x \in V$. Define $x_0 := \{ \xi \in \mu : \exists z \ (\xi,z) \in x \}$. Then $x_0$ is a countable subset of $\mu$ in $V$, so $k \restriction x_0 \in V$. It is easy to check that $k \cap x = (k \restriction x_0) \cap x$, and hence $k \cap x \in V$. Conversely, suppose that whenever $\mu$ is an ordinal, $k : \mu \to On$ is in $W$, and for any set $a$ in $V$ which is countable in $V$, $k \restriction a \in V$, then $k \in V$. We will prove that $(V,W)$ has the $\omega_1$-approximation property. Let $X$ be a subset of $V$ in $W$ which is countably approximated by $V$, and we will show that $X \in V$. By the comments preceding the lemma, we may assume that $X \subseteq \mu$ for some ordinal $\mu$. Let $k : \mu \to 2$ be the characteristic function of $X$, so that $k(\alpha) = 1$ iff $\alpha \in X$. Then $k \in V$ iff $X \in V$, so it suffices to show that $k \in V$. To show that $k \in V$, it suffices to show that whenever $a$ is a countable set in $V$, then $k \restriction a \in V$. So let $a$ be a countable set in $V$. Then $X \cap a \in V$ by assumption. But $k \restriction a$ is equal to the function with domain $a \cap \mu$ such that for all $\alpha \in a \cap \mu$, $\alpha$ is mapped to $1$ iff $\alpha \in X \cap a$. Since $a \cap \mu$ and $X \cap a$ are in $V$, so is $k \restriction a$. \end{proof} \begin{lemma} Let $\mathbb{P}$ be a regular suborder of $\mathbb{Q}$, and suppose that $\mathbb{Q}$ forces that the pair $(V[\dot G_{\mathbb{Q}} \cap \mathbb{P}],V[\dot G_{\mathbb{Q}}])$ has the $\omega_1$-approximation property. Then $\mathbb{P}$ forces that $\mathbb{Q} / \dot G_{\mathbb{P}}$ has the $\omega_1$-approximation property. \end{lemma} \begin{proof} Let $G$ be a $V$-generic filter on $\mathbb{P}$. Then by Lemma 1.2(2), whenever $H$ is a $V[G]$-generic filter on $\mathbb{Q} / G$, then $H$ is a $V$-generic filter on $\mathbb{Q}$, $H \cap \mathbb{P} = G$, and $V[H] = V[G][H]$. By assumption, the pair $(V[H \cap \mathbb{P}],V[H])$ has the $\omega_1$-approximation property. But $(V[H \cap \mathbb{P}],V[H]) = (V[G],V[G][H])$. Thus, for any $V[G]$-generic filter $H$ on $\mathbb{Q} / G$, the pair $(V[G],V[G][H])$ has the $\omega_1$-approximation property. This means that $\mathbb{Q} / G$ has the $\omega_1$-approximation property in $V[G]$. \end{proof} \bigskip \addcontentsline{toc}{section}{2. Side conditions} \textbf{\S 2. Side conditions} \stepcounter{section} \bigskip In this section, we lay out the basic framework of side conditions which will serve as the foundation for almost everything in the paper. Our goal is to make this material as self-contained as possible. However, we do not want to prove all of the results from scratch, since that has already been done in other papers, and some of the proofs are tedious. So several of the assumptions and results in this section will be stated without proof; we will provide specific references here and in Section 13 so that an interested reader can easily find the complete details. \bigskip The basic objects we introduce are the cardinals $\kappa$ and $\lambda$, the set $\Lambda$, the structure $\mathcal A$ on $H(\lambda)$, the classes of countable models $\mathcal X$ and uncountable models $\mathcal Y$, and the comparison point $\beta_{M,N}$, for all $M$ and $N$ in $\mathcal X$. \begin{notation} For the remainder of the paper, $\kappa$ is a regular cardinal satisfying that $\omega_2 \le \kappa$, and $\lambda$ is a regular cardinal such that $\kappa \le \lambda$. \end{notation} In this paper, our interest will be in the cases where either $\kappa = \omega_2$, or $\kappa$ is an inaccessible cardinal which is intended to become $\omega_2$ in some generic extension. In the context of adding a single object by forcing, it is natural to let $\lambda = \kappa$. If multiple objects are being added by forcing, then $\lambda$ will be at least $\kappa^+$. \begin{notation} Fix a set $\Lambda$ such that for some club $C^* \subseteq \kappa$, $$ \Lambda = C^* \cap \mathrm{cof}(>\! \omega). $$ \end{notation} \begin{notation} Fix a structure $\mathcal A$, whose underlying set is $H(\lambda)$, which has a well-ordering of $H(\lambda)$ as a predicate, and for which the sets $\kappa$ and $\Lambda$ are definable predicates. \end{notation} Since $\mathcal A$ has a well-ordering as a predicate, it has definable Skolem functions. For any set $a \subseteq H(\lambda)$, let $Sk(a)$ denote the Skolem hull of $a$ in $\mathcal A$ under some (any) complete set of definable Skolem functions. \begin{notation} Fix sets $\mathcal X$ and $\mathcal Y$ satisfying: \begin{enumerate} \item for all $M \in \mathcal X$, $M$ is a countable elementary substructure of $\mathcal A$; \item for all $P \in \mathcal Y$, $P$ is an elementary substructure of $\mathcal A$, $|P| < \kappa$, $P \cap \kappa \in \kappa$, and $\mathrm{cf}(P \cap \kappa) > \omega$. \end{enumerate} \end{notation} The next assumption describes some closure properties of $\mathcal X$ and $\mathcal Y$. \begin{assumption} \begin{enumerate} \item If $P$ and $Q$ are in $\mathcal Y$, then $P \cap Q \in \mathcal Y$. \item If $M \in \mathcal X$ and $P \in \mathcal Y$, then $M \cap P \in \mathcal X$. \end{enumerate} \end{assumption} Following Friedman \cite{friedman}, we say that a stationary set $T^* \subseteq P_{\omega_1}(\kappa)$ is \emph{thin} if for all $\beta < \kappa$, $$ | \{ a \cap \beta : a \in T^* \} | < \kappa. $$ Implicit in our listed assumptions is the existence of a thin stationary set $T^*$, which is used in the actual definitions of $\mathcal A$ and $\mathcal X$. For example, it will be the case that for all $M \in \mathcal X$, $M \cap \kappa \in T^*$. In particular, there are a limited number of sets of the form $M \cap \kappa$, where $M \in \mathcal X$. The next assumption follows as a consequence; see \cite[Proposition 1.11]{jk21} for the details. \begin{assumption} If $M \in \mathcal X$, $\alpha \in \Lambda \cup \{ \kappa \}$, and $Sk(\alpha) \cap \kappa = \alpha$ if $\alpha < \kappa$, then $M \cap \alpha \in Sk(\alpha)$. \end{assumption} Note that there are club many $\alpha < \kappa$ such that $Sk(\alpha) \cap \kappa = \alpha$. Also, $Sk(\kappa) \cap \kappa = \kappa$ is immediate. We have enough information now to derive some useful properties. \begin{lemma} Let $M \in \mathcal X$ and $\alpha \in \Lambda \cup \{ \kappa \}$. If $\alpha < \kappa$, assume that $Sk(\alpha) \cap \kappa = \alpha$. Then: \begin{enumerate} \item $Sk(M \cap \alpha) = M \cap Sk(\alpha)$; \item $Sk(M \cap \alpha) \cap \kappa = M \cap \alpha$. \end{enumerate} \end{lemma} \begin{proof} (1) The forward inclusion is immediate. For the reverse inclusion, suppose that $x \in M \cap Sk(\alpha)$. Then since $x \in Sk(\alpha)$, there are ordinals $\beta_0,\ldots,\beta_{n-1}$ in $\alpha$ and a definable Skolem function $f$ of $\mathcal A$ such that $x = f(\beta_0,\ldots,\beta_{n-1})$. Since $\kappa$ and $f$ are definable in $\mathcal A$, the lexicographically least tuple of ordinals $\gamma_0,\ldots,\gamma_{n-1}$ in $\kappa$ such that $x = f(\gamma_0,\ldots,\gamma_{n-1})$ is definable in $\mathcal A$ from $x$. Since $x \in M \cap Sk(\alpha)$, it follows that $\gamma_0,\ldots,\gamma_{n-1}$ are in $M \cap Sk(\alpha)$. But $Sk(\alpha) \cap \kappa = \alpha$, so $\gamma_0,\ldots,\gamma_{n-1} \in M \cap \alpha$. Therefore, $x = f(\gamma_0,\ldots,\gamma_{n-1}) \in Sk(M \cap \alpha)$. (2) Using (1) and our assumption about $\alpha$, we have $$ Sk(M \cap \alpha) \cap \kappa = (M \cap Sk(\alpha)) \cap \kappa = M \cap (Sk(\alpha) \cap \kappa) = M \cap \alpha. $$ \end{proof} \begin{lemma} Let $M$ and $N$ be in $\mathcal X$ and $\alpha \in \Lambda \cup \{ \kappa \}$. If $\alpha < \kappa$, assume that $Sk(\alpha) \cap \kappa = \alpha$. Then: \begin{enumerate} \item If $M \cap \alpha \in N$, then $M \cap \alpha \in Sk(N \cap \alpha)$; \item If $M \cap \kappa \subseteq \alpha$ and $M \cap \kappa \in N$, then $M \cap \kappa \in Sk(N \cap \alpha)$. \end{enumerate} \end{lemma} \begin{proof} (1) By Assumption 2.6, $M \cap \alpha \in Sk(\alpha)$. By Lemma 2.7(1), $M \cap \alpha \in N \cap Sk(\alpha) = Sk(N \cap \alpha)$. (2) If $M \cap \kappa \subseteq \alpha$, then $M \cap \kappa = M \cap \alpha$. So (2) follows immediately from (1). \end{proof} We now introduce the \emph{comparison point} $\beta_{M,N}$, for all $M$ and $N$ in $\mathcal X$. The actual definition of $\beta_{M,N}$ is not important for us in this paper. The only properties of $\beta_{M,N}$ which we will need are stated in Lemma 2.10 and Proposition 2.11 below. \begin{notation} For $M$ and $N$ in $\mathcal X$, $\beta_{M,N}$ will denote the \emph{comparison point} of $M$ and $N$, as defined in \cite[Definition 1.14]{jk27}. \end{notation} \begin{lemma} Let $M$ and $N$ be in $\mathcal X$. Then: \begin{enumerate} \item $\beta_{M,N} \in \Lambda$ and $\beta_{M,N} = \beta_{N,M}$; \item if $\beta < \beta_{M,N}$ and $\beta \in \Lambda$, then $M \cap [\beta,\beta_{M,N}) \ne \emptyset$; \item if $K \in \mathcal X$ and $K \subseteq M$, then $\beta_{K,N} \le \beta_{M,N}$. \end{enumerate} \end{lemma} \begin{proof} See \cite[Definition 1.14, Lemma 1.16(1,3)]{jk27}. \end{proof} For a set of ordinals $a$, let $\mathrm{cl}(a)$ denote the union of $a$ together with the set of limit points of $a$. \begin{proposition} Let $M$ and $N$ be in $\mathcal X$. Then $$ \mathrm{cl}(M \cap \kappa) \cap \mathrm{cl}(N \cap \kappa) \subseteq \beta_{M,N}. $$ \end{proposition} \begin{proof} See \cite[Lemma 1.15]{jk27}. \end{proof} Since the property described in Proposition 2.11 is extremely important for what follows, let us review it for emphasis. The property says that if $\gamma$ is an ordinal which is either in $M \cap \kappa$ or is a limit point of $M \cap \kappa$, and at the same time, is either in $N \cap \kappa$ or is a limit point of $N \cap \kappa$, then $\gamma < \beta_{M,N}$. \begin{lemma} Let $P \in \mathcal Y$ and $M \in P \cap \mathcal X$. Then for all $K \in \mathcal X$, $\beta_{K,M} < P \cap \kappa$. \end{lemma} \begin{proof} Roughly speaking, the reason why this is true is because, given $M$, there are only countably many possibilities for the value of $\beta_{K,M}$, and hence they are all in $P$ by elementarity. See \cite[Lemma 1.34(1)]{jk27} for the proof. \end{proof} We now use the comparison point $\beta_{M,N}$ to introduce a way of comparing models in $\mathcal X$. \begin{definition} Let $M$ and $N$ be in $\mathcal X$. \begin{enumerate} \item Let $M < N$ if $M \cap \beta_{M,N} \in N$. \item Let $M \sim N$ if $M \cap \beta_{M,N} = N \cap \beta_{M,N}$. \item Let $M \le N$ if either $M < N$ or $M \sim N$. \end{enumerate} \end{definition} If $M < N$, then by elementarity, the set $\mathrm{cl}(M \cap \beta_{M,N})$ is a member of $N$. Since $\mathrm{cl}(M \cap \beta_{M,N})$ is countable, it follows that $\mathrm{cl}(M \cap \beta_{M,N}) \subseteq N$. Also, every initial segment of the set of ordinals $M \cap \beta_{M,N}$ is in $N$, since there are only countably many initial segments. \begin{definition} A finite set $A \subseteq \mathcal X$ is said to be \emph{adequate} if for all $M$ and $N$ in $A$, either $M < N$, $M \sim N$, or $N < M$. \end{definition} Note that $A$ is adequate iff for all $M$ and $N$ in $A$, $\{ M, N \}$ is adequate. If $A$ is adequate and $B \subseteq A$, then $B$ is adequate. If $M$ and $N$ are in an adequate set $A$, then either $M \le N$ or $N \le M$. \begin{lemma} Suppose that $M \le N$. Then $$ M \cap \beta_{M,N} = M \cap N \cap \kappa = M \cap N \cap \beta_{M,N}. $$ \end{lemma} \begin{proof} Since $M \le N$, either $M \cap \beta_{M,N} \in N$ or $M \cap \beta_{M,N} = N \cap \beta_{M,N}$. In either case, $M \cap \beta_{M,N} \subseteq N$. So $M \cap \beta_{M,N} \subseteq M \cap N \cap \kappa$. Conversely, by Proposition 2.11, $M \cap N \cap \kappa \subseteq \beta_{M,N}$, so $M \cap N \cap \kappa \subseteq M \cap \beta_{M,N}$. This proves the first equality. For the second equality, the reverse inclusion is trivial, and the forward inclusion follows from Proposition 2.11. \end{proof} \begin{lemma} If $A$ is adequate, $N \in \mathcal X$, and $A \in N$, then $A \cup \{ N \}$ is adequate. \end{lemma} \begin{proof} Since $A$ is finite, $A \subseteq N$. So it suffices to show that if $M \in N \cap \mathcal X$, then $M < N$. As $M$ is countable, $M \cap \kappa \subseteq N$. So by Proposition 2.11, $M \cap \kappa \subseteq \beta_{M,N}$. Therefore, $M \cap \kappa = M \cap \beta_{M,N}$. As $M \in N$, we have that $M \cap \kappa \in N$ by elementarity. Hence, $M \cap \beta_{M,N} \in N$, so $M < N$. \end{proof} It turns out that if $\{ M, N \}$ is adequate, then which relation holds between $M$ and $N$ is determined by comparing the ordinals $M \cap \omega_1$ and $N \cap \omega_1$. \begin{lemma} Let $\{ M, N \}$ be adequate. Then: \begin{enumerate} \item $M < N$ iff $M \cap \omega_1 < N \cap \omega_1$; \item $M \sim N$ iff $M \cap \omega_1 = N \cap \omega_1$; \item $M \le N$ iff $M \cap \omega_1 \le N \cap \omega_1$. \end{enumerate} \end{lemma} \begin{proof} Suppose that $M < N$, and we will show that $M \cap \omega_1 < N \cap \omega_1$. Since $\beta_{M,N}$ has uncountable cofinality, $\omega_1 \le \beta_{M,N}$. Therefore, $M \cap \omega_1$ is an initial segment of $M \cap \beta_{M,N}$, and hence is in $N$. So $M \cap \omega_1 < N \cap \omega_1$. Suppose that $M \sim N$, and we will show that $M \cap \omega_1 = N \cap \omega_1$. Then $M \cap \beta_{M,N} = N \cap \beta_{M,N}$. Since $\omega_1 \le \beta_{M,N}$, $M \cap \omega_1 = N \cap \omega_1$. Conversely, if $M \cap \omega_1 < N \cap \omega_1$, then the implications which we just proved rule out the possibilities that $N < M$ and $N \sim M$. Therefore, $M < N$. This completes the proof of (1) and (2), and (3) follows immediately. \end{proof} \begin{lemma} Let $A$ be an adequate set. Then the relation $<$ is irreflexive and transitive on $A$, $\sim$ is an equivalence relation on $A$, $\le$ is transitive on $A$, and the relations $<$ and $\le$ respect $\sim$. \end{lemma} \begin{proof} Immediate from Lemma 2.17. \end{proof} We state a closure property of $\mathcal X$ as an assumption. \begin{assumption} Suppose that $M$ and $N$ are in $\mathcal X$ and $\{ M, N \}$ is adequate. Then $M \cap N \in \mathcal X$. \end{assumption} The proof of this assumption depends on the actual definition of $\mathcal X$, which we are not giving here. We note that this assumption is fairly easy to verify in the case that $\kappa = \lambda$. However, when $\lambda > \kappa$, the assumption is highly nontrivial, and requires a very careful definition of $\mathcal X$. Similar comments apply to the existence of stationarily many simple models, which we describe next. \begin{definition} A set $N \in \mathcal X$ is said to be \emph{simple} if for all $M \in \mathcal X$, if $M < N$ then $M \cap N \in N$. \end{definition} \begin{definition} A set $P \in \mathcal Y$ is said to be \emph{simple} if for all $M \in \mathcal X$, $M \cap P \in P$. \end{definition} \begin{assumption} The set of $N \in P_{\omega_1}(H(\lambda))$ such that $N \in \mathcal X$ and $N$ is simple is stationary. \end{assumption} \begin{assumption} The set of $P \in P_{\kappa}(H(\lambda))$ such that $P \in \mathcal Y$ and $P$ is simple is stationary. \end{assumption} We now state the main theorems of the basic theory of adequate sets. The proofs of these results depend on technical, and sometimes tedious, facts about comparison points, so we will omit them. First, we handle countable models. \begin{proposition} Let $A$ be adequate and $N \in A$. Then the set $$ B := A \cup \{ M \cap N : M \in A, \ M < N \} $$ is adequate. Moreover, for all $M \in B$, if $M < N$ then $M \cap N \in B$. \end{proposition} \begin{proof} The first statement is proven in \cite[Proposition 1.25]{jk27}. The second statement is easy to prove. \end{proof} \begin{proposition}[Amalgamation over countable models] Let $A$ be adequate, $N \in A$, and suppose that for all $M \in A$, if $M < N$ then $M \cap N \in A$. Assume that $N$ is simple. Suppose that $B$ is adequate and $$ A \cap N \subseteq B \subseteq N. $$ Then $A \cup B$ is adequate. \end{proposition} \begin{proof} See \cite[Proposition 1.29]{jk27}. \end{proof} Let us derive an easy consequence of Proposition 2.24. \begin{lemma} Suppose that $M < N$. Then $M \sim M \cap N$. \end{lemma} Note that by Assumption 2.19, $M \cap N \in \mathcal X$. \begin{proof} Applying Proposition 2.24 to the adequate set $\{ M, N \}$, we get that the set $\{ M, N, M \cap N \}$ is adequate. In particular, $\{ M, M \cap N \}$ is adequate. By Lemma 2.15, $M < N$ implies that $M \cap \beta_{M,N} = M \cap N \cap \beta_{M,N}$. Since $\omega_1 \le \beta_{M,N}$, it follows that $M \cap \omega_1 = M \cap N \cap \omega_1$. As $\{ M, M \cap N \}$ is adequate, Lemma 2.17(2) implies that $M \sim M \cap N$. \end{proof} Next, we handle uncountable models. \begin{proposition} Let $A$ be adequate and $P \in \mathcal Y$. Then the set $$ B := A \cup \{ M \cap P : M \in A \} $$ is adequate. Moreover, for all $M \in B$, $M \cap P \in B$. \end{proposition} \begin{proof} The first statement is proven in \cite[Proposition 1.33]{jk27}. The second statement is easy to prove. \end{proof} \begin{proposition}[Amalgamation over uncountable models] Let $A$ be adequate, $P \in \mathcal Y$, and suppose that for all $M \in A$, $M \cap P \in A$. Assume that $P$ is simple. Suppose that $B$ is adequate and $$ A \cap P \subseteq B \subseteq P. $$ Then $A \cup B$ is adequate. \end{proposition} \begin{proof} See \cite[Proposition 1.35]{jk27}. \end{proof} \begin{lemma} Suppose that $M \in \mathcal X$ and $P \in \mathcal Y$. Then $M \sim M \cap P$. \end{lemma} Note that by Assumption 2.5(2), $M \cap P \in \mathcal X$. \begin{proof} Applying Proposition 2.27 to the adequate set $\{ M \}$, we get that $\{ M, M \cap P \}$ is adequate. Since $\omega_1 \le P \cap \kappa$, we have that $M \cap \omega_1 = M \cap P \cap \omega_1$. Hence, by Lemma 2.17(2), $M \sim M \cap P$. \end{proof} We will need one more result about simple models. \begin{lemma} Suppose that $N \in \mathcal X$ is simple and $P \in \mathcal Y \cap N$ is simple. Then $N \cap P$ is simple. \end{lemma} Note that by Assumption 2.5(2), $N \cap P \in \mathcal X$. \begin{proof} Let $M \in \mathcal X$ be such that $M < N \cap P$, and we will show that $M \cap (N \cap P) \in N \cap P$. It suffices to show that $M \cap N \cap P < N$. For then, since $N$ is simple, $$ M \cap (N \cap P) = (M \cap N \cap P) \cap N \in N, $$ and since $P$ is simple, $$ M \cap (N \cap P) = (M \cap N \cap P) \cap P \in P. $$ So $M \cap (N \cap P) \in N \cap P$. Since $M < N \cap P$, we have that $M \cap \beta_{M,N \cap P} \in N \cap P$. In particular, $M \cap \beta_{M,N \cap P} \in N$. We claim that $\beta_{M \cap N \cap P,N} \le \beta_{M,N \cap P}$. If not, then by Lemma 2.10(2), we can fix $\gamma \in (M \cap N \cap P) \cap [\beta_{M,N \cap P},\beta_{M \cap N \cap P,N})$. Then by Proposition 2.11, $\gamma < \beta_{M,N \cap P}$, which is a contradiction. Since $M \cap \beta_{M,N \cap P} \in N$ and $\beta_{M \cap N \cap P,N} \le \beta_{M,N \cap P}$, it follows that $$ M \cap \beta_{M \cap N \cap P,N} \in N. $$ But $M < N \cap P$ implies that $$ M \cap \beta_{M \cap N \cap P,N} \subseteq M \cap \beta_{M,N \cap P} \subseteq N \cap P. $$ Thus, $$ M \cap \beta_{M \cap N \cap P,N} = (M \cap N \cap P) \cap \beta_{M \cap N \cap P,N}, $$ and so this set is in $N$. Hence, $M \cap N \cap P < N$. \end{proof} We end this section by deriving some specialized consequences of adequacy, which will play a role in the arguments concerning the forcing poset developed in Part II. \begin{definition} A set $a \subseteq P_{\omega_1}(\kappa)$ is said to be an \emph{$\in$-chain} if for all $x$ and $y$ in $a$, either $x = y$, $x \in Sk(y)$, or $y \in Sk(x)$. \end{definition} \begin{lemma} Suppose that $a \subseteq P_{\omega_1}(\kappa)$ is an $\in$-chain. Assume that for all $x \in a$, $Sk(x) \cap \kappa = x$. Then for all $x$ and $y$ in $a$, $x \in Sk(y)$ iff $\sup(x) < \sup(y)$, and $x = y$ iff $\sup(x) = \sup(y)$. \end{lemma} \begin{proof} Straightforward. \end{proof} \begin{lemma} Let $A$ be an adequate set, $\alpha \in \Lambda$, and assume that $Sk(\alpha) \cap \kappa = \alpha$. Then the set $$ C := \{ M \cap \alpha : M \in A, \ \alpha \in M \} $$ is a finite $\in$-chain and a subset of $Sk(\alpha)$. \end{lemma} \begin{proof} By Assumption 2.6, $C$ is a subset of $Sk(\alpha)$. Let $M$ and $N$ be in $A$ such that $\alpha \in M \cap N$. We will show that either $M \cap \alpha = N \cap \alpha$, $M \cap \alpha \in Sk(N \cap \alpha)$, or $N \cap \alpha \in Sk(M \cap \alpha)$. Without loss of generality, assume that $M \le N$. Then either $M \cap \beta_{M,N} = N \cap \beta_{M,N}$, or $M \cap \beta_{M,N} \in N$. Since $\alpha \in M \cap N \cap \kappa$, it follows that $\alpha < \beta_{M,N}$ by Proposition 2.11. So if $M \cap \beta_{M,N} = N \cap \beta_{M,N}$, then intersecting both sides of this equation with $\alpha$ we get that $M \cap \alpha = N \cap \alpha$. Assume that $M \cap \beta_{M,N} \in N$. Then since $M \cap \alpha$ is an initial segment of $M \cap \beta_{M,N}$, $M \cap \alpha \in N$. By Lemma 2.8(1), $M \cap \alpha \in Sk(N \cap \alpha)$. \end{proof} \bigskip \addcontentsline{toc}{section}{3. Remainders} \textbf{\S 3. Remainders} \stepcounter{section} \bigskip In the standard development of the basic ideas of adequate sets, the next topic which comes up is the idea of a remainder point (see \cite[Section 2]{jk27}). In this paper, however, only a particular kind of remainder point will be relevant, namely, those which are in the set $r^*$ defined next. \begin{definition} Let $A$ be an adequate set. Define $r^*(A)$ as the set of ordinals $\gamma$ satisfying that for some $K$ and $M$ in $A$ with $K \sim M$, $$ \gamma = \min((M \cap \kappa) \setminus \beta_{K,M}). $$ \end{definition} Note that $r^*(A)$ is finite. Also, $A \subseteq B$ implies that $r^*(A) \subseteq r^*(B)$. Before analyzing the set $r^*(A)$, we first prove a very useful lemma. \begin{lemma} Suppose that $M$ and $N$ are in $\mathcal X$ and $\{ M, N \}$ is adequate. Assume that $\alpha$ and $\gamma$ are uncountable ordinals satisfying: \begin{enumerate} \item $\alpha \in M \cap \kappa$; \item $\gamma \in (N \cap \kappa) \cup \{ \kappa \}$; \item $\alpha \ne \gamma$; \item $M \cap \alpha = N \cap \gamma$. \end{enumerate} Then $M \sim N$ and $\alpha = \min((M \cap \kappa) \setminus \beta_{M,N})$. \end{lemma} \begin{proof} Since $\alpha$ and $\gamma$ are uncountable, $M \cap \alpha = N \cap \gamma$ implies that $M \cap \omega_1 = N \cap \omega_1$. As $\{ M, N \}$ is adequate, Lemma 2.17(2) implies that $M \sim N$. In particular, $M \cap \beta_{M,N} = N \cap \beta_{M,N}$. Since $\alpha \ne \gamma$, either $\gamma < \alpha$ or $\alpha < \gamma$. First, assume that $\gamma < \alpha$. Then $\gamma \in N \cap \kappa$. Since $\gamma < \alpha$ and $M \cap \alpha = N \cap \gamma$, it follows that $\gamma \notin M$. But $\gamma \in N \cap \kappa$. As $M \cap \beta_{M,N} = N \cap \beta_{M,N}$ and $\gamma \notin M$, $\beta_{M,N} \le \gamma$. So $\beta_{M,N} \le \gamma < \alpha$. We claim that $\alpha = \min((M \cap \kappa) \setminus \beta_{M,N})$. If not, then there is $\alpha_0 \in M \cap \kappa$ such that $\beta_{M,N} \le \alpha_0 < \alpha$. Then $\alpha_0 \in M \cap \alpha = N \cap \gamma$, so $\alpha_0 \in N$. By Proposition 2.11, $\alpha_0 \in M \cap N \cap \kappa \subseteq \beta_{M,N}$. So $\alpha_0 < \beta_{M,N}$, which contradicts the choice of $\alpha_0$. Secondly, assume that $\alpha < \gamma$. Since $M \cap \alpha = N \cap \gamma$, we have that $\alpha \notin N$. As $M \cap \beta_{M,N} = N \cap \beta_{M,N}$ and $\alpha \in M \cap \kappa$, it follows that $\beta_{M,N} \le \alpha$. We claim that $\alpha = \min((M \cap \kappa) \setminus \beta_{M,N})$. If not, then there is $\alpha_0 \in M \cap \kappa$ such that $\beta_{M,N} \le \alpha_0 < \alpha$. But then $\alpha_0 \in M \cap \alpha = N \cap \gamma$. By Proposition 2.11, it follows that $\alpha_0 \in M \cap N \cap \kappa \subseteq \beta_{M,N}$. So $\alpha_0 < \beta_{M,N}$, which contradicts the choice of $\alpha_0$. \end{proof} The main goal of this section is to prove Proposition 3.5, which handles models in $\mathcal X$, and Proposition 3.8, which handles models in $\mathcal Y$. First, we analyze $r^*$ in the context of models in $\mathcal X$. \begin{lemma} Suppose that $M < N$ and $(M \cap \kappa) \setminus \beta_{M,N} \ne \emptyset$. Then $$ \min((M \cap \kappa) \setminus \beta_{M,N}) = \min((M \cap \kappa) \setminus \beta_{M,M \cap N}). $$ \end{lemma} Note that by Assumption 2.19, $M < N$ implies that $M \cap N \in \mathcal X$. \begin{proof} By Lemma 2.26, $M \sim M \cap N$. Let $\alpha := \min((M \cap \kappa) \setminus \beta_{M,N})$. Then by Lemma 2.15 and the minimality of $\alpha$, $$ M \cap N \cap \kappa = M \cap \beta_{M,N} = M \cap \alpha. $$ So $M \cap \alpha = M \cap N \cap \kappa$. Applying Lemma 3.2 to $M$, $M \cap N$, $\alpha$, and $\kappa$, we get that $\alpha = \min((M \cap \kappa) \setminus \beta_{M,M \cap N})$. \end{proof} \begin{lemma} Let $\{ K, M, N \}$ be adequate, and assume that $M \in N \cap \mathcal X$ and $K \sim M$. Then: \begin{enumerate} \item $K < N$ and $M \sim K \cap N$; \item if $\alpha = \min((M \cap \kappa) \setminus \beta_{K,M})$, then $\alpha = \min((M \cap \kappa) \setminus \beta_{M,K \cap N})$; \item if $\alpha = \min((K \cap \kappa) \setminus \beta_{K,M})$, then either $\alpha = \min((K \cap N \cap \kappa) \setminus \beta_{K \cap N,M})$, or $\alpha = \min((K \cap \kappa) \setminus \beta_{K,K \cap N})$. \end{enumerate} \end{lemma} \begin{proof} (1) Since $K \sim M$, it follows that $K \cap \omega_1 = M \cap \omega_1$ by Lemma 2.17(2). As $M \in N$, $M \cap \omega_1 < N \cap \omega_1$. Therefore, $K \cap \omega_1 < N \cap \omega_1$. By Lemma 2.17(1), $K < N$. By Proposition 2.24, $\{ K, M, N, K \cap N \}$ is adequate. By Lemma 2.26, $K \sim K \cap N$. So $M \sim K \sim K \cap N$, which by Lemma 2.18 implies that $M \sim K \cap N$. (2) We apply Lemma 3.2 to the objects $M$, $K \cap N$, $\alpha$, and $\gamma$, where $$ \gamma := \min(((K \cap N \cap \kappa) \cup \{ \kappa \}) \setminus \beta_{K,M}). $$ Provided that the assumptions of Lemma 3.2 are true for these objects, we get that $\alpha = \min((M \cap \kappa) \setminus \beta_{M,K \cap N})$, which finishes the proof of (2). Assumptions (1) and (2) of Lemma 3.2 are immediate, and (3) follows from Proposition 2.11 since $\alpha$ and $\gamma$ are greater than or equal to $\beta_{K,M}$. For assumption (4), we need to show that $M \cap \alpha = K \cap N \cap \gamma$. By the minimality of $\alpha$ and $\gamma$, we have that $M \cap \alpha = M \cap \beta_{K,M}$ and $K \cap N \cap \gamma = K \cap N \cap \beta_{K,M}$. So it suffices to show that $$ M \cap \beta_{K,M} = K \cap N \cap \beta_{K,M}. $$ Since $K \sim M$, $M \cap \beta_{K,M} = K \cap \beta_{K,M}$. This last equation implies the reverse inclusion of the displayed equation, and noting that $M \in N$ implies that $M \subseteq N$, it implies the forward inclusion as well. (3) Let $$ \gamma := \min((M \cap \kappa) \cup \{ \kappa \}) \setminus \beta_{K,M}). $$ Then by the minimality of $\alpha$ and $\gamma$ and the fact that $K \sim M$, we have that $$ K \cap \alpha = K \cap \beta_{K,M} = M \cap \beta_{K,M} = M \cap \gamma. $$ \bigskip First, assume that $\alpha \in N$. We apply Lemma 3.2 to the objects $K \cap N$, $M$, $\alpha$, and $\gamma$. Provided that the assumptions of Lemma 3.2 hold for these objects, we can conclude that $$ \alpha = \min((K \cap N \cap \kappa) \setminus \beta_{K \cap N,M}), $$ which completes the proof of (3). Assumptions (1) and (2) of Lemma 3.2 are immediate, and (3) follows from Proposition 2.11 since $\alpha$ and $\gamma$ are greater than or equal to $\beta_{K,M}$. For (4), we need to show that $K \cap N \cap \alpha = M \cap \gamma$. From the above, we already know that $K \cap \alpha = M \cap \gamma$, so it suffices to show that $$ K \cap \alpha = K \cap N \cap \alpha. $$ Since $K < N$, $K \cap \beta_{K,N} = K \cap N \cap \beta_{K,N}$ by Lemma 2.15. As $\alpha \in K \cap N \cap \kappa$, we have that $\alpha < \beta_{K,N}$ by Proposition 2.11. Hence, $$ K \cap \alpha = K \cap \beta_{K,N} \cap \alpha = K \cap N \cap \beta_{K,N} \cap \alpha = K \cap N \cap \alpha. $$ \bigskip Secondly, assume that $\alpha \notin N$. Since $K < N$ by (1), we have that $K \cap \beta_{K,N} \subseteq N$, and therefore $\beta_{K,N} \le \alpha$. To show that $\alpha = \min((K \cap \kappa) \setminus \beta_{K,K \cap N})$, by Lemma 3.3 it suffices to show that $\alpha = \min((K \cap \kappa) \setminus \beta_{K,N})$. Suppose for a contradiction that $\alpha$ is not equal to $\min((K \cap \kappa) \setminus \beta_{K,N})$. Then there is $\alpha_0 \in K$ such that $\beta_{K,N} \le \alpha_0 < \alpha$. But then $\alpha_0 \in K \cap \alpha$. Since $\alpha = \min((K \cap \kappa) \setminus \beta_{K,M})$, it follows that $\alpha_0 < \beta_{K,M}$. Now $K \sim M$ means that $K \cap \beta_{K,M} = M \cap \beta_{K,M}$. As $\alpha_0 \in K \cap \beta_{K,M}$, we have that $\alpha_0 \in M$. But $M \in N$, so $M \subseteq N$, and hence $\alpha_0 \in N$. Therefore, $\alpha_0 \in K \cap N \cap \kappa$, which implies that $\alpha_0 < \beta_{K,N}$ by Proposition 2.11, contradicting the choice of $\alpha_0$. \end{proof} \begin{proposition} Assume that $A$ is adequate, $N \in A$ is simple, and for all $M \in A$, if $M < N$ then $M \cap N \in A$. Let $B$ be adequate, and suppose that $$ A \cap N \subseteq B \subseteq N. $$ Then $$ r^*(A \cup B) = r^*(A) \cup r^*(B). $$ \end{proposition} \begin{proof} By Proposition 2.25, $A \cup B$ is adequate. The reverse inclusion is immediate. For the forward inclusion, it suffices to show that if $K \in A$, $M \in B$, and $K \sim M$, then $r^*(\{ K, M \}) \subseteq r^*(A) \cup r^*(B)$. Since $M \in B \subseteq N$, we have that $M \in N$. Since $K$ and $N$ are in $A$ and $K \sim M$, Lemma 3.4(1) implies that $K < N$ and $M \sim K \cap N$. So by our assumptions about $A$, $K \cap N \in A$. And since $N$ is simple, $K \cap N \in N$. Therefore, $K \cap N \in A \cap N \subseteq B$. So $K \cap N$ and $M$ are both in $B$. Also, by Lemma 2.26, $K \sim K \cap N$. Suppose that $\alpha \in r^*(\{ K, M \})$, and we will show that $\alpha \in r^*(A) \cup r^*(B)$. First, assume that $\alpha = \min((M \cap \kappa) \setminus \beta_{K,M})$. Then by Lemma 3.4(2), $\alpha = \min((M \cap \kappa) \setminus \beta_{M,K \cap N})$. Since $K \cap N$ and $M$ are in $B$ and $M \sim K \cap N$, $\alpha \in r^*(B)$. Secondly, assume that $\alpha = \min((K \cap \kappa) \setminus \beta_{K,M})$. Then by Lemma 3.4(3), either $\alpha = \min((K \cap N \cap \kappa) \setminus \beta_{K \cap N,M})$, or $\alpha = \min((K \cap \kappa) \setminus \beta_{K,K \cap N})$. In the first case, $\alpha \in r^*(B)$, since $K \cap N$ and $M$ are in $B$ and $M \sim K \cap N$. In the second case, $\alpha \in r^*(A)$, since $K$ and $K \cap N$ are in $A$ and $K \sim K \cap N$. \end{proof} Next, we analyze the set $r^*$ in the context of models in $\mathcal Y$. \begin{lemma} Suppose that $M \in \mathcal X$, $P \in \mathcal Y$, and $(M \cap \kappa) \setminus (P \cap \kappa) \ne \emptyset$. Then $$ \min((M \cap \kappa) \setminus (P \cap \kappa)) = \min((M \cap \kappa) \setminus \beta_{M \cap P,M}). $$ \end{lemma} \begin{proof} By Lemma 2.29, $M \sim M \cap P$. Let $\alpha := \min((M \cap \kappa) \setminus (P \cap \kappa))$. Then by the minimality of $\alpha$, $$ M \cap \alpha = M \cap P \cap \kappa. $$ Applying Lemma 3.2 to $M$, $M \cap P$, $\alpha$, and $\kappa$, we get that $\alpha = \min((M \cap \kappa) \setminus \beta_{M,M \cap P})$. \end{proof} \begin{lemma} Let $K \in \mathcal X$, $P \in \mathcal Y$, and assume that $M \in P \cap \mathcal X$ and $K \sim M$. Then: \begin{enumerate} \item $K \cap P \sim M$; \item if $\alpha = \min((M \cap \kappa) \setminus \beta_{K,M})$, then $\alpha = \min((M \cap \kappa) \setminus \beta_{M,K \cap P})$; \item if $\alpha = \min((K \cap \kappa) \setminus \beta_{K,M})$, then either $\alpha = \min((K \cap P \cap \kappa) \setminus \beta_{K \cap P,M})$, or $\alpha = \min((K \cap \kappa) \setminus \beta_{K \cap P,K})$. \end{enumerate} \end{lemma} \begin{proof} (1) Since $K \sim M$, $K \cap \beta_{K,M} = M \cap \beta_{K,M}$. By Lemma 2.12, since $M \in P$, $\beta_{K,M} < P \cap \kappa$. Thus, $$ K \cap P \cap \beta_{K,M} = M \cap \beta_{K,M}. $$ By Lemma 2.10(3), $\beta_{K \cap P,M} \le \beta_{K,M}$. Therefore, $$ K \cap P \cap \beta_{K \cap P,M} = M \cap \beta_{K \cap P,M}. $$ So $K \cap P \sim M$. (2) We apply Lemma 3.2 to the objects $M$, $K \cap P$, $\alpha$, and $\gamma$, where $$ \gamma := \min(((K \cap P \cap \kappa) \cup \{ \kappa \}) \setminus \beta_{K,M}). $$ Provided that the assumptions of Lemma 3.2 are true for these objects, it follows that $\alpha = \min((M \cap \kappa) \setminus \beta_{M,K \cap P})$, and we are done. Assumptions (1) and (2) are immediate, and (3) follows from Proposition 2.11 since $\alpha$ and $\gamma$ are greater than or equal to $\beta_{K,M}$. It remains to show that $M \cap \alpha = K \cap P \cap \gamma$. By the minimality of $\alpha$ and $\gamma$, we have that $M \cap \alpha = M \cap \beta_{K,M}$ and $K \cap P \cap \gamma = K \cap P \cap \beta_{K,M}$. So it suffices to show that $$ M \cap \beta_{K,M} = K \cap P \cap \beta_{K,M}. $$ Since $K \sim M$, $M \cap \beta_{K,M} = K \cap \beta_{K,M}$. Hence, it is enough to show that $$ K \cap \beta_{K,M} = K \cap P \cap \beta_{K,M}. $$ But this follows immediately from the fact that $\beta_{K,M} < P \cap \kappa$, which is true by Lemma 2.12. (3) Let $$ \gamma := \min(((M \cap \kappa) \cup \{ \kappa \}) \setminus \beta_{K,M}). $$ By Proposition 2.11, $\alpha \ne \gamma$, since $\alpha$ and $\gamma$ are greater than or equal to $\beta_{K,M}$. By the minimality of $\alpha$ and $\gamma$ and the fact that $K \sim M$, we have that $$ K \cap \alpha = K \cap \beta_{K,M} = M \cap \beta_{K,M} = M \cap \gamma. $$ First, assume that $\alpha < P \cap \kappa$. Then $K \cap \alpha = K \cap P \cap \alpha$. Therefore, by the last paragraph, $$ K \cap P \cap \alpha = M \cap \gamma. $$ Applying Lemma 3.2 to the objects $K \cap P$, $M$, $\alpha$, and $\gamma$, it follows that $$ \alpha = \min((K \cap P \cap \kappa) \setminus \beta_{K \cap P,M}). $$ Secondly, assume that $P \cap \kappa \le \alpha$. To show that $\alpha = \min((K \cap \kappa) \setminus \beta_{K \cap P,K})$, by Lemma 3.6 it suffices to show that $\alpha = \min((K \cap \kappa) \setminus (P \cap \kappa))$. Suppose for a contradiction that $\alpha$ is not equal to $\min((K \cap \kappa) \setminus (P \cap \kappa))$. Then there is $\alpha_0 \in K \cap \kappa$ such that $P \cap \kappa \le \alpha_0 < \alpha$. Then $\alpha_0 \in K \cap \alpha$. Since $\alpha = \min((K \cap \kappa) \setminus \beta_{K,M})$, it follows that $\alpha_0 < \beta_{K,M}$. Hence, $\alpha_0 \in K \cap \beta_{K,M}$. As $K \sim M$, we have that $K \cap \beta_{K,M} = M \cap \beta_{K,M}$. So $\alpha_0 \in M$. But $M \in P$, so $M \subseteq P$. Therefore, $\alpha_0 \in P \cap \kappa$, which contradicts the choice of $\alpha_0$. \end{proof} \begin{proposition} Suppose that $A$ is adequate, $P \in \mathcal Y$ is simple, and for all $M \in A$, $M \cap P \in A$. Let $B$ be adequate, and suppose that $$ A \cap P \subseteq B \subseteq P. $$ Then $$ r^*(A \cup B) = r^*(A) \cup r^*(B). $$ \end{proposition} \begin{proof} By Proposition 2.28, $A \cup B$ is adequate. The reverse inclusion is immediate. For the forward inclusion, it suffices to show that if $K \in A$, $M \in B$, and $K \sim M$, then $r^*(\{ K, M \}) \subseteq r^*(A) \cup r^*(B)$. Since $M \in B \subseteq P$, we have that $M \in P$. Thus, the assumptions of Lemma 3.7 are satisfied. By Lemma 3.7(1), $M \sim K \cap P$. By the assumptions on $A$, $K \cap P \in A$. As $P$ is simple, $K \cap P \in P$. So $K \cap P \in A \cap P \subseteq B$. Hence, $K \cap P$ is in $B$. By Lemma 2.29, $K \sim K \cap P$. Suppose that $\alpha \in r^*(\{ K, M \})$, and we will show that $\alpha \in r^*(A) \cup r^*(B)$. First, assume that $\alpha = \min((M \cap \kappa) \setminus \beta_{K,M})$. Then by Lemma 3.7(2), $\alpha = \min((M \cap \kappa) \setminus \beta_{M,K \cap P})$. Since $K \cap P$ and $M$ are in $B$ and $M \sim K \cap P$, it follows that $\alpha \in r^*(B)$. Secondly, assume that $\alpha = \min((K \cap \kappa) \setminus \beta_{K,M})$. Then by Lemma 3.7(3), either $\alpha = \min((K \cap P \cap \kappa) \setminus \beta_{K \cap P,M})$, or $\alpha = \min((K \cap \kappa) \setminus \beta_{K \cap P,K})$. In the first case, since $K \cap P$ and $M$ are in $B$ and $K \cap P \sim M$, it follows that $\alpha \in r^*(B)$. In the second case, since $K$ and $K \cap P$ are in $A$ and $K \sim K \cap P$, it follows that $\alpha \in r^*(A)$. \end{proof} \bigskip \part{The Single Forcing} \bigskip \addcontentsline{toc}{section}{4. The forcing poset} \textbf{\S 4. The forcing poset} \stepcounter{section} \bigskip We introduce a forcing poset for adding a partial square sequence to a stationary set $S \subseteq \kappa \cap \mathrm{cof}(>\! \omega)$. This forcing poset will preserve $\omega_1$, is $\kappa$-c.c., and if $\kappa > \omega_2$, then it will collapse $\kappa$ to become $\omega_2$. \begin{notation} Fix, for the remainder of Part 2, a set $S \subseteq \Lambda$ which is stationary in $\kappa$ and satisfies that for all $\alpha \in S$, $Sk(\alpha) \cap \kappa = \alpha$. \end{notation} \begin{definition} Let $\mathbb{P}$ be the forcing poset whose conditions are triples of the form $p = (f_p,g_p,A_p)$ satisfying the following requirements:\footnote{We will sometimes refer to $f_p$, $g_p$, and $A_p$ as the $f$, $g$, and $A$ components of $p$.} \begin{enumerate} \item $A_p$ is an adequate set; \item $f_p$ is a function with a finite domain, and for all $x \in \mathrm{dom}(f_p)$, either $x \in S$, or there is $M \in A_p$ and $$ \alpha \in (M \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \} $$ satisfying that $x = M \cap \alpha$; moreover, for all $x \in \mathrm{dom}(f_p)$, $f_p(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$; \item if $x \in \mathrm{dom}(f_p)$, then $f_p(x) \subseteq \mathrm{dom}(f_p)$, and for all $K \in f_p(x)$, $$ f_p(K) = f_p(x) \cap Sk(K); $$ \item $g_p$ is a function whose domain is the set of all pairs $(K,x)$ such that $K \in f_p(x)$, and for all $(K,x) \in \mathrm{dom}(g_p)$, $g_p(K,x)$ is a finite subset of $x \setminus \sup(K)$; \item if $K \in f_p(L)$ and $L \in f_p(x)$, then $g_p(K,x) \subseteq g_p(K,L)$;\footnote{Note that if $K \in f_p(L)$ and $L \in f_p(x)$, then $K \in f_p(x)$ by requirement (3).} \item if $\alpha \in \mathrm{dom}(f_p) \cap S$, $M \in A_p$, and $\alpha \in M$, then $M \cap \alpha \in f_p(\alpha)$; \item $r^*(A_p) \cap S \subseteq \mathrm{dom}(f_p)$. \end{enumerate} For conditions $p$ and $q$ in $\mathbb{P}$, we let $q \le p$ if: \begin{enumerate} \item[(a)] $A_p \subseteq A_q$; \item[(b)] $\mathrm{dom}(f_p) \subseteq \mathrm{dom}(f_q)$, and for all $x \in \mathrm{dom}(f_p)$, $f_p(x) \subseteq f_q(x)$; \item[(c)] for all $(K,x) \in \mathrm{dom}(g_p)$, $g_p(K,x) \subseteq g_q(K,x)$; \item[(d)] if $K$ and $x$ are in $\mathrm{dom}(f_p)$ and $K \in f_q(x)$, then $K \in f_p(x)$. \end{enumerate} \end{definition} Let us summarize some of the main properties which we will prove about $\mathbb{P}$. The forcing poset $\mathbb{P}$ will be shown to be strongly proper on a stationary set, and thus preserve $\omega_1$, and to be $\kappa$-c.c. In particular, $\mathbb{P}$ preserves the stationarity of $S$. If $\kappa > \omega_2$, then $\mathbb{P}$ collapses all cardinals $\mu$ such that $\omega_1 < \mu < \kappa$ to have size $\omega_1$, and hence forces that $\kappa = \omega_2$. Finally, $\mathbb{P}$ forces that there exists a partial square sequence on $S$, and in particular, forces that $S$ is in the approachability ideal $I[\omega_2]$. The properties of $\mathbb{P}$ just listed will be proved in Sections 4--7. In Sections 8 and 9, we will derive some additional information about the forcing poset $\mathbb{P}$, and use this information to show that certain quotients of $\mathbb{P}$ have the $\omega_1$-approximation property. In the remainder of the current section, we will prove some basic facts about $\mathbb{P}$ which we will need. \bigskip \begin{lemma} Let $p \in \mathbb{P}$ and $x \in \mathrm{dom}(f_p)$. Then: \begin{enumerate} \item $Sk(x) \cap \kappa = x$; \item for all $K \in f_p(x)$, $K \subseteq x$; \item if $N \in \mathcal X$, $\alpha \in S$, $x \in N \setminus S$, and $x \subseteq \alpha$, then $x \in Sk(N \cap \alpha)$; \item if $P \in \mathcal Y$ and $\sup(x) < P \cap \kappa$, then $x \in P$. \end{enumerate} \end{lemma} \begin{proof} (1) By Definition 4.2(2), either $x \in S$, or there is $M \in A_p$ and $\alpha \in (M \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$. Then $Sk(x) \cap \kappa = x$ holds by Notation 4.1 in the first case, and by Lemma 2.7(2) in the second case. (2) Suppose that $K \in f_p(x)$, and we will show that $K \subseteq x$. By Definition 4.2(2,3), $K$ is a countable subset of $\kappa$ in $Sk(x)$. By the elementarity of $Sk(x)$, $K \subseteq Sk(x)$. As $K \subseteq \kappa$, it follows by (1) that $K \subseteq Sk(x) \cap \kappa = x$. (3) Fix $M \in A_p$ and $\beta \in (M \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \beta$. Since $M \cap \beta = x \in N$, we have that $x = M \cap \beta \in Sk(N \cap \beta)$ by Lemma 2.8(1). If $\beta \le \alpha$, then $x \in Sk(N \cap \beta) \subseteq Sk(N \cap \alpha)$, and we are done. If $\alpha < \beta$, then since $x = M \cap \beta \subseteq \alpha$, $M \cap \beta = M \cap \alpha$. So $M \cap \alpha = x \in N$. Hence, $x \in Sk(N \cap \alpha)$ by Lemma 2.8(1). (4) If $x \in S$, then $x = \sup(x) \in P$. Otherwise by Definition 4.2(2) there is $M \in A_p$ and $\beta \in (M \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \beta$. Let $\alpha := P \cap \kappa$. Then by the elementarity of $P$, $Sk(\alpha) \cap \kappa = \alpha$. By Assumption 2.6, $M \cap \alpha \in Sk(\alpha) \subseteq P$. So $M \cap \alpha \in P$. If $\beta \le \alpha$, then $x = M \cap \beta$ is an initial segment of $M \cap \alpha$, and hence is in $P$. If $\alpha < \beta$, then since $\sup(x) = \sup(M \cap \beta) < \alpha$, $M \cap \alpha = M \cap \beta = x$, which is in $P$. \end{proof} \begin{lemma} Let $p \in \mathbb{P}$ and $z \in \mathrm{dom}(f_p)$. Then for all $x$ and $y$ in $f_p(z)$, $x \in Sk(y)$ iff $\sup(x) < \sup(y)$, and $x = y$ iff $\sup(x) = \sup(y)$. \end{lemma} \begin{proof} By Definition 4.2(3), $f_p(z) \subseteq \mathrm{dom}(f_p)$. So by Lemma 4.3(1), if $x \in f_p(z)$ then $Sk(x) \cap \kappa = x$. The lemma now follows from Lemma 2.32, letting $a = f_p(z)$. \end{proof} \begin{lemma} Let $p \in \mathbb{P}$ and $\alpha \in S \setminus \mathrm{dom}(f_p)$. If $M \in A_p$ and $\alpha \in M$, then $M \cap \alpha$ is not in $\mathrm{dom}(f_p)$. \end{lemma} \begin{proof} Suppose for a contradiction that $M \cap \alpha \in \mathrm{dom}(f_p)$. Then by Definition 4.2(2), there is $M_1 \in A_p$ and $\beta \in (M_1 \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $M \cap \alpha = M_1 \cap \beta$. Since $\beta \in \mathrm{dom}(f_p)$ and $\alpha \notin \mathrm{dom}(f_p)$, $\alpha \ne \beta$. By Lemma 3.2 applied to $M$, $M_1$, $\alpha$, and $\beta$, we have that $M \sim M_1$ and $$ \alpha = \min((M \cap \kappa) \setminus \beta_{M,M_1}). $$ In particular, $\alpha \in r^*(A_p) \cap S$. So by Definition 4.2(7), $\alpha \in \mathrm{dom}(f_p)$, which contradicts our assumptions. \end{proof} We show next that for any condition $p$ and any ordinal $\alpha$ in $S$, there is $q \le p$ with $\alpha \in \mathrm{dom}(f_q)$. Among other things, this fact will allow us to prove that $\mathbb{P}$ adds a partial square sequence whose domain is all of $S$. \begin{lemma} Let $p \in \mathbb{P}$, and let $\alpha$ and $\beta$ be distinct ordinals in $S \setminus \mathrm{dom}(f_p)$. Then the sets $$ \mathrm{dom}(f_p), \ \{ M \cap \alpha : M \in A_p, \ \alpha \in M \}, \ \{ N \cap \beta : N \in A_p, \ \beta \in N \} $$ are pairwise disjoint. \end{lemma} \begin{proof} By Lemma 4.5, the first and second sets are disjoint, and the first and third sets are disjoint. If the second and third sets are not disjoint, then $M \cap \alpha = N \cap \beta$, for some $M$ and $N$ in $A_p$ with $\alpha \in M$ and $\beta \in N$. Applying Lemma 3.2 to $M$, $N$, $\alpha$, and $\beta$, we get that $M \sim N$ and $\alpha = \min((M \cap \kappa) \setminus \beta_{M,N})$. In particular, $\alpha \in r^*(A_p) \cap S$. So by Definition 4.2(7), $\alpha \in \mathrm{dom}(f_p)$, which contradicts our assumptions. \end{proof} \begin{definition} Let $p \in \mathbb{P}$, and let $x$ be a finite subset of $S \setminus \mathrm{dom}(f_p)$. Define $p + x$ as the triple $(f,g,A)$ satisfying: \begin{enumerate} \item $A := A_p$; \item $\mathrm{dom}(f) := \mathrm{dom}(f_p) \cup x \cup \{ M \cap \alpha : M \in A_p, \ \alpha \in M \cap x \}$; \item for each $z \in \mathrm{dom}(f_p)$, $f(z) := f_p(z)$; \item for each $\alpha \in x$, $f(\alpha) := \{ M \cap \alpha : M \in A_p, \ \alpha \in M \}$; \item for each $\alpha \in x$ and $M \in A_p$ with $\alpha \in M$, $f(M \cap \alpha) := f(\alpha) \cap Sk(M \cap \alpha)$, where $f(\alpha)$ was defined in (4); \item the domain of $g$ is the set of pairs $(K,z)$ such that $K \in f(z)$; \item $g(K,z) = g_p(K,z)$ if $(K,z) \in \mathrm{dom}(g_p)$, and $g(K,z) = \emptyset$ if $(K,z) \in \mathrm{dom}(g) \setminus \mathrm{dom}(g_p)$. \end{enumerate} \end{definition} Note that by Lemma 4.6, $f(M \cap \alpha)$ in (5) is well-defined, since the set $M \cap \alpha$ is not in $\mathrm{dom}(f_p)$ and uniquely determines $\alpha$. \begin{lemma} Let $p \in \mathbb{P}$, and let $x$ be a finite subset of $S \setminus \mathrm{dom}(f_p)$. Then $p + x$ is a condition in $\mathbb{P}$, and $p + x \le p$. \end{lemma} \begin{proof} Let $q := p + x$. Assuming that $q$ is a condition, it is easy to check that $q \le p$ from the definition of $q$ and the fact that for all $M$ and $z$ in $\mathrm{dom}(f_p)$, $M \in f_p(z)$ iff $M \in f_q(z)$. To show that $q$ is a condition, we verify requirements (1)--(7) of Definition 4.2. (1), (4), (6), and (7) are immediate from the definition of $q$, and (3) is easy to check using the definition of $f_q$. It remains to prove (2) and (5). (2) Clearly $f_q$ is a function with a finite domain, and every $z \in \mathrm{dom}(f_q)$ has the right form. Let $z \in \mathrm{dom}(f_q)$, and we will show that $f_q(z)$ is a finite $\in$-chain and $f_q(z) \subseteq Sk(z) \setminus S$. This is immediate if $z \in \mathrm{dom}(f_p)$. If $z = M \cap \alpha$, for some $M \in A_p$ with $\alpha \in M \cap x$, then $f_q(M \cap \alpha) = f_q(\alpha) \cap Sk(M \cap \alpha)$ is obviously a subset of $Sk(M \cap \alpha)$, and will be a finite $\in$-chain disjoint from $S$ provided that $f_q(\alpha)$ is. So it suffices to show that $f_q(\alpha) = \{ M \cap \alpha : M \in A_p , \ \alpha \in M \}$ is a finite $\in$-chain and is a subset of $Sk(\alpha) \setminus S$. This set is obviously disjoint from $S$, and it is a finite $\in$-chain and a subset of $Sk(\alpha)$ by Lemma 2.33. (5) Suppose that $K \in f_q(L)$ and $L \in f_q(z)$. We will show that $g_q(K,z) \subseteq g_q(K,L)$. If $z \in \mathrm{dom}(f_p)$, then $f_q(z) = f_p(z)$; this implies that $K$ and $L$ are in $\mathrm{dom}(f_p)$ as well, so $K \in f_p(L)$ and $L \in f_p(z)$. Hence, $g_q(K,z) = g_p(K,z) \subseteq g_p(K,L) = g_q(K,L)$. On the other hand, if $z$ is not in $\mathrm{dom}(f_p)$, then $(K,z) \notin \mathrm{dom}(g_p)$. Therefore, $g_q(K,z) = \emptyset \subseteq g_q(K,L)$. \end{proof} The next lemma will be needed in the amalgamation arguments of Section 7. \begin{lemma} Let $p$ be a condition. Then there is $q \le p$ satisfying that whenever $K \in f_q(x)$ and $x \in f_q(y)$, then $$ g_q(K,x) \subseteq g_q(K,y).\footnote{Note that by Definition 4.2(5), any such condition satisfies that whenever $K \in f_q(x)$ and $x \in f_q(y)$, then $g_q(K,x) = g_q(K,y)$.} $$ Moreover, $f_q = f_p$ and $A_q = A_p$. \end{lemma} \begin{proof} Define $q$ as follows. Let $f_q := f_p$ and $A_q := A_p$. For any $K$ and $y$ such that $K \in f_p(y)$, define $$ g_q(K,y) := \bigcup \{ g_p(K,x) : x = y, \ \textrm{or} \ (K \in f_p(x) \ \textrm{and} \ x \in f_p(y)) \}. $$ It is trivial to check that if $q$ is a condition, then $q \le p$. To show that $q$ is a condition, we verify requirements (1)--(7) of Definition 4.2. (1), (2), (3), (6), and (7) are immediate. It remains to prove (4) and (5). (4) The domain of $g_q$ is equal to the set of pairs $(K,y)$, where $K \in f_q(y)$. Let $K \in f_q(y)$, and we will show that $g_q(K,y)$ is a finite subset of $y \setminus \sup(K)$. By the definition of $g_q(K,y)$ and the fact that $p$ is a condition, it is clear that $g_q(K,y)$ is finite and every ordinal in $g_q(K,y)$ is greater than or equal to $\sup(K)$. It remains to show that $g_q(K,y) \subseteq y$. Let $\xi \in g_q(K,y)$, and we will show that $\xi \in y$. By the definition of $g_q$, either $\xi \in g_p(K,y)$, or else there is some $x$ satisfying that $K \in f_p(x)$, $x \in f_p(y)$, and $\xi \in g_p(K,x)$. In the first case, $\xi \in y$, since $p$ is a condition. In the second case, $\xi \in g_p(K,x) \subseteq x$. But $x \in f_p(y)$, which implies that $x \subseteq y$ by Lemma 4.3(2). So $\xi \in y$. (5) Suppose that $K \in f_q(L)$ and $L \in f_q(y)$. We will show that $g_q(K,y) \subseteq g_q(K,L)$. Since $f_q = f_p$, we have that $K \in f_p(L)$ and $L \in f_p(y)$. Let $\xi \in g_q(K,y)$, and we will show that $\xi \in g_q(K,L)$. By the definition of $g_q(K,y)$, either \begin{enumerate} \item[(a)] $\xi \in g_p(K,y)$, or \item[(b)] there is $x$ such that $K \in f_p(x)$, $x \in f_p(y)$, and $\xi \in g_p(K,x)$. \end{enumerate} In case a, since $p$ is a condition, $g_p(K,y) \subseteq g_p(K,L)$. Also, $g_p(K,L) \subseteq g_q(K,L)$ by the definition of $g_q$. So $\xi \in g_p(K,y) \subseteq g_p(K,L) \subseteq g_q(K,L)$. Consider case b. Since $x$ and $L$ are both in $f_p(y)$, either $L \in f_p(x)$, $x = L$, or $x \in f_p(L)$. Assume that $L \in f_p(x)$. Then, since $p$ is a condition, $$ \xi \in g_p(K,x) \subseteq g_p(K,L) \subseteq g_q(K,L). $$ Assume that $x = L$. Then $$ \xi \in g_p(K,x) = g_p(K,L) \subseteq g_q(K,L). $$ Finally, assume that $x \in f_p(L)$. Then by the definition of $g_q(K,L)$, since $K \in f_p(x)$ and $x \in f_p(L)$, $g_p(K,x) \subseteq g_q(K,L)$. So $\xi \in g_q(K,L)$. \bigskip This completes the proof that $q$ is a condition. To show that $q$ is as required, suppose that $K \in f_q(x)$ and $x \in f_q(y)$, and we will show that $g_q(K,x) \subseteq g_q(K,y)$. Then $K \in f_p(x)$ and $x \in f_p(y)$. Let $\xi \in g_q(K,x)$, and we will show that $\xi \in g_q(K,y)$. By the definition of $g_q(K,x)$, either $\xi \in g_p(K,x)$, or $\xi \in g_p(K,x_0)$ for some $x_0$ with $K \in f_p(x_0)$ and $x_0 \in f_p(x)$. In the second case, $x_0 \in f_p(x)$ and $x \in f_p(y)$ imply that $x_0 \in f_p(y)$. Let $x'$ be equal to $x$ or $x_0$ depending on the first or the second case. Then in either case, $K \in f_p(x')$ and $x' \in f_p(y)$, and also $\xi \in g_p(K,x')$. By the definition of $g_q$, $g_p(K,x') \subseteq g_q(K,y)$. Thus, $\xi \in g_q(K,y)$, as required. \end{proof} \bigskip \addcontentsline{toc}{section}{5. A partial square sequence} \textbf{\S 5. A partial square sequence} \stepcounter{section} \bigskip In Sections 6 and 7, we will prove that $\mathbb{P}$ preserves $\omega_1$, is $\kappa$-c.c., and collapses $\kappa$ to become $\omega_2$. The proofs of these facts are quite involved. So it makes sense, from the expositional point of view, to assume for the time being that they are true, and show that the forcing poset $\mathbb{P}$ does what it is intended to do, namely, to add a partial square sequence on $S$. To be precise, in this section we will assume exactly that $\mathbb{P}$ preserves $\omega_1$, forces that $\kappa$ is equal to $\omega_2$, and that Lemma 7.1 from Section 7 below holds. \bigskip Let $\dot f$ be a $\mathbb{P}$-name for a function such that $\mathbb{P}$ forces that for all $x$, $$ \dot f(x) = \bigcup \{ f_p(x) : p \in \dot G_\mathbb{P}, \ x \in \mathrm{dom}(f_p) \}. $$ It is easy to check that $\mathbb{P}$ forces that $\dot f(x)$ is an $\in$-chain of countable subsets of $\kappa$ and is a subset of $Sk(x) \setminus S$. Note that by Lemma 4.4, $\mathbb{P}$ forces that if $J$ and $K$ are in $\dot f(x)$, then $\sup(J) < \sup(K)$ iff $J \in Sk(K)$. For each $\alpha \in S$, let $\dot c_\alpha$ be a $\mathbb{P}$-name such that $\mathbb{P}$ forces that $$ \dot c_\alpha = \{ \sup(M) : M \in \dot f(\alpha) \}. $$ We will prove that $\mathbb{P}$ forces that the sequence $$ \langle \dot c_\alpha : \alpha \in S \rangle $$ is a partial square sequence on $S$. \begin{lemma} Let $\alpha \in S$. Then $\mathbb{P}$ forces that for all $K \in \dot f(\alpha)$, $$ \dot f(K) = \dot f(\alpha) \cap Sk(K) $$ and $$ \dot c_\alpha \cap \sup(K) = \{ \sup(J) : J \in \dot f(K) \}. $$ \end{lemma} \begin{proof} Straightforward. \end{proof} \begin{lemma} Let $\alpha \in S$. Then $\mathbb{P}$ forces that $\dot c_\alpha$ is a cofinal subset of $\alpha$ with order type $\omega_1$. \end{lemma} \begin{proof} We first show that $\dot c_\alpha$ is forced to be a cofinal subset of $\alpha$. So let $\gamma < \alpha$ and $p \in \mathbb{P}$. Using Lemma 4.8, we can fix $q \le p$ with $\alpha \in \mathrm{dom}(f_q)$. By Lemma 7.1, fix $r \le q$ such that for some $N \in \mathcal X$ with $\gamma$ and $\alpha$ in $N$, $N \in A_r$. By Definition 4.2(6), $N \cap \alpha \in f_r(\alpha)$. So $r$ forces that $\gamma < \sup(N \cap \alpha) \in \dot c_\alpha$. Now we show that $\dot c_\alpha$ is forced to have order type equal to $\omega_1$. Since $\alpha$ has uncountable cofinality and $\dot c_\alpha$ is forced to be cofinal in $\alpha$, clearly $\dot c_\alpha$ is forced to have an order type of uncountable cofinality. If it is not forced to have order type equal to $\omega_1$, then some condition forces that it has a proper initial segment of order type $\omega_1$. Hence, for some $p \in \mathbb{P}$ and $K \in f_p(\alpha)$, $p$ forces that $\dot c_\alpha \cap \sup(K)$ has order type equal to $\omega_1$. Let $G$ be a generic filter on $\mathbb{P}$ which contains $p$, and let $c_\alpha := \dot c_\alpha^G$ and $f := \dot f^G$. By Lemma 5.1, $c_\alpha \cap \sup(K) = \{ \sup(J) : J \in f(K) \}$. Since $c_\alpha \cap \sup(K)$ is uncountable, it follows that $f(K)$ is uncountable. But since $K \in f(\alpha)$, by Lemma 5.1 we have that $f(K) = f(\alpha) \cap Sk(K) \subseteq Sk(K)$. As $Sk(K)$ is countable, so is $f(K)$, and we have a contradiction. \end{proof} \begin{proposition} Let $\alpha \in S$. Suppose that $p \in \mathbb{P}$ and $p$ forces that $\xi < \alpha$ is a limit point of $\dot c_\alpha$. Then there is $q \le p$ such that for some $M \in f_q(\alpha)$, $\sup(M) = \xi$. In particular, $\mathbb{P}$ forces that $\dot c_\alpha$ is closed. \end{proposition} \begin{proof} Note that by Lemma 5.2, $\xi$ must have cofinality $\omega$. Extend $p$ to $q$ so that for some $M \in f_{q}(\alpha)$, $q$ forces that $M$ is the membership least element of $\dot f(\alpha)$ with $\xi \le \sup(M)$. We will prove that $\sup(M) = \xi$, which finishes the proof. \bigskip \noindent \emph{Claim 1:} If $K \in f_q(\alpha)$ and $\sup(K \cap \xi) < \xi$, then $\sup(K) < \xi$. \bigskip Suppose for a contradiction that $K \in f_q(\alpha)$, $\sup(K \cap \xi) < \xi$, but $\xi \le \sup(K)$. Since $\xi$ is forced by $q$ to be a limit point of $\dot c_\alpha$, there exist $t \le q$ and $N$ such that $N \in f_t(\alpha)$ and $$ \sup(K \cap \xi) < \sup(N) < \xi. $$ As $K$ and $N$ are in $f_t(\alpha)$ and $\sup(N) < \xi \le \sup(K)$, it follows that $N \in Sk(K)$ by Lemma 4.4. By elementarity, $\sup(N) \in Sk(K)$. By Lemma 4.3(1), $Sk(K) \cap \kappa = K$. So $\sup(N) \in K$. Thus, $\sup(N) \in K \cap \xi$, contradicting that $\sup(K \cap \xi) < \sup(N)$. \bigskip It easily follows from Claim 1 that $f_q(\alpha)$ is the union of the sets $A_1$ and $A_2$ defined by $$ A_1 := \{ K \in f_q(\alpha) : \sup(K) < \xi \}, $$ $$ A_2 := \{ K \in f_q(\alpha) : \sup(K \cap \xi) = \xi \}. $$ Namely, if $K \in f_q(\alpha)$, then either $\sup(K \cap \xi) < \xi$, in which case $\sup(K) < \xi$ by Claim 1, or else $\sup(K \cap \xi) = \xi$. Note that since $f_q(\alpha)$ is an $\in$-chain, if $K \in A_1$ and $L \in A_2$, then $\sup(K) < \sup(L)$, and therefore $K \in Sk(L)$ by Lemma 4.4. Observe that since $M \in f_q(\alpha)$ and $\xi \le \sup(M)$, we have that $M \in A_2$. So for all $K \in A_1$, $K \in Sk(M)$. Also, since $M$ is the membership least element of $f_q(\alpha)$ with $\xi \le \sup(M)$, we have that for all $N \in A_2$, either $M = N$ or $M \in Sk(N)$. In particular, for all $N \in A_2$, $M \subseteq N$. \bigskip Now we prove the proposition. Assume for a contradiction that $\sup(M) \ne \xi$. Then $M \cap [\xi,\alpha) \ne \emptyset$. Fix $\gamma < \xi$ large enough so that if $J \in A_1$, then $\sup(J) < \gamma$. This is possible since $A_1$ is finite. As $q$ forces that $\xi$ is a limit point of $\dot c_\alpha$, we can fix $s \le q$ and $K$ such that: \begin{enumerate} \item $K$ is the membership largest element of $f_s(\alpha)$ with $\sup(K) < \xi$; \item $K \notin \mathrm{dom}(f_q)$; \item $\gamma < \sup(K)$; \item $K \cap \omega_1$ is different from $L \cap \omega_1$, for all $L \in A_q$. \end{enumerate} By Definition 4.2(2), fix $K_1 \in A_s$ and $\beta \in (K_1 \cap \mathrm{dom}(f_s) \cap S) \cup \{ \kappa \}$ such that $K = K_1 \cap \beta$. We use $s$, $K$, $K_1$, and $\beta$ to define an extension $r$ of $q$. Let $$ A_r := A_q \cup \{ K_1 \}. $$ If $\beta = \kappa$, then let $$ \mathrm{dom}(f_r) := \mathrm{dom}(f_q) \cup \{ K \} \cup \{ K_1 \cap \delta : \delta \in K_1 \cap \mathrm{dom}(f_q) \cap S \}, $$ and if $\beta < \kappa$, then let \begin{multline*} \mathrm{dom}(f_r) := \mathrm{dom}(f_q) \cup \{ K \} \cup \{ K_1 \cap \delta : \delta \in K_1 \cap \mathrm{dom}(f_q) \cap S \} \ \cup \\ \{ \beta \} \cup \{ L \cap \beta : L \in A_q, \ \beta \in L \}. \end{multline*} Note that the domain of $f_r$ is a subset of the domain of $f_s$, since $s \le q$, $K \in f_s(\alpha)$, $K_1 \in A_s$, and $\beta \in (K_1 \cap \mathrm{dom}(f_s) \cap S) \cup \{ \kappa \}$. Thus, it makes sense to define, for each $x \in \mathrm{dom}(f_r)$, $$ f_r(x) := f_s(x) \cap \mathrm{dom}(f_r). $$ Observe that since $K \in f_s(\alpha)$ and $K$ and $\alpha$ are in $\mathrm{dom}(f_r)$, we have that $K \in f_r(\alpha)$. Let $J \in f_r(x)$, and we will define $g_r(J,x)$. We let $g_r(J,x) := g_s(J,x)$, unless $J = K$ and $x \in f_r(\alpha) \cup \{ \alpha \}$, in which case we let $$ g_r(J,x) := g_s(J,x) \cup \{ \zeta \}, $$ where $$ \zeta := \min(M \setminus \xi). $$ The ordinal $\zeta$ exists because we are assuming for a contradiction that $\sup(M) > \xi$. Note that in either case, we have that $g_s(J,x) \subseteq g_r(J,x)$. \bigskip We will prove that $r$ is a condition and $r \le q$. Let us see that this gives us a contradiction. If $r$ is a condition and $r \le q$, then $r$ forces that $\xi$ is a limit point of $\dot c_\alpha$. But $\sup(K) < \xi$, so we can find $u \le r$ and $L \in f_u(\alpha)$ such that $\sup(K) < \sup(L) < \xi$. Then $K$ and $L$ are in $f_u(\alpha)$, and since $\sup(K) < \sup(L)$, $K \in Sk(L)$ by Lemma 4.4. So $K \in f_u(\alpha) \cap Sk(L) = f_u(L)$. Therefore, by Definition 4.2(4,5), $$ g_u(K,\alpha) \subseteq g_u(K,L) \subseteq L. $$ By the definition of $g_r$, $\zeta \in g_r(K,\alpha)$. Hence, $$ \zeta \in g_r(K,\alpha) \subseteq g_u(K,\alpha) \subseteq L. $$ So $\zeta \in L$. This is a contradiction, since $\sup(L) < \xi \le \zeta$. \bigskip Suppose for a moment that $r$ is a condition, and let us prove that $r \le q$. We verify properties (a)--(d) of Definition 4.2. (a,b) By the definition of $r$, $A_q \subseteq A_r$ and $\mathrm{dom}(f_q) \subseteq \mathrm{dom}(f_r)$. If $x \in \mathrm{dom}(f_q)$, then since $s \le q$, $$ f_q(x) \subseteq f_s(x) \cap \mathrm{dom}(f_q) \subseteq f_s(x) \cap \mathrm{dom}(f_r) = f_r(x). $$ (c) Suppose that $(J,x) \in \mathrm{dom}(g_q)$. Then $J \ne K$, since $K \notin \mathrm{dom}(f_q)$. So by the definition of $g_r$, $g_r(J,x) = g_s(J,x)$. Since $s \le q$, we have that $g_q(J,x) \subseteq g_s(J,x) = g_r(J,x)$. (d) Assume that $J$ and $x$ are in $\mathrm{dom}(f_q)$ and $J \in f_r(x)$. Then by the definition of $f_r$, $J \in f_s(x)$. Since $s \le q$, it follows that $J \in f_q(x)$. \bigskip It remains to prove that $r$ is a condition. We verify requirements (1)--(7) of Definition 4.2. \bigskip (1) We have that $A_r = A_q \cup \{ K_1 \} \subseteq A_s$. Since $A_s$ is adequate, so is $A_r$. \bigskip (2) It is obvious that $f_r$ is a function with a finite domain. Let $x \in \mathrm{dom}(f_r)$, and we will show that either $x \in S$, or there is $L \in A_r$ and $\delta \in (L \cap \mathrm{dom}(f_r) \cap S) \cup \{ \kappa \}$ such that $x = L \cap \delta$. If $x \in \mathrm{dom}(f_q)$, then this statement follows from the fact that $q$ is a condition. If $x = K$, then $x = K_1 \cap \beta$, where $K_1 \in A_r$ and $\beta \in (K_1 \cap \mathrm{dom}(f_r) \cap S) \cup \{ \kappa \}$. If $x = K_1 \cap \delta$, where $\delta \in K_1 \cap \mathrm{dom}(f_q) \cap S$, then we are done since $K_1 \in A_r$ and $\mathrm{dom}(f_q) \subseteq \mathrm{dom}(f_r)$. If $\beta = \kappa$, then we have already handled all possibilities for $x$. Suppose that $\beta < \kappa$. Then we also have the possibility that $x = \beta$, in which case $x \in S$, or $x = L \cap \beta$, where $\beta \in L$ and $L \in A_q$. In the second case, $L \in A_r$ and $\beta \in L \cap \mathrm{dom}(f_r) \cap S$, so we are done. Let $x \in \mathrm{dom}(f_r)$, and we will show that $f_r(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$. But by the definition of $f_r$, $f_r(x) \subseteq f_s(x)$. Since $s$ is a condition, $f_s(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$. Hence, $f_r(x)$ is as well. \bigskip (3) Let $x \in \mathrm{dom}(f_r)$. Then $f_r(x) = f_s(x) \cap \mathrm{dom}(f_r)$, and therefore $f_r(x) \subseteq \mathrm{dom}(f_r)$. Let $K \in f_r(x)$, and we will show that $f_r(K) = f_r(x) \cap Sk(K)$. But $f_r(K) = f_s(K) \cap \mathrm{dom}(f_r)$, and since $s$ is a condition, $f_s(K) = f_s(x) \cap Sk(K)$. Thus, \begin{multline*} f_r(K) = f_s(K) \cap \mathrm{dom}(f_r) = (f_s(x) \cap Sk(K)) \cap \mathrm{dom}(f_r) = \\ (f_s(x) \cap \mathrm{dom}(f_r)) \cap Sk(K) = f_r(x) \cap Sk(K). \end{multline*} \bigskip (4) Consider $J \in f_r(x)$, and we will show that $g_r(J,x)$ is a finite subset of $x \setminus \sup(J)$. If $g_r(J,x) = g_s(J,x)$, then since $s$ is a condition, $g_s(J,x)$ is a finite subset of $x \setminus \sup(J)$, and we are done. Suppose that $g_r(J,x) \ne g_s(J,x)$. Then by the definition of $g_r$, $J = K$, $x \in f_r(\alpha) \cup \{ \alpha \}$, and $g_r(J,x) = g_s(K,x) \cup \{ \zeta \}$, where $\zeta = \min(M \setminus \xi)$. Since $s$ is a condition, $g_s(J,x)$ is a finite subset of $x \setminus \sup(J)$. So it suffices to show that $\zeta \in x \setminus \sup(K)$. We already know that $\sup(K) < \xi \le \zeta$, so $\zeta \notin \sup(K)$. It remains to show that $\zeta \in x$. If $x = \alpha$, then certainly $\zeta < \alpha$, since $\zeta \in M$ and $M \subseteq \alpha$. Assume that $x \in f_r(\alpha)$. We consider the different possibilities for why $x$ is in $f_r(\alpha)$. First, assume that $x \in \mathrm{dom}(f_q)$. Then, since $x \in f_r(\alpha) \subseteq f_s(\alpha)$, we have that $x \in f_s(\alpha)$. But $x$ and $\alpha$ are in $\mathrm{dom}(f_q)$. Since $s \le q$, it follows that $x \in f_q(\alpha)$. So either $x \in A_1$ or $x \in A_2$. By the choice of $\gamma$ and $K$, for all $J \in A_1$, $\sup(J) < \gamma < \sup(K)$. Since $K \in f_r(x)$, $K \in Sk(x)$, and so $\sup(K) < \sup(x)$. Therefore, $x \notin A_1$. So $x \in A_2$. As noted above, the minimality of $M$ implies that for all $J \in A_2$, $M \subseteq J$. So $M \subseteq x$. Since $\zeta \in M$, $\zeta \in x$, and we are done. Secondly, assume that $x \notin \mathrm{dom}(f_q)$. Since $K \in f_r(x)$, $K \in Sk(x)$. So $$ K_1 \cap \omega_1 = K \cap \omega_1 < x \cap \omega_1. $$ It follows that $x$ is not equal to $K$, and $x$ is not equal to $K_1 \cap \delta$ for any $\delta \in K_1 \cap \mathrm{dom}(f_q) \cap S$. The remaining possibility is that $\beta < \kappa$ and $x = L \cap \beta$, where $L \in A_q$ and $\beta \in L$. Since $x \in f_r(\alpha) \subseteq f_s(\alpha)$, we have that $x \in f_s(\alpha)$. As $x$ and $M$ are both in $f_s(\alpha)$, they are membership comparable. Since $K \in f_s(x) = f_s(L \cap \beta)$ and $K$ is the membership largest member of $f_s(\alpha)$ with $\sup(K) < \xi$, we have that $\xi \le \sup(L \cap \beta)$. But recall that $q$ forces that $M$ is the membership least element of $\dot f(\alpha)$ with $\xi \le \sup(M)$, and $s \le q$. Hence, it is not the case that $L \cap \beta \in Sk(M)$. So either $L \cap \beta = M$ or $M \in Sk(L \cap \beta)$. In either case, $\zeta \in M \subseteq L \cap \beta$, so $\zeta \in L \cap \beta = x$. \bigskip (5) Suppose that $J \in f_r(L)$ and $L \in f_r(x)$. Then $J \in f_s(L)$ and $L \in f_s(x)$. We will show that $g_r(J,x) \subseteq g_r(J,L)$. If $g_r(J,x) = g_s(J,x)$, then since $s$ is a condition and $g_s(J,L) \subseteq g_r(J,L)$, we have that $$ g_r(J,x) = g_s(J,x) \subseteq g_s(J,L) \subseteq g_r(J,L). $$ Assume that $g_r(J,x) \ne g_s(J,x)$. Then by the definition of $g_r$, we have that $J = K$, $x \in f_r(\alpha) \cup \{ \alpha \}$, and $g_r(K,x) = g_s(K,x) \cup \{ \zeta \}$, where $\zeta = \min(M \setminus \xi)$. Again, $g_s(K,x) \subseteq g_s(K,L) \subseteq g_r(K,L)$. So it suffices to show that $\zeta \in g_r(K,L)$. By the definition of $g_r$, in order to show that $\zeta \in g_r(K,L)$, it is enough to show that $L \in f_r(\alpha) \cup \{ \alpha \}$, for then $g_r(K,L)$ is defined as $g_s(K,L) \cup \{ \zeta \}$. But $L \in f_r(x)$ and $x \in f_r(\alpha) \cup \{ \alpha \}$. So if $x = \alpha$, then $L \in f_r(\alpha)$, and if $x \in f_r(\alpha)$, then $L \in f_r(x) = f_r(\alpha) \cap Sk(x)$ by requirement (3), so $L \in f_r(\alpha)$. \bigskip (6) Suppose that $\delta \in \mathrm{dom}(f_r) \cap S$, $L \in A_r$, and $\delta \in L$. We will show that $L \cap \delta \in f_r(\delta)$. Since $\mathrm{dom}(f_r) \subseteq \mathrm{dom}(f_s)$, we have that $\delta \in \mathrm{dom}(f_s) \cap S$. And $A_r \subseteq A_s$. As $s$ is a condition, it follows that $L \cap \delta \in f_s(\delta)$. Since $f_r(\delta) = f_s(\delta) \cap \mathrm{dom}(f_r)$, it suffices to show that $L \cap \delta \in \mathrm{dom}(f_r)$. By the definition of $\mathrm{dom}(f_r)$, $\delta$ being in $\mathrm{dom}(f_r) \cap S$ implies that either $\delta \in \mathrm{dom}(f_q)$ or $\delta = \beta$. Also, $A_r = A_q \cup \{ K_1 \}$, so $L$ being in $A_r$ means that either $L \in A_q$ or $L = K_1$. If $\delta \in \mathrm{dom}(f_q)$ and $L \in A_q$, then $L \cap \delta \in f_q(\delta) \subseteq \mathrm{dom}(f_q) \subseteq \mathrm{dom}(f_r)$, since $q$ is a condition. If $\delta \in \mathrm{dom}(f_q)$ and $L = K_1$, then $L \cap \delta = K_1 \cap \delta$ is in $\mathrm{dom}(f_r)$ by the definition of $\mathrm{dom}(f_r)$. If $\delta = \beta$ and $L \in A_q$, then $L \cap \delta = L \cap \beta$ is in $\mathrm{dom}(f_r)$ by the definition of $\mathrm{dom}(f_r)$. And if $\delta = \beta$ and $L = K_1$, then $L \cap \delta = K_1 \cap \beta = K$ is in $\mathrm{dom}(f_r)$ by the definition of $\mathrm{dom}(f_r)$. \bigskip (7) Suppose that $\tau \in r^*(A_r) \cap S$, and we will show that $\tau \in \mathrm{dom}(f_r)$. Fix $J$ and $L$ in $A_r$ such that $J \sim L$ and $\tau = \min((J \cap \kappa) \setminus \beta_{J,L})$. Then obviously $J$ and $L$ are different, and since $J \sim L$, $J \cap \omega_1 = L \cap \omega_1$ by Lemma 2.17(2). Moreover, $A_r = A_q \cup \{ K_1 \}$, and by the choice of $K$, $K_1 \cap \omega_1 = K \cap \omega_1$ is different from $N \cap \omega_1$ for all $N \in A_q$. Therefore, $J$ and $L$ must both be in $A_q$. Thus, $\tau \in r^*(A_q) \cap S \subseteq \mathrm{dom}(f_q) \subseteq \mathrm{dom}(f_r)$. \end{proof} \begin{proposition} The forcing poset $\mathbb{P}$ forces that $\langle \dot c_\alpha : \alpha \in S \rangle$ is a partial square sequence. \end{proposition} \begin{proof} By Lemma 5.2 and Proposition 5.3, for each $\alpha \in S$, $\mathbb{P}$ forces that $\dot c_\alpha$ is a club subset of $\alpha$ with order type equal to $\omega_1$. Let $G$ be a generic filter on $\mathbb{P}$. Consider $\alpha$ and $\beta$ in $S$, and let $c_\alpha := \dot c_\alpha^G$ and $c_\beta := \dot c_\beta^G$. Assume that $\xi$ is a common limit point of $c_\alpha$ and $c_\beta$. We will show that $c_\alpha \cap \xi = c_\beta \cap \xi$. Since $c_\alpha$ and $c_\beta$ are closed, it follows that $\xi \in c_\alpha \cap c_\beta$. Thus, there are $K \in f(\alpha)$ and $L \in f(\beta)$ such that $\sup(K) = \xi = \sup(L)$. By Lemma 5.1, $$ c_\alpha \cap \xi = c_\alpha \cap \sup(K) = \{ \sup(J) : J \in f(K) \}, $$ and $$ c_\beta \cap \xi = c_\beta \cap \sup(L) = \{ \sup(J) : J \in f(L) \}. $$ Thus, to show that $c_\alpha \cap \xi = c_\beta \cap \xi$, it suffices to show that $f(K) = f(L)$. We will prove, in fact, that $K = L$. Since $K \in f(\alpha)$ and $L \in f(\beta)$, we can fix $p \in G$ such that $K \in f_p(\alpha)$ and $L \in f_p(\beta)$. Then $K$ and $L$ are in $\mathrm{dom}(f_p)$, by Definition 4.2(3). By Definition 4.2(2), fix $K_1$ in $A_p$ and $\theta \in (K_1 \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $K = K_1 \cap \theta$, and $L_1$ in $A_p$ and $\tau \in (L_1 \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $L = L_1 \cap \tau$. Since $\sup(K_1 \cap \theta) = \sup(K) = \xi$, it follows that $\xi \le \theta$, and similarly, $\xi \le \tau$. As $\theta$ and $\tau$ are in $S \cup \{ \kappa \}$, they have uncountable cofinality. On the other hand, since $\xi$ is the supremum of the countable set $K$, $\xi$ has countable cofinality. Therefore, $\xi < \theta$ and $\xi < \tau$. Since the sets $K = K_1 \cap \theta$ and $L = L_1 \cap \tau$ are closed under successor ordinals by elementarity, $\xi$ is not a member of $K_1$ nor $L_1$. Therefore, $$ K = K_1 \cap \theta = K_1 \cap \xi, $$ and $$ L = L_1 \cap \tau = L_1 \cap \xi. $$ The ordinal $\xi$, which is the supremum of $K$ and $L$, is a common limit point of $K_1 \cap \kappa$ and $L_1 \cap \kappa$. So by Proposition 2.11, $\xi < \beta_{K_1,L_1}$. We claim that $K_1 \sim L_1$. Suppose not, and without loss of generality, assume that $K_1 < L_1$. Then $K_1 \cap \beta_{K_1,L_1}$ is in $L_1$. Since $\xi < \beta_{K_1,L_1}$, and $\xi$ is a limit point of $K = K_1 \cap \theta$, we have that $\xi$ is a limit point of $K_1 \cap \beta_{K_1,L_1}$. But $K_1 \cap \beta_{K_1,L_1}$ is in $L_1$, and therefore $\xi$ is in $L_1$ by elementarity. So $\xi \in L_1 \cap \tau = L$, which is a contradiction. So indeed, $K_1 \sim L_1$. Hence $K_1 \cap \beta_{K_1,L_1} = L_1 \cap \beta_{K_1,L_1}$. Since $\xi < \beta_{K_1,L_1}$, it follows that $K_1 \cap \xi = L_1 \cap \xi$. Thus, $K = K_1 \cap \xi = L_1 \cap \xi = L$. \end{proof} We point out that if $S$ is chosen so that $S = D \cap \mathrm{cof}(> \! \omega)$, for some club set $D \subseteq \kappa$, then adding a partial square sequence on $S$ will imply that $\Box_{\omega_1}$ holds in the generic extension (see the end of \cite{jk23}). So as a special case, our forcing poset provides another way to force $\Box_{\omega_1}$ with finite conditions. \bigskip \addcontentsline{toc}{section}{6. Amalgamation over uncountable models} \textbf{\S 6. Amalgamation over uncountable models} \stepcounter{section} \bigskip We now turn to proving that $\mathbb{P}$ is strongly proper on a stationary set, and hence preserves $\omega_1$, and is $\kappa$-c.c. Strong properness is proven using amalgamation of conditions over countable models, and the $\kappa$-c.c.\ is proven using amalgamation of conditions over uncountable models. The uncountable case is similar to, but not as complicated as, the countable case, so we will handle the uncountable case first. Many of the results which we will prove in Sections 6 and 7 will be used again in Sections 8 and 9, where the approximation property of certain quotients of $\mathbb{P}$ is verified. For this reason, it will be helpful to develop the notation and results of Sections 6 and 7 in great detail. Let us give a brief outline of the main ideas presented in this section. The goal is to show that for any simple model $Q \in \mathcal Y$, the maximum condition in $\mathbb{P}$ is strongly $Q$-generic. This fact will imply that $\mathbb{P}$ is $\kappa$-c.c. Let $D_Q$ denote the set of conditions $s \in \mathbb{P}$ such that for all $M \in A_s$, $M \cap Q \in A_s$. We will show that $D_Q$ is dense in $\mathbb{P}$. For each $s \in D_Q$, we will define a condition $s \restriction Q$ in $Q \cap \mathbb{P}$. This condition will satisfy that for all $w \le s \restriction Q$ in $Q \cap \mathbb{P}$, $w$ and $s$ are compatible. Since $D_Q$ is dense, it will follow that the maximum condition of $\mathbb{P}$ is strongly $Q$-generic. \begin{lemma} Let $q \in \mathbb{P}$ and $Q \in \mathcal Y$. Then there is $s \le q$ such that for all $M \in A_s$, $M \cap Q \in A_s$. Moreover, $A_s = A_q \cup \{ M \cap Q : M \in A_q \}$. \end{lemma} Recall that by Assumption 2.5(2), if $M \in \mathcal X$ and $Q \in \mathcal Y$, then $M \cap Q \in \mathcal X$. \begin{proof} By Proposition 2.27, the set $A_q \cup \{ M \cap Q : M \in A_q \}$ is adequate. Define $$ x_0 := r^*(A_q \cup \{ M \cap Q : M \in A_q \}) \cap S, $$ and define $$ x := x_0 \setminus \mathrm{dom}(f_q). $$ Let $r := q + x$. By Definition 4.7 and Lemma 4.8, $r$ is a condition, $r \le q$, and $A_r = A_q$. Also, $x_0 \subseteq \mathrm{dom}(f_r)$. Define $s$ as follows. Let $f_s := f_r$, $g_s := g_r$, and $$ A_s := A_r \cup \{ M \cap Q : M \in A_r \}. $$ We claim that $s$ is as required. By Proposition 2.27, for all $M \in A_s$, $M \cap Q \in A_s$. It is trivial to check that if $s$ is a condition, then $s \le q$. It remains to show that $s$ is a condition. We verify requirements (1)--(7) of Definition 4.2. (1) follows from Proposition 2.27. Requirements (2)--(5) follow immediately from $r$ being a condition, together with the fact that $f_s = f_r$, $g_s = g_r$, and $A_r \subseteq A_s$. (6) Suppose that $\alpha \in \mathrm{dom}(f_s) \cap S$, $M \in A_s$, and $\alpha \in M$. We will show that $M \cap \alpha \in f_s(\alpha)$. Since $f_r = f_s$, we have that $\alpha \in \mathrm{dom}(f_r) \cap S$. First, assume that $M \in A_r$. Then since $r$ is a condition, $M \cap \alpha \in f_r(\alpha) = f_s(\alpha)$, and we are done. Secondly, assume that $M = M_1 \cap Q$ for some $M_1 \in A_r$. Then $\alpha \in M = M_1 \cap Q$. So $\alpha \in M_1$ and $\alpha < Q \cap \kappa$. Thus, $$ M \cap \alpha = M_1 \cap Q \cap \alpha = M_1 \cap \alpha. $$ Since $M_1 \in A_r$, $\alpha \in \mathrm{dom}(f_r) \cap S$, and $\alpha \in M_1$, it follows that $M \cap \alpha = M_1 \cap \alpha \in f_r(\alpha) = f_s(\alpha)$. (7) We need to show that $r^*(A_s) \cap S \subseteq \mathrm{dom}(f_s)$. But since $A_r = A_q$, we have that $$ r^*(A_s) \cap S = r^*(A_q \cup \{ M \cap Q : M \in A_q \}) \cap S = x_0 \subseteq \mathrm{dom}(f_r) = \mathrm{dom}(f_s). $$ \end{proof} \begin{definition} For each $Q \in \mathcal Y$, let $D_Q$ denote the set of conditions $q \in \mathbb{P}$ such that for all $M \in A_q$, $M \cap Q \in A_q$. \end{definition} \begin{lemma} Let $Q \in \mathcal Y$. Then $D_Q$ is dense in $\mathbb{P}$. \end{lemma} \begin{proof} Immediate from Lemma 6.1. \end{proof} Note that if $q \in Q \cap \mathbb{P}$, then $q \in D_Q$. Namely, for all $M \in A_q$, $M \in Q$, and therefore $M \cap Q = M \in A_q$. \begin{definition} Suppose that $Q \in \mathcal Y$ is simple and $q \in D_Q$. Let $q \restriction Q$ denote the triple $(f,g,A)$ satisfying: \begin{enumerate} \item $\mathrm{dom}(f) := \mathrm{dom}(f_q) \cap Q$, and for all $x \in \mathrm{dom}(f)$, $f(x) := f_q(x)$; \item $\mathrm{dom}(g) := \mathrm{dom}(g_q) \cap Q$, and for all $(K,x) \in \mathrm{dom}(g)$, $g(K,x) := g_q(K,x)$; \item $A := A_q \cap Q$. \end{enumerate} \end{definition} Note that in (1) above, if $x \in \mathrm{dom}(f_q) \cap Q$, then $Sk(x) \subseteq Q$. So $f_q(x) = f(x)$ is a finite subset of $Q$, and therefore is in $Q$. In (2), if $K \in f_q(x)$ and $K$ and $x$ are in $Q$, then $g_q(K,x) \subseteq x \subseteq Q$. So $g_q(K,x)$ is a finite subset of $Q$, and hence is in $Q$. Similarly, $A = A_q \cap Q$ is in $Q$. It easily follows from these observations that $q \restriction Q \in Q$. \begin{lemma} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Then $q \restriction Q$ is in $Q \cap \mathbb{P}$ and $q \le q \restriction Q$. \end{lemma} \begin{proof} Let $q \restriction Q = (f,g,A)$. We already observed that $q \restriction Q \in Q$. It is trivial to check that if $q \restriction Q$ is a condition, then $q \le q \restriction Q$. So it suffices to show that $q \restriction Q$ is a condition. We verify requirements (1)--(7) of Definition 4.2. (1), (5), (6), and (7) are immediate. It remains to prove (2), (3), and (4). \bigskip (2) Obviously $f$ is a function with a finite domain. Let $x \in \mathrm{dom}(f) = \mathrm{dom}(f_q) \cap Q$. Since $f(x) = f_q(x)$, it follows that $f(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$. We claim that either $x \in S$, or there is $M \in A$ and $\alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$. Since $x \in \mathrm{dom}(f_q)$ and $q$ is a condition, we have that either $x \in S$, or there is $M_1 \in A_q$ and $\alpha \in (M_1 \cap \mathrm{dom}(f_q) \cap S) \cup \{ \kappa \}$ such that $x = M_1 \cap \alpha$. If $x \in S$, then we are done, so assume the second case. Since $q \in D_Q$ and $Q$ is simple, $$ M_1 \cap Q \in A_q \cap Q = A. $$ So it suffices to show that $x = M_1 \cap Q \cap \beta$, for some $\beta$ with $$ \beta \in (M_1 \cap Q \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}. $$ We split the proof into the cases of whether $\alpha < Q \cap \kappa$ or $Q \cap \kappa \le \alpha$. First, assume that $\alpha < Q \cap \kappa$. Then $\alpha \in \mathrm{dom}(f_q) \cap S \cap Q = \mathrm{dom}(f) \cap S$, and $$ x = M_1 \cap \alpha = M_1 \cap \alpha \cap Q = M_1 \cap Q \cap \alpha. $$ Hence, $x = M_1 \cap Q \cap \beta$, where $\beta = \alpha \in M_1 \cap Q \cap \mathrm{dom}(f) \cap S$. Secondly, assume that $Q \cap \kappa \le \alpha$. Since $x \in Q$, $x = M_1 \cap \alpha \subseteq Q \cap \kappa$. Therefore, $$ x = M_1 \cap \alpha = (M_1 \cap \alpha) \cap (Q \cap \kappa) = M_1 \cap Q \cap \kappa. $$ Thus, $x = M_1 \cap Q \cap \beta$, where $\beta = \kappa$. \bigskip (3) Let $x \in \mathrm{dom}(f) = \mathrm{dom}(f_q) \cap Q$, and we will show that $f(x) \subseteq \mathrm{dom}(f)$. We have that $f_q(x) \subseteq Sk(x) \subseteq Q$. And since $q$ is a condition, $$ f(x) = f_q(x) \subseteq \mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f). $$ So $f(x) \subseteq \mathrm{dom}(f)$, as required. Now consider $K \in f(x) = f_q(x)$. Then since $q$ is a condition, $$ f(K) = f_q(K) = f_q(x) \cap Sk(K) = f(x) \cap Sk(K). $$ \bigskip (4) We have that $(K,x) \in \mathrm{dom}(g)$ iff $(K,x) \in \mathrm{dom}(g_q) \cap Q$ iff ($K$ and $x$ are in $Q$ and $K \in f_q(x)$) iff $K \in f(x)$. For each $(K,x) \in \mathrm{dom}(g)$, $g(K,x) = g_q(K,x) \subseteq x \setminus \sup(K)$. \end{proof} The next lemma will not be used until Section 8. \begin{lemma} Let $Q \in \mathcal Y$ be simple. \begin{enumerate} \item Suppose that $p \in Q \cap \mathbb{P}$, $q \in D_Q$, and $q \le p$. Then $q \restriction Q \le p$. \item Suppose that $q$ and $r$ are in $D_Q$ and $r \le q \restriction Q$. Then $r \restriction Q \le q \restriction Q$. \item Suppose that $p$ and $q$ are in $D_Q$ and $q \le p$. Then $q \restriction Q \le p \restriction Q$. \end{enumerate} \end{lemma} \begin{proof} (1) We verify properties (a)--(d) of Definition 4.2. (a) Since $p \in Q$, $A_p \subseteq Q$. As $q \le p$, we have that $$ A_p \subseteq A_q \cap Q = A_{q \restriction Q}. $$ (b) Since $p \in Q$, $\mathrm{dom}(f_p) \subseteq Q$. As $q \le p$, we have that $$ \mathrm{dom}(f_p) \subseteq \mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q}). $$ Let $x \in \mathrm{dom}(f_p)$. Then $$ f_p(x) \subseteq f_q(x) = f_{q \restriction Q}(x). $$ (c) Let $(K,x) \in \mathrm{dom}(g_p)$. Since $q \le p$, $$ g_p(K,x) \subseteq g_q(K,x) = g_{q \restriction Q}(K,x). $$ (d) Assume that $K$ and $x$ are in $\mathrm{dom}(f_p)$ and $K \in f_{q \restriction Q}(x)$. Then $K \in f_{q \restriction Q}(x) = f_q(x)$. Since $q \le p$, it follows that $K \in f_p(x)$. \bigskip (2) We know that $q \restriction Q \in Q \cap \mathbb{P}$, $r \in D_Q$, and $r \le q \restriction Q$. By (1), it follows that $r \restriction Q \le q \restriction Q$. \bigskip (3) By Lemma 6.5, we know that $p \le p \restriction Q$. So $q \le p \le p \restriction Q$. Hence, $q \le p \restriction Q$. Thus, $p$ and $q$ are in $D_Q$ and $q \le p \restriction Q$. By (2), it follows that $q \restriction Q \le p \restriction Q$. \end{proof} We will now begin analyzing the situation where $q \in D_Q$ and $w \le q \restriction Q$ is in $Q \cap \mathbb{P}$. \begin{lemma} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Then: \begin{enumerate} \item $A_q \cap Q \subseteq A_w$; \item $\mathrm{dom}(f_q) \cap Q \subseteq \mathrm{dom}(f_w)$, and for all $x \in \mathrm{dom}(f_q) \cap Q$, $f_q(x) \subseteq f_w(x)$; \item $\mathrm{dom}(g_q) \cap Q \subseteq \mathrm{dom}(g_w)$, and for all $(K,x) \in \mathrm{dom}(g_q) \cap Q$, $g_q(K,x) \subseteq g_w(K,x)$. \end{enumerate} \end{lemma} \begin{proof} Immediate from the definition of $q \restriction Q$ and the fact that $w \le q \restriction Q$. \end{proof} As discussed at the beginning of the section, we are going to show that whenever $w \le q \restriction Q$, where $q \in D_Q$ and $w \in Q \cap \mathbb{P}$, then $w$ and $q$ are compatible. We now begin the construction of a specific lower bound of $w$ and $q$, which we will denote by $w \oplus_Q q$. In order to define the amalgam $w \oplus_Q q$, we will need to define the $f$, $g$, and $A$ components of $w \oplus_Q q$. The amalgam of the $A$-components will be $A_w \cup A_q$. We handle the $f$-components next. \begin{definition} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Define $f_w \oplus_Q f_q = f$ as follows. The domain of $f$ is equal to $\mathrm{dom}(f_w) \cup \mathrm{dom}(f_q)$. The values of $f$ are defined by the following cases: \begin{enumerate} \item for all $x \in \mathrm{dom}(f_w)$, $f(x) := f_w(x)$; \item for all $x \in \mathrm{dom}(f_q) \setminus Q$, if $f_q(x) \cap Q = \emptyset$, then $f(x) := f_q(x)$; \item for all $x \in \mathrm{dom}(f_q) \setminus Q$, if $f_q(x) \cap Q \ne \emptyset$, then $f(x) := f_q(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_q(x) \cap Q$. \end{enumerate} \end{definition} It is easy to see that cases 1--3 describe all of the possibilities for a set being in $\mathrm{dom}(f)$, since $\mathrm{dom}(f_q) \cap Q \subseteq \mathrm{dom}(f_w)$ by Lemma 6.7(2). Moreover, cases 1--3 are obviously disjoint. The next three lemmas describe some important properties of $f_w \oplus_Q f_q$. The first two lemmas are easy, but the third is quite involved. \begin{lemma} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Let $f := f_w \oplus_Q f_q$. Then: \begin{enumerate} \item if $x \in \mathrm{dom}(f_w)$, then $f_w(x) = f(x)$; \item if $x \in \mathrm{dom}(f_q)$, then $f_q(x) \subseteq f(x)$. \end{enumerate} \end{lemma} \begin{proof} (1) is by Definition 6.8(1), and (2) follows immediately from Definition 6.8(2,3). \end{proof} \begin{lemma} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Let $f := f_w \oplus_Q f_q$. Then: \begin{enumerate} \item $\mathrm{dom}(f) \cap Q = \mathrm{dom}(f_w)$; \item if $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_w)$, then $K \in f_w(x)$; \item if $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_q)$, then $K \in f_q(x)$. \end{enumerate} \end{lemma} \begin{proof} (1) By Lemma 6.7(2), $\mathrm{dom}(f_q) \cap Q \subseteq \mathrm{dom}(f_w)$. Hence, $$ \mathrm{dom}(f) \cap Q = (\mathrm{dom}(f_w) \cup \mathrm{dom}(f_q)) \cap Q = \mathrm{dom}(f_w), $$ where the last equality follows from the fact that $\mathrm{dom}(f_w) \cap Q = \mathrm{dom}(f_w)$ and $\mathrm{dom}(f_q) \cap Q \subseteq \mathrm{dom}(f_w)$. (2) Suppose that $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_w)$. By Definition 6.8(1), $f(x) = f_w(x)$, so $K \in f_w(x)$. (3) Assume that $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_q)$. We will show that $K \in f_q(x)$. The proof splits into the three cases of Definition 6.8 for how $f(x)$ is defined. In case 2, $f(x) = f_q(x)$, so $K \in f_q(x)$. In case 1, $f(x) = f_w(x)$. So $K \in f_w(x)$. In particular, $K$ and $x$ are in $Q$. So $K$ and $x$ are in $\mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q})$. Since $w \le q \restriction Q$ and $K \in f_w(x)$, it follows that $K \in f_{q \restriction Q}(x) = f_q(x)$. In case 3, $f(x) = f_q(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_q(x) \cap Q$. So either $K \in f_q(x)$, or $K \in f_w(M)$. In the first case we are done, so assume that $K \in f_w(M)$. Then $K$ and $M$ are in $\mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q})$. Since $w \le q \restriction Q$ and $K \in f_w(M)$, it follows that $K \in f_{q \restriction Q}(M) = f_q(M)$. So $K \in f_q(M)$ and $M \in f_q(x)$. Therefore, $K \in f_q(x)$. \end{proof} The next lemma will be used to verify that $f_w \oplus_Q f_q$ satisfies requirements (2) and (3) of Definition 4.2 for $w \oplus_Q q$. \begin{lemma} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Let $f := f_w \oplus_Q f_q$. Then: \begin{enumerate} \item $f$ is a function with a finite domain, and for all $x \in \mathrm{dom}(f)$, either $x \in S$, or there is $M \in A_w \cup A_q$ and $$ \alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \} $$ such that $x = M \cap \alpha$; moreover, for all $x \in \mathrm{dom}(f)$, $f(x)$ is a finite $\in$-chain and $f(x) \subseteq Sk(x) \setminus S$; \item if $x \in \mathrm{dom}(f)$, then $f(x) \subseteq \mathrm{dom}(f)$, and for all $K \in f(x)$, $f(K) = f(x) \cap Sk(K)$. \end{enumerate} \end{lemma} \begin{proof} (1) The domain of $f$ is equal to $\mathrm{dom}(f_w) \cup \mathrm{dom}(f_q)$, which is finite. Let $x \in \mathrm{dom}(f)$. Then either $x \in \mathrm{dom}(f_w)$ or $x \in \mathrm{dom}(f_q)$. If $x \in \mathrm{dom}(f_w)$, then either $x \in S$, or there is $M \in A_w$ and $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$. If $x \in \mathrm{dom}(f_q)$, then either $x \in S$, or there is $M \in A_q$ and $\alpha \in (M \cap \mathrm{dom}(f_q) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$. In either case, either $x \in S$, or there is $M \in A_w \cup A_q$ and $\alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$. Let $x \in \mathrm{dom}(f)$, and we will show that $f(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$. We consider the three cases in the definition of $f(x)$ given in Definition 6.8. In cases 1 and 2, $f(x)$ is equal to either $f_w(x)$ or $f_q(x)$. Since $w$ and $q$ are conditions, then in either case, $f(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$. Consider case 3, which says that $x \in \mathrm{dom}(f_q) \setminus Q$ and $f(x) = f_q(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_q(x) \cap Q$. Since $w$ and $q$ are conditions, it follows that $$ f(x) \subseteq (Sk(x) \cup Sk(M)) \setminus S. $$ But $M \in f_q(x)$ implies that $M \in Sk(x)$, and therefore $Sk(M) \subseteq Sk(x)$. Hence, $$ f(x) \subseteq Sk(x) \setminus S. $$ Since $w$ and $q$ are conditions, $f_q(x)$ and $f_w(M)$ are each finite $\in$-chains. So to prove that $f(x)$ is a finite $\in$-chain, it suffices to show that whenever $K \in f_w(M)$ and $L \in f_q(x) \setminus f_w(M)$, then $K \in Sk(L)$. If $L = M$, then since $K \in f_w(M)$, $K \in Sk(M) = Sk(L)$, and we are done. Suppose that $L \ne M$. As $L$ and $M$ are different elements of $f_q(x)$, either $L \in f_q(M)$ or $M \in f_q(L)$. But $f_q(M) \subseteq f_w(M)$ by Lemma 6.7(2), and we assumed that $L \notin f_w(M)$. Thence, $M \in f_q(L)$. But $K \in f_w(M)$ implies that $K \in Sk(M)$, and $M \in f_q(L)$ implies that $M \in Sk(L)$. Therefore, $K \in Sk(M) \subseteq Sk(L)$, so $K \in Sk(L)$. \bigskip (2) Let $x \in \mathrm{dom}(f)$. We claim that $f(x) \subseteq \mathrm{dom}(f)$. In cases 1 and 2 of Definition 6.8, either $f(x) = f_w(x)$ or $f(x) = f_q(x)$. Since $w$ and $q$ are conditions, $f(x) \subseteq \mathrm{dom}(f_w) \subseteq \mathrm{dom}(f)$ in the first case, and $f(x) \subseteq \mathrm{dom}(f_q) \subseteq \mathrm{dom}(f)$ in the second case. In the third case, $f(x) = f_q(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_q(x) \cap Q$. Since $w$ and $q$ are conditions, $f(x) \subseteq \mathrm{dom}(f_q) \cup \mathrm{dom}(f_w) = \mathrm{dom}(f)$. \bigskip Assume that $K \in f(x)$, and we will show that $f(K) = f(x) \cap Sk(K)$. We split the proof into the three cases of Definition 6.8 for the definition of $f(x)$. In case 1, $x \in \mathrm{dom}(f_w)$ and $f(x) = f_w(x)$. So $K \in f_w(x) \subseteq \mathrm{dom}(f_w)$. Hence, $f(K) = f_w(K)$. Since $w$ is a condition, $$ f(K) = f_w(K) = f_w(x) \cap Sk(K) = f(x) \cap Sk(K). $$ In case 2, $x \in \mathrm{dom}(f_q) \setminus Q$, $f_q(x) \cap Q = \emptyset$, and $f(x) = f_q(x)$. Since $K \in f(x) = f_q(x)$, we have that $f_q(K) = f_q(x) \cap Sk(K)$, since $q$ is a condition. In particular, $K \in \mathrm{dom}(f_q) \setminus Q$ and $f_q(K) \cap Q = \emptyset$. Therefore, by definition, $f(K) = f_q(K)$. So $$ f(K) = f_q(K) = f_q(x) \cap Sk(K) = f(x) \cap Sk(K). $$ In case 3, $x \in \mathrm{dom}(f_q) \setminus Q$ and $f(x) = f_q(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_q(x) \cap Q$. Then either $K \in f_q(x)$, or $K \in f_w(M)$. Since $M$ is the largest element of $f_q(x) \cap Q$ and $q$ is a condition, $$ f_q(x) \cap Q = (f_q(x) \cap Sk(M)) \cup \{ M \} = f_q(M) \cup \{ M \} \subseteq f_w(M) \cup \{ M \}, $$ where the inclusion holds by Lemma 6.7(2). So $f_q(x) \cap Q \subseteq f_w(M) \cup \{ M \}$. It easily follows that either $K \in f_q(x) \setminus Q$, $K = M$, or $K \in f_w(M)$. First, assume that $K \in f_q(x) \setminus Q$. Then, since $f_q(K) = f_q(x) \cap Sk(K)$, $M$ is the membership largest element of $f_q(K) \cap Q$. So by definition, $f(K) = f_q(K) \cup f_w(M)$. Since $f_w(M) \subseteq Sk(M) \subseteq Sk(K)$, we have that \begin{multline*} f(K) = f_q(K) \cup f_w(M) = (f_q(x) \cap Sk(K)) \cup f_w(M) = \\ (f_q(x) \cup f_w(M)) \cap Sk(K) = f(x) \cap Sk(K). \end{multline*} Secondly, assume that $K = M$. Then $f(K) = f(M) = f_w(M)$. Hence, it suffices to show that $$ f_w(M) = (f_q(x) \cup f_w(M)) \cap Sk(M). $$ The forward inclusion is immediate. For the reverse inclusion, let $J \in (f_q(x) \cup f_w(M)) \cap Sk(M)$, and we will show that $J \in f_w(M)$. So either $J \in f_q(x) \cap Sk(M)$ or $J \in f_w(M) \cap Sk(M)$. In the latter case, we are done. In the former case, by Lemma 6.7(2) we have that $$ J \in f_q(x) \cap Sk(M) = f_q(M) \subseteq f_w(M), $$ so $J \in f_w(M)$. Thirdly, assume that $K \in f_w(M)$. Then $K \in \mathrm{dom}(f_w)$, so $f(K) = f_w(K)$. Since $w$ is a condition, $f_w(K) = f_w(M) \cap Sk(K)$. But by the case in the previous paragraph, $f_w(M) = f(M) = f(x) \cap Sk(M)$. And since $K \in Sk(M)$, $Sk(K) \subseteq Sk(M)$. Therefore, $$ f(K) = f_w(K) = f_w(M) \cap Sk(K) = (f(x) \cap Sk(M)) \cap Sk(K) = f(x) \cap Sk(K). $$ \end{proof} We now handle the amalgamation of the $g$-components of $w$ and $q$. \begin{definition} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Let $f := f_w \oplus_Q f_q$. Define $g_w \oplus_Q g_q$ as the function $g$ with domain equal to the set of pairs $(K,x)$ such that $K \in f(x)$, such that for all $(K,x) \in \mathrm{dom}(g)$, $$ g(K,x) := \bigcup \{ g_w(K,y) \cup g_q(K,y) : x = y, \ \textrm{or} \ x \in f(y) \}.\footnote{When working with the $g$-components of a condition, we will adopt the convention that when $(K,x)$ is not a member of the domain of $g$, then $g(K,x)$ will denote the empty set. In particular, when verifying an inclusion of the form $g(K,x) \subseteq y$, the inclusion is trivial in the case that $(K,x)$ is not in $\mathrm{dom}(g)$.} $$ \end{definition} Note that $g(K,x)$ is finite. The next lemma will be used to show that $g_w \oplus_Q g_q$ satisfies requirement (4) of Definition 4.2 for $w \oplus_Q q$. \begin{lemma} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Let $g := g_w \oplus_Q g_q$. Then for any $(K,x) \in \mathrm{dom}(g)$, $$ g(K,x) \subseteq x \setminus \sup(K). $$ \end{lemma} \begin{proof} Let $(K,x) \in \mathrm{dom}(g)$, which means that $K \in f(x)$. By definition, any ordinal in $g(K,x)$ is either in $g_w(K,x) \cup g_q(K,x)$, or in $g_w(K,y) \cup g_q(K,y)$ for some $y$ with $x \in f(y)$. In the first case, the ordinal is in $x \setminus \sup(K)$, since $w$ and $q$ are conditions. In the second case, the ordinal is not less than $\sup(K)$, because $w$ and $q$ are conditions. Thus, it suffices to show that whenever $x \in f(y)$, then $$ g_w(K,y) \cup g_q(K,y) \subseteq x. $$ We split the proof into the three cases of Definition 6.8 for how $f(y)$ is defined. Note that by Lemma 6.11(2), $K \in f(x)$ and $x \in f(y)$ implies that $K \in f(y)$. \bigskip (1) Suppose that $y \in \mathrm{dom}(f_w)$, so that $f(y) = f_w(y)$. Then $x \in f(y) = f_w(y) \subseteq \mathrm{dom}(f_w)$. So $x \in \mathrm{dom}(f_w)$, and therefore $f(x) = f_w(x)$ Hence, $K \in f_w(x)$ and $x \in f_w(y)$. By Definition 4.2(4,5), $$ g_w(K,y) \subseteq g_w(K,x) \subseteq x. $$ Now assume that $(K,y) \in \mathrm{dom}(g_q)$, and we will show that $g_q(K,y) \subseteq x$. Then $(K,y) \in \mathrm{dom}(g_q) \cap Q$, so by Lemma 6.7(3), $g_q(K,y) \subseteq g_w(K,y)$. But we just proved that $g_w(K,y) \subseteq x$. So $g_q(K,y) \subseteq x$. \bigskip (2) Suppose that $y \in \mathrm{dom}(f_q) \setminus Q$ and $f_q(y) \cap Q = \emptyset$. Then $f(y) = f_q(y)$, so $x \in f_q(y)$. And since $K$ and $x$ are in $f(y) = f_q(y)$, $K$ and $x$ are not in $Q$. Since $f_q(x) = f_q(y) \cap Sk(x)$, we have that $f_q(x) \cap Q = \emptyset$. Therefore, by definition, $f(x) = f_q(x)$. Hence, $K \in f_q(x)$ and $x \in f_q(y)$. By Definition 4.2(4,5), it follows that $$ g_q(K,y) \subseteq g_q(K,x) \subseteq x. $$ On the other hand, since $y$ is not in $Q$, $(K,y)$ is not in the domain of $g_w$. So the inclusion $g_w(K,y) \subseteq x$ is trivial. \bigskip (3) Suppose that $y \in \mathrm{dom}(f_q) \setminus Q$, and $f(y) = f_q(y) \cup f_w(M)$, where $M$ is the membership largest element of $f_q(y) \cap Q$. Since $y \notin Q$, $(K,y)$ is not in $\mathrm{dom}(g_w)$, and therefore $g_w(K,y) = \emptyset$, which is a subset of $x$. It remains to show that $g_q(K,y) \subseteq x$. This is trivial if $(K,y) \notin \mathrm{dom}(g_q)$, so assume that $(K,y) \in \mathrm{dom}(g_q)$, which means that $K \in f_q(y)$. Since $f(y) = f_q(y) \cup f_w(M)$, either $x \in f_q(y)$, or $x \in f_w(M)$. First, assume that $x \in f_q(y)$. Then $K \in f_q(y)$, $x \in f_q(y)$, and $K \in Sk(x)$, which implies that $K \in f_q(x)$, since $q$ is a condition. Therefore, $$ g_q(K,y) \subseteq g_q(K,x) \subseteq x, $$ since $q$ is a condition. Secondly, assume that $x \in f_w(M)$. So $x \in Q$, and since $K \in Sk(x)$, it follows that $K \in Q$ as well. Now $x \in f_w(M)$ implies that $x$, and hence $K$, are in $Sk(M)$. Also, $K$ and $M$ are both in $f_q(y)$. So $K \in f_q(M)$. By Lemma 6.7(2), $f_q(M) \subseteq f_w(M)$, so $K \in f_w(M)$. Since $K$ and $M$ are in $f_q(y) \cap Q$ and $K \in f_q(M)$, $$ g_q(K,y) \subseteq g_q(K,M) \subseteq g_w(K,M), $$ where the last inclusion holds by Lemma 6.7(3). Also, $K \in f_w(M)$, $x \in f_w(M)$, and $K \in Sk(x)$ imply that $K \in f_w(x)$, since $w$ is a condition. So $$ g_w(K,M) \subseteq g_w(K,x) \subseteq x, $$ since $w$ is a condition. Thence, $g_q(K,y) \subseteq g_w(K,M) \subseteq x$. \end{proof} We are ready to define the amalgam $w \oplus_Q q$. \begin{definition} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Let $w \oplus_Q q$ be the triple $(f,g,A)$ defined by: \begin{enumerate} \item $f := f_w \oplus_Q f_q$; \item $g := g_w \oplus_Q g_q$; \item $A := A_w \cup A_q$. \end{enumerate} \end{definition} We will now show that $w \oplus_Q q$ is a condition below $w$ and $q$. We have done most of the work of the proof already. \begin{proposition} Let $Q \in \mathcal Y$ be simple and $q \in D_Q$. Suppose that $w \in Q \cap \mathbb{P}$ and $w \le q \restriction Q$. Then $w$ and $q$ are compatible. In fact, $w \oplus_Q q$ is in $\mathbb{P}$ and $w \oplus_Q q \le w, q$. \end{proposition} \begin{proof} We will prove that $w \oplus_Q q$ is a condition and that $w \oplus_Q q$ is below $w$ and $q$. Let $w \oplus_Q q = (f,g,A)$. To show that $w \oplus_Q q$ is a condition, we verify requirements (1)--(7) of Definition 4.2. \bigskip (1) We apply Proposition 2.28. Since $q \in D_Q$, we have that for all $M \in A_q$, $M \cap Q \in A_q$. Also, $A_w$ is adequate, and by Lemma 6.7(1), $$ A_q \cap Q \subseteq A_w \subseteq Q. $$ By Proposition 2.28, $A_w \cup A_q = A$ is adequate. \bigskip (2,3) These statements are immediate from Lemma 6.11. \bigskip (4) By Definition 6.12, $g$ is a function whose domain is the set of pairs $(K,x)$ such that $K \in f(x)$. And by Lemma 6.13, for all $(K,x) \in \mathrm{dom}(g)$, $g(K,x) \subseteq x \setminus \sup(K)$. Also $g(K,x)$ is finite, by Definition 6.12. \bigskip (5) Let $K \in f(L)$ and $L \in f(x)$, and we will show that $g(K,x) \subseteq g(K,L)$. Let $\xi \in g(K,x)$. Then by Definition 6.12, either $\xi \in g_w(K,x) \cup g_q(K,x)$, or for some $y$ with $x \in f(y)$, $\xi \in g_w(K,y) \cup g_q(K,y)$. In the second case, $L \in f(x)$ and $x \in f(y)$ imply by requirement (3) that $L \in f(y)$. So letting $z := x$ in the first case, and $z := y$ in the second case, we have that $L \in f(z)$ and $\xi \in g_w(K,z) \cup g_q(K,z)$. By Definition 6.12, it follows that $\xi \in g(K,L)$. \bigskip (6) Let $\alpha \in \mathrm{dom}(f) \cap S$, $M \in A$, and suppose that $\alpha \in M$. We will show that $M \cap \alpha \in f(\alpha)$. Since $\mathrm{dom}(f) = \mathrm{dom}(f_w) \cup \mathrm{dom}(f_q)$, either $\alpha \in \mathrm{dom}(f_w)$ or $\alpha \in \mathrm{dom}(f_q)$. As $A = A_w \cup A_q$, either $M \in A_w$ or $M \in A_q$. First, assume that $\alpha \in \mathrm{dom}(f_w)$. Then $f(\alpha) = f_w(\alpha)$ by Definition 6.8. If $M \in A_w$, then $M \cap \alpha \in f_w(\alpha)$, since $w$ is a condition. But $f(\alpha) = f_w(\alpha)$, so $M \cap \alpha \in f(\alpha)$. Suppose that $M \in A_q$. Since $q \in D_Q$ and $Q$ is simple, $M \cap Q \in A_q \cap Q \subseteq A_w$, by Lemma 6.7(1). So $M \cap Q \in A_w$. As $\alpha \in \mathrm{dom}(f_w)$, $\alpha \in Q$. Thus, $\alpha \in (M \cap Q) \cap \mathrm{dom}(f_w)$. Since $w$ is a condition, it follows that $(M \cap Q) \cap \alpha \in f_w(\alpha) = f(\alpha)$. But $\alpha \in Q$ implies that $M \cap Q \cap \alpha = M \cap \alpha$. So $M \cap \alpha \in f(\alpha)$, as required. Secondly, assume that $\alpha \in \mathrm{dom}(f_q) \setminus \mathrm{dom}(f_w)$. Then $\alpha \notin Q$, for otherwise $\alpha \in \mathrm{dom}(f_q) \cap Q \subseteq \mathrm{dom}(f_w)$ by Lemma 6.7(2). Note that this implies that $M \notin A_w$. For otherwise, $\alpha \in M \in A_w \subseteq Q$, which implies that $\alpha \in Q$. So $M \in A_q$. Since $q$ is a condition, $M \cap \alpha \in f_q(\alpha)$. But $f_q(\alpha) \subseteq f(\alpha)$, by cases 2 and 3 of Definition 6.8. So $M \cap \alpha \in f(\alpha)$. \bigskip (7) As in (1) above, the assumptions of Proposition 3.8 hold for $A_q$ and $A_w$. Therefore, $$ r^*(A) = r^*(A_w \cup A_q) = r^*(A_w) \cup r^*(A_q). $$ As $w$ and $q$ are conditions, $$ r^*(A_w) \cap S \subseteq \mathrm{dom}(f_w), \ \ \ r^*(A_q) \cap S \subseteq \mathrm{dom}(f_q). $$ But $\mathrm{dom}(f_w) \subseteq \mathrm{dom}(f)$ and $\mathrm{dom}(f_q) \subseteq \mathrm{dom}(f)$. Hence, $$ r^*(A) \cap S = (r^*(A_w) \cap S) \cup (r^*(A_q) \cap S) \subseteq \mathrm{dom}(f). $$ \bigskip This completes the proof that $w \oplus_Q q$ is a condition. Now we show that $w \oplus_Q q \le w, q$. First, we prove that $w \oplus_Q q \le w$ by verifying properties (a)--(d) of Definition 4.2 for $w$. (a) Since $A = A_w \cup A_q$, clearly $A_w \subseteq A$. (b) follows from Lemma 6.9(1), (c) is immediate from Definition 6.12, and (d) was proved in Lemma 6.10(2). Secondly, we prove that $w \oplus_Q q \le q$ by verifying properties (a)--(d) of Definition 4.2 for $q$. (a) Since $A = A_w \cup A_q$, clearly $A_q \subseteq A$. (b) was proved in Lemma 6.9(2), (c) is immediate from Definition 6.12, and (d) was proved in Lemma 6.10(3). \end{proof} \begin{corollary} The forcing poset $\mathbb{P}$ is $\kappa$-c.c. \end{corollary} \begin{proof} Let $A$ be an antichain of $\mathbb{P}$, and suppose for a contradiction that $A$ has size at least $\kappa$. Without loss of generality, assume that $A$ is maximal. By Assumption 2.23, there are stationarily many simple models in $\mathcal Y$, so we can fix a simple model $Q \in \mathcal Y$ such that $Q \prec (H(\lambda),\in,\mathbb{P},A)$. As $A$ has size at least $\kappa$ and $|Q| < \kappa$, we can fix $s \in A \setminus Q$. By Lemma 6.3, fix $q \le s$ such that $q \in D_Q$. Then $q \restriction Q$ is a condition in $Q \cap \mathbb{P}$. By the elementarity of $Q$ and the maximality of $A$, there is $t \in A \cap Q$ which is compatible with $q \restriction Q$. By elementarity, fix $w \in Q \cap \mathbb{P}$ such that $w \le q \restriction Q, t$. By Proposition 6.15, $w$ and $q$ are compatible, so fix $v \le w, q$. Then $v \le w \le t$, and $v \le q \le s$. Hence, $s$ and $t$ are compatible. But $s$ and $t$ are in $A$ and $A$ is an antichain. Therefore, $s = t$. This is impossible, since $t \in Q$ and $s \notin Q$. \end{proof} \bigskip \addcontentsline{toc}{section}{7. Amalgamation over countable models} \textbf{\S 7. Amalgamation over countable models} \stepcounter{section} \bigskip In this section we will prove that the forcing poset $\mathbb{P}$ is strongly proper on a stationary set. We will show that for any simple model $N \in \mathcal X$, for any $p \in N \cap \mathbb{P}$, there is $q \le p$ which is strongly $N$-generic. For each $p \in N \cap \mathbb{P}$, we will show that there is $q \le p$ such that $N \in A_q$. We will argue that $q$ is strongly $N$-generic as follows. We will define a set $D_N$ which satisfies, among other things, that for all $r \in D_N$, if $M \in A_r$ and $M < N$, then $M \cap N \in A_r$. The set $D_N$ will be dense below $q$. For each $r \in D_N$, we will define a condition $r \restriction N$ in $N$ satisfying that for all $w \in N \cap \mathbb{P}$, if $w \le r \restriction N$, then $w$ and $r$ are compatible. The arguments given in this section are very similar to those in the previous section. However, since $N \cap \kappa$ is a countable set, rather than an ordinal as in the uncountable case, the arguments given in this section are more complicated. \bigskip The first thing we will prove is that any condition $p \in N$ can be extended to a condition containing $N$. \begin{lemma} Let $p \in \mathbb{P}$, $N \in \mathcal X$, and suppose that $p \in N$. Then there is $q \le p$ such that $N \in A_q$. \end{lemma} \begin{proof} Define $q$ as follows. Let $A_q := A_p \cup \{ N \}$. Define $f_q$ by letting $$ \mathrm{dom}(f_q) := \mathrm{dom}(f_p) \cup \{ N \cap \alpha : \alpha \in \mathrm{dom}(f_p) \cap S \}. $$ For each $\alpha \in \mathrm{dom}(f_p) \cap S$, define $$ f_q(\alpha) := f_p(\alpha) \cup \{ N \cap \alpha \}. $$ For each $M \in \mathrm{dom}(f_p) \setminus S$, define $f_q(M) := f_p(M)$. Finally, for each $\alpha \in \mathrm{dom}(f_p) \cap S$, define $$ f_q(N \cap \alpha) := f_p(\alpha). $$ For $K \in f_p(x)$, define $g_q(K,x) := g_p(K,x)$. Let $\alpha \in \mathrm{dom}(f_p) \cap S$. For $K \in f_p(\alpha)$, let $g_q(K,N \cap \alpha) := g_p(K,\alpha)$ and $g_q(N \cap \alpha,\alpha) := \emptyset$. \bigskip It is easy to verify that if $q$ is a condition, then $q \le p$. Also, $N \in A_q$ by definition. It remains to prove that $q$ is a condition. We verify requirements (1)--(7) of Definition 4.2. For (1), $A_q$ is adequate by Lemma 2.16. (4), (5), and (6) are easy. It remains to prove (2), (3), and (7). \bigskip (2) Clearly $f_q$ is a function, and every member of $\mathrm{dom}(f_q)$ is of the required form. Let $x \in \mathrm{dom}(f_q)$, and we will show that $f_q(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$. If $x \in \mathrm{dom}(f_p) \setminus S$, then $f_q(x) = f_p(x)$, so we are done since $p$ is a condition. Suppose that $x = \alpha \in \mathrm{dom}(f_p) \cap S$. Then $f_q(\alpha) = f_p(\alpha) \cup \{ N \cap \alpha \}$. Since $p$ is a condition, $f_p(\alpha)$ is a finite $\in$-chain and a subset of $Sk(\alpha) \setminus S$. So it suffices to show that $f_p(\alpha) \subseteq Sk(N \cap \alpha)$ and $N \cap \alpha \in Sk(\alpha)$. By Assumption 2.6, $N \cap \alpha \in Sk(\alpha)$. Let $K \in f_p(\alpha)$, and we will show that $K \in Sk(N \cap \alpha)$. By Definition 4.2(2), fix $K_1 \in A_p$ and $\beta \in (K_1 \cap \mathrm{dom}(f_p) \cap S) \cup \{ \kappa \}$ such that $K = K_1 \cap \beta$. Since $p \in N$, we have that $K_1$ and $\alpha$ are in $N$. Therefore, $K_1 \cap \alpha \in N$. By Lemma 2.8(1), $K_1 \cap \alpha \in Sk(N \cap \alpha)$. Note that $K = K_1 \cap \beta$ is an initial segment of $K_1 \cap \alpha$ if $\beta \le \alpha$, and since $K \subseteq \alpha$, $K = K_1 \cap \alpha$ if $\alpha < \beta$. In either case, $K$ is an initial segment of $K_1 \cap \alpha$. Since $K_1 \cap \alpha$ is in $Sk(N \cap \alpha)$, by elementarity so is $K$. Finally, assume that $x = N \cap \alpha$, where $\alpha \in \mathrm{dom}(f_p) \cap S$. Then $f_q(x) = f_p(\alpha)$. We just showed that $f_p(\alpha)$ is a subset of $Sk(N \cap \alpha)$, and since $p$ is a condition, it is a finite $\in$-chain disjoint from $S$. \bigskip (3) Let $x \in \mathrm{dom}(f_q)$. It is easy to check by cases that $f_q(x) \subseteq \mathrm{dom}(f_q)$. Let $K \in f_q(x)$, and we will show that $f_q(K) = f_q(x) \cap Sk(K)$. First, assume that $x \in \mathrm{dom}(f_p) \setminus S$. Then $f_q(x) = f_p(x)$, so $K \in f_p(x) \subseteq \mathrm{dom}(f_p) \setminus S$. So by definition, $f_q(K) = f_p(K)$. Therefore, $$ f_q(K) = f_p(K) = f_p(x) \cap Sk(K) = f_q(x) \cap Sk(K). $$ Secondly, assume that $x = \alpha \in \mathrm{dom}(f_p) \cap S$. Then $f_q(\alpha) = f_p(\alpha) \cup \{ N \cap \alpha \}$. If $K \in f_p(\alpha)$, then $K \in \mathrm{dom}(f_p) \setminus S$, so by definition, $f_q(K) = f_p(K)$. Thence, $$ f_q(K) = f_p(K) = f_p(\alpha) \cap Sk(K) = f_q(\alpha) \cap Sk(K), $$ where the last equality follows from the fact that $N \cap \alpha \notin Sk(K)$. If $K = N \cap \alpha$, then $$ f_q(K) = f_q(N \cap \alpha) = f_p(\alpha) = f_q(\alpha) \cap Sk(N \cap \alpha), $$ where the last equality follows from the fact that $f_p(\alpha) \subseteq Sk(N \cap \alpha)$, as shown above. Thirdly, assume that $x = N \cap \alpha$. Then $f_q(N \cap \alpha) = f_p(\alpha)$. Hence, $K \in f_p(\alpha) \setminus S$. So by definition, $f_q(K) = f_p(K)$. Thus, $$ f_q(K) = f_p(K) = f_p(\alpha) \cap Sk(K) = f_q(N \cap \alpha) \cap Sk(K). $$ \bigskip (7) Note that $r^*(A_p \cup \{ N \}) = r^*(A_p)$. Namely, if $\gamma \in r^*(A_p \cup \{ N \})$, then $\gamma \in r^*(\{ K, M \})$ for some distinct $K$ and $M$ in $A_p \cup \{ N \}$ such that $K \sim M$. But for all $K \in A_p$, $K \not \sim N$. Hence, $K$ and $M$ are in $A_p$, and $\gamma \in r^*(A_p)$. It follows that $$ r^*(A_q) \cap S = r^*(A_p \cup \{ N \}) \cap S = r^*(A_p) \cap S \subseteq \mathrm{dom}(f_p) \subseteq \mathrm{dom}(f_q). $$ \end{proof} The next lemma will be used to show that the set $D_N$, which we will define shortly, is dense below any condition which contains $N$. \begin{lemma} Let $q \in \mathbb{P}$ and let $N \in A_q$. Then there is $s \le q$ such that for all $M \in A_s$, if $M < N$ then $M \cap N \in A_s$. \end{lemma} Recall that if $M < N$ are in $\mathcal X$, then $\{ M, N \}$ is adequate, and therefore $M \cap N \in \mathcal X$ by Assumption 2.19. \begin{proof} By Proposition 2.24, the set $A_q \cup \{ M \cap N : M \in A_q, \ M < N \}$ is adequate. Define $$ x_0 := r^*(A_q \cup \{ M \cap N : M \in A_q, \ M < N \}) \cap S, $$ and define $$ x := x_0 \setminus \mathrm{dom}(f_q). $$ Let $r := q + x$. By Lemma 4.8, $r$ is a condition and $r \le q$. By Definition 4.7, $A_r = A_q$, and easily, $x_0 \subseteq \mathrm{dom}(f_r)$. Define $s$ as follows. Let $f_s := f_r$, $g_s := g_r$, and $$ A_s := A_r \cup \{ M \cap N : M \in A_r, \ M < N \}. $$ We claim that $s$ is as required. By Proposition 2.24, for all $M \in A_s$, if $M < N$ then $M \cap N \in A_s$. It is trivial to check that if $s$ is a condition, then $s \le r$, and therefore $s \le q$. It remains to prove that $s$ is a condition. We verify requirements (1)--(7) of Definition 4.2. (1) follows from Proposition 2.24. (2)--(5) follow immediately from $r$ being a condition, together with the fact that $f_s = f_r$, $g_s = g_r$, and $A_r \subseteq A_s$. (6) Suppose that $\alpha \in \mathrm{dom}(f_s) \cap S$, $M \in A_s$, and $\alpha \in M$. We will show that $M \cap \alpha \in f_s(\alpha)$. Since $f_s = f_r$, we have that $\alpha \in \mathrm{dom}(f_r) \cap S$. So if $M \in A_r$, then $M \cap \alpha \in f_r(\alpha) = f_s(\alpha)$, since $r$ is a condition. Assume that $M \in A_s \setminus A_r$, which means that $M = M_1 \cap N$ for some $M_1 \in A_r$ with $M_1 < N$. Then $\alpha \in M \cap \kappa = M_1 \cap N \cap \kappa$. By Proposition 2.11, it follows that $\alpha < \beta_{M_1,N}$. Since $M_1 < N$, $M_1 \cap \beta_{M_1,N} \in N$, so $M_1 \cap \alpha \in N$. In particular, $M_1 \cap \alpha \subseteq N$. So $$ M \cap \alpha = M_1 \cap N \cap \alpha = M_1 \cap \alpha. $$ But $M_1 \in A_r$ and $\alpha \in M_1$. Since $r$ is a condition, $$ M \cap \alpha = M_1 \cap \alpha \in f_r(\alpha) = f_s(\alpha). $$ (7) We need to show that $r^*(A_s) \cap S \subseteq \mathrm{dom}(f_s)$. Since $f_s = f_r$, it suffices to show that $r^*(A_s) \cap S \subseteq \mathrm{dom}(f_r)$. But by the definition of $A_s$ and since $A_r = A_q$, we have that $$ r^*(A_s) \cap S = r^*(A_q \cup \{ M \cap N : M \in A_q, \ M < N \}) \cap S = x_0, $$ and as noted above, $x_0 \subseteq \mathrm{dom}(f_r)$. So $r^*(A_s) \cap S \subseteq \mathrm{dom}(f_r)$. \end{proof} \begin{definition} For any $N \in \mathcal X$, let $D_N$ denote the set of conditions $r \in \mathbb{P}$ satisfying: \begin{enumerate} \item $N \in A_r$; \item for all $M \in A_r$, if $M < N$ then $M \cap N \in A_r$; \item whenever $K \in f_r(x)$ and $x \in f_r(y)$, then $$ g_r(K,x) \subseteq g_r(K,y). $$ \end{enumerate} \end{definition} The next lemma says that $D_N$ is dense below any condition which contains $N$. \begin{lemma} Let $N \in \mathcal X$. Then for any condition $q \in \mathbb{P}$, if $N \in A_q$, then there is $s \le q$ such that $s \in D_N$. \end{lemma} \begin{proof} Let $q \in \mathbb{P}$ be such that $N \in A_q$. By Lemma 7.2, there is $r \le q$ such that for all $M \in A_r$, if $M < N$ then $M \cap N \in A_r$. By Lemma 4.9, there is $s \le r$ such that $f_s = f_r$, $A_s = A_r$, and whenever $K \in f_s(x)$ and $x \in f_s(y)$, then $g_s(K,x) \subseteq g_s(K,y)$. Then $s \le q$ and $s \in D_N$. \end{proof} \begin{definition} Suppose that $N \in \mathcal X$ is simple and $r \in D_N$. Define $r \restriction N$ as the triple $(f,g,A)$ satisfying: \begin{enumerate} \item $\mathrm{dom}(f) = \mathrm{dom}(f_r) \cap N$, and for all $x \in \mathrm{dom}(f)$, $f(x) := f_r(x) \cap N$; \item $\mathrm{dom}(g) = \mathrm{dom}(g_r) \cap N$, and for all $(K,x) \in \mathrm{dom}(g)$, $g(K,x) := g_r(K,x)$; \item $A := A_r \cap N$. \end{enumerate} \end{definition} Observe that in (1), if $x \in \mathrm{dom}(f) \setminus S$, then $x \in N$ implies that $Sk(x) \subseteq N$. Therefore, $f_r(x) \subseteq N$. So in this case, $f(x) = f_r(x) \cap N$ is equal to $f_r(x)$. Let us prove that $r \restriction N$ is in $N$. Obviously $A = A_r \cap N$ and $\mathrm{dom}(f) = \mathrm{dom}(f_r) \cap N$ are in $N$, and for all $x \in \mathrm{dom}(f)$, $f(x) = f_r(x) \cap N$ is in $N$. Consider $K \in f(x)$. Then $K$ and $x$ are in $N$. If $x \notin S$, then $x \subseteq N$. Therefore, $g(K,x) = g_r(K,x) \subseteq x \subseteq N$. So $g(K,x)$ is a finite subset of $N$, and hence is in $N$. Finally, suppose that $x = \alpha \in S$. Then $\alpha \in N$, and as $N \in A_r$, $N \cap \alpha$ is in $f_r(\alpha)$ by Definition 4.2(6). Also, $$ g_r(K,\alpha) \subseteq g_r(K,N \cap \alpha) \subseteq N \cap \alpha $$ by Definition 4.2(4,5). Hence, $g(K,\alpha) = g_r(K,\alpha)$ is a finite subset of $N$, and hence is in $N$. We have proven that all of the components of $r \restriction N$ are in $N$. Therefore, $r \restriction N$ is in $N$. \begin{lemma} Suppose that $N \in \mathcal X$ is simple and $r \in D_N$. Then $r \restriction N$ is in $N \cap \mathbb{P}$ and $r \le r \restriction N$. \end{lemma} \begin{proof} Let $r \restriction N = (f,g,A)$. We have already observed that $r \restriction N \in N$. It is trivial to check that if $r \restriction N$ is a condition, then $r \le r \restriction N$. So it suffices to show that $r \restriction N$ is a condition. We verify requirements (1)--(7) of Definition 4.2. (1), (4), (5), (6), and (7) are easy to check. It remains to prove (2) and (3). (2) Obviously $f$ is a function with a finite domain. Let $x \in \mathrm{dom}(f) = \mathrm{dom}(f_r) \cap N$. Then $x \in N$. We will show that either $x \in S$, or there is $M \in A$ and $\alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$, and moreover, $f(x)$ is a finite $\in$-chain and $f(x) \subseteq Sk(x) \setminus S$. We begin by showing that $f(x)$ is a finite $\in$-chain and $f(x) \subseteq Sk(x) \setminus S$. Since $r$ is a condition, $f_r(x)$ is a finite $\in$-chain and $f_r(x) \subseteq Sk(x) \setminus S$. But $f(x) = f_r(x) \cap N \subseteq f_r(x)$. Therefore, $f(x)$ is a finite $\in$-chain and $f(x) \subseteq Sk(x) \setminus S$. Now we show that either $x \in S$, or there is $M \in A$ and $\alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}$ such that $x = M \cap \alpha$. Since $x \in \mathrm{dom}(f_r)$ and $r$ is a condition, we have that either $x \in S$, or there is $M_1 \in A_r$ and $\alpha \in (M_1 \cap \mathrm{dom}(f_r) \cap S) \cup \{ \kappa \}$ such that $x = M_1 \cap \alpha$. In the first case, we are done, so assume the second case. Since $x = M_1 \cap \alpha \in N$, clearly $M_1 \cap \omega_1 \in N$. By Lemma 2.17(1), it follows that $M_1 < N$. Therefore, as $r \in D_N$, we have that $M_1 \cap N \in A_r$. But $N$ is simple, so also $M_1 \cap N \in N$. So $M_1 \cap N \in A_r \cap N = A$. Hence, to complete the proof, it suffices to show that $x = M_1 \cap N \cap \beta$, for some $\beta \in (M_1 \cap N \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}$. If $\alpha = \kappa$, then $x = M_1 \cap \kappa \in N$, so $M_1 \cap \kappa \subseteq N$. Hence $$ x = M_1 \cap \kappa = M_1 \cap N \cap \kappa, $$ and we are done. Suppose that $\alpha < \kappa$. First, assume that $\alpha \in N$. Then $\alpha \in \mathrm{dom}(f_r) \cap S \cap N = \mathrm{dom}(f) \cap S$. Also, $\alpha \in M_1 \cap N \cap \kappa$, which implies that $\alpha < \beta_{M_1,N}$ by Proposition 2.11. Since $M_1 < N$, it follows that $M_1 \cap \alpha \subseteq N$. Hence, $$ x = M_1 \cap \alpha = M_1 \cap N \cap \alpha. $$ As $\alpha \in M_1 \cap N \cap \mathrm{dom}(f) \cap S$, we are done. Secondly, assume that $\alpha \notin N$. We claim that $$ M_1 \cap \alpha = M_1 \cap N \cap \kappa, $$ which will finish the proof. Since $x = M_1 \cap \alpha$ is in $N$, the forward inclusion is immediate. For the reverse inclusion, let $\gamma \in M_1 \cap N \cap \kappa$, and we will show that $\gamma \in M_1 \cap \alpha$. Then $\gamma < \beta_{M_1,N}$ by Proposition 2.11, and therefore, since $M_1 < N$, we have that $$ M_1 \cap \gamma \subseteq M_1 \cap \beta_{M_1,N} \subseteq N. $$ Since $\alpha \notin N$, it follows that $\alpha \notin M_1 \cap \gamma$. But $\alpha \in M_1$. Therefore, $\gamma \le \alpha$. Also, $\gamma \in N$ and $\alpha \notin N$ implies that $\gamma \ne \alpha$, so $\gamma < \alpha$. Thus, $\gamma \in M_1 \cap \alpha$, completing the proof. (3) Let $x \in \mathrm{dom}(f) = \mathrm{dom}(f_r) \cap N$. Since $r$ is a condition, we have that $$ f(x) = f_r(x) \cap N \subseteq \mathrm{dom}(f_r) \cap N = \mathrm{dom}(f). $$ Thus, $f(x) \subseteq \mathrm{dom}(f)$. Let $K \in f(x) = f_r(x) \cap N$, and we will show that $f(K) = f(x) \cap Sk(K)$. Since $K \in f_r(x)$, $K \notin S$, so $Sk(K) \subseteq N$. Therefore, $f_r(K) \subseteq Sk(K) \subseteq N$. So $$ f(K) = f_r(K) \cap N = f_r(K). $$ As $r$ is a condition, we have that $f(K) = f_r(K) = f_r(x) \cap Sk(K)$. But $Sk(K) \subseteq N$ implies that $$ f_r(x) \cap Sk(K) = f_r(x) \cap Sk(K) \cap N = f(x) \cap Sk(K). $$ Thus, $f(K) = f(x) \cap Sk(K)$. \end{proof} The next lemma will not be used until Section 8. \begin{lemma} Suppose that $N \in \mathcal X$ is simple and $r \in D_N$. Assume that $p \in N \cap \mathbb{P}$ and $r \le p$. Then $r \restriction N \le p$. \end{lemma} \begin{proof} We verify properties (a)--(d) of Definition 4.2. (a) Since $p \in N$, $A_p \subseteq N$. As $r \le p$, we have that $$ A_p \subseteq A_r \cap N = A_{r \restriction N}. $$ (b) Since $p \in N$, $\mathrm{dom}(f_p) \subseteq N$ and for all $x \in \mathrm{dom}(f_p)$, $f_p(x) \subseteq N$. As $r \le p$, we have that $$ \mathrm{dom}(f_p) \subseteq \mathrm{dom}(f_r) \cap N = \mathrm{dom}(f_{r \restriction N}). $$ Let $x \in \mathrm{dom}(f_p)$. Then $$ f_p(x) \subseteq f_r(x) \cap N = f_{r \restriction N}(x). $$ (c) Let $(K,x) \in \mathrm{dom}(g_p)$. Then since $r \le p$, $$ g_p(K,x) \subseteq g_r(K,x) = g_{r \restriction N}(K,x). $$ (d) Assume that $K$ and $x$ are in $\mathrm{dom}(f_p)$ and $K \in f_{r \restriction N}(x)$. We claim that $K \in f_p(x)$. But $$ K \in f_{r \restriction N}(x) = f_r(x) \cap N \subseteq f_r(x). $$ So $K \in f_r(x)$. Since $r \le p$, it follows that $K \in f_p(x)$. \end{proof} We will now begin analyzing the situation where $r \in D_N$ and $w \le r \restriction N$ is in $N \cap \mathbb{P}$. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Then: \begin{enumerate} \item $A_r \cap N \subseteq A_w$; \item $\mathrm{dom}(f_r) \cap N \subseteq \mathrm{dom}(f_w)$, and for all $x \in \mathrm{dom}(f_r) \cap N$, $f_r(x) \cap N \subseteq f_w(x)$; \item $\mathrm{dom}(g_r) \cap N \subseteq \mathrm{dom}(g_w)$, and for all $(K,x) \in \mathrm{dom}(g_r) \cap N$, $g_r(K,x) \subseteq g_w(K,x)$. \end{enumerate} \end{lemma} \begin{proof} Immediate from the definition of $r \restriction N$ and the fact that $w \le r \restriction N$. \end{proof} As we discussed at the beginning of the section, we are going to show that whenever $w \le r \restriction N$, where $r \in D_N$ and $w \in N \cap \mathbb{P}$, then $w$ and $r$ are compatible. As in the previous section, we will construct a specific lower bound $w \oplus_N r$ of $w$ and $r$. We will describe separately the $f$, $g$, and $A$ components of $w \oplus_N r$. The $A$-component of $w \oplus_N r$ will be defined as $A_w \cup A_r$. We handle the $f$-component next. Unfortunately, the definition of $f_w \oplus_N f_r$ is much more complicated than in the previous section. The domain of $f_w \oplus_N f_r$ will include not only $\mathrm{dom}(f_w) \cup \mathrm{dom}(f_r)$, but also some additional sets. Before defining $f_w \oplus_N f_r$, we prove two lemmas which will help us handle its domain. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Then the set $$ \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r) \} $$ is disjoint from $N$, $\mathrm{dom}(f_w)$, and $\mathrm{dom}(f_r)$. \end{lemma} \begin{proof} It is clear that any member $M \cap \alpha$ of the displayed set is not in $N$, since $N \le M$ implies that $N \cap \omega_1 \le M \cap \omega_1 = M \cap \alpha \cap \omega_1$. Since $w \in N$, and hence $\mathrm{dom}(f_w) \subseteq N$, it follows that the displayed set is disjoint from $\mathrm{dom}(f_w)$. Suppose for a contradiction that for some $M \in A_r$ with $N \le M$ and some $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$, $M \cap \alpha$ is in $\mathrm{dom}(f_r)$. By Definition 4.2(2), fix $M_1 \in A_r$ and $\beta \in (M_1 \cap \mathrm{dom}(f_r) \cap S) \cup \{ \kappa \}$ such that $M \cap \alpha = M_1 \cap \beta$. Since $\alpha \in \kappa \setminus \mathrm{dom}(f_r)$ and $\beta \in \mathrm{dom}(f_r) \cup \{ \kappa \}$, $\alpha \ne \beta$. Applying Lemma 3.2 to $M$, $M_1$, $\alpha$, and $\beta$, we get that $M \sim M_1$ and $\alpha = \min((M \cap \kappa) \setminus \beta_{M,M_1})$. Since $M$ and $M_1$ are in $A_r$ and $\alpha \in S$, we have that $\alpha \in r^*(A_r) \cap S$. By Definition 4.2(7), it follows that $\alpha \in \mathrm{dom}(f_r)$. But this contradicts the choice of $\alpha$. \end{proof} \begin{lemma} Suppose that $r \in \mathbb{P}$, $N \in A_r$, and $x \in \mathrm{dom}(f_r)$. Then there is at most one ordinal $\alpha$ such that $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$ and $N \cap \alpha \in f_r(x) \cup \{ x \}$. \end{lemma} \begin{proof} Suppose for a contradiction that $\alpha < \beta$ are in $\mathrm{dom}(f_r) \cap S \cap N$, and $N \cap \alpha$ and $N \cap \beta$ are both in $f_r(x) \cup \{ x \}$. Then $N \cap \alpha$ and $N \cap \beta$ are membership comparable. Since $N \cap \alpha$ and $N \cap \beta$ have the same intersection with $\omega_1$, they must be equal. But then $\alpha \in N \cap \beta = N \cap \alpha$, which is impossible. \end{proof} We are ready to define $f_w \oplus_N f_r$. \begin{definition} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Define $f_w \oplus_N f_r = f$ as follows. The domain of $f$ is equal to the union of $\mathrm{dom}(f_w)$, $\mathrm{dom}(f_r)$, and the set $$ \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r) \}. $$ The values of $f$ are defined by the following cases: \begin{enumerate} \item for all $x \in \mathrm{dom}(f_w) \setminus S$, $$ f(x) := f_w(x); $$ \item for all $\alpha \in \mathrm{dom}(f_w) \cap S \cap \mathrm{dom}(f_r)$, $$ f(\alpha) := f_w(\alpha) \cup f_r(\alpha); $$ \item for all $\alpha \in (\mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$, $$ f(\alpha) := f_w(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}; $$ \item if $x \in \mathrm{dom}(f_r) \setminus N$, and for some $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$, $N \cap \alpha \in f_r(x) \cup \{ x \}$, then $$ f(x) := f_w(\alpha) \cup f_r(x); $$ \item for all $x \in \mathrm{dom}(f_r) \setminus N$ such that (4) fails, if $f_r(x) \cap N = \emptyset$, then $$ f(x) := f_r(x); $$ \item for all $x \in \mathrm{dom}(f_r) \setminus N$ such that (4) fails, if $f_r(x) \cap N \ne \emptyset$, then $$ f(x) := f_r(x) \cup f_w(M), $$ where $M$ is the membership largest element of $f_r(x) \cap N$; \item for a set of the form $M \cap \alpha$, where $M \in A_r$, $N \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$, $$ f(M \cap \alpha) := f(\alpha) \cap Sk(M \cap \alpha), $$ where $f(\alpha)$ was defined in (3). \end{enumerate} \end{definition} It is easy to see that cases 1--7 describe all of the possibilities for a set being in $\mathrm{dom}(f)$, using the fact that $\mathrm{dom}(f_r) \cap N \subseteq \mathrm{dom}(f_w)$ by Lemma 7.8(2). Moreover, cases 1--6 are obviously disjoint, and they are also disjoint from case 7 by Lemma 7.9. Finally, note that the ordinal $\alpha$ in case 4 is unique by Lemma 7.10, so $f(x)$ is well-defined in this case. The next four lemmas describe some important properties of $f_w \oplus_N f_r$. Lemmas 7.12 and 7.13 will be used to show that $w \oplus_N r$ is below $w$ and $r$ in $\mathbb{P}$. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Let $f := f_w \oplus_N f_r$. Then: \begin{enumerate} \item $\mathrm{dom}(f_w) \cup \mathrm{dom}(f_r) \subseteq \mathrm{dom}(f)$; \item if $x \in \mathrm{dom}(f_w)$, then $f_w(x) \subseteq f(x)$; \item if $x \in \mathrm{dom}(f_r)$, then $f_r(x) \subseteq f(x)$. \end{enumerate} \end{lemma} \begin{proof} (1,2) By the definition of $f$ in Definition 7.11, $\mathrm{dom}(f_w)$ and $\mathrm{dom}(f_r)$ are subsets of $\mathrm{dom}(f)$. By Cases 1, 2, and 3 of Definition 7.11, for all $x \in \mathrm{dom}(f_w)$, $f_w(x) \subseteq f(x)$. (3) Suppose that $x \in \mathrm{dom}(f_r)$, and we will show that $f_r(x) \subseteq f(x)$. We consider each of the cases 1--7 of Definition 7.11 in the definition of $f(x)$. If $f(x)$ is defined by cases 2, 4, 5, or 6, then $f_r(x) \subseteq f(x)$ by definition. Since $x \in \mathrm{dom}(f_r)$, case 3 does not hold, and case 7 does not hold by Lemma 7.9. It remains to consider case 1. For case 1, suppose that $x \in \mathrm{dom}(f_w) \setminus S$ and $f(x) = f_w(x)$. Then $x \in N$. So $f_r(x) \subseteq Sk(x) \subseteq N$. Hence, by Lemma 7.8(2), $f_r(x) = f_r(x) \cap N \subseteq f_w(x) = f(x)$. \end{proof} \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Let $f := f_w \oplus_N f_r$. Then: \begin{enumerate} \item $\mathrm{dom}(f) \cap N = \mathrm{dom}(f_w)$; \item if $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_w)$, then $K \in f_w(x)$; \item if $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_r)$, then $K \in f_r(x)$. \end{enumerate} \end{lemma} \begin{proof} (1) The inclusion $\mathrm{dom}(f_w) \subseteq \mathrm{dom}(f) \cap N$ is immediate, so it suffices to show that if $x \in \mathrm{dom}(f) \cap N$, then $x \in \mathrm{dom}(f_w)$. Note that $x$ is not equal to $M \cap \alpha$, for any $M \in A_r$ with $N \le M$ and $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$, since such a set is not in $N$ by Lemma 7.9. So by the definition of $\mathrm{dom}(f)$, we have that either $x \in \mathrm{dom}(f_w)$, in which case we are done, or $x \in \mathrm{dom}(f_r)$. In the second case, $x \in \mathrm{dom}(f_r) \cap N \subseteq \mathrm{dom}(f_w)$ by Lemma 7.8(2). \bigskip (2) Suppose that $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_w)$. We will show that $K \in f_w(x)$. Since $K$ and $x$ are in $\mathrm{dom}(f_w)$, they are in $N$. Hence, we are in cases 1, 2, or 3 of Definition 7.11. In case 1, $f(x) = f_w(x)$, so $K \in f_w(x)$. In case 2, $f(x) = f_w(\alpha) \cup f_r(\alpha)$, where $x = \alpha \in \mathrm{dom}(f_w) \cap S \cap \mathrm{dom}(f_r)$. So either $K \in f_w(\alpha)$, in which case we are done, or $K \in f_r(\alpha)$. In the second case, by Lemma 7.8(2) we have that $$ K \in f_r(\alpha) \cap N \subseteq f_w(\alpha) = f_w(x). $$ In case 3, $f(x)$ is equal to the union of $f_w(x)$ together with a collection of sets which are not members of $N$. Since $K \in N$, $K \in f_w(x)$. \bigskip (3) Suppose that $K \in f(x)$ and $K$ and $x$ are in $\mathrm{dom}(f_r)$. We will show that $K \in f_r(x)$. First, assume that $K$ and $x$ are both in $N$. Then $K$ and $x$ are in $\mathrm{dom}(f_r) \cap N \subseteq \mathrm{dom}(f_w)$ by Lemma 7.8(2). By (2) just proven, $K \in f_w(x)$. Yet $K$ and $x$ are in $\mathrm{dom}(f_r) \cap N = \mathrm{dom}(f_{r \restriction N})$. Since $w \le r \restriction N$, it follows that $K \in f_{r \restriction N}(x) \subseteq f_r(x)$. So $K \in f_r(x)$. Next, we consider each of the cases 1--7 of Definition 7.11 for $f(x)$. Case 1 is immediate, since it implies that $K$ and $x$ are in $N$. In case 2, $x = \alpha$ is in $N$ and $f(x) = f_w(\alpha) \cup f_r(\alpha)$. If $K \in f_w(\alpha)$, then $K$ and $x$ are both in $N$, and we are done. Otherwise $K \in f_r(\alpha)$, and we are also done. Case 3 does not apply, since it says that $x = \alpha$ is not in $\mathrm{dom}(f_r)$. In case 5, $f(x) = f_r(x)$, so $K \in f_r(x)$, and we are done. Case 7 does not apply, since any set of the form described there is not in $\mathrm{dom}(f_r)$ by Lemma 7.9. It remains to consider cases 4 and 6. In case 4, there is $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$ such that $N \cap \alpha \in f_r(x) \cup \{ x \}$, and $f(x) = f_w(\alpha) \cup f_r(x)$. So either $K \in f_w(\alpha)$, or $K \in f_r(x)$. In the second case we are done, so assume that $K \in f_w(\alpha)$. By Lemma 7.12(2), $f_w(\alpha) \subseteq f(\alpha)$, so $K \in f(\alpha)$. Since $K$ and $\alpha$ are in $\mathrm{dom}(f_r)$, it follows by case 2 just handled that $K \in f_r(\alpha)$. But also $K \in f_w(\alpha)$ means that $K \in N$. Since $N \cap \alpha \in f_r(\alpha)$ by Definition 4.2(6) and $K \in N$, we must have that $K \in f_r(N \cap \alpha)$. But $N \cap \alpha \in f_r(x) \cup \{ x \}$, so $K \in f_r(x)$. In case 6, $f(x) = f_r(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_r(x) \cap N$. If $K \in f_r(x)$ then we are done, so assume that $K \in f_w(M)$. Since $M$ is in $\mathrm{dom}(f_r) \cap N \subseteq \mathrm{dom}(f_w)$ by Lemma 7.8(2), $M$ is in $\mathrm{dom}(f_w) \setminus S$. So $f(M)$ is defined as in case 1 of Definition 7.11. Hence, $f(M) = f_w(M)$. Therefore, $K \in f(M)$. Since $K$ and $M$ are in $\mathrm{dom}(f_r)$, it follows by case 1 handled above that $K \in f_r(M)$. Thus, $K \in f_r(M)$ and $M \in f_r(x)$, so $K \in f_r(x)$. \end{proof} The next lemma will be used to show that $f_w \oplus_N f_r$ satisfies requirement (2) of Definition 4.2 for $w \oplus_N r$. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Let $f := f_w \oplus_N f_r$. Then $f$ is a function with a finite domain, and for all $x \in \mathrm{dom}(f)$, either $x \in S$, or there is $M \in A_w \cup A_r$ and $$ \alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \} $$ satisfying that $x = M \cap \alpha$; moreover, for all $x \in \mathrm{dom}(f)$, $f(x)$ is a finite $\in$-chain and $f(x) \subseteq Sk(x) \setminus S$. \end{lemma} \begin{proof} It is immediate that $f$ is a function with a finite domain. By the definition of the domain of $f$ in Definition 7.11, together with the fact that $\mathrm{dom}(f_w) \cup \mathrm{dom}(f_r) \subseteq \mathrm{dom}(f)$ and $w$ and $r$ are conditions, it is easy to see that if $x \in \mathrm{dom}(f)$, then either $x \in S$, or there is $M \in A_w \cup A_r$ and $\alpha \in (M \cap \mathrm{dom}(f) \cap S) \cup \{ \kappa \}$ satisfying that $x = M \cap \alpha$. Let $x \in \mathrm{dom}(f)$, and we will show that $f(x)$ is a finite $\in$-chain and $f(x) \subseteq Sk(x) \setminus S$. We will consider each of the cases 1--7 of Definition 7.11. We are done if either $f(x) = f_w(x)$ or $f(x) = f_r(x)$ as in cases 1 and 5, since $w$ and $r$ are conditions. In case 7, $f(x) \subseteq Sk(x)$ by definition, and $f(x)$ is a finite $\in$-chain and disjoint from $S$ provided that the result is true for case 3. It remains to handle cases 2, 3, 4, and 6. \bigskip \noindent \emph{Case 2:} $x = \alpha \in \mathrm{dom}(f_w) \cap S \cap \mathrm{dom}(f_r)$ and $f(\alpha) = f_w(\alpha) \cup f_r(\alpha)$. Since $w$ and $r$ are conditions, $f_w(\alpha)$ and $f_r(\alpha)$ are themselves finite $\in$-chains and subsets of $Sk(\alpha) \setminus S$. So it suffices to show that if $K \in f_w(\alpha)$ and $M \in f_r(\alpha) \setminus f_w(\alpha)$, then $K \in Sk(M)$. Since $\alpha \in \mathrm{dom}(f_w)$, $\alpha$ is in $N$. Thus, $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$. Since $N \in A_r$, we have that $N \cap \alpha \in f_r(\alpha)$, because $r$ is a condition. As $M$ is also in $f_r(\alpha)$, $M$ and $N \cap \alpha$ are membership comparable. We claim that $M$ is not in $Sk(N \cap \alpha)$. Otherwise, $M \in N$, so $M \in f_r(\alpha) \cap N \subseteq f_w(\alpha)$ by Lemma 7.8(2). Hence, $M \in f_w(\alpha)$, which contradicts the choice of $M$. Thus, either $M = N \cap \alpha$, or $N \cap \alpha \in Sk(M)$. In either case, $Sk(N \cap \alpha) \subseteq Sk(M)$. Since $K \in f_w(\alpha)$, $K \subseteq \alpha$. Therefore, $K \in N$ implies that $K \in Sk(N \cap \alpha)$ by Lemma 4.3(3). Hence, $K \in Sk(N \cap \alpha) \subseteq Sk(M)$, so $K \in Sk(M)$. \bigskip \noindent \emph{Case 3:} $x = \alpha \in (\mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$ and $$ f(\alpha) = f_w(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}. $$ Since $w$ is a condition, $f_w(\alpha)$ is a finite $\in$-chain and a subset of $Sk(\alpha) \setminus S$. By Lemma 2.33, the second set in the above union is also a finite $\in$-chain and a subset of $Sk(\alpha)$, and it is obviously disjoint from $S$. So it suffices to show that if $K \in f_w(\alpha)$, $M \in A_r$, $N \le M$, and $\alpha \in M$, then $K \in Sk(M \cap \alpha)$. Since $K \in f_w(\alpha)$, we have that $K \in N$ and $K \subseteq \alpha$. By Lemma 4.3(3), $K \in Sk(N \cap \alpha)$. Since $\alpha \in M \cap N$, $\alpha < \beta_{M,N}$ by Proposition 2.11. As $N \le M$, we have that $N \cap \alpha \subseteq M \cap \alpha$. Thus, $Sk(N \cap \alpha) \subseteq Sk(M \cap \alpha)$. So $K \in Sk(M \cap \alpha)$. \bigskip \noindent \emph{Case 4:} $x \in \mathrm{dom}(f_r) \setminus N$, $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$, and $N \cap \alpha \in f_r(x) \cup \{ x \}$. Then $f(x) = f_w(\alpha) \cup f_r(x)$. Since $w$ and $r$ are conditions, $f_w(\alpha)$ and $f_r(x)$ are themselves finite $\in$-chains, and subsets of $Sk(\alpha) \setminus S$ and $Sk(x) \setminus S$ respectively. Consider $K \in f_w(\alpha)$, and we will show that $K \in Sk(x)$. So $K \in N$ and $K \subseteq \alpha$. By Lemma 4.3(3), $K \in Sk(N \cap \alpha)$. But $N \cap \alpha \in f_r(x) \cup \{ x \}$, so in particular, $Sk(N \cap \alpha) \subseteq Sk(x)$. Thus, $K \in Sk(x)$. This completes the proof that $f(x) \subseteq Sk(x) \setminus S$. Since $f_w(\alpha)$ and $f_r(x)$ are themselves finite $\in$-chains, in order to show that $f(x)$ is a finite $\in$-chain, it suffices to show that if $K \in f_w(\alpha)$ and $L \in f_r(x) \setminus f_w(\alpha)$, then $K \in Sk(L)$. Since $K \in N$ and $K \subseteq \alpha$, $K \in Sk(N \cap \alpha)$ by Lemma 4.3(3). As $L \in f_r(x)$ and $N \cap \alpha \in f_r(x) \cup \{ x \}$, $L$ and $N \cap \alpha$ are membership comparable. We claim that $L$ is not in $Sk(N \cap \alpha)$. Suppose for a contradiction that $L \in Sk(N \cap \alpha)$. Then $L \in N$, and $L \in f_r(N \cap \alpha) \subseteq f_r(\alpha)$. So $L \in f_r(\alpha) \cap N \subseteq f_w(\alpha)$ by Lemma 7.8(2). Hence, $L \in f_w(\alpha)$, which contradicts the choice of $L$. Thence, either $L = N \cap \alpha$ or $N \cap \alpha \in Sk(L)$. In either case, $Sk(N \cap \alpha) \subseteq Sk(L)$, and in particular, $K \in Sk(L)$. \bigskip \noindent \emph{Case 6:} $x \in \mathrm{dom}(f_r) \setminus N$ and $f(x) = f_r(x) \cup f_w(M)$, where $M$ is the membership largest element of $f_r(x) \cap N$. Since $r$ and $w$ are conditions, $f_r(x)$ is a finite $\in$-chain and a subset of $Sk(x) \setminus S$, and $f_w(M)$ is a finite $\in$-chain and a subset of $Sk(M) \setminus S$. Since $M \in f_r(x)$, $M \in Sk(x)$. Therefore, $f_w(M) \subseteq Sk(M) \subseteq Sk(x)$. Hence, both $f_r(x)$ and $f_w(M)$ are finite $\in$-chains and subsets of $Sk(x) \setminus S$. So it suffices to show that if $K \in f_w(M)$ and $L \in f_r(x) \setminus f_w(M)$, then $K \in Sk(L)$. Since $L$ and $M$ are both in $f_r(x)$, they are membership comparable. But if $L \in Sk(M)$, then by Lemma 7.8(2), $$ L \in (f_r(x) \cap Sk(M)) = f_r(M) \subseteq f_w(M). $$ So $L \in f_w(M)$, which contradicts the choice of $L$. Therefore, either $L = M$, or $M \in Sk(L)$. In either case, $Sk(M) \subseteq Sk(L)$. As $K \in f_w(M)$, $K \in Sk(M)$, so $K \in Sk(M) \subseteq Sk(L)$. \end{proof} The next lemma will be used to show that $f_w \oplus_N f_r$ satisfies requirement (3) of Definition 4.2 for $w \oplus_N r$. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Let $f := f_w \oplus_N f_r$. If $x \in \mathrm{dom}(f)$, then $f(x) \subseteq \mathrm{dom}(f)$, and for all $K \in f(x)$, $f(K) = f(x) \cap Sk(K)$. \end{lemma} \begin{proof} We begin by proving that $f(x) \subseteq \mathrm{dom}(f)$. This is immediate in cases 1, 2, 4, 5, and 6 of Definition 7.11, since the fact that $w$ and $r$ are conditions implies that for any $y \in \mathrm{dom}(f_w)$ and $z \in \mathrm{dom}(f_r)$, $f_w(y) \subseteq \mathrm{dom}(f_w) \subseteq \mathrm{dom}(f)$ and $f_r(z) \subseteq \mathrm{dom}(f_r) \subseteq \mathrm{dom}(f)$. Also, case 7 follows from case 3. It remains to consider case 3. In case 3, $\alpha$ is in $(\mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$ and $f(\alpha) = f_w(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}$. Again, we know that $f_w(\alpha) \subseteq \mathrm{dom}(f_w) \subseteq \mathrm{dom}(f)$. And if $M \in A_r$, $N \le M$, and $\alpha \in M$, then $M \cap \alpha$ is in $\mathrm{dom}(f)$ by Definition 7.11. \bigskip Let $K \in f(x)$, and we will prove that $f(K) = f(x) \cap Sk(K)$. The proof splits into the seven cases of Definition 7.11 in the definition of $f(x)$. Note that since $K \in f(x)$, it follows that $K \notin S$ by Lemma 7.14. \bigskip \noindent \emph{Case 1:} $x \in \mathrm{dom}(f_w) \setminus S$ and $f(x) = f_w(x)$. Then $K \in f_w(x)$. So $K \in \mathrm{dom}(f_w) \setminus S$ as well. Hence, $f(K) = f_w(K)$. Therefore, $$ f(K) = f_w(K) = f_w(x) \cap Sk(K) = f(x) \cap Sk(K). $$ \bigskip \noindent \emph{Case 2:} $x = \alpha \in \mathrm{dom}(f_w) \cap S \cap \mathrm{dom}(f_r)$ and $f(\alpha) = f_w(\alpha) \cup f_r(\alpha)$. So either $K \in f_w(\alpha)$ or $K \in f_r(\alpha)$. First, assume that $K \notin N$. Then $K \in f_r(\alpha) \setminus N$. By Definition 4.2(6), $N \cap \alpha \in f_r(\alpha)$. So both $K$ and $N \cap \alpha$ are in $f_r(\alpha)$. As $K \notin N$, $N \cap \alpha \in f_r(K) \cup \{ K \}$. So we are in case 4 of Definition 7.11 in the definition of $f(K)$. Therefore, $f(K) = f_w(\alpha) \cup f_r(K)$. So it suffices to show that $$ f_w(\alpha) \cup f_r(K) = (f_w(\alpha) \cup f_r(\alpha)) \cap Sk(K). $$ Since $w \in N$, $f_w(\alpha) \subseteq N \cap Sk(\alpha) = Sk(N \cap \alpha)$ by Lemma 2.7(1). Since $N \cap \alpha \in f_r(K) \cup \{ K \}$, $N \cap \alpha \subseteq K$, so $Sk(N \cap \alpha) \subseteq Sk(K)$. Hence, $f_w(\alpha) \subseteq Sk(K)$. Also, as $K \in f_r(\alpha)$, $f_r(K) = f_r(\alpha) \cap Sk(K)$. Thence, \begin{multline*} $$ (f_w(\alpha) \cup f_r(\alpha)) \cap Sk(K) = \\ = (f_w(\alpha) \cap Sk(K)) \cup (f_r(\alpha) \cap Sk(K)) = f_w(\alpha) \cup f_r(K). \end{multline*} Secondly, assume that $K \in N$. Then either $K \in f_w(\alpha)$, or $K \in f_r(\alpha) \cap N \subseteq f_w(\alpha)$ by Lemma 7.8(2). In either case, $K \in f_w(\alpha)$. So $K \in \mathrm{dom}(f_w) \setminus S$, and therefore, by definition, $f(K) = f_w(K)$. So $$ f(K) = f_w(K) = f_w(\alpha) \cap Sk(K). $$ Since $K$ and $w$ are $N$, and $f_r(\alpha) \cap N \subseteq f_w(\alpha)$ by Lemma 7.8(2), we have that \begin{multline*} f(\alpha) \cap Sk(K) \subseteq f(\alpha) \cap N = \\ = (f_w(\alpha) \cup f_r(\alpha)) \cap N \subseteq f_w(\alpha) \cup (f_r(\alpha) \cap N) \subseteq f_w(\alpha). \end{multline*} So $f(\alpha) \cap Sk(K) \subseteq f_w(\alpha)$. On the other hand, by Lemma 7.12(2), $f_w(\alpha) \subseteq f(\alpha)$. Therefore, $$ f(\alpha) \cap Sk(K) = f_w(\alpha) \cap Sk(K) = f(K). $$ \bigskip \noindent \emph{Case 3:} $x = \alpha \in (\mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$ and $$ f(\alpha) = f_w(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}. $$ First, assume that $K \in f_w(\alpha)$. Then $f(K)$ is defined as in case 1 of Definition 7.11, so $f(K) = f_w(K)$. Since $K \in N$, and the members of the second set in the displayed union are not in $N$ by Lemma 7.9, and hence not in $Sk(K)$, we have that $$ f(\alpha) \cap Sk(K) = f_w(\alpha) \cap Sk(K) = f_w(K) = f(K). $$ Secondly, assume that $K = M \cap \alpha$, where $M \in A_r$, $N \le M$, and $\alpha \in M$. Then $f(K)$ is defined as in case 7 of Definition 7.11, namely, $$ f(K) = f(M \cap \alpha) = f(\alpha) \cap Sk(M \cap \alpha) = f(\alpha) \cap Sk(K). $$ \bigskip \noindent \emph{Case 4:} $x \in \mathrm{dom}(f_r) \setminus N$, $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$, and $N \cap \alpha \in f_r(x) \cup \{ x \}$. Then $f(x) = f_w(\alpha) \cup f_r(x)$. Note that by Definition 4.2(6), $N \cap \alpha \in f_r(\alpha)$. First, assume that $K \in N$. Then either $K \in f_w(\alpha)$, or $K \in f_r(x) \cap N$. Consider the second case. Since $K$ and $N \cap \alpha$ are both in $f_r(x) \cup \{ x \}$ and $K \in N$, we must have that $K \in f_r(N \cap \alpha)$. Then $K \in f_r(N \cap \alpha) \subseteq f_r(\alpha)$. So $K \in f_r(\alpha) \cap N \subseteq f_w(\alpha)$ by Lemma 7.8(2). Hence, in either case, $K \in f_w(\alpha)$. By Definition 7.11(1), $f(K) = f_w(K)$. Since $K \in f_w(\alpha)$, $f(K) = f_w(K) = f_w(\alpha) \cap Sk(K)$. Therefore, \begin{multline*} f(x) \cap Sk(K) = (f_w(\alpha) \cup f_r(x)) \cap Sk(K) = \\ = (f_w(\alpha) \cap Sk(K)) \cup (f_r(x) \cap Sk(K)) = f_w(K) \cup (f_r(x) \cap Sk(K)). \end{multline*} Since $f(K) = f_w(K)$, it suffices to show that $f_r(x) \cap Sk(K) \subseteq f_w(K)$, for then $$ f(x) \cap Sk(K) = f_w(K) \cup (f_r(x) \cap Sk(K)) = f_w(K) = f(K). $$ Let $z \in f_r(x) \cap Sk(K)$, and we will show that $z \in f_w(K)$. Since $K \in N$, $z \in N$. As $z \in f_r(x) \cap N$ and $N \cap \alpha \in f_r(x) \cup \{ x \}$, we have that $z \in f_r(N \cap \alpha)$. So $z \in f_r(N \cap \alpha) \cap N \subseteq f_r(\alpha) \cap N \subseteq f_w(\alpha)$ by Lemma 7.8(2). So $z \in f_w(\alpha)$. Hence, $z \in f_w(\alpha) \cap Sk(K) = f_w(K)$, which completes the proof. Secondly, assume that $K \notin N$. Then since $K \in f(x) = f_w(\alpha) \cup f_r(x)$, and $w \in N$, we must have that $K \in f_r(x)$. As $N \cap \alpha \in f_r(x) \cup \{ x \}$ and $K \notin N$, it follows that $N \cap \alpha \in f_r(K) \cup \{ K \}$. So we are in case 4 of Definition 7.11 in the definition of $f(K)$. That means that $f(K) = f_w(\alpha) \cup f_r(K)$. We have that \begin{multline*} f(x) \cap Sk(K) = (f_w(\alpha) \cup f_r(x)) \cap Sk(K) = \\ = (f_w(\alpha) \cap Sk(K)) \cup (f_r(x) \cap Sk(K)) = (f_w(\alpha) \cap Sk(K)) \cup f_r(K). \end{multline*} So $f(x) \cap Sk(K) = (f_w(\alpha) \cap Sk(K)) \cup f_r(K)$. On the other hand, \begin{multline*} f(K) = f(K) \cap Sk(K) = (f_w(\alpha) \cup f_r(K)) \cap Sk(K) = \\ = (f_w(\alpha) \cap Sk(K)) \cup (f_r(K) \cap Sk(K)) = (f_w(\alpha) \cap Sk(K)) \cup f_r(K). \end{multline*} So $f(K) = (f_w(\alpha) \cap Sk(K)) \cup f_r(K) = f(x) \cap Sk(K)$. \bigskip \noindent \emph{Case 5:} $x \in \mathrm{dom}(f_r) \setminus N$, case 4 fails for $x$, $f_r(x) \cap N = \emptyset$, and $f(x) = f_r(x)$. Then $K \in f_r(x)$. Since $f_r(x) \cap N = \emptyset$, $K \notin N$. Also, since $f_r(K) = f_r(x) \cap Sk(K)$, we have that $f_r(K) \cap N = \emptyset$. And as $f_r(K) \cup \{ K \} \subseteq f_r(x)$, we cannot have that $N \cap \alpha \in f_r(K) \cup \{ K \}$ for any $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$, since otherwise case 4 would be true for $x$. So we are in case 5 of Definition 7.11 in the definition of $f(K)$, and therefore $f(K) = f_r(K)$. So $$ f(K) = f_r(K) = f_r(x) \cap Sk(K) = f(x) \cap Sk(K). $$ \bigskip \noindent \emph{Case 6:} $x \in \mathrm{dom}(f_r) \setminus N$, case 4 fails, $M$ is the largest member of $f_r(x) \cap N$, and $f(x) = f_r(x) \cup f_w(M)$. By the maximality of $M$, and since $f_r(M) = f_r(x) \cap Sk(M)$, we have that $$ f_r(x) \cap N = f_r(M) \cup \{ M \} \subseteq f_w(M) \cup \{ M \}, $$ where the inclusion holds by Lemma 7.8(2) and the fact that $f_r(M) = f_r(M) \cap N$. Therefore, $$ f(x) \cap N = (f_r(x) \cup f_w(M)) \cap N = f_w(M) \cup \{ M \}. $$ Also, since $M \in N$, clearly $f_r(M) \setminus N = \emptyset$, and therefore, $$ f(x) \setminus N = (f_r(x) \cup f_w(M)) \setminus N = f_r(x) \setminus N. $$ To summarize, $$ f(x) \cap N = f_w(M) \cup \{ M \}, \ \ \ f(x) \setminus N = f_r(x) \setminus N. $$ We split into the two cases of whether $K \in N$ or $K \notin N$. First, assume that $K \in N$. Then $K \in f_w(M) \cup \{ M \}$. So $f(K)$ is defined as in case 1 of Definition 7.11 and $f(K) = f_w(K)$. So it suffices to show that $$ f_w(K) = f(x) \cap Sk(K). $$ For the forward inclusion, $K \in f_w(M) \cup \{ M \}$ implies that $$ f_w(K) \subseteq f_w(M) \subseteq f_r(x) \cup f_w(M) = f(x). $$ So $f_w(K) \subseteq f(x) \cap Sk(K)$. For the reverse inclusion, let $J \in f(x) \cap Sk(K)$, and we will show that $J \in f_w(K)$. Then $$ J \in f(x) \cap N = f_w(M) \cup \{ M \}. $$ Since $K \in f_w(M) \cup \{ M \}$ and $J \in Sk(K)$, clearly $J \ne M$. Therefore, $J \in f_w(M)$. Since $w$ is a condition, $$ J \in f_w(M) \cap Sk(K) = f_w(K). $$ Secondly, assume that $K \notin N$. Then $K \in f(x) \setminus N = f_r(x) \setminus N$. Hence, $f_r(K) = f_r(x) \cap Sk(K)$, since $r$ is a condition. As $K$ and $M$ are both in $f_r(x)$, they are membership comparable. But as $M \in N$ and $K \notin N$, clearly we must have that $M \in Sk(K)$. Since $M$ is the membership largest element of $f_r(x) \cap N$ and $M \in f_r(K)$, $M$ is the membership largest element of $f_r(K) \cap N$. Moreover, since $K \in f_r(x)$, $f_r(K) \cup \{ K \} \subseteq f_r(x)$, so there is no $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$ with $N \cap \alpha \in f_r(K) \cup \{ K \}$. So $f(K)$ is defined as in case 6 of Definition 7.11, which means that $f(K) = f_r(K) \cup f_w(M)$. So \begin{multline*} f(K) = f_r(K) \cup f_w(M) = (f_r(x) \cap Sk(K)) \cup f_w(M) = \\ = (f_r(x) \cup f_w(M)) \cap (Sk(K) \cup f_w(M)) = f(x) \cap Sk(K), \end{multline*} where the last equality follows from the definition of $f(x)$ and the fact that $f_w(M) \subseteq Sk(M) \subseteq Sk(K)$. So $f(K) = f(x) \cap Sk(K)$. \bigskip \noindent \emph{Case 7:} $x = M \cap \alpha$, where $M \in A_r$, $N \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$. Then $$ f(x) = f(\alpha) \cap Sk(M \cap \alpha), $$ where $f(\alpha)$ is defined as in case 3 of Definition 7.11. So $K \in f(\alpha)$. Also, $x = M \cap \alpha \in f(\alpha)$ by the definition of $f(\alpha)$. By case 3 handled above, $$ f(K) = f(\alpha) \cap Sk(K), \ \ \ f(x) = f(\alpha) \cap Sk(x). $$ We also know from Lemma 7.14 that $K \in f(x)$ implies that $K \in Sk(x)$, and hence $Sk(K) \subseteq Sk(x)$. Therefore, $Sk(x) \cap Sk(K) = Sk(K)$. Consequently, $$ f(K) = f(\alpha) \cap Sk(K) = f(\alpha) \cap Sk(x) \cap Sk(K) = f(x) \cap Sk(K). $$ \end{proof} This completes our analysis of $f_w \oplus_N f_r$. We now turn to amalgamating the $g$-components of $w$ and $r$. \begin{definition} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Let $f := f_w \oplus_N f_r$. Define $g_w \oplus_N g_r$ as the function $g$ with domain equal to the set of pairs $(K,x)$ such that $K \in f(x)$, such that for all $(K,x) \in \mathrm{dom}(g)$, $$ g(K,x) := \bigcup \{ g_w(K,y) \cup g_r(K,y) : x = y, \textrm{or} \ x \in f(y) \}.\footnote{As in the previous section, we let $g_w(K,y)$ denote the empty set in the case that $(K,y) \notin \mathrm{dom}(g_w)$, and similarly with $g_r$.} $$ \end{definition} The next lemma will be used to show that $g_w \oplus_N g_r$ satisfies requirement (4) of Definition 4.2 for $w \oplus_N r$. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Let $g := g_w \oplus_N g_r$. Then for any $(K,x)$ in $\mathrm{dom}(g)$, $$ g(K,x) \subseteq x \setminus \sup(K). $$ \end{lemma} \begin{proof} Let $(K,x) \in \mathrm{dom}(g)$. Then $K \in f(x)$. By definition, we have that $$ g(K,x) = \bigcup \{ g_w(K,y) \cup g_r(K,y) : x = y, \textrm{or} \ x \in f(y) \}. $$ Since $g_w(K,x)$ and $g_r(K,x)$ are subsets of $x \setminus \sup(K)$, it suffices to show that if $x \in f(y)$, then $g_w(K,y)$ and $g_r(K,y)$ are subsets of $x \setminus \sup(K)$. Since $w$ are $r$ are conditions, $g_w(K,y)$ and $g_r(K,y)$ are subsets of $y \setminus \sup(K)$. Hence, it suffices to show that $g_w(K,y)$ and $g_r(K,y)$ are subsets of $x$. \bigskip \noindent \emph{Claim 1:} The result holds if $K$, $x$, and $y$ are all in $N$. \bigskip If $K$, $x$, and $y$ are all in $N$, then $K \in f(x)$ and $x \in f(y)$ imply by Lemma 7.13(1,2) that $K \in f_w(x)$ and $x \in f_w(y)$. Since $w$ is a condition, it follows that $g_w(K,y) \subseteq g_w(K,x) \subseteq x$. To show that $g_r(K,y) \subseteq x$, assume that $(K,y) \in \mathrm{dom}(g_r)$, which means that $K \in f_r(y)$. Then by Lemma 7.8(3), $g_r(K,y) \subseteq g_w(K,y)$. But we just showed that $g_w(K,y) \subseteq x$, so we are done. \bigskip \noindent \emph{Claim 2:} The result holds if $K$, $x$, and $y$ are all in $\mathrm{dom}(f_r)$. \bigskip If $K$, $x$, and $y$ are all in $\mathrm{dom}(f_r)$, then $K \in f(x)$ and $x \in f(y)$ imply by Lemma 7.13(3) that $K \in f_r(x)$ and $x \in f_r(y)$. Since $r$ is a condition, $$ g_r(K,y) \subseteq g_r(K,x) \subseteq x. $$ To show that $g_w(K,y) \subseteq x$, assume that $(K,y) \in \mathrm{dom}(g_w)$, which means that $K \in f_w(y)$. In particular, $K$ and $y$ are in $N$. So if $x$ is also in $N$, then we are done by Claim 1. Assume that $x \notin N$. Note that $y \in S$. For if $y \notin S$, then $y$ is countable. So $y \in N$ implies that $Sk(y) \subseteq N$, and hence $x \in N$, which is false. Therefore, $y = \alpha$ for some $\alpha \in \mathrm{dom}(f_r) \cap S$. Since $N \in A_r$ and $\alpha \in N$, $N \cap \alpha \in f_r(\alpha)$ by Definition 4.2(6). As $x \in f_r(\alpha)$, $x$ and $N \cap \alpha$ are membership comparable. But $x \notin N$. Therefore, either $x = N \cap \alpha$ or $N \cap \alpha \in Sk(x)$. In either case, $N \cap \alpha \subseteq x$. As $w \in N$, we have that $$ g_w(K,y) = g_w(K,\alpha) \subseteq N \cap \alpha \subseteq x. $$ \bigskip The rest of the proof splits up into the cases 1--7 of Definition 7.11 in the definition of $f(y)$. \bigskip \noindent \emph{Case 1:} $y \in \mathrm{dom}(f_w) \setminus S$ and $f(y) = f_w(y)$. Then $K$, $x$, and $y$ are in $N$. So we are done by Claim 1. \bigskip \noindent \emph{Case 2:} $y = \alpha \in \mathrm{dom}(f_w) \cap S \cap \mathrm{dom}(f_r)$ and $f(y) = f_w(\alpha) \cup f_r(\alpha)$. Then $y \in N$. By Definition 4.2(6), $N \cap \alpha \in f_r(\alpha)$. Assume that $(K,\alpha) \in \mathrm{dom}(g_w)$, which means that $K \in f_w(\alpha)$. We will show that $g_w(K,\alpha) \subseteq x$. In particular, $K \in N$. If $x$ is also in $N$, then we are done by Claim 1. So assume that $x \notin N$. Since $x \in f(y) = f_w(\alpha) \cup f_r(\alpha)$ and $x \notin N$, we have that $x \in f_r(\alpha)$. So both $N \cap \alpha$ and $x$ are in $f_r(\alpha)$. As $x \notin N$, either $x = N \cap \alpha$ or $N \cap \alpha \in Sk(x)$. In either case, $N \cap \alpha \subseteq x$. But $g_w(K,\alpha) \subseteq N \cap \alpha$, since $w \in N$. Therefore, $g_w(K,\alpha) \subseteq x$. Assume that $(K,\alpha) \in \mathrm{dom}(g_r)$, which means that $K \in f_r(\alpha)$. We will show that $g_r(K,\alpha) \subseteq x$. So $K$ and $\alpha$ are both in $\mathrm{dom}(f_r)$. If $x$ is also in $\mathrm{dom}(f_r)$, then we are done by Claim 2. So assume that $x \notin \mathrm{dom}(f_r)$. But $x \in f(y) = f_w(\alpha) \cup f_r(\alpha)$, so $x \in f_w(\alpha)$. So $x \in N$. As $K \in Sk(x)$, also $K \in N$. So $K$, $x$, and $\alpha$ are all in $N$, and we are done by Claim 1. \bigskip \noindent \emph{Case 3:} $y = \alpha \in (\mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$. Since $\alpha \notin \mathrm{dom}(f_r)$, we have that $(K,\alpha) \notin \mathrm{dom}(g_r)$. Therefore, $g_r(K,\alpha) = \emptyset \subseteq x$. Assume that $(K,\alpha) \in \mathrm{dom}(g_w)$, which means that $K \in f_w(\alpha)$. We will show that $g_w(K,\alpha) \subseteq x$. In particular, $K$ and $\alpha$ are both in $N$. So if $x$ is also in $N$, then we are done by Claim 1. Suppose that $x \notin N$. Then by the definition of $f(\alpha)$, we have that $x = M \cap \alpha$, for some $M \in A_r$ with $N \le M$ and $\alpha \in M$. Then $\alpha \in M \cap N \cap \kappa$, so $\alpha < \beta_{M,N}$ by Proposition 2.11. Since $N \le M$, we have that $N \cap \alpha \subseteq M \cap \alpha = x$. As $w \in N$, it follows that $g_w(K,\alpha) \subseteq N \cap \alpha \subseteq x$. \bigskip Before handling cases 4--7, let us note that in each of these cases, $y \notin N$. This is immediate in cases 4, 5, and 6, and follows from Lemma 7.9 in case 7. Consequently, $(K,y) \notin \mathrm{dom}(g_w)$, and hence $g_w(K,y) = \emptyset \subseteq x$. Thus, we only need to show that $g_r(K,y) \subseteq x$. Assume that $(K,y) \in \mathrm{dom}(g_r)$, which means that $K \in f_r(y)$. We will show that $g_r(K,y) \subseteq x$. If $x \in \mathrm{dom}(f_r)$, then we are done by Claim 2. Hence, we may assume that $x \notin \mathrm{dom}(f_r)$. \bigskip \noindent \emph{Case 4:} For some $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$, $N \cap \alpha \in f_r(y) \cup \{ y \}$, and $f(y) = f_w(\alpha) \cup f_r(y)$. By Definition 4.2(6), $N \cap \alpha \in f_r(\alpha)$. Since $x \notin \mathrm{dom}(f_r)$, it follows that $x \in f_w(\alpha)$. In particular, $x \in N$, and therefore also $K \in N$. By Lemma 7.12(2), since $\alpha \in \mathrm{dom}(f_w)$, $f_w(\alpha) \subseteq f(\alpha)$. Hence, $x \in f(\alpha)$. We also know that $K \in f(x)$. Thus, by case 2 handled above, $$ g_r(K,\alpha) \subseteq x. $$ Since $K \in f_r(y)$, $N \cap \alpha \in f_r(y) \cup \{ y \}$, and $K \in N$, we have that $K \in f_r(N \cap \alpha)$. Therefore, $K \in f_r(\alpha)$. By Definition 4.2(5), it follows that $$ g_r(K,y) \subseteq g_r(K,N \cap \alpha). $$ Since $r \in D_N$, by Definition 7.3(3) we have that $$ g_r(K,N \cap \alpha) \subseteq g_r(K,\alpha). $$ Putting it all together, $$ g_r(K,y) \subseteq g_r(K,N \cap \alpha) \subseteq g_r(K,\alpha) \subseteq x. $$ \bigskip \noindent \emph{Case 5:} $y \in \mathrm{dom}(f_r) \setminus N$, $f_r(y) \cap N = \emptyset$, and $f(y) = f_r(y)$. Since $x \in f(y) = f_r(y)$, we have that $x \in \mathrm{dom}(f_r)$. But this contradicts the fact that $x \notin \mathrm{dom}(f_r)$. \bigskip \noindent \emph{Case 6:} $y \in \mathrm{dom}(f_r) \setminus N$, and $f(y) = f_r(y) \cup f_w(M)$, where $M$ is the membership largest element of $f_r(y) \cap N$. So either $x \in f_r(y)$ or $x \in f_w(M)$. Since $x \notin \mathrm{dom}(f_r)$, we have that $x \in f_w(M)$. Then $x \in Sk(M)$, and thus $K \in Sk(M)$. Since $K$ and $M$ are in $f_r(y)$ and $K \in Sk(M)$, it follows that $K \in f_r(M)$. Since $r$ is a condition, $g_r(K,y) \subseteq g_r(K,M)$. Now $x \in f_w(M)$, and $f(M) = f_w(M)$ by Definition 7.11(1). Thus, $K \in f(x)$ and $x \in f(M)$. By case 1 handled above, $g_r(K,M) \subseteq x$. Thus, $g_r(K,y) \subseteq g_r(K,M) \subseteq x$. \bigskip \noindent \emph{Case 7:} $y = M \cap \alpha$, where $M \in A_r$, $N \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$. By Lemma 7.9, $y$ is not in $\mathrm{dom}(f_r)$. But we assumed that $K \in f_r(y)$, so we have a contradiction. Thus, this case does not occur. \end{proof} We are ready to define the amalgam $w \oplus_N r$. \begin{definition} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $w \in N \cap \mathbb{P}$ and $w \le r \restriction N$. Define $w \oplus_N r$ as the object $(f,g,A)$ satisfying: \begin{enumerate} \item $f := f_w \oplus_N f_r$; \item $g := g_w \oplus_N g_r$; \item $A := A_w \cup A_r$. \end{enumerate} \end{definition} We will now show that $w \oplus_N r$ is a condition which is below $w$ and $r$. We have completed most of the work for this proof in the preceding lemmas. \begin{proposition} Let $N \in \mathcal X$ be simple and $r \in D_N$. Then for all $w \le r \restriction N$ in $N \cap \mathbb{P}$, $w$ and $r$ are compatible. In fact, $w \oplus_N r$ is in $\mathbb{P}$ and $w \oplus_N r \le w, r$. \end{proposition} \begin{proof} Let $w \oplus_N r = (f,g,A)$. We will prove that $w \oplus_N r$ is a condition which is below $w$ and $r$. To show that $w \oplus_N r$ is a condition, we verify requirements (1)--(7) of Definition 4.2. \bigskip (1) We apply Proposition 2.25. Since $r \in D_N$, we have that for all $M \in A_r$, if $M < N$ then $M \cap N \in A_r$. Also, $A_w$ is adequate, and by Lemma 7.8(1), $$ A_r \cap N \subseteq A_w \subseteq N. $$ By Proposition 2.25, $A_w \cup A_r = A$ is adequate. \bigskip (2,3) These statements are immediate from Lemmas 7.14 and 7.15. \bigskip (4) By Definition 7.16, $g$ is a function whose domain is the set of pairs $(K,x)$ such that $K \in f(x)$. And by Lemma 7.17, for all $(K,x) \in \mathrm{dom}(g)$, $g(K,x) \subseteq x \setminus \sup(K)$. Also, $g(K,x)$ is finite by Definition 7.16. \bigskip (5) Let $K \in f(L)$ and $L \in f(x)$. We will show that $g(K,x) \subseteq g(K,L)$. So let $\xi \in g(K,x)$. Then by Definition 7.16, either $\xi \in g_w(K,x) \cup g_r(K,x)$, or for some $y$ with $x \in f(y)$, $\xi \in g_w(K,y) \cup g_r(K,y)$. In the second case, $L \in f(x)$ and $x \in f(y)$ imply by requirement (3) that $L \in f(y)$. So letting $z := x$ in the first case, and $z := y$ in the second case, we have that $L \in f(z)$ and $\xi \in g_w(K,z) \cup g_r(K,z)$. By Definition 7.16, it follows that $\xi \in g(K,L)$. \bigskip (6) Let $\alpha \in \mathrm{dom}(f) \cap S$ and $M \in A$ with $\alpha \in M$. We will show that $M \cap \alpha \in f(\alpha)$. By the definition of $\mathrm{dom}(f)$ given in Definition 7.11, clearly $\alpha \in \mathrm{dom}(f_w) \cup \mathrm{dom}(f_r)$. Also, $A = A_w \cup A_r$, so either $M \in A_w$ or $M \in A_r$. First, assume that $\alpha \in \mathrm{dom}(f_r)$. If $M \in A_r$, then since $r$ is a condition, $M \cap \alpha \in f_r(\alpha) \subseteq f(\alpha)$. Suppose that $M \in A_w$. Then $\alpha \in M \in N$, so $\alpha \in \mathrm{dom}(f_r) \cap N \subseteq \mathrm{dom}(f_w)$ by Lemma 7.8(2). Since $w$ is a condition, $M \cap \alpha \in f_w(\alpha) \subseteq f(\alpha)$. Secondly, assume that $\alpha \in \mathrm{dom}(f_w) \setminus \mathrm{dom}(f_r)$. If $M \in A_w$, then since $w$ is a condition, $M \cap \alpha \in f_w(\alpha) \subseteq f(\alpha)$. Suppose that $M \in A_r$. If $N \le M$, then $M \cap \alpha \in f(\alpha)$ by Definition 7.11(3). Suppose that $M < N$. Then since $r \in D_N$, we have that $M \cap N \in A_r \cap N \subseteq A_w$ by Lemma 7.8(1). Since $\alpha \in \mathrm{dom}(f_w)$ and $\alpha \in M \cap N$, it follows that $M \cap N \cap \alpha \in f_w(\alpha) \subseteq f(\alpha)$. Now $\alpha \in M \cap N \cap \kappa$ implies that $\alpha < \beta_{M,N}$ by Proposition 2.11. Since $M < N$, Lemma 2.15 implies that $$ M \cap \alpha = M \cap \beta_{M,N} \cap \alpha = M \cap N \cap \alpha. $$ Thus, $M \cap \alpha \in f(\alpha)$. \bigskip (7) As in (1) above, the assumptions of Proposition 3.5 hold for $A_r$ and $A_w$. Therefore, $$ r^*(A) = r^*(A_w \cup A_r) = r^*(A_w) \cup r^*(A_r). $$ As $w$ and $r$ are conditions, $$ r^*(A_w) \cap S \subseteq \mathrm{dom}(f_w), \ \ \ r^*(A_r) \cap S \subseteq \mathrm{dom}(f_r). $$ But $\mathrm{dom}(f_w) \subseteq \mathrm{dom}(f)$ and $\mathrm{dom}(f_r) \subseteq \mathrm{dom}(f)$. Hence, $$ r^*(A) \cap S = (r^*(A_w) \cap S) \cup (r^*(A_r) \cap S) \subseteq \mathrm{dom}(f). $$ \bigskip This completes the proof that $w \oplus_N r$ is a condition. Now we show that $w \oplus_N r \le w, r$. First, we prove that $w \oplus_N r \le w$ by verifying properties (a)--(d) of Definition 4.2 for $w$. (a) Since $A = A_w \cup A_r$, clearly $A_w \subseteq A$. (b) follows from Lemma 7.12(1,2). (c) is immediate from Definition 7.16, and (d) was proved in Lemma 7.13(2). Secondly, we prove that $w \oplus_N r \le r$ by verifying properties (a)--(d) of Definition 4.2 for $r$. (a) Since $A = A_w \cup A_r$, clearly $A_r \subseteq A$. (b) follows from Lemma 7.12(1,3). (c) is immediate from Definition 7.16, and (d) was proved in Lemma 7.13(3). \end{proof} \begin{corollary} The forcing poset $\mathbb{P}$ is strongly proper on a stationary set. In particular, it preserves $\omega_1$. \end{corollary} \begin{proof} By Assumption 2.22, the set of $N \in \mathcal X$ such that $N$ is simple is stationary in $P_{\omega_1}(H(\lambda))$. So it suffices to show that for all simple $N \in \mathcal X$, for all $p \in N \cap \mathbb{P}$, there is $q \le p$ such that $q$ is strongly $N$-generic. Let $p \in N \cap \mathbb{P}$. By Lemma 7.1, fix $q \le p$ with $N \in A_q$. We claim that $q$ is strongly $N$-generic. So let $D$ be a dense subset of $N \cap \mathbb{P}$, and we will show that $D$ is predense below $q$. Let $r \le q$, and we will find $w \in D$ which is compatible with $r$. Since $N \in A_{r}$, we can apply Lemma 7.4 and fix $s \le r$ such that $s \in D_N$. Then by Lemma 7.6, $s \restriction N$ is in $N \cap \mathbb{P}$. As $D$ is dense in $N \cap \mathbb{P}$, fix $w \le s \restriction N$ in $D$. By Proposition 7.19, $w$ and $s$ are compatible. Since $s \le r$, it follows that $w$ and $r$ are compatible. \end{proof} It follows from Corollaries 6.16 and 7.20 that $\mathbb{P}$ preserves $\omega_1$ and $\kappa$. By the next lemma, no cardinal in between $\omega_1$ and $\kappa$ survives. \begin{lemma} If $\mu$ is a cardinal and $\omega_1 < \mu < \kappa$, then $\mathbb{P}$ forces that $\mu$ is not a cardinal. \end{lemma} \begin{proof} Let $G$ be a generic filter on $\mathbb{P}$. Define $$ A := \{ M : \exists p \in G \ M \in A_p, \ \mu \in M \}. $$ Consider $M$ and $N$ in $A$. Then there is $p \in G$ with $M$ and $N$ in $A_p$. Since $\mu \in M \cap N \cap \kappa$, $\mu < \beta_{M,N}$ by Proposition 2.11. If $M \cap \omega_1 = N \cap \omega_1$, then $M \sim N$ by Lemma 2.17, and therefore, $M \cap \beta_{M,N} = N \cap \beta_{M,N}$. Since $\mu < \beta_{M,N}$, it follows that $M \cap \mu = N \cap \mu$. This proves that the map which sends a member of the set $$ A^* := \{ M \cap \mu : M \in A \} $$ to its intersection with $\omega_1$ is an injective function from $A^*$ into $\omega_1$. Hence, in $V[G]$, $A^*$ has size less than or equal to $\omega_1$. A density argument using Lemma 7.1 shows that for all $\xi < \mu$, there is $N \in A$ with $\xi \in N$. It follows that $\bigcup A^* = \mu$. So in $V[G]$, $\mu$ is the union of a collection of countable sets of size at most $\omega_1$. This implies that in $V[G]$, $\mu$ has size at most $\omega_1$. Since $\omega_1 < \mu$, $\mu$ is not a cardinal in $V[G]$. \end{proof} \begin{corollary} The forcing poset $\mathbb{P}$ forces that $\kappa = \omega_2$. \end{corollary} \begin{proof} Immediate from Corollaries 6.16 and 7.20 and Lemma 7.21. \end{proof} \bigskip \addcontentsline{toc}{section}{8. Further analysis} \textbf{\S 8. Further analysis} \stepcounter{section} \bigskip The goal of the next two sections is to prove that certain quotients of the forcing poset $\mathbb{P}$ satisfy the $\omega_1$-approximation property. This fact will follow from the equation $$ (q \oplus_N p) \restriction Q = (q \restriction Q) \oplus_{N \cap Q} (p \restriction Q), $$ which is proved in Proposition 8.6. Lemmas 8.1 and 8.2 provide some additional information about $\mathbb{P}$ which we will need to prove the approximation property in Section 9. Then Lemmas 8.3, 8.4, and 8.5 prepare us for proving Proposition 8.6. The information which we provide here will be used again in Part III to analyze products of the partial square forcing poset. \begin{lemma} Let $N \in \mathcal X$ be simple and $r \in D_N$. Suppose that $v$ and $w$ are in $N \cap \mathbb{P}$ and $$ w \le v \le r \restriction N. $$ Then $w \oplus_N r \le v \oplus_N r$. \end{lemma} The proof of this lemma is straightforward, but due to the multitude of cases in Definition 7.11, it is also somewhat lengthy. \begin{proof} Let $s := v \oplus_N r$ and $t := w \oplus_N r$. We will prove that $t \le s$. We verify properties (a)--(d) of Definition 4.2 for $s$ and $t$. \bigskip (a) By Definition 7.18, $A_s = A_v \cup A_r$ and $A_t = A_w \cup A_r$. Since $w \le v$, $A_v \subseteq A_w$. Therefore, $A_s \subseteq A_t$. \bigskip (b,d) The domain of $f_s$ is equal to the union of $\mathrm{dom}(f_v)$, $\mathrm{dom}(f_r)$, and the set $$ \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in (M \cap \mathrm{dom}(f_v) \cap S) \setminus \mathrm{dom}(f_r) \}. $$ The domain of $f_t$ is equal to the union of $\mathrm{dom}(f_w)$, $\mathrm{dom}(f_r)$, and the set $$ \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r) \}. $$ But $w \le v$ implies that $\mathrm{dom}(f_v) \subseteq \mathrm{dom}(f_w)$. It is easy to check from the above definitions and the fact that $\mathrm{dom}(f_v) \subseteq \mathrm{dom}(f_w)$ that $\mathrm{dom}(f_s) \subseteq \mathrm{dom}(f_t)$. \bigskip Let $x \in \mathrm{dom}(f_s)$, and we will prove that $f_s(x) \subseteq f_t(x)$. At the same time, we will also show that if $K$ is in $\mathrm{dom}(f_s)$ and $K \in f_t(x)$, then $K \in f_s(x)$. Note that these assertions imply (b) and (d). The proof will split into the seven cases of Definition 7.11 for how $f_s(x)$ is defined. First, we prove a claim. \bigskip \noindent \emph{Claim 1:} If $K \in \mathrm{dom}(f_s)$ and $K \in f_w(x)$, then $K \in f_s(x)$. \bigskip Since $K \in f_w(x)$, $K$ and $x$ are in $N$. By Lemma 7.13(1), $K$ and $x$ are in $\mathrm{dom}(f_s) \cap N = \mathrm{dom}(f_v)$. So $K$ and $x$ are in $\mathrm{dom}(f_v)$ and $K \in f_w(x)$. Since $w \le v$, it follows that $K \in f_v(x)$. But $s \le v$ implies that $f_v(x) \subseteq f_s(x)$. So $K \in f_s(x)$. \bigskip \noindent \emph{Case 1:} $x \in \mathrm{dom}(f_v) \setminus S$ and $f_s(x) = f_v(x)$. Since $w \le v$, $x \in \mathrm{dom}(f_w) \setminus S$, and so $f_t(x) = f_w(x)$. But $w \le v$ implies that $f_v(x) \subseteq f_w(x)$, hence $f_s(x) \subseteq f_t(x)$. Suppose that $K \in \mathrm{dom}(f_s)$ and $K \in f_t(x)$. Then $K \in f_t(x) = f_w(x)$. So $K \in f_s(x)$ by Claim 1. \bigskip \noindent \emph{Case 2:} $x = \alpha \in \mathrm{dom}(f_v) \cap S \cap \mathrm{dom}(f_r)$ and $f_s(\alpha) = f_v(\alpha) \cup f_r(\alpha)$. Since $\mathrm{dom}(f_v) \subseteq \mathrm{dom}(f_w)$, $\alpha \in \mathrm{dom}(f_w) \cap S \cap \mathrm{dom}(f_r)$, so $f_t(\alpha) = f_w(\alpha) \cup f_r(\alpha)$. Also, $w \le v$ implies that $f_v(\alpha) \subseteq f_w(\alpha)$. Therefore, $f_s(\alpha) \subseteq f_t(\alpha)$. Suppose that $K \in \mathrm{dom}(f_s)$ and $K \in f_t(\alpha)$. Then $K \in f_t(\alpha) = f_w(\alpha) \cup f_r(\alpha)$. So either $K \in f_w(\alpha)$ or $K \in f_r(\alpha)$. If $K \in f_w(\alpha)$, then $K \in f_s(\alpha)$ by Claim 1. If $K \in f_r(\alpha)$, then $K \in f_s(\alpha)$ by definition. \bigskip \noindent \emph{Case 3:} $x = \alpha \in (\mathrm{dom}(f_v) \cap S) \setminus \mathrm{dom}(f_r)$ and $$ f_s(\alpha) = f_v(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}. $$ Since $\mathrm{dom}(f_v) \subseteq \mathrm{dom}(f_w)$, $\alpha \in (\mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$, and therefore $$ f_t(\alpha) = f_w(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}. $$ Since $f_v(\alpha) \subseteq f_w(\alpha)$, it follows that $f_s(\alpha) \subseteq f_t(\alpha)$. Suppose that $K \in \mathrm{dom}(f_s)$ and $K \in f_t(\alpha)$. Then $K \in f_t(\alpha) = f_w(\alpha) \cup \{ M \cap \alpha : M \in A_r, \ N \le M, \ \alpha \in M \}$. If $K$ is in the second set of this union, then $K \in f_s(\alpha)$ by definition. If $K \in f_w(\alpha)$, then $K \in f_s(\alpha)$ by Claim 1. \bigskip \noindent \emph{Case 4:} $x \in \mathrm{dom}(f_r) \setminus N$, and for some $\alpha \in \mathrm{dom}(f_r) \cap S \cap N$, $N \cap \alpha \in f_r(x) \cup \{ x \}$. Then $f_s(x) = f_v(\alpha) \cup f_r(x)$. Clearly we are also in case 4 in the definition of $f_t(x)$. So $f_t(x) = f_w(\alpha) \cup f_r(x)$. Since $w \le v$, $f_v(\alpha) \subseteq f_w(\alpha)$. Therefore, $f_s(x) \subseteq f_t(x)$. Suppose that $K \in \mathrm{dom}(f_s)$ and $K \in f_t(x)$. Then $K \in f_w(\alpha) \cup f_r(x)$. If $K \in f_r(x)$, then $K \in f_s(x)$ by definition. Suppose that $K \in f_w(\alpha)$. Then $K$ is in $N$. So $K$ and $\alpha$ are in $\mathrm{dom}(f_s) \cap N = \mathrm{dom}(f_v)$ by Lemma 7.13(1). Since $K \in f_w(\alpha)$ and $w \le v$, it follows that $K \in f_v(\alpha)$. So $K \in f_s(x)$ by definition. \bigskip \noindent \emph{Case 5:} $x \in \mathrm{dom}(f_r) \setminus N$, case 4 is false, and $f_r(x) \cap N = \emptyset$. Then clearly we are also in case 5 in the definition of $f_t(x)$. So by definition, $f_s(x)$ and $f_t(x)$ are both equal to $f_r(x)$. In particular, $f_s(x) \subseteq f_t(x)$. Also, if $K \in \mathrm{dom}(f_s)$ and $K \in f_t(x)$, then $K \in f_t(x) = f_s(x)$. \bigskip \noindent \emph{Case 6:} $x \in \mathrm{dom}(f_r) \setminus N$, case 4 is false, and $f_s(x) = f_r(x) \cup f_v(M)$, where $M$ is the membership largest element of $f_r(x) \cap N$. Then we are obviously also in case 6 in the definition of $f_t$. So $f_t(x) = f_r(x) \cup f_w(M)$. Since $w \le v$, we have that $f_v(M) \subseteq f_w(M)$. Therefore, $f_s(x) \subseteq f_t(x)$. Suppose that $K \in \mathrm{dom}(f_s)$ and $K \in f_t(x)$. Then $K \in f_t(x) = f_r(x) \cup f_w(M)$. If $K \in f_r(x)$, then $K \in f_s(x)$ by definition. Otherwise, $K \in f_w(M)$. Since $M \in \mathrm{dom}(f_r) \subseteq \mathrm{dom}(f_s)$, Claim 1 implies that $K \in f_s(M)$. But $M \in f_r(x) \subseteq f_s(x)$. So $K \in f_s(M)$ and $M \in f_s(x)$. Since $s$ is a condition, $K \in f_s(x)$. \bigskip \noindent \emph{Case 7:} $x = M \cap \alpha$, where $M \in A_r$, $N \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_v) \cap S) \setminus \mathrm{dom}(f_r)$. Then $$ f_s(M \cap \alpha) = f_s(\alpha) \cap Sk(M \cap \alpha). $$ Since $w \le v$, $\alpha \in (M \cap \mathrm{dom}(f_w) \cap S) \setminus \mathrm{dom}(f_r)$. So we are also in case 7 in the definition of $f_t(x)$. Therefore, $$ f_t(M \cap \alpha) = f_t(\alpha) \cap Sk(M \cap \alpha). $$ Since $f_s(\alpha) \subseteq f_t(\alpha)$ by case 3 handled above, it follows that $f_s(M \cap \alpha) \subseteq f_t(M \cap \alpha)$. Suppose that $K \in \mathrm{dom}(f_s)$ and $K \in f_t(M \cap \alpha)$. Then by the definition of $f_t(M \cap \alpha)$, $K \in f_t(\alpha)$ and $K \in Sk(M \cap \alpha)$. Since $\alpha \in \mathrm{dom}(f_v)$ and $s \le v$, $\alpha \in \mathrm{dom}(f_s)$. By case 3 handled above, it follows that $K \in f_s(\alpha)$. So $K \in f_s(\alpha) \cap Sk(M \cap \alpha) = f_s(M \cap \alpha)$. This completes the proof of (b) and (d). \bigskip (c) Let $(K,x) \in \mathrm{dom}(g_s)$, and we will show that $g_s(K,x) \subseteq g_t(K,x)$. By Definition 7.16, $K \in f_s(x)$ and $$ g_s(K,x) = \bigcup \{ g_v(K,y) \cup g_r(K,y) : x = y, \textrm{or} \ x \in f_s(y) \}. $$ Since $f_s(x) \subseteq f_t(x)$ as just shown, $K \in f_t(x)$, and by Definition 7.16, $$ g_t(K,x) = \bigcup \{ g_w(K,y) \cup g_r(K,y) : x = y, \textrm{or} \ x \in f_t(y) \}. $$ Let $\xi \in g_s(K,x)$. To show that $\xi \in g_t(K,x)$, we will consider the different possibilities for why $\xi$ is in $g_s(K,x)$. \bigskip \noindent \emph{Case 1:} $\xi \in g_v(K,x)$. Then $(K,x) \in \mathrm{dom}(g_v)$, which means that $K \in f_v(x)$. Since $w \le v$, $K \in f_w(x)$ and $g_v(K,x) \subseteq g_w(K,x)$. So $\xi \in g_w(K,x) \subseteq g_t(K,x)$. \bigskip \noindent \emph{Case 2:} $\xi \in g_v(K,y)$, where $x \in f_s(y)$. Then $(K,y) \in \mathrm{dom}(g_v)$, which means that $K \in f_v(y)$. Since $w \le v$, $K \in f_w(y)$ and $g_v(K,y) \subseteq g_w(K,y)$. So $\xi \in g_w(K,y)$. But $x \in f_s(y) \subseteq f_t(y)$. So by definition, $g_w(K,y) \subseteq g_t(K,x)$. So $\xi \in g_t(K,x)$. \bigskip \noindent \emph{Case 3:} $\xi \in g_r(K,x)$. Since $t \le r$, it follows that $g_r(K,x) \subseteq g_t(K,x)$, so $\xi \in g_t(K,x)$. \bigskip \noindent \emph{Case 4:} $\xi \in g_r(K,y)$, where $x \in f_s(y)$. By (b), $f_s(y) \subseteq f_t(y)$. So $x \in f_t(y)$. Since $\xi \in g_r(K,y)$ and $x \in f_t(y)$, it follows by definition that $\xi \in g_t(K,x)$. \end{proof} \begin{lemma} Let $N \in \mathcal X$ and $Q \in \mathcal Y$. Let $p \in \mathbb{P}$, and suppose that $N \in A_p$. Then there is $s \le p$ such that $s \in D_N \cap D_Q$. \end{lemma} \begin{proof} We construct $s$ in several steps. We begin by applying Lemma 7.2 to find $q \le p$ such that for all $M \in A_q$, if $M < N$ then $M \cap N \in A_q$. Next, we apply Lemma 6.1 to fix $r \le q$ such that for all $M \in A_r$, $M \cap Q \in A_r$, and moreover, $A_r = A_q \cup \{ M \cap Q : M \in A_q \}$. We claim that for all $M \in A_r$, if $M < N$ then $M \cap N \in A_r$. This is certainly true if $M \in A_q$, so assume that $M = M_1 \cap Q$, where $M_1 \in A_q$. By Lemma 2.29, $M_1 \sim M_1 \cap Q = M$. Since $M < N$, it follows that $M_1 < N$ by Lemma 2.18. As $M_1 \in A_q$, we have that $M_1 \cap N \in A_q$ by the choice of $q$. Hence, $$ M \cap N = (M_1 \cap Q) \cap N = (M_1 \cap N) \cap Q. $$ But $M_1 \cap N \in A_q$ implies that $M \cap N = (M_1 \cap N) \cap Q \in A_r$ by the definition of $A_r$. Now apply Lemma 4.9 to find $s \le r$ such that $A_s = A_r$ and whenever $K \in f_s(x)$ and $x \in f_s(y)$, then $g_s(K,x) \subseteq g_s(K,y)$. Then $s \le p$ and $s \in D_N \cap D_Q$. \end{proof} The next three lemmas will prepare us for proving Proposition 8.6. \begin{lemma} Let $N \in \mathcal X$ be simple, $Q \in \mathcal Y \cap N$ be simple, and $p \in D_N \cap D_Q$. Then $p \restriction N \in D_Q$ and $p \restriction Q \in D_{N \cap Q}$. \end{lemma} Recall that by Lemma 2.30, $N \cap Q$ is simple. \begin{proof} We prove first that $p \restriction N \in D_Q$, which by Definition 6.2 means that for all $M \in A_{p \restriction N}$, $M \cap Q \in A_{p \restriction N}$. So let $M \in A_{p \restriction N}$. Then by Definition 7.5, $M \in A_{p \restriction N} = A_p \cap N$. As $p \in D_Q$, we have that $M \cap Q \in A_p$. And since $M$ and $Q$ are in $N$, $M \cap Q \in N$. Thus, $M \cap Q \in A_p \cap N = A_{p \restriction N}$. Next, we prove that $p \restriction Q \in D_{N \cap Q}$. First, we show that $N \cap Q \in A_{p \restriction Q}$. Since $A_{p \restriction Q} = A_p \cap Q$ by Definition 6.4, we need to show that $N \cap Q \in A_p \cap Q$. Since $Q$ is simple, $N \cap Q \in Q$. As $p \in D_N$, $N \in A_p$, and since $p \in D_Q$, it follows that $N \cap Q \in A_p$. So $N \cap Q \in A_p \cap Q$. Secondly, let $M \in A_{p \restriction Q} = A_p \cap Q$ be such that $M < N \cap Q$, and we will show that $$ M \cap (N \cap Q) \in A_{p \restriction Q} = A_p \cap Q. $$ Since $M \cap N \in \mathcal X$ and $Q$ is simple, we have that $M \cap N \cap Q \in Q$. By Lemma 2.29, $N \sim N \cap Q$. Since $M < N \cap Q$, Lemma 2.18 implies that $M < N$. As $p \in D_N$ and $M \in A_p$, it follows that $M \cap N \in A_p$. Since $p \in D_Q$, $M \cap N \cap Q \in A_p$. So $M \cap N \cap Q \in A_p \cap Q$, as required. Thirdly, let $K \in f_{p \restriction Q}(x)$ and $x \in f_{p \restriction Q}(y)$, and we will show that $g_{p \restriction Q}(K,x) \subseteq g_{p \restriction Q}(K,y)$. By the definition of $p \restriction Q$, we have that $K \in f_p(x)$ and $x \in f_p(y)$. Since $p \in D_N$, $g_p(K,x) \subseteq g_p(K,y)$. As $g_p(K,x) = g_{p \restriction Q}(K,x)$ and $g_p(K,y) = g_{p \restriction Q}(K,y)$ by the definition of $p \restriction Q$, we are done. \end{proof} \begin{lemma} Let $N \in \mathcal X$ be simple, $Q \in \mathcal Y \cap N$ be simple, and $p \in D_N \cap D_Q$. Then $$ (p \restriction N) \restriction Q = (p \restriction Q) \restriction (N \cap Q). $$ \end{lemma} Note that we needed Lemma 8.3 to see that $(p \restriction N) \restriction Q$ and $(p \restriction Q) \restriction (N \cap Q)$ are defined. \begin{proof} By Definitions 6.4 and 7.5, we have that \begin{multline*} \mathrm{dom}(f_{(p \restriction N) \restriction Q}) = \mathrm{dom}(f_{p \restriction N}) \cap Q = \mathrm{dom}(f_p) \cap N \cap Q = \\ = (\mathrm{dom}(f_p) \cap Q) \cap (N \cap Q) = \mathrm{dom}(f_{p \restriction Q}) \cap (N \cap Q) = \mathrm{dom}(f_{(p \restriction Q) \restriction (N \cap Q)}). \end{multline*} And for each $x \in \mathrm{dom}(f_{(p \restriction N) \restriction Q})$, $$ f_{(p \restriction N) \restriction Q}(x) = f_{p \restriction N}(x) = f_{p}(x) \cap N = f_{p \restriction Q}(x) \cap N, $$ which, since $f_{p \restriction Q}(x) \subseteq Q$, is equal to $$ f_{p \restriction Q}(x) \cap (N \cap Q) = f_{(p \restriction Q) \restriction (N \cap Q)}(x). $$ Thus, $f_{(p \restriction N) \restriction Q} = f_{(p \restriction Q) \restriction (N \cap Q)}$. Again by Definitions 6.4 and 7.5, we have that \begin{multline*} \mathrm{dom}(g_{(p \restriction N) \restriction Q}) = \mathrm{dom}(g_{p \restriction N}) \cap Q = \mathrm{dom}(g_p) \cap N \cap Q = \\ = (\mathrm{dom}(g_p) \cap Q) \cap (N \cap Q) = \mathrm{dom}(g_{p \restriction Q}) \cap (N \cap Q) = \mathrm{dom}(g_{(p \restriction Q) \restriction (N \cap Q)}). \end{multline*} And for each $(K,x) \in \mathrm{dom}(g_{(p \restriction N) \restriction Q})$, $$ g_{(p \restriction N) \restriction Q}(K,x) = g_{p \restriction N}(K,x) = g_p(K,x) = g_{p \restriction Q}(K,x) = g_{(p \restriction Q) \restriction (N \cap Q)}(K,x). $$ This proves that $g_{(p \restriction N) \restriction Q} = g_{(p \restriction Q) \restriction (N \cap Q)}$. Finally, by Definitions 6.4 and 7.5, \begin{multline*} A_{(p \restriction N) \restriction Q} = A_{p \restriction N} \cap Q = A_p \cap N \cap Q = \\ = (A_p \cap Q) \cap (N \cap Q) = A_{p \restriction Q} \cap (N \cap Q) = A_{(p \restriction Q) \restriction (N \cap Q)}. \end{multline*} \end{proof} \begin{lemma} Let $N \in \mathcal X$ be simple, $Q \in \mathcal Y \cap N$ be simple, and $p \in D_N \cap D_Q$. Suppose that $q \in N \cap D_Q$ and $q \le p \restriction N$. Then: \begin{enumerate} \item $q \oplus_{N} p$ is in $D_Q$; \item $q \restriction Q \in N \cap Q$ and $$ q \restriction Q \le (p \restriction Q) \restriction (N \cap Q). $$ \end{enumerate} \end{lemma} \begin{proof} (1) Since $q \le p \restriction N$, $q \oplus_N p$ is a condition which is below $q$ and $p$. We claim that $q \oplus_N p$ is in $D_Q$, which means that for all $M \in A_{q \oplus_N p}$, $M \cap Q \in A_{q \oplus_N p}$. Now $A_{q \oplus_N p} = A_q \cup A_p$. So if $M \in A_{q \oplus_N p}$, then either $M \in A_q$ or $M \in A_p$. But $q$ and $p$ are both in $D_Q$, so in the first case, $M \cap Q \in A_q$, and in the second case, $M \cap Q \in A_p$. In either case, $M \cap Q \in A_q \cup A_p = A_{q \oplus_N p}$. (2) Since $q$ and $Q$ are in $N$, $q \restriction Q \in N$. Also, $q \restriction Q \in Q$, so $q \restriction Q \in N \cap Q$. By Lemmas 8.3 and 8.4, $p \restriction N \in D_Q$ and $$ (p \restriction N) \restriction Q = (p \restriction Q) \restriction (N \cap Q). $$ Now $q \le p \restriction N$, so by Lemma 6.6(3), we have that $$ q \restriction Q \le (p \restriction N) \restriction Q = (p \restriction Q) \restriction (N \cap Q). $$ \end{proof} \begin{proposition} Let $N \in \mathcal X$ be simple, $Q \in \mathcal Y \cap N$ be simple, and suppose that $Q \cap \kappa \notin S$. Let $p \in D_N \cap D_Q$. Assume that $q \in N \cap D_Q$ and $q \le p \restriction N$. Then $$ (q \oplus_N p) \restriction Q = (q \restriction Q) \oplus_{N \cap Q} (p \restriction Q). $$ \end{proposition} Note that Lemma 8.5 implies that $(q \oplus_N p) \restriction Q$ and $(q \restriction Q) \oplus_{N \cap Q} (p \restriction Q)$ are defined. \begin{proof} Let $$ s := (q \oplus_N p) \restriction Q $$ and $$ t := (q \restriction Q) \oplus_{N \cap Q} (p \restriction Q). $$ Our goal is to prove that $s = t$. The proof will be split into three steps, namely, showing that $A_s = A_t$, $f_s = f_t$, and $g_s = g_t$. \bigskip By Definitions 6.4 and 7.18, we have that \begin{multline*} A_s = A_{(q \oplus_N p) \restriction Q} = A_{q \oplus_N p} \cap Q = (A_q \cup A_p) \cap Q = \\ = (A_q \cap Q) \cup (A_p \cap Q) = A_{q \restriction Q} \cup A_{p \restriction Q} = A_{(q \restriction Q) \oplus_{N \cap Q} (p \restriction Q)} = A_t. \end{multline*} So $A_s = A_t$. \bigskip We begin the proof of $f_s = f_t$ by showing that $\mathrm{dom}(f_t) \subseteq \mathrm{dom}(f_s)$. So let $x \in \mathrm{dom}(f_t)$, and we will show that $x \in \mathrm{dom}(f_s)$. By Definition 7.11, either \begin{enumerate} \item[(a)] $x \in \mathrm{dom}(f_{q \restriction Q})$, or \item[(b)] $x \in \mathrm{dom}(f_{p \restriction Q})$, or \item[(c)] $x = M \cap \alpha$, where $M \in A_{p \restriction Q}$, $N \cap Q \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_{q \restriction Q}) \cap S) \setminus \mathrm{dom}(f_{p \restriction Q})$. \end{enumerate} By Definition 6.4, the domain of $f_s$ is equal to $\mathrm{dom}(f_{q \oplus_N p}) \cap Q$. So it suffices to show that $x \in \mathrm{dom}(f_{q \oplus_N p})$ and $x \in Q$. We consider cases a, b, and c separately. (a) $x \in \mathrm{dom}(f_{q \restriction Q}) = \mathrm{dom}(f_q) \cap Q$. So $x \in Q$. Also, $\mathrm{dom}(f_q) \subseteq \mathrm{dom}(f_{q \oplus_N p})$, so $x \in \mathrm{dom}(f_{q \oplus_N p})$. (b) $x \in \mathrm{dom}(f_{p \restriction Q}) = \mathrm{dom}(f_p) \cap Q$. Then $x \in Q$. Also, $\mathrm{dom}(f_p) \subseteq \mathrm{dom}(f_{q \oplus_N p})$, so $x \in \mathrm{dom}(f_{q \oplus_N p})$. (c) $x = M \cap \alpha$, where $M \in A_{p \restriction Q}$, $N \cap Q \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_{q \restriction Q}) \cap S) \setminus \mathrm{dom}(f_{p \restriction Q})$. Then $M \in A_{p \restriction Q} = A_p \cap Q$ and $\alpha \in \mathrm{dom}(f_{q \restriction Q}) = \mathrm{dom}(f_q) \cap Q$. In particular, $M$ and $\alpha$ are in $Q$, so $M \cap \alpha = x$ is in $Q$. It remains to show that $M \cap \alpha \in \mathrm{dom}(f_{q \oplus_N p})$. By the definition of the domain of $f_{q \oplus_N p}$ in Definition 7.11, it suffices to show that $M \in A_p$, $N \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_q) \cap S) \setminus \mathrm{dom}(f_p)$. We know that $M \in A_p$ from the last paragraph. By Lemma 2.29, $N \sim N \cap Q$, and since $N \cap Q \le M$, it follows by Lemma 2.18 that $N \le M$. In the previous paragraph, we observed that $\alpha \in \mathrm{dom}(f_q) \cap Q$. Also, by the choice of $\alpha$, $\alpha \in M \cap S$ and $\alpha \notin \mathrm{dom}(f_{p \restriction Q}) = \mathrm{dom}(f_p) \cap Q$. Since $\alpha \in Q$, the latter statement implies that $\alpha \notin \mathrm{dom}(f_p)$. To summarize, $\alpha \in (M \cap \mathrm{dom}(f_q) \cap S) \setminus \mathrm{dom}(f_p)$, as required. This completes the proof that $\mathrm{dom}(f_t) \subseteq \mathrm{dom}(f_s)$. To finish the proof that $f_s = f_t$, we show that if $x \in \mathrm{dom}(f_s)$, then $x \in \mathrm{dom}(f_t)$ and $f_s(x) = f_t(x)$. Let $x \in \mathrm{dom}(f_s)$. Then $$ x \in \mathrm{dom}(f_s) = \mathrm{dom}(f_{q \oplus_N p}) \cap Q $$ and $$ f_s(x) = f_{(q \oplus_N p) \restriction Q}(x) = f_{q \oplus_N p}(x). $$ Thus, it suffices to prove that $x \in \mathrm{dom}(f_t)$ and $$ f_{q \oplus_N p}(x) = f_{(q \restriction Q) \oplus_{N \cap Q} (p \restriction Q)}(x). $$ The proof splits into the seven cases of Definition 7.11 for how $f_{q \oplus_N p}(x)$ is defined. We remind the reader that $x \in Q$, as just noted. \bigskip (1) $x \in \mathrm{dom}(f_q) \setminus S$ and $f_s(x) = f_q(x)$. Then $$ x \in \mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q}) \subseteq \mathrm{dom}(f_t). $$ Also, $x \in \mathrm{dom}(f_{q \restriction Q}) \setminus S$ implies that $f_t(x)$ is defined as in case 1 of Definition 7.11. So $f_t(x) = f_{q \restriction Q}(x) = f_q(x) = f_s(x)$. \bigskip (2) $x = \alpha \in \mathrm{dom}(f_q) \cap S \cap \mathrm{dom}(f_p)$ and $f_s(\alpha) = f_q(\alpha) \cup f_p(\alpha)$. Then $\alpha \in \mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q}) \subseteq \mathrm{dom}(f_t)$ and $\alpha \in \mathrm{dom}(f_p) \cap Q = \mathrm{dom}(f_{p \restriction Q})$. Hence, $$ \alpha \in \mathrm{dom}(f_{q \restriction Q}) \cap S \cap \mathrm{dom}(f_{p \restriction Q}), $$ and therefore $f_t(\alpha)$ is defined as in case 2 of Definition 7.11. So $$ f_t(\alpha) = f_{q \restriction Q}(\alpha) \cup f_{p \restriction Q}(\alpha) = f_q(\alpha) \cup f_p(\alpha) = f_s(\alpha). $$ \bigskip (3) $x = \alpha \in (\mathrm{dom}(f_q) \cap S) \setminus \mathrm{dom}(f_p)$ and $$ f_s(x) = f_q(\alpha) \cup \{ M \cap \alpha : M \in A_p, \ N \le M, \ \alpha \in M \}. $$ Then $\alpha \in \mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q}) \subseteq \mathrm{dom}(f_t)$. Also, $\alpha \notin \mathrm{dom}(f_p)$ implies that $\alpha \notin \mathrm{dom}(f_p) \cap Q = \mathrm{dom}(f_{p \restriction Q})$. To summarize, we have that $\alpha \in (\mathrm{dom}(f_{q \restriction Q}) \cap S) \setminus \mathrm{dom}(f_{p \restriction Q})$, which means that we are in case 3 in the definition of $f_t(\alpha)$. So $$ f_t(\alpha) = f_{q \restriction Q}(\alpha) \cup \{ K \cap \alpha : K \in A_{p \restriction Q}, \ N \cap Q \le K, \ \alpha \in K \}. $$ Since $f_{q \restriction Q}(\alpha) = f_q(\alpha)$, in order to show that $f_s(x) = f_t(x)$, the above equations imply that it suffices to show that \begin{multline*} \{ M \cap \alpha : M \in A_p, \ N \le M, \ \alpha \in M \} = \\ = \{ K \cap \alpha : K \in A_{p \restriction Q}, \ N \cap Q \le K, \ \alpha \in K \}. \end{multline*} Let $K \cap \alpha$ be in the set on the right, where $K \in A_{p \restriction Q}$, $N \cap Q \le K$, and $\alpha \in K$. Then $K \in A_{p \restriction Q} = A_p \cap Q$. By Lemma 2.29, $N \sim N \cap Q$. Since $N \cap Q \le K$, Lemma 2.18 implies that $N \le K$. So $K \in A_p$, $N \le K$, and $\alpha \in K$. Thus, $K \cap \alpha$ is in the set on the left. Conversely, let $M \cap \alpha$ be a member of the set on the left, where $M \in A_p$, $N \le M$, and $\alpha \in M$. We will show that $M \cap \alpha$ is a member of the set on the right. It suffices to show that $M \cap \alpha$ is equal to $K \cap \alpha$, for some $K \in A_{p \restriction Q}$ such that $N \cap Q \le K$ and $\alpha \in K$. Let $K := M \cap Q$. Since $p \in D_Q$ and $M \in A_p$, $M \cap Q \in A_p$. As $Q$ is simple, $M \cap Q \in Q$. So $K = M \cap Q \in A_p \cap Q = A_{p \restriction Q}$. By Lemma 2.29, $M \sim M \cap Q$ and $N \sim N \cap Q$. Since $N \le M$, Lemma 2.18 implies that $N \cap Q \le M \cap Q = K$. And $\alpha \in Q$, so $\alpha \in M \cap Q = K$. It remains to show that $M \cap \alpha = K \cap \alpha$. But $\alpha \in Q$ implies that $K \cap \alpha = M \cap Q \cap \alpha = M \cap \alpha$. \bigskip (4) $x \in \mathrm{dom}(f_p) \setminus N$, and for some $\alpha \in \mathrm{dom}(f_p) \cap S \cap N$, $N \cap \alpha \in f_p(x) \cup \{ x \}$. Then $f_s(x) = f_q(\alpha) \cup f_p(x)$. We claim that $\alpha \in Q$. Suppose for a contradiction that $\alpha \notin Q$. Since $\alpha \in S$ and $Q \cap \kappa \notin S$, it follows that $Q \cap \kappa < \alpha$. As $N \cap \alpha \in f_p(x) \cup \{ x \}$, $N \cap \alpha$ is either equal to $x$ or is in $f_p(x)$. In particular, since $x \in Q$, we have that $$ \sup(N \cap \alpha) \le \sup(x) < Q \cap \kappa < \alpha. $$ So $\sup(N \cap \alpha) < Q \cap \kappa$. But $Q \in N$, so $Q \cap \kappa \in N \cap \alpha$, which contradicts that $\sup(N \cap \alpha) < Q \cap \kappa$. This completes the proof that $\alpha \in Q$. Now we argue that we are in case 4 of Definition 7.11 for $f_t(x)$. We know that $\alpha$ and $x$ are in $\mathrm{dom}(f_p) \cap Q = \mathrm{dom}(f_{p \restriction Q})$. So $x \in \mathrm{dom}(f_{p \restriction Q}) \setminus (N \cap Q)$, $\alpha \in \mathrm{dom}(f_{p \restriction Q}) \cap S \cap (N \cap Q)$, and $N \cap \alpha \in f_p(x) \cup \{ x \}$. Since $\alpha \in Q$, $N \cap \alpha = N \cap Q \cap \alpha$, so $N \cap Q \cap \alpha \in f_p(x) \cup \{ x \} = f_{p \restriction Q}(x) \cup \{ x \}$. This proves that we are in case 4 in the definition of $f_t(x)$. So $$ f_t(x) = f_{q \restriction Q}(\alpha) \cup f_{p \restriction Q}(x) = f_q(\alpha) \cup f_p(x) = f_s(x). $$ \bigskip (5,6) $x \in \mathrm{dom}(f_p) \setminus N$ and case 4 fails. Then $x \in \mathrm{dom}(f_p) \cap Q = \mathrm{dom}(f_{p \restriction Q})$, so $x \in \mathrm{dom}(f_{p \restriction Q}) \setminus (N \cap Q)$. In particular, $x \in \mathrm{dom}(f_t)$. We claim that case 4 of Definition 7.11 fails for $f_t(x)$. Suppose for a contradiction that for some $\alpha \in \mathrm{dom}(f_{p \restriction Q}) \cap S \cap (N \cap Q)$, we have that $(N \cap Q) \cap \alpha \in f_{p \restriction Q}(x) \cup \{ x \}$. Then $\alpha \in \mathrm{dom}(f_{p \restriction Q}) = \mathrm{dom}(f_p) \cap Q$. Since $\alpha \in Q$, $N \cap Q \cap \alpha = N \cap \alpha$. So $\alpha \in \mathrm{dom}(f_p) \cap S \cap N$ and $N \cap \alpha \in f_{p \restriction Q}(x) \cup \{ x \} = f_p(x) \cup \{ x \}$. Hence, we are in case 4 of Definition 7.11 in the definition of $f_{q \oplus_N p}(x)$, which is a contradiction. It follows that we are in either case 5 or 6 in the definition of $f_t(x)$. Assume that we are in case 5 in the definition of $f_{q \oplus_N p}(x)$. Then $f_p(x) \cap N = \emptyset$ and $f_s(x) = f_p(x)$. It follows that $$ f_{p \restriction Q}(x) \cap (N \cap Q) = f_p(x) \cap (N \cap Q) \subseteq f_p(x) \cap N = \emptyset. $$ So we are also in case 5 in the definition of $f_t(x)$. Hence, $$ f_t(x) = f_{p \restriction Q}(x) = f_p(x) = f_s(x). $$ Now assume that we are in case 6 in the definition of $f_{q \oplus_N p}(x)$. Then $f_s(x) = f_p(x) \cup f_q(M)$, where $M$ is the membership largest element of $f_p(x) \cap N$. Since $M \in f_p(x)$, it follows that $M \in Sk(x)$. And since $x \in Q$, $M \in Q$. Thus, $M \in f_p(x) \cap Q = f_{p \restriction Q}(x) \cap Q$. Since $M \in N$, $M$ is in $f_{p \restriction Q}(x) \cap (N \cap Q)$. In particular, $f_{p \restriction Q}(x) \cap (N \cap Q)$ is nonempty, so we are in case 6 in the definition of $f_t(x)$. We claim that $M$ is the membership largest element of $f_{p \restriction Q}(x) \cap (N \cap Q)$. Otherwise there is $M_1 \in f_{p \restriction Q}(x) \cap (N \cap Q)$ such that $M \in Sk(M_1)$. But then $$ M_1 \in f_{p \restriction Q}(x) \cap (N \cap Q) = f_p(x) \cap N \cap Q \subseteq f_p(x) \cap N. $$ Since $M$ is the membership largest element of $f_p(x) \cap N$, this is a contradiction. By the definition of $f_t(x)$, we have that $$ f_t(x) = f_{p \restriction Q}(x) \cup f_{q \restriction Q}(M) = f_p(x) \cup f_q(M) = f_s(x). $$ \bigskip (7) $x = M \cap \alpha$, where $M \in A_p$, $N \le M$, and $\alpha \in (M \cap \mathrm{dom}(f_q) \cap S) \setminus \mathrm{dom}(f_p)$. Then $f_s(x) = f_s(\alpha) \cap Sk(M \cap \alpha)$. We claim that $M \cap \alpha$ is equal to $K \cap \alpha$, for some $K \in A_{p \restriction Q}$ such that $N \cap Q \le K$ and $\alpha \in (K \cap \mathrm{dom}(f_{q \restriction Q}) \cap S) \setminus \mathrm{dom}(f_{p \restriction Q})$. Let $K := M \cap Q$. Since $M \in A_p$ and $p \in D_Q$, $M \cap Q \in A_p$. As $Q$ is simple, $M \cap Q \in Q$. So $K = M \cap Q \in A_p \cap Q = A_{p \restriction Q}$. By Lemma 2.29, $M \sim M \cap Q$ and $N \sim N \cap Q$. Since $N \le M$, it follows by Lemma 2.18 that $N \cap Q \le M \cap Q = K$. Suppose for a moment that $\alpha \in Q$. Then $\alpha \in M \cap Q = K$ and $\alpha \in \mathrm{dom}(f_q) \cap Q = \mathrm{dom}(f_{q \restriction Q})$. Also, since $\alpha \notin \mathrm{dom}(f_p)$ and $\mathrm{dom}(f_{p \restriction Q}) = \mathrm{dom}(f_p) \cap Q$, it follows that $\alpha \notin \mathrm{dom}(f_{p \restriction Q})$. So $\alpha \in (K \cap \mathrm{dom}(f_{q \restriction Q}) \cap S) \setminus \mathrm{dom}(f_{p \restriction Q})$. Finally, assuming that $\alpha \in Q$, we have that $$ K \cap \alpha = M \cap Q \cap \alpha = M \cap \alpha. $$ Thus, assuming that $\alpha \in Q$, we have shown that $K \cap \alpha = M \cap \alpha$ is in the domain of $f_t$ and $f_t(K \cap \alpha)$ is defined as in case 7 of Definition 7.11. So $f_t(K \cap \alpha) = f_t(\alpha) \cap Sk(K \cap \alpha)$. By case 3 handled above, $f_s(\alpha) = f_t(\alpha)$. So \begin{multline*} f_t(M \cap \alpha) = f_t(K \cap \alpha) = f_t(\alpha) \cap Sk(K \cap \alpha) = \\ = f_s(\alpha) \cap Sk(M \cap \alpha) = f_s(M \cap \alpha). \end{multline*} It remains to prove that $\alpha \in Q$. Suppose for a contradiction that $\alpha \notin Q$. Since $p \in D_Q$ and $M \in A_p$, we have that $M \cap Q \in A_p$. By Lemma 2.29, $M \sim M \cap Q$. As $M \cap \alpha = x \in Q$ and $\alpha \notin Q$, clearly $\alpha = \min((M \cap \kappa) \setminus (Q \cap \kappa))$. By Lemma 3.6, we have that $$ \alpha = \min((M \cap \kappa) \setminus \beta_{M \cap Q,M}). $$ Now $M$ and $M \cap Q$ are both in $A_p$ and $M \sim M \cap Q$. Therefore, $\alpha \in r^*(A_p) \cap S$. Since $p$ is a condition, $\alpha \in \mathrm{dom}(f_p)$. But this contradicts our original choice of $\alpha$. \bigskip This completes the proof that $f_s = f_t$. \bigskip Now we show that $g_s = g_t$. Since the domain of $g_s$ is equal to the set of pairs $(K,x)$ such that $K \in f_s(x)$, and the domain of $g_t$ is equal to the set of pairs $(K,x)$ such that $K \in f_t(x)$, the fact that $f_s = f_t$ implies that $\mathrm{dom}(g_s) = \mathrm{dom}(g_t)$. Let $(K,x) \in \mathrm{dom}(g_s)$, and we will show that $g_s(K,x) = g_t(K,x)$. So $K \in f_s(x) = f_{q \oplus_N p}(x)$. By Definition 6.4, we have that $$ g_s(K,x) = g_{(q \oplus_N p) \restriction Q}(K,x) = g_{q \oplus_N p}(K,x). $$ Hence, by Definition 7.16, $$ g_s(K,x) = \bigcup \{ g_q(K,y) \cup g_p(K,y) : x = y, \textrm{or} \ x \in f_{q \oplus_N p}(y) \}. $$ Also, since $g_{q \restriction Q} = g_q \restriction Q$ and $g_{p \restriction Q} = g_p \restriction Q$, we have that \begin{multline*} g_t(K,x) = g_{(q \restriction Q) \oplus_{N \cap Q} (p \restriction Q)}(K,x) = \\ = \bigcup \{ g_{q}(K,y) \cup g_p(K,y) : x = y, \textrm{or} \ x \in f_t(y) \}. \end{multline*} Therefore, to show that $g_s(K,x) = g_t(K,x)$, it suffices to show that for any ordinal $\xi$, the following are equivalent: \begin{enumerate} \item $\xi \in g_q(K,y) \cup g_p(K,y)$, for some $y$ such that either $x = y$ or $x \in f_{q \oplus_N p}(y)$; \item $\xi \in g_q(K,y) \cup g_p(K,y)$, for some $y$ such that either $x = y$ or $x \in f_t(y)$. \end{enumerate} Obviously (1) and (2) are equivalent in the special case that $x = y$. \bigskip (2) implies (1): Suppose that $\xi \in g_q(K,y) \cup g_p(K,y)$, where $x \in f_t(y)$. Then $$ x \in f_t(y) = f_s(y) = f_{(q \oplus_N p) \restriction Q}(y) = f_{q \oplus_N p}(y). $$ So $\xi \in g_q(K,y) \cup g_p(K,y)$, where $x \in f_{q \oplus_N p}(y)$. Hence, (1) holds. \bigskip (1) implies (2): Suppose that $\xi \in g_q(K,y) \cup g_p(K,y)$, where $x \in f_{q \oplus_N p}(y)$. If $y \in Q$, then $$ y \in \mathrm{dom}(f_{q \oplus_N p}) \cap Q = \mathrm{dom}(f_{(q \oplus_N p) \restriction Q}) = \mathrm{dom}(f_s) $$ and $$ x \in f_{q \oplus_N p}(y) = f_{(q \oplus_N p) \restriction Q}(y) = f_s(y) = f_t(y). $$ Therefore, (2) holds. The more difficult case is when $y$ is not in $Q$. We split the proof into the two cases of whether $\xi$ is in $g_q(K,y)$ or in $g_p(K,y)$. \bigskip \noindent \emph{Case 1:} $\xi \in g_q(K,y)$. Then in particular, $g_q(K,y)$ is nonempty, which implies that $K \in f_q(y)$. So $K$ and $y$ are in $\mathrm{dom}(f_q)$. Since $q \in N$, $K$ and $y$ are in $N$. We claim that $x \in \mathrm{dom}(f_q)$. Assume for a moment that this claim is true, and we finish the proof. We have that $K$, $x$, and $y$ are in $\mathrm{dom}(f_q)$, and also, $K \in f_s(x) = f_{q \oplus_N p}(x)$ and $x \in f_{q \oplus_N p}(y)$. Since $q \oplus_N p \le q$, it follows that $K \in f_q(x)$ and $x \in f_q(y)$. As $q$ is a condition, we have that $g_q(K,y) \subseteq g_q(K,x)$. Thus, $\xi \in g_q(K,x)$, which implies that (2) holds, as required. Suppose for a contradiction that $x$ is not in $\mathrm{dom}(f_q)$. If $x \in N$, then $x \in \mathrm{dom}(f_{q \oplus_N p}) \cap N = \mathrm{dom}(f_q)$ by Lemma 7.13(1). So $x \notin N$. Since $x \in f_{q \oplus_N p}(y)$ and $y \in N$, the only way that $x$ would not be in $N$ is if $y \in S$. So $y = \alpha$, for some $\alpha \in \mathrm{dom}(f_q) \cap S$. Since $N \in A_p \subseteq A_{q \oplus_N p}$, we have that $N \cap \alpha \in f_{q \oplus_N p}(\alpha)$. And since $x \in f_{q \oplus_N p}(\alpha)$ and $x \notin N$, $N \cap \alpha$ is either equal to $x$ or is in $f_{q \oplus_N p}(x)$. In particular, $\sup(N \cap \alpha) \le \sup(x)$. Since $s \in Q$ and $x \in \mathrm{dom}(f_s)$, $x \in Q$. Hence, $\sup(x) < Q \cap \kappa$. Since $y = \alpha$ is not in $Q$, $Q \cap \kappa \le \alpha$. But $\alpha \in S$ and $Q \cap \kappa \notin S$, so $Q \cap \kappa < \alpha$. Therefore, $$ \sup(N \cap \alpha) \le \sup(x) < Q \cap \kappa < \alpha. $$ But $Q \in N$ implies that $Q \cap \kappa \in N \cap \alpha$, which contradicts that $\sup(N \cap \alpha) < Q \cap \kappa$. \bigskip \noindent \emph{Case 2:} $\xi \in g_p(K,y)$. Then in particular, $g_p(K,y)$ is nonempty, which implies that $K \in f_p(y)$. So $K$ and $y$ are in $\mathrm{dom}(f_p)$. The easier case is when $x \in \mathrm{dom}(f_p)$. Then since $K \in f_s(x) = f_{q \oplus_N p}(x)$, $x \in f_{q \oplus_N p}(y)$, and $q \oplus_N p \le p$, it follows that $K \in f_p(x)$ and $x \in f_p(y)$. Since $p$ is a condition, it follows that $g_p(K,y) \subseteq g_p(K,x)$. So $\xi \in g_p(K,x)$, and (2) holds. \bigskip Assume that $x \notin \mathrm{dom}(f_p)$. Since $x \in \mathrm{dom}(f_{q \oplus_N p})$, Definition 7.11 implies that either $x \in \mathrm{dom}(f_q)$, or $x = M \cap \alpha$ for some $M \in A_p$ with $N \le M$ and some $\alpha \in (M \cap \mathrm{dom}(f_q) \cap S) \setminus \mathrm{dom}(f_p)$. \bigskip \noindent \emph{Case 2a:} $x \in \mathrm{dom}(f_q)$. Then $x \in N$. Since $K \in f_s(x)$ and $x \notin S$, $K \in N$ as well. By Lemma 7.8(2), $K \in \mathrm{dom}(f_p) \cap N \subseteq \mathrm{dom}(f_q)$. So $K \in \mathrm{dom}(f_q)$. Since $K$ and $x$ are in $\mathrm{dom}(f_q)$, $K \in f_{q \oplus_N p}(x)$, and $q \oplus_N p \le q$, it follows that $K \in f_q(x)$. \bigskip First, assume that $y \in N$. By Lemma 7.8(2), $y \in \mathrm{dom}(f_p) \cap N \subseteq \mathrm{dom}(f_q)$. So $x$ and $y$ are in $\mathrm{dom}(f_q)$. Since $x \in f_{q \oplus_N p}(y)$ and $q \oplus_N p \le q$, we have that $x \in f_q(y)$. So $K \in f_q(x)$ and $x \in f_q(y)$. Since $q$ is a condition, $g_q(K,y) \subseteq g_q(K,x)$. On the other hand, by Lemma 7.8(3), $(K,y) \in \mathrm{dom}(g_p) \cap N \subseteq \mathrm{dom}(g_q)$ and $g_p(K,y) \subseteq g_q(K,y)$. Thus, $$ \xi \in g_p(K,y) \subseteq g_q(K,y) \subseteq g_q(K,x). $$ So (2) holds. \bigskip Secondly, assume that $y \notin N$. Let us consider cases 1--7 of Definition 7.11 in the definition of $f_{q \oplus_N p}(y)$. Since $y \notin N$, cases 1, 2, and 3 are false. As $K \in f_p(y) \cap N$, case 5 is false. Case 7 is false by Lemma 7.9, since $y \in \mathrm{dom}(f_p)$. So we are left with cases 4 and 6. In case 4, $y \in \mathrm{dom}(f_p) \setminus N$, and $N \cap \alpha \in f_p(y) \cup \{ y \}$ for some $\alpha \in \mathrm{dom}(f_p) \cap S \cap N$. Then $f_{q \oplus_N p}(y) = f_q(\alpha) \cup f_p(y)$. Since $x \in f_{q \oplus_N p}(y)$ and $x \notin \mathrm{dom}(f_p)$, it follows that $x \in f_q(\alpha)$. Since $K \in f_p(y) \cap N$ and $N \cap \alpha \in f_p(y) \cup \{ y \}$, we have that $K \in f_p(N \cap \alpha)$. Since $p$ is a condition, $g_p(K,y) \subseteq g_p(K,N \cap \alpha)$. As $N \cap \alpha \in f_p(\alpha)$ and $p \in D_N$, it follows that $g_p(K,N \cap \alpha) \subseteq g_p(K,\alpha)$. So $g_p(K,y) \subseteq g_p(K,\alpha)$. Since $K$ and $\alpha$ are in $N$, $g_p(K,\alpha) \subseteq g_q(K,\alpha)$ by Lemma 7.8(3). Now $K \in f_q(x)$ and $x \in f_q(\alpha)$, so $g_q(K,\alpha) \subseteq g_q(K,x)$ since $q$ is a condition. Putting it all together, $$ \xi \in g_p(K,y) \subseteq g_p(K,\alpha) \subseteq g_q(K,\alpha) \subseteq g_q(K,x). $$ So (2) holds. In case 6, we have that case 4 fails, and $f_{q \oplus_N p}(y) = f_p(y) \cup f_q(M)$, where $M$ is the membership largest element of $f_p(y) \cap N$. Since $x \in f_{q \oplus_N p}(y)$ and $x \notin \mathrm{dom}(f_p)$, we have that $x \in f_q(M)$. Also, $K \in f_p(y) \cap N$, so $K \in f_p(M) \cup \{ M \}$. But $K \in Sk(x)$ and $x \in f_q(M)$, so $K \ne M$. Hence, $K \in f_p(M)$. As $p$ is a condition, $g_p(K,y) \subseteq g_p(K,M)$. Also, since $K \in f_p(M)$ and $K$ and $M$ are in $N$, $K \in f_q(M)$ by Lemma 7.8(2). As $x \in f_q(M)$ and $K \in Sk(x)$, $K \in f_q(x)$. Since $q$ is a condition, $g_q(K,M) \subseteq g_q(K,x)$. And by Lemma 7.8(3), $g_p(K,M) \subseteq g_q(K,M)$. Therefore, $$ \xi \in g_p(K,y) \subseteq g_p(K,M) \subseteq g_q(K,M) \subseteq g_q(K,x). $$ So (2) holds. \bigskip \noindent \emph{Case 2b:} $x = M \cap \alpha$ for some $M \in A_p$ with $N \le M$ and some $\alpha \in (M \cap \mathrm{dom}(f_q) \cap S) \setminus \mathrm{dom}(f_p)$. Note that $\alpha \in \mathrm{dom}(f_q)$ implies that $\alpha \in N$. We will show, in fact, that this case is impossible. We claim that $\alpha < Q \cap \kappa$. If not, then since $\alpha \in S$ and $Q \cap \kappa \notin S$, we have that $Q \cap \kappa < \alpha$. Now $M \cap \alpha = x \in Q$, so $\sup(x) < Q \cap \kappa$. And $\alpha \in M \cap N \cap \kappa$, so $\alpha < \beta_{M,N}$ by Proposition 2.11. Therefore, $Q \cap \kappa < \alpha < \beta_{M,N}$. Since $N \le M$, we have that $N \cap \beta_{M,N} \subseteq M$. So $N \cap \alpha \subseteq M \cap \alpha$. Therefore, $$ \sup(N \cap \alpha) \le \sup(M \cap \alpha) = \sup(x) < Q \cap \kappa < \alpha. $$ So $\sup(N \cap \alpha) < Q \cap \kappa$. Since $Q \in N$, we have that $Q \cap \kappa \in N \cap \alpha$, which contradicts that $\sup(N \cap \alpha) < Q \cap \kappa$. So indeed, $\alpha < Q \cap \kappa$. Since $y \notin Q$, Lemma 4.3(4) implies that $Q \cap \kappa \le \sup(y)$. Hence, $\alpha < \sup(y)$. But we know that $M \cap \alpha = x \in f_{q \oplus_N p}(y)$. And by Definition 7.11 and Lemma 4.6, the only value of $y$ for which $M \cap \alpha$ is in $f_{q \oplus_N p}(y)$ is either $y = \alpha$ or $y = L \cap \alpha$ for some $L$. Neither of these cases is possible, since $\alpha < \sup(y)$. \end{proof} \bigskip \addcontentsline{toc}{section}{9. The approximation property} \textbf{\S 9. The approximation property} \stepcounter{section} \bigskip In this section we will prove that certain quotients of the forcing poset $\mathbb{P}$ have the $\omega_1$-approximation property. \begin{lemma} Suppose that $Q \in \mathcal Y$ is simple and $Q \prec (H(\lambda),\in,\mathbb{P})$. Then $Q \cap \mathbb{P}$ is a regular suborder of $\mathbb{P}$. \end{lemma} \begin{proof} If $p$ and $q$ are in $Q \cap \mathbb{P}$ and are compatible in $\mathbb{P}$, then by the elementarity of $Q$, there is $r \in Q \cap \mathbb{P}$ such that $r \le p, q$. So $p$ and $q$ are compatible in $Q \cap \mathbb{P}$. Let $B$ be a maximal antichain of $Q \cap \mathbb{P}$, and we will show that $B$ is predense in $\mathbb{P}$. So let $p \in \mathbb{P}$, and we will find $s \in B$ which is compatible with $p$. Since the set $D_Q$ is dense by Lemma 6.3, fix $q \le p$ in $D_Q$. Then $q \restriction Q$ exists and is in $Q \cap \mathbb{P}$. Since $B$ is maximal, we can find $s \in B$ such that $s$ and $q \restriction Q$ are compatible in $Q \cap \mathbb{P}$. Fix $w \le s, q \restriction Q$ in $Q \cap \mathbb{P}$. By Proposition 6.15, $w$ and $q$ are compatible in $\mathbb{P}$. Fix $t \le w, q$. Then $t \le q \le p$ and $t \le w \le s$. So $p$ and $s$ are compatible. \end{proof} \begin{thm} Suppose that $Q \in \mathcal Y$ is simple, $Q \prec (H(\lambda),\in,\mathbb{P})$, and $Q \cap \kappa \notin S$. Then $Q \cap \mathbb{P}$ forces that $\mathbb{P} / \dot G_{Q \cap \mathbb{P}}$ has the $\omega_1$-approximation property.\footnote{As discussed in the introduction of the paper, the proof of this result is similar to the proof of \cite[Lemma 2.22]{mitchell}, except that we replace the tidy condition property of \cite[Definition 2.20]{mitchell} with Proposition 8.6.} \end{thm} \begin{proof} By Lemma 1.4, it suffices to show that $\mathbb{P}$ forces that the pair $$ (V[\dot G_\mathbb{P} \cap Q],V[\dot G_\mathbb{P}]) $$ has the $\omega_1$-approximation property. So let $p$, $\mu$, and $\dot k$ be given such that $\mu$ is an ordinal, and $p$ forces in $\mathbb{P}$ that $\dot k : \mu \to On$ is a function satisfying that for any countable set $a$ in $V[\dot G_\mathbb{P} \cap Q]$, $\dot k \restriction a \in V[\dot G_\mathbb{P} \cap Q]$. We will find an extension of $p$ which forces that $\dot k$ is in $V[\dot G_\mathbb{P} \cap Q]$. Fix a regular cardinal $\theta$ large enough so that $\mathbb{P}$, $\mu$, and $\dot k$ are members of $H(\theta)$. By the stationarity of the simple models in $\mathcal X$ as described in Assumption 2.22, we can fix a countable set $M^* \prec H(\theta)$ which contains the parameters $\mathbb{P}$, $Q$, $p$, $\mu$, $\dot k$, and satisfies that the set $M^* \cap H(\lambda)$ is in $\mathcal X$ and is simple. Let $M := M^* \cap H(\lambda)$. Note that since $\mathbb{P} \subseteq H(\lambda)$, $M^* \cap \mathbb{P} = M \cap \mathbb{P}$. In particular, $p \in M \cap \mathbb{P}$. Also, observe that $Q \in M^* \cap H(\lambda) = M$. By Lemma 7.1, we can fix $p_0 \le p$ such that $M \in A_{p_0}$. Since $M^* \cap \mu$ is in $V$, by the choice of $p$ and $\dot k$ we can fix $p_1 \le p_0$ and a $(Q \cap \mathbb{P} )$-name $\dot s$ such that $$ p_1 \Vdash_\mathbb{P} \dot k \restriction (M^* \cap \mu) = \dot s^{\dot G_\mathbb{P} \cap Q}. $$ Since $M \in A_{p_1}$, apply Lemma 8.2 to fix $p_2 \le p_1$ such that $p_2 \in D_{M} \cap D_Q$. Note that since $p_2 \le p$ and $p \in M$, it follows that $p_2 \restriction M \le p$ by Lemma 7.7. So it will suffice to prove that $p_2 \restriction M$ forces that $\dot k$ is in $V[\dot G_\mathbb{P} \cap Q]$. \bigskip \noindent \emph{Claim 1:} If $t \le p_2$ is in $D_Q$, $\nu \in M^* \cap \mu$, and $t \Vdash_{\mathbb{P}} \dot k(\nu) = x$ (or $t \Vdash_{\mathbb{P}} \dot k(\nu) \ne x$, respectively) then $t \restriction Q \Vdash_{Q \cap \mathbb{P}} \dot s(\nu) = x$ (or $t \restriction Q \Vdash_{Q \cap \mathbb{P}} \dot s(\nu) \ne x$, respectively). \bigskip We will prove only the main part of Claim 1, since the parenthetical part has the essentially the same proof. So assume that $t \Vdash_\mathbb{P} \dot k(\nu) = x$. If $t \restriction Q \not \Vdash_{Q \cap \mathbb{P}} \dot s(\nu) = x$, then there is $u \le t \restriction Q$ in $Q \cap \mathbb{P}$ such that $u \Vdash_{Q \cap \mathbb{P}} \dot s(\nu) \ne x$. By Proposition 6.15, $u$ and $t$ are compatible in $\mathbb{P}$. Fix $z \in \mathbb{P}$ such that $z \le u, t$. Fix a generic filter $G$ on $\mathbb{P}$ with $z \in G$. Then $t \in G$, which implies that $\dot k^G(\nu) = x$. Also $p_1 \in G$, which implies that $$ \dot k^G \restriction (M^* \cap \mu) = \dot s^{G \cap Q}. $$ It follows that $\dot s^{G \cap Q}(\nu) = x$. But $z \le u$, so $u \in G \cap Q$. By the choice of $u$, $\dot s^{G \cap Q}(\nu) \ne x$, which is a contradiction. This completes the proof of Claim 1. \bigskip \noindent \emph{Claim 2:} For all $q \le p_2 \restriction M$ in $D_Q$, $\nu < \mu$, and $x$, $$ q \Vdash_{\mathbb{P}} \dot k(\nu) = x \implies \forall r \in \mathbb{P} ( ( r \le p_2 \restriction M \land r \le q \restriction Q ) \implies (r \Vdash_\mathbb{P} \dot k(\nu) = x)). $$ \bigskip Note that $p_2 \restriction M$, $Q$, $D_Q$, $\mu$, $\dot k$, and $\mathbb{P}$ are in $M^*$. So by the elementarity of $M^*$, it suffices to show that the statement holds in $M^*$. Suppose for a contradiction that there exists $q \le p_2 \restriction M$ in $M^* \cap D_Q$, $\nu \in M^* \cap \mu$, and $x \in M^*$ such that $$ q \Vdash_\mathbb{P} \dot k(\nu) = x, $$ but there is $r_0 \in M^* \cap \mathbb{P}$ with $r_0 \le p_2 \restriction M$ and $r_0 \le q \restriction Q$ such that $$ r_0 \not \Vdash_\mathbb{P} \dot k(\nu) = x. $$ By the elementarity of $M^*$, we can fix $r \le r_0$ in $M^* \cap D_Q$ such that $$ r \Vdash_\mathbb{P} \dot k(\nu) \ne x. $$ Then $r \le p_2 \restriction M$ and $r \le q \restriction Q$. Since $r \le q \restriction Q$, it follows that $r \restriction Q \le q \restriction Q$ by Lemma 6.6(2). We have that $q \le p_2 \restriction M$ and $q \in M^* \cap \mathbb{P} = M \cap \mathbb{P}$. By Proposition 7.19, $q \oplus_{M} p_2$ is a condition which is below $q$ and $p_2$. Similarly, $r \le p_2 \restriction M$ and $r \in M \cap \mathbb{P}$. By Proposition 7.19, $r \oplus_{M} p_2$ is a condition which is below $r$ and $p_2$. By Proposition 8.6, we have that $$ (q \oplus_{M} p_2) \restriction Q = (q \restriction Q) \oplus_{M \cap Q} (p_2 \restriction Q), $$ and $$ (r \oplus_{M} p_2) \restriction Q = (r \restriction Q) \oplus_{M \cap Q} (p_2 \restriction Q). $$ We would like to apply Lemma 8.1 to $M \cap Q$, $p_2 \restriction Q$, $q \restriction Q$, and $r \restriction Q$. Let us check that the assumptions of Lemma 8.1 hold for these objects. By Lemma 2.30, $M \cap Q$ is simple. Since $p_2 \in D_{M} \cap D_Q$, it follows that $p_2 \restriction Q \in D_{M \cap Q}$ by Lemma 8.3. As $q$, $r$, and $Q$ are in $M^*$, we have that $q \restriction Q$ and $r \restriction Q$ are in $M^* \cap Q = M \cap Q$. We observed above that $r \restriction Q \le q \restriction Q$. Finally, $q \le p_2 \restriction M$ implies that $q \restriction Q \le (p_2 \restriction Q) \restriction (M \cap Q)$ by Lemma 8.5. Thus, all of the assumptions of Lemma 8.1 hold. So by Lemma 8.1, $$ (r \restriction Q) \oplus_{M \cap Q} (p_2 \restriction Q) \le (q \restriction Q) \oplus_{M \cap Q} (p_2 \restriction Q). $$ Combining this with the equalities above, we have that $$ (r \oplus_{M} p_2) \restriction Q \le (q \oplus_{M} p_2) \restriction Q. $$ We claim that this last inequality is impossible. In fact, we will show that $(r \oplus_{M} p_2) \restriction Q$ and $(q \oplus_{M} p_2) \restriction Q$ are incompatible. This contradiction will complete the proof of Claim 2. We know that $r \Vdash_\mathbb{P} \dot k(\nu) \ne x$, and therefore, since $r \oplus_{M} p_2 \le r$, $r \oplus_{M} p_2 \Vdash_\mathbb{P} \dot k(\nu) \ne x$. By Lemma 8.5(1), $r \oplus_M p_2$ is in $D_Q$. So by Claim 1, $$ (r \oplus_{M} p_2) \restriction Q \Vdash_{Q \cap \mathbb{P}} \dot s(\nu) \ne x. $$ Similarly, $q \Vdash_\mathbb{P} \dot k(\nu) = x$, and therefore, since $q \oplus_{M} p_2 \le q$, $q \oplus_{M} p_2 \Vdash_\mathbb{P} \dot k(\nu) = x$. By Lemma 8.5(1), $q \oplus_M p_2$ is in $D_Q$. So by Claim 1, $$ (q \oplus_{M} p_2) \restriction Q \Vdash_{Q \cap \mathbb{P}} \dot s(\nu) = x. $$ So indeed, $(r \oplus_{M} p_2) \restriction Q$ and $(q \oplus_{M} p_2) \restriction Q$ are incompatible, since they force contradictory information. \bigskip Now we finish the proof that $p_2 \restriction M$ forces that $\dot k \in V[\dot G_\mathbb{P} \cap Q]$. Let $G$ be a generic filter on $\mathbb{P}$ with $p_2 \restriction M \in G$, and we will show that $\dot k^G \in V[G \cap Q]$. In the model $V[G \cap Q]$, define a partial function $h : \mu \to V$ by letting, for every $\nu < \mu$, $h(\nu) = x$ iff there exists $t \in G \cap Q$ such that for every $r \in \mathbb{P}$, if $r$ is below both $p_2 \restriction M$ and $t$, then $$ r \Vdash_\mathbb{P}^V \dot k(\nu) = x. $$ We claim that $h = \dot k^G$. First, let us show that $h$ is well-defined. So assume that $t_0$ and $t_1$ are in $G \cap Q$ and witness respectively that $h(\nu) = x_0$ and $h(\nu) = x_1$. We will show that $x_0 = x_1$. Fix $u \le t_0, t_1$ in $G \cap Q$. Since $p_2 \restriction M$ is in $G$, we can fix $r \le u, p_2 \restriction M$ in $G$. Now $r$ is below both $p_2 \restriction M$ and $t_0$, so by the choice of $t_0$, $r \Vdash_\mathbb{P}^V \dot k(\nu) = x_0$. Similarly, $r$ is below both $p_2 \restriction M$ and $t_1$, so by the choice of $t_1$, $r \Vdash_\mathbb{P}^V \dot k(\nu) = x_1$. Thus, $r \Vdash_\mathbb{P}^V x_0 = x_1$, which implies that $x_0 = x_1$. Secondly, we prove that $h = \dot k^G$. As just shown, $h$ is a well-defined function whose domain is a subset of $\mu$. So it suffices to show that for all $\nu < \mu$, $\nu \in \mathrm{dom}(h)$ and $h(\nu) = \dot k^G(\nu)$. Fix $\nu < \mu$, and let $x := \dot k^G(\nu)$. Fix $q \le p_2 \restriction M$ in $G \cap D_Q$ such that $q \Vdash_\mathbb{P} \dot k(\nu) = x$. Then by Claim 2, for all $r \in \mathbb{P}$, if $r$ is an extension of both $p_2 \restriction M$ and $q \restriction Q$, then $r \Vdash_\mathbb{P} \dot k(\nu) = x$. Let $t := q \restriction Q$. By Lemma 6.5, $q \le t$. Since $q \in G$, it follows that $t \in G \cap Q$. So by the definition of $h$, in order to show that $h(\nu) = x$, it suffices to show that for all $r \in \mathbb{P}$, if $r$ is an extension of both $p_2 \restriction M$ and $t = q \restriction Q$, then $r \Vdash_\mathbb{P}^V \dot k(\nu) = x$. But this statement is exactly what we observed to be true at the end of the previous paragraph. \end{proof} Recall that in Theorem 9.2 we assumed that $Q \cap \kappa \notin S$. Suppose, on the other hand, that $Q \in \mathcal Y$ is simple, $Q \prec (H(\lambda),\in,\mathbb{P})$, and $Q \cap \kappa \in S$. Let $G$ be a generic filter on $\mathbb{P}$. Let $\langle c_\alpha : \alpha \in S \rangle$ be the partial square sequence in $V[G]$ as defined in Section 5. Since $Q \cap \kappa \in S$, we have that $c_{Q \cap \kappa}$ is defined. Using the coherence property of the partial square sequence, one can show that every proper initial segment of $c_{Q \cap \kappa}$ is in $V[G \cap Q]$. But by a density argument, $c_{Q \cap \kappa}$ is not in $V[G \cap Q]$. Thus, the quotient $\mathbb{P} / (G \cap Q)$ does not have the $\omega_1$-approximation property in $V[G \cap Q]$. \bigskip \part{Combining Forcings} \bigskip \addcontentsline{toc}{section}{10. A product forcing} \textbf{\S 10. A product forcing} \stepcounter{section} \bigskip We now develop a forcing poset which simultaneously adds partial square sequences on multiple stationary subsets of $\kappa$. This forcing poset will be a kind of side condition product forcing. Before we get started, we need to make some additional assumptions. \begin{assumption} The cardinal $\lambda$ introduced in Notation 2.1 is at least $\kappa^+$. \end{assumption} \begin{notation} Fix an ordinal $\lambda^* \le \lambda$. Fix a sequence $\langle S_i : i < \lambda^* \rangle$ such that for all $i < \lambda^*$, $S_i \subseteq \Lambda$ is stationary in $\kappa$ and for all $\alpha \in S_i$, $Sk(\alpha) \cap \kappa = \alpha$; moreover, for all $i < j < \lambda^*$, $S_i \cap S_j$ is nonstationary. \end{notation} \begin{assumption} For all $i < j < \lambda^*$, there is a club set $C_{i,j} \subseteq \kappa$ satisfying that $S_i \cap S_j \cap C_{i,j} = \emptyset$, and moreover, $C_{i,j}$ is definable in the structure $\mathcal A$ of Notation 2.3 from $i$ and $j$. \end{assumption} \begin{notation} For each $i < \lambda^*$, let $\mathbb{P}_i$ denote the forcing poset defined in Definition 4.2 which adds a partial square sequence on $S_i$. \end{notation} We introduce a side condition product forcing which combines the forcing posets $\mathbb{P}_i$, for all $i < \lambda^*$. Before giving the definition, we need the following technical lemma to make sure that the definition makes sense. \begin{lemma} Suppose that $M$ and $N$ are in $\mathcal X$, $M \le N$, and $\gamma = \min((M \cap \kappa) \setminus \beta_{M,N})$. Then there is at most one ordinal $i < \lambda^*$ such that $i \in M \cap N$ and $\gamma \in S_i$. \end{lemma} \begin{proof} Suppose for a contradiction that there are $i < j$ in $M \cap N \cap \lambda^*$ such that $\gamma \in S_i \cap S_j$. By Assumption 10.3, $S_i \cap S_j \cap C_{i,j} = \emptyset$, and by the elementarity of $M$ and $N$, $C_{i,j} \in M \cap N$. By the elementarity of $M \cap N$, it is easy to show that $C_{i,j}$ is cofinal in $M \cap N \cap \kappa$. Since $M \le N$, by Lemma 2.15 and the minimality of $\gamma$, $$ M \cap N \cap \kappa = M \cap \beta_{M,N} = M \cap \gamma. $$ Therefore, $C_{i,j}$ is cofinal in $M \cap \gamma$. Since $C_{i,j} \in M$, by the elementarity of $M$ it follows that $C_{i,j}$ is cofinal in $\gamma$. As $C_{i,j}$ is a club, $\gamma \in C_{i,j}$. But then $\gamma \in S_i \cap S_j \cap C_{i,j}$, contradicting the choice of $C_{i,j}$. \end{proof} \begin{definition} Let $\mathbb{Q}$ be the forcing poset consisting of pairs $p = (F_p,A_p)$ satisfying: \begin{enumerate} \item $A_p$ is an adequate set; \item $F_p$ is a function whose domain is a finite subset of $\lambda^*$; \item for all $i \in \mathrm{dom}(F_p)$, $F_p(i) \in \mathbb{P}_i$ and $$ \{ M \in A_p : i \in M \} \subseteq A_{F_p(i)};\footnote{Recall from Definition 4.2 that $F_p(i) = (f_{F_p(i)},g_{F_p(i)},A_{F_p(i)})$.} $$ \item if $M$ and $N$ are in $A_p$, $M \sim N$, $i \in M \cap N \cap \lambda^*$, and the ordinal $\min((M \cap \kappa) \setminus \beta_{M,N})$ exists and is in $S_i$, then $i \in \mathrm{dom}(F_p)$. \end{enumerate} Let $q \le p$ if $A_p \subseteq A_q$, $\mathrm{dom}(F_p) \subseteq \mathrm{dom}(F_q)$, and for all $i \in \mathrm{dom}(F_p)$, $F_q(i) \le F_p(i)$ in $\mathbb{P}_i$. \end{definition} Note that since $A_p$ is finite, Lemma 10.5 implies that there are only finitely many ordinals $i$ as described in (4). Let us see that we can add any ordinal in $\lambda^*$ to the domain of $F_p$. \begin{definition} Let $p \in \mathbb{Q}$, and let $x$ be a finite subset of $\lambda^* \setminus \mathrm{dom}(F_p)$. Define $$ p \uplus x $$ as the pair $(F,A)$ satisfying: \begin{enumerate} \item $A := A_p$; \item $\mathrm{dom}(F) := \mathrm{dom}(F_p) \cup x$; \item for all $j \in \mathrm{dom}(F_p)$, $F(j) := F_p(j)$, and for all $i \in x$, $$ F(i) := (\emptyset,\emptyset,B_i), $$ where $$ B_i := \{ M \in A_p : i \in M \}. $$ \end{enumerate} \end{definition} \begin{lemma} Let $p \in \mathbb{Q}$, and let $x$ be a finite subset of $\lambda^* \setminus \mathrm{dom}(F_p)$. Then: \begin{enumerate} \item $p \uplus x \in \mathbb{Q}$; \item $p \uplus x \le p$; \item $x \subseteq \mathrm{dom}(F_{p \uplus x})$; \item whenever $q \le p$ and $x \subseteq \mathrm{dom}(F_q)$, then $q \le p \uplus x$. \end{enumerate} \end{lemma} \begin{proof} (3) is immediate. \bigskip (1) To see that $p \uplus x$ is a condition, requirements (1), (2), and (4) of Definition 10.6 are immediate. For requirement (3), for all $j \in \mathrm{dom}(F_p)$, $F_{p \uplus x}(j) = F_p(j) \in \mathbb{P}_j$ and $\{ M \in A_p : j \in M \} \subseteq A_{F_p(j)} = A_{F_{p \uplus x}(j)}$, since $p$ is a condition. Consider $i \in x$. Then by definition, $F_{p \uplus x}(i) = (\emptyset,\emptyset,B_i)$, where $B_i = \{ M \in A_p : i \in M \}$. Thus, it suffices to show that $(\emptyset,\emptyset,B_i) \in \mathbb{P}_i$. We check that $(\emptyset,\emptyset,B_i)$ satisfies properties (1)--(7) of Definition 4.2. (1) $B_i$ is adequate, because it is a subset of $A_p$. (2)--(6) are vacuously true. We claim that (7) is vacuously true as well. Suppose that $\gamma \in r^*(B_i) \cap S_i$. Then for some $M$ and $N$ in $B_i$, $M \sim N$ and $\gamma = \min((M \cap \kappa) \setminus \beta_{M,N})$. By the definition of $B_i$, $M$ and $N$ are in $A_p$ and $i \in M \cap N \cap \lambda^*$. So $M$ and $N$ are in $A_p$, $M \sim N$, $i \in M \cap N \cap \lambda^*$, and $\min((M \cap \kappa) \setminus \beta_{M,N})$ exists and is in $S_i$. Since $p$ is a condition, Definition 10.6(4) implies that $i \in \mathrm{dom}(F_p)$. But $i \in x$ and $x \cap \mathrm{dom}(F_p) = \emptyset$, which is a contradiction. \bigskip (2) It is trivial to check that $p \uplus x \le p$. \bigskip (4) Assume that $q \le p$ and $x \subseteq \mathrm{dom}(F_q)$, and we will show that $q \le p \uplus x$. Since $q \le p$, $A_{p \uplus x} = A_p \subseteq A_q$ and $\mathrm{dom}(F_p) \subseteq \mathrm{dom}(F_q)$. Since $x \subseteq \mathrm{dom}(F_q)$, $\mathrm{dom}(F_{p \uplus x}) = \mathrm{dom}(F_p) \cup x \subseteq \mathrm{dom}(F_q)$. Let $j \in \mathrm{dom}(F_{p \uplus x})$, and we will show that $F_q(j) \le F_{p \uplus x}(j)$. If $j \in \mathrm{dom}(F_p)$, then by definition, $F_{p \uplus x}(j) = F_p(j)$. And since $q \le p$, $F_q(j) \le F_p(j)$. So $F_q(j) \le F_{p \uplus x}(j)$. Suppose that $i \in x$. We claim that $F_q(i) \le F_{p \uplus x}(i)$, that is, $F_q(i) \le (\emptyset,\emptyset,B_i)$, where $B_i = \{ M \in A_p : i \in M \}$. We verify properties (a)--(d) of Definition 4.2. Note that (b), (c), and (d) are vacuously true. For (a), since $i \in \mathrm{dom}(F_q)$, by Definition 10.6(3) we have that $\{ M \in A_q : i \in M \} \subseteq A_{F_q(i)}$. But $q \le p$ implies that $A_p \subseteq A_q$. Hence, $B_i = \{ M \in A_p : i \in M \} \subseteq \{ M \in A_q : i \in M \} \subseteq A_{F_q(i)}$. \end{proof} The next two easy lemmas will be useful in what follows. \begin{lemma} Let $p \in \mathbb{Q}$. For each $i \in \mathrm{dom}(F_p)$, suppose that $r_i$ is a condition in $\mathbb{P}_i$ such that $r_i \le F_p(i)$ in $\mathbb{P}_i$. Define $r$ as follows: \begin{enumerate} \item $A_r := A_p$; \item $\mathrm{dom}(F_r) := \mathrm{dom}(F_p)$; \item for all $i \in \mathrm{dom}(F_r)$, $F_r(i) := r_i$. \end{enumerate} Then $r \in \mathbb{Q}$ and $r \le p$. \end{lemma} \begin{proof} Straightforward. \end{proof} \begin{lemma} Let $x$ be a finite subset of $\lambda^*$, and assume that for each $i \in x$, $D_i$ is a dense subset of $\mathbb{P}_i$. Then for any $p \in \mathbb{Q}$, there is $r \le p$ satisfying: \begin{enumerate} \item $A_r = A_p$; \item $\mathrm{dom}(F_r) = \mathrm{dom}(F_p) \cup x$; \item for each $i \in x$, $F_r(i) \in D_i$. \end{enumerate} \end{lemma} \begin{proof} Let $q := p \uplus (x \setminus \mathrm{dom}(F_p))$. Then $q$ is a condition, $A_q = A_p$, $\mathrm{dom}(F_q) = \mathrm{dom}(F_p) \cup x$, and $q \le p$. For each $i \in x$, fix $r_i \le F_q(i)$ in $D_i$. By Lemma 10.9, there is $r \le q$ such that $A_r = A_q = A_p$, $\mathrm{dom}(F_r) = \mathrm{dom}(F_q) = \mathrm{dom}(F_p) \cup x$, and for all $i \in \mathrm{dom}(F_q)$, if $i \in x$ then $F_r(i) = r_i$, and if $i \notin x$ then $F_r(i) = F_q(i)$. Then $r$ is as required. \end{proof} The next result justifies our informal use of the word ``product'' to describe $\mathbb{Q}$. \begin{proposition} For each $i < \lambda^*$, there is a projection mapping from a dense subset of $\mathbb{Q}$ into $\mathbb{P}_i$. \end{proposition} \begin{proof} Let $D$ be the set of conditions $q \in \mathbb{Q}$ such that $i \in \mathrm{dom}(F_q)$, together with the maximum condition $(\emptyset,\emptyset)$. By Lemma 10.8, if $p \in \mathbb{Q}$, then there is $q \le p$ such that $i \in \mathrm{dom}(F_q)$. Thus, $D$ is dense in $\mathbb{Q}$. Define $\pi_i : D \to \mathbb{P}_i$ as follows. Let $\pi_i(\emptyset,\emptyset)$ be the maximum condition of $\mathbb{P}_i$, namely, $(\emptyset,\emptyset,\emptyset)$. If $q \in D$ and $q$ is not the maximum condition, then $i \in \mathrm{dom}(F_q)$. In that case, let $\pi_i(q) := F_q(i)$. We claim that $\pi_i$ is a projection mapping. Obviously, $\pi_i$ maps the maximum condition of $\mathbb{Q}$ to the maximum condition of $\mathbb{P}_i$. Suppose that $q \le p$ in $D$, and we will show that $\pi_i(q) \le \pi_i(p)$ in $\mathbb{P}_i$. This is immediate if $p$ is the maximum condition of $\mathbb{Q}$, so assume that $i \in \mathrm{dom}(F_p)$. Then since $q \le p$, we have that $\pi_i(q) = F_q(i) \le F_p(i) = \pi_i(p)$. Suppose that $v \le \pi_i(p)$ in $\mathbb{P}_i$, and we will find $r \le p$ in $D$ such that $\pi_i(r) = v$. First, assume that $p$ is not the maximum condition of $\mathbb{Q}$. Then $v \le \pi_i(p) = F_p(i)$. By Lemma 10.9, there exists $r \le p$ satisfying that $F_r(i) = v$ and $F_r(j) = F_p(j)$ for all $j \in \mathrm{dom}(F_p) \setminus \{ i \}$. Then $\pi_i(r) = v$, as required. Secondly, assume that $p$ is the maximum condition of $\mathbb{Q}$. We construct a condition $r$ as follows. Let $A_r := \emptyset$, and let $F_r$ be the function with domain equal to $\{ i \}$ such that $F_r(i) = v$. Since $A_r = \emptyset$, it is easy to check that $r$ is a condition, with most properties of Definition 10.6 being vacuously true. Also, $r \le p$, since $p$ is the maximum condition, and $\pi_i(r) = F_r(i) = v$. \end{proof} It follows that if $G$ is a generic filter on $\mathbb{Q}$, then $\pi_i[G \cap D]$ generates a generic filter on $\mathbb{P}_i$, where $D$ is the dense subset of $\mathbb{Q}$ which is the domain of $\pi_i$. We will prove in Section 11 that $\mathbb{Q}$ preserves $\omega_1$ and is $\kappa$-c.c. It follows by Corollary 7.22 that $\mathbb{Q}$ collapses $\kappa$ to become $\omega_2$. And by Proposition 5.4, $\mathbb{Q}$ adds a partial square sequence on $S_i$, for all $i < \lambda^*$. Since $\mathbb{Q}$ is $\kappa$-c.c., it also preserves the stationarity of $S_i$. Hence, for all $i < \lambda^*$, $\mathbb{Q}$ forces that $S_i$ is a stationary subset of $\omega_2 \cap \mathrm{cof}(\omega_1)$ in the approachability ideal $I[\omega_2]$. See Corollary 11.22 below for more details. \bigskip We conclude this section by introducing a set $s^*$ which is analogous to the set $r^*$ from Parts I and II. \begin{definition} Let $A$ be an adequate set. Define $s^*(A)$ as the set of $i < \lambda^*$ such that for some $M$ and $N$ in $A$, $M \sim N$, $i \in M \cap N$, and the ordinal $\min((M \cap \kappa) \setminus \beta_{M,N})$ exists and is in $S_i$. \end{definition} Note that requirement (4) of Definition 10.6 is equivalent to the statement that $s^*(A_p) \subseteq \mathrm{dom}(F_p)$. Observe that if $A \subseteq B$, then $s^*(A) \subseteq s^*(B)$. The following is an analogue of Propositions 3.5 and 3.8 for $s^*$. \begin{proposition} Let $A$ be an adequate set, $N \in \mathcal X \cup \mathcal Y$ be simple, and suppose that: \begin{enumerate} \item if $N \in \mathcal X$, then $N \in A$ and for all $M \in A$, if $M < N$ then $M \cap N \in A$; \item if $N \in \mathcal Y$, then for all $M \in A$, $M \cap N \in A$. \end{enumerate} Let $B$ be an adequate set such that $$ A \cap N \subseteq B \subseteq N. $$ Then $$ s^*(A \cup B) = s^*(A) \cup s^*(B). $$ \end{proposition} \begin{proof} By Propositions 2.25 and 2.28, $A \cup B$ is adequate. The reverse inclusion is immediate. For the forward inclusion, let $K \in A$ and $M \in B$, and assume that $K \sim M$, $i \in K \cap M \cap \lambda^*$, $\gamma$ is equal to either $\min((K \cap \kappa) \setminus \beta_{K,M})$ or $\min((M \cap \kappa) \setminus \beta_{K,M})$, and $\gamma \in S_i$. We will show that $i \in s^*(A) \cup s^*(B)$. We claim that $K \cap N \in B$. First, assume that $N \in \mathcal X$. Since $M \in N$, $M \cap \omega_1 < N \cap \omega_1$. As $K \sim M$, $K \cap \omega_1 = M \cap \omega_1$ by Lemma 2.17(2). So $K \cap \omega_1 < N \cap \omega_1$, and hence $K < N$ by Lemma 2.17(1). By (1), $K \cap N \in A$. As $N$ is simple, $K \cap N \in N$. So $K \cap N \in A \cap N \subseteq B$. Secondly, assume that $N \in \mathcal Y$. Then since $K \in A$, $K \cap N \in A$ by (2). As $N$ is simple, $K \cap N \in N$. So $K \cap N \in A \cap N \subseteq B$. Since $M \in N$, $i \in N$. So $i \in (K \cap N) \cap M$. Also note that by Lemmas 2.18, 2.26, and 2.29, $K \sim K \cap N \sim M$. We consider the two possibilities for $\gamma$. Suppose that $\gamma = \min((K \cap \kappa) \setminus \beta_{K,M})$. Then by Lemma 3.4(3) in the case that $N \in \mathcal X$ and Lemma 3.7(3) in the case that $N \in \mathcal Y$, either $\gamma = \min((K \cap N \cap \kappa) \setminus \beta_{K \cap N,M})$, or $\gamma = \min((K \cap \kappa) \setminus \beta_{K,K \cap N})$. In the first case, since $\gamma \in S_i$, $i \in (K \cap N) \cap M$, and $K \cap N$ and $M$ are in $B$, it follows that $i \in s^*(B)$. In second case, since $\gamma \in S_i$, $i \in (K \cap N) \cap K$, and $K \cap N$ and $K$ are in $A$, we have that $i \in s^*(A)$. Now suppose that $\gamma = \min((M \cap \kappa) \setminus \beta_{K,M})$. Then by Lemma 3.4(2) in the case that $N \in \mathcal X$ and Lemma 3.7(2) in the case that $N \in \mathcal Y$, $\gamma = \min((M \cap \kappa) \setminus \beta_{M,K \cap N})$. Since $\gamma \in S_i$, $i \in (K \cap N) \cap M$, and $K \cap N$ and $M$ are in $B$, it follows that $i \in s^*(B)$. \end{proof} \bigskip \addcontentsline{toc}{section}{11. Amalgamation} \textbf{\S 11. Amalgamation} \stepcounter{section} \bigskip In this section we will prove cardinal preservation results for $\mathbb{Q}$, namely, that $\mathbb{Q}$ is strongly proper on a stationary set and is $\kappa$-c.c. The arguments are simpler than those from Sections 6 and 7, and as a result we are able to handle the amalgamation arguments for countable and uncountable models at the same time. The order of topics and results is similar to that of Sections 6 and 7. \begin{lemma} Let $p \in \mathbb{Q}$, $N \in \mathcal X$, and suppose that $p \in N$. Then there is $r \le p$ such that $N \in A_r$. \end{lemma} \begin{proof} Since $p \in N$, for all $i \in \mathrm{dom}(F_p)$, $F_p(i) \in N$. So we can apply Lemma 7.1 and fix, for each $i \in \mathrm{dom}(F_p)$, a condition $q_i \le F_p(i)$ in $\mathbb{P}_i$ such that $N \in A_{q_i}$. Now apply Lemma 10.9 and fix $q \le p$ such that $A_q = A_p$, $\mathrm{dom}(F_q) = \mathrm{dom}(F_p)$, and for each $i \in \mathrm{dom}(F_p)$, $F_q(i) = q_i$. Define $r$ by letting $F_r := F_q$ and $A_r := A_q \cup \{ N \}$. It is easy to see that if $r$ is a condition, then $r \le q$ and $N \in A_r$. So we will be done if we can prove that $r$ is a condition. We verify requirements (1)--(4) of Definition 10.6. (1) Since $A_p = A_q$ and $p \in N$, $A_r$ is adequate by Lemma 2.16. (2) is immediate. (4) Since $M \in N$ for all $M \in A_q$, easily $s^*(A_r) = s^*(A_q \cup \{ N \}) = s^*(A_q)$. Since $q$ is a condition, $s^*(A_q) \subseteq \mathrm{dom}(F_q) = \mathrm{dom}(F_r)$. (3) Let $i \in \mathrm{dom}(F_r)$. Then $F_r(i) = F_q(i) = q_i$, which is in $\mathbb{P}_i$. Let $M \in A_r = A_q \cup \{ N \}$ and suppose that $i \in M$. We will show that $M \in A_{q_i}$. Since $M \in A_r$, either $M \in A_q$ or $M = N$. First, assume that $M = N$. Then by the choice of $q_i$, $M = N \in A_{q_i}$, and we are done. Secondly, assume that $M \in A_q$. Now $A_q = A_p$ and $\mathrm{dom}(F_r) = \mathrm{dom}(F_q) = \mathrm{dom}(F_p)$. So $M \in A_p$ and $i \in \mathrm{dom}(F_p)$. Since $p$ is a condition, $M \in A_{F_p(i)}$. But $q_i \le F_p(i)$, so $A_{F_p(i)} \subseteq A_{q_i}$. Hence, $M \in A_{q_i}$. \end{proof} \begin{lemma} Let $p \in \mathbb{Q}$ and $N \in \mathcal X \cup \mathcal Y$. Suppose that if $N \in \mathcal X$, then $N \in A_p$. Then there is $s \le p$ satisfying: \begin{enumerate} \item if $N \in \mathcal X$, then for all $M \in A_s$, if $M < N$ then $M \cap N \in A_s$, and moreover, $A_s = A_p \cup \{ M \cap N : M \in A_p, \ M < N \}$; \item if $N \in \mathcal Y$, then for all $M \in A_s$, $M \cap N \in A_s$, and moreover, $A_s = A_p \cup \{ M \cap N : M \in A_p \}$. \end{enumerate} \end{lemma} \begin{proof} Define $$ B := A_p \cup \{ M \cap N : M \in A_p, \ M < N \} $$ in the case that $N \in \mathcal X$, and $$ B := A_p \cup \{ M \cap N : M \in A_p \} $$ in the case that $N \in \mathcal Y$. By Propositions 2.24 and 2.27, $B$ is adequate. Define $$ q := p \uplus (s^*(B) \setminus \mathrm{dom}(F_p)). $$ By Definition 10.7 and Lemma 10.8, $q \in \mathbb{Q}$, $q \le p$, $A_q = A_p$, and $\mathrm{dom}(F_q) = \mathrm{dom}(F_p) \cup s^*(B)$. If $N \in \mathcal X$, then since $N \in A_q$, Definition 10.6(3) implies that for all $i \in \mathrm{dom}(F_q) \cap N$, $N \in A_{F_q(i)}$. Applying Lemma 7.2 in the case that $N \in \mathcal X$ and Lemma 6.1 in the case that $N \in \mathcal Y$, we can fix, for each $i \in \mathrm{dom}(F_q) \cap N$, a condition $r_i \le F_q(i)$ in $\mathbb{P}_i$ satisfying: \begin{enumerate} \item if $N \in \mathcal X$, then for all $M \in A_{r_i}$, if $M < N$ then $M \cap N \in A_{r_i}$; \item if $N \in \mathcal Y$, then for all $M \in A_{r_i}$, $M \cap N \in A_{r_i}$. \end{enumerate} Now apply Lemma 10.9 and fix $r \le q$ such that $A_r = A_q$, $\mathrm{dom}(F_r) = \mathrm{dom}(F_q)$, for each $i \in \mathrm{dom}(F_q) \cap N$, $F_r(i) = r_i$, and for each $i \in \mathrm{dom}(F_q) \setminus N$, $F_r(i) = F_q(i)$. Finally, define $$ s := (F_r, B). $$ We claim that $s$ is as required. Note that if $s$ is a condition, then clearly $s \le r$. Also, by Propositions 2.24 and 2.27, since $A_p = A_q = A_r$, $A_s = B$ satisfies (1) and (2) of the lemma. It remains to show that $s$ is a condition. We verify requirements (1)--(4) of Definition 10.6. (1) We already observed that $B$ is adequate. (2) is immediate. (4) We have that $$ s^*(A_s) = s^*(B) \subseteq \mathrm{dom}(F_q) = \mathrm{dom}(F_r) = \mathrm{dom}(F_s). $$ (3) Let $i \in \mathrm{dom}(F_s)$. Then $F_s(i) = F_r(i) \in \mathbb{P}_i$, since $r$ is a condition. Suppose that $M \in A_s = B$ and $i \in M$. We will show that $M \in A_{F_s(i)}$. If $M \in A_p = A_r$, then since $i \in \mathrm{dom}(F_s) = \mathrm{dom}(F_r)$, it follows that $M \in A_{F_r(i)} = A_{F_s(i)}$ since $r$ is a condition. Suppose that $M \in B \setminus A_p$. Then $M = M_1 \cap N$ for some $M_1 \in A_p$, where $M_1 < N$ in the case that $N \in \mathcal X$. Then $i \in M = M_1 \cap N$, so $i \in M_1$ and $i \in N$. Now $M_1$ is in $A_p = A_r$. Since $i$ is in $M_1 \cap \mathrm{dom}(F_r)$ and $r$ is a condition, it follows that $M_1$ is in $A_{F_r(i)}$. Also, $i \in \mathrm{dom}(F_r) \cap N = \mathrm{dom}(F_q) \cap N$, so by the choice of $r$, $F_r(i) = r_i$. And by the choice of $r_i$, $M_1 \cap N \in A_{r_i}$. So $M = M_1 \cap N \in A_{r_i} = A_{F_r(i)} = A_{F_s(i)}$. \end{proof} \begin{notation} Let $N \in \mathcal X$. For each $i < \lambda^*$, let $D_{i,N}$ denote the set of conditions in $\mathbb{P}_i$ defined as $D_N$ in Definition 7.3. \end{notation} We introduce an analogue $D(N)$ of $D_{i,N}$ for $\mathbb{Q}$. \begin{definition} Let $N \in \mathcal X$. Define $D(N)$ as the set of conditions $p \in \mathbb{Q}$ satisfying: \begin{enumerate} \item $N \in A_p$; \item for all $M \in A_p$, if $M < N$ then $M \cap N \in A_p$; \item for all $i \in \mathrm{dom}(F_p) \cap N$, $F_p(i) \in D_{i,N}$. \end{enumerate} \end{definition} \begin{lemma} Let $N \in \mathcal X$. Then for any condition $p \in \mathbb{Q}$, if $N \in A_p$, then there is $r \le p$ such that $r \in D(N)$. \end{lemma} \begin{proof} Let $p \in \mathbb{P}$. By Lemma 11.2(1), fix $q \le p$ such that for all $M \in A_q$, if $M < N$ then $M \cap N \in A_q$. Since $N \in A_q$, for each $i \in \mathrm{dom}(F_q) \cap N$, $N \in A_{F_q(i)}$ by Definition 10.6(3). So by Lemma 7.4, for each $i \in \mathrm{dom}(F_q) \cap N$, we can fix $r_i \le F_q(i)$ such that $r_i \in D_{i,N}$. By Lemma 10.9, there is $r \le q$ such that $A_r = A_q$, $\mathrm{dom}(F_r) = \mathrm{dom}(F_q)$, for all $i \in \mathrm{dom}(F_r) \cap N$, $F_r(i) = r_i$, and for all $i \in \mathrm{dom}(F_r) \setminus N$, $F_r(i) = F_q(i)$. Then $r \le p$ and $r \in D(N)$. \end{proof} \begin{lemma} Let $N \in \mathcal X$ and $q \in D(N)$. Let $x$ be a finite subset of $\lambda^* \setminus \mathrm{dom}(F_q)$. Then $q \uplus x \in D(N)$. \end{lemma} \begin{proof} Let $r := q \uplus x$. Since $q \in D(N)$ and $A_q = A_r$, we have that $N \in A_r$, and for all $M \in A_r$, if $M < N$ then $M \cap N \in A_r$. Also, since $q \in D(N)$, for all $i \in \mathrm{dom}(F_q) \cap N$, $F_r(i) = F_q(i) \in D_{i,N}$. It remains to show that for all $i \in x \cap N$, $F_r(i) \in D_{i,N}$. Let $i \in x \cap N$. By Definition 10.7, $F_r(i) = (\emptyset,\emptyset,B_i)$, where $B_i = \{ M \in A_q : i \in M \}$. We will verify that $(\emptyset,\emptyset,B_i)$ satisfies requirements (1), (2), and (3) of Definition 7.3. (1) Since $q \in D(N)$, $N \in A_q$. As $i \in N$, $N \in B_i$ by definition. (3) is immediate, since $f_{F_r(i)} = \emptyset$. (2) Suppose that $M \in B_i$ and $M < N$, and we will show that $M \cap N \in B_i$. By the definition of $B_i$, $M \in A_q$ and $i \in M$. So $i \in M \cap N$. Since $q \in D(N)$ and $M < N$, $M \cap N \in A_q$. So $M \cap N \in A_q$ and $i \in M \cap N$, which by definition implies that $M \cap N \in B_i$. \end{proof} \begin{notation} Let $P \in \mathcal Y$. For each $i < \lambda^*$, let $D_{i,P}$ denote the set of conditions in $\mathbb{P}_i$ defined as $D_P$ in Definition 6.2. \end{notation} We introduce an analogue $D(P)$ of $D_{i,P}$ for $\mathbb{Q}$. \begin{definition} Let $P \in \mathcal Y$. Define $D(P)$ as the set of conditions $p \in \mathbb{Q}$ satisfying: \begin{enumerate} \item for all $M \in A_p$, $M \cap P \in A_p$; \item for all $i \in \mathrm{dom}(F_p) \cap P$, $F_p(i) \in D_{i,P}$. \end{enumerate} \end{definition} \begin{lemma} Let $P \in \mathcal Y$. Then $D(P)$ is dense in $\mathbb{Q}$. \end{lemma} \begin{proof} Let $p \in \mathbb{Q}$. By Lemma 11.2(2), we can find $q \le p$ such that for all $M \in A_q$, $M \cap P \in A_q$. Let $x := \mathrm{dom}(F_q) \cap P$. For each $i \in x$, the set $D_{i,P}$ is dense in $\mathbb{P}_i$ by Lemma 6.3. By Lemma 10.10, fix $r \le q$ such that $A_r = A_q$, $\mathrm{dom}(F_r) = \mathrm{dom}(F_q) \cup x = \mathrm{dom}(F_q)$, and for each $i \in x$, $F_r(i) \in D_{i,P}$. Then $r \le p$ and $r \in D(P)$. \end{proof} \begin{lemma} Let $P \in \mathcal Y$ and $q \in D(P)$. Let $x$ be a finite subset of $\lambda^* \setminus \mathrm{dom}(F_q)$. Then $q \uplus x \in D(P)$. \end{lemma} \begin{proof} Let $r := q \uplus x$. Then by Definition 10.7, $A_r = A_q$. Since $q \in D(P)$, it follows that for all $M \in A_r$, $M \cap P \in A_r$. It remains to show that for all $i \in \mathrm{dom}(F_r) \cap P$, $F_r(i) \in D_{i,P}$. By Definition 10.7, $\mathrm{dom}(F_r) = \mathrm{dom}(F_q) \cup x$, for all $i \in \mathrm{dom}(F_q)$, $F_r(i) = F_q(i)$, and for all $i \in x$, $F_r(i) = (\emptyset,\emptyset,B_i)$, where $B_i = \{ M \in A_q : i \in M \}$. Since $q \in D(P)$, for all $i \in \mathrm{dom}(F_q) \cap P$, $F_r(i) = F_q(i) \in D_{i,P}$. It remains to show that for all $i \in x \cap P$, $F_r(i) = (\emptyset,\emptyset,B_i)$ is in $D_{i,P}$. By Definition 6.2, we need to show that for all $M \in B_i$, $M \cap P \in B_i$. So let $M \in B_i$. Then $M \in A_q$ and $i \in M$. So $i \in M \cap P$. Since $q \in D(P)$, $M \cap P \in A_q$. Hence, $M \cap P \in A_q$ and $i \in M \cap P$, which means that $M \cap P \in B_i$. \end{proof} \begin{definition} Suppose that $N \in \mathcal X \cup \mathcal Y$ is simple and $q \in D(N)$. Define $q \restriction N$ as the pair $(F,A)$ satisfying: \begin{enumerate} \item $A := A_q \cap N$; \item $\mathrm{dom}(F) := \mathrm{dom}(F_q) \cap N$; \item for all $i \in \mathrm{dom}(F)$, $F(i) := F_q(i) \restriction N$, as defined in Definition 7.5 if $N \in \mathcal X$, and as defined in Definition 6.4 if $N \in \mathcal Y$. \end{enumerate} \end{definition} Note that (3) makes sense because $F_q(i) \in D_{i,N}$, for all $i \in \mathrm{dom}(F_q) \cap N$. \begin{lemma} Suppose that $N \in \mathcal X \cup \mathcal Y$ is simple and $q \in D(N)$. Then $q \restriction N$ is in $N \cap \mathbb{Q}$ and $q \le q \restriction N$. \end{lemma} \begin{proof} Let $q \restriction N = (F,A)$. Then $A = A_q \cap N$ and $\mathrm{dom}(F) = \mathrm{dom}(F_q) \cap N$ are finite subsets of $N$, and hence are in $N$. For each $i \in \mathrm{dom}(F)$, $F(i) = F_q(i) \restriction N$ is in $N$ by Lemmas 6.5 and 7.6. So $F$ is in $N$. Since $A$ and $F$ are in $N$, so is $q \restriction N$. To prove that $q \restriction N$ is in $\mathbb{Q}$, we verify requirements (1)--(4) of Definition 10.6. (1) and (2) are immediate. For (3), let $i \in \mathrm{dom}(F)$. Then $F(i) = F_q(i) \restriction N$ is in $\mathbb{P}_i$ by Lemmas 6.5 and 7.6. Suppose that $M \in A$ and $i \in M$, and we will show that $M \in A_{F(i)}$. Then $M \in A = A_q \cap N$, so $M \in A_q$ and $M \in N$. Since $q$ is a condition, the fact that $M \in A_q$ and $i \in M \cap \mathrm{dom}(F_q)$ implies that $M \in A_{F_q(i)}$. Since $F(i) = F_q(i) \restriction N$, Definitions 6.4 and 7.5 imply that $$ M \in A_{F_q(i)} \cap N = A_{F_q(i) \restriction N} = A_{F(i)}. $$ For (4), suppose that $K$ and $L$ are in $A$, $K \sim L$, $i \in K \cap L \cap \lambda^*$, and the ordinal $\min((K \cap \kappa) \setminus \beta_{K,L})$ exists and is in $S_i$. We will show that $i \in \mathrm{dom}(F)$. Since $A = A_q \cap N$, $K$ and $L$ are in $A_q$ and in $N$. As $q$ is a condition, $i$ must be in $\mathrm{dom}(F_q)$. Since $K \in N$ and $i \in K$, we have that $i \in N$. So $i \in \mathrm{dom}(F_q) \cap N = \mathrm{dom}(F)$. This completes the proof that $q \restriction N$ is a condition. Now we show that $q \le q \restriction N$. We have that $A = A_q \cap N \subseteq A_q$ and $\mathrm{dom}(F) = \mathrm{dom}(F_q) \cap N \subseteq \mathrm{dom}(F_q)$. Let $i \in \mathrm{dom}(F)$, and we will show that $F_q(i) \le F(i)$ in $\mathbb{P}_i$. But $F(i) = F_q(i) \restriction N$ and $F_q(i) \le F_q(i) \restriction N$ in $\mathbb{P}_i$ by Lemmas 6.5 and 7.6. \end{proof} The next lemma will be needed in Section 12. \begin{lemma} Let $N \in \mathcal X \cup \mathcal Y$ be simple and $q \in D(N)$. \begin{enumerate} \item Suppose that $p \in N \cap \mathbb{Q}$ and $q \le p$. Then $q \restriction N \le p$. \item Suppose that $p$ is in $D(N)$ and $q \le p$. Then $q \restriction N \le p \restriction N$. \end{enumerate} \end{lemma} \begin{proof} (1) Since $p \in N$ and $q \le p$, we have that $A_{p} \subseteq A_q \cap N = A_{q \restriction N}$ and $\mathrm{dom}(F_p) \subseteq \mathrm{dom}(F_q) \cap N = \mathrm{dom}(F_{q \restriction N})$. Let $i \in \mathrm{dom}(F_p)$, and we will show that $F_{q \restriction N}(i) \le F_p(i)$ in $\mathbb{P}_i$. Since $q \le p$, we have that $F_q(i) \le F_p(i)$. As $F_p(i) \in N \cap \mathbb{P}_i$, it follows that $F_{q \restriction N}(i) = F_q(i) \restriction N \le F_p(i)$ by Lemmas 6.6(1) and 7.7. (2) By Lemma 11.12, $p \le p \restriction N$. So $q \le p \restriction N$. Now $p \restriction N \in N$, so by (1), $q \restriction N \le p \restriction N$. \end{proof} We will now begin analyzing the situation where $q \in D(N)$ and $w \le q \restriction N$ is in $N \cap \mathbb{Q}$. \begin{lemma} Let $N \in \mathcal X \cup \mathcal Y$ be simple and $q \in D(N)$. Suppose that $w \in N \cap \mathbb{Q}$ and $w \le q \restriction N$. Then: \begin{enumerate} \item $A_q \cap N \subseteq A_w$; \item $\mathrm{dom}(F_q) \cap N \subseteq \mathrm{dom}(F_w)$, and for all $i \in \mathrm{dom}(F_q) \cap N$, $F_w(i) \le F_q(i) \restriction N$ in $\mathbb{P}_i$. \end{enumerate} \end{lemma} \begin{proof} Immediate from the definition of $q \restriction N$ and the fact that $w \le q \restriction N$. \end{proof} Note that in (2) above, if $i \in \mathrm{dom}(F_q) \cap N$, then since $w \in N$, $F_w(i)$ is a condition in $N \cap \mathbb{P}_i$ which is below $F_q(i) \restriction N$ in $\mathbb{P}_i$. As $F_q(i) \in D_{i,N}$, it follows by Propositions 6.15 and 7.19 that $F_w(i) \oplus_N F_q(i)$ is a condition in $\mathbb{P}_i$ which is below $F_w(i)$ and $F_q(i)$. As in Sections 6 and 7, we are going to show that whenever $w \le q \restriction N$, where $q \in D(N)$ and $w \in N \cap \mathbb{Q}$, then $w$ and $q$ are compatible. We will define a specific lower bound of $w$ and $q$, namely, $w \oplus^N q$. However, unlike the situation in Sections 6 and 7, the condition $w \oplus^N q$ will exist only under the assumption that $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$. \begin{definition} Let $N \in \mathcal X \cup \mathcal Y$ be simple and $q \in D(N)$. Suppose that $w \in N \cap \mathbb{Q}$ and $w \le q \restriction N$. Assume, moreover, that $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$. Define $w \oplus^N q$ as the pair $(F,A)$ satisfying: \begin{enumerate} \item $A := A_w \cup A_q$; \item $\mathrm{dom}(F) := \mathrm{dom}(F_q)$; \item for all $i \in \mathrm{dom}(F_q) \setminus \mathrm{dom}(F_w)$, $F(i) := F_q(i)$, and for all $i \in \mathrm{dom}(F_w) \cap \mathrm{dom}(F_q)$, $F(i) := F_w(i) \oplus_N F_q(i)$, as defined in Definition 7.18 if $N \in \mathcal X$, and as defined in Definition 6.14 if $N \in \mathcal Y$. \end{enumerate} \end{definition} \begin{proposition} Let $N \in \mathcal X \cup \mathcal Y$ be simple and $q \in D(N)$. Suppose that $w \in N \cap \mathbb{Q}$ and $w \le q \restriction N$. Assume, moreover, that $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$. Then $w$ and $q$ are compatible. In fact, $w \oplus^N q$ is in $\mathbb{Q}$ and $w \oplus^N q \le w, q$. \end{proposition} \begin{proof} Let $w \oplus^N q = (F,A)$. To prove that $w \oplus^N q$ is a condition, we verify requirements (1)--(4) of Definition 10.6. For (1), the set $A = A_w \cup A_q$ is adequate by Propositions 2.25 and 2.28. For (2), obviously $F$ is a function whose domain is a finite subset of $\lambda^*$. For (4), by Proposition 10.13 we have that \begin{multline*} s^*(A) = s^*(A_w \cup A_q) = s^*(A_w) \cup s^*(A_q) \subseteq \\ \subseteq \mathrm{dom}(F_w) \cup \mathrm{dom}(F_q) = \mathrm{dom}(F_q) = \mathrm{dom}(F). \end{multline*} It remains to prove (3). Let $i \in \mathrm{dom}(F)$. If $i \in \mathrm{dom}(F_q) \setminus \mathrm{dom}(F_w)$, then $F(i) = F_q(i)$, which is in $\mathbb{P}_i$ since $q$ is a condition. If $i \in \mathrm{dom}(F_q) \cap \mathrm{dom}(F_w)$, then $F(i) = F_w(i) \oplus_N F_q(i)$, which is in $\mathbb{P}_i$ by Propositions 6.15 and 7.19. Assume that $i \in \mathrm{dom}(F)$, and we will show that $$ \{ M \in A_{w \oplus^N q} : i \in M \} \subseteq A_{F(i)}, $$ that is, $$ \{ M \in A_w \cup A_q : i \in M \} \subseteq A_{F(i)}. $$ First, assume that $i \in \mathrm{dom}(F_q) \setminus \mathrm{dom}(F_w)$, so $F(i) = F_q(i)$. Since $\mathrm{dom}(F_q) \cap N \subseteq \mathrm{dom}(F_w)$ by Lemma 11.14(2), it follows that $i \notin N$. In particular, if $M \in A_w$ then $M \in N$, so $i$ cannot be in $M$ since otherwise it would be in $N$. It follows that $$ \{ M \in A_w \cup A_q : i \in M \} = \{ M \in A_q : i \in M \}. $$ Since $F(i) = F_q(i)$, $q$ being a condition implies that $$ \{ M \in A_q : i \in M \} \subseteq A_{F_q(i)} = A_{F(i)}. $$ Secondly, assume that $i \in \mathrm{dom}(F_q) \cap \mathrm{dom}(F_w)$. Then $F(i) = F_w(i) \oplus_N F_q(i)$. By Definitions 6.14 and 7.18, $$ A_{F(i)} = A_{F_w(i)} \cup A_{F_q(i)}. $$ Since $w$ and $q$ are conditions, $$ \{ M \in A_w : i \in M \} \subseteq A_{F_w(i)}, \ \ \ \{ M \in A_q : i \in M \} \subseteq A_{F_q(i)}. $$ Therefore, $$ \{ M \in A_w \cup A_q : i \in M \} \subseteq A_{F_w(i)} \cup A_{F_q(i)} = A_{F(i)}. $$ This completes the proof that $w \oplus^N q$ is in $\mathbb{Q}$. \bigskip It remains to show that $w \oplus^N q \le q, w$. First, we prove that $w \oplus^N q \le w$. We have that $A_w \subseteq A_w \cup A_q = A$. Since $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$ by assumption, it follows that $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q) = \mathrm{dom}(F)$. Let $i \in \mathrm{dom}(F_w)$, and we will show that $F(i) \le F_w(i)$ in $\mathbb{P}_i$. But $F(i) = F_w(i) \oplus_N F_q(i)$, which is less than or equal to $F_w(i)$ in $\mathbb{P}_i$ by Propositions 6.15 and 7.19. Secondly, we prove that $w \oplus^N q \le q$. We have that $A_q \subseteq A_w \cup A_q = A$, and $\mathrm{dom}(F) = \mathrm{dom}(F_q)$. Let $i \in \mathrm{dom}(F_q)$, and we will show that $F(i) \le F_q(i)$ in $\mathbb{P}_i$. If $i \notin \mathrm{dom}(F_w)$, then $F(i) = F_q(i)$, and we are done. If $i \in \mathrm{dom}(F_w)$, then $F(i) = F_w(i) \oplus_N F_q(i)$, which is less than or equal to $F_q(i)$ in $\mathbb{P}_i$ by Propositions 6.15 and 7.19. \end{proof} In the above amalgamation result, we assumed that $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$. To prove the amalgamation result in general, we need a lemma. \begin{lemma} Suppose that $N \in \mathcal X \cup \mathcal Y$ is simple and $q \in D(N)$. Let $x$ be a finite subset of $\lambda^* \setminus \mathrm{dom}(F_q)$. Then $$ (q \uplus x) \restriction N = (q \restriction N) \uplus (x \cap N). $$ \end{lemma} Recall that by Lemmas 11.6 and 11.10, $q \uplus x$ is in $D(N)$. \begin{proof} By Definitions 10.7 and 11.11, $$ A_{(q \uplus x) \restriction N} = A_{q \uplus x} \cap N = A_q \cap N = A_{(q \restriction N)} = A_{(q \restriction N) \uplus (x \cap N)}. $$ Also, \begin{multline*} \mathrm{dom}(F_{(q \uplus x) \restriction N}) = \mathrm{dom}(F_{q \uplus x}) \cap N = (\mathrm{dom}(F_q) \cup x) \cap N = \\ (\mathrm{dom}(F_q) \cap N) \cup (x \cap N) = \mathrm{dom}(F_{q \restriction N}) \cup (x \cap N) = \mathrm{dom}(F_{(q \restriction N) \uplus (x \cap N)}). \end{multline*} Let $i \in \mathrm{dom}(F_{(q \uplus x) \restriction N})$, and we will show that $$ F_{(q \uplus x) \restriction N}(i) = F_{(q \restriction N) \uplus (x \cap N)}(i). $$ By the above equalities, we have that either $i \in \mathrm{dom}(F_{q \restriction N})$ or $i \in x \cap N$. First, assume that $i \in \mathrm{dom}(F_{q \restriction N})$. Then by Definitions 10.7 and 11.11, we have that $F_{(q \restriction N) \uplus (x \cap N)}(i) = F_{q \restriction N}(i) = F_q(i) \restriction N$. On the other hand, since $\mathrm{dom}(F_{q \restriction N}) = \mathrm{dom}(F_q) \cap N$, it follows that $i \in \mathrm{dom}(F_q)$, and hence $F_{(q \uplus x) \restriction N}(i) = F_{q \uplus x}(i) \restriction N = F_q(i) \restriction N$. Now assume that $i \in x \cap N$. Then by definition, $$ F_{(q \restriction N) \uplus (x \cap N)}(i) = (\emptyset,\emptyset,B), $$ where $$ B = \{ M \in A_{q \restriction N} : i \in M \}. $$ Also, $$ F_{(q \uplus x) \restriction N}(i) = F_{q \uplus x}(i) \restriction N = (\emptyset,\emptyset,C) \restriction N, $$ where $$ C = \{ M \in A_q : i \in M \}. $$ Hence, it suffices to show that $$ (\emptyset,\emptyset,B) = (\emptyset,\emptyset,C) \restriction N. $$ By Definitions 6.4 and 7.5, $$ (\emptyset,\emptyset,C) \restriction N = (\emptyset,\emptyset,C \cap N). $$ So it suffices to show that $$ B = C \cap N. $$ But $M \in B$ iff ($M \in A_{q \restriction N} = A_q \cap N$ and $i \in M$) iff $M \in C \cap N$. \end{proof} \begin{proposition} Let $N \in \mathcal X \cup \mathcal Y$ be simple and $q \in D(N)$. Then for all $w \le q \restriction N$ in $N \cap \mathbb{Q}$, $w$ and $q$ are compatible. In fact, let $x := \mathrm{dom}(F_w) \setminus \mathrm{dom}(F_q)$. Then $w \le (q \uplus x) \restriction N$, $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_{q \uplus x})$, and $w \oplus^N (q \uplus x)$ is less than or equal to $w$, $q \uplus x$, and $q$. \end{proposition} \begin{proof} Note that $x \subseteq N$. By Lemma 11.17, $$ (q \uplus x) \restriction N = (q \restriction N) \uplus (x \cap N). $$ By Lemma 10.8(4), since $w \le q \restriction N$ and $x \cap N = x \subseteq \mathrm{dom}(F_w)$, it follows that $$ w \le (q \restriction N) \uplus (x \cap N). $$ Hence, $w \le (q \uplus x) \restriction N$. As $x = \mathrm{dom}(F_w) \setminus \mathrm{dom}(F_q)$, clearly $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q) \cup x = \mathrm{dom}(F_{q \uplus x})$. By Proposition 11.16, it follows that $w \oplus^N (q \uplus x)$ is a condition which is less than or equal to $w$ and $q \uplus x$. Since $q \uplus x \le q$, also $w \oplus^N (q \uplus x) \le q$. \end{proof} \begin{corollary} The forcing poset $\mathbb{Q}$ is strongly proper on a stationary set. In particular, it preserves $\omega_1$. \end{corollary} \begin{proof} By Assumption 2.22, the set of $N \in \mathcal X$ such that $N$ is simple is stationary. So it suffices to show that for all simple $N \in \mathcal X$, for all $p \in N \cap \mathbb{Q}$, there is $q \le p$ such that $q$ is strongly $N$-generic. Let $p \in N \cap \mathbb{Q}$. By Lemma 11.1, fix $q \le p$ with $N \in A_q$. We claim that $q$ is strongly $N$-generic. So let $D$ be a dense subset of $N \cap \mathbb{Q}$, and we will show that $D$ is predense below $q$. Let $r \le q$, and we will find $w \in D$ which is compatible with $r$. Since $N \in A_r$, we can apply Lemma 11.5 to fix $s \le r$ such that $s \in D(N)$. Then by Lemma 11.12, $s \restriction N$ is in $N \cap \mathbb{Q}$. As $D$ is dense in $N \cap \mathbb{Q}$, fix $w \le s \restriction N$ in $D$. By Proposition 11.18, $w$ and $s$ are compatible. Since $s \le r$, it follows that $w$ and $r$ are compatible. \end{proof} \begin{corollary} Suppose that $P \in \mathcal Y$ is simple and $P \prec (H(\lambda),\in,\mathbb{Q})$. Then the maximum condition of $\mathbb{Q}$ is strongly $P$-generic. Moreover, $P \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$. \end{corollary} \begin{proof} Let $D$ be a dense subset of $P \cap \mathbb{Q}$, and we will show that $D$ is predense in $\mathbb{Q}$. So let $p \in \mathbb{Q}$, and we will find $w$ in $D$ which is compatible with $p$. By Lemma 11.9, fix $q \le p$ in $D(P)$. Then $q \restriction P$ is in $P \cap \mathbb{Q}$ by Lemma 11.12. Since $D$ is dense in $P \cap \mathbb{Q}$, fix $w \le q \restriction P$ in $D$. By Proposition 11.18, $w$ and $q$ are compatible. Since $q \le p$, it follows that $w$ and $p$ are compatible. This completes the proof that the maximum condition in $\mathbb{Q}$ is strongly $P$-generic. Now we show that $P \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$. If $p$ and $q$ are in $P \cap \mathbb{Q}$ and are compatible in $\mathbb{Q}$, then by the elementarity of $P$, there is $r \in P \cap \mathbb{Q}$ with $r \le p, q$. So $p$ and $q$ are compatible in $P \cap \mathbb{Q}$. Let $B \subseteq P \cap \mathbb{Q}$ be a maximal antichain of $P \cap \mathbb{Q}$, and we will prove that $B$ is predense in $\mathbb{Q}$. Let $D$ be the set of conditions in $P \cap \mathbb{Q}$ which are below some member of $B$. Then $D$ is dense in $P \cap \mathbb{Q}$. Since the maximum condition is strongly $P$-generic, $D$ is predense in $\mathbb{Q}$. It easily follows that $B$ is predense in $\mathbb{Q}$. \end{proof} \begin{corollary} The forcing poset $\mathbb{Q}$ is $\kappa$-c.c. \end{corollary} \begin{proof} Let $A$ be an antichain of $\mathbb{Q}$, and suppose for a contradiction that $A$ has size at least $\kappa$. Without loss of generality, assume that $A$ is maximal. By Assumption 2.23, there are stationarily many simple models in $\mathcal Y$. So we can fix a simple model $P \in \mathcal Y$ such that $P \prec (H(\lambda),\in,\mathbb{Q},A)$. As $A$ has size at least $\kappa$ and $|P| < \kappa$, fix $s \in A \setminus P$. By Lemma 11.9, fix $q \le s$ such that $q \in D(P)$. By Lemma 11.12, $q \restriction P$ is a condition in $P \cap \mathbb{Q}$. By the elementarity of $P$ and the maximality of $A$, there is $t \in A \cap P$ which is compatible with $q \restriction P$. By elementarity, fix $w \in P \cap \mathbb{Q}$ such that $w \le q \restriction P, t$. By Proposition 11.18, $w$ and $q$ are compatible. Fix $v \le w, q$. Then $v \le w \le t$ and $v \le q \le s$. Hence, $s$ and $t$ are compatible. But $s$ and $t$ are in $A$ and $A$ is an antichain. Therefore, $s = t$. This is impossible, since $t \in P$ and $s \notin P$. \end{proof} The next result summarizes the main properties which we have proven about the forcing poset $\mathbb{Q}$. \begin{corollary} The forcing poset $\mathbb{Q}$ preserves $\omega_1$, is $\kappa$-c.c., forces that $\kappa$ is equal to $\omega_2$, and forces that for all $i < \lambda^*$, $S_i \in I[\omega_2]$. \end{corollary} \begin{proof} By Corollaries 11.19 and 11.21, $\mathbb{Q}$ preserves $\omega_1$ and is $\kappa$-c.c. Let $i < \lambda^*$, and consider a generic filter $G$ on $\mathbb{Q}$. Then by Proposition 10.11 and the comments which follow, there is a generic filter $H$ on $\mathbb{P}_i$ such that $V[H] \subseteq V[G]$. By Corollary 7.22, $\kappa$ is equal to $\omega_2$ in $V[H]$. Since $V[H] \subseteq V[G]$, it follows that any cardinal $\mu$ such that $\omega_1 < \mu < \kappa$ has size $\omega_1$ in $V[G]$. Therefore, $\kappa = \omega_2$ in $V[G]$. By Proposition 5.4, there is a partial square sequence on $S_i$ in $V[H]$. But being a partial square sequence is upwards absolute between $V[H]$ and $V[G]$, since they have the same $\omega_1$ and $\omega_2$. So there is a partial square sequence on $S_i$ in $V[G]$. Therefore, $S_i \in I[\omega_2]$ in $V[G]$. \end{proof} \bigskip \addcontentsline{toc}{section}{12. Approximation} \textbf{\S 12. Approximation} \stepcounter{section} \bigskip In this section we will prove that if $P \in \mathcal Y$ is simple, $P \prec (H(\lambda),\in,\mathbb{Q})$, and for all $i \in P \cap \lambda^*$, $P \cap \kappa \notin S_i$, then $P \cap \mathbb{Q}$ forces that $\mathbb{Q} / \dot G_{P \cap \mathbb{Q}}$ has the $\omega_1$-approximation property. The proof is similar to the analogous result given in Sections 8 and 9 for the forcing poset $\mathbb{P}$, albeit somewhat easier. The order of topics and results follows along the same lines as in those previous sections. \begin{lemma} Let $N \in \mathcal X$ be simple and $q \in D(N)$. Suppose that $v$ and $w$ are in $N \cap \mathbb{Q}$ and $$ w \le v \le q \restriction N. $$ Assume, moreover, that $\mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$. Then $w \oplus^N q \le v \oplus^N q$. \end{lemma} Note that since $w \le v$, $\mathrm{dom}(F_v) \subseteq \mathrm{dom}(F_w) \subseteq \mathrm{dom}(F_q)$. So $v \oplus^N q$ is defined. \begin{proof} Let $s := v \oplus^N q$ and $t := w \oplus^N q$. We will prove that $t \le s$. Since $w \le v$, $A_v \subseteq A_w$. By Definition 11.15, we have that $$ A_s = A_v \cup A_q \subseteq A_w \cup A_q = A_t. $$ So $A_s \subseteq A_t$. Also, by Definition 11.15, $\mathrm{dom}(F_s) = \mathrm{dom}(F_q) = \mathrm{dom}(F_t)$. Let $i \in \mathrm{dom}(F_s)$, and we will show that $F_t(i) \le F_s(i)$ in $\mathbb{P}_i$. First, assume that $i \notin N$. Then by Definition 11.15, $F_s(i) = F_q(i)$ and $F_t(i) = F_q(i)$, and we are done. Secondly, assume that $i \in N$. Then by Lemma 11.14, $i \in \mathrm{dom}(F_q) \cap N \subseteq \mathrm{dom}(F_v) \subseteq \mathrm{dom}(F_w)$. So by Definition 11.15, $F_s(i) = F_v(i) \oplus_N F_q(i)$ and $F_t(i) = F_w(i) \oplus_N F_q(i)$. Also, since $v$, $w$, and $i$ are in $N$, so are $F_v(i)$ and $F_w(i)$. As $q \in D(N)$ and $i \in \mathrm{dom}(F_q) \cap N$, $F_q(i) \in D_{i,N}$. Also, since $w \le v \le q \restriction N$, we have that in $\mathbb{P}_i$, $$ F_w(i) \le F_v(i) \le F_{q \restriction N}(i) = F_q(i) \restriction N. $$ By Lemma 8.1, it follows that $$ F_t(i) = F_w(i) \oplus_N F_q(i) \le F_v(i) \oplus_N F_q(i) = F_s(i). $$ \end{proof} \begin{lemma} Let $N \in \mathcal X$ and $P \in \mathcal Y$. Let $p \in \mathbb{Q}$, and suppose that $N \in A_p$. Then there is $s \le p$ such that $s \in D(N) \cap D(P)$. \end{lemma} \begin{proof} By Lemma 11.2(1), there is $q \le p$ such that for all $M \in A_q$, if $M < N$ then $M \cap N \in A_q$. By Lemma 11.2(2), there is $r \le q$ such that for all $M \in A_r$, $M \cap P \in A_r$, and moreover, $$ A_r = A_q \cup \{ M \cap P : M \in A_q \}. $$ We claim that for all $M \in A_r$, if $M < N$ then $M \cap N \in A_r$. This is certainly true if $M \in A_q$, so assume that $M = M_1 \cap P$, where $M_1 \in A_q$. By Lemma 2.29, $M_1 \sim M_1 \cap P = M$. Since $M < N$, it follows that $M_1 < N$ by Lemma 2.18. As $M_1 \in A_q$, we have that $M_1 \cap N \in A_q$ by the choice of $q$. Now $M \cap N = (M_1 \cap P) \cap N = (M_1 \cap N) \cap P$. But $M_1 \cap N \in A_q$ implies that $M \cap N = (M_1 \cap N) \cap P \in A_r$, by the definition of $A_r$. Since $r$ is a condition and $N \in A_r$, we have that for all $i \in \mathrm{dom}(F_r) \cap N$, $N \in A_{F_r(i)}$. Let $x_1 := \mathrm{dom}(F_r) \cap N$ and $x_2 := \mathrm{dom}(F_r) \cap P$. Then: \begin{enumerate} \item For each $i \in x_1 \setminus x_2$, since $N \in A_{F_r(i)}$ we can fix, by Lemma 7.4, a condition $s_i \le F_r(i)$ in $D_{i,N}$. \item For each $i \in x_1 \cap x_2$, since $N \in A_{F_r(i)}$ we can fix, by Lemma 8.2, a condition $s_i \le F_r(i)$ in $D_{i,N} \cap D_{i,P}$. \item For each $i \in x_2 \setminus x_1$, we can fix, by Lemma 6.3, a condition $s_i \le F_r(i)$ in $D_{i,P}$. \end{enumerate} Now apply Lemma 10.9 to fix $s \le r$ satisfying that $A_s = A_r$, $\mathrm{dom}(F_s) = \mathrm{dom}(F_r)$, for all $i \in x_1 \cup x_2$, $F_s(i) = s_i$, and for all $i \in \mathrm{dom}(F_r) \setminus (x_1 \cup x_2)$, $F_s(i) = F_r(i)$. Then $s \le p$ and $s \in D(N) \cap D(P)$. \end{proof} The next three lemmas will be used in the proof of Proposition 12.6. \begin{lemma} Let $N \in \mathcal X$ be simple, $P \in \mathcal Y \cap N$ be simple, and $p \in D(N) \cap D(P)$. Then $p \restriction N \in D(P)$ and $p \restriction P \in D(N \cap P)$. \end{lemma} \begin{proof} We prove first that $p \restriction N \in D(P)$, which means that for all $M \in A_{p \restriction N}$, $M \cap P \in A_{p \restriction N}$, and for all $i \in \mathrm{dom}(F_{p \restriction N}) \cap P$, $F_{p \restriction N}(i) \in D_{i,P}$. Let $M \in A_{p \restriction N}$. Then $M \in A_{p \restriction N} = A_p \cap N$, so $M \in A_p \cap N$. Since $p \in D(P)$, we have that $M \cap P \in A_p$. And as $M$ and $P$ are in $N$, $M \cap P \in N$. Therefore, $M \cap P \in A_p \cap N = A_{p \restriction N}$. Now let $i \in \mathrm{dom}(F_{p \restriction N}) \cap P = \mathrm{dom}(F_p) \cap N \cap P$, and we will show that $F_{p \restriction N}(i) \in D_{i,P}$. Since $i \in N \cap P$ and $p \in D(N) \cap D(P)$, $F_p(i) \in D_{i,N} \cap D_{i,P}$. By Lemma 8.3, it follows that $F_p(i) \restriction N \in D_{i,P}$. But $F_p(i) \restriction N = F_{p \restriction N}(i)$. This completes the proof that $p \restriction N \in D(P)$. Next, we prove that $p \restriction P \in D(N \cap P)$. First, we show that $N \cap P \in A_{p \restriction P}$. Since $p \in D(N)$, $N \in A_p$. As $p \in D(P)$, $N \cap P \in A_p$. Since $P$ is simple, $N \cap P \in P$. So $N \cap P \in A_p \cap P = A_{p \restriction P}$. Secondly, we prove that if $M \in A_{p \restriction P}$ and $M < N \cap P$, then $M \cap N \cap P \in A_{p \restriction P}$. Let $M \in A_{p \restriction P}$, and assume that $M < N \cap P$. Then $M \in A_{p \restriction P} = A_p \cap P$. By Lemma 2.29, $N \sim N \cap P$. Since $M < N \cap P$, it follows by Lemma 2.18 that $M < N$. Since $p \in D(N)$ and $M < N$, it follows that $M \cap N \in A_p$. And as $p \in D(P)$, $M \cap N \cap P \in A_p$. Since $P$ is simple, $M \cap N \cap P \in P$. So $M \cap N \cap P \in A_p \cap P = A_{p \restriction P}$. Thirdly, we show that if $i \in \mathrm{dom}(F_{p \restriction P}) \cap (N \cap P)$, then $F_{p \restriction P}(i) \in D_{i,N \cap P}$. Since $p \in D(P) \cap D(N)$ and $i \in N \cap P$, we have that $F_p(i) \in D_{i,N} \cap D_{i,P}$. By Lemma 8.3, $F_p(i) \restriction P$ is in $D_{i,N \cap P}$. But $F_p(i) \restriction P = F_{p \restriction P}(i)$. \end{proof} \begin{lemma} Let $N \in \mathcal X$ be simple, $P \in \mathcal Y \cap N$ be simple, and $p \in D(N) \cap D(P)$. Then $$ (p \restriction N) \restriction P = (p \restriction P) \restriction (N \cap P). $$ \end{lemma} Note that we needed Lemma 12.3 to see that $(p \restriction N) \restriction P$ and $(p \restriction P) \restriction (N \cap P)$ are defined. \begin{proof} By Definition 11.11, we have that \begin{multline*} A_{(p \restriction N) \restriction P} = A_{p \restriction N} \cap P = A_p \cap N \cap P = \\ (A_p \cap P) \cap (N \cap P) = A_{p \restriction P} \cap (N \cap P) = A_{(p \restriction P) \restriction (N \cap P)}. \end{multline*} And \begin{multline*} \mathrm{dom}(F_{(p \restriction N) \restriction P}) = \mathrm{dom}(F_{p \restriction N}) \cap P = \mathrm{dom}(F_p) \cap N \cap P = \\ (\mathrm{dom}(F_p) \cap P) \cap (N \cap P) = \mathrm{dom}(F_{p \restriction P}) \cap (N \cap P) = \mathrm{dom}(F_{(p \restriction P) \restriction (N \cap P)}). \end{multline*} Let $i \in \mathrm{dom}(F_{(p \restriction N) \restriction P})$, and we will show that $$ F_{(p \restriction N) \restriction P}(i) = F_{(p \restriction P) \restriction (N \cap P)}(i). $$ By the above equations, $i \in N \cap P$. Since $p \in D(N) \cap D(P)$, we have that $F_p(i) \in D_{i,N} \cap D_{i,P}$. So by Definition 11.11 and Lemma 8.4, \begin{multline*} F_{(p \restriction N) \restriction P}(i) = F_{p \restriction N}(i) \restriction P = (F_{p}(i) \restriction N) \restriction P = \\ (F_{p}(i) \restriction P) \restriction (N \cap P) = F_{p \restriction P}(i) \restriction (N \cap P) = F_{(p \restriction P) \restriction (N \cap P)}(i). \end{multline*} \end{proof} \begin{lemma} Let $N \in \mathcal X$ be simple, $P \in \mathcal Y \cap N$ be simple, and $p \in D(N) \cap D(P)$. Suppose that $q \in N \cap D(P)$, $q \le p \restriction N$, and $\mathrm{dom}(F_q) \subseteq \mathrm{dom}(F_p)$. Then: \begin{enumerate} \item $q \oplus^{N} p$ is in $D(P)$; \item $q \restriction P \in N \cap P$ and $$ q \restriction P \le (p \restriction P) \restriction (N \cap P). $$ \end{enumerate} \end{lemma} \begin{proof} (1) Let us prove that $q \oplus^N p$ is in $D(P)$, which means that for all $M \in A_{q \oplus^N p}$, $M \cap P \in A_{q \oplus^N p}$, and for all $i \in \mathrm{dom}(F_{q \oplus^N p}) \cap P$, $F_{q \oplus^N p}(i) \in D_{i,P}$. Now $A_{q \oplus^N p} = A_q \cup A_p$. So if $M \in A_{q \oplus^N p}$, then either $M \in A_q$ or $M \in A_p$. But $q$ and $p$ are both in $D(P)$, so in the first case, $M \cap P \in A_q$, and in the second case, $M \cap P \in A_p$. In either case, $M \cap P \in A_q \cup A_p = A_{q \oplus^N p}$. Now let $i \in \mathrm{dom}(F_{q \oplus^N p}) \cap P$, and we will show that $F_{q \oplus^N p}(i) \in D_{i,P}$. By Definition 11.15, $\mathrm{dom}(F_{q \oplus^N p}) = \mathrm{dom}(F_p)$, for all $i \in \mathrm{dom}(F_p) \setminus \mathrm{dom}(F_q)$, $F_{q \oplus^N p}(i) = F_p(i)$, and for all $i \in \mathrm{dom}(F_q) \cap \mathrm{dom}(F_p)$, $F_{q \oplus^N p}(i) = F_q(i) \oplus_N F_p(i)$. First, assume that $i \in \mathrm{dom}(F_p) \setminus \mathrm{dom}(F_q)$. Then $F_{q \oplus^N p}(i) = F_p(i)$. Since $p \in D(P)$ and $i \in \mathrm{dom}(F_p) \cap P$, we have that $F_p(i) \in D_{i,P}$. Secondly, assume that $i \in \mathrm{dom}(F_p) \cap \mathrm{dom}(F_q)$. Since $q \in N$, we have that $i \in N$. So $F_{q \oplus^N p}(i) = F_q(i) \oplus_N F_p(i)$. Thus, it suffices to show that $F_q(i) \oplus_N F_p(i) \in D_{i,P}$. This will follow from Lemma 8.5, provided that the assumptions of this lemma are true for $F_p(i)$ and $F_q(i)$. Since $i \in N \cap P$ and $p \in D(N) \cap D(P)$, $F_p(i) \in D_{i,N} \cap D_{i,P}$. As $q$ and $i$ are in $N$, $F_q(i) \in N$, and since $q \in D(P)$ and $i \in P$, $F_q(i) \in N \cap D_{i,P}$. Finally, as $q \le p \restriction N$, $F_q(i) \le F_{p \restriction N}(i) = F_p(i) \restriction N$. This completes the verification of the assumptions of Lemma 8.5. By Lemma 8.5(1), we have that $F_q(i) \oplus_N F_p(i)$ is in $D_{i,P}$. (2) Since $q$ and $P$ are in $N$, $q \restriction P \in N$. Also, $q \restriction P \in P$, so $q \restriction P \in N \cap P$. By Lemmas 12.3 and 12.4, we have that $p \restriction N \in D(P)$ and $$ (p \restriction N) \restriction P = (p \restriction P) \restriction (N \cap P). $$ As $q \le p \restriction N$, it follows by Lemma 11.13(2) that $$ q \restriction P \le (p \restriction N) \restriction P = (p \restriction P) \restriction (N \cap P). $$ \end{proof} \begin{proposition} Let $N \in \mathcal X$ be simple, $P \in \mathcal Y \cap N$ be simple, and suppose that $P \cap \kappa \notin S_i$, for all $i \in P \cap \lambda^*$. Let $p \in D(N) \cap D(P)$, $q \in N \cap D(P)$, and $q \le p \restriction N$. Assume, moreover, that $\mathrm{dom}(F_q) \subseteq \mathrm{dom}(F_p)$. Then $$ (q \oplus^N p) \restriction P = (q \restriction P) \oplus^{N \cap P} (p \restriction P). $$ \end{proposition} Note that since $\mathrm{dom}(F_q) \subseteq \mathrm{dom}(F_p)$, we also have that $$ \mathrm{dom}(F_{q \restriction P}) = \mathrm{dom}(F_q) \cap P \subseteq \mathrm{dom}(F_p) \cap P = \mathrm{dom}(F_{p \restriction P}). $$ By this fact and Lemma 12.5, it follows that $(q \oplus^N p) \restriction P$ and $(q \restriction P) \oplus^{N \cap P} (p \restriction P)$ are defined. \begin{proof} Let $$ s := (q \oplus^N p) \restriction P $$ and $$ t := (q \restriction P) \oplus^{N \cap P} (p \restriction P). $$ Our goal is to prove that $s = t$. \bigskip We have that \begin{multline*} A_s = A_{(q \oplus^N p) \restriction P} = A_{q \oplus^N p} \cap P = (A_q \cup A_p) \cap P = \\ = (A_q \cap P) \cup (A_p \cap P) = A_{q \restriction P} \cup A_{p \restriction P} = A_{(q \restriction P) \oplus^{N \cap P} (p \restriction P)} = A_t. \end{multline*} Thus, $A_s = A_t$. Similarly, \begin{multline*} \mathrm{dom}(F_s) = \mathrm{dom}(F_{(q \oplus^N p) \restriction P}) = \mathrm{dom}(F_{q \oplus^N p}) \cap P = \\ = \mathrm{dom}(F_p) \cap P = \mathrm{dom}(F_{p \restriction P}) = \mathrm{dom}(F_{(q \restriction P) \oplus^{N \cap P} (p \restriction P)}) = \mathrm{dom}(F_t). \end{multline*} So $\mathrm{dom}(F_s) = \mathrm{dom}(F_t)$. Let $i \in \mathrm{dom}(F_s)$, and we will show that $F_s(i) = F_t(i)$. Note that $\mathrm{dom}(F_s) \subseteq P$, so $i \in P$. By definition, we have that $$ F_s(i) = F_{(q \oplus^N p) \restriction P}(i) = F_{q \oplus^N p}(i) \restriction P. $$ The definition of $F_{q \oplus^N p}(i)$ splits into two cases, depending on whether $i \in \mathrm{dom}(F_p) \setminus \mathrm{dom}(F_q)$, or $i \in \mathrm{dom}(F_p) \cap \mathrm{dom}(F_q)$. First, assume that $i \in \mathrm{dom}(F_p) \setminus \mathrm{dom}(F_q)$. Then $F_{q \oplus^N p}(i) = F_p(i)$ by Definition 11.15. Thus, by the above, $$ F_s(i) = F_p(i) \restriction P. $$ Since $i \notin \mathrm{dom}(F_q)$, also $i \notin \mathrm{dom}(F_{q}) \cap P = \mathrm{dom}(F_{q \restriction P})$. Thus, by Definition 11.15, $$ F_t(i) = F_{(q \restriction P) \oplus^{N \cap P} (p \restriction P)}(i) = F_{p \restriction P}(i) = F_p(i) \restriction P = F_s(i). $$ Secondly, assume that $i \in \mathrm{dom}(F_p) \cap \mathrm{dom}(F_q)$. Then $i \in N$. By Definition 11.15 and the above, $$ F_s(i) = F_{q \oplus^N p}(i) \restriction P = (F_q(i) \oplus_N F_p(i)) \restriction P. $$ Also, $i \in \mathrm{dom}(F_p) \cap \mathrm{dom}(F_q) \cap P = \mathrm{dom}(F_{p \restriction P}) \cap \mathrm{dom}(F_{q \restriction P})$. So by Definition 11.15, \begin{multline*} F_t(i) = F_{(q \restriction P) \oplus^{N \cap P} (p \restriction P)}(i) = F_{q \restriction P}(i) \oplus_{N \cap P} F_{p \restriction P}(i) = \\ (F_{q}(i) \restriction P) \oplus_{N \cap P} (F_p(i) \restriction P). \end{multline*} Thus, to show that $F_s(i) = F_t(i)$, it suffices to show that $$ (F_q(i) \oplus_N F_p(i)) \restriction P = (F_{q}(i) \restriction P) \oplus_{N \cap P} (F_p(i) \restriction P). $$ This equation follows immediately from Proposition 8.6 for the conditions $F_q(i)$ and $F_p(i)$, so it is enough to verify that the assumptions of Proposition 8.6 hold. Since $i \in P$, $P \cap \kappa \notin S_i$. As $i \in N \cap P$ and $p \in D(N) \cap D(P)$, $F_p(i) \in D_{i,N} \cap D_{i,P}$. Since $q \in N \cap D(P)$, $F_q(i) \in N \cap D_{i,P}$. And as $q \le p \restriction N$, $F_q(i) \le F_{p \restriction N}(i) = F_p(i) \restriction N$. Thus, all of the assumptions of Proposition 8.6 are true, and we are done. \end{proof} \begin{thm} Let $P \in \mathcal Y$ be simple, $P \prec (H(\lambda),\in,\mathbb{Q})$, and suppose that for all $i \in P \cap \lambda^*$, $P \cap \kappa \notin S_i$. Then $P \cap \mathbb{Q}$ forces that $\mathbb{Q} / \dot G_{P \cap \mathbb{Q}}$ has the $\omega_1$-approximation property. \end{thm} Recall that by Corollary 11.20, $P \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$. The proof of this theorem is almost identical in several places to the proof of Theorem 9.2. In those places, we will ask the reader to refer to the proof of Theorem 9.2 for some of the details instead of repeating everything here. \begin{proof} By Lemma 1.4, it suffices to show that $\mathbb{Q}$ forces that the pair $$ (V[\dot G_\mathbb{Q} \cap P],V[\dot G_\mathbb{Q}]) $$ has the $\omega_1$-approximation property. So let $p$, $\mu$, and $\dot k$ be given such that $\mu$ is an ordinal, and $p$ forces in $\mathbb{Q}$ that $\dot k : \mu \to On$ is a function satisfying that for any countable set $a$ in $V[\dot G_\mathbb{Q} \cap P]$, $\dot k \restriction a \in V[\dot G_\mathbb{Q} \cap P]$. We will find an extension of $p$ which forces that $\dot k$ is in $V[\dot G_\mathbb{Q} \cap P]$. Fix a regular cardinal $\theta$ large enough so that $\mathbb{Q}$, $\mu$, and $\dot k$ are members of $H(\theta)$. By the stationarity of the simple models in $\mathcal X$ as described in Assumption 2.22, fix a countable set $M^* \prec H(\theta)$ such that $M^*$ contains the parameters $\mathbb{Q}$, $P$, $p$, $\mu$, and $\dot k$, and satisfies that $M^* \cap H(\lambda)$ is in $\mathcal X$ and is simple. Let $M := M^* \cap H(\lambda)$. Note that since $\mathbb{Q} \subseteq H(\lambda)$, $M \cap \mathbb{Q} = M^* \cap \mathbb{Q}$. In particular, $p \in M \cap \mathbb{Q}$. Also, note that since $P \in H(\lambda)$, we have that $P \in M$ and $M \cap P = M^* \cap P$. By Lemma 11.1, fix $p_0 \le p$ such that $M \in A_{p_0}$. By the choice of $p$ and $\dot k$, and since $M^* \cap \mu$ is in $V$, we can fix $p_1 \le p_0$ and a $(P \cap \mathbb{Q})$-name $\dot s$ such that $$ p_1 \Vdash_\mathbb{Q} \dot k \restriction (M^* \cap \mu) = \dot s^{\dot G_\mathbb{Q} \cap P}. $$ Since $M \in A_{p_1}$, by Lemma 12.2 we can fix $p_2 \le p_1$ such that $p_2 \in D(M) \cap D(P)$. Since $p_2 \le p$ and $p \in M$, it follows that $p_2 \restriction M \le p$ by Lemma 11.13(1). So it suffices to prove that $p_2 \restriction M$ forces that $\dot k$ is in $V[\dot G_\mathbb{Q} \cap P]$. \bigskip \noindent \emph{Claim 1:} If $t \le p_2$ is in $D(P)$, $\nu \in M^* \cap \mu$, and $t \Vdash_{\mathbb{Q}} \dot k(\nu) = x$ (or $t \Vdash_{\mathbb{Q}} \dot k(\nu) \ne x$, respectively) then $t \restriction P \Vdash_{P \cap \mathbb{Q}} \dot s(\nu) = x$ (or $t \restriction P \Vdash_{P \cap \mathbb{Q}} \dot s(\nu) \ne x$, respectively). \bigskip The proof of Claim 1 is identical to the proof of Claim 1 of Theorem 9.2, except that the reference to Proposition 6.15 is replaced with a reference to Proposition 11.18. \bigskip \noindent \emph{Claim 2:} For all $q \le p_2 \restriction M$ in $D(P)$, $\nu < \mu$, and $x$, $$ q \Vdash_{\mathbb{Q}} \dot k(\nu) = x \implies \forall r \in \mathbb{Q} ( ( r \le p_2 \restriction M \land r \le q \restriction P ) \implies (r \Vdash_\mathbb{Q} \dot k(\nu) = x)). $$ \bigskip Note that $p_2 \restriction M$, $P$, $D(P)$, $\mu$, $\dot k$, and $\mathbb{Q}$ are in $M^*$. So by the elementarity of $M^*$, it suffices to show that the statement holds in $M^*$. Suppose for a contradiction that there exists $q \le p_2 \restriction M$ in $M^* \cap D(P)$, $\nu \in M^* \cap \mu$, and $x \in M^*$ such that $$ q \Vdash_\mathbb{Q} \dot k(\nu) = x, $$ but there is $r_0 \in M^* \cap \mathbb{Q}$ with $r_0 \le p_2 \restriction M$ and $r_0 \le q \restriction P$ such that $$ r_0 \not \Vdash_\mathbb{Q} \dot k(\nu) = x. $$ By the elementarity of $M^*$, we can fix $r \le r_0$ in $M^* \cap D(P)$ such that $$ r \Vdash_\mathbb{Q} \dot k(\nu) \ne x. $$ Then $r \le p_2 \restriction M$ and $r \le q \restriction P$. Since $r \le q \restriction P$ and $q \restriction P \in P$, it follows that $r \restriction P \le q \restriction P$ by Lemma 11.13(1). Observe that if we let $$ q' := q \uplus (\mathrm{dom}(F_r) \setminus \mathrm{dom}(F_q))$$ and $$ r' := r \uplus (\mathrm{dom}(F_q) \setminus \mathrm{dom}(F_r)), $$ then $q'$ and $r'$ satisfy exactly the same properties which we stated that $q$ and $r$ satisfy, and moreover, $\mathrm{dom}(F_{q'}) = \mathrm{dom}(F_{r'})$. Let us check this observation carefully. Since $q$ and $r$ are in $M^*$, so are $q'$ and $r'$. And by Definition 10.7, $$ \mathrm{dom}(F_{q'}) = \mathrm{dom}(F_q) \cup \mathrm{dom}(F_r) = \mathrm{dom}(F_{r'}). $$ Since $q' \le q \le p_2 \restriction M$, we have that $q' \le p_2 \restriction M$. As $q$ and $r$ are in $D(P)$, so are $q'$ and $r'$ by Lemma 11.10. And since $q' \le q$ and $r' \le r$, we have that $q' \Vdash_\mathbb{Q} \dot k(\nu) = x$ and $r' \Vdash_\mathbb{Q} \dot k(\nu) \ne x$. Finally, $r' \le r \le p_2 \restriction M$ implies that $r' \le p_2 \restriction M$. Letting $y := \mathrm{dom}(F_r) \setminus \mathrm{dom}(F_q)$, the fact that $r \le q \restriction P$ implies by Lemmas 10.8(4) and 11.17 that $$ r' \le r \le (q \restriction P) \uplus (y \cap P) = (q \uplus y) \restriction P = q' \restriction P, $$ so $r' \le q' \restriction P$. And by Lemma 11.13(1), this last inequality implies that $r' \restriction P \le q' \restriction P$. By replacing $q$ and $r$ with $q'$ and $r'$ respectively if necessary, we can assume without loss of generality that $\mathrm{dom}(F_q) = \mathrm{dom}(F_r)$. Let $x := \mathrm{dom}(F_q) \setminus \mathrm{dom}(F_{p_2})$. Define $$ p_3 := p_2 \uplus x. $$ Then by Proposition 11.18, $q$ and $r$ are below $p_3 \restriction M$. Also, $q \oplus^{M} p_3$ is a condition below $q$ and $p_3$, and $r \oplus^{M} p_3$ is a condition below $r$ and $p_3$. Also, by Lemmas 11.6 and 11.10, $p_3 \in D(M) \cap D(P)$. Since $\mathrm{dom}(F_q)$ and $\mathrm{dom}(F_r)$ are subsets of $\mathrm{dom}(F_{p_3})$, by Proposition 12.6 we have that $$ (q \oplus^{M} p_3) \restriction P = (q \restriction P) \oplus^{M \cap P} (p_3 \restriction P) $$ and $$ (r \oplus^{M} p_3) \restriction P = (r \restriction P) \oplus^{M \cap P} (p_3 \restriction P). $$ We would like to apply Lemma 12.1 to $M \cap P$, $p_3 \restriction P$, $q \restriction P$, and $r \restriction P$. Let us check that the assumptions of Lemma 12.1 hold for these objects. By Lemma 2.30, $M \cap P$ is simple. Since $p_3 \in D(M) \cap D(P)$, it follows that $p_3 \restriction P \in D(M \cap P)$ by Lemma 12.3. As $q$, $r$, and $P$ are in $M^*$, we have that $q \restriction P$ and $r \restriction P$ are in $M^* \cap P = M \cap P$. Finally, we observed above that $r \restriction P \le q \restriction P$, and $q \le p_3 \restriction M$ implies that $q \restriction P \le (p_3 \restriction P) \restriction (M \cap P)$ by Lemma 12.5. Thus, all of the assumptions of Lemma 12.1 hold. Consequently, $$ (r \restriction P) \oplus^{M \cap P} (p_3 \restriction P) \le (q \restriction P) \oplus^{M \cap P} (p_3 \restriction P). $$ Combining this with the equalities above, we have that $$ (r \oplus^{M} p_3) \restriction P \le (q \oplus^{M} p_3) \restriction P. $$ We claim that this last inequality is impossible. In fact, we will show that $(r \oplus^M p_3) \restriction P$ and $(q \oplus^M p_3) \restriction P$ are incompatible. This contradiction will complete the proof of Claim 2. We know that $r \Vdash_\mathbb{Q} \dot k(\nu) \ne x$, and therefore, since $r \oplus^{M} p_3 \le r$, we have that $r \oplus^{M} p_3 \Vdash_\mathbb{Q} \dot k(\nu) \ne x$. By Claim 1, $$ (r \oplus^{M} p_3) \restriction P \Vdash_{P \cap \mathbb{Q}} \dot s(\nu) \ne x. $$ Similarly, $q \Vdash_\mathbb{Q} \dot k(\nu) = x$, and therefore, since $q \oplus^M p_3 \le q$, we have that $q \oplus^M p_3 \Vdash_\mathbb{Q} \dot k(\nu) = x$. By Claim 1, $$ (q \oplus^{M} p_3) \restriction P \Vdash_{P \cap \mathbb{Q}} \dot s(\nu) = x. $$ Thus, indeed $(r \oplus^M p_3) \restriction P$ and $(q \oplus^M p_3) \restriction P$ are incompatible, since they force contradictory information. This completes the proof of Claim 2. \bigskip The proof that $p_2 \restriction M$ forces that $\dot k$ is in $V[\dot G_\mathbb{Q} \cap P]$ follows from Claim 2 in exactly the same way that the analogous conclusion in Theorem 9.2 followed from Claim 2 there. \end{proof} \bigskip \addcontentsline{toc}{section}{13. The consistency result} \textbf{\S 13. The consistency result} \stepcounter{section} \bigskip We now fulfill the mission of the paper and prove that it is consistent, relative to the consistency of a greatly Mahlo cardinal, that the approachability ideal $I[\omega_2]$ does not have a maximal set modulo clubs. We work in a ground model $V$ in which $\kappa$ is a greatly Mahlo cardinal, $2^\kappa = \kappa^+$, and $\Box_\kappa$ holds. The consistency of a greatly Mahlo cardinal easily implies the consistency of these assumptions. It is a standard fact that $\kappa$ being greatly Mahlo implies that there exists a sequence $\langle B_i : i < \kappa^+ \rangle$ of stationary subsets of $\kappa$ satisfying the following properties: \begin{enumerate} \item for each $i < \kappa^+$, for all $\beta \in B_i$, $\beta$ is strongly inaccessible; \item for each $i < \kappa^+$, for all $\beta \in B_{i+1}$, $B_i \cap \beta$ is stationary in $\beta$; \item for all $i < j < \kappa^+$, there is a club set $C \subseteq \kappa$ such that $B_j \cap C \subseteq B_i$; \item for each $i < \kappa^+$, $B_i \setminus B_{i+1}$ is stationary. \end{enumerate} Such a sequence is obtained by iterating the Mahlo operation $$ M(A) := \{ \alpha \in \kappa \cap \mathrm{cof}(>\! \omega) : A \cap \alpha \ \textrm{is stationary in $\alpha$} \}, $$ starting with the set of inaccessibles in $\kappa$, and taking diagonal intersections of some form at limit stages. We refer the reader to \cite[Section 4]{baumgartner} for more information about greatly Mahlo cardinals. \bigskip The results of this paper up to now were made in the context of several fixed objects, together with some assumptions about these objects. Specifically, in Section 2 we fixed $\kappa$, $\lambda$, $\Lambda$, $\mathcal A$, $\mathcal X$, and $\mathcal Y$, satisfying the properties described in Notations 2.1, 2.2, 2.3, and 2.4, and Assumptions 2.5, 2.6, 2.19, 2.22, 2.23, and 10.1. In addition, in Section 10 we fixed an ordinal $\lambda^*$ and a sequence $\langle S_i : i < \lambda^* \rangle$ satisfying Notation 10.2 and Assumption 10.3. We now specify such objects explicitly and justify the properties which we have been assuming about them. We will refer to our previous paper \cite{jk27} for some of the definitions and proofs. The greatly Mahlo cardinal $\kappa$ which we fixed at the beginning of this section is the cardinal described in Notation 2.1. The cardinal $\lambda$ described in Notation 2.1 is equal to $\kappa^+$. We refer to \cite[Notation 1.7]{jk27} for the definition of $\Lambda$. In that paper, we have that $\Lambda = C^* \cap \mathrm{cof}(>\! \omega)$, where $C^*$ is a club subset of $\kappa$. The club set $C^*$, in turn, is defined in terms of a thin stationary set $T^* \subseteq P_{\omega_1}(\kappa)$. We must justify, therefore, the existence of a thin stationary set. But $\kappa$ is strongly inaccessible, so we can let $T^*$ be equal to the entire set $P_{\omega_1}(\kappa)$. The properties of $\kappa$, $\lambda$, and $\Lambda$ described in Notations 2.1 and 2.2 and Assumption 10.1 are now immediate. We refer to \cite[Section 7]{jk27} for the definitions of $\mathcal A$, $\mathcal X$, and $\mathcal Y$. At the beginning of that section, it is assumed that $2^\kappa = \kappa^+$ and $\Box_\kappa$, which are exactly the same assumptions which we made above. Let $\mathcal A$ denote the structure which is obtained by expanding the structure on $H(\kappa^+)$ specified in \cite[Notation 7.6]{jk27} by adding the sequence $\langle B_i : i < \kappa^+ \rangle$ as a predicate. This structure has a well-ordering of $H(\kappa^+)$ as a predicate, and therefore has definable Skolem functions. It also has $\kappa$ and $\Lambda$ as constants. Thus, the description of $\mathcal A$ made in Notation 2.3 is satisfied. We define $\mathcal X$ exactly as in \cite[Notation 7.7]{jk27}. Then by definition, for all $M \in \mathcal X$, $M$ is a countable elementary substructure of $\mathcal A$. We define $\mathcal Y$ as the set of models $P$ which are in the set defined in \cite[Notation 7.8]{jk27} and also satisfy that $\mathrm{cf}(P \cap \kappa) > \omega$. Then by definition, for all $P \in \mathcal Y$, $P$ is an elementary substructure of $\mathcal A$, $|P| < \kappa$, and $P \cap \kappa \in \kappa$. Thus, the properties of $\mathcal X$ and $\mathcal Y$ described in Notation 2.4 are satisfied. The next lemma verifies Assumptions 2.5, 2.6, and 2.19. \begin{lemma} \begin{enumerate} \item If $P$ and $Q$ are in $\mathcal Y$, then $P \cap Q \in \mathcal Y$; \item if $M \in \mathcal X$ and $P \in \mathcal Y$, then $M \cap P \in \mathcal X$; \item if $M$ and $N$ are in $\mathcal X$ and $\{ M, N \}$ is adequate, then $M \cap N \in \mathcal X$; \item if $M \in \mathcal X$, $\alpha \in \Lambda \cup \{ \kappa \}$, and $Sk(\alpha) \cap \kappa = \alpha$ if $\alpha < \kappa$, then $M \cap \alpha \in Sk(\alpha)$. \end{enumerate} \end{lemma} \begin{proof} (1), (2), and (3) follow immediately from \cite[Lemma 7.16]{jk27}. (4) Note that $Sk(\kappa) \cap \kappa = \kappa$. Let $M \in \mathcal X$ and $\alpha \in \Lambda \cup \{ \kappa \}$ be as in (4). Since $\kappa$ is strongly inaccessible, for all $\beta < \alpha$, the cardinality of $P(\beta)$ is in $Sk(\alpha) \cap \kappa = \alpha$ by elementarity. So again by elementarity, $P(\beta) \subseteq Sk(\alpha)$. As $M$ is countable and $\mathrm{cf}(\alpha) > \omega$, $M \cap \alpha \in P(\beta)$ for some $\beta < \alpha$. Hence, $M \cap \alpha \in Sk(\alpha)$. \end{proof} In \cite[Definition 7.18]{jk27}, the notion of a simple model in $\mathcal X \cup \mathcal Y$ is defined. This notion is different from what we are calling simple in this paper, so let us momentarily refer to the property from \cite[Definition 7.18]{jk27} as strongly simple. By \cite[Lemma 8.2]{jk27}, a set in $\mathcal X \cup \mathcal Y$ which is strongly simple in also simple in the sense that we are using in the present paper. The next lemma verifies Assumption 2.22. \begin{lemma} There are stationarily many sets $N \in P_{\omega_1}(H(\kappa^+))$ such that $N \in \mathcal X$ and $N$ is simple. \end{lemma} \begin{proof} By \cite[Proposition 7.20]{jk27}, there are stationarily many strongly simple models in $\mathcal X$. Since strongly simple implies simple, there are stationarily many simple models in $\mathcal X$. \end{proof} The next lemma gives a sufficient criterion for a set $P$ being a simple model in $\mathcal Y$. \begin{lemma} Suppose that $P \in P_{\kappa}(H(\kappa^+))$ and $P$ satisfies: \begin{enumerate} \item $P \prec \mathcal A$; \item $P \cap \kappa \in \kappa$; \item $\mathrm{cf}(\sup(P \cap \kappa^+)) = P \cap \kappa$. \end{enumerate} Then $P \in \mathcal Y$ and $P$ is simple. \end{lemma} \begin{proof} By \cite[Lemma 7.15]{jk27}, assumptions (1), (2), and (3) imply that $P \in \mathcal Y$. By \cite[Lemma 8.3]{jk27}, assumption (3) implies that $P$ is strongly simple, and hence simple. \end{proof} The next lemma justifies Assumption 2.23. \begin{lemma} The set of $P \in P_{\kappa}(H(\kappa^+))$ such that $P \in \mathcal Y$ and $P$ is simple is stationary. \end{lemma} \begin{proof} Given a function $F : H(\kappa^+)^{<\omega} \to H(\kappa^+)$, build a membership increasing and continuous chain $\langle P_i : i < \kappa \rangle$ of elementary substructures of $\mathcal A$ which have size less than $\kappa$ and are closed under $F$. Then there is a club of $\alpha$ such that $P_\alpha \cap \kappa = \alpha$. Fix a strongly inaccessible cardinal $\alpha$ in this club. Then $P_\alpha$ is closed under $F$ and satisfies properties (1), (2), and (3) of Lemma 13.3. Hence, $P_\alpha$ is in $\mathcal Y$ and is simple. \end{proof} Let the ordinal $\lambda^*$ from Notation 10.2 be equal to $\kappa^+$. Define, for each $i < \kappa^+$, $$ S_i := (B_i \setminus B_{i+1}) \cap C^* \cap C', $$ where $C^*$ and $C'$ are club subsets of $\kappa$ such that $\Lambda = C^* \cap \mathrm{cof}(>\! \omega)$, and for all $\alpha \in C'$, $Sk(\alpha) \cap \kappa = \alpha$. By the properties described at the beginning of the section for $\langle B_i : i < \kappa^+ \rangle$, each $S_i$ is a stationary subset of $\kappa \cap \mathrm{cof}(>\!\omega)$, and for all $\alpha \in S_i$, $\alpha \in \Lambda$ and $Sk(\alpha) \cap \kappa = \alpha$. Consider $i < j < \kappa^+$. We claim that there is a club set $C_{i,j} \subseteq \kappa$, which is definable in $\mathcal A$ from $i$ and $j$, such that $S_i \cap S_j \cap C_{i,j} = \emptyset$. Since $i +1 \le j$, we know that there exists a club $C$ such that $B_j \cap C \subseteq B_{i+1}$. Let $C_{i,j}$ be the least such club in the well-ordering of $H(\kappa^+)$ which is a predicate of $\mathcal A$. Since $S_j \subseteq B_j$, $S_j \cap C_{i,j} \subseteq B_{i+1}$. But by definition, $B_{i+1}$ is disjoint from $S_i$. Thus, $S_i \cap S_j \cap C_{i,j} = \emptyset$. This completes the verification of the properties described in Notation 10.2 and Assumption 10.3. Finally, for each $i < \kappa^+$, let $\mathbb{P}_i$ denote the forcing poset defined in Definition 4.2 for adding a partial square sequence on $S_i$, and let $\mathbb{Q}$ be the product forcing defined in Definition 10.6. \bigskip This completes the choice of all of the background objects and the verification of all of the assumptions which we made about them. By Corollary 11.22, the forcing poset $\mathbb{Q}$ preserves $\omega_1$, is $\kappa$-c.c., forces that $\kappa$ is equal to $\omega_2$, and forces that for all $i < \kappa^+$, $S_i \in I[\omega_2]$. \bigskip It remains to show that $\mathbb{Q}$ forces that $I[\omega_2]$ does not have a maximal set modulo clubs. We will prove a technical lemma about names and then finish the proof of the consistency result. \begin{lemma} Suppose that $P \in \mathcal Y$ is simple and $P \prec (H(\kappa^+),\in,\mathbb{Q})$. Let $\dot a \in P$ be a nice $\mathbb{Q}$-name for a countable subset of $\kappa$. Then for any generic filter $G$ on $\mathbb{Q}$, $\dot a^G \in V[G \cap P]$. \end{lemma} Recall that by Corollary 11.20, $P \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$. Therefore, $G \cap P$ is a generic filter on $P \cap \mathbb{Q}$ and $V[G \cap P] \subseteq V[G]$. \begin{proof} Let $\alpha := P \cap \kappa$. Since $\mathbb{Q}$ is $\kappa$-c.c., by elementarity we can fix a set $b \in P$ such that $b$ is a bounded subset of $\kappa$ and $\mathbb{Q}$ forces that $\dot a \subseteq b$. Note that $b \subseteq \alpha$. Since $\dot a$ is a nice name, for each $\gamma < \kappa$ there is a unique antichain $A_\gamma$ such that $$ (p,\check \gamma) \in \dot a \iff p \in A_\gamma. $$ Moreover, as $\mathbb{Q}$ forces that $\dot a \subseteq \alpha$, $A_\gamma = \emptyset$ for all $\gamma \in \kappa \setminus \alpha$. Let $\gamma < \alpha$. Since $\dot a \in P$, by elementarity $A_\gamma$ is in $P$. As $\mathbb{Q}$ is $\kappa$-c.c., $|A_\gamma| < \kappa$. Since $P \cap \kappa \in \kappa$, we have that $A_\gamma \subseteq P \cap \mathbb{Q}$. It follows that $\dot a$ is actually a $(P \cap \mathbb{Q})$-name. Since $P \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$, $\dot a^G = \dot a^{G \cap P}$. Thus, $\dot a^G \in V[G \cap P]$. \end{proof} \begin{thm} The forcing poset $\mathbb{Q}$ forces that the approachability ideal $I[\omega_2]$ does not have a maximal set modulo clubs. \end{thm} \begin{proof} Suppose for a contradiction that there is a condition $p$ and a sequence $\vec{\dot a} = \langle \dot a_i : i < \kappa \rangle$ of $\mathbb{Q}$-names for countable subsets of $\kappa$ such that $p$ forces that $S_{\vec{\dot a}}$ is a maximal set in $I[\omega_2]$ modulo clubs. This means that $p$ forces that whenever $S \in I[\omega_2]$, then there is a club $C \subseteq \omega_2$ such that $S \cap C \subseteq S_{\vec{\dot a}}$. Without loss of generality, we may assume that each $\dot a_i$ is a nice $\mathbb{Q}$-name for a countable subset of $\kappa$. Since $\mathbb{Q}$ is $\kappa$-c.c., it follows that each name $\dot a_i$, and therefore the entire sequence of names $\vec{\dot a}$, is a member of $H(\kappa^+)$. In the ground model $V$, fix a set $X$ satisfying: \begin{enumerate} \item $X \prec \mathcal A$ and $X \prec (H(\kappa^+),\in,\mathbb{Q},\vec{\dot a})$; \item $|X| = \kappa$; \item $\tau := X \cap \kappa^+$ is an ordinal in $\kappa^+$; \item $X^{<\kappa} \subseteq X$. \end{enumerate} This is possible since $\kappa$ is strongly inaccessible. Note that by (3) and (4), $\mathrm{cf}(\tau) = \kappa$. Fix a membership increasing and continuous sequence $\langle P_i : i < \kappa \rangle$ of sets of size less than $\kappa$, whose union is equal to $X$, such that each $P_i$ is an elementary substructure of $\mathcal A$ and $(H(\kappa^+),\in,\mathbb{Q},\vec{\dot a})$. This is possible since $X^{< \kappa} \subseteq X$ by (4). Note that by elementarity, $\sup(P_i \cap \kappa^+) \in P_{i+1}$ for all $i < \kappa^+$. Using the properties of the sequence $\langle B_i : i < \kappa^+ \rangle$ described at the beginning of this section, we can fix, for each $i < \tau$, a club set $C_i \subseteq \kappa$ such that $$ B_\tau \cap C_i \subseteq B_i. $$ Define a function $F : \tau \times \kappa \to \kappa$ by letting $$ F(i,\gamma) := \min(C_i \setminus \gamma), $$ for all $(i,\gamma) \in \tau \times \kappa$. Note that $F \subseteq X$. Also, fix a club $C^\tau \subseteq \kappa$ such that $B_{\tau+1} \cap C^\tau \subseteq B_\tau$. Fix a club $D \subseteq \kappa$ such that for all $\beta \in D$, $P_\beta \cap \kappa = \beta$, $\beta \in C^\tau$, and $P_\beta$ is closed under the function $F$. \bigskip \noindent \emph{Claim 1:} Suppose that $\beta \in D$ and $\beta$ is strongly inaccessible. Then $P_\beta \in \mathcal Y$, $P_\beta$ is simple, $P_\beta \prec (H(\kappa^+),\in,\mathbb{Q})$, $\beta \in C^\tau$, and for all $i \in P_\beta \cap \kappa^+$, $\beta \in C_i$. \bigskip The fact that $P_\beta$ is closed under $F$ easily implies that for all $i \in P_\beta \cap \kappa^+ = P_\beta \cap \tau$, $P_\beta \cap \kappa = \beta$ is a limit point of $C_i$, and therefore is in $C_i$. And $\beta \in C^\tau$ by the definition of $D$. The set $P_\beta$ is an elementary substructure of $\mathcal A$ and $(H(\kappa^+),\in,\mathbb{Q})$ by the choice of the sequence $\langle P_i : i < \kappa \rangle$. Since $P_\beta$ is the union of the sequence $\langle P_i : i < \beta \rangle$, and $\sup(P_i \cap \kappa^+) \in P_{i+1}$ for all $i < \beta$, it follows that $$ \mathrm{cf}(\sup(P_\beta \cap \kappa^+)) = \mathrm{cf}(\beta) = \beta = P_\beta \cap \kappa. $$ By Lemma 13.3, it follows that $P_\beta \in \mathcal Y$ and $P_\beta$ is simple. This completes the proof of Claim 1. \bigskip Let $G$ be a generic filter on $\mathbb{Q}$ with $p \in G$. For each $i < \kappa$, let $a_i := \dot a_i^G$, and let $\vec a := \langle a_i : i < \kappa \rangle$. Then by assumption, in $V[G]$ the set $S_{\vec a}$ is maximal in $I[\omega_2]$ modulo clubs. In $V[G]$ the set $S_{\tau+1}$ is in $I[\omega_2]$. Since $S_{\vec a}$ is maximal modulo clubs, fix a club $C \subseteq \kappa$ in $V[G]$ such that $$ S_{\tau+1} \cap C \subseteq S_{\vec a}. $$ As the set $S_{\tau+1}$ is stationary in $V[G]$, we can fix $$ \beta \in S_{\tau+1} \cap \lim(D) \cap C. $$ Then $\beta \in S_{\tau+1} \cap C \subseteq S_{\vec a}$. Let $P := P_\beta$. Note that since $\beta \in S_{\tau+1}$, $\beta$ is strongly inaccessible in $V$. Also, $\beta \in D$. So by Claim 1, $P \in \mathcal Y$, $P$ is simple, $P \prec (H(\kappa^+),\in,\mathbb{Q})$, $\beta \in C^\tau$, and for all $i \in P \cap \kappa^+$, $\beta \in C_i$. Since $\beta \in C^\tau$, $S_{\tau+1} \subseteq B_{\tau+1}$, and $B_{\tau+1} \cap C^\tau \subseteq B_\tau$, it follows that $\beta \in B_\tau$. Also, for all $i \in P \cap \kappa^+$, $\beta \in B_\tau \cap C_i \subseteq B_i$, so $\beta \in B_i$. Therefore, for all $i \in P \cap \kappa^+$, since $i+1 \in P \cap \kappa^+$ by elementarity, $\beta \in B_{i+1}$. But $S_i \subseteq B_{i} \setminus B_{i+1}$. So for all $i \in P \cap \kappa^+$, $P \cap \kappa = \beta \notin S_i$. By Corollary 11.20 and Theorem 12.7, it follows that $P \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$ and $P \cap \mathbb{Q}$ forces that $\mathbb{Q} / \dot G_{P \cap \mathbb{Q}}$ has the $\omega_1$-approximation property. \bigskip \noindent \emph{Claim 2:} The forcing poset $P \cap \mathbb{Q}$ is $\beta$-c.c. \bigskip Let $A$ be an antichain of $P \cap \mathbb{Q}$, and we will prove that $|A| < \beta$. Without loss of generality, assume that $A$ is maximal. Since $\beta$ is strongly inaccessible and is a limit point of $D$, we have that $D \cap \beta$ is a club subset of $\beta$. As $P$ is the union of the $\subseteq$-increasing and continuous sequence $\langle P_i : i < \beta \rangle$, there is a club $E \subseteq \beta$ such that for all $\gamma \in E$: \begin{enumerate} \item $\gamma \in D \cap \beta$; \item $P_\gamma \prec (P,\in,P \cap \mathbb{Q},A)$. \end{enumerate} Now $\beta \in B_{\tau+1}$, which implies that $B_\tau \cap \beta$ is stationary in $\beta$. Since $E$ is a club subset of $\beta$, we can fix $\gamma \in E \cap B_\tau$. Then in particular, $\gamma \in D$ and $\gamma$ is strongly inaccessible. By Claim 1, it follows that $P_\gamma \in \mathcal Y$, $P_\gamma$ is simple, and $P_\gamma \prec (H(\kappa^+),\in,\mathbb{Q})$. Therefore, by Corollary 11.20, $P_\gamma \cap \mathbb{Q}$ is a regular suborder of $\mathbb{Q}$. Since $P_\gamma \cap \mathbb{Q} \subseteq P \cap \mathbb{Q}$, it follows from Lemma 1.1 that $P_\gamma \cap \mathbb{Q}$ is a regular suborder of $P \cap \mathbb{Q}$. We claim that $P_\gamma \cap A$ is a maximal antichain of $P_\gamma \cap \mathbb{Q}$. It is obviously an antichain. Let $v \in P_\gamma \cap \mathbb{Q}$. Then since $A$ is a maximal antichain of $P \cap \mathbb{Q}$, there is $s \in A$ such that $s$ is compatible in $P \cap \mathbb{Q}$ with $v$. But $P_\gamma$ is an elementary substructure of $(P,\in,P \cap \mathbb{Q},A)$. So by elementarity, there is $s \in A \cap P_\gamma$ which is compatible in $P \cap \mathbb{Q}$ with $v$. Again by elementarity, $s$ and $v$ are compatible in $P_\gamma \cap \mathbb{Q}$. This completes the proof that $P_\gamma \cap A$ is a maximal antichain of $P_\gamma \cap \mathbb{Q}$. As $P_\gamma \cap A$ is a maximal antichain of $P_\gamma \cap \mathbb{Q}$ and $P_\gamma \cap \mathbb{Q}$ is a regular suborder of $P \cap \mathbb{Q}$, it follows that $P_\gamma \cap A$ is predense in $P \cap \mathbb{Q}$. But $A$ is a maximal antichain of $P \cap \mathbb{Q}$; therefore, it must be the case that $A = P_\gamma \cap A$. So $A \subseteq P_{\gamma}$. But as $P_\gamma \in P$, we have that $|P_\gamma| \in P \cap \kappa = \beta$ by elementarity. So $|A| < \beta$. This completes the proof of Claim 2. \bigskip Recall that $\beta \in S_{\vec a}$ in $V[G]$. Therefore, in $V[G]$ there is a set $c$ which is cofinal in $\beta$ with order type $\omega_1$, and for all $\xi < \beta$, $c \cap \xi \in \{ a_i : i < \beta \}$. Now for all $i < \beta$, the name $\dot a_i$ is in $P$ by elementarity. By Lemma 13.5, it follows that $\dot a_i^G = a_i$ is in $V[G \cap P]$. So $c$ is a cofinal subset of $\beta$ with order type $\omega_1$, and every proper initial segment of $c$ is in $V[G \cap P]$. It easily follows that whenever $x$ is a countable set in $V[G \cap P]$, then $c \cap x \in V[G \cap P]$. Since $\mathbb{Q} / (G \cap P)$ has the $\omega_1$-approximation property in $V[G \cap P]$, we have that $c \in V[G \cap P]$. As $c$ has order type $\omega_1$ and $\beta$ is a strongly inaccessible cardinal in $V$, it follows that $\beta$ is no longer regular in $V[G \cap P]$. But by Claim 2, $P \cap \mathbb{Q}$ is $\beta$-c.c., and so $P \cap \mathbb{Q}$ preserves the regularity of $\beta$. Since $V[G \cap P]$ is a generic extension of $V$ by the forcing poset $P \cap \mathbb{Q}$, we have a contradiction. \end{proof} \bibliographystyle{plain}
6e964ae3be26a56c939b5985e33af46f1e30cd66
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The concept of synchronistion is based on the adjustment of rhythms of oscillating systems due to their interaction \cite{pikovsky01}. Synchronisation phenomenon was recognised by Huygens in the 17th century, time when he performed experiments to understand this phenomenon \cite{bennett02}. To date, several kinds of synchronisation among coupled systems were reported, such as complete \cite{li16}, phase \cite{pereira07,batista10}, lag \cite{huang14}, and collective almost synchronisation \cite{baptista12}. Neuronal synchronous rhythms have been observed in a wide range of researches about cognitive functions \cite{wang10,hutcheon00}. Electroencephalography and magnetoencephalography studies have been suggested that neuronal synchronization in the gamma frequency plays a functional role for memories in humans \cite{axmacher06,fell11}. Steinmetz et al. \cite{steinmetz00} investigated the synchronous behaviour of pairs of neurons in the secondary somatosensory cortex of monkey. They found that attention modulates oscillatory neuronal synchronisation in the somatosensory cortex. Moreover, in the literature it has been proposed that there is a relationship between conscious perception and synchronisation of neuronal activity \cite{hipp11}. We study spiking and bursting synchronisation betwe\-en neuron in a neuronal network model. A spike refers to the action potential generated by a neuron that rapidly rises and falls \cite{lange08}, while bursting refers to a sequence of spikes that are followed by a quiescent time \cite{wu12}. It was demonstrated that spiking synchronisation is relevant to olfactory bulb \cite{davison01} and is involved in motor cortical functions \cite{riehle97}. The characteristics and mechanisms of bursting synchronisation were studied in cultured cortical neurons by means of planar electrode array \cite{maeda95}. Jefferys $\&$ Haas discovered synchronised bursting of CA1 hippocampal pyramidal cells \cite{jefferys82}. There is a wide range of mathematical models used to describe neuronal activity, such as the cellular automaton \cite{viana14}, the Rulkov map \cite{rulkov01}, and differential equations \cite{hodgkin52,hindmarsh84}. One of the simplest mathematical models and that is widely used to depict neuronal behaviour is the integrate-and-fire \cite{lapicque07}, which is governed by a linear differential equation. A more realistic version of it is the adaptive exponential integrate-and-fire (aEIF) model which we consider in this work as the local neuronal activity of neurons in the network. The aEIF is a two-dimensional integrate-and-fire model introduced by Brette $\&$ Gerstner \cite{brette05}. This model has an exponential spike mechanism with an adaptation current. Touboul $\&$ Brette \cite{touboul08} studied the bifurcation diagram of the aEIF. They showed the existence of the Andronov-Hopf bifurcation and saddle-node bifurcations. The aEIF model can generate multiple firing patterns depending on the parameter and which fit experimental data from cortical neurons under current stimulation \cite{naud08}. In this work, we focus on the synchronisation phenomenon in a randomly connected network. This kind of network, also called Erd\"os-R\'enyi network \cite{erdos59}, has nodes where each pair is connected according to a probability. The random neuronal network was utilised to study oscillations in cortico-thalamic circuits \cite{gelenbe98} and dynamics of network with synaptic depression \cite{senn96}. We built a random neuronal network with unidirectional connections that represent chemical synapses. We show that there are clearly separated ranges of parameters that lead to spiking or bursting synchronisation. In addition, we analyse the robustness to external perturbation of the synchronisation. We verify that bursting synchronisation is more robustness than spiking synchronisation. However, bursting synchronisation requires larger chemical synaptic strengths, and larger voltage potential relaxation reset to appear than those required for spiking synchronisation. This paper is organised as follows: in Section II we present the adaptive exponential integrate-and-fire model. In Section III, we introduce the neuronal network with random features. In Section IV, we analyse the behaviour of spiking and bursting synchronisation. In the last Section, we draw our conclusions. \section{Adaptive exponential integrate-and-fire} As a local dynamics of the neuronal network, we consider the adaptive exponential integrate-and-fire (aEIF) model that consists of a system of two differential equations \cite{brette05} given by \begin{eqnarray}\label{eqIF} C \frac{d V}{d t} & = & - g_L (V - E_L) + {\Delta}_T \exp \left(\frac{V - V_T}{{\Delta}_T} \right) \nonumber \\ & & +I-w , \nonumber \\ \tau_w \frac{d w}{d t} & = & a (V - E_L) - w, \end{eqnarray} where $V(t)$ is the membrane potential when a current $I(t)$ is injected, $C$ is the membrane capacitance, $g_L$ is the leak conductance, $E_L$ is the resting potential, $\Delta_T$ is the slope factor, $V_T$ is the threshold potential, $w$ is an adaptation variable, $\tau_w$ is the time constant, and $a$ is the level of subthreshold adaptation. If $V(t)$ reaches the threshold $V_{\rm{peak}}$, a reset condition is applied: $V\rightarrow V_r$ and $w\rightarrow w_r=w+b$. In our simulations, we consider $C=200.0$pF, $g_L=12.0$nS, $E_L=-70.0$mV, ${\Delta}_T=2.0$mV, $V_T=-50.0$mV, $I=509.7$pA, $\tau_w=300.0$ms, $a=2.0$nS, and $V_{\rm{peak}}=20.0$mV \cite{naud08}. The firing pattern depends on the reset parameters $V_r$ and $b$. Table \ref{table1} exhibits some values that generate five different firing patterns (Fig. \ref{fig1}). In Fig. \ref{fig1} we represent each firing pattern with a different colour in the parameter space $b\times V_r$: adaptation in red, tonic spiking in blue, initial bursting in green, regular bursting in yellow, and irregular in black. In Figs. \ref{fig1}a, \ref{fig1}b, and \ref{fig1}c we observe adaptation, tonic spiking, and initial burst pattern, respectively, due to a step current stimulation. Adaptation pattern has increasing inter-spike interval during a sustained stimulus, tonic spiking pattern is the simplest regular discharge of the action potential, and the initial bursting pattern starts with a group of spikes presenting a frequency larger than the steady state frequency. The membrane potential evolution with regular bursting is showed in Fig. \ref{fig1}d, while Fig. \ref{fig1}e displays irregular pattern. \begin{table}[htbp] \caption{Reset parameters.} \centering \begin{tabular}{c c c c c} \hline Firing patterns & Fig. & b (pA) & $V_r$ (mV) & Layout \\ \hline adaptation &\ref{fig1}(a) & 60.0 & -68.0 & red \\ tonic spiking & \ref{fig1}(b) & 5.0 & -65.0 & blue\\ initial burst & \ref{fig1}(c) & 35.0 & -48.8 & green \\ regular bursting & \ref{fig1}(d) & 40.0 & -45.0 & yellow\\ irregular & \ref{fig1}(e) & 41.2 & -47.4 & black \\ \hline \end{tabular} \label{table1} \end{table} \begin{figure}[hbt] \centering \includegraphics[height=7cm,width=10cm]{fig1.eps} \caption{(Colour online) Parameter space for the firing patterns as a function of the reset parameters $V_r$ and $b$. (a) Adaptation in red, (b) tonic spiking in blue, (c) initial bursting in green, (d) regular bursting in yellow, and (e) irregular in black.} \label{fig1} \end{figure} As we have interest in spiking and bursting synchronisation, we separate the parameter space into a region with spike and another with bursting patterns (Fig. \ref{fig2}). To identify these two regions of interest, we use the coefficient of variation (CV) of the neuronal inter-spike interval (ISI), that is given by \begin{eqnarray}\label{CV} {\rm CV}=\frac{{\sigma}_{\rm{ISI}}}{\rm{\overline{ISI}}}, \end{eqnarray} where ${\sigma}_{\rm{ISI}}$ is the standard deviation of the ISI normalised by the mean $\bar{\rm ISI}$ \cite{gabbiani98}. Spiking patterns produce $\rm{CV}<0.5$. Parameter regions that represent the neurons firing with spiking pattern are denoted by gray colour in Fig. \ref{fig2}. Whereas, the black region represents the bursting patterns, which results in $\rm{CV} \geq 0.5$. \begin{figure}[hbt] \centering \includegraphics[height=7cm,width=9cm]{fig2.eps} \caption{Parameter space for the firing patterns as a function of the reset parameters $V_r$ and $b$. Spike pattern in region I ($\rm{CV}<0.5$) and bursting pattern in region II ($\rm{CV}\geq 0.5$) are separated by white circles.} \label{fig2} \end{figure} \section{Spiking or bursting synchronisation} In this work, we constructed a network where the neurons are randomly connected \cite{erdos59}. Our network is given by \begin{eqnarray}\label{eqIFrede} C \frac{d V_i}{d t} & = & - g_L (V_i - E_L) + {\Delta}_T \; \rm{exp} \left(\frac{V_i - V_T}{{\Delta}_T} \right) \nonumber \\ & + & I_i - w_i + g_{\rm{ex}} (V_{\rm{ex}} - V_i) \sum_{j=1}^N A_{ij} s_j + \Gamma_i, \nonumber \\ \tau_w \frac{d w_i}{d t} & = & a_i (V_i - E_L) - w_i, \nonumber \\ \tau_{\rm{ex}} \frac{d s_i}{d t} & = & - s_i. \end{eqnarray} where $V_i$ is the membrane potential of the neuron $i$, $g_{\rm{ex}}$ is the synaptic conductance, $V_{\rm{ex}}$ is the synaptic reversal potential, $\tau_{\rm{ex}}$ is the synaptic time constant, $s_i$ is the synaptic weight, $A_{ij}$ is the adjacency matrix, $\Gamma_i$ is the external perturbation, and $a_i$ is randomly distributed in the interval $[1.9,2.1]$. The schematic representation of the neuronal network that we have considered is illustrated in Fig \ref{fig3}. Each neuron is randomly linked to other neurons with a probability $p$ by means of directed connections. When $p$ is equal to 1, the neuronal network becames an all-to-all network. A network with this topology was used by Borges et al. \cite{borges16} to study the effects of the spike timing-dependent plasticity on the synchronisation in a Hodgkin-Huxley neuronal network. \begin{figure}[hbt] \centering \includegraphics[height=6cm,width=9cm]{fig3.eps} \caption{Schematic representation of the neuronal network where the neurons are connected according to a probability $p$.} \label{fig3} \end{figure} A useful diagnostic tool to determine synchronous behaviour is the complex phase order parameter defined as \cite{kuramoto03} \begin{equation} z(t)=R(t)\exp({\rm i}\Phi(t))\equiv\frac{1}{N}\sum_{j=1}^{N}\exp({\rm i}\psi_{j}), \end{equation} where $R$ and $\Phi$ are the amplitude and angle of a centroid phase vector, respectively, and the phase is given by \begin{equation} \psi_{j}(t)=2\pi m+2\pi\frac{t-t_{j,m}}{t_{j,m+1}-t_{j,m}}, \end{equation} where $t_{j,m}$ corresponds to the time when a spike $m$ ($m=0,1,2,\dots$) of a neuron $j$ happens ($t_{j,m}< t < t_{j,m+1}$). We have considered the beginning of the spike when $V_j>-20$mV. The value of the order parameter magnitude goes to 1 in a totally synchronised state. To study the neuronal synchronisation of the network, we have calculated the time-average order-parameter, that is given by \begin{equation} \overline{R}=\frac{1}{t_{\rm fin}-{t_{\rm ini}}}\sum_{t_{\rm ini}}^{t_{\rm fin}}R(t), \end{equation} where $t_{\rm fin}-t_{\rm ini}$ is the time window for calculating $\bar{R}$. Figs. \ref{fig4}a, \ref{fig4}b, and \ref{fig4}c show the raster plots for $g_{\rm ex}=0.02$nS, $g_{\rm ex}=0.19$nS, and $g_{\rm ex}=0.45$nS, respectively, considering $V_r=-58$mV, $p=0.5$, and $b=70$pA, where the dots correspond to the spiking activities generated by neurons. For $g_{\rm ex}=0.02$nS (Fig. \ref{fig4}a) the network displays a desynchonised state, and as a result, the order parameter values are very small (black line in Fig. \ref{fig4}d). Increasing the synaptic conductance for $g_{\rm ex}=0.19$nS, the neuronal network exhibits spike synchronisation (Fig. \ref{fig4}b) and the order parameter values are near unity (red line in Fig. \ref{fig4}d). When the network presents bursting synchronisation (Fig. \ref{fig4}c), the order parameter values vary between $R\approx 1$ and $R\ll 1$ (blue line in Fig. \ref{fig4}d). $R\ll 1$ to the time when the neuron are firing. \begin{figure}[hbt] \centering \includegraphics[height=11cm,width=10cm]{fig4.eps} \caption{(Colour online) Raster plot for (a) $g_{\rm ex}=0.02$nS, (b) $g_{\rm ex}=0.19$nS, and (c) $g_{\rm ex}=0.45$nS, considering $V_r = -58$mV, $p=0.5$, and $b=70$pA. In (d) the order parameter is computed for $g_{\rm ex}=0.02$nS (black line), $g_{\rm ex}=0.19$nS (red line), and $g_{\rm ex}=0.19$nS (blue line).} \label{fig4} \end{figure} In Fig. \ref{fig5}a we show ${\bar R}$ as a function of $g_{\rm ex}$ for $p=0.5$, $b=50$pA (black line), $b=60$pA (red line), and $b=70$pA (blue line). The three results exhibit strong synchronous behaviour (${\bar R}>0.9$) for many values of $g_{\rm ex}$ when $g_{\rm ex}\gtrsim 0.4$nS . However, for $g_{\rm ex}\lesssim 0.4$nS, it is possible to see synchronous behaviour only for $b=70$pA in the range $0.15{\rm nS}<g_{\rm ex}<0.25{\rm nS}$. In addition, we calculate the coefficient of variation (CV) to determine the range in $g_{\rm ex}$ where the neurons of the network have spiking or bursting behaviour (Fig. \ref{fig5}b). We consider that for CV$<0.5$ (black dashed line) the neurons exhibit spiking behaviour, while for CV$\geq 0.5$ the neurons present bursting behaviour. We observe that in the range $0.15{\rm nS}<g_{\rm ex}<0.25{\rm nS}$ for $b=70$pA there is spiking sychronisation, and bursting synchronisation for $g_{\rm ex}\gtrsim 0.4$nS. \begin{figure}[hbt] \centering \includegraphics[height=7cm,width=9cm]{fig5.eps} \caption{(Colour online) (a) Time-average order parameter and (b) CV for $V_r=-58$mV, $p=0.5$, $b=50$pA (black line), $b=60$pA (red line), and $b=70$pA (blue line).} \label{fig5} \end{figure} \section{Parameter space of synchronisation} The synchronous behaviour depends on the synaptic conductance and the probability of connections. Fig. \ref{fig6} exhibits the time-averaged order parameter in colour scale as a function of $g_{\rm ex}$ and $p$. We verify a large parameter region where spiking and bursting synchronisation is strong, characterised by ${\bar R}>0.9$. The regions I and II correspond to spiking and bursting patterns, respectively, and these regions are separated by a white line with circles. We obtain the regions by means of the coefficient of variation (CV). There is a transition between region I and region II, where neurons initially synchronous in the spike, loose spiking synchronicity to give place to a neuronal network with a regime of bursting synchronisation. \begin{figure}[hbt] \centering \includegraphics[height=6cm,width=9cm]{fig6.eps} \caption{(Colour online) $g_{\rm ex} \times p$ for $V_r=-58$mV and $b=70$pA, where the colour bar represents the time-average order parameter. The regions I (spike patterns) and II (bursting patterns) are separated by the white line with circles.} \label{fig6} \end{figure} We investigate the dependence of spiking and bursting synchronisation on the control parameters $b$ and $V_r$. To do that, we use the time average order parameter and the coefficient of variation. Figure \ref{fig7} shows that the spike patterns region (region I) decreases when $g_{\rm ex}$ increases. This way, the region I for $b<100$pA and $V_r=-49$mV of parameters leading to no synchronous behaviour (Fig. \ref{fig7}a), becomes a region of parameters that promote synchronised bursting (Fig. \ref{fig7}b and \ref{fig7}c). However, a large region of desynchronised bursting appears for $g_{\rm ex}=0.25$nS about $V_r=-45$mV and $b>100$pA in the region II (Fig. \ref{fig7}b). For $g_{\rm ex}=0.5$nS, we see, in Fig. \ref{fig7}c, three regions of desynchronous behaviour, one in the region I for $b<100$pA, other in region II for $b<200$pA, and another one is located around the border (white line with circles) between regions I and II for $b>200$pA. \begin{figure}[hbt] \centering \includegraphics[height=12cm,width=7cm]{fig7.eps} \caption{(Colour online) Parameter space $b \times V_r$ for $p=0.5$, $\gamma=0$ (a) $g_{\rm ex}=0.05$nS, (b) $g_{\rm ex}=0.25$nS, and (c) $g_{\rm ex}=0.5$nS, where the colour bar represents the time-average order parameter. The regions I (spike patterns) and II (bursting patterns) are separated by white circles.} \label{fig7} \end{figure} It has been found that external perturbations on neuronal networks not only can induce synchronous behaviour \cite{baptista06,zhang15}, but also can suppress synchronisation \cite{lameu16}. Aiming to study the robustness to perturbations of the synchronous behaviour, we consider an external perturbation $\Gamma_i$ (\ref{eqIFrede}). It is applied on each neuron $i$ with an average time interval of about $10$ms and with a constant intensity $\gamma$ during $1$ms. Figure \ref{fig8} shows the plots $g_{\rm ex} \times p$ for $\gamma>0$, where the regions I and II correspond to spiking and bursting patterns, respectively, separated by white line with circles, and the colour bar indicates the time-average order parameter values. In this Figure, we consider $V_r=-58$mV, $b=70$pA, (a) $\gamma=250$pA, (b) $\gamma=500$pA, and (c) $\gamma=1000$pA. For $\gamma=250$pA (Fig. \ref{fig8}a) the perturbation does not suppress spike synchronisation, whereas for $\gamma=500$pA the synchronisation is completely suppressed in region I (Fig. \ref{fig8}b). In Fig. \ref{fig8}c, we see that increasing further the constant intensity for $\gamma=1000$pA, the external perturbation suppresses also bursting synchronisation in region II. Therefore,the synchronous behavior in region II is more robustness to perturbations than in the region I, due to the fact that the region II is in a range with high $g_{\rm ex}$ and $p$ values, namely strong coupling and high connectivity. \begin{figure}[hbt] \centering \includegraphics[height=12cm,width=7cm]{fig8.eps} \caption{(Colour online) $g_{\rm ex} \times p$ for $V_r=-58$mV, $b=70$pA, (a) $\gamma=250$pA, (b) $\gamma=500$pA, and (c) $\gamma=1000$pA.} \label{fig8} \end{figure} In order to understand the perturbation effect on the spike and bursting patterns, we consider the same values of $g_{\rm ex}$ and $p$ as Fig. \ref{fig7}a. Figure \ref{fig9} exhibits the space parameter $b\times V_r$, where $\gamma$ is equal to $500$pA. The external perturbation suppresses synchronisation in the region I, whereas we observe synchronisation in region II. The synchronous behaviour in region II can be suppressed if the constant intensity $\gamma$ is increased. Therefore, bursting synchronisation is more robustness to perturbations than spike synchronisation. \begin{figure}[hbt] \centering \includegraphics[height=5cm,width=7cm]{fig9.eps} \caption{(Colour online) $b \times V_r$ for $g_{\rm ex}=0.05$nS, $p=0.5$, and $\gamma=500$pA, where the colour bar represents the time-average order parameter. The regions I (spike patterns) and II (bursting patterns) are separated by white line with circles.} \label{fig9} \end{figure} \section{Conclusion} In this paper, we studied the spiking and bursting synchronous behaviour in a random neuronal network where the local dynamics of the neurons is given by the adaptive exponential integrate-and-fire (aEIF) model. The aEIF model can exhibit different firing patterns, such as adaptation, tonic spiking, initial burst, regular bursting, and irregular bursting. In our network, the neurons are randomly connected according to a probability. The larger the probability of connection, and the strength of the synaptic connection, the more likely is to find bursting synchronisation. It is possible to suppress synchronous behaviour by means of an external perturbation. However, synchronous behaviour with higher values of $g_{\rm ex}$ and $p$, which typically promotes bursting synchronisation, are more robust to perturbations, then spike synchronous behaviour appearing for smaller values of these parameters. We concluded that bursting synchronisation provides a good environment to transmit information when neurons are stron\-gly perturbed (large $\Gamma$). \section*{Acknowledgements} This study was possible by partial financial support from the following Brazilian government agencies: CNPq, CAPES, and FAPESP (2011/19296-1 and 2015/07311-7). We also wish thank Newton Fund and COFAP.
1ba0a8c597a5b6ee0830ca23c5b9ab8ce3b62efd
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\subsection{Power} \noindent PMTs that meet the required specifications in terms of pulse rise time, dark current and counting rates, and quantum efficiency require applied high voltages (HV) between $1-2.5$~kV and have maximum current ratings of $0.2-0.5$~mA. For the detector design using 12 read-out channels per module, the HV power supply (HVPS) must provide approximately 10~mA per module. In order to minimize costs, we aim to use one HV power supply to power 10 modules (120 channels), and thus we require a HVPS rated to approximately 100~mA and 500~W. For a 100 module detector, 10 HVPS are required and the total power requirement would thus be approximately 5~kW. Several commercial HVPS systems exist that meet these requirements. For example, the \href{http://theelectrostore.com/shopsite_sc/store/html/PsslashEK03R200-GK6-Glassman-New-refurb.html}{Glassman model number PS/EK03R200-GK6} provides an output of $\pm3$~kV with a maximum of 200 mA, and features controllable constant current / constant voltage operation. Regulation and monitoring of the power supplied to the detector will be required on both the module distribution boards and the front-end distribution boards. In both cases, over-current and over-voltage protection will be necessary both for safety and in order to protect the front-end electronics from damage. The monitoring may be accomplished by a measurement circuit that digitizes and transmits the measured voltages and currents over a serial bus to the slow control system for the detector by a generic, CERN built data acquisition board called an Embedded Local Monitoring Board (ELMB)~\cite{ELMB}. Energy calibration will be done in situ using an $^{241}$Am source, which yields a 60~keV $X$-ray. Calibration runs performed at specified intervals will track the PMT+scintillator response as a function of time. In addition to energy calibration, an LED pulser that can deliver a stable light pulse into each scintillator will also be deployed. The LED system will be used to monitor drift in response of the PMT+scintillator as a function of time in between $^{241}$Am source calibrations as well as detect any inefficient or non-functional readout channels. \subsection{Power} \noindent PMTs that meet the required specifications in terms of pulse rise time, dark current and counting rates, and quantum efficiency require applied high voltages (HV) between $1-2.5$~kV and have maximum current ratings of $0.2-0.5$~mA. For the detector design using 12 read-out channels per module, the HV power supply (HVPS) must provide approximately 10~mA per module. In order to minimize costs, we aim to use one HV power supply to power 10 modules (120 channels), and thus we require a HVPS rated to approximately 100~mA and 500~W. For a 100 module detector, 10 HVPS are required and the total power requirement would thus be approximately 5~kW. Several commercial HVPS systems exist that meet these requirements. For example, the \href{http://theelectrostore.com/shopsite_sc/store/html/PsslashEK03R200-GK6-Glassman-New-refurb.html}{Glassman model number PS/EK03R200-GK6} provides an output of $\pm3$~kV with a maximum of 200 mA, and features controllable constant current / constant voltage operation. Regulation and monitoring of the power supplied to the detector will be required on both the module distribution boards and the front-end distribution boards. In both cases, over-current and over-voltage protection will be necessary both for safety and in order to protect the front-end electronics from damage. The monitoring may be accomplished by a measurement circuit that digitizes and transmits the measured voltages and currents over a serial bus to the slow control system for the detector by a generic, CERN built data acquisition board called an Embedded Local Monitoring Board (ELMB)~\cite{ELMB}. Energy calibration will be done in situ using an $^{241}$Am source, which yields a 60~keV $X$-ray. Calibration runs performed at specified intervals will track the PMT+scintillator response as a function of time. In addition to energy calibration, an LED pulser that can deliver a stable light pulse into each scintillator will also be deployed. The LED system will be used to monitor drift in response of the PMT+scintillator as a function of time in between $^{241}$Am source calibrations as well as detect any inefficient or non-functional readout channels. \section{Introduction \label{sec:Intro}} \input{intro} \section{Site Selection \label{sec:site}} \input{site.tex} \section{Relationship with CMS \label{sec:cms}} \input{cms.tex} \section{Detector Concept \label{sec:det}} \input{det.tex} \section{Mechanics, Cooling, and Magnetic Shielding \label{sec:infra}} \input{infra.tex} \section{Power and Calibrations \label{sec:pow}} \input{calib.tex} \section{Trigger and Readout \label{sec:daq}} \input{readout.tex} \section{Backgrounds \label{sec:bkg}} \input{bkg.tex} \section{Simulations and Sensitivity \label{sec:sens}} \input{sim.tex} \section{Timeline and Next Steps \label{sec:timeline}} \noindent We aim to have the experiment ready for physics during Run 3. To that end, we envisage the following timeline: \begin{itemize} \item Construct small fraction of detector ($\sim10\%$) in next 2 yrs \item Install partial detector in PX56 by end of Run 2 (YETS 2017 + TS in 2018) \item Commission and take data in order to evaluate beam-on backgrounds {\it in situ} \item Construction + Installation of remainder of detector during LS2 (2019--2020) \item Final commissioning by spring 2021 \item Operate detector for physics for duration of Run 3 and HL-LHC (mid 2021--) \end{itemize} \noindent The next step in the milliQan project is to seek external funding to enable at least the 10\% construction. No such funding has yet been secured for this project, but one or more proposals to one or more funding agencies are being prepared for the near future. \section{Summary \label{sec:end}} \noindent In this LOI we have proposed a dedicated experiment that would detect ``milli-charged" particles produced by pp collisions at LHC Point 5. The experiment would be installed during LS2 in the vestigial drainage gallery above UXC and would not interfere with CMS operations. Our calculations and simulations indicate that with 300~fb$^{-1}$ of integrated luminosity, sensitivity to a particle with charge $\mathcal{O}(10^{-3})~e$ can be achieved for masses of $\mathcal{O}(1)$~GeV, and charge $\mathcal{O}(10^{-2})~e$ for masses of $\mathcal{O}(10)$~GeV. This would greatly extend the parameter space explored for particles with small charge and masses above 100 MeV. We have performed sufficient R\&D to encourage us to proceed with securing funding for the project, and with this letter of intent we express the intention to do so. \begin{acknowledgments} \noindent We wish to thank Tiziano Camporesi, Joel Butler, and the CMS collaboration for their encouragement. We would also like to thank Vladimir Ivanchenko, Andrea Dotti and Mihaly Novak for useful discussions regarding {\sc Geant4}. \end{acknowledgments} \section{Introduction \label{sec:Intro}} \input{intro} \section{Site Selection \label{sec:site}} \input{site.tex} \section{Relationship with CMS \label{sec:cms}} \input{cms.tex} \section{Detector Concept \label{sec:det}} \input{det.tex} \section{Mechanics, Cooling, and Magnetic Shielding \label{sec:infra}} \input{infra.tex} \section{Power and Calibrations \label{sec:pow}} \input{calib.tex} \section{Trigger and Readout \label{sec:daq}} \input{readout.tex} \section{Backgrounds \label{sec:bkg}} \input{bkg.tex} \section{Simulations and Sensitivity \label{sec:sens}} \input{sim.tex} \section{Timeline and Next Steps \label{sec:timeline}} \noindent We aim to have the experiment ready for physics during Run 3. To that end, we envisage the following timeline: \begin{itemize} \item Construct small fraction of detector ($\sim10\%$) in next 2 yrs \item Install partial detector in PX56 by end of Run 2 (YETS 2017 + TS in 2018) \item Commission and take data in order to evaluate beam-on backgrounds {\it in situ} \item Construction + Installation of remainder of detector during LS2 (2019--2020) \item Final commissioning by spring 2021 \item Operate detector for physics for duration of Run 3 and HL-LHC (mid 2021--) \end{itemize} \noindent The next step in the milliQan project is to seek external funding to enable at least the 10\% construction. No such funding has yet been secured for this project, but one or more proposals to one or more funding agencies are being prepared for the near future. \section{Summary \label{sec:end}} \noindent In this LOI we have proposed a dedicated experiment that would detect ``milli-charged" particles produced by pp collisions at LHC Point 5. The experiment would be installed during LS2 in the vestigial drainage gallery above UXC and would not interfere with CMS operations. Our calculations and simulations indicate that with 300~fb$^{-1}$ of integrated luminosity, sensitivity to a particle with charge $\mathcal{O}(10^{-3})~e$ can be achieved for masses of $\mathcal{O}(1)$~GeV, and charge $\mathcal{O}(10^{-2})~e$ for masses of $\mathcal{O}(10)$~GeV. This would greatly extend the parameter space explored for particles with small charge and masses above 100 MeV. We have performed sufficient R\&D to encourage us to proceed with securing funding for the project, and with this letter of intent we express the intention to do so. \begin{acknowledgments} \noindent We wish to thank Tiziano Camporesi, Joel Butler, and the CMS collaboration for their encouragement. We would also like to thank Vladimir Ivanchenko, Andrea Dotti and Mihaly Novak for useful discussions regarding {\sc Geant4}. \end{acknowledgments}
a5fd2f310d6714cde82b7c04bb3e6f796c5aae70
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction Throughout this paper, let $R$ denote a commutative Noetherian ring with identity and ${\mathfrak a}$ a proper ideal of $R$. For any non-zero $R$-module $M$, the ith local cohomology module of $M$ is defined as $$H^{i}_{{\mathfrak a}}(M):=\lim_{\underset{n\geq 1}{\longrightarrow}} \operatorname{Ext}^{i}_{R}(R/{{\mathfrak a}^n},M).$$ $\operatorname{V}({\mathfrak a})$ denotes the set of all prime ideals of $R$ containing ${\mathfrak a}$. For an $R$-module $M$, the {\it cohomological dimension} of $M$ with respect to ${\mathfrak a}$ is defined as $\operatorname{cd}({\mathfrak a},M):=\sup \{i\in \mathbb{Z}\mid H^{i}_{{\mathfrak a}}(M)\neq 0\}$ which is known that for a local ring $(R,{\mathfrak m})$ and ${\mathfrak a}={\mathfrak m}$, this is equal to the dimension of $M$. For unexplained notation and terminology about local cohomology modules, we refer the reader to \cite{BH} and \cite{BSH}. The notion of {\it cohomological dimension filtration} (abbreviated as {\it cd-filtration}) of $M$ introduced by A. Atazadeh and et al. \cite{ASN} which is a generalization of the concept of dimension filtration that is defined by P. Schenzel \cite{Sch} in local case. For any integer $0\leq i\leq \operatorname{cd}({\mathfrak a},M)$, let $M_{i}$ denote the largest submodule of $M$ such that $\operatorname{cd}({\mathfrak a},M_{i})\leq i$. Because of the maximal condition of a Noetherian $R$-module the submodules $M_{i}$ of $M$ are well-defined. Moreover, it follows that $M_{i-1}\subseteq M_{i}$ for all $1\leq i\leq\operatorname{cd}({\mathfrak a},M)$. In the present article, we will use the concept of relative Cohen-Macaulay modules. An $R$-module $M$ is {\it relative Cohen-Macaulay} w.r.t ${\mathfrak a}$ whenever $H^{i}_{{\mathfrak a}}(M)=0$ for all $i\neq\operatorname{height}_{M}({\mathfrak a})$. In other words, $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $\operatorname{grade}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},M)$ (see \cite{Z}). Notice that this concept has a connection with a notion which has been studied under the title of {\it cohomologically complete intersection ideals} in \cite{HSch}. It is well-known that $\operatorname{height}_R({\mathfrak a})\leq\operatorname{cd}({\mathfrak a},R)\leq\operatorname{ara}({\mathfrak a})$. The ideal ${\mathfrak a}$ is called a set-theoretic complete intersection ideal whenever $\operatorname{height}_R {\mathfrak a}=\operatorname{ara}({\mathfrak a})$. A set-theoretic complete intersection ideal is a cohomologically complete intersection ideal. Recently, relative Cohen-Macaulay modules have been studied also in \cite{HS}.\\ Sharp \cite{SH} and some other authors have shown that a Cohen-Macaulay local ring $R$ admits a canonical module if and only if it is the homomorphic image of a Gorenstein local ring. In particular, if $R$ is a complete Cohen-Macaulay local ring of dimension $n$, then $w_{R}=\operatorname{Hom}_{R}(H^{n}_{{\mathfrak m}}(R),E(R/ {\mathfrak m}))$ is a canonical module of $R$. The outline of the paper is as follows.\\ Section 2 is devoted to discuss main topics of this paper. We initiate this section by showing that if $(R,{\mathfrak m})$ is a complete local ring and $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},M)=c>0$, and $\operatorname{Supp}_{R}(H^{c}_{{\mathfrak a}}(M))\subseteq\operatorname{V}({\mathfrak m})$, then ${\mathfrak a}$ contains a regular element on $D_{R}(H^{c}_{{\mathfrak a}}(M))$ (see Corollary~\ref{co}). In this horizon, we prove the following theorem (see Theorem~\ref{p3}). \begin{thm}\label{p0} Let $(R,{\mathfrak m})$ be a local ring and let $M$ be a finite $R$-module with $\operatorname{cd}({\mathfrak a},M)=c>0$. Assume that $\underline{x}=x_{1},\ldots ,x_{n}\in{\mathfrak a}$ is a regular sequence on both $M$ and $D_R(H^{c}_{{\mathfrak a}}(M))$. Then $$\operatorname{cd}({\mathfrak a},M/\underline{x}M)=\operatorname{cd}({\mathfrak a},M)-n.$$ \end{thm} As an application of Theorem~\ref{p0}, we bring the next result (see Corollary~\ref{c3}). \begin{cor} Let $(R,{\mathfrak m})$ be a local ring and $M$ be a finite $R$-module with $\operatorname{cd}({\mathfrak a},M)=c>0$. \begin{itemize} \item[(i)] Let $x\in{\mathfrak a}$ be a regular element on both $M$ and $D_R(H^{c}_{{\mathfrak a}}(M))$. Then $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $M/xM$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \item[(ii)] Let $\underline{x}=x_{1},\ldots ,x_{n}\in{\mathfrak a}$ be a regular sequence on both $M$ and $D_R(H^{c}_{{\mathfrak a}}(M))$. Then $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $M/ \underline{x}M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \end{itemize} \end{cor} We define the {\it cohomological deficiency modules} of $M$ as the matlis duality of $H^{i}_{{\mathfrak a}}(M)$ for $i\neq \operatorname{cd}({\mathfrak a},M)$, denoted by $K^{i}_{{\mathfrak a}}(M)$ and further {\it cohomological canonical module} of $M$ if $i=\operatorname{cd}({\mathfrak a},M)$. One of the main purposes of this section is to introduce a new class of modules over $R$ called {\it relative Cohen-Macaulay filtered modules} ({\it or relative sequentially Cohen-Macaulay modules}) w.r.t ${\mathfrak a}$, abbreviated by RCMF modules. This is an extension of the concept of {\it Cohen-Macaulay filtered modules} ({\it sequentially Cohen-Macaulay modules}) introduced by P. Schenzel \cite{Sch} for local case. The concept of sequentially Cohen-Macaulay modules was introduced by Stanley \cite{S} for graded modules. It is interesting that any relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ is an RCMF $R$-module w.r.t ${\mathfrak a}$. But, any RCMF $R$-module w.r.t ${\mathfrak a}$ is not a relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ necessarily. We derive permanence properties of RCMF modules. More precisely, basic properties of RCMF modules with respect to non-zerodivisors, localization and completion are discussed. One of the main results of this section is the following theorem (see Theorem~\ref{Th1}). \medskip \begin{thm} Let $(R,{\mathfrak m})$ be a local ring and $M$ be a finite $R$-module with the cd-filtration $\mathcal{M}=\{M_{i}\}^{c}_{i=0}$ where $c=\operatorname{cd}({\mathfrak a},M)$. \begin{itemize} \item[(i)] Let $x\in{\mathfrak a}$ be a regular element on $M$, $D_R(H^{c}_{{\mathfrak a}}(M))$, and $D_R(H^{i}_{{\mathfrak a}}(M_i))$ for all $0\leq i\leq c$. Then $M$ is an RCMF module w.r.t ${\mathfrak a}$ if and only if $M/xM$ is an RCMF module w.r.t ${\mathfrak a}$. \item[(ii)] Let $\underline{x}=x_{1},\ldots ,x_{n}\in{\mathfrak a}$ be a regular sequence on $M$, $D_R(H^{c}_{{\mathfrak a}}(M))$, and $D_R(H^{i}_{{\mathfrak a}}(M_i))$ for all $0\leq i\leq c$. Then $M$ is an RCMF module w.r.t ${\mathfrak a}$ if and only if $M/ \underline{x}M$ is an RCMF module w.r.t ${\mathfrak a}$. \end{itemize} \end{thm} \medskip As another main result of this section, we provide a necessary and sufficient condition for a filtration to be cd-filtration of a module by the associated prime ideals of its factors (see Theorem~\ref{Th4}). \begin{thm} Let $\mathcal{M}=\{M_i\}_{i=0}^{c}$ be a filtration of the finite $R$-module $M$ and $\operatorname{cd}({\mathfrak a},M_{0})=0$. The following conditions are equivalent: \begin{itemize} \item[(i)] $\operatorname{Ass}(M_{i}/M_{i-1})=\operatorname{Ass}^{i}(M)$ for all $1\leq i\leq c$; \item[(ii)] $\mathcal{M}$ is the cd-filtration of $M$. \end{itemize} \end{thm} Atazadeh and et al. \cite[Theorem 1.1]{ASN} proved that if a finite $R$-module $M$ has a cd-filtration, then this filtration is uniquely determined by a reduced primary decomposition of the zero submodule in $M$. In the present paper, without such a condition on $M$, we present a cd-filtration for all $R$-modules whose zero submodule has a primary decomposition (see Corollary~\ref{c5}). \\ In section 3, we study relative Cohen-Macaulayness in rings and modules. In view of \cite[Proposition 2.3]{M}, it follows that if $(R,{\mathfrak m})$ is a Cohen-Macaulay ring, then for a maximal Cohen-Macaulay non-zero module $M$, relative Cohen-Macaulayness of the ring $R$ and the module $M$ is equivalent, whenever $\operatorname{Supp}_R(M)=\operatorname{Spec}(R)$. We determine equivalency between two classes of relative Cohen-Macaulay rings and modules, ``multiplication" and ``semidualizing" modules, in Corollaries~\ref{c1} and ~\ref{cc}. Among other things, in Proposition~\ref{prop31}, comparing with \cite[Proposition 5.1]{M}, we show that $R$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if its canonical module is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. Finally, as any relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ is an RCMF module w.r.t ${\mathfrak a}$, if $(R,{\mathfrak m})$ is a relative Cohen-Macaulay local ring w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},R)=c$, then for all $0\leq i\leq c$ the $R$-modules $K^{i}_{{\mathfrak a}}(R)$ are either zero or $i$-cohomological dimensional relative Cohen-Macaulay modules w.r.t ${\mathfrak a}$ (see Proposition~\ref{prop6}). \section{Cohomological dimension filtration and relative Cohen-Macaulay filtered modules} In this section, we define and study relative Cohen-Macaulay filtered modules. As the main objective, a characterization for such modules is presented in Theorem~\ref{Th1} and also a characterization of cd-filtration is presented in Theorem~\ref{Th4}. We begin by recalling the definition of cohomological dimension filtration due to Atazadeh and et al. in \cite{ASN} and the concept of relative Cohen-Macaulayness due to Zargar in \cite{Z}. \begin{dfn}(see \cite{ASN}) Let $M$ be a finite $R$-module. The increasing filtration $\mathcal{M}=\{M_{i}\}^{c}_{i=0}$ of submodules of $M$, where $c=\operatorname{cd}({\mathfrak a},M)$ is called the {\it cohomological dimension filtration} of $M$ if for all integer $0\leq i\leq c$, $M_{i}$ is the largest submodule of $M$ such that $\operatorname{cd}({\mathfrak a},M_{i})\leq i$. \end{dfn} \medskip \begin{dfn}(see \cite{Z}) A finite $R$-module $M$ is called {\it relative Cohen-Macaulay} w.r.t ${\mathfrak a}$ if there is precisely one non-vanishing local cohomology module w.r.t ${\mathfrak a}$. Clearly, this is the case if and only if $\operatorname{grade}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},M)$. \end{dfn} \medskip These definitions motivate us to introduce the following concept. \begin{dfn}\label{defin} Let $M$ be a finite $R$-module and $\mathcal{M}=\{M_{i}\}^{c}_{i=0}$ be the cohomological dimension filtration of submodules of $M$, where $c=\operatorname{cd}({\mathfrak a},M)$. $M$ is called a {\it relative Cohen-Macaulay filtered module} ({\it relative sequentially Cohen-Macaulay module}) w.r.t ${\mathfrak a}$, whenever $\mathcal{M}_{i}={M_{i}}/{M_{i-1}}$ is either zero or an $i$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ for all $1\leq i\leq c$. Let us abbreviate this notion by RCMF. \end{dfn} \medskip Related to the definition of RCMF modules, we state the notion of {\it relative Cohen-Macaulay filtration} w.r.t ${\mathfrak a}$ which will be useful in the process. \begin{dfn} Let $M$ be a fininte $R$-module with $\operatorname{cd}({\mathfrak a},M)=c$. An increasing filtration ${\mathcal{C}}=\{C_{i}\}^{c}_{i=0}$ of submodules of $M$ is called a {\it relative Cohen-Macaulay filtration} of $M$ w.r.t ${\mathfrak a}$ whenever $C_{c}=M$ and ${\mathcal{C}_{i}=C_{i}/C_{i-1}}$ is either zero or an $i$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ for all ${1\leq i\leq c}$. \end{dfn} \medskip \begin{prop}\label{prop1} Let ${\mathcal{C}}=\{C_{i}\}^{c}_{i=0}$ be the relative Cohen-Macaulay filtration of the $R$-module $M$ w.r.t ${\mathfrak a}$. Then ${\mathcal{C}}$ coinsides with the cohomological dimension filtration. \end{prop} \begin{proof} First, it is clear that $\operatorname{cd}({\mathfrak a},C_i)\leq i$ for all $0\leq i\leq c$. Also we have $$\operatorname{Ass}_R C_i=\{{\mathfrak p}\in\operatorname{Ass}_R M\mid \operatorname{cd}({\mathfrak a},R/{\mathfrak p})\leq i\}$$ by \cite[Proposition 2.6]{ASN}. This implies that $C_i=H^{0}_{{\mathfrak a}_i}(C_j)$ in which ${\mathfrak a}_i=\prod_{\operatorname{cd}({\mathfrak a},R/ {\mathfrak p}_{j})\leq i}{\mathfrak p}_{j}$ for all $0\leq i\leq c$. Now let $0\leq i\leq c$ and $j\geq i$. Consider the following exact sequence $$0\longrightarrow C_j\longrightarrow C_{j+1}\longrightarrow \mathcal{C}_{j+1}\longrightarrow 0.$$ As $\mathcal{C}_{j+1}$ is either zero or $(j+1)$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$, it follows that $H^{0}_{{\mathfrak a}_i}(C_j)\cong H^{0}_{{\mathfrak a}_i}(C_{j+1})$. Thus $C_i=H^{0}_{{\mathfrak a}_i}(C_j)$ for all $j\geq i$. Also since $M=C_c$, the proof will be completed by \cite[Proposition 2.3]{ASN}. \end{proof} \begin{rem}\label{R2} Let $\mathcal{M}=\{M_{i}\}^{c}_{i=0}$ be the cd-filtration of $M$ where $c=\operatorname{cd}({\mathfrak a},M)$. Let $1\leq i\leq c$. Considering the exact sequence $0\longrightarrow M_{i-1}\longrightarrow M_{i}\longrightarrow M_{i}/M_{i-1}\longrightarrow 0$, we have $$\operatorname{cd}({\mathfrak a},M_{i})=\max\{\operatorname{cd}({\mathfrak a},M_{i-1}), \operatorname{cd}({\mathfrak a},M_{i}/M_{i-1})\}$$ by \cite[Corollary 2.3 (i)]{DNT}. Thus $\operatorname{cd}({\mathfrak a},M_{i})=\operatorname{cd}({\mathfrak a},M_{i}/M_{i-1})$ for all $1\leq i\leq c$. \end{rem} \medskip Now recall that the {\it finiteness dimension} of $M$ relative to ${\mathfrak a}$, $f_{{\mathfrak a}}(M)$, is defined by $$ f_{{\mathfrak a}}(M):=\inf\{i\in\mathbb{N}_{0}: H^{i}_{{\mathfrak a}}(M)\ \text{is not finitely generated}\}. $$ Another formulation is the {\it ${\mathfrak b}$-finiteness dimension} of $M$ relative to ${\mathfrak a}$ which is defined by $$ f_{{\mathfrak a}}^{{\mathfrak b}}(M):=\inf\{i\in\mathbb{N}_{0}: {\mathfrak b}\nsubseteq\operatorname{rad} (0:_{R}H^{i}_{{\mathfrak a}}(M))\}. $$ Moreover, {\it the ${\mathfrak b}$-minimum ${\mathfrak a}$-adjusted depth} of $M$, denoted by $\lambda_{{\mathfrak a}}^{{\mathfrak b}}(M)$, is defined by $$ \lambda_{{\mathfrak a}}^{{\mathfrak b}}(M):=\inf\{\operatorname{depth} M_{{\mathfrak p}}+\operatorname{height} ({\mathfrak a}+{\mathfrak p})/{\mathfrak p}: {\mathfrak p}\in\operatorname{Spec}(R)\setminus\operatorname{V}({\mathfrak b})\}, $$ where ${\mathfrak b}$ is the second ideal of $R$ without assuming ${\mathfrak b}\subseteq{\mathfrak a}$ in general. By convention, the infimum of the empty set of integers is interpreted by $\infty$. \medskip \begin{lem}\label{r3} Let $M$ be a relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},M)>0$. Then $$ \operatorname{cd}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},R/{\mathfrak p})\ \text{for all}\ {\mathfrak p}\in\operatorname{Ass}_{R}(M). $$ \end{lem} \begin{proof} First, note that we have $f_{{\mathfrak a}}(M)=\operatorname{cd}({\mathfrak a},M)$ as $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ and so $\operatorname{cd}({\mathfrak a},M)\leq \lambda_{{\mathfrak a}}^{{\mathfrak a}}(M)$ by \cite[Theorem 9.3.7]{BSH}. Let ${\mathfrak p}\in\operatorname{Ass}_R(M)$. If ${\mathfrak p}\in\operatorname{Ass}_{R}(M)\cap\operatorname{V}({\mathfrak a})$, then $$\operatorname{cd}({\mathfrak a},M)\leq\lambda_{{\mathfrak a}}^{{\mathfrak a}}(M)\leq\operatorname{depth} M_{{\mathfrak p}}+\operatorname{height}({\mathfrak a}+{\mathfrak p})/{\mathfrak p}.$$ That is $\operatorname{cd}({\mathfrak a},M)\leq 0$ which is a contradiction. Thus ${\mathfrak p}\in\operatorname{Ass}_{R}(M)\setminus \operatorname{V}({\mathfrak a})$. Now, by the Independence Theorem \cite[Theorem 4.2.1]{BSH}, we have \[\begin{array}{ll} \operatorname{cd}({\mathfrak a},M)\leq\lambda_{{\mathfrak a}}^{{\mathfrak a}}(M)&\leq\operatorname{depth} M_{{\mathfrak p}}+\operatorname{height}({\mathfrak a}+{\mathfrak p})/ {\mathfrak p}\\&=\operatorname{height}({\mathfrak a}+{\mathfrak p})/ {\mathfrak p}\\&\leq \operatorname{cd}(({\mathfrak a}+{\mathfrak p})/ {\mathfrak p},R/ {\mathfrak p})\\&=\operatorname{cd}({\mathfrak a},R/ {\mathfrak p})\\&\leq\operatorname{cd}({\mathfrak a},M). \end{array}\] Therefore $\operatorname{cd}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},R/{\mathfrak p})$. \end{proof} \medskip Here several examples of RCMF modules are provided. In order to prove part (e), we bring the following remark. \begin{rem}\label{r8} Let $R\longrightarrow R'$ be a faithfully flat ring homomorphism and $M$ be an $R$-module. Then by using the Flate Base Change Theorem \cite[Theorem 4.3.2]{BSH} we get $\operatorname{cd}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a} R',M\otimes_R R')$ and $\operatorname{grade}({\mathfrak a},M)=\operatorname{grade}({\mathfrak a} R',M\otimes_R R')$. \end{rem} \medskip \begin{exmps}\label{E1} \begin{itemize} \item[(a)] Any relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ is an RCMF module w.r.t ${\mathfrak a}$. \item[(b)] Let $M$ be an $R$-module with $\operatorname{cd}({\mathfrak a},M)=1$. Then $M$ is an RCMF module w.r.t ${\mathfrak a}$. \item[(c)] Let $R=k[X,Y]/(XY)$ and ${\mathfrak a}=(x)$ be an ideal of $R$ in which $x$ is the image of $X$ in $R$. Then $R$ is RCMF w.r.t ${\mathfrak a}$. \item[(d)] Let $R$ be a ring with $\operatorname{cd}({\mathfrak a},R)=c$. Let $N_{i}$, $0\leq i\leq c$ be a family of $R$-modules such that either $N_{i}=0$ or $N_{i}$ is $i$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$. Then $M=\oplus_{i=0}^{c}N_{i}$ is an RCMF module w.r.t ${\mathfrak a}$ over $R$. \item[(e)] Let $R[[x]]$ be the formal power series ring in one variable $x$ over the ring $R$. Then a finite $R$-module $M$ is an RCMF module w.r.t ${\mathfrak a}$ if and only if $M[[x]]$ is an RCMF module over the ring $R[[x]]$ w.r.t ${\mathfrak a} R[[x]]$. \end{itemize} \end{exmps} \begin{proof} (a) Let $M$ be a relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},M)=c$. We can assume that $c>0$. Since $\operatorname{cd}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},R/{\mathfrak p})$ for all ${\mathfrak p}\in\operatorname{Ass}_{R}(M)$ by Lemma~\ref{r3}, we have $M_c=M$, and $M_{i}=0$ for all $0\leq i<c$. It follows that $M$ is an RCMF module w.r.t ${\mathfrak a}$.\\ To prove part (b), let $M_0 \subseteq M_1=M$ be the cd-filtration of $M$. We show that $M/M_0$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. By Remark~\ref{R2}, $\operatorname{cd}({\mathfrak a},M/M_0)=\operatorname{cd}({\mathfrak a},M)=1$. On the other hand, since $\operatorname{cd}({\mathfrak a},\Gamma_{{\mathfrak a}}(M))\leq 0$, we have $\Gamma_{{\mathfrak a}}(M_0)=\Gamma_{{\mathfrak a}}(M)$ as $M_0$ is the largest submodule of $M$ with $\operatorname{cd}({\mathfrak a},M_0)\leq 0$. Therefore considering the following exact sequence $$ 0\longrightarrow H^{0}_{{\mathfrak a}}(M_0)\longrightarrow H^{0}_{{\mathfrak a}}(M)\longrightarrow H^{0}_{{\mathfrak a}}(M/M_0)\longrightarrow H^{1}_{{\mathfrak a}}(M_0)\longrightarrow\cdots, $$ we get $H^{0}_{{\mathfrak a}}(M/M_0)=0$ and so $\operatorname{grade}({\mathfrak a},M/M_0)=1$ as desired. \\ For part (c), as it is seen in \cite[Example 2.3]{V}, $\operatorname{cd}({\mathfrak a},R)=1\neq 0=\operatorname{height}_R({\mathfrak a})$. Hence $R$ is not relative Cohen-Macaulay w.r.t ${\mathfrak a}$. On the other hand, $R$ is RCMF w.r.t ${\mathfrak a}$ as (b) holds. This example shows that being RCMF w.r.t ${\mathfrak a}$ does not lead to relative Cohen-Macaulayness w.r.t ${\mathfrak a}$ necessarily.\\ Part (d) follows by Proposition~\ref{prop1} as $M$ admits a filtration $M_{i}=\oplus_{j=0}^{i} N_{j}$ such that $M_{i}/M_{i-1}\cong N_{i}$, is either zero or an $i$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ for all $1\leq i\leq c$. One can prove this by definitions and the fact that $\operatorname{cd}({\mathfrak a},\oplus_{j=0}^{i}N_j)=i$ for all $0\leq i\leq c$. \\ For the last part, apply Remark~\ref{r8} as $R[[x]]$ is isomorphic to the $I$-adic completion of $R[x]$, where $I=(x)$ and $R[x]$ is faithfully flat over $R$. \end{proof} \medskip \begin{dfn}\label{def} Let $(R,{\mathfrak m},k)$ be a local ring, ${\mathfrak a}$ an ideal of $R$ and $M$ be a finite $R$-module with $\operatorname{cd}({\mathfrak a},M)=c$. For $i \neq c$, the {\it ith cohomological deficiency module} of $M$ is defined by $$K^{i}_{{\mathfrak a}}(M):=D_{R}(H^{i}_{{\mathfrak a}}(M))=\operatorname{Hom}_{R}(H^{i}_{{\mathfrak a}}(M),E_{R}(k)).$$ The module $K(M):=K^{c}_{{\mathfrak a}}(M)$ is called the {\it cohomological canonical module} of $M$. Note that $K^{i}_{{\mathfrak a}}(M)=0$ for all $i<0$ or $i>c$. \end{dfn} \medskip According to the above definition, we bring the following property of the cohomological canonical module of $R$. \begin{lem} \label{lemm1} Let $(R,{\mathfrak m})$ be a relative Cohen-Macaulay local ring w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},R)=c$. Then $K^{c}_{{\mathfrak a}}(R)$ is a faithful relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ of finite injective dimension of type one. \end{lem} \begin{proof} Firstly, in \cite[Theorem 4.3 (ii), (iii)]{Z}, it is shown that $K^{c}_{{\mathfrak a}}(R)$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ of finite injective dimension of type one. Next, as $R$ is a relative Cohen-Macaulay ring w.r.t ${\mathfrak a}$, we have $\operatorname{Ann}_{R}H^{c}_{{\mathfrak a}}(R)=0$ by \cite[Theorem 3.3]{L}. Hence $\operatorname{Ann}_{R}K^{c}_{{\mathfrak a}}(R)=0$ because the annihilators of an $R$-module and its Matlis dual are equal by \cite[Remarks 10.2.2]{BSH}. This completes the proof. \end{proof} \medskip In the next proposition, we provide a cohomological result of RCMF modules. \begin{prop}\label{lemma} Let $M$ be an RCMF module w.r.t ${\mathfrak a}$ with ${\mathcal{M}}=\{M_{i}\}^{c}_{i=0}$ its cd-filtration, where $c=\operatorname{cd}({\mathfrak a},M)$. Then for all $1\leq i\leq c$, $$H^{i}_{{\mathfrak a}}(M)\cong H^{i}_{{\mathfrak a}}(M_{i})\cong H^{i}_{{\mathfrak a}}({\mathcal{M}}_{i}),$$ where $\mathcal{M}_i=M_i/M_{i-1}$ for all $1\leq i\leq c$. In particular, it follows that $K^{i}_{{\mathfrak a}}(M)\cong K^{i}_{{\mathfrak a}}({\mathcal{M}}_{i})$ for all ${1\leq i\leq c}.$ \end{prop} \begin{proof} Let $1\leq i\leq c$. By using the short exact sequence $$0\longrightarrow M_{i-1}\longrightarrow M_{i}\longrightarrow {\mathcal{M}}_{i}\longrightarrow 0$$ and since $\operatorname{cd}({\mathfrak a},M_{i-1})\leq i-1$, we have an isomorphism $H^{i}_{{\mathfrak a}}(M_{i})\cong H^{i}_{{\mathfrak a}}({\mathcal{M}}_{i})$. Now, as ${\mathcal{M}}_{i}=M_{i}/M_{i-1}$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$, it yields isomorphisms $H^{j}_{{\mathfrak a}}(M_{i})\cong H^{j}_{{\mathfrak a}}(M_{i-1})$ for all $j<i$. By induction, it follows that $$H^{i}_{{\mathfrak a}}(M)=H^{i}_{{\mathfrak a}}(M_{c})\cong H^{i}_{{\mathfrak a}}(M_{c-1})\cong \ldots \cong H^{i}_{{\mathfrak a}}(M_{i+1})\cong H^{i}_{{\mathfrak a}}(M_{i}),$$ which completes the first part of the assertion. Now by virtue of Definition~\ref{def}, we obtain $K^{i}_{{\mathfrak a}}(M)\cong K^{i}_{{\mathfrak a}}({\mathcal{M}}_{i})$ for all $1\leq i\leq c$ as desired. \end{proof} We will provide the definitions and results which are needed in the process. The following definition is a generalization of the concept of ${\mathfrak a}$-filter regular $M$-sequences which has been stated in \cite{CSCHT}. \begin{dfn}(see \cite{T}) A sequence $x_1,\ldots,x_n$ of elements of $R$ is called an {\it ${\mathfrak a}$-filter regular $M$-sequence} if $x_i\notin{\mathfrak p}$ for all $$ {\mathfrak p}\in\operatorname{Ass}_R(M/(x_1,\ldots,x_{i-1})M)\setminus\operatorname{V}({\mathfrak a})\ \text{for all} \ 1\leq i\leq n. $$ \end{dfn} \medskip By definition, it deduces that every regular $M$-sequence is an ${\mathfrak a}$-filter regular $M$-sequence and any $R$-filter regular $M$-sequence is a poor regular $M$-sequence. \begin{rem}\label{re} \begin{itemize} \item[(i)] Let $n$ be a positive integer. By definition, we can find $n$ elements of ${\mathfrak a}$ which form an ${\mathfrak a}$-filter regular $M$-sequence as follows. If $\operatorname{Ass}_R(M)\subseteq\operatorname{V}({\mathfrak a})$, then choose $y_1\in{\mathfrak a}$ arbitrarily. If not, since ${\mathfrak a}\nsubseteq\cup_{{\mathfrak p}\in\operatorname{Ass}_R(M)\setminus\operatorname{V}({\mathfrak a})}{\mathfrak p}$, there exists $y_1\in{\mathfrak a}$ such that $y_1\notin{\mathfrak p}$ for all ${\mathfrak p}\in\operatorname{Ass}_R(M)\setminus\operatorname{V}({\mathfrak a})$. Again, if $\operatorname{Ass}_R(M/y_1 M)\subseteq\operatorname{V}({\mathfrak a})$, then choose $y_2\in{\mathfrak a}$ arbitrarily. If not, since ${\mathfrak a} \nsubseteq\cup_{{\mathfrak p}\in\operatorname{Ass}_R(M/y_1 M)\setminus\operatorname{V}({\mathfrak a})}{\mathfrak p}$, there esists $y_2\in{\mathfrak a}$ such that $y_2\notin{\mathfrak p}$ for all ${\mathfrak p}\in\operatorname{Ass}_R(M/y_1 M)\setminus\operatorname{V}({\mathfrak a})$. Proceeding the same way, we can find $y_1,\ldots,y_n\in{\mathfrak a}$ which form an ${\mathfrak a}$-filter regular $M$-sequence. \item[(ii)] For any positive integer $n$, there are $n$ elements of $R$ which form a poor regular $M$-sequence. \end{itemize} \end{rem} \medskip \begin{dfn} (see \cite{O} and \cite{R}) For a ring $R$, let $E_{R}$ be the injective hull of the direct sum $\oplus_{{\mathfrak m}\in\operatorname{Max}(R)}R/{\mathfrak m}$ of all simple $R$-modules and $D_{R}(-)$ be the functor $\operatorname{Hom}_{R}(-,E_{R})$. (Note that $D_{R}(-)$ is a natural generalization of Matlis duality functor to non-local rings.) \end{dfn} \medskip \begin{lem}\label{lem} Let $M$ be a finite dimensional finite $R$-module and $n$ be a positive integer such that $\operatorname{Ext}^{j}_{R}(R/{\mathfrak a},D_{R}(H^{t}_{{\mathfrak a}}(M)))=0$ for all $t>n$ and all $j\in \mathbb{N}_{0}$. Then for any ${\mathfrak a}$-filter regular $M$-sequence $x_{1},\ldots ,x_{i}$ with $i>n$, $\operatorname{Ext}^{j}_{R}(R/{\mathfrak a},D_{R}(H^{i}_{(x_{1},\ldots ,x_{i})}(M)))=0$ for all $j\in \mathbb{N}_{0}$. In particular, it holds for $n=\operatorname{cd}({\mathfrak a},M)$. \end{lem} \begin{proof} In the proof of \cite[Lemma 3.3]{KhKh}, it is used the fact that for every exact sequence of $R$-modules, the finiteness of sided modules lead to the finiteness of the middle one. Thus the same method of the proof works exactly replacing ``finite modules" by ``zero modules". \end{proof} \medskip Applying Lemma~\ref{lem}, we obtain the following result which is needed for Corollary~\ref{co}. \begin{lem}\label{lemm} Let $n$ be a positive integer such that $\operatorname{Ext}^{j}_{R}(R/{\mathfrak a},D_{R}(H^{t}_{{\mathfrak a}}(M)))=0$ for all $t>n$ and all $j\in \mathbb{N}_{0}$. Then $\operatorname{Hom}_{R}(R/{\mathfrak a},D_{R}(H^{n}_{{\mathfrak a}}(M)))=0$. In particular, it holds for $n=\operatorname{cd}({\mathfrak a},M)$. \end{lem} \begin{proof} By Grothendieck's Vanishing Theorem \cite[Theorem 6.1.2]{BSH}, we may assume that $n\leq\dim M$. In view of Remark~\ref{re}(i) and using Lemma~\ref{lem}, we can trace the same method of the proof of \cite[Theorem 3.4]{KhKh} by replacing ``finite modules" by ``zero modules" to get the assertion. \end{proof} Notice that the assumption ``complete local" in \cite[Lemma 3.3 and Theorem 3.4]{KhKh} is due to show that the finiteness properties of local cohomology modules. But we eliminated it in Lemma~\ref{lemm} because we do not need this assumption for vanishing of local cohomology modules. \medskip \begin{dfn} (see \cite{H}) An $R$-module $M$ is called ${\mathfrak a}$-{\it cofinite} if $\operatorname{Supp}_{R}(M)\subseteq\operatorname{V}({\mathfrak a})$ and $\operatorname{Ext}^{j}_{R}(R/{\mathfrak a},M)$ is finite for all $j\geq 0$. \end{dfn} \medskip \begin{rem}\label{r2} (see \cite[Theorem 2.1]{DY}) For a finite $R$-module $M$ and a non-negative integer $c$ if $H^{i}_{{\mathfrak a}}(M)$ is ${\mathfrak a}$-cofinite for all $i<c$, then $\operatorname{Hom}_{R}(R/{\mathfrak a},H^{c}_{{\mathfrak a}}(M))$ is finite. \end{rem} \medskip \begin{lem}\label{l3} Let $(R,{\mathfrak m})$ be a complete local ring, $c$ a non-negative integer. Let $H^{i}_{{\mathfrak a}}(M)=0$ for all $i<c$ and $\operatorname{Supp}_{R}(M/{\mathfrak a} M)\subseteq\operatorname{V}({\mathfrak m})$. If $\operatorname{Hom}_{R}(R/{\mathfrak a},D_{R}(H^{c}_{{\mathfrak a}}(M)))=0$, then ${\mathfrak a}$ contains a regular element on $D_{R}(H^{c}_{{\mathfrak a}}(M))$. \end{lem} \begin{proof} By assumption, $H^{i}_{{\mathfrak a}}(M)$ is ${\mathfrak a}$-cofinite for all $i<c$. Thus $\operatorname{Hom}_{R}(R/{\mathfrak a},H^{c}_{{\mathfrak a}}(M))$ is finite by Remark~\ref{r2}. As $\operatorname{Supp}_{R}(M/{\mathfrak a} M)\subseteq\operatorname{V}({\mathfrak m})$, then $\operatorname{Supp}_{R}(\operatorname{Hom}_{R}(R/{\mathfrak a},H^{c}_{{\mathfrak a}}(M)))\subseteq\operatorname{V}({\mathfrak m})$ and so $\operatorname{Hom}_{R}(R/{\mathfrak a},H^{c}_{{\mathfrak a}}(M))$ is Artinian. Thus $H^{c}_{{\mathfrak a}}(M)$ is Artinian by \cite[Theorem 7.1.2]{BSH}. Now, as $R$ is a complete local ring, $D_{R}(H^{c}_{{\mathfrak a}}(M))$ is finite by \cite[Theorem 10.2.12]{BSH}. Also, since $\operatorname{Hom}_{R}(R/{\mathfrak a},D_{R}(H^{c}_{{\mathfrak a}}(M)))=0$ by our assumption, ${\mathfrak a}$ contains a regular element on $D_{R}(H^{c}_{{\mathfrak a}}(M))$ as required. \end{proof} \medskip Motivated by \cite[Lemma 4.3]{HSch}, we are particularly interested to prove the existance of a regular element on $D_{R}(H^{c}_{{\mathfrak a}}(M))$ as follows. \begin{cor}\label{co} Let $(R,{\mathfrak m})$ be a complete local ring and $M$ be a relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},M)=c>0$ and $\operatorname{Supp}_{R}(M/{\mathfrak a} M)\subseteq\operatorname{V}({\mathfrak m})$. Then ${\mathfrak a}$ contains a regular element on $D_{R}(H^{c}_{{\mathfrak a}}(M))$. \end{cor} \begin{proof} It is straightforward by using Lemma~\ref{lemm} and Lemma~\ref{l3}. \end{proof} \medskip Now, we consider the behavior of cohomological dimension of an $R$-module under non-zerodivisors. \begin{thm}\label{p3} Let $(R,{\mathfrak m})$ be a local ring and $M$ be a finite $R$-module with $\operatorname{cd}({\mathfrak a},M)=c>0$. If $\underline{x}=x_{1},\ldots ,x_{n}\in{\mathfrak a}$ is a regular sequence on both $M$ and $D_R(H^{c}_{{\mathfrak a}}(M))$, then $$ \operatorname{cd}({\mathfrak a},M/\underline{x}M)=\operatorname{cd}({\mathfrak a},M)-n. $$ \end{thm} \begin{proof} We prove by induction on the length of regular sequence $\underline{x}$. Let $n=1$ and $\widehat{R}$ be the ${\mathfrak m}$-adic completion of $R$. In view of the Flat Base Change Theorem \cite[Theorem 4.3.2]{BSH} as $\widehat{R}$ is faithfully flat over $R$, we have $\operatorname{cd}_R({\mathfrak a},M)=\operatorname{cd}_{\widehat{R}}({\mathfrak a}\widehat{R},\widehat{M})$ and $\operatorname{cd}_R({\mathfrak a},M/x_1 M)=\operatorname{cd}_{\widehat{R}}({\mathfrak a}\widehat{R},\widehat{M}/x_1\widehat{M})$. Thus we may assume that $R$ is a complete local ring. Let $\operatorname{cd}({\mathfrak a},M)=c$. It is clear that $\operatorname{cd}({\mathfrak a},M/x_1 M)\leq c$. We claim that $c-1\leq\operatorname{cd}({\mathfrak a},M/x_1 M)$. Suppose that $\operatorname{cd}({\mathfrak a},M/x_1 M)<c-1$. Then the multiplication map by $x_1$ on $H^{c}_{{\mathfrak a}}(M)$ is injective. As $H^{c}_{{\mathfrak a}}(M)$ is ${\mathfrak a}$-torsion, we get $H^{c}_{{\mathfrak a}}(M)=0$ which is a contradiction. Thus $c-1\leq\operatorname{cd}({\mathfrak a},M/x_1 M)\leq c$. Now, we show that $H^{c}_{{\mathfrak a}}(M/x_1 M)=0$. From the exact sequence $$ H^{c}_{{\mathfrak a}}(M)\overset{x_1}\longrightarrow H^{c}_{{\mathfrak a}}(M)\longrightarrow H^{c}_{{\mathfrak a}}(M/x_1 M)\longrightarrow 0, $$ we obtain the following exact sequence $$ D_{R}(H^{c}_{{\mathfrak a}}(M/x_1 M))\longrightarrow D_{R}(H^{c}_{{\mathfrak a}}(M))\overset{x_1}\longrightarrow D_{R}(H^{c}_{{\mathfrak a}}(M))\longrightarrow 0. $$ Since $x_1$ is a regular element on $D_R(H^{c}_{{\mathfrak a}}(M))$, we get $D_{R}(H^{c}_{{\mathfrak a}}(M/x_1 M))=0$. Since the annihilators of $H^{c}_{{\mathfrak a}}(M/x_1 M)$ and its Matlis dual are equal by \cite[Remark 10.2.2]{BSH}, it deduces that $H^{c}_{{\mathfrak a}}(M/x_1 M)=0$. Therefore $\operatorname{cd}({\mathfrak a},M/x_1 M)=c-1$, as desired. Now, assume that the assertion holds for any sequences of length $n-1$. Then the claim will be proved by using inductive hypothesis. \end{proof} \medskip We are now ready to bring an effective application of the above theorem which is needed in Theorem~\ref{Th1}. \begin{cor}\label{c3} Let $(R,{\mathfrak m})$ be a local ring and $M$ be a finite $R$-module with $\operatorname{cd}({\mathfrak a},M)=c>0$. \begin{itemize} \item[(i)] Let $x\in{\mathfrak a}$ be a regular element on both $M$ and $D_R(H^{c}_{{\mathfrak a}}(M))$. Then $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $M/xM$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \item[(ii)]Let $\underline{x}=x_{1},\ldots ,x_{n}\in{\mathfrak a}$ be a regular sequence on both $M$ and $D_R(H^{c}_{{\mathfrak a}}(M))$. Then $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $M/ \underline{x}M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \end{itemize} \end{cor} \begin{proof} (i) As $\operatorname{cd}({\mathfrak a},M/xM)=\operatorname{cd}({\mathfrak a},M)-1$ by Theorem~\ref{p3}, and $\operatorname{grade}({\mathfrak a},M/xM)=\operatorname{grade}({\mathfrak a},M)-1$, the assertion follows easily.\\ Part (ii) follows by part (i) and using induction on $n$. \end{proof} \medskip We are now in a position to bring a non-zerodivisor characterization of RCMF modules. \begin{thm}\label{Th1} Let $(R,{\mathfrak m})$ be a local ring and $M$ be a finite $R$-module with the cd-filtration $\mathcal{M}=\{M_{i}\}^{c}_{i=0}$ where $\operatorname{cd}({\mathfrak a},M)=c$. \begin{itemize} \item[(i)] Let $x\in{\mathfrak a}$ be a regular element on $M$, $D_R(H^{c}_{{\mathfrak a}}(M))$, and $D_R(H^{i}_{{\mathfrak a}}(M_i))$ for all $0\leq i\leq c$. Then $M$ is an RCMF module w.r.t ${\mathfrak a}$ if and only if $M/xM$ is an RCMF module w.r.t ${\mathfrak a}$. \item[(ii)] Let $\underline{x}=x_{1},\ldots ,x_{n}\in{\mathfrak a}$ be a regular sequence on $M$, $D_R(H^{c}_{{\mathfrak a}}(M))$, and $D_R(H^{i}_{{\mathfrak a}}(M_i))$ for all $0\leq i\leq c$. Then $M$ is an RCMF module w.r.t ${\mathfrak a}$ if and only if $M/ \underline{x}M$ is an RCMF module w.r.t ${\mathfrak a}$. \end{itemize} \end{thm} \begin{proof} For part (i), let $\mathcal{M}_{i}=M_i/M_{i-1}$. First, as $x\in{\mathfrak a}$ is an $M$-regular element, we show that $M_{0}=0$. By the notion of \cite[Proposition 2.3]{ASN}, we know that $M_{0}=H^{0}_{{\mathfrak a}_{0}}(M)$ where ${\mathfrak a}_{0}=\prod_{\operatorname{cd}({\mathfrak a},R/ {\mathfrak p}_{j})=0}{\mathfrak p}_{j}$. Let $\operatorname{cd}({\mathfrak a},R/ {\mathfrak p}_{j})=0$. Then we have $H^{0}_{{\mathfrak a}}(R/ {\mathfrak p}_{j})\neq 0$ and so there exists $r\in R\setminus {\mathfrak p}_{j}$ and $n\in\mathbb{N}$ such that $r{\mathfrak a}^{n}\subseteq {\mathfrak p}_{j}$. Thus ${\mathfrak a}\subseteq {\mathfrak p}_{j}$ that is impossible because ${\mathfrak p}_{j}\in\operatorname{Ass}_{R}(M)$. Therefore $\{{\mathfrak p}_{j}\mid \operatorname{cd}({\mathfrak a},R/ {\mathfrak p}_{j})=0\}=\emptyset$ and hence $H^{0}_{{\mathfrak a}_{0}}(M)=H^{0}_{R}(M)=0$. That is, $M_{0}=0$ as we claimed. Now, let $1\leq i\leq c$. Then $\operatorname{Ass}_{R}(\mathcal{M}_{i})\subseteq \operatorname{Ass}_{R}(M)$ by \cite[Proposition 2.6]{ASN}. Therefore $x$ is also $\mathcal{M}_{i}$-regular element for all $i\geq 1$. Also, it is easy to see that $M_{i}\cap xM=xM_{i}$ and $xM_i\cap M_{i-1}=xM_{i-1}$ for all $1\leq i \leq c$. (Note that if $m\in M$ and $xm\in M_i$, then $m\in M_i$ because $x$ is a non-zerodivisor on $\mathcal{M}_i$. Hence $xM\cap M_i\subseteq xM_i$.) Now, let $M$ be an RCMF module w.r.t ${\mathfrak a}$. Then we have \[\begin{array}{ll}\label{a1} \mathcal{M}_{i}/x\mathcal{M}_{i}\cong M_i/(xM_i+M_{i-1}) & \cong (M_i/xM_i)/(M_{i-1}/(xM_i \cap M_{i-1})) \\ & \cong (M_{i}/(M_{i}\cap xM))/(M_{i-1}/(xM\cap M_{i-1})) \\ & \cong ((M_{i},xM)/xM)/((M_{i-1},xM)/xM).\end{array}\] Therefore as $x$ is $\mathcal{M}_i$-regular element, $\{(M_{i+1},xM)/xM\}^{c-1}_{i=0}$ is a relative Cohen-Macaulay filtration of $M/xM$ w.r.t ${\mathfrak a}$ by Corollary~\ref{c3}. Notice that $H^{i}_{{\mathfrak a}}(M_i)\cong H^{i}_{{\mathfrak a}}(\mathcal{M}_i)$ for all $i$. Therefore by virtue of Proposition~\ref{prop1}, $M/xM$ is an RCMF module w.r.t ${\mathfrak a}$. Moreover, by Theorem~\ref{p3}, $M/xM$ is a $(c-1)$-cohomological dimensional RCMF module w.r.t ${\mathfrak a}$. Conversely, suppose that $M/xM$ is an RCMF module w.r.t ${\mathfrak a}$. Then the cohomological dimension filtration $\{M'_{i}\}^{c-1}_{i=0}$ of $M/xM$ has the property that $\mathcal {M'}_{i}=M'_{i}/M'_{i-1}$ is either zero or $i$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$. Let $M_{i+1}$ denote the preimage of $M'_{i}$ in $M$, $\mathcal{M}_{i+1}:=M_{i+1}/M_i$ for $0\leq i\leq c-1$, and $M_{0}:=0$. Since $\operatorname{Ass}_R\mathcal{M'}_i\subseteq\operatorname{Ass}_R M$ by \cite[Proposition 2.6]{ASN}, it deduces that $\operatorname{Ass}_R\mathcal{M}_i\subseteq\operatorname{Ass}_R M$ and so $x$ is an $\mathcal{M}_{i}$-regular element. By the isomorphisms as we mentioned in above, $\mathcal{M}_i/x\mathcal{M}_i\cong M'_{i-1}/M'_{i-2}$ for all $1\leq i\leq c$, as all $M_i$'s contain $xM$ (here put $M'_{-1}:=0$). Thus $\mathcal{M}_i/x\mathcal{M}_i$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ and $\operatorname{cd}({\mathfrak a},\mathcal{M}_i/x\mathcal{M}_i)=\operatorname{cd}({\mathfrak a},M'_{i-1}/M'_{i-2})$ for all $1\leq i\leq c$. Now, by Theorem~\ref{p3} and Corollary~\ref{c3}, $\mathcal{M}_i$ is $i$-cohomological dimensional relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ as $x$ is non-zerodivisor on $\mathcal{M}_i$. Therefore $M$ is RCMF module w.r.t ${\mathfrak a}$.\\ Part (ii) follows by part (i) and applying the induction on $n$. \end{proof} \medskip Another property of RCMF modules is about localization behaviour. It is clear that if $M$ is a relative Cohen-Macaulay $R$-module w.r.t ${\mathfrak a}$ and ${\mathfrak p}\in \operatorname{Supp}_{R}(M)\cap\operatorname{V}({\mathfrak a})$, then $M_{{\mathfrak p}}$ is relative Cohen-Macaulay $R_{{\mathfrak p}}$-module w.r.t ${\mathfrak a} R_{{\mathfrak p}}$ and $\operatorname{cd}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},M_{{\mathfrak p}})$. \begin{prop} Let $M$ be an RCMF $R$-module w.r.t ${\mathfrak a}$. Then $M_{{\mathfrak p}}$ is an RCMF $R_{{\mathfrak p}}$-module w.r.t ${\mathfrak a} R_{{\mathfrak p}}$ for any prime ideal ${\mathfrak p}\in \operatorname{Supp}_{R}(M)\cap\operatorname{V}({\mathfrak a})$. \end{prop} \begin{proof} Let $\mathcal{M}=\{M_{i}\}^{c}_{i=0}$ denote the cd-filtration of $M$, where $c=\operatorname{cd}({\mathfrak a},M)$. Let ${\mathfrak p}\in\operatorname{V}({\mathfrak a})$. Consider the filtration $\{M_{i}\otimes_{R} R_{{\mathfrak p}}\}^{c}_{i=0}$. We claim that this is a cd-filtration of $M_{\mathfrak p}$. First, as $$ \operatorname{cd}({\mathfrak a},M)=\max\{\operatorname{cd}({\mathfrak a},M_{c-1}),\operatorname{cd}({\mathfrak a},M/M_{c-1})\}, $$ and $$ \operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},M_{{\mathfrak p}})=\max\{\operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},(M_{c-1})_{{\mathfrak p}}),\operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},(M/M_{c-1})_{{\mathfrak p}})\}, $$ and $M/M_{c-1}$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$, we have $\operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},(M/M_{c-1})_{{\mathfrak p}})=\operatorname{cd}({\mathfrak a},M/M_{c-1})=c$ and $\operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},(M_{c-1})_{{\mathfrak p}})\leq \operatorname{cd}({\mathfrak a},M)=c$. Thus $\operatorname{cd}({\mathfrak a} R_{{\mathfrak p}},M_{{\mathfrak p}})=c$. Further, $M_{i}/M_{i-1}\otimes_{R} R_{{\mathfrak p}}$ is either zero or a relative Cohen-Macaulay $R_{{\mathfrak p}}$-module w.r.t ${\mathfrak a} R_{{\mathfrak p}}$ of cohomological dimension $i$ for all $1\leq i\leq c$. Therefore in view of Proposition~\ref{prop1}, the claim is proved. \end{proof} \medskip We can also show that passage to completion preserves the property of RCMF, as illustrated below. \begin{prop} Let $M$ be a finite RCMF $R$-module w.r.t ${\mathfrak a}$. Then $\widehat{M}$ is an RCMF $\widehat{R}$-module w.r.t ${\mathfrak a}\widehat{R}$, where $\widehat{}$ is the ${\mathfrak a}$-adic completion. \end{prop} \begin{proof} Let $\{M_{i}\}^{c}_{i=0}$ denote the relative Cohen-Macaulay filtration of $M$, where $c=\operatorname{cd}({\mathfrak a},M)$. Then $\{\widehat{M_{i}}\}^{c}_{i=0}$ is a relative Cohen-Macaulay filtration of the $\widehat{R}$-module $\widehat{M}$ w.r.t ${\mathfrak a}\widehat{R}$ by Flate Base Change Theorem \cite[Theorem 4.3.2]{BSH}. Thus by Proposition~\ref{prop1}, $\widehat{M}$ is an RCMF module over $\widehat{R}$ w.r.t ${\mathfrak a}\widehat{R}$. \end{proof} \medskip The other main result of this section is a characterization of the cd-filtration of $M$ in terms of associated prime ideals of its factors. For all $i$, set $\operatorname{Ass}^{i}_R(M)=\{{\mathfrak p}\in\operatorname{Ass}_R(M)\mid \operatorname{cd}({\mathfrak a},R/{\mathfrak p})=i\}$. \begin{thm}\label{Th4} Let $\mathcal{M}=\{M_i\}_{i=0}^{c}$ be a filtration of the finite $R$-module $M$ and $\operatorname{cd}({\mathfrak a},M_{0})=0$. The following conditions are equivalent: \begin{itemize} \item[(i)] $\operatorname{Ass}_R(M_{i}/M_{i-1})=\operatorname{Ass}^{i}_R(M)$ for all $1\leq i\leq c$; \item[(ii)] $\mathcal{M}$ is the cd-filtration of $M$. \end{itemize} \end{thm} \begin{proof} By virtue of \cite[Proposition 2.6 (iii)]{ASN}, we only have to prove the implication $(i)\Rightarrow (ii)$. First, we claim that $$ \operatorname{Ass}_R(M_{i-1})\cap\operatorname{Ass}_R(M_{i}/M_{i-1})=\emptyset \ \text{for all}\ 1\leq i\leq c. $$ Contrarily, assume that for some $1\leq i\leq c$, there is ${\mathfrak p}\in\operatorname{Ass}_R(M_{i-1})\cap \operatorname{Ass}_R(M_{i}/M_{i-1})$. Then $\operatorname{cd}({\mathfrak a},M_{i-1})\geq i$ by (i). If $i>1$, by hypothesis, $\operatorname{Ass}_R(M_{i-1}/M_{i-2})=\operatorname{Ass}^{i-1}_R(M)$ and so ${\mathfrak p}\notin \operatorname{Ass}_R(M_{i-1}/M_{i-2})$. Thus, considering the exact sequence $0\longrightarrow M_{i-2}\longrightarrow M_{i-1}\longrightarrow M_{i-1}/M_{i-2}\longrightarrow 0$, we have ${\mathfrak p}\in \operatorname{Ass}_R(M_{i-2})$. As $\operatorname{cd}({\mathfrak a},R/ {\mathfrak p})=i$, we have $\operatorname{cd}({\mathfrak a},M_{i-2})\geq i$. By repeating this descending process, $$ \operatorname{cd}({\mathfrak a},M_{0}), \operatorname{cd}({\mathfrak a},M_{1}), \ldots , \operatorname{cd}({\mathfrak a},M_{i-2}), \operatorname{cd}({\mathfrak a},M_{i-1}) $$ are all not less than $i$. Hence $\operatorname{cd}({\mathfrak a},M_{0})\geq i>0$ which is a contradiction. Now, consider the exact sequence $0\longrightarrow M_{c-1}\longrightarrow M\longrightarrow M/M_{c-1}\longrightarrow 0$, we have $\operatorname{cd}({\mathfrak a},M_{c-1})\leq c-1$ as $\operatorname{Ass}_R(M_{c-1})\cap\operatorname{Ass}_R(M/M_{c-1})=\emptyset$. Now, let $N$ be the largest submodule of $M$ such that $\operatorname{cd}({\mathfrak a},N)\leq c-1$ and ${\mathfrak p}\in\operatorname{Ass}_R(N/M_{c-1})$. Since $\operatorname{Ass}_R(N/M_{c-1})\subseteq\operatorname{Ass}_R^{c}(M)$, we have $\operatorname{cd}({\mathfrak a},R/{\mathfrak p})=c$. But ${\mathfrak p}\in\operatorname{Supp}_R(N/M_{c-1})\subseteq\operatorname{Supp}_R(N)$ and so $\operatorname{cd}({\mathfrak a},R/{\mathfrak p})\leq\operatorname{cd}({\mathfrak a},N)\leq c-1$ which is impossible. Therefore $\operatorname{Ass}_R(N/M_{c-1})=\emptyset$ and $M_{c-1}$ is the largest submodule of $M$ such that $\operatorname{cd}({\mathfrak a},M_{c-1})\leq c-1$. Now, descendingly, we proceed this method to prove that $\mathcal{M}$ is the cd-filtration of $M$. \end{proof} We end this section by a consequence of the above theorem which gives us a cd-filtration for certain modules. \begin{cor} \label{c5} (compare \cite[Proposition 2.3]{ASN}) Let $\cap^{n}_{i=1}N_{i}$ be a reduced primary decomposition of $(0)$ in $M$, where $N_{i}$ is ${\mathfrak p}_{i}$-primary, and $M_{i}=\cap_{\operatorname{cd}({\mathfrak a},R/{\mathfrak p}_{j})>i} N_{j}$ for all $0\leq i\leq c=\operatorname{cd}({\mathfrak a},M)$. Assume that $\operatorname{cd}({\mathfrak a},\cap_{\operatorname{cd}({\mathfrak a},R/{\mathfrak p}_{j})>0} N_{j})=0$. Then $\{M_{i}\}^{c}_{i=0}$ is the cd-filtration of $M$. \end{cor} \begin{proof} Let $1\leq i\leq c$. It is easy to see that $M_{i-1}=M_i \cap L_i$, where $L_i$ is the intersection of all $N_j$'s such that $\operatorname{cd}({\mathfrak a},R/{\mathfrak p}_j)=i$. By rewriting the indices, let $L_i=N_1 \cap\ldots\cap N_m$. By Theorem~\ref{Th4}, we need to show that $\operatorname{Ass}_R(M_i/M_{i-1})=\{{\mathfrak p}_1,\ldots,{\mathfrak p}_m\}$. First, we note that $\operatorname{Ass}_R(M_i/M_{i-1})=\operatorname{Ass}_R(M_i+L_i/L_i)\subseteq\operatorname{Ass}_R(M/L_i)$. Also, $\operatorname{Ass}_R(M/L_i)=\operatorname{Ass}_R(\oplus_{j=1}^{n}M/N_j)=\{{\mathfrak p}_1,\ldots,{\mathfrak p}_m\}$. Thus, it is enough to show that $\{{\mathfrak p}_1,\ldots,{\mathfrak p}_m\}\subseteq\operatorname{Ass}_R(M_i/M_{i-1})$. We have $M_{i-1}=L_i\cap L_{i+1}\cap\ldots\cap L_c$ and $M_i=L_{i+1}\cap L_{i+2}\cap\ldots\cap L_c$. Let $1\leq r\leq m$. We show that ${{\mathfrak p}}_r\in\operatorname{Ass}_R(M_i/M_{i-1})$. As $(0)=\cap^{n}_{j=1}N_{j}$, is a reduced primary decomposition, it deduces that \begin{equation}\label{ee} M_{i-1}\subsetneqq(N_1\cap\ldots\cap\widehat{N_r}\cap\ldots\cap N_m)\cap L_{i+1}\cap\ldots\cap L_c. \end{equation} For convenience, we denote the right side of (1) by $A$ for the rest. So there exists $x\in A$ such that $x\notin M_{i-1}$. Notice that $(M_{i-1}:x)=(N_r:x)$. Now, as $N_r$ is ${{\mathfrak p}}_r$-primary, there exists $t>0$ such that ${{{\mathfrak p}}_r}^t M\subseteq N_r$. Hence ${{{\mathfrak p}}_r}^t x\subseteq M_{i-1}$. Suppose that $s\geq 0$ is the least integer such that ${{\mathfrak p}_r}^{s+1}x\subseteq M_{i-1}$ and ${{\mathfrak p}_r}^s\nsubseteq M_{i-1}$. This implies that there exists $y\in{{\mathfrak p}_r}^s x$ such that $y\notin M_{i-1}$. Now, it is easy to see that ${\mathfrak p}_r=(M_{i-1}:y)$, i.e., ${\mathfrak p}_r\in\operatorname{Ass}_R(M_i/M_{i-1})$. This completes the proof. \end{proof} \section{Relative Cohen-Macaulayness in rings and modules} In this section, we prove some results concerning relative Cohen-Macaulay rings and modules. We begin by determining two classes of modules for which relative Cohen-Macaulayness is equivalent for rings and modules. \begin{dfn}(see \cite{ES}) An $R$-module $M$ is called {\it multiplication} if for every submodule $N$ of $M$ there exists an ideal ${\mathfrak a}$ of $R$ such that $N={\mathfrak a} M$. Moreover, if $\operatorname{Ann}_R M=0$, then $M$ is called faithful multiplication. \end{dfn} \medskip \begin{cor}\label{c1} Let $M$ be a faithful multiplication $R$-module. Then $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $R$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \end{cor} \begin{proof} Using \cite[Theorem 2.2 (b)]{AFM}, we have $\operatorname{Supp}(M)=\operatorname{Spec}(R)$. Thus $\operatorname{cd}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},R)$ by \cite[Theorem 2.2]{DNT}. On the other hand, $\operatorname{grade}({\mathfrak a},M)=\operatorname{grade}({\mathfrak a},R)$ by \cite[Theorem 2.2 (a)]{AFM}. Therefore $\operatorname{grade}({\mathfrak a},M)=\operatorname{cd}({\mathfrak a},M)$ if and only if $\operatorname{grade}({\mathfrak a},R)=\operatorname{cd}({\mathfrak a},R)$ as required. \end{proof} \medskip \begin{dfn}(see \cite{F} and \cite{V2}) The $R$-module $M$ is {\it semidualizing} if it satisfies the following: \begin{itemize} \item[(i)] $M$ is finitely generated, \item[(ii)] The homotopy map ${\chi}^{R}_{M}: R\longrightarrow \operatorname{Hom}_{R}(M,M)$, defined by $r\mapsto [s\mapsto rs]$, is an isomorphism, and \item[(iii)] $\operatorname{Ext}^{i}_{R}(M,M)=0$ for all $ i>0.$ \end{itemize} \end{dfn} \medskip \begin{cor}\label{cc} Let $M$ be a semidualizing $R$-module. Then $M$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $R$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \end{cor} \begin{proof}The assertion easily follows by \cite[Proposition 2.1.16]{W} and \cite[Theorem 2.2.6]{W}. \end{proof} \medskip Comparing with \cite[Proposition 5.1]{M}, we prove the following result on canonical modules by getting benefit from the concept of semidualizing modules. \begin{prop}\label{prop31} Let $(R,{\mathfrak m})$ be a local ring and $w_{R}$ be its canonical module. Then $R$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $w_{R}$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$. \end{prop} \begin{proof} As the canonical modules are semidualizing, we have $\operatorname{grade}({\mathfrak a},w_{R})=\operatorname{grade}({\mathfrak a},R)$ by \cite[Theorem 2.2.6]{W}. On the other hand, $\operatorname{Supp}_{R}(w_{R})=\operatorname{Spec}(R)$ by \cite[Proposition 2.1.16]{W} and so $\operatorname{cd}({\mathfrak a},w_{R})=\operatorname{cd}({\mathfrak a},R)$ by \cite[Theorem 2.2]{DNT}. Therefore $\operatorname{grade}({\mathfrak a},R)=\operatorname{cd}({\mathfrak a},R)$ if and only if $\operatorname{grade}({\mathfrak a},w_{R})=\operatorname{cd}({\mathfrak a},w_{R})$, as required. \end{proof} \medskip Next result is about the cohomological deficiency modules of a relative Cohen-Macaulay ring. \begin{prop}\label{prop6} Let $(R,{\mathfrak m})$ be a relative Cohen-Macaulay local ring w.r.t ${\mathfrak a}$ with $\operatorname{cd}({\mathfrak a},R)=c$. Then for all $0\leq i\leq c$, the $R$-modules $K^{i}_{{\mathfrak a}}(R)$ are either zero or $i$-cohomological dimensional relative Cohen-Macaulay modules w.r.t ${\mathfrak a}$. \end{prop} \begin{proof} By assumption and Example~\ref{E1}(a), we deduce that the quotient ideal $I_{i}:={{\mathfrak a}_{i}}/{{\mathfrak a}_{i-1}}$ in the cd-filtration $\{{\mathfrak a}_i\}_{i=0}^{c}$ of $R$ is either zero or an $i$-cohomological dimensional relative Cohen-Macaulay ideal w.r.t ${\mathfrak a}$ for all $1\leq i\leq c$ . In view of Proposition~\ref{lemma}, it follows that $K^{i}_{{\mathfrak a}}(R)\cong K^{i}_{{\mathfrak a}}({I}_{i})$ for all $0\leq i\leq c$. Since $I_{i}$ is either zero or an $i$-cohomological dimensional relative Cohen-Macaulay ideal w.r.t ${\mathfrak a}$, we have $K^{i}_{{\mathfrak a}}(I_{i})$ is either zero or the cohomological canonical module of $I_{i}$. But the cohomological canonical module of $I_{i}$ is relative Cohen-Macaulay module w.r.t ${\mathfrak a}$ by \cite[Theorem 4.3 (i)]{Z}. Thus the assertion follows. \end{proof} \medskip At the end, we give some results about relative Cohen-Macaulay rings under some mild assumptions. \begin{prop}\label{ppp} Let $f:R\longrightarrow R'$ be a faithfully flat homomorphism of Noetherian rings. Then $R$ is relative Cohen-Macaulay w.r.t ${\mathfrak a}$ if and only if $R'$ is relative Cohen-Macauly ring w.r.t ${\mathfrak a} R'$. \end{prop} \begin{proof} As $H^{i}_{{\mathfrak a}}(R)\otimes_R R'\cong H^{i}_{{\mathfrak a} R'}(R')$ for all $i\geq 0$, we get $\operatorname{cd}({\mathfrak a},R)=\operatorname{cd}({\mathfrak a} R',R')$ and $\operatorname{grade}({\mathfrak a},R)=\operatorname{grade}({\mathfrak a} R',R')$. Therefore the assertion follows. \end{proof} \medskip \begin{cor} Let $f:R\longrightarrow R'$ be a faithfully flat homomorphism of Noetherian local rings and $R$ be a relative Cohen-Macaulay w.r.t ${\mathfrak a}$. Then $H^{i}_{{\mathfrak a}}(R')\neq 0$ if and only if $(0:_{R}H^{i}_{{\mathfrak a} R'}(R'))=0$. \end{cor} \begin{proof} Apply Proposition~\ref{ppp} and \cite[Theorem 3.3]{L}. \end{proof} \medskip \begin{exmp} Let ${\mathfrak a}\subseteq Jac(R)$. Then $\widehat{R}$, the ${\mathfrak a}$-adic completion of $R$, is relative Cohen-Macaulay ring w.r.t ${\mathfrak a}\widehat{R}$ if and only if $R$ is relative Cohen-Macauly ring w.r.t ${\mathfrak a}$. \end{exmp} \medskip \begin{exmp} The polynomial ring $R[x]$ is relative Cohen-Macaulay ring w.r.t ${\mathfrak a} R[x]$ if and only if $R$ is relative Cohen-Macaulay ring w.r.t ${\mathfrak a}$. \end{exmp} {\bf Acknowledgements.} The authors are grateful to the reviewer for suggesting several improvements to the manuscript. Moreover, the authors would like to express their thanks to Dr. Raheleh Jafari from Kharazmi University for her useful comments.
2f59f7879be9c80f1df4909ebccace08db257c28
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} To write high-quality program code for a Multi-Processor System-on-Chip (MPSoC), software developers must fully understand how their code will be executed on-chip. Debugging and tracing tools can help developers to gain this understanding. They are a keyhole through which developers can peek and observe the software execution. Today, and even more in the future, this keyhole narrows as MPSoCs integrate more functionalities, while at the same time the amount of software increases dramatically. Furthermore, concurrency and deep interaction of software with hardware components beyond the instruction set architecture (ISA) boundary are on the rise. Therefore more, not less, insight into the system is needed to keep up or even increase developer productivity. Many of today's MPSoCs are executing concurrent code on multiple cores, interact with the physical environment (cyber-physical systems), or must finish execution in a bounded amount of time (hard real-time). In these scenarios, a non-intrusive observation of the software execution is required, like it is provided by tracing. Instead of stopping the system for observation, as done in run-control debugging, the observed data is transferred off-chip for analysis. Unfortunately, observing a full system execution would generate data streams in the range of petabits per second~\cite[p.~16]{vermeulen_debugging_2014}. This is the most significant drawback of tracing: the system insight is limited by the off-chip bottleneck. Today's tracing systems, like ARM CoreSight~\cite{CoreSightWebsite} or NEXUS 5001~\cite{_nexus_2003} are designed to efficiently capture the operation of a functional unit (like a CPU) as compressed trace stream. With filters and triggers it is possible to configure which and when a functional unit is traced (observed). The trace streams (or short, traces) are then transported across an off-chip interface (and possibly other intermediate devices) to a host PC. Upon arrival the compressed trace streams are first decompressed (reconstructed) using the program binary and other static information which was removed before. The reconstructed trace streams are then fed to a data analysis application, which extracts information out of the data. This information can then be presented to a developer or it can be used by other tools, e.g. for runtime verification. \medskip The \textbf{main idea} in this work is to move the data analysis (at least partially) from the host PC into the chip. Bringing the computation closer to the data sources reduces the off-chip bandwidth requirements, and ultimately increases insight into software execution. To realize this idea, we introduce \emph{DiaSys}, a replacement for the tracing system in an MPSoC. DiaSys does not stream full execution traces off-chip for analysis. Instead, it first creates events from observations on the chip. Events can signal any interesting state change of the observed system, like the execution of a function in the program code, a change in interconnect load beyond a threshold, or the read of a data word from a certain memory address. A \emph{diagnosis application} then processes the observed events to give them ``meaning.'' Given an appropriate diagnosis application, a software developer might not be presented with any more a sequence of events like ``a read/write request was issued'', but with the more meaningful output of the diagnosis application ``a race condition bug was observed.'' Analyzing the data on-chip is not only beneficial to reduce the off-chip bandwidth requirements, but also enables new use cases in the future, such as self-adapting or self-healing systems. However, doing all this processing on-chip would, in some cases (and markets), be too costly in terms of chip area. Therefore, we describe the diagnosis applications so that they can be transparently split into multiple parts: one part executing on-chip in dedicated, distributed hardware components close to the data source, and another part running on a host PC. In summary, our \textbf{key contributions} are: \begin{itemize} \item an architecture and component library of on-chip infrastructure to collect and analyze diagnosis data created during the software execution, and \item a model of computation which allows developers to describe data analysis tasks (the ``diagnosis application'') in a way which is independent of the specific hardware implementation of the diagnosis system. \end{itemize} Combining these two contributions, we show that \begin{itemize} \item DiaSys is a viable alternative to tracing systems in the two major fields where tracing is employed today: hypothesis testing (debugging) and the collection of runtime statistics. Two case studies explore these use cases (Section~\ref{sec:usage}). \item the diagnosis applications introduced by DiaSys are a beneficial representation of a data analysis task: they abstract from the implementation through a clearly defined model of computation to foster re-use and portability (Section~\ref{sec:method:diagnosis_applications}). \item DiaSys is implementable in hardware with reasonable system cost (Section~\ref{sec:hwimpl}). \end{itemize} In the following, we explore our diagnosis system in depth. We start with a thorough analysis of the state of the art in tracing systems in Section~\ref{sec:related_work}, based on which we developed our concept of the diagnosis system presented in Section~\ref{sec:method}. We include a detailed model of a diagnosis application and a discussion of its semantics in Section~\ref{sec:method:diagnosis_applications}. The architecture of our diagnosis system is presented next in Section~\ref{sec:arch}, followed by a discussion of its possible limitations. In Section~\ref{sec:hwimpl} we present our hardware implementation. Combining all parts, we show two usage examples in Section~\ref{sec:usage}, one to find a multi-core race condition bug, and one to create an application profile. \paragraph{A word on terminology} We use the terms ``diagnosis'' and ``diagnosis system'' to stress the fact that we integrate the on-chip observation of the software execution with the data analysis. We use ``tracing'' to refer to the method of obtaining insight into the SoC by transferring a stream of observations from a functional unit off-chip. ``Debugging'' is used as a synonym for ``hypothesis testing,'' the process of (usually manually) checking (by various means) if the software behaves as expected. \section{Background and Related Work} \label{sec:related_work} Our approach touches and integrates two usually separated topics: obtaining a software execution trace from a SoC, and processing the obtained information in order to generate useful information. In this section we present background and related work on both topics. \subsection{Gaining Insight into SoCs} \label{sec:related_work:insight} \begin{figure} \centering \includesvgnolatex{img/diagnosis_applications_existing} \caption{A schematic view of a common tracing system like ARM CoreSight or NEXUS 5001.} \label{fig:diagnosis_applications_existing} \end{figure} Today's tracing solutions for SoCs are designed to capture and transfer as much as possible of the SoC's internal state to an external observer. They are generally structured as shown in Figure~\ref{fig:diagnosis_applications_existing}. First, trace data streams are obtained from the observation of various functional units in the system, like CPUs, buses and memories. Then, this data is spatially and temporally reduced through the use of filters and triggers. Finally, the redundancy in the data is removed by the use of compression algorithms. The resulting trace data stream is then transferred off-chip (live or delayed through an on-chip memory buffer). On a host PC, the original trace streams are reconstructed using information from the program binary and other static information, which was discarded as part of the compression process. All major commercial SoC vendors offer tracing solutions based on this template. ARM provides its licensees the CoreSight intellectual property (IP) blocks~\cite{CoreSightWebsite}. They are used in SoCs from Texas Instruments, Samsung and STMicroelectronics, among others. Vendors such as Qualcomm (formerly Freescale) include tracing solutions based on the IEEE-ISTO~5001 (Nexus) standard~\cite{_nexus_2003}, while Infineon integrates the Multi-Core Debug Solution (MCDS) into its automotive microcontrollers~\cite{ipextreme_infineon_2008}. Since 2015 Intel also includes a tracing solution in their desktop, server and embedded processors called Intel Processor Trace (PT)~\cite{_intel_2015}. The main differentiator between the solutions is the configurability of the filter and trigger blocks. Driven by the off-chip bottleneck, a major research focus are lossless trace compression schemes. Program trace compression available in commercial solutions typically requires 1 to 4~bit per executed instruction~\cite{hopkins_debug_2006, orme_debug_2008}, while solutions proposed in academia claim compression ratios down to $0.036$~bit per instruction~\cite{uzelac_real-time_2010}. Even though data traces contain in general no redundancy, in practice compression rates of about 4:1 have been achieved \cite{hopkins_debug_2006}. \subsection{Analyzing System Behavior} \label{sec:related_work:analysis} A human is easily overwhelmed when asked to analyze multiple gigabits of trace data each second. Instead, automated analysis tools are used to extract useful information out of the vast amount of trace data. Such tools have one common goal: to help a developer better understand the software execution on the target system. The means to achieve this goal, however, vary widely. Diagnosis applications which analyze non-functional issues such as performance bugs often generate results in the form of ordered lists. They list for example applications which consume most processing or memory resources, or which generate most I/O traffic. This report can then be a starting point for a more fine-grained analysis of the problem. Diagnosis applications which target functional bugs are usually more specialized; in many cases, a diagnosis application is created just to confirm or negate one single hypothesis about the software execution on the chip. For example, a developer might want to confirm that a certain variable stays within defined bounds, e.g. to check if an array overflow occurred. Most analysis tools for SoCs are not stand-alone applications, but part of debugging and tracing software packages from vendors like Lauterbach, Green Hills or ARM. They are usually controlled through a graphical user interface. Of course, analysis applications used to understand software execution are not only developed for SoCs and other embedded systems. Most tools in this domains are intrusive: they run as part of the analyzed system and obtain the required system state through instrumentation. However, the general concepts are also relevant for the diagnosis of SoCs. This is especially true for scriptable or programmable debugging, which applies the concept of event-driven programming to debugging. Whenever a defined \emph{probe point} is hit, an event is triggered and an \emph{event handler} executes. Common probe points are the execution of a specific part of the program (like entering a certain program function), or the access to a given memory location. The best-known current implementations of this concept are DTrace and SystemTap, which run on, or are part of, BSDs, Linux, and macOS (where DTrace is integrated into the ``Apple Instruments'' product) \cite{cantrill_dynamic_2004,eigler_architecture_2005}. The concept, however, is much older. Dalek~\cite{olsson_dataflow_1991} is built on top of the GNU Debugger (GDB) and uses a dataflow approach to combine events and generate higher-level events out of primitive events. Marceau et al. extend the dataflow approach and apply it to the debugging of Java applications~\cite{marceau_dataflow_2004}. Coca~\cite{ducasse_coca_1999}, on the other hand, uses a language based on Prolog to define conditional breakpoints as a sequence of events described through predicates for debugging C programs. In a work targeting early multi-processor systems, but otherwise closely related to our approach, Lumpp et. al. present a debugging system which is based on an event/action model~\cite{lumpp_specification_1990}. A specification language is used to describe events in the system trigger which an action, and hardware units can be used to identify these events. None of the presented works directly tackle the observability problem in SoCs by moving the data analysis partially on-chip. However, they form a strong foundation of ideas, which inspired us in the design of the diagnosis system. It is presented in the following sections. \section{DiaSys, Our Diagnosis System} \label{sec:method} We have designed our diagnosis system to address the shortcomings of today's tracing systems. Based on a set of requirements, we discuss the design of the diagnosis system in depth, followed by a hardware/software architecture implementing the diagnosis system. First, however, we define some terms used in the following discussion. \subsection{Definitions} \paragraph{Observed system} The part of the SoC which is observed or monitored by the diagnosis system. In other works, the term ``target system'' is used. \paragraph{Functional unit} A subset of the observed system which forms a logical unit to provide a certain functionality. Examples for functional units are CPUs, memories, or interconnect resources such as a bus or NoC routers. \paragraph{State} The state of a system is the unity of all stored information in that system at a given point in time which is necessary to explain its future behavior.~\cite[p.~103]{harris_digital_2012} In a sequential circuit, the state is equal to the memory contents of the system. \subsection{Design Requirements for the Diagnosis System} \label{sec:method:requirements} A set of requirements guides the design of the diagnosis system. \paragraph{Distributed} The diagnosis system must be able to reduce the amount of observation data as close to the source, i.e. the functional units, as possible. Since the data sources are distributed across the chip, the diagnosis system must also be distributed appropriately. \paragraph{Non-Intrusive} The diagnosis system must be non-intrusive (passive). Non-intrusive observation preserves the event ordering and temporal relationships in concurrent executions, a requirement for debugging multi-core, real-time, or cyber-physical systems~\cite{fidge_fundamentals_1996}. Non-intrusiveness also gives a developer the confidence that he or she is observing a bug in the program code, not chasing a problem caused by the observation (a phenomenon often called ``Heisenbug''~\cite{gray_why_1986}). \paragraph{Flexible On-Chip/Off-Chip Cost Split} The diagnosis system must be flexible to implement. The implementation of the diagnosis system involves a trade-off between the provided level of observability and the system cost. The two main cost contributions are the off-chip interface and the chip area spent on diagnosis extensions. The diagnosis system concept must be flexible enough to give the chip designer the freedom to configure the amount of chip resources, the off-chip bandwidth and the pin count in a way that fits the chip's target market. At the same time, to provide flexibility on the observation, the system must be able to adapt to a wide range of bugs. \paragraph{Relaxed Timing Constraints} The diagnosis system must not assume a defined timing relationship between the individual distributed components. Today's larger SoCs are designed as globally asynchronous, locally synchronous (GALS) systems with different power and clock domains, where no fixed time relationship between components can be given. \subsection{The Concept of the Diagnosis System} \label{sec:method:concept} \begin{figure} \centering \includesvgnolatex{img/diasys_model} \caption{A schematic view of the diagnosis system.} \label{fig:diasys_model} \end{figure} Based on the discussed requirements this section gives an overview on the diagnosis system as shown in Figure~\ref{fig:diasys_model}. The \textit{input} to the diagnosis system is the state of the observed system over time, the \textit{output} are the diagnosis results, which can be represented in various forms. With respect to the input and output interfaces, the diagnosis system is identical to a traditional tracing system. The difference lies in the components which generate the output from the input. Three main components are responsible for this processing: the event generators, the diagnosis application together with its execution platform, and the event sinks. Between these components, data is exchanged as diagnosis events. \textit{Diagnosis events are the container for data} exchanged in the diagnosis system. In the general case, an event consists of a type identifier and a payload. Events are self-contained, i.e. they can be decoded without the help from previous or subsequent events. \textit{Event generators produce} primary events based on state changes in the observed system. They continuously compare the state of the observed system with a \emph{trigger condition}. If this condition holds, they \emph{trigger} the generation of a primary event. A \emph{primary event} is a specialized diagnosis event in which the type identifier is equal to a unique identifier describing the event generator. The payload contains a partial snapshot of the state of the observed system at the same instant in time as the event was triggered. Which parts of the state are attached to the event is specified by the event generator. For example, a CPU event generator might produce primary events when it observes a function call and attach the current value of a CPU register as payload. A primary event answers two questions: why was the event generated, and in which state was the observed system at this moment in time. \textit{The diagnosis application analyzes} the software execution on the observed system. It is modeled as transformational dataflow application, which transforms primary events into output events. The goal of this transformation is to interpret the state changes represented in primary events in a way that yields useful information for a developer or an automated tool. We describe diagnosis applications in more detail in Section~\ref{sec:method:diagnosis_applications}. \textit{The diagnosis application execution platform} executes diagnosis applications. The execution platform can span (transparent to the diagnosis application developer) across the chip boundary. On the chip, it consists of specialized hardware blocks which are able to execute (parts of) the diagnosis application. On the host PC, a software runtime environment enables execution of the remaining parts of the diagnosis application. The on- and off-chip part of the execution platform are connected through the off-chip interface. This split design of the execution platform allows hardware designers to trade off chip area with the bandwidth provided for the off-chip interface, while retaining the same level of processing power, and in consequence, system observability. \textit{Event sinks consume} output events produced by the diagnosis application. Their purpose is to present the data either to a human user in a suitable form (e.g. as a simple log of events, or as visualization), or to format the events in a way that makes them suitable for consumption by an automated tool, or possibly even for usage by an on-chip component. An example usage scenario for an automated off-chip user is runtime validation, in which data collected during the runtime of the program is used to verify properties of the software. Together, event generators, the diagnosis application and the event sink build a processing chain which provides a powerful way to distill information out of observations in the SoC. \subsection{Diagnosis Applications} \label{sec:method:diagnosis_applications} Diagnosis applications are the heart of the diagnosis system, as they perform the ``actual work'' of interpreting what happens on the observed system during the software execution. Diagnosis applications are transformational dataflow applications. We chose this model to enable the transparent mapping of the diagnosis application to an execution platform spanning across the chip boundary. Our goal is that the developer of the diagnosis application does not need to explicitly partition the diagnosis application into an on-chip and an off-chip part; instead, this mapping could be performed in an automated way. (Currently, however, we do not perform an automated mapping.) No matter how the diagnosis application is mapped onto the execution platform, the behavior of the application follows identical rules, i.e. the semantics of the application stay the same. The diagnosis application is a \emph{transformational application}, in contrast to reactive or interactive applications~\cite{halbwachs_synchronous_1991}. This means, starting from a given set of inputs, the application \emph{eventually} produces an output. The application code only describes the functional relationship between the input and the output, not the timing when the output is generated. The application also does not influence or interact in another way with the observed system from which its inputs are derived. The diagnosis application is structured as \emph{dataflow application}. Its computation is represented by a directed graph, in which the nodes model the computation, and the edges model communication links. In diagnosis applications we call the graph nodes \emph{transformation actors}, and the graph edges \emph{channels}. \begin{figure} \centering \includesvgnolatex{img/transformation_actor} \caption{A transformation actor with $n=3$ input channels and $m = 2$ output channels.} \label{fig:transformation_actor} \end{figure} Each transformation actor reads events from $n \in \mathbb{N}_0$ input channels, and writes events to $m \in \mathbb{N}_0$ output channels, as shown in Figure~\ref{fig:transformation_actor}. A transformation actor starts its processing, it ``fires,'' if a \emph{sufficient} number of events are available at its inputs. The definition of ``sufficient'' depends on the individual transformation actor. For example, one transformation actor might always read one event from each input before starting the processing, while another one might always read two events from input 1 and one event from input 2. When firing, the transformation actor applies an arbitrary transformation function $f$ to the input events. The generated output depends on \begin{itemize} \item the read input events, \item the ordering of the input events, and \item the internal state of the transformation actor. \end{itemize} Transformation actors may communicate only through the input and output channels, but not through additional side channels (e.g. shared variables). Diagnosis applications built out of such transformation actors are \emph{nondeterministic}, as defined by Kahn~\cite{kahn_semantics_1974,lee_dataflow_1995}. This means, the output not only depends on the history of inputs (i.e. the current input and the state of the actor), but also on the relative timing (the ordering) of events. Nondeterministic behavior of diagnosis applications is, in most cases, the expected and wanted behavior; it gives its authors much needed flexibility. An example of nondeterministic diagnosis applications are applications which aggregate data over time, like the lock contention profiling presented in Section~\ref{sec:usage:lockprofiling}. These applications consume an unspecified amount of input events and store an aggregate of these inputs. After a certain amount of time, they send a summary of the observations to an output channel. But at the same time, nondeterministic diagnosis applications prevent the static analysis of event rates, bandwidth and processing requirements. If wanted, application authors can therefore create deterministic diagnosis applications, if they restrict themselves to \begin{itemize} \item always reading the input channels in the same order without testing for data availability first (instead, block and wait until the data arrives), \item connecting one channel to exactly one input and one output of an actor, and \item using only transformation functions which are deterministic themselves. \end{itemize} Note that we only describe the diagnosis application itself as nondeterministic. Its execution, after being mapped to an execution platform, can be deterministic, i.e. multiple identical runs produce the same diagnosis result. \subsection{Diagnosis System Architecture} \label{sec:arch} \begin{figure} \includesvgnolatex{img/diasys_hwarch} \caption{The architecture of our diagnosis system implementation. Event generators (EG) observe the functional units (FU). They connect via an interconnect to different processing nodes (PN) and an off-chip interface. On a host PC processing nodes implemented in software further process the events. Finally, event sinks prepare the data for developers or automation. A diagnosis application can be mapped flexibly on the execution platform; a sample mapping is shown.} \label{fig:diasys_hwarch} \end{figure} In the previous sections we presented the diagnosis system from a functional perspective. We continue now by presenting an implementation architecture of the diagnosis system. It consists of extensions to the SoC, as well as software on a host PC; an exemplary architecture is shown in Figure~\ref{fig:diasys_hwarch}. In the SoC, different functional units (FU) can be observed, like a CPU, a memory, or a bus. Each functional unit can be attached to one or multiple event generators (EG). The resulting events are transmitted over a diagnosis interconnect to on-chip processing nodes. The processing nodes form the execution platform for the diagnosis application; they will be discussed in depth in Section~\ref{sec:arch:execution_platform}. Invisible to the diagnosis application, the execution platform spans across the chip boundary. Through an I/O interface, a host PC is connected to the SoC. The PC contains a software runtime environment to provide further processing nodes as part of the diagnosis application execution platform. If all processing has been accomplished, the output events are sent to an event sink application on the host PC, which formats the output events for developers or automation. Depending on the features and computational power provided by the processing nodes, a diagnosis application can be mapped to the execution platform in a flexible way. \subsection{Diagnosis Application Execution Platform} \label{sec:arch:execution_platform} The heart of the diagnosis system are the diagnosis applications, which are executed on the diagnosis application execution platform. As discussed before, this platform spans across the SoC and the host PC. On the SoC, it consists of processing nodes of different types which are connected by a shared interconnect (such as a NoC or a bus). Each processing node has an input and output interface to receive and send out events on the interconnect. Different types of processing nodes can offer different degrees of flexibility regarding their computation. Some might be able to perform only a single functionality specified at hardware design time, like a counter or a statistical aggregator, while others might be freely programmable. As an example, we present in Section~\ref{sec:arch:diagnosis_processor} the Diagnosis Processor, a programmable general-purpose processing node. As the chip area (economically) available for on-chip diagnosis processing is limited, the diagnosis application execution platform extends to the host PC. Connected through an arbitrary off-chip interface, a runtime layer in software provides a virtually unlimited number of ``soft'' processing nodes. Such PNs are implemented in software on the host PC and accept, like their on-chip counterpart, events as input and produce events as output. By being executed on a host PC, they provide more compute and memory resources. \medskip The transformation or computation in a diagnosis application is represented by transformation actors. For execution, they are mapped to the available processing nodes, as shown exemplary in Figure~\ref{fig:diasys_hwarch}. An $n$:1 mapping of transformation actors to processing nodes is possible, if the combined transformation of all $n$ transformation actors can be executed by the processing node. To achieve the greatest possible reduction in off-chip traffic, as much computation as possible should be mapped to on-chip processing nodes. The remainder of processing is then mapped to processing nodes on a host PC, where significantly more processing power is available. \subsubsection{The Diagnosis Processor: A Multi-Purpose Processing Node} \label{sec:arch:diagnosis_processor} The diagnosis processor is a freely programmable general-purpose processing node. Like any processor design, it sacrifices computational density for flexibility. Its design is inspired by existing scriptable debugging solutions, like SystemTap or DTrace, which have shown to provide a very useful tool for software developers in a growingly complex execution environment. The usage scenario for this processing node are custom or one-off data analysis tasks. This scenario is very common when searching for a bug in software. First, a hypothesis is formed by the developer why a problem might have occurred. Then, this hypothesis must be validated in the running system. For this validation, a custom data analysis script must be written, which is highly specific to the problem (or the system state is manually inspected). This process is repeated multiple times, until the root cause of the problem is found. As this process is approached differently by every developer (and often also influenced by experience and luck), a very flexible processing node is required. We present the hardware design of our diagnosis processor implementation in Section~\ref{sec:hwimpl:diagnosis_processor}. We envision the programming of the diagnosis processor being done through scripts similar to the ones used by SystemTap or DTrace. They allow to write trace analysis tasks on a similar level of abstraction as the analyzed software itself, leading to good developer productivity. \subsection{Discussion} The presented diagnosis system is designed to fulfill the requirements outlined in Section~\ref{sec:method:requirements}. In the following, we discuss the consequences of the design decisions, which can limit the applicability of the diagnosis system approach in some cases. By transforming the observed system state close to the source into denser information, the off-chip bottleneck can be circumvented. As a downside, this lossy transformation thwarts a usage scenario of today's tracing systems. In many of these systems, it is possible to capture a trace once, store it, and run different analysis tasks on it. If major parts of the captured data are dismissed early, this is not possible any more. Instead, the analysis task must be defined (as diagnosis application) before the system is run. If the problem hypothesis changes and a different diagnosis application is required, the system must be run again. The severity of this limitation strongly depends on how hard it is to reproduce a bug or behavior across runs. Another feature present in many of today's tracing systems, which is explicitly not supported by the diagnosis system, are cross-triggers. Cross-triggers are a mechanism in the tracing system to start or stop the observation, or to observe different components, based on another observation in the system. For example, memory accesses could be traced only after a CPU executed a certain program counter. Cross-triggers are most useful if their timing behavior is predictable. For example, memory accesses are traced ``in the next cycle'' after the specified program counter was executed. In GALS SoCs, such timing guarantees cannot be given; for a diagnosis application spanning across a SoC and a host PC, it is equally impossible to give (reasonably low bounded) timing guarantees. We make this property explicit by modeling the diagnosis system as a transformational system, not a reactive system. The commercially available tracing systems today are less specific about this. For example, ARM CoreSight uses a handshaking protocol for cross-triggers delivered across clock boundaries, which guarantees save delivery of the signal, but does not guarantee any latency. Instead of relying on cross-triggers to collect data from different sources at the same instant in time, we capture this data already when creating primary events through event generators. The payload of primary events is the only way to pass multiple state observations with a defined timing relation to the diagnosis system. For example, an event generator attached to a CPU can trigger an event based on a program counter value, and attach current contents of certain CPU registers or stack contents to it. Using this method, it is possible to generate for example an event which informs about a function being called, and which function arguments (stored in CPU registers or on the stack) have been passed to it. We show an example of such an event generator as part of our hardware implementation. Finally, we discuss the system behavior in overload situations, i.e. if more input data is received than the diagnosis system can process. Given the generally unknown input data, and the generally nondeterministic behavior of the diagnosis application, it is not possible to statically dimension the diagnosis system to be able to handle all possible input sequences. Therefore, overload situations are unavoidable in the general (and most common) case. If an overload situation is detected, the diagnosis system can react in multiple ways. First, it could temporarily stall the observed system. This gives the diagnosis system time to process outstanding events without new events being produced. This approach is only feasible in a synchronous non-realtime system. A more common approach is to discard incoming data until further processing resources are available. Depending on the diagnosis application, a recovery strategy needs to be formulated. Some applications can deal easily with incomplete input data, e.g. diagnosis applications creating statistics. Others are not able to work with an incomplete input sequence and in consequence fail to be executed properly. \medskip This ends the discussion of DiaSys in general. In the following, we present a hardware implementation of our approach, and then continue then with two usage examples how DiaSys can be put to work. \section{Implementing DiaSys in Hardware} \label{sec:hwimpl} \begin{figure} \centering \scriptsize \includesvgnolatex{img/prototype_sys} \caption{Block diagram of the prototype implementation. The diagnosis extensions added to the $2\times2$ multi-core system are drawn in blue.} \label{fig:prototype_sys} \end{figure} DiaSys, as presented in the previous section, can be implemented in various ways in hardware. Our implementation, which we present in the following, is one such implementation. It was created to answer two questions: first, to show that DiaSys can be implemented in hardware, and second, to give resource usage numbers for one specific implementation. As dimensioning and optimization for speed or area usage strongly depends on how DiaSys is used, a general answer to this question must remain out of scope for this work. \medskip The diagnosis system extends a $2 \times 2$ tiled multi-core system as shown in Figure~\ref{fig:prototype_sys}. Our implementation runs on an FPGA and uses the OpTiMSoC framework~\cite{wallentowitz_open_2013}. The observed system consists of four mor1kx CPU cores (an implementation of the OR1K or ``OpenRISC'' ISA), each connected to a distributed memory and a mesh NoC interconnect (components with white background). This system is representative of the multi- and many-core architecture template currently in research and available early products, such as the Intel SCC or the Mellanox (formerly Tilera and EZchip) Tile processors. The diagnosis system, depicted in blue, consists of the following components. \begin{itemize} \item Four event generators attached to the CPUs (marked ``EG''). \item A single diagnosis processor. \item A 16~bit wide, unidirectional ring NoC, the ``diagnosis NoC,'' to connect the components of the diagnosis system. It carries both the event packets as well as the configuration and control information for the event generators and processing nodes. \item A USB~2.0 off-chip interface. \item Software support on the host PC to control the diagnosis system, and to display the results. \end{itemize} All components connected to the diagnosis NoC follow a common template to increase reusability. Common parts are the NoC interface and a configuration module, which exposes readable and writable configuration registers over the NoC. In the following, we explain the implementation of the main components in detail. \subsection{CPU Event Generator} \label{sec:hwimpl:implementation:eventgen_cpu} \begin{figure} \centering \scriptsize \includesvgnolatex{img/eventgen_cpu} \caption{Block diagram of the CPU event generator.} \label{fig:eventgen_cpu} \end{figure} The CPU event generator is attached to a single CPU core. Its main functionality is implemented in the trigger and the state capture modules. The trigger unit of the CPU event generator triggers on two types of conditions: either the value of the program counter (PC), or the return from a function call (the jump back to the caller). At each point in time, 12 independent trigger conditions can be monitored. The number of monitored trigger conditions is proportional to the used hardware resources. Our dimensioning was determined by statistical analysis of large collections of SystemTap and DTrace scripts: $\leq 9$ concurrent probes are used in 95~\% of SystemTap scripts, and $\leq 12$ concurrent probes cover 92~\% of the DTrace scripts. The partial system state snapshot can be configured to contain the CPU register contents and the function arguments passed to the function. A block diagram of the CPU event generator is shown in Figure~\ref{fig:eventgen_cpu}. The PC trigger is implemented as simple comparator. The ``function return'' trigger requires a special implementation, because no unique point in the program flow, i.e. no single PC value, describes the return from a function (a function can have multiple call sites and can return to the caller from different points in the function body). Instead, we use the following method: \begin{enumerate} \item A PC trigger is set to the first instruction of the called function. \item If the trigger fires, the link (a.k.a. return) address is pushed to a memory structure inside the return monitor, the ``return address stack.'' The link address is the program counter to jump to if the function has finished its execution and the execution returns to the caller. On OR1K (as common on RISC architectures, including ARM and MIPS) the link address is stored in a CPU register. On other architectures and calling conventions (such as x86 and x86\_64), the link address is pushed to the stack. \item Now the system monitors the program flow for the topmost PC value in the return address stack. If this PC is executed, a function returned to its caller and the function return monitor triggers the generation of a primary event. \end{enumerate} To capture the values inside CPU registers, the register writeback signal of the mor1kx CPU is observed, and a copy of the register file is created. This copy can then be included in the event packet if a trigger fires. The passing of function arguments to functions depends on the calling convention. On OR1K, the first six data words passed to a function are available in CPU registers, all other arguments are pushed to the stack before calling the function. This is common for RISC architectures; other architectures and calling conventions might pass almost all arguments on the stack (such as x86). To record the function arguments as part of the primary event we therefore need to create a copy of the stack memory that can be accessed non-intrusively. We do this by observing CPU writes to the stack memory location. In our implementation for the mor1kx CPU we create a copy of the stack memory by monitoring the instruction stream for a store word (\texttt{l.sw}) instruction with a target address \texttt{rA} equal to the stack pointer \texttt{R1}. The data in the source register \texttt{(rB)}, together with a write offset \texttt{I} (with $\texttt{I} \geq 0$, i.e. targeting the previous stack frame) can be then used to recreate the stack frame.% \footnote{In theory, data can be written to the stack in a different way. However, the described way is common across compilers and used by the default GCC compiler for OR1K.} \subsection{Diagnosis Processor} \label{sec:hwimpl:diagnosis_processor} \begin{figure} \centering \includesvgnolatex{img/dbg_coprocessor} \caption{Block diagram of the diagnosis processor, a freely programmable processing node.} \label{fig:diagnosis_processor} \end{figure} The diagnosis processor design is extended from a standard processor template like it is used in the observed system. The main components, shown in Figure~\ref{fig:diagnosis_processor}, are a single mor1kx CPU core and an SRAM block as program and data memory. This system is extended with components to reduce the runtime overhead of processing event packets. First, the network adapter, which connects the CPU to the diagnosis NoC, directly stores the incoming event packets in the memory through a DMA engine. All event packets are processed in a run-to-completion fashion. We can therefore avoid interrupting the CPU and instead store the address of the event to be processed next in a hardware ``run queue.'' A ``discard queue'' signals the hardware scheduler which events have been processed and can be purged from memory. \subsection{Resource Usage} \begin{table} \centering \footnotesize \begin{tabulary}{\linewidth}{lLrrr} \toprule \multicolumn{2}{l}{\textbf{Module}} &\textbf{LUTS} & \textbf{REGS} & \textbf{RAMS}\\ \midrule \multicolumn{2}{l}{observed system} & 40625 & 29638 & 80 \\ & 1 compute tile (system contains 4)& $\sim 7232$ & $\sim$ 4763 & 20\\ & $2\times2$ mesh NoC & 10791 & 9964 & 0\\ & support infrastructure (DRAM if, clock/reset mgr) & 904 & 623 & 0\\ \midrule \multicolumn{2}{l}{diagnosis extensions} & 19556 & 19140 & 147\\ & 1 CPU Event Generator (fully featured)& 3603 & 6521 & 2 \\ & 1 CPU Event Generator (reduced CoreSight-like functionality) & 1365 & 1594 & 0 \\ & 1 Diagnosis Processor & 8614 & 4549 & 145\\ & diagnosis NoC & 2520 & 2926 & 0\\ \bottomrule \end{tabulary} \medskip \caption{The resource usage of a CPU diagnosis unit. Either the fully-featured or the reduced-functionality CPU Event Generator can be used.} \label{tab:resourceusage} \end{table} The prototype of the tiled MPSoC with the diagnosis extensions was synthesized for a ZTEX~1.15d board with a Xilinx Spartan-6 XC6SLX150 FPGA. The relevant hardware utilization numbers as obtained from a Synplify Premier FPGA synthesis are given in Table~\ref{tab:resourceusage}. The functional system, even though it consists of four CPU cores, is relatively small, as the used mor1kx CPU cores are lightweight (comparable to small ARM Cortex M cores). The functional system contains no memory, but uses an external DDR memory. In this scenario, the full diagnosis system is rather large. We have implemented two types of CPU event generators. A ``lite'' variant of the event generator can trigger only on a program counter value, and not on the return from a function call. This reduced functionality makes the event generator comparable to the feature set of the ARM CoreSight ETM trace unit, which is said to use $\sim$~7,000 NAND gate equivalents~\cite{orme_debug_2008}, making it similarly sized as our event generator. The possibility to trigger also on the return from a function call significantly increases the size of the event generator, mostly due to additional memory. The diagnosis processor is about 20~percent larger than a regular compute tile, as it contains an additional DMA engine and the packet queues. It also contains 30~kByte of SRAM as program and data memory, which is not present in a regular compute tile. In summary, the resource usage of the diagnosis system is acceptable, especially if used in larger functional systems with more powerful CPU cores. At the same time, the implementation still contains many opportunities for optimization, which we plan to explore in the future. Also, a full system optimization to determine a suitable number of diagnosis processors and other processing nodes for a given number of CPU cores is future work. \section{Usage Examples} \label{sec:usage} We designed DiaSys as general-purpose approach to gain insight into SoCs, similar to today's tracing systems. Unfortunately, no benchmarks exist to evaluate such systems in a standardized way. We therefore rely on two spotlight usage examples to highlight important aspects of our approach. The first example is a hypothesis testing or ``debugging'' scenario which could also be performed using a trace-based debugger. We included this example as a demonstration of the flexibility of our approach: the process of debugging is usually a one-time effort, and the debugger is used as a tool to observe the program execution at various places in order to validate an hypothesis in a developer's head. In the second example we show how to create a lock contention profile with DiaSys. This example is taken from the area of runtime analysis, the other major area in which tracing is employed today. While the analysis tasks in this area are more standardized (thus need a less flexible diagnosis system), they usually require the long-time observation of the whole program execution, therefore producing large data rates in todays implementations. The creation of a lock contention profile therefore serves as a good example for the data reduction capabilities of DiaSys. \subsection{Hypothesis Testing: Finding a Race Condition} \label{sec:usage:racecondition} Hypothesis testing, or simply ``debugging,'' is the most common scenario in which software developers need to get insight into the software execution. While in many cases an intrusive debugging tool is sufficient, the more tricky bugs are related to timing, and thus require non-intrusive system insight. Today, developers use trace-based debugging for this task. In this example we show that DiaSys is equally suitable for such a scenario. In the following we discuss the debugging of a race condition which occurs in an application distributed over three compute tiles. We implement this example on our hardware implementation of DiaSys, as discussed in Section~\ref{sec:hwimpl}. \subsubsection{Problem Description} The application in this usage example consists of three tasks, running on three different processors concurrently. Core 0 runs the task \texttt{bank}, which is holds a variable \texttt{balance}. The other two cores 1 and 2 run the tasks \texttt{atm0} and \texttt{atm1}, respectively. All communication is handled through message passing. A message \texttt{get\_balance} reads the value of \texttt{balance} from core 0, and \texttt{set\_balance} writes it back. The tasks \texttt{atm0} and \texttt{atm1} periodically wait for a random amount of time, then get \texttt{balance}, decrement it by 1, and write it back. When running the application, we notice that sometimes $n$ calls to \texttt{set\_balance} do not, as expected, decrement \texttt{balance} by $n$, but by $m < n$. \subsubsection{Debugging Approach I: Observe exchanged messages} \begin{figure} \footnotesize \begin{sequencediagram} \newthread{atm1}{:atm0 (core 1)} \newinst[1.5]{bank}{:bank (core 0)} \newinst[1.5]{atm2}{:atm1 (core 2)} \stepcounter{threadnum} \node[below of=inst\theinstnum,node distance=0.8cm] (thread\thethreadnum) {}; \tikzstyle{threadcolor\thethreadnum}=[fill=gray!30] \tikzstyle{instcoloratm2}=[fill=gray!30] \begin{call}{atm1}{\shortstack{get\_balance\_req}}{bank}{\shortstack{get\_balance\_resp}} \end{call} \begin{call}{atm2}{\shortstack{get\_balance\_req}}{bank}{\shortstack{get\_balance\_resp}} \end{call} \begin{messcall}{atm1}{\shortstack{set\_balance}}{bank} \end{messcall} \begin{messcall}{atm2}{\shortstack{set\_balance}}{bank} \end{messcall} \end{sequencediagram} \caption{Sequence diagram showing the race condition in the first case study.} \label{fig:evaluation:racecondition_msc} \end{figure} Initially, we don't know where the problem is located. However, we assume that something in the exchange of messages goes wrong. We therefore use DiaSys to print out all incoming and outgoing messages at task \texttt{bank} for manual analysis. We start by configuring the CPU event generator at core 0 (running the task \texttt{bank}) to generate types of primary events: \begin{enumerate} \item One event if the message passing send function is called. As payload we capture the identifier of the destination core and the type of the message (e.g. \texttt{get\_balance\_resp}) \item Another event if the message passing receive function is called. As payload we capture the identifier of the source core and the type of the message. \end{enumerate} To create an event log, all events are sent directly to the host PC, where they are displayed to the developer in the form of a text log. For the purpose of easier understanding in this paper, Figure~\ref{fig:evaluation:racecondition_msc} presents the interesting section of this text log in the form of a sequence diagram. Looking at this diagram, experienced developers will notice the bug: two read-modify-write sequences are interleaved, causing the value written by \texttt{atm0} to be overwritten by \texttt{atm1}. Such behavior is a textbook example of a race condition. \subsubsection{Debugging Approach II: Transaction checking} The first debugging approach used DiaSys only to gather data, not to analyze it. The less often the race condition occurs, the more trace data is generated in the first approach, which must be transferred and manually checked. In our second approach, we also automate this checking. As result, the developer is only informed if an actual race condition occurred. In the correct scenario, the sequence of getting the balance, modifying it, and writing it back is an atomic transaction. We therefore form the hypothesis ``the read-modify-write sequence is atomic,'' and use DiaSys to check it. If the hypothesis does not hold, we have found a race condition. We first configure the event generator at core 0 (running the task \texttt{bank}) to create two primary events: \begin{itemize} \item An event \texttt{EV\_GET\_BALANCE\_CALL} when entering the function \texttt{get\_balance()} on core 0. This function is called if a \texttt{get\_balance\_req} message is received. \item Another event \texttt{EV\_SET\_BALANCE\_RETURN} when returning from the function \texttt{set\_balance()}. This function returns when the \texttt{set\_balance} message has been fully processed. \end{itemize} For both events, the source of the message (i.e. \texttt{atm0} or \texttt{atm1}) is included as payload. Furthermore, we program the diagnosis processor to execute a transformation actor shown in pseudo code in Listing~\ref{lst:tn_check_transaction}. (In our hardware implementation, we programmed the diagnosis processor in C with an equivalent program.) \begin{lstlisting}[label=lst:tn_check_transaction, caption={Pseudo code of the transformation actor checking the balance updating transaction.},float] TA_CHECK_BALANCE_TRANS { bool in_transaction = false; bool transaction_owner = NULL; event = wait(EV_GET_BALANCE_CALL or EV_SET_BALANCE_RETURN); // get_balance(src) and set_balance(src) // are both passed the source of the request // message as first function argument msg_src = event.data.args['src']; if (in_transacation && transaction_owner != msg_src) { // race condition found event_type = EV_RACE_DETECTED; event_data = {}; return new Event(event_type, event_data); } if (event.type == EV_GET_BALANCE_CALL) { // start of new transaction in_transacation = true; transaction_owner = msg_src; return; } if (event.type == EV_SET_BALANCE_RETURN && in_transacation && transaction_owner != msg_src) { // race condition found event_type = EV_RACE_DETECTED; event_data = {}; return new Event(event_type, event_data); } } \end{lstlisting} The two primary events are now routed from core 0 to the diagnosis processor, which checks the hypothesis that all transactions are atomic. If a violation, i.e. a race condition, is found, a new event \texttt{EV\_RACE\_DETECTED} is generated. This event is sent to the host PC and displayed as an event log to the developer. \subsubsection{Event and Data Rates} The most significant benefit from automating the hypothesis checking, as shown in the previous section, is the increase in productivity for the developer: even in long-running programs, testing the hypothesis becomes straightforward. At the same time, the automation also reduces the required trace bandwidth. The first approach of creating events for all received and transmitted messages for manual inspection required a trace bandwidth of 57.09~KBit/s. When automating the checking in the second approach, the bandwidth was reduced to 5.66~KBit/s, a reduction by $10.1\times$. The bandwidth requirements are heavily dependent on the program characteristics. In the program executions on our hardware platform running at 50~MHz, we observed 44~\% of transactions were interleaved, i.e. a race condition. Doing the same analysis in a tracing system would require an instruction and a data trace to be captured. Even though our evaluation platform is not able to produce such traces, we can estimate the required trace bandwidth by measuring the number of executed instructions and data memory accesses. Assuming an instruction trace compressed to 2~bit/instruction and a data trace compressed to 16 bit/access (for data and address, following \cite{hopkins_debug_2006}), we get a total trace stream of 246.31~MBit/s. It must be noted, however, that this number is an upper bound on the required bandwidth for a tracing system. Depending on the used tracing system implementation (such as CoreSight), not a full instruction or data trace is captured, but filters and triggers can be used to reduce this stream to just relevant parts, similar to what our event generators do. \subsubsection{Outlook: Programming DiaSys} In our first case study, we have shown how to use DiaSys to obtain data for manual analysis from the SoC, and how to automate the manual checking as hypothesis test running on-chip on the diagnosis processor. In this example, we wrote the transformation actor manually as C program, which is called upon receiving events. In the future, we envision also other means to describe the diagnosis application executed by DiaSys. In this example, a description of the desired system behavior as linear temporal logic (LTL) expression could be applied, as it commonly done in runtime validation systems. Using such LTL expressions for the validation of the correct hardware configuration by the software has been presented in \cite{li_rule-based_2016}. \subsection{Runtime Analysis: Creating a Lock Contention Profile} \label{sec:usage:lockprofiling} In our second usage example, we present a diagnosis application which generates a ``profile,'' a statistic about some software behavior. Examples for such commonly used profiles are the various \texttt{*top} Linux commands which present an ordered list of processes or threads with the highest usage of CPU, I/O, memory, or other resources. We include this usage example as it represents, next to debugging, the second large motivation to use a tracing tool today: runtime analysis. Other examples from this category of applications are code coverage and runtime profiling. Most runtime analysis tasks today require a full (instruction) trace to be recorded, which requires a high off-chip bandwidth. Using DiaSys, we can show how this trace bandwidth can be reduced by on-chip processing. To give meaningful insight into data rates generated when analyzing large real-life applications, we use in this example not a self-created example application, but an application from the PARSEC benchmark suite. \subsubsection{Evaluation Prototype} \label{sec:usage:lockprofiling:software_prototype} \begin{figure} \includesvgnolatex{img/software_prototype} \caption{The software prototype of the diagnosis system. All calls in the observed application are recorded in an event log file by an preloaded library. The event log file is read by the diagnosis system implemented as Python application. The figure shows the monitoring of all \texttt{pthread\_mutex\_lock()} calls and returns as used by the usage example in Section~\ref{sec:usage:lockprofiling}.} \label{fig:software_prototype} \end{figure} The hardware implementation prototype of DiaSys presented in Section~\ref{sec:hwimpl} is only able to run a very limited set of applications, as it only provides bare-metal programming support (comparable to a microcontroller without operating system), and no full POSIX environment as it can be found on Linux for example. To run larger applications, such as standard benchmarks, we therefore created a software prototype of the diagnosis system. It runs purely in software on a Linux PC and is best suited for an evaluation of event rates inside the diagnosis system, as well as the design of new diagnosis applications. Since no hardware extensions are used, its operation is intrusive, i.e. the timing of the observed application is slightly changed. The prototypical event generators can only trigger on the call of and return from a C library function, and the function arguments can be included in the event as data items. The software prototype consists of two parts, which are shown in Figure~\ref{fig:software_prototype}. The first part is a ``preload library.'' It is a small software library written in C which is able to monitor all calls to C library functions and write them into an event log file. This event log file is then used by a prototype of the diagnosis system implemented in Python. It consists of event generators, which read the event log file. A set of Python functions connected by channel objects represent the transformation actors. (We assume a one-to-one mapping of transformation actors to processing nodes in this prototype.) The output of the diagnosis application is directly printed to a console, as specialized event sinks are not necessary for our evaluations. We now use this software prototype to create the lock contention profile. \subsubsection{Problem Description} A \emph{lock contention} occurs in concurrent programs if multiple threads try to acquire a mutex lock at the same time~\cite[p. 147]{herlihy_art_2008}. In this case, all but one threads have to wait for the lock to be released before they can continue processing. Therefore, the lock acquisition time is a good metric for program efficiency: the less time it takes, the earlier the thread is done with its work. In order for a developer to get insight into the lock contention behavior of the program, a contention profile can be created. It lists all acquired locks, together with the summarized and averaged times the acquisition took. Such a profile can be generated in an intrusive way with tools like Intel VTune Amplifier or mutrace\footnote{\url{http://0pointer.de/blog/projects/mutrace.html}}, and is traditionally formatted as shown in Listing~\ref{lst:lockprofile_output}. \subsubsection{Measurement Approach} The lock acquisition time can be measured by obtaining the time the mutex lock function took to execute. In applications using pthreads, as it is the case for almost all applications running on Linux, macOS or BSD, the mutex lock function is named \verb|pthread_mutex_lock()|. \begin{lstlisting}[label=lst:mutex_lock, caption={A sketch of the \texttt{pthread\_mutex\_lock()} function. This function must be executed atomically, i.e. without interruption.}] int pthread_mutex_lock(pthread_mutex_t *mutex) { blocking_wait_until_mutex_is_free(mutex); lock_mutex(mutex); return 0 /* success */; } \end{lstlisting} As shown in the simplified code sketch in Listing~\ref{lst:mutex_lock}, the function blocks for an indefinite amount of time until a lock is available. If it is available, it acquires the lock and returns. To create a lock contention profile, we need to measure the execution times of all \verb|pthread_mutex_lock()| function calls in all threads. We then group this measurement by lock, given by the argument \verb|mutex| of the lock function, to obtain the number of times a lock was acquired, how long all lock acquisitions took in summary, and on average. \subsubsection{Diagnosis Application} \begin{figure} \includesvgnolatex{img/evaluation_lockprofile_appmodel} \caption{A diagnosis application to create a lock contention profile.} \label{fig:evaluation_lockprofile_appmodel} \end{figure} To perform the analysis outlined in the previous section, we configure the diagnosis system as shown in Figure~\ref{fig:evaluation_lockprofile_appmodel}. First, we attach two event generators to each CPU in the observed system. We configure them to generate two primary events which together measure the execution time of the lock acquisition function \texttt{pthread\_mutex\_lock()}. \begin{itemize} \item One primary event \texttt{EV\_LOCK\_CALL} is triggered if the CPU \emph{enters (calls)} the \texttt{pthread\_mutex\_lock()} function. The first function argument to \texttt{pthread\_\-mutex\_\-lock}, the \texttt{mutex}, is attached to the event as data item, together with a timestamp containing the current time. \item Another primary event \texttt{EV\_LOCK\_RETURN} is triggered if the CPU \emph{returns} from the \texttt{pthread\_mutex\_lock()} function. For this event, only a timestamp is attached as event data. \end{itemize} To calculate the execution time of the function \texttt{pthread\_\-mutex\_\-lock()}, we create a transformation actor \texttt{TA\_DIFF}, the pseudo code of which is shown in Listing~\ref{lst:tn_time_diff}. \begin{lstlisting}[label=lst:tn_time_diff, caption={Pseudo code of the transformation actor calculating the lock acquisition time.},float] TA_DIFF { lock_call = wait(EV_LOCK_CALL); lock_return = wait(EV_LOCK_RETURN); uint16_t time = lock_return.data.timestamp - lock_call.data.timestamp; uint16_t mutex_hash = hash(lock_call.data.args['mutex']); event_type = lock_acq_time; event_data = { time: time, lock: mutex_hash }; return new Event(event_type, event_data); } \end{lstlisting} It waits for both primary events \texttt{EV\_\-LOCK\_\-CALL} and \texttt{EV\_\-LOCK\_\-RETURN}, calculates the difference between the timestamps, and creates a new event \texttt{EV\_LOCK\_ACQ\_TIME} with two data items, the lock acquisition time and a hash of the \texttt{mutex} argument to reduce the data size. As last step in the processing, all \texttt{EV\_LOCK\_ACQ\_TIME} events are aggregated by another transformation actor called \texttt{TA\_STAT}. Again, a pseudo code implementation is given in Listing~\ref{lst:tn_stat}. \begin{lstlisting}[label=lst:tn_stat, caption={Pseudo code showing the functionality of the transformation actor creating a lock profile.},float] TA_STAT { event = wait(EV_LOCK_ACQ_TIME or EV_SEND_LOCK_PROFILE); // aggregate if (event.type == EV_LOCK_ACQ_TIME) { m = event.data.mutex; stat[m]['cnt']++; stat[m]['t_sum'] += event.data.time; return; } // send statistics output to host PC if (event.type == EV_SEND_LOCK_PROFILE) { event_type = EV_LOCK_PROFILE; event_data = {stat: stat}; return new Event(event_type, event_data); } } \end{lstlisting} If an event of type \texttt{EV\_LOCK\_ACQ\_TIME} is received, the timestamp is added to a hash data structure which records, grouped by the mutex, the number of calls to the lock function and the total time these calls took. After the program run, on request of the developer running the diagnosis, or in regular time intervals, a \texttt{EV\_SEND\_LOCK\_PROFILE} is generated. If this event is received, the aggregated statistics are sent to the event sink, which then presents the aggregated results to the developer. \subsubsection{Evaluation of the Diagnosis Application} In the evaluation of this usage example we focus on the event and data rates between the event generators and transformation nodes. In order to provide realistic inputs, we profiled the \emph{dedup} application from the PARSEC~3.0 Benchmark Suite with the large input data sets~\cite{bienia11benchmarking}. As PARSEC does not run on our custom-built prototype MPSoC platform, we used the software prototype described in Section~\ref{sec:usage:lockprofiling:software_prototype}. All transformation actors were implemented in Python code equivalent to the pseudo code listings presented in Listings~\ref{lst:tn_time_diff} and~\ref{lst:tn_stat}. \paragraph{Output of the Diagnosis Application} Before we analyze the diagnosis application itself, we discuss the output it generates, i.e. the lock contention profile shown in Listing~\ref{lst:lockprofile_output}. PARSEC was instructed to use at least 4 threads; ultimately 16 threads were spawned by the dedup application. (There is no option in PARSEC to specify the exact number of threads used.) The execution of the observed application took 2.68~s. \begin{lstlisting}[label=lst:lockprofile_output, caption={Output of the lock contention profile diagnosis application observing the PARSEC dedup application.}] mutex # acq. sum [ns] avg [ns] (01) 0x7fd9ac018988 47785 8835387 184.90 (02) 0x7fd9d1ed2978 47784 226012031 4729.87 (03) 0x1c36500 9426 53724035 5699.56 (04) 0x1c36660 9423 21904608 2324.59 (05) 0x1c36710 4638 12528702 2701.32 (06) 0x1c365b0 105 46999 447.61 (07) 0x7fd9d2091430 8 1974 246.75 (08) 0x7fd9b41948f8 8 2277 284.62 (09) 0x7fd9b42b9ad8 8 2560 320.00 (10) 0x7fd9d20f8928 8 2215 276.88 \end{lstlisting} The output shows the top ten most acquired mutexes, together with the total and averaged lock acquisition time. Notable in this profile are mutexes 2 to 5, which take on average significantly longer to acquire: these locks are called to be ``contended.'' A profile helps to understand the program behavior and serves as a starting point to fix possible bugs or inefficiencies. If lock contention is observed (and performance goals of the application are not met), it is common to replace coarse-grained locks with more fine-grained locks, i.e. locks which protect a shorter critical section. However, fixing a bug is not in the scope of this work. Instead, we now turn our discussion to the event and data rates generated when executing the diagnosis application that generated the profile as shown. \paragraph{Event and Data Rates} We designed the diagnosis system to reduce the off-chip traffic by moving the data analysis partially into the SoC. To evaluate if the data rates are in fact reduced, we analyze event rates within the diagnosis application. We use the following event sizes: \begin{itemize} \item An \texttt{EV\_LOCK\_CALL} event requires 14 bytes: two bytes for the event type identifier, four bytes for the timestamp, and eight bytes for the \texttt{mutex} argument. \item An \texttt{EV\_LOCK\_RETURN} event requires six bytes: two bytes for the event type identifier and four bytes for the timestamp. \item An \texttt{EV\_LOCK\_ACQ\_TIME} event requires six bytes: two bytes for the event type identifier, two bytes for the lock acquisition time, and two bytes for the hashed \texttt{mutex} argument. \end{itemize} Over the whole program run, the event generators attached to the 16 CPUs generate a total of 516,254 events, which equals 4.9~MByte of transmitted data or, over the program runtime, an average data rate of 14.7~MBit/s. The \texttt{TA\_DIFF} transformation actors half the number of events, resulting in a data rate of 4.4~MBit/s, or a reduction to 30~\%. Finally, after being aggregated by \texttt{TA\_STAT}, the full result can be transferred off-chip with roughly 204~bytes \medskip A direct comparison of our results to existing tracing systems is challenging. For our analysis we need access to the \texttt{mutex} function argument through a data trace, which is not supported by higher-speed tracing implementations such as CoreSight PTM and Intel PT. However, as a first lower-bound estimation of the data rate generated by a state-of-the-art tracing system, we created a full instruction trace using Intel PT. The same PARSEC dedup application created a trace file of 1.82~GB, which corresponds to 5.4~GBit/s over the program runtime. In summary, DiaSys was able to reduce the required trace bandwidth compared to an Intel PT instruction trace significantly due to on-chip analysis. When transferring data off-chip after processing in the \texttt{TA\_DIFF} processing nodes, the bandwidth is reduced from more than 5.4~GBit/s to 4.4~MBit/s, a reduction by $1233 \times$. \paragraph{Discussion} Depending on the feature set and timestamp granularity of the various tracing implementations, the bandwidth reduction that DiaSys is able to achieve can vary. However, a general observation holds: the most significant bandwidth savings result from the fact that we very precisely capture only data in the event generators which is relevant to our problem. The subsequent processing step \texttt{TA\_DIFF} of calculating the time difference between two events is further able to discard roughly $\nicefrac{2}{3}$ of the data. Both operations are simple enough to be implemented even in resource-constraint on-chip environments. The final step \texttt{TA\_STAT} is again able to give large percentage-wise reductions in data rate, however the absolute savings might not justify an on-chip processing any more. This last step could therefore be executed on the host PC -- without changing the diagnosis application. \section{Conclusions} \label{sec:conclusion} In this paper we introduced DiaSys, a diagnosis system which aims to replace tracing systems in MPSoCs, where software observability is limited by the off-chip bottleneck. To avoid this bottleneck, we move parts of the data analysis into the chip and closer to the data source. The diagnosis system consists of event generators, which observe the functional units executing software on the SoC, the diagnosis application and event sinks. Diagnosis applications describe the data analysis task in a way that is understandable for the developer and portable across different SoCs. In detail we discussed their properties and semantics. Diagnosis applications are portable by design, because components of the application can be freely mapped to distributed diagnosis extensions inside the SoC, or to a runtime environment on the host PC. The implementation of such a mapping tool is future work. In our evaluation we showed on two prototypes that the implementation of the required diagnosis extensions is feasible with reasonable hardware overhead. We also showed in two usage examples from different domains that the envisioned reduction in off-chip bandwidth requirements can be achieved. In the future, we plan to extend this system with more specialized processing nodes, which are suited for common analysis tasks. We also investigate how machine-learning approaches can be used to dynamically adjust the analysis tasks during runtime. \subsection*{Acknowledgments} This work was funded by the Bayerisches Staatsministerium für Wirtschaft und Medien, Energie und Technologie (StMWi) as part of the project ``SoC Doctor,'' and by the German Research Foundation (DFG) as part of the Transregional Collaborative Research Centre ``Invasive Computing'' (SFB/TR 89). The responsibility for the content remains with the authors. We would like to thank Ignacio Alonso and Markus Göhrle for their contributions to the implementation of the evaluation platform. We especially thank Stefan Wallentowitz for the creation and ongoing support of OpTiMSoC. \bibliographystyle{elsarticle-num}
c06ec0f30465dd21ff3ea5027d2bbfa652a2e3e5
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} The Type~Ia supernova (SN~Ia) SN~2011fe was discovered on 2011 August 24, just 11~hr after explosion \citep{nugent11}. It is among the nearest ($\sim 6.9$~Mpc) and youngest ($\sim 11$~hr) SNe~Ia ever discovered. Extensive spectroscopic and photometric studies of SN~2011fe indicate that it is ``normal'' in nearly every sense: in luminosity, spectral and color evolution, abundance patterns, etc. \citep{parrent12,richmond12,roepke12,vinko12,munari13,pereira13}. Its unremarkable nature coupled with the wealth of observations made over its lifetime render it an ideal laboratory for understanding the physical processes which govern the evolution of normal SNe~Ia. Indeed, these data have allowed observers to place numerous and unprecedented constraints on the progenitor system of a particular SN~Ia \citep[e.g.,][]{li11,nugent11,bloom12,chomiuk12,horesh12,margutti12}. Equally as information-rich as observations taken at early times are those taken much later, when the supernova's photosphere has receded and spectrum formation occurs deep in the SN core. For example, \citet{shappee13} used late-time spectra to further constrain the progenitor system of SN~2011fe, namely that the amount of hydrogen stripped from the putative companion must be $< 0.001~M_\odot$. \citet{mcclelland13} found that the luminosity from SN~2011fe in the 3.6~$\mu$m channel of \textit{Spitzer}/IRAC fades almost twice as quickly as in the 4.5~$\mu$m channel, which they argue is a consequence of recombination from doubly ionized to singly ionized iron peak elements. In addition, \citet{kerzendorf14} used photometric observations near 930~d post-maximum light to construct a late-time quasi-bolometric light curve, and showed that the luminosity continues to trace the radioactive decay rate of $^{56}$Co quite closely, suggesting that positrons are fully trapped in the ejecta, disfavoring a radially combed or absent magnetic field in this SN. \citet{graham15} presented an optical spectrum at 981~d post-explosion and used constraints on both the mass of hydrogen as well as the luminosity of the putative secondary star as evidence against a single-degenerate explosion mechanism. \citet{taubenberger15} presented an optical spectrum at 1034~d post-explosion, and speculated about the presence of [\ion{O}{1}] lines near 6300~\AA, which, if confirmed, would provide strong constraints on the mass of unburned material near the center of the white dwarf progenitor of SN~2011fe. Non-detections of the H$\alpha$ line at both of these very late epochs also strengthened the constraints on the presence of hydrogen initially posed by \citet{shappee13}. Finally, \citet{mazzali15} used spectrum synthesis models of SN~2011fe from 192 to 364 days post-explosion to argue for a large central mass of stable iron and a small mass of stable nickel -- about 0.23~$M_\odot$ and 0.01~$M_\odot$, respectively. We complement these various late-time analyses with a series of radiative transfer models corresponding to a series of optical and ultraviolet (UV) spectra of SN~2011fe. \section{Observations} \label{sec:obs} \begin{table} \begin{tabular}{lll} UT Date & Phase & Telescope \\ & (days) & $+$Instrument \\ 2011 Dec 19 & $+$100 & WHT$+$ISIS \\ 2012 Apr 2 & $+$205 & Lick 3-m$+$KAST \\ 2012 Jul 17 & $+$311 & Lick 3-m$+$KAST \\ 2012 Aug 23 & $+$349 & Lick 3-m$+$KAST \\ 2013 Apr 8 & $+$578 &Lick 3-m$+$KAST \end{tabular} \caption{Observing log of spectra that appear here for the first time. The phase is with respect to maximum light.} \label{tab:obs} \end{table} We obtained optical spectra of SN~2011fe at days +100, +205, +311, +349, and +594 (Dec 19, 2011, Apr 2, 2012, Jul 17, 2012, Aug 23, 2012, Mar 27, 2013); the observations are shown in Figure~\ref{fig:all_optical_spectra_11fe} and described in Table~\ref{tab:obs}. The day +205 and +311 spectra were presented in \citet{mazzali15}. We also obtained an ultraviolet spectrum with \textit{Hubble Space Telescope} at day +360 (GO 12948; PI: R.~Thomas). This latter observation consisted of ten orbits, the first nine using the STIS/NUV-MAMA configuration, and the last with the STIS/CCD G430L and G750L configurations. The data from one of the NUV-MAMA orbits was unrecoverable, and so the final spectrum, shown in Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_no_model}, represents co-addition of the nine remaining observations. \begin{figure*} \centering \includegraphics[scale=0.5]{f1} \caption{Five optical spectra of SN~2011fe used in this work. The fluxes are scaled arbitrarily in order to facilitate spectral feature comparisons.\label{fig:all_optical_spectra_11fe} } \end{figure*} Also shown in Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_no_model} is a smoothed version of the \textit{HST} spectrum. We used the algorithm presented in \citet{marion09}, which consists of applying a low-pass filter to the signal. The motivation for this approach is the notion that the physical features in SN~Ia spectra are broad, while most noise in the spectrum is narrow. Therefore, if one can suppress the high-``frequency'' features (the noise), what will remain will be the pure signal from the SN. To accomplish this task, one calculates the power spectrum of the original spectrum using a Fourier transformation, suppresses the power spectrum at all high ``frequencies'' in which information is deemed to be noise, and applies an inverse Fourier transformation to recover the smoothed spectrum. An especially useful feature of this technique is its insensitivity to spikes in \textit{HST} spectra due to cosmic rays. \begin{figure*} \centering \includegraphics[scale=0.5]{f2} \caption{Smoothed spectrum using the low-pass filter technique outlined in \citet{marion09}.} \label{fig:sn11fe_combined_hst_spectra_ft_smoothed_no_model} \end{figure*} \section{Radiative transfer models} We used the \texttt{PHOENIX/1D} code \citep{hb99} with the same modifications discussed in \citet{friesen14} to capture the most important physical processes at late times in SNe~Ia. The underlying explosion model was a spherically symmetric delayed-detonation model presented in \citet{dominguez01}. For each observation presented in \S\ref{sec:obs} we calculated a corresponding synthetic spectrum, assuming a 16~day rise time for the model \citep{nugent11}. \section{Discussion} \label{sec:discussion} The theory of spectrum formation at late times in normal SNe~Ia has broadly converged to a scenario in which electron configurations of atoms in the rarefied ejecta are primarily in their ground state, and are excited by collisions with free electrons to low-lying metastable levels, which in turn emit forbidden lines as they return to the ground state \citep[e.g.,][]{meyerott78,meyerott80,axelrod80,ruiz-lapuente92,kuchner94,ruiz-lapuente95,bowers97,mazzali11,mazzali12,silverman13}. Little to no continuum emission is present in SN~Ia spectra at these epochs. This stands in contrast to the spectrum formation mechanism at early times, near maximum light, in which the optical depth to Thomson scattering on free electrons is large, leading to the formation of a photosphere on top of which atoms undergo line scattering via strong permitted lines, giving rise to P~Cygni spectral features. \citet{kirshner75} argued that the P~Cygni mechanism is no longer active at late times because the photosphere has disappeared (indicated by the absence of continuum) and there are no longer enough photons for these strong lines to scatter. Curiously, good spectral fits have been obtained for relatively late SN~Ia spectra with the parameterized spectrum synthesis code \texttt{SYNOW}, which treats only line scattering by permitted lines. Such fits require only a few ions -- \ion{Na}{1}, \ion{Ca}{2}, and \ion{Fe}{2} -- and fit optical spectra fairly well, especially blueward of $\sim 6000$~\AA. Examples include the normal SN~1994D at day +115 \citep{branch05}, the normal SN~2003du at day +84 \citep{branch08}, the subluminous SN~1991bg at day +91 \citep{branch08}, and the peculiar SN~2002cx at day +227 \citep{jha06}. While the parameterized approach of \texttt{SYNOW} to solving the radiative transfer equation restricts analysis of those fits to putative line identifications and velocity measurements, they nevertheless demonstrate that either SN~Ia spectra exhibit a remarkable degeneracy with respect to forbidden and permitted line formation, or that permitted lines continue to drive emergent spectrum formation at late times, regardless of what physical mechanisms generate the underlying flux \citep{branch08,friesen12}. These two competing analyses of late-time SN~Ia spectra agree that the majority of the spectrum is formed by Fe lines, but they predict dramatically different velocities of the line-forming regions in the ejecta. For example, \citet{branch08} argue that Fe, Ca, and Na are located at 7000~km~s$^{-1}$ and higher in the day +84 spectrum of SN~2003du. In contrast, \citet{bowers97} argue for velocities from 1000~--~3000~km~s$^{-1}$ in the +95 spectrum of the same object. Identifying the correct velocity of the line-forming region has important consequences for constraining the structure of the inner regions of SN~Ia ejecta, which in turn constrain the as-yet unknown explosion mechanism. \subsection{Day +100} \label{subsec:11fe_p100} The day +100 spectrum of SN~2011fe and the corresponding synthetic spectrum from \texttt{PHOENIX} are shown in Figure~\ref{fig:pah_std_d116_delta_t_no_forb_lines_vs_11fe_p100}. Overall the fit is good, although a few features in our model do not match those in the observed spectrum, namely the emission feature near 5900~\AA. In addition, the peak at 4700~\AA\ in the synthetic spectrum is too weak, the blue side of the broad emission at 7200~\AA\ is absent in the model, and the flux in the blue and near-UV is too high. Most other features are well reproduced, both in strength and in shape. \begin{figure*} \centering \includegraphics[scale=0.5]{f3} \caption{Synthetic spectrum of delayed-detonation model of \citet{dominguez01} at day 116 vs. SN~2011fe at day +100. No forbidden lines were included in this calculation.} \label{fig:pah_std_d116_delta_t_no_forb_lines_vs_11fe_p100} \end{figure*} In the context of much of the literature which concerns late-time SN~Ia spectra, the fidelity of this fit is peculiar because the calculation used the most current atomic database of \citet{kurucz95,kurucz02}, which includes no forbidden line data for any ions. This stands in contrast to the most common interpretations of spectra of ``old'' SN~Ia, which were discussed earlier. If the purely permitted line identifications are correct they are difficult to reconcile with kinematic analyses such as that of \citet{maeda10}, which assume that the emission peaks correspond to forbidden lines forming within a few 100~km~s$^{-1}$ of $v \simeq 0$~km~s$^{-1}$. Rather, the strong emission peaks at $4700$~\AA\ and $5200$~\AA, which each have previously been identified as a mixture of [\ion{Fe}{2}] and [\ion{Fe}{3}] lines, may instead be formed by the handful of permitted lines of \ion{Fe}{2} whose upper levels are among the crowded $3d^6(^5D)4p$ configuration, with energies between 5~eV and 6~eV, and whose lower levels are, coincidentally, the handful of metastable levels around 3~eV which are purportedly responsible for the aforementioned forbidden emission features. However, it is important to note that the analysis of \citet{maeda10} only requires that the ejecta is optically thin at the rest wavelength of the line and as we show below in the redder parts of the spectra that condition is met at later epochs. In the line-scattering interpretation of spectrum formation, this would imply that the dips just to the blue of these two strong emission features are the corresponding absorptions, rather than regions lacking in emission. These absorptions would correspond to line velocities of $\sim 6000$~km~s$^{-1}$, similar to that found in the +115 spectrum of SN~1994D by \citep{branch05}. Although below we illustrate some complications with this permitted-line-only model, it is instructive first to entertain the idea that this is, in fact, representative of late-time spectrum formation physics in SNe~Ia. Given the contrast of this result with those found elsewhere in the literature, it is important to evaluate the late-time line scattering scenario within the context of other analyses of SN~2011fe, in order to determine whether or not it is copacetic with what is already known about the spectral evolution of this object. We consider three such pieces of analysis. First, \citet{parrent12} traced the velocity evolution of \ion{Fe}{2} in the early spectra of SN~2011fe using the automated spectrum code \texttt{SYNAPPS} \citep{thomas11} and found that at day +15 the minimum velocity of that ion was $\sim 8000$~km~s$^{-1}$ (see their Figure~3). Furthermore, after maximum light, the rate of change of line velocities in SN~2011fe, and in most SNe~Ia in general, slows dramatically. Second, in the hydrodynamical model used in our calculation, Fe remains the most abundant species in the ejecta from the center of the ejecta out to $\sim 12 000$~km~s$^{-1}$ \citep[see][Figure~2]{friesen14}; our putative line velocity estimate of $\sim 6000$~km~s$^{-1}$ falls well within this range. Finally, \citet{iwamoto99} show that the optical depth of \ion{Fe}{2} computed in local thermodynamic equilibrium (LTE) peaks between 5000~K and $10 000$~K, roughly the same temperature range as that of the ejecta in our models. (One would be remiss to read too much into this corroboration, as the radiation field in the SN~Ia ejecta at this epoch is far from LTE.) Although none of these results offer conclusive evidence that the strong features in the +100 spectrum of SN~2011fe are indeed P~Cygni profiles, they do show that it is quite reasonable to consider that possibility. We caution that it is unlikely that the \emph{entire} optical spectrum consists of overlapping P~Cygni line profiles due to resonance-scattering, as is the case at very early (photospheric) epochs in SNe~Ia. \citet{branch08} attempted to fit the day +84 spectrum of SN~2003du with the resonance-scattering code \texttt{SYNOW}, and found that P~Cygni lines fit the blue part of the spectrum (blueward of 6600~\AA) quite well, but failed quite severely redward of that. As they discuss, the likely explanation is that resonance-scattering near this epoch is very influential at blue wavelengths, but forbidden emission is prominent in the red. (We find a similar result in our attempts to fit the optical and UV spectra at +349 and +360, respectively, which we discuss below.) In fact, to argue that spectrum formation consists of \emph{either} resonance-scattering by optically thick permitted lines \emph{or} emission from optically thin forbidden lines is somewhat of a false dilemma, as both scenarios assume a degree of locality in the radiative transfer which is probably unphysical. In particular, the former assumes that the source function $S$ depends only on the local mean intensity $J$, while the latter assumes that emitting lines are well separated in wavelength such that they act independently of each other. Each of these approximations is valid in certain regimes, i.e., resonance-scattering at photospheric epochs and forbidden emission at \emph{very} late times ($> 1$~yr) and in wavelength regions far from the forest of iron-peak lines, such as the infrared, but there exists a wide range between those extremes, in which all of these effects compete to form the emergent spectrum. \citet{bongard08} addressed this topic in detail by calculating a grid of \texttt{PHOENIX} spectra using the hydrodynamical model W7 \citep{nomoto84} at 20 days post-explosion. They found that even at very low velocities and high optical depth ($\tau > 3$), the ``spectrum''\footnote{The radiation flux $F_\lambda$.} at those velocities is already highly distorted from that of a blackbody, due to line and continuum interactions of the radiation field with iron-peak elements deep in the core of the SN. The ions found at higher velocities, near the photosphere, then further distort this underlying spectrum through additional absorption, emission, and line scattering, leading to an emergent spectrum containing a complicated mixture of P~Cygni, continuum, and thermal components which are difficult to disentangle. That our spectral model which explicitly omits forbidden line data fits the day +100 spectrum of SN~2011fe reasonably well, suggests that day +100 is simply too early for collisionally excited forbidden emission to be the primary driver of spectrum formation. It appears that line scattering processes continue to contribute significantly, even this late in the lifetime of this SN. In short, there are many physically-motivated reasons to suspect that permitted lines play an important role in SN~Ia spectrum formation at this epoch. However, since forbidden lines are frequently identified in spectra of SNe~Ia of this age, we tested this theory by expanding our atomic database to include collisional and radiative data of forbidden lines, as described in \citet{friesen14}. We then repeated the radiative transfer calculation with this expanded database, and compare the two results in Figure~\ref{fig:pah_std_d116_delta_t_forb_lines_vs_no_forb_lines_vs_11fe_p100}. The results are quite similar, except that the model with forbidden lines has a lower UV flux and stronger emission at 7300~\AA\ and 8600~\AA. \begin{figure*} \centering \includegraphics[scale=0.5]{f4} \caption{Comparison of \texttt{PHOENIX} spectra with and without forbidden lines at day 116.} \label{fig:pah_std_d116_delta_t_forb_lines_vs_no_forb_lines_vs_11fe_p100} \end{figure*} The most notable shortcoming of both synthetic spectra is the lack of emission at 5900~\AA. This feature has been identified alternatively as \ion{Na}{1}~D \citep{branch08,mazzali08} or [\ion{Co}{3}]~$\lambda 5888$~\AA\ \citep{dessart14}. The explosion model used in these calculations contains little \ion{Na}{1}, so it is not surprising that we do not recover a strong Na~I~D emission feature there. However, at day 116 the model contains several 0.1~$M_\odot$ of $^{56}$Co, and yet the forbidden emission at 5900~\AA\ does not appear. This discrepancy may be related to underestimating the gas temperature in the model at this epoch (see \S\ref{subsec:11fe_p205}). Identifying whether a feature is an ``emission'' or ``absorption'' is not a straightforward task in \texttt{PHOENIX} calculations. This is because the algorithm calculates emissivities and opacities of NLTE species by adding up all contributions to each at each wavelength point \citep[e.g.,][]{hb14}, with no regard to the underlying atomic processes which produced them. Such an approach captures naturally the notion that spectrum formation is inherently multi-layered in supernovae: one region deep in the ejecta may be strongly in emission at one wavelength, but a region above it may be optically thick at that wavelength, absorbing much of the underlying emission \citep{bongard08}. The emergent spectrum is then a convolution of both processes, and such classifications as ``absorption'' or ``emission,'' while relevant to single interactions, no longer describe adequately the complete process of spectrum formation. We are therefore relegated to using more indirect methods for isolating the sources of features in synthetic spectra. The single-ion spectrum method \citep{bongard08} can help one identify the particular ion or ions which influence particular parts of a synthetic spectrum, but it cannot, e.g., isolate the effects of permitted lines from forbidden lines, which is desirable in this context. We found that the only useful way to accomplish this was to remove the forbidden lines from the calculation entirely and re-compute the entire model. This can unfortunately broaden the parameter space of the model, since forbidden lines affect the temperature structure by acting as coolants \citep{friesen14,dessart14}. Unfortunately we are aware of no more targeted method of accomplishing this goal. We computed single-ion spectra for both synthetic spectra (with and without forbidden lines), for the most influential ions. For the spectrum without forbidden lines, these are shown in Figure~\ref{fig:pah_std_d116_delta_t_no_forb_lines_vs_11fe_p100_single_ion_spectra}. For the spectrum with forbidden lines, these are shown in Figure~\ref{fig:pah_std_d116_delta_t_forb_coll_vs_11fe_p100_single_ion_spectra}. \begin{figure*} \centering \includegraphics[scale=0.9]{f5} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 116, compared to SN~2011fe at day +100. No forbidden lines were included in this calculation.} \label{fig:pah_std_d116_delta_t_no_forb_lines_vs_11fe_p100_single_ion_spectra} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.9]{f6} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 116, compared to SN~2011fe at day +100.} \label{fig:pah_std_d116_delta_t_forb_coll_vs_11fe_p100_single_ion_spectra} \end{figure*} Both with and without forbidden lines, the synthetic spectra indicate that most of the optical spectrum at day +100 is formed by \ion{Fe}{2}. In addition, the \ion{Ca}{2} H \& K doublet at $\lambda \lambda 3934, 3968$~\AA, a pair of strong resonance lines, contributes significantly to the emission at 4000~\AA. However, comparison between the two sets of single-ion spectra indicate a fascinating result which \citet{branch08} found highly improbable: it appears that entirely different combinations of atomic lines can conspire to produce similar optical spectra. Furthermore, the synthetic spectra in Figure~\ref{fig:pah_std_d116_delta_t_no_forb_lines_vs_11fe_p100} and Figure~\ref{fig:pah_std_d116_delta_t_forb_lines_vs_no_forb_lines_vs_11fe_p100} are the \emph{natural endpoints} of calculations subject to otherwise identical parameters. For example, when forbidden lines are included, the emission at 4000~\AA\ is due entirely to the \ion{Ca}{2} H \& K doublet; when they are absent, it is a combination of that same doublet with contributions also from \ion{Fe}{2} and \ion{Co}{2}. The double-horned emissions at 7250~\AA\ and 7500~\AA\ in the synthetic spectra lacking forbidden lines are due to emission from \ion{Fe}{2}; but the double-horned features at 7150~\AA\ and 7400~\AA\ in the spectra containing forbidden lines are due to \ion{Fe}{2} (possibly [\ion{Fe}{2}]~$\lambda \lambda 7155, 7171$~\AA)and [\ion{Ni}{2}]~$\lambda 7374, 7412$~\AA. This is likely stable $^{58}$Ni, since the radioactive $^{56}$Ni has mostly decayed by this point. The degeneracy among these various features is the likely explanation for the conflicting results of, e.g., \citet{bowers97} and \citet{branch08}. One would expect that adding forbidden lines is always favorable over neglecting them: if a calculation captures all relevant atomic processes and if forbidden lines are truly unimportant in some scenario, they will naturally ``deactivate''. And in fact, adding in the forbidden atomic data did address some problems in the synthetic spectra which lacked them. The lower UV flux and the 8600~\AA\ emission in the synthetic spectrum forbidden lines can be explained by their cooling effects: lower temperatures generally lead to lower opacities in the UV, and a lower temperature allows more \ion{Ca}{3} to recombine to \ion{Ca}{2}. However, the cooling effects also introduced a new problem: the infrared triplet (IR3) of \ion{Ca}{2} $\lambda \lambda 8498, 8542, 8662$~\AA\ (a trio of strong permitted lines) is responsible for the emission at 8600~\AA, but the model overestimates the strength of the emission at this wavelength. In the observation there is a pair of weaker emission features at the same location, and it is possible that at least one of these two emissions is due to the \ion{Ca}{2} IR3, although probably not both, since they are spread too far apart in wavelength. It is therefore not entirely clear which of the two synthetic spectra are ``better,'' and it is possible therefore that both permitted \emph{and} forbidden lines affect the optical spectra of SNe~Ia at this epoch. \subsection{Day +205} \label{subsec:11fe_p205} The day +205 spectrum of SN~2011fe and the corresponding \texttt{PHOENIX} spectrum are shown in Figure~\ref{fig:pah_std_d221_delta_t_forb_coll_vs_11fe_p205}. Attempts to calculate a spectrum at this epoch without forbidden lines, as was done in \S\ref{subsec:11fe_p100}, led to unrecoverable numerical instabilities in the code. It seems, then, that by this age forbidden lines play an important role. The single-ion spectra are shown in Figure~\ref{fig:pah_std_d221_delta_t_forb_coll_vs_11fe_p205_single_ion_spectra}. The emission feature at 4700~\AA\ is primarily [\ion{Fe}{3}]~$\lambda \lambda 4607, 4658$~\AA\ and [\ion{Fe}{2}]~$\lambda \lambda 4640, 4664$~\AA. The weak but clearly separate features around 4300~\AA\ are [\ion{Fe}{2}]~$\lambda \lambda 4287, 4359$~\AA, and the emission at 5300~\AA\ is [\ion{Fe}{2}]~$\lambda 5300$~\AA. The double-horned feature in the synthetic spectrum centered around 7300~\AA\ consists of [\ion{Fe}{2}]~$\lambda \lambda 7155, 7172$~\AA\ on the left, and [\ion{Ni}{2}]~$\lambda 7412$~\AA\ on the right; the shape of this pair of features is too exaggerated in the synthetic spectrum compared to the day +100 spectrum of SN~2011fe, but at later epochs the shape is a good match to the observations. At this epoch the ratio of \ion{Fe}{2} to \ion{Fe}{3} is well reproduced, with the strength of the 4700~\AA\ emission from \ion{Fe}{3} improved over that from the day +100 spectrum. However, the 5200~\AA\ emission, also from \ion{Fe}{3}, is somewhat weak. The strong \ion{Ca}{2} IR3 emission which was overestimated in strength in they day +100 synthetic spectrum is now absent entirely. Coincidentally, the forbidden line [\ion{Fe}{2}]~$\lambda 8617$~\AA, at nearly the same wavelength as IR3, has grown in strength at day +205, and fits quite well to the observation. The \ion{Ca}{2} H \& K doublet, which was quite strong at day +100, has diminished in strength and is replaced mostly by [\ion{Fe}{3}]~$\lambda 4008$~\AA. It seems, then, that the serendipitous degeneracy among permitted and forbidden lines which \citet{branch05} found unlikely, may actually be realized, at least for some features in the optical spectra of SN~2011fe. The double-horned pair of emissions centered at 7200~\AA\ is better reproduced at this epoch as well. Curiously, the emission at 5900~\AA\ is now quite well fit with [\ion{Co}{3}]~$\lambda 5888$~\AA, while at day +100, when most of the $^{56}$Ni had decayed to $^{56}$Co, the feature was absent entirely. It is possible that the temperature in the day +100 model was too low, which would explain the underabundance of \ion{Fe}{3} emitting at 4700~\AA\ and \ion{Co}{3} emitting at 5900~\AA. A higher temperature would also reduce the abundance of \ion{Ca}{2} in favor of \ion{Ca}{3}, explaining the reduced strength of both the H \& K doublet and the IR3. \begin{figure*} \centering \includegraphics[scale=0.5]{f7} \caption{Synthetic spectrum of delayed-detonation model of \citet{dominguez01} at day 221 vs. SN~2011fe at day +205.} \label{fig:pah_std_d221_delta_t_forb_coll_vs_11fe_p205} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.9]{f8} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 221, compared to SN~2011fe at day +205.} \label{fig:pah_std_d221_delta_t_forb_coll_vs_11fe_p205_single_ion_spectra} \end{figure*} \subsection{Day +311} The observed and synthetic spectra at day +311 are shown in Figure~\ref{fig:pah_std_d331_delta_t_forb_coll_vs_11fe_p311}, and the single-ion spectra are shown in Figure~\ref{fig:pah_std_d331_delta_t_forb_coll_vs_11fe_p311_single_ion_spectra}. At this epoch the spectra look similar to those at day +205. The [\ion{Fe}{3}] emission at 4700~\AA\ is still strong, although at day +311 the [\ion{Fe}{2}] emission at 4400~\AA\ has grown in strength, and continues to do so at later epochs. This is likely a reflection of some (but not much) recombination from \ion{Fe}{3} to \ion{Fe}{2} at this age. The \ion{Ca}{2} H \& K emission is still present at 4000~\AA, but somewhat weak, just as at day +205. \begin{figure*} \centering \includegraphics[scale=0.5]{f9} \caption{Synthetic spectrum of delayed-detonation model of \citet{dominguez01} at day 331 vs. SN~2011fe at day +311.} \label{fig:pah_std_d331_delta_t_forb_coll_vs_11fe_p311} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.7]{f10} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 331, compared to SN~2011fe at day +311.} \label{fig:pah_std_d331_delta_t_forb_coll_vs_11fe_p311_single_ion_spectra} \end{figure*} \subsection{Day +349} \label{subsec:11fe_p349} The model and observations at day +349 are displayed in Figure~\ref{fig:pah_std_d376_delta_t_vs_11fe_p349}. The corresponding single-ion spectra are shown in Figure~\ref{fig:pah_std_d376_delta_t_forb_coll_vs_11fe_p349_single_ion_spectra}. The \ion{Ca}{2} H \& K doublet emission at 4000~\AA\ is of similar strength as at day +205, and it may still contribute to that feature in SN~2011fe, although the [\ion{Fe}{3}] appears to be stronger at that wavelength. At this and later epochs the model begins to exhibit some problems. A sharp emission feature forms in the near-UV in the synthetic spectrum which is not observed in SN~2011fe. In addition, in the synthetic spectrum the emission at 4400~\AA\ (due mostly to \ion{Fe}{2}) is nearly as strong as the 4700~\AA\ feature (mostly \ion{Fe}{3}), while in the observed spectrum the latter remains considerably stronger. This problem is likely not one of radiative transfer effects in the model, but rather one of atomic physics. The recombination rate for ions scales with the free electron density $n_e$, which dilutes geometrically roughly as $t^{-3}$ \citep{de10a}. Thus at these very late times the recombination time scale for, e.g., \ion{Fe}{2}, can be of the same order as the dynamical time scale, i.e., the age of the SN. In this case, time-dependent effects of ion recombination can become influential on spectrum formation \citep{sollerman04,taubenberger15}. In the calculations used to generate the above figures, we neglected time-dependence in both the radiation field and the ion populations: both are assumed to be in steady-state. Assuming a steady-state radiation field is a valid approximation at late times --- since at most wavelengths the optical depths in the ejecta are low, the radiative transfer time scale is effectively the light-crossing time, which is many orders of magnitude shorter than the dynamical time scale. Thus the radiation field equilibrates with the ejecta almost instantaneously at any given time $t$. However, by assuming steady-state in the ion populations, we overestimate the rate of recombination from, e.g., \ion{Fe}{3} to \ion{Fe}{2}. This manifests in synthetic spectra as \ion{Fe}{2} features which are too strong, as in Figure~\ref{fig:pah_std_d376_delta_t_vs_11fe_p349}. Presumably the same pathology affects the day +578 synthetic spectrum more severely (see Figure~\ref{subsec:11fe_p578}). Time-dependent effects in the ion populations may not be the only source of the discrepant features in the synthetic spectra at very late times. The density profile of the ejecta in the explosion model also strongly affect $n_e$. Thus the \ion{Fe}{2}-to-\ion{Fe}{3} population ratio may provide a constraint on the initial conditions of the explosion model. For example, Figure~\ref{fig:pah_std_d376_delta_t_vs_11fe_p349} may indicate that the density of the iron-rich core of the model is too high, leading to an $n_e$ which is too high, inducing recombination from \ion{Fe}{3} to \ion{Fe}{2} too soon. \begin{figure*} \centering \includegraphics[scale=0.5]{f11} \caption{Synthetic spectrum of delayed-detonation model of \citet{dominguez01} at day 376 vs. SN~2011fe at day +349.} \label{fig:pah_std_d376_delta_t_vs_11fe_p349} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.7]{f12} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 376, compared to SN~2011fe at day +349.} \label{fig:pah_std_d376_delta_t_forb_coll_vs_11fe_p349_single_ion_spectra} \end{figure*} \subsection{The UV spectrum at day +360} \label{subsec:11fe_p360} The UV spectrum from \textit{HST} at day +360, as well as the best fitting spectrum from \texttt{PHOENIX}, are shown in Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_vs_pah_std_d376_fixed_tcor}. The single-ion spectra are shown in Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_vs_pah_std_d376_fixed_tcor_single_ion}. From these one finds that \ion{Fe}{2} is responsible for most of the spectral features in the UV at day +360, just as it was at day +100. However, \ion{Fe}{3}, \ion{Co}{3}, and \ion{Ni}{3} all contribute significantly to the bluest portion of the spectrum as well. The most interesting result, however, is that the \ion{Ca}{2} H \& K doublet continues to contribute significantly to the emission around 4000~\AA, despite being over 1~yr since explosion. It seems, then, that the extreme strength of this line overcomes both the small total abundance of \ion{Ca}{2} in the ejecta, as well as the large amount of geometric dilution which accompanies a year of expansion. \begin{figure*} \centering \includegraphics[scale=0.5]{f13} \caption{Synthetic spectrum of delayed-detonation model of \citet{dominguez01} at day 376, compared to SN~2011fe obtained with HST at day +360.} \label{fig:sn11fe_combined_hst_spectra_ft_smoothed_vs_pah_std_d376_fixed_tcor} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.7]{f14} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 376, compared to SN~2011fe at day +360.} \label{fig:sn11fe_combined_hst_spectra_ft_smoothed_vs_pah_std_d376_fixed_tcor_single_ion} \end{figure*} A second notable feature of this result is that nearly all of the \ion{Fe}{2} features are permitted lines, not forbidden; the bluest forbidden lines for any ion in this version of the \texttt{PHOENIX} atomic database is about 3200~\AA. Identifying these features is no simple feat, however, because \ion{Fe}{2} has thousands of lines between $\sim 1600 - 4000$~\AA. Furthermore, the contributions from several other ions in the UV at late times, each with several thousands of lines themselves, are at some wavelengths of similar strength as \ion{Fe}{2}. The convolution of all of these lines from different species renders the identification of individual features in the UV a difficult task. However, even without identifying particular lines, we can nevertheless learn a great deal about the UV line forming region using other methods (see Figure~\ref{sec:opacity_late_times}). \subsection{Day +578} \label{subsec:11fe_p578} Figure~\ref{fig:pah_std_d594_vs_11fe_p578} shows the observed and synthetic spectra of SN~2011fe at day +578. The single-ion spectra are shown in Figure~\ref{fig:pah_std_d594_delta_t_forb_coll_vs_11fe_p578_single_ion_spectra}. The fit of the \texttt{PHOENIX} synthetic spectrum to the observation is poor, and has resisted improvement even experimenting with a variety of different temperature-correction algorithms. Possible culprits for this include time-dependent effects as discussed in \S\ref{subsec:11fe_p349}, as well as other physical processes which \texttt{PHOENIX} currently does not treat, including dielectric recombination and charge-exchange reactions. However, analysis of this result nevertheless reveals some useful information. For example, the unphysical spike in flux around 3250~\AA\ is due entirely to \ion{Fe}{2}, although to which line in particular is not clear. In addition, the emission at 8600~\AA\, formerly produced by \ion{Ca}{2} IR3, has been replaced by [\ion{Fe}{2}]~$\lambda 8617$~\AA. This is yet another example of a truly remarkable degeneracy among permitted lines and forbidden lines at similar wavelength, but which become active at very different times. Although our model spectra predict the recombination to \ion{Fe}{2} too early, the event eventually does happen in SN~2011fe. In particular, in the day +594 spectrum, the strong \ion{Fe}{3} emission peak at 4700~\AA\ has disappeared entirely, with only a handful of \ion{Fe}{2} features remaining. Indeed, \citet{taubenberger15} and \citet{graham15} have tentatively identified features of \ion{Fe}{1} in $\sim 1000$~d spectrum of SN~2011fe, perhaps heralding a concurrent recombination transition from \ion{Fe}{2} to \ion{Fe}{1}. \begin{figure*} \centering \includegraphics[scale=0.7,trim=2cm .5cm .5cm .5cm,clip=true]{f15} \caption{Synthetic spectrum of delayed-detonation model of \citet{dominguez01} at day 594 vs. SN~2011fe at day +578.} \label{fig:pah_std_d594_vs_11fe_p578} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.7]{f16} \caption{Single-ion spectra corresponding to the composite spectrum of the delayed-detonation model of \citet{dominguez01} at day 594, compared to SN~2011fe at day +578.} \label{fig:pah_std_d594_delta_t_forb_coll_vs_11fe_p578_single_ion_spectra} \end{figure*} \section{Velocity Shifts} In a recent paper, \citet{Black16} examined the wavelength shifts of prominent features at late times in a series of SNe~Ia spectra, including the spectra of SN~2011fe. They found, in particular, a redward shift of of the prominent 4700~\AA feature with no signs of the redward drift slowing down at epochs up to day +400. Figure~\ref{fig:feature_shifts} shows that our models show no such general trend. In fact, the 4700~\AA feature after a strong redward shift, begins to move back to the blue. Since our models do not show a strong fidelity with the observations it is difficult to draw firm conclusions. \citet{Black16} suggest that the redward shift is primarily caused by temporal variations due to lines of [Fe III] and [Co II]. Since these lines are very temperature sensitive, this could be indicative of our general model uncertainties. \begin{figure*} \centering \includegraphics[scale=0.7]{f17} \caption{The position of the central wavelength peak of the 4700~\AA, 5300~\AA, 7150~\AA, and 7400~\AA\ feature (clockwise from upper left) as a function of epoch.} \label{fig:feature_shifts} \end{figure*} \section{Opacity at late times} \label{sec:opacity_late_times} Our \texttt{PHOENIX} calculations have been fairly successful at reproducing the late-time optical and UV spectra of the normal SN~2011fe. In addition, one of the great advantages of using first-principles codes such as \texttt{PHOENIX} is that one may glean a great deal of information about the underlying physics which drives the formation of the synthetic spectra. For example, in Figure~\ref{fig:optical_depths_optical_p349} we show the optical depths along the $\mu = -1$ (radially inward) ray in the day +349 model whose spectrum was shown in Figure~\ref{fig:pah_std_d376_delta_t_vs_11fe_p349}. For reference, the black dashed line shows $\tau = 1$, the division between optically thick and optically thin. The dashed red line shows the optical depth due only to Thomson scattering; at early times this is the dominant opacity source and gives rise to the photosphere in SNe. At late times, however, the geometric dilution of the free electron density $n_e$ leads to a very low Thomson scattering opacity, falling well below $\tau = 1$ even all the way to the center of the ejecta. From this alone we may infer that there is likely very little continuum radiation present this late in a SN~Ia's lifetime, as reflected in the spectra. Furthermore, at only a select few wavelengths -- mostly on the blue edge of the optical band -- does the optical depth reach $\tau = 1$ at all; at most wavelengths $\lambda \gtrsim 4500$~\AA, the ejecta are quite optically thin, in agreement with previous studies. Our calculations therefore indicate that most of the optical spectrum at day +349 in SN~2011fe consists of blended emission features from collisionally excited, optically thin forbidden lines. \begin{figure*} \centering \includegraphics[scale=0.6,trim=1.5cm .5cm .0cm .5cm,clip=true]{f18} \caption{Optical depths at a collection of optical wavelengths in the day +100 model (cf. Figure~\ref{fig:pah_std_d116_delta_t_forb_lines_vs_no_forb_lines_vs_11fe_p100}).} \label{fig:optical_depths_optical_p100} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.6,trim=1.5cm .5cm 0.0cm .5cm,clip=true]{f19} \caption{Optical depths at a collection of optical wavelengths in the day +349 model (cf. Figure~\ref{fig:pah_std_d376_delta_t_vs_11fe_p349}).} \label{fig:optical_depths_optical_p349} \end{figure*} The UV portion of the spectrum of SN~2011fe behaves entirely differently than the optical, however. One may suspect as much simply by noticing the significant degree of structure and complexity in the observed UV spectrum (Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_no_model}). These suspicions are confirmed by analogous calculations of optical depths at various UV wavelengths, shown in Figure~\ref{fig:optical_depths_UV_p349}. The black and red dashed lines are the same as in Figure~\ref{fig:optical_depths_optical_p349}. Unlike the optical band, however, most UV wavelengths are \emph{extremely} optically thick, with many reaching $\tau \sim 10^5$ at the center of the ejecta. Another surprising result is that many UV wavelengths become optically thick at quite high velocity, crossing the $\tau = 1$ threshold at $v \sim 10 000 - 15000$~km~s$^{-1}$. This result is corroborated by the presence of the emission component of the \ion{Ca}{2} H \& K doublet near 4000~\AA\ in Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_vs_pah_std_d376_fixed_tcor_single_ion}. \begin{figure*} \centering \includegraphics[scale=0.6,trim=1.5cm .5cm .0cm .5cm,clip=true]{f20} \caption{Optical depths at a collection of UV wavelengths in the day +100 model (cf. Figure~\ref{fig:pah_std_d116_delta_t_forb_lines_vs_no_forb_lines_vs_11fe_p100}).} \label{fig:optical_depths_UV_p100} \end{figure*} \begin{figure*} \centering \includegraphics[scale=0.60,trim=1.5cm .5cm .0cm .5cm,clip=true]{f21} \caption{Optical depths at a collection of UV wavelengths in the day +349 model (cf. Figure~\ref{fig:pah_std_d376_delta_t_vs_11fe_p349}).} \label{fig:optical_depths_UV_p349} \end{figure*} If the UV remains as optically thick as Figure~\ref{fig:optical_depths_optical_p349} and Figure~\ref{fig:optical_depths_UV_p349} suggest, then it appears that the transition from the ``photospheric'' to the ``nebular'' phase in SNe~Ia is far more complex than expected. Specifically, there are likely few or no forbidden emission lines which are active in the UV; this precludes the possibility of measuring asymmetric bulk motion of the inner regions of the SN ejecta, since the assumption behind such measurements is that the emission lines are optically thin and centered at the line rest wavelengths \citep{maeda10}. On the other hand, since the UV is optically thick, the spectrum may consist of the same overlapping P~Cygni line scattering profiles which characterize maximum-light spectra of SNe; if this is the case we may be able to infer ejecta velocities of the iron-rich core of SN~2011fe by measuring the location of the absorption minima of the features in Figure~\ref{fig:sn11fe_combined_hst_spectra_ft_smoothed_no_model}. However, as discussed in \S\ref{subsec:11fe_p360}, the multitude of UV lines, as well as likely blending among several atomic species, make this challenging. If we entertain the possibility that some of the UV spectrum is forming at the UV photosphere at velocities of order $10 000$~km~s$^{-1}$, the rest wavelengths of such lines would still lie within a crowded space of UV and optical transitions of iron-peak elements. The problem with the later fits may be related to time-dependent effects in the ionization of the gas, as the recombination time scale at late times is likely of order of the age of the SN. By assuming steady-state in our calculations, we likely overestimated the recombination rate, leading to an overabundance of \ion{Fe}{2} with respect to \ion{Fe}{3}. Re-computing these models with time-dependence in the NLTE rate equations is possible in principle, but the inherent ``noisiness'' of the root-finding algorithm we used to calculate the temperatures becomes amplified with each time step, resulting in a large amount of spurious temperature oscillations in the model at very late times, and a poorly fitting spectrum. \section{Conclusions} We extended \texttt{PHOENIX} to calculate radiative transfer models well into the late-time epochs of SNe~Ia, with an eye toward obtaining good fits to the high-quality optical and UV spectra of SN~2011fe. Doing so required similar methods to those discussed in \citet{friesen14}, in particular using an alternative method to that of Uns\"old-Lucy for calculating the temperature structure of the gas, as well as accounting for the collisional and radiative rate data for forbidden lines, which behave quite differently than permitted lines. The resulting synthetic spectra, ranging from +100 to +578 days post-maximum light, vary in degrees of fidelity to corresponding observed spectra of SN~2011fe, with the earlier epochs fitting quite well and the later epochs less so. At day +100 we found that radiative transfer calculations which neglect forbidden lines and those which include them can produce remarkably similar optical spectra, but with quite different atomic species and combinations of lines forming the various features. We found that, at least as late as day +360, permitted lines such as \ion{Ca}{2} H \& K and IR3 continue to influence spectrum formation in the optical, and permitted lines of \ion{Fe}{2} form much of the spectrum in the UV. In addition, these models indicate that some emission features from permitted lines are replaced by other emission features of forbidden lines at nearly the same wavelength as the SN evolves. For example, the emission from \ion{Ca}{2} H \& K at 4000~\AA\ is replaced around day +205 by [\ion{Fe}{3}]~$\lambda 4008$~\AA, and the emission from \ion{Ca}{2} IR3 at around 8600~\AA\ is replaced by [\ion{Fe}{2}]~$\lambda 8617$~\AA. \section*{Acknowledgments} The work has been supported in part by support for programs HST-GO-12948.004-A was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. This work was also supported in part by NSF grant AST-0707704, by NASA Grant NNX16AB25G and DOE Grant DE-SC0009956. The work of EB was also supported in part by SFB 676, GRK 1354 from the DFG. R.J.F.\ gratefully acknowledges support from NASA grant 14-WPS14-0048, NSF grant AST-1518052, and the Alfred P.\ Sloan Foundation. This research used resources of the National Energy Research Scientific Computing Center (NERSC), which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231; and the H\"ochstleistungs Rechenzentrum Nord (HLRN). We thank both these institutions for a generous allocation of computer time. \clearpage
d18a74adc6f563a5494f9cf0ee38af9746eada63
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{High-field instability from spin-wave theory} In the asymptotic high-field limit all spins are aligned along the field axis. Magnon excitations are suppressed by a large energy gap. By lowering the field strength the magnon gap decreases and eventually vanishes at some critical field strength $h_{\mathrm{c}0}$. Below $h_{\mathrm{c}0}$ the high-field state becomes unstable, indicating a transition towards one of the various intermediate-field phases. (This is true as long as the continuous transition is not preempted by a first-order transition at some higher field $h_{\mathrm c} > h_{\mathrm{c}0}$.) The wavevector at which the magnon gap closes then determines the ordering wavevector of the intermediate-field phase. We parameterize the magnon excitations above the polarized ground state at high field by Holstein-Primakoff bosons $a_i$ and $b_i$ on the A and B sublattices of the honeycomb lattice. It is convenient to use a spin-space frame obtained by rotating the cubic-axes basis $\vec e_x$, $\vec e_y$, $\vec e_z$ such that the magnetic field lies in the $3$-direction, \begin{align} \label{eq:spin-frame} \vec e_1 & = \frac{(\vec e_z \times \vec h) \times \vec h}{\lvert (\vec e_z \times \vec h) \times \vec h \rvert}, & \vec e_2 & = \frac{\vec e_z \times \vec h}{\lvert \vec e_z \times \vec h \rvert}, & \vec e_3 & = \frac{\vec h}{\lvert \vec h \rvert}. \end{align} E.g., for field in the diagonal $[111]$ direction we choose the new spin-basis vectors $\vec e_1 = (\vec e_x + \vec e_y - 2 \vec e_z)/\sqrt{6}$, $\vec e_2 = (-\vec e_x + \vec e_y)/\sqrt{2}$, and $\vec e_3 = (\vec e_x + \vec e_y + \vec e_z)/\sqrt{3}$. To leading order in the $1/S$ expansion the spin operators in this basis read: \begin{figure*}[!p] \includegraphics[scale=0.66]{magnon-dispersion_001_phi062_h2983.pdf}\hfill \includegraphics[scale=0.66]{magnon-dispersion_001_phi1687_h2217.pdf} \caption{Magnon excitation spectrum $\varepsilon_{\vec q}$ from linear spin-wave theory in the polarized phase for field in the $[001]$ direction and $h=h_{\mathrm{c}0}$ along high-symmetry lines in the Brillouin zone (see inset). The magnon gap vanishes at M$_1$ and M$_3$ for $\pi/2 < \varphi < \varphi_{\mathrm{c}2}$, when the transition is towards the canted zigzag phase (left panel), as well as for $3\pi/2 < \varphi < \varphi_{\mathrm{c}4}$, when the transition is towards the canted stripy phase (right panel).} \label{fig:magnon-spectrum-001} \end{figure*} \begin{figure*}[!p] \includegraphics[scale=0.65]{magnon-dispersion_111_phi062_h2615.pdf}\hfill \includegraphics[scale=0.65]{magnon-dispersion_111_phi1687_h1663.pdf}\hfill \includegraphics[scale=0.65]{magnon-dispersion_111_phi1922_h4850.pdf} \caption{Same as Fig.~\ref{fig:magnon-spectrum-001} for field in the $[111]$ direction. The magnon gap vanishes at the K points in the Brillouin zone for $\pi/2 < \varphi < \varphi_{\mathrm{c}1}$, when the transition is towards the AF vortex phase (left panel), as well as for $3\pi/2 < \varphi < 7\pi/4$, when the transition is towards the vortex phase (middle panel). For $\varphi_{\mathrm{c}3} < \varphi < 2\pi$ the instability wave vector is at the $\Gamma$ point, indicating the transition towards the canted N\'eel phase. Note that due to the unbroken $\mathbb{Z}_3$ symmetry the magnon spectrum is the same at all three $\mathrm M$ points, in contrast to the situation when $\vec h \nparallel [111]$. } \label{fig:magnon-spectrum-111} \end{figure*} \begin{figure*}[!p] \includegraphics[scale=0.65]{magnon-dispersion_1107_phi062_h2629.pdf}\hfill \includegraphics[scale=0.65]{magnon-dispersion_1107_phi1687_h1696.pdf} \caption{Same as Fig.~\ref{fig:magnon-spectrum-001} for field in the $[11\frac{7}{10}]$ direction, when the magnon gap vanishes at incommensurate wavevectors between M$_2$ and K, K'.} \label{fig:magnon-spectrum-1107} \end{figure*} \begin{table*}[!p] \caption{Instability field strength $h_{\mathrm{c}0}$, instability wavevector $\vec{Q}$, and intermediate-field phase below $h_{\mathrm{c}0}$ for different values of $\varphi$ and field directions. The transitions between the polarized and intermediate-field phases are always continuous except where indicated. Here $\varphi_{\mathrm{c}1} \simeq 0.715\pi$, $\varphi_{\mathrm{c}2} = \pi - \arctan(1/2) \simeq 0.852\pi$, $\varphi_{\mathrm{c}3} \simeq 1.812\pi$, and $\varphi_{\mathrm{c}4} = 2\pi - \arctan(1/2) \simeq 1.852\pi$. For $\vec h \parallel [111]$ the high-symmetry points $\mathrm M_1$, $\mathrm M_2$, and $\mathrm M_3$ have equivalent spectrum, but not for $\vec h \nparallel [111]$.} \label{tab:instability-field} \renewcommand{\arraystretch}{1.2} \begin{tabular*}{\linewidth}{@{\extracolsep{\fill}} >{$}l<{$} | >{$}l<{$} >{$}l<{$} l | >{$}l<{$} >{$}l<{$} l | >{$}l<{$} >{$}l<{$} l} \hline\hline \multirow{2}{*}{$\varphi$} & \multicolumn{3}{c|}{$\vec h \parallel [001]$} & \multicolumn{3}{c|}{$\vec h \parallel [110]$} & \multicolumn{3}{c}{$\vec h \parallel [111]$}\\ & h_{\mathrm{c}0}(\varphi)/(AS) & \vec{Q} & phase($h \nearrow h_{\mathrm{c}0}$) & h_{\mathrm{c}0}(\varphi)/(AS) & \vec{Q} & phase($h \nearrow h_{\mathrm{c}0}$) & h_{\mathrm{c}0}(\varphi)/(AS) & \vec{Q} & phase($h \nearrow h_{\mathrm{c}0}$) \\ \hline 0\dots \frac{\pi}{2} & 6 \cos\varphi + 4\sin\varphi & \Gamma & canted N\'eel & 6 \cos\varphi + 4\sin\varphi & \Gamma & canted N\'eel & 6 \cos\varphi + 4\sin\varphi & \Gamma & canted N\'eel \\ \frac{\pi}{2} \dots \varphi_{\mathrm{c}1} & 2 \cos\varphi + 4\sin\varphi & \mathrm{M}_1, \mathrm{M}_3 & canted zigzag & 2 \cos\varphi + 4\sin\varphi & \mathrm{M}_2 & canted zigzag & 3 \cos\varphi + 4\sin\varphi & \mathrm{K} & AF vortex \\ \varphi_{\mathrm{c}1} \dots \varphi_{\mathrm{c}2} & 2 \cos\varphi + 4\sin\varphi & \mathrm{M}_1, \mathrm{M}_3 & canted zigzag & 2 \cos\varphi + 4\sin\varphi & \mathrm{M}_2 & canted zigzag & \multicolumn{3}{c}{\it ---discontinuous $(h_\mathrm{c} > h_{\mathrm{c}0})$---} \\ \varphi_{\mathrm{c}2} \dots \frac{3\pi}{2} & 0 & \Gamma & FM & 0 & \Gamma & FM & 0 & \Gamma & FM \\ \frac{3\pi}{2} \dots \frac{7\pi}{4} & 4 \cos \varphi & \mathrm{M}_1, \mathrm{M}_3 & canted stripy & 4 \cos \varphi & \mathrm{M}_2 & canted stripy & 3 \cos \varphi & \mathrm{K} & vortex \\ \frac{7\pi}{4} \dots \varphi_{\mathrm{c}3} & 4 \cos \varphi & \mathrm{M}_1, \mathrm{M}_3 & canted stripy & 4 \cos \varphi & \mathrm{M}_2 & canted stripy & \multicolumn{3}{c}{\it ---discontinuous $(h_\mathrm{c} > h_{\mathrm{c}0})$---} \\ \varphi_{\mathrm{c}3} \dots \varphi_{\mathrm{c}4} & 4 \cos \varphi & \mathrm{M}_1, \mathrm{M}_3 & canted stripy & 4 \cos \varphi & \mathrm{M}_2 & canted stripy & 6 \cos \varphi + 4 \sin \varphi & \Gamma & canted N\'eel\\ \varphi_{\mathrm{c}4} \dots 2\pi & 6 \cos\varphi + 4\sin\varphi & \Gamma & canted N\'eel & 6 \cos\varphi + 4\sin\varphi & \Gamma & canted N\'eel & 6 \cos\varphi + 4\sin\varphi & \Gamma & canted N\'eel \\ \hline\hline \end{tabular*} \end{table*} \begin{widetext} \begin{align} \vec S_i & = (S - a^\dagger_i a_i) \vec e_3 + \sqrt{\frac{S}{2}} (a_i + a^\dagger_i) \vec e_1 + \ii \sqrt{\frac{S}{2}} (a_i - a^\dagger_i) \vec e_2 + \mathcal O(1/\sqrt{S}), \qquad i \in \mathrm A, \\ \vec S_j & = (S - b^\dagger_j b_j) \vec e_3 + \sqrt{\frac{S}{2}} (b_j + b^\dagger_j) \vec e_1 + \ii \sqrt{\frac{S}{2}} (b_j - b^\dagger_j) \vec e_2 + \mathcal O(1/\sqrt{S}), \qquad j \in \mathrm B, \end{align} The spin-wave Hamiltonian in Fourier space then becomes (up to constant terms) \begin{align} \mathcal H_\mathrm{SW} = S \sum_{\vec q \in \mathrm{BZ}} \biggl\{ \left(\frac{h}{S}-3J-2K\right) \left(a^\dagger_{\vec q} a_{\vec q} + b^\dagger_{\vec q} b_{\vec q}\right) + \lambda_0(\vec q) a^\dagger_{\vec q} b_{\vec q} + \lambda_0^*(\vec q) b^\dagger_{\vec q} a_{\vec q} + \lambda_1(\vec q) a_{-\vec q} b_{\vec q} + \lambda_1^*(-\vec q) a^\dagger_{\vec q} b^\dagger_{-\vec q} \biggr\} + \mathcal{O}(1/S^0), \end{align} with \begin{equation} \lambda_0(\vec q) = \begin{cases} (J + K) \left(\ee^{\ii \vec q \cdot \vec \delta_x} + \ee^{\ii \vec q \cdot \vec \delta_y}\right) + J \ee^{\ii \vec q \cdot \vec \delta_z}, & \quad \text{for } \vec h \parallel [001], \\ \left(J + \frac{K}{2}\right) \left(\ee^{\ii \vec q \cdot \vec \delta_x} + \ee^{\ii \vec q \cdot \vec \delta_y}\right) + (J+K) \ee^{\ii \vec q \cdot \vec \delta_z}, & \quad \text{for } \vec h \parallel [110], \\ \left(J+\frac{2}{3}K\right)\left(\ee^{\ii \vec q \cdot \vec \delta_x} + \ee^{\ii \vec q \cdot \vec \delta_y} + \ee^{\ii \vec q \cdot \vec \delta_z}\right), & \quad \text{for } \vec h \parallel [111], \end{cases} \end{equation} and \begin{equation} \lambda_1(\vec q) = \begin{cases} K \left(\ee^{\ii \vec q \cdot \vec \delta_x} - \ee^{\ii \vec q \cdot \vec \delta_y} \right), & \quad \text{for } \vec h \parallel [001], \\ \frac{K}{2} \left(-\ee^{\ii \vec q \cdot \vec \delta_x} - \ee^{\ii \vec q \cdot \vec \delta_y} + 2\ee^{\ii \vec q \cdot \vec \delta_z} \right), & \quad \text{for } \vec h \parallel [110], \\ \frac{2}{3}K \left( \ee^{\ii \vec q \cdot \vec \delta_x - \frac{2\pi\ii}{3}} + \ee^{\ii \vec q \cdot \vec \delta_y + \frac{2\pi\ii}{3}} + \ee^{\ii \vec q \cdot \vec \delta_z} \right), & \quad \text{for } \vec h \parallel [111]. \end{cases} \end{equation} \end{widetext} Here, $\vec \delta_x$, $\vec \delta_y$, and $\vec \delta_z$ are the nearest-neighbor vectors on $x$, $y$, and $z$ bonds, respectively, of the honeycomb lattice. The leading-order piece of $\mathcal H_\mathrm{SW}$ is quadratic in boson operators and can be diagonalized analytically by means of a Bogoliubov transformation. Resulting magnon spectra with $h \equiv \lvert \vec h \rvert$ tuned to the instability field strength $h_{\mathrm{c}0}$ are depicted for different coupling parameters $\varphi$ and field directions $\hat h \equiv \vec h / h$ in Figs.\ \ref{fig:magnon-spectrum-001}--\ref{fig:magnon-spectrum-1107}. For large $h/S \gg |J|, |K|$ the minimum of the magnon dispersion is always at the $\Gamma$ point in the Brillouin zone. As long as $|K| \ll |J|$ (i.e., when the zero-field ground state is a simple N\'eel or FM state), it remains at the $\Gamma$ point upon decreasing $h \searrow h_{\mathrm{c}0}$ at which it eventually vanishes; cf.\ right panel of Fig.~\ref{fig:magnon-spectrum-111}. Above the zigzag and stripy zero-field ground states, however, the minimum of the dispersion shifts discontinuously from $\Gamma$ towards a finite wavevector as a function of field. For field in the $[001]$ direction ($[110]$ direction) the instability wavevector at which the magnon gap eventually vanishes is at two (one) of the three inequivalent $\mathrm M$ points in the Brillouin zone, indicating a direct continuous transition towards the canted zigzag or stripy phase; see Fig.~\ref{fig:magnon-spectrum-001}. For field in the $[111]$ direction (left and middle panel of Fig.~\ref{fig:magnon-spectrum-111}), by contrast, the instability wavevector is at the $\mathrm K$ points, forbidding a direct continuous transition towards a simple canted deformation of the zero-field ground state. Remarkably, intermediate field directions lead to magnon softening at incommensurate wavevectors, see Fig.~\ref{fig:magnon-spectrum-1107} for field in the $[11 \frac{7}{10}]$ direction; we leave a detailed study of the resulting ordered states for future work. We note that there is a linear band crossing point for field in the $[001]$ direction when $-0.148\pi < \varphi < 0.687\pi$. (An analogous band crossing point occurs when $0.852\pi < \varphi < 1.687\pi$.) This can be understood as a ``Dirac magnon'' that is located at the K point (and finite energy) in the Heisenberg limit $\varphi = 0$, and shifted from K towards the $\Gamma$ point (M$_2$ point) for finite $\varphi>0$ (finite $\varphi < 0$).\cite{magnonspectrum} Another such bosonic Dirac point is located at the opposite [with respect to the $\Gamma$ point (M$_2$ point) for $\varphi > 0$ ($\varphi < 0$)] wavevector. At $\varphi = 0.687\pi$ ($\varphi = -0.148\pi$) both merge and annihilate at the $\Gamma$ point (M$_2$ point). Explicit values for the instability field strength $h_{\mathrm{c}0}$ and corresponding instability wavevectors are given in Table~\ref{tab:instability-field}. There, we have also indicated the special cases when the instability of the high-field magnon is preempted by a discontinuous transition, as obtained from the analytical parameterization of phases (Sec.~\ref{sec:parametrization}). In all other cases, the transition from polarized towards intermediate-field phases is continuous, and we have checked that the instability field strength $h_{\mathrm{c}0}$ indeed then always coincides with the critical field strength $h_\mathrm{c}$ as obtained from the parametrization (Sec.~\ref{sec:parametrization}), as well as with $h_\mathrm{c}$ from the MC data (Sec.~\ref{sec:mc}). This also serves as an independent verification of the numerics. We also note that the magnetization process in the HK model on the 3D hyperhoneycomb lattice in $[111]$ field appears to be similarly complex as found here. This is because the magnon instability at $h_{\mathrm{c}0}$ (which happens to coincide with $h_{\mathrm{c}0}$ for the 2D honeycomb lattice) occurs \emph{above} the metamagnetic first-order transitions found in the MC simulations. This has apparently been overlooked in the previous analysis.\cite{lee2014} \section{Monte-Carlo simulations}\label{sec:mc} To identify the intermediate-field phases, we study the large-$S$ limit of the HK model by employing a combination of classical Monte Carlo (MC) simulations and energy minimization. We work on honeycomb lattices of size $L\times L$ with periodic boundary conditions. The lattices are spanned by the primitive lattice vectors $\vec{a}_{1\left(2\right)}=\left(3/2,\pm\sqrt{3}/2\right)$, with each unit cell containing two sites amounting to a total number of spins of $N=2L^{2}$. We perform equilibrium MC simulations using single-site updates with a combination of the heat-bath and microcanonical (or over-relaxation) algorithms,\cite{berg_mc} with typically $10^{7}$ MC steps per spin. We combine these updates with the parallel-tempering algorithm \cite{partemp_jpsj} in order to efficiently equilibrate the MC configurations at very low $T$. From the MC data, we compute the uniform magnetization in the field direction (Fig.~4\ in the main text) \begin{equation}\label{eq:mag} \frac{\vec m \cdot \hat h}{S}=\left \langle \frac{1}{N}\sum_{i}\frac{\vec{S}_{i}}{S}\cdot\frac{\vec h}{\lvert \vec h \rvert}\right\rangle, \end{equation} where $\langle \cdots \rangle $ denotes MC average, as well as the static spin structure factor (Fig.~2\ in the main text) \begin{equation}\label{eq:sfactor} S_{\vec{k}} = \left\langle \vec{S}(\vec{k})\cdot \vec{S}(-\vec{k})\right\rangle, \end{equation} where \begin{equation}\label{eq:sfourier} \vec{S}(\vec{k}) = \frac{1}{\sqrt{N}}\sum_{i}\vec{S}_{i}e^{-i\vec{k}\cdot\vec{R}_{i}}, \end{equation} is the Fourier transform of a given spin configuration and $\vec R_i$ is the lattice vector at site $i$. To find the classical ground state, we start from a MC spin configuration obtained at low $T$ (typically $T/\lvert J S^2 \rvert \sim 0.005$) and then iteratively align the spins with their local fields\cite{hloc} $\vec{h}_{i}^\text{loc}$, \begin{equation}\label{eq:min} \vec{S}_{i} = \frac{\vec{h}_{i}^\text{loc}}{\left|\vec{h}_{i}^\text{loc}\right|} S. \end{equation} Convergence is reached after the largest update in a lattice sweep, $|\vec{S}_{i}^\text{new}-\vec{S}_{i}^\text{old}|_\text{max}/S$, is smaller than $10^{-12}$. Because of the several competing ground states, it is important to start from unbiased MC configurations in order to obtain the correct classical ground state. \begin{figure*}[btp] \includegraphics[width=\textwidth]{spin-configurations-mc.pdf} \caption{Real-space spin configuration snapshots from cooled MC data, projected onto the plane perpendicular to $\vec h \parallel [111]$, for $N=2 \cdot 18^2$ spins. The colors of the arrows refer to the spin directions in this perpendicular plane as a code to guide the eye (see inset); the arrows' lengths refer to the relative spin magnitudes in this plane (cf.\ the spin projections in Fig.~\ref{fig:spin-configs}). } \label{fig:mc-snapshots} \end{figure*} We performed extensive field scans at $\varphi = 0.57\pi$, $0.62\pi$, $0.733\pi$, $0.83\pi$, $1.578\pi$, $1.687\pi$, $1.813\pi$, and $1.922\pi$ for the $[111]$ field direction, $\varphi = 0.62\pi$ and $1.687\pi$ for the $[001]$ direction, as well as $\varphi = 1.687\pi$ for the $[110]$ field direction, with system sizes up to $L=24$. For field in the $[111]$ direction we find a total of 10 phases at finite $h$. Spin-configuration snapshots from cooled MC data are depicted for selected parameter values in Fig.~\ref{fig:mc-snapshots}. \section{Parametrization of phases}\label{sec:parametrization} The low-$T$ spin configurations obtained from the MC simulations allows the deduction of the symmetries, unit-cell sizes,\cite{largecell} and sublattice structure of the different phases. We make use of this information by parametrizing the spin configurations in terms of a set of angles, which then are optimized at fixed model parameters $\varphi$ and $\vec h$ to determine the state of lowest energy. Doing this for all phases enables a comparison of energies from which we deduce the classical phase diagram in the low-temperature limit. \begin{table*}[p!] \caption{Ans\"atze for angles $\phi_i$ and $\theta_i$ in Eq.~\eqref{eq:spin-ansatz} for parametrization of spin $\vec S_i$ at lattice site $\vec R_i$ within one magnetic unit cell. Here, $\vec R_i$ is measured in units of the lattice constant from the center of the first hexagon.} \label{tab:angles} \begin{minipage}[t]{\columnwidth} \begingroup \renewcommand{\arraystretch}{1.077} \mbox{}\\[-\baselineskip] \begin{tabular*}{\columnwidth}{@{\extracolsep{\fill}} lclll} \hline\hline Phase & $i$ & $\vec R_i$ & $\phi_i$ & $\theta_i$ \\ \hline polarized & 1 & $(1,0)$ & 0 & 0 \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & 0 & 0 \\ canted zigzag & 1 & $(1,0)$ & $\pi$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\pi$ & $\theta$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $0$ & $\theta'$ \\ & 4 & $(-1,0)$ & $0$ & $\theta'$ \\ AF star & 1 & $(1,0)$ & $0$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\frac{\pi}{3}$ & $\theta'$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $\frac{2\pi}{3}$ & $\theta$ \\ & 4 & $(-1,0)$ & $\pi$ & $\theta'$ \\ & 5 & $(\cos \frac{4\pi}{3},\sin \frac{4\pi}{3})$ & $\frac{4\pi}{3}$ & $\theta$ \\ & 6 & $(\cos \frac{5\pi}{3},\sin \frac{5\pi}{3})$ & $\frac{5\pi}{3}$ & $\theta'$ \\ & 7 & $(2,0)$ & $0$ & $\pi$ \\ & 8 & $(2\cos \frac{\pi}{3}, 2\sin \frac{\pi}{3})$ & $0$ & $0$ \\ AF vortex & 1 & $(1,0)$ & $\frac{2\pi}{3} - \delta$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\frac{5\pi}{3} + \delta$ & $\theta$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $\frac{4\pi}{3}-\delta$ & $\theta$ \\ & 4 & $(-1,0)$ & $\frac{\pi}{3}+\delta$ & $\theta$ \\ & 5 & $(\cos \frac{4\pi}{3},\sin \frac{4\pi}{3})$ & $-\delta$ & $\theta$ \\ & 6 & $(\cos \frac{5\pi}{3},\sin \frac{5\pi}{3})$ & $\delta$ & $\theta$ \\ vortex & 1 & $(1,0)$ & $\frac{5\pi}{3} - \delta$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\frac{5\pi}{3} + \delta$ & $\theta$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $\frac{\pi}{3}-\delta$ & $\theta$ \\ & 4 & $(-1,0)$ & $\frac{\pi}{3}+\delta$ & $\theta$ \\ & 5 & $(\cos \frac{4\pi}{3},\sin \frac{4\pi}{3})$ & $\pi-\delta$ & $\theta$ \\ & 6 & $(\cos \frac{5\pi}{3},\sin \frac{5\pi}{3})$ & $\pi+\delta$ & $\theta$ \\ zigzag star & $1, \dots, 9$ & $(\frac{3i-1}{2},\frac{\sqrt{3}(i-1)}{2})$ & $\phi_i$ & $\theta_i$ \\ & $10, \dots, 18$ & $(\frac{3i-29}{2},\frac{\sqrt{3}(i-9)}{2})$ & $\phi_i$ & $\theta_i$ \\ & $19, \dots, 27$ & $(\frac{3i-55}{2},\frac{\sqrt{3}(i-17)}{2})$ & $\phi_i$ & $\theta_i$ \\ & $28, \dots, 36$ & $(\frac{3i-83}{2},\frac{\sqrt{3}(i-25)}{2})$ & $\phi_i$ & $\theta_i$ \\ \hline\hline \end{tabular*}\endgroup \end{minipage}\hfill \begin{minipage}[t]{\columnwidth} \mbox{}\\[-\baselineskip] \begin{tabular*}{\columnwidth}{@{\extracolsep{\fill}} lclll} \hline\hline Phase & $i$ & $\vec R_i$ & $\phi_i$ & $\theta_i$ \\ \hline canted N\'eel & 1 & $(1,0)$ & 0 & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\pi$ & $\theta$ \\ canted stripy & 1 & $(1,0)$ & $\pi$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $0$ & $\theta'$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $0$ & $\theta'$ \\ & 4 & $(-1,0)$ & $\pi$ & $\theta$ \\ FM star & 1 & $(1,0)$ & $\pi$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\frac{\pi}{3}$ & $\theta$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $\frac{5\pi}{3}$ & $\theta$ \\ & 4 & $(-1,0)$ & $\pi$ & $\theta$ \\ & 5 & $(\cos \frac{4\pi}{3},\sin \frac{4\pi}{3})$ & $\frac{\pi}{3}$ & $\theta$ \\ & 6 & $(\cos \frac{5\pi}{3},\sin \frac{5\pi}{3})$ & $\frac{5\pi}{3}$ & $\theta$ \\ & 7 & $(2,0)$ & $0$ & $0$ \\ & 8 & $(2\cos \frac{\pi}{3}, 2\sin \frac{\pi}{3})$ & $0$ & $0$ \\ diluted star & 1 & $(1,0)$ & $\pi$ & $\theta$ \\ & 2 & $(\cos \frac{\pi}{3},\sin \frac{\pi}{3})$ & $\frac{\pi}{3}$ & $\theta$ \\ & 3 & $(\cos \frac{2\pi}{3},\sin \frac{2\pi}{3})$ & $\frac{5\pi}{3}$ & $\theta$ \\ & 4 & $(-1,0)$ & $\pi$ & $\theta$ \\ & 5 & $(\cos \frac{4\pi}{3},\sin \frac{4\pi}{3})$ & $\frac{\pi}{3}$ & $\theta$ \\ & 6 & $(\cos \frac{5\pi}{3},\sin \frac{5\pi}{3})$ & $\frac{5\pi}{3}$ & $\theta$ \\ & 7 & $(2,0)$ & $0$ & $\theta$ \\ & 8 & $(2\cos \frac{\pi}{3}, 2\sin \frac{\pi}{3})$ & $\frac{4\pi}{3}$ & $\theta$ \\ & 9 & $(2\cos \frac{2\pi}{3}, 2\sin \frac{2\pi}{3})$ & $\frac{2\pi}{3}$ & $\theta$ \\ & 10 & $(-2,0)$ & $0$ & $\theta$ \\ & 11 & $(2\cos \frac{4\pi}{3}, 2\sin \frac{4\pi}{3})$ & $\frac{4\pi}{3}$ & $\theta$ \\ & 12 & $(2\cos \frac{5\pi}{3}, 2\sin \frac{5\pi}{3})$ & $\frac{2\pi}{3}$ & $\theta$ \\ & 13 & $(2,-\sqrt{3})$ & $\frac{4\pi}{3}$ & $\theta'$\\ & 14 & $(\frac{5}{2},-\frac{\sqrt{3}}{2})$ & $\frac{4\pi}{3}$ & $\theta'$ \\ & 15 & $(\frac{5}{2},\frac{\sqrt{3}}{2})$ & $\frac{2\pi}{3}$ & $\theta'$ \\ & 16 & $(2,\sqrt{3})$ & $\frac{2\pi}{3}$ & $\theta'$ \\ & 17 & $(\frac{1}{2},\frac{3\sqrt{3}}{2})$ & $0$ & $\theta'$ \\ & 18 & $(-\frac{1}{2},\frac{3\sqrt{3}}{2})$ & $0$ & $\theta'$ \\ \hline\hline \end{tabular*} \end{minipage} \end{table*} Using the rotated basis as defined in Eq.~\eqref{eq:spin-frame}, the spin $\vec S_i$ at site $i$ can be parametrized as \begin{equation} \label{eq:spin-ansatz} \vec S_i = S \left( \vec e_1 \sin \theta_i \cos \phi_i + \vec e_2 \sin \theta_i \sin \phi_i + \vec e_3 \cos \theta_i \right). \end{equation} In the polarized phase we have $\theta_i \equiv 0$, while $\theta_i > 0$ defines a canted state. For given coupling parameter $\varphi$ and magnetic field $\vec h = h \vec e_3$ our ans\"atze for the angles $\theta_i$ and $\phi_i$ as obtained from the MC simulations are given in Table \ref{tab:angles}. Except for the vortex, AF vortex, and zigzag star phases the spin projections onto the plane perpendicular to $\vec h$ ($\vec e_1$-$\vec e_2$ plane) are locked on the directions of the cubic-axes projections $\vec e_3 \times (\vec e_x \times \vec e_3)$, $\vec e_3 \times (\vec e_y \times \vec e_3)$, and $\vec e_3 \times (\vec e_z \times \vec e_3)$, see Fig.~\ref{fig:mc-snapshots}. For these phases we therefore have $\phi_i \in \{0, \frac{\pi}{3}, \frac{2\pi}{3}, \pi, \frac{4\pi}{3}, \frac{5\pi}{3}\}$, and we may minimize with respect to the field-dependent canting angles $\theta_i$ only. In each case we in fact find that there are at most only two different possible $\theta$ angles [indicated by the at most two different lengths of the spin projections in Fig.~\ref{fig:mc-snapshots}(a), (d), and (e)]. This makes the computation of the minimized energy of a given classical state and their comparison among different states numerically cheap. In the cases of the vortex and AF vortex phases the $\phi$ angles are not locked onto the projection of the cubic-axes direction. However, we find that the classical energy in these cases in fact becomes independent of the angle $\delta$ that determines the (uniform) deviation from the cubic-axes locking. (The MC data show that thermal fluctuations lift this degeneracy by an order-from-disorder mechanism.) By contrast, for the zigzag star phase we do not use any particular ansatz for the configuration, except for the fact (again as obtained from the cooled MC data) that the magnetic unit cell spans $2 \times 9$ crystallographic unit cells.\cite{largecell} The explicit assumptions for $\theta_i$ and $\phi_i$ for all states are summarized in Table~\ref{tab:angles}. \begin{figure*}[!p] \includegraphics[width=\textwidth]{spin-configurations-ana-suppl.pdf} \caption{Real-space spin configurations from analytical parametrization for various states. Here we have aligned the $[111]$ spin-space axis perpendicular to the real-space lattice. Dashed: magnetic unit cell. Gray lattices: spin projections onto the plane perpendicular to $\vec h$. The colors of the arrows refer to the spin directions in this perpendicular plane (see inset of Fig.~\ref{fig:mc-snapshots}). } \label{fig:spin-configs} \end{figure*} \begin{figure*}[!p] \includegraphics[width=0.45\linewidth]{strctr-fctr_ana_phi062-phi083.pdf} \caption{ Single-domain static spin structure factor from analytical parametrization for different values of $\varphi$ and $\vec h \parallel [111]$, allowing us to distinguish between (a) the single-$Q$ order of the canted zigzag phase and (b) the triple-$Q$ AF star pattern. The zigzag star phase (c) has a total of 18 inequivalent Bragg peaks within the first Brillouin zone. Averaging over the six symmetry-related ground states, obtained by $2\pi/3$ rotation and inversion, yields the hexagram pattern as seen in the MC spin structure, Fig.~{2}(b) in the main text. The inner white dashed hexagon indicates the location of the first Brillouin zone of the honeycomb lattice. In all cases, the signal at $\Gamma$ arises from the uniform magnetization component in field direction. For visualization purposes, we have replaced the $\delta$ peaks of the infinite-size system by finite-width Gaussian distributions. } \label{fig:strctr-fctr_ana} \end{figure*} The parametrization allows the straightforward comparison of the minimized energies of the various states and the deduction of the phase boundaries for arbitrary coupling parameter $\varphi$ and field strength $h$ under the assumption that no further states (not parametrized in Table~\ref{tab:angles}, and missed by the MC scans) are stabilized somewhere in the phase diagram. The result is depicted in Fig.~1\ in the main text. In addition to the quadruple point at $(\varphi, h/(AS)) = (7\pi/4, 3/\sqrt{2})$ we find 8 triple points at ($\pi/2$, 4), ($0.55\pi$, 2.17), ($0.698\pi$, 1.38), ($0.715\pi$, 1.25), ($0.722\pi$, 1.01), ($1.673\pi$, 1.35), ($1.812\pi$, 2.76), and ($1.825\pi$, 2.62). The total magnetization in field direction $\hat h = \vec h / \lvert \vec h \rvert$ is given by $\vec m \cdot \hat h/S = N^{-1} \sum_i \cos \theta_i = -N^{-1} (\partial E/\partial h)$ with $E\equiv E(\varphi,h)$ as the ground-state energy for given $\varphi$ and $h$. The magnetization curves agree very well with the MC measurements, see Fig.~4\ in the main text. Exceptions are a few data points very close to first-order transitions; we attribute these deviations to hysteresis effects in the MC simulations. We have explicitly checked that the minimized energy from the analytical parametrization is always less than or equal the one from the cooled MC configuration for the same parameters. We visualize several magnetic unit cells of the spin configurations for all canted phases in Fig.~\ref{fig:spin-configs}. Fig.~3\ in the main text analogously shows one respective magnetic unit cell for selected phases. In Fig.~\ref{fig:spin-configs}, we display also the projections of the spin configurations onto the plane perpendicular to $\vec h$ (to be compared with the cooled MC spin configurations in Fig.~\ref{fig:mc-snapshots}). We can also use the parametrized spin configurations to compute static spin structure factors, allowing a comparison with the MC structure factors (Fig.~2\ in the main text). However, an efficient MC simulation (in our case with parallel tempering) averages over the full ground-state manifold. It consequently does not allow the direct distinction between single- and multi-$Q$ states. For example, while a pure ``$+z$ zigzag'' state with the spins of a particular zigzag line on the honeycomb lattice pointing along the $+z$ direction would exhibit a Bragg peak at only one out of the three inequivalent M points in the first Brillouin zone (M$_2$), the simulations always average over $\pm x$, $\pm y$, and $\pm z$ zigzag states (as long as these are degenerate), and the MC structure factors exhibit Bragg peaks at all three M points. Experimentally, this is equivalent to having multiple magnetic domains in a large sample. Using the analytical parametrization, by contrast, we can compute ``single-domain'' structure factors for \emph{fixed} states without averaging over the ground-state manifold, allowing us to distinguish between single-$Q$ and multi-$Q$ phases in a direct way. In Fig.~\ref{fig:strctr-fctr_ana} we show examples for the canted zigzag phase with a Bragg peak at only one out of the three M points in the first Brillouin zone~(a), to be compared with the AF star phase which exhibits Braggs peaks at all three M points~(b). Fig.~\ref{fig:strctr-fctr_ana}(c) shows the single-domain structure factor of the zigzag star phase, with a total of 18 inequivalent Bragg peaks in the first Brillouin zone, to be compared with the MC averaged structure factor of Fig.~2 (b) in the main text. \section{Klein duality and star vs. zigzag/stripy phases} We explain how the fact that the AF star and zigzag states, and analogously the FM star and stripy states, are classically degenerate for all $\varphi$ can be understood in terms of the Klein duality~\cite{Cha10, Cha13, kimchi2014}. This will also allow us to gain useful insight into the quantum-fluctuation effects on the phase diagram for $S=1/2$. We introduce the dual spins $\vec{S}_i'$ by dividing the honeycomb lattice into four sublattices $\mathrm A$, $\mathrm B$, $\mathrm C$, $\mathrm D$ and identifying~\cite{Cha10} \begin{equation} \label{eq:duality} \vec{S}_i' \equiv \begin{cases} \vec S_i & \text{for } i \in \mathrm A,\\ \diag(1,-1,-1)\, \vec S_i & \text{for } i \in \mathrm B,\\ \diag(-1,1,-1)\, \vec S_i & \text{for } i \in \mathrm C,\\ \diag(-1,-1,1)\,\vec S_i & \text{for } i \in \mathrm D. \end{cases} \end{equation} In terms of the dual spins the Heisenberg-Kitaev Hamiltonian [Eq.~(1)\ in the main text] can be written as \begin{equation} \label{eq:hk-dual} \mathcal{H} = - J \sum_{\left\langle ij\right\rangle} \vec{S}_{i}' \cdot \vec{S}_{j}' + 2(K+J) \sum_{\left\langle ij\right\rangle_{\gamma}} S_{i}^{\gamma}{}' S_{j}^{\gamma}{}' - \sum_{i} \vec h_i' \cdot \vec{S}_i'. \end{equation} with $\vec h_i'$ denoting the dual magnetic field, obtained by a duality transformation that is analogous to the spin transformation in Eq.~\eqref{eq:duality}. Eq.~\eqref{eq:hk-dual} describes a Heisenberg-Kitaev model in a nonuniform field $\vec{h}_i'$. For $\vec{h}_i' = 0$ and $K=-J$, i.e., $\varphi \in \{3\pi/4, 7\pi/4\}$, it features a spin $\mathrm{SU}(2)$ symmetry that is hidden in the original basis.\cite{Cha10} For finite field, a $\mathrm U(1)$ part of the hidden symmetry is left intact, if and only if $\vec h$ points along one of the cubic axes $\vec e_x$, $\vec e_y$, or $\vec e_z$ (and thus $\vec{h}_i'$ is parallel or antiparallel to this axis). For other field directions, no continuous spin symmetry remains at finite $\vec h$. \begin{figure*}[t!] \includegraphics[width=\textwidth]{magnetization_LSWT_111.pdf} \caption{ Total magnetization in field direction as function of applied field for $S=1/2$ in the high-field phase, $h>h_{\mathrm c}(\varphi)$, for different fixed values of $\varphi$ (a) from $\pi/2$ (red) to $0.85\pi$ (violet) above the zigzag zero-field ground state and (b) from $3\pi/2$ (red) to $1.85\pi$ (violet) above the stripy zero-field ground state. Black lines: magnetization at critical field $h_{\mathrm c}(\varphi)$. The magnetization vanishes above $h_{\mathrm c}$ for (a) $0.49\pi<\varphi<0.54\pi$ as well as for (b) $1.49\pi<\varphi<1.54\pi$, indicating strong quantum fluctuations and the breakdown of the semiclassical approximation. For $\varphi = \pi/2$ ($\varphi = 3\pi/2$) the magnetization vanishes at $h_{\mathrm{c,QSL}} \sim 4.10 AS$ ($h_{\mathrm{c,QSL}} \sim 0.053 AS$), at which one might expect a transition towards a topologically ordered spin liquid. The dashed lines for $0.72\pi < \varphi < 0.85\pi$ and $1.75\pi < \varphi < 1.81\pi$ denote the magnetization in the metastable high-field state when the magnon instability at $h_{\mathrm{c}0}(\varphi)$ is preempted by a first-order transition at $h_\mathrm{c} > h_{\mathrm{c}0}$; here the dashed black line denotes the magnetization at $h_{\mathrm{c}0}(\varphi)$ and the thick black line the magnetization at $h_\mathrm{c}(\varphi)$. } \label{fig:magnetization-111} \end{figure*} Consider the exactly solvable ``stripy Klein point'' for $K=-J < 0$ (i.e., $\varphi = 7\pi/4$) and $\vec h = 0$. The quantum ground state is a ferromagnet with the dual spins pointing along a fixed, but arbitrary direction $\vec{S}_i' = S \vec n$ in spin space. Only the six states with $\vec n \in \{\pm \vec e_x, \pm \vec e_y, \pm \vec e_z\}$ out of this $\mathrm{SU}(2)$ degenerate ground-state manifold correspond to collinear spin configurations in the original basis. These are precisely the six possible stripy quantum ground states of the Heisenberg-Kitaev model. States for which only one (no) component of $\vec n$ in the cubic-axes basis vanishes correspond to coplanar (noncoplanar) spin textures in the original basis. Due to the hidden $\mathrm{SU}(2)$ symmetry an order-from-disorder mechanism can lift this quantum-ground-state degeneracy only away from the Klein point, e.g., when we consider a different set of couplings with $\varphi \notin \{3\pi/4, 7\pi/4\}$ or switch on an external field $\vec h \neq 0$. In fact, these states belong to the highly-degenerate ground-state manifold of the classical Kitaev model,\cite{baskaran2008} and thus have the same \emph{classical} energy for all~$\varphi$. For $7\pi/4 \leq \varphi < 1.85\pi$, we find classically that an infinitesimally small field $\vec h \parallel [111]$ lifts the degeneracy in favor of a state in which also $\vec n \parallel [111]$. In the original spin basis this state corresponds to the FM star configuration. Upon inclusion of quantum fluctuations, one may expect that an order-from-disorder mechanism will shift the phase boundary between the stripy phase and FM star phase from zero field for $S \to \infty$ to finite values of the field for $S = 1/2$ if $\varphi > 7\pi/4$. Directly at $\varphi = 7\pi/4$, however, the degeneracy survives in the quantum case because of the presence of the hidden $\mathrm{SU}(2)$ symmetry. We infer (in the sense of degenerate perturbation theory in small $h$) that the FM star phase reaches all the way down to $h \searrow 0$, with the triple point at $(\varphi,h) = (7\pi/4,0)$ staying at zero field also for $S=1/2$. Note that the above argument does not rely on the fact that quantum fluctuations are absent in the zero-field ground state of the dual FM model. An analogous mechanism should therefore be expected at the ``zigzag Klein point'' for $K=-J > 0$, i.e., $\varphi = 3\pi/4$. Here, the zero-field ground state in the dual basis is a N\'eel antiferromagnet with $\vec{S}_i' = (-1)^i S \vec n$ along an arbitrary direction~$\vec n$. $\vec n \in \{\pm \vec e_x, \pm \vec e_y, \pm \vec e_z\}$ corresponds to one of the six possible zigzag states in the original basis. $\vec n \parallel [111]$ corresponds to the noncoplanar AF star phase. Again, we find that a finite $\vec h \parallel [111]$ lifts the degeneracy in favor of the state with $\vec n \parallel [111]$. Due to the absence of an order-from-disorder mechanism in the hidden-$\mathrm{SU}(2)$-symmetric model when $\varphi = 3\pi/4$ and $\vec h=0$, we expect that the degenerate zero-field ground state gives way to an AF star ground state at infinitesimal field in the $[111]$ direction also in the quantum limit when $S=1/2$. We conclude that finite regions of both FM star and AF star phases exist for field in the $[111]$ direction not only classically, but also in the quantum phase diagram for $S=1/2$, at least in the vicinity of the Klein points at $\varphi = 3\pi/4$ or $7\pi/4$.% \cite{noteduality} \section{High-field magnetization for $S=1/2$} \begin{figure*}[!t] \includegraphics[width=\textwidth]{magnetization_LSWT_001.pdf} \caption{Same as Fig.~\ref{fig:magnetization-111} for $\vec h \parallel [001]$. For $\varphi = \pi/2$ ($\varphi = 3\pi/2$) the magnetization now vanishes at $h_{\mathrm{c,QSL}} \sim 4.03 AS$ ($h_{\mathrm{c,QSL}} \sim 0.032 AS$). It also vanishes for $\varphi = 1.852 \pi$ and $h = h_\mathrm{c} = 3.58 AS$, when the instability wavevector of the high-field magnon spectrum changes between $\vec Q = \Gamma$ and $\vec Q = \mathrm M$. Here, no metastable states exist since the transition from the polarized phase towards a canted phase is always continuous for any $\varphi$. } \label{fig:magnetization-001} \end{figure*} In the polarized phase, the influence of quantum fluctuations for $S=1/2$ may be estimated by again employing spin-wave theory. To this end, we compute the $1/S$ correction to the total magnetization in field direction: \begin{align} \frac{\vec m \cdot \hat h}{S} & = \frac{1}{N} \left[\sum_{i \in \mathrm A}\left(1-\frac{1}{S} \langle a_i^\dagger a_i \rangle \right) + \sum_{j \in \mathrm B}\left(1-\frac{1}{S} \langle b_j^\dagger b_j \rangle \right) \right] \nonumber \\ & \quad + \mathcal O(1/S^2), \end{align} where $\langle a_i^\dagger a_i \rangle$ and $\langle b_j^\dagger b_j \rangle$ are the magnon densities at lattice sites $i \in \mathrm A$ and $j \in \mathrm B$, respectively. The magnetization curves in linear spin-wave theory for $S=1/2$ as function of $h>h_{\mathrm c}$ in the polarized phase are depicted for selected values of $\varphi$ and $\vec h \parallel [111]$ in Fig.~4\ in the main text. Analogous curves for full ranges of $\varphi$ in the polarized phase above the zigzag and stripy zero-field ground states are given in Fig.~\ref{fig:magnetization-111} for $\vec h \parallel [111]$ and Fig.~\ref{fig:magnetization-001} for $\vec h \parallel [001]$. In the nonfrustrated FM and AF Heisenberg cases the corrections vanish, $\langle a_i^\dagger a_i \rangle = \langle b_j^\dagger b_j \rangle = 0$, but they become enhanced by increasing the Kitaev exchange $K$. Consider the limit $|K|\gg|J|$: Here the zero-field state for $S=1/2$ is a gapless $\mathbb{Z}_2$ spin liquid, and it is known \cite{kitaev06, jiang2011} that this state is unstable towards a gapped topologically ordered spin liquid for infinitesimal field in the $[111]$ direction. Increasing $h$ eventually drives a transition towards the polarized phase at some finite $h = h_\mathrm{c,QSL} > 0$. While a quantitative analysis of this topological quantum transition is beyond the realm of linear spin-wave theory, a simple estimate for the transition points may be obtained by computing the parameter sets $(\varphi, h)$ at which the magnetization (to first order in $1/S$) vanishes. This way, we find, e.g., for $\varphi = 3\pi/2$ (FM Kitaev model) the critical field strength as $h_\mathrm{c,QSL}/(AS) \sim 0.053$, which is in about $30\%$ agreement with the value from density-matrix renormalization group calculations.\cite{jiang2011} For $\varphi = \pi/2$ (AF Kitaev model) we find a significantly higher estimate of $h_\mathrm{c,QSL} / (AS) \sim 4.10$, i.e., the spin liquid in the AF Kitaev model is much more stable against uniform applied field as compared to the FM Kitaev model. For intermediate $|K| \sim |J|$ (e.g., near the ``Klein'' points at $\varphi = 3\pi/4$ and $7\pi/4$) the magnetization in the polarized phase is finite for all $h > h_\mathrm{c}(\varphi)$, with the leading-order correction to the saturated magnetization of the order of $50\%$. (The exception is $\varphi = 1.852\pi$ for field in the $[001]$ direction, when the instability wavevector changes from $\Gamma$ to M, the lower magnon band becomes flat with $\varepsilon_{\vec q} = 0$ between $\Gamma$ and M for $h \searrow h_\mathrm{c}$, and the leading-order magnetization correction diverges.) For the case of possible experimental relevance, $\varphi \sim (0.6\dots0.72) \pi$, we hence expect that the gapped high-field phase is reached at a magnetization of about half the saturation magnetization.
185e9f90d93d078196d46d0c79174714237f3669
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The black hole information puzzle is the puzzle of whether black hole formation and evaporation is unitary, and debate on this issue has continued for more than 36 years \cite{Page:1993up, Giddings:2006sj, Mathur:2008wi}, since Hawking radiation was discovered \cite{Hawking:1974sw}. Hawking originally used local quantum field theory in the semiclassical spacetime background of an evaporating black hole to deduce \cite{Hawking:1976ra} that part of the information about the initial quantum state would be destroyed or leave our Universe at the singularity or quantum gravity region at or near the centre of the black hole, so that what remained outside after the black hole evaporated would not be given by unitary evolution from the initial state. However, this approach does not fully apply quantum theory to the gravitational field itself, so it was objected that the information-loss conclusion drawn from it might not apply in quantum gravity \cite{Page:1979tc}. Maldacena's AdS/CFT conjecture \cite{Maldacena:1997re} has perhaps provided the greatest impetus for the view that quantum gravity should be unitary within our Universe and give no loss of information. If one believes in local quantum field theory outside a black hole and also that one would not experience extreme harmful conditions (`drama') immediately upon falling into any black hole sufficiently large that the curvature at the surface would not be expected to be dangerous, then recent papers by Almheiri, Marolf, Polchinski, and Sully (AMPS) \cite{Almheiri:2012rt}, and by them and Stanford (AMPSS) \cite{Almheiri:2013hfa}, give a new challenge to unitarity, as they argued that unitarity, locality, and no drama are mutually inconsistent. It seems to us that locality is the most dubious of these three assumptions. Nevertheless, locality seems to be such a good approximation experimentally that we would like a much better understanding of how its violation in quantum gravity might be able to preserve unitarity and yet not lead to the drama of firewalls or to violations of locality so strong that they would be inconsistent with our observations. Giddings (occasionally with collaborators) has perhaps done the most to investigate unitary nonlocal models for quantum gravity \cite{Giddings:2006sj, Giddings:2006be, Giddings:2007ie, Giddings:2007pj, Giddings:2009ae, Giddings:2011ks, Giddings:2012bm, Giddings:2012dh, Giddings:2012gc, Giddings:2013kcj, Giddings:2013jra, Giddings:2013noa, Giddings:2014nla, Giddings:2014ova, Giddings:2015uzr, Donnelly:2016rvo, Giddings:2017mym, Donnelly:2017jcd}. For other black hole qubit models, see \cite{Terno:2005ff, Levay:2006pt, Levay:2007nm, Duff:2008eei, Levay:2008mi, Borsten:2008wd, Rubens:2009zz, Levay:2010ua, Duff:2010zz, Duff:2012nd, Borsten:2011is, Levay:2011bq, Avery:2011nb, Dvali:2011aa, Borsten:2012sga, Borsten:2012fx, Dvali:2012en, Duff:2013xna, Levay:2013epa, Verlinde:2013vja, Borsten:2013vea, Duff:2013rma, Borsten:2013uma, Dvali:2013lva, Prudencio:2014ypa, Pramodh:2014jha, Chatwin-Davies:2015hna, Dai:2015dqt, Belhaj:2016yyq, Belhaj:2016yfo}. Here we present a qubit toy model for how a black hole might evaporate unitarily and without firewalls, but with nonlocal gravitational degrees of freedom. We model radiation modes emitted by a black hole as localized qubits that interact locally with these nonlocal gravitational degrees of freedom. Similar models were first investigated by Giddings in his previously referred papers, particularly in \cite{Giddings:2011ks,Giddings:2012bm,Giddings:2012dh}. Nomura and his colleagues also have a model \cite{Nomura:2014woa,Nomura:2014voa,Nomura:2016qum} with some similarities to ours. In this way we can go from modes near the horizon that to an infalling observer appear to be close to a vacuum state (and hence without a firewall), and yet the modes that propagate outward can pick up information from the nonlocal gravitational field they pass through so that they transfer that information out from the black hole. \section{Qualitative Description of Our Qubit Model} Using Planck units in which $\hbar = c = G = k_\mathrm{Boltzmann} = 1$, a black hole that forms of area $A$ and Bekenstein-Hawking entropy $S_\mathrm{BH} = A/4$ may be considered to have $e^{S_\mathrm{BH}} = 2^{S_\mathrm{BH}/(\ln{2})}$ orthonormal states, which is the same number as the number of orthonormal states of $n = S_\mathrm{BH}/(\ln{2}) = A/(4\ln{2})$ qubits if this is an integer, which for simplicity we shall assume. We shall take the state of these $n$ qubits as being the state of the gravitational field of the black hole. We assume that this state is rapidly scrambled by highly complex unitary transformations, so that generically a black hole formed by collapse, even if it is initially in a pure state, will have these $n$ qubits highly entangled with each other. However, in our model we shall assume that there are an additional $n$ qubits of outgoing radiation modes just outside the horizon, and a third set of $n$ qubits of outgoing but infalling radiation modes just inside the horizon. We shall assume that these two sets of qubits have a unique pairing (as partner modes in the beginning of the Hawking radiation) and further that each pair is in the singlet Bell state that we shall take to represent the vacuum state as seen by an infalling observer, so that all of these $2n$ qubits of radiation modes near the black hole horizon are in the vacuum pure state and hence give no contribution to the Bekenstein-Hawking entropy $S_\mathrm{BH} = n\ln{2}$. We thus explicitly assume that the infalling observer sees only the vacuum and no firewall in crossing the event horizon. See \cite{Page:2013mqa} for one argument for justifying this assumption. Now we assume that the Hawking emission of one mode corresponds to one of the $n$ outgoing radiation modes from just outside the horizon propagating to radial infinity. However, the new assumption of this model is that the radiation qubit that propagates outward interacts (locally) with one of the $n$ nonlocal qubits representing the black hole gravitational field, in just such a way that when the mode gets to infinity, the quantum state of that radiation qubit is interchanged with the quantum state of the corresponding black hole gravitational field qubit. This is a purely unitary transformation, not leading to any loss of information. Assume for simplicity that the black hole forms in a pure state that becomes highly scrambled by a unitary transformation. Therefore, as an early outgoing radiation qubit propagates out to become part of the Hawking radiation, when it interchanges its state with that of the corresponding gravitational field qubit, it will become nearly maximally entangled with the black hole state and will have von Neumann entropy very nearly $\ln{2}$, the maximum for a qubit. So the early Hawking radiation qubits will each have nearly the maximum entropy allowed, and there will be very little entanglement between the early radiation qubits themselves. Meanwhile, the black hole qubit corresponding to each outgoing radiation qubit will have taken on the state that the outgoing radiation qubit had when it was just outside the horizon and hence be in the unique singlet Bell state with the infalling radiation qubit just inside the horizon that was originally paired with the outgoing qubit. This vacuum singlet Bell state can then be omitted from the analysis without any loss of information. In this way we can model the reduction in the size of the black hole as it evaporates by the reduction of the number of black hole qubits. We might say that each such vacuum Bell pair falls into the singularity, but what hits the singularity in this model is a unique quantum state, similar to the proposal of Horowitz and Maldacena \cite{Horowitz:2003he}. Therefore, if we start with $n$ black hole gravitational field qubits, $n$ outgoing radiation qubits just outside the horizon, and $n$ infalling radiation qubits just inside the horizon, after the emission of $n_r$ outgoing radiation qubits, $n_r$ of the infalling radiation qubits will have combined into a unique quantum state with the $n_r$ black hole qubits that were originally interacting with the $n_r$ outgoing radiation qubits that escaped, so that we can ignore them as what we might regard as merely vacuum fluctuations. This leaves $n-n_r$ pairs of outgoing radiation qubits just outside the horizon and infalling qubits just inside the horizon (each pair being in the singlet Bell state), and $n-n_r$ black hole gravitational field qubits. Eventually the number of Hawking radiation qubits, $n_r$, exceeds the number of black hole qubits remaining, $n-n_r$, when $n_r > n/2$, and the black hole becomes `old.' At this stage, the remaining black hole qubits all become nearly maximally entangled with the Hawking radiation qubits, so that the von Neumann entropy of the black hole becomes very nearly $(n-n_r)\ln{2}$, which we shall assume is very nearly $A/4$ at that time. Since the whole system is assumed to be in a pure state, and since we have assumed unitary evolution throughout, the von Neumann entropy of the Hawking radiation at this late stage is also very nearly $(n-n_r)\ln{2}$, but now this is less than the maximum value, which is $n_r\ln{2}$. Thus each of the $n_r$ Hawking radiation qubits can no longer be maximally entangled with the remaining $n-n_r$ black hole qubits, and significant entanglement begins to develop between the Hawking radiation qubits themselves. Nevertheless, for any collection of $n' < n/2$ qubits of the Hawking radiation, the von Neumann entropy of that collection is expected \cite{Page:1993df, Page:1993wv} to be very nearly $n'\ln{2}$, so one would still find negligible quantum correlations between any collection of $n'$ Hawking radiation qubits. Finally, when all $n$ of the original outgoing radiation qubits have left the black hole and propagated to infinity to become Hawking radiation qubits, there are no qubits left for the black hole; hence it has completely evaporated away. The $n$ Hawking radiation qubits now form a pure state, just as the original quantum state that formed the black hole was assumed to be. Of course, the unitary scrambling transformation of the black hole qubits means that the pure state of the final Hawking radiation can look quite different from the initial state that formed the black hole, but the two are related by a unitary transformation. The net effect is that the emission of one outgoing radiation qubit gives the transfer of the information in one black hole qubit to one Hawking radiation qubit. But rather than simply saying that this transfer is nonlocal, from the inside of the black hole to the outside, we are saying that the black hole qubit itself is always nonlocal, and that the outgoing radiation qubit picks up the information in the black hole qubit locally, as it travels outward through the nonlocal gravitational field of the black hole. Therefore, in this picture in which we have separated the quantum field theory qubits of the radiation from the black hole qubits of the gravitational field, we do not need to require any nonlocality for the quantum field theory modes, but only for the gravitational field. In this way the nonlocality of quantum gravity might not have much observable effect on experiments in the laboratory focussing mainly on local quantum field theory modes. \section{Mathematics of Qubit Transport} Before the black hole forms, we assume that we have a Hilbert space of dimension $2^n$ in which each state collapses to form a black hole whose gravitational field can be represented by $n$ nonlocal qubits. We assume that we have a pure initial state represented by the set of $2^n$ amplitudes $A_{q_1q_2\ldots q_n}$, where for each $i$ running from 1 to $n$, the corresponding $q_i$ can be 0 or 1, representing the two basis states of the $i$th qubit. Once the black hole forms, without changing the Hilbert space dimension, we can augment this Hilbert space by taking its tensor product with a 1-dimensional Hilbert space for the vacuum state of $n$ infalling and $n$ outgoing radiation modes just inside and just outside the event horizon. We shall assume that this vacuum state is the tensor product of vacuum states for each pair of modes, with each pair being in the singlet Bell state that we shall take to represent the vacuum for that pair of modes. That is, once the black hole forms, we assume that we have $n$ nonlocal qubits for the gravitational field of the black hole, labeled by $a_i$, where $i$ runs from 1 to $n$, $n$ localized qubits for the infalling radiation modes just inside the horizon, labeled by $b_i$, and $n$ localized qubits for the outgoing radiation modes just outside the horizon, labeled by $c_i$. Suppose that each qubit has basis states $\ket{0}$ and $\ket{1}$, where subscripts (either $a_i$, $b_i$, or $c_i$) will label which of the $3n$ qubits one is considering. We assume that each pair of infalling and outgoing radiation qubits is in the vacuum singlet Bell state \begin{equation} \ket{B}_{b_i c_i} = \frac{1}{\sqrt{2}}\Bigl(\ket{0}_{b_i}\ket{1}_{c_i} -\ket{1}_{b_i}\ket{0}_{c_i}\Bigr). \label{Bell} \end{equation} Initially the quantum state of the black hole gravitational field and radiation modes is \begin{equation} \ket{\Psi_0}=\sum_{q_1=0}^1\sum_{q_2=0}^1\cdots\sum_{q_n=0}^1 A_{q_1q_2\ldots q_n}\prod_{i=1}^n\ket{q_i}_{a_i}\prod_{i=1}^n\ket{B}_{b_ic_i}, \label{initial state} \end{equation} where the $A_{q_1q_2\ldots q_n}$ are the amplitudes for the $2^n$ product basis states for the black hole gravitational field. Note that the entire quantum state is the product of a state of all the black hole gravitational qubits and a single pure vacuum state for the radiation modes. During the emission of the $i$th radiation mode to become a mode of Hawking radiation at radial infinity, the basis state for the subsystem of the $i$th black hole, infalling radiation, and outgoing radiation qubits changes as \begin{equation} \ket{q_i}_{a_i}\ket{B}_{b_ic_i} \mapsto -\ket{B}_{a_ib_i}\ket{q_i}_{c_i}, \label{transfer} \end{equation} where $\ket{B}_{a_ib_i}$ is the analogue of $\ket{B}_{b_ic_i}$ given by Eq.\ (\ref{Bell}) with $b_i$ replaced by $a_i$ and $c_i$ replaced by $b_i$. As is obvious from the expressions on the right hand sides, this just interchanges the state of the $i$th black hole qubit with the state of the $i$th outgoing radiation qubit. If $P_{a_ic_i} = \ket{B}_{a_ic_i}\bra{B}_{a_ic_i}$ multiplied by the identity operator in the $b_i$ subspace, then for $\theta = \pi$ the continuous sequence of unitary transformations \begin{equation} U(\theta)=\exp\Bigl(-i\theta P_{a_ic_i}\Bigr)={\rm I}+(e^{-i\theta}-1)P_{a_ic_i} \label{Unitary operator for qubit transfer} \end{equation} becomes $U(\pi) = {\rm I}-2P_{a_ic_i}$, which gives the unitary transformation \eqref{transfer}, interchanging the states of the $i$th black hole qubit with the state of the outgoing radiation qubit. We might suppose that as the radiation qubit moves outward, the $\theta$ parameter of the unitary transformation is a function of the radius $r$ that changes from 0 at the horizon to $\pi$ at radial infinity. For example, one could take $\theta = \pi(1 - K/K_h)$, where $K$ is some curvature invariant (such as the Kretschmann invariant, $K = R^{\mu\nu\rho\sigma}R_{\mu\nu\rho\sigma}$) that decreases monotonically from some positive value at the horizon (where its value is $K_h$) to zero at infinity. We now assume that after the emission of the $i$th mode, the vacuum Bell state of the $i$th black hole qubit and the $i$th infalling radiation qubit can be dropped from the analysis, so that one only has the Hawking radiation qubit remaining for that $i$. Then the state of the subsystem for that $i$ goes from $-\ket{B}_{a_ib_i}\ket{q_i}_{c_i}$ given by Eq.\ (\ref{transfer}) to simply $\ket{q_i}_{c_i}$ for the qubit representing the Hawking radiation mode. Therefore, after all of the $n$ outgoing radiation modes propagate out to infinity while interacting with the black hole gravitational field, and after all the Bell vacua left inside the black hole are omitted, one is left with no black hole and the Hawking radiation in the final pure state \begin{equation} \ket{\Psi_1}=\sum_{q_1=0}^1\sum_{q_2=0}^1\cdots\sum_{q_n=0}^1 A_{q_1q_2\ldots q_n}\prod_{i=1}^n\ket{q_i}_{c_i}. \label{final state} \end{equation} As a note we require that nonlocal gravitational qubits $a_i$ do not create firewalls by themselves. That is, even though the vacuum states on the horizon $b_i,c_i$ are in the range of nonlocal effects, they remain to be constrained in the singlet state unless systems $c_i$ are propagating away to infinity as Hawking radiation by Eq.\ \eqref{Unitary operator for qubit transfer}. This is consistent with the above assumption that the parameter $\theta$ in Eq.\ \eqref{Unitary operator for qubit transfer} is a function of the radius $r$. Conversely, it seems plausible to assume that any incoming mode gradually \emph{drops off} some of its information during propagation through this nonlocal gravitational field. \subsection{Mining Issue} AMPSS \cite{Almheiri:2013hfa}, whose Eq.\ (3.3) is essentially the same as our \eqref{transfer}, raised the following issue with subsystem transfer models as resolutions of the firewall paradox. Suppose there exists an ideal mining equipment that can approach arbitrarily close to the horizon without falling into it, and then the equipment interacts with one of systems $c_i$ just outside the horizon. Note that this can be done without any exchange of energy due to the infinite redshift, and it is assumed that there is no entangling either. For example, the mining equipment can unitarily acts on the system $c_i$ as \begin{eqnarray} U_{\text{mine}}&:&\ket{0}_{c_i}\mapsto e^{i\phi}\ket{0}_{c_i},\;\;\;\ket{1}_{c_i}\mapsto e^{-i\phi}\ket{1}_{c_i}.\\ U_{\text{mine}}&:&\ket{B}_{b_ic_i}\mapsto\frac{\cos\phi}{\sqrt{2}}\Bigl(\ket{0}_{b_i}\ket{1}_{c_i} -\ket{1}_{b_i}\ket{0}_{c_i}\Bigr)+\frac{i\sin\phi}{\sqrt{2}}\Bigl(\ket{0}_{b_i}\ket{1}_{c_i} +\ket{1}_{b_i}\ket{0}_{c_i}\Bigr).\label{mine} \end{eqnarray} Thus the system on the horizon has one bit of information after this mining process and is thus no longer in the vacuum state. First of all, it seems implausible that such an ideal equipment can be physically realistic. Since the equipment is accelerating in order to stay outside the horizon without falling into the black hole, it has an Unruh temperature that becomes very high near the horizon. Then the equipment and the modes it interacts with, $c_i$ in this case, should strongly couple and would be expected to be approximately in a thermal state. As a consequence it seems plausible that energy must be transferred between the mining equipment and the modes $c_i$. Also, notice that the AMPSS mining argument does not take nonlocality into account. That is, the mining equipment would interact with the nonlocal gravitational degrees of freedom even if it could avoid the objection of the previous paragraph. As discussed previously, interactions with nonlocal gravitational degrees of freedom transfer part of the quantum information of the mining system into the gravitational degrees of freedom as the equipment approaches to the horizon. We can think of this transferred part as now being a part of the temporarily enlarged nonlocal gravitational degrees of freedom when the equipment is very near to the horizon. Then in this picture the mining equipment can still produce the phase change Eq.\ \eqref{mine} on the system just outside the horizon, but this excitation will be eventually absorbed into the nonlocal gravitational degrees of freedom. This absorption is possible regardless of how old the black hole is, because the nonlocal degrees of freedom are temporarily enlarged by the partially transferred degrees of freedom of the mining equipment. In summary, the AMPSS mining argument is not problematic for our model. \section{Giddings' Physical Conditions} Giddings \cite{Giddings:2012bm} has proposed a list of physical constraints on models of black hole evaporation. We shall write each constraint in italics below and then follow that with comments on how our qubit model can satisfy the proposed constraint. (i) \emph{Evolution is unitary.} Our model explicitly assumes unitary evolution. (ii) \emph{Energy is conserved.} Our model is consistent with a conserved energy given by the asymptotic behavior of the gravitational field. The unitary transformation $U(\theta(r))$ during the propagation of each radiation qubit can be written in terms of a radially dependent Hamiltonian without any explicit time dependence, so there is nothing in our model that violates energy conservation. (iii) \emph{The evolution should appear innocuous to an infalling observer crossing the horizon; in this sense the horizon is preserved.} We explicitly assume that the radiation modes are in their vacuum states when they are near the horizon, so there is no firewall or other drama there. (iv) \emph{Information escapes the black hole at a rate $dS/dt\sim1/R$.} Although we did not discuss the temporal rates above, if one radiation qubit propagates out through some fiducial radius, such as $r = 3M$, during a time period comparable to the black hole radius $R$, since during the early radiation each qubit carries an entropy very nearly $\ln{2}$, indeed one would have $dS/dt\sim1/R$. (v) \emph{The coarse-grained features of the outgoing radiation are still well-approximated as thermal.} Because of the scrambling of the black hole qubits so that each one is very nearly in a maximally mixed state, when the information is transferred from the black hole qubits to the Hawking radiation qubits, each one of these will also be very nearly in a maximally mixed state, which in the simplified toy model represents thermal radiation. Furthermore, one would expect that any collection of $n' < n/2$ qubits of the Hawking radiation also to be nearly maximally mixed, so all the coarse-grained features of the radiation would be well-approximated as thermal. (vi) \emph{Evolution of a system ${\cal H}_A\otimes{\cal H}_B$ saturates the subadditivity inequality $S_A+S_B \geq S_{AB}$.} Here it is assumed that $A$ and $B$ are subsystems of $n_A$ and $n_B$ qubits respectively of the black hole gravitational field and of the Hawking radiation, not including any of the infalling and outgoing radiation qubits when they are near the horizon. Then for $n_A + n_B < n/2$, $A$, $B$, and $AB$ are all nearly maximally mixed, so $S_A \approx n_A\ln{2}$, $S_B \approx n_B\ln{2}$, and $S_{AB} \approx (n_A+n_B)\ln{2}$, thus approximately saturating the subadditivity inequality. (Of course, for any model in which the total state of $n$ qubits is pure and any collection of $n' < n/2$ qubits has nearly maximal entropy, $S \approx n'\ln{2}$, then if $n_A < n/2$, $n_B < n/2$, but $n_A + n_B > n/2$, then $S_A \approx n_A\ln{2}$ and $S_B \approx n_B\ln{2}$, but $S_{AB} \approx (n-n_A-n_B)\ln{2}$, so $S_A+S_B-S_{AB} \approx 2n_A+2n_B-n > 0$, so that the subadditivity inequality is generically not saturated in this case.) \section{Conclusions} We have given a toy qubit model for black hole evaporation that is unitary and does not have firewalls. It does have nonlocal degrees of freedom for the black hole gravitational field, but the quantum field theory radiation modes interact purely locally with the gravitational field, so in some sense the nonlocality is confined to the gravitational sector. The model has no mining issue and also satisfies all of the constraints that Giddings has proposed, though further details would need to be added to give the detailed spectrum of Hawking radiation. The model is in many ways {\it ad hoc}, such as in the details of the qubit transfer, so one would like a more realistic interaction of the radiation modes with the gravitational field than the simple model sketched here. One would also like to extend the model to include possible ingoing radiation from outside the black hole. \section*{Acknowledgments} DNP acknowledges discussions with Beatrice Bonga, Fay Dowker, Jerome Gauntlett, Daniel Harlow, Adrian Kent, Donald Marolf, Jonathan Oppenheim, Subir Sachdev, and Vasudev Shyam at the Perimeter Institute, where an early version of this paper was completed. We also benefited from emails from Steven Avery, Giorgi Dvali, Steven Giddings, Yasunori Nomura, and Douglas Stanford. Revisions were made while using Giorgi Dvali's office during the hospitality of Matthew Kleban at the Center for Cosmology and Particle Physics of New York University. This research was supported in part by the Natural Sciences and Engineering Research Council of Canada, and in part by Perimeter Institute for Theoretical Physics. Research at Perimeter Institute is supported by the Government of Canada through the Department of Innovation, Science and Economic Development and by the Province of Ontario through the Ministry of Research, Innovation and Science. \section*{References}
c7dfc98266d9a9c36e71efeb0e655803ef3a7190
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Results} The actuation concept consists of two elements. First, the conversion of rotation into time-reversible translation is achieved by each particle separately, hence, a single-particle problem. Second, the breaking of the time reversibility arises from the interaction between the particles in a cluster. In the following, these two elements are demonstrated qualitatively with a combination of analytic and numerical calculations using sd-particles. Afterwards, the investigations are used to explain our experimental observations of actuated clusters from magnetic Janus spheres, which move under oscillating fields. \subsection{Oscillation of single sd-particles} We access the role of magnetic asymmetry of a single dipolar particle by considering its reciprocal motion under oscillating fields for four different scenarios (Fig.~\ref{pics:sk-trans-sd}). If, in an isotropic medium, a sphere with a central dipole (case i, Fig.~\ref{pics:sk-trans-sd}) is exposed to an oscillating field $B^{z}=B_0^{z}\sin(\omega_B t)$, the dipole follows the field via up and down rotation by angle $\theta\in [-\theta_A,\theta_A]$. As a magnetic object rotates around its magnetic center this results only in particle rotation. Asymmetry is introduced by sd-particles, where the magnetic center is shifted away from the geometric center (case ii, Fig.~\ref{pics:sk-trans-sd}). Here, the dipole is shifted radially outwards by $\xi\,\frac{d_p}{2}$ along the dipole orientation, where $\xi\in[0,1]$ is the shift parameter and $d_p$ is the particle diameter. Under oscillating fields, the rotation around the magnetic center displaces the particle center periodically. In an isotropic environment, the particle center moves on an arc \cite{Che09} with a radius determined by the shift $\xi$. The behavior changes in the proximity of a wall, where anisotropic drag is present. We consider this case here because the experiments described later in this work are performed on particles moving close to a planar substrate. We emphasize that the presence of a wall is, however, not essential for the actuation mechanism presented here. \begin{figure} \centerline{\includegraphics[width=8.7cm]{sketch1.pdf} } \caption{Time-reversible motion of a single dipolar sphere under oscillating magnetic fields $B^{\text{z}}$. The scenarios depict a sphere with central dipole (i, iii) and shifted dipole (ii, iv) in an isotropic medium (i, ii) and close to a wall (iii, iv). The dipole (light red arrow) with moment $m$ oscillates up (green) and down (yellow), which results in a displacement of the sphere (dotted green and yellow circles) under asymmetric conditions (ii\,-\,iv). } \label{pics:sk-trans-sd} \end{figure} Already for a sphere with central dipole (case iii, Fig.~\ref{pics:sk-trans-sd}) in a liquid medium close to a wall, rotation is converted into reversible back and forth displacement. An sd-particle also rolls back and forth, and additionally performs a periodic displacement away from the wall and towards the wall (case iv, Fig.~\ref{pics:sk-trans-sd}) because the magnetic center is shifted away from the geometric center of the sphere. In contrast to case iii, this motion is not symmetric with respect to $\theta$ due to the hindrance by the wall. For an sd-particle in an isotropic medium, we can derive an analytic expression for the displacement of the particle center that arises from the rotation by an infinitesimal angle $\mathrm{d}\theta$ around the magnetic center (see SI Appendix). The infinitesimal displacement, separated into components perpendicular ($\mathrm{d}r$) and parallel ($\mathrm{d}z$) to the oscillating field $B^{z}$, is given by \begin{subequations} \begin{align} \mathrm{d}r &= - \xi \frac{d_p}{2} c(\theta) \sin \theta\, \mathrm{d}\theta , \\ \mathrm{d}z &= \xi \frac{d_p}{2} c(\theta) \cos \theta\, \mathrm{d}\theta . \end{align} \label{eq:conv-med}% \end{subequations} The term $c(\theta)$ contains potential sources of anisotropic hindrance or enhancement. For a single particle, this is given by, e.g., a wall. The determination of $c(\theta)$ close to a wall is a complex problem. Since the rotation around the shifted magnetic center periodically changes the distance between particle center and wall, also the hydrodynamic friction coefficient changes periodically. To illustrate the concept, here, it is sufficient to approximate the surface-induced drag with a step function of two values $c_1,c_2$ that alternate with the sign of $\theta$. Distinguishing between translation parallel ($r$) and perpendicular ($z$) to the surface gives \begin{equation} c_r(\theta)= \begin{cases} c_{r,1} & \text{if } \theta \geq 0 \\ c_{r,2} & \text{if } \theta < 0 \end{cases} \text{ ,} \quad c_z(\theta)= \begin{cases} c_{z,1} & \text{if } \theta \geq 0 \\ c_{z,2} & \text{if } \theta < 0 \end{cases}. \label{eq:conv-sf} \end{equation} During $\theta > 0$ (green circle in case iv, Fig.~\ref{pics:sk-trans-sd}) the surface-induced translation (ii) enhances the translation $r$ caused by the off-centered dipole (iii). During $\theta < 0$ (yellow circle), the directions of the two translations are opposite to each other as can be seen by the comparison between cases ii and iii. Therefore, for oscillations close to a wall (iv) the displacement during $\theta > 0$ is always higher than during $\theta < 0$ and, thus, $|c_1|>|c_2|$. If the influence of the surface on the displacement is stronger than that of the shifted magnetic center, then during $\theta<0$ the overall translation direction is even inverted with respect to the one in an isotropic medium (ii), and $c_{r,2}<0$. Otherwise $c_{r,2}>0$ holds. Note that due to the time reversibility the integral of Eq.~\ref{eq:conv-med} over one oscillation cycle ($\theta=0\rightarrow \theta_{\text{A}}\rightarrow 0\rightarrow -\theta_{\text{A}}\rightarrow 0$) becomes zero, excluding directed motion. \subsection{Sd-particles in a ring cluster} Next, we transfer the considerations for a single sd-particle to sd-particles that form a two-dimensional cluster in the plane $z=0$. We assume that within the cluster the particles are free to rotate in response to magnetic stimuli but that the relative distances between all particles are kept constant during field exposure. The simplest case to study interacting sd-particles is given if they form a ring. Then, ideally all particles have identical boundary conditions and behave uniformly. As an example, a ring of three particles (triple ring, Fig. \ref{graph:sk-rot}\,a) will be studied. In analogy to a single sd-particle, rotation by $\mathrm{d} \theta$ displaces each sd-particle by $\mathrm{d} r$ and $\mathrm{d} z$. However, since the particles are constrained to a circle by the magnetic interaction (Fig.~\ref{graph:sk-rot}\,a) and move uniformly, the displacement $\mathrm{d} r$ is forced along an orbit spanned by the particle centers. As a consequence, the displacement of each sd-particle, induced by the rotation $\mathrm{d} \theta$, is hindered sterically by the presence of neighboring particles. The hindrance depends on the angle $\varphi$ that is enclosed by the tangent to that orbit and the in-plane component of the dipole (Fig.~\ref{graph:sk-rot}\,a). $\mathrm{d}r$ becomes zero if $\varphi=90^{\circ}$, and it is completely unobstructed if $\varphi=0^{\circ}$. Such an anisotropic hindrance is taken into account by introducing the reduction factor $c_{\text f}(\theta)=\cos \varphi(\theta)$. This hindrance indirectly affects also the displacement $\mathrm{d}z$ The factor $c_{\text f}(\theta)$ suggests that the hindrance between interacting particles depends exclusively on $\varphi$. To determine $\varphi$ during the driven oscillations of sd-particles in a triple ring, we have studied the rotational (angular) motion of the particles separately from their translational motion. The rotational motion $\varphi(\theta(t))$ has been obtained by numerically solving the equation of rotation of the interacting sd-particles under an oscillating field $B^{z}$, which points perpendicular to the plane of the ring cluster. \begin{figure*} \centerline{\includegraphics{sim-rot_copy3.pdf}} \caption{Rotation of a ring of sd-particles. (\textbf{a}) Triple ring (side and top view), and orientation angles $\varphi$ and $\theta$ of sd-particles. For better visualization, the hemisphere that contains the dipole (red arrow) is colored in grey. The dotted ring indicates the orbit of motion spanned by the particle centers. (\textbf{b}) Numerically calculated trajectory in $\theta-\varphi$ coordinates of an sd-particle ($\xi=0.55$) in a triple ring under an oscillating field ($B_0^{\text{z}}=6\,\frac{\mu_0 m_{\text{p}}}{4\pi d_{\text{p}}^3}$). The arrows indicate the time evolution. (\textbf{c}) Sketched top view of a triple ring during one movement cycle. For clarity, the drawn variations in $\varphi$ (dark green segments) are much larger than the ones plotted in (b). (\textbf{d}) Orbital translation $r$ versus time $t'$ of an sd-particle in a triple ring that is located in an isotropic medium or on a surface ($c_1=2$ and $c_2=-0.4$). The time $t'$ is measured in period $T$. } \label{graph:sk-rot} \end{figure*} The periodic variation of $B^z$ leads to an oscillatory behaviour of $\theta (t)$. Additionally, we have observed that the variation of $\theta$ involves a periodic change of $\varphi$ for interacting sd-particles in a ring. The trajectories of the dipoles in $\theta$\,-\,$\varphi$ coordinates are double loops (Fig.~\ref{graph:sk-rot}\,b). Starting from the field-free equilibrium state (0), the dipoles perform a short transient oscillation until they reach a steady-state oscillation along the path I\,$\rightarrow$\,IV. The loop implies that the dipoles obtain larger angles $\varphi$ while $|\theta|$ increases (I/III) than during the part of the trajectory in which $|\theta|$ decreases (II/IV) (Fig.~\ref{graph:sk-rot}\,c). Such a loop trajectory $\varphi(\theta)$ breaks time reversibility since the sense of direction of the loop is uniquely defined. Inserting $\varphi(\theta)$ into the reduction term $c_{\text f}(\theta)=\cos\varphi(\theta)$ gives rise to a time-irreversible translation of the sd-particles along the orbit. During increase of $|\theta|$, the displacement $\mathrm{d}r$ is hindered more strongly than during decrease of $|\theta|$. Based on that a non-reciprocal displacement $\mathrm{d}r$ of the sd-particles in the triple ring can be predicted, which leads to an effective rotation of the whole ring (Fig.~\ref{graph:sk-rot}\,c). Inserting the term $c(\theta)=c_{\text f}(\theta)=\cos\varphi(\theta)$ into Eq.~\ref{eq:conv-med} gives the infinitesimal displacement of a particle in a ring located in an isotropic environment. The time dependence of the displacement $\mathrm{d}r(t)$ can be obtained by inserting analytic expressions for $\varphi(\theta(t))$ and $\theta(t)$. Fitting the numerical data (Fig.~\ref{graph:sk-rot}\,b) to Lissajous curves (see SI Appendix) yields suitable approximate functional terms. The net translation is obtained by integration, $r(t')=\int_0^{t'} \mathrm{d}r(t)$ (Fig.~\ref{graph:sk-rot}\,d). For a ring in an isotropic medium, the sd-particles periodically translate back and forth twice per movement cycle (Fig.~\ref{graph:sk-rot}\,d) similar to a single particle (case ii, Fig.~\ref{pics:sk-trans-sd}). In addition, they gain an overall distance $\Delta r$ traveled after each movement cycle I\,$\rightarrow$\,IV. This is a very small effect since $r$ is in the range of a few $\% \, d_{\text p}$, and $\Delta r$ is less than one tenth of that range The translation $r(t)$ during one cycle can be detailed as follows. During increase of $|\theta|$ (I/III), the particle translates with its magnetic side ahead (Fig.~\ref{graph:sk-rot}\,c), defined as back motion. Forth translation with the non-magnetic side ahead occurs during sections II and IV. As derived from the rotational trajectory (Fig.~\ref{graph:sk-rot}\,b), the factor $\cos \varphi(\theta)$ during I/III is smaller than during II/IV. Therefore, $-\mathrm{d}r(\theta_{\rm I})=-\mathrm{d}r(\theta_{\rm III})<\mathrm{d}r(\theta_{\rm II})=\mathrm{d}r(\theta_{\rm IV})$. The integration of $\mathrm{d}r$ over one field cycle is, thus, positive. The particles in a ring perform a net translation along the orbit with the non-magnetic side facing forward. Finally, the uniform translation of all particles along the orbit gives an effective rotation of the whole ring in the plane of the particle centers. In addition to the translation $\mathrm{d}r$ in the plane of the triple ring, also the motion $\mathrm{d}z$ perpendicular to the ring has to be considered. Collectively, the particles move up and down, resulting in an oscillating up-and-down translation of the ring. Due to the broken symmetry caused by $\cos \varphi(\theta)$, the trajectory in $z$-direction is not reversible either. However, the vertical component $z$ that is gained during $\mathrm{d}\theta>0$ is reversed during $\mathrm{d}\theta<0$, and the ring gains no overall distance along $z$. For a ring that is located on a surface, the translation $r(t)$ has been obtained analogously (Fig.~\ref{graph:sk-rot}\,d). In this case, the factor $c(\theta)=c_{\text f}(\theta)\cdot c_{r}(\theta)$ has been applied in Eq.~\ref{eq:conv-med}\,a. The translational behavior differs qualitatively from the one in an isotropic medium if a value $c_{r,2}<0$ is applied, i.e. if the surface drag dominates over the effect of the dipole shift on the translation. Then, negative values of $c_{r,2}$ reverse the sign of the displacements in sections III and IV. Thus, forth displacement takes place during II/III and back displacement occurs during I/IV (Fig.~\ref{graph:sk-rot}\,d). During I/II, where $c_{r,1}>0$ is effective, the translation $r(t)$ looks similar to the one without surface. Further, since on the surface $|c_1|>|c_2|$, $\mathrm{d}r(\theta_{\rm I})+\mathrm{d}r(\theta_{\rm II})<\mathrm{d}r(\theta_{\rm III})+\mathrm{d}r(\theta_{\rm IV})$ holds. Also in this scenario the integrated forth displacement $r$ (facing the non-magnetic side) is larger than the back displacement (magnetic side) during one field cycle. In the example presented in Fig.~\ref{graph:sk-rot}\,d, the ratio $c_2/c_1=-0.2$ has been applied. This ratio is also visible in the curve $r(t)$ by the ratio between the amplitude during I/II and the one during III/IV. The presented actuation concept is, of course, not limited to the triple cluster, which is, however, the smallest entity showing this effect. The concept can be applied to other clusters, too, as will be shown next by experimental studies. \subsection{Actuated clusters of capped particles} \label{ssec:rot-caps} The proposed actuation of interacting sd-particles under oscillating fields can indeed be found in an experimental system. We use particles with a diameter of $d_{\text{p}}=4.54\,\mu$m that are equipped with a hemispherical magnetic coating, so-called capped or Janus particles. Due to the hemispherical coating, the magnetic center of mass is shifted away from the particle center. A ferromagnetic coating with perpendicular magnetic anisotropy provides a stray field with dipolar characteristic \cite{Alb05}, and the net dipole points along the Janus director. For these particles, an effective dipole shift of about $\xi=0.6$ has been estimated from comparison with simulations \cite{Kan11, Ste16b}. In solution, the particles sediment on a substrate and assemble in a great variety of open and compact two-dimensional structures with diverse magnetic order \cite{Ste16a,Ste16b,Bar08a}. The particle orientation can be visualized by transmission light microscopy via the optical contrast between the transparent, uncoated and the intransparent, coated hemisphere. In contrast to the sd-particles with a single point dipole, the capped particles provide an extended magnetization distribution. This, however, leads only to a slight modification of the dipole interaction potential \cite{Ste16b}. Quantitative consequences on the motion will be pointed out accordingly. \begin{figure} \begin{center} \includegraphics{graph-rotation1.pdf} \caption{Actuation of colloidal clusters. (\textbf{a}) Image sequence of the rotation of a self-assembled triple ring (R3) under an oscillating field. (\textbf{b}) Orientation $\varphi(t)$, the sine fit curve (solid), and the orbital translation $r(t)$ of a capped particle in R3 during four periods $T$ ($B_0^{\text{z}}=0.34$\,mT, $\omega_B/2\pi= 2$\,Hz at 20\,fps). Dotted lines are spline curves through the data points. (\textbf{c}) Microscopy images of a five-particle ring (R5) and a linear chain (C5) that form under weak oscillating fields. (\textbf{d}) $r(t)$ of a particle in R5 and C5 ($B_0^{\text{z}} = 1.22$\,mT, $\omega_B/2\pi= 10$\,Hz). The superimposed beat results from the stroboscopic recording with a digital camera (20\,fps). (Scale bars: 5\,$\mu$m)} \label{graph:3-4ring} \end{center} \end{figure} Here, the actuation of selected structures under oscillating fields will be presented. Spontaneously, the capped particles self-assemble into triple rings (R3). Under oscillating fields $B^{\text{z}}$, one can observe that the ring rotates (Fig.~\ref{graph:3-4ring}\,a). In agreement with the theoretical predictions, the uncapped, non-magnetic hemisphere always faces towards the rotation direction. The translations $r(t)$ and the orientations $\varphi(t)$ of the capped particles in R3 have been measured with image analysis (Fig.~\ref{graph:3-4ring}\,b). Their trends coincide qualitatively with numerical findings for sd-particles in a ring on the substrate (Fig.~\ref{graph:sk-rot}\,d). During each cycle, $r(t)$ exhibits a motion sequence of fast forth, slow forth, slow back, and fast back translation. In analogy to the simulation, the ratio between the amplitudes during the section of slow and fast translation in the experimental curve (Fig.~\ref{graph:3-4ring}\,b) gives a measure for the ratio between the conversion factors, suggesting a value of $c_2/c_1\approx -0.2$. The mean relative increase $\Delta r/d_{\rm p}$ per cycle is $1.8\,\%$. This is one order of magnitude larger than in the simulation (Fig.~\ref{graph:sk-rot}\,b). The difference results from the modified dipole interaction between the capped particles, which will be confirmed next. The capped particles perform radial oscillations $\varphi(t)$ twice per field cycle T (Fig.~\ref{graph:3-4ring}\,b), which also coincides with the sd-particles. Fitting $\varphi(t)$ to a sine function gives a mean value of $\varphi=32^{\circ}$ and an oscillation amplitude of $10^{\circ}$. This radial amplitude of capped particles is about 8 times as large as for sd-particles, assuming same polar amplitudes in $\theta$. Therefore, also the double loop trajectory of capped particles is broadened by this factor. Since the area enclosed by the double loop is proportional to the propulsion efficiency $\Delta r/d_{\rm p}$ , the experimental value of $\Delta r/d_{\rm p}$ must be about one order of magnitude larger than in the simulation. The theoretically proposed vertical displacement of the particle centers are not visible in the vertical projections of the clusters. However, indirect evidence is given by the experimentally obtained value of $|c_2/c_1|\neq 1$, which implies that the translational friction differs between $\theta<0$ and $\theta>0$. In order to exclude an influence based on two chemically different surfaces of the Janus particles, we have examined two sets of particles, one with a Pd layer and one with a silicon oxide layer on top of the magnetic film. Under same experimental conditions, triple rings of both types of particles exhibit the same angular speed, thus, surface effects can be excluded. Next, we have tested whether the actuation concept also holds for other clusters. Due to the magnetic shift, capped particles (and also sd-particles) do not form larger rings spontaneously. However, we have shown earlier that closed rings and linear chains (R5 and C5 in Fig.~\ref{graph:3-4ring}\,c) can form if weak, low-frequency oscillating fields are applied \cite{Ste16a}. The presence of these clusters is, thus, linked directly to any motion of these clusters under oscillating fields. We have observed that also R5 rotates under oscillating fields, leading to a significant slope in $r(t)$ (Fig.~\ref{graph:3-4ring}\,d). In contrast, under the same environmental conditions, chains (C5) only perform periodic back and forth displacement along the chain direction with almost no net translation. As reported previously, the loop trajectory (\ref{graph:sk-rot}\,b), which breaks time-reversal symmetry, vanishes for a collinear orientation of the dipolar particles \cite{Ste16a}. The particles perform reciprocal oscillations $\theta(t)$ without variation in $\varphi$. This implies that clusters of collinear sd-particles cannot move under oscillating fields. The small, but yet visible, slope in the trajectory of C5 (Fig.~\ref{graph:3-4ring}\,d) suggests that the particles in a chain are not perfectly collinear at all times. This might be caused by thermally induced bending (deformation) of the chain. \begin{figure}[t] \includegraphics{translator.pdf} \caption{Trajectory of the center of mass of two compact clusters (insets) (\textbf{a,b}) under an oscillating field with $B^{\text{z}} = 0.53$\,mT, $\omega_B/2\pi = 5$\,Hz (a: 88\,s; b: 251\,s) and (\textbf{c}) under an oscillating field ($B^{\text{z}} = 1.13$\,mT, $\omega_B/2\pi = 20$\,Hz) and a constant in-plane field ($B^{\text{x/y}} = 0.03$\,mT). The orientation of $B^{\text{x/y}}$ is switched from the $y$-axis to the $x$-axis during the recording, leading to a change of the translation direction. (Scale bars: 5\,$\mu$m)} \label{graph:rot-trans} \end{figure} So far, the actuation of interacting sd-particles has been demonstrated for ring clusters to take advantage of symmetric boundary conditions. Of course, the presented concept also holds if the assembled cluster is not rotationally symmetric. Since the motion of a cluster arises from the sum of the displacements $\mathrm{d}r$ of the constituting particles, the lack of rotational symmetry can result in more diverse motion. As an example, two compact clusters consisting of five particles (Fig.~\ref{graph:rot-trans}\,a, b) have been examined under oscillating fields. The capped particles can self-assemble into compact clusters with various magnetic configurations as a result of the magnetic shift (SI) \cite{Kan11,Kli13}. Here, we present two stable clusters (Fig.~\ref{graph:rot-trans}\,a, b) that exhibit the same spatial shape but slightly different magnetic configurations. Specifically, they differ in the azimuthal orientation of only one particle (most left one in the insets in Fig.~\ref{graph:rot-trans}) while all other four particles are similar. Interestingly, this small difference in the magnetic configuration leads to a drastic difference in the trajectory of the clusters when actuated by oscillating fields (Fig.~\ref{graph:rot-trans}\,a, b). One of the clusters (a) performs a screw-like motion while the other (b) exhibits an almost linear path. Considering that each particle in a cluster contributes with $\mathrm{d}r$ along the in-plane orientation of its cap, the net translation arises approximately from the sum of the dipole vectors. In cluster (a), the dipole vectors of the particles form a bent ring, which leads to the observed helical motion. In cluster (b), the vectorial sum of the cap orientations gives a significant cluster moment, resulting in directed translation of the cluster. The net cluster moment additionally can be used to orient the cluster in fields parallel to the cluster plane, provided that the field intensity is too low to alter the cluster structure. This enables a control over the translation direction (Fig.~\ref{graph:rot-trans}\,c). After orienting the cluster along a weak parallel field, it performs directed translation under an additional oscillating field. Switching the in-plane field orientation by $90^{\circ}$ results in a change of the translation direction. A noteworthy feature in the example presented in Fig.~\ref{graph:rot-trans}\,c is that the translation path is not parallel to the direction of the applied field, because the net magnetic orientation of this cluster does not coincide with its effective translation direction. \section{Discussion} At low Reynolds numbers, locomotion under an oscillating drive is possible only if the system provides flexibility. Here, we have shown that this can be realized even in the absence of structural body deformations, a scenario that is impossible within the usually considered actuation concept of hydrodynamic drag. The newly introduced propulsion strategy requires two ingredients. First, an object must be built from multiple spheres that can rotate internally in response to a field. Second, the spheres must exhibit an anisotropic magnetization distribution. Under oscillating fields, the interaction between these spheres activates torsional oscillation perpendicular to the field direction, which breaks time reversal symmetry. While this strategy does not rely on hydrodynamic drag, the presence of drag close to a surface modifies the motion. We propose that this actuation concept can be applied also to other magnetically anisotropic particles, which are currently widely discussed for complex self-assembly \cite{Tie14,Che09,Yen16,Sac12,Abr13,Yan12}. For practical applications, the presented actuation concepts combines three essential benefits. First and most striking, the overall motion can take different paths depending on the magnetic configuration of a cluster. Second, the translation direction can be controlled. Third, from the equation of motion (Eq.~\ref{eq:erot-coup}) it can be deduced that the speed of motion is adjustable by varying the amplitude of the oscillating field. Ring clusters perform rotation under oscillating fields due to their rotational symmetry. Other, asymmetric clusters perform directed translation as demonstrated with magnetically capped particles. The motion crucially depends on the position and orientation of each constituting particle in the cluster. This suggests that, beyond the presented examples, by proper choice of asymmetric particles and assembly of a cluster any combination of rotation and translation can be realized. For a rigorous theoretical description of a translating cluster with asymmetric shape, additionally, the anisotropic drag must be taken into consideration, which provides a complex problem on its own \cite{Kuem13,Wen14}. Further, a translating cluster was shown to exhibit a residual magnetic moment, which can be used to steer the cluster orientation with a weak parallel field. A superimposed oscillating field leads to orientation-controlled directed translation. Thus, the actuation with a single field component advantageously gives two remaining field components free for controlling the translation direction. Finally, the field amplitude directly controls the actuation speed. With increasing fields, the oscillation amplitude $\theta_A$ increases. Since $\theta_A$ sets the integration bounds for the displacement $\Delta r$ per field cycle (Eq.~\ref{eq:conv-med}), the field amplitude determines $\Delta r$. The presented clusters move with 1-5\,$\mu$m/s, and, thus, the actuation efficiency is comparable to those of swimmers under anisotropic drag \cite{Bec16}. In conclusion, the presented results show that internal rotations in multi-component objects can play a crucial role for directed locomotion in overdamped systems. This provides new actuation strategies that complement established concepts based on anisotropic drag. Considering the simplicity of the oscillating field pattern that drives the motion, the question remains whether similar actuation can be found elsewhere, e.g., in animate systems. In addition to anisotropic potentials, also entropic interactions between components with anisotropic shape \cite{Glo07} might enable similar propulsion mechanisms based on the presented strategy of internal rotations. \section{Methods} \subsection{Numerical Methods} The rotational motion of interacting sd-particles in a ring cluster exposed to an oscillating field $\mathbf B^{\rm{z}}=\mathbf B^{\rm{z}}_0\sin (\omega t)$ has been calculated by numerically solving the equation of rotation. In the simulation, the particles are spatially fixed at their center. This assumption is reasonable when studying the rotation of the particles in the reference system of a cluster that does not change its spatial shape. Assuming sd-particles with dipole moment $\textbf{m}$ (unit vector $\hat{\textbf{m}}$) and stray field $\textbf{B}^{\rm{p}}$, the equation of rotation is obtained by balancing the rotational drag, ${\tau}^{\rm d}$, with the magnetic torques between the sd-particles and between sd-particles and the field, ${\tau}^{\rm B}$. Inertial forces have been neglected on account of the dominating drag forces. Due to the dipole shift, besides the aligning torque ${\tau}^{\rm p}=\textbf{m} \times \textbf{B}^{\rm{p}}$, also the gradient force $(\textbf{m} \cdot \nabla) \textbf{B}^{\rm{p}}$ between the sd-particles is relevant. The latter converts into an effective torque, ${\tau}^{ F}$, such that the equation of rotation is given by \begin{eqnarray} {\tau}^{\rm d}&=& {\tau}^{\rm p}+ {\tau}^{F} + {\tau}^{\rm B}\\ f_{\rm{r}} \dot{\Theta} (t)&=& \textbf{m} \times \textbf{B}^{\rm{p}} + \xi \hat{\textbf{m}} \times ((\textbf{m} \cdot \nabla)\textbf{B}^{\rm{p}}) + \textbf{m} \times \textbf{B}^{\rm{z}} \label{eq:erot-coup} \end{eqnarray} $\dot{\Theta}$ is the angular velocity of an object with rotational friction coefficient $f_{\rm{r}}$ and orientation $\Theta=(\theta,\,\varphi)$ (Fig.~\ref{graph:sk-rot}\,a). An ensemble of $n$ particles gives a system of $n$ coupled equations of rotation (Eq.~\ref{eq:erot-coup}), which are solved iteratively in a self-consistent cycle. From the state at time $t$, the magnetic moment $\mathbf{m}_i$ of particle $i$ at time step $t+1$ is obtained b \begin{equation} \hat{\textbf{m}}_i^{(t+1)}={\rm{norm}}\left[\hat{\textbf{m}}_i^{(t)} + \frac{1}{f_{\rm{r}}}(\tau^p + {\tau}^{F} + {m}_i\times ({B}^{z}_0 \, \cos(\omega t)) )\right] \label{eq:rot-mot} \end{equation} The function 'norm$[\cdot]$' scales the updated magnetic moments $\mathbf{m}^{(t+1)}$ to retain the initial magnitude \textit{m} of the dipoles. In the calculation, a rotational friction coefficient of $f_{\rm r}=100\,\frac{\mu_0 m^2}{32\pi r_{\rm p}^3}$ has been applied. It has the dimension of energy since the numerical time steps $\Delta t=(t+1)-t$ are dimensionless and set to 1. The angular frequency $\omega$ is also a dimensionless quantity and is set to 0.05. \subsection{Experimental Details} The particle preparation and experimental setup has been described in detail in reference \cite{Ste16b}. In short, silica spheres with a radius of $r_{\rm p} = (2.27 \pm 0.23)\,\mu$m have been coated on one hemisphere with a magnetic thin film of $[{\rm Co}(0.28\,{\rm nm})/{\rm Pd}(0.9\,{\rm nm})]_8$ \cite{Alb05}. The magnetic thin film causes mass imbalance in the subatomic range (SI) and, thus, can be neglected here. This film is known to exhibit strong perpendicular magnetic anisotropy \cite{Car85}; the magnetic orientation points perpendicular to the film plane. When depositing the film on a sphere, the magnetic anisotropy follows the curvature of the particle surface. After magnetic saturation, the particle becomes magnetically single domain and obtains a radially symmetric anisotropy distribution \cite{Alb05}. This leads to a magnetic stray field with dipolar characteristic, and the net dipole moment points along the Janus director. To test the influence of the chemically distinct surfaces, one set of particles has additionally been coated with $\text{Si}_3\text{N}_4$ on top of the metal film. This top layer oxidizes and becomes amorphous silicon oxide, which is chemically similar to the uncoated silica hemisphere. A suspension of the coated particles in distilled water is studied via transmission light microscopy. Due to the density mismatch the particles sediment on the ground of the sample cell, providing a two-dimensional particle system. An electromagnetic coil is mounted above the sample cell, providing perpendicular, low-frequency fields. An additional set of two pairs of coils attached beneath the sample provide constant fields to orient translating clusters. The microscopy recordings of the moving clusters have been analyzed by optical image analysis. Using the open-source software ImageJ, a customized detection algorithm based on Hough transformation was built and employed. \subsection{Acknowledgments} Financial support by the German Research Foundation (DFG, Grant Nos. ER 341/9-1, AL618/11-1, and FOR 1713 GE 1202/9-1) is gratefully acknowledged. We are thankful to Bjoergvin Hj\"orvarsson for pointing to the rotating clusters in the microscopy recordings. We thank Henrique Moyses and Jeremy Palacci for fruitful discussions. \subsection{Author contributions} G.S. and A.E. conceived the experiments, G.S. and S.G. conceived the simulations. M.A. devised the routine for particle coating and D.N. prepared the particles. G.S. conducted the experiments and the simulations and interpreted the results. M.S. gave guidance and advice. G.S., A.E. and S.G. wrote the paper. All authors reviewed the manuscript. \subsection{Correspondence} Correspondence and requests for materials should be addressed to G.S. ([email protected]) and to A.E. ([email protected]).
5410a4b3a1b09f1912adf6ec69edc991a5231d5d
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Currents of spin angular momentum play a central role in the field of spintronics. Significant contributions have been made by spin currents, such as control of magnetizations by spin transfer torque,\citep{PhysRevLett.84.3149,demidov2011control,berger1996emission,ralph2008spin} transmission of electric signals through insulators,\citep{kajiwara2010transmission,cornelissen2015long,:/content/aip/journal/apl/107/17/10.1063/1.4935074} thermoelectric conversion,\citep{uchida2008observation,uchida2010spin,jaworski2010observation,kirihara2012spin,7452553} and electric probing of insulator magnetization.\citep{nakayama2013spin,chen2013theory,althammer2013quantitative} In order to detect and utilize these spin based-phenomena, conversion between spin and charge currents is necessary. For realizing efficient spin-to-charge current conversion, a wide range of materials are currently being investigated, including metals,\citep{saitoh2006conversion,kimura2007room,miao2013inverse,liu2012spin,mosendz2010quantifying,wang2014scaling,valenzuela2006direct,mendes2014large,niimi2015reciprocal,Azevedo:2005kw} semiconductors,\citep{wunderlich2005experimental,sinova2004universal,kato2004observation,ando2011electrically,ando2012observation,murakami2003dissipationless,chen2013direct} organic materials,\citep{SunSchootenKavandEtAl2016,ando2013solution} carbon-based materials,\citep{dushenko2015experimental} and topological insulators.\citep{shiomi2014spin,AndoHamasakiKurokawaEtAl2014,mellnik2014spin} Finding materials suitable to the spin-to-charge conversion is thus indispensable to making spintronic devices. One of the popular methods of the spin-to-charge conversion is the inverse spin Hall effect (ISHE),\citep{Azevedo:2005kw,saitoh2006conversion} which is the reciprocal effect of the spin Hall effect\citep{hirsch1999spin,kato2004observation,sinova2004universal,murakami2003dissipationless} caused by spin-orbit interaction. In the ISHE, a spin current generates a transverse charge current in a conductor such as Pt. Since the first demonstration of the ISHE in Pt and Al,\citep{saitoh2006conversion,valenzuela2006direct} it has been extensively studied because of its versatility.\citep{hoffmann2013spin,RevModPhys.87.1213} Dynamical generation of spin currents can be achieved by the spin pumping (SP).\citep{mizukami2002effect,Tserkovnyak:2005fr,tserkovnyak2002enhanced} At the interface between a normal conductor (N) and a ferromagnet (F), the SP causes emission of spin currents into the N layer from magnetization dynamics in the adjacent F layer. Such the magnetization dynamics is typically triggered by applying a microwave field; at the ferromagnetic resonance (FMR) or the spin wave resonance (SWR) condition, the magnetization resonantly absorbs the microwave power and exhibits a coherent precessional motion. A part of the angular momenta stored in this precessional motion is the source of the spin current generated by the SP. The combination of the ISHE and the SP enables electric detection and generation of spin currents.\citep{saitoh2006conversion,Azevedo:2005kw} This is the setup commonly used to study the properties of spin-to-charge current conversion and spin transport in materials. A spin current is injected into an N layer by the SP and is converted to a measurable electromotive force by the ISHE. The conversion efficiency between spin and charge currents in this process can be determined by estimating the density of the injected spin current by analyzing microwave spectra.\citep{ando2011inverse,mosendz2010quantifying} The spin transport property of a material can be investigated by constructing a heterostructure in which the material of interest is placed between a spin-current injector and detector layers.\citep{shikoh2013spin,dushenko2015experimental,watanabe2014polaron} \begin{figure} \begin{centering} \includegraphics{Fig1SpandRectHeat} \par\end{centering} \caption{(Color online) A schematic illustration of the SP and ISHE processes. The SP induces an spin current, $\mathbf{j}_{{\rm s}}$, and the ISHE convert $\mathbf{j}_{{\rm s}}$ into an charge current, $\mathbf{j}_{{\rm c,ISHE}}$. A microwave driving the FMR induces an rf current, $\mathbf{j}_{{\rm rf}}$, causing a dc current $j_{{\rm rect}}$ via galvanomagnetic effects. The absorbed microwave power at FMR induces a temperature gradient, $\nabla T$, causing a thermoelectric voltage. These process results in unwanted signals.\label{fig:A-schematic-illustration}} \end{figure} It should be noted that the voltage signal from the ISHE can be contaminated by other contributions in practice experiments. We thus need to extract the ISHE contribution by separating or minimizing the unwanted signals in order to ensure the validity of the measurements.\citep{inoue2007detection,mosendz2010quantifying,azevedo2011spin,chen2013direct,iguchi2013effect,PhysRevLett.111.217602} The ISHE signal is characterized by Lorentzian spectral shape and sign change under the magnetization reversal,\citep{ando2011inverse} and some of the unwanted signals show the same spectral shape and the sign change in configurations commonly used in the SP experiments.\citep{chen2013dc,chen2013direct} Such signals can be induced by a temperature gradient via thermoelectric effects\citep{bakker2012thermoelectric,schultheiss2012thermoelectric,shiomi2014spin,qiu2012all} and by an rf current via rectification effects (See Fig. \ref{fig:A-schematic-illustration}).\citep{Egan:1963gr,1970PhRvB...2.2651M,2007ApPhL..90r2507Y,Gui:2007fb} The heat emitted by excited magnetization dynamics induces a thermal gradient in a sample, resulting in an electromotive force due to the conventional Seebeck effect. This type of heat induced signals can be eliminated if one designs experimental conditions appropriately. Rectification effects comes from interplay between stray rf currents induced by an incident microwave and galvanomagnetic effects coming from oscillating magnetizations. The direction, magnitude, and phase of the stray rf current contain uncertainty because they depend on the details of an experimental setting, so the signals from the rectification effects tend to be complicated. Since the rectification effects were first observed in 1963 in a Ni film with the anisotropic magnetoresistance (AMR) and the anomalous Hall effect (AHE),\citep{Egan:1963gr} we can not make light of the effects. A number of works regarding to the extraction of the ISHE signal from other contributions in electric measurements on the SP have been reported.\citep{inoue2007detection,mosendz2010quantifying,azevedo2011spin,chen2013direct,iguchi2013effect,PhysRevLett.111.217602} In this article, we review the previous studies of the voltage signals induced by the SP and further introduce some methods to analyze the signals on the FMR. Here, we will focus on the experiments in a microwave cavity, so that small density of induced rf currents and thus small rectification contribution can be expected. The original field distribution in a cavity is minimally disturbed by placing a sample in it because the empty region in the cavity does not carry rf currents. In experiments, it is often difficult to identify the origin of the stray rf currents because it depends on an individual setup: the sample structure including wires for electric measurements. In this paper, we describe methods to separate the SP contribution from other artifacts in the electric measurements by introducing parameter dependence to the voltages in the presence of the stray rf currents. The methods can be applied to systems with a microstrip line or coplanar waveguide, but the rf-current-induced magnetization excitation due to the SHE, or the spin transfer torque FMR (STT-FMR), \citep{chiba2014current,liu2011spin} is neglected. This article is organized as follows. In Section \ref{sec:signals-due-to}, we show the analytical descriptions of the signals from the SP, rectification effects, and heating effects based on the magneto-circuit theory\citep{Tserkovnyak:2005fr} and the Landau-Lifshitz-Gilbert (LLG) equation\citep{gilbert2004phenomenological}. In Section \ref{sec:Separation-methods}, we discuss the dependence of the signals on the FMR spectrum, sample geometry, magnetization orientation, and excitation frequency. Here, we point out the similarity between the voltage signals due to the SP and rectification effects. Finally, in Section \ref{sec:Summary}, we give a summary of the methods based on the dependences described in Section \ref{sec:Separation-methods} and emphasize that the voltage measurement of the in-plane magnetization angular dependence with out-of-plane microwave magnetic field in a properly designed system is the most reliable way and thus enables quantitative studies of the SP on various kinds of materials. \section{Signals due to spin pumping and microwave effects in bilayer systems\label{sec:signals-due-to}} In this section, we discuss the analysis of the voltage signals induced by the SP, rectification effects, and heating effects. For the calculation of the SP-induced ISHE signal, we consider a spin current generated by magnetization dynamics via the SP in a bilayer film consisting of a normal conductor (N) and a ferromagnet (F). The rectification signals are expressed using the derived magnetization dynamics for the SP. For the heating-induced signals, we discuss thermoelectric effects due to the heating at the FMR and SWR of magnetostatic surface spin waves. \begin{figure} \begin{centering} \includegraphics{FigSP} \par\end{centering} \caption{(Color online) (a) F/N bilayer system under the SP. F (N) denotes the ferromagnet (normal conductor) layer. (b) Normalized FMR spectra of ${\rm Ni}_{81}{\rm Fe}_{19}$ and ${\rm Ni}_{81}{\rm Fe}_{19}$/Pt systems, where $H_{0}$ denotes the strength of an applied field and $H_{{\rm r}}$ denotes the FMR field. \label{fig:pumped-and-back}} \end{figure} \subsection{Spin current induced by spin pumping} The spin pumping (SP) is the phenomenon that a magnetically excited F layer induces a spin current into the N layer placed adjacent to it.\citep{mizukami2002effect,tserkovnyak2002enhanced} Here, let us suppose the N and the F layers span $xy$ plane, and they are stacked in z direction {[}Fig. \ref{fig:pumped-and-back}(a){]}. The spin current density through the F/N interface due to the SP is given by\citep{Tserkovnyak:2005fr} \begin{equation} \mathbf{j}_{{\rm s}}^{{\rm {\rm pump}}}=\frac{\hbar}{4\pi}g_{{\rm r}}^{\uparrow\downarrow}\left(\mathbf{m}\times\dot{\mathbf{m}}\right)+\frac{\hbar}{4\pi}g_{{\rm i}}^{\uparrow\downarrow}\dot{\mathbf{m}},\label{eq:sp} \end{equation} where $\mathbf{m}$ denotes the unit vector along the magnetization in the F layer, $\mathbf{\dot{m}}$ the time derivative of $\mathbf{m}$, and $g_{{\rm r}}^{\uparrow\downarrow}$ ($g_{{\rm i}}^{\uparrow\downarrow}$) the real (imaginary) part of mixing conductance per unit area, $\hbar$ the Planck constant. The spin polarization of $\mathbf{j}_{{\rm s}}^{{\rm pump}}$ is represented by its vector direction, and its flow direction is the interface normal $\boldsymbol{z}$. In diffusive N layers, the spin accumulation $\boldsymbol{\mu}_{{\rm s}}$ is formed owing to the pumped spin current. This spin accumulation induces a back-flow spin current into the F layer, and it renormalizes the mixing conductance in Eq. (\ref{eq:sp}) to an effective one denoted by $g_{{\rm eff}}^{\uparrow\downarrow}$. The back-flow spin current density from the magneto-circuit theory is given by\citep{Tserkovnyak:2005fr} \begin{equation} \mathbf{j}_{{\rm s}}^{{\rm back}}=\frac{1}{4\pi}g_{{\rm r}}^{\uparrow\downarrow}\mathbf{m}\times\left(\boldsymbol{\mu}_{{\rm s}}^{{\rm F/N}}\times\mathbf{m}\right)+\frac{1}{4\pi}g_{{\rm i}}^{\uparrow\downarrow}\boldsymbol{\mu}_{{\rm s}}^{{\rm F/N}}\times\mathbf{m}\label{eq:back} \end{equation} when the spin relaxation is fast enough and $\boldsymbol{\mu}_{{\rm s}}^{{\rm F/N}}\propto\mathbf{j_{{\rm s}}^{{\rm pump}}}$ holds,\citep{jiao2013spin} where $\boldsymbol{\mu}_{{\rm s}}^{{\rm F/N}}$ denotes the spin accumulation at the F/N interface ($z=0$). The solution without the approximation can be found in Ref. \citealp{jiao2013spin}. Then, combining Eqs. (\ref{eq:sp}) and (\ref{eq:back}), one can find the net dc spin current density, \begin{equation} \mathbf{j}_{{\rm s}}^{{\rm {\rm F/N}}}=\mathbf{j}_{{\rm s}}^{{\rm pump}}-\mathbf{j}_{{\rm s}}^{{\rm back}}=\frac{\hbar}{4\pi}g_{{\rm r,eff}}^{\uparrow\downarrow}\mathbf{m}\times\dot{\mathbf{m}}+\frac{\hbar}{4\pi}g_{{\rm i,eff}}^{\uparrow\downarrow}\dot{\mathbf{m}},\label{eq:jsdc0} \end{equation} where $g_{{\rm r\left(i\right),eff}}^{\uparrow\downarrow}$ represents the real (imaginary) part of the effective mixing conductance per unit area. The mixing conductance at the F/N interfaces has been widely investigated in many combinations of materials.\citep{weiler2013experimental,rojas2014spin,zhang2015role} Hereafter $g_{{\rm i}{\rm ,eff}}^{\uparrow\downarrow}$ is omitted for simplicity because it is much smaller than the real part.\citep{Xia:2002ug,Jia:2011gm} The expression of $g_{{\rm r,eff}}^{\uparrow\downarrow}$ can be obtained in terms of the parameters of the N layer. $\boldsymbol{\mu}_{{\rm s}}^{{\rm F/N}}$ is calculated from the spin accumulation profile $\boldsymbol{\mu}_{{\rm s}}\left(z\right)$ determined by the spin diffusion equation\citep{jiao2013spin} \begin{equation} \frac{\partial}{\partial t}\boldsymbol{\mu}_{{\rm s}}\left(z\right)=-\gamma_{{\rm N}}\boldsymbol{\mu}_{{\rm s}}\left(z\right)\times\mu_{0}\mathbf{H}_{0}+D\nabla^{2}\boldsymbol{\mu}_{{\rm s}}\left(z\right)-\frac{\boldsymbol{\mu}_{{\rm s}}\left(z\right)}{\tau_{{\rm s}}}\label{eq:diffeq} \end{equation} with the boundary conditions: $-\frac{\hbar\sigma_{{\rm N}}}{4e^{2}}\nabla\boldsymbol{\mu}_{{\rm s}}\left(0\right)=\mathbf{j}_{{\rm s}}^{{\rm F/N}}$ at the interface ($z=0$) and\linebreak{} $-\frac{\hbar\sigma_{{\rm N}}}{4e^{2}}\nabla\boldsymbol{\mu}_{{\rm s}}\left(d_{{\rm N}}\right)=0$ at the outer boundary of the N layer ($z=d_{{\rm N}}$). $\sigma_{{\rm N}\left({\rm F}\right)}$ and $d_{{\rm N}\left({\rm F}\right)}$ are the conductivity and thickness of the N (F) layer. $\gamma_{{\rm N}}$ denotes the gyromagnetic ratio of electrons in the N layer, $e$ the electron charge, $\mu_{0}$ the permittivity of a vacuum, $\mathbf{H}_{0}$ an external field, $D$ the diffusion constant, and $\tau_{{\rm s}}$ the spin-relaxation time. We focus on the regime where the Hanle effect is negligibly small (a rigorous treatment can be found in Ref. \citealp{ando2011electrically}). Then, $\boldsymbol{\mu}_{{\rm s}}\left(z\right)$ is obtained as \begin{equation} \boldsymbol{\mu}_{{\rm s}}\left(z\right)=\frac{4e^{2}}{\hbar\sigma_{{\rm N}}}\lambda\frac{\cosh\left[\left(z-d_{{\rm N}}\right)/\lambda\right]}{\sinh\left(d_{{\rm N}}/\lambda\right)}\mathbf{j}_{{\rm s}}^{{\rm {\rm F/N}}},\label{eq:mus} \end{equation} where $\lambda\equiv\sqrt{D\tau_{{\rm s}}}$ denotes the spin diffusion length. Using Eqs. (\ref{eq:back}), (\ref{eq:jsdc0}), and \eqref{eq:mus}, we find that the effective mixing conductance is given by \begin{equation} g_{{\rm r,eff}}^{\uparrow\downarrow}=\left(\frac{1}{g_{r}}+\frac{\pi\hbar\sigma_{{\rm N}}}{e^{2}\lambda}\text{tanh}\frac{d_{{\rm N}}}{\lambda}\right)^{-1}.\label{eq:geff} \end{equation} \subsection{Magnetization dynamics and spin current} Next, let us examine the effect of the SP on magnetization dynamics. We will calculate the effective mixing conductance and the magnitude of a pumped spin current in terms of observable parameters in experiments. \begin{figure} \begin{centering} \includegraphics{FigConfig} \par\end{centering} \caption{(Color online) (a) Spatial coordinate of the system, where $\mathbf{m}_{0}$ denotes unit vector in the direction of the equilibrium magnetization and is identical with $\mathbf{e}_{r}$. (b) Spectral shapes of ${\rm Lor}\left(\omega,\omega_{{\rm r}}\right)$ and ${\rm Asym}\left(\omega,\omega_{{\rm r}}\right)$. \label{fig:(a)-Spatial-coordinate}} \end{figure} The SP affects magnetization dynamics. When the incoming and outgoing spin currents, given by $\gamma{\rm div}\left({\rm \mathbf{j}_{s}^{{\rm F/N}}}\right)/\left(d_{{\rm F}}I_{{\rm s}}\right)$, are included, the LLG equation\citep{gilbert2004phenomenological} is modified as \begin{eqnarray} \dot{\mathbf{m}} & = & -\gamma\mathbf{m}\times\mu_{0}\left[\mathbf{H}_{{\rm eff}}\left(\mathbf{m}\right)+\mathbf{h}_{{\rm rf}}\left(t\right)\right]\nonumber \\ & & +\left(\alpha_{0}+\frac{\gamma}{d_{{\rm F}}I_{{\rm s}}}\frac{\hbar}{4\pi}g_{{\rm r,eff}}^{\uparrow\downarrow}\right)\mathbf{m}\times\dot{\mathbf{m}}\label{eq:LLG-1} \end{eqnarray} where $\gamma$ denotes the gyromagnetic ratio of the ferromagnet F, $\mathbf{H}_{{\rm eff}}\left(\mathbf{m}\right)$ the effective field, $\mathbf{h}_{{\rm rf}}\left(t\right)$ an microwave field, $\alpha_{0}$ the Gilbert damping constant without spin current exchange, and $I_{{\rm s}}$ the saturation magnetization. The effective field is given by $\mathbf{H}_{{\rm eff}}\left(\mathbf{m}\right)=-\nabla_{\mathbf{m}}F_{m}/I_{{\rm s}}$, where the magnetostatic energy $F_{m}$ includes magnetocrystalline anisotropy. The second term in the second line of Eq. (\ref{eq:LLG-1}) indicates that the SP acts as an additional damping term.\citep{mizukami2002effect} The reverse process is also demonstrated; a spin current injected from the N layer reduces the damping of the F layer.\citep{ando2008electric} Finding the enhanced damping thus directly relates to $g_{{\rm r,eff}}^{\uparrow\downarrow}$. In practical experiments, the total damping term represented by $\alpha\mathbf{m}\times\dot{\mathbf{m}}$ is observed. By measuring $\alpha$ from the FMR spectra of F/N and F films,\citep{tserkovnyak2002spin,mizukami2002effect} the change in the Gilbert damping constant can be found , i.e. $\Delta\alpha=\alpha_{{\rm F/N}}-\alpha_{{\rm F}}$. Figure \ref{fig:pumped-and-back}(b) shows an example of the comparison of the FMR microwave spectrum for ${\rm Ni}_{81}{\rm Fe}_{19}$/Pt (F/N) bilayer and ${\rm Ni}_{81}{\rm Fe}_{19}$ (F) single-layer samples, where broadening of the F/N spectral peak can be seen. By fitting the spectral peak using Lorentzian, $\alpha$ is obtained from the full-width at the half-maximum (FWHM) which has the relation $\Delta H=\left(\partial\omega_{{\rm r}}/\partial H_{0}\right)^{-1}\alpha\left(\omega_{\theta}+\omega_{\phi}\right)$ for field strength $H_{0}$ swept measurements,\citep{mizukami2002effect} where $\omega$ and $\omega_{{\rm r}}$ respectively denote the angular frequency of the magnetization precession and that at the resonance determined by $\omega_{\theta}$ and $\omega_{\phi}$ derived below. Finding the value of $\Delta\alpha$, one obtains \begin{equation} g_{{\rm r,eff}}^{\uparrow\downarrow}=4\pi\frac{I_{{\rm s}}}{\gamma\hbar}\Delta\alpha d_{{\rm F}}.\label{eq:geffexp} \end{equation} The measurement of $\Delta\alpha$ requires some care. Since the magnetic properties of a ferromagnetic film in a heterostructure are affected by the other part of the structure,\citep{johnson1996magnetic} there can be contributions on $\Delta\alpha$ other than the SP through inhomogeneous broadening and two-magnon scattering. Thus careful comparison of $\alpha$ is required.\citep{mizukami2002effect,PhysRevB.73.144424,zakeri2007spin} In order to find the spectral shape and magnitude of the spin current $\mathbf{j}_{{\rm s}}^{{\rm F/N}}$, we calculate the magnetization dynamics $\mathbf{m}\times\dot{\mathbf{m}}$ from Eq. (\ref{eq:LLG-1}). Here, we consider precessing magnetization $\mathbf{m}_{{\rm rf}}\equiv{\rm Re}\left[\left(m_{\theta}\mathbf{e}_{\theta}+m_{\phi}\mathbf{e}_{\phi}\right)\exp\left(i\omega t\right)\right]$ around the equilibrium magnetization vector $\mathbf{m}_{0}$ excited by an microwave field $\mathbf{h}_{{\rm rf}}\equiv{\rm Re}\left[\left(h_{\theta}\mathbf{e}_{\theta}+h_{\phi}\mathbf{e}_{\phi}\right)\exp\left(i\omega t\right)\right]$ as shown in Fig. \ref{fig:(a)-Spatial-coordinate}. $\mathbf{m}_{0}$ is determined by the condition $\mathbf{m}_{0}\times\mathbf{H}_{{\rm eff}}=0$. The unit vectors of the polar coordinate $\mathbf{e}_{p}\left(p=r,\theta,\phi\right)$ in which $\mathbf{e}_{r}$ points $\mathbf{m}_{0}$ have a relation to the unit vectors of the Cartesian coordinate $\mathbf{e}_{i}\left(i=x,y,z\right)$, $\mathbf{e}_{p}=\sum_{i=x,y,z}u_{pi}\mathbf{e}_{i}$ with \begin{equation} u=\left(\begin{array}{ccc} \sin\theta_{{\rm M}}\cos\phi_{{\rm M}} & \sin\theta_{{\rm M}}\sin\phi_{{\rm M}} & \cos\theta_{{\rm M}}\\ \cos\theta_{{\rm M}}\cos\phi_{{\rm M}} & \cos\theta_{{\rm M}}\sin\phi_{{\rm M}} & -\sin\theta_{{\rm M}}\\ -\sin\phi_{{\rm M}} & \cos\phi_{{\rm M}} & 0 \end{array}\right),\label{eq:u} \end{equation} where $\theta_{{\rm M}}$ and $\phi_{{\rm M}}$ denote the polar angle between the $z$ axis and $\mathbf{m}_{0}$, and the azimuthal angle measured from the $x$ axis, respectively. Taking the time average of the pumped spin current in Eq. \eqref{eq:jsdc0} yields \begin{equation} \left\langle \mathbf{j}_{{\rm s}}^{{\rm F/N}}\right\rangle _{t}=\frac{\hbar\omega}{4\pi}g_{r,{\rm eff}}^{\uparrow\downarrow}{\rm Im}\left[m_{\theta}m_{\phi}^{*}\right]\mathbf{m}_{0},\label{eq:spin pump} \end{equation} where $\left\langle \cdots\right\rangle _{t}$ and $a^{*}$ mean temporal average and complex conjugate of $a$, respectively. The relation between the magnitude of $\mathbf{m}_{{\rm rf}}$ and $\mathbf{h}_{{\rm rf}}$ obtained from the Eq. \eqref{eq:LLG-1} is reduced to $(m_{\theta},m_{\phi})=\boldsymbol{\chi}.(h_{\theta},h_{\phi})$, where the susceptibility $\boldsymbol{\chi}$ is given by\citep{smit1959ferrites} \begin{equation} \boldsymbol{\chi}\approx\frac{\gamma\mu_{0}}{\alpha\omega}\frac{S\left(\omega,\omega_{{\rm r}}\right)}{\omega_{\theta}+\omega_{\phi}}\left(\begin{array}{cc} \omega_{\theta} & -\omega_{\theta\phi}+i\omega\\ -\omega_{\theta\phi}-i\omega & \omega_{\phi} \end{array}\right)\label{eq:chi-1} \end{equation} with $\omega_{\theta}\equiv\omega_{\phi\phi}-\frac{\partial}{\partial m_{r}}F_{m}$, $\omega_{\phi}\equiv\omega_{\theta\theta}-\frac{\partial}{\partial m_{r}}F_{m}$, and \begin{equation} \omega_{pq}\equiv\frac{\gamma\mu_{0}}{I_{{\rm s}}}\frac{\partial}{\partial m}_{p}\frac{\partial}{\partial m}_{q}F_{m}\ \left(p,q\in\{\theta,\phi\}\right).\label{eq:omegath} \end{equation} $S\left(\omega,\omega_{{\rm r}}\right)$ represents a spectrum function, \begin{equation} S\left(\omega,\omega_{{\rm r}}\right)\equiv\frac{\alpha\omega}{\left(\omega_{{\rm r}}-\omega\right)^{2}+\left(\alpha\omega\right)^{2}}\left[\left(\omega_{{\rm r}}-\omega\right)-i\alpha\omega\right], \end{equation} which real part represents asymmetric spectrum, ${\rm Asym\left(\omega,\omega_{{\rm r}}\right)\equiv{\rm Re}}\left[S\left(\omega,\omega_{{\rm r}}\right)\right]$ (known as asymmetric Lorentzian), and which imaginary part represents symmetric spectrum, ${\rm Lor}\left(\omega,\omega_{{\rm r}}\right)\equiv{\rm Im}\left[S\left(\omega,\omega_{{\rm r}}\right)\right]$ (Lorentzian). The resonance angular frequency $\omega_{{\rm r}}$ is given by \begin{equation} \omega_{{\rm r}}\equiv\sqrt{\omega_{\theta}\omega_{\phi}-\omega_{\theta\phi}^{2}}.\label{eq:omegares} \end{equation} Calculation using Eq. (\ref{eq:chi-1}) yields \begin{equation} \left\langle \mathbf{j}_{{\rm s}}^{{\rm F/N}}\right\rangle _{t}=\frac{\hbar g_{r,{\rm eff}}^{\uparrow\downarrow}}{4\pi}\frac{\gamma^{2}\mu_{0}^{2}\left(\omega_{\theta}\left|h_{\theta}\right|^{2}+\omega_{\phi}\left|h_{\phi}\right|^{2}+{\rm Im}\left[h_{\theta}h_{\phi}^{*}\left\{ \omega_{\theta}\omega_{\phi}-\left(\omega_{\theta\phi}+i\omega\right)^{2}\right\} /\omega\right]\right)}{\alpha^{2}\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}{\rm Lor}\left(\omega,\omega_{{\rm r}}\right)\mathbf{m}_{0},\label{eq:jsdc} \end{equation} where we used a relation, $\left|S\left(\omega,\omega_{{\rm r}}\right)\right|^{2}={\rm Lor}\left(\omega,\omega_{{\rm r}}\right)$. Equation \eqref{eq:jsdc} means that the spectral shape of $\left\langle \mathbf{j}_{{\rm s}}^{{\rm F/N}}\right\rangle _{t}$ is Lorentzian.\citep{saitoh2006conversion} Equation \eqref{eq:jsdc} can be also expected in terms of the elliptic precession with the cone angle $\Theta$ in Refs. \citealp{mosendz2010quantifying,Ando:2009gh,czeschka2011scaling}: at resonance $\mathbf{m}_{{\rm rf}}={\rm Re}\left[\sin\Theta\exp\left(i\omega t\right)\mathbf{e}_{\theta}+iA\sin\Theta\exp\left(i\omega t\right)\mathbf{e}_{\phi}\right]$ and thus ${\rm Im}\left[m_{\theta}m_{\phi}^{*}\right]=A\sin^{2}\Theta$, where $A$ is an correction factor for the elliptical precession motion. The magnetization excitation leads to decreased power of the incident microwave and thus microwave measurements are useful for determining $\alpha$ and other parameters related to the resonance condition. The microwave power absorption on the FMR per unit volume is calculated by \begin{eqnarray} \Delta P & = & -\frac{\omega}{2\pi}\int_{0}^{\frac{2\pi}{\omega}}\mathbf{h}\cdot\dot{\mathbf{m}}dt\nonumber \\ & = & \frac{\gamma\mu_{0}}{2}\frac{\left(\omega_{\theta}\left|h_{\theta}\right|^{2}+\omega_{\phi}\left|h_{\phi}\right|^{2}-2{\rm Re}\left[h_{\theta}h_{\phi}^{*}\left(\omega_{\theta\phi}+i\omega\right)\right]\right)}{\alpha\left(\omega_{\theta}+\omega_{\phi}\right)}{\rm Lor}\left(\omega,\omega_{{\rm r}}\right).\label{eq:delP} \end{eqnarray} The transferred energy from the microwave to the magnetization dynamics finally results in heat via damping processes of the dynamics, which causes thermoelectric signals. In this calculation, we assumed that $\mathbf{h}_{{\rm rf}}$ is only induced by an applied microwave field and is not affected by an induced rf current. Rf currents are known to trigger the FMR through generation of rf spin currents via the SHE.\citep{liu2011spin} This contribution can be included to $\mathbf{h}_{{\rm rf}}$ by calculating the spin-transfer torque due to the absorption of the rf spin current.\citep{liu2011spin,chiba2014current} Such a contribution will result in a phase shift between the actual rf current and the rf current determined by the analysis provided in this article. \subsection{Voltage generated by inverse spin Hall effect } Here, we describe the electromotive force generated by the SP and ISHE in the N layer by taking account of the spin current profile in the N layer. The pumped spin current in the N layer gives rise to an electromotive force due to the ISHE. The ISHE induces a charge current density transverse to both the spin polarization ($\propto\mathbf{j}_{{\rm s}}$) and its flow direction $\left(\propto\mathbf{z}\right)$, which can be expressed as\citep{ando2011inverse} \begin{equation} \mathbf{j}_{{\rm c,ISHE}}=\frac{2e}{\hbar}\theta_{{\rm SHE}}\mathbf{j}_{{\rm s}}\times\mathbf{z}. \end{equation} Therefore, as $\left\langle \mathbf{j}_{{\rm s}}^{{\rm F/N}}\right\rangle _{t}\propto\mathbf{m}_{0}$ holds in the SP, the direction of the ISHE current is reversed under the magnetization reversal ($\mathbf{m}_{{\rm 0}}\rightarrow-\mathbf{m}_{{\rm 0}}$), which is an important feature of the ISHE induced by the SP. According to the short circuit model,\citep{nakayama2012geometry} the electromotive force is calculated by the sum of the induced current. The total dc current induced by the ISHE is given by \begin{equation} \mathbf{J}_{{\rm ISHE}}=w\frac{2e}{\hbar}\theta_{{\rm SHE}}\int_{0}^{d_{{\rm N}}}\mathbf{z}\times\left\langle \mathbf{j}_{{\rm s}}\left(z\right)\right\rangle _{t}dz\label{eq:JISHE} \end{equation} with $w$ being the width of a sample {[}See Fig. \ref{fig:Typical-configuration-of}(a){]}. One can observe the electromotive force $\mathbf{E}_{{\rm ISHE}}$ induced by $\mathbf{J}_{{\rm ISHE}}$, which satisfies the relation $\tilde{R}_{{\rm tot}}\mathbf{J}_{{\rm ISHE}}+\mathbf{E}_{{\rm {\rm ISHE}}}=0$ for an open circuit condition, where $\tilde{R}_{{\rm tot}}$ is the total resistance per unit length of the system, e.g. $\tilde{R}_{{\rm tot}}^{-1}=w\left(\sigma_{{\rm N}}d_{{\rm N}}+\sigma_{{\rm F}}d_{{\rm F}}\right)$ for F/N bilayer systems. The spin current profile is obtained from Eqs. (\ref{eq:jsdc0}) and (\ref{eq:mus}) as \begin{eqnarray} \mathbf{j}_{{\rm s}}\left(z\right) & = & -\frac{\hbar}{2e}\frac{\sigma_{{\rm N}}}{2}\nabla\boldsymbol{\mu}_{{\rm s}}\left(z\right)\\ & = & \mathbf{j}_{{\rm s}}^{{\rm {\rm F/N}}}\frac{\sinh\left(\left[d_{{\rm N}}-z\right]/\lambda\right)}{\sinh\left(d_{{\rm N}}/\lambda\right)}.\label{eq:jsz} \end{eqnarray} Then we yields \begin{equation} \mathbf{E}_{{\rm ISHE}}=w\tilde{R}_{{\rm tot}}\theta_{{\rm SHE}}\frac{2e}{\hbar}\lambda\tanh\frac{d_{N}}{2\lambda}\left\langle j_{{\rm s}}^{{\rm {\rm F/N}}}\right\rangle _{t}\mathbf{z}\times\mathbf{m}_{0}\label{eq:EISHE} \end{equation} from Eqs. (\ref{eq:jsdc0}), (\ref{eq:JISHE}), and (\ref{eq:jsz}). \begin{figure} \begin{centering} \includegraphics{FigSampleRod} \par\end{centering} \caption{(Color online) (a) Typical configuration of sample setup for the ISHE measurement by using the SP. (b) Schematic illustration of experimental configuration in a cavity. \label{fig:Typical-configuration-of} } \end{figure} A typical experimental setup for observing the ISHE voltage is depicted in Fig. \ref{fig:Typical-configuration-of}(b). A bilayer sample is placed on a cutout of a quartz rod, which has a through hole in the center for wires connected to a voltmeter. The wires are covered by an insulating polymer and connected to the electrodes in the $y$ axis. These wires measure the $y$ component of Eq. \eqref{eq:EISHE}, so that the maximized ISHE voltage is detected when the magnetization points along the $x$ axis. $\mathbf{m}_{0}$ is controlled by an applied field, $\mathbf{H}_{0}$. During the measurements, while $\omega$ is fixed, the field strength $H_{0}$ is swept so that $\omega_{{\rm r}}$ is changed. \subsection{Voltage generated by rectification effects} Rectification effects at the FMR are caused by an rf current in a sample possessing a galvanomagnetic effect, i.e. magnetization dependent resistivity $\rho\left(\mathbf{m}\right)$. Describing the resistivity which oscillates due to the precessing magnetization as $\tilde{\rho}\left(\mathbf{m}\right)\propto m_{p}\cos\left(\omega t\right)$ and an rf current as $j_{{\rm rf}}=j_{{\rm rf}}^{0}\cos\left(\omega t+\psi\right)$, a rectified dc electromotive force, $E_{{\rm rect}}$, is given by \begin{eqnarray} E_{{\rm rect}} & \propto & m_{p}\cos\left(\omega t\right)\cdot j_{{\rm rf}}^{0}\cos\left(\omega t+\psi\right)\nonumber \\ & \propto & \frac{jm_{p}}{2}\left[\cos\psi+O\left(t\right)\right], \end{eqnarray} where $\psi$ denotes the phase difference between the precessing magnetization and the rf current, as depicted in Fig. \ref{fig:possible-origins-of}(a). Here, we will consider the anisotropic magnetoresistance (AMR), the anomalous Hall effect (AHE), and the spin Hall magnetoresistance (SMR), which are common examples of galvanomagnetic effects in the bilayer systems used in the SP measurements. Note that any effect leading to magnetization dependent resistance gives a dc voltage signal. Therefore the tunnel magnetoresistance,\citep{yuasa2004giant,ikeda2008tunnel,miyazaki1995giant} the colossal magnetoresistance,\citep{ramirez1997colossal} and the spin accumulation Hall effect\citep{hou2015hall} can give the rectified dc voltages which can be calculated in the way described below. The induced voltage is calculated from the Ohm's law $E_{i}=\rho_{ij}(\mathbf{m})j_{{\rm rf}}^{j}$, where $j_{{\rm rf}}^{i}$ denotes the $i$-th component of an rf current passing through a sample.\citep{Egan:1963gr} For an F layer with the AMR and AHE, the resistivity tensor can be expanded in terms of the magnetization by \begin{equation} \rho_{ij}^{{\rm F}}\left(\mathbf{m}\right)=\rho_{ij}^{0}+\Delta\rho_{{\rm AMR}}m_{i}m_{j}+\rho_{{\rm AHE}}\sum_{l=x,y,z}\epsilon_{ijl}m_{l},\label{eq:restensor} \end{equation} where $m_{i}$ denotes the component in the $i$ axis and $\epsilon$ denotes the Levi-Civita tensor. $\rho_{ij}^{0}$ denotes the resistivity part insensitive to $\mathbf{m}$, $\Delta\rho_{{\rm AMR}}$ the AMR coefficient, and $\rho_{{\rm AHE}}$ the AHE coefficient. For an N layer with the SMR, whose magnitude is $\Delta\rho_{{\rm SMR}}$, the resistivity tensor is given by\citep{0953-8984-28-10-103004} \begin{equation} \rho_{ij}^{{\rm N}}\left(\mathbf{m}\right)=\rho_{ij}^{0}-\Delta\rho_{{\rm SMR}}\sum_{k=x,y}\epsilon_{ikz}m_{k}\sum_{l=x,y}\epsilon_{jlz}m_{l}+\rho_{{\rm AHE}}^{{\rm N}}\epsilon_{ijz}m_{z}.\label{eq:restensor-1} \end{equation} Substituting $m_{i}=\sum_{p=\theta,\phi}u_{pi}m_{p}$ with $u_{pi}$ in Eq. (\ref{eq:u}) and extracting the components proportional to precessing magnetization, $m_{\theta}$ and $m_{\phi}$, yields the time-dependent resistivity tensor $\tilde{\rho}^{{\rm F}\left({\rm N}\right)}$ for the F (N) layer {[}cf. Ref. \citealp{iguchi2013effect}{]}. Then the $i$-th component of the rectified dc current density from the $j$-th component of the rf current is given by $\sigma_{{\rm F(N)}}{\rm Re}\left[\tilde{\rho}_{ij}^{{\rm F(N)}}j_{{\rm rf}}^{j*}\right]/2$ for the F(N) layer. Considering the short circuit model, the electromotive force $\mathbf{E}_{{\rm rect}}$ is obtained as \begin{equation} \mathbf{E}_{{\rm rect}}=\tilde{R}_{{\rm tot}}\left(\int_{-d_{{\rm F}}}^{0}\frac{\sigma_{{\rm F}}}{2}{\rm Re}\left[\tilde{\rho}^{{\rm F}}\cdot\mathbf{j}_{{\rm rf}}^{*}\left(z\right)\right]dz+\int_{0}^{d_{{\rm N}}}\frac{\sigma_{{\rm N}}}{2}{\rm Re}\left[\tilde{\rho}^{{\rm N}}\cdot\mathbf{j}_{{\rm rf}}^{*}\left(z\right)\right]dz\right).\label{eq:Erect} \end{equation} An origin of $z$ dependence of $\mathbf{j}_{{\rm rf}}\left(z\right)$ comes from the skin effect (which appears as the Dyson effect\citep{dyson1955electron} for the microwave resonance experiments). $\mathbf{j}_{{\rm rf}}\left(z\right)$ is localized within the skin depth: $\delta_{{\rm skin}}=2\left(\sigma\mu\omega\right)^{-1/2}$, where $\mu$ denotes the permeability. $\delta_{{\rm skin}}$ is typically smaller than the thickness of systems used for the SP experiments; for example, $\delta_{{\rm skin}}$ for Cu at $\omega/2\pi$=10 GHz is estimated to be 0.6 $\mu$m, and the thickness scale used in the experiments is less than a few hundred nanometers. \begin{figure} \begin{centering} \includegraphics{FigRecTShape} \par\end{centering} \caption{(Color online) Spectral shapes of the rectification signal $E_{{\rm rect}}$ induced by $m_{\theta}$ and $m_{\phi}$ at $t=0$ (real part) and $t=\pi/\left(2\omega\right)$ (imaginary part)\label{fig:(c)-the-spectral}. $\psi$ is the phase difference between the rf field $h_{{\rm rf}}\exp\left(i\omega t\right)$ and the rf current $j_{{\rm rf}}\exp\left(i\left[\omega t+\psi\right]\right)$. } \end{figure} The in-plane electromotive force induced by an rf current which lies in the film plane ($xy$ plane) and homogeneous over $z$ is \begin{align} \left(\begin{array}{c} E_{{\rm rect}}^{x}\\ E_{{\rm rect}}^{y} \end{array}\right)= & \frac{\tilde{R}_{{\rm tot}}}{2}\sin\theta_{{\rm M}}\sum_{L={\rm F},{\rm N}}\sigma_{L}d_{L}\times\nonumber \\ & \left\{ \left[\cos\theta_{{\rm M}}\rho_{L}^{\theta}+\rho_{{\rm AHE}}^{L}\left(\begin{array}{cc} 0 & 1\\ -1 & 0 \end{array}\right)\right]{\rm Re}\left[m_{\theta}\left(\begin{array}{c} j_{{\rm rf}}^{x*}\\ j_{{\rm rf}}^{y*} \end{array}\right)\right]+\rho_{L}^{\phi}{\rm Re}\left[m_{\phi}\left(\begin{array}{c} j_{{\rm rf}}^{x*}\\ j_{{\rm rf}}^{y*} \end{array}\right)\right]\right\} \label{eq:Erectex} \end{align} with \begin{equation} \rho_{{\rm F(N)}}^{\theta}=\Delta\rho_{{\rm AMR(SMR)}}\left(\begin{array}{cc} +(-)1+\cos\left(2\phi_{{\rm M}}\right) & \sin\left(2\phi_{{\rm M}}\right)\\ \sin\left(2\phi_{{\rm M}}\right) & +(-)1-\cos\left(2\phi_{{\rm M}}\right) \end{array}\right), \end{equation} \begin{equation} \rho_{{\rm F(N)}}^{\phi}=\left(\begin{array}{cc} -\sin\left(2\phi_{{\rm M}}\right) & \cos\left(2\phi_{{\rm M}}\right)\\ \cos\left(2\phi_{{\rm M}}\right) & \sin\left(2\phi_{{\rm M}}\right) \end{array}\right), \end{equation} where + (-) sign on the first term of the diagonal parts corresponds to the F (N) layer. The magnitude and spectral shape are determined by ${\rm Re}\left[m_{p}j_{{\rm rf}}^{i{\rm *}}\right]$ through Eq. (\ref{eq:chi-1}). Figure \ref{fig:(c)-the-spectral} shows possibly induced spectra with various $j_{{\rm rf}}^{*}$ direction in response to the dynamic magnetizations, which exhibits both the Lorentzian and asymmetric Lorentzian. The rectification signals show linear dependence to the incident power because $m_{p}\propto h_{{\rm rf}}$ and $j_{{\rm rf}}\propto h_{{\rm rf}}$, which is same as the ISHE signal since $\left\langle j_{{\rm s}}^{{\rm F/N}}\right\rangle _{t}\propto{\rm Im}\left[m_{\theta}m_{\phi}^{*}\right]\propto h_{{\rm rf}}^{2}$. \begin{figure} \centering{}\includegraphics{FigRfcurrent}\caption{(Color online) (a) Schematic illustration of mechanism of the rectification effect. (b) Induced rf current in an isolated conductor under an rf electric field, $E_{{\rm rf}}$. $\varepsilon_{0}$ denotes the permittivity of a vacuum (c,d) Possible origins of stray rf currents: induced by in-plane (c) and out-of-plane (d) rf fields. (e,f) Rf currents picked up by the wires for the electric measurements. \label{fig:possible-origins-of}} \end{figure} In the following, we discuss the origin of the stray rf currents causing the extrinsic signals in the previous experiments. One possible origin of the rf current in a cavity is the generation due to the Ohm's law because of a non-zero rf electric field, $\mathbf{E}_{{\rm rf}}$, which is considered in most of the previous studies.\citep{saitoh2006conversion,ando2011inverse,chen2013direct,inoue2007detection} However, in an open circuit condition, a conductor smaller than the microwave wavelength, typically $\sim1$ cm for the X-band microwaves, can screen $\mathbf{E}_{{\rm rf}}$ below the plasma frequency, which results in a displacement current $\mathbf{j_{{\rm dis}}}\approx i\varepsilon\omega\mathbf{E}_{{\rm rf}}$ rather than the rf current $\mathbf{j}_{{\rm rf}}=\sigma\mathbf{E}_{{\rm rf}}$ induced by the Ohm's law {[}See Fig. \ref{fig:possible-origins-of}(b){]}. Here $\varepsilon$ is the permittivity. A simulation in Ref. \citealp{peligrad1998cavity} shows that the original cavity field distribution is modified so that $\mathbf{E}_{{\rm rf}}$ in a conductor is zero. $\mathbf{j}_{{\rm dis}}$ is usually much smaller than the $\mathbf{j}_{{\rm rf}}$, e.g. by a factor of $10^{8}$ for Cu at $\omega/2\pi=10$ GHz, and the phase of $\mathbf{j}_{{\rm dis}}$ is $90^{\circ}$ different from the $\mathbf{j}_{{\rm rf}}$, which indicates other consideration is necessary to explain previous results. The Faraday's law causes an rf current from a microwave magnetic field. An in-plane microwave magnetic field induces an eddy current in the cross section of the bilayer systems {[}See Fig. \ref{fig:possible-origins-of}(c){]}. Though the current directions are opposite in the F and N layers, the microwave current can induce a non-zero rectification signal due to the difference between $\tilde{\rho}^{{\rm F}}\left(\mathbf{m}\right)$ and $\tilde{\rho}^{{\rm N}}\left(\mathbf{m}\right)$. This is one of plausible contributions to the experiments. For out-of-plane microwave field, the voltage appears only when the electrodes are placed off center from the rf-eddy-current distribution {[}See Fig. \ref{fig:possible-origins-of}(d){]}. Such an effect was studied in Ref. \citealp{EuO}, which remarks that this effect can be eliminated by making the sample structure symmetric. Consideration on wires connected to samples for the ISHE measurements leads to two additional contributions by the Ohm's law and Faraday's law, which are shown in Fig. \ref{fig:possible-origins-of}(e,f). In one case, $\mathbf{j}_{{\rm rf}}$ can be proportional to $\sigma\mathbf{E}_{{\rm rf}}$ with a factor considering the modification of the field due to a conducting sample. $\mathbf{j}_{{\rm rf}}$ is generated and is transmitted through the paired or twisted wires forming an microwave transmission line. In the other case, $\mathbf{j}_{{\rm rf}}$ is generated by the induction around a sample forming a pick-up coil with wires. In both the cases, rf currents only appear along the electrode direction. As we discussed above, there are several mechanisms for the stray rf-current generation in cavities. The suppression of the rf current is not straightforward and thus the analysis based on $\mathbf{j}_{{\rm rf}}$ by leaving its magnitude, direction and phase undefined parameters, which are to be fitted with experimental data, is appropriate. The consideration of in-plane rf currents is enough for the analysis because the aforementioned mechanisms do not induce an rf current along the $z$ direction unless a pathway for the rf current is formed in the $z$ direction. This also holds for rf currents originating from the magnonic charge pumping,\citep{azevedo2015electrical,ciccarelli2015magnonic} the ac ISHE current due to the ac SP,\citep{wei2014spin,weiler2014phase} and other generation effects. \subsection{Voltage generated by heating\label{subsec:Induced-voltage-by}} \begin{figure} \begin{centering} \includegraphics{FigHeating} \par\end{centering} \caption{(Color online) (a,b) Spectral shape of electromotive forces due to heat induced by the FMR (a) and SWR (b). (c) Induced $\nabla T$ due to MSSWs. When the magnetization is reversed, $\nabla T$ can change its sign, resulting in the same symmetry as the ISHE ($\mathbf{E}_{{\rm ISHE}}\propto\mathbf{z}\times\mathbf{m}_{0}$). (d) Schematic illustration of difference in the dispersion relation for thin and thick ferromagnetic films. \label{fig:Electromotive-force-due}} \end{figure} Damping processes of magnetization dynamics generate heat, which can result in an electromotive force via thermoelectric effects. Thermoelectric effects, such as the Seebeck and Nernst effects, are seen in conducting materials by themselves regardless of the presence of the F layer. When a magnetization emits the absorbed power $\Delta P$ to phonons, a temperature gradient, $\nabla T\propto\Delta P\propto{\rm Lor}\left(\omega,\omega_{{\rm r}}\right)$, can be induced. $\nabla T$ is then formed into an electromotive force, e.g. $E_{{\rm SE}}\propto\nabla T$ via the Seebeck effect. While the spectral shape of $E_{{\rm SE}}$ on the FMR is Lorentzian, it becomes complicated on the SWR in thick ferromagnetic films, which is also used to drive the SP.\citep{sandweg2010enhancement} The SWR gives a different heat profile on each resonance,\citep{An:2013jn} so that the sign of $\nabla T$ easily changes. When the neighboring peaks have different signs, the signal should look like the asymmetric Lorentzian, as shown in Fig. \ref{fig:Electromotive-force-due}(b). As a result, the total thermoelectric voltage becomes the superposition of the symmetric and asymmetric Lorentzian for the SWR case. The Seebeck effect contributes to the voltage signal when the heat profile produced by the FMR or SWR is not symmetrically distributed with respect to the electrodes for the voltage detection.In most systems, it does not have explicit dependence on the magnetization direction. However, there can be a case where $\nabla T$ is sensitive to the magnetization direction as is seen in a system comprised of a thick ferromagnetic film.\citep{an2013unidirectional} In such films, spin waves known as magnetostatic surface spin waves (MSSWs) localize on the top and bottom surfaces of the F layer and propagate non-reciprocally.\citep{stancil2009spin} MSSWs are demonstrated to convey heat to an arbitrary direction controlled by the magnetization polarization utilizing the non-reciprocity by unbalanced excitation of spin waves with wavevectors $+\mathbf{k}$ and $-\mathbf{k}$.\citep{an2013unidirectional} This results in $\nabla T\propto\mathbf{k}$ and thus induces a Seebeck voltage in the N layer, $\mathbf{E}_{{\rm SE}\left({\rm MSSW}\right)}\propto\Delta P\mathbf{k}$. In microstrip antenna excitation, $\mathbf{k}$ of the dominant MSSWs is reversed under the magnetization reversal, $\mathbf{m}_{0}\rightarrow-\mathbf{m}_{0}$.\citep{an2013unidirectional,schneiderphase} This is because $\mathbf{k}$ of the MSSWs localized at a surface with its normal $\mathbf{n}$ is determined by $\mathbf{m}_{0}\times\mathbf{n}$.\citep{stancil2009spin} Consequently, the sign of the induced heat current by the MSSW is reversed, resulting in a thermoelectric signal with the same symmetry as the ISHE, i.e. \begin{equation} \mathbf{E}_{{\rm SE}\left({\rm MSSW}\right)}=A_{{\rm SE\left(MSSW\right)}}\Delta P\mathbf{z}\times\mathbf{m}_{0},\label{eq:Ese} \end{equation} where $A_{{\rm SE\left(MSSW\right)}}$ is a constant determined by the Seebeck coefficient and the temperature profile due to the MSSW. In the cavity experiments on the SWR, though the spin waves with $+\mathbf{k}$ and $-\mathbf{k}$ are equally excited, asymmetries between the surfaces that these two modes are localized on can give the thermoelectric voltage in the same form as Eq. (\ref{eq:Ese}); as we place a N layer on top of the F layer, the inversion symmetry between the top and bottom layers is broken, so that the contributions from $+\mathbf{k}$ and $-\mathbf{k}$ give unequal contributions to $\nabla T$ in the N layer, which results in $\nabla T\propto\mathbf{k}$. Moreover, the existence of the substrate at the bottom surface may promote the asymmetry of the thermal conduction, possibly growing $\nabla T$ {[}See Fig. \ref{fig:Electromotive-force-due}(c){]}. Thus, the MSSW heating effect can appear regardless of the excitation methods in the F/N bilayer systems. This effect can be significant in materials with high thermoelectric conversion efficiency, such as low carrier density conductors. Other contributions come from the transverse thermoelectric effects reflecting field or magnetization direction, such as the Nernst-Ettingshausen, anomalous Nernst effect (ANE) and spin Seebeck effect (SSE). Neglecting the angular difference between an applied field and $\mathbf{m}_{0}$, the induced voltage is proportional to $\mathbf{m}_{0}\times\nabla T$. Thus when $\nabla T$ is formed in the thickness direction, it gives an in-plane electromotive force \begin{equation} \mathbf{E}_{{\rm TTE}}=A_{{\rm TTE}}\Delta P\mathbf{z}\times\mathbf{m}_{0}, \end{equation} where $A_{{\rm TTE}}$ denotes a proportionality constant determined by the magnitude of the transverse thermoelectric effects and $\nabla T$ along $z$ direction. Importantly, $\mathbf{E}_{{\rm TTE}}$ shows the same symmetry as $\mathbf{E}_{{\rm ISHE}}$. \section{Separation methods of SP-induced ISHE signal\label{sec:Separation-methods}} In this section, we will introduce a guideline to select proper materials for the F layer and four methods to extract the spin current contribution from observed signals. Here, we discuss the microwave contribution to the voltage signals in terms of spectral shape, thickness, magnetization angle, and excitation frequency dependences. Understanding these dependences, the rectification effects can be isolated by a measurement of magnetization angular dependence, and the heating effects can be isolated by that of frequency dependence. \subsection{Suitable sample design for measurements of ISHE driven by SP\label{subsec:Suitable-materials-for}} For electric measurements of the SP, an appropriate choice of materials for reducing the rectification and heating effects can improve the performance of the experiments. The first step of experiments of the ISHE induced by the SP thus is to consider the right choice of materials for the spin injector. The rectification effects can be suppressed by using a material with low galvanomagnetic coefficients. The coefficients represented by $\Delta\rho_{{\rm AMR}}$, $\rho_{{\rm AHE}}^{{\rm F}}$, $\Delta\rho_{{\rm SMR}}$, and $\rho_{{\rm AHE}}^{{\rm N}}$ in Eq. (\ref{eq:restensor}) are proportional to the signals. For the metallic spin injector, ${\rm Ni}_{81}{\rm Fe}_{19}$, so called Permalloy, is often used but other materials such as CoFe alloys with the low AMR ratio is a good candidate for the SP.\citep{haidar2015reducing,ganguly2014thickness} Similarly, the SMR is known to be small compared to the AMR, and thus the use of a ferrimagnetic insulator is effective.\citep{iguchi2013effect} The FMR and MSSW heating effects due to the conventional Seebeck effect can be minimized by making the sample structure symmetric about the electrodes and by reducing the thickness of the ferromagnetic layer $d_{{\rm F}}$, respectively. The feature due to magnetostatic interaction is dominant around $\left|\mathbf{k}\right|d_{{\rm F}}\sim1$. \citep{stancil2009spin} When $d_{{\rm F}}$ is decreased, such a value of $\mathbf{k}$ increases and eventually reaches the exchange regime where the magnetostatic feature is lost. Figure \ref{fig:Electromotive-force-due} illustrates the dispersion relation of spin waves for thick and thin ferromagnetic films. The manifold of the dispersion shrinks as the film thickness reduces. The group velocity of the MSSWs correspondingly becomes smaller,\citep{iguchi2013spin} and the heat conveyer effect eventually disappears. Depending on the strength of the Seebeck effect, the appropriate thickness is below $100$ nm for the measurements free from the MSSWs heating, which can be fabricated by pulsed laser deposition,\citep{PhysRevLett.107.066604} sputtering,\citep{:/content/aip/journal/apl/101/15/10.1063/1.4759039} or metal-organic decomposition.\citep{ishibashi2010magneto} The MSSW contribution can also be confirmed by a control experiment with the insertion of a thin nonmagnetic insulator layer between the N and F layers because the nonmagnetic insulator cuts the spin transport but allows heat transport.\citep{PhysRevLett.111.247202} \subsection{Spectral shape dependence } First, let us introduce a way to separate a measured electric signal into symmetric and antisymmetric parts with respect to reversal of magnetization. We will explain why this simple method does not work for isolating the microwave effects. In addition to the spectral shape separation introduced here, measurements on the other dependences are strongly recommended. \begin{figure} \begin{centering} \includegraphics{FigShapeSeparation} \par\end{centering} \caption{(Color online) (a) Common procedure for extracting the ISHE contribution. (b) The origins which can show the sign change under the magnetization reversal. Lorentzian part with the sign change is not only due to the ISHE, but also due to the microwave extrinsic effects. \label{fig:Analysis-of-obtained}} \end{figure} As it is derived in Eqs. (\ref{eq:EISHE}) and (\ref{eq:Erect}), the dc electromotive force on the FMR $\mathbf{E}_{{\rm tot}}=\mathbf{E}_{{\rm ISHE}}+\mathbf{E}_{{\rm rect}}+\mathbf{E}_{{\rm SE}{\rm \left(MSSW\right)}}+\mathbf{E}_{{\rm TTE}}$ has the two distinct parts proportional to ${\rm Lor}\left(\omega,\omega_{r}\right)$ and ${\rm Asym}\left(\omega,\omega_{r}\right)$. By fitting an observed signal by \begin{equation} \mathbf{E}_{{\rm tot}}=\mathbf{E}_{{\rm sym}}{\rm Lor}\left(\omega,\omega_{{\rm r}}\right)+\mathbf{E}_{{\rm asym}}{\rm Asym}\left(\omega,\omega_{{\rm r}}\right), \end{equation} the separation can be done, where $E_{{\rm sym}\left({\rm asym}\right)}$ is the magnitude of the ${\rm Lor}\left(\omega,\omega_{{\rm r}}\right)$\linebreak{} (${\rm Asym}\left(\omega,\omega_{{\rm r}}\right)$) part (See Fig. \ref{fig:Analysis-of-obtained}). The earlier naive discussions attribute the whole $E_{{\rm sym}}$ due to the ISHE, but this assumption does not hold as is discussed below. Generally, $E_{{\rm sym}}$ includes not only the ISHE component but also the rectification contribution.\citep{azevedo2011spin,chen2013direct} For example, in a ${\rm TE}_{011}$ cavity with the rf field $h_{{\rm rf}}$ in the $y$ direction and an rf current along the $x$ direction with $j_{{\rm rf}}\propto ih_{{\rm rf}}$, the $y$-component of Eq. \eqref{eq:Erectex} leads to \begin{align} E_{{\rm sym},{\rm ISHE}}^{y}\propto & \frac{\omega_{\phi}}{\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}\sin\theta_{{\rm M}},\nonumber \\ E_{{\rm sym},{\rm rect}}^{y}\propto & \frac{\left(\rho_{{\rm AHE}}\omega_{\theta\phi}-\Delta\rho_{{\rm AMR}}\omega_{\phi}\right)}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}\sin\theta_{{\rm M}}\label{eq:ESMRy} \end{align} at $\phi_{{\rm M}}=0$.\citep{chen2013direct} The configuration is depicted in Fig. \ref{fig:Typical-configuration-of}(b). $E_{{\rm sym}}$ includes signals due to the rectification effects and, importantly, possesses the same symmetry as the ISHE signal, i.e. $\sin\theta_{{\rm M}}$. Therefore, the part of the Lorentzian signal with the sign change following to the magnetization reversal cannot be attributed to only the ISHE without further examinations. In Ref. \citealp{chen2013direct}, a separation of the contributions based on the difference in the pre-factors, i.e. $\omega_{\theta\phi}$, $\omega_{\phi}$, $\omega$, darling the $\theta_{{\rm M}}$ scan is suggested and will be introduced in Sect. \ref{subsec:Angular-dependence-of}. Note that Eq. (\ref{eq:ESMRy}) holds only when the rf current is constant during the scan. However, the stray rf current often shows angular dependence.\citep{lustikova2015vector} The heating effect due to transverse thermoelectric effects can also induce the similar signal to the ISHE, which is given by \begin{equation} E_{{\rm sym,T{\rm TE}}}^{y}\propto\frac{\omega_{\phi}}{\omega_{\theta}+\omega_{\phi}}{\rm sin\theta_{{\rm M}}.} \end{equation} The FMR heating contribution discussed in Sect. \ref{subsec:Induced-voltage-by} can be extracted by the frequency dependence measurement. The MSSW heating contribution discussed in Sect. \ref{subsec:Induced-voltage-by} is unable to be removed until the F layer thickness is reduced. One possible solution for handling this difficulty is finding this contribution from the the calculation based on the Seebeck coefficient following a temperature profile measurement as is done in Ref. \citealp{qiu2015spin}. An asymmetric component is a sign of contribution from the rectification effects although the reverse is not true because rectification signals can have only the Lorentzian component. If one knows the direction of the rf current, then the asymmetric component might be a good measure to determine its magnitude and thus the rectification contribution. The direction may be estimated by measuring the voltage along other directions as shown in Ref. \citealp{lustikova2015vector,tsukahara2014self}. \subsection{Thickness dependence\label{subsec:Thickness-dependence-of}} \begin{figure} \begin{centering} \includegraphics{FigThickness} \par\end{centering} \caption{(Color online) (a,b) Thickness $d_{{\rm N}}$ (a) and $d_{{\rm F}}$ (b) dependence of the ISHE and rectification signals. We assume $\lambda=10$ nm and normalize the curves by the maximum value in the plot regime. \label{fig:Thickness-dependence-of}} \end{figure} The voltage signals from the ISHE and rectification effects have different dependence on the thickness of the F and N layers.\citep{nakayama2012geometry} For a bilayer where the galvanomagnetic effects in the F layer is dominant, the symmetric Lorentzian signal after taking the difference between $\mathbf{m}_{0}$ and $-\mathbf{m}_{0}$ is expressed in the following form, \begin{equation} E_{{\rm sym},{\rm ISHE}}^{y}=\frac{E_{{\rm ISHE}}^{0}}{\sigma_{{\rm N}}d_{{\rm N}}+\sigma_{{\rm F}}d_{{\rm F}}}\tanh\frac{d_{{\rm N}}}{2\lambda},\label{eq:thEyishe} \end{equation} \begin{equation} E_{{\rm sym},{\rm rect}}^{y}=\frac{d_{{\rm F}}E_{{\rm rect}}^{{\rm 0}}}{\sigma_{{\rm N}}d_{{\rm N}}+\sigma_{{\rm F}}d_{{\rm F}}}, \end{equation} where $E_{{\rm ISHE}}^{0}$ and $E_{{\rm rect}}^{0}$ are respectively determined by Eqs. (\ref{eq:EISHE}) and (\ref{eq:Erectex}). Regarding $E_{{\rm ISHE}}^{0}$ (or $\left\langle j_{{\rm s}}^{{\rm F/N}}\right\rangle _{t}$) and $E_{{\rm rect}}^{0}$ as constants, the $d_{{\rm N}}$ dependence of Eq. (\ref{eq:thEyishe}) reads a tangent hyperbolic function divided by the total resistance, which shows a positive peak around $d_{{\rm N}}\approx2\lambda$ {[}See Fig. \ref{fig:Thickness-dependence-of}(a){]}. The $d_{{\rm N}}$ dependence of the rectification contribution shows monotonic decrease as $d_{{\rm N}}$ increases. The distinction between these two becomes difficult for $d_{{\rm N}}>\lambda_{{\rm N}}$, because both show just decrease as $d_{{\rm N}}$ increases {[}See Fig. \ref{fig:Thickness-dependence-of}(b){]}. On the other hand, the $d_{{\rm F}}$ dependence shows clear difference; as $d_{{\rm F}}$ increases, the ISHE and rectification contributions respectively show monotonic decrease and increase before saturation. This is because the SP (the rectification effects) induced voltage is electrically-shorted by the additional F (N) layer. This contrary dependence of the SP and rectification contributions is very useful to separate them. For a practice analysis, thickness dependence of parameters should be considered, such as $g_{{\rm eff,r}}^{\uparrow\downarrow}$, $\tilde{\rho}^{{\rm F}}$, and $j_{{\rm rf}}\left(z\right)$. The FMR and MSSW heating effects can complicate the thickness dependence of the signal because the induced temperature gradient depends on sample structure and environment. Therefore, it is highly recommended to carefully consider an appropriate design of the samples as described in Sect. \ref{subsec:Suitable-materials-for}. \subsection{Angular dependence\label{subsec:Angular-dependence-of}} The magnetization angular dependence of $\mathbf{E}_{{\rm ISHE}}$ is derived in Ref. \citealp{ando2008angular} and that of $\mathbf{E}_{{\rm rect}}$ due to the AMR (PHE), AHE and SMR can be found in Refs. \citealp{chen2013direct,iguchi2013effect,Egan:1963gr,mecking2007microwave,yamaguchi2008broadband}. Here, we will focus on the ISHE and rectification effects, because the heating effects can not be isolated by the angular dependence. The angular dependence can be studied by two types of rotation: out-of-plane angular dependence (OP) and in-plane angular dependence (IP), which are shown in Figs. \ref{fig:OP-angular-dependence} and \ref{fig:IP-angular-dependence}. In addition, there are two different means to excite the ferromagnet: applying an rf excitation field in OP or IP. Here, these four configurations are considered in a film system with uniaxial anisotropy perpendicular to the film plane, which is relevant to most of experiments. Such systems are described by the magnetostatic energy: \begin{equation} F_{m}=-\mathbf{m}\cdot\mathbf{H}_{0}+\frac{I_{{\rm s}}^{2}}{2\mu_{0}}m_{z}^{2}-K_{{\rm u}}m_{z}^{2},\label{eq:Um} \end{equation} where $\mathbf{H}_{0}$ is an applied external field, given by\linebreak{} $\mathbf{H}_{0}=H_{0}\left(\sin\theta_{{\rm H}}\cos\phi_{{\rm H}}\mathbf{e}_{x}+\sin\theta_{{\rm H}}\sin\phi_{{\rm H}}\mathbf{e}_{y}+\cos\theta_{{\rm H}}\mathbf{e}_{z}\right)$, and $K_{{\rm u}}$ denotes the perpendicular anisotropy constant. Equation (\ref{eq:Um}) reduces Eq. (\ref{eq:omegath}) to \begin{align} \omega_{\theta} & =\gamma\left[\mu_{0}H_{0}\cos\left(\theta_{{\rm H}}-\theta_{{\rm M}}\right)-I_{{\rm eff}}\cos^{2}\theta_{{\rm M}}\right],\\ \omega_{\phi} & =\gamma\left[\mu_{0}H_{0}\cos\left(\theta_{{\rm H}}-\theta_{{\rm M}}\right)-I_{{\rm eff}}\cos\left(2\theta_{{\rm M}}\right)\right], \end{align} and $\omega_{\phi\theta}=0$, where $I_{{\rm eff}}$ denotes the effective magnetization $I_{{\rm eff}}=I_{{\rm s}}-2\mu_{0}K_{{\rm u}}/I_{{\rm s}}$, and $\theta_{{\rm H}}$ $\left(\phi_{{\rm H}}\right)$ denotes the polar (azimuthal) angle in the spherical coordinate as shown in Fig. \ref{fig:OP-angular-dependence}(a). The resonance field $H_{{\rm r}}$ is determined by $H_{0}$ which simultaneously satisfies $\omega_{{\rm r}}=\sqrt{\omega_{\theta}\omega_{\phi}}$ and $\mathbf{m}_{0}\times\mathbf{H}_{{\rm eff}}=0$, which is reduced to \begin{equation} 2\mu_{0}H_{0}\sin\left(\theta_{{\rm H}}-\theta_{{\rm M}}\right)+I_{{\rm eff}}\sin\left(2\theta_{{\rm M}}\right)=0 \end{equation} and $\phi_{{\rm H}}=\phi_{{\rm M}}$. Note that in the following calculation we still use $\omega_{{\rm \theta}\left(\phi\right)}$ for simple notation but impose $\omega_{{\rm \theta\phi}}=0$. \begin{figure}[t] \begin{centering} \includegraphics{FigOPangdep} \par\end{centering} \caption{(Color online) (a) Schematic illustration of out-of-plane (OP) angular dependence measurement, where $\mathbf{m}_{0}$ and $\mathbf{H}_{0}$ points different direction due to the demagnetizing and uniaxial anisotropy fields. (b) $H_{{\rm r}}$ and $\theta_{{\rm M}}$ as a function of $\theta_{{\rm H}}$. (c,d) angular dependence of the ISHE and rectification signals for the in-plane excitation (c) and OP excitation (d). The ISHE and rectification signals are normalized to be 1 and 0.5, respectively.\label{fig:OP-angular-dependence}} \end{figure} In the setup for the OP angular dependence measurements, the magnetization is rotated in the $xz$ plane ($\phi_{{\rm M}}=0$) and electrodes for detecting the ISHE voltage are placed in the $y$ axis {[}See Fig. \ref{fig:OP-angular-dependence}(a){]}. An IP excitation field, $h_{{\rm rf}}$ along the $y$ axis, induces \begin{equation} E_{{\rm sym},{\rm ISHE}}^{y}=A_{{\rm ISHE}}\frac{\omega_{\phi}}{\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}\sin\theta_{{\rm M}}, \end{equation} \begin{equation} E_{{\rm sym},{\rm rect}}^{y}=A_{{\rm rect}}\frac{\left(\omega\rho_{{\rm AHE}}{\rm Re}\left[j_{{\rm rf}}^{x*}\right]+\omega_{\phi}\Delta\rho_{{\rm AMR}}{\rm Im}\left[j_{{\rm rf}}^{x*}\right]\right)}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}\sin\theta_{{\rm M}}, \end{equation} where the coefficients $A$ are given by \begin{equation} A_{{\rm ISHE}}=w\tilde{R}_{{\rm tot}}\theta_{{\rm SHE}}\frac{2e}{\hbar}\lambda\tanh\frac{d_{{\rm N}}}{2\lambda}\frac{\hbar g_{r,{\rm eff}}^{\uparrow\downarrow}}{4\pi}\frac{\gamma^{2}\mu_{0}^{2}h_{{\rm rf}}^{2}}{\alpha^{2}}, \end{equation} \begin{equation} A_{{\rm rect}}=\tilde{R}_{{\rm tot}}\frac{\sigma_{{\rm F}}d_{{\rm F}}}{2}\frac{\gamma\mu_{0}h_{{\rm rf}}}{\alpha}, \end{equation} both of which are proportional to $\sin\theta_{{\rm M}}$. In Ref. \citealp{chen2013direct}, the separation is done based on the difference in the dependence on $\omega_{\theta}\left(\theta_{{\rm M}}\right)$ and $\omega_{\phi}\left(\theta_{{\rm M}}\right)$ between the ISHE and rectification voltages, which is because they have different responses to the effective field sweeping over $\theta_{{\rm H}}$. Figure \ref{fig:OP-angular-dependence}(b) shows $H_{{\rm r}}$ and $\theta_{{\rm M}}$ as a function of $\theta_{{\rm H}}$ calculated with $\gamma=1.79\times10^{11}$ ${\rm T^{-1}s^{-1}}$ and $I_{{\rm eff}}=1$ T. Figure \ref{fig:OP-angular-dependence}(c) shows the $\theta_{{\rm M}}$ dependence of $E_{{\rm ISHE}}$ and $E_{{\rm rect}}$, which possess similar form, but slight difference seen in the solid and dashed curves. This similarity can be a large source of error in data fitting. An OP excitation field, $h_{z}$, induces \begin{equation} E_{{\rm sym},{\rm ISHE}}^{y}=A_{{\rm ISHE}}\frac{\omega_{\theta}}{\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}\sin^{3}\theta_{{\rm M}}, \end{equation} \begin{equation} E_{{\rm sym},{\rm rect}}^{y}=A_{{\rm rect}}\frac{\left(\omega\Delta\rho_{{\rm AMR}}{\rm Re}\left[j_{{\rm rf}}^{x*}\right]-\omega_{\theta}\rho_{{\rm AHE}}{\rm Im}\left[j_{{\rm rf}}^{x*}\right]\right)}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}\sin^{2}\theta_{{\rm M}}, \end{equation} {[}See Fig. \ref{fig:OP-angular-dependence}(d){]}. At a glance, the difference in the two signals are clear, namely $E_{{\rm rect}}^{y}\propto\sin^{2}\theta_{{\rm M}}$ and $E_{{\rm ISHE}}\propto\sin^{3}\theta_{{\rm M}}$. However, when $j_{{\rm rf}}\propto\sin\theta_{{\rm M}}$ holds, the contributions cannot be separated by the harmonic functions. \begin{figure} \begin{centering} \includegraphics{FigIPangdep} \par\end{centering} \caption{(Color online) (a, b) In-plane (IP) angular dependence of the ISHE and rectification signals for the IP excitation (a) and out-of-plane excitation (b). The ISHE and rectification signals are normalized to be 1 and 0.5, respectively. PHE denotes the AMR signal caused by $j_{{\rm rf}}^{x}$.\label{fig:IP-angular-dependence}} \end{figure} Similarly to the OP angular dependence in the OP excitation, the IP angular dependence excited by an IP rf field has the same difficulty. In the setup, the magnetization is rotated in the $xy$ plane ($\theta_{{\rm M}}=90^{\circ}$) and electrodes for detecting the ISHE voltage are placed in the $y$ axis. The electromotive forces in the $y$ direction are \begin{equation} E_{{\rm sym},{\rm ISHE}}^{y}=A_{{\rm ISHE}}\frac{\omega_{\phi}}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}\cos^{3}\phi_{{\rm M}}, \end{equation} \begin{equation} E_{{\rm sym},{\rm rect}}^{y}=A_{{\rm rect}}\cos\phi_{{\rm M}}\frac{\omega_{\phi}\Delta\rho_{{\rm AMR}}\left(\cos\left(2\phi_{{\rm M}}\right){\rm Im}\left[j_{{\rm rf}}^{x*}\right]+\sin\left(2\phi_{{\rm M}}\right){\rm Im}\left[j_{{\rm rf}}^{y*}\right]\right)+\omega\rho_{{\rm AHE}}{\rm Re}\left[j_{{\rm rf}}^{x*}\right]}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}\label{eq:EsymIPIP} \end{equation} {[}See Fig. \ref{fig:IP-angular-dependence}(a){]}. In this configuration, the ISHE and AMR contributions mix even for the simplest condition that the rf current is angular-independent. The IP angular dependence excited by an OP rf field has an advantage to the previous three angular dependences because the AMR and AHE show different symmetric angular dependences in the IP configuration. An OP excitation field, $h_{z}$, induces \begin{equation} E_{{\rm sym},{\rm ISHE}}^{y}=A_{{\rm ISHE}}\frac{\omega_{\theta}}{\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}\cos\phi_{{\rm M}},\label{eq:IP-1} \end{equation} \begin{equation} E_{{\rm sym},{\rm rect}}^{y}=A_{{\rm rect}}\frac{\omega\Delta\rho_{{\rm AMR}}\left(\cos\left(2\phi_{{\rm M}}\right){\rm Re}\left[j_{{\rm rf}}^{x*}\right]+\sin\left(2\phi_{{\rm M}}\right){\rm Re}\left[j_{{\rm rf}}^{y*}\right]\right)-\omega_{\theta}\rho_{{\rm AHE}}{\rm Im}\left[j_{{\rm rf}}^{x*}\right]}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}.\label{eq:IP-2} \end{equation} For the simplest case where $j_{x\left(y\right)}^{{\rm rf}}$ is constant during the rotation, importantly, the ISHE, AMR and AHE show the different angular dependences, $\cos\phi_{{\rm M}}$, $\cos\left(2\phi_{{\rm M}}\right)$, and constant {[}Fig. \ref{fig:IP-angular-dependence}(b){]}. Thus, fitting the result using the harmonic functions gives the ISHE contribution directly. Figure \ref{fig:IP-angular-dependence} shows the calculation result based on Eqs. (\ref{eq:IP-1}) and (\ref{eq:IP-2}). \begin{figure} \begin{centering} \includegraphics{FigFourier} \par\end{centering} \caption{(Color online) (a,b) Analysis scheme based on the Fourier cosine coefficients: case for the constant rf current (a) and case for angular dependent rf current (b) during the magnetization angle scan in the film plane. The inset shows the magnitude of the Fourier cosine coefficient of the complex $j_{{\rm rf}}^{x}$. \label{fig:Analysis-scheme-based}} \end{figure} When $j_{x\left(y\right)}^{{\rm rf}}$ has an angular dependence, an analysis method based on the Fourier series coefficient is effective in the measurements on the IP angular dependence with OP excitation. The $n$-th Fourier cosine coefficient of the voltage is given by \begin{align} F_{{\rm sym}}^{{\rm cos}}\left(n\right) & =\frac{\omega_{\theta}A{}_{{\rm ISHE}}}{\left(\omega_{\theta}+\omega_{\phi}\right)^{2}}\delta_{n-1}+\frac{A_{{\rm rect}}}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}\left(\omega\Delta\rho_{{\rm AMR}}{\rm Re}\left[J_{{\rm rf}}^{x}\left(n-2\right)\right]-\omega_{\theta}\rho_{{\rm AHE}}{\rm Im}\left[J_{{\rm rf}}^{x}\left(n\right)\right]\right),\\ F_{{\rm asym}}^{{\rm cos}}\left(n\right) & =\frac{A_{{\rm rect}}}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}\left(-\omega\Delta\rho_{{\rm AMR}}{\rm Im}\left[J_{{\rm rf}}^{x}\left(n-2\right)\right]-\omega_{\theta}\rho_{{\rm AHE}}{\rm Re}\left[J_{{\rm rf}}^{x}\left(n\right)\right]\right),\label{eq:Fsymasym} \end{align} where $\delta_{n}$ denotes the Kronecker delta function and $J_{{\rm rf}}^{x}\left(n\right)$ is the $n$-th Fourier cosine coefficient of $j_{{\rm rf}}^{x}$. Figure \ref{fig:Analysis-scheme-based}(a) shows an expected intensity of the coefficients for $J_{{\rm rf}}^{x}\left(n\right)\propto\delta_{n}$. The contributions are clearly separated. Figure \ref{fig:Analysis-scheme-based}(b) shows a calculation of the coefficients for $J_{{\rm rf}}^{x}\left(n\right)$ with $J_{{\rm rf}}^{x}\left(1\right)\neq0$. $J_{{\rm rf}}^{x}\left(1\right)$ can induces an ISHE-like signal via the AHE. However, this contribution can be removed by comparing symmetric and antisymmetric components because there is a relation, \[ F_{{\rm sym}}^{{\rm cos}}\left(1\right)=-\frac{\omega_{\theta}}{\omega}\frac{\rho_{{\rm AHE}}}{\Delta\rho_{{\rm AMR}}}F_{{\rm asym}}^{{\rm cos}}\left(3\right), \] in the absence of the ISHE. Therefore, the factor above gives the upper limit of the Lorentzian part due to the AHE. When $\omega_{{\rm \theta}}\left(\omega_{\phi}\right)$ possesses $\phi_{{\rm M}}$ dependence because of a magnetic anisotropy field in-plane, Eq. (\ref{eq:Fsymasym}) should be recalculated by considering the Fourier coefficients of $\omega_{\theta}$ and $\omega_{{\rm \phi}}$. Note that when $\Delta\rho=0$ but $\rho_{{\rm AHE}}\neq0$, this method can not be applied, and it is better to change a material for the F layer or to try a measurement on the ferromagnetic layer thickness dependence described in Sect. \ref{subsec:Thickness-dependence-of}. \subsection{Frequency dependence } Here, we focus on the difference in the frequency dependence of the signals from the ISHE and the microwave effects. The frequency dependence has attracted much attention for its nonlinear physics coming from magnon-magnon interactions.\citep{kurebayashi2011controlled,castel2012frequency,harii2011frequency,sakimura2014nonlinear,chumak2015magnon} Our interest is the linear excitation regime in which the derived equations for the FMR can be used.\citep{iguchi2012spin} The heating effects show a clear difference from the ISHE and rectification effects, so that this method works effectively for removing the heating contribution. \begin{figure} \centering{}\includegraphics{FigFreq}\caption{(Color online) Frequency dependence of signals induced by the ISHE, rectification, and heating effects. The curves are normalized at $\omega/2\pi=1$ GHz. Other parameters are same as those used for the OP angular dependence calculation.\label{fig:frequency-dependence}} \end{figure} In the calculation we consider a system with field along the $x$ and rf field along the $y$ axis, i.e. the IP excitation at $\theta_{{\rm M}}=90^{\circ}$ and $\phi_{{\rm M}}=0^{\circ}$. This configuration is often used in measurements with a microstrip line or a coplanar waveguide. The frequency dependence of the ISHE and rectification signals are respectively given by \begin{equation} E_{{\rm sym},{\rm ISHE}}^{y}=A_{{\rm ISHE}}\frac{\omega_{\phi}}{\left(\omega_{\theta}+\omega_{\phi}\right)^{2}} \end{equation} and \begin{equation} E_{{\rm sym},{\rm rect}}^{y}=A{}_{{\rm rect}}\frac{\left(\omega\rho_{{\rm AHE}}{\rm Re}\left[j_{{\rm rf}}^{x*}\right]+\omega_{\phi}\Delta\rho_{{\rm AMR}}{\rm Im}\left[j_{{\rm rf}}^{x*}\right]\right)}{\omega\left(\omega_{\theta}+\omega_{\phi}\right)}. \end{equation} The electromotive force due to the heating, proportional to microwave absorption $\Delta P$, is given by \begin{equation} E_{{\rm sym},{\rm TE}}^{y}=A{}_{{\rm TE}}\frac{\omega_{\phi}}{\left(\omega_{\theta}+\omega_{\phi}\right)} \end{equation} with $A_{{\rm TE}}=\frac{\gamma\mu_{0}h_{{\rm rf}}^{2}}{2\alpha}\left(A_{{\rm TTE}}+A_{{\rm SE}\left({\rm MSSW}\right)}\right)$. Figure \ref{fig:frequency-dependence} shows the calculated frequency dependence of the ISHE, rectification effects, and heating effects. Since the ISHE and AHE are similar, the separation between the ISHE and rectification signals based on the frequency will not be so accurate. The difference between the SP and the rectification effects comes from the ellipticity of magnetization precession due to demagnetizing and anisotropic fields. At frequencies $\omega>\gamma I_{{\rm eff}}$,the large external field necessary for the FMR makes the precession trajectory circular, so that both of them become proportional to $1/\omega$. The heating is proportional to $\Delta P$ {[}Eq. (\ref{eq:delP}){]} and is proportional to $\omega_{\theta\left(\phi\right)}/\omega$, which reaches constant at high frequencies. This feature is distinct from the others. The studies on the frequency dependence\citep{kurebayashi2011controlled,castel2012frequency,harii2011frequency,sakimura2014nonlinear,iguchi2012spin} indicates that the thermoelectric contribution in the ferrimagnetic insulator/Pt bilayer systems is not dominant for the microwave spin pumping experiments. \section{Summary\label{sec:Summary}} \begin{figure} \centering{}\includegraphics[width=12cm]{FigSummr}\caption{(Color online) Summary of methods for measuring the SP-induced ISHE. It is written which effect can be separated (\Checkmark ) or cannot be separated ($\times$) by using combination of the corresponding material for the spin injector (F) layer and the measurement. $-$ denotes the non applicable separation method. $\triangle$ denotes the moderate separation method when galvanomagnetic or transverse thermoelectric effects are not negligibly small. Here, we assume that the MSSW heating is dominant among the heating-induced signals. The use of a thin ferrimagnetic insulator is the best way to explore spin physics by using the spin pumping, because the magnitude of the galvanomagnetic effects and MSSW heating is expected to be small compared to that in metallic and thick ferromagnets. \label{fig:Summary-of-how}} \end{figure} In this article, we reviewed voltage generation by the SP-induced ISHE, the rectification effects due to galvanomagnetic effects, and the heating effects due to thermoelectric effects. The electric detection of a spin current induced by the SP using the ISHE is a strong method to study spin physics in a material of interest. The key for the study is clear separation between the ISHE and the other extrinsic contributions. In some configurations, they look similar to each other, but by employing the separation methods introduced here, one can perform a reliable measurement with high accuracy. Figure \ref{fig:Summary-of-how} summarizes the recommended method by which the accuracy can be easily obtained. For a better experiment, a material of interest should be on top of a thin ferrimagnetic insulator film, which reduces both the rectification and heating effects. For systems with metallic ferromagnet, the IP magnetization angular dependence is the best configuration to clarify the differences among the voltage signals of the ISHE and the rectification effects, because the ISHE is sensitive to spin polarization while the AMR is sensitive to the magnitude than the polarization. The separation schemes discussed in this article provides a better way to extract the SP-originated signals. We thus anticipate that the experimental schemes help further material investigations and contribute to the development of novel spintronic devices. \section*{Acknowledgment} The authors thank K. Sato, K. Uchida, and D. Hirobe for valuable discussions and suggestions. This work was supported by Grant-in-Aid for Scientific Research on Innovative Area, ``Nano Spin Conversion Science\textquotedblright{} (No. 26103002, 26103005) from MEXT, Japan, and E-IMR, Tohoku University. \bibliographystyle{jpsj}
842d24bc33ac769220affce1ce229e3d0c36375b
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{introduction} \setcounter{equation}{0} \numberwithin{equation}{section} In the present paper, we consider the equation \begin{equation}\label{1.1} -y''(x)+q(x)y(x )=f(x),\quad x\in \mathbb R \end{equation} where $f\in L_p^{\loc}(\mathbb R)$, $p\in[1,\infty)$ and \begin{equation}\label{1.2} 0\le q \in L_1^{\loc}(\mathbb R). \end{equation} Our general goal is to determine a space frame within which equation \eqref{1.1} always has a unique stable solution. To state the problem in a more precise way, let us fix two positive continuous functions $\mu(x)$ and $\theta(x),$ $x\in\mathbb R,$ a number $p\in[1,\infty)$, and introduce the spaces $L_p(\mathbb R,\mu)$ and $L_p(\mathbb R,\theta):$ \begin{align} &L_p(\mathbb R,\mu)=\left\{f\in L_p^{\loc}(\mathbb R):\|f\|_{L_p(\mathbb R ,\mu)}^p=\int_{-\infty}^\infty|\mu(x)f(x)|^pdx<\infty\right\}\label{1.3}\\ &L_p(\mathbb R,\theta)=\left\{f\in L_p^{\loc}(\mathbb R):\|f\|_{L_p(\mathbb R ,\theta)}^p=\int_{-\infty}^\infty|\theta(x)f(x)|^pdx<\infty\right\}.\label{1.4} \end{align} For brevity, below we write $L_{p,\mu}$ and $L_{p,\theta},$ \ $\|\cdot\|_{p,\mu}$ and $\|\cdot\|_{p,\theta}$, instead of $L_p(\mathbb R,\mu),$ $L_p(\mathbb R,\theta)$ and $\|\cdot\|_{L_p(\mathbb R,\mu)}$, $\|\cdot\|_{L_p(\mathbb R,\theta)},$ respectively (for $\mu=1$ we use the standard notation $L_p$ $(L_p:=L_p(\mathbb R))$ and $\|\cdot\|_p$ $(\|\cdot\|_p:=\|\cdot\|_{L_p}).$ In addition, below by a solution of \eqref{1.1} we understand any function $y,$ absolutely continuous together with its derivative and satisfying equality \eqref{1.1} almost everywhere on $\mathbb R$. Let us introduce the following main definition (see \cite[Ch.5, \S50-51]{12}: \begin{defn}\label{defn1.1} We say that the spaces $L_{p,\mu}$ and $L_{p,\theta}$ make a pair $\{L_{p,\mu},L_{p,\theta}\}$ admissible for equation \eqref{1.1} if the following requirements hold: I) for every function $f\in L_{p,\theta}$ there exists a unique solution $y\in L_{p,\mu}$ of \eqref{1.1}; II) there is a constant $c(p)\in (0,\infty)$ such that regardless of the choice of a function $f\in L_{p,\theta}$ the solution $y\in L_{p,\mu}$ of \eqref{1.1} satisfies the inequality \begin{equation}\label{1.5} \|y\|_{p,\mu}\le c(p)\|f\|_{p,\theta}. \end{equation} \end{defn} Let us in addition we make the following conventions: For brevity we say ``problem I)--II)" or ``question on I)--II)" instead of ``problem (or question) on conditions for the functions $\mu$ and $\theta$ under which requirements I)--II) of Definition \ref{defn1.1} hold." We say ``the pair $\{L_{p,\mu};L_{p,\theta}\}$ admissible for \eqref{1.1}" instead of ``the pair of spaces $\{L_{p,\mu};L_{p,\theta}\}$ admissible for equation \eqref{1.1}", and we often omit the word ``equation" before \eqref{1.1}. By $c,\, c(\cdot)$ we denote absolute positive constants which are not essential for exposition and may differ even within a single chain of calculations. Our general requirement \eqref{1.2} is assumed to be satisfied throughout the paper, is not referred to, and does not appear in the statements. Let us return to Definition \ref{defn1.1}. The question on the admissibility of the pair $\{L_p,L_p\}$ for \eqref{1.1} was studied in \cite{3,6} (in \cite{3,6} for $\mu\equiv\theta\equiv1$ in the case where I)--II) were valid, we said that equation \eqref{1.1} is correctly solvable in $L_p.$ We maintain this terminology in the present paper.) Let us quote the main result of \cite{3,6} (in terms of Definition \ref{defn1.1}). \begin{thm} \label{thm1.2} \cite{3} The pair $\{L_p,L_p\}$ is admissible for \eqref{1.1} if and only if there is $a\in(0,\infty)$ such that $q_0(a)>0.$ Here \begin{equation}\label{1.6} q_0(a)=\inf_{x\in\mathbb R}\int_{x-a}^{x+a}q(t)dt. \end{equation} \end{thm} Below we continue the investigation started in \cite{3,6}. Our goal is as follows: given equation \eqref{1.1}, to determine requirements to the weights $\mu$ and $\theta$ under which the pair $\{L_{p,\mu};L_{p,\theta}\},$ $p\in[1,\infty),$ is admissible for \eqref{1.1}. Such an approach to the inversion of \eqref{1.1} allows to study this equation also in the case where \thmref{thm1.2} is not applicable, for example, in the following three cases: \begin{enumerate} \item[1)]\ $q_0(a)>0$ for some $a\in(0,\infty),$ $f\notin L_p,$ $p\in[1,\infty);$ \item[2)]\ $q_0(a)=0$ for all $a\in(0,\infty),$ $f\in L_p,$ $p\in[1,\infty);$ \item[2)]\ $q_0(a)=0$ for all $a\in(0,\infty),$ $f\notin L_p,$ $p\in[1,\infty).$ \end{enumerate} Our main result (see \thmref{thm4.3} in \S4 below) reduces the stated problem to the question on the boundedness of a certain integral operator $S: L_p\to L_p$ (see \eqref{4.3} in \S4). {}From this criterion, under additional requirements to the functions $\mu,$ $\theta$ and $q,$ one can deduce some concrete particular conditions which control the solution of our problem. See \S4 for such restrictions. We now describe the structure of the paper. Section~2 contains preliminaries; in Section~3 we give various technical assertions; all our results and relevant comments are presented in Section~4; all the proofs are collected in Section~5; and Section~6 contains an example of the presented statements. \section{Preliminaries} Recall that our standing assumption \eqref{1.2} is not included in the statements. \begin{lem}\label{lem2.1} \cite{4} Suppose that the following condition holds: \begin{equation}\label{2.1} \int_{-\infty}^x q(t)dt>0,\quad \int_x^\infty q(t)dt>0,\quad \forall x\in\mathbb R. \end{equation} Then for any given $x\in\mathbb R,$ each of the equations in $d\ge0$ \begin{equation}\label{2.2} \int_0^{\sqrt 2 d}\int_{x-t}^{x+t}q(\xi)d\xi dt=2,\qquad d\int_{x-d}^{x+d}q(\xi)d\xi=2 \end{equation} has a unique finite positive solution. Denote these solutions by $d(x)$ and $\hat d(x)$, respectively. We have the inequalities \begin{equation}\label{2.3} \frac{d(x)}{\sqrt 2}\le \hat d(x)\le \sqrt 2 d(x),\quad x\in\mathbb R. \end{equation} \end{lem} Note that the functions $d(x)$ and $\hat d(x)$ were introduced by the authors (see \cite{1,4}) and M. Otelbaev (see \cite{10}), respectively. Analysing our assertions and requirements (see \S4 below), it is useful to take into account that the function $q^*(x)\doe d^{-2}(x)$ $(d^{-2}:=1/d^2)$ can be interpreted as a composed (in the sense of function theory) average of the function $q(\xi),$ $\xi\in\mathbb R$, at the point $\xi=x$ with step $d(x).$ Indeed, denote $$S_x(q)(t)=\frac{1}{2t}\int_{x-t}^{x+t}q(\xi)d\xi,\quad t>0, \quad x\in\mathbb R,$$ $$M(f)(\eta)=\frac{1}{\eta^2}\int_0^{\sqrt 2\eta}tf(t)dt,\quad \eta>0.$$ Clearly, $S_x(q)(t)$ is the Steklov average with step $t>0$ of the function $q(\xi),$ $\xi\in\mathbb R$, at the point $\xi=x,$ and $M(f)(\eta)$ is the average of the function $f(t),$ $t>0$ with step $\eta>0$ at the point $t=0.$ Now, using \begin{align*}q^*(x)&=\frac{1}{d^2(x)}=\frac{1}{2d^2(x)}\int_0^{\sqrt 2d(x)}\int_{x-t}^{x+t}q(\xi)d\xi dt\\ &=\frac{1}{d^2(x)}\int_0^{\sqrt 2 d(x)}t\left[\frac{1}{2t}\int_{x-t}^{x+t}q(\xi)d\xi\right]dt=M(S_x(q))(d(x)). \end{align*} Similarly, the function $\hat q^*(x)\doe \hat d(x)^{-2}$ $x\in\mathbb R,$ can be interpreted as the Steklov average of the function $q(\xi),$ $\xi\in\mathbb R,$ at the point $\xi=x$ with step $\hat d(x).$ Indeed (see \eqref{2.1}), we have $$\hat q^*(x)=\frac{1}{\hat d^2(x)}=\frac{1}{2\hat d(x)}\int_{x-\hat d(x)}^{x+\hat d(x)} q(\xi)d\xi=S_x(q)(\hat d(x)).$$ \begin{thm}\label{thm2.2} \cite{2} Suppose that \eqref{2.1} holds. Then the equation \begin{equation}\label{2.4} z''(x)=q(x)z(x),\quad x\in\mathbb R, \end{equation} has a fundamental system of solutions (FSS) $\{u(x),v(x)\},$ $x\in\mathbb R,$ such that \begin{gather} u(x)>0,\quad v(x)>0,\quad u'(x)<0,\quad v'(x)>0,\quad \forall x\in\mathbb R,\label{2.5}\\ v'(x)u(x)-u'(x)v(x)=1,\quad \forall x\in\mathbb R,\label{2.6}\\ \lim_{x\to-\infty}\frac{v(x)}{u(x)}=\lim_{x\to\infty}\frac{u(x)}{v(x)}=0,\label{2.7}\\ |\rho'(x)|<1,\quad \forall x\in\mathbb R,\quad \rho(x)\doe u(x)v(x).\label{2.8} \end{gather} \end{thm} Let us introduce the Green function of equation \eqref{1.1}: \begin{equation}\label{2.9} G(x,t)=\begin{cases} u(x)v(t),\ & x\ge t\\ u(t)v(x),\ &x\le t\end{cases} \end{equation} \begin{thm}\label{thm2.3}\cite{8} For $x,t\in\mathbb R$, we have the Davies-Harrell representations for the solution $\{u(x),v(x)\}$ and the Green function $G(x,t):$ \begin{gather}\label{2.10} u(x)=\sqrt{\rho(x)}\exp\left(-\frac{1}{2}\int_{x_0}^x\frac{d\xi}{\rho(\xi)}\right),\quad v(x)=\sqrt{\rho(x)}\exp\left(\frac{1}{2}\int_{x_0}^x\frac{d\xi}{\rho(\xi)}\right),\\ \label{2.11} G(x,t)=\sqrt{\rho(x)\rho(t)}\exp\left(-\frac{1}{2}\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right|\right). \end{gather} \end{thm} Here $x_0$ is a unique solution of the equation $u(x)=v(x),$ $x\in\mathbb R$ (see \cite{2}), the function $\rho$ is defined in \eqref{2.8}. \begin{thm}\label{thm2.4} \cite{4} Suppose that \eqref{2.1} holds. Then we have the Otelbaev ienqualities: \begin{equation}\label{2.12} \frac{d(x)}{2\sqrt 2}\le\rho(x)\le\sqrt 2 d(x),\quad x\in\mathbb R. \end{equation} \end{thm} Two-sided, sharp by order estimates of the function $\rho$ were first obtained by M. Otelbaev (see \cite{10}), and therefore all such inequalities are referred to by his name. Note that the inequalities given in \cite{10} are expressed in terms of another auxiliary function, more complicated than $d(x),$ $x\in\mathbb R$, and are proven under auxiliary requirements to the function $q.$ Let us introduce the Green operator \begin{equation}\label{2.13} (Gf)(x)=\int_{-\infty}^\infty G(x,t)f(t)dt,\quad x\in\mathbb R. \end{equation} \begin{thm}\label{thm2.5} \cite{6} Suppose that \eqref{2.1} holds, and let $p\in[1,\infty).$ Then equation \eqref{1.1} is correctly solvable in $L_p$ (or, in other words, the pair $\{L_p,L_p\}$ is admissible for \eqref{1.1}) if and only if the operator $G:L_p\to L_p$ is bounded. In the latter case, for $f\in L_p,$ the solution $y\in L_p$ of \eqref{1.1} is of the form $y=Gf.$ \end{thm} \begin{thm}\label{thm2.6} \cite{3} For $p\in[1,\infty)$, equation \eqref{1.1} is correctly solvable in $L_p$ (i.e., the pair $\{L_p,L_p\}$ is admissible for \eqref{1.1}) if and only if equalities \eqref{2.1} hold and $\hat d_0<\infty.$ Here \begin{equation}\label{2.14} \hat d_0=\sup_{x\in\mathbb R}\hat d(x). \end{equation} \end{thm} \begin{thm}\label{thm2.7} \cite{11} Let $\mu$ and $\theta$ be continuous positive functions in $\mathbb R,$ and let $H$ be an integral operator \begin{equation}\label{2.15} (Hf)(t)=\mu(t)\int_t^\infty\theta(\xi)f(\xi)d\xi,\quad t\in\mathbb R. \end{equation} For $p\in(1,\infty)$, the operator $H:L_p\to L_p$ is bounded if and only if $H_p<\infty.$ Here $H_p=\sup\limits_{x\in\mathbb R}H_p(x),$ \begin{equation}\label{2.16} H_p(x)=\left(\int_{-\infty}^x\mu(t)^pdt\right)^{1/p}\cdot\left(\int_x^\infty\theta(t)^{p'}dt\right) ^{1/p'},\quad p'=\frac{p}{p-1}. \end{equation} In addition, \begin{equation}\label{2.17} H_p\le\|H\|_{p\to p}\le (p)^{1/p}(p')^{1/p'}H_p. \end{equation} \end{thm} \begin{thm}\label{thm2.8} \cite{11} Let $\mu$ and $\theta$ be continuous positive functions in $\mathbb R,$ and let $\tilde H$ be an integral operator \begin{equation}\label{2.18} (\tilde Hf)(t)=\mu(t)\int_{-\infty}^t\theta(\xi)f(\xi)d\xi,\quad t\in\mathbb R. \end{equation} For $p\in(1,\infty)$ the operator $\tilde H:L_p\to L_p$ is bounded if and only if $\tilde H_p<\infty.$ Here $\tilde H_p=\sup_{x\in\mathbb R}\tilde H_p(x)$ \begin{equation}\label{2.19} \tilde H_p(x)=\left[\int_{-\infty}^x\theta(t)^{p'}dt\right]^{1/p'}\cdot\left[\int_x^\infty\mu(t)^p dt\right]^{1/p},\quad p'=\frac{p}{p-1}. \end{equation} In addition, \begin{equation}\label{2.20} \tilde H_p\le\|\tilde H\|_{p\to p}\le(p)^{1/p}(p')^{1/p'}\tilde H_p. \end{equation} \end{thm} \begin{thm}\label{thm2.9} \cite{10a} Let $-\infty\le a<b\le \infty,$ let $K(x,t)$ be a continuous function for $s,t\in(a,b), $ and let $K$ be an integral operator \begin{equation}\label{2.21} (Kf)(t)=\int_a^bK(s,t)f(s)ds,\quad t\in(a,b). \end{equation} Then we have the inequality \begin{equation}\label{2.22} \|K\|_{L_1(a,b)\to L_1(a,b)}=\sup_{s\in(a,b)}\int_a^b|K(s,t)|dt. \end{equation} \end{thm} \section{Auxiliary assertions} In this section, we mainly present the properties of the function $d(x)$, $x\in\mathbb R$ (see \lemref{lem2.1}). Here we assume that condition \eqref{2.1} is satisfied, and we do not include it in the statements. \begin{lem}\label{lem3.1} The function $d(x)$ is continuously differentiable for all $x\in\mathbb R,$ and the following inequality holds: \begin{equation}\label{3.1} \sqrt 2|d'(x)|\le 1,\quad x\in\mathbb R. \end{equation} \end{lem} \begin{remark}\label{rem3.2} It is interesting to compare estimate \eqref{2.8} (see also \eqref{2.12}) with estimate \eqref{3.1}. \end{remark} \begin{lem}\label{lem3.3} For $x\in\mathbb R,$ we have the inequalities \begin{equation}\label{3.2} 4^{-1}d(x)\le d(t)\le 4d(x),\quad\text{if}\quad |t-x|\le d(x). \end{equation} \end{lem} \begin{lem}\label{lem3.4} For $x\in\mathbb R,$ we have the inequalities (see \thmref{thm2.2}): \begin{equation}\label{3.3} c^{-1}\le\frac{u(t)}{u(x)};\ \frac{v(t)}{v(x)};\quad \frac{\rho(t)}{\rho(x)}\le c\quad\text{if}\quad |t-x|\le d(x). \end{equation} \end{lem} \begin{lem}\label{lem3.5} For a given $x\in\mathbb R,$ consider the function \begin{equation}\label{3.4} F(\eta)=\int_0^{\sqrt 2\eta}\int_{x-t}^{x+t}q(\xi)d\xi dt,\quad \eta\ge 0. \end{equation} The function $F(\eta)$ is differentiable and non-negative, together with its derivative, and \begin{equation}\label{3.5} F(0)=0,\qquad F(\infty)=\infty. \end{equation} In addition, the inequality $\eta\ge d(x)$ $(0\le\eta\le d(x))$ holds if and only if $F(\eta)\ge2$ $(F(\eta)\le 2).$ \end{lem} \begin{lem}\label{lem3.6} Let a function $f$ be defined on $\mathbb R$ and absolutely continuous together with its derivative. Then for all $x\in\mathbb R$ and $t\ge0,$ we have the equality \begin{equation}\label{3.6} \int_{x-t}^{x+t}f(\xi)d\xi=2f(x)t+\int_0^t\int_0^{t_1}\int_{x-t_2}^{x+t_2}f''(t_3)dt_3dt_2dt_1. \end{equation} \end{lem} \begin{thm}\label{thm3.7} Suppose that condition \eqref{2.1} holds and the function $q(x)$ can be written in the form \begin{equation}\label{3.7} q(x)=q_1(x)+q_2(x),\quad x\in\mathbb R, \end{equation} where $q_1(x),$ $x\in\mathbb R,$ is positive and absolutely continuous together with its derivative, and $q_2\in L_1^{\loc}(\mathbb R).$ Denote \begin{gather}\label{3.8} A(x)=\left[0,\frac{2}{\sqrt{q_1(x)}}\right],\quad x\in\mathbb R,\\ \label{3.9} \varkappa_1(x)=\frac{1}{q_1(x)^{3/2}}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}q_1''(\xi)d\xi\right|,\quad x\in\mathbb R,\\ \label{3.10} \varkappa_2(x)=\frac{1}{\sqrt{q_1(x)}}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}q_2(\xi)d\xi\right|,\quad x\in\mathbb R. \end{gather} If we have the condition \begin{equation}\label{3.11} \varkappa_1(x)\to0,\quad \varkappa_2(x)\to0\quad\text{as}\quad |x|\to\infty, \end{equation} then the following relations hold: \begin{gather} d(x)\sqrt{q_1(x)}=1+\varepsilon(x),\qquad |\varepsilon(x)|\le 2(\varkappa_1(x)+\varkappa_2(x)),\qquad |x|\gg 1,\label{3.12}\\ c^{-1}\le d(x)\sqrt{q_1(x)}\le c\quad\text{for all}\quad x\in\mathbb R.\label{3.13} \end{gather} \end{thm} \section{Main results} Throughout the sequel we assume that our standing requirements to the functions $q$ (see \eqref{1.2}), and $\mu$ and $\theta$ (see \S1) are satisfied, and we do not mention them in the statements. \begin{thm}\label{thm4.1} Suppose that the function $q$ is nonnegative and continuous at every point of the real axis. Suppose that for a given $p\in[1,\infty)$ the following condition holds: \begin{equation}\label{4.1} \int_{-\infty}^0\mu(t)^pdt=\int_0^\infty \mu(t)^pdt=\infty. \end{equation} Then the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1} only if inequalities \eqref{2.1} hold. \end{thm} To make our a priori requirements independent of the parameter $p\in[1,\infty)$, throughout the sequel we assume that together with \eqref{1.2}, condition \eqref{2.1} holds. Similar to \eqref{1.2}, below this condition is not quoted and does not appear in the statements. \begin{lem}\label{lem4.2} Suppose that the following condition holds: \begin{equation}\label{4.2} \int_{-\infty}^0\mu(t)dt=\int_0^\infty \mu(t)dt=\infty. \end{equation} Then for every $p\in[1,\infty)$ equation \eqref{2.4} has no solutions $z\in L_{p,\mu}$ apart from $z\equiv0.$ \end{lem} Note that for $\mu\equiv1$ \lemref{lem4.2} was proved in \cite{2}. Our main result is the following. \begin{thm}\label{thm4.3} Suppose that condition \eqref{4.2} holds. Then the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1} if and only if the operator $S:L_p\to L_p$ is bounded. Here \begin{equation}\label{4.3} (Sf)(x)=\mu(x)\int_{-\infty}^\infty\frac{G(x,t)}{\theta(t)}f(t)dt,\quad x\in\mathbb R,\quad f\in L_p. \end{equation} \end{thm} Note that for $\mu\equiv\theta\equiv1$ \thmref{thm4.3} was proved in \cite{6}. Thus, this theorem reduces the original problem on the admissibility of the pair $\{L_{p,\mu};L_{p,\theta}\}$ for \eqref{1.1} to the boundedness of the integral operator $S:L_p\to L_p$ (see \eqref{4.3}). This result is clearly useful for the investigation of \eqref{1.1} for the following reason. Consider, say, the case $p\in(1,\infty)$. The operator $S$ is a sum of two operators of Hardy type (see \eqref{2.9}, \eqref{2.15} and \eqref{2.18}): \begin{align} &(S_1f)(x)=\mu(x)u(x)\int_{-\infty}^x\frac{v(t)}{\theta(t)}f(t)dt,\quad x\in\mathbb R,\label{4.4}\\ &(S_2f)(x)=\mu(x)v(x)\int^{\infty}_x\frac{u(t)}{\theta(t)}f(t)dt,\quad x\in\mathbb R.\label{4.5} \end{align} For the norms $\|S_1\|_{p\to p},$ $\|S_2\|_{p\to p},$ we know sharp by order two-sided estimates (see \eqref{2.17} and \eqref{2.20}), which can be expressed in terms of the weights $\mu,\theta$ and a FSS $\{u,v\}$ of equation \eqref{2.4}. The solutions $\{u,v\}$ can, in turn, be expressed in terms of the implicit function $\rho$ (see \eqref{2.10}), for which in turn one has sharp by order estimates in terms of the function $d$ (see \eqref{2.12} and \eqref{2.2}). Finally, for the implicit function $d,$ which is, in general, not computable, as well as the function $\rho,$ we have sharp by order two-sided estimates, which can be expressed in terms of the original function $q$ (see \eqref{3.12}, \eqref{3.13}). Thus, this long chain of estimates yields some information allowing us to find conditions for the boundedness of the operator $S_i: L_p\to L_p,$ $i=1,2$ (and hence of the operator $S:L_p\to L_p,$ $p\in (1,\infty)),$ which are expressed in terms of the weights $\mu,\theta$ and the function $q.$ We want to emphasize that these conditions become precise if we are able to use the information obtained from the estimates in an ingenious way (see, say, \cite{6} where similar arguments were used). One can compare this approach to that of applying the Cauchy criterion for the convergence of a number series to getting various working criteria, convenient for practical investigation of a given number series. In a similar way, \thmref{thm4.3} can be used for deducing convenient particular tests for the admissibility of the pair $\{L_{p,\mu};L_{p,\theta}\}$, $p\in[1,\infty)$, for a given equation \eqref{1.1}. Here is an example. The assertion given below (\thmref{thm4.7}) is obtained by using one of the possible ways for practical implementation of the approach to the study of \eqref{1.1} presented above. To formulate \thmref{thm4.7}, we need some new definitions, auxiliary assertions and comments. \begin{defn}\label{defn4.4} We say that the function $q$ belongs to the class $H$ (and write $q\in H)$ if the following equality holds: \begin{equation}\label{4.6} \lim_{|x|\to\infty}\nu(x)=0. \end{equation} Here \begin{equation}\label{4.7} \nu(x)=d(x)\int_0^{\sqrt 2d(x)}(q(x+t)-q(x-t))dt,\quad x\in\mathbb R. \end{equation} \end{defn} In the next assertion, we state an important property of the functions $q\in H.$ \begin{lem}\label{lem4.5} Let $q\in H.$ Then for any $\varepsilon>0$ there is a constant $c(\varepsilon)\in[1,\infty)$ such that for all $x,t\in\mathbb R$ the following inequalities hold: \begin{equation}\label{4.8} c(\varepsilon)^{-1}\exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right)\le\frac{d(t)} {d(x)}\le c(\varepsilon)\exp\left(\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right). \end{equation} \end{lem} Note that for $\varepsilon\ge 1/\sqrt 2$ inequalities \eqref{4.8} hold regardless of condition \eqref{4.6}. Indeed, under conditions \eqref{1.2} and \eqref{2.1}, the function $d(x), $ $x\in\mathbb R$ is well-defined, differentiable, and satisfies the following relations (see Lemmas \ref{lem2.1} and \ref{lem3.1}: \begin{align} -\varepsilon\le-\frac{1}{\sqrt2}&\le d'(\xi)\le\frac{1}{\sqrt2}\le\varepsilon,\quad\xi\in\mathbb R\quad\Rightarrow\nonumber\\ -\frac{\varepsilon}{d(\xi)}&\le\frac{d'(\xi)}{d(\xi)} \le\frac{\varepsilon}{d(\xi)},\quad\xi\in\mathbb R \quad\Rightarrow\nonumber\\ \exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right)&\le\frac{d(t)}{d(x)}\le \exp\left(\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right),\quad x,t\in\mathbb R. \label{4.9} \end{align} This means that in contrast with \eqref{4.9}, for $\varepsilon\in(0,1\sqrt 2)$ estimates \eqref{4.8} arise because of condition \eqref{4.6}. \begin{defn}\label{defn4.6} Let $q\in H.$ We say that a pair of weights (weight functions) $\{\mu,\theta\}$ agrees with the function $q$ if for any $\varepsilon>0$ there is a constant $c(\varepsilon)\in[1,\infty)$ such that for all $t,x\in\mathbb R$ one has the inequalities \begin{equation}\label{4.10} c(\varepsilon)^{-1}\exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right)\le\sqrt{ \frac{d(t)}{d(x)}}\frac{\mu(t)}{\mu(x)};\ \sqrt{\frac{d(t)}{d(x)}}\frac{\theta(x)}{\theta(t)}\le c(\varepsilon)\exp\left(\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right). \end{equation} In the latter case, we say that the pair $\{L_{p,\mu};L_{p,\theta}\},$ $p\in [1,\infty)$, agrees with equation \eqref{1.1} \end{defn} \begin{thm}\label{thm4.7} Suppose that conditions \eqref{4.2} hold. Let $q\in H.$ Suppose that the pair $\{L_{p,\mu};L_{p,\theta}\},$ $p\in [1,\infty)$ agrees with equation \eqref{1.1}. Then this pair is admissible for \eqref{1.1} if and only if $m(q,\mu,\theta)<\infty.$ Here \begin{equation}\label{4.11} m(q,\mu,\theta)=\sup_{x\in\mathbb R}\left(\frac{\mu(x)}{\theta(x)}d^2(x)\right). \end{equation} \end{thm} To prove inequalities \eqref{4.10}, the following lemma can be useful. \begin{lem}\label{lem4.8} Suppose that a function $\mu(x)$ is defined, positive and differentiable for all $x\in\mathbb R,$ let $q\in H,$ and let $d(x),$ $x\in\mathbb R,$ denote the auxiliary function from \lemref{lem2.1}. Then, if the equality \begin{equation}\label{4.12} \lim_{|x|\to\infty}\frac{\mu'(x)}{\mu(x)}d(x)=0 \end{equation} holds, then for any given $\varepsilon>0$ there is a constant $c(\varepsilon)\in(0,\infty)$ such that for all $t,x\in\mathbb R$ inequalities \eqref{4.10} hold. \end{lem} The next assertions are convenient for the study of concrete equations. They are obvious and are given without proofs. \begin{thm}\label{4.9} Let $q\in H,$ and suppose that \begin{gather} d_0\doe\sup_{x\in\mathbb R}d(x)=\infty,\label{4.13}\\ \int_{-\infty}^0q^*(x)dx=\int_0^\infty q^*(x)dx=\infty,\quad q^*(x)=\frac{1}{d^2(x)},\quad x\in\mathbb R.\label{4.14} \end{gather} Then the following assertions hold: \begin{enumerate} \item[A)] for $p\in[1,\infty)$ the pair $\{L_p;L_p\}$ is not admissible for \eqref{1.1}; \item[B)] for $p\in[1,\infty)$ the pair $\{L_{p,q^*};L_p\}$ is admissible for \eqref{1.1}. \end{enumerate} \end{thm} \begin{thm}\label{4.10} Let $q\in H,$ and suppose that the weight function $\theta(x),$ $x\in\mathbb R,$ is such that $m_0>0$ where \begin{equation}\label{4.15} m_0=\inf_{x\in\mathbb R}(q^*(x)\theta(x)),\quad q^*(x)=\frac{1}{d^2(x)}. \end{equation} Then for $p\in[1,\infty)$ the pair $\{dL_p;L_{p,\theta}\}$ is admissible for \eqref{1.1}. \end{thm} \section{Proofs} \begin{proof}[Proof of \lemref{lem3.1}] The existence of the derivative $d'(x),$ $x\in\mathbb R$ is a consequence of the theory of implicit functions \cite[Ch.II,\S1,no.3]{7}. It is proven in the same way as in \cite{5}. The following relations are deduced from \eqref{2.2}: $$\int_0^{\sqrt 2d(x)}\int_{x-t}^{x+t}q(\xi)d\xi dt=2\quad \Rightarrow$$ \begin{align*} 0&=\sqrt2d'(x)\int_{x-\sqrt 2d(x)}^{x+\sqrt2d(x)}q(\xi)d\xi+\int_0^{\sqrt 2 d(x)}[q(x+t)-q(x-t)]dt\\ &=\sqrt2d'(x)\int_{x-\sqrt2d(x)}^{x+\sqrt2d(x)}q(\xi)d\xi+\left[\int_x^{x+\sqrt2d(x)}q(\xi)d\xi- \int_{x-\sqrt2d(x)}^xq(\xi)d\xi\right]\quad\Rightarrow\\ |d'(x)|&=\frac{1}{\sqrt 2}\left|\int_x^{x+\sqrt2d(x)}q(\xi)d\xi-\int_{x-\sqrt2d(x)}^x q(\xi)d\xi\right| \left(\int_{x-\sqrt2d(x)} ^{x+\sqrt2d(x)}q(\xi)d\xi\right)^{-1}\le\frac{1}{\sqrt2}. \end{align*} \end{proof} \begin{proof}[Proof of \lemref{lem3.3}] Below we use Lagrange's formula and \eqref{3.1}: \begin{align*} &|d(t)-d(x)|=|d'(\theta)|\, |t-x|\le\frac{d(x)}{\sqrt2}\quad\Rightarrow\\ &d(t)\le\left(1+\frac{1}{\sqrt2}\right)d(x)\le 4d(x)\quad\text{for}\quad t\in[x-d(x),\, x+d(x)]\\ &d(t)\ge\left(1-\frac{1}{\sqrt2}\right)d(x)\ge \frac{d(x)}{4}\quad\text{for}\quad t\in[x-d(x),\, x+d(x)]. \end{align*} \end{proof} \begin{proof}[Proof of \lemref{lem3.4}] Below we use \eqref{2.12} and \eqref{3.2}: \begin{align*} \int_{x-d(x)}^{x+d(x)}\frac{d(\xi)}{\rho(\xi)}&=\int_{x-d(x)}^{x+d(x)}\frac{d(\xi)}{\rho(\xi)}\cdot \frac{d(x)}{d(\xi)}\cdot\frac{d\xi}{d(x)}\le 2\sqrt 2\cdot 4\cdot 2=c<\infty,\\ \int_{x-d(x)}^{x+d(x)}\frac{d\xi}{\rho(\xi)}&=\int_{x-d(x)}^{x+d(x)} \frac{d(\xi)}{\rho(\xi)} \cdot \frac{d(x)}{d(\xi)}\cdot\frac{d\xi}{d(x)}\ge \frac{1}{\sqrt 2}\cdot \frac{1}{4}\cdot 2\ge c^{-1}>0. \end{align*} Now we use this together with \eqref{2.10} and obtain \begin{align*} \frac{u(t)}{u(x)}&\ge\sqrt{\frac{\rho(t)}{\rho(x)}}\exp\left(-\frac{1}{2}\left|\int_x^t\frac{d\xi} {\rho(\xi)}\right|\right)\ge\sqrt{\frac{d(x)}{\rho(x)}\cdot\frac{d(t)}{d(x)}\cdot\frac{\rho(t)} {d(t)}}\exp\left(-\frac{1}{2}\left|\int_{x-d}^{x+d}\frac{d\xi}{\rho(\xi)}\right|\right)\\ &\ge c^{-1}>0;\\ \frac{u(t)}{u(x)}&\le\sqrt{\frac{\rho(t)}{\rho(x)}}\exp\left(\frac{1}{2}\left|\int_x^t\frac{d\xi} {\rho(\xi)}\right|\right)\le\sqrt{\frac{d(x)}{\rho(x)}\cdot\frac{d(t)}{d(x)}\cdot\frac{\rho(t)} {d(t)}}\exp\left(\frac{1}{2}\left|\int_{x-d}^{x+d}\frac{d\xi}{\rho(\xi)}\right|\right)\\ &\ge c<\infty. \end{align*} Inequalities \eqref{3.3} for the solution $v$ are checked similarly, and estimates \eqref{3.3} for $\rho$ follow from the estimates of $u$ and $v$ and \eqref{2.8}. \end{proof} \begin{proof}[Proof of \lemref{lem3.5}] To prove that the function $F(\eta)$ is differentiable and the functions $F(\eta)$ and $F'(\eta)$ are non-negative for $\eta\ge0$, we use properties of integral. The last assertion of the lemma follows from Lagrange's formula and the relations $$F(\eta)-2=F(\eta)-F(d(x))=F'(\theta)(\eta-d(x)).$$ \end{proof} \begin{proof}[Proof of \lemref{lem3.6}] To obtain \eqref{3.6}, we use the following simple transformations \begin{align*} \int_{x-t}^{x+t}f(\xi)d\xi&=\int_0^t[f(x+t_1)+f(x-t_1)]dt_1=2f(x)t+\int_0^t[f(x+t_1)-f(x)]dt\\ &\quad -\int_0^t[f(x)-f(x-t_1)]dt_1=2f(x)t+\int_0^t\left[\int_0^{t_1}(f(x+t_2))'dt_2\right]dt_1\\ &\quad-\int_0^t\left[\int_0^{t_1}(f(x-t_2))'dt_2\right]dt_1=2f(x)t+\int_0^t\int_0^{t_1} [f(x+t_2)-f(x-t_2)]'dt_2dt_1\\ &=2f(x)t+\int_0^t\int_0^{t_1}\int_{x-t_2}^{x+t_2}f''(t_3)dt_3dt_2dt_1. \end{align*} \end{proof} \begin{proof}[Proof of \thmref{thm3.7}] Set $$\eta(x)=\frac{1-\delta(x)}{\sqrt{q_1(x)}},\qquad \delta(x)=2(\varkappa_1(x)+\varkappa_2(x)), \qquad |x|\gg1.$$ Then by \eqref{3.4}, \eqref{3.6}, \eqref{3.7}, \eqref{3.8}, \eqref{3.9}, \eqref{3.10} and \eqref{3.11}, we have \begin{align*} F(\eta(x))&=\int_0^{\sqrt2\eta(x)}\int_{x-t}^{x+t}q_1(\xi)d\xi dt+\int_0^{\sqrt2\eta(x)}\int_{x-t}^{x+t}q_2(\xi)d\xi dt\\ &\le \int_0^{\sqrt 2\eta(x)}\left[2q_1(x)t+\int_0^t\int_0^{t_1}\int_{x-t_2}^{x+t_2}q_1''(t_3)dt_3dt_2dt_1\right]\\ &\quad +\sqrt 2\eta(x)\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}q_2(\xi)d\xi\right|\le\left(\sqrt 2\eta(x)\right)^2q_1(x)\\ &\quad +\frac{\left(\sqrt 2\eta(x)\right)^3}{6}\sup_{t_2\in A(x)}\left|\int_{x-{t_2}}^{x+{t_2}}q_1''(\xi)d\xi\right|+\sqrt 2(1-\delta(x))\varkappa_2(x)\\ &\le 2(1-\delta(x))^2+\frac{\sqrt 2}{3}(1-\delta(x))^3\varkappa_1(x)+\sqrt 2\varkappa_2(x)\\ &\le 2[(1-\delta(x))^2+\varkappa_1(x)+\varkappa_2(x)]\\ &=2\left[1-\frac{\delta(x)}{2}-\left(\frac{\delta (x)}{2}-\delta^2(x)\right)-\varkappa_1(x)-\varkappa_2(x)\right]\le 2. \end{align*} Hence $d(x)\ge \eta(x)$ for $|x|\gg1$ by \lemref{lem3.5}. Let now $$\eta(x)=\frac{1+\delta(x)}{\sqrt{q_1(x)}},\qquad \delta(x)=2(\varkappa_1(x)+\varkappa_2(x)),\qquad |x|\gg1.$$ Then by the same arguments we obtain: \begin{align*} F(\eta(x))&=\int_0^{\sqrt 2\eta(x)}\int_{x-t}^{x+t}q_1(\xi)d\xi dt+\int_0^{\sqrt 2\eta(x)} \int_{x-t}^{x+t}q_2(\xi)d\xi dt\\ &\ge \int_0^{\sqrt \eta(x)}\int_{x-t}^{x+t}\left[2q_1(x)t+\int_0^t\int_0^{t_1}\int_{x-t_2}^{x+t_2}q_1''(t_3) dt_3dt_2dt_1\right]dt\\ &\quad -\sqrt2 \eta(x)\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}q_2(\xi)d\xi\right|\ge\left(\sqrt 2\eta(x)\right)^2q_1(x)\\ &\quad -\frac{\left(\sqrt 2\eta(x)\right)^3}{6}\sup_{t_2\in A(x)}\left|\int_{x-t_2}^{x+t_2}q_1''(t_3)dt_3\right|-\sqrt 2(1+\delta(x))\varkappa_2(x)\\ &\ge 2(1+\delta(x))^2-\frac{\sqrt 2}{3}(1+\delta(x))^3\varkappa_1(x)-2\varkappa_2(x)\\ &\ge 2(1+\delta(x))+\varkappa_1(x)+\varkappa_2(x)\ge 2. \end{align*} Hence $d(x)\le\eta(x)$ for $|x|\gg1$ by \lemref{lem3.5}, and equality \eqref{3.12} is proven. Further, since the function $d(x)\sqrt{q_1(x)}$ is continuous and positive for all $x\in\mathbb R,$ for all $x_0\in (0,\infty)$ we the inequalities: \begin{gather*} 0<m\le f(x)\le M<\infty,\quad |x|\le x_0\\ m=\min_{|x|\le x_0}f(x),\quad M=\max_{|x|\le x_0} f(x),\quad f(x)=d(x)\sqrt{q_1(x)}. \end{gather*} Together with \eqref{3.12}, this implies \eqref{3.13}. \end{proof} \begin{proof}[Proof of \thmref{thm4.1}] Assume the contrary. Then \eqref{4.1} holds, the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1}, and there exists $x_0\in\mathbb R$ such that one of inequalities \eqref{2.1}, say, the second one, does not hold: \begin{equation}\label{5.1} \int_{x_0}^\infty q(t)dt=0\quad \Rightarrow\quad q(x)\equiv 0,\qquad x\in[x_0,\infty). \end{equation} Without loss of generality, in what follows we assume $x_0\ge 1.$ Let us introduce the functions $\varphi$ and $f_0.$ \begin{align} &1)\quad\varphi\in C^\infty(\mathbb R),\quad \supp\varphi=[x_0,\infty),\quad 0\le\varphi(x)\le 1\quad for \quad x\in\mathbb R,\label{5.2}\\ &\qquad\qquad\qquad\qquad \varphi(x)\equiv 1\quad \text{for}\quad x\ge x_0+1\label{5.3}\\ &2)\quad f_0(x):=-\varphi''(x)+q(x)\varphi(x),\quad x\in\mathbb R.\label{5.4} \end{align} {}From 1)--2) we obtain the equality \begin{gather}q(x)\varphi(x)\equiv0,\qquad x\in\mathbb R\quad\Rightarrow\nonumber\\ f_0(x)=-\varphi''(x),\quad x\in\mathbb R\quad\Rightarrow \quad \supp f_0=[x_0,x_0+1].\label{5.5} \end{gather} According to \eqref{5.5}, we conclude that $f_0\in L_{p,\theta}:$ $$\|f_0\|_{L_{p,\theta}}^p=\int_{-\infty}^\infty |\theta(x)f_0(x)|^pdx=\int_{x_0}^{x_0+1}|\theta(x)\varphi''(x)|^pdx=c(x_0)<\infty.$$ Since the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1}, we conclude that \eqref{1.1} for $f=f_0$ has a unique solution $y_0\in L_{p,\mu}.$ Then (see \eqref{5.4} and \eqref{5.5}) \begin{equation}\label{5.6} y_0(x)=\varphi(x)+z(x),\qquad x\in\mathbb R, \end{equation} where $z(x),$ $x\in\mathbb R,$ is some soluton of \eqref{2.4}. From \eqref{2.4} and \eqref{5.1}, we obtain the equality \begin{equation}\label{5.7} z''(x)=0\quad\text{for}\quad x\in [x_0,\infty)\quad \Rightarrow\quad z(x)=c_1+c_2x\quad\text{for}\quad x\ge x_0. \end{equation} Let us show that $c_2=0.$ Assume to the contrary that $c_2\ne0. $ Choose $x_1$ so that to have the inequality \begin{equation}\label{5.8} \frac{|1+c_1|}{|c_2|}\cdot\frac{1}{x}\le\frac{1}{2}\quad\text{for}\quad x\ge x_1\ge x_0+1. \end{equation} Then (see \eqref{5.3}) \begin{align*} \infty&>\|y_0\|_{p,\mu}^p\ge\int_{x_1}^\infty \mu(x)^p|\varphi(x)+z(x)|^pdx=\int_{x_1}^\infty\mu(x)^p|1+c_1+c_2x|^pdx\\ &\ge |c_2x_1|^p\int_{x_1}^\infty \mu(x)^p\left|1-\left|\frac{1+c_1}{c_2}\right|\frac{1}{x} \right|^pdx\ge\left|\frac{c_2x_1}{2}\right|^p\int_{x_1}^\infty\mu(x)^pdx=\infty, \end{align*} and we get a contradiction. Hence $c_2=0.$ Let us check that also $c_1=0.$ Assume that $c_1\ne0.$ Since $\varphi\in C^\infty(\mathbb R),$ from \eqref{5.2} it follows that $\varphi(x_0)=\varphi'(x_0)=0$ and therefore (see \eqref{5.7}): \begin{align*} y(x_0)&=\varphi(x_0)+z(x_0)=c_1,\\ y'(x_0)&=\varphi'(x_0)+z'(x_0)=0. \end{align*} In addition, $\varphi(x)\equiv0$ for $x\le x_0,$ and therefore from \eqref{5.5} and \eqref{5.6} it follows that the function $z$ is a solution of the Cauchy problem \begin{numcases} {} z''(x)=q(x)z(x),\qquad\qquad x\le x_0\label{5.9}\\ z(x_0)=c_1,\quad z'(x_0)=0.\label{5.10} \end{numcases} Further, without loss of generality, we assume that $c_1=1.$ Let us check that then we have the inequality \begin{equation}\label{5.11} z(x)\ge1\qquad\text{for}\qquad x\le x_0. \end{equation} Towards this end, first note that since $z(x_0)=1,$ we have $z(x)>0$ in some left half-neighborhood of the point $x_0$ (i.e., for $x\in(x_0-\varepsilon,x_0]$ for some $\varepsilon>0).$ But then $z(x)>0$ for all $x<x_0.$ Indeed, if this is not the case, then $z(x)$ has at least one zero on $(-\infty,x_0).$ Let $\tilde x$ be the first zero of $z(x)$ to the left from $x_0.$ Then $z'(\tilde x)\ge0.$ Indeed, if $z'(\tilde x)<0$ and $z(\tilde x)=0,$ then $z(x)<0$ in some right half-neighborhood of the $\tilde x.$ But $z(x_0)=1$ and $\tilde x<x_0.$ Hence, the interval $(\tilde x,x_0)$ contains a zero of $z(x),$ contrary to the definition of the point $\tilde x.$ Thus $z'(\tilde x)\ge0.$ On the other hand, \begin{align*} &z'(x_0)-z'(\tilde x)=\int_{\tilde x}^{x_0}q(\xi)z(\xi)d\xi\qquad\Rightarrow\\ &z'(\tilde x)=-\int_{\tilde x}^{x_0}q(\xi)z(\xi)\le 0. \end{align*} Hence $z'(\tilde x)=0.$ But then the function $z(x)$ is a solution of the Cauchy problem \begin{align*} z''(x)=q(x)z(x),\quad x\le x_0 \\ z(\tilde x)=z'(\tilde x)=0 \end{align*} $$\qquad\Rightarrow\qquad z(x)\equiv 0,\quad x\le x_0.$$ We get a contradiction because $z(x_0)=1.$ Thus $z(x)>0$ for $x\le x_0.$ Then for $x\le x_0$, we have $$-z'(x)=z'(x_0)-z'(x)=\int_x^{x_0}q(\xi)z(\xi)d\xi\ge0\quad\Rightarrow \quad z'(x)\le 0,\quad x\le x_0.$$ Hence $z(x)\ge z(x_0)=1$ for $x\le x_0.$ This implies that \begin{align*} \infty&>\|y_0\|_{p,\mu}=\int_{-\infty}^\infty |\mu(x)y_0(x)|^pdx\ge\int_{-\infty}^{x_0}|\mu(x)y_0(x)|^pdx\\ &=\int_{-\infty}^{x_0}|\mu(x)z(x)|^pdx\ge\int_{-\infty}^{x_0}\mu(x)^pdx=\infty. \end{align*} We get a contradiction. Hence $c_1=0,$ and we obtain the equality $$y_0(x)=\varphi(x),\qquad x\ge x_0 \qquad \Rightarrow$$ \begin{align*} \infty&>\|y_0\|_{p,\mu}^p\ge\int_{x_0+1}^\infty|\mu(x)y_0(x)|^pdx=\int_{x_0+1}^\infty|\mu(x)\varphi (x)|^pdx\\ &=\int_{x_0+1}^\infty\mu(x)^pdx=\infty. \end{align*} We get a contradiction. Hence \eqref{5.1} does not hold. \end{proof} \begin{proof}[Proof of \lemref{lem4.2}] Let us show that in the case of \eqref{4.2} for all $p\in[1,\infty)$ we have the equalities \begin{equation}\label{5.12} \int_{-\infty}^0(\mu(t)u(t))^pdt=\int_0^\infty(\mu(t)v(t)^pdt=\infty. \end{equation} We only consider the second equality because the first one can be proved in the same way. For $p=1$ equality \eqref{5.11} follows from \thmref{thm2.2} and \eqref{4.2} is a straightforward manner. Let $p\in(1,\infty)$, $p'=p(p-1)^{-1}.$ The following relations rely only on \thmref{thm2.2}: \begin{align} \int_0^\infty&\frac{dt}{v(t)^{p'}} =\int_0^\infty\frac{v'(t)v(t)^{-p'}}{v'(t)}dt\le \frac{1}{v'(0)}\int_0^\infty v'(t)v(t)^{-p'}dt\nonumber\\ &=\frac{1}{p'-1}\frac{1}{v'(0)}\left(\frac{1}{v(0)^{p'-1}}-\frac{1}{v(\infty)^{p'-1}}\right) \le \frac{1}{p'-1}\frac{1}{v'(0)v(0)^{p'-1}}=c(p)<\infty.\label{5.13} \end{align} Let $A>0.$ Below we use H\"older's inequality and \eqref{5.13}: $$\int_0^A\mu(t)dt\le\left[\int_0^A(\mu(t)v(t))^pdt\right]^{1/p}\cdot\left[\int_0^A\frac{dt}{v(t)p'}\right] ^{1/p'}\le c(p)\left[\int_0^A (\mu(t)v(t))^pdt\right]^{1/p}.$$ Now, to obtain \eqref{5.12}, in the last inequality we let $A$ tend to infinity. Let us now go over to the proof of the lemma. By \thmref{thm2.2}, the general solution of \eqref{2.4} is of the form $$z(x)=c_1u(x)+c_2v(x),\qquad x\in\mathbb R.$$ Let $z\in L_{p,\mu}.$ Then $c_2=0$. Indeed, if $c_2\ne0,$ then denote $x_1\gg1,$ a number such that for all $x\ge x_1$ we have the inequality (see \eqref{2.7}): \begin{equation}\label{5.14} \left|\frac{c_1}{c_2}\right|\frac{u(x)}{v(x)}\le\frac{1}{2},\qquad x\ge x_1. \end{equation} Now from \eqref{5.12}, \eqref{5.14} and \thmref{thm2.2} it follows that \begin{align*} \infty&>\|z\|_{p,\mu}^p=\int_{-\infty}^\infty|\mu(x)(c_1u(x)+c_2v(x)|^pdx\\ &\ge|c_2 |^p\int_{x_1}^\infty(\mu(x)v(x))^p\left| 1-\left|\frac{c_1}{c_2}\right|\, \frac{u(x)}{v(x)}\right|^pdx\ge\left|\frac{c_2}{2}\right|^p\int_{x_1}^\infty(\mu(x)v(x))^pdx=\infty. \end{align*} We get a contradiction. Hence $c_2=0.$ The equality $c_1=0$ now follows from \eqref{2.5} and \eqref{5.12}. \end{proof} \begin{proof}[Proof of \thmref{thm4.3} for $p\in(1,\infty)$] \textit{Necessity}. We need the following lemma. \begin{lem}\label{lem5.1} Let $p\in[1,\infty)$. Suppose that conditions \eqref{4.2} hold, and the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1}. Then, if $f\in L_p$ and $\supp f=[x_1,x_2],$ $x_2-x_1<\infty,$ then $f\in L_{p,\theta}$ and the solution $y\in L_{p,\mu}$ of \eqref{1.1} which corresponds to $f$ is of the form \eqref{2.13}. \end{lem} \renewcommand{\qedsymbol}{} \begin{proof} Below we only consider the case $p\in (1,\infty)$ (for $p=1$ the arguments are similar). Let us continue the function $f$ by zero beyond the segment $[x_1,x_2]$ and maintain the original notation. From the obvious inequalities \begin{equation}\label{5.15} c^{-1}\le\theta(x)\le c,\quad x\in[x_1,x_2],\quad c=c(x_1,x_2), \end{equation} it follows that $f\in L_{p,\theta}.$ Set (see \eqref{2.9}, \eqref{2.13}) \begin{align} \tilde y(x)&=\int_{-\infty}^\infty G(x,t)f(t)dt\nonumber\\ &=u(x)\int_{-\infty}^xv(t)f(t)dt+v(x)\int_x^\infty u(t)f(t)dt,\quad x\in\mathbb R.\label{5.16} \end{align} Let us estimate the integrals in \eqref{5.16}: \begin{align} \int_{-\infty}^xv(t)|f(t)|dt&\le\left[\int_{x_1}^{x_2}\left(\frac{v(t)}{\theta(t)}\right)^{p'}dt \right]^{1/p'}\cdot\left[\int_{x_1}^{x_2}|\theta(t)f(t)|^pdt\right]^{1/p}\nonumber\\ &\le c\left(\int_{x_1}^{x_2}v(t)^{p'}dt\right)^{1/p'}\cdot\|f\|_{p,\theta},\quad x\in\mathbb R,\label{5.17}\\ \int_{x}^\infty u(t)|f(t)|dt&\le\left[\int_{x_1}^{x_2}\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt \right]^{1/p'}\cdot\left[\int_{x_1}^{x_2}|\theta(t)f(t)|^pdt\right]^{1/p}\nonumber\\ &\le c\left(\int_{x_1}^{x_2}u(t)^{p'}dt\right)^{1/p'}\cdot\|f\|_{p,\theta},\quad x\in\mathbb R.\label{5.18}. \end{align} {}From \eqref{5.17} and \eqref{5.18} it follows that the function $\tilde y(x),$ $s\in\mathbb R,$ is well-defined. It is also easy to see that the function $\tilde y(x),$ $x\in\mathbb R$ is a particular solution of \eqref{1.1}. But, since $f\in L_{p,\theta},$ \eqref{1.1} has a unique solution $y\in L_{p,\theta}.$ This means that we have the equality $$y(x)=\tilde y(x)+c_1u(x)+c_2v(x),\qquad x\in\mathbb R.$$ Let us check that $c_1=c_2=0.$ Assume, say, that $c_2\ne0.$ Then for $x\ge x_2$, we get \begin{align*} |y(x)|&\ge |c_2|v(x)-|c_1|u(x)-u(x)\int_{x_1}^{x_2}v(t)|f(t)|dt\\ &=|c_1|v(x)\left[1-\left|\frac{c_1}{c_2}\right|\frac{u(x)}{v(x)}-\frac{u(x)}{v(x)}\int_{x_1}^{x_2} v(t)|f(t)|dt\right]. \end{align*} {}From \eqref{2.7} and \eqref{5.17} it follows that there exists $x_3\ge\max\{1,x_2\}$ such that $$|y(x)|\ge\frac{1}{2}|c_2|v(x)\qquad\text{for}\quad x\ge x_3\qquad\Rightarrow \qquad \text{(see \eqref{5.12}):}$$ $$\infty>\|y\|_{p,\mu}^p\ge\int_{x_1}^\infty|\mu(x)y(x)|^pdx\ge\left|\frac{c_2}{2}\right|^p\int_{x_3}^\infty |\mu(x)v(x)|^pdx=\infty.$$ We get a contradiction. Hence $c_2=0.$ Similarly, we prove that also $c_1=0,$ and therefore $y=\tilde y$ (see \eqref{5.16}). Let $[x_1,x_2]$ be any finite segment. Set \begin{equation}\label{5.19} f(t)=\begin{cases} \theta(t)^{-p'}\cdot u(t)^{p'-1},&\quad t\in [x_1,x_2]\\ 0,&\quad t\notin [x_1,x_2]\end{cases} \end{equation} Then \begin{equation}\label{5.20} \|f\|_{L_{p,\theta}}^p=\int_{x_1}^{x_2}|\theta(t)f(t)|^pdt=\int_{x_1}^{x_2}\frac{\theta(t)^pu^{p(p'-1)}(t)}{\theta(t)^{p'p}} dt=\int_{x_1}^{x_2}\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt<\infty. \end{equation} Therefore, since the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1}, in the case of \eqref{5.19} equation \eqref{1.1} has a solution $y\in L_{p,\mu}.$ This solution is of the form \eqref{2.13} (see \lemref{lem5.1}). This implies that \begin{align} &\infty>\|y\|_{p,\mu}^p=\int_{-\infty}^\infty|\mu(x)y(x)|^pdx\nonumber\\ &=\left\{\int_ {-\infty}^\infty\mu(x)^p\left[u(x)\int_ {-\infty}^x v(t)f(t)dt+v(x) \int_x^\infty u(t)f(t)dt\right]^pdx\right\}\nonumber\\ &\ge\int_{-\infty}^\infty(\mu(x)v(x))^p\left(\int_x^\infty u(t)f(t)dt\right)^pdx\ge\int_{-\infty}^{x_1}(\mu(x)v(x))^p\left(\int_x^\infty u(t)f(t)dt\right)^pdx\nonumber\\ &\ge\int_{-\infty}^{x_1}(\mu(x)v(x))^pdx \left(\int_{x_1}^{x_2}u(t)f(t)dt\right)^p=\int_{-\infty}^{x_1}(\mu(x)v(x))^pdx \left(\int_{x_1}^{x_2}\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt\right)^p\label{5.21} \end{align} Now, using \eqref{5.21}, \eqref{5.20} and \eqref{1.5}, we obtain \begin{align*} \left[\int_{-\infty}^{x_1}(\mu(x)v(x))^pdx\right]^{1/p}\int_{x_1}^{x_2}\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt&\le \|y\|_{p,\mu}\le c(p)\|f\|_{p,\theta}\\ &=c(p)\left[\int_{x_1}^{x_2}\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt\right]^{1/p}\quad\Rightarrow\\ \end{align*} $$\left(\int_{-\infty}^{x_1}(\mu(t)v(t))^pdt\right)^{1/p}\left(\int_{x_1}^{x_2}\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt \right)^{1/p'}\le c(p)<\infty.$$ Since in this inequality $x_1$ and $x_2$ $(x_1\le x_2)$ are arbitrary numbers, we conclude that $$M=\sup_{x\in\mathbb R}\left(\int_{-\infty}^{x}(\mu(t)v(t))^pdt\right)^{1/p}\cdot\left(\int_x^\infty\left( \frac{u(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}\le c(p)<\infty.$$ This inequality means that the operator $S_2:L_p\to L_p,$ \begin{equation}\label{5.22} (S_2f)(x)=\mu(x)v(x)\int_x^\infty\frac{u(t)}{\theta(t)}f(t)dt,\quad x\in\mathbb R \end{equation} is bounded (see \thmref{thm2.7}). Similarly, we use \thmref{thm2.8} to conclude that the operator $S_1:L_p\to L_p,$ \begin{equation}\label{5.23} (S_1f)(x)=\mu(x)u(x)\int_{-\infty}^x\frac{v(t)}{\theta(t)}f(t)dt,\quad x\in\mathbb R \end{equation} is bounded. Since we have the equality (see \eqref{2.9} and \eqref{4.3}) \begin{equation}\label{5.24} S=S_1+S_2 \end{equation} our assertion now follows from the triangle inequality for norms. \end{proof} \end{proof} \newpage \begin{proof}[Proof of \thmref{thm4.3}] \textit{Sufficiency.} \renewcommand{\qedsymbol}{\openbox} \begin{lem}\label{lem5.2} Let $p\in[1,\infty),$ and let $S,$ $S_1,$ $S_2$ be operators \eqref{4.3}, \eqref{5.23} and \eqref{5.22}, respectively. Then we have the inequalities \begin{equation}\label{5.25} \frac{\|S_1\|_{p\to p}+\|S_2\|_{p\to p}}{2}\le \|S\|_{p\to p}\le\|S_1\|_{p\to p}+\|S_2\|_{p\to p}. \end{equation} \end{lem} \begin{proof} The upper estimate in \eqref{5.25} follows from \eqref{5.24}. To prove the lower estimate in \eqref{5.25}, we use the following obvious relations: \begin{align*} \|S_1(f)\|_p^p&=\int_{-\infty}^\infty \mu(x)^p\left|u(x)\int_{-\infty}^x\frac{v(t)}{\theta(t)}f(t)dt\right|^pdx\\ &\le \int_{-\infty}^\infty\mu(x)^p\left(u(x)\int_{-\infty}^x\frac{v(t)}{\theta(t)}|f(t)|dt \right)^pdx\\ &\le\int_{-\infty}^\infty\mu(x)^p\left[u(x)\int_{-\infty}^x\frac{v(t)}{\theta(t)}|f(t)|dt+v(x)\int_x^\infty\frac{u(t)}{\theta (t)}|f(t)|dt\right]^pdx\\ &=\int_{-\infty}^\infty\left|\mu(x)\int_{-\infty}^\infty\frac{G(x,t)}{\theta(t)}|f(t)| dt\right|^pdx=\|S(|f|)\|_p^p \le\|S\|_{p\to p}^p\cdot\|f\|_p^p. \end{align*} This implies that $\|S_1\|_{p\to p}\le\|S\|_{p\to p}.$ Similarly, we check that $\|S_2\|_{p\to p}\le\|S\|_{p\to p}.$ These inequalities imply the lower estimate in \eqref{5.25}. \end{proof} Let us now go over to the proof of the theorem. Since \eqref{2.1} holds, equation \eqref{2.4} has a FSS $\{u,v\}$ with the properties from \thmref{thm2.2}. Since the operator $S:L_p\to L_p $ is bounded, so are also the operators $S_i: L_p\to L_p,$ $i=1,2$ (see \eqref{5.25}). Then, by Theorems \thmref{thm2.7} and \thmref{thm2.8}, we obtain the inequalities \begin{align} \tilde M_p&\doe \sup_{x\in\mathbb R}\left(\int_{-\infty}^x\left(\frac{v(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}\cdot\left(\int_x^\infty(\mu(t)u(t)^pdt\right) ^{1/p}<\infty,\label{5.26}\\ M_p&\doe \sup_{x\in\mathbb R} \left(\int_{-\infty}^x(\mu(t)v(t)^pdt\right) ^{1/p} \cdot \left(\int^{ \infty}_x\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}<\infty.\label{5.27} \end{align} These inequalities imply that the function \begin{equation}\label{5.28} y(x)=(Gf)(x)=u(x)\int_{-\infty}^x v(t)f(t)dt+v(x)\int_x^\infty u(t)f(t)dt,\quad x\in\mathbb R \end{equation} is well-defined because the integrals in \eqref{5.28} converge: \begin{align*} \int_{-\infty}^x v(t)|f(t)|dt&\le\left(\int_{-\infty}^x\left(\frac{v(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}\cdot\|f\|_{p,\theta},\quad x\in\mathbb R, \\ \int^{\infty}_x u(t)|f(t)|dt&\le\left(\int_{-\infty}^x\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}\cdot\|f\|_{p,\theta},\quad x\in\mathbb R.\\ \end{align*} Further, one can check in a straightforward manner (see \thmref{thm2.2}) that the function $y(x),$ $x\in\mathbb R$ is a solution of \eqref{1.1}. In addition, \begin{align*} \|y\|_{p,\mu}&=\left[\int_{-\infty}^\infty\left(\mu(x)\left|\int_{-\infty}^\infty G(x,t)f(t)dt\right|\right)^pdx\right]^{1/p}\\ &=\left[\int_{-\infty}^\infty\left(\mu(x)\left|\int_{-\infty}^\infty\frac{G(x,t)}{\theta(t)}(\theta(t)f(t))dt\right|\right)^p dx\right]^{1/p}\\ &=\|S(\theta f)\|_p\le\|S\|_{p\to p}\cdot\|\theta f\|_p=\|S\|_{p\to p}\cdot\|f\|_{p,\theta}, \end{align*} i.e., \eqref{1.5} holds. It only remains to refer to \lemref{lem4.2}. \end{proof} \renewcommand{\qedsymbol}{} \begin{proof}[Proof of \thmref{thm4.3} for $p=1$] \textit{Necessity}. Let $[x_1,x_2]$ be an arbitrary finite segment, and let $f\in L_1$ be such that $\supp f=[x_1,x_2].$ Then (see \eqref{5.15}) $f\in L_{1,\theta}$ and therefore equation \eqref{1.1} with such a right-hand side has a unique solution $y\in L_{1,\mu}.$ By \lemref{lem5.1}, this solution is given by formula \eqref{2.13} and satisfies \eqref{1.5}. Let us introduce the operator $\tilde S:$ $$(\tilde Sg)(x)=\mu(x)\int_{x_1}^{x_2}\frac{G(x,t)}{\theta(t)}g(t)dt,\quad x\in[x_1,x_2],\quad g\in L_1(x_1,x_2)$$ and the function $g$ given on the sequence $[x_1,x_2]$ by the formula $$g(x)=\theta(x)f(x),\qquad x\in [x_1,x_2].$$ Then we have \begin{align*} \|\tilde Sg\|_{L_1(x_1,x_2)}&=\int_{x_1}^{x_2}\left|\mu(x)\int_{x_1}^{x_2}\frac{G(x,t)}{\theta(t)}g(t)dt\right|dx\\ &=\int_{x_1}^{x_2}\mu(x)\left|\int_{x_1}^{x_2}G(x,t)f(t)dt\right|dx=\int_{x_1}^{x_2}\mu(x)\left|\int_{-\infty}^\infty G(x,t)f(t)dt\right|dx\\ &=\int_{x_1}^{x_2}\mu(x)|y(x)|dx\le\int_{-\infty}^\infty \mu(x)|y(x)|dx=\|y\|_{1,\mu}\le c(1)\|f\|_{1,\theta}\\ &=c(1)\int_{-\infty}^\infty\theta(t)|f(t)|dt=c(1)\int_{x_1}^{x_2}|\theta(t)f(t)|dt=c(1)\|g\|_{L_1(x_1,x_2)}. \end{align*} Together with \eqref{2.22} and \eqref{2.9}, this implies that \begin{align*} &\|\tilde S\|_{L_1(x_1,x_2)\to L_1(x_1,x_2)}\le c(1)\qquad \Rightarrow\\ &\sup_{x\in[x_1,x_2]}\frac{1}{\theta(x)}\int_{x_1}^{x_2}\mu(t)G(x,t)dt=\|\tilde S\|_{L_1(x_1,x_2)\to L_1(x_1,x_2)}\le c(1). \end{align*} In the last inequality, $x_1$ and $x_2$ are arbitrary numbers. Hence $$\sup_{x\in\mathbb R}\frac{1}{\theta(x)}\int_{-\infty}^\infty \mu(t)G(x,t)dt\le c(1)<\infty.$$ But then by \thmref{thm2.9} we obtain that $\|S\|_{L_1\to L_2}\le c(1)<\infty,$ as required. \end{proof} \renewcommand{\qedsymbol}{\openbox} \begin{proof}[Proof of \thmref{thm4.3} for $p=1$] \textit{Sufficiency.} {}From \eqref{2.1} it follows that equation \eqref{2.4} has a FSS $\{u,v\}$ (see \thmref{thm2.2}), the Green function and the operator $S$ are defined (see \eqref{2.9} and \eqref{4.3}). Further, the operators $S_i,$ $i=1,2$ (see \eqref{4.4}, \eqref{4.5}) are bounded because so is the operator $S:L_1\to L_1$ (see \lemref{lem5.2}). Let now $f\in L_{1,\theta}$ and $g=\theta\cdot|f|.$ Then $0\le g\in L_1,$ $S_ig\in L_1,$ $i=1,2,$ and one has the inequalities \begin{equation}\label{5.29} 0\le (S_ig)(x)<\infty,\qquad \forall x\in\mathbb R,\qquad i=1,2. \end{equation} We will prove \eqref{5.29} for $i=1$ (the case $i=2$ is considered in a similar way). Assume to the contrary that there exists $x_1\in\mathbb R$ such that $(S_1g)(x_1)=\infty.$ Let $x_2>x_1.$ Then, since the functions $\mu$ and $u$ are continuous, we have \begin{align*} (S_1g)(x_2)&=\mu(x_2)u(x_2)\int_{-\infty}^{x_2}\frac{v(t)}{\theta(t)}g(t)dt\\ &\ge \frac{\mu(x_2)u(x_2)}{\mu(x_1)u(x_1)}\left[\mu(x_1)u(x_1)\int_{-\infty}^{x_1}\frac{v(t)}{\theta(t)}g(t)dt\right]= \frac{\mu(x_2)u(x_2)}{\mu(x_1)u(x_1)}(S_1g)(x_1)=\infty\\ \Rightarrow\qquad\qquad\qquad\\ \infty&>\|Sg\|_1=\int_{-\infty}^\infty\mu(x)u(x)\left|\int_{-\infty}^x\frac{v(t)}{\theta(t)}g(t)dt\right|dx\\ &\ge\int_{x_1}^\infty \mu(x)u(x)\left(\int_{-\infty}^x\frac{v(t)} {\theta(t)}g(t)dt\right)dx=\int_{x_1}^\infty(S_1g)(x)dx=\infty. \end{align*} We get a contradiction. Hence, inegualities \eqref{5.29} hold. From \eqref{5.29} and the definition of $g$ we obtain \begin{equation}\label{5.30} \int_{-\infty}^xv(t)|f(t)|dt<\infty,\qquad \int_x^\infty u(t)|f(t)|dt<\infty \qquad \forall x\in\mathbb R. \end{equation} For instance, \begin{align*} \int_{-\infty}^x v(t)|f(t)|dt&=\frac{1}{\mu(x)u(x)}\left[\mu(x)u(x)\int_{-\infty}^x \frac{v(t)}{\theta(t)}\cdot(\theta(t) |f(t)|)dt\right]\\ &=\frac{1}{\mu(x)u(x)}(S_1g)(x)<\infty\quad\Rightarrow\quad\eqref{5.30} \end{align*} Thus, if $f\in L_{1,\theta},$ then by \eqref{5.30} the following integrals converge: $$\int_{-\infty}^x v(t)f(t)dt,\qquad \int_x^\infty u(t)f(t)dt,\qquad x\in\mathbb R$$ and therefore, for $x\in\mathbb R,$ the function $$y(x)=(Gf)(x)=u(x)\int_{-\infty}^x v(t)f(t)dt+v(x)\int_x^\infty u(t)f(t)dt,\quad x\in\mathbb R$$ is well-defined. This immediately implies that $y(x)$ is a solution of \eqref{1.1}. In addition, \eqref{1.5} holds: \begin{align*} \|\mu y\|_1&=\int_{-\infty}^\infty \mu(x)\left|\int_{-\infty}^\infty \frac{G(x,t)}{\theta(t)}(\theta(t)f(t))dt\right|dx \le \int_{-\infty}^\infty \mu(x)\int_{-\infty}^\infty\frac{G(x,t)}{\theta(t)}|g(t)|dt\, d\\ &=\|Sg\|_1\le\|S\|_{1\to 1}\cdot \|g\|_1 =\|S\|_{1\to 1}\cdot\|f\|_{1,\theta}\quad \Rightarrow\quad\eqref{1.5}. \end{align*} It remains to note that by \lemref{lem4.2} this solution is unique in the class $L_{1,\mu}.$ \end{proof} \begin{proof}[Proof of \lemref{lem4.5}] {}From \eqref{2.2} we obtain the inequality $$2\le\sqrt 2 d(x)\int_{x-\sqrt 2 d(x)}^{x+\sqrt 2 d(x)}q(\xi)d\xi,\qquad x\in\mathbb R.$$ Together with the formula for $|d'(x)|$ (see the proof of \lemref{lem3.1}), this implies that \begin{align*} |d'(x)|&\le \frac{d(x)}{\sqrt 2}\left|\int_x^{x+\sqrt 2 d(x)}q(\xi)d\xi-\int_{x-\sqrt 2 d(x)}^x q(\xi)d\xi\right|\cdot\left(d(x)\int_{x-\sqrt 2 d(x)}^{x+\sqrt 2 d(x)}q(\xi)d\xi\right)^{-1} \\ &\le\frac{1}{2}d(x)\left|\int_x^{x+\sqrt 2 d(x)}q(\xi)d\xi-\int^x_{x-\sqrt 2 d(x)}\right|=\frac{\nu(x)}{2},\quad x\in\mathbb R\quad \Rightarrow \end{align*} \begin{equation} \lim_{|x|\to \infty}d'(x)=0.\label{5.31} \end{equation} Let us now go to \eqref{4.8}. Fix $\varepsilon\in(0,1/\sqrt 2)$ (see \eqref{4.9} regarding the case $\varepsilon\ge 1/\sqrt 2$). Then there exists $x_0=x_0(\varepsilon)\gg 1$ such that we have the inequality (see \eqref{5.31} \begin{equation}\label{5.32} |d'(x)|\le\varepsilon\qquad\text{if}\qquad |x|\ge x_0. \end{equation} It is easy to see that all possible cases of placing the numbers $t,x\in\mathbb R$ and the segments $(-\infty, x_0],$ \ $[-x_0, x_0]$ and $[x_0, \infty]$ can be put in the following table: \begin{equation}\vbox{\offinterlineskip \hrule \hrule \halign{&\vrule#& \stru #\tabskip=.09em\cr height4pt&\omit&&\omit&&\omit &\cr &\qquad \qquad$1.1$ \hfil && \qquad \quad$1.2$\quad&&\qquad \qquad$ 1.3$ \hfil &\cr &\quad $x\in(-\infty,-x_0]$\hfil&& \quad $x\in(-\infty,-x_0]$\hfil&& \quad $x\in(-\infty,-x_0]$\hfil &\cr &\quad $t\in(-\infty,-x_0]\hfil$&& \quad $t\in[-x_0,x_0]$\hfil&& \quad $t\in[x_0,\infty]$\hfil &\cr \noalign{\hrule} &\qquad \qquad$2.1$ \hfil && \qquad \quad$2.2$\quad&&\qquad \qquad$ 2.3$ \hfil &\cr &\quad $x\in[-x_0,x_0]$\hfil&& \quad $x\in[-x_0,x_0]$\hfil&& \quad $x\in[-x_0,x_0]$\hfil &\cr &\quad $t\in(-\infty,-x_0]\hfil$&& \quad $t\in[-x_0,x_0]$\hfil&& \quad $x\in[x_0,\infty)$\hfil &\cr \noalign{\hrule} &\qquad \qquad$3.1$ \hfil && \qquad \quad$3.2$\quad&&\qquad \qquad$ 3.3$ \hfil &\cr &\quad $x\in(x_0,\infty]$\hfil&& \quad $x\in(x_0,\infty)$\hfil&& \quad $x\in(x_0,\infty)$\hfil &\cr &\quad $t\in(-\infty,-x_0]\hfil$&& \quad $t\in[-x_0,x_0]$\hfil&& \quad $t\in[-x_0,\infty)$\hfil &\cr \noalign{\hrule} } \hrule} \raisetag{3\baselineskip} \label{5.33} \end{equation} \bigskip We check inequalities \eqref{4.8} separately in each case appearing in \eqref{5.33}. \newpage \noindent \textit{Cases 1.1 and 3.3}. Both cases are treated in the same way. Let us introduce the standing notation for the whole proof: \begin{gather*} m(\varepsilon)=\min_{t\in[-x_0,x_0]} d(t),\qquad M(\varepsilon)=\max_{t\in[-x_0,x_0]}d(t)\\ c(\varepsilon)=\max\left\{\frac{1}{m(\varepsilon)}, M(\varepsilon) \right\},\\ a=\min\{x,t),\qquad b=\max\{x,t\}. \end{gather*} Consider, say, Case 3.3. The following implications are obvious: \begin{gather*} -\varepsilon \le d'(\xi)\le\varepsilon\quad\text{for}\quad \xi\in[a,b]\Rightarrow -\frac{\varepsilon}{d(\xi)}\le \frac{d'(\varepsilon)}{d(\xi)}\le\frac{\varepsilon}{d(\xi)},\quad \xi\in[a,b]\\ \Rightarrow - \varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right| =-\varepsilon\int_a^b\frac{d\xi}{d(\xi)}\le \ln\frac{d(b)}{d(a)}\le\varepsilon\int_a^b\frac{d\xi}{d(\xi)}=\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\quad\Rightarrow\\ \exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right) \le \frac{d(b)}{d(a)},\qquad \frac{d(a)}{d(b)}\le \exp\left(\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right)\quad\Rightarrow \quad \eqref{4.8}. \end{gather*} \noindent \textit{Cases 1.2 and 2.1}. Both cases are treated in the same way. For instance, in Case 1.2 we have \begin{align*} \frac{d(t)}{d(x)}&=\frac{d(t)}{d(-x_0)}\cdot \frac{d(-x_0)}{d(x)}\le c(\varepsilon)^2\exp\left(\varepsilon\left|\int_x^{-x_0}\frac{d\xi}{d(\xi)}\right|\right)\\ &\le c(\varepsilon)^2\exp\left(\varepsilon\left|\int_x^{-x_0}\frac{d\xi}{d(\xi)}+\int_{-x_0}^t\frac{d\xi}{d(\xi)}\right|\right)= c(\varepsilon)^2\exp\left(\varepsilon\left|\int_x^t\frac{d\xi)}{d(\xi)}\right|\right); \end{align*} \begin{align*} \frac{d(t)}{d(x)}&=\frac{d(t)}{d(-x_0)}\cdot \frac{d(-x_0)}{d(x)}\ge c(\varepsilon)^{-2}\exp\left(-\varepsilon\left|\int_x^{-x_0}\frac{d\xi}{d(\xi)}\right|\right)\\ &\ge c(\varepsilon)^2\exp\left(-\varepsilon\left|\int_x^{-x_0}\frac{d\xi}{d(\xi)}+\int_{-x_0}^t\frac{d\xi}{d(\xi)}\right|\right)= c(\varepsilon)^2\exp\left(-\varepsilon\left|\int_x^t\frac{d\xi)}{d(\xi)}\right|\right)\quad\Rightarrow\quad\eqref{4.8}. \end{align*} \noindent \textit{Cases 1.3 and 3.1}. Both cases are treated in the same way. For instance, in Case 1.3 we have \begin{align*} \frac{d(t)}{d(x)}&=\frac{d(-x_0)}{d(x)}\cdot \frac{d(x_0)}{d(-x_0)}\cdot\frac{d(t)}{d(x_0)}\le \frac{M}{m}\exp\left(\varepsilon\left|\int_x^{-x_0}\frac{d\xi}{d(\xi)}\right|+\varepsilon \left|\int_{x_0}^t\frac{d\xi}{d(\xi)}\right| \right)\\ &\le c(\varepsilon)^2\exp\left[\varepsilon\left(\int_x^{-x_0}\frac{d\xi}{d(\xi)}+\int_{-x_0}^x\frac{d\xi}{d(\xi)} \right)+\int_{x_0}^t\frac{d\xi}{d(\xi)}\right]\\ &= c(\varepsilon)^2\exp\left(\varepsilon\left|\int_x^t\frac{d\xi)}{d(\xi)}\right|\right); \end{align*} \begin{align*} \frac{d(t)}{d(x)}&=\frac{d(-x_0)}{d(x)}\cdot \frac{d(x_0)}{d(-x_0)} \cdot\frac{d(t)}{d(x_0)}\ge \frac{m}{M}\exp\left(-\varepsilon\left|\int_x^{-x_0}\frac{d\xi}{d(\xi)}\right|-\varepsilon \left|\int_{x_0} ^t\frac{d\xi}{d\xi}\right|\right)\\ &\ge c(\varepsilon)^{-2}\exp\left[-\varepsilon\left(\int_x^{-x_0}\frac{d\xi}{d(\xi)}+\int_{-x_0}^{x_0}\frac{d\xi}{d(\xi)} +\int_{x_0}^t\frac{d\xi}{d(\xi)}\right)\right ]\ge c(\varepsilon)^{-2}\exp\left(-\varepsilon\left|\int_x^t\frac{d\xi)}{d(\xi)}\right|\right) . \end{align*} \noindent \textit{Case 2.2}. We have \begin{align*} \frac{d(t)}{d(x)}&\le \frac{M(\varepsilon)}{m(\varepsilon)}\le c(\varepsilon)^2\exp\left(\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right);\\ \frac{d(t)}{d(x)}&\ge \frac{m(\varepsilon)}{M(\varepsilon)}\ge c(\varepsilon)^{-2}\exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right).\\ \end{align*} \noindent \textit{Cases 2.3 and 3.2}. Both cases are treated in the same way. For instance, in Case 2.3 we have \begin{align*} \frac{d(t)}{d(x)}&=\frac{d(x_0)}{d(x)}\cdot\frac{d(t)}{d(x_0)}\le \frac{M(\varepsilon)}{m(\varepsilon)} \exp\left(\varepsilon\left|\int_{x_0}^t\frac{d\xi}{d(\xi)}\right|\right) \\ &\le c(\varepsilon)^2\exp\left[\varepsilon\left(\int_x^{x_0}\frac{d\xi}{d(\xi)}+\int_{x_0}^t\frac{d\xi}{d(\xi)}\right)\right] = c(\varepsilon)^2\exp\left(\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right); \end{align*} \begin{align*} \frac{d(t)}{d(x)}&=\frac{d(x_0)}{d(x)}\cdot\frac{d(t)}{d(x_0)}\ge \frac{m(\varepsilon)}{M(\varepsilon)} \exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right) \\ &\ge c(\varepsilon)^{-2}\exp\left[-\varepsilon\left(\int_x^{x_0}\frac{d\xi}{d(\xi)}+\int_{x_0}^t\frac{d\xi}{d(\xi)}\right)\right]=c (\varepsilon)^{-2}\exp\left(-\varepsilon\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right). \end{align*} \end{proof} \begin{proof}[Proof of \thmref{thm4.7} for $p\in(1,\infty)$] \textit{Necessity}. We need some auxiliary assertions. \begin{lem}\label{lem5.3} Let $p\in[1,\infty)$, $p'=p(p-1)^{-1}.$ Denote \begin{align} M_p(x)&=\left(\int_{-\infty}^x(\mu(t)v(t))^pdt\right)^{1/p}\cdot\left(\int_x^\infty\left(\frac{u(t)}{\theta(t)}\right)^{1/p'}dt \right),\quad x\in\mathbb R,\label{5.34}\\ \tilde M_p(x)&=\left(\int_{-\infty}^x\left(\frac{v(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}\cdot\left( \int_x^\infty(\mu(t)u(t)) ^pdt\right)^{1/p},\quad x\in\mathbb R.\label{5.35} \end{align} Then we have the equalities (see \eqref{2.8}): \begin{align} M_p(x)&=\left[\int_{-\infty}^x\left(\sqrt{\rho(t)}\mu(t)\right)^p\exp\left(-\frac{p}{2}\int_t^x\frac{d\xi}{\rho(\xi)}\right) dt\right]^{1/p}\nonumber\\ &\quad\cdot\left[\int_x^\infty\left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right)^{p'}\exp\left(-\frac{p'}{2}\int_x^t\frac{d\xi}{\rho(\xi)}\right) dt\right]^{1/p'},\quad x\in\mathbb R,\label{5.36} \end{align} \begin{align} \tilde M_p(x)&=\left[\int_{-\infty}^x\left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right)^{p'} \exp\left(-\frac{p'}{2}\int_t^x\frac{d\xi}{\rho(\xi)}\right) dt\right]^{1/p'} \nonumber\\ &\quad\cdot\left[\int_x^\infty\left(\mu(t)\sqrt{\rho(t)} \right)^{p}\exp\left(-\frac{p}{2}\int_x^t\frac{d\xi}{\rho(\xi)}\right) dt\right]^{1/p},\quad x\in\mathbb R.\label{5.37} \end{align} \end{lem} \renewcommand{\qedsymbol}{\openbox} \begin{proof} Equalities \eqref{5.36} and \eqref{5.37} are proved in the same way. Consider, say, \eqref{5.36}. This equality can be obtained by substituting formulas \eqref{2.10} in \eqref{5.34}: \begin{align*} M_p(x)&=\left[\int_{-\infty}^x\left(\mu(t)\sqrt{\rho(t)}\right)^p\exp\left(\frac{p}{2}\int_{x_0}^t\frac{d\xi}{\rho(\xi)} \right) dt\right]^{1/p}\\ &\quad\cdot\left[\int_x^\infty\left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right) ^{p'} \exp\left(-\frac{p'}{2}\int_{x_0}^t\frac{d\xi}{\rho(\xi)}\right) dt\right]^{1/p'} \\ &=\left[\int_{-\infty}^x\left(\mu(t)\sqrt{\rho(t)}\right)^p\exp\left(-\frac{p}{2}\int^{x}_t\frac{d\xi}{\rho(\xi)} \right) \cdot\exp\left(\frac{p}{2}\int_{x_0}^x\frac{d\xi}{\rho(\xi)}\right)dt\right]^{1/p}\\ &\quad\cdot\left[\int_x^\infty\left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right) ^{p'} \exp\left(-\frac{p'}{2}\int_x^t\frac{d\xi}{\rho(\xi)}\right)\cdot\exp\left(-\frac{p'}{2}\int_{x_0}^x\frac{d\xi}{\rho(\xi)} \right) dt\right]^{1/p'}\\ &=\left[\int_{-\infty}^x\left(\mu(t)\sqrt{\rho(t)}\right)^p\exp\left(-\frac{p}{2}\int_{t}^x\frac{d\xi}{\rho(\xi)} \right) dt\right]^{1/p}\\ &\quad\cdot\left[ \int_x^\infty \left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right) ^{p'}\exp\left(-\frac{p'}{2} \int_{x}^t\frac{d\xi}{\rho(\xi)} \right)dt \right]^{1/p'}. \end{align*} \end{proof} Let us introduce some more notation: \begin{equation}\label{5.38} \varphi(x,t)=\begin{cases} \displaystyle\frac{\mu(x)v(x)}{\mu(t)v(t)},&\quad\text{if}\quad x\le t\\ \\ \displaystyle\frac{\mu(t)v(t)}{\mu(x)v(x)},&\quad\text{if}\quad x\ge t\end{cases},\qquad \psi(x,t)=\begin{cases} \displaystyle\frac{\theta(x)u(t)}{\theta(t)u(x)},&\quad\text{if}\quad x\le t\\ \\ \displaystyle\frac{\theta(t)u(x)}{\theta(x)u(t)},&\quad\text{if}\quad x\ge t\end{cases}. \end{equation} \begin{lem}\label{lem5.4} Under the hypotheses of the theorem, for a given $\varepsilon>0$ and for all $t,x\in\mathbb R,$ we have the inequality \begin{equation}\label{5.39} \max\{\varphi(x,t);\psi(x,t)\}\le c(\varepsilon)\exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right)\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right|\right). \end{equation} \end{lem} \begin{proof} We will check inequality \eqref{5.39} for the function $\varphi$ (for the function $\psi$ the proof of \eqref{5.39} is similar). Below we use \eqref{2.10}, \eqref{2.12} and \eqref{4.10}. Let $x\ge t.$ Then \begin{align*} \frac{\mu(t)}{\mu(x)}\cdot\frac{v(t)}{v(x)}&=\frac{\mu(t)}{\mu(x)}\cdot\sqrt{\frac{\rho(t)}{\rho(x)}}\exp\left(-\frac{1}{2}\int_t^x\frac{d\xi}{\rho(\xi)} \right)\le c\frac{\mu(t)}{\mu(x)}\cdot\sqrt{\frac{d(t)}{d(x)}}\exp\left(-\frac{1}{2} \int_t^x\frac{d\xi}{\rho(\xi)}\right)\\ &\le c(\varepsilon)\exp\left(\varepsilon\int_t^x\frac{d\xi}{d(\xi)}-\frac{1}{2}\int_t^x\frac{d\xi}{\rho(\xi)}\right)\le c(\varepsilon)\exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right)\int_t^x\frac{d\xi}{\rho(\xi)}\right)\\ &=c(\varepsilon)\exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right)\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right|\right); \end{align*} Similarly, for $x\le t,$ we have: \begin{align*} \frac{\mu(x)}{\mu(t)}\cdot\frac{v(x)}{v(t)}&=\frac{\mu(x)}{\mu(t)}\sqrt{\frac{\rho(x)}{\rho(t)}}\exp\left(-\frac{1}{2} \int_x^t\frac{d\xi}{\rho(\xi)}\right)\le c\frac{\mu(x)}{\mu(t)}\sqrt{\frac{d(x)}{d(t)}}\exp\left(-\frac{1}{2}\int_x^t\frac{d\xi}{\rho(\xi)}\right)\\ &\le c(\varepsilon)\exp\left(\varepsilon\int_t^x\frac{d\xi}{d(\xi)}-\frac{1}{2} \int_t^x\frac{d\xi}{\rho(\xi)}\right)\le c(\varepsilon)\exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right) \int_t^x\frac{d\xi}{\rho(\xi)} \right)\\ &=c(\varepsilon)\exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right)\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right|\right). \end{align*} \end{proof} \begin{lem}\label{lem5.5} Under conditions \eqref{1.1} and \eqref{2.1}, we have the inequality \begin{equation}\label{5.40} \int_{x-d(x)}^{x+d(x)}\frac{d\xi}{d(\xi)}\le 8,\quad\forall\ x\in\mathbb R. \end{equation} \end{lem} \begin{proof} Estimate \eqref{5.40} follows from \eqref{3.2}: $$\int_{x-d(x)}^{x+d(x)}\frac{d\xi}{d(\xi)}=\int_{x-d(x)}^{x+d(x)}\frac{d(x)}{d(\xi)}\cdot\frac{d\xi}{d(x)}\le\int_{x-d(x)} ^{x+d(x)}4\frac{d\xi}{d(x)}=8. $$ \end{proof} \begin{lem}\label{lem5.6} Under the hypotheses of the theorem, we have the inequalities \begin{equation}\label{5.41} c^{-1}\le\frac{\mu(t)}{\mu(x)},\quad \frac{\theta(t)}{\theta(x)}\le c;\quad\text{if}\quad t\in[x-d(x),x+d(x)],\quad x\in\mathbb R. \end{equation} \end{lem} \begin{proof} We will only check inequalities \eqref{5.41} for the function $\mu$ (the proof of \eqref{5.41} for the function $\theta$ is similar). In \eqref{4.10}, set $\varepsilon=\frac{1}{2}.$ Now for $|t-x|\le d(x),$ $x\in\mathbb R$, we use \eqref{3.2}, \eqref{4.10} and \eqref{5.40}: \begin{align*} \frac{\mu(t)}{\mu(x)}&\le c\sqrt{\frac{d(x)}{d(t)}}\exp\left(\frac{1}{2}\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right) \le c\exp\left(\frac{1}{2}\int_{x-d(x)}^{x+d(x)}\frac{d\xi}{d(\xi)}\right)\le c<\infty,\\ \frac{\mu(t)}{\mu(x)}&\ge c^{-1}\sqrt{\frac{d(x)}{d(t)}}\exp\left(-\frac{1}{2}\left|\int_x^t\frac{d\xi}{d(\xi)}\right|\right) \ge c^{-1}\exp\left(-\frac{1}{2}\int_{x-d(x)}^{x+d(x)}\frac{d\xi}{d(\xi)}\right)\le c^{-1}>0. \end{align*} \end{proof} Let us now go over to the theorem. Since condition \eqref{2.1} holds, by \thmref{thm2.2}, a FSS $\{u,v\}$ of equation \eqref{2.4} is defined, and thus the operator $S$ (see \eqref{4.3}) is also defined. Since the pair $\{L_{p,\mu};L_{p,\theta}\}$ is admissible for \eqref{1.1}, by \thmref{thm4.3} the operator $S:L_p\to L_p,$ $p\in[1,\infty)$ is bounded. Then so are the operators $S_i: L_p\to L_p$, $i=1,2$ (see \eqref{5.25}). Let $p\in(1,\infty).$ Consider, say, the operator $S_2: L_p\to L_p.$ Since it is bounded, we have $M_p<\infty$ by \thmref{thm2.7} (see \eqref{5.27} and \eqref{5.34}. Below we use this fact together with \lemref{lem2.1}, \eqref{2.12}, \eqref{5.40} and \eqref{5.41}: \begin{align*} \infty&>M_p=\sup_{x\in\mathbb R} M_p(x)=\sup_{x\in\mathbb R}\left(\int_{-\infty}^x(\mu(t)v(t))^pdt\right)^{1/p}\left( \int_x^\infty\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt\right)^{1/p'}\\ &=\sup_{x\in\mathbb R}\left[\int_{-\infty}^x\left(\sqrt{\rho(t)\mu(t)}\right)^p\exp\left(-\frac{p}{2}\int_t^x\frac{d\xi} {\rho(\xi)}\right)dt\right]^{1/p}\\ &\quad \cdot\left[\int_x^\infty\left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right)^{p'} \exp\left(- \frac{p'}{2}\int_x^t\frac{d\xi}{\rho(\xi)} \right) dt\right]^{1/p'}\\ &\ge\sup_{x\in\mathbb R}\left[\int_{x-d(x)}^x\left(\sqrt{\rho(t)}\mu(t)\right)^p\exp\left(-\frac{p}{2}\int_t^x\frac{d\xi} {\rho(\xi)}\right)dt\right]^{1/p}\\ &\quad \cdot\left[\int_x^{x+d(x)}\left(\frac{\sqrt{\rho(t)}}{\theta(t)}\right)^{p'}\exp\left(-\frac{p'}{2} \int_x^t\frac{d\xi}{\rho(\xi)}\right)dt\right]^{1/p'}\\ &\ge c^{-1}\sup_{x\in\mathbb R}\left[\int_{x-d(x)}^x\left(\sqrt{d(t)}\mu(t)\right)^p\exp\left(-\sqrt 2p\int_t^x\frac{d\xi}{d(\xi)}\right)dt\right]^{1/p}\\ &\quad \cdot\left[\int_x^{x+d(x)}\left(\frac{\sqrt{d(t)}}{\theta(t)}\right)^{p'} \exp\left(-\sqrt 2p'\int_x^t\frac{d\xi}{d(\xi)}\right)dt\right]^{1/p'}\\ &\ge c^{-1}\sup_{x\in\mathbb R}\left[\int_{x-d(x)}^x\left(\sqrt{d(t)}\mu(t)\right)^p\exp\left(-\sqrt{2p}\int_{x-d(x)}^{x+d(x)}\frac{d\xi}{d(\xi)}\right) dt\right]^{1/p}\\ &\quad \cdot\left[\int_x^{x+d(x)}\left(\frac{\sqrt{d(t)}}{\theta(t)}\right)^{p'}\exp\left(-\sqrt2p'\int_{x-d(x)}^{x+d(x)} \frac{d\xi}{d(\xi)}\right)dt\right]^{1/p'}\\ &\ge c^{-1}\sup_{x\in\mathbb R}\frac{\mu(x)}{\theta(x)}d^2(x)=c^{_1} m(q,\mu,\theta), \end{align*} as required. Let $p=1.$ Since the operator $S:L_1\to L_1$ is bounded (\thmref{thm4.3}), so are the operators $S_i:L_1\to L_1, $ $i=1,2$ (see \lemref{lem5.2}). Let, say, $i=2.$ Below we use \thmref{thm2.9}, \eqref{5.22}, \lemref{lem2.1}, \eqref{2.10}, \eqref{2.12}, \eqref{5.40}, \eqref{5.41} and \eqref{3.2}: \begin{align*} \infty&>\|S_2\|_{1\to1}=\sup_{x\in\mathbb R}\frac{u(x)}{\theta(x)}\int_{-\infty}^x\mu(t)v(t)dt\ge\sup_{x\in\mathbb R}\frac{u(x)}{\theta(x)}\int_{x-d(x)}^x\mu(t)v(t)dt\\ &=\sup_{x\in\mathbb R}\frac{\sqrt{\rho(x)}}{\theta(x)}\int_{x-d(x)}^x\mu(t)]\sqrt{\rho(t)}\exp\left(-\frac{1}{2}\int_t^x\frac{d\xi}{\rho(\xi)}\right)\\ &\ge c^{-1}\sup_{x\in\mathbb R}\frac{\sqrt{d(x)}}{\theta(x)}\int_{x-d(x)}^x\mu(t)\sqrt{d(t)}\exp\left(-\sqrt2\int_t^x\frac{d\xi}{d(\xi)}\right)dt\\ &\ge c^{-1}\sup_{x\in\mathbb R}\frac{\sqrt{d(x)}}{\theta(x)}\int_{x-d(x)}^x\mu(t)\sqrt{d(t)}\exp\left(-\sqrt2\int_{x-d(x)}^{x+d(x)}\frac{d\xi}{d(\xi)}\right)dt\\ &\ge c^{-1}\sup_{x\in\mathbb R}\frac{\sqrt{d(x)}}{\theta(x)}\int_{x-d(x)}^x\mu(t)\sqrt{d(t)}dt\ge c^{-1}\sup_{x\in\mathbb R}\frac{\mu(x)}{\theta(x)}d^2(x)=c^{-1}m(q,\mu,\theta). \end{align*} \renewcommand{\qedsymbol}{} \begin{proof}[Proof of \thmref{thm4.7}] \textit{Sufficiency.} It is enough to show that the operators $S_i: L_p\to L_p,$ $p\in[1,\infty),$ $i=1,2,$ are bounded. Indeed, then so is the operator $S:L_p\to L_p,$ $p\in[1,\infty)$ (see \eqref{5.25}), and then by \thmref{thm4.3} the pair $\{L_{p,\mu};\, L_{p,\theta}$ is admissible for \eqref{1.1}. Both operators $S_i,$ $i=1,2,$ are treated in the same way, and therefore below we only consider the operator $S_2$ (see \eqref{4.5}, \eqref{5.22}). Below, when estimating $\|S_2\|_{p\to p},$ $p\in(1,\infty)$, we use \thmref{thm2.7}, \eqref{5.22}, \eqref{5.34}, \eqref{5.38}, \eqref{5.39}, \eqref{4.10} for $\varepsilon=1/4\sqrt2,$ \eqref{2.12} and \eqref{4.11}: \begin{align*} &\|S_2\|_{p\to p} \le c(p)\sup_{x\in\mathbb R}\left[\int_{-\infty}^x(\mu(t)v(t))^pdt\right]^{1/p}\cdot\left[\int_x^\infty\left(\frac{u(t)}{\theta(t)}\right)^{p'}dt\right] ^{1/p'}\\ &=c(\varepsilon)\sup_{x\in\mathbb R}(\mu(x)v(x))^{1/p'}\left[\int_{-\infty}^x\left(\frac{\mu(t)v(t)}{\mu(x)v(x)}\right)^{p-1}(\mu(t)v(t))dt\right]^{1/p}\\ &\quad\cdot\left(\frac{u(x)}{\theta(x)}\right)^{1/p}\left[\int_x^\infty\left(\frac{u(t)}{\theta(t)}\cdot \frac{\theta(x)}{u(x)}\right)^{p'-1}\left(\frac{u(t)}{\theta(t)}\right)dt\right]^{1/p'}\\ &=c(\varepsilon)\sup_{x\in\mathbb R}\left[\frac{u(x)}{\theta(x)}\int_{-\infty}^x\varphi(x,t)^{p-1}(\mu(t)v(t))dt\right]^{1/p}\\ &\quad \cdot\left[\mu(x)v(x)\int_x^\infty\psi(x,t)^{p'-1}\left(\frac{u(t)}{\theta(t)}\right) dt\right]^{1/p'}\\ &\le c(\varepsilon)\sup_{x\in\mathbb R}\left[\int_{-\infty}^x\left(\frac{u(x)}{\theta(x)}\cdot\frac{\theta(t)}{u(t)}\right)\cdot\varphi(x,t)^{p-1}\frac{ \mu(t)}{\theta(t)}\rho(t)dt\right]^{1/p}\\ &\quad \cdot\left[\int_x^\infty\left(\frac{\mu(x)v(x)}{\mu(t)v(t)}\right)\cdot\psi(x,t)^{p'-1}\cdot\frac{\mu(t)}{\theta(t)}\rho(t)dt \right]^{1/p'}\\ &=c(\varepsilon)\sup_{x\in\mathbb R}\left[\int_{-\infty}^x\psi(x,t)\cdot\varphi(x,t)^{p-1}\frac{\mu(t)}{\theta(t)}\rho(t)dt\right]^{1/p}\\ &\quad\cdot \left[\int_x^\infty\varphi(x,t)\psi(x,t)^{p'-1}\frac{\mu(t)}{\theta(t)}\rho(t)dt\right]^{1/p'}\\ &\le c(\varepsilon)\sup_{x\in\mathbb R} \left[ \int_{-\infty}^x\left(\frac{\mu(t)}{\theta(t)}d^2(t)\right)\cdot\left(\frac{\rho(t)}{d(t)} \right)^2 \cdot\frac{1}{\rho(t)} \exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right) p\int_t^x\frac{d\xi}{\rho(\xi)}\right)dt \right]^{1/p}\\ &\quad\cdot\left[\int_x^\infty\left(\frac{\mu(t)}{\theta(t)}d^2(t)\right)\cdot\left(\frac{\rho(t)}{d(t)}\right)^2 \frac{1}{\rho(t)} \exp\left(\left(\sqrt2\varepsilon-\frac{1}{2}\right){p'}\int_x^t\frac{d\xi}{\rho(\xi)} \right)dt\right]^{1/p'}\\ &\le c(\varepsilon)m(q,\mu,\theta)\sup_{x\in\mathbb R}\left[\int_{-\infty}^x\frac{1}{\rho(t)}\exp\left(-\frac{p}{4}\int_t^x \frac{d\xi}{\rho(\xi)}\right)dt\right]^{1/p}\\ &\quad\cdot\left[\int_x^\infty\frac{1}{\rho(t)}\exp\left(-\frac{p'}{4}\int_x^t\frac{d\xi}{\rho(\xi)}\right)dt\right]^{1/p'} \le cm(q,\mu,\theta)<\infty. \end{align*} Consider the case $p=1.$ Below, when estimating $\|S\|_{1\to1}, $ we use \eqref{2.21}, \eqref{4.3}, \eqref{2.11} and \eqref{4.10} for $\varepsilon=1/4\sqrt2$, and \eqref{2.12}: \begin{align*} \|S\|_{1\to1}&=\sup_{x\in\mathbb R}\frac{1}{\theta(x)}\int_{-\infty}^\infty\mu(t)G(x,t)dt=\sup_{x\in\mathbb R}\frac{\sqrt {\rho(x)}}{\theta(x)}\int_{-\infty}^\infty\mu(t)\sqrt{\rho(t)}\exp\left(-\frac{1}{2}\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right| \right)dt\\ &=\sup_{x\in\mathbb R}\int_{-\infty}^\infty\left(\frac{\mu(t)}{\theta(t)}d^2(t)\right)\cdot\left(\frac{\rho(t)} {d(t)}\right)^2 \frac{\theta(t)\sqrt{d(x)}} {\theta(x)\sqrt{d(t)}}\cdot \sqrt{\frac{ \rho(x)}{d(x) } \cdot\frac{d(t)}{\rho(t)}}\cdot\frac{1}{\rho(t)} \exp\left(-\frac{1}{2}\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right| \right)dt\\ &\le cm(q,\mu,\theta)\int_{-\infty}^\infty\frac{1}{\rho(t)}\exp\left(\varepsilon \left|\int_x^t\frac{d\xi}{d(\xi)}\right|-\frac{1}{2}\left| \int_x^t\frac{d\xi}{\rho(\xi)}\right|\right)dt\\ &\le cm(q,\mu,\theta)\int_{-\infty}^\infty\frac{1}{\rho(t)}\exp\left(\left(\sqrt 2\varepsilon-\frac{1}{2}\right)\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right|\right)dt\\ &=cm(q,\mu,\theta)\int_{-\infty}^\infty\frac{1}{\rho(t)} \exp\left(-\frac{1}{4}\left|\int_x^t\frac{d\xi}{\rho(\xi)}\right|\right)dt= cm(q,r,\mu)<\infty. \end{align*} Thus the operator $S:L_p\to L_p,$ $p\in[1,\infty)$ is bounded, and it remains to refer to \thmref{thm4.3}. \end{proof} \end{proof} \renewcommand{\qedsymbol}{\openbox} \begin{proof}[Proof of \lemref{lem4.8}] Fix $\varepsilon>0$ and choose $x_0=x_0(\varepsilon)>>1$ in order to have the inequalities \begin{equation}\label{5.42} -\frac{\varepsilon}{3}\le\frac{\mu'(\xi)}{\mu(\xi)}d(\xi);\quad d'(\xi)\le\frac{\varepsilon}{3}\quad\text{for all}\quad |\xi|\ge x_0. \end{equation} {}From \eqref{5.42}, one can easily deduce the estimates \begin{equation}\label{5.43} -\frac{2}{3}\le\frac{(\mu(\xi)d(\xi))'}{\mu(\xi)d(\xi)}\le \frac{2\varepsilon}{3}\cdot \frac{1}{d(\xi)}\quad\text{for all}\quad |\xi|\ge x_0. \end{equation} Let, say, $t\ge x\ge x_0.$ Then from \eqref{5.43}, we obtain \begin{equation}\label{5.44} \exp\left(-\frac{2\varepsilon}{3}\int_x^t\frac{d\xi}{d(\xi)}\right)\le\frac{\mu(t)d(t)}{\mu(x)d(x)}\le\exp\left(\frac{2\varepsilon} {3}\int_x^t\frac{d\xi}{d(\xi)}\right),\quad t\ge x\ge x_0. \end{equation} Let us write \eqref{5.44} in a different way: $$\sqrt{\frac{d(x)}{d(t)}}\exp\left(-\frac{2\varepsilon}{\varepsilon}\int_x^t\frac{d\xi}{d(\xi)}\right)\le\frac{\mu(t)}{\mu(x)} \sqrt{\frac{d(t)}{d(x)}}\le\sqrt{\frac{d(x)}{d(t)}} \exp\left(\frac{2\varepsilon}{3}\int_x^t\frac{d\xi}{d(\xi)}\right),\quad t\ge x\ge x_0.$$ We now combine the latter estimates with inequalities \eqref{4.8} written for $\frac{2\varepsilon}{3}$ instead of $\varepsilon:$ $$ c\left(\frac{2\varepsilon}{3}\right)^{-1/2}\exp\left(-\frac{2\varepsilon}{3}\int_x^t\frac{d\xi}{d(\xi)}\right)\le\sqrt{ \frac{d(x)}{d(t)}} \le c\left(\frac{2\varepsilon}{3}\right)^{1/2}\exp\left(\frac{\varepsilon}{3}\int_x^t\frac{d\xi}{d(\xi)}\right). $$ We easily obtain that for $t\ge x\ge x_0$ we have the inequalities $$c\left(\frac{2}{3}\varepsilon\right)^{-1/2}\exp\left(-\varepsilon\int_x^t\frac{d\xi}{d(\xi)}\right)\le\frac{\mu(t)}{\mu(x)} \sqrt{\frac{d(t)}{d(x)}}\le c\left(\frac{2\varepsilon}{3}\right)^{1/2}\exp\left(\varepsilon\int_x^t\frac{d\xi}{d(\xi)}\right),$$ as required. The cases $x\ge t\ge x_0$ and the cases $t\le x\le-x_0,$ $x\le t\le-x_0$ are considered in a similar way. We then continue the proof as in \lemref{lem4.5}, with obvious modifications, similar to those presented above. \end{proof} \section{Example} In this final section, we consider equation \eqref{1.1} with \begin{equation}\label{6.1} q(x)=\frac{1}{\sqrt{1+x^2}}+\frac{\cos(e^{|x|})}{\sqrt{1+x^2}},\qquad x\in\mathbb R. \end{equation} Using the results obtained above, we show that the following assertions hold: \begin{enumerate} \item[A)] Equation \eqref{1.1} in the case of \eqref{6.1} is not correctly solvable in $L_p,$ for any $p\in[1,\infty);$ \item[B)] For equation \eqref{1.1} in the case of \eqref{6.1}, for any $p\in[1,\infty)$, the following pair of spaces $\{L_{p,\mu};L_{p,\theta}\}$ is admissible, where \begin{equation}\label{6.2} \mu(x)=\frac{1}{\sqrt{1+x^2}\ln(2+x^2)},\qquad \theta(x)=\frac{1}{ \ln(2+x^2)},\qquad x\in\mathbb R. \end{equation} \end{enumerate} \begin{remark*} Below we present an algorithm for the study of \eqref{1.1} for a given pair of spaces (cases \eqref{6.1} and $\{L_p,L_p\}$ and $\{L_{p,\mu};L_{p,\theta}\}$ in the case of \eqref{6.2}). We do not consider the question of the description of all pairs of spaces admissible for \eqref{1.1} in the case of \eqref{6.1}. \end{remark*} For the reader's convenience, we enumerate the main steps of the proof of assertions A) and B). Note that since the functions in \eqref{6.1} and \eqref{6.2} are even, all proofs are only given for $x\in[0,\infty)$ or for $x\in[x_0,\infty),$ $x_0\gg1$. \noindent 1) \textit{Checking condition \eqref{2.1}}. Let us check that in the case of \eqref{6.1} condition \eqref{2.1} holds. Assume to the contrary that there is $x_0\in\mathbb R $ such that \begin{equation}\label{6.3} \int_{x_0}^\infty q(t)dt=0. \end{equation} The function $q$ in \eqref{6.1} is continuous and non-negative. Therefore, from \eqref{6.3} it follows that $q(t)\equiv0$ for $t\in[x_0,\infty)$ which is obviously false. This contradiction implies \eqref{2.1}. \noindent 2) \textit{Existence of the function $d(x), $ $x\in\mathbb R$, and its estimates}. From 1) and \lemref{lem2.1}, it follows that the function $d(x)$ is defined for all $x\in\mathbb R.$ To obtain its estimates, we use \thmref{thm3.7}. Denote (see \eqref{3.7} and \eqref{3.8}) \begin{gather} q_1(x)=\frac{1}{\sqrt{1+x^2}};\qquad q_2(x)=\frac{\cos(e^{|x|})}{\sqrt{1+x^2}},\qquad x\in\mathbb R;\label{6.4}\\ A(x)=\left[0,2\sqrt[4]{1+x^2}\right];\qquad \omega(x)=\left[x-2\sqrt[4]{1+x^2},x+2\sqrt[4]{1+x^2}\right],\qquad x\in\mathbb R.\label{6.5} \end{gather} Let us check \eqref{3.11} for the function $\varkappa_1$ (see \eqref{3.9}): \begin{align} \varkappa_1(x)&=\frac{1}{q_1(x)^{3/2}}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}q_1''(\xi)d\xi\right| =(1+x^2)^{3/4}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}\left(\frac{1}{\sqrt{1+\xi^2}}\right)''d\xi\right|\nonumber\\ &=(1+x^2)^{3/4}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}\frac{1-2\xi^2}{1+\xi^2}\cdot \frac{d\xi}{(1+\xi^2)^{3/2}}\right|.\label{6.6} \end{align} Note the obvious inequalities \begin{equation}\label{6.7} \left|\frac{1-2\xi^2}{1+\xi^2}\right|\le\frac{1+2\xi^2}{1+\xi^2}\le 2,\qquad \xi\in\mathbb R. \end{equation} In addition, for $\xi\in A(x),$ $x\gg1,$ we have \begin{align} \frac{1+\xi^2}{1+x^2}&\le 1+\frac{|\xi-x|\, |\xi+x|}{1+x^2}\le 1+c\frac{x^{3/2}}{1+x^2}\le 2;\label{6.8}\\ \frac{1+\xi^2}{1+x^2}&\ge 1-\frac{|\xi-x|\, |\xi+x|}{1+x^2}\ge 1 -c\frac{x^{3/2}}{1+x^2}\ge \frac{1}{2}.\label{6.9} \end{align} {}From \eqref{6.6}, \eqref{6.7}, \eqref{6.8} and \eqref{6.9}, it now follows that \begin{align*} \varkappa_1(x)&\le (1+x^2)^{3/4}\sup_{t\in A(x)}\left[\int_{x-t}^{x+t}\left|\frac{1-2\xi^2}{1+\xi^2}\right|\frac{1}{(1+x^2)^{3/2}}\left(\frac{1+x^2}{1+\xi^2}\right)^{3/2}d\xi\right]\\ &\le c\frac{(1+x^2)^{3/4}}{(1+x^2)^{3/2}}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}1d\xi\right|=\frac{c}{\sqrt{1+x^2}}\to 0,\quad x\to\infty. \end{align*} Let us now check \eqref{3.11} for $\varkappa_2(x)$, $x\gg1.$ First show that for $x\gg1$ we have the inequality (see \eqref{6.5}): \begin{equation}\label{6.10} \sup_{[\alpha,\beta]\subseteq\omega(x)}\left|\int_\alpha^\beta\frac{\cos e^t}{\sqrt{1+t^2}}\right|\le c\frac{e^{-x/2}}{\sqrt{1+x^2}},\qquad x\gg 1. \end{equation} We need the following simple assertions, given without proof: \begin{enumerate} \item [a)] $x-\sqrt[4]{1+x^2}\to\infty$ as $x\to\infty;$ \item[b)] the function $\varphi(\xi)$ where $$\varphi(\xi)=\frac{e^{-\xi}}{\sqrt{1+\xi^2}},\qquad\xi\in\mathbb R$$ is monotone decreasing for all $\xi\in\mathbb R.$ \end{enumerate} Let $t$ be any point in the interval $(\alpha,\beta).$ Below we use assertions a), b) and the second mean theorem (see \cite{15}): \begin{align} \sup_{[\alpha,\beta]\subseteq \omega(x)}&\left|\int_\alpha^\beta\frac{\cos e^\xi}{\sqrt{1+\xi^2}}d\xi\right| =\sup_{[\alpha,\beta]\subseteq\omega(x)}\left|\int_\alpha^\beta\frac{e^{-\xi}}{\sqrt{1+\xi^2}} (e^\xi cos e^\xi)d\xi\right|\nonumber\\ &=\sup_{[\alpha,\beta]\subseteq\omega(x)}\frac{e^{-\alpha}}{\sqrt{1+\alpha^2}}\left|\int_\alpha^t e^\xi\cos e^\xi d\xi\right|\le c\left.\frac{c^{-\xi}}{\sqrt{1+\xi^2}}\right|_{\xi=x-2\sqrt[4]{1+x^2}}\le c\frac{e^{-x/2}}{\sqrt{1+x^2}}.\label{6.12} \end{align} Now, from \eqref{6.12} for $x\gg1$ we obtain \begin{align*} \varkappa_2(x)&=\frac{1}{\sqrt{q_1(x)}}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}q_2(\xi)d\xi\right|=\sqrt[4]{1+x^2}\sup_{t\in A(x)}\left|\int_{x-t}^{x+t}\frac{\cos e^\xi}{\sqrt{1+\xi^2}}d\xi\right|\\ &\le \sqrt[4]{1+x^2}\sup_ {[\alpha,\beta]\subseteq \omega(x)}\left|\int_\alpha^\beta\frac{\cos e^\xi}{\sqrt{1+\xi^2}} d\xi\right|\le c \frac{e^{-x/2}}{\sqrt{1+x^2}}\quad\Rightarrow \quad\eqref{3.11}. \end{align*} Since \eqref{3.11} is proven, by \thmref{thm3.7} we obtain \begin{gather} d(x)=\sqrt[4]{1+x^2}(1+\varepsilon(x)),\qquad |\varepsilon(x)|\le 2(\varkappa_1(x)+\varkappa_2(x)),\quad |x|\gg 1,\label{6.13}\\ c^{-1}\sqrt[4]{1+x^2}\le d(x)\le c \sqrt[4]{1+x^2},\qquad x\in\mathbb R.\label{6.14} \end{gather} \noindent 3) \textit{Proof of assertion A)}. {}From \eqref{6.14}, it follows that $d_0=\infty$ (see \eqref{2.3} and \eqref{2.14}). It remains to refer to \thmref{thm2.6}. \hfil\quad $\square$ Let us now go to assertion B). \noindent 4) \textit{Checking the inclusion $q\in H.$} To prove \eqref{4.6}, we need estimates of $\tau_1(x)$ and $\tau_2(x)$ for $x\gg 1$ where (see \eqref{6.1} and \eqref{6.4}) \begin{align} \tau_1(x)&=\left|\int_0^{\sqrt 2d(x)}(q_1(x+t)-q_1(x-t))dt\right|;\label{6.15}\\ \tau_2(x)&=\left|\int_0^{\sqrt 2d(x)}(q_2(x+t)-q_2(x-t))dt\right|.\label{6.16} \end{align} To estimate $\tau_1(x)$, we use below \eqref{6.7}, \eqref{6.8}, \eqref{6.9} and \eqref{6.13}: \begin{align} \tau_1(x)&=\left|\int_0^{\sqrt 2d(x)}\left(\int_{x-t}^{x+t}q_1'(\xi)d\xi\right)dt\right|\le\sqrt 2d(x)\sup_{|t|\le\sqrt 2d(x)} \left|\int_{x-\xi}^{x+\xi}q_1'(t)dt\right|\nonumber\\ &\le c\sqrt[4]{1+x^2}\sup_{|\xi|\le 2\sqrt[t]{1+t^2}}\left|\int_{x-\xi}^{x+\xi}\frac{t}{\sqrt{1+t^2}}\cdot\frac{1+x^2}{1+t^2}\cdot\frac{dt}{1+x^2}\right|\nonumber\\ &\le c\frac{\sqrt[4]{1+x^2}}{1+x^2}\sup_{|t|\le 2\sqrt[4]{1+x^2}}|t|\le \frac{c}{\sqrt{1+x^2}},\qquad x\gg1.\label{6.17} \end{align} The estimate for $\tau_2(x),$ $x\gg1,$ follows from \eqref{6.10} and \eqref{6.1}: \begin{align} |\tau_2(x)|&\le \left|\int_0^{\sqrt 2d(x)}q_2(x+t)dt\right|+\left|\int_0^{\sqrt 2d(x)}q_2(x-t)dt\right|\nonumber\\ &=\left|\int_{x-\sqrt 2d(x)}^xq_2(\xi)d\xi\right|+\left|\int_x^{x+\sqrt 2d(x)}q_2(\xi)d\xi\right|\nonumber\\ &\le 2\sup_{[\alpha,\beta]\subseteq\omega(x)}\left|\int_\alpha^\beta\frac{\cos e^\xi}{1+\xi^2}d\xi\right|\le c\frac{e^{-x}}{\sqrt{1+x^2}},\quad x\gg 1.\label{6.18} \end{align} From \eqref{6.17}, \eqref{6.18} and \eqref{6.14}, we obtain \eqref{4.6}, and therefore $q\in H.$ \noindent 5. \textit{Checking that the weights $\mu(x)$ and $\theta(x)$ agree with the function $q.$} Equalities \eqref{4.12} for the functions $\mu(x)$ and $1/\theta(x)$ (see \eqref{6.2}) are easily proved with the help of estimates \eqref{6.14}. \noindent 6. \textit{Proof of assertion B).} Below we use \thmref{thm4.7}. Let us check that in case \eqref{6.2} requirements \eqref{4.2} are satisfied. Let $x_0\gg1.$ Then \begin{align*} \int_0^\infty \mu(t)dt&=\int_0^\infty\frac{dt}{\sqrt{1+t^2}\ln(2+t^2)}\ge\int_{x_0}^\infty\frac{1}{t\sqrt{1+t^{-2}}}\cdot \frac{dt}{2\ln t+\ln(1+2t^{-2})}\\ &\ge c^{-1}\int_{x_0}^\infty\frac{dt}{t\ln t}=\infty\quad\Rightarrow\quad \eqref{4.2}. \end{align*} Since the weights $\mu$ and $\theta$ agree with the function $q,$ and one has the relations (see assertion \eqref{6.14}): $$m(q,\mu,\theta)=\sup_{x\in\mathbb R}\left(\frac{\mu(x)}{\theta(x)}d^2(x)\right)\le c\sup_{x\in\mathbb R}\left(\frac{\mu(x)}{\theta(x)}\sqrt{1+x^2}\right)=c<\infty,$$ assertion B) follows from \thmref{thm4.7}. \hfill\quad\qed
7c6ec6b07f6e9cc57c84968ebb95269aa2395b34
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Spectral embedding methods are based on analyzing Markov chains on a high-dimensional data set $\left\{x_i\right\}_{i=1}^{n} \subset \mathbb{R}^d$. There are a variety of different methods, see e.g. Belkin \& Niyogi \cite{belk}, Coifman \& Lafon \cite{coif1}, Coifman \& Maggioni \cite{coif2}, Donoho \& Grimes \cite{donoho}, Roweis \& Saul \cite{rs}, Tenenbaum, de Silva \& Langford \cite{ten}, and Sahai, Speranzon \& Banaszuk \cite{sahai}. A canonical choice for the weights of the graph is declare that the probability $p_{ij}$ to move from point $x_j$ to $x_i$ to be $$ p_{ij} = \frac{ \exp\left(-\frac{1}{\varepsilon}\|x_i - x_j\|^2_{\ell^2(\mathbb{R}^d)}\right)}{\sum_{k=1}^{n}{ \exp\left(-\frac{1}{\varepsilon}\|x_k - x_j\|^2_{\ell^2(\mathbb{R}^d)}\right)}},$$ where $\varepsilon > 0$ is a parameter that needs to be suitably chosen. This Markov chain can also be interpreted as a weighted graph that arises as the natural discretization of the underlying 'data-manifold'. Seminal results of Jones, Maggioni \& Schul \cite{jones} justify considering the solutions of $$ -\Delta \phi_n = \lambda_n^2 \phi_n$$ as measuring the intrinsic geometry of the weighted graph. Here we always assume Neumann-boundary conditions whenever such a graph approximates a manifold. \begin{figure}[h!] \begin{tikzpicture}[scale=0.2\textwidth, node distance=1.2cm,semithick] \node[origVertex] (0) {}; \node[origVertex] (1) [right of=0] {}; \node[origVertex] (2) [above of=0] {}; \node[origVertex] (3) [above of=1] {}; \node[origVertex] (4) [above of=2] {}; \path (0) edge[origEdge, out=-45, in=-135] node[newVertex] (m0) {} (1) edge[origEdge, out= 45, in= 135] node[newVertex] (m1) {} (1) edge[origEdge] node[newVertex] (m2) {} (2) (1) edge[origEdge] node[newVertex] (m3) {} (3) (2) edge[origEdge] node[newVertex] (m4) {} (3) edge[origEdge] node[newVertex] (m5) {} (4) (3) edge[origEdge, out=125, in= 55, looseness=30] node[newVertex] (m6) {} (3); \path (m0) edge[newEdge, out= 135, in=-135] (m1) edge[newEdge, out= 45, in= -45] (m1) edge[newEdge, out=-145, in=-135, looseness=1.7] (m2) edge[newEdge, out= -35, in= -45, looseness=1.7] (m3) (m1) edge[newEdge] (m2) edge[newEdge] (m3) (m2) edge[newEdge] (m4) edge[newEdge, out= 135, in=-135] (m5) (m3) edge[newEdge] (m4) edge[newEdge, out= 45, in= 15] (m6) (m4) edge[newEdge] (m5) edge[newEdge, out= 90, in= 165] (m6) ; \draw [thick, xshift=0.006cm,yshift=0.005cm] plot [smooth, tension=1] coordinates { (0.03,0.01) (0.04,-0.01) (0.06,0.01) (0.055,0.02) (0.05, 0.01) (0.04, 0.01) (0.035, 0.01) (0.03, 0.02) (0.03,0.01) }; \end{tikzpicture} \caption{Graphs that approximate smooth manifolds.} \end{figure} The cornerstone of spectral embedding is the realization that the map \begin{align*} \Phi: \left\{x_i\right\}_{i=1}^{n} &\rightarrow \mathbb{R}^k \\ x &\rightarrow (\phi_1(x), \phi_2(x), \dots, \phi_k(x)). \end{align*} can be used as an effective way of reducing the dimensionality. One useful explanation that is often given is to observe that the Feynman-Kac formula establishes a link between random walks on the weighted graph and the evolution of the heat equation. We observe that random walks have a tendency to be trapped in clusters and are unlikely to cross over bottlenecks and, simultaneously, that the evolution of the heat equation can be explicitely given as $$ \left[e^{t \Delta} f\right](x) = \sum_{n=1}^{\infty}{e^{-\lambda_n^2 t} \left\langle f, \phi_n \right\rangle \phi_n(x)}.$$ The exponential decay $e^{-\lambda_n^2 t}$ implies that the long-time dynamics is really governed by the low-lying eigenfunctions who then have to be able to somehow reconstruct the random walks' inclination for getting trapped in clusters and should thus be able to reconstruct the cluster. We believe this intuition to be useful and our further exposition will be based on this. \section{The Wave equation} \subsection{Introduction.} Once the eigenfunctions of the Laplacian have been understood, they imply complete control over the Cauchy problem for the wave equation \begin{align*} (\partial_t^2 - \Delta)u(x,t) = 0 \\ u(x,0) = f(x) \\ \partial_t u(x,0) = g(x) \end{align*} given by the eigenfunction expansion $$ u(t,x) = \sum_{n = 1}^{\infty}{ \left[ \cos{(\lambda_ n t)} \left\langle f, \phi_n \right\rangle + \sin{(\lambda_ n t)} \left\langle g, \phi_n \right\rangle \right] \phi_n(x) }.$$ Throughout the rest of the paper, we will understand a solution of a wave equation as an operator of that form, which is meaningful on both smooth, compact manifolds equipped with the Laplace-Beltrami operator $\Delta_g$ as well as on discrete weighted graphs equipped with the Graph Laplacian $\mathcal{L}$. A notable difference is the lack of decay associated with the contribution coming from higher eigenfunctions -- this is closely related to the fact that the heat equation is highly smoothing while the wave equation merely preserves regularity. In one dimension, this is is easily seen using $$ (\partial_t^2 - \partial_x^2) = (\partial_t - \partial_x)( \partial_t + \partial_x)$$ implying that translations $u(x,t) = f(x+t)$ and $u(x,t) = f(x-t)$ are particular solutions of the wave equation which preserve their initial roughness). However, the dynamics is still controlled by low-lying eigenfunctions in a time-averaged sense: note that $$ \frac{1}{b-a}\int_{a}^{b}{u(t,x) dt} = \sum_{n = 1}^{\infty}{ \left[ \left( \frac{1}{b-a} \int_{a}^{b}{\cos{(\lambda_ n t)} dt}\right) \left\langle f, \phi_n \right\rangle + \left( \frac{1}{b-a} \int_{a}^{b}{ \sin{(\lambda_ n t)} dt}\right) \left\langle g, \phi_n \right\rangle \right] \phi_n(x) }$$ where the integrals decay as soon as $\lambda_n^{-1} \lesssim b-a$ since $$ \frac{1}{b-a} \int_{a}^{b}{ \sin{(\lambda_ n t)} dt} \lesssim \min \left( 1, \frac{1}{\lambda_n} \frac{1}{b-a} \right).$$ Put differently, the average behavior over a certain time interval is much smoother than the instantenous behavior. We will now prove that 'average' considerations within the framework of the wave equation allow us to reconstruct the classical distance used in spectral embedding: then, after seeing that 'average' considerations recover the known framework, we will investigate the behavior on shorter time-scales and use that as a way of deriving a finer approximation of the underlying geometry of the given data. \subsection{Recovering the spectral distance.} We start by defining the usual spectral distance between two elements $x_0, y_0 \in \mathcal{M}$ w.r.t. the first $N$ eigenfunctions as $$ d_N(x_0, y_0)^2 = \sum_{n=1}^{N}{(\phi_n(x_0) - \phi_n(y_0))^2}.$$ Equivalently, this may be understood as the Euclidean distance of the embedding $$ d_N(x_0, y_0)^2 = \|\Phi_N(x_0) - \Phi_N(y_0)\|_{\ell^2(\mathbb{R}^N)}^2.$$ Given the dynamical setup of a wave equation, there is another natural way of measuring distances. Given a point $z \in \mathcal{M}$, we define $u_{z}(x,t)$ as the solution of \begin{align*} (\partial_t^2 - \Delta)u_{z}(x,t) &= 0 \\ u_{z}(x,0) &= \delta_{z} \\ \partial_t u_{z}(x,0) &= 0, \end{align*} where $\delta_z$ is the Dirac $\delta-$function in the point $z$. The solution starts out being centered at $z$ and then evolves naturally according to the heat equation. Since we are mainly interested in computational aspects, we will use $u_{z,N}$ to denote the projection of $u_z$ onto the first $N$ Laplacian eigenfunctions. It is natural to assume that if $x_0, y_0 \in \mathcal{M}$ are close, then $u_{x_0}(x,t)$ and $u_{y_0}(x,t)$ should be fairly similar on most points of the domain for most of the time. \begin{figure}[h!] \minipage{0.5\textwidth} \begin{center} \includegraphics[width=0.9\textwidth]{realwave.pdf} \end{center} \endminipage\hfill \minipage{0.45\textwidth} \begin{center} \begin{tikzpicture}[scale=1.2] \draw [ultra thick] (0,0) circle (0.02cm); \draw [ultra thick] (0,0) circle (0.2cm); \draw [ultra thick] (0,0) circle (0.4cm); \draw [ultra thick] (0,0) circle (0.6cm); \draw [ultra thick] (0,0) circle (0.8cm); \draw [ultra thick] (0,0) circle (1cm); \draw [ultra thick] (0,0) circle (1.2cm); \draw [ultra thick] (0,0) circle (1.4cm); \draw [ultra thick] (1.3,0) circle (0.02cm); \draw [ultra thick] (1.3,0) circle (0.2cm); \draw [ultra thick] (1.3,0) circle (0.4cm); \draw [ultra thick] (1.3,0) circle (0.6cm); \draw [ultra thick] (1.3,0) circle (0.8cm); \draw [ultra thick] (1.3,0) circle (1cm); \draw [ultra thick] (1.3,0) circle (1.2cm); \draw [ultra thick] (1.3,0) circle (1.4cm); \end{tikzpicture} \end{center} \endminipage\hfill \captionsetup{width=0.9\textwidth} \caption{An evolving Dirac $\delta-$function and the overlap between two solutions.} \end{figure} We will now prove the Main Theorem stating that this notion fully recovers the spectral distance. \begin{thm}[Wave equation can recover spectral distance.] Assume $\mathcal{M}$ is connected (in the sense of $\lambda_1 > 0$). Then the average $L^2-$distance of the wave equation arising from Dirac measures placed in $x_0, y_0$ allows to reconstruct the spectral distance $d_N(x_0,y_0)$ via $$ \lim_{T \rightarrow \infty}{\frac{1}{T} \int_{0}^{T}{ \int_{\Omega}{(u_{x_0,N}(x,t) - u_{x_0,N}(x,t))^2 dx} dt}} = \frac{1}{2} d_N(x_0, y_0)^2.$$ \end{thm} \begin{proof} By definition, we have that $$ u_{N,x_0}(x,t) = \sum_{n = 0}^{N}{\cos{(\lambda_ n t)} \left\langle \phi_n, \delta_{x_0} \right\rangle \phi_n(x) } \quad \mbox{and} \quad u_{N,y_0}(x,t) = \sum_{n = 0}^{N}{\cos{(\lambda_ n t)} \left\langle \phi_n, \delta_{y_0} \right\rangle \phi_n(x) } $$ We explicitly have that $$ \phi_0(x) = \frac{1}{\sqrt{|\mathcal{M}|}} \qquad \mbox{and} \quad \left\langle \phi_n, \delta_{z} \right\rangle = \phi_n(z).$$ Since the $\phi_n$ are orthonormal in $L^2(\Omega)$, the Pythagorean theorem applies and \begin{align*} \int_{\Omega}{(u_N(x,t) - w_N(x,t))^2 dx} &= \int_{\Omega}{ \left( \sum_{n = 1}^{N}{ \cos{(\lambda_ n t)} (\phi_n(x_0) - \phi_n(y_0)) \phi_n(x) } \right)^2 dx}\\ &= \sum_{n=1}^{N}{\cos{(\lambda_ n t)}^2 (\phi_n(x_0) - \phi_n(y_0))^2 \int_{\Omega}{\phi_n(x)^2 dx}}\\ &= \sum_{n=1}^{N}{\cos{(\lambda_ n t)}^2 (\phi_n(x_0) - \phi_n(y_0))^2} \end{align*} and, since $\lambda_k \geq \lambda_1 > 0$, we easily see that $$ \lim_{T \rightarrow \infty}{\frac{1}{T} \int_{0}^{T}{\cos{(\lambda_n z)}^2dz}} = \frac{1}{2}$$ and therefore $$ \lim_{T \rightarrow \infty}{\frac{1}{T} \int_{0}^{T}{ \sum_{n=1}^{N}{\cos{(\lambda_ n t)}^2 (\phi_n(x_0) - \phi_n(y_0))^2} dt}} = \frac{1}{2} \sum_{n=1}^{N}{ (\phi_n(x_0) - \phi_n(y_0))^2}.$$ \end{proof} \textit{Remark.} If $\mathcal{M}$ is not connected but has multiple connected components, then the argument shows $$ \frac{1}{2} d_N(x_0, y_0)^2 \leq \lim_{T \rightarrow \infty}{\frac{1}{T} \int_{0}^{T}{ \int_{\Omega}{(u_{x_0,N}(x,t) - u_{x_0,N}(x,t))^2 dx} dt}} \leq d_N(x_0, y_0)^2.$$ \section{The Algorithm} \begin{flushright} \textit{If you want to see something, you send waves in its general direction, you don't throw heat at it.}\\ -- attributed to Peter Lax \end{flushright} \subsection{Spectral Echolocation.} The Theorem discussed in the preceeding section suggests that we lose no information when using distances induced by the wave equation. The main underlying idea of our approach is that we naturally obtain \textit{additional} information. We emphasize that the algorithm we describe here is \textit{not} a dimension-reduction algorithm -- instead, it can be regarded as a natural pre-processing step to enhance the effectiveness of spectral methods. Furthermore, it is more appropriate to speak of an entire family of algorithms: there are a variety of parameters and norms one could define and the optimal choice is not a priori clear.\\ \begin{mdframed} \textbf{Spectral Echolocation Algorithm.} \begin{quote} \begin{enumerate} \item \textbf{Input.}~A weighted graph $G = (V,E,w)$. \item Compute the first $N$ Laplacian eigenfunctions $\left\{\phi_1, \phi_2, \dots, \phi_N\right\}$. \item Pick $k$ random points $v_1, v_2, \dots, v_k \in V$. \item Define $k$ functions $f_1, \dots, f_k: V \rightarrow \mathbb{R}$ as 'mollifications' of the indicator functions associated to the $k$ points. We propose taking the existing affinities given by the weights $$ f_i(x) = w_{i,x}. $$ \item Pick $\varepsilon > 0$. The projection of the solution of the attenuated wave equation with $f_i$ as initial datum onto the first $N$ eigenfunctions is $$ u_i(t,x) = \sum_{n=1}^{N}{ \cos{(\lambda_k t)} e^{-\varepsilon \lambda_k t} \left\langle f_i, \phi_n\right\rangle\phi_n(x)}$$ \item Define a new weight between any two points $v_1, v_2 \in V$ given by $$ d_i(v_1, v_2) =\| u_i(t,v_1) - u_i(t,v_2)\|^{\alpha}_{X} + \| (u_i)_t(t,v_1) - (u_i)_t(t,v_2)\|^{\beta}_{Y},$$ where $u_t$ is the derivative in time and $X,Y$ are any norms on the space of continuous functions $C[0,T]$ and $\alpha, \beta > 0$. \item \textbf{Output.} A distance $d:V \times V \rightarrow \mathbb{R}_{+}$ synthesized out of $d_1, \dots, d_k$, examples being $$d(v_1,v_2) = \min_{1 \leq i \leq k}{ d_i(v_1, v_2)} \quad \mbox{and} \quad d(v_1,v_2) = \frac{1}{k}\sum_{i=1}^{k}{ d_i(v_1, v_2)}.$$ \end{enumerate} \end{quote} \end{mdframed} \vspace{10pt} The underlying idea is quite simple: we start with various initial distributions of 'water' at rest. We want these initial configurations to be relatively smooth so as to avoid drastic shocks. Given this initial configuration, we follow the evolution of the wave equation at our desired level of resolution (given by restricting to $N$ eigenfunctions). Points that are nearby should always have comparable levels of water and comparable levels of change in water level and this is measured by the integral norm. The exponentially decaying term $\exp(-\varepsilon \lambda_k t)$ in the evolution of the solution $$ u_i(t,x) = \sum_{n=1}^{N}{ \cos{(\lambda_k t)} e^{-\varepsilon \lambda_k t} \left\langle f_i, \phi_n\right\rangle \phi_n(x)}$$ comes from actually solving for the attenuated wave equation which further reduces high-frequency shocks and increases stability. As described above, setting $X = L^2$, $\varepsilon = 0$, squaring the norm, ignoring the derivative term completely and letting $T \rightarrow \infty$ recovers the original weights of the graph completely. In practice, we have found that $T = 1, \varepsilon = \lambda_1^{-1}$, $X = Y = L^1[0,1]$ and $\alpha=\beta=1$ yield the best results, however, this is a purely experimental finding -- identifying the best parameters and giving a theoretical justification for their success is still an open problem. \section{Examples of Noisy Clustering} \subsection{Parameters} We always consider $N=50$ eigenfunctions and $k=10$ randomly chosen initial spots from which to send out waves. The attenuation factor is always chosen as $\varepsilon = \lambda_1^{-1}$ and time is chosen so that the first eigenfunction performs one oscillation $T=\lambda_1^{-1}$. Further parameters are $X = L^1[0,T] = Y$ and $\alpha = 1 = \beta$. This uniquely defines the $1 \leq i \leq k$ individually induced distances, we always condense them into one distance using either $$d(v_1,v_2) = \min_{1 \leq i \leq k}{ d_i(v_1, v_2)} \qquad \mbox{or} \qquad d(v_1,v_2) = \frac{1}{k}\sum_{i=1}^{k}{ d_i(v_1, v_2)}.$$ Generally, continuous geometries benefit from taking the minimum because of increased smoothness whereas clustering problems are better treated using the second type of combined distance. \subsection{Geometric Clusters with Erroneous Edges}\label{geomClusterErroneousEdge} A benefit of the refined wave echolocation metric is that, unlike heat, the transmission between two clusters does not simply depend on the number of edges but also their topology. We consider two clusters in $\mathbb{R}^2$ each of which consists of a 1000 points arranged in a unit disk and the two unit disks are well-separated -- the obstruction comes from a large number of random edges; specifically, every point is randomly connected to 4\% in the other cluster. Heat diffuses quickly between these two clusters due to the large number of intra cluster connections. For this reason, the heat embedding of the fails to separate the clusters (however, it is does preserve some aspects of the topology, see Fig. \ref{fig:randErrorEmbeddings}). In contrast, however, the wave echolocation metric manages a clear separation of objects. \begin{figure}[h!] \begin{tabular}{cc} \includegraphics[width=.48\textwidth]{laplacianEmbedding2cluster.png} & \includegraphics[width=.48\textwidth]{waveEmbedding_affStart2.png} \end{tabular} \caption{Heat kernel embedding (left) and embedding of the wave echolocation metric (right). We used averaging across 10 starting positions.}\label{fig:randErrorEmbeddings} \end{figure} \subsection{Social Networks} Social networks are a natural application of spectral methods, and mirror the synthetic example in Section \ref{geomClusterErroneousEdge}. We examine spectral echolocation on the Facebook social circles dataset from \cite{facebookNetwork}, which consists of 4039 people in 10 friend groups. While there exist clear friend groups, edges within the clusters are still somewhat sparse, and there exist erroneous edges between clusters. One goal is to propagate friendship throughout the network and emphasize clusters. Figure \ref{fig:socialNetworkAffinity} shows the original affinity matrix, sorted by cluster number. We also compute the diffusion distance and spectral echolocation distance, and display the affinity matrix $W_{i,j} = exp(-d(v_i, v_j)^2)$ for both. Spectral echolocation not only compresses the inter cluster distances, it also discovers weak similarity between different clusters that share a number of connections. \begin{figure}[!h] \begin{tabular}{ccc} \includegraphics[width=.32\textwidth]{orginialAffinity.png} & \includegraphics[width=.32\textwidth]{laplacianAffinity.png} & \includegraphics[width=.32\textwidth]{waveEmbedding.png} \\ \includegraphics[width=.32\textwidth]{orginialAffinity_zoomed.png} & \includegraphics[width=.32\textwidth]{laplacianAffinity_zoomed.png} & \includegraphics[width=.32\textwidth]{waveEmbeddingAffinity_zoomed2.png} \\ \end{tabular} \captionsetup{width=0.95\textwidth} \caption{Original Facebook affinity matrix (left), heat kernel embedding affinity (center), and wave embedding affinity (right). Bottom is zoomed in version of the top.}\label{fig:socialNetworkAffinity} \end{figure} \textit{Substructures.} Another natural goal is to detect small ``friendship circles'' within the larger network. These circles are based off of the features that brought the group together (e.g. same university, same work, etc). Overall there are 193 circles, though many only contain two or three people, and many of the larger circles are nowhere close to a dense network. We compare the average affinity within the 100 largest circles across several techniques. For both the standard heat kernel embedding as well as the wave embedding, we build a new graph between people based on whether two people are ``10-times closer than chance'', which is defined as \begin{eqnarray*} \{x,y\}\in E \iff e^{-d(x,y)^2/\epsilon} > 10\cdot \mathbb{E}_{x'}\mathbb{E}_{y'} [e^{-d(x',y')^2/\epsilon}]. \end{eqnarray*} We now compare the typical number of edges in each circle for the original data as well as the two embeddings -- we observe a dramatic improvement. The results are displayed in Figure \ref{fig:circles}. \begin{figure}[h!] \begin{tabular}{ccc} \includegraphics[width=.32\textwidth]{origAffinityInCircle.png} & \includegraphics[width=.32\textwidth]{lapAffinityInCircle.png} & \includegraphics[width=.32\textwidth]{waveAffinityInCircle.png} \end{tabular} \caption{Number of friendship edges in original network (left), heat kernel embedding network (center), wave embedding network (right).}\label{fig:circles} \end{figure} \section{Examples with Heterogeneous Dimensionality} \subsection{Plane with Holes} We examine the behavior of waves in a porous medium. Figure \ref{fig:planeWithHoles} shows that the wave equation travels more quickly through the bridges (the wave speeds up while in a bottleneck), and gets caught in the intersections. preserves the topology of the data and emphasizes the holes. \begin{figure}[!h] \begin{tabular}{ccc} \includegraphics[height=.25\textwidth]{originalData2.png} & \includegraphics[height=.25\textwidth]{laplacianEmbedding2.png} & \includegraphics[height=.25\textwidth]{waveEmbedding3.png} \end{tabular} \captionsetup{width=0.95\textwidth} \caption{Original Data (left), heat kernel Embedding (center), and Wave Embedding (right). We used the minimum distance across 10 starting positions.}\label{fig:planeWithHoles} \end{figure} \subsection{Union of Manifolds} Another interesting property of the wave equation is that the change in position $u_t$ undergoes a dramatic change whenever the dimensionality of the manifold changes: waves are suddenly forced into a very narrow channel or -- going the other direction -- are suddenly evolving in many different directions. We demonstrate this first in Figure \ref{fig:6D1D}. The data consists of two six-dimensional spheres, connected by a one-dimensional line. The low frequency eigenfunctions of the heat kernel travel from one end to the other without much recognition of the varying dimensionality. However, the wave embedding creates a gap between the bridge and the two spheres, with the variation of the first non-trivial eigenfunction being supported almost entirely on the bridge. \begin{figure}[!h] \begin{tabular}{ccc} \includegraphics[width=.33\textwidth]{6DSpheres1DBridge2.png} & \includegraphics[width=.33\textwidth]{6DSpheres1DBridge_laplacianEmbedding2.png} & \includegraphics[width=.33\textwidth]{6DSpheres1DBridge_waveEmbedding2.png} \end{tabular} \caption{6D spheres with 1D bridge (left), heat kernel embedding (center), and wave embedding (right). For this problem, we use the min distance across 10 starting positions.}\label{fig:6D1D} \end{figure} \subsection{Union of Manifolds with different dimensions} We also consider the same problem with spheres of different dimensions as in Figure \ref{fig:6D3D1D}. The data consists of a six-dimensional sphere and a three-dimensional sphere connected by a one-dimensional line. The figure displays the affinity matrices of the points, with the first block representing the six-dimensional sphere, the second block representing the three-dimensional sphere, and the third small block for the bridge. Notice that, in the heat kernel affinity, the bridge has affinity to more points in the lower dimensional sphere than the higher dimensional sphere. Also notice that the wave embedding separates the six-dimensional sphere much further from the bridge than the three-dimensional sphere.\\ \begin{figure}[!h] \begin{tabular}{cc} \includegraphics[width=.48\textwidth]{6D3D1DBridge2.png} & \includegraphics[width=.48\textwidth]{6D3D1D_waveEmbeddingSorted2.png} \\ \includegraphics[width=.48\textwidth]{6D3D1D_lapAffinitySorted.png} & \includegraphics[width=.48\textwidth]{6D3D1D_waveAffinitySorted.png} \end{tabular} \captionsetup{width=0.95\textwidth} \caption{6D sphere in cluster 1 and 3D sphere in cluster 2 with 1D bridge (top left), wave embedding (top right), heat kernel affinity matrix (bottom left), and wave affinity matrix (bottom right). For this problem, we use the min distance across 10 starting positions.}\label{fig:6D3D1D} \end{figure} Finally, we consider two six-dimensional spheres connected via a two-dimensional bridge in Figure \ref{fig:6D2D}. Specifically, we examine the local affinities of several points on the bridge. Notice that, for the wave equation, the affinities of points on the bridge are far from isotropic and clearly distinguish the direction the wave is traveling between the two spheres. Moveover, points on the bridge near the spheres have much lower affinity to points on the sphere than their heat kernel counterparts. \begin{figure}[!h] \begin{tabular}{cc} \includegraphics[width=.48\textwidth]{waveAff1.png} & \includegraphics[width=.48\textwidth]{waveAff2.png} \\ \includegraphics[width=.48\textwidth]{lapAff1.png} & \includegraphics[width=.48\textwidth]{lapAff2.png} \end{tabular} \captionsetup{width=0.95\textwidth} \caption{Neighborhoods of chosen points in 6D spheres--2D bridge example for wave embedding using averaging across 10 starting positions (top) and heat kernel affinity (bottom).}\label{fig:6D2D} \end{figure} \section{Comments and Remarks} \subsection{Other partial differential equations} Spectral echolocation has two novel components: \begin{enumerate} \item the evolution of a dynamical system on an existing weighted graph \item and the construction of a refined metric using information coming from the behavior of the dynamical system. \end{enumerate} Our current presentation had its focus mainly on the case where the dynamical system is given by the wave equation, however, it is not restricted to that. Let us quickly consider a general linear partial differential equation of the type $$ \frac{\partial}{\partial t} u(t,x) = D u(t,x) \qquad \mbox{on}~\mathbb{R},$$ where $D$ is an arbitrary differential operator. The Fourier transform in the space variable yields a separation of frequencies $$ \frac{\partial}{\partial t} \widehat{u}(t,\xi) = i p(D, \xi) \widehat{u}(t, \xi),$$ where $p(D,\xi)$ is the symbol of the differential operator at frequency $\xi$. This is a simple ordinary differential equation whose solution can be written down explicitly as $$ \widehat{u}(t, \xi) = e^{i t p(D, \xi)} \widehat{u}(0, \xi)$$ and taking the Fourier transform again allows us to write the solution as $$ u(t,x) = \int_{\mathbb{R}}{ e^{i x \cdot \xi + i t p(D, \xi)} \widehat{u}(0, \xi) d\xi}.$$ Differential equations for which this scheme is applicable include the classical heat equation ($D = \Delta$) but also variants that include convolution with a sufficiently nice potential ($Du = \Delta u + V * u$), the Airy equation ($D = \partial_{xxx}$) and, more generally, any sufficiently regular pseudo-differential operator (for example $\sqrt{-\Delta + c^2}$). The crucial insight is that the abstract formulation via the Fourier transform has a direct analogue on weighted graphs: more precisely, given eigenfunctions $\phi_1, \dots, \phi_N$ associated to eigenvalues $\lambda_1, \dots, \lambda_N$, the natural `frequency' associated to $\phi_k$ is, of course, $\lambda_k$ and we may define the solution of $$ \frac{\partial}{\partial t} u(t,x) = D u(t,x)$$ in the same way via $$ u(t,x) = \sum_{k=1}^{N}{ e^{ p(\lambda) t} \left\langle u(0,x), \phi_k \right\rangle_{L^2} \phi_k(x)}.$$ \begin{figure}[h!] \begin{tabular}{ccc} \includegraphics[width=.32\textwidth]{waveAff12.png} & \includegraphics[width=.32\textwidth]{waveAff22.png} & \includegraphics[width=.32\textwidth]{waveAff3.png} \\ \includegraphics[width=.32\textwidth]{heatAff1.png} & \includegraphics[width=.32\textwidth]{heatAff2.png} & \includegraphics[width=.32\textwidth]{heatAff3.png} \\ \includegraphics[width=.32\textwidth]{airyAff1.png} & \includegraphics[width=.32\textwidth]{airyAff2.png} & \includegraphics[width=.32\textwidth]{airyAff3.png} \\ \includegraphics[width=.32\textwidth]{schrodAff1.png} & \includegraphics[width=.32\textwidth]{schrodAff2.png} & \includegraphics[width=.32\textwidth]{schrodAff3.png} \end{tabular} \caption{Wave (top), Heat ($2^{nd}$ row), Airy ($3^{rd}$ row), and Schrodinger (bottom).} \end{figure} Natural `symbols' include heat $p(\lambda) = - \lambda^2$, Airy $p(\lambda) = i \lambda^3$ or Schrodinger $p(\lambda) = i \lambda$. Naturally, the same analysis goes through for equations of the type $u_{tt} = Du$ and our analysis of the attenuated wave equation above follows that scheme. The analysis of partial differential equations on graphs is still in its infancy and our original motivation for using the wave equation is a number of desirable properties that seem uniquely suited for the task at hand: no dissipation of energy and finite speed of propagation. Numerical examples show that different symbols $p(\lambda)$ can induce very similar neighborhoods: we believe that this merits further study; in particular, a thorough theoretical analysis of the proposed family of algorithms is highly desired. \subsection{Special case: Heat equation} We want to emphasize that our approach is novel even when we chose to emulate the classical heat propagation. This method can outperform the classical (unrefined) embedding via Laplacian eigenmaps even in relatively simple toy examples: we consider the classical 2D dumbbell example in Figure \ref{fig:dumbbell2}. \begin{figure}[h!] \begin{tabular}{c} \includegraphics[width=.5\textwidth]{dumbbellData.png} \end{tabular} \caption{The dumbbell domain in our experiment.}\label{fig:dumbbell2} \end{figure} This example has a small Cheeger constant due to the bottleneck, which means the first non-trivial eigenfunction will be essentially constant on the boxes and change rapidly on the bottleneck. This classical examples illustrates well how the first nontrivial eigenfunction can be used as a classifier and the classical Laplacian eigenmap works spectacularly well without any further modifications. \begin{figure}[h!] \begin{tabular}{c} \includegraphics[width=.6\textwidth]{dumbbellEigenfunctions.png} \end{tabular} \caption{The values of the first eigenfunction for classical Laplacian eigenmaps (blue) vs. the refined heat metric (blue).}\label{fig:dumbbell} \end{figure} Figure \ref{fig:dumbbell} compares the eigenfunction $\phi_1$ of the Laplacian compared to the first non-trivial eigenfunction of the heat equation distance matrix. We observe that the refined heat metric is a much better approximation to the function \[f(x) = \begin{cases} -1 & x_1\leq -0.5 \\ 2x & -0.5< x_1< 0.5 \\ 1 & x_1\geq 0.5 \end{cases} \] than $\phi_1$ and allows for a more accurate reconstruction of the bridge. We also observe that the nontrivial eigenfunction is essentially and to a remarkable degree constant on the two clusters which further increases its value as a classifier. \section{Conclusions} \textit{Summary.} We have presented a new pre-processing technique for classical dimensionality reduction techniques based on spectral methods. The underlying new idea comes in three parts: (1) if one computes eigenfunctions of the Laplacian, then one might just as well use them so simulate the evolution of a partial differential equation on the existing weighted graph, (2) especially for physically meaningful dynamical systems such as the wave equation one would expect points with high affinity to behave similarly throughout time and (3) this motivates the construction of a refined metric extracting information coming from the behavior of the dynamical system. \\ \textit{The wave equation.} We were originally motivated by a series of desirable properties of the wave equation on $\mathbb{R}^n$: preservation of regularity and finite speed of propagation. Recall that one of the fundamental differences between the heat equation and the wave equation is that solutions of the heat equation experience an instanteneous gain in smoothness while the wave equation merely preserves the smoothness of the initial datum (and sometimes not even that). Our main point is to show that this is not arbitrary but due to physical phenomena whose counterparts in the world of data can provide a refined measurement: \textit{the lack of regularity can be helpful}! However, as we have shown, there are similar effects for most other partial differential equations and theoretical justifications on a precise enough level that they would distinguish between various dynamical systems are still missing -- we believe this to be a fascinating open problem.\\ \textit{Refined metrics.} Similarily, our way of refining metrics, either by taking the minimum or by compiling an average, is motivated by considerations (see also \cite{stein}) that are not specifically tuned to our use of partial differential equations -- another fascinating open question is whether there is a more natural and attuned way of extracting information.\\ \textbf{Acknowledgement.} The authors are grateful to Raphy Coifman for a series of fruitful discussions and helpful suggestions. A.C. is supported by an NSF Postdoctoral Fellowship \#1402254, S.S. is supported by an AMS Simons Travel Grant and INET Grant \#INO15-00038.
48196a99e6c64a2673a380447e415ed30b20c95a
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{sintro} \section{Background: CUR and low-rank approximation}\label{sbcgr} {\em Low-rank approximation} of an $m\times n$ matrix $W$ having a small numerical rank $r$, that is, having a well-conditioned rank-$r$ matrix nearby, is one of the most fundamental problems of numerical linear algebra \cite{HMT11} with a variety of applications to highly important areas of modern computing, which range from the machine learning theory and neural networks \cite{DZBLCF14}, \cite{JVZ14} to numerous problems of data mining and analysis \cite{M11}. One of the most studied approaches to the solution of this problem is given by $CUR$ {\em approximation} where $C$ and $R$ are a pair of $m\times l$ and $k\times n$ submatrices formed by $l$ columns and $k$ rows of the matrix $W$, respectively, and $U$ is a $k\times l$ matrix such that $W\approx CUR$. Every low-rank approximation allows very fast approximate multiplication of the matrix $W$ by a vector, but CUR approximation is particularly transparent and memory efficient. The algorithms for computing it are characterized by the two main parameters: (i) their complexity and (ii) bounds on the error norms of the approximation. We assume that $r\ll \min\{m,n\}$, that is, the integer $r$ is much smaller than $\min\{m,n\}$, and we seek algorithms that use $o(mn)$ flops, that is, much fewer than the information lower bound $mn$. \section{State of the art and our progress}\label{ssartpr} The algorithms of \cite{GE96} and \cite{P00} compute CUR approximations by using order of $mn\min\{m,n\}$ flops.\footnote{Here and hereafter {\em ``flop"} stands for ``floating point arithmetic operation".} \cite{BW14} do this in $O(mn\log(mn))$ flops by using randomization. These are record upper bounds for computing a CUR approximation to {\em any input matrix} $W$, but the user may be quite happy with having a close CUR approximations to {\em many matrices} $W$ that make up the class of his/her interest. The information lower bound $mn/2$ (a flop involves at most two entries) does not apply to such a restricted input classes, and we go well below it in our paper \cite{PSZa} (we must refer to that paper for technical details because of the limitation on the size of this submission). We first formalize the problem of CUR approximation of an average $m\times n$ matrix of numerical rank $r\ll \min\{m,n\}$, assuming the customary Gaussian (normal) probability distribution for its $(m+n)r$ i.i.d. input parameters. Next we consider a two-stage approach: (i) first fix a pair of integers $k\le m$ and $l\le n$ and compute a CUR approximation (by using the algorithms of \cite{GE96} or \cite{P00}) to a random $k\times l$ submatrix and then (ii) extend it to computing a CUR approximation of an input matrix $W$ itself. We must keep the complexity of Stage (i) low and must extend the CUR approximation from the submatrix to the matrix $W$. We prove that for a specific class of input matrices $W$ these two tasks are in conflict (see Example 11 of \cite{PSZa}), but such a class of hard inputs is narrow, because we prove that our algorithm produces a close approximation to the average $m\times n$ input matrix $W$ having numerical rank $r$. (We define such an average matrix by assuming the standard Gaussian (normal) probability distribution.) By extending our two-stage algorithms with the technique of \cite{GOSTZ10}, which we call {\em cross-approximation}, we a little narrow the class of hard inputs of Example 11 of \cite{PSZa} to the smaller class of Example 14 of \cite{PSZa} and moreover deduce a sharper bounds on the error of approximation by maximizing the {\em volume} of an auxiliary $k\times l$ submatrix that defines a CUR approximation In our extensive tests with a variety of real world input data for regularization of matrices from Singular Matrix Database, our fast algorithms consistently produce close CUR approximation. Since our fast algorithms produce reasonably accurate CUR approximation to the average input matrix, the class of hard input matrices for these algorithms must be narrow, and we studied a tentative direction towards further narrowing this input class. We prove that the algorithms are expected to output a close CUR approximation to any matrix $W$ if we pre-process it by applying Gaussian multipliers. This is a nontrivial result of independent interest (proven on more than three pages), but its formal support is only for application of Gaussian multipliers, which is quite costly. We hope, however, that we can still substantially narrow the class of hard inputs even if we replace Gaussian multipliers with the products of reasonable numbers of random bidiagonal matrices and if we partly curb the permutation of these matrices. If we achieve this, then preprocessing would become non-costly. This direction seems to be quite promising, but still requires further work. Finally, our algorithms can be extended to the acceleration of various computational problems that are known to have links to low-rank approximation, but in our concluding Section \ref{scncl} we describe a novel and rather unexpected extensions to the acceleration of the Fast Multipole Method and Conjugate Gradient Algorithms,\footnote{Hereafter we use the acronyms FMM and CG.} both being among the most celebrated achievements of the 20th century in Numerical Linear Algebra. \subsection{Some related results on matrix algorithms and our progress on other fundamental subjects of matrix computations}\label{srltwr} A huge bibliography on CUR and low-rank approximation, including the known best algorithms, which we already cited, can be accessed from the papers \cite{HMT11}, \cite{M11}, \cite{BW14} and \cite{W14}. Our main contribution is dramatic acceleration of the known algorithms. Some of our techniques extend the ones of \cite{PZ16}, \cite{PZ17}, and \cite{PZa}, where we also show duality of randomization and derandomization and apply it to fundamental matrix computations. In \cite{PZ16} we prove that preprocessing with almost any well-conditioned multiplier of full rank is as efficient on the average for low-rank approximation as preprocessing with a Gaussian one, and then we propose some new highly efficient sparse and structured multipliers. Besides providing a new insight into the subject, this motivates the design of more efficient algorithms and shows specific direction to this goal. We obtain similar progress in \cite{PZa} for and \cite{PZ17} for preprocessing Gaussian elimination with no pivoting and block Gaussian elimination. We recall that Gaussian elimination with partial pivoting is performed millions time per day, where pivoting, required for numerical stabilization, is frequently a bottleneck because it interrupts the stream of arithmetic operations with foreign operations of comparison, involves book-keeping, compromises data locality, and increases communication overhead and data dependence. Randomized preprocessing is a natural substitution for pivoting, and in \cite{PZa} we show that Gaussian elimination with no pivoting as well as block Gaussian elimination (which is another valuable algorithm and which also requires protection against numerical problems) are efficient on the average input with preprocessing by any nonsingular and well-conditioned multipliers. \cite{PZ17} obtains similar progress for the important subject of the approximation of trailing singular spaces associated with the $\nu$ smallest singular values of a matrix having numerical nullity $\nu$. Our current progress greatly supersedes these earlier results, however, in terms of the scale of the acceleration of the known algorithms. Our technique of representing random Gaussian multipliers as a product of random bidiagonal factors, our extension of CUR approximation to FMM and CG algorithms, and our analysis of CUR approximation for the average input are new and can have some independent interest. \section{Conclusions}\label{scncl} We dramatically accelerated the known algorithms for the fundamental problems of CUR and low-rank approximation in the case of the average input matrix and then pointed out a direction towards heuristic extension of the resulting fast algorithm to a wider class of inputs by applying quasi Gaussian preprocessing. Our extensive tests for benchmark matrices of discretized PDEs have consistently supported the results of our formal analysis. Our study can be extended to a variety of important subjects of matrix computations. Some of such extensions have been developed in papers \cite{PZ16}, \cite{PZ17} and \cite{PZa}, and there are various challenging directions for further progress. In particular our accelerated CUR and low-rank approximation enables faster solution of some new important computational problems, thus extending the long list of the known applications. In the concluding section of \cite{PSZa}, we add two new highly important subjects to this long list.
ff44ea6d25a77da54b74b5060a441c09c58b4dfa
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} NGC 4660 is an E5 elliptical galaxy with a strong incrusted disk component, and has frequently been used as a `classic' example of this type of galaxy. As such it has appeared frequently in the literature. It is located in the Virgo cluster, and has been identified as a member of the nearest compact group satisfying the Hickson criterion \citep{mam89}, along with other Virgo galaxies like M59 and M60, but \citet{mam08} concluded from surface brightness fluctuation measurements that M59 and NGC 4660 form a pair $\sim 1 - 2$ Mpc closer to us than the other three galaxies of the potential compact group. Hence we adopt a distance of 18 Mpc for this galaxy, giving a scale of 87 pc/arcsec and 5.2 kpc/arcmin. Generally it has been assumed that the disk in NGC 4660 is primordial and that the galaxy has not undergone a significant interaction or merger, but here we report the discovery of a long, curved filament apparently emanating from NGC 4660, in the direction of M59, which implies that it has experienced some kind of major event with a dynamical timescale of a few $\times 10^{8}$ years. The filament was discovered on a co-added array of scanned Kodak Technical Pan films of the SE region of the Virgo cluster taken with the UK Schmidt Telescope \citep{kat98, kat01} which have previously been used to identify or confirm other filamentary structures, e.g. the filamentary `ellipse' of IC 3481 \citep{per09}, the filamentary connections between NGC 4410 A/B, C and D \citep{per08} etc. Although such Schmidt plate/film studies have become almost obselete in modern astronomy, this discovery of the present, previously unsuspected, filament of NGC 4660 shows that they could still be put to productive uses. Here we report the discovery of the filament using the Schmidt data, and describe subsequent multiband CCD observations with the 2.1m telescope at the San Pedro M\'artir Observatory, which provide colour information for this galaxy and further imaging of the filament. In Section 2 we provide a summary of the previous work on the galaxy NGC 4660, while in Section 3 we describe the observational data and their processing. Results of the photometry of NGC 4660 and of the filament are given in Section 4, and in Section 5 we discuss the age and the formation of the filament and give general conclusions. \section {Previous work on NGC 4660} NGC 4660 (VCC 2000) has been studied in various ways, and has formed part of many samples of early-type galaxies over the years, on account of its proximity to us in the Virgo cluster and its incrusted disk component. It is located at RA = 12:44:32.0 Dec. = +11:11:26 (J2000), has a recessional velocity of 1083 km/s, and redshift of 0.003612 (NASA/IPAC Extragalactic Database). Although its classification in the SIMBAD database is E5, the Virgo Cluster Catalogue \citep{bin85} gives a classification of E3/S0--1(3), which emphasises its transitionary nature. There have been a few previous photometric and structural studies of NGC 4660, though three of them have used the same observational dataset. We compare our photometric results with theirs in Section 4. \citet{benet88} carried out CCD photometry and isophote analysis on NGC 4660 in the V, R and I bands. They found an effective radius $r_{e}$ between 9--9.8 arcsec and an isophote twist of $5^{\circ}$. Peak ellipticity was $\sim 0.5$ while $a_{4}/a$ (used as the 4th cosine coefficient in this case) rises to 0.04 at 30 arcsec. \citet{rix90} performed photometric modelling of galaxies with 2 components (spheroidal $r^{1/4}$ law bulge plus exponential disk). They applied their techniques to \citet{benet88}'s data on NGC 4660 and found they achieved better fits with a significant disk component included ($L_{disk}$ $\sim 50$\% $L_{bulge}$). \citet{ben88} described NGC 4660 as a rapidly-rotating elliptical while \citet{rix90} conclude that it is the disk rotation which is the major contributor to the detected velocity. \citet{sco95}, again using the same photometric data of \citet{benet88} for NGC 4660, find disk signatures in the ellipticity and $a_{4}/a$ profiles (again used as the 4th cosine coefficient). The ellipticity is $\sim 0.2$ at the centre and peaks at 0.5 at 10--30 arcsec, while $a_{4}/a$ is $< 0.01$ in the centre, rising to 0.02--0.04 between 10--40 arcsec. The velocity dispersion $\sigma$ is $\sim 200$ km s$^{-1}$ at $< 5$ arcsec (0.43 kpc) and 100 km s$^{-1}$ at 30 arcsec (2.6 kpc), while the disk rotation velocity is 180 km s$^{-1}$ at 10--25 arcsec (0.87--2.17 kpc) and $\sim 100$ km s$^{-1}$ at 0--10 arcsec ( 0--0.87 kpc). They find evidence of an isophotal twist, with PAs in the range $90^{\circ}$ to $100^{\circ}$, between 10--20 arcsec (0.87--1.74 kpc), taking this as evidence of a disk with inclined bar extending to 12 arcsec (1.04 kpc) from the centre. Its surface brightness profile is similar to early-type barred galaxies \citep{com93}. For the disk, $v/{\sigma}$ is very high, $\sim 3.3$, and D/B within the bar is 0.5. \citet{sco95} suggest that the disk may be unstable to bar formation, and the bar may have a long evolution time. The dataset of \citet{sco95} comprises mainly disky elliptical galaxies i.e. E galaxies with `pointed' isophotes (of which, NGC 4660 has the highest D/B ratio, 0.28). These galaxies are found to lie on the same correlation between central surface brightness and disk scale length defined by lenticular and spiral galaxies. This implies that disky E's are not inclined S0's but form `transition' objects between E's and S0's. The disk profile is not generally exponential. In NGC 4660 the disk and bulge rotate in the same direction and have similar surface brightness profiles, suggesting that they formed at about the same time out of the same material. This galaxy was observed by \citet{fer06} with HST/ACS as part of a sample of 100 early-type galaxies in Virgo. These data deal with the central regions of the galaxy at higher spatial resolution, and so the present data are complementary to this dataset. Their image shows the presence of a possible stripped satellite galaxy, 2.5 arcsec across and 4.5 arcsec south of the centre, described as blue, `resembling a nucleus with two spiral arms'. They fit the luminosity profile of NGC 4660 with one component with S\'ersic indices of 4.0 in the $g$ band and $4.5$ in the $z$ band and an average $g-z$ colour of 1.51. Structural parameters such as ellipticity, PA, $a_{3}, a_{4}, b_{3}, b_{4}$ are reported, giving results quite similar to previous references. $b_{4}$ is used here as the 4th cosine coeffcient and provides a clear separation of the boxy bulge and the disk. The $g-z$ colour profile shows a blueward tendency from 0-3 -- 20 arcsec then rises slightly to 50 arcsec. NGC 4660 forms part of the SAURON sample, a survey of 72 early-type galaxies using the integral-field spectrograph SAURON at the William Herschel Telescope. Stellar population studies of this sample give us ages and metallicities of the dominant stellar populations in central regions of this galaxy \citep{bac01, zee02}. \citet{kun10} carried out a stellar population analysis of absorption line strength maps of 48 of these galaxies including NGC 4660. They find only old stellar populations in this galaxy, $12.2 ^{+ 1.2} _{-0.6}$ Gyr at $R_{e}/8$ (1.4 arcsec, 1.3 kpc) and $13.4 ^{+1.3} _{-1.2}$ Gyr at $R_{e}$ (11.5 arcsec, 1.0 kpc). The metallicity [Fe/H] varies from $0.15 \pm 0.02$ at $R_{e}/8$ to $0.11 \pm 0.02$ at $R_{e}$, while the $\alpha$-element composition varies slightly, [$\alpha$/Fe] is $0.24 \pm 0.04$ at $R_{e}/8$ and $0.29 \pm 0.05$ at $R_{e}$. Although the [$\alpha$/Fe] variation is within the errors, it is typical of the other galaxies in the sample to have depressed [$\alpha$/Fe] values in the centre corresponding to higher metallicities. The velocity dispersion $\sigma$ is 221 km s$^{-1}$ at $R_{e}/8$ and 181 km s$^{-1}$ at $R_{e}$. All galaxies which are fast rotators in the sample and which have flattened components with disk-like kinematics are found to have different stellar populations in these flattened components. Those with young populations ($ < 3$ Gyr) frequently have circumnuclear disks and rings which are still forming stars. Meanwhile NGC 4660 belongs to the other extreme, in which the structure with disk-like kinematics has slightly older ages and lower [${\alpha}$/Fe] ratios in which the `secondary' star-formation event which formed the disk-like structure is presumably nearly as old as the `elliptical' component of the galaxy. We compare these results with populations indicated by our photometric colours in Section 5. Another interesting line of work carried out concerning NGC 4660 is its possible membership of a Compact Group (CG). \citet{mam89} considered it, alongside M59, M60, NGC 4638 and NGC 4647, as a possible CG involving members of the Virgo cluster, as it was found to satisfy Hickson (1982)'s criteria according to new magnitude measurements. However, \citet{mam08} concluded from surface brightness fluctuation measurements of individual early-type galaxies that NGC 4660 may form a pair with M59, at a distance of $\sim 2$ Mpc closer to us than most of the rest of the Virgo cluster. Although the various statistical calibrations differ, M59 and NGC 4660 as a pair are at least 440 kpc closer to us than M60 and NGC 4638. This line-of-sight depth would be too great for a CG \citep{mam08}. Our new data do not provide anything new on this interesting possibility. In the following section we consider the new observational data and their processing. \section{Observations and Data Reduction} \subsection{The Schmidt films} The use of deep, wide-field imaging to detect faint material in groups and clusters is well-established, e.g. \citet{kem93}. \citet{kat98, kat01} produced a stack of 13 digitally co-added Kodak Technical Pan films obtained with the UK Schmidt telescope, using the `OR' filter (equivalent to the Cousins R band). The films covered the SE part of the Virgo cluster, with the aim of detecting and mapping the faint material in galaxy haloes and filamentary structures in this cluster. Faint material is detected down to 28 $R$ mag arcsec$^{-2}$. Figure 1 is a section of these data containing NGC 4660. The curved filament is clearly seen to the NW of the galaxy. Our CCD fields are indicated and an arrow gives the direction of M59. \subsection{Optical CCD photometry and data reduction} We carried out direct imaging of a field containing NGC 4660 (at its SE corner, to include part of the area occupied by the filament) on 2005 March 30 with the 2.1m telescope of the Observatorio Astron\'omico Nacional, San Pedro M\'artir, Baja California, Mexico. The SITe1 CCD camera, with Johnson BVRI filters was used. We obtained total combined exposure times of 900, 600, 450, 300s in BVRI respectively in photometric conditions. The spatial scale for these data was 0.31 arcsec pixel$^{-1}$ and the CCD field of view was $5.3 \times 5.3$ arcmin$^2$, with a seeing of $\sim 2$ arcsec. 6 images of 600s in R were also taken of a field to the W which is centred on the brightest parts of the filament (as seen in the Schmidt data). The standard procedures of bias-subtraction, dark-correction and flat-fielding were carried out using IRAF. \footnote{IRAF is distributed by the National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy (AURA) under a cooperative agreement with the National Science Foundation.} For the broad-band observations, we observed Landolt standard fields containing Palomar Green stars \citep{lan92} at intervals during the night at a range of airmasses. Estimates of the extinction coefficients were obtained by plotting instrumental magnitudes against airmass for multiple observations of the same field of standard stars. These values of the extinction coefficients were used to obtain photometric zero points and colour terms. The photometric transformation equations obtained were \begin{equation} B = b + 24.24 - 0.24X + 0.13(B-V), \end{equation} \begin{equation} V = v + 24.94 - 0.15X + 0.02(B-V), \end{equation} \begin{equation} R = r + 24.67 - 0.01X + 0.05(B -V), \end{equation} \begin{equation} I = i + 24.17 -0.04X + 0.05(B-V), \end{equation} where $X$ is the effective airmass of the exposure. The errors associated with our data have two main contributions, a) the systematic error in obtaining the calibration equation, b) the error in determining the true sky value. The first is obtained by analysing the residuals in the fits used to obtain the parameters of the calibration equation. The second is estimated by measuring the sky level in several different areas of the image (free from bright objects) and calculating the $\sigma$ for these independent measurements of the sky value. Error a) is the most significant in brighter regions whilst b) is most significant for fainter regions. Errors in colours are calculated by adding in quadrature errors in individual bands. We note that for the 2.1m telescope data we have used Landolt standards defined in the Johnson-Kron-Cousins system while the filters used for the observations are Johnson filters, and in R and I there are significant differences between the two filter systems. \citet{per08} reported no significant differences between the surface brightnesses obtained over a range of galaxy colours, using previous observations of galaxies in both systems. The typical seeing FWHM in each filter was measured and the images smoothed with a gaussian filter until the image in each filter had the same FWHM. The calibration equations were applied to the counts in the images to obtain colour maps by direct subtraction of images. We performed surface photometry on NGC 4660 using the final images in counts with the stars masked out. Radial colour profiles were created by subtraction of individual calibrated surface brightness profiles, with errors calculated as above. Ellipse-fitting was carried out using the program {\it ellipse} in the isophote package in IRAF, with an outer isophote of 25.8 $B$ mag arcsec$^{-2}$ in B (4\% of sky level). The best-fit isophotal models generated can be subtracted from the final reduced image (in counts) to produce residual images, in which non-axisymmetrical structures previously hidden in the brighter parts of the galaxy can be revealed. We also used GALFIT \citep{pen02} which is an algorithm that can analyse the light distribution profiles of astronomical objects in digital images, using analytic functions, using single-component models to describe the overall morphology or using a combination of different structural components such as bulges, disks, bars, spiral arms, to model more precisely the structure of particular galaxies. Each component can be described by scale lengths, total magnitudes, ellipticity, position angle, S\'ersic index etc. We used both a one-component S\'ersic model and a two-component model (S\'ersic and exponential disk) to fit the images of NGC 4660 in all photometric bands, obtaining S\'ersic indices, bulge-to-disk luminosity ratios (B/D), effective radii, scale lengths etc. A star taken from the same field was selected as the input PSF image in GALFIT, which is included in the modelling fits to the image. This PSF image is 101 x 101 pixels with its peak flux at the 51st pixel in X and Y. \section{Results} \subsection{NGC 4660} An image of the CCD field in V containing NGC 4660 is presented in Figure 2a. In order to show details across the entire range of brightness and the brightest core regions, the image was represented in logarithmic scale and processed through layer masks. As mentioned in Section 3.2 the galaxy appears in the SE corner, to allow the inclusion of part of the area occupied by the filament in the field, though this part of the filament was not detected even at high contrast. The `pointy' isophotes indicating the presence of a disk component are quite obvious. There is a small diffuse object about 1 arcmin north-east of the centre of NGC 4660, VCC 2002 (LEDA 49295) which is classified as a dE \citep{bin85}. The residual image in V is shown in Fig. 2b. A characteristic pattern due to the disk component is evident, with excesses of light (bright areas) along the major and minor axes and minima (dark areas) at $\sim 45^{\circ}$ to these. The stripped satellite galaxy observed by \citet{fer06} can be seen at about 5 arcsec South of the centre of NGC 4660. The surface brightness profiles in the 4 filters are shown in Figure 3a. The outer ellipse has a semi-major axis of approximately 100 arcsec. The centre of this galaxy is particularly bright, reaching a central surface brightness of 14.7 $I$ mag arcsec$^{-2}$, 17.1 $B$ mag arcsec$^{-2}$. Fig. 3b shows the surface brightness profiles plotted against $a^{1/4}$, showing that this galaxy does not obey the de Vaucouleurs $r^{1/4}$ law, while at the same time, does not show huge deviations from this law. The profiles in each of the 4 filters show the same broad tendencies. Figures 4a and 4b show the $B-V$ and $B-R$ colour maps, respectively. The most distinctive feature is the nuclear and central region, including much of the disk, which displays slightly redder colours to a radius of about 20 arcsec. NNE of the nucleus, both $B-V$ and $B-R$ display a slightly yellower zone. The `halo', i.e. the main body of the elliptical, is much noisier but is bluer than the centre. In $B-R$ the eastern half of the disk appears slightly less red than the western, by $\approx 0.10 - 0.15$ mag. The small galaxy VCC 2002, 1 arcmin NE of the centre, also can be seen in these colour maps. There is no clear evidence of the feature at 4.5 arcsec south of the centre as in the HST/ACS data of \citet{fer06}, though there does seem to be a slight and abrupt change in colour, becoming redder going from east to west, at its position South of the centre of NGC 4660. There does seem to be some subtle colour changes over the disk area, which may support the `tidal stream' hypothesis (see Section 5). Figure 5 shows the radial colour profiles of NGC 4660 in $B-V$, $B-R$ and $B-I$. The error bars include both systematic and sky noise contributions. In the central regions the sky noise is insignificant relative to the galaxy counts, so changes in colour of order 0.05 mag are significant in these regions. $B-V$ is at about 0.90-0.95 in the disk region, rising slowly to $\sim 1.15$ at $a \sim 70-80$ arcsec then declining again. In $B-R$ the disk colour is $\approx 1.7$, rising to 1.90-2.00 at $\sim 70-80$ arcsec, then declining further out. In $B-I$ the regions occupied by the disk have a colour of $\approx 2.3$, and this rises steadily to about 2.8 at 65--80 arcsec. In general there is evidence for a slight but significant redward gradient in the fitted isophotes from about 35 to 80 arcsec, while the region occupied by the disk has a flatter colour profile. The colour maps however do appear to show that the central plane of the inner disk component may be $\sim 0.1$ mag redder than neighbouring parts of the galaxy. However the galaxy light is not dominated by the disk even at small radii and the overall colours of the isophotes (rounder than the disk component) may be less red than implied by the disk colours. The $g -z$ colour profile of \citet{fer06} is dominated by a blueward tendency between 0.3 and 20 arcsec but is redward to 80 arcsec and so is broadly in agreement with our profiles. For all the structural parameters investigated, the $V$ band data are used, although no significant variation was seen between this and other filters. The radial variations of the structural parameters $a_{4}/a$ and $b_{4}/a$ are plotted in Figures 6a and b respectively. We use $b_4/a$ as the 4th cosine coefficent. $a_{4}/a$ is predominantly slighty positive a (though $< 0.01$) between 10 and 30 arcsec where the disk dominates, and predominantly slightly negative between 30--45 arcsec. The value of $b_{4}/a$ rises steadily from 0.01 at $\sim 10$ arcsec to 0.05 at $\sim 30$ arcsec, indicating clearly disky isophotes in the whole radial range occupied by the disk. Beyond $\sim 45$ arsec the values of $b_{4}/a$ are negative (boxy), implying a disk embedded in a boxy halo. The same tendencies with radius are shown in the HST/ACS data of \citet{fer06} between 5--30 arcsec (this paper also uses $b_4/a$ as the 4th cosine coefficient.) The radial ellipticity profile of NGC 4660 is shown in Figure 7a. The Position Angle profile is shown in Figure 7b. Both the ellipticity and PA profiles are similar to other results in the literature including \citet{fer06}, the twist at 5--10 arcsec being taken as evidence for a disk with inclined bar \citep{sco95}. The bulge/halo therefore has the same position angle as the disk component. Results from the GALFIT program show that the bulge component and the galaxy as a whole has a $(B-V) \approx 1.0$ while the disk is significantly bluer with $(B-V) \approx 0.7$. The D/B ratio is 0.09 for B, and 0.06 for V i.e. confirming that the disk is somewhat bluer than the bulge component. These are rather lower D/B ratios than previously reported. S\'ersic indices for the bulge component are $\approx 6$. The disk scale length of around 11 arcsec is similar to the galaxy effective radius of previous authors, e.g.\ \citet{kun10}, while the bulge component effective radius is twice as large, maybe due to our relatively deeper exposures detecting more light associated with the filament around the galaxy. This may also explain the lower D/B ratios. For a summary of photometric parameters measured, see Table 1. \begin{table} \caption{ Photometric parameters measured for the whole galaxy, bulge and disk component in the $V$ filter} \begin{tabular}{crrr} Parameter & Total & Bulge & Disk \\ Magnitude & 10.9 & 11.0 & 14.1 \\ $r_e$ (arcsec) & 22 & 21 & \\ $r_s$ (arcsec) & & & 11\\ $n$ & 5.8 & 6.2 & 1 \\ Ellipticity & 0.3 & 0.3 & 0.6 \\ PA (degrees) & 91 & 91 & 91 \\ color $B-V$ & 1.0 & 1.0 & 0.7 \\ D/B & 0.06 & & \\ \end{tabular} \end{table} \subsection{The filament} The brightest parts of the filament as seen in the Schmidt image (Fig. 1) are detected in the CCD field (Figure 8) as an area of diffuse emission in the north-centre of the field (running NNE-SSW) and just S of centre between two stars. There is also an area of emission in the SE of the field corresponding to emission from the NW part of the halo of NGC 4660. The northern detection of the filament has two peaks of surface brightness of approximately 26.6 $R$ mag arcsec$^{-2}$ (RA = 12:44:18.5 Dec. = +11:15:27 (J2000)) and 27.4 $R$ mag arcsec$^{-2}$ (RA = 12:44:20.2 Dec. = +11:16:04, corresponding to only about 0.4\% and 0.2\% respectively of the sky brightness in $R$. These diffuse features may correspond to Tidal Dwarf galaxies (TDGs). Figure 9 shows the Schmidt field containing NGC 4660 and the filament without any lines marked on the image, giving a clearer unobstructed view of the filament, in which the brightest areas in the CCD filament field can be seen more clearly (with help from Fig.\ 1). The northern object (TDG-1) has an apparent $R$ mag of approximately 20.9 while that of TDG-2 is 21.5, corresponding to absolute magnitudes of --10.4 and --9.8 respectively. The scale lengths are of the order of a few pixels (100-200 pc). There are further candidate objects to the South marked as `TDGs?'. The straightened-out length of the filament is about 12 arcmin, corresponding to about 60 kpc at a distance of 18 Mpc to NGC 4660. Moving at a typical velocity for material in filaments of 100 km s$^{-1}$ (for example \citet{hib95}), stellar material at the tip of the filament would have taken $\sim 6 \times 10^{8}$ years to reach this position. \section{Discussion and Conclusions} Tidal tails and other fine low surface brightness features around galaxies can only be formed in gravitational interactions between galaxies (usually major mergers with proportion of masses less than 4) over a timescale of a few $\times 10^8$ years \citep{tom72}. Therefore the detection of the filament near to NGC 4660 appears to indicate a past gravitational interaction for this galaxy, for which there was no previous evidence. Properties of tidal tails can be used to date the merger event which formed them. Tidal tails may be long-lived after a gravitational encounter. NGC 4660 may have had an encounter, or a `dry' interaction (not merger), with another early-type galaxy. The nearest candidate galaxy for such an interaction with NGC 4660 is LEDA 42878 (VCC 1991) which can be seen at about 6 arcmin { West-SouthWest of NGC 4660 (seen in Figs.\ 1 and 9), at 12:44:09.2 +11:10:32 (J2000), is classified as a dE in the SIMBAD database and identified as a nucleated dwarf galaxy by \citet{bin85}. This has a radial velocity of 1681 km s$^{-1}$ \citep{san11}, approximately 600 km s$^{-1}$ more than that of NGC 4660, which would make any significant interaction between them unlikely. There is also the dE 1 arcmin NE of the centre of NGC 4660, VCC 2002, but this appears to be too small to produce such a long filament in an interaction with NGC 4660. Such prominent tidal tails are more likely to be formed in `wet mergers' (in which at least one progenitor is a gas-rich spiral). There are no H I observations which could demonstrate whether the filament observed near NGC 4660 is gas-rich. The presence of only one tail may indicate that only one of the progenitors of NGC 4660 was a gas-rich spiral, The period of time for which a tail may be observed is important in studies of galactic history and evolution, especially in the case of NGC 4660 where it was the only direct evidence of the galaxy having experienced a tidal interaction. Tails are formed on the dynamical timescale of $\sim 10^{8}$ years and the existence of a bright tail does not imply that the interaction which formed it was recent, bright tails may survive for a few Gyr. The galaxy evolution numerical simulations of \citet{pei10} show early-type/spiral mergers producing peaks in Star Formation Rate (SFR) for $\sim 2$ Gyr after closest approach, while shells are seen at $\sim 1.5$ Gyr after closest approach. \citet{con09} predicts maximum merger timescales of $1.1 \pm 0.3$ Gyr, without taking into account the detailed history of the fallback of tidal features. \citet{hib95} modelled the spatial morphology and velocity structure of the famous merger product NGC 7252, suggesting that the merger occurred 0.6 Gyr ago and 80\% of the mass would fall back to the merger remnant in 2.5 Gyr. So detection of tidal debris would get difficult after about 3 Gyr, while if the merger were older there would be time for more minor mergers to occur, disrupting the filament \citep{duc11}. Most of the tail material would and does remain bound -- there are currently velocity reversals along the tails indicating material which has already reached the turnaround point in its orbit and has started falling to smaller radii. 20\% of current tail particles will not fall back to $< 5 R_{e}$ in a Hubble time. A filament as relatively narrow as the one in NGC 4660 would require at least one of the parent galaxies to be a spiral galaxy, though the other may have been an early-type dynamically hot galaxy, which would only have produced plume-like debris which may disappear more rapidly. Alternatively, if the other galaxy was indeed another spiral galaxy, the other tail has evaporated or was destroyed in a subsequent minor merger interaction with the general gravitational potential well of the cluster/group environment. So in Virgo one may expect tails to survive for maybe less than 2 Gyr. The presence of a spiral galaxy would make this a gas-rich (`wet') merger. The two peaks in surface brightness in the North of the CCD field (Figure 8) may correspond to Tidal Dwarf Galaxies (TDGs) which are gravitationally bound small systems of stars and gas formed in major mergers (mass ratio $> 1:4$), see \citet{mir92}. TDGs are generally thought to form from gas clouds pulled out of galaxies during mergers and so do not represent previously existing stellar systems. Here we only have data in red filters for these TDGs, which are often found to have a bluish colour \citep{duc11}. Studies of the colours and spectra of the candidate TDGs in NGC 4660 would prove interesting, but are made difficult by their faintness. However the candidate old tidal dwarfs studied by \citet{duc14} have typical sizes of 0.8--2.3 kpc and absolute magnitudes corresponding to --13.5 to --17.5 in $R$, while their central surface brightnesses are 23.5--26.5 $R$ mag arcsec$^{-2}$. Younger dwarfs in the same article have scale lengths of 2--6 kpc, absolute $R$ magnitudes of --14 to --17.5, and central surface brightnesses ranging from 20.5 to 25 $R$ mag arcsec$^{-2}$. While our candidate TDGs in NGC 4660 are at the low end of the surface brightness range for old dwarfs they are much smaller objects, with luminosities at least 40 times less than previously identified TDGs, and only a few pixels in size, so they inevitably have small scale lengths given the detection threshold. The implied masses will also roughly be 2 orders of magnitude less, so while the objects in \citet{duc14} are of the order of $10^8 M_{\odot}$, the present objects may be only about $10^6 M_{\odot}$ in mass. \citet{duc11} consider that the TDGs in the eastern tidal tail of NGC 5557 may be at least 2 Gyr old and they are still 1.3 mag bluer than the rest of this tail, while \citet{duc14} give a spectroscopic age of 4 Gyr for one of these TDGs. If the filament in NGC 4660 is several Gyr old then these could represent faint, evolved, even older TDGs, however they are so much smaller that they seem to represent extreme lower-luminosity cases of the class of TDGs or a different class of objects. Evidence for long lived TDGs, or detection of many fainter but similar objects, would be significant in terms of studies of numbers of dwarf satellite galaxies. The `stripped satellite' detected by \citet{fer06} may provide an alternative origin for the filament, formed by stars tidally stripped from this satellite galaxy during the closest parts of its orbit to NGC 4660, making the filament a tidal stream, comparable to the tidal stream associated with the Sagittarius dwarf galaxy satellite of the Milky Way \citep{iba94}. This would account for the presence of only one detectable filament. \citet{sta15} note that detection of extragalactic tidal streams from the ground is complicated by their low surface brightnesses, below 28 mag arcsec$^{-2}$ and this seems just about compatible with the surface brightness of the filament of NGC 4660 away from the possible TDGs. The colour maps of Figure 4a and 4b do imply some complicated structure in the inner disk area, with gradients from east to west, on the South side of the disk. Studies of tidal streams are now commonly made with the 3.6$\mu$m and 4.5$\mu$m filters of the {\it Spitzer Space Telescope} on its warm mission \citep{sta15}. It would be interesting to map the area around NGC 4660 in these filters. While the present filament is relatively bright, this tidal stream hypothesis does provide a possible explanation for the origin of the filament. The ages of stellar populations in the galaxy given by the SAURON data \citep{kun10} are 12--14 Gyr, with metallicities of 1.3--1.4 times solar (with marginal evidence for lower central metallicities) though this only covers the galaxy interior to and at $R_e$ (11.5 arcseconds). The $(B-V)$ colour profile varies from $\sim 0.9$ in the disk region to $\sim 1.15$ in the exterior regions, which is a bigger colour change than implied by the marginal metallicity gradient in the SAURON data (according to the Tables in \citet{bre94} the change in metallicity will only raise the $B-V$ colour by a few hundredths of a magnitude). However, the GALFIT results do show a rather bluer colour for the disk ($B-V \approx 0.7$ compared with $\approx 1.0$ for the much larger and brighter S\'ersic component), and the disk is progressively less prominent in the 2-component fit from B to I. The models of \citet{bre94} give $B-V$ colours of 1.03--1.13 for a 12.5 Gyr population with metallicity 1.0--2.5 times solar, slightly redder than the S\'ersic component here but compatible with the colours in the inner colour profile. Meanwhile the disk component colours can be reproduced with models of age 1-2 Gyr for solar metallicity and 1 Gyr for 2.5 times solar metallicity. So this may imply relatively young stellar populations in the disk, masked by the fact that the elliptical `bulge' component dominates at all radii, and because the SAURON data use circular apertures, and now revealed by the ability of GALFIT to separate the light from the two components. This may therefore be evidence for star formation produced by a merger 1-2 Gyr ago and now detectable through a bluer colour in the stellar population of the disk component separated by GALFIT. In conclusion, we have detected a tidal filament in the vicinity of NGC 4660 using the enhanced Schmidt film material, and detected brightness peaks of the filament in subsequent deep CCD data. This is not a galaxy expected to have undergone a tidal interaction/merger recently, and the SAURON data indicate only old populations, and a disk coeval with, or only slightly younger than, the elliptical component of the galaxy. However the two-component fit shows that the disk component has bluer colours, indicating that the disk may have been formed in a merger 1-2 Gyr ago, or that this interaction caused enhanced star formation in a pre-existing disk. There are brighter concentrations within the filament which resemble Tidal Dwarf Galaxies, although they are at least 40 times fainter. These may represent faint, evolved TDGs or may be another class of object. A previously detected stripped satellite south of the nucleus it detected in our residual image and may imply that the filament is really a tidal stream produced by perigalactic passages of this satellite. It would be interesting to carry out deep H I observations in the region of the filament, and its brightness peaks, to determine its gas content, and to obtain {\it Spitzer} images of the filament, mapping the old stellar popualtions to fainter levels. \acknowledgments We thank CONACyT for providing financial help for the project ``Stellar populations in early-type galaxies''. We thank the staff of the UKST for obtaining the Schmidt films between 1991 and 1994, the APM group for scanning the films, and Mike Irwin for helpful advice at that time. The co-addition of the APM scans of the Schmidt films was carried out at the Manchester STARLINK node by Dave Berry and Thanassis Katsiyannis. We thank John Meaburn for having the idea of stacking Schmidt plates. We also thank all the staff in San Pedro M\'artir for all their technical support, and Violeta Guzman and Nely Cerda for their help with the observations. Thanks to Andr\'es Rodr\'\i guez for computer management and software help. We thank the anonymous referee for many thorough and helpful suggestions, leading to a great improvement in this paper.
daee9918f4f93de04946c46745d1c314288d3d3e
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section*{Supplemental Material} In this Supplemental Material, we provide more numerical data for the ground-state entanglement entropy and entanglement spectrum. \subsection*{Ground-state entanglement entropy} In the main text, we have discussed the ground-state entanglement entropy $S(\overline{\rho})$ obtained by averaging the density matrices of the three ground states, i.e., $\overline{\rho}=\frac{1}{3}\sum_{i=1}^3|\Psi_i\rangle\langle\Psi_i|$. Now we compute the corresponding result $S(|\Psi_i\rangle)$ and its derivative $dS(|\Psi_i\rangle)/dW$ of the three individual states. The sample-averaged results are shown in Fig.~\ref{Spsi}. The data of three individual states have some differences, but are qualitatively the same: for all of them, the entanglement decreases with $W$, and the derivative with respect to $W$ has a single minimum that becomes deeper for larger system sizes. For the finite systems that we have studied, the location of the minimum does depend somewhat on the individual states, but the value does not deviate much from $W=0.6$. To incorporate the effects of all of the three states, we compute the mean $\overline{S}=\frac{1}{3}\sum_{i=1}^3 S(|\Psi_i\rangle)$. This is an alternative averaging method to the one ($\overline{\rho}=\frac{1}{3}\sum_{i=1}^3|\Psi_i\rangle\langle\Psi_i|$) that we use in the main text. The sample-averaged results are shown in Fig.~\ref{Sbar}. The minimum of $\langle d\overline{S}/dW\rangle$ is located at $W\approx0.6$ for $N=5-9$ electrons [Fig.~\ref{Sbar}(c)], and its depth diverges as $h\propto N^{1.33}$ with the system size [Fig.~\ref{Sbar}(d)]. The scaling $d\overline{S}/dW\propto N^{\frac{1}{2}+\frac{1}{2\nu}}f'[N^{\frac{1}{2\nu}}(W-W_c)]$ suggests $\nu\approx 0.6$. $\langle\overline{S}\rangle$ agrees with an area law at all $W$'s, and the entanglement density starts to drop at $W\approx0.4$ [Fig.~\ref{Sbar}(b)]. All of these results are very similar to those shown in Figs.~1 and 2 in the main text. This means both averaging methods, i.e., $\overline{\rho}=\frac{1}{3}\sum_{i=1}^3|\Psi_i\rangle\langle\Psi_i|$ and $\overline{S}=\frac{1}{3}\sum_{i=1}^3 S(|\Psi_i\rangle)$, can identify the ground-state phase transitions and give the same critical $W$. However, we observe larger finite-size effects of $h$ and error bars in $\langle d\overline{S}/dW\rangle$ (especially at small $W$). \begin{figure*} \centerline{\includegraphics[width=\linewidth]{entropy_psi.pdf}} \caption{$\langle S(|\Psi_i\rangle)\rangle$ and $\langle dS(|\Psi_i\rangle)/dW\rangle$ for (a,d) $|\Psi_1\rangle$, (b,e) $|\Psi_2\rangle$ and (c,f) $|\Psi_3\rangle$, where $|\Psi_1\rangle$, $|\Psi_2\rangle$ and $|\Psi_3\rangle$ are the three states with ascending energies in the ground-state manifold. Here we averaged $20000$ samples for $N=4-7$, $5000$ samples for $N=8$, and $800$ samples for $N=9$ electrons. The data at $W=\infty$, i.e., the noninteracting limit are also given.} \label{Spsi} \end{figure*} \begin{figure} \centerline{\includegraphics[width=\linewidth]{entropy_psi_avg.pdf}} \caption{We measure the ground-state entanglement by $\overline{S}=\frac{1}{3}\sum_{i=1}^3 S(|\Psi_i\rangle)$. (a) $\langle\overline{S}\rangle$, (b) the entanglement density $\alpha$, and (c) $\langle d\overline{S}/dW\rangle$ versus the disorder strength $W$. (d) The depth $h$ of $\langle d\overline{S}/dW\rangle$ versus the number of electrons $N$ on a double logarithmic plot. The linear fit (dashed line) shows $h\propto N^{1.33}$. Here we averaged $20000$ samples for $N=4-7$, $5000$ samples for $N=8$, and $800$ samples for $N=9$ electrons. The data at $W=\infty$, i.e., the noninteracting limit are also given in (a) and (b).} \label{Sbar} \end{figure} \subsection*{Ground-state entanglement spectrum (ES)} In the main text, we consider the density of states (DOS) $\overline{D}(\xi)$ and level statistics $\overline{P}(s)$ of the ES averaged over three ground states. We find that the results of each individual state are almost the same as those obtained by averaging over three ground states, which justifies the procedures of doing an average. Here, we demonstrate the results [$D_1(\xi)$ and $P_1(s)$] of $|\Psi_1\rangle$ for completeness (Fig.~\ref{oespsi1}). The results for $|\Psi_2\rangle$ and $|\Psi_3\rangle$ are almost the same as $|\Psi_1\rangle$, thus we do not show them here. \begin{figure} \centerline{\includegraphics[width=\linewidth]{oes_N_9_psi1_v2.pdf}} \caption{The sample-averaged DOS $\langle D_1(\xi)\rangle$ and the level-spacing distribution $P_1(s)$ of the ground-state ES below $\xi=40$ for $|\Psi_1\rangle$ of $N=9$ electrons at (a) $W=0.4$, (b) $W=0.6$, (c) $W=1$, (d) $W=10$, (e) $W=100$ and (f) $W=\infty$. At each $W$, we choose three windows to compute $P_1(s)$, plotted versus $s$ in the insets. The blue crosses correspond to numerical data, while the red lines give the theoretical prediction for the Gaussian unitary ensemble (GUE), semi-Poisson (S.~P.) and the Poisson distribution, for which $P(s)=\frac{32}{\pi^2}s^2 e^{-\frac{4}{\pi}s^2}$, $P(s)=4se^{-2s}$ and $P(s)=e^{-s}$, respectively. Data from $800$ realizations of disorder. } \label{oespsi1} \end{figure} We should also consider the problem of numerical noise in the ES obtained by singular value decomposition of the many-body eigenstates. The machine precision for double precision variables is $2^{-53}$. This implies that those singular values $\sqrt{\xi}$ below $2^{-53}$ have the danger to be ruined by the numerical noise, which corresponds to $\xi=-\ln2^{-53\times 2}\approx 73.5$ in the ES. Considering that the entries of the many-body eigenstates are complex numbers (two double precision variables) in our systems and the many-body eigenstates themselves also contain numerical error from Lanczos iterations, the numerical noise in the ES may appear at lower $\xi$. In order to detect the critical $\xi$ at which the machine precision problem starts to dominate, we check the DOS $\overline{D}(\xi)$ of the ES at different disorder strengths. We expect that the ES levels generated by numerical noise always assemble around the same energy. This will correspond to a peak in the DOS that does not move with the change of disorder strength. In Fig.~\ref{oesdos}, we indeed observe such a situation deeply in the localized phase. There is always a peak around $\xi\approx50$ that does not move for $W=100,1000$ and $\infty$, meaning that the machine precision problem has occured at these disorder strengths. Therefore, we only focus on those ES levels with $\xi\leq40$ for safety. \begin{figure} \centerline{\includegraphics[width=\linewidth]{oesdos_N_9_v2.pdf}} \caption{The sample-averaged DOS $\langle\overline{D}(\xi)\rangle$ of the ground-state ES of $N=9$ electrons at $W=0.1,1,10,100,1000$ and $\infty$. $\langle\overline{D}(\xi)\rangle$ is averaged over the three ground states using $800$ samples.} \label{oesdos} \end{figure} \end{document}
5fab997cf8b27e973a64a061be0aa50cb3e2044f
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The shallow water equations (Saint-Venant \cite{Saint-Venant1871}) are a common $(d - 1)$ approximation to the $d$--dimensional Navier-Stokes equations to model incompressible, free surface flows. Due to the ability of high-order Galerkin methods to keep dissipation and dispersion errors low (Ainsworth et al.\ \cite{ainsworthEtAl2006}) and their flexibility with arbitrary geometries and {\it hp}-adaptivity, these methods are proving their mettle for solving the shallow water equations (SW) in the modeling of non-linear waves in different geophysical flows (Ma \cite{ma1993}, Iskandarani \cite{iskandaraniEtAl1995}, Taylor et al.\ \cite{taylorTribbia1997}, Giraldo \cite{giraldo2001}, Giraldo et al.\ \cite{giraldoHesthavenWarburton2002}), Dawson and Aizinger \cite{dawsonAizinger2005}, Kubatko et al.\ \cite{kubatkoEtAl2006}, Nair et al.\ \cite{nairThomasLoft2007}, Giraldo and Restelli \cite{giraldoRestelli2010}, Xing et al.\ \cite{xingZhangShu2010}, K\"arn\"a et al.\ \cite{karnaEtAl2011}, Hendricks et al.\ \cite{hendricksKoperaGiraldo2015}, Hood \cite{karolineTHESIS2016}). Although it is typically assumed that high-order Galerkin methods are not strictly necessary, they do offer many advantages over their low-order counterparts. Examples include their ability to resolve fine scale structures and to do so with fewer degrees of freedom, as well as their strong scaling properties on massively parallel computers (M\"uller et al.\ \cite{mullerEtAl2016}, Abdi et al.\ \cite{abdiEtAl2016}, Gandhem et al.\ \cite{gandhamEtAl2015}). High-order methods are often attributed with some disadvantages as well. For example, they are constrained by small time-steps. To overcome this restriction, we follow K\"arn\"a et al.\ \cite{karnaEtAl2011} and implement an implicit Runge-Kutta scheme based on Giraldo et al.\ \cite{giraldoEtAl2013}. Furthermore, the numerical approximation of non-linear hyperbolic systems via high-order methods is susceptible to unphysical Gibbs oscillations that form in the proximity of strong gradients such as those of sharp wave fronts (e.g. bores). Filters such as Vandeven's and Boyd's \cite{vandeven1991, boyd1996} are the most common tools to handle this problem, as well as artificial diffusion of some sort. However, we noticed that filtering may not be sufficient as the flow strengthens and the wave sharpness intensifies. We have recently shown this issue with the solution of the non-linear Burgers' equation by high order spectral elements in \cite[\S 5]{marrasEtAl2015b} and will show how a proper dynamic viscosity does indeed improve on filters for the shallow water system as well. To preserve numerical stability without compromising the overall quality of the solution, Pham Van et al.\ \cite{phanVanEtAl2014} and Rakowsky et al.\ \cite{rakowskyEtAl2013} utilized a Lilly-Smagorinsky model \cite{lilly1962, smagorinsky1963}. To account for sub-grid scale effects, viscosity was also utilized in the DG model described by Gourgue et al.\ \cite{gourgueEtal2009} to improve their inviscid simulations. Recently, Pasquetti et al.\ \cite{pasquettiGuermondPopoICOSAHOM2014} stabilized the high-order spectral element solution of the one-dimensional Saint-Venant equations via the entropy viscosity method. Michoski et al.\ \cite{michoskiEtAl2016} compared artificial viscosity, limiters, and filters for the (modal) DG solution of SW, concluding that a dynamically adaptive diffusion may be the most effective means of regularization at higher orders. The diffusion model that we present in this work stems from the sub-grid scale (SGS) model that was proposed by Nazarov and Hoffman \cite{nazarovHoffman2013} to stabilize the linear finite element solution of compressible flows with shock waves. This was later applied to stabilize high-order continuous and discontinuous Galerkin (CG/DG) in the context of stratified, low Mach number atmospheric flows by Marras et al.\ in \cite{marrasEtAl2015b}. This approach is based on the ideas of scales splitting in large eddy simulation, where the scales of physical importance are split into grid resolvable and unresolvable. The unresolved scales are then parameterized via the SGS model at hand ({\it Dyn-SGS}). {\it Dyn-SGS} is applicable to any numerical method and is not limited to Galerkin approximations; its implementation in existing numerical models is straightforward. Stabilization becomes more important as the grid resolution is decreased. The metric that we utilize to evaluate its importance is the spectrum of kinetic energy across wave numbers. To understand the impact of diffusion with respect to the energetic behavior of the numerical solution, we compare the spectra obtained for the stabilized continuous and discontinuous Galerkin solutions against their inviscid counterparts. As viscosity is accounted for, the solution improves at coarser resolutions for both CG and DG. In the case of DG, however, dissipation is already built-in by the flux computation across elements. For this reason, the amount of artificial diffusion that DG requires for stabilization is smaller than the one required by CG. This property of DG has been exploited to build implicit Large Eddy Simulation (ILES) models (e.g., Uranga et al.\ \cite{urangaEtAl2011}). ILES relies on the numerical dissipation of the discretization method to model the effects on the unresolved, sub-grid scale eddies rather than using an explicit sub-grid scale (SGS) model, under the requirement of high-resolution \cite{rider2006, drikakisEtAl2007}. The MPDATA method of Smolarkiewicz \cite{smolarkiewicz1983,smolarkiewicz1984} is an example of such an ILES model used for modeling geophysical flows. Finally, there is a known difficulty in including wetting and drying algorithms (typically designed for low-order methods) while preserving high-order accuracy. The application of wetting/drying with discontinuous Galerkin using low-order Lagrange polynomials can be found in, e.g., Bunya et al.\ \cite{bunyaEtAl2009}, Vater et al.\ \cite{vaterEtAl2015}, K\"arn\"a et al.\ \cite{karnaEtAl2011}, or Gourgue et al . cite{gourgueEtal2009}, and using Bernstein polynomials up to order three in Beisiegel and Behrens \cite{beisiegelBehrens2015}. The positivity preserving limiter of Xing et al.\ \cite{xingZhangShu2010} (Xing-Zhang-Shu limiter from now on) was designed for high-order discontinuous Galerkin to solve this problem in particular. Because it is mass conservative and preserves the global high order accuracy of the solution, we implemented it in our model and extended it to continuous Galerkin as well. This limiter is guaranteed to work if and only if the mean water height is positive. Its designers recommend using the total variation bounding (TVB) limiter of Shu \cite{shu1987, cockburnShu1989} to make the solution positive definite. Because the TVB limiter requires a bound on the derivative of the solution, which we do not know a priori, we rely on the artificial dissipation introduced above instead. Looking at the larger picture of things, the ever increasing interest in the high-order accuracy of inundation models stems from a long list of coastal disasters in the last 10 years alone. From the 2004 tsunami in the Indian Ocean that caused 280,000 deaths, to the 2011 T\-ohoku tsunami, followed by Superstorm Sandy in 2012, the devastating 2013 typhoons in the Philippines and India, and the 2014 hurricane Odile; the list seems to be getting longer rapidly as the global climate patterns are changing. To study the impact of similar events in the future, coastal planners around the world are more and more relying on numerical tools such as the one that we present in this paper. The rest of the paper is organized as follows. The shallow water equations are defined in \S \ref{equationsSct}. Their numerical solution is described in \S \ref{discreteSCT}. The inundation model is described in \S \ref{wetDrySECT} and is followed by the derivation of the SGS model in \S \ref{stabilizationSCT}. Numerical results are reported in \S \ref{testsSCT}. The conclusions are drawn in \S \ref{conclusionsSCT}. \section{Equations} \label{equationsSct} Let $\Omega\in\mathbb{R}^d$ be a fixed domain of space dimension $d$ with boundary $\Gamma$ and Cartesian coordinates ${\bf x}=x$ in 1D and ${\bf x}=[x,y]$ in 2D and let $t\in\mathbb{R}^+$ identify time. Let $z$ always identify the direction of gravity. The absolute free surface level and bathymetry are identified by the symbols $H_s(t,{\bf x})$ and, respectively, $H_b({\bf x})$, so that $H=H_s + H_b$. The viscous shallow water equations are then given as: \begin{subequations} \label{SWE} \begin{equation} \label{massEqn} \frac{\partial H}{\partial t} + \nabla\cdot \left(H{\bf u}\right) = \delta\nabla\cdot(\mu_{SGS}\nabla H), \end{equation} \begin{equation} \label{momeEqn} \frac{\partial H{\bf u}}{\partial t} + \nabla\cdot\left(H{\bf u}\otimes {\bf u} + \frac{g}{2}(H^2 - H^2_{b}) {\bf I} \right) + g H_s\nabla\cdot\left(H_{b} {\bf I} \right) = \nabla\cdot \left(H_s\mu_{SGS}\nabla{\bf u}\right), \end{equation} \end{subequations} where $g=9.81\,{\rm m\,s^{-2}}$ is the acceleration of gravity, ${\bf I}$ is the $d\times d$ identity matrix, and $\mu_{SGS}$ is the dynamic dissipation coefficient to be defined shortly. In (\ref{massEqn}), the $\delta$ coefficient defines whether viscosity is turned on ($\delta=1$) or off ($\delta=0$) in the continuity equation. We will return to this topic shortly. \section{Space and time discretization} \label{discreteSCT} High-order spectral element (SEM or CG, for continuous Galerkin) and discontinuous Galerkin (DG) approximations on quadrilateral elements are used to discretize Eq.\ (\ref{SWE}). The numerical model used in this paper is in fact the two-dimensional version of the NUMA model described in Giraldo et al.\ \cite{giraldoHesthavenWarburton2002} and in Abdi and Giraldo \cite{abdiGiraldo2016}. Furthermore, the model is derived from the model described in Kopera and Giraldo \cite{koperaGiraldo2013a, koperaGiraldo2013b} which is here used for the shallow water equations. The solution is advanced in time using a fully implicit Runge-Kutta scheme (see \S \ref{sec:time:integration}). \subsection{Spectral element and discontinuous Galerkin approximations} We leave the details of the discretization to the work of Giraldo et al. \cite{giraldoHesthavenWarburton2002} and Kopera and Giraldo \cite{koperaGiraldo2013a, koperaGiraldo2013b}. Nonetheless, we introduce some notation that we are going to use later in the paper. To solve the shallow water equations by element-based Galerkin methods on a domain $\Omega$, we proceed by defining the weak form of (\ref{SWE}) that we first recast in compact notation as \begin{equation} \label{compactSW} \frac{\partial {\bf q}}{\partial t} + \nabla\cdot{\bf F}({\bf q}) = {\bf S}({\bf q}), \end{equation} where ${\bf q} = [H, H{\bf u}]^{\rm T}$ is the transposed array of the solution variables and ${\bf F}$ and ${\bf S}$ are the flux and source terms. In the case of spectral elements, the space discretization yields the semi-discrete matrix problem \begin{equation} \label{matrixprobCG} \frac{\partial {\bf q}}{\partial t} = \widehat{\bf D}^{\rm T}{\bf F}({\bf q}) + {\bf S}({\bf q})\,, \end{equation} where, for the global mass and differentiation matrices, ${\bf M}$ and ${\bf D}$, we have that $\widehat{\bf D}={\bf M}^{-1}{\bf D}$. The global matrices are obtained from their local counterparts, ${\bf M}^e$ and ${\bf D}^e$, by direct stiffness summation, which maps the local degrees of freedom of an element $\Omega^h_e$ to the corresponding global degrees of freedom in $\Omega^h$, and adds the element values in the global system. By construction, ${\bf M}$ is diagonal (assuming inexact integration), with an obvious advantage if explicit time integration is used. In the discontinuous Galerkin approximation, the problem at hand is solved only locally, and unlike the case of spectral elements, the flux integral that stems from the integration by-parts must be discretized as well. The element-wise counterpart of the matrix problem (\ref{matrixprobCG}) is then written as: \begin{equation} \label{matrixprobDG} \frac{\partial {\bf q}^e}{\partial t} = - (\widehat{\bf M}^{\Gamma,e})^{\rm T}\breve{\bf F}({\bf q}^e) + (\widehat{\bf D}^e)^{\rm T}{\bf F}({\bf q}^e) + {\bf S}({\bf q}^e), \end{equation} where we obtain $\widehat{\bf M}^{\Gamma,e}=({\bf M}^e)^{-1}{\bf M}^{\Gamma,e}$ from the element boundary matrix, ${\bf M}^{\Gamma,e}$, and the element mass matrix, ${\bf M}^e$. Out of various possible choices for the definition of the numerical flux $\breve{\bf F}({\bf q})$ in Eq.\ (\ref{matrixprobDG}), we adopted the Rusanov flux. The Laplace operator of viscosity is approximated using the Symmetric Interior Penalty method (SIP; the reader is referred to Arnold \cite{arnold1982} for details on its definition). \subsection{Time integration\label{sec:time:integration}} Equation \eqref{matrixprobCG} is integrated in time by an implicit Runge-Kutta (RK) scheme that corresponds to the implicit part of the implicit-explicit scheme used in \cite{giraldoEtAl2013} (see also \cite{Butcher_1999}). The method coefficients in standard ($A=a_{ij},\,b,\,c$) tableaux form are the following \newcommand\ST{\rule[-0.75em]{0pt}{2em}} \begin{align} \label{eq:ARK:s3:p2:q2:LSTABLE} & \begin{array}{c|ccc} \ST 0&0\\ \ST 2-\sqrt{2}&1-\frac{1}{\sqrt{2}}&1-\frac{1}{\sqrt{2}}\\ \ST 1&\frac{1}{2\sqrt{2}}&\frac{1}{2\sqrt{2}}&1-\frac{1}{\sqrt{2}}\\ \cline{1-4} \ST &\frac{1}{2\sqrt{2}}&\frac{1}{2\sqrt{2}}&1-\frac{1}{\sqrt{2}} \end{array}\,,\qquad \begin{array}{c|c}c&A\\ \hline &b\end{array}. \end{align} Scheme \eqref{eq:ARK:s3:p2:q2:LSTABLE} is a three stage second order explicit-first-stage singly diagonally implicit RK (ESDIRK) scheme. This scheme has desirable accuracy and stability properties: $(i)$ all stages are second order accurate, $(ii)$ it is stiffly accurate and $L$-stable, and $(iii)$ it is strong-stability-preserving (SSP) \cite{gottlieb2001strong} with SSP coefficient of 2. These properties allow us to take large time-steps with high accuracy as well as alleviate the instability issues associated with sharp solution gradients \cite{gottlieb2001strong}. The two-dimensional tests presented later in this paper showed to be the most demanding in terms of stability constraints. Method \eqref{eq:ARK:s3:p2:q2:LSTABLE} allows us to gain up to one order of magnitude in terms of maximum admissible advective CFL when compared to an explicit method (explicit part of ARK3, \cite{Kennedy_2001}). In particular, the explicit four-stage Runge-Kutta solution of the solitary wave against one isolated obstacle described in \S \ref{OneislandTest} preserved stability for up to CFL=0.21 using both CG and DG approximations. Although we were not able to use arbitrarily large time-steps with the ESDIRK with the current implementation (we will address this issue in a future work), we resolved the same problem at CFL = 1.8. Schemes with a subset of these properties are employed by K\"{a}rn\"{a} et al.\ \cite{karnaEtAl2011} and shown to be robust in this context. Method \eqref{eq:ARK:s3:p2:q2:LSTABLE} used in this study satisfies all properties ($i$-$iii$). Computationally, at each of the two implicit stages we have to solve a nonlinear equation ${\bf G}({\bf Q}^{(i)})=0$, where ${\bf Q}^{(i)}$ are the stage values, $i=1,2$. We do so by using Newton iterations with a stopping criterion based on the relative decrease in the residual; that is, stop at iteration $k$ if $||{\bf G}({\bf Q}^{(i)}_k)||/||{\bf G}({\bf Q}^{(i)}_0)||<Tol_N$. At each Newton iteration we have to solve a linear system ${\bf J}({\bf Q}^{(i)}_k - {\bf Q}^{(i)}_{k-1}) = - {\bf G}({\bf Q}^{(i)}_{k-1})$, where ${\bf J}$ is the Jacobian matrix of ${\bf G}({\bf Q}^{(i)})$. We approximate the Jacobian using directional finite differences and iterate with the generalized minimal residual (GMRES) method, which is effectively a Jacobian-free Newton--Krylov method \cite{knoll2004jacobian}. The GMRES stopping criterion is also based on the relative residual. The first stage is explicit and equal to the last stage of the previous step, effectively making it a two-stage method, which saves some computational time. \section{The Dyn-SGS model for the shallow water equations} \label{stabilizationSCT} There are different ways to derive the viscous model described by Eq.\ (\ref{SWE}) from the inviscid Saint-Venant equations. Analogous to our previous work on the large eddy simulation of stratified atmospheric flows \cite{marrasEtAl2015b}, the current model builds on the separation between grid resolved (indicated as $\overline{f}({\bf x})$ for any quantity $f({\bf x})$) and unresolved (sub-grid) scales. The unresolved scales are modeled via the eddy viscosity scheme described in this paper. Given an element $\Omega_e$ of order $N$ and with side lengths $\Delta x, \Delta y$ of comparable orders of magnitude, we define the following characteristic length (and hence filter width \cite{sagautBook}): \[ \overline\Delta = \min{\left(\Delta x, \Delta y\right)}/(N+1). \] The value of $\Delta$ sets the size of the smallest resolvable scales in the same way as cut-off filters do in large eddy simulation models. The application of this filter to the continuity equation (\ref{massEqn}) results in the presence of an additional sub-grid term on the right-hand side. It is often debated whether diffusion should be applied to the continuity equation \cite{pasquettiGuermondPopoICOSAHOM2014, gerbeauEtAl2001,Guermond_Popov_2014}; clearly, should the discrete viscous operator not be conservative, mass dissipation should not be used. However, by relying on spectral elements with integration by parts of the second-order diffusion operator, the discrete viscous operator is conservative, as shown in \cite{gubaEtAl2014}. To get a sense of how necessary a stabilized continuity equation may be, we will show a few results for both conditions in \S \ref{OneislandTest}. Scale separation in the momentum equation yields a new equation that includes the gradient of the quantity \begin{equation} \label{tauDef} {\bm \tau}^{SGS} \approx \overline{H} \mu_{SGS} \nabla\widetilde{\bf u}, \end{equation} where the $\widetilde{\bf u}$ indicates sub-grid velocity. The coefficient $\mu_{SGS}$ is defined element-wise and is given as: \begin{equation} \label{mun} \mu_{SGS} = \max\left(0.0, \min (\mu_{\rm max}|_{\Omega_e}, \mu_{{\rm res}}|_{\Omega_e} ) \right), \end{equation} where \begin{equation} \label{mu1} \mu_{{\rm res}}|_{\Omega_e} = \overline{\Delta}^2 \, \max \left(\frac{\| R(H)\|_{\infty,\Omega_e}}{\| H - \widehat{H}\|_{\infty,\Omega}} , \frac{\| R(H{\bf u})\|_{\infty,\Omega_e}}{\| H{\bf u} - \widehat{H \bf u} \|_{\infty,\Omega}} \right) \end{equation} and \begin{equation} \label{mumax} \mu_{\rm max}|_{\Omega_e} = 0.5\overline{\Delta}\left\Vert |{\bf u}| + \sqrt{g H_s} \right\Vert_{\infty,\Omega_e}. \end{equation} In (\ref{mu1}, \ref{mumax}), $\ \widehat{\cdot}\ $ indicates the spatially averaged value of the quantity at hand over the global domain $\Omega$, the norms $\|\cdot \|_{\infty,\Omega}$ at the denominator are used to preserve the physical dimension of the resulting equation, and $R(H, H{\bf u})$ are the residuals of the inviscid governing equations. At each time-step, the residuals are known. The presence of $R$ makes the artificial diffusion mathematically consistent\footnote{In the context of stabilization, {\it mathematical consistency} is the property of artificial diffusion such that it vanishes when the residual is zero.}. The quantity $|{\bf u}| + \sqrt{g H_s}$ in (\ref{mu1}, \ref{mumax}) is the maximum wave speed. \paragraph{\bf Remark 1: physical dimensions} We would like to emphasize the necessity for the physically correct dimensions of diffusion. This is an important issue that is often underestimated and not accounted for in the design of artificial diffusion methods for stabilization purposes.\\ \section{Wetting and drying interface} \label{wetDrySECT} The dry regions are handled by adding an infinitesimal layer of water on the dry surfaces. We utilized a threshold value of 1e-3 m\footnote{One millimeter of water is physically negligible.} for all the tests that we ran. Furthermore, the limiter of Xing et al.\ \cite{xingZhangShu2010} is applied on the velocity and water depth. Even at high order, this approach is suitable for wetting and drying problems solved via both continuous and discontinuous Galerkin methods. Albeit simple, the Xing-Zhang-Shu limiter works well. However, it is guaranteed to work if and only if the mean water height is positive. Xing et al.\ \cite{xingZhangShu2010} recommend using the total variation bounding (TVB) limiter of Shu \cite{shu1987, cockburnShu1989} to make the solution positive definite. We rely on artificial dissipation to achieve this as the TVB limiter requires a bound on the derivative of the solution, which we do not know a priori. \section{Numerical tests} \label{testsSCT} We verify the correctness of our numerical model through five standard benchmarks in both one- and two-dimensions. In 1D, given the N-wave by Carrier et al.\ \cite{carrierEtAl2003} that mimics a wave generated by an offshore submarine landslide, we compute the solution of the tsunami approaching a sloping beach. See \S \ref{1drunupCarrier}. The numerical solution is evaluated against a set of tabulated surface height and momentum data provided by \cite{tsunamiWorkshopData}. The next 1D problem consists of the oscillation of a flat lake in a parabolic bowl \cite{gallardoEtAl2007,xingZhangShu2010}. The existence of an analytic solution to this problem allows us to measure the accuracy of the numerical solutions. See \S \ref{gridConvergence}. In addition, various 1D wetting and drying test cases were analyzed in the thesis of Hood \cite{karolineTHESIS2016} to assess our model. In 2D, we analyzed the results for the following tests. The vertical oscillation of a column of water in an axisymmetric paraboloid with analytic solution \cite{thacker1981}. We describe this problem in \S \ref{2dparaboloidTest} to test the inundation against the two-dimensional effects of a varying topography. The second two-dimensional test involves the flooding in a closed channel as presented by Gallardo et al.\ \cite{gallardoEtAl2007} and by Xing and Zhang \cite{xingZhang2013}. No exact solution is presented for this test, but the results can be compared against previous studies. This is shown in \S \ref{3islands}. We conclude the set of 2D problems with the simulation of the flow in a two-dimensional channel with an isolated obstacle \cite{synolakis1987}. We use this test to gain insight into the practical -- other than theoretical as stated in \cite{gerbeauEtAl2001} -- need to include viscosity by analyzing the energetics of the solutions. The test is run during a sufficiently long time to assess the robustness of the inundation and viscosity schemes to handle fast motion with interacting waves that impinge against a steep obstacle. See \S \ref{OneislandTest}. The description and analysis of more tests can be found in the collection recently compiled by Delestre et al.\ \cite{delestreEtAl2014}. \subsection{1D tsunami runup over a sloping beach} \label{1drunupCarrier} The runup of a long wave on a uniformly sloping beach was originally proposed at the third international workshop on long-wave runup models \cite{tsunamiWorkshopData}. The one-dimensional computational domain is defined as $\Omega = x = [-500, 50000]$ m. The dry initial beach is 500 m long. The initial waveform was defined by Carrier et al.\ in \cite{carrierEtAl2003} for an $L=8$ m domain as: \begin{equation} \label{carrierWaveEQN} \eta = a_1 \exp\{ -\hat{k}_1(x - \hat{x}_1)^2\} - a_2\exp\{ \hat{k}_2(x - \hat{x}_2)^2 \}, \end{equation} with constants $(a_1,a_2,\hat{k}_1,\hat{k}_2, \hat{x}_1, \hat{x}_2) = (0.006, 0.018, 0.4444, 4.0, 4.1209, 1.6384)$. To scale the wave to the $L=50000$ m long domain used for the current test, the scaling factor $\delta = L/8$ is introduced and Eq.\ (\ref{carrierWaveEQN}) is re-expressed with respect to $x_{1,2} = \hat{x}_{1,2} \delta$ and $k_{1,2} = \hat{k}_{1,2}/\delta^2$, with larger amplitudes $(a_1, a_2)=(3.0, -8.8)$. The initial wave is plotted in Fig.\ \ref{runuptinitialwave}. \begin{figure} \centering \includegraphics[width=0.7\textwidth]{./runup-2500x4th-visc15-filter_Htot_INITIAL.png} \caption{1D tsunami runup over a sloping beach: initial N-wave.} \label{runuptinitialwave} \end{figure} The solutions at times $t=[0,\, 110,\, 220]$ s are plotted in Fig.\ \ref{runupSolution220s_2500el} and are compared against the tabulated data available in \cite{tsunamiWorkshopData}. Fig.\ \ref{runupSolution220s_2500el} shows that the effect of diffusion on the water surface solution is clearly negligible. This can be explained by looking at the structure and values of $\mu_{SGS}$ in Fig.\ \ref{runup1D_SGS}. With a water surface that is smooth almost everywhere, the dynamic diffusion coefficient is so small that its effect becomes minimal. We will see later that this will not be the case in problems with a greater degree of irregularity of the surface. We plot the space-time evolution of the shoreline in Fig.\ \ref{runupXTspaceFullDomain} where the wave elevation and total depth are plotted in the proximity of the shore. The dashed red curve represents the tabulated shoreline \cite{tsunamiWorkshopData}. By direct comparison with Carrier's results \cite{carrierEtAl2003}, the patterns of the water surface elevation ($\eta({x,t})$) and total water depth are in full agreement. \begin{figure} \centering \includegraphics[width=0.495\textwidth]{./SGS_runup-1250x4th-visc15-filter_Htot_t160s.png} \includegraphics[width=0.495\textwidth]{./SGS_runup-2500x4th-visc15-filter_Htot_t160s.png}\\ \includegraphics[width=0.495\textwidth]{./SGS_runup-1250x4th-visc15-filter_Htot_t175s.png} \includegraphics[width=0.495\textwidth]{./SGS_runup-2500x4th-visc15-filter_Htot_t175s.png}\\ \includegraphics[width=0.495\textwidth]{./SGS_runup-1250x4th-visc15-filter_Htot_t220s.png} \includegraphics[width=0.495\textwidth]{./SGS_runup-2500x4th-visc15-filter_Htot_t220s.png} \caption{1D tsunami runup over a sloping beach. Computed and exact solutions at $t=[160,175,220]$ s. Left column: $\Delta x \approx 10$ m (1250 elements of order $4$). Right column: $\Delta x \approx 5$ m (2500 elements of order $4$). The computed inviscid and viscous (SGS) solutions appear almost perfectly superimposed. The problem is smooth almost everywhere so that the intensity of the dynamic dissipation is minimal. This translates into an almost imperceptible effect of diffusion on the solution. The structure of the dissipation coefficient is plotted in Fig.\ \ref{runup1D_SGS}.} \label{runupSolution220s_2500el} \end{figure} \begin{figure} \centering \includegraphics[width=0.7\textwidth]{./SGSvsH-2500x4th-visc15-filter_t160s.png} \includegraphics[width=0.7\textwidth]{./SGSvsH-2500x4th-visc15-filter_t175s.png} \includegraphics[width=0.7\textwidth]{./SGSvsH-2500x4th-visc15-filter_t220s.png} \caption{1D tsunami runup over a sloping beach. Dynamic $\mu_{SGS}$ (red, dashed line) and water surface (blue, solid line) in the full domain. The effect of diffusion on the solution of Fig.\ \ref{runupSolution220s_2500el} is minimal as the value of the coefficient is indeed very small. The solution is smooth almost everywhere, which is the reason for the very small values of the dynamic diffusion coefficient.} \label{runup1D_SGS} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./runup_5000x4th_CHARACTERISTICS_elevation.png} \includegraphics[width=0.49\textwidth]{./runup_5000x4th_CHARACTERISTICS_with_SHORE_LINE_seaDepth.png} \caption{1D tsunami runup over a sloping beach. Left: wave trajectory in the full 50 km long domain. Right: $x-t$ variation of the water depth in the proximity of the coast. The shoreline is at the interface between the white area (dry shore) and the color shading (water surface). The dashed red curve is the exact shoreline.} \label{runupXTspaceFullDomain} \end{figure} \subsection{Grid convergence rate} \label{gridConvergence} To measure the convergence rate of the model, we compare the computed solutions against the analytic solution of the flow in a one-dimensional parabolic bowl \cite{gallardoEtAl2007,thacker1981}. The parabolic topography is defined as: \[ H_b(x) = h_0\left(\frac{1}{a^2} x^2\right) - 0.5 \] where $h_0=2$ m and $a=1$ in $\Omega=x=[-1,1]$ m. The initial velocity is $u=0\, {\rm m s^{-1}}$ and the water surface begins to oscillate due to gravity only. The solution is computed using 8, 16, 32, 64, and 128 elements of order 4. The CG solution using 128 elements is plotted in Fig.\ \ref{1DparabolicBowl128CG}, where we compare the inviscid solution against the viscous computation at $t=[2.5,\,5,\,10]$ s. The solution preserves its smoothness at all times as long as the resolution is sufficiently fine. At high resolution, it is evident that the CG approximation to this problem does not require stabilization. As the resolution is decreased by a factor of 4 (Fig.\ \ref{1DparabolicBowlCGlowRes}), the inviscid solution begins to develop oscillations although stability is not compromised. Unlike the inviscid one, the stabilized solution preserves the surface flatness although it moves slower due to a viscous source term in the momentum equation. This fact is also reflected in the computation of the normalized $L_2$ error norms plotted in Fig.\ \ref{L2error1DparabolicBowl}. The analytic solution against which the error is calculated is defined for an inviscid flow. A difference between the viscous and inviscid computation is hence expected. The same observation applies to the CG and DG computations alike. The DG solutions to the same problem are plotted in Fig.\ \ref{1DparabolicBowl128DG} and \ref{1DparabolicBowlDGlowRes}. \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_noLES_128_timestep_125.png} \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_LES_128_timestep_125.png}\\ \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_noLES_128_timestep_250.png} \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_LES_128_timestep_250.png}\\ \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_noLES_128_timestep_500.png} \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_LES_128_timestep_500.png}\\ \caption{1D flow in a parabolic bowl. {\bf CG} solution. Left column: inviscid. Right column: viscous. Computed water level (blue, solid line), exact solution (blue, dashed line with open squares), and $\mu_{SGS}$ (red, solid line) at different times using 128 elements of order 4. For visualization, $\mu_{SGS}$ is scaled by a factor of 1000. The solution is sufficiently smooth that the dynamic viscosity is negligible. On the other hand, the limiter$+$wetting/drying scheme is responsible for the positivity of the water surface in the proximity of the moving boundary.} \label{1DparabolicBowl128CG} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_noLES_16_timestep_500.png} \includegraphics[width=0.49\textwidth]{./Parabola_CG_H_LES_16_timestep_500.png} \caption{1D flow in a parabolic bowl. Low resolution {\bf CG} solutions using 16 elements of order 4. Left column: inviscid. Right column: viscous. Computed water level (blue, solid line), exact solution (blue, dashed line with open squares), and $\mu_{SGS}$ (red, solid line) at $t=10$ s. For visualization, $\mu_{SGS}$ is scaled by a factor of 1000.} \label{1DparabolicBowlCGlowRes} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./Parabola1D_L2err_CGLES.png} \includegraphics[width=0.49\textwidth]{./Parabola1D_L2err_DGLES.png} \caption{Normalized $L_2$ error of water surface at $t=10$ s for {\bf CG} (left) and {\bf DG} (right). The $-3$ and $-4$ curves indicate the reference rates. } \label{L2error1DparabolicBowl} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_noLES_128_timestep_125.png} \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_LES_128_timestep_125.png}\\ \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_noLES_128_timestep_250.png} \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_LES_128_timestep_250.png}\\ \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_noLES_128_timestep_500.png} \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_LES_128_timestep_500.png}\\ \caption{1D flow in a parabolic bowl. Like Fig.\ \ref{1DparabolicBowl128CG} but using {\bf DG}.} \label{1DparabolicBowl128DG} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_noLES_16_timestep_500.png} \includegraphics[width=0.49\textwidth]{./Parabola_DG_H_LES_16_timestep_500.png}\\ \caption{Like Fig.\ \ref{1DparabolicBowlCGlowRes}, but using {\bf DG}.} \label{1DparabolicBowlDGlowRes} \end{figure} \subsection{2D oscillation in a symmetrical paraboloid} \label{2dparaboloidTest} This test was defined by Thacker \cite{thacker1981} who computed the exact solution of the problem. The paraboloid of revolution is defined as : \[ H_b(x,y) = h_0\left(1 - \frac{\sqrt{x^2 + y^2}}{a^2}\right) - 0.1 \] in the domain $\Omega=[-2,2]\times [-2,2]\,{\rm m}^2$, with $h_0=0.2$ m. The initial condition is a reversed paraboloid (See top row of Fig.\ \ref{2DparabolicBowlDG}). The variation of the bed slope and the radial symmetry are a perfect test to assess the inundation scheme with two dimensional effects. The inviscid CG and DG solutions are plotted in Figs.\ \ref{2DparabolicBowlCG} and \ref{2DparabolicBowlDG}, respectively, between $t=0$ and $t=10$ s. The radial symmetry of the solution is preserved throughout the simulation, and no spurious modes can be observed in the proximity of the moving shoreline. Although it is not fully visible from the plots, we observed a small discrepancy between the exact and computed solutions that occurs at the very center of the bowl at the latest time. More specifically, the numerical solution tends to move slightly slower than the exact solution along the bowl centerline. This occurs for CG and DG alike. \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./1D_XZ_VIEW_briggs1995_dxdy_H_1D_XY100X100CG_noLES_000.png} \includegraphics[width=0.49\textwidth]{./1D_YZ_VIEW_briggs1995_dxdy_H_1D_XY100X100CG_noLES_000.png}\\ \includegraphics[width=0.49\textwidth]{./1D_XZ_VIEW_briggs1995_dxdy_H_1D_XY100X100CG_noLES_020.png} \includegraphics[width=0.49\textwidth]{./1D_YZ_VIEW_briggs1995_dxdy_H_1D_XY100X100CG_noLES_020.png}\\ \includegraphics[width=0.49\textwidth]{./1D_XZ_VIEW_briggs1995_dxdy_H_1D_XY100X100CG_noLES_040.png} \includegraphics[width=0.49\textwidth]{./1D_YZ_VIEW_briggs1995_dxdy_H_1D_XY100X100CG_noLES_040.png}\\ \caption{2D oscillation in a paraboloid. Inviscid {\bf CG} solution. Left: $x-z$ view. Right: $y-z$ view. From top to bottom: $t=[0,\; 5,\; 10]\,s$. Solution with $100\times 100$ elements of order 4.} \label{2DparabolicBowlCG} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./1D_XZ_VIEW_briggs1995_dxdy_H_1D_XY100X100DG_noLES_000.png} \includegraphics[width=0.49\textwidth]{./1D_YZ_VIEW_briggs1995_dxdy_H_1D_XY100X100DG_noLES_000.png}\\ \includegraphics[width=0.49\textwidth]{./1D_XZ_VIEW_briggs1995_dxdy_H_1D_XY100X100DG_noLES_020.png} \includegraphics[width=0.49\textwidth]{./1D_YZ_VIEW_briggs1995_dxdy_H_1D_XY100X100DG_noLES_020.png}\\ \includegraphics[width=0.49\textwidth]{./1D_XZ_VIEW_briggs1995_dxdy_H_1D_XY100X100DG_noLES_040.png} \includegraphics[width=0.49\textwidth]{./1D_YZ_VIEW_briggs1995_dxdy_H_1D_XY100X100DG_noLES_040.png}\\ \caption{2D oscillation in a paraboloid. Inviscid {\bf DG} solution. Left: $x-z$ view. Right: $y-z$ view. From top to bottom: $t=[0,\; 5,\; 10]\,s$. Solution with $100\times 100$ elements of order 4.} \label{2DparabolicBowlDG} \end{figure} \subsection{2D flooding problem in a channel with three mounds} \label{3islands} This test \cite{kawaharaUmetsu1986,brufauEtAl2002} was used by Xing and Zhang \cite{xingZhang2013} and Gallardo et al.\ \cite{gallardoEtAl2007} to assess, on unstructured grids, the same positivity-preserving method that we are applying here \cite{xingZhangShu2010}. We reproduce their results here to verify its correct implementation in our code. The three mounds in the channel $\Omega=75 \times 30\, {\rm m^2}$, are defined by the function $H_b({\bf x})=\max(0.0, m_1,m_2,m_3)$ where \begin{subequations} \[ m_1=1.0-0.10\sqrt{(x-30.0)^2 + (y-22.5)^2}, \] \[ m_2=1.0-0.10\sqrt{(x-30.0)^2 + (y-7.50)^2}, \] \[ m_3=2.8-0.28\sqrt{(x-47.5)^2 + (y-15.0)^2}, \] \end{subequations} The flooding is triggered by a dam break at $t=0$ s. The evolution of the flooding from $t=0$ to $t=40$ s is shown in Fig. \ref{humpConfig1flow}. We computed the solution using a relatively coarse grid made of $15\times 6$ elements of order 4. The results are in agreement with \cite{gallardoEtAl2007} and \cite{xingZhang2013}: the flow separation is well resolved and occurs after 10 seconds from the dam breaking. After 15 seconds, the first water front has reached the back wall and is fully reflecting back by 20 seconds. By 40 seconds, the flow has almost reached full rest. \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./case1000-configure1-3humps_Htot_t0s.png} \includegraphics[width=0.49\textwidth]{./case1000-configure1-3humps_Htot_t5s.png} \includegraphics[width=0.49\textwidth]{./case1000-configure1-3humps_Htot_t10s.png} \includegraphics[width=0.49\textwidth]{./case1000-configure1-3humps_Htot_t20s.png} \includegraphics[width=0.49\textwidth]{./case1000-configure1-3humps_Htot_t30s.png} \includegraphics[width=0.49\textwidth]{./case1000-configure1-3humps_Htot_t40s.png} \caption{Dam break problem in a closed channel with three mounds. Viscous {\bf CG} solution at $t=[0,\,5,\,10,\,20\,30\,40]$ s. Solution computed using $15\times 6$ elements of order 4 for the domain $\Omega=75\times30\,{\rm m^2}$.} \label{humpConfig1flow} \end{figure} \subsection{2D solitary wave runup and run-down on a circular island} \label{OneislandTest} A solitary wave runup on a circular island was studied in \cite{briggsEtAl1995} at the Waterways Experiment Station of the US Army Corps of Engineers. In this example, the initial wave is modeled via the following analytic definition by Synolakis \cite{synolakis1987}: \begin{equation} \label{synowave} \eta({\bf x}, 0) = \frac{A}{h_0} {\rm sech}^2 \left( \gamma(x - x_c) \right), \end{equation} where $A=0.064$ m is the wave amplitude, $x_c=2.5$ m, $h_0=0.32$ m is the initial still water level, and \[ \gamma = \sqrt{\frac{3A}{4h_0}}. \] The island is a cone given as \[ H_b = 0.93 \left( 1 - \frac{r}{r_c} \right), \quad {\rm if}\; r\leq r_c, \] where $r=\sqrt{ (x - x_c)^2 + (y-y_c)^2}$, $r_c=3.6$ m, and is centered at $(x_c, y_c)=(12.5, 15)$ m. The cone is installed on a flat bathymetry. The fluid is confined within four solid walls. To understand how the proposed diffusion and numerical approximation depend on the grid, we ran the simulation at the four resolutions $\Delta {\bf x} \approx [0.05\,,0.10\,,0.20\,,0.40]$ m. Fig.\ \ref{briggs4Resolutions} shows that the stabilized DG solution is converging to the same solution and the main features of the interacting waves are reproduced almost equally across the four grids. Certainly, the 40 cm grid spacing is the most dissipative, although it is encouraging to see how the important features resolved at 5 cm are still well represented on the coarsest grid. The same observation applies to the CG solution (plot not shown). \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./briggs1995_dxdy_H_5cm_DG_LES_200.png} \includegraphics[width=0.49\textwidth]{./briggs1995_dxdy_H_10cm_DG_LES_200.png} \includegraphics[width=0.49\textwidth]{./briggs1995_dxdy_H_20cm_DG_LES_200.png} \includegraphics[width=0.49\textwidth]{./briggs1995_dxdy_H_40cm_DG_LES_200.png} \caption{{\bf DG} solution of water depth for the single-hill configuration. Results obtained with the four grid resolutions $\Delta{\bf x} = [0.05, 0.10, 0.20, 0.40]$ m (indicated in the figures) using $4^{th}$-order elements. The color bar is cut at 0.25 to preserve the visibility of the smallest features. The dark blue coloring within the region of the cone corresponds to the water depth equal to the threshold water layer of 1e-3 m.} \label{briggs4Resolutions} \end{figure} In Figs.\ \ref{briggs4Resolutions1DsliceXZ_CG}-\ref{briggs4Resolutions1DsliceXZ_DG_UVELO} we plot the projection of the 2D solution on the plane $y=0$. The spurious modes that characterize the water surface in the proximity of the sharpest wave front are fully removed by {\it Dyn-SGS} (Fig.\ \ref{briggs4Resolutions1DsliceXZ_CG}) without weakening the front sharpness. We observe the same in the velocity field plotted in Fig.\ \ref{briggs4Resolutions1DsliceXZ_CG_UVELO}. This is in full agreement with the application of {\it Dyn-SGS} to non-linear wave problems with strong discontinuities, as previously shown in \cite[Figs.\ 16, 17]{marrasEtAl2015b} for the solution of the Burgers' equation. We briefly mentioned above how DG already has built-in dissipation. This is clearly visible from the plot of Fig.\ \ref{briggs4Resolutions1DsliceXZ_DG}; the unstabilized DG solution shows no oscillations except for, at most, some minimal under- and over-shooting. This implies that its residual is so small that the effect of SGS eddy viscosity reduces to a minimum value. This explains why the inviscid and viscous DG solutions look similar. It is important to notice the same space distribution of the CG and DG velocity waves shown in Figs.\ \ref{briggs4Resolutions1DsliceXZ_CG_UVELO} and \ref{briggs4Resolutions1DsliceXZ_DG_UVELO}. This is a visual confirmation that the two solutions present the same dispersion properties, which is to be expected from two analogous numerical approximations to the same problem. This is indicative of a correct implementation of the unified CG/DG framework. \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./1DXZ_VIEW_briggs1995_dxdy_H_5cm_CG_noLES_200.png} \includegraphics[width=0.49\textwidth]{./1DXZ_VIEW_briggs1995_dxdy_H_5cm_CG_LES_200.png} \caption{$x-z$ view of the 2D {\bf CG} solution of water depth $(H)$ for the single-hill configuration. Inviscid (left) against viscous solution using SGS (right). Solutions obtained using $4^{th}$-order elements. The dark blue coloring within the region of the cone corresponds to the water depth equal to the threshold water layer of 1e-3 m.} \label{briggs4Resolutions1DsliceXZ_CG} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./1DXZ_UVELO_VIEW_briggs1995_dxdy_H_5cm_CG_noLES_200.png} \includegraphics[width=0.49\textwidth]{./1DXZ_UVELO_VIEW_briggs1995_dxdy_H_5cm_CG_LES_200.png} \caption{Like Fig. \ref{briggs4Resolutions1DsliceXZ_CG}, but for $u$-velocity component. Inviscid (left) against viscous solution using SGS (right).} \label{briggs4Resolutions1DsliceXZ_CG_UVELO} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./1DXZ_VIEW_briggs1995_dxdy_H_5cm_DG_noLES_200.png} \includegraphics[width=0.49\textwidth]{./1DXZ_VIEW_briggs1995_dxdy_H_5cm_DG_LES_200.png} \caption{$x-z$ view of the 2D {\bf DG} solution of water depth $(H)$ for the single-hill configuration. Inviscid (left) against viscous solution using SGS (right). Solution obtained using $4^{th}$-order elements. The dark blue coloring within the region of the cone corresponds to the water depth equal to the threshold water layer of 1e-3 m. This plot shows the power of DG. Without SGS it still almost captures the bore sharply.} \label{briggs4Resolutions1DsliceXZ_DG} \end{figure} \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./1DXZ_UVELO_VIEW_briggs1995_dxdy_H_5cm_DG_noLES_200.png} \includegraphics[width=0.49\textwidth]{./1DXZ_UVELO_VIEW_briggs1995_dxdy_H_5cm_DG_LES_200.png} \caption{Like Fig. \ref{briggs4Resolutions1DsliceXZ_DG}, but for $u$-velocity component. Inviscid (left) against viscous solution using SGS (right). A comparison of this figure with Fig.\ \ref{briggs4Resolutions1DsliceXZ_CG_UVELO} shows the same dispersion properties of the CG and DG solutions.} \label{briggs4Resolutions1DsliceXZ_DG_UVELO} \end{figure} The solution of the top plot in Fig.\ \ref{briggsCGstableVSunstableH3dview}, was computed with stabilization applied to the continuity ($\delta=1$ in Eq.\ (\ref{SWE})) and momentum equations. When compared against the un-stabilized solution (bottom plot in the same figure), we notice that the features of the fronts of the interacting waves are fully preserved. Furthermore, the fronts are not excessively smeared out as the high frequency modes are removed. The effect of momentum stabilization on the velocity field is shown in Fig.\ \ref{briggsCGstableVSunstableUVELO3dview}. \begin{figure} \centering \includegraphics[width=0.9\textwidth]{./3D_VIEW_briggs1995_dxdy_H_3D_5cm_CG_noLES_168.png} \includegraphics[width=0.9\textwidth]{./3D_VIEW_briggs1995_dxdy_H_3D_5cm_CG_LES_168.png} \caption{{\bf CG} solution for the single-hill configuration. Instantaneous perspective view of the unstabilized (top) and stabilized (bottom) water surface for $\Delta{\bf x} = 5$ cm using $4^{th}$-order elements. The high frequency instabilities are removed by Dyn-SGS without compromising the overall sharpness of the interacting waves. Both solutions are characterized by the same wave features at all scales.} \label{briggsCGstableVSunstableH3dview} \end{figure} \begin{figure} \centering \includegraphics[width=0.9\textwidth]{./3D_VIEW_briggs1995_dxdy_UVELO_3D_5cm_CG_noLES_168.png} \includegraphics[width=0.9\textwidth]{./3D_VIEW_briggs1995_dxdy_UVELO_3D_5cm_CG_LES_168.png} \caption{Like Fig.\ \ref{briggsCGstableVSunstableH3dview}, but surface of horizontal velocity, $u$. For best view of the velocity surface, the view angle is different from the one of Fig.\ \ref{briggsCGstableVSunstableH3dview}.} \label{briggsCGstableVSunstableUVELO3dview} \end{figure} The instantaneous energy spectra of the viscous and inviscid CG and DG solutions at $t=50$ s are plotted in Fig.\ \ref{briggs1995SpectraCGvsDG}. The difference between the CG and DG curves is striking. The viscous and inviscid DG spectra overlap almost fully and show approximately the same decay across the entire spectrum, from a -5/3 slope in the inertial sub-range to a -3 slope in the dissipation wave numbers (refer to \cite{boffettaEcke2012} for a review on two-dimensional flows and their energetics). This is only true as long as the resolution is not too coarse, especially so in the case of CG. At very coarse resolutions ($\Delta {\bf x} \geq 0.4$ m), neither CG or DG can avoid energy from building up in the highest modes unless artificial diffusion is used. The inherent dissipation of DG is no longer sufficient. \begin{figure} \centering \includegraphics[width=0.49\textwidth]{./Time_AVE_energy_spec_kx_2DSW_CG_SGS_vs_NOSGS10cm.png} \includegraphics[width=0.49\textwidth]{./Time_AVE_energy_spec_kx_2DSW_DG_SGS_vs_NOSGS10cm.png} \includegraphics[width=0.49\textwidth]{./Time_AVE_energy_spec_kx_2DSW_CG_SGS_vs_NOSGS20cm.png} \includegraphics[width=0.49\textwidth]{./Time_AVE_energy_spec_kx_2DSW_DG_SGS_vs_NOSGS20cm.png} \includegraphics[width=0.49\textwidth]{./Time_AVE_energy_spec_kx_2DSW_CG_SGS_vs_NOSGS40cm.png} \includegraphics[width=0.49\textwidth]{./Time_AVE_energy_spec_kx_2DSW_DG_SGS_vs_NOSGS40cm.png} \caption{Instantaneous 1D energy spectra of the single hill problem of Fig.\ \ref{briggs4Resolutions} at $t=50$ s. Left: CG with and without viscosity. Right: DG with and without viscosity. From top to bottom the resolution decreases.} \label{briggs1995SpectraCGvsDG} \end{figure} We stated above that $\mu_{SGS}$ is only active where the equation residuals (i.e. gradients) are important. In the case of water waves, this occurs in the proximity of the wave fronts. In Fig.\ \ref{nusgs_timeevolution}, we plot $\mu_{SGS}$ to show its spatial structure and its evolution between $t=0$ and $t=50$ seconds. This plots clearly show how diffusion is equally zero away from the fronts and only activates where really necessary. It may not be so obvious to achieve this by using an artificial diffusion that is not residual-based. To provide a visual correlation between $\mu_{SGS}$ and the wave features, in Fig.\ \ref{H_timeevolution} we plot the stabilized spectral element solution of the water surface at a grid resolution $\Delta x \approx 0.05$ m. \begin{sidewaysfigure} \centerfloat \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_010.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_020.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_030.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_040.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_050.png}\\ \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_060.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_070.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_080.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_090.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_100.png}\\ \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_110.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_120.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_130.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_140.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_150.png}\\ \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_160.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_170.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_180.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_190.png} \includegraphics[trim=2cm 2cm 2cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_SGS_5cm_CG_LES_200.png} \caption{Time evolution of $\mu_{SGS}$ from $t=0$ s to $t=50$ s. The colorscale ranges between 0 (blue) and 5 (red) m$^2/$s. The plotted domain is $\Omega=[0, 25]\times [-15, 15]$ m$^2$.} \label{nusgs_timeevolution} \end{sidewaysfigure} \begin{sidewaysfigure} \centerfloat \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_010.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_020.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_030.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_040.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_050.png}\\ \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_060.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_070.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_080.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_090.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_100.png}\\ \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_110.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_120.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_130.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_140.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_150.png}\\ \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_160.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_170.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_180.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_190.png} \includegraphics[trim=2cm 2cm 2.5cm 0cm, clip=true,width=0.2\textwidth]{./briggs1995_dxdy_H_5cm_CG_LES_200.png} \caption{Time evolution of $H$ from $t=0$ s to $t=50$ s. The color scale ranges between 0.25 (blue) and 0.5 (red) m. The plotted domain is $\Omega=[0, 25]\times [-15, 15]$ m$^2$.} \label{H_timeevolution} \end{sidewaysfigure} \section{Conclusions} \label{conclusionsSCT} We presented the numerical solution of the shallow water equations via continuous and discontinuous Galerkin (CG/DG) methods to model problems involving inundation. Careful handling of the transition between dry and wet surfaces is particularly challenging for high-order methods. The most adopted solution in these cases consists of lowering the approximation order in either the transition zone alone, or in the whole domain. By extending to CG a simple slope limiter originally designed for DG \cite{xingZhangShu2010}, and by combining it with a thin water layer always present in dry regions, we showed that the nominal high-order accuracy of the underlying space approximations was preserved and that the solution smoothness in the proximity of moving boundaries was maintained in one- and two-dimensions. The simplicity of this approach is effective for CG as well as DG. More important, we demonstrated its effectiveness using higher order elements. This limiter, unless embedded with a positivity-preserving scheme for water height, does not prevent the high-order solution from triggering unphysical Gibbs oscillations in the proximity of strong gradients. To overcome this problem, we presented a dynamically adaptive dissipation based on a residual-based sub-grid scale eddy viscosity model ({\it Dyn-SGS}). By numerical examples, we demonstrated the following properties of this model when solving the shallow water equations: $(i)$ It removes the Gibbs oscillations that form in the proximity of sharp wave fronts while preserving the overall accuracy and sharpness of the solution everywhere else. This is possible thanks to the residual-based definition of the dynamic diffusion coefficient. $(ii)$ For coarse grids, it prevents energy from building up at small wave-numbers; this is very important to preserve numerical stability in the flow regimes we are interested in. $(iii)$ The model has no user tunable parameter, which is of great advantage when the model is to be used by an external user. It is important to underline that the natural, built-in dissipation of DG be may be large enough that the contribution of {\it Dyn-SGS} becomes irrelevant. When this happens, the dynamic dissipation detects it from the residual, and hence limits its own strength. Finally, a three stage, second order explicit-first-stage, singly diagonally implicit Runge-Kutta (ESDIRK) time integration scheme was implemented to overcome the small time-step restriction incurred by high-order Galerkin approximations. \section{Acknowledgments} The authors would like to acknowledge the contribution of Haley Lane, who implemented the one-dimensional version of the wetting and drying algorithm used in this work. The authors would also like to acknowledge Karoline Hood who tested the correctness of the implicit solver in her NPS Master's thesis (\cite{karolineTHESIS2016}). The authors are also thankful to Oliver Fringer and the Rogers for discussions regarding coastal flows, and to Stephen R. Guimond for providing his MATLAB functions to compute the energy spectra. FXG acknowledges the support of the ONR Computational Mathematics program, and FXG and EMC acknowledge the support of AFOSR Computational Mathematics. \newpage
d53bbf469ca04bb5fbd7062a450cba2c00479b10
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction and Summary of Results} The problem of admissible functional-classes has been of recent interest in the context of higher-spin (HS) theories \cite{Vasiliev:1990en}. In particular, in \cite{Kessel:2015kna,Boulanger:2015ova} the quadratic interaction term sourcing Fronsdal's equations was extracted from Vasiliev's equations obtaining an expression of the schematic form:\footnote{Notice that in the following we use the schematic notation $\Box^l\sim \dots\nabla_{\mu(l)}\phi\ldots\nabla^{\mu(l)}\phi$. We give precise formulas for the above contractions in the spinor language in the following section in eq.~\eqref{dic}. In this section all formulas are schematic and provide some intuition on their generic structure.} \begin{equation}\label{back} \Box\phi_{\mu(s)}+\ldots=\sum_{l=0}^\infty\frac{j_l}{l!l!}\,\Box^l\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)_{\mu(s)}\,, \end{equation} which is sometime in the literature referred to as pseudo-local or quasi-local, meaning that it is a formal series in derivatives.\footnote{This terminology originates from the fact that formal series allow truncations to finitely many terms which are always local.} The extracted Fronsdal current have coefficients whose asymptotic behaviour is given by $j_l\sim \frac1{l^3}$ for $l\to \infty$ for any choice of the spin $s$. The asymptotic behaviour of the coefficients raised the important question whether the backreaction extracted is or not strongly coupled\footnote{Preliminary questions of this type were raised in \cite{Boulanger:2008tg}.}. Furthermore, a key question whose study was undertaken in \cite{Skvortsov:2015lja,Taronna:2016ats} was whether it is possible to extract the coefficients of the canonical Metsaev vertices with finitely many derivatives \cite{Metsaev:2005ar,Joung:2011ww} from the above tails. Indeed most of the coefficients at the cubic order are unphysical, since they can be removed by local field redefinitions. In this respect, the full list of Metsaev-like couplings was indeed extracted holographically in \cite{Sleight:2016dba} and amounts to a finite number of coefficients for any triple of spins, to be contrasted to the above infinite set (see also \cite{Taronna:2010qq,Sagnotti:2010at} for the analogous string theory computation and corresponding cubic couplings).\footnote{It is important to stress that in a fully non-linear HS theory it is expected that the appropriate field frame which makes HS geometry manifest will entail \emph{all} of the above coefficients. The situation should be similar to the Einstein-Hilbert cubic couplings which are dressed by improvement terms that can be removed by a field redefinition at the cubic order. This is a further key motivation to understand these higher-derivative tails.} Remarkably, the pseudo-local nature of the above currents implies that the only way to relate them to their Metsaev-like counterparts is via a pseudo-local field redefinition of the same schematic form: \begin{align}\label{pseudored} \phi_{\mu(s)}\rightarrow\phi^\prime_{\mu(s)}=\phi_{\mu(s)}+\sum_{l=0}^\infty \frac{a_l}{l!l!}\,\Box^l\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)_{\mu(s)}\,, \end{align} involving a sum over infinitely many terms unbounded in derivatives (i.e. pseudo-local). This result has motivated a renewed interest in the analysis of the admissible functional classes in HS theories. Indeed, an arbitrary pseudo-local redefinition defined in \eqref{pseudored} is sufficient to remove all pseudo-local current interactions \cite{Prokushkin:1999xq,Kessel:2015kna} and some further condition on the coefficients $a_l$ should be imposed on top of quasi-locality. A proposal\footnote{The proposal of \cite{Taronna:2016ats} is based on jet space and on the convergence of the infinite derivative expansion. It turns out that this proposal ensures the invariance of the Witten diagrams under the corresponding admissible field redefinitions.} based on the invariance of the holographic Witten-diagrams was put forward in \cite{Skvortsov:2015lja,Taronna:2016ats}, while in \cite{Vasiliev:2016xui} (see \cite{Vasiliev:2015wma} for further details) it was proposed to study classes of functions in $z$ and $y$ oscillators which are closed under star product multiplication. The aim of this note is to elaborate on these results from various perspectives, and to present the explicit form of the field redefinitions mapping the pseudo-local back-reaction \eqref{back} to its canonical (local) form. In the following we list/summarise some relevant points of this analysis, together with the main results of this note, leaving the details of the derivation to the following sections: \begin{enumerate} \item Defining the canonical $s$-derivative current made out of two scalars as: \begin{equation}\label{canJ} J^{\text{can}}_{\mu(s)}=i^s\,\phi\,\overset{\leftrightarrow}{\nabla}_{\mu(s)}\phi\,, \end{equation} the redefinition which allows to bring the pseudo-local backreaction \eqref{back} to its canonical form: \begin{equation}\label{caneq1} \underbrace{\Box\phi_{\mu(s)}+\ldots}_{\mathcal{F}_{\mu(s)}}=\alpha_s \,J^{\text{can}}_{\mu(s)}\,, \end{equation} has the structure \eqref{pseudored} with the following choice of coefficients: \begin{equation}\label{genred} a_{l-1}=\frac{2l+s+2}{2}\left(\frac{l!}{(l+s+1)!}\right)^2[p_s(l)+\#\, \alpha_s]\,. \end{equation} Above $p_s(l)$ a polynomial of degree $2(s-1)$ in $l$ while we have left the coefficients $\alpha_s$ arbitrary. As detailed in the following sections the above discussion generalises to a current involving HS linearised curvatures of any spin. In this way it is transparent to compare redefinitions that give different answers for the overall coefficient of the canonical current in \eqref{caneq1}. Moreover, only one of the redefinitions considered above gives the coupling constant which matches the one derived in \cite{Bekaert:2015tva,Sleight:2016dba}. Keeping track of field normalisations (see Appendix~\ref{normaliz}), in the type A theory the choice expected from holography is: \begin{equation}\label{recon} \alpha_s=2^{1-s}\,g^2\,N_0^2\,, \end{equation} which has a simple spin-dependence up to a spin-independent factor proportional to the normalisation of the scalar field kinetic term and to the HS coupling constant $g$. The leading asymptotic behaviour for $l\to\infty$ of the coefficients in the field redefinition \eqref{genred} is spin-independent and equal to $\frac{1}{l^3}$. Therefore, all the redefinitions considered above belong to the \emph{same} functional space as the current in eq.~\eqref{back} itself. \item The value \eqref{recon} for the coupling constant can be fixed using Noether procedure to the quartic-order (see e.g. \cite{Kessel:2015kna} for the 3d computation using admissibility condition) and was reconstructed from Holography in \cite{Bekaert:2015tva,Sleight:2016dba}. So far, however, it was not possible to fix all cubic couplings using only the Noether procedure. That this should be possible in principle is suggested by the result of Metsaev in the light-cone gauge \cite{Metsaev:1991nb,Metsaev:1991mt}.\footnote{In covariant language cubic couplings, including highest derivative ones, should be fixed by global part of the HS symmetry from the equation $\delta^{(1)}S^{(3)}\approx0$.} Furthermore, the implications of the field redefinition mapping the theory to its canonical form at cubic order should be analysed at the quartic order. Such non-local redefinitions would generate a non-local quartic coupling. To appreciate the issue it might be worth noting that the above redefinitions can generate quartic couplings which differ from each other by single-trace blocks in the corresponding conformal block expansion (see e.g. \cite{Bekaert:2014cea,Bekaert:2015tva,Taronna:2016ats,Bekaert:2016ezc,Sleight:2016hyl}). It is also conceivable that, at the quartic order another non-local redefinition will be needed to compensate the cubic redefinition and the additional non-local tails which would arise. The problem of finding a \emph{non-perturbative} redefinition which relates the above tails to standard HS equations is so far open. In this note we restrict the attention to the lowest non-trivial order. \item Notice that choosing a different coupling constant for the canonical current amounts to a subleading contribution in \eqref{genred} with spin-dependent behaviour $\sim 1/l^{2s+1}$ for the coefficients $a_l$ (recall that in the minimal type A theory we restrict the attention to even spins $s>0$). Changing the overall coefficient $\alpha_s$ of the current in \eqref{caneq1} by $\epsilon$ \emph{does not} change the leading asymptotic behaviour $\sim 1/l^3$ of the series expansion of the redefinition \eqref{genred}. This implies that the specification of an asymptotic behaviour for the coefficients $a_l$ is \emph{not} sufficient to specify a proper functional class beyond the proposal of \cite{Taronna:2016ats}. Allowing redefinitions whose asymptotic behaviour is $a_l\sim 1/l^3$ does \emph{not} fix a unique value for $\alpha_s$. Some further condition on the redefinitions must be introduced in order recover a unique admissible choice for $\alpha_s$ when enlarging the functional space beyond the proposal of \cite{Taronna:2016ats}. Notice that the above analysis of the coefficients has a simple interpretation. Given a certain field redefinition with coefficients $a_l$ bound to have a certain asymptotic behaviour, the corresponding improvement can be obtained by a simple action of the covariant adjoint derivative whose effect is to produce some other pseudo-local tail with coefficients $\tilde a_l$ which can be expressed linearly in terms of the coefficients $a_l$, $a_{l-1}$ and $a_{l-2}$: \begin{equation} \tilde a_l= A^{(s)}_l\,a_l+B^{(s)}_{l}\,a_{l-1}+C^{(s)}_{l}\,a_{l-2}\,, \end{equation} with \begin{align} A^{(s)}_l&\sim (l+s+2)^2\,,& B^{(s)}_l&\sim 2ls+2(l+1)^2+s^2\,,& C^{(s)}_l\sim l^2 \end{align} This means that $\tilde a_l \prec l^2 a_l$. On the other hand it is \emph{impossible} to obtain coefficients $\tilde a_l$ growing for $l\to\infty$ much more slowly than the original set of coefficients $a_l$. The only possibility is to have some fine-tuning so that the coefficients $\tilde a_l$ go much faster to zero than the original coefficients. This implies that in order to remove by a field redefinition a backreaction with a given asymptotic behaviour for its coefficients the best one can do is to have a redefinition with similar asymptotic behaviour $\tilde a_l \sim a_l$. This simple argument indirectly implies that the redefinition proposed in \cite{Vasiliev:2016xui} should also be compatible with the \emph{spin-independent} asymptotic behaviour presented in this note. To conclude, a simple test of the functional class proposal of \cite{Vasiliev:2015wma,Vasiliev:2016xui} would be to check if the redefinitions \eqref{genspin} for different values of $\alpha_s$ than \eqref{recon} are indeed not admissible. \item It might be of some interest also to consider a different perspective on the same problem. It is indeed possible to avoid to talk about the subtle issue of field redefinitions and study the limit of the finite derivative truncations of a given back-reaction: \begin{equation}\label{Jsum} J=\sum_{l=0}^\infty \frac{j_l}{l!l!}\,\Box^l\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)\equiv \lim_{k\to\infty}\underbrace{\sum_{l=0}^k \frac{j_l}{l!l!}\,\Box^l\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)}_{J_k}\,. \end{equation} In the above procedure each finite-derivative truncation is well-defined and one can extract the canonical-current piece of each truncation unambiguously. Analogously, one can compute for each truncation the corresponding Witten diagram using standard techniques from local field theories and take the limit only afterwards \cite{Skvortsov:2015lja,Taronna:2016ats}. We declare that the limit exists when the limit is finite and is \emph{independent} of local redefinitions $f_k$ or $g_k$ performed on each given truncation under the assumption that $f_k$ and $g_k$ converge to admissible redefinitions $f_\infty$ and $g_\infty$ according to\footnote{We briefly recall that in order to check whether a redefinition $f_\infty$ belongs to the functional class of \cite{Taronna:2016ats} one first needs to compute the associated improvement $J^{(f)}$ generated by the field redefinition at this order. The corresponding redefinition is then considered admissible iff the limit of the projections of each local truncation of $J^{(f)}$ on the local canonical coupling is vanishing: \begin{align} \lim_{k\to\infty}J_k^{(f)}&=(\lim_{k\to\infty} a_k)\,\mathbf{J}+\lim_{k\to\infty}\Delta J_k\,,& \lim_{k\to\infty} a_k&=0\,. \end{align} Here $\mathbf{J}$ is a fixed, but otherwise arbitrary, (\emph{local}) representative for the non-trivial canonical coupling.} \cite{Taronna:2016ats}. Using a diagrammatic language, the existence of the limit can be summarised by the following commutative diagram: \begin{equation} \begin{matrix}\includegraphics[width=10cm,keepaspectratio]{DiagramC.png}\end{matrix}\,, \end{equation} where $f_\infty$ and $g_\infty$ belong to the functional class defined in \cite{Taronna:2016ats} while $\widetilde{J}_k$ and $J^\prime_k$ are different local forms of the truncation which differ by a \emph{local} field redefinition. If this limit exists we can resum the higher-derivative tail and extract the coefficient of the canonical Metsaev-like coupling. In \cite{Skvortsov:2015lja} it was observed that the above limit for the backreaction \eqref{back} does \emph{not} exist. In this case, it might still be possible to define the sum via some analytic continuation. This is a standard situation where one can define the sum of infinite series formally introducing a cut off procedure. An example of this procedure is: \begin{equation} \sum_{l=1}^\infty 1\,, \end{equation} which can be regularised introducing a regulator as: \begin{equation} \sum_{l=1}^\infty e^{-\epsilon (l+\Lambda)}=\frac{e^{-\epsilon \Lambda}}{e^{\epsilon }-1}\sim \frac{1}{\epsilon}-\left(\frac{1}{2}+\Lambda\right)+O\left(\epsilon\right)\,. \end{equation} The choice $\Lambda=0$ reproduces the standard $\zeta$-function regularisation. As expected, the finite part of the result is regulator dependent and hence ambiguous. For a backreaction \eqref{Jsum} with a divergent sum one similarly ends up with expressions which can be defined formally by analytic continuation. For each given spin there exist a choice of regulator which reproduces the result expected by the holographic reconstruction \eqref{recon}. The question then becomes the same which is usually asked about a renormalisable theory. Namely, whether the choice of regularisation which gives results compatible with holographic reconstruction \eqref{recon} is spin-dependent. If the choice of regulator is spin \emph{independent}, the choice for spin $2$ will fix at the same time the whole backreaction unambiguously. However, if the proper choice of regulator compatible with the holographic reconstruction is spin-dependent the corresponding analytic continuation is not predictive. In the following we give the regularised results for the backreaction for $s=2,4,6$ using the results of \cite{Skvortsov:2015lja,Taronna:2016ats}: \begin{align} \alpha_2(\Lambda)&=\frac{1}{36} \left(1-6 \Lambda ^2\right)\,,\\ \alpha_4(\Lambda)&=\frac{-2100 \Lambda ^6+14280 \Lambda ^5-31290 \Lambda ^4+26600 \Lambda ^3-1680 \Lambda ^2+10080 \Lambda +34567}{1058400}\,,\\ \alpha_6(\Lambda)&=\frac{1}{92207808000}\Big(-291060 \Lambda ^{10}+6338640 \Lambda ^9-57387330 \Lambda ^8+280637280 \Lambda ^7\nonumber\\&-802849740 \Lambda ^6+1351860048 \Lambda ^5-1257850440 \Lambda ^4+525866880 \Lambda ^3\nonumber\\&-3991680 \Lambda ^2+79833600 \Lambda +415046341\Big)\,. \end{align} It is now easy to verify that the choice of the regulator which matches the holographically reconstructed result \eqref{recon}, in the appropriate normalisation, for $\alpha_2$ is not consistent with $\alpha_4$ or $\alpha_6$ with the same normalisation. Therefore, the regulator $\Lambda$ must be spin-dependent, to compensate the highly spin-dependent form of the above regularised expressions. This makes the corresponding analytic continuations unpredictive. This result is not in contradiction with the analysis of field redefinitions presented in this note. The above feature may be a different reincarnation of the fact that all redefinitions removing the higher-derivative tail, regardless the value of $\alpha_s$, have the same asymptotic behaviour at $l\to\infty$. Similarly, this does not allow to single out a unique value of $\alpha_s$. \item It might be interesting to compare the complicated redefinition \eqref{genred} which matches the holographically reconstructed result starting from \eqref{back} with other redefinitions which would generate the required coupling constant but from a free theory. One may indeed start with \emph{free} Fronsdal equations and find the non-local redefinition which would generate the appropriate cubic couplings. This redefinition should not be admissible but, remarkably, it has a faster asymptotic behaviour for $l\to \infty$ than \eqref{genred}: \begin{equation}\label{genredfree} a_{l-1}=\alpha_s\,s!(s-1)!\,(2l+s+2)\left(\frac{l!}{(l+s+1)!}\right)^2\,. \end{equation} This expression is simpler than \eqref{genred}, and falls off faster as $l\to\infty$: $a_l\sim\frac{1}{l^{2s+1}}$\,. Notice that the above redefinition \eqref{genredfree} allows to generate the holographic backreaction from the free theory with the choice \eqref{recon}. The above redefinition however should not be considered admissible as it does not leave the cubic Witten diagrams invariant. \item In the parity violating case, the backreaction \eqref{back} is multiplied by a factor proportional to the parity violating phase $\theta$. Surprisingly (see \cite{Kristiansson:2003xx,Boulanger:2015ova}) this factor is given by $\cos(2\theta)$: \begin{equation}\label{back2} J_{\mu(s)}(\theta)=\sum_{l=0}^\infty\,\frac{j_l\,\cos(2\theta)}{l!l!}\,\Box^l\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)_{\mu(s)}\,, \end{equation} It was then observed in \cite{Boulanger:2015ova} that term by term each element of the pseudo-local series in \eqref{back} vanishes identically for $\theta=\frac{\pi}{4}$. The interpretation of this $\theta$-dependence is at the moment unclear as it seems to be in contradiction with the holographic expectations \cite{Maldacena:2012sf}. In \cite{Vasiliev:2016xui} it was proposed that up to an admissible field redefinition one is left with a canonical current and a $\theta$-independent coefficient. Let us assume that the field redefinition reproducing the holographically expected coupling constants is admissible. This means that the difference between the above non-local current \eqref{back2} and a local canonical current with a fixed non-vanishing coefficient\footnote{Whether or not $\alpha_s(\theta)$ depends on $\theta$ is not important for the following argument. The only assumption is that $\alpha_s$ does not vanish for any value of $\theta$.} $\alpha_s(\theta)$ is an admissible improvement: \begin{equation} J_s(\theta)-\alpha_s(\theta) J_s^{\,\text{can.}}=\Delta^{(J)}_s(\theta)\,. \end{equation} Since the above must be true for any value of $\theta$, we can now set $\theta=\tfrac{\pi}{4}$ and use that $J_s(\tfrac{\pi}{4})=0$. This however implies that the canonical current itself is an admissible improvement: \begin{equation} -\alpha_s(\theta) J_s^{\text{can.}}=\Delta^{(J)}_s(\tfrac{\pi}{4})\,, \end{equation} This further implies that $\Delta^{(J)}_s(\theta)+\frac{x}{\alpha_s(\theta)}\,\Delta^{(J)}_s(\tfrac{\pi}{4})$ is an admissible improvement whose associated admissible field redefinition puts \emph{any} coefficient (parametrised by $x$) in front of the canonical current. However, this contradicts our assumption on the admissibility of the above field-redefinition. Furthermore, the above shows that in the $\theta=\frac{\pi}{4}$ case the initial redefinition itself must generate the canonical backreaction \emph{from} the free Fronsdal equations, and therefore has to match \eqref{genredfree}. Let us also stress that the only assumptions we used is the existence of a field redefinition which is both allowed and capable of changing the $\theta$ dependence of the initial pseudo-local current \eqref{back2}. The contradiction we find seems to imply that no such admissible redefinition may exist. \end{enumerate} In the following we give some details on the results summarised above. This note is organised as follows. After a brief review of the main formalism in Section~\ref{review_part}, we move to the analysis of the redefinitions and to the study of their structure in Section~\ref{Redefs}. We describe the analytic continuations of the higher-derivative tails in Section~\ref{analytic}. We end with a short summary and some outlook in Section~\ref{Conclusions}. In the Appendix we summarise various conventions and derive the normalisation of the Fronsdal kinetic term used in the unfolded language. \section{Pseudo-local Currents}\label{review_part} In \cite{Prokushkin:1999xq,Kessel:2015kna,Boulanger:2015ova,Skvortsov:2015lja,Taronna:2016ats} a convenient generating function formalism was developed, first in 3d and then in 4d, to manipulate quasi-local current interactions and corresponding field redefinitions. In this section we recall the basic ingredients of the formalism, and refer to \cite{Kessel:2015kna,Skvortsov:2015lja} for further details. The main object is the zero-form $C(y,\bar{y}|x)$, which is a formal expansion in the spinorial oscillators $y_\alpha$ and $\bar{y}_{\dot{\alpha}}$ satisfying the linearised unfolded equations \begin{equation} \tadD C(y,{{\bar{y}}}|x)=0\,.\label{twisteq} \end{equation} Here $\adD$ and $\tadD$ are the adjoint and twisted-adjoint covariant derivatives expressed in terms of spinorial oscillators as:\footnote{$h^{\alpha{\dot{\alpha}}}$ is the vierbein, $\nabla=d -\omega^{\alpha\ga}y_\alpha{\partial}_\alpha-\varpi^{{\dot{\alpha}}\gad} {{\bar{y}}}_{\dot{\alpha}}\bar{{\partial}}_{\dot{\alpha}}$ is the AdS$_4$ covariant derivative, while $\varpi^{\alpha\ga}$ and $\varpi^{{\dot{\alpha}}\gad}$ are the (anti-)self-dual components of the spin-connection of AdS$_4$.} \begin{align} \adD &=\nabla-h^{\alpha{\dot{\alpha}}}(y_\alpha \bar{{\partial}}_{\dot{\alpha}}-{{\bar{y}}}_{\dot{\alpha}} {{\partial}}_\alpha)\,,\\ \tadD &=\nabla+ih^{\alpha{\dot{\alpha}}}(y_\alpha {{\bar{y}}}_{\dot{\alpha}}-{\partial}_\alpha \bar{{\partial}}_{\dot{\alpha}})\,. \end{align} In the unfolded language, the zero-form is the main ingredient upon which one constructs ordinary current interactions. Furthermore, upon solving the twisted adjoint covariant constancy condition one recovers the relations between components of the zero form and derivatives of the linearised Weyl tensors associated to the HS fields: \begin{align} C_{\alpha(k+2s),{\dot{\alpha}}(k)}(x)&\sim\nabla^k C_{\alpha(2s)}(x)\,,& C_{\alpha(k),{\dot{\alpha}}(k+2s)}(x)&\sim\nabla^k C_{{\dot{\alpha}}(2s)}(x)\,, \end{align} Above, $C_{\alpha(2s)}(x)$ and $C_{{\dot{\alpha}}(2s)}(x)$ are the self-dual and anti-self dual part of the linearised HS Weyl tensor for $s>0$ and $\Phi(x)\equiv C(0,0|x)$ is the actual scalar field. A current which sources the HS Fronsdal operator is a bilinear functional $J_{\alpha(s){\dot{\alpha}}(s)}(C,C)$ of the zero-form $C$, which is conserved on the equations of motion \eqref{twisteq}: \begin{align} \nabla^{\beta{\dot{\beta}}}J_{\beta\alpha(s-1){\dot{\beta}}{\dot{\alpha}}(s)}\approx0\,. \end{align} For practical purposes, it is convenient to specify the most general form for a current in terms of a generating function Kernel $\bar{J}(Y,\xi,\eta)$: \begin{equation} J(C,C)= \int d^4\xi \, d^4\eta\, \bar{J}(Y,\xi,\eta)\,C(\xi|x) C(\eta|x)\,. \end{equation} Above we have introduced the Fourier transform of the $0$-form with respect to the spinorial variables $y$ and ${{\bar{y}}}$, \begin{equation} C(y,{{\bar{y}}}|x)=\int d^2\xi d^2{{\bar{\xi}}}\,e^{iy^\alpha \xi_\alpha+i{{\bar{y}}}^{\dot{\alpha}}{{\bar{\xi}}}_{{\dot{\alpha}}}}\,C(\xi,{{\bar{\xi}}}|x)\,, \end{equation} where for ease of notation we denote both Fourier transforms and original $0$-forms by the same letter. In this way, contractions of indices are encoded as simple monomials: \begin{align} (y\xi)^n(y\eta)^m({{\bar{y}}}{{\bar{\xi}}})^{\bar{n}}({{\bar{y}}}{{\bar{\eta}}})^{\bar{m}}(\xi\eta)^l({{\bar{\xi}}}{{\bar{\eta}}})^{\bar{l}}\sim C_{\alpha(n)\nu(l);{\dot{\alpha}}(\bar{n})\dot\nu(\bar{l})}C_{\alpha(m)}{}^{\nu(l)}{}_{{\dot{\alpha}}(m)}{}^{\dot{\nu}(\bar{l})}\,. \end{align} In addition to the above way of representing a generic current interaction it is also convenient to introduce generating functions of coefficients via contour integrals, as originally proposed in \cite{Prokushkin:1999xq}. Restricting the attention to the canonical current sector, which in four-dimensions is uniquely specified by the absence of trace components, one can then write the most general current kernel as \begin{align}\label{can4d} \bar{J}(Y,\xi,\eta)=\oint_{\tau_i,s,r} j(\alpha_1,\alpha_2,\beta,\gamma) \, e^{i(s y\zeta^-+\tau_1\xi\eta+r {{\bar{y}}}{{\bar{\zeta}}}^++\tau_2{{\bar{\xi}}}{{\bar{\eta}}})}\,, \end{align} with contour integrals in $\tau_i$, $s$ and $r$ and in terms of a function of $4$ complex variables: \begin{align} \alpha_1&=\tau_1^{-1}\,,& \alpha_2&=\tau_2^{-1}\,,&\beta&=s^{-1}\,,&\gamma&=r^{-1}\,,\\ &&\zeta^{\pm}&=\xi\pm\eta\,,&{{\bar{\zeta}}}^{\pm}&={{\bar{\xi}}}\pm{{\bar{\eta}}}\,. \end{align} These parametrise the four contractions of indices relevant to the canonical current sector in 4d. One can easily translate between the contour integral form and the generating function form via: \begin{equation} \alpha_2^{m+1} \alpha_1^{n+1} \beta ^{s_1+1} \gamma ^{s_2+1}\rightarrow \frac{(i{{\bar{\xi}}}{{\bar{\eta}}}){}^{m} (i\xi\eta){}^{n} (i{{\bar{y}}}{{\bar{\zeta}}}^+){}^{s_2} (iy\zeta^-){}^{s_1}}{m!\, n!\, s_1!\, s_2!}\,. \end{equation} Notice that in this generating function calculus the function $j(\alpha_1,\alpha_2,\beta,\gamma)$ is defined as a formal series, and this should be thought at this level as a regularity condition. This regularity condition will be assumed here since it is equivalent to pseudo-locality of the back-reaction. Notice also that a constant term or terms proportional to $1/\alpha_1$ drop out of the contour integration. In particular, two generating functions will give the same pseudo-local current if they differ by terms of this type. For the details we refer to \cite{Prokushkin:1999xq,Kessel:2015kna} and use the symbol $\sim$ to indicate equality modulo the above equivalence relation. To conclude this section we present the corresponding expressions for the explicit Fronsdal currents of the type $s$-$s_1$-$s_2$: \begin{equation}\label{Fr1} \frac{1}{4(s-1)}\underbrace{\left[\Box+2(s^2-2s-2)+\ldots\right]\phi_s}_{\mathcal{F}}=J_s(C,C)\,, \end{equation} as extracted from Vasiliev's equations\footnote{The normalisation for the Fronsdal tensor comes from the solution to torsion as described in \cite{Boulanger:2015ova}.} in \cite{Boulanger:2015ova}. The generating function $j_s$ for the spin-s current is: \begin{equation}\label{Res} j_s(\beta,\gamma,\alpha_1,\alpha_2)=\frac{i}2e^{2i\theta}(\beta\gamma)^{s+1}\sum_{n,m}c_{n,m}^{(s)}\left(\frac{ p_1(n)-m\, p_2(n)}{(n+s-1)^2 (n+s)^2 (n+s+1)}\right)\alpha_1^n\alpha_2^m+h.c.\,, \end{equation} to be plugged into \eqref{can4d} with \begin{align} p_1^{(s)}(n)&=\Big[[2(s-1)]n^3+[2+s (6 s-5)]n^2+[s \left(1-4 s+6 s^2\right)]n\nonumber\\&+[s (1+s) (2+s (2 s-3))]\Big]\,,\\ p_2^{(s)}(n)&=\Big[[2(s-1)]n^2+[2+s (4 s-3)]n+s (1+s (2 s-1))\Big]\,, \end{align} and \begin{align} c_{n,m}^{(s)}=\frac{1}{4(s-1)}\frac{(-1)^{m+s}+(-1)^{n}}{2}\,. \end{align} Notice that powers of $\alpha_1$ and $\alpha_2$ translate into contractions among the $0$-form, hence powers of $\alpha_1\alpha_2$ parametrise powers of $\Box$ in the metric-like language. We can then define $\tau=\alpha_1\alpha_2$ parametrising the pseudo-local tail of the given interaction term. In general, when restricting attention to sources to the Fronsdal equation in the spin-s sector, one fixes the dependence on $\beta$ and $\gamma$ as in \eqref{Res} while working with a generating function of the type: \begin{align}\label{Generic} j_{s}(\beta,\gamma,\alpha_1,\alpha_2)&=i \cos(2\theta)(\beta\gamma)^{s+1}\,\alpha_1^n\,g(z)\,,& z&=\alpha_1\alpha_2\,. \end{align} The function $g(z)$ parametrises the infinite non-local tail while the dependence on $\beta$, $\gamma$ and a single $\alpha_1$ ($\alpha_2$) fixes the canonical current tensor structure and the spin of the zero-forms (see Appendix \ref{Dictionary}). The canonical current with no higher-derivative tail $J^{\text{can}}(C,C)$ is simply encoded by the choice $g(z)\sim \alpha\, z$. In the $s$-$0$-$0$ case the dictionary can be given quite explicitly as: \begin{align}\label{dic} j_s&\rightarrow \cos(2\theta)\sum_{l,k}a_{l,k}\nabla_{\mu(s-k)\nu(l)}\Phi\nabla_{\mu(k)}{}^{\nu(l)}\Phi\,,\\ \nabla_{\mu(s-k)\nu(l)}\Phi\nabla_{\mu(k)}{}^{\nu(l)}\Phi&\equiv h^{\alpha{\dot{\alpha}}}_{\mu}\ldots h^{\alpha{\dot{\alpha}}}_{\mu}C_{\alpha(s-k)\nu(l),{\dot{\alpha}}(s-k)\dot{\nu}(l)}C_{\alpha(k)}{}^{\nu(l)}{}_{,{\dot{\alpha}}(s-k)}{}^{\dot{\nu}(l)}\,,\label{def} \end{align} with \begin{align}\label{factor} a_{l,k}&=\frac{(-1)^{k+l}s!s!}{l!l!k!k!(s-k)!(s-k)!}\,g_l\,,& g(z)=\sum_{l=0}^\infty g_{l}\,z^{l+1}\,. \end{align} The above dictionary holds also for redefinitions of the Fronsdal field and allows to easily translate from the generating function language to the standard tensorial language. Notice that the notation $\nabla_{\mu(s-k)\nu(l)}\Phi\nabla_{\mu(k)}{}^{\nu(l)}\Phi$ is defined by eq.~\eqref{def} and includes symmetrisation and traceless projection. In particular $\nabla_{\mu(s)}\neq \nabla_\mu\ldots \nabla_\mu$. We give further details on more general current interactions included in the generating function \eqref{Generic} in Appendix~\ref{Dictionary}. In the following we shall restrict our attention to the function $g(z)$ encoding the higher-derivative tail. It is important to stress that the $\frac1{(l!)^2}$ factor in \eqref{factor} arises via the above contour integrations and does not appear in the function $g(z)$. \section{Pseudo-Local Field Redefinitions}\label{Redefs} In this section we employ the generating function formalism to study the non-local field redefinitions which relate the back-reaction extracted from Vasiliev's equation in \cite{Boulanger:2015ova} to canonical currents. For simplicity we work in the A-type model, setting $\theta=0$. The discussion generalises straightforwardly to any choice of $\theta$ as this only appears as an overall factor. The effect of a field redefinition of the spin-$s$ field quadratic in the $0$-form can be encoded in an arbitrary function $k(z)$ (analogous to $g(z)$ above) via the differential operator\footnote{The corresponding redefinition at the level of unfolded equations reads: \begin{align} \delta\omega_s(y,{{\bar{y}}})&=h^{\alpha{\dot{\alpha}}}{\partial}_\alpha\bar{{\partial}}_{\dot{\alpha}} k_s(y,{{\bar{y}}})\,. \end{align}} \begin{multline}\label{canonical} \Delta j_s=\frac{(\beta\gamma)^{s+1}}{4(s-1)}\Big[z^2 (z +1)^2 k_s''(z)+z(z+1)(2 s+(3+\bar{n}) z +3+\bar{n}) k_s'(z)\\+ \left(\bar{n}(z+1)(z+s+1)+s^2 (z +1)+2 s+(z +1)^2\right)k_s(z)\Big]\,, \end{multline} where $\bar{n}=n-m$ with $n$ and $m$ giving the power of $\alpha_1$ and $\alpha_2$ respectively as in \eqref{Res}. In this way, the problem of finding redefinitions which remove the pseudo-local tail is thus mapped into a ordinary differential equation. In the following we apply this formalism to the current \eqref{Res}. We first give the explicit field redefinition which maps the pseudo-local current \eqref{Res} to a canonical current with an arbitrary overall coefficient: \begin{equation}\label{caneq} \underbrace{\Box\phi_s+\ldots}_{\mathcal{F}}=\alpha_s \,J_s^{\text{can}}(C,C)\,. \end{equation} Notice that in principle $\alpha_s$ can depend on the spins $s_1$ and $s_2$ of the zero-forms. However, since the result of the redefinition depends only on $s$ we do not write explicitly the dependence on $s_1$ and $s_2$. We then compare redefinitions which give different overall coefficients $\alpha_s$ for the canonical current. For ease of notation we only restrict ourselves to the currents with $\bar{n}=0$ (see Appendix \ref{Dictionary}). \paragraph{Spin-2:} Using the generating function formalism reviewed above, the back-reaction \eqref{Res} for $s=2$ and $\bar{n}=0$ can be encoded by (see eq.~\eqref{Generic}): \begin{equation} g_2(z)=\sum_{l=0}^\infty \frac{1}{2\, (l+1)^2 (l+2)}\,(-z)^l=\frac{z -z\,\text{Li}_2(-z )-(z +1) \log (z +1)}{2 z ^2}\,. \end{equation} The field redefinition that reduces the above pseudo-local current to its canonical form can be obtained solving the differential equation: \begin{align} z (z +1) \left(z (z +1) k_2''(z)+(3 z +7) k_2'(z )\right)+(z +3)^2 k_2(z )=-4\,g_2(z)+\alpha_2\, z+\beta_2\,. \end{align} The constant $\beta_2$ parametrises non-trivial pseudo-local redefinitions which result in a vanishing\footnote{We have not considered terms on the right-hand side of the type $z^{-m}$, since the corresponding redefinitions vanish upon performing the contour integration and are thus equivalent under the equivalence relation $\sim$. The inequivalent solutions are parametrised by one constant $\beta_s$. This can be checked by solving the differential equation (see \cite{Kessel:2015kna} for further details).} contribution to the backreaction upon performing the contour integration. On the other hand, $\alpha_2$ is an arbitrary constant in front of the canonical current (which in this case is the stress tensor \eqref{caneq}). In the following we fix the constant $\beta_2=2\alpha_2 +2$ in such a way that the slowest contribution to the coefficients in the $l\to\infty$ limit is set to zero. Furthermore, one should carefully fix the freedom in the homogeneous solutions choosing the unique solution to the above differential equation which is analytic in $z$, as this corresponds to the standard regularity condition on pseudo-local functionals \cite{Prokushkin:1999xq,Skvortsov:2015lja}. The final result has the following series expansion around $z=0$: \begin{equation} k_2(z)=\sum_{l=0}^\infty\frac{(4 \alpha_2 +l (l+3)+2)}{(l+1)^2 (l+2) (l+3)^2}\,(-z)^l\,, \end{equation} where we have left $\alpha_2$ arbitrary. So far the discussion applies to any current interaction with $\bar{n}=0$ regardless the spin of the 0-forms. Considering the $2$-$0$-$0$ case, the choice compatible with holography is given by \eqref{recon} and is $\alpha_2=2 g^2\,N_0^2$. \paragraph{Spin-4:} The source to the spin-4 Fronsdal operator extracted from \eqref{Res} for $\bar{n}=0$ can be encoded in the following generating-function: \begin{equation} g_4(z)=\sum_{n=0}^\infty \frac{(3 l+11)}{3 (l+3)^2 (l+4)^2}\,(-z)^l\sim-\frac1{3\,z^4}\,\left[(2 z+1) \text{Li}_2(-z) + (z+1) \log (z+1)\right]\,. \end{equation} The redefinition bringing the above back-reaction to its canonical form \eqref{caneq} with an arbitrary coefficient $\alpha_4$, can be obtained solving the differential equation: \begin{multline} z (z +1) \left(z (z +1) k_4''(z)+(3 z +11) k_4'(z )\right)+(z^2+18z+25) k_4(z )\\=-12\,g_4(z)+\alpha_4 z+\beta_4\,. \end{multline} Again, $\beta_4$ parametrises pseudo-local redefinitions which do not change the back-reaction, and can be chosen to improve the asymptotic behaviour. In this case we choose $\beta_4=\frac{1}{2} (3 \alpha_4 +4)$. The solution to the above differential equation is then: \begin{equation} k_4(z)=\sum_{l=0}^\infty\frac25\frac{p_4(l)+120 (6 \alpha_4 +5)}{(l+1)^2 (l+2)^2 (l+3) (l+4)^2 (l+5)^2}\,(-z)^l\,, \end{equation} with $p_4(l)$ a polynomial of order $5$ in the variable $l$: \begin{equation} p_4(l)=l \left(5 l^5+77 l^4+470 l^3+1445 l^2+2345 l+1898\right)\,. \end{equation} \paragraph{Spin-6:} The source to the spin-6 Fronsdal operator extracted from \eqref{Res} for $\bar{n}=0$ can be encoded in the generating-function \begin{multline} g_6(z)=\sum_{l=0}^\infty \frac{3 (5 l+28)}{10 (l+5)^2 (l+6)^2}\,(-z)^l\\\sim-\frac3{10\,z^6}\left[(z +1) \log (z +1)+ (3 z +2) \text{Li}_2(-z )\right]\,. \end{multline} The redefinition bringing the above back-reaction to its canonical form \eqref{caneq} with an arbitrary coefficient $\alpha_6$ can be obtained solving the differential equation: \begin{multline} z (z +1) \left(z (z +1) k_6''(z)+3(z +5) k_6'(z )\right)+z (z +38)+49 k_6(z )\\=-20\,g_6(z)+\alpha_6 z+\beta_6\,. \end{multline} Again, $\beta_6$ parametrises pseudo-local redefinitions which do not change the back-reaction. In this case we choose $\beta_6=\frac{1}{3} (4\, \alpha_6 +6)$. The solution to the above differential equation is then given by: \begin{equation} k_6(z)=\sum_{l=0}^\infty\frac{p_6(l)+2\cdot 6!\cdot 5!\, (\alpha_6 +1)}{(l+1)^2 (l+2)^2 (l+3)^2 (l+4) (l+5)^2 (l+6)^2 (l+7)^2}\,(-z)^l\,, \end{equation} with $p_6(l)$ a polynomial of order $10$ in the variable $l$: \begin{multline} p_6(l)=l\, \Big(3 l^9+\frac{323 l^8}{3}+1686 l^7+\frac{105862 l^6}{7}+85755 l^5+320009 l^4\\+792684 l^3+\frac{3844612 l^2}{3}+1288512 l+\frac{5072640}{7}\Big)\,. \end{multline} \paragraph{Generic even spin:} In the generic spin-$s$ case, the source quadratic in the $0$-form to the Fronsdal operator extracted in \cite{Boulanger:2015ova} can be encoded by the generating-function \begin{multline} g_s(z)=\sum_{l=0}^\infty \frac{s (2 l (s-1)+s (2 s-3)+2)}{8 (s-1) (l+s-1)^2 (l+s)^2}\,(-z)^l\\\sim-\frac{c_s}{z^s}\left[(z +1) \log (z +1)+\tfrac12 (s(z+1)-2) \text{Li}_2(-z )\right]\,, \end{multline} with \begin{equation} c_s=\frac14\frac{s}{s-1}\,. \end{equation} The corresponding redefinition mapping the above pseudo-local back-reaction to a canonical current \eqref{caneq} with overall coefficient $\alpha_s$ \eqref{caneq} is: \begin{equation}\label{genspin} k_s(z)=\sum_{l=0}^\infty\frac{2l+s+2}{2}\left(\frac{l!}{(l+s+1)!}\right)^2[p_s(l)+X_s\, \alpha_s+Y_s]\,(-z)^l\,, \end{equation} with $p_s(l)$ a polynomial of degree $2(s-1)$ in the variable $l$ and $X_s$ and $Y_s$ two spin-dependent constant. The general form of the polynomial for arbitrary spin is cumbersome, and we do not present it explicitly. To summarise, in this section we have calculated explicitly the field redefinition relating the backreaction extracted from Vasiliev's equations to their canonical form. The main result is that the coefficient of the canonical current turns out to contribute a subleading term in the field redefinition and cannot be fixed by prescribing an asymptotic behaviour for the corresponding coefficients. \section{Analytic Continuation}\label{analytic} In this section upon reviewing the results of \cite{Skvortsov:2015lja,Taronna:2016ats}, we study the analytic continuation of the formal series obtained by considering the limit of the finite derivative truncations of the backreaction \eqref{back}. The main observation of \cite{Skvortsov:2015lja,Taronna:2016ats} is that for any truncation of the pseudo-local interaction term \eqref{Res} it is possible to compute the corresponding projection on the canonical current piece which is parametrised by finitely many structures in correspondence with Metsaev's cubic couplings. Each higher-derivative term gives a contribution to the canonical coupling weighted by some proportionality factor $C_l^{(s)}$ which measures the overlap of the higher-derivative term on the canonical structure. Schematically the projection reads: \begin{equation} \mathcal{P}\left[\Box^l\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)_{\mu(s)}\right]=C_{l}^{(s)}\,\left(\nabla\ldots\nabla\phi\ \nabla\ldots\nabla\phi\right)_{\mu(s)}\,, \end{equation} where indeed the right-hand side is proportional to the canonical structure times a certain overall coefficient which was computed in \cite{Skvortsov:2015lja,Taronna:2016ats}. In generating functions terms, the above amounts to the following projection (we restrict again to $\bar{n}=0$): \begin{equation} \mathcal{P}[z^l]=C_l^{(s)}\,z\,, \end{equation} with coefficients given by: \begin{equation} C_l^{(s)}=-\frac{(-1)^l s\, \Gamma (l+s+1) \, _3F_2(1-s,1-s,-2 s;2-2 s,l-s+1;1)}{2 (2 s-1) \Gamma (s+1)^2 \Gamma (l-s+1)}\,. \end{equation} After the projection the canonical structure $g(z)=z$ factorises and one is left with an overall coefficient which combines together all contributions from each higher-derivative term. Below we give the corresponding coefficient for some low spin examples: \begin{align} s&=2\,,& &-\frac1{12}\sum_{l=1}^\infty l\,,\\ s&=4\,,& &-\frac1{3\cdot 7!}\sum_{l=1}^\infty \frac{l (l+1) (l+2)^2 (3 l+11) (5 l (l+4)+3)}{(l+3) (l+4)}\,,\\ s&=6\,,& &-\frac{3}{5\cdot 11!}\sum_{l=1}^\infty \frac{(l+4)!}{(l-1)!}\frac{(l+3) (5 l+28) (7 l (l+6) (3 l (l+6)+19)+20)}{(l+5) (l+6)}\,,\\ s&=8\,,& &-\frac{7}{6\cdot 15!}\sum_{l=1}^\infty \frac{(l+6)!}{(l-1)!}\tfrac{(l+4) (7 l+53) (l (l+8) (11 l (l+8) (13 l (l+8)+274)+14631)+420)}{ (l+7) (l+8)}\,. \end{align} These series are divergent and can be regularised by a standard $\zeta$-function regularisation introducing a regulator of the type $e^{-\epsilon(l+\Lambda)}$ which allows to resum them. Dropping the divergent parts and setting $\epsilon$ to zero one then obtains the following regularised expressions for the corresponding overall coefficients of the canonical current:\footnote{In the formulas for $\alpha_s(\Lambda)$ we take into account the normalisation $\frac1{4(s-1)}$ in \eqref{Fr1} so to arrive to a source of the type $\mathcal{F}_{\mu(s)}=\alpha_s(\Lambda) J_{\mu(s)}^{\text{can.}}$.} \begin{align} \alpha_2(\Lambda)&=\frac{1}{36} \left(1-6 \Lambda ^2\right)\\ \alpha_4(\Lambda)&=\frac{-2100 \Lambda ^6+14280 \Lambda ^5-31290 \Lambda ^4+26600 \Lambda ^3-1680 \Lambda ^2+10080 \Lambda +34567}{1058400}\\ \alpha_6(\Lambda)&=\frac{1}{92207808000}\Big(-291060 \Lambda ^{10}+6338640 \Lambda ^9-57387330 \Lambda ^8+280637280 \Lambda ^7\nonumber\\&-802849740 \Lambda ^6+1351860048 \Lambda ^5-1257850440 \Lambda ^4+525866880 \Lambda ^3\nonumber\\&-3991680 \Lambda ^2+79833600 \Lambda +415046341\Big)\\ \alpha_8(\Lambda)&=\frac{1}{39269461271040000}\Big(-51531480 \Lambda ^{14}+2307024720 \Lambda ^{13}-45901996140 \Lambda ^{12}\nonumber\\&+535422167280 \Lambda ^{11}-4065466421016 \Lambda ^{10}+21086692426800 \Lambda ^9\nonumber\\&-76212086580630 \Lambda ^8+191985197049360 \Lambda ^7-330826659683520 \Lambda ^6\nonumber\\&+372883114251648 \Lambda ^5-249836835568320 \Lambda ^4+77398236115200 \Lambda ^3\nonumber\\&-62270208000 \Lambda ^2+2615348736000 \Lambda +16930453296697\Big)\,. \end{align} These results allow in principle to fix the regulator in order to recover the holographically reconstructed result of \cite{Bekaert:2015tva,Sleight:2016dba}. The main observation is however that the associated choice for the regulator is \emph{spin-dependent}. Furthermore the regularised coupling constant is given by complicated polynomials of order $\Lambda^{2(s-1)}$. The lack of a spin-independent regularisation makes the corresponding analytic continuations unpredictive as a consequence of the strongly coupled nature of the higher-derivative tails. It is therefore not clear how to identify a regularisation which preserves all HS symmetries without solving the Noether procedure up to the quartic order. On the other hand, the simple form of the local holographically reconstructed couplings \cite{Sleight:2016dba} predicts a very simple structure for the corresponding local interactions. \section{Conclusions}\label{Conclusions} In this note we have determined explicitly the general form of the redefinition which reduces the back-reaction extracted in \cite{Boulanger:2015ova} from Vasiliev's equations to a canonical current with an arbitrary overall coefficient \eqref{caneq}. We have also given some details about the analytic continuation of the formally divergent sums. Our analysis provides a convenient test ground to probe functional class proposals for admissible non-local interactions. The main conclusion of this note is that a full non-perturbative functional class able to fix the overall coefficient in front of the canonical interaction terms should be spelled out to complete the dictionary between Vasiliev's equations and standard HS equations like Fronsdal's equations. Furthermore, it is worth stressing that in this note we have restricted the attention to the cubic order which admits a local completion at least for fixed spins. At the quartic and higher orders we expect the situation to be even more subtle, since non-localities may not be anymore removed by redefinitions, and very restrictive consistency requirements will be put in place by consistency \cite{Taronna:2011kt,Taronna:2012gb,Taronna:2016ats,Rahman:2016tqc}. The study of locality is at the moment incomplete at quartic and higher orders and this complicates the definition and the test of possible non-perturbative functional classes. We conclude this note with a short list of observations and comments: \begin{itemize} \item The main difficulty of HS theories is the absence of a scale beyond the AdS-radius. This implies that the behaviour of a pseudo-local tail can only be controlled by its coefficients owing to $\Lambda[\nabla,\nabla]\sim1$. A functional class proposal would then prescribe a given asymptotic behaviour for the coefficients of higher and higher order terms as in the proposal of \cite{Taronna:2016ats}. Enlarging the functional class beyond the latter proposal seem to lead to unpredictive results unless some further condition on the redefinition is imposed. \item It would be interesting to study the limit $\alpha^\prime\to\infty$ at the level of effective cubic string field theory couplings in AdS. Some issues about commutativity of limits may arise in this context when taking the limit $\alpha^\prime\to\infty$ before or after removing the higher-derivative tails. It might well be that if one considers the naive $\alpha^\prime\to\infty$ limit of a non-local string coupling this would indeed be strongly coupled. This would mean that the tensionless limit should be well-defined only in a particular field frame, while singular in others. This kind of situation suggests that allowed redefinitions for any finite value of $\alpha^\prime$ could become not allowed after the limit. If so the infinities observed in this note could be resolved upon taking the limit from string theory using $\alpha^\prime$ as regulator for the strongly coupled series. On the other hand, the fact that the $\alpha^\prime\to\infty$ limit could display such subtleties requires particular care in the definition itself of tensionless strings. \item Beyond the particular problem of mapping the tail in \eqref{back} to a canonical form, the simplicity of the 4d theory in the spinorial language is expected to manifest in a simple form of all quadratic sources when rewritten in the unfolded form. Such rewriting, and the structures involved beyond the $\star$-product, is at the moment unknown (see however \cite{Vasiliev:1988sa,Vasiliev:1989xz,Vasiliev:1989yr}). For this reason an interesting problem would be to unfold the non-linear Fronsdal equations in 4d coming from the cubic couplings extracted holographically in \cite{Sleight:2016dba}. This result would provide us with the complete list of cubic vertices in the unfolded equations possibly giving us a hint of higher-order completions. \end{itemize} \section*{Acknowledgments} \label{sec:Aknowledgements} I am grateful to N. Boulanger, S. Didenko, D. Francia, A. Sagnotti, E. Sezgin, C. Sleight, Z. Skvortsov, D. Sorokin, P. Sundell and M. Vasiliev for useful discussions. The research of M. Taronna is partially supported by the Fund for Scientific Research-FNRS Belgium, grant FC 6369 and by the Russian Science Foundation grant 14-42-00047 in association with Lebedev Physical Institute. This research was also supported by the Munich Institute for Astro- and Particle Physics (MIAPP) of the DFG cluster of excellence “Origin and Structure of the Universe”. \begin{appendix} \section{Canonical Currents Quadratic in the Curvatures}\label{Dictionary} In this appendix we would like to give a few more details on the explicit tensorial form of the currents discussed in this note. Canonical currents are encoded as generating functions by $(\beta\gamma)^{s+1}\alpha_1^n\,z$, since any dependence on $\alpha_1\alpha_2$ can be removed by a local field redefinition. Such monomial corresponds to the following generating function kernels: \begin{multline} (\beta\gamma)^{s+1}\alpha_1^{n}\,z\rightarrow\frac1{s!^2n!}\,(-i{\partial}_{\xi}\cdot{\partial}_{\eta})^n(-iy({\partial}_\xi-{\partial}_\eta))^s(-i{{\bar{y}}}({\partial}_{{\bar{\xi}}}+{\partial}_{{\bar{\eta}}}))^s\,\\\sum_{p_1,q_1}\frac1{p_1!q_1!}C_{\alpha(p_1){\dot{\alpha}}(q_1)}(x)\xi^{\alpha(p_1)}{{\bar{\xi}}}^{{\dot{\alpha}}(q_1)}\sum_{p_2,q_2}\frac1{p_2!q_2!}C_{\beta(p_2){\dot{\beta}}(q_2)}\eta^{\beta(p_2)}{{\bar{\eta}}}^{{\dot{\beta}}(q_2)}\Big|_{\eta=0,\xi=0}\,. \end{multline} It is then easy to perform all required differentiations ending up with \begin{equation} (\beta\gamma)^{s+1}\alpha_1^{n}\,z\\\rightarrow(-1)^s\,\frac{y^{\alpha(s)}{{\bar{y}}}^{{\dot{\alpha}}(s)}}{s!^2}\sum_{p,q=0}^s(-1)^q\binom{s}{p}\binom{s}{q}\,\frac{(-i)^{n}}{n!}C_{\alpha(s-p)\beta(n){\dot{\alpha}}(s-q)}C_{\alpha(p)}{}^{\beta(n)}{}_{{\dot{\alpha}}(q)}\,, \end{equation} together with its conjugate piece which can be obtained by replacing $\alpha_1$ with $\alpha_2$. Notice that the spin of a zero form $C_{\alpha(n){\dot{\alpha}}(m)}$ is given by $\tfrac{|n-m|}{2}$ so that the first 0-form has spin $s_1=\tfrac{|n-(p-q)|}2$ while the second has spin $s_2=\tfrac{|n+(p-q)|}{2}$. The number of derivatives is instead $s_1+s_2+\text{min}(s-p+n,s-q)+\text{min}(p+n,q)$. A particularly interesting case is when the canonical current has the maximum number of derivatives $s+s_1+s_2$. For $s_1=s_2$ this coupling is realised with $p=q$ and $s_1=s_2=\tfrac{n}2$. The $s$-$0$-$0$ current is a particular case of the latter for $n=0$. In the $s$-$s_1$-$s_2$ case $n=0$ reproduces Bel-Robinson type currents \cite{Gelfond:2006be}.\footnote{In the $n=0$ case the same generating function can also be written in more conventional form as $$(\beta\gamma)^{s+1}\,z\qquad\longrightarrow\qquad C(y,{{\bar{y}}})C(-y,{{\bar{y}}})\,.$$} \section{Fixing the 2pt Normalisations in Vasiliev's Equations Holographically}\label{normaliz} An important subtlety to correctly interpret coefficients of currents at the equations of motion level is to determine the kinetic term normalisation for the Fronsdal fields. Such normalisation indeed play a key role to fix the cubic couplings in \cite{Sleight:2016dba}. In the following we will study these normalisations fixing the notation. Furthermore we will fix the kinetic-term normalisation for all HS fields by matching holographically the sources coming from the $0$-$0$-$s$ cubic couplings. First of all it is important to fix the convention for spinorial indices as in \cite{Boulanger:2015ova} with \begin{align}\label{viel} h_\mu^{\alpha{\dot{\alpha}}}h^\mu_{\beta{\dot{\beta}}}&= \epsilon_{\beta}{}^{\alpha}\epsilon_{{\dot{\beta}}}{}^{{\dot{\alpha}}}\,, && h_\mu^{\alpha{\dot{\alpha}}}h^\nu_{\alpha{\dot{\alpha}}}=\delta^\nu_\mu\,, \end{align} and in Poincar\'e coordinates: \begin{align} h_\mu^{\alpha{\dot{\alpha}}}\,dx^\mu&=\frac1{2z} \sigma_\mu^{\alpha{\dot{\alpha}}} dx^\mu\,, && h^\mu_{\alpha{\dot{\alpha}}}=z \sigma^\mu_{\alpha{\dot{\alpha}}}\,, && g_{\mu\nu}=\frac1{2z^2} \eta_{\mu\nu}\,. \end{align} Here $\epsilon^{\alpha\beta}=-\epsilon^{\beta\alpha}$, $\epsilon^{12}=1$ and $\epsilon=i\sigma_2$ with $\sigma_i^{\alpha{\dot{\alpha}}}$, $i=1,2,3$ being the Pauli matrices. Notice that in this note we work with the choice $\Lambda=2$. Fronsdal equations then read: \begin{equation} \frac1{4(s-1)}\,\left[\square +2(s^2-2s-2)+\ldots\right] \phi_{\alpha(s){\dot{\alpha}}(s)}=J_{\alpha(s){\dot{\alpha}}(s)}\,. \end{equation} where we have included the factor $\frac1{4(s-1)}$ coming from solving torsion as described in \cite{Boulanger:2015ova}. The mass term is dressed by a factor of $2$ coming from our conventions for $\Lambda$. The mapping between spinorial and vectorial indices is achieved via the vielbein as: \begin{equation} \phi_{\mu(s)}\equiv \phi_{\alpha(s){\dot{\alpha}}(s)}(x)\,\underbrace{h^{\alpha{\dot{\alpha}}}_{\mu}\cdots h^{\alpha{\dot{\alpha}}}_{\mu}}_s\,, \end{equation} which follows from \eqref{viel}. The holographically reconstructed equations can be extracted from the $0$-$0$-$s$ cubic coupling: \begin{equation} \mathcal{L}\sim\sum_s \left[\frac{N_s^2}{2^{s+2}}\,\phi_{\mu(s)}\Box\,\phi^{\mu(s)}+\ldots\right]-\underbrace{\frac{2^4}{\sqrt{N}}}_{g}\sum_s N_sN_0^2\,\frac{2^{-\tfrac{s}2}}{\Gamma(s)}\,i^s\,\phi^{\mu(s)}\,\left(\tfrac1{2^s}\phi\,\overset{\leftrightarrow}{\nabla}_{\mu(s)}\phi+\ldots\right)\,, \end{equation} with an arbitrary normalisation $N_s$ for the Fronsdal kinetic term. They read:\footnote{Notice that a factor $2^{-s}$ comes from the non-canonically normalised metric in the contractions between HS field and derivatives. A further factor of $2$ comes from the non canonical normalisation for the Laplacian and another factor of $2$ comes from the variation with respect to the scalar field.} \begin{align} (\Box-4)\phi(x)&=g\,N_s\,\frac{2^{2-\tfrac{s}{2}}\,i^s}{\Gamma(s)}\,\phi^{\mu(s)}\,\nabla_{\mu(s)}\phi=g\,N_s\,\frac{2^{2-\tfrac{s}{2}}}{\Gamma(s)}\,\phi^{\alpha(s){\dot{\alpha}}(s)}\,C_{\alpha(s){\dot{\alpha}}(s)}\,,\label{arbitraryN} \end{align} where we have used that on the scalar sector of the zero form: \begin{equation} \tadD C=0\qquad\longrightarrow\qquad C_{\alpha(s){\dot{\alpha}}(s)}(x)=i^n\underbrace{\nabla_{\alpha{\dot{\alpha}}}\cdots\nabla_{\alpha{\dot{\alpha}}}}_s\,\phi(x)\,. \end{equation} The above equations allow to determine holographically the normalisation of the kinetic term used by Vasiliev's equations simply from the $0$-$0$-$s$ source to the scalar equation which can be extracted from the standard twisted-adjoint structure constants: \begin{equation} \tadD C=\omega\star C-C\star\pi(\omega)\,. \end{equation} Upon translating the above equations to standard Klein-Gordon equations we arrive to: \begin{equation} (\Box-4)\phi(x)=\frac{4}{\Gamma(s)^2}\,\phi^{\alpha(s){\dot{\alpha}}(s)}\,C_{\alpha(s){\dot{\alpha}}(s)}\,, \end{equation} from which comparing with \eqref{arbitraryN} we can determine the normalisation $N_s$ used by Vasiliev's equations: \begin{align} N_s&=\frac1{g}\,\frac{2^{\tfrac{s}{2}}}{\Gamma(s)}\,,& s&>0\,. \end{align} In the latter normalisation one then gets the following coupling constants for the source to the Fronsdal equations: \begin{equation} \left[\square +2(s^2-2s-2)+\ldots\right]\phi_{\mu(s)}(x)\,=\,\underbrace{\frac{g\,N_0^2}{N_s}\,\frac{2^{\tfrac{s}{2}+1}}{\Gamma(s)}}_{=\,2\,g^2\, N_0^2}\,i^s\left(\tfrac{1}{2^s}\phi\,\overset{\leftrightarrow}{\nabla}_{\mu(s)}\phi+\ldots\right)\,,\label{alphas} \end{equation} which, taking into account the normalisation \eqref{canJ} for the canonical current, determines the value for the coefficient $\alpha_s$ in \eqref{caneq1} up to a spin-independent constant: \begin{equation} \alpha_s=2^{1-s}g^2\,N_0^2\,. \end{equation} Similar results can be obtained for all other couplings of \cite{Sleight:2016dba}. \end{appendix}
9b897dee32341df521b91e8a88f9d2b9a04e562f
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \blfootnote{ % % % % \hspace{-0.65cm} This work is licensed under a Creative Commons Attribution 4.0 International License. License details: \url{http://creativecommons.org/licenses/by/4.0/} } Argumentation mining is a relatively new challenge in corpus-based discourse analysis that involves automatically identifying argumentative structures within a corpus. Many tasks in argumentation mining~\cite{argumentsurvey2015} and debating technologies~\cite{colingdemo2014} involve categorizing a sequence in the context of another sequence. For example, in \emph{context dependent claim detection}~\cite{cdcd2014}, given a sentence, one task is to identify whether the sentence contains a claim relevant to a particular debatable topic (generally given as a context sentence). Similarly in \emph{context dependent evidence detection}~\cite{evidence2015}, given a sequence (possibly multiple sentences), one task is to detect if the sequence contains an evidence relevant to a particular topic. We refer to such class of problems in computational argumentation as \emph{bi-sequence classification} problems---given two sequences $s$ and $c$ we want to predict the label for the target sequence $s$ in the context of another sequence $c$\footnote{In this paper, we shall ignore the subtle distinction between sentence and sequence and both will mean just a text segment composed of words.}. Apart from the debating tasks, several other natural language inference tasks fall under the same paradigm of having a pair of sequences. For example, \emph{recognizing textual entailment}~\cite{snli2015}, where the task is to predict if the meaning of a sentence can be inferred from the meaning of another sentence. Another class of problems originated from question-answering systems also known as \emph{answer selection}, where given a question, a candidate answer needs to be classified as an answer to the question at hand or not. Recently, deep learning approaches have obtained very high performance across many different natural language processing tasks. These models can often be trained in an end-to-end fashion and do not require traditional, task-specific feature engineering. For many single sequence classification tasks, the state-of-the-art approaches are based on recurrent neural networks (RNN variants like Long Short-Term Memory (LSTM)~\cite{lstm97} and Gated Recurrent Unit (GRU)~\cite{cho14}) and convolution neural network based models (CNN)~\cite{Kim14}. Whereas for bi-sequence classification, the context sentence $c$ has to be explicitly taken into account when performing the classification for the target sentence $s$. The context can be incorporated into the RNN and CNN based models in various ways. However there is not much understanding in current literature as to the best way to handle context in these deep learning based models. In this paper, we empirically evaluate(see Section ~\ref{sec:experiments}) the performance of five different ways of handling context in conjunction with target sentence(see Section~\ref{sec:models}) for multiple bi-sequence classification tasks(see Section~\ref{sec:tasks}) using architectures composed of RNNs and(/or) CNNs. In a nutshell, this paper makes the two major novel contributions: \begin{enumerate} \item We establish the first deep learning based baselines for three bi-sequence classification tasks relevant to argumentation mining with zero feature engineering. \item We empirically compare the performance of several ways handling context for bi-sequence classification problems in RNN and CNN based models. While some of these variants are used in various other tasks, there has been no formal comparison of different variants and this is the first attempt to actually list all the variants and compare them on several publicly available benchmark datasets. \end{enumerate} \section{Bi-Sequence classification tasks}\label{sec:tasks} In this section, we will briefly mention the various bi-sequence tasks of interest in the literature of argumentation mining and in the broader natural language inference domain. \subsection{Argumentation Mining} We mainly consider two prominent tasks in argumentation mining, namely, detecting the claims~\cite{cdcd2014} and evidences~\cite{evidence2015}, within a given corpus, which are related to a pre–specified topic. These two tasks together helps to automatically construct persuasive arguments out of a given corpora. We will define the following four concepts: \noindent \textbf{Motion} - The topic under debate, typically a short phrase that frames the discussion. \noindent \textbf{Claim} - A general, typically concise statement that directly supports or contests the motion. \noindent\textbf{Motion text} - A document/article/discourse that contain claims with high probability. \noindent \textbf{Evidence} - A set of statements that directly supports the claim for a given motion. \subsubsection{Context Dependent Claim Detection (CDCD)} \label{sec:cdcd} \emph{Given a sentence in a motion text the task is to identify whether the sentence contains a claim relevant to the motion or not}. This is the claim sentence task introduced by ~\newcite{cdcd2014}. For example, each of the following sentences includes a claim, marked in italic, for the motion topic in brackets. \begin{enumerate} \item \small (\textbf{the sale of violent video games to minors}) Recent research has suggested that some \emph{violent video games may actually have a pro-social effect in some contexts, for example, team play}. \item (\textbf{the right to bear arms}) Some gun control organizations say that \emph{increased gun ownership leads to higher levels of crime, suicide and other negative outcomes}. \end{enumerate} \subsubsection{Context Dependent Evidence Detection (CDED)} \label{sec:cded} \emph{Given a segment in a motion text the task is to identify whether the segment contains an evidence relevant to the motion or not}~\cite{evidence2015}. We consider evidences of two types in this paper, \emph{Study} and \emph{Expert}. Evidences of type study are generally results of a quantitative analysis of data given as numbers, or as conclusions. The following are two examples for study evidence relevant to the motion topic in brackets. \begin{itemize} \item \small (\textbf{the sale of violent video games to minors}) A 2001 study found that exposure to violent video games causes at least a temporary increase in aggression and that this exposure correlates with aggression in the real world. \item (\textbf{the right to bear arms}) In the South region where there is the highest number of legal guns per citizen only $59\%$ of all murders were caused by firearms in contrast to $70\%$ in the Northeast where there is the lowest number of legal firearms per citizen. \end{itemize} Evidence of type expert is a testimony by a person/group/commitee/organization with some known expertise/authority on the topic. The following are two examples for expert evidence relevant to the motion topic in brackets. \begin{enumerate} \item \small (\textbf{the sale of violent video games to minors}) This was also the conclusion of a meta-analysis by psychologist Jonathan Freedman, who reviewed over 200 published studies and found that the majority did not find a causal link. \item (\textbf{the right to bear arms}) University of Chicago economist Steven Levitt argues that available data indicate that neither stricter gun control laws nor more liberal concealed carry laws have had any significant effect on the decline in crime in the 1990s. \end{enumerate} \subsection{Textual Entailment (TE)} This task \cite{snli2015} corresponds to a multiclass setting, where given a pair of sentences (\emph{premise} and \emph{hypothesis}), the task is to identify whether one of them (\emph{premise}) entails, contradicts or is neutral with respect to the other sentence (\emph{hypothesis}). Unlike the other debating tasks seen previously, we cannot call these pair of sentences as context and target as these are more symmetric in nature. Typical examples\footnote{\url{http://nlp.stanford.edu/projects/snli/}} are the following (premise followed by hypothesis): \begin{itemize} \item \small Entailment: A soccer game with multiple males playing - Some men are playing a sport. \item Contradiction: A black race car starts up in front of a crowd of people - A man is driving down a lonely road. \item Neutral: A smiling costumed woman is holding an umbrella - A happy woman in a fairy costume holds an umbrella. \end{itemize} \subsection{Answer Selection for Questions} Question Answering (QA) System is a natural extension to the traditional commercial search engines as it is concerned with fetching answers to natural language queries and returning the information accurately in natural human language. A QA system can be either closed-domain or open-domain, the former being restricted to a particular domain while the latter is not. \emph{Answer sentence selection is a crucial subtask of the open-domain question answering problem, with the goal of extracting answers from a set of pre-selected sentences} \cite{wikiqa}. This is again bi-sequence classification task where the pair of sequences being a question and a candidate answer to be selected. \section{Deep Learning models for sequence pairs}\label{sec:models} All the tasks described in the previous section can be formulated as bi-sequence classification problems where we have to predict the label for the given pair of sequences. For simpler single sequence text classification tasks, RNN or CNN based architectures have become standard baselines. In this section, we will briefly introduce RNN and CNN and then subsequently describe RNN and CNN based architectures for bi-sequence classification tasks. Specifically, we talk about five different ways of handling context along with the target sentence. \subsection{Continuous Bag of Words (CBOW)}\label{sec:cbow} One of the simplest forms of sequence representation is the CBOW model, where every word in the sequence produces some word embedding (say, based on word2vec \cite{word2vec13}) and the average of the word embedding vectors over the words produces the representation of the sequence. As is evident, this form of representation totally disregards the word order of the sequence. \subsection{Recurrent neural networks (RNNs)} The RNN model provides a framework for conditioning on the entire history of the sequence without resorting to the Markov assumption traditionally used for modelling sequences. Unlike CBOW, RNNs encode arbitrary length sequences as fixed size vectors without disregarding the word order. Given an ordered list of $n$ input vectors $\textbf{x}_1,...,\textbf{x}_n$ and an initial state vector $\textbf{s}_0$, a RNN generates an ordered list of $n$ state vectors $\textbf{s}_0,...,\textbf{s}_n$ and an ordered list of $n$ output vectors $\textbf{y}_1,...,\textbf{y}_n$, that is, $RNN(\textbf{s}_0,\textbf{x}_1,...,\textbf{x}_n)=\textbf{s}_1,...,\textbf{s}_n,\textbf{y}_1,...,\textbf{y}_n$. The input vectors $\textbf{x}_i$ (which corresponds to a fixed dimensional representation for each word in the sequence) are presented to the RNN in a sequential fashion and $\textbf{s}_i$ represents the state of the RNN after observing the inputs $\textbf{x}_1,...,\textbf{x}_i$. The output vector $\textbf{y}_i$ is a function of the corresponding state vector $\textbf{s}_i$ and is then used for further prediction. An RNN is given by the following update equations: \begin{eqnarray} \textbf{s}_i &=& R(\textbf{x}_i,\textbf{s}_{i-1}) \\ \textbf{y}_i &=& O(\textbf{s}_{i}) \end{eqnarray} The recursively defined function $R$ takes as input the previous state vector $\textbf{s}_{i-1} $ and the current input vector $\textbf{x}_i \in \mathbb{R}^{d_{x}}$ and results in an updated state vector $\textbf{s}_i \in \mathbb{R}^{d_{s}}$. An additional function $O$ maps the state vector $\textbf{s}_i$ to an output vector $\textbf{y}_i \in \mathbb{R}^{d_{y}}$. Different instantiations of $R$ and $O$ will result in the different network structures (Simple RNN, LSTM~\cite{lstm97}, GRU~\cite{cho14}, etc.). The final state vector $\textbf{s}_n$ can be thought of as encoding the entire input sequence into a fixed size vector, which can be passed to a softmax layer to produce class probabilities. \subsection{Convolutional Neural Networks (CNNs)} CNNs are built on the premise of locality and parameter sharing which has proven to produce very effective feature representation for images. Following the groundbreaking work by \newcite{Kim14}, there has been a lot of interest shown by the text community towards applying CNNs for modelling text representation. As in case of RNNs defined above, an $n$-word sentence consists of embedding vectors $\textbf{x}_1,...,\textbf{x}_n \in \mathbb{R}^{d_{x}}$, one for each word in the sentence. Let $\textbf{x}_{i:i+j}$ denote the concatenation of words $\textbf{x}_i,\textbf{x}_{i+1},...,\textbf{x}_{i+j}$. A convolution operation defined by a non-linear function $f$ applies a filter $\textbf{w} \in \mathbb{R}^{h d_{x}}$ to a window of $h$ words to produce a single feature value as given below: \begin{eqnarray} c_i = f (\textbf{w} . \textbf{x}_{i:i+h-1} + b) \\ \textbf{c} = [c_1,c_2,...,c_{n-h+1}] \end{eqnarray} In the next step, max-pooling is applied which essentially produces a single feature value $\hat{c} = max\{\textbf{c}\}$, corresponding to one filter that has been used. The model can have multiple feature values, one for each applied filter, thus producing a feature representation for the input sentence, which can again be passed to a softmax layer to produce class probabilities. \subsection{Bi-Sequence RNN models} \label{sec:biseq-rnn} For bi-sequence classification tasks we use two RNNs, one RNN to encode the context sentence (\emph{context RNN}) and another separate RNN encode the target sentence (\emph{target RNN}). We define the following five different variants of combining these two RNNs for bi-sequence classification tasks (see Figure~\ref{fig:models} for illustration of these variants). \begin{enumerate} \item \small \textbf{conditional-state}: The final state of the context RNN is fed as the initial state of the target RNN. This way of handling context for RNNs has been previously used in conversational systems~\cite{Vinyals:15}, image description~\cite{vinyals2015show} and image question answering~\cite{Mengye:15} systems. \item \textbf{conditional-input}: The final state of the context RNN is fed as auxiliary input (concatenated with every input) for the target RNN. This way of handling context has been previously used in machine translation tasks~\cite{sutskever2014sequence}. \item \textbf{conditional-state-input}: The final state of the context RNN is fed as the initial state of the target RNN and also fed as input for target RNN concatenated with every input. \item \textbf{concat}: The final states of both the context and the target RNN are concatenated and then fed to a softmax layer for the label prediction. \item \textbf{bi-linear}: The final states of both the context and the target RNN are combined using a bi-linear form ($\textbf{x}^{\top}\textbf{W}\textbf{y}$) with a softmax function for the final prediction. There are different $\textbf{W}$ for different classes under consideration. \end{enumerate} From here on, we would refer to architecture types 1, 2 and 3 as \emph{conditional} variants while the others will be addressed as is. In addition, we consider another baseline variant \textbf{concat-sentence}, in which we concatenate the context and the sentence with a special separator token and feed the entire concatenated sequence to a single RNN. For all these variants we use a common embedding layer. Also note that the conditional variants require a common RNN size for both the context and the target RNNs. Even though that restriction is not there for other variants, we choose the same RNN size anyways for convenience. \begin{figure*}[t] \centering \subfigure[conditional-state]{\includegraphics[width=0.32\textwidth]{figs/conditional_state}} \subfigure[conditional-input]{\includegraphics[width=0.32\textwidth]{figs/conditional_input}} \subfigure[conditional-state-input]{\includegraphics[width=0.32\textwidth]{figs/conditional_input_state}} \subfigure[concat]{\includegraphics[width=0.32\textwidth]{figs/concat}} \subfigure[bilinear]{\includegraphics[width=0.32\textwidth]{figs/bilinear}} \subfigure[concat-sentence]{\includegraphics[width=0.32\textwidth]{figs/concat_sentence}} \caption{RNN based Architectures for bi-sequence classification.}\label{fig:models} \end{figure*} \begin{figure*}[t] \centering \subfigure[concat-bilinear-all-combinations]{\label{fig:models11}\includegraphics[width=0.45\textwidth]{figs/concat-bilinear}} \subfigure[conditional-all-combinations]{\label{fig:models12}\includegraphics[width=0.45\textwidth]{figs/cond-all}} \caption{Multiple model variants for bi-sequence classification.}\label{fig:models1} \end{figure*} \subsection{Bi-Sequence model variants} \label{sec:biseq-all} In this paper, we consider multiple ways of extending the bi-sequence architectures mentioned in section ~\ref{sec:biseq-rnn}, by replacing RNN with CBOW or CNN either for context or target or both. For the variations \emph{concat} and \emph{bi-linear} (see Fig. ~\ref{fig:models11}), we have considered CBOW, RNN and CNN for context representation whereas we have RNN and CNN for target representation. In tasks where context has very few words (say debating tasks or question-answering), a simple representation like CBOW may work for context. However, we haven't considered modelling target using CBOW as targets are usually of larger length. For the \emph{conditional} variants (see Fig. ~\ref{fig:models12}), we haven't considered CBOW due to their limited modelling capacity and there is no softmax layer directly on top of context representation to compensate for it (even though softmax is only on top of target RNN). Moreover, target can only be RNN as there is no concept of hidden state for CNNs. Hence, we use CNN and RNN for context whereas RNN for target. In addition, we consider the concat-sentence (mentioned in section ~\ref{sec:biseq-rnn}) as a baseline. This leads to 19 architectures for empirical comparison (12 from Fig. ~\ref{fig:models11} and 6 from Fig. ~\ref{fig:models12} and the baseline). \section{Experiments} \label{sec:experiments} We have carried out extensive evaluation of the above architecture variants over a wide range of datasets related to argumentation mining as well as datasets appealing to the larger natural language community like textual entailment and question answering data. We consider data with class imbalance problem as well as balanced and we do not restrict ourselves to binary classification by working with multiclass dataset as well. As can be found in the Table~\ref{debater-data}, we consider the following tasks related to the domain of argumentation mining\cite{aharoni14}, which is available here \footnote{\url{https://www.research.ibm.com/haifa/dept/vst/mlta_data.shtml}} : \begin{itemize} \item \small Claim Sentence : This is the dataset for the CDCD task defined in section ~\ref{sec:cdcd}. This is the current benchmark dataset for the Claim Detection task. There are a total 47183 canditate claims distributed among 33 motions. \item EXPERT Evidence : This is corresponding to the CDED task defined in section ~\ref{sec:cded} for evidence type EXPERT. There are 56985 labelled candidates for 57 different motion topics. \item STUDY Evidence : For evidence type STUDY, the dataset consists of 33534 labelled candidates for 49 motion topics. \end{itemize} Table~\ref{debater-data} summarizes all of the datasets above. Interesting point to be noted here is that all the datasets above have very low number of positives and the architectures we are evaluating need to be resilient to the class imbalance problem for these datasets. Other than the debating datasets listed above, we also consider two datasets related to more popular problems in the natural language processing community: \begin{itemize} \item \small Textual Entailment (TE) \footnote{\url{http://nlp.stanford.edu/projects/snli/}} \cite{snli2015} :This dataset consist of around 500K instances evenly distributed across all three classes. So, here we have a multiclass problem in a balanced setting. \item{WikiQA \footnote{\url{http://aka.ms/WikiQA}} \cite{wikiqa} : There are around 29K labelled question/answer pairs at our disposal.} \end{itemize} \begin{table}[!h] \small \parbox{.4\linewidth}{ \centering \begin{tabular}{|l|l|l|l|} \hline \bf \small{Task} & \bf \small{Motions} & \bf \small{Data Size} & \bf \small{Positives} \\ \hline \small{Claim Sentence} & 33 & 47183 & 2.77\% \\ \small{EXPERT Evidence} & 57 & 56985 & 4.56\% \\ \small{STUDY Evidence} & 49 & 33534 & 3.74\% \\ \hline \end{tabular} \caption{\label{debater-data} Argument Mining Datasets.} } \hfill \parbox{.5\linewidth}{ \centering \begin{tabular}{|l|r|l|l|l|l|} \hline \bf \small{Task} & \bf \small{Train} & \bf \small{Dev} & \bf \small{Test} & \bf \small{Problem} & \bf \small{Class} \\ \hline \small{TE} & \small{549367} & \small{9842} & \small{9824} & \small{Multiclass}& \small{Balance} \\ \small{WikiQA} & \small{20360} & \small{2733} & \small{6165} & \small{Binary} & \small{Imbalance} \\ \hline \end{tabular} \caption{\label{public-data} More Datasets.} } \end{table} \subsection{Experimental Setup} For each of the architectures mentioned in section ~\ref{sec:biseq-all}, we choose the best configuration of hyperparameters based on the validation portion of the particular dataset (For Claim and Evidence datasets, we consider a train:valid:test split of 60:10:30 while for the TE and WikiQA datasets we consider their corresponding given split). Performance of the best performing configuration for every architecture is reported on the test data using the appropriate metric. As the claim and expert and study Evidences had similar data characteristics (in terms of data size and context and target lengths), we did extensive hyperparam tuning only on the claim dataset and applied the best configurations without further tuning to the expert and study datasets. Exhaustive hyperparam tuning was done on the TE dataset as well because its data characteristics are very different from other datasets. In addition to reporting the test metrics for argumentation datasets, we carried out Leave-One-Out(leaving one motion out for testing) Mode training and evaluation which is more appropriate for this problem setting as it is crucial that we generalize well to totally unseen motion topics. In this case, we report the macro-average metrics over all motions. \subsection{Hyperparameter Tuning} Considering the number of variations of combinations of architectures we have considered, we have a huge hyperparameter space to deal with. Hence, we decided to fix insignificant hyperparameters and focus only on the relevant ones. We have decided to use word2vec \cite{word2vec13} pretrained models for initializing the word embeddings across CBOW, CNN and RNNs and made them trainable specific to task at hand. In addition we have found through minimal tuning that the Adam \cite{adam14} optimizer seems to work best. We have also found that a learning rate of 0.001 works best in most scenarios except when the parameter space in some architectures (for ex, \emph{bi-linear}) is large, in which case lower learning rates of 0.0001 or 0.00001 worked well. Rather than tuning the maximum sequence length for context and target sentence, we tried to fix it by getting reasonable values by plotting histogram of sequence lengths and had a cut-off at around 98-99 percentile. The max lengths turned out to be 14 for context and 60 for target for Claim, Evidence and WikiQA datasets while for textual entailment, they turned out to be 30 in both due to the symmetrical nature between premise and hypothesis. We tuned the following hyperparameters:\\ \small \textbf{RNN Model} : GRU or LSTM.\\ \textbf{RNN Size} : 50,100,200,300,400,500,1000.\\ \textbf{CNN Filter Sizes} : 3,3+4,3+4+5,2+3+4+5.\\ \textbf{CNN Number of Filters} : 10,20,40,64,128.\\ \textbf{L2 Reg coeff for CNN} : 0, 0.01, 0.001, 0.0001.\\ \normalsize For every architecture type, we carried out the optimization of the relevant hyperparams from the above list over the whole grid. One point to note is that for certain architectures like conditional-state, as output of context is fed in to the hidden state of target RNN, there are some restrictions in the allowable context RNN/CNN hyperparam configurations based on the target RNN settings as the output dimension of the context RNN/CNN needs to match the hidden state dimension of the target RNN. \subsection{Evaluation Metrics} For the datasets Claim, Expert, Study Evidence and WikiQA datasets, we have used standard evaluation measures like Average Precision (Area under Precision Recall Curve) and AUC to choose best hyperparam configurations based on validation data as well as report test metrics. For argument mining specific tasks, we have reported other additional metrics like P@200, R@200, F1@200, P@50, R@50 and F1@50 \cite{cdcd2014} in addition to reporting AUC and Average Precision. Please note for leave-one-out mode, the reported metrics are macro-average over all motion topics. For Textual entailment, since it is a more balanced dataset, reporting valid and test accuracies are standard in the literature and we have done the same. \begin{table*}[!ht] \scriptsize \centering \begin{tabular}{|l|l|l|l|r|} \hline \bf Task &\bf Context & \bf Target & \bf Architecture & \bf Test AVGP\\ \hline \multirow{3}{*}{Claim Sentence} & RNN & CNN & \bf Concat & \bf 0.307 \\ \cline{2-5} & CNN & CNN & Concat & 0.304 \\ \cline{2-5} & \multicolumn{3}{l|}{Concat-Sentence baseline} & 0.17 \\ \hline \hline \multirow{3}{*}{EXPERT Evidence} & RNN & RNN & \bf Conditional-State-Input & \bf 0.257 \\ \cline{2-5} & CNN & CNN & Concat & 0.254 \\ \cline{2-5} & \multicolumn{3}{l|}{Concat-Sentence baseline} & 0.225 \\ \hline \hline \multirow{3}{*}{STUDY Evidence} & CNN & CNN & \bf Concat & \bf 0.297 \\ \cline{2-5} & RNN & CNN & Concat & 0.29 \\ \cline{2-5} & \multicolumn{3}{l|}{Concat-Sentence baseline} & 0.236 \\ \hline \hline \multirow{2}{*}{WikiQA} & CBOW & RNN & \bf Concat & \bf 0.187 \\ \cline{2-5} & CNN & RNN & Conditional-State-Input & 0.186 \\ \hline \end{tabular} \caption{Empirical evaluation based on Average Precision on assymetric datasets.} \label{table-avgp-summary} \end{table*} \begin{table*}[!ht] \scriptsize \centering \begin{tabular}{|l|l|l|l|r|} \hline \bf Task &\bf Context & \bf Target & \bf Architecture & \bf Test AUC\\ \hline \multirow{3}{*}{Claim Sentence} & CNN & CNN & \bf Concat & \bf 0.873 \\ \cline{2-5} & CNN & RNN & Conditional-State & 0.873 \\ \cline{2-5} & \multicolumn{3}{l|}{Concat-Sentence baseline} & 0.831 \\ \hline \hline \multirow{3}{*}{EXPERT Evidence} & RNN & RNN & \bf Conditional-State & \bf 0.832 \\ \cline{2-5} & RNN & RNN & Conditional-State-Input & 0.823 \\ \cline{2-5} & \multicolumn{3}{l|}{Concat-Sentence baseline} & 0.805 \\ \hline \hline \multirow{3}{*}{STUDY Evidence} & CNN & CNN & \bf Concat & \bf 0.87 \\ \cline{2-5} & CBOW & CNN & Concat & 0.864 \\ \cline{2-5} & \multicolumn{3}{l|}{Concat-Sentence baseline} & 0.844 \\ \hline \hline \multirow{2}{*}{WikiQA} & CNN & CNN & \bf Concat & \bf 0.74 \\ \cline{2-5} & CBOW & RNN & Concat & 0.74 \\ \hline \end{tabular} \caption{Empirical evaluation based on AUC on assymetric datasets.} \label{table-auc-summary} \end{table*} \begin{table*}[!ht] \centering \resizebox{\textwidth}{!}{\begin{tabular}{|l|l|r|r|} \hline \bf Method &\bf Model & \bf TrainAcc(\%) & \bf TestAcc(\%) \\ \hline \cite{snli2015} & Use of features incl unigrams and bigrams & 99.7 & 78.2\\ \cite{vendrov15} & 1024D GRU encoders w/ unsupervised 'skip-thoughts' pre-training & 98.8 & 81.4\\ \cite{mou15} & 300D Tree-based CNN encoders & 83.3 & 82.1 \\ \hline \cite{cheng16} & 450D LSTMN with deep attention fusion & 88.5 & 86.3\\ \cite{parikh16} & 200D decomposable attention model with intra-sentence attention & 90.5 & \bf 86.8\\ \hline \small{\bf Conditional-State-RNN-RNN} & Simple architecture with RNNs without attention & 89.97 & 82.36 \\ \hline \end{tabular}} \caption{Comparison with the state-of-the-art in Textual Entailment dataset.} \label{table-te} \end{table*} \begin{table*}[!ht] \small \centering \resizebox{\textwidth}{!}{\begin{tabular}{|l|r|r|r|r|r|r|r|r|} \hline \bf Method &\bf P@200 & \bf R@200 & \bf F1@200 & \bf P@50 & \bf R@50 & \bf F1@50 & \bf AVGP & \bf AUC \\ \hline CDCD \cite{cdcd2014}** & 9.0 & \bf 73.0 & - & 18.0 & 40.0 & - & - & - \\ BoW \cite{lippi2015} & 8.2 & 51.7 & 14.2 & - & - & - & 0.117 & 0.771\\ TK \cite{lippi2015} & 9.8 & 58.7 & 16.8 & - & - & - & 0.161 & 0.808\\ TK+Topic \cite{lippi2015} & \bf 10.5 & 62.9 & \bf 18.0 & - & - & - & 0.178 & 0.823\\ \hline \bf Concat-CNN-CNN & 9.64 & 61.5 & 15.8 & 17.1 & 27.7 & 19.2 & 0.173 & 0.812 \\ Conditional-State-Input-RNN-RNN & 9.56 & 60.0 & 15.6 & 16.6 & 26.9 & 18.5 & 0.162 & 0.801 \\ \hline \end{tabular}} \caption{\small Results in Leave-One-Motion-Out mode for Claim Sentence Task. **\newcite{cdcd2014} used a smaller version of the dataset consisting of only 32 motions and also less number of claims. For fair comparison, we also use the same version of dataset as in CDCD and report the results in Appendix ~\ref{sec:supp}. } \label{table-lomo-claim} \end{table*} \begin{table*}[!ht] \small \centering \resizebox{\textwidth}{!}{\begin{tabular}{|l|r|r|r|r|r|r|r|r|r|r|r|} \hline \bf Task &\bf P@200 & \bf R@200 & \bf F1@200 & \bf P@50 & \bf R@50 & \bf F1@50 & \bf P@20 & \bf P@10 & \bf P@5 & \bf AVGP & \bf AUC \\ \hline Claim Sentence Task & 9.64 & 61.5 & 15.8 & 17.1 & 27.7 & 19.2 & 22.4 & 27.9 & 28.5 & 0.173 & 0.812\\ EXPERT Evidence Task & 9.53 & 64.0 & 14.5& 14.5 & 35.0 & 15.7& 18.6 & 21.1 & 22.5 & 0.160 & 0.750\\ STUDY Evidence Task & 8.33 & 79.5 & 13.5 & 15.5 & 53.9 & 18.9 & 20.8 & 25.3 & 31.8 & 0.298 & 0.836\\ \hline \end{tabular}} \caption{Numbers in Leave-One-Motion-Out mode for all three debating tasks using our approach.} \label{table-lomo-all} \end{table*} \subsection{Results and Discussion} Tables ~\ref{table-avgp-summary} and ~\ref{table-auc-summary} report the two top ranking architectures for four datasets based on Test AUC and Test Avg Precision. We find that \textbf{Concat} is the winning architecture variant across majority of the datasets considered. Moreover, the runner-up architecture type \textbf{Conditional-State-Input} is also similar to 'Concat' in the sense that concatenation of context representation is done at the input of the sentence RNN. Now the four datasets we considered are asymmetric in nature as there are significantly fewer contexts (motions or questions) than the targets. Hence the context model does not see enough data for learning and hence, if the learnt context model is fed directly to the hidden state of the target RNN, the improperly learnt context model can play a big role. In contrast if a Concat kind of architecture is used, the linear plus softmax layer can decide on how much importance to give to the context model. Hence, Concat is doing better in this case. From Textual entailment dataset(which is symmetric in nature), we found that conditional type of architectures are doing better at the Test accuracies. In fact, the winning architecture was \textbf{Conditional-State} with RNN-RNN combo, which did better in terms of test accuracy than the feature based models \cite{snli2015} and one tree-based model \cite{mou15}. However, it came close to the state-of-the-art attention based model \cite{parikh16}. In our work we are empirically evaluating simple architectures for bisequence classification without using more sophisticated tree-based or attention-based models. It is possible that adding attention on top of this will improve the results further. The \emph{bi-linear} model, is supposed to capture the interaction between the context and target reps via a quadratic form (section ~\ref{sec:biseq-rnn}). For the asymmetric datasets, this is not doing well again due to insufficient data for context. Whereas, it does well for the TE data. However, due to the huge parameter space for bi-linear, training times are considerably higher and requires lower learning rate than other architecture types. The runtimes are comparable for the other architecture variants. From Table ~\ref{table-lomo-claim}, the main takeaway is that we are the only deep learning based method with zero feature engineering and we have come very close to the state-of-the-art systems \cite{cdcd2014} and \cite{lippi2015}, which are heavily feature-engineered. Here again the winner is a 'Concat' based combination of architecture. Moreover, Tables ~\ref{table-lomo-claim} and ~\ref{table-lomo-all} are the first deep learning zero feature engineered baselines for all argument mining datasets. Appendix ~\ref{sec:supp} contains the details of the exhaustive experiments on all architectures on the different datasets in terms of the best hyperparameters used. \section{Conclusion} In this work, we have considered taking up multiple architectures for bisequence classification tasks, for which not much understanding is there in the current literature. In addition to suggesting winning architecture recipes for different kinds of datasets, we have established deep learning based baselines for argument mining tasks with zero feature engineering. As future work, it remains to be seen how adding attention on top of winning simple architectures fare in terms of benchmark performance. \small \section{Acknowledgements} We would like to thank Mitesh M. Khapra for the multiple discussions that we had with him leading to this paper. In addition, we are also grateful to our colleagues at IBM Debating Technologies for the numerous suggestions and feedback at different points of time. \bibliographystyle{acl}
03df1d25322bbd4914a5768042c64fe6efc97101
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Menger co-analytic groups} We shall assume all spaces are completely regular. \begin{defn} A topological space is \textbf{analytic} if it is a continuous image of $\mathbb{P}$, the space of irrationals. A space is \textbf{Lusin} if it is an injective continuous image of $\mathbb{P}$. (This is the terminology of \cite{RJ}. This term is currently used for a different concept.) A space is \textbf{$K$-analytic} if it is a continuous image of a Lindel\"of \v{C}ech-complete space. A space is \textbf{$K$-Lusin} if it is an injective continuous image of a Lindel\"of \v{C}ech-complete space. \end{defn} \begin{defn} A space is \textbf{co-analytic} if $\beta X\setminus X$ is analytic. In general, we call $\beta X\setminus X$ \textbf{the remainder} of $X$. $b X\setminus X$, for any compactification $b X$ of $X$, is called \textbf{a remainder} of $X$. \end{defn} \begin{defn} A space is \textbf{Menger} if whenever $\{\mathcal{U}_n : n < \omega\}$ is a sequence of open covers, there exist finite $\mathcal{V}_n$, $n < \omega$, such that $\mathcal{V}_n \subseteq \mathcal{U}_n$ and $\bigcup \{\mathcal{V}_n : n < \omega\}$ is a cover. \end{defn} Arhangel'ski\u\i ~\cite{Ar2} proved that Menger analytic spaces are $\sigma$-compact, generalizing Hurewicz's classic theorem that Menger completely metrizable spaces are $\sigma$-compact. Menger's conjecture was disproved in \cite{MF}, where Miller and Fremlin also showed it undecidable whether Menger co-analytic sets of reals are $\sigma$-compact. In \cite{TT} we proved that Menger \v{C}ech-complete spaces are $\sigma$-compact and obtained various sufficient conditions for Menger co-analytic topological spaces to be $\sigma$-compact. We continue that study here. In \cite{TT} we observed that $\mathbf{\Pi}_{1}^1$-determinacy -- which we also call \textbf{CD}: the \emph{Axiom of Co-analytic Determinacy} -- implies Menger co-analytic sets of reals are $\sigma$-compact. Indeed, \textbf{PD} (\emph{the Axiom of Projective Determinacy}) implies Menger projective sets of reals are $\sigma$-compact \cite{T2}, \cite{TT}. When one goes beyond co-analytic spaces in an attempt to generalize Arhangel'ski\u\i's theorem, one runs into ZFC counterexamples, but it is not clear whether there is a ZFC co-analytic counterexample. Assuming $V=L$, there is a counterexample which is a subset of $\mathbb{R}$ \cite{MF}, \cite{TT}. Here we prove: \begin{thm}\label{thm1} \textbf{CD} implies every Menger co-analytic topological group is $\sigma$-compact. \end{thm} \begin{rem} \textbf{CD} follows from the existence of a measurable cardinal \cite{M}. \end{rem} We first slightly generalize the \textbf{CD} result quoted above. \begin{lem}\label{lem1} \textbf{CD} implies every separable metrizable Menger co-analytic space is $\sigma$-compact. \end{lem} In order to prove this, we need some general facts about analytic spaces and perfect maps. \begin{lem}\label{lemA} Metrizable perfect pre-images of analytic spaces are analytic. \end{lem} \begin{proof} Rogers and Jayne \cite[5.8.9]{RJ} prove that perfect pre-images of metrizable analytic spaces are $K$-analytic, and that $K$-analytic metrizable spaces are analytic \cite[5.5.1]{RJ}. \end{proof} \begin{lem}[{ \cite[3.7.6]{E}}]\label{lemB} If $f:X\to Y$ is perfect, then for any $B\subseteq Y$, $f_B:f^{-1}(B)\to B$ is perfect. \end{lem} \begin{lem}[{ \cite[5.2.3]{RJ}}]\label{lemC} If $f$ is a continuous map of a compact Hausdorff $X$ onto a Hausdorff space $Y$ and the restriction of $f$ to a dense subspace $E$ of $X$ is perfect, then $f^{-1}\circ f(E)=E$. \end{lem} \begin{lem}\label{lemD} Metrizable perfect pre-images of co-analytic spaces are co-analytic. \end{lem} \begin{proof} Let $M$ be a metrizable perfect pre-image of a co-analytic $X$. Let $p$ be the perfect map. Extend $p$ to $P$ mapping $\beta M$ onto $\beta X$. Then by Lemma \ref{lemC}, $P^{-1}\circ P(M)=M$, i.e. $P^{-1}(X)=M$. Then $P(\beta M\setminus M)=\beta X\setminus X$, since $P$ is onto and points in $M$ map into $X$. By Lemma \ref{lemB}, $P|P^{-1}(\beta X\setminus X)$ is perfect. But then $\beta M\setminus M$ is analytic by Lemma \ref{lemA}, so $M$ is co-analytic. \end{proof} \begin{proof}[Proof of Lemma \ref{lem1}] Let $X$ be separable metrizable Menger co-analytic. It is folklore (see e.g. \cite{K}) that every separable metrizable space $X$ is a perfect image of a $0$-dimensional one, and hence of a subspace $M$ of the Cantor space $\mathbb{K}\subseteq\mathbb{R}$. Then $M$ is Menger co-analytic, so by \textbf{CD} is $\sigma$-compact. But then so is $X$. \end{proof} \begin{lem}[{ \cite{Ar3}}]\label{lem2} A topological group with Lindel\"of remainder is a perfect pre-image of a metrizable space. \end{lem} Since analytic spaces are Lindel\"of, a co-analytic group is a perfect pre-image of a metrizable space. Since Menger spaces are Lindel\"of, a Menger co-analytic topological group $G$ is a perfect pre-image of a separable metrizable space $M$. In \cite{TT}, we proved \textit{perfect images of co-analytic spaces are co-analytic}, so $M$ is co-analytic and Menger and therefore $\sigma$-compact by \textbf{CD} and Lemma \ref{lem1}. Then $G$ is $\sigma$-compact as well.\qed After hearing about Theorem \ref{thm1}, S. Tokg\"oz \cite{To} proved: \begin{prop}\label{prop1} $V=L$ implies there is a Menger co-analytic group which is not $\sigma$-compact. \end{prop} \section{Productively Lindel\"of co-analytic spaces} \begin{defn} A space $X$ is \textbf{productively Lindel\"of} if for every Lindel\"of space $Y$, $X \times Y$ is Lindel\"of. \end{defn} We have extensively studied productively Lindel\"of spaces \cite{AAJT, AT, BT, DTZ, T2, T1, TTs}, as have other authors. Since productively Lindel\"of spaces consistently are Menger \cite{T1, AAJT, T2, RZ} it is natural to ask: \begin{prob}\label{prob1} Are productively Lindel\"of co-analytic spaces $\sigma$-compact? \end{prob} \begin{defn} A \textbf{Michael space} is a Lindel\"of space whose product with the space $\mathbb{P}$ of irrationals is not Lindel\"of. \end{defn} It is consistent that there is a Michael space, but it is not known whether there is one from ZFC. If there is no Michael space, then the space $\mathbb{P}$ of irrationals is productively Lindel\"of, co-analytic, nowhere locally compact, but not $\sigma$-compact. We shall prove: \begin{thm}\label{thm4} CH implies productively Lindel\"of co-analytic spaces which are nowhere locally compact are $\sigma$-compact. \end{thm} I do not know whether the unwanted ``nowhere locally compact" clause can be removed. It assures us that $\beta X \setminus X$ is dense in $\beta X$. Laying the groundwork for proving Theorem \ref{thm4}, we need some definitions and previous results. \begin{defn}[\cite{Ar1}] A space is of \textbf{countable type} if each compact set is included in a compact set of countable character. \end{defn} \begin{lem}[{~\cite{HI}}]\label{lem5} A completely regular space is of countable type if and only if some (all) remainder(s) are Lindel\"of. \end{lem} \begin{defn}[{\cite{A}}] A space is \textbf{Alster} if each cover by $G_\delta$'s such that each compact set is included in the union of finitely many members of the cover has a countable subcover. \end{defn} \begin{lem}[{~\cite{AAJT, T2}}]\label{lem6} Alster spaces of countable type are $\sigma$-compact. \end{lem} \begin{lem}[{~\cite{A}}]\label{lem7} CH implies productively Lindel\"of spaces of weight $\leq \aleph_1$ are Alster. \end{lem} We can now prove Theorem \ref{thm4}. Let $X$ be productively Lindel\"of, co-analytic, and nowhere locally compact. $\beta X \setminus X$ is analytic and hence Lindel\"of and separable. It is dense in $\beta X$, so $w(\beta X)$ and hence $w(X) \leq 2^{\aleph_0}=\aleph_1$. Then $X$ is Alster. Since $\beta X \setminus X$ is Lindel\"of, $X$ has countable type, so it is $\sigma$-compact. \hfill\qed For metrizable spaces, Repov\'s and Zdomskyy \cite{RZ} proved: \begin{prop}\label{prop8} If there is a Michael space and \textbf{CD} holds, then every co-analytic productively Lindel\"of metrizable space is $\sigma$-compact. \end{prop} We would like to drop the metrizability assumption, using: \begin{lem}[{~\cite{RZ}}]\label{lem9} If there is a Michael space, then productively Lindel\"of spaces are Menger. \end{lem} As in \cite{TT}, we run up against the unsolved problem: \begin{prob}\label{prob2} Is it consistent that co-analytic Menger spaces are $\sigma$-compact? \end{prob} However, we can apply the various partial results in the previous section and \cite{TT} to obtain: \begin{thm}\label{thm10} Suppose there is a Michael space and \textbf{CD} holds. Then if $X$ is co-analytic and productively Lindel\"of, then $X$ is $\sigma$-compact if either: \begin{enumerate} \item{ closed subspaces of $X$ are $G_\delta$'s,} \item[]{or} \item{$X$ is a $\mathbf{\Sigma}$-space,} \item[]{or} \item{$X$ is a $p$-space,} \item[]{or} \item{$X$ is a topological group.} \end{enumerate} \end{thm} \begin{proof} These conditions all imply under \textbf{CD} that Menger co-analytic spaces are $\sigma$-compact. $\mathbf{\mathop{\pmb{\sum}}}$-spaces and p-spaces are discussed in Section 3. \end{proof} The two hypotheses of Theorem \ref{thm10} are compatible, since it is well-known that CH is compatible with the existence of a measurable cardinal, and that CH implies the existence of a Michael space \cite{Mi}. Various other hypotheses about cardinal invariants of the continuum also imply the existence of a Michael space -- see e.g. \cite{Moore2}. These are all compatible with \textbf{CD}. We also have: \begin{thm}\label{thm12} There is a Michael space if and only if productively Lindel\"of \v{C}ech-complete spaces are $\sigma$-compact. \end{thm} \begin{proof} If there is no Michael space, the space of irrationals is productively Lindel\"of, and of course it is \v{C}ech-complete but not $\sigma$-compact. If there is a Michael space, productively Lindel\"of spaces are Menger, but we showed in \cite{TT} that Menger \v{C}ech-complete spaces are $\sigma$-compact. \end{proof} Repov\'s and Zdomskyy \cite{RZ} prove: \begin{prop}\label{prop13} Suppose $\operatorname{cov}(\mathcal{M}) > \omega_1$, and there is a Michael space. Then every productively Lindel\"of $\Sigma_{2}^1$ subset of the Cantor space is $\sigma$-compact. \end{prop} \begin{ex} The metrizability condition cannot be removed; \emph{Okunev's space} is a productively Lindel\"of continuous image of a co-analytic space, but is not $\sigma$-compact (see \cite{BT}). In more detail, consider the Alexandrov duplicate $A$ of the space $\mathbb{P}$ of irrationals. $A$ is co-analytic, since it has a countable remainder with a countable base. A countable metrizable space is homeomorphic to an $F_\sigma$ in the Cantor space, and so is analytic. Okunev's space is obtained by collapsing the non-discrete copy of $\mathbb{P}$ in $A$ to a point. Note that Okunev's space is not co-analytic. To see this, if it were, it would be of countable type by Lemma \ref{lem5}. In \cite{BT} we showed that this space is Alster but not $\sigma$-compact, which would contradict Lemma \ref{lem6}. \end{ex} \section{K-analytic and K-Lusin spaces} We take the opportunity to make some observations about \textit{K-analytic}, \textit{K-Lusin}, \textit{absolute Borel}, \emph{Frol\'ik}, and what {Arhangel'ski\u\i} \cite{Ar4} calls \emph{Borelian of the first type} spaces. These are all attempts to generalize concepts of Descriptive Set Theory beyond separable metrizable spaces. \begin{defn}[{ \cite{BT}}] A space is \textbf{Frol\'ik} if it is homeomorphic to a closed subspace of a countable product of $\sigma$-compact spaces. \end{defn} \begin{defn} A space $X$ is \textbf{absolute Borel} if it is in the $\sigma$-algebra generated by the closed sets of $\beta X$. A space $X$ is \textbf{Borelian of the first type} if it is in the $\sigma$-algebra generated by the open sets of $\beta X$. \end{defn} \begin{defn} A space is \textbf{projectively $\sigma$-compact} (\textbf{projectively countable}) if its continuous images in separable metrizable spaces are all $\sigma$-compact (countable). \end{defn} Frol\'ik \cite{F} showed that each Frol\'ik space is absolute $K_{\sigma\delta}$ (and therefore Lindel\"of), i.e. the intersection of countably many $\sigma$-compact subspaces of its \v{C}ech-Stone compactification (and conversely), and also is the continuous image of a \v{C}ech-complete Frol\'ik space, so that Frol\'ik spaces are absolute Borel and {$K$-analytic}. $K$-Lusin spaces are clearly $K$-analytic; $K$-Lusin spaces are also Frol\'ik \cite[5.8.6]{RJ}. Since $K$-analytic metrizable spaces are analytic and analytic Menger spaces are $\sigma$-compact \cite{Ar2}, we see that Menger K-analytic spaces are projectively $\sigma$-compact \cite{BT}. In \cite{T1} we proved that projectively $\sigma$-compact Lindel\"of spaces are \emph{Hurewicz}, so we conclude: \begin{thm} Menger K-analytic spaces are Hurewicz. \end{thm} \emph{Hurewicz} is a property strictly between $\sigma$-compact and Menger. A space is \textbf{Hurewicz} if every \v{C}ech-complete space including it includes a $\sigma$-compact subspace including it (This is equivalent to the usual definition -- see \cite{T2}). This theorem may give some inkling as to why it seems to be hard to find topological properties that imply Hurewicz spaces are $\sigma$-compact which don't in fact imply Menger spaces are $\sigma$-compact. There are, however, Hurewicz subsets of $\mathbb{R}$ which are not $\sigma$-compact --- see e.g. \cite{Ts}. There is a projectively $\sigma$-compact Frol\'ik space which is not $\sigma$-compact (Okunev's space -- see \cite{BT}). Okunev's space is also not \v{C}ech-complete, since Menger \v{C}ech-complete spaces are $\sigma$-compact \cite{TT}. There is a Frol\'ik subspace of $\mathbb{R}$ which is not \v{C}ech-complete, since ``\v{C}ech-complete" translates into being a $G_{\delta}$, and we know the Borel hierarchy is non-trivial. There are of course analytic subsets of $\mathbb{R}$ which are not absolute Borel and hence not Frol\'ik. Moore's L-space \cite{Moore} is projectively countable but not K-analytic. The reason is that all its points are $G_\delta$'s, which contradicts projectively countable for K-analytic spaces \cite[5.4.3]{RJ}. Since K-Lusin spaces are Frol\'ik, it is worth mentioning that: \begin{prop}[{ \cite{BT}}] There are no Michael spaces if and only if every Frol\'ik space is productively Lindel\"of. \end{prop} We could add to this ``if and only if every K-Lusin space is productively Lindel\"of''. \begin{proof} $\mathbb{P}$ is K-Lusin. \end{proof} Also of interest is: \begin{prop}[ {\cite[2.5.5]{RJ}}] K-analytic spaces are \textit{powerfully Lindel\"of}, i.e. their countable powers are Lindel\"of -- in fact they are K-analytic. \end{prop} \begin{thm}\label{thm32} Co-analytic Menger K-analytic spaces are $\sigma$-compact. \end{thm} \begin{cor} Suppose there is a Michael space. Then co-analytic productively Lindel\"of K-analytic spaces are $\sigma$-compact. \end{cor} Compare with \ref{prop8}. The Corollary follows from \ref{lem9}. In order to prove \ref{thm32} we need to know: \begin{defn}[{\cite{Ar4}}] A completely regular space is called an \textbf{s-space} if there exists a \textbf{countable open source} for $X$ in some compactification $bX$ of $X$, i.e. a countable collection $\mathcal{S}$ of open subsets of $bX$ such that $X$ is a union of some family of intersections of non-empty subfamilies of $\mathcal{S}$. \end{defn} We also need to know about \textbf{p-spaces} and \textbf{$\mathbf{\mathop{\pmb{\sum}}}$-spaces}, but do not need their internal characterizations. What we need are: \begin{lem}[{ \cite{Ar1}}] A completely regular space is \emph{\textbf{Lindel\"of {p}}} if it is the perfect pre-image of a separable metrizable space. \end{lem} \begin{lem}[{ \cite{Naga}}] A completely regular space is \textbf{\emph{Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$}} if and only if it is the continuous image of a Lindel\"of p-space. \end{lem} \begin{lem} An analytic space has a countable network and hence (see e.g \cite{G}) is Lindel\"of and a $\mathbf{\mathop{\pmb{\sum}}}$-space. \end{lem} \begin{lem}[{ \cite{Ar4}}]\label{lem37} $X$ is a Lindel\"of p-space if and only if it is a Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$-space and an s-space. \end{lem} \begin{lem}[{ \cite{Ar4}}] $X$ is Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$ if and only if its remainder is an s-space. \end{lem} \begin{lem}[{ \cite{Ar3}}] $X$ is a Lindel\"of p-space if and only if its remainder is. \end{lem} \begin{proof}[Proof of Theorem \ref{thm32}] Such a space $X$ is a Lindel\"of p-space, since both it and its remainder are Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$. Let $X$ map perfectly onto a metrizable $M$. Then $M$ is analytic and Menger, so is $\sigma$-compact, so $X$ is also. \end{proof} \begin{thm}\label{thm310} Co-analytic Menger absolute Borel spaces are $\sigma$-compact. \end{thm} To see this, we introduce: \begin{defn} Given a family of sets $\mathcal{S}$, Rogers and Jayne \cite{RJ} say that a set is a \textbf{Souslin $\mathcal{S}$-set} if it has a representation in the form \begin{equation*} \bigcup_{\sigma\in{^\omega\omega}}\bigcap_n\textbf{S}(\sigma|n) \end{equation*} with \textbf{S}$(\sigma|n)\in\mathcal{S}$ for all finite sequences of positive integers. \end{defn} Rogers and Jayne prove: \begin{lem}[{ \cite[2.5.4]{RJ}}]\label{lem314} The family $\mathcal{A}$ of $K$-analytic subsets of a completely regular space is closed under the Souslin operation i.e. every Souslin $\mathcal{A}$-set is in $\mathcal{A}$; if a family is closed under the Souslin operation, it is closed under countable intersections and countable unions. \end{lem} \begin{cor}\label{cor313} Absolute Borel spaces are $K$-analytic. \end{cor} \begin{proof} This is well-known. In $\beta X$, closed subsets are compact; compact spaces are $K$-analytic. \end{proof} Theorem \ref{thm310} now follows from \ref{thm32}. \begin{thm}\label{thm314} Every Lindel\"of Borelian space of the first type is K-analytic. \end{thm} \begin{proof} We proceed by induction on subspaces of a fixed compact space. For the basis step, note that open subspaces of a compact space are locally compact, while Lindel\"of locally compact spaces are $\sigma$-compact. For the successor stage, assume a Lindel\"of Borelian set of the first type is the union (intersection) of countably many K-analytic subspaces. By Lemma \ref{lem314}, the union (intersection) is K-analytic and hence Lindel\"of. The limit stage is trivial. \end{proof} {Arhangel'ski\u\i} \cite{Ar4} proved that Borelian sets of the first type are s-spaces. This is interesting because: \begin{thm}\label{lem39} Every absolute Borel s-space is a Lindel\"of p-space. \end{thm} \begin{proof} We induct on Borel order. The basis step is trivial. We need to show s-spaces which are the countable union (intersection) of Lindel\"of p-spaces are Lindel\"of \emph{p}. By \ref{lem37} it suffices to show they are Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$. Let $\{X_n\}_{n<\omega}$ be Lindel\"of \textit{p}. Let $\sum_{n<\omega} X_n$ be the disjoint sum of the $X_n$'s. Then $\sum_{n<\omega} X_n$ is clearly Lindel\"of \textit{p}. Consider the natural map $\sigma$ from $\sum_{n<\omega} X_n$ to $\bigcup_{n<\omega} X_n$ obtained by identifying all copies of a point $x\in\bigcup_{n<\omega}X_n$ which are in $\sum_{n<\omega} X_n$. $\sigma$ is continuous, so $\bigcup_{n<\omega} X_n$ is Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$. Now consider $\prod_{n<\omega} X_n$. This is also Lindel\"of \textit{p} \cite{Ar1} and so then is the diagonal $\Delta$. Define $\pi\left(\langle x,x, \dots\rangle\right)=x$. Then $\pi$ is continuous and maps $\Delta$ onto $\bigcap_{n<\omega} X_n$, which is therefore Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$. \end{proof} Note Okunev's space is Lindel\"of absolute $F_{\sigma\delta}$ but is not s, since it is not of countable type \cite{BT}, while s-spaces are \cite{Ar4}. By Theorem \ref{thm32}, Okunev's space is not co-analytic. Borel sets of reals are of course analytic; Okunev's space shows that Lindel\"of absolute Borel spaces need not be analytic, since it is Menger but not $\sigma$-compact. Compact spaces are Borelian of the first type, so the latter spaces need not be analytic. A somewhat smaller class of spaces than the $K$-analytic ($K$-Lusin) ones is comprised of what Rogers and Jayne call the \textit{proper $K$-analytic} (\textit{proper $K$-Lusin}) spaces. \begin{defn} A space is \textbf{proper $K$-analytic} if it is the perfect pre-image of an analytic subspace of $\mathbb{R}^\omega$. A space is \textbf{proper K-Lusin} if it is the perfect pre-image of a Lusin subspace of $\mathbb{R}^\omega$. \end{defn} Rogers and Jayne \cite{RJ} prove that a space is proper K-Lusin if and only if both it and its remainder are K-analytic. It follows that a space is proper K-Lusin if and only if it and its remainder are K-Lusin. They also prove that K-Lusin spaces are absolute $K_{\sigma\delta}$, i.e. what we have called Frol\'ik. It follows that proper K-Lusin spaces are both $K_{\sigma\delta}$ and $G_{\delta\sigma}$, i.e. countable unions of \v{C}ech-complete spaces. We shall provide a large number of equivalences for ``proper K-Lusin'' below. Proper K-analytic spaces are p-spaces, and their continuous real-valued images are analytic, so: \begin{thm} Menger proper K-analytic spaces are $\sigma$-compact. \end{thm} \begin{cor} Menger proper K-Lusin spaces are $\sigma$-compact. \end{cor} \begin{lem}[{ \cite{RJ}}]\label{lem319} Let $\mathfrak{Z}(Y)$ be the collection of zero-sets of $Y$. Then $X$ is proper $K$-analytic if and only if $X\in\textbf{S}(\mathfrak{Z}(\beta X))$. \end{lem} \begin{thm} A space is proper K-analytic if and only if it is a K-analytic p-space. \end{thm} \begin{proof} By definition, a proper K-analytic space is a p-space. By \ref{lem319} and \ref{lem314} it is K-analytic. Conversely, if $X$ is a K-analytic p-space, it maps perfectly onto a separable metrizable analytic space, which embeds into $\mathbb{R}^\omega$. \end{proof} Note that zero-sets are closed $G_\delta$'s, so that the \textit{absolute Baire sets}, i.e. the elements of the $\sigma$-algebra generated by the zero-sets, are both Lindel\"of Borelian of the first type and absolute Borel. \begin{cor} Menger absolute Baire spaces are $\sigma$-compact. \end{cor} Mixing Rogers and Jayne with {Arhangel'ski\u\i}, we have: \begin{thm}\label{thm318} The following are equivalent: \begin{itemize} \item[(a)] $X$ is proper K-Lusin, \item[(b)] $X$ and its remainder are K-Lusin, \item[(c)] $X$ and its remainder are both Frol\'ik, \item[(d)] $X$ is Lindel\"of Borelian of the first type, \item[(e)] $X$ is absolute Borel and Lindel\"of p, \item[(f)] $X$ is absolute Borel and of countable type. \end{itemize} \end{thm} \begin{proof} We have already proved that (a), (b) and (c) are equivalent. (c) implies (d), since $X$ is Lindel\"of absolute $G_{\delta\sigma}$. If $X$ is Lindel\"of Borelian of the first type, it is K-analytic, but so is its remainder, so (d) implies (b). If $X$ is absolute Borel, it is K-analytic and its remainder is Borelian of the first type. If $X$ is Lindel\"of p, so is its remainder, so (e) implies (b). (b) implies a proper K-Lusin space and its remainder are both K-analytic spaces, hence Lindel\"of $\mathbf{\mathop{\pmb{\sum}}}$ spaces, so they are p-spaces. Thus (b) implies (e). (e) implies (f) since p-spaces are of countable type \cite{Ar1}. (f) implies the remainder of $X$ is Lindel\"of Borelian of the first type, and so is K-analytic. Then since $X$ is K-analytic, (f) implies (b). \end{proof} We know that Menger proper K-analytic (a fortiori, proper K-Lusin) spaces are $\sigma$-compact, but Menger K-analytic spaces may not be. \begin{prob} Are Menger K-Lusin spaces $\sigma$-compact? \end{prob} An interesting fact about K-Lusin spaces is that: \begin{lem}[{ \cite[5.4.3]{RJ}}]\label{lem322} The following are equivalent for a K-Lusin $X$: \begin{itemize} \item[(a)] $X$ includes a compact perfect set; \item[(b)] $X$ admits a continuous real-valued function with uncountable range; \item[(c)] $X$ is not the countable union of compact subspaces which include no perfect subsets. In particular, if $X$ is not $\sigma$-compact, it includes a compact perfect set. \end{itemize} \end{lem} From this, we can conclude that Okunev's space is not K-Lusin, since it is not $\sigma$-compact but doesn't include a compact perfect set. Indeed we have: \begin{defn} A space is \textbf{Rothberger} if whenever $\{\mathcal{U}_n\}_{n<\omega}$ are open covers, there exists a cover $\{{U}_n\}_{n<\omega},U_n\in\mathcal{U}_n$. \end{defn} Thus \textit{Rothberger} is a strengthening of \textit{Menger}. \begin{lem}[{ \cite{AA}}] Rothberger spaces do not include a compact perfect set. \end{lem} \begin{thm} K-analytic Rothberger spaces are projectively countable. \end{thm} \begin{proof} They are projectively $\sigma$-compact. \end{proof} \begin{cor} K-Lusin Rothberger spaces are $\sigma$-compact. \end{cor} \begin{proof} This follows from \ref{lem322}. \end{proof} \begin{rem} Projectively countable Lindel\"of spaces are always Rothberger \cite{T1}; thus Okunev's space is Rothberger \cite{BT}. The assertion that Rothberger spaces are projectively countable is equivalent to \textit{Borel's Conjecture} \cite{T1}. \end{rem} Here are some more problems we have not been able to solve: \begin{prob} Does \textbf{CD} imply co-analytic Hurewicz spaces are $\sigma$-compact? \end{prob} \begin{prob} Are Lindel\"of co-analytic projectively $\sigma$-compact spaces $\sigma$-compact? \end{prob} Note $V=L$ implies there is a co-analytic Hurewicz group of reals that is not $\sigma$-compact \cite{To}. \nocite{*} \bibliographystyle{acm}
ff4a6df1a52f4891034852c14c652312e4fea285
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} There are many open questions regarding the strength and geometry of the magnetic field in radio galaxies and their relation to other properties of the radio source. The observed degree of polarization depends on the intrinsic properties, such as the regularity and orientation of the source magnetic fields as well as the Faraday effects from the intervening regions of ionized gas along the line of sight. The largest current sample of polarized sources is the NRAO/VLA all sky survey, NVSS, at 1.4 GHz \citep{1998AJ....115.1693C}. It shows that the majority of extragalactic radio sources are only a few percent polarized. Polarization studies of small samples of extragalactic radio sources at other frequencies also show a similar weak average polarization, and suggest the fractional polarization increases at frequencies higher than 1.4 GHz \citep[e.g.][]{2009A&A...502...61M}. It is not clear which mechanism is dominant in reducing the fractional polarization at lower frequencies and depolarizing the sources, although several models have been suggested \citep{1966MNRAS.133...67B,1991MNRAS.250..726T,1998MNRAS.299..189S,2008A&A...487..865R,2015MNRAS.450.3579S}. One key cause for depolarization is Faraday rotation, which can be characterized to first order by a change in the angle of the linear polarization: \begin{equation} \Delta \chi=\left(0.812 \int \frac{n_e{\bf B}}{(1+z)^2}\cdot \frac{d{\bf l}}{dz} \,dz\right) \lambda^2 \equiv \phi \lambda^2 \end{equation} where $\Delta \chi$ is the amount of the rotation of the polarization vector in rad, $\lambda$ is the observation wavelength in m, $z$ is the redshift of the Faraday screen, ${\bf B}$ is the ionized medium magnetic field vector in $\mu$G, $n_e$ is the number density of electrons in the medium in cm$^{-3}$ and $\,d{\bf l}$ is the distance element along the line of sight in pc. The term in parentheses is called the Faraday depth, $\phi$. For a single line of sight through a thin ionized screen, this is equivalent to the rotation measure, $\textrm{RM}$, defined by $\textrm{RM} \equiv \frac{\Delta \chi}{\Delta \lambda^2}$ which can be measured observationally. Different lines of sight to the source all within the observing beam can have different values of $\phi$. Typically, this progressively depolarizes the source at longer wavelengths, but it can also lead to constructive interference and re-polarization, i.e., higher fractional polarizations at longer wavelengths. There are at least three separate possible Faraday screens with different $\textrm{RM}$ distributions along the line of sight: the Galactic component, intervening extragalactic ionized gas, and material local to the source. Multiple studies such as \cite{2005MNRAS.359.1456G,2008ApJ...676...70K,2010MNRAS.409L..99S,2012ApJ...761..144B,2012arXiv1209.1438H,2013ApJ...771..105B,2014ApJ...795...63F,2014MNRAS.444..700B,2014PASJ...66...65A,2015aska.confE.114V,2015arXiv150900747V} have tried to identify and distinguish these separate components and study the evolution of the magnetic field of galaxies through cosmic time. When many lines of sight each have independent single Faraday depths, this problem is approached statistically. Another long standing puzzle is the anti-correlation between the total intensity of radio sources and their degree of polarization, as observed by many groups such as \cite{2002A&A...396..463M}, \cite{2004MNRAS.349.1267T}, \cite{2006MNRAS.371..898S}, \cite{2007ApJ...666..201T}, \cite{2010ApJ...714.1689G}, \cite{2010MNRAS.402.2792S} and \cite{2014ApJ...787...99S}. The physical nature of this relation has been a mystery for almost a decade, and is confused by the dependency on other source properties. \cite{2010ApJ...714.1689G} found that most of their highly polarized sources are steep spectrum, show signs of resolved structure on arc-second scales, and are lobe dominated. However, they found no further correlation between the spectral index and fractional polarization. The anti-correlation between total intensity and fractional polarization seems to become weak for very faint objects with 1.4 GHz total intensities between 0.5 mJy $< I <$ 5 mJy as suggested in \cite{2014ApJ...785...45R}, based on a small sample of polarized radio galaxies in the GOODS-N field \citep{2010ApJS..188..178M}. Recently, \cite{2015arXiv150406679O} studied a sample of 796 radio-loud AGNs with $z < 0.7$. They found that low-excitation radio galaxies have a wide range of fractional polarizations up to $\sim$ 30 \%, and are more numerous at faint Stokes I flux densities while high-excitation radio galaxies are limited to polarization degrees less than 15\%. They suggest that the ambient gas density and magnetic fields local to the radio source might be responsible for the difference. Using WISE colors, \cite{2014MNRAS.444..700B} suggested that the observed anti-correlation primarily reflects the difference between infrared AGN and star-dominated populations. Large samples of polarization data at multiple frequencies are required to understand the magnetic field structures and depolarization mechanisms responsible for the low observed polarization fractions. \cite{2013ApJ...771..105B} have showed the polarization fraction of compact sources decreases significantly at 189 MHz compared to 1.4 GHz. They studied a sample of 137 sources brighter than 4 mJy and only detected one polarized source with probably a depolarization mechanism intrinsic to the source. Recently, \cite{2014ApJS..212...15F} used the \cite{2009ApJ...702.1230T} (hereafter TSS09) catalog, and assembled polarization spectral energy distributions for 951 highly polarized extragalactic sources over the broad frequency range, 0.4 GHz to 100 GHz. They showed that objects with flat spectra in total intensity have complicated polarization spectral energy distributions (SEDs), and are mostly re-polarized somewhere in the spectrum, while steep spectrum sources show higher average depolarization. As a result, they claimed that the dominant source of depolarization should be the local environment of the source, since the spectral index is an intrinsic property of these highly polarized sources. The current work follows up on their discovery, using a sample selected only on the basis of total intensity at 2.3 GHz. In this work, we use the data from the S-PASS survey, conducted by the Australian Parkes single dish radio telescope at 2.3 GHz. We cross match the data with the NVSS catalog and generate a new independent depolarization catalog of bright extragalactic radio sources. Unlike other polarization studies such as \cite{2014ApJS..212...15F} and \citet{2012arXiv1209.1438H} our catalog is not selected based on high polarized intensity which enables us to include objects with low fractional polarizations as well. We study the evolution and possible correlation between quantities such as depolarization, spectral indices and $\textrm{RM}$s. We will tackle the nature of the well-known observed anti-correlation between total intensity and fractional polarization as well as the origin of the dominant component of depolarization. Section \ref{sec:obs} presents the 1.4 GHz and 2.3 GHz observations. Section \ref{sec:mapanalysis} explains the steps in our analysis of the S-PASS total intensity and polarization maps as well as the cross matching with the NVSS catalog. In Section \ref{quantities} we derive quantities such as spectral index, residual rotation measure, fractional polarization and depolarization. The main results and their implications are discussed in sections \ref{result} and \ref{discussion} respectively. At the end, Section \ref{summary} summarizes the main findings and conclusions. Throughout this paper we employ the $\Lambda$CDM cosmology with parameters of H$_0=70$ km.s$^{-1}$Mpc$^{-1}$, $\Omega_m=0.3$ and $\Omega_{\Lambda}=0.7$. \section{Observations} \label{sec:obs} \subsection{The 2.3 GHz Data}\label{spass} The S-PASS is a project to map the southern sky at Dec $<-1.0$ deg in total intensity and linear polarization. The observations were conducted with the 64-m Parkes Radio Telescope, NSW Australia. A description of S-PASS is given in \cite{2013Natur.493...66C} and \cite{2010ASPC..438..276C}; here we report a summary of the main details. The S-band receiver used is a circular polarization package with system temperature T$_{sys}$ = 20 K, and beam width FHWM= 8.9 arcmin at 2300 MHz. Data were collected with the digital correlator Digital Filter Banks mark 3 (DFB3) recording the two autocorrelation products (RR* and LL*) and their complex cross-correlation (RL*). The sources PKS B1934-638 and PKS B0407-658 were used for flux density calibration and PKS B0043-424 for polarization calibration. After Radio Frequency Interference (RFI) flagging, frequency channels were binned together covering the ranges 2176-2216 and 2256-2400 MHz, for an effective central frequency of 2307 MHz and bandwidth of 184 MHz. As described in \cite{2010ASPC..438..276C}, the observing strategy is based on long azimuth scans taken at the elevation of the south celestial pole at Parkes covering the entire Dec range (-89 deg to -1 deg) in each scan. For the current work, the spatial large scale component has been removed from each Stokes parameter, applying a high pass spatial filter to optimize for compact source finding and analysis. A median filter with a window of 45 arc-min was used. The final product was a set of 15$\times$15 deg$^2$ zenithal projection maps covering the entire sky observed by S-PASS. Final maps are convolved to a beam of FWHM = 10.75 arcmin. Stokes I , Q , and U sensitivity is better than 1.0 mJy beam$^{-1}$. Details of scanning strategy, map-making, and final maps are in \cite{2010ASPC..438..276C} and \cite{2013Natur.493...66C} and will be presented in full details in a forthcoming paper (Carretti et al. 2016, in preparation). The confusion limit is 6 mJy in Stokes I \citep{2013MNRAS.430.1414C} and much lower in polarization (average polarization fraction in compact sources is around 2\%, see this work). The instrumental polarization leakage is 0.4\% on-axis \citep{2010ASPC..438..276C} and less than 1.5\% off-axis. \subsection{The 1.4 GHz Data} The NVSS is a 1.4 GHz radio survey with the Very Large Array (VLA) covering the entire sky north of -40 degrees declination at a resolution of 45 arcsec (FWHM). The rms brightness fluctuations are approximately uniform across the sky at $\sim$0.45 mJy per beam in Stokes I and $\sim$0.29 mJy per beam in Stokes Q and U. The astrometry is accurate to within $<1$ arcsec for point sources with flux densities $>15$ mJy, and to $<7$ arcsec for the faintest detectable sources ($\sim$2.3 mJy in Stokes I). The survey has a completeness limit of 2.5 mJy, which resulted in a catalog of over 1.8 million discrete sources in Stokes I. More details about the NVSS can be found in \cite{1998AJ....115.1693C}. \section{Creating the new sample} \subsection{Cross-matching and selection criteria}\label{sec:mapanalysis} We first attempted to construct a joint S-PASS/NVSS catalog using NVSS I,Q, and U images convolved to the processed S-PASS resolution of $\sim$11'. However, upon convolution, the resulting NVSS images were very heavily mottled because of the lack of short interferometer baselines, and the noise level increased dramatically above the full resolution images. We therefore followed an alternative approach, viz., measuring the contributions of all individual NVSS sources at the position of each NVSS source, as described further below. There are rare situations where very-low level diffuse NVSS emission could also have contributed significantly to the S-PASS flux \citetext{e.g, cluster halos, \citealt{2001ApJ...548..639K}}, and would be missed by our procedure, but this very minor possible contribution to our strong total intensity sources has been ignored. We constructed the initial S-PASS catalog by searching the S-PASS maps at the position of all NVSS sources with $I_\mathrm{NV}~>~$10~mJy in the overlap region between the two surveys, and fitting Gaussian functions to the S-PASS total intensity images. For sources with a spectral index of -0.7 (-0.3) this would correspond to a 4(5) $\sigma$ detection in S-PASS. However, in order to have adequate sensitivity to sources with low fractional polarizations in S-PASS, we adopted a much higher threshold of $I_\mathrm{SP} >$ 420~mJy for the catalog. Duplicate sources were eliminated. Additional sources were eliminated from the catalog if they had either of these data quality issues:\\ a) Excess noise ($>$0.75~mJy per beam rms, 1.5 $\times$ the mode calculated in bins of 0.01 mJy) in the 7.5' - 11.25' annulus around the total intensity NVSS source;\\ b) Excess noise ($>$3~mJy per beam rms. 2$\times$ the average rms value) in the 45'-90' annulus in either Q or U maps in S-PASS.\\ We verified by visual inspection that the above selection criteria have successfully eliminated the NVSS and S-PASS regions with instrumental artifacts. At the processed S-PASS resolution of $\sim$11', many sources identified by the above procedure are actually blends of multiple NVSS sources. In order to derive meaningful information from the sample, we therefore needed to eliminate sources with significant contributions from blending. To do this, we defined a search radius of 16' (i.e., to the 3.5$\sigma$, 2$\times 10^{-3}$ level of the S-PASS beam) around each S-PASS source, and calculated the I,Q, and U contributions of each NVSS source (with $I_\mathrm{NV}>$10 mJy) at the position of the S-PASS source. Thus, for the NVSS portion of the catalog, we have two values for each Stokes parameter: X$_{Ntarget}$, the flux (I,Q, or U) of the NVSS source with the largest Stokes I contribution at the S-PASS position, and X$_{Ncont}$, the I,Q, or U flux from all other NVSS sources within the 16' search radius, scaled by their distance from the S-PASS peak position using a Gaussian kernel representing the S-PASS beam. The final values for comparison with S-PASS are then X$_{Ntotal} \equiv$ X$_{Ntarget}$ + X$_{Ncont}$. Figure \ref{contI2} shows the distribution of the percent contamination of the target source in NVSS total intensity, $\frac{I_{\text{cont}}}{I_{\text{target}}}$, and polarization, $\frac{P_{\text{cont}}}{P_{\text{target}}}$. We then adopted a 10\% polarization contamination threshold, and only selected sources with $\frac{P_{\text{cont}}}{P_{\text{target}}} < 0.1$. The joint S-PASS/NVSS catalog contains 533 sources meeting all of the above criteria. A description of the biases that could result from our contamination threshold is discussed in Section \ref{bias}. \begin{figure}[h] \centering \includegraphics[scale=0.51]{cont5.eps} \caption{ Distributions of the percentage of contamination in the NVSS total intensity, $100\times\frac{I_{cont}}{I_{target}}$, in black solid line and polarization flux density, $100\times\frac{P_{cont}}{P_{target}}$ in red dashed line are shown. The catalog only contains sources with $\frac{P_{cont}}{P_{target}} < 0.1$. } \label{contI2} \end{figure} \subsection{Derived quantities}\label{quantities} \subsubsection{NVSS and S-PASS polarized flux density, fractional polarization and depolarization} We calculated the polarization intensity (averaged over the entire bandwidth) for the NVSS and S-PASS surveys separately. The effect of bandwidth depolarization is discussed in section \ref{bwdepol}. We used Stokes $Q$ and $U$ to calculate the polarized intensity, $P$, in both NVSS and S-PASS as following: \begin{equation} P=\sqrt{Q^2+U^2} \end{equation} where for NVSS the $Q$ and $U$ include both the target and contamination flux density, $Q=Q_{\text{target}}+Q_{\text{cont}}$ and $U=U_{\text{target}}+U_{\text{cont}}$. The bias corrected polarized flux density, $P_{bc}$, is approximated as follows: \begin{equation} P_{\text{bc}}=\sqrt{Q^2+U^2-\sigma_{p}^2-\sigma_{\text{cont}}^2} \end{equation} where $\sigma_{p}$ is the global rms of $U$ or $Q$ maps ($\sigma^{NV}_{U,Q} \approx 0.3$ mJy per beam and $\sigma^{SP}_{U,Q} \approx 1.7$ mJy per 3-arcmin pixel), measured through the entire $Q$ and $U$ maps, and $\sigma_{\text{cont}}$ is the total contribution of the contaminant apertures rms noise to the bias in NVSS, scaled for their separation from the target. We also calculated the fractional polarization, $\pi$, \begin{equation} \pi_\mathrm{SP}=\frac{P^{SP}_{\text{bc}}}{I_\mathrm{SP}} \end{equation} \begin{equation} \pi_\mathrm{NV}=\frac{P^{NV}_{\text{bc}}}{I_\mathrm{NV}} \end{equation} where the NVSS total intensity is equal to the target plus the contamination flux density, $I_\mathrm{NV}=I_{\text{target}}+I_{\text{cont}}$. The NVSS residual instrumental polarization percentage peaks at $\epsilon_\mathrm{NV} \approx 0.12\%$ for a sample of strong and unpolarized sources \citep{1998AJ....115.1693C}. We used this value as a cutoff for the NVSS fractional polarization; for any sources below this threshold we report upper limits as the maximum of $\left(\frac{3\sigma_p}{I}, \epsilon_\mathrm{NV} \right)$. To estimate the S-PASS residual instrumental polarization we selected the 27 objects with $\pi_\mathrm{NV} < 0.12\%$ in our final sample, and plotted the distribution of their $\pi_\mathrm{SP}$ values (Figure \ref{fig:sprespi}). The median of the distribution, $\bar{\pi}_\mathrm{SP}=0.55\%$, which we assumed to be a good estimator of the S-PASS residual instrumental polarization percentage, $\epsilon_\mathrm{SP}$. Note that if the residual instrumental polarizations were zero, then the rms noise of 1.7 mJy per beam would result in much smaller fractional polarizations than $0.55\%$ for the 27 mentioned objects. On the other hand, objects with $\pi_\mathrm{NV} < 0.12\%$ can potentially be more polarized at higher frequencies, so we could be overestimating the instrumental contribution. We ignored this possibility, and chose the more conservative approach of assuming $\pi_\mathrm{SP}=0.55\%$ is only due to instrument leakage. \begin{figure}[h] \centering \includegraphics[scale=0.51]{sprespi.eps} \caption{ Distribution of S-PASS fractional polarization for 27 objects with $\pi_\mathrm{NV} < \epsilon_\mathrm{NV} $.} \label{fig:sprespi} \end{figure} Out of $533$ objects, 416 objects are successfully detected ($P_{\text{bc}} > 3\sigma_p$ and $\pi > \epsilon$) in both NVSS and S-PASS polarized flux densities. There are 90 sources that are not detected in polarization in S-PASS but are detected in NVSS, whereas 12 objects with no detection in NVSS polarization are detected in S-PASS. There are 15 objects that do not have polarization above our threshold in either survey. The depolarization, D, is defined to be the ratio between S-PASS and NVSS fractional polarizations: \begin{equation} D \equiv \frac{\pi_\mathrm{SP}}{\pi_\mathrm{NV}} \end{equation} We calculated the depolarization of all objects with $3\sigma_p$ polarization detection and $\pi > \epsilon$ in both S-PASS and NVSS. Upper/lower limits on $D$ are also calculated for sources as appropriate. \subsubsection{Polarization angle and rotation measure} Assuming that the contaminating sources have very little impact on the polarization angle of the target source, we used NVSS and S-PASS $Q$ and $U$ flux densities to derive the polarization angles $\chi_\mathrm{NV}$ and $\chi_\mathrm{SP}$; where \begin{equation} \chi=\frac{1}{2}\tan^{-1}\frac{U}{Q} \end{equation} These angles are used to estimate the amount of the rotation measure, $\textrm{RM}_\mathrm{NS}$, between the NVSS and S-PASS. The median uncertainty on the derived rotation measures is on the order of 1.6 rad m$^{-2}$. The polarization angle can be wrapped by a positive or negative integer coefficient, $n$, of $\pi$ radians from the true angle, the so-called $n\pi$ ambiguity. In this case, the true rotation measure is $\textrm{RM}_\mathrm{NS}=\textrm{RM}_0 \pm n\pi/\lambda^2$ rad m$^{-2}$. We used the TSS09 rotation measure catalog ( $\textrm{RM}_\mathrm{T}$) to fix $n$ by minimizing the absolute values of $\Delta \textrm{RM}$, where $\Delta\textrm{RM} \equiv \textrm{RM}_\mathrm{T}-\textrm{RM}_\mathrm{NS}-n\pi/\lambda^2$ for the 364 sources in common. These are not necessarily the correct $\textrm{RM}$s, since TSS09 has its own $n\pi$ ambiguity of 653 rad m$^{-2}$, while this ambiguity for $\textrm{RM}_\mathrm{NS}$ is about 108 rad m$^{-2}$. However, they provide the most conservative estimate of $\Delta \textrm{RM}$, the inferred non-linearity in the Faraday rotation as a function of $\lambda$. The parameter $n$ took values of $-1, 0, 1$ for all objects except one with $n=-2$. Note that, including the polarization contamination and recalculating the RMs based on the two NVSS sub-bands could introduce offsets as large as 42 rad m$^{-2}$. As a result, using the uncontaminated NVSS $\textrm{RM}_\mathrm{T}$ is appropriate. \subsubsection{Bandwidth depolarization}\label{bwdepol} When the $\textrm{RM}$ is high, the rotation of the polarization angle across a fixed bandwidth reduces the net degree of polarization, which is called bandwidth depolarization. To evaluate the importance of this effect for our sample, we used the 364 sources overlapping with TSS09. We predicted the NVSS and S-PASS bandwidth depolarizations for our objects based on the measured TSS09 $\textrm{RM}_\mathrm{T}$ and our $\textrm{RM}_\mathrm{NS}$, respectively. As shown in Figure \ref{fig:BWdep2} the ratio between the observed fractional polarization and the true degree of polarization $\pi_\mathrm{obs}/\pi_\mathrm{true}$ never gets smaller than 0.95 for S-PASS, and only 3\% of objects have NVSS $\pi_\mathrm{obs}/\pi_\mathrm{true}$ smaller than 0.9. The median $\pi_\mathrm{obs}/\pi_\mathrm{true}$ for both S-PASS and NVSS are 0.999 and 0.996 respectively, and therefore, bandwidth depolarization will not affect our analysis throughout this work. \begin{figure}[h] \centering \includegraphics[scale=0.51]{BWdep3.eps} \caption{The ratio of the observed and true fractional polarizations, $\pi_\mathrm{obs}/\pi_\mathrm{true}$, based on the NVSS and S-PASS bandwidth depolarizations is shown for 364 objects as a function of the cumulative percentile. Only 3\% of objects in our sample experience NVSS bandwidth depolarization which results in $\pi_\mathrm{obs}/\pi_\mathrm{true}$ smaller than 0.9.} \label{fig:BWdep2} \end{figure} \subsubsection{Spectral index} We used I$_{NV}$ and peak S-PASS (11' beam) intensities, and calculated the power law spectral index, $\alpha$, where $I \propto \nu^{\alpha}$. Figure \ref{fig:sphst} shows the distribution of spectral indices for our 533 objects. The median is $\bar\alpha\sim -0.83$. The contaminating flux contributing to the NVSS intensities can be a small source of uncertainty in the calculated spectral indices; we estimated its median to be $\sigma_{\alpha,Cont} \sim 0.01$ while total uncertainties on the derived spectral indices has median value of $\sigma_{\alpha,Tot}=0.05$. \begin{figure}[h] \centering \includegraphics[scale=0.51]{sphst2.eps} \caption{ Distribution of spectral indices,$\alpha$, calculated based on NVSS and S-PASS total intensities. The median spectral index is $\bar{\alpha} \approx -0.83$.} \label{fig:sphst} \end{figure} \subsubsection{Surface area of the object} We used the NVSS catalog de-convolved minor, $\theta_m$, and major $\theta_J$ axes of the target object, and calculated the effective area, $A$ as follows.: \begin{equation} A \equiv \frac{1}{4}\pi\theta_m\theta_J \end{equation} One must note that almost all sources remain unresolved in S-PASS due to the very large beam size. \subsubsection{Uncertainties} We used the measured local rms values as uncertainties of the NVSS $Q$, $U$ and S-PASS $I$, $Q$ and $U$ flux densities. The uncertainty of the NVSS total intensities are extracted from the NVSS catalog. Error propagation is used to approximate the uncertainty on all the other derived quantities such as polarized flux density and rotation measure. We note that \cite{2011ApJ...726....4S} showed that the rotation measure uncertainties reported in the TSS09 catalog might be underestimated. As a result, we multiplied all the $\textrm{RM}_\mathrm{T}$ uncertainties by 1.22 as described in \cite{2011ApJ...726....4S}. \subsection{Selection Bias}\label{bias} We do not select objects based on their polarization intensities or fractional polarizations. However, we apply a threshold cut on the contribution of polarized contaminants. There is a higher probability for objects with low polarized intensity, either intrinsic or due to depolarization, to suffer from contaminating neighbors, and to be dropped from our final sample. To investigate a possible missing population, we compared two different sub-samples a) sources in our catalog with $\frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.1 (533 sources, 416 detected in both NVSS and S-PASS) and b) objects rejected from our catalog with 0.1 $\le \frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.25 (75 sources, 40 detected in both NVSS and S-PASS). We compared the fractional polarization and the depolarization properties of these two sub-samples. If we were \emph{not} creating a selection bias, then they should have similar properties. Figure \ref{polcont} shows the results. Objects with larger polarization contamination have on average lower 2.3 GHz fractional polarization (median $\bar \pi_\mathrm{SP} = 1.5\%$) while less contaminated sources have $\bar \pi_\mathrm{SP}= 2.5\%$. Moreover, the fraction of sources with $\pi_\mathrm{SP} < 1\%$ is 2.5 times higher (50\%) among objects with 0.1 $\le \frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.25 than sources with $\frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.1. The Spearman rank test between $D$ and $\frac{P_{\text{cont}}}{P_{\text{target}}}$ with $r= 0.22$ and $p<0.00001$ rejects the null hypothesis of no correlation. Thus, we are likely to be missing a population of highly depolarized sources. Figure \ref{polcont} suggests that around 30\% of sources with polarized contamination 0.1 $\le \frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.25 have depolarizations $\log(D)>0.47$. Assuming this fraction is also valid for sources with contaminations larger than 25\% we estimate that we have missed $\sim 50$ depolarized objects in our final sample due to the polarized contamination threshold cut. Therefore, our final sample of 533 sources is missing a population ($\sim 50$ objects) of heavily depolarized sources due to our contamination threshold cut. However, we can not correct for such an effect since the amount of contamination in our 2.3 GHz polarization intensities can not be measured. As a result, one should treat the number of depolarized sources in our sample as a strong lower limit and consider this in interpreting all the other related conclusions. \begin{figure}[h] \centering \includegraphics[scale=0.51]{polcont2.eps} \includegraphics[scale=0.51]{fphistnewpol1.eps} \caption{Distribution of $\log(D)$ (top) and 2.3 GHz fractional polarization $\pi_\mathrm{SP}$ (bottom) of objects with $\frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.1 and 0.1 $\le \frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.25 are shown with solid and dashed lines respectively. Black, red and blue colors represent objects with detection, upper limits in S-PASS and upper limits in NVSS polarizations. The area under the $\log(D)$ histogram of objects with $\frac{P_{\text{cont}}}{P_{\text{target}}} <$ 0.1 and detected polarization flux densities is colored in gray for clarity. } \label{polcont} \end{figure} In addition, it is possible that our total intensity and polarization contamination thresholds have resulted in a bias toward less dense regions of the sky. We measured the surface number density of the contaminating neighbors in our sample and the parent NVSS--S-PASS overlap sample with $I_\textrm{NV} > 10$ mJy. We used the same aperture with a radius of 16 arcmin and found that the contaminant surface number density in our final sample ($4\times 10^{-3}$ arcmin$^{-2}$) is on average 20\% less than our parent sample ($5\times 10^{-3}$ arcmin$^{-2}$). It is unlikely that the results of this work are affected by such a bias. \subsection{Statistical tests}\label{stat} Throughout this work we adopted two nonparametric statistical tests. We calculated the Spearman rank correlation coefficient ($r_s$) to measure the strength of any possible correlation. The two-sample Kolmogorov-Smirnov (KS) test is also used to check the null hypothesis that two sub-samples, divided by a parameter of interest, are drawn from the same parent distribution. The significance of each test is estimated by performing bootstrap sampling simulations and constructing $10^5$ random samples from the initial distribution. We have assigned two-tail p-values based on the results of our simulations. Table \ref{table2} summarizes the result of all the statistical tests performed in this work. In the case of a single hypothesis test we would reject the null hypothesis if the p-value $\le0.01$. However, we have performed a total 90 tests, counting both KS and Spearman. To avoid the multiple hypothesis testing problem, we adopted the Bonferroni correction as discussed in \cite{2016arXiv160203765A} and chose a conservative significance level threshold of p-value $\le 10^{-4}$. We therefore rejected the null hypothesis of the KS or the Spearman rank tests only if the corresponding p-value is less than or equal to $10^{-4}$. In addition, to test the robustness of correlations with p-value less than $10^{-4}$ and to identify any possible influence of the total intensity and polarization contaminations on the results we repeated the relevant statistical tests on smaller (by a factor of $\sim$0.4) but clean samples of objects with less than 1\% contamination. Although the strength of some correlations became stronger or did not change, their p-values increased up to $2\times 10^{-3}$ due to the much smaller sample size. We therefore, adopted the robustness probability p$_\textrm{robust}$ of $2\times 10^{-3}$ as a second threshold and treated the correlations with original p-value$\le10^{-4}$ and $2\times 10^{-3} < $p$_\textrm{robust}< 0.05$ as suggestive relations only, and did not draw any conclusion based on them. These are marked in Table \ref{table2} for completeness and potentially future work. The correlations with original p-value$ <10^{-4}$ but p$_\textrm{robust}> 0.05$ are rejected. \section{Results}\label{result} We have derived a polarization catalog of 533 extragalactic radio sources, which can be downloaded for public use through the VizieR catalog access tool. The description of the entries in the online catalog is listed in table \ref{table3}. \tabcolsep=0.1cm \tabletypesize{\scriptsize} \begin{deluxetable}{ll}[h] \tablecolumns{2} \tablewidth{85mm} \tablecaption{Description of the entries in the online catalog. \label{table3}} \tablehead{ \colhead{{Column index}} & \colhead{Description} } 1 & NVSS name tag \\ 2 & NVSS RA in decimal degrees (J2000) \\ 3 & NVSS Dec in decimal degrees (J2000) \\ 4 & Galactic longitude \\ 5 & Galactic latitude \\ 6 & NVSS total intensity, $I_\mathrm{NV}$ \\ 7 & Uncertainty on the $I_\mathrm{NV}$ \\ 8 & NVSS polarized intensity \\ 9 & Uncertainty on the NVSS polarized intensity \\ 10 & NVSS fractional polarization, $\pi_\mathrm{NV}$\\ 11 & Uncertainty on the $\pi_\mathrm{NV}$ \\ 12 & Upper limit flag on the $\pi_\mathrm{NV}$ \\ 13 & NVSS polarization angle \\ 14 & Uncertainty on the NVSS polarization angle \\ 15 & NVSS catalog fitted deconvolved major axis \\ 16 & Upper limit flag on the deconvolved major axis \\ 17 & NVSS catalog fitted deconvolved minor axis \\ 18 & Upper limit flag on the deconvolved minor axis \\ 19 & S-PASS peak intensity, $I_\mathrm{SP}$ \\ 20 & Uncertainty on the $I_\mathrm{SP}$ \\ 21 & S-PASS polarized intensity \\ 22 & Uncertainty on the S-PASS polarized intensity \\ 23 & S-PASS fractional polarization, $\pi_\mathrm{SP}$ \\ 24 & Uncertainty on the $\pi_\mathrm{SP}$ \\ 25 & Upper limit flag on the $\pi_\mathrm{SP}$ \\ 26 & S-PASS polarization angle \\ 27 & Uncertainty on the S-PASS polarization angle \\ 28 & Spectral index derived from NVSS and S-PASS \\ 29 & Uncertainty on the spectral index \\ 30 & Depolarization, $D$ \\ 31 & Uncertainty on the $D$ \\ 32 & Taylor et al. 2009 rotation measure, $\textrm{RM}_\mathrm{T}$ \\ 33 & Uncertainty on the $\textrm{RM}_\mathrm{T}$ multiplied by 1.22 \\ 34 & The NVSS \& S-PASS rotation measure, $\textrm{RM}_\mathrm{NS}$ \\ 35 & Uncertainty on the $\textrm{RM}_\mathrm{NS}$ \\ 36 & Rotation measure difference, $\Delta RM$ \\ 37 & Uncertainty on the $\Delta RM$ \\ 38 & Median $\textrm{RM}_\mathrm{T}$ \\ 39 & Number of sources contributed to the median $\textrm{RM}_\mathrm{T}$ \\ 40 & Redshift from Hammond et al. 2012 \\ 41 & WISE catalog W1 (3.4 micron) magnitude \\ 42 & Uncertainty on the W1 \\ 43 & W1 detection signal to noise ratio \\ 44 & WISE catalog W2 (4.6 micron) magnitude \\ 45 & Uncertainty on the W2 \\ 46 & W2 detection signal to noise ratio \\ 47 & WISE catalog W3 (12 micron) magnitude \\ 48 & Uncertainty on the W3 \\ 49 & W3 detection signal to noise ratio \\ \end{deluxetable} \subsection{Rotation measures}\label{rm0} The distribution of $\textrm{RM}_\mathrm{NS}$ calculated based on NVSS and S-PASS (black) and the Taylor et al. rotation measures, $\textrm{RM}_\mathrm{T}$, (red) for the same objects are shown in Figure \ref{fig:rmhist}. Both distributions are very similar in shape. Their medians are $3.6\pm 2.0$ and $0.5 \pm 1.9 $ rad m$^{-2}$ , respectively, while their standard deviations are 38.4 and 36.4 rad m$^{-2}$. Some of the scatter in the $\textrm{RM}$ distributions could be due to the uncertainty of the measurements. \citep{2011ApJ...726....4S}. However, the median error on the $\textrm{RM}_\mathrm{T}$ for the small bright sample of 364 objects in this work is only $\sigma_\mathrm{T}=3.5$ rad m$^{-2}$. The median measurement uncertainty estimated for $\textrm{RM}_\mathrm{NS}$ is even smaller, $\sigma_\mathrm{NS}=1.6$ rad m$^{-2}$. Subtracting the median errors from the observed standard deviation of the $\textrm{RM}$ distributions in quadrature result in residual standard deviations of $36.2$ rad m$^{-2}$ and $38.36$ rad m$^{-2}$ for $\textrm{RM}_\mathrm{T}$ and $\textrm{RM}_\mathrm{NS}$ respectively, and largely represent the spread in Galactic foregrounds. \begin{figure}[h] \centering \includegraphics[scale=0.51]{rmhist.eps} \includegraphics[scale=0.51]{RM2new.eps} \caption{Top: The distributions (top) and the scatter diagram (bottom) of the NVSS, S-PASS rotation measures, RM$_\mathrm{NS}$ versus TSS09 RM$_\mathrm{T}$ for the 364 common objects. The three red solid lines in the bottom show one-to-one relations for the three cases of $n= [-1,0,1]$.} \label{fig:rmhist} \end{figure} \subsection{Distribution of fractional polarization and depolarization}\label{D_distribution} The median NVSS (S-PASS) fractional polarization of all 533 objects is $\bar{\pi}=0.017$ $(0.020)$ including the upper limits. There are $505$ ($428$) objects with detected NVSS (S-PASS) polarization ($P > 3\sigma_p$ and $\pi > \epsilon $). However, 416 of these objects are detected in both NVSS and S-PASS. The distributions of NVSS and S-PASS fractional polarization of these 416 objects are shown in Figure \ref{fig:fphist}. The median (and standard deviation) of NVSS and S-PASS fractional polarization of these common objects are 0.022 (.022) and 0.025 (0.023), respectively. Although the median values of $\pi_\mathrm{SP}$ and $\pi_\mathrm{NV}$ are very close, the median value of their ratio (the median depolarization) is not necessarily equal to one. The TSS09 catalog was limited to sources with sufficient signal:noise in polarization, and is thus biased towards much higher fractional polarizations (median $\bar{\pi}_T \sim 0.06$) than our catalog, which is $\sim$3.5 times lower, including both measurements and upper limits. \begin{figure}[h] \centering \includegraphics[scale=0.51]{fphistnew_2.eps} \caption{Normalized histograms of fractional polarizations, $\pi$, for 416 objects with detected polarization in both NVSS and S-PASS and the upper limits. The black and red solid lines represent the NVSS and S-PASS distributions of objects with detected polarizations while the dashed blue and red lines sketches the distribution of upper limits of NVSS and S-PASS polarizations. For comparison we also show the NVSS fractional polarization distribution of the TSS09 catalog 37543 sources with dotted-dashed magenta line. \label{fig:fphist}} \end{figure} Figure \ref{fig:depol} shows the normalized distribution of $\log(D)$ for steep and flat spectrum sources separately. Objects with both S-PASS and NVSS detected polarizations are shown in solid black, and have median depolarizations of $\bar{D}=1.4$ and $\bar{D}=0.9$ for 315 steep and 101 flat sources respectively. The depolarization distribution of steep spectrum sources is skewed toward large values of $D$. Almost 28\% of steep spectrum (24\% of all) objects have $D \ge 2$, and only 2\% have $D \le 0.5$. On the other hand, flat spectrum sources include both depolarized and re-polarized objects. There are 17\% and 13\% of flat spectrum sources with $D \ge 2$ and $D \le 0.5$ respectively. The results of the statistical tests presented in Table \ref{table2} confirm that steep and flat spectrum sources do not have the same depolarization distributions. The red dashed histogram in Figure \ref{fig:depol} shows the normalized distribution of 58 steep spectrum and 31 flat spectrum objects with upper limits on the depolarization. These sources have S-PASS polarizations less than $3\sigma$ or $\pi_\mathrm{SP} < \epsilon_\mathrm{SP}$ but are detected in NVSS polarization. The 12 steep spectrum objects with NVSS $P < 3\sigma_p$ or $\pi_\mathrm{NV} < \epsilon_\mathrm{NV}$ and detected S-PASS polarization are treated as lower limits on the depolarization. The dotted dashed blue line show the distribution of the lower limits in Figure \ref{fig:depol}. In total, 16 objects are detected in neither NVSS nor in S-PASS polarizations and we do not show them in Figure \ref{fig:depol}. \begin{figure}[h] \centering \includegraphics[scale=0.51]{Depol_st.eps} \includegraphics[scale=0.51]{Depol_ft.eps} \caption{Distributions of $\log(D)$ normalized to the total number of objects for steep (top) and flat (bottom) spectrum sources. Black solid histogram represents objects with detected polarization in both NVSS and S-PASS. The red histogram with dashed line is the distribution of the upper limits in depolarization. The lower limits are shown with dotted-dashed blue line. The two red and blue arrows show the direction of movement for the upper and lower limits. } \label{fig:depol} \end{figure} \cite{2014ApJS..212...15F} used their multi wavelength polarization spectra and derived an equivalent power law polarization spectral index $\beta$, where $\pi \propto \lambda^{\beta}$. As long as the power law model is assumed our depolarization parameter $D$ and $\beta$ are related such that $\log(D)=\log(\frac{\lambda_\mathrm{SP}}{\lambda_\mathrm{NV}}) \beta$ where the $\lambda_\mathrm{SP}$ and $\lambda_\mathrm{NV}$ are the average wavelengths of the S-PASS and NVSS surveys respectively. \cite{2014ApJS..212...15F} found weak evidence of a bimodal distribution for $\beta$ of steep spectrum objects. We do not see any sign of bimodal depolarization within objects with $\alpha < -0.5$, as shown in Figure \ref{beta}. The $\beta$ distribution of steep spectrum objects is single-peaked but asymmetric with a longer tail toward depolarized objects. As will be discussed later, the majority of steep spectrum sources in our sample can be classified as IR AGNs according to their infrared colors. A more complete sample which also includes radio galaxies with infrared colors of normal ellipticals can confirm if the weak bimodal depolarization observed by \cite{2014ApJS..212...15F} is real. \begin{figure}[h] \centering \includegraphics[scale=0.51]{Depol3_2.eps} \caption{Distribution of the polarization spectral index $\beta$ as introduced in \cite{2014ApJS..212...15F} assuming a power law depolarization model. The solid blue and dashed red lines represent the steep and flat spectrum sources. \label{beta}} \end{figure} We also looked at the combined sample of steep and flat spectrum sources and classified them into three depolarization categories. The choice of the depolarization boundaries is somewhat arbitrary. However, we designed the three depolarization categories to isolate the peak observed in Figure \ref{fig:depfp}, as is discussed below. Sources with $ 0.6 < D < 1.7 $ have median spectral index of $\bar{\alpha} \sim -0.82$ while sources with $D \ge 1.7$ shows a slightly steeper median spectrum with $\bar{\alpha} \sim -0.9$. The spectral slope is mostly flat for re-polarized objects with $D \le 0.6$, with a median $\bar{\alpha} \sim -0.1$. However, there are 14 re-polarized objects with steep spectral indices, $\alpha < -0.5$. This is consistent with \cite{2014ApJS..212...15F} who also found a small population of steep spectrum re-polarized sources. Figure \ref{sphstrep} shows the distribution of the spectral indices of re-polarized objects. We also included 24 objects with detection in $\pi_\mathrm{NV}$ but only upper limits on $\pi_\mathrm{SP}$. Figure \ref{sphstrep} suggests there are two separate populations of re-polarized sources with flat and steep spectra. Including the mentioned upper limits on $D$, 61\% of re-polarized sources have $\alpha \ge -0.5$ (i.e., flat). \begin{figure}[h] \centering \includegraphics[scale=0.51]{sphstrep.eps} \caption{ Spectral index distribution of the re-polarized objects, $D< 0.6$, including 24 sources with detection in $\pi_\mathrm{NV}$ but only upper limits on $\pi_\mathrm{SP}$. While it seems there are two separate populations of re-polarized sources with flat and steep spectrums, the majority of them, 61\%, have $\alpha \ge -0.5$. \label{sphstrep}} \end{figure} To understand the relation between fractional polarization and depolarization, we plotted $\pi_\mathrm{SP}$ versus $\log(D)$, and calculated the running medians in bins of 30 objects (Figure \ref{fig:depfp}). There is an apparent peak for S-PASS fractional polarization at $log(D )\sim 0$, while both depolarized and re-polarized sources show weaker $\pi_\mathrm{SP}$ than sources with fractional polarization above 6\%. Both KS and Spearman rank coefficient tests on the $|\log(D)|$ and $\pi_\mathrm{SP}$ confirm this anti-correlation. We also used two subsamples with $\log(D) >0$ and $\log(D) <0$ and performed the two KS and Spearman rank tests on each subsample separately. The results confirmed that fractional polarizations are higher in the vicinity of $\log(D)\sim 0$ in each subsample. However, the correlation between $|\log(D)|$ and $\pi_\mathrm{SP}$ of the subsample with $\log(D) >0$ became uncertain when only including the contamination clean sample of the robustness test. Table \ref{table2} summarizes the results of these statistical tests. \begin{figure}[h] \centering \includegraphics[scale=0.51]{Depfp_shade.eps} \caption{S-PASS fractional polarization versus $\log(D)$. The red solid line represents the running median of $\pi_\mathrm{SP}$ calculated in bins of $N=30$ objects in $\log(D)$ space and the dark-pink shaded region is the estimated uncertainty on the running medians calculated as $|M-[p16, p84]|/\sqrt(N)$ where $M$ is the median value and $[p16, p84]$ are the 16 and 84 percentiles. The error bars on the left and right upper corners are the medians of the intrinsic uncertainties in $\pi_\mathrm{SP}$ for the two half of data in $\log(D)$. } \label{fig:depfp} \end{figure} Figure \ref{fig:fphist23} shows the S-PASS (top) and NVSS (bottom) fractional polarization distributions for three sub-samples with $|\log(D)| \le 0.23$, $\log(D) > 0.23$ and $\log(D)<-0.23$. Objects with $\log(D) \sim 0$ have almost the same distribution in both S-PASS and NVSS (by construction) with median fractional polarizations of $\bar{\pi}_\mathrm{SP}=0.030$ and $\bar{\pi}_\mathrm{NV}=0.028$ while depolarized sources have smaller medians, $\bar{\pi}_\mathrm{SP}=0.024$ and $\bar{\pi}_\mathrm{NV}=0.009$ with an offset between NVSS and S-PASS as expected. Objects with re-polarization show more complicated behavior. They have a median $\bar{\pi}_\mathrm{SP}=0.015$ and $\bar{\pi}_\mathrm{NV}=0.035$. By definition the median degree of polarization of a sample of re-polarized sources is expected to be higher at 1.4 GHz than 2.3 GHz. It is possible that the true median $\bar \pi_\textrm{SP}$ and $\bar \pi_\textrm{NV}$ are lower than the above values because we would have systematically excluded re-polarized objects with $\pi_\mathrm{SP}$ less than the detection limit. This results in over estimating the median fractional polarization of re-polarized sources in both NVSS and S-PASS. \begin{figure}[h] \centering \includegraphics[scale=0.51]{fphist2.eps} \includegraphics[scale=0.51]{fphist3.eps} \caption{Normalized S-PASS (upper) and NVSS (lower) fractional polarization distribution for objects with $|\log(D) < 0.23|$ (solid black), $\log(D) > 0.23$ (dashed red) and $\log(D)<-0.23$ (dotted-dashed blue). \label{fig:fphist23}} \end{figure} \subsection{Total intensity and fractional polarization}\label{ip} Our sample includes total intensities from 0.42 to 10 Jy, which gives us the opportunity to study possible correlations between the fractional polarization and total intensity. As listed in Table \ref{table2} both KS and Spearman tests suggest there is a weak anti-correlation between $\pi_\mathrm{SP}$ and $I_\mathrm{SP}$ of the whole sample of sources at 2.3 GHz. More investigation revealed that is true for steep spectrum ($\alpha < -0.5$) sources alone, while it disappears for flat spectrum ($\alpha \ge -0.5$) objects. The anti-correlation among steep spectrum sources became weaker and more uncertain when only including the contamination clean sample of the robustness test, and thus should be treated as a suggestive trend only. Figure \ref{fig:Ifp} shows the S-PASS $\pi_\mathrm{SP}$ of only steep spectrum sources versus their logarithm of total intensity. The calculated running medians (including the upper limits on $\pi_\mathrm{SP}$ to avoid any selection bias due to our total intensity cut) are shown as well. Objects with $\alpha < -0.5$ and $\log(I_\mathrm{SP}) < 2.9$ have median of $\bar{\pi}_\mathrm{SP} \sim 0.03$ while sources with larger total intensity are less polarized with medians of $\bar{\pi}_\mathrm{SP} \sim 0.02$. \begin{figure}[h] \centering \includegraphics[scale=0.51]{Ifp2_2_shade.eps} \caption{S-PASS fractional polarization of only steep spectrum($\alpha < -0.5$) versus their total intensity. The open circles represent the upper limits on the degree of polarization. The black solid line is the running medians of $\pi_\mathrm{SP}$ including the upper limits and the dark-pink shaded region is the estimated uncertainty on the running medians. The red error bars in upper right and left corners show the median intrinsic uncertainties of $\pi_\mathrm{SP}$ for the two half of the data in $\log(I)$ space. \label{fig:Ifp}} \end{figure} To shed light on a possible physical origin of the observed anti-correlation we calculated the luminosities, based on the 261 objects in our sample which have redshifts in the \cite{2012arXiv1209.1438H} catalog. 222 of these sources are detected in both NVSS and S-PASS polarization. Using our spectral indices, we calculated the K-corrected 2.3 GHz luminosities. The 141 steep spectrum objects have median luminosity of $L_{steep}=1.7\times 10^{27}$ WHz$^{-1}$. Although there is a nominal difference between $\bar{\pi}_\mathrm{SP}$ for higher and lower luminosities (2.6\% and 2.2\%, respectively), these do not appear statistically significant. There is also no statistically significant difference in $|\log(D)|$ for the high and low luminosity steep spectrum sources. The 81 flat spectrum sources are at higher redshifts, on average, and have a median luminosity of $\bar L_{flat}=3.0 \times 10^{27}$ W Hz$^{-1}$. \subsection{Correlation between $\textrm{RRM}$, $\Delta \textrm{RM}$, $\pi$ and $D$}\label{deltarm} There are two measures to characterize the Faraday effects that are either local to the source or in the intervening IGM medium, the residual rotation measure $\textrm{RRM}$, which takes out the Galactic foreground contribution to the observed $\textrm{RM}$, and $\Delta \textrm{RM} \equiv \textrm{RM}_\mathrm{T}-\textrm{RM}_\mathrm{NS}$, which sheds light on the frequency dependency of the $\textrm{RM}$. The absolute value of $|\Delta RM|$ is an indicator of the Faraday complexity of the source and its environment. As explained in the following, we found that $\Delta \textrm{RM}$ is anti-correlated with $\pi$ and correlated with $|\log(D)|$. Faraday complex sources, i.e, those with multiple $\textrm{RM}$ components should be both depolarized and have polarization angles which may not vary linearly with $\lambda^2$. We therefore examined the possible correlation between depolarization and $|\Delta \textrm{RM}|$. Figure \ref{fig:drm2new} shows $|\Delta \textrm{RM}|$ versus $|\log(D)|$ for all objects with detected polarization in both NVSS and S-PASS. The running medians of the $|\Delta \textrm{RM}|$ calculated in bins of $|\log(D)|$ show an evolution. To quantify this, we calculated the Spearman rank, which yielded a correlation coefficient of $r_s=0.23$ and p-value of $p = 0.00003$ establishing that depolarization and non-$\lambda^2$ polarization angle behavior are related. \begin{figure}[h] \centering \includegraphics[scale=0.51]{DRM2new_shade.eps} \caption{Absolute difference between rotation measures calculated in this work and in TSS09, $|\Delta \textrm{RM}|$ versus $|\log(D)|$. Black and green crosses represent depolarized and re-polarized objects respectively. The solid red line is the running median of $|\Delta \textrm{RM}|$ calculated for bins of 23 objects in $|\log(D)|$ space which include both depolarized and re-polarized sources and the dark-pink shaded region is the estimated uncertainty on the running medians. The error bars on the left and right upper corners are the medians of intrinsic uncertainties in $|\Delta \textrm{RM}|$ for the two halves of the data. \label{fig:drm2new}} \vskip 1mm \end{figure} A large $\textrm{RM}$ beyond the Galactic foreground $\textrm{RM}$ screen could also indicate the presence of Faraday complexity and depolarization. To estimate this, we removed the Galactic contribution by subtracting the median $\bar{\textrm{RM}}$ within 3 degrees of each target (excluding the target itself), using the TSS09 catalog. This yields the residual rotation measure, $\textrm{RRM}_\mathrm{T} \equiv \textrm{RM} -\bar{\textrm{RM} }$. Subtracting the median $\bar{\textrm{RM}}$ is not the best method to estimate the extragalactic component of the $\textrm{RM}$ as discussed in \cite{2015A&A...575A.118O}. However, for objects above the Galactic latitude of $|b| > 20$ degrees, which is true for most of our sample, the difference between the \cite{2015A&A...575A.118O} recipe and our method is small. As shown in Figure \ref{rrmdep}, we find the Spearman rank coefficient of $r_s=0.21$ and p-value of $7\times10^{-5}$ which suggests a correlation between $|\textrm{RRM}_\mathrm{T}|$ and $|$log(D)$|$. However, our robustness test on the clean sample failed to confirm such a trend. Thus, only $|\Delta \textrm{RM}|$ shows a clear sign of a correlation with depolarization. \begin{figure}[h] \centering \includegraphics[scale=0.51]{rrmtdepol_shade.eps} \caption{ The absolute residual rotation measures, $|\textrm{RRM}_\mathrm{T}|$ versus the $|\log(D)|$. The red solid line which represent the running medians of $|\log(D)|$, shows an increase with raising $|\textrm{RRM}_\mathrm{T}|$. The dark-pink shaded region is the estimated uncertainty on the running medians.\label{rrmdep}} \end{figure} We also found, the 1.4 GHz and 2.3 GHz fractional polarizations show moderate anti-correlations with $|\Delta \textrm{RM} |$, as shown in Figure \ref{drmp} and listed in Table \ref{table2}. Thus, depolarization does reduce the fractional polarizations at these frequencies, although the dominant role of field disorder is discussed in Section \ref{obs}. Moreover, the Spearman rank test with $r_s=-0.25$ and p-value of $<10^{-5}$ suggest an anti-correlation between $|\textrm{RRM}_\mathrm{T}|$ and $\pi_\textrm{NV}$. However, our robustness test failed to confirm this significance. \begin{figure}[h] \centering \includegraphics[scale=0.51]{drmp_shade.eps} \caption{ S-PASS fractional polarization versus the $|\Delta \textrm{RM} |$ which is a representation of the Faraday structure. \label{drmp}} \end{figure} \subsection{Polarization, depolarization and the object angular extent}\label{extent} To study how the morphology of a system affects the depolarization, we used total intensity deconvolved areas ($A$) derived from the NVSS catalog \cite{1998AJ....115.1693C}. Flat spectrum objects in our sample are unresolved in the NVSS synthesized beam while steep spectrum objects include both resolved and unresolved sources. For the steep spectrum sources, Figure \ref{fig:dsize} shows the distributions of the absolute $|\log(D)|$ for two sub-samples - unresolved and resolved sources with the dividing line at $\log(A) = 2.5$ arcsec$^2$. On average, resolved sources have smaller $|\log(D)|$ with median of $0.12$ compared to $0.20$ for unresolved sources. The scatter of the two samples is almost the same with standard deviation of $0.21$. Beam depolarization should only play a small role, because most resolved sources are only slightly resolved. We also looked at the dependence of fractional polarization on size. Figure \ref{fig:sizefp} shows the distributions of the S-PASS fractional polarization for the unresolved and extended samples of steep spectrum objects. On average, resolved and extended steep spectrum objects have 2.3 GHz fractional polarizations, $\bar \pi_\mathrm{SP} \sim 4\%$, two times larger than their unresolved counterparts. Both KS and Spearman tests confirm a strong strong positive correlation between $A$ and $\pi_\mathrm{SP}$ of steep spectrum objects. \begin{figure}[h] \centering \includegraphics[scale=0.51]{sizedepolhist.eps} \caption{The $|\log(D)|$ distributions of the unresolved (black solid) and extended (dashed red) steep spectrum objects in the NVSS survey. The de-convolved surface area thresholds $\log(A) \le 2.5$ arcsec$^2$ and $\log(A) > 2.5$ arcsec$^2$ are used to separate unresolved and extended sources, and the two vertical blue solid and dashed lines represent the medians of $|\log(D)|$ for the two samples respectively. \label{fig:dsize}} \end{figure} \begin{figure}[h] \centering \includegraphics[scale=0.51]{sizefphist.eps} \caption{The $\pi_\mathrm{SP}$ distributions of the unresolved (black solid) and extended (dashed red) steep spectrum objects in the NVSS survey. The de-convolved surface area thresholds $\log(A) \le 2.5$ arcsec$^2$ and $\log(A) > 2.5$ arcsec$^2$ are used to separate unresolved and extended sources, and the two vertical blue solid and dashed lines represent the medians of $\pi_\mathrm{SP}$ for the two samples respectively. \label{fig:sizefp}} \end{figure} \subsection{Spatial distribution of depolarization in the sky}\label{galD} We carried out a brief investigation to see if the depolarization properties in our sample were related to their position in Galactic coordinates. Figure \ref{fig:galactic} shows the distribution of 533 objects in the sky, color coded with respect to their depolarization. Visual inspection does not reveal any obvious over-density of depolarized or re-polarized objects. \begin{figure} \centering \includegraphics[scale=0.51]{galactic.eps} \caption{Distribution of the 533 objects in the sky, color coded with the depolarization. $l$ and $b$ are the Galactic longitude and latitude coordinates in degrees. Black dots are objects that are not detected in either NVSS or S-PASS polarization. Green, red and blue triangles are objects with depolarization $0.5< D <2$, $D > 2$ and $D < 0.5$ respectively. \label{fig:galactic}} \end{figure} We also calculated the auto correlation between depolarization and angular separation, and the two point angular correlation function for the most depolarized and least depolarized sources. None of these showed any evidence for clustering of depolarization in space. Similarly, the two point angular correlation functions for the highest and lowest fractional polarizations at 2.3~GHz revealed no clustering. Other work has identified some positional dependence to polarizations in the NVSS catalog. \cite{2007ApJ...663L..21S} discovered regions with angular scales of $\sim 10$ degrees in which the density of the polarized sources drops by a factor of 2-4. They named these regions the ``polarization shadows,'' and found that some of them are associated with the Galactic HII regions while the rest are related to the depolarized areas in the diffuse Galactic radio emission. All polarization shadows in \cite{2007ApJ...663L..21S} are located within the Galactic plane at $|b| < 20 $ degrees except one which is at Galactic ($l=5$, $b=+24$). Almost all of the objects in our sample have Galactic latitudes of $|b| > 20 $ degrees, and none are located around ($l=5$, $b=+24$), so the Galactic polarization shadows likely do not affect the current work. However, it is interesting to search for high latitude Galactic diffuse emissions in smaller scales and their probable signature on the depolarization of the extragalactic sources in future surveys and larger samples with higher number density. \tabcolsep=0.04cm \tabletypesize{\scriptsize} \begin{deluxetable*}{ccccc|cc}[h] \centering \tablecolumns{7} \tablewidth{180mm} \tablecaption{Results of non-parametric statistical tests with simulated p-values. \label{table2}} \tablehead{ \colhead{{Parameters}} & \colhead{Constraint}&\colhead{KS distribution} & \colhead{KS samples} & \colhead{KS p-value} & \colhead{Spearman rank} & \colhead{p-value} \\ \colhead{} & & & \colhead{ } & \colhead{ simulated} & \colhead{correlation coefficient} &\colhead{simulated} } \startdata *$\pi_\mathrm{SP}$ \& $\alpha$ & - & $\pi_\mathrm{SP}$ & $\alpha \lessgtr -0.5$ &$<0.00001 $ & $-0.24$& $<0.00001$ \\[5pt] *$\pi_\mathrm{SP}$ \& $Area$ & $\alpha <-0.5$ & $\pi_\mathrm{SP}$ & $A \lessgtr \bar A$ & $<0.00001$ & $0.36$ & $<0.00001 $ \\[5pt] *$\pi_\mathrm{SP}$ \& $|\log(D)|$ & - & $|\log(D)|$ & $\pi_\mathrm{SP} \lessgtr \bar{\pi}_\mathrm{SP}$ & $<0.00001 $ & $-0.28$ & $<0.00001 $ \\[5pt] ?$|\log(D)|$ \& $\pi_\mathrm{SP}$ & $\alpha<-0.5$ & $|\log(D)|$ & $\pi_\mathrm{SP} \lessgtr \bar{\pi}_\mathrm{SP}$ & $0.00050$& $-0.26$ & $ <0.00001$ \\[5pt] $|\log(D)|$ \& $\pi_\mathrm{SP}$ & $\alpha \ge -0.5$ & $|\log(D)|$ & $\pi_\mathrm{SP} \lessgtr \bar{\pi}_\mathrm{SP}$& $0.022$&$-0.37$ & $0.00019 $ \\[5pt] ?$D$ \& $\pi_\mathrm{SP}$ &$D>1$ & $D$ & $\pi_\mathrm{SP} \lessgtr \bar{\pi}_\mathrm{SP}$ & $ 0.00004$ &$-0.25$ & $0.0095 $ \\[5pt] *$D$ \& $\pi_\mathrm{SP}$ & $D<1$ & $D$ & $\pi_\mathrm{SP} \lessgtr \bar{\pi}_\mathrm{SP}$ & $ 0.00005$ & $0.50$ & $< 0.00001 $ \\[5pt] ?$I_\mathrm{SP}$ \& $\pi_\mathrm{SP}$ & - & $\pi_\mathrm{SP}$ & $I_\mathrm{SP} \lessgtr \bar{I}_\mathrm{SP} $ & $<0.00001 $ & $-0.25$ & $<0.00001 $ \\[5pt] ?$I_\mathrm{SP} $ \& $\pi_\mathrm{SP}$ & $\alpha<-0.5$ & $\pi_\mathrm{SP}$ & $I_\mathrm{SP} \lessgtr \bar{I}_\mathrm{SP} $& $<0.00001 $ & $-0.25$ & $<0.00001 $ \\[5pt] {$I_\mathrm{NV}$ \& $\pi_\mathrm{NV}$} & - &$\pi_\mathrm{NV}$ & $I_\mathrm{NV} \lessgtr \bar{I}_\mathrm{NV} $& $0.050 $ &-0.13 & $ 0.0021$ \\[5pt] {$I_\mathrm{NV} $ \& $\pi_\mathrm{NV}$} &$\alpha<-0.5$ & $\pi_\mathrm{NV}$ & $I_\mathrm{NV} \lessgtr \bar{I}_\mathrm{NV} $& $0.094 $ &-0.16 & $0.0013 $ \\[5pt] {$I_\mathrm{NV} $ \& $\pi_\mathrm{NV}$} & $\alpha \ge -0.5$& $\pi_\mathrm{NV}$ & $I_\mathrm{NV} \lessgtr \bar{I}_\mathrm{NV} $& $ 0.96$ &-0.04 & $0.67 $ \\[5pt] {$I_\mathrm{NV} $ \& $|\log(D)|$} &$\alpha<-0.5$ & $|\log(D)|$& $I_\mathrm{NV} \lessgtr \bar{I}_\mathrm{NV} $& $ 0.33$ &0.08 & $0.18 $ \\[5pt] {$I_\mathrm{SP} $ \& $\pi_\mathrm{SP}$} &$\alpha \ge -0.5$ & $\pi_\mathrm{SP}$ & $I_\mathrm{SP} \lessgtr \bar{I}_\mathrm{SP} $& $0.26 $ &-0.06 & $ 0.47$ \\[5pt] {$L_\mathrm{SP}$ \& $|\log(D)|$} & $\alpha<-0.5$ &$L_\mathrm{SP}$ & $|\log(D)| \lessgtr 0.13$ &$0.010 $ & 0.12& $ 0.16$ \\[5pt] {$L_\mathrm{SP}$ \& $\pi_\mathrm{SP}$} & $\alpha<-0.5$ & $\pi_\mathrm{SP}$ &$L_\mathrm{SP} \lessgtr \bar{L}_\mathrm{SP}$ & $0.21 $ & -0.13& $ 0.11$ \\[5pt] {$L_\mathrm{SP}$ \& $\pi_\mathrm{SP}$} & $\alpha \ge -0.5$ & $\pi_\mathrm{SP}$ &$L_\mathrm{SP} \lessgtr \bar{L}_\mathrm{SP}$ & $ 0.32 $ & 0.07& $ 0.53$ \\[5pt] *$|\Delta \textrm{RM} |$ \& $|\log(D)|$ & - & $|\log(D)|$ & $|\Delta \textrm{RM} | \lessgtr |\Delta \overline{\textrm{R}}\textrm{M}| $ & $ 0.0010$ & $0.23$ & $0.00003 $ \\[5pt] *$|\Delta \textrm{RM} | $ \& $\pi_\mathrm{SP}$ & - & $|\Delta \textrm{RM} | $ & $\pi_\mathrm{SP} \lessgtr \bar{\pi}_\mathrm{SP}$ & $<0.00001 $ & $-0.40$ & $ <0.00001$ \\[5pt] *$D$ \& $|\Delta \textrm{RM} |$ & $D>1$& $D$& $|\Delta \textrm{RM} | \lessgtr |\Delta \overline{\textrm{R}}\textrm{M}| $ & $ 0.017 $ & $0.26$ & $ 0.00009 $ \\[5pt] $D$ \& $|\Delta \textrm{RM}|$ & $D<1$ &$D$ & $|\Delta \textrm{RM}| \lessgtr |\Delta \overline{\textrm{R}}\textrm{M}| $ & $ 0.12$ & -0.14 & $ 0.12$ \\[5pt] $|\log(D)|$ \& $|\textrm{RRM}_\mathrm{T}| $ & - &$|\log(\mathrm{D})|$ & $|\textrm{RRM}_\mathrm{T}| \lessgtr |\textrm{R}\overline{\textrm{RM}}_\mathrm{T}|$&$ 0.019$ & $0.21$ & $0.00007 $ \\[5pt] $|\textrm{RRM}_\mathrm{T}| $ \& $\pi_\mathrm{NV}$ & - &$\pi_\mathrm{NV}$ &$|\textrm{RRM}_\mathrm{T}| \lessgtr |\textrm{R}\overline{\textrm{RM}}_\mathrm{T}|$ & $0.0020 $ & $-0.25$ & $ <0.00001$ \\[5pt] *$|\Delta \textrm{RM}| $ \& $\pi_\mathrm{NV}$ & - & $|\Delta \textrm{RM}| $ & $\pi_\mathrm{NV} \lessgtr \bar{\pi}_\mathrm{NV}$ & $<0.00001 $ & $-0.44$ & $<0.00001 $ \\[5pt] $I_\mathrm{SP} $ \& $|\log(D)|$ &$\alpha<-0.5$ & $|\log(D)|$& $I_\mathrm{SP} \lessgtr \bar{I}_\mathrm{SP} $ &$ 0.40$ & $0.01$ & $ 0.80$ \\[5pt] *$D$ \& $\alpha$ & - & $D$ & $\alpha \lessgtr-0.5$ & $<0.00001$ & $-0.26$ & $< 0.00001$ \\[5pt] $D$ \& $z$ & $\alpha<-0.5$~\&~$D\ge1.5$ &$D$ & $z \lessgtr \bar{z}$ &$0.015$ &$-0.36$ &$ 0.011$ \\[5pt] {$I_\mathrm{SP}$ \& $z$} & $\alpha<-0.5$ &$I_\mathrm{SP}$ & $z \lessgtr \bar{z}$ &$ 0.014$ &-0.12 & $0.14 $ \\[5pt] {$I_\mathrm{SP}$ \& $z$} & $\alpha \ge -0.5$ &$I_\mathrm{SP}$ & $z \lessgtr \bar{z}$ &$0.14 $ &-0.28 & $0.0062 $ \\[5pt] {$D$ \& $z$} & $\alpha<-0.5$ &$D$ & $z \lessgtr \bar{z}$ &$0.44 $ &-0.03 & $ 0.75$ \\[5pt] {$D$ \& $z$} & $\alpha \ge -0.5$ &$D$ & $z \lessgtr \bar{z}$ &$ 0.15$ &0.10 & $ 0.36$ \\[5pt] {$|\textrm{RRM}_\mathrm{T}|$ \& $z$} & - &$|\textrm{RRM}_\mathrm{T}|$ & $z \lessgtr \bar{z}$ &$0.074 $ &0.11 & $ 0.10$ \\[5pt] {$\pi_\mathrm{SP}$ \& $z$} & $\alpha < -0.5$ &$\pi_\mathrm{SP}$ & $z \lessgtr \bar{z}$ &$ 0.73$ &-0.05 & $ 0.51$ \\[5pt] {$\pi_\mathrm{SP}$ \& $z$} & $\alpha<-0.5 ~\&~ D \ge 1.5$ &$\pi_\mathrm{SP}$ & $z \lessgtr \bar{z}$ &$ 0.79$ &0.12 & $ 0.41$ \\[5pt] {$\pi_\mathrm{SP}$ \& $z$} & $\alpha \ge -0.5$ &$\pi_\mathrm{SP}$ & $z \lessgtr \bar{z}$ &$0.59 $ &0.01 & $ 0.96$ \\[5pt] {$\pi_\mathrm{NV}$ \& $z$} & $\alpha<-0.5 ~\&~ D \ge 1.5$ &$\pi_\mathrm{NV}$ & $z \lessgtr \bar{z}$ &$ 0.15$ &0.26 & $ 0.075$ \\[5pt] {$|\Delta \textrm{RM}|$ \& $z$} & - &$|\Delta \textrm{RM}|$ & $z \lessgtr \bar{z}$ &$ 0.47$ &0.04 & $0.55 $ \\[5pt] *$W1-W2$ \& $\alpha$ & - & $W1-W2$ & $\alpha \lessgtr -0.5$ & $<0.00001$ & $0.27$ & $<0.00001$ \\[5pt] {$W1-W2$ \& $D$} & - &$W1-W2$ & $D \lessgtr 0.6$&$ 0.022$ &$ -0.12$&$0.045$ \\[5pt] {$W1-W2$ \& $D$} & - &$W1-W2$ & $0.6<D <1.7$ \& $D>1.7$& $0.31$ & -0.06& $0.38$ \\[5pt] {$W2-W3$ \& $D$} & - &$W2-W3$ & $0.6<D <1.7$ \& $D>1.7$& $0.025$ & 0.06& $0.38$ \\[5pt] {$W1-W2$ \& $D$} & $\alpha <-0.5$ &$D$ & $W1-W2 \lessgtr 0.6$ & $0.27$ & $-0.06$ & $0.42$ \\[5pt] {$W1-W2$ \& $\pi_\mathrm{SP}$} & $\alpha <-0.5$ &$\pi_\mathrm{SP}$ & $W1-W2 \lessgtr 0.6$ & $ 0.62$ & $-0.07$ & $0.36$ \\[5pt] \enddata \tablenotetext{}{Note: The * symbol in the beginning of some of the rows indicates that at least one of the tests resulted in p-value $ \le 10^{-4}$ and p$_\textrm{robust} \le 2\times10^{-3}$. The ? symbol represents suggestive correlations with p-value $ \le 10^{-4}$ and $2\times10^{-3}<$p$_\textrm{robust}< 0.05$. } \end{deluxetable*} \subsection{WISE colors and polarization}\label{spwise} We matched our catalog to the Wide-field Infrared Survey Explorer, WISE, catalog, \cite{2010AJ....140.1868W}, with a search radius of five arc-seconds. Out of 533 objects, 455 have WISE counterparts. All of them are detected with at least $5\sigma$ in the WISE $3.4 \mu m$ band, W1, while 445 (323) have $> 5\sigma$ detection in $4.6 \mu m$, W2, ($12 \mu m$, W3) band. $W1-W2$ and $W2-W3$ colors can be used to separate different galaxy populations such as AGNs and ellipticals \citep{2010AJ....140.1868W,2011ApJ...735..112J}. Recently, \cite{2015MNRAS.453.2326B} studied WISE colors of a large sample of resolved radio galaxies from the Radio Galaxy Zoo project, and found that most radio objects can be classified as ellipticals, AGNs and LIRGs. Figure \ref{fig:wise5arc2} shows the WISE color-color diagram of objects in our sample for which we have depolarization measurements and WISE counterparts. All objects used in Figure \ref{fig:wise5arc2} have W1 and W2 detections larger than $5\sigma$ and with small errors in the $W_2-W_3$ colors $\sigma_{(W2-W3)} < 0.4$. We investigated the possible dependence of the polarization and depolarization on WISE colors. The WISE dependence is difficult to isolate, since flat and steep spectrum objects have different WISE and different depolarization distributions. We therefore looked at steep spectrum objects only, and found that neither $\pi_\mathrm{SP}$ or $|\log(D)|$ were significantly correlated with WISE colors (Table \ref{table2}). We do not sample the ``elliptical'' region of WISE color space, which makes up a distinct population in the \cite{2015MNRAS.453.2326B} study. \begin{figure}[h] \centering \includegraphics[scale=0.51]{wise5arc2.eps} \caption{Distribution of objects with steep, $\alpha<-0.5$ and flat, $\alpha \ge -0.5$ spectral indices in the WISE color-color diagram. \label{fig:wise5arc2}} \end{figure} \subsection{Redshift Dependence}\label{zev} There are 222 objects in our sample that are detected in both NVSS and S-PASS polarization maps and have redshifts in \cite{2012arXiv1209.1438H} catalog. Figure \ref{zhist} shows the redshift distribution of the 222 matched sources, as well as the separated distributions of steep and flat spectrum objects. Steep spectrum objects are located within $0 \le z \le 2.34$ with median redshift of $\bar z=0.64$ while flat spectrum sources, as expected for a flux limited sample, tend to have larger redshifts, $0.22 \le z \le 2.81$, with median of $\bar z=1.18$. \begin{figure}[h] \centering \includegraphics[scale=0.51]{zhist.eps} \caption{Redshift distribution of the matched radio sources with \cite{2012arXiv1209.1438H} catalog. Histograms of steep ($\alpha < -0.5$) and flat ($\alpha \ge -0.5$) spectrum sources are shown in dashed blue and dotted-dashed red lines. \label{zhist}} \end{figure} As discussed earlier, steep and flat spectrum objects have different depolarization distributions and therefore, we studied their redshift evolution separately. We examined the redshift dependence of only depolarized steep spectrum sources ($D \ge 1.5$), since we expected to see a change in polarization properties due to the change in rest frame wavelength. We used the threshold $D=1.5$ to choose as many highly depolarized sources as possible while excluding the scattered objects that are in the vicinity of the observed peak at $D\sim1$ in Figure \ref{fig:depfp}. We found weak evidence for a decrease in depolarization of these sources as redshift increases (Spearman $r_s=-0.36$, p=0.011), which does not cross our conservative detection threshold. The average $\pi_\mathrm{NV}$ of 49 steep spectrum sources with $D \ge 1.5$ seems to increase from $\bar\pi_\mathrm{NV}=0.46\%$ at $z \le 0.5$ to $\bar\pi_\mathrm{NV}=1.02\%$ at $z \ge 0.5$, while their observed depolarization decreases and $\pi_\mathrm{SP}$ stays almost fixed. Figure \ref{depolsample} shows the running median of $\pi_\mathrm{NV}$ and $D$ calculated in bins of redshift as well as the expected evolutionary behavior of the three depolarizing scenarios. We will discuss this more in Section \ref{redshift}. \begin{figure}[h] \centering \includegraphics[scale=0.51]{depolsamplenew.eps} \includegraphics[scale=0.51]{depolsample2new.eps} \caption{Top: Fractional polarizations at 1.4 GHz, $\pi_\mathrm{NV}$, of depolarized steep spectrum sources with $D \ge 1.5$ versus redshift. Bottom: depolarization, $D$, of the same sample of sources versus redshift. The solid red lines represent the running medians of the $\pi_\mathrm{NV}$ (top) and $D$ (bottom) in bins of redshift. The green dotted, dashed blue and purple dashed-dotted lines are representations of the following three cases with B66 depolarization models: 1. A depolarizing screen located at the redshift of the source, 2. Combination of two depolarizing components, one Galactic and one at the redshift of the source, and 3. An evolving $\sigma_{\phi}$ at the depolarizing screen at the source redshift. \label{depolsample}} \end{figure} On the other hand, we do not find any change with redshift in depolarization of separate samples of steep or flat spectrum sources which include all re-polarized and depolarized sources. The median, $\log(D) \approx 0.1$, and standard deviation $\sigma_{\log(D)} \approx 0.26$, of steep spectrum objects stay almost constant with increasing redshift. Flat spectrum sources appear to be mostly re-polarized at $z<1$ while at higher redshifts the number of re-polarized and depolarized flat spectrum objects are almost the same. However, as listed in Table \ref{table2}, none of the KS and Spearman tests could confirm such a redshift dependence among flat spectrum sources. We also performed both KS and Spearman rank tests on $|\textrm{RRM}|$ and $|\Delta \textrm{RM}|$, and did not detect any noticeable redshift dependence (Figure \ref{zrrmt}). The 2.3 GHz fractional polarization of steep and flat spectrum sources also stays fixed at all cosmic times, although have different average values for populations of steep and flat objects. \begin{figure}[h] \centering \includegraphics[scale=0.51]{zrrmt.eps} \caption{ Distribution of the $|\textrm{RRM}|$ for the 206 objects is plotted versus redshift, $z$. Blue and red crosses represent objects with $\alpha <-0.5$ and $\alpha \ge -0.5$. The solid black line shows the running medians of the $|\textrm{RRM}|$ of all sources. The orange filled circles are the data points extracted from Figure 3 of \cite{2008ApJ...676...70K} as discussed in Section \ref{redshift}. Each circle represents the median value of their $|\textrm{RRM}|$ for each redshift bin. \label{zrrmt}} \end{figure} \subsection{Summary of major results} \begin{enumerate} \item The majority of extragalactic radio objects with $I_\mathrm{SP} \ge 420$ mJy have degrees of polarization on the order of 2\% to 3\% at both 1.4 GHz and 2.3 GHz. \item $\pi_\mathrm{SP}$ and $|\log(D)|$ are anti-correlated. On average, objects that are not depolarized ($|\log(D)| \le 0.23$), have median fractional polarizations of $\bar \pi_\mathrm{SP} \approx \bar \pi_\mathrm{NV} \approx 3\% - 4\%$, with $\bar \pi_\mathrm{SP} \approx 2\%$ for more depolarized objects and $\bar \pi_\mathrm{SP} \approx 1\%$ for re-polarized sources. Objects with high fractional polarizations ($\pi_\mathrm{SP}\approx \pi_\mathrm{NV} \approx 10\%$) are not depolarized ($|\log(D)|\approx 0$). \item Flat and steep spectrum objects have different polarization properties. 55\% of flat spectrum sources are re-polarized, compared to only 24\% for steep spectrum sources. Steep spectrum sources have larger degrees of polarization as well as stronger average depolarization. \item Extended objects ($>20''$) have higher fractional polarizations ($\bar{\pi}_\mathrm{SP}=4\%$) and smaller depolarizations ($|\log(D)| \sim 0.13$) than compact sources ($\bar{\pi}_\mathrm{SP} \sim 2 \%$, $|\log(D)| \sim 0.20$). \item Almost 24\% of the objects with detected polarization have $D > 2$. An additional 10\% of all sources may be too depolarized to be included in our sample. \item On average, sources with large $|\log(D)|$ (depolarized or re-polarized) show larger changes in $\textrm{RM}$ with wavelength ($\Delta \textrm{RM}$). \item We find weak evidence for a redshift dependence of the depolarization in a sub-sample of sources, those with steep spectra and $D\ge1.5 $. \item We do not find any evidence for changes of the observed 2.3 GHz fractional polarization, depolarization, $|\textrm{RRM}_\mathrm{T}|$ and $\Delta \textrm{RM}$ from $z=0$ to $z=2$ when all sources are considered. The median degree of polarization of both steep (141) and flat (81) spectrum sources with known redshift remain almost constant at $\pi_\mathrm{SP} \approx 2.5 \%$ and $\pi_\mathrm{SP} \approx 2.0 \%$ respectively. \item A large scatter in both depolarization and fractional polarization is seen at all redshifts. \item We did not find any evidence for angular clustering in the distribution of the depolarized sources. \item Both $\pi$ and $|\log(D)|$ of steep spectrum sources are independent of WISE $W_1-W_2$ color. \end{enumerate} \section{Discussion}\label{discussion} \subsection{Radio source field disorder}\label{obs} While radio synchrotron radiation can potentially be highly polarized, the NVSS and S-PASS fractional polarizations of most objects in our sample are around $2\% - 3\%$, and very rarely exceed $10\%$ (Figure \ref{fig:fphist}). Depolarization due to the presence of an irregular Faraday screen between the source and the observer, e.g., can potentially reduce the initial degree of the polarization, generally leading to higher fractional polarizations at higher frequencies \citep{1966MNRAS.133...67B,1991MNRAS.250..726T}. However, between 1.4 GHz and 2.3 GHz we find that the majority of extragalactic objects experience only small depolarizations, with $60\%$ of the objects have $0.6< D < 1.7$. Moreover, objects with the strongest fractional polarizations ($\pi\approx 10\%$) have little depolarization. The reduction from a theoretical maximum of $\sim$40-70\% to either $\approx$10\% with no depolarization, or $\approx$3\%, with modest depolarization, must therefore be due to field disorder. To approximate the necessary number of randomly oriented magnetic field patches within an unresolved source, we performed a simple simulation. We considered a uniform brightness two dimensional source, with equal fractional polarizations $\pi_0=50\%$ in each patch. By randomizing the polarization angles, we estimated that sources currently unresolved in our beam should contain approximately 70 to 80 independent magnetic patches to reduce the observed fractional polarization to $\sim$4\%. There is a subset of sources where depolarization does play a significant role. Almost, 24\% of sources with detected polarizations have $D>2$. Moreover, we estimated a missing $\approx$10\% population of heavily depolarized sources. It is not clear how strong an effect field disorder has for that subset. \subsection{Prospects for high frequency surveys} One important implication of these results is for surveys at higher frequencies, where one might expect to increase number counts by a large factor because of less depolarization. However, changing the frequency of observation from L to S band will not result in a major increase in the number of polarized detections. the number of polarized objects. As an example, the number of sources with polarized flux densities larger than 10 mJy in our sample is almost equal at both 2.3 GHz and 1.4 GHz (368 in S band and 363 in L band). Future polarization surveys and the Square Kilometer Array, SKA \citep{2011arXiv1111.5802B} precursors such as Polarization Sky Survey of the Universe's Magnetism, POSSUM \citep{2010AAS...21547013G}, Westerbork Observations of the Deep APERTIF Northern sky, WODAN \citep{2012MNRAS.427.2079C}, MeerKAT International GigaHertz Tiered Extragalactic Exploration survey, MIGHTEE \citep{2012AfrSk..16...44J}, Very Large Array Sky Survey, VLASS \citep{2014arXiv1401.1875M} and VLASS Deep will detect hundreds of thousands of polarized sources in different frequencies. The VLASS will operate at S band from 2 to 4 GHz and has angular resolution and sensitivity of $\sim 3.5$ arcsec and 0.7 mJy per beam respectively. The number density of flat spectrum sources is expected to be similar in L and S bands since their flux density is almost independent of the frequency, and their median depolarization is $\bar D \sim 1$ as shown in Figure \ref{fig:depol}. On the other hand, steep spectrum, $\alpha < -0.5$, sources in our sample with median $\bar{\alpha}=-0.9$ are on average fainter at S band by a factor of 1.4. Therefore, their number density at a fixed signal to noise reduces. However, the median polarization of steep spectrum objects in our sample is approximately 1.3 times higher at 2.3 GHz than 1.4 GHz at resolutions as low as S-PASS, $\sim 9$ arcmin. This indicates that the median polarization flux density of these objects should have been reduced by $\sim$17\%. \cite{2014ApJ...785...45R} showed at 1.6 arcsec resolution there are $\sim 6$ polarized sources per squared degree at 0.7 mJy per beam and S:N $>$ 10 in L band, and the integrated number density of objects with polarization flux density larger than $p$ goes as $N_p\propto p^{-0.6} $. As a result, one can expect to detect roughly 11\% less polarized objects at S band compared to L band at 1.6 arcsec resolution. All in all, considering the larger beam size of the VLASS all sky survey one can expect to detect approximately the same number of polarized sources in S band as the calculation of \cite{2014ApJ...785...45R} in L band. This is already a factor of six above the existing surface density of polarized sources from the NVSS catalog in L band. \subsection{Prospects for $\textrm{RM}$ grid experiments} There is strong interest in measuring and estimating the intergalactic magnetic field in clusters of galaxies or in cosmic filaments through $\textrm{RM}$ analysis and tomography, e.g. \cite{2014PASJ...66...65A}. In the presence of a single Faraday screen along the line of sight, the rotation angle of the radio polarization vector of extra-galactic sources depends linearly on $\lambda^2$. This simple relation makes it possible to estimate the magnetic field of the medium with some assumptions for the electron density, after subtracting out a Galactic component. However, any complication in the structure of the Faraday screen within the observation beam or along the line of sight through the emitting source will result in non-$\lambda^2$ behavior, and an inability to isolate the foreground screen of interest. We have measured the non-$\lambda^2$ behavior using $\Delta \textrm{RM}$. As shown in Figure \ref{drmp}, large $\Delta \textrm{RM}$s occur preferentially at low fractional polarizations. In order to avoid large values of $\Delta \textrm{RM}$, which would compromise any foreground experiment, it is necessary to use only fractional polarizations ($\ge 3-4\%$). This will cause a reduction in the number of available sources; only 33\% of sources in our sample have $\pi_{SP} >$ 3\%. However, if reliable $\chi(\lambda^2)$ were available for some subset of sources, then it might be possible to increase this number. \subsection{Origins of depolarization} As shown in Section \ref{galD} we did not detect any angular clustering of sources by fractional polarization or depolarization, that would have implied a Galactic origin. We can not rule out the possibility of Galactic $\textrm{RM}$ fluctuations on arcsec scales, but these are likely to be extremely small and we do not consider them further here. The dependence of depolarization on spectral index shows that it must primarily occur local to the source. If depolarization is local to the environment of the source, then it may show signs of dependence to some intrinsic characteristics of the source such as spectral index or the luminosity. The results found here on the spectral behavior are consistent with \cite{2014ApJS..212...15F} who did a multi-wavelength polarization study on sources selected from the TSS09 catalog. The dependence of polarization properties of objects on their angular extent (Section \ref{extent}) also supports the local depolarization scenario. As shown in Figure \ref{fig:dsize}, compact sources seem to have larger depolarizations ( $|\log(D)| \sim 0.20$ vs. $\sim 0.13$) and smaller fractional polarizations ($\bar{\pi}_\mathrm{SP}=4\%$ vs. 2\%) than sources extended in NVSS. This is inconsistent with irregular screens either Galactic or extragalactic, which should yield higher fractional polarizations and less depolarization for compact sources. Thus, the depolarization must arise in a Faraday component directly related to the source. If Galactic or intervening Faraday screens were the dominant depolarizing components then we expect to see larger depolarization in a sample of extended sources. \subsubsection{The origin of the total intensity and fractional polarization anti-correlation}\label{origin} The anti-correlation between total intensity and fractional polarization at 1.4 GHz has been extensively discussed (such as \citealt{2002A&A...396..463M, 2004MNRAS.349.1267T, 2007ApJ...666..201T, 2010ApJ...714.1689G,2010MNRAS.402.2792S,2014ApJ...787...99S}). Recently, \cite{2014MNRAS.444..700B} used WISE colors to suggest that the anti-correlation was due to the difference in environments between WISE-AGNs (IR colors dominated by AGN) and WISE-Ellipticals (IR colors dominated by starlight). These effects are likely confused by the fact that the anti-correlation is found only among steep-spectrum sources, as discussed in Section \ref{ip}. The WISE-AGN class contains a large fraction of flat spectrum objects, for which we find no anti-correlation, while the WISE-Ellipticals are largely steep-spectrum \citep{2014MNRAS.444..700B}. The dependence we found on the spectral index is also consistent with \cite{2004MNRAS.349.1267T} and the stacking analysis of \cite{2015arXiv150100390S}. The limited range of $I_\textrm{SP}$ in our sample makes it difficult to study these effects. However, to illuminate the underlying issues, we note that the suggestive anti-correlation between $I_\mathrm{SP}$ and $\pi_\mathrm{SP}$ of steep spectrum sources must arise from some physical difference in properties between the bright and faint sources that are not expected in fair, uniform samples. We have not been able to identify this underlying parameter. We find no statistically significant anti-correlation between $L_\mathrm{SP}$ and $\pi_\mathrm{SP}$. We attempted to correct for the size dependence, in case that was a confounding variable, but the anti-correlation remained. Size could still be an important factor, since the resolution of even the NVSS is much larger than the typical source size. Higher resolution observations of this sample could reveal, e.g., that the bright sources are much more compact and dominated by central AGN, as opposed to fainter, lobe-dominated structures with more ordered fields. Depolarization might also be playing a role, since $\pi_\mathrm{SP}$ is correlated with the $|\log(D)|$. However, again, the anti-correlation breaks down when we look at $L_\mathrm{SP}$ and $|\log(D)|$. This leaves us back, again, at some as yet undetermined physical difference between the faint and bright sources. \subsubsection{Re-polarized objects} We showed that most re-polarized objects have flat spectra ($\alpha \ge -0.5$), and are therefore concentrated in the WISE-AGN population (Figure \ref{fig:wise5arc2}). This makes it likely that they contain a high proportion of compact nuclei with polarization SEDs influenced by self-absorbed, and perhaps Faraday thick components. This is consistent with \cite{2014ApJS..212...15F} who also found flat spectrum objects have complex polarization behaviors. While 61\% of re-polarized objects have flat spectra and are optically thick sources, the remaining 39\% have steep spectra. The nature of these objects is not clear. However, there are few proposed models in the literature. re-polarization can occur when there is interference between two (or a few) unresolved and separate Faraday patches in the beam of the telescope. This can result in an oscillatory behavior of the fractional polarization with changing frequency as discussed in \cite{2011AJ....141..191F} and \cite{1984ApJ...283..540G}.\cite{2012AJ....144..105H} studied the AGN jet structure of 191 extragalactic radio objects, and found multiple regions along the jets of a few objects show signs of re-polarization. As discussed in \cite{2012ApJ...747L..24H} they argue that both internal Faraday rotation in the jet medium as well as the configuration of the magnetic fields can explain the observed re-polarization in these optically thin jets. In Faraday thick regions the rotation of the polarization angles might align the polarization vectors from the far and near sides along the line of sight which can potentially result in re-polarization. \subsection{Redshift Evolution}\label{redshift} The evolution of the magnetic properties of galaxies with time has been subject of multiple studies (such as \citealt{2012arXiv1209.1438H, 2008ApJ...676...70K,2008Natur.454..302B,2005MNRAS.359.1456G,1995ApJ...445..624O,1984ApJ...279...19W}). We distinguish here between two different quantities, an \emph{observed} redshift dependence and an \emph{inferred} redshift evolution, based on applying the polarization equivalent of a K-correction (redshift dilution). As discussed in Section \ref{zev}, we found weak evidence that the average observed depolarization of steep spectrum depolarized sources with $D \ge 1.5$ decreases with increasing redshift, while the 1.4~GHz fractional polarization increases (the 2.3~GHz fractional polarization shows no change). The detected redshift variations are weak, compared to the scatter, and their probability (0.011) does not cross our conservative detection threshold. However, given the importance of this issue, we discuss the causes and consequences of redshift dependencies to help clarify the underlying issues. Polarization SEDs are often complex, especially for flat spectrum sources. This is seen in our numerous detections of re-polarization, and the broad wavelength SEDs cataloged by \cite{2014ApJS..212...15F}. In such cases, it is impossible to predict the trends of depolarization and fractional polarization with redshift expected from the K-correction. In the case where $D \sim 1$, no redshift dependence is expected, since there is no wavelength dependence to the fractional polarization. Therefore, the fact that we observe decreasing depolarization and increasing 1.4~GHz fractional polarization at increased redshift only for steep-spectrum sources with $D > 1.5$ is consistent with K-corrections only, without any physical redshift evolution. We now look at this more quantitatively, assuming the simplest case of an unresolved source with an irregular depolarizing Faraday screen \citep{1966MNRAS.133...67B}(B66), external to, but at the same redshift as the source. The expected fractional polarization behavior is then \begin{equation} \label{B66} \pi=\pi_0 \exp(-C\lambda_{rest}^4) \end{equation} where $\pi_0$ is the initial fractional polarization and $C \propto \sigma_{\phi}^2$ is a function of the dispersion in the Faraday depth. For a region with electron density $n$ and magnetic field component parallel to the line of sight $B_z$, fluctuations in the parameter $nB_z$ over the extent of the region is represented by $\sigma_{\phi}$. Assuming no physical change in $\sigma_{\phi}$ with time, the redshift dilution effect results in an increase in the observed fractional polarization, $\pi\propto exp\left(-C\lambda^4(1+z)^{-4}\right)$. The observed depolarization also decreases with redshift since $D\propto exp\left(C(\lambda_\mathrm{NV}^4-\lambda_\mathrm{SP}^4)(1+z)^{-4}\right)$. This simplest picture (Model 1), however, is not quantitatively consistent with our observations (Figure \ref{depolsample}). We therefore considered two additional models based on the B66 screen. Model 2: A combination of two depolarizing components, one Galactic or relatively local to us, and one at the redshift of the source, and Model 3: A physical change in $\sigma_{\phi}$ of the depolarizing screen at the source redshift. As shown in Figure \ref{depolsample}, the general behavior of the observed $\pi_\mathrm{NV}$, and $D$ as well as $\pi_\mathrm{SP}$ (not shown) of the depolarized steep spectrum sources and their evolution with redshift can be explained by models 2 and 3. However, a single depolarizing component, local to the source, with no evolution in $\sigma_{\phi}$ does not seem to be consistent with the observation. Larger samples, and resolved polarization maps where the Faraday structure can be directly seen, are needed to clarify these results. As an alternative to the B66 screen, \cite{1991MNRAS.250..726T} suggested depolarization can be modeled as power law $\pi \propto \lambda^{-4/m}$ at wavelengths larger than $\lambda_{1/2}$, at which the degree of polarization is equal to half of its maximum value. The above relation only holds under certain condition in which the Faraday screen $\textrm{RM}$ structure function varies as a power law across the source $S(\delta x) \propto \delta x^m$ where $S(\delta x)\equiv <[\textrm{RM}(x+\delta x)-\textrm{RM}(x)]^2>$ and $x$ is the angular coordinate. If we assume the fractional polarization of unresolved objects follows any power law model with arbitrary exponent $-4/m$ and a constant related to the $\textrm{RM}$ dispersion, $\pi =C \lambda^{-4/m}$, then the observed depolarization, $D=\pi_{SP}/\pi_{NV}$, and both the redshift and the $\sigma_{\phi}$ dependences cancel out. Therefore, one can expect to observe no evolution in the average $D$ even if $\sigma_{\phi}$ changes with redshift, contrary to what we observe. \subsubsection{Comparisons to previous work} Earlier work has been based on samples including sources with both flat and steep spectra, and without selections based on depolarization. For our full sample, we find no redshift trends in fractional polarizations or depolarization, or measures of increased Faraday structure such as $|\textrm{RRM}_\mathrm{T}|$ and $|\Delta \textrm{RM}|$. This is consistent with the negative results from \cite{2012ApJ...761..144B} and \cite{2012arXiv1209.1438H}. In addition, their samples were taken from the TSS09 catalog, which is biased towards high fractional polarizations, and thus, towards depolarizations D$\sim$1, for which no redshift evolution is expected. Our data are inconsistent with the analysis of \cite{2008ApJ...676...70K}, who claimed that the rotation measure of galaxies at redshifts larger than $z=1$ are on average larger (by $\sim 10$ rad m$^{-2}$) than the low redshift objects, despite the redshift dilution effect. In Figure \ref{zrrmt} we show $|\textrm{RRM}_\mathrm{T}|$ versus the redshift of objects in our sample and overlay the \cite{2008ApJ...676...70K} median $|\textrm{RRM}|$ values from their Figure 3. Our data are consistent with theirs, and show no evidence for the claimed increase in $\textrm{RRM}$. It is possible that a physical increase in $\sigma_{\phi}$ and depolarization as a function of redshift could mask the redshift dilution effect, leaving no observed redshift dependence to fractional polarization, $\textrm{RRM}$, $\Delta \textrm{RM}$ or depolarization. This is discussed with more details in \cite{2012arXiv1209.1438H}, \cite{2008ApJ...676...70K}, \cite{2008Natur.454..302B}, \cite{1995ApJ...445..624O} and \cite{1984ApJ...279...19W}. \cite{2005MNRAS.359.1456G} studied the redshift evolution of the depolarization of 26 resolved, powerful radio galaxies and quasars over the cosmic time. They applied corrections to the measured depolarizations based on models of the wavelength and resolution effects at different redshifts. They claim a physical evolution in $\sigma_{\phi}$ and depolarization as a function of redshift, but we cannot compare their results to ours, since neither the original data nor the details of the models are shown. \section{Conclusions} \label{summary} We constructed a depolarization ($D=\pi_{2.3}/\pi_{1.4}$) catalog of extragalactic radio sources brighter than $420$ mJy at 2.3 GHz including total intensities, spectral indices, observed and residual rotation measures, fractional polarization, depolarization as well as the redshift, 2.3 GHz luminosity and WISE magnitudes for almost half of the objects. We looked for possible correlations between these quantities and found that the fractional polarization of extragalactic radio sources depends on the spectral index, morphology, the intrinsic magnetic field disorder as well as the depolarization of these sources. We summarize our main conclusions as follows: \\ Consistent with previous studies over half of flat spectrum sources in our sample are re-polarized while the majority of steep spectrum objects are depolarized. There is also a significant population of steep-spectrum sources that are repolarized; their underlying physical structure is currently unknown. Although steep objects are more polarized at 2.3~GHz, they are fainter in total intensity, and therefore future surveys at higher frequencies will result in approximately the same number of sources at fixed sensitivity as the lower frequencies. Depolarization, and thus fractional polarizations, are related to the presence of Faraday structures indicated by the non-$\lambda^2$ behavior of polarization angles ($\Delta \textrm{RM}$). Future studies using polarized sources as background probes need to minimize $\textrm{RM}$ structures intrinsic to the sources. Such clean samples require high fractional polarizations ($\pi \ge 4\%$), which will severely limit the number of available sources. Sources with little or no depolarization between 1.4 GHz and 2.3 GHz have fractional polarizations ranging from a few to 10\%. This is much lower than the theoretical maximum, and therefore shows the dominant role of field disorder in creating low polarizations. Compact steep spectrum objects in the NVSS catalog have more Faraday structure, and are $\sim 2$ times less polarized at 2.3 GHz than the extended sources. We found suggestive evidence for a decrease in the depolarization from $z=0$ to $z=2.3$, but only when the sample is restricted to the steep spectrum, $\alpha < -0.5$, depolarized, $D \ge 1.5$ objects. More investigation is needed to confirm the depolarization trend. Assuming that it's real, it is likely the result of the redshift dilution effect (at least partially) but requires more than a simple depolarizing screen local to the source.\\ The National Radio Astronomy Observatory is a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc. Partial support for ML and LR comes from National Science Foundation grant AST-1211595 to the University of Minnesota. B.M.G. has been supported by the Australian Research Council through the Centre for All-sky Astrophysics (grant CE110001020) and through an Australian Laureate Fellowship (grant FL100100114). The Dunlap Institute is funded through an endowment established by the David Dunlap family and the University of Toronto. We would like to thank G. Bernardi and D. H. F. M. Schnitzeler and the referee for a number of useful conversations and comments on the manuscript. \bibliographystyle{apj}
31ec072526eada9187ae9e5d9bca9e7b1fe6f89c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{sec:intro} The constraint satisfaction problem (CSP) provides a framework in which it is possible to express, in a natural way, many combinatorial problems encountered in computer science and AI~\cite{Cohen06:handbook,Creignou01:book,Feder98:monotone}. An instance of the CSP consists of a set of variables, a domain of values, and a set of constraints on combinations of values that can be taken by certain subsets of variables. The basic aim is then to find an assignment of values to the variables that satisfies the constraints (decision version) or that satisfies the maximum number of constraints (optimization version). Since CSP-related algorithmic tasks are usually hard in full generality, a major line of research in CSP studies how possible algorithmic solutions depend on the set of relations allowed to specify constraints, the so-called {\em constraint language}, (see, e.g.~\cite{Bulatov??:classifying,Cohen06:handbook,Creignou01:book,Feder98:monotone,Krokhin17:book}). The constraint language is denoted by $\Gamma$ and the corresponding CSP by $\CSP\Gamma$. For example, when one is interested in polynomial-time solvability (to optimality, for the optimization case), the ultimate sort of results are dichotomy results~\cite{Bulatov17:dichotomy,Bulatov??:classifying,Feder98:monotone,Kolmogorov17:complexity,Thapper16:finite,Zhuk17:dichotomy}, pioneered by~\cite{Schaefer78:complexity}, which characterise the tractable restrictions and show that the rest are NP-hard. Classifications with respect to other complexity classes or specific algorithms are also of interest (e.g.~\cite{Barto14:jacm,Barto12:NU,Kolmogorov15:power,Larose09:universal}). When approximating (optimization) CSPs, the goal is to improve, as much as possible, the quality of approximation that can be achieved in polynomial time, see e.g. surveys~\cite{Khot10:UGCsurvey,MM17:cspsurvey}. Throughout the paper we assume that P$\ne$NP. The study of {\em almost satisfiable} CSP instances features prominently in the approximability literature. On the hardness side, the notion of approximation resistance (which, intuitively, means that a problem cannot be approximated better than by just picking a random assignment, even on almost satisfiable instances) was much studied recently, e.g.~\cite{Austrin13:usefulness,Chan13:resist,Hastad14:maxnot2,Khot14:strong}. Many exciting developments in approximability in the last decade were driven by the {\em Unique Games Conjecture} (UGC) of Khot, see survey~\cite{Khot10:UGCsurvey}. The UGC states that it is NP-hard to tell almost satisfiable instances of $\CSP\Gamma$ from those where only a small fraction of constraints can be satisfied, where $\Gamma$ is the constraint language consisting of all graphs of permutations over a large enough domain. This conjecture (if true) is known to imply optimal inapproximability results for many classical optimization problems~\cite{Khot10:UGCsurvey}. Moreover, if the UGC is true then a simple algorithm based on semidefinite programming (SDP) provides the best possible approximation for all optimization problems $\CSP\Gamma$~\cite{Prasad08:optimal}, though the exact quality of this approximation is unknown. On the positive side, Zwick~\cite{Zwick98:finding} initiated the systematic study of approximation algorithms which, given an almost satisfiable instance, find an almost satisfying assignment. Formally, call a polynomial-time algorithm for CSP {\em robust} if, for every $\eps>0$ and every $(1-\eps)$-satisfiable instance (i.e. at most a $\eps$-fraction of constraints can be removed to make the instance satisfiable), it outputs a $(1-g(\eps))$-satisfying assignment (i.e. that fails to satisfy at most a $g(\eps)$-fraction of constraints). Here, the {\em loss} function $g$ must be such that $g(\eps)\rightarrow 0$ as $\eps\rightarrow 0$. Note that one can without loss of generality assume that $g(0)=0$, that is, a robust algorithm must return a satisfying assignment for any satisfiable instance. The running time of the algorithm should not depend on $\eps$ (which is unknown when the algorithm is run). Which problems $\CSP\Gamma$ admit robust algorithms? When such algorithms exist, how does the best possible loss $g$ depend on $\Gamma$? \subsection*{Related Work} In~\cite{Zwick98:finding}, Zwick gave an SDP-based robust algorithm with $g(\eps)=O(\eps^{1/3})$ for {\sc 2-Sat} and an LP-based robust algorithm with $g(\eps)=O(1/\log(1/\eps))$ for {\sc Horn $k$-Sat}. Robust algorithms with $g(\eps)=O(\sqrt{\eps})$ were given in~\cite{Charikar09:near} for {\sc 2-Sat}, and in~\cite{Charikar06:near} for {\sc Unique Games($q$)} where $q$ denotes the size of the domain. For {\sc Horn-2-Sat}, a robust algorithm with $g(\eps)=2\eps$ was given in~\cite{Guruswami12:tight}. These bounds for {\sc Horn $k$-Sat} ($k\ge 3$), {\sc Horn $2$-Sat}, {\sc 2-Sat}, and {\sc Unique Games($q$)} are known to be optimal~\cite{Guruswami12:tight,Khot02:power,Khot07:optimal}, assuming the UGC. The algebraic approach to CSP~\cite{Bulatov??:classifying,Cohen06:handbook,Jeavons97:closure} has played a significant role in the recent massive progress in understanding the landscape of complexity of CSPs. The key to this approach is the notion of a {\em polymorphism}, which is an $n$-ary operation (on the domain) that preserves the constraint relations. Intuitively, a polymorphism provides a uniform way to combine $n$ solutions to a system of constraints (say, part of an instance) into a new solution by applying the operation component-wise. The intention is that the new solution improves on the initial solutions in some problem-specific way. Many classifications of CSPs with respect to some algorithmic property of interest begin by proving an algebraic classification stating that every constraint language either can simulate (in a specific way, via gadgets, -- see e.g.~\cite{Barto16:robust,Dalmau13:robust,Larose09:universal} for details) one of a few specific basic CSPs failing the property of interest or else has polymorphisms having certain nice properties (say, satisfying nice equations). Such polymorphisms are then used to obtain positive results, e.g. to design and analyze algorithms. Getting such a positive result in full generality in one step is usually hard, so (typically) progress is made through a series of intermediate steps where the result is obtained for increasingly weaker algebraic conditions. The algebraic approach was originally developed for the decision CSP~\cite{Bulatov??:classifying,Jeavons97:closure}, and it was adapted for robust satisfiability in~\cite{Dalmau13:robust}. One such algebraic classification result~\cite{Larose09:ability} gives an algebraic condition (referred to as $\mathrm{SD}(\wedge)$ or ``omitting types \textbf{1} and \textbf{2}'' -- see~\cite{Barto14:jacm,Kozik15:maltsev,Larose09:ability} for details) equivalent to the \underline{in}ability to simulate {\sc Lin-$p$} -- systems of linear equations over $Z_p$, $p$ prime, with 3 variable per equation. H\aa stad's celebrated result~\cite{Hastad01:optimal} implies that {\sc Lin-$p$} does not admit a robust algorithm (for any $g$). This result carries over to all constraint languages that can simulate (some) {\sc Lin-$p$}~\cite{Dalmau13:robust}. The remaining languages are precisely those that have the logico-combinatorial property of CSPs called ``{\em bounded width}'' or ``{\em bounded treewidth duality}''~\cite{Barto14:jacm,Bulatov09:bounded,Larose06:bounded}. This property says, roughly, that all unsatisfiable instances can be refuted via local propagation -- see~\cite{Bulatov08:duality} for a survey on dualities for CSP. Barto and Kozik used $\mathrm{SD}(\wedge)$ in~\cite{Barto14:jacm}, and then in~\cite{Barto16:robust} they used their techniques from~\cite{Barto14:jacm} to prove the Guruswami-Zhou conjecture~\cite{Guruswami12:tight} that each bounded width CSP admits a robust algorithm. The general bound on the loss in~\cite{Barto16:robust} is $g(\eps)=O((\log\log(1/\eps))/\log(1/\eps))$. It is natural to ask when a better loss can be achieved. In particular, the problems of characterizing CSPs where linear loss $g(\eps)=O(\eps)$ or polynomial loss $g(\eps)=O(\eps^{1/k})$ (for constant $k$) can be achieved have been posed in~\cite{Dalmau13:robust}. Partial results on these problems appeared in~\cite{Dalmau13:robust,Dalmau15:towards,Kun12:robust}. For the Boolean case, i.e. when the domain is $\{0,1\}$, the dependence of loss on $\Gamma$ is fully classified in~\cite{Dalmau13:robust}. \subsection*{Our Contribution} We study CSPs that admit a robust algorithm with polynomial loss. As explained above, the bounded width property is necessary for admitting any robust algorithm. {\sc Horn 3-Sat} has bounded width, but does not admit a robust algorithm with polynomial loss (unless the UGC fails)~\cite{Guruswami12:tight}. The algebraic condition that separates {\sc Lin-$p$} and {\sc Horn 3-Sat} from the CSPs that can potentially be shown to admit a robust algorithm with polynomial loss is known as $\mathrm{SD}(\vee)$ or ``omitting types \textbf{1}, \textbf{2} and \textbf{5}''~\cite{Dalmau13:robust}, see Section~\ref{sec:algebra} for the description of $\mathrm{SD}(\vee)$ in terms of polymorphisms. The condition $\mathrm{SD}(\vee)$ is also a necessary condition for the logico-combinatorial property of CSPs called ``{\em bounded pathwidth duality}'' (which says, roughly, that all unsatisfiable instances can be refuted via local propagation in a linear fashion), and possibly a sufficient condition for it too~\cite{Larose09:universal}. It seems very hard to obtain a robust algorithm with polynomial loss for every CSP satisfying $\mathrm{SD}(\vee)$ all in one step. From the algebraic perspective, the most general natural condition that is (slightly) stronger than $\mathrm{SD}(\vee)$ is the {\em near-unanimity (NU)} condition~\cite{Baker75:chinese-remainder}. CSPs with a constraint language having an NU polymorphism received a lot of attention in the literature (e.g.~\cite{Feder98:monotone,Jeavons98:consist,Barto12:NU}). Bounded pathwidth duality for CSPs admitting an NU polymorphism was established in a series of papers~\cite{Dalmau05:linear,Dalmau08:majority,Barto12:NU}, and we use some ideas from~\cite{Dalmau08:majority,Barto12:NU} in this paper. We prove that any CSP with a constraint language having an NU polymorphism admits a randomized robust algorithm with loss $O(\eps^{1/k})$, where $k$ depends on the size of the domain. It is an open question whether this dependence on the size of the domain is necessary. We prove that, for the special case of a ternary NU polymorphism known as {\em dual discriminator} (the corresponding CSP is a common generalisation of {\sc Unique Games} with a fixed domain and {\sc 2-Sat}), we can always choose $k=2$. Our algorithms use the standard SDP relaxation for CSPs. The algorithm for the general NU case follows the same general scheme as~\cite{Barto16:robust,Kun12:robust}: \begin{enumerate} \item Solve the LP/SDP relaxation for a $(1-\eps)$-satisfiable instance $\inst I$. \item Use the LP/SDP solution to remove certain constraints in $\inst I$ with total weight $O(g(\eps))$ (in our case, $O(\eps^{1/k})$) so that the remaining instance satisfies a certain consistency condition. \item Use the appropriate polymorphism (in our case, NU) to show that any instance of $\CSP\Gamma$ with this consistency condition is satisfiable. \end{enumerate} Steps 1 and 2 in this scheme can be applied to any CSP instance, and this is where essentially all work of the approximation algorithm happens. Polymorphisms are not used in the algorithm, they are used in Step 3 only to prove the correctness. While the above general scheme is rather simple, applying it is typically quite challenging. Obviously, Step 2 prefers weaker conditions (achievable by removing not too many constraints), while Step 3 prefers stronger conditions (so that they can guarantee satisfiability), so reaching the balance between them is the main (and typically significant) technical challenge in any application of this scheme. Our algorithm is somewhat inspired by~\cite{Barto16:robust}, but it is also quite different from the algorithm there. That algorithm is designed so that Steps 1 and 2 establish a consistency condition that, in particular, includes the 1-minimality condition, and establishing 1-minimality alone requires removing constraints with total weight $O(1/\log{(1/\eps)})$~\cite{Guruswami12:tight}, unless UGC fails. To get the right dependency on $\eps$ we introduce a new consistency condition somewhat inspired by~\cite{Barto12:NU,Kozik16:circles}. The proof that the new consistency condition satisfies the requirements of Steps 2 and 3 of the above scheme is one of the main technical contributions of our paper. \subsubsection*{Organization of the paper} After some preliminaries, we formulate the two main results of this paper in Section \ref{sec:main}. Section \ref{sec:SDP} then contains a~description of SDP relaxations that we will use further on. Sections \ref{sec:overview1} and \ref{sec:overview-thm2} contain the description of the algorithms for constraint languages compatible with NU polymorphism and dual discriminator, respectively; the following chapters prove the correctness of the two algorithms. \section{Preliminaries} \subsection{CSPs} Throughout the paper, let $D$ be a {\em fixed finite} set, sometimes called the {\em domain}. An {\em instance} of the $\csp$ is a pair $\inst I=(V,{\mathcal C})$ with $V$ a finite set of {\em variables} and ${\mathcal C}$ is a finite set of constraints. Each constraint is a pair $(\overline{x},R)$ where $\overline{x}$ is a tuple of variables (say, of length $r>0$), called the {\em scope} of $C$ and $R$ an $r$-ary relation on $D$ called the {\em constraint relation} of $C$. The arity of a constraint is defined to be the arity of its constraint relation. In the weighted optimization version, which we consider in this paper, every constraint $C\in{\mathcal C}$ has an associated {\em weight} $w_C\geq 0$. Unless otherwise stated we shall assume that every instance satisfies $\sum_{C\in{\mathcal C}} w_C=1$. An {\em assignment} for $\inst I$ is a mapping $s:V\rightarrow D$. We say that $s$ satisfies a constraint $((x_1,\dots,x_r),R)$ if $(s(x_1),\dots,s(x_r))\in R$. For $0\leq \beta\leq 1$ we say that assignment $s$ $\beta$-satisfies $\inst I$ if the total weight of the constraints satisfied by $s$ is at least $\beta$. In this case we say that $\inst I$ is $\beta$-satisfiable. The best possible $\beta$ for $\inst I$ is denoted by $\mathrm{Opt}(\inst I)$. A {\em constraint language} on $D$ is a {\em finite} set $\Gamma$ of relations on $D$. The problem $\csp(\Gamma)$ consists of all instances of the CSP where all the constraint relations are from $\Gamma$. Problems {\sc $k$-Sat}, {\sc Horn $k$-Sat}, {\sc Lin-$p$}, {\sc Graph $H$-colouring}, and {\sc Unique Games$(|D|)$} are all of the form $\CSP\Gamma$. The {\em decision problem} for $\csp(\Gamma)$ asks whether an input instance $\inst I$ of $\csp(\Gamma)$ has an assignment satisfying all constraints in $\inst I$. The {\em optimization problem} for $\csp(\Gamma)$ asks to find an assignment $s$ where the weight of the constraints satisfied by $s$ is as large as possible. Optimization problems are often hard to solve to optimality, motivating the study of {\em approximation} algorithms. \subsection{Algebra}\label{sec:algebra} An $n$-ary operation $f$ on $D$ is a map from $D^n$ to $D$. We say that $f$ {\em preserves} (or is a {\em polymorphism} of) an $r$-ary relation $R$ on $D$ if for all $n$ (not necessarily distinct) tuples $(a^i_1,\dots,a_r^i)\in R$, $1\leq i\leq n$, the tuple $(f(a_1^1,\dots,a_n^1),\dots,f(a_1^r,\dots,a_n^r))$ belongs to $R$ as well. Say, if $R$ is the edge relation of a digraph $H$, then $f$ is a polymorphism of $R$ if and only if, for any list of $n$ (not necessarily distinct) edges $(a_1,b_1),\ldots,(a_n,b_n)$ of $H$, there is an edge in $H$ from $f(a_1,\ldots,a_n)$ to $f(b_1,\ldots,b_n)$. If $f$ is a polymorphism of every relation in a constraint language $\Gamma$ then $f$ is called a polymorphism of $\Gamma$. Many algorithmic properties of $\CSP\Gamma$ depend only on the polymorphisms of $\Gamma$, see survey~\cite{Barto17:poly}, also~\cite{Bulatov??:classifying,Dalmau13:robust,Jeavons97:closure,Larose09:universal}. An $(n+1)$-ary ($n\ge 2$) operation $f$ is a {\em near-unanimity (NU)} operation if, for all $x,y\in D$, it satisfies \begin{multline*} f(x,x,\ldots,x,x,y)=f(x,x,\ldots,x,y,x)=\dots =f(y,x,\ldots,x,x,x)=x. \end{multline*} Note that the behaviour of $f$ on other tuples of arguments is not restricted. An NU operation of arity 3 is called a {\em majority} operation. We mentioned in the introduction that (modulo UGC) only constraint languages satisfying condition $\mathrm{SD}(\vee)$ can admit robust algorithms with polynomial loss. The condition $\mathrm{SD}(\vee)$ can be expressed in many equivalent ways: for example, as the existence of ternary polymorphisms $d_0,\ldots, d_t$, $t\ge 2$, satisfying the following equations~\cite{Hobby88:structure}: \begin{align} d_0(x,y,z) &= x,\quad d_t(x,y,z) = z, \\ d_i(x,y,x) &= d_{i+1}(x,y,x) \text{ for all even $i<t$}, \label{sdjoin3}\\ d_i(x,y,y) &= d_{i+1}(x,y,y) \text{ for all even $i<t$},\\ d_i(x,x,y) &= d_{i+1}(x,x,y) \text{ for all odd $i<t$}. \end{align} If line (\ref{sdjoin3}) is strengthened to $d_i(x,y,x)=x$ for all $i$, then, for any constraint language, having such polymorphisms would be equivalent to having an NU polymorphism of some arity~\cite{Barto13:NU} (this is true only when constraint languages are assumed to be finite). NU polymorphisms appeared many times in the CSP literature. For example, they characterize the so-called ``bounded strict width'' property~\cite{Feder98:monotone,Jeavons98:consist}, which says, roughly, that, after establishing local consistency in an instance, one can always construct a solution in a greedy way, by picking values for variables in any order so that constraints are not violated. \begin{theorem}\cite{Feder98:monotone,Jeavons98:consist} \label{the:maj} \label{the:nu} Let $\Gamma$ be a constraint language with an NU polymorphism of some arity. There is a polynomial-time algorithm that, given an instance of $\csp(\Gamma)$, finds a satisfying assignment or reports that none exists. \end{theorem} Every relation with an $(n+1)$-ary NU polymorphism is {\em $n$-decomposable} (and in some sense the converse also holds)~\cite{Baker75:chinese-remainder}. We give a formal definition only for the majority case $n=2$. Let $R$ be a $r$-ary ($r\ge 2$) relation. For every $i,j\in\{1,\dots,r\}$, let $\pr_{i,j} R$ be the binary relation $\{(a_i,a_j)\mid (a_1,\dots,a_r)\in R\}$. Then $R$ is called $2$-{\em decomposable} if the following holds: a tuple $(a_1,\dots,a_r)\in D^r$ belongs to $R$ if and only if $(a_i,a_j)\in \pr_{i,j} R$ for every $i,j\in\{1,\dots,r\}$. The {\em dual discriminator} is a majority operation $f$ such that $f(x,y,z)=x$ whenever $x,y,z$ are pairwise distinct. Binary relations preserved by the dual discriminator are known as {\em implicational}~\cite{Bibel88:deductive} or {\em 0/1/all}~\cite{Cooper94:characterising} relations. Every such relation is of one of the four following types: \begin{enumerate} \item $(\{a\}\times D)\cup (D\times\{b\})$ for $a,b\in D$, \item $\{(\pi(a),a)\mid a\in D\}$ where $\pi$ is a permutation on $D$, \item $P\times Q$ where $P,Q\subseteq D$, \item a intersection of a relation of type 1 or 2 with a relation of type 3. \end{enumerate} The relations of the first kind, when $D=\{0,1\}$, are exactly the relations allowed in {\sc 2-Sat}, while the relations of the second kind are precisely the relations allowed in {\sc Unique Games $(|D|)$}. We remark that having such an explicit description of relations having a given polymorphism is rare beyond the Boolean case. \section{Main result} \label{sec:main} \begin{theorem} \label{the:main} Let $\Gamma$ be a constraint language on $D$. \begin{enumerate} \item If $\Gamma$ has a near-unanimity polymorphism then $\CSP\Gamma$ admits a randomized polynomial-time robust algorithm with loss $O(\eps^{1/k})$ for $k = 6|D|^r +7$ where $r$ is the maximal arity of a~relation in $\Gamma$. Moreover, if $\Gamma$ contains only binary relations then one can choose $k=6|D|+7$. \item If $\Gamma$ has the dual discriminator polymorphism then $\CSP\Gamma$ admits a randomized polynomial-time robust algorithm with loss $O(\sqrt{\eps})$. \end{enumerate} \end{theorem} It was stated as an open problem in~\cite{Dalmau13:robust} whether every CSP that admits a robust algorithm with loss $O(\eps^{1/k})$ admits one where $k$ is bounded by an absolute constant (that does not dependent on $D$). In the context of the above theorem, the problem can be made more specific: is dependence of $k$ on $|D|$ in this theorem avoidable or there is a strict hierarchy of possible degrees there? The case of a majority polymorphism is a good starting point when trying to answer this question. As mentioned in the introduction, robust algorithms with polynomial loss and bounded pathwidth duality for CSPs seem to be somehow related, at least in terms of algebraic conditions. The condition $\mathrm{SD}(\vee)$ is the common necessary condition for them, albeit it is conditional on UGC for the former and unconditional for the latter. Having an NU polymorphism is a sufficient condition for both. Another family of problems $\CSP\Gamma$ with bounded pathwidth duality was shown to admit robust algorithms with polynomial loss in~\cite{Dalmau13:robust}, where the parameter $k$ depends on the pathwidth duality bound (and appears in the algebraic description of this family). This family includes languages not having an NU polymorphism of any arity -- see~\cite{Carvalho10:caterpillar,Carvalho11:lattice}. It is unclear how far connections between the two directions go, but consistency notions seem to be the common theme. Returning to the discussion of a possible hierarchy of degrees in polynomial loss in robust algorithms -- there was a similar question about a hierarchy of bounds for pathwidth duality, and the hierarchy was shown to be strict~\cite{Dalmau08:majority}, even in the presence of a majority polymorphism. \section{SDP relaxation}\label{sec:SDP} Associated to every instance $\inst I=(V,{\mathcal C})$ of CSP there is a standard SDP relaxation. It comes in two versions: maximizing the number of satisfied constraints and minimizing the number of unsatisfied constraints. We use the latter. We define it assuming that all constraints are binary, this will be sufficient for our purposes. The SDP has a variable $\mathbf{x}_a$ for every $x\in V$ and $a\in D$. It also contains a special unit vector $\mathbf{v}_0$. The goal is to assign $(|V\|D|)$-dimensional real vectors to its variables minimizing the following objective function: \begin{equation} \sum_{C=((x,y),R)\in {\mathcal C}} w_C\sum_{(a,b)\not\in R} \mathbf{x}_a\mathbf{y}_b \label{sdpobj} \end{equation} subject to: \begin{align} &\mathbf{x}_a\mathbf{y}_b\geq 0 & x,y\in V, a,b\in D \label{sdp1} \\ &\mathbf{x}_a\mathbf{x}_b= 0 & x\in V, a,b\in D, a\neq b \label{sdp2} \\ &\textstyle\sum_{a\in D} \mathbf{x}_a=\mathbf{v}_0 & x\in V \label{sdp3} \\ &\|\mathbf{v}_0\|=1 & \label{sdp4} \end{align} In the intended integral solution, $x=a$ if $\mathbf{x}_a=\mathbf{v}_0$. In the fractional solution, we informally interpret $\|\mathbf{x}_a\|^2$ as the probability of $x=a$ according to the SDP (the constraints of the SDP ensure that $\sum_{a\in D} \|\mathbf{x}_a\|^2 =1$). If $C=((x,y),R)$ is a constraint and $a,b\in D$, one can think of $\mathbf{x}_a\mathbf{y}_b$ as the probability given by the solution of the SDP to the pair $(a,b)$ in $C$. The optimal SDP solution, then, gives as little probability as possible to pairs that are not in the constraint relation. For a constraint $C=((x,y),R)$, conditions (\ref{sdp3}) and (\ref{sdp4}) imply that $\sum_{(a,b)\in R} \mathbf{x}_a\mathbf{y}_b$ is at most $1$. Let $\loss(C)=\sum_{(a,b)\not\in R} \mathbf{x}_a\mathbf{y}_b$. For a subset $A\subseteq D$, let $\mathbf{x}_A=\sum_{a\in A} \mathbf{x}_a$. Note that $\mathbf{x}_D=\mathbf{y}_D(=\mathbf{v}_0)$ for all $x,y\in D$. Let $\mathrm{SDPOpt}(\inst I)$ be the optimum value of (\ref{sdpobj}). It is clear that, for any instance $\inst I$, we have $\mathrm{Opt}(\inst I)\ge \mathrm{SDPOpt}(\inst I)\ge 0$. There are algorithms ~\cite{Vandenberghe96:semidefinite} that, given an SDP instance $\inst I$ and some additive error $\delta>0$, produce in time $\operatorname{\textit{poly}}\, (|\inst I|, \log(1/\delta))$ an output vector solution whose value is at most $\mathrm{SDPOpt}(\inst I)+\delta$. There are several ways to deal with the error $\delta$. In this paper we deal with it by introducing a preprocessing step which will also be needed to argue that the algorithm described in the proof of Theorem~\ref{the:main}(1) runs in polynomial time. {\bf Preprocessing step 1.}\quad Assume that ${\mathcal C}=\{C_1,\dots,C_m\}$ and that $w_{C_1}\ge w_{C_2}\ge \ldots \ge w_{C_m}$. Using the algorithm from Theorem~\ref{the:nu}, find the largest $j$ such that the subinstance $\inst I_j=(V,\{C_1,\dots,C_j\})$ is satisfiable. If the total weight of the constraints in $\inst I_j$ is at least $1-1/m$ then return the assignment $s$ satisfying $\inst I_j$ and stop. \begin{lemma} \label{lem:prep1} \label{le:prep1} Assume that $\inst I$ is $(1-\eps)$-satisfiable. If $\epsilon\leq 1/m^2$ then preprocessing step 1 returns an assignment that $(1-\sqrt{\epsilon})$-satisfies~$\inst I$. \end{lemma} \begin{proof} Assume $\eps\leq 1/m^2$. Let $i$ be maximum with the property that $w_{C_i}>\eps$. It follows that the instance $\inst I_i=(V,\{C_1,\dots,C_i\})$ is satisfiable since the assignment $(1-\eps)$-satisfying $\inst I$ must satisfy every constraint with weight larger than $\eps$. It follows that $i\leq j$ and, hence, the value of the assignment satisfying $\inst I_j$ is at least $1-w_{C_{i+1}}-\cdots -w_{C_m}\geq 1-mw_{C_{i+1}}\geq 1-m\eps\geq 1-\sqrt{\eps}$. \end{proof} If the preprocessing step returns an assignment then we are done. So assume that it did not return an assignment. Then we know that $\epsilon\ge 1/m^2$. We then solve the SDP relaxation with $\delta=1/m^2$ obtaining a solution with objective value at most $2\epsilon$ which is good enough for our purposes. \section{Overview of the proof of Theorem \ref{the:main}(1)} \label{sec:overview1} We assume throughout that $\Gamma$ has a near-unanimity polymorphism of arity $n+1$ ($n\ge 2$). It is sufficient to prove Theorem \ref{the:main}(1) for the case when $\Gamma$ consists of binary relations and $k=6|D|+7$. The rest will follow by Proposition~4.1 of~\cite{Barto16:robust} (see also Theorem~24 in~\cite{Barto17:poly}), which shows how to reduce the general case to constraint languages consisting of unary and binary relations in such a way that the domain size increases from $|D|$ to $|D|^r$ where $r$ is the maximal arity of a~relation in $\Gamma$. Note that every unary constraint $(x,R)$ can be replaced by the binary constraint $((x,x),R')$ where $R'=\{(a,a)\mid a\in R\}$. Throughout the rest of this section, let $\inst I=(V,{\mathcal C})$ be a $(1-\eps)$-satisfiable instance of $\CSP\Gamma$. \subsection{Patterns and realizations}\label{sect:pattern} A~\emph{pattern in $\inst I$} is defined as a~directed multigraph $p$ whose vertices are labeled by variables of $\inst I$ and edges are labeled by constraints of $\inst I$ in such a way that the beginning of an edge labeled by $((x,y),R)$ is labeled by $x$ and the end by $y$. Two of the vertices in $p$ can be distinguished as the \emph{beginning} and the \emph{end} of~$p$. If these two vertices are labeled by variables $x$ and $y$, respectively, then we say that $p$ is a~pattern is from $x$ to $y$. For two patterns $p$ and $q$ such that the end of $p$ and the beginning of $q$ are labeled by the same variable, we define $p+q$ to be the pattern which is obtained from the disjoint union of $p$ and $q$ by identifying the end of $p$ with the beginning of $q$ and choosing the beginning of $p+q$ to be the beginning of $p$ and the end of $p+q$ to be the end of $q$. We also define $jp$ to be $p + \dots + p$ where $p$ appears $j$ times. A~pattern is said to be a~\emph{path pattern} if the underlying graph is an~oriented path with the beginning and the end being the two end vertices of the path, and is said to be an~\emph{$n$-tree pattern} if the underlying graph is an orientation of a tree with at most $n$ leaves, and both the beginning and the end are leaves. A~\emph{path of $n$-trees pattern} is then any pattern of the form $t_1+\dots+t_j$ for some $n$-tree patterns $t_1,\dots,t_j$. A~\emph{realization of a~pattern} $p$ is a~mapping $r$ from the set of vertices of $p$ to $D$ such that if $(v_x,v_y)$ is an edge labeled by $((x,y),R)$ then $(r(v_x),r(v_y)) \in R$. Note that $r$ does not have to map different vertices of $p$ labeled with same variable to the same element in $D$. A~\emph{propagation} of a~set $A\subseteq D$ along a pattern $p$ whose beginning vertex is $b$ and ending vertex is $e$ is defined as follows. For $A\subseteq D$, define $A + p=\{r(e) \mid \text{ $r$ is a realization of $p$ with $r(b) \in A$}\}$. Also for a~binary relation $R$ we put $A+R=\{b\mid (a,b)\in R \mbox{ and } a\in A\}$. Observe that we have $(A+p)+q = A+(p+q)$. Further, assume that we have non-empty sets $D_x^\ell$ where $1\leq \ell\leq |D|+1$ and $x$ runs through all variables in an instance $\inst I$. Let $p$ be a pattern in $\inst I$ with beginning $b$ and end $e$. We call a realization $r$ of $p$ an \emph{$\ell$-realization} (with respect to the family $\{D_x^\ell\}$) if, for any vertex $v$ of $p$ labeled by a~variable $x$, we have $r(v)\in D_x^{\ell+1}$. For $A\subseteq D$, define $A +^\ell p=\{r(e) \mid r \text{ is an $\ell$-realization of $p$ with $r(b) \in A$}\}$. Also, for a constraint $((x,y),R)$ or $((y,x),R^{-1})$ and sets $A,B\subseteq D$, we write $B=A+^\ell (x,R,y)$ if $B=\{b\in D_y^{\ell+1} \mid (a,b)\in R \mbox{ for some } a\in A\cap D_x^{\ell+1}\}$. \subsection{The consistency notion} Recall that we assume that $\Gamma$ contains only binary relations. Before we formally introduce the new consistency notion, which is the key to our result, as we explained in the introduction, we give an example of a similar simpler condition. We mentioned before that {\sc 2-Sat} is a special case of a CSP that admits an NU polymorphism (actually, the only majority operation on $\{0,1\}$). There is a textbook consistency condition characterizing satisfiable {\sc 2-Sat} instances, which can be expressed in our notation as follows: for each variable $x$ in a {\sc 2-Sat} instance $\inst I$, there is a value $a_x$ such that, for any path pattern $p$ in $\inst I$ from $x$ to $x$, we have $a_x\in \{a_x\}+p$. Let $\inst I$ be an instance of $\CSP\Gamma$ over a set $V$ of variables. We say that $\inst I$ satisfies condition \IPQ{n} if the following holds: \begin{description} \item{\IPQ{n}} For every $y\in V$, there exist non-empty sets $D_y^1 \subseteq \ldots \subseteq D_y^{|D|} \subseteq D_y^{|D|+1}=D$ such that for any $x\in V$, any $\ell \leq |D|$, any $a\in D_x^\ell$, and any two patterns $p,q$ which are paths of $n$-trees in $\inst I$ from $x$ to $x$, there exists $j$ such that \[ a \in \{a\} +^{\ell} (j(p+q) + p). \] \end{description} Note that $+$ between $p$ and $q$ is the pattern addition and thus independent of $\ell$. Note also that $a$ in the above condition belongs to $D_x^\ell$, while propagation is performed by using $\ell$-realizations, i.e., inside sets $D_y^{\ell+1}$. The following theorem states that this consistency notion satisfies the requirements of Step~3 of the general scheme (for designing robust approximation algorithms) discussed in the introduction. \begin{theorem}\label{thm:nicelevels} Let $\Gamma$ be a constraint language containing only binary relations such that $\Gamma$ has an $(n+1)$-ary NU polymorphism. If an instance $\inst I$ of $\CSP\Gamma$ satisfies \IPQ{n}, then $\inst I$ is satisfiable. \end{theorem} \subsection{The algorithm} Let $k = 6|D| + 7$. We provide an algorithm which, given a~$(1-\epsilon)$-satisfiable instance $\inst I$ of $\CSP\Gamma$, removes $O(\epsilon^{1/k})$ constraints from it to obtain a~subinstance $\inst I'$ satisfying condition (IPQ)$_n$. It then follows from Theorem~\ref{thm:nicelevels} that $\inst I'$ is satisfiable, and we can find a satisfying assignment by Theorem~\ref{the:nu}. \subsubsection{More preprocessing} \label{sec:preprocessing} By Lemma \ref{lem:prep1} we can assume that $\epsilon\geq 1/m^2$. We solve the SDP relaxation with error $\delta=1/m^2$ and obtain an solution $\{\vec x_a\}$ $(x\in V,a\in D)$ whose objective value $\epsilon'$ is at most $2\epsilon$. Let us define $\alpha$ to be $\max\{\epsilon',1/m^2\}$. It is clear that $\alpha=O(\epsilon)$. Furthermore, this gives us that $1/\alpha\le m^2$. This will be needed to argue that the main part of the algorithm runs in polynomial time. Let $\kappa=1/k$ (we will often use $\kappa$ to avoid overloading formulas). {\bf Preprocessing step 2.}\quad For each $x\in V$ and $1\leq \ell\leq |D|+1$, compute sets $D_x^{\ell}\subseteq D$ as follows. Set $D_x^{|D|+1}=D$ and, for $1\leq \ell\leq |D|$, set $D_x^{\ell}=\{a\in D\mid \|\vec x_a\|\ge r_{x,\ell}\}$ where $r_{x,\ell}$ is the smallest number of the form $r=\alpha^{3\ell \kappa}(2|D|)^{i/2}$, $i\ge 0$ integer, with $\{b\in D\mid r(2|D|)^{-1/2}\le \|\vec x_b\|< r\}=\emptyset$. It is easy to check that $r_{x,\ell}$ is obtained with $i\le |D|$. It is clear that the sets $D_x^{\ell}\subseteq D$, $x\in V$, $1\leq \ell\leq |D|$, can be computed in polynomial time. The sets $D_x^\ell$ are chosen such that $D_x^\ell$ contains relatively ``heavy'' elements ($a$'s such that $\|\vec x_a\|^2$ is large). The thresholds are chosen so that there is a~big gap (at least by a~factor of $2|D|$) between ``heaviness'' of an element in $D_x^\ell$ and outside. \subsubsection{Main part} \label{sec:algorithm} Given the preprocessing is done, we have that $1/\alpha \leq m^2$, and we precomputed sets $D_x^\ell$ for all $x\in V$ and $1\leq \ell \leq |D|+1$. The description below uses the number $n$, where $n+1$ is the arity of the NU polymorphism of $\Gamma$. \textbf{Step 0.}\quad Remove every constraint $C$ with $\loss(C) > \alpha^{1-\kappa}$. \textbf{Step 1.}\quad For every $ 1 \leq \ell \leq |D| $ do the following. Pick a~value $r_\ell \in (0,\alpha^{(6\ell +4)\kappa})$ uniformly at random. Here we need some notation: for $x,y \in V$ and $A,B\subseteq D$, we write $\vec x_A \preceq^\ell \vec y_B$ to indicate that there is \textbf{no} integer $j$ such that \( \| \vec y_B \|^2 < r_\ell +j\alpha^{(6\ell +4)\kappa} \leq \| \vec x_A \|^2. \) Then, remove all constraints $((x,y),R)$ such that there are sets $A,B \subseteq D$ with $B = A +^\ell (x,R,y)$ and $\vec x_A \not\preceq^\ell \vec y_B$, or with $B = A +^\ell (y,R^{-1},x)$ and $\vec y_A \not\preceq^\ell \vec x_B$. \textbf{Step 2.}\quad For every $1 \leq \ell \leq |D|$ do the following. Let $m_0 = \lfloor \alpha^{-2\kappa} \rfloor$. Pick a~value $s_\ell \in \{0,\dots, m_0-1\}$ uniformly at random. We define $\vec x_A \preceq^{\ell}_w \vec y_B$ to mean that there is \textbf{no} integer $j$ such that \( \|\vec y_B\|^2 < r_\ell + (s_\ell +jm_0)\alpha^{(6\ell +4)\kappa} \leq \|\vec x_A\|^2. \) Obviously, if $\vec x_A \preceq^{\ell} \vec y_B$ then $\vec x_A \preceq^{\ell}_w \vec y_B$. Now, if $A\subseteq B \subseteq D_x^{\ell+1}$ are such that $\|\vec x_B - \vec x_A\|^2 \leq (2n-3)\alpha^{(6\ell +4)\kappa}$ and $\vec x_B \not\preceq^{\ell}_w \vec x_A$, then remove all the constraints in which $x$ participates. \textbf{Step 3.}\quad For every $1\leq \ell \leq |D|$ do the following. Pick $m_\ell = \lceil \alpha ^ {-(3\ell +1)\kappa} \rceil$ unit vectors independently uniformly at random. For $x,y\in V$ and $A,B \subseteq D$, say that $\vec x_A$ and $\vec y_B$ are \emph{cut} by a vector $\vec u$ if the signs of $\vec u \cdot (\vec x_A - \vec x_{D\setminus A})$ and $\vec u \cdot (\vec y_B - \vec y_{D\setminus B})$ differ. Furthermore, we say that $\vec x_A$ and $\vec y_B$ are $\ell$-cut if there are cut by at least one of the chosen $m_\ell$ vectors. For every variable $x$, if there exist subsets $A,B \subseteq D$ such that $A \cap D_x^\ell \neq B \cap D_x^\ell$ and the vectors $\vec x_A$ and $\vec x_B$ are not $\ell$-cut, then remove all the constraints in which $x$ participates. \textbf{Step 4.}\quad For every $1 \leq \ell \leq |D|$, remove every constraint $((x,y),R)$ such that there are sets $A,B \subseteq D$ with $B = A +^\ell (x,R,y)$, and $\vec x_A$ and $\vec y_B$ are $\ell$-cut, or with $B = A +^\ell (y,R^{-1},x)$, and $\vec y_A$ and $\vec x_B$ are $\ell$-cut. \textbf{Step 5.}\quad For every $1 \leq \ell \leq |D|$ do the following. For every variable $x$, If $A,B\subseteq D_x^{\ell +1}$ such that $\| \vec x_B - \vec x_A \|^2 \leq (2n-3)\alpha^{(6\ell+4)\kappa}$ and $\vec x_A$ and $\vec x_B$ are $\ell$-cut, remove all constraints in which $x$ participates. \textbf{Step 6.}\quad By Proposition~\ref{prop:consistence} and Theorem~\ref{thm:nicelevels}, the remaining instance $\inst I'$ is satisfiable. Use the algorithm given by Theorem \ref{the:nu} to find a~satisfying assignment for $\inst I'$. Assign all variables in $\inst I$ that do not appear in $\inst I'$ arbitrarily and return the obtained assignment for $\inst I$. \bigskip Note that we chose to define the cut condition based on $\mathbf{x}_A-\mathbf{x}_{D\setminus A}$, rather than on $ \mathbf{x}_A$, because the former choice has the advantage that $\|\mathbf{x}_A-\mathbf{x}_{D\setminus A}\|=1$, which helps in some calculations. In Step~0 we remove constraints such that, according to the SDP solution, have a high probability to be violated. Intuitively, steps 1 and 2 ensure that the loss in $\|\vec x_A\|$ after propagating $A$ by a~path of $n$-trees is not too big. This is achieved first by ensuring that by following a~path we do not lose too much (step~1) which also gives a~bound on how much we can lose by following an~$n$-tree pattern (see Lemma \ref{lem:tree-loss}). Together with the removal of constraints in step 2, this guarantees that following a~path of $n$-trees we do not lose too much. This ensures that $\{a\} +^\ell (j(p+q) + p)$ is non-vanishing as $j$ increases. Steps 3--5 ensure that if $A$ and $B$ are connected by paths of $n$-trees in both directions (i.e. $\vec x_A = \vec x_B +^{\ell} p_1$ and $\vec x_B = \vec x_A +^{\ell} p_2$), then $\vec x_A$ and $\vec x_B$ do not differ too much (i.e. $A \cap D_x^\ell = B \cap D_x^\ell$). This is achieved by separating the space into cones by cutting it using the $m_\ell$ chosen vectors, removing the variables which have two different sets that are not $\ell$-cut (step 3), and then ensuring that if we follow an edge (step 4), or if we drop elements that do not extend to an~$n$-tree (step 5) we do not cross a~border to another cone. This gives us both that the sequence $A_j = \{a\} +^\ell (j(p+q) +p)$ stabilizes and that, after it stabilizes, $A_j$ contains $a$. This provides condition \IPQ{n} for the remaining instance $\inst I'$. The algorithm runs in polynomial time. Since $D$ is fixed, it is clear that the steps 0--2 can be performed in polynomial time. For steps 3--5, we also need that $m_\ell$ is bounded by a polynomial in $m$, which holds because $\alpha \geq 1/m^2$. The correctness of the algorithm is given by (Theorem~\ref{thm:nicelevels} and) the two following propositions whose proof can be found in Section \ref{sec:correctness-of-algorithm}. These propositions show that our new consistency notion satisfies the requirements of Step~2 of the general scheme for designing robust approximation algorithms discussed in the introduction. \begin{proposition} \label{prop:expected-lost-weight} The expected total weight of constraints removed by the algorithm is $O(\alpha^\kappa)$. \end{proposition} \begin{proposition} \label{prop:consistence} The instance $\inst I'$ obtained after steps 0--5 satisfies the condition \IPQ{n} (with the sets $D_x^\ell$ computed by preprocessing step 2 in Section \ref{sec:preprocessing}). \end{proposition} \section{Overview of the proof of Theorem~\ref{the:main}(2)}\label{sec:overview-thm2} Since the~dual discriminator is a~majority operation, every relation in $\Gamma$ is 2-decomposable. Therefore, it follows, e.g.\ from Lemma 3.2 in~\cite{Dalmau13:robust}, that to prove that $\CSP\Gamma$ admits a robust algorithm with loss $O(\sqrt\eps)$, it suffices to prove this for the case when $\Gamma$ consists of all unary and binary relations preserved by the dual discriminator. Such binary constraints are of one of the four kinds described in Section~\ref{sec:algebra}. Using this description, it follows from Lemma~3.2 of~\cite{Dalmau13:robust} that it suffices to consider the following three types of constraints: \begin{enumerate} \item Disjunction constraints of the form $x= a \vee y = b$, where $a,b\in D$; \item Unique game (UG) constraints of the form $x = \pi(y)$, where $\pi$ is any permutation on~$D$; \item Unary constraints of the form $x \in P$, where $P$ is an arbitrary non-empty subset of $D$. \end{enumerate} We present an algorithm that, given a $(1-\eps)$-satisfiable instance $\inst I=(V,{\mathcal C})$ of the problem, finds a solution satisfying constraints with expected total weight $1- O(\sqrt{\eps\log{|D|}})$ (the hidden constant in the $O$-notation does not depend on $\eps$ and $|D|$). We now give an informal and somewhat imprecise sketch of the algorithm and its analysis. We present details in Section~\ref{sec:thm2}. We use the SDP relaxation from Section~\ref{sec:SDP}. Let us call the value $\|\mathbf{x}_a\|^2$ the SDP weight of the value $a$ for variable $x$. \subsubsection*{Variable Partitioning Step} The algorithm first solves the SDP relaxation. Then, it partitions all variables into three groups $\VV{0}$, $\VV{1}$, and $\VV{2}$ using a threshold rounding algorithm with a random threshold. If most of the SDP weight for $x$ is concentrated on one value $a\in D$, then the algorithm puts $x$ in the set $\VV{0}$ and assigns $x$ the value $a$. If most of the SDP weight for $x$ is concentrated on two values $a,b\in D$, then the algorithm puts $x$ in the set $\VV{1}$ and restricts the domain of $x$ to the set $D_x = \{a,b\}$ (thus we guarantee that the algorithm will eventually assign one of the values $a$ or $b$ to $x$). Finally, if the SDP weight for $x$ is spread among 3 or more values, then we put $x$ in the set $\VV{2}$; we do not restrict the domain for such $x$. After we assign values to $x\in \VV{0}$ and restrict the domain of $x\in \VV{1}$ to $D_x$, some constraints are guaranteed to be satisfied (say, the constraint $(x=a)\vee (y=b)$ is satisfied if we assign $x$ the value $a$ and the constraint $x\in P$ is satisfied if $D_x\subseteq P$). Denote the set of such constraints by ${\cal C}_s$ and let ${\mathcal C}'={\mathcal C}\setminus{\mathcal C}_s$. We then identify a set ${\cal C}_v\subseteq {\mathcal C}'$ of constraints that we conservatively label as violated. This set includes all constraints in ${\mathcal C}'$ except those belonging to one of the following 4 groups: \begin{enumerate} \item disjunction constraints $(x = a) \vee (y = b)$ with $x, y \in {\cal V}_1$ and $a\in D_x$, $b\in D_y$; \item UG constraints $x = \pi (y)$ with $x, y \in {\cal V}_1$ and $D_x = \pi(D_y)$; \item UG constraints $x = \pi (y)$ with $x, y \in {\cal V}_2$; \item unary constraints $x \in P$ with $x\in{\cal V}_2$. \end{enumerate} Our construction of sets $\VV{0}$, $\VV{1}$, and $\VV{2}$, which is based on randomized threshold rounding, ensures that the expected total weight of constraints in ${\cal C}_v$ is $O(\eps)$ (see Lemma~\ref{lem:preproc-violated}). The constraints from the 4 groups above naturally form two disjoint sub-instances of $\inst I$: $\inst I_1$ (groups 1 and 2) on the set of variables $\VV{1}$, and $\inst I_2$ (groups 3 and 4) on $\VV{2}$. We treat these instances independently as described below. \subsubsection*{Solving Instance $\inst I_1$} The instance $\inst I_1$ with the domain of each $x$ restricted to $D_x$ is effectively an instance of Boolean 2-CSP (i.e. each variable has a 2-element domain and all constraints are binary). A robust algorithm with quadratic loss for this problem was given by Charikar et al.~\cite{Charikar09:near}. This algorithm finds a solution violating an $O(\sqrt{\varepsilon})$ fraction of all constraints if the optimal solution violates at most $\varepsilon$ fraction of all constraints or $\mathrm{SDPOpt}(\inst I_1)\leq \eps$. However, we cannot apply this algorithm to the instance $\inst I_1$ as is. The problem is that the weight of violated constraints in the optimal solution for $\inst I_1$ may be greater than $\omega(\varepsilon)$. Note that the unknown optimal solution for the original instance $\inst I$ may assign values to variables $x$ outside of the restricted domain $D_x$, and hence it is not a feasible solution for $\inst I_1$. Furthermore, we do not have a feasible SDP solution for the instance $\inst I_1$, since the original SDP solution (restricted to the variables in $\VV{1}$) is not a feasible solution for the Boolean 2-CSP problem (because $\sum_{a\in D_x}\mathbf{x}_a$ is not necessarily equal to $\mathbf{v}_0$ and, consequently, $\sum_{a\in D_x}\|\mathbf{x}_a\|^2$ may be less than 1). Thus, our algorithm first transforms the SDP solution to obtain a feasible solution for $\inst I_1$. To this end, it partitions the set of vectors $\{\mathbf{x}_a: x\in \VV{1}, a\in D_x\}$ into two sets $H$ and $\bar{H}$ using a modification of the hyperplane rounding algorithm by Goemans and Williamson~\cite{Goemans95:improved}. In this partitioning, for every variable $x$, one of the two vectors $\{\mathbf{x}_a: a\in D_x\}$ belongs to $H$ and the other belongs to $\bar H$. Label the elements of each $D_x$ as $\alpha_x$ and $\beta_x$ so that so that $\mathbf{x}_{\alpha_x}$ is the vector in $H$ and $\mathbf{x}_{\beta_x}$ is the vector in $\bar H$. For every $x$, we define two new vectors $\mathbf{\tilde{x}}_{\alpha_x} = \mathbf{x}_{\alpha_x}$ and $\mathbf{\tilde{x}}_{\beta_x} = \mathbf{v}_0-\mathbf{x}_{\alpha_x}$. It is not hard to verify that the set of vectors $\{\mathbf{\tilde{x}}_{a}: x\in\VV{1}, a\in D_x\}$ forms a feasible SDP solution for the instance $\inst I_1$. We show that for each disjunction constraint $C$ in the instance $\inst I_1$, the cost of $C$ in the new SDP solution is not greater than the cost of $C$ in the original SDP solution (see Lemma~\ref{lem:new-SDP-feasible}). The same is true for all but $O(\sqrt{\eps})$ fraction of UG constraints. Thus, after removing UG constraints for which the SDP value has increased, we get an SDP solution of cost $O(\varepsilon)$. Using the algorithm~\cite{Charikar09:near} for Boolean 2-CSP, we obtain a solution for $\inst I_1$ that violates constraints of total weight at most $O(\sqrt{\eps})$. \subsubsection*{Solving Instance $\inst I_2$} The instance $\inst I_2$ may contain only unary and UG constraints as all disjunction constraints are removed from $\inst I_2$ at the variable partitioning step. We run the approximation algorithm for Unique Games by Charikar et al.~\cite{Charikar06:near} on $\inst I_2$ using the original SDP solution restricted to vectors $\{\mathbf{x}_a: x\in \VV{2}, a\in D\}$. This is a valid SDP relaxation because in the instance $\inst I_2$, unlike the instance $\inst I_1$, we do not restrict the domain of variables $x$ to $D_x$. The cost of this SDP solution is at most $\eps$. As shown in~\cite{Charikar06:near}, the weight of constraints violated by the algorithm~\cite{Charikar06:near} is at most $O(\sqrt{\eps\log |D|})$. We get the solution for $\inst I$ by combining solutions for $\inst I_1$ and $\inst I_2$, and assigning values chosen at the variable partitioning step to the variables from the set $\VV{0}$. \section{Proof of Theorem~\ref{thm:nicelevels}}\label{sect:proofof3} In this section we prove Theorem~\ref{thm:nicelevels}. The proof will use constraint languages with relations of arity greater than two. In order to talk about such instances we need to extend the definition of a pattern. Note that patterns~(in the sense of Section~\ref{sect:pattern}) are instances~ (with some added structure) and the realizations of patterns are solutions. We use the pattern/instance and solution/realization duality to generalize the notion of a pattern. Moreover we often treat patterns as instances and~(whenever it makes sense) instances as patterns. We will often talk about path/tree instances; they are defined using the {\em incidence multigraph}. The incidence multigraph of an instance $\inst J$ is bipartite, its the vertex set consists of variables and constraints of $\inst J$ (which form the two parts), and if a variable $x$ appears $j$ times in a constraint $C$ then the vertices corresponding to $x$ and $C$ have $j$ edges between them. An instance is {\em connected} if its incidence multigraph is connected; an~instance is a~\emph{tree instance} if it is connected and its incidence multigraph has no multiple edges and no cycles. A~\emph{leaf variable} in a tree instance is a~variable which corresponds to a~leaf in the incidence multigraph, and we say that two variables are \emph{neighbours} if they appear together in a~scope of some constraint (i.e., the corresponding vertices are connected by a~path of length 2 in the incidence multigraph). Note that the incidence multigraph of a path pattern in a binary instance~(treated as an instance, as described in the first paragraph of this section) is a path, and of an $n$-tree pattern is a tree with $n$ leaves. The next definition captures, among other things, the connection between the pattern~(treated as an instance) and the instance in which the pattern is defined. Let $\inst J_1$ and $\inst J_2$ be two instances over the same constraint language. An~\emph{(instance) homomorphism} $e\colon \inst J_1 \to \inst J_2$ is a~mapping that maps each variable of $\inst J_1$ to a variable of $\inst J_2$ and each constraint of $\inst J_1$ to constraint of $\inst J_2$ in such a~way that every constraint $((y_1,\dots,y_k),R)$ in $\inst J_1$ is mapped to $((e(y_1),\dots,e(y_k)),R)$. Using these new notions, a path pattern in an instance $\inst I$ (see the definition in Section~\ref{sect:pattern}) can alternatively be defined as an instance, with beginning and end chosen among the leaf variables, whose incidence graph is a path from beginning to end, together with a homomorphism into $\inst I$. Similarly we define a {\em path pattern} in a (not necessarily binary) instance $\inst I$ as an instance $\inst J$, with chosen beginning/end leaf variables, whose incidence graph, after removing all the other vertices of degree one, is a path from beginning to end, together with a homomorphism $e\colon \inst J\rightarrow\inst I$. Addition of path patterns and propagation are defined in an analogous way as for patterns with binary constraints~(see Section~\ref{sect:pattern}). For a $k$-ary relation $R$, let $\pr_i(R)=\{a_i\mid (a_1,\ldots,a_i,\ldots,a_k)\in R\}$. A CSP instance $\inst J$ is called {\em arc-consistent} in sets $D_x$~($x$ ranges over variables of $\inst J$) if, for any variable $x$ and any constraint $((x_1,\ldots,x_k),R)$ in $\inst J$, if $x_i=x$ then $\pr_i(R)=D_x$. We say that a~CSP instance $\inst J$ satisfies condition {\em (PQ)} in sets $D_x$ if \begin{enumerate} \item $\inst J$ is arc-consistent in these sets and \label{cond:pqac} \item for any variable $x$, \label{cond:pqpq} any path patterns $p,q$ from $x$ to $x$, and any $a\in D_x$ there exists $j$ such that \( a\in \{a\} + (j(p+q) + p) \). \end{enumerate} Note that if the instance $\inst J$ is binary then (PQ) implies \IPQ{n} for all $n$~ (setting $D_x^i=D$ if $i=|D|+1$ and $D_x^i=D_x$ if $i<|D|+1$). The following fact, a special case of Theorem A.2 in~\cite{Kozik16:circles}, provides solutions for (PQ) instances. \begin{theorem} \label{thm:nice} If $\Gamma'$ is a~constraint language with a near-unanimity polymorphism, then every instance of $\CSP{\Gamma'}$ satisfying condition (PQ) is satisfiable. \end{theorem} Finally, a standard algebraic notion has not been defined yet: having fixed $\Gamma$ over a set $D$, a subset $A\subseteq D$ is a {\em subuniverse} if, for any polymorphism $g$ of $\Gamma$, we have $g(a_1,a_2,\ldots)\in A$ whenever $a_1,a_2,\ldots \in A$. For any $S\subseteq D$, the subuniverse {\em generated by} $S$ is defined as \[ \{g(a_1,\ldots,a_r)\mid r\ge 1, a_1,\ldots,a_r\in S, \text{$g$ is an $r$-ary polymorphism of $\Gamma$} \} \] \subsection{Into the proof} We begin the proof of Theorem~\ref{thm:nicelevels}. We fix a binary language $\Gamma$ compatible with a $(n+1)$-ary NU polymorphism and an instance $\inst I$ of in $\CSP\Gamma$ which satisfies $\IPQ{n}$ with sets $D_x^\ell$. Note that we can assume that all $D_x^\ell$'s are subuniverses. If this is not the case, we replace each $D_x^\ell$ with the subuniverse generated by it. It is easy to check that~(after the change) the instance $\inst I$ still satisfies $\IPQ{n}$ with such enlarged $D_x^\ell$'s. For each variable $x$, choose and fix an arbitrary index $i$ such that $D_x^i=D_x^{i+1}$ and call it the {\em level} of $x$. Note that each variable has a level~ (since the sets $D_x^\ell$ are non-empty and $\ell$ ranges from 1 to $|D|+1$). Let $V^i$ denote the set of variables of level $i$ and $V^{< i}, V^{\le i},\dots$ be defined in the natural way. Our proof of Theorem~\ref{thm:nicelevels} will proceed by applying Theorem~\ref{thm:nice} to $\inst I$ restricted to $V^1$, then to $V^2$ and so on. However, in order to obtain compatible solutions, we will add constraints to the restricted instances. \subsection{The instances in levels} Let $\inst I^i$~(for $i\leq |D|$) be an instance such that: \begin{enumerate} \item $V^i$ is the set of variables of $\inst I^i$; \item $\inst I^i$ contains, for every every $n$-tree pattern $t$ of $\inst I$, the constraint $((x_1,\dots,x_k),R)$ defined in the following way: \label{enum:newconstraints} let $v_1,\dotsc,v_k$ be all the vertices of $t$ labeled by variables from $V^i$, then $x_1,\dots,x_k$ are the labels of $v_1,\dots,v_k$ respectively and \[ R = \{ (r(v_1),\dots,r(v_k)) \mid r \text{ is a $i$-realization of $t$ (i.e. inside sets $D_x^{i+1}$)} \}. \] \end{enumerate} This definition has a number of immediate consequences: First, every binary constraint between two variables from $V^i$ is present in $\inst I^i$~ (as it defines a two-element $n$-tree). Second, note that if for some $n$-tree contains a vertex $v_j$ in $V^i$ which is not a leaf then by splitting the tree $t$ at $v_j$~(with $v_j$ included in both parts) we obtain two trees defining constraints which together are equivalent to the constraint defined by $t$. This implies that by including only the constraints defined by $n$-trees $t$ such that only the leaves can be from $V^i$, we obtain an equivalent~(i.e. having the same set of solutions) instance. Throughout most of the proof we will be working with such a restricted instance. In this instance the arity of constraints is bounded by $n$. Since the arity of a constraint in $\inst I^i$ is bounded and the size of the universe is fixed, $\inst I^i$ is a finite instance, even though some constraints in it can be defined via infinitely many $n$-tree patterns. It is easy to see that all the relations in the constraints are preserved by all the polymorphisms of $\Gamma$. The instance $\inst I^i$ is arc-consistent with sets $D_x^i(=D_x^{i+1})$: Let $((x_1,\dotsc,x_k),R)$ be a constraint defined by $v_1,\dotsc, v_k$ in $t$ and let $a\in D_{x_j}^i$. By \IPQ{n} there is a realization of $t$ in $D_x^{i+1}$ mapping $v_j$ to $a$ and thus $D_{x_j}^i\subseteq \pr_j R$. On the other hand, as $D_{x_j}^i=D_{x_j}^{i+1}$ and every tuple in $R$ comes from a realization inside the sets $D_x^{i+1}$'s, we get $\pr_j R\subseteq D_{x_j}^i$. Next we show that $\inst I^i$ has property (PQ). Part \ref{cond:pqac} of the definition was established in the paragraph above. For part \ref{cond:pqpq}, let $p$ and $q$ to be arbitrary path patterns from $x$ to $x$ in $\inst I^i$. Define $p'$ and $q'$ to be the paths of trees in $\inst I$ obtained, from $p$ and $q$, respectively, by replacing (in the natural way) each constraint in $p$ and $q$ with the~tree that defines it (we use the fact that each constraint is defined by leaves of a tree). We apply property \IPQ{n} for $\inst I$ with $\ell=i$ and patterns $p'$ and $q'$ to get that, for any $x\in V^i$ and any $a\in D_x^i$, there is a number $j$ such that $a\in \{a\}+^i(j(p'+q')+p')$. The property (PQ) follows immediately. Since $\inst I^i$ has the property (PQ) then, by Theorem~\ref{thm:nice}, it has a solution. The solution to $\inst I$ will be obtained by taking the union of appropriately chosen solutions to $\inst I^1,\dotsc, \inst I^{|D|}$. \subsection{Invariant of the iterative construction} A global solution, denoted $\sol\colon V\to D$, is constructed in steps. At the start, we define it for the variables in $V^1$ by choosing an arbitrary solution to $\inst I^1$. In step $i$ we extend the definition of $\sol$ from $V^{<i}$ to $V^{\le i}$, using a carefully chosen solution to $\inst I^i$. Our construction will maintain the following condition: \begin{description} \item{($E_i$)} every $n$-tree pattern in $\inst I$ has a realization inside the sets $D_x^{i+1}$ which agrees with $\sol$ on $V^{\le i}$. \end{description} Note that, after the first step, the condition $(E_1)$ is guaranteed by the constraints of $\inst I^1$. Assume that we are in step $i$: we have already defined $\sol$ on $V^{<i}$ and condition $(E_{i-1})$ holds. Our goal is to extend $\sol$ by a~solution of $\inst I^i$ in such a~way that $(E_{i})$ holds. The remainder of Section~\ref{sect:proofof3} is devoted to proving that such a solution exists. Once we accomplish that, we are done with the proof: Condition $(E_{i})$ implies that $\sol$ is defined on $V^{\le i}$, and for every constraint $((x,y),R)$ between $x,y\in V^{\le i}$ the pattern from $x$ to $y$ containing a~single edge labeled by $((x,y),R)$ is an~$n$-tree. This implies that $\sol$ satisfies $((x,y),R)$ i.e. it is a solution on $V^{\le i}$. After establishing $(E_{|D|})$ we obtain a solution to $\inst I$. \subsection{Restricting \texorpdfstring{$\inst I^i$}{the instance in level i}} We begin by defining a new instance $\inst K^i$: it is defined almost identically to $\inst I^i$, but in part~\ref{enum:newconstraints} of the definition we require that the realization $r$ sends vertices from $V^{<i}$ according to $\sol$. As in the case of $\inst I^i$ we can assume that all the constraints are defined by leaves of the tree. Thus every $n$-tree pattern with no internal vertices in $V^{i}$ defines one constraint in $\inst I^i$ and another in $\inst K^i$. Just like $\inst I^i$, the instance $\inst K^i$ is finite. Note that we yet need to establish that constraints of $\inst K^i$ are non-empty, but the following claim, where $f$ is the fixed $(n+1)$-ary near unanimity polymorphism, holds independently. \begin{claim}\label{claim:absorbing} Let $((x_1,\dotsc,x_k),R)$ and $((x_1,\dotsc,x_k),R')$ be constraints defined by the same tree $t$ in $\inst I^i$ and $\inst K^i$~(respectively). If $\tuple a_1,\dotsc,\tuple a_{n+1}\in R'$, $\tuple a\in R$, and $j\in \{1,\dots,n+1 \}$ then $f(\tuple a_1,\dotsc,\tuple a_{j-1},\tuple a,\tuple a_{j+1},\dotsc,\tuple a_{n+1})$~ belongs to $R'$. \end{claim} \begin{proof} Let $r_i$ be a realization of $t$ defining $\tuple a^i$; this realization sends all the vertices of $t$ labeled by variables from $V^{<i}$ according to $\sol$. Let $r$ be a realization of $t$ defining $\tuple a$. Fix a function, from vertices of $t$ into $D$, sending vertex $v$ to \[ f(r_1(v),\dotsc,r(v),\dotsc,r_{n+1}(v) \] (where $r(v)$ is in position $j$). This is clearly a realization, and if $v$ is labeled by $x\in V^{<i}$ it sends $v$ according to $\sol$~ (since $f$ is a near-unanimity operation). The new realization witnesses that $f(\tuple a_1,\dotsc,\tuple a_{j-1},\tuple a,\tuple a_{j+1},\dotsc,\tuple a_{n+1})$ belongs to $R'$. \end{proof} In order to proceed we need to show that the instance $\inst K^i$ contains a non-empty, arc-consistent subinstance, i.e. an arc-consistent instance (in some non-empty sets $D_x$) obtained from $\inst K^i$ by restricting every constraint in it so that each coordinate can take value only in the appropriate set $D_x$. A proof of this claim is the subject of the next section. \subsection{Arc-consistent subinstance of~$\inst K^i$} In order to proceed with the proof we need an additional definition. Let $e\colon \inst J_1 \to \inst J_2$ be an instance homomorphism. If for any variable $y$ of $\inst J_1$ and any constraint $((x_1,\dots,x_k),R)$ of $\inst J_2$ with $e(y) = x_i$~(for some $i$) the constraint $((x_1,\dots,x_k),R)$ has exactly one preimage $((y_1,\dots,y_k),R)$ with $y = y_i$, we say that $e$ is a~\emph{covering}. A~\emph{universal covering tree instance} $\uct(\inst J)$ of a connected instance $\inst J$ is a~(possibly countably infinite) tree instance $\inst T$ together with a~covering $e\colon\inst T\to \inst J$ satisfying some additional properties. If $\inst J$ is a tree instance, then one can take $\uct(\inst J)=\inst J$, otherwise $\uct(\inst J)$ is always infinite. If an instance $\inst J$ is disconnected then $\uct(\inst J)$ is a disjoint union of universal covering tree instances for connected components of $\inst J$. Several equivalent (precise) definitions of $\UCT$ can be found in Section 5.4 of~\cite{Kozik16:circles} or Section 4 of~\cite{Kun12:robust}. For our purposes, it is enough to mention that, for any $\inst J$, the instance $\uct(\inst J)$~(with covering $e$) has the following two properties. For any two variables $v,v'$ satisfying $e(v)=e(v')$ there exists an endomorphism $h$ of $\uct(\inst I)$~ (i.e a~homomorphism into itself) sending $v$ to $v'$ and such that $e\circ h =e$. Similarly for constraints $C$ and $C'$ if $e(C)=e(C')$ then there is an endomorphism $h$ such that $h(C)=C'$ and $e\circ h = e$. It is well known that $\UCT(\inst J)$ has a solution if and only if $\inst J$ has an arc-consistent subinstance. Consider $\UCT(\inst K^i)$ and fix a covering $e': \UCT(\inst K^i) \rightarrow \inst K^i$. Let $\mathcal{T}^i$ be an instance obtained from $\UCT(\inst K^i)$ by replacing each constraint $C$ in it by a~tree that defines $e'(C)$, each time introducing a~fresh set of variables for the internal vertices of the trees. Let $e$ be the instance homomorphism from $\mathcal{T}^i$ to $\inst I$ defined in the natural way. We call a solution~(or a partial solution) to $\mathcal{T}^i$ {\em nice} if it maps each $v$ into $D_{e(v)}^{i+1}$ and moreover if $e(v)\in V^{<i}$ then $v$ is mapped to $\sol(e(v))$. It should be clear that nice solutions to $\mathcal{T}^i$ correspond to solutions of $\UCT(\inst K^i)$~ (although the correspondence is not one-to-one). \begin{claim}\label{claim:almostAC} There exists a~nice solution of $\mathcal{T}^i$. \end{claim} \begin{proof} If $\mathcal{T}^i$ is not connected, we consider each connected component separately and then take the union of nice solutions. Henceforth we assume that $\mathcal{T}^i$ is connected. By a standard compactness argument, it suffices to find a nice solution for every finite subtree of $\mathcal{T}^i$. Suppose, for a contradiction, that $\inst T$ is a minimal finite subtree of $\mathcal{T}^i$ without nice solutions. First, only the leaf vertices of $\inst T$ can be mapped, by $e$, into variables from $V^{<i}$. Indeed, if an internal vertex is mapped to a variable in $V^{<i}$, we can split the tree at this vertex into two parts, obtain~(from the minimality of $\inst T$) nice solutions to both parts~(which need to map the splitting vertex according to $\sol$, i.e., to the same element) and merge these solutions to obtain a nice solution to $\inst T$. This is a contradiction. Second we show that $\inst T$ has more than $n$ leaves mapped by $e$ into $V^{<i}$. Assume that $\inst T$ has $n$ or fewer leaves mapped to $V^{<i}$ and let $\inst T'$ be the smallest subtree of $\inst T$ with these leaves. Then $\inst T'$ is an $n$-tree and by $(E_{i-1})$ we obtain a solution $s$ to $\inst T'$ in $D^i_x$'s which sends leaves of $\inst T'$ according to $\sol$. It remains to extend $s$ to a solution of $\inst T$ in $D^{i+1}_x$'s. This extension is done in a sequence of steps. In each step $s$ is defined for increasingly larger subtrees of $\inst T$. Furthermore, at each step the following condition (*) is satisfied by $s$: if a vertex $v$ has a value assigned by $s$ and a neighbour without such value then $s(v)$ belongs to $D^i_{e(v)}$. Clearly, this condition holds in the beginning. At each step we pick a~constraint $C$ on a~vertex $v$ with an assigned value and a~vertex $v'$ without such a value. (Note that the constraints of $\inst T^i$, and consequently of $\inst T$, are binary.) $C$ has been added to $\inst T^i$ by replacing a constraint of $\uct(\inst K^i)$ with an $n$-tree ${\inst T_C}$ that defines it. Let $\inst S$ be a~maximal subtree of ${\inst T}$ such that it contains $C$, it has $v$ as a leaf, and all other nodes in $\inst S$ have not been assigned by $s$ and belong to ${\inst T_C}$. Since ${\inst T_C}$ is a $n$-tree, $\inst S$ is also an $n$-tree, and we can use $\IPQ{n}$ to derive that there exists a solution, $s'$, of ${\inst S}$ in $D_x^{i+1}$'s that sends $v$ to $s(v)\in D^i_{e(v)}$. More specifically, we apply $\IPQ{n}$ with $x=v$, $a=s(v)$, and both $p$ and $q$ being the same pattern $t_1+t_2$ such that $t_1$ is $\inst S$ with beginning $v$ and end being any other leaf of $\inst S$, and $t_2$ is $t_1$ with beginning and end swapped. This solution $s'$ can be added to $s$ (as the values on $v$ are the same). It remains to see that condition (*) is preserved after extending $s$ with $s'$. Indeed, let $u$ be any vertex such that after adding solution $s'$ has a neighbour $u'$ that has not yet been assigned. We can assume that $u$ is one of the new variables assigned by $s'$. If $e(u)\in V^i$ then the claim follows from the fact that $D^{i+1}_{e(u)}=D^i_{e(u)}$ so we can assume that $e(u)\not\in V^i$. However, in this case, all neighbours of $u$ in $\inst T$ must be in $\inst T_C$, so the constraint in ${\mathcal T}$ containing both $u$ and $u'$ must be also in ${\inst T_C}$ contradicting the maximality of ${\inst S}$. So the counterexample $\inst T$ must have at least $n+1$ leaves mapped into $V^{<i}$. Fix any $n+1$ of such leaves $v_1,\dotsc,v_{n+1}$ and let $\inst T_i$ denote a~subinstance of $\inst T$ obtained by removing, $v_i$ together with the single constraint containing $v_i$ : $((v_i,v'_i),R_i)$ from $\inst T$. Clearly $v'_i$ is not a leaf~(as it would make our $\inst T$ a two-element instance) and by the fact that only leaves can be mapped into $V^{<i}$ we get that $e(v'_i) \in V^{i}$ or $e(v'_i)\in V^{>i}$ and, in the last case, $i\neq|D|$. By minimality each $\inst T_i$ has a nice realization, say $s_i$. Now either $e(v'_i)\in V^i$ and $s_i(v'_i)\in D_{e(v'_i)}^i = D_{e(v'_i)}^{i+1}$ or $e(v'_i)\in V^{>i}, s_i(v'_i)\in D_{e(v'_i)}^{i+1}$ and $i+1\neq |D|+1$. In both cases $s_i(v'_i)\in D_{e(v'_i)}^j$ for $j\leq |D|$ and thus, by \IPQ{n}, there exists $a_i\in D$ such that $(s_i(v'_i),a_i)\in R_i$. We let $s'_i$ be the realization of $\inst T$ obtained by extending $s_i$ by mapping $v_i$ to $a_i$. The last step is to apply the $n+1$-ary near unanimity operation coordinatewise to $s'_i$'s~ (in a way identical to the one in the proof of Claim~\ref{claim:absorbing}). The application produces a nice realization of $\inst T$. This contradiction finishes the proof of the claim. \end{proof} We will denote the arc-consistent subinstance of $\inst K^i$~ (which is about to be constructed) by $\inst L^i$. The variables of $\inst L^i$ and $\inst K^i$~(or indeed $\inst I^i$) are the same. For every constraint $(\tuple x, R)$ in $\inst K^i$ we introduce a constraint $(\tuple x, R')$ into $\inst L^i$ where \begin{equation*} R' = \{\tuple{a}\colon \tuple a = s(\tuple y) \text{ where $s$ is a solution to $\UCT(\inst K^i)$ and $e'((\tuple y,R)) = (\tuple x, R)$}\} \end{equation*} where $e'$ is an instance homomorphism mapping $\UCT(\inst K^i)$ to $\inst K^i$. In other words we restrict a relation in a constraint of $\inst K^i$ by allowing only the tuples which appear in a solution of the $\UCT(\inst K^i)$~ (at this constraint). All the relations of $\inst L^i$ are preserved by all the polymorphisms of $\Gamma$, and are non-empty~(by Claim 2). The fact that $\inst L^i$ is arc-consistent is an easy consequence of the endomorphism structure of universal covering trees. Finally Claim~\ref{claim:absorbing} holds for $\inst L^i$: \begin{claim}\label{claim:absorbingL} Let $((x_1,\dotsc,x_k),R)$ and $((x_1,\dotsc,x_k),R')$ be constraints defined by the same tree $t$ in $\inst I^i$ and $\inst L^i$, respectively. Let $\tuple a_1,\dotsc,\tuple a_{n+1}\in R'$ and $\tuple a\in R$, then $f(\tuple a_1,\dotsc,\tuple a,\dotsc,\tuple a_{n+1})$, where $f$ is the $n+1$-ary near unanimity operation and $\tuple a$ is in position $j$, belongs to $R'$. \end{claim} \begin{proof} By Claim~\ref{claim:absorbing} the tuple $f(\tuple a_1,\dotsc,\tuple a,\dotsc,\tuple a_{n+1})$ belongs to the relation in the corresponding constraint in $\inst K^i$. Thus if it extends to a solution of $\UCT(\inst K^i)$ it belongs to $R'$. However each $\tuple a^i$ extends to a solution of $\UCT(\inst K^i)$ and $\tuple a$ extends to a solution of $\UCT(\inst I^i)$. By applying the near-unanimity operation $f$ to these extensions~(coordinatewise), we obtain the required evaluation. \end{proof} \subsection{A solution to $\inst K^i$} In order to find a solution to $\inst L^i$, we will use Corollary B.2 from~\cite{Kozik16:circles}. We state it here in a~simplified form using the following notation: for subuniverses $A'\subseteq A$, we say that $A'$ {\em nu-absorbs} $A$ if, for some NU polymorphism $f$, $f(a_1,\ldots,a_n)\in A'$ whenever $a_1,\ldots,a_n\in A$ and at most one $a_i$ is in $A\setminus A'$. Similarly, if $R'\subseteq R$ are relations preserved by all polymorphisms of $\Gamma$ we say $R'$ nu-absorbs $R$, if for some near-unanimity operation $f$ taking all arguments from $R'$ except for one which comes from $R$ produces a result in $R'$. \begin{corollary}[Corollary B.2 from~\cite{Kozik16:circles}] Let $\inst I$ satisfy (PQ) condition in sets $A_x$. Let $\inst I'$ be an arc-consistent instance in sets $A'_x$ on the same set of variables as $\inst I$ such that: \begin{enumerate} \item for every variable $x$ the subuniverse $A'_x$ nu-absorbs $A_x$, and \item for every constraint $((x_1,\dotsc,x_n),R')$ in $\inst I'$ there is a corresponding constraint $((x_1,\dotsc,x_n),R)$ in $\inst I$ such that $R'$ nu-absorbs $R$~(and both respect the NU operation). \end{enumerate} Then there are subuniverses $A_x''$ of $A_x'$~(for every $x$) such that the instance $\inst I''$ obtained from $\inst I'$ by restricting the domain of each variable to $A''_x$ and by restricting the constraint relations accordingly satisfies the condition (PQ). \end{corollary} We will apply the corollary above using $\inst I^i$ for $\inst I$ and $\inst L^i$ for $\inst I'$. By our construction, $\inst I^i$ satisfies condition (PQ), and the sets $D_x^i$~ (which play the role of $A_x$) are subuniverses of $D$. On the other hand $\inst L^i$ is arc-consistent and all the relations involved in it are closed under the polymorphisms of $\Gamma$. Claim~\ref{claim:absorbingL} shows that each relation $R'$ nu-absorbs the corresponding $R$. By arc-consistency, the projection of $R'$ on a variable $x$ is the same for each constraint $((x_1,\ldots,x_n),R')$ containing $x$, call the corresponding sets $A'_x$. Since each $R'$ nu-absorbs $R$, it follows that each $A'_x$ nu-absorbs the corresponding $A_x$. The corollary implies that we can restrict the instance $\inst L^i$ to obtain an instance satisfying (PQ). By Theorem~\ref{thm:nice} such an instance, and thus both $\inst K^i$ and $\inst L^i$, has a solution. \subsection{Finishing the proof} We choose any solution to $\inst K^i$ and extend the global solution $\sol$ to $V^i$ according to it. There exists a solution on $V^{\le i}$, because every constraint between two variables from this set is either in $V^{<i}$ or defines a two-variable $n$-tree which was used to define a constraint in $\inst K^i$. It remains to prove that, with such an extension, condition $(E_{i})$ holds. Let $t$ be an $n$-tree pattern in $\inst I$. If it has no variables mapped to $V^{i}$, then $(E_i)$ follows from $(E_{i-1})$. Assume that it has such variables. By splitting $t$ at internal vertices mapped to $V^i$, it is enough to consider the case when only leaves of $t$ are mapped to $V^i$. Then $t$ defines a constraint $(\tuple x,R)$ in $\inst K^i$. The solution to $\inst K^i$ mapping $\tuple x$ to $\tuple a\in R$ and the evaluation of $t$ witnessing that $\tuple a$ belongs to $R$ can be taken to satisfy $(E_i)$ for $t$. Theorem~\ref{thm:nicelevels} is proved. \section{Full proof of Theorem \ref{the:main}(1)} \label{sec:correctness-of-algorithm} In this subsection we prove Propositions \ref{prop:expected-lost-weight} and~\ref{prop:consistence}. The following equalities, which can be directly verified, are used repeatedly in this section: for any subsets $A,B$ of $D$ and any feasible solution $\{\mathbf{x}_a\}$ of the SDP relaxation of $\inst I$ it holds that $\|\mathbf{x}_A\|^2=\mathbf{x}_A\mathbf{y}_D$ and $\|\mathbf{y}_B-\mathbf{x}_A\|^2=\mathbf{x}_{D\setminus A}\mathbf{y}_B+\mathbf{x}_A\mathbf{y}_{D\setminus B}$. \subsection{Analysis of Preprocessing step 2}\label{sec:preproc} In some of the proofs it will be required that $\alpha\leq c_0$ for some constant $c_0$ depending only on $|D|$. This can be assumed without loss of generality, since we can adjust constants in $O$-notation in Theorem~\ref{the:main}(1) to ensure that $\eps\le c_0$ (and we know that $\alpha\le \eps$). We will specify the requirements on the choice of $c_0$ as we go along. \begin{lemma} \label{le:defH} \label{lem:maj2} There exists a constant $c>0$ that depends only on $|D|$ such that the sets $D_x^{\ell}\subseteq D$, $x\in V$, $1\leq \ell\leq |D|$, obtained in Preprocessing step 2, are non-empty and satisfy the following conditions: \begin{enumerate} \item for every $a\in D_x^{\ell}$, $\|\mathbf{x}_a\|\geq \alpha^{3\ell \kappa}$, \item for every $a\not\in D_x^{\ell}$, $\|\mathbf{x}_a\|\leq c\alpha^{3\ell \kappa}$. \item for every $a\in D_x^{\ell}$, $\|\mathbf{x}_a\|^2\geq 2 \|\mathbf{x}_{D\setminus{D_x^{\ell}}}\|^2$ \item $D_x^{\ell}\subseteq D_x^{\ell+1}$ (with $D_x^{|D|+1}=D$) \end{enumerate} \end{lemma} \begin{proof} Let $c=(2|D|)^{(|D|/2)}$. It is straightforward to verify that conditions (1)--(3) are satisfied. Let us show condition (4). Since $c$ only depends on $|D|$ we can choose $c_0$ (an upper bound on $\alpha$) so that $c\alpha^{3\kappa}<1$. It follows that $c\alpha^{3(\ell+1)\kappa}<\alpha^{3\ell \kappa}$. It follows from conditions (1) and (2) that $D_x^{\ell}\subseteq D_x^{\ell+1}$. Finally, let us show that $D_x^{\ell}$ is non-empty. By condition (4) we only need to take care of case $\ell=1$. We have by condition (2) that \[ \sum_{a\in {D\setminus D_x^1}} \|\mathbf{x}_a\|^2\leq |D|c^2\alpha^{6\kappa} \] Note that we can adjust $c_0$ to also satisfy $|D|c^2\alpha^{6\kappa}<1$ because, again, $c$ only depends on $|D|$. \end{proof} \subsection{Proof of Proposition \ref{prop:expected-lost-weight}} We will prove that the total weight of constraints removed in each step 0-5 of the algorithm in Section~\ref{sec:algorithm} is $O(\alpha^\kappa)$. \begin{lemma} \label{le:step0} The total weight of the constraints removed at step $0$ is at most $\alpha^{\kappa}$. \end{lemma} \begin{proof} We have \[ \alpha\geq \sum_{C\in{\mathcal C}} w_C \loss(C)\geq \sum_{\substack{C\in{\mathcal C} \\ \loss(C)\geq\alpha^{1-\kappa}}} w_C \alpha^{1-\kappa}, \] from which the lemma follows. \end{proof} \begin{lemma} \label{le:path-loss} Let $((x,y),R)$ be a constraint not removed at step $0$, and let $A,B$ be such that $B=A+^{\ell}(x,R,y)$. Then $\|\mathbf{y}_B\|^2\geq \|\mathbf{x}_A\|^2-c\alpha^{(6\ell+6)\kappa}$ for some constant $c>0$ depending only on $|D|$. The same is also true for a~constraint $((y,x),R)$ and $A = B+^\ell (y,R^{-1},x)$. \end{lemma} \begin{proof} Consider the first case, i.e., a~constraint $((x,y),R)$ and $B = A+^\ell (x,R,y)$. We have \[ \mathbf{x}_A\mathbf{y}_{D\setminus B} = \sum_{\substack{a\in A, b\in D\setminus B \\ (a,b)\not\in R}} \mathbf{x}_a\mathbf{y}_b + \sum_{\substack{a\in A, b\in D\setminus B \\ (a,b)\in R}} \mathbf{x}_a\mathbf{y}_b. \] The first term is bounded from above by the loss of constraint $((x,y),R)$, and hence is at most $\alpha^{1-\kappa}$, since the constraint has not been removed at step $0$. Since $B=A+^{\ell}(x,R,y)$ it follows that for every $(a,b)\in R$ such that $a\in A$ and $b\in D\setminus{B}$ we have that $a\not\in D_x^{\ell+1}$ or $b\not\in D_y^{\ell+1}$. Hence, the second term is at most \[ \mathbf{x}_{D\setminus D_x^{\ell+1}}\mathbf{y}_D+\mathbf{x}_D\mathbf{y}_{D\setminus D_y^{\ell+1}}=\|\mathbf{x}_{D\setminus D_x^{\ell+1}}\|^2+\|\mathbf{y}_{D\setminus D_y^{\ell+1}}\|^2 \] which, by Lemma~\ref{le:defH}(2), is bounded from above by $d\alpha^{(6\ell+6)\kappa}$ for some constant $d>0$. From the definition of $\kappa$ it follows that $(6\ell+6)\kappa\leq 1-\kappa$, and hence we conclude that $\mathbf{x}_A\mathbf{y}_{D\setminus B}\leq (d+1)\alpha^{(6\ell+6)\kappa}$. Then, we have that $$ \|\mathbf{y}_B\|^2 = \mathbf{x}_A\mathbf{y}_B+\mathbf{x}_{D\setminus A}\mathbf{y}_B \geq \mathbf{x}_A\mathbf{y}_B = \mathbf{x}_A\mathbf{y}_D-\mathbf{x}_A\mathbf{y}_{D\setminus B} \geq \|\mathbf{x}_A\|^2-(d+1)\alpha^{(6\ell+6)\kappa}. $$ \end{proof} \begin{lemma} \label{le:step1} The expected weight of the constraints removed at step $1$ is $O(\alpha^{\kappa})$. \end{lemma} \begin{proof} Let $((x,y),R)$ be a constraint not removed at step $0$. We shall see that the probability that it is removed at step $1$ is at most $c \alpha^{\kappa}$ where $c>0$ is a constant. Let $A,B$ be such that $B=A+^{\ell} (x,R,y)$. It follows from Lemma \ref{le:path-loss} that $\|\mathbf{y}_B\|^2\geq \|\mathbf{x}_A\|^2-d\alpha^{(6\ell+6)\kappa}$ for some constant $d>0$. Hence, the probability that a value $r_\ell$ in step $1$ makes that $\mathbf{y}_B\not\preceq^{\ell} \mathbf{x}_A$ is at most \[ \frac{d\alpha^{(6\ell+6)\kappa}}{\alpha^{(6\ell+4)\kappa}} = d\alpha^{2\kappa}\leq d \alpha^{\kappa}. \] We obtain the same bound if we switch $x$ and $y$, and consider sets $A,B$ such that $A=B+^{\ell} R^{-1}$. Taking the union bound for all sets $A,B$ and all values of $\ell$ we obtain the desired bound. \end{proof} \begin{lemma} \label{lem:removing-variable} If there exists a~constant $c>0$ depending only on $|D|$ such that for every variable $x$, the probability that all constraints involving $x$ are removed in Step~2, Step~3, or Step~5 is at most $c\alpha^\kappa$, then the total expected weight of constraints removed this way in the corresponding is at most $2c\alpha^\kappa$. \end{lemma} \begin{proof} Let $w_x$ denote the total weight of the constraints in which $x$ participates. The expected weight of constraints removed is at most \[ \sum_{x\in V} w_x c \alpha^{\kappa} = (\sum_{x\in V} w_x) c\alpha^{\kappa} = 2 c\alpha^{\kappa} \] and the lemma is proved. \end{proof} \begin{lemma} The expected weight of the constraints removed at step 2 is $O(\alpha^\kappa)$. \end{lemma} \begin{proof} Let $x$ be a~variable. According to Lemma~\ref{lem:removing-variable} it is enough to prove that the probability that we remove all constraints involving $x$ at step 2 is at most $c\alpha^\kappa$ for some constant $c>0$. Suppose that $A\subseteq B$ are such that $\| \vec x_B \|^2 - \| \vec x_A \|^2 = \|\vec x_B - \vec x_A \|^2 \leq (2n -3) \alpha^{(6\ell+4)\kappa}$. Then the probability that one of the bounds of the form $r_\ell + (s_\ell +jm_0)\alpha^{(6\ell +4)\kappa}$ separates $\| \vec x_B \|^2$ and $\| \vec x_A \|^2$ is at most \[ {(2n-3)}/{m_0} \leq (2n-3)/ (\alpha ^{-2\kappa} - 1) \] which is at most $c\alpha^\kappa$ for some constant $c>0$ whenever $\alpha^\kappa < 1/2$. Taking the union bound for all sets $A,B$ and all values of $\ell$ we obtain the desired bound. \end{proof} \begin{lemma} \label{le:cut} There exist constants $c,d>0$ depending only on $|D|$ such that for every pair of variables $x$ and $y$ and every $A,B\subseteq D$, the probability, $p$, that a unit vector $\mathbf{u}$ chosen uniformly at random cuts $\mathbf{x}_A$ and $\mathbf{y}_B$ satisfies \[ c\cdot \| \mathbf{y}_B-\mathbf{x}_A \| \leq p\leq d\cdot \|\mathbf{y}_B-\mathbf{x}_A \|. \] \end{lemma} \begin{proof} Let $0\leq x\leq 1$ and let $0\le \theta\le \pi$ be an angle such that $x=\cos(\theta)$. There exist constants $a,b>0$ such that $$a\cdot \sqrt{1-x}\leq \theta\leq b\cdot \sqrt{1-x}.$$ Now, if $\theta$ is the angle between $\mathbf{x}_A-\mathbf{x}_{D\setminus A}$ and $\mathbf{y}_B-\mathbf{y}_{D\setminus B}$ then \begin{multline*} 1-\cos(\theta) = 1-(\mathbf{x}_A-\mathbf{x}_{D\setminus A})(\mathbf{y}_B-\mathbf{y}_{D\setminus B}) \\ = 2(\mathbf{x}_{D\setminus A}\mathbf{y}_B+\mathbf{x}_A\mathbf{y}_{D\setminus B}) = 2\|\mathbf{y}_B-\mathbf{x}_A \|^2 \end{multline*} Since $p=\theta/\pi$, the result follows. \end{proof} \begin{lemma} \label{le:maj-step3} \label{le:step4} The expected weight of the constraints removed at step 3 is $O(\alpha^{\kappa})$. \end{lemma} \begin{proof} According to Lemma \ref{lem:removing-variable}, it is enough to prove that the probability that we remove all constraints involving $x$ at step 3 is at most $c\alpha^{\kappa}$ for some constant $c$. Let $A$ and $B$ such that $A\cap D_x^{\ell}\neq B\cap D_x^{\ell}$. Let $a$ be an element in symmetric difference $(A\cap D_x^{\ell})\triangle (B\cap D_x^{\ell})$. Then we have $\|\mathbf{x}_B-\mathbf{x}_A \|=\sqrt{\mathbf{x}_{D\setminus A}\mathbf{x}_B+\mathbf{x}_A\mathbf{x}_{D\setminus B}}\geq \|\mathbf{x}_a\|\geq \alpha^{3\ell \kappa}$, where the last inequality is by Lemma~\ref{le:defH}(1). Then by Lemma \ref{le:cut} the probability that $\mathbf{x}_A$ and $\mathbf{x}_B$ are not $\ell$-cut is at most \[ (1-c\alpha^{3\kappa\ell})^{m_{\ell}} \leq \frac{1}{\exp(c\alpha^{3\kappa\ell}m_{\ell})} \leq \frac{1}{\exp(c\alpha^{-\kappa})} \leq c\alpha^{\kappa}. \] where $c$ is the constant given in Lemma \ref{le:cut}. Taking the union bound for all sets $A,B$ and all values of $\ell$ we obtain the desired bound. \end{proof} \begin{lemma} \label{le:maj-step2} \label{le:step3} The expected weight of the constraints removed at step~4 is $O(\alpha^{\kappa})$. \end{lemma} \begin{proof} Let $((x,y),R)$ be a constraint not removed at steps $0$ and $1$. We shall prove that the probability that it is removed at step 4 is at most $c\alpha^{\kappa}$ for some constant $c>0$. Fix $\ell$ and $A,B$ such that $B=A+^{\ell}(x,R,y)$. Since the constraint has not been removed at step 1, we have $\mathbf{y}_B\preceq^{\ell} \mathbf{x}_A$. Since $B=A+^{\ell} p$ we have that $\mathbf{x}_A \mathbf{y}_{D\setminus B}\leq c_1\alpha^{(6\ell+6)\kappa}$, as shown in the proof of Lemma~\ref{le:path-loss}. Since $\|\mathbf{x}_A\|^2=\mathbf{x}_A(\mathbf{y}_B+ \mathbf{y}_{D\setminus B})$, it follows that $\mathbf{x}_A \mathbf{y}_B\geq \|\mathbf{x}_A\|^2-c_1\alpha^{(6\ell+6)\kappa}$. Also, we have $\|\mathbf{y}_B\|^2=(\mathbf{x}_A\mathbf{y}_B+\mathbf{x}_{D\setminus A}\mathbf{y}_B)$ is at most $\|\mathbf{x}_A\|^2+\alpha^{(6\ell+4)\kappa}$ because $\mathbf{y}_B\preceq^{\ell} \mathbf{x}_A$. Using the bound on $\mathbf{x}_A \mathbf{y}_B$ obtained above, it follows that $\mathbf{x}_{D\setminus A}\mathbf{y}_B$ is at most $\alpha^{(6\ell+4)\kappa}+c_1\alpha^{(6\ell+6)\kappa}\leq (c_1+1)\alpha^{(6\ell+4)\kappa}$. Putting the bounds together, we have that \begin{multline*} \|\mathbf{y}_B-\mathbf{x}_A \| = \sqrt{\mathbf{x}_{D\setminus A}\mathbf{y}_B + \mathbf{x}_A\mathbf{y}_{D\setminus B}} \leq \\ \sqrt{c_1\alpha^{(6\ell+6)\kappa} + (c_1+1)\alpha^{(6\ell+4)\kappa}} \leq c_2 \alpha^{(3\ell+2)\kappa} \end{multline*} for some constant $c_2>0$. Applying the union bound and Lemma \ref{le:cut} we have that the probability that $\mathbf{x}_A$ and $\mathbf{y}_B$ are $\ell$-cut is at most $m_{\ell} dc_2\alpha^{(3\ell+2)\kappa}=O(\alpha^{\kappa})$. We obtain the same bound if we switch $x$ and $y$, and take $R^{-1}$ instead of~$R$. Taking the union bound for all sets $A,B$ and all values of $\ell$ we obtain the desired bound. \end{proof} \begin{lemma} The expected weight of the constraints removed at step 5 is $O(\alpha^\kappa)$. \end{lemma} \begin{proof} Again, according to Lemma \ref{lem:removing-variable}, it is enough to prove that the probability that we remove all constraints involving $x$ at step~5 is at most $c_1\alpha^{\kappa}$ for some constant $c_1$. Suppose that $A$, $B$ are such that $\| \vec x_A - \vec x_B \|^2 \leq (2n-3)\alpha^{(6\ell+4)\kappa}$. Hence, by Lemma \ref{le:cut} and the union bound the probability that $\vec x_A$ and $\vec x_B$ are $\ell$-cut is at most \[ m_\ell d(2n-3)^{1/2}\alpha^{(3\ell+2)\kappa} \leq d(2n-3)^{1/2}\alpha^ \kappa \] where $d$ is the constant from Lemma \ref{le:cut}. Taking the union bound for all sets $A$, $B$ and all values of $\ell$, we obtain the desired bound. \end{proof} \subsection{Proof of Proposition \ref{prop:consistence}} All patterns appearing in this subsection are in $\inst I'$. The following notion will be used several times in our proofs: Let $t$ be a tree and let $y$ be one of its nodes. We say that a subtree $t'$ of $t$ is {\em separated by vertex $y$} if $t'$ is maximal among all the subtrees of $t$ that contain $y$ as a leaf. \begin{lemma} \label{le:notdecreasing} \label{lem:maj18} \label{lem:path-loss} Let $1\leq \ell\leq |D|$, let $p$ be a~path pattern from $x$ to $y$, and let $A,B$ be such that $B=A+^{\ell} p$. Then $\mathbf{x}_A\preceq^{\ell} \mathbf{y}_B$, and in particular, $\|\mathbf{x}_A \| \leq \|\mathbf{y}_B \| + \alpha^{(6\ell+4)\kappa}$. \end{lemma} \begin{proof} Since the relation $\preceq^\ell$ is transitive, it is enough to prove the lemma for path patterns containing only one constraint. But this is true, since all the constraints $((x,y),R)$ or $((y,x),R)$ which would invalidate the lemma have been removed in step 1. \end{proof} \begin{lemma} \label{lem:lost-on-a-tree} If $p$ is a~tree pattern with at most $j+1$ leaves starting at $x$, and $A \subseteq D_x^{\ell+1}$ is such that $A +^\ell p = \emptyset$ then \( \| \vec x_A \|^2 \leq (2j-1) \alpha^{(6\ell +4) \kappa} \). \end{lemma} \begin{proof} We will prove the statement by induction on the number of leaves. For $j=1$ this follows from Lemma \ref{lem:maj18}. Suppose then that $p$ is a~tree pattern with $j+1>2$ leaves and the statement is true for any tree pattern with at most $j$ leaves. Choose $y$ to be the first branching vertex in the unique path in $p$ from $x$ to the end of $p$, and let $p_0,t_1,\dots,t_h$ be all subtrees of $p$ separated by $y$ where $p_0$ is the subtree containing $x$. We turn $p_0$ into a pattern by choosing $x$ as beginning and $y$ as end. Similarly, we turn every $t_i$ into a pattern by choosing $y$ as beginning and any other arbitrary leaf as end. Since $y$ is a~branching vertex, we have that $h \geq 2$, every $t_i$ has $j_i + 1 < j + 1$ leaves, and $\sum_{i=1}^h{j_i} = j$. Now, let $B_i$ denote the set $\{ a\in D_y^{\ell +1} : a +^\ell t_i = \emptyset \}$. Since $j_i < j$, we know that $\| \vec y_{B_i} \|^2 \leq (2j_i -1) \alpha^{(6\ell + 4)\kappa}$. Further, for $B = \bigcup_{i=1}^h B_i$, we have, using inductive assumption, that \begin{multline*} \|\vec y_B\|^2 \leq \sum_{i=1}^h \|\vec y_{B_i}\|^2 \leq \sum_{i=1}^h (2j_i -1) \alpha^{(6\ell + 4)\kappa} \\ = (2j - h) \alpha^{(6\ell + 4)\kappa} \leq (2j - 2) \alpha^{(6\ell +4)\kappa}. \end{multline*} Finally, since $A +^\ell p = \emptyset$ then $A +^\ell p_0 \subseteq B$, and the claim follows from Lemma~\ref{lem:maj18}. \end{proof} \begin{lemma} \label{lem:tree-loss} Let $1\leq \ell \leq |D|$, let $p$ be a pattern from $x$ to $y$ which is a~path of $n$-trees. If $A,B\subseteq D$ are such that $B +^\ell p = A$, then \( \|\vec y_{A}\|^2 \geq \|\vec x_B\|^2 - \alpha^{(6\ell +2)\kappa} \). \end{lemma} \begin{proof} We claim that for any $n$-tree pattern $t$ and $A,B$ with $B+^\ell t = A$, we have $\vec x_B \preceq_w^\ell \vec y_A$. Since the relation $\preceq_w^\ell$ is transitive, the lemma is then a~direct consequence. For a~contradiction, suppose that $t$ is a~smallest (by inclusion) $n$-tree that does not satisfy the claim. Observe that $t$ is not a~path, due to Lemma~\ref{lem:path-loss} and the fact that $\vec x_B\preceq^\ell \vec y_A$ implies $\vec x_B\preceq_w^\ell \vec y_A$. Let $v_x$ and $v_y$ denote the beginning and the end vertex of $t$, respectively; and let $v_z$ be the last branching vertex that appears on the path connecting $v_x$ and $v_y$, and let it be labeled by $z$. Let $t_1,t_2,p_1,\dots,p_j$ be all subtrees of $t$ separated by $v_z$, where $t_1$ and $t_2$ are the subtrees containing $v_x$ and $v_y$ respectively. Let us turn $p_1,\dots,p_j$ into patterns by choosing $v_j$ as beginning and any other leaf as end. Furthermore, choose $x$ and $z$ to be the beginning and end, respectively, of $t_1$ and $z$ and $y$ to be the beginning and end, respectively, of $t_2$. Note that $t_2$ is a~path. Further, we know that for $C = B +^\ell t_1$ we have $\vec x_B \preceq_w^\ell \vec z_C$ by minimality of $t$. Now, let $C_i = \{ a \in D_z^{\ell +1} : a+^\ell p_i = \emptyset \}$. Then by Lemma \ref{lem:lost-on-a-tree}, we get that $\|\vec z_{C_i} \|^2 \leq (2j_i -1)\alpha^{(6\ell +4)\kappa}$ where $j_i+1$ is the number of leaves of $p_i$, therefore for $C' = \bigcup C_i$ we have $\| \vec z_{C'} \|^2 \leq \sum\| \vec z_{C_i} \|^2\le (2n - 3) \alpha^{(6\ell +4)\kappa}$. This implies that $\| \vec z_{C \setminus C'} \|^2 \geq \| \vec z_C \|^2 - (2n - 3) \alpha^{(6\ell +4)\kappa}$, and consequently $\vec z_C \preceq_w^\ell \vec z_{C \setminus C'}$ as otherwise all constraints containing $z$ would have been removed at step 2. Finally, observe that $A = (C\setminus C') +^\ell t_2$, and therefore $\vec z_{C \setminus C'} \preceq^\ell \vec y_A$ and, hence, $\vec z_{C \setminus C'} \preceq_w^\ell \vec y_A$. Putting this together with all other derived $\preceq^\ell_w$-relations, we get the required claim. \end{proof} \begin{lemma} \label{lem:maj19} Let $1 \leq \ell \leq |D|$, let $p$ be a pattern from $x$ to $x$ which is a path of $n$-trees, and let $A,B$ be such that $B +^\ell p = A$. If $A \cap D_x^\ell \subseteq B \cap D_x^\ell$ then $A \cap D_x^\ell = B \cap D_x^\ell$. \end{lemma} \begin{proof} For a~contradiction, suppose that there is an~element $a \in (D_x^\ell \cap B) \setminus A$. From Lemma~\ref{lem:maj2} we get that \( \| \vec x_{B\setminus A} \|^2 \geq \|\vec x_a \|^2 \geq 2 \|\vec x_{D\setminus D_x^\ell} \|^2 \geq 2 \|\vec x_{A\setminus B} \|^2. \) Therefore, we have \begin{multline*} \|\vec x_A\|^2 = \|\vec x_B\|^2 - \|\vec x_{B\setminus A}\|^2 + \|\vec x_{A\setminus B}\|^2 \leq \|\vec x_B \|^2 - \| \vec x_a \|^2 + (1/2)\| \vec x_a \|^2 = \\ \|\vec x_B\|^2-(1/2)\|\vec x_a\|^2 \leq \|\vec x_B\|^2-(1/2)\alpha^{6\ell \kappa}. \end{multline*} On the other hand, since $p$ is a~path of $n$-trees, we get from the previous lemma that \( \|\vec x_A\|^2 \geq \|\vec x_B\|^2 - \alpha^{(6\ell +2)\kappa} \). If we adjust constant $c_0$ from Section~\ref{sec:preproc} so that $1/2> \alpha^{2\kappa}$, the above inequalities give a~contradiction. \end{proof} \begin{lemma}\label{lem:consist} Let $x$ be a~variable, let $p$ and $q$ be two patterns from $x$ to $x$ which are paths of $n$-trees, let $1 \leq \ell \leq |D|$, and let $A\subseteq D_x^\ell$. Then there exists some $j$ such that $A \subseteq A +^\ell (j(p + q) + p)$. \end{lemma} \begin{proof} For every $A$, define $A_0,A_1,\dots$ in the following way. If $i = 2j$ is even then $A_i = A +^\ell (j(p+q))$. Otherwise, if $i = 2j+1$ is odd then $A_i = A +^\ell (j(p+q) + p)$. We claim that for every sufficiently large $u$, we have $A_u \cap D_x^\ell = A_{u+1} \cap D_x^\ell$. From the finiteness of $D$, we get that for every sufficiently large $u$ there is $u' > u$ such that $A_u = A_{u'}$. It follows that there exists some path of $n$-trees pattern $p'$ starting and ending in $x$ such that $A_u = A_{u+1} +^\ell p'$. To prove the claim we will show that $\vec x_{A_u}$ and $\vec x_{A_{u+1}}$ are not $\ell$-cut. Then the claim follows as otherwise we would have removed all constraints involving $x$ at step 3. Consider the path $x_1,\dots,x_u$ in $p'$ which connects the beginning and end vertices. Further, let $R_i = R$ if the $i$-th edge of the path is labeled by $((x_i,x_{i+1}),R)$, and let $R_i = R^{-1}$ if the $i$-th edge is labeled by $((x_{i+1},x_i),R)$. Now define a~sequence $B_1,B_2',B_2,\dots,B_m$ inductively by setting $B_1 = A_{u+1}$, $B'_{i+1} = B_i +^\ell (x_i,R_i,x_{i+1})$. Further, if $x_{i+1}$ is not a~branching vertex, put $B_{i+1} = B_{i+1}'$. If $x_{i+1}$ is a~branching vertex, then let $\Phi_i$ be the set of all subtrees separated by $x_{i+1}$ in $p'$, excluding the two such subtrees containing the beginning and the end of $p'$. Then, turn each subtree in $\Phi_i$ into a pattern by choosing $x_{i+1}$ as beginning and any other leaf as end, and define \( B_{i+1} = \{ b\in B_{i+1}' : b +^\ell t \neq \emptyset \mbox{ for all } t \in \Phi_i \}. \) Since $p'$ is a~path of $n$-trees, we know that the sum of the numbers of leaves of the trees from $\Phi_i$ that are also leaves of $p'$ is strictly less than $n-1$. Finally, if $\vec x_{A_u}$ are $\vec x_{A_{u+1}}$ are $\ell$-cut then, for some $i$, vectors ${\vec x_i}_{B_i}$ and ${\vec x_{i+1}}_{B'_{i+1}}$ are $\ell$-cut, or vectors ${\vec x_i}_{B_i}$ and ${\vec x_i}_{B_i'}$ are $\ell$-cut. The former case is impossible since $B'_{i+1} = B_i +^\ell (x_i,R_i,x_{i+1})$, and hence if $\vec x_{B'_{i+1}}$ and $\vec x_{B_i}$ are $\ell$-cut, then either of the constraints $((x_i,x_{i+1}), R_i)$ or $((x_{i+1},x_i),R^{-1})$ would have been removed at step 4. We now show that the latter case is impossible either. Clearly, in this case $x_i$ is a branching vertex. For $t\in \Phi_i$, let $C_t =\{ b \in B_i' : b +^\ell t = \emptyset \}$ and let $j_t$ be the number of leaves of $t$. By Lemma~\ref{lem:lost-on-a-tree} we get \( \| {\vec x_i}_{C_t} \|^2 \leq (2j_t -1) \alpha^{(6\ell + 4)\kappa} \) for any $t \in \Phi_i$, and consequently, \[ \| {\vec x_i}_{B_i'} - {\vec x_i}_{B_i} \|^2 \leq \sum_{t\in \Phi_i} \| {\vec x_i}_{C_t} \|^2 \leq \sum_{t\in \Phi_i} (2j_t - 1) \alpha^{(6\ell +4)\kappa} \leq (2n - 3) \alpha^{(6\ell +4)\kappa}. \] Therefore, if ${\vec x_i}_{B_i}$ and ${\vec x_i}_{B_i'}$ were $\ell$-cut, then all constraints that include $x_i$ would have been removed at step 5. We conclude that indeed we have $A_u \cap D_x^\ell = A_{u+1} \cap D_x^\ell$ for all sufficiently large $u$. Now, take $u = 2j+1$ large enough. We have that \( (A \cup A_{u+1}) +^\ell (j(p+q) + p) = A_u \cup A_{2u+1}. \) And also $(A_u \cup A_{2u+1}) \cap D_x^\ell = A_{u+1} \cap D_x^\ell \subseteq (A \cup A_{u+1})\cap D_x^\ell$, hence by Lemma \ref{lem:maj19} we get that \( (A \cup A_{u+1}) \cap D_x^\ell = A_{u+1} \cap D_x^\ell \). Since $A\subseteq D_x^\ell$ by assumption of the lemma, we have \( A \subseteq A_{u+1} \cap D_x^\ell \subseteq A_{u} = A +^\ell (j(p+q) + p) \). \end{proof} Finally, setting $A=\{a\}$ in Lemma~\ref{lem:consist} gives Proposition~\ref{prop:consistence}. \section{Full proof of Theorem~\ref{the:main}(2)}\label{sec:thm2} In this section, we prove Theorem~\ref{the:main}(2). A brief outline of the proof is given in Section~\ref{sec:overview-thm2}. Throughout this section, $\inst I=(V,{\mathcal C})$ is a $(1-\eps)$-satisfiable instance of $\CSP\Gamma$ where $\Gamma$ consists of implicational constraints. \subsection{SDP Relaxation}\label{sec:thm2:SDP} We use SDP relaxation (\ref{sdpobj})--(\ref{sdp4}) from Section~\ref{sec:SDP}. For convinience, we write the SDP objective function as follows. \begin{multline} \sum_{C \in {\mathcal C}\text{ equals } (x = a) \vee (y = b)} w_C \vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})} \\ {}+\frac{1}{2}\sum_{C \in {\mathcal C}\text{ equals } x = \pi(y)} \,\sum_{a\in D} w_C \| \mathbf{x}_{\pi(a)} - \mathbf{y}_{a}\|^2 \\ {}+ \sum_{C \in {\mathcal C}\text{ equals }x \in P} w_C \left(\sum_{a\in D\setminus P} \|\mathbf{x}_{a}\|^2 \right). \label{SDP} \end{multline} This expression equals (\ref{sdpobj}) because of SDP constraint~(\ref{sdp3}). \iffalse Minimize \begin{multline} \sum_{C \in {\mathcal C}\text{ equals } (x = a) \vee (y = b)} w_C \vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})} \\ {}+\frac{1}{2}\sum_{C \in {\mathcal C}\text{ equals } x = \pi(y)} \,\sum_{a\in D} w_C \| \mathbf{x}_{\pi(a)} - \mathbf{y}_{a}\|^2 \\ {}+ \sum_{C \in {\mathcal C}\text{ equals }x \in P} w_C \left(\sum_{a\in D\setminus P} \|\mathbf{x}_{a}\|^2 \right) \label{SDP} \end{multline} subject to \begin{align} &\vprod{ \mathbf{x}_{a}}{ \mathbf{y}_{b}} \geq 0 & x,y\in V,\ a,b\in D \label{sdp:pos} \\ &\vprod{ \mathbf{x}_{a}}{ \mathbf{x}_{b}} = 0 & x\in V,\ a,b\in D,\ a\neq b \\ &\sum_{a\in D} \mathbf{x}_{a} = \mathbf{v}_0 & x\in V \label{sdp:sum-to-one}\\ &\|\mathbf{x}_{a} - \mathbf{z}_{c}\|^2 \leq \|\mathbf{x}_{a} - \mathbf{y}_{b}\|^2 + \|\mathbf{y}_{b} - \mathbf{z}_{c}\|^2 \hskip-8em\label{sdp:triangle-ineq}\\ && x,y,z\in V,\ a,b,c\in D\notag\\ &\|\mathbf{v}_0\|^2 = 1.\label{sdp:unitnorm} \end{align} This SDP and the SDP we presented in Section~\ref{sec:SDP} are almost identical. Their objective functions are equal, because of constraints~(\ref{sdp3}) and~(\ref{sdp:sum-to-one}). For convinience, we write the objective function differently in SDP (\ref{SDP})--(\ref{sdp:unitnorm}). The only difference between the SDPs is the presence of the ``triangle inequalities'' (\ref{sdp:triangle-ineq}). We introduce them, because we use the algorithm from~\cite{Charikar06:near} for Unique Games, which assumes that the SDP has triangle inequalities. \fi As discussed before (Lemma \ref{le:prep1}) we can assume that $\eps\geq 1/m^2$ where $m$ is the number of constraints. We solve SDP with error $\delta=1/m^2$ obtaining a solution, denoted $\mathsf{SDP}$, with objective value $O(\epsilon)$. Note that every feasible SDP solution satisfies the following conditions. \begin{align} &\| \mathbf{x}_{a} \|^2 = \vproddot{ \mathbf{x}_{a}}{\bigl(\mathbf{v}_0 - \sum_{b\neq a} \mathbf{x}_{b}\bigr)} = \vproddot{ \mathbf{x}_{a}}{\mathbf{v}_0}-\sum_{b\neq a}\vproddot{\mathbf{x}_a}{\mathbf{x}_b} = \vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 },\label{eq:length}\\[1mm] &\vprod{ \mathbf{x}_{a} }{ \mathbf{y}_{b}} = \vproddot{ \mathbf{x}_{a}}{(\mathbf{v}_0 - \sum_{b'\neq b}\mathbf{y}_{b'})} \label{eq:2SATrequirement} = \|\mathbf{x}_{a}\|^2 - \sum_{b'\neq b} \vprod{ \mathbf{x}_{a}}{ \mathbf{y}_{b'}} \leq \|\mathbf{x}_{a}\|^2, \\[1mm] & \|\mathbf{x}_{a}\|^2 - \|\mathbf{y}_{b}\|^2 = \|\mathbf{x}_{a} - \mathbf{y}_{b}\|^2 + 2(\vprod{\mathbf{x}_{a}}{\mathbf{y}_{b}} - \|\mathbf{y}_{b}\|^2) \leq \|\mathbf{x}_{a} - \mathbf{y}_{b}\|^2, \label{eq:triangle}\\[1mm] & \vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{( \mathbf{v}_0 - \mathbf{y}_{b})} = \vprod{ \sum_{a'\neq a} \mathbf{x}_{a'}} {\sum_{b'\neq b} \mathbf{y}_{b'}} \geq 0.\label{eq:positivity} \end{align} \subsection{Variable Partitioning Step}\label{sec:2pre} In this section, we describe the first step of our algorithm. In this step, we assign values to some variables, partition all variables into three groups ${\cal V}_0$, ${\cal V}_1$ and ${\cal V}_2$, and then split the instance into two sub-instances ${\cal I}_1$ and ${\cal I}_2$. \paragraph{Vertex Partitioning Procedure.} \noindent Choose a number $r \in (0, 1/6)$ uniformly at random. Do the following for every variable $x$. \begin{enumerate} \item Let $D_x = \{a: 1/2 -r < \vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0}\}$. \item Depending on the size of $D_x$ do the following: \begin{enumerate} \item If $|D_x| = 1$, add $x$ to ${\cal V}_0$ and assign $x = a$, where $a$ is the single element of $D_x$. \item If $|D_x| > 1$, add $x$ to ${\cal V}_1$ and restrict $x$ to $D_x$ (see below for details). \item If $D_x = \varnothing$, add $x$ to ${\cal V}_2$. \end{enumerate} \end{enumerate} Note that each variable in ${\cal V}_0$ is assigned a value; each variable $x$ in ${\cal V}_1$ is restricted to a set $D_x$; each variable in ${\cal V}_2$ is not restricted. \begin{lemma} (i) If $\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0} > \frac{1}{2} + r$ then $x\in {\cal V}_0$. (ii) For every $x\in {\cal V}_1$, $|D_x| = 2$. \end{lemma} \begin{proof} (i) Note that for every $b\neq a$, we have $\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0} + \vprod{ \mathbf{x}_{b}}{ \mathbf{v}_0} \leq 1$ and, therefore, $\vprod{\mathbf{x}_{b}}{\mathbf{v}_0} < 1/2 -r$. Hence, $b\notin D_x$. We conclude that $D_x = \{a\}$ and $x\in {\cal V}_0$. \smallskip \noindent(ii) Now consider $x \in {\cal V}_1$. We have, $$ |D_x| < 3(1/2 -r) |D_x| = 3\sum_{a\in D_x} (1/2 -r) \leq 3\sum_{a\in D_x} \vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0} \leq 3. $$ Therefore, $|D_x| \leq 2$. Since $x \in {\cal V}_1$, $|D_x| > 1$. Hence $|D_x| = 2$. \end{proof} We say that an assignment is admissible if it assigns a value in $D_x$ to every $x \in {\cal V}_1$ and it is consistent with the partial assignment to variables in ${\cal V}_0$. From now on we restrict our attention only to admissible assignments. We remove those constraints that are satisfied by every admissible assignment (our algorithm will satisfy all of them). Specifically, we remove the following constraints: \begin{enumerate} \item UG constraints $x=\pi(y)$ with $x, y \in {\cal V}_0$ that are satisfied by the partial assignment; \item disjunction constraints $(x = a) \vee (y = b)$ such that either $x \in {\cal V}_0$ and $x$ is assigned value $a$, or $y \in {\cal V}_0$ and $y$ is assigned value $b$; \item unary constraints $x \in P$ such that either $x \in {\cal V}_0$ and the value assigned to $x$ is in $P$, or $x \in {\cal V}_1$ and $D_x \subseteq P$. \end{enumerate} We denote the set of satisfied constraints by ${\cal C}_s$. Let ${\cal C}'={\cal C}\setminus {\cal C}_s$ be the set of remaining constraints. We now define a set of \textit{violated} constraints --- those constraints that we conservatively assume will not be satisfied by our algorithm (even though some of them might be satisfied by the algorithm). We say that a constraint $C\in {\cal C}'$ is violated if at least one of the following conditions holds: \begin{enumerate} \item $C$ is a unary constraint on a variable $x \in {\cal V}_0 \cup {\cal V}_1$. \item $C$ is a disjunction constraint $(x = a) \vee (y = b)$ and either $x \notin {\cal V}_1$, or $y \notin {\cal V}_1$ (or both). \item $C$ is a disjunction constraint $(x = a) \vee (y = b)$, and $x, y \in {\cal V}_1$, and either $a\notin D_x$, or $b\notin D_y$ (or both). \item $C$ is a UG constraint $x = \pi(y)$, and at least one of the variables $x$, $y$ is in ${\cal V}_0$. \item $C$ is a UG constraint $x = \pi(y)$, and one of the variables $x$, $y$ is in ${\cal V}_1$ and the other is in ${\cal V}_2$. \item $C$ is a UG constraint $x = \pi(y)$, $x, y\in {\cal V}_1$ but $D_x \neq \pi(D_y)$. \end{enumerate} We denote the set of violated constraints by ${\cal C}_v$ and let ${\cal C}'' = {\cal C}'\setminus {\cal C}_v$. \begin{lemma}\label{lem:preproc-violated} $\mathbb{E}[w({\mathcal C}_v)] = O(\eps)$. \end{lemma} \begin{proof} We analyze separately constraints of each type in ${\mathcal C}_v$. \subsubsection*{Unary constraints} A unary constraint $x \in P$ in $\cal C$ is violated if and only if $x \in {\cal V}_0 \cup {\cal V}_1$ and $D_x \not\subseteq P$ (if $D_x \subseteq P$ then $C\in {\cal C}_s$ and thus $C$ is not violated). Thus the SDP contribution of each violated constraint $C$ of the form $x \in P$ is at least $$ w_C\sum_{a\in D\setminus P} \|\mathbf{x}_{a}\|^2 \geq w_C\sum_{a\in D_x\setminus P}\|\mathbf{x}_{a}\|^2 = w_C\sum_{a\in D_x\setminus P} \vproddot{ \mathbf{x}_{a}}{ \mathbf{v}_0} \geq w_C\Bigl(\frac{1}{2} - r\Bigr)\geq \frac{w_C}{3}. $$ The last two inequalities hold because the set $D_x\setminus P$ is nonempty; $\mathbf{x}_{a} \mathbf{v}_0 \geq 1/2-r$ for all $a\in D_x$ by the construction; and $r\leq 1/6$. Therefore, the expected total weight of violated unary constraints is at most $3\,\mathsf{SDP}= O(\eps)$. \subsubsection*{Disjunction constraints} Consider a disjunction constraint $(x = a) \vee (y = b)$. Denote it by $C$. Assume without loss of generality that $\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 } \geq \vprod{ \mathbf{y}_{b}}{ \mathbf{v}_0}$. Consider several cases. If $\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 } > 1/2 +r$ then $x \in {\cal V}_0$ and $x$ is assigned value $a$. Thus, $C$ is satisfied. If $\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 } \leq 1/2 +r$ and $\vprod{ \mathbf{y}_{b}}{ \mathbf{v}_0} > 1/2 -r$ then we also have $\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0} > 1/2 - r$ and hence $x, y\in {\cal V}_0 \cup {\cal V}_1$ and $a\in D_x$, $b\in D_y$. Thus, $C$ is not violated (if at least one of the variables $x$ and $y$ is in ${\cal V}_0$, then $C\in {\cal C}_s$; otherwise, $C \in {\cal C}'$). Therefore, $C$ is violated only if $$\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 } \leq 1/2 +r \text{ and } \vprod{ \mathbf{y}_{b}}{ \mathbf{v}_0} \leq 1/2 -r ,$$ or equivalently, \begin{equation} \vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 } - 1/2 \leq r \leq 1/2 - \vprod{ \mathbf{y}_{b}}{ \mathbf{v}_0}.\label{eq:bad-event} \end{equation} Since we choose $r$ uniformly at random in $(0,1/6)$, the probability density of the random variable $r$ is 6 on $(0,1/6)$. Thus the probability of event (\ref{eq:bad-event}) is at most \begin{multline*} 6 \max\Bigl(\bigl(( 1/2 - \vprod{ \mathbf{y}_{b}}{ \mathbf{v}_0}\bigr)- \bigl(\vprod{ \mathbf{x}_{a}}{ \mathbf{v}_0 } - 1/2)\bigr),0\Bigr) = 6\max\Bigl( \vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})} -\vprod{\mathbf{x}_{a}}{\mathbf{y}_{b}},0\Bigr)\\ {}\stackrel{\text{by (\ref{sdp1}) and (\ref{eq:positivity})}}{\leq} 6\vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})}. \end{multline*} The expected weight of violated constraints is at most, $$ \sum_{\substack{C\in{\mathcal C} \text{ equals }\\(x = a) \vee (y = b)}} 6w_C\vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})} \leq 6\,\mathsf{SDP} = O(\eps). $$ \subsubsection*{UG constraints} Consider a UG constraint $x = \pi(y)$. Assume that it is violated. Then $D_x \neq \pi(D_y)$ (note that if $x$ and $y$ do not lie in the same set ${\cal V}_t$ then $|D_x| \neq |D_y|$ and necessarily $D_x \neq \pi(D_y)$). Thus, at least one of the sets $\pi(D_y) \setminus D_x$ or $D_x \setminus \pi(D_y)$ is not empty. If $\pi(D_y) \setminus D_x\neq \varnothing$, there exists $c \in \pi(D_y) \setminus D_x$. We have, \begin{align*} \Prob{c \in \pi(D_y) \setminus D_x} &\leq \Prob{\|\mathbf{y}_{\pi^{-1}(c)}\|^2 > 1/2 - r \text{ and } \|\mathbf{x}_{c}\|^2 \leq 1/2 - r} \\ &= \Prob{1/2 - \|\mathbf{y}_{\pi^{-1}(c)}\|^2 < r \leq 1/2 -\|\mathbf{x}_{c}\|^2} \\ &\leq 6\max(\|\mathbf{y}_{\pi^{-1}(c)}\|^2 - \|\mathbf{x}_{c}\|^2, 0) \\ &\stackrel{\text{by (\ref{eq:triangle})}}{\leq} 6\|\mathbf{y}_{\pi^{-1}(c)} - \mathbf{x}_{c}\|^2. \end{align*} By the union bound, the probability that there is $c \in \pi(D_y) \setminus D_x$ is at most $6\sum_{c\in D}\|\mathbf{y}_{\pi^{-1}(c)} - \mathbf{x}_{c}\|^2=6\sum_{b\in D}\|\mathbf{y}_{b} - \mathbf{x}_{\pi(b)}\|^2$. Similarly, the probability that there is $b \in D_x \setminus \pi(D_y)$ is at most $6\sum_{b\in D}\|\mathbf{y}_{b} - \mathbf{x}_{\pi(b)}\|^2$. Therefore, the probability that the constraint $x=\pi(y)$ is violated is upper bounded by $12\sum_{b\in D}\|\mathbf{y}_{b} - \mathbf{x}_{\pi(b)}\|^2$. Consequently, the total expected weight of all violated UG constraints is at most \begin{multline*}\sum_{C \in {\mathcal C}\text{ equals } x = \pi(y)} w_C \left(12 \,\sum_{b\in D} \| \mathbf{x}_{\pi(b)} - \mathbf{y}_{b}\|^2\right) \\ = 24 \times \left(\frac12 \sum_{C \in {\mathcal C}\text{ equals } x = \pi(y)} w_C \,\sum_{b\in D} \| \mathbf{x}_{\pi(b)} - \mathbf{y}_{b}\|^2\right) \leq 24\,\mathsf{SDP} = O(\eps), \end{multline*} here we bound the value of the SDP by the second term of the objective function~(\ref{SDP}). \end{proof} \noindent We restrict our attention to the set ${\cal C}''$. There are four types of constraints in ${\cal C}''$. \begin{enumerate} \item disjunction constraints $(x = a) \vee (y = b)$ with $x, y \in {\cal V}_1$ and $a\in D_x$, $b\in D_y$; \item UG constraints $x = \pi (y)$ with $x, y \in {\cal V}_1$ and $D_x = \pi(D_y)$; \item UG constraints $x = \pi (y)$ with $x, y \in {\cal V}_2$; \item unary constraints $x \in P$ with $x\in{\cal V}_2$. \end{enumerate} Denote the set of type 1 and 2 constraints by ${\cal C}_1$, and type 3 and 4 constraints by ${\cal C}_2$. Let ${\cal I}_1$ be the sub-instance of ${\cal I}$ on variables ${\cal V}_1$ with constraints ${\cal C}_1$ in which every variable $x$ is restricted to $D_x$, and ${\cal I}_2$ be the sub-instance of ${\cal I}$ on variables ${\cal V}_2$ with constraints ${\cal C}_2$. In Sections~\ref{solve-i1} and~\ref{solve-i2}, we show how to solve ${\cal I}_1$ and ${\cal I}_2$, respectively. The total weight of constraints violated by our solution for ${\cal I}_1$ will be at most $O(\sqrt{\eps})$; The total weight of constraints violated by our solution for ${\cal I}_2$ will be at most $O(\sqrt{\eps\log{|D|} })$. Thus the combined solution will satisfy a subset of the constraints of weight at least $1 - O(\sqrt{\eps \log{|D|}})$. \subsection{Solving Instance \texorpdfstring{${\cal I}_1$}{I1}}\label{solve-i1} In this section, we present an algorithm that solves instance ${\cal I}_1$. The algorithm assigns values to variables in ${\cal V}_1$ so that the total weight of violated constraints is at most $O(\sqrt{\eps})$. \begin{lemma}\label{lem:consistentUG} There is a randomized algorithm that, given instance ${\cal I}_1$ and the SDP solution $\{\mathbf{x}_a\}$ for ${\cal I}$, finds a set of UG constraints ${\cal C}_{bad}\subseteq {\cal C}_1$ and values $\alpha_x,\beta_x \in D_x$ for every $x\in {\cal V}_1$ such that the following conditions hold. \begin{itemize} \item $D_x = \{\alpha_x, \beta_x\}$. \item for each UG constraint $x=\pi(y)$ in ${\cal C}_1\setminus {\cal C}_{bad}$, we have $\alpha_x =\pi(\alpha_y)$ and $\beta_x =\pi(\beta_y)$. \item The expected weight of ${\cal C}_{bad}$ is $O(\sqrt{\eps}).$ \end{itemize} \end{lemma} \begin{proof} We use the algorithm of Goemans and Williamson for Min Uncut~\cite{Goemans95:improved} to find values~$\alpha_x$, $\beta_x$. Recall that in the Min Uncut problem (also known as Min 2CNF$\equiv$ deletion) we are given a set of Boolean variables and a set of constraints of the form $(x = a) \leftrightarrow (y = b)$. Our goal is to find an assignment that minimizes the weight of unsatisfied constraints. Consider the set of UG constraints in ${\cal C}_1$. Since $|D_x| = 2$ for every variable $x \in {\cal V}_1$, each constraint $x = \pi(y)$ is equivalent to the Min Uncut constraint $(x =\pi(a)) \leftrightarrow (y =a)$ where $a$ is an element of $D_y$ (it does not matter which of the two elements of $D_y$ we choose). We define an SDP solution for the Goemans---Williamson relaxation of Min Uncut as follows. Consider $x \in {\cal V}_1$. Denote the elements of $D_x$ by $a$ and $b$ (in any order). Let $$\mathbf{x}^*_{a} = \frac{\mathbf{x}_{a}-\mathbf{x}_{b}}{\|\mathbf{x}_{a}-\mathbf{x}_{b}\|} \quad\text{and}\quad \mathbf{x}^*_{b} = - \mathbf{x}^*_{a} = \frac{\mathbf{x}_{b}-\mathbf{x}_{a}}{\|\mathbf{x}_{a}-\mathbf{x}_{b}\|}.$$ Note that the vectors $\mathbf{x}_a$ and $\mathbf{x}_b$ are nonzero orthogonal vectors, and, thus, $\|\mathbf{x}_a - \mathbf{x}_b\|$ is nonzero. The vectors $\mathbf{x}^*_{a}$ and $\mathbf{x}^*_{b}$ are unit vectors. Now we apply the random hyperplane rounding scheme of Goemans and Williamson: We choose a random hyperplane and let $H$ be one of the half-spaces the hyperplane divides the space into. Note that for every $x$ exactly one of the two antipodal vectors in $\{\mathbf{x}^*_{a}:a\in D_x\}$ lies in $H$ (almost surely). Define $\alpha_x$ and $\beta_x$ so that $\mathbf{x}^*_{\alpha_x}\in H$ and $\mathbf{x}^*_{\beta_x}\notin H$. Let ${\cal C}_{bad}$ be the set of UG constraints such that $\alpha_x \neq \pi(\alpha_y)$, or equivalently $\mathbf{x}_{\pi(\alpha_y)}^* \notin H$. Values $\alpha_x$ and $\beta_x$ satisfy the first condition. If a UG constraint $x=\pi(y)$ is in ${\cal C}_1\setminus {\cal C}_{bad}$, then $\alpha_x = \pi(\alpha_y)$; also since $D_x = \pi(D_y)$, $\beta_x = \pi(\beta_y)$. So the second condition holds. Finally, we verify the last condition. Consider a constraint $x = \pi(y)$. Let $\mathbf{A} = \mathbf{x}_{\pi(\alpha_y)} - \mathbf{x}_{\pi(\beta_y)}$ and $\mathbf{B} = \mathbf{y}_{\alpha_y} - \mathbf{y}_{\beta_y}$. Since $x\in {\cal V}_1$, we have $\|\mathbf{x}_{\pi(\alpha_y)}\|^2 > 1/2 -r > 1/3$ and $\strut\|\mathbf{x}_{\pi(\beta_y)}\|^2 > 1/3$. Hence $\|\mathbf{A}\|^2 = \|\mathbf{x}_{\pi(\alpha_y)}\|^2 + \|\mathbf{x}_{\pi(\beta_y)}\|^2 > 2/3$. Similarly, $\|\mathbf{B}\|^2> 2/3$. Assume first that $\|\mathbf{A}\| \geq \|\mathbf{B}|$. Then, \begin{align*} \|\mathbf{x}^*_{\pi(\alpha_y)} - \mathbf{y}^*_{\alpha_y}\|^2 &= \left\|\frac{\mathbf{A}}{\|\mathbf{A}\|} - \frac{\mathbf{B}}{\|\mathbf{B}\|}\right\|^2 = 2 - \frac{2\vprod{\mathbf{A}}{\mathbf{B}}}{\|\mathbf{A}\| \|\mathbf{B}\|} \\ &=\frac{2}{\|\mathbf{B}\|^2} \times \left( \|\mathbf{B}\|^2 - \frac{\|\mathbf{B}\|}{\|\mathbf{A}\|} \, \vprod{\mathbf{A}}{\mathbf{B}}\right). \end{align*} We have $2\Bigl(\|\mathbf{B}\|^2 - \frac{\|\mathbf{B}\|}{\|\mathbf{A}\|}\,\vprod{\mathbf{A}}{\mathbf{B}} \Bigr) \leq \|\mathbf{A} -\mathbf{B}\|^2$, since $$ \|\mathbf{A} -\mathbf{B}\|^2 - 2\Bigl(\|\mathbf{B}\|^2 - \frac{\|\mathbf{B}\|}{\|\mathbf{A}\|}\,\vprod{\mathbf{A}}{\mathbf{B}} \Bigr) = \Bigl(\|\mathbf{A}\|-\|\mathbf{B}\|\Bigr)\Bigl(\|\mathbf{A}\|+\|\mathbf{B}\| - \frac{2 \vprod{\mathbf{A}}{\mathbf{B}}}{\|\mathbf{A}\|}\Bigr) \geq 0, $$ because $\|\mathbf{A}\| \geq \vprod{\mathbf{A}}{\mathbf{B}}/\|\mathbf{A}\|$ and $\|\mathbf{B}\| \geq \vprod{\mathbf{A}}{\mathbf{B}}/\|\mathbf{A}\|$. We conclude that \begin{align*} \|\mathbf{x}^*_{\pi(\alpha_y)} &- \mathbf{y}^*_{\alpha_y}\|^2 \leq \frac{\|\mathbf{A}-\mathbf{B}\|^2}{\|\mathbf{B}\|^2} \leq \frac{3}{2}\|\mathbf{A}-\mathbf{B}\|^2\\ &= \frac{3}{2}\, \|(\mathbf{x}_{\pi(\alpha_y)} -\mathbf{y}_{\alpha_y}) - (\mathbf{x}_{\pi(\beta_y)} - \mathbf{y}_{\beta_y}) \|^2 \\ &\leq 3\, \|\mathbf{x}_{\pi(\alpha_y)} -\mathbf{y}_{\alpha_y}\|^2 + 3\, \|\mathbf{x}_{\pi(\beta_y)} - \mathbf{y}_{\beta_y}\|^2. \end{align*} If $\|\mathbf{A}\|\leq \|\mathbf{B}\|$, we get the same bound on $\|\mathbf{x}^*_{\pi(\alpha_y)} - \mathbf{y}^*_{\alpha_y}\|^2$ by swapping $\mathbf{A}$ and $\mathbf{B}$ in the formulas above. Therefore, \[ \sum_{\substack{C\in{\cal C}_1 \\ \text{ is of the form } \\ x = \pi(y)}} w_C\|\mathbf{x}^*_{\pi(\alpha_y)} - \mathbf{y}^*_{\alpha_y}\|^2\leq 3\, \mathsf{SDP} = O(\eps). \] The analysis by Goemans and Williamson shows that the expected total weight of the constraints of the form $x = \pi(y)$ such that $$\mathbf{x}^*_{\pi(\alpha_y)}\notin H \text{ and } \mathbf{y}^*_{\alpha_y}\in H$$ is at most $O(\sqrt{\eps})$, see Section 3 in~\cite{Goemans95:improved} for the original analysis or Section 2 in survey~\cite{MM17:cspsurvey} for presentation more closely aligned with our notation. Therefore, the expected total weight of ${\cal C}_{bad}$ is $O(\sqrt{\eps})$. \end{proof} We remove all constraints ${\cal C}_{bad}$ from ${\cal I}_1$ and obtain an instance ${\cal I}_1'$ (with the domain for each variable $x$ now restricted to $D_x$). We construct an SDP solution $\{\mathbf{\tilde x}_{a}\}$ for ${\cal I}_1'$. We let $$\mathbf{\tilde x}_{\alpha_x} = \mathbf{x}_{\alpha_x} \quad\text{and}\quad \mathbf{\tilde x}_{\beta_x} = \mathbf{v}_0 - \mathbf{x}_{\alpha_x}.$$ We define $S_{x\alpha_x} = \{\alpha_x\}$ and $S_{x\beta_x} = D\setminus S_{x\alpha_x}$. Since $\mathbf{\mathbf{\tilde x}}_{\beta_x} = \mathbf{v}_0 - \mathbf{x}_{\alpha_x} = \mathbf{x}_{S_{x\beta_x}}$, we have, \begin{equation} \mathbf{\tilde x}_{a} = \mathbf{x}_{S_{xa}} \quad \text{for every } a\in D_x.\label{eq:redef-x-a} \end{equation} Note that $a\in S_{xa}$ for every $a\in D_x$. \begin{lemma}\label{lem:new-SDP-feasible} The solution $\{\mathbf{\tilde x}_{a}\}$ is a feasible solution for SDP relaxation (\ref{sdpobj})--(\ref{sdp4}) for ${\cal I}'_1$. Its cost is $O(\eps)$. \end{lemma} \begin{proof} We verify that the SDP solution is feasible. First, we have $\sum_{a\in D_x} \mathbf{\tilde x}_{a} = \mathbf{v}_0$ and $$\vprod{ \mathbf{\tilde x}_{\alpha_x}}{ \mathbf{\tilde x}_{\beta_x}} = \vproddot{ \mathbf{x}_{\alpha_x}}{ (\mathbf{v}_0 - \mathbf{x}_{\alpha_x})} = \vprod{ \mathbf{x}_{\alpha_x}}{ \mathbf{v}_0 } - \|\mathbf{x}_{\alpha_x}\|^2 = 0.$$ Then for $a\in D_x$ and $b\in D_y$, we have $\vprod{ \mathbf{\tilde x}_{a}}{ \mathbf{\tilde y}_{b}} = \sum_{a'\in S_{xa}, b'\in S_{yb}} \vprod{ \mathbf{x}_{a'}}{ \mathbf{y}_{b'}} \geq 0$. We now show that the SDP cost is $O(\eps)$. First, we consider disjunction constraints. We prove that the contribution of each constraint $(x = a) \vee (y =b)$ to the SDP for ${\cal I}_1'$ is at most its contribution to the SDP for $\cal I$. That is, \begin{equation} \vprod{(\mathbf{v}_0 - \mathbf{\tilde x}_{a})}{(\mathbf{v}_0 - \mathbf{\tilde y}_{b})} \leq \vprod{ (\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})}. \label{ineq:SDP} \end{equation} Observe that $(\mathbf{v}_0 - \mathbf{\tilde x}_{a}) = \mathbf{x}_{D\setminus S_{xa}}$, $(\mathbf{v}_0 - \mathbf{\tilde y}_{b}) = \mathbf{y}_{D\setminus S_{yb}}$, $(\mathbf{v}_0 - \mathbf{x}_{a}) = \mathbf{x}_{D\setminus \{a\}}$, and $(\mathbf{v}_0 - \mathbf{y}_{b}) = \mathbf{y}_{D\setminus \{b\}}$. Then, $D\setminus S_{xa}\subseteq D\setminus\{a\}$ and $D\setminus S_{yb}\subseteq D\setminus\{b\}$. Therefore, by (\ref{sdp1}), \begin{multline*} \vprod{(\mathbf{v}_0 - \mathbf{\tilde x}_{a})}{(\mathbf{v}_0 - \mathbf{\tilde y}_{b})} = \sum_{(a',b')\in (D\setminus S_{xa})\times (D\setminus S_{yb})} \vprod{\mathbf{x}_{a'}}{\mathbf{y}_{b'}} \leq \\ \leq \sum_{(a',b')\in (D\setminus \{a\})\times (D\setminus \{b\})}\vprod{\mathbf{x}_{a'}}{\mathbf{y}_{b'}} = \vprod{ (\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})}. \end{multline*} \iffalse Observe that \begin{multline*} \vprod{(\mathbf{v}_0 - \mathbf{x}_{a})}{(\mathbf{v}_0 - \mathbf{y}_{b})} - \vprod{(\mathbf{v}_0 - \mathbf{\tilde x}_{a})}{(\mathbf{v}_0 - \mathbf{\tilde y}_{b})} = \\ \vprod{(\mathbf{v}_0 - \mathbf{\tilde x}_{a})}{(\mathbf{\tilde y}_{b} - \mathbf{y}_{b})} + \vprod{{(\mathbf{\tilde x}_{a}- \mathbf{x}_{a})}(\mathbf{v}_0 - \mathbf{\tilde y}_{b})} \\ {}+ \vprod{(\mathbf{\tilde x}_a - \mathbf{x}_{a})}{(\mathbf{\tilde y}_b - \mathbf{y}_{b})} . \end{multline*} We prove that all terms on the right hand side are nonnegative, and thus inequality (\ref{ineq:SDP}) holds. Using the identities (\ref{eq:redef-x-a}) and $\sum_{a'\in D} \mathbf{x}_{a'} = \mathbf{v}_{0}$ as well as the inequality $\mathbf{x}_{a'}\mathbf{y}_{b'}\geq 0$ (for all $a',b'\in D$), we get $$ \vprod{(\mathbf{v}_0 - \mathbf{\tilde x}_{a})}{(\mathbf{\tilde y}_{b} - \mathbf{y}_{b})} = \sum_{\substack{a'\in D\setminus S_{xa}\\b'\in S_{yb}\setminus \{b\}}} \mathbf{x}_{a'}\mathbf{y}_{b'} \geq 0. $$ Similarly, $\vprod{{(\mathbf{\tilde x}_{a}- \mathbf{x}_{a})}(\mathbf{v}_0 - \mathbf{\tilde y}_{b})}\geq 0$, and $$ \vprod{(\mathbf{\tilde x}_a - \mathbf{x}_{a})}{(\mathbf{\tilde y}_b - \mathbf{y}_{b})}= \sum_{\substack{a'\in S_{xa}\setminus \{a\}\\b'\in S_{yb}\setminus \{b\}}} \mathbf{x}_{a'}\mathbf{y}_{b'} \geq 0. $$ \fi Now we consider UG constraints. The contribution of a UG constraint $x = \pi(y)$ in ${\cal C}_1 \setminus {\cal C}_{bad}$ to the SDP for ${\cal I}_1'$ equals the weight of the constraint times the following expression. \begin{multline*} \|\mathbf{\tilde x}_{\pi(\alpha_y)} - \mathbf{\tilde y}_{\alpha_y}\|^2 + \|\mathbf{\tilde x}_{\pi(\beta_y)} - \mathbf{\tilde y}_{\beta_y}\|^2 \\ = \|\mathbf{\tilde x}_{\alpha_x} - \mathbf{\tilde y}_{\alpha_y}\|^2 + \|\mathbf{\tilde x}_{\beta_x} - \mathbf{\tilde y}_{\beta_y}\|^2 \\ = \|\mathbf{x}_{\alpha_x} - \mathbf{y}_{\alpha_y}\|^2 + \|(\mathbf{v}_0 - \mathbf{x}_{\alpha_x}) - (\mathbf{v}_0-\mathbf{y}_{\alpha_y})\|^2 \\ = 2 \|\mathbf{x}_{\alpha_x} - \mathbf{y}_{\alpha_y}\|^2 = 2 \|\mathbf{x}_{\pi(\alpha_y)} - \mathbf{y}_{\alpha_y}\|^2. \end{multline*} Thus, by the choice of $\alpha_x$ and $\alpha_y$ (Lemma \ref{lem:consistentUG}) the contribution is at most twice the contribution of the constraint to the SDP for ${\cal I}$. We conclude that the SDP contribution of all the constraints in ${\cal C}_1 \setminus {\cal C}_{bad}$ is at most $2\,\mathsf{SDP} = O(\eps)$. \end{proof} Finally, we note that ${\cal I}_1'$ is a Boolean 2-CSP instance. We round solution $\{\mathbf{\tilde x}_{a}\}$ using the rounding procedure by Charikar et al. for Boolean 2-CSP~\cite{Charikar09:near} (when $|D| = 2$, the SDP relaxation used in~\cite{Charikar09:near} is equivalent to SDP (\ref{sdpobj})--(\ref{sdp4})). We get an assignment of variables in ${\cal V}_1$. The weight of constraints in ${\cal C}_{1}\setminus {\cal C}_{bad}$ violated by this assignment is at most $O(\sqrt{\eps})$. Since $w({\cal C}_{bad}) = O(\sqrt{\eps})$, the weight of constraints in ${\cal C}_{1}$ violated by the assignment is at most $O(\sqrt{\eps})$. \subsection{Solving Instance \texorpdfstring{${\cal I}_2$}{I2}}\label{solve-i2} Instance ${\cal I}_2$ is a unique games instance with additional unary constraints. We restrict the SDP solution for $\cal I$ to variables $x \in {\cal V}_2$ and get a solution for the unique game instance ${\cal I}_2$. Note that since we do not restrict the domain of variables $x\in \VV{2}$ to $D_x$, the SDP solution we obtain is feasible. The SDP cost of this solution is at most $\mathsf{SDP}$. We round this SDP solution using a variant of the algorithm by Charikar et al.~\cite{Charikar06:near} that is presented in Section 3 of the survey~\cite{MM17:cspsurvey}; this variant of the algorithm does not need $\ell_2^2$-triangle-inequality SDP constraints. Given a $(1-\eps)$-satisfiable instance of Unique Games, the algorithm finds a solution with the weight of violated constraints at most $O(\sqrt{\eps \log{|D|}})$. We remark that paper~\cite{Charikar06:near} considers only unique game instances. However, in~\cite{Charikar06:near}, we can restrict the domain of any variable $x$ to a set $S_x$ by setting $\mathbf{x}_a = 0$ for $a\in D\setminus S_x$. Hence, we can model unary constraints as follows. For every unary constraint $x\in P$, we introduce a dummy variable $z_{x,P}$ and restrict its domain to the set $P$. Then we replace each constraint $x\in P$ with the equivalent constraint $x = z_{x,P}$. The weight of the constraints violated by the obtained solution is at most $O(\sqrt{\eps \log{|D|}})$. \medskip \noindent Finally, we combine results proved in Sections~\ref{sec:2pre}, \ref{solve-i1}, and~\ref{solve-i1} and obtain Theorem~\ref{the:main}(2).
3831f5d43651b93443f7b97153ed8ff9384f90a1
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:Introduction} Topological insulators (TIs) form a class of materials with unique properties, associated with a non-trivial topology of their quasiparticle band structure (for a review, see Refs.~\cite{Zhang:rev, Hasan-Kane:rev, Hasan-Moore:rev, Ando:rev}). The key feature of two-dimensional (2D) and three-dimensional (3D) TIs is the existence of special gapless edge and surface states, respectively, while the bulk states of those materials are gapped. The hallmark property of the surface states is their topological protection. Mathematically, the nontrivial topological properties of time-reversal (TR) invariant TIs are generally described \cite{Moore:2007} by multiple copies of the $Z_2$ invariants found by Kane and Mele \cite{Kane-Mele}. This implies that the energy band gap should close at the boundary between topological and trivial insulator (e.g., vacuum) giving rise to the occurrence of the gapless interface states and the celebrated bulk-boundary correspondence. The discovery of the $Z_2$ topology in TIs is an important breakthrough because it showed that nontrivial topology can be embedded in the band structure and that the presence of an external magnetic field is not mandatory for the realization of topological phases. Another distinctive feature of the 3D TIs is a relativistic-like energy spectrum of the surface states, whose physical origin is related to a strong spin-orbit coupling \cite{Hsieh:2009}. Indeed, the surface states on each of the surfaces are described by 2D massless Dirac fermions in an irreducible 2$\times$2 representation, with a single Dirac point in the reciprocal space. For comparison, quasiparticles in graphene demonstrate similar properties, but have four inequivalent Dirac cones due to a spin and valley degeneracy \cite{Castro:2009} that makes certain aspects of their physics very different from those of the surface states in TIs. In our study below, we will concentrate only on the case of the strong 3D TIs whose surface states are protected by the topology of the bulk bands in combination with the TR symmetry. This leads to the locking of momenta and spin degrees of freedom and, consequently, to the formation of a helical Dirac (semi)metal state \cite{Hsieh:2009}. Such a state is characterized by the electron antilocalization and the absence of backscattering. The phenomenon of antilocalization has deep mathematical roots and is usually explained by an additional Berry's phase $\pi$ that is acquired when an electron circles a Dirac point. From the physical viewpoint, when scattering on an impurity, an electron must change its spin in order to preserve its chirality. Such a process is possible only in the case of magnetic impurities which break explicitly the TR symmetry. Experimentally, a linear relativistic-like dispersion law of the surface states is observed in Bi$_{1-x}$Sb$_x$, Bi$_2$Se$_3$, Bi$_2$Te$_3$, Sb$_2$Te$_3$, Bi$_2$Te$_2$Se, and other materials by using angle resolved photoemission spectroscopy (ARPES) \cite{Hsieh:2008, Zhang:2009, Hsieh:2009, Chen:2009, Cava-Hasan}. Furthermore, scanning tunneling microscopy and scanning tunneling spectroscopy provide additional information about the topological nature of the surface states, such as the quasiparticles interference patterns around impurities and defects. The Fourier analysis of these patterns has shown that the backscattering between $\mathbf{k}$ and $-\mathbf{k}$ is highly suppressed in Bi$_{1-x}$Sb$_x$ \cite{Roushan:2009} and Bi$_2$Te$_3$ \cite{Zhang-Cheng:2009} in accord with the TR symmetry protection. The existence of an odd number of Dirac nodes leads to other exotic properties associated with surface states of TIs, e.g., an axion electromagnetic response \cite{Qi:2008}, an unusual surface Hall conductance \cite{Fu:2007-76,Xu:2014}, etc. It is well known that electrons confined to two dimensions can form numerous interaction-induced phases. By using numerical calculations, it was shown in Ref.~\cite{Koulakov:1996} that it is energetically favorable for the 2D electron liquid in a weak magnetic field to form domains with empty and fully filled higher Landau levels. Depending on the number of Landau levels filled, the corresponding charge density wave (CDW) phase is realized with a ``stripe'' or ``bubble'' pattern. By using the simplest model of the surface states in a magnetic field with strong local repulsion and a long-range Coulomb interaction included perturbatively, it was suggested that a similar CDW phase with a ``stripe'' or ``bubble'' pattern can be also realized on the surface of 3D TIs for supercritical values of a local repulsion constant \cite{Vishwanath:2010}. For subcritical local repulsion, the composite Fermi liquid (CFL) \cite{HLR} phase is expected \cite{Rezayi:2000, Vishwanath:2010}. It is interesting that composite fermions in conventional 2D electron gas at half-filling were recently suggested to be massless Dirac (quasi)particles \cite{Son:2015} similar to the surface quasiparticles of TIs. This result was also checked numerically in Ref.~\cite{Geraedts:2016}, where it was shown that at the half-filling the particle-hole symmetry for composite fermions plays the same role as the TR symmetry for the 2D Dirac fermions and, consequently, the backscattering off symmetry preserving impurities is also forbidden. The influence of an external electric field on the exciton condensation in thin films of TIs was studied in Refs.~\cite{Franz:2009, Efimkin:2012}, where it was shown that the electron condensate effectively joins the surfaces of a thin film and leads to the formation of a pairing gap. However, this is important only in thin ($l_z \lesssim 8~\mbox{nm}$) films of TIs and can be ignored in sufficiently thick slabs \cite{Linder:2009}. The exciton condensate exhibits unusual properties including a stable zero mode and a fractional charge $\pm e/2$ carried by a singly quantized vortex in the exciton condensate \cite{Franz:2009}. The dynamical gap generation in a simple model of TIs was also considered in Ref.~\cite{Cea:2016}. Just like a magnetic field, an external electric field may play an important role in the dynamics of the surface states in a 3D TI slab. In this paper, we study the dynamical gap generation and the phase diagram of a TI slab placed in external magnetic and electric fields perpendicular to the slab surfaces. (Note that the case of the parallel fields is rather trivial. While a parallel magnetic field does not affect the orbital motion, a parallel electric field produces a current on the surface.) We argue that a uniform phase with both dynamically generated Dirac and Haldane gaps is realized in sufficiently strong (weak) electric (magnetic) fields. Although the explicit calculations performed in this paper use the model parameters suitable for Bi$_2$Se$_3$, the main qualitative conclusions should be valid for all similar TIs. The paper is organized as follows. The effective Hamiltonian of the surface states in the simplest model of a topological insulator with short- and long-range interactions is described in Sec.~\ref{sec:model-all}. The set of gap equations at finite temperature is derived in Sec.~\ref{sec:gap-equation-Coulomb} and its solutions in electric and magnetic fields are obtained numerically in Sec.~\ref{sec:gap-equation-Coulomb-results-kappa}. The qualitative description of the inhomogeneous phase with two stripes is given in Sec.~\ref{sec:stripe}. The main results are discussed and summarized in Secs.~\ref{sec:Discussion} and \ref{sec:Conclusion}, respectively. For convenience, throughout the paper, we set $\hbar=c=1$. \section{Model} \label{sec:model-all} By projecting the 3D bulk Hamiltonian onto the subspace of surface states (see Refs.~\cite{Zhang:rev, Shen:book} for a detailed consideration), the following effective Hamiltonian for the top surface of a 3D TI is obtained \cite{Zhang:2009,Shan:2010, Zhang:rev,Shen:book}: \begin{eqnarray} \mathcal{H}_{\mathrm{top}\,\,\mathrm{surf}}(\mathbf{k})=C+v_F\left(\bm{\sigma}\times\mathbf{k}\right)_z +\mathcal{O}\left(\mathbf{k}^2\right) = C+\left( \begin{array}{cc} 0 & v_F(ik_x+k_y) \\ v_F(-ik_x+k_y) & 0 \\ \end{array} \right)+ \mathcal{O}\left(\mathbf{k}^2\right), \label{model-H-2D} \end{eqnarray} where $C$ is a constant, $\bm{\sigma}=\left(\sigma_x, \sigma_y\right)$ are the Pauli matrices, $v_F=4.1~\mbox{eV\AA}=6.2\times10^5~\mbox{m/s}$ is the Fermi velocity, and $\mathbf{k}=\left(k_x, k_y\right)$ is the surface momentum. The effective surface Hamiltonian for the bottom surface is obtained from the Hamiltonian of the top surface by the inversion $\mathbf{k} \to -\mathbf{k}$ (see Sec.~III.C in Ref.~\cite{Zhang:rev}). It is worth noting that the effective surface Hamiltonian is valid only at sufficiently small chemical potentials, when the bulk states are gapped. Therefore, the corresponding energy cutoff can be approximated by the bulk band gap, i.e., $\Lambda\simeq \Delta_{\rm bulk}$. In the case of Bi$_2$Se$_3$, for example, $\Delta_{\rm bulk} \approx 0.35~\mbox{eV}$ \cite{Mooser, Black}. The resulting model Hamiltonian, describing quasiparticle states on the top and bottom surfaces of the 3D TI in constant electric and magnetic fields applied perpendicular to the surfaces of the slab, is given by $H^{(0)}=H^{(0)}_{+}\oplus H^{(0)}_{-}$, where \begin{equation} H^{(0)}_{\lambda}=\int d^2\mathbf{r}\,\,\psi^{\dag}_{\lambda}(\mathbf{r})\left( \begin{array}{cc} m^{(0)} -\mu_{\lambda}^{(0)} & iv_F(\pi_x-i\pi_y) \\ -iv_F(\pi_x +i\pi_y) & -m^{(0)} -\mu_{\lambda}^{(0)} \\ \end{array} \right)\psi_{\lambda}(\mathbf{r}). \label{model-H-s-0} \end{equation} Here $\lambda=\pm$ denotes the top and bottom surfaces, respectively, $\mu_{\lambda }^{(0)}$ is the surface electrochemical potential, $\bm{\pi} \equiv -i \bm{\nabla} + e\mathbf{A}$ is the canonical momentum, $\mathbf{A}=\left(0, Bx\right)$ is the vector potential that describes the constant magnetic field $\mathbf{B}$ pointing in the $z$ direction, and $e$ is the electron charge. Note that in Eq.~(\ref{model-H-s-0}) we redefined the wave function on the bottom surface by replacing $\psi_{-}\to\sigma_z\psi_{-}$. As the notation suggests, the value of $\mu^{(0)}_{\lambda}$ may depend on the surface index $\lambda$. Indeed, this is quite natural in the model at hand since fixing charge densities on the top and bottom surfaces requires an introduction of the corresponding local electrochemical potentials. In view of a large surface $g$-factor, $g_{s}=18\pm4$ \cite{Fu:2016}, the Zeeman splitting is important in TIs. This spin splitting is included in Hamiltonian (\ref{model-H-s-0}) as the bare gap parameter $m^{(0)} =g_s\mu_{B} B/2$, where $\mu_{B}=5.788 \times 10^{-5}~\mbox{eV/ T}$ is the Bohr magneton. Before proceeding with the analysis of the model, it is convenient to rewrite the model Hamiltonian (\ref{model-H-s-0}) in terms of the Dirac matrices. It is well known that there are two irreducible representations of the Clifford-Dirac algebra in (2+1)-dimensions, e.g., see Ref.~\cite{Appelquist:1986}. One of them is \begin{equation} \gamma^0=\sigma_z, \quad \gamma^1=i\sigma_x, \quad \gamma^2=i\sigma_y \label{model-gamma-matrices} \end{equation} and the other irreducible representation is obtained by changing $\gamma^{\mu}\to-\gamma^{\mu}$ with $\mu=0, 1, 2$ in Eq.~(\ref{model-gamma-matrices}). In terms of the Dirac matrices (\ref{model-gamma-matrices}), the free Hamiltonian (\ref{model-H-s-0}) takes the following form: \begin{equation} H^{(0)}_{\lambda}=\int d^2\mathbf{r}\,\,\bar{\psi}_{\lambda}(\mathbf{r})\left( -\mu^{(0)}_{\lambda}\gamma^0+v_F(\bm{\pi}\cdot\bm{\gamma})+m^{(0)} \right)\psi_{\lambda}(\mathbf{r}), \label{model-H-0-matrices} \end{equation} where $\bar{\psi}_{\lambda}(\mathbf{r})=\psi^{\dag}_{\lambda}(\mathbf{r})\gamma^0$. When an external electric field is applied perpendicularly to the surfaces of the TI slab, the gapless surface states will tend to completely screen the field out. Indeed, from a physics viewpoint, the TI slab is like a Faraday cage made of gapless (metallic) surface states enclosing a gapped (insulating) interior. This implies that there should be no electric field inside a (sufficiently thick) TI slab. Enforcing this condition allows one to determine the charge densities and electrochemical potentials on the surfaces. In terms of the charge densities on the top and bottom surfaces, one has \begin{equation} \rho_{\lambda }=\lambda \epsilon_0 \mathcal{E}, \label{model-DOS-zero-field} \end{equation} where $\mathcal{E}$ is the external electric field pointing in the $z$-direction, $\epsilon_0 \approx 8.854 \times10^{-12}~\mbox{F/m}$ is the permittivity of free space, and $2\epsilon_0 \mathcal{E}$ corresponds to the difference of the charge densities of the top and bottom surfaces needed to compensate the external electric field. Under the parity transformation $P$ in (2+1) dimensions, which changes the sign of a spatial coordinate, i.e., $(x,y) \to (-x,y)$, the two-component spinors transform as follows: $P\psi(t,x,y)P^{-1}=\sigma_x\psi(t,-x,y)$. Clearly, the last term in Hamiltonian (\ref{model-H-0-matrices}) breaks parity, as well as the TR symmetry $T\psi(t,x,y)T^{-1}=\sigma_y\psi^{*}(-t,x,y)$. This mass term is known in the literature as the Haldane mass $\sum_{\lambda}m_H\,\bar{\psi}_{\lambda}\psi_{\lambda}$ \cite{Haldane}. A parity and TR invariant mass is also possible in the model with two irreducible representations. It is given by the Dirac mass term $\sum_{\lambda}m_D\,\lambda\bar{\psi}_{\lambda}\psi_{\lambda}$ with the parity transformation defined by $\psi_{\lambda=+1} \to \sigma_x \psi_{\lambda=-1}$ and $\psi_{\lambda=-1} \to \sigma_x \psi_{\lambda=+1}$. (Note that, in the TI slab model, this transformation interchanges the states on the different spatially separated surfaces.) While a Chern--Simons mass term for the gauge field is induced via one-loop polarization when the Haldane mass is present, the Chern-Simons term is absent in the case of the Dirac mass. The spontaneous breaking of parity in $(2+1)$-dimensional QED was studied in Ref.~\cite{Appelquist1:1986}. In this study, the model interaction Hamiltonian $H_{\rm int}$ includes both a long-range Coulomb and a short-range local four-fermion interactions \begin{equation} H_{\rm int} = \frac{e^2}{8\pi \epsilon_0 \kappa_{\rm surf}}\int d^2\mathbf{r}d^2\mathbf{r}^{\prime}\,\frac{\Psi^{\dagger}(\mathbf{r}) \Psi(\mathbf{r}) \Psi^{\dagger}(\mathbf{r}^{\prime})\Psi(\mathbf{r}^{\prime})}{|\mathbf{r}-\mathbf{r}^{\prime}|} +\frac{G_{\rm int}}{2}\int d^2\mathbf{r}\, \Psi^{\dagger}(\mathbf{r})\Psi(\mathbf{r}) \Psi^{\dagger}(\mathbf{r})\Psi(\mathbf{r}), \label{model-H-int} \end{equation} where $\Psi(\mathbf{r})=\left(\psi_{\lambda=+1}(\mathbf{r}),\psi_{\lambda=-1}(\mathbf{r})\right)^T$. The first term in $H_{\rm int}$ describes the long-range Coulomb interaction and takes into account the effective surface dielectric constant $\kappa_{\rm surf}=\left(1+\kappa_{\rm bulk}\right)/2 \approx 56$, where the bulk dielectric constant $\kappa_{\rm bulk}\approx113$ for Bi$_2$Se$_3$ \cite{Madelung}. The second term captures the on-site local repulsion, parametrized by the dimensionful coupling constant $G_{\rm int}$. In view of the large bulk dielectric constant and assuming a large slab thickness, we neglect the intersurface interaction and the possible formation of an intersurface exciton condensate \cite{Franz:2009, Efimkin:2012}. Thus, the full Hamiltonian of our model is given by the sum of the free and interaction Hamiltonians in Eqs.~(\ref{model-H-0-matrices}) and (\ref{model-H-int}). \section{Gap equations} \label{sec:gap-equation-Coulomb} In this section, we study the gap generation in the effective model of a sufficiently thick TI slab described in the previous section. The inverse free surface fermion propagator is given by \begin{equation} iS^{-1}_{\lambda}(u,u^\prime) = \left[(i\partial_t+\mu^{(0)}_{\lambda })\gamma^0 -v_F(\bm{\pi}\cdot\bm{\gamma})-m^{(0)}\right]\delta^{3}(u-u^\prime), \label{gapEq-Coulomb-sinverse} \end{equation} where $u=\left(t, \mathbf{r}\right)$ denotes a space-time coordinate. By using this as a guide, we assume the following rather general ansatz for the inverse full surface fermion propagator: \begin{eqnarray} iG^{-1}_{\lambda}(u,u^\prime)= \left[(i\partial_t+\mu_{\lambda })\gamma^0 -v_F(\bm{\pi}\cdot\bm{\gamma})-m_{\lambda}\right]\delta^{3}(u-u^\prime), \label{gapEq-Coulomb-ginverse} \end{eqnarray} where $m_{\lambda }$ is a dynamically generated gap (mass) which, in general, includes both Haldane and Dirac gaps and $\mu_{\lambda }$ denotes the dynamical electrochemical potential. Note that all dynamical parameters in the full propagator are assumed to be functions of $(\bm{\pi}\cdot\bm{\gamma})^2l^2$, where $l = 1/\sqrt{|eB|}$ is the magnetic length. Therefore, in the end, they all depend on the Landau level index $n$. Because of the long-range interaction, in principle, the renormalization of the wave function should be included in the full propagator (\ref{gapEq-Coulomb-ginverse}). This can be formally done by replacing the Fermi velocity $v_F$ with a dynamical function $F_{\lambda}$. It is well justified, however, to neglect the renormalization of the Fermi velocity and replace it with $v_F$. Indeed, even in the case of graphene with an unscreened Coulomb interaction, the renormalized Fermi velocity is generically $10\%$ to $30\%$ larger than the corresponding bare value $v_F$ \cite{2011NatPh7,Gonzalez:1993uz,Gorbar:2011kc}. Because of a much larger surface dielectric constant and, consequently, a much smaller coupling constant, the Coulomb interaction will play a minor role in the Fermi velocity renormalization and, as we will show below, in the generation of dynamical gaps in TIs. In order to represent the inverse propagator in the form of a Landau-level expansion, we use the following complete set of eigenstates (for details, see Appendix~A in Ref.~\cite{Gorbar:2011kc}): \begin{equation} \psi_{n, k_y}(\mathbf{r})=\frac{1}{\sqrt{2^{n+1} \pi l n!}} H_n\left(k_yl+\frac{x}{l}\right)e^{-\frac{1}{2l^2}\left(x+k_yl^2\right)^2} e^{is_Bk_yy}, \label{gapEq-Coulomb-psi-nky} \end{equation} where $H_n(x)$ are the Hermite polynomials and $s_{B}=\mbox{sign}(eB)$. By making use of the results in Appendix~\ref{sec:wf-Green}, we derive the following inverse fermion propagators in the mixed frequency-momentum representation: \begin{eqnarray} \label{gapEq-Coulomb-sinverse-LL-no-phase} S^{-1}_{\lambda}(\omega,\mathbf{r},\mathbf{r}^\prime)&=& e^{i\Phi(\mathbf{r},\mathbf{r}^\prime)}\tilde{S}_{\lambda}^{-1}(\omega,\mathbf{r}-\mathbf{r}^\prime) ,\\ \label{gapEq-Coulomb-ginverse-LL-no-phase} G^{-1}_{\lambda}(\omega,\mathbf{r},\mathbf{r}^\prime)&=& e^{i\Phi(\mathbf{r},\mathbf{r}^\prime)}\tilde{G}_{\lambda}^{-1}(\omega,\mathbf{r}-\mathbf{r}^\prime). \end{eqnarray} Here $\Phi(\mathbf{r}, \mathbf{r}^{\prime})= -eB(x+x^{\prime})(y-y^{\prime})/2$ is the famous Schwinger phase and the translation invariant parts of the inverse propagators are given by \begin{eqnarray} \label{gapEq-Coulomb-sinverse-LL} iS^{-1}_{\lambda}(\omega,\mathbf{r}-\mathbf{r}^\prime)&=& \frac{e^{-\eta/2}}{2\pi l^2}\sum_{n=0}^{\infty} \Big\{ s_B (\omega+\mu^{(0)}_{\lambda })\left[P_{+} L_{n-1}(\eta)-P_{-} L_{n}(\eta)\right]-m^{(0)}\left[P_{+} L_{n-1}(\eta)+P_{-} L_{n}(\eta) \right] \nonumber \\ &-&\frac{i}{l^2}v_F (\bm{\gamma}\cdot\mathbf{r}) L^1_{n-1}(\eta)\Big\},\\ \label{gapEq-Coulomb-ginverse-LL} iG^{-1}_{\lambda}(\omega,\mathbf{r}-\mathbf{r}^\prime)&=& \frac{e^{-\eta/2}}{2\pi l^2}\sum_{n=0}^{\infty} \Big\{ s_B (\omega+\mu_{n,\lambda })\left[P_{+} L_{n-1}(\eta)-P_{-} L_{n}(\eta)\right]-m_{n,\lambda }\left[P_{+} L_{n-1}(\eta)+P_{-} L_{n}(\eta) \right]\nonumber \\ &-&\frac{i}{l^2}v_F (\bm{\gamma}\cdot\mathbf{r}) L^1_{n-1}(\eta)\Big\}, \end{eqnarray} where $P_{\pm} =\left(1 \pm s_B \gamma^0\right)/2$, $\eta= (\mathbf{r}-\mathbf{r}^{\prime})^2/(2l^2)$, and $L^{j}_{n} \left(x\right)$ are the generalized Laguerre polynomials (by definition $L_n\equiv L_n^0$). In order to study the dynamical gap generation, we utilize the Baym--Kadanoff (BK) formalism \cite{BK}, which leads to a self-consistent Schwinger-Dyson equation for the fermion propagator. In contrast to a perturbative analysis, the BK formalism can capture nonperturbative effects such as spontaneous symmetry breaking. To leading order in coupling, the BK effective action in the model under consideration is given by Eq.~(\ref{app-action-BK}) in Appendix~\ref{sec:App-action-true}. In view of the geometry of conducting states of our TI system, it should not be too surprising that the effective action (\ref{app-action-BK}) has a form similar to that in bilayer graphene (compare with Eq.~(9) in Ref.~\cite{bilayer}). The extremum of the effective action $\frac{\delta \Gamma(G)}{\delta G_{\lambda}} =0$ defines the following Schwinger-Dyson equation for the full fermion propagator (for details, see Appendix~\ref{sec:App-action-true}): \begin{equation} iG^{-1}_{\lambda}(u,u^\prime) = iS^{-1}_{\lambda}(u,u^\prime) - e^2 \gamma^0 G_{\lambda}(u,u^{\prime}) \gamma^0 D(u^{\prime}-u) -G_{\rm int}\left\{ \gamma^0 G_{\lambda}(u,u) \gamma^0 - \gamma^0\, \mbox{tr}[\gamma^0G_{\lambda}(u,u)]\right\}\delta^{3}(u-u^{\prime}), \label{gapEq-Coulomb-gap} \end{equation} where the trace in the last term is taken over the spinor indices and the Hartree term due to the Coulomb interaction is absent. This is justified because of the overall neutrality of the sample, i.e., \begin{equation} Q_b- e\, \sum_{\lambda=\pm}\mbox{tr}[\gamma^0G_{\lambda}(u,u)]=0, \label{gapEq-Coulomb-Hartree} \end{equation} where $Q_b$ denotes the background charge due to the external gates. We note, however, that it does not make sense to drop the Hartree-type term due to the contact interaction. Therefore, the corresponding term is kept in curly brackets in Eq.~(\ref{gapEq-Coulomb-gap}). The propagator mediating the Coulomb interaction is denoted by $D(u)$. Its explicit expression is given by \begin{equation} D(u)=\int\frac{d\omega d^2\mathbf{k}}{(2\pi)^3}D(\omega, \mathbf{k})e^{-i\omega t+i\mathbf{k}\cdot\mathbf{r}}\approx\delta(t) \frac{1}{4\pi\epsilon_0\kappa_{\rm surf}} \int \frac{dk}{2\pi} \frac{kJ_0(kr)}{k+\Pi(0,k)}, \label{gapEq-Coulomb-D} \end{equation} where $J_0(x)$ is the Bessel function. In the last expression, we neglected the dependence of the polarization function $\Pi(\omega,k)$ on $\omega$, which corresponds to an instantaneous approximation. Such an approximation may be reasonable for the TI surfaces, where charge carriers propagate much slower than the speed of light and, thus, the retardation effects are negligible. It is worth noting, however, that the instantaneous approximation has a tendency to underestimate the strength of the Coulomb interaction \cite{Gorbar:2002iw}. Just like the inverse propagators in Eqs.~(\ref{gapEq-Coulomb-sinverse-LL-no-phase}) and (\ref{gapEq-Coulomb-ginverse-LL-no-phase}), the propagators themselves have the same Schwinger phase. The full propagator, in particular, takes the following explicit form: \begin{eqnarray} \label{gapEq-Coulomb-G-phase} G_{\lambda}(\omega, \mathbf{r}, \mathbf{r}^{\prime}) &=& e^{i\Phi(\mathbf{r},\mathbf{r}^\prime)}\tilde{G}_{\lambda}(\omega, \mathbf{r}-\mathbf{r}^{\prime}) ,\\ \label{gapEq-Coulomb-G-no-phase} \tilde{G}_{\lambda}(\omega, \mathbf{r}-\mathbf{r}^{\prime})&=& \frac{e^{-\eta/2}}{2\pi l^2}\sum_{n=0}^{\infty} \Bigg\{\frac{s_B\left(\omega +\mu_{n,\lambda } \right)\left[ L_{n-1}(\eta)P_{+} -L_{n}(\eta)P_{-}\right]}{\left(\omega+\mu_{n,\lambda } +i0\,\mbox{sign}{(\omega)}\right)^2-M_n^2} \nonumber\\ &+&\frac{m_{n,\lambda } \left[L_{n-1}(\eta)P_{+} +L_{n}(\eta)P_{-}\right] -i\frac{v_F}{l^2}L_{n-1}^1(\eta) \left(\bm{\gamma}\cdot(\mathbf{r}-\mathbf{r}^{\prime})\right) }{\left(\omega+\mu_{n,\lambda } +i0\,\mbox{sign}{(\omega)}\right)^2-M_n^2} \Bigg\}, \end{eqnarray} where $M_n=\sqrt{\left(m_{n, \lambda }\right)^2+\epsilon_{B}^2n}$ and $\epsilon_{B}= \sqrt{2v_F^2|eB|}$ is the Landau energy scale. The inverse and full fermion propagators at finite temperature are easily obtained through the standard replacement $\omega\to i\omega_{m^{\prime}}=i\pi T(2m^{\prime}+1)$. By factorizing the Schwinger phase on both sides of Eq.~(\ref{gapEq-Coulomb-gap}), we arrive at the following gap equation for the translation invariant part of the full propagator: \begin{equation} i\tilde{G}^{-1}_{\lambda}(\omega, \mathbf{r}) = i\tilde{S}^{-1}_{\lambda}(\omega, \mathbf{r}) - \alpha v_F \!\! \int\frac{d\Omega}{2\pi} \frac{dk}{2\pi} \frac{kJ_0(kr)}{k+\Pi(0,k)} \gamma^0 \tilde{G}_{\lambda}(\Omega, \mathbf{r}) \gamma^0 -G_{\rm int}\!\int\frac{d\Omega}{2\pi} \delta^2(\mathbf{r}) \left\{ \gamma^0 \tilde{G}_{\lambda}(\Omega, \mathbf{r}) \gamma^0 - \gamma^0\, \mbox{tr}[\gamma^0\tilde{G}_{\lambda}(\Omega, \mathbf{r})]\right\}. \label{gapEq-Coulomb-gap-1} \end{equation} Here we introduced the following notation $\alpha=e^2/(4\pi \epsilon_0 v_F \kappa_{\rm surf})$. In the case of Bi$_2$Se$_3$, in particular, $\alpha\approx0.062$. Although it is hard to estimate $G_{\rm int}$ reliably, its origin is the Coulomb repulsion on distance scales comparable to the lattice spacing. Thus, it may be reasonable to use the following approximate model value: \begin{equation} G_{\rm int}=\frac{\alpha v_F^2\kappa_{\rm surf}}{\Delta_{\rm bulk}} \approx 168.7~\mbox{eV\AA}^2, \label{gapEq-Coulomb-Gint} \end{equation} where the factor $\kappa_{\rm surf}$ was introduced in order to compensate for polarization effects in the definition of $\alpha$. Indeed, polarization effects should be negligible at small distances. It is worth noting that the corresponding dimensionless constant \begin{equation} g_{\rm int}= \frac{G_{\rm int}\Delta_{\rm bulk}}{8\sqrt{2\pi} v_F^2}\approx 0.18 \label{gapEq-Coulomb-g-def} \end{equation} is rather small. In fact, it is an order of magnitude smaller than the critical value $g_{cr}=\sqrt{\pi}$ needed for generating a gap in a (2+1)-dimensional model in the absence of a magnetic field \cite{Gusynin:1994re}. Because of this and because of the strong suppression of the Coulomb interaction by the large dielectric constant, no dynamical generation of a gap is expected in such a TI material in the absence of an external magnetic field. Consequently, the magnetic catalysis \cite{Gusynin:1994re} will play a crucial role in the generation of dynamical gaps in TIs. (For a recent review on magnetic catalysis, see Ref.~\cite{Miransky:2015ava}.) By using the explicit form of the fermion propagator (\ref{gapEq-Coulomb-G-no-phase}) on the right-hand side of Eq.~(\ref{gapEq-Coulomb-gap-1}), we can easily calculate the integral over $\Omega$ (or the sum over the Matsubara frequency at nonzero temperature, see Appendix \ref{sec:Greens-function}). Afterwards, by multiplying both sides of the gap equation (\ref{gapEq-Coulomb-gap-1}) by $e^{-\eta/2}L_{n^{\prime}}(\eta)$ or $e^{-\eta/2}(\bm{\gamma}\cdot\mathbf{r})L_{n^{\prime}}^1(\eta)$ and then integrating over $\mathbf{r}$, the complete set of equations for the dynamical parameters can be straightforwardly obtained. In particular, the gap equations for the lowest Landau level (LLL) parameters are given by \begin{eqnarray} \Delta_{\rm eff, \lambda }&=&\mu_{\lambda}^{(0)}+s_Bm^{(0)}+ \alpha \frac{v_F}{2l} \Bigg\{ \mathcal{K}^{(0)}_{0,0}\left[ 1-2n_F\left(\Delta_{\mathrm{eff}, \lambda}\right) \right] -\sum_{n^{\prime}=1}^{\infty}\mathcal{K}^{(0)}_{n^{\prime},0}\Big[ n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}} -\mu_{n^{\prime}, \lambda })\nonumber\\ &-& s_B m_{n^{\prime}, \lambda } \frac{1-n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda }) -n_F(M_{n^{\prime}} -\mu_{n^{\prime}, \lambda })}{M_{n^{\prime}}} \Big] \Bigg\} +\frac{G_{\rm int}}{4\pi l^2}\Bigg\{ \sum_{n^{\prime}=1}^{\infty} \left[n_F\left(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda }\right)-n_F\left(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda }\right)\right]\nonumber\\ &+& \sum_{n^{\prime}=1}^{\infty} s_B m_{n^{\prime}, \lambda } \frac{1-n_F\left(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda }\right)-n_F \left(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda } \right)}{M_{n^{\prime}}} \Bigg\}, \label{gapEq-Coulomb-gap-Delta} \end{eqnarray} where $n_{F}(x)=1/\left(e^{x/T}+1\right)$ is the Fermi-Dirac distribution. Notice that we introduced an effective LLL electrochemical potential $\Delta_{\mathrm{eff}, \lambda} =\mu_{0, \lambda}+ s_{B}m_{0, \lambda}$ because the LLL parameters $\mu_{0,\lambda}$ and $m_{0,\lambda}$ cannot be unambiguously defined separately and only their combination $\Delta_{\mathrm{eff}, \lambda}$ has a well-defined physical meaning \cite{Gorbar:2008hu,Goerbig}. Similarly, the equations for the dynamical parameters associated with higher Landau levels read as \begin{eqnarray} m_{n, \lambda }&=&m^{(0)}+ s_B \alpha \frac{v_F}{4l} \Bigg\{ \mathcal{K}^{(0)}_{0,n}\left[ 1-2n_F\left(\Delta_{\mathrm{eff}, \lambda}\right) \right]+ \sum_{n^{\prime}=1}^{\infty}\mathcal{K}^{(0)}_{n^{\prime}-1,n-1} \Big[ n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda }) \nonumber\\ &+&s_B m_{n^{\prime},\lambda } \frac{1-n_F(M_{n^{\prime}}+\mu_{n^{\prime},\lambda })-n_F(M_{n^{\prime}}-\mu_{n^{\prime},\lambda })} {M_{n^{\prime}}} \Big] \nonumber\\ &-& \sum_{n^{\prime}=1}^{\infty}\mathcal{K}^{(0)}_{n^{\prime},n}\left[ n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}} -\mu_{n^{\prime}, \lambda }) -s_B m_{n^{\prime}, \lambda } \frac{1-n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}} -\mu_{n^{\prime}, \lambda })}{M_{n^{\prime}}} \right] \Bigg\} \nonumber\\ &+& \frac{G_{\rm int}}{8\pi l^2}\left\{ s_B\left[1-2n_F\left(\Delta_{\mathrm{eff}, \lambda}\right)\right] +2\sum_{n^{\prime}=1}^{\infty} m_{n^{\prime}, \lambda } \frac{1-n_F\left(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda }\right)-n_F\left(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda } \right)}{M_{n^{\prime}}}\right\}, \label{gapEq-Coulomb-gap-m} \end{eqnarray} \begin{eqnarray} \mu_{n, \lambda }&=&\mu^{(0)}_{\lambda }+\alpha \frac{v_F}{4l} \Bigg\{ \mathcal{K}^{(0)}_{0,n} \left[ 1-2n_F\left(\Delta_{\mathrm{eff}, \lambda}\right) \right] -\sum_{n^{\prime}=1}^{\infty}\mathcal{K}^{(0)}_{n^{\prime}-1,n-1} \Big[ n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda }) \nonumber\\ &+&s_B m_{n^{\prime}, \lambda } \frac{1-n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda })} {M_{n^{\prime}}} \Big] -\sum_{n^{\prime}=1}^{\infty}\mathcal{K}^{(0)}_{n^{\prime},n}\Big[ n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}} -\mu_{n^{\prime}, \lambda }) \nonumber\\ &-&s_B m_{n^{\prime}, \lambda } \frac{1-n_F(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda })-n_F(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda })} {M_{n^{\prime}}} \Big] \Bigg\} \nonumber\\ &-&\frac{G_{\rm int}}{8\pi l^2}\left\{ \left[1-2n_F\left(\Delta_{\mathrm{eff}, \lambda}\right)\right] -2\sum_{n^{\prime}=1}^{\infty} \left[n_F\left(M_{n^{\prime}}+\mu_{n^{\prime}, \lambda }\right)-n_F\left(M_{n^{\prime}}-\mu_{n^{\prime}, \lambda }\right)\right] \right\}. \label{gapEq-Coulomb-gap-mu} \end{eqnarray} The kernel coefficients $\mathcal{K}^{(0)}_{m,n}$ that capture the long-range interaction effects in the gap equations are defined in Eq.~(\ref{app-K-def}). In this study, for simplicity, we neglect all screening effects, i.e., we set $\Pi(0,k)=0$. Then, the numerical analysis greatly simplifies because the coefficients $\mathcal{K}^{(0)}_{m,n}$ can be calculated analytically, see Eq.~(\ref{app-K-Pi0}). In addition to the gap equation (\ref{gapEq-Coulomb-gap-1}), the constraints for the surface charge densities in Eq.~(\ref{model-DOS-zero-field}) should be also satisfied. In terms of the model parameters, the explicit form of the constraint reads as \begin{equation} -\frac{e}{4\pi l^2}\left\{ \left[1-2n_F\left(\Delta_{\mathrm{eff}, \lambda}\right)\right] -2\sum_{n=1}^{\infty} \left[n_F\left(M_n+\mu_{n, \lambda }\right)-n_F\left(M_n-\mu_{n, \lambda }\right)\right] \right\} =\lambda \epsilon_0 \mathcal{E}, \label{gapEq-Coulomb-DOS} \end{equation} where we used the definition for the surface charge densities in terms of the fermion propagator, i.e., $\rho_{\lambda }=e\,\mbox{tr}[G_{\lambda}(u,u) \gamma^0]$. Because the surface charge density is fixed by the external electric field, the electrochemical potential $\mu^{(0)}_{\lambda}$ is not an independent parameter. It is determined together with the other dynamical parameters by solving the system of Eqs.~(\ref{gapEq-Coulomb-gap-Delta}) through (\ref{gapEq-Coulomb-DOS}). \section{Numerical results} \label{sec:gap-equation-Coulomb-results-kappa} In this section, we present our numerical solutions of gap equations (\ref{gapEq-Coulomb-gap-Delta}) through (\ref{gapEq-Coulomb-gap-mu}), together with the constraint in Eq.~(\ref{gapEq-Coulomb-DOS}). Before proceeding to the analysis, it is convenient to give the formal definition of the Dirac and Haldane gaps in the TI model at hand. While the original surface gaps $m_{n,+}$ and $m_{n,-}$ (with $n \ge 1$) have a straightforward physical meaning, the symmetry properties of the ground state can be better understood in terms of Dirac and Haldane gaps, i.e., \begin{equation} m_{n, D}= \frac{m_{n,+}-m_{n,-}}{2}, \qquad m_{n, H}= \frac{m_{n,+}+m_{n,-}}{2}. \label{gap-Coulomb-results-Dirac-Haldanem-tmu} \end{equation} Strictly speaking, these gaps cannot be associated with the usual Dirac and Haldane masses in (2+1)-dimensional QED (see, Ref.~\cite{Appelquist:1986}), because $m_{\pm}$ in TIs correspond to spatially separated surfaces. Since the free Hamiltonian (\ref{model-H-s-0}) contains the bare Haldane gap $m^{(0)}$ due to the Zeeman interaction, it is also convenient to define the dynamical part of the total Haldane gap $\Delta m_{n, H}\equiv m_{n,H}-m^{(0)}$. In order to provide an insight into relation (\ref{gap-Coulomb-results-Dirac-Haldanem-tmu}) between $m_{\pm}$ and Dirac and Haldane gaps, let us recall the reducible 4$\times$4 representation for $\mbox{QED}_{2+1}$ considered in Ref.~\cite{Appelquist:1986} \begin{eqnarray} \tilde{\gamma}^0=\left( \begin{array}{cc} \gamma^0 & 0 \\ 0 & -\gamma^0 \\ \end{array} \right), \quad \tilde{\gamma}^1=\left( \begin{array}{cc} \gamma^1 & 0 \\ 0 & -\gamma^1 \\ \end{array} \right), \quad \tilde{\gamma}^2=\left( \begin{array}{cc} \gamma^2 & 0 \\ 0 & -\gamma^2 \\ \end{array} \right). \label{gap-Coulomb-results-reducible} \end{eqnarray} In this representation, in addition to the $\gamma$-matrices in Eq.~(\ref{gap-Coulomb-results-reducible}), there exist two other matrices, \begin{eqnarray} \tilde{\gamma}^3=i\left( \begin{array}{cc} 0 & 1 \\ 1 & 0 \\ \end{array} \right), \quad \tilde{\gamma}^5=i\left( \begin{array}{cc} 0 & 1 \\ -1 & 0 \\ \end{array} \right), \label{gap-Coulomb-results-reducible-35} \end{eqnarray} which anticommute with $\tilde{\gamma}^0$, $\tilde{\gamma}^1$, and $\tilde{\gamma}^2$. In terms of these $4\times 4$ matrices, the existence of a $U(2)$ symmetry in the model of a TI slab, defined by Eqs.~(\ref{model-H-0-matrices}) and (\ref{model-H-int}), is transparent. The corresponding group generators are given by \begin{equation} 1, \quad i{\cal R}_{\mu}\tilde{\gamma}^3, \quad {\cal R}_{\mu}\tilde{\gamma}^5, \quad \mbox{and} \quad \tilde{\gamma}^3\tilde{\gamma}^5, \label{gap-Coulomb-results-generators} \end{equation} where ${\cal R}_{\mu}$ is the operator which interchanges $\mu_{+}^{(0)}\leftrightarrow\mu_{-}^{(0)}$ in the low-energy free Hamiltonian (\ref{model-H-0-matrices}). (Note that, in the absence of an external electric field, there is no need in the operator ${\cal R}_{\mu}$.) As is easy to check, the Dirac gap $m_D\bar{\Psi}\Psi$ breaks the $U(2)$ symmetry down to $U_{+}(1)\times U_{-}(1)$, where $\bar{\Psi}=\Psi^{\dagger}\tilde{\gamma}^0$ and the subscript $\lambda=\pm$ labels the two irreducible representations or the surfaces of the TI slab. The Haldane gap $m_H\bar{\Psi}\tilde{\gamma}^3\tilde{\gamma}^5\Psi$ is invariant with respect to the $U(2)$ symmetry, but, unlike the Dirac gap, it breaks the parity $P$ and $T$ symmetries. Since external electric and magnetic fields break $P$ and $T$ symmetries, the generation of the Haldane gap has no effect on symmetry breaking. Therefore, only the dynamically generated Dirac gap will spontaneously break the symmetry of our model. As we will see below, such a gap is indeed generated due to the electrified magnetic catalysis. For numerical calculations, it is useful to estimate energy scales in the problem at hand \begin{eqnarray} \Delta_{\rm bulk}\approx 350~\mbox{meV},\quad\quad \frac{g_s\mu_B B}{2}\approx 0.5B[T]~\mbox{meV},\nonumber \\ \epsilon_B=\sqrt{2v_F^2|eB|}\approx 22.6\sqrt{B[T]}~\mbox{meV},\quad\quad l\approx25.7~\mbox{nm}/\sqrt{B[T]}. \label{gap-Coulomb-results-energy-scales} \end{eqnarray} By solving numerically the gap equations (\ref{gapEq-Coulomb-gap-Delta}), (\ref{gapEq-Coulomb-gap-m}) and (\ref{gapEq-Coulomb-gap-mu}) together with constraint (\ref{gapEq-Coulomb-DOS}), we straightforwardly obtain the electrochemical potentials $\mu_{n,\pm}$ and the gaps $m_{n,\pm}$ as functions of the magnetic field. The results for the lowest and first Landau level parameters are shown in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa} for fixed values of the electric field and temperature, $\mathcal{E}=1~\mbox{mV/\AA}$ and $T=5\times10^{-3}\Delta_{\rm bulk} \approx20~\mbox{K}$, respectively. In the calculation, we truncated the system of equations by including only $n_{\rm max}=26$ Landau levels. \begin{figure*}[!t] \begin{center} \includegraphics[width=0.32\linewidth]{Fig1a.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth]{Fig1b.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth]{Fig1c.eps} \end{center} \caption{(Color online) The lowest and first Landau level parameters as functions of the magnetic field for fixed values of the external electric field, $\mathcal{E}=1~\mbox{mV/\AA}$, and temperature, $T=5\times10^{-3}\Delta_{\rm bulk}\approx20~\mbox{K}$. The results for the (effective) electrochemical potentials $\Delta_{\mathrm{eff}, \pm}$, and $\mu_{1,\pm}$ are shown in the left panel, the gaps $m_{1,\pm}$ are shown in the middle panel, and Dirac and Haldane gaps $m_{1,D}$, $m_{1,H}$, and $\Delta m_{1,H}$ are shown in the right panel.} \label{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa} \end{figure*} As we see from the left panel in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa}, the absolute values of the electrochemical potentials $\Delta_{\mathrm{eff}, \pm}$ and $\mu_{1,\pm}$ experience a large jump around $|B|\approx5~\mbox{T}$. The jump corresponds to the point at which the filling of the first Landau level starts. We checked that the position of the jump shifts to larger values of the magnetic field with increasing the external electric field. Of course, this behavior is expected, since larger electric fields require higher charge densities on the TI surfaces. In addition to the large jump around $|B|\approx5~\mbox{T}$, we also observe additional features in the dependence of $\Delta_{\mathrm{eff}, \pm}$ and $\mu_{1,\pm}$ at smaller values of the magnetic field. They generically correspond to the onset of filling of higher Landau levels. Here it is appropriate to mention that, in all regimes studied, the electrochemical potentials $\mu^{(0)}_{\pm}$ are very similar quantitatively to $\mu_{1,\pm}$ and, therefore, we do not show them in our figures. Let us now turn to the discussion of the dynamically generated gaps. The results in the middle panel of Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa} clearly demonstrate that the surface gaps $m_{1,\pm}$ monotonically increase with the magnetic field. More interestingly, however, we find that the values of the gaps on the two surfaces, $m_{1,+}$ and $m_{1,-}$, remain comparable, although not identical to each other for sufficiently weak electric fields. The importance of this observation becomes obvious in the context of the $U(2)$ symmetry discussed earlier. Indeed, if the values of $m_{1,+}$ and $m_{1,-}$ were exactly the same, they would describe a pure Haldane solution. As is clear from the definition in Eq.~(\ref{gap-Coulomb-results-Dirac-Haldanem-tmu}), a small difference between $m_{1,+}$ and $m_{1,-}$ implies the existence of a dynamically generated Dirac gap. Such a gap is induced by the applied electric field. This conclusion is further supported by the dependence of the dynamical gaps on the electric field, shown in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa} and discussed below Eq.~(\ref{gapEq-results-Coulomb-DOS}). The results in the right panel of Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa} demonstrate that the absolute value of the Dirac gap $m_{1, D}$ increases with the magnetic field at sufficiently small fields, $|B|\lesssim 5~\mbox{T}$, when the LLL is fully filled. At larger magnetic fields, $|B|\gtrsim5~\mbox{T}$, when the LLL is not fully filled, the Dirac gap remains nearly constant (or increases very slowly). In contrast, the dynamical part of the Haldane gap $\Delta m_{1, H}$ increases approximately as $B^2$. Note that, because of the linear dependence of $m^{(0)}$ on the magnetic field, the total Haldane gap $m_{1, H}$ grows almost linearly with $B$. \begin{figure*}[!t] \begin{center} \includegraphics[width=0.32\linewidth]{Fig2a.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth]{Fig2b.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth]{Fig2c.eps} \end{center} \caption{(Color online) The lowest and first Landau level parameters as functions of the external electric field for fixed values of the magnetic field, $B=5~\mbox{T}$, and temperature, $T=5\times10^{-3}\Delta_{\rm bulk}\approx20~\mbox{K}$. The results for the (effective) electrochemical potentials $\Delta_{\mathrm{eff}, \pm}$, and $\mu_{1,\pm}$ are shown in the left panel, the gaps $m_{1,\pm}$ are shown in the middle panel, and Dirac and Haldane gaps $m_{1,D}$, $m_{1,H}$ and $\Delta m_{1,H}$ are shown in the right panel.} \label{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa} \end{figure*} The dependencies of the lowest and first Landau level parameters on the external electric field are presented in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa} for fixed values of the magnetic field ($B=5~\mbox{T}$) and temperature ($T=5\times10^{-3}\Delta_{\rm bulk}\approx20~\mbox{K}$). As we see from the left panel in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa}, the absolute values of the electrochemical potentials slowly increase with electric field at first, and then experience a substantial jump at $|\mathcal{E}|\approx1~\mbox{mV/\AA}$. The jump corresponds to the field at which the filling of the first Landau level begins. From the middle panel in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa}, we see that the surface gaps $m_{1,\pm}$ have a linear dependence at weak fields (i.e., in the regime of a partially filled LLL) and stay approximately constant at higher electric fields. As might have been expected, the Haldane gap $m_{1,H}$, which is shown in the right panel of Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa}, depends very weakly on the applied electric field. This is in contrast to the behavior of the Dirac gap $m_{1, D}$ (see the right panel in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa}), which is linear in $\mathcal{E}$ at small fields and stays approximately constant at large fields. It should be also emphasized that the Dirac gap vanishes at $\mathcal{E}=0$. As we argue below, this fact is important from the viewpoint of symmetry properties in the model. As already suggested earlier, the generation of the Dirac gap is directly connected with the applied external electric field. In order to demonstrate this in the simplest possible setting, it is instructive to consider an approximate form of the gap equation (\ref{gapEq-Coulomb-gap-m}) in the limit of a large magnetic field. By rewriting it in terms of the Haldane and Dirac gaps, we obtain \begin{eqnarray} m_{n, H}&\approx &m^{(0)}+\frac{1}{4l} \sum_{n^{\prime}=1}^{\infty} \left(\alpha v_F \mathcal{K}^{(0)}_{n^{\prime}-1,n-1} + \alpha v_F \mathcal{K}^{(0)}_{n^{\prime},n} +\frac{G_{\rm int}}{\pi l} \right) \frac{m_{n^{\prime},H} }{M_{n^{\prime}}} , \label{gapEq-results-Coulomb-gap-mH} \\ m_{n, D}&\approx& -\pi l s_B \epsilon_0\left(\alpha v_F \mathcal{K}^{(0)}_{0,n}+\frac{G_{\rm int}}{2\pi l}\right)\frac{\mathcal{E}}{e} + \frac{1}{4l} \sum_{n^{\prime}=1}^{\infty} \left(\alpha v_F \mathcal{K}^{(0)}_{n^{\prime}-1,n-1} + \alpha v_F \mathcal{K}^{(0)}_{n^{\prime},n} +\frac{G_{\rm int}}{\pi l} \right) \frac{m_{n^{\prime},D} }{M_{n^{\prime}}} , \label{gapEq-results-Coulomb-gap-mD} \end{eqnarray} where we took into account that $n_F(M_{n}\pm \mu_{n, \lambda }) \ll 1$ for $n\geq 1$ and assumed that $M_{n^{\prime}}$ is almost independent of the small Dirac gap. Note that in order to rewrite the LLL contributions in the gap equations in terms of the electric field $\mathcal{E}$, we used the following approximate expression for the surface charge densities: \begin{equation} \rho_{+}= -\rho_{-}=\frac{\rho_{+}-\rho_{-}}{2}\approx\frac{e}{4\pi l^2} \left[n_F\left(\Delta_{\mathrm{eff}, +}\right) -n_F\left(\Delta_{\mathrm{eff}, -}\right) \right] =\epsilon_0\mathcal{E}. \label{gapEq-results-Coulomb-DOS} \end{equation} By comparing the gap equations (\ref{gapEq-results-Coulomb-gap-mH}) and (\ref{gapEq-results-Coulomb-gap-mD}), we see that the external electric field plays the role of a ``seed" for the Dirac gap $m_{n, D}$, just as the bare gap parameter $m^{(0)}$ for the Haldane gap $m_{n,H}$. This explains why the external electric field is the key factor in generating the Dirac gap and breaking the $U(2)$ symmetry in the slab of 3D TIs. It may be instructive to study the dependence of the electrochemical potentials $\mu_{n,\pm}$ and gaps $m_{n,\pm}$ on the Landau level index $n$. The corresponding results for two different values of the electric field are presented in Fig.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-n-kappa} for $B=5~\mbox{T}$ and $T=5\times10^{-3}\Delta_{\rm bulk}\approx20~\mbox{K}$. As we see, all dynamical parameters depend very weakly on the Landau level index $n$. In view of the large surface dielectric constant and, consequently, weak Coulomb interaction, this result is not surprising. Moreover, it strongly suggests that the long-range interaction indeed plays a minor role compared to the local interaction. \begin{figure*}[!t] \begin{center} \includegraphics[width=0.32\linewidth]{Fig3a.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth]{Fig3b.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth]{Fig3c.eps} \end{center} \caption{(Color online) The electrochemical potentials $\mu_{n,\pm}$ (left panel), the gaps $m_{n,\pm}$ (middle panel), and the Dirac and Haldane gaps (right panel) as functions of the Landau level index $n$. The results for $\mathcal{E}=2~\mbox{mV/\AA}$ are represented by red solid and green dotted lines. The results for $\mathcal{E}=1~\mbox{mV/\AA}$ are represented by blue dashed and brown dash-dotted lines. The values of the magnetic field and temperature are $B=5~\mbox{T}$ and $T=5\times10^{-3}\Delta_{\rm bulk}\approx20~\mbox{K}$, respectively.} \label{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-n-kappa} \end{figure*} By using the above results, we can also obtain the quasiparticle energy levels as functions of the magnetic and electric fields \begin{eqnarray} \omega_{0,\lambda}=-\Delta_{\rm eff, \lambda}, \quad \omega_{n>0, \lambda}=-\mu_{n, \lambda}\pm M_{n}, \label{gap-Coulomb-results-spectrum} \end{eqnarray} where $M_n$ were given below Eq.~(\ref{gapEq-Coulomb-G-no-phase}). The corresponding numerical results are summarized in Fig.~\ref{fig:gapEq-Coulomb-results-surface-spectrum}. As we see from the left panel in Fig.~\ref{fig:gapEq-Coulomb-results-surface-spectrum}, there is a rather large splitting between the energy levels on the top and bottom surfaces at small values of the magnetic field. This corresponds to the regime with higher Landau levels being occupied. With increasing the magnetic field, the magnitude of splitting quickly diminishes and becomes rather small when the LLL regime is reached. In contrast, the increase of the electric field tends to amplify the splitting between the Landau levels. The existence of such a splitting may lead to an observation of new plateaus in the Hall conductivity. The large jumps in the energy spectrum at $|B|\approx5~\mbox{T}$ and $|\mathcal{E}|\approx1~\mbox{mV/\AA}$ correspond to the onset of and the exit from the LLL regime, respectively. As is clear, these features are directly connected with the corresponding jumps in the electrochemical potentials, seen in the left panels of Figs.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa} and \ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa}. \begin{figure*}[!t] \begin{center} \includegraphics[width=0.42\linewidth]{Fig4a.eps}\hspace*{0.1\linewidth} \includegraphics[width=0.42\linewidth]{Fig4b.eps} \end{center} \caption{(Color online) The quasiparticle energies for the first three Landau levels as functions of the magnetic field at fixed $\mathcal{E}=1~\mbox{mV/\AA}$ (left panel) and as functions of the electric field at fixed $B=5~\mbox{T}$ (right panel). Red and blue lines denote the quasiparticle energies on the top and bottom surfaces, respectively. Solid lines represent the LLL, dashed and dotted lines correspond to the first and second Landau levels, respectively. The temperature is $T=5\times10^{-3}\Delta_{\rm bulk}\approx20~\mbox{K}$.} \label{fig:gapEq-Coulomb-results-surface-spectrum} \end{figure*} Before concluding this section, let us briefly discuss the role of finite temperature in our solution. As expected, the main results remain qualitatively the same for a whole range of sufficiently small values of the temperature. With increasing (decreasing) the temperature, however, the jumps that correspond to the onset of and the exit from the LLL regime become smoother (sharper) in the dependence of the electrochemical potentials on the fields, shown in the left panels of Figs.~\ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-B-kappa} and \ref{fig:gapEq-Coulomb-results-surface-tmu0-HLL-E-kappa}. It is also worth pointing that a weak dependence of electrochemical potentials on the fields in the regions between the jumps is caused by thermal broadening of Landau levels. It vanishes in the limit $T\to0$. \section{Inhomogeneous phase with two stripes: qualitative approach} \label{sec:stripe} In the previous section, we advocated the homogeneous phase with dynamically generated gaps as the ground state of 3D TIs in a sufficiently strong external electric field. On the other hand, the inhomogeneous CDW phase considered in Ref.~\cite{Vishwanath:2010} is likely to be more favorable in weak electric fields. In order to provide a qualitative analytic description of the inhomogeneous CDW phase with a ``stripe'' pattern, in this section we consider a simple configuration of two stripes with an infinitely thin transition region, or a domain wall at $x=0$. This is modeled by an inhomogeneous gap $m(x)=|m|\,\mbox{sign}{(x)}$ for the top surface, i.e., $\lambda=+1$ and $s_{B}=+1$. (Note that such a gap function with asymptotes of opposite sign at $x\to\pm\infty$, but without a magnetic field, is qualitatively similar to the famous Jackiw-Rebbi solution in 1D \cite{Jackiw-Rebbi}.) The solution to the Dirac equation with the gap function in the form $m(x)=|m|\,\mbox{sign}{(x)}$ is discussed in Appendix~\ref{sec:domain-wall}. The corresponding numerical results for the quasiparticle energy spectrum as a function of $k_y$, as well as the chiral condensate and charge density as functions of the spatial coordinate $x$, are shown in Fig.~\ref{fig:stripe-zero-modes-chiral-condensate}. In order to plot the results, we fixed the model parameters as follows: $m=5~\mbox{meV}$, $\mu=0$, and $B=5~\mbox{T}$. In the calculation, we also limited the sum over Landau levels ($n_{\rm max}=26$) and cut off the integration over $k_y$ ($-6/l \leq k_y \leq 6/l$). \begin{figure*}[!t] \begin{center} \includegraphics[width=0.32\linewidth]{Fig5a.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth] {Fig5b.eps}\hspace{0.012\linewidth} \includegraphics[width=0.32\linewidth] {Fig5c.eps} \end{center} \caption{(Color online) The energy spectrum as a function of $k_y$ (left panel), the chiral condensate (middle panel) and the charge density (right panel) as functions of $x$ in the Dirac problem with the inhomogeneous gap $m(x)=|m|\,\mbox{sign}{(x)}$. For simplicity, only the first 6 Landau levels are presented in the left panel. The red solid lines in the middle and right panels represent the numerical data. The blue dashed lines denote the corresponding fits. The model parameters are $m=5~\mbox{meV}$, $\mu=0$, and $B=5~\mbox{T}$.} \label{fig:stripe-zero-modes-chiral-condensate} \end{figure*} As we see from Fig.~\ref{fig:stripe-zero-modes-chiral-condensate}, the chiral condensate and the charge density have a kink and antikink structure, respectively. Therefore, the existence of zero energy states on the domain wall agrees with the inhomogeneous form of the gap function. The chiral condensate and the charge density can be fitted well by the following functions: \begin{eqnarray} \label{stripe-zero-modes-chircond-fit} \mbox{tr}[G(u,u)]=\sum_{n} \frac{\,\mbox{sign}{(\omega_n)}}{2} \bar{\psi}_{\omega_n}(x)\psi_{\omega_n}(x) \approx \frac{1}{4\pi l^2} 2.5 \tanh{\left(2.6\frac{x}{l}\right)}, \\ \label{stripe-zero-modes-DOS-fit} \mbox{tr}[\gamma^0 G(u,u)]=\sum_{n} \frac{\,\mbox{sign}{(\omega_n)}}{2} \psi^{\dag}_{\omega_n}(x)\psi_{\omega_n}(x) \approx -\frac{1}{4\pi l^2} 1.0 \tanh{\left(1.3\frac{x}{l}\right)}, \end{eqnarray} where the sum runs over the complete set of eigenstates, given by the solutions to the spectral equation (\ref{stripe-zero-modes-spectral-eq}). These fits are shown in Fig.~\ref{fig:stripe-zero-modes-chiral-condensate} alongside with the numerical solutions. Let us now consider the case of a nonzero external electric field, applied perpendicularly to the slab. By considering a sufficiently thick slab, we will assume that the average electric field inside the slab vanishes. In the homogeneous case, the field is screened by the uniform surface charge densities. This generically requires specific nonzero electrochemical potentials $\mu_{\pm}$ on the top and bottom TIs surfaces. In the inhomogeneous striped phase, however, the simplest way to achieve a nonzero average surface charge densities is to vary the width of stripes by $\Delta l_x$. The value of $\Delta l_x$ can be estimated from the following expression: \begin{equation} \epsilon_0\mathcal{E} = \rho(l_x+\Delta l_x) - \rho(l_x)+ \rho(-l_x+\Delta l_x) - \rho(-l_x) \approx \frac{e}{4\pi l^2} \left[\tanh{\left(1.3\frac{l_x+\Delta l_x}{l}\right)}-\tanh{\left(1.3\frac{l_x-\Delta l_x}{l}\right)}\right]. \label{stripe-energy-diff-stripes-DOS} \end{equation} where we used the standard definition for the surface charge density $\rho=e\,\mbox{tr}[\gamma^0 G(u,u)]$ together with Eq.~(\ref{stripe-zero-modes-DOS-fit}). \begin{figure*}[!t] \begin{center} \includegraphics[width=0.45\linewidth]{Fig6.eps} \end{center} \caption{(Color online) The ratio of the width correction $\Delta l_x$ to the stripe half-width $l_x$ as a function of the external electric field (red line). The model parameters are $m=5~\mbox{meV}$, $\mu=0$, $B=5~\mbox{T}$, and $l_x=500~\mbox{\AA}$.} \label{fig:stripe-energy-diff-nonzero-field-Lky-kappa} \end{figure*} Our numerical result for the ratio of the correction $\Delta l_x$ to the stripe half-width $l_x$ is plotted in Fig.~\ref{fig:stripe-energy-diff-nonzero-field-Lky-kappa}. We see that the correction to the stripe width $\Delta l_x$ becomes significant at sufficiently strong electric fields. In fact, it is comparable to $l_x$ already for $\mathcal{E} \gtrsim 1~\mbox{mV/\AA}$. This suggests that the stripe phase is unstable when the electric field exceeds a certain critical value $\mathcal{E}_{\rm cr}$. Quantitatively, the critical value roughly corresponds to the beginning of the first Landau level filling, \begin{equation} \mathcal{E}_{\rm cr}\approx\frac{e^2B}{4\pi \epsilon_{0} \hbar c}, \label{stripe-energy-diff-critical-E} \end{equation} where we used Eq.~(\ref{gapEq-Coulomb-DOS}) and restored Plank's constant $\hbar$ and the speed of light $c$. This relation implies that $\mathcal{E}_{\rm cr}\approx0.22B[\mbox{T}]~\mbox{mV}/\mbox{\AA}$, which is in a good agreement with our previous numerical estimate $\mathcal{E} \gtrsim 1~\mbox{mV/\AA}$ at $B=5~\mbox{T}$. Another way to estimate the critical value of the electric field is to compare the free energy density in the homogeneous phase, which is given by Eq.~(\ref{app-free-energy-T}), with the energy density of the stripe phase estimated in Ref.~\cite{Vishwanath:2010}. In other words, the value of $\mathcal{E}_{\rm cr}$ is given by the solution to the following equation: \begin{equation} 0=\sum_{\lambda}\Omega_{\lambda} - \left[\frac{\sqrt{\alpha \gamma}}{l^2}-\frac{m}{l^2}\right], \label{stripe-energy-diff-critical-field-Eq} \end{equation} where the term in square brackets corresponds to the energy cost of creating the stripe phase \cite{Vishwanath:2010}. The latter is characterized by the domain wall tension $\gamma\sim1/l^2$ and the magnetic mass $m\sim 1/l$. The value of $\alpha$ is given below Eq.~(\ref{gapEq-Coulomb-gap-1}). The solution to Eq.~(\ref{stripe-energy-diff-critical-field-Eq}) can be easily obtained numerically and appears to agree quite well with the estimate in Eq.~(\ref{stripe-energy-diff-critical-E}). One can also obtain an approximate analytical solution to Eq.~(\ref{stripe-energy-diff-critical-field-Eq}) by using the LLL approximation for the free energy density (\ref{app-free-energy-T}), i.e., \begin{equation} \Omega_{\lambda} \approx \frac{\lambda \epsilon_0 \mathcal{E} \Delta_{\rm eff}}{e} , \label{stripe-energy-diff-free-energy-LLL-1} \end{equation} where we used the LLL approximation for the charge density (\ref{gapEq-Coulomb-DOS}). Then, by substituting this into Eq.~(\ref{stripe-energy-diff-critical-field-Eq}) and estimating the effective electrochemical potential as $\Delta_{\rm eff}\sim \lambda/l$, we find the following critical value of the electric field: \begin{equation} \mathcal{E}_{\rm cr} \approx \frac{e^2B (1-\sqrt{\alpha})}{2\epsilon_0 \hbar c}. \label{stripe-energy-diff-critical-E-2} \end{equation} This result is qualitatively the same as the estimate in Eq.~(\ref{stripe-energy-diff-critical-E}), although quantitatively appears to be somewhat larger, $\mathcal{E}_{\rm cr} \approx 1.03B[\mbox{T}]~\mbox{mV}/\mbox{\AA}$. We conclude, therefore, that the critical electric field scales linearly with the magnetic field, $\mathcal{E}_{\rm cr} \sim e^2B/(\epsilon_0 \hbar c)$, but the coefficient of proportionality is determined only up to an overall factor of order $1$. \section{Discussion} \label{sec:Discussion} In this section, we discuss the range of validity and limitations of our study, and compare our main results with those existing in the literature. Let us start by pointing the limitation of our model used for the description of the TI surface states. While the model captures the Dirac nature of the low-energy quasiparticles, it does not describe the hexagonal warping of the Fermi surface that occurs away from the Dirac point \cite{Chen:2009, Hsieh:2009}. The corresponding effect was taken into account in the study of gap generation in Ref.~\cite{Baum:2012} and could play an essential role in some TIs. For example, this may be the case in Bi$_2$Te$_3$ [e.g., see Fig.~1(c) in Ref.~\cite{Chen:2009}], in which the band gap is about three times smaller than in Bi$_2$Se$_3$ and the trigonal potential $\sim k^3$ is rather strong. In the case of Bi$_2$Se$_3$, however, the hexagonal warping could be safely neglected, except for the case of rather high values of the chemical potential [e.g., see Fig.~8 in Ref.~\cite{Ando:rev} and Fig.~3(b) in Ref.~\cite{Cava-Hasan}]. The model used in this study also ignores a Schr\"{o}dinger-type term $\sim k^2$, which describes an asymmetry between the electron and hole bands \cite{Carbotte:2015} (see also Fig.~1 in Ref.~\cite{Cava-Hasan}). When the quadratic term is sufficiently small, it is not expected to substantially affect the dynamics of the gap generation. The study here did not include the effects of the intersurface tunneling on the dynamical generation of gaps. According to Ref.~\cite{Linder:2009}, tunneling between the opposite surfaces may be quite important only for sufficiently thin ($l_z\lesssim8~\mbox{nm}$) TI slabs. Therefore, neglecting the intersurface tunneling is expected to be a good approximation in the case of thick samples. It would be interesting, however, to rigorously study the corresponding effects in thin TI films in external electric and magnetic fields. One of the uncertainties of the model Hamiltonian used in this study is the strength of the local interaction $G_{\rm int}$. Although the order of magnitude of this coupling constant could be estimated by using general arguments, its precise value is unknown. Despite this, we argue that the simplified model (\ref{model-H-0-matrices}) that includes both short- and long-range interactions (\ref{model-H-int}) is sufficient for the qualitative analysis of the electrified magnetic catalysis in 3D TIs. Moreover, we might even suggest that, irrespective of the specific value of the coupling constants, the qualitative features established here should be rather universal. It is interesting to compare our results with those obtained in Ref.~\cite{Vishwanath:2010}, where the phase diagram was studied in 3D TIs in a magnetic field, but without an external electric field. The authors of Ref.~\cite{Vishwanath:2010} argued that, depending on the strength of local interaction, the CFL or CDW (``stripe'' or ``bubble'') phases can be realized. Our results here suggest that neither of those two phases describe the ground state of the TI slab in a sufficiently strong external electric field. The CFL phase with the half-filled LLL on each TI surface cannot be easily deformed to screen out the external electric field from penetrating into the TI bulk. This would imply a large energy cost and disfavor the CFL phase. The CDW phase could perhaps survive when a relatively weak electric field is applied. In this case, the average charge densities on the TI surfaces, which are needed to screen the electric field out from the bulk, could be simply obtained by the formation of positive and negative stripes (or bubbles) of unequal size. (Note that it is energetically favorable to have either completely filled or empty LLL inside the stripes \cite{Vishwanath:2010}.) As we showed in Sec.~\ref{sec:stripe}, a simple estimate suggests that the charge imbalance obtained by the variation of the stripe widths can compensate only relatively weak electric fields. Therefore, a sufficiently strong electric field $\mathcal{E}>\mathcal{E}_{\rm cr}$ also destroys the CDW phase. Our parametric estimate for the critical electric field strength is $\mathcal{E}_{\rm cr} \sim e^2B/(\epsilon_0 \hbar c)$. In view of the above arguments, we claim that the ground state of the TI slab in a nonzero magnetic field and a sufficiently strong electric field is a homogeneous phase with equal in magnitude, but opposite in sign surface charge densities. It is also characterized by the presence of both Dirac and Haldane gaps. While our qualitative conclusion seems rather rigorous, this study is insufficient to establish the precise structure of the phase diagram in the plane of the applied electric and magnetic fields. It would be very interesting to clarify the details of the corresponding phase diagram either experimentally or numerically. It may be instructive to note that the thermal broadening of Landau levels plays a relatively important technical role in our analysis and in the description of the electrified homogeneous phases. Indeed, by using a nonzero temperature, we were able to unambiguously describe the surface ground states with adjustable partial fillings of Landau levels, needed to screen the external electric field. Certainly, the corresponding ground states allow a well defined zero temperature limit, but their description may become more subtle. By noting that surface impurities also broaden Landau levels, we suggest that their presence could lead to a realization of the electrified magnetic catalysis similar to that in Sec.~\ref{sec:gap-equation-Coulomb-results-kappa}. As is clear from our study, the low-energy model for the surface states of 3D TIs is essentially a (2+1)-dimensional QED, supplemented by certain constraints. The generation of different types of gaps, such as those describing spontaneous parity breaking and chiral symmetry breaking were studied in QED$_{2+1}$ without background electromagnetic fields in Refs.~\cite{Appelquist:1986,Appelquist1:1986,Wijewardhana} a long time ago. Moreover, it was shown that the Dirac mass can be spontaneously generated, while the Haldane mass is energetically disfavored \cite{Appelquist1:1986}. Clearly, this is not the case in the problem at hand, where both types of gaps are generated on the TI surfaces. This is due to the fact that the TR and inversion symmetries are explicitly broken by the external magnetic and electric fields. Furthermore, we find that the Haldane gap dominates at small values of the electric field. This situation is reminiscent of the dynamically enhanced Zeeman splitting in graphene \cite{Gorbar:2011kc}. \section{Conclusion} \label{sec:Conclusion} In this study, we considered the dynamical generation of gaps in a slab of a 3D TI, such as Bi$_2$Se$_3$, placed in the magnetic and electric fields perpendicular to its surfaces. (Although we used the model parameters for Bi$_2$Se$_3$, the main conclusions should be valid for all similar TIs.) Note that the conducting states on the TI slab surfaces and the overall geometry of the system are rather similar to bilayer graphene. On the other hand, the degeneracy connected with the valley and spin degrees of freedom, which is responsible for a variety of quantum Hall states in bilayer graphene, is absent in a TI slab. Still, there are notable similarities in the dynamics of gap generation in these two physical systems. For example, the valley quantum Hall (or layer polarized) state, which is realized in a sufficiently strong external electric field in bilayer graphene, resembles the homogeneous state considered in this paper. By solving the gap equations for the surface quasiparticle propagators in a simple model with short- and long-range interactions, we found that {\em both} the Dirac and Haldane gaps are dynamically generated in the TI slab in external electric and magnetic fields. The underlying mechanism is a different version of the magnetic catalysis. Because of a large surface dielectric constant, the Coulomb interaction appears to play a minor role in the dynamics. Unlike the Dirac gap, the Haldane gap respects the $U(2)$ symmetry with the generators given in Eq.~(\ref{gap-Coulomb-results-generators}), but breaks the parity and TR symmetries. Since both discrete symmetries are explicitly broken by external electric and magnetic fields, the generation of the Haldane gap does not break any symmetries. The Dirac gap, on the other hand, is generated only in the presence of an electric field. The result of such an {\em electrified} magnetic catalysis is a spontaneous breaking of the $U(2)$ symmetry. By comparing our results with the findings in Ref.~\cite{Vishwanath:2010}, we argued that the homogeneous phase with dynamically generated Dirac and Haldane gaps is the true ground state in the TI slab in nonzero magnetic and sufficiently strong electric fields. The precise structure of the phase diagram in the plane of applied electric and magnetic fields remains to be clarified, however. \acknowledgments The authors are grateful to V.P. Gusynin for useful discussions. The work of E.V.G. was supported partially by the Ukrainian State Foundation for Fundamental Research. The work of V.A.M. and P.O.S. was supported by the Natural Sciences and Engineering Research Council of Canada. The work of I.A.S. was supported in part by the U.S. National Science Foundation under Grant No.~PHY-1404232.
bcd73857a29c93a81082cd2b5634975a37343c29
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Gibbs' Canonical Ensemble} From Gibbs' 1902 text {\it Elementary Principles in Statistical Mechanics}, page 183 : \begin{quotation} ``If a system of a great number of degrees of freedom is microcanonically distributed in phase, any very small part of it may be regarded as canonically distributed.'' \end{quotation} Thus J. Willard Gibbs pointed out that the energy states of a ``small'' system weakly coupled to a larger ``heat reservoir'' with a temperature $T$ have a ``canonical'' distribution : $$ f(q,p) \propto e^{-{\cal H}(q,p)/kT} \ . $$ with the Hamiltonian ${\cal H}(q,p)$ that of the small system. Here $(q,p)$ represents the set of coordinates and momenta of that system. `` {\it Canonical} '' means simplest or prototypical. The heat reservoir coupled to the small system and responsible for the canonical distribution of energies is best pictured as an ideal-gas thermometer characterized by an unchanging kinetic temperature $T$ . The reservoir gas consists of many small-mass classical particles engaged in a chaotic and ergodic state of thermal and mechanical equilibrium with negligible fluctuations in its temperature and pressure. Equilibrium within this thermometric reservoir is maintained by collisions as is described by Boltzmann's equation. His ``H Theorem'' establishes the Maxwell-Boltzmann velocity distribution found in the gas. See Steve Brush's 1964 translation of Boltzmann's 1896 text {\it Vorlesungen \"uber Gastheorie}. Prior to fast computers texts in statistical mechanics were relatively formal with very few figures and only a handful of numerical results. In its more than 700 pages Tolman's 1938 tome {\it The Principles of Statistical Mechanics} includes only two Figures. [ The more memorable one, a disk colliding with a triangle, appears on the cover of the Dover reprint volume. ] Today the results-oriented graphics situation is entirely different as a glance inside any recent issue of {\it Science} confirms. \section{Nos\'e-Hoover Canonical Dynamics -- Lack of Ergodicity} In 1984, with the advent of fast computers and packaged computer graphics software already past, Shuichi Nos\'e set himself the task of generalizing molecular dynamics to mimic Gibbs' canonical distribution\cite{b1,b2}. In the end his approach was revolutionary. It led to a new form of heat reservoir described by a single degree of freedom with a logarithmic potential, rather than the infinitely-many oscillators or gas particles discussed in textbooks. Although the theory underlying Nos\'e's approach was cumbersome Hoover soon pointed out a useful simplification\cite{b3,b4} : Liouville's flow equation in the phase space provides a direct proof that the ``Nos\'e-Hoover'' motion equations are consistent with Gibbs' canonical distribution. Here are the motion equations for the simplest interesting system, a single one-dimensional harmonic oscillator : $$ \dot q = (p/m) \ ; \ \dot p = -\kappa q - \zeta p \ ; \ \dot \zeta = [ \ (p^2/mkT) - 1 \ ]/\tau^2 \ . $$ The ``friction coefficient'' $\zeta$ stabilizes the kinetic energy $(p^2/2m)$ through integral feedback, extracting or inserting energy as needed to insure a time-averaged value of precisely $(kT/2)$ . The parameter $\tau$ is a relaxation time governing the rate of the thermostat's response to thermal fluctuations. In what follows we will set all the parameters and constants $(m,\kappa,k,T,\tau)$ equal to unity, purely for convenience. Then the Nos\'e-Hoover equations have the form : $$ \dot q = p \ ; \ \dot p = -q -\zeta p \ ; \ \dot \zeta = p^2 - 1 \ [ \ {\rm NH} \ ] \ . $$ Liouville's phase-space flow equation, likewise written here for a single degree of freedom, is just the usual continuity equation for the three-dimensional flow of a probability density in the ($q,p,\zeta$) phase space : $$ \dot f = (\partial f/\partial t) + \dot q(\partial f/\partial q) + \dot p(\partial f/\partial p) + \dot \zeta(\partial f/\partial \zeta) = -f(\partial \dot q/\partial q) -f(\partial \dot p/\partial p)-f(\partial \dot \zeta/\partial \zeta) \ . $$ This approach leads directly to the simple [ NH ] dynamics described above. It is easy to verify that Gibbs' canonical distribution needs only to be multiplied by a Gaussian distribution in $\zeta$ in order to satisfy Liouville's equation. $$ e^{-q^2/2}e^{-p^2/2}e^{-\zeta^2/2} \propto f_{NH} \propto f_Ge^{-\zeta^2/2} \longrightarrow (\partial f_{NH}/\partial t) \equiv 0 \ . $$ Hoover emphasized that the simplest thermostated system, a harmonic oscillator, does {\it not} fill out the entire Gibbs' distribution in $(q,p,\zeta)$ space. It is not ``ergodic'' and fails to reach all of the oscillator phase space. In fact, with {\it all} of the parameters ( mass, force constant, Boltzmann's constant, temperature, and relaxation time $\tau$ ) set equal to unity only six percent of the Gaussian distribution is involved in the chaotic sea\cite{b5}. See {\bf Figure 1} for a cross section of the Nos\'e-Hoover sea in the $p=0$ plane. The complexity in the figure, where the ``holes'' correspond to two-dimensional tori in the three-dimensional $(q,p,\zeta)$ phase space, is due to the close relationship of the Nos\'e-Hoover thermostated equations to conventional chaotic Hamiltonian mechanics with its infinitely-many elliptic and hyperbolic points. \begin{figure} \includegraphics[width=4.5in,angle=-90.]{fig1.ps} \caption{ The $p=0$ cross section of the chaotic sea for the Nos\'e-Hoover harmonic oscillator. 502 924 crossings of the plane are shown. The fourth-order Runge-Kutta integration used a timestep $dt = 0.0001$. A point was plotted whenever the product $p_{old}p_{new}$ was negative. } \end{figure} \section{More General Thermostat Ideas} New varieties of thermostats, some of them Hamiltonian and some not, appeared over the ensuing 30-year period following Nos\'e's work\cite{b6,b7,b8,b9,b10,b11,b12,b13,b14,b15, b16,b17,b18}. This list is by no means complete. Though important, simplicity is not the sole motivation for abandoning purely-Hamiltonian thermostats. Relatively recently we pointed out that Hamiltonian thermostats are incapable of generating or absorbing heat flow\cite{b6,b7}. The close connection between changing phase volume and entropy production guarantees that Hamiltonian mechanics is fundamentally inconsistent with irreversible flows. At equilibrium Bra\'nka, Kowalik, and Wojciechowski\cite{b8} followed Bulgac and Kusnezov\cite{b9,b10} in emphasizing that {\it cubic} frictional forces, $-\zeta^3p$, which also follow from a novel Hamiltonian, promote a much better coverage of phase space, as shown in {\bf Figure 2} . The many small holes in the $p=0$ cross section show that this approach also lacks ergodicity. \begin{figure} \includegraphics[width=4.5in,angle=-90.]{fig2.ps} \caption{ The $p=0$ cross section of the chaotic sea for an oscillator governed by Bra\'nka, Kowalik, and Wojciechowski's choice of the motion equation, $\ddot q = \dot p = -q -\zeta^3p \ ; \ \dot \zeta = p^2 - 1$ . 20 billion timesteps, with $dt = 0.0001$, resulted in 636 590 crossings of the $p=0$ section, using the integration procedure of Figure 1. } \end{figure} \subsection{Joint Control of Two Velocity Moments} Attempts to improve upon this situation led to a large literature with the most useful contributions applying thermostating ideas with two or more thermostat variables\cite{b9,b10}. An example, applied to the harmonic oscillator, was tested by Hoover and Holian\cite{b11} and found to provide all of Gibbs' distribution : $$ \dot q = p \ ; \ \dot p = -q - \zeta p - \xi p^3\ ; \ \dot \zeta = p^2 - 1 \ ; \ \dot \xi = p^4 - 3p^2 \ {\rm [ \ HH \ ]} $$ The two thermostat variables $(\zeta,\xi)$ together guarantee that both the second and the fourth moments of the velocity distribution have their Maxwell-Boltzmann values [ 1 and 3 ] . Notice that two-dimensional cross sections like those in the Figures are no longer useful diagnostics for ergodicity once the phase-space dimensionality exceeds three. \subsection{Joint Control of Coordinates and Velocities} In 2014 Patra and Bhattacharya\cite{b12} suggested thermostating both the coordinates and the momenta : $$ \dot q = p - \xi q \ ; \ \dot p = -q - \zeta p \ ; \ \dot \zeta = p^2 - 1 \ ; \ \dot \xi = q^2 - 1 \ {\rm [ \ SEPB \ ]} \ . $$ an approach already tried by Sergi and Ezra in 2001\cite{b13}. A slight variation of the Sergi-Ezra-Patra-Bhattacharya thermostat takes into account Bulgac and Kusnezov's observation that cubic terms favor ergodicity : $$ \dot q = p - \xi^3 q \ ; \ \dot p = -q - \zeta p \ ; \ \dot \zeta = p^2 - 1 \ ; \ \dot \xi = q^2 - 1 \ {\rm [ \ PB_{var} \ ]} \ . $$ These last two-thermostat equations appear to be a good candidate for ergodicity, reproducing the second and fourth moments of $(q,p,\zeta,\xi)$ within a fraction of a percent. We have not carried out the thorough investigation that would be required to establish their ergodicity as the single-thermostat models are not only simpler but also much more easily diagnosed because their sections are two-dimensional rather than three-dimensional. \section{Single-Thermostat Ergodicity} Combining the ideas of ``weak control'' and the successful simultaneous thermostating of coordinates and momenta\cite{b14} led to further trials attempting the weak control of two different kinetic-energy moments\cite{b15}. One choice out of the hundreds investigated turned out to be successful for the harmonic oscillator : $$ \dot q = p \ ; \ \dot p = - q -\zeta( 0.05p + 0.32p^3) \ ; \ \dot \zeta = 0.05(p^2 - 1) + 0.32(p^4 - 3p^2) \ [ \ {\rm ``0532 \ Model''} \ ] \ . $$ These three oscillator equations passed all of the following tests for ergodicity : \noindent [ 1 ] The moments $\langle \ p^2 \ \rangle = 1 \ ; \ \langle \ p^4\ \rangle = 3 \ ; \ \langle \ p^6 \ \rangle = 15 $ were confirmed. \noindent [ 2 ] The independence of the largest Lyapunov exponent to the initial conditions indicated the absence of the toroidal solutions. \noindent [ 3 ] The separation of two nearby trajectories had an average value of 6 :\\ $\langle \ (q_1-q_2)^2 + (p_1-p_2)^2 + (\zeta_1-\zeta_2)^2 \ \rangle = 2 + 2 + 2 = 6 $ . \noindent [ 4 ] The times spent at positive and negative values of $\{ \ q,p,\zeta \ \}$ were close to equal. \noindent [ 5 ] The times spent in regions with each of the 3! orderings of the three dependent variables were equal for long times. These five criteria were useful tools for confirming erogidicity. Evidently weak control is the key to efficient ergodic thermostating of oscillator problems. \begin{figure} \includegraphics[width=4.5in,angle=-90.]{fig3.ps} \caption{ $p=0$ cross section for a singly-thermostated quartic oscillator, with motion equations $ \ddot q = \dot p = -q^3 -\zeta p^3 \ ; \ \dot \zeta = p^4 - 3p^2$ . Runge-Kutta integration as in Figures 1 and 2 with 503 709 crossings of the $p=0$ plane. Several hundred singly-thermostated attempts failed to obtain canonical ergodicity for the quartic oscillator. } \end{figure} \section{A Fly in the Ointment, the Quartic Potential} The success in thermostating the harmonic oscillator led to like results for the simple pendulum but {\it not} for the quartic potential\cite{b15}. See {\bf Figure 3}. This somewhat surprising setback motivates the need for more work and is the subject of the Ian Snook Prize for 2016. This Prize will be awarded to the author(s) of the most interesting original work exploring the ergodicity of single-thermostated statistical-mechanical systems. The systems are not at all limited to the examples of the quartic oscillator and the Mexican Hat potential but are left to the imagination and creativity of those entering the competition. \begin{figure} \includegraphics[width=2.0in,angle=-0.]{fig4.ps} \caption{ Shuichi Nos\'e ( 1951-2005 ) and Ian Snook ( 1945-2013 ) } \end{figure} \section{Conclusions -- Ian Snook Prize for 2016} It is our intention to reward the most interesting and convincing entry submitted for publication to Computational Methods in Science and Technology ( www.cmst.eu ) prior to 31 January 2017. The 2016 Ian Snook prize of \$500 dollars will be presented to the winner in early 2017. An Additional Prize of the same amount will likewise be presented by the Institute of Bioorganic Chemistry of the Polish Academy of Sciences ( Poznan Supercomputing and Networking Center ). We are grateful for your contributions. This work is dedicated to the memories of our colleagues, Ian Snook ( 1945-2013 ) and Shuichi Nos\'e ( 1951-2005 ), shown in {\bf Figure 4} . \pagebreak
4df0043dbd096d6c69998163b90855e98340756c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} In recent decades acoustic techniques in solid state physics demonstrate serious progress, especially by moving to previously unattainable high-frequency band, up to a terahertz frequencies \cite{ps-ultrasonics}. Considerable efforts in this direction, called often picosecond ultrasonics, are stimulated by the short-wavelength character of acoustic waves in this band, and, in some cases, efficient coupling of acoustic strain to electronic, optical, magnetic excitations in solid state. This allows application of high-frequency acoustic signals for testing and control of nanodimensional solid-state structures. From practical point of view, the most serious restriction of picosecond ultrasonics is the use of ultrafast (femtosecond) lasers for both excitation of acoustic signals and detection of its coupling to a solid-state nanostructure, usually with the use of pump-probe technique. In spite of considerable improvement of such lasers characteristics and growth of their availability, development of a robust electrically controlled picosecond acoustic technique would be an essential breakthrough in the field. Speaking about the high-frequency acoustic wave excitation, terahertz sasers could be a solution of the problem \cite{saser1,saser2,saser3}. For detection purposes, several options are available. The superconductor based detectors are in use from 70th. The robust bolometers used to be widely employed for acoustic spectroscopy are currently less popular since, in contrast to optical methods, they are hardly sensitive to the spectrum of an acoustic signal. The superconductor contacts do possess spectral selectivity \cite{super-contacts} but their fabrication is quite sophisticated. Semiconductor-based approaches are more preferable. A photo-electric acoustic wave detection by {\it p-i-n} diodes with a quantum well embedded into the {\it i} region demonstrated high efficiency, but, although based on electric current measurement, requires use of femtosecond laser for temporal signal sampling \cite{pin}. An alternative, also semiconductor-based, method using Schottky diodes has been demonstrated recently \cite{schottky}. It is purely electrical and is based on induction of displacement current by propagating acoustic wave. Considering such factors as all-electrical detection principle, use of robust well-studied devices technology which can be integrated with various solid-state structures, possible room-temperature applications, this method looks an attractive candidate for wide use as high-frequency acoustic detector. In this paper the main physical principles of Schottky diode acoustic detection are considered theoretically in details. The developed model allows to address such issues as feasible magnitude of the electrical signal, fundamental restrictions on detectable acoustic signal frequency, possible ways of the diode structure optimization. The paper is organized as follows. In section I the expression for the accumulated electrical charge due to the acoustic strain perturbation is obtained for important cases of piezoelectric and deformation potential coupling. It is used then in section II for analysis of the electrical response of the Schottky diode. Then, the conclusions follow. \section{Expression for the acoustic wave induced charge in a diode} \begin{figure} \centering \includegraphics[width=0.6\linewidth]{fig1} \caption{The schematics of the energy diagram of $n$-type Schottky diode with the used coordinate frame. $z=z_i$ corresponds to the metal-semiconductor interface. The insert shows the model electrical circuit which is used for the electrical detection of acoustic signals. } \label{fig:1} \end{figure} The energy diagram of the Schottky diode is shown in Fig.\ref{fig:1} for particular case of n-doped semiconductor. We consider the range of external biases $V$, for which the Schottky barrier is much higher than temperature measured in energy units. In this case the electrical current is small and it is possible to assume that electron distributions in semiconductor and metal regions correspond to quasi-equilibrium and can be characterized by quasi-Fermi levels shifted by the value of electrical bias $eV$ assuming positive sign for the direct bias of the diode. While an acoustic wave propagates through the structure, the related strain induces the potential acting on electrons. Such a potential can be described within the deformation potential model \cite{Gantmakher-Levinson}. Redistribution of the charge carriers in this potential gives rise to the perturbation of the electric filed in the system. In addition, in a piezoelectric semiconductor electric field is perturbed due to the lattice polarization induced by the acoustic wave. The perturbed potential $\delta \varphi$ satisfies the Poisson equation, which in one-dimensional limit, corresponding to an acoustic wave propagating along $z$-axis which is normal to the flat metal-semiconductor interface, is \begin{equation} \label{eq:Poisson} \frac{d \delta \varphi}{dz} =\frac{e}{\varepsilon_s \varepsilon_0} \delta n + \frac{1}{\varepsilon_s \varepsilon_0} \frac{d P_z}{dz}, \end{equation} where $\delta n$ is perturbation of the electron concentration, $\varepsilon_s$ and $\varepsilon_0$ are the dielectric constant of the semiconductor and the absolute permittivity, and $P_z$ is the $z$-component of the peizoelectric polarization. The important assumption we are going to use is that all perturbations caused by the acoustic wave are much slower than the electron relaxation processes in both metal and semiconductor. The latter can be characterized by the dielectric relaxation time $\varepsilon_{s,m} \varepsilon_0/\sigma$, where $\sigma$ is conductivity and $\varepsilon_m$ is the lattice dielectric permittivity of metal. Such time is usually within subpicosecond band for semiconductor and even shorter for metal. Thus, we may use a quasi-static approach determining $\delta n$ while dealing with sub-terahertz acoustic waves. This means that at any time instant the electron density perturbation is the same as in the case of static nonuniform strain distribution corresponding to this particular time. Specifically, dropping the time dependence for brevity, in the linear approach for semiconductor region $z<z_i$ we have: \begin{equation} \label{eq:dn_s} \delta n (z) = e (\delta \varphi (z) - U_{DP}(z)/e - \delta V_s) \frac{dn_s}{dE_F}, \end{equation} where $n_s (E_F)$ is the electron concentration dependence on the Fermi energy, $U_{DP}$ is the deformation potential energy of electrons, and we allow perturbation of the semiconductor reference potential, $\delta V_s$, caused by the acoustic wave. Note, that the value of the derivative in right hand side of Eq.(\ref{eq:dn_s}) depends on coordinate. Analogously, in metal, $z>z_i$, we have \begin{equation} \label{eq:dn_m} \delta n (z) = e (\delta \varphi (z) - U_{DP}(z)/e - \delta V_m) \frac{dn_m}{dE_F}. \end{equation} With Eqs.(\ref{eq:dn_s},\ref{eq:dn_m}), the Poisson equation becomes a linear inhomogeneous differential equation. It is convenient to perform its solution separately for $z<z_i$ and $z>z_i$, applying then the boundary conditions at $z=z_i$. Using standard variation of constants method and taking into account that $\varphi(z=-\infty) = \delta V_s$, $\varphi(z=\infty) = \delta V_m$ , we obtain \begin{eqnarray} \label{eq:pt-sol} \delta \varphi (z) = \delta V_s +c_s \phi_{s2} (z) +\frac{\phi_{s2} (z)}{w_s} \int_z^{z_i} dz' \phi_{s1}(z') \left(k_s^2 (z') U_{DP}(z')/e- \frac{1}{\varepsilon\varepsilon_0} \frac{d P_z}{dz'}\right) + \\ \frac{\phi_{s1} (z)}{w_s} \int_{-\infty}^{z} dz' \phi_{s2}(z') \left(k_s^2 (z') U_{DP}(z')/e- \frac{1}{\varepsilon\varepsilon_0} \frac{d P_z}{dz'}\right), \mbox{~for~} z<z_i \nonumber \\ \delta \varphi (z) = \delta V_m +c_m \phi_{s2} (z) -\frac{\phi_{m2} (z)}{w_m} \int_{z_i}^z dz' \phi_{m1}(z') k_m^2 (z') U_{DP}(z')/e - \nonumber \\ \frac{\phi_{m1} (z)}{w_m} \int_z^{\infty} dz' \phi_{m2}(z') k_m^2 (z') U_{DP}(z')/e, \mbox{~for~} z>z_i. \nonumber \end{eqnarray} Here $c_s$ and $c_m$ are constants, $k_{s,m}^2 =e^2dn_{s,m}/dE_{F} (\varepsilon_{s,m} \varepsilon_0)^{-1}$, and $\phi_{m1,2}$ are fundamental solutions of the homogeneous versions of equations for $\delta \varphi$: \begin{eqnarray} \label{eq:pt-hom} \frac{d \phi_{s1,2}}{dz^2} =k_s^2 (z) \phi_{s1,2} \mbox{~for~} z<z_i \\ \frac{d \phi_{m1,2}}{dz^2} =k_m^2 (z) \phi_{m1,2} \mbox{~for~} z>z_i \nonumber \end{eqnarray} These functions are selected such that $\phi_{s2} (-\infty) =0$, $\phi_{m2} (\infty) =0$ and Wroskians in Eq.(\ref{eq:pt-sol}) are $w_{s,m} = \phi_{s,m1} \phi'_{s,m2} - \phi'_{s,m1} \phi_{s,m2}$. The constants $c_s$ and $c_m$ are determined via the boundary conditions at $z=z_i$, requiring continuity of potential and electrical induction. Then, it is straightforward to calculate the perturbation of the accumulated charge, $\delta Q$: \begin{equation} \label{eq:charge-def} \delta Q=\varepsilon\varepsilon_0 S \int_{z_i}^{\infty} dz k_m^2 (-\delta \varphi (z) + U_{DP}(z)/e + \delta V_m), \end{equation} where $S$ is the diode cross-section. After some algebra from the expressions for the potential we obtain \begin{eqnarray} \label{eq:charge-expr} \delta Q=C \left( \delta V - V_{PZ}(z_i) + \int_{-\infty}^{z_i} dz G_s (z) \left( V_{DP}(z) +V_{PZ} (z)\right) - \right. \nonumber \\ \left. \int_{z_i}^\infty dz G_m (z) V_{DP} (z) \right), \end{eqnarray} where we introduced the effective potential due to the deformation potential acousto-electric coupling $V_{DP} \equiv - U_{DP}/e$, potential induced due to poiezoelectric action of the aoustic wave $V_{PZ}$ such that $V'_{PZ} = P_z/(\varepsilon_s \varepsilon_0)$, the kernel functions \begin{eqnarray} \label{eq:kernel} G_s (z)= \frac{1}{\phi'_{s2}(z_i)}\phi_{s2}(z)k_s^2(z), \\ G_m (z)= \frac{1}{w_m} \left( \phi_{m1}(z_i) \phi'_{m2} (z_i)-\frac{\varepsilon_s}{\varepsilon_m}\phi_{m2} (z_i) \phi'_{m1}\right) \frac{1}{\phi'_{m2}(z_i)} \phi_{m2}(z)k_m^2(z) \nonumber \end{eqnarray} and the diode capacitance $C=\varepsilon_s\varepsilon_0 S/L_{eff}$ with \begin{equation} \label{eq:thickness} L_{eff}=\frac{\phi_{s2}(z_i)}{\phi'_{s2}(z_i)} - \frac{\varepsilon_s}{\varepsilon_m} \frac{\phi_{m2}(z_i)}{\phi'_{m2}(z_i)}. \end{equation} In Fig.\ref{fig:kernel} we plot the spatial dependence of the kernel function $G_s$ calculated for GaAs Schottky diodes with doping $10^{17}~cm^{-3}$ and $10^{18}~cm^{-3}$ and temperatures $10K$ and $300K$. The steady-state potential profile and the screening parameter were determined with the standard approach assuming low value of the diode current \cite{Sze}. As we see, the charge is controlled by the perturbation near the edge of the depletion layer. This result is expectable: indeed, the used boundary conditions assume no acoustic perturbation for $z=-\infty$. In this case although variation of strain inside the spatially uniform portion of semiconductor leads to charge redistribution, it does not change the total charge in it. Only if strain changes near the inhomogeneous region near the edge of the depletion layer, the total charge experiences the perturbation. For comparison, we show the kernel function for a rough model of step-like dependence of $k_s$, where it is set to zero in the depletion region and to the value of the bulk semiconductor to the left of its edge, assumed to be infinitely sharp. The approximation allows analytical determination of $G_s$. As we see, for semiconductor this model is not very good, especially at room temperature where depletion region edge is not well-defined. However, it is good for the metal region since here any energetic perturbation is much less then the Fermi energy. As a result, for metal we can use the analytical expression for $G_m$, which is $G_m=k_m \exp (-k_m (z-z_i))$ for $z>z_i$. It is important to mention useful normalization conditions, which hold for any distribution of potential in the diode: \begin{eqnarray} \label{eq:kernel-normailzation} \int_{-\infty}^{z_i} G_s (z) dz = 1, \\ \int_{z_i}^{\infty} G_m (z) dz = \xi_m \equiv \frac{1}{w_m} \left( \phi_{m1}(z_i) \phi'_{m2} (z_i)-\frac{\varepsilon_s}{\varepsilon_m}\phi_{m2} (z_i) \phi'_{m1}\right) \nonumber \end{eqnarray} \begin{figure} \centering \includegraphics[width=0.7\linewidth]{kernels} \caption{The kernel function $G_s$ for GaAs Schottky diodes with doping level $10^{17}$~cm$^{-3}$ and $10^{18}$~cm$^{-3}$ and different temperatures. $z=0$ corresponds to the metal-GaAs interface, {\it i.e.} $z_i=0$. For comparison, the results for model step-like spatial dependence of $k_s^2$ are shown. } \label{fig:kernel} \end{figure} Let us discuss the in some details the deformation potential and piezoelectric couplings. For semiconductor-contribution this is straightforward. The deformation coupling describes the shift of the bottom of the conduction band minima. Its specific form depends on the crystal symmetry and the momentum position of the conduction band \cite{Ivchenko-Pikus}. In any case, $V_{DP}$ is proportional to strain. Below, to be specific, we will provide expressions for the case of GaAs with $z$-axis parallel to its [111] crystallographic direction and longitudinal acoustic wave propagating along $z$. In this case \begin{equation} \label{eq:DP111} V_{DP}= \frac{E_1}{e} u_{zz}, \end{equation} where $E_1$ is the deformation potential constant and $u_{zz}$ is the only present component of strain. The piezoelectric potential is determined by the strain-induced piezoelectric polarization. For the mentioned geometry and acoustic wave polarization we obtain \begin{equation} \label{eq:PZ111} V_{PZ}= \frac{2 e_{14}}{\sqrt{3}\varepsilon \varepsilon_0} u_z, \end{equation} where $e_{14}$ is the piezoelectric constant of a cubic material and $u_z$ is the only present component of displacement in the considered longitudinal acoustic wave. As we see, the piezoelectric effect induces charge not only because of charge redistribution, but also due to direct induction of potential (the second term in the brackets of Eq.(\ref{eq:charge-expr})). It is worth to mention that for the case of different crystallographic orientation, crystal symmetry or acoustic wave polarization the general structure of the expressions for deformation and piezoelectric potentials remains the same with the former proportional to strain and the latter proportional to displacement. Of course, in some cases some contributions vanish. For example, in GaAs there is no piezoelectric coupling for acoustic wave of any polarization propagating along [100] direction; deformation potential in this case is absent for transverse wave. For the metal region consideration of the coupling of acoustic wave to electrons is more complicated than for semiconductor. This is because the deformation potential in a metal is considered as a perturbation of electron spectrum in some momentum point near the Fermi surface. Therefore, this value is, strictly speaking, momentum dependent. While considering screening, a momentum-averaged value is introduced to determine the charge perturbation \cite{Gantmakher-Levinson,Abrikosov}. Its dependence on the strain components is determined by the symmetry of the metal Fermi surface. In fact, the corresponding constants are hardly known. This is because experiments on electron transport or ultrasound attenuation in metals provide {\it screened} value of electron-phonon coupling averaged in a specific way \cite{Abrikosov}. In the following, we will use in metal \begin{equation} \label{eq:DP-metal} V_{DP}= \frac{E_m}{e} u_{zz}, \end{equation} keeping in mind that the effective constant $E_m$ has specific value dependent on the metal crystallographic orientation (for metal single crystals) and acoustic wave polarization. By the order of magnitude, one can expect $E_m$ to be about few electronvolts. It is worth mentioning that in general we should not discard piezoelectric-like coupling in metal. It is usually done while considering electron scattering by phonons since efficient screening in metals cancels any macroscopic potential. However, the magnitude of the space charge induced under the screening does not vanish. In particular, this is seen from the expression for $G_m$, which provides finite value for the induced charge regardless of large value of $k_m$. In the following we do not include piezoelectric contribution in metal into consideration since no info is available of its presence and strength. However, one has to keep in mind that high-frequency acoustic wave detection by Schottky diode could reveal possible piezoelectric-like coupling in metals. In principle, it can be distinguish from the deformation potential, since, similar to the semiconductor, the it should be proportional to the displacement rather than strain \section{Detection of the acoustic wave by the diode} Naturally, the signal induced by an acoustic wave passing through the diode depends both on its intrinsic characteristics and the properties of the electrical circuit which includes the Schottky diode. We consider simple model circuit consisting of the diode and series resistance $R$ (see the insert of Fig.\ref{fig:1}). Using Eq.(\ref{eq:charge-expr}) we can easily obtain equation for $\delta V$: \begin{eqnarray} \label{eq:circuit} \frac{d \delta V}{dt} +\frac{\delta V}{RC} =\frac{dS}{dt} \\ S=\left( V_{PZ}(z_i) - \int_{-\infty}^{z_i} dz G_s (z) \left( V_{DP}(z) +V_{PZ} (z)\right) + \xi_m V_{DP} ^{(m)}(z_i) \right), \nonumber \end{eqnarray} where the right hand side can be considered as a source caused by an acoustic wave, smallness of the screening length in the metal is taken into account, and superscript $(m)$ indicates the deformation potential in the metal. The particular form of the acoustic signal depends on the kind of the acoustic source. In high-frequency band the most popular one is a bipolar strain pulse generated with the use of picosecond ultrasonics technique \cite{ps-ultrasonics}. Alternatively, quasi-monochromatic acoustic waves can be produced by semiconductor superlattices illuminated by femtosecond laser pulses \cite{ps-ultrasonics} or sasers \cite{saser1,saser2,saser3}. Since in the linear response regime any acoustic signal can be presented as a plane wave superposition, in Eq.({\ref{eq:circuit}) we switch to the frequency domain and obtain \begin{equation} \label{eq:circuit-freq} \delta V_\omega = \frac{1}{1+i (\omega RC)^{-1}} S_\omega. \end{equation} The intrinsic detection properties of the diode are reflected by the frequency dependence of $S_\omega$. In fact, it is determined by the spatial broadening of the kernel function $G_s$. Assuming the plane-wave strain, we obtain \begin{equation} \label{eq:S_omega} S_\omega= - i \tilde{V}_{PZ} \left(1-J_s \exp(i\theta)\right) +\xi_m \tilde{V}_{DP}^{(m)} - \tilde{V}_{DP}^{(s)}J_s \exp(i\theta), \end{equation} where $\tilde{V}_{PZ}$ and $\tilde{V}_{DP}^{(s,m)}$ are the amplitudes of the piezoelectric and deformation potentials (with the superscript labeling semiconductor and metal contributions). For the specific case of [111]-oriented semiconductor (Eqs.(\ref{eq:DP111},\ref{eq:PZ111},\ref{eq:DP-metal})) we have $\tilde{V}_{PZ}=2 e_{14} u_{zz}^{(0)}s \left(\sqrt{3} \varepsilon_s \varepsilon_0 \omega\right)^{-1}$ and $\tilde{V}_{DP}^{(s,m)}= E_{s,m} u_{zz}^{(0)}/e$, where $s$ is sound velocity and $u_{zz}^{(0)}$ is the strain amplitude. In Eq.(\ref{eq:S_omega}) the overlap integral is introduced: \begin{equation} \label{eq:overlap} J_s \exp (i\theta) =\int_{-\infty} ^{z_i} dz G_s(z) \exp (i\omega z/s). \end{equation} \begin{figure} \centering \includegraphics[width=0.7\linewidth]{freq-sens-amp} \caption{The calculated overlap $J_s$ for various doping levels and temperature. } \label{fig:overlap} \end{figure} The calculated frequency dependence of $J_s$ is shown in Fig.\ref{fig:overlap}. As it is expected, $J_s$ is suppressed for frequencies corresponding to the acoustic wavelength smaller than the spatial localization length of the kernel $G_s$. The frequency dependence of $\theta$, which is not shown in a graph, reflects the phase shift of the acoustic signal at the edge of the depletion layer and at the metal-semiconductor interface and corresponds roughly to $2\pi$ variation for frequency increase about $90$ and $26$~GHz for doping $10^{18}$ and $10^{17}$~cm$^{-3}$, respectively. \begin{figure} \centering \includegraphics[width=0.7\linewidth]{freq-sens-amp-pz} \caption{The calculated value of $J_s^{(PZ)}$ for various doping levels and temperature. The lines' legend is the same as in Fig.\protect\ref{fig:overlap}. } \label{fig:overlap-pz} \end{figure} If piezoelectric coupling is present in the structure, it commonly exceeds the deformation one for frequencies below a hundred gigahertz. For separate analysis of the piezoelectric contribution it is convenient to introduce the value $J_s^{(PZ)} \equiv |1 -J_s \exp (i \theta)|$. The frequency dependence of $J_s^{(PZ)}$ is shown in Fig.\ref{fig:overlap-pz}. Naturally, it shows resonances corresponding to in-phase perturbation at the edge of the semiconductor depletion region and metal-semiconductor interface. Positions of these resonances can be easily predicted since the piezoelectric contribution to the diode response is determined by the parameters of semiconductor only, which are usually well-known. It is worth to mention a special case of piezoelectric coupling and relatively low frequency acoustic wave, for which the acoustic wavelength is larger than both the broadening of $G_s$ and the thickness of the depletion layer. Here, $S$ becomes proportional to strain. So, for the particular case of Eq.(\ref{eq:PZ111}) we have $S= 2 e_{14} u_{zz} L_{eff}(\sqrt{3}\varepsilon_0 \varepsilon_s)^{-1}$. If, in addition, if $(RC)^{-1}$ exceeds considerably the characteristic acoustic frequency, $\delta V = S$. In other words, the electrical signal measures directly the value of strain in near-interface region. For doping $10^{18}$~cm$^{-3}$, this approach can be valid for frequency up to several tens of gigahertz. In diodes where piezoelectric coupling is absent, for example those employing non-piezoelectric semiconductors, like Si or Ge, or grown along certain crystallographic directions, like [001] GaAs, the situation is different. The resonances are expected in this case as well, but their location is difficult to predict because of unknown value of the effective deformation potential constant in metal. For higher frequencies the deformation potential coupling is most efficient. In addition, as we see from Fig.\ref{fig:overlap}, the semiconductor contribution is suppressed for high frequencies. However, the metal contribution persists for any realistic frequency. This means that the actual frequency restrictions are set by the ability of high-frequency electronics to measure the high-frequency electric signals. Summarizing the obtained results we can conclude that the acoustic wave detection by Schottky diodes can be described by a simple model where electrical response of the diode is caused by the displacement current induced by electrons screening the strain-induced perturbation. The actual upper frequency limit is set by the parameters of the current-registering equipment rather than internal diode properties due to the fast electronic response and small screening length in metal contact of the diode. On the other hand, the semiconductor-side signal contributions are efficient, for common diode structures for frequencies below few hundreds of gigahertz. These results will be an important guide for interpretation of the measured electrical diode response to an acoustic perturbation as well as optimization of the Schottky diode acoustic wave detectors. \begin{acknowledgments} \end{acknowledgments}
fe9f0cbedace3b1bc230475543afbb1bc2881730
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} Atmospheric metals are not expected to be present in isolated white dwarfs (WDs) with effective temperatures below $\sim25,000$ K. At these temperatures, radiative forces become too weak \citep{chayer95} to significantly counteract the quick gravitational settling that sinks material heavier than helium in extremely short timescales compared to the typical cooling ages of WDs \citep{FM79,K09}. However, it has been found that $\sim25\%-50\%$ of all field WDs exhibit spectral lines that are indicative of the presence of metals in their atmospheres \citep{Z03,Z10,K14}. The high-metallicity material found in the atmospheres of most of these ``polluted'' WDs is consistent with the composition of rock-forming material \citep{Z07,G12,farihi13,JY14}. This observation suggesting that pollution comes from minor rocky bodies (e.g., asteroids). One possibility is that these rocky bodies get very close to the WD so they can be tidally disrupted and then accreted. Further support of this picture comes from observations of circumstellar disks --revealed by infrared excess in the stellar spectrum-- around many polluted WDs (see \citealt{farihi16} for a recent review). These disks orbit within ${\sim}1R_\odot$, roughly the distance at which the material would reside after the tidal disruption (the Roche radius). All the WDs with detected disks have atmospheric pollution. More recently, this picture has been reinforced by the recent observation of minor bodies transiting the polluted WD 1145+017 \citep{vanderburg15,alonso16,G16,rap16,xu16}. Although the leading explanation for WDs pollution --the accretion of tidally disrupted asteroids-- seems robust and well supported by observations, the underlying dynamical mechanism responsible for placing these rocky bodies in star grazing orbits remains much less constrained and understood. A better understanding of this mechanism can lead to new insights into initial conditions leading to WD pollution, as well as into the long-term dynamics and evolution of the planetary systems around WDs and/or their progenitors (typically A and F stars; see \citealt{veras16} for a recent review on this subject). A theoretical model to explain the WD pollution from planetary dynamical instabilities was put forward by \citet{DS02}. According to their model, a planetary system that is marginally stable throughout the main sequence can become unstable due to stellar mass loss during post-MS evolution. This global instability can then promote some asteroids into star grazing orbits. This idea has been explored in more detail using realistic numerical $N$-body integrations of multi-planet systems (no asteroids) and stellar evolution \citep{veras13,MVV14,VG15}. Similarly, the mass loss of the host star can widen the region around mean-motion resonances where chaotic diffusion of asteroids acts efficiently, leading to their posterior tidal disruption \citep{B11,DWS12,FH14}. As well, mass loss in close binary systems can drive the outermost planetesimals into the chaotic orbits, with one of the possible outcomes being collisions with either one of the stars \citep{kratter12}. Thus far, these proposed dynamical mechanisms rely on generally short-timescale instabilities (either scattering or mean-motion-resonance overlap) triggered (or enhanced) by mass loss or simply by the aging of the planetary systems, and still face some difficulties. In particular, these mechanisms are subject to the following constraints: \begin{enumerate} \item {\it the delivery of material must happen for WDs of all ages.} The observations seem to show that neither the rate of polluted WDs, nor that the level of pollution decreases with the WD cooling age \citep{KGF14,wyatt14}. Thus, to explain the observed pollution rate, the underlying mechanism should be able to deliver enough material into the WD's atmosphere independently of how much time it has passed since the stellar mass loss phase. \item {\it The supply of material into white dwarf grazing orbits must be a steady process.} Both the large observed rate of polluted WDs and the short timescales that follow a disruption event (or order the orbital timescale) require of a sustained process to deliver bodies toward disruption. The formation of a debris disks following disruption can extend the duration of the delivery toward the stellar atmosphere, but its associated timescale is still short compared to the cooling ages of most polluted WDs \citep{veras14b,veras15}. \item {\it The reservoir of rocky material has to be long-lived.} The amount of material waiting to be delivered toward the star cannot be arbitrarily large. A planetesimal disk can be destroyed by a collisional cascade, shattering the rocky bodies down to dust, which can be blown out during the RG and AGB phases by radiation pressure (e.g., \citealt{BW10}). All else being equal, disks with lower surface densities and at larger separations can survive for longer timescales, possibly avoiding this fate (e.g., \citealt{wyatt07,HT10,BW10}). \end{enumerate} In this paper, we propose a new mechanism that overcomes (or at least alleviates) these difficulties. We propose that the nature of the instabilities, which drives the material in a planetesimal disk into disrupting orbits, is secular (not scattering nor driven by mean-motion resonances) and that the instabilities are initiated only at the very end of the the stellar evolution (AGB phase) once a stabilizing, pre-existing planetary system is engulfed by an extended stellar envelope. This mechanism can provide steady pollution over all ages of the WD (overcoming the difficulties 1 and 2), while working for a low surface density disk that remains dynamically cold during the main sequence, and that gets gradually depleted long after mass loss has taken place (addressing difficulty 3). We illustrate how the instabilities arise due to the Kozai-Lidov (KL) mechanism in wide ($\gtrsim100$ AU) stellar binaries, although our proposal is more general and sub-stellar companions and other sources of secular excitation are allowed. We expect that for these wide binaries the possible WD pollution associated with post-AGB dust disks and stellar winds might be negligible (e.g., \citealt{ruyter06,VW09,bilik12,clayton14}). \section{Planet Engulfment as a Trigger for ``Dormant" Secular Instabilities} \subsection{Planetary Systems as Suppressors of Secular Instabilities} White dwarf pollution by tidally disrupted minor rocky bodies requires a mechanism to deliver asteroids from distant orbital separations into the stars's tidal disruption radius ($\sim1R_\odot$). Nearly radial orbits may result from secular instabilities, which in some cases are capable of exciting eccentricities up to values of $\sim1$. One well-known example of such instabilities is the KL mechanism (\citealt{kozai,lidov}; see \citealt{naoz16} for a recent review), which takes place when a distant stellar-mass companion is highly inclined respect to the orbit of the minor body. However, it is also known that additional bodies in the system may affect or entirely suppress the effect of the KL mechanism \citep[e.g.][]{holman97}. In the simplest scenario of one planet in a circular orbit with mass $M_{\rm p}$ and semi-major axis $a_{\rm p}$ inside a planetesimal's orbit ($a_{\rm p}<a$), the effect of the additional quadrupole potential due to the planet's time-averaged orbit will overcome that of the outer stellar companion if the planet-induced apsidal precession frequency \begin{eqnarray} \dot{\varpi}_\mathrm{in} \simeq \frac{1}{2} n \left(\frac{M_{\rm p}}{M_{\rm s}}\right)\left(\frac{a_{\rm p}}{a}\right)^2 \approx n\epsilon\in, \label{eq:v_in} \end{eqnarray} is larger than that induced by a binary with mass $M_{\rm b}$ and semi-major axis $a_{\rm b}$ of \begin{eqnarray} \label{eq:omega_out} \dot{\varpi}_\mathrm{out} \simeq n \left(\frac{M_{\rm b}}{M_{\rm s}}\right)\,\left(\frac{a}{a_{\rm b}}\right)^3(1-e_{\rm b}^2)^{-3/2} = n\epsilon_\mathrm{out}, \label{eq:varpi_out} \end{eqnarray} with $n$ being the mean motion frequency of the planetesimal and where we have used the definition of two dimensionless quantities $\epsilon\in$ and $\epsilon_\mathrm{out}$ that represent the relative strength of the tidal potentials (see the Appendix and \citealt{mun15b}). When $\dot{\varpi}_\mathrm{in}=\dot{\varpi}_\mathrm{out}$, then $a=r_L$, where $r_L$ is the ``Laplace radius'', defined as \begin{eqnarray} r_{\rm L}&\equiv&\left(\frac{M_{\rm p}}{2M_{\rm b}}a_{\rm p}^2 a_{\rm b}^3\left[1-e_{\rm b}^2\right]^{3/2} \right)^{1/5}\nonumber\\ &\simeq&16.2 ~\mbox{AU}~\left(\frac{M_{\rm p}}{M_J}\right)^{1/5} \left(\frac{M_{\rm b}}{0.5M_\odot}\right)^{-1/5} \left(\frac{a_p}{2~\mbox{AU}}\right)^{2/5}\nonumber\\ &&\times\left(\frac{a_{\rm b}\sqrt{1-e_{\rm b}^2}} {600\sqrt{1-0.5^2}~\mbox{AU}}\right)^{3/5}. \label{eq:rl} \end{eqnarray} For $a<r_L$, the dynamics of the asteroid will be dominated by the planet's quadrupole potential, such that the planetesimal's angular momentum vector $\propto\mathbf{j}$ will precess around the planet's, with perfect alignment being the equilibrium solution. Conversely, for $a>r_L$, the dynamics of the asteroid will be dominated by the binary companion, with $\mathbf{j}$ precessing around the binary's angular momentum vector (with the possibility of being Kozai-unstable), with perfect alignment being the equilibrium solution. The smooth transition between these two regimes place takes rapidly around $a\simeq r_L$, and the general equilibrium solution of the equilibrium inclination $i_{\rm eq}$ for all values of $a$ is known as the ``Laplace surface". For a test particle in a circular orbit, the Laplace surface is given by (e.g., \citealt{tremaine09,tamayo13}): \begin{eqnarray} \tan 2 i_{\rm eq}=\frac{\sin 2i_{\rm b}}{\cos 2 i_{\rm b}+2(r_{\rm L}/a)^5} \label{eq:i_lap} \end{eqnarray} where $i$ ($i_{\rm b}$) is the inclination of the test particle (binary) relative to the planetary system Thus, for as long as there is a planet (or a planetary system) such that $r_L$ is large enough to accommodate a (nearly) coplanar population of planetesimals/asteriods, such bodies will be protected from the tidal potential from the binary companion, largely ignoring its presence throughout the main sequence (MS) evolution of the host star. \subsection{Triggering of secular instabilities} \label{sec:trigger} Any reduction of the quadrupole potential due to the planet will reduce the extent of the ``safe zone" defined by the Laplace radius, progressively exposing bodies to the influence of the binary companion's tidal potential. One possible cause of such a change is planetary engulfment during the post main sequence stages of stellar evolution. During the red giant branch (RGB) and asymptotic giant branch (AGB) phases of the post-MS, low-to-intermediate mass stars can reach radii of a fraction of, or up to few AU, presumably engulfing all planets within this distance \citep{MV12,villa14}. In particular, the AGB phase is during which most of the mass in the stellar envelope is lost, resulting in the expansion of all the orbits in the system. The engulfment of a planet consists of both its evaporation and its spiraling in during stellar expansion (potentially aided by the tidal interaction with the extended stellar envelope; \citealp{villa14}). For simplicity, here we simply model the engulfment as the gradual reduction of the planetary semi-major axis $a_p$: \begin{equation} a_p(t)=a_{p,0}e^{-t/\tau_a}\;\;\; \text{ for }\; t_{\rm MS}<t<t_{\rm WD} \label{eq:aplanet} \end{equation} where $\tau_a$ represents the in-spiral timescale of the planet, $t_{\rm MS}$ is the duration of the stellar MS and $t_{\rm WD}$ is the time at which the WD is formed. As $r_L\propto a_{\rm p}^{2/5}$ (Eq.~\ref{eq:rl}), planetary engulfment causes the Laplace radius the decrease. In addition, we consider mass loss. If the expulsion of the stellar outer layers happens on timescales much longer than all the orbital periods in the system, then angular momentum conservation dictates that all semi-major axes evolve as $a/\dot{a}=-M_{\rm s}/\dot{M_{\rm s}}$ \citep[e.g.,][]{hadjidemetriou63,veras11}. Thus, introducing another timescale $\tau_{\rm ml}$ we can write: \begin{equation} M_{\rm s}(t)=\left\{ \begin{array}{lc} M_{\rm s,0}e^{-t/\tau_{\rm ml}}\;\text{ if }& M_{\rm s,0}e^{-t/\tau_{\rm ml}}> M_{\rm WD}\\ M_{\rm WD} & \sim \end{array} \right. \label{eq:massloss} \end{equation} for $t>t_{\rm MS}$. Which in turn implies $a(t)=a(0)e^{t/\tau_{\rm ml}}$ and $a_b(t)=a_{b}(0)(M_{{\rm s},0}+M_{\rm b})/(M_{{\rm s},0}e^{-t/\tau_{\rm ml}}+M_{\rm b})$ for $t_{\rm MS}<t<\tau_{\rm ml}\ln(M_{{\rm s},0}/M_{\rm WD})$. The effect of mass loss has an opposite effect to engulfment on the value of the Laplace radius ($r_L\propto a_b^{3/5}$). For $M_b\ll M_s$, one can write an approximate evolution of the Laplace radius: \begin{eqnarray} \frac{r_{\rm L}(t)}{r_{\rm L}(0)}= \exp{\Big[-\cfrac{2t}{5\tau_{\rm a}}+\cfrac{3t}{5\tau_{\rm ml}}\Big]}, ~t_{\rm MS}<t<\tau_{\rm ml}\ln\left(\tfrac{M_{{\rm s},0}}{M_{\rm WD}}\right)\nonumber\\ \end{eqnarray} which shows how mass loss and engulfment have opposite effects. Ultimately, engulfment is the determining factor, since $r_L,\rightarrow0$ as $a_{\rm p}\rightarrow0$, while mass loss stops when the stellar remnant mass reaches $M_{\rm WD}$\footnote{Note that a planet is engulfed when it reaches the stellar envelope at $a_{\rm p}\simeq R_s$ at some point of the evolution (e.g., \citealt{MV12,villa14}), which is equivalent to setting $a_{\rm p}=0$ in $r_{\rm L}$ because the quadrupole moment provided by the planet vanishes after engulfment (i.e., $\dot{\varpi}_{\rm in}=0$ in Equation [\ref{eq:v_in}]).}. The reduction of $r_L$ has a direct impact on the shape of the Laplace surface in Eq.~(\ref{eq:i_lap}). (1) If the ``parameter" $r_L$ changes {\it very slowly}, the $i_{\rm eq}$ curve will evolve smoothly, and any bodies initially lying on the Laplace surface will remain on the final Laplace surface\footnote{Note that a dynamical solution can ``follow'' a slowly changing fixed point {\it provided} this equilibrium remains stable. If the stability of the fixed point changes -- i.e., it experiences a bifurcation -- then adiabaticity is broken. The Laplace surface is known to become unstable in eccentricity (circular orbits are not allowed) in a narrow region of $a$ around $r_L$ only if $i_{\rm b}\gtrsim 69^\circ$ \citep{tremaine09} }. This implies full coplanarity with the binary companion when $r_L\rightarrow0$, i.e., $i_{\rm final}\sim i_{\rm eq}\rightarrow i_{\rm b}$. (2) On the contrary, if $r_L$ is changed {\it very rapidly}, then the bodies initially lying on the Laplace surface will not be able to catch up, thus being frozen in their initial inclinations $i_{\rm final}\sim i_{\rm initial}\sim0$. These two limits entail completely opposite consequences for the planetesimals: (1) if the planet is removed slowly, planetesimals will reach coplanarity with the companion, thus being Kozai-stable; (2) if the planet is removed suddenly, minor bodies will see their initial inclinations unchanged, thus being susceptible to KL oscillations. \subsubsection{Adiabaticity}\label{sec:adiabaticity} The speed at which $r_L$ is changed (either ``very slow'' or ``very fast'' as defined below) will determine whether configurations near equilibrium evolve ``adiabatically" or not \citep[e.g.][]{landau}. Quantitatively, the rate of change in $\epsilon\in$ (or in $r_L$) must be much smaller than the linear oscillation frequency $\omega_0$ around the equilibrium solution of the Laplace surface. Rewriting eq. (31) in \citet{tremaine09} as \begin{equation}\label{eq:osc_freq} \begin{split} \omega_0&=\frac{3}{2\sqrt{2}}n\epsilon\in\Bigg[ 1+\cos2i_{\rm eq}-\sin2i_{\rm eq} \\ &~~~~~~~~~+\frac{1}{2}\left(\frac{\epsilon_\mathrm{out}}{\epsilon\in}\right) \left(\cos2i_{\rm eq}+2\cos2i_{\rm b}\right)\\ &~~~~~~~~~+\frac{1}{2}\left(\frac{\epsilon_\mathrm{out}}{\epsilon\in}\right)^2\cos^2i_{\rm b} \Bigg]^{1/2}~~. \end{split} \end{equation} The degree of adiabaticity can be represented by the ratio $|(\dot{r_L}/r_L)|/\omega_0$ \citep{landau}, which is roughly \begin{equation} \frac{2}{3}\frac{\tfrac{2}{5}\tau_a^{-1}}{n\epsilon\in} \sim \frac{1}{\dot{\varpi}\in\tau_a} \;\;\text{if}\;\;\;\epsilon_\mathrm{out}\ll\epsilon\in~~, \end{equation} or \begin{equation} \frac{4}{3}\frac{\tfrac{2}{5}\tau_a^{-1}}{n\epsilon_\mathrm{out}\cos i_{\rm b}}\sim\frac{1}{\dot{\varpi}_\mathrm{out} \tau_a\cos i_{\rm b}}\sim \frac{\tau_{\rm KL}}{\tau_{\rm a}\cos i_{\rm b}} \;\;\text{if}\;\;\;\epsilon\in\ll\epsilon_\mathrm{out}~~, \end{equation} where we have introduced the KL timescale \citep[e.g.,][]{anto15} \begin{equation}\label{eq:tau} \begin{split} \tau_{\rm KL}&=\frac{16}{15\pi}\left(\frac{a_{\rm b} \sqrt{1-e_{\rm b}^2}}{a}\right)^{3} \frac{M_{\rm s}}{M_{\rm b}}P\\ &\simeq2.3~ \mbox{Myr} \bigg(\frac{a}{10\mbox{ AU}}\bigg)^{-\tfrac{3}{2}} \bigg(\frac{a_{\rm b}}{600\mbox{ AU}}\bigg)^{3}({1-e_{\rm b}^2})^{3/2}\\ &~~~\times\bigg(\frac{M_{\rm s}}{M_\odot}\bigg)^{\tfrac{1}{2}} \bigg(\frac{M_{\rm b}}{M_\odot}\bigg)^{-1}. \end{split} \end{equation} Thus, after $r_L$ has shrunk below $a$, the two relevant timescales to compare will be $\tau_{\rm a}$ and $\tau_{\rm KL}$, and for evolution on the Laplace surface to be adiabatic, it is required that $\tau_{\rm a}\gg\tau_{\rm KL}$. As we will see below, this requirement will be rarely met by planet engulfment, and thus adiabaticity will be most commonly broken as $r_L\rightarrow0$. For this reason, in most situations, planetesimals will be suddenly exposed to the KL mechanism once planet engulfment has been completed (see Fig.~\ref{fig:phases} for a schematic depiction). Thus, only after engulfment has removed the protection against eccentricity excitation, will the external perturber be enabled to place planetesimals into orbits leading to their disruption. \subsection{Toy Model of Four Bodies with Octupole Terms} In the absence of a planet, planetesimals will be susceptible to KL oscillations induced by the stellar binary provided that the relative inclination between the planetesimal and the binary $i_{\rm b}$ is in the range of $40^\circ-140^\circ$. Nevertheless, for planetesimals with $a\sim10$~AU to reach the tidal disruption distance of $\sim R_\odot$ at pericenter, their eccentricities must be such that $a(1-e)\sim R_\odot=5\times10^{-3}$~AU, i.e $(1-e)\sim 10^{-3}-10^{-4}$, which is difficult to achieve during quadrupole-order KL oscillations. The eccentric KL mechanism (\citealp{naoz16}, which includes octupole-order terms in the tidal potential when $e_{\rm b}\neq0$), on the other hand, is capable of producing such high eccentricities, thus satisfying the requirements for tidal disruption of planetesimals. In order to study the evolution of planetesimals during the stellar MS and through the post-MS we integrated the (double-averaged) secular equations of a hierarchical four-body system \citep{ham15,mun15b} consisting of a star (of mass $M_{\rm s}$), a gas giant (of mass $M_{\rm p}$ and semi-major axis $a_{\rm p}$), a massless particle (semi-major axis $a$) and a distant companion (of mass $M_{\rm b}$ and semi-major axis $a_{\rm b}$). The initial setup is reminiscent of that of \citet{mun15b} but in this case we have included octupole-level terms in the potentials following \citet{ham15} (see the Appendix of the current manuscript). \subsubsection{Equations of Motion} We evolve the planetesimal's dimensionless specific angular momentum vector ${\mathbf{j}}=\sqrt{1-e^2}~\hat{\bf{j}}$ and eccentricity vector $\mathbf{e}$ according to the equations of motion \citep[e.g.][]{tremaine14}: \begin{subequations}\label{eq:eom} \begin{align} \label{eq:motion_tp_j} \frac{d{\mathbf{j}}}{dt}=\frac{1}{\sqrt{\mathcal{G}M\in a}}\Big( {\mathbf{j}}\times\nabla_{\mathbf{j}} \Phi +\mathbf{e}\times\nabla_\mathbf{e} \Phi \Big),\\ \label{eq:motion_tp_e} \frac{d\mathbf{e}}{dt}=\frac{1}{\sqrt{\mathcal{G}M\in a}}\Big( {\mathbf{j}}\times\nabla_\mathbf{e} \Phi +\mathbf{e}\times\nabla_{\mathbf{j}} \Phi \Big),\ \end{align} \end{subequations} where $M\in=M_{\rm s}+M_{\rm p}$ and $\Phi=\Phi\in+\Phi_\mathrm{out}$ is the combined tidal potential from the inner star-planet pair and the outer stellar companion including octupole-order terms (see Eqs~[\ref{eq:potential_av}]-[\ref{eq:potential_av_gr}]). The star-planet angular momentum and eccentricity vectors $\propto{\mathbf{j}}\in$ and $\mathbf{e}\in$ are evolved following an analogous set of equation of motion, except that the tidal potential is entirely due to the binary companion, with the contribution of the planetesimal being neglected. Finally, the angular momentum and eccentricity vectors of the star-binary system $\propto{\mathbf{j}}_\mathrm{out}$ and $\mathbf{e}_\mathrm{out}$ -- although in principle subject to evolution under the tidal potential from the inner star-planet pair -- are held constant and only the semimajor axis of this outermost orbit $a_\mathrm{out}$ is evolved consistently with the process of mass loss of the central star. \begin{figure*} \centering \includegraphics[width=16cm]{fig1.eps} \caption{The orbital architectures at the different phases of the stellar evolution considered in our example. {\it Panel a}: a $2M_{\odot}$ MS star orbited by a giant planet at $a_{\rm p}=2$ AU, a coplanar planetesimal disk at $a=3-12$ AU, and inclined binary companion ($i_{\rm b}=80^\circ$) at $a_{\rm b}=600$ AU. The disk remains stable against the KL mechanism because of the planet perturbations. {\it Panel b}: the planet is engulfed by the host star during the giant phase (GB or AGB phases). {\it Panel c}: the orbits of the planetesimal disk and the binary expand due to mass loss. {\it Panel d}: the disk is subject to the KL mechanism and the planetesimals are tidally disrupted outside-in. } \label{fig:phases} \end{figure*} \subsubsection{Initial set-up: main-sequence configuration} \label{sec:setup} We consider a host star with a Zero Age Main Sequence mass of $M_{\rm s}=2M_\odot$, which is a typical progenitor for the currently-observed WD population in the Milky Way. We assume that this star is orbited by a Jupiter-mass planet at $a_{\rm p}=2$ AU in a circular orbit, and a wide stellar binary companion with a mass of $M_{\rm b}=0.5M_\odot$, a semi-major axis of $a_{\rm b}=600 $ AU and an eccentricity of $e_{\rm b}=0.5$ (see panel a in Figure \ref{fig:phases}). The inclination relative to the planetary orbit is $i_{\rm b}=80^\circ$. The planet is subject to the tidal field from the stellar companion and apsidal precession due to General Relativity (GR). At 2~AU, the apsidal precession period due to GR ($\sim10$ Myr) is shorter than that due to the companion ($\sim30$~Myr; Eq. [\ref{eq:tau}]), and thus KL oscillations are suppressed. In practice, other sources of pericenter precession such as additional planets can make the planetary orbit long-term stable against perturbations from the inclined companion. In this configuration, $r_L\approx16$~AU, and thus any bodies interior to this distance will be protected by the planet against perturbations from the binary companion. The fourth body in the system is a massless planetesimal located at $a=10$~AU -- such that $a<r_L$ -- in near coplanarity with the planetary orbit ($i_{\rm eq}\simeq0.46^\circ$ from Eq. [\ref{eq:i_lap}]) \subsubsection{Post-main sequence evolution} The system is evolved at once (including MS and Post-MS stages, see the sequence depicted in Figure \ref{fig:phases}) under one set of equations (Eq.~[\ref{eq:eom}]). After some time $t_{\rm MS}$, mass loss and planetary engulfment are triggered (Eqs.~[\ref{eq:aplanet}] and [\ref{eq:massloss}]), affecting directly the semi-major axes $a_{\rm p}$, $a_{\rm b}$ and $a$ (which are constant in the secular evolution of the MS stage), in addition to the central mass $M_s$ (see Eqs.~(\ref{eq:ml_in})-(\ref{eq:ml_out}) in the Appendix). The timescale for planetary removal and mass loss is expected to be of a few Myr, and to take place primarily during the AGB phase of stellar evolution (e.g., \citealt{hurley00}). We set $\tau_{\rm ml}=4\tau_a$ to ensure that the mass loss and the orbit's shrinkage happen simultaneously, but the former takes place mostly at the end of the planet's orbital decay. During this phase, planets inside $\sim3-5$ AU are expected to be engulfed by the expanded envelope of the host star \citep{MV12,villa14}. By using the secular equations of motion it is implicitly assumed that the mass loss and engulfment timescales are much longer than all the orbital timescales in the system. For outer companion separations of up to $\sim2000$~AU, this is a reasonable approximation. In such case, all the orbital elements in the system, except the semi-major axes, will remain unchanged \citep[e.g.,][]{hadjidemetriou63,veras11}. In our example, we use that the final mass -- the mass of the WD -- is $M_{\rm WD}\simeq0.64M_\odot$ \citep{hurley00}. The semi-major axes of the small bodies expands in factor of $M_{\rm s}(t=0)/M_{\rm WD}\simeq3.1$, while the binary does so in a factor of $[M_{\rm s}(t=0)+M_{\rm b}]/(M_{\rm WD}+M_{\rm b})\simeq2.2$. Initially, $\epsilon\in/\epsilon_\mathrm{out}=(r_{L,0}/a)^5\approx11$. As the planet is engulfed, $\epsilon\in/\epsilon_\mathrm{out}\rightarrow0$; in practice, the potential from the planet is ignored after $\epsilon\in/\epsilon_\mathrm{out}$ reaches $10^{-5}$, or when $r_L=0.1 a$. After the planet influence becomes negligible, the planetesimals pericenter precession will be given by $\dot{\varpi}_\mathrm{out}$ (Eq.~\ref{eq:omega_out}). If the removal of the planet is non-adiabatic (as defined in Section~\ref{sec:adiabaticity}), the planetesimal may be subject to KL oscillations, which take place with a characteristic period of $\tau_{\rm KL}\sim \dot{\varpi}_\mathrm{out}^{-1}$. In classic Kozai oscillations (quadrupole-order perturbations $\propto a^2/a_{\rm b}^3$) the planetesimals eccentricity may reach $e_{\rm max}=(1-5/3\cos^2 i_{\rm b})^{1/2}\simeq0.97$, implying a minimum pericenter distance of $a(1-e_{\rm max})\simeq1$ AU. However, for $e_{\rm b}\neq0$, the planetesimal is subject to strong forcing due to higher-order (mostly octupole-order $\propto e_{\rm b}a^3/a_{\rm b}^4$) perturbations, which happen in timescales longer than the KL timescale \citep[e.g.][]{naoz16}. These {\it very long timescale} effects can drive the eccentricity up to much higher values ($1-e\lesssim0.001$). If high enough eccentricities are reached such that $a(1-e)<2R_\odot$, then the planetesimal is assumed to be tidally disrupted. \begin{figure*} \centering \includegraphics[width=17cm]{fig2.eps} \caption{Orbital evolution of the planetary system initially composed of an inner Jupiter-mass planet at 2 AU and a massless planetesimal at 10 AU orbiting a 2$M_\odot$ star (final WD mass of 0.64 $M_\odot$). We show the results after the main sequence ($t>t_{\rm MS}$) and the vertical dashed lines indicate the zero WD's cooling age. The planetesimal and the planets start with zero mutual inclination and circular orbits, while the binary companion with $M_b=0.5M_\odot$ and $a_b=600$ AU has an inclination of $80^\circ$ relative to the planetary system. The eccentricity of the binary is $e_b=0.5$, and the initial ascending nodes and arguments of pericenter are 0 for all the orbits. The upper panels (a, b, and c) show the semi-major axes (solid lines) and pericenter distances (dotted lines). The lower panels (d, e, and f) show the inclination of the planetesimal relative to the planets (solid red line) and relative to the binary (solid black line). The different columns show the different planet's semi-major axes decaying timescale $\tau_a$ in Equation (\ref{eq:aplanet}): $\tau_a=0.5$ Myr (panels a and d), $\tau_a=10$ Myr (panels b and e), and $\tau_a=100$ Myr (panels c and f). The planet is assumed to be engulfed at 0.2 AU and the planetesimal is assumed to be tidally disrupted when it reaches $a(1-e)<2R_\odot$, which only happens for $\tau_a=0.5$ Myr (dotted black line in panel a). We set the mass loss timescale $\tau_{\rm ml}=M_s/|\dot{M}_s|$ equal to $4\tau_a$ to ensure that mass is lost most efficiently after the planet is engulfed. } \label{fig:tau_a} \end{figure*} \subsubsection{Fast and slow planet engulfment} To directly test the qualitative predictions of Section~\ref{sec:trigger}, we integrate the 4-body equations of motion while varying the engulfment time $\tau_{\rm a}$. As described in Section~\ref{sec:adiabaticity}, the ratio $\tau_{\rm a}/\tau_{\rm KL}$ will determine whether or not the planetesimal will be susceptible to the influence of the eccentric KL mechanism. \paragraph{Fast engulfment ($\tau_{\rm a}=0.5$~Myr)} For $\tau_{\rm a}\ll \tau_{\rm KL}\sim 8$~Myr (left panels, Fig.~\ref{fig:tau_a}), we expect adiabaticity to be broken and the planetesimal to be impulsively removed from the Laplace surface. In this case, orbits expand according to the mass loss prescription (top panel), but this takes place before the planetesimal inclination has been altered significantly (bottom panel). The planetesimal-to-binary inclination is this the same as in the initial condition ($\sim80^\circ$), which is capable of triggering KL oscillations. After planet removal, common KL oscillations ensue (with period of $\simeq10$~Myr), reaching a maximum planetesimal eccentricity of 0.9. In the longer run, the slower octupole-level oscillations cause dramatic eccentricity growth, reaching $e\simeq0.999$ and beyond, sufficient to guarantee tidal disruption. \paragraph{Trans-adiabatic engulfment ($\tau_{\rm a}=10$~Myr)} When $\tau_{\rm a}\sim\tau_{\rm KL}$ (Fig.~\ref{fig:tau_a}, middle panels), the orbital behavior of the planet is markedly different from the ``fast engulfment'' case from above. In this case, the planetesimal ``tries to follow'' the Laplace surface solution $i_{\rm eq}(r_L)$ (Eq.~[\ref{eq:i_lap}]) as $r_L$ shrinks with time. Initially, the planetesimal can follow closely; the initial oscillation amplitude ($\lesssim2^\circ$) is seeded by the imperfect alignment of $i_0$ with $i_{\rm eq}$ at $t=0$. If adiabaticity were to be preserved, this initial amplitude should grow as the oscillation frequency decreases from $\sim\dot{\varpi}\in|_{t{=}0}$ to $\sim\dot{\varpi}_\mathrm{out} \footnote{Linear oscillations around the stable equilibrium should behave like a harmonic oscillator of time-varying frequency $\omega_0(t)$ (Eq.~\ref{eq:osc_freq}), for which the action is $E/\omega_0=\omega_0A^2/2$, where $A$ is the oscillation amplitude. For very slowly varying $\omega_0$, the action $E/\omega_0$ is an adiabatic invariant.}. Nevertheless, planet engulfment is still too fast, as the planetesimal leaves the Laplace surface before the planet is fully engulfed. In this case, the planetesimal is decoupled from the planet at $t{\simeq}20$~Myr (the planet is finally eliminated at $t{\simeq}30$~Myr), with an inclination relative to the binary of $\gtrsim40^\circ$. This small inclination ($>39.23^\circ$) is still large enough to trigger some mild KL oscillations, but evidently it is far from the inclinations needed to obtain tidal disruptions as $e_{\rm max}$ is only $\simeq0.45$. \paragraph{Slow engulfment ($\tau_{\rm a}=100$~Myr)} For slow engulfment ($\tau_{\rm a}\gg\tau_{\rm KL}$, right panels, Fig.~\ref{fig:tau_a}), the planetesimal nearly follows the Laplace surface to the end of the integration, reaching a final inclination of only $\sim13^\circ$ and a finite constant eccentricity of $\simeq0.3$. The finite eccentricity is reached at $\simeq60$~Myr when $a\sim r_L$, and is due to the bifurcation experienced in the Laplace equilibrium solution, which makes circular orbits unstable \citep{tremaine09,tamayo13}. For even slower engulfment ($\tau_{\rm a}=1$~Gyr), the final state is much more steady, i.e., inclination oscillations are small, and the planetesimal never leaves the vicinity of the Laplace surface, even if it gains a finite eccentricity as it crosses the bifurcation. We note that this bifurcation exists only when the planet-binary inclination is $\gtrsim69^\circ$. In addition, the finite octupole potential from the binary introduces modifications to the classical Laplace equilibrium analysis \citep[e.g.][]{mun15b}. In an analogous example with $e_{\rm b}=0$ and a lower inclination we obtain an end-state where the planetesimal is in perfect alignment with the binary and retains zero eccentricity throughout the integration. \vspace{5mm} Of these three scenarios, only the first one (fast engulfment) is expected to resemble the AGB phase, which lasts less than $\sim10$ Myr. This introduces an important difference with the work of \citet{mun15b}, which finds that the rate of reduction of $r_L$ is always slower than the oscillations around the equilibrium solution. In the present case, adiabaticity is an unlikely outcome, which implies that planetesimals can ``instantaneously" see themselves in a Kozai-unstable configurations even though throughout the entire MS lifetime of the host star they were protected against such instabilities. Finally, we caution that our numerical calculations are based on the double orbit averaging approximation and might not represent the dynamics properly, leading to spurious extreme eccentricities required to disrupt the planetesimal \citep{LKD16}. To this extent, we have repeated the three-body integrations after planet engulfment using the direct high-order N-body integrator IAS15 \citep{RS15}, which is part of the REBOUND package \citep{RL12}. We find that the evolution looks very similar, but the planetesimal is disrupted slightly later after two extra oscillation cycles compared with the secular code. \section{Evolution of the planetesimal disk} We integrate the orbital evolution of a disk of collision-less planetesimals orbiting a white dwarf and perturbed by a distant companion based on the orbital configuration of our fiducial system described in \S\ref{sec:trigger}. This phase corresponds to panel d in Figure \ref{fig:phases}. \subsection{Initial conditions} After the AGB phase we are left with a a WD of mass $M_{\rm WD}\simeq0.64M_\odot$ orbited by a planetesimal disk and stellar binary companion with mass of $M_{\rm b}=0.5M_\odot$, semi-major axis $a_{\rm b}=1300$ AU, and inclination of $i_{\rm b}=80^\circ$ (relative to the planetary orbit before engulfment). We shall assume that the planetesimal disk has the following power-law profile for the surface density \begin{eqnarray} \Sigma(a)=\frac{M_{\rm disk}} {2\pi\left(a_{\rm out}-a_{\rm in}\right)}\cdot\frac{1}{a}, \end{eqnarray} where $M_{\rm disk}$ is the total mass of the disk, while $a_{\rm in}$ and $a_{\rm out}$ are its inner and outer boundaries. This power-law profile has a uniform mass distribution as a function of semi-major axis, which is a convenient choice to easily read out our results for the accretion rates. The inner edge of the disk $a_{\rm in}$ is set by the dynamical stability due to the planetary perturbations during the Main Sequence. Since the planet is initially at $a_{\rm p}=2$ AU, the long-term stability is guaranteed for planetesimals at $\gtrsim3$ AU\footnote{The test particles are Hill stable for $a/a_p\gtrsim1.4$ (e.g., \citealt{gladman93}).}. Since the semi-major axis of the planetesimals expands by a factor of $\simeq3.1$ relative to the initial value during the MS, we set $a_{\rm in}=10$ AU. The outer edge is set by the location of the Laplace radius, beyond which the secular perturbations due to the binary can excite the eccentricities and inclinations of the planetesimals. More specifically, as shown by \citet{tremaine09} for a binary's inclination of $i_{\rm b}=80^\circ$ a test particle is stable against eccentricity perturbations when $a<0.9r_{\rm L}\simeq14.6$ AU. Thus, we conservatively consider the planetesimals with $a<12$ AU in our calculations so the disk remains nearly aligned with the orbit of the planet and with low-eccentricity excitation during the planet engulfment. For reference, the Laplace surface of a planetesimal with $a=12$ AU has an inclination relative to the planetary orbit of $i_{\rm eq}\sim1.1^\circ$ (Eqs. [\ref{eq:i_lap}] and [\ref{eq:rl}]). Again, since the semi-major axis of the planetesimals expands by a factor of $\simeq3.1$ relative to the initial value during the MS, we set $a_{\rm out}=35$ AU. The planetesimal orbits are initialized with zero inclination, random longitudes of the ascending node and arguments of pericenter, and eccentricities from a Rayleigh distribution with parameter 0.01. We leave the mass of the disk as a free parameter, but keep in mind that the collisional evolution of the disk limits the maximum mass for a given age (e.g., \citealt{wyatt07,HT10}). We discuss the constraints on the disk mass in \S\ref{sec:discussion}. \begin{figure*} \centering \includegraphics[width=18cm]{fig3.eps} \caption{Number of particles that are tidally disrupted by the WD as a function of their initial semi-major axis during the WD phase (i.e., after their orbits expanded by a factor of $\simeq3.1$ relative to their MS values) and the time at which they cross the disruption distance (initial time is the zero cooling age time). {\it Panel a:} two-dimensional histogram. The binning in the horizontal axis is evenly-spaced in the log of the disruption time. {\it Panel b:} fraction of tidally disrupted particles as a function of semi-major axis. {\it Panel c:} number of tidally disrupted particles normalized by the tallest bin as a function of the disruption time. } \label{fig:pop_dis} \end{figure*} \subsection{Results} We evolve the disk up to 10 Gyr using $50,000$ particles and record the time at which a particle is tidally disrupted, which we define to takes place once $a(1-e)<2R_\odot$. The results are not sensitive to the choice of the disruption distance, and typical values within $\sim1-3R_\odot$ \citep{veras14b} give similar results. In Figure \ref{fig:pop_dis} we show the number of disrupted bodies as function of the initial semi-major axes and the disruption times. From panel a we observe that the bodies at larger semi-major axes tend to be disrupted first. This is expected because the KL timescale decreases with semi-major axes as $a^{-3/2}$ (see Equation \ref{eq:tau}). However, we point out from panel a that for a given semi-major axis the disruptions happen at many different times, not just $\tau_{\mbox{\tiny{KL}}}$. This is because the binary is eccentric ($e_{\rm b}=0.5$) and we expect the dynamics of the disk to be affected by the octupole-level eccentricity modulations of the KL mechanism\footnote Note that the maximum eccentricity with $e_{\rm b}=0$ is $e_{\rm max}=(1-5/3\cos^2 i_{\rm b})^{1/2}=0.975$, which it does not lead to disruptions since $a(1-e_{\rm max})>0.3$ AU. (see \citealt{naoz16} for a recent review), which lead to extreme eccentricities on timescales longer than $\tau_{\rm KL}$. The strength of these octupole-level perturbations (relative to the quadrupole-level) is proportional to $\epsilon_{\rm oct}=e_{\rm b}a/a_{\rm b}$, which implies that the planetesimals with smaller semi-major axes are expected to be less affected by the octupole modulations. Consistently, we observe from panels a and b that the number of disruptions decreases from $\gtrsim90\%$ at $a\lesssim20$ AU to nearly zero for $a\lesssim14$ AU. Overall, $70\%$ of the planetesimals are tidally disrupted (see panel b of Figure \ref{fig:pop_dis}) . Similarly, the timescale is of these octupole-level oscillations is $\sim \epsilon_{\rm oct}^{-1/2}\tau_{\mbox{\tiny{KL}}}\propto a^{-2}$ \citep{anto15}. Therefore, the small bodies are disrupted after several secular timescales $\tau_{\mbox{\tiny{KL}}}$ giving rise to a wide range disruptions times for a fixed semi-major axis. From panel c we observe that the disruption times have a wide distribution in the range of $\sim0.05-10$ Gyr. The distribution peaks at $\sim0.1$ Gyr because the first modulation of the octupole is driving most of the planetesimals to extremely large eccentricities around this time. Beyond this peak, the distribution flattens in log of the time, $dN/d\log(t)\sim \mbox{cst.}$, meaning that it decays as $\propto 1/t$ at late times. This slow decay is due to extra octupole-level modulations of the planetesimals that survived the first one high eccentricity phase, which happens preferentially for smaller values of $a$, as expected. In summary, most of the planetesimal disk ($\sim70\%$ of the mass) is tidally disrupted in a wide range of timescales due to both the long-term octupole-level perturbations and the large radial extent of the disk. The rate of disruption events decays slowly as $\propto1/t$ at late times. \section{Discussion} \label{sec:discussion} We have proposed a new mechanism to explain the observed pollution in WDs through the tidal disruption of planetesimals orbiting these stars. We propose that a planetary system (one or several planets inside $\sim2-5$ AU) shield the planetesimals orbits against the KL mechanism due a distant stellar companion. Once the planetary system is engulfed during the late stages of stellar evolution (e.g., the AGB phase), the orbits of the planetesimals become (secularly) unstable, leading to extreme eccentricities ($e\gtrsim0.999$) and, therefore, to tidal disruptions. This mechanism has the following properties: \begin{enumerate} \item {\it pollution takes place for WDs of all cooling ages}. This property, required by the observational evidence, is inherent to the nature of the eccentric KL mechanism, which leads to the excitation into extreme eccentricities over very long timescales. \item {\it It provides a steady flux of tidally disrupted rocky bodies}. This property is due the large radial extent of the planetesimal disk, widely spanning different disruption timescales. Each part of the disk has a disruption timescale $\propto a^{-2}$ (the eccentric KL timescale). Furthermore, disruptions can happen after multiple eccentric KL cycles. This property is required in order to have a non-negligible probability of observing the metals in the WDs atmosphere, since both the accretion and settling timescales are much shorter than the cooling ages of the observed systems. We quantify this probability in \S\ref{sec:fpoll} (see Equation [\ref{eq:fpoll}]). % \item {\it The planetesimal disk can have low surface densities and be long-lived}. Since the planetesimal disk can have a large radial extent ($\Delta a/a\sim3.5$ in our example in Figure \ref{fig:pop_dis}) and most of the disk can disrupted ($\sim0.7$ of the mass), our mechanism can explain the observed accretion rates even for low surface densities (see constraints in \S \ref{sec:acc}). These low surface density disks can live for longer timescales avoiding grinding down to dust, which would be easily blown out during the RG and AGB phases by radiation pressure (e.g., \citealt{BW10}). \end{enumerate} Regarding point 1 above, we note that a similar idea, relying on long-timescale secular instabilities to explain the pollution of the oldest WDs have been presented by \citet{BV15} and \citet{HP16}. These models, however, have not yet shown to provide the required steady flux of disrupting rocky bodies (point 2 above), nor consider the shielding effect against external perturbations that planetary systems would have during the MS phase of the host star. In what follows, we discuss further constraints on our model from both observations and theoretical expectations. \subsection{Constraints on the disk mass $M_{\rm disk}$} \label{sec:mdisk} Disks with high enough masses can quench our mechanism by (at least) the following two reasons: (1) pericenter precession rate due to self-gravity limits the KL mechanism; (2) dynamically hot disks become highly collisional, likely shattering the small bodies. The pericenter precession timescale of a planetesimal at 10 AU (the inner edge of our disk) due to the self-gravity of the disk has been estimated by \citet{batygin11} to be (see also \citealt{R13,TTMR15}): \begin{eqnarray} \tau_{\rm s-g}\sim 13 \mbox{ Myr} \left(\frac{M_\oplus} {M_{\rm disk}}\right). \label{eq:tau_sg} \end{eqnarray} This timescale should be compared to the timescale of pericenter precession due to the external companion, $\tau_{\mbox{\tiny{KL}}}$ (Equation~\ref{eq:tau}). For the example with $a_{\rm b}=1300$ AU and $a\sim10-35$ AU, $\tau_{\rm KL}\sim4-24$ Myr, implying that, for the KL mechanism to operate, we require $M_{\rm disk}\lesssim0.5M_\oplus$. However, when $\tau_{\rm s-g}\gtrsim\tau_{\mbox{\tiny{KL}}}$, the maximum eccentricity reached by the KL mechanism is still reduced \citep[e.g.][]{liu15} and thus the planetesimals do not reach star-grazing orbits. We checked the effect of disk self-gravity by adding a pericenter precession term to equations of motion with a rate $\sim1/\tau_{\rm s-g}$, finding that a range of maximum disk masses in $\sim0.01-0.1M_\oplus$ still allows for numerous tidal disruptions. After KL oscillations stir up the planetesimal disk, the relative velocities between the planetesimals are high ($e\sim i\sim1$), and the collisions between planetesimals can be highly disruptive. \citet{HT10} have estimated that a dynamically hot planetesimal disk at $\sim10$ AU can survive the disruptive collisions or the gravitational scattering events for Gyr timescales if $M_{\rm disk}\lesssim M_\oplus$. However, such disks should be comprised of $\lesssim10^4$ small bodies, each with at least a Ceres mass of $\sim10^{24}\rm{g}\sim10^{-4}M_\oplus$. These constraints relax dramatically for disks at larger semi-major axes. For example, at 100 AU, the disk can be as massive 100 $M_\oplus$ and allow for body numbers of up to $10^{10}$ and individual masses of $10^{19}$ g. In our model, a disk with radial extent of $\sim10-40$ AU and a mass of $M_{\rm disk}\lesssim M_\oplus$ should survive in the long term. In conclusion, our mechanism is expected to operate in disks with masses of $\lesssim0.1M_\oplus$. Higher masses are expected to either quench the KL oscillations or disrupt the disk via collisions between the planetesimals. \begin{figure*} \centering \includegraphics[width=0.99\textwidth]{fig4.eps} \caption{Accretion rates derived from the number of small bodies disrupted in our model in Figure \ref{fig:pop_dis} for different disk masses $M_{\rm disk}$ and cooling ages. We assume that all the particles have the same mass and that $f_{\rm acc}=1$ in Equation (\ref{eq:Macc}), i.e., all the mass in tidally disrupted particles is accreted by the WD. {\it Panel a:} we set $M_{\rm disk}=0.1M_\oplus$ (solid black line), which is the largest mass allowed by the precession due to the self-gravity of the disk, and also show lower values of $M_{\rm disk}=10^{-3}M_\oplus$ (dark gray dashed line) and $M_{\rm disk}=10^{-5}M_\oplus$ (light grey dotted line). A fit to the observations taken from \citet{wyatt14} is shown in red dashed line. {\it Panel b:} we divide the accretion rates in panel a in two different populations: cooling ages $<500$ Myr with $M_{\rm disk}=2.4\times10^{-4}M_\oplus$ (solid black line) and cooling ages $>500$ Myr $M_{\rm disk}=0.01M_\oplus$ (gray dashed line). The disk masses in each population are set to reproduce the observed peak in the observations (red dashed line).} \label{fig:hist_acc} \end{figure*} \subsection{Accretion rates} \label{sec:acc} Having constrained the initial mass of the disk to $M_{\rm disk}\lesssim0.1M_\oplus$, we can estimate the maximum accretion rate of planetesimals predicted by our model as a function of age. For simplicity, we shall assume that the size distribution of the planetesimals does not change with semi-major axes. The total mass accreted by the WD can estimated as: \begin{eqnarray} M_{\rm acc}=f_{\rm td}\cdot f_{\rm acc}\cdot M_{\rm disk} \label{eq:Macc} \end{eqnarray} where $f_{\rm td}$ is the fraction of tidally disrupted objects and $f_{\rm acc}$ is the mass fraction of disrupted particles that is accreted and reaches the WD's surface. We estimate the former fraction directly from our simulation (see panel b of Figure \ref{fig:pop_dis}) to be $f_{\rm td}\simeq 0.7$. The latter fraction is largely unconstrained. The accretion rate can be calculated as \begin{eqnarray} \frac{dM_{\rm acc}}{dt}=\frac{df_{\rm td}}{dt} \cdot f_{\rm acc}\cdot M_{\rm disk}, \end{eqnarray} where $df_{\rm td}/dt$ is shown in panel c of Figure \ref{fig:pop_dis} (normalized by the maximum rate). From this figure we point out that the frequency of disrupted bodies and, therefore the accretion rate, is highest at $\sim50-300$ Myr and it flattens (in log space, i.e., decays as $\propto1/t$) at $\gtrsim500$ Myr. In Figure \ref{fig:hist_acc}, we show the accretion rate derived from our calculations by computing the time between tidal disruptions and assuming that all the bodies have the same mass. We set $f_{\rm acc}=1$ and quote values of the disk mass $M_{\rm disk}$. In panel a, we show the accretion rate for different disk masses in a range of $M_{\rm disk}=10^{-5}-10^{-1}M_\oplus$, where the upper limit is roughly the maximum mass imposed by precession due to disk self-gravity (see \S\ref{sec:mdisk}). The distribution of accretion for $M_{\rm disk}=0.1M_\oplus$ (solid black line) peaks at $\sim3\times10^{10}~\mbox{g}~\mbox{s}^{-1}$, while there is a secondary bump at $\sim10^{9}~\mbox{g}~\mbox{s}^{-1}$. This bi-modality is due to the accretion rate change at early and late times in the evolution of the WD (see panel c of Figure \ref{fig:pop_dis}). By splitting the sample into early stages (cooling ages $<500$ Myr) and late stages (cooling ages $>500$ Myr), we obtain two symmetric (log-normal) distributions. For reference, we show a fit to the observations using a log-normal distribution with $\mu=8$ and $\sigma=1.3$ taken from \citet{wyatt14}. In panel b, we show the accretion rates at early and late stages for disk masses of $M_{\rm disk}=2.5\times10^{-4}M_\oplus$ and $M_{\rm disk}=0.01M_\oplus$, both coinciding with the observed peak at $10^{8}~\mbox{g}~\mbox{s}^{-1}$. We also note from the panel b that our model predicts a smaller dispersion of the accretion rates relative to the observations. In reality, we expect that for an ensemble of systems our predicted accretion rates should broaden significantly by considering a distribution of disk masses and particle masses, as well different orbital separations of the binary companion. This calculation is beyond the scope of this paper and it will be worth studying in a future work. In summary, we find that disk masses in the range $M_{\rm disk}=10^{-4}-10^{-2}M_\oplus$ produce accretion rates consistent with observations. These disk masses are small enough that this mechanism is not expected to be limited by either by disk self-gravity or the collisional destruction of small bodies (see \S\ref{sec:mdisk}). The accretion rate is expected to peak at $\sim10^8$ yr and decays as $\propto1/t$ at late times. \subsection{Estimate of the pollution rate} \label{sec:fpoll} Here we estimate the rate at which this mechanism might contribute to the observed pollution of WDs and assess whether it can account the high observed rates of $\sim25-50\%$. The fraction of WDs for which our mechanism can contribute to the observed levels of pollution can be estimated as: \begin{eqnarray} f_{\rm poll}\equiv f_T\cdot f_{\rm KL}\cdot f_b\cdot f_p, \label{eq:fpoll} \end{eqnarray} where $f_T$ is the fractional time in the WD's cooling age during which the rocky material is being supplied, $f_{\rm KL}$ is the fraction of systems that leading to planetesimal disruptions due to the eccentric KL mechanism, $f_b$ is the fraction of stars with wide ($a_b>100$ AU) binary companions, and $f_p$ is the fraction of planetary systems with planets within $\sim3$ AU and planetesimal disks during the main sequence of A and F stars. We can optimistically estimate that $f_T\sim1$ since the proposed system provides a steady delivery of asteroids into disrupting orbits (see panel c in Figure \ref{fig:pop_dis}). In reality, we have to account for the finite number of planetesimals in the system which would lead to discontinuous events of tidal disruptions. However, this effect is compensated by the finite timescales involved in the circularization of the disrupted material and subsequent accretion onto the WD's atmosphere (see \citealt{veras16}). We emphasize that one of the virtues of this new mechanism is that $f_T\sim1$, which is not the case in many others models found in the literature. The fraction of systems for which the planetesimals can be tidally disrupted due the eccentric KL mechanism $f_{\rm KL}$ can be estimated from previous studies in the context of planet disruptions in main sequence stars (e.g., \citealt{naoz12,petro15,ASL15,MLL16}). These studies find that for a population of wide binaries ($a_{\rm b}\gtrsim100$ AU) with isotropic inclinations and a thermal eccentricity distribution\footnote{\citet{TK15} find that that the eccentricity distribution for solar-type wide binaries ($a_{\rm b}>50$ AU) is $f(e_{\rm b})=1.2e_{\rm b}+0.4$, while $f(e_{\rm b})=2e_{\rm b}$ \citet{K07} for very wide binaries ($a_{\rm b}\gtrsim1000$ AU)} roughly up to $\sim30\%$ of them can lead to planetary tidal disruptions (assuming no tidal circularization). This estimate is consistent with the observation that $5/17\sim30\%$ of the WDs with wide binary companions are metal-polluted \citep{Z14}. Thus, $f_{\rm KL}\simeq0.3$. The fraction of A and F stars (progenitors of most WDs) with binary companions has been measured to be $\sim70\%$ \citep{K07,peter12,rosa12}, while the semi-major axis distribution follows a log-normal distribution peaked at $\sim300$ AU \citep{rosa12}, which is significantly wider than the peak of Solar-type stars at $\sim40$ AU \citep{ragha10}. Using the semi-major axis distribution from \citet{rosa12}, we find that $\sim70\%$ of the binaries have $a_{\rm b}>100$ AU, implying that the fraction of A stars with wide binary companions is $\sim0.7\times0.7\sim0.5$. Thus, we estimate that the fraction of A stars with wide binary companions is $f_b\sim0.5$. The fraction of A and F stars with planetary systems and outer small rocky bodies $f_p$ can be estimated --conservatively-- from the occurrence of gas giant planets (assuming they all have small rocky bodies). The RV planet searches find that $\gtrsim10\%$ of Solar-type stars have gas giant planets (e.g., \citealt{WF15}), while the occurrence seems increase linearly with the host star mass, reaching $\sim30\%\pm15\%$ for A-stars with 2$M_\odot$ stars \citep{JJ10}. Lower-mass planets around Solar-type stars are more abundant than giants, and their occurrence, which is not well constrained at AU separations and for A and F stars, is roughly 0.5. The extrapolation of these results suggest that the occurrence of planets in A and F stars can be as high unity. Thus, we use $f_p\sim0.3-1$. Putting these numbers together, we estimate that the fraction of polluted WDs that can be explained by our mechanism is $f_{\rm poll}= f_T\cdot f_{\rm KL}\cdot f_b\cdot f_p\sim 1\times0.3\times0.5\times(0.3-1)\sim 0.05-0.15$. This result implies that our mechanism can explain a significant portion, but not all, of the observed polluted WDs, which amount to fraction of $f_{\rm poll,obs}\gtrsim0.25$. In summary, our mechanism can only explain a fraction, although still significant (up to $50\%$), of the observed polluted WDs. Complete searches for companions, for which GAIA will play a crucial role, might shed light on the significance of this new model to explain the pollution of WDs. \subsection{Outer companions in polluted white dwarfs} The engulfment-aided KL mechanism to explain WD pollution requires the presence of a massive body in a wide orbit (either a stellar companion or a planet). We note that both the distance and the mass of the distant perturber enter into the calculation mainly through the timescale of the gravitational perturbations in the form of the dimensionless parameter $\epsilon_\mathrm{out}\propto M_\mathrm{out}/a_\mathrm{out}^3$ (see Equations [\ref{eq:varpi_out}] and [\ref{eq:eps_out}]). Thus, the evolution of the system with a Solar-mass companion will be very similar to that with a Jupiter-mass companion but 10 times closer, providing an alternative version of the mechanism that may increment the explained pollution rate by accommodating different outer companions. We note that the presence of a low-mass main sequence star like in the example of Figure \ref{fig:tau_a} would most likely be detected had the WD been observed in the first place. However, the fraction of polluted WDs with wide main sequence star companions is small (e.g., \citealt{Z14}). This implies that either the model presented here can only explain a subset of the polluted WDs, or that the current sample of outer companions of polluted WDs is largely incomplete. We briefly discuss the latter alternative, namely, other types of companions might still escape detection and might contribute to the incompleteness of the current sample of companions. First, outer planetary companions and brown dwarfs at $\sim10-100$ AU distances can remain undetected (e.g., \citealt{farihi05,debes11,day13}). Also, these outer companions can drive eccentricities to nearly unity values not only by the KL mechanism (for which large inclinations respect to the planetesimal disk are required), but also by either a nearly coplanar and eccentric body \citep{li14,petro15b} or secular chaotic diffusion due to two or more eccentric and/or inclined bodies \citep{LW11,WL11,batygin15}. Second, stellar-mass companions such as other fainter WDs or neutron stars and black holes can also escape detection. One intriguing observational puzzle is the mismatch between the measured binary fraction of WDs \citep[$\sim30\%$;][]{farihi05,holberg16} and that of their progenitors ($\sim 70-100\%$, e.g., \citealt{K07}). As noted by \citet{ferr12}, a relatively flat mass ratio distribution for the progenitors of WDs gives a better fit the mass distribution of the detected WDs companions, typically M-dwarfs. However, this same distribution also implies that $\sim30\%$ of the WDs should be in double WD systems, most of them ``hiding" as singles. Similarly, based on the completeness of the sample of A-stars in \citet{rosa12}, \citet{KK16} argued that $\sim10\%$ of these WD progenitors are likely to host undetected companions that will become WDs within the age of our galaxy. In this picture, current catalogs are typically missing the fainter WDs in the WD-WD system. \subsection{Effects ignored and simplifications} \label{sec:extra} We discuss some of the relevant effects ignored in this work that might change the dynamics of the system. \paragraph{Extra planets in the system} For simplicity, we have considered a planetary systems with only one planet. We expect our results not to be significantly altered if extra planets within $\sim1-5$ AU are present in the system (those would also be engulfed). In particular, the extra planets enhance the precession rate of the small bodies in the disk and shield the bodies from outer perturbations at even larger distances (i.e., the Laplace radius in Equation [\ref{eq:rl}] increases\footnote{By adding $i=1,..,N$ bodies with masses $m_i$ and semi-major axes $a_i$ we just replace $m_{\rm p}a_{\rm p}^2$ by $m_{\rm p}a_{\rm p}^2+\sum_{i=1}^Nm_ia_i^2$ in the Laplace radius in Equation (\ref{eq:rl}).}). If one or more distant enough planets do survive the AGB phase, these can quench the secular instabilities and our mechanism would not operate (at least in its cleanest version described here). However, the surviving planets themselves could be susceptible to secular perturbations due to the distant perturber. This effect can, in principle, destabilize a planetary system and/or excite the eccentricities of the planets, and such processes in turn can lead to the tidal disruption of small bodies by the WD. \paragraph{Mass loss and galactic tides on the very wide binaries} For wide enough binaries ($\sim10^4$ AU), mass loss can happen on timescales that are not too short compared to the orbital timescales and the response of the orbital elements would be different \citep{veras11,BV15} from what is described in Section~\ref{sec:trigger}. In general, this implies that either the companion becomes more eccentric (or even unbound), which can enhance the eccentric KL mechanism (e.g., \citealt{naoz16}). In addition, these wide binaries will be affected by galactic tides and have their angular momentum altered, which again has the potential of enhancing the secular interactions. \paragraph{Stellar evolution} We have modeled the planetary engulfment by shrinking the orbit in a prescribed way (see Equation \ref{eq:aplanet}) to see how the orbital elements of the planetesimal respond to a gradual planet removal. In reality, the process is much more complicated and the radius of the star during the AGB can undergo pulsations, while tides in the star lead to planet inspiral. These effects have been modeled in detail by \citet{MV12} and \citet{villa14}. Their results indicate that the engulfment generally happens on timescales of $\sim0.1$ Myr. This timescale is shorter than the shortest secular timescale\footnote{The shortest secular timescale can be obtained from evaluating $\tau_{\mbox{\tiny{KL}}}$ in Equation (\ref{eq:tau}) at a semi-major axis and period of a planetesimal located at the Laplace radius in Equation (\ref{eq:rl}).} $\tau_{\mbox{\tiny{KL}}}$ of planetesimals during engulfment for binaries with $a_{\rm b}\gtrsim100$ AU. Therefore, we expect our model to operate for these wide binaries regardless of the details of planet engulfment. In a future work, we will calculate the effects of mass loss and the planetary engulfment using realistic stellar evolution models and incorporate these in a population synthesis model. By doing this, we will be able to better address the significance of our model. \section{Conclusions} We have studied a new mechanism to explain the observed metal pollution in white dwarfs through the tidal disruption of small rocky bodies in a planetesimal disk. We propose that one or several planets can shield a planetesimal disk against the KL mechanism due a distant binary companion. Once the host star evolves off the main sequence to become a WD, these planets can be engulfed (most likely during the AGB phase), thus triggering the KL mechanism, and leading to the tidal disruption of the rocky bodies in the planetesimal disk. We have shown that this mechanism can account for the observed accretion rates for WDs with all cooling ages provided that the disks have masses $\sim10^{-4}-10^{-2}M_\oplus$. Our model allows for planetesimal disks with large radial extents, and as a consequence, it presents the following advantages compared to other models: \begin{itemize} \item it provides a steady supply of material (each part of the disk has a different and long disruption timescale), enhancing the probability of observing the pollution of WD atmospheres; \item it allows for low-density surface disks, which can survive internal disruptive collisions over long timescales. \end{itemize} This mechanism is only triggered after the host star has left the main sequence, providing a self-consistent explanation as to why the KL mechanism does not act on the planetesimal disk for the prior few Gyrs. Our estimates indicate that this model can account for a significant fraction of the polluted WDs. Complete searches for companions of WDs might shed light on the significance of our proposal. \acknowledgements{ We are grateful to Daniel Tamayo, Dimitri Veras, Dong Lai, Norm Murray, Nicholas Stone, Roman Rafikov, and Yanqin Wu for enlightening discussions. C.P. acknowledges support from the Gruber Foundation Fellowship.}
f059ade8ca1b6c7457fa8310cb89acd964da485d
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} In his book ``Proximal Flows''~\cite[Section~\RNum{2}.3, p.\ 19]{glasner1976proximal} Glasner defines the notion of a {\em strongly amenable group}: A group is strongly amenable if each of its proximal actions on a compact space has a fixed point. A continuous action $G \curvearrowright X$ of a topological group on a compact Hausdorff space is proximal if for every $x, y \in X$ there exists a net $\{g_n\}$ of elements of $G$ such that $\lim_n g_n x = \lim_n g_n y$. Glasner shows that virtually nilpotent groups are strongly amenable and that non-amenable groups are not strongly amenable. He also gives examples of amenable --- in fact, solvable --- groups that are not strongly amenable. Glasner and Weiss~\cite{glasner2002minimal} construct proximal minimal actions of the group of permutations of the integers, and Glasner constructs proximal flows of Lie groups~\cite{glasner1983proximal}. To the best of our knowledge there are no other such examples known. Furthermore, there are no other known examples of minimal proximal actions that are not also {\em strongly proximal}. An action $G \curvearrowright X$ is strongly proximal if the orbit closure of every Borel probability measure on $G$ contains a point mass measure. This notion, as well as that of the related Furstenberg boundary~\cites{furstenberg1963poisson, furstenberg1973boundary, furman2003minimal}, have been the object of a much larger research effort, in particular because a group is amenable if and only if all of its strongly proximal actions on compact spaces have fixed points. Richard Thompson's group $F$ has been alternatively ``proved'' to be amenable and non-amenable (see, e.g.,~\cite{cannon2011thompson}), and the question of its amenability is currently unresolved. In this paper we pursue the less ambitious goal of showing that is it not strongly amenable, and do so by directly constructing a proximal action that has no fixed points. This action does admit an invariant measure, and thus does not provide any information about the amenability of $F$. It is a new example of a proximal action which is not strongly proximal. \vspace{0.3in} The authors would like to thank Eli Glasner and Benjamin Weiss for enlightening and encouraging conversations. \section{Proofs} Let $F$ denote Thompson's group $F$. In the representation of $F$ as a group of piecewise linear transformations of $\mathbb{R}$ (see, e.g.,~\cite[Section 2.C]{kaimanovich2016thompson}), it is generated by $a$ and $b$ which are given by \begin{align*} a(x) &= x-1\\ b(x) &= \begin{cases} x& x \leq 0\\ x/2& 0 \leq x \leq 2\\ x-1& 2 \leq x. \end{cases} \end{align*} The set of dyadic rationals $\Gamma =\mathbb{Z}[\frac{1}{2}]$ is the orbit of $0$. The Schreier graph of the action $G \curvearrowright \Gamma$ with respect to the generating set $\{a,b\}$ is shown in Figure~\ref{fig:schreier} (see~\cite[Section 5.A, Figure 6]{kaimanovich2016thompson}). The solid lines denote the $a$ action and the dotted lines denote the $b$ action; self-loops (i.e., points stabilized by a generator) are omitted. This graph consists of a tree-like structure (the blue and white nodes) with infinite chains attached to each node (the red nodes). \begin{figure}[ht] \centering \includegraphics[scale=0.6]{schreier.pdf} \caption{\label{fig:schreier}The action of $F$ on $\Gamma$.} \end{figure} Equipped with the product topology, $\{-1,1\}^\Gamma$ is a compact space on which $F$ acts continuously by shifts: \begin{align} \label{shift-action} [f x](\gamma) = x(f^{-1}\gamma). \end{align} \begin{proposition} \label{prop:pre_proximal} Let $c_{-1}, c_{+1} \in \{-1,1\}^{\Gamma}$ be the constant functions. Then for any $x \in \{-1,1\}^{\Gamma}$ it holds that at least one of $c_{-1},c_{+1}$ is in the orbit closure $\overline{F x}$. \end{proposition} \begin{proof} It is known that the action $F \curvearrowright \Gamma$ is highly-transitive (Lemma 4.2 in ~\cite{cannon1994notes}), i.e. for every finite $V, W \subset \Gamma$ of the same size there exists a $f \in F$ such that $f(V)=W$. Let $x\in \{-1,1\}^{\Gamma}$. There is at least one of -1 and 1, say $\alpha$, for which we have infinitely many $\gamma \in \Gamma$ with $x(\gamma)=\alpha$. Given a finite $W \subset \Gamma$ choose a $V \subset \Gamma$ of the same size and such that $x(\gamma) = \alpha$ for all $\gamma \in V$. Then there is some $f \in F$ with $f(V) = W$, and so $f x$ takes the value $\alpha$ on $W$. Since $W$ is arbitrary we have that $c_\alpha$ is in the orbit closure of $x$. \end{proof} Given $x_1,x_2 \in \{-1,1\}^{\Gamma}$, let $d$ be their pointwise product, given by $d(\gamma) = x_1(\gamma) \cdot x_2(\gamma)$. By Proposition~\ref{prop:pre_proximal} there exists a sequence $\{f_n\}$ of elements in $F$ such that either $\lim_n f_n d = c_{+1}$ or $\lim_n f_n d = c_{-1}$. In the first case $\lim_n f_n x_1 = \lim_n f_n x_2$, while in the second case $\lim_n f_n x_1 = -\lim_n f_n x_2$, and so this action resembles a proximal action. In fact, by identifying each $x \in \{-1,1\}^{\Gamma}$ with $-x$ one attains a proximal action, and indeed we do this below. However, this action has a fixed point --- the constant functions --- and therefore does not suffice to prove our result. We spend the remainder of this paper in deriving a new action from this one. The new action retains proximality but does not have fixed points. Consider the path $(\rfrac{1}{2}, \rfrac{1}{4},\rfrac{1}{8},\ldots,\rfrac{1}{2^n},\ldots)$ in the Schreier graph of $\Gamma$ (Figure~\ref{fig:schreier}); it starts in the top blue node and follows the dotted edges through the blue nodes on the rightmost branch of the tree. The pointed Gromov-Hausdorff limit of this sequence of rooted graphs\footnote{The limit of a sequence of rooted graphs $(G_n,v_n)$ is a rooted graph $(G,v)$ if each ball of radius $r$ around $v_n$ in $G_n$ is, for $n$ large enough, isomorphic to the ball of radius $r$ around $v$ in $G$ (see, e.g.,~\cite[p.\ 1460]{aldous2007processes}).} is given in Figure~\ref{fig:schreier2}, and hence is also a Schreier graph of some transitive $F$-action $F \curvearrowright F/K$. In terms of the topology on the space $\mathrm{Sub}_F \subset \{0,1\}^F$ of the subgroups of $F$, the subgroup $K$ is the limit of the subgroups $K_n$, where $K_n$ is the stabilizer of $\rfrac{1}{2^n}$. It is easy to verify that $K$ is the subgroup of $F$ consisting of the transformations that stabilize $0$ and have right derivative $1$ at $0$ (although this fact will not be important). Let $\Lambda = F/K$. \begin{figure}[ht] \centering \includegraphics[scale=0.6]{schreier2.pdf} \caption{\label{fig:schreier2}The action of $F$ on $\Lambda$.} \end{figure} We can naturally identify with $\mathbb{Z}$ the chain black nodes at the top of $\Lambda$ (see Figure~\ref{fig:schreier2}). Let $\Lambda'$ be the subgraph of $\Lambda$ in which the dotted edges connecting the black nodes have been removed. Given a black node $n \in \mathbb{Z}$, denote by $T_n$ the connected component of $n$ in $\Lambda'$; this includes the black node $n$, the chain that can be reached from it using solid edges, and the entire tree that hangs from it. Each graph $T_n$ is isomorphic to the Schreier graph of $\Gamma$, and so the graph $\Lambda$ is a covering graph of $\Gamma$ (in the category of Schreier graphs). Let \begin{align*} \Psi \colon \Lambda \to \Gamma \end{align*} be the covering map. That is, $\Psi$ is a graph isomorphism when restricted to each $T_n$, with the black nodes in $\Lambda$ mapped to the black node $0 \in \Gamma$. Using the map $\Psi$ we give names to the nodes in $\Lambda$. Denote the nodes in $T_0$ as $\{(0, \gamma) \,:\, \gamma \in \Gamma\}$ so that $\Psi(0,\gamma) = \gamma$. Likewise, in each $T_n$ denote by $(n,\gamma)$ the unique node in $T_n$ that $\Psi$ maps to $\gamma$. Hence we identify $\Lambda$ with \begin{align*} \mathbb{Z} \times \Gamma = \{(n, \gamma)\,:\, n \in \mathbb{Z}, \gamma \in \Gamma\} \end{align*} and the $F$-action is given by \begin{align} \label{a-action-on-Lambda} a (n,\gamma) &= (n, a \gamma)\\ \label{b-action-on-Lambda} b (n,\gamma) &= \begin{cases} (n, b \gamma)&\mbox{if }\gamma \neq 0\\ (n+1, 0)&\mbox{if }\gamma= 0 \end{cases} \end{align} Equip $\{-1,1\}^\Lambda$ with the product topology to get a compact space. As usual, the $F$-action on $\Lambda$ (given explicitly in ~\ref{a-action-on-Lambda} and ~\ref{b-action-on-Lambda}) defines a continuous action on $\{-1,1\}^\Lambda$. Consider $\pi:\{-1,1\}^\Gamma \to \{-1,1\}^\Lambda$, given by $\pi(x)(n, \gamma) = (-1)^n x(\gamma)$. Let $Y = \pi(\{-1,1\}^\Gamma) \subseteq \{-1,1\}^\Lambda$. \begin{claim} \label{clm:compact-and-invariant} $Y$ is compact and $F$-invariant. \end{claim} \begin{proof} $\pi$ is injective and continuous, so $Y = \pi(\{-1,1\}^\Gamma) \subseteq \{-1,1\}^\Lambda$ is compact and isomorphic to $\{-1,1\}^\Gamma$. Moreover, $Y$ is invariant to the action of $F$, because $a^{\pm 1}\pi(x) = \pi (a^{\pm 1}x)$ and $b^{\pm 1}\pi(x) = \pi(b^{\pm}\bar{x})$ where $\bar{x}(\gamma) = \begin{cases} x(\gamma)&\mbox{if }\gamma \neq 0\\ -x(\gamma)&\mbox{if } \gamma = 0 \end{cases}$. \end{proof} The last $F$-space we define is $Z$, the set of pairs of mirror image configurations in $Y$: \begin{align} \label{the-space-Z} Z = \left\{\{y, -y\}\,:\,y\in Y \right\}. \end{align} Now it is clear that equipped with the quotient topology, $Z$ is a compact and Hausdorff $F$-space. Furthermore, we now observe that $Z$ admits an invariant measure. Consider the i.i.d.\ Bernoulli $1/2$ measure on $\{-1,1\}^\Gamma$, i.e. the unique Borel measure on $\{-1,1\}^\Gamma$, for which \begin{align*} X_\gamma \colon & \{-1,1\}^\Gamma \to \{0, 1\},\quad x\mapsto \frac{x(\gamma)+1}{2} \end{align*} are independent Bernoulli $1/2$ random variables for all $\gamma \in \Gamma$. Clearly, it is an invariant measure and hence it is pushed forward to an invariant measure on $Y$, and then on $Z$. In particular, this shows that $Z$ is not strongly proximal. \begin{claim} \label{clm:no-fixed-points} The action $F \curvearrowright Z$ does not have any fixed points. \end{claim} \begin{proof} Pick $\hat{y} = \{y, -y\}\in Z$. We have $[by](0, -1) = y(0, -1) \neq -y(0, -1)$, so $by\neq -y$. Similarly, $[b y](0, 0) = y(-1, 0) = -y(0, 0) \neq y(0, 0)$, and so $by \neq y$. Hence $b\hat{y}\neq \hat{y}$. \end{proof} \begin{proposition} \label{thm:proximal} The action $F \curvearrowright Z$ is proximal. \end{proposition} \begin{proof} Let $\hat{y_1}=\{y_1, -y_1\}$ and $\hat{y_2}=\{y_2,-y_2\}$ be two points in $Z$, and let $y_i=\pi(x_i)$. Let $x_1 \cdot x_2$ denote the pointwise product of $x_1$ and $x_2$. Now by Proposition~\ref{prop:pre_proximal} there is a sequence of elements $\{f_n\}_n$ in $F$ such that $\{f_n (x_1 \cdot x_2)\}_n$ tends to either $c_{-1}$ or $c_{+1}$ in $\{-1,1\}^\Gamma$. Since $Y$ is compact, we may assume that $\{f_n y_1\}_n$ and $\{f_n y_2\}_n$ have limits, by descending to a subsequence if necessary. It is straightforward to check that $f_n y_1 \cdot f_n y_2 = f_n\pi(x_1)\cdot f_n\pi(x_2)=\pi(f_n x_1) \cdot \pi(f_n x_2)$. So: \begin{align*} [f_n y_1 \cdot f_n y_2](n,\gamma) &= [\pi(f_n x_1) \cdot \pi(f_n x_2)](n, \gamma)\\ &= (-1)^{2n}\;[f_n x_1](\gamma)\;[f_n x_2](\gamma)\\ &=[f_n x_1 \cdot f_n x_2](\gamma) = [f_n (x_1 \cdot x_2)](\gamma) \end{align*} So $\lim_n f_n y_1 = \pm \lim_n f_n y_2$, which implies $\lim_n f_n \hat{y_1} = \lim_n f_n \hat{y_2}$. \end{proof} \begin{theorem} Thompson's group $F$ is not strongly amenable. \end{theorem} \begin{proof} Since the space $Z$ we constructed above is proximal (Proposition~\ref{thm:proximal}), and has no fixed points (Claim~\ref{clm:no-fixed-points}), we conclude that $F$ has a proximal action with no fixed points, so $F$ is not strongly amenable. \end{proof}
307a7516e620bdb53ec72d9db06c57a1b2448573
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:1} The influence of social groups in pedestrian dynamics, especially in evacuation scenarios, is an area of recent interest, see e.g. \cite{mueller2,mueller} and other contributions in these proceedings. The situations that are considered are widespread and well-known in everyday life. For example, many people visit concerts or soccer matches not alone, but together with family and friends in so-called social groups. In case of emergency, these groups will try to stay together during an evacuation. The strength of this cohesion depends on the composition of the social group. Several adult friends would form a loose group that is mainly connected via eye contact, whereas a mother would take her child's hand and form a strong or even fixed bond. In addition, even the size of the social groups could have an effect on the evacuation behaviour. In order to consider these phenomena in a more detailed way, a cooperation of researchers of the universities of Cologne and Wuppertal and the Forschungszentrum J\"ulich has performed several experiments aiming at the determination of the general influence of inhomogeneities on pedestrian dynamics. They contained two series of experiments with pupils of different ages in two schools in Wuppertal. The first series focussed on the determination of the fundamental diagram of inhomogeneous groups, i.e. pedestrians of different size. The second series of experiments considered evacuation scenarios. In several runs the parameters of the crowd of evacuating pupils were varied, i.e. the size of the social group and its structure and the interaction between the group members. Here we present first results for these evacuation experiments. \section{Teaching units} \label{sec:2} The experiments were accompanied by teaching units for all involved students providing an introduction into the topic of traffic and pedestrian dynamics. In classes of fifth and sixth grade, the focus of the classes was on the important quantities of pedestrian dynamics, especially density, time and bottleneck situations. This introduction to crowd effects and pedestrian behaviour was intended to raise awareness for their relevance for their everyday lives and safety issues. Therefore we arranged little experiments the students could perform themselves, e.g. the panic experiment according to Mintz \cite{mintz} (see Fig.~\ref{fig:1}). In small groups the pupils had to pull several wooden wedges out of a bottle with a narrow neck as fast as possible and observe the blocking of the wedges when every students pulls at the same time. This experiment was supposed to indicate that coordination can lead to better results compared to selfish behavior. The older pupils of classes 10 and 11 participated in an introduction to cellular automata and the physics of traffic. They received several worksheets on the Game of Life and other cellular automata, especially the Nagel-Schreckenberg model \cite{nasch}. The aim of these lessons was to obtain a first qualitative and quantitative understanding of the collective effects in traffic systems. This should help to increase the identification with the experiments they later participated in and raise awareness about the relevance of this kind of research for everyday life. \begin{figure}[t] \sidecaption \includegraphics[scale=.045]{P1020628.jpg} \caption{Panic experiment according to Mintz. Every pupil is assigned a cord with a wedge on its end lying in a bottle with a narrow neck. If every student pulls at the same time and as fast as possible, the wedges block at the bottleneck. On the other hand, behaving in a coordinate way leads to a smooth process that is significantly faster. (Photo by V.~Ziemer, U Wuppertal)} \label{fig:1} \end{figure} \section{Experimental set-up and procedure} \label{sec:3} The experiments were performed in two schools in Wuppertal. Students of four classes participated as part of project work. The experimental room was built in the school's assembly hall. \subsection{Experimental set-up} The experimental area was a square room of $5\times 5$~m$^2$ bounded by several small buckets. In the center of this area there was a square starting area of $3\times 3$~m$^2$ denoted by the white marks. The students stepped into the room through the entrance that is shown below in Fig.~\ref{fig:2} and assembled in the starting area. During the evacuation they had to leave the room using the exit on the left side. The exit door was built by two upstanding platforms and had a variable width changing between 0.8~m and 1.2~m. The area behind the door was connected to the waiting area before the entrance, so the students could walk on a closed path. For the collection of data all experiments were recorded by a camera system. This system was mounted on the hall's ceiling and contained customary digital cameras and GoPros. All students wore caps of different colour. Each colour represented a certain interval of body heights. The body height of each pupil was measured before the experiments started. This information is needed to determine the position accurately, but the different colours can also be used to draw conclusions about the composition of the group of evacuating pupils later in the video. All caps showed also a black point at the middle of the head. That allows to recognize and track each person in the video. \begin{figure}[t] \sidecaption \includegraphics[scale=.13]{RaumSchueler.png} \caption{The experimental area consists of a square room of $5\times 5$~m$^2$, with a starting area in the middle. The pupils had to leave this room through the exit at the left. They wore colourful caps to distinguish the different body heights.} \label{fig:2} \end{figure} \subsection{Experimental procedure} In general, the pupils had to perform several evacuation runs. For each run, a group of 32 to 46 persons assembled in the starting area, distributed nearly uniformly. During the evacuation, the students were allowed to use the whole experimental area. After a starting signal, the participants had to leave the room using the exit door. They should walk briskly and evacuate the room as fast as possible. The pupils were told to imagine there would be a kind of danger, like fire or smoke. However, they were not allowed to run, scramble or push each other. After leaving the room they had to assemble again in the waiting area in front of the entrance and to wait for the next run. The group of pupils that was placed into the experiment was compound in different ways to consider different parameters. The first parameter that was varied in the experiment was the composition of the entire group. At all, there were two different age classes allowing for three different group compositions. The crowd could consist only of children aged 10 to 12 years, only of young adults aged 15 to 17, or a mixture of both groups whereby children and youths were equally represented. The second parameter concerned the social group size. In several runs, the students had either to evacuate on their own without regarding the others around them, or to form pairs, or larger social groups. These groups could contain four, six or eight persons. Within one pair or social group the students had to try to stay together during the evacuation run. As a third parameter we considered the interaction within the social group. The nature of the interaction can be specified by (i) its strength and (ii) the hierarchy of group members. Regarding the interaction strength, the group members could either be connected loosely, by just trying to stay together via eye contact, or they could have a fixed bond. A fixed bond was realized by holding each other's hand or some other physical contact. Furthermore, hierarchy of the group members could be different. In the first case, all partners were treated equally. Each group member had to leave the room and to stay together with their partners. In the other case, one student was declared as the "leader", the other one as the "follower". The leader had to leave the room without regarding its partner or the other students, whereas the follower just had to follow the leader through the room. This leads to four different ways to form pairs during the evacuation run. In the case of age-matched partners, the leader was chosen randomly. For the runs with the mixed crowd, the pairs were composed of one child and one teenager that took the part of the leader in the runs they were needed. All runs with larger social groups were done with loose bonds. In social groups of same age, there was no leader, but in mixed social groups one of the youths was declared as the leader. To analyze the experiments the videos of the camera system that was mounted on the hall's ceiling was available. For each run of the experiment there is a video sequence. Using the \texttt{PeTrack} software \cite{boltes}, it is possible to extract the trajectories for each person and each run. The students were recognized via the black point in the middle of their coloured caps. The position of this point was tracked in each frame, generating the trajectory of each participant. \section{Analysis} \label{sec:4} First, we focus on the analysis of the data regarding the influence of group size on the evacuation scenario. Therefore, we use the data of one school and of the runs with larger groups. Most of these experiments were performed only with the older pupils with loose bonds and no leader-follower relationship, to which we restrict our analysis for now. In different runs, the students formed groups of four, six and eight persons. In addition, one run with groups of six students and with an explicitly cooperative behaviour within the group was performed. They should concentrate a bit more on their group members and try to leave the room together. For comparison, we also consider the run with pairs and a loose bond that can be seen as a smaller group of two persons. \subsection{Evacuation times for large groups} First we consider the evacuation time. In Fig.~\ref{fig:3}, we plot the number of evacuated persons against the time needed to leave the room. The results can be compared between the different runs. The evacuation time for each person is defined as the time difference between the beginning of the evacuation and the moment when the person passes the door, exactly when he/she leaves the aisle that is formed by the two platforms. The beginning of the evacuation can be determined only approximately because the starting signal is not audible in the videos that are used for the extraction of the trajectories. For extracting the evacuation times we set the beginning on the moment of the first movement towards the door. However, for the analysis the influences of this inaccurate definition, the pre-movement time or other delays should be minimized. In doing so, we take the evacuation time of the very first person that left the room and subtract it from all the other times. Thereby, all plots start at zero for the first person and it is easier to compare different runs. For the analysis of the runs with larger groups the evacuation times are shown in Fig.~\ref{fig:3}. All graphs show a nearly linear behaviour that could be expected. At the beginning of the evacuation all evacuation times are roughly the same. Between three and six evacuated persons the curves start to split into two groups. After increasing slightly, the difference between the two progresses remains nearly constant until the end of the evacuation. The main insight is that there are several runs that are clearly faster than other ones. The upper two curves represent the evacuation in pairs and in groups of six with very cooperative behaviour. The lower graphs show the runs with larger groups of four, six and eight persons. Within the two groups of curves the differences are not large enough to separate the runs from each other. However, in the lower group, the run with eight participants per group seems to be a bit faster at the end of the evacuation. The run with six participants per group and cooperative behaviour is clearly slower than the run with same group size but without this instruction. These results indicate that forming groups is advantageous for the evacuation, whereas behaving cooperatively inhibits this effect. \begin{figure}[t] \sidecaption \includegraphics[width=\textwidth]{Zeiten.png} \caption{Evacuation times for larger groups of one of the schools. The splitting of the curves into two groups is obvious. The shorter evacuation times belong to the runs with larger groups of four, six and eight persons, the longer ones to those with pairs and cooperative behaviour.} \label{fig:3} \end{figure} While looking for reasons for the differences in evacuation times, one first approach could be to determine the density distribution. Therefore, we determined the Voronoi cells \cite{steffen,voronoi} within the experimental room for each person at different times. As a measure of density we coloured all cells dependent on their size: smaller cells are coloured in shades of red, larger ones in blue. In Fig.~\ref{fig:4} the density distributions for the run with pairs and with groups of four persons are shown. It is clearly seen that the distribution for the run in (b) is a bit narrower than the other one at the same time step. That means when forming groups, the children order rather behind each other than next to each other in front of the door. This behaviour seems to be advantageous for evacuating the room as it leads to a shorter evacuation time. \begin{figure}[t] \begin{center} \sidecaption \subfloat[Pairs at 4 s]{\includegraphics[scale=.30]{Voronoi-2er-4s.png}} \qquad \subfloat[Groups of four at 4 s]{\includegraphics[scale=.30]{Voronoi-4er-4s.png}} \caption{Voronoi cells for each person in the room at time 4~s for the runs with pairs and social groups of four persons. The colour of each cell depends on its size as a measure of density.} \label{fig:4} \end{center} \end{figure} \subsection{First attempts to interpretation} \label{sec:5} The results obtained so far suggest certain interpretations which, however, need to be substantiated by further experiments with better statistics. It is obvious from the plot of the evacuation times that increasing the group size leads to a decrease in evacuation times. The density distributions show the pupils ordered in different ways for forming groups than for pairs. A possible explanation is that the persons subordinate within the group and just follow the other group members. Because of that, there may be less conflicts between persons that meet at the door in competing for space. A person is just in competition with persons of other groups, not with own group members. Increasing the group size reduces the number of possible competitors. This reduction of conflicts may have a positive influence on the evacuation time. When the children have to show cooperative behaviour, the evacuation is slower than without this instruction. It is a possible explanation that here the effort to stay together is larger and reduces the effort to leave the room. \section{Summary and outlook} \label{sec:6} We performed experiments under laboratory conditions to determine the influence of social groups on evacuations. A comparison of evacuation times between runs with different group sizes shows that increasing the group sizes lowers the evacuation time. The participants order in a different way for larger groups. These first preliminary results have to be analyzed in more detail. The statistics need to be improved by further experiments. However, with the help of the density distributions, photographs of the finish and the data of the second school we hope to get more information from the present experiments, e.g. about the microscopic mechanisms especially close to the exit. In addition, there are some few parameters that should also be analyzed, e.g. the effect of body size and age. \begin{acknowledgement} We dedicate this contribution in grateful memory to our friend and colleague Matthias Craesmeyer.\\ We thank the team from the Forschungszentrum J\"ulich and Wuppertal University and the students and teachers of Gymnasium Bayreuther Stra\ss{}e and Wilhelm-D\"orpfeld-Gymnasium in Wuppertal for the help with the experiments. Financial support by the DFG under grant SCHA~636/9-1 is gratefully acknowledged. \end{acknowledgement} \bibliographystyle{spphys.bst}
e446ef41a60af69ad04b265c37552ed3922672e1
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} Policy search algorithms based on supervised learning from a computational or human ``teacher'' have gained prominence in recent years due to their ability to optimize complex policies for autonomous flight \cite{rmswd-lmruc-13}, video game playing \cite{rgb-rilsp-11,gsllw-amcts-14}, and bipedal locomotion \cite{mlapt-icdcc-15}. Among these methods, guided policy search algorithms \cite{levine2015end} are particularly appealing due to their ability to adapt the teacher to produce data that is best suited for training the final policy with supervised learning. Such algorithms have been used to train complex deep neural network policies for vision-based robotic manipulation \cite{levine2015end}, as well as a variety of other tasks \cite{zkla-ldcpa-15,mlapt-icdcc-15}. However, convergence results for these methods typically follow by construction from their formulation as a constrained optimization, where the teacher is gradually constrained to match the learned policy, and guarantees on the performance of the final policy only hold at convergence if the constraint is enforced exactly. This is problematic in practical applications, where such algorithms are typically executed for a small number of iterations. In this paper, we show that guided policy search algorithms can be interpreted as approximate variants of mirror descent under constraints imposed by the policy parameterization, with supervised learning corresponding to a projection onto the constraint manifold. Based on this interpretation, we can derive a new, simplified variant of guided policy search, which corresponds exactly to mirror descent under linear dynamics and convex policy spaces. When these convexity and linearity assumptions do not hold, we can show that the projection step is approximate, up to a bound that depends on the step size of the algorithm, which suggests that for a small enough step size, we can achieve continuous improvement. The form of this bound provides us with intuition about how to adjust the step size in practice, so as to obtain a simple algorithm with a small number of hyperparameters. The main contribution of this paper is a simple new guided policy search algorithm that can train complex, high-dimensional policies by alternating between trajectory-centric reinforcement learning and supervised learning, as well as a connection between guided policy search methods and mirror descent. We also extend previous work on bounding policy cost in terms of KL divergence \cite{rgb-rilsp-11,slmja-trpo-15} to derive a bound on the cost of the policy at each iteration, which provides guidance on how to adjust the step size of the method. We provide empirical results on several simulated robotic navigation and manipulation tasks that show that our method is stable and achieves similar or better performance when compared to prior guided policy search methods, with a simpler formulation and fewer hyperparameters. \section{Guided Policy Search Algorithms} \label{sec:gps} We first review guided policy search methods and background. Policy search algorithms aim to optimize a parameterized policy $\pi_\theta(\action_t|\state_t)$ over actions $\action_t$ conditioned on the state $\state_t$. Given stochastic dynamics $p(\mathbf{x}_{t+1}|\state_t,\action_t)$ and cost $\ell(\state_t,\action_t)$, the goal is to minimize the expected cost under the policy's trajectory distribution, given by \mbox{$J(\theta) = \sum_{t=1}^T E_{\pi_\theta(\state_t,\action_t)} [\ell(\state_t,\action_t)]$}, where we overload notation to use $\pi_\theta(\state_t,\action_t)$ to denote the marginals of $\pi_\theta(\tau) = p(\mathbf{x}_1)\prod_{t=1}^T p(\mathbf{x}_{t+1}|\state_t,\action_t) \pi_\theta(\action_t|\state_t)$, where $\tau =\{\mathbf{x}_1,\mathbf{u}_1,\dots,\mathbf{x}_T,\mathbf{u}_T\}$ denotes a trajectory. A standard reinforcement learning (RL) approach to policy search is to compute the gradient $\nabla_\theta J(\theta)$ and use it to improve $J(\theta)$ \cite{w-ssgfa-92,ps-rlmsp-08}. The gradient is typically estimated using samples obtained from the real physical system being controlled, and recent work has shown that such methods can be applied to very complex, high-dimensional policies such as deep neural networks \cite{slmja-trpo-15,lhphe-ccdrl-16}. However, for complex, high-dimensional policies, such methods tend to be inefficient, and practical real-world applications of such model-free policy search techniques are typically limited to policies with about one hundred parameters \cite{dnp-spsr-13}. Instead of directly optimizing $J(\theta)$, guided policy search algorithms split the optimization into a ``control phase'' (which we'll call the C-step) that finds multiple simple local policies $p_i(\action_t|\state_t)$ that can solve the task from different initial states $\mathbf{x}_1^i \sim p(\mathbf{x}_1)$, and a ``supervised phase'' (S-step) that optimizes the global policy $\pi_\theta(\action_t|\state_t)$ to match all of these local policies using standard supervised learning. In fact, a variational formulation of guided policy search \cite{lk-vpsto-13} corresponds to the EM algorithm, where the C-step is actually the E-step, and the S-step is the M-step. The benefit of this approach is that the local policies $p_i(\action_t|\state_t)$ can be optimized separately using domain-specific local methods. Trajectory optimization might be used when the dynamics are known \cite{zkla-ldcpa-15,mlapt-icdcc-15}, while local RL methods might be used with unknown dynamics \cite{la-lnnpg-14,levine2015end}, which still requires samples from the real system, though substantially fewer than the direct approach, due to the simplicity of the local policies. This sample efficiency is the main advantage of guided policy search, which can train policies with nearly a hundred thousand parameters for vision-based control using under 200 episodes \cite{levine2015end}, in contrast to direct deep RL methods that might require orders of magnitude more experience \cite{slmja-trpo-15,lhphe-ccdrl-16}. \begin{algorithm}[t] \caption{Generic guided policy search method \label{alg:gps}} \begin{algorithmic}[1] \FOR{iteration $k \in \{1, \dots, K\}$} \STATE C-step: improve each $p_i(\action_t|\state_t)$ based on surrogate cost $\tilde{\ell}_i(\state_t,\action_t)$, return samples $\mathcal{D}_i$ \label{algline:gps_local} \STATE S-step: train $\pi_\theta(\action_t|\state_t)$ with supervised learning on the dataset $\mathcal{D} = \cup_i \mathcal{D}_i$ \label{algline:gps_global} \STATE Modify $\tilde{\ell}_i(\state_t,\action_t)$ to enforce agreement between $\pi_\theta(\action_t|\state_t)$ and each $p(\action_t|\state_t)$ \label{algline:gps_dual} \ENDFOR \end{algorithmic} \end{algorithm} A generic guided policy search method is shown in Algorithm~\ref{alg:gps}. The C-step invokes a local policy optimizer (trajectory optimization or local RL) for each $p_i(\action_t|\state_t)$ on line~\ref{algline:gps_local}, and the S-step uses supervised learning to optimize the global policy $\pi_\theta(\action_t|\state_t)$ on line~\ref{algline:gps_global} using samples from each $p_i(\action_t|\state_t)$, which are generated during the C-step. On line~\ref{algline:gps_dual}, the surrogate cost $\tilde{\ell}_i(\state_t,\action_t)$ for each $p_i(\action_t|\state_t)$ is adjusted to ensure convergence. This step is crucial, because supervised learning does not in general guarantee that $\pi_\theta(\action_t|\state_t)$ will achieve similar long-horizon performance to $p_i(\action_t|\state_t)$ \cite{rgb-rilsp-11}. The local policies might not even be reproducible by a single global policy in general. To address this issue, most guided policy search methods have some mechanism to force the local policies to agree with the global policy, typically by framing the entire algorithm as a constrained optimization that seeks at convergence to enforce equality between $\pi_\theta(\action_t|\state_t)$ and each $p_i(\action_t|\state_t)$. The form of the overall optimization problem resembles dual decomposition, and usually looks something like this: \begin{align} \min_{\theta,p_1,\dots,p_N} \sum_{i=1}^N \sum_{t=1}^T E_{p_i(\state_t,\action_t)}[\ell(\state_t,\action_t)] \text{ such that } p_i(\action_t|\state_t) = \pi_\theta(\action_t|\state_t) \,\,\,\forall \state_t, \action_t, t, i. \label{eqn:gps} \end{align} Since $\mathbf{x}_1^i \sim p(\mathbf{x}_1)$, we have $J(\theta) \approx \sum_{i=1}^N \sum_{t=1}^T E_{p_i(\state_t,\action_t)}[\ell(\state_t,\action_t)]$ when the constraints are enforced exactly. The particular form of the constraint varies depending on the method: prior works have used dual gradient descent \cite{lwa-lnnpg-15}, penalty methods \cite{mlapt-icdcc-15}, ADMM \cite{mt-cbfat-14}, and Bregman ADMM \cite{levine2015end}. We omit the derivation of these prior variants due to space constraints. \subsection{Efficiently Optimizing Local Policies} A common and simple choice for the local policies $p_i(\action_t|\state_t)$ is to use time-varying linear-Gaussian controllers of the form $p_i(\action_t|\state_t) = \mathcal{N}(\mathbf{K}_t\state_t + \mathbf{k}_t, \mathbf{C}_t)$, though other options are also possible \cite{mt-cbfat-14,mlapt-icdcc-15,zkla-ldcpa-15}. Linear-Gaussian controllers represent individual trajectories with linear stabilization and Gaussian noise, and are convenient in domains where each local policy can be trained from a different (but consistent) initial state $\mathbf{x}_1^i \sim p(\mathbf{x}_1)$. This represents an additional assumption beyond standard RL, but allows for an extremely efficient and convenient local model-based RL algorithm based on iterative LQR \cite{lt-ilqr-04}. The algorithm proceeds by generating $N$ samples on the real physical system from each local policy $p_i(\action_t|\state_t)$ during the C-step, using these samples to fit local linear-Gaussian dynamics for each local policy of the form $p_i(\mathbf{x}_{t+1}|\state_t,\action_t) = \mathcal{N}(f_{\state t}\state_t + f_{\action t}\action_t + f_{c t},\mathbf{F}_t)$ using linear regression, and then using these fitted dynamics to improve the linear-Gaussian controller via a modified LQR algorithm \cite{la-lnnpg-14}. This modified LQR method solves the following optimization problem: \begin{equation} \min_{\mathbf{K}_t,\mathbf{k}_t,\mathbf{C}_t} \sum_{t=1}^T E_{p_i(\state_t,\action_t)}[\tilde{\ell}_i(\state_t,\action_t)] \text{ such that } D_\text{KL}(p_i(\tau) \| \bar{p}_i(\tau)) \leq \epsilon,\label{eqn:lqrkl} \end{equation} where we again use $p_i(\tau)$ to denote the trajectory distribution induced by $p_i(\action_t|\state_t)$ and the fitted dynamics $p_i(\mathbf{x}_{t+1}|\state_t,\action_t)$. Here, $\bar{p}_i(\action_t|\state_t)$ denotes the previous local policy, and the constraint ensures that the change in the local policy is bounded, as proposed also in prior works \cite{bs-cps-03,ps-rlmsp-08,pma-reps-10}. This is particularly important when using linearized dynamics fitted to local samples, since these dynamics are not valid outside of a small region around the current controller. In the case of linear-Gaussian dynamics and policies, the KL-divergence constraint $D_\text{KL}(p_i(\tau) \| \bar{p}_i(\tau)) \leq \epsilon$ can be shown to simplify, as shown in prior work~\cite{la-lnnpg-14} and Appendix~\ref{app:kldiv}: \[ D_\text{KL}(p_i(\tau) \| \bar{p}_i(\tau)) \!=\! \sum_{t=1}^T D_\text{KL}(p_i(\action_t|\state_t) \| \bar{p}_i(\action_t|\state_t) ) \!=\! \sum_{t=1}^T \!-\! E_{p_i(\state_t,\action_t)} [ \log \bar{p}_i(\action_t|\state_t) ] - \mathcal{H}(p_i(\action_t|\state_t)), \] and the resulting Lagrangian of the problem in Equation~(\ref{eqn:lqrkl}) can be optimized with respect to the primal variables using the standard LQR algorithm, which suggests a simple method for solving the problem in Equation~(\ref{eqn:lqrkl}) using dual gradient descent~\cite{la-lnnpg-14}. The surrogate objective $\tilde{\ell}_i(\state_t,\action_t) = \ell(\state_t,\action_t) + \phi_i(\theta)$ typically includes some term $\phi_i(\theta)$ that encourages the local policy $p_i(\action_t|\state_t)$ to stay close to the global policy $\pi_\theta(\action_t|\state_t)$, such as a KL-divergence of the form $D_\text{KL}(p_i(\action_t|\state_t) \| \pi_\theta(\action_t|\state_t))$. \subsection{Prior Convergence Results} Prior work on guided policy search typically shows convergence by construction, by framing the C-step and S-step as block coordinate ascent on the (augmented) Lagrangian of the problem in Equation~(\ref{eqn:gps}), with the surrogate cost $\tilde{\ell}_i(\state_t,\action_t)$ for the local policies corresponding to the (augmented) Lagrangian, and the overall algorithm being an instance of dual gradient descent \cite{lwa-lnnpg-15}, ADMM \cite{mt-cbfat-14}, or Bregman ADMM \cite{levine2015end}. Since these methods enforce the constraint $p_i(\action_t|\state_t) = \pi_\theta(\action_t|\state_t)$ at convergence (up to linearization or sampling error, depending on the method), we know that $\frac{1}{N}\sum_{i=1}^N E_{p_i(\state_t,\action_t)}[\ell(\state_t,\action_t)] \approx E_{\pi_\theta(\state_t,\action_t)}[\ell(\state_t,\action_t)]$ at convergence.\footnote{As mentioned previously, the initial state $\mathbf{x}_1^i$ of each local policy $p_i(\action_t|\state_t)$ is assumed to be drawn from $p(\mathbf{x}_1)$, hence the outer sum corresponds to Monte Carlo integration of the expectation under $p(\mathbf{x}_1)$.} However, prior work does not say anything about $\pi_\theta(\action_t|\state_t)$ at intermediate iterations, and the constraints of policy search in the real world might often preclude running the method to full convergence. We propose a simplified variant of guided policy search, and present an analysis that sheds light on the performance of both the new algorithm and prior guided policy search methods. \section{Mirror Descent Guided Policy Search} \label{sec:md} \vspace{-0.1in} \begin{algorithm}[t] \caption{Mirror descent guided policy search (MDGPS): convex linear variant \label{alg:mdgps}} \begin{algorithmic}[1] \FOR{iteration $k \in \{1, \dots, K\}$} \STATE C-step: $p_i \leftarrow \arg\min_{p_i} E_{p_i(\tau)}\left[\sum_{t=1}^T \ell(\state_t,\action_t)\right] \text{ such that } D_\text{KL}(p_i(\tau) \| \pi_\theta(\tau)) \leq \epsilon$ \STATE S-step: $\pi_\theta \leftarrow \arg\min_\theta \sum_i D_\text{KL}(p_i(\tau) \| \pi_\theta(\tau))$ (via supervised learning) \ENDFOR \end{algorithmic} \end{algorithm} In this section, we propose our new simplified guided policy search, which we term mirror descent guided policy search (MDGPS). This algorithm uses the constrained LQR optimization in Equation~(\ref{eqn:lqrkl}) to optimize each of the local policies, but instead of constraining each local policy $p_i(\action_t|\state_t)$ against the previous local policy $\bar{p}_i(\action_t|\state_t)$, we instead constraint it directly against the global policy $\pi_\theta(\action_t|\state_t)$, and simply set the surrogate cost to be the true cost, such that $\tilde{\ell}_i(\state_t,\action_t) = \ell(\state_t,\action_t)$. The method is summarized in Algorithm~\ref{alg:mdgps}. In the case of linear dynamics and a quadratic cost (i.e. the LQR setting), and assuming that supervised learning can globally solve a convex optimization problem, we can show that this method corresponds to an instance of mirror descent \cite{bt-mdnps-03} on the objective $J(\theta)$. In this formulation, the optimization is performed on the space of trajectory distributions, with a constraint that the policy must lie on the manifold of policies with the chosen parameterization. Let $\Pi_\Theta$ be the set of all possible policies $\pi_\theta$ for a given parameterization, where we overload notation to also let $\Pi_\Theta$ denote the set of trajectory distributions that are possible under the chosen parameterization. The return $J(\theta)$ can be optimized according to $\pi_\theta \leftarrow \arg\min_{\pi \in \Pi_\Theta} E_{\pi(\tau)}[\sum_{t=1}^T \ell(\state_t,\action_t)]$. Mirror descent solves this optimization by alternating between two steps at each iteration $k$: \vspace{-0.05in} \begin{align*} p^k \leftarrow \arg\min_p E_{p(\tau)}\left[\sum_{t=1}^T \ell(\state_t,\action_t) \right] \text{ s. t. } D\left(p,\pi^k\right) \leq \epsilon ,\hspace{0.4in} \pi^{k + 1} \leftarrow \arg\min_{\pi \in \Pi_\Theta} D\left(p^k, \pi\right). \end{align*} The first step finds a new distribution $p^k$ that minimizes the cost and is close to the previous policy $\pi^k$ in terms of the divergence $D\left(p,\pi^k\right)$, while the second step projects this distribution onto the constraint set $\Pi_\Theta$, with respect to the divergence $D(p^k, \pi)$. In the linear-quadratic case with a convex supervised learning phase, this corresponds exactly to Algorithm~\ref{alg:mdgps}: the C-step optimizes $p^k$, while the S-step is the projection. Monotonic improvement of the global policy $\pi_\theta$ follows from the monotonic improvement of mirror descent \cite{bt-mdnps-03}. In the case of linear-Gaussian dynamics and policies, the S-step, which minimizes KL-divergence between trajectory distributions, in fact only requires minimizing the KL-divergence between policies. Using the identity in Appendix~\ref{app:kldiv}, we know that \begin{equation} \vspace{-0.05in} D_\text{KL}(p_i(\tau) \| \pi_\theta(\tau)) = \sum_{t=1}^T E_{p_i(\state_t)}\left[D_\text{KL}(p_i(\action_t|\state_t) \| \pi_\theta(\action_t|\state_t))\right]. \label{eqn:mirrorkl} \end{equation} \subsection{Implementation for Nonlinear Global Policies and Unknown Dynamics} \label{sec:implementation} \vspace{-0.05in} In practice, we aim to optimize complex policies for nonlinear systems with unknown dynamics. This requires a few practical considerations. The C-step requires a local quadratic cost function, which can be obtained via Taylor expansion, as well as local linear-Gaussian dynamics $p(\mathbf{x}_{t+1}|\state_t,\action_t) = \mathcal{N}(f_{\state t}\state_t + f_{\action t}\action_t + f_{c t},\mathbf{F}_t)$, which we can fit to samples as in prior work \cite{la-lnnpg-14}. We also need a local time-varying linear-Gaussian approximation to the global policy $\pi_\theta(\action_t|\state_t)$, denoted $\bar{\pi}_{\theta i}(\action_t|\state_t)$. This can be obtained either by analytically differentiating the policy, or by using the same linear regression method that we use to estimate $p(\mathbf{x}_{t+1}|\state_t,\action_t)$, which is the approach in our implementation. In both cases, we get a different global policy linearization around each local policy. Following prior work \cite{la-lnnpg-14}, we use a Gaussian mixture model prior for both the dynamics and global policy fit. \begin{algorithm}[t] \caption{Mirror descent guided policy search (MDGPS): unknown nonlinear dynamics \label{alg:mdgpsfull}} \begin{algorithmic}[1] \FOR{iteration $k \in \{1, \dots, K\}$} \STATE Generate samples $\mathcal{D}_i = \{\tau_{i,j}\}$ by running either $p_i$ or $\pi_{\theta i}$ \STATE Fit linear-Gaussian dynamics $p_i(\mathbf{x}_{t+1}|\state_t,\action_t)$ using samples in $\mathcal{D}_i$ \STATE Fit linearized global policy $\bar{\pi}_{\theta i}(\action_t|\state_t)$ using samples in $\mathcal{D}_i$ \STATE C-step: $p_i \leftarrow \arg\min_{p_i} E_{p_i(\tau)}[\sum_{t=1}^T \ell(\state_t,\action_t)] \text{ such that } D_\text{KL}(p_i(\tau) \| \bar{\pi}_{\theta i}(\tau)) \leq \epsilon$ \STATE S-step: $\pi_\theta \leftarrow \arg\min_\theta \sum_{t,i,j} D_\text{KL}(\pi_\theta(\action_t|\mathbf{x}_{t,i,j}) \| p_i(\action_t|\mathbf{x}_{t,i,j}))$ (via supervised learning) \STATE Adjust $\epsilon$ (see Section~\ref{sec:step}) \ENDFOR \end{algorithmic} \end{algorithm} The S-step can be performed approximately in the nonlinear case by using the samples collected for dynamics fitting to also train the global policy. Following prior work \cite{levine2015end}, our S-step minimizes\footnote{Note that we flip the KL-divergence inside the expectation, following~\cite{levine2015end}. We found that this produced better results. The intuition behind this is that, because $\log p_i(\action_t|\state_t)$ is proportional to the Q-function of $p_i(\action_t|\state_t)$ (see Appendix~\ref{app:cstep}), $D_\text{KL}(\pi_\theta(\action_t|\mathbf{x}_{t,i,j}) \| p_i(\action_t|\mathbf{x}_{t,i,j})$ minimizes the cost-to-go under $p_i(\action_t|\state_t)$ with respect to $\pi_\theta(\action_t|\state_t)$, which provides for a more informative objective than the unweighted likelihood in Equation~(\ref{eqn:mirrorkl}).} \[ \sum_{i,t} E_{p_i(\state_t)}\left[D_\text{KL}(\pi_\theta(\action_t|\state_t) \| p_i(\action_t|\state_t)) \right] \approx \frac{1}{|\mathcal{D}_i|}\sum_{i,t,j} D_\text{KL}(\pi_\theta(\action_t|\mathbf{x}_{t,i,j}) \| p_i(\action_t|\mathbf{x}_{t,i,j})), \] \noindent where $\mathbf{x}_{t,i,j}$ is the $j^\text{th}$ sample from $p_i(\state_t)$ obtained by running $p_i(\action_t|\state_t)$ on the real system. For linear-Gaussian $p_i(\action_t|\state_t)$ and (nonlinear) conditionally Gaussian $\pi_\theta(\action_t|\state_t) = \mathcal{N}(\mu^\policy(\state_t),\Sigma^\policy(\state_t))$, where $\mu^\policy$ and $\Sigma^\policy$ can be any function (such as a deep neural network), the KL-divergence $D_\text{KL}(\pi_\theta(\action_t|\mathbf{x}_{t,i,j}) \| p_i(\action_t|\mathbf{x}_{t,i,j}))$ can easily be evaluated and differentiated in closed form~\cite{levine2015end}. However, in the nonlinear setting, minimizing this objective no longer minimizes the KL-divergence between trajectory distributions $D_\text{KL}(\pi_\theta(\tau) \| p_i(\tau))$ exactly, which means that MDGPS does not correspond exactly to mirror descent: although the C-step can still be evaluated exactly, the S-step now corresponds to an approximate projection onto the constraint manifold. In the next section, we discuss how we can bound the error in this projection. A summary of the nonlinear MDGPS method is provided in Algorithm~\ref{alg:mdgpsfull}, and additional details are in Appendix~\ref{app:mdgpsdetails}. The samples for linearizing the dynamics and policy can be obtained by running either the last local policy $p_i(\action_t|\state_t)$, or the last global policy $\pi_\theta(\action_t|\state_t)$. Both variants produce good results, and we compare them in Section~\ref{sec:experiments}. \subsection{Analysis of Prior Guided Policy Search Methods as Approximate Mirror Descent} The main distinction between the proposed method and prior guided policy search methods is that the constraint $D_\text{KL}(p_i(\tau) \| \bar{\pi}_{\theta i}(\tau)) \leq \epsilon$ is enforced on the local policies at each iteration, while in prior methods, this constraint is iteratively enforced via a dual descent procedure over multiple iterations. This means that the prior methods perform approximate mirror descent with step sizes that are adapted (by adjusting the Lagrange multipliers) but not constrained exactly. In our empirical evaluation, we show that our approach is somewhat more stable, though sometimes slower than these prior methods. This empirical observation agrees with our intuition: prior methods can sometimes be faster, because they do not exactly constrain the step size, but our method is simpler, requires less tuning, and always takes bounded steps on the global policy in trajectory space. \section{Analysis in the Nonlinear Case} Although the S-step under nonlinear dynamics is not an optimal projection onto the constraint manifold, we can bound the additional cost incurred by this projection in terms of the KL-divergence between $p_i(\action_t|\state_t)$ and $\pi_\theta(\action_t|\state_t)$. This analysis also reveals why prior guided policy search algorithms, which only have asymptotic convergence guarantees, still attain good performance in practice even after a small number of iterations. We will drop the subscript $i$ from $p_i(\action_t|\state_t)$ in this section for conciseness, though the same analysis can be repeated for multiple local policies $p_i(\action_t|\state_t)$. \subsection{Bounding the Global Policy Cost} \label{sec:bounding} The analysis in this section is based on the following lemma, which we prove in Appendix~\ref{app:distro_bound}, building off of earlier results by Ross et al.~\cite{rgb-rilsp-11} and Schulman et al.~\cite{slmja-trpo-15}: \begin{lemma} \label{lemma:distro_bound} Let $\epsilon_t = \max_{\state_t} D_\text{KL}(p(\action_t|\state_t) \| \pi_\theta(\action_t|\state_t)$. Then $D_\text{TV}(p(\state_t)\| \pi_\theta(\state_t)) \leq 2 \sum_{t=1}^T \sqrt{2\epsilon_t}$. \end{lemma} This means that if we can bound the KL-divergence between the policies, then the total variation divergence between their state marginals (given by $D_\text{TV}(p(\state_t)\| \pi_\theta(\state_t)) = \frac{1}{2}\|p(\state_t) - \pi_\theta(\state_t)\|_1$) will also be bounded. This bound allows us in turn to relate the total expected costs of the two policies to each other according to the following lemma, which we prove in Appendix~\ref{app:cost_bound}: \begin{lemma} \label{lemma:cost_bound} If $D_\text{TV}(p(\state_t)\| \pi_\theta(\state_t)) \leq 2 \sum_{t=1}^T \sqrt{2\epsilon_t}$, then we can bound the total cost of $\pi_\theta$ as \[ \sum_{t=1}^T E_{\pi_\theta(\state_t,\action_t)}[\ell(\state_t,\action_t)] \leq \sum_{t=1}^T \left[ E_{p(\state_t,\action_t)}[\ell(\state_t,\action_t)] + \sqrt{2\epsilon_t} \max_{\state_t, \action_t} \ell(\state_t, \action_t) + 2\sqrt{2\epsilon_t} Q_{\text{max},t} \right] \] where $Q_{\text{max},t} = \sum_{t^\prime=t}^T \max_{\mathbf{x}_{t^\prime},\mathbf{u}_{t^\prime}}\ell(\mathbf{x}_{t^\prime},\mathbf{u}_{t^\prime})$, the maximum total cost from time $t$ to $T$. \end{lemma} This bound on the cost of $\pi_\theta(\action_t|\state_t)$ tells us that if we update $p(\action_t|\state_t)$ so as to decrease its total cost or decrease its KL-divergence against $\pi_\theta(\action_t|\state_t)$, we will eventually reduce the cost of $\pi_\theta(\action_t|\state_t)$. For the MDGPS algorithm, this bound suggests that we can ensure improvement of the global policy within a small number of iterations by appropriately choosing the constraint $\epsilon$ during the C-step. Recall that the C-step constrains $\sum_{t=1}^T \epsilon_t \leq \epsilon$, so if we choose $\epsilon$ to be small enough, we can close the gap between the local and global policies. Optimizing the bound directly turns out to produce very slow learning in practice, because the bound is very loose. However, it tells us that we can either decrease $\epsilon$ toward the end of the optimization process or if we observe the global policy performing much worse than the local policies. We discuss how this idea can be put into action in the next section. \subsection{Step Size Selection} \label{sec:step} In prior work \cite{lwa-lnnpg-15}, the step size $\epsilon$ in the local policy optimization is adjusted by considering the difference between the predicted change in the cost of the local policy $p(\action_t|\state_t)$ under the fitted dynamics, and the actual cost obtained when sampling from that policy. The intuition is that, because the linearized dynamics are local, we incur a larger cost the further we deviate from the previous policy. We can adjust the step size by estimating the rate at which the additional cost is incurred and choose the optimal tradeoff. Let $\ell_{k-1}^{k-1}$ denote the expected cost under the previous local policy $\bar{p}(\action_t|\state_t)$, $\ell_{k-1}^k$ the cost under the current local policy $p(\action_t|\state_t)$ and the previous fitted dynamics (which were estimated using samples from $\bar{p}(\action_t|\state_t)$ and used to optimize $p(\action_t|\state_t)$), and $\ell_k^k$ the cost of the current local policy under the dynamics estimated using samples from $p(\action_t|\state_t)$ itself. Each of these can be computed analytically under the linearized dynamics. We can view the difference $\ell_k^k - \ell_{k-1}^k$ as the additional cost we incur from imperfect dynamics estimation. Previous work suggested modeling the change in cost as a function of $\epsilon$ as following: $\ell_k^k - \ell_{k-1}^{k-1} = a\epsilon^2 + b\epsilon$, where $b$ is the change in cost per unit of KL-divergence, and $a$ is additional cost incurred due to inaccurate dynamics \cite{lwa-lnnpg-15}. This model is reasonable because the integral of a quadratic cost under a linear-Gaussian system changes roughly linearly with KL-divergence. The additional cost due to dynamics errors is assumes to scale superlinearly, allowing us to solve for $b$ by looking at the difference $\ell_k^k - \ell_{k-1}^k$ and then solving for a new optimal $\epsilon^\prime$ according to $\epsilon^\prime = -b/2a$, resulting in the update $\epsilon^\prime = \epsilon (\ell_{k-1}^k - \ell_{k-1}^{k-1}) / 2(\ell_{k-1}^k - \ell_k^k)$. In MDGPS, we propose to use two step size adjustment rules. The first rule simply adapts the previous method to the case where we constrain the new local policy $p(\action_t|\state_t)$ against the global policy $\pi_\theta(\action_t|\state_t)$, instead of the previous local policy $\bar{p}(\action_t|\state_t)$. In this case, we simply replace $\ell_{k-1}^{k-1}$ with the expected cost under the previous global policy, given by $\ell_{k-1}^{k-1,\pi}$, obtained using its linearization $\bar{\pi}_\theta(\action_t|\state_t)$. We call this the ``classic'' step size: $\epsilon^\prime = \epsilon (\ell_{k-1}^{k} - \ell_{k-1}^{k-1,\pi}) / 2(\ell_{k-1}^k - \ell_k^k)$. However, we can also incorporate intuition from the bound in the previous section to obtain a more conservative step adjustment that reduces $\epsilon$ not only when the obtained local policy improvement doesn't meet expectations, but also when we detect that the global policy is unable to reproduce the behavior of the local policy. In this case, reducing $\epsilon$ reduces the KL-divergence between the global and local policies which, as shown in the previous section, tightens the bound on the global policy return. As mentioned previously, directly optimizing the bound tends to perform poorly because the bound is quite loose. However, if we estimate the cost of the global policy using its linearization, we can instead adjust the step size based on a simple model of \emph{global} policy cost. We use the same model for the change in cost, given by $\ell_k^{k,\pi} - \ell_{k-1}^{k-1,\pi} = a\epsilon^2 + b\epsilon$. However, for the term $\ell_k^k$, which reflects the actual cost of the new policy, we instead use the cost of the new global policy $\ell_k^{k,\pi}$, so that $a$ now models the additional loss due to \emph{both} inaccurate dynamics and inaccurate projection: if $\ell_k^{k,\pi}$ is much worse than $\ell_{k-1}^{k}$, then either the dynamics were too local, or S-step failed to match the performance of the local policies. In either case, we decrease the step size.\footnote{Although we showed before that the discrepancy depends on $\sum_{t=1}^T \sqrt{2\epsilon}_t$, here we use $\epsilon^2$. This is a simplification, but the net result is the same: when the global policy is worse than expected, $\epsilon$ is reduced.} As before, we can solve for the new step size $\epsilon^\prime$ according to $\epsilon^\prime = \epsilon (\ell_{k-1}^{k} - \ell_{k-1}^{k-1,\pi}) / 2(\ell_{k-1}^{k} - \ell_k^{k,\pi})$. We call this the ``global'' step size. Details of how each quantity in this equation is computed are provided in Appendix~\ref{app:step}. \section{Relation to Prior Work} \label{sec:prior_work} While we've discussed the connections between MDGPS and prior guided policy search methods, in this section we'll also discuss the connections between our method and other policy search methods. One popular supervised policy learning methods is DAGGER~\cite{rgb-rilsp-11}, which also trains the policy using supervised learning, but does not attempt to adapt the teacher to provide better training data. MDGPS removes the assumption in DAGGER that the supervised learning stage has bounded error against an arbitrary teacher policy. MDGPS does not need to make this assumption, since the teacher can be adapted to the limitations of the global policy learning. This is particularly important when the global policy has computational or observational limitations, such as when learning to use camera images for partially observed control tasks or, as shown in our evaluation, blind peg insertion. When we sample from the global policy $\pi_\theta(\action_t|\state_t)$, our method resembles policy gradient methods with KL-divergence constraints \cite{ps-rlmsp-08,pma-reps-10,slmja-trpo-15}. However, policy gradient methods update the policy $\pi_\theta(\action_t|\state_t)$ at each iteration by linearizing with respect to the policy parameters, which often requires small steps for complex, nonlinear policies, such as neural networks. In contrast, we linearize in the space of time-varying linear dynamics, while the policy is optimized at each iteration with many steps of supervised learning (e.g. stochastic gradient descent). This makes MDGPS much better suited for quickly and efficiently training highly nonlinear, high-dimensional policies. \section{Experimental Evaluation} \label{sec:experiments} We compare several variants of MDGPS and a prior guided policy search method based on Bregman ADMM (BADMM) \cite{levine2015end}. We evaluate all methods on one simulated robotic navigation task and two manipulation tasks. Guided policy search code, including BADMM and MDGPS methods, is available at \texttt{https://www.github.com/cbfinn/gps}. \begin{wrapfigure}{r}{0.25\textwidth} \vspace{-0.18in} \includegraphics[width=0.24\textwidth]{tasks.png} \vspace{-0.30in} \end{wrapfigure} \vspace{-0.12in} \paragraph{Obstacle Navigation.} In this task, a 2D point mass (grey) must navigate around obstacles to reach a target (shown in green), using velocities and positions relative to the target. We use $N=5$ initial states, with 5 samples per initial state per iteration. The target and obstacles are fixed, but the starting position varies. \vspace{-0.12in} \paragraph{Peg Insertion.} This task, which is more complex, requires controlling a 7 DoF 3D arm to insert a tight-fitting peg into a hole. The hole can be in different positions, and the state consists of joint angles, velocities, and end-effector positions relative to the target. This task is substantially more challenging physically. We use $N=9$ different hole positions, with 5 samples per initial state per iteration. \vspace{-0.12in} \paragraph{Blind Peg Insertion.} The last task is a blind variant of the peg insertion task, where the target-relative end effector positions are provided to the local policies, but not to the global policy $\pi_\theta(\action_t|\state_t)$. This requires the global policy to search for the hole, since no input to the global policy can distinguish between the different initial state $\mathbf{x}_1^i$. This makes it much more challenging to adapt the global and local policies to each other, and makes it impossible for the global learner to succeed without adaptation of the local policies. We use $N=4$ different hole positions, with 5 samples per initial state per iteration. The global policy for each task consists of a fully connected neural network with two hidden layers with $40$ rectified linear units. The same settings are used for MDGPS and the prior BADMM-based method, except for the difference in surrogate costs, constraints, and step size adjustment methods discussed in the paper. Results are presented in Figure~\ref{fig:results}. On the easier point mass and peg tasks, all of the methods achieve similar performance. However, the MDGPS methods are all substantially easier to apply to these tasks, since they have very few free hyperparameters. An initial step size must be selected, but the adaptive step size adjustment rules make this choice less important. In contrast, the BADMM method requires choosing an initial weight on the augmented Lagrangian term, an adjustment schedule for this term, a step size on the dual variables, and a step size for local policies, all of which have a substantial impact on the final performance of the method (the reported results are for the best setting of these parameters, identified with a hyperparameter sweep). On the harder blind peg task, MDGPS consistently outperforms BADMM when sampling from the local policies (``off policy''), with both the classic and global step sizes. This is particularly apparent in the success rates in Table~\ref{tbl:success}, which shows that the MDGPS policies succeed at actually inserting the peg into the hole more often and on more conditions. This suggests that our method is better able to improve global policies particularly in situations where informational or representational constraints make na\"{i}ve imitation of the local policies insufficient to solve the task. On-policy sampling tends to learn slower, since the approximate projection causes the global policy to lag behind the local policy in performance, but this method is still able to consistently improve the global policies. Sampling from the global policies may be desirable in practice, since the global policies can directly use observations at runtime instead of requiring access to the state \cite{levine2015end}. The global step size also tends to be more conservative, but produces more consistent and monotonic improvement. \begin{figure} \centering \includegraphics[width=\textwidth]{results.png} \caption{Results for MDGPS variants and BADMM on each task. MDGPS is tested with local policy (``off policy'') and global policy (``on policy'') sampling (see Section~\ref{sec:implementation}), and both the ``classic'' and ``global'' step sizes (see Section~\ref{sec:step}). The vertical axis for the obstacle task shows the average distance between the point mass and the target. The vertical axis for the peg tasks shows the average distance between the bottom of the peg and the hole. Distances above 0.1, which is the depth of the hole (shown as a dotted line) indicate failure. All experiments are repeated three times, with the average performance and standard deviation shown in the plots.} \label{fig:results} \vspace{-0.1in} \end{figure} \begin{table} {\footnotesize \begin{tabular}{| c | c | c | c | c | c | c |} \hline & Iteration &BADMM & Off Pol., Classic & Off Pol., Global & On Pol., Classic & On Pol., Global\\ \hline \multirow{4}{*}{\rotatebox{90}{\scriptsize{Peg}}} & 3 & 0.00~\% & 0.00~\% & 0.00~\% & 0.00~\% & 0.00~\% \\ & 6 & 51.85~\% & \textbf{62.96~\%} & 22.22~\% & 48.15~\% & 33.33~\% \\ & 9 & 51.85~\% & 77.78~\% & 74.07~\% & 77.78~\% & \textbf{81.48~\%} \\ & 12 & 77.78~\% & 70.73~\% & \textbf{92.59~\%} & \textbf{92.59~\%} & 85.19~\% \\ \hline \hline \multirow{4}{*}{\rotatebox{90}{\scriptsize{Blind Peg}}} & 3& 0.00~\% & 0.00~\% & 0.00~\% & 0.00~\% & 0.00~\% \\ & 6& 50.00~\% & \textbf{58.33~\%} & 25.00~\% & 33.33~\% & 25.00~\% \\ & 9& 58.33~\% & \textbf{75.00~\%} & 50.00~\% & 58.33~\% & 33.33~\% \\ & 12& 75.00~\% & 83.33~\% & \textbf{91.67~\%} & 58.33~\% & 58.33~\% \\ \hline \end{tabular} } \vspace{0.1in} \caption{ Success rates of each method on each peg insertion task. Success is defined as inserting the peg into the hole with a final distance of less than 0.06. Results are averaged over three runs.} \label{tbl:success} \vspace{-0.3in} \end{table} \section{Discussion and Future Work} \label{sec:discussion} We presented a new guided policy search method that corresponds to mirror descent under linearity and convexity assumptions, and showed how prior guided policy search methods can be seen as approximating mirror descent. We provide a bound on the return of the global policy in the nonlinear case, and argue that an appropriate step size can provide improvement of the global policy in this case also. Our analysis provides us with the intuition to design an automated step size adjustment rule, and we illustrate empirically that our method achieves good results on a complex simulated robotic manipulation task while requiring substantially less tuning and hyperparameter optimization than prior guided policy search methods. Manual tuning and hyperparameter searches are a major challenge across a range of deep reinforcement learning algorithms, and developing scalable policy search methods that are simple and reliable is vital to enable further progress. As discussed in Section~\ref{sec:prior_work}, MDGPS has interesting connections to other policy search methods. Like DAGGER \cite{rgb-rilsp-11}, MDGPS uses supervised learning to train the policy, but unlike DAGGER, MDGPS does not assume that the learner is able to reproduce an arbitrary teacher's behavior with bounded error, which makes it very appealing for tasks with partial observability or other limits on information, such as learning to use camera images for robotic manipulation \cite{levine2015end}. When sampling directly from the global policy, MDGPS also has close connections to policy gradient methods that take steps of fixed KL-divergence \cite{ps-rlmsp-08,slmja-trpo-15}, but with the steps taken in the space of trajectories rather than policy parameters, followed by a projection step. In future work, it would be interesting to explore this connection further, so as to develop new model-free policy gradient methods. \bibliographystyle{plain}
b2626f3873b25d41c2de31c91c507e7e8082993f
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Spatio--temporal models are widely used by practitioners. Explaining economic, environmental, social, or biological phenomena, such as peer influence, neighbourhood effects, contagion, epidemics, interdependent preferences, climate change, and so on, are only some of the interesting applications of such models. A widely used spatio--temporal model is the spatial dynamic panel data model (SDPD) proposed and analysed by \cite{LeeYu10a}. See \cite{LeeYu10b} for a survey. To improve adaptivity of SDPD models, \cite{DouAlt15} recently proposed a generalized model that assigns different coefficients to varied locations and assumes heteroskedastic and spatially correlated errors. The model is \begin{equation}\label{eqn1} {\mathbf y}_t = D(\boldsymbol{\lambda}_0){\mathbf W}{\mathbf y}_t + D({\boldsymbol{\lambda}_1}){\mathbf y}_{t-1} + D(\boldsymbol{\lambda}_2){\mathbf W}{\mathbf y}_{t-1} + \mbox{\boldmath$\varepsilon$}_t, \end{equation} where the vector ${\mathbf y}_t$ is of order $p$ and contains the observations at time $t$ from $p$ different locations; the errors $\mbox{\boldmath$\varepsilon$}_t$ are serially uncorrelated; the \emph{spatial matrix} ${\mathbf W}$ is a weight matrix with zero main diagonal and is assumed to be known; $D(\boldsymbol{\lambda}_j)$ with $j=0,1,2$ are diagonal matrices, and $\boldsymbol{\lambda}_j$ are the vectors with the spatial coefficients $\lambda_{ji}$ for $i=1,\ldots,p$. The \emph{generalized SDPD} model in (\ref{eqn1}) guarantees adaptivity by means of its $3p$ parameters. It is characterized by the sum of three terms: the \emph{spatial component}, driven by matrix ${\mathbf W}$ and the spatial parameter $\boldsymbol{\lambda}_0$; the \emph{dynamic component}, driven by the autoregressive parameter $\boldsymbol{\lambda}_1$; and the \emph{spatial--dynamic component}, driven by matrix ${\mathbf W}$ and the spatial--autoregressive parameter $\boldsymbol{\lambda}_2$. If the vectors $\boldsymbol{\lambda}_j$ are scalars for all $j$, then model (\ref{eqn1}) reduces to the classic SDPD of \cite{LeeYu10a}. The errors $\mbox{\boldmath$\varepsilon$}_t$ in model (\ref{eqn1}) are serially uncorrelated and may show heteroskedasticity and cross-correlation over space, so that $\mathop{var}(\mbox{\boldmath$\varepsilon$}_t)$ is a full matrix. This is a novelty compared with the \emph{SDPD} model of \cite{LeeYu10a}, where the errors must be cross-uncorrelated and homoskedastic in order to get consistency of the estimators. A setup similar to ours for the errors has been also considered by \cite{KelPru10} and \cite{Su12}, but not for panel models. However, their estimators are generally based on the instrumental variables technique, in order to overcome the endogeneity of the \emph{zero-lag} regressor. For the \emph{generalized SDPD} model, instead, \cite{DouAlt15} propose a new estimation procedure based on a generalized Yule--Walker approach. They show the consistency of the estimators under regularity assumptions. They also derive the convergence rate and the conditions under which the estimation procedure does not suffer for high-dimensional setups, notwithstanding the large number of parameters to be estimated (which become infinite with the dimension $p$). In real data applications, it is important to check the validity of the assumptions required for the consistency of the estimation procedure. See, for example, the assumptions and asymptotic analysis in \cite{LeeYu10a} and \cite{DouAlt15} as well as the references therein. Checking such assumptions on real data is often not easy; at times, they are clearly violated. Moreover, the spatial matrix ${\mathbf W}$ is assumed to be known, but in many cases, this is not true, and it must be estimated. For example, the spatial weights can be associated with ``similarities'' between spatial units and measured by estimated correlations. Another example is when the spatial weights are zeroes/ones, depending on the ``relationships'' between the spatial units, but the neighbourhood structure of ${\mathbf W}$ is unknown (\emph{i.e.}, it is not known where the ones must be allocated). In such cases, the performance of the \emph{SDPD} models needs to be investigated. Readers are advised to refer to recent papers on spatial matrix estimation (see, among others, \cite{LamSou16}). Motivated by the above considerations, we propose a new version of the \emph{ SDPD} model obtained by adding a constraint to the spatial parameters of the \emph{generalized SDPD} of \cite{DouAlt15}. New estimators of the parameters are proposed and investigated theoretically and empirically. The new model is called \emph{stationary SDPD} and has several advantages. First, the structure of the model and the interpretation of the parameters are simpler than the \emph{generalized SDPD} model, with the consequence that the assumptions underlying the theoretical results are clearer and can be checked easily with real data. Moreover, the estimation procedure is fast and simple to implement. Second, the proposed estimators of the parameters are always unbiased and reach the $\sqrt{T}$ convergence rate (where $T$ is the temporal length of the time series) even in the high-dimensional case, although the number of parameters tends to infinity with the dimension $p$. Last, but not least, our model allows wider application than the classic \emph{SDPD} model, and it is general enough to represent a wide range of multivariate linear processes that can be implicitly interpreted (when they are not explicitly interpretable) as spatio--temporal processes, with respect to a ``latent spatial matrix,'' which needs to be estimated. A big implication of this is that our model is not necessarily confined to the representation of strict spatio--temporal processes (where the spatial matrix is known), but it can also be considered as a valid alternative to the general \emph{VAR} models (where there is no spatial matrix), with two relevant advantages: i) more efficient estimation of the model and ii) the possibility of estimating the model even when $p>T$, thus avoiding the \emph{curse of dimensionality} that characterizes the \emph{VAR} models. Surprisingly, the simulation results show the remarkably better performance of our model and the new estimation procedure compared with the standard VAR model and the standard estimation procedure, even when the spatial matrix is latent and, therefore, to be estimated (see section \ref{matrixA}). The rest of the paper is organized as follows. Section \ref{sdpd} presents the new model and discusses the issue of identifiability. Section \ref{est_alg} describes the estimation procedure. The theoretical results are shown in section \ref{asymptotic}. The empirical performance of the estimation procedure is investigated in section \ref{simulazioni}. Finally, all the proofs are provided in the Appendix. \section{A constrained spatio--temporal model: the stationary SDPD}\label{sdpd} In the sequel, we assume that ${\mathbf y}_1, \cdots, {\mathbf y}_T$ are the observations from a stationary process defined by (\ref{eqn3}). The transpose of a matrix ${\mathbf A}$ is denoted with ${\mathbf A}^T$. We assume that the process has mean zero and denote with $\boldsymbol{\Sigma}_j=\mathop{cov}({\mathbf y}_t,{\mathbf y}_{t-j})=E({\mathbf y}_t{\mathbf y}_{t-j}^T)$ the covariance matrix of the process at the lag $j$. The \emph{generalized SDPD} model in (\ref{eqn1}) can be reformulated as follows. \begin{equation} \left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]{\mathbf y}_t = D({\boldsymbol{\lambda}_1})\left[{\mathbf I}_p - D(\boldsymbol{\lambda}^+_2){\mathbf W}\right]{\mathbf y}_{t-1} + \mbox{\boldmath$\varepsilon$}_t, \label{eqn3} \end{equation} where $\boldsymbol{\lambda}^+_2$ is a vector obtained by dividing the elements of $\boldsymbol{\lambda}_2$ by the corresponding elements of $\boldsymbol{\lambda}_1$ (assuming, for now, that all the coefficients in $\boldsymbol{\lambda}_1$ are different from zero). Note that model (\ref{eqn3}) is equivalent to a multivariate (auto)regression between a linear combination of ${\mathbf y}_t$ and a linear combination of the lag ${\mathbf y}_{t-1}$, where the weights of the two linear combinations depend on ${\mathbf W}$ and the coefficients $\boldsymbol{\lambda}_0$ and $\boldsymbol{\lambda}^+_2$, respectively. \begin{equation}\label{zeta} {\mathbf z}_t^{(\boldsymbol{\lambda}_0,{\mathbf W})} = D({\boldsymbol{\lambda}_1}){\mathbf z}_{t-1}^{(\boldsymbol{\lambda}^+_2,{\mathbf W})} + \mbox{\boldmath$\varepsilon$}_t. \end{equation} Some special cases may arise from model (\ref{eqn3}) by adding restrictions on the parameters $\boldsymbol{\lambda}_{j}$. First, if we assume that the spatial parameters are constant over space, that is, $\boldsymbol{\lambda}_{j}$ is scalar for $j=0,1,2$, then we obtain the classic \emph{SDPD} model of \cite{LeeYu10a}. Another constrained model, proposed and analysed in this paper, may be derived by assuming that $\boldsymbol{\lambda}_0=\boldsymbol{\lambda}^+_2$. The reason underlying the choice of this constraint is a generalized assumption of stationarity. In time series analysis, stationarity means that the dependence structure of the process is constant (in some sense) over time. In particular, second-order stationarity assumes that correlations between the observations $({\mathbf y}_t,{\mathbf y}_{t-j})$ depend on the lag $j$ but not on $t$, implying that $\mathop{var}({\mathbf y}_t)$ is constant for all $t$. However, in spatio--temporal time series, there are two kinds of correlations: \emph{temporal correlations}, involving observations at different time points, and \emph{spatial correlations}, involving observations at different spatial units. As we refer to stationarity, it makes sense to assume that spatial correlations are also time-invariant, which means that the weights in (\ref{zeta}) must not change over time, thus, $\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]=\left[{\mathbf I}_p-D(\boldsymbol{\lambda}^+_2){\mathbf W}\right]$, also implying that $\mathop{var}({\mathbf z}_t)$ is the same for all $t$. Therefore, we add the constraint $\boldsymbol{\lambda}_0=\boldsymbol{\lambda}^+_2$, and the model becomes \begin{equation} \label{b1} \left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]{\mathbf y}_t = D(\boldsymbol{\lambda}_1)\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]{\mathbf y}_{t-1} + \mbox{\boldmath$\varepsilon$}_t. \end{equation} We denote the model as \emph{stationary SDPD}. Model (\ref{b1}) has several advantages that will be shown in the following sections. Above all, imposing spatio--temporal stationarity helps gain efficiency while still preserving the spatial adaptability that characterizes the \emph{generalized SDPD} model of \cite{DouAlt15}. Moreover, our model allows representation of a wide range of multivariate processes by means of a simple model subject to few assumptions that can be easily checked using real data. Finally, it is worthwhile to stress the difference between the \emph{SDPD} model of \cite{LeeYu10a}, the \emph{generalized SDPD} model of \cite{DouAlt15}, and the \emph{stationary SDPD} model proposed here. The first model imposes that the spatial relationships be the same for all units, since the coefficients $\lambda_j$ (with $j=0,1,2$) do not change with $i=1,\ldots,p$. Instead, the \emph{stationary SDPD} model in (\ref{b1}) allows varied coefficients for different spatial units, as in the \emph{generalized SDPD} of \cite{DouAlt15}, but they are assumed to be time-invariant thanks to a constraint on the time-lagged parameters. Of course, the estimation procedures vary for the three cases in terms of the convergence rates. The constrained model underlying our \emph{stationary SDPD} allows the estimation procedure to reach the $\sqrt{T}$ convergence rate and to guarantee unbiased estimators, whatever the dimension $p$ and even when $p\rightarrow\infty$ at any rate. This is a big improvement with respect to the other two models. In fact, for the classic \emph{SDPD} model, the estimators are characterized by a $\sqrt{Tp}$ convergence rate (which is faster than that of our model, since they have only three parameters to estimate instead of $2p$), but a bias of order $T^{-1}$ exists, and it does not vanish when $p/T\rightarrow\infty$ (see Theorem 3 of \cite{LeeYu10a}). On the other hand, the convergence rates of the estimators in the \emph{generalized SDPD} model are slower than those of our model and deteriorate when $p\rightarrow\infty$ at a rate faster than $\sqrt{T}$ (see Theorem 2 of \cite{DouAlt15}). \subsection{Identification of parameters in the case of cross-uncorrelated errors} In this section, we assume, for simplicity, that the matrix $\boldsymbol{\Sigma}_0^{\varepsilon}=\mathop{var}(\mbox{\boldmath$\varepsilon$}_t)$ is diagonal (\emph{i.e.}, there is heteroskedasticity but no cross-correlation in the error process) and discuss the identifiability of the model. In the next section, we generalize the problem by also adding some cross-correlations in the error process. Defining ${\mathbf z}_t^{(0)}=\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]{\mathbf y}_t$, model (\ref{b1}) can be reformulated as \begin{eqnarray} {\mathbf z}_t^{(0)} &=& D(\boldsymbol{\lambda}_1){\mathbf z}_{t-1}^{(0)} + \mbox{\boldmath$\varepsilon$}_t, \label{b1ter} \end{eqnarray} which is a transformed \emph{VAR} process with uncorrelated components, since $D(\boldsymbol{\lambda}_1)$ is diagonal. Given that we assume that $\boldsymbol{\Sigma}_0^{\varepsilon}$ is also diagonal, the coefficients $\lambda_{1i}$ for $i=1,\ldots,p$, represent the slopes of $p$ univariate autoregressive models with respect to the latent variables $z_{it}^{(0)}$. Therefore, $\lambda_{1i}\equiv\mathop{cor}(z_{it}^{(0)}, z_{i,t-1}^{(0)})$. From (\ref{b1}), it follows that \begin{equation} \left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]\boldsymbol{\Sigma}_1 = D(\boldsymbol{\lambda}_1)\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]\boldsymbol{\Sigma}_0 \label{seconda}\\ \end{equation} and for the $i$-th equation, \begin{equation}\label{vincolo} ({\mathbf e}^T_i-\lambda_{0i}{\mathbf w}^T_i)\boldsymbol{\Sigma}_{1}=\lambda_{1i}({\mathbf e}^T_i-\lambda_{0i}{\mathbf w}^T_i)\boldsymbol{\Sigma}_0, \end{equation} where ${\mathbf e}_i$ is the column vector with its $i$-th component equal to one and all the others equal to zero, while ${\mathbf w}_i$ is the column vector containing the $i$-th row of matrix ${\mathbf W}$. Under general assumptions, (\ref{vincolo}) admits only one solution with respect to $\lambda_{0i}$ and $\lambda_{1i}$ (see Theorem \ref{theorem1}), which can be found among the extreme points of $\lambda_{1i}=\mathop{cor}(z_{it}^{(0)}, z_{i,t-1}^{(0)})$ as a function of $\lambda_{0i}$. To provide insight into this, the first two plots of figure \ref{figure2} show two examples based on model 1 used in the simulation study. Denote with ($\lambda_{0i}^*,\lambda_{1i}^*)$ the true values of the coefficients used in model 1 for a given location $i$ (in particular, in figure \ref{figure2}, the first two plots refer to locations $i=6$ and $i=8$). The solid line shows $\lambda_{1i}=\mathop{cor}(z_{it}^{(0)}, z_{i,t-1}^{(0)})$ as a function of $\lambda_{0i}$. The two dots show the points of this function where the first derivative is zero. The vertical and horizontal dashed lines identify which one of the two points satisfies the sufficient condition in (\ref{vincolo}). As expected, it coincides with the true values ($\lambda_{0i}^*,\lambda_{1i}^*)$ used to generate the time series. Theorem \ref{theorem1}, shown in the Appendix, formalizes this result. \begin{theorem}\label{theorem1} Consider model (\ref{b1}) for a stationary process ${\mathbf y}_t$ with mean zero, and assume that the error process $\mbox{\boldmath$\varepsilon$}_t$ is such that $\boldsymbol{\Sigma}^0_\varepsilon=\mathop{var}(\mbox{\boldmath$\varepsilon$}_t)$ is diagonal (\emph{i.e.}, there is heteroskedasticity but no cross-correlation in the errors). Under assumptions $A1-A4$ in section \ref{asymptotic}, the following results hold: \begin{enumerate} \item There exist a unique couple of values $(\lambda_{0i}^*,\lambda_{1i}^*)$ satisfying the following system of equations: \begin{equation}\label{vinc} ({\mathbf e}^T_i-\lambda_{0i}{\mathbf w}^T_i)\boldsymbol{\Sigma}_{1}-\lambda_{1i}({\mathbf e}^T_i-\lambda_{0i}{\mathbf w}^T_i)\boldsymbol{\Sigma}_0={\bf 0}^T, \qquad\qquad i=1,\ldots,p, \end{equation} where ${\mathbf e}_i$ is the $i$-th unit vector, and ${\mathbf w}_i$ contains the $i$-th row of the spatial matrix ${\mathbf W}$. \item Such a point, $(\lambda_{0i}^*,\lambda_{1i}^*)$, is also the solution of the following second-order polynomial equation: \begin{eqnarray} \label{nec_cond} \left.\frac{\partial\mathop{cov}(z_{it}^{(0)}, z_{i,t-1}^{(0)})}{\partial\lambda_{0i}}\right|_{\lambda_{0i}=\lambda_{0i}^*}- \lambda_{1i}^*\left.\frac{\partial\mathop{var}(z_{i,t-1}^{(0)})}{\partial\lambda_{0i}}\right|_{\lambda_{0i}=\lambda_{0i}^*} &=& 0. \end{eqnarray} \end{enumerate} \end{theorem} \noindent\textbf{Remark 1:} Theorem \ref{theorem1} not only shows that the \emph{stationary SDPD} model is well identified, because there is a unique solution for $(\boldsymbol{\lambda}_0,\boldsymbol{\lambda}_1)$, but it also suggests a way to estimate such parameters. In fact, we can find all the solutions to equation (\ref{nec_cond}) and then check which one satisfies the sufficient condition in (\ref{vincolo}). This estimation procedure is described in section \ref{est_alg}. \subsection{Identification of parameters in the case of cross-correlated errors} Now, we relax the assumption on the error $\mbox{\boldmath$\varepsilon$}_t$ by letting $\boldsymbol{\Sigma}_0^\varepsilon$ be a full matrix (i.e., there is heteroskedasticity and cross-correlation in the error process). This setup allows the process ${\mathbf y}_t$ to include some {spurious cross-correlation} not explained by ${\mathbf W}$. In this case, the coefficients $\lambda_{i1}$ still give the correlations between the latent variables $z_{i,t}^{(0)}$ and $z_{i,t-1}^{(0)}$, but now, the $p$ equations in model (\ref{b1ter}) are correlated. The main consequence of this is that the true values $(\lambda_{0i}^*, \lambda_{1i}^*)$ do not identify an extreme point of the correlation function (see case $i=2$ in figure \ref{figure2}). Anyway, the sufficient condition in (\ref{vincolo}) is still valid, and the true coefficients $(\lambda_{0i}^*, \lambda_{1i}^*)$ can be identified by introducing a ``constrained'' condition. \begin{theorem}\label{theorem1bis} Consider model (\ref{b1}) for a stationary process ${\mathbf y}_t$ with mean zero, and assume that the error process $\mbox{\boldmath$\varepsilon$}_t$ is such that $\boldsymbol{\Sigma}^0_\varepsilon=\mathop{var}(\mbox{\boldmath$\varepsilon$}_t)$ is a full matrix (\emph{i.e.}, there is heteroskedasticity and cross-correlation in the errors). Under assumptions $A1-A4$ in section \ref{asymptotic}, the following results hold: \begin{enumerate} \item There exist a unique couple of values $(\lambda_{0i}^*,\lambda_{1i}^*)$ satisfying the following system of equations \[ ({\mathbf e}^T_i-\lambda_{0i}{\mathbf w}^T_i)\boldsymbol{\Sigma}_{1}-\lambda_{1i}({\mathbf e}^T_i-\lambda_{0i}{\mathbf w}^T_i)\boldsymbol{\Sigma}_0={\bf 0}^T, \qquad\qquad i=1,\ldots,p, \] where ${\mathbf e}_i$ is the $i$-th unit vector, and ${\mathbf w}_i$ contains the $i$-th row of the spatial matrix ${\mathbf W}$; \item such a point, $(\lambda_{0i}^*,\lambda_{1i}^*)$, is also the solution of the following second-order polynomial equation. \begin{eqnarray*} \left.\frac{\partial\mathop{cov}(z_{it}^{(0)}, z_{i,t-1}^{(0)})}{\partial\lambda_{0i}}\right|_{\lambda_{0i}=\lambda_{0i}^*}- \lambda_{1i}^*\left.\frac{\partial\mathop{var}(z_{i,t-1}^{(0)})}{\partial\lambda_{0i}}\right|_{\lambda_{0i}=\lambda_{0i}^*} &=& ({\mathbf e}_i^T-\lambda_{0i}^*{\mathbf w}^T_i)(\boldsymbol{\Sigma}_1^T-\boldsymbol{\Sigma}_1){\mathbf w}_i. \end{eqnarray*} \end{enumerate} \end{theorem} \noindent\textbf{Remark 2:} When the errors $\mbox{\boldmath$\varepsilon$}_t$ are not cross-correlated, the matrix $\boldsymbol{\Sigma}_1$ is symmetric by Lemma \ref{lemma1} in the Appendix, so that point 2 in Theorem \ref{theorem1bis} becomes the same as in Theorem \ref{theorem1}. Therefore, Theorem \ref{theorem1bis} includes Theorem \ref{theorem1} as a special case. \section{Estimation procedure}\label{est_alg} We present here a simple algorithm for the estimation of the parameters $(\lambda_{0i},\lambda_{1i})$ for $i=1,\ldots,p$. First, estimate the matrices $\boldsymbol{\Sigma}_1$ and $\boldsymbol{\Sigma}_0$ through some consistent estimators $\hat\boldsymbol{\Sigma}_0$ and $\hat\boldsymbol{\Sigma}_1$. For example, $\hat\boldsymbol{\Sigma}_0=(n-1)^{-1}{\mathbf Y}_0{\mathbf Y}_0^T$ and $\hat\boldsymbol{\Sigma}_1=(n-2)^{-1}{\mathbf Y}_0{\mathbf Y}_1^T$, where ${\mathbf Y}_l=({\mathbf y}_{1+l}, \cdots, {\mathbf y}_{n-l})$. Alternatively, the threshold estimator analyzed in \cite{CheAlt13} can be considered in the high dimensional setup. Then, for each location $i=1,\ldots,p$, implement the following steps. \begin{enumerate} \item Define ${\mathbf e}_i$ as the $i$-th unit vector and ${\mathbf w}^T_i={\mathbf e}^T_i{\mathbf W}$, then compute: \begin{eqnarray*} \hat a_{0i} &=& {\mathbf e}_i^T\hat\boldsymbol{\Sigma}_0{\mathbf e}_i, \quad \hat a_{1i} = {\mathbf e}_i^T\hat\boldsymbol{\Sigma}_1 {\mathbf e}_i, \quad \hat a_{2i} = {\mathbf e}_i^T(\hat\boldsymbol{\Sigma}_1^T-\hat\boldsymbol{\Sigma}_1){\mathbf w}_i, \\ \hat b_{0i} &=& -2{\mathbf e}_i^T\hat\boldsymbol{\Sigma}_0{\mathbf w}_i, \quad \hat b_{1i} = -{\mathbf e}^T_i(\hat\boldsymbol{\Sigma}_1+\hat\boldsymbol{\Sigma}_1^T){\mathbf w}_i, \\ \hat c_{0i} &=& {\mathbf w}^T_i\hat\boldsymbol{\Sigma}_0{\mathbf w}_i,\quad \hat c_{1i} = {\mathbf w}^T_i\hat\boldsymbol{\Sigma}_1{\mathbf w}_i. \end{eqnarray*} \item Find the two roots $\lambda_{0i}^{(1)}$ and $\lambda_{0i}^{(2)}$ of the following two-order polynomial equation. \begin{equation}\label{eqq} \hat t_{0i} + \hat t_{1i}\lambda_{i0} + \hat t_{2i}\lambda_{i0}^2 =0, \end{equation} where $\hat t_{0i} = \hat b_{1i}\hat a_{0i}-\hat b_{0i}\hat a_{1i}+\hat a_{0i}\hat a_{2i}$, $\hat t_{1i} = 2(\hat a_{0i}\hat c_{1i}-\hat c_{0i}\hat a_{1i})+\hat a_{2i}\hat b_{0i}$, and $\hat t_{2i} = \hat c_{1i}\hat b_{0i}-\hat c_{0i}\hat b_{1i}+\hat a_{2i}\hat c_{0i}$. \item Estimate $\lambda_{0i}$ and $\lambda_{1i}$ by \begin{eqnarray}\label{stimatore0} \hat\lambda_{0i}&=&\arg\min_{j=1,2}{\mathbf v}_{ij}^T{\mathbf v}_{ij}, \\ \hat\lambda_{1i}&=&\frac{({\mathbf e}_i^T-\hat\lambda_{0i}{\mathbf w}^T_i)\hat\boldsymbol{\Sigma}_1({\mathbf e}_i-\hat\lambda_{0i}{\mathbf w}_i)}{({\mathbf e}_i^T-\hat\lambda_{0i}{\mathbf w}^T_i)\hat\boldsymbol{\Sigma}_0({\mathbf e}_i-\hat\lambda_{0i}{\mathbf w}_i)}, \label{stimatore1} \end{eqnarray} where $ {\mathbf v}^T_{ij} = ({\mathbf e}_i^T-\lambda_{0i}^{(j)}{\mathbf w}^T_i)\hat\boldsymbol{\Sigma}_1-\lambda_{1i}^{(j)}({\mathbf e}_i^T-\lambda_{0i}^{(j)}{\mathbf w}^T_i)\hat\boldsymbol{\Sigma}_0, $, and $\lambda_{1i}^{(j)} = ({\mathbf e}_i^T-\lambda_{0i}^{(j)}{\mathbf w}^T_i)\hat\boldsymbol{\Sigma}_1({\mathbf e}_i-\lambda_{0i}^{(j)}{\mathbf w}_i)/({\mathbf e}_i^T-\lambda_{0i}^{(j)}{\mathbf w}^T_i)\hat\boldsymbol{\Sigma}_0({\mathbf e}_i-\lambda_{0i}^{(j)}{\mathbf w}_i)$. \end{enumerate} \vspace{10pt}\noindent\textbf{Remark 3:} Assumption $A2$ in section \ref{asymptotic} guarantees that matrix $\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]$ has full rank. However, the above estimation procedure may suffer for some locations if matrix $\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]$ is near singularity. Such a case may come about because of the presence of some almost linearly dependent rows in the matrix, which may cause the quantity ${\mathbf w}_i^T[\boldsymbol{\Sigma}_1-\lambda_{1i}\boldsymbol{\Sigma}_0]{\mathbf w}_i$ to be almost zero for those rows (see Lemma \ref{lemma2}). As a result, the procedure loose efficiency for the estimation of $\lambda_{i0}$ for those locations (but it still works for $\lambda_{1i}$). Something similar may happen if there are some (almost) zero rows in ${\mathbf W}$, which is excluded by assumption $A4$. Anyway, it is worthwhile to stress that the estimation procedure works efficiently for all the other locations. In fact, the procedure does not require the inversion of matrix $\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]$, so it is able to isolate and separate the effects of ``collinear'' locations (or uncorrelated locations) from the other locations and to guarantee consistent and efficient estimations for the last locations. \section{Theoretical results}\label{asymptotic} In this section, we show the theoretical foundations of our proposal. In particular, we present the assumptions and show the consistency and the asymptotic normality of the estimators, for the cases of finite dimension and high dimension. Moreover, we show that the \emph{stationary SDPD} model can be used to represent a wide range of multivariate linear processes with respect to a ``latent spatial matrix,'' and therefore, it is of wider application than classic spatio--temporal contexts. The reduced form of model (\ref{b1}) is \begin{equation}\label{b1bis} {\mathbf y}_t={\mathbf A}^*{\mathbf y}_{t-1}+\mbox{\boldmath$\varepsilon$}_t^*, \end{equation} where $\mbox{\boldmath$\varepsilon$}_t^*=\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]^{-1}\mbox{\boldmath$\varepsilon$}_t$ and \begin{equation}\label{diagonalize} {\mathbf A}^*=\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]^{-1}D(\boldsymbol{\lambda}_1)\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]. \end{equation} Note that the errors $\mbox{\boldmath$\varepsilon$}_t^*$ have mean zero and are serially uncorrelated. Model (\ref{b1bis}) has a \emph{VAR} representation, so it is stationary when all the eigenvalues of matrix ${\mathbf A}^*$ are smaller than one in absolute value. From (\ref{diagonalize}), we can note that $\boldsymbol{\lambda}_1$ contains the eigenvalues of ${\mathbf A}^*$ whereas $\boldsymbol{\lambda}_0$ only affects its eigenvectors (see the proof of Theorem \ref{theorem1}). Therefore, we must consider the following assumptions: \begin{itemize} \item[A1)] $\lambda_{1i}\in\mathbb{R}$ and $|\lambda_{1i}|<1$, for all $i$, and vector $\boldsymbol{\lambda}_1$ is not scalar; \item[A2)] $\lambda_{0i}\in\mathbb{R}$ for all $i$ and vector $\boldsymbol{\lambda}_0$ is such that matrix $\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]$ has full rank; \item[A3)] the errors $\varepsilon_{it}$ are serially independent and such that $E(\varepsilon_{it})=0$ and $E|\varepsilon_{it}|^\delta<\infty$ for all $i,t$, for some $\delta>4$; \item[A4)] the spatial matrix ${\mathbf W}$ is nonsingular and has zero main diagonal. \end{itemize} Assumption $A1$ implies stationarity. Moreover, it guarantees that there are at least two distinct values in vector $\boldsymbol{\lambda}_1$ so that model (\ref{b1}) is identifiable, as shown in Theorem \ref{theorem1}. Assumption $A2$ is clearly necessary to assure that matrix $\left[{\mathbf I}_p-D(\boldsymbol{\lambda}_0){\mathbf W}\right]$ can be inverted so that the reduced model in (\ref{b1bis}) is well defined (Remark 3 indicates what happens when this assumption is not satisfied). Incidentally, it is worthwhile to note that our setup automatically solves the problem concerning the parameter space of $\boldsymbol{\lambda}_0$, highlighted at the end of section 2.2 by \cite{KelPru10}. So, in the empirical applications of our model, it is possible to use any kind of normalization for ${\mathbf W}$ (\emph{i.e.}, row-factor normalization or single-factor normalization), since the vector $\boldsymbol{\lambda}_0$ would automatically rescale accordingly (see section \ref{high} for more details on this aspect). This means that the coefficients $\lambda_{0i}$ can also take values outside the classic interval $[-1,1]$. Assumption $A3$ assures that the results in \cite{Han76} can be applied to show the asymptotic normality of the estimators. Assumption $A4$ is classic in spatio--temporal models and guarantees that the model is well defined and identifiable with respect to all the parameters, also for $p\rightarrow\infty$ (see Lemma \ref{lemma2} and Theorem \ref{theorem4}). Under assumptions $A1-A4$, it is immediately evident that the estimators $\hat\boldsymbol{\lambda}_{0}$ and $\hat\boldsymbol{\lambda}_{1}$, presented in section \ref{est_alg}, are both consistent following Theorem 11.2.1 in \cite{BroDav86}. For asymptotic normality, the following theorem can be stated. \begin{theorem}\label{theorem3} Consider $\hat\lambda_{0i}$ and $\hat\lambda_{1i}$, the estimators obtained by the algorithm in section \ref{est_alg}. Under assumptions $A1-A4$, we have for finite $p$ \begin{equation} \sqrt{T}(\hat\lambda_{ji}-\lambda_{ji}) \stackrel{d}{\longrightarrow}N(0,{\mathbf D}_{ji}^T{\mathbf V}_{ji}{\mathbf D}_{ji}) \nonumber\qquad\qquad j=0,1;\quad i=1,\ldots,p, \end{equation} where ${\mathbf D}_{ji}$ are the $K_i\times 1$ vectors, and ${\mathbf V}_{ji}$ are the matrices of order $K_i$ with $K_i \le 2p^2$ (see the proof). \end{theorem} Note that the estimators $\widehat\lambda_{ji}$ are unbiased for all $i,j$ and for all $p$. In the high dimension, we have infinite parameters to estimate ($2p$ in total, where $p\rightarrow\infty$). Therefore, we must assure that the consistency of the estimators is still valid in such a case. As expected, the properties of matrix ${\mathbf W}$ influence the consistency and the convergence rates of the estimators $\hat\lambda_{ij}$ when $p\rightarrow\infty$. For example, denote with $k_i$ the number of nonzero elements in vector ${\mathbf w}_i$. If $k_i=O(1)$ as $p\rightarrow\infty$, for all $i$, then the effective dimension of model (\ref{b1}) is finite and Theorem \ref{theorem3} can still be applied for the consistency and the asymptotic normality of the estimators $\hat\lambda_{ji}$, even if $p\rightarrow\infty$. The following Theorem \ref{theorem4}, instead, shows the consistency of the estimators under more general vectors ${\mathbf w}_i$, with $k_i\rightarrow\infty$ as $p\rightarrow\infty$. \subsection{Asymptotics for high dimensional setups}\label{high} In model (\ref{b1}), the spatial correlation between a given location $i$-th and the other locations is summarized by $\lambda_{0i}{\mathbf w}_i$. If the vector ${\mathbf w}_i$ is rescaled by a factor $\delta_i$, then we can have an equivalent model by rescaling the spatial coefficient $\lambda_{0i}$ by the inverse of the same factor, since $\lambda_{0i}{\mathbf w}_i=\delta_i^{-1}\lambda_{0i}{\mathbf w}_i\delta_i=\lambda_{0i,\delta}{\mathbf w}_{i,\delta}$. In such a way, we may consider irrelevant a row-normalization of matrix ${\mathbf W}$ if we let the coefficients in $D(\boldsymbol{\lambda}_{0})$ rescale accordingly. Such an approach is not new and follows the idea of \cite{KelPru10}. We use this approach here in order to simplify the analysis and the interpretation of the \emph{stationary SDPD} model in the high dimensional setup. In fact, when $p\rightarrow\infty$ and $k_i=O(p)$, the vectors ${\mathbf w}_i$ may change with $p$ and this may have an influence on the scale order of the process. This happens, for example, if we consider a row-normalized spatial matrix ${\mathbf W}$, since the weights become infinitely small for infinitely large $p$. Looking at the (\ref{diagonalize}), model (\ref{b1}) appears to become spatially uncorrelated for $p\rightarrow\infty$ because matrix ${\mathbf W}$ tends to be asymptotically diagonal (for $p\rightarrow\infty$ and $T$ given). As a consequence, the model appears to become not identifiable in the high dimension with respect to the parameters $\lambda_{0i}$. To avoid this, we assume here that also the coefficients $\lambda_{0i}$ may depend on the dimension $p$, borrowing the idea of \cite{KelPru10}. In such a way, we can derive the conditions for the identifiability of the model in the high dimension and better convergence rates for the estimators. This is shown by the following theorem. \begin{theorem}\label{theorem4} Consider $\hat\lambda_{0i}$ and $\hat\lambda_{1i}$, the estimators obtained by the algorithm in section \ref{est_alg}. Assume that the number of nonzero values in ${\mathbf w}_i$ is $k_i=O(p)$ for all $i=1,\ldots,p$. Under assumptions $A1-A4$, for $p\rightarrow\infty$ we have the following cases: \begin{itemize} \item[(i)] if the vectors ${\mathbf w}_i$ are normalized by $L_1$ norm then \[ \left|\hat\lambda_{ji}-\lambda_{ji}\right| =O_p(T^{-1/2}) \nonumber\qquad\qquad {\rm for\ } j=0,1;i=1,\ldots,p, \] provided that $\lambda_{0i}=O(p)$; \item[(ii)] if the vectors ${\mathbf w}_i$ are normalized by $L_2$ norm and $\lambda_{0i}=O(1)$ then \[ \left|\hat\lambda_{ji}-\lambda_{ji}\right| =O_p(T^{-1/2}) \nonumber\qquad\qquad {\rm for\ } j=0,1;i=1,\ldots,p; \] \item[(iii)] for generic (not normalized but bounded) vectors ${\mathbf w}_i$ and $\lambda_{0i}=O(1)$ we have \[ \left|\hat\lambda_{ji}-\lambda_{ji}\right| =O_p(pT^{-1/2}) \nonumber\qquad\qquad {\rm for\ } j=0,1;i=1,\ldots,p. \] \end{itemize} \end{theorem} As shown by Theorem \ref{theorem4}, cases (i) and (ii), if we consider a row-normalized spatial matrix ${\mathbf W}$, our estimation procedure is consistent for any value of $p$ and with $p\rightarrow\infty$ at any rate. In other words, the convergence rate is not affected by the dimension $p$. However, there are some differences between the two cases of $L_1$ and $L_2$ normalization. In the first case, we need to impose that the spatial coefficients $\lambda_{0i}$ increases in the order $O(p)$ as $p\rightarrow\infty$ (otherwise the model becomes not identifiable in the high dimension), whereas in the last case of $L_2$ norm they can remain constant for $p\rightarrow\infty$. In case (iii), which is more general because it is valid for any ${\mathbf W}$, we need to impose $k_i=o(T^{1/2})$ in order to guarantee the consistency of the estimators. To complete this section, we want to show the class of processes that can be analysed by our \emph{stationary SDPD} model. Under assumption $A2$, any \emph{stationary SDPD} model can be equivalently represented as a VAR process as in (\ref{b1bis}), with respect to an autoregressive matrix coefficient ${\mathbf A}^*$ defined in (\ref{diagonalize}). Now, by exploiting the simple structure of our model, we can show the conditions under which the opposite is true. The following corollary derives from standard results. \begin{corollary}\label{corollary1} Given a stationary multivariate process ${\mathbf y}_t={\mathbf A}^*{\mathbf y}_{t-1}+\mbox{\boldmath$\varepsilon$}_t^*$, with $\mbox{\boldmath$\varepsilon$}_t^*$ satisfying assumption $A3$, a necessary and sufficient condition to represent the process ${\mathbf y}_t$ by a stationary SDPD model is that matrix ${\mathbf A}^*$ is diagonalizable. Therefore, matrix ${\mathbf A}^*$ must have $p$ linearly independent eigenvectors. This is (alternatively) assured by one of the following sufficient conditions: \begin{itemize} \item the eigenvalues $\lambda_{11},\ldots,\lambda_{1p}$ of matrix ${\mathbf A}^*$ are all distinct, or \item the eigenvalues $\lambda_{11},\ldots,\lambda_{1p}$ of matrix ${\mathbf A}^*$ consist of $h$ distinct values $\mu_1,\ldots,\mu_h$ having geometric multiplicities $r_1,\ldots,r_h$, such that $r_1+\ldots+r_h=p$. \end{itemize} \end{corollary} By corollary \ref{corollary1} and assumptions $A1-A4$, the VAR processes that cannot be represented and consistently estimated by our \emph{stationary SDPD} model are those characterized by a matrix ${\mathbf A}^*$ with linear dependent eigenvectors (i.e., those with algebraic multiplicities) or those with complex eigenvalues. In order to apply our model to those cases also, we should generalize the estimation procedure using the Jordan decomposition of matrix ${\mathbf A}^*$. However, we leave this topic to future study. \section{Simulation study}\label{simulazioni} This section contains the results of a simulation study implemented to evaluate the performance of the proposed estimation procedure. In section 5.1, we describe the settings and check the validity of the assumptions for the simulated models. Then, in section 5.2, we evaluate the consistency of the estimation procedure and the convergence rate for the estimators using a known spatial matrix. Finally, section 5.3, we analyse the case when the spatial matrix ${\mathbf W}$ is unknown, and therefore, to be estimated. \subsection{Settings} We consider three different spatial matrices. In the first, we randomly generate a matrix of order $p\times p$, and we post-multiply this matrix by its transpose in order to force symmetry. The resulting spatial matrix is denoted with ${\mathbf W}_1$. Note that such a matrix is \emph{full}, and it may have positive and negative elements. In the other two cases, the spatial matrix is \emph{sparse} and has only positive entries: ${\mathbf W}_2$ is generated by setting to one only four values in each row while ${\mathbf W}_3$ is generated by setting to one $2\sqrt{p}$ elements in each row. For all three cases, we check the rank to guarantee that the spatial matrix has $p$ linearly independent rows. Moreover, we set to zero the main diagonal, and we rescale the elements so that each row has norm equal to one ($L_2$ row-normalization). For the error process, we generate $p$ independent Gaussian series $e_{ti}$ with mean zero and standard error ${\sigma}_i$, where the values ${\sigma}_i$ are generated randomly from a uniform distribution $U(0.5, 1.5)$ for $i=1,\ldots,p$. Then, we define the cross-correlated error process $\mbox{\boldmath$\varepsilon$}_t=\{\varepsilon_{it},t=1,\ldots,T\}$, where \[ \left\{ \begin{array}{ll} \varepsilon_{ti} = e_{ti} -0.7*e_{t2} & \mbox{for }i=3,\ldots,p, \\ \varepsilon_{ti} = e_{ti} & \mbox{otherwise}. \\ \end{array} \right. \] We generate all $\lambda_{ji}$ from a uniform distribution $U(-0.7, 0.7)$. The settings above guarantee that assumptions $A1-A4$ hold. We generate different models with dimensions $p = (10, 50, 100, 500)$ and sample sizes $T = (50, 100, 500, 1000)$. Note that we may have $T<<p$. For each configuration of settings, we generate 500 Monte Carlo replications of the model and report the estimation results. All the analyses have been made in R. \subsection{Empirical performance of the estimators when ${\mathbf W}$ is known} Figure \ref{figure5} shows the box plots of the estimations for increasing sample sizes $T = (50, 100, 500, 1000)$ and fixed dimension $p = 100$. The four plots at the top refer to the estimation of $\lambda_{0i}$ while the four plots at the bottom refer to that of $\lambda_{1i}$. Each plot focuses on a different {location} $i$, where $i = 97,\ldots,100$. The true values of the coefficients $\lambda_{ji}$ are shown through the horizontal lines. Note that we have $T\leq p$ for the first two box plots in each plot, since $p = 100$ for this model. The box plots are centred on the true value of the parameters, and the variance reduces for increasing values of $T$, showing consistency of the estimators and a good performance for small $T$/large $p$ also. To evaluate the estimation error, for each realized time series, we compute the average error $AE$ and the average squared error $ASE$ using the equations below. \begin{equation}\label{mse} AE(\widehat\boldsymbol{\lambda}_j) = \frac{1}{p}\sum_{i=1}^p{(\hat\lambda_{ji}-\lambda_{ji})}, \qquad ASE(\widehat\boldsymbol{\lambda}_j) = \frac{1}{p}\sum_{i=1}^p{(\hat\lambda_{ji}-\lambda_{ji})^2}, \qquad j=1,2. \end{equation} Table \ref{tabella1} reports the mean values of $ASE(\widehat\boldsymbol{\lambda}_j)$ (with the standard deviations in brackets) computed over 500 simulated time series for different values of $T$ and $p$. As shown in the table, the estimation error decreases when the sample size $T$ increases. It is interesting to note that the estimation error does not increase for increasing values of the dimension $p$. This is more evident from figure \ref{increasing_p_global}, which shows the box plots of the average errors $AE(\boldsymbol{\lambda}_0)$ (at the top) and $AE(\boldsymbol{\lambda}_1)$ (at the bottom) computed for 500 replications of the model, with varying values of $p$, sample sizes $T$, and spatial matrix ${\mathbf W}_1$. We can note from the figure that $\hat\boldsymbol{\lambda}_0$ and $\hat\boldsymbol{\lambda}_1$ are unbiased for all $n$ and $p$. Moreover, the variability of the box plots decreases for $p\rightarrow\infty$ and fixed $T$: this is a consequence of averaging the absolute error over the $p$ locations using equation (\ref{mse}). \subsection{Estimation results when the spatial matrix is unknown}\label{matrixA} In this section, we evaluate the performance of the proposed estimation procedure when the spatial matrix ${\mathbf W}$ is unknown and needs to be estimated. In this case, the estimation error has to be evaluated with respect to matrix ${\mathbf A}^*$ in order to include the effects of both $\hat\boldsymbol{\lambda}_{j}$ and $\hat{\mathbf W}$ on the final estimations. So, using (\ref{diagonalize}), we define the two estimators \begin{eqnarray} \hat{\mathbf A}_{SDPD}^*({\mathbf W}) &=& \left[{\mathbf I}_p-D(\hat\boldsymbol{\lambda}_0){\mathbf W}\right]^{-1}D(\hat\boldsymbol{\lambda}_1)\left[{\mathbf I}_p-D(\hat\boldsymbol{\lambda}_0){\mathbf W}\right] and \label{AW}\\ \hat{\mathbf A}_{SDPD}^*(\hat{\mathbf W}) &=& \left[{\mathbf I}_p-D(\hat\boldsymbol{\lambda}_0)\hat{\mathbf W}\right]^{-1}D(\hat\boldsymbol{\lambda}_1)\left[{\mathbf I}_p-D(\hat\boldsymbol{\lambda}_0)\hat{\mathbf W}\right],\label{AWhat} \end{eqnarray} where matrix ${\mathbf W}$ is assumed to be known in the first case and unknown in the second. When ${\mathbf W}$ is unknown, we estimate it by the (row-normalized) correlation matrix at lag zero, but other more efficient estimators of ${\mathbf W}$ can be considered alternatively. For the sake of comparison, remembering the \emph{VAR} representation of our model in (\ref{b1bis}), we also estimate matrix ${\mathbf A}^*$ using the classic Yule--Walker estimator of the VAR model $\hat{\mathbf A}_{VAR}^*=\hat\boldsymbol{\Sigma}_0^{-1}\hat\boldsymbol{\Sigma}_1$. To give a measure of the estimation error, we define \begin{equation}\label{mse2} ASE({\mathbf A}^*_{(1)})= \frac{1}{p}\sum_{i=1}^p{(\hat A^*_{1i}-A^*_{1i})^2}, \end{equation} where $A^*_{1i}$ for $i=1,\ldots,p$ are the true coefficients in the first row of matrix ${\mathbf A}^*$, and $\hat A^*_{1i}$ are their estimated values. The box plots in figure \ref{figure6} summarize the results of the estimations from 500 replications of the model with $p=100$ (at the top) and $p=500$ (at the bottom). We report the average squared error computed by (\ref{mse2}) in three different cases: the classic Yule--Walker estimator of the VAR model $\hat{\mathbf A}_{VAR}^*$ on the left, our estimator $\hat{\mathbf A}_{SDPD}^*({\mathbf W})$ proposed in (\ref{AW}) with the known spatial matrix in the middle, and our estimator $\hat{\mathbf A}_{SDPD}^*(\hat{\mathbf W})$ proposed in (\ref{AWhat}) with the estimated spatial matrix on the right. Figure \ref{figure6} shows interesting results. First, note that the classic estimator $\hat{\mathbf A}_{VAR}^*$ cannot be applied when $T\leq p$, and this is a serious drawback of the classic VAR models. On the other hand, the \emph{stationary SDPD} model is equivalently used to represent the same process but it can always generate an estimation result for all values of $T$ and $p$ regardless of whether ${\mathbf W}$ is known or unknown. Moreover, if we compare the box plots, we can note that both the median and the variability of the estimators $\hat{\mathbf A}_{SDPD}^*({\mathbf W})$ and $\hat{\mathbf A}_{SDPD}^*(\hat{\mathbf W})$ are remarkably lower than those relative to the classic estimator $\hat{\mathbf A}_{VAR}^*$ (when available) for all sample sizes $T$ and dimensions $p$. This deserves a further remark: while it is expected that the estimator $\hat{\mathbf A}_{SDPD}^*({\mathbf W})$ performs better than $\hat{\mathbf A}_{VAR}^*$ (given that it exploits the knowledge of the true spatial matrix ${\mathbf W}$), it is surprising to also see that the estimator $\hat{\mathbf A}_{SDPD}^*(\hat{\mathbf W})$ outperforms the classic estimator $\hat{\mathbf A}_{VAR}^*$, notwithstanding the fact that they function under the same conditions (only the time series ${\mathbf y}_t$ is observed and no spatial matrix is known). Of course, the ASE of the estimator $\hat{\mathbf A}_{SDPD}^*(\hat{\mathbf W})$ slightly increases compared to that of the estimator $\hat{\mathbf A}_{SDPD}^*(\hat{\mathbf W})$, but its variability remains more or less the same.
c2084d91152fccc47f8e0c5b3c965e679a0d55ea
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section*{Abstract}{\small Here is considered the full evolution of a spherical supernova remnant. We start by calculating the early time ejecta-dominated stage and continue through the different phases of interaction with the circumstellar medium, and end with the dissipation and merger phase. The physical connection between the phases reveals new results. One is that the blast wave radius during the adiabatic phase is significantly smaller than it would be, if one does not account for the blast wave interaction with the ejecta. \vspace{10mm} \normalsize} \end{minipage} \section{Introduction} $\,\!$\indent A supernova remnant (SNR), the aftermath of a supernova explosion, is an important phenomenon of study in astrophysics. The typical $10^{51}$ erg of energy released in the explosion is transferred primarily into the interstellar medium during the course of evolution of a SNR. SNR are also valuable as tools to study the evolution of star, the evolution of the Galaxy, and the evolution of the interstellar medium. A SNR emits in X-rays from its hot shocked gas, in infrared from heated dust, and in radio continuum. The latter is via synchrotron emission from relativistic electrons accelerated at the SNR shock. The evolution of a single SNR can be studied and calculated using a hydrodynamics code. However to study the physical conditions of large numbers of SNR, it is desirable to have analytic methods to obtain input parameters needed to run a detailed hydrodynamic simulation. The short paper describes the basic ideas behind the analytic methods, the creation of software to carry out the calculations and some new results of the calculations. \section{Theory and calculation methods} $\,\!$\indent The general time sequence of events that occur after a supernova explosion, which comprise the supernova remnant can be divided into a number of phases of evolution (Chevalier, 1977). These are summarized as follows. The ejecta dominated (ED) phase is the earliest phase when the ejecta from the explosion are not yet strongly decelerated by interaction. Self-similar solutions were found for the ejecta phase for the case of a supernova with ejecta with a power-law density profile occurring in a circumstellar medium with a power-law density profile (Chevalier, 1982). Solutions were given for ejecta power-law indices of 7 and 12, and circumstellar medium power-law indices of 0 and 2. The latter correspond to uniform a circumstellar medium and one caused by a stellar wind with constant mass-loss rate. The non-self similar evolution between ED to the Sedov-Taylor (ST) self-similar phase was treated by Truelove and McKee (1999). They found the so-called unified solution for the evolution of the forward and reverse shock waves during this phase. The Sedov-Taylor (ST) self-similar phase is that for which the shocked ISM mass dominates over the shocked ejecta mass and for which radiative energy losses from the hot interior supernova remnant gas remain negligible. These solutions are reviewed in numerous works, and are based on the original work on blast waves initiated by instantaneous point energy injection in a uniform medium (Taylor, 1946; Sedov, 1946). The next stage occurs when radiative losses from the post-shock gas become important enough to affect the post-shock pressure and the dynamics of expansion of the supernova remnant. This phase is called the pressure-driven snowplow phase (PDS phase). Cooling sets in most rapidly for the interior gas closest to the outer shock front, so that a thin cold shell forms behind the shock. Interior to the thin shell, the interior remains hot and has significant pressure, so it continues to expand the shell. The shell decelerates because it is gaining mass continually while being acted upon by the interior pressure. Here we refer the review of this phase of evolution by Cioffi, McKee and Bertschinger (1988) This work also compares the analytic solutions to numerical hydrodynamic solutions for verification. When the interior pressure has dropped enough, it no longer influences the evolution of the massive cool shell. After this time, the supernova remnant is in the momentum conserving shell (MCS phase. The shell slows down according to the increase in swept up mass from the interstellar medium. The final fate of a supernova remnant is merger with the interstellar medium, when the shock velocity drops low enough the the expanding shell is no longer distinguishable from random motions in the interstellar medium. To create an analytic model, or its realization in software, the different phases of evolution were joined. This problem is not simple, as pointed out in the work of Truelove and McKee (1999). The evolution of the SNR is determined by the distribution of mass, pressure and velocity within the SNR and the shock jump conditions where there are any shocks. We follow similar methods to those in Truelove and McKee (1999), to ensure that the SNR evolution has continuous shock velocity and radius with time and closely follows that of more detailed hydrodynamic calculations. \section{Results} $\,\!$\indent Analytic solutions have been created which cover the evolution of the SNR from early ED phase through ED-ST transition, ST phase, ST to PDS transition and final dissolution of the SNR. We have taken care to properly join the different phases as noted above. These solutions allow variation in the input physical parameters, such as explosion energy, ejected mass, ejecta and circumstellar medium density profiles and age. The numerical implementation of the solutions provides various output quantities, such as forward and reverse shock radius, and shock velocities and temperatures. These can be compared to the observed properties of a given SNR. Adjustment of the input parameters to match the observed properties yields estimates of the physical properties of the SNR, and also allows estimates in uncertainties in these properties. One of the new results from the analytic calculations is that the shock radius at any given time during the ST phase is significantly less than it is for the standard analytic ST solution. The reduced shock radius is a real physical effect and is understood as caused by interaction of the reverse shock wave with the (initially unshocked) ejecta. This result has not been pointed out previously, and will change SNR parameter estimates that have been made with the standard ST solution. Results of some of the calculations with the full-evolution model are shown in Figures 1 and 2. Figure 1 shows the forward and reverse shock radii and velocities for the ED phase, ED to ST phase and ST phase, for a SNR in a uniform circumstellar medium, and the parameters listed in the figure caption. Figure 2 shows similar plots for a SNR in a stellar wind circumstellar medium. \begin{figure} \center \includegraphics[width=\textwidth]{s0n7.JPG} \caption{Left panel: forward and reverse shock radius vs. time for a SNR with energy $E=10^{51}$erg, ejected mass $2M_{\odot}$, in a uniform circumstellar medium ($s=0$) with density 1 cm$^{-3}$ and temperature 100 K. The ejecta density power-law index is $n=7$. Right panel: forward and reverse shock velocity vs. time.} \end{figure} \begin{figure} \center \includegraphics[width=\textwidth]{s2n7.JPG} \caption{Left panel: forward and reverse shock radius vs. time for a SNR with energy $E=10^{51}$erg, ejected mass $2M_{\odot}$, in a stellar wind ($s=2$) with wind velocity 30 km/s and mass loss rate $10^{-6}M_{\odot}$/yr. The ejecta density power-law index is $n=7$. Right panel: forward and reverse shock velocity vs. time.} \end{figure} \small \section*{Acknowledgments} Support for this work was provided the Natural Sciences and Engineering Research Council of Canada. \section*{References} \bibliographystyle{aj} \small
ea26d50ef9855d66aef5d2296d47fb43cf42e00c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The purpose of this paper is to provide a convenient operadic framework for the cumulants of free probability theory. In~\cite{DrummondColeParkTerilla:HPTI,DrummondColeParkTerilla:HPTII}, the author and his collaborators described an operadic framework for so-called Boolean and classical cumulants. In those papers, the fundamental object of study is an algebra $A$ equipped with a linear map $E$, called \emph{expectation}, to some fixed algebra $B$. The expectation is not assumed to be an algebra homomorphism; rather one measures the degree to which $E$ fails to be an algebra homomorphism with a sequence of multilinear maps $\kappa_n$ from powers of $A$ to $B$, called cumulants. The cumulants, in many cases, can be defined recursively in terms of the expectation map via a formula of the form: \begin{equation}\label{outline of cumulants} E(x_1\cdots x_n)= \sum \kappa_{i_1}(\cdots)\cdots \kappa_{i_k}(\cdots). \end{equation} Depending on what kind of probability theory is under consideration, the summation on the left may be over a different index set. See, e.g.,~\cite{Speicher:OUP,Muraki:FIQUP,HasebeSaigo:JCNI}. In~\cite{DrummondColeParkTerilla:HPTI,DrummondColeParkTerilla:HPTII}, these recursive definitions for the collection of cumulants (in the Boolean and classical regimes, respectively) were reinterpreted as the collection of linear maps determining a coalgebraic map into a cofree object. In the Boolean case, the cofree object is the tensor coalgebra. In the classical case it is the symmetric coalgebra. This reformulation is intended as the background for a homotopical enrichment of probability theory; adding a grading, a filtration, and a differential to this coalgebraic picture leads to a rich theory with applications to quantum field theory~\cite{Park:HTPSICIHLA}. This application is motivational and will play no role in this paper. None of the work mentioned above treats the case of \emph{free cumulants}, arguably the most important kind of cumulant in noncommutative probability theory. When the target algebra $B$ is commutative, there is a formula similar to those above and the framework outlined above can be used directly, employing a more exotic type of coalgebra than the tensor or symmetric coalgebra. This point of view is taken in~\cite{DrummondCole:NCWCFFHPT}. However, there is a flaw in this point of view, which is that assuming the target to be commutative is external to the theory; internally it makes perfect sense for the target itself to be noncommutative. This is called \emph{operator-valued} free probability theory because the expectation is valued in a noncommutative algebra, such as an operator algebra. Operator-valued free cumulants, as defined by Speicher,~\cite{Speicher:CTFPAOVFPT} are somewhat more cumbersome to describe explicitly than in the commutative case using classical combinatorial methods. Consequently Speicher develops an \emph{operator-valued $R$-transform} to collect the information concisely. In our setting, there is one evident related obstruction to extending the framework developed in~\cite{DrummondColeParkTerilla:HPTI,DrummondColeParkTerilla:HPTII} to operator-valued free cumulants. The defining formulas for classical and Boolean cumulants and for free cumulants valued in a commutative algebra share a certain property. Namely, they are \emph{string-like}, meaning that the right-hand side of Equation~(\ref{outline of cumulants}) is a product of cumulants. However, the defining formulas for operator-valued free cumulants (that is, free cumulants valued in a not necessarily commutative algebra) contain terms like \[ \kappa_2(x_1\kappa_1(x_2)\otimes x_3). \] or more generally \[ \kappa_{n_1}(x_1\kappa_{n_2}(x_2\kappa_{n_3}(\cdots)\kappa_{n_4}(\cdots),\cdots),\cdots). \] In a word, they are not string-like but \emph{tree-like}. This is precisely the issue that leads Speicher to develop the operator-valued $R$-transform. Here, this tool is avoided by using an operadic reformulation. Tree-like formulas can be obtained by passing from algebras and coalgebras, which have a string-like structure, to nonsymmetric operads and cooperads, which have a tree-like structure. The main result of this paper shows how the relationship between the moments and free cumulants, realized as cooperadic maps $M$ and $K$, is encapsulated quite simply in terms of a canonical twist: \begin{equation*} M=\Phi\circ K. \end{equation*} As phrased in this paper, the moments and cumulants are defined in some other manner and this is a theorem, but it is probably better to consider this as an alternative definition which is quite simple from the operadic viewpoint. This reformulation is part of a campaign to explore applications of the operadic language in probability theory; the result contained herein is modest and is intended to serve as further advertisment and evidence (following~\cite{Male:DTLRMTFP,DrummondColeParkTerilla:HPTI,DrummondColeParkTerilla:HPTII,DrummondColeTerilla:CIHPT,DrummondCole:NCWCFFHPT}) of potentially deeper connections between the two areas. It is possible that both this reformulation and those attempted in the author's previous work (cited above) are reflections of a combinatorial relationship between operads and M\"obius inversion with respect to a poset. This is not pursued further here, but see~\cite[3.3]{Mendez:SOCCS} for some discussion and further references on this topic. The remainder of the paper is organized as follows. Section~\ref{sec: operads} reviews the parts of operadic theory that are used in the paper. Section~\ref{sec: partitions and trees} goes over the combinatorics of non-crossing partitions, and Section~\ref{sec:freeprob} applies this to define free cumulants. Finally, Section~\ref{sec: main result} states and proves the reformulation of free cumulants in operadic terms. \subsection{Conventions} Everything linear occurs over a fixed ground ring. Algebras are generally not assumed to be commutative or unital. Every finite ordered set is canonically isomorphic to $[n]\coloneqq\{1,\ldots, n\}$, and this canonical isomorphism will be routinely abused. A graph is a finite set of vertices, a finite set of half-edges, a source map from half-edges to vertices, and an involution on the half-edges; a half-edge is a leaf it is a fixed point of the involution. A graph is connected if every two vertices can be joined by a path of half-edges connected by having the same source or via the involution. A connected graph is a tree if it has more vertices than edges. A root is a choice of leaf of a tree (this is no longer considered a leaf). The root of a vertex is the unique half-edge ``closest'' to the overall root. The root vertex is the unique vertex whose root is the overall root. A planar tree has a cyclic order on the half-edges of each vertex. \section{Operads and cooperads}\label{sec: operads} Aside from some minor changes, conventions of~\cite{LodayVallette:AO} are used for operadic algebra. This section reviews standard definitions (more details can be seen in~\cite[5.9]{LodayVallette:AO}). \begin{defi} A \emph{collection} $M=\{M_n\}_{n\ge 0}$ is a set of modules indexed by nonnegative numbers (the index is called \emph{arity}). Given a collection $M$, a graph \emph{decorated} by $M$ is a pair $(G,D)$ where $G$ is a graph and $D=\{D_v\}$ is a collection of elements of $M$ indexed by the vertices of $G$; for a vertex of valence $k+1$ the decoration $D_v$ should be in the module $M_k$. There is a \emph{composition product} denoted $\circ$ on collections \[ (M\circ N)_n=\bigoplus_k M_k\otimes \left(\bigoplus_{i_1+\cdots+i_k=n} N_{i_1}\otimes\cdots \otimes N_{i_k}\right). \] This product has a unit $I$, where $I_1$ is the ground ring and $I_{n\ne 1}$ is $0$, and together $\circ$ and $I$ make the category of collections into a monoidal category. \end{defi} \begin{defi} A \emph{nonsymmetric operad} is a monoid $\mathbf{P}$ in this monoidal category. Its data can be specified by giving a collection $P$, a \emph{composition} map $\gamma:P\circ P\to P$, and a \emph{unit} map $\eta:I\to P$ satisfying associativity and unital constraintns. A \emph{nonsymmetric cooperad} is a comonoid $\mathbf{C}$ in this monoidal category. Its data can be specified by giving a collection $C$, a \emph{decomposition} map $\Delta:C\to C\circ C$, and a \emph{counit} map $\epsilon:C\to I$ satisfying coassociativity and counital constraints. The collection $I$ has a canonical nonsymmetric cooperad structure, denoted $\mathbf{I}$. \end{defi} In this paper everything will be nonsymmetric and the adjective will be omitted. \begin{defi} Let $\mathbf{C}$ be a cooperad with underlying collection $C$. A \emph{coaugmentation} of $\mathbf{C}$ is a map of cooperads $\eta:\mathbf{I}\to \mathbf{C}$. The decomposition map $\Delta$ induces a decomposition map $\widetilde{\Delta}:C\to C\circ C$ realized by $\Delta-\eta\mathbf{I}\circ\id+\eta\epsilon\circ\eta\mathbf{I}$. The notation $\widetilde{\Delta}^n$ is used for the map $C\to C^{\circ n+1}$ given by composition of the maps \[(\widetilde{\Delta}\circ\underbrace{\id\circ\cdots\circ \id}_{n-1}) : {C}^{\circ n}\to {C}^{\circ n+1} .\] A coaugmented cooperad $\mathbf{C}$ is \emph{conilpotent} if for every element $c\in\mathbf{C}$, there is a natural number $N$ such that $\Delta^N c = (\Delta^{N-1}c)\circ \epsilon\mathbf{I}$. \end{defi} \begin{example} \begin{itemize} \item The motivating example of an operad is the \emph{endomorphism operad} of a vector space $B$, denoted $\End B$. The module $(\End B)_n$ is $\Hom(B^{\otimes n},B)$ and the image of the unit is the identity map of $B$. Composition is given by composition of maps among tensor powers of $B$. \item The category of modules is a full subcategory of the category of cooperads (or operads) where for a module $M$, the cooperad $\mathbf{M}$ has $M_0=M$, $M_1=I_1$, and only trivial compositions. \item The coassociative cooperad has $M_n$ equal to the ground field for all $n$ with every decomposition map induced by the canonical isomorphism between the ground field and its tensor powers. \end{itemize} \end{example} \begin{defi} Operads have a forgetful functor to collections whose left adjoint is called the \emph{free} operad on a collection. Conilpotent coaugmented cooperads have a forgetful functor to collections whose right adjoint is called the \emph{cofree} cooperad on a collection (suppressing conilpotence and coaugmentation). \end{defi} Both the cofree and free functor on $M$ can be realized at the collection level as the collection of rooted planar trees with vertices decorated by elements of $M$, denoted $\mathcal{T}(M)$. This implies the following. \begin{enumerate} \item Fix an operad $\mathbf{P}$ (with underlying collection $P$) and an element of $\mathcal{T}(M)$. That is, take a planar rooted tree $T$ and an element of $\mathbf{P}(n)$ for every vertex of $T$ of valence $n+1$ (collectively called \emph{a decoration of $T$ by $\mathbf{P}$}). Then there is a canonical element of $\mathbf{P}$ called \emph{the composition of the decoration} induced by the counit of the forgetful free adjunction $\mathcal{F}(P)\xrightarrow{\boldsymbol{\epsilon}} \mathbf{P}$. Since this is a monad, this operation is associative, in the sense that this composition can be done subtree by subtree and the output is insensitive to the choice of subtrees or order of composition. \item Dually, given a cooperad $\mathbf{C}$ with underlying collection $C$, the unit of the cofree forgetful adjunction $\mathbf{C}\xrightarrow{\boldsymbol{\eta}} \free^c(C)$ yields the following. For every planar rooted tree $T$ with $n$ leaves and vertices $\{v_i\}$ where $v_i$ has valence $n_i+1$, and every element $c\in \mathbf{C}(n)$, there is a canonical set of elements $c_i\in \mathbf{C}(n_i)$. This procedure is called the \emph{decomposition} of $c$ into a decoration of $T$ by $\mathbf{C}$. It can be realized as follows. Let $\widetilde{\Delta}^N c$ stabilize as in the definition of conilpotence. Each summand corresponds to a tree with levels and decorations. Forget the levels and any decorations by $\mathbf{I}$ and project onto the summand corresponding to the tree $T$. See~\cite[5.8.7]{LodayVallette:AO} for more details. \item There is a canonical linear isomorphism $\psi$ between the free operad on a collection and the conilpotent cofree cooperad on the same collection. \end{enumerate} \begin{lemma}\label{lemma:isomorphismsmatch} Let $M$ be a collection. An endomorphism $\free^c(M)\to\free^c(M)$ is an isomorphism if and only if its restriction $M\subset \free^c(M)\to M$ is an isomorphism of collections. \end{lemma} This is in precise parallel to the situation with power series, where a power series is invertible if and only if its constant term is invertible. \begin{proof} For $F$ an endomorphism, let $F_r$ denote its restriction. Note that $(F\circ G)_r = F_r\circ G_r$, which implies that if $F$ is an isomorphism, so is $F_r$. On the other hand, if $F_r$ is an isomorphism, then induction on the number of vertices in a tree in $\free^c(M)$ allows one to build an inverse $F^{-1}$. \end{proof} \begin{defi}\label{defi:canonical twist} Let $\mathbf{P}$ be an operad with underlying collection $P$ The \emph{canonical twist} $\Phi_\mathbf{P}:\free^c(P)\to\free^c(P)$ is the cooperad map induced by the composition $\phi_\mathbf{P}=\boldsymbol{\epsilon}\circ \psi$: \[\free^c(P)\xrightarrow{\psi} \mathcal{F}(P)\xrightarrow{\boldsymbol{\epsilon}} \mathbf{P}\to P.\] \end{defi} \begin{lemma} Let $\mathbf{P}$ be an operad. Then the canonical twist is an isomorphism. \end{lemma} \begin{proof} Restricted to $P$, the canonical twist is the identity. Then Lemma~\ref{lemma:isomorphismsmatch} implies the result. \end{proof} \section{Partitions and trees}\label{sec: partitions and trees} \begin{defi} Let $[n]$ be an ordered set and let $\pi=(p_1,\ldots, p_k)$ be a partition of it, so that $[n]$ is the disjoint union of the blocks $p_i$. Blocks in our partitions are always ordered so that $\min p_i<\min p_j$ whenever $i<j$. A partition $\pi$ is \emph{crossing} if there exist $w$ and $y$ in $p_i$ and $x$ and $z$ in $p_j$ (with $i\ne j$) such that $w<x<y<z$. A partition $\pi$ is \emph{non-crossing} if it is not crossing. The notation $NC(n)$ (respectively $NC_k(n)$) refers to the set of noncrossing partitions of $[n]$ (with $k$ blocks). The unique partition with a single block is called the trivial partition. \end{defi} Noncrossing partitions are important in combinatorics and there are many bi-indexed sets of combinatorial objects in canonical bijection with them. For our purposes, the following such bijection will be useful. \begin{lemma} The set $NC_k(n)$ is in bijection with the set of planar rooted trees with $n$ leaves and $k+1$ vertices (including the root) satisfying the conditions that \begin{enumerate} \item Every non-root vertex has at least one leaf attached to it, and \item the root has no leaves attached to it. \end{enumerate} The bijection from trees to partitions is given explicitly by numbering the leaves clockwise starting from the root and then letting two numbers share a block if the corresponding leaves are incident on the same vertex. Thus blocks are in bijection with non-root vertices. \end{lemma} See Figure~\ref{figure: partitions trees}. Henceforth partitions will be freely identified with the corresponding trees. \begin{figure} \begin{tikzpicture} \draw (0,0) -- (0,-1); \draw (0,0) -- (0,1); \draw (2.5,2) -- (0,1); \draw (1.5,2) -- (0,1); \draw (.5,2) -- (0,1); \draw (-.5,2) -- (0,1); \draw (-1.5,2) -- (0,1); \draw (-2.5,2) -- (0,1); \draw (-1.5,2) -- (-1.5,3); \draw (-.5,2) -- (-.5,3); \draw (1.5,2) -- (1.5,3); \draw [fill] (0,0) circle [radius=.05]; \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (-1.5,2) circle [radius=.05]; \draw [fill] (-.5,2) circle [radius=.05]; \draw [fill] (1.5,2) circle [radius=.05]; \node [above] at (-2.5,2) {$1$}; \node [above] at (-1.5,3) {$2$}; \node [above] at (-.5,3){$3$}; \node [above] at (.5,2){$4$}; \node [above] at (1.5,3) {$5$}; \node [above] at (2.5,2) {$6$}; \node [below] at (0,-1) {$(146)(2)(3)(5)$}; \draw (6,0) -- (6,-1); \draw (6,0) -- (7,1); \draw (6,0) -- (5,1); \draw (5,1) -- (6,2); \draw (5,1) -- (5,2); \draw (5,1) -- (4,2); \draw (5,2) -- (4,3); \draw (5,2) -- (5,3); \draw (5,2) -- (6,3); \draw (5,3) -- (5,4); \draw (7,1) -- (7,2); \draw [fill] (6,0) circle [radius=.05]; \draw [fill] (5,1) circle [radius=.05]; \draw [fill] (5,2) circle [radius=.05]; \draw [fill] (5,3) circle [radius=.05]; \draw [fill] (7,1) circle [radius=.05]; \node [above] at (4,2) {$1$}; \node [above] at (4,3) {$2$}; \node [above] at (5,4) {$3$}; \node [above] at (6,3) {$4$}; \node [above] at (6,2) {$5$}; \node [above] at (7,2) {$6$}; \node [below] at (6,-1) {$(15)(24)(3)(6)$}; \end{tikzpicture} \caption{Two non-crossing partitions and the corresponding planar rooted trees} \label{figure: partitions trees} \end{figure} It will be useful later to modify this construction. \begin{construction}\label{construction: messy tree} Let $\pi$ be a noncrossing partition of $[n]$. There is a function $h:\{0,\ldots,n\}\to \{*,B_1,\ldots B_k\}$ defined by letting $h(i)$ be the maximal block (should one exist), which contains elements $x$ and $y$ of $[n]$ such that $x\le i$ and $y>i$. Should no such block exist, then $h(i)=*$. Now let $i_0,\ldots, i_n$ be non-negative numbers. The tree $\pi_{i_0,\ldots,i_n}$ is obtained from $\pi$ by the following procedure: \begin{enumerate} \item For each $j$, attach $i_j$ new leaves at the vertex $h(j)$ of $\pi$ (if $h(j)=*$, attach to the root vertex) in the unique possible way so that the new leaves are after the $j$th original leaf and before the $j+1$st original leaf of $\pi$, using the clockwise order around leaves. \item Consider a non-root vertex $v$. Its incoming half-edges are of the form $e_{0,1},\ldots,e_{0,k_0},\ell_1,e_{1,1},\ldots, e_{1,k_1},\ell_2,\ldots, \ell_r, e_{r_1,\ldots, e_{r,k_r}}$, where $\ell_i$ are original leaves and $e_{i,j}$ are new leaves or parts of edges. Let $S_-$ be the set of indices $j$ in $\{0,\ldots, r\}$ such that $k_j\ge 1$ and let $S_+$ be the set of indices $j$ such that $k_j>1$. Now replace $v$ with a tree which has \begin{enumerate} \item one ``bottom'' vertex with incoming half-edges in ordered bijection with $S_-$ and \item \label{item: bottom top} ``top'' vertices in ordered bijection with $S_+$ where the vertex $v_j$ has $k_j$ incoming half-edges. \end{enumerate} Join the root of a top vertex with the corresponding incoming half-edge of the bottom vertex; the other incoming half-edges of the bottom vertex and all incoming half-edges of the top vertex are identified with the incoming half-edges of $v$. \item Delete all of the original leaves of $\pi$; delete the root if it is bivalent at this point in the construction. \end{enumerate} The tree obtained after the intermediate step~\ref{item: bottom top} will also be important and will be called $\overline{\pi}_{i_0,\ldots, i_n}$. \end{construction} See Figure~\ref{fig: messy tree}. \begin{figure} \begin{tikzpicture} \draw(0,0)--(0,1); \draw (0,1)--(0,2); \draw(0,2)--(-1,3); \draw(0,2)--(1,3); \draw(0,2)--(0,3); \draw(0,3)--(1,4); \draw(0,3)--(-1,4); \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (0,2) circle [radius=.05]; \draw [fill] (0,3) circle [radius=.05]; \node[below] at (0,0){$\pi=(14)(23)$}; \node[above] at (-1,3){$1$}; \node[above] at (-1,4){$2$}; \node[above] at (1,4){$3$}; \node[above] at (1,3){$4$}; \begin{scope}[shift ={(3,4)}] \draw(0,-5)--(0,-4); \draw (0,-4)--(1,-3); \draw (0,-4)--(-1,-3); \draw (1,-3)--(0,-2); \draw(1,-3)--(1,-2); \draw (1,-3)--(2,-2); \draw(1,-2)--(2,-1); \draw(1,-2)--(0,-1); \draw(1,-2)--(1,-1); \draw(0,0)--(1,-1); \draw(1,0)--(1,-1); \draw(2,0)--(1,-1); \draw [fill] (0,-4) circle [radius=.05]; \draw [fill] (1,-3) circle [radius=.05]; \draw [fill] (1,-2) circle [radius=.05]; \draw [fill] (1,-1) circle [radius=.05]; \node[right] at (0,-4){$r$}; \node[right] at (1,-3){$b$}; \node[right] at (1,-2){$t$}; \node[right] at (1,-1){$b$}; \node[below] at (0,-5){$\overline{\pi}_{1,1,1,1,0}$}; \node[above] at (0,-2){$1$}; \node[above] at (0,0){$2$}; \node[above] at (2,0){$3$}; \node[above] at (2,-2){$4$}; \end{scope} \begin{scope}[shift ={(8,4)}] \draw(0,-5)--(0,-4); \draw (0,-4)--(1,-3); \draw (0,-4)--(-1,-3); \draw (1,-3)--(1,-2); \draw (1,-2)--(0,-1); \draw(1,-2)--(1,-1); \draw (1,-2)--(2,-1); \draw(1,-1)--(1,0); \draw [fill] (0,-4) circle [radius=.05]; \draw [fill] (1,-3) circle [radius=.05]; \draw [fill] (1,-2) circle [radius=.05]; \draw [fill] (1,-1) circle [radius=.05]; \node[right] at (0,-4){$r$}; \node[right] at (1,-3){$b$}; \node[right] at (1,-2){$t$}; \node[right] at (1,-1){$b$}; \node[below] at (0,-5){${\pi}_{1,1,1,1,0}$}; \end{scope} \begin{scope}[shift ={(3,-8.5)}] \draw(0,0)--(0,1); \draw (0,1)--(0,2); \draw(0,2)--(-1,3); \draw(0,2)--(1,3); \draw(0,2)--(0,3); \draw(0,3)--(-1,4); \draw(0,3)--(0,4); \draw(0,3)--(1,4); \draw (1,4)--(1,5); \draw (1,4)--(0,5); \draw (1,4)--(2,5); \draw (1,5)--(2,6); \draw(1,5)--(0,6); \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (0,2) circle [radius=.05]; \draw [fill] (0,3) circle [radius=.05]; \draw [fill] (1,4) circle [radius=.05]; \draw [fill] (1,5) circle [radius=.05]; \node[right] at (0,1){$r$}; \node[right] at (0,2){$b$}; \node[right] at (0,3){$t$}; \node[right] at (1,4){$b$}; \node[right] at (1,5){$t$}; \node[below] at (0,0){$\overline{\pi}_{0,2,2,0,0}$}; \node[above] at (-1,3){$1$}; \node[above] at (0,5){$2$}; \node[above] at (2,5){$3$}; \node[above] at (1,3){$4$}; \end{scope} \begin{scope}[shift ={(7,-8.5)}] \draw (0,1)--(0,2); \draw(0,2)--(0,3); \draw(0,3)--(-1,4); \draw(0,3)--(0,4); \draw(0,3)--(1,4); \draw (1,4)--(1,5); \draw (1,5)--(2,6); \draw(1,5)--(0,6); \draw [fill] (0,2) circle [radius=.05]; \draw [fill] (0,3) circle [radius=.05]; \draw [fill] (1,4) circle [radius=.05]; \draw [fill] (1,5) circle [radius=.05]; \node[right] at (0,2){$b$}; \node[right] at (0,3){$t$}; \node[right] at (1,4){$b$}; \node[right] at (1,5){$t$}; \node[below] at (0,1){${\pi}_{0,2,2,0,0}$}; \end{scope} \end{tikzpicture} \caption{Examples of Construction~\ref{construction: messy tree}. Top, bottom, and root vertices are labelled $t$, $b$, and $r$ respectively.}\label{fig: messy tree} \end{figure} \begin{remark*} Note that the first and last incoming half-edge at each ``bottom'' vertex of $\overline{\pi}_{i_0,\ldots, i_n}$ are always original leaves of $\pi$. \end{remark*} \section{Free probability and operator-valued free cumulants}\label{sec:freeprob} \begin{defi} Let $B$ be an algebra. A \emph{$B$-valued probability space} consists of a pair $(A,E)$ where $A$ is a $B$-algebra and $E$ is a $B$-linear map, called \emph{expectation} $A\to B$ such that the composition $B\to A\to B$ is the identity. By abuse of notation, $E$ will usually be omitted. \end{defi} Classically $B$ is the ground field but for a general theory it is necessary to allow more general algebras, in particular, non-commutative algebras. To be precise, a $B$-algebra is a $B$-bimodule $A$ equipped with a product $A\otimes_B A\to A$ and a $B$-linear map $\eta:B\to A$ which respects the product structure. Let $A$ be a $B$-valued probability space and let $f:A^{\otimes_B n}\to B$ be a $B$-multilinear map. For an $(n+1)$-tuple $(i_0,\ldots, i_n)$ of non-negative integers with sum $N$, define a map \[ f_{i_0,\ldots, i_n}: Hom(A^{\otimes n}, Hom(B^{\otimes N}, B)) \] whose evaluation on $a_1\otimes a_n$ is given by the composition \[ B^{\otimes N}\to B^{\otimes i_0}\otimes A\otimes B^{\otimes i_1}\otimes A\otimes\cdots\otimes A\otimes B^{\otimes i_n} \to A^{\otimes_B n}\xrightarrow{f} B. \] Where the first map inserts $a_j$ in the $j$th $A$ place and the second map is given by repeated use of the $B$-bimodule structure on $A$. The map $f_{i_0,\ldots, i_n}(a_1,\ldots, a_n)$ can be realized as the composition in $\End B$ along a decoration of the tree $\pi_{\min\{i_0,1\},\ldots, \min\{i_n,1\}}$, where $\pi$ is the trivial partition of $[n]$. Decorate ``top'' vertices and the root, should it exist, with the product in $B$ and decorate the single vertex corresponding to the single block of $\pi$ with $f_{\min\{i_0,1\},\ldots, \min\{i_n,1\}}(a_1,\ldots, a_n)$. The following definition is Definition 2.1.1 of~\cite{Speicher:CTFPAOVFPT}, restricted to $B$-algebras. It has been reworded to use operadic language. \begin{defi}\label{defi: multiplicative function} Let $A$ be a $B$-algebra. For $n\ge 1$, let $f^{(n)}:A^{\otimes_B n}\to B$ be a $B$-linear map. Then the \emph{multiplicative function} \[\hat{f}:\bigcup_n NC(n)\times A^{\otimes_B n}\to B\] is defined on $(\pi, a_1\otimes\cdots\otimes a_n)$ as the composition in $\End B$ along a decoration of the tree $\pi_{0,\ldots, 0}$. ``Top'' vertices and the root vertex of $\pi$, if it survives in $\pi_{0,\ldots,0}$, are decorated with the product in $B$. Let $v$ be a ``bottom'' vertex of $\overline{\pi}_{0,\ldots, 0}$ with ordered incoming half-edge set \[ \ell_0,e_{1,1},\ldots,e_{1,k_1},\ell_1,\ldots,\ell_{k-1},e_{k,i_k},\ell_k \] where $\ell_i$ is an original leaf of $\pi$ which is numbered $n(\ell_i)$ in $\pi$ and $i_j$ are non-negative numbers. Then the corresponding ``bottom'' vertex of $\pi_{0,\ldots 0}$ is decorated with \[f^{(k)}_{0,i_1,\ldots, i_k,0}(a_{n(\ell_0)}\otimes\cdots\otimes a_{n(\ell_k)}).\] Then $\hat{f}(\pi,a_1\otimes\cdots\otimes a_n)$ is the composition of this decoration of $\pi_{0,\ldots, 0}$ in the operad $\End B$, viewed as an element of $(\End B)(0)\cong B$. \end{defi} See Figure~\ref{figure: multiplicative function}. \begin{figure} \begin{tikzpicture} \draw (0,0) -- (0,-1); \draw (0,0) -- (0,1); \draw (2.5,2) -- (0,1); \draw (1.5,2) -- (0,1); \draw (.5,2) -- (0,1); \draw (-.5,2) -- (0,1); \draw (-1.5,2) -- (0,1); \draw (-2.5,2) -- (0,1); \draw (-1.5,2) -- (-1.5,3); \draw (-.5,2) -- (-.5,3); \draw (1.5,2) -- (1.5,3); \draw [fill] (0,0) circle [radius=.05]; \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (-1.5,2) circle [radius=.05]; \draw [fill] (-.5,2) circle [radius=.05]; \draw [fill] (1.5,2) circle [radius=.05]; \node [above] at (-2.5,2) {$1$}; \node [above] at (-1.5,3) {$2$}; \node [above] at (-.5,3){$3$}; \node [above] at (.5,2){$4$}; \node [above] at (1.5,3) {$5$}; \node [above] at (2.5,2) {$6$}; \node [below] at (0,-1) {$\pi=(146)(2)(3)(5)$}; \begin{scope}[shift = {(6,0)}] \draw (0,0) -- (0,-1); \draw (0,0) -- (0,1); \draw (2,2) -- (0,1); \draw (1,2) -- (0,1); \draw (0,2) -- (0,1); \draw (-1,2) -- (0,1); \draw (-2,2) -- (0,1); \draw (-1,2) -- (-2,3); \draw (-1,2) -- (0,3); \draw (0,3) -- (0,4); \draw (-2,3) -- (-2,4); \draw (1,2) -- (1,3); \draw [fill] (0,0) circle [radius=.05]; \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (-1,2) circle [radius=.05]; \draw [fill] (1,2) circle [radius=.05]; \draw [fill] (-2,3) circle [radius=.05]; \draw [fill] (0,3) circle [radius=.05]; \node [above] at (-2,2) {$1$}; \node [above] at (-2,4) {$2$}; \node [above] at (0,4){$3$}; \node [above] at (0,2){$4$}; \node [above] at (1,3) {$5$}; \node [above] at (2,2) {$6$}; \node [below] at (0,-1) {$\overline{\pi}_{0,\ldots, 0}$}; \end{scope} \begin{scope}[shift = {(0,-6)}] \draw (0,0) -- (0,1); \draw (1,2) -- (0,1); \draw (-1,2) -- (0,1); \draw (-1,2) -- (-2,3); \draw (-1,2) -- (0,3); \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (-1,2) circle [radius=.05]; \draw [fill] (1,2) circle [radius=.05]; \draw [fill] (-2,3) circle [radius=.05]; \draw [fill] (0,3) circle [radius=.05]; \node [below] at (0,0) {${\pi}_{0,\ldots, 0}$}; \end{scope} \begin{scope}[shift = {(6,-6)}] \draw (0,0) -- (0,1); \draw (1,2) -- (0,1); \draw (-1,2) -- (0,1); \draw (-1,2) -- (-2,3); \draw (-1,2) -- (0,3); \draw [fill] (0,1) circle [radius=.05]; \draw [fill] (-1,2) circle [radius=.05]; \draw [fill] (1,2) circle [radius=.05]; \draw [fill] (-2,3) circle [radius=.05]; \draw [fill] (0,3) circle [radius=.05]; \node [right] at (0,1) {$f^{(3)}_{0,1,1,0}(a_1\otimes a_4\otimes a_6)$}; \node [right] at (1,2) {$f^{(1)}_{0,0}(a_5)$}; \node [right=2] at (-1,2) {$\mu_B$}; \node [above] at (-2,3) {$f^{(1)}_{0,0}(a_2)$}; \node [above] at (0,3) {$f^{(1)}_{0,0}(a_3)$}; \node [below] at (0,0) {$\hat{f}((146)(2)(3)(5),a_1\otimes\cdots\otimes a_6)$}; \end{scope} \end{tikzpicture} \caption{This figure demonstrates the evaluation of the multiplicative function $\hat{f}$ on the element \[(146)(2)(3)(5),a_1\otimes\cdots\otimes a_6.\] The eventual output is \[f^{(3)}(a_1f^{(1)}(a_2) f^{(1)}(a_3)\otimes a_4f^{(1)}(a_5)\otimes a_6). \] } \label{figure: multiplicative function} \end{figure} The following definition combines Example 1.2.2, Definition 2.1.6, and Proposition 3.2.3 of~\cite{Speicher:CTFPAOVFPT}. \begin{defi} Let $A$ be a $B$-valued probability space. The free cumulant $\kappa_n:A^{\otimes_B n}\to B$ is defined recursively in terms of the expectation as follows: \[ E(a_1\cdots a_n)=\sum_{\pi\in NC(n)} \hat{\kappa}(\pi,a_1\otimes\cdots\otimes a_n) \] \end{defi} It will be useful in the next section to record a version of this defining relationship viewed in $\End B$. The following is a direct application of the definitions. \begin{lemma}\label{lemma: moment-cumulant formula with insertions} The moment and cumulant satisfy the following relations for nonnegative $i_0,\ldots, i_n$ and $a_1,\ldots, a_n$ in $A$: \[ E_{i_0,\ldots, i_n}(a_1,\ldots, a_n) = \left(\sum_{\pi\in NC(n)}\hat{\kappa}(\pi,\bullet)\right)_{i_0,\ldots, i_n}(a_1,\ldots, a_n). \] which in turn is the composition in $\End B$ along the decorated tree $\pi_{i_0,\ldots,i_n}$ with decoration as in Definition~\ref{defi: multiplicative function}. \end{lemma} \section{Main result}\label{sec: main result} \begin{defi} The cooperad $\textsf{coAs}_A$ is the categorical product of the coassociative cooperad and the cooperad which is the algebra $A$ concentrated in arity $0$. \end{defi} Let $V=\langle *\rangle$ be a one-dimensional free module. There is an explicit presentation of the sum of the modules of the underlying collection of $\textsf{coAs}_A$ as $\bigoplus_{n=1}^\infty (A\oplus V)^{\otimes n}$. Here the arity $n$ module consists of those elements that are degree $n$ in the generator $*$ of $V$. In this presentation, the cocomposition map is given as follows. Let $w$ be a word in $*$ and $A$; let $F(W)$ be the the set of all ways of writing $w$ as the concatenation $b_0a_1b_1\cdots a_nb_n$ where \begin{itemize} \item The words $b_i$ are (possibly empty) words in $A$ \item The words $a_i$ are nonempty words in $*$ and $A$. \end{itemize} Then \[ \Delta w =\sum_{F(W)}(b_0*b_1*\cdots *b_n)\circ (a_1\otimes\cdots \otimes a_n). \] The projection map to $\textsf{coAs}$ is given by projecting to $\bigoplus V^{\otimes n}$ and identifying $*^{\otimes n}$ with $1$ in the ground ring. The projection map to $A$ is given by projecting to ${A\oplus V}^{\otimes 1}$, identifying $A$ with itself and $*$ with the image of $I$. \begin{defi} Let $A$ be a $B$-valued probability space. The \emph{moment morphism} $M:\textsf{coAs}_A\to \free^c(\End B)$ is the map of cooperads determined by its linear restriction $m:\textsf{coAs}_A\to \End B$ which is defined on the word \[ w=\underbrace{*\ldots *}_{i_0}a_1\underbrace{*\ldots *}_{i_1}a_2\ldots a_m\underbrace{*\ldots *}_{i_m} \] with $\sum i_j=n$ as \[ m(w)(b_1\otimes\cdots\otimes b_n)=(E\circ\mu_A)_{i_0,\ldots,i_m}(a_1\otimes\ldots\otimes a_m) \] or more explicitly \[ m(w)(b_{0,1}\otimes\cdots\otimes b_{m,i_m})= E(b_{0,1}\cdots b_{0,i_0}a_1b_{1,1}\cdots b_{1,i_1}a_2\cdots a_{m}b_{m,1}\cdots b_{m,i_m}). \] The \emph{free cumulant morphism} $K:\textsf{coAs}_A\to \free^c(\End B)$ is the map of cooperads $\Phi^{-1}\circ M$. \end{defi} See Figure~\ref{fig: moment cumulant diagram}. \begin{figure}\label{fig: moment cumulant diagram} \[ \xymatrix{ &\End B \\\\ &\free^c(\End B)\ar[uu]&\mathcal{F}(\End B)\ar@{=>}[uul]_{\boldsymbol{\epsilon}} \\ \textsf{coAs}_A\ar[uuur]^{m}\ar@{:>}[ur]_M\ar@{:>}[dr]^K\ar@{.>}[dddr]_k \\ &\free^c(\End B)\ar@{:>}[uu]^\Phi\ar[uur]_{\psi}\ar[dd] \\\\ &\End B } \] \caption{This figure shows the relationship between moments and cumulants. In the diagram, single arrows are linear maps and double arrows are cooperad and/or operad maps.} \end{figure} \begin{thm}[Main Result] Let $A$ be a $B$-valued probability space. The restriction $k$ of the free cumulant morphism $K$ consists of the following: \begin{enumerate} \item The map $k(\underbrace{*\ldots*}_n)$ is $(-1)^n$ times multiplication $B^{\otimes n}\to B$ for $n>1$. \item Let $\alpha_1,\ldots, \alpha_{m-1}$ be in $\{0,1\}$. Then \[ k(a_1\underbrace{*\ldots*}_{\alpha_1}a_2\ldots a_{m-2}\underbrace{*\ldots *}_{\alpha_{m-1}}a_m) \] (this description is slightly misleading because $\alpha_j\in \{0,1\}$) is \[(\kappa_m)_{0,\alpha_1,\ldots,\alpha_{m-1},0}(a_1\otimes\cdots\otimes a_m) \] (in particular $k(a_1\ldots a_m)=\kappa(a_1\otimes\cdots \otimes a_m)$). \item Applied to any word which contains an element of $A$ and also two consecutive $*$ or a word beginning or ending with $*$ which contains an element of $A$, the map $k$ vanishes. \end{enumerate} \end{thm} \begin{proof} For the duration of the proof, let $w$ be a word in $\overline{\textsf{coAs}_A}(n)$ which contains precisely the $m$ letters (in order) $a_1,\ldots, a_m$ from $A$ and $n$ $*$ symbols, say \[ w= \underbrace{*\ldots *}_{i_0}a_1\underbrace{*\ldots *}_{i_1}a_2\ldots a_m\underbrace{*\ldots *}_{i_m} \] with $i_j\ge 0$ and $\sum i_j=n$. We give this word weight $2m-1+ \sum i_j$. Weight is nonnegative, positive on the cokernel of the coaugmentation, and preserved by the decomposition map of $\textsf{coAs}_A$. The proof will proceed by induction on weight in each case. Using the equivalent characterization $M=\Phi\circ K$ gives a recursive definition of $k$ on the weight $L$ component in terms of the value of $m$ on the weight $L$ component and the value of $k$ on components of strictly smaller weight. That is, since the codomain $\free^c(\End B)$ is cofree, it suffices to project to $\End B$ for the definition, which yields $m=\phi\circ K$. Then via~\cite[Prop. 5.8.6]{LodayVallette:AO} (or more properly speaking, its nonsymmetric version), the map $K$ can be written at the level of collections as the composition \[ \textsf{coAs}_A\xrightarrow{\boldsymbol{\eta}} \mathcal{T}(\textsf{coAs}_A) \xrightarrow{\mathcal{T}(k)} \mathcal{T}(\End B) \] Since $\phi$ is just $\boldsymbol{\epsilon}\circ \psi$, and $\psi$ is the identity at the level of collections, we can the relationship between $m$ and $k$ as the commutativity of the following diagram: \begin{equation}\label{equation: square} \begin{gathered} \xymatrix{ \textsf{coAs}_A\ar[r]^m\ar[d]_{\boldsymbol{\eta}}&\End B\\ \mathcal{T}(\textsf{coAs}_A)\ar[r]_{\mathcal{T}(k)}&\mathcal{T}(\End B)\ar[u]_{\boldsymbol{\epsilon}} } \end{gathered} \end{equation} where $\boldsymbol{\eta}$ and $\boldsymbol{\epsilon}$ are the canonical decomposition and composition maps. Now $\boldsymbol{\eta}(w)$ consists of a sum of trees decorated with elements of $\overline{\textsf{coAs}_A}(j)$ for $j\le n$. There is one summand corresponding to a tree with a single vertex decorated by $w$ itself. This summand will contribute $k(w)$ to the eventual equation, and it is the purpose of the inductive step to determine its value. We will call this summand the \emph{trivial} summand and call the tree $T_{\triv}$. We call decorated trees with a vertex decoration which contains an element of $A$ and either the string $**$ or the symbol $*$ at the beginning or end of the decoration \emph{degenerate}. In the first and second case of the statement of the theorem, by the inductive premise, only summands corresponding to nondegenerate decorated trees can contribute te $\boldsymbol{\eta}(w)$. In the third case, $T_{\triv}$ is the only degenerate decorated tree that may contribute. Any nondegenerate decorated tree $T$ induces a partition $\pi_T$ of $[m]$ by saying $p$ and $q$ are in the same block if $a_p$ and $a_q$ are in the same vertex decoration. This partition is necessarily non-crossing because the original tree was planar. For $\pi$ a partition, let $\mathcal{T}_\pi$ be the set of nondegenerate decorated trees $T$ such that $\pi_T=\pi.$ Then the sum calculating $\boldsymbol{\eta}(w)$ polarizes into subsums: \[ \sum_T \boldsymbol{\eta}(w)_T = \sum_{\pi\in NC(m)} \sum_{T\in \mathcal{T}_{\pi}}\boldsymbol{\eta}(w)_T \] in the first two cases in the statement of the theorem (in fact, in the first case $m$ is always $0$), and \[ \sum_T \boldsymbol{\eta}(w)_T = T_{\triv} + \sum_{\pi\in NC(m)} \sum_{T\in \mathcal{T}_{\pi}}\boldsymbol{\eta}(w)_T \] in the third case of the statement of the theorem. Then given a partition $\pi$ arising from a nondegenerate decorated tree in the sum, the set of vertex decorations of $T\in \mathcal{T}_\pi$ which contain a letter of $A$ is independent of $T$. This set of vertex decorations can be recovered from $\pi$ as follows. Let $a_{i_0},\ldots, a_{i_j}$ be a block of $\pi$. Then necessarily the vertex decoration is of the form $a_{i_0}*^{\alpha_1}\cdots *^{\alpha_j} a_{i_j}$ where each $\alpha_p$ is either $0$ or $1$. If $a_{i_\ell}$ and $a_{i_{\ell+1}}$ are adjacent in $w$, then necessarily $\alpha_\ell=0$. On the other hand, if $a_{i_\ell}$ and $a_{i_{\ell+1}}$ are separated in $w$, then necessarily $\alpha_\ell=1$. Fix a non-crossing partition $\pi$. Then there is a unique decorated tree in $T_\pi$ with a minimal number of vertices, obtained as a decoration of $\pi_{i_0,\ldots, i_n}$. The ``bottom'' vertices of this tree are decorated by the unique decoration described in the previous paragraph and all other vertices are decorated by $*\ldots *$. Then it is straightforward to verify that the set of nongenerate decorated trees $T_\pi$ consists of all trees obtained from this decoration of $\pi_{i_0,\ldots, i_n}$ by replacing a vertex decorated by $*\ldots*$ by a tree of the same overall arity, all of whose vertices are at least trivalent, and all of whose vertices are decorated by $*\ldots*$. At this point, it may be better to split into cases corresponding to the cases in the statement of the theorem. \begin{enumerate} \item On the word $w_n=(\underbrace{*\ldots*}_n)$ with $n>1$, the canonical decomposition $\boldsymbol{\eta}(w_n)$ is then the sum over all planar rooted trees with $n$ leaves; for each such tree the labels are all $*\ldots *$. On all nontrivial trees, $\boldsymbol{\epsilon}\circ \mathcal{T}(k)$ is, up to sign, just multiplication $\mu_n:B^{\otimes n}\to B$. Then by induction we have the formula \[ \mu_n = m(w_n)=k(w_n)+ \sum \varepsilon_T \mu_n \] In fact the indexing set of trees $T$ for the sum is in canonical bijection with the non-top cells of the $n$-dimensional associahedron, and the sign $\varepsilon_T$ is just the dimension of the corresponding cell, so this is essentially a sum which calculates the Euler characteristic of the associahedron. The associahedron is contractible so we get \[ \mu_n = k(w_n) + \mu_n(1- (-1)^n). \] \item Each individual set of trees $\mathcal{T}_\pi$ has its summands in bijection, as in the previous case, with faces of associahedra. To be precise, in this case there is a product of associahedra, one for each vertex of $\pi_{i_0,\ldots, i_n}$ decorated by $*\ldots *$. As in the previous case, the signs of $k$ applied to these decorations are such that after applying $\boldsymbol{\epsilon}$ to the subtrees where each vertex is decorated by $*\ldots *$, what is obtained is a redecoration of the tree $\pi_{i_0,\ldots, i_n}$, now by $\End B$, as follows. \begin{enumerate} \item Vertices that were previously decorated by $*\ldots *$ are now decorated by the product in $\End B$, with no sign. \item Vertices that were previously decorated by \[a_{i_0}\underbrace{*\ldots *}_{\alpha_1}a_{i_1}\ldots a_{i_{j-1}}\underbrace{*\ldots*}_{\alpha_{j}}a_{i_j}\] (this description is slightly misleading because $\alpha_j\in \{0,1\}$) are now decorated by induction by \[(\kappa_{j})_{0,\alpha_0,\ldots, \alpha_{j},0}(a_{i_0}\otimes\cdots \otimes a_{i_j})\] as in the statement of the theorem, except for the following special case. \item the single tree $T_{\triv}$ is decorated by $k(w)$. \end{enumerate} Then by Lemma~\ref{lemma: moment-cumulant formula with insertions}, the equation $m(w)=\boldsymbol{\epsilon}\circ \mathcal{T}(k)\circ\boldsymbol{\eta}(w)$ is the same as the moment-cumulant formula for $E_{{i_0},{i_1},\ldots, {i_m},0}(a_1\otimes\cdots \otimes a_m)$, up to the difference \[ k(w)-(\kappa_m)_{i_0,i_1,\ldots,i_m}(a_1\otimes\cdots\otimes a_m). \] So these two expressions are equal, as desired. \item This is similar to the second case. Again, each set of trees $\mathcal{T}_\pi$ is in bijection with faces of products of associahedra and by the same trick one obtains a redecoration of $\pi_{i_0,\ldots, i_n}$. In this case the tree $T_{\triv}$ is not part of any $\mathcal{T}_\pi$ but instead is its own separate summand. Then in this case the equation $m(w)=\boldsymbol{\epsilon}\circ \mathcal{T}(k)\circ\boldsymbol{\eta}(w)$ is the same as the moment-cumulant formula for $E_{{i_0},{i_1},\ldots, {i_m},0}(a_1\otimes\cdots \otimes a_m)$, up to the difference \[ k(w) \] so $k(w)$ is zero, as desired. \end{enumerate} \end{proof} \subsection*{Concluding remarks} \begin{remark*} Equation~(\ref{equation: square}) in the preceding proof suggests a different interpretation of the main result. The maps $k$ and $m$ can be understood as maps of collections between the underlying collection of the cooperad $\textsf{coAs}_A$ and the underlying collection of the operad $\End B$. The space of maps of collections from a cooperad to an operad possesses a rich natural structure (see~\cite[6.4,10.2.3]{LodayVallette:AO} for details and notation). Apparently the relational equation can be expressed in terms of the convolution by the following expression: \[m=\sum_{n=1}^\infty k^{\circledcirc n}\approx\frac{k}{1-k}. \] \end{remark*} \begin{remark*} The entire paper could be modified to work with symmetric operads; in this case it would be reasonable to replace $\textsf{coAs}_A$ with a commutative version $\textsf{coCom}_A$. This would only make sense in the case that both $A$ and $B$ are commutative. As one might reasonably expect, this analagous procedure seems to describe the \emph{classical} cumulants as the canonical twist of the moments. As there are many direct combinatorial presentations~\cite{RotaShen:OCC} for classical cumulants and even a significantly more direct approach from the operadic point of view~\cite{DrummondColeParkTerilla:HPTII}, any details or verification have been omitted. \end{remark*} \bibliographystyle{amsalpha}
ba21779ca4df037124955b1d43a715e5c701dcfa
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The Navier-Stokes equations (NSEs) represent a formulation of the Newton's laws of motion for a continuous distribution of matter in a fluid state, characterized by an inability to support shear stresses, see \cite{Doering-Gibbon}. The NSEs allow to determine the velocity field and the pressure of fluids confined in regions of the space, and they are used to describe many different physics phenomena as weather, water flow in tubes, ocean currents and others. Moreover, these equations are useful in several fields of knowledge such as petroleum industry, plasma physics, meteorology, thermo-hydraulics, among others (see \cite{RT} for instance). Due to this fact, these equations have been attracted to the attention of several mathematicians since they play an important role for applications. See \cite{boldrini, Alexandre, Doering-Gibbon, GRR, GRR1, GRR2, Jiu-Wang-Xin, rosa, RT, Teman} and the references therein. On the other hand, the theory of impulsive dynamical systems has been shown to be a powerful tool to model real-world problems in physics, technology, biology, among others. Because of this fact, the interest in the study of impulsive dynamical systems has increasing considerably. For recents trends on this subject we indicate the works \cite{BonottoDemuner, bonotto1, BBCC, Cortes, Davis, Feroe, Yang, Zhao} and the references therein. However, the study of Navier-Stokes equations with impulse effects is really scarce. Motivated by this fact, in this paper, we investigate existence and uniqueness of mild solutions for the impulsive NSEs \begin{equation}\label{Eq5} \displaystyle\left\{\begin{array}{ll} \displaystyle\frac{\partial u}{\partial t} + q(t)(u \cdot \nabla)u - \nu\Delta u +\nabla p = \phi(t,u), & (t,x) \in \left((0, +\infty)\setminus \displaystyle\bigcup_{k=1}^{+\infty}\{t_k\}\right) \times \Omega, \vspace{1mm}\\ {\rm div}\, u = 0, & (t,x) \in (0, +\infty) \times \Omega, \vspace{1mm}\\ u = 0, & (t,x) \in (0, +\infty) \times \partial\Omega, \vspace{1mm}\\ u(0, \cdot)= u_0 & x \in \Omega, \vspace{1mm}\\ u(t_k^+, \cdot) - u(t_k^-, \cdot) = I_k (u(t_k, \cdot)), & x\in\Omega, \; k=1, 2,\ldots , \end{array} \right. \end{equation} where $\Omega$ is a bounded smooth domain in $\mathbb{R}^2$. Here $u = (u_1,u_2)$ denotes the velocity field of a fluid filling $\Omega$, $p$ is its scalar pressure and $\nu > 0$ is its viscosity. We will assume that $q$ is a bounded function, $\phi$ is a nonlinearity which will be specified later, $\{t_k\}_{k \in \mathbb{N}} \subset (0, +\infty)$ is a sequence of impulse times such that $\displaystyle\lim_{t \rightarrow +\infty} t_k = +\infty$, $u(t_k, \cdot) = u(t_k^+, \cdot) = \displaystyle\lim_{\delta \rightarrow 0+}u(t_k + \delta, \cdot)$, $u(t_k^-, \cdot) = \displaystyle\lim_{\delta \rightarrow 0+}u(t_k - \delta, \cdot)$ and $I_k$, $k \in \mathbb{N}$, are the impulse operators. Besides to impulsive actions in the system \eqref{Eq5}, we also allow that the external force $\phi$ is not continuous and depends on the solution $u$. We point out that the Navier-Stokes equations with impulses make sense physically and allow to describe more precisely the phenomena modeled by these equations, since $u$ represents the velocity of the field of a fluid and moreover, the external force $\phi$ in this case does not need to be continuous. It is well known that the phenomena which occur in the environment have impulsive behavior and the functions which model them have several discontinuities. Therefore, with this impulsive model, we intend to give a more precisely description of the Navier-Stokes equations. The system \eqref{Eq5} without impulse conditions was studied in the classical monograph \cite{Cheban}, where $\phi$ is a function of time $t\in\mathbb{R}$. More precisely, the author studies existence and uniqueness of global mild solutions for the non-impulsive equation \[ \displaystyle\frac{\partial u}{\partial t} + q(t)(u \cdot \nabla)u - \nu\Delta u +\nabla p = \phi(t), \] subject to the conditions $\textrm{div}\, u = 0$ and $u|_{\partial \Omega} = 0$, where $\Omega$ is a bounded smooth domain in $\mathbb{R}^2$. Our goal here is to write a weaker formulation of the system \eqref{Eq5} and then, we intend to investigate the existence and uniqueness of mild solutions. In order to do this, we start by considering some notations which can be found in \cite{rosa} and \cite{RT}, for instance. Let $\mathbb{L}^2 (\Omega) = (L^2 (\Omega))^2$ and $\mathbb{H}_0^1 (\Omega) = (H_0^1 (\Omega))^2$ endowed, respectively, with the inner products $$(u,v) = \displaystyle \sum_{j=1}^2 \int_{\Omega} u_j \cdot v_j \ dx, \ \ \ u = (u_1, u_2), \ v = (v_1, v_2) \in \mathbb L^2 (\Omega),$$ and $$((u,v)) = \displaystyle \sum_{j=1}^2 \int_{\Omega} \nabla u_j \cdot \nabla v_j dx, \ \ \ u = (u_1, u_2), \ v = (v_1, v_2) \in \mathbb{H}^1_0 (\Omega)$$ and norms $| \cdot | = ( \cdot, \cdot)^{1/2}$ and $\| \cdot \| = (( \cdot, \cdot))^{1/2}$. Now, we consider the following sets: $$\mathcal{E} = \{ v \in (C_0^{\infty}(\Omega))^2: \; \nabla \cdot v = 0 \ \textrm{in} \ \Omega\},$$ $$ V = \textrm{closure of} \ \mathcal{E} \ \textrm{in} \ \mathbb{H}_0^1 (\Omega)$$ and $$ H = \textrm{closure of} \ \mathcal{E} \ \textrm{in} \ \mathbb{L}^2 (\Omega).$$ The space $H$ is a Hilbert space with the scalar product $(\cdot, \cdot)$ induced by $\mathbb{L}^2 (\Omega)$ and the space $V$ is a Hilbert space with the scalar product $((u, v))$ induced by $\mathbb{H}_0^1 (\Omega)$. The space $V$ is contained in $H$, it is dense in $H$ and by the Poincare's Inequality, the inclusion $i:V \hookrightarrow H$ is continuous. Denote by $V'$ and $H'$ the dual spaces of $V$ and $H$, respectively. The adjoint operator $i^*$ is linear and continuous from $H'$ to $V'$, $i^*(H')$ is dense in $V'$ and $i^*$ is one to one since $i(V) = V$ is dense in $H$. Moreover, by the Riesz representation Theorem, we can identify $H$ and $H'$ and write $$V \subset H \equiv H' \subset V',$$ where each space is dense in the following one and the injections are continuous. As a consequence of the previous identifications, the scalar product in $H$, $(f,u)$, of $f \in H$ and $u \in V$ is the same as the duality product between $V'$ and $V$, $\langle f, u \rangle$, i.e., $$ \langle f, u \rangle = (f, u), \ \ \ \text{for all} \; f \in H \; \text{and} \; \text{for all} \; u \in V.$$ Also, for each $u \in V$, the form $$v \in V \mapsto \nu((u, v)) \in \mathbb R$$ is linear and continuous on $V$. Therefore, there exists an element of $V'$ which we denote by $Au$ such that $$ \langle Au, v \rangle = \nu((u, v)), \; \text{for all} \; v \in V.$$ Notice that the mapping $u \mapsto Au$ is linear, continuous and it is an isomorphism from $V$ to $V'$. Based on it, we consider the following weak formulation of \eqref{Eq5}: \begin{equation}\label{weak-Eq} \left\{ \begin{array}{lll} \displaystyle\frac{d}{dt} (u, v) + \nu ((u, v)) + b(t)(u,u,v) = \langle \phi(t,u), v \rangle, \ \ v \in V, \; t > 0, \; t \neq t_k, \vspace{1mm}\\ u(t_k) - u(t_k^-) = I_k(u(t_k^-)), \ \ k \in \mathbb N, \vspace{1mm}\\ u(0) = u_0 \in H, \end{array} \right. \end{equation} where $\phi(t,u) \in V'$ and $b(t): V \times V \times V \to \mathbb R$ is given by $$b(t) (u, v, w) = q(t) \displaystyle\sum_{i, j = 1}^2 \displaystyle\int_{\Omega} u_i \displaystyle\frac{\partial v_j}{\partial x_i} w_j dx.$$ The weak formulation \eqref{weak-Eq} is equivalent to the impulsive system \begin{equation}\label{IS} \left\{ \begin{array}{lll} u'+ Au + B(t)(u, u) = \phi, \ \ \ \textrm{in} \ \ V', \ \ t > 0, \; t \neq t_k,\vspace{1mm}\\ u(t_k) - u(t_k^-) = I_k(u(t_k^-)), \ \ k \in \mathbb{N}, \vspace{1mm}\\ u(0) = u_0 \in H, \end{array} \right. \end{equation} where $u'= du/dt$, $A: V \to V'$ is the Stokes operator defined by $$\langle Au, v \rangle = \nu((u,v)), \; \text{for all} \; u, v \in V,$$ and $B(t): V \times V \to V'$ is a bilinear operator defined by $$\langle B(t)(u, v), w \rangle = b(t) (u, v, w), \; \text{for all} \; u, v, w \in V.$$ In Section 2, we consider the following general impulsive system \begin{equation}\label{IntroNS1} \left\{ \begin{array}{lll} u' + Au + B(\sigma(\cdot,\omega))(u,u) = f(\cdot, \sigma(\cdot,\omega), u), \quad t > 0, \; t \in I, \; t \neq t_k, \; k\in\mathbb{N}, \vspace{1mm}\\ u(t_k) - u(t_k^-) = I_k (u(t_k^-)), \ \ k \in \mathbb{N},\vspace{1mm}\\ u(0) = u_0 \in H, \end{array} \right. \end{equation} where $f: I\times\mathcal{M}\times H\rightarrow H$ is a piecewise continuous function with respect to $t\in\mathbb{R}$, non-stationary and also depends on the solution $u$. All the conditions of system \eqref{IntroNS1} will be specified later. We prove the existence and uniqueness of global mild solutions for the system \eqref{IntroNS1} when $\mathcal{M}$ is compact, see Theorems \ref{T1.1} and \ref{T1.2}. The case when $\mathcal{M}$ is not compact is considered in Theorem \ref{T1.3}. In Section 3, we prove existence and uniqueness of global mild solutions for the 2D NSEs with impulses \eqref{Eq5} via system \eqref{IntroNS1}. We also give sufficient conditions to obtain dissipativity for the system \eqref{Eq5}. All the results from this paper hold for the non-impulsive case. \section{Preliminaries} Let $(\mathcal{M},d)$ be a compact metric space and $(\mathcal{M}, \mathbb R, \sigma)$ be a dynamical system on $\mathcal{M}$, i.e., $\sigma:\mathbb{R} \times \mathcal{M} \to \mathcal{M}$ is a continuous mapping which satisfies the following properties: \begin{itemize} \item[$i)$] $\sigma(0, \omega)= \omega$, \quad $\omega \in \mathcal{M}$; \item[$ii)$] $\sigma(s,\sigma(t,\omega))=\sigma(s+t,\omega)$, \quad $t,s \in \mathbb{R}$, $\omega \in \mathcal{M}$. \end{itemize} Let $H$ be a real or complex Hilbert space and $\mathcal{L}(H) = \{T: H \rightarrow H: \, T \; \text{is linear }$ $\text{and bounded}\}$ equipped with the operational norm. Let $A:D(A) \subset H \to H$ be a self-adjoint operator such that \begin{equation}\label{Coer} {\rm Re}\, \langle Au, u \rangle \geq a |u|^2_H, \end{equation} for all $u\in D(A)$ and $a > 0$. It follows by \cite[Lemma 6.20]{Alexandre} that $\mathbb{C}\setminus (-\infty, a]\subset \rho(A)$ (resolvent of $A$), and there exists a constant $M\geq 1$ such that $$ \|(\lambda - A)^{-1}\|_{\mathcal{L}(H)} \le \frac{M}{|\lambda -a |}, \quad \lambda \in \Sigma_{a,\varphi}, \, \lambda \neq a, $$ where $\Sigma_{a,\varphi}=\{\lambda \in \mathbb{C}: |{\rm arg}\, (\lambda-a)| \le \varphi\}$, $\varphi < \pi$. In particular, $A$ is sectorial. It follows that $-A$ generates an analytic semigroup $\{e^{-At}: t\geq 0\} \subset \mathcal{L}(H)$ which satisfies \begin{equation}\label{EQ0} \|e^{-At}\|_{\mathcal{L}(H)} \le K e^{-at}, \end{equation} for some constant $K > 0$, where $a > 0$ comes from \eqref{Coer}. Assuming that $0 \in \rho(A)$, we consider the scale of Hilbert spaces $X^\alpha=D(A^\alpha)$ of fractional power of the operator $A$ endowed with the norm $\|\cdot \|_{X^\alpha}= \|A^\alpha \cdot \|_H$ ($X^0=H$). If $ \beta > \alpha \geq 0$, it is well known that $X^\beta$ is a dense subspace of $X^\alpha$ with continuous inclusion and \begin{equation} \label{eq:fractpow} \|e^{-At}\|_{\mathcal{L}(X^\alpha, X^\beta)} \le C_{\alpha,\beta} \, t^{\alpha-\beta} e^{-at}. \end{equation} Consider $F$ a Hilbert space such that $H \subset F$ with inclusion dense and continuous. Denote by $\mathcal{L}(F, H)=\{T: F \rightarrow H: \, T \; \text{is linear and bounded}\}$ equipped with the operational norm. We will assume that the semigroup $\{e^{-At}: t\geq 0\}$ satisfies: \begin{enumerate} \item[$i)$] $e^{-At} \in \mathcal{L}(F, H)$, \quad for all $t >0$; \item[$ii)$] There exists $0 \le \alpha_1 <1$, such that \begin{equation}\label{eq:sg-est-F} \|e^{-At} \|_{\mathcal{L}(F, H)} \le K_1 t^{-\alpha_1} e^{-at}, \quad \text{for all}\; t >0. \end{equation} \end{enumerate} We denote by $\mathcal{L}^2 (H, F)$ the space of all continuous bilinear operators $\mathcal{B} : H \times H \to F$ equipped with the norm \[ \|\mathcal{B}\|_{\mathcal{L}^2 (H, F)} = \sup \{ |\mathcal{B}(u, v)|_F : |u|_H \leq 1, \ |v|_H \leq 1 \}. \] Let $C(\mathcal{M}, \mathcal{L}^2(H, F))$ be the space of all continuous mapping $B: \mathcal{M} \to \mathcal{L}^2 (H, F)$ endowed with the norm \[ \|B\|_{\infty} = \sup_{\omega \in \mathcal{M}} \|B(\omega)\|_{\mathcal{L}^2 (H, F)}. \] The space $\Big( C(\mathcal{M}, \mathcal{L}^2(H, F)), \|\cdot\|_{\infty} \Big)$ is a Banach space, see \cite{Cheban}. For all $u, v \in H$ and $\omega \in \mathcal{M}$, we have that \begin{equation}\label{EQ5} |B(\omega)(u, u) - B(\omega)(v, v)|_F \leq \|B\|_{\infty}(|u|_H + |v|_H ) |u - v|_H, \end{equation} and also \begin{equation}\label{EQ51} |B(\omega)(u,u)|_F \leq \|B\|_{\infty}|u|_H^2. \end{equation} Let $\{t_k\}_{k \in \mathbb{N}}$ be a strictly increasing sequence in $(0, +\infty)$ such that $\displaystyle\lim_{t \rightarrow +\infty} t_k = +\infty$. Let $I \subset \mathbb{R}$ be an interval and $f: I\times \mathcal{M}\times H \rightarrow H$ and $I_k: H \rightarrow H$, $k \in \mathbb{N}$, be functions satisfying the following conditions: \begin{enumerate} \item[(C1)] For each fixed $t \in I$, $f(t, \cdot, \cdot)$ is continuous on $\mathcal{M}\times H$. \item[(C2)] Let $\omega \in \mathcal{M}$ and $u \in H$. Then $\displaystyle\lim_{s \rightarrow t} f(s, \omega, u) = f(t, \omega, u)$ for all $t \neq t_k$, $k \in \mathbb{N}$, the limit $\displaystyle\lim_{s \rightarrow t_k-} f(s, \omega, u)$ exists and $\displaystyle\lim_{s \rightarrow t_k+} f(s, \omega, u) = f(t_k, \omega,u)$, for all $k \in \mathbb{N}$. \item[(C3)] There is a bounded function $M : \mathbb R \to \mathbb R_+$, such that for any interval $[a,b] \subset I$, we have $$\displaystyle \int_a^b |\phi(s)| | f(s, \omega, u)|_H ds \leq \displaystyle\int_a^b M(s) |\phi(s)| ds$$ for all $\phi \in L^1 [a,b]$, $\omega \in \mathcal{M}$ and $u \in H$. \item[(C4)] There is a bounded function $L: \mathbb{R} \rightarrow \mathbb{R}_+$, such that for any interval $[a, b] \subset I$, we have \[ \int_a^b |\phi(s)| |f(s, \omega_1, u_1) - f(s, \omega_2, u_2)|_H \, ds \leq \int_a^b L(s)|\phi(s)|(d(\omega_1, \omega_2) + |u_1 - u_2|_H)ds \] for all $\phi \in L^1[a, b]$, $\omega_1, \omega_2 \in \mathcal{M}$ and $u_1, u_2 \in H$. \item[(C5)] There exists a constant $K_2>0$ such that \[ \sup_{k\in\mathbb{N}}\sup_{u\in H}|I_k(u)|_H \le K_2. \] \item[(C6)] There exists a constant $K_{3}>0$ such that \[ |I_k(u) - I_k(v)|_H \le K_3 |u- v|_{H} \] for all $u,v \in H$ and for all $k\in\mathbb{N}$. \end{enumerate} Now, given $\omega \in \mathcal{M}$ and assuming all the conditions above, we consider the following impulsive system in the state space $H$: \begin{equation}\label{NS1} \left\{ \begin{array}{lll} u' + Au + B(\sigma(\cdot,\omega))(u,u) = f(\cdot, \sigma(\cdot,\omega), u), \quad t > 0, \; t \in I, \; t \neq t_k, \; k\in\mathbb{N}, \vspace{1mm}\\ u(t_k) - u(t_k^-) = I_k (u(t_k^-)), \ \ k \in \mathbb{N},\vspace{1mm}\\ u(0) = u_0 \in H. \end{array} \right. \end{equation} \begin{remark} Since $\displaystyle\lim_{k \rightarrow +\infty} t_k = +\infty$, it is clear that given a closed interval $[0, T]$, there exists at most a finite number of moments of impulses $t_1, t_2,\ldots, t_n \in [0, T]$ such that $0 < t_1 < t_2 < \ldots < t_n \leq T$. Thus, given $T > 0$ there is an integer $n_T > 0$ such that $t_{n_T} \leq T < t_{n_T+1}$. \end{remark} Given $T>0$, we consider the space $PC^+ ([0,T], H)= \{u:[0, T]\to H : \ u \text{ is continuous at }\linebreak t \neq t_k, \text{ right-continuous at } t=t_k \text{ and the limit } \displaystyle \lim_{t \to t_k^-} u(t) \;\text{exists for all} \, k = 1, \ldots, n_T \} $. It is well known that the space $PC^+ ([0,T], H)$ endowed with the norm $\displaystyle \|u\|_{PC^+}= \sup_{t \in [0,T]} |u(t)|_{H}$ is a Banach space. In the sequel, we present the definition of a mild solution for the system \eqref{NS1}. \begin{definition}\label{mild-solution} Let $[0, T]\subset I$. We say that $u \in PC^+ ([0,T], H)$ is a \emph{mild solution} of \eqref{NS1} if $u$ satisfies the following integral equation: \begin{equation}\label{NS2} u(t) = \displaystyle\left\{ \begin{array}{lcc} e^{-At} u_{0} + \displaystyle \int_{0}^t e^{-A(t-s)}g(s, \omega, u(s)) ds, & \text{if} & 0 \leq t < t_1, \\ e^{-A(t-t_1)}[u(t_1^-) + I_1(u(t_1^-))] +\displaystyle \int_{t_1}^t e^{-A(t-s)}g(s, \omega, u(s)) ds, & \text{if} & t_1 \leq t < t_2, \\ e^{-A(t-t_2)}[u(t_2^-) + I_2(u(t_2^-))] +\displaystyle \int_{t_2}^t e^{-A(t-s)}g(s, \omega, u(s)) ds, & \text{if} & t_2 \leq t < t_3, \\ & \vdots & \\ e^{-A(t-t_{k})}[u(t_{k}^-) + I_k(u(t_{k}^-))] +\displaystyle \int_{t_{k}}^t e^{-A(t-s)}g(s, \omega, u(s)) ds, &\text{if} & t_{k} \leq t \leq T, \end{array} \right. \end{equation} where $0 < t_1 < \ldots < t_k \leq T < t_{k+1}$ are the impulse times ($k=n_T$) and $g(s, \omega, u(s)) = -B({\sigma} (s,\omega))(u(s),u(s)) + f(s,{\sigma} (s, \omega), u(s))$, $s \in [0, T]$. System \eqref{NS2} can be rewritten in the following way \[ u(t) = e^{-At} u_{0} + \displaystyle\int_{0}^{t} e^{-A(t-s)}g(s, \omega, u(s))ds + \displaystyle\sum_{0 < t_i <t} e^{-A(t-t_i)}I_i(u(t_i^-)). \] \end{definition} Given $K \subset H$, we consider the following space of functions: \begin{align*} P&C^{+}_1([0,T]\times {K} \times \mathcal{M}, H) =\{ \varphi:[0,T]\times {K}\times \mathcal{M} \to H: \, \text{ for all } (u,\omega) \in {K} \times \mathcal{M}, \\ & \varphi(\cdot,u,\omega) \in PC^+( [0, T], H) \text{ and for all } t \in [0,T], \, \varphi(t,\cdot,\cdot): K \times \mathcal{M} \to H \text{ is continuous} \}. \end{align*} Theorem \ref{T1.1} ensures that the nonautonomous system \eqref{NS1} admits a unique mild solution in the sense of Definition \ref{mild-solution}. \begin{theorem}\label{T1.1} Let $u_0 \in H$ and $r >0$. Assume that \eqref{EQ0}, \eqref{eq:sg-est-F}, \eqref{EQ5} and conditions (C1) - (C6) hold. Then there exist positive numbers $\delta = \delta(u_0, r) > 0$, $T = T(u_0, r) > 0$ and a function $\varphi: [0, T]\times \overline{B(u_0, \delta)}\times \mathcal{M} \rightarrow H$ satisfying the following conditions: \begin{enumerate} \item[$i)$] $\varphi (0, u_0, \omega) = u_0$, for all $\omega \in \mathcal{M}$; \item[$ii)$] $| \varphi (t, u, \omega) - u_0|_H \leq r$ for all $(t,u,\omega) \in [0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M}$; \item[$iii)$] $\varphi \in PC^{+}_1([0,T] \times \overline{B(u_0, \delta)} \times \mathcal{M}, \overline{B(u_0, r)})$. \end{enumerate} Moreover, the function $u:[0,T] \to H$ defined by $u(t)=\varphi(t,u_0,\omega)$ is the unique mild solution of system \eqref{NS1}. \end{theorem} \begin{proof} Let $\delta > 0$ and $T > 0$ be such that $[0, T]\subset I$. Given $\varphi \in PC^{+}_1 ([0,T]\times \overline{B(u_0, \delta)}\times \mathcal{M}, H)$, we define \[ S\varphi(t, u, \omega) = e^{-At}u + \int_{0}^t e^{-A(t-s)} g(s,\omega,\varphi(s)) ds + \displaystyle\sum_{0 < t_i <t} e^{-A(t-t_i)}I_i(\varphi(t_i^-)), \] where $\varphi(s) = \varphi(s, u, \omega)$ and $g(s, \omega, \varphi(s)) = -B(\sigma(s,\omega))(\varphi(s),\varphi(s)) + f(s, \sigma(s, \omega), \varphi(s))$, for all $s\in [0, T]$, $u \in \overline{B(u_0, \delta)}$ and $\omega \in \mathcal{M}$. Since functions in $PC^{+}_1([0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M},\overline{B(u_0, r)})$ are bounded, we can consider the distance \[ d_{\infty}(\varphi_1, \varphi_2) = \sup\{|\varphi_1(t, u, \omega) - \varphi_2(t, u, \omega)|_H: 0 \leq t \leq T, \, u \in \overline{B(u_0, \delta)}, \, \omega \in \mathcal{M}\}, \] for $\varphi_1, \varphi_2 \in PC^{+}_1([0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M},\overline{B(u_0, r)})$. It is not difficult to see that $(PC^{+}_1([0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M},\overline{B(u_0, r)}), d_{\infty})$ is a complete metric space. For convenience, let us denote $\Gamma(\delta, T, r) = PC^{+}_1([0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M},\overline{B(u_0, r)})$ and $\Gamma(\delta, T) = PC^{+}_1([0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M}, H)$. \vspace{0.3cm} \textbf{Assertion 1:} $S \in C(\Gamma(\delta, T, r), \Gamma(\delta, T))$. \vspace{0.3cm} In fact, at first note that $S\varphi \in \Gamma(\delta, T)$ for all $\varphi \in \Gamma(\delta, T, r)$. Now, let $\varphi_1, \varphi_2 \in \Gamma(\delta, T, r)$ and $(t, u, \omega) \in [0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M}$. By Condition (C4) there is a bounded function $L: \mathbb{R} \rightarrow \mathbb{R}_+$ such that \[ \int_0^t e^{-a(t-s)} |f(s,\sigma(s, \omega), \varphi_1(s)) - f(s, \sigma(s, \omega), \varphi_2(s))|_H \,ds \leq \] \begin{equation}\label{EQ2.9} \leq \int_0^t L(s)|\varphi_1(s) - \varphi_2(s)|_H \,ds \leq N Td_{\infty}(\varphi_1, \varphi_2), \end{equation} where $N = \displaystyle\sup_{s \in [0, T]}|L(s)|$. Then, using \eqref{EQ0}, \eqref{eq:sg-est-F}, \eqref{EQ5}, \eqref{EQ2.9} and Condition (C6), we have \[ |S\varphi_1(t, u, \omega) - S\varphi_2(t, u, \omega)|_H \leq \] \[ \leq \displaystyle\int_{0}^{t} \left|e^{-A(t - s)} \left[B(\sigma(s, \omega))(\varphi_1(s), \varphi_1(s)) - B(\sigma(s, \omega))(\varphi_2(s), \varphi_2(s)) \right]\right|_H ds + \] \[ + \int_{0}^{t} \left|e^{-A(t -s)}\left[f(s, \sigma(s, \omega), \varphi_1(s)) - f(s, \sigma(s, \omega), \varphi_2(s))\right]\right|_H ds+ \] \[ + \displaystyle\sum_{0 < t_i <t} |e^{-A(t-t_i)}[I_i(\varphi_1(t_i^-)) - I_i(\varphi_2(t_i^-))]|_H \leq \] \[ \leq 2\|B\|_{\infty}K_1(r + |u_0|_H) d_{\infty}(\varphi_1, \varphi_2)\int_0^t(t-s)^{-\alpha_1}e^{-a(t-s)}ds + \] \[ +KNT d_{\infty}(\varphi_1, \varphi_2) + KK_3d_{\infty}(\varphi_1, \varphi_2)\sum_{0 < t_i <t} e^{-a(t-t_i)} \leq \] \begin{equation}\label{EQNS1} \leq \left(2\|B\|_{\infty}K_1(r + |u_0|_H)\dfrac{T^{-\alpha_1 + 1}}{-\alpha_1 + 1} + KNT + KK_3n_{T}\right)d_{\infty}(\varphi_1, \varphi_2), \end{equation} where $n_{T}$ is the number of impulses on the interval $[0, T]$. Hence, $S \in C(\Gamma(\delta, T, r), \Gamma(\delta, T))$. \vspace{0.3cm} \textbf{Assertion 2:} There are $\delta_1 = \delta_1(u_0, r) \in (0, \delta)$ and $T_1 = T_1(u_0, r)\in(0, T)$ such that $S: \Gamma(\delta_1, T_1, r) \rightarrow \Gamma(\delta_1, T_1, r)$. \vspace{0.3cm} In fact, let $\varphi \in \Gamma(\delta, T, r)$ and $(t, u, \omega) \in [0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M}$. By \eqref{EQ0} and Condition $(C4)$, one can obtain a bounded function $L: \mathbb{R}\rightarrow \mathbb{R}_+$ such that \[ \left| \int_{0}^t e^{-A(t-s)}\left[f(s, \sigma(s, \omega), \varphi(s)) - f(s, \sigma(s, \omega), 0)\right] ds \right|_H \leq \] \[ \int_0^t K e^{-a(t-s)} |f(s, \sigma(s, \omega), \varphi(s)) - f(s, \sigma(s, \omega), 0)|_Hds \leq \] \begin{equation}\label{EQ2.11} \leq K\int_0^t L(s) |\varphi(s)|_H ds \leq K N (|u_0|_H + r)T, \end{equation} where $N = \displaystyle\sup_{s \in [0, T]}|L(s)|$. Let $m(\delta, T) = \sup\left\{|e^{-At}u - u_0|_H: \, t \in [0, T], \; u \in \overline{B(u_0, \delta)}\right\}$ and $M = \displaystyle\sup_{s \in [0, T]} | M(s)|$, where $M$ is the function given by Condition (C3). Then, using \eqref{EQ0}, \eqref{eq:sg-est-F}, \eqref{EQ51}, \eqref{EQ2.11}, Condition (C3) and Condition (C5), we obtain $$ \left|S \varphi(t, u, \omega) - u_0 \right|_H \leq \left| e^{-At}u - u_0 \right|_{H} + \left| \int_{0}^t e^{-A(t -s)} B(\sigma(s, \omega))(\varphi(s), \varphi(s)) ds \right|_H+ $$ $$ + \left| \int_{0}^t e^{-A(t-s)} f(s, \sigma(s, \omega), \varphi(s)) ds \right|_H + \left|\displaystyle\sum_{0 < t_i <t} e^{-A(t-t_i)}I_i(\varphi(t_i^-))\right|_H \leq$$ \[ \leq m(\delta, T) + \int_{0}^t K_1e^{-a(t-s)} (t - s)^{-\alpha_1} \|B\|_{\infty} | \varphi(s)|^2_H ds + \] \[ + \left | \int_{0}^t e^{-A(t-s)}\left[f(s, \sigma(s, \omega), \varphi(s)) - f(s, \sigma(s, \omega), 0)\right] ds \right|_H + \left| \int_{0}^t e^{-A(t-s)}f(s, \sigma(s, \omega), 0)ds \right|_H + \] \[ + \sum_{0 < t_i <t} Ke^{-a(t-t_i)}|I_i(\varphi(t_i^-))|_H \leq \] \[ \leq m(\delta, T) + K_1\|B\|_{\infty} (|u_0|_H + r)^2 \frac{T^{-\alpha_1 + 1}}{1 - \alpha_1}+ KK_2n_T + \] \[ +KNT(|u_0|_H + r) + \int_{0}^t K e^{-a(t-s)} M(s) ds \leq \] \[ \leq m(\delta, T) + K_1\|B\|_{\infty} (|u_0|_H + r)^2 \frac{T^{-\alpha_1 + 1}}{1 - \alpha_1} + KN (|u_0|_H + r)T + \] \[ + KMT + KK_2n_T := d_1 (u_0, r, \delta, T). \] Now, we note that $d_1 (u_0, r, \delta, T) \rightarrow 0$ as $\delta \to 0$ and $ T \to 0$. Thus, there are $\delta_1 = \delta_1(u_0, r) > 0, \delta_1 < \delta$, and $T_1 = T_1(u_0, r) > 0, T_1 < T$, such that $d_1(u_0, r, \delta', T') \leq r$ for all $\delta' \in (0, \delta_1]$ and $T' \in (0, T_1]$. \vspace{0.3cm} \textbf{Assertion 3:} There exist $T_0 = T_0(u_0, r)>0$ and $\delta_0 = \delta_0(u_0, r)> 0$ such that \linebreak$S: \Gamma(\delta_0, T_0, r) \rightarrow \Gamma(\delta_0, T_0, r)$ is a contraction. \vspace{0.3cm} In fact, take $T_2 > 0$ such that \[ 2\|B\|_{\infty}K_1(r + |u_0|_H)\dfrac{T_2^{-\alpha_1 + 1}}{-\alpha_1 + 1} + KNT_2 + KK_3n_{T_2} < 1. \] It is enough to take $\delta_0 = \delta_1$ and $T_0 = \min\{T_1, T_2\}$ to conclude Assertion 3. In conclusion, by the Banach fixed point Theorem, there exists a unique function $\varphi \in \Gamma(\delta_0, T_0, r)$ satisfying the system \eqref{NS2} on the interval $[0,T_0]$ and the result follows. \end{proof} Theorem \ref{T1.2} gives sufficient conditions for the mild solution of system \eqref{NS1} to be prolongated on $\mathbb{R}_+$. \begin{theorem}\label{T1.2} Suppose that $I = \mathbb{R}_+$ and the conditions of Theorem \ref{T1.1} hold. If the mild solution $\varphi(t, u_0, \omega)$ of system \eqref{NS1} is bounded, then it may be prolonged on $\mathbb{R}_+$. \end{theorem} \begin{proof} By Theorem \ref{T1.1}, $\varphi(t, u_0, \omega)$ is the unique solution of system \eqref{NS1} passing through the point $u_0\in H$ at time $t=0$. This solution is defined on some maximal interval $[0, \alpha_{(u_0, \omega)})$. Let $\varphi(t, u_0, \omega)$ be bounded and suppose that $\alpha_{(u_0, \omega)} < \infty$. If $\alpha_{(u_0, \omega)} \neq t_k$ for all $k\in \mathbb{N}$, then defining $\varphi(\alpha_{(u_0, \omega)}, u_0, \omega) = \displaystyle\lim_{t\rightarrow \alpha_{(u_0, \omega)}-}\varphi(t, u_0, \omega)$, it follows that $\varphi(t, u_0, \omega)$ may be extend on the interval $[0, \alpha_{(u_0, \omega)}]$ which is a contradiction. Now, suppose that $\alpha_{(u_0, \omega)} = t_k$ for some $k\in\mathbb{N}$. Since the limit $\displaystyle\lim_{t\rightarrow t_k-}\varphi(t, u_0, \omega) = \varphi(t_k^-, u_0, \omega)$ exists and $I_k(\varphi(t_k^-, u_0, \omega)) \in H$, then we may use the proof of Theorem \ref{T1.1} and extend $\varphi(t, u_0, \omega)$ in some interval $[t_k, t_k + \epsilon)$, $\epsilon > 0$, with $\varphi(t_k, u_0, \omega) = \varphi(t_k^-, u_0, \omega) + I_k(\varphi(t_k^-, u_0, \omega))$ which is a contradiction. Hence, $\alpha_{(u_0, \omega)} = +\infty$. \end{proof} By following the proofs of Theorems \ref{T1.1} and \ref{T1.2}, we can state the next result which deals with existence and uniqueness of global mild solutions for the system \eqref{NS1} when $\mathcal{M}$ is not necessarily compact. \begin{theorem}\label{T1.3} Let $u_0 \in H$ and $r >0$. Assume that \eqref{EQ0}, \eqref{eq:sg-est-F}, \eqref{EQ5} and conditions (C1) - (C6) hold. Suppose that $\mathcal{M}$ is not necessarily compact and $\|B\|_{\infty} < \infty$. Then there exist positive numbers $\delta = \delta(u_0, r) > 0$, $T = T(u_0, r) > 0$ and a function $\varphi: [0, T]\times \overline{B(u_0, \delta)}\times \mathcal{M} \rightarrow H$ satisfying the following conditions: \begin{enumerate} \item[$i)$] $\varphi (0, u_0, \omega) = u_0$ for all $\omega \in \mathcal{M}$; \item[$ii)$] $| \varphi (t, u, \omega) - u_0|_H \leq r$ for all $(t,u,\omega) \in [0, T]\times \overline{B(u_0, \delta)} \times \mathcal{M}$; \item[$iii)$] $\varphi \in PC^{+}_1([0,T] \times \overline{B(u_0, \delta)} \times \mathcal{M}, \overline{B(u_0, r)})$. \end{enumerate} Moreover, the function $u:[0,T] \to H$ defined by $u(t)=\varphi(t,u_0,\omega)$ is the unique mild solution of system \eqref{NS1}. If $u(t)$ is bounded and $I=\mathbb{R}_+$, then it can be prolonged on $\mathbb{R}_+$. \end{theorem} \vspace{.2cm} \begin{remark} Suppose that \eqref{EQ0}, \eqref{eq:sg-est-F}, \eqref{EQ5} and conditions (C1) - (C4) hold. By the proofs of the previous results we obtain the existence and uniqueness of global mild solutions for the following non-impulsive system: \[ \left\{ \begin{array}{lll} u' + Au + B(\sigma(t,\omega))(u,u) = f(t, \sigma(t,\omega), u), \quad t > 0, \vspace{1mm}\\ u(0) = u_0 \in H. \end{array} \right. \] \end{remark} \section{The 2D Navier-Stokes equations with impulses} In this section, we present conditions to obtain the existence and uniqueness of global mild solutions for the following 2D Navier-Stokes equations with impulses \begin{equation}\label{l1} \displaystyle\left\{\begin{array}{ll} \displaystyle\frac{\partial u}{\partial t} + q(t)(u \cdot \nabla)u - \nu\Delta u +\nabla p = \phi(t, u), & (t,x) \in \left((0, +\infty)\setminus \displaystyle\bigcup_{k=1}^{+\infty}\{t_k\}\right) \times \Omega, \vspace{1mm}\\ {\rm div}\, u = 0, & (t,x) \in (0, +\infty) \times \Omega, \vspace{1mm}\\ u = 0, & (t,x) \in (0, +\infty) \times \partial\Omega, \vspace{1mm}\\ u(0, \cdot)= u_0(\cdot) & x \in \Omega, \vspace{1mm}\\ u(t_k, \cdot) - u(t_k^-, \cdot) = I_k(u(t_k^-, \cdot)), & x\in\Omega, \; k=1, 2,\ldots , \end{array} \right. \end{equation} where $\Omega$ is an open and bounded set in $\mathbb R^2$ with $\partial \Omega \in C^2$, $u = (u_1,u_2)$ is the velocity field of a fluid, $p$ is the scalar pressure, $\nu > 0$ is the kinematic viscosity of the fluid, $\phi=\phi(t, u) \in \mathbb R^2$ is the external body force, $q(t)$ is a bounded function, $\{t_k\}_{k \in \mathbb{N}} \subset (0, +\infty)$ is a sequence of impulses such that $\displaystyle\lim_{t \rightarrow +\infty} t_k = +\infty$ and $I_k$ is the impulse operator for each $k \in \mathbb N$. Let $$\mathcal{E} = \{u \in (C_0^{\infty}(\Omega))^2: \; \nabla \cdot u = 0 \ \textrm{in} \ \Omega\},$$ $$ V = \textrm{closure of} \ \mathcal{E} \ \textrm{in} \ \mathbb{H}_0^1 (\Omega)$$ and $$ H = \textrm{closure of} \ \mathcal{E} \ \textrm{in} \ \mathbb{L}^2 (\Omega),$$ where $\mathbb{L}^2 (\Omega) = (L^2 (\Omega))^2$ and $\mathbb{H}_0^1 (\Omega) = (H_0^1 (\Omega))^2$ are endowed, respectively, with the inner products \[ (u,v) = \displaystyle \sum_{j=1}^2 \int_{\Omega} u_j \cdot v_j \ dx, \ \ \ u = (u_1, u_2), \ v = (v_1, v_2) \in \mathbb L^2 (\Omega), \] and $$((u,v)) = \displaystyle \sum_{j=1}^2 \int_{\Omega} \nabla u_j \cdot \nabla v_j dx, \ \ \ u = (u_1, u_2), \ v = (v_1, v_2) \in \mathbb{H}^1_0 (\Omega),$$ and norms $| \cdot | = ( \cdot, \cdot)^{1/2}$ and $\| \cdot \| = (( \cdot, \cdot))^{1/2}$. We assume the following general hypotheses throughout this section: \begin{enumerate} \item[(H1)] $\phi: \mathbb{R}_+\times \mathbb{R}^2 \rightarrow \mathbb{R}^2$ is a bounded function such that for each fixed $t \in \mathbb{R}_+$, $\phi(t, \cdot)$ is continuous on $\mathbb{R}^2$. \item[(H2)] Let $x \in \mathbb{R}^2$. Then $\displaystyle\lim_{s \rightarrow t} \phi(s, x) = \phi(t, x)$ for all $t \neq t_k$, $k \in \mathbb{N}$, the limit $\displaystyle\lim_{s \rightarrow t_k-} \phi(s, x)$ exists and $\displaystyle\lim_{s \rightarrow t_k+} \phi(s, x) = \phi(t_k, x)$, for all $k \in \mathbb{N}$. \item[(H3)] There is $C > 0$ such that $|\phi(s, x) - \phi(s, y)| \leq C|x-y|$ for all $s \in \mathbb{R}_+$ and for all $x, y \in \mathbb{R}^2$. \item[(H4)] There exists a constant $C_1>0$ such that \[ \sup_{k\in\mathbb{N}}\sup_{x\in \mathbb{R}^2}|I_k(x)| \le C_1. \] \item[(H5)] There exists a constant $C_{2}>0$ such that \[ |I_k(x) - I_k(y)| \le C_2 |x- y| \] for all $x,y \in \mathbb{R}^2$ and for all $k\in\mathbb{N}$. \end{enumerate} Now, denote by $P$ the corresponding orthogonal projection $P: \mathbb{L}^2 (\Omega) \rightarrow H$ and set the operators \[ A = -\nu P \Delta \] and \[ \mathcal{B}(t)(u, v) = q(t)P( (u \cdot \nabla) v). \] It is well known that the Stokes operator $A$ is positive self-adjoint with domain $D(A)$ dense in $H$, $0 \in \rho(A)$ and $A^{-1}$ is compact. Also, there exists $\alpha>0$ such that \begin{equation}\label{JER1} \langle Au, u \rangle \geq \alpha |u|^2_H, \end{equation} for all $u \in H$. We also have the following orthogonality property of the nonlinear term which is fundamental and expresses the conservation of energy by the inertial forces: \begin{equation}\label{JER2} \langle \mathcal{B}(t)(u,v), v \rangle = 0 \end{equation} for all $u, v \in H$ and for all $t\in\mathbb{R}_+$. For the above properties see, for instance, \cite{Cheban}, \cite{Constantin} and \cite{Teman}. We set the Hilbert spaces $X^\alpha$, $\alpha \in(0, 1]$, as the domain of the powers of $A$ and we have \[ V = X^{\frac{1}{2}} \quad \text{and} \quad |u|_{V} = |\nabla u|. \] Applying $P$ in the equation \[ \displaystyle\frac{\partial u}{\partial t} + q(t)(u \cdot \nabla)u - \nu\Delta u +\nabla p = \phi(t, u), \] we obtain the evolution equation \begin{equation}\label{I2} u'+ Au + \mathcal{B}(t)(u, u) = \mathcal{F}(t, u), \end{equation} where $\mathcal{F}(t, u) = P\phi(t, u)$ for all $t>0$ and $u\in H$, $\langle Au, v \rangle = \nu((u, v))$ for all $u, v \in H$ and \[ \langle \mathcal{B}(t)(u, u), w \rangle = q(t) \displaystyle\sum_{i, j = 1}^2 \displaystyle\int_{\Omega} u_i \displaystyle\frac{\partial u_j}{\partial x_i} w_j dx \quad \text{for all} \quad u, w \in H. \] We also assume that: \begin{enumerate} \item[(A)] $\mathcal{F} \in PC^+(\mathbb{R}_+\times H, H)$; \item[(B)] $\mathcal{B} \in C(\mathbb{R}_+, L^2(H, F))$, where $F = D(A^{-\delta})$ for some $0 < \delta < 1$. \end{enumerate} Denote $Y$ by $C(\mathbb{R}_+, L^2(H, F))\times PC^+(\mathbb{R}_+\times H, H)$ and let $(Y, \mathbb{R}_+, \sigma)$ be the semidynamical system of translations, that is, $\sigma(t, g) = g_t$ for all $g \in Y$ and $t\geq 0$. Now, set \[ \mathcal{M} := \mathcal{H}(\mathcal{B}, \mathcal{F}) = \overline{\{(\mathcal{B}_{\tau}, \mathcal{F}_{\tau}): \, \tau \in \mathbb{R}_+\}}, \] where $\mathcal{B}_{\tau}(t) = \mathcal{B}(t + \tau)$ for all $t \in \mathbb{R}_+$ and $\mathcal{F}_{\tau}(t, u) = \mathcal{F}(t + \tau, u)$ for all $t \in \mathbb{R}_+$ and $u\in H$. If $(\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}}) \in \mathcal{M}$ and $\tau\geq 0$ we consider $\widetilde{\mathcal{B}}_{\tau}(t) = \widetilde{\mathcal{B}}(t+\tau)$ and $\widetilde{\mathcal{F}}_{\tau}(t, u) = \widetilde{\mathcal{F}}(t+\tau, u)$ for all $(t, u) \in \mathbb{R}_+\times H$. According to \cite{Cheban}, the equation \begin{equation}\label{L2} u' + Au + \widetilde{\mathcal{B}}(t)(u,u) = \widetilde{\mathcal{F}}(t, u), \end{equation} where $(\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}}) \in \mathcal{H}(\mathcal{B}, \mathcal{F})$, is called the $\mathcal{H}-$class along with the equation \eqref{I2}. Define the mapping $B: \mathcal{M} \rightarrow L^2(H, F)$ by \[ B(\omega) = B(\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}}) := \widetilde{\mathcal{B}}(0) \] and the mapping $f: \mathbb{R}_+\times\mathcal{M}\times H \rightarrow H$ by \[ f(t, \omega, u) = f(t, (\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}}), u) := \widetilde{\mathcal{F}}(0, u). \] Then equation \eqref{L2} can be rewritten in the form \begin{equation}\label{NSNA} u' + Au + B(\sigma(t, \omega))(u,u) = f(t, \sigma(t, \omega), u). \end{equation} From \eqref{JER2}, we obtain \begin{equation}\label{JER3} \langle B(\omega)(u,v), v \rangle = 0 \end{equation} for all $u, v \in H$ and for all $\omega \in \mathcal{M}$. \begin{lemma}\label{LEMMA1} $\displaystyle\sup_{\omega \in \mathcal{M}} \|B(\omega)\|_{\mathcal{L}^2 (H, F)} < \infty$. \end{lemma} \begin{proof} Since $q(t)$ is bounded, there is $L > 0$ such that $|q(t)| \leq L$ for all $t\geq 0$. Then \[ \sup_{\omega \in \mathcal{M}} \|B(\omega)\|_{\mathcal{L}^2 (H, F)} = \sup_{\omega\in\{(\mathcal{B}_{\tau}, \mathcal{F}_{\tau}): \, \tau \in \mathbb{R}_+\}}\|B(\omega)\|_{\mathcal{L}^2 (H, F)} =\sup_{\tau\in \mathbb{R}_+} \|\mathcal{B}(\tau)\|_{\mathcal{L}^2 (H, F)} = \] \[ =\sup_{\tau\in \mathbb{R}_+} \sup_{|u|_H\leq 1, \, |v|_H\leq 1}|\mathcal{B}(\tau)(u, v)|_F = \sup_{\tau\in \mathbb{R}_+} \sup_{|u|_H\leq 1, \, |v|_H\leq 1}|q(\tau)P( (u \cdot \nabla) v)|_F \leq \] \[ \leq L\sup_{|u|_H\leq 1, \, |v|_H\leq 1} \|(u \cdot \nabla) v) \|_{H} \leq L\sup_{|u|_H\leq 1, \, |v|_H\leq 1}\|u\|_H \|\nabla v \|_H \leq \] \[ \leq L\sup_{|u|_H\leq 1, \, |v|_H\leq 1} \|u\|_H \| v \|_V = L\sup_{|u|_H\leq 1, \, |v|_V\leq 1} \|u\|_H \| v \|_V \leq L. \] Therefore, we have the desired result. \end{proof} Since $\phi$ is bounded and we have Lemma \ref{LEMMA1}, we may consider in $\mathcal{M}$ the metric $d_{\mathcal{M}}$ given by \[ d_{\mathcal{M}}(\omega_1, \omega_2) = d_{\mathcal{M}}((\widetilde{\mathcal{B}}_1, \widetilde{\mathcal{F}}_1), (\widetilde{\mathcal{B}}_2, \widetilde{\mathcal{F}}_2)) = \|\widetilde{\mathcal{B}}_1 - \widetilde{\mathcal{B}}_2\|_{\infty} + \|\widetilde{\mathcal{F}}_1 -\widetilde{\mathcal{F}}_2\|_{PC^+_{\infty}}, \] where $\displaystyle \|\widetilde{\mathcal{F}}\|_{PC^+_{\infty}}= \sup_{(t, u) \in [0,+\infty)\times H} |\widetilde{\mathcal{F}}(t, u)|_{H}$ and $\|\widetilde{\mathcal{B}}\|_{\infty} = \displaystyle\sup_{\omega \in \mathcal{M}} \|\widetilde{\mathcal{B}}(\omega)\|_{\mathcal{L}^2 (H, F)}$. \vspace{.3cm} Let us consider the following impulsive system associated to \eqref{NSNA}: \begin{equation}\label{WeakEq} \left\{ \begin{array}{lll} u' + Au + B(\sigma(t, \omega))(u,u) = f(t, \sigma(t, \omega), u), \ \ u \in H, \; t > 0, \; t \neq t_k, \vspace{1mm}\\ u(t_k) - u(t_k^-) = I_k(u(t_k^-)), \ \ k \in \mathbb N, \vspace{1mm}\\ u(0) = u_0, \end{array} \right. \end{equation} which is a weak formulation of \eqref{l1}. In the next lines, we show that $f$ satisfies the conditions (C1), (C2), (C3) and (C4) presented in Section 2. This will help us to show that system \eqref{WeakEq} admits a unique global mild solution, see Theorems \ref{TeoNS} and \ref{TEO2.6} in the sequel. \begin{lemma}\label{LNS1} The mapping $f: \mathbb{R}_+\times\mathcal{M}\times H \rightarrow H$ satisfies the conditions (C1), (C2), (C3) and (C4). \end{lemma} \begin{proof} First, let us show that $f$ satisfies condition (C1). Let $t \in \mathbb{R}_+$ be fixed. Take $(\omega_n, u_n), (\omega_0, u_0) \in \mathcal{M}\times H$, $n=1,2,\ldots$, such that \[ d_{\mathcal{M}}(\omega_n, \omega_0) \rightarrow 0 \quad \text{and} \quad |u_n - u_0|_H \rightarrow 0 \] as $n \rightarrow +\infty$. Note that $\omega_0 = (\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}})$ and $\omega_n = (\widetilde{\mathcal{B}}_{n}, \widetilde{\mathcal{F}}_{n})$, $n=1,2,3,\ldots$. Moreover, \[ \|\widetilde{\mathcal{F}}_n -\widetilde{\mathcal{F}}\|_{PC^+_{\infty}} \rightarrow 0 \quad \text{as} \quad n\rightarrow +\infty. \] Now, since $(\widetilde{\mathcal{B}}_{n}, \widetilde{\mathcal{F}}_{n}), (\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}}) \in \mathcal{M}$, $n=1,2,3,\ldots$, then there are sequences $\{\tau_k^n\}_{k\in\mathbb{N}}$ and $\{s_k\}_{k\in\mathbb{N}}$ in $\mathbb{R}_+$ such that \[ \widetilde{\mathcal{F}}_{n}(r, u) = \lim_{k\rightarrow +\infty}\mathcal{F}_{\tau_k^n}(r, u) \quad \text{and} \quad \widetilde{\mathcal{F}}(r, u) = \lim_{k\rightarrow +\infty}\mathcal{F}_{s_k}(r, u) \] for each $(r, u) \in \mathbb{R}_+\times H$ and $n\in\mathbb{N}$. Then, \[ |f(t, \omega_n, u_n) - f(t, \omega_0, u_0)|_H = |\widetilde{\mathcal{F}}_n(0, u_n) - \widetilde{\mathcal{F}}(0, u_0)|_H = \lim_{k\rightarrow +\infty} |\mathcal{F}(\tau^n_ k, u_n) - \mathcal{F}(s_k, u_0)|_H \] \[ \leq \lim_{k\rightarrow +\infty} |\mathcal{F}(\tau^n_ k, u_n) - \mathcal{F}(\tau^n_ k, u_0)|_H + \lim_{k\rightarrow +\infty} |\mathcal{F}(\tau^n_ k, u_0) - \mathcal{F}(s_k, u_0)|_H \] \[ = \lim_{k\rightarrow +\infty} |P\phi(\tau^n_ k, u_n) - P\phi(\tau^n_ k, u_0)|_H + |\widetilde{\mathcal{F}}_{n}(0, u_0) - \widetilde{\mathcal{F}}(0, u_0)|_H \] \[ \leq \|P\|C|u_n - u_0|_H + |\widetilde{\mathcal{F}}_{n}(0, u_0) - \widetilde{\mathcal{F}}(0, u_0)|_H, \] where the last inequality follows by Condition (H3). Hence, $|f(t, \omega_n, u_n) - f(t, \omega_0, u_0)|_H \rightarrow 0$ as $n\rightarrow +\infty$. The Condition (H2) implies in Condition (C2). In order to show that Condition (C3) holds, we define the function $M : \mathbb R \to \mathbb R_+$ by $M(t) = \sup\{|\phi(s, u)|_H: \, s \in\mathbb{R}_+, u \in H\}$ for all $t\in\mathbb{R}$, which is well defined since $\phi$ is bounded by Condition (H1). Note that $M$ is constant. Given $\omega = (\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}})\in \mathcal{M}$, there is a sequence $\{r_n\}_{n\in\mathbb{N}}$ in $\mathbb{R}_+$ such that \[ f(t, \omega, u) = \widetilde{\mathcal{F}}(0, u) = \lim_{n\rightarrow +\infty}\mathcal{F}_{r_n}(0, u) = \lim_{n\rightarrow +\infty}P\phi(r_n, u), \] for all $t\geq 0$ and $u\in H$. Then, for $[a, b] \subset \mathbb{R}_+$, we have \[ \displaystyle\int_a^b |\psi(s)| | f(s, \omega, u)|_H ds = \int_a^b|\psi(s)||\widetilde{\mathcal{F}}(0, u)|_Hds \leq\displaystyle\int_a^b M(s)|\psi(s)|ds, \] for all $\psi \in L^1 [a,b]$, $\omega \in \mathcal{M}$ and $u \in H$. Finally, we need to verify the Condition (C4). Define $L: \mathbb{R} \rightarrow \mathbb{R}_+$ by $L(t) = C + 1$, $t\in\mathbb{R}$, where $C$ is the constant of the Condition (H3). Then, given $[a, b] \subset \mathbb{R}_+$, we have \[ \int_a^b |\psi(s)| |f(s, \omega_1, u_1) - f(s, \omega_2, u_2)|_Hds = \int_a^b|\psi(s)||\widetilde{\mathcal{F}}_1(0, u_1) - \widetilde{\mathcal{F}}_2(0, u_2)|_Hds \leq \] \[ \leq \int_a^b|\psi(s)|\left(|\widetilde{\mathcal{F}}_1(0, u_1) - \widetilde{\mathcal{F}}_1(0, u_2)|_H + |\widetilde{\mathcal{F}}_1(0, u_2) - \widetilde{\mathcal{F}}_2(0, u_2)|_H\right)ds \leq \] \[ \leq \int_a^b |\psi(s)|\left(C\|P\||u_1-u_2|_H + \|\widetilde{\mathcal{F}}_1 -\widetilde{\mathcal{F}}_2\|_{PC^+_{\infty}}\right)ds \leq \] \[ \leq \int_a^b L(s)|\psi(s)| (d_{\mathcal{M}}(\omega_1, \omega_2) + |u_1 - u_2|_{H})ds, \] for all $\psi \in L^1[a, b]$, $\omega_1 = (\widetilde{\mathcal{F}}_1, \widetilde{\mathcal{B}}_1), \omega_2 = (\widetilde{\mathcal{F}}_2, \widetilde{\mathcal{B}}_2) \in \mathcal{M}$ and $u_1, u_2 \in H$. \end{proof} \vspace{.3cm} \begin{lemma}\label{LNS222} $\sup\{|f(t, \omega, u)|_H: \, t\geq0, \omega \in\mathcal{M}, u\in H\}< \infty$. \end{lemma} \begin{proof} If $t\geq 0$, $u\in H$ and $\omega = (\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}})\in\mathcal{M}$, we have \[ |f(t, \omega, u)|_H = |\widetilde{\mathcal{F}}(0, u)|_H = \lim_{n\rightarrow +\infty}|\mathcal{F}_{\tau_n}(0, u)|_H = \lim_{n\rightarrow +\infty}|P\phi(\tau_n, u)|_H \leq \|P\|\eta \leq \eta, \] where $\eta>0$ is a bound of $\phi$ since it is bounded by Condition (H1). Hence, the result follows. \end{proof} By Lemma \ref{LNS222}, we may define $\|f\|_1 = \sup\{|f(t, \omega, u)|_H: \, t\geq0, \omega \in\mathcal{M}, u\in H\}$ From Theorem \ref{T1.3}, we have the following straightforward result of existence and uniqueness of mild solutions. \begin{theorem}\label{TeoNS} Under conditions (H1)-(H5), (A) and (B), the system \eqref{WeakEq} admits a unique mild solution $u: [0, T]\rightarrow H$ defined in some interval $[0, T]$ satisfying $u(0)=u_0$. \end{theorem} The mild solution of system \eqref{WeakEq} may be prolonged on $\mathbb{R}_+$, see Theorem \ref{TEO2.6}. \begin{theorem}\label{TEO2.6} Suppose that $\displaystyle\sum_{i=1}^{+\infty}|I_i(u)|_H = \Gamma < \infty$ for all $u \in H$. Then: \begin{enumerate} \item[$a)$] The solution $\varphi(t, u_0, \omega)$ of the impulsive system \eqref{WeakEq} is bounded and therefore, it may be prolonged on $\mathbb{R}_+$; \item[$b)$] $|\varphi(t, u_0, \omega)|_H \leq 2C(|u_0|_H) + \Gamma$, for all $t \geq 0$, $\omega \in \mathcal{M}$ and $u_0 \in H$, where \[ C(r) = \displaystyle\left\{ \begin{array}{ccc} r & \text{if} & r \geq \dfrac{\|f\|_1}{\alpha}, \\ \\ \dfrac{\|f\|_1}{\alpha} & \text{if} & r\leq \dfrac{\|f\|_1}{\alpha} \end{array} \right. \] and $\alpha$ is given by \eqref{JER1}. \end{enumerate} \end{theorem} \begin{proof} $a)$ Let $u_0\in H$ and $\omega \in \mathcal{M}$. By Theorem \ref{TeoNS}, there exists a unique solution $\varphi(t, u_0, \omega)$ of equation \eqref{WeakEq} passing through $u_0$ at time $t=0$ and defined on some interval $[0, T_{(u_0, \omega)})$. Let $\psi(t; 0, u_0, \omega)$ be the solution of the non-impulsive equation \eqref{NSNA} such that $\psi(0) = u_0$. Note that \[ \varphi(t, u_0, \omega) = \psi(t; 0, u_0, \omega) \quad \text{for} \quad t \in [0, T_{(u_0, \omega)}) \cap [0, t_1). \] Define $\eta(t) = |\psi(t; 0, u_0, \omega)|_H^2$ for $t \in [0, T_{(u_0, \omega)})\cap [0, t_1)$ and denote $\psi(t; 0, u_0, \omega)$ by $\psi(t)$. Using \eqref{JER1} and \eqref{JER3}, we obtain \[ \eta'(t) = 2 \langle \psi'(t), \psi(t) \rangle = - 2 \langle A \psi(t), \psi(t) \rangle - 2 \langle B(\sigma(t, w))(\psi(t), \psi(t)), \psi(t) \rangle + \] \[ + \ 2 \langle f(t, \sigma(t, \omega), \psi(t)), \psi(t)\rangle \leq -2\alpha|\psi(t)|_H^2 + 2\|f\|_1|\psi(t)|_H. \] Thus, $$\eta' \leq - 2 \alpha \eta + 2 \| f \|_1\eta^{1/2},$$ which implies that $\eta(t) \leq v(t)$ for all $t \in [0, T_{(u_0, \omega)})\cap [0, t_1)$, where $v(t)$ is an upper solution of $v' = -2\alpha v + 2\|f\|_1v^{1/2}$ such that $v(0) = \eta(0) = |u_0|^2_H.$ Then \[ \eta(t) \leq \left[ \left( |u|_H - \displaystyle\frac{\| f\|_1}{\alpha} \right) e^{-\alpha t} + \displaystyle\frac{\| f\|_1}{\alpha} \right]^2, \] that is, \[ |\varphi(t, u_0, \omega)|_H = |\psi(t; 0, u_0, \omega)|_H \leq \left(|u_0|_H - \dfrac{\|f\|_1}{\alpha}\right)e^{-\alpha t} + \dfrac{\|f\|_1}{\alpha}, \quad t \in [0, T_{(u_0, \omega)})\cap [0, t_1). \] Now, if $t \in [0, T_{(u_0, \omega)})\cap [t_1, t_2)$, we use the previous argument and we obtain \begin{eqnarray*} |\varphi(t, u_0, \omega)|_H & \leq & \left(|\psi(t_1; 0, u_0, \omega) + I_1(\psi(t_1; 0, u_0, \omega))|_H - \dfrac{\|f\|_1}{\alpha}\right)e^{-\alpha(t-t_1)} + \dfrac{\|f\|_1}{\alpha}\\ &\leq & \left(|u_0|_H - \dfrac{\|f\|_1}{\alpha}\right)e^{-\alpha t} + |I_1(\psi(t_1; 0, u_0, \omega))|_He^{-\alpha(t-t_1)} + \dfrac{\|f\|_1}{\alpha}. \end{eqnarray*} Continuing with this reasoning, if $t \in [0, T_{(u_0, \omega)})\cap [t_k, t_{k+1})$, we get \begin{equation}\label{EQ3.5} |\varphi(t, u_0, \omega)|_H \leq \left(|u_0|_H - \dfrac{\|f\|_1}{\alpha}\right)e^{-\alpha t} + \sum_{i=1}^{k}|I_i(\psi(t_i; t_{i-1}, u_{i-1}^+, \omega))|_H e^{-\alpha(t-t_i)} + \dfrac{\|f\|_1}{\alpha}, \end{equation} where we denote $t_0 = 0$, $u_0^+ = u_0$ and $u_i^+ = \psi(t_i; t_{i-1}, u_{i-1}^+, \omega) + I_i( \psi(t_i; t_{i-1}, u_{i-1}^+, \omega))$ for $i=1,2,\ldots$. Since there exists a finite number of impulses on the interval $[0, T_{(u_0, \omega)})$ and $\displaystyle\sum_{i=1}^{+\infty}|I_i(u)|_H = \Gamma < \infty$ for all $u \in H$, then by \eqref{EQ3.5} we get that $\varphi(t, u_0, \omega)$ is bounded. Consequently, by Theorem \ref{T1.3}, it can be prolonged on $\mathbb{R}_+$. $b)$ Note that $|\varphi(t, u_0, \omega)|_H \leq |u_0|_H + \dfrac{\|f\|_1}{\alpha} + \Gamma$, for all $t\geq 0$, $u_0\in H$ and $\omega \in \mathcal{M}$. Therefore, the result holds. \end{proof} The system \eqref{WeakEq} is called \textit{bounded dissipative} if there is a nonempty bounded set $B_0 \subset H$ such that for each bounded set $B \subset H$ there exists $T = T(B) > 0$ such that $\varphi(t, u_0, \omega) \in B_0$ for all $t\geq T$, $u_0\in B$ and $\omega\in\mathcal{M}$. In this case, $B_0$ is called a \textit{bounded attractor} for the system \eqref{WeakEq}. In the next result, we obtain dissipativity for the system \eqref{WeakEq}. \begin{theorem}\label{TEO2.7} Suppose that $\displaystyle\sum_{i=1}^{+\infty}|I_i(u)|_H = \Gamma < \infty$ for all $u \in H$. Then the system \eqref{WeakEq} is bounded dissipative. \end{theorem} \begin{proof} By the proof of Theorem \ref{TEO2.6}, we have \[ \displaystyle\lim_{t\rightarrow +\infty}\sup_{|u_0|_H \leq r, \; \omega \in \mathcal{M}}|\varphi(t, u_0, \omega)|_H \leq \dfrac{\|f\|_1}{\alpha} + \Gamma, \] for all $r > 0$. Hence, the set $B_0 = \left\{u \in H: \, |u|_H \leq \dfrac{\|f\|_1}{\alpha} + \Gamma\right\}$ is a bounded attractor for the system \eqref{WeakEq}. \end{proof} In the last result, we present an estimative between two solutions with different initial data in the same fiber $\omega \in \mathcal{M}$. \begin{theorem}\label{TEO2.8} Suppose that $\displaystyle\sum_{i=1}^{+\infty}|I_i(u)|_H = \Gamma < \infty$ for all $u \in H$. Let $r_0 = 2\dfrac{\|f\|_1}{\alpha} + \Gamma$. Then, for each $k \in \mathbb{N}$, we have \[ |\varphi(t, u_1, \omega) - \varphi(t, u_2, \omega)|_H \leq (1 + C_2)^ke^{-(\alpha - 2\|B\|_{\infty}r_0 - C)t}|u_1 - u_2|_H, \] for all $t \in [t_k, t_{k+1})$, $u_1, u_2 \in \overline{B(0, \alpha^{-1}\|f\|_1)}$ and $\omega \in \mathcal{M}$. \end{theorem} \begin{proof} Let $t \in \mathbb{R}_+$, $\omega \in \mathcal{M}$, $u_1, u_2 \in \overline{B(0, \alpha^{-1}\|f\|_1)}$ and define $\eta(t) = \psi(t; 0, u_1, \omega) - \psi(t; 0, u_2, \omega)$, where $\psi_i := \psi(t; 0, u_i, \omega)$ is the solution of equation \eqref{NSNA} without impulses defined on $\mathbb{R}_+$ and passing through $u_i$ at time $t=0$, $i=1,2$. By Theorem \ref{TEO2.6}, we have \[ |\psi_i(t)|_H \leq 2\dfrac{\|f\|_1}{\alpha} + \Gamma = r_0, \quad \text{for all} \quad t \in [0, t_1] \quad \text{and} \quad i=1,2. \] Given $\omega = (\widetilde{\mathcal{B}}, \widetilde{\mathcal{F}}) \in \mathcal{M}$, there is a sequence $\{s_n\}_{n\in\mathbb{N}}\subset\mathbb{R}_+$ such that $\widetilde{\mathcal{F}}(t, u) = \displaystyle\lim_{n\rightarrow+\infty}\mathcal{F}_{s_n}(t, u)$ for all $(t, u) \in \mathbb{R}_+\times H$. By the definition of $f$, we have \[ |f(t, \sigma(t, \omega), \psi_1) - f(t,\sigma(t, \omega), \psi_2)|_H = |\widetilde{\mathcal{F}}_t(0, \psi_1) - \widetilde{\mathcal{F}}_t(0, \psi_2)|_H = \] \[ = \lim_{n\rightarrow +\infty}|\mathcal{F}_{s_n}(t, \psi_1) - \mathcal{F}_{s_n}(t, \psi_2)|_H = \lim_{n\rightarrow +\infty}|P\phi(s_n + t, \psi_1) - P\phi(s_n + t, \psi_2)|_H \leq \|P\|C|\eta(t)|_H. \] Then, using \eqref{JER1} and the above estimative, we have \[ \dfrac{d}{dt}|\eta(t)|^2_H = -2\langle A\eta(t), \eta(t)\rangle + 2\langle B(\sigma(t, \omega))(\psi_2, \psi_2) - B(\sigma(t, \omega))(\psi_1, \psi_1), \eta(t)\rangle + \] \[ + 2\langle f(t, \sigma(t, \omega), \psi_1) - f(t,\sigma(t, \omega), \psi_2), \eta(t)\rangle \leq \] \[ \leq -2\alpha|\eta(t)|^2_H + 4r_0\|B\|_{\infty}|\eta(t)|_H^2 + 2C\|P\||\eta(t)|^2_H \] \[ \leq -2\left(\alpha - 2\|B\|_{\infty}r_0 - C\right)|\eta(t)|^2_H = -2\beta |\eta(t)|^2_H, \] for all $t\in [0, t_1]$, where $\beta = \alpha - 2\|B\|_{\infty}r_0 - C$ and $\|P\|\leq 1$. Hence, $|\eta(t)|^2_H \leq e^{-2\beta t}|\eta(0)|^2_H$, that is, \begin{equation}\label{EQ} |\psi(t; 0, u_1, \omega) - \psi(t; 0, u_2, \omega)|_H \leq e^{-\beta t}|u_1-u_2|_H \end{equation} for all $t\in [0, t_1]$. Thus, if $0\leq t < t_1$, we get \[ |\varphi(t, u_1, \omega) - \varphi(t, u_2, \omega)|_H = |\psi(t; 0, u_1, \omega) - \psi(t; 0, u_2, \omega)|_H \leq e^{-\beta t}|u_1-u_2|_H. \] Let $t_1 \leq t < t_2$ and $\eta_1(t) = \varphi(t, u_1, \omega) - \varphi(t, u_2, \omega) = \psi(t; t_1, u_1^+, \omega) - \psi(t; t_1, u_2^+, \omega)$, where $\psi(t; t_1, u_i^+, \omega)$ is the solution of \eqref{NSNA} such that $\psi(t_1) = u_i^+$ and $u_i^+ = \psi(t_1; 0, u_i, \omega) + I_1( \psi(t_1; 0, u_i, \omega))$, $i=1,2$. Following the steps above to show \eqref{EQ}, we obtain \[ |\eta_1(t)|_H \leq e^{-\beta (t - t_1)}|\eta_1(t_1)|_H, \quad \text{for all} \; \; t \geq t_1. \] On the other hand, we have \begin{eqnarray*} |\eta_1(t_1)|_H & = & |\psi(t_1; 0, u_1, \omega) + I_1(\psi(t_1; 0, u_1, \omega)) - \psi(t_1; 0, u_2, \omega) - I_1(\psi(t_1; 0, u_2, \omega))|_H \\ & \leq & |\psi(t_1; 0, u_1, \omega) - \psi(t_1; 0, u_2, \omega)|_H+ |I_1(\psi(t_1; 0, u_1, \omega)) - I_1(\psi(t_1; 0, u_2, \omega))|_H\\ & \leq & (1 + C_2)|\psi(t_1; 0, u_1, \omega) - \psi(t_1; 0, u_2, \omega)|_H\\ & \leq & (1 + C_2)e^{-\beta t_1}|u_1 - u_2|_H, \end{eqnarray*} where $C_2$ comes from Condition (H5). Consequently, \[ |\varphi(t, u_1, \omega) - \varphi(t, u_2, \omega)|_H \leq (1+C_2)e^{-\beta t}|u_1-u_2|_H, \quad \text{for all} \quad t_1 \leq t < t_2. \] Continuing with this process, if $t_k \leq t < t_{k+1}$ we get \[ |\varphi(t, u_1, \omega) - \varphi(t, u_2, \omega)|_H \leq (1+C_2)^ke^{-\beta t}|u_1-u_2|_H, \] for all $u_1, u_2\in \overline{B(0, \alpha^{-1}\|f\|_1)}$ and $\omega\in\mathcal{M}$. \end{proof} \vspace{.2cm} \begin{remark} Theorems \ref{TeoNS}, \ref{TEO2.6}, \ref{TEO2.7} and \ref{TEO2.8} also hold for the non-impulsive system \[ \left\{ \begin{array}{lll} u' + Au + B(\sigma(t, \omega))(u,u) = f(t, \sigma(t, \omega), u), \ \ u \in H, \; t > 0, \vspace{1mm}\\ u(0) = u_0, \end{array} \right. \] with the obvious adaptations. \end{remark}
c119738c8140213d35cc0c262585ef2842765f45
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Germanium is widely used as a detector material in experiments searching for a rare process like the interaction of weakly interacting massive particles (WIMPs)~\cite{dmreview}. It is possible to build detectors with very good energy resolution based on the measurement of the ionization produced in the particle interaction, or of the increase of temperature~\cite{gebolo}. In addition, the combination of the ionization and heat signals is a powerful tool to distinguish nuclear recoils from electron recoils. Moreover, the crystal-growing process used in the semiconductor industry purifies the material to a high level that matches well the stringent ultra-low radioactivity requirements of rare event searches. The potential of germanium detectors for achieving very low threshold below 1~keV is particularly attractive for searches of WIMPs with masses below 10~GeV/c$^{2}$. The background at energies below 20~keV in such a detector is thus of particular interest. Notably, the contribution from tritium beta decays may have a significant impact on the sensitivity of the next generation of these detectors. The crystallization process removes all cosmogenically-produced radioactive atoms, with the exception of unstable germanium isotopes like $^{68}$Ge (see below). Their populations grow back again when the crystal is kept above ground, and therefore exposed to cosmic rays and the associated hadronic showers. Short-lived isotopes decay rapidly as soon as the detectors are stored underground, deep enough to suppress the hadronic component of the cosmic rays~\cite{Farley2006451}. The isotopes that merit attention have lifetimes exceeding a few tens of days, since shorter-lived nuclei can be eliminated just by storing the detectors in the underground site for some reasonable time before starting data taking. The cosmogenic products that have the most noticeable effect on the low-energy spectrum recorded in germanium detectors are those that decay via electronic capture (EC). The capture is often followed by the emission of a $K$-shell X-ray with characteristic energy between 4 and 11~keV. $L$- and $M$-shell captures will produce weaker lines at approximately 1 and 0.1~keV, respectively. The sharp line shapes and known $K$:$L$:$M$ intensity ratios can be used to identify and subtract the associated events. However, it is preferable to reduce their initial intensities to the lowest possible level. Measurements of the production rates of EC-decaying isotopes is helpful in designing a detector-production procedure that limits these backgrounds to acceptable levels, and, more generally, to constrain models predicting the production rates of all isotopes, including those that may prove to elude direct measurements. Another type of background of particular interest is the beta decay of tritium ($^{3}$H) originated from nuclear reactions induced by the interaction of the hadronic component of cosmic rays with atoms in the material~\cite{avignone}. The electron emitted in the beta decay of tritium has an end point $Q_{\beta}$ of only 18.6~keV, and thus contributes to the background of low-energy events over the entire energy range relevant for low-mass WIMP searches. The lifetime of $^{3}$H is particularly long ($\tau$ = 17.79~y), so the tritium activity can essentially be expected to remain almost the same throughout the life of the detector. The only way to reduce this background is to limit the exposure of the crystal between the time it is grown and its installation underground. There are large uncertainties in model predictions for the production rate of $^{3}$H, and available measurements can only provide crude upper limits~\cite{mei}. The EDELWEISS collaboration has operated an array of 24 germanium heat-and-ionization detectors with the objective to perform searches for WIMPs with a total exposure close to 3000~kg$\cdot$d, and more specific searches for low-mass WIMPs with a subset of its detectors with the best experimental energy thresholds~\cite{lowmass}. The experiment is located in the Laboratoire Souterrain de Modane (LSM) and protected by a mean rock overburden of 1800~m (4800~m.w.e.) that reduces the cosmic ray flux to about 5~$\mu$/m$^2$/day~\cite{Schmidt:2013gdc}, i.e.\ 10$^6$ times less than at the surface. The detectors are covered by interleaved electrodes that provide an efficient tool to reject surface events (i.e.\ particle interactions taking place within $\sim$2 mm from the detector surface) down to energies of $\sim$1~keV~\cite{lowmass}. The resolutions achieved with these detectors, the reduction of the external gamma-ray background and the excellent surface-event rejection performance of the interleaved electrodes~\cite{broniatowski}, have made possible a precise measurement of decay rates of different nuclei in the bulk volume of germanium detectors, and in particular, for the first time, to measure unambiguously the intensity of the tritium spectrum. Efforts were made to keep to a minimum the exposure of each crystal to cosmic rays throughout the detector production. A history of the key steps in the detector production process is available. Despite this, there are non-negligible systematic uncertainties in the recorded history of exposure times. However, these uncertainties can be tested, because unforeseen production delays\footnote{ These delays occurred to solve problems related to surface current leakage, as described in \protect{\cite{leakage}}.} led to a relatively large spread (up to a factor of 4) in the exposures of the different detectors to cosmic rays. It was therefore possible to check on isotopes with the largest statistics that the observed activation rates scaled with the expectations from the recorded history of exposure times. In the following, we detail the EDELWEISS-III setup relevant to this measurement (Section~\ref{sec:setup}), as well as the data selection (Section~\ref{sec:data}). We present the expected properties of the activation of tritium and other isotopes and of the energy spectrum of the emitted electrons, and describe the analysis used to extract their intensities from the data~(Section \ref{sec:analysis}). These results are then converted into production rates during the exposure above ground (Section~\ref{sec-history}) and compared to a previous measurement and calculations (Section~\ref{sec:pr}). \section{Experimental setup}\label{sec:setup} The active target of the EDELWEISS-III WIMP search experiment consists of twenty-four 800-g {\em Fully-InterDigit} (FID) germanium detectors, cooled down to an operating temperature of 18 mK. All materials surrounding the detectors have been tested for their radiopurity~\cite{scorza}. The cryostat is surrounded by 20~cm of lead and 50~cm of polyethylene shielding. A constant flow of de-radonized air in the vicinity of the cryostat reduces the radon level down to 30 mBq/m$^3$. The shielding is surrounded by a 100 m$^2$ muon veto made of plastic scintillator modules with a geometrical coverage of more than 98\%~\cite{Schmidt:2013gdc}. The detectors are high-purity germanium cylindrical crystals of 70 mm in diameter and 40 mm in height. All surfaces are entirely covered with interleaved Al electrodes, biased at alternate values of potentials. The electrodes are 150 $\mu$m in width and separated by 2 mm. The potentials applied to the electrodes are chosen to determine an axial electric field in the bulk of the detector~\cite{broniatowski}, while in the volume within about 2~mm from the surfaces the electric field is parallel to them. As a consequence, electron-hole pairs created in the bulk volume are collected in the axial field by the fiducial electrodes on both sides of the detector, while surface events will be collected by adjacent electrodes. This scheme extends also to the cylindrical surfaces. Fiducial (or bulk) events can thus be selected on the basis of the presence of signals of opposite signs on the fiducial electrodes on each side of the detector, and on the absence of signals on all other electrodes. The fiducial ionization $E_{fid}$ is defined as the average of the signals on the two fiducial electrodes. Calibrations with a $^{210}$Pb source in equilibrium with its Bi and Po daughters have shown that this technique can reject surface $\beta$'s, $\alpha$'s and $^{206}$Pb recoils with an efficiency of the order of 99.999\%~\cite{surfacerejection}. Nuclear recoils are identified by comparing the ionization signal with the $\sim\mu K$ rise in temperature accompanying each interaction, measured with two 15~mm$^{3}$ germanium NTD (Neutron Transmutation Doped) thermistors glued on each side of the detector. The charge and temperature signals are calibrated using a $^{133}$Ba $\gamma$-ray source: the units of both signals are thus in keV-electron-equivalent (keV$_{ee}$). Since this work is devoted to the study of bulk electron recoil populations, the subscript $ee$ will be omitted in the following. The signals on both thermistors are combined into a single heat measurement $E_c$. The average dispersion in ionization signals $E_{fid}$ at 0 keV is characterized by $\sigma_{fid}$ = 230 eV (baseline resolution). For the heat signal $E_c$, the corresponding values of $\sigma_{c}$ range from 150 to 500~eV depending on detectors. The resolution increases with the energy of the signal, with a linear term that is dominated by charge trapping effects~\cite{quentin}. At 356 keV, the resolutions $\sigma(E)$ are larger than the baseline $\sigma_{fid}$ and $\sigma_{c}$ values and are approximately 4 keV. The data acquisition triggers if the amplitude of one of the heat signals is above a threshold value\footnote{The online filtering on the heat signals is less efficient than the offline version, resulting in a slight degradation of resolution. Here, $E_c$ and $\sigma_{c}$ refer to the offline-filtered heat signals. As the trigger is applied to the signals calculated online, some $E_c$ values may be less than their corresponding online trigger threshold by as much as $\sigma_{c}$.}. These values are automatically adjusted every minute for each heat sensor according to the event rates recorded minute by minute. This, together with the 0.5 kHz sampling of the heat channels, results in the possibility to set the thresholds to levels that are close to $\sim$4.5$\sigma_{c}$, while keeping the trigger rates per detector at approximately 50 mHz. \section{Data selection}\label{sec:data} \subsection{Data set and detector selection} The data was recorded over a period of 280 calendar days, from July 2014 to April 2015. During that period, 160 days were devoted to WIMP searches. All twenty-four detectors were used to define coincidences between detectors. The average heat trigger threshold and baseline resolutions were monitored hour by hour. For this analysis, only hours when this trigger threshold is below 2 keV were selected. The hourly fiducial baseline resolution on the fiducial ionization measurement was also required to be less than 400 eV. Two detectors with a failing ionization channel, preventing an efficient application of the fiducial cut, were excluded from this exposure. Three detectors had less than one day of running time with an online threshold of below 2 keV and were also excluded from further analysis. The dead-time corrected exposure of the remaining 19 detectors is 1853 detector$\cdot$day. This exposure is considered for different global fits used in order to evaluate systematic uncertainties. As each detector had a different history of exposure to cosmic rays, the final results will be based on the fits to individual detector data for the 13 detectors with individual exposure greater than 60 days, corresponding to 87.0\% of the total exposure. Similarly, a sub-sample of events with an online threshold below 0.8 keV is also defined, for precision tests of the efficiency correction and of the sample purity at energies lower than used in the final analysis. This sample corresponds to 499 detector$\cdot$day, to which 10 detectors contribute. \subsection{Event selection} \label{sect-neg} An event is included in the analysis if its $E_{fid}$ ($E_c$) value is larger than 3.5$\sigma_{fid}$ ($3\sigma_{c}$). The values of $\sigma_{fid}$ and $\sigma_{c}$ are measured hour by hour from Gaussian fits to the distribution of amplitudes observed in events that have not triggered the online threshold. Cuts are also performed on the $\chi^2$ of the fit of template pulse shapes to the ionization signals, resulting in an efficiency loss of less than 1\%. Fig.~\ref{fig-datasel}a) shows the distribution of the fiducial ionization $E_{fid}$ as a function of the heat signal $E_c$ for selected events in a sample where the cut on $E_{fid}$ has been relaxed from $3.5\sigma_{fid}$ to $3\sigma_{fid}$. On this figure, events in gray are those rejected by the fiducial cut. Namely, events are considered as having occurred in the fiducial volume if the signals on the two non-fiducial electrodes, as well as the difference between the two fiducial electrodes, are each consistent with noise within $\pm$2.574$\sigma$ (99\%). The rejected events are mostly distributed along the two dashed lines on Fig.~\ref{fig-datasel}a) representing the expected location for surface beta events (blue dotted line) and surface gamma events (red dotted line). Fig.~\ref{fig-datasel}b) shows the same distribution after the above-described selection of fiducial events. The events in gray are those where the values of $E_{fid}$ and $E_c$ are not consistent with each other within 2.574$\sigma$ of the experimental resolutions. They are associated with the background of heat-only events discussed in Ref.~\cite{lowmass}. Their origin is illustrated in Figs.~\ref{fig-datasel}c) and d), where the low-energy part is shown together with events selected with cuts calculated using $|E_{fid}|$ instead of $E_{fid}$, namely, $|E_{fid}|$ $>$ 3$\sigma_{fid}$ and the consistency cut is applied on ($E_c-|E_{fid}|$). For clarity, events with trigger threshold below 0.8 keV (Fig.~\ref{fig-datasel}c) are shown separately from those with trigger thresholds between 0.8 and 1.5~keV (Fig.~\ref{fig-datasel}d). The distribution of $E_{fid}$ for heat-only events is a Gaussian centered at $E_{fid}=0$. In Figs.~\ref{fig-datasel}c) and d), what is therefore observed after the rejection of events with $|E_{fid}| <3 \sigma_{fid}$ are two populations of events, symmetric in $\pm~E_{fid}$. A leakage from the population of surface gamma and beta events would be centered around the red and blue dotted lines, respectively, on Fig.~\ref{fig-datasel}c) and d), and not distributed symmetrically at $\pm E_{fid}$. Such a symmetry is clearly observed for the events in gray, corresponding to those rejected by the consistent cut on $E_{c}-|E_{fid}|$. Most of the potential backgrounds below 2 keV thus comes from heat-only events. The population of heat-only events with $E_{fid}>0$ that are not rejected by the consistency cut will be estimated by mirroring the equivalent population with $E_{fid}<0$. For instance, in Fig.~\ref{fig-datasel}c) and for $E_c<$0.8 keV, the population of gray points with $E_{fid}<0$ is nearly equal to those with $E_{fid}>0$. This is also true for the events in black, indicating that most events with positive ionization energies and $E_c$ values below 0.8 keV remaining after applying the consistency cut are due to the tail of heat-only events. % Figure~\ref{fig-datasel}c) suggests that subtracting a sideband corresponding to events with $E_{fid}<$0 should confirm the clear observation of the peak of events at $\sim$1.2~keV due to the capture of $L$-shell electrons in Ga, Ge and Zn atoms (Table~\ref{tab-decay}). The efficiency of the entire analysis procedure at low energies has been tested (see Section~\ref{sec-yield}) by comparing the $L/K$ intensity ratio obtained with the known theoretical values, in the data sample with an online trigger threshold below 0.8 keV. It can also be observed on Fig.~\ref{fig-datasel}c) that the 1.2 keV peak may extend to the red dashed line corresponding to surface gamma events. This effect will be taken into account when calculating the energy dependence of the efficiency in Section~\ref{sec-eff}. It can also be observed from Fig.~\ref{fig-datasel} that the background above 2 keV is expected to be very small. \subsection{Selection of multiple-hit events} \label{sec-coinc} As can be deduced from the electron capture decay data listed in Table~\ref{tab-decay}, events associated with $^{49}$V, $^{55}$Fe, $^{68}$Ga and $^{68}$Ge are expected to produce an $L$- or $K$-shell energy deposit in a single FID detector. The 18.6 keV electron emitted in the beta decay of tritium is also expected not to escape the fiducial volume of the detector where it occurs. Backgrounds unrelated to these decays can be reduced by rejecting events where more than one detector have triggered simultaneously. An event is considered as a single if no other detectors have triggered within an interval of 10~ms, or if the sum of the ionization energies of all other detectors having triggered in this interval is less than 1~keV. However, Table~\ref{tab-decay} also shows that events associated with $^{65}$Zn (and $^{54}$Mn) are accompanied by the emission of a $\gamma$ ray with an energy of 1115.5~keV (834.9~keV) with a branching ratio of 50.0\% (100\%). This $\gamma$ ray (or internal conversion electron) may be detected simultaneously in the same detector, in which case no $L$- or $K$-shell peaks are observed. A high-energy $\gamma$ ray can also escape the FID fiducial volume and be detected in one of the 23 other detectors. Thus, some of the $^{54}$Mn and $^{65}$Zn decays are associated with coincident events. As a result, the line intensity for these events will be obtained by a simultaneous fit of the spectra of single- and multiple-hit events. In addition, the intensity for $^{65}$Zn will be corrected for the fraction of unobserved events at 8.98 keV due to the absorption of the 1115.5~keV $\gamma$-ray inside the detector itself, as estimated by Monte Carlo simulation, accounting for 25$\pm$1\% of all $^{65}$Zn decays. The $^{54}$Mn line is included in the peak model, but its intensity is too weak to extract a reliable rate. \subsection{Energy spectrum} \label{sec-eff} The energy of an event is taken as the average of $E_c$ and $E_{fid}$ weighted by the inverse of the square of their associated baseline resolutions $\sigma_{c}$ and $\sigma_{fid}$, as measured hour-by-hour. A side-band spectrum corresponding to the contribution of heat-only events, obtained by replacing $E_{fid}$ by $-E_{fid}$ in the weighted average, as described in Section~\ref{sect-neg}, has been subtracted from the resulting spectrum. The spectrum is then corrected for the energy dependence of the efficiency due to {\em i)} the fiducial volume cut, {\em ii)} the cut on $E_{fid}>$3.5$\sigma_{fid}$, and {\em iii)} the online trigger threshold on the heat energy\footnote{ The cut on $E_c>$3$\sigma_{c}$ has a negligible effect on the efficiency once the effect of the online trigger threshold is taken into account.}. The fiducial mass value per detector~\cite{lowmass} is measured with the $K$-line intensities, and thus effectively includes an inefficiency due to the rejection of events where the apparent signal on a given electrode comes from a gaussian fluctuation of its baseline noise. As the energy of the event approaches zero keV, the expected signal on the non-fiducial electrodes for surface events becomes smaller than the baseline noise, and the fiducial cut can no longer reject them. The selected volume becomes the entire detector, and the effective volume increases from 70\% to 100\% of the detector. Fig.~\ref{fig-datasel}c) and d) hint at the presence of this effect as the distribution of events around the peak at 1.2 keV extends out to the line where surface gamma rays are expected. This energy dependence is calculated using the measured baseline resolutions on the signal on the non-fiducial electrodes and on the difference between those on the two fiducial electrodes. This efficiency correction is applied to the detector energy spectra, but as a convention, the fiducial exposures in all the following will be quoted for a $\gamma$ ray of 10 keV. The energy-dependence of the efficiency induced by the 3.5$\sigma_{fid}$ cut on $E_{fid}$ and the online heat threshold have been calculated for each detector using the hour-by-hour measurements of both values, and taking into account the smearing effect due to the difference between the online and offline reconstruction of $E_c$. The calculated corrections are very close to 1 over the entire energy range down to the analysis threshold, set at 2 keV. The validity of these corrections are thoroughly tested by applying them to the sample of events where the online threshold is below 0.8 keV. The test consists of a comparison of the measured and theoretical $L/K$ intensity ratio for this data sample. The model predicts that the $L/K$ efficiency ratio varies from 0.92 to 1.01 between 1.1 and 1.3 keV (the energies of the $L$-shell lines for $^{65}$Zn and $^{68}$Ge, respectively: see Table~\ref{tab-decay}). The resulting efficiency-corrected spectrum for the 1853 detector$\cdot$day sample is shown in Fig.~\ref{fig-spectrum}. The inset shows the efficiency-corrected spectrum for the data set with an online trigger threshold below 0.8 keV (499 detector$\cdot$day), used to test the efficiency model and the heat-only sideband correction down to 1 keV. The electron capture and the tritium beta-decay intensities described below are taken from the fit of the sample of events with the 2 keV threshold cut. \section{Decay rate measurements}\label{sec:analysis} The decay rates of the different cosmogenically produced isotopes are obtained from simultaneous fits of the energy spectra of single and multiple hit events, as defined in Section~\ref{sec-coinc}. The fit is first applied to the co-added spectra of Fig.~\ref{fig-spectrum} in order to test its validity and estimate global systematics, and then to individual detector spectra in order to obtain individual count rates per kilogram and per day for each of the isotopes of interest. The fit model has three components: {\em i)} the tritium spectrum, {\em ii)} a Compton background and {\em iii)} $K$- and $L$- spectral lines. \subsection{Tritium beta spectrum} The energy spectrum of emitted electrons has an end-point value of $Q_{\beta}$~=~18.6 keV. It is described by~\cite{Brown:1995xn}: \begin{equation} \frac{dN}{dt} \propto \sqrt{T^2+2mc^2T} \; (T+mc^2) \; (Q_{\beta}-T)^2 \; F(T,Z=2) \end{equation} \noindent where $T$ is the kinetic energy, $m$ is the mass of the electron, $c$ the speed of light and \mbox{$F(T,Z=2)$} is the Fermi function for tritium decay. This function can be approximated in the non-relativistic limit as $x/(1-e^{-x})$, where $x=\frac{4\pi\alpha c}{v}$, $\alpha$ is the fine structure constant and $v$ is the electron velocity. With this Fermi function, and $T$ expressed in keV, the spectral shape becomes: \begin{equation} \frac{dN}{dt} \propto (T+mc^2) \; (Q_{\beta}-T)^2 \; (1-e^{-\frac{1.466}{\sqrt{T}}})^{-1} \label{eqn-shape} \end{equation} This function is used to describe the tritium component in the fit. The tritium intensity is taken as the integral of the fitted component from zero to 18.6 keV. The spectral shape of Eq.~(2) was not smeared with the detector-dependent energy resolutions, as it was verified that this procedure has an impact only below the analysis threshold, and in a region around the end point that has a negligible statistical weight in the determination of the intensity. \subsection{Compton background} All the peaks visible below 12~keV on Fig.~\ref{fig-spectrum} are due to the electron capture decays of different isotopes (mainly Ge, Ga and Zn, see Table~\ref{tab-decay}) that will be detailed in Section~4.3. Between 20 and 50~keV, the Compton background is constant within statistical errors, with an average of 0.090 $\pm$ 0.002~counts/kg/day/keV. Studies of the Compton plateau below 50 keV in $^{133}$Ba calibrations and GEANT4 Monte Carlo simulations also indicate that this flat behavior between 20 and 50 keV does not hint to a possible slope below 20 keV. Nevertheless, variations as much as $\sim$10\% of the Compton rate between 0 and 20 keV cannot be excluded, but a similar slope should also be present in multiple-hit events. The energy spectrum of multiple-hit events is also shown on Fig.~\ref{fig-spectrum}. Apart from the expected contributions of the $^{65}$Zn and, possibly, $^{54}$Mn peaks (discussed below), the spectrum between 2 and 50 keV appears to be relatively consistent with a flat Compton plateau. The statistical significance of this sample is sufficient to provide an alternative template for the shape of this background. It can be noted that rejecting multiple events reduces the Compton background by almost a factor of two while having no effect on tritium decays. In the following, the tritium and EC line intensities are obtained from a fit that includes a flat Compton background. As the fitted tritium intensity is somewhat more sensitive to the assumed shape of the Compton continuum, a systematic uncertainty is taken into account as the difference between this result and the intensity obtained with different background assumptions, namely {\em i)} a first order polynomial and {\em ii)} the multiple-hit spectrum scaled by the ratio of the number of counts between 20 and 40~keV in the single and multiple-hit spectra. The first test is performed for each individual detector. The second test requires more statistics and is only applied to the spectrum of Fig.~\ref{fig-spectrum}. \subsection{Electron capture lines} The model for the peaks in the spectrum is not only important for getting the EC decay intensities but also to constrain the amplitude of the underlying tritium component. The structure of peaks considered in the fit is the following. The energy-dependent width of a peak is given by $\sigma(E)$ = $\sqrt{\sigma^2_0+(\alpha E)^2}$, where $\alpha$ is a free parameter and $\sigma_0$ is fixed to the value derived from the measured average baseline resolutions $\sigma_{fid}$ and $\sigma_{c}$: $\sigma_0=1/\sqrt{\sigma^{-2}_{fid}+\sigma^{-2}_{c}}$. The global energy scale is also a free parameter of the fit. Table~\ref{tab-decay} lists all the electron capture (EC) decays with $t_{1/2}>$ 10~days, and their daughter decays, for 21$\leq Z \leq$33. All these can potentially lead to peaks at the $K$-shell binding energy $E_K$ in a germanium detector. The detectors are fully efficient to electrons, positrons and all forms of radiation down to a few eV. The energies of all prompt ($<1$~ms) cascades of radiation are summed up into the same event, so only pure EC decays can produce a peak at energy $E_K$. By applying the fiducial selection, we ensure that the radiation is emitted at least a few millimeters below the detector surface, and only $\gamma$ rays with energies above some hundred keV can escape from being summed up with the X-ray energies. Since the detectors have been installed at LSM between 6 and 18 months before the run start in July 2014, only isotopes with half-lives longer than 100 days need to be considered. An exception is $^{71}$Ge. This nucleus has a half-life of 11~days, but it is produced by $(n,\gamma)$ absorption when the detector is exposed to a neutron source. This decay produces a peak at the same energy $E_K$ = 10.37~keV as the EC decay of $^{68}$Ge. Consequently, some care must be taken to exclude data following neutron calibrations when extracting the $^{68}$Ge line intensity. The EC decay of $^{68}$Ge produces a daughter with a half-life $t_{1/2}$ = 67.7~min, $^{68}$Ga, that can also undergo an EC decay to $^{68}$Zn with a line at $E_K$ = 9.66~keV with a branching ratio of 11.12\%. This value should correspond to the ratio of the peak intensities from $^{68}$Ga and $^{68}$Ge EC decays. Other observable peaks should be those at the $E_K$ values of the EC decay of $^{65}$Zn, $^{55}$Fe, $^{54}$Mn and $^{49}$V. Therefore, peaks at those energies are included in the fit, together with those at 9.66 and 10.37~keV. No peak associated to the EC decay of $^{57}$Co is expected to be observed at $E_K$ = 7.11~keV in our detectors, since this decay is accompanied with 122 and 136~keV $\gamma$-rays that are easily absorbed inside the fiducial volume and the $\sim$2~mm layer of Ge that surrounds it. The same is true for $^{44}$Ti decays, unobservable in the fiducial volume of the detector because of the accompanying 78.3 and 67.9~keV transitions. However, the EC decay of its daughter $^{44}$Sc, with a half-life $t_{1/2}$ = 4.0~h, is in principle observable at $E_K$ = 4.04~keV, as the accompanying 1157.0~keV $\gamma$-ray may escape detection inside the detector. The total spectrum of Fig.~\ref{fig-spectrum} was fitted including peaks at $E_K$ = 4.04 and 7.11 keV. The resulting intensities were consistent with zero within their statistical uncertainties, and had no significant impact on the fitted tritium intensities. The heights of these two peaks have been fixed to zero for all subsequent fits. The $L$-shell peak structure only needs to be included in the fit of the data sample with the $<$0.8~keV online threshold requirement (inset of Fig.~\ref{fig-spectrum}). This multiplet structure is included in the fit as peaks at 1.10, 1.19 and 1.30~keV (Table~\ref{tab-decay}) with intensities taken as those of the peaks at 8.98, 9.66 and 10.37~keV scaled by the same $L/K$ ratio, which is a free parameter of the fit. \subsection{Fit to the spectra and uncertainties} The simultaneous fit to the single- and multiple-hit data shown in Fig.~\ref{fig-spectrum} have relatively good $\chi^2/d.o.f.$ values (1.12 and 1.03 for the fit to the data with an online trigger threshold below 2 and 0.8~keV, respectively). The fitted number of tritium counts varies by 1\% whether the Compton background is assumed to be flat or to be a first order polynomial. Changing the binning of the data has a similar effect. A larger difference appears if the Compton shape is taken from that of the multiple-hit spectra instead of being assumed to be flat. In this case the fitted number of tritium counts decreases by 11\%. This value is considered as a systematic uncertainty of the fitted tritium intensity associated with the uncertainty in the shape of the Compton background spectrum at low energy. The fitted $L/K$ ratio of the intensities of the peaks shown in the inset of Fig.~\ref{fig-spectrum} is 0.113 $\pm$ 0.008 (statistical error only), and agrees well with the theoretical value of 0.11~\cite{PhysRev.132.362}. It can also be interpreted that the efficiency correction recovers at least 93\% (90\% C.L.) of the full efficiency at an energy of 1.25 keV, combination of the three $E_L$ peaks of Table~\ref{tab-decay}. The efficiency model described in Section~\ref{sec-eff} is more robust for energies above 2.0 keV, as turning on or off entirely each of the three elements in the efficiency model changes the fitted tritium intensity by less than 2\%. The associated systematics is neglected as it is much smaller than the 11\% uncertainty described in the previous paragraph. \subsection{Activity measurements} \label{sec-yield} The fit of the entire 2~keV threshold data set results in an observed tritium activity of 0.94~$\pm$ ~0.06~(stat.)~$\pm$ ~0.10 (syst.)~events per day and per detector, extrapolated down to 0~keV using Eq.~(\ref{eqn-shape}). However, the rate can be expected to vary from detector to detector. The fit is therefore performed for each detector individually. The tritium rate is obtained from the fit with the flat Compton background. The systematic uncertainty associated with the Compton shape is taken as the quadratic sum of the common 11\% contribution discussed in the previous section and the difference between the fitted value with the flat and linear Compton backgrounds in individual detectors. The same systematic tests applied to the peak intensities showed that they are not significantly affected by uncertainties in the shape of the Compton background. The $^{68}$Ge rate is obtained from a fit that excludes the 90 days following the calibration of the detector with an AmBe neutron source. Fig.~\ref{fig-time} shows that this cut should essentially remove all contaminations of the $^{68}$Ge peak by $^{71}$Ge decays. The $^{68}$Ga/$^{68}$Ge ratio derived from the data outside this interval is 0.117 $\pm$ 0.006 and is compatible with the value of 0.1112 from Table~\ref{tab-decay}. Fig.~\ref{fig-time} also shows that the decay rates of the $^{68,71}$Ge and $^{65}$Zn decrease with the proper time constant, while the tritium rate is constant, as expected from its very long half-life. As the total length of the data taking period is close to one year, the decay rate of isotopes with half-lives $t_{1/2}$ <~3~y can not be considered constant. Then, the activity of an isotope with lifetime $\tau= t_{1/2} / \ln 2$ has been obtained from a fit where each event is weighted by $\exp(\frac{t-t_{ref}}{\tau})$, where $t_{ref}$ corresponds to the middle of the data-taking period, i.e.\ October 30$^{th}$, 2014. The fitted activity is thus the activity on this date. The resulting decay rates per detector are listed in Table~\ref{tab:rates}. The measured rates are expressed in terms of events per kilogram and per day, using the specific fiducial mass of each detector. The activities for $^{65}$Zn are calculated including the count rates observed in both single- and multiple-hit events (see Section~\ref{sec-coinc}) and they are corrected for the effect of the 1115.5~keV gamma to obtain total activities. Monte Carlo simulations have shown that the probability for this gamma to escape the fiducial volume without interacting is 50.1\%. Given the 50.0\% branching ratio for the emission of this gamma, it can be estimated that 25\% of all $^{65}$Zn EC decays do not produce a visible 8.98 keV peak, and the observed rate is corrected accordingly. \section{Exposure history} \label{sec-history} The population of cosmogenically produced radioactive isotopes in germanium is expected to be reduced to a negligible level during the crystallization process, except for $^{68}$Ge. The activation occurs while the crystal is stored above ground, induced mostly by high-energy neutron interactions. After a time $\Delta t$ = $t$$-t_{g}$ after crystal growth happening at time $t_{g}$, the isotope activity per unit mass $a(t)$ increases as: \begin{equation} \label{eq:1} a(t) = a_{max} \; (1-e^{-\frac{t-t_g}{\tau}}) \end{equation} \noindent where $a_{max}$ is the maximal activation, obtained when it becomes equal to the production rate, and $\tau$ is the lifetime. The value of $a_{max}$ for a given isotope depends on the flux of cosmic rays and the cross section for its production (including decay chains). Once the detector is shielded from cosmic rays by a significant thickness of rock so further activation can be neglected, the decay rate changes as: \begin{equation} \label{eq:2} a(t) = a_{max} \; (1-e^{-\frac{t_s-t_g}{\tau}}) \; e^{-\frac{t-t_s}{\tau}} \end{equation} \noindent where $t_{s}$ is the time at which the detector is screened from cosmic rays. A precise history detailing the succession and the length of the periods of time during which the detector is exposed or protected from cosmic rays is thus essential to evaluate isotope activities at a given time. The crystals have been produced gradually from September 2011 to December 2012, and they have been installed at the LSM in three batches: in January 2013, in September 2013 and in January 2014. The time periods of exposure to cosmic rays and underground storage of the Ge detectors are listed per detector in Table~\ref{tab:history}. The germanium crystals were first exposed above ground in a period $t_{exp1}$ starting from their production by UMICORE in Belgium, during their polishing at BSI in Latvia, up to their shipment to different shallow underground sites at CEA in France. The shallow sites were at least 10~m.w.e. deep. Considering the attenuation length of the neutron flux of 148~g/cm$^{2}$~\cite{ziegler}, such depths result in a reduction of activation by a factor 850. Cosmogenic production during the time $t_{dec1}$ when the crystals were stored in the shallow sites is therefore neglected. % The second exposure time $t_{exp2}$ occurs when the detector is shipped from the shallow sites to the nearby laboratories of CSNSM and CEA, where electrodes are evaporated, the surfaces are etched with XeF$_2$~\cite{leakage}, the NTD heat sensors are glued to them, and the detectors are shipped to an underground site. This site is either the LSM, or, for three of the detectors, one of the shallow sites, where they were stored prior to a grouped shipment to LSM. The last decay period $t_{dec2}$ corresponds to the interval between their arrival at LSM and the date of October 30$^{th}$, 2014 ($t_{ref}$). For the three detectors stored in the shallow site before their shipment to LSM, this period is included in $t_{dec2}$, and the short shipment time is included in the systematic uncertainty. Confirmed dates are those of the crystal production, the end of their polishing, the start and end of the electrode evaporation, the XeF$_2$ etching processes, the glueing of the NTD, and of their arrival at the LSM underground laboratory. Tracking of detector shipments and in-and-out of the shallow sites are not documented as precisely. This is taken into account by ascribing a $\pm$7 day uncertainty on the duration of each shipment, resulting in the systematic uncertainty shown in Table~\ref{tab:history}. This uncertainty is anti-correlated between two consecutive periods. The time periods $t_{exp1}$ and $t_{dec2}$ are bound by only one transport, and therefore have a smaller uncertainty. For two of the detectors, FID827 and FID828, there is a significantly larger uncertainty on the time interval between detector fabrication and shipment to underground site. Considering the succession of exposure times ($t_{exp1}$ and $t_{exp2}$) and cooling-off periods ($t_{dec1}$ and $t_{dec2}$), and using Eqs.~(\ref{eq:1}) and (\ref{eq:2}) the decay rate observed on October 30$^{th}$, 2014 ($t_{ref}$) in a detector ($\frac{dN}{dt}$) is related to the production rate ($P$) by: \begin{equation}\label{eq:pr} \frac{dN}{dt} = \nonumber\\ P \times [(1-e^{\frac{-t_{exp1}}{\tau}})(e^{\frac{-(t_{dec1}+t_{exp2}+t_{dec2})}{\tau}}) + (1-e^{\frac{-t_{exp2}}{\tau}})(e^{\frac{-t_{dec2}}{\tau}})] \end{equation} The expression in brackets in Eq.~(\ref{eq:pr}) corresponds to the saturation fraction $f_{s}(t_{ref})$, i.e.\ the activation at the time $t_{ref}$ expressed as a fraction of $a_{max}$, for a given detector history. Table~\ref{tab:saturationfraction} lists $f_{s}$ values calculated from Table~\ref{tab:history}, together with their systematic uncertainties $\sigma(f_{s})$. The calculation of $f_s$ can also be done at any other time $t$ between $t_g$ and $t_{ref}$. This is what is shown on Fig.~\ref{fig:satfrac} for $^{65}$Zn (top) and $^{3}$H (bottom), for the two detectors FID844 (left) and FID827 (right). The time axis starts at crystallization and ends on the reference day $t_{ref}$. The red lines correspond to the history described in Table~\ref{tab:history}, while the blue dotted lines are calculated using the $\pm$1$\sigma$ uncertainties. The detector FID844 has a standard history, where the two exposures $t_{exp1}$ and $t_{exp2}$ are separated by a relatively long cooling-down period $t_{dec1}$. This period is sufficiently long compared to the mean lifetime of $^{65}$Zn that the activity at $t_{ref}$ is dominated by the effect of the last exposure $t_{exp2}$, and consequently the uncertainty on $f_{s}(t_{ref})$ is dominated by that on $t_{exp2}$. This is not the case for $^{3}$H ($t_{dec1}\ll \tau$), where both exposure times $t_{exp1}$ and $t_{exp2}$ contribute equally to $f_{s}(t_{ref})$. FID827 is an example of a different history, where $t_{dec1}$ is also small compared to the $^{65}$Zn lifetime. As a first consistency test of the detector history model with the observed count rates, the correlation factor between the measured rate and the saturation fraction of $^{68}$Ge and $^{65}$Zn, for which measurements provide higher statistics, were evaluated. The correlation factors, calculated as Pearson product-moment correlation coefficient, are 0.90 and 0.83, respectively. This is a good indication that the observed count rate variations from detector to detector are indeed caused by the known history differences between them. In the final analysis (next section), the validity of this correlation is tested taking into account the uncertainties of the measurements and detector history. \section{Production rates}\label{sec:pr} \subsection{Experimental Results} Figures~\ref{fig:tritium} to \ref{fig:ge68} show the measured decay rates of Table~\ref{tab:rates}, as a function of the saturation fraction of Table~\ref{tab:saturationfraction}, for the thirteen FID detectors. Each plot corresponds to a different isotope. The data are fitted with a first degree polynomial: $y = P\cdot~x$, where $x$ is the saturation fraction, $y$ is the measured decay rate and the coefficient $P$ is the production rate of this particular isotope. The fits are shown by solid red lines. Errors in the count rates and saturation fraction have been taken into account. The fit parameters $P$ corresponding to the best estimates of the production rate are listed in Table~\ref{tab:pr}, together with the $\chi^2/d.o.f.$ of the fit. The final production rate values and their total errors are listed in Table~\ref{tab:results}. Most fits result in good reduced $\chi^2$ values, except $^{65}$Zn and $^{3}$H. To investigate this further, the fits were repeated with different scaling factors $S$ applied to the systematic uncertainty in the saturation fraction. The reduced $\chi^2$ values for $^{65}$Zn and $^{3}$H are equal or less than one for $S=2$. We therefore include as an additional systematic uncertainty in the production rate of these two isotopes the difference between the results with $S$ = 1 and $S$ = 2. The final production rate values and their total errors are listed in Table~\ref{tab:results}. Additional checks for consistency of the results have been carried out. Three detectors (FID826, 827 and 828) have been removed from the analysis because they were stored at a shallow site after being fully assembled but before being shipped to LSM, which could induce additional uncertainty relative to other detectors that were shipped directly to LSM. Production rates per isotopes agree within statistical errors with initial results. Although no activation is assumed to occur at the shallow sites, we repeated the analysis considering an activation rate equivalent to either 2\% or 5\% of the value on the surface. This resulted in small changes in the saturation fraction at $t=t_{ref}$ which are within the systematic uncertainty arising from the shipment history alone. \subsection{Calculated production rates} Figures~\ref{fig:tritium} to \ref{fig:ge68} also show the lines corresponding to production rate estimates from different models. The production rate, $P_{i}$, of a radioactive isotope $i$ can be calculated as follows: \begin{equation}\label{eq:pr1} P_{i} = \sum_{j} N_{j} \int\phi(E)\sigma_{ij}(E) dE \end{equation} where $N_{j}$ is the number of target nuclear isotope $j$, $\sigma _{ij}$ is the excitation function of isotope $i$ produced by neutrons on stable isotopes of material $j$, and $\phi$ is the cosmic neutron flux. Estimates of the production rate have been carried out with the ACTIVIA code~\cite{activia}. The code does not include proton activation of materials; the hadronic part of the cosmic-ray spectrum at the surface is dominated by neutrons~\cite{ziegler}. The original ACTIVIA code uses the neutron energy spectrum at the surface from~\cite{ziegler} but, as shown in Ref.~\cite{Zhang:2016rlz}, this spectrum does not match recent experimental data~\cite{gordon}. We carried out the calculations of production rates by replacing the original neutron spectrum with that from Ref.~\cite{gordon}. These calculations are performed using a cosmic ray neutron spectrum in the northern hemisphere parametrized as follows~\cite{gordon}: \begin{eqnarray}\nonumber\label{eq:nflux} \phi(E) &=& 1.006 \times 10^{-6}e^{-0.35\ln^{2}E+2.1451\ln{E} } \\ && + 1.011 \times 10^{-3}e^{-0.4106\ln^{2}E-0.667\ln{E}} \end{eqnarray} where $E$ is the neutron kinetic energy in MeV and $\phi$ is in units of cm$^{-2}$~s$^{-1}$~MeV$^{-1}$. We have also used semi-empirical cross sections from Refs.~\cite{silb,silb1,silb2,silb3,silb4} and the cross sections from the MENDL-2P libraries~\cite{mendl}, shown, respectively, as dashed and dash-dotted light-blue lines in Figs.~\ref{fig:tritium} to~\ref{fig:ge68}. The results of the calculations are listed in the third column of Table~\ref{tab:results}. They are labeled (a) or (b) if they use the semi-empirical cross sections~\cite{silb,silb1,silb2,silb3,silb4} or the MENDL-2P~\cite{mendl} database, respectively. For $^{49}$V both calculations give the same production rate. Our calculations with ACTIVIA, quoted in the second column of Table~\ref{tab:results}, show that excitation functions may account for up to a factor of 2 difference in the production rate of $^{68}$Ge. The difference increases with the atomic number of the isotope produced. Effects on the production rates caused by differences in cross sections and cosmogenic neutron energy spectra in ACTIVIA and GEANT4 codes~\cite{geant} have been estimated in Ref.~\cite{Zhang:2016rlz}. There, the neutron spectrum from~\cite{gordon} was taken as measured values for New York location whereas in the present work the parameterization~\cite{gordon} of these data has been used as given by Eq.~(\ref{eq:nflux}). The two approaches (data vs parameterization of these data) differ by no more than 2\%. Comparing production rates from calculations in Ref.~\cite{ceb}, listed in the third and fourth columns of Table~\ref{tab:results}, we can conclude that different input cosmic-ray neutron spectra, from Refs.~\cite{ziegler} and \cite{gordon}, can lead to a variation in production rates of about 20-30\% for the isotopes considered here. \subsection{Comparison with previous measurements and models} Different model estimates have been compared to the measurements presented in the previous sections. In Figs.~\ref{fig:tritium} to~\ref{fig:ge68}, the EDELWEISS data and the best fit to these data are compared to the ACTIVIA calculations performed within this work considering semi-empirical~\cite{silb,silb1,silb2,silb3,silb4} (dashed light-blue line) and MENDL-2P~\cite{mendl} (dash-dotted light-blue line) cross sections. Calculations from previous works are also shown in dashed lines: Cebrian et al.~\cite{ceb} (green), Klapdor-Kleingrothaus et al.~\cite{KK} (blue), and Mei et al.~\cite{mei} using TALYS cross sections (purple), the update of this work from Zhang et al.~\cite{Zhang:2016rlz} (olive-green) and the estimates from Avignone et al.~\cite{avignone} (orange). The calculations from Ref.~\cite{ceb} that assume Ziegler~\cite{ziegler} and Gordon et al.~\cite{gordon} cosmic neutron spectra are shown as dotted and dashed-dotted lines, respectively. All these calculated rates are also listed in Table~\ref{tab:results}. In addition, the previous experimental measurements for $^{68}$Ge and $^{65}$Zn reported in Ref.~\cite{avignone} are shown as solid orange line in Fig.~\ref{fig:zn} and Fig.~\ref{fig:ge68} and appear in the last column of Table~\ref{tab:results}. Among themselves, these models agree within a factor of 2-3 for $^{65}$Zn, for $^{68}$Ge and $^{55}$Fe, whereas the difference for $^{3}$H can be close to one order of magnitude, ranging from 27.7 to 210 nuclei/kg/day. Ref.~\cite{mei} quotes an experimental upper limit on the $^{3}$H production rate of 21~nuclei/kg/day of $^{3}$H from a by-eye fit to the IGEX data, in tension with an estimate of 27.7~nuclei/kg/day from the model using the TALYS cross sections and the cosmic neutron spectrum parameterization from Ref.~\cite{gordon}. A comparison of the latter value with most recent calculations is available in Ref.~\cite{Zhang:2016rlz}. However, the tritium production rates estimated with GEANT4 (I) and ACTIVIA (II) are larger: 52.4~nuclei/kg/day and 48.3~nuclei/kg/day, respectively, and are closer to our measurement of 82$\pm$21~nuclei/kg/day with statistical and systematic uncertainties included. The measured production rates for all nuclei are within a factor two of our ACTIVIA calculations, with preferred cross sections from MENDL-2P library, especially for the nuclei $^{65}$Zn and $^{68}$Ge. The closest agreement between data and our ACTIVIA calculations is for $^{55}$Fe. Concerning the other calculations listed in Table~\ref{tab:results}, those predicting the largest rates are systematically favored, except for the $^{3}$H prediction from Ref.~\cite{avignone}. The significance of the tension between our measurements and ACTIVIA calculations can be assessed by comparing the difference in their values to the quoted experimental uncertainties. The ACTIVIA estimates from this work show a 1.7$\sigma$ to 1.8$\sigma$ deviation from the measured production rate of $^{3}$H of 82~$\pm$~21~nuclei/kg/day. The disagreement with the prediction from Ref.~\cite{avignone} using the prescription from \cite{Hess} is stronger, at 6$\sigma$. The agreement between the measured rates and those predicted by ACTIVIA (average value), for isotopes other than $^{3}$H, is better: deviations of less than 1.7$\sigma$ (statistical error only) are observed for $^{55}$Fe, $^{49}$V and $^{65}$Zn. It should be noted that the present measurement for $^{65}$Zn disagrees with that in Ref.~\cite{avignone} by 5.9$\sigma$ (statistical error only). Including systematic uncertainty, the disagreement is still strong at a 4.8$\sigma$ level. Since the population of $^{68}$Ge is not reduced to zero during the germanium crystallization, there is an uncertainty in its saturation fraction associated to the amount of time the germanium ore was exposed to cosmic rays before the crystal growing. The first value for $^{68}$Ge in Table~\ref{tab:pr} (202 $\pm$ 16) is calculated assuming no exposure prior to the crystallization. The second assumes a 3-year exposure before the crystal growth, reaching a population close to saturation. In this second scenario, the production rate is reduced to 84~$\pm$~3~nuclei/kg/day. The $\chi^{2}$ of the fit is reduced from 3.2 to 1.0 assuming a scaling factor of $\sigma(f_s)$ of 2 (see Section 6.1) resulting in a production rate of 84 $\pm$ 6, from which is obtained a lower limit on the production rate of 74~nuclei/kg/day at 90\% C.L.\. It is, nevertheless, 3.6$\sigma$ above the average value of the ACTIVA results, considering the difference between the two predictions as a systematic uncertainty. Ref.~\cite{avignone} reports a value of 30~$\pm$~7~nuclei/kg/day at saturation, 6.3$\sigma$ below our lower limit. The present $^{65}$Zn production rate measurement is 2.8~$\pm$~0.6 times larger than that of Ref.~\cite{avignone}. The lower limit on the $^{68}$Ge value is also larger than the measurement in that same reference, by a similar factor of 2.5. In order to evaluate the possible source of this discrepancy, more detailed information on the analysis of the data of Ref.~\cite{avignone} are needed. Our measurements, extending to more than one cosmogenic activation product, can help to improve the models by constraining them better and thus contribute to the reduction of the systematic uncertainties associated with the wide variation of their predictions. The discrepancies of predictions from nucleus to nucleus, and the significant difference in the ACTIVIA calculations with semi-empirical or MENDL-2P cross sections suggest that an important source of uncertainties in the calculations comes from the different excitation functions of an isotope produced by neutrons on the stable parent isotope of the material. \section{Conclusion} The cosmogenic activation of various isotopes in the germanium detectors of the EDELWEISS-III experiment has been measured. The data for five isotopes and thirteen detectors with different exposure times lead to a consistent set of measurements. The first measurement of the $^{3}$H decay rate in germanium detectors is presented. It has been interpreted in terms of production rate of 82~$\pm$~21~nuclei/kg/day with statistical and systematic uncertainties included. The tritium production due to cosmic-ray neutrons is thus important and the present measurement provides valuable information needed to evaluate the reduction of the exposure to cosmic rays necessary for germanium detector arrays used for dark matter searches. The measured production rates on $^{49}$V, $^{55}$Fe and $^{65}$Zn of 2.8 $\pm$ 0.6~nuclei/kg/day, 4.6 $\pm$ 0.7~nuclei/kg/day and 106 $\pm$ 13~nuclei/kg/day, respectively, presented here are the most accurate to-date. A lower limit of 74~nuclei/kg/day at 90\% C.L.\ on production rate of $^{68}$Ge is discussed. The measured $^{65}$Zn production rate and the lower limit on that of $^{68}$Ge are a factor 2.7 $\pm$ 0.6 larger than the measurements reported in Ref.~\cite{avignone}. The origin of this discrepancy is unknown. The measurements agree within a factor of two with estimates performed with the ACTIVIA code within this work. The best agreement is found for $^{49}$V and $^{55}$Fe. The estimates for $^{3}$H, $^{65}$Zn and $^{68}$Ge tend to underestimate the measured rates, with significance ranging from 1.7$\sigma$ to 6$\sigma$. The difference between these predictions and those from other models can also differ by as much as a factor of two in most cases, with no single model giving a satisfying description for all measured isotopes. It can be foreseen that the precision of the present measurements will help constrain and further improve the models. \section{Acknowledgments} The help of the technical staff of the Laboratoire Souterrain de Modane and the participant laboratories is gratefully acknowledged. The EDELWEISS project is supported in part by the German ministry of science and education (BMBF Verbundforschung ATP Proj.-Nr.\ 05A14VKA), by the Helmholtz Alliance for Astroparticle Physics (HAP), by the French Agence Nationale pour la Recherche (ANR) and the LabEx Lyon Institute of Origins (ANR-10-LABX-0066) of the Universit{\'e} de Lyon within the program ``Investissements d'Avenir'' (ANR-11-IDEX-00007), by the P2IO LabEx (ANR-10-LABX-0038) in the framework "Investissements d'Avenir'' (ANR\-11\-IDEX\-0003\-01) managed by the ANR (France), by Science and Technology Facilities Council (UK), and the Russian Foundation for Basic Research (grant No. 15-02-03561). \clearpage \newpage \nocite{*} \bibliographystyle{aipnum-cp}
d1c27e8f8bf8c3fc658b94c78ecb9ad84b736d71
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction.} Katz \cite[5.3.1]{Katz90ESDE} discovered that the hypergeometric $\sD$-modules on $\bA^1_{\bC}\setminus\{0\}$ can be described as the multiplicative convolution of hypergeometric $\sD$-modules of rank one. Precisely speaking, Katz proved the statement (ii) in the following theorem (the statement (i) is trivial but put to compare with another theorem later). \begin{cvthm} Let $\balpha=(\alpha_1,\dots,\alpha_m)$ and $\bbeta=(\beta_1,\dots,\beta_n)$ be two sequences of complex numbers and assume that $\alpha_i-\beta_j$ is not an integer for any $i,j$. Let $\sHyp(\balpha;\bbeta)$ be the $\sD$-module on $\bG_{\rmm,\bC}$ defined by the hypergeometric operator \[ \Hyp(\balpha;\bbeta)=\prod_{i=1}^m(x\partial-\alpha_i)-x\prod_{j=1}^n(x\partial-\beta_j), \] that is, \[ \sHyp(\balpha;\bbeta)\defeq\sD_{\bA^1_{\bC}\setminus\{0\}}/\sD_{\bA^1_{\bC}\setminus\{0\}}\Hyp(\balpha;\bbeta). \] Then, $\sHyp(\balpha;\bbeta)$ has the following properties. \textup{(i)} If $m\neq n$, then $\sHyp(\balpha;\bbeta)$ is a free $\sO_{\bG_{\rmm,\bC}}$-module of rank $\max\{m,n\}$. If $m=n$, then the restriction of $\sHyp(\balpha;\bbeta)$ to $\bG_{\rmm,\bC}\setminus\{1\}$ is a free $\sO_{\bG_{\rmm,\bC}\setminus\{1\}}$-module of rank $m$. \textup{(ii)} We have an isomorphism \[ \sHyp(\balpha;\bbeta)\cong \sHyp(\alpha_1;\emptyset)\ast\dots\ast\sHyp(\alpha_m;\emptyset) \ast\sHyp(\emptyset;\beta_1)\ast\dots\ast\sHyp(\emptyset;\beta_n), \] where $\ast$ denotes the multiplicative convolution of $\sD_{\bG_{\rmm,\bC}}$-modules. \end{cvthm} Besides the hypergeometric $\sD$-modules over the complex numbers, Katz also studied the $\ell$-adic theory of hypergeometric sheaves. Let $k$ be a finite field with $q$ elements, let $\psi$ be a non-trivial additive character on $k$ and let $\bchi=(\chi_1,\dots,\chi_m), \brho=(\rho_1,\dots,\rho_n)$ be sequences of characters on $k^{\times}$ satisfying $\chi_i\neq\rho_j$ for all $i,j$. Then, he \emph{defined} the $\ell$-adic hypergeometric sheaves $\sH_{\psi,!}^{\ell}(\bchi,\brho)$ on $\bG_{\rmm,k}$ by using the multiplicative convolution of $\sH_{\psi,!}^{\ell}(\chi_i;\emptyset)$'s and $\sHyp_{\psi,!}^{\ell}(\emptyset;\rho_j)$'s, where these convolvends are defined by using Artin--Schreier sheaves and Kummer sheaves. This $\ell$-adic sheaf $\sH_{\psi,!}^{\ell}(\bchi;\brho)$ has a property similar to (i) in the above theorem. Namely, it is a smooth sheaf on $\bG_{\rmm,k}$ of rank $\max\{m,n\}$ if $m\neq n$, and its restriction to $\bG_{\rmm,k}\setminus\{1\}$ is a smooth sheaf of rank $m$ if $m=n$ \cite[Theorem 8.4.2]{Katz90ESDE}. Moreover, by definition, $\sH_{\psi,!}^{\ell}(\bchi;\brho)$ has a Frobenius structure. The Frobenius trace functions of the $\ell$-adic hypergeometric sheaves are called the ``hypergeometric functions over finite field''. This function gives a generalization of the classical Kloosterman sums. Moreover, this function has an intimate connection with the Frobenius action on the \'etale cohomology of a certain class of algebraic varieties (for example, Calabi--Yau varieties) over finite fields. (The hypergeometric function over finite field is also called the ``Gaussian hypergeometric function'' by Greene \cite{Greene}, who independently of Katz found this function based on a different motivation.) The purpose of this article is to develop a $p$-adic counterpart of these complex and $\ell$-adic hypergeometric objects. This $p$-adic hypergeometric object will have a presentation in terms of the ($p$-adic) differential equation, and at the same time has a Frobenius structure. For its formalisation, we exploit the theory of arithmetic $\sD$-modules introduced by Berthelot. The main theorem of this article is stated as follows. \begin{mainthm} Let $K$ be a complete discrete valuation field of mixed characteristic $(0,p)$ with residue field $k$, a finite field with $q$ elements. Let $\pi$ be an element of $K$ that satisfies $\pi^{q-1}=(-p)^{(q-1)/(p-1)}$. Let $\balpha=(\alpha_1,\dots,\alpha_m)$ and $\bbeta=(\beta_1,\dots,\beta_n)$ be two sequences of elements of $\frac{1}{q-1}\bZ$, and assume that $\alpha_i-\beta_j$ is not an integer for any $i,j$. Let $\sHyp_{\pi}(\balpha;\bbeta)$ be the $\sD^{\dag}_{\widehat{\bP}^1,\bQ}(\pdag{\{0,\infty\}})$-module defined by the $p$-adic hypergeometric differential operator \[ \Hyp_{\pi}(\balpha;\bbeta)=\prod_{i=1}^m(x\partial-\alpha_i)-(-1)^{m+np}\pi^{m-n}x\prod_{j=1}^n(x\partial -\beta_j), \] that is, \[ \sHyp_{\pi}(\balpha;\bbeta)\defeq\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{0,\infty\}})/ \sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{0,\infty\}})\Hyp_{\pi}(\balpha;\bbeta). \] Then, $\sHyp_{\pi}(\balpha;\bbeta)$ has the following properties. \textup{(i)} If $m\neq n$, then $\sHyp_{\pi}(\balpha;\bbeta)$ has a structure of an overconvergent $F$-isocrystal on $\bG_{\rmm,k}$ of rank $\max{\{m,n\}}$. If $m=n$, then the restriction of $\sHyp_{\pi}(\balpha;\bbeta)$ to $\widehat{\bP^1_V}\setminus\{0,1,\infty\}$ has a structure of an overconvergent $F$-isocrystal on $\bG_{\rmm, k}\setminus\{1\}$ of rank $m$. \textup{(ii)} We have an isomorphism \[ \sHyp_{\pi}(\balpha;\bbeta)\cong\sHyp_{\pi}(\alpha_i;\emptyset)\ast\dots\ast\sHyp_{\pi}(\alpha_m;\emptyset)\ast \sHyp_{\pi}(\emptyset;\beta_1)\ast\dots\ast\sHyp_{\pi}(\emptyset;\beta_n). \] \textup{(iii)} Let $x$ be a closed point in $\bG_{\rmm,k}$ if $m\neq n$, and in $\bG_{\rmm,k}\setminus\{1\}$ of $m=n$. Then, the Frobenius trace of the overconvergent $F$-isocrystal obtained in (i) at $x$ equals that of the $\ell$-adic hypergeometric sheaf $\sH_{\psi,!}^{\ell}(\bchi;\brho)$, where $\psi$, $\bchi$ and $\brho$ are the corresponding (sequence of) characters to $\pi$, $\balpha$ and $\bbeta$ respectively. \end{mainthm} In this theorem, (i) states not only that it is a free $\sO$-module of the desired rank, but also that this $\sD$-module satisfies a kind of convergence condition and that it has a Frobenius structure. Despite the triviality of (i) of Convolution Theorem over $\bC$, this is therefore a highly non-trivial statement. Another difference from the over-$\bC$ case is that we restricted the parameters to rational coefficients. We expect that, even for other $p$-adic parameters not necessarily coming from characters, the arithmetic hypergeometric $\sD$-modules are overconvergent isocrystals (not necessarily with Frobenius structure) if these parameters satisfy the non-Liouville difference condition. However, since the overholonomicity of arithmetic $\sD$-modules is not preserved by tensor product, our method is not applicable at least verbatim. \vspace{10pt} We briefly explain the strategy of the proof of Main Theorem. We firstly prove (ii). By a similar argument to the theory of $\sD$-module over the complex numbers, the convolution with $\sHyp_{\pi}(\alpha_i;\emptyset)$ can be described by using the $p$-adic Fourier transform of $\sD$-modules ``on $\widehat{\bA}^1$''. We compare $\sHyp_{\pi}(\balpha;\bbeta)$, which is an object ``on $\widehat{\bG_{\rmm}}$'', with the $\sD$-module ``on $\widehat{\bA}^1$'' associated to $\Hyp_{\pi}(\balpha;\bbeta)$. (This is a point where we need some $p$-adic calculation.) By this comparison, we may compute the right-hand side of the isomorphism in (ii) with the aid of the explicit description of $p$-adic Fourier transform given by Huyghe. Secondly, (iii) is a corollary of (ii) (see Remark \ref{rem:explicittraces}). We finally prove (i). It is an essentially classical result that (i) holds for $\sHyp_{\pi}(\alpha_i;\emptyset)$'s and $\sHyp_{\pi}(\emptyset;\beta_j)$'s. In particular, they are overholonomic $F$-$\sD$-modules. Because the property of being an overholonomic $F$-$\sD$-module is preserved by six functors as proved by D.\ Caro, (ii) shows that $\sHyp_{\pi}(\balpha;\bbeta)$ is also an overholonomic $F$-$\sD$-module. An overholonomic $F$-$\sD$-module is known to be an overconvergent $F$-isocrystal on a dense open subset, but we have to show that it is so on the ``correct'' open subset. To do this, we use (iii) to show that $\sHyp_{\pi}(\balpha;\bbeta)$ have the desired rank as an $\sO_{\bX,\bQ}$-module. By combining this result with some properties of $F$-isocrystals, we can finish the proof of (ii). \vspace{10pt} Recently, there appeared an article by Crew \cite{Crew16}, which studies the $p$-adic differential equations from the point of view of rigidity. He proves the existence of the Frobenius structure of the regular singular rigid overconvergent isocrystals under suitable conditions. We note that our Main Theorem (i) in the case where $m=n$ thus proves that the regular singular differential equations as above are examples of the objects which he is dealing with. There are other studies on the $p$-adic hypergeometric equations. Another study by Crew \cite{Crew94CM} deals with them in the case where $n=0$, all $\alpha_i$'s are zero, $p$ is not divisible by $m$ and $p\neq 2$; our Main Theorem (i) is proved there in this case. A theorem due to Tsuzuki \cite[3.3.1, 3.3.3]{Tsuzuki03} is applicable to proving the overconvergence and the existence of a Frobenius structure for ``Picard--Fuchs equations'' such as $\Hyp(1/2,1/2;0,0)$, and we may prove Main Theorem (i) by it in these cases. Our approach is quite different from theirs in the point that we extensively use the multiplicative convolution and systematically deal with general case including these two cases. At last, we note that the $p$-adic theory of hypergeometric functions (or exponential sums) are also studied in terms of partial linear differential equations of rank one on a higher dimensional torus, for example by Dwork \cite{Dwork} and Adolphson \cite{Adolphson}. \vspace{10pt} We conclude this introduction by explaining the structure of the article. Section 1 is a quick review of the theory of arithmetic $\sD$-modules. Section 2 is devoted to giving a fundamental properties of multiplicative convolution of arithmetic $\sD$-modules. Section 3 concerns the arithmetic hypergeometric $\sD$-modules. In Subsection 3.1, we give a definition of the $p$-adic hypergeometric $\sD$-modules by using the $p$-adic hypergeometric differential operators, and prove fundamental properties of them. In Subsection 3.2, we give another definition of the $p$-adic hypergeometric $\sD$-modules by using the multiplicative convolution, and compare it with the one defined in Subsection 3.1; Main Theorem (i) is proved here (Theorem \ref{thm:hypandconv}). Subsection 3.3 gives additional properties of $p$-adic hypergeometric $\sD$-modules concerning multiplicative convolutions. Section 4 deals with the hypergeometric isocrystals. In Subsection 4.1, we proved that the $p$-adic hypergeometric $\sD$-modules are in fact overconvergent $F$-isocrystals on the desired open subsets. Here, we finish the proof of Main Theorem ((ii) is proved in Theorem \ref{thm:ocFisoc}. For (iii), see Remark \ref{rem:explicittraces}.) In the last Subsection 4.2, we prove the irreducibility of the hypergeometric isocrystals and give a characterization of them. \subsection{Acknowledgements} The author would like to thank Jeng-Daw Yu for his discussions and a lot of useful comments about this work. The author would also like to thank Tomoyuki Abe for his answer to my questions about the theory of arithmetic $\sD$-modules. \subsection{Conventions and Notations.} Throughout this article, $V$ denotes a complete discrete valuation ring of mixed characteristic $(0,p)$ whose residue field $k$ is a finite field with $q=p^s$ elements. The fraction field of $V$ is denoted by $K$. We denote by $|\cdot|$ the norm on $K$ normalized as $|p|=p^{-1}$. We will assume that $K$ has a primitive $p$-th root of unity after Subsection 3.2. If $\sP$ is a smooth formal scheme over $\Spf(V)$, then $\sD^{\dag}_{\sP,\bQ}$ denotes the sheaf of rings of arithmetic differential operators introduced by Berthelot \cite{BerthelotI}. Moreover, if $T$ is a divisor of the special fiber of $\sP$, then $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$ denotes \emph{the sheaf of arithmetic differential operators with overconvergent singularity along $T$} \cite[4.2]{BerthelotI}. Let $\sA$ be a sheaf of (not necessarily commutative) rings. If we say ``$\sA$-module'', we always mean a sheaf of left $\sA$-modules. $D^{\rb}(\sA)$ denotes the bounded derived category of the category of $\sA$-modules. $D^{\rb}_{\coh}(\sA)$ denotes the full subcategory of $D^{\rb}(\sA)$ consisting of the complexes whose cohomologies are coherent. Finally, in this article, if $\kappa$ is a field, a \emph{$\kappa$-variety} means a separated $\kappa$-scheme of finite type. \section{Arithmetic $\sD$-modules.} In this section, we recall the theory of arithmetic $\sD$-modules that will be used in this article. \subsection{Six functors.} In this subsection, we fix terminologies and notations, and quickly summarize the theory of six-functor formalism of arithmetic $\sD$-modules. For a detailed explanation, the reader may consult \cite[Section 1]{AbeCaro}. Let $\sP$ be a smooth formal scheme over $\Spf(V)$. Caro \cite[D\'efinition 2.1]{Caro09ASENS} defined the subcategory $D^{\rb}_{\ovhol}(\sD^{\dag}_{\sP,\bQ})$ of $D^{\rb}_{\coh}(\sD^{\dag}_{\sP,\bQ})$ consisting of \emph{overholonomic $\sD^{\dag}_{\sP,\bQ}$-complexes}. Moreover, let $T$ be a divisor of the special fiber of $\sP$. The subcategory of $D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)$ consisting of the objects that are overholonomic as $\sD^{\dag}_{\sP,\bQ}$-complex is denoted by $D^{\rb}_{\ovhol}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)$. If a coherent $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$-module is an object of $D^{\rb}_{\ovhol}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)$, then we say that it is an overholonomic $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$-module. It is convenient to introduce here the following terminology. \begin{definition} (i) A \emph{d-couple} is a pair $(\sP,T)$, where $\sP$ is a smooth formal scheme over $\Spf(V)$ and where $T$ is a divisor of the special fiber of $\sP$ (an empty set is also a divisor). (ii) A \emph{morphism of d-couples} $\widetilde{f}\colon(\sP',T')\to(\sP,T)$ is a morphism $\overline{f}\colon\sP'\to\sP$ such that $\overline{f}(\sP'\setminus T')\subset\sP\setminus T$ and that $\overline{f}^{-1}(T)$ is a divisor (or empty). \end{definition} \begin{definition} (i) A d-couple $(\sP, T)$ \emph{realizes} the $k$-variety $X$ if $\sP$ is proper and $X$ is the special fiber of $\sP\setminus T$. (ii) A morphism of d-couples $\widetilde{f}\colon(\sP',T')\to(\sP,T)$ \emph{realizes} the morphism of $k$-varieties $f\colon X'\to X$ if $(\sP', T')$ \resp{$(\sP, T)$} realizes $X'$ \resp{$X$} and if $\overline{f}\colon\sP'\to\sP$ induces $f$. \end{definition} Now, let $X$ be a $k$-variety, and assume that there exists a d-couple $(\sP, T)$ realizing $X$. Then, the category $D^{\rb}_{\ovhol}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)$ is proved to be independent of the choice of $(\sP, T)$ up to a canonical equivalence. This category is also denoted by $D^{\rb}_{\ovhol}(X/K)$. Even if such a realizing d-couple does not exist, the triangulated category $D^{\rb}_{\ovhol}(X/K)$ can be defined for arbitrary realizable $k$-varieties (Recall that a $k$-variety is said to be \emph{realizable} if it can be embedded into a proper smooth formal scheme over $\Spf(V)$.) We do not explain the construction of $D^{\rb}_{\ovhol}(X/K)$ for realizable varieties because we have only to use formal properties of this category and do not need to refer to the construction. \begin{definition} Let $X$ be a realizable variety over $k$. (Recall that $k$ is a finite field with $q=p^s$ elements.) We denote by $F^{(s)}_{X/k}\colon X\to X$ the $s$-th Frobenius morphism on $X$. (i) Let $\sM$ be an object of $D^{\rb}_{\ovhol}(X/K)$. A \emph{Frobenius structure} on $\sM$ is an isomorphism $\Phi\colon\sM\to (F^{(s)}_{X/k})^{\ast}\sM$. (ii) We define the category $F\hyphen D^{\rb}_{\ovhol}(X/K)$ as follows: the objects are the pairs $(\sM, \Phi)$ consisting of an object $\sM$ of $D^{\rb}_{\ovhol}(X/K)$ and a Frobenius structure $\Phi$ on $\sM$; the morphisms are those of $D^{\rb}_{\ovhol}(X/K)$ compatible with the Frobenius structures. \end{definition} The category $F\hyphen D^{\rb}_{\ovhol}(X/K)$ is equipped with Grothendieck's six operations. \vspace{4px} \textbf{Dual functor.} Let $X$ be a realizable $k$-variety. Then, we have the \emph{dual functor} \[ \bD_X\colon F\hyphen D_{\ovhol}^{\rb}(X/K)\longrightarrow F\hyphen D^{\rb}_{\ovhol}(X/K). \] The functor $\bD_X\circ\bD_X$ is naturally isomorphic to the identity functor on $F\hyphen D^{\rb}_{\ovhol}(X/K)$. \vspace{4px} \textbf{Pull-back functors.} Let $f\colon X'\to X$ be a morphism of realizable $k$-varieties. Then, we have the extraordinary pull-back functor \[ f^!\colon F\hyphen D^{\rb}_{\ovhol}(X/K)\longrightarrow F\hyphen D^{\rb}_{\ovhol}(X'/K). \] Moreover, we define the ordinary pull-back functor \[ f^{\plus}\colon F\text{-}D^{\rb}_{\ovhol}(X/K)\longrightarrow F\text{-}D^{\rb}_{\ovhol}(X'/K) \] by $f^{\plus}=\bD_{X'}\circ f^!\circ\bD_X$. If $f$ is an open immersion, then the two functors $f^!$ and $f^{\plus}$ are isomorphic \cite[5.5]{Abe14}; in this case, these two functors are also denoted by $f^{\ast}$. Now, let us assume that $f$ is realized by a smooth morphism $\widetilde{f}\colon(\sP',T')\to(\sP,T)$ of d-couples. (A morphism of d-couple is said to be smooth if the morphism $\overline{f}\colon \sP'\to\sP$ is smooth.) In this case, we have a functor $\widetilde{f}^!\colon D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)\to D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP',\bQ}(\pdag{T'})\big)$ compatible with the functor $f^!$ above. In fact, Berthelot's construction \cite[4.3.3]{Berthelot02Ast} gives a functor $\overline{f}^!\colon D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)\to D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP',\bQ}(\pdag{f^{-1}(T)})\big)$, and we have $\widetilde{f}^!(\sM)=\sD^{\dag}_{\sP',\bQ}(\pdag{T'})\otimes\overline{f}^!(\sM)$. In the case where $f\colon \sX'\hookrightarrow\sX$ is an open immersion, where $\sP'=\sP$ and where $\overline{f}$ is the identity morphism, we also denote $\widetilde{f}^!$ by $\widetilde{f}^{\ast}$. The functor $\widetilde{f}^{\ast}$ is exact on the category of coherent $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$-modules because a sequence of coherent $\sD^{\dag}_{\sP,\bQ}(\pdag{T'})$-modules is exact if and only if its restriction on $\sX'$ is \cite[Proposition 4.3.12 (ii)]{BerthelotI}. \vspace{4px} \textbf{Push-forward functors.} Let $f\colon X'\to X$ be a morphism of realizable $k$-varieties. Then, we have the push-forward functor $f_{\plus}$: \[ f_{\plus}\colon F\hyphen D^{\rb}_{\ovhol}(X'/K)\longrightarrow F\hyphen D^{\rb}_{\ovhol}(X/K). \] Moreover, we define the extraordinary push-forward functor \[ f_!\colon F\hyphen D^{\rb}_{\ovhol}(X'/K)\longrightarrow F\hyphen D^{\rb}_{\ovhol}(X/K) \] by $f_!=\bD_X\circ f_{\plus}\circ\bD_{X'}$. We have a natural morphism $f_!\to f_{\plus}$, and it is an isomorphism if $f$ is proper. We mention here a special case. Assume that $f$ is realized by a morphism of d-couples $\widetilde{f}\colon(\sP',T')\to(\sP,T)$. Then, we have a push-forward functor \cite[1.1.6]{Caro06CM} \[ \widetilde{f}_{\plus}\colon D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP',\bQ}(\pdag{T'})\big)\to D^{\rb}\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big) \] and it is compatible with the $f_{\plus}$ above. If $f$ is an open immersion, if $\sP'=\sP$ and if $\overline{f}$ is the identity morphism on $\sP$, then $\widetilde{f}_{\plus}$ is obtained by considering the complex of $\sD^{\dag}_{\sP,\bQ}$-modules as a complex of $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$-module via the inclusion $\sD^{\dag}_{\sP,\bQ}(\pdag{T})\hookrightarrow\sD^{\dag}_{\sP,\bQ}(\pdag{T'})$. \vspace{4px} \textbf{Tensor products.} Let $X$ be a realizable $k$-variety. Then we have the twisted tensor functor \[ \widetilde{\otimes}\colon F\hyphen D^{\rb}_{\ovhol}(X/K)\times F\hyphen D^{\rb}_{\ovhol}(X/K)\longrightarrow F\hyphen D^{\rb}_{\ovhol}(X/K). \] If $f\colon X'\to X$ is a morphism of $k$-varieties, then we have an isomorphism $f^!( \text{ -- } \widetilde{\otimes} \text{ -- } )\cong f^!(\text{ -- })\widetilde{\otimes} f^!(\text{ -- })$. If $X$ is realized by a d-couple $(\sP, T)$, then for two coherent $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$-modules $\sM$ and $\sN$, the module $\sM\widetilde{\otimes}\sN$ is defined by $\sM\otimes^{\dag}_{\sO_{\sP}(\pdag{T})_{\bQ}}\sN[-\dim\sP]$. Here, $\sO_{\sP,\bQ}(\pdag{T})$ denotes $\sO_{\sP}(\pdag{T})\otimes\bQ$, where $\sO_{\sP}(\pdag{T})$ is the sheaf of functions on $\sP$ with overconvergent singularities along $T$ \cite[4.2.4]{BerthelotI}, and $\otimes^{\dag}$ denotes the overconvergent tensor product. The tensor functor $-\otimes-\defeq \bD_X\big(\bD_X(-)\widetilde{\otimes}\bD_X(-)\big)$ is also defined. \vspace{4px} \textbf{Exterior tensor product.} Let $X, Y$ be two realizable $k$-varieties. Then, we have the \emph{exterior tensor product} \[ \boxtimes\colon F\hyphen D_{\ovhol}^{\rb}(X/K)\times F\hyphen D_{\ovhol}^{\rb}(Y/K) \longrightarrow F\hyphen D^{\rb}_{\ovhol}(X\times Y/K) \] defined by $\sM\boxtimes\sN\defeq \pr_1^{!}\sM\widetilde{\otimes}\pr_2^{!}\sN$, where $\pr_1\colon X\times Y\to X$ \resp{$\pr_2\colon X\times Y\to Y$} denotes the first \resp{second} projection. The functor $-\boxtimes-$ is isomorphic to $\pr_1^{\plus}(-)\otimes\pr_2^{\plus}(-)$ \cite[Proposition 1.3.3]{AbeCaro}. \label{para:boxtimesandpush} We also have the relative K\"unneth formula \cite[1.1.5]{Abe13Langlands}. Namely, if $f\colon X'\to X$, $g\colon Y'\to Y$ are morphisms of realizable $k$-varieties, then there exists a natural isomorphism $(f\times g)_{\plus}(\sM\boxtimes\sN)\cong(f_{\plus}\sM)\boxtimes (g_{\plus}\sN)$ for each object $\sM$ of $F\hyphen D^{\rb}_{\ovhol}(X'/K)$ and $\sN$ of $F\hyphen D^{\rb}_{\ovhol}(Y'/K)$. By using the dual functor, we also get a natural isomorphism $(f\times g)_!(\sM\boxtimes\sN)\cong(f_!\sM)\boxtimes(g_!\sN)$. \vspace{4px} Projection formula is also available \cite[A.6]{AbeCaro}. Namely, let $f\colon X'\to X$ be a morphism of realizable $k$-varieties. Then, for each $\sM\in F\hyphen D^{\rb}_{\ovhol}(X'/K)$ and $\sN\in F\hyphen D^{\rb}_{\ovhol}(X/K)$, we have natural isomorphisms $f_{\plus}(\sM\widetilde{\otimes}f^{!}\sN)\cong f_{\plus}(\sM)\widetilde{\otimes}\sN$. \vspace{4px} The base change theorem is also available; assume that we are given the cartesian diagram \[ \begin{tikzpicture}[description/.style={fill=white,inner sep=2pt}] \matrix (m) [matrix of math nodes, row sep=0.4em, column sep=1em, text height=1.5ex, text depth=0.25ex] { X_2' & & X_2 \\ & \square & \\ X_1' & & X_1 \\ }; \path[->,font=\scriptsize] (m-1-1) edge node[above] {$g'$} (m-1-3) (m-1-1) edge node[left] {$f'$} (m-3-1) (m-1-3) edge node[right] {$f$} (m-3-3) (m-3-1) edge node[below] {$g$} (m-3-3); \end{tikzpicture} \] of realizable $k$-varieties. Then, we have natural isomorphisms $g^!\circ f_{\plus}\cong f'_{\plus}\circ g'^!$ and $g^{\plus}\circ f_!\cong f'_!\circ g'^{\plus}$ of functors $F\hyphen D^{\rb}_{\ovhol}(X_2/K)\to F\hyphen D^{\rb}_{\ovhol}(X_1'/K)$. In fact, the first isomorphism is the base change theorem \cite[1.3.10]{AbeCaro}, and the second one is derived from the first by using the dual functors. \vspace{4px} \textbf{Scalar Extension.} We introduce a functor of scalar extension. Let $k'$ be a finite extension of $k$ with $q'=p^{s'}$ elements, and put $V'\defeq V\otimes_{W(k)}W(k')$ and $K'\defeq\Frac(V')$. Let $X$ be a realizable variety over $k$ and put $X'\defeq X\times_kk'$. Under this setting, we have the ``changing the base field'' functor \[ \iota_{k'/k}\colon F\hyphen D^{\rb}_{\ovhol}(X/K)\to F\hyphen D^{\rb}_{\ovhol}(X'/K'). \] (Note that a Frobenius structure on an object $\sN$ of $D^{\rb}_{\ovhol}(X'/K')$ is defined to be an isomorphism $\sN\to(F^{(s')}_{X'})^{\ast}\sN$.) For later use, we describe the construction of $\iota_{k'/k}$ assuming that $X$ is realized by a d-couple $(\sP,T)$ realizing $X$. We first fix notations. Put $\sP'\defeq\sP\times_{\Spf(V)}\Spf(V')$ and $T'\defeq T\times_kk'$, and denote by $f\colon\sP'\to\sP$ the projection. In this paragraph, we also denote $\sD^{\dag}_{\sP,\bQ}$ by $\sD^{\dag}_{\sP/V,\bQ}$ to make it explicit that it is defined for the morphism $\sP\to\Spf(V)$. Now, take an object $(\sM,\Phi)$ of $F\hyphen D^{\rb}_{\ovhol}(X/K)$. Let us put \[ \sM'\defeq\sD^{\dag}_{\sP'/V',\bQ}\otimes_{f^{-1}\sD^{\dag}_{\sP/V,\bQ}} f^{-1}\sM. \] Since $f^{\ast}\sD^{\dag}_{\sP/V,\bQ}\cong\sD^{\dag}_{\sP'/V',\bQ}$ \cite[2.2.2]{Berthelot02Ast}, $\sM'$ is isomorphic to $f^!\sM$ as objects of $D^{\rb}_{\ovhol}(X'/K)$ by definition of Berthelot's functor $f^!$. Because of the presence of the Frobenius structure, $\sM'$ is in fact an object of $D^{\rb}_{\ovhol}(X'/K')$ \cite[1.2]{AbeCaro14}. Moreover, by using the (semi-linear) morphism $\Phi'\colon \sM'\to (F_{X'}^{(s)})^{\ast}\sM'$ induced by $\Phi$, we define an isomorphism $\Phi''\colon \sM'\to (F_{X'}^{(s')})^{\ast}\sM'$ in $D^{\rb}_{\ovhol}(X'/K')$ by \[ \Phi''\defeq (F_{X'}^{(s'-s)})^{\ast}\Phi'\circ(F_{X'}^{(s'-2s)})^{\ast}\Phi'\dots\circ(F_{X'}^{(s)})^{\ast}\Phi'\circ\Phi'. \] We define $\iota_{k'/k}\big((\sM,\Phi)\big)$ to be the pair $(\sM',\Phi'')$. By the properties of six functors, the scalar extension functor $\iota_{k'/k}$ commutes with six functors. \vspace{4pt} \textbf{Frobenius trace function.} Let $X$ be a realizable $k$-variety, let $(\sM, \Phi)$ be an object of $F\hyphen D^{\rb}_{\ovhol}(X/K)$ and let $k'$ be a finite extension of $k$. In this paragraph, we make explicit the term ``Frobenius trace of $(\sM,\Phi)$ at a $k'$-valued point $x\in X(k')$''. Let $i_x\colon\Spec(k')\hookrightarrow X'$ denote the morphism of $k'$-varieties induced by $x$. Under the notation in the previous paragraph, put $(\sM',\Phi')\defeq i_x^{\plus}\circ \rho_{k'/k}\big( (\sM,\Phi)\big)$; it is thus an object of $F\hyphen D^{\rb}_{\ovhol}(\Spec(k')/K')$. $\sM'$ is a complex of $K'$-vector spaces each of whose cohomology is finite-dimensional, and the natural morphism $\sM'\to(F^{(s')})^{\ast}\sM'$ composed with $(\Phi')^{-1}\colon(F^{(s')})^{\ast}\sM'\to\sM'$ is a $K'$-linear isomorphism of $\sM'$, which we denote by $F_{\sM,x}$. Now, the Frobenius trace of $(\sM,\Phi)$ at $x\in X(k')$ is defined to be $\sum_{i\in\bZ}(-1)^i\tr\big(F_{\sM,x}|\sH^i(\sM')\big)$. \vspace{4pt} \textbf{Overconvergent $F$-isocrystals.} At last, we recall the relationship between the theory of isocrystals and that of arithmetic $\sD$-modules. Let $X$ be a smooth scheme of dimension $d$ which is separated of finite type over $k$, and assume that there exists a d-couple $(\sP, T)$ realizing $X$. Then, there exists a fully faithful functor \[ \sp_{\plus}\colon F\hyphen\Isoc^{\dag}(X/K)\longrightarrow F\hyphen\Coh\big(\sD^{\dag}_{\sP,\bQ}(\pdag{T})\big), \] where the target denotes the category of coherent $\sD_{\sP,\bQ}(\pdag{T})$-modules with Frobenius structure. The essential image of this functor is the coherent $\sD_{\sP,\bQ}(\pdag{T})$-modules whose restriction to $\sP\setminus T$ is coherent as an $\sO_{\sP\setminus T,\bQ}$-module \cite[Th\'eor\`eme 2.2.12]{Caro06CM}. We say that an object $F\hyphen\Coh\big(\sD_{\sP,\bQ}^{\dag}(\pdag{T})\big)$ is an overconvergent $F$-isocrystal if it belongs to the essential image of $\sp_{\plus}$. If $f\colon X\to Y$ is a morphism between smooth schemes separated of finite type, and if we put $d\defeq \dim X-\dim Y$, we have $\sp_{\plus}\circ f^{\ast}\cong f^{\plus}([-d])\circ\sp_{\plus}$ \cite[4.1.1]{Abe13Langlands}. Moreover, if we $\sP$ is of constant dimension, then we have $\sp_{\plus}(-\otimes -)\cong \sp_{\plus}(-)\widetilde{\otimes}\sp_{\plus}(-)[\dim\sP]$ \cite[3.1.8]{Caro15MM}. \subsection{$\sD^{\dag}$-affinity.} As in the classical case, when we discuss about the coherent $\sD^{\dag}$-modules, it often suffices that we may just discuss on the global sections. In this subsection, we recall a fundamental theorem about this point and give two particular examples which we regularly use in this article. \begin{theorem}[{\cite[5.3.3]{Huyghe98AIF}}] Let $\sP$ be a projective smooth formal scheme over $\Spf(V)$, and let $T$ be an ample divisor of the special fiber of $\sP$. Then, the functor $\Gamma(\sP, -)$ is exact and gives an equivalence of categories from the category of coherent $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$-modules to the category of coherent $\Gamma\big(\sP, \sD^{\dag}_{\sP,\bQ}(\pdag{T})\big)$-modules. \label{thm:Daffine} \end{theorem} We particularly apply this theorem to the following two d-couples: $(\sP, T)=(\widehat{\bP^1_V}, \{\infty\})$ and $(\sP, T)=(\widehat{\bP^1_V}, \{0,\infty\})$. In these cases, the ring of global sections of the sheaf $\sD^{\dag}_{\sP,\bQ}(\pdag{T})$ has an explicit description (cf. \cite[p. 915]{Huyghe98AIF}), and we immediately get the following corollaries. \begin{corollary} \label{para:DoverA1} Let $A_1(K)^{\dag}$ be the ring defined by \[ A_1(K)^{\dag} \defeq \Set{\sum_{l,k\in\bN}a_{l,k}x^l\partial^{[k]} | a_{l,k}\in K, \exists C>0, \exists \eta<1, |a_{l,k}|_p < C\eta^{l+k}}. \] Then, the functor $\Gamma(\widehat{\bP^1_V}, -)$ is exact and gives an equivalence of categories from the category of coherent $\sD_{\widehat{\bP^1_V},\bQ}^{\dag}(\pdag\{\infty\})$-modules to that of coherent $A_1(K)^{\dag}$-modules. \end{corollary} \begin{corollary} \label{para:DoverGm} Let $B_1(K)^{\dag}$ be the ring defined by \[ B_1(K)^{\dag} \defeq \Set{\sum_{l\in\bZ,k\in\bN}a_{l,k}x^l\partial^{[k]} | a_{l,k}\in K, \exists C>0, \exists \eta<1, |a_{|l|_{\infty},k}|_p < C\eta^{|l|_{\infty}+k}}. \] Here, $|l|_{\infty}$ denotes the Euclid norm. Then, the functor $\Gamma(\widehat{\bP^1_V}, -)$ is exact and gives an equivalence of categories from the category of coherent $\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag\{0,\infty\})$-modules to that of coherent $B_1(K)^{\dag}$-modules. \end{corollary} In the rest of this article, we implicitly use these equivalences and identify the source and the target. \subsection{Examples of overholonomic arithmetic $\sD$-modules.} Now, we introduce some specific examples of overholonomic $F$-$\sD^{\dag}$-modules. \vspace{4px} \textbf{Delta modules.} Let $X$ be an open subscheme of $\bA^1_k$, let $\lambda$ be an element of $k$ and let $\iota_{\lambda}\colon \{\lambda\}\hookrightarrow X$ denote the inclusion. We define an object $\delta_{\lambda}$ of $F\hyphen D^{\rb}_{\ovhol}(X/K)$, the ``Dirac delta module'' at $\lambda$. Note that the sheaf $\sO_{\{\lambda\},\bQ}\defeq\sO_{\{\lambda\}}\otimes\bQ$ can be seen as an object of $F\hyphen D^{\rb}_{\ovhol}(\{\lambda\}/K)$. We define $\delta_{\lambda}\defeq \iota_{\lambda,\plus}(\sO_{\{\lambda\},\bQ})$. If $\widetilde{\lambda}$ denotes the Teichm\"uller lift of $\lambda$, this object is isomorphic to $A_1(K)^{\dag}/A_1(K)^{\dag}(x-\widetilde{\lambda})$ as a coherent $A_1(K)^{\dag}$-module. \vspace{4px} \textbf{Dwork module associated with a non-trivial additive character.} Let $\psi$ be a non-trivial additive character on $k$ and assume that $K$ contains a primitive $p$-th root of unity. We associate with $\psi$ an overholonomic $F\hyphen\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})$-module (therefore an object of $F\hyphen D^{\rb}_{\ovhol}(\bA^1_k)$) $\sL_{\psi}$, which we call ``the Dwork module associated with $\psi$''. Before we review its definition, let us recall the fact that $\psi$ defines an element $\pi_{\psi}$ of $K$ as follows \cite[(1.3)]{Berthelot84Ast}. Firstly, remark that the character $\psi$ takes value in $K$ because we assumed that $K$ has a primitive $p$-th root of unity. Now, if $k$ is a prime field, then $\pi_{\psi}$ is defined to be the unique root of $X^{p-1}+p$ that satisfies \[ \psi(1)\equiv 1+\pi_{\psi} \pmod{\pi_{\psi}^2}. \] For general $k$, let $k_0$ be the prime field contained in $k$ and fix a nontrivial additive character $\psi_0$ on $\bF_p$. Then, there exists a unique element $a\in\bF_q$ that satisfies \[ \psi(x) = \psi_0\big(\Tr_{k/k_0}(ax)\big)\quad \forall x\in k. \] The element $\pi_{\psi}$ is defined by $\pi_{\psi}=\pi_{\psi_0}\widetilde{a}$, where $\widetilde{a}$ is the Teichm\"uller lift of $a$ in $K$. Conversely, we may recover $\psi$ from $\pi_{\psi}$ as follows. In fact, the radius of convergence of the formal power series $\theta_{\psi}(z)=\exp\big(\pi_{\psi}(z-z^q)\big)$ is strictly greater that $1$, and for each $x\in\bF_q$, we have the equation $\theta_{\psi}(\widetilde{x})=\psi(x)$ \cite[Lemme (1.4)]{Berthelot84Ast}. We note that an element $\pi$ of $K$ is of the form $\pi_{\psi}$ for some non-trivial additive character $\psi$ on $k$ if and only if $\pi^{q-1}=(-p)^{(q-1)/(p-1)}$. In fact, $\pi_{\psi}^{q-1}=(-p)^{(q-1)/(p-1)}$ by the definition of $\pi_{\psi}$, and since different two non-trivial characters on $k$ give different $\pi_{\psi}$'s, each root of $X^{q-1}-(-p)^{(q-1)/(p-1)}$ must be of the form $\pi_{\psi}$. We are now ready for defining the Dwork module $\sL_{\psi}$ associated with the non-trivial additive character $\psi$. It is defined to be the coherent $\sD_{\widehat{\bP^1_V},\bQ}^{\dag}(\pdag{\{\infty\}})$-module \[ \sL_{\psi}\defeq \sD_{\widehat{\bP^1_V},\bQ}^{\dag}(\pdag{\{\infty\}})/\sD_{\widehat{\bP^1_V},\bQ}^{\dag}(\pdag{\{\infty\}})(\partial+\pi_{\psi}) \] with the Frobenius structure \begin{equation} \Phi\colon\sL_{\psi}\to (F^{(s)}_{\bP^1_k/k})^{\ast}\sL_{\psi};\quad e_{\psi}\mapsto \exp\big(-\pi_{\psi}(x-x^q)\big)\otimes e_{\psi}; \label{eq:FrobeniusofDwork} \end{equation} here, $e_{\psi}$ denotes the global section of $\sL_{\psi}$ defined by the global section $1\in A_1(K)^{\dag}$ and $x$ denotes the coordinate. We also denote $\sL_{\psi}$ by $\sL_{\pi_{\psi}}$ and call it the Dwork module associated with $\pi_{\psi}$. We recall some fundamental properties of the Dwork module. \begin{proposition} Let $\psi$ be a non-trivial additive character on $k$. \textup{(i)} Let $L_{\psi}$ be the Dwork $F$-isocrystal associated to $\psi$ \cite{Berthelot84Ast}. Then, ${\sp}_{\plus}L_{\psi}$ is isomorphic to $\sL_{\psi}$. \textup{(ii)} Let $k'$ be a finite extension of $k$. Then, the base extension $\iota_{k'/k}(\sL_{\psi})$ of $\sL_{\psi}$ is isomorphic to $\sL_{\psi\circ\Tr_{k'/k}}$. \textup{(iii)} Let $x$ be an element of $k$ and consider it as a $k$-valued point $i_{x}\colon \Spec(k)\hookrightarrow\bA_k^1$. Then, Frobenius trace of $\sL_{\psi}$ at $x$ is $-q\psi(x)$. \label{prop:propertiesofDwork} \end{proposition} \begin{proof} (i) It is a classical fact \cite[5.2]{Berthelot90}. (ii) Because $\iota_{k'/k}$ is exact and sends $\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})$ to $\sD^{\dag}_{\widehat{\bP^1_{V'}},\bQ}(\pdag{\{\infty\}})$, we have \[ \iota_{k'/k}(\sL_{\psi})=\sD^{\dag}_{\widehat{\bP^1_{V'}}}(\pdag{\{\infty\}})/\sD^{\dag}_{\widehat{\bP^1_{V'}}}(\pdag{\{\infty\}})(\partial+\pi_{\psi}) \] as a $\sD^{\dag}_{\widehat{\bP^1_{V'}}}(\pdag{\{\infty\}})$-module. Since $\pi_{\psi}=\pi_{\psi\circ\Tr_{k'/k}}$, it is the same object as $\sL_{\psi\circ\Tr_{k'/k}}$. The Frobenius structures are also compatible because $\theta_{\psi\circ\Tr_{k'/k}}(z)=\prod_{i=1}^{s'-s}\theta_{\psi}(z^{q^{i-1}})$. (iii) Because of the isomorphism $i_x^{\plus}\sp_{\plus}(L_{\psi})\cong \sp_{\plus}(i_x^{\ast}L_{\psi})(1)[1]$, it suffices to prove that the Frobenius trace of $L_{\psi}$ at $x$ equals $\psi(x)$. Since $L_{\psi}$ is an isocrystal of rank one, $i_x^{\ast}L_{\psi}$ is a one-dimensional $K$-vector space. If $x\neq 0$, let $\widetilde{x}$ denote the Teichm\"uller lift of $x$; then the Frobenius structure is defined by \[ i_x^{\ast}L_{\psi}\to i_x^{\ast}L_{\psi}; \quad e\mapsto \exp\big(\pi_{\psi}(\widetilde{x}-\widetilde{x}^q)\big)e=\theta_{\psi}(\widetilde{x})e=\psi(x)e, \] which shows the claim. A similar calculation shows the claim in the case where $x=0$. \end{proof} \vspace{4px} \textbf{Kummer module associated with a multiplicative character.} Let $\alpha$ be an element of $\frac{1}{q-1}\bZ$. We associate with $\alpha$ an overholonomic $F\hyphen\sD^{\dag}_{\widehat{\bP^1}_V,\bQ}(\pdag{\{0,\infty\}})$-module (therefore an object of $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k})$) $\sK_{\alpha}$, which we call ``the Kummer module associated with $\alpha$'', in the following way. We define $\sK_{\alpha}$ to be $B_1(K)^{\dag}/B_1(K)^{\dag}(x\partial-\alpha)$ (considered as a coherent $\sD^{\dag}_{\bP^1,\bQ}(\pdag\{0,\infty\})$-module) with the Frobenius structure \begin{equation} \sK_{\alpha}\to (F_{\bP^1_k/k}^{(s)})^{\ast}\sK_{\alpha}; \quad e_{\alpha}\mapsto x^{-\alpha(q-1)}\otimes e_{\alpha}; \label{eq:FrobeniusofKummer} \end{equation} here, $e_{\alpha}$ denotes the global section of $\sK_{\alpha}$ defined by the element $1\in B_1(K)^{\dag}$. As we did in the case of Dwork modules, we may consider the Kummer modules as being associated with characters. Let $\chi$ be a (multiplicative) character on $k^{\times}$. Let $\alpha_{\chi}$ be an element of $\frac{1}{q-1}\bZ$ satisfying $\chi(\xi)=\widetilde{\xi}^{(q-1)\alpha_{\chi}}$. (This condition does not uniquely determine $\alpha_{\chi}\in\frac{1}{q-1}\bZ$, but $\alpha_{\chi} \bmod \bZ$ is a uniquely determined element of $\frac{1}{q-1}\bZ/\bZ$.) $\sK_{\alpha_{\chi}}$ is also denoted by $\sK_{\chi}$; this does not depend on the choice of $\alpha_{\chi}$ up to isomorphism because $\sK_{\alpha}\cong\sK_{\alpha+1}$ for any $\alpha\in\frac{1}{q-1}\bZ$. The following properties, which correspond to those for Dwork modules (Proposition \ref{prop:propertiesofDwork}), are also available. The proof is also parallel to that of Proposition \ref{prop:propertiesofDwork} and we omit it. \begin{proposition} Let $\chi$ be a character on $k^{\times}$. \textup{(i)} Let $K_{\alpha}$ be the Kummer convergent $F$-isocrystal on $\bG_{\rmm,k}$ overconvergent along $\{0,\infty\}$ defined by $\alpha$. Then, $\sp_{\plus}K_{\alpha}$ is isomorphic to $\sK_{\chi}$. \textup{(ii)} Let $k'$ a finite extension of $k$. Then, the base extension $\iota_{k'/k}(\sK_{\psi})$ of $\sK_{\psi}$ is isomorphic to $\sK_{\psi\circ\Norm_{k'/k}}$. \textup{(iii)} Let $x$ be an element of $k'^{\times}$ and consider it as a $k$-valued point $i_x\colon\{x\}\hookrightarrow\bG_{\rmm,k}$. Then, the Frobenius trace of $\sK_{\chi}$ at $x$ is $-q\chi(x)$. \label{prop:propertiesofKummer} \end{proposition} \subsection{Fourier transform.} Let $\psi$ be a non-trivial additive character on $k$, and we fix it throughout this subsection. We assume that $K$ contains a primitive $p$-th root of unity. Recall from the previous subsection that then an element $\pi_{\psi}$ of $K$ is associated with $\psi$ and that we have the Dwork module $\sL_{\psi}$ in $F\hyphen D^{\rb}_{\ovhol}(\bA^1_k/K)$. The purpose of this subsection is recalling the theory of Fourier transforms of arithmetic $\sD$-modules, which is closely related to the theory of multiplicative convolution. The basic references are articles of Huyghe \cite{NootHuyghe04, NootHuyghe13}. Her theory of $p$-adic Fourier transforms is the one with respect to the character $\psi$ which can be expressed by $\psi_0\circ\Tr_{k/k_0}$ (in other words, for which $\pi_{\psi}$ is a root of $X^{p-1}+p$). However, her argument remains valid for general $\psi$, we use the theory of $p$-adic Fourier transform for a general $\psi$. First, we define the ``integral kernel'' of the arithmetic Fourier transform. Let $\mu\colon \bA^1_k\times\bA^1_k\to\bA^1_k$ be the multiplication morphism $(x,y)\mapsto xy$, and define $\sL_{\mu,\psi}\defeq \mu^!(\sL_{\psi}[-1])$. Then, $\sL_{\mu,\psi}$ is an object of $F\hyphen D^{\rb}_{\ovhol}(\bA^1_k\times\bA^1_k/K)$ concentrated on degree $0$. Let $(\sP', T')$ denote the d-couple defined by $\sP'=\widehat{\bP^1_V}\times\widehat{\bP^1_V}$ and $T'=(\{\infty\}\times\bP^1_k)\cup(\bP^1_k\times\{\infty\})$. We also regard $\sL_{\mu,\psi}$ as an object of $D^{\rb}_{\coh}\big(\sD^{\dag}_{\sP',\bQ}(\pdag{T'})\big)$. \begin{definition} The functor \[ \FT_{\psi}\colon D^{\rb}_{\coh}\big(\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})\big) \longrightarrow D^{\rb}_{\coh}\big(\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})\big) \] is defined by sending $\sM$ in $D^{\rb}_{\coh}\big(\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})\big)$ to \[ \FT_{\psi}(\sM) = \widetilde{\pr_2}_{,\plus}\big(\sL_{\mu,\psi}\otimes^{\dag}_{\sO_{\sP',\bQ}(\pdag{T'})}\widetilde{\pr_1}^!\sM\big)[-2], \] where $\widetilde{\pr_i}\colon (\sP',T')\to(\widehat{\bP^1_V},\{\infty\})$ denotes the smooth morphism of d-couples defined by the $i$-th projection $\widehat{\bP^1_V}\times\widehat{\bP^1_V}\to\widehat{\bP^1_V}$ for $i\in\{1,2\}$. This object $\FT_\psi(\sM)$ is called the \emph{geometric Fourier transform} of $\sM$. We also define the geometric Fourier transform on the category $F\hyphen D^{\rb}_{\ovhol}(\bA^1_k/K)$. Namely, we define \[ \FT_{\psi}\colon F\hyphen D^{\rb}_{\ovhol}(\bA^1_k/K)\to F\hyphen D^{\rb}_{\ovhol}(\bA^1_k/K);\quad \sM\mapsto\pr_{2,+}\big(\sL_{\mu,\psi}\widetilde{\otimes}\pr_1^!\sM\big), \] where $\pr_i$'s are $i$-th projections $\bA_k^1\times\bA_k^1\to\bA^1_k$. \end{definition} These two functors are compatible with each other because it is true for each functor appearing in the definitions. In defining the geometric Fourier transform, we could have used $\pr_{2,!}$ instead of $\pr_{2,\plus}$, and Huyghe proved that these two ``geometric Fourier transforms'' coincide. The following proposition is a special case of her result. \begin{proposition}[{\cite[Theorem 3.2]{NootHuyghe13}}] Let $\sM$ be an object of $F\hyphen D_{\ovhol}^{\rb}(\bA^1/K)$. Then, the natural morphism \[ \pr_{2,!}\big(\sL_{\mu,\psi}\widetilde{\otimes}\pr_1^!\sM\big) \longrightarrow \FT_{\psi}(\sM) \] is an isomorphism. \label{prop:miracleofft} \end{proposition} \begin{remark} The original form \cite[Theorem 3.2]{NootHuyghe13} of this proposition by Huyghe is quite stronger. It states that for all objects of $D_{\coh}^{\rb}\big(\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag\{\infty\})\big)$, the source of this morphism is well-defined, that so is the morphism itself, and that it is an isomorphism. \end{remark} Under the identification in \ref{para:DoverA1}, the geometric Fourier transform of an overholonomic $\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag\{\infty\})$-module is explicitly described as follows. \begin{proposition}[{\cite[Th\'eor\`eme 5.3.1]{NootHuyghe04}}] \label{prop:geomftandnaiveft} Let $\varphi_{\pi_{\psi}}\colon A_1(K)^{\dag}\to A_1(K)^{\dag}$ be the ring automorphism defined by $\varphi_{\pi_{\psi}}(x)=-\partial/\pi_{\psi}$ and $\varphi_{\pi_{\psi}}(\partial)=\pi_{\psi} x$. Let $\sM$ be a coherent $A_1(K)^{\dag}$-module and denote by $\varphi_{\pi_{\psi},\ast}\sM$ the coherent $A_1(K)^{\dag}$-module obtained by letting $A_1(K)^{\dag}$ act on $\sM$ via $\varphi_{\pi_{\psi}}$. Then, we have a natural isomorphism $\FT_{\psi}(\sM) \cong \varphi_{\pi_{\psi},\ast}\sM[-1].$ \label{prop:globalsectionofft} \end{proposition} \section{Multiplicative Convolution of Arithmetic $\sD$-modules.} In this section, we construct a theory of multiplicative convolutions of arithmetic $\sD$-modules. The contents of this section is basically a direct translation of the classical arguments for the convolution of complex $\sD$-modules \cite[Chapter 5]{Katz90ESDE} and for $\ell$-adic perverse sheaves \cite[Chapter 7]{Katz90ESDE} to the $p$-adic setting. \subsection{Definition and basic properties.} \begin{definition} Let $\sM, \sN$ be objects of $F$-$D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$. Then, we define two \emph{convolutions}, $\sM\ast_{!}\sN$ and $\sM\ast_{\plus}\sN$, by \[ \sM\ast_{!}\sN\defeq\mu_{!}\big(\sM\boxtimes\sN\big),\quad \sM\ast_{\plus}\sN\defeq \mu_{\plus}\big(\sM\boxtimes\sN\big), \] where $\mu\colon\bG_{\rmm,k}\times\bG_{\rmm,k}\to\bG_{\rmm,k}$ denotes the multiplication morphism. \end{definition} We here state some basic properties of these convolutions. \begin{proposition} Let $\ast$ be the convolution $\ast_{!}$ or $\ast_{\plus}$. Then, the following properties hold. \begin{enumerate} \item[\textup{(i)}] The delta module $\delta_1$ is the unit of the convolution, that is, we have natural isomorphisms $\delta_1\ast \sM\cong\sM$ and $\sM\ast\delta_1\cong\sM$ for all $\sM\in F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$. \item[\textup{(ii)}] The convolution is commutative, that is, we have a natural isomorphism $\sM_1\ast\sM_2\cong\sM_2\ast\sM_1$ for all $\sM_1,\sM_2\in F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$. \item[\textup{(iii)}] The convolution is associative, that is, we have a natural isomorphism $(\sM_1\ast\sM_2)\ast\sM_3\cong\sM_1\ast(\sM_2\ast\sM_3)$ for all $\sM_1,\sM_2,\sM_3\in F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$. \end{enumerate} \end{proposition} \begin{proof} This is an easy exercise of six functors. As an example, let us prove (i). Let $\iota_1\colon\{1\}\hookrightarrow \bG_{\rmm,k}$ be the inclusion. Then, we have $\delta_1=\iota_{1,\plus}(\sO_{\{1\},\bQ})$, and because $\iota_1$ is proper we also have $\delta_1=\iota_{1,!}(\sO_{\{1\},\bQ})$. Then, the K\"unneth formula shows that $\delta_1\boxtimes\sM\cong (\iota_1\times\id_{\bG_{\rmm}})_{?}(\sO_{\{1\},\bQ}\boxtimes\sM)$ for each $?\in\{\plus, !\}$, where the exterior tensor product in the right-hand side is taken on $\{1\}\times\bG_{\rmm,k}$. Under the identification $\{1\}\times\bG_{\rmm,k}\cong\bG_{\rmm,k}$, the object $\sO_{\{1\},\bQ}\boxtimes\sM$ is identified with $\sM$, and the product map $\mu$ is identified with the identity map, which shows that $\delta_1\ast\sM\cong\sM$. The isomorphism $\sM\ast\delta_1\cong\sM$ is proved in the same way. \end{proof} \begin{proposition} \label{prop:frobeniusofconvolution} Let $\sM, \sN$ be objects of $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$, let $x$ be an element of $k$ and consider it as a $k$-valued point of $\bG_{\rmm,k}$. For each element $y$ of $k$, we denote by $\Tr(\Phi|\sM_y)$ \resp{$\Tr(\Phi|\sN_y)$} the Frobenius trace of $\sM$ \resp{$\sN$} at the $k$-valued point defined by $y$. Then, the Frobenius trace of $\sM\ast_!\sN$ at the $k$-valued point defined by $x$ equals \[ \sum_{x_1,x_2\in k, x_1x_2=x}\Tr(\Phi|\sM_{x_1})\Tr(\Phi|\sN_{x_2}). \] \end{proposition} \begin{proof} Let $\mu\colon\bG_{\rmm,k}\times\bG_{\rmm,k}\to\bG_{\rmm,k}$ denote the multiplication map and let $\pr_i\colon\bG_{\rmm,k}\times\bG_{\rmm,k}\to\bG_{\rmm,k}$ denote the $i$-th projection for $i\in\{1,2\}$. Recall that $\sM\ast_!\sN=\mu_!(\sM\boxtimes\sN)\cong\mu_!(\pr_1^{\plus}\sM\otimes\pr_2^{\plus}\sN)$. Let $G_x$ denote the fiber product of $i_x\colon \Spec(k)\hookrightarrow\bG_{\rmm,k}$ and $\mu$, let $i_G$ denote the inclusion $G_x\hookrightarrow\bG_{\rmm,k}\times\bG_{\rmm,k}$, and let $f\colon G_x\to\Spec(k)$ denote the structure morphism. Then, by the base change theorem, we have \[ i_x^{\plus}(\sM\ast_!\sN)\cong f_!i_{G_x}^{\plus}(\pr_1^{\plus}\sM\otimes\pr_2^{\plus}\sN). \] The Frobenius trace of this object at a closed point can be calculated by the trace formula \cite[A.4.1]{Abe13Langlands}. Each $k$-valued point of $G_x$ is of the form $(i_{x_1},i_{x_2})$, where $x_1, x_2\in k^{\times}$ with $x_1x_2=x$, and where $i_{x_j}$ is the $k$-valued point $\Spec(k)\hookrightarrow\bG_{\rmm,k}$ defined by $x_j$. For this $k$-valued point, we have \[ (i_{x_1},i_{x_2})^{\plus}i_{G_x}^{\plus}(\pr_1^{\plus}\sM\otimes\pr_2^{\plus}\sN) \cong (i_{x_1}^{\plus}\sM)\otimes(i_{x_2}^{\plus}\sN) \cong (i_{x_1}^{\plus}\sM)\widetilde{\otimes}(i_{x_2}^{\plus}\sN), \] where the last isomorphism follows from \cite[Proposition after 5.8]{Abe11}. This shows the claim. \end{proof} \subsection{Relation with Fourier Transforms.} In this subsection, we will show that the convolutions are ``symmetric'' generalizations of the geometric Fourier transform. In this subsection, $\inv\colon\bG_{\rmm,k}\to\bG_{\rmm,k}$ denotes the morphism defined by $x\mapsto x^{-1}$, $\pr_i\colon\bG_{\rmm,k}\times\bG_{\rmm,k}\to\bG_{\rmm,k}$ denotes the $i$-th projections for $i\in\{1,2\}$. \begin{lemma} \label{lem:mcandintegral} For each objects $\sM, \sN\in F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$, we have canonical isomorphisms \[ \sM\ast_{\plus}\sN \cong \pr_{2,\plus}\big(\pr_1^{!}(\inv^{\ast}\sM)\widetilde{\otimes}\mu^{!}\sN\big) \quad\text{and}\quad \sM\ast_{!}\sN \cong \pr_{2,!}\big(\pr_1^{!}(\inv^{\ast}\sM)\widetilde{\otimes}\mu^{!}\sN\big). \] \end{lemma} \begin{proof} The proof goes as in \cite[(5.1.10)]{Katz90ESDE}. Let $\sigma\colon \bG_{\rmm,k}\times\bG_{\rmm,k}\to\bG_{\rmm,k}\times\bG_{\rmm,k}$ the morphism defined by $(x,y)\mapsto (x^{-1},xy)$. Since $\mu=\pr_2\circ\sigma$, $\sM\ast_{\plus}\sN$ \resp{$\sM\ast_{!}\sN$} is isomorphic to $\pr_{2,\plus}\circ\sigma_{\plus}(\pr_1^{!}\sM\widetilde{\otimes}\pr_2^{!}\sN)$ \resp{$\pr_{2,!}\circ\sigma_{\plus}(\pr_1^{!}\sM\widetilde{\otimes}\pr_2^{!}\sN)$ because $\sigma_{\plus}=\sigma_!$}. By using $\inv\circ\pr_1\circ\sigma=\pr_1$, we have for each $?\in\{\plus,!\}$ \[ \sM\ast_{?}\sN \cong \pr_{2,?}\circ\sigma_{\plus} \big(\sigma^{\ast}\pr_1^!(\inv^{\ast}\sM)\widetilde{\otimes}\pr_2^!\sN\big)\cong \pr_{2,?}\big(\pr_1^!(\inv^{\ast}\sM)\widetilde{\otimes}\sigma_{\plus}(\pr_2^!\sN)\big), \] where the second isomorphism is the projection formula applied to $\sigma$. Since $\sigma\circ\sigma=\id$, we have a natural isomorphism $\sigma^{\ast}\cong\sigma_{\plus}$ of functors, which shows that $\sigma_{\plus}\circ\pr_2^!\cong (\pr_2\circ\sigma)^!=\mu^!$. This shows the claim. \end{proof} \begin{proposition} \label{prop:fourierandconv} Assume that $K$ contains a primitive $p$-th root of unity. We denote by $j\colon \bG_{\rmm,k}\hookrightarrow\bA^1_k$ the natural inclusion. Then, for each object $\sM$ of $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$, we have natural isomorphisms \[ j^{\ast}\big(\FT_{\psi}(j_{\plus}\inv^{\ast}\sM)\big) \cong \sM\ast_{\plus} (j^{\ast}\sL_{\psi})[-1] \quad\text{and}\quad j^{\ast}\big(\FT_{\psi}(j_!\inv^{\ast}\sM)\big) \cong \sM\ast_{!} (j^{\ast}\sL_{\psi})[-1]. \] \end{proposition} \begin{proof} The proof of the first isomorphism goes precisely as in the over-$\bC$ case \cite[5.2.3]{Katz90ESDE} and is omitted here. In the same way, we may construct an isomorphism \[ j^{\ast}\bigg( \pr_{2,!}\big(\sL_{\mu,\psi}\widetilde{\otimes}\pr_1^!j_!\inv^{\ast}\sM\big)\bigg)\cong\sM\ast_{!}(j^{\ast}\sL_{\psi}). \] Now, the isomorphism $\pr_{2,!}(\sL_{\mu,\psi}\widetilde{\otimes}\pr_1^!\sN)\cong \FT_{\psi}(\sN)$ (Proposition \ref{prop:miracleofft}) applied to $\sN=j_!\inv^{\ast}\sM$ proves the second isomorphism. \end{proof} \section{$p$-adic hypergeometric $\sD$-modules.} In this section, we introduce the $p$-adic hypergeometric $\sD$-modules in two ways; one is given by explicit hypergeometric equations, and the other uses the multiplicative convolutions. After that, we compare these two $\sD$-modules and investigate the properties of them. \subsection{Hypergeometric differential operators.} First, let us define a hypergeometric differential $\sD^{\dag}$-module on $\bG_{\rmm,k}$ by using hypergeometric differential operators. \begin{definition} \label{def:hgoperators} Let $\pi$ be a non-zero element of $K$, and let $\alpha_1,\ldots,\alpha_m,\beta_1,\ldots,\beta_n$ be elements of $K$. We write the sequence $\alpha_1,\ldots,\alpha_m$ by $\balpha$ and $\beta_1,\ldots,\beta_n$ by $\bbeta$. \begin{itemize} \item[\textup{(i)}] We define the hypergeometric operator $\Hyp_{\pi}(\balpha;\bbeta)=\Hyp_{\pi}(\alpha_1,\ldots,\alpha_m;\beta_1,\ldots,\beta_n)$ to be \[ \Hyp_{\pi}(\balpha;\bbeta) \defeq \prod_{i=1}^m (x\partial-\alpha_i) - (-1)^{m+np}\pi^{m-n}x\prod_{j=1}^n(x\partial-\beta_j) \] \item[\textup{(ii)}] We define a $B_1(K)^{\dag}$-module $\sH_{\pi}(\balpha;\bbeta)=\sH_{\pi}(\alpha_1,\ldots,\alpha_m;\beta_1,\ldots,\beta_n)$ by \[ \sH_{\pi}(\balpha;\bbeta) \defeq B_1(K)^{\dag}/B_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta). \] \end{itemize} \end{definition} \begin{remark} Recall from \ref{para:DoverGm} that we are identifying the category of coherent $\sD_{\widehat{\bP^1_V},\bQ}^{\dag}(\pdag{\{0,\infty\}})$-modules and that of coherent $B_1(K)^{\dag}$-modules. Since $\sH_{\pi}(\balpha; \bbeta)$ is a coherent $B_1(K)^{\dag}$-module by definition, it is also regarded as a coherent $\sD_{\bP^1,\bQ}^{\dag}(\pdag{\{0,\infty\}})$-module. However, the fact that $\sH_{\pi}(\balpha; \bbeta)$ is a coherent $\sD_{\widehat{\bP^1_V},\bQ}^{\dag}(\pdag{\{0,\infty\}})$-module does not mean that $\sH_{\pi}(\balpha; \bbeta)$ is coherent as a $\sD_{\widehat{\bP^1_V},\bQ}^{\dag}$-module (\cite[Remarque after Th\'eor\`eme 4.4.12]{BerthelotI}). This is a reason why we cannot immediately conclude that it is an overholonomic $\sD_{\widehat{\bP^1_V},\bQ}^{\dag}$-module. We later show that it is in fact overholonomic if $\pi$ is the element associated with a non-trivial additive character and if $\alpha_i$'s and $\beta_j$'s are the elements associated with multiplicative characters under a suitable condition. \end{remark} \begin{lemma} \label{lem:calcofhyp} Under the notation in Definition \ref{def:hgoperators}, $\sH_{\pi}(\balpha; \bbeta)$ has the following properties. \begin{itemize} \item[\textup{(i)}] Let $\widetilde{\inv}\colon(\widehat{\bP^1_V},\{0,\infty\})\to(\widehat{\bP^1_V},\{0, \infty\})$ denote the morphism of d-couples defined by the inversion morphism $\overline{\inv}\colon\widehat{\bP^1_V}\to\widehat{\bP^1_V}$. (This morphism $\widetilde{\inv}$ realizes the inversion morphism $\inv\colon\bG_{\rmm,k}\to\bG_{\rmm,k}$.) Then, $\widetilde{\inv}^{\ast}\sH_{\pi}(\balpha; \bbeta)$ is isomorphic to $\sH_{(-1)^p\pi}(-\bbeta, -\balpha)$, where $-\balpha$ \resp{$-\bbeta$} denotes the sequence $-\alpha_1,\ldots,-\alpha_m$ \resp{$-\beta_1,\ldots,-\beta_n$}. \item[\textup{(ii)}] Let $\gamma$ be an element of $\frac{1}{q-1}\bZ$. Then, $\sH_{\pi}(\balpha;\bbeta)\otimes^{\dag}_{\sO_{\bP^1,\bQ}(\pdag{\{0,\infty\}})}\sK_{\gamma}$ is isomorphic to $\sH_{\pi}(\balpha+\gamma;\bbeta+\gamma)$, where $\balpha+\gamma$ \resp{$\bbeta+\gamma$} denotes the sequence $\alpha_1+\gamma,\ldots,\alpha_m+\gamma$ \resp{$\beta_1+\gamma,\ldots,\beta_n+\gamma$}. \end{itemize} \end{lemma} \begin{proof} (i) Since $\widetilde{\inv}^{\ast}$ is identified with the base extension via $B_1(K)^{\dag}\to B_1(K)^{\dag}$ defined by $x\mapsto x^{-1}$ and $\partial\mapsto -x^2\partial$, it suffices to show that the left ideal generated by \[ \prod_{i=1}^m(-x\partial-\alpha_i)-(-1)^{m+np}\pi^{m-n}x^{-1}\prod_{j=1}^n(-x\partial-\beta_j) \] equals that generated by $\Hyp_{(-1)^p\pi}(-\bbeta;-\balpha)$. This can be seen by a direct calculation: \begin{align*} & \prod_{i=1}^m(-x\partial-\alpha_i)-(-1)^{m+np}\pi^{m-n}x^{-1}\prod_{j=1}^n(-x\partial-\beta_j)\\ =& (-1)^{m+np+n+1}\pi^{m-n}x^{-1} \left\{\prod_{j=1}^n(x\partial+\beta_j)-(-1)^{n+mp}\big((-1)^p\pi\big)^{n-m}x\prod_{i=1}^m(x\partial+\alpha_i)\right\}\\ =& (-1)^{m+np+n+1}\pi^{m-n}x^{-1}\Hyp_{(-1)^p\pi}(-\bbeta;-\balpha). \end{align*} (ii) We know that $\sK_{\gamma}=B_1(K)^{\dag}/B_1(K)^{\dag}(x\partial-\gamma)$ is isomorphic to $\sO_{\widehat{\bP^1_V},\bQ}(\pdag{\{0,\infty\}})$ as an $\sO_{\widehat{\bP^1_V},\bQ}(\pdag{\{0,\infty\}})$-module. Therefore, the functor $\text{-- }\otimes^{\dag}_{\sO_{\widehat{\bP^1_V},\bQ}(\pdag{\{0,\infty\})}}\sK_{\gamma}$ on the category of coherent $B_1(K)^{\dag}$-modules does not change the underlying $\sO_{\widehat{\bP^1_V},\bQ}(\pdag{\{0,\infty\}})$-module, and $\partial$ acts on the resulting $B_1(K)^{\dag}$-module as the action of $\partial + \gamma x^{-1}$ on the original $B_1(K)^{\dag}$-module. By this change of the action, $x\partial$ goes $x\partial-\gamma$, which proves the assertion. \end{proof} Later, we also have to consider the hypergeometric differential operator ``on $\bA^1_k$'', not only ``on $\bG_{\rmm,k}$''. The following proposition describes the first essential relationships between them. \begin{proposition} Let $\pi$ be a non-zero element of $K$ and let $\alpha_1,\ldots,\alpha_m,\beta_1,\ldots,\beta_n$ be elements of $K$. Let $\widetilde{j}\colon(\bP^1_k,\{0,\infty\})\hookrightarrow(\bP^1_k,\{\infty\})$ denote the morphism of d-couples defined by $\id_{\bP^1_k}$. (This morphism realizes the inclusion morphism $j\colon\bG_{\rmm,k}\hookrightarrow\bA^1_k$.) \begin{itemize} \item[\rm (i)] $\widetilde{j}^{\ast}\big(A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)\big)$ is isomorphic to $\sH_{\pi}(\balpha;\bbeta)$. \item[\rm (ii)] Assume that $|\pi|=|p|^{1/(p-1)}$, that $\alpha_i$'s belong to $\frac{1}{q-1}\bZ\setminus\bZ$ and that $\beta_j$'s belong to $\frac{1}{q-1}\bZ$. Then, the natural morphism \[ A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta) \to \widetilde{j}_{\plus}\widetilde{j}^{\ast}\big(A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)\big) \] of $A_1(K)^{\dag}$-modules is an isomorphism. \end{itemize} \label{prop:connectiontype} \end{proposition} \begin{proof} (i) The assertion follows from the fact that the functor $\widetilde{j}^{\ast}$ is exact on the category of coherent $\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})$-modules. (ii) Let $\sL$ denote the $A_1(K)^{\dag}$-module $A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)$ and let $\varphi\colon \sL\to\widetilde{j}_{\plus}\widetilde{j}^{\ast}\sL$ denote the natural morphism in question. In order to prove that $\varphi$ is an isomorphism, it suffices to prove that the left multiplication $l_x\colon\sL\to\sL$ by $x$ on $\sL$ is bijective. In fact, suppose that we have proved the bijectivity of $l_x$. Since the restriction of $\varphi$ (regarded as a morphism of $\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag{\{\infty\}})$-modules) to $\widehat{\bP_V^1}\setminus\{0\}$ is an isomorphism, it suffices to show that the restriction of $\varphi$ on $\widehat{\bA^1_V}$ is an isomorphism. Let us denote by $\widetilde{i}\colon(\{0\},\emptyset)\to(\widehat{\bP^1_V},\{\infty\})$ the morphism of d-couples defined by the closed immersion $\bar{i}\colon\{0\}\hookrightarrow\widehat{\bP^1_V}$. Then, by the definition of extraordinary pull-back \cite[(1.1.6.1)]{Caro06CM} (cf. \cite[4.3.2]{Berthelot02Ast}), $\widetilde{i}^!\sL$ is the complex $\left[\sL\xrightarrow{l_x}\sL\right]$, where the target is placed at degree zero. Therefore, the bijectivity of $l_x$ is equivalent to $\widetilde{i}^!\sL=0$. Moreover, if we denote by $i'\colon \{0\}\hookrightarrow\bA^1_k$ the inclusion, then $\widetilde{i}^!\sL=0$ implies $i'^!(\sL|_{\widehat{\bA^1_V}})=0$. In the localization triangle \cite[(1.1.6.5)]{Caro06CM} in $D^{\rb}_{\coh}(\sD^{\dag}_{\widehat{\bA^1_V},\bQ})$, \[ \bR\underline{\Gamma}^{\dag}_{\{0\}}\big(\sL|_{\widehat{\bA^1_V}}\big)\longrightarrow \sL|_{\widehat{\bA^1_V}}\longrightarrow \big(\widetilde{j}_{\plus}\widetilde{j}^{\ast}\sL\big)|_{\widehat{\bA^1_V}}\longrightarrow +1, \] the first term is isomorphic to $i'_{\plus}\circ i'^{!}\big(\sL|_{\widehat{\bA^1_V}}\big)$ \cite[Corollaire 3.4.7, Th\'eor\`eme 3.4.9 and the first point in Remarques 3.4.10]{Caro12}. This shows that the restriction of $\varphi$ on $\widehat{\bA^1_V}$ is an isomorphism, which concludes the proof. Now, we show that $l_x$ is bijective. We firstly work on the injectivity. Let $P, Q$ be elements of $A_1(K)^{\dag}$ that satisfies $xP=Q\Hyp_{\pi}(\balpha;\bbeta)$. We show that $Q\in xA_1(K)^{\dag}$; then, since $x$ is not a zero-divisor in $A_1(K)^{\dag}$, we get that $P\in A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)$ and the injectivity follows. In order to show that $Q\in xA_1(K)^{\dag}$, we may assume that $Q$ is of the form $Q=\sum_{l=0}^{\infty}c_l\partial^{[l]}$, where $c_l$'s are elements of $K$ satisfying $\exists C>0, \exists\eta<1, \forall l, |c_l|<C\eta^l$. Then, because $\Hyp_{\pi}(\balpha;\bbeta)=\prod_i(x\partial-\alpha_i)-(-1)^{m+np}\pi^{m-n} x\prod_j(x\partial-\beta_j)$, and because $\partial^{[l]}x\equiv\partial^{[l-1]} \pmod{xA_1(K)^{\dag}}$, we have \begin{align*} Q\Hyp_{\pi}(\balpha;\bbeta)&\equiv \sum_{l=0}^{\infty}c_l\prod_{i=1}^m(l-\alpha_i)\partial^{[l]} -(-1)^{m+np}\pi^{m-n}\sum_{l=1}^{\infty}c_l\prod_{j=1}^n(l-1-\beta_j)\partial^{[l-1]}\\ & \hspace{200pt} \pmod{xA_1(K)^{\dag}}. \end{align*} By assumption, the left-hand side belongs to $xA_1(K)^{\dag}$, which shows that, for each $l$, \[ c_l\prod_{i=1}^m(l-\alpha_i) = (-1)^{m+np}\pi^{m-n}c_{l+1}\prod_{j=1}^n(l-\beta_j). \] Fix a positive integer $l$ that exceeds all $\alpha_j$'s and $\beta_j$'s. Then, for each natural number $k$, we have \[ c_{l+k}=(-1)^{k(m+np)}\pi^{-k(m-n)}\frac{\prod_{i=1}^m(l+k-1-\alpha_i)(l+k-2-\alpha_i)\dots(l-\alpha_i)} {\prod_{j=1}^n(l+k-1-\beta_j)(l+k-2-\beta_j)\dots(l-\beta_j)}c_l. \] Lemma \ref{lem:valofvp} below shows that $\left|(l+k-1-\beta_j)\ldots(l-\beta_j)\right|^{-1}\geq p^{k/(p-1)-1}k^{-1}$ and $\big|(l+k-1-\alpha_i)\ldots(l-\alpha_i)\big|\geq p^{-k/(p-1)}(q-1)^{-1}(l+k-1-\alpha_i)^{-1}$. By these inequalities and $|\pi|=p^{-1/(p-1)}$, we have \[ |c_{l+k}| \geq p^{-m}(q-1)^{-2m}k^{-n}\prod_{i=1}^m(l+k-1-\alpha_i)^{-1}|c_l| \] for each $k$. Since $|c_{l+k}|<C\eta^{l+k}$ for all $k$, we must have $|c_l|=0$, therefore $c_l=0$. Since $l$ supposed to be an arbitrary positive integer exceeding all $\alpha_j$'s and $\beta_j$'s, we get that $Q$ is a finite sum. Now, the recurrence relation for $c_l$ and the assumption that $\alpha_i$'s are not integers, we know that $Q=0$. Next, we show the surjectivity. Given $P\in A_1(K)^{\dag}$, we have to show that there exists $Q, R\in A_1(K)^{\dag}$ such that $xQ=P+R\Hyp_{\pi}(\balpha;\bbeta)$. We may assume that $P$ is of the form $P=\sum_{l=0}^{\infty}c_l\partial^{[l]}$, where $c_l$'s are elements of $K$ satisfying $\exists C>0, \exists\eta<1, \forall l, |c_l|<C\eta^l$. We show that there exists $R\in A_1(K)^{\dag}$ of the form $R=\sum_{d=0}^{\infty}d_l\partial^{[l]}$ that satisfies $P+R\Hyp_{\pi}(\balpha;\bbeta)\in xA_1(K)^{\dag}$. Let $l_0$ be the maximum of the elements in $\Set{\beta_j+1 | j\in\{1,\ldots,n\}}\cap\bZ_{\geq 0}$ if this set is not empty; if it is empty, then let $l_0=0$. Then, we may assume that $c_l=0$ if $l<l_0$ by the following reason. If $A_1(K)$ denotes the usual Weyl algebra with coefficients in $K$, then by our assumption on the parameters, the right multiplication by $\Hyp_{\pi}(\balpha;\bbeta)$ is bijective on $A_1(K)/xA_1(K)$ \cite[2.9.4, (3)$\Rightarrow$(2)]{Katz90ESDE}. This shows that there exists $R'\in A_1(K)$ such that $\sum_{l=0}^{l_0-1}c_l\partial^{[l]}+R'\Hyp_{\pi}(\balpha;\bbeta)\in xA_1(K)$ (The proof in the reference \cite{Katz90ESDE} is given over $\bC$, but it remains valid for all field of characteristic $0$). Now, we assume that $c_l=0$ if $l<l_0$. We put $d_l=0$ if $l<l_0$, and for each $s\geq 0$ we put \begin{equation} d_{l_0+s} = \sum_{t=s}^{\infty}(-1)^{(t-s)(m+np+1)}\pi^{(t-s)(m-n)}\frac{\prod_{j=1}^n(l_0+t-1-\beta_j)\ldots(l_0+s-\beta_j)}{\prod_{i=1}^m(l_0+t-\alpha_i)\ldots(l_0+s-\alpha_i)}c_{l_0+t}; \label{eq:defofd} \end{equation} This infinite series actually converges; in fact, Lemma \ref{lem:valofvp} shows that $\big|(l_0+t-1-\beta_j)\ldots(l_0+s-\beta_j)\big|\leq p^{-(t-s)(p-1)+1}(t-s)$ and that \[ \big|(l_0+t-\alpha_i)\ldots(l_0+s-\alpha_i)\big|^{-1}\leq p^{(t-s-1)/(p-1)}(q-1)(l_0+t-\alpha_i), \] and therefore the norm of each summand in the left-hand side is bounded from above by \[ p^{m/(p-1)+n}(q-1)^mC\cdot (t-s)^m\prod_{i=1}^m(l_0+t-\alpha_i)\cdot\eta^{l_0+t}. \] This converges to $0$ as $t\to\infty$, which shows that $d_{l_0+s}$ is well-defined. Now, we put $R=\sum_{l=0}^{\infty}d_l\partial^{[l]}$. Then, by the bound calculated above, $R$ is an element of $A_1(K)^{\dag}$. In fact, we have $|d_{l_0+s}|\leq C'\max\big\{(t-s)^m\prod_i(l_0+t-\alpha_i)\eta^{l_0+t}\big\}$ for a constant $C'>0$, where the max is taken for $t\geq s$. By choosing $\eta'$ satisfying $\eta>\eta'>1$, we have $|d_{l_0+s}|\leq C'\eta'^{l_0+t}$ for sufficiently large $s$, which shows that $R\in A_1(K)^{\dag}$. Finally, we show that $R$ satisfies $P+R\Hyp_{\pi}(\balpha;\bbeta)\in xA_1(K)^{\dag}$. This is equivalent to showing that \[ d_l\prod_{i=1}^m(l-\alpha_i)-(-1)^{(m+np)}\pi^{m-n}d_{l+1}\prod_{j=1}^n(l-\beta_j)+c_l=0 \] for all $l\geq 0$. It trivially holds if $l<l_0-1$ because $d_l=d_{l+1}=c_l=0$ in this case; it also holds if $l=l_0-1$ because $d_l=c_l=0$ and $l-\beta_j=0$ for some $j$; otherwise, we may check it directly by using (\ref{eq:defofd}). This completes the proof of the surjectivity of $l_x$, and therefore of the proposition. \end{proof} \begin{lemma} Let $l, N$ be natural numbers that satisfies $l\leq N$, and let $\alpha$ be an element of $\frac{1}{q-1}\bZ$. Then, we have \[ \left|\prod_{s=l}^N(s-\alpha)\right|\leq p^{-(N-l+1)/(p-1)+1}(N-l+1) \] If $\alpha<l$, then we also have \[ \left|\prod_{s=l}^N(s-\alpha)\right|\geq p^{-(N-l+1)/(p-1)}(q-1)^{-1}(N-\alpha)^{-1} \] \label{lem:valofvp} \end{lemma} \begin{proof} Let $m$ be a positive integer, and let $t_m$ be the number of $(s-\alpha)$'s for $s=l,\ldots,N$ that belongs to $p^m\bZ_{(p)}$: \[ t_m \defeq \#\Set{ s\in\{l,\ldots,N\} | s-\alpha\in p^m\bZ_{(p)} }. \] Then, if $v_p$ denotes the $p$-adic valuation so that $v_p(p)=1$, we have $v_p\left(\prod_{s=l}^N(s-\alpha)\right) = \sum_{m=1}^{\infty} t_m$. Because $\alpha\in\bZ_{(p)}$, there is exactly one multiple of $p^m$ in every $p^m$ successive $(s-\alpha)$'s, and as a result we have $\floor*{\frac{N-l+1}{p^m}} \leq t_m\leq \floor*{\frac{N-l+1}{p^m}}+1$. Moreover, in case $\alpha<l$, we have $t_m=0$ unless $p^m\leq(q-1)(N-\alpha)$; in fact, $s-\alpha$ (for $s=l,\ldots,N$) is a multiple of $p^m$ if and only if so is the integer $(q-1)(s-\alpha)$. Now, since \[ \frac{N-l+1}{p-1}-\log_p(N-l+1)-1\leq \sum_{m=1}^{\infty}\floor*{\frac{N-l+1}{p^m}} \leq\frac{N-l+1}{p-1}, \] $v_p\left(\prod_{s=l}^N(s-\alpha)\right)=\sum_{m=1}^{\infty}t_m$ satisfies \[ \sum_{m=1}^{\infty}t_m\geq \sum_{m=1}^{\infty}\floor*{\frac{N-l+1}{p^m}}\geq \frac{N-l+1}{p-1}-\log_p(N-l+1)-1 \] in general. Moreover, if $\alpha<l$, then by the discussion above we have \[ \displaystyle\sum_{m=1}^{\infty}t_m=\sum_{m=1}^{\log_p(q-1)(N-\alpha)}t_m \] and it satisfies \[ \sum_{m=1}^{\log_p(q-1)(N-\alpha)}t_m \leq \sum_{m=1}^{\infty}\floor*{\frac{N-l+1}{p^m}}+\log_p(q-1)(N-\alpha)\leq\frac{N-l+1}{p-1}+\log_p(q-1)(N-\alpha). \] This shows the assertion. \end{proof} \subsection{Arithmetic hypergeometric $\sD$-modules and convolution} In this section, firstly, we give another construction of arithmetic hypergeometric $\sD$-modules. Under this definition, these $\sD$-modules are overholonomic and have Frobenius structures by nature. Secondly, we compare these $\sD$-modules with the ones given in the previous subsection. The comparison is the main part of this article. \emph{In the remaining part of this article, we always assume that $K$ has a primitive $p$-th root of unity.} \begin{definition} \label{def:defofh} Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\dots,\chi_m$ and $\rho_1,\dots,\rho_n$ be multiplicative characters on $k^{\times}$. Let $j\colon\bG_{\rmm,k}\hookrightarrow\bA^1_k$ denote the inclusion. We denote the sequence $\chi_1,\ldots,\chi_m$ by $\bchi$ and $\rho_1,\ldots,\rho_n$ by $\brho$; the empty sequence is denoted by $\emptyset$. Then, we define an object $\sHyp_{\psi,\plus}(\bchi;\brho)$ of $F\hyphen D_{\ovhol}^{\rb}(\bG_{\rmm,k}/K)$ as follows. (i) If $(m,n)=(0,0)$, then $\sHyp_{\psi,\plus}(\emptyset;\emptyset)\defeq \delta_1$. (ii) If $(m,n)=(1,0)$, then $\sHyp_{\psi,\plus}(\chi_1;\emptyset)\defeq j^{\ast}\sL_{\psi}\widetilde{\otimes}\sK_{\chi_1}[2]$. (iii) If $(m,n)=(0,1)$, then $\sHyp_{\pi,\plus}(\emptyset;\rho_1)\defeq\inv^{\ast}\big(j^{\ast}\sL_{\psi^{-1}}\widetilde{\otimes}\sK_{\rho_1^{-1}}\big)[2]$. (iv) Otherwise, $\sHyp_{\psi,\plus}(\bchi;\brho)$ is defined by \[ \sHyp_{\psi,\plus}(\chi_1;\emptyset)\ast_{\plus}\ldots\ast_{\plus}\sHyp_{\psi,\plus}(\chi_m;\emptyset) \ast_{\plus} \sHyp_{\psi,\plus}(\emptyset;\rho_1)\ast_{\plus}\ldots\ast_{\plus}\sHyp_{\psi,\plus}(\emptyset;\rho_n). \] We also define an object $\sHyp_{\psi,!}(\bchi;\brho)$ of $F\hyphen D_{\ovhol}^{\rb}(\bG_{\rmm,k}/K)$ in a similar way. Namely, we put $\sHyp_{\psi,!}(\bchi;\brho)\defeq\sHyp_{\psi,\plus}(\bchi;\brho)$ if $(m,n)\in\big\{(0,0), (1,0), (0,1)\big\}$, and otherwise $\sHyp_{\psi,!}(\bchi;\brho)$ is defined by (iv) above but $\ast_{\plus}$'s are replaced by $\ast_{!}$. \end{definition} \begin{remark} \label{rem:hypandh} If $(m,n)\in\big\{(0,0),(1,0),(0,1)\big\}$, then we immediately get a concrete description of $\sHyp_{\psi,\plus}(\bchi;\brho)$ (and therefore of $\sHyp_{\psi,!}(\bchi;\brho)$) as a $B_1(K)^{\dag}$-module: \begin{align*} \sHyp_{\psi,\plus}(\emptyset;\emptyset) & = B_1(K)^{\dag}/B_1(K)^{\dag}(1-x),\\ \sHyp_{\psi,\plus}(\chi_1 ;\emptyset) & = \big(B_1(K)^{\dag}/B_1(K)^{\dag}(x\partial-\alpha_{\chi_1}+\pi_{\psi} x)\big)[1], \text{ and}\\ \sHyp_{\psi,\plus}(\emptyset;\rho_1) & = \big(B_1(K)^{\dag}/B_1(K)^{\dag}(1-(-1)^p\pi_{\psi}^{-1}x(x\partial-\alpha_{\rho_1}))\big)[1]. \end{align*} Here, $\alpha_{\chi_1}$ \resp{$\alpha_{\rho_1}$} is an element of $\frac{1}{q-1}\bZ$ satisfying $\chi_1(\xi)=\widetilde{\xi}^{(q-1)\alpha_{\chi_1}}$ \resp{$\rho_1(\xi)=\widetilde{\xi}^{(q-1)\alpha_{\rho_1}}$}. The first equation follows from the definition of the delta module, and the other two equations can be proved by the same calculation as we did in Lemma \ref{lem:calcofhyp}. \end{remark} This remark shows that, for these $(m,n)$'s, then $\sHyp_{\psi,\plus}(\bchi;\brho)$ and $\sHyp_{\psi,!}(\bchi;\brho)$ are isomorphic to $\sH_{\pi_{\psi}}(\balpha;\bbeta)$ as $B_1(K)^{\dag}$-modules modulo a shift of degree. The aim of this section is to generalize this fact to the case where $m$ and $n$ are larger. Before doing that, we prove the similar formula for $\sHyp_{\psi}$'s as Lemma \ref{lem:calcofhyp}. \begin{lemma} Under the notation in Definition \ref{def:defofh}, assume that $(m,n)\neq(0,0)$. Then, $\sHyp_{\psi,?}(\bchi;\brho)$ has the following properties where $?$ denotes $\plus$ or $!$. \label{lem:calcofh} \begin{itemize} \item[(i)] $\inv^{\ast}\sHyp_{\psi,?}(\bchi;\brho)$ is isomorphic to $\sHyp_{\psi^{-1},?}(\brho^{-1};\bchi^{-1})$, where $\brho^{-1}$ \resp{$\bchi^{-1}$} denotes the sequence $\rho_1^{-1},\dots,\rho_m^{-1}$ \resp{$\chi_1^{-1},\dots,\chi_n^{-1}$}. \item[(ii)] Let $\gamma$ be a character on $k^{\times}$. If $(m,n)\neq(0,0)$, then $\sHyp_{\psi,?}(\bchi;\brho)\widetilde{\otimes}(\sK_{\gamma}[1])$ is isomorphic to $\sHyp_{\psi,?}(\bchi\gamma;\brho\gamma)$, where $\bchi\gamma$ \resp{$\brho\gamma$} denotes the sequence $\chi_1\gamma,\dots,\chi_m\gamma$ \resp{$\rho_1\gamma,\dots,\rho_n\gamma$}. \end{itemize} \end{lemma} \begin{proof} (i) If $(m,n)=(1,0)$ or $(m,n)=(0,1)$, then the assertion is obvious by definition. Otherwise, it follows from the formula $\inv^{\ast}(\sM\ast_{?}\sN)\cong(\inv^{\ast}\sM)\ast_{?}(\inv^{\ast}\sN)$, whose proof is immediate. (ii) If $(m,n)=(1,0)$, then the claim directly follows from the formula $\sK_{\rho}\widetilde{\otimes}\sK_{\chi}[1]\cong \big(\sK_{\rho}\otimes^{\dag}_{\sO(\pdag\{0,\infty\})}\sK_{\chi}[1]\big)[-1]\cong\sK_{\rho\chi}$. If $(m,n)=(0,1)$, then it suffices to use this formula and the formula $\inv^{\ast}\sK_{\chi^{-1}}\cong\sK_{\chi}$ twice. In order to prove the assertion in the general case, we first note that $\mu^!\sK_{\gamma}\cong\sK_{\gamma}\boxtimes\sK_{\gamma}[1]$. Therefore, for arbitrary objects $\sM$ and $\sN$ in $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm}/K)$, we have \begin{align*} (\sM\widetilde{\otimes}\sK_{\gamma}[1])\ast_{\plus}(\sN\widetilde{\otimes}\sK_{\gamma}[1]) &\cong\mu_{\plus}\big((\sM\boxtimes\sN)\widetilde{\otimes}(\sK_{\gamma}\boxtimes\sK_{\gamma})\big)[2]\\ & \cong (\sM\ast_{\plus}\sN)\otimes\sK_{\gamma}[1]. \end{align*} This fact and the assertion for $(m,n)=(1,0),(0,1)$ finish the proof. \end{proof} \begin{lemma} Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\dots,\chi_m$ and $\rho_1,\dots,\rho_n$ be characters on $k^{\times}$. Let $k'$ be a finite extension of $k$. We put $\psi'\defeq\psi\circ\Tr_{k'/k}$, $\chi'_i\defeq\chi_i\circ\Norm_{k'/k}$ and $\rho'_j\defeq\rho_j\circ\Norm_{k'/k}$. Moreover, we denote by $\bchi$ \resp{$\brho$} the sequence $\chi_1,\dots,\chi_m$ \resp{$\rho_1,\dots,\rho_n$} and by $\bchi'$ \resp{$\brho'$} the sequence $\chi'_1,\dots,\chi'_m$ \resp{$\rho'_1,\dots,\rho'_n$}. Then, we have \[ \iota_{k'/k}\big(\sHyp_{\psi,\plus}(\bchi,\brho)\big)=\sHyp_{\psi',\plus}(\bchi',\brho') \] and \[ \iota_{k'/k}\big(\sHyp_{\psi,!}(\bchi,\brho)\big)=\sHyp_{\psi',\plus}(\bchi',\brho'). \] \label{lem:scalarextofhyp} \end{lemma} \begin{proof} If $(m,n)=(0,0)$, the assertion is obvious. If $(m,n)\neq (0,0)$, then it follows from the description of scalar extension of Dwork modules and of Kummer modules (Subsection 1.3) and from the fact that the scalar extension commutes with the six-functor formalism. \end{proof} \begin{theorem} Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\dots,\chi_m$, $\rho_1,\ldots,\rho_n$ be characters on $k^{\times}$. Let $\balpha$ \resp{$\bbeta$} denote the sequence $\alpha_1,\dots,\alpha_m$ \resp{$\beta_1,\dots,\beta_n$}, where $\alpha_i$ \resp{$\beta_j$} is an element of $\frac{1}{q-1}\bZ$ that satisfies $\chi_i(\xi)=\widetilde{\xi}^{(q-1)\alpha_i}$ \resp{$\rho_j(\xi)=\widetilde{\xi}^{(q-1)\beta_j}$}. We assume that $(m,n)\neq(0,0)$ and that $\chi_i\neq\rho_j$ for any $i, j$. Then, we have an isomorphism \[ \sHyp_{\psi,\plus}(\bchi;\brho) \cong \sH_{\pi_{\psi}}(\balpha;\bbeta)[m+n]. \] as $\sD^{\dag}_{\widehat{\bP^1_V},\bQ}(\pdag\{0,\infty\})$-modules. \label{thm:hypandconv} \end{theorem} \begin{proof} We prove it by induction on $m+n$. The proof in the case $(m,n)=(1,0), (0,1)$ is already explained in Remark \ref{rem:hypandh}. First, by Lemma \ref{lem:calcofhyp} (i) and Lemma \ref{lem:calcofh} (i), we may assume that $m>0$. Moreover, by Lemma \ref{lem:calcofhyp} (ii) and Lemma \ref{lem:calcofh} (ii), we may assume that $\chi_1=\boldsymbol{1}$ and $\alpha_1=0$. Now, note that $\sHyp_{\psi,\plus}(\boldsymbol{1};\emptyset)=j^{\ast}\sL_{\pi}\widetilde{\otimes}\sK_{\boldsymbol{1}}[2]=j^{\ast}\sL_{\pi}[1]$. It therefore suffices to show the isomorphism $\sHyp_{\psi,\plus}(\bchi';\brho)\ast_{\plus} (j^{\ast}\sL_{\psi}[1])\cong \sH_{\pi_{\psi}}(\balpha;\bbeta)[1]$ of $B_1(K)^{\dag}$-modules, where $\bchi'$ denotes the sequence $\chi_2,\ldots,\chi_m$. Firstly, Proposition \ref{prop:fourierandconv} shows that \[ \sHyp_{\psi,\plus}(\bchi';\brho)\ast_{\plus}(j^{\ast}\sL_{\psi}[1]) \cong j^{\ast}\big(\FT_{\psi}(j_{\plus}\inv^{\ast}\sHyp_{\psi,\plus}(\bchi';\brho))\big). \] By Lemma \ref{lem:calcofh} (i), we have \[ \inv^{\ast}\sHyp_{\psi,\plus}(\bchi';\brho) \cong\sHyp_{\psi^{-1},\plus}(\brho^{-1};\bchi'^{-1}), \] and by induction hypothesis, it is isomorphic to \[ B_1(K)^{\dag}/B_1(K)^{\dag}\Hyp_{\pi_{\psi^{-1}}}(-\bbeta-1;-\balpha'-1)[m+n-1] \] as a $B_1(K)^{\dag}$-module. Moreover, by Proposition \ref{prop:connectiontype} (i), it is isomorphic to \[ \widetilde{j}^{\ast}\left(A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi_{\psi^{-1}}}(-\bbeta-1;-\balpha'-1)\right)[m+n-1]. \] Since there are no integers in the sequence $-\bbeta-1$ by our assumption, we conclude by using Proposition \ref{prop:connectiontype} (ii) that \[ j_{\plus}\inv^{\ast}\sHyp_{\psi,\plus}(\bchi';\brho)[m+n]\cong A_1(K)^{\dag}/A_1(K)^{\dag} \Hyp_{\pi_{\psi^{-1}}}(-\bbeta-1;-\balpha'-1)[m+n-1]. \] Now, we may compute the Fourier transform of this object by using Proposition \ref{prop:geomftandnaiveft} and the formula $\pi_{\psi^{-1}}=(-1)^p\pi_{\psi}$, and the result is $A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi_{\psi}}(\balpha;\bbeta)[m+n]$. In fact, since under the notation in Proposition \ref{prop:globalsectionofft} we have $\varphi_{\pi_{\psi}}(x\partial)=-\partial x=-x \partial - 1$, \begin{align*} & \varphi_{\pi_{\psi}}\Hyp_{\pi_{\psi^{-1}}}(-\bbeta-1;-\balpha'-1) \\ = & \varphi_{\pi_\psi}\left(\prod_{j=1}^n(x\partial+\beta_j+1)-(-1)^{n+(m-1)p}\big((-1)^p\pi_{\psi}\big)^{n-m+1}x\prod_{i=2}^m(x\partial+\alpha_i+1)\right)\\ = & \prod_{j=1}^n(-x\partial+\beta_j)-(-1)^{n(p-1)}\pi_{\psi}^{n-m+1}\frac{-\partial}{\pi_{\psi}}\prod_{i=2}^m( -x\partial+\alpha_i)\\ = & \frac{(-1)^{n(p-1)+m-1}}{\pi_{\psi}^{m-n}x}\left\{x\partial\prod_{i=2}^m(x\partial-\alpha_i)-(-1)^{m+np}\pi_{\psi}^{m-n}\prod_{j=1}^n(x\partial-\beta_j)\right\}. \end{align*} Again by using Proposition \ref{prop:connectiontype} (i), it shows the assertion. \end{proof} \begin{corollary} Let $\pi$ be an element of $K$ that satisfies $\pi^{q-1}=(-p)^{(q-1)/(p-1)}$, and let $\alpha_1,\dots,\alpha_m,\beta_1,\dots,\beta_n$ be elements of $\frac{1}{q-1}\bZ$ that satisfy $\alpha_i-\beta_j\not\in\bZ$ for any $i,j$. Then, the $B_1(K)^{\dag}$-module $\sH_{\pi}(\balpha; \bbeta)$ and the $A_1(K)^{\dag}$-module $A_1(K)^{\dag}/\Hyp_{\pi}(\balpha;\bbeta)A_1(K)^{\dag}$ are overholonomic and have a Frobenius structure. \label{cor:overholonomic} \end{corollary} \begin{proof} By assumption and Theorem \ref{thm:hypandconv}, $\sH_{\pi}(\balpha;\bbeta)$ is isomorphic to $\sHyp_{\psi,\plus}(\bchi;\brho)$ for some $\psi$, $\chi_i$'s and $\rho_j$'s satisfying the assumptions of Theorem \ref{thm:hypandconv}. Therefore, the assertion for $\sH_{\pi}(\balpha; \bbeta)$ now follows. The assertion for $A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)$ follows from the theorem together with Proposition \ref{prop:connectiontype} (i), (ii). \end{proof} \subsection{Comparison of two hypergeometric $\sD$-modules associated with characters} In Definition \ref{def:defofh}, we defined two versions of hypergeometric $\sD$-modules associated with characters; one uses $\ast_{!}$ and the other uses $\ast_{\plus}$. The goal of this subsection is to prove that these are naturally isomorphic to each other under the hypothesis in Theorem \ref{thm:hypandconv}. We begin with a variant of Proposition \ref{prop:connectiontype} (ii). \begin{proposition} Let $\alpha_1,\ldots,\alpha_m,\beta_1,\ldots,\beta_n$ be elements of $\frac{1}{q-1}\bZ$. Assume that $\alpha_i\not\in\bZ$ for any $i$ and moreover that $\alpha_i-\beta_j\not\in\bZ$ for any $i,j$. Let $j\colon\bG_{\rmm}\hookrightarrow\bA^1$ be the inclusion. Then, the natural morphism \[ j_!j^{\ast}\big(A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)\big)\longrightarrow A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta) \] is an isomorphism. \label{prop:connectiontype2} \end{proposition} \begin{proof} We put $\sL\defeq A_1(K)^{\dag}/A_1(K)^{\dag}\Hyp_{\pi}(\balpha;\bbeta)$; it is overholonomic and has a Frobenius structure by Corollary \ref{cor:overholonomic}. By the localization triangle $j_!j^{\ast}\to\id\to i_{\plus}i^{\plus}\to$ \cite[(3.1.9.1)]{AbeMarmora}, it suffices to prove that $i^{\plus}\sL=0$. Since $i^{\plus}\sL$ is isomorphic to the complex $\left[\sL^{\an}\xrightarrow{\partial\cdot} \sL^{\an}\right]$ \cite[5.1.3]{Crew12CM}, where $\sL^{\an}$ denotes the analytification of $\sL$, it suffices to prove that $\partial$ bijectively acts on $\sL^{\an}$. Moreover, since the kernel and the cokernel of this morphism have the same dimension \cite[3.1.10 and Remark after that]{AbeMarmora}, it only remains to show that it is injective. Recall that the ring $\sD^{\an}$ of analytic differential operators \cite[4.1]{Crew12CM} is described as \[ \sD^{\an}=\Set{\sum_{k=0}^{\infty}a_k\partial^{[k]} | a_k\in A[0,1[\text{ and } \exists \eta<1, \forall r<1, \exists C_r>0, |a_k|_r\leq C_r\eta^k}. \] Here, $A[0,1[$ denotes the ring of analytic functions on the open unit disk, and $|\cdot|_r$ denotes the $r$-Gauss norm on $A[0,1[$. Since we have $\sL^{\an}=\sD^{\an}/\sD^{\an}\Hyp_{\pi}(\balpha;\bbeta)$, in order to show the injectivity of $\partial\cdot$, it suffices to prove the following property: if $P, Q\in \sD^{\an}$ satisfies $\partial P=Q\Hyp_{\pi}(\balpha;\bbeta)$, then $Q$ belongs to $\partial \sD^{\an}$. We may assume that $Q\in A[0,1[$. We put $Q=\sum c_lx^l$. Since \[ Q\Hyp_{\pi}(\balpha;\bbeta)\equiv \sum_{l=0}^{\infty}\left(\prod_{i=1}^m(-l-1-\alpha_i)x^l-(-1)^{m+np}\pi^{m-n}\prod_{j=1}^n(-l-2-\beta_j)x^{l+1}\right)c_l \] modulo $\partial\sD^{\an}$, the assumption that $\alpha_i\not\in\bZ$ shows inductively that all $c_l$'s are zero. \end{proof} \begin{corollary} Under the situation in the previous proposition, the natural morphism \[ j_{!}\sH_{\pi}(\balpha; \bbeta)\longrightarrow j_{\plus}\sH_{\pi}(\balpha;\bbeta) \] is an isomorphism. \label{cor:middleext} \end{corollary} \begin{proof} Combine Proposition \ref{prop:connectiontype} and Proposition \ref{prop:connectiontype2}. \end{proof} Now, we get the following proposition. \begin{proposition} Let $\psi$ be a non-trivial additive character on $k$, let $\chi_1,\ldots,\chi_m$ and $\rho_1,\ldots,\rho_n$ be characters on $k^{\times}$ and assume that $\chi_i\neq\rho_j$ for any $i,j$. Then, the natural morphism \[ \sHyp_{\psi,!}(\bchi; \brho)\longrightarrow\sHyp_{\psi,\plus}(\bchi; \brho) \] in $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$ is an isomorphism. \label{prop:twohypareisom} \end{proposition} \begin{proof} If $(m,n)=(0,0),(1,0),(0,1)$, the assertion is obvious by definition. Let us prove the assertion by induction on $m+n$. As in the proof of Theorem \ref{thm:hypandconv}, we may assume that $m>0$ and that $\chi_1$ is trivial. Then, (again by the similar argument as in the proof of Theorem \ref{thm:hypandconv},) the claim reduces to showing that \[ j^{\ast}\big(\FT_{\psi}(j_{!}\sHyp_{\pi_{\psi^{-1}},!}(\brho^{-1};\bchi'^{-1}))\big) \longrightarrow j^{\ast}\big(\FT_{\psi}(j_{\plus}\sHyp_{\pi_{\psi},\plus}(\brho^{-1};\bchi'^{-1}))\big) \] is an isomorphism. By induction hypothesis, the natural morphism \[ \sHyp_{\pi_{\psi^{-1}},!}(\brho^{-1};\bchi'^{-1})\longrightarrow\sHyp_{\pi_{\psi^{-1}},\plus}(\brho^{-1};\bchi'^{-1}) \] is an isomorphism, and moreover Theorem \ref{thm:hypandconv} shows that both are isomorphic to $\sH_{\pi}(\balpha;\bbeta)$ for some $\pi$, $\alpha_i$'s and $\beta_j$'s satisfying the assumption of Proposition \ref{prop:connectiontype2}. Therefore, the morphism \[ j_{!}\sHyp_{\pi_{\psi^{-1}},!}(\brho^{-1};\bchi'^{-1})\longrightarrow j_{\plus}\sHyp_{\pi_{\psi^{-1}},\plus}(\brho^{-1};\bchi'^{-1}) \] is an isomorphism because the underlying morphism in $D^{\rb}_{\ovhol}(\bA^1_k/K)$ is an isomorphism by the previous corollary. This shows the assertion. \end{proof} \section{Hypergeometric Isocrystals.} In this section, we study the isocrystals (in the classical sense) defined by hypergeometric differential operators. The first subsection compares it with the arithmetic hypergeometric $\sD$-modules in the previous section, and as a result, proves that the latter is has a structure of overconvergent $F$-isocrystals. In the second subsection, we discuss the irreducibility and a characterization of these overconvergent $F$-isocrystals. Recall that we are always assuming that $K$ has a primitive $p$-th root of unity. \subsection{Overconvergence of hypergeometric isocrystals.} Throughout this subsection, we fix a $\pi\in K^{\times}$ with $\pi^{q-1}=(-p)^{(q-1)/(p-1)}$, fix elements $\alpha_1,\dots,\alpha_m$ and $\beta_1,\dots,\beta_n$ in $\frac{1}{q-1}\bZ$ such that $\alpha_i-\beta_j\not\in\bZ$ for any $i,j$, and assume that $(m,n)\neq(0,0)$. We put $r\defeq\max\{m,n\}$. We denote by $T$ the divisor $\{0,\infty\}$ of $\bP^1_k$ if $m\neq n$ and $\{0,1,\infty\}$ if $m=n$, and we put $\sX\defeq \widehat{\bP^1_V}\setminus T$ and $X\defeq\bP^1_k\setminus T$. In this subsection, $h_l(x)\in K[x] (l\in\bN)$ denote the polynomial defined by \[ \Hyp_{\pi}(\balpha;\bbeta)=\sum_{l=0}^{\infty}h_l(x)\partial^l. \] \begin{lemma} \label{lem:coefficientofhgoperator} We have \[ h_r(x) = \begin{cases} x^m & \text{if }m>n,\\ x^m\big(1-(-1)^{n(p+1)}x\big) & \text{if }m=n,\\ -(-1)^{m+np}\pi^{m-n}x^{n+1} & \text{if }m<n.\end{cases} \] If $l>r$, then $h_l(x)=0$. \label{lem:hr} \end{lemma} \begin{proof} $(x\partial)^l$ is the sum of $x^l\partial^l$ and a differential operator of differential order $<l$. The assertion follows from this fact. \end{proof} This lemma shows that $\frac{h_l(x)}{h_r(x)}$'s are global sections of $\sO_{\sX_K}$. With this in mind, we define the hypergeometric isocrystals $H_{\pi}(\balpha;\bbeta)$ on $X$ as follows. \begin{definition} The hypergeometric isocrystal $H_{\pi}(\balpha;\bbeta)$ is the free $\sO_{\sX_K}$-module $(\sO_{\sX_K})^{\oplus r}$ equipped with the connection \[ \nabla = \begin{pmatrix} 0 & 1 & & & \\ & 0 & 1 & & \\ & & & & \\ & & & 0 & 1 \\ -\frac{h_0(x)}{h_r(x)} & -\frac{h_1(x)}{h_r(x)} & \dots & \dots & -\frac{h_{r-1}(x)}{h_r(x)} \end{pmatrix} \otimes dx. \] \end{definition} The following theorem, which relates the isocrystal $H_{\pi}(\balpha;\bbeta)$ and the arithmetic $\sD$-module $\sH_{\pi}(\balpha;\bbeta)$, is the goal of this subsection. For the convenience of the reader, we repeat the hypothesis in this subsection. \begin{theorem} Let $\pi$ be an element of $K^{\times}$ satisfying $\pi^{q-1}=(-p)^{(q-1)/(p-1)}$, and let $\alpha_1,\dots,\alpha_m$ and $\beta_1,\dots,\beta_n$ be elements of $\frac{1}{q-1}\bZ$ satisfying $\alpha_i-\beta_j$ for any $i,j$. Then, we have an isomorphism \[ \sp_{\plus}H_{\pi}(\balpha;\bbeta)\cong \sH_{\pi}(\balpha;\bbeta). \] In particular, $H_{\pi}(\balpha;\bbeta)$ has a structure of a convergent $F$-isocrystal on $X$ overconvergent along $T$. \label{thm:ocFisoc} \end{theorem} In the course of the proof, we need the theory of hypergeometric functions over finite fields (or hypergeometric sums) \cite[(8.2.7)]{Katz90ESDE} and the $\ell$-adic hypergeometric sheaves \cite[Theorem 8.4.2]{Katz90ESDE}. \begin{definition} (i) Let $m, n$ be natural numbers and let $t$ be an element of $k$. We define the hypersurface $V(m,n,t)$ of $(\bG_{\rmm,k})^{m+n}$ by the equation \[ \prod_{i=1}^m x_i \cdot \prod_{j=1}^n y_j^{-1} = t, \] where $(x_1,\dots,x_m,y_1,\dots,y_n)$ is the coordinate of $(\bG_{\rmm,k})^{m+n}$. (ii) Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\ldots,\chi_m,\rho_1,\ldots,\rho_n$ be multiplicative characters on $k^{\times}$. We denote the sequence $\chi_1,\dots,\chi_m$ \resp{$\rho_1,\dots,\rho_n$} by $\bchi$ \resp{$\brho$}. Then, the function $\Hyp_{\psi}(\bchi; \brho)$ on $k^{\times}$ is defined by \[ \Hyp_{\psi}(\bchi;\brho)(t) \defeq \sum\psi\left(\sum_{i=1}^m x_i-\sum_{j=1}^n y_j\right)\cdot\prod_{i=1}^m\chi_i(x_i)\cdot \prod_{j=1}^n\rho_j^{-1}(y_j), \] where the sum runs over $(x_1,\dots,x_m,y_1,\dots,y_n)\in V(m,n,t)$. \end{definition} \begin{proposition}[{\cite[(8.2.7), Theorem 8.4.2]{Katz90ESDE}}] Under the notation in Definition 4.1.4 (ii), assume that $\chi_i\neq\rho_j$ for each $i,j$. Then, there exists a smooth $\ell$-adic sheaf $\sH^{\ell}_{\psi,!}(\bchi;\brho)$ of rank $r=\max\{m,n\}$ on $X$, whose Frobenius trace at a closed point $x$ of degree $h$ equals \[ (-1)^{m+n+1}\Hyp_{\psi\circ\Tr_{\kappa(x)/k}}(\bchi\circ\Norm_{\kappa(x)/k};\brho\circ\Norm_{\kappa(x)/k}). \] \end{proposition} We refer the smooth $\ell$-adic sheaf $\sH^{\ell}_{\psi,!}(\bchi;\brho)$ in the previous proposition as ``the $\ell$-adic hypergeometric sheaf''. We are now ready for the proof of Theorem \ref{thm:ocFisoc}. We start by calculating the Frobenius trace of $\sHyp_{\psi,!}(\bchi;\brho)$. \begin{proposition} Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\dots,\chi_m$ and $\rho_1,\dots,\rho_n$ be characters on $k^{\times}$. Let $x$ be a $k$-rational point of $\bG_{\rmm,k}$. Then, the Frobenius trace of $\sHyp_{\psi,!}(\bchi;\brho)$ at $x$ equals $q^{2(m+n)}\Hyp_{\psi}(\bchi;\brho)(x)$. \label{prop:frobeniustrace} \end{proposition} \begin{proof} If $(m,n)=(1,0)$, Proposition \ref{prop:propertiesofDwork} (iii) and \ref{prop:propertiesofKummer} (iii) show that the Frobenius trace of $\sHyp_{\psi,!}(\chi;\emptyset)$ at $x$ equals $q^2\psi(x)\chi(x)$, which coincides with $q^2\Hyp_{\psi}(\chi;\emptyset)(x)$. The case where $(m,n)=(0,1)$ can be proved similarly. In general, for any sequences $\bchi$ \resp{$\bchi',\brho,\brho'$} of characters on $k^{\times}$, we have \[ \Hyp_{\psi}(\bchi,\bchi';\brho,\brho')(t) = \sum_{t_1t_2=t}\Hyp_{\psi}(\bchi;\brho)(t_1)\Hyp_{\psi}(\bchi',\brho')(t_2). \] The claim is proved by this formula, Proposition \ref{prop:frobeniusofconvolution} and the previous two cases. \end{proof} \begin{lemma} \label{lem:dimofstalks} Let $T'$ be a divisor of $\bP^1_k$ that includes $T$, put $\sY\defeq\widehat{\bP^1_V}\setminus T'$, $Y\defeq \bP^1_k\setminus T'$, and assume that $\sH_{\pi}(\balpha;\bbeta)|_{\sY}$ is a coherent $\sO_{\sY,\bQ}$-module. Then, $\sH_{\pi}(\balpha;\bbeta)|_{\sY}$ is a free $\sO_{\sY,\bQ}$-module of rank $r$. \end{lemma} \begin{proof} By assumption, there exists a convergent isocrystal $E$ on $Y$ satisfying $\sp_{\plus}E\cong\sH_{\pi}(\balpha;\bbeta)|_{\sY}$. Since $E$ is free as an $\sO_{\sY_K}$-module (being locally free over the principal ideal ring $\Gamma(\sY_K, \sO_{\sY_K})$), $\sH_{\pi}(\balpha;\bbeta)|_{\sY}$ is also free as an $\sO_{\sY,\bQ}$-module. Therefore, it remains to prove that its rank equals $r$. Let $x$ be a closed point of $Y$ with residue field $\kappa(x)$. Then, $i_x^{\plus}\iota_{\kappa(x)/k}\sH_{\pi}(\balpha;\bbeta)$ is concentrated at degree $-1$ \cite[(1.2.14.1)]{AbeCaro}; for proving the lemma, it suffices to show that the $(-1)$-st cohomology is of dimension $r$. By the scalar extension, we may assume that $x$ is a $k$-rational point. Let $\psi$ \resp{$\chi_i$'s and $\rho_j$'s} be a non-trivial character on $k$ \resp{characters on $k^{\times}$} corresponding to $\pi$ \resp{$\alpha_i$'s and $\beta_j$'s}, so that $\sH_{\pi}(\balpha;\bbeta)\cong \sHyp_{\psi,\plus}(\bchi;\brho)[-(m+n)]$. For each positive integer $h$, let $k_h$ be a field extension of $k$ of degree $h$. By using the previous proposition, we see just as in \cite[(8.2.7)]{Katz90ESDE} that the Frobenius trace at $x$ of $\iota_{k_h/k}\big(\sHyp_{\psi,!}(\bchi,\brho)\big)[-(m+n)]$ equals \begin{equation} (-1)^{m+n}(q^h)^{2(m+n)}\Hyp_{\psi\circ\Tr_{k_h/k}}(\bchi\circ\Norm_{k_h/k};\brho\circ\Norm_{k_h/k}). \label{eq:traceatx} \end{equation} The assumption and the purity \cite[(1.2.14.1)]{AbeCaro} shows that the trace of the Frobenius action on this $(-1)$-st cohomology is the minus of (\ref{eq:traceatx}). Since it equals the Frobenius trace of the smooth $\ell$-adic sheaf $\sH^{\ell}_{\psi,!}(\bchi;\brho)$ of rank $r$ at $x$, there exist ($q$-Weil) numbers $\gamma_1,\dots,\gamma_r$ independent of $h$ satisfying \[ (-1)^{m+n+1}(q^h)^{2(m+n)}\Hyp_{\psi\circ\Tr_{k_h/k}}(\bchi\circ\Norm_{k_h/k};\brho\circ\Norm_{k_h/k}) = \gamma_1^h+\gamma_2^h+\dots+\gamma_r^h. \] Since the left-hand side is the trace of the $h$-th iterate of the Frobenius action on the $(-1)$-st cohomology of $i_x^{\plus}\sHyp_{\psi,\plus}(\bchi;\brho)[-(m+n)]$, this action has exactly $r$ eigenvalues (Note that there are no zero eigenvalue.) This shows that its dimension equals $r$. \end{proof} \begin{lemma} \label{lem:phiisisom} Let $T'$ be a divisor of $\bP^1_k$ that includes $T$, and put $\sY\defeq\widehat{\bP^1_V}\setminus T'$. Assume that $\sH_{\pi}(\balpha;\bbeta)|_{\sY}$ is a coherent $\sO_{\sY,\bQ}$-module. Then, the composition \[ \sD_{\sY,\bQ}\longrightarrow \sD_{\sY,\bQ}^{\dag} \longrightarrow \sH_{\pi}(\balpha;\bbeta)|_{\sY} \] induces an isomorphism \[ \varphi\colon \sD_{\sY,\bQ}/\sD_{\sY,\bQ}\Hyp_{\pi}(\balpha;\bbeta)\longrightarrow \sH_{\pi}(\balpha;\bbeta)|_{\sY} \] of $\sD_{\sY,\bQ}$-modules. \end{lemma} \begin{proof} Since the leading coefficient $h_r(x)$ of $\Hyp_{\pi}(\balpha;\bbeta)$ is invertible on $\sX$ by Lemma \ref{lem:coefficientofhgoperator}, $\sD_{\sX,\bQ}/\sD_{\sX,\bQ}\Hyp_{\pi}(\balpha;\bbeta)$ is a free $\sO_{\sX,\bQ}$-module of rank $r$; The previous lemma shows that the target of $\varphi$ is also a free $\sO_{\sY,\bQ}$-module of rank $r$. Therefore, it suffices to show that $\varphi$ is surjective. Since $\varphi$ is a morphism of coherent $\sO_{\sY,\bQ}$-modules, it suffices to prove that the morphism $\Gamma(\sY,\varphi)$ of the global sections is surjective. It follows because it has a dense image and because the source is complete. \end{proof} \begin{proof}[Proof of Theorem \ref{thm:ocFisoc}] Let $T'$ be a divisor of $\bP^1_k$ containing $T$ that satisfies the assumption in Lemma \ref{lem:phiisisom}, which exists because $\sH_{\pi}(\balpha;\bbeta)|_{\sY}$ is an overholonomic $F$-$\sD^{\dag}_{\sY,\bQ}$-module \cite[Th\'eor\`eme 2.2.17]{Caro06CM}. Since $\sD_{\sY,\bQ}/\sD_{\sY,\bQ}\Hyp_{\pi}(\balpha;\bbeta)$ is the restriction of $\sp_{\plus}H_{\pi}(\balpha;\bbeta)$ to $\sY$, the isocrystal $H_{\pi}(\balpha;\bbeta)|_{\sY_K}$ is a convergent isocrystal on $Y$ \cite[Th\'eor\`eme 2.2.12]{Caro06CM}. Since the convergent property of an isocrystal can be checked by the restricting to a dense open subset \cite[Theorem 2.16]{Ogus84}, $H_{\pi}(\balpha;\bbeta)$ itself is a convergent isocrystal. This shows that the $\sD_{\sX,\bQ}$-module structure on $\sp_{\plus}H_{\pi}(\balpha;\bbeta)$ extends to a $\sD^{\dag}_{\sX,\bQ}$-module structure, and we obtain an isomorphism $\sp_{\plus}H_{\pi}(\balpha;\bbeta)\cong\sH_{\pi}(\balpha;\bbeta)|_{\sX}$ of $\sD^{\dag}_{\sX,\bQ}$-modules. In particular, $\sH_{\pi}(\balpha;\bbeta)|_{\sX}$ is $\sO_{\sX,\bQ}$-coherent, which shows the claim \cite[Th\'eor\`eme 2.2.12]{Caro06CM}. \end{proof} \begin{remark} Let $\psi$ \resp{$\chi_i$'s and $\rho_j$'s} be a non-trivial character on $k$ \resp{characters on $k^{\times}$} corresponding to $\pi$ \resp{$\alpha_i$'s and $\beta_j$'s}. Then, Proposition \ref{prop:frobeniustrace} shows that the Frobenius trace of $H_{\pi}(\balpha;\bbeta)$ equals $(-1)^{m+n+1}\Hyp_{\psi}(\bchi;\brho)$ (The difference from the Frobenius trace of $\sHyp_{\psi,!}(\bchi;\brho)$ comes from the isomorphism of purity \cite[(1.2.14.1)]{AbeCaro}). As a result, the Frobenius trace of $H_{\pi}(\balpha;\bbeta)$ coincides with that of $\ell$-adic hypergeometric sheaf, which we denoted by $\sH_{\psi,!}^{\ell}(\bchi;\brho)$ in the proof of Lemma \ref{lem:dimofstalks}. \label{rem:explicittraces} \end{remark} \subsection{Irreducibility and a characterization of hypergeometric isocrystals.} We firstly prove that the arithmetic hypergeometric $\sD$-modules are irreducible. \begin{proposition} Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\dots,\chi_m$ and $\rho_1,\dots,\rho_n$ be characters on $k^{\times}$ that satisfies $\chi_i\neq\rho_j$ for any $i,j$. Then, the following statements hold. {\rm (i)} $\sHyp_{\psi,\plus}(\bchi;\brho)$ is an irreducible object of $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$. {\rm (ii)} $j_{\plus}\sHyp_{\psi,\plus}(\bchi;\brho)$ is an irreducible object of $F\hyphen D^{\rb}_{\ovhol}(\bA^1_k/K)$. \label{prop:irredofD} \end{proposition} \begin{proof} If $(m,n)=(0,0)$, then both assertions are trivial. We prove both (i) and (ii) by induction on $m+n$. In order to prove (i), we first note that the irreducibility is preserved under taking $\inv^{\ast}$ and under the tensorisation of $\sK_{\gamma}$'s because these operations are equivalences of categories from $F\hyphen D^{\rb}_{\ovhol}(\bG_{\rmm,k}/K)$ to itself. Therefore, by the same argument as in the proof of Theorem \ref{thm:hypandconv}, we may assume that $m>0$ and that $\chi_1$ is trivial. In the proof of Theorem \ref{thm:hypandconv}, we see that \[ \sHyp_{\psi,\plus}(\bchi;\brho)\cong j^{\ast}\big(\FT_{\psi}(j_{\plus}\sHyp_{\pi_{\psi^{-1}},\plus}(\brho^{-1};\bchi'^{-1}))\big) \] if $\bchi'$ denotes the sequence $\chi_2,\dots,\chi_m$. By induction hypothesis, $j_{\plus}\sHyp_{\psi,\plus}(\brho^{-1};\bchi'^{-1})$ is irreducible in $F\hyphen D^{\rb}_{\ovhol}(\bA^1_k/K)$. Because the geometric Fourier transform $\FT_{\psi}$ is an equivalence of categories, $\FT_{\psi}\big(j_{\plus}\sHyp_{\psi,\plus}(\brho^{-1};\bchi'^{-1})\big)$ is also irreducible. Because $j^{\ast}=j^!$ and because it preserves the property of being ``irreducible or zero'' \cite[1.4.6]{AbeCaro}, the proof of (i) is done. To prove (ii), note that Corollary \ref{cor:middleext} shows that $j_{\plus}\sHyp_{\psi,\plus}(\bchi;\brho)$ is the intermediate extension \cite[1.4.1]{AbeCaro} of $\sHyp_{\psi,\plus}(\bchi;\brho)$. Since the intermediate extension preserves the irreducibility \cite[1.4.7]{AbeCaro}, (ii) is proved. \end{proof} This proposition also gives the irreducibility of hypergeometric isocrystals. \begin{proposition} Let $\pi$ be an element of $K^{\times}$ satisfying $\pi^{q-1}=(-p)^{(q-1)/(p-1)}$, and let $\alpha_1,\dots,\alpha_m$ and $\beta_1,\dots,\beta_n$ be elements of $\frac{1}{q-1}\bZ$ satisfying $\alpha_i-\beta_j\not\in\bZ$ for any $i,j$. If $(m,n)\neq (0,0)$, then the overconvergent $F$-isocrystal $H_{\pi}(\balpha;\bbeta)$ on $X$ is irreducible, where $X$ denotes $\bG_{\rmm,k}$ if $m\neq n$ and $\bG_{\rmm,k}\setminus\{1\}$ if $m=n$. \end{proposition} \begin{proof} If $m\neq n$, then it follows from Proposition \ref{prop:irredofD}. If $m=n$, then it follows from the irreducibility of $k^{!}\sHyp_{\psi,\plus}(\bchi;\brho)$, where $k\colon \bG_{\rmm,k}\setminus\{1\}\hookrightarrow\bG_{\rmm,k}$ is the inclusion, which is a consequence of Proposition \ref{prop:irredofD} and \cite[Lemma 1.4.6]{AbeCaro}. \end{proof} At last, we give a characterization of hypergeometric isocrystals in terms of Frobenius trace function. \begin{proposition} Let $\psi$ be a non-trivial additive character on $k$, and let $\chi_1,\dots,\chi_m$ and $\rho_1,\dots,\rho_n$ be characters on $k^{\times}$ that satisfies $\chi_i\neq\rho_j$ for any $i,j$; let $\pi$ \resp{$\alpha_i$'s and $\beta_j$'s} be elements of $K$ \resp{$\frac{1}{q-1}\bZ$} corresponding to $\psi$ \resp{$\chi_i$'s and $\rho_j$'s}. Let $X$ be $\bG_{\rmm,k}$ if $m\neq n$ and be $\bG_{\rmm,k}\setminus\{1\}$ if $m=n$. Let $\iota\colon \overline{K}\cong\bC$ is an isomorphism. Let $\sF$ be an $\iota$-mixed overconvergent $F$-isocrystal on $X$ whose Frobenius trace at each closed point $x$ equals $(-1)^{m+n+1}\Hyp_{\psi}(\bchi;\brho)$. Then, there exists an isomorphism $\sF\to H_{\pi}(\balpha;\bbeta)$ of overconvergent $F$-isocrystals. Moreover, we have \[\dim_K\Hom(\sF, H_{\pi}(\balpha;\bbeta))=1.\] \end{proposition} \begin{proof} First, note that $H_{\pi}(\balpha;\bbeta)$ is pointwise $\iota$-pure because the Frobenius traces coincide with those of a smooth $\ell$-adic sheaf $\sH^{\ell}_{\psi}(\bchi;\brho)$ and the latter is so \cite[Theorem 8.4.2 (4)]{Katz90ESDE}. Therefore, by the ``\v{C}ebotarev density theorem'' for overconvergent $F$-isocrystals \cite[3]{Abe11}, the semi-simplifications of $\sF$ and $H_{\pi}(\balpha;\bbeta)$ are isomorphic to each other. Because $H_{\pi}(\balpha;\bbeta)$ is irreducible by the previous proposition, $\sF$ is itself isomorphic to $H_{\pi}(\balpha;\bbeta)$. In particular, $\sF$ is irreducible, which shows the last part of the proposition. \end{proof}
8b38a56a8d3dfec99cda595719feed3583d351f9
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Many problems, particularly in combinatorics, reduce to asking whether some graph with a given property exists, or alternatively, asking how many such non-isomorphic graphs exist. Such graph search and graph enumeration problems are notoriously difficult, in no small part due to the extremely large number of symmetries in graphs. In practical problem solving, it is often advantageous to eliminate these symmetries which arise naturally due to graph isomorphism: typically, if a graph $G$ is a solution then so is any other graph $G'$ that is isomorphic to $G$. General approaches to graph search problems typically involve either: \emph{generate and test}, explicitly enumerating all (non-isomorphic) graphs and checking each for the given property, or \emph{constrain and generate}, encoding the problem for some general-purpose discrete satisfiability solver (i.e. SAT, integer programming, constraint programming), which does the enumeration implicitly. % In the explicit approach, one typically iterates, repeatedly applying an extend and reduce approach: First \emph{extend} the set of all non-isomorphic graphs with $n$ vertices, in all possible ways, to graphs with $n+1$ vertices; and then \emph{reduce} the extensions to their non-isomorphic (canonical) representatives. % In the constraint based approach, one typically first encodes the problem and then applies a constraint solver in order to produce solutions. The (unknown) graph is represented in terms of Boolean variables describing it as an adjacency matrix $A$. The encoding is a conjunction of constraints that constitute a model, $\varphi_A$, such that any satisfying assignment to $\varphi_A$ is a solution to the graph search problem. Typically, symmetry breaking constraints~\cite{Crawford96,CodishMPS13} are added to the model to reduce the number of isomorphic solutions, while maintaining the correctness of the model. It remains unknown whether a polynomial time algorithm exists to decide the graph isomorphism problem. Nevertheless, finding good graph isomorphism algorithms is critical when exploring graph search and enumeration problems. Recently an algorithm was published by \citeN{Babai15} which runs in time $O\left( {\exp \left( log^c(n) \right) } \right)$, for some constant $c>1$, and solves the graph isomorphism problem. Nevertheless, top of the line graph isomorphism tools use different methods, which are, in practice, faster. \citeN{nauty} introduces an algorithm for graph canonization, and its implementation, called \texttt{nauty}\ (which stands for \emph{no automorphisms, yes?}), is described in \cite{nauty_impl}. % In contrast to earlier works, where the canonical representation of a graph was typically defined to be the smallest graph isomorphic to it (in the lexicographic order), \texttt{nauty}\ introduced a notion which takes structural properties of the graph into account. For details on how \texttt{nauty}~defines canonicity and for the inner workings of the \texttt{nauty}\ algorithm see~\cite{nauty,nauty_impl,hartke_nauty,nautyII}. % In recent years \texttt{nauty}~has gained a great deal of popularity and success. Other, similar tools, are \textsf{bliss}~\cite{bliss} and \textsf{saucy}~\cite{saucy}. The \texttt{nauty}\ graph automorphism tool consists of two main components. (1) a C library, \texttt{nauty}, which may be linked to at runtime, that contains functions applicable to find the canonical labeling of a graph, and (2) a collection of applications, \texttt{gtools}, that implement an assortment of common tasks that \texttt{nauty}\ is typically applied to. % When downloading the tool both components are included. During compilation static library files are created for the C library. These files may be linked to at runtime, and header files are provided which may be included in foreign C code. During compilation, the applications of \texttt{gtools}\ are compiled into a set of command line applications. This paper presents a lightweight Prolog interface to both components of \texttt{nauty}\ which we term \texttt{pl-nauty}\ and \texttt{pl-gtools}. % The implementation of \texttt{pl-nauty}\ is by direct use of Prolog's foreign language interface. The implementation of \texttt{pl-gtools}\ is slightly more complex. Each \texttt{gtools}\ application is run as a child process with the input and output controlled via unix pipes. The \texttt{pl-gtools}\ framework provides a set of general predicates to support this type of application integration. The integration of \texttt{nauty}\ into Prolog facilitates programming with the strengths of the two paradigms: logic programming for solving graph search problems on the one hand, and efficient pruning of (intermediate) solutions modulo graph isomorphism, on the other. % It enables Prolog programs which address graph search problems to apply \texttt{nauty}~natively, through Prolog, in the process of graph search and enumeration. Graphs may be generated non-deterministically and may be canonized deterministically. % It also facilitates the interaction with various graph representations: those used in \texttt{nauty}, and those more natural for use with Prolog. The interface for \texttt{nauty}\ from within Prolog combines well also with other tools and techniques typically applied when addressing graph search problems, such as constraint and SAT based programming. % For example, recent work~\cite{Codish2016}, presents a computer-based proof that the Ramsey number $R(4,3,3)=30$, thus closing a long open problem concerning the value of $R(4,3,3)$. That paper made extensive use of SAT solvers, symmetry breaking techniques, and the \texttt{nauty}\ library. It was this experience that led us to implement \texttt{pl-nauty}. The remaining sections of this paper are organized in the following manner: Section~(\ref{sec:prelim}) introduces the definitions used throughout the paper, as well as the running example of Ramsey graphs. Section~(\ref{sec:pln}) introduces the core of the \texttt{pl-nauty}\ library by examples. Section~(\ref{sec:plg}) details the \texttt{pl-gtools}\ framework, and details the template used to integrate \texttt{gtools}\ applications with Prolog. Section~(\ref{sec:tech}) closes some technical loose ends, including details of supported platforms, package availability, and additional references to source code. Finally, Section~\ref{sec:conclude} concludes. \section{Preliminaries} \label{sec:prelim} A graph $G=(V,E)$ consists of a set of vertices $V=[n]=\set{1,\ldots,n}$ and a set of edges $E\subseteq V \times V$. In the examples presented in this paper graphs are always simple. Meaning that they are undirected, there are no self loops, and no multiple edges. % The tools we present allow also directed graphs and support vertex-coloring. % Two graphs $G=([n],E)$ and $G'=([n],E')$ are said to be isomorphic if the vertices of one graph may be permuted to obtain the other. Namely, if there exists a permutation $\pi\colon[n]\to[n]$ such that $(u,v) \in E \iff (\pi(u),\pi(v)) \in E'$. % % Graph isomorphism is an equivalence relation. As such, it induces equivalence classes on any set of graphs, wherein graphs $G$, $G'$ are in the same equivalence class if $G$ and $G'$ are isomorphic. % The canonical representation of a graph $G$ is some fixed value $can(G)$ such that for every graph $G'$ isomorphic to $G$ we have $can(G) = can(G')$. The running example we use throughout this paper concerns the generation of Ramsey graphs: A $R(s,t;n)$ Ramsey graph, where $s,t,n\in\mathbb{N}$, is a graph $G$ with $n$ vertices such that $G$ contains no clique of size $s$ nor an independent set of size $t$. We denote by ${\cal R}(s,t;n)$ the set of all non-isomorphic Ramsey $R(s,t;n)$ graphs. The Ramsey number $R(s,t)$ is the smallest natural number $n$ for which no $R(s,t;n)$ graph exist. \section{Interfacing Prolog with \texttt{nauty}'s C library} \label{sec:pln} The \texttt{pl-nauty}\ interface is implemented using the foreign language interface of SWI-Prolog~\cite{swipl}. The \texttt{nauty}\ C library is linked with corresponding C code written for Prolog, which involves four low-level Prolog predicates: (1)~\texttt{densenauty/8}, (2)~\texttt{canonic\_graph/6}, (3)~\texttt{isomorphic\_graphs/6}, and (4)~\texttt{graph\_convert/5}. % The experienced \texttt{nauty}\ user will find \texttt{densenauty/8} to be a direct interface to the corresponding C function in \texttt{nauty}. The \texttt{canonic\_graph/6} predicate performs graph canonization only. The \texttt{isomorphic\_graphs/6} predicate tests two graphs for isomorphism, and \texttt{graph\_convert/5} converts between the supported graph formats such as between the \texttt{graph6}~\cite{graph6} format often used in \texttt{nauty}\ and the Boolean adjacency matrices natural in logic programming. We present several examples of the \texttt{pl-nauty}\ library in Prolog. The first two examples revolve around enumerating Ramsey graphs modulo isomorphism. The rest are simple demonstrations of the core \texttt{pl-nauty}\ predicates in various cases. % In the first example we apply a straightforward iterative approach to enumerate all solutions modulo isomorphism. % The second example illustrates how \texttt{nauty}~ integrates into an existing tool-chain, all specified as part of the Prolog process. Here we first construct a constraint model, infused with a partial symmetry breaking predicate. Then, apply the finite domain constraint compiler \textsf{BEE}~\cite{BEE,jair2013} (written in Prolog) to obtain a CNF model, apply a SAT solver (through its Prolog interface), and then generate all solutions of constraint model. At the end of each iteration we apply predicates from the \texttt{pl-nauty}\ library to remove isomorphic solutions. % The core of the code, with an emphasis on using the \texttt{pl-nauty}~library is presented below. The complete code is available for download as part of the \texttt{pl-nauty}~library, in the \verb|examples| directory. \subsection{The First Example: Generate and Test} In the code below, % the predicate \verb!genRamseyGT(S, T, N, Graphs)! iterates starting from the empty graph to generate in \verb!Graphs!, the set of all canonical Ramsey $(S,T;N)$ colorings. We represent graphs as Boolean adjacency matrices: a list of \verb!N! length-\verb!N! lists. % At iteration \verb!I! it takes, \verb!Acc!, the canonical set of Ramsey $(S,T;I)$ colorings computed thus far and calls the predicate \verb!extendRamsey(S, T, I, Acc, NewAcc)! to obtain, \verb!NewAcc!, the canonical set of Ramsey $(S,T;I+1)$ colorings. % {\scriptsize \begin{verbatim} genRamseyGT(S, T, N, Graphs) :- genRamsey(0, S, T, N, [[]], Graphs). genRamsey(I, S, T, N, Acc, Graphs) :- I < N, !, I1 is I+1, extendRamsey(S, T, I, Acc, NewAcc), genRamsey(I1, S, T, N, NewAcc, Graphs). genRamsey(N, _, _, N, Graphs, Graphs). \end{verbatim} } The predicate \verb!extendRamsey(S, T, N, Graphs, NewGraphs)! takes a list, \verb!Graphs! of (canonical) Ramsey $(S,T;N)$ graphs. Then, a new vertex is added in all possible ways to each graph in \verb!Graphs! and those new graphs that are Ramsey $(S,T;N+1)$ colorings are canonized. Finally, the resulting graphs are sorted to remove duplicates, resulting in \verb!NewGraphs!. It is the call to \verb!canonic_graph/3! that interfaces to our Prolog integration of the \texttt{nauty}\ tool. % {\scriptsize \begin{verbatim} extendRamsey(S, T, N, Graphs, NewGraphs) :- N1 is N+1, findall(Canonic, (member(Graph, Graphs), addVertex(Graph, NewGraph), isRamsey(S,T,N1,NewGraph), /* #1 (test)*/ canonic_graph(N1, NewGraph, Canonic) /* #2 (reduce)*/ ), GraphsTmp), sort(GraphsTmp, NewGraphs). \end{verbatim} } The predicate \verb!addVertex(Matrix,ExtendedMatrix)! extends non-deterministically an adjacency \verb!Matrix! with a new vertex by adding a new first row and equal first column. % {\scriptsize \begin{verbatim} addVertex(Matrix,[NewRow|NewRows]) :- NewRow = [0|Xs], addFirstCol(Matrix,Xs,NewRows). addFirstCol([],[],[]). addFirstCol([Row|Rows],[X|Xs],[[X|Row]|NewRows]) :- member(X,[0,1]), addFirstCol(Rows,Xs,NewRows). \end{verbatim} } To complete the example, we illustrate the test predicate \verb!isRamsey(S,T,N,Graph)! which succeeds if the given \verb!Graph! is a Ramsey $(S,T;N)$ coloring. This is so if it is not possible to \verb!choose! \verb!S! vertices from the graph, the edges between which are all ``colored'' 0, or \verb!T! vertices from the graph, the edges between which are all ``colored'' 1. % {\scriptsize \begin{verbatim} isRamsey(S,T,N,Graph) :- forall( choose(N, S, Vs), mono(0, Vs, Graph) ), forall( choose(N, T, Vs), mono(1, Vs, Graph) ). mono(Color, Vs, Graph) :- cliqeEdges(Vs,Graph,Es), maplist(==(Color), Es). cliqeEdges([],_,[]). choose(N,K,C) :- cliqeEdges([I|Is],Graph,Es) :- numlist(1,N,Ns), cliqeEdges(I, Is, Graph, Es0), length(C,K), cliqeEdges(Is, Graph, Es). choose(C,Ns). cliqeEdges(_,[],_,[]). choose([],[]). cliqeEdges(I,[J|Js], Graph, [E|Es]) :- choose([I|Is],[I|Js]) :- nth1(I, Graph, Gi), choose(Is,Js). nth1(J, Gi, E), choose(Is,[_|Js]) :- cliqeEdges(I,Js,Graph,Es). choose(Is,Js). \end{verbatim} } We first demonstrate the application of the \verb!genRamseyGT! to the so called, ``party problem''. What is the smallest number of people that must be invited to a party so that at least three know each other, or at least three do not know each other. This is the smallest $N$ for which there is no $(3,3;N)$ coloring. The following two calls illustrate that there is a single canonical coloring when $N=5$ and none when $N=6$. So, the answer to the party problem (as well-known) is 6. {\scriptsize \begin{verbatim} ?- genRamseyGT(3,3,5,Gs). Gs = [ [[0,1,1,0,0], [1,0,0,1,0], [1,0,0,0,1], [0,1,0,0,1], [0,0,1,1,0]]]. ?- genRamseyGT(3,3,6,Gs). Gs = []. \end{verbatim} } We make three observations regarding the generation of graphs in this example. Consider the predicate \verb!extendRamsey/5!. \begin{enumerate} \item If the call \verb!canonic_graph(N1, NewGraph, Canonic)!, at the line marked \verb!/* #2 */!, is replaced by the line \verb!Canonic = NewGraph!, then all solutions are found, not just the canonical ones. For example, when $N=5$ there are 12 solutions, all of them isomorphic. {\scriptsize \begin{verbatim} ?- genRamseyGT(3,3,5,Gs), length(Gs,M). M = 12. \end{verbatim} } \item If the call to \verb!isRamsey(S,T,N1,NewGraph)!, at the line marked \verb!/* #1 */!, is removed then we generate all non-isomorphic graphs on \verb!N! vertices. For example, {\scriptsize \begin{verbatim} ?- genRamseyGT(3,3,5,Gs), length(Gs,M). M = 34. \end{verbatim} } \item If both changes are made, then we generate all graphs on \verb!N! vertices. {\scriptsize \begin{verbatim} ?- genRamseyGT(3,3,5,Gs), length(Gs,M). M = 1024. \end{verbatim} } \end{enumerate} We now demonstrate the application of the \verb!genRamseyGT! to generate incrementally all non-isomorphic $(3,5;N)$ Ramsey colorings. It is known \cite{Rad2014} that $R(3,5) = 14$. Table~(\ref{tab:35n}) summarizes the enumeration of all non-isomorphic $(3,5;N)$ colorings graphs. The first row indicates the number of (non-isomorphic) colorings generated. The next rows detail the time (in seconds) to compute these colorings and the time spent in the calls to \verb!canonic_graph!. It is notable that the time spent to reduce solutions modulo isomorphism using \texttt{nauty}\ is negligible. \begin{table}[t] \centering {\tiny \begin{tabular}{c|cccccccccccccc} $n$ & $1$ & $2$ & $3$ & $4$ & $5$ & $6$ & $7$ & $8$ & $9$ & $10$ & $11$ & $12$ & $13$ & $14$ \\ \hline $|{\cal R}(3,5;n)|$ & 1 & 2 & 3 & 7 & 13 & 32 & 71 & 179 & 290 & 313 & 105 & 12 & 1 & 0 \\ time (sec) & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.03 & 0.20 & 0.90 & 4.66 & 16.61 & 39.24 & 52.72 & 55.75 & 56.20 \\ nauty (sec) & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.01 & 0.02 & 0.06 & 0.09 & 0.11 & 0.11 & 0.11 & 0.11 \end{tabular} } \caption{Enumerating $R(3,5;n)$ graphs: Generate, Test \& Reduce.} \label{tab:35n} \end{table} To summarize this section, we stress that this is a toy application with the intention to illustrate an application of the integration of Prolog with the \texttt{nauty}\ package. % A more elaborate solution of this problem would, for example, combine the calls {\scriptsize \begin{verbatim} addVertex(Graph, NewGraph), isRamsey(S,T,N1,NewGraph) \end{verbatim} } in \verb!extendRamsey! to add edges connecting the new vertex to the rest of the graph incrementally so as not to violate the \verb!isRamsey! condition. This combination could also perform various propagation based optimizations. \subsection{The Second Example: Constrain and Generate} In the code below, % the predicate \verb!genRamseyCG(S, T, N, Graphs)! encodes an instance \verb!ramsey(S,T,N)! to a finite domain constraint model. We adopt \textsf{BEE}\ ~\cite{BEE,jair2013} for this purpose. The call to \verb!encode/3! generates a constraint model, \verb!Constraints! and the $\mathtt{N\times N}$ \verb!Matrix! of Boolean (Prolog) variables. The \verb!Matrix! structure serves as a mapping between the instance variables, which talks about the search for Ramsey colorings, and the \verb!Constraints! variables. It specifies the connection between variables in the constraint model and edges in the unknown graph we are searching for. The call to \verb!bCompile/2! compiles the constraints to a corresponding \verb!CNF!. The call to \verb!solveAll/3! iterates with the underlying SAT solver to provide all satisfying \verb!Assignments! of the \verb!CNF! (modulo the variables of interest in the list \verb!Booleans!). Satisfying assignments are then decoded back to the world of graphs in the call to \verb!decode/3!, and finally it is here that we call on the predicate \verb!canonic_graph/3! from the \texttt{pl-nauty}\ interface to restrict solutions to their canonical forms and remove isomorphic solutions by sorting these. {\scriptsize \begin{verbatim} genRamseyCG(S, T, N, Graphs) :- encode(ramsey(S,T,N), Matrix, Constraints), bCompile(Constraints,CNF), projectVariables(Matrix, Booleans), solveAll(CNF,Booleans,Assignments), decode(Assignments,Matrix,Graphs0), maplist(canonic_graph(N), Graphs0, Graphs1), sort(Graphs1, Graphs). \end{verbatim} } The predicate \verb!encode/3! is presented below. It first creates an $\mathtt{N\times N}$ adjacency \verb!Matrix! with Boolean variables representing the object of the search for a Ramsey(S,T;N) graph. It then imposes three sets of constraints: (1) the call to \verb!lex_star/2! constrains the rows of \verb!Matrix! to be pairwise lexicographically ordered. This implements the symmetry break described in~\cite{CodishMPS13}; (2) the first call to \verb!no_clique/4! constrains the graph represented by \verb!Matrix! to contain no independent set of size \texttt{S}, and (3) the second call to \verb!no_clique/4! constrains the graph represented by \verb!Matrix! to contain no clique of size \texttt{T}. The full details of the example are available for download as part of the \texttt{pl-nauty}~library, in the \verb|examples| directory. {\scriptsize \begin{verbatim} encode(ramsey(S,T,N), map(Matrix), Constraints) :- adj_matrix_create(N, Matrix), lex_star(Matrix, Cs1-Cs2), /* #1 */ no_clique(0, S, Matrix, Cs2-Cs3), /* #2 */ no_clique(1, T, Matrix, Cs3-Cs4), /* #3 */ Cs4 = [], Constraints = Cs1. \end{verbatim} } The following illustrates the \textsf{BEE}\ constraint model, with the associated adjacency matrix, produced by a call to the \texttt{encode/3} predicate for a Ramsey $R(3,3;5)$ instance. Note that the elements on the diagonal of the matrix are $-1$ which is how \textit{false} is represented in \textsf{BEE}. The constraint model consists of three types of constraints corresponding to the three annotated calls in \verb!encode/3!. {\scriptsize \begin{verbatim} [[-1,A,B,C,D], [A,-1,E,F,G], [B,E,-1,H,I], [C,F,H,-1,J], [D,G,I,J,-1]] bool_arrays_lex([B,C,D],[E,F,G]), bool_array_or([A,B,E]), bool_array_or([-A,-B,-E]), bool_arrays_lex([A,B,D],[F,H,J]), bool_array_or([A,C,F]), bool_array_or([-A,-C,-F]), bool_arrays_lex([A,F,G],[B,H,I]), bool_array_or([A,D,G]), bool_array_or([-A,-D,-G]), bool_arrays_lex([A,E,F],[D,I,J]), bool_array_or([B,C,H]), bool_array_or([-B,-C,-H]), bool_arrays_lex([B,E,I],[C,F,J]), bool_array_or([B,D,I]), bool_array_or([-B,-D,-I]), bool_arrays_lex([C,F,H],[D,G,I]), bool_array_or([C,D,J]), bool_array_or([-C,-D,-J]), bool_array_or([E,F,H]), bool_array_or([-E,-F,-H]), bool_array_or([E,G,I]), bool_array_or([-E,-G,-I]), bool_array_or([F,G,J]), bool_array_or([-F,-G,-J]), bool_array_or([H,I,J]), bool_array_or([-H,-I,-J]) \end{verbatim} } Table~(\ref{tab:35n_sat}) summarizes the enumeration of all non-isomorphic $(3,5;N)$ colorings graphs using the constrain and generate approach. The first row indicates the number of (non-isomorphic) colorings generated. The second row indicates the number of colorings found when solving the constraint model (with the partial symmetry break). The next rows detail the time (in seconds) to compute these colorings and the time spent in the calls to \verb!canonic_graph!. It is notable that the time spent to reduce solutions modulo isomorphism using \texttt{nauty}\ is negligible. \begin{table}[h] \centering {\tiny \begin{tabular}{ c|cccccccccccccc} $n$ & $1$ & $2$ & $3$ & $4$ & $5$ & $6$ & $7$ & $8$ & $9$ & $10$ & $11$ & $12$ & $13$ & $14$ \\ \hline $|{\cal R}(3,5;n)|$ & 1 & 2 & 3 & 7 & 13 & 32 & 71 & 179 & 290 & 313 & 105 & 12 & 1 & 0 \\ $\#SAT$ & 1 & 2 & 3 & 7 & 18 & 63 & 255 & 1100 & 3912 & 7319 & 3806 & 272 & 2 & 0 \\ time (sec) & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.02 & 0.02 & 0.12 & 0.74 & 1.97 & 1.16 & 1.15 & 0.07 \\%$< 3$ sec \\ nauty (sec) & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.00 & 0.01 & 0.05 & 0.16 & 0.05 & 0.00 & 0.00 & 0.00 \end{tabular} } \caption{Enumerating $R(3,5;n)$ graphs: Constrain, Generate \& Reduce.} \label{tab:35n_sat} \end{table} \subsection{The \texttt{graph\char`_convert/5} predicate} The \texttt{graph\char`_convert/5} predicate performs conversions between the different graph formats that are supported by \texttt{pl-nauty}. Supported formats include: adjacency matrices, adjacency lists, edge lists, and the \texttt{graph6} format. % As an example, to convert a graph, or a list of graphs, from the \texttt{graph6} format, to Prolog's adjacency matrix format: {\scriptsize \begin{verbatim} ?- Graph = `DqK', graph_convert(5, graph6_atom, adj_matrix, Graph, AdjMatrix). Graph = `DqK', AdjMatrix = [[0,1,1,0,0], [1,0,0,1,0], [1,0,0,0,1], [0,1,0,0,1], [0,0,1,1,0]] \end{verbatim} } % {\scriptsize \begin{verbatim} ?- Graphs = [`DRo',`Dbg',`DdW',`DLo',`D[S',`DpS',`DYc',`DqK',`DMg',`DkK',`Dhc',`DUW'], maplist(graph_convert(5, graph6_atom, adj_matrix), Graphs, AdjMatrices). Graphs = [`DRo',`Dbg',`DdW',`DLo',`D[S',`DpS',`DYc',`DqK',`DMg',`DkK',`Dhc',`DUW'], AdjMatrices = [[[0,0,1,0,1],[0,0,0,1,1],[1,0,0,1,0],[0,1,1,0,0],[1,1,0,0,0]], [[0,1,0,0,1],[1,0,0,1,0],[0,0,0,1,1],[0,1,1,0,0],[1,0,1,0,0]], [[0,1,0,1,0],[1,0,0,0,1],[0,0,0,1,1],[1,0,1,0,0],[0,1,1,0,0]], [[0,0,0,1,1],[0,0,1,0,1],[0,1,0,1,0],[1,0,1,0,0],[1,1,0,0,0]], [[0,0,1,1,0],[0,0,1,0,1],[1,1,0,0,0],[1,0,0,0,1],[0,1,0,1,0]], [[0,1,1,0,0],[1,0,0,0,1],[1,0,0,1,0],[0,0,1,0,1],[0,1,0,1,0]], [[0,0,1,0,1],[0,0,1,1,0],[1,1,0,0,0],[0,1,0,0,1],[1,0,0,1,0]], [[0,1,1,0,0],[1,0,0,1,0],[1,0,0,0,1],[0,1,0,0,1],[0,0,1,1,0]], [[0,0,0,1,1],[0,0,1,1,0],[0,1,0,0,1],[1,1,0,0,0],[1,0,1,0,0]], [[0,1,0,1,0],[1,0,1,0,0],[0,1,0,0,1],[1,0,0,0,1],[0,0,1,1,0]], [[0,1,0,0,1],[1,0,1,0,0],[0,1,0,1,0],[0,0,1,0,1],[1,0,0,1,0]], [[0,0,1,1,0],[0,0,0,1,1],[1,0,0,0,1],[1,1,0,0,0],[0,1,1,0,0]]] \end{verbatim} } \subsection{The \texttt{canonic\_graph/6} predicate} The \texttt{canonic\_graph/6} predicate performs graph canonization and it takes the form \texttt{canonic\_graph(N, InputFmt, OutputFmt, Graph, Perm, Canonic)} where \texttt{InputFmt} is the format of the \texttt{N} vertex input graph (\texttt{Graph}), \texttt{OutputFmt} is the format of the canonical graph (\texttt{Canonic}), and \texttt{Perm} is the permutation whose application to the input graph renders the canonical representative. For example: {\scriptsize \begin{verbatim} ?- N = 5, Graph = [[0,1,0,0,0], [1,0,1,0,1], [0,1,0,1,0], [0,0,1,0,1], [0,1,0,1,0]], canonic_graph(N,adj_matrix,adj_matrix,Graph,Perm,Canonic). N = 5, Graph = [[0,1,0,0,0],[1,0,1,0,1],[0,1,0,1,0],[0,0,1,0,1],[0,1,0,1,0]], Canonic = [[0,0,0,0,1],[0,0,0,1,1],[0,0,0,1,1],[0,1,1,0,0],[1,1,1,0,0]], Perm = [1, 5, 2, 4, 3] \end{verbatim} } A compact version of \texttt{canonic\_graph/6} is also included in \texttt{pl-nauty}\ in the form of the predicate \texttt{canonic\char`_graph/3}. The predicate \texttt{canonic\char`_graph/3} takes the form \texttt{canonic\char`_graph(NVert, Graph, Canonic)} and it is equivalent to \texttt{canonic\char`_graph(NVert, adj\char`_matrix, adj\char`_matrix, Graph, \char`_, Canonic)}. For example: {\scriptsize \begin{verbatim} ?- N = 5, Graph = [[0,1,0,0,0], [1,0,1,0,1], [0,1,0,1,0], [0,0,1,0,1], [0,1,0,1,0]], canonic_graph(N,Graph,Canonic). N = 5, Graph = [[0,1,0,0,0],[1,0,1,0,1],[0,1,0,1,0],[0,0,1,0,1],[0,1,0,1,0]], Canonic = [[0,0,0,0,1],[0,0,0,1,1],[0,0,0,1,1],[0,1,1,0,0],[1,1,1,0,0]] \end{verbatim} } \subsection{The \texttt{isomorphic\char`_graphs/6} predicate} The \texttt{isomorphic\char`_graphs/6} predicate tests for graph isomorphism. It takes the form: \texttt{isomorphic\char`_graphs(N, Graph1, Graph2, Perm, Canonic, Opts)} and tests if the two \texttt{N} vertex input graphs, \texttt{Graph1} and \texttt{Graph2} are isomorphic via a permutation \texttt{Perm}. If they are then \texttt{Canonic} is the canonical form they share. The predicate takes a list \texttt{Opts} of options to customize the behavior of this predicate. Options include any of the following: \texttt{fmt1(Fmt1)} the format of \texttt{Graph1}, \texttt{fmt2(Fmt2)} the format of \texttt{Graph2}, \texttt{cgfmt(CgFmt)} the format of \texttt{Canonic}. % In the case where \texttt{Graph1} and \texttt{Graph2} are not isomorphic the predicate will fail silently. % For example: {\scriptsize \begin{verbatim} ?- N = 5, Graph1 = [[0,1,0,1,1], [1,0,1,0,0], [0,1,0,1,0], [1,0,1,0,1], [1,0,0,1,0]], Graph2 = [[0,1,0,1,1], [1,0,1,0,0], [0,1,0,0,1], [1,0,0,0,1], [1,0,1,1,0]], isomorphic_graphs(N, Graph1, Graph2, Perm, Canonic, []). N = 5, Graph1 = [[0,1,0,1,1],[1,0,1,0,0],[0,1,0,1,0],[1,0,1,0,1],[1,0,0,1,0]], Graph2 = [[0,1,0,1,1],[1,0,1,0,0],[0,1,0,0,1],[1,0,0,0,1],[1,0,1,1,0]], Perm = [1,2,3,5,4], Canonic = [[0,1,0,1,0],[1,0,0,0,1],[0,0,0,1,1],[1,0,1,0,1],[0,1,1,1,0]] ?- N = 5, Graph1 = [[0,1,1,0,1],[1,0,0,0,1],[1,0,0,0,0],[0,0,0,0,0],[1,1,0,0,0]], Graph2 = [[0,1,0,0,1],[1,0,1,1,0],[0,1,0,0,1],[0,1,0,0,1],[1,0,1,1,0]], isomorphic_graphs(N, Graph1, Graph2, Perm, Canonic, []). false. \end{verbatim} } \subsection{The \texttt{densenauty/8} predicate} Most of the core predicates of \texttt{pl-nauty}\, and many of the examples described above are based on the \texttt{densenauty/8} predicate. The \texttt{densenauty/8} predicate is a direct interface to the \texttt{nauty}\ C library function of the same name. The predicate is called in a similar fashion to its counterpart in the \texttt{nauty}\ C library. A complete documentation of \texttt{densenauty/8} may be found in the source code provided with \texttt{pl-nauty}, and in the \texttt{nauty}\ user guide \cite{nauty_guide}. Briefly, the predicate \texttt{densenauty/8} takes the following form: \begin{verbatim} densenauty(NVert, Graph, Labeling, Partition, Permutation, Orbits, Canonic, Opts) \end{verbatim} where \texttt{NVert} is the number of vertices in the input graph, \texttt{Graph} is the input graph, \texttt{Labeling}, \texttt{Partition} and \texttt{Orbits} are the labeling, partition and orbits of the input graph, as described in the \texttt{nauty}\ user guide \cite{nauty_guide}, \texttt{Canonic} is the canonical form of the input graph, and \texttt{Permutation} is the permutation of the nodes of the input graph which may be applied to obtain the Canonic representative. The last argument, \texttt{Opts} is used to modify the behavior of \texttt{densenauty}. For example, it may be used to control the format of the input graph, and Canonic representative. \section{Interfacing Prolog and \texttt{gtools}} \label{sec:plg} The \texttt{nauty}\ graph automorphism tool comes with a collection of applications called \texttt{gtools}, that implement an assortment of common tasks that \texttt{nauty}\ is typically applied to. During installation (of \texttt{nauty}) these are compiled into a set of command line applications. % These applications cannot simply be loaded using the foreign language interface. Each application is like a black box. We do not wish to access its source code. % One straightforward approach to integrate \texttt{gtools}\ with Prolog is to run each such application from within Prolog, write its output to a temporary file, and then to read the file, and continue with the task that the Prolog program is addressing. A more elegant solution makes use of unix pipes to skip that intermediate step of writing and reading from files. The output is directly read/written via Prolog. The voodoo is using pipes (which are like in-memory files). We have implemented a Prolog library called \texttt{pl-gtools}, which provides a framework for calling the applications in \texttt{gtools}\ using unix pipes. % The \texttt{pl-gtools}\ framework supports two types of \texttt{gtools}\ applications which take any number of command line arguments and write their output to standard output. The first type does not require any input, and the second requires some form of input (from standard input). % We present a general template to support the two ``sides'' of the pipe: a child predicate (which typically executes a \texttt{gtools}\ command), and a parent predicate (which typically reads the output of the child). The framework includes predicates: % \verb!gtools_exec/6! and \verb!gtools_fetch/2!, and two additional predicates for applications which respectively require uni- and bi-directional communication: \verb!gtools_fork_exec/2! and \verb!gtools_fork_exec_bidi/2!. % For uni-directional communication, a call to \verb!gtools_fork_exec(Parent, Child)! will fork and execute the \texttt{Parent} goal as the parent process and the \texttt{Child} goal as the child process. It assumes that both \texttt{Parent} and \texttt{Child} take an additional argument which is unified with the corresponding input/output streams (to support communication from child to parent). % For bi-directional communication, a call to \verb!gtools_fork_exec_bidi(Parent, Child)! is exactly the same, except that the \texttt{Parent} and \texttt{Child} take two additional arguments to support two way communication. The predicate \verb!gtools_fetch/2! reads the next line from the output stream of the child and converts it to an atom. When the end of the stream is reached then the predicate fails. % A call to \verb!gtools_exec/6! takes the form {\small \texttt{gtools\_exec(NautyDir, Cmd, Args, InputStream, OutputStream, ErrorStream)} } where: \verb!NautyDir! is the directory in the file system which contains the \texttt{gtools}\ applications, \verb!Cmd! is the name of the \texttt{gtools}\ command that we which to execute, and \verb!Args! is its argument list. The final three arguments specify the standard input, output and error streams. % The call to \verb!gtools_exec/6! invokes the \texttt{exec/1} predicate of SWI-Prolog, replacing the current process image with \texttt{Cmd} and its \texttt{Args}. We present two example uses of \texttt{pl-gtools}. The first, calls \texttt{geng} from \texttt{gtools}, which iterates over all non-isomorphic graphs with a given number of vertices. The second, calls \texttt{shortg} from \texttt{gtools}, which reduces a set of graphs to non-isomorphic members. \subsection{Example 1: \texttt{geng}} \label{sec:geng} This example illustrates how the framework is applied for an application which reads no input. % The \texttt{gtools}\ application \texttt{geng} receives an argument \texttt{n} and outputs one line for each non-isomorphic graph with $n$ vertices. % Its Prolog implementation consists of three predicates: \texttt{geng/2}, \texttt{parent\_geng/2} and \texttt{child\_geng/2}. The predicate \texttt{geng/2} is the main predicate which backtracks over all results of the \texttt{gtools}\ application. The predicates \texttt{parent\_geng/2} and \texttt{child\_geng/2} implement respectively the parent and child sides of the pipe. {\scriptsize \begin{verbatim} geng(N, Graph) :- gtools_fork_exec(geng:parent_geng(Graph), geng:child_geng(N)). parent_geng(Graph,Read) :- gtools_fetch(Read, Graph). child_geng(N,Stream) :- gtools_exec(`nauty26r3', geng, [`-q', N], _, Stream, _). \end{verbatim} } \subsection{Example 2: \texttt{shortg}} This example illustrates how the framework is applied for an application which reads from standard input. % The \texttt{shortg} application reads a list of graphs in the \texttt{graph6} format~\cite{graph6} from standard input, and removes all isomorphic duplicates, writing to standard output. It can be applied as follows: After integrating \texttt{shortg} with \texttt{pl-gtools}\ it could be called from Prolog like so: {\scriptsize \begin{verbatim} ?- InputGraphs = [`DRo',`Dbg',`DdW', `DLo',`D[S',`DpS', `DYc',`DqK',`DMg', `DkK',`Dhc',`DUW'], shortg(InputGraphs, OutputGraphs). InputGraphs = [`DRo',`Dbg',`DdW',`DLo',`D[S',`DpS', `DYc',`DqK' | ... ], OutputGraphs = [`DqK']. \end{verbatim} } The implementation of \texttt{shortg} in Prolog consists of three predicates and is very similar to that for \texttt{geng} except that communication between the child and parent processes is bi-directional. {\scriptsize \begin{verbatim} shortg(In, Out) :- gtools_fork_exec_bidi(shortg:parent_shortg(In, Out), shortg:child_shortg). parent_shortg(In, Out, PRead, PWrite) :- maplist(writeln(PWrite), In), flush_output(PWrite), close(PWrite), findall(O, gtools_fetch(PRead, O), Out), close(PRead). child_shortg(CRead, CWrite) :- gtools_exec(`nauty26r3', shortg, [`-q'], CRead, CWrite, _). \end{verbatim} } In this example, \texttt{shortg/2} takes two arguments: \texttt{In} a list of input graphs in the \texttt{graph6} format, to be reduced modulo isomorphism, and \texttt{Out} will be unified with the set of reduced graphs. The predicate calls the \texttt{gtools\_fork\_exec\_bidi/2} predicate. Pipes are opened to setup two way communication between the parent and child. Two additional predicates are implemented: one for the parent process and one for the child process. Each predicate takes, as its last two arguments the read and write ends of the pipes, so communication may be established. In our case, the parent writes the set of input graphs to the write end of the pipe, and then reads the results from the read end of the child's pipe. The child calls \texttt{gtools\_exec/6}, and executes \texttt{shortg/2}. \section{Technical Details} \label{sec:tech} A short overview of some technical details regarding \texttt{pl-nauty}\ and \texttt{pl-gtools}\ follows. The package containing \texttt{pl-nauty}\ and \texttt{pl-gtools}\ is available for download from the \texttt{pl-nauty}\ homepage at: \url{http://www.cs.bgu.ac.il/~frankm/plnauty}. The package contains a \texttt{README} file, which contains usage and installation instructions, as well as an \texttt{examples} directory containing the examples discussed in this paper. The C code for \texttt{pl-nauty}\ may be found in the \texttt{src} directory. Also in the \texttt{src} directory are the two module files for \texttt{pl-nauty}\ and \texttt{pl-gtools}. Both \texttt{pl-nauty}\ and \texttt{pl-gtools}\ were compiled and tested on Debian Linux and Ubuntu Linux using the 7.x.x branch of SWI-Prolog. It is important to mention that both \texttt{pl-nauty}\ and \texttt{pl-gtools}\ contain Linux specific features, and are oriented towards SWI-Prolog. It should also be noted that \texttt{pl-nauty}\ is not thread-safe, for reasons of performance. If you require a thread-safe version of \texttt{pl-nauty}\ you should synchronize calls to the predicates of the \texttt{pl-nauty}\ module. \section{Conclusion} \label{sec:conclude} We have presented, and made available, a Prolog interface to the core components of the \texttt{nauty}~graph-automorphism tool~\cite{nauty} which is often cited as ``The world's fastest isomorphism testing program'' (see for example {\small\url{http://www3.cs.stonybrook.edu/~algorith/implement/nauty/implement.shtml}}). The contribution of the paper is in the utility of the tool which we expect to be widely used. The tool facilitates programming with the strengths of two paradigms: logic programming for solving graph search problems on the one hand, and efficient pruning of (intermediate) solutions modulo graph isomorphism, on the other. % It enables Prolog programs which address graph search problems to apply \texttt{nauty}~natively, through Prolog, in the process of graph search and enumeration. Graphs may be generated non-deterministically and may be canonized deterministically. %
461c005ff63aee7aa7a74b332faadbc07ca57726
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{\label{intro}Introduction} The ``instanton calculus'' is a common approach for studying the non-perturbative semiclassical effects in gauge theories and sigma models. One of the first and perhaps the best known illustration of this approach is the $O(3)$ Non-Linear Sigma Model (NLSM) in two dimensions, where multi-instanton configurations admit a simple analytic form \cite{Polyakov:1975yp}. It is less known that the $O(3)$ NLSM provides an opportunity to explore a mechanism of exact summation of the instanton configurations in the path integral. In order to explain the purpose of this paper, we start with a brief overview of the main ideas behind this summation. The instanton contributions in the $O(3)$ NLSM were calculated in a semiclassical approximation in the paper \cite{Fateev:1979dc}. It was shown that the effect of instantons with positive topological charge can be described in terms of the non-interacting theory of Dirac fermions. Moreover, every instanton has its anti-instanton counterpart with the same action and opposite topological charge. Thus, neglecting the instanton-anti-instanton interaction, one arrives to the theory with two non-interacting fermions. Although the classical equation has no solutions containing both instanton-anti-instanton configurations, such configurations must still be taken into account. In ref.\,\cite{Bukhvostov:1980sn} Bukhvostov and Lipatov (BL) have found that the weak instanton-anti-instanton interaction is described by means of a theory of two Dirac fermions, $\psi_\sigma \ (\sigma=\pm)$, with the Lagrangian \begin{eqnarray}\label{Lagr1} {\cal L}= \sum_{\sigma=\pm }{\bar \psi}_\sigma \big({\rm i} \gamma^\mu\partial_\mu-M\big){ \psi}_\sigma- g\, \big({\bar \psi}_+\gamma^\mu{ \psi}_+\big) \big({\bar \psi}_- \gamma_\mu{ \psi}_-\big)\ . \end{eqnarray} The perturbative treatment of \eqref{Lagr1} leads to ultraviolet (UV) divergences and requires renormalization. The renormalization can be performed by adding the following counterterms to the Lagrangian which preserve the invariance w.r.t. two independent $U(1)$ rotations $\psi_\pm\mapsto \mbox{e}^{{\rm i}\alpha_\pm} \, \psi_\pm$, as well as the permutation $\psi_+\leftrightarrow\psi_-$: \begin{eqnarray}\label{Lagr2} {\cal L}_{\rm BL}={\cal L}-\sum_{\sigma=\pm}\Big(\,\delta M\, {\bar \psi}_\sigma{ \psi}_\sigma+ \frac{g_1}{2}\, \big({\bar \psi}_\sigma\gamma^\mu{ \psi}_\sigma\big)^2\Big)\ . \end{eqnarray} In fact the cancellation of the UV divergences leaves undetermined one of the counterterm couplings. It is possible to use the renormalization scheme where the renormalized mass $M$, the bare mass $M_0=M+\delta M$ and UV cut-off energy scale $\Lambda_{\rm UV}$ obey the relation \begin{eqnarray}\label{aoisasosa} \frac{M}{M_0}=\bigg(\frac{M}{\Lambda_{\rm UV}}\bigg)^\nu\ , \end{eqnarray} where the exponent $\nu$ is a renormalization group invariant parameter as well as dimensionless coupling $g$. For $\nu=0$ the fermion mass does not require renormalization and the only divergent quantity is the zero point energy. The theory, in a sense, turns out to be UV finite in this case. Then the specific {\it logarithmic} divergence of the zero point energy can be interpreted as a ``small-instanton'' divergence in the context of $O(3)$ NLSM. Recall, that the standard lattice description of the $O(3)$ sigma model has problems -- for example, the lattice topological susceptibility does not obey naive scaling laws. L\"uscher has shown \cite{Luscher:1981tq} that this is because of the so-called ``small instantons'' -- field configurations such as the winding of the $O(3)$-field around plaquettes of lattice size, giving rise to spurious contribution to quantities related to the zero point energy. To the best of our knowledge, there is no any indication that the fermionic QFT is integrable for general values of the parameters $(g,\nu)$ \cite{Ameduri:1998ah}. However, it is expected to be an integrable theory for $\nu=0$, which is of prime interest for the problem of instanton summation. The corresponding factorizable scattering theory was proposed in \cite{Fateev:1996ea}, by extending previous results of \cite{ZZ79,Polyakov:1983tt,Faddeev:1985qu}. According to the work \cite{Fateev:1996ea} the spectrum of the model contains a fundamental quadruplet of mass $M$ whose two-particle $S$-matrix is given by the direct product $(-S_{a_1}\otimes S_{a_2})$ of two $U(1)$-symmetric solutions of the $S$-matrix bootstrap. Each of the factors $S_a$ coincides with the soliton $S$-matrix in the quantum sine-Gordon theory with the renormalized coupling constant $a$. The couplings are not independent but satisfy the condition $a_1+a_2=2$, so that, without loss of generality, one can set $a_1=1-\delta$ and $a_2=1+\delta$ with $\delta\geq 0$. A relation between $\delta$ and the four-fermion coupling $g$ is not universal, i.e., depends on regularization procedure involved in the perturbative calculations. Nevertheless, $g=\frac{\pi \delta }{1-\delta^2}$ if one uses the regularization that preserves the underlying $U(1)\otimes U(1)$ symmetry of the BL model. Together with the fundamental particles of mass $M$, there are also bound states whose masses are given by $M_n=2M\, \sin\big(\textstyle{\frac{\pi n }{2}}\,(1-\delta) \big)$, where the integer $n$ run from $1$ to an integer part of $\frac{1}{1-\delta}$. As $\delta\to 1^-$, the fermion coupling $g$ approaches infinity $g\to\infty$, and an increasing number of particles with vanishing mass occur in the theory. The theory can also be continued into the strong coupling regime with $\delta>1$ by means of the bosonization technique. Namely, the fermionic BL model can be equivalently formulated as a theory of two Bose scalars $\varphi_i$ governed by the Lagrangian \cite{Bukhvostov:1980sn} \begin{eqnarray}\label{jssksjsak} {\tilde {\cal L}}_{\rm BL}={\textstyle \frac{1}{16\pi}}\ \big(\, (\partial_\nu\varphi_1)^2+(\partial_\nu\varphi_2)^2\,\big) +4\mu\ \cos\big({\textstyle \frac{\sqrt{a_1}}{2}}\varphi_1\big) \cos\big({\textstyle \frac{\sqrt{a_2}}{2}}\varphi_2\big)\ . \end{eqnarray} The interacting term here can be written as $4\mu\, \cosh\big({\textstyle \frac{\sqrt{\delta-1}}{2}}\varphi_1\big) \cos\big({\textstyle \frac{\sqrt{\delta+1}}{2}}\varphi_2\big)$, and, hence, the bosonic description is still applicable as $\delta>1$. As it was pointed out by Al.B. Zamolodchikov (unpublished, see also \cite{Fateev:1996ea}), the Lagrangian \eqref{jssksjsak} with $a_1=2-a_2<0$ provides a dual description of the so-called sausage model \cite{Fateev:1992tk}, which is a NLSM whose target space has a geometry of a deformed 2-sphere. As $a_1\to-\infty$ the sausage metric gains the $O(3)$-invariance and we come back to the $O(3)$ NLSM. Notice that the formal substitution $\delta\equiv 1-a_1=\infty$ into the relation $g=\frac{\pi\delta}{1-\delta^2}$ leads to the vanishing fermionic coupling in the initial Lagrangian \eqref{Lagr1}. Putting the theory on a finite segment $x^{1}\in [0,R]$, one should impose boundary conditions on the fundamental fermion fields. We shall consider the twisted (quasiperiodic) boundary conditions, which preserve the $U(1)\otimes U(1)$ invariance of the bulk Lagrangian, \begin{eqnarray}\label{apssspps} {\psi}_\pm(x^0,x^1+R)=-\mbox{e}^{2\pi{\rm i} k_\pm}\, {\psi}_\pm(x^0,x^1)\ ,\ \ \ \ {\bar \psi}_\pm(x^0,x^1+R)=-\mbox{e}^{-2\pi{\rm i} k_\pm}\, {\bar \psi}_\pm(x^0,x^1)\ . \end{eqnarray} The pair of real numbers $(k_+,k_-)$ labels different sectors of the theory and, therefore, one can address the problem of computing of vacuum energy $E_{\bf k}$ in each sector. Notice that twisted boundary conditions is of special interest for application of resurgence theory to the problem of instanton summation \cite{Dunne:2012ae}. There is no doubt to say that the above scenario of the instanton summation deserves a detailed quantitative study. Perhaps the simplest question in this respect concerns an exact description of finite volume energy spectrum for the theory \eqref{jssksjsak} in both regimes $0<\delta< 1$ and $\delta>1$. In this work we will focus on the perturbative regime $0<\delta< 1$, where the fermionic description \eqref{Lagr2} can be applied. We propose an exact formula which expresses the vacuum energies in terms of certain solutions of the classical sinh-Gordon equation. The formula is perfectly matching both the conformal perturbation theory as well as the standard renormalized perturbation theory for the Lagrangian \eqref{Lagr2}. The result also agrees with the original coordinate Bethe ansatz solution of ref.\cite{Bukhvostov:1980sn} and the associated non-linear integral equations derived in \cite{Saleur:1998wa}. The aim of this paper is to review and further develop all these approaches to facilitate future considerations of the NLSM regime of the theory with $\delta>1$. The paper is organized as follows. In the first two sections we discuss the perturbative approaches for calculating $E_{\bf k}$. In Sec.\,\ref{sec2}, the small-$R$ behavior of $E_{\bf k}$ is studied by means of the conformal perturbation theory for the bosonic Lagrangian \eqref{jssksjsak}. Then, in Sec.\,\ref{sec3}, using the fermionic Lagrangian \eqref{Lagr2}, the vacuum energies are calculated within the second order of standard renormalized perturbation theory. The exact formula for the vacuum energies expressed through solutions of the classical sinh-Gordon equation is presented in Sec.~\ref{secnew5}. Our considerations there are essentially based on the previous works \cite{Lukyanov:2013wra, Bazhanov:2013cua,Bazhanov:2014joa}. These connections allows one to derive a system non-linear integral equations which is well suited for perturbative analysis around $\delta=0$. Finally, Sec.\,\ref{sec5} contains a summary of the original coordinate Bethe ansatz results \cite{Bukhvostov:1980sn} and the corresponding non-linear integral equations \cite{Saleur:1998wa}, as well as their numerical comparison with our calculation. \section{\label{sec2}Small-$R$ expansion} In this paper we shall mainly focus on the BL model with the vanishing exponent $\nu$ \eqref{aoisasosa}. Nevertheless it is useful to start with the theory characterized by a general set $(g,\nu)$. In the bosonic formulation, the model is still described by the Lagrangian \eqref{jssksjsak}, where the couplings $(a_1,a_2)$ substitute the pair $(g,\nu)$. These two pairs of renormalization group invariants are related as follows \cite{Bukhvostov:1980sn,Fateev:1996ea}:\footnote{Here, again, it is assumed that we are dealing with the regularization of the fermionic theory which preserves the $U(1)\otimes U(1)$ invariance.} \begin{eqnarray}\label{aikallaalsi} \nu=\frac{1}{2}\ (a_1+a_2-2)\ ,\ \ \ \ \ \ \ \ \frac{g}{\pi}=\frac{a_2-a_1}{2 a_1 a_2}\ . \end{eqnarray} Due to the periodicity of the potential term in $\varphi_i$, the space of states splits into the orthogonal subspaces ${\cal H}_{\bf k}$ characterized by two ``quasimomenta'' ${\bf k}=(k_1,k_2)$, \begin{eqnarray} \varphi_i\mapsto \varphi_i+\frac{4\pi}{\sqrt{a_i}}\ :\ \ \ \ |\,\Psi_{\bf k}\,\rangle\mapsto \mbox{e}^{2\pi{\rm i} k_i}\, |\,\Psi_{\bf k}\,\rangle\ . \end{eqnarray} As usual in the bosonization, the quasimomenta are related to the fermionic twists \eqref{apssspps}: \begin{eqnarray} k_\pm=\frac{1}{2}\ (k_1\pm k_2)\ .\label{apssspps2} \end{eqnarray} The neutral (w.r.t. $U(1)\otimes U(1)$) sector of the theory is described by the Bose fields with periodic boundary conditions: \begin{eqnarray}\label{bound} \varphi_i(x^0,x^1+R)=\varphi_i(x^0,x^1)\ . \end{eqnarray} In the Euclidean version of \eqref{jssksjsak}, the periodic boundary corresponds to the geometry of infinite (or very long in the``time'' direction $x^0$) flat cylinder \begin{eqnarray}\label{oasiosai} D=\big\{{\bf x}=(x^0,x^1)\,|\, -\infty<x^0<\infty,\ x^1\equiv x^1+R\big\}\ . \end{eqnarray} Then the ratio $E_{\bf k}/R$ would correspond to the specific (per unit length of the cylinder) free energy with the scalar operator $\exp\big({\rm i} (k_1\varphi_1+k_2\varphi_2)\big)$ ``flowing'' along the cylinder. The UV conformal dimension of this operator is $\Delta=\frac{1}{4}\sum_{i=1}^2a_ik_i^2$. Therefore, we expect that at $R\to 0$ \begin{eqnarray}\label{aosiaioas} E_{\bf k}\sim -\frac{\pi}{6R}\ c_{\bf k}\ ,\ \ \ \ \ c_{\bf k}=\sum_{i=1}^2\big(1-6 a_ik_i^2\big)\ . \end{eqnarray} The conformal perturbation theory for $E_{\bf k}$ is constructed in the usual way \cite{Zamolodchikov:1994uw} and yields an expansion in the dimensionless variable $\lambda=2\pi\mu\, \big(\frac{R}{2\pi}\big)^{1-\nu}$, \begin{eqnarray}\label{aasioaisaasi} E_{\bf k}=\frac{\pi}{R}\ \sum_{n=0}^\infty e_{n}^{(\nu)}\ \lambda^{2n}\ . \end{eqnarray} Here the first coefficient $e^{(\nu)}_0$ coincides with $-\frac{c_{\bf k}}{6}$, while the subsequent ones are given by the perturbative integrals. In particular $e_1^{(\nu)}=I(p_+)+I(p_-)$, where $p_\pm=\frac{1}{2}(a_1k_1\pm a_2k_2)$ and \begin{eqnarray} \label{sssai} I(p)&=& \int_D\frac{\mbox{d}^2 x}{ R^2} \frac{ 4^{-\nu}\, \pi\, \mbox{e}^{-\frac{2\pi}{R} (\nu+2p) x^0}}{ \big(\sinh\big(\frac{\pi}{R}(x^0+{\rm i} x^1)\big)\sinh\big(\frac{\pi}{R}(x^0-{\rm i} x^1)\big)\big)^{1+\nu}}\nonumber \\[.3cm] &=& \frac{\Gamma(\frac{1}{2}+p +\nu) \Gamma(\frac{1}{2}-p)\Gamma(-\nu)} {\Gamma(\frac{1}{2}-p-\nu) \Gamma(\frac{1}{2}+p) \Gamma(1+\nu)}\ . \end{eqnarray} In the opposite large-$R$ limit, the vacuum energy is composed of an extensive part which is proportional to the spatial size of the system and does not depends on the quasimomenta. The specific bulk energy, ${\cal E}\equiv\lim_{R\to\infty}E_{\bf k}/R$, has dimension $[\,mass\,]^2$, i.e., ${\cal E}/M^2$ is a certain function of the dimensionless couplings $(g,\nu)$. This universal ratio, along with another dimensionless combinations $\mu/M^{1-\nu}$, are fundamental characteristic of the theory, which allows one to glue together the small- and large-$R$ asymptotic expansions. It is convenient to extract the extensive part from $E_{\bf k}$ and introduce the scaling function \begin{eqnarray}\label{osaoasisoais} {\mathfrak F}(r,{\bf k})=\frac{R}{\pi}\ ( E_{\bf k}-R\ {\cal E})\ . \end{eqnarray} Notice that it is a dimensionless function of the dimensionless variables $r\equiv MR$ and ${\bf k}$ (and, of course, the couplings), satisfying the normalization condition \begin{eqnarray}\label{aosioaioasas} \lim_{r\to+\infty}{\mathfrak F}(r,{\bf k})= 0\ . \end{eqnarray} Also, since the value of $c_{\rm eff}\equiv-6\, {\mathfrak F}(r,{\bf k})$ at $r=0$ coincides with the UV effective central charge \eqref{aosiaioas}, this function can be interpreted as an effective central charge for the off-critical theory. After this preparation let us turn to the case $\nu=0$. Now, as it follows from the relations \eqref{aikallaalsi}, the parameters of the bosonic Lagrangian \eqref{jssksjsak} obey the constraint \begin{eqnarray} a_1+a_2=2\ , \end{eqnarray} which can be resolved as \begin{eqnarray} a_1=1-\delta\ ,\ \ \ a_2=1+\delta\ . \end{eqnarray} We will assume that $0<a_1\leq 1\leq a_2$, i.e., $0\leq \delta<1$. A formal substitution of $\nu=0$ in \eqref{sssai} leads to a divergent expression. In order to regularize $I(p)$, we cut a small disk $|{\bf x}|<\epsilon$ in the integration domain $D$. As $\epsilon\to 0$, the regularized integral diverges logarithmically: \begin{eqnarray}\label{saisssaoo} I^{(\epsilon)}(p)|_{\nu=0}=-2\, \log\big({\textstyle \frac{2\pi}{R}}\, \epsilon\big) -\psi\big({\textstyle\frac{1}{2}}+p\big)-\psi\big({\textstyle\frac{1}{2}}-p\big)-2\gamma_{\rm E}+o(1)\ , \end{eqnarray} where $\psi$ stands for the logarithmic derivative of the $\Gamma$-function and $\gamma_{\rm E}$ is the Euler constant. In the case $\nu=0$, the general small-$R$ expansion is substituted by the asymptotic series of the form \begin{equation}\label{aolissiao} \frac{RE_{\bf k}}{\pi}\asymp -\frac{1}{3}+\frac{4p_1^2}{1-\delta}+\frac{4p_2^2}{1+\delta}- (\mu R)^2\ \Big(e_1(0)-4\, \log\big({\textstyle \frac{2\pi}{R}}\, \epsilon\,\mbox{e}^{\gamma_E-\frac{1}{2}}\big)\Big)- \sum_{n=2}^\infty e_n(\delta)\ (\mu R)^{2n}\ , \end{equation} where explicitly \begin{eqnarray}\label{apspaspaspas} e_1(0)=-2-\psi\big({\textstyle\frac{1}{2}}+p_1+p_2\big)-\psi\big({\textstyle\frac{1}{2}}-p_1-p_2\big)- \psi\big({\textstyle\frac{1}{2}}+p_1-p_2\big)-\psi\big({\textstyle\frac{1}{2}}-p_1+p_2\big) \end{eqnarray} and \begin{eqnarray} \label{kasoisau} p_1=\frac{1}{2}\ (1-\delta)\ k_1\ ,\ \ \ \ \ p_2=\frac{1}{2}\ (1+\delta)\ k_2\ . \end{eqnarray} In ref.\,\cite{Fateev:1996ea} Fateev presented strong arguments supporting the integrability of the BL model with $\nu=0$ and found an exact $\mu-M$ relation, \begin{eqnarray}\label{jshsyusy} \mu={\textstyle \frac{M}{2\pi}}\ \cos\big({\textstyle\frac{\pi\delta}{2}}\big)\ . \end{eqnarray} Using his results it is straightforward to obtain (see Sec.\,\ref{secnew5} bellow) the following expression for the bulk specific energy \begin{eqnarray}\label{oaisaiasoso} {\cal E}=\pi\mu^2\ \Big(\,4\, \log\big( \pi\mu\epsilon\,\mbox{e}^{\gamma_{\rm E}-\frac{1}{2}}\big)+ \psi\big({\textstyle\frac{1+\delta}{2}}\big)+ \psi\big({\textstyle\frac{1-\delta}{2}}\big)-2\, \psi\big({\textstyle\frac{1}{2}}\big)\Big)\ . \end{eqnarray} One can see now that ${\mathfrak F}$, defined by eq.\,\eqref{osaoasisoais}, does not contain any UV divergences, i.e., it is an universal scaling function of the dimensionless variable $r=MR$. Its small-$R$ expansion can be written in the form \begin{eqnarray}\label{apspassp} {\mathfrak F}(r,{\bf k})\asymp -\frac{1}{3}+2k_+^2+2k_-^2-4\delta\,k_+k_-- 16\,\rho^2\, \log(\rho)- \sum_{n=1}^\infty e_n(\delta)\ (2\rho)^{2n}\ , \end{eqnarray} where $k_\pm=\frac{1}{2}(k_1\pm k_2)$, $\rho={\textstyle \frac{r}{4\pi}}\ \cos\big({\textstyle\frac{\pi\delta}{2}}\big)$ and \begin{eqnarray}\label{aosiosaaso} e_1(\delta)=e_1(0)+\psi\big({\textstyle\frac{1+\delta}{2}}\big)+ \psi\big({\textstyle\frac{1-\delta}{2}}\big)-2\, \psi\big({\textstyle\frac{1}{2}}\big)\ . \end{eqnarray} A few comments are in order here. As it was already mentioned in the introduction, the logarithmic divergence of ${\cal E}$ is well expected in the context of application of the BL model to the problem of instanton summation in the $O(3)$ sigma model. The integration over the instanton moduli space leads to the divergent contribution of the small-size instantons \cite{Luscher:1981tq}. So that $\epsilon$ can be interpreted as a cut-off parameter which allows one to exclude the divergent contribution of the small-instantons. Another comment concerns to the symbol $\asymp$, which is used in eqs.\,\eqref{aolissiao} and \eqref{apspassp} to emphasize the asymptotic nature of these power series expansions. To see that they have zero radius of convergence, it is sufficient to consider the case $\delta=0$. Returning to the fermionic description, the model \eqref{Lagr1} with $g=0$ constitutes a pair of non-interaction Dirac fermions, so that there exists a closed analytic expression for the scaling function ${\mathfrak F}_{0}={\mathfrak F}|_{\delta=0}$. Namely, $ {\mathfrak F}_0(r,{\bf k})= {\mathfrak f}(r,k_+)+ {\mathfrak f}(r,k_-)$, where $\pi {\mathfrak f}/R^2 $ coincides with the specific free energy of the free Dirac fermion at the temperature $1/R$ and (imaginary) chemical potential $2\pi{\rm i} k/R$, i.e., \begin{eqnarray}\label{apoasoasp} {\mathfrak f}\big(r,k)= -\frac{r}{2\pi^2}\ \int_{-\infty}^\infty\mbox{d}\theta\,\cosh(\theta) \ \log\Big[\Big(1+\mbox{e}^{2\pi{\rm i} k}\mbox{e}^{-r\cosh(\theta)}\Big) \Big(1+\mbox{e}^{-2\pi{\rm i} k}\mbox{e}^{-r\cosh(\theta)}\Big)\,\Big]\ . \end{eqnarray} It is now straightforward to see that the power series \eqref{apspassp} for $\delta=0$ is indeed an asymptotic expansion and \begin{eqnarray} e_n(0)&=&-2\,\delta_{n,1}+ \frac{(-1)^n\,n}{4^{n-1} (n!)^2}\ \ \Big(\, \psi^{(2n-2)}\big({\textstyle \frac{1}{2}}+p_1+p_2\big)+ \psi^{(2n-2)}\big({\textstyle \frac{1}{2}}-p_1-p_2\big)\nonumber \\ &+& \psi^{(2n-2)}\big({\textstyle \frac{1}{2}}+p_1-p_2\big)+ \psi^{(2n-2)}\big({\textstyle \frac{1}{2}}-p_1+p_2\big)\,\Big)\ , \end{eqnarray} where the superscript stands for derivative of $(2n-2)$-order w.r.t. the argument. For nonzero $\delta$ the asymptotic coefficients $e_n(\delta)$ with $n\geq 2$ can be expressed in terms of the multiple integrals. Unfortunately such representation can not be used for any practical purposes. The only exclusion is $e_2(\delta)$, whose integral representation can be simplified dramatically. For future references we describe here major steps in this calculation. First of all, using the complex coordinate $z=\exp(2\pi (x^0+{\rm i} x^1)/R)$, the asymptotic coefficient $e_2(\delta)$ can be represented as a 6-fold integral, \begin{eqnarray}\label{aosiosoaaa} e_2(\delta)&=& e_2(0)+ 2\ \int\prod_{i=1}^3\frac{\mbox{d}^2 z_i}{2\pi}\ |z_1|^{-1+2p_1+2p_2}\ |z_2|^{-1-2 p_1+2p_2}\ |z_3|^{-1-2p_1-2p_2}\ \nonumber\\ &\times& \bigg(\,\bigg|\frac{(z_1-z_2)(1-z_3)}{(z_3-z_2)(1-z_1)}\bigg|^{2 \delta}-1\bigg)\ \big|(1-z_2)(z_1-z_3)\big|^{-2}\ . \end{eqnarray} Now let us substitute the integration variable $z_2$ with $\zeta=\frac{(1-z_1)(z_2-z_3)}{(1-z_2)(z_1-z_3)}$, and integrate over $z_1$ by means of the identity \begin{eqnarray}\label{opasssaaspsp} \int\frac{\mbox{d}^2 z}{\pi}\, |z|^{-1+2p_1+2p_2}\, |1-z|^{-1+2 p_1-2p_2}\, |z-w|^{-1-2p_1+2p_2}= |w|^{-1+2p_2}\ |1-w|\ \tau_{p_1 p_2}\big({\textstyle\frac{1}{1-w}}\big)\ . \end{eqnarray} Here \begin{eqnarray}\label{tau12} \tau_{p_1p_2}(\zeta) &=& \frac{\Omega(-p_1,p_2) }{2p_1}\ |\zeta |^{1-2p_1}|1-\zeta|^{1+2 p_2} \big|{}_2F_1\big({\textstyle \frac{1}{2}}- p_1+p_2, {\textstyle \frac{1}{2}}- p_1+p_2, 1-2 p_1;\zeta\big)\big|^2\nonumber\\ &-&\frac{\Omega(p_1,p_2) }{2p_1}\ |\zeta|^{1+2p_1}|1-\zeta|^{1-2 p_2} \big|{}_2F_1\big({\textstyle \frac{1}{2}}+ p_1-p_2, {\textstyle \frac{1}{2}}+ p_1-p_2, 1+2 p_1;\zeta\big)\big|^2 , \end{eqnarray} ${}_2F_1$ stands for the conventional hypergeometric function, and \begin{eqnarray} \Omega(p_1,p_2)=\frac{\Gamma(\frac{1}{2}+p_1-p_2)\Gamma(\frac{1}{2}+p_1+p_2)} {\Gamma(\frac{1}{2}-p_1-p_2)\Gamma(\frac{1}{2}-p_1+p_2)}\ \frac{\Gamma(1-2p_1)}{\Gamma(1+2p_1)}\ . \end{eqnarray} Finally, the integral over $z_3$ can be performed using a remarkable relation \begin{eqnarray}\label{ahssaajh} \big(\tau_{p_1 p_2}(\zeta)\big)^2= |\zeta|^2\ \int \frac{\mbox{d}^2 z}{\pi|z|^2}\ \Big|\frac{1-\zeta z }{z (z-\zeta)}\Big|^{2p_1}\ \frac{\tau_{p_1 p_2}\big(X(z)\big)}{|X(z)|}\ , \end{eqnarray} where $X(z)=\frac{(\zeta-z)(1-\zeta z)}{\zeta(1-z)^2}$. As a final result one obtains the following integral representation \begin{eqnarray} \label{aasopaps} e_2(\delta)=e_2(0)+\frac{1}{4\pi}\ \int\frac{\mbox{d}^2\zeta}{|\zeta|^{2}|1-\zeta|^{2}}\ \Big( |\zeta|^{-2\delta}|1-\zeta|^{2\delta}-1\Big)\ \tau^2_{p_1 p_2}(\zeta)\ . \end{eqnarray} This formula allows one to achieve a reliable accuracy in the numerical calculation of $e_2(\delta)$. For illustration, we present in Fig.\ref{fig6bb} the numerical results for $k_1=k_2=0$. Notice that in this case the corresponding function $\tau_{00}(\zeta)$ in \eqref{aasopaps} can be expressed in terms of the complete elliptic integral of the first order $K(\zeta)= \frac{\pi}{2}\ {}_2F_{1}(\frac{1}{2},\frac{1}{2},1;\zeta)$: \begin{eqnarray} \tau_{00}(\zeta)=\frac{8}{\pi}\ |\zeta(1-\zeta)|\ \Re e\big(K^*( \zeta)K(1-\zeta)\big)\ . \end{eqnarray} Note that $\tau_{p_1 p_2}(\zeta)$ given in \eqref{tau12} is a particular case of a more general function $\tau_{p_1 p_2 p_3}(\zeta)$ defined by \eqref{oaissisaoi}, namely $\tau_{p_1 p_2}(\zeta)\equiv \tau_{p_1 p_2 p_3}(\zeta)|_{p_3=0}$. This function defines a real solution \eqref{liouville1} of the Liouville equation \eqref{liouville2}, satisfying the asymptotic conditions \eqref{paspapsoasa}. \begin{figure} \centering \includegraphics[width=9.5cm]{plot_delta_e2.eps} \caption{ The difference $\Delta e_2\equiv e_2(\delta)-e_2(0)$, defined by \eqref{aasopaps}, as a function of $\delta$ for $k_1=k_2=0$. Note that in this case $\Delta e_2=48.21714061416\ldots\times \delta^2 + 60.427986409885\ldots\times\, \delta^4+O(\delta^6)$ as $\delta\to 0$. } \label{fig6bb} \end{figure} \section{\label{sec3}Weak coupling expansion} We now consider a weak coupling expansion of the scaling function ${\mathfrak F}$. Since $g=\frac{\pi\delta}{1-\delta^2}= \pi\delta+O(\delta^3)$, no needs to distinguish $\frac{g}{\pi}$ and $\delta$ within the first two perturbative orders. It is convenient to define the perturbative coefficients through the relation: \begin{eqnarray}\label{apospapaso} {\mathfrak F}={\mathfrak F}_{0} + {\mathfrak F}_{1}\, \delta+ {\mathfrak F}_{2}\, \delta^2+O(\delta^3)\ . \end{eqnarray} Here ${\mathfrak F}_0(r,{\bf k})={\mathfrak f}(r,k_+)+ {\mathfrak f}(r,k_-)$ with ${\mathfrak f}$ given by \eqref{apoasoasp} (recall that $k_\pm=\frac{1}{2}\,(k_1\pm k_2)$). The results obtained in the previous section allows one to predict the leading small-$R$ behavior of ${\mathfrak F}_{i}$. Generally speaking the coefficients in the power series \eqref{apspassp} admit the Taylor expansion $ e_n(\delta)=e_n(0)+ \big(\frac{\partial e_n(0)}{\partial \delta}\big)_{p}\ \delta+ \frac{1}{2}\ \big(\frac{\partial^2 e_n(0)}{\partial \delta}\big)_{p}\ \delta^2+O(\delta^3)$. In particular, as it follows from eq.\eqref{aosiosaaso}, \begin{eqnarray} {\textstyle \big(\frac{\partial e_1(0)}{\partial \delta}\big)_{p}=0\ ,\ \ \ \ \big(\frac{\partial e_1(0)}{\partial \delta}\big)_{p}=\frac{1}{2}\ \psi''''\big({\textstyle\frac{1}{2}}\big)\ .} \end{eqnarray} Also, using the original integral representation \eqref{aosiosoaaa} for $e_2(\delta)$, one can show that \begin{eqnarray} \big({\textstyle \frac{\partial e_2(0)}{\partial \delta}}\big)_{p}= {\textstyle\frac{1}{4}}\, \big(\, \psi'\big({\textstyle \frac{1}{2}}+p_1+p_2\big)- \psi'\big({\textstyle \frac{1}{2}}-p_1-p_2\big)\big)\,\big( \psi'\big({\textstyle \frac{1}{2}}+p_1-p_2\big)- \psi'\big({\textstyle \frac{1}{2}}-p_1+p_2\big)\,\big)\, . \end{eqnarray} In the case $p_1=p_2=0$ the weak coupling expansion includes only even powers of $\delta$ (see Fig.\,\ref{fig6bb}) and \begin{eqnarray} {\textstyle \big(\frac{\partial^2 e_2(0)}{\partial \delta^2}\big)_{p_1=p_2=0}=-{\textstyle\frac{1}{8}}\ \psi''''\big({\textstyle\frac{1}{2}}\big)\ .} \end{eqnarray} All of these can be used to study the short distance expansion of ${\mathfrak F}_{i}$ in eq.\eqref{apospapaso}.\footnote{ Recall that the relations \eqref{kasoisau} between $k_i$ and $p_i$ involve the perturbative coupling. This should be taken into account since it is assumed that the expansion \eqref{apospapaso} is performed for fixed values of $k_i$ rather then $p_i$.} In particular, it is possible to show that \begin{eqnarray}\label{papospasoa} {\mathfrak F}_1(r,{\bf k})=- 4\,{\mathfrak q}(r,k_+)\, {\mathfrak q}(r,k_-)\ , \end{eqnarray} where \begin{equation}\label{paospspasopsa} {\mathfrak q}(r,k)= k+ \big(\psi'\big({\textstyle\frac{1}{2}} +k\big)-\psi'\big({\textstyle\frac{1}{2}} -k\big)\big) \Big({\frac{r}{4\pi}}\Big)^2-{\frac{1}{2}}\, \big(\psi'''\big({\textstyle\frac{1}{2}} +k\big)- \psi'''\big({\textstyle\frac{1}{2}} -k\big)\big)\, \Big({\frac{r}{4\pi}}\Big)^4 +O(r^6) \end{equation} and also \begin{eqnarray} { {\mathfrak F}_{2}= \frac{r^2}{4}\ \log\Big(\frac{r}{4\pi}\Big) +A\ \Big({\frac{r}{4\pi}}\Big)^2 - B\ \Big({\frac{r}{4\pi}}\Big)^4+O(r^6)\ ,} \end{eqnarray} where \begin{eqnarray} A&=&-\psi''\big({\textstyle\frac{1}{2}}\big)- \pi^2\ \Big(\psi\big({\textstyle\frac{1}{2}} +k_+\big)+\psi\big({\textstyle\frac{1}{2}} +k_-\big)+ \psi\big({\textstyle\frac{1}{2}} -k_+\big)+\psi\big({\textstyle\frac{1}{2}} -k_-\big)\,\Big)\\ &+& 2 k_+^2\, \Big(\psi''\big({\textstyle\frac{1}{2}} +k_-\big)+\psi''\big({\textstyle\frac{1}{2}} -k_-\big)\,\Big)+ 2k^2_-\, \Big(\psi''\big({\textstyle\frac{1}{2}} +k_+\big)+\psi''\big({\textstyle\frac{1}{2}} -k_+\big)\,\Big)\ . \nonumber \end{eqnarray} In the case $k_1=k_2=0$, \begin{eqnarray}\label{asaposoosap} A|_{k_1=k_2=0}= -\psi''\big({\textstyle\frac{1}{2}}\big)- 4\pi^2\, \psi\big({\textstyle\frac{1}{2}}\big)\ ,\ \ \ \ B|_{k_1=k_2=0}=-\psi''''\big({\textstyle\frac{1}{2}}\big)-4\pi^2\, \psi''\big({\textstyle\frac{1}{2}}\big)\ . \end{eqnarray} For finite values of $r$ the perturbative coefficients ${\mathfrak F}_i$ can be calculated within the renormalized perturbation theory based on Lagrangian \eqref{Lagr2}. Let ${\boldsymbol S}_\sigma({\bf x})\equiv \langle\, \psi_\sigma ({\bf x})\otimes {\bar \psi}_\sigma({\bf 0})\,\rangle $ (${\bf x}=(x^0,x^1),\,\sigma=\pm)$ be the fermionic Matsubara propagator with the temperature $1/R$ and chemical potential $2\pi{\rm i} k_\sigma/R$. It can be expressed in terms of the modified Bessel function of the second kind $K_s(z)=\frac{1}{2}\ \int_{-\infty}^\infty\mbox{d}\theta\ \mbox{e}^{s\theta-z\cosh(\theta)}$, \begin{eqnarray} {\boldsymbol S}_\sigma({\bf x})=\big(M-\gamma^a\partial_a\big)\ G_\sigma({\bf x})\ , \end{eqnarray} where $\gamma^a$ are Euclidean $\gamma$-matrices, $\{\gamma^a,\gamma^b\}=2\,\delta^{ab}$, and \begin{eqnarray} G_\sigma({\bf x})=\frac{1}{2\pi}\ \sum_{n=-\infty}^\infty(-1)^{n}\ \mbox{e}^{2\pi{\rm i} n k_\sigma }\ K_0\big(|w-{\rm i}\,n r|\big)\ \ \ \ \ \ \ {\rm with}\ \ \ \ \ w=M\, (x^0+{\rm i}\,x^1)\ . \end{eqnarray} At the first perturbative order one has (see Fig.\,\ref{fig2a}) \begin{eqnarray} {\mathfrak F}_1= R^2\ {\rm Tr}\big({\boldsymbol S}_+(0)\gamma_a\big)\,{\rm Tr}\big({\boldsymbol S}_-(0)\gamma^a\big)= 4R^2\ \langle\, \Psi^\dagger_+\, \Psi_+(0)\,\rangle\, \langle\, {\bar \Psi}^\dagger_-\, {\bar \Psi}_-(0)\,\rangle\ . \end{eqnarray} Here $\Psi_\sigma$ and ${\bar \Psi}_\sigma$ stand for the components of the Dirac bispinors $\psi_\sigma$ with the Lorentz spin $+\frac{1}{2}$ and $-\frac{1}{2}$, respectively. \begin{figure} \centering \psfrag{a}{$-$} \psfrag{b}{$+$} \includegraphics[width=4.5cm]{1pert.eps} \caption{A diagrammatic representation of ${\mathfrak F}_1$ in eq.\,\eqref{apospapaso}. The signs $\pm$ label the fermion ``colors'' $\psi_\pm$ propagating along the loops (see Lagrangian \eqref{Lagr2}). } \label{fig2a} \end{figure} In zero-temperature limit the Lorentz invariance is restored and hence $ \langle\, {\bar \Psi}^\dagger_\sigma\, {\bar \Psi}_\sigma(0)\,\rangle= -\langle\, \Psi^\dagger_\sigma\, \Psi_\sigma(0)\,\rangle\to 0$. Introducing function ${\mathfrak q}$ through the relation \begin{eqnarray}\label{paopsopas} \langle\, {\bar \Psi}^\dagger_\sigma\, {\bar \Psi}_\sigma(0)\,\rangle=- \langle\, \Psi^\dagger_\sigma\, \Psi_\sigma(0)\,\rangle= \frac{1}{R}\ {\mathfrak q}(r,k_\sigma)\ , \end{eqnarray} one observes that ${\mathfrak F}_1$ takes the form \eqref{papospasoa}. It is also easy to see that \begin{eqnarray}\label{laaaksussa} {\mathfrak q}=\frac{1}{4}\ \frac{\partial {\mathfrak f}}{\partial k}\ , \end{eqnarray} where ${\mathfrak f}={\mathfrak f}(r,k)$ is given by \eqref{apoasoasp}. This is in a complete agreement with the short distance prediction \eqref{paospspasopsa}. The second-order diagrams are depicted in Fig.\,\ref{fig3a}. \begin{figure} \centering \psfrag{a}{$+$} \psfrag{b}{$-$} \psfrag{c}{$-\sigma$} \psfrag{d}{$\sigma$} \psfrag{G}{${\rm (I)}$} \psfrag{H}{${\rm (II)}$} \psfrag{W}{${\rm (III)}$} \includegraphics[width=16cm]{2pert.eps} \caption{The diagrams contributing to the second perturbative order. The contribution of the counterterm $\propto g_1$ in \eqref{Lagr2} is visualized by the type III diagrams (as $\nu=0$, there is no mass renormalization, i.e. $\delta M=0$). } \label{fig3a} \end{figure} The type I diagram gives the contribution \begin{eqnarray}\label{appsopsaoyss} {\mathfrak F}_2^{\rm (I)}=-\frac{\pi}{2}\ r^2\, \int_{D_\epsilon}\mbox{d}^2x\ {\rm Tr}\big({\boldsymbol S}_+(-{\bf x})\,\gamma_a\, {\boldsymbol S}_+({\bf x})\, \gamma_b\big)\ {\rm Tr}\big({\boldsymbol S}_-(-{\bf x})\,\gamma^a\, {\boldsymbol S}_-({\bf x})\, \gamma^b\big)\ . \end{eqnarray} Because of the UV divergence at ${\bf x}=0$, the integration domain $D_\epsilon$ here is chosen to be the cylinder \eqref{oasiosai} without an infinitesimal hole $|{\bf x}|<\epsilon$. One can show that, as $\epsilon$ tends to zero, \begin{eqnarray} {\mathfrak F}^{\rm (I)}_2=-\Big(\frac{R}{2\pi\epsilon}\Big)^2+ \sum_{\sigma=\pm}\Big( {\mathfrak t}(r,k_\sigma)- \frac{r}{2\pi}\, \log\big({\textstyle\frac{M\epsilon}{2}}\, \mbox{e}^{\gamma_E- \frac{1}{2}}\big) \Big)^2+{\rm finite}\ , \end{eqnarray} where \begin{eqnarray} {\mathfrak t}=-\pi\ \frac{\partial {\mathfrak f}}{\partial r}\ . \end{eqnarray} In fact, since $E_{\bf k}=R\,{\cal E}+\frac{\pi}{R}\, {\mathfrak F}$, the quadratic divergence $\propto 1/\epsilon^2$ should be relocated to the specific bulk energy. Generally speaking, the specific bulk energy has a form \begin{eqnarray}\label{paopsaoaops} {\cal E}= w(g)\, \Lambda^2+\frac{M^2}{\pi}\, \cos^2\big({\textstyle\frac{\pi\delta}{2}}\big)\ \log\big(M/\Lambda\big)+o(1)\ , \end{eqnarray} where $\Lambda\gg M$ is some lattice energy scale and $w$ is some (nonuniversal) function of the coupling $g$. Notice that, in writing eq.\,\eqref{oaisaiasoso}, the quadratic divergence was omitted (as usual in QFT). The type II diagrams from Fig.\,\ref{fig3a} leads to the UV finite integral over the whole cylinder $D$: \begin{eqnarray} {\mathfrak F}^{\rm (II)}_2=\frac{\pi}{2}\ r^2\, \int_{D}\mbox{d}^2x\ \sum_{\sigma=\pm} {\rm Tr}\big({\boldsymbol S}_{\sigma}(0)\,\gamma_a\big)\, {\rm Tr}\big({\boldsymbol S}_{-\sigma}(-{\bf x})\,\gamma^a\, {\boldsymbol S}_{-\sigma}({\bf x})\, \gamma^b\big)\, {\rm Tr}\big({\boldsymbol S}_{\sigma}(0)\,\gamma_b\big) \ . \end{eqnarray} Finally, the counterterm $\propto g_1$ in \eqref{Lagr2} contributes through the type III diagrams, schematically visualized in Fig.\,\ref{fig3a}. This can be written in the form $ \frac{2g_1}{\pi}\ {\mathfrak F}^{\rm (III)}_2 $ with \begin{eqnarray}\label{aapasosaosa} {\mathfrak F}^{\rm (III)}_2&=& \frac{1}{4}\ R^2\ \sum_{\sigma=\pm}\Big( {\rm Tr}\big({\boldsymbol S}_\sigma (0)\gamma^a\big)\, {\rm Tr}\big({\boldsymbol S}_\sigma(0)\gamma_a\big) -{\rm Tr}\big({\boldsymbol S}_\sigma(0)\gamma^a{\boldsymbol S}_\sigma(0)\gamma_a\big)\Big)\nonumber\\ &=&R^2\sum_{\sigma=\pm }\big(\langle\, \Psi^\dagger_\sigma\, \Psi_\sigma(0)\,\rangle \langle\, {\bar \Psi}^\dagger_\sigma\, {\bar \Psi}_\sigma(0)\,\rangle- \langle\, {\bar \Psi}_\sigma\, \Psi^\dagger_\sigma(0)\,\rangle^2\big)\ . \end{eqnarray} Contrary to the one point functions \eqref{paopsopas}, the condensate $\langle\, {\bar \Psi}_\sigma\, \Psi^\dagger_\sigma(0)\,\rangle$ diverges logarithmically: \begin{eqnarray}\label{aaoisosaoa} \langle\, {\bar \Psi}_\sigma\, \Psi^\dagger_\sigma(0)\,\rangle= \langle\, { \Psi}_\sigma\, {\bar \Psi}^\dagger_\sigma(0)\,\rangle= \frac{1}{R}\ {\mathfrak t}(r,k_\sigma) - \frac{M}{2\pi}\ \log\Big(\frac{M\epsilon}{2} \, \mbox{e}^{\gamma_E-\frac{1}{2}+C} \Big)\ , \end{eqnarray} where $C$ is some constant. Since \begin{eqnarray}\label{spsopsop} {\mathfrak F}^{\rm (I)}_2+ {\mathfrak F}^{\rm (III)}_2+\Big(\frac{R}{2\pi\epsilon}\Big)^2=-\frac{r^2 }{\pi^2 }\ C\ \log(M\epsilon)+{\rm finite}\ , \end{eqnarray} the UV divergence $\propto \log^2(\epsilon)$ is canceled from the sum of types I and III diagrams if we choose $g_1=\frac{g^2}{2\pi}+O(g^3)$. As well as the quadratic divergence, the remaining logarithmic divergence should be relocated to the specific bulk energy. Expanding $\cos^2\big({\textstyle\frac{\pi\delta}{2}}\big)$ in \eqref{paopsaoaops} one can find the value of the constant $C$: \begin{eqnarray}\label{aisiaosai} C=\frac{\pi^2}{4}\ . \end{eqnarray} This way the second order correction takes the form \begin{eqnarray}\label{apaspospa} {\mathfrak F}_{2}= \frac{r^2}{4\pi^2}\ {\cal C}_2+ \lim_{\epsilon\to 0}\bigg[\, \sum_{\alpha={\rm I,II,III}}{\mathfrak F}^{(\alpha)}_2+ \Big(\frac{R}{2\pi\epsilon}\Big)^2+ \frac{r^2 }{4 }\ \log\Big(\frac{M\epsilon}{2} \, \mbox{e}^{\gamma_{\rm E}-\frac{1}{2}}\Big) \,\bigg]\ , \end{eqnarray} where the finite constant should be adjusted to satisfy the normalization condition \eqref{aosioaioasas}. It reads explicitly as \begin{eqnarray} {\cal C}_2= \frac{\pi^4}{8}-\frac{1}{2}-\frac{1}{4}\ \psi''\big({\textstyle\frac{1}{2}}\big)\ . \end{eqnarray} Further calculations show that \begin{eqnarray}\label{sauiuias} {\mathfrak F}_{2}(r,{\bf k})&=& -\frac{1}{2}\,\ \big(1+c(2k_1)+ c(2k_2)\big)\ \, r^2\,K_0(2 r) \\ &-&\big(1-c(2k_1) c(2k_2)\big)\ r\, \int_{-\infty}^\infty\frac{\mbox{d} \nu}{\pi}\ \frac{\nu^2\, K_{{\rm i}\nu}(r)K_{1+{\rm i}\nu}(r)} {\sinh^2(\frac{\pi\nu}{2})} +o\big(\mbox{e}^{-2r}\big)\ .\nonumber \end{eqnarray} Here the shortcut notation $c(k)=\cos(\pi k)$ is used and $K_{s}(z)$ denotes the modified Bessel function of the second kind: \begin{eqnarray} K_s(z)=\frac{1}{2}\ \int_{-\infty}^\infty\mbox{d}\theta\ \mbox{e}^{s\theta-z\cosh(\theta)}\ . \end{eqnarray} Also in eq.\,\eqref{sauiuias} and bellow, the symbol $o\big(\mbox{e}^{-2r}\big)$ denotes a remaining term that decays faster than $r^{-N}\ \mbox{e}^{-2r}$ for any positive $N$ as $r\to+\infty$. Notice that the normalization condition ${\mathfrak F}_{2}=o\big(\mbox{e}^{-r}\big)$ implies an absence of the finite renormalization of the fermion mass. It can be used for fixing the constant $C$ in \eqref{aaoisosaoa} and hence avoid any reference to the exact relation \eqref{paopsaoaops}. For $k_1=k_2=0$, the result of perturbative calculation is presented in Fig.\,\ref{fig4}. \begin{figure} \centering \includegraphics[width=8cm]{plot_perturbation_theory_0.eps} \includegraphics[width=8cm]{plot_perturbation_theory_2.eps} \caption{The perturbative coefficient ${\mathfrak F}_{i}$ \eqref{apospapaso} for $k_1=k_2=0$ $({\mathfrak F}_{1}=0$ in this case). The left panel shows ${\mathfrak F}_{0}= -\frac{2r}{\pi^2}\ \int_{-\infty}^\infty\mbox{d}\theta\,\cosh(\theta) \, \log\big(1+\mbox{e}^{-r\cosh(\theta)}\big)$. At the right panel ${\mathfrak F}_{2}$ is compared against its large-$r$ asymptotic ${\mathfrak F}_{2}=-\frac{3}{2}\,r^2\,K_0(2r)+o(\mbox{e}^{-2r})$ (blue dashed line) and the small-$r$ asymptotic ${\mathfrak F}_{2}= \frac{r^2}{4} \, \log\big(\frac{r}{4\pi}\big)+A\, \big(\frac{r}{4\pi}\big)^2-B\, \big(\frac{r}{4\pi}\big)^4+O(r^6)$ (red dashed line). The numerical coefficients $A$ and $B$ are given by eq.\eqref{asaposoosap}. } \label{fig4} \end{figure} Using eqs.\,\eqref{papospasoa} and \eqref{laaaksussa} it is easy to show that \begin{eqnarray}\label{oaaaaasisao} {\mathfrak F}_{1}(r,{\bf k}) =\frac{2}{\pi^2}\ \big(c(2k_1)-c(2k_2)\big) \ r^2\,K_1^2(r)+o\big(\mbox{e}^{-2r}\big)\ . \end{eqnarray} Thus, at least at the first-two perturbative orders, the leading large-$r$ behavior of the scaling function ${\mathfrak F}$ is defined by ${\mathfrak F}_{0}$ only and therefore \begin{eqnarray}\label{aosioasiasoas} {\mathfrak F}(r,{\bf k})=-\frac{4}{\pi^2}\ c(k_1)\, c(k_2)\ r\,K_1(r)+o(\mbox{e}^{-r})\ . \end{eqnarray} This can be understood as follows. The leading large-$R$ behavior comes from the virtual fermions trajectories winding once around the Matsubara circle. Such trajectories should be counted with the phase factor $\mbox{e}^{{\rm i}\pi (\sigma_1k_1+\sigma_2 k_2)}$ and, therefore, the summation over four possible sign combinations with $\sigma_{1,2}=\pm 1$ gives rise eq.\eqref{aosioasiasoas}. Thus we may expect that the asymptotic formula \eqref{aosioasiasoas} holds true as the mass of the first bound state $M_1=2M\,\cos(\frac{\pi\delta}{2})$ is greater than $M$, i.e., for $\delta\in[0, \frac{2}{3})$. Before concluding this section let us make a few remarks about the (non-integrable) case with a non-zero value of $\nu$. Instead of adjusting the counterterm coupling $g_1$, the logarithmic divergences can be absorbed by the mass counterterm with $\delta M=\nu\ \log(M/\Lambda_{\rm UV})$, where $\nu=\frac{g^2}{\pi^2}-\frac{2 g_1}{\pi} $ and $\Lambda_{\rm UV}=\frac{2}{\epsilon}\ \exp(\frac{1}{2}-\gamma_{\rm E}-\frac{\pi^2}{4})$. (This is an infinitesimal version of eq.\,\eqref{aoisasosa} where $M_0=M+\delta M$.) As it was mentioned in the introduction, the exponent $\nu$ and the four-fermion coupling $g$ can be thought of as independent parameters for the family of BL models. Using eq.\,\eqref{aapasosaosa}, it is easy to see that \begin{eqnarray} \frac{\partial {\mathfrak F}}{\partial\nu}\Big|_{\nu=0}= \sum_{\sigma=\pm}\big( {\mathfrak q}^2(r,k_\sigma)+ {\mathfrak t}^2(r,k_\sigma)\big)+O(g)\ . \end{eqnarray} Finally we note that for $\nu\not=0$, (an universal part of) the specific bulk energy has a valid Laurent expansion of the form \begin{eqnarray} {\cal E}(g,\nu)=M^2\, \Big(\,h_{-1}(g)\, \nu^{-1}+\sum_{n=0}^\infty h_{n}(g)\ \nu^n\, \Big)\ , \end{eqnarray} where $h_{n}(g)$ admit power series expansions in $g^2$. \section{\label{secnew5}Exact formula for ${\mathfrak F}(r,{\bf k})$} The BL model with non-vanishing $\nu=\frac{1}{2}\, (a_1+a_2-2)$ can be thought as a sort of analytical regularization of the model with $\nu=0$ -- the integrals appearing in the conformal perturbation theory converge for negative values of $\nu$, but become singular at $\nu\to 0^-$. A brief inspection of eq.\,\eqref{sssai} shows that a simple pole $\frac{1}{\nu}$ replaces the logarithmic divergence $2\, \log(\epsilon \mu)+const$ in \eqref{saisssaoo} which occurs when the integral is regularized by excluding a neighborhood of the singular point from the integration domain. The BL with non-vanishing $\nu$ is a well defined QFT and it is interesting in itself in a context of applications in condensed matter physics \cite{Lesage:1997jq}. However, as it was already mentioned in the introduction, the ``$\nu$-deformation'' spoils the integrability. Remarkably that there exists an integrable deformation of the BL model with $\nu=0$. The corresponding model was introduced by Fateev in the works \cite{Fateev:1995ht,Fateev:1996ea} and it will be referred to bellow as the Fateev model. Contrary to the BL model, the Fateev (F) model involves three Bose fields governed by the Lagrangian \begin{eqnarray}\label{aposoasio} {\tilde {\cal L}}_{\rm F}&=& \frac{1}{16\pi}\ \sum_{i=1}^3 \big(\, (\partial_0\varphi_i)^2-(\partial_1\varphi_i)^2\,\big)\\ &+&2\mu\ \big(\, \mbox{e}^{{\rm i}\, \alpha_3\varphi_3}\ \cos(\alpha_1\varphi_1+\alpha_2\varphi_2)+\mbox{e}^{-{\rm i} \alpha_3\varphi_3}\ \cos(\alpha_1\varphi_1-\alpha_2\varphi_2)\,\big)\, . \nonumber \end{eqnarray} Here $\alpha_i=\frac{1}{2}\sqrt{a_i}$ and the coupling constants $a_i$ satisfy a single constraint \begin{eqnarray}\label{aposapoas} a_1+a_2+a_3=2\ , \end{eqnarray} which implies that the parameter $\mu$ has a dimension of mass. As $\alpha_3\to 0$, the field $\varphi_3$ decouples in \eqref{aposoasio} and the interacting part coincides with the bosonic version of the BL Lagrangian \eqref{jssksjsak} with $a_1+a_2=2$. In fact, this observation requires a more careful assessment. Performing the limit $\alpha_3\to 0$, one should expand the exponentials $\mbox{e}^{\pm{\rm i}\, \alpha_3\varphi_3}$ in \eqref{aposoasio} to the terms $\propto a_3=4 \alpha^2_3$. The mass of the decoupled field is given by the relation $m^2=8\pi\mu a_3\, \langle\, \cos(\alpha_1\varphi_1)\cos(\alpha_2\varphi_2)\,\rangle$, where the vacuum expectation value is taken for the BL model with $\nu=0$. This expectation value is simply related to the corresponding specific bulk energy, $\langle\, \cos(\alpha_1\varphi_1)\cos(\alpha_2\varphi_2)\,\rangle= -\frac{1}{4}\ \frac{\partial {\cal E}}{\partial \mu}$, and hence \begin{eqnarray} \label{ossiis} m^2=-2\pi \mu\, \lim_{a_3\to 0} \,\Big( a_3\ \frac{\partial {\cal E}}{\partial \mu}\,\Big)\ . \end{eqnarray} Eq.\,\eqref{oaisaiasoso} shows that ${\cal E}=4\pi\mu^2\,\log(\mu\epsilon)+\ldots$ and, as has been argued above, should be replaced by ${\cal E}= \frac{4\pi\mu^2}{a_1+a_2-2}+\ldots$ within the analytical regularization. This, combined with \eqref{ossiis} and the constraint $a_3=2-a_1-a_2$, means that the field $\varphi_3$ has the mass $m=4\pi\mu$ in the decoupling limit. Taking into account $M-\mu$ relation \eqref{jshsyusy}, one finally obtains \begin{eqnarray} m= 2M\, \cos\big({\textstyle\frac{\pi\delta}{2}}\big)\ , \end{eqnarray} where we use $\delta=1-a_1=a_2-1$. One of Fateev's important results concerning the theory \eqref{aposoasio} is an elegant analytical expression for the specific bulk energy \cite{Fateev:1996ea}: \begin{eqnarray}\label{akjsaskj} {\cal E}_{\rm F}=-\pi\mu^2\ \prod_{i=1}^3\frac{\Gamma(\frac{a_i}{2})}{\Gamma(1-\frac{a_i}{2})}\ . \end{eqnarray} The linear constraint imposed on parameters $a_i$, can be resolved by setting $a_1=1-\delta-\frac{a_3}{2}$ and $a_2=1+\delta-\frac{a_3}{2}$, and, therefore, as $a_3\to 0$ one has \begin{eqnarray}\label{saoasiosiosao} {\cal E}_{\rm F}=\pi \mu^2\, \Big( -\frac{2}{a_3}-4\log 2+ \psi\big({\textstyle\frac{1+\delta}{2}}\big)+\psi\big({\textstyle\frac{1-\delta}{2}}\big)- 2\psi\big({\textstyle\frac{1}{2}}\big)+o(1)\,\Big)\ . \end{eqnarray} Keeping in mind that $\frac{1}{a_3}$ can be substituted by $(-\log(\mu\epsilon))$ one find the relation \begin{eqnarray} {\cal E}_{\rm F}\to {\cal E}+const\ m^2\ \ \ \ {\rm as}\ \ \ \ a_3\to 0\ , \end{eqnarray} where ${\cal E}$ is the specific bulk energy for the BL model \eqref{oaisaiasoso}, whereas the term $\propto m^2$ is a contribution of the free massive field. Notice that $const$ does not depend on the coupling $\delta$, and it can be always set to zero. We can consider now the Fateev model in finite volume with the periodic boundary conditions $\varphi_i(x^0,x^1+R)= \varphi_i(x^0,x^1)$ imposed on all three fields $\varphi_i$. Similar to the definition \eqref{osaoasisoais} for the BL model, let us introduce ${\mathfrak F}_{\rm F}=R\, ( {E}_{\bf k}-R{\cal E}_{\rm F})/\pi$. Then the above consideration suggests that \begin{eqnarray}\label{apossoasop} \lim_{a_3\to 0^-}{\mathfrak F}_{\rm F}={\mathfrak F}(r,{\bf k})+ {\mathfrak f}_{\rm B}\big(2r c\big({\textstyle\frac{\delta}{2}}\big)\big)\ , \end{eqnarray} where the second term in the r.h.s. with \begin{eqnarray} {\mathfrak f}_{\rm B}(\beta)=\frac{\beta}{2\pi^2}\ \int_{-\infty}^\infty\mbox{d}\theta\ \cosh(\theta)\ \log\big(1-\mbox{e}^{-\beta\cosh(\theta)}\big)\ , \end{eqnarray} corresponds to a contribution of the free boson of mass $2M c\big({\textstyle\frac{\delta}{2}}\big)$ with $c(x)\equiv\cos(\pi x)$. Notice that the limit in \eqref{apossoasop} should be taken from negative values of $a_3$, so that the Lagrangian \eqref{aposoasio} is real. For $a_i>0 \ (i=1,2,3)$ the Lagrangian is complex, but the QFT is still well defined. In this case the potential term in \eqref{aposoasio} is periodic w.r.t. all fields $\varphi_i$ and the space of states splits on the orthogonal subspaces characterized by a triple of quasimomenta ${\bf k}=(k_1,k_2,k_3)$. For $a_3<0$ different sectors of the theory are labeled by a pair of quasimomenta, similar to the case of the BL model, so that eq.\,\eqref{apossoasop} can be understood literally as a relation between the vacuum energies in the Fateev and BL models characterized by the same ${\bf k}=(k_1,k_2)$. A major advantage of the case with all positive $a_i$ is that the general structure of the small-$R$ expansion in this regime is considerably simple compared to the case $a_3<0$. For $a_i>0$ the potential term in the Lagrangian \eqref{aposoasio} is a uniformly bounded perturbation for any finite value of the dimensionless product $\mu R$. Therefore the conformal perturbation theory yields an expansion of the form \begin{eqnarray}\label{asopisspaasop} \frac{R}{\pi}\, E_{\rm F}=-\sum_{n=0}^\infty e^{(\rm F)}_{2n}\ (\mu R)^{4n}\ \ \ \ \ \ \ (a_i>0)\ . \end{eqnarray} Here $e^{(\rm F)}_{0}=\frac{1}{6}\sum_{i=1}^3\big(1-6\,a_i\,k_i^2\,\big)$, whereas the coefficients $e^{(\rm F)}_{2n}$ for $n\geq 1$ are expressed in terms of convergent 2D Coulomb-type integrals, for example \begin{eqnarray}\label{asapasosapp} e^{(\rm F)}_2&=&2 \int\prod_{i=1}^3\frac{\mbox{d}^2 z_i}{2\pi}\ |z_1|^{-1+2 p_1+2 p_2+2p_3 }\ |z_2|^{-1-2 p_1+2 p_2-2p_3}\ |z_3|^{-1-2p_1-2p_2+2p_3}\ \\ &\times& \big|(z_1-1)(z_2-z_3)\big|^{2 a_1-2}\ \big|(z_1-z_2)(z_3-1)\big|^{2 a_2-2}\ \big|(z_1-z_3)(z_2-1)\big|^{2 a_3-2}\ ,\nonumber \end{eqnarray} where $p_i=\frac{1}{2}\, a_ik_i$. Notice that the integral diverges at $a_3\to 0^+$ and formula \,\eqref{aosiosoaaa} for the asymptotic coefficient $e_2(\delta)$ in the BL model is a regularized version of $e^{(\rm F)}_2$ with $p_3=0$. Similarly to the expression for $e_2(\delta)$, eq.\,\eqref{asapasosapp} can be brought to the form \begin{eqnarray} \label{sopaps} e^{(\rm F)}_2=\frac{1}{4\pi}\ \int\mbox{d}^2\zeta\ |\zeta|^{2 a_1-4}|1-\zeta|^{2 a_2-4}\ \big(\tau_{p_1 p_2 p_3}(\zeta)\big)^2\ , \end{eqnarray} where \begin{eqnarray}\label{oaissisaoi} \tau_{p_1p_2p_3}(\zeta)= -\frac{1}{2p_1}\sum_{\sigma=\pm} \sigma\, \sqrt{\Omega(\sigma p_1,p_2+p_3) \Omega(\sigma p_1,p_2-p_3)} \ \ \big|\chi_{\sigma p_1,p_2,p_3}(\zeta)\big|^2 \end{eqnarray} and \begin{eqnarray} \chi_{p_1p_2p_3}(\zeta)= \zeta ^{\frac{1}{2}+p_1}(1-\zeta)^{\frac{1}{2}+p_2}\ \, {}_2F_1\big({\textstyle \frac{1}{2}}+ p_1+p_2+p_3, {\textstyle \frac{1}{2}}+ p_1+p_2-p_3, 1+2 p_1;\zeta\big)\ . \end{eqnarray} The derivation follows the same steps outlined in Sec.\,\ref{sec2}; Fist of all, one should substitute the integration variables $z_2$ by $\zeta=\frac{(1-z_1)(z_2-z_3)}{(1-z_2)(z_1-z_3)}$. Then the integral over $z_1$ is performed using the identity \eqref{opasssaaspsp} where $p_1$ is substituted by $p_1+p_3$. Finally one should use the identity generalizing \eqref{ahssaajh}: \begin{eqnarray} \big(\tau_{p_1 p_2 p_3}(\zeta)\big)^2= |\zeta|^2\ \int \frac{\mbox{d}^2 z}{\pi|z|^2}\ \Big|\frac{1-\zeta z }{z (z-\zeta)}\Big|^{2p_1+2p_3}\ |z|^{4p_3}\ \frac{\tau_{p_1+p_3, p_2,0}\big(X(z)\big)}{|X(z)|}\ , \end{eqnarray} where $X(z)=\frac{(\zeta-z)(1-\zeta z)}{\zeta(1-z)^2}$. An important observation is that $\tau_{p_1p_2p_3}(\zeta)$, considered as a function on the Riemann sphere, is regular except for three points $\zeta=0,\,1,\infty$ and \begin{eqnarray}\label{liouville1} \eta_{\rm L}= -\log \tau_{p_1p_2p_3}(\zeta) \end{eqnarray} is a real solution of the Liouville equation \begin{eqnarray}\label{liouville2} \partial_\zeta\partial_{\bar \zeta}\eta_{\rm L}-\mbox{e}^{2\eta_{\rm L}}=0 \end{eqnarray} for $|p_i|<\frac{1}{2},\ \sum_i |p_i|<\frac{1}{2}$ (for details, see e.g. ref.\,\cite{Zamolodchikov:1995aa}). Notice that $\tau_{p_1p_2p_3}(\zeta)=\tau_{p_1p_2p_3}(1-\zeta)=|\zeta|^2\,\tau_{p_3p_2p_1}(\zeta^{-1})$ and therefore $\eta_{\rm L}$ satisfy the following asymptotic conditions at the punctures: \begin{eqnarray}\label{paspapsoasa} \eta_{\rm L}\to \begin{cases} (2|p_1|-1)\, \log|\zeta|+O(1)\ \ \ &{\rm as}\ \ \ \zeta\to 0\\ (2|p_2|-1)\, \log|\zeta-1|+O(1)\ \ \ &{\rm as}\ \ \ \zeta\to 1\\ (2| p_3|+1)\, \log|1/\zeta|+O(1)\ \ \ &{\rm as}\ \ \ \zeta\to\infty \end{cases} \ . \end{eqnarray} This way the result of conformal perturbation theory can be expressed in terms of solution of the Liouville equation on the three-punctured sphere $ {\mathbb S}^2/\{0,\,1,\infty\}$: \begin{eqnarray}\label{asopisspaop} \frac{R}{\pi}\ E_{\rm F}=-\frac{1}{6}\ \sum_{i=1}^3\Big(1-\frac{24}{a_i}\ p_i^2\,\Big)- \frac{1}{4\pi}\ \int\mbox{d}^2\zeta\ |P(\zeta)|^2\ \mbox{e}^{-2\eta_{\rm L}}+O(\rho^8)\ \ \ \ (a_i>0)\ , \end{eqnarray} where $\rho=\frac{1}{2}\ \mu R$ and \begin{eqnarray} P(\zeta)=\rho^2\ \zeta^{ a_1-2}(1-\zeta)^{ a_2-2}\ .\label{Pzeta-def} \end{eqnarray} In ref.\,\cite{Lukyanov:2013wra} it was conjectured that \begin{eqnarray}\label{asopisspaopsdd} \frac{R}{\pi}\ E_{\rm F}=-\frac{1}{6}\ \sum_{i=1}^3\Big(1-\frac{24}{a_i}\ p_i^2\,\Big)- \frac{1}{4\pi}\ \int\mbox{d}^2\zeta\ |P(\zeta)|^2\ \mbox{e}^{-2\eta}\ \ \ \ \ \ \ (a_i>0)\ , \end{eqnarray} where $\eta$ is a real solution of the so-called modified sinh-Gordon equation \begin{eqnarray}\label{aopsasapso} \partial_\zeta\partial_{\bar \zeta}\eta-\mbox{e}^{2\eta}+|P(\zeta)|^2\ \mbox{e}^{-2\eta}=0\ , \end{eqnarray} satisfying the the same asymptotic conditions as \eqref{paspapsoasa} (i.e., $\eta_{\rm L}$ should be substituted by $\eta$ in \eqref{paspapsoasa}). The last term in \eqref{aopsasapso} $\propto \rho^4$ and can be treated perturbatively for $|p_i|<\frac{a_i}{4}$. Therefore the small-$R$ behavior \eqref{asopisspaop} follows immediately from the exact formula \eqref{asopisspaopsdd}. One can show that the leading large-$R$ asymptotic of \eqref{asopisspaopsdd} correctly reproduces the specific bulk energy \eqref{akjsaskj} (see ref.\cite{Lukyanov:2013wra} for details). Additional arguments in support of eq.\,\eqref{asopisspaopsdd} were presented in the work \cite{Bazhanov:2013cua}. Eq.\,\eqref{asopisspaopsdd} can be transformed to a formula for the scaling function ${\mathfrak F}_{\rm F}\equiv R\,(E_{\rm F}-R{\cal E}_{\rm F})/\pi$. For this purpose, one should consider the Schwarz-Christoffel mapping \begin{eqnarray}\label{soiosa} w(\zeta)=\int\mbox{d} \zeta\ \sqrt{P(\zeta)}\ , \end{eqnarray} which maps the upper half plane $\Im m(\zeta)\geq 0$ to the triangle $(w_1,\,w_2,\,w_3)$ in the complex $w$-plane (see Fig.\,\ref{fig8}). \begin{figure} \centering \psfrag{a}{$\pi a_1$} \psfrag{b}{$\pi a_2$} \psfrag{c}{$\frac{\pi a_1}{2}$} \psfrag{d}{$\frac{\pi a_2}{2}$} \psfrag{e}{$\frac{\pi a_3}{2}$} \psfrag{w1}{$w_1$} \psfrag{w2}{$w_2$} \psfrag{w3}{$w_3$} \psfrag{bw3}{${\bar w}_3$} \includegraphics[width=4.5cm]{triangle1.eps} \caption{Triangle $(w_1, w_2, w_3)$ is a $w$-image of the upper half plane $\Im m(\zeta)>0$ under the Schwarz-Christoffel mapping \eqref{soiosa} with $a_i>0$. The point ${\bar w}_3$ is a reflection of $ w_3$ w.r.t. the straight line $(w_1,w_2)$. The domain ${\mathbb D}_{\rm F}^{(+)}$ is obtained from the 4-polygon $(w_1, w_3, w_2,{\bar w}_3)$ by the identification of the sides $[w_1,w_3]\sim[w_1,{\bar w}_3]$ and $[w_2,w_3]\sim [w_2,{\bar w}_3]$.} \label{fig8} \end{figure} The lower half plane $\Im m(\zeta)\leq 0$ is mapped into the congruent triangle $(w_1,w_2,{\bar w}_3)$. It is straightforward to show that the real function ${\hat \eta}={ \eta}-{\textstyle \frac{1}{4}}\ \log(P{\bar P})$ is a solution of the sinh-Gordon equation \begin{eqnarray}\label{sinh-eq} \partial_w\partial_{\bar w}{\hat \eta}-\mbox{e}^{2{ \hat \eta}}+ \mbox{e}^{-2{ \hat \eta}}=0 \end{eqnarray} in the open domain ${\mathbb D}_{\rm F}^{(+)}$, which is obtained by gluing together the triangles along their sides, as it shown in Fig.\,\ref{fig8}. At the singular points $w=w_i\ (i=1,2,3)$ the solution has the following asymptotic behavior: \begin{eqnarray}\label{osasail} {\hat\eta}=(2 |k_i|-1)\ \log|w-w_i|+O(1)\ \ \ \ \ \ \ \ {\rm as}\ \ \ \ \ w\to w_i\ . \end{eqnarray} In ref.\,\cite{Lukyanov:2013wra} it was shown that formula \eqref{asopisspaopsdd} implies the relation \begin{eqnarray}\label{soapsopsa} {\mathfrak F}_{\rm F}=-{ \frac{8}{\pi}}\ \int_{{\mathbb D}_{\rm F}^{(+)}}\mbox{d}^2 w\ \sinh^2({\hat\eta})+ \sum_{i=1}^3 a_i \, \big(|k_i|-{\textstyle\frac {1}{2}}\big)^2\ \ \ \ \ \ (a_i>0)\ . \end{eqnarray} Then, in the consequent paper \cite{Bazhanov:2014joa}, it was argued that \eqref {soapsopsa}, with some minor modifications, also applies to the case $a_{1,2}>0$, $a_3<0$. Namely, \begin{eqnarray}\label{soapsopsadddf} {\mathfrak F}_{\rm F}=-{ \frac{8}{\pi}}\ \int_{{\mathbb D}^{(-)}_{\rm F}}\mbox{d}^2 w\ \sinh^2({\hat\eta})+ \sum_{i=1}^2 a_i\, \big(|k_i|-{\textstyle\frac {1}{2}}\big)^2 \ \ \ \ \ \ (a_1,a_2>0,\ a_3<0)\ , \end{eqnarray} where now ${\hat \eta}$ is a solution of the sinh-Gordon equation \eqref{sinh-eq} in the domain shown in Fig.\,\ref{fig3}, satisfying the asymptotic conditions \eqref{osasail} at the vertices $w_1$ and $w_2$, and \begin{eqnarray} {\hat \eta}\to 0\ \ \ \ \ {\rm as}\ \ \ \ |w|\to\infty\ . \end{eqnarray} \begin{figure} \centering \psfrag{a}{$\pi a_1$} \psfrag{b}{$\pi a_2$} \psfrag{c}{$\frac{\pi a_1}{2}$} \psfrag{d}{$\frac{\pi a_2}{2}$} \psfrag{e}{$\frac{\pi a_3}{2}$} \psfrag{w1}{$w_1$} \psfrag{w2}{$w_2$} \psfrag{w3}{$w_3$} \psfrag{bw3}{${\bar w}_3$} \includegraphics[width=4cm]{domain1.eps} \caption{Domain ${\mathbb D}_{\rm F}^{(-)}$ -- the image of the thrice-punctured sphere for the case of Schwarz-Christoffel mapping \eqref{soiosa} with $a_{1,2}>0$, $a_3<0$.} \label{fig3} \end{figure} As $a_3\to 0^-$, the domain ${\mathbb D}_{\rm F}^{(-)}$ tends to the region ${\mathbb D}_{\rm BL}$ shown in Fig.\,\ref{fig3sgsg}. \begin{figure} \centering \psfrag{a}{$\pi a_1$} \psfrag{b}{$\pi a_2$} \psfrag{c}{$r/4$} \psfrag{w1}{$w_1$} \psfrag{w2}{$w_2$} \psfrag{w}{$w$} \includegraphics[width=4.5 cm]{domain.eps} \caption{Domain ${\mathbb D}_{\rm BL}$ -- the image of the thrice-punctured sphere for the case of Schwarz-Christoffel mapping \eqref{soiosa} with $a_1+a_2=2$. The overall size of ${\mathbb D}_{\rm BL}$ is controlled by a length of the segment $(w_1, w_2)$, which coincides with $r/4$} \label{fig3sgsg} \end{figure} With the relation \eqref{apossoasop}, this leads to the following exact formula for the scaling function ${\mathfrak F}(r,{\bf k})$ in the BL model, \begin{eqnarray}\label{soapsopsaddf} {\mathfrak F}(r,{\bf k})=-{\mathfrak f}_{\rm B}\big(2r c\big({\textstyle\frac{\delta}{2}}\big)\big) -{ \frac{8}{\pi}}\ \int_{{\mathbb D}_{\rm BL}}\mbox{d}^2 w\ \sinh^2({\hat\eta})+ \sum_{i=1}^2 a_i\, \big(|k_i|-{\textstyle\frac {1}{2}}\big)^2 \ . \end{eqnarray} As we shall see below this formula is in a perfect agreement with all perturbation theory calculations, considered in Sec.\,\ref{sec2} and Sec.\,\ref{sec3} of this paper, as well as with all other known results on the BL model, including the Bethe ansatz results of \cite{Bukhvostov:1980sn,Saleur:1998wa}. The sinh-Gordon equation \eqref{sinh-eq} is a classical integrable equation which can be treated by the inverse scattering transform method. Thus the relation \eqref{soapsopsaddf} allows one to apply this powerful method to the problem of determining the vacuum energies. The working is very similar to that for the Fateev model, considered in \cite{Bazhanov:2013cua}, where all $a_1,a_2,a_3>0$, though contains a few original details. We postpone these derivations to our future publication \cite{BLR:2016b} but present the final result here. The scaling function \eqref{soapsopsaddf} is expressed through the solution of a system of two Non-Linear Integral Equations (NLIE): \begin{eqnarray}\label{DDV} \varepsilon_\sigma(\theta)=r\,\sinh(\theta-{\rm i}\chi_\sigma)-2\pi k_\sigma+ \sum_{\sigma'=\pm} \int_{-\infty}^{\infty}\frac{\mbox{d}\theta'}{\pi}\, {G}_{\sigma\sigma'}(\theta-\theta')\ \Im m\Big[\log\big(1+e^{-{\rm i}\varepsilon_{\sigma'}(\theta'-{\rm i} 0)}\big)\Big]. \end{eqnarray} Here $\sigma=\pm$, $(\chi_+,\chi_-)=(0,\,\pi a_1/2)$ and the kernels are given by the relations \begin{eqnarray}\label{kerdef} G_{\pm\pm}(\theta)=G_{a_1}(\theta)+G_{a_2}(\theta)\ ,\ \ \ \ \ \ { G}_{\pm\mp}(\theta)={\hat G}_{a_1}(\theta)-{\hat G}_{a_2}(\theta)\ . \end{eqnarray} with \begin{eqnarray} G_a(\theta)&=& \int_{-\infty}^\infty\mbox{d}\nu \ \frac{\mbox{e}^{{\rm i}\nu\theta}\, \sinh(\frac{\pi\nu}{2}(1-a))} {2\cosh(\frac{\pi \nu}{2})\sinh(\frac{\pi \nu a}{2})}\\ {\hat G}_a(\theta)&=& \int_{-\infty}^\infty \mbox{d}\nu\ \frac{\mbox{e}^{{\rm i}\nu\theta}\, \sinh(\frac{\pi\nu }{2})} {2\cosh(\frac{\pi \nu}{2})\sinh(\frac{\pi \nu a}{2})}\ .\nonumber \end{eqnarray} Once the numerical data for $\varepsilon_\pm(\theta)$ are available, ${\mathfrak F}(r,{\bf k})$ \eqref{soapsopsaddf} can be computed by means of the relation \begin{eqnarray}\label{aooisaosa} {\mathfrak F}(r,{\bf k})=\pm \frac{ r}{\pi}\ \Im m\Big[L_+(\pm {\rm i})+ \mbox{e}^{\mp\frac{{\rm i}\pi}{2}a_1}\,L_-(\pm{\rm i})\, \Big]\ , \end{eqnarray} where \begin{equation}\label{kajsjas} L_\sigma(\nu)=\int_{-\infty}^\infty\frac{\mbox{d} \theta}{\pi}\ \mbox{e}^{-{\rm i} \nu \theta}\, \log\big(1+e^{-{\rm i}\varepsilon_\sigma(\theta-{\rm i} 0)}\big)\ . \end{equation} Notice that \eqref{aooisaosa} is valid for both choices of the sign $\pm$. Eq.\eqref{aooisaosa} can be compared against the predictions of renormalized perturbation theory in several ways. First, note that the integral equation \eqref{DDV} have a smooth limit for $\delta\to0$ (its kernel vanishes linearly in $\delta$). Using this property we have verified that the function ${\mathfrak F}_2$ in eq.\,\eqref{apospapaso}, extracted from the numerical solution of \eqref{DDV}-\eqref{kajsjas} for $k_1=k_2=0$ and $0.1\leq r\leq 5$, within nine significant digits coincides with the result of the perturbative calculations, shown with the solid line in the right panel of Fig.\,\ref{fig4}. Second, one can show that the exact formula \eqref{aooisaosa} implies the following large-$R$ asymptotics \begin{eqnarray}\label{hauasyu} &&{\mathfrak F}(r,{\bf k})={\mathfrak F}_{0}(r,{\bf k})+ {\mathfrak f}_{ \rm B}(2r)-{\mathfrak f}_{\rm B}\big(2r c\big({\textstyle\frac{\delta}{2}}\big)\big) + \frac{16 r}{\pi^2}\ \sum_{i=1}^2\int_{-\infty}^\infty \frac{\mbox{d}\nu}{2\pi}\ \\ &&\times\ \Big(c^2(k_1)\,c^2(k_2)-c^2(k_{3-i})\,\cosh^2\big({\textstyle \frac{\pi\nu}{2}}\big) \Big)\ K_{{\rm i}\nu}(r)K_{1-{\rm i}\nu}(r) \ \frac{\sinh(\frac{\pi\nu}{2}(1-a_i))} {\cosh(\frac{\pi\nu}{2})\sinh(\frac{\pi\nu}{2} a_i)}+o\big(\mbox{e}^{-2r}\big)\ ,\nonumber \end{eqnarray} where $k_1=k_++k_-,\,k_2=k_+-k_-$ and $c(x)\equiv\cos(\pi x)$. Expanding this relation to the second order in $\delta=1-a_1=a_2-1$, one finds that the result is consistent with eqs.\,\eqref{apospapaso},\,\eqref{oaaaaasisao} and \eqref{sauiuias} from Sec.\,\ref{sec3}. Third, the numerical values for ${\mathfrak F}(r,{\bf k})$ obtained from \eqref{aooisaosa} and presented in Fig.\,\ref{fig5bb} and Tab.\,\ref{tab1} on page~\pageref{tab1}, show an excellent agreement with the large-$R$ asymptotic formula \eqref{hauasyu} and also with the predictions of the conformal perturbation theory, given by \eqref{apspassp}, \eqref{aosiosaaso} and \eqref{aasopaps}. \begin{figure} \centering \includegraphics[height=7.5cm]{plot_ddv_asymptotics.eps} \caption{ The scaling function ${\mathfrak F}(r,{\bf k})$ as a function of $r= MR$ for $\delta=\frac{17}{47}=0.36\ldots,\ k_1=\frac{47}{150},\ k_2=\frac{47}{640}$. The solid line was obtained from numerical integration of \eqref{DDV},\,\eqref{aooisaosa}. The blue dashed and red dotted lines represent, respectively, the large-$r$ approximation\ \eqref{hauasyu} and the small-$r$ expansion \eqref{apspassp}. For the chosen set of parameters the latter becomes ${\mathfrak F}=-0.263322916666667-0.0719853960038915\, r^2\,\log(r)+0.092255549888030\, r^2+ 0.0000477491676\, r^4+O(r^6)$\,. The numerical values for ${\mathfrak F}$ and its asymptotics are given in Tab.~\ref{tab1} on page~\pageref{tab1}. } \label{fig5bb} \end{figure} Finally, the exact expressions \eqref{soapsopsaddf} and \eqref{aooisaosa} perfectly agree with the Bethe ansatz results, considered in the next section. \section{Bethe ansatz results\label{sec5} } As shown already in the original BL paper \cite{Bukhvostov:1980sn} the fermionic model \eqref{Lagr1} could be solved by the coordinate Bethe ansatz. In this section we review and extend their results. Within the Bethe ansatz approach the eigenvectors and eigenvalues of the Hamiltonian are parameterized through rapidities of pseudoparticles filling the bare vacuum state. These rapidities are determined by the Bethe Ansatz Equations (BAE). In the context of relativistic QFT models the number of pseudoparticles is infinite and, therefore, the related BAE require some regularization which makes that number finite. Following the BL paper \cite{Bukhvostov:1980sn} here we will impose a straightforward cutoff to the number of pseudoparticles. An alternative and in many respects more efficient lattice-type regularization is considered in our next paper \cite{BLR:2016b}. Let $N\ge2$ be an even integer. The BAE of ref.\cite{Bukhvostov:1980sn} involve two sets of unknown rapidities (called Bethe roots) $\{u_\ell\}$ and $\{\theta_{\mathscr J}\}$, containing $N$ and $2N$ variables, where \begin{equation} \ell \in\big\{\textstyle-\frac{N}{2}+1,-\frac{N}{2}+2,\ldots,\frac{N}{2}\big\}\,, \qquad {{\mathscr J}}\in\big\{-N+1,-N+2,\ldots,N\big\}\,.\label{set2} \end{equation} Throughout this section we will assume that the indices $\ell$ and ${{\mathscr J}}$ always run over the above sets of values, respectively. With a slight change of notations and some minor corrections\footnote{% The parameter $g$ in \cite{Bukhvostov:1980sn} is related to our $\delta=-g$; their integer $n$ is replaced here by $N$ (we assume that this number is even); we have restored a missing minus sign in the LHS of eqs.\,(82) of \cite{Bukhvostov:1980sn}, which corresponds to ours eq.\,\eqref{bae1}; the case of untwisted boundary conditions, considered in \cite{Bukhvostov:1980sn}, corresponds to $p_1=p_2=0$ here.} the Bethe ansatz equations of ref.\cite{Bukhvostov:1980sn} (generalized for the twisted boundary conditions \eqref{apssspps}) can be written as \begin{subequations}\label{bae} \begin{eqnarray} -1&=&\displaystyle\mbox{e}^{2\pi{\rm i}(p_1-p_2)}\, \mbox{e}^{{\rm i} {\mathcal M} R\sinh \theta_{\mathscr J}}\ \prod_\ell\ \frac{\sinh\big(\theta_{\mathscr J}-u_\ell-{1/2}\,{\rm i} \pi \delta\big)} {\sinh\big(\theta_{\mathscr J}-u_\ell+{1/2}\,{\rm i} \pi \delta\big)}\,\label{bae1} \\[.8cm] -1&=&\displaystyle\mbox{e}^{-4\pi{\rm i} p_1}\,\prod_{\ell'} \frac{\sinh\big(u_\ell-u_{\ell'}+{\rm i} \pi \delta\big)} {\sinh\big(u_\ell-u_{\ell'}-{\rm i} \pi \delta\big)}\ \prod_{\mathscr J}\ \frac{\sinh\big(u_\ell-\theta_{\mathscr J}-{1/2}\,{\rm i} \pi \delta\big)} {\sinh\big(u_\ell-\theta_{\mathscr J}+{1/2}\,{\rm i} \pi \delta\big)}\,,\label{bae2} \end{eqnarray} \end{subequations} where and the indices $\ell,\ell',{\mathscr J}$ take the integer values \eqref{set2}. The parameters $p_1$ and $p_2$ are defined by eqs.\,\eqref{apssspps}, \eqref{apssspps2} and \eqref{kasoisau}. Altogether there are $3N$ equations for $3N$ unknown $\theta$'s and $u$'s. When the cutoff is removed, $N\to\infty$,\ the number of Bethe roots becomes infinite. The parameter ${\cal M}$ is the {\em bare mass} parameter entering the coordinate Bethe ansatz calculation of \cite{Bukhvostov:1980sn} (denoted as ``$m$'' therein). Its relationship with the physical fermion mass $M$ used in the previous sections follows from the requirement that the scaling function, determined by the BAE, at large distances should vanish as $\propto\exp(-M R)$, i.e., exactly as the one in \eqref{aosioasiasoas}. As we shall see below this is achieved if one sets (see remarks after eq.\,\eqref{ceff-lat}) \begin{equation}\label{physmass} {\cal M}=M\cos\big({\textstyle\frac{\pi\delta}{2}}\big)\,. \end{equation} This relation will be assumed in what follows. For practical purposes it is useful to rewrite BAE \eqref{bae} in the logarithmic form \begin{subequations}\label{baelog} \begin{eqnarray} m_{\mathscr J}&=&{{1/2}}+p_1-p_2+\frac{{\mathcal M} R} {2\pi}\ \sinh(\theta_{\mathscr J})+ \sum_\ell\ {{\phi}}_{2\delta}(\theta_{\mathscr J}-u_\ell) \label{baelog1} \\[.8cm] {\overline m}_\ell&=&{{1/2}}-2p_1 -\sum_{\ell'} {{\phi}}_{4\delta}(u_\ell-u_{\ell'}) +\sum_{\mathscr J}{{\phi}}_{2\delta}(u_\ell-\theta_{\mathscr J})\,,\label{baelog2} \end{eqnarray} \end{subequations} where \begin{equation}\label{e-def} {{\phi}}_\alpha(\theta)=\frac{1}{2\pi{\rm i}}\log\left[\frac{\sinh \big(\frac{1}{4}\, {\rm i} \pi \alpha-\theta\big)} {\sinh\big(\frac{1}{4}\,{\rm i} \pi \alpha+\theta\big)}\right]\,, \end{equation} and the integer phases $\{m_{\mathscr J}\}$ and $\{{\overline m}_\ell\}$ play the r${\hat {\rm o}}$le of quantum numbers, which uniquely characterize solutions of the BAE. Different solutions define different eigenstates of the Hamiltonian. The energy of the corresponding state reads \begin{equation} E=-{\mathcal M} \,\sum_{\mathscr J} \cosh({\theta_{\mathscr J}})\,. \label{e-lip} \end{equation} As usual, the most difficult question in the analysis of BAE is to determine patterns of zeroes and the corresponding phase assignment in \eqref{baelog} for different states, in particular for the vacuum state. For the untwisted boundary conditions, $p_1=p_2=0$, this question was studied in \cite{Bukhvostov:1980sn}. It was shown that for small values of $|\delta|\ll 1$ the vacuum roots $\{u_\ell\}$ are real and their positions are given by an asymptotic formula \begin{equation}\label{uass} {\mathcal M} R\,\sinh u_\ell=(2\ell-1)\pi\, +O(\delta)\,\qquad\ \ \ (|\delta|\ll 1)\,, \end{equation} whereas the roots $\{\theta_{\mathscr J}\}$ split into pairs \begin{equation} \label{pairs} \theta_{2\ell-{1/2}\pm\frac{1}{2} }=u_\ell\pm\sqrt{\frac{\pi\delta}{r \cosh( u_\ell)}}+O(\delta^{\frac{3}{2}})\,, \end{equation} centered around $u$'s. This description is valid for both signs of delta. For $\delta>0$ the $\theta$-roots are real and the phases in \eqref{baelog} take consecutive integer values \begin{equation}\label{phases} m_{\mathscr J}={\mathscr J}\,,\qquad {\overline m}_\ell=\ell\ \qquad (\delta>0)\,, \end{equation} within the range defined in \eqref{set2}. For $\delta<0$ the $u$-roots remain real and retain the same phases as in \eqref{phases}, \begin{subequations}\label{phases2} \begin{equation}\label{phases2a} \qquad {\overline m}_\ell=\ell\ \qquad ( \delta<0)\,. \end{equation} The $\theta$-roots become complex and form the so-called 2-strings with a more subtle phase assignment. Near the origin $\big|\Re e (\theta_{\mathscr J})\big|<2/(\pi^2\delta)$ the phases are still consecutive, as stated in \cite{Bukhvostov:1980sn}\footnote{% The phases of complex roots are not uniquely defined. Here we adopt the convention that the functions \eqref{e-def} entering \eqref{baelog} should not have jumps under small variation of roots near their exact positions. For that reason for $\delta<0$ we replace ${ {\phi}}_{2\delta}(\theta)$ in \eqref{baelog} with $\tilde{{\phi}}_{2\delta}(\theta)$, where $$ \tilde{{\phi}}_\alpha(\theta) =\frac{1}{2\pi{\rm i}}\, \log\left[\frac{\sinh \big(\theta-\frac{1}{4} {\rm i} \pi \alpha\big)} {\sinh\big(\theta+\frac{1}{4}{\rm i} \pi \alpha\big)}\right]\,, $$ differs from \eqref{e-def} by the sign of the argument of the logarithm. As a result our 2-strings phases assignment in \eqref{phases2b} looks different, but nevertheless equivalent to the corresponding eq.\,(92) in \cite{Bukhvostov:1980sn}. } \begin{equation}\label{phases2b} m_{2\ell-{1/2}\pm\frac{1}{2}} =-\ell+1 \ \qquad (\delta<0)\,, \end{equation} however for larger $\big|\Re e(\theta_{\mathscr J})\big|$ this is no longer true and the consecutive phase segments are divided by regions of ``holes'', where the RHS of \eqref{phases2b} jumps over several integers. A general description of this pattern is unknown. \end{subequations} The arguments of \cite{Bukhvostov:1980sn} are based on the perturbation theory around the free fermion case with the untwisted boundary conditions (corresponding to $\delta=0$ and $p_1=p_2=0$) and expected to work well for sufficiently small $\delta$'s and vanishing $p$'s. We have verified this picture numerically. The arrangement of the vacuum roots for $N=16$ and $|\delta|=0.05$ is illustrated in Fig.\,\ref{figsimple}, where only a part of complex plane, containing a half of the roots is shown. For $\delta<0$ the formula \eqref{pairs} is valid for $|\ell|<2/(\pi^2\delta)$. For larger values of $\ell$ the $\theta$-roots form almost perfect 2-strings \begin{equation} \label{pairs2} \theta_{2\ell-{1/2}\pm\frac{1}{2} }=u_\ell\pm {\textstyle\frac{ 1}{2}}\, {\rm i}\pi\delta \, \big(1+O(\ell^{-1})\big)\, ,\qquad \ell\gg2/(\pi^2|\delta|)\,\qquad (\delta<0)\, . \end{equation} Our numerical analysis shows that essentially the same picture of zeroes\footnote{% When $p_1,p_2\not=0$, eqs.\,\eqref{uass} and \eqref{pairs} should be modified, but \eqref{phases}, \eqref{phases2} and \eqref{pairs2} remain intact.} holds also for small non-zero values of $p_1$ and $p_2$. In particular, the integer phases \eqref{phases} and \eqref{phases2} remains the same, as they cannot change under continuous deformations of the boundary conditions. \begin{figure} \centering \hspace*{-1cm} \includegraphics[width=16cm]{plot_deviation.eps} \caption{The arrangements of the Bethe roots solving \eqref{baelog} with $N=16$, $p_1=p_2=0$ and $|\delta|=0.05$. The (green) crosses show the roots $\{u_\ell\}$, (blue) dots show the roots $\{\theta_{\mathscr J}\}$ for $\delta>0$ and (red) asterisks show the (complex) roots $\{\theta_{\mathscr J}\}$ for $\delta<0$ (the roots $\{u_\ell\}$ remains the same). Only a part of complex plane, containing a half of the roots is shown. The dashed lines and circles illustrate the pairing of $\theta$-roots described by eqs.\,\eqref{pairs} and \eqref{pairs2}.} \label{figsimple} \end{figure} Using BAE \eqref{baelog} one can show \cite{BLR:2016b} that the vacuum energy \eqref{e-lip} diverges quadratically for large $N$ (cf. eq.\,\eqref{paopsaoaops}) \begin{equation} \frac{RE}{\pi}= \epsilon_{2}\, N^2 +{\epsilon}_0 \,r^2\,\log\big({4N/r}\big)+ O(1)\ \ \ \qquad (N\to \infty)\,,\label{qdiv} \end{equation} where \begin{equation} \epsilon_2=-(1+\delta)\,,\qquad {\epsilon}_0= -{\textstyle\frac{1}{\pi^2}}\ \cos^2\big(\textstyle{\frac{\pi\delta}{2}}\big)\,,\qquad r=M R\,. \end{equation} Then from the finite-size scaling arguments (applied in the context of the Bethe ansatz regularization of massive field theory models \cite{Destri:1994bv,Lukyanov:2011wd}) one expects that for $N\to\infty$ \ the regularized expression for the energy \begin{equation} {\mathfrak F}(r,{\bf k})=-\frac{c_{\bf{k}}}{6}+ \lim_{\scriptstyle{\begin{subarray}{c} N\to\infty\\ r-\mbox{\scriptsize{fixed}} \end{subarray}}}\left(\frac{RE }{\pi} -\epsilon_2\, N^2-{\epsilon}_0\, r^2\ \Big(\log({4N/r})+C\,\Big)\right)\,, \label{ceff-lat} \end{equation} where $c_{\bf{k}}=\sum_{i=1}^2\big(1-6 a_ik_i^2\big)$, reduces to the scaling function ${\mathfrak F}(r,{\bf k})$ for the integrable case of the QFT model \eqref{Lagr1},\,\eqref {Lagr2}. The constant $C$ is non-universal, it is determined by the requirement ${\mathfrak F}(r,{\bf k})\to0$ as $r\to \infty$. The relation \eqref{physmass} follows from the requirement that \eqref{ceff-lat} has the same large distance decay exponent as in \eqref{aosioasiasoas}. For $\delta>0$ the formula \eqref{ceff-lat} has been verified numerically. The values ${\mathfrak F}(r,{\bf k})$ obtained from \eqref{e-lip} with the solution of \eqref{baelog1}, \eqref{baelog2} and \eqref{phases} for $N=500$ display a good agreement (to within at least three decimal places) with the more accurate results obtained from the NLIE \eqref{DDV}, see Fig.\,\ref{ceff-pic}. \begin{figure} \centering \includegraphics[height=7.5cm]{plot_ddv_lipatov.eps} \caption{Dots show values of ${\mathfrak F}(r,{\bf k})$ for $\delta=17/47$, $p_1=1/10,p_2=1/20$ calculated from \eqref{e-lip} and \eqref{ceff-lat} with $N=500$ and the value of $C=0.9658605$. The continuous curve represents the results obtained from \eqref{aooisaosa} and the NLIE \eqref{DDV}.} \label{ceff-pic} \end{figure} Finally note that, as shown by Saleur \cite{Saleur:1998wa}, the BAE \eqref{baelog1}, \eqref{baelog2}, describing the $\delta>0$ vacuum state \eqref{phases} filled by the real $\theta$-roots, can be converted to a set of NLIE. After some minor corrections\footnote{% In the untwisted case $k_1=k_2=0$ our eq.\,\eqref{sddv} is equivalent to eq.\,(7) of \cite{Saleur:1998wa} where one should restore a missed factor $1/(2\pi)$ in front of the kernel $\Phi_{ij}$ therein; our eq.\,\eqref{senergy} is equivalent to eq.\,(8) of \cite{Saleur:1998wa} where one should remove an extra factor $L$ in the LHS.} these NLIE (generalized for the twisted boundary conditions \eqref{apssspps}) can be written as \begin{equation} \tilde{\varepsilon}_j(\theta)={\tilde r}_j\, \sinh(\theta) -2\pi {\tilde k}_j+ \sum_{l=1,2}\, \int_{-\infty}^{\infty}\frac{\mbox{d}\theta'}{\pi}\ \widetilde{G}_{jl}(\theta-\theta')\ \Im m\Big[ \log\big(1+\mbox{e}^{-{\rm i}\tilde{\varepsilon}_l(\theta'-{\rm i} 0)}\big)\,\Big]\, ,\label{sddv} \end{equation} where $j=1,2$, \begin{equation} {\tilde r}_1=2r\,\cos\big({\textstyle\frac{\pi\delta}{2}}\big)\,,\quad {\tilde r}_2=r\,,\quad {\tilde k}_1= k_2\,,\quad {\tilde k}_2=k_+\,,\quad k_\pm={\textstyle\frac{1}{2}}\, (k_1\pm k_2)\,. \end{equation} The kernel $\widetilde{G}_{jl}$ reads \begin{equation} \widetilde{G}_{11}(\theta)=\int_{-\infty}^\infty\,\mbox{d}\nu\ \frac{\mbox{e}^{{\rm i} \nu \theta}\,\sinh(\frac{\pi\nu a_1}{2})} {\sinh(\frac{\pi\nu a_2}{2})}\,,\qquad \widetilde{G}_{22}(\theta)=\int_{-\infty}^\infty\,\mbox{d}\nu\ \frac{\mbox{e}^{{\rm i} \nu \theta}\,\sinh^2(\frac{\pi\nu \delta}{2})} {\sinh(\frac{\pi\nu a_1}{2})\sinh(\frac{\pi\nu a_2}{2})}\,, \end{equation} and \begin{equation} \widetilde{G}_{12}(\theta)=\widetilde{G}_{21}(\theta) =\int_{-\infty}^\infty\,\mbox{d}\nu\ \frac{\mbox{e}^{{\rm i} \nu \theta}\,\sinh(\frac{\pi\nu}{2})} {\sinh(\frac{\pi\nu a_2}{2})}\ . \end{equation} Note that $\widetilde{G}_{22}(\theta)$ coincides with ${G}_{++}(\theta)$ defined in \eqref{kerdef}. With these notations the scaling function \eqref{ceff-lat} can be written as \begin{equation} {\mathfrak F}(r,{\bf k}) =\frac{1}{\pi^2} \sum_{j=1,2}\, {\tilde r}_j\, \int_{-\infty}^\infty\, \mbox{d} \theta\, \sinh(\theta) \,\Im m\Big[\log\big(1+\mbox{e}^{-{\rm i} \tilde{\varepsilon}_j(\theta-{\rm i} 0)}\big)\,\Big]\ .\label{senergy} \end{equation} \begin{table}[h] \begin{center} \begin{tabular}{| c || l| l | l|} \hline \rule{0mm}{3.6mm} $r=MR$&$\ \ \ \ \ \ \ {\mathfrak F}(r,{\bf k})$& $\ \ \ \ \ {\mathfrak F}(r,{\bf k})_{\rm UV}$& $\ \ \ \ \ {\mathfrak F}(r,{\bf k})_{\rm IR}$\\ \hline $0.1$& $-0.2607428309788$ &$-0.2607428313953$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.2$& $-0.2549983506999$ &$-0.2549983772536$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.3$& $-0.2472190685352$ &$-0.2472193690897$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.4$& $-0.2380056043350$ &$-0.2380072781157$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.5$& $-0.2277756139968$ &$-0.2277819263012$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.6$& $-0.2168482299168$ &$-0.2168668158811$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.7$& $-0.2054791982549$ &$-0.2055252929751$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.8$& $-0.1938786835672$ &$-0.1939794374587$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $0.9$& $-0.1822213607225$ &$-0.1824212140898$ & $\ \ \ \ \ \ \ \ \star\star\star$\\ $1.0$& $-0.1706526112907$ &$-0.1710196176110$ & $-0.2014349564662$\\ $1.1$& $-0.1592926194023$ &$-0.1599255303999$ & $-0.1847842117398$\\ $1.2$& $-0.1482393124726$ &$-0.1492751769847$ & $-0.1692047325819$\\ $1.3$& $-0.1375706804828$ &$-0.1391926693109$ & $-0.1547042796504$\\ $1.4$& $-0.1273467826821$ &$-0.1297919365909$ & $-0.1412670230341$\\ $1.5$& $-0.1176116153451$ &$-0.1211782235966$ & $-0.1288603066874$\\ $1.6$& $-0.1083949276224$ &$-0.1134492772872$ & $-0.1174399287778$\\ $1.7$& $-0.0997140161108$ &$-0.1066963026919$ & $-0.1069542273502$\\ $1.8$& $-0.0915754934205$ &$-0.1010047442851$ & $-0.0973472061921$\\ $1.9$& $-0.0839770063651$ &$-0.0964549329272$ & $-0.0885608931075$\\ $2.0$& $-0.0769088715072$ &$-0.0931226275591$ & $-0.0805370881568$\\ $2.4$& $-0.0535707439372$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0549560607657$\\ $2.8$& $-0.0369408994984$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0374531791217$\\ $3.2$& $-0.0253532059419$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0255388456222$\\ $3.6$& $-0.0173705422556$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0174369596054$\\ $4.0$& $-0.0118987032850$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0119222709924$\\ $4.4$& $-0.0081537173228$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0081620342332$\\ $4.8$& $-0.0055903459505$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0055932695721$\\ $5.2$& $-0.0038344869615$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0038355117142$\\ $5.6$& $-0.0026307619638$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0026311203213$\\ $6.0$& $-0.0018049899423$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0018051150200$\\ $6.4$& $-0.0012382494264$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0012382930098$\\ $6.8$& $-0.0008492127577$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0008492279219$\\ $7.2$& $-0.0005821697868$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0005821750559$\\ $7.6$& $-0.0003989064681$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0003989082967$\\ $8.0$& $-0.0002731844458$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0002731850797$\\ $8.4$& $-0.0001869773822$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0001869776017$\\ $8.8$& $-0.0001278976621$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0001278977381$\\ $9.2$& $-0.0000874328116$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0000874328378$\\ $9.6$& $-0.0000597346969$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0000597347060$\\ $10.$& $-0.0000407872423$ &$\ \ \ \ \ \ \ \ \ \star\star\star$ & $-0.0000407872454$\\ \hline \end{tabular} \end{center} \caption{Numerical data for Fig.\,\ref{fig5bb}. The first column contains numerical values of ${\mathfrak F}(r,{\bf k})$ obtained by solving the NLIE \eqref{DDV},\,\eqref{aooisaosa} for $\delta=\frac{17}{47}=0.36\ldots,\ k_1=\frac{47}{150},\ k_2=\frac{47}{640}$. The second and third columns contain the short- and large-distance asymptotics of ${\mathfrak F}(r,{\bf k})$, given by \eqref{apspassp} and \eqref{hauasyu}, respectively. } \label{tab1} \end{table} Note, that even though the equations \eqref{sddv} look totally different from \eqref{DDV} the resulting expression \eqref{senergy} for the scaling function is, in fact, exactly equivalent to \eqref{aooisaosa}. A complete proof of this equivalence is presented in our next paper \cite{BLR:2016b}. It is also worth noting that from the point of view of numerical analysis the system \eqref{DDV} displays a much faster convergence than \eqref{sddv} and, therefore, requires lesser computational resources. Moreover, the system \eqref{DDV} is well suited for small $\delta$ analysis, whereas the eq.\,\eqref{sddv} becomes singular for $\delta\to0$ (the latter fact has already been noted in \cite{Saleur:1998wa}, where the NLIE \eqref{sddv} for the untwisted case $k_\pm=0$ were originally derived). \section{Conclusion} The Bukhvostov-Lipatov (BL) model \cite{Bukhvostov:1980sn} describes weakly interacting instantons and anti-instantons in the $O(3)$ non-linear sigma model in two dimensions. In this paper we have studied various aspects of the BL model with twisted boundary conditions, using all well-established approaches to 2D massive integrable QFT, including the conformal perturbation theory (Sec.\,\ref{sec2}), the standard renormalized perturbation theory (Sec.\,\ref{sec3}) and the Bethe ansatz (Sec.\,\ref{sec5}). Moreover, in Sec.\,\ref{secnew5} we have proposed an exact formula \eqref{soapsopsaddf} for the vacuum energy of the model, expressing it via a special solution of the sinh-Gordon equation \eqref{sinh-eq} in the domain ${\mathbb D}_{\rm BL}$ (see Fig.\,\ref{fig3sgsg}). The required solution $\hat\eta(w)$ decays at $|w|\to\infty$ and obey the boundary conditions \eqref{osasail} at the singular points $w_1$ and $w_2$. The connection to the classically integrable sinh-Gordon equation is rather powerful, since it allows one to obtain the non-linear integral equations \eqref{DDV}, determining the vacuum energy in the form \eqref{aooisaosa}. We have shown that this formula perfectly matches all our perturbation theory calculations as well as the previously known coordinate Bethe ansatz results of Bukhvostov and Lipatov \cite{Bukhvostov:1980sn}, and Saleur \cite{Saleur:1998wa}. The comparisons were done both analytically (where possible) and numerically. Complete proofs and derivations of our exact results are postponed into the forthcoming publication \cite{BLR:2016b}. The main idea of that work is to connect the functional equations for connection coefficients for the auxiliary linear problem for the sinh-Gordon equation \eqref{sinh-eq} to the Bethe ansatz equations \eqref{bae}, arising from the coordinate Bethe ansatz \cite{Bukhvostov:1980sn}. This requires rather substantial works involving the particle-hole transformation and lattice-type regularization of the BAE, as well as some generalization of arguments of ref.\cite{Bazhanov:2013cua}, devoted to the Fateev model. Clearly, further study of the BL model is desirable. Indeed, almost all the considerations in this paper concerns the weak coupling regime $0<\delta<1$. However, the most interesting regime is the strong coupling regime $\delta>1$, where the BL model admits a dual description as the so-called sausage model \cite{Fateev:1992tk}. Interestingly, this model turns into the $O(3)$ NLSM, in the limit $\delta\to\infty$. This suggests that the instanton counting becomes exact in the strong coupling limit of the BL model. We intend to address this problem in the future. The description of the vacuum state energy of the BL model in terms of the classical sinh-Gordon equation can be viewed as an instance of a remarkable, albeit unusual correspondence between {\em integrable quantum field theories} and {\em integrable classical field theories} in two dimensions, which cannot be expected from the standard quantum--classical correspondence principle. In the past two decades this topic has undergone various conceptual developments, which can be traced through the works \cite{DT99b,Bazhanov:1998wj,Suzuki:2000fc,Dorey:2006an,Feigin:2007mr, Bazhanov:2003ni,Fioravanti:2004cz,Lukyanov:2010rn,Dorey:2012bx, Lukyanov:2013wra,Bazhanov:2013cua,Masoero:2015lga,Ito:2015nla}. The commonly accepted mystery of this correspondence is slightly unveiled by our conformal perturbation theory calculations in Sec.\,\ref{sec2} and Sec.\,\ref{secnew5}. Indeed eqs.\,\eqref{aasopaps} and \eqref{asopisspaop}, expressing the vacuum energy in terms of the solutions \eqref{oaissisaoi}, \eqref{liouville1} of the Liouville equation \eqref{liouville2} arise as a direct result of calculations, without any additional assumptions. It would be interesting to check whether these calculations can be generalized to other integrable QFTs where the correspondence to classical integrable equations is known. More generally, it would be very important to better understand connections of the above correspondence to mathematical structures arising in 4D gauge theories \cite{Gaiotto:2009hg,Nekrasov:2009rc,Litvinov:2013zda}, calculations of amplitudes of high energy scattering \cite{Alday:2009dv,Basso:2013vsa,Bartels:2008ce} and dualities in finite dimensional quantum-mechanical systems \cite{Mironov:2012ba}. \section*{Acknowledgment} The authors thank G.V.~Dunne, L.D.~Faddeev, A.R.~Its, L.N.~Lipatov, L.A.~Takhtajan, \newline V.O.~Tarasov, A.M.~Tsvelick and A.B.~Zamolodchikov for their interest to this work and useful remarks. \bigskip \noindent The research of SL is supported by the NSF under grant number NSF-PHY-1404056. \providecommand{\href}[2]{#2} \begingroup\raggedright
50d97946ad5e25c8d148a3d665dd68f161cea3ba
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The motivation for this note was the observation that the basic recursion relation for the modified Bessel function $K$[1], $$K_0(z)+\left(\frac{2}{z}\right)K_1(z)=K_2(z)$$ can be expressed as the symmetry with respect to $m=0$ and $n=1$ of the sum $$\sum_{k=0}^n K_{k-m-1}(z)\left(\frac{z}{2}\right)^{k+m}.\eqno(1)$$ The attempt to generalize this to arbitrary $m$ and $n$ led to our principal result\vskip .1in \noindent {\bf Theorem 1}\vskip .1in For positive integers $m$ and $n$ the expression $$(n+1)!\sum_{k=0}^n\frac{1}{k!}{m+k+1\choose{m}}K_{k-m-1}(z)\left(\frac{z}{2}\right)^{k+m}\eqno(2)$$ is symmetric with respect to $m$ and $n$.\vskip .1in \noindent This will be proven in the following section and some similar results presented in the concluding paragraph. \section{Calculation} Consider the sum $$F(n,p,q)=\frac{(n+q+1)!}{q!(q+1)!}\sum_{k=0}^p \frac{(q+k+1)!}{(k+1)!}\frac{(n+k)!}{k!}\eqno(3)$$ for $p,q,n\in {\cal{Z}}^+$. One finds that, e.g. $$F(1,p,q)=\frac{(p+q+2)!}{p!q!}$$ $$F(2,p,q)=\frac{(p+q+2)!}{p!q!}[6+2(p+q)+pq]$$ and by induction on $n$ one obtains \vskip .1in \newpage \noindent {\bf Lemma 1}\vskip .1in $$\frac{p!q!}{(p+q+2)!}F(n,p,q)$$ is a polynomial $P(p,q)=P(q,p)$ of degree $n-1$ in $p$ and $q$. Next, by interchanging the order of summation and invoking lemma 1, one has \vskip .1in \noindent {\bf Lemma 2}\vskip .1in $$G(p,q,z)=\sum_{n=0}^{\infty}\frac{1}{(n!)^2}F(n,p,q)z^n=\sum_{k=0}^p {q+k+1\choose{q}}\;_2F_1(k+1,q+2;1;z)$$ is analytic for $ |z|<1$ and symmetric with respect to $p$ and $q$. \vskip .1in Finally, noting that[2] $$\int_0^{\infty}J_0(z\sqrt{x})\;_2F_1(k+1,q+2;1;-x)dx=\frac{2^{-k-q} z^{k+q+1}}{k!(q+1)!}K_{k-q-1}(z)\eqno(4)$$ (changing $q$ to $m$ and $p$ to $n$) we have Theorem 1. For example, with $m=0$ we get the possibly new summation $$\sum_{k=0}^n\frac{1}{k!}K_{k-1}(z)(z/2)^k=\frac{1}{n!}K_{n+1}(z)(z/2)^n.\eqno(5)$$ Setting $z=-ix$ in the relation $$K_{\nu}(z)= \frac{\pi}{2}i^{\nu+1}[J_{\nu}(iz)+iY_{\nu}(iz)]\eqno(6)$$ after a small manipulation one obtains \vskip .1in \noindent {\bf Theorem 2}\vskip .1in $$(-1)^m(n+1)!\sum_{k=0}^n\frac{1}{k!}{m+k+1\choose{m}}\, J_{k-m-1}(x)(x/2)^{k+m}\eqno(7)$$ $$(-1)^m(n+1)!\sum_{k=0}^n\frac{1}{k!} {m+k+1\choose{m}}\, Y_{k-m-1}(x)(x/2)^{k+m}\eqno(8)$$ are both symmetric with respect to $m$ and $n$.\vskip .1in \newpage \noindent {\bf Corollary }\vskip .1in $$\sum_{k=0}^{n}\frac{1}{k!}\, {\cal{C}}_{k-1}(x)(x/2)^k =-\frac{1}{n!}\, {\cal{C}}_{n+1}(x)(x/2)^n\eqno(9)$$ where ${\cal{C}}=aJ +b Y$. \vskip .2in \section{Discussion} Analogous sum relations can be obtained by other means. For example, let us start with the hypergeometric summation formula[3] $$\;_3F_2(-n,1,a;3-a,n+3;-1)=\frac{(n+2)n!}{2(a-1)\Gamma(a-2)}\left[\frac{\Gamma(a-1)}{(n+1)!}+(-1)^n\Gamma(a-n-2)\right].\eqno(9)$$ But, $$\;_3F_2(-n,1,a;3-a,n+3;-1)=\frac{n!(n+2)!}{\Gamma(a-2)\Gamma(a)}\sum_{k=1}^{n+1} (-1)^{k+1}\frac{\Gamma(a-1+k)\Gamma(a-1-k)}{\Gamma(n+k)\Gamma(n-k)}.\eqno(10)$$ With $n$ replaced by $n-1$ and $a=(s+n)/2+1$, the first term of (9) is half of what would be the $k=0$ term of the sum in (10) and one has $$\sum_{k=0}^n(-1)^k(2-\delta_{k,0})\frac{\Gamma\left(\frac{s+n}{2}-k\right)\Gamma\left(\frac{s+n}{2}+k\right)}{(n-k)!(n+k)!}=\frac{(-1)^n}{n!}\Gamma\left(\frac{s+n}{2}\right)\Gamma\left(\frac{s-n}{2}\right).\eqno(11)$$ Next we take the inverse Mellin transform of both sides, noting that $$\int_{c-i\infty}^{c+i\infty}\frac{ds}{2\pi i}(2/x)^s\Gamma\left(\frac{s+n}{2}-k\right)\Gamma\left(\frac{s+n}{2}+k\right)=4x^nK_{2k}(x)\eqno(12)$$ $$\int_{c-i\infty}^{c+i\infty}\frac{ds}{2\pi i}(2/x)^s\Gamma\left(\frac{s-n}{2}\right)\Gamma\left(\frac{s+n}{2}\right)=4K_n(x).\eqno(13)$$ Consequently, $$K_n(x)=\left(\frac{x}{2}\right)^n\sum_{k=0}^{n}(-1)^{k+n}n!\frac{(2-\delta_{k,0})}{(n-k)!(n+k)!}K_{2k}(x).\eqno(14)$$ Since many integrals of the Gauss hypergeometric function are known, one of the most extensive tabulations being[2], Lemma 2 is the gateway to a myriad of unexpected finite sum identities involving various classes of special functions. We conclude by listing a small selection.. From[2] $$\int_0^{\infty}(1-e^{-t})^{\lambda-1}e^{-xt}\;_2F_1(k+1,m+2;1;ze^{-t})dt$$ $$=B(x,\lambda)\;_3F_2(k+1,m+2,x;1,x+\lambda; z) \eqno(15)$$ and one has the symmetry of $$ \sum_{k=0}^n {m+k+1\choose{m}}\;_3F_2(k+1,m+2,x;1,x+\lambda;z)\eqno(16)$$ For example for $m=0$ $$\sum_{k=0}^n\;_3F_2(k+1,2,x;1,x+\lambda;z)=(n+1)\;_2F_1(n+2,x;x+\lambda;z).\eqno(17)$$ Similarly, $$\frac{n!(n+1)!}{\Gamma(n+2-a)}\sum_{k=0}^n\frac{(m+k+1)!\Gamma(k+1-a)}{k!(k+1)!}.$$ $$=\frac{m!(m+1)!}{\Gamma(m+2-a)}\sum_{k=0}^m\frac{(n+k+1)!\Gamma(k+1-a)}{k!(k+1)!}\eqno(18)$$ $$\sum_{k=0}^n{m+k+1\choose{m}}=\sum_{k=0}^m{n+k+1\choose{n}}.\eqno(19)$$ $$\sum_{k=0}^n{m+k+1\choose{m}}\;_3F_2(k+1,m+2,a;1,a+b;z)$$ $$=\sum_{k=0}^m{n+k+1\choose{n}}\;_3F_2(k+1,n+2,a;1,a+b;z).\eqno(20)$$ $$\sum_{k=0}^n\;_3F_2(k+1,2,a;1,a+b;1)=\frac{(n+1)\Gamma(b-n-2)\Gamma(a+b)}{\Gamma(a+b-n-2)\Gamma(b)}.\eqno(21)$$ $$\sum_{k=0}^n\frac{(p+k)!}{k!}=\frac{(n+p+1)!}{(p+1)n!},\quad p=0,1,2,\cdots\eqno(22)$$ $$\sum_{k=0}^n{m+k+1\choose{m}}z^{(k+m)/2}S_{-k-m-2,k-m-1}(z)$$ $$=\sum_{k=0}^m{n+k+1\choose{n}}z^{(k+n)/2}S_{-k-n-2,k-n-1}(z).\eqno(23)$$ $$\sum_{k=0}^n{m+k+1\choose{m}}z^{(k+m)/2}W_{-k-m-2,k-m-1}(z)$$ $$=\sum_{k=0}^m{n+k+1\choose{n}}z^{(k+n)/2}W_{-k-n-2,k-n-1}(z).\eqno(24)$$ \section{References} \noindent [1] G.E. Andrews, R. Askey and R. Roy, {\it Special Functions} [Cambridge University Press, 1999] \noindent [2] A.P. Prudnikov, Yu. A.Brychkov and O.I. Marichev,{\it Integrals and Series, Vol. 3} [Gordon and Breach, NY 1986] Section 2.21.1. \noindent [3] Ibid. Section (2.4.1). \end{document}
702ece0f7a9b408770cf6d81d20d81b11c11d0e5
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} For decades, the microscopic process which causes a linear-in-temperature term in the electrical resistivity of pure ferromagnetic metals (Fe, Co and Ni) at low temperatures---which is clearly observed around liquid-helium temperatures \cite{Campbell,Volkenshtein}---has remained unclear. In this temperature region, the $T^2$ dependence of the electrical resistivity characteristic of the transition metals at low temperatures, due to the $s$-$d$ exchange interaction \cite{Kasuya2,Goodings,Mannari} and inter-electronic collisions, \cite{Baber} ceases to be the only dominant contribution. The most known intrinsic mechanism giving a linear term in the resistivity is the spin-orbit interaction between the orbits of the $4s$ conduction electrons and the spins of the nearly localized $3d$ ferromagnetic electrons. \cite{Turov,Turov2,Turov3} However, this predicts a linear coefficient which is about a thousand of times smaller than observed. \cite{Turov2,Goodings,Taylor} Despite other mechanisms have been proposed \cite{Volkenshtein} to explain this anomalous behavior, including e.g. electron-magnon scattering taking into account the electronic spin polarization, and scattering of the conduction electrons by 2D spin-wave excitations on the magnetic domain walls; it is believed, \cite{Campbell,Volkenshtein} based on a series of experiments, that the anomaly is caused by the scattering of conduction electrons by the internal magnetic induction present in the ferromagnetic metals, observed as an internal magnetoresistance effect. However, no explanation of this fact has been given so far using quantum mechanics. In this article, I propose a simple picture of an internal magnetoresistance effect in the ferromagnetic metals which predicts the correct magnitude of the linear coefficient. This is realized as the contribution to the electrical resistivity coming from electronic spin-flip transitions in the conduction band---which is Zeeman-split by the internal magnetic induction---and mediated by the isotropic spin-phonon interaction of the conduction-electron spins with the orbital \emph{contact} (hyperfine) field these electrons produce at the ionic positions. This mechanism, which accounts for the observed spin-lattice relaxation times of pure ferromagnetic metals at room temperatures, complements the existing theories of spin relaxation of conduction electrons in metals, \cite{Overhauser,Fabian,Boross,Mokrousov} which do not deal with the ferromagnetic case. The electronic spin-flip transitions introduced here portrait phonons as carriers of angular momentum. The macroscopic consequences of this were first discussed by Zhang and Niu\cite{Zhang} in their consideration of the Einstein-de Hass effect in a magnetic crystal, leading to the envisioning of \emph{chiral} phonons\cite{ZhangNiu2} as lattice modes supporting left-handed and right-handed excitations and spin\cite{Garanin,Holanda}. The direct observation of chiral phonons has been done very recently\cite{SKim,HZhu}; however, as far as I know, the role played by them in the electrical resistivity of a metal has not been considered before. \section{Description of the model} Consider a system of itinerant and interacting electrons \emph{magnetically} coupled to the localized ions of the material. The Hamiltonian of this system is \begin{equation}\label{Hked} H_e = \sum_{\bm{k} s}E_k n_{\bm{k} s}+\dfrac{1}{2}\sum_{\bm{k} \neq 0}J(\bm{k})\rho_{\bm{k}}\rho_{-\bm{k}}+H_{\tm{dd}}. \end{equation} Here, the first term represents the kinetic energy of these electrons, which have wave number $\bm{k}$ and spin index $s$, with $E_k=\hbar^2k^2/2m$ and $n_{\bm{k} s}=\hat{c}_{\bm{k} s}^{\dagger}\hat{c}_{\bm{k} s}$ being the electron number operator. The second term represents the electron-electron Coulomb interactions, with $\rho_{\bm{k}}=\sum_{\bm{l}s}\hat{c}_{\bm{l}+\bm{k},s}^{\dagger}\hat{c}_{\bm{l} s}$ being a Fourier component of the electronic density, and $J(\bm{k})$ being the Fourier transform of the Coulomb electric potential. The third term in \eqref{Hked} represents the magnetic dipole-dipole interactions between electron pairs and between electron-ion pairs. This is given by \begin{equation}\label{dd} H_{\tm{dd}}=-\sum_{r}\bm{\mu}_r\cdot\bm{B}(\bm{x}_r)=-\sum_{rq}\bm{\mu}_r\cdot \bm{D}(\bm{x}_r-\bm{x}_q)\cdot\bm{\mu}_q, \end{equation} where $\bm{\mu}_r$ is the magnetic moment of the $r^{\tm{th}}$ dipole at position $\bm{x}_r$, which interacts with the magnetic dipole field $\bm{B}(\bm{x}_r)=\sum_q \bm{D}(\bm{x}_r-\bm{x}_q)\cdot\bm{\mu}_q$ generated by the other dipoles. Here $\bm{D}(\bm{x}_r-\bm{x}_q)$ is a dyad representing the dipole kernel \begin{equation}\label{dk} \bm{D}(\bm{x}_r-\bm{x}_q) = \dfrac{3\,\hat{\bm{x}}_{rq}\hat{\bm{x}}_{rq}-\bm{1}}{|\bm{x}_r-\bm{x}_q|^3}+\dfrac{8\pi}{3}\delta(\bm{x}_r-\bm{x}_q)\bm{1}, \end{equation} with $\hat{\bm{x}}_{rq}$ a unit vector from $\bm{x}_r$ to $\bm{x}_q$---note that the second term in \eqref{dk} is necessary to account for the volume integral of the magnetic dipole field $\bm{B}(\bm{x})$ over a region containing all the dipoles.\cite{Jackson} Let me divide now the magnetic dipoles in two classes: those belonging to the ions, in which case the label is changed to $r_i$, and those belonging to itinerant electrons, in which case the label is changed to $r_e$. Furthermore, by separating the orbital and spin contributions to the magnetic moments as \begin{equation}\label{mur} \bm{\mu}_r = \bm{\mu}_r^{\tm{spin}}+\bm{\mu}_r^{\tm{orb}}=-\mu_B(2\bm{S}_r+\bm{L}_r), \end{equation} where $\bm{S}_r$ and $\bm{L}_r$ are, respectively, the spin and orbital angular momentum operators (in units of $\hbar$) corresponding to the $r^{\tm{th}}$ dipole---for simplicity assume an electronic $g$-factor of 2---and substituting \eqref{mur} into \eqref{dd}, the following contributions to the dipole-dipole Hamiltonian turn out to be sufficient for the discussion \begin{equation}\label{Hbi} \begin{split} H_{\tm{\,spin-i}}^{\tm{\,spin-e}}&=-\sum_{r_e}\bm{\mu}_{r_e}^{\tm{spin}}\cdot\bm{B}_i^{\tm{spin}}(\bm{x}_{r_e})\\ &=-\sum_{r_e,q_i}\bm{\mu}_{r_e}^{\tm{spin}}\cdot \bm{D}(\bm{x}_{r_e}-\bm{x}_{q_i})\cdot\bm{\mu}_{q_i}^{\tm{spin}}, \end{split} \end{equation} which represents the interaction of the spins of the itinerant electrons with the magnetic field created by the spins of the ions; \begin{equation}\label{Hci} \begin{split} H_{\tm{\,orb-e}}^{\tm{\,spin-e}}&=-\sum_{r_e}\bm{\mu}_{r_e}^{\tm{spin}}\cdot\bm{B}_e^{\tm{orb}}(\bm{x}_{r_e})\\ &=-\sum_{r_e,q_e}\bm{\mu}_{r_e}^{\tm{spin}}\cdot \bm{D}(\bm{x}_{r_e}-\bm{x}_{q_e})\cdot\bm{\mu}_{q_e}^{\tm{orb}}, \end{split} \end{equation} which represents the interaction of the spins of the itinerant electrons with the magnetic field created by the orbital motion of these electrons; and \begin{equation}\label{Hoo} \begin{split} H_{\tm{\,orb-i}}^{\tm{\,orb-e}}&=-\sum_{r_e}\bm{\mu}_{r_e}^{\tm{orb}}\cdot\bm{B}_i^{\tm{orb}}(\bm{x}_{r_e})\\ &=-\sum_{r_e,q_i}\bm{\mu}_{r_e}^{\tm{orb}}\cdot \bm{D}(\bm{x}_{r_e}-\bm{x}_{q_i})\cdot\bm{\mu}_{q_i}^{\tm{orb}}, \end{split} \end{equation} which represents the interaction of the orbital moments of the itinerant electrons with the magnetic field created from these same orbits. The terms that I have neglected from $H_{\tm{dd}}$ are the spin-orbit interactions $H_{\tm{\,orb-i}}^{\tm{\,spin-e}}$ and $H_{\tm{\,spin-i}}^{\tm{\,orb-e}}$ as well as $H_{\tm{\,orb-e}}^{\tm{\,orb-e}}$ and $H_{\tm{\,spin-e}}^{\tm{\,spin-e}}$, whose contributions to the electrical resistivity of metals are well known \cite{Turov2,Taylor,Overhauser} and therefore do not play an important role in the appearance of the effect here described. To proceed further, I make the standard assumption that the magnetization of the ferromagnetic body is entirely due to the unbalanced spins of the $3d$ electrons in the ions---the $4s$ electrons being the itinerants---and replace $\bm{B}_i^{\tm{spin}}(\bm{x})$ in \eqref{Hbi} with the average internal magnetic induction $\bm{B}_i(\bm{x})=4\pi \bm{M}(\bm{x})$ in the absence of an externally applied magnetic field, where $\bm{M}(\bm{x})$ is the magnetization field---this is just the magnetic constitutive relation involved in the macroscopic Maxwell's equations. At the low temperatures of interest and zero applied fields, the spatial dependence of the internal magnetic induction may be suppresed---corresponding to the neglection of the magnetic domain structure, whose effect was mentioned in the introduction to be irrelevant. As a result \begin{equation}\label{Bi} \bm{B}_{i}=4\pi \bm{M}_s, \end{equation} with $\bm{M}_s$ being the magnetization within an arbitrary magnetic domain---typically measured to be the saturation magnetization \cite{Gurevich}---here taken to point along an arbitrary direction consistent with the neglection of magnetic anisotropies. According to de Haas-van Alphen oscillation experiments, \cite{Berger,Anderson,Joseph} Eq. \eqref{Bi} is practically \emph{the} internal magnetic induction seen by conduction electrons in the ferromagnetic metals within each magnetic domain. Having approximated the effect of \eqref{Hbi} by introducing the average field \eqref{Bi} acting on the itinerant-electron spins, I turn now to \eqref{Hci}. In the magnetic field $\bm{B}_e^{\tm{orb}}(\bm{x})$ due to the electronic orbital motion, the underlying isotropy of the present model calls for the neglection of magnetic anisotropy effects related to the first term in \eqref{dk}. Consequently, taking the orbital part of \eqref{mur}, I consider only the contact (hyperfine) field due to the itinerant-electron orbital motion \begin{equation}\label{Borb} \bm{B}_e^{\tm{orb}}(\bm{x})=-\dfrac{8\pi}{3}\mu_B\sum_{q_e}\delta(\bm{x}-\bm{x}_{q_e})\,\bm{L}_{q_e}. \end{equation} It is in the treatment of this field that the main results of this paper rest upon. This is done next. \subsection{Main assumption regarding the electronic orbital motion} My claim is that the magnetic field contribution \eqref{Borb} from the orbital motion of the itinerant electrons gives a noticeable effect provided these electrons are in a state where they \emph{rigidly} move with the ions---while still being able to drift in the transport of electricity. This is expressed mathematically by writing in \eqref{Borb} \begin{equation}\label{xL} \bm{x}_{q_e}=\bm{R}_n+\bm{r}_n,\hspace{0.5cm} \bm{L}_{q_e}=\bm{L}_{n}, \end{equation} where $\bm{L}_n$ is the orbital angular momentum (in units of $\hbar$) of the $n^{\tm{th}}$ \emph{ion}, which at a given time is displaced $\bm{r}_n$ from its equilibrium position $\bm{R}_n$. For the low temperatures of interest, the displacement $\bm{r}_n$ is very small compared to interatomic distances and then \eqref{Borb} is, to leading order, \begin{equation}\label{Bc} \bm{B}_c(\bm{x})=-\dfrac{8\pi}{3}\mu_B\sum_n\delta(\bm{x}-\bm{R}_n)\,\bm{L}_n, \end{equation} where the subscript $c$ stands for ``contact'' field which, although it is entirely due to the orbital motion of the itinerant electrons, it is expressed---through my main assumption---in terms of variables related to the lattice dynamics, bringing in this way the phonons into the description. An electron rigidly moving with a given ion, as implied by \eqref{xL}---therefore giving the impression of being attached to or localized on that ion---but this electron still being able to wander around as itinerant electrons do, reminds me of the strong correlations intertwining the atomic and band behavior of electrons in transition metals, as first discussed by Hubbard \cite{Hubbard}. These correlations relate to the electron spins, as may be noticed from the following extract from Hubbard's original paper: ``As a guide one may note that Hund's first rule for atoms indicates that the intra-atomic interactions are of such a nature as to align the electron spins on an atom, so one may expect a similar effect in a metal. Suppose now ... that at some instant a given atom has its total spin in the up direction. Then the intra-atomic interactions are, according to Hund's rule, of such a nature that this atom tends to attract electrons with spin up and repel those with spin down. In this way the property of an atom on having total spin at some instant tend to be self-perpetuating ... This persistence of the atomic spin state is not due to the same up-spin electrons being localized on the atom. The actual electrons on the atom are always changing as a result of their band motion, but the electron motions are correlated in such a way as to keep a preponderance of up-spin electrons on the atom. In these circumstances (i.e. if the correlations are strong enough) one can think of the spin as being associated with the atom rather than with the electrons ...'' I believe that the aforementioned phenomenon related to the electron spin should also happen to the orbital degree of freedom of the electrons, with Hund's rule---which is just a consequence of interactions of the form $-\bm{\mu}_{r_e}^{\tm{spin}}\cdot\bm{\mu}_{q_i}^{\tm{spin}}$---replaced by the isotropic part of \eqref{Hoo}, obtained when the first term in \eqref{dk} is neglected. That is, due to interactions of the form $-\bm{\mu}_{r_e}^{\tm{orb}}\cdot\bm{\mu}_{q_i}^{\tm{orb}}$, the energy associated with the isotropic orbital state of the electrons is minimized when a given electron $r_e$ ``belongs'' to an ion $q_i$, the maximum electron-ion attraction occurring when these have the same orbital angular momentum, as assumed in \eqref{xL}---the equality of orbital angular momentum being possible since, as discussed later, the orbital angular momentum of the ions is mass-independent. In the above statement `` ... minimized when a given electron $r_e$ ``belongs'' to an ion $q_i$'' there is no restriction as to what ion should that electron belong so, under these orbital interactions, it has the freedom to wander from ion to ion as long as the correlations keep a preponderance of conduction electrons sharing the same orbital angular momentum as the ions where they might happen to be at any given moment in time. As, I will show later in this paper, this correlated state of the electrons is particularly possible at temperatures coinciding with that of liquid helium. In the following I will therefore follow the attitude, that the above conjectured orbital state of the itinerant electrons really takes place in nature, deriving its consequences, as manifested in the electrical resistivity and spin-relaxation. Should this assumption not be true, the current understanding of transport in ferromagnetic metals may be regarded as incomplete, since no quantum-mechanical explanation would then exist for the anomaly discussed in this paper. \subsection{Mean-field Hamiltonian} Wrapping up the above discussion, the present model assumes itinerant electrons which are magnetically coupled to lattice ions, from the mentioned approximations to \eqref{Hbi} and \eqref{Hci}, according to \begin{equation}\label{Hei} H_{\tm{e-i}}= -\int \bm{\mu}(\bm{x})\cdot [\bm{B}_{i}+\bm{B}_c(\bm{x})]\,d\bm{x}, \end{equation} where the magnetic moment density due to the itinerant-electron spin is written in second quantization notation as $\bm{\mu}(\bm{x})=-2\mu_B\sum_{ss',\bm{k}\kk'}\varphi_{\bm{k}'s'}^{*}(\bm{x})\varphi_{\bm{k} s}(\bm{x})\,\hat{c}_{\bm{k}' s'}^{\dagger}\bm{\sigma}_{s's}\hat{c}_{\bm{k} s}$, with $\varphi_{\bm{k} s}(\bm{x})$ being the Bloch wavefunctions and $\bm{\sigma}_{s's}$ being the Pauli spin-$1/2$ matrices. Eq. \eqref{Hei} describes the mean-field approximation to the interaction energy of the conduction-electron spins with the internal magnetic induction, corrected by the spin-orbit interaction arising from these electrons being in ``contact'' with the ions for sufficiently enough time during their band motion. It is now desirable to extract the mean-field effect of the first two terms in \eqref{Hked}. This leads us to the Stoner model for the itinerant electrons, in which the kinetic and exchange energies of these electrons are described by \cite{Izuyama2,YosidaB} \begin{equation}\label{e0} H_{\tm{kin+ex}} = \sum_{\bm{k}s}\varepsilon_{\bm{k}s}^0\,n_{\bm{k}s},\hspace{0.5cm}\varepsilon_{\bm{k}s}^0=\dfrac{\hbar^2 k^2}{2m}-s\,(\Delta_{\textrm{ex}}/2), \end{equation} where $\Delta_{\textrm{ex}}$ is the exchange spin-splitting of the conduction band and, when not a subindex, $s=\pm$ according to a conduction electron having its spin $\uparrow$ or $\downarrow$ with respect to the quantization direction, given by that parallel to the majority spins, i.e., the spin up direction ($-\bm{M}_s/M_s$) for the electrons. The total Hamiltonian that I consider---taking into account the harmonic displacements of the ions from the lattice positions---is then \begin{equation*} H= \sum_{\bm{k}s}\varepsilon_{\bm{k}s}^0\,n_{\bm{k}s}+\sum_{\bm{q}\alpha}\hbar\omega_{\bm{q}\alpha} n_{\bm{q}\alpha}+H_{\tm{e-i}}, \end{equation*} where $n_{\bm{q}\alpha}=\hat{a}_{\bm{q}\alpha}^{\dagger}\hat{a}_{\bm{q}\alpha}$ is the operator for the number of phonons with wavevector $\bm{q}$ and polarization $\alpha$. For simplicity, I use the isotropic Debye model, having spectrum with transverse $\omega_{\bm{q} 1,2}=\omega_{\bm{q} T}=c_T\,q$ and longitudinal $\omega_{\bm{q} 3}=\omega_{\bm{q} L}=c_L\,q$ excitations, with $c_T$ and $c_L$ the corresponding speeds of sound. Extracting the diagonal part of \eqref{Hei} with respect to the Fock basis---due entirely to the Zeeman splitting caused by the internal magnetic induction---the total Hamiltonian can be rewritten as \begin{equation}\label{Htot} H= \sum_{\bm{k}s}\varepsilon_{\bm{k}s}\,n_{\bm{k}s}+\sum_{\bm{q}\alpha}\hbar\omega_{\bm{q}\alpha} n_{\bm{q}\alpha}+H_{\tm{c}}, \end{equation} where I have denoted $H_c=-\int \bm{\mu}(\bm{x})\cdot \bm{B}_c(\bm{x})d\bm{x}$, and the spin-split conduction-electron bands are \begin{equation}\label{ef} \varepsilon_{\bm{k}s}=\varepsilon_{\bm{k}s}^0-s\,(\hbar\omega_c/2),\hspace{0.5cm}\hbar\omega_c=2\mu_B B_{\tm{i}}, \end{equation} where $\omega_c$ is the conduction-electron cyclotron frequency due to the internal magnetic induction $B_{\tm{i}}=4\pi M_s$ in \eqref{Bi}, with $\mu_B$ being the Bohr magneton. The term $H_c$ is nondiagonal with respect to the Fock basis and then causes electron scattering. Furthermore, since both the electronic and the phonon variables appear in this term, it plays the role of an electron-phonon interaction, which is treated here as a perturbation to which the standard perturbation theory in transport phenomena is to be applied. Before doing this, I anticipate that only the electron scattering events with spin flip contribute to the anomaly sought for---the other processes contributing as $T^3$ to the electrical resistivity---so I only concentrate on these processes. The unit of angular momentum gained or released in such transitions by the itinerant electrons then requires that phonon modes with spin 1 exist, which can give away or absorb that unit of angular momentum. I describe this modes next. \subsection{Chiral phonons} The operator $\bm{\hat{L}}_n=\bm{\hat{r}}_n\times\bm{\hat{p}}_n$ in \eqref{Bc} is the orbital angular momentum (in units of $\hbar$) of the $n^{\tm{th}}$ ion, where $\bm{\hat{r}}_n$ is the displacement of the ion from the lattice point $\bm{R}_n$, and $\bm{\hat{p}}_n=M\bm{\dot{\hat{r}}}_n$, with $M$ the ionic mass. In second quantization notation, this can be expressed \cite{Zhang} as $\bm{\hat{L}}_n=(1/N)\sum_{\bm{qq}'}\bm{\hat{S}}_{\bm{qq}'}e^{i(\bm{q}-\bm{q}')\cdot \bm{R}_n}$, where $N$ is the number of ions and \begin{equation}\label{Ln} \begin{split} \bm{\hat{S}}_{\bm{qq}'}=\dfrac{1}{2}\sum_{\alpha\alpha'}&\sqrt{\dfrac{\omega_{\bm{q}'\alpha'}}{\omega_{\bm{q}\alpha}}}\,(\hat{a}_{\bm{q}\alpha}+\hat{a}_{-\bm{q}\alpha}^{\dagger})\,\bm{S}_{\bm{q}\q'(\alpha\alpha')}\\ &\ \times(\hat{a}_{-\bm{q}'\alpha'}-\hat{a}_{\bm{q}'\alpha'}^{\dagger}), \end{split} \end{equation} with $\hat{a}_{\bm{q}\alpha}^{\dagger}$ ($\hat{a}_{\bm{q}\alpha}$) being the creation (annihilation) operator of a phonon with wave vector $\bm{q}$, angular frequency $\omega_{\bm{q}\alpha}$, and \emph{linear} polarization in the direction of the real unit vector $\bm{e}_{\bm{q}\alpha}$, and $\bm{S}_{\bm{q}\q'(\alpha\alpha')}=-i\,(\hat{\bm{e}}_{\bm{q}\alpha}\times\hat{\bm{e}}_{\bm{q}'\alpha'})$. Note the independence of \eqref{Ln} on the ionic mass, making plausible referring to $\bm{\hat{L}}_n$ as the orbital angular momentum of a conduction electron rigidly moving with the $n^{\tm{th}}$ ion. With the notation $\bm{S}_{\bm{q}(\alpha\alpha')}=\bm{S}_{\bm{q}\q(\alpha\alpha')}$, we can write the total angular momentum of the phonon system as $\bm{\hat{S}}_{\textrm{ph}}\equiv\sum_n\bm{\hat{L}}_n=\sum_{\bm{q}}\bm{\hat{S}}_{\bm{q},\textrm{ph}}$, where the angular momentum operator of a single phonon, $\bm{\hat{S}}_{\bm{q},\textrm{ph}}=\bm{\hat{S}}_{\bm{q}\q}$, with wave vector $\bm{q}$ is given by \begin{equation} \bm{\hat{S}}_{\bm{q},\textrm{ph}}=\dfrac{1}{2}\sum_{\alpha\alpha'}\sqrt{\dfrac{\omega_{\bm{q}\alpha'}}{\omega_{\bm{q}\alpha}}}(\hat{a}_{\bm{q}\alpha}+\hat{a}_{-\bm{q}\alpha}^{\dagger})\,\bm{S}_{\bm{q}(\alpha\alpha')}\,(\hat{a}_{-\bm{q}\alpha'}-\hat{a}_{\bm{q}\alpha'}^{\dagger}). \end{equation} The sum in this expression can be seen as a matrix multiplication. In fact, with $\bm{e}_{\bm{q} 1}=(1,0,0)$, $\bm{e}_{\bm{q} 2}=(0,1,0)$, and $\bm{e}_{\bm{q} 3}=\bm{q}/q=(0,0,1)$, the matrices $\bm{S}_{\bm{q}}$ constitute a representation of the infinitesimal generators of rotations in three dimensions, with $\bm{S}_{\bm{q}}^{2}=\sum_aS_{\bm{q}}^{a}S_{\bm{q}}^{a}=2\cdot\bm{1}$, and the commutation relations $\left[ S_{\bm{q}}^{a},S_{\bm{q}}^{b}\,\right]=i\,\sum_c\epsilon_{abc}S_{\bm{q}}^{c}$, and $\left[ S_{\bm{q}}^{a},\bm{S}_{\bm{q}}^{2}\,\right]=0$. As usual in the theory of angular momentum, it is convenient to work in a representation which simultaneously diagonalizes $S_{\bm{q}}^{3}$ and $\bm{S}_{\bm{q}}^{2}$. This is done by changing basis from $\left\lbrace \bm{e}_{\bm{q}\alpha}\right\rbrace$ to the helicity basis $\left\lbrace \bm{\epsilon}_{\bm{q}\sigma}\right\rbrace$ defined by the circular $\bm{\epsilon}_{\bm{q}\pm}=(1/\sqrt{2})(\bm{e}_{\bm{q} 2}\mp i\bm{e}_{\bm{q} 1})$ and longitudinal $\bm{\epsilon}_{\q0}=i\bm{e}_{\bm{q} 3}$ polarization vectors, where $\bm{\epsilon}_{-\bm{q}\sigma}^{*}=\bm{\epsilon}_{\bm{q}\sigma}$. The corresponding unitary transformation can be shown to map the matrices $\bm{S}_{\bm{q}(\alpha\alpha')}$ to $\bm{\mathcal{S}}_{\bm{q}(\sigma\sigma')}=-i\,(\hat{\bm{\epsilon}}_{\bm{q}\sigma}^{*}\times\hat{\bm{\epsilon}}_{\bm{q}\sigma'})$, which are the spin matrices \begin{equation} \mathcal{S}_{\bm{q}}^3 = \begin{pmatrix} 1 & 0 & 0 \\ 0 & 0 & 0\\ 0 & 0 & -1 \end{pmatrix} ,\hspace{0.1cm} \mathcal{S}_{\bm{q}}^{+} = \begin{pmatrix} 0 & \sqrt{2} & 0 \\ 0 & 0 & \sqrt{2}\\ 0 & 0 & 0 \end{pmatrix} , \hspace{0.1cm} \mathcal{S}_{\bm{q}}^{-} = (\mathcal{S}_{\bm{q}}^{+})^{\dagger}. \end{equation} for a spin-1 particle: the \emph{chiral} phonon. In the new representation, the chiral-phonon operators are $\hat{b}_{\bm{q}\pm}=(1/\sqrt{2})(\hat{a}_{\bm{q} 2}\mp i\hat{a}_{\bm{q} 1})$, which annihilate phonons with $\pm$ helicities (circular polarization); and $\hat{b}_{\q0}=i\hat{a}_{\bm{q} 3}$, which annihilates zero helicity (or longitudinal) modes. We have, for instance, $\hat{S}_{\bm{q},\textrm{ph}}^3=\hat{b}_{\bm{q}+}^{\dagger}\hat{b}_{\bm{q}+}-\hat{b}_{\bm{q}-}^{\dagger}\hat{b}_{\bm{q}-}$. These phonons have definite spin projections $\pm1,0$ along the propagation direction and, making an analogy with circularly polarized light, the displacement field of the ions must rotate perpendicular to the propagation direction in a circularly polarized elastic wave. \cite{Garanin} It is important to emphasize---and this has long been known\cite{ADlevine} by studying, in the Lagrangian formalism, the transformation properties under rotation of the quantized phonon field---that the spin of the phonon is well-defined in isotropic media and has a value of 1. In a real crystal, it is well-defined only along certain directions of propagation, such as along the lattice vectors of a cubic crystal or along trigonal axes. Since I am using the isotropic Debye model for the phonons in the ferromagnetic metals, phonons with spin 1 are therefore available in all directions for electron scattering. The magnetic nature of the chiral phonons is not displayed in an isotropic medium in the absence of a external magnetic field---as assumed in the present case---since the transverse bands of the phonon spectrum remain degenerate. For this reason, the itinerant electrons do not experience a further shift of their energy bands due to the contact interaction discussed here, since its diagonal matrix elements in the space of the electrons, being proportional to the thermal-averaged total phonon angular momentum $\bar{S}_{\textrm{ph}}^3=\bar{N}_{+}-\bar{N}_{-}$ along the quantization direction of the electron spin, \emph{vanish}, due to the same thermal number $\bar{N}_{\pm}$ of right-handed and left-handed phonons caused by the degeneracy. Therefore, for a direct observation of chiral phonons, the degeneracy of the transverse phonon bands has to be lifted either by the application of an external magnetic field\cite{Holz} or by a spatial symmetry breaking. The latter has been achieved in the recent experiments observing chiral phonons, in a system with broken inversion symmetry of the crystal lattice\cite{HZhu} or with disorder\cite{SKim}. In the present case, however, the circular phonons only play the subsidiary role of being ``reservoirs'' of angular momentum for the electronic spin-flip transitions. Returning to our technical discussion, I use the convention $\mathcal{PT}\,\hat{b}_{\bm{q}\sigma}=\hat{b}_{-\bm{q}\sigma}^{*}=\hat{b}_{\bm{q}\sigma}$, inherited from $\bm{\epsilon}_{-\bm{q}\sigma}^{*}=\bm{\epsilon}_{\bm{q}\sigma}$, which states that the wavefunctions of the crystal lattice vibrations are even under the $\mathcal{PT}$ transformation (complex conjugation $+$ parity). When the change to the helicity basis is performed in the tensor product spin space corresponding to $\bm{q}$ and $\bm{q}'$, I transform $\bm{S}_{\bm{q}\q'}\rightarrow\bm{\mathcal{S}}_{\bm{q}}\otimes \bm{1}_{\bm{q}'}=\bm{1}_{\bm{q}}\otimes\bm{\mathcal{S}}_{\bm{q}'}$ in \eqref{Ln}, since the image of $\bm{S}_{\bm{q}\q'}$ under the transformation must behave as an angular momentum upon rotations and, by definition, must reduce to $\bm{\mathcal{S}}_{\bm{q}}$ when $\bm{q}=\bm{q}'$. The ladder operators corresponding to \eqref{Ln}, in the helicity basis, then read \begin{equation}\label{Sqm} \begin{split} \hat{S}_{\bm{q}\q'}^{-}= &\,\dfrac{1}{2}\Biggl[\sqrt{\dfrac{2\omega_{\bm{q}'L}}{\omega_{\bm{q} T}} }(\hat{b}_{\bm{q}-}^{\dagger}+\hat{b}_{\bm{q}+})(\hat{b}_{\bm{q}'0}^{\dagger}+\hat{b}_{\bm{q}'0})\\ &\ - \sqrt{\dfrac{2\omega_{\bm{q}'T}}{\omega_{\bm{q} L}} }(\hat{b}_{\q0}^{\dagger}-\hat{b}_{\q0})(\hat{b}_{\bm{q}'-}^{\dagger}-\hat{b}_{\bm{q}'+})\Biggr], \end{split} \end{equation} and $\hat{S}_{\bm{q}\q'}^{+}=(\hat{S}_{\bm{q}\q'}^{-})^{\dagger}$ which enter the terms causing electron spin-flip scattering. By using the Bloch theorem for $\varphi_{\bm{k} s}(\bm{x})=e^{i\bm{k} \cdot \bm{x}}u_{\bm{k} s}(\bm{x})$, that is, $u_{\bm{k} s}(\bm{x+\bm{R}_n})=u_{\bm{k} s}(\bm{x})$ to express $u_{\bm{k} s}(\bm{\bm{R}_n})=u_{\bm{k} s}(0)=\varphi_{\bm{k} s}(0)$, and neglecting umklapp processes, the relevant terms in $H_c$ are mainly \begin{equation}\label{Hc} \frac{1}{2}\sum_{\bm{k} \bm{k}', \bm{q} \bm{q}'} A_{\bm{k} \downarrow,\bm{k}'\uparrow}\,\delta_{\Delta\bm{q},-\Delta \bm{k}}\;\hat{c}_{\bm{k}'\uparrow}^{\dagger}\hat{c}_{\bm{k} \downarrow}\;\hat{S}_{\bm{qq}'}^{-}, \end{equation} where $\Delta \bm{k}=\bm{k}'-\bm{k}$, $\Delta\bm{q}=\bm{q}'-\bm{q}$, and the matrix elements $A_{\bm{k} s,\bm{k}' s'}=-\textstyle{\frac{16\pi}{3}}\mu_B^2 \varphi_{\bm{k}' s'}^{*}(0)\varphi_{\bm{k} s}(0)$ give the strength of the resulting electron-phonon interaction. These terms account for processes where spin-$\downarrow$ electrons transition to the majority-spin conduction band. The reverse processes can be shown to be exponentially suppressed at low temperatures. \section{Results and discussion} The transition probability rate for a conduction electron undergoing a collision from the state $\bm{k}\hspace{-0.1cm}\downarrow$ to an unoccupied state $\bm{k}'\hspace{-0.1cm}\uparrow$ is obtained from \eqref{Sqm} and \eqref{Hc}, in the leading order of perturbation theory, using Fermi's golden rule. \cite{Solano2} Since for temperatures $T\rightarrow0$ the average occupation of a phonon mode $\bar{N}_{\bm{q}\sigma}$ is exponentially small \emph{except} for the lowest energy mode supported by the crystal lattice---with wavevector magnitude $q_{\tm{min}}=2\pi/L=0^{+}$, where $L$ is the largest linear size of the sample---the processes happening more frequently at very low temperatures are those in which these lowest energy modes are involved, for which conservation of energy reads \begin{equation}\label{cE} \hbar c_T |\Delta\bm{k}| -|\Delta \varepsilon_{\bm{k}}|=0, \end{equation} up to a negligible term of $O(1/L)$. Therefore, neglecting contributions from exponentially small terms as well as the temperature-independent term (discussed later), the transition probability rate per electron, for temperatures $T\rightarrow0$, is \begin{equation}\label{w} \begin{split} w_{\bm{k}\downarrow\rightarrow\bm{k}'\uparrow}&=\dfrac{3\pi}{16}\dfrac{|A_{\bm{k} \downarrow,\bm{k}' \uparrow}|^2|\Delta\epsilon_{\bm{k}}|^2}{(k_B\Theta)^4}\dfrac{k_F}{|\Delta \bm{k}|}\dfrac{c_s^4}{c_L^2c_T^2}\left(\dfrac{2}{z}\right)^{4/3}\\ &\ \times \dfrac{(k_BT)}{\hbar}\ln\left(\dfrac{k_BT}{\hbar c_L q_{\tm{min}}}\right)+O(T^2), \end{split} \end{equation} where $\Theta$ is the Debye temperature, $k_F$ is the Fermi wave vector, $z=N_e/N$ is the number of conduction electrons per ion and $c_s$ is average speed of sound, determined from $3/c_s^3=2/c_T^3+1/c_L^3$. The dominant (first) term in \eqref{w} comes from processes wherein a lowest-energy longitudinal mode of the lattice is absorbed and a circular mode is \emph{spontaneously} emitted to satisfy the conservation laws. The most probable transitions per unit time are obtained from \eqref{w} when $|\Delta\epsilon_{\bm{k}}|$ is maximum and $|\Delta \bm{k}|$ is minimum, corresponding respectively to $|\Delta\epsilon_{\bm{k}}|=|E_{F\uparrow}-E_{F\downarrow}|=\hbar\omega_c$ and $|\Delta \bm{k}|=k_{F\uparrow}-k_{F\downarrow}\equiv\Delta k_F$. The most probable processes then satisfy \eqref{cE} in the form $\hbar \omega_c=\hbar c_T \Delta k_F$ or, equivalently, by defining the characteristic temperature $T_{\tm{i}}=\hbar \omega_c/k_B$ associated with the internal magnetic induction and $T_{\tm{res}}=\hbar c_T \Delta k_F/k_B$ as that associated with the aforementioned spontaneous excitation of circular-phonon modes, \eqref{cE} may be written in the form $T_{\tm{i}}=T_{\tm{res}}$. By writing \cite{Yosida} $\Delta k_{F}/k_F=\Delta_{\tm{ex}}/2E_F$, we can express $T_{\tm{res}}$ as \begin{equation}\label{Ts} T_{\tm{res}} = \dfrac{c_T}{c_s}(z/2)^{1/3}(\Delta_{\tm{ex}}/2E_F)\,\Theta, \end{equation} whose agreement with $T_{\tm{i}}$, as shown in table \ref{t1}, is remarkably good for the pure ferromagnetic metals, with the characteristic temperatures involved being around liquid-helium temperatures. \begin{table*}[t] \begin{center} \begin{tabular}{cccccccccccccc} \ \vspace{-3mm}\\\rowcolor{snow} Metal & $z$ & $n$ $(10^{28}$m$^{-3})$ & $c_L$ $(m/s)$ & $T_{\tm{res}}$ ($^{\circ}$K) & $T_{\tm{i}}$ ($^{\circ}$K) & $\Theta$ ($^{\circ}$K) & $H_a$(MG) &$\xi$ & $\delta\Omega_F$ & \multicolumn{2}{c}{$\gamma$ ($10^{-12}\Omega$ cm/$^{\circ}$K)}& \multicolumn{2}{c}{$\tau_{\downarrow\uparrow}$ (ns)} \\ \rowcolor{snow} & & & $c_T=c_L/\sqrt{2}$& & & & & & & theory & exp. & theory & exp. \\\rowcolor{lgray} Fe& 0.22 & 8.5 & 5960& 4.0 & 2.9 &470 & 1.95 &0.35& 0.5&16.8&11$-$49.3& 0.04&0.03$-$0.1\\ \rowcolor{gwhite} Co & 0.7 & 9.1 & 4720& 5.4 & 2.4 &445 & 2.2&0.21&8.0 & 6.4&3$-$32& 0.03&$-$\\\rowcolor{lgray} Ni & 0.6& 9.2 &6040& 1.0 & 0.8 &450 & 2.4 &0.55& 8.0&7.9&5.8$-$16& 0.03&0.02$-$0.1 \\%\hline \end{tabular} \caption{Predicted against observed values of the linear coefficient $\gamma$ (from \eqref{gamma}) in the electrical resistivity of the ferromagnetic metals at low temperatures, and the spin-lattice relaxation time $\tau_{\downarrow\uparrow}$ at $T=300^{\circ}$K (from \eqref{utau}). See the text for a discussion of the appropriate values of the other quantities presented here. \label{t1}} \end{center} \end{table*} In order to estimate the spin-lattice relaxation time corresponding to the transition rates in \eqref{w}, it is necessary to account for less probable transitions involving conduction-electron energy losses and crystal momentum transfers in the ranges $0\le|\Delta\epsilon_{\bm{k}}|<\hbar\omega_c$ and $\Delta k_F<|\Delta \bm{k}|\le 2k_F $, respectively. Since only order of magnitudes are of interest here, no more sophistication than averaging \eqref{w} over the solid angle is required. That is $1/\tau_{\downarrow\uparrow}\equiv4\pi \delta\Omega_F\,\tm{mean}(\avg{w_{\bm{k}\downarrow\rightarrow\bm{k}'\uparrow}^{\tm{max}}},\avg{w_{\bm{k}\downarrow\rightarrow\bm{k}'\uparrow}^{\tm{min}}})=2\pi\delta\Omega_F\,\avg{w_{\bm{k}_F\downarrow\rightarrow\bm{k}_F\uparrow}}$, where the average $\avg{\cdot}$ is taken over the solid angle between $\bm{k}$ and $\bm{k}'$. The factor $\delta\Omega_F$ is included to account for the realistic reduction ($\delta\Omega_F<1$) or increase ($\delta\Omega_F>1$) in effective solid angle from that subtended by a spherical Fermi surface ($\delta\Omega_F=1$). Since $\avg{1/|\Delta \bm{k}|}=1/\tm{max}(k,k')$ in \eqref{w}, the factor $\delta\Omega_F$ is understood as referring to the majority-spin Fermi surface shape. With this, the spin-lattice relaxation time is \begin{equation}\label{utau} \begin{split} \dfrac{1}{\tau_{\downarrow\uparrow}}&=\dfrac{3\pi^2}{8}\delta\Omega_F\dfrac{\avg{|A_{k_F\downarrow,k_F\uparrow}|^2}(\hbar\omega_c)^2}{(k_B\Theta)^4}\left(\dfrac{2}{z}\right)^{4/3}\dfrac{c_s^4}{c_L^2c_T^2}\\ &\ \times \dfrac{(k_BT)}{\hbar}\ln\left(\dfrac{k_BT}{\hbar c_L q_{\tm{min}}}\right)+O(T^2). \end{split} \end{equation} Due to the isotropy of the model, the contribution of the discussed scattering mechanism to the ideal electrical resistivity (excluding the residual term at $T=0$, discussed in the Appendix) of the ferromagnetic metals at low temperatures can be calculated from the Drude-Mott formula \cite{Mott3} \begin{equation}\label{rhop} \rho(T)=\dfrac{m}{e^2z\, n}\dfrac{1}{\tau_{\uparrow\downarrow}}=\gamma\,T + \eta\left(\dfrac{T}{\Theta}\right)\ln\left(\dfrac{T}{\Theta}\right)+O(T^2), \end{equation} with $n=N/V$ the atomic density and the coefficients $\gamma$ and $\eta$ readily obtained from \eqref{utau}, the first term being the dominant term around liquid-helium temperatures since $T/\Theta$ is then much smaller than $c_sq_D/c_Lq_{\tm{min}}$, with $q_D$ being he Debye cutoff---I use a typical length $L= 35$ mm for the largest dimension of the samples in the electrical resistivity measurements. \cite{Semenenko} \subsection{Numerical estimates} In order to obtain numerical estimates, let me discuss the values of the physical quantities involved in \eqref{utau} and \eqref{rhop}, which are described, in the following, in the order shown in Table \ref{t1} (from left to right). The number of conduction electrons per ion $z$ consistent with the observed values of the saturation magnetization, are taken from ref. [\onlinecite{Mott}]; the atomic densities $n$ as well as the Fermi energies $E_F\simeq 7$ eV for the three elements are taken from ref. [\onlinecite{Goodings}]; the longitudinal speeds of sound $c_L$ are taken from ref. [\onlinecite{Lide}], and I use $c_T=c_L/\sqrt{2}$, which is valid for elastically isotropic bodies (coming from the corresponding relation \cite{Fuchs} $c_{11}=2\,c_{44}$ for the elastic constants, with $c_{12}=0$), for consistency with the isotropic model. The exchange splittings of the $4s$ band $\Delta_{\tm{ex}}$ are estimated as the energy difference between the bottoms of the majority-spin and minority-spin bands at the $\Gamma$ point (center of Brillouin zone) extracted from the band-structure calculations of ref. [\onlinecite{Nautiyal}] for Fe, ref. [\onlinecite{Singal}] for Co and ref. [\onlinecite{Wang}] for Ni, with the approximate values 0.13, 0.27 and 0.05 eV, respectively; from these the temperatures $T_{\tm{res}}$ are obtained from Eq. \eqref{Ts}. The temperatures $T_{\tm{i}}$ have been estimated from the well-known\cite{Berger} magnitudes of the internal magnetic induction $B_{\textrm{i}}$ of 22, 18, and 6 kG for Fe, Co and Ni, respectively. The low temperature limits of the Debye temperature $\Theta$ are taken from ref. [\onlinecite{Kittel}], and the outermost $s$-electron hyperfine fields in the free atom $H_a$ from ref. [\onlinecite{Watson}]. It is convenient to write $\avg{|A_{k_F\uparrow,k_F\downarrow}|}=2\mu_B \xi H_a$, with the Knight ratio \cite{Townes} defined as $\xi=\langle|\bar{\varphi}_{k_F}(0)|^2\rangle/|\psi_a(0)|^2$, where $\langle|\bar{\varphi}_{k_F}(0)|^2\rangle$ is the average probability density at the nucleus of electronic states on the Fermi surface, and $\psi_a(0)$ the wavefunction at the nucleus of the outermost $s$ electron in the free atom which, as it is known, produces a hyperfine field of magnitude $H_a=(8\pi/3)\mu_B|\psi_a(0)|^2$. Note that, in writing $\avg{|A_{k_F\uparrow,k_F\downarrow}|}$ in this form, I have assumed that $\varphi_{k_F\uparrow}(0)$ and $\varphi_{k_F\downarrow}(0)$ deviate only slightly \cite{Song,Duff2} from their arithmetic mean $\bar{\varphi}_{k_F}(0)$. The Knight ratio accounts for any deviation in hyperfine coupling from atomic behavior and may deviate from $\xi=1$ for two reasons: \cite{Bennett2} (i) the reduction of $s$-character of the wavefunctions at the Fermi surface and (ii) the fact that the wavefunctions in a metal are normalized within volumes smaller than in the free atom, causing the conduction electron density in the metal greater than in the free atom. For ``simple'' metals the reduction tends to predominate over the normalization effect, with $\xi$ taking values between 0.1 and 0.8. We take both effects into account, in their simplest form, by taking $\xi=(1/N_{3d4s})a^3/(4\pi r^3/3)$, with $N_{3d4s}$ the number of electrons per atom in the $3d$ and $4s$ subshells, with values 8, 9, and 10 for Fe, Co, and Ni, respectively, $a$ is the lattice constant with values 2.87, 2.51 and 3.52 \AA{} for Fe, Co and Ni, respectively \cite{Kittel}, and $r$ is the atomic radius with values 1.26, 1.25 and 1.24 \AA{} for Fe, Co and Ni, respectively \cite{Kittel}. Since $|\bar{\varphi}|^2$ and $|\psi|^2$ have units of 1/volume, the normalization effect is then taken into account by the ratio of unit cell to atomic volumes $a^3/(4\pi r^3/3)$, and the reduction in $s$-character by the factor $1/N_{3d4s}$, the latter because of the $N_{3d4s}$ electrons in the $3d$ and $4s$ subshells of the free atom, only a fraction $z$ remains in the $4s$ band in the solid, \cite{Mott} and we need to count all the electrons which can make this donation to the $4s$ band. With this simple rule, the values of $\xi$ shown in Table \ref{t1} are in the range of those for ``simple'' metals. Moreover, the $4s$ contribution to the effective hyperfine fields $\xi H_a$ obtained in this way are in agreement (except for Ni, for which it is an order of magnitude higher) with the theoretical estimates of ref. [\onlinecite{Muto}] aimed at explaining the observed hyperfine fields in the $3d$ ferromagnetic metals from M\"{o}ssbauer and NMR experiments. Finally, we need to estimate the factors $\delta \Omega_F$. For Ni (fcc structure), the sheet of the Fermi surface coming from the $4s$ majority-spin band is, as in copper, spherical-like, with necks touching the Brillouin zone faces near the $L$ points \cite{Gold}. A \emph{rough} picture of this surface may then be drawn by considering the union of a major sphere of radius $k_{F\uparrow}$, centered at the $\Gamma$ point, with little spheres with such radii as to touch the major sphere and the $L$ points, i.e. the centers of the hexagonal faces of the truncated octahedron constituting the Brillouin zone of a fcc structure. Since there are 8 such faces, we have 8 little spheres and then we need to multiply the value $\delta\Omega_F=1$ corresponding to the major sphere by 8, as shown in Table \ref{t1}. For Co (hcp structure), similar results apply due to the correspondence between energy bands (and Fermi surfaces) of the fcc and hcp structures when the hcp double zone is rotated until its [0001] axis coincides with the [111] axis of the fcc zone \cite{Gold}. For Fe (bcc structure), the situation is much more complex since the Fermi surface corresponding to the $4s$-band breaks up into small regions of electron and hole pockets \cite{Gold,Stearns}. Nevertheless, I consider this as a reduction in effective solid angle and take for $\delta\Omega_F$ the neutral value shown in Table \ref{t1}, having in mind that a more detailed investigation should not considerably change the overall result. \subsection{Comparison with experiments} Having described the magnitudes of the relevant quantities defining the linear coefficient \begin{equation}\label{gamma} \begin{split} \gamma&=\dfrac{3\pi^2}{4}\delta\Omega_F\dfrac{(\mu_B \xi H_a)^2(\hbar\omega_c)^2}{(k_B\Theta)^4}\left(\dfrac{2}{z}\right)^{7/3}\dfrac{c_s^4}{c_L^2c_T^2}\\ &\ \hspace{1cm}\times \dfrac{mk_B}{e^2n\hbar}\ln\left(\dfrac{k_B\Theta}{\hbar c_L q_{\tm{min}}}\right), \end{split} \end{equation} in \eqref{rhop}, I show the theoretical estimates from the present model in Table \ref{t1}, as well as the predicted spin-lattice relaxation time $\tau_{\downarrow\uparrow}$ at $T=300^{\circ}$K from \eqref{utau}. These are compared with the experimental values. The observed values of the coefficient $\gamma$ from electrical resistivity measurements are taken from the review article of\; \textcite{Volkenshtein}. The ranges shown represent the minimum and maximum values of measurements of the linear coefficient by multiple authors under similar experimental conditions. In the three materials, the predicted values then agree with the measurements within the statistical error. As for the spin-lattice relaxation time at 300$^{\circ}$K, from the line width of the ferromagnetic resonance signal a value of about $0.1$ ns is found to fit the data best \cite{Bloembergen2,Holmes} for Fe and Ni, although magneto-optic Kerr effect measurements \cite{Agranat} reveal a spin-lattice relaxation time between 20 ps and 40 ns for Ni, and spin-polarized photoemission experiments \cite{Vaterlaus} reveal this to be greater than 30 ps for Fe. These measurements are then shown as ranges in Table \ref{t1}, within which the theoretical predictions fall again. I was not able to find measurements of the spin-lattice relaxation time of hcp cobalt but, from the theoretical estimates, this is believed to be nearly the same as that of iron and nickel. The agreement of the theoretical predictions with the observed values from independent experiments and for different materials then indicate that the main assumption of this paper about the correlated motion of the itinerant electrons is most likely to be true. This, however, should be subjected to further studies. \begin{acknowledgments} I would like to acknowledge the support from the Fulbright-Colciencias Fellowship, the Columbia GSAS Faculty Fellowship, and express my gratitude to Prof. Andrew Millis for valuable discussions. \end{acknowledgments}
e618a3994cacdcd915be6c81aa7aff50b5abc60a
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Handling batch insertions} In this section, we study the dynamic DFS tree problem in the batch insertion setting. The goal of this section is to prove Theorem \ref{batch-ins}. Our algorithm basically follows the same framework for fully dynamic DFS proposed in \cite{baswana2016dynamic}. Since we are only interested in the dynamic DFS tree problem in the batch insertion setting, the algorithms \textsf{BatchInsert} and \textsf{DFS} presented below is a moderate simplification of the original algorithm in \cite{baswana2016dynamic}, by directly pruning those details unrelated to insertions. \begin{algorithm}[H] \caption{\textsf{BatchInsert}} \KwData{a DFS tree $T$ of $G$, set of insertions $U$} \KwResult{a DFS tree $T^*$ of $G + U$} Add each inserted vertex $v$ into $T$, set $\mathit{par}(v) = r$\; Initialize $L(v)$ to be $\emptyset$ for each $v$\; Add each inserted edge $(u, v)$ to $L(u)$ and $L(v)$\; Call $\mathrm{\textsf{DFS}}(r)$\; \end{algorithm} \begin{algorithm}[H] \caption{\textsf{DFS}} \KwData{a DFS tree $T$ of $G$, the entering vertex $v$} \KwResult{a partial DFS tree} Let $u = v$\; \While{$\mathit{par}(u)$ is not visited} { Let $u = \mathit{par}(u)$\; } Mark $\mathit{path}(u, v)$ to be visited\; Let $(w_1, \dots, w_t) = \mathit{path}(u, v)$\; \For{$i \in [t]$} { \If{$i \ne t$} { Let $\mathit{par}^*(w_i) = w_{i + 1}$\; } \For{child $x$ of $w_i$ in $T$ except $w_{i + 1}$} { Let $(y, z) = Q(T(x), u, v)$, where $y \in \mathit{path}(u, v)$\; Add $z$ into $L(y)$\; } } \For{$i \in [t]$} { \For{$x \in L(w_i)$} { \If{$x$ is not visited} { Let $\mathit{par}^*(x) = w_i$\; Call $\mathrm{\textsf{DFS}}(x)$\; } } } \end{algorithm} In Algorithm \textsf{BatchInsert}, we first attach each inserted vertex to the super root $r$, and pretend it has been there since the very beginning. Then only edge insertions are to be considered. All inserted edges are added into the reduced adjacency lists of corresponding vertices. We then use \textsf{DFS}{} to traverse the graph starting from $r$ based on $T$, $L$, and build the new DFS tree while traversing the entire graph and updating the reduced adjacency lists. In Algorithm \textsf{DFS}, the new DFS tree is built in a recursive fashion. Every time we enter an untouched subtree, say $T(u)$, from vertex $v \in T(u)$, we change the root of $T(u)$ to $v$ and go through $\mathit{path}(v, u)$; i.e., we wish to reverse the order of $\mathit{path}(u, v)$ in $T^*$. One crucial step behind this operation is that we need to find a new root for each subtree $T(w)$ originally hanging on $\mathit{path}(u, v)$. The following lemma tells us where the $T(w)$ should be rerooted on $\mathit{path}(u, v)$ in $T^*$. \begin{lemma}[\cite{baswana2016dynamic}] \label{feasible_edge} Let $T^*$ be a partially constructed DFS tree, $v$ the current vertex being visited, $w$ an (not necessarily proper) ancestor of $v$ in tree $T^*$, and $C$ a connected component of the subgraph induced by unvisited vertices. If there are two edges $e$ and $e'$ from $C$ incident on $v$ and $w$, then it is sufficient to consider only $e$ during the rest of the DFS traversal. \end{lemma} Let $Q(T(w), u, v)$ be the edge between the highest vertex on $\mathit{path}(u, v)$ incident to a vertex in subtree $T(w)$, and the corresponding vertex in $T(w)$. $Q(T(w), u, v)$ is defined to be $\mathsf{Null}$ if such an edge does not exist. By Lemma \ref{feasible_edge}, it suffices to ignore all other edges but just keep the edge returned by $Q(T(w), u, v)$; this is because we have reversed the order of $\mathit{path}(u, v)$ in $T^*$ and thus $Q(T(w), u, v)$ connects to the lowest possible position in $T^*$. Hence $T(w)$ should be rerooted at $Q(T(w), u, v)$. Denote $(x, y)$ to be the edge returned by $Q(T(w), u, v)$ where $x \in \mathit{path}(u, v)$, and then we add $y$ into $L(x)$. After finding an appropriate entering edge for each hanging subtree, we process each vertex $v \in \mathit{path}(u, v)$ in ascending order of depth (with respect to tree $T$). For every unvisited $w \in L(v)$, we set $\mathit{par}^*(w) = v$, and recursively call $\mathrm{\textsf{DFS}}(w)$. \begin{theorem} \textsf{BatchInsert}{} correctly reports a feasible DFS tree $T^*$ of graph $G + U$. \end{theorem} \begin{proof} We argue that in a single call $\mathrm{\textsf{DFS}}(v)$, where $u$ is the highest unvisited ancestor of $v$, every unvisited (at the moment of being enumerated) subtree $T(w)$ hanging from $\mathit{path}(u, v)$, as well as every vertex on $\mathit{path}(u, v)$ except $v$, will be assigned an appropriate parent such that these parent-child relationships constitute a DFS tree of $G$ at the termination of \textsf{BatchInsert}{}. When the traversal reaches $v$, the entire $T(u)$ is untouched, or else $u$ would have been marked by a previous visit to some vertex in $T(u)$. We could therefore choose to go through $\mathit{path}(v, u)$ to reach $u$ first. By Lemma~\ref{feasible_edge}, if a subtree $T(w)$ is reached from some vertex on $\mathit{path}(u, v)$, it suffices to consider only the edge $Q(T(w), u, v)$. After adding the query results of all hanging subtrees into the adjacency lists of vertices on $\mathit{path}(u, v)$, every hanging subtree visited from some vertex $x$ on $\mathit{path}(u, v)$ should be visited in a correct way through edges in $L(x)$ solely. Since every vertex will eventually be assigned a parent, \textsf{BatchInsert}{} does report a feasible DFS tree of graph $G + U$. \end{proof} For now we have not discussed how to implement $Q(T(w), u, v)$ and the above algorithm only assumes blackbox queries to $Q(T(\cdot), \cdot, \cdot)$. The remaining problem is to devise a data structure $\mathcal{D}$ to answer all the queries demanded by Algorithm \textsf{DFS} in $O(n)$ total time. We will show in the next section that there exists a data structure $\mathcal{D}$ with the desired performance, which is stated as the following lemma. \begin{lemma} \label{query_time} There exists a data structure $\mathcal{D}$ with preprocessing time $O\left(\min\{m \log n, n^2\}\right)$ time and space complexity $O\left( \min \{ m \log n, n^2 \} \right)$ that can answer all queries $Q(T(w), x, y)$ in a single run of \textsf{BatchInsert}{} in $O(n)$ time. \end{lemma} \begin{proof}[Proof of Theorem \ref{batch-ins}] By Lemma~\ref{query_time}, the total time required to answer queries is $O(n)$. The total size of reduced adjacency lists is bounded by $O(n + |U|)$, composed by $O(|U|)$ edges added in \textsf{BatchInsert}{} and $O(n)$ added during DFS. Thus, the total time complexity of $\textsf{BatchInsert}$ is $O(n + |U|)$. During preprocessing, we use depth first search on $G$ to get the initial DFS tree $T$, and build $\mathcal{D}$ in time $O\left( \min \{ m \log n, n^2 \} \right)$. The total time for preprocessing is $O\left( \min \{ m \log n, n^2 \} \right)$. \end{proof} \section{Introduction} Depth First Search (DFS) is one of the most renowned graph traversal techniques. After Tarjan's seminal work~\cite{tarjan1972depth}, it demonstrates its power by leading to efficient algorithms to many fundamental graph problems, e.g., biconnected components, strongly connected components, topological sorting, bipartite matching, dominators in directed graph and planarity testing. Real world applications often deal with graphs that keep changing with time. Therefore it is natural to study the dynamic version of graph problems, where there is an online sequence of updates on the graph, and the algorithm aims to maintain the solution of the studied graph problem efficiently after seeing each update. The last two decades have witnessed a surge of research in this area, like connectivity~\cite{eppstein1997sparsification,henzinger1999randomized,holm2001poly,kapron2013dynamic}, reachability~\cite{roditty2008improved,sankowski2004dynamic}, shortest path~\cite{demetrescu2004new,roditty2012dynamic}, bipartite matching~\cite{baswana2011fully,neiman2016simple}, and min-cut~\cite{thorup2001fully}. We consider the dynamic maintenance of DFS trees in undirected graphs. As observed by Baswana et al.~\cite{baswana2016dynamic} and Nakamura and Sadakane~\cite{nakamura2017space}, the {\em incremental} setting, where edges/vertices are added but never deleted from the graph, is arguably easier than the {\em fully dynamic} setting where both kinds of updates can happen --- in fact, they provide algorithms for incremental DFS with $\tilde{O}(n)$ worst case update time, which is close to the trivial $\Omega(n)$ lower bound when it is required to explicitly report a DFS tree after each update. {\bf\em So, is there an algorithm that requires nearly linear preprocessing time and space, and reports a DFS tree after each incremental update in $O(n)$ time?} In this paper, we study the problem of maintaining a DFS tree in the incremental setting, and give an affirmative answer to this question. \subsection{Previous works on dynamic DFS} Despite the significant role of DFS tree in static algorithms, there is limited progress on maintaining a DFS tree in the {\em dynamic} setting. Many previous works focus on the {\em total time} of the algorithm for any arbitrary updates. Franciosa et al.~\cite{franciosa1997incremental} designed an incremental algorithm for maintaining a DFS tree in a DAG from a given source, with $O(mn)$ total time for an arbitrary sequence of edge insertions; Baswana and Choudhary~\cite{baswana2015dynamic} designed a decremental algorithm for maintaining a DFS tree in a DAG with expected $O(mn\log n)$ total time. For undirected graphs, Baswana and Khan~\cite{baswana2014incremental} designed an incremental algorithm for maintaining a DFS tree with $O(n^2)$ total time. These algorithms used to be the only results known for the dynamic DFS tree problem. However, none of these existing algorithms, despite that they are designed for only a partially dynamic environment, achieves a worst case bound of $o(m)$ on the update time. That barrier is overcome in the recent breakthrough work of Baswana et al.~\cite{baswana2016dynamic}, they provide, for undirected graphs, a fully dynamic algorithm with worst case $O(\sqrt{mn} \log^{2.5} n)$ update time, and an incremental algorithm with worst case $O(n \log^{3} n)$ update time. Due to the rich information in a DFS tree, their results directly imply faster worst case fully dynamic algorithms for subgraph connectivity, biconnectivity and 2-edge connectivity. The results of Baswana et al.~\cite{baswana2016dynamic} suggest a promising way to further improve the worst case update time or space consumption for those fully dynamic algorithms by designing better dynamic algorithms for maintaining a DFS tree. In particular, based on the framework by Baswana et al.~\cite{baswana2016dynamic}, Nakamura and Sadakane~\cite{nakamura2017space} propose an algorithm which takes $O(\sqrt{mn} \log^{1.75} n / \sqrt{\log \log n})$ time per update in the fully dynamic setting and $O(n \log n)$ time in the incremental setting, and $O(m \log n)$ bits of space. \subsection{Our results} In this paper, following the approach of~\cite{baswana2016dynamic}, we improve the update time for the incremental setting, also studied in~\cite{baswana2016dynamic}, by combining a better data structure, a novel tree-partition lemma by Duan and Zhang~\cite{duan2016improved} and the fractional-cascading technique by Chazelle and Guibas~\cite{chazelle1986fractional,chazelle1986fractional2}. For any set $U$ of incremental updates (insertion of a vertex/an edge), we let $G + U$ denote the graph obtained by applying the updates in $U$ to the graph $G$. Our results build on the following main theorem. \begin{theorem}\label{batch-ins} There is a data structure with $O(\min\{m \log n, n^2\})$ size, and can be built in $O(\min\{m \log n, n^2\})$ time, such that given a set $U$ of $k$ insertions, a DFS tree of $G + U$ can be reported in $O(n+k)$ time. \end{theorem} By the above theorem combined with a de-amortization trick in~\cite{baswana2016dynamic}, we establish the following corollary for maintaining a DFS tree in an undirected graph with incremental updates. \begin{corollary}[\textbf{Incremental DFS tree}]\label{cor-incre-dfs} Given a sequence of online edge/vertex insertions, a DFS tree can be maintained in $O(n)$ worst case time per insertion. \end{corollary} \subsection{Organization of the Paper} In Section~2 we introduce frequently used notations and review two building blocks of our algorithm --- the tree partition structure \cite{duan2016improved} and the fractional cascading technique \cite{chazelle1986fractional,chazelle1986fractional2}. In Section~3, we consider a batched version of the incremental setting, where all incremental updates are given at once, after which a single DFS tree is to be reported. Given an efficient scheme to answer queries of form $Q(T(\cdot), \cdot, \cdot)$, we prove Theorem~\ref{batch-ins}, which essentially says there is an efficient algorithm, which we call \textsf{BatchInsert}{}, for the batched incremental setting. In Section~4, we elaborate on the implementation of the central query subroutine $Q(T(\cdot), \cdot, \cdot)$ used in the batched incremental algorithm. We first review a standard de-amortization technique, applying which our algorithm for the batched setting directly implies the efficient algorithm for the incremental setting stated in Corollary~\ref{cor-incre-dfs}. We then, in Sections~\ref{sec:logn}~and~\ref{sec:nsquare} respectively, introduce (1) an optimized data structure that takes $O(m \log n)$ time for preprocessing and answers each query in $O(\log n)$ time, and (2) a relatively simple data structure that takes $O(n^2)$ time for preprocessing and answers each query in $O(1)$ time. One of these two structures, depending on whether $m \log n > n^2$ or not, is then used in Section~\ref{sec:mlogn} to implement a scheme that answers each query in amortized $O(1)$ time. This is straightforward when the $(n^2, 1)$ structure is used. When instead the $(m \log n, \log n)$ structure is used, we apply a nontrivial combination of the tree partition structure and the fractional cascading technique to bundle queries together, and answer each bundle using a single call to the $(m \log n, \log n)$ structure. We show that the number of such bundles from queries made by \textsf{BatchInsert}{} cannot exceed $O(n / \log n)$, so the total time needed for queries is $O(n)$. This finishes the proof of Theorem~\ref{batch-ins} and Corollary~\ref{cor-incre-dfs} and concludes the paper. \section{Preliminaries} Let $G = (V, E)$ denote the original graph, $T$ a corresponding DFS tree, and $U$ a set of inserted vertices and edges. We first introduce necessary notations. \begin{itemize} \item $T(x)$: The subtree of $T$ rooted at $x$. \item $\mathit{path}(x, y)$: The path from $x$ to $y$ in $T$. \item $\mathit{par}(v)$: The parent of $v$ in $T$. \item $N(x)$: The adjacency list of $x$ in $G$. \item $L(x)$: The reduced adjacency list for vertex $x$, which is maintained during the algorithm. \item $T^*$: The newly generated DFS tree. \item $\mathit{par}^*(v)$: The parent of $v$ in $T^*$. \end{itemize} Our result uses a tree partition lemma in \cite{duan2016improved} and the famous fractional cascading structure in \cite{chazelle1986fractional,chazelle1986fractional2}, which are summarized as the following two lemmas. \begin{lemma}[Tree partition structure \cite{duan2016improved}] \label{tree_partition_lem} Given a rooted tree $T$ and any integer parameter $k$ such that $2 \le k \le n = |V(T)|$, there exists a subset of vertices $M \subseteq V(T)$, $|M|\le 3n/k-5$, such that after removing all vertices in $M$, the tree $T$ is partitioned into sub-trees of size at most $k$. We call every $v \in M$ an $M$-marked vertex, and $M$ a marked set. Also, such $M$ can be computed in $O(n \log n)$ time. \end{lemma} \begin{lemma}[Fractional cascading \cite{chazelle1986fractional,chazelle1986fractional2}] \label{fractional_cascading} Given $k$ sorted arrays $\{A_i\}_{i \in [k]}$ of integers with total size $\sum_{i=1}^k |A_i| = m$. There exists a data structure which can be built in $O(m)$ time and using $O(m)$ space, such that for any integer $x$, the successors of $x$ in all $A_i$'s can be found in $O(k + \log m)$ time. \end{lemma} \section{Dealing with queries in \textsf{BatchInsert}} In this section we prove Lemma~\ref{query_time}. Once this goal is achieved, the overall time complexity of batch insertion taken by Algorithm~\textsf{BatchInsert}{} would be $O(n + |U|)$. In the following part of this section, we will first devise a data structure in Section \ref{sec:logn}, that answers any single query $Q(T(w), u, v)$ in $O(\log n)$ time, which would be useful in other parts of the algorithm. We will then present another simple data structure in Section \ref{sec:nsquare}, which requires $O(n^2)$ preprocessing time and $O(n^2)$ space and answers each query in $O(1)$ time. Finally, we propose a more sophisticated data structure in Section \ref{sec:mlogn}, which requires $O(m\log n)$ preprocessing time and $O(m \log n)$ space and answer all queries $Q(T(w), x, y)$ in a single run of \textsf{BatchInsert}{} in $O(n)$ time. Hence, we can always have an algorithm that handles a batch insertion $U$ in $O(n + |U|)$ time using $O(\min\{m\log n, n^2\})$ preprocessing time and $O(\min\{m\log n, n^2\})$ space, thus proving Theorem \ref{batch-ins}. We can then prove Corollary \ref{cor-incre-dfs} using the following standard de-amortization argument. \begin{lemma}{(Lemma 6.1 in \cite{baswana2016dynamic})} \label{deamortization} Let $\mathcal{D}$ be a data structure that can be used to report the solution of a graph problem after a set of $U$ updates on an input graph $G$. If $\mathcal{D}$ can be initialized in $O(f)$ time and the solution for graph $G + U$ can be reported in $O(h + |U| \times g)$ time, then $\mathcal{D}$ can be modified to report the solution after every update in worst-case $O\left(\sqrt{fg} + h\right)$ update time after spending $O(f)$ time in initialization, given that $\sqrt{f / g} \le n$. \end{lemma} \begin{proof}[Proof of Corollary \ref{cor-incre-dfs}] Taking $f = \min\{m\log n, n^2 \}$, $g = 1$, $h = n$ and directly applying the above lemma will yield the desired result. \end{proof} \subsection{Answering a single query in $O(\log n)$ time} \label{sec:logn} We show in this subsection that the query $Q(T(\cdot), \cdot, \cdot)$ can be reduced efficiently to the range successor query (see, e.g., \cite{nekrich2012sorted}, for the definition of range successor query), and show how to answer the range successor query, and thus any individual query $Q(T(\cdot), \cdot, \cdot)$, in $O(\log n)$ time. To deal with a query $Q(T(w), x, y)$, first note that since $T$ is a DFS tree, all edges not in $T$ but in the original graph $G$ must be ancestor-descendant edges. Querying edges between $T(w)$ and $\mathit{path}(x, y)$ where $x$ is an ancestor of $y$ and $T(w)$ is hanging from $\mathit{path}(x, y)$ is therefore equivalent to querying edges between $T(w)$ and $\mathit{path}(x, \mathit{par}(w))$, i.e., $Q(T(w), x, y) = Q(T(w), x, \mathit{par}(w))$. From now on, we will consider queries of the latter form only. Consider the DFS sequence of $T$, where the $i$-th element is the $i$-th vertex reached during the DFS on $T$. Note that every subtree $T(w)$ corresponds to an interval in the DFS sequence. Denote the index of vertex $v$ in the DFS sequence by $\mathit{first}(v)$, and the index of the last vertex in $T(v)$ by $\mathit{last}(v)$. During the preprocessing, we build a 2D point set $S$. For each edge $(u, v) \in E$, we add a point $p = (\mathit{first}(u), \mathit{first}(v))$ into $S$. Notice that for each point $p \in S$, there exists exactly one edge $(u, v)$ associated with $p$. Finally we build a 2D range tree on point set $S$ with $O(m\log n)$ space and $O(m\log n)$ preprocessing time. To answer an arbitrary query $Q(T(w), x, \mathit{par}(w))$, we query the point with minimum $x$-coordinate lying in the rectangle $\Omega = [\mathit{first}(x), \mathit{first}(w) - 1] \times [\mathit{first}(w), \mathit{last}(w)]$. If no such point exists, we return \textsf{Null} for $Q(T(w), x, \mathit{par}(w))$. Otherwise we return the edge corresponding to the point with minimum $x$-coordinate. Now we prove the correctness of our approach. \begin{itemize} \item If our method returns \textsf{Null}, $Q(T(w), x, \mathit{par}(w))$ must equal \textsf{Null}. Otherwise, suppose $Q(T(w), x, \mathit{par}(w)) = (u, v)$. Noticing that $(\mathit{first}(u), \mathit{first}(v))$ is in $\Omega$, it means our method will not return \textsf{Null} in that case. \item If our method does not return \textsf{Null}, denote $(u', v')$ to be the edge returned by our method. We can deduce from the query rectangle that $u' \in T(x) \backslash T(w)$ and $v' \in T(w)$. Thus, $Q(T(w), x, \mathit{par}(w)) \neq \textsf{Null}$. Suppose $Q(T(w), x, \mathit{par}(w)) = (u, v)$. Notice that $(\mathit{first}(u), \mathit{first}(v))$ is in $\Omega$, which means $\mathit{first}(u') \le \mathit{first}(u)$. If $u' = u$, then our method returns a feasible solution. Otherwise, from the fact that $\mathit{first}(u') < \mathit{first}(u)$, we know that $u'$ is an ancestor of $u$, which contradicts the definition of $Q(T(w), x, \mathit{par}(w))$. \end{itemize} \subsection{An $O(n^2)$-space data structure}\label{sec:nsquare} \label{pre_all} In this subsection we propose a data structure with quadratic preprocessing time and space complexity that answers any $Q(T(\cdot), \cdot, \cdot)$ in constant time. Since we allow quadratic space, it suffices to precompute and store answers to all possible queries $Q(T(w), u, \mathit{par}(w))$. For preprocessing, we enumerate each subtree $T(w)$, and fix the lower end of the path to be $v = \mathit{par}(w)$ while we let the upper end $u$ go upward from $v$ by one vertex at a time to calculate $Q(T(w), u, v)$ incrementally, in order to get of the form $Q(T(w), \cdot, \cdot)$ in $O(n)$ total time. As $u$ goes up, we check whether there is an edge from $T(w)$ to the new upper end $u$ in $O(1)$ time; for this task we build an array (based on the DFS sequence of $T$) for each vertex, and insert an 1 into the appropriate array for each edge, and apply the standard prefix summation trick to check whether there is an 1 in the range corresponding to $T(w)$. To be precise, let $A_u: [n] \rightarrow \{0, 1\}$ denote the array for vertex $u$. Recall that $\mathit{first}(v)$ denotes the index of vertex $v$ in the DFS sequence, and $\mathit{last}(v)$ the index of the last vertex in $T(v)$. For a vertex $u$, we set $A_u[\mathit{first}(v)]$ to be 1 if and only if there is an edge $(u, v)$ where $u$ is the higher end. Now say, we have the answer to $Q(T(w), u, v)$ already, and want to get $Q(T(w), u', v)$ in $O(1)$ time, where $u' = \mathit{par}(u)$. If there is an edge between $T(w)$ and $u'$, then it will be the answer. Or else the answer to $Q(T(w), u', v)$ will be the same as to $Q(T(w), u, v)$. In order to know whether there is an edge between $T(w)$ and $u'$, we check the range $[\mathit{first}(w), \mathit{last}(w)]$ in $A_{u'}$, and see if there is an $1$ in $O(1)$ time using the prefix summation trick. \begin{lemma} \label{preprocess_all} The preprocessing time and query time of the above data structure are $O(n^2)$ and $O(1)$ respectively. \end{lemma} \begin{proof} The array $A_u$ and its prefix sum can be computed for each vertex $u$ in total time $O(n^2)$. For each subtree $T(w)$, we go up the path from $w$ to the root $r$, and spend $O(1)$ time for each vertex $u$ on $\mathit{path}(r, w)$ to get the answer for $Q(T(w), u, \mathit{par}(w))$. There are at most $n$ vertices on $\mathit{path}(r, w)$, so the time needed for a single subtree is $O(n)$, and that needed for all subtrees is $n \cdot O(n) = O(n^2)$ in total. On the other hand, for each query, we simply look it up and answer in $O(1)$ time. Hence we conclude that the preprocessing time and query time are $O(n^2)$ and $O(1)$ respectively. \end{proof} \subsection{An $O(m\log n)$-space data structure}\label{sec:mlogn} Observe that in \textsf{BatchInsert}{} (and \textsf{DFS}), a bunch of queries $\{Q(T(w_i), x, y)\}$ are always made simultaneously, where $\{T(w_i)\}$ is the set of subtrees hanging from $\mathit{path}(x, y)$. We may therefore answer all queries for a path in one pass, instead of answering them one by one. By doing so we confront two types of hard queries. First consider an example where the original DFS tree $T$ is a chain $L$ where $a_1$ is the root of $L$ and for $1 \le i \le n - 1$, $a_{i+1}$ is the unique child of $a_i$. When we invoke $\textsf{DFS}(a_1)$ on $L$, $path(u, v)$ is the single node $a_1$. Thus, we will call $Q(T(a_2), a_1, a_1)$ and add the returned edge into $L(a_1)$. Supposing there are no back-edges in this graph, the answer of $Q(T(a_2), a_1, a_1)$ will be the edge $(a_1, a_2)$. Therefore, we will recursively call the $\textsf{DFS}(a_2)$ on the chain $(a_2, a_n)$. Following further steps of \textsf{DFS}, we can see that we will call the query $Q(T(w), x, y)$ for $\Omega(n)$ times. For the rest of this subsection, we will show that we can deal with this example in linear time. The idea is to answer queries involving short paths in constant time. For instance, in the example shown above, $path(u, v)$ always has constant length. We show that when the length of $path(u, v)$ is smaller than $2\log n$, it is affordable to preprocess all the answers to queries of this kind in $O(m\log n)$ time and $O(n \log n)$ space \begin{figure}[!ht] \centering \includegraphics[width=0.6\linewidth]{many_children} \captionsetup{justification=centering} \caption{In this example, if we stick to the 2D-range-based data structure introduced before, then computing all $Q(T(a_i), r, r^\prime)$ would take as much as $O(n\log n)$ time.} \label{fig:heavy} \end{figure} The second example we considered is given as Figure \ref{fig:heavy}. In this tree, the original root is $r$. Suppose the distance between $r$ and $r'$ is $n / 2$. When we invoke $\textsf{DFS}(r')$, $path(u, v)$ the path from $r$ to $r'$. Thus, we will call $T(a_1, r, r')$, $T(a_2, r, r')$, $\ldots$, $T(a_{n - 2}, r, r')$, which means we make $\Omega(n)$ queries. In order to deal with this example in linear time, the main idea is using fractional cascading to answer all queries $Q(T(w), x, y)$ with a fixed $path(u, v)$, for all subtrees $T(w)$ with small size. In the examples shown above, all subtrees cut off $path(u, v)$ have constant size and thus the total time complexity for this example is $O(n)$. We will finally show that, by combining the two techniques mentioned above, it is enough to answer all queries $Q(T(w), x, y)$ in linear time, thus proving Lemma~\ref{query_time}. \subsection*{Data structure} The data structure consists of the following parts. \begin{enumerate}[(\romannumeral1)] \item Build the 2D-range successor data structure that answers any $Q(T(\cdot), \cdot, \cdot)$ in $O(\log n)$ time. \item For each ancestor-descendent pair $(u, v)$ such that $u$ is at most $2\log n$ hops above $v$, precompute and store the value of $Q(T(v), u, \mathit{par}(v))$. \item Apply Lemma \ref{tree_partition_lem} with parameter $k = \log n$ and obtain a marked set of size $O(n / \log n)$. Let $M$ be the set of all marked vertices $x$ such that $|T(x)| \geq \log n$. For every $v\notin M$, let $\mathit{anc}_v\in M$ be the nearest ancestor of $v$ in set $M$. Next we build a fractional cascading data structure for each $u\in M$ in the following way. Let $M_u$ be the set of all vertices in $T(u)$ whose tree paths to $u$ do not intersect any other vertices $u^\prime \neq u$ from $M$, namely $M_u = \{v\mid \mathit{anc}_v = u \}$; see Figure \ref{ds} for an example. Then, apply Lemma \ref{fractional_cascading} on all $N(v), v\in M_u$ where $N(v)$ is treated as sorted array in an ascending order with respect to depth of the edge endpoint opposite to $v$; this would build a fractional cascading data structure that, for any query encoded as a $w\in V$, answers for every $v\in M_u$ its highest neighbour below vertex $w$ in total time $O(|M_u| + \log n)$. \end{enumerate} Here is a structural property of $M$ that will be used when answering queries. \begin{lemma}\label{struct-marked} For any ancestor-descendent pair $(u, v)$, if $\mathit{path}(u, v)\cap M = \emptyset$, then $\mathit{path}(u, v)$ has $\leq 2\log n$ hops. \end{lemma} \begin{proof} Suppose otherwise. By definition of marked vertices there exists a marked vertex $w\in \mathit{path}(u, v)$ that is $\leq \log n$ hops below $u$. Then since $\mathit{path}(u, v)$ has $>2\log n$ many hops, it must be $T(w)\geq \log n$ which leads to $w\in M$, contradicting $\mathit{path}(u, v)\cap M = \emptyset$. \end{proof} \begin{center} \begin{figure} \begin{subfigure}{0.5\textwidth} \includegraphics[width=3in, height=2.4in]{ds} \caption{In this example, each blue node represents a vertex $v_i (1\leq i\leq 4)$ from set $M$, and $M_{v_i}$'s are drawn as yellow triangles. For each triangle, a fractional cascading data structure is built on adjacency lists of all vertices inside.} \label{ds} \end{subfigure} \quad \begin{subfigure}{0.5\textwidth} \includegraphics[width=3.3in, height=2.5in]{qry} \caption{In this picture, sets $M$ and $X\cup \{r\}$ are drawn as blue nodes and black nodes respectively, and each yellow triangle is a subtree rooted at a leaf of $T[X]$, which has size $\geq \log n$. Note that every ancestor-descendent tree path between two black nodes contains a blue node.} \label{qry} \end{subfigure} \end{figure} \end{center} \subsection*{Preprocessing time} First of all, for part (\romannumeral1), as discussed in a previous subsection, 2D-range successor data structure takes time $O(m\log n)$ to initialize. Secondly, for part (\romannumeral3), on the one hand by Lemma \ref{tree_partition_lem} computing a tree partition takes time $O(n\log n)$; on the other hand, by Lemma \ref{fractional_cascading}, initializing the fractional cascading with respect to $u\in M$ costs $O(\sum_{v\in M_u}|N(v)|)$ time. Since, by definition of $M_u$, each $v\in V$ is contained in at most one $M_u, u\in M$, the overall time induced by this part would be $O(\sum_{u\in M}\sum_{v\in M_u}|N(v)|) = O(m)$. Preprocessing part (\romannumeral2) requires a bit of cautions. The procedure consists of two steps. \begin{enumerate}[(1)] \item For every ancestor-descendent pair $(u, v)$ such that $u$ is at most $2\log n$ hops above $v$, we mark $(u, v)$ if $u$ is incident to $T(v)$. Here goes the algorithm: for every edge $(u, w)\in E$ ($u$ being the ancestor), let $z\in \mathit{path}(u, w)$ be the vertex which is $2\log n$ hops below $u$ (if $\mathit{path}(u, w)$ has less than $2\log n$ hops, then simply let $z = w$); note that this $z$ can be found in constant time using the level-ancestor data structure \cite{bender2000lca} which can be initialized in $O(n)$ time. Then, for every vertex $v\in \mathit{path}(u, z)$, we mark the pair $(u, v)$. The total running time of this procedure is $O(m\log n)$ since each edge $(u, w)$ takes up $O(\log n)$ time. \item Next, for each $v\in V$, we compute all entries $Q(T(v), u, \mathit{par}(v))$ required by (\romannumeral2) in an incremental manner. Let $u_1, u_2, \cdots, u_{2\log n}$ be the nearest $2\log n$ ancestors of $v$ sorted in descending order with respect to depth, and then we directly solve the recursion $Q(T(v), u_{i+1}, \mathit{par}(v)) = \begin{cases} Q(T(v), u_{i}, \mathit{par}(v)) & (u_{i+1}, v)\text{ is not marked}\\ u_{i+1} & i=0\text{ or }(u_{i+1}, v)\text{ is marked}\\ \end{cases}$ for all $0\leq i < 2\log n$ in $O(\log n)$ time. The total running time would thus be $O(n\log n)$. \end{enumerate} Summing up (\romannumeral1)(\romannumeral2)(\romannumeral3), the preprocessing time is bounded by $O(m\log n)$. \subsection*{Query algorithm and total running time} We show how to utilize the above data structures (\romannumeral1)(\romannumeral2)(\romannumeral3) to implement $Q(T(\cdot), \cdot, \cdot)$ on line 9-11 in Algorithm \textsf{DFS} such that the overall time complexity induced by this part throughout a single execution of Algorithm \textsf{BatchInsert}{} is bounded by $O(n)$. Let us say we are given $(w_1, w_2, \cdots, w_t) = \mathit{path}(u, v)$ and we need to compute $Q(T(x), u, v)$ for every subtree $T(x)$ that is hanging on $\mathit{path}(u, v)$. There are three cases to discuss. \begin{enumerate}[(1)] \item If $\mathit{path}(u, v)\cap M = \emptyset$, by Lemma \ref{struct-marked} we claim $\mathit{path}(u, v)$ has at most $2\log n$ hops, and then we can directly retrieve the answer of $Q(T(x), u, v)$ from precomputed entries of (\romannumeral2), each taking constant query time. \item Second, consider the case where $\mathit{path}(u, v)\cap M \neq \emptyset$. Let $s_1, s_2, \cdots, s_l, l\geq 1$ be the consecutive sequence (in ascending order with respect to depth in tree $T$) of all vertices from $M$ that are on $\mathit{path}(u, v)$. For those subtrees $T(x)$ that are hanging on $\mathit{path}(u, \mathit{par}(s_1))$, we can directly retrieve the value of $Q(T(x), u, \mathit{par}(x))$ from (\romannumeral2) in constant time, as by Lemma~\ref{struct-marked} $\mathit{path}(u, \mathit{par}(s_1))$ has at most $2\log n$ hops. \item Third, we turn to study the value of $Q(T(x), u, \mathit{par}(x))$ when $\mathit{par}(x)$ belongs to a $\mathit{path}(s_i, \mathit{par}(s_{i+1})), i<l$ or $\mathit{path}(s_l, v)$. The algorithm is two-fold. \begin{enumerate}[(a)] \item First, we make a query of $u$ to the fractional cascading data structure built at vertex $s_i$ ($1\leq i\leq l$), namely part (\romannumeral3), which would give us, for every descendent $y\in M_{s_i}$, the highest neighbour of $y$ below $u$. Using this information we are able to derive the result of $Q(T(x), u, v)$ if $|T(x)| < \log n$, since in this case $T(x)\cap M = \emptyset$ and thus $T(x)\subseteq M_{s_i}$. By Lemma \ref{fractional_cascading} the total time of this procedure is $O(|M_{s_i}| + \log n)$. \item We are left to deal with cases where $|T(x)| \geq \log n$. In this case, we directly compute $Q(T(x), u, v)$ using the 2D-range successor built in (\romannumeral1) which takes $O(\log n)$ time. \end{enumerate} \end{enumerate} Correctness of the query algorithm is self-evident by the algorithm. The total query time is analysed as following. Throughout an execution of Algorithm \textsf{BatchInsert}, (1) and (2) contribute at most $O(n)$ time since each $T(x)$ is involved in at most one such query $Q(T(x), u, v)$ which takes constant time. As for (3)(a), since each marked vertex $s\in M$ lies in at most one such path $(w_1, w_2, \cdots, w_t) = \mathit{path}(u, v)$, the fractional cascading data structure associated with $M_s$ is queried for at most once. Hence the total time of (3)(a) is $O(\sum_{s\in M}(|M_s| + \log n)) = O(n + |M|\log n) = O(n)$; the last equality holds by $|M|\leq O(n / \log n)$ due to Lemma \ref{tree_partition_lem}. Finally we analyse the total time taken by (3)(b). It suffices to upper-bound by $O(n / \log n)$ the total number of such $x$ with the property that $|T(x)| \geq \log n$ and $\mathit{path}(u, \mathit{par}(x))\cap M \neq \emptyset$. Let $X$ be the set of all such $x$'s. \begin{lemma}\label{struct-large-tree} Suppose $x_1, x_2\in X$ and $x_1$ is an ancestor of $x_2$ in tree $T$. Then $\mathit{path}(x_1, x_2)\cap M \neq \emptyset$. \end{lemma} \begin{proof} Suppose otherwise $\mathit{path}(x_1, x_2)\cap M = \emptyset$. Consider the time when query $Q(T(x_2), u, v)$ is made and let $\mathit{path}(u, v)$ be the path being visited by then. As $x_2\in X$, by definition it must be $\mathit{path}(u, \mathit{par}(x_2))\cap M\neq \emptyset$. Therefore, $\mathit{path}(u, x_2)$ is a strict extension of $\mathit{path}(x_1, x_2)$, and thus $x_1, \mathit{par}(x_1)\in \mathit{path}(u, x_2)$, which means $x_1$ and $\mathit{par}(x_1)$ become visited in the same invocation of Algorithm \textsf{DFS}. This is a contradiction since for any query of form $Q(T(x_1), \cdot, \cdot)$ to be made, by then $\mathit{par}(x_1)$ should be tagged ``visited'' while $x_1$ is not. \end{proof} Now we prove $|X| = O(n / \log n)$. Build a tree $T[X]$ on vertices $X\cup \{r\}$ in the natural way: for each $x\in X$, let its parent in $T[X]$ be $x$'s nearest ancestor in $X\cup \{r\}$. Because of $$|X| < 2\#\text{leaves of }T[X] + \#\text{vertices with a unique child in }T[X]$$ it suffices to bound the two terms on the right-hand side: on the one hand, the number of leaves of $T[X]$ is at most $n / \log n$ since for each leave $x$ it has $|T(x)|\geq \log n$; on the other hand, for each $x\in T[X]$ with a unique child $y\in T[X]$, by Lemma \ref{struct-large-tree} $\mathit{path}(x, y)\cap M \neq \emptyset$, and so we can charge this $x$ to an arbitrary vertex in $\mathit{path}(x, y)\cap M$, which immediately bounds the total number of such $x$'s by $|M| = O(n / \log n)$; see Figure \ref{qry} for an illustration. Overall, $|X| \leq O(n / \log n)$.
34973f407d5ca4e750c9372e7cee1f1af47f89b1
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{section:Introduction} Since 2012, the focus of Galactic center (GC) observations has been set on investigating an infrared (IR) excess source detected by \cite{Gillessen2012} as a fast-moving object approaching the position of the central supermassive black hole (SMBH) of the Milky Way, Sagittarius A* (Sgr~A*). It has been interpreted as a combination of dust and core-less gas cloud called G2 \citep{Gillessen2012, Gillessen2013a, Pfuhl2015} and also DSO, standing for Dusty S-cluster Object \citep{Eckart2013}. This source moves on a highly eccentric orbit and passed its closest approach to the SMBH in May 2014 \citep{meyer2014,Valencia2015}. Given the short distance of its periapse, it has been suspected that it might produce extraordinary accretion events on to the galaxy's central black hole \citep[e.g.][]{Shcherbakov2014, Abarca2014, Scoville2013, Sadowski2013}. If the DSO is a pure gas cloud of a few Earth masses \citep{Gillessen2012}, it might have formed in the stellar cluster, possibly within the disk of young stars at a distance of few arcseconds from the GC. After forming there, it might have moved on its current remarkably eccentric orbit by gravitational interaction with massive stars \citep{Murray-Clay2012, Scoville2013}. This scenario must have happened recently (1990-2000), therefore it should have been observed during the total time of its existence. As a consequence, the pure gas scenario seems unlikely, and several authors have proposed scenarios suggesting the presence of a central star for this source \citep[e.g.,][]{Murray-Clay2012, Eckart2013, Scoville2013, Ballone2013, Phifer2013, Zajacek2014, Witzel2014, Valencia2015}. \cite{Scoville2013} proposed that the DSO is a T~Tauri star that was formed in the young stellar ring and then inserted into its current orbit. They suggested that a very dense bow~shock is produced for the T~Tauri star wind and modeled it numerically. \cite{Valencia2015} also discussed that the bright observed Br$\gamma$ emission of the DSO with a large line width might be the result of infalling material shaping a disk around the central star, which may be a T~Tauri star with an age of $\sim10^5$ yr. Considering the stellar nature, DSO would not be disrupted when reaching its closest point to the SMBH and did not need a recent formation. Using hydrodynamical simulations, \cite{Jalali2014} have shown that young stars could form very close to SMBHs within small molecular clumps on eccentric orbits around the black hole. They showed that for such orbital configurations, the gravitational potential of the SMBH and orbital (geometrical) compression increase the density of cold gas clumps to reach the threshold values suitable for star formation (see also \cite{mapelli2016} for a recent review). The $L$-band observations of DSO/G2 close to the peribothron support the idea of the compactness of this source, which means that the source cannot be a pure gas cloud \citep{Ghez2014, Witzel2014}. \cite{Eckart2013} revealed the first $K_\mathrm{s}$-band identification of \textcolor[rgb]{1,0.501961,0}{\textcolor[rgb]{1,0.501961,0}{\textcolor[rgb]{0,0,0}{the} \textcolor[rgb]{0,0,0}{DSO}} }with a magnitude of $\sim$18.9 from the ESO Very Large Telescope (VLT) continuum imaging data. Using the spectral decomposition of this source, they obtained an upper limit of $\sim 30~L_{\odot}$ for its luminosity. The $H$, $K_\mathrm{s}$, and $L$-band continuum measurements can be matched either by an unusually warm dust component at a temperature of 550-650 K or by a stellar source enclosed in the dust at a temperature of $\sim$ 450 K \citep[see Fig.~15 in][]{Eckart2013}. The $H-K_\mathrm{s} > 2.3$ color limit supports the scenario that the DSO is a dust-embedded star and not a core-less cloud of gas and dust \citep{Eckart2013}. The mass of this object is higher than what was assumed for a pure gas source, but lower than the typical mass of S-cluster stars ($20 M_{\odot}$). \cite{Valencia2015} reported NIR observations of the DSO during its approach to the SMBH at the GC, which were carried out with SINFONI at the VLT from February to September 2014. They detected spatially compact Br$\gamma$ line emission from the DSO before and after its peribothron passage and also a Br$\gamma$ line width increase, which may indicate that the DSO is a young accreting star with a dust envelope. The observational data were used to obtain the orbital parameters of this object. Comparable to the previous estimates \citep[e.g.,][]{Meyer2014a}, \cite{Valencia2015} obtained a peribothron distance of about $163\pm16$ AU with a half-axis length of about 33 mpc and an ellipticity $e=0.976$. When the DSO reaches the peribothron, tidal stretching and disruption of the envelope lead to a velocity dispersion enhancement in the accretion flow toward the central star \citep{Eckart2013, Zajacek2014}. Based on a study by \cite{Witzel2014}, the $L$-band emission of the DSO compared to the Br$\gamma$ emission measured by \cite{Pfuhl2015} is more compact. This shows that the $L$-band emission originates from an optically thick dust envelope around a central star, while the Br$\gamma$ emission is coming from the hot gas that is externally heated by ionized photons of the stars close to the DSO. However, \cite{Valencia2015} did not find evidence for significantly extended and tidally stretched Br$\gamma$ emission. The extra emission of Br$\gamma$ close to the DSO position is not connected to the DSO and is most likely emitted from other sources in the field (Peissker et al., in prep.). The DSO is not the only infrared excess source in the S-cluster of the GC, and there are more dusty sources in this region \citep{Eckart2013,Meyer2014a}. These sources might be dust-enshrouded pre-main-sequence stars that form a bow shock ahead of their path when they move through the medium with a supersonic speed. Other candidates have also been observed in the radio continuum observations of the GC \citep{yusef2015}. \\ Imaging polarimetry is a powerful technique for studying dusty environments such as core-less dusty objects and/or circumstellar dusty regions. The analysis of polarization allows us to quantitatively evaluate the object geometries and the dust properties. Intrinsic polarization can be generated only if the system is not symmetric. The asymmetry can occur when the radiation field of the star is not isotropic as a result of a geometric distortion, for instance, when the star develops a bow~shock ahead of its path, or when its photosphere surface brightness is not uniform, or in other words, is influenced by bright spots. Therefore, supplementary to considering the continuum and line emissions from the DSO, studying the light polarization can be very helpful in determining the nature and properties of this source. If the DSO is a bow~shock, the polarization is determined by the bow~shock morphology. Subsequently, the E-vectors are predicted to be perpendicular to the direction of motion if the medium is homogeneous. If the dust shell surrounding the DSO is a disk, then the resulting polarization depends on the disk inclination. In this paper we analyze the NIR polarimetric imaging data taken with NACO at the ESO VLT using its Wollaston prism to study the polarization properties of the DSO. In Sect. \ref{section:Observations} we begin with the details about the observations, then describe the data reduction and determine the position of the DSO in the images. In Sect. \ref{section:results} we present the results of the applied flux density calibration method: light curves, polarimetry measurements, and their statistical analysis. We present the performed radiative transfer model to describe the DSO polarization in Sect. \ref{dso-model}. We discuss the implications of our results in Sect. \ref{section:discussion} and finally summarize the main results of the NIR polarimetry of the DSO in Sect. \ref{section:summary}. \begin{figure*}[ \begin{center} \subfigure{% \includegraphics[width=0.45\textwidth]{2008_referee_new.eps} } \subfigure{% \includegraphics[width=0.45\textwidth]{2009_referee_new.eps} }\\ \subfigure{% \includegraphics[width=0.45\textwidth]{2011_referee_new.eps} } \subfigure{% \includegraphics[width=0.45\textwidth]{2012_referee_new.eps} }\\ \end{center} \caption[$K_\mathrm{s}$-band deconvolved median images of the central arcsecond of the GC in polarimetry mode]{Final $K_\mathrm{s}$-band deconvolved median images of the central arcsecond at the GC in polarimetry mode ($90^{\circ}$ polarization channel) in 2008, 2009, 2011, and 2012 from top left to bottom right. The position of the DSO is shown by a circle on its orbit.} The asterisk indicates the position of Sgr~A*. In all the images north is up and east is to the left. \label{fig:allepochs-dso} \end{figure*} \section{Observations and data reduction} \label{section:Observations} \subsection{Data set and reduction process} \label{section:the data set} The NIR observations have been carried out at the ESO VLT on Paranal, Chile. The data were obtained using the NAOS adaptive optics (AO) module and NIR camera CONICA \citep[together NACO;][]{Lenzen2003, Rousset2003, Brandner2002} at the VLT on Paranal, Chile. The AO loop was locked on IRS7, a supergiant with $K_\mathrm{s}$ $\sim 6.5-7.0$~mag and located $\sim 5.5''$ north of Sgr~A*, using the NIR wavefront sensor. We collected all $K_\mathrm{s}$-band (2.2 $\mu m$) data of the GC taken with S13 camera in 13 mas pixel scale polarimetry mode from 2004 to 2012. The Wollaston prism with the combination of a half-wave retarder plate in NACO provides the possibility of simultaneously measuring two orthogonal directions of the electric field vector. We used the reduced data sets as presented in \cite{shahzamanian2015b}. Standard data reduction was applied with flat-fielding, sky subtraction, and bad-pixel correction. All the dithered exposures were aligned using a cross-correlation algorithm \citep[ESO Eclipse Jitter;][]{Devillard1999}. The sky background was measured on a dark cloud located at $713''$ east and $400''$ north of GC. The final sky background was obtained by getting the median of the dithered science epochs. During the observations, the PSF changed because the weather condition varied. Therefore, the quality of each epoch was determined based on the PSF measured from the stars in the field of view at the observing time. We created a data cube from the combination of the best-quality exposures (with seeing $<2''$) of each observing night. To have a final image with longer integration time on the source and higher signal-to-noise-ratio ($S/N$), we obtained the median of the spatial pixels of the combined cube images. In this case, we cannot study the flux density and polarimetry variability on short times of minutes to months. For 2008, we combined all of the observing nights, since the position of the DSO does not change significantly within a few months. Table \ref{table:log} shows observation dates, number of exposures, and integration times of the data sets used for the analysis here before the data were combined. We used the Lucy-Richardson deconvolution algorithm on the resulting image of the individual years created from the data cube for the aperture photometry. The PSF was extracted using the IDL-based StarFinder routine \citep{Diolaiti2000} from isolated stars close to the DSO position. We aligned all the resulting median cubes of the four polarization channels of the individual observing years by using a cross-correlation algorithm. The image was restored by convolving the deconvolved image with a Gaussian beam of a FWHM of about 60 mas. For the 90-degree channel in 2008, 2009, 2011, and 2012 the resulting images are shown in Fig.~1. In all the years the DSO was clearly detected in its continuum emission in all channels taken with the NACO Wollaston prisms. To substantiate the detection of the DSO, we also show in the appendix the results of a high-pass-filtered (smooth-subtracted) image analysis in Fig. A.1 for 2008 and 2012. In all these images the DSO flux density contribution can clearly be identified as an individual source component. While possible PSF contributions have not been cleaned in the high-pass filtered images, we have conducted our analysis of the DSO emission using the Lucy deconvolved images. \begin{table* \caption[Galactic center Observations Log]{Galactic center observations log \centerin \begin{tabular}{c c c c c \hline \hline Date & Start time & Stop time & Number of frames& Integration time\\[0.5ex (DD.MM.YYYY)&(UT)&(UT)&&(sec)\\ \hlin \\ 25.05.2008 & 06:05:20.32 & 10:35:38.65 & 250 & 40\\ 27.05.2008 & 04:52:04.92 & 08:29:38.07 & 184 & 40\\ 30.05.2008 & 08:24:33.51 & 09:45:25.69 & 80 & 40\\ 01.06.2008 & 06:04:51.56 & 10:10:26.78 & 240 & 40\\ 03.06.2008 & 08:37:23.56 & 09:58:58.85 & 80 & 40\\ 18.05.2009 & 04:37:55.08 & 10:19:54.10 & 286 & 40\\ 27.05.2011 & 04:49:39.82 & 10:27:25.65 & 334 & 45\\ 17.05.2012 & 04:49:20.72 & 09:52:57.08 & 256 & 45\\[1ex \hlin \end{tabular} \label{table:log} \end{table*} \subsection{Position of the DSO} \label{dso-position} We obtained the position of the DSO in each year based on the Br$\gamma$ traced orbit from \cite{Meyer2014a} and \cite{Valencia2015}. Between 2004 and 2007, this source was confused with S63 and could not be distinguished from the background. From 2008 on, it starts to be clear and resolved from stellar confusion. The crowded region of the S-cluster may be the reason of this confusion \citep{sabha2012}. An example of the $K_\mathrm{s}$-band identification of the DSO in median polarimetry images for 2008 and 2012 is presented in Fig.~\ref{fig:allepochs-dso}. We used the data set presented in Table \ref{table:log} to investigate the flux density and polarimetry of DSO. \section{Analysis and results} \label{section:results} \subsection{Light curves} \label{section:lightcurves} \begin{figure}[b \centering \includegraphics[width=\columnwidth]{dso_total_newest_new.eps} \caption[Light curve of the DSO observed in polarimetry mode]{NIR $K_\mathrm{s}$-band light curve of the DSO observed in polarimetry mode in 2008, 2009, 2011, and 2012.} \label{fig:dso_alone} \end{figure} We measured the flux density of the DSO in each observing year by aperture photometry using 30~mas radius circular apertures. Flux calibration was made by aperture photometry using circular apertures of 40~mas radius and based on the known flux values of 13 S-stars. Eight background apertures were placed on an area devoid of individual sources to measure the background flux in the deconvolved images. We placed two of these background apertures very close to the DSO position. We used the same stellar calibrators as were presented in Table 1 in \cite{witzel2012} and Fig. 2 in \cite{shahzamanian2015b}. We would like to note that considering the proper background apertures is critical for an accurate flux density estimation of the DSO, since it is a faint source in the $K_\mathrm{s}$ band and background emission can have a prominent effect on its flux density value estimation. We added the photon counts in each aperture and then added the resulting values of two orthogonal polarimetry channels to obtain the total flux densities. Then we corrected them for the background contribution. This was done for the DSO and the calibrators close to it. The obtained flux densities were corrected for the extinction, that is, the $K_\mathrm{s}$-band extinction correction magnitude of 2.46 derived for the innermost arc-second by \cite{schoedel2010} was adopted. Subsequently, the light curve was obtained for 2008, 2009, 2011, and 2012, as presented in Fig. \ref{fig:dso_alone}. The uncertainties of the measured DSO flux densities in the light curve were estimated by obtaining the observational noise by setting ten apertures in different positions of the background close to the DSO position. The DSO does not show any intrinsic flux density variability in the $K_\mathrm{s}$ band based on our data set. The reason might be either a period of variability that does not match the time resolution of our data or an irregular variability. However, considering the limited data sample, any conclusion on the flux density variability and the effect of stellar contribution on the flux density estimation cannot be made at this point. \subsection{Polarimetry} \label{section:polarimetry} We derived the polarization degree and angle by obtaining normalized Stokes parameters ($I$, $Q$, $U$, $V$) from the observed flux densities,\\ \noindent \begin{equation} I = 1\\ \end{equation} \begin{equation} Q =\frac{f_{0}-f_{90}}{f_{0}+f_{90}} \end{equation} \begin{equation} U =\frac{f_{45}-f_{135}}{f_{45}+f_{135}} \end{equation} \begin{equation} V = 0,\end{equation} where $f_{0}$, $f_{90}$ , and $f_{45}$, $f_{135}$ are flux density pairs of orthogonally polarized channels. It is not possible to measure circular polarization in normalized Stokes $V$ since NACO is not provided with a quarter-wave plate. The circular polarization of stellar sources in the GC is very small \citep{Bailey1984}, therefore we assumed that it can be ignored and set to 0. However, in the case of dusty sources that have a high dust density, circular polarization may be produced by multiple scattering. Polarization degree $p$ and angle $\phi$ can be obtained as follows: \noindent \begin{equation}\label{eq:deg} p =\sqrt{Q^{2} + U^{2}} \end{equation} \begin{equation} \phi =\frac{1}{2}~\arctan\left(\frac{U}{Q} \right) ,\end{equation} \noindent \begin{table}[b \caption[Polarimetry results of the DSO]{Polarimetry measurements of the DSO \centerin \begin{tabular}{c c c \hline \hline Observing year & $p$ & $\phi$ \\[0.5ex \hlin 2008 & $30.14\%$ & $ -62.87^{\circ}$\\ 2009 & $32.6\%$ & $ 42.92^{\circ}$\\ 2011 & $29.9\%$ & $ ~18.125^{\circ}$\\ 2012 & $37.64\%$ & $ -9.67^{\circ}$ \\[1ex \hlin \end{tabular} \label{table:pol-dso} \end{table} where the polarization angle $\phi$ is measured from north to east and samples a range between 0$^\circ$ and 180$^\circ$ or -90$^\circ$ and 90$^\circ$. The instrumental polarization induced in the measured polarization can be corrected for by a model obtained from \cite{Witzel2011} that multiplies a combination of Mueller matrices for different elements of the telescope with the derived stokes vectors. We applied their model to correct for the instrumental polarization effects, where the measured polarization parameters can be estimated with an accuracy of $\sim1\%$ in polarization degree and $\sim5^\circ$ in polarization angle \citep{Witzel2011}. We show the results of the polarimetry measurements for our data set in Table \ref{table:pol-dso}. Based on the calculated values of $p$ and $\phi$ for different years, the DSO is polarized with a high polarization degree ($p>20\%$) and a polarization angle that changes as the source moves on its orbit. Figure \ref{fig:sim-4} presents a schematic illustration of the DSO polarization angle variation when the source moves on an eccentric orbit around the position of Sgr~A* before it reaches the peribothron. Standard error propagation for calculating the polarization uncertainties cannot be applied here since the errors on Q and U are not small compared to the measured values. Moreover, Q and U can be close to zero even if the flux density values in different polarimetry channels are large. In addition, the polarization degree and angle are mainly not Gaussian distributions. \\ \begin{figure}[!t] \centering \includegraphics[width=\columnwidth]{dso-ban-ok-referee.eps} \caption[A sketch showing the DSO polarization angle variation]{Sketch of the DSO polarization angle variation when it moves on its eccentric orbit around the position of Sgr~A* for four different years. The orange shaded areas show the range of possible values of the polarization angle based on our observation and the measurement uncertainties (see Table 3).} \label{fig:sim-4} \end{figure} In general, the effects that are responsible for producing NIR polarization in a GC source can be considered to be intrinsic to the source itself or to foreground effects caused by grain alignment along the line of sight, that is, dichroic extinction. Dichroic extinction can also play a significant role in polarization as a local effect for sources that are surrounded by an optically thick dust envelope \citep{Whitney2002}. Figure \ref{fig:stars-pol} presents the polarization degree and angle of the DSO compared to those of several S-stars located close to the DSO position obtained in this work. It also shows the GC foreground polarization from \cite{Buchholz2013}, which is $(6.1 \pm 1.3)\%$ at $20^{\circ}\pm 7^{\circ}$ in $K_\mathrm{s}$ band, widely parallel to the Galactic plane. Based on this figure, the measured polarization degree and angle of the DSO is not the foreground polarization and can be interpreted as intrinsic quantities.\\ In addition to the foreground polarization, some sources that are in the Northern Arm and the bow-shock sources have intrinsic polarization in $K_\mathrm{s}$ band as well as at longer wavelengths \citep{rauch2013}. The intrinsic polarization of the stellar sources can be produced by the following processes: emission from aligned dust grains, dichroic extinction, and scattering on spherical grains and/or magnetically aligned dust grains. It is still an open question which of these processes is responsible for producing the observed intrinsic polarization of GC sources. We discuss two of these processes that may produce the DSO polarization in Sect. 4: scattering and dichroic extinction.\\ \begin{figure* \begin{center} \subfigure{% \includegraphics[width=0.45\textwidth]{degree_stars.eps} } \subfigure{% \includegraphics[width=0.45\textwidth]{angle_stars.eps} }\\ \end{center} \caption[]{Left: Comparison of the polarization degree of the DSO (black dots) with those of GC S-stars located close to the DSO position (S7, S57, S19, S20, S40, S23, and S63; shown with X, asterisk, plus sign, diamond, triangle, violet point, and square, respectively). Right: Comparison of the polarization angle of the DSO (black dots) with those of the S-stars similar to the left panel. In both panels, some of the considered stars are not isolated in some years in which it is difficult to calculate their polarization parameters, therefore we did not show them as data points. The regions between two dashed red lines and dotted lines present the 1 and 3$\sigma$ confidence intervals of the $K_{\rm{s}}$-band polarization degree and angle distributions of the stars reported in \cite{Buchholz2013}, respectively. For the significance and uncertainties of the measured points see Sects.3.3 and 3.4. } \label{fig:stars-pol} \end{figure*} \cite{Buchholz2011} found that the $K_{\rm{s}}$-band PSF shows a polarized structure. The authors found that for the core of the PSF (see their Fig~B.2) this effect is basically negligible. Here, the variations of the intrinsic PSF polarization amount to a value of only about 1-2\%. The polarized nature of the PSF is of some importance for extended structures. We find that the continuum flux distribution of the DSO is unresolved. However, it is moving within the central stellar cluster. \cite{Buchholz2011} reported that on the first and second Airy ring (see their Fig.~B.2) polarization degrees of up to 15\% are possible. At this location the flux density levels correspond to less than about 3\% of the peak for a Strehl ratio of 50\%\ in 0.5'' seeing. In this case the polarized contributions of these neighboring sources are lower than 2\% of the DSO peak flux. In 2009 and 2011 the DSO is at even larger distances from similarly bright sources, hence, the expected polarized flux contributions in these years is even lower than in 2008. In 2012 the distance of the DSO to two sources that are about 50 times brighter is close to the second Airy ring with flux contributions well below 1\%. Hence, the polarized flux contribution from other sources is well below 15\%. This implies that in all years the polarization estimates are to more than 85\% (even more than 98\% in 2008, 2009, and 2011) dominated by the DSO itself. Therefore, we did not consider PSF polarization variations in our present investigation. Assuming the DSO has a bow-shock structure, a comprehensive model of bow~shock polarization using observation results can enable us to analyze the influence of different polarization scenarios (emission, scattering, and dichroic extinction) that play a role in producing the intrinsic polarization of a bow~shock. \\ \subsection{Statistical significance of the measured values} The polarization degree is a positive defined quantity that becomes biased to higher values (than the intrinsic $p$ of the source) at low $S/N$ measurements. For this reason, it is possible to measure polarization degrees that are unphysically higher than 100\% due to the observational uncertainties. Thus, it is important to keep in mind that the measured (or observed) polarization degree and angle of any astrophysical object are a good estimation of the source intrinsic properties only when the $S/N$ is high. The DSO $K_\mathrm{s}$-band emission from the yearly stacked images analyzed here have a medium $S/N$ of $\sim6-10$ in 2008, 2009, and 2012 and a low $S/N \sim2.5$ in 2011. In the first three cases, the distribution of the observed polarization degree can be described by a Rice function \citep{serkowski1958, vinokur1965}; while in the latter, even this function fails to properly model the high skewness of the polarization degree distribution \citep{simmons1985}. In turn, the observed polarization-angle distribution depends on the flux-density $S/N$ and on the intrinsic $p$ of the source. \begin{figure*}[] \centering \includegraphics[width=1.\textwidth]{newfig5.eps} \caption{Polarization degree distribution for the observations in 2008, 2009, 2011, and 2012 when considering the null hypothesis ($p=0$\%, $\phi=0^{\circ}$), the observed flux density $I,$ and its uncertainty. The most probable observed polarization degree, $p_{\rm obs}$ in each case is the mode of the distributions. The range that contains 68\% of the most probably observed values, $1 \sigma p_{\rm obs}$ is delineated with vertical dashed lines. The smooth line over the histograms corresponds to the best-fit Rice function. } \label{fig:figjj} \end{figure*} To estimate the significance of the measured DSO polarization, we assumed as null hypothesis that the source is intrinsically not polarized, meaning that the observed polarization angle $\phi_{\rm obs}$ and degree $p_{\rm obs}$ are solely produced by the observational uncertainties. For this test, we therefore calculated $f_0$, $f_{90}$, $f_{45}$, and $f_{135}$ from the observed total flux density $I_{obs}$ using $p=0$\% and $\phi=0^{\circ}$ and assuming the intrinsic flux density to be the same as the observed $I=I_{obs}$. We simulated the effect of the noise in the polarization channels by adding a randomly selected amount to each channel that was drawn from a normal distribution with $\sigma(f_{X}) = 0.7\,\sigma(I)$, where $\sigma(I)$ is the uncertainty of the observed total flux density. Using a Monte Carlo approach, we calculated 10\,000 times the observed polarization degree and angle for such a setup using Eqs.\,(1)-(6). The normalized histograms of $p_{\rm obs}$ for the observations in 2008, 2009, 2011, and 2012 are shown in Fig.~\ref{fig:figjj}. These distributions represent the probability of measuring certain polarization degrees under the null hypothesis (i.e., the source is intrinsically not polarized) given the previously estimated $I$ and their uncertainties. Figure \ref{fig:figjj} shows that in 2008, where $S/N\approx10$, the most probable observed polarization degree is $\sim 9$\% with 68/100 of the possible measurements within $(4.3, 17.0)$\%. Under these conditions, measuring a DSO polarization degree higher than $30$\% when the source is intrinsically not polarized is only probable in less than 1/100 cases. In 2009 and 2012, where $S/N\approx6$, the probability of finding an intrinsically unpolarized source of the same flux density as the DSO, and $p_{\rm obs} \geq 35$\% is $\sim 10/100$ cases. Only in 2011, the low $S/N$ broadens the distribution of $p_{\rm obs}$, implying that the intrinsic polarization degree cannot be estimated in this data set. In all tests of the null hypothesis, the observed polarization angle $\phi_{\rm obs}$ can be any value between $-90^{\circ}$ and $90^{\circ}$ because the intrinsic polarization degree was set to zero. We conclude that three of the measurements of the DSO polarization degree are statistically significant at the confidence levels 0.99 in 2008, and 0.9 in 2009 and 2012. Assuming as starting values for the simulation $p=6$\% and $\phi=20^{\circ}$, i.e. an intrinsically unpolarized source located at the GC, which, therefore, will display $p$ and $\phi$ values equal to that of the foreground polarization, the significance levels of the polarization measurements change to 0.98 for the 2008, and to 0.88 for 2009 and 2011. This means that intrinsic polarization of the DSO has been detected in 2008, 2009, and 2012. \begin{figure*}[] \centering \includegraphics[width=1.\textwidth]{FigUncer2.eps} \caption{Left: Marginalized distributions of intrinsic polarization degree and angle, assuming that the observed values are the most probable ones. Here the case of 2008 data is shown. Right: Combined probability obtained from the two distributions at the left. The bottom and right panels show transversal cuts to the combined distribution at the position of the marked lines. } \label{fig:figuncer} \end{figure*} \subsection{Measurement uncertainties} After establishing the statistical significance of the polarization measurements of three observed epochs, our next goal is to estimate the uncertainty of the intrinsic parameters given the observed values. Several measurements of an intrinsically non-variable source would be ideal to assess the uncertainties in the polarization quantities. However, it is unknown whether the DSO is variable or not, and we can only rely on the stacked images of three years. Our alternative approach consists of using the Monte Carlo simulations presented before to estimate the range of most probable intrinsic values that would yield the observed DSO polarization. We sampled the whole range of possible DSO intrinsic polarization degree and angle in a grid of $p \in [0.0, 0.7]$ in steps of 0.005, and $\phi \in [-90^{\circ},90^{\circ}]$ in steps of $2^{\circ}$, calculating each case 10\,000 times. In this way, we constructed two data cubes. One of them has in the first axis the intrinsic polarization degree $p$, in the second axis the intrinsic polarization angle $\phi$, and in the third the observed polarization degree $p_{\rm obs}$. The value of $i{\rm-th}$ cell corresponds to the probability of observing $p_{{\rm obs}, i}$ when the source has intrinsic $p_i$, $\phi_i$, and the observed flux density (with its uncertainty). Similarly, the second cube has as axes $p$, $\phi,$ and $\phi_{\rm obs}$, and any $i{\rm-th}$ cell equals the probability of observing $\phi_{{\rm obs},i}$ given $p$, $\phi$ and $I \pm\sigma(I)$. From the sheets of $(p, \phi)$ at a constant $p_{\rm obs}$ in one cube, and at a constant $\phi_{\rm obs}$ in the other, we can deduce the range of intrinsic polarization values that most probably yield the observed the polarization degree and angle. Figure~\ref{fig:figuncer} shows the $(p, \phi)$ sheets for $p_{\rm obs}=30\%$ and $\phi_{\rm obs}=-63^{\circ}$, as is the case for the DSO in 2008. As expected, the probability of measuring certain $p_{\rm obs}$ is independent of the intrinsic polarization angle of the source and peaks at the most probable intrinsic $p_{\rm best}$ -- which does not necessarily equal $p_{\rm obs}$. In contrast, the probability of observing a particular $\phi_{\rm obs}$ depends on the intrinsic polarization degree and is symmetric with respect to $\phi$. The most probable intrinsic polarization angle is the same as the observed value $\phi_{\rm best} = \phi_{\rm obs}$. The 68\% ranges of intrinsic values that most probably produce the observed ones, $\sigma(p)$ and $\sigma(\phi)$ were calculated from the figures making transversal cuts at the positions of the most probable intrinsic $p_{\rm best}$ and $\phi_{\rm best}$, as shown in Fig.~\ref{fig:figuncer} (right). The results for all years are compiled in Table 3. Under the assumption that the DSO polarization degree is approximately constant over the studied years, the significance of the polarization degree measurement is larger than $1/100/,000$. The values for 2011 are reported in the table for completeness, but because of the low $S/N$ of the flux-density measurement that year, they cannot be interpreted as intrinsic properties of the source. \begin{table} \caption{Most probable DSO intrinsic polarization values.} \label{tyy} \centering \begin{threeparttable} \begin{tabular}{c c c c c c} \hline\hline Year & $I \pm \sigma(I) /{\rm mJy}$ & $p_{\rm obs}$ & $\phi_{\rm obs}$ & $p_{\rm best} \pm \sigma(p)$ & $\sigma(\phi)$ \\ \hline 2008 & $0.25 \pm 0.02$ & $30.1$ & $117.1$ & $28_{-10}^{+10}$ & $9$ \\ 2009 & $0.19 \pm 0.03$ & $32.6$ & $42.9$ & $28_{-10}^{+16}$ & $15$ \\ 2011\tnote{a} & $0.23 \pm 0.09$ & $29.9$ & $18.1$ & $0.5_{-0.5}^{+38}$ & $58$\\ 2012 & $0.25 \pm 0.05$ & $37.6$ & $170.3$ & $32_{-18}^{+18}$ & $15$ \\ \hline \end{tabular} \begin{tablenotes} \item[a] These values cannot be interpreted as intrinsic because of the low flux-density $S/N$ obtained this year. \end{tablenotes} \end{threeparttable} \end{table} \section{DSO model: a young, supersonic star} \label{dso-model} Two mechanisms are responsible for the origin of the polarized flux density: Mie scattering, that is, photon scattering on spherical grains, and dichroic extinction, or selective extinction of photons due to the non-spherical shape of dust grains. Using the dust continuum radiative transfer modeling, we focused on the process of Mie scattering. The model of the DSO was required to meet the observed characteristics: total SED constraints--integrated flux densities in H, $K_{\rm{s}}$, L', and M bands, total polarization degree higher than $\sim 20^{\circ}$, and changes of polarization degree and angle along the orbit.\\ We considered different circumstellar geometries to assess whether they can explain the detected polarized signal and the significant color excess in NIR bands. To test the properties of different constituents, we performed a series of dust continuum radiative transfer simulations using the Monte Carlo code Hyperion \citep{2011A&A...536A..79R}. First, we considered only a stellar source and a bow shock layer in the calculation, which did not yield a significant polarized emission. Adding a spherical dusty shell around a stellar source provided enough extinction to match the NIR excess, but the total polarization degree remained below $10\%$. Finally, we added a flared disk with bipolar outflows into the density distribution, which caused the source to deviate more from the spherical geometry (see Fig. \ref{fig:fig_model_DSO}). The polarized emission in this composite model (star+flared disk+bipolar outflows+bow shock) is dominated by dust and stellar emission that is scattered on spherical dust grains. The largest contribution of polarized flux density originates in the region where the outflow intersects the dense bow-shock layer. An overview of different possible circumstellar geometries is provided in Table~\ref{table:geometry}, where the values of the total linear polarization degree in $K_{\rm s}$ band and the color excess $K_{\rm s}-L'$ are included. A high value of the total polarization degree in $K_{\rm s}$ band and an intrinsic NIR excess between $K_{\rm s}$ and $L'$ NIR bands are matched by the composite model star+disk+cavities+dense bow shock, whose properties are studied in more detail. This scenario resembles proplyd-like objects, which seem to be present in the Sgr~A West region \citep[see][for VLA continuum observations]{2015ApJ...801L..26Y}. Therefore, the DSO might be the first detected and monitored young stellar object in the vicinity of the SMBH that would manifest a very recent low-mass star-formation event close to the GC. \begin{table*}[] \centering \resizebox{\textwidth}{!}{ \begin{tabular}{c c c c} \hline \hline \\ \textbf{Geometry} & \textbf{$K_{\rm s}$-band Total Linear Polarization Degree} $\overline{p}_{\rm K,L}$ [\%] & $\mathbf{(K_{\rm s}-L')_{\rm int}}$ (intrinsic) & $\mathbf{(K_{\rm s}-L')_{\rm ext}}$ (with dust extinction)\\ \\ \hline \\ Star & $0$ ($\sim 6\%$ foreground pol. at the GC) & $-0.9$ & $0.4$ \\ Star+rotationally flattened envelope ($50^{\circ}$ inclination) & $0.2$ & $0.1$ & $1.4$ \\ Star+flared disk & $3.2$ & $0.3$ & $1.6$\\ Star+dense bow shock (inclined) & $4.1$ & $1.9$ & $3.2$\\ Star+spherical dusty envelope+dense bow shock (inclined) & $1.0$ & $1.6$ & $2.9$\\ Star+flattened envelope+cavities ($90\%$ inclination) & $10.1$ & $1.2$ & $2.4$\\ Star+flared disk+flattened envelope+cavities ($90\%$ inclination) & $10.5$ & $-1.3$ & $-0.03$\\ Star+disk+cavities+dense bow shock & $25.0$ & $1.9$ & $3.2$\\ \\ \hline \label{table:geometry} \end{tabular} } \caption{Different circumstellar geometries with a list of constituents. Important parameters are the total linear polarization degree in $K_{\rm s}$ band $(2.2\,\rm{\mu m})$ and the color index $K_{\rm s}-L'$. The observed values of the degree and the color index are matched by the composite model star+disk+cavities+dense bow shock that is used in the following analysis.} \end{table*} \subsection{Monte Carlo radiative transfer simulations} We performed 3D Monte Carlo radiative transfer calculations using the code Hyperion \citep{2011A&A...536A..79R} to study under which conditions the model of a supersonic, young stellar object would yield the observed DSO properties. The advantage of the Monte Carlo approach is that it is suitable for arbitrary 3D geometry (stellar source surrounded by circumstellar envelope and bow~shock), and we automatically obtain a full Stokes vector $(I,Q,U,V)$, which allows us to easily calculate the total and the polarized flux density. In our simulation runs we set up a spherical polar grid containing $400\times 200 \times 10$ grid points and subsequently added a density grid of gaseous-dusty mixture with a gas-to-dust ratio of $100:1$. A power-law distribution with a slope of $-3.5$ \citep{1994ApJ...422..164K} of spherical dust grains with a radius of between $0.01$ and $10\ \mu\rm{m}$ was considered. The distance to the GC was taken to be $8\,\rm{kpc}$. \begin{figure}[!t] \centering \includegraphics[height=5cm,width=\columnwidth]{bow_shock2_new.eps} \caption{Schematic view of the proplyd-like model of the DSO structure. See the text for a description of the model components.} \label{fig:fig_model_DSO} \end{figure} The density of the circumstellar envelope was assumed to have a simple radial profile, $\rho_{\rm{disc}}=\rho_0(r/1\,\rm{AU})^{-1}$ with $\rho_{0}=10^{-14}\,\rm{g\,cm^{-3}}$, and the bipolar cavities were assumed to have an opening angle of $90^{\circ}$ and a uniform density of $10^{-20}\,\rm{g\,cm^{-3}}$ \citep{Sanchez2016}. For the bow-shock component and its evolution, we considered the treatment of \citet{Zajacek2016} for the DSO based on the analytical axisymmetric model of \citet{wilkin1996}, keeping the same density and temperature profiles for the hot corona around Sgr~A*. A two-dimensional bow-shock shape is obtained by the model of \cite{wilkin1996} \noindent \begin{equation}\label{eq:bowmodel} R(\theta) = R_{0}\csc\theta\sqrt{3(1 - \theta\cot\theta)}, \end{equation} \noindent where $\theta$ is the polar angle from the axis of symmetry, as seen by the star at the coordinate origin. $R_{0}$ is the so-called stand-off distance obtained by balancing the ram pressures of the stellar wind and ambient medium at $\theta = 0,$ and it is given by \noindent \begin{equation}\label{eq:offdistance} R_{0} =\sqrt{\frac{\dot{m_{w}}v_{w}}{\Omega \rho_{a}v_{\star}^{2}}} .\end{equation} \noindent Here, $\dot{m}_{\rm w}$ is the stellar mass-loss rate, $v_{\rm w}$ the terminal velocity of the stellar wind, $\rho_{\rm a}$ the ambient medium mass density where $\rho_{\rm a} = m_{\rm H}n_{\rm H}$ (with $n_{\rm H} \approx n_{\rm e}$), and $m_{\rm H}$ the mean molecular weight of hydrogen, and $v_{\star}$ is the velocity at which the star moves through the medium (i.e., the relative velocity of the star with respect to the ambient medium when the ambient medium is not stationary). The solid angle $\Omega=2\pi(1-\cos\theta_0)$, where $\theta_0$ stands for the half-opening angle of the bipolar outflow, represents the direction into which the stellar wind is blown, outside $\Omega$ there is no outflow \citep{1997ApJ...474..719Z}. For an isotropic stellar wind, $\theta_0=\pi$, the solid angle is naturally $\Omega=4\pi$, whereas for our non-spherical bipolar stellar outflow, $\theta_0=\pi/4$, and the solid angle is then reduced to $\Omega=2\pi(1-\sqrt{2}/2)\approx 1.84$. The bow~shock consists of two layers: (a) a hot and sparse forward shock, and (b) a cold and denser reverse shock \citep{Scoville2013}. We considered only the colder and denser layer, whose density is about four orders of magnitude higher than for the hot shock \citep{Scoville2013}, and therefore it is more significant in terms of scattering on spherical dust particles. The initial value for the bow-shock density is set to $10^{-16}\,\rm{g\,cm^{-3}}$, in accordance with \citet{Scoville2013} for the distance of $10^{16}\,\rm{cm}$ from Sgr~A*. For the typical parameters of a young star potentially associated with the DSO, which means a mass-loss rate of $\dot{m}_{\rm{w}}=10^{-8}\,\rm{M_{\odot}\,yr^{-1}}$ and a wind velocity of $v_{\rm{w}}=100\,\rm{km\,s^{-1}}$ , the stand-off distance is of about $10\,\rm{AU}$ \citep{Zajacek2016}, which sets the basic length-scale of the model. The typical angular scale of such an object is $1\,\rm{mas}$ at the distance of the GC, in accordance with the compactness of the DSO observed along the orbit \citep{Valencia2015}. The outer radius of the circumstellar envelope was set to $3\,\rm{AU}$ and the inner radius was set to the typical dust sublimation radius \citep{2002ApJ...579..694M}, \begin{equation} R_{\rm{sub}}=1.1 \sqrt{Q_{\rm{R}}}\left(\frac{L_{\star}}{1000\,L_{\odot}}\right)^{1/2} \left(\frac{T_{\rm{sub}}}{1500\,\rm{K}} \right)^{-2}\,\rm{AU}\,, \end{equation} which for the upper limit on the luminosity of the DSO $L_{\star}=30\,L_{\odot}$ and a ratio of dust absorption efficiencies $Q_{\rm{R}}=1$ leads to the estimate of $R_{\rm{sub}}\approx 0.1\,\rm{AU}$. \subsection{Comparison with observations} The described model of a supersonic, young stellar object with the adopted density distribution can reproduce the inferred flux densities in $H$, $K_{\rm{s}}$, $L'$, and $M$ NIR bands: $F_{H}\leq0.14$~mJy, $F_{K}\sim 0.3$~mJy, $F_{L}\sim 1.9$~mJy, and $F_{M} \sim 2.8$~mJy, respectively \citep{Eckart2013,Witzel2014,Gillessen2012}. In Fig.~\ref{fig_sed} we compare the observed flux densities with the best-fit SED obtained from simulations. \\ \begin{figure} \centering \includegraphics[width=\columnwidth]{bowshock_shell_sed.eps} \caption{SED for the young stellar object described in the text at the GC. The points label the flux densities inferred from the observations.} \label{fig_sed} \end{figure} The obtained proper motion of the DSO between 2008 and 2009 is $v_{\alpha} = (-400)\mathrm{km\,s^{-1}}$, $v_{\delta} = (800)\mathrm{km\, s^{-1}}$ and between 2011 and 2012 it is $v_{\alpha} = (-2000)\mathrm{km\, s^{-1}}$, $v_{\delta} = (600)\mathrm{km\, s^{-1}}$. Therefore, $v_{\star}$ changes from $894.42\mathrm{km\, s^{-1}}$ to $2088.06\mathrm{km\, s^{-1}}$. The uncertainty of the proper motion is of about $1\mathrm{mas\, yr^{-1}}$ \citep{eckart2012a}. Using the relative velocity values from 2008 to 2012 in Eq. \ref{eq:offdistance}, the stand-off distance becomes about half of its value \citep[see also Fig. 3 in][]{Zajacek2016}. Moreover, the density profile of the central region based on X-ray data derived in \cite{shcherbakov2010} indicates that the particle number density increases within the central arcseconds from the SMBH. \\ Owing to the increasing orbital velocity of the DSO and higher ambient pressure toward the pericenter, the bow~shock shrinks and becomes denser. An increment by a factor of four and potentially even more in the bow-shock number density is expected between 2008 and 2012 under the assumption that the bow-shock mass stays approximately constant. This leads to the progressively higher scattering from the non-spherical bow shock for smaller distances from the SMBH. The increase in the density of the bow-shock layer leads to the progressive increase in polarization degree from an initial $30\%$ in 2008 to almost $40\%$ in 2012, which can be reproduced by radiative transfer simulations of the dust continuum. The observed change in the polarization angle $\Phi$ is given by the combination of the \textit{\textup{external}} factors (i.e., motion of the DSO through the external accretion flow) and/or the \textit{\textup{internal}} factors (geometry of the circumstellar environment: disk, bipolar wind, and the bow~shock). Using the Monte Carlo radiative transfer calculations, we investigated the effect of the change in orientation of the bipolar wind with respect to the symmetry axis of the bow~shock. The angle between the symmetry axis of the bow~shock and the axis of the bipolar wind is denoted by $\delta$. In our simulations, we gradually increased $\delta$, starting at $0^{\circ}$ (bipolar wind aligned with the bow-shock axis), and we stopped at $90^{\circ}$ (bipolar wind perpendicular to the bow-shock axis). The increment is $10^{\circ}$ , and we performed the simulations in both clockwise and counter-clockwise directions. In Fig.~\ref{img_delta} we plot the dependency $\Phi=f(\delta)$ for the two directions (upper and lower part). In the calculations the bow shock lies in the orbital plane and we observe it from above. The calculated values are represented by points and the lines stand for linear interpolation (blue and orange lines) and the linear regression (green dashed lines). The basic trend in Fig.~\ref{img_delta} is the following: for the bipolar wind aligned with the bow-shock axis, $\delta=0^{\circ}$, the polarization angle reaches values $\pm 90^{\circ}$; when the angle $|\delta|$ deviates from zero and approaches $90^{\circ}$, the polarization angle approaches zero. Therefore, the change of the orientation of the bipolar wind covers the whole range $(-90^{\circ},+90^{\circ})$. The fitted linear relations are $\Phi_1=-0.97(\pm 0.06)\delta+84(\pm 3)$ (clockwise rotation) and $\Phi_2=1.02(\pm 0.07)\delta-88(\pm 4)$ (counter-clockwise direction), which in principle may be approximated in the following way: $\Phi \approx -(+) \delta +(-) 90$. The change in orientation of the bipolar wind takes place as a consequence of the torques induced by the supermassive black hole, which leads to its precession when the circumstellar disk is misaligned with the orbital plane. This naturally affects the inclination of the outflow that originates in the disk, and the precession takes place on the precession timescale, which is longer than the orbital timescale, $T_{\rm{prec}}>T_{\rm{orb}}$. The wobbling of the disk occurs on timescales shorter than one orbital period, approximately $T_{\rm{wob}}\sim 1/2\,T_{\rm{orb}}$ \citep{2000MNRAS.317..773B}. \begin{figure}[!t] \centering \includegraphics[width=\columnwidth]{bowshock_shell_delta_illustration.eps} \caption{Linear polarization angle as a function of bipolar wind orientation $\delta$. The plot consists of two parts: the upper part shows the dependence for the clockwise rotation of the bipolar wind; the lower part is the same dependence for the counter-clockwise rotation of the bipolar wind. The lines correspond to the linear interpolation of the data; linear regression lines are depicted by green dashed lines.} \label{img_delta} \end{figure} In addition, the change in polarization angle may be the combination of the change in internal geometry of the outflow and the external interaction of a supersonic star with an ambient wind with a certain velocity field. When the bipolar wind is aligned with the symmetry axis of the bow~shock, the total polarization angle is perpendicular to the bow~shock symmetry axis. Subsequently, when the supersonic star interacts with the external outflow or inflow, the bow-shock orientation changes because of the change in relative velocity. This would naturally lead to the corresponding alternation of the polarization angle. \\ For completeness, we also produce the images of scattered emission in $K_{\rm{s}}$ band, maps of linear polarization degree, and the distribution of the polarization angle for the configurations with $\delta=0^{\circ}$, $\delta=45^{\circ}$, and $\delta=90^{\circ}$ (see Fig.~\ref{fig_linpol3}). Most of the polarized flux density originates in the bow-shock layer where the bipolar stellar wind intercepts the bow shock. The change in tilt of the bipolar cavity naturally leads to the modification of the polarization angle. All in all, the main observed characteristics of the DSO, that is, compactness, the total flux density, the polarization degree and angle, and their corresponding changes, can be reproduced by the model of a supersonic YSO. A high polarization degree in $K_{\rm{s}}$ band is due to the scattering of stellar and dust photons, and the main contribution is due to the self-scattering of dust emission. The reconstructed RGB image of the DSO model from the radiative transfer is shown in Fig.~\ref{fig_rgb_dso}, where blue stands for the $K_{\rm{s}}$ band, green for the $L'$ band, and red for the $M$-band emission. \begin{figure} \centering \includegraphics[width=\columnwidth]{bowshock_pol_deg_angle_delta0.png} \includegraphics[width=\columnwidth]{bowshock_pol_deg_angle_delta45.png} \includegraphics[width=\columnwidth]{bowshock_pol_deg_angle_delta90.png} \caption{Emission map of scattered light in $K_{\rm{s}}$ band, the distribution of the polarization degree, and the angle in the left, middle, and right panels, respectively, for three different configurations of the star--outflow system: $\delta=0^{\circ}$, $45^{\circ}$, and $90^{\circ}$ from top to bottom.} \label{fig_linpol3} \end{figure} \subsection{Effect of dichroic extinction} A fraction of the polarized flux density of the DSO may arise from dichroic extinction. Dust grains might not be necessarily spherical, but they might be cylindrical and aligned to a certain extent by a magnetic or radiation field. A significant polarized signal may then be created by dichroic extinction: photons with electric vectors that are parallel to the grain axis experience higher extinction than those with electric vectors that are perpendicular. Therefore, a polarized emission emerges with a polarization degree $p$ that may be expressed in the following way \citep{krugel2002physics}: \begin{equation} p=\frac{e^{-\tau_{\rm{min}}}-e^{-\tau_{\rm{max}}}}{e^{-\tau_{\rm{min}}}+e^{-\tau_{\rm{max}}}}\,, \end{equation} where $\tau_{\rm{max}}$ is the highest optical depth in the direction of the strongest attenuation and $\tau_{\rm{min}}$ is the lowest optical depth in the perpendicular direction. The difference is usually small, $\tau_{\rm{max}}-\tau_{\rm{min}}\rightarrow 0$, which then leads to \begin{equation} p=\frac{1}{2}(\tau_{\rm{max}}-\tau_{\rm{min}})=\frac{1}{2} \sigma_{\rm{gr}}(Q_{\rm{ext}}^{\rm{max}}-Q_{\rm{ext}}^{\rm{min}})\int_{0}^{l} n(l) \mathrm{d}l\,, \label{eq_dichroic_extinction} \end{equation} where $\sigma_{\rm{gr}}$ is the effective cross-section of the grain, $Q_{\rm{ext}}^{\rm{max}}$ and $Q_{\rm{ext}}^{\rm{min}}$ are the highest and lowest extinction efficiencies, and the integral $N=\int_{0}^{l} n(l) \mathrm{d}l$ stands for the column density of dust along the line of sight. Since the polarization degree $p$ in Eq.~\ref{eq_dichroic_extinction} is proportional to the column density, which tends to increase toward the pericenter as a result of the increasing surface density of the bow-shock layer (see Zaja\v{c}ek et al., 2016, for detailed calculations), the increase in polarization degree toward the pericenter (2008--2012) might be partially caused by dichroic extinction. \begin{figure}[tbh] \centering \includegraphics[width=0.75\columnwidth]{bowshock_rgb_blurred.png} \caption{RGB image of the DSO source model. Blue stands for the $K_{\rm{s}}$ band, green for the $L'$ band, and red for the $M$-band emission.} \label{fig_rgb_dso} \end{figure} \section{Discussion} \label{section:discussion} The obtained polarization degree of the DSO in different observing years is significantly higher than the foreground polarization ($6.1\%$ in $K_\mathrm{s}$ band), which shows that it is an intrinsically polarized source. Emission or absorption from elongated grains that are aligned magnetically and (Mie-) scattering on dust grains in the dusty stellar envelopes are the most probable processes to produce the intrinsic polarization. In the NIR, emission and absorption processes can take place at the same time and cancel each other out. When scattering on the spherical dust grains occurs, part of the light that is scattered forward is unpolarized, and the rest would be scattered perpendicular to the direction of incident light, which produces linear polarization. If a central star is embedded in an isotropic envelope of spherical grains, the scattered light produces polarization with polarization vectors tangential to concentric circles about the central star that causes the overall polarization to be zero. Polarization patterns can be more complex for more complex geometries such as YSOs that have dusty disks around the central source \citep[see, e.g.,][]{murakawa2010}, and bow-shock sources. In a system of a star that has dusty disk around it, the polarization values of different regions around the source cancel each other out when the source is viewed face-on and no polarization would be detected, while when it is viewed edge-on, the source shows polarization \citep{Buchholz2011}. Scattering can also occur on elongated and non-spherical dust grains and has been modeled \citep[see, e.g.,][]{Whitney2002}, although in our radiative transfer model we assumed only spherical dust grains. All in all, polarization produced by light scattering depends on the source viewing angle and its geometry. Since the measured polarization degree of the DSO that we obtain is high ($p>20\%$), it might be caused by non-spherical geometry, for example, by a star with a bipolar wind and a bow~shock. In general, if the DSO is moving through the hot, X-ray emitting region of the interstellar medium with a supersonic velocity, a bow~shock can form. Collisional ionization and high densities in the shocked layer of the stellar wind can generate its observed emission lines \citep{Scoville2013}. However, no increase in the Br$\gamma$ luminosity has been detected, which would be expected for the bow-shock dominated emission \citep{Valencia2015}. \\ The polarization angle varies during the years, that is to say, the direction of the polarization vector changes from its expected direction, which is perpendicular to the direction of motion. The modification of the polarization angle can be caused by the change in orientation of the bipolar wind of the star that occurs when the circumstellar disk is not aligned with the orbital plane. It can also be explained by a combination of the source motion and the interaction with the surrounding medium. T~Tauri stars generally exhibit an intrinsic polarization, and one of the main origins of their polarization is scattering by dust grains in circumstellar shells \citep[see, e.g.,][]{yudin1998}. In polarimetric images, they show variations in polarization degree and angle \citep{Appenzeller1989}. Many T~Tauri stars have magnetospheric accretion flows since their circumstellar magnetic fields are powerful and globally ordered to maintain this type of large-scale accretion flows \citep{Symington2005}.\\ Some circumstellar disks of the low-mass stars have a bow-shock appearance like the proplyds (protoplanetary disks) of the Orion nebula cluster \citep{o'dell1994, Johnstone1998, Storzer1999}. NIR polarimetry is a tool to constrain the disk parameters of the proplyds even if the disk structure itself is not resolvable \citep{rost2008}. Based on our data set, we cannot spatially resolve a disk or a bow-shock structure because the source appears as a point source with the current angular resolution of 8~m class telescopes in $K_\mathrm{s}$ band. It might be similar to X3 and X7, the bow-shock sources within the inner $5 ''$ at the GC \citep{muzic2010}. Moreover, as the source is not extended, it is not possible to measure the polarization in individual regions of the source, as was done in \cite{rauch2013} for IRS 8. \\ \section{Summary and conclusions} \label{section:summary} We have analyzed the NIR polarized observations of a Dusty S-cluster Object (DSO/G2) on an eccentric orbit around Sgr A*. $K_\mathrm{s}$-band polarization data were available for 2008, 2009, 2011, and 2012. In all these years we clearly detected $K_\mathrm{s}$-band continuum emission in the different channels of the Wollaston prism. The data cover the polarization information of the DSO before its pericenter passage (May 2014). The source does not show significant variability in the overall $K_\mathrm{s}$-band flux density during the observed years, and its polarization degree is mostly above 20\%, which is higher than the foreground polarization measured on the surrounding stars. It appears that the polarization degree is approximately constant and the polarization angle varies as it approaches the position of Sgr A*. Based on our significance analysis, the polarization measurements of the DSO are significant in 2008, 2009, and 2012 and can be interpreted as source-intrinsic properties.\\ Since the total polarization degree is noticeably high, higher than $20\%$ for all epochs, the DSO structure is expected to deviate from the spherical symmetry. Moreover, the analysis of \citet{Valencia2015} showed that the source remains compact, meaning that it is not effected by tidal forces close to the pericenter of its orbit. \citet{Gillessen2012} and \citet{Eckart2013} discussed a NIR excess of $K_{\rm{s}}-L' > 3$, which implies the presence of a dense gaseous-dusty envelope. All of these basic observed parameters, the NIR excess, the compactness, and the significant polarization, may be reconciled within the model of a dust-enshrouded young star, to be precise, a pre-main-sequence star of class 1 \citep{Zajacek2014,2015wds..conf...27Z}, that forms a dense bow-shock layer by its supersonic motion upon approaching the supermassive black hole \citep{Zajacek2016}. The obtained polarization properties of the DSO in this work can be caused by the non-spherical geometry of a bow~shock and a bipolar wind of the star. We used the 3D radiative transfer model implementing the code Hyperion \citep{2011A&A...536A..79R} to compare the observed measurements to the model of the young stellar object forming a bow~shock. We conclude that the varying polarization angle is related to the intrinsic change of the circumstellar configuration. The change in bipolar outflow orientation may be due to the accretion disk wobbling or precession in the gravitational field of the SMBH. It can be also produced by external interaction of the DSO with the accretion flow. Although our model is simple, it can reproduce many observed properties of the DSO obtained in this work, such as the total flux density and the polarization degree. A more detailed analysis of the model will be provided in Zaja\v{c}ek et al., in prep.\\ \cite{shahzamanian2015b} showed that the Sgr~A* system exhibits a stable geometry and accretion process that is consistent with the preferred jet or wind directions. The close fly-by of the DSO, or similar dusty sources (Peissker et al., in prep.), might have an effect on the stable accretion flow onto Sgr~A* that depends on the nature of these objects. However, after the pericenter passage of the DSO, the object remained compact and its orbit Keplerian \citep{Witzel2014,Valencia2015}. Consequently, it did not lose a noticeable amount of energy and angular momentum during its closest approach to Sgr~A*, and as a result experienced weak interactions with the central black hole \citep{park2015}. However, based on hydrodynamical simulations, it may take several years for their interaction and to see a change in the activity of Sgr~A* \citep{burkert2012, schartmann2012}, either as an increase in the accretion flow rate or in the appearance of jets \citep{yuan2014}. Therefore, polarization and variability measurements of Sgr~A* are needed to be continued as they are the ideal tool to probe any change in the apparently stable system as a function of the DSO fly-by. Moreover, future polarized observations of the DSO, that is, after the pericenter passage, in the NIR can help us to better constrain the source polarization and structure.\\ \begin{acknowledgements} The authors would like to thank the anonymous referee for the helpful comments on this paper. We would like to thank G. Witzel for fruitful discussions. This work was supported in part by the Deutsche Forschungsgemeinschaft (DFG) via the Cologne Bonn Graduate School (BCGS) and the Max Planck Society through the International Max Planck Research School (IMPRS) for Astronomy and Astrophysics. B. Shahzamanian has been supported by IMPRS and the BCGS. N. Sabha has been supported by BCGS. M. Zajacek and M. Parsa are members of the IMPRS. Part of this work was supported by fruitful discussions with members of the Czech Science Foundation DFG collaboration (No. 13-00070J). We also received funding from the European Union Seventh Framework Program (FP7/2007- 2013) under grant agreement n312789; Strong gravity: Probing Strong Gravity by Black Holes Across the Range of Masses. \end{acknowledgements} \vspace*{0.5cm} \bibliographystyle{aa}
287255ae470416fd57cbbc11d592576739e71a00
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} A concordance of observations support a universe whose energy budget is dominated by the unknown elements of dark energy (DE) and dark matter (DM), while baryonic matter occupies only around 5\% of the total energy \cite{concord1,concord2,wmap9,planck2013,planck2015}. Focusing on DM, all concrete evidence for its existence is solely based upon its gravitational interactions, prompting some to hypothesize alternative explanations to DM, like modified gravitational interactions (e.g., Ref~\cite{mond}), but these alternatives are not viable in light of observations of colliding galaxy clusters where it is shown that the bulk of the clusters consists of non-luminous matter that does not interact (except gravitationally) \cite{bullet}. Though there is no evidence that DM interacts through non-gravitational channels, theories beyond the standard model that incorporate DM candidates often contain a rich panoply of interactions between dark and standard model particles. In fact, DM will interact electromagnetically even if it is electrically neutral, provided that it couples to charged particles. Granted, these electromagnetic interactions are suppressed, but they are generally nonzero. Given this, a DM medium will have an optical index of refraction which is generally dispersive; i.e., the phase velocity of light in the medium is frequency dependent. In Ref.~\cite{dm_n}, we computed the refractive index for various particulate DM models. The forward Compton scattering amplitude links the medium's optical properties with the particle-level interaction between the photon and DM \cite{goldwatson,fermi}. Rather general considerations (namely, Lorentz covariance and invariance under charge conjugation, parity, and time-reversal symmetries) restrict the structure of this forward Compton amplitude at low photon energies, $\omega$. As a result, the leading order contributions to the forward scattering amplitude are model independent, attributable to the charge, mass, and magnetic dipole moment of the scatterer \cite{GGT,low,GG,G,lapidus}, and the higher order, model dependent, terms follow a known form such that the index is $n(\omega) = 1 - A\omega^{-2} +B +C \omega^2 +\dots$ with each coefficient non-negative assuming small $\omega$ \cite{cosmicn, dm_n}. For electrically neutral DM candidates, the $\mathcal{O}(\omega^{-2})$ term in the index of refraction vanishes because $A = 2 q^2$ where $q$ is the electric charge of the DM. The resulting index of refraction simplifies to $n(\omega) \approx 1 + B + C \omega^2$. In principle, one can experimentally assess the coefficient $C$ through astrophysical observation. Given the normally dispersive nature of the DM in the cosmos, high energy photons will travel more slowly than ones with lower energy. If a broadband pulse of photons travels over a sufficient baseline through the DM medium, then, statistically, the arrival time of photons from that pulse will be energy dependent. For this study, a near ideal source of photons is a gamma ray burst (GRB), observable out to redshifts of $z>9$ \cite{Cucchiara:2011pj}. So, if the arrival time of photons from a large sample of GRBs shows energy and baseline dependence characteristic of matter dispersion, then one can observationally assess the coefficient $C$. The brightness of bursts is a boon for measuring dispersive matter effects, but their varied spectra \cite{Gruber:2014iza} and lower frequency afterglows \cite{Meszaros:1996sv,Cenko:2010cg,Zaninoni:2013hca} are a significant confounding factor. Because dispersion measurements rely on temporal knowledge of the emission spectra, a single GRB event cannot yet be used to constrain DM properties. But with a large number of GRB observations located at a variety of redshifts, it is expected that, statistically, random variations amongst the sources should wash out, and the expected redshift and energy dependence that indicate dispersion should survive. The Fermi Gamma-Ray Space Telescope \cite{fermi_lat} is dedicated to the task of gamma-ray observations, so that in the future, a sufficient number of GRBs may be used to provide meaningful constraints. Regardless, from our computations of the refractive index for several neutral pointlike DM models \cite{dm_n}, we expect the DM dispersion, i.e., the coefficient $C$, to be extremely small; as a result, any time lags would be immeasurable. But, the theoretical landscape is rife with DM candidates, and the work in Ref.~\cite{dm_n} only considered a small subset. Here, we expound upon our previous work by computing the optical dispersion of composite DM comprised of millicharge constituents. Assessing the size of DM dispersive effects is crucial to evaluating potential claims of beyond-SM physics because matter dispersion is not the only mechanism by which one can achieve such energy-dependent photon time lags. In theories with Lorentz invariance violation (LIV), the photon's dispersion relation is modified \cite{myers-pospelov,kosto,kost_mewes_em,kost_mewes_grb}. Following Ref.~\cite{jacobpiran}, LIV effects can modify the dispersion relation for photons at an energy scale $E_\text{LIV}$ \begin{equation} E^2 - p^2 = \pm p^2 \left( \frac{p}{E_\text{LIV}} \right)^n \label{disp_rel} \end{equation} for some integer $n\ge 1$. If the modification in the dispersion relation in Eq.~(\ref{disp_rel}) comes with the minus sign (rather than the plus sign), then high energy photons will lag simultaneously emitted lower energy photons. Photon time lags due to matter dispersion have a different baseline dependence relative to LIV effects (i.e., they share a different dependence upon the source's redshift $z$), but if $n=2$ in Eq.~(\ref{disp_rel}), the two effects have a common photon energy dependence. For this reason, it is useful to know the relative magnitudes of the LIV and matter effects at a given baseline so that matter effects cannot confound potential claims of LIV gleaned from GRB photon arrival times. Because DM comprises the bulk of the matter in the universe, we will consider its impact on dispersion in detail. In Ref.~\cite{dm_n}, we found that the potential LIV effects would dwarf any dispersion due to various models of pointlike DM, assuming $E_\text{LIV}$ is around the Planck scale. In fact, we found that dispersion due to matter effects was irrelevant until energies around $10^{29}$ GeV. But composite DM models \cite{Faraggi:2000pv, Gudnason:2006ug, Gudnason:2006yj, feng_hiddendm, kaplan_da1, Hamaguchi:2009db, Barbieri:2010mn, Lisanti:2009am, cidm_alves, quirky, formfactor_dm, kaplan_da2, Hur:2007uz, Blennow:2010qp, DelNobile:2011je, Frigerio:2012uc, Cline:2012bz, cline_da, Kumar:2011iy, kouvaris, wallemacq, Holthausen:2013ota, Buckley:2012ky, Higaki:2013vuv, cline_strong, Cline:2014eaa, Bhattacharya:2013kma, Boddy:2014yra,bro, Carmona:2015haa, Choquette:2015mca, Wallemacq:2014sta, Antipin:2015xia} might prove to be more reactive, particularly if the DM is comprised of charged constituents \cite{quirky, Lisanti:2009am, DelNobile:2011je, Buckley:2012ky, kouvaris, Cline:2012bz, cline_strong, Choquette:2015mca, Cline:2014eaa, Gudnason:2006ug, Gudnason:2006yj, Wallemacq:2014sta, wallemacq, cline_da}. The motivation for composite dark matter models is varied. Some models are constructed so as to explain possible photon signals of indirect dark matter detection \cite{kouvaris, Cline:2012bz, Cline:2014eaa,bro}. Others introduce composite systems designed to smooth out the cuspiness of simulated DM galactic halo profiles \cite{cline_strong,Boddy:2014yra}. Furthermore, several models attempt to rectify seemingly contradictory results in the experimental search for DM. The DAMA/LIBRA experiment \cite{dama_libra} reports a statistically significant annual modulation in its detector which could be attributed to the relative motion of the detector through the galaxy's dark matter halo. Furthermore, the CoGeNT experiment \cite{cogent1} reports signals above background in its detector which, if due to dark matter, would be consistent with the apparent signal from DAMA/LIBRA. If these results are due to DM interactions, they occupy a region of parameter space that has been seemingly ruled out by the CDMS-II \cite{cdmsii_ge} and XENON100 \cite{xe100} experiments. To reconcile results from DAMA with null results from other experiments, the notion that DM could interact through inelastic channels has been proposed \cite{idm_tuckersmith,idm_chang,idm_marchrussell}. A natural way to incorporate inelastic interactions into a model is to allow DM to be composite, rather than point-like, and a host of models take this tack as means to accommodate the DAMA or CoGeNT results in light of other DM constraints \cite{kaplan_da1,cidm_alves,formfactor_dm,kaplan_da2,cline_da,kouvaris,wallemacq,Lisanti:2009am, DelNobile:2011je,Kumar:2011iy, Blennow:2010qp, Wallemacq:2014sta}. In what follows, we will consider composite dark matter particles which are electrically neutral, but comprised of millicharged constituents. Taken {\em en masse}, we are interested in the dispersive refractive index of such a medium. As the DM is electrically neutral, it is essentially invisible to low energy photons. However, for photons which are near the threshold energy needed to transition the composite DM to an excited state, the photon interaction is substantive. Considering dark matter as a bulk medium, the interaction between dark matter and light can be characterized in terms of an electric susceptibility and index of refraction. For photon energies below the transition energy $\omega \ll \omega_0$, the medium will rather generically exhibit dispersion quadratic in the photon energy $n(\omega) \approx 1 + B + C\omega^2$ for constants $B$ and $C$ as with the pointlike DM, because the low-energy theorems of Compton scattering can be generalized to composite structures \cite{brodsky}. \section{Index of refraction} Classically, a linear dielectric medium, such as a dilute gas of dark matter, will acquire a polarization when subjected to an external electric field. The degree of polarization, or dipole moment per unit volume, can be characterized through the electric susceptibility $\mathbf{P} = \chi_e \mathbf{E}$. From the susceptibility, we can compute the medium's index of refraction $n =\sqrt{1+\chi_e}$. To compute the susceptibility of a dark-matter medium, we take a semiclassical approach in which the quantum mechanical DM system interacts with a classical electromagnetic wave via electric dipole transitions. The constituents which comprise the DM will be assumed to be effectively non-relativistic so that they can be described via the Schr\"odinger equation. We assume the system consists of two constituents of masses $m_{a}, m_{b}$ with electric millicharge $\pm \epsilon e$ bound via a potential $V(r)$ with $r$ the relative separation between the particles. Though we assume a pair of constituent particles, the analysis can be extended to bound states consisting of more particles if need be. Defining the reduced mass $m := m_a m_b/(m_a + m_b)$ and relative momentum $\mathbf{p} := m\dot{\mathbf{r}}=(m_b \mathbf{p}_a-m_a \mathbf{p}_b)/(m_a +m_b)$, we construct the unperturbed Hamiltonian $H_0 := - \nabla^2/(2m) + V(r)$ for the system. To determine the interaction between the DM and light, a classical electromagnetic wave of frequency $\omega$ interacts with the quantum mechanical electric dipole moment of the DM, $\pmb{\mathpzc{p}} := -\epsilon e \mathbf{r}$, which introduces to the Hamiltonian a perturbation, $H' = - \pmb{\mathpzc{p}} \cdot \mathbf{E}$. In the long wavelength limit, the spatial variation of the electric field is irrelevant leaving the time-dependent perturbation $H' =-\pmb{\mathpzc{p}} \cdot \mathbf{E}_0 \cos \omega t$. \subsection{Millicharged atomic dark matter} To create composite particles in a dark sector, modelers introduce a new dark gauge group which results in a binding force among the composite's constituents. The simplest gauge group is $U(1)$ \cite{feng_hiddendm,kaplan_da1,formfactor_dm,kaplan_da2,cline_da,wallemacq,cline_strong, Cline:2014eaa,Wallemacq:2014sta}. If the symmetry is unbroken, then the dark photon is massless and can kinetically mix with the Standard Model (SM) photon. In such models, the particles can effectively couple to the SM photon thereby acquiring a fractional electric charge \cite{holdom}. This permits the existence of dark atoms which are overall electrically neutral, but made of constituents with electric millicharge \cite{cline_da,wallemacq,cline_strong,Cline:2014eaa,Wallemacq:2014sta}. A neutral dark atom comprised of millicharged particles consists of two fermions $\psi_\mathbf{p}$ and $\psi_\mathbf{e}$, charged under the unbroken gauge group $U(1)'$, coupling to the dark photon with opposite charges $\pm \mathbf{e}$. [NB: The boldface type is meant to refer to the particles and couplings in the dark sector.] The dark ``proton" and ``electron" can form bound states under the dark Coulombic potential $V(r) = -\boldsymbol{\alpha}/r$, where we define the dark fine structure constant $\boldsymbol{\alpha} := \mathbf{e}^2/ (4\pi)$, and the relative separation between the particles is $r$. We can use non-relativistic quantum mechanics to describe this dark ``hydrogen", $\mathbf{H}$. Without loss of generality, we assume $m_\mathbf{e}\le m_\mathbf{p}$ and define the reduced mass and relative momentum as above with $m_a= m_\mathbf{e}$ and $m_b= m_\mathbf{p}$. The unperturbed Hamiltonian $H_0$ yields the usual hydrogenic eigenstates $\psi_{n\ell m}$ and energy spectrum $E_n = - \boldsymbol{\alpha}^2 m/(2n^2)$ indexed by principal quantum number $n$, a positive integer. The dark and SM sectors are coupled through photon kinetic mixing which gives rise to electric millicharges $\pm \epsilon e$ of the dark particles. Electromagnetic waves can induce transitions between dark atomic energy states, but we argue that the bulk of the dark atoms exist in the ground state. There are three main mechanisms by which the atoms can be excited beyond the ground state: dark atom self-interactions, absorption of dark photons, or absorption of SM photons. The existence of elliptical DM halos severely constrains the DM self-interaction cross section; from the limits in Ref.~\cite{MiraldaEscude:2000qt}, the ratio of the DM self interaction to the dark atom's mass must be $\sigma/m_\mathbf{H} < 0.02$ cm$^2$/g though more recent studies have relaxed this bound to 0.1 cm$^2$/g \cite{Peter:2012jh}. These limits can be satisfied either through tuning the model parameters or dilution of the dark atom component of DM. For models which satisfy this constraint, we can estimate the mean free time between collisions for dark atoms in the Milky Way's galactic halo. The mean free path can be estimated as $\lambda \sim (\sigma N)^{-1}$ where $N$ the number density of dark atoms; the number density is related to the DM mass density via $N = \rho/m_\mathbf{H}$. Then the time between collisions is $t_\text{fp} \sim \lambda/v = m_\mathbf{H} /(\sigma \rho v) $. Taking as typical parameters the local dark matter density $\rho \sim 0.3$ GeV/cm$^3$ and $v\sim 200$ km/s, we find $t_\text{fp} \gtrsim 10^{18}$ s; i.e., they are non-interacting. Dark-atom absorption of dark photons also produces excited states. The greatest energy density of dark radiation is found in the dark analog of the cosmic microwave background (CMB). Viable models require the dark radiation to be slightly cooler than the CMB \cite{adm_cosmo}, so these dark photons will not have sufficient energy to excite the dark atoms. All that remains is the interaction with SM electromagnetic waves, which we discuss below. We consider a dark atom in its ground state which can be excited by SM photons. We restrict our study to a two-state system, limiting the electromagnetic wave frequency to $\omega \lesssim \omega_0 := E_2-E_1$. In the presence of the electromagnetic wave, the Hamiltonian for the dark atom is $H= H_0 + H'$, and a general state is $\Psi(t) = c_1(t) e^{-i E_1t} \psi_1+c_2(t) e^{-i E_2t} \psi_2$, with stationary eigenstates $\psi_{1,2}$. Using the Schr\"odinger equation, we can develop coupled differential equations for the coefficients $c_1$ and $c_2$. These equations must be amended to account for spontaneous emission of a dark photon from the excited state. With this extra term proportional to the decay constant $\Gamma$, we have the equation for $c_2$ \begin{equation} \frac{\mathrm{d}}{\mathrm{d}t} c_2(t) = -i \, (\pmb{\mathpzc{p}}_{21}\cdot \mathbf{E}_0) e^{i\omega_0t} \cos(\omega t) c_1 (t) -\Gamma c_2(t) \end{equation} where $ \pmb{\mathpzc{p}}_{21} := \langle \psi_2 | \pmb{\mathpzc{p}} | \psi_1 \rangle$ and $\Gamma= \frac{2^7}{3^8}\boldsymbol{\alpha}^5 m$, adapted from atomic hydrogen \cite{loudon}. We solve perturbatively for the coefficients $c_{j}(t) = c_{j}^{(0)}(t)+ c_{j}^{(1)}(t)+\cdots$ with initial conditions $c_1(0)=1$ and $c_2(0)=0$. To determine the {\em linear} response of the atom to the field $E_0$, we only need the zeroth order approximation for $c_1 \approx 1$ and the first order approximation for $c_2$ \begin{equation} c_2(t) \approx -\frac{1}{2} (\pmb{\mathpzc{p}}_{21}\cdot \mathbf{E}_0) \left[ \frac{e^{i(\omega_0 + \omega)t}}{\omega_0 + \omega - i \Gamma}+\frac{e^{i(\omega_0 - \omega)t}}{\omega_0 - \omega - i \Gamma }\right]. \end{equation} The induced dipole moment for $\Psi$ is thus \begin{equation} \pmb{\mathpzc{p}}(t) = -\epsilon e \langle \Psi(t) | \mathbf{r} | \Psi(t) \rangle = -2\epsilon e\, \mathrm{Re} [ \mathbf{r}_{12} c_1^*c_2 e^{-i \omega_0 t}]. \end{equation} We note that the polarization $P(t)$ will be the product of the average induced dipole moment in the direction of $\mathbf{E}_0$ and the number density $N$. Averaging over the relative orientation between $\mathbf{r}_{12}$ and $\mathbf{E}_0$, the susceptibility is \begin{equation} \chi_e = N\pi\frac{2^{18} }{3^{11}} \frac{\epsilon^2 \alpha }{\boldsymbol{\alpha}^2 m^2}\frac{\omega_0}{\omega_0^2 -(\omega+i\Gamma)^2}. \end{equation} Given that the DM medium is weakly interacting and dilute, $\chi_e$ is nearly zero. As such we can approximate the index of refraction as $n \approx 1 +\frac{1}{2} \chi_e$. For frequencies below the transition energy $\omega < \omega_0$, the index of refraction exhibits normal dispersion; that is, $n(\omega)$ increases with $\omega$. In fact, for $\omega \ll \omega_0$, the dispersion is quadratic in frequency \begin{equation} \mathrm{Re}(n) \approx 1 + N\pi\frac{2^{20} }{3^{12}} \frac{\epsilon^2 \alpha }{\boldsymbol{\alpha}^4 m^3}\left( 1 + \frac{\omega^2}{\omega_0^2}\right) , \label{adm_index} \end{equation} neglecting the small term proportional to $\Gamma^2$. \subsection{Other millicharged composite particles \label{sho_sect}} Millicharged atomic DM represents only a fraction of proposed composite DM models. Many models posit composite states strongly bound by a nonabelian gauge force \cite{Holthausen:2013ota, quirky, Lisanti:2009am,DelNobile:2011je,Buckley:2012ky,kouvaris,Cline:2012bz,cline_strong,Choquette:2015mca,Barbieri:2010mn,Hur:2007uz,Faraggi:2000pv,Kumar:2011iy,Blennow:2010qp,Hamaguchi:2009db,Higaki:2013vuv, Bhattacharya:2013kma, Carmona:2015haa,Gudnason:2006ug,Gudnason:2006yj,Antipin:2015xia,Boddy:2014yra,bro}, some of which contain electrically charged constituents \cite{quirky,Lisanti:2009am, DelNobile:2011je,Buckley:2012ky, kouvaris, Cline:2012bz, cline_strong,Choquette:2015mca, Gudnason:2006ug, Gudnason:2006yj}. The details of these strongly composite systems with charged constituents are model dependent, so we will only sketch an approach as to how one would estimate the susceptibility of such a DM medium. As with the atomic DM system, we will assume a composite state consisting of two particles with effective (dressed) masses of $m_a$ and $m_b$ with electric millicharges $\pm \epsilon e$. To model a tightly bound system, we approximate their interaction potential as that of a harmonic oscillator, $V(r) = \frac{1}{2}m\omega_0^2 r^2$ with $m$ the reduced mass. As above, we will only consider a two state system consisting of the ground state and first excited state; the energy difference between these states is $\omega_0$. Our na\"ive assumptions result in a simplistic model, yet for a system as complex as a nucleon, the SHO potential adapted to three constituent quarks yields an order of magnitude estimate of the nucleon polarizability \cite{holstein}. With this new potential, we merely need to compute the transition dipole moment $\mathpzc{p}_{21}$ and the decay constant $\Gamma$. We find $\mathpzc{p}_{21} = \epsilon e/\sqrt{2m \omega_0}$ and $\Gamma =\epsilon^2 \alpha \omega_0^2/(3m)$. Given our assumption of nonrelativistic QM, $\Gamma/\omega_0$ is small so that we can approximate the susceptibility of this DM medium as \begin{equation} \chi_e = \frac{4}{3} \pi N \frac{\epsilon^2 \alpha }{ m \omega_0^2}\left( 1 + \frac{\omega^2}{\omega_0^2}\right). \end{equation} Finally, because the susceptibility is small, the refractive index can be approximated by $n \approx 1 + \frac{1}{2} \chi_e$ for $\omega \ll \omega_0$. \section{Observational consequences for GRB photons} As a broadband pulse of electromagnetic radiation travels through a dispersive medium, the pulse shape spatially broadens because the phase speed of each component wave is frequency dependent. The DM medium we consider is normally dispersive, $n \approx 1 + B + C \omega^2$, so higher frequency components of the pulse will lag the lower frequency components. Over large distances, a time lag can accrue between these two components. Gamma-ray bursts are apt photon sources for dispersive studies because they are explosive events that occur over short time scales and their brightness permits observation over cosmological distances. Because cosmological distances are involved, we must account for the universe's expansion as a light pulse travels from source to observer \cite{jacobpiran}. There are three factors that need to be considered. First, the light-travel time between source and detector is dependent upon the redshift of the source and the local expansion rate. Second, at redshift $z$ the number density of dark matter increases by a factor of $(1+z)^3$; a factor of present day DM number density $N$ is contained in the coefficient $C$. Finally, as we look into the past, the wavelength of light blue shifts relative to its value $\omega$ at the detector (at $z=0$). Incorporating these three factors, the time lag accrued between (detected) frequencies $\omega_\text{hi}$ and $\omega_\text{lo}$ simultaneously emitted from a source at redshift $z$ becomes \begin{equation} \tau \approx C\left( \omega_\text{hi}^2 -\omega_\text{lo}^2 \right) \int_0^z \frac{(1+z')^5 \mathrm{d}z'}{H(z')}, \label{matter_lag1} \end{equation} where the Hubble expansion rate at redshift $z'$ is $H(z') = H_0 \sqrt{(1+z')^3\Omega_M + \Omega_\Lambda}$, assuming a simple $\Lambda$CDM cosmology. If the two photon energies are well separated, we can neglect the low energy term in the difference in Eq.~(\ref{matter_lag1}) and set $\omega = \omega_\text{hi} \gg \omega_\text{lo}$ so that \begin{equation} \tau \approx C\omega^2 K_5 (z), \label{matter_lag2} \end{equation} where we define the integral over the baseline $K_j (z) := \int_0^z \frac{(1+z')^j \mathrm{d}z'}{H(z')}$. We employ the cosmological parameters in the 2015 data release from the Planck satellite \cite{planck2015}. The Hubble constant today is $H_0 = 67.8 \pm 0.9 \,\hbox{km/s/Mpc}$, whereas the fraction of the energy density in matter relative to the critical density today is $\Omega_M = 0.308\pm 0.012$. For the simple $\Lambda$CDM model, the universe is flat so that the corresponding fraction of the energy density in the cosmological constant $\Lambda$ is $\Omega_\Lambda = 1- \Omega_M$. The high energy photons from distant GRBs have already been used to place lower bounds on the scale at which Lorentz-invariance violating effects could modify the photon's dispersion relation \cite{AmelinoCamelia:1997gz,ellis, Boggs:2003kxa,Ellis:2005wr,jacobpiran, ellis2,Ellis:2009yx,AmelinoCamelia:2009pg,Vasileiou:2013vra}. For some LIV models, the modified dispersion relation acquires terms quadratic in the photon energy, i.e., $n=2$ in Eq.~(\ref{disp_rel}). For these models, the following time lag can accrue between the low and high energy photons emitted by a GRB \begin{equation} \Delta t_\text{LIV} \approx \frac{3}{2} \left(\frac{\omega}{E_\text{LIV}}\right)^2 K_2(z). \label{t_liv} \end{equation} where the (present day) photon energy $\omega = \omega_\text{hi} \gg \omega_\text{lo} $ was emitted at redshift $z$. The energy $E_\text{LIV}$ characterizes the scale at which LIV effects become appreciable. Data from GRBs and AGNs have placed limits on this scale (for quadratic dependence) to be $E_\text{LIV} > 1.3 \times 10^{11}$ GeV \cite{bolmont}. We wish to assess the size of dispersive matter effects relative to those attributable to LIV. From Eqs.~(\ref{matter_lag2}) and (\ref{t_liv}), we will consider the ratio $\tau/\Delta t_\text{LIV}\sim \frac{2}{3} C E_\text{LIV}^2 K_5(z)/K_2(z)$ for millicharged composite DM candidates. The factor $C$ is dependent on the DM model, whereas the ratio of the $K_j$ integrals depends on the redshift of the source with $K_5(z)/K_2(z)\sim \mathcal{O}((1+z)^3)$. The GRB constraint in Ref.~\cite{bolmont} is derived from GRB 090510 which is located at a reshift of $z=0.903\pm 0.003$ with a high energy photon $\omega_\text{hi}=30.53^{+5.79}_{-2.56}$ GeV detected 0.829 s after the trigger of the GRB monitor \cite{fermi_nat}. In this case, the ratio of integrals $K_j$ is $\mathcal{O}(10)$. On the other hand, there have been two GRB observations with confirmed redshift $z>8$: GRB 090423 with $z=8.3\,$ \cite{Tanvir:2009zz} and GRB 090429B with $z=9.4$ \cite{Cucchiara:2011pj}. For these, we find the maximum value of $K_5/K_2 \sim \mathcal{O}(10^3)$. Thus, matter effects are enhanced dramatically relative to LIV effects for higher redshift sources. As stated previously, we work within the context of a simple $\Lambda$CDM model; dark energy is attributed to a cosmological constant with an equation of state $w= -1$, where $w$ is the ratio of DE's pressure to energy density. This is consistent with the 2015 Planck data which, along with external astrophysical data, determines $w=-1.006 \pm 0.045$, if a constant $w$ is assumed \cite{planck2015}. However, when the Planck data are combined with weak lensing data, cosmologies with a time-dependent equation of state are at least marginally preferred \cite{planck2015de}. Extensions of the simple $\Lambda$CDM model will affect the expansion rate of the universe, $H(z')$, and thus the integrals $K_j(z)$. The impact of cosmology upon GRB photon and neutrino time lags due to LIV was previously considered in Refs.~\cite{Biesiada:2007zzb,Biesiada:2009zz}, where the authors compare results from a simple $\Lambda$CDM model to quintessence, Chaplygin gas, and braneworld cosmologies. Amongst the models, differences in time lags exist which could affect the measurability of the effect or the interpretation of a measurement if the cosmology is unknown. Using the models considered in Refs.~\cite{Biesiada:2007zzb,Biesiada:2009zz}, we determine the impact of cosmology upon the {\em relative} size of dispersive matter effects with those from LIV. We find the ratio of integrals $K_5(z)/K_2(z)$ varies little (less than $5\%$) out to $z=10$ for the $\Lambda$CDM, quintessence, Chaplygin gas, and braneworld models. On the other hand, for the variable quintessence model, we find that $K_5(z)/K_2(z)$ is commensurate in size with this ratio in the other models up to $z=2$, but beyond this redshift, the ratio of integrals approaches an asymptotic value around 18 (while the ratio scales as $\mathcal{O}((1+z)^3)$ for the other models). For GRBs at a distant redshift of $z=10$, the difference between the variable quintessence model and the others will be a factor of 25. On the face of it, this difference is substantial and could perhaps be even larger for different models of quintessence; however, when we examine the size of DM dispersive effects below, we will see that the choice of cosmology is of subleading significance. Given this, we opt to use the simple $\Lambda$CDM model, though acknowledge that other cosmologies are likely and will quantitatively affect our results to a degree. \subsection{Millicharged atomic dark matter} We now examine the size of the dispersive coefficient $C$ in the refractive index relative to $E_\text{LIV}$. This coefficient sets the energy scale, $C^{-\frac{1}{2}}$, at which DM dispersion becomes appreciable. For atomic dark matter, we find from Eq.~(\ref{adm_index}) this coefficient to be $C= N\pi\frac{2^{20} }{3^{12}} \frac{\epsilon^2 \alpha }{\boldsymbol{\alpha}^4 m^3 \omega_0^2}$. The DM number density is $N = \rho/m_\mathbf{H}$ where the mass of the dark atom is $m_\mathbf{H} = m_\mathbf{p} + m_\mathbf{e} -\frac{1}{2}\boldsymbol{\alpha}^2 m$. For propagation of light across cosmological distances, the average DM density is $\rho \simeq 1.25\times 10^{-6}$ GeV/cm$^3$\cite{planck2015}. As a benchmark, it is useful to estimate the value of the coefficient for various atomic dark matter models. In Ref.~\cite{wallemacq}, the best-fit model employs a heavy dark proton $m_\mathbf{p} = 650$ GeV and much lighter dark electron $m_\mathbf{e}= 0.426$ MeV with an electric millicharge $\epsilon = 6.7\times 10^{-5}$ and a dark fine structure constant which takes the SM value, $\boldsymbol{\alpha} = \alpha$. For this model, the dispersive term is $C = 1.8 \times 10^{-25}$ GeV$^{-2}$. As a consequence, the energy scale at which the DM dispersion is appreciable is $C^{-\frac{1}{2}} = 2.4 \times 10^{12}$ GeV, which is commensurate with the limit on $E_\text{LIV}$ from Ref.~\cite{bolmont}. On the face of it, it would seem that for large $z$ the $K_5$ integral could make dispersion due to millicharge atomic DM competitive in magnitude with potential LIV dispersive effects, but we must recall one crucial point--the binding energy of the dark atom. Our estimate of the DM refractive index is only valid in the limit in which observed photon energies are much smaller than the energy difference between the ground and excited state of the dark atom, $\omega \ll \omega_0 = \frac{3}{8} m \boldsymbol{\alpha}^2$. The dark electron in Ref.~\cite{wallemacq} has a mass similar to the SM electron, so the ionization energy of the dark hydrogen is on the scale of electronvolts with $\omega_0 = 8.5$ eV. As such, the limits derived from high energy GRB photons are not relevant for this model of atomic dark matter because the dark electrons are so weakly bound. Let us now consider another model of atomic dark matter which employs very different masses and electric millicharge. Reference \cite{cline_da} has an atomic DM model which can account for the CoGeNT experimental results. The authors' dark atom consists of massive particles $m_\mathbf{p} = m_\mathbf{e} = 3$ GeV with a slightly larger dark fine structure constant $\boldsymbol{\alpha}= 0.062$ and much larger millicharge $\epsilon = 10^{-2}$. In this case, the relevant energy scale for matter dispersive effects is $C^{-\frac{1}{2}} = 6 \times10^{21}$ GeV with transition threshold of $\omega_0 = 2.2$ MeV. Due to the larger masses and dark fine structure constant, the threshold energy $\omega_0$ rises, allowing the dark matter to be probed with more energetic photons, but for such massive models, the energy scale at which matter effects become appreciable is well beyond the current bounds on $E_\text{LIV}$. These two specific examples are exemplars of a general trend. Namely, for light DM masses, the energy scale at which matter dispersion is operative is commensurate with limits on the LIV energy scale, but the binding energy is too small to actually probe such dispersion. For more massive models, the threshold energy $\omega_0$ increases, but $C^{-\frac{1}{2}}$ increases at a greater rate. To confirm these generalities, we explore more fully the allowed parameter space of exotic millicharged dark atoms without regard to their feasibility as a dark matter candidate. In particular, we compute the dispersive energy scale $C^{-\frac{1}{2}}$ for a range of dark electron and proton masses subject only to the provision that the existence of the constituent millicharged particles have not been excluded through other considerations. There are, in fact, stringent constraints on electric millicharge \cite{davidson,Vogel:2013raa}. Figure 1 of Ref.~\cite{Vogel:2013raa} shows a current summary of these constraints for particle masses ranging from 100 eV to 100 TeV. Stellar evolution severely constrains electric millicharge $\epsilon < 2\times 10^{-14}$ for masses below 10 keV. Around 100 keV to 1 GeV, BBN and CMB constraints upon the light degrees of freedom limit electric millicharge $\epsilon \lesssim 10^{-9}$. For masses between 1 MeV and 100 GeV, collider constraints limit $\epsilon < 0.2$. Dispersive effects scale as $\epsilon^2$, so they will be maximized for the largest allowed $\epsilon$. To explore the largest possible matter dispersion, we choose, for a given particle mass, the millicharge which saturates the bounds in Fig.~1 of Ref.~\cite{Vogel:2013raa}. Setting the dark fine structure constant equal to the SM value $\boldsymbol{\alpha} = \alpha$, we plot in Fig.~\ref{fig1} contours representing the energy scale $C^{-\frac{1}{2}}$ for various dark electron and proton masses. These contours are superimposed upon filled contours which represent the threshold energy $\omega_0$. For masses ranging from 100 eV to 1 GeV, we find the energy scale charactering matter dispersion range from $10^9$ GeV to $10^{21}$ GeV, spanning twelve decades of energy, while $\omega_0$ ranges from meV to keV, spanning six decades. \begin{figure} \includegraphics[width=10cm]{fig1.pdf} \caption{The solid contour lines depict $C^{-\frac{1}{2}}$ for millicharged atomic dark matter; this is superimposed upon filled contours which represent the threshold energy $\omega_0$. The dark atom consists of a dark electron and proton with mass $m_{\mathbf e}$ and $m_{\mathbf p}$ and dark fine structure constant equal to that of the SM, $\boldsymbol{\alpha} = \alpha$. All contours carry units of electronvolts. \label{fig1}} \end{figure} The broad scan of parameter space depicted in Fig.~\ref{fig1} confirms our preliminary conjecture. We see rather generally that if the energy scale at which atomic DM dispersive effects are near current limits on $E_\text{LIV}$ then the binding energy of the dark atom is sub-eV. The photon energy needed to probe the DM dispersion would ionize the dark atoms. For higher threshold energies, matter dispersive effects quickly become irrelevant. Supposing a LIV scale around the Planck mass, $\sim 10^{19}$ GeV, the ionization energy of the dark atom is on the order of keV. Thus, in a search for LIV dispersive effects, dispersive effects due to atomic dark matter are not a confounding source of background. \subsection{Other millicharged composite particles} We now turn to our millicharged composite system bound under the harmonic oscillator potential. There is wide variation amongst the strongly bound composite DM models, but we will explore the dispersive matters effects for a limited range of particle mass and oscillator energy $\omega_0$. Again, to determine the maximal effect, we assume electric millicharge values which saturate the bounds in Fig.~1 of Ref.~\cite{Vogel:2013raa}. From above, we find the dispersive coefficent to be $C = \frac{2}{3} \pi N \frac{\epsilon^2 \alpha }{ m \omega_0^4}$. We will assume that the constituent millicharged particles have the same effective mass $m_a = m_b$ so that the reduced mass is one-half this mass $m = \frac{1}{2} m_a$ and the total mass of the system is roughly $2m_a$. With these simplifying assumptions, the dispersive energy scale depends rather simply upon the composite parameters, $C^{-\frac{1}{2}} \sim m_a \omega_0^2/\epsilon$. As a result, this energy is lowest for masses $m_a$ in the GeV to 100 GeV region where $\epsilon \lesssim 0.2$. \begin{figure}[h] \includegraphics[width=10cm]{fig2.pdf} \caption{ Plot of $C^{-\frac{1}{2}}$ as a function of constituent mass $m_a$. We assume $m_a = m_b$ and a threshold energy of $\omega_0 = 1$ keV. \label{fig2}} \end{figure} In Fig.~\ref{fig2}, we plot the dispersive energy scale as a function of the mass $m_a$, choosing $\omega_0 =1$ keV. The threshold energy could take a host of values, but we choose 1 keV because inelastic DM models with an excitation scale of order $\mathcal{O}(1-100$ keV) can accommodate the DAMA anomalous experimental results or potential indirect detection signals \cite{Lisanti:2009am,Cline:2014eaa, Kumar:2011iy, bro}. Given that $C^{-\frac{1}{2}}$ is proportional to $\omega_0^2$, it is not difficult to extrapolate these results to different threshold values. Referring to Fig.~\ref{fig2}, we find that the lowest dispersive energy scale, $C^{-\frac{1}{2}}= 1.7 \times 10^{13}$ GeV, occurs for a millicharged particle mass of 1.3 GeV. This energy is a few decades above the current limit on the LIV scale, but over long baselines the size of matter dispersive effects could rival those due to LIV. Of course, unlike LIV, the quadratic dispersive terms is only relevant for photon energies below the threshold of 1 keV. This difference between the threshold and dispersive energy scales makes the matter dispersive effects immeasurable. Indeed, for a nearby GRB, $z\sim 1$, the time lag between keV and lower energy photons is on the order of $10^{-20}$ s. Given the same millicharged particle mass of 1.3 GeV, a dispersive scale near the Planck mass, $C^{-\frac{1}{2}} \sim 10^{19}$ GeV, involves a threshold energy near 1 MeV, but probing the DM with near threshold photons results in the same time lag for the $\omega_0 = 1$ keV case. As with atomic dark matter, whenever the energy scale of the dispersive coefficient is commensurate with the LIV scale, the threshold energy of the composite system is too small to probe this scale. Though one has some freedom to tune the threshold of the SHO model independent of the millicharged particle properties, this freedom is not sufficient to construct a composite DM candidate that can have measurable dispersive effects competitive with potential LIV effects. \subsection{Baryonic matter} One potential source of background to the dispersive matter effects from composite millicharge DM is ordinary baryonic matter which, for the most part, consists of the hydrogen and helium produced in big-bang nucleosynthesis. Dispersion due to baryonic matter occurs on varying scales, the first being the atomic scale, which is relevant for photon energies in the eV range. As we discussed above, the probes of LIV are high-energy (MeV or greater) photons that would ionize the hydrogen and helium of baryonic matter, so these cannot be confounded with LIV effects. Beyond the atomic energy scale, a medium of hydrogen and helium will effectively appear as a plasma to high energy photons, and the plasma's optical properties will be predominantly determined by its electron component. Such a medium possesses a dispersive index of refraction of the form $n \approx = 1- A\omega^{-2}$ with $A= e^2 N/(2m_e)$ where $N$ is the number density of the electrons \cite{cosmicn,dm_n}. This does not have the $\mathcal{O}(\omega^2)$ dispersive behavior typical for a neutral scatterer, so it cannot be confused with dispersion due to DM or LIV. Furthermore, for a low density cloud of atoms, dispersion is negligible for high energy photons because $n -1 \sim 1/\omega^2$. For photon energies approaching the nuclear scale, there are $\mathcal{O}(\omega^2)$ dispersive effects to consider arising from the polarizabilities of the nucleons, but this physics is reasonably well understood both from an experimental and theoretical standpoint \cite{holstein_rev}. As a result, the impact of baryonic dispersion can be studied in detail, but here we opt to estimate the effect by combining measured values of a nucleon's electric polarizability with our SHO model for strongly bound systems in Sec.~\ref{sho_sect}. From Ref.~\cite{holstein_rev}, we find electric polarizabilities for the proton and neutron, $\alpha_\text{E}^p=11 \times 10^{-4}$ fm$^3$ and $\alpha_\text{E}^n=12 \times 10^{-4}$ fm$^3$, so for a nucleon we take $\alpha_\text{E}^{N}=12 \times 10^{-4}$ fm$^3$. The small size of the polarizability indicates a tightly bound system, and we expect dispersive effects to be small. With our SHO model for dispersion, we find $n \approx 1 + 2\pi N \alpha_\text{E}^N(1 + \omega^2/\omega_0^2)$ where $N$ is the number density of nucleons; we estimate the excitation energy to be $\omega_0 \approx 300$ MeV \cite{holstein}. From the Planck 2015 data, the baryonic component of the universes's energy budget is $\Omega_b =0.048$ so that number density of nucleons is $N = 2.5 \times 10^{-7}$ cm$^{-3}$. This renders a dispersive term of $n -1 \approx (2 \times 10^{-48})\, \omega^2/\omega_0^2$. For photon energies well below $\omega_0$, the dispersive energy scale is $C^{-\frac{1}{2}} \sim 2\times 10^{26}$ MeV. In short, this is not competitive with LIV probes and likely immeasurable. But, even if a suspected baryonic signal arose from a photon time lag, one can assess the presence of hydrogen and helium along the line of sight through the absorption spectrum that will arise in the optical afterglow of the burst. This absorption spectrum can aid in determining the presence of atomic scale physics, e.g., a Lyman-alpha system, differentiating the matter from millicharged DM. \section{Conclusion} Models of composite dark matter represent an attractive alternative to simple pointlike DM candidates. Through their inelastic interactions, one can account for potential DM direct and indirect detection signals, produce more realistic DM galactic halo profiles, and provide a natural explanation for the relative abundance of dark to baryonic matter. In addition to these physical motivations, composite systems are aesthetically pleasing because the dark sector mirrors some of the complexity of the SM sector. In this work, we focused upon composite systems which consist of electrically charged constituents. Such particles naturally couple to the electromagnetic field via an electric dipole moment, rendering the cosmos with a dispersive optical index of refraction. We computed this index of refraction for atomic DM and more strongly bound composite systems, modeled through a harmonic oscillator potential. Given the dispersive nature of the refractive index, higher energy photons will lag those with lower energy as they travel through the universe from a distant GRB. A time lag accrued over long baselines could be confused with similar dispersive effects which result from theories of LIV. However, we found that, for both atomic and more strongly bound composite DM, whenever the energy scale of matter dispersion is commensurate with that of LIV dispersion, the threshold energy $\omega_0$ is too small to actually probe the matter dispersive effects. As a result, potential claims of LIV achieved through time lags cannot be confused for matter dispersion due to composite DM comprised of charged constituents. \section{ACKNOWLEDGMENTS} AK acknowledges partial support during the completion of this work from the Washington NASA Space Grant Consortium and the University of Puget Sound. DL thanks Bernie Bates for useful discussions.
b26025642b0a26f49d0c14cd2ade145b9ea460f9
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section*{References}} \usepackage{graphicx} \usepackage{amsmath} \usepackage{natbib} \usepackage{amssymb} \usepackage{amsthm} \usepackage{lineno} \usepackage{subfig} \usepackage{enumerate} \usepackage{fullpage} \usepackage{algorithm} \usepackage{algpseudocode} \usepackage{xcolor} \usepackage[colorinlistoftodos]{todonotes} \usepackage{hyperref} \biboptions{sort,compress} \usepackage{xcolor} \newcommand*\patchAmsMathEnvironmentForLineno[1]{% \expandafter\let\csname old#1\expandafter\endcsname\csname #1\endcsname \expandafter\let\csname oldend#1\expandafter\endcsname\csname end#1\endcsname \renewenvironment{#1}% {\linenomath\csname old#1\endcsname}% {\csname oldend#1\endcsname\endlinenomath} \newcommand*\patchBothAmsMathEnvironmentsForLineno[1]{% \patchAmsMathEnvironmentForLineno{#1}% \patchAmsMathEnvironmentForLineno{#1*}}% \AtBeginDocument{% \patchBothAmsMathEnvironmentsForLineno{equation}% \patchBothAmsMathEnvironmentsForLineno{align}% \patchBothAmsMathEnvironmentsForLineno{flalign}% \patchBothAmsMathEnvironmentsForLineno{alignat}% \patchBothAmsMathEnvironmentsForLineno{gather}% \patchBothAmsMathEnvironmentsForLineno{multline}% } \usepackage{color,soul} \usepackage{enumitem} \usepackage{mathtools} \usepackage{booktabs} \definecolor{lightblue}{rgb}{.90,.95,1} \definecolor{darkgreen}{rgb}{0,.5,0.5} \newcommand\assignment[1]{\todo[inline,color=red!10,size=\normalsize]{#1}} \newcommand\guide[2]{\sethlcolor{lightblue}\hl{#2}\todo[color=lightblue,size=\tiny]{#1}} \newcommand\guidenoa[1]{\sethlcolor{lightblue}\hl{#1}} \definecolor{lightgreen}{rgb}{.90,1,0.90} \newcommand\bcon[1]{\todo[color=lightgreen,size=\tiny]{$\downarrow\downarrow\downarrow$ #1}} \newcommand\econ[1]{\todo[color=lightgreen,size=\tiny]{$\uparrow\uparrow\uparrow$ #1}} \newcommand\commofA[2]{\todo[color=red!50,size=\small,inline]{{\bf \color{blue} {#1}'s comments}: #2}} \newcommand\commofB[2]{\todo[color=blue!50,size=\small,inline]{{\bf \color{blue} {#1}'s comments}: #2}} \newcommand\commofC[2]{\todo[color=purple!50,size=\small,inline]{{\bf \color{blue} {#1}'s comments}: #2}} \newcommand{\boldsymbol{\tau}}{\boldsymbol{\tau}} \newcommand{\tilde{\boldsymbol{\tau}}^{rans}}{\tilde{\boldsymbol{\tau}}^{rans}} \newcommand{\bs}[1]{\boldsymbol{#1}} \newcommand{\textbf{[ref]}}{\textbf{[ref]}} \usepackage{changes} \definechangesauthor[name={Reviewer 1}, color = red]{Editor} \usepackage{array} \newcolumntype{P}[1]{>{\centering\arraybackslash}m{#1}} \newcommand{\delete}[1]{\textcolor{gray}{\sout{#1}}} \graphicspath{ {./figs/} } \linespread{1.5} \journal{Flow Turbulence Combust} \begin{document} \begin{frontmatter} \clearpage \title{A~Priori Assessment of Prediction Confidence for Data-Driven Turbulence Modeling} \author[vt]{Jin-Long Wu} \ead{[email protected]} \author[vt]{Jian-Xun Wang} \author[vt]{Heng Xiao\corref{corxh}} \cortext[corxh]{Corresponding author. Tel: +1 540 231 0926} \ead{[email protected]} \author[snl]{Julia Ling} \address[vt]{Department of Aerospace and Ocean Engineering, Virginia Tech, Blacksburg, VA 24060, USA} \address[snl]{Thermal/Fluid Science and Engineering, Sandia National Laboratories, Livermore, California 94551, USA} \begin{abstract} Although Reynolds-Averaged Navier--Stokes (RANS) equations are still the dominant tool for engineering design and analysis applications involving turbulent flows, standard RANS models are known to be unreliable in many flows of engineering relevance, including flows with separation, strong pressure gradients or mean flow curvature. With increasing amounts of 3-dimensional experimental data and high fidelity simulation data from Large Eddy Simulation (LES) and Direct Numerical Simulation (DNS), data-driven turbulence modeling has become a promising approach to increase the predictive capability of RANS simulations. However, the prediction performance of data-driven models inevitably depends on the choices of training flows. This work aims to identify a quantitative measure for \textit{a priori} estimation of prediction confidence in data-driven turbulence modeling. This measure represents the distance in feature space between the training flows and the flow to be predicted. Specifically, the Mahalanobis distance and the kernel density estimation (KDE) technique are used as metrics to quantify the distance between flow data sets in feature space. To examine the relationship between these two extrapolation metrics and the machine learning model prediction performance, the flow over periodic hills at $Re=10595$ is used as test set and seven flows with different configurations are individually used as training sets. The results show that the prediction error of the Reynolds stress anisotropy is positively correlated with Mahalanobis distance and KDE distance, demonstrating that both extrapolation metrics can be used to estimate the prediction confidence \textit{a priori}. A quantitative comparison using correlation coefficients shows that the Mahalanobis distance is less accurate in estimating the prediction confidence than KDE distance. The extrapolation metrics introduced in this work and the corresponding analysis provide an approach to aid in the choice of data source and to assess the prediction performance for data-driven turbulence modeling. \end{abstract} \begin{keyword} turbulence modeling\sep Mahalanobis distance\sep kernel density estimation\sep random forest regression\sep extrapolation\sep machine learning \end{keyword} \end{frontmatter} \section{Introduction} \label{sec:intro} Even with the rapid growth of available computational resources, numerical models based on Reynolds-Averaged Navier--Stokes (RANS) equations are still the dominant tool for engineering design and analysis applications involving turbulent flows. However, the development of turbulence models has stagnated--the most widely used general-purpose turbulence models (e.g., $k$-$\varepsilon$ models, $k$-$\omega$ models, and the S--A model) were all developed decades ago. These models are known to be unreliable in many flows of engineering relevance, including flows with three-dimensional structures, swirl, pressure gradients, or curvature~\cite{Craft}. This lack of accuracy in complex flows has diminished the utility of RANS as a predictive simulation tool for use in engineering design, analysis, optimization, and reliability assessments. Recently, data-driven turbulence modeling has emerged as a promising alternative to traditional modeling approaches. While data-driven methods come in many formulations and with different assumptions, the basic idea is that a model or correction term is determined based on data. In the context of turbulence, this data can come from either experiments or high fidelity simulations such as Direct Numerical Simulations (DNS) or well-resolved Large Eddy Simulations (LES). Koumoutsakos~\cite{Milano} trained neural networks on channel flow DNS data and applied them to the modeling of near-wall turbulence structures. Tracey et al.~\cite{Tracey2013} used neural networks to predict the Reynolds stress anisotropy and source terms for transport equations of turbulence quantities (e.g., $\tilde{\nu}_t$ for the S--A model and $\omega$ for $k$-$\omega$ models). Duraisamy et al.~\cite{Tracey2013, Duraisamy2015} have used Gaussian processes to predict the turbulence intermittency and correction terms for the turbulence transport equations. Ling and Templeton~\cite{ling2015evaluation} trained random forest classifiers to predict when RANS assumptions would fail. Ling et al.~\cite{Ling2016,ling2016reynolds} further used random forest regressors and neural networks to predict the Reynolds stress anisotropy. Wang et al.~\cite{Wang2016} have recently investigated the use of random forests to predict the discrepancies of RANS modeled Reynolds stresses in separated flows. These studies show the significant and growing interest in applying data-driven machine learning techniques to turbulence modeling. However, Tracey et al.~\cite{Tracey2013}, Ling and Templeton~\cite{ling2015evaluation}, and Wang et al.~\cite{Wang2016} all reported that their data-driven closures had diminished performance on flows that were significantly different from the ones on which they were trained. Such findings underline the importance of properly choosing the training flows. For instance, if a machine learning model for the eddy viscosity is trained on a database of attached boundary layer flows, then it would not be surprising if the model had poor performance when making predictions on a flow with separation. In the general context of data-driven modeling, the \textit{training set} is the set of data to which the model is fit or calibrated. The \textit{test set} is the set of data on which the model makes predictions. It is expected that the prediction performance will not be satisfactory if the test set is significantly different from the training sets. A key question, then, is how to determine whether the test flow is ``different'' from the training flows. Test flow that might appear very different could, in fact, be well-supported in the training set. For example, a model trained on flow around an airfoil might perform well on a channel flow because it will have encountered developing boundary layers in its training set. Conversely, a model trained on attached flow around an airfoil might perform very poorly on stalled flow around an airfoil, even though the flow configurations appear quite similar. Because many of these data-driven models are formulated such that their inputs are the local flow variables, it is the local flow regimes that must be well-supported in the training set, not any specific global geometry. The degree to which the test flow is well-supported by training data will in large part determine the reliability of the model closure. In deploying these data-driven models, then, it will be crucial to have efficient metrics of determining if a test flow is an extrapolation from the training set. Ling and Templeton~\cite{ling2015evaluation} presented one promising option for quantifying model extrapolation. They used a statistical metric called the Mahalanobis distance to calculate the probability that a test point was drawn from the training distribution, and showed that their machine learning model error was significantly higher on test points with high Mahalanobis distances. However, they did not carry out a thorough investigation of the correlation between the Mahalanobis distance and machine learning model accuracy for different test sets and training sets. Because the Mahalanobis distance assumes that the training data have a multivariate Gaussian distribution, it is not clear how generally applicable this metric will be for different flow cases. Therefore, this paper will investigate two different statistical metrics of extrapolation, the Mahalanobis distance and the distance based on Kernel Density Estimation (KDE), for a variety of training sets. These extrapolation metrics will be analyzed and compared in the context of a machine learning framework for the prediction of Reynolds stresses developed by Wang et al.~\cite{Wang2016}. The ultimate goal of this work is to provide quantitative metrics to assess the prediction confidence \textit{a priori}, in order to guide the choice of training set when applying data-driven turbulence modeling. The rest of the paper is organized as follows. Section~\ref{sec:ml-methodology} will describe the machine learning methodology of Wang et al.~\cite{Wang2016} and the database of flows used for training and testing. Section~\ref{sec:extrapolation} will describe different extrapolation metrics and their relative advantages and disadvantages. Section~\ref{sec:results} will analyze the performance of these extrapolation metrics in predicting the regions of high uncertainty in the machine learning models, and Section~\ref{sec:conclusion} will present conclusions and ideas for next steps. \section{Machine Learning Methodology} \label{sec:ml-methodology} In this section we summarize the Physics-Informed Machine Learning (PIML) framework proposed by Wang et al.~\cite{Wang2016}, which will be used to evaluate the efficacy of the proposed extrapolation metrics. However, note that extrapolation metrics proposed in this work are not specific to the PIML framework. Particular emphasis is placed on applications where the feature space input has high dimensions (10 to 100 features), which is typical in computational mechanics problems and in other complex physical systems (see e.g., \cite{ling2016jcp}). The overarching goal of the work of Wang et al.~\cite{Wang2016} is a physics-informed machine learning framework for turbulence modeling. The problem can be formulated as follows: given high-fidelity data for the Reynolds stresses from DNS or well-resolved LES on a database of flows, predict the Reynolds stresses on a new flow for which only RANS data are available. The flows used to fit the machine learning model are referred to as \emph{training flows}, while the flow for which the model is evaluated is referred to as the \emph{test flow}. It is assumed that the training flows and the test flow have similar flow physics. Wang et al.~\cite{Wang2016} trained machine learning models to predict a corrector for the Reynolds stress tensor. The procedure is summarized as follows: \begin{enumerate} \item Perform baseline RANS simulations on both the training flows and the test flow. \item Compute the feature vector $\mathbf{q}(\mathbf{x})$ based on the RANS state variables. \item Compute the discrepancies field $\Delta \tau_\alpha(\mathbf{x})$ of the RANS modeled Reynolds stresses for the training flows based on the high-fidelity data, where $\Delta \tau_\alpha = \tau^{RANS}_\alpha - \tau^{truth}_\alpha$. \item Construct regression functions $ f_\alpha: \mathbf{q} \mapsto \Delta \tau_\alpha$ for the discrepancies based on the training data prepared in Step 3. These regression functions were constructed using random forest regressors~\cite{liaw2002classification}. \item Compute the Reynolds stresses discrepancies for the test set by evaluating the regression functions. The Reynolds stresses can subsequently be obtained by correcting the baseline RANS predictions with the evaluated discrepancies. \end{enumerate} In machine learning terminology the discrepancies $\Delta \tau_i$ here are referred to as \emph{responses}, the feature vector $\mathbf{q}$ as the \emph{input}, and the mappings $ f_\alpha: \mathbf{q} \mapsto \Delta \tau_\alpha$ as \emph{regression functions}. There are three essential components in the physics-based machine learning framework outlined above: (1) identification of mean flow features as input, (2) representation of Reynolds stresses as responses, and (3) construction of regression functions from training data. The three components of the framework are presented below. The reader is referred to~\cite{Wang2016} for further details. \subsection{Identification of Mean Flow Features as Regression Input} Ten features based on the RANS computed mean flow fields (velocity $U_i$ and pressure $P$) are identified as inputs to the regression function, which is consistent with the mean flow features listed in the work by Wang et al.~\cite{Wang2016}. Most of these features are adopted from the work by Ling and Templeton~\cite{ling2015evaluation}, with an additional feature of mean streamline curvature. It is because that RANS models tend to be less reliable at the regions with large streamline curvature, thus including it as a mean flow feature helps in detecting those regions. Turbulence intensity is another mean flow feature in the work by Wang et al.~\cite{Wang2016}, since it is an important feature in describing turbulence. Similarly, the turbulence time scale is also chosen as a feature to better describe the turbulence. Wang et al.~\cite{Wang2016} also chose the wall-distance based Reynolds number as a mean flow feature, since the presence of wall dampens the wall normal fluctuation and thus has a important impact upon the Reynolds stress anisotropy. The pressure gradient along streamline is also used as a mean flow feature, even though a uniform acceleration of an incompressible flow does not alter the turbulence. The main reason is that Reynolds stress discrepancy between the RANS simulation and DNS/LES simulation is to be predicted, and this discrepancy is influenced by $dp/dx_i$. For example, the RANS simulation is usually more reliable for the equilibrium boundary layer, but is less reliable for the boundary layer with strong acceleration. Therefore, larger Reynolds stress discrepancy can be expected for the boundary layer with strong acceleration, and the inclusion of $dp/dx_i$ helps in distinguishing such scenario. Most of the mean flow features are normalized within the range between -1 to 1, except for the wall-distance based on Reynolds number that lies within the range between 0 to 2. With the same length of range, the Euclidean distance along different features is comparable to each other. The normalization also facilitate machine learning and is a common practice there. These mean flow features are adopted from the data-driven turbulence modeling framework~\cite{Wang2016} to ensure consistency. A detailed list of the ten mean flow features used in this work can be found in the work by Wang et al.~\cite{Wang2016}. However, it should be noted that the current work focuses on the investigation of a priori assessing the closeness of flow features between the training flows and the test flow, and the \textit{a priori} confidence assessment metrics in this work is directly applicable to other choice of mean flow features. All these features are independent under rotation, translation or reflection of the coordinate system. However, some of them are not Galilean invariant, e.g., the normalization factor $U_iU_i$ and the streamline curvature $D \bold{\Gamma}/ Ds$ that are dependent on the velocity of a moving reference frame. Therefore, the authors recommend the use of fixed coordinate systems for both the training and prediction flows. \subsection{Representation of Reynolds Stress Discrepancy as Regression Response} The discrepancies of RANS predicted Reynolds stresses, or more precisely the magnitude, shape and orientation thereof, are identified as responses of the regression functions. It has been shown that these discrepancies are likely to be universal among flows of the same characteristics, and thus the regression function constructed based on them can be extrapolated to new flows~\cite{Wang2016}. To obtain the components, the Reynolds stress tensor is decomposed as follows~\cite{emory2011modeling,xiao-mfu}: \begin{equation} \label{eq:tau-decomp} \boldsymbol{\tau} = 2 k \left( \frac{1}{3} \mathbf{I} + \mathbf{A} \right) = 2 k \left( \frac{1}{3} \mathbf{I} + \mathbf{V} \Lambda \mathbf{V}^T \right) \end{equation} where $k$ is the turbulent kinetic energy, which indicates the magnitude of $\boldsymbol{\tau}$; $\mathbf{I}$ is the second order identity tensor; $\mathbf{A}$ is the anisotropy tensor; $\mathbf{V} = [\mathbf{v}_1, \mathbf{v}_2, \mathbf{v}_3]$ and $\Lambda = \textrm{diag}[\lambda_1, \lambda_2, \lambda_3]$ with $\lambda_1+\lambda_2+\lambda_3=0$ are the orthonormal eigenvectors and eigenvalues of $\mathbf{A}$, respectively, indicating its shape and orientation. In the Barycentric triangle shown schematically in Fig.~\ref{fig:bary}, the eigenvalues $\lambda_1$, $\lambda_2$, and $\lambda_3$ are mapped to the Barycentric coordinates as follows~\cite{banerjee2007presentation}: \begin{subequations} \label{eq:lambda2c} \begin{align} C_1 & = \lambda_1 - \lambda_2 \\ C_2 & = 2(\lambda_2 - \lambda_3) \\ C_3 & = 3\lambda_3 + 1 \ , \end{align} \end{subequations} where $C_1 + C_2 + C_3 = 1$. Placing the triangle in a Cartesian coordinate $\boldsymbol{\xi} \equiv (\xi, \eta)$, the location of any point within the triangle is a convex combination of those of the three vertices, i.e., \begin{equation} \boldsymbol{\xi} = \boldsymbol{\xi}_{1c}C_1 + \boldsymbol{\xi}_{2c}C_2 + \boldsymbol{\xi}_{3c}C_3 \end{equation} where $\boldsymbol{\xi}_{1c}$, $\boldsymbol{\xi}_{2c}$, and $\boldsymbol{\xi}_{3c}$ denote coordinates of the three vertices of the triangle. Consequently, the coordinate $\boldsymbol{\xi} \equiv (\xi, \eta)$ uniquely identifies the shape of the anisotropy tensor. In this work, the discrepancies of the coordinate $\Delta \boldsymbol{\xi} \equiv (\Delta \xi, \Delta \eta)$ are chosen as the regression responses, where $\Delta \xi = \xi_{DNS}-\xi_{RANS}$ and $\Delta \eta = \eta_{DNS} - \eta_{RANS}$ represent the discrepancy between the RANS predicted Reynolds stress anisotropy and the DNS data. \subsection{Random Forest for Building Regression Functions} With the input (mean flow features $\mathbf{q}$) and responses (Reynolds stress discrepancies $\Delta\tau_i$) identified above, an algorithm is needed to map from the input to the responses. In this work, random forest regression is employed~\cite{breiman2001random}. Random forest regression is an ensemble learning technique that aggregates predictions from a number of decision trees. In decision tree learning, a tree-like model is built to predict the response variable by learning simple decision rules from the training data. While decision trees have the advantages of being computationally efficient and amenable to interpretation, they tend to overfit the data, i.e., yield models that reproduce the training data very well but predict poorly for unseen data. In random forest regression, an ensemble of trees is built with bootstrap aggregation samples (i.e., sampling with replacement) drawn from the training data~\cite{friedman2001elements}. Moreover, only a subset of randomly chosen features is used when constructing each split in each tree, which reduces the correlation among the trees in the ensemble. By aggregating a large number of trees obtained in this manner, random forests can achieve significantly improved predictive performance and largely avoid overfitting. In addition, random forests can provide an relative importance score for each input feature by counting the times of splitting within the decision trees based on the given flow feature. These importance scores reflect the influence of the choice of flow features on the training-prediction performance. More detailed discussion of the feature importance can be found in the work by Wang et al.~\cite{Wang2016}. Random forest regression is a widely used regression method in machine learning community. Compared to the neural network, the random forest is less prone to overfitting and is thus more robust as pointed out by Breiman~\cite{breiman2001random}. \begin{figure}[!htbp] \centering \includegraphics[width=0.5\textwidth]{bary-triangle} \caption{The Barycentric triangle that encloses all physically realizable states of the Reynolds stresses~\cite{banerjee2007presentation,emory2013modeling}. The position within the Barycentric triangle represents the anisotropy state of the Reynolds stress. The three corners represent the limiting states.} \label{fig:bary} \end{figure} \section{Extrapolation Metrics} \label{sec:extrapolation} \subsection{Motivation with Machine Learning Based Predictive Turbulence Modeling} The machine learning framework as summarized in Section~\ref{sec:ml-methodology} was used by Wang et al.~\cite{Wang2016} to predict Reynolds stresses in conjunction with standard RANS models. The objective was to predict the Reynolds stress in the flow over period hills at $Re=10595$. Training flows were chosen from the NASA benchmark database~\cite{nasa-web} including (1) the flow over periodic hills at $Re=1400$ (PH1400), 2800 (PH2800) and 5600 (PH5600)~\cite{breuer2009flow}, (2) the flow past a curved backward facing step (CBFS13200) at $Re=13200$~\cite{bentaleb2012large}, (3) the flow in a converging-diverging channel (CDC11300) at Re=$11300$~\cite{laval2011direct}, (4) the flow past a backward facing step (BFS4900) at $Re=4900$~\cite{le1997direct}, and (5) the flow in a channel with wavy bottom wall (WC360) at $Re=360$~\cite{maass1996direct}. The geometries and the flow characteristics of these flows are illustrated in Fig.~\ref{fig:case-intro}. The Reynolds numbers are defined based on the bulk velocity $U_b$ at the narrowest cross-section in the flow and the height $H$ of the crest or step. \begin{figure}[!htbp] \centering \subfloat[Periodic hills, $Re=10595$]{\includegraphics[width=0.45\textwidth]{Ucontour_PH_crop}}\hspace{0.5em}\\ \subfloat[Periodic hills, $Re=5600$]{\includegraphics[width=0.45\textwidth]{Ucontour_PH_Re5600}}\hspace{0.5em} \subfloat[Curved backward step, $Re=13200$]{\includegraphics[width=0.35\textwidth]{Ucontour_CBFS}}\\ \subfloat[Converging--diverging channel, $Re=11700$]{\includegraphics[width=0.48\textwidth, height=0.08\textwidth]{Ucontour_CDC}}\hspace{0.5em} \subfloat[Backward facing step $Re=4700$]{\includegraphics[width=0.32\textwidth, height=0.08\textwidth]{Ucontour_BFS}}\\ \subfloat[Wavy channel, $Re=360$]{\includegraphics[width=0.7\textwidth, height=0.1\textwidth]{Ucontour_WC}} \caption{The mean velocity field and separation bubble of the test set and different training sets. The test set is (a) the flow over periodic hills at $Re=10595$. The training sets include (b) the flow over periodic hills at $Re=5600$, $Re=2800$ and $Re=1400$ (only the flow at $Re=5600$ is shown here for simplicity), (c) the flow over curved backward facing step at $Re =13200$, (d) the flow in a converging-diverging channel at $Re = 11300$, (e) the flow over a backward facing step at $Re = 4900$, and (f) the flow in a channel with wavy bottom wall at $Re = 360$. The test set is the flow over periodic hills at $Re=10595$. All the flow fields illustrated in this figure are obtained based on RANS simulations, in which Launder-Sharma $k$-$\epsilon$ model~\citep{launder1974application} is used. The lines with arrows in panel (a) indicate the profile locations for the anisotropy presented in Figs.~\ref{fig:bary2} and~\ref{fig:bary4}.} \label{fig:case-intro} \end{figure} In this study we conducted an extensive evaluation of the machine learning based prediction of Reynolds stresses, in the context of extrapolation detection. In order to isolate the contribution of the data from each flow to the predictive capability and to simplify the performance assessment, each flow above is individually used for training as opposed to combining several flows. In general, our experiences suggest that better predictive performance is obtained when the training flow and the prediction flow are more similar. For example, since the test case is the flow over periodic hills at $Re=10595$, the predictive performance is the best when the training case is the flow in the same geometry but at a different Reynolds number $Re=5600$. In contrast, the performance is the least favorable when the backward step flow $Re=4900$ or the wavy channel flow $Re=360$ is used as training flow. Physical intuition suggests that the backward step and wavy channel cases are the furthest from the prediction flow in feature space. This is because the backward step has a sudden expansion (as opposed to the gradual expansion in the periodic hill geometry) and the wavy channel flow has a Reynolds number ($Re=360$) that is drastically different from that in the test case ($Re=10595$). The other two flows (curved step and converging--diverging channel) fall between the two extremes above because their geometries are qualitatively similar to that of the test case, and their Reynolds numbers (13200 and 11700) are comparable to 10595 as well. Hence, the predictive performances of different training cases agree well with our physical understanding of degrees of similarity between the training and test flows. However, when data-driven models are used in practical flows, the similarity between the prediction flow and various candidate sets of training flows is usually not clear. The performances of data-driven models depend on the similarity between the training flows and the test flow. Predicting the test flow with significantly different flow physics from the training flows is potentially catastrophic. To prevent these consequences that may stem from bad judgment from users of data-driven models, in this work we propose the extrapolation metrics that objectively quantifies the similarity between the training flows and the test flow. These extrapolation metrics can assess the prediction performance of the data-driven model \textit{a priori}, and they also have the potential to provide guidelines for selecting more suitable training flows to improve the prediction. \subsection{Extrapolation Metrics} \label{sec:bad-metrics} Before presenting the details of the two metrics investigated in this work, we examine a few apparently attractive candidates of extrapolation metrics based on nearest neighbor distances and marginal distributions. We discuss and illustrate why they are not suitable. The idea behind extrapolation metrics is to determine the extrapolation distance between a given test point or test set and the training data. There are several different approaches for quantifying this distance. One metric would be the \emph{nearest neighbor distance}. The nearest neighbor distance is the Euclidean distance in feature space between the test point and the nearest point in the training set. Because this nearest neighbor distance is susceptible to noise, a common variation is the $K^{th}$ nearest neighbor distance~\cite{Cover1967}, which is the Euclidean distance to the $K^{th}$ nearest point in the training set, where $K$ is some pre-determined integer. Unfortunately, these methods are unwieldy--they require retaining the entire training database to compare against. In turbulence simulations, even the mean flow data from a single simulation can consume many gigabytes of memory, so transferring the training database to each user of the machine learning model based on these training data sets is impractical. Another seemingly appealing yet equally unsuitable indicator of distance between two sets of points is the \emph{marginal probability density functions}. This is illustrated in Fig.~\ref{fig:marginal} for the simple example of a two-dimensional feature space. This figure shows two data sets $S_1$ and $S_2$ have identical marginal densities (indicated by the bell-shaped curves on the two axes) but cover distinctly different regions in the feature space. If $S_1$ is used for training to predict the response of points in $S_2$, most of the evaluations would involve aggressive extrapolations, and thus poor predictive performance would be expected. The situation will be even more pronounced in higher-dimensional feature spaces. \begin{figure}[!htbp] \centering \includegraphics[width=0.45\textwidth]{marginalPDF.pdf} \caption{Illustration of two data sets that have identical marginal distributions but cover different regions in feature space.} \label{fig:marginal} \end{figure} After evaluating a number of alternatives, we identified the Mahalanobis distance and Kernel Density Estimation (KDE) as two promising metrics for evaluating the predictive performance of training--prediction methods. In both metrics, only the inputs of the training and test data are used and information of the response is not needed. More importantly, only the statistical quantities (e.g., mean, covariance, estimated probability density function of the training set) are used, and the complete raw training data set is not needed. This characteristic leads to much lower memory consumption, which is in contrast to the nearest neighbor distance methods. The Mahalanobis distance~\cite{mahalanobis1936generalized} is an efficient method of representing the extrapolation distance that does not rely on marginal distributions. The Mahalanobis distance is defined as the distance between a point $\tilde{\mathbf{q}}$ and the mean of the training points $\boldsymbol{\mu}$, scaled by the covariance matrix $\Sigma$ of the training points, \begin{equation} \label{eq:mdist} D = \sqrt{(\tilde{\mathbf{q}} - \boldsymbol{\mu})^T \Sigma^{-1} (\tilde{\mathbf{q}} -\boldsymbol{\mu})} \end{equation} The basic idea of the Mahalanobis distance is that it measures the distance between a point and a distribution--in this case, the distribution of training points. The larger the Mahalanobis distance, the greater the degree of extrapolation. To determine the Mahalanobis distance of a test point, it is not necessary to compare against all of the training data--only the mean and covariance matrix of the training data must be saved, which is highly memory efficient. In this paper, the Mahalanobis distance has been normalized based on percentiles from the training set. The normalized Mahalanobis distance of a test point is given by $1 - \gamma_{dm}$, where $\gamma_{dm}$ is the fraction of training points with a larger raw Mahalanobis distance than the test point. This normalization ensures that the distance lies in the range between 0 and 1, with a normalized distance of 0 indicating no extrapolation and a distance of 1 indicating very high extrapolation. With this normalization convention, larger raw Mahalanobis distance leads to larger normalized distance. This is because the raw Mahalanobis distance takes into account the scattering of the training flow points by incorporating the covariance matrix $\Sigma$. Therefore, larger raw distance means the distance from the test flow point to the mean of the training flow points is greater relative to the scattering of the training flow points. In other words, fewer training points have a greater raw distance than the raw distance of the test flow point, and thus a larger normalized Mahalanobis distance can be expected. An underlying assumption for the Mahalanobis distance is that the training set distribution is a multivariate Gaussian. However, this may be a poor assumption in many turbulent flows, where multi-modal distributions may be common. Therefore, the performance of the Mahalanobis distance as an extrapolation metric will be compared against Kernel Density Estimation (KDE)~\cite{Silverman}. KDE is a method of approximating the distribution of the training data. The KDE at a point $\tilde{\mathbf{q}}$ from a distribution of training data $\mathbf{q}_{i}$ for $i=1, \cdots, n$ is: \begin{eqnarray} \hat{f} = \frac{1}{n \sigma} \sum\limits_{i=1}^{n} K\left(\frac{ \tilde{\mathbf{q}} - {\mathbf{q}}_{i}}{\sigma}\right) \end{eqnarray} Often, a Gaussian kernel is used, and the bandwidth $\sigma$ is the standard deviation of the Gaussian kernel. In this work, the bandwidth is determined based on Scott's rule~\cite{scott2015multivariate}. Unlike the Mahalanobis distance, the Gaussian KDE distance does not assume that the training set has a multivariate Gaussian distribution. Instead, it has a much less stringent assumption--that the training data distribution can be approximated as a sum of multivariate Gaussians. KDE can be used as an extrapolation metric by determining the probability that a given test point was drawn from the training data. The KDE distance has been normalized by comparing the KDE probability estimation to a uniform distribution, as shown in Eq.~\ref{KDENormalizationEq}. \begin{eqnarray} D_{KDE} = 1 - \frac{P_{KDE}}{P_{KDE} + 1/A} \label{KDENormalizationEq} \end{eqnarray} In Eq.~\ref{KDENormalizationEq}, $D_{KDE}$ is the normalized KDE distance and $P_{KDE}$ is the KDE probability estimate. $A$ is the area in state space covered by the training set: $A = \prod\limits_{i} (q_{i, max} - q_{i, min})$. This normalization therefore compares how likely a point is given the KDE probability distribution estimate versus a uniform distribution. As with the normalized Mahalanobis distance, this normalized KDE distance varies from 0 (no extrapolation) to 1 (high extrapolation). The trade-off between the Mahalanobis distance and the KDE distance is between memory efficiency and flexibility. While the Mahalanobis distance is more memory efficient--it only requires retaining the mean and covariance matrix of the training data--it also makes strong assumptions about the Gaussian nature of the training data. The KDE distance requires much more memory usage, since it stores the convolution of the Gaussian kernel with the entire training set. However, it is able to account for strongly non-Gaussian training data distributions. Therefore, a key question addressed by this paper is whether the Mahalanobis distance is an effective extrapolation metric, or whether the Gaussian assumption undermines its efficiency. The KDE method , which does not make this assumption, is less memory efficient and more computationally costly. In this paper we will investigate the effectiveness of these two extrapolation metrics in detecting regions of extrapolation for data-driven turbulence closures. It should be noted that the calculations of these two extrapolation metrics only involve the RANS simulated mean flow field. This is because the high fidelity data are usually unavailable for the flow to be predicted. For the RANS simulations, the same RANS model is used for both the training flows and the test flow, and the extrapolation metrics are calculated based on these RANS simulations. Compared to DNS/LES, RANS simulations may fail to detect a flow feature and miss the correct flow topology. For instance, the separation would be falsely suppressed with a much shallower hill for the flow over periodic hills. Assuming shallower hill in both the training flow and the test flow, a similar suppression effect would exist for both flows if $k$-$\varepsilon$ model is used, and the closeness between these two flows can still be detected. However, if such suppression only occurs in the training flows and is absent in the test flow, the extrapolation metrics would not suggest a close relationship between the training flows and the test flow, which can be seen as a conservative choice of estimating closeness between flow physics. As we used RANS simulations, the calculated extrapolation metrics would depend on the choice of RANS model. This impact of RANS model selection is especially desirable for the data-driven turbulence modeling, since the mean flow features used as inputs in the existing data-driven turbulence modeling~\cite{Wang2016,ling2016reynolds} are obtained based on RANS simulations. The extrapolation metrics are thus calculated based on the RANS simulations to assess the closeness of these inputs between the training flows and the test flow. In this work we explore the correlation of these extrapolation metrics with the machine learning prediction error for the test case where the high-fidelity data is available. By demonstrating such positive correlation, these two metrics can be used as indicators of the machine learning performance for the flows where the high-fidelity data are absent. \section{Numerical Results} \label{sec:results} In this work, the extrapolation metrics based on Mahalanobis distance and kernel density estimation are assessed for a given test set and several different training sets. The relationship between machine learning prediction performance and these extrapolation metrics is also studied. The flow over periodic hills~\cite{breuer2009flow} at $Re=10595$ is chosen as the test set. As shown in Section~\ref{sec:extrapolation}, seven different training sets are employed in this work. The flow configurations of training sets are illustrated in Fig.~\ref{fig:case-intro}. It should be noted that only the lower part ($y/H<1.2$) of the training set is used for the random forest regression due to the available high fidelity data from training sets. For all the baseline RANS simulations, Launder-Sharma $k$-$\epsilon$ model~\cite{launder1974application} is used. It should be noted that the method proposed in this work is also directly applicable to other RANS models. However, the same RANS model needs to be employed for both the calculation of extrapolation metrics and the data-driven turbulence modeling procedure to ensure consistency. The $y^+$ of the first cell center is kept less than 1 and thus no wall model is applied. The RANS simulations are performed in an open-source CFD platform OpenFOAM, using a built-in steady-state incompressible flow solver \texttt{simpleFoam}~\citep{weller1998tensorial}, in which the SIMPLE algorithm is used. We choose the steady-state solver because the flow problems investigated in this work are all steady-state problems. For the numerical schemes, second-order central difference scheme is chosen for all terms except for the convection term, which is discretized with second-order upwind scheme. The random forest regression is performed based on each of the training sets. The number of max features is set as 6, considering that there are 10 input features in this work. The number of trees is set as 100. This number is chosen by observing the out-of-bag (OOB) error to avoid possible overfitting on the training sets. We have observed the OOB error with different numbers of trees (50, 100, 150) and this error is not sensitive based on our current setting of the number of trees. Representative results for the Reynolds stress anisotropy from four of the training sets are shown in Figs.~\ref{fig:bary2} and~\ref{fig:bary4}. The Reynolds stress anisotropy profiles were taken along two lines at $x/H=2$ and $x/H=4$, both of which are within the recirculation region. It can be seen in Figs.~\ref{fig:bary2}(a) and~\ref{fig:bary4}(a) that the predicted Reynolds stress anisotropies are in good agreement with the benchmark data if the flow over periodic hills at $Re=5600$ is used as the training set. The prediction is less accurate but still captures the general pattern of benchmark data if the flow at $Re=1400$ is used as training set. If the training set is the flow over curved step, the predicted Reynolds stress anisotropy is still significantly improved versus the default RANS predictions and the pattern of the benchmark data is generally predicted as shown in Figs.~\ref{fig:bary2}(c) and~\ref{fig:bary4}(c). The predicted Reynolds stress anisotropy state is significantly less accurate if the training set is the flow in a channel with a wavy bottom wall as shown in Figs.~\ref{fig:bary2}(d) and~\ref{fig:bary4}(d). In this work, these quantities that directly predicted by the trained discrepancy functions are compared with the extrapolation metrics. The reconstructed Reynolds stress components can be found in a separate work by Wang et al.~\cite{Wang2016}. It should be noted that the specification of DNS Reynolds stress into RANS equations would reduce the robustness of the numerical simulation~\cite{thompson2016methodology,poroseva2016accuracy}. For the data-driven turbulence modeling, some attempts to propagate the predicted Reynolds stress to the mean velocity via RANS equations can be found in the work by Wang et al.~\cite{Wang2017}. The comparison in Figs.~\ref{fig:bary2} and~\ref{fig:bary4} indicates that the prediction performance depends on the choice of the training set. If the training set is closely related to the test set, e.g., the geometry is the same and the Reynolds number is slightly different, the predicted Reynolds stress anisotropy is more reliable. To assess the confidence of the prediction in the practical scenario where benchmark data are not available, the closeness between different flows needs to be defined. \begin{figure}[!htbp] \centering \includegraphics[width=0.7\textwidth]{baryPlot-legend}\\ \subfloat[Training set: PH5600]{\includegraphics[width=0.4\textwidth]{bary_Re5600_x2}}\hspace{0.15em} \subfloat[Training set: PH1400]{\includegraphics[width=0.4\textwidth]{bary_Re1400_x2}}\\ \subfloat[Training set: CBFS13200]{\includegraphics[width=0.4\textwidth]{bary_CBFS_x2}}\hspace{0.15em} \subfloat[Training set: WC360]{\includegraphics[width=0.4\textwidth]{bary_WC_x2}}\\ \caption{Prediction of Reynolds stress anisotropy for the flow over periodic hills at $Re=10595$ along the line at $x/H=2$. The panels show the Reynolds stress anisotropy (a) with the flow over periodic hills at $Re=5600$ (PH5600) as training set, (b) with the flow over periodic hills at $Re=1400$ (PH1400) as training set, (c) with the flow over curved backward facing step (CBFS13200) as training set and (d) with the flow in wavy channel (WC360) as training set.} \label{fig:bary2} \end{figure} \begin{figure}[!htbp] \centering \includegraphics[width=0.7\textwidth]{baryPlot-legend}\\ \subfloat[Training set: PH5600]{\includegraphics[width=0.4\textwidth]{bary_Re5600_x4}}\hspace{0.15em} \subfloat[Training set: PH1400]{\includegraphics[width=0.4\textwidth]{bary_Re1400_x4}}\\ \subfloat[Training set: CBFS13200]{\includegraphics[width=0.4\textwidth]{bary_CBFS_x4}}\hspace{0.15em} \subfloat[Training set: WC360]{\includegraphics[width=0.4\textwidth]{bary_WC_x4}}\\ \caption{Prediction of Reynolds stress anisotropy for the flow over periodic hills at $Re=10595$ along the line at $x/H=4$. The panels show the Reynolds stress anisotropy (a) with the flow over periodic hills at $Re=5600$ (PH5600) as training set, (b) with the flow over periodic hills at $Re=1400$ (PH1400) as training set, (c) with the flow over curved backward facing step (CBFS13200) as training set and (d) with the flow in wavy channel (WC360) as training set.} \label{fig:bary4} \end{figure} We use the Mahalanobis distance and the KDE distance as extrapolation metrics to gauge the closeness between different flows in feature space. The distribution of normalized Mahalanobis distance between the flow over periodic hills at $Re=10595$ and each training set is shown in Fig.~\ref{fig:Mdist-pdf}. For the training set with same geometry but different Reynolds numbers, it can be seen in Fig.~\ref{fig:Mdist-pdf} that the mean Mahalanobis distance increases as the difference between the Reynolds numbers becomes larger. This is consistent with the empirical knowledge that flows with same geometric configuration are often closely related if the Reynolds number difference is small. For the training sets with a different geometry, it can be seen in Fig.~\ref{fig:Mdist-pdf}(b) that the Mahalanobis distances are generally greater than that shown in Fig.~\ref{fig:Mdist-pdf}(a). In addition, the Mahalanobis distances based on the flow in a wavy channel are generally larger than the Mahalanobis distances based on the flow over curved step. Compared with the prediction performance as shown in Figs.~\ref{fig:bary2} and~\ref{fig:bary4}, this suggests that the Mahalanobis distance can be an indicator to estimate the prediction performance when the benchmark data are not available. In practice, the normalized Mahalanobis distances can be obtained \textit{a priori} based on the mean flow features from training set and test set. Generally speaking, smaller mean Mahalanobis distance indicates that the RANS simulation of the training set is more similar to the RANS simulation of the test set. Assuming that similar RANS simulation results indicate similar flow physics, better prediction performance can be expected. \begin{figure}[!htbp] \centering \subfloat[Reynolds number extrapolation]{\includegraphics[width=0.495\textwidth]{distribution-Dist-Re}}\hspace{0.1em} \subfloat[Geometry extrapolation]{\includegraphics[width=0.495\textwidth]{distribution-Dist-Geo}}\\ \caption{Probability density function of Mahalanobis distance based on different training sets. (a) The distribution of Mahalanobis distance based on Reynolds number extrapolation. (b) The distribution of Mahalanobis distance based on geometry extrapolation. All the Mahalanobis distances have been normalized into the range from zero to one, as detailed in Section~\ref{sec:extrapolation}.} \label{fig:Mdist-pdf} \end{figure} Similar to the Mahalanobis distance, the KDE distances are generally smaller for Reynolds number extrapolation as shown in Fig.~\ref{fig:kde-dist-pdf}(a), compared with the KDE distances based on geometry extrapolation as shown in Fig.~\ref{fig:kde-dist-pdf}(b). This is consistent with the Mahalanobis distance results as shown in Fig.~\ref{fig:Mdist-pdf}, indicating that Mahalanobis distance can provide an overall reliable extrapolation estimation in spite of its Gaussian distribution assumption. \begin{figure}[!htbp] \centering \subfloat[Reynolds number extrapolation]{\includegraphics[width=0.495\textwidth]{kde-distance-re}}\hspace{0.1em} \subfloat[Geometry extrapolation]{\includegraphics[width=0.495\textwidth]{kde-distance-Geo}}\\ \caption{Probability density function of KDE distance based on different training sets. (a) The distribution of KDE distance based on Reynolds number extrapolation. (b) The distribution of KDE distance based on geometry extrapolation. All the KDE distances have been normalized by the estimated area occupied by training set in feature space.} \label{fig:kde-dist-pdf} \end{figure} By analyzing the mean prediction error of Reynolds stress anisotropy based on different training sets, we can see that there is a positive correlation between mean prediction error of Reynolds stress anisotropy and the extrapolation metrics as shown in Fig.~\ref{fig:MSE}. Figure~\ref{fig:MSE} also shows that the Mahalanobis distance based on geometry extrapolation is close to one and the standard deviation is much smaller than that based on Reynolds number extrapolation. A possible explanation is that the normalization procedure may lead to a dense clustering near the value of one. Compared to the KDE distance as shown in Fig.~\ref{fig:MSE}(b), it can be seen in Fig.~\ref{fig:MSE}(a) that the trend of Mahalanobis distance is similar to that of KDE distance for different training sets. This result suggests that the Mahalanobis distance, while simpler, may still be effective as an extrapolation metric on turbulence data sets. \begin{figure}[!htbp] \centering \subfloat[Mahalanobis distance]{\includegraphics[width=0.495\textwidth]{mean-Mdist-error.pdf}}\hspace{0.1em} \subfloat[KDE distance]{\includegraphics[width=0.495\textwidth]{mean-kdeDist-error.pdf}} \caption{ The relationship between the mean error of the prediction of Reynolds stress anisotropy and the mean value of extrapolation metrics, i.e., (a) Mahalanobis distance and (b) KDE distance. The standard deviations of both extrapolation metrics are shown as the horizontal bars. The standard deviations of the prediction error of anisotropy are also shown as the vertical bars.} \label{fig:MSE} \end{figure} After demonstrating the positive correlation between prediction error and extrapolation metrics through an integral view as shown in Fig.~\ref{fig:MSE}, the next step is to investigate the relationship between local prediction performance and extrapolation metrics as shown in Figs.~\ref{fig:bary_Re5600} to~\ref{fig:bary_WC}. The predictions of Reynolds stress anisotropy as shown in Figs.~\ref{fig:bary_Re5600}(a) and~\ref{fig:bary_Re5600}(b) have a good agreement with the benchmark data, which demonstrates that the prediction performance is satisfactory. The training set is the flow over periodic hills at $Re=5600$. It should be noted that such satisfactory prediction performance is not necessarily guaranteed for Reynolds number extrapolation. In this work, there is no significant change of flow physics for the flows over periodic hills from $Re=5600$ to $Re=10595$, which explains the successful Reynolds number extrapolation as shown in Fig.~\ref{fig:bary_Re5600}. For other flows such as boundary layer transition, it is unlikely to achieve the similar quality of prediction by training on the boundary layer without transition and predicting one with a transition. Therefore, the satisfactory Reynolds number extrapolation results shown here should be interpreted with caution. The KDE distance is close to zero in most areas as shown in Fig.~\ref{fig:bary_Re5600}(d), indicating that the extrapolation extent from training set to test set is small. It is consistent with the prediction performance of Reynolds stress anisotropy as shown in Figs.~\ref{fig:bary_Re5600}(a) and~\ref{fig:bary_Re5600}(b). In addition, it can be seen in Fig.~\ref{fig:bary_Re5600}(d) that the KDE distance is large near the bottom wall at the regions from $x/H=1$ to $x/H=2$ and from $x/H=2$ to $x/H=3$. The reason is that the training set only has data along $x/H=1$, 2 and 3, and the extrapolation extent is expected to be high between these lines due to the flow separation, which leads to a rapid change of flow features. Compared to KDE distance, the Mahalanobis distance is generally high near the bottom wall at the region from $x/H=1$ to $x/H=3$, which indicates that Mahalanobis distance is a more rough estimation of extrapolation compared with KDE distance. \begin{figure}[!htbp] \centering \includegraphics[width=0.5\textwidth]{pred-legend}\\ \subfloat[Prediction of $\xi$]{\includegraphics[width=0.495\textwidth]{pehill10595-Xi-Re5600}}\hspace{0.1em} \subfloat[Prediction of $\eta$]{\includegraphics[width=0.495\textwidth]{pehill10595-Eta-Re5600}}\\ \includegraphics[width=0.35\textwidth]{contour-legend}\\ \vspace{-0.8em} \subfloat[Mahalanobis distance]{\includegraphics[width=0.45\textwidth, height=0.12\textwidth]{Mdist-Re5600-crop}}\hspace{0.1em} \subfloat[KDE distance]{\includegraphics[width=0.45\textwidth, height=0.12\textwidth]{kde-Re5600-crop}}\\ \caption{Prediction of Reynolds stress anisotropy for the flow over periodic hills at $Re=10595$ along (a) horizontal direction $\xi$ and (b) vertical direction $\eta$ of Barycentric triangle. \textbf{The training set is the flow over periodic hills at \boldmath{$Re=5600$}}. The profiles are shown at 10 locations, $x/H$ = 0, 0.5, 1, 2, ..., 8. The benchmark discrepancy is obtained based on the Reynolds stress from LES simulation~\cite{breuer2009flow}. The contours of (c) Mahalanobis distance and (d) KDE distance are presented for comparison.} \label{fig:bary_Re5600} \end{figure} The prediction of Reynolds stress anisotropy as shown in Fig.~\ref{fig:bary_CBFS} is less satisfactory than the prediction as shown in Fig.~\ref{fig:bary_Re5600}, when the flow over curved backward facing step at $Re=13200$ is used as the training set. Specifically, the prediction of anisotropy shown in Figs.~\ref{fig:bary_CBFS}(a) and~\ref{fig:bary_CBFS}(b) still have a good agreement with the benchmark data at the regions from $x/H=2$ to $x/H=4$ (the mean squared error of $\xi$ and $\eta$ are 0.0046 and 0.0095, respectively). The prediction of anisotropy is less satisfactory at the downstream region from $x/H=5$ to $x/H=7$ (the mean squared error of $\xi$ and $\eta$ are 0.0124 and 0.0097, respectively), and the prediction is even less satisfactory near the hill crest at inlet and within the contraction region from $x/H=7$ to $x/H=9$ (the mean squared error of $\xi$ and $\eta$ are 0.036 and 0.037, respectively). A possible reason is that the mean flow feature of contraction and the favorable pressure gradient is not covered in the training set, which means that the extrapolation extent at the contraction region is greater. Due to the periodic inlet boundary condition, it is expected that the extrapolation extent near the inlet would also be high. From the contour of KDE distance as shown in Fig.~\ref{fig:bary_CBFS}(d), it can be seen that such a pattern of extrapolation extent is faithfully represented. Compared to the KDE distance, the Mahalanobis distance as shown in Fig.~\ref{fig:bary_CBFS}(c) is less informative. It demonstrates that the normalized Mahalanobis distance is less able to estimate the extrapolation extent based on geometry extrapolation. \begin{figure}[!htbp] \centering \includegraphics[width=0.5\textwidth]{pred-legend}\\ \subfloat[Prediction of $\xi$]{\includegraphics[width=0.495\textwidth]{pehill10595-Xi-CBFS}}\hspace{0.1em} \subfloat[Prediction of $\eta$]{\includegraphics[width=0.495\textwidth]{pehill10595-Eta-CBFS}}\\ \includegraphics[width=0.35\textwidth]{contour-legend}\\ \vspace{-0.8em} \subfloat[Mahalanobis distance]{\includegraphics[width=0.45\textwidth, height=0.12\textwidth]{Mdist-CBFS-crop}}\hspace{0.1em} \subfloat[KDE distance]{\includegraphics[width=0.45\textwidth, height=0.12\textwidth]{kde-CBFS-crop}}\\ \caption{Prediction of Reynolds stress anisotropy for the flow over periodic hills at $Re=10595$ along (a) horizontal direction $\xi$ and (b) vertical direction $\eta$ of Barycentric triangle. \textbf{The training set is the flow over curved backward facing step at \boldmath{$Re=13200$}}. The profiles are shown at 10 locations, $x/H$ = 0, 0.5, 1, 2, ..., 8. The benchmark discrepancy is obtained based on the Reynolds stress from LES simulation~\cite{breuer2009flow}. The contours of (c) Mahalanobis distance and (d) KDE distance are presented for comparison.} \label{fig:bary_CBFS} \end{figure} Compared to the prediction of anisotropy as shown in Figs.~\ref{fig:bary_Re5600} and~\ref{fig:bary_CBFS}, the prediction performance is much less satisfactory as shown in Fig.~\ref{fig:bary_WC}, where the training case is the flow in a wavy channel at $Re=360$. The main reason is that the Reynolds number of the training set is much smaller than that of the prediction set, and extrapolation extent is expected to be high across the whole domain. From the contours of Mahalanobis distance and KDE distance as shown in Figs.~\ref{fig:bary_WC}(c) and~\ref{fig:bary_WC}(d), it can be seen that such a pattern of extrapolation extent is represented by both the Mahalanobis distance and KDE distance. Both metrics are near one for almost the entire extent of the domain. \begin{figure}[!htbp] \centering \includegraphics[width=0.5\textwidth]{pred-legend}\\ \subfloat[Prediction of $\xi$]{\includegraphics[width=0.495\textwidth]{pehill10595-Xi-WC}}\hspace{0.1em} \subfloat[Prediction of $\eta$]{\includegraphics[width=0.495\textwidth]{pehill10595-Eta-WC}}\\ \includegraphics[width=0.35\textwidth]{contour-legend}\\ \vspace{-0.8em} \subfloat[Mahalanobis distance]{\includegraphics[width=0.45\textwidth, height=0.12\textwidth]{Mdist-WC-crop}}\hspace{0.1em} \subfloat[KDE distance]{\includegraphics[width=0.45\textwidth, height=0.12\textwidth]{kde-WC-crop}}\\ \caption{Prediction of Reynolds stress anisotropy for the flow over periodic hills at $Re=10595$ along (a) horizontal direction $\xi$ and (b) vertical direction $\eta$ of Barycentric triangle. \textbf{The training set is the flow in a wavy channel at \boldmath{$Re=360$}}. The profiles are shown at 10 locations, $x/H$ = 0, 0.5, 1, 2, ..., 8. The benchmark discrepancy is obtained based on the Reynolds stress from LES simulation~\cite{breuer2009flow}. The contours of (c) Mahalanobis distance and (d) KDE distance are presented for comparison.} \label{fig:bary_WC} \end{figure} The scatter plot of local prediction error of Reynolds stress anisotropy and extrapolation metrics are presented in Figs.~\ref{fig:Mdist-err} and~\ref{fig:kde-err}. The correlation coefficients are also presented in Table.~\ref{tab:corr} for a more quantitative comparison. The scatter pattern is more random as shown in Fig.~\ref{fig:Mdist-err}(a) when the Mahalanobis distance is employed as extrapolation metric and the training set is the flow over periodic hills at $Re=5600$. A possible explanation is that when Mahalanobis distance becomes small, it is expected that the extrapolation error is also small. Therefore, other error sources, such as non-local effect that is not correlated with Mahalanobis distance, will dominate and the relationship between prediction error and Mahalanobis distance tends to become more random. \begin{table}[!ht] \begin{center} \caption {Correlation coefficients between extrapolation metrics and the prediction error of anisotropy} \label{tab:corr} \begin{tabular}{P{6cm}P{4cm}P{3cm}} \toprule Training Set & Mahalanobis Distance & KDE Distance \\ \midrule Periodic Hills ($Re=5600$) & 0.31 & 0.40 \\ Periodic Hills ($Re=2800$) & 0.38 & 0.66 \\ Periodic Hills ($Re=1400$) & 0.40 & 0.70 \\ Curved Step ($Re=13200$) & 0.28 & 0.53 \\ C-D Channel ($Re=11300$) & 0.42 & 0.54 \\ Backward Step ($Re=4900$) & 0.41 & 0.57 \\ Wavy Channel ($Re=360$) & $-0.06$ & 0.08 \\ \bottomrule \end{tabular} \end{center} \end{table} Compared to the correlation shown in Fig.~\ref{fig:Mdist-err}(a), it can be seen in Figs.~\ref{fig:Mdist-err}(b) and~\ref{fig:Mdist-err}(c) that the correlation is stronger when the training set is the flow over periodic hills at $Re=1400$ or the flow over curved backward facing step, which is also confirmed by the quantitative comparison of correlation coefficients as shown in Table~\ref{tab:corr}. In these cases, therefore, it can be seen that the extrapolation error is a dominant source of error in the regression models. If the extrapolation extent further increases, it is shown in Fig.~\ref{fig:Mdist-err}(d) that the correlation between Mahalanobis distance and prediction error becomes weak again. This is due to the clustering of Mahalanobis distance near one, which significantly reduces the resolution of Mahalanobis distance. In such a scenario, the correlation between prediction error of Reynolds stress anisotropy and Mahalanobis distance is weaker, which leads to a more random relationship between Mahalanobis distance and prediction error. \begin{figure}[!htbp] \centering \subfloat[Training set: PH5600]{\includegraphics[width=0.495\textwidth]{corr-Mdist-normalized-Re5600}}\hspace{0.1em} \subfloat[Training set: PH1400]{\includegraphics[width=0.495\textwidth]{corr-Mdist-normalized-Re1400}}\\ \subfloat[Training set: CBFS13200]{\includegraphics[width=0.495\textwidth]{corr-Mdist-normalized-CBFS}}\hspace{0.1em} \subfloat[Training set: WC360]{\includegraphics[width=0.495\textwidth]{corr-Mdist-normalized-WC}} \caption{The correlation between the Mahalanobis distance and the prediction error of Reynolds stress anisotropy. The results based on four training sets are presented, including (a) the flow over periodic hills at $Re=5600$ (PH5600), (b) the flow over periodic hills at $Re=1400$ (PH1400), (c) the flow over curved backward facing step at $Re=13200$ (CBFS13200) and (d) the flow in a wavy channel at $Re=360$ (WC360). Points are colored by the local density of the scatter plot, and the brighter color indicates the higher density.} \label{fig:Mdist-err} \end{figure} The pattern of correlation between the prediction error of Reynolds stress anisotropy and KDE distance as shown in Fig.~\ref{fig:kde-err} is similar to that based on Mahalanobis distance. Specifically, there exists a positive correlation if the training set is the flow over periodic hills at $Re=1400$ or the flow over curved backward step at $Re=13200$. The correlation is weaker if the training set is the flow over periodic hills at $Re=5600$ or the flow in a wavy channel at $Re=360$. Although the correlation pattern is similar based on Mahalanobis distance and KDE distance, the clustering pattern is different. The Mahalanobis distance based on Reynolds number extrapolation is more evenly distributed from zero to one, while the KDE distance based on Reynolds number extrapolation tends to cluster near zero. On the other hand, the Mahalanobis distance based on geometry extrapolation is more likely to cluster near one as shown in Fig.~\ref{fig:Mdist-err}(d), while such clustering is less noticeable for KDE distance as shown in Fig.~\ref{fig:kde-err}(d). \begin{figure}[!htbp] \centering \subfloat[Training set: PH5600]{\includegraphics[width=0.495\textwidth]{corr-kde-dist-Bary-Re5600}}\hspace{0.1em} \subfloat[Training set: PH1400]{\includegraphics[width=0.495\textwidth]{corr-kde-dist-Bary-Re1400}}\\ \subfloat[Training set: CBFS13200]{\includegraphics[width=0.495\textwidth]{corr-kde-dist-Bary-CBFS}}\hspace{0.1em} \subfloat[Training set: WC360]{\includegraphics[width=0.495\textwidth]{corr-kde-dist-Bary-WC}} \caption{The correlation between the KDE distance and the prediction error of Reynolds stress anisotropy. The results based on four training sets are presented, including (a) the flow over periodic hills at $Re=5600$ (PH5600), (b) the flow over periodic hills at $Re=1400$ (PH1400), (c) the flow over curved backward facing step at $Re=13200$ (CBFS13200) and (d) the flow in a wavy channel at $Re=360$ (WC360). Points are colored by the local density of the scatter plot, and the brighter color indicates the higher density.} \label{fig:kde-err} \end{figure} The correlation coefficient is smaller based on Mahalanobis distance as shown in Table~\ref{tab:corr}, indicating that the prediction error of Reynolds stress anisotropy is less correlated with Mahalanobis distance. In addition, the correlation coefficient based on Mahalanobis distance can become negative if the extrapolation extent is high, e.g., the flow in a wavy channel at $Re=360$ is used as training set. Therefore, Mahalanobis distance is less accurate in estimating the extrapolation extent and prediction performance compared with the KDE distance. This is consistent with our expectation since the Gaussian distribution assumption in calculating Mahalanobis distance may not be appropriate. However, it should be noted that the Gaussian distribution assumption reduces the memory usage and the computational cost of extrapolation metrics, which is the advantage of the Mahalanobis distance. \section{Conclusion} \label{sec:conclusion} Recently, the increasing interest in data-driven turbulence modeling has created a demand for simple metrics for evaluating extrapolation. Such \textit{a priori} evaluation can provide an estimate of the predictive performance in real applications where high fidelity data are not available for the flow to be predicted. In addition, it can guide the choice of training flows to achieve better prediction performance. In the present work we discuss the evaluation of extrapolation between different flows in feature space. The Mahalanobis distance and KDE distance are used as two extrapolation metrics in feature space to measure the closeness between different flows. Specifically, the flow over periodic hills at $Re=10595$ is used as the test set in this work. Three training sets at different Reynolds numbers and four training sets with different geometries are individually used to investigate the relationship between the prediction error and the extrapolation metrics. In particular, two training sets at different Reynolds numbers and two training sets with different geometries are chosen for detailed analysis. The results demonstrate that the relationship between extrapolation metrics and the prediction error is less correlated in two extreme scenarios, i.e., when the training set is very similar to the test set or very different from the test set. In the former case, other sources of error besides extrapolation error dominate the prediction uncertainty. In the later case, the degree of extrapolation is so high that the machine learning algorithm is just ``guessing'', leading to a plateau in the error rate. Except for these two extreme scenarios, both the Mahalanobis distance and the KDE distance are positively correlated with the prediction error, demonstrating that both extrapolation metrics can be used in estimating the extrapolation extent and prediction performance. However, the quantitative comparison of correlation coefficient shows that the prediction error is less correlated with the Mahalanobis distance, indicating that the estimation of extrapolation extent based on Mahalanobis distance is less accurate. On the other hand, the Gaussian distribution assumption of Mahalanobis distance reduces the memory usage and the computational cost, which is the advantage of using Mahalanobis distance. In conclusion, the KDE distance is preferable if the accuracy of extrapolation estimation is more important, while the Mahalanobis distance is still acceptable in some applications in which the memory usage or the computational cost of high dimensional kernel density estimation is not affordable. \section*{Acknowledgment} HX would like to thank Dr. Eric G. Paterson for numerous helpful discussions during this work. Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000. SAND2016-6700J \section*{Compliance with Ethical Standards} Conflict of Interest: The authors declare that they have no conflict of interest. \bibliographystyle{elsarticle-num}
acfcf9ae15b9762d15d4cf273fb5b434dfd8a101
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Optical remote sensing has made significant advances in recent years. Among these has been the deployment and wide-spread use of hyperspectral imagery on a variety of platforms (including manned and unmanned aircraft and satellites) for a wide variety of applications, ranging from environmental monitoring, ecological forecasting, disaster relief to applications pertaining to national security. With rapid advancements in sensor technology, and the resulting reduction of size, weight and power requirements of the imagers, it is also now common to deploy multiple sensors on the same platform for multi-sensor imaging. As a specific example, it is appealing for a variety of remote sensing applications to acquire hyperspectral imagery and Light Detection and Ranging (LiDAR) data simultaneously --- hyperspectral imagery offers a rich characterization of object specific properties, while LiDAR provides topographic information that complements Hyperspectral imagery \cite{dalponte2008fusion,YP2015,brennan2006object,shimoni2011detection,pedergnana2011fusion}. Modern LiDAR systems provide the ability to record entire waveforms for every return signal as opposed to providing just the point cloud. This enables a richer representation of surface topography. While feature reduction is an important preprocessing to analysis of single-sensor high dimensional passive optical imagery (particularly hyperspectral imagery), it becomes particularly important with multi-sensor data where each sensor contributes to high dimensional raw features. A variety of feature projection approaches have been used for feature reduction, including classical approaches such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and their many variants, manifold learning approaches such as Supervised and Unsupervised Locality Preserving Projections \cite{lunga2014manifold}. Several of these methods are implemented in both the input (raw) feature space and the Reproducible Kernel Hilbert Space (RKHS) for data that are nonlinearly separable. Further, most traditional approaches to feature extraction are designed for single-sensor data --- a unique problem with multi-sensor data is that feature spaces corresponding to each sensor often have different statistical properties, and a single feature projection may hence be sub-optimal. It is hence desired to have a projection for feature reduction that preserves the underlying information from each sensor in a lower dimensional subspace. More recently, we developed a feature projection approach, referred to as Angular Discriminant Analysis (ADA) \cite{CP2015_ADA_JSTSP,PC2013Asilomar_Sparse,PrasadCuiICASSP2013}, that was optimized for hyperspectral imagery and demonstrated robustness to spectral variability. Specifically, the approach sought a lower dimensional subspace where classes were maximally separated in an angular sense, preserving important spectral shape related characteristics. We also developed a local variant of the algorithm (LADA) that preserved angular locality in the subspace. In this paper, we propose a composite kernel implementation of this framework and demonstrate for the purpose of feature projection in multi-sensor frameworks. Specifically, by utilizing a composite kernel (a dedicated kernel for each sensor), and ADA (or LADA) for each sensor, the resulting projection is highly suitable for classification. The proposed approach serves as a very effective feature reduction algorithm for sensor fusion --- it optimally fuses multi-sensor data and projects it to a lower dimensional subspace. A traditional classifier can be employed following this, for supervised learning. We validate the method with the University of Houston multi-sensor dataset comprising of Hyperspectral and LiDAR data and show that the proposed method significantly outperforms other approaches to feature fusion. The outline of the remainder of this paper is as follows. In sec. \ref{sec:related}, we review related work. In sec. \ref{sec:proposed}, we describe the proposed approach for multi-sensor feature extraction. In sec. \ref{sec:results}, we describe the experimental setup and present results with the proposed method, comparing it to several state-of-the-art techniques to feature fusion. \section{Related Work} \label{sec:related} Traditional approaches to feature projection based dimensionality reduction such as PCA, LDA and their variants largely rely on Euclidean measures. Manifold learning approaches \cite{lunga2014manifold} also seek to preserve manifold structures and neighborhood locality through projections that preserve such structures. Other projection based approaches to feature reduction, such as Locality Preserving Projections (LPP), Local Fisher's Discriminant Analysis (LFDA) \cite{lunga2014manifold,prasad2014SMoG,MPWB2013} etc. integrate ideas of local neighborhoods through affinity matrices, into classical projection based analysis approaches such as PCA, LDA etc. As a general feature extraction approach, Euclidean distance is a reasonable choice, including for remotely sensed image analysis. However, by noting the well understood benefits of spectral angle for hyperspectral image analysis, in previous work, we developed an alternate feature projection paradigm that worked with angular distance measures instead of euclidean distance measures \cite{CP2015_ADA_JSTSP,PC2013Asilomar_Sparse,PrasadCuiICASSP2013} --- we demonstrated that when projecting hyperspectral data through this class of transformations, the resulting subspaces were very effective for downstream classification and significantly outperformed their Euclidean distance counterparts. In addition to benefits with classification, we also demonstrated other benefits of this class of methods, including robustness to illumination differences --- something that is very important for remote sensing. In other previous work, it has been shown that a reproducible kernel Hilbert space (RKHS) generated by composite kernels (a weighted linear combination of basis kernels) is very effective for multi-source fusion \cite{YP2015,wang2013multiple,tuia2010multisource,camps2006composite}. Here, we briefly review the developments related to angular discriminant analysis. This will provide a context and motivation for the proposed work in this paper that seeks to demonstrate the benefits of composite kernel angular discriminant analysis for multi-source image analysis. \subsection{Angular Discriminant Analysis} Here, we briefly review Angular Discriminant Analysis (ADA) and its locality preserving counterpart, Local Angular Discriminant Analysis (LADA). Consider a $d$-dimensional feature space (e.g. hyperspectral imagery with $d$ spectral channels). Let $\{{x}_{i} \in \mathbb{R}^{d}, y_i \in \{1,2, \ldots, c\}\}$ be the $i$-th training sample with an associated class label $y_i$, where $c$ is the number of classes. The total number of training samples in the library is $n = \sum_{l=1}^{c}{n_{l}}$, where $n_{l}$ denotes the number of training samples from class $l$. Let ${\mathit{T}} \in \mathbb{R}^{d \times r}$ be the desired projection matrix, where $r$ denotes the reduced dimensionality. We also denote symbols having unit norm with a \emph{tilde} --- this will be useful where we normalize the data to a unit norm to focus on angular separability. \subsubsection{ADA} Traditional LDA seeks to find a subspace that maximizes between-class scatter while minimizing within-class scatter, where the scatter is measured using Euclidean distances. While similar in philosophy, ADA is an entirely new approach to subspace learning that is based on \emph{angular} scatter --- it seeks a subspace where within-class angular scatter is maximized, and the between-class angular scatter is maximized. Just like LDA, the ADA optimization problem can be posed as a generalized eigenvalue problem. Specifically, ADA seeks to find a projection where the ratio of between-class inner product to within-class inner product of data samples is minimized. The within-class outer product matrix ${\mathit{O}}^{(\text{w})}$ and between-class outer product matrix ${\mathit{O}}^{(\text{b})}$ are defined as \begin{align} {\mathit{O}}^{(\text{w})} &= \sum^{c}_{l=1}\sum^{}_{i:y_i=l} \tilde{{\mu}}_{l}\tilde{{x}}_{i}^{t}, \label{eq:Ow} \\ {\mathit{O}}^{(\text{b})} &= \sum^{c}_{l=1}n_l\tilde{{\mu}}\tilde{{\mu}}_{l}^{t}. \label{eq:Ob} \end{align} {where $\tilde{{\mu}}_{l} = \frac{1}{n_l}\sum_{i:y_i=l}^{}{\tilde{{x}}_{i}}$ is the normalized mean of $l$-th class samples, and $\tilde{{\mu}} = \frac{1}{n}\sum_{i=1}^{n}{\tilde{{x}}_{i}}$ is defined as the normalized total mean.} It was shown in \cite{CP2015_ADA_JSTSP} that the projection matrix ${\mathit{T}}_{\textit{ADA}}$ of ADA can be approximated as the solution to the following trace ratio problem \begin{align} {\mathit{T}}_{\textit{ADA}} &\approx \operatornamewithlimits{argmin}_{{\mathit{T}}\in \mathbb{R}^{d \times r}}\left[\operatorname{tr}\big(({\mathit{T}}^{t}{\mathit{O}}^{(\text{w})}{\mathit{T}})^{-1}{\mathit{T}}^{t}{\mathit{O}}^{(\text{b})}{\mathit{T}}\big)\right]. \label{eq:ada2} \end{align} The projection matrix ${\mathit{T}}_{\textit{ADA}}$ can be obtained by solving the generalized eigenvalue problem involving ${\mathit{O}}^{(\text{w})}$ and ${\mathit{O}}^{(\text{b})}$. \subsubsection{LADA} Similar to LDA, ADA is a ``global'' projection in that it does not specifically promote preservation of local (neighborhood) angular relationships under the projection. We hence developed LADA in \cite{CP2015_ADA_JSTSP}, which is a local variant of ADA. The within and between-class outer product matrices of LADA are obtained as follows \begin{align} {\mathit{O}}^{(\text{lw})} &= \sum_{i,j=1}^{n}\mathit{\tilde{W}}_{ij}^{(\text{lw})}\tilde{{x}}_{i}\tilde{{x}}_{j}^{t}, \label{eq:lada_ow} \\ {\mathit{O}}^{(\text{lb})} &= \sum_{i,j=1}^{n}\mathit{\tilde{W}}_{ij}^{(\text{lb})}\tilde{{x}}_{i}\tilde{{x}}_{j}^{t}, \label{eq:lada_ob} \end{align} where the normalized weight matrices are defined as \begin{align} \mathit{\tilde{W}}^{(\text{lw})}_{ij}&= \begin{cases} \mathit{\tilde{A}}_{ij}/n_l, & \text{if $y_i,y_j=l$}, \\ 0, & \text{if $y_i\ne y_j$}, \end{cases} \label{eq:lada_w} \\ \mathit{\tilde{W}}^{(\text{lb})}_{ij}&= \begin{cases} \mathit{\tilde{A}}_{ij}(1/n-1/n_l), & \text{if $y_i,y_j=l$}, \\ 1/n, & \text{if $y_i\ne y_j$}. \end{cases} \label{eq:lada_b} \end{align} The normalized affinity matrix $\mathit{\tilde{A}}_{ij}\in [0,1]$ between $\tilde{{x}}_i$ and $\tilde{{x}}_j$ is defined as \begin{equation} \mathit{\tilde{A}}_{ij}=\exp\left(-\frac{( 2-2\tilde{{x}}_i^{t}\tilde{{x}}_j)}{\tilde{\gamma}_i\tilde{\gamma}_j}\right) , \label{eq:aff} \end{equation} where $\tilde{\gamma}_{i}= \sqrt{2-2\tilde{{x}}_i^{t}\tilde{{x}}^{(\text{knn})}_i}$ denotes the \emph{local angular scaling} of data samples in the angular neighborhood of $\tilde{{x}}_i$, and $\tilde{{x}}^{(\text{knn})}_i$ is the \emph{K}-th nearest neighbors of $\tilde{{x}}_i$. Similar to ADA, the projection matrix of LADA can be defined as \begin{align} {\mathit{T}}_{\textit{LADA}} &= \operatornamewithlimits{argmin}_{{\mathit{T}}\in \mathbb{R}^{d \times r}}\left[\operatorname{tr}\big(({\mathit{T}}^{t}{\mathit{O}}^{(\text{lw})}{\mathit{T}})^{-1}{\mathit{T}}^{t}{\mathit{O}}^{(\text{lb})}{\mathit{T}}\big)\right]. \label{eq:lada} \end{align} \section{Composite Kernel Angular Discriminant Analysis for Image Fusion} \label{sec:proposed} In this section, we develop and describe the proposed approach to multi-source feature extraction --- composite kernel angular discriminant analysis (CKADA) and its locality preserving counterpart (CKLADA). Our underlying hypothesis with this work is that even when angular information is important for optical image analysis, in a multi-source (e.g. multi-sensor scenario), having dedicated kernels (specific to each source) would result in a superior projection that addresses source-specific nonlinearities. With that goal, we extend our previous work with angular discriminant analysis by implementing it in a composite kernel reproducible kernel Hilbert space and demonstrate for a specific application of multi-sensor image analysis that the resulting subspace is highly discriminative and outperforms other subspace learning approaches. Consider a nonlinear mapping $\Phi(\cdot)$ from the input space to a RKHS $\mathcal{H}$ as follows: \begin{equation} {\Phi}: \mathbb{R}^d \rightarrow \mathcal{H}, {x} \rightarrow {\Phi}({x}). \end{equation} \noindent and a kernel function ${K}$ defined as: \begin{equation} {K}({{{x}}_i},{{{x}}_j}) = \left\langle {\Phi ({{{x}}_i}),\Phi ({{{x}}_j})} \right\rangle, \end{equation} where $\left\langle {\cdot,\cdot} \right\rangle$ is the inner product of two vectors. Consider next a set of $M$ co-registered multi-source images resulting in the following $M$-Tuple of feature vectors from co-registered images for every geolocation (co-registered pixels): $\{{x}^1, {x}^2, \hdots, {x}^M\}$, where ${x}^m \in \mathbb{R}^{d_m}$. Associated with every pixel (geolocation) for which ground truth is available, there is a class label $y \in \{1, 2, \hdots, c\}$. A composite kernel RKHS can then be constructed as \begin{equation} {K}({x}_i, {x}_j) = \sum_{m=1}^M \alpha_m {K}_m ({x}^m_i, {x}^m_j) \label{eq:kernelmix} \end{equation} \noindent where $K_m$ is a basis kernel for the $m$'th source, formed by any valid Mercer's kernel. To implement Composite Kernel ADA (CKADA), note that ${\mathit{O}}^{(\text{w})}$ and ${\mathit{O}}^{(\text{b})}$ can be reformulated as \begin{align} {\mathit{O}}^{(\text{w})} &= {\mathit{X}}{\mathit{W}}^{(\text{w})}{\mathit{X}}^{t}, \\ {\mathit{O}}^{(\text{b})} &= {\mathit{X}}{\mathit{W}}^{(\text{b})}{\mathit{X}}^{t}. \end{align} \noindent where ${\mathit{W}}^{(\text{w})}$ is given as \begin{align} \mathit{W}_{ij}^{(\text{w})}&= \begin{cases} 1/n_{l}, & \text{if $y_i,y_j=l$}, \\ 0, & \text{if $y_i\ne y_j$}. \end{cases} \label{eq:ww2} \end{align} \noindent and ${\mathit{W}}^{(\text{b})}$ is given as \begin{align} \mathit{W}_{ij}^{(\text{b})}&= \begin{cases} 1/n-1/n_l, & \text{if $y_i,y_j=l$}, \\ 1/n, & \text{if $y_i\ne y_j$}. \end{cases} \label{eq:wb2} \end{align} ADA can hence be re-expressed as the solution to the following generalized eigenvalue problem \begin{align} {\mathit{X}}{\mathit{W}}^{(\text{b})}{\mathit{X}}^{t}{\nu} = \lambda{\mathit{X}}{\mathit{W}}^{(\text{w})}{\mathit{X}}^{t}{\nu}. \label{eq:keig} \end{align} Since ${\nu}$ can be represented as a linear combination of columns of ${\mathit{X}}$, it can be formulated using a vector ${\varphi} \in \mathbb{R}^{n}$ as \begin{align} {\mathit{X}}^{t}{\nu} = {\mathit{X}}^{t}{\mathit{X}}{\varphi} = {\mathit{K}}{\varphi}, \end{align} where ${\mathit{K}}$ is a $n \times n$ symmetric kernel (Gram) matrix. Here $\mathit{K}_{ij} = \kappa({x}_{i},{x}_{j}) = \langle {x}_{i}, {x}_{j} \rangle$ represents a simple inner product kernel, but can be replaced by \eqref{eq:kernelmix} by utilizing the kernel trick. Multiplying ${\mathit{X}}^{t}$ on both sides of \eqref{eq:keig}, results in the following generalized eigenvalue problem. \begin{align} {\mathit{K}}{\mathit{W}}^{(\text{b})}{\mathit{K}}{\varphi} &= \lambda{\mathit{K}}{\mathit{W}}^{(\text{w})}{\mathit{K}}{\varphi}. \label{eq:kgen} \end{align} Let ${\Psi}=\{{\varphi}_{k}\}_{k=1}^{r}$ be the $r$ generalized eigenvectors associated with the $r$ smallest eigenvalues $\lambda_{1} \le \lambda_{2}, \ldots, \le \lambda_{r}$. A test sample ${x}_{\textit{test}}$ can be embedded in $\mathcal{H}$ via \begin{align} ({\mathit{X}}{\Psi})^{t}{x}_{\textit{test}} = {\Psi}^{t}{\mathit{X}}^{t}{x}_{\textit{test}} = {\Psi}^{t}{\mathit{K}}_{{\mathit{X}},{x}_{\textit{test}}}, \label{eq:ckladaeig} \end{align} where ${\mathit{K}}_{{\mathit{X}},{x}_{\textit{test}}}$ is a $n \times 1$ vector. Composite Kernel Local ADA (CKLADA) can likewise be implemented by replacing the weight matrices (${\mathit{W}}^{(\text{w})}$ and ${\mathit{W}}^{(\text{b})}$) above with their local counterparts defined in \eqref{eq:lada_w} and \eqref{eq:lada_b}. We note that in the proposed approach, the empirical kernel (Gram) matrix from \eqref{eq:kernelmix} that is formed as a weighted linear combination over all sources is used in the generalized eigenvalue problem for CKLADA \eqref{eq:ckladaeig}. The algorithm projects the data from $M$ sources onto a \emph{unified RKHS} through a bank of kernels individually optimized for each source. The final embedding seeks to optimally separate (in an angular sense) data in the RKHS. The linear mixture of kernel enables us to optimize each kernel (for example the kernel parameters) for each source instead of applying a single kernel for all sources, and to specify source importance (via mixing weights) to the overall analysis task at hand. \emph{Practical Considerations:} We note the following free parameters in the overall embedding that affect the subspace that is generated: Embedding dimension, $r$, mixture weights used in the composite kernel, $\{\alpha_m\}_{m=1}^M$, choice of kernel and related kernel parameters. We note that unlike some other embedding techniques such as LDA and its variants where the embedding dimension is upper bounded due to rank deficiency of the between class scatter, with composite kernel local ADA, the between class angular scatter is not rank limited, and as a result, the projection matrix resulting from the solution to the generalized eigenvalue problem does not enforce an upper bound on the embedding dimension. Hence, $r$ is a free parameter that represents the \emph{unified subspace} generated by all $M$ sources. The choice of $r$ should hence be governed by the information content (as quantified for example in the eigenspectra of the decomposition). The choice of weights can be made through cross validation or techniques such as kernel alignment --- in our experience, there is often a very wide plateau over a range of values of the weights, and hence we chose to use simple cross validation to learn weights from our training data. We utilized a standard radial basis function (RBF) kernel for each source ($K_m$), but the kernel parameter (width of the RBF kernel) is optimized for each source individually via cross validation. \emph{Classification:} We note that following a CKLADA projection, a simple classifier can be utilized for down-stream analysis. This follows from the observation that applying kernel projections while simultaneously ensuring preservation of angular locality will result in subspaces where class-specific data are compactly clustered. We validate and measure the efficacy of subspaces resulting from CKLADA by utilizing the following classifiers: (1) A K Nearest neighbor (KNN) classifier, (2) A Gaussian maximum-likelihood (ML) classifier, and a (3) sparse representation based classifier (SRC) \cite{PAMI2009_SRC}. While the choice of KNN and ML are obvious for subspaces formed by Kernel projections as noted in \cite{YP2015}, we make a remark on choice of SRC as an additional classifier to measure efficacy of subspaces --- this choice is motivated not only by the observation that SRC has emerged as a powerful classification approach for high dimensional remote sensing data \cite{Cui2014,yuan2012visual,chen2011hyperspectral,Nasser2011SRC,zhang2011sparse} and that it exploits the inherent sparsity when representing samples using training dictionaries, but also because popular solvers used (e.g. Orthogonal Matching Pursuit, OMP) are driven by inner products to learn the sparse representation and hence they essentially exploit angular information. They implicitly seek a representation where a test sample is represented sparsely in a dictionary of training data such that the atoms that eventually contribute (have non-zero, significant representation coefficients) to the representation are angularly similar to the test data samples. We hence contend that CKLADA is particularly well suited for SRC and its variants. \section{Experimental Setup and Results} \label{sec:results} \subsection{Dataset} The dataset we utilize represents a sensor fusion scenario, comprising of LiDAR pseudo-waveforms, and a hyperspectral image cube, and is popular in the remote sensing community as a benchmark. The data were acquired over the University of Houston campus and the neighboring urban area. All the images are at the same spatial resolution of 2.5 m and have the same spatial size of $349 \times 1340$. The hyperspectral image was acquired with the ITRES CASI sensor, containing 144 spectral bands, ranging from 380 nm to 1050 nm. The LiDAR DSM data was acquired by an Optech Gemini sensor and then co-registered to the hyperspectral image. The laser pulse wavelength and repetition rate were 1064 nm and 167 kHz, respectively. The instrument can make up to 4 range measurements. The total number of ground reference samples is 2832, covering 15 classes of interest, with approximately 200 samples for each class --- these were determined by photo-interpretation of high resolution optical imagery.. The groundtruth map is overlaid with the gray scale image showing one channel of the hyperspectral image in Fig. \ref{fig:img_uh} \begin{figure*}[ht] \centering \includegraphics[width=15cm]{uh_casi.png} \\ \includegraphics[width=17cm]{gt_img.jpg} \\ \includegraphics[width=15cm]{color_bar.jpg} \caption{ True color image of hyperspectral University of Houston data, and the ground truth.} \label{fig:img_uh} \end{figure*} From the dense LiDAR point cloud, a pseudo-waveform was generated for each geolocation, that is co-registered with the hyperspectral image. The pseudo-waveform was generated by quantizing the elevation into uniform sized bins, and determining the average intensity of points as a function of elevation bins. This provides us with a co-registered cube of waveform-like LiDAR data that is coregistered with our hyperspectral image. We note that like spectral reflectance profiles that have unique shapes depending on the material in the pixel, shapes of pseudo-waveform also correlate with the material and topographic properties in the image. Hence, angular measures (such as provided by CKLADA) would be appropriate for such analysis compared to Euclidean measures. \subsection{Baselines} To validate the efficacy of the subspaces generated by CKLADA and CKADA, we setup classification experiments using the University of Houston multi-sensor dataset. We used popular and commonly employed embeddings as baselines to compare against, including CKLFDA and KPCA. Each of these embeddings was used with 3 classifiers: KNN, ML and SRC. We note that CKLFDA is a composite kernel counterpart of LFDA based on Euclidean distance measures, and is the best possible multi-source embedding that can be compared with CKLADA --- a comparison of CKLFDA vs. CKLADA provides a direct understanding of the benefits of angular information for multi-source embeddings, and of the resulting algorithmic framework proposed in sec. \ref{sec:proposed}. With CKLFDA, CKADA and CKLADA, we treat each sensor (hyperspectral imagery and pseudo-waveform LiDAR) as a source, each getting its dedicated base kernel. With a single kernel KPCA, we stack features from the two sensors and project them via a single transformation based on these methods. In all cases, we use RBF kernels as our base kernels, and the width of the kernel is determined via cross validation. Other free parameters including sparsity level used in SRC, number of nearest neighbors in $K-NN$ are also determined empirically from the training data via cross-validation. \subsection{Visualization of Embeddings} To provide a visual demonstration on the power of composite kernel angular discriminant analysis for geospatial image analysis, we provide visualization of composite kernel projections CKLADA, CKADA (both angular discriminant analysis) and CKLFDA. These results are depicted in fig. \ref{fig:img_vis}. The figure depicts false color images generated by projecting the multi-sensor data onto the first three most significant eigenvectors learned from CKLADA, CKADA and CKLFDA respectively. It can be clearly seen that CKLADA (and CKADA to some degree) preserve object specific properties throughout the image (for example, the highly textured objects such as urban vegetation, residential areas etc. have their spatial context significantly preserved in the lower dimensional subspace). On the contrary, CKLFDA, which can be considered as the closest benchmark/baseline competitor does not perform as well. Further, towards the right corner of the image, we point the reader to the substantial benefit of CKLADA under cloud shadows - spatial structures under cloud shadows are visible under CKLADA (and CKADA to some extent), but not when using CKLFDA. \begin{figure*}[ht] \centering \begin{subfigure}{\linewidth} \centering \includegraphics[height=4cm]{cklada.jpg} \\ \caption{CKLADA (Proposed)} \end{subfigure} \begin{subfigure}{\linewidth} \centering \includegraphics[height=4cm]{ckada.jpg} \\ \caption{CKADA (Proposed)} \end{subfigure} \begin{subfigure}{\linewidth} \centering \includegraphics[height=4cm]{cklfda.jpg} \caption{CKLFDA \cite{YP2015}} \end{subfigure} \caption{Visualizing the projections as false color images resulting from projecting the multi-sensor data onto the first 3 significant eigenvectors using (a) CKLADA (proposed); (b) CKADA (proposed); and (c) CKLFDA.} \label{fig:img_vis} \end{figure*} \subsection{Comparative Results} Experimental results comparing performance of CKADA and CKLADA with baseline embeddings are provided. As mentioned previously, free parameters were determined empirically via cross-validation. Tab. \ref{tab:res_ss} depicts overall accuracy as a function of training sample size, ranging from a small to a sufficiently large value for the proposed and baseline embeddings with various classifiers. We notice that the proposed composite kernel angular discriminant analysis approaches (CKADA and CKLADA) provide anywhere from $5-10\%$ improvement in performance compared to state of the art (CKLFDA), and provide even higher accuracies compared to a traditional single-kernel baseline, KPCA. We note that even when using very limited training data, CKLADA is able to substantially outperform other composite kernel and single-kernel methods (using just 10 samples per class, we obtain as much as a $10\%$ gain in performance with CKLADA). Even when using a sufficiently large training sample set (e.g. 50 samples per class), CKLADA and CKADA outperform other methods. In Tab. {\ref{tab:res_src}, Tab. \ref{tab:res_ml}, and Tab. \ref{tab:res_knn},} we depict class specific accuracies, overall and average accuracies using the proposed methods and baselines using SRC, ML and KNN classifiers respectively. Once again, it is clear that CKADA and CKLADA consistently provide robust classification, particularly for the ``difficult'' classes (such as residential buildings, commercial buildings, roads, parking lots etc.). We also provide classification maps in fig. \ref{fig:map_uh} using the proposed method (CKLADA) and its closest competitor, CKLFDA, using the SRC classifier. We note that CKLADA results in a map with very little noise and misclassifications, and is particularly robust under the very challenging area in the right corner of the image that is under a cloud shadow (for e.g., when using CKLFDA, the area under a cloud shadow get systematically misclassified as water --- something that is visibly remedied by CKLADA). The improvements to misclassifications occurring over difficult classes is even more apparent with these ground cover classification maps. \begin{table*}[ht] \centering \caption{Comparison of various feature embedding algorithms for multi-sensor image analysis as a function of training sample size} \begin{tabular}{c c c c c c} \hline \multirow{2}{*}{Method} & \multicolumn{5}{c}{Number of training samples per class} \\ & 10 & 20 & 30 & 40 & 50\\ \hline CKLADA-SRC & 73.2$\pm$3.8 & 87.1$\pm$1.1 & 90.4$\pm$1.1 & 92.3$\pm$0.9 & 93.3$\pm$0.9 \\ CKADA-SRC & 74.5$\pm$2.3 & 85.8$\pm$1.2 & 89.2$\pm$1.2 & 91.2$\pm$1.1 & 92.3$\pm$0.8 \\ CKLFDA-SRC & 66.6$\pm$2.5 & 81.6$\pm$1.6 & 86.2$\pm$1.3 & 87.8$\pm$1.0 & 88.9$\pm$1.0 \\ KPCA-SRC & 67.9$\pm$1.4 & 79.3$\pm$1.3 & 84.1$\pm$1.1 & 86.7$\pm$0.8 & 88.5$\pm$0.9 \\ CKLADA-ML & 74.33$\pm$2.4 & 85.7$\pm$1.6 & 91.1$\pm$1.2 & 93.3$\pm$1.0 & 94.3$\pm$0.7 \\ CKADA-ML & 77.2$\pm$2.7 & 86.7$\pm$1.6 & 91.6$\pm$1.2 & 93.0$\pm$1.0 & 94.0$\pm$0.8 \\ CKLFDA-ML & 70.4$\pm$2.4 & 81.2$\pm$1.7 & 86.7$\pm$1.6 & 88.9$\pm$1.1 & 90.2$\pm$1.0 \\ KPCA-ML & 72.15$\pm$2.7 & 85.1$\pm$1.9 & 91.2$\pm$1.2 & 93.3$\pm$1.0 & 94.3$\pm$0.8 \\ CKLADA-KNN & 80.3$\pm$1.7 & 88.1$\pm$1.3 & 91.4$\pm$1.0 & 93.0$\pm$0.9 & 93.9$\pm$0.7 \\ CKADA-KNN & 80.3$\pm$1.6 & 87.0$\pm$1.2 & 90.1$\pm$1.0 & 91.4$\pm$0.9 & 92.4$\pm$0.8 \\ CKLFDA-KNN & 70.7$\pm$2.2 & 82.5$\pm$1.6 & 86.5$\pm$1.2 & 88.1$\pm$1.0 & 89.3$\pm$0.9 \\ KPCA-KNN & 69.7$\pm$1.6 & 79.2$\pm$1.1 & 83.7$\pm$1.3 & 86.4$\pm$0.9 & 87.8$\pm$0.9 \\ \hline \end{tabular} \label{tab:res_ss} \end{table*} \begin{table*}[ht] \centering \caption{Using proposed and baseline feature embeddings with SRC} \begin{tabular}{c||c|c||c c c c} \hline \hline \multirow{2}{*}{Class} & \multicolumn{2}{c||}{Samples} & \multicolumn{4}{c}{Methods} \\ &Train & Test & CKLADA & CKADA & CKLFDA & KPCA \\ \hline \hline Grass-healthy & 30 & 168 & 99.4$\pm$1.0 & 99.0$\pm$2.0 & 98.6$\pm$1.7 & 98.0$\pm$1.8 \\ \cline{2-3} Grass-stressed & 30 & 160 & 97.7$\pm$1.4 & 96.0$\pm$1.6 & 98.0$\pm$1.2 & 95.9$\pm$3.1 \\ \cline{2-3 Grass-synthetic & 30 & 162 & 99.7$\pm$0.5 & 99.7$\pm$0.4 & 95.8$\pm$2.5 & 98.3$\pm$1.3 \\ \cline{2-3 Tree & 30 & 158 & 98.1$\pm$1.2 & 95.9$\pm$2.7 & 98.8$\pm$1.2 & 99.5$\pm$0.5 \\ \cline{2-3 Soil & 30 & 156 & 99.8$\pm$0.3 & 98.6$\pm$0.9 & 98.5$\pm$1.7 & 97.3$\pm$2.4 \\ \cline{2-3 Water & 30 & 152 & 98.7$\pm$2.3 & 95.1$\pm$3.0 & 97.0$\pm$2.3 & 96.9$\pm$2.3 \\ \cline{2-3 Residential & 30 & 166 & 85.8$\pm$5.5 & 81.3$\pm$5.0 & 80.3$\pm$4.6 & 77.1$\pm$6.1 \\ \cline{2-3 Commercial & 30 & 161 & 86.8$\pm$5.5 & 82.9$\pm$6.4 & 77.2$\pm$6.4 & 79.5$\pm$5.6 \\ \cline{2-3 Road & 30 & 163 & 80.2$\pm$5.7 & 78.5$\pm$6.1 & 73.7$\pm$5.5 & 64.4$\pm$5.1 \\ \cline{2-3 Highway & 30 & 161 & 90.9$\pm$3.2 & 90.8$\pm$3.7 & 86.9$\pm$3.9 & 77.9$\pm$5.2 \\ \cline{2-3 Railway & 30 & 151 & 88.5$\pm$3.9 & 86.2$\pm$4.5 & 82.6$\pm$4.1 & 76.1$\pm$5.8 \\ \cline{2-3 Parking Lot 1 & 30 & 162 & 75.2$\pm$5.7 & 74.2$\pm$6.3 & 67.8$\pm$6.4 & 62.3$\pm$5.0 \\ \cline{2-3 Parking Lot 2 & 30 & 154 & 65.5$\pm$4.7 & 74.5$\pm$4.5 & 58.4$\pm$6.8 & 50.7$\pm$5.7 \\ \cline{2-3 Tennis Court & 30 & 151 & 99.5$\pm$0.5 & 98.5$\pm$1.1 & 98.6$\pm$2.0 & 96.6$\pm$2.2 \\ \cline{2-3 Running Track & 30 & 157 & 98.9$\pm$0.5 & 98.7$\pm$0.7 & 98.4$\pm$0.9 & 99.8$\pm$0.4 \\ \cline{2-3 \hline \hline OA & -- & -- & 91.0$\pm$1.0 & 89.9$\pm$0.9 & 87.3$\pm$1.2 & 84.7$\pm$1.1 \\%& 81.4$\pm$1.8 & 72.6$\pm$7.7 \\ AA & -- & -- & 91.0$\pm$2.8 & 90.0$\pm$3.2 & 87.4$\pm$3.4 & 84.7$\pm$3.5 \\%& 81.5$\pm$4.5 & 72.7$\pm$10.8 \\ \hline \end{tabular} \label{tab:res_src} \end{table*} \begin{table*}[ht] \centering \caption{Using proposed and baseline feature embeddings with Gaussian ML} \begin{tabular}{c||c|c||c c c c} \hline \hline \multirow{2}{*}{Class} & \multicolumn{2}{c||}{Samples} & \multicolumn{4}{c}{Methods} \\ &Train & Test & CKLADA & CKADA & CKLFDA & KPCA \\ \hline \hline Grass-healthy & 30 & 168 & 94.9$\pm$5.4 & 95.1$\pm$4.1 & 92.6$\pm$5.2 & 95.9$\pm$4.4 \\ \cline{2-3} Grass-stressed & 30 & 160 & 99.4$\pm$0.8 & 97.8$\pm$1.9 & 99.8$\pm$0.3 & 98.6$\pm$1.9 \\ \cline{2-3} Grass-synthetic & 30 & 162 & 96.5$\pm$0.3 & 96.8$\pm$2.7 & 88.2$\pm$4.8 & 96.7$\pm$2.7 \\ \cline{2-3} Tree & 30 & 158 & 99.7$\pm$0.8 & 99.5$\pm$0.8 & 98.9$\pm$1.7 & 99.9$\pm$0.3 \\ \cline{2-3} Soil & 30 & 156 & 97.6$\pm$2.7 & 93.2$\pm$4.4 & 95.0$\pm$4.4 & 97.4$\pm$2.5 \\ \cline{2-3} Water & 30 & 152 & 95.1$\pm$2.9 & 92.6$\pm$4.0 & 94.0$\pm$3.5 & 96.0$\pm$2.8 \\ \cline{2-3} Residential & 30 & 166 & 86.1$\pm$6.9 & 80.8$\pm$6.9 & 82.4$\pm$7.9 & 82.2$\pm$7.8 \\ \cline{2-3} Commercial & 30 & 161 & 86.1$\pm$8.3 & 91.3$\pm$6.6 & 73.5$\pm$11.8 & 86.5$\pm$9.7 \\ \cline{2-3} Road & 30 & 163 & 83.5$\pm$8.6 & 87.8$\pm$6.8 & 76.8$\pm$7.2 & 82.6$\pm$8.9 \\ \cline{2-3} Highway & 30 & 161 & 82.8$\pm$8.3 & 83.9$\pm$6.9 & 76.7$\pm$7.8 & 83.4$\pm$8.3 \\ \cline{2-3} Railway & 30 & 151 & 84.5$\pm$7.7 & 81.8$\pm$7.6 & 80.6$\pm$8.0 & 83.6$\pm$8.6 \\ \cline{2-3} Parking Lot 1 & 30 & 162 & 71.7$\pm$8.7 & 49.8$\pm$11.6 & 64.5$\pm$7.6 & 71.8$\pm$10.2 \\ \cline{2-3} Parking Lot 2 & 30 & 154 & 92.2$\pm$3.3 & 96.7$\pm$1.5 & 84.8$\pm$6.7 & 92.5$\pm$4.3 \\ \cline{2-3} Tennis Court & 30 & 151 & 98.0$\pm$2.0 & 96.7$\pm$3.4 & 95.5$\pm$2.8 & 98.5$\pm$1.9 \\ \cline{2-3} Running Track & 30 & 157 & 96.6$\pm$2.0 & 97.1$\pm$1.9 & 94.9$\pm$2.9 & 97.9$\pm$1.5 \\ \cline{2-3} \hline \hline OA & -- & -- & 90.9$\pm$1.2 & 89.3$\pm$1.3 & 86.5$\pm$1.3 & 84.6$\pm$1.6 \\ AA & -- & -- & 91.0$\pm$4.7 & 89.4$\pm$4.7 & 86.6$\pm$5.5 & 90.9$\pm$5.0 \\ \hline \end{tabular} \label{tab:res_ml} \end{table*} \begin{table*}[ht] \centering \caption{Using proposed and baseline feature embeddings with KNN} \begin{tabular}{c||c|c||c c c c} \hline \hline \multirow{2}{*}{Class} & \multicolumn{2}{c||}{Samples} & \multicolumn{4}{c}{Methods} \\ &Train & Test & CKLADA & CKADA & CKLFDA & KPCA \\ \hline \hline Grass-healthy & 30 & 168 & 98.6$\pm$3.0 & 99.4$\pm$2.3 & 97.4$\pm$2.9 & 98.2$\pm$2.8 \\ \cline{2-3} Grass-stressed & 30 & 160 & 97.5$\pm$1.5 & 96.9$\pm$1.4 & 97.3$\pm$2.0 & 97.2$\pm$2.6 \\ \cline{2-3} Grass-synthetic & 30 & 162 & 99.5$\pm$0.8 & 99.7$\pm$0.4 & 96.9$\pm$1.9 & 97.9$\pm$1.6 \\ \cline{2-3} Tree & 30 & 158 & 98.1$\pm$1.8 & 96.0$\pm$2.7 & 96.9$\pm$2.9 & 99.7$\pm$0.4 \\ \cline{2-3} Soil & 30 & 156 & 99.7$\pm$0.5 & 98.5$\pm$0.8 & 98.1$\pm$2.6 & 98.1$\pm$1.1 \\ \cline{2-3} Water & 30 & 152 & 98.4$\pm$2.4 & 94.8$\pm$2.4 & 97.3$\pm$2.2 & 96.4$\pm$1.7 \\ \cline{2-3} Residential & 30 & 166 & 84.6$\pm$5.2 & 79.6$\pm$6.3 & 73.8$\pm$6.0 & 64.4$\pm$6.7 \\ \cline{2-3} Commercial & 30 & 161 & 82.1$\pm$8.1 & 76.4$\pm$8.2 & 77.5$\pm$7.8 & 79.2$\pm$6.7 \\ \cline{2-3} Road & 30 & 163 & 83.1$\pm$4.9 & 78.7$\pm$5.1 & 74.0$\pm$5.8 & 64.3$\pm$5.3 \\ \cline{2-3} Highway & 30 & 161 & 94.3$\pm$2.7 & 93.3$\pm$3.2 & 89.0$\pm$3.4 & 80.8$\pm$3.8 \\ \cline{2-3} Railway & 30 & 151 & 93.4$\pm$3.2 & 90.3$\pm$4.2 & 83.9$\pm$5.3 & 76.0$\pm$6.3 \\ \cline{2-3} Parking Lot 1 & 30 & 162 & 78.8$\pm$5.9 & 76.6$\pm$6.4 & 70.7$\pm$6.5 & 67.8$\pm$6.5 \\ \cline{2-3} Parking Lot 2 & 30 & 154 & 65.4$\pm$5.0 & 69.6$\pm$4.1 & 54.2$\pm$4.9 & 41.0$\pm$4.7 \\ \cline{2-3} Tennis Court & 30 & 151 & 99.6$\pm$0.5 & 99.2$\pm$0.8 & 98.3$\pm$1.7 & 99.3$\pm$0.7 \\ \cline{2-3} Running Track & 30 & 157 & 99.1$\pm$0.4 & 98.8$\pm$0.6 & 97.7$\pm$1.0 & 99.2$\pm$0.7 \\ \cline{2-3} \hline \hline OA & -- & -- & 91.5$\pm$1.1 & 89.8$\pm$1.0 & 86.8$\pm$1.1 & 83.9$\pm$1.1 \\ AA & -- & -- & 91.5$\pm$3.1 & 89.8$\pm$3.3 & 86.7$\pm$3.8 & 84.0$\pm$3.4 \\ \hline \end{tabular} \label{tab:res_knn} \end{table*} \begin{figure*}[ht] \centering \begin{subfigure}{\linewidth} \centering \includegraphics[width=16.4cm]{cklada_src_20.png} \\ \caption{Classification map generated using CKLADA (Proposed) with 20 training samples per class.} \end{subfigure} \begin{subfigure}{\linewidth} \centering \includegraphics[width=16.2cm]{cklfda_src_20.png} \\ \caption{Classification map generated using CKLFDA \cite{YP2015} with 20 training samples per class.} \vspace{12pt} \includegraphics[width=15cm]{color_bar.jpg} \end{subfigure} \caption{Classification maps obtained from the multi-source (hyperspectral and pseudo-waveform LiDAR) dataset using (a) the proposed CKLADA embedding and (b) a CKLFDA (bottom) embedding, both with an SRC classifier following the embedding, and with 20 samples per class for training.} \label{fig:map_uh} \end{figure*} \section{Conclusions \label{sec:Conclusions} We presented a composite kernel variant of angular discriminant analysis and local angular discriminant analysis. Angular discriminant analysis was previously shown to be very beneficial for high dimensional hyperspectral classification. In this paper, we expanded those developments via a composite kernel and demonstrated that this paradigm can be a very useful feature embedding algorithm in multi-source scenarios, such as when fusing multiple geospatial images. We validated our results with a popular multi-sensor benchmark and demonstrated that composite kernel angular discriminant analysis consistently outperforms other feature embeddings. \bibliographystyle{IEEEtran}
45ba93c295302d5ab9571db456ed48805d2172e7
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section*{\refname}% \@mkboth{\MakeUppercase\refname}{\MakeUppercase\refname}% \list{\@biblabel{\@arabic\c@enumiv}}% {\settowidth\labelwidth{\@biblabel{#1}}% \leftmargin\labelwidth \advance\leftmargin\labelsep \@openbib@code \usecounter{enumiv}% \let\p@enumiv\@empty \itemsep=0pt \parsep=0pt \leftmargin=\parindent \itemindent=-\parindent \renewcommand\theenumiv{\@arabic\c@enumiv}}% \sloppy \clubpenalty4000 \@clubpenalty \clubpenalty \widowpenalty4000% \sfcode`\.\@m} {\def\@noitemerr {\@latex@warning{Empty `thebibliography' environment}}% \endlist} \makeatother \begin{document} \graphicspath{{Figs/}} \maketitle \begin{abstract} \noindent We propose a flexible means of estimating vector autoregressions with time-varying parameters (TVP-VARs) by introducing a latent threshold process that is driven by the absolute size of parameter changes. This enables us to dynamically detect whether a given regression coefficient is constant or time-varying. When applied to a medium-scale macroeconomic US dataset our model yields precise density and turning point predictions, especially during economic downturns, and provides new insights on the changing effects of increases in short-term interest rates over time. \end{abstract} \textbf{\small Keywords:} {\small Change point model, Threshold mixture innovations, Structural breaks, Shrinkage, Bayesian statistics, Monetary policy.}\\[-1em] \textbf{\small JEL Codes}: C11, C32, C52, E42.\\[-1em] \newpage \section{Introduction} \label{sec:intro} In the last few years, economists in policy institutions and central banks were criticized for their failure to foresee the recent financial crisis that engulfed the world economy and led to a sharp drop in economic activity. Critics argued that economists failed to predict the crisis because models commonly utilized at policy institutions back then were too simplistic. For instance, the majority of forecasting models adopted were (and possibly still are) linear and low dimensional. The former implies that the underlying structural mechanisms and the volatility of economic shocks are assumed to remain constant over time -- a rather restrictive assumption. The latter implies that only little information is exploited which may be detrimental for obtaining reliable predictions. In light of this criticism, practitioners started to develop more complex models that are capable of capturing salient features of time series commonly observed in macroeconomics and finance. Recent research \citep{stock1996evidence,cogley2002evolving,cogley2005drifts,primiceri2005time,sims2006were} suggests that, at least for US data, there is considerable evidence that the influence of certain variables appears to be time-varying. This raises additional issues related to model specification and estimation. For instance, do all regression parameters vary over time? Or is time variation just limited to a specific subset of the parameter space? Moreover, as is the case with virtually any modeling problem, the question whether a given variable should be included in the model in the first place naturally arises. Apart from deciding whether parameters are changing over time, the nature of the process that drives the dynamics of the coefficients also proves to be an important modeling decision. In a recent contribution, \cite{fruhwirth2010stochastic} focus on model specification issues within the general framework of state space models. Exploiting a non-centered parametrization of the model allows them to rewrite the model in terms of a constant parameter specification, effectively capturing the steady state of the process along with deviations thereof. The non-centered parameterization is subsequently used to search for appropriate model specifications, imposing shrinkage on the steady state part and the corresponding deviations. Recent research aims to discriminate between inclusion/exclusion of elements of different variables and whether the associated regression coefficient is constant or time-varying \citep{belmonte2014hierarchical, eisenstat2016stochastic,koop2012forecasting,koop2013large, kalli2014time}. Another strand of the literature asks whether coefficients are constant or time-varying by assuming that the innovation variance in the state equation is characterized by a change point process \citep{mcculloch1993bayesian,gerlach2000efficient, koop2009evolution, giordani2012efficient}. However, the main drawback of this modeling approach is the severe computational burden originating from the need to simulate additional latent states for each parameter. This renders estimation of large dimensional models like vector autoregressions (VARs) unfeasible. To circumvent such problems, \cite{koop2009evolution} estimate a single Bernoulli random variable to discriminate between time constancy and parameter variation for the autoregressive coefficients, the covariances, and the log-volatilities, respectively. This assumption, however, implies that either all autoregressive parameters change over a given time frame, or none of them. Along these lines, \cite{mah-son:eff} allow for independent breaks in regression coefficients and the volatility parameters. However, they show that their multivariate approach is inferior to univariate change point models when out-of-sample forecasts are considered and conclude that allowing for independent breaks in each series is important. In the present paper, we introduce a method that can be applied to a highly parameterized VAR model by combining ideas from the literature of latent threshold models \citep{nee-dun:bay, nakajima2013bayesian,nakajima2013dynamic,zhou2014bayesian,kimura2016identifying} and mixture innovation models. Specifically, we introduce a set of latent thresholds that controls the degree of time-variation separately for each parameter and for each point in time. This is achieved by estimating variable-specific thresholds that allow for movements in the autoregressive parameters if the proposed change of the parameter is large enough. We show that this can be achieved by assuming that the innovations of the state equation follow a threshold model that discriminates between a situation where the innovation variance is large and a case with an innovation variance set (very close) to zero. The proposed framework nests a wide variety of competing models, most notably the standard time-varying parameter model, a change-point model with an unknown number of regimes, mixtures between different models, and finally the simple constant parameter model. To assess systematically, in a data-driven fashion, which predictors should be included in the model, we impose a set of Normal-Gamma priors \citep{griffin2010inference} in the spirit of \cite{bitto2015achieving} on the initial state of the system. We illustrate the empirical merits of our approach by carrying out an extensive forecasting exercise based on a medium-scale US dataset. Our proposed framework is benchmarked against two constant parameter Bayesian VAR models with stochastic volatility and hierarchical shrinkage priors. The findings indicate that the threshold time-varying parameter VAR excels in crisis periods while being only slightly inferior during ``normal'' periods in terms of one-step-ahead log predictive likelihoods. Considering turning point predictions for GDP growth, our model outperforms the constant parameter benchmarks when upward turning points are considered while yielding similar forecasts for downward turning points. In the second part of the application, we provide evidence on the degree of time-variation of the underlying causal mechanisms for the USA. Considering the determinant of the time-varying variance-covariance matrix of the state innovations as a global measure for the strength of parameter movements, we find that those movements reach a maximum in the beginning of the 1980s while displaying only relatively modest movements before and after that period. Consequently, we investigate the effects of a monetary policy shock for the pre- and post-1980 periods separately. This exercise reveals a considerable prize puzzle in the 1960s which starts disappearing in the early 1980s. Moreover, considering the most recent part of our sample period, we find evidence for increased effectiveness of monetary policy. This is especially pronounced during the aftermath of the global financial crisis in 2008/09. The paper is structured as follows. Section~\ref{sec:framework} introduces the modeling approach, the prior setup and the corresponding MCMC algorithm for posterior simulation. Section~\ref{sec:illustration} illustrates the behavior of the model by showcasing scenarios with few, moderately many, and many jumps in the state equation. In Section~\ref{sec:application}, we apply the model to a medium-scale US macroeconomic dataset and assess its predictive capabilities against a range of competing models in terms of density and turning point predictions. Moreover, we investigate during which periods VAR coefficients display the largest amount of time-variation and consider the associated implications on dynamic responses with respect to a monetary policy shock. Finally, Section~\ref{sec:conclusion} concludes. \section{Econometric framework} \label{sec:framework} We begin by specifying a flexible model that is capable of discriminating between constant and time-varying parameters at each point in time. \subsection{A threshold mixture innovation model} \label{sec:model} Consider the following dynamic regression model, \begin{equation} y_t = \boldsymbol{x}_t' \boldsymbol{\beta}_t + u_t, ~u_t \sim \mathcal{N}(0, \sigma_t^2)\label{eq:obs}, \end{equation} where $\boldsymbol{x}_t$ is a $K$-dimensional vector of explanatory variables and $\boldsymbol{\beta}_t =(\beta_{1t},\dots,\beta_{Kt})'$ a vector of regression coefficients. The error term $u_t$ is assumed to be independently normally distributed with (potentially) time-varying variance. This model assumes that the relationship between elements of $\boldsymbol{x}_t$ and $y_t$ is not necessarily constant over time, but changes subject to some law of motion for $\boldsymbol{\beta}_t$. Typically, researchers assume that the $j$th element of $\boldsymbol{\beta}_t$ ($j = 1,\dots,K)$ follows a random walk process, \begin{equation} \beta_{jt} = \beta_{j,t-1}+e_{jt},~e_{jt} \sim \mathcal{N}(0,\vartheta_j), \label{eq:states1} \end{equation} with $\vartheta_j$ denoting the innovation variance of the latent states. \autoref{eq:states1} implies that parameters evolve gradually over time, ruling out abrupt changes. While being conceptually flexible, in the presence of only a few breaks in the parameters, this model generates spurious movements in the coefficients that could be detrimental for the empirical performance of the model \citep{d2013macroeconomic}. Thus, we deviate from \autoref{eq:states1} by specifying the innovations of the state equation $e_{jt}$ to be a mixture distribution. More concretely, let \begin{align} e_{jt} &\sim \mathcal{N}(0, \theta_{jt}),\\ \label{eq:threshold_1} \theta_{jt} &=s_{jt} \vartheta_{j1}+(1-s_{jt}) \vartheta_{j0}, \end{align} where $s_{jt}$ is an indicator variable with an unconditional Bernoulli distribution. This mechanism is closely related to an absolutely continuous spike-and-slab prior where the slab has variance $\vartheta_{j1}$ and the spike has variance $\vartheta_{j0}$ with $\vartheta_{j1} \gg \vartheta_{j0}$ \citep[see e.g.,][for an excellent survey on this class of priors]{mal-wag:com}. In the present framework we assume that the conditional distribution $p(s_{jt}|\Delta \beta_t)$ follows a threshold process, \begin{equation} s_{jt} = \begin{cases} 1 ~\text{ if }~ |\Delta \beta_{jt}|>d_j, \\ 0 ~\text{ if } ~ |\Delta \beta_{jt}|\le d_j, \end{cases} \label{eq:threshold_2} \end{equation} where $d_j$ is a coefficient-specific threshold to be estimated and $\Delta \beta_{jt} := \beta_{jt}-\beta_{j,t-1}$. Equations (\ref{eq:threshold_1}) and (\ref{eq:threshold_2}) state that if the absolute period-on-period change of $\beta_{jt}$ exceeds a threshold $d_j$, we assume that the change in $\beta_{jt}$ is normally distributed with zero mean and variance $\vartheta_{j1}$. On the contrary, if the change in the parameter is too small, the innovation variance is set close to zero, effectively implying that $\beta_{jt} \approx \beta_{j,t-1}$, i.e.,~almost no change from period $(t-1)$ to $t$. This modeling approach provides a great deal of flexibility, nesting a plethora of simpler model specifications. The interesting cases are characterized by situations where $s_{jt}$ equals unity only for some $t$. For instance, it could be the case that parameters tend to exhibit strong movements at given points in time but stay constant for the majority of the time. An unrestricted time-varying parameter model would imply that the parameters are gradually changing over time, depending on the innovation variance in \autoref{eq:states1}. Another prominent case would be a structural break model with an unknown number of breaks \citep[for a recent Bayesian exposition, see][]{koop2007estimation}. The mixture innovation component in \autoref{eq:threshold_1} implies that we discriminate between two regimes. The first regime assumes that changes in the parameters tend to be large and important to predict $y_t$ whereas in the second regime, these changes can be safely regarded as zero, thus effectively leading to a constant parameter model over a given period of time. Compared to a standard mixture innovation model that postulates $s_{jt}$ as a sequence of independent Bernoulli variables, our approach assumes that regime shifts are governed by a (conditionally) deterministic law of motion. The main advantage of our approach relative to mixture innovation models is that instead of having to estimate a full sequence of $s_{jt}$ for all $j$, the threshold mixture innovation model only relies on a single additional parameter per coefficient. This renders estimation of high dimensional models such as vector autoregressions (VARs) feasible. The additional computational burden turns out to be negligible relative to an unrestricted TVP-VAR, see Section~\ref{sec:mcmc} for more information. Our model is also closely related to the latent thresholding approach put forward in \cite{nakajima2013bayesian} within the time series context. While in their model latent thresholding discriminates between the inclusion or exclusion of a given covariate at time $t$, our model detects whether the associated regression coefficient is constant or time-varying. \subsection{The threshold mixture innovation TTVP-VAR model} The model proposed in the previous subsection can be straightforwardly generalized to the VAR case with stochastic volatility (SV) by assuming that $\boldsymbol{y}_t$ is an $m$-dimensional response vector. In this case, \autoref{eq:obs} becomes, \begin{equation} \boldsymbol{y}_t = \boldsymbol{x}_t' \boldsymbol{\beta}_t+\boldsymbol{u}_t, \end{equation} with $\boldsymbol{x}'_t = \{\boldsymbol{I}_M \otimes \boldsymbol{z}'_t\}$, where $\boldsymbol{z}_t = (\boldsymbol{y}'_{t-1}, \dots, \boldsymbol{y}'_{t-P})'$ includes the $P$ lags of the endogenous variables.\footnote{In the empirical application, we also include an intercept term which we omit here for simplicity.} The vector $\boldsymbol{\beta}_t$ now contains the dynamic autoregressive coefficients with dimension $K=M^2p$ where each element follows the state equation given by Eqs. (\ref{eq:states1}) to (\ref{eq:threshold_2}). The vector of white noise shocks $\boldsymbol{u}_t$ is distributed as \begin{equation} \boldsymbol{u}_t \sim \mathcal{N}(\boldsymbol{0}_m, \boldsymbol{\Sigma}_t). \end{equation} Hereby, $\boldsymbol{0}_m$ denotes an $m$-variate zero vector and $\boldsymbol{\Sigma}_t = \boldsymbol{V}_t \boldsymbol{H}_t \boldsymbol{V}_t'$ is a time-varying variance-covariance matrix. The matrix $\boldsymbol{V}_t$ is a lower triangular matrix with unit diagonal and $\boldsymbol{H}_t = \text{diag}(e^{h_{1t}},\dots,e^{h_{mt}})$. We assume that the logarithm of the variances evolves according to \begin{equation} h_{it} = \mu_i + \rho_i (h_{i,t-1}+\mu_i) + \nu_{it},~\text{ for }~i=1,\dots,m, \end{equation} with $\mu_i$ and $\rho_i$ being equation-specific mean and persistence parameters and $\nu_{it} \sim \mathcal{N}(0,\zeta_i)$ is an equation-specific white noise error with variance $\zeta_i$. For the covariances in $\boldsymbol{V}_t$ we impose the random walk state equation with error variances given by \autoref{eq:threshold_1}. Conditional on the ordering of the variables it is straightforward to estimate the TTVP model on an equation-by-equation basis, augmenting the $i$th equation with the contemporaneous values of the preceding $(i-1)$ equations (for $i>1$), leading to a Cholesky-type decomposition of the variance-covariance matrix. Thus, the $i$th equation (for $i=2,\dots,m$) is given by \begin{equation} y_{it} = \tilde{\boldsymbol{z}}'_{it} \tilde{\boldsymbol{\beta}}_{it}+ u_{it}. \label{eq: vareqspecific} \end{equation} We let $\tilde{\boldsymbol{z}}_{it}= (\boldsymbol{z}'_t, y_{1t},\dots, y_{i-1, t})'$, and $\tilde{\boldsymbol{\beta}}_{it}=(\boldsymbol{\beta}_{it}', \tilde{v}_{i1, t}, \dots, \tilde{v}_{i i-1, t})'$ is a vector of latent states with dimension $K_i= Mp+i-1$. Here, $\tilde{v}_{ij, t}$ denotes the dynamic regression coefficients on the $j$th (for $j<i$) contemporaneous value showing up in the $i$th equation. Note that for the first equation we have $\tilde{\boldsymbol{z}}_{1t}= \boldsymbol{z}_t$ and $\tilde{\boldsymbol{\beta}}_{1t}=\boldsymbol{\beta}_{1t}$. The law of motion of the $j$th element of $\tilde{\boldsymbol{\beta}}_{it}$ reads \begin{equation} \tilde{\beta}_{ij,t} = \tilde{\beta}_{ij,t-1}+\sqrt{\theta_{ij,t}} r_t, ~ r_t \sim \mathcal{N}(0,1). \end{equation} Hereby, $\theta_{ij,t}$ is defined similarly to \autoref{eq:threshold_1}. In what follows, it proves to be convenient to stack the states of all equations in a $K$-dimensional vector $\tilde{\boldsymbol{\beta}}_t = (\boldsymbol{\beta}'_{1t},\dots, \tilde{\boldsymbol{\beta}}'_{Mt})'$ and let $\tilde{\beta}_{jt}$ denote the $j$th element of $\tilde{\boldsymbol{\beta}}_t$. While being clearly not order-invariant, this specific way of stating the model yields two significant computational gains. First, the matrix operations involved in estimating the latent state vector become computationally less cumbersome. Second, we can exploit parallel computing and estimate each equation simultaneously on a grid. \subsection{Prior specification} \label{sec:priors} Since our approach to estimation and inference is Bayesian, we have to specify suitable prior distributions for all parameters of the model. We impose a Normal-Gamma prior \citep{griffin2010inference} on each element of $\tilde{\boldsymbol{\beta}}_{i0}$, the initial state of the $i$th equation, \begin{equation} \tilde{\beta}_{0i,j}|\tau_{j} \sim \mathcal{N}\left(0, \frac{2}{\lambda_i^2} \tau^2_{ij}\right),~\tau^2_{ij} \sim \mathcal{G}(a_{i},a_{i}), \end{equation} for $i=1,\dots,m; j=1,\dots, K_i$. Hereby, $\lambda_i^2$ and $a_{i}$ are hyperparameters and $\tau^2_{ij}$ denotes an idiosyncratic scaling parameter that applies an individual degree of shrinkage on each element of $\tilde{\boldsymbol{\beta}}_{i0}$. The hyperparameter $\lambda_i^2$ serves as an equation-specific shrinkage parameter that shrinks all elements of $\tilde{\boldsymbol{\beta}}_{i0}$ that belong to the $i$th equation towards zero while the local shrinkage parameters $\tau_{ij}$ provide enough flexibility to also allow for non-zero values of $\tilde{\beta}_{0i, j}$ in the presence of a tight global prior specification. For the equation-specific scaling parameter $\lambda_i^2$ we impose a Gamma prior, $ \lambda_i^2 \sim \mathcal{G}(b_0,b_1), $ with $b_0$ and $b_1$ being hyperparameters chosen by the researcher. In typical applications we specify $b_0$ and $b_1$ to render this prior effectively non-influential. If the innovation variances of the observation equation are assumed to be constant over time, we impose a Gamma prior on $\sigma_i^{-2}$ with hyperparameters $c_0$ and $c_1$, i.e.,~$\sigma_i^{-2} \sim \mathcal{G}(c_0, c_1)$. By contrast, if stochastic volatility is introduced we follow \cite{kastner2014ancillarity} and impose a normally distributed prior on $\mu_i$ with mean zero and variance $100$, a Beta prior on $\rho_i$ with $(\rho_i+1)/2\sim \mathcal{B}(a_\rho,b_\rho)$, and a Gamma distributed prior on $\zeta_i \sim \mathcal{G}(1/2, 1/(2B_\zeta))$. In principle, the spike variance $\vartheta_{ij,0}$ could be estimated from the data and a suitable shrinkage prior could be employed to push $\vartheta_{ij,0}$ towards zero. However, we follow a simpler approach and estimate the slab variance $\vartheta_{ij, 1}$ only while setting $\vartheta_{ij,0} = \xi \times \hat{\vartheta}_{ij}$. Here, $\hat{\vartheta}_{ij}$ denotes the variance of the OLS estimate for automatic scaling which we treat as a constant specified a priori. The multiplier $\xi$ is set to a fixed constant close to zero, effectively turning off any time-variation in the parameters. As long as $\vartheta_{ij,0}$ is not chosen too large, the specific value of the spike variance proves to be rather non-influential in the empirical applications that follow. We use a Gamma distributed prior on the inverse of the innovation variances in the state specification in \autoref{eq:states1}, i.e.,~$ \vartheta_{ij,1}^{-1} \sim \mathcal{G}(r_{ij,0}, r_{ij, 1})$ for $i=1,\dots,m; j=1,\dots,K_i$.\footnote{Of course, it would also be possible to use a (restricted) Gamma prior on $\vartheta_{ij,1}$ in the spirit of \cite{fruhwirth2010stochastic}. However, we have encountered some issues with such a prior if the number of observations in the regime associated with $s_{ij,t}=1$ is small. This stems from the fact that the corresponding conditional posterior distribution is generalized inverse Gaussian, a distribution that is heavy tailed and under certain conditions leads to excessively large draws of $\vartheta_{ij,1}$.} Again, $r_{ij, 0}$ and $r_{ij, 1}$ denote scalar hyperparameters. This choice implies that we artificially bound $\vartheta_{ij,1}$ away from zero, implying that in the upper regime we do not exert strong shrinkage. This is in contrast to a standard time-varying parameter model, where this prior is usually set rather tight to control the degree of time variation in the parameters \citep[see, e.g.,][]{primiceri2005time}. Note that in our model the degree of time variation is governed by the thresholding mechanism instead. Finally, the prior specification of the baseline model is completed by imposing a uniform distributed prior on the thresholds, \begin{equation} d_{ij} \sim \mathcal{U}(\pi_{ij,0}, \pi_{ij,1}) \text{ for } j=1,\dots,K_i. \label{eq:priorthresholds} \end{equation} Here, $\pi_{ij,0}$ and $\pi_{ij,1}$ denote the boundaries of the prior that have to be specified carefully. In our examples, we use $\pi_{0i,j} = 0.1 \times \sqrt{\vartheta_{ij,1}}$ and $\pi_{ij,1} = 1.5 \times \sqrt{\vartheta_{ij,1}}$. This prior bounds the thresholds away from zero, implying that a certain amount of shrinkage is always imposed on the autoregressive coefficients. Setting $\pi_{ij,0}=0$ for all $i,j$ would also be a feasible option but we found in simulations that being slightly informative on the presence of a threshold improves the empirical performance of the proposed model markedly. It is worth noting that even under the assumption that $\pi_{0j}>0$, our framework performs well in simulations where the data is obtained from a non-thresholded version of our model. This stems from the fact that in a situation where parameters are expected to evolve smoothly over time, the average period-on-period change of $\beta_{ij,t}$ is small, implying that $0.1 \times \sqrt{\vartheta_{ij,1}}$ is close to zero and the model effectively shrinks small parameter movements to zero. \subsection{Posterior simulation} \label{sec:mcmc} We sample from the joint posterior distribution of the model parameters by utilizing a Markov chain Monte Carlo (MCMC) algorithm. Conditional on the thresholds $d_{ij}$, the remaining parameters can be simulated in a straightforward fashion. After initializing the parameters using suitable starting values we iterate between the following six steps. \begin{enumerate} \item We start with equation-by-equation simulation of the full history $\{\tilde{\boldsymbol{\beta}}_{it}\}_{t=0,1,\dots,T}$ by means of a standard forward filtering backward sampling algorithm \citep{carter1994gibbs, fruhwirth1994data} while conditioning on the remaining parameters of the model \item The inverse of the innovation variances of \autoref{eq:states1}, $\vartheta^{-1}_{ij},~i=1,\dots,m; j=1,\dots,K_i$ have conditional density \[ p(\vartheta^{-1}_{ij}|\bullet)=p(\vartheta^{-1}_{ij}|d_{ij},\boldsymbol{\beta}) \propto p(\boldsymbol{\beta}|\vartheta^{-1}_{ij},d_{ij})p(d_{ij}|\vartheta^{-1}_{ij})p(\vartheta^{-1}_{ij}), \] which turns out to be a Gamma distribution, i.e., \begin{equation} \vartheta^{-1}_{ij}|\bullet \sim \mathcal{G}\left(r_{ij,0} + \frac{T_{ij,1}}{2} + \frac{1}{2},r_{ij,1}+\frac{\sum_{t=1}^{T}s_{ij,t}(\tilde{\beta}_{ij,t}-\tilde{\beta}_{ij,t-1})^2}{2}\right), \end{equation} with $T_{ij,t}=\sum_{t=1}^T s_{ij, t}$ denoting the number of time periods that feature time variation in the $j$th parameter and the $i$th equation. \item Combining the Gamma prior on $\tau_{ij}^2$ with the Gaussian likelihood yields a Generalized Inverted Gaussian (GIG) distribution \begin{equation} \tau_{ij}^2|\bullet \sim \mathcal{GIG}\left(a_{ij}-\frac{1}{2}, \tilde{\beta}_{ij, 0}^2, a_{ij} \lambda_i^2\right), \end{equation} where the density of the GIG$(\kappa, \chi,\psi)$ distribution is proportional to \begin{equation} z^{\kappa-1} \exp\left\lbrace - \frac{1}{2}\left( \frac{\chi}{z}+\psi z\right)\right\rbrace. \end{equation} To sample from this distribution, we use the R package GIGrvg \citep{GIGrvg} implementing the efficient rejection sampler proposed by \cite{hoermann2013generating}. \item The global shrinkage parameter $\lambda_i^2$ is sampled from a Gamma distribution given by \begin{equation} \lambda_i^2| \bullet \sim \mathcal{G}\left(b_0+a_i K_i, b_1+\frac{a_i}{2}\sum_{j=1}^{K_i} \tau^2_{ij}\right). \end{equation} \item We update the thresholds by applying $K_i$ Griddy Gibbs steps \citep{ritter1992facilitating} per equation. Due to the structure of the model, the conditional distribution of $\tilde{\boldsymbol{\beta}}_{ij,1:T} = (\beta_{ij,1},\dots,\beta_{ij,T})'$ is \begin{equation} p\left(\tilde{\boldsymbol{\beta}}_{ij,1:T} | d_{ij}, \vartheta_{ij}\right) \propto \prod_{t=1}^T \frac{1}{\sqrt{2 \pi \theta_{ij, t} }} \exp \left\lbrace -\frac{(\tilde{\beta}_{ijt}-\tilde{\beta}_{ij,t-1})^2}{2 \theta_{ij, t}}\right\rbrace. \end{equation} This expression can be straightforwardly combined with the prior in \autoref{eq:priorthresholds} to evaluate the conditional posterior of $d_{ij}$ at a given candidate point. The procedure is repeated over a fine grid of values that is determined by the prior and an approximation to the inverse cumulative distribution function of the posterior is constructed. Finally, this approximation is used to perform inverse transform sampling. \item The coefficients of the log-volatility equation and the corresponding history of the log-volatilities are sampled by means of the algorithm brought forward by \cite{kastner2014ancillarity} which is efficiently implemented in the R package \texttt{stochvol} \citep{kastner2016dealing}. Under homoscedasticity, $\sigma_i^{-2}$ is simulated from $\sigma_i^{-2}|\bullet \sim \mathcal{G}\left(c_0+T/2, c_1+{\sum_{t=1}^T (y_{it}-\boldsymbol{z}_{it}' \tilde{\boldsymbol{\beta}}_{it})^2}/{2}\right).$ \end{enumerate} After obtaining an appropriate number of draws, we discard the first $N$ as burn-in and base our inference on the remaining draws from the joint posterior. In comparison with standard TVP-VARs, Step (5) is the only additional MCMC step needed to estimate the proposed TTVP model. Moreover, note that this update is computationally cheap, increasing the amount of time needed to carry out the analysis conducted in Section~\ref{sec:application} by around five percent. For larger models (i.e.,\ with $m$ being around $15$) this step becomes slightly more intensive but, relative to the additional computational burden introduced by applying the FFBS algorithm in Step (1), its costs are still comparably small relative to the overall computation time needed. We found that mixing and convergence properties of our proposed algorithm are similar to standard Bayesian TVP-VAR estimators. In other words, the sampling of the thresholds does not seem to substantially increase the autocorrelation of the MCMC draws. The TTVP algorithm is bundled into the R package \texttt{threshtvp} which is available from the authors upon request. \section{An illustrative example} \label{sec:illustration} In this section we illustrate our approach by means of a rather stylized example that emphasizes how well the mixture innovation component for the state innovations performs when used to approximate different data generating processes (DGPs). For demonstration purposes it proves to be convenient to start with the following simple DGP with $K=1$: \begin{align*} y_t &= x_{1t}' \beta_{1t} + u_t, ~u_t \sim \mathcal{N}(0, 0.01^2), \\ \beta_{1t} &= \beta_{1t-1} + e_{1t}, ~e_{1t} \sim \mathcal{N}(0,s_{1t} \times 0.15^2), \end{align*} where $s_{1t} \in \{0,1\}$ is chosen at random to yield paths which are characterized by many, moderately many, and few breaks. Finally, independently for all $t$, we generate $x_{1t} \sim \mathcal{U}(-1,1)$ and set $\beta_{1,0}=0$. \autoref{fig:examples} shows three possible realizations of $\beta_{1t}$ and the corresponding estimates obtained from a standard TVP model and our TTVP model. To ease comparison between the models we impose a similar prior setup for both models. Specifically, for $\sigma^{-2}$ we set $c_0=0.01$ and $c_1=0.01$, implying a rather vague prior. For the shrinkage part on $\beta_{1,0}$ we set $\lambda^2 \sim \mathcal{G}(0.01,0.01)$ and $a_1 = 0.1$, effectively applying heavy shrinkage on the initial state of the system. The prior on $\vartheta_1$ is specified as in \cite{nakajima2013bayesian}, i.e., $\vartheta^{-1}_1 \sim \mathcal{G}(3,0.03)$. To complete the prior setup for the TTVP model we set $\pi_{1,0}=0.1\times \sqrt{\vartheta_1}$ and $\pi_{1,1}=1.5\times \sqrt{\vartheta_1}$. \begin{figure}[p] \includegraphics[width=.98\textwidth, trim=30 35 25 50, clip=true]{beta_2.pdf}\\ \includegraphics[width=.98\textwidth, trim=30 35 25 50, clip=true]{beta_2_5.pdf}\\ \includegraphics[width=.98\textwidth, trim=30 35 25 50, clip=true]{beta_3.pdf} \caption{Left: Evolution of the actual state vector (dotted black) along with the posterior medians of the TVP model (dashed blue) and the TTVP model (solid red). The TTVP posterior moving probability is indicated by areas shaded in gray. Right: Demeaned posterior distribution of the TVP model (90\% credible intervals in shaded blue) and the TTVP model (90\% credible intervals in red).} \label{fig:examples} \end{figure} The left panel of \autoref{fig:examples} displays the evolution of the posterior median of a standard TVP model (in dotted blue) and of the TTVP model (in solid red) along with the actual evolution of the state vector (in dotted black). In addition, the areas shaded in gray depict the probability that a given coefficient moves over a certain time frame (henceforth labeled as posterior moving probability, PMP). The right panel shows de-meaned 90\% credible intervals of the coefficients from the TVP model (blue shaded area) and the TTVP model (solid red lines). At least two interesting findings emerge. First, note that in all three cases, our approach detects parameter movements rather well, with the PMP reaching unity in virtually all time points that feature a structural break of the corresponding parameter. The TVP model also tracks the actual movement of the states well but does so with much more high frequency variation. This is a direct consequence of the inverted Gamma prior on the state innovation variances that bound $\vartheta_1$ artificially away from zero, irrespective of the information contained in the likelihood \citep[see][for a general discussion of this issue]{fruhwirth2010stochastic}. Second, looking at the uncertainty surrounding the median estimate (right panel of \autoref{fig:examples}) reveals that our approach succeeds in shrinking the posterior variance. This is due to the fact that in periods where the true value of $\beta_{1t}$ is constant, our model successfully assumes that the estimate of the coefficient at time $t$ is also constant, whereas the TVP model imposes a certain amount of time variation. This generates additional uncertainty that inflates the posterior variance, possibly leading to imprecise inference. Thus, the TTVP model detects change points in the parameters in situations where the actual number of breaks is small, moderate and large. In situations where the DGP suggests that the actual threshold equals zero, our approach still captures most of medium to low frequency noise but shrinks small movements that might, in any case, be less relevant for econometric inference. \section{Empirical application: Macroeconomic forecasting and structural change} \label{sec:application} \subsection{Model specification and data} We use an extended version of the US macroeconomic data set employed in \cite{smets2007shocks}, \cite{geweke2012prediction} and \cite{amisano2017prediction}. Data are on a quarterly basis, span the period from 1947Q2 to 2014Q4, and comprise the log differences of consumption, investment, real GDP, hours worked, consumer prices and real wages. Last, and as a policy variable, we include the Federal Funds Rate (FFR) in levels. In the next subsections we investigate structural breaks in macroeconomic relationships by means of forecasting and impulse response analysis. Following \cite{primiceri2005time}, we include $p=2$ lags of the endogenous variables. The prior setup is similar to the one adopted in the previous sections, except that now all hyperparameters are equation-specific and feature an additional index $i=1,\dots,m$. More specifically, for all applicable $i$ and $j$, we use the following values for the hyperparameters. For the shrinkage part on the initial state of the system, we again set $\lambda_i^2 \sim \mathcal{G}(0.01,0.01)$ and $a_i = 0.1$, and the prior on $\vartheta_{ij}$ is specified to be informative with $\vartheta^{-1}_{ij} \sim \mathcal{G}(3,0.03)$. For the parameters of the log-volatility equation we use $\mu_i \sim \mathcal{N}(0, 10^2), \frac{\rho_i+1}{2} \sim \mathcal{B}(25,5)$, and $\zeta_i \sim \mathcal{G}(1/2, 1/2)$. The last ingredient missing is the prior on the thresholds where we set $\pi_{ij, 0}=0.1 \times \sqrt{\vartheta_{ij,1}}$ and $\pi_{ij, 1}=1.5 \times \sqrt{\vartheta_{ij,1}}$. For the seven-variable VAR we draw $500\,000$ samples from the joint posterior and discard the first $400\,000$ draws as burn-in. Finally, we use thinning such that inference is based on $5000$ draws out of $100\,000$ retained draws. \subsection{Forecasting evidence} We start with a simple forecasting exercise of one-step-ahead predictions. For that purpose we use an expanding window and a hold-out sample of 100 quarters. Forecasts are evaluated using log-predictive Bayes factors, which are defined as the difference of log predictive scores (LPS) of a specification of interest and a benchmark model. The log-predictive score is a widely used metric to measure density forecast accuracy \citep[see e.g.,][]{geweke2010comparing}. As the benchmark model, we use a TVP-VAR with relatively little shrinkage. This amounts to setting the thresholds equal to zero and specify the prior on $\vartheta_{ij}^{-1}\sim\mathcal{G}(3,0.03)$. We, moreover, include two additional constant parameter competitors, namely a Minne\-sota-type VAR \citep{Doan1984} and a Normal-Gamma (NG) VAR \citep{Huber2017}. All models feature stochastic volatility. In order to assess the impact of different prior hyperparameters on $\vartheta_{ij}$ and the impact of $\xi$, we estimate the TTVP model over a grid of meaningful values. \autoref{tab:LPS_1} depicts the LPS differences between a given model and the benchmark model. First, we see that all models outperform the no-shrinkage time-varying parameter VAR as indicated by positive values of the log-predictive Bayes factors. Second, constant parameter VARs with shrinkage turn out to be hard to beat. Especially the hierarchical Minnesota prior does a good job with respect to one-quarter-ahead forecasts. For the TTVP model we see that forecast performance also varies with the prior specification. More specifically, the results show that increasing $\xi$, which implies more time variation in the lower regime a-priori, deteriorates the forecasting performance. This is especially true if a large value for $\xi$ is coupled with small values of $r_{ij,0}$ and $r_{ij,1}$ -- the latter referring to the a priori belief of large swings of coefficients in the upper regime of the model. \begin{table}[t] \centering \begin{tabular}{lcrr} &$ \begin{aligned} r_{ij,0}&=3 \\ r_{ij,1}&=0.03 \end{aligned}$ & $ \begin{aligned} r_{ij,0}&=1.5 \\ r_{ij,1}&=1 \end{aligned}$ & $ \begin{aligned} r_{ij,0}&=0.001 \\ r_{ij,1}&=0.001 \end{aligned}$ \\\midrule $\xi=\xi_1 = 10^{-6}$ & 169.06 & 168.75 & 169.16 \\ $\xi = \xi_2= 10^{-5}$& 170.80 & 170.60 & 173.87 \\ $\xi =\xi_3 = 10^{-4}$ & 170.95 & 172.31 & 158.44 \\ $\xi = \xi_4 = 10^{-3}$ & 130.45 & 163.78 & 137.53 \\ \midrule & NG & Minnesota & \\ BVAR & 173.77 & 177.20 & \\ \bottomrule \end{tabular} \caption{Log predictive Bayes factors relative to a time-varying parameter VAR without shrinkage for different key parameters of the model. The final row refers to the log predictive Bayes factor of a BVAR equipped with a Normal-Gamma (NG) shrinkage prior and a hierarchical Minnesota prior. All models estimated with stochastic volatility. Numbers greater than zero indicate that a given model outperforms the benchmark.} \label{tab:LPS_1} \end{table} To investigate the predictive performance of the different model specifications further, \autoref{fig:lps_a} shows the log predictive Bayes factors relative to the benchmark model over time. The plot shows that the specifications with $\xi_1$ and $\xi_2$ excel during most of the sample period, irrespective of the prior on $\vartheta_{ij}$. The constant parameter models, by contrast, dominate only during two very distinct periods of our sample, namely at the beginning and at the end of the time span covered. In both periods, no severe up or downswings in economic activity occur and the constant parameter models with SV display excellent predictive capabilities. By contrast, during volatile periods -- such as the global financial crisis -- our modeling approach seems to pay off in terms of predictive accuracy. To investigate this in more detail, we focus on the forecasting performance of the different model specifications during the period from 2006Q1 to 2010Q1 in \autoref{fig:lps_b}. Here we see that TTVP specifications with $\xi_j$ for $j<4$ outperform all remaining competitors. This additional, and more detailed, look at the forecasting performance during turbulent times thus reveals that the TTVP model is a valuable alternative to simpler models. Put differently, we observe that during more volatile periods the TTVP model can severely outperform constant parameter models, while in tranquil times its forecasts are never far off. \begin{figure}[p] \centering \begin{subfigure}{\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 45 20 40]{LPS_benchmarks.pdf} \caption{Full evaluation period (1995Q1 to 2014Q4).}\label{fig:lps_a} \end{subfigure}\\ \begin{subfigure}{\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 45 25 40]{LPS_benchmark_crisis.pdf} \caption{Crisis period only (2006Q1 to 2010Q1).}\label{fig:lps_b} \end{subfigure} \caption{Log predictive Bayes factor relative to a TVP-VAR-SV model.}\label{fig:lps} \end{figure} Next, we examine turning point forecasts, since the detection of structural breaks might be a further useful application of the TTVP framework. We focus on turning points in real GDP growth and follow \cite{canova2004forecasting} to label time point $(t+1)$ a \emph{downward turning point} -- conditional on the information up to time $t$ -- if $S_{t+1}$, the growth rate of real GDP at time $(t+1)$, satisfies that $S_{t-2} < S_t$, $S_{t-1} < S_{t}$, and $S_t > S_{t+1}$. Analogously, the time point $(t+1)$ is labeled an \emph{upward turning point} if $S_{t-2} > S_t$, $S_{t-1}>S_{t}$, and $S_t < S_{t+1}$. Equipped with these definitions, we then can split the total number of turning points up into upturns and downturns and compute the quadratic probability (QPS) scores as an accuracy measure of upturn and downturn probability forecasts. The results are provided in \autoref{tab:QPS}. \begin{table}[t] \centering \resizebox{\columnwidth}{!}{% \begin{tabular}{lccccccc} &\multicolumn{3}{c}{Downturns} & & \multicolumn{3}{c}{Upturns}\\ &$ \begin{aligned} r_{ij,0}&=3 \\ r_{ij,1}&=0.03 \end{aligned}$ & $ \begin{aligned} r_{ij,0}&=1.5 \\ r_{ij,1}&=1 \end{aligned}$ & $ \begin{aligned} r_{ij,0}&=0.001 \\ r_{ij,1}&=0.001 \end{aligned}$ & & $ \begin{aligned} r_{ij,0}&=3 \\ r_{ij,1}&=0.03 \end{aligned}$ & $ \begin{aligned} r_{ij,0}&=1.5 \\ r_{ij,1}&=1 \end{aligned}$ & $ \begin{aligned} r_{ij,0}&=0.001 \\ r_{ij,1}&=0.001 \end{aligned}$ \\ \midrule $\xi=\xi_1 = 10^{-6}$ & 0.66 & 0.67 & 0.67 & & 0.84 & 0.83 & 0.83 \\ $\xi=\xi_2 = 10^{-5}$ & 0.66 & 0.66 & 0.67 & & 0.83 & 0.83 & 0.81 \\ $\xi=\xi_3 = 10^{-4}$ & 0.64 & 0.65 & 0.68 & & 0.83 & 0.84 & 0.80 \\ $\xi=\xi_4 = 10^{-3}$ & 0.87 & 0.67 & 0.78 & & 0.81 & 0.83 & 0.78 \\ \midrule & NG & Minnesota & & & NG & Minnesota& \\ BVAR &0.62&0.62 & & & 0.84 & 0.93 & \\ \bottomrule \end{tabular}} \caption{QPS scores relative to a time-varying parameter VAR without shrinkage for different key parameters of the model. The final row refers to the QPS score of a BVAR equipped with a Normal-Gamma (NG) shrinkage prior and a hierarchical Minnesota prior. All models estimated with stochastic volatility. Numbers below unity indicate that a given model outperforms the benchmark.} \label{tab:QPS} \end{table} The picture that arises is similar to that of the density forecasting exercise: all variants of the TTVP model beat the no-shrinkage time-varying parameter VAR model. Turning point forecasts deteriorate for larger values of $\xi$ and especially so if they are coupled with small choices for $r_{ij}$, yielding a relatively uninformative prior on $\vartheta_{ij}$ and consequently little shrinkage. Forecast gains relative to the benchmark model are more sizable for downward than for upward turning points. In comparison to the two constant parameter competitors, the TTVP model excels in predicting upward turning points (for which there are more observations in the sample), while forecast performance is slightly inferior for downward forecasts. Also note that for downward predictions, penalizing time-variation seems to be essential and consequently the strongest performance among TTVP specifications is achieved for small values of $\xi$. The opposite is the case for upward turning points where reasonable predictions can be also achieved with a rather loose prior. \subsection{Detecting structural breaks in US data} In this section we aim to have a closer and more systematic look at changes in the joint dynamics of our seven variable TTVP-VAR model for the US economy. To that end, we examine the posterior mean of the determinant of the time-varying variance-covariance matrix of the innovations in the state equation \citep{cogley2005drifts}. For each draw of $\boldsymbol{\Omega}_{it} = \text{diag}(\theta_{i1,t },\dots,\theta_{K_i1,t})$ we compute its log-determinant and subtract the mean across time. Large values of this measure point towards a pronounced degree of time-variation in the autoregressive coefficients of the corresponding equations. The results are provided in \autoref{fig:concovtrace} for each equation and the full system. \begin{figure}[p] \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{concovtrace.pdf} \caption{Consumption}\label{fig:2a} \end{subfigure} \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{invcovtrace.pdf} \caption{Investment}\label{fig:2b} \end{subfigure} \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{outcovtrace.pdf} \caption{Output}\label{fig:2c} \end{subfigure}\\[.7em] \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{houcovtrace.pdf} \caption{Hours worked}\label{fig:2d} \end{subfigure} \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{infcovtrace.pdf} \caption{Inflation}\label{fig:2e} \end{subfigure} \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{reacovtrace.pdf} \caption{Real wages}\label{fig:2f} \end{subfigure}\\[.7em] \centering \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{intcovtrace.pdf} \caption{FFR}\label{fig:2g} \end{subfigure} \begin{subfigure}{.327\textwidth} \includegraphics[width=\textwidth, clip, trim = 30 40 30 40]{det_overall.pdf} \caption{Overall}\label{fig:2h} \end{subfigure} \caption{Posterior mean of the determinant of time-varying variance-covariance matrix of the innovations to the state equation from 1947Q2 to 2014Q4. Values are obtained by taking the exponential of the demeaned log-determinant across equations. Gray shaded areas refer to US recessions dated by the NBER business cycle dating committee.} \label{fig:concovtrace} \end{figure} For all variables we see at least one prominent spike during the sample period indicating a structural break. Most spikes in the determinant occur around 1980, when then Fed chairman Paul Volcker sharply increased short-term interest rates to fight inflation. Other breaks relate to the dot-com bubble in the early 2000s (consumption), the oil price crisis and stock market crash in the early 1970s (hours worked) and another oil price related crisis in the early 1990s. Also, the transition from positive interest rates to the zero lower bound in the midst of the global financial crisis is indicated by a spike in the determinant. That we can relate spikes to historical episodes of financial and economic distress lends further confidence in the modeling approach. Among these periods, the early 1980s seem to have constituted by far the most severe rupture for the US economy. \subsection{Impulse responses to a monetary policy shock} In this section we examine the dynamic responses of a set of macroeconomic variables to a contractionary monetary policy shock. The monetary policy shock is calibrated as a 100 basis point (bp) increase in the FFR and identified using a Cholesky ordering with the variables appearing in exactly the same order as mentioned above. This ordering is in the spirit of \cite{christiano2005} and has been subsequently used in the literature \citep[see][for an excellent survey]{Coibion2012}. Drawing on the results of the previous section, we focus on two sub-sets of the sample, namely the pre-Volcker period from 1947Q4 to 1979Q1 and the rest of the sample.\footnote{The split into two sub-sets is conducted for interpretation purposes only. For estimation, the entire sample has been used.} The time-varying impulse responses -- as functions of horizons -- are displayed in \autoref{fig:irf_volcker}. Additionally, we also include impulse responses for different horizons -- as functions of time -- over the full sample period in \autoref{fig:irf1} and \autoref{fig:irf2}. \begin{figure}[p] \begin{minipage}{1\linewidth}~\\ \centering \textbf{1947Q4 to 1979Q1} \end{minipage}\\[.5em] \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_consumption.pdf} Consumption \end{minipage}% \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_investment.pdf} Investment \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_output.pdf} Output \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_hours.pdf} Hours worked \end{minipage} \vspace{-.5em} \centering \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_inflation.pdf} Inflation \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_real_wage.pdf} Real wages \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{pre_volcker_interest_rate.pdf} FFR \end{minipage}\\ \vspace{2em} \begin{minipage}{1\linewidth}~\\ \centering \textbf{1979Q2 to 2014Q4} \end{minipage}\\[.5em] \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_consumption.pdf} Consumption \end{minipage}% \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_investment.pdf} Investment \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_output.pdf} Output \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_hours.pdf} Hours worked \end{minipage} \vspace{.5em} \centering \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_inflation.pdf} Inflation \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_real_wage.pdf} Real wages \end{minipage} \begin{minipage}[b]{0.246\linewidth} \centering \includegraphics[clip, trim=20 45 20 50, width=\linewidth]{post_volcker_interest_rate.pdf} FFR \end{minipage} \caption{Posterior median impulse response functions over two sample splits, namely the pre-Volcker period (1947Q4 to 1979Q1) and the rest of the sample period (1979Q2 to 2014Q4). The coloring of the impulse responses refer to their timing: light yellow stands for the beginning of the sample split, dark red stands for the end of sample split. For reference, 68\% credible intervals over the average of the sample period provided (dotted black lines).}\label{fig:irf_volcker} \end{figure} \begin{sidewaysfigure}[p] \includegraphics[width=\textwidth]{RA.pdf} \caption{Posterior median responses to a $+100$ bp monetary policy shock, after 4 (top panels), 8 (middle panels) and 12 (bottom panels) quarters. Shaded areas correspond to 90\% (dark red) and 68\% (light red) credible sets.}\label{fig:irf1} \end{sidewaysfigure} \begin{sidewaysfigure}[p] \includegraphics[width=\textwidth]{supply.pdf} \caption{Posterior median responses to a $+100$ bp monetary policy shock, after 4 (top panels), 8 (middle panels) and 12 (bottom panels) quarters. Shaded areas correspond to 90\% (dark red) and 68\% (light red) credible sets.}\label{fig:irf2} \end{sidewaysfigure} In \autoref{fig:irf_volcker} we investigate whether the size and the shape of responses varies between and within the two sub-samples. For that purpose we show median responses over the first sample split in the top row and for the second part of the sample in the bottom row of \autoref{fig:irf_volcker}. Impulse responses that belong to the beginning of a sample split are depicted in light yellow, those that belong to the end of the sample period in dark red. To fix ideas, if the size of a response increases continuously over time we should see a smooth darkening of the corresponding impulse from light yellow to dark red. Considering, for instance, hours worked, this phenomenon can clearly be seen in in the second sample period from 1979Q2 to 2014Q4, where the median response changes gradually from slightly negative to substantially negative. On the other hand, abrupt changes are also clearly visible, see e.g.,\ the drastic change of the inflation response from 1979Q1 (the last quarter in the first sample) to 1979Q2 (the first quarter in the second sampler), dropping from substantially positive to just above zero within one quarter (see also \autoref{fig:irf1}). Considering the dynamic responses across different angles, we find three regularities which are worth emphasizing. The first concerns the overall effects of the monetary policy shock. Note that an unexpected rate increase deters investment growth, hours worked and consequently overall output growth for both sample splits. These results are reasonable from an economic perspective. Also, estimated effects on output growth and inflation are comparable to those of \citet{Baumeister2013} who use a TVP-VAR framework and US data. Responses of consumption growth tend to be accompanied by wide credible sets. The same applies to inflation and real wages. Second, we examine changes in responses over time for the first sub-period. One of the variables that shows a great deal of variation in magnitudes is the response of inflation. Here, effects become increasingly positive the further one moves from 1947Q4 to 1979Q1 and the shades of the responses turn continuously darker. These results imply a severe ``price puzzle''. While overall credible sets for the sub-sample are wide, positive responses for inflation and thus the price puzzle are estimated over the period from the mid-1960s to the beginning of the 1980s (see also \autoref{fig:irf1}). A similar picture arises when looking at consumption growth. During the first sample split, effects become increasingly more negative, but responses are only precisely estimated for the period from the mid-1960s to the beginning of the 1980s. This might be explained by the fact that the monetary policy driven increase in inflation spurs consumption since saving becomes less attractive. Third, we focus on the results over the more recent second sample split from 1979Q2 to 2014Q4. Paul Volcker's fight against inflation had some bearings on overall macroeconomic dynamics in the USA. With the onset of the 1980s, the aforementioned price puzzle starts to disappear (in the sense that effects are surrounded by wide credible sets and medium responses increasingly negative). There is also a great deal of time variation evident in other responses, mostly becoming increasingly negative. Put differently, the effectiveness of monetary policy seems to be higher in the more recent sample period than before. This can be seen by effects on hours worked, investment growth and output growth. That the effects of a hypothetical monetary policy shock on output growth are particular strong after the crisis corroborates findings of \citet{Baumeister2013} and \citet{Feldkircher2016}. The latter argue that this is related to the zero lower bound period: after a prolonged period of unaltered interest rates, a deviation from the (long-run) interest rate mean can exert considerable effects on the macroeconomy. \section{Closing remarks} \label{sec:conclusion} This paper puts forth a novel approach to estimate time-varying parameter models in a Bayesian framework. We assume that the state innovations are following a threshold model where the threshold variable is the absolute period-on-period change of the corresponding states. This implies that if the (proposed) change is sufficiently large, the corresponding variance is set to a value greater than zero. Otherwise, it is set close to zero which implies that the states remained virtually constant from $(t-1)$ to $t$. Our framework is capable of discriminating between a plethora of competing specifications, most notably models that feature many, moderately many, and few structural breaks in the regression parameters We also propose a generalization of our model to the VAR framework with stochastic volatility. In an application to the US macroeconomy, we examine the usefulness of the TTVP-VAR in terms of forecasting, turning point prediction, and structural impulse response analysis. Our results show that the model yields precise forecasts, especially so during more volatile times such as witnessed in 2008. For that period, the forecast gain over simpler constant parameter models is particularly high. We then proceed by investigating turning point predictions, and observe excellent performance of the TTVP model, in particular for upturn predictions. Finally, we examine impulse responses to a $+100$ basis points contractionary monetary policy shock focusing on two sub-periods of our sample span, the pre-Volcker period and the rest of the sample. Our results reveal significant evidence for a severe price puzzle during episodes of the pre-Volcker period. The positive effect of the rate increase in inflation disappears in the second half of our sample. Modeling changes in responses over the two sub-periods only, such as in a regime switching model, however, would be too simplistic, as we do find also a lot of time variation within each sub-period. For example, we find increasing effectiveness of monetary policy in terms of output, investment growth, and hours worked in the more recent sub-period. This is especially true for the period after the global financial crisis in which the Federal Funds Rate has been tied to zero. For that period, a hypothetical deviation from the zero lower bound would create pronounced effects on the wider macroeconomy \section{Acknowledgments} We sincerely thank the participants of the WU Brown Bag Seminar of the Institute of Statistics and Mathematics, the 3rd Vienna Workshop on High-Dimensional Time Series in Macroeconomics and Finance 2017, the NBP Workshop on Forecasting 2017, and in particular Sylvia Fr\"uhwirth-Schnatter, for many helpful comments and suggestions that improved the paper significantly. \singlespacing \bibliographystyle{./bibtex/econometrica}
ff93b19cf23f2c010c37ca986e573725d1d60754
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{sec:intro} A useful representation of the occupied states in a periodic insulator is the Wannier function. Wannier functions (WFs) provide a localized real-space description of the extended Bloch states.~\cite{MLWFs-Review} In particular, WFs give a chemical picture of the bonding nature of a material, an alternative real-space formalism for many quantities, and can also be used for interpolating various physical properties on a fine mesh in the Brillouin zone~\cite{PhysRevB.74.195118,PhysRevB.76.165108}. For examples, WFs can be used to compute electronic polarization, orbital magnetization, the component of isotropic magnetoelectric coupling, and various transport properties. However, an exponentially localized Wannier function representation does not exist for insulators with a non-zero Chern number $C$.\cite{PhysRevB.74.235111,PhysRevLett.98.046402} Insulators with a non-zero Chern number are called integer quantum Hall insulators (or Chern insulators) and are characterized with a non-zero Hall conductance $\sigma = C e^2/h$. (Three-dimensional insulators are characterized by a triplet of Chern numbers.) In the past several years there has been significant interest in a group of materials related to the Chern insulator. These are called $\mathbb{Z}_2$ topological insulators (TIs). In two dimensions these $\mathbb{Z}_2$ topological insulators can be seen as topologically equivalent to two copies of a Chern insulator, one with $C=1$ and another with $C=-1$, \begin{equation} H_{\rm TI} = H_{+1} \oplus H_{-1}. \label{eq:pmC} \end{equation} Therefore the Chern number for a $\mathbb{Z}_2$ topological insulator is zero\cite{PhysRevLett.98.046402} which guarantees that it allows exponentially localized WFs ($C$ is additive over bands so in this case we have $C=1-1=0$). However, WFs of $\mathbb{Z}_2$ topological insulators do not preserve time reversal (TR) symmetry~\cite{PhysRevB.83.035108} even though the underlying Hamiltonian $H_{\rm TI}$ itself is time-reversal symmetric. This can be seen by realizing that constructing TR-preserving WFs would be equivalent to constructing WFs individually for the bands given by $H_{+1}$ and $H_{-1}$ separately, which is not possible as bands with non-zero $C$ don't have exponentially localized WFs. Therefore, the only possibility for constructing a smooth gauge of a compound system ($H_{\rm TI}$) is to break TR symmetry in the gauge by mixing eigenstates of $H_{+1}$ with those of $H_{-1}$. \subsection{Constructing Wannier functions from a guess}\label{sec:wannier90} Generalized WFs~\cite{MLWFs-MV} are obtained as the Fourier transform of the Bloch states $\psi_{m\mathbf{k}}$ (here we consider the case of three dimensions) \begin{equation}\label{eq:wf} \Ket{\mathbf{R} n}=\frac{V}{\left(2\pi\right)^3}\int\text{d}\mathbf{k} e^{-i\mathbf{k}\cdot\mathbf{R}}\sum_m u^{(\mathbf{k})}_{mn}\Ket{\psi_{m\mathbf{k}}}, \end{equation} where $u^{(\mathbf{k})}$ is an arbitrary unitary matrix that mixes different bands for a given $\mathbf{k}$-point, $\mathbf{R}$ is a translation vector, and $n$ is an integer running over the number of bands considered. This gauge freedom can be used to construct WFs with minimal possible spatial extent. These so called maximally localized Wannier functions (MLWFs) minimize the spread functional $\Omega$, \begin{equation}\label{eq:spread_r} \Omega = \sum_n \left(\Braket{r^2}_n-\Braket{\mathbf{r}}^2_n\right), \end{equation} with \begin{align} \Braket{r^2}_n &= \Braket{\mathbf{0}n|r^2|\mathbf{0}n}, \\ \Braket{\mathbf{r}}_n &= \Braket{\mathbf{0}n|\mathbf{r}|\mathbf{0}n}. \end{align} Due to the properties of the Fourier transform, localization of Wannier function $\ket{\mathbf{0} n}$ in real space is equivalent to the smoothness of Bloch states $\sum_m u^{(\mathbf{k})}_{mn}\Ket{\psi_{m\mathbf{k}}}$ in $\mathbf{k}$-space. Within the standard approach~\cite{MLWFs-MV} MLWFs are constructed for a set of $N$ composite bands by first guessing a set of $N$ localized orbitals $g_n(\mathbf{r})$ that are close to the $N$ target Wannier functions, \begin{equation}\label{eq:good-guess} \Ket{g_n}\approx\Ket{\bm{0}n}. \end{equation} Now given a set of Bloch states $\Ket{\psi_{m\mathbf{k}}}$ that are potentially not smooth in $\mathbf{k}$-space (or equivalently its WFs are not localized) one can try smoothening them by first projecting them into these guess orbitals $g_n$ \begin{equation} a^{(\mathbf{k})}_{mn} = \Braket{\psi_{m\mathbf{k}} | g_n} \end{equation} and then constructing the unitary matrices $u^{(\mathbf{k})}$ via the L{\"o}wdin orthonormalization procedure~\cite{Lowdin}, \begin{equation}\label{eq:gauge0} u^{(\mathbf{k})} = a^{(\mathbf{k})}\left[a^{(\mathbf{k})\dagger}a^{(\mathbf{k})}\right]^{-1/2}. \end{equation} If the overlap matrix appearing above under the inverse square root \begin{equation}\label{eq:def-sk} s^{(\mathbf{k})} \equiv a^{(\mathbf{k})\dagger}a^{(\mathbf{k})} \end{equation} has large singular values then this procedure is well-defined and matrices $u^{(\mathbf{k})}$ constructed in this manner can be used to rotate Bloch states into a smooth gauge $u^{(\mathbf{k})} \Ket{\psi}$. It is trivial to show that if $\Ket{g_n}$ are MLWFs, that singular values of $s^{(\mathbf{k})}$ all equal $1$ and this procedure gives back rotated Bloch states $u^{(\mathbf{k})} \Ket{\psi}$ that correspond to the original MLWFs. Therefore, one can hope that if the $\Ket{g_n}$ are at least close to MLWFs the resulting rotated Bloch states again correspond to localized---but not necessarily maximally localized---WFs. Given this starting point one can use procedure from Ref.~\onlinecite{MLWFs-MV} to arrive at MLWFs if needed. On the other hand if the $\Ket{g_n}$ are not close to MLWFs then the resulting rotated Bloch states need not be smooth. This is easily seen in the case of a single band. In this case if the orbital $\Ket{g}$ fails to capture the character of the Bloch state $\Ket{\psi_{\mathbf{k}}}$ for some $\mathbf{k}$-point then the complex number $a^{(\mathbf{k})}$ will have a small norm at that $\mathbf{k}$. If the norm of $a^{(\mathbf{k})}$ is exactly zero for some $\mathbf{k}$ then the procedure involving a negative square root in Eq.~\eqref{eq:gauge0} is ill-defined as it involves division by zero. However, if the norm of $a^{(\mathbf{k})}$ is small but non-zero then the procedure is potentially numerically unstable as small noise in $a^{(\mathbf{k})}$ might get amplified when taking the negative square root of $s=a^{\dagger} a$. This analysis generalizes to the case of multiple occupied bands ($N > 1$) in the following way. If one of the rotated $N$ Bloch states are not captured well by guess $\Ket{g_n}$ then the overlap matrix (which is now a full $N\!\times\!N$ matrix) will have one small singular value and again the process of taking the negative square root of $s^{(\mathbf{k})}$ in Eq.~\eqref{eq:gauge0} will be ill-defined or unstable. In fact this is precisely the way in which the L{\"o}wdin procedure fails if one tries to apply it to the case of a Chern insulator.~\cite{PhysRevB.74.235111} For any localized trial orbital $\Ket{g_n}$ the overlap matrix $s^{(\mathbf{k})}$ for a Chern insulator will have at least one zero singular value at least at one point in the Brillouin zone. This will also happen in the case of a $\mathbb{Z}_2$ topological insulators if one chooses trial orbitals that form a time-reversal symmetric pair. In Ref.~\citenum{PhysRevB.83.035108} it was recognized that projecting trial orbitals that break TR symmetry is necessary to ensure that all singular values of $s^{(\mathbf{k})}$ are nonzero everywhere in the Brillouin zone. In practice this approach still requires an initial guess of orbitals that approximate the target WFs. This guessing is somewhat harder than in the case of a non-topological insulator since it must break TR and potentially some other crystalline symmetries. In Ref.~\citenum{PhysRevB.83.035108} this was achieved for a tight-binding model by an educated guess of trial orbitals based on the orbital character of the bands at the band inversion points and symmetries present in the model. Another approach for constructing WFs for $\mathbb{Z}_2$ topological insulators was introduced in Ref.~\citenum{PhysRevB.93.035453}. This approach relies on constructing a smooth gauge in a closely related non-topological insulator phase and then transporting that gauge to the TI of interest by following a path in the parameter space that explicitly breaks time-reversal symmetry (and potentially other symmetries such as inversion). This parameter space has to be chosen for each system at hand by adding terms to the Hamiltonian that break TR (and potentially crystalline) symmetry and keep the electron band gap open. In this manuscript we will present a method that can automatically construct WFs for topologically nontrivial insulators. \section{Our approach, the OPFM} \label{sec:opfm} In a recent manuscript we introduced the optimized projection functions method (OPFM)\cite{OPFM} that allows automatic construction of MLWFs. We will now give a brief review of the OPFM and then discuss why this approach is suitable for constructing WFs for $\mathbb{Z}_2$ topological insulators. As opposed to the standard approach, which requires $N$ trial orbitals for $N$ composite electron bands, in the OPFM one selects a larger set of $M\!>\!N$ orbitals $h_i(\mathbf{r})$ that approximately span the space of $N$ Wannier functions in a home cell, \begin{equation}\label{eq:span} \text{Span}(\Ket{h_i})\supseteq\text{Span}(\Ket{\bm{0}n}). \end{equation} This can easily be achieved by including in $\{h\}$ valence atomic orbitals. Given a set of projection orbitals $\{h\}$, we use the OPFM to find a semiunitary $M\!\times\!N$ matrix W such that the $N$ orbitals \begin{equation} \Ket{\widetilde{g}_j}=\sum_{i=1}^M W_{ij}\Ket{h_i} \end{equation} are as close as possible to localized WFs. Given the functions $\widetilde{g}_j$ one can construct the smooth gauge by first expanding the original functions into Bloch states, \begin{equation} A^{(\mathbf{k})}_{mn}=\Braket{\psi_{m\mathbf{k}} | h_n} \end{equation} and then rotating them into the optimal subset, \begin{equation} a^{(\mathbf{k})}_{mn} = A^{(\mathbf{k})}W \end{equation} which can then be used in the L{\"o}wdin procedure. Now we discuss why the OPFM is suitable for constructing WFs in topological insulators. In $\mathbb{Z}_2$ topological insulators, the spin-orbit interaction induces a band-inversion between states of different orbital character. For example, in the case of Bi$_2$Se$_3$ the topologically nontrivial state is induced by a band inversion at the $\Gamma$ point between Se and Bi states. Therefore one can expect that the MLWFs for the occupied bands in this system will contain a mixture of both Se and Bi states. Guessing such a mixture is nontrivial for several reasons. First, as we will show later, this mixture includes complex imaginary components. Second, the mixture typically contains a contribution from more than two atoms. Third, the mixture must break all relevant symmetries which enforce the topologically nontrivial state. However, our OPFM can find this mixture since $\{h\}$ in the case of Bi$_2$Se$_3$ can include both Se and Bi atomic orbitals as $M$, the number of orbitals in set $\{h\}$, can be larger than the number of electron bands $N$. In this manuscript we follow the notation we introduced in Ref.~\onlinecite{OPFM} where square $N\!\times\!N$ matrices are represented by lowercase letters (e.g.\ $a^{(\mathbf{k})}$ and $u^{(\mathbf{k})}$), and larger rectangular $N\!\times\!M$, $M\!\times\!N$, or square $M\!\times\!M$ matrices are represented by uppercase letters (e.g.\ $A^{(\mathbf{k})}$ and $W$). \subsection{Selecting the set \texorpdfstring{$\{h\}$}{h}}\label{sec:basis-h} Now we will discuss a choice of orbitals $h_i$ that satisfy the condition given in Eq.~\eqref{eq:span}. Mathematically speaking, without knowing anything about chemical bonding in the insulator of interest, one would have to include in set $\{h\}$ all atomic orbitals on all atoms in the crystal to guarantee a complete basis for expansion of the WFs. Luckily, in an ionic or a covalent insulator it is enough to choose set $\{h\}$ to include only valence atomic orbitals, as they are typically the ones forming atomic bonds. In addition, since WFs are typically exponentially localized it is enough to choose orbitals in the home cell, and possibly few atoms in the neighboring unit cells. (For example, as discussed in Ref.~\onlinecite{OPFM} in the case of cubic silicon one has to include in set $\{h\}$ atomic orbitals centered on two atoms in the basis as well as three neighboring atoms, so that, for each of the four Si-Si bonds, both Si atoms forming a particular bond are included in the set $\{h\}$.) However, bonding in the case of $\mathbb{Z}_2$ TIs is more involved than in a typical insulator. As we will show in this manuscript, presence of spin-orbit induced band inversion induces an intricate bonding network so that some WFs extends over more than two atoms (as in the case of the bonds in silicon) and thus one needs to use a somewhat larger set $\{h\}$ than in a conventional covalent material. However, in all cases we tested, it was enough to include in $\{h\}$ orbitals in the home-cell along with the orbitals in a single neighboring cell. \subsection{Finding matrix \texorpdfstring{$W$}{W}}\label{sec:offdiag} In Ref.~\citenum{OPFM} the problem of finding $W$ that minimizes the WF spread $\Omega$ was reduced to minimizing the Lagrangian \begin{align}\label{eq:lagrangian_old} \mathcal{L} \left(W,\lambda\right) = \ &\widetilde{\Omega}_\text{I,OD}(W) \\ &+ \lambda w \sum_{\mathbf{k}}\sum_{i=1}^N \abs*{\left[W^{\dagger}\left( S^{(\mathbf{k})} -I_M \right) W\right]_{ii}}^2, \notag \end{align} where we define the large overlap matrix \begin{equation} S^{(\mathbf{k})} \equiv A^{(\mathbf{k})\dagger}A^{(\mathbf{k})}. \end{equation} The first term in Eq.~\eqref{eq:lagrangian_old} approximates the sum of the invariant and offdiagonal parts of the spread $\Omega$. However, this approximation is valid only when rotated overlap matrix $W^\dagger S^{(\mathbf{k})} W$ is close to the identity matrix (see discussion in Ref.~\onlinecite{OPFM}). For simply bonded insulators this condition is enforced by the second term in Eq.~\eqref{eq:lagrangian_old}. While strictly speaking the entire matrix $W^\dagger S^{(\mathbf{k})} W$ should be close to the identity matrix, the second term in Eq.~\eqref{eq:lagrangian_old} only penalizes the deviation of diagonal elements of $W^\dagger S^{(\mathbf{k})} W$ away from $1$. This simplification is adequate for the case of simply bonded insulators where only a small number of atoms are needed to span the space of WFs centered in the home cell. However, for the case of $\mathbb{Z}_2$ TIs, one needs to use a somewhat larger set $\{h\}$ and this simplification is insufficient since now some contributions to a WF could potentially be duplicated by more than one element in the set $\{h\}$. Therefore, in this manuscript we will construct $W$ by minimizing the following Lagrangian that penalizes the offdiagonal elements of $W^\dagger S^{(\mathbf{k})} W$ as well, \begin{align}\label{eq:lagrangian_new} \begin{split} \mathcal{L} \left(W,\lambda\right) = \ &\widetilde{\Omega}_\text{I,OD}(W) \\ &+ \lambda w \sum_{\mathbf{k}}\sum_{i=1}^N\sum_{j=1}^N \abs*{\left[W^{\dagger} \left( S^{(\mathbf{k})} - I_M \right) W\right]_{ij}}^2 . \end{split} \end{align} We describe the algorithm to minimize such Lagrangian in App.~\ref{sec:appndx-offdiag}. We confirmed that with this extended Lagrangian WF spreads for conventional insulators investigated in Ref.~\citenum{OPFM} are unaffected. For example, the initial spread in the case of cubic silicon changes by less than 0.2\% when offdiagonal elements are included in $\mathcal{L}$. The only difference with respect to Ref.~\citenum{OPFM} is that now---with the Lagrangian from Eq.~\eqref{eq:lagrangian_new}---it doesn't matter whether some WFs can be represented by orbitals from $\{h\}$ in more than one way, at a small additional cost in the computational time. For example, in the case of the cubic silicon if we include 6 (instead of 3) neighboring atoms in the set $\{h\}$ so that three out of four Si-Si bonds can be represented in duplicated ways, the total initial spread is changed only by 0.6\%. We note here that it is numerically straightforward to construct an arbitrary set $\{h\}$ given a set of orbitals $h_i$ in the home cell. For a particular orbital $h_i(\mathbf{r})$ on the basis atom, a projection onto another orbital given by the same orbital but translated by lattice vector $\mathbf{R}$ is simply, \begin{equation}\label{eq:expand-A} \Braket{\psi_{{n\mathbf{k}}} | h_i(\mathbf{r}-\mathbf{R})} = e^{-i\mathbf{k}\cdot\mathbf{R}} \Braket{\psi_{{n\mathbf{k}}} | h_i(\mathbf{r})} \end{equation} by the virtue of Bloch's theorem. \section{Examples}\label{sec:examples} In the following subsections, we apply the optimized projection functions method to three examples of $\mathbb{Z}_2$ topological insulators. The first is the Kane-Mele model~\cite{PhysRevLett.95.146802}, which is a two-dimensional tight-binding model on the honeycomb structure. The second is a three-dimensional tight-binding model of a strong topological insulator that was introduced in Ref.~\onlinecite{Qi2008}. The third example is a realistic case of a three-dimensional strong topological insulator (Bi$_2$Se$_3$) as calculated within the density functional theory approach.~\cite{Xia2009,Zhang2009} \subsection{Two-dimensional model} The Kane-Mele model is a two-dimensional model of a $\mathbb{Z}_2$ topological insulator. It contains four electron bands, two of which are considered to be occupied. It is defined on a honeycomb structure that can be described in terms of the hexagonal lattice with primitive lattice vectors $\mathbf{a}_{1,2} = \frac{a}{2}(\sqrt{3}\hat{\mathbf{y}}\pm\hat{\mathbf{x}})$, and a basis of two sites, $A$ and $B$, located at $\bm{\tau}_A=a\hat{\mathbf{y}}/\sqrt{3}$ and $\bm{\tau}_B=2a\hat{\mathbf{y}}/\sqrt{3}$, respectively. In what follows, we choose $a=1$~\AA\ for convenience. The Kane-Mele Hamiltonian is \begin{align} \begin{split} H = t&\sum_{\langle ij \rangle}c_i^\dagger c_j + i\lambda_{\text{SO}}\sum_{\langle\langle ij \rangle\rangle}\nu_{ij}c_i^\dagger s^z c_j \\ +& i\lambda_{\text{R}}\sum_{\langle ij \rangle}c_i^\dagger(\mathbf{s}\times\hat{\mathbf{d}}_{ij})_z c_j + \lambda_v\sum_i \xi_i c_i^\dagger c_i. \end{split} \label{eq:KM} \end{align} We suppressed spin indices on the electron creation and annihilation operators. Symbol $\langle ij \rangle$ indicates a sum over nearest neighbors and $\langle\langle ij \rangle\rangle$ indicates a sum over next-nearest neighbors. The first term in the Hamiltonian is the nearest neighbor hopping, with hopping strength $t$ that we set equal to $1$ for convenience (i.e., all energies are in units of $t$). The second term describes spin-dependent second nearest neighbor hopping, which emulates a spin-orbit interaction. Here, $\nu_{ij}$ takes on the value $\pm1$ depending on the sign of $(\hat{\mathbf{d}}_1\times\hat{\mathbf{d}}_2)_z$, where $\hat{\mathbf{d}}_1$ and $\hat{\mathbf{d}}_2$ are the unit vectors along the bonds traversed as the electron hops from site $j$ to $i$, and $s^z$ is the Pauli spin matrix. The third term describes nearest neighbor Rashba coupling, where $\hat{\mathbf{d}}_{ij}$ is the unit vector along the bond from $j$ to $i$. Lastly, the fourth term introduces a staggered on-site potential ($\xi_i=\pm1$) between the $A$ and $B$ sublattices; we choose $\xi_i$ so that the on-site potential is negative on the $B$ sublattice and the occupied bands in the normal phase have dominant $B$ character. In the following we set the staggered on-site term $\lambda_v=1$ and the Rashba term $\lambda_\text{R}=0.5$. Increasing the strength of the spin-orbit term $\lambda_\text{SO}$ tunes the model from describing the normal to the topological insulator phase, with the transition at $\lambda_\text{SO}\approx0.27$. For calculations in the topological phase we use $\lambda_\text{SO}=0.6$. The Kane-Mele model is solved using the PythTB~\footnote{\url{http://physics.rutgers.edu/pythtb/}} package with a basis of two orbitals per site, one each for spin-up and spin-down, \begin{equation}\label{eq:km-orbitals} \Ket{A;\uparrow_z}, \quad \Ket{A;\downarrow_z}, \quad \Ket{B;\uparrow_z}, \quad \Ket{B;\downarrow_z}. \end{equation} In Figure~\ref{fig:km-bands} we plot the band structures in both the normal phase ($\mathbb{Z}_2$ even) and the topological phase ($\mathbb{Z}_2$ odd). The bands are colored according to the character of the state, with red corresponding to a state of $B$-orbital character and blue corresponding to $A$-orbital character, and gray indicating a mixture. \begin{figure} \includegraphics{fig1a} \par\vspace{0.25cm} \includegraphics{fig1b} \caption{\label{fig:km-bands}% Band structures of the Kane-Mele model in the normal phase (top) and topological phase (bottom). The part of the bands colored red correspond to $B$-orbital character, while blue corresponds to $A$-orbital character. The thickness of the line corresponds to the spin component, with thicker indicating mostly spin-up, and thinner indicating spin-down. The zero of energy is set to the middle of the gap.} \end{figure} The thickness of the line indicate the spin of the state along the $z$-axis, with thicker corresponding to mostly spin-up and thinner line corresponding to mostly spin-down; an intermediate thickness indicates mixed spin state due to the Rashba-like term. In the topological phase, there is a clear inversion of the character of the states near the $K$ and $K^\prime$ points. \subsubsection{Selecting the set \texorpdfstring{$\{h\}$}{h}}\label{sec:cluster} Our results of the OPFM applied to the case of a Kane-Mele model are shown at the top of Table~\ref{tab:km} along with the result form previous work.\cite{PhysRevB.83.035108} In the previous work the following guess orbitals were used to construct WFs for the Kane-Mele model in the topological phase: $\Ket{A;\uparrow_x}$~and~$\Ket{B;\downarrow_x}$. Note that spins here point in plane ($x$) while the basis functions have spins pointing perpendicular to the plane ($z$). To construct localized WFs for the Kane-Mele model using the OPFM we consider several different sets $\{h\}$ of basis functions. The smallest set consisted of the four orbitals on the two atoms in the home cell defined in Eq.~\eqref{eq:km-orbitals}. As expected, this small set is unable to capture the extended nature of the WF in the $\mathbb{Z}_2$ topological insulator. The fact that this set is too small is numerically indicated with a small minimal singular value of $s^{(\mathbf{k})}=W^{\dagger}S^{(\mathbf{k})}W$ (its value is only 0.03, not given in Table~\ref{tab:km}) which then results in an ill-defined L{\"o}wdin orthonormalization procedure. This procedure produces a gauge with a WF spread $\Omega^0=0.333$, significantly higher than that of a MLWF ($\Omega^{\rm GM}=0.189$). If we try smoothening this gauge further with the Marzari-Vanderbilt procedure\cite{MLWFs-MV} it remains stuck in a local minimum as the spread is only slightly reduced to $\Omega^{\rm GM}=0.319$. Since the figure of merit (i.e. minimal singular value of $s^{(\mathbf{k})}$) was small for this set $\{h\}$ we decided to use the OPFM with larger sets $\{h\}$. The next set we considered, labeled $\{0,1^{\text{st}}\}$, includes---in addition to orbitals in the home cell---orbitals on their four first-neighboring atoms. An even larger set we tried $\{0,1^{\text{st}},2^{\text{nd}}\}$ includes both first and second nearest neighboring atoms. \begin{table*} \caption{\label{tab:km}% Results of the OPFM applied to the examples in Section~\ref{sec:examples}. The first column gives the square moduli of $s^{(\mathbf{k})}-I_N$ averaged over $\mathbf{k}$-points and matrix elements. The second column lists the minimal singular value of $s^{(\mathbf{k})}$ over all $\mathbf{k}$-points. The third column show the spread $\Omega^0$ after L{\"o}wdin procedure while the fourth column shows the spread $\Omega^{\rm GM}$ when L{\"o}wdin procedure is followed up by the Marzari-Vanderbilt\cite{MLWFs-MV} procedure. For three-dimensional cases, the fifth column gives the value for the Chern-Simons $\theta$ term. See text for a description of the sets $\{h\}$ used in the OPFM. } \begin{ruledtabular} \begin{tabular}{lccccc} & \multirow{2}{*}{Average $\abs*{\left(s^{(\mathbf{k})} - I_N \right)_{ij} }^2$} & \multirow{2}{*}{Min. $\text{sing}\left( s^{(\mathbf{k})} \right) $} & \multicolumn{2}{c}{Spread (\AA$^2$)} & \multirow{2}{*}{Chern-Simons $\theta$} \\ \cline{4-5} & & & \multicolumn{1}{c}{$\Omega^0$} & \multicolumn{1}{c}{$\Omega^\text{GM}$} & \\ \hline \vspace{-6pt}\\ \multicolumn{6}{l}{Two-dimensional model}\\ \quad\quad Previous work, Ref.~\onlinecite{PhysRevB.83.035108} & 0.148 & \multicolumn{1}{d{1.5}}{0.11} & 0.212 & 0.189 & \\ \quad\quad OPFM using set $\{ 0, 1^{\text{st}} \}$ & 0.017 & \multicolumn{1}{d{1.5}}{0.40} & 0.244 & 0.189 & \\ \quad\quad OPFM using set $\{ 0, 1^{\text{st}}, 2^{\text{nd}} \}$ & 0.006 & \multicolumn{1}{d{1.5}}{0.71} & 0.207 & 0.189 & \\ \vspace{-6pt}\\ \multicolumn{6}{l}{Three-dimensional model}\\ \quad\quad OPFM using set $\{ 0, 1 \}$ & 0.0184 & \multicolumn{1}{d{1.5}}{0.0472} & 0.142 & 0.135 & 0.96$\pi$ \\ \quad\quad OPFM using set $\{ 0, 1, 2, 3 \}$ & 0.0133 & \multicolumn{1}{d{1.5}}{0.1022} & 0.142 & 0.135 & 0.96$\pi$ \\ \vspace{-6pt}\\ \multicolumn{6}{l}{Density functional theory, Bi$_2$Se$_3$}\\ \quad\quad Previous work, Ref.~\onlinecite{Coh-CSOMP} & 0.0068 & \multicolumn{1}{d{1.5}}{0.0003} & 109.80 & 95.84 & 0.32$\pi$ \\ \quad\quad Previous work, Ref.~\onlinecite{PhysRevB.93.035453} & 0.0057 & \multicolumn{1}{d{1.5}}{0.0002} & 126.12 & 95.85 & 0.35$\pi$ \\ \quad\quad OPFM using set $\{ 0,1 \}$ & 0.0026 & \multicolumn{1}{d{1.5}}{0.0001} & 133.43 & 95.84 & 0.34$\pi$ \\ \quad\quad OPFM using set $\{ 0,1,2,3 \}$ & 0.0017 & \multicolumn{1}{d{1.5}}{0.0240} & 310.29 & 95.83 & 0.34$\pi$ \\ \end{tabular} \end{ruledtabular} \end{table*} As soon as we include first or second neighbor atoms into the set $\{h\}$ the resulting minimal singular value of $s^{(\mathbf{k})}$ increases from 0.03 to 0.40 and 0.71, respectively for the two sets, and the resulting spread $\Omega^0$ decreases. Final spread $\Omega^{\rm GM}$ agrees with previous result\cite{PhysRevB.83.035108} up to numerical precision. Table~\ref{tab:km} contains also the average distance between the overlap matrix $s^{(\mathbf{k})}$ and the identity matrix. However, that averaged quantity masks the fact that the overlap matrix is typically different from an identity matrix only in a small part of the Brillouin zone where inversion occurs (see band characters near $K$ and $K'$ points in Fig.~\ref{fig:km-bands}). Therefore for the purpose of presentation we give in Fig.~\ref{fig:km-overlap-cluster} the distribution of $\abs*{\left(s^{(\mathbf{k})}-I\right)_{ij}}^2$ over all $\mathbf{k}$-points and all its matrix elements $ij$. Note that this quantity is the same as the second term in Eq.~\eqref{eq:lagrangian_new}. As can be seen from Fig.~\ref{fig:km-overlap-cluster} in all cases distance of $s^{(\mathbf{k})}$ from identity is small for nearly all $\mathbf{k}$-points (note that the vertical scale is logarithmic). However, in the case of the guess orbitals from Ref.~\onlinecite{PhysRevB.83.035108} there is a fraction of $\mathbf{k}$ points for which matrix elements of $s^{(\mathbf{k})}$ are quite far away from identity matrix (up to 0.7). These $\mathbf{k}$-points correspond to the small part of the Brillouin zone with inverted bands. However, singular values of $s^{(\mathbf{k})}$ are large enough (smallest one is 0.11) so that the L{\"o}wdin procedure is well behaved even for the guess orbital from Ref.~\onlinecite{PhysRevB.83.035108}. In the case of the OPFM the deviation of $s^{(\mathbf{k})}$ from identity is significantly smaller (the maximum value is only 0.1 for the cluster $\{ 0, 1^{\text{st}}, 2^{\text{nd}} \}$). \begin{figure} \includegraphics[width=\columnwidth]{fig2} \caption{\label{fig:km-overlap-cluster}% Histograms of the square moduli of the elements of $s^{(\mathbf{k})}-I$ for guess orbitals from Ref.~\citenum{PhysRevB.83.035108} (top panel) and two sets $\{h\}$ within the OPFM (middle and bottom panel). } \end{figure} We also confirmed that OPFM gives automatically good spread as one varies $\lambda_\text{SO}$ through the transition from the topological all the way to the normal phase. For this two-dimensional model we constructed the projection matrices on a $15^2$ $\mathbf{k}$-point grid. We find that the optimal value for the Lagrange multiplier ($\lambda$) is 0.03. We also note that while in the case of normal insulators studied previously one can often initialize $W$ with the identity matrix, in the case of $\mathbb{Z}_2$ TIs we sometimes need to start off the minimization from a random matrix so that the initial $W$ breaks all symmetries. (We confirmed that, in the case of normal insulators studied in Ref.~\citenum{OPFM}, starting minimization procedure from a random matrix does not affect the final spread. For example, in the case of cubic silicon the total spread is unaffected within numerical precision.) \subsubsection{Analysis of the WF} As expected, we find that the WF in the $\mathbb{Z}_2$ topological insulator case extends well beyond the home cell. This finding is expected since band inversion usually occurs over a small region in $\mathbf{k}$-space. To analyze the extent of the Wannier function in more detail we show in Fig.~\ref{fig:km-wfs} the MLWFs for the Kane-Mele model in the real space. We write the WF amplitude on a particular sublattice $j=\{A,B;\mathbf{R}\}$ in the crystal as, \begin{equation} \left(\alpha_1 + i\alpha_2\right)\Ket{j\uparrow} + \left(\beta_1 + i\beta_2\right)\Ket{j\downarrow}. \end{equation} These amplitudes can be computed from the projections of Bloch states into basis functions ($A^{(\mathbf{k})}$) and the smooth gauge for the WFs $u^{(\mathbf{k})}$ as, \begin{align*} \Braket{h_j | \mathbf{R} n} &= \bra{h_j}\sum_{m\mathbf{k}}u^{(\mathbf{k})}_{mn}\Ket{\psi_{m\mathbf{k}}} = \sum_{\mathbf{k}}A^{(\mathbf{k})\dagger}u^{(\mathbf{k})}, \end{align*} Figure~\ref{fig:km-wfs} shows the amplitudes $\alpha_1$, $\alpha_2$, $\beta_1$, and $\beta_2$ on each site $j$ in the crystal for both occupied bands (labelled \#1 and \#2). The size of the circles are proportional to the absolute value of the magnitude of $\alpha_1,\alpha_2,\beta_1,\beta_2$ while color denotes their sign (red for positive and blue for negative). The cross symbols ($\bm\times$) denote $A$ sites while plus symbols ($\bm+$) denote $B$ sites. In the normal phase, both WFs are centered near the $B$ site in the home cell, with components of opposite sign on the first nearest neighbors, and small components on second nearest neighbors. The topological phase has WFs that are centered near different sites ($A$ and $B$ in the home cell), with both being a mixture of spin-up and spin-down. Most importantly, in the topological phase the WF amplitude extends well beyond the home cell into the first and second nearest neighbors. The beyond-home-cell component of the WF in addition has a significant imaginary part. Therefore, from here we confirm once again that the set $\{h\}$ in the case of a $\mathbb{Z}_2$ topological insulator must include orbitals beyond those in the home cell. \begin{figure*} \includegraphics{fig3a} \par\vspace{0.75cm} \includegraphics{fig3b} \caption{\label{fig:km-wfs}% Plot of the Wannier functions obtained via the OPFM on set $\{ 0, 1^{\text{st}}, 2^{\text{nd}} \}$, for the two occupied bands of the Kane-Mele model in the normal insulator phase (top) and the topological insulator phase (bottom). The crosses ($\bm \times$) indicate the $A$ sites, while the plus signs ($\bm +$) indicate the $B$ sites. Red circles correspond to a component that is positive and blue circles correspond to negative. The size (area) of the circle is proportional to the magnitude of the component $\alpha_1,\alpha_2,\beta_1,\beta_2$. Here, the WFs plotted are for a Kane-Mele model with $\lambda_v=1$ and $\lambda_\text{R}=0$. } \end{figure*} \subsection{Three-dimensional model}\label{sec:3d-ti} We now turn to the model of a three-dimensional strong $\mathbb{Z}_2$ topological insulator. A simple model of such an insulator is given in Ref.~\citenum{Qi2008} by constructing a higher dimensional insulator with a non-zero second Chern number and then restricting the model to three dimensions. Similarly as in the case of a Kane-Mele model this model consists of four orbitals in the basis and two occupied electronic bands. One difference with respect to the Kane-Mele model is that the only hopping terms in the model are either between orbitals in the same unit cell, or between the first neighboring cells. The Kane-Mele model includes hopping to second nearest neighboring cells as well. As is done for the Kane-Mele model, here we performed OPFM with sets $\{h\}$ of increasing size. Once again we find that as soon as a neighboring cell is included in the set $\{h\}$, the OPFM procedure produces a smooth gauge. Similarly, as in the case of Kane-Mele model, we find that with larger sets $\{h\}$ the minimal singular value of the overlap matrix is increased. Here we adopt a notation by which the set $\{0, 1\}$ corresponds to orbitals in the home cell as well as neighboring cells translated along the first lattice vector. Similarly, the set $\{0, 1, 2, 3\}$ corresponds to orbitals in the home cell as well as those translated along all three lattice vectors. In addition to the quantities reported in Table~\ref{tab:km} for the Kane-Mele model, here we also report the value of the Chern-Simons orbital magnetoelectric coupling $\theta$.~\cite{Qi2008,PhysRevLett.102.146805} The $\theta$ term takes on the value 0 or $\pi$ (modulo $2\pi$) in the normal and topological phase, respectively. However, these values would be obtained only in the limit of infinitely dense $\mathbf{k}$-meshes, as the discretized expression for $\theta$ we used is not gauge invariant (gauge invariant discrete form of $\theta$ is unknown, as far as we are aware). On a finite mesh the calculated value of $\theta$ in the topological phase is typically smaller than $\pi$ and it converges very slowly to $\pi$ as the $\mathbf{k}$-mesh gets denser. We used a $20^3$ mesh of $\mathbf{k}$ points for this calculation. Here the value of the Lagrange multiplier ($\lambda$) is $1$. The expressions for $\theta$ in terms of WFs given in Ref.~\onlinecite{Coh-CSOMP} clearly shows that $\theta$ must be $0$ if the WFs are purely real. Therefore, just as in the case of two-dimensional model, the WFs in the three-dimensional topological insulator must contain large imaginary components so that $\theta$ can be non-zero ($\theta=\pi$ modulo $2\pi$ to be precise). Table~\ref{tab:km} contains some of the results of the OPFM applied to the three-dimensional model. In the normal phase of that model (not shown in the Table~\ref{tab:km}) our OPFM finds projection functions that well approximate the WFs even when we use the set $\{0\}$ with orbitals only in the home cell. The minimal singular value of the $s^{(\mathbf{k})}$ matrix is close to identity (0.92) and the spread after the L{\"o}wdin procedure agrees with the spread at the global minimum within the first four non-zero significant digits. In the topological phase of the model, the set $\{0\}$ results in an overlap matrix with a very small singular value ($10^{-29}$) but with an inclusion of a larger set $\{h\}$ all figures of merit improve, as in the case of the Kane-Mele model. Therefore we conclude that even in the case of three-dimensional models the WF in a $\mathbb{Z}_2$ topological insulator extends well beyond the home cell. We note here that for relatively small sets such as $\{0,1\}$ the minimal singular value is rather small (0.0472); however, the resulting final spread is very close to the spread at the global minimum and the value of $\theta$ is close to $\pi$. The minimal singular value increases to 0.1022 in the set $\{0,1,2,3\}$. We also tried using an even larger set where orbitals are translated both in positive and negative direction of the lattice vector $\{0,1,\bar{1},2,\bar{2},3,\bar{3}\}$ and we find that the minimal singular value increases to 0.514. Despite having a larger minimal singular value, $\theta$ and $\Omega^{\rm GM}$ computed from this set are up to numerical precision equal to those obtained using a much smaller set, $\{0,1\}$. Therefore we conclude that the set $\{0,1\}$ is adequate for this system even though it yields a somewhat small minimal singular value of the overlap matrix (0.0472). To further test our method we generalized the tight-binding model from Ref.~\citenum{Qi2008} to higher number of bands. The model from Ref.~\citenum{Qi2008} was constructed from $2n$-dimensional Clifford algebra where $n=2$. If we use $n=3$ or $n=4$ algebras and again perform dimensional reduction to three dimensions, the resulting tight-binding model will have $2^{n-1}$ occupied bands out of $2^n$ bands. This means that in the $n=3$ case we have $8$-band model with $4$ occupied bands, while with $n=4$ we have $16$-band model with $8$ occupied bands. Applying the OPFM to these models using the set $\{0,1,\bar{1},2,\bar{2},3,\bar{3}\}$ again produces a smooth gauge. While the minimal singular value in the $4$-band model $(n=2$) discussed earlier is 0.514, with $8$-band model ($n=3)$ it is 0.356, and with $16$-band model ($n=4$) it is 0.360. \subsection{Density functional theory, \texorpdfstring{B\lowercase{i}$_{2}$}{Bi2}\texorpdfstring{S\lowercase{e}$_{3}$}{Se3}}\label{sec:bi2se3} We now turn from the tight-binding models to some realistic calculations based on density functional theory. As an example of a prototypical strong 3D TI we use Bi$_2$Se$_3$.~\cite{Xia2009,Zhang2009} Its crystal structure is described by a rhombohedral lattice within the $D^5_{3d}$ space group. The material is made up of units of quintuple layers of Bi and Se. Each of the five layers in the quintuple forms a hexagonal sheet in plane. The topological phase is realized due to the strong spin-orbit coupling causing a band inversion of Se $p$ and Bi $p$ character around the $\Gamma$ point.~\cite{Zhang2009} This inversion is evident in Fig.~\ref{fig:bi2se3-bands}. To construct localized WFs for Bi$_2$Se$_3$, we first perform fully relativistic density functional theory calculations with the {\sc Quantum ESPRESSO} package.~\cite{QE-2009} The ground state properties are obtained using a $6^3$ $\mathbf{k}$-point grid and a kinetic energy cutoff of 60~Ry. The projection matrices $A^{(\mathbf{k})}_{mn}$ are obtained on a $12^3$ $\mathbf{k}$-point grid by projecting the top 28 valence bands into atomic Bi and Se $s$ and $p$ orbitals. We use Eq.~\eqref{eq:expand-A} to construct projections into orbitals translated by a lattice vector. In all calculations for Bi$_2$Se$_3$ we used the value $\lambda=1$ for the Lagrange multiplier. Again we consider several sets $\{h\}$ generated by translating the basis atoms by different lattice vectors, as using only orbitals in the home cell once again gave very small ($10^{-6}$) minimal singular value of $s^{(\mathbf{k})}$. As soon as a neighboring cells are included in the set $\{h\}$ the minimal singular values increase as well as the spread $\Omega^{0}$. We used the same translation vectors as in the three-dimensional model case: $\{0,1\}$ and $\{0,1,2,3\}$. Here again $0$ represents orbitals in the home cell while non-zero integers $1,2,3$ represent translations along the three equivalent rhombohedral lattice vectors. For completeness, we note that we chose as basis atoms those for which the reduced coordinates in the rhombohedral frame are as small as possible (between $-1/2$ and $1/2$). The results of OPFM in the case of Bi$_2$Se$_3$ are shown in Table~\ref{tab:km} along with the results from previous work. One of the previous works\cite{Coh-CSOMP} guessed WFs by trying out various initial projections that break symmetries while the other\cite{PhysRevB.93.035453} found it by constructing a path in parameter space that breaks time-reversal and inversion symmetry. In both earlier works the Bloch states were projected into hydrogen-like orbitals. We find very good agreement in both $\theta$ and $\Omega^{\rm GM}$ between our approach and two earlier works. The computed value for $\theta$ in all three cases is close to $\theta\approx 0.3\pi$ since we used a relatively small $\mathbf{k}$-point grid (it was $12^3$). With larger $\mathbf{k}$-grids $\theta$ converges towards $\pi$. We note here that the minimal singular value of $s^{(\mathbf{k})}$ using relatively large set $\{ 0,1,2,3 \}$ in OPFM is still somewhat small (0.0240) even though it is two orders of magnitude larger than those in previous works.\cite{Coh-CSOMP,PhysRevB.93.035453} Nevertheless $\Omega^{\rm GM}$ agrees well with each other in all cases and the value of $\theta$ is what is expected for a three-dimensional strong $\mathbb{Z}_2$ topological insulator. (Some of the difference between the minimal singular values in these approaches might originate from use of hydrogen-like projection functions in Refs.~\onlinecite{Coh-CSOMP,PhysRevB.93.035453}) \begin{figure} \includegraphics{fig4} \caption{\label{fig:bi2se3-bands}% Band structure of Bi$_2$Se$_3$. The part of the bands colored red correspond to Se $p$ character, while blue corresponds to Bi $p$ character. The zero of energy is set to the middle of the gap. } \end{figure} \section{Outlook} In this paper we described a procedure for automated construction of maximally localized Wannier functions for topologically nontrivial set of bands. We expect that this method can be applied to any topological insulator, either protected by time-reversal symmetry, or by crystalline symmetry, as long as there exists a localized representation, i.e. as long as the first Chern numbers are all zero. Similarly, we expect that this method could be applied to topologically nontrivial bands not only of electrons, but also of phonons, photons, cold atoms, or other particles. \begin{acknowledgments} We thank Bradford A. Barker for providing the pseudopotentials for Bi and Se. We also thank Georg W. Winkler for clarifying the initial projections used in Ref.~\citenum{PhysRevB.93.035453}. This research was supported by the Theory Program at the Lawrence Berkeley National Lab through the Office of Basic Energy Sciences, U.S. Department of Energy under Contract No. DE-AC02-05CH11231 which provided the tight binding and DFT calculations; and by the National Science Foundation under grant DMR15-1508412 which provided for basic theory and method development. Computational resources have been provided by the DOE at Lawrence Berkeley National Laboratory's NERSC facility. \end{acknowledgments}
1d17908881d0423d55305dc62465e3e4f84bfac9
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:introduction} Inference of the underlying point sources in photon count maps is a recurring problem in astronomy. Potential challenges include poorly known backgrounds, detector noise, shot noise from faint or modeling degeneracies from overlapping point sources. The resulting symptom is that of flux incompleteness, where the faintest sources are not resolved, but instead absorbed into the diffuse background prediction. Hence the flux distribution of an incomplete catalog exhibits an unphysical roll off at the faint end, even when there are yet fainter true point sources in the image. A commonly applied approach to point source inference is the frequentist method of asking whether an additional point source increases the maximum likelihood compared to the null model, i.e., without the point source. Iteratively performing this exercise over the image while potentially floating other parameters such as the background normalization, one can produce a map of delta log-likelihood, which can be used to identify features in the count data that are unlikely to come from a diffuse emission component. A standard in gamma-ray astronomy is to reject model point sources that yield a test statistic below 25, which, for a $\chi^2$ distribution with four degrees of freedom (spatial coordinates, flux and spectral index), corresponds to just above 4$\sigma$. This approach is computationally cheap and has been the standard algorithm to reduce full-sky maps to catalogs such as the 3FGL catalog of point sources in the gamma-ray sky \citep{Acero2015a}. In the limit where an image is populated by well-separated point sources, this method can capture the spatial and spectral uncertainties of individual sources in the form of ellipsoids assuming Gaussian covariance. However as sources start to overlap, covariances between positions and spectra of model point sources cannot be captured by whether it is favorable to reject the null hypothesis for a single source. Instead, there can be multiple sources consistent with the observed data with a potentially complicated spatial and spectral covariance. Generalization of the frequentist approach to pairs and even multiplets of point sources can in principle probe such covariances between point source positions and fluxes. However the computational complexity quickly increases before a significant fraction of the parameter space volume can be explored. Also, the best fit delta log-likelihood comparison of models with different numbers of point sources becomes ill-defined since a point source model can fit the data at least as well as another with fewer sources. Therefore one needs to balance the goodness of fit of a point source model with its predictivity. If this type of across-model comparison is not properly handled, the maximum likelihood catalog will either blend point sources or over-fit the image by introducing spurious point sources. Traditional (deterministic) catalogs can reduce large amounts of observations to relatively concise lists of point sources, precluding false positives with the use of a hard significance cut that discards subthreshold information. However it is important to keep in mind that catalogs are still models that describe our state of knowledge consistent with the given data up to statistical and systematic errors. Especially in the crowded field limit, the uncertainties on the number, localizations and spectra of candidate point sources are more complicated than the usually adopted Gaussian form. Therefore keeping a fair ensemble of realizations of the underlying catalog space properly propagates the uncertainties to subsequent analyses that rely on the catalog. This is in contrast with the frequentist approach of representing our state of knowledge about the point sources with an estimator of the most likely catalog. These concerns make a case for adopting a Bayesian approach to point source inference. In this paper we construct a Bayesian framework, Probabilistic Cataloger (\texttt{PCAT}), to perform probabilistic point source inference. In this setting, the hypothesis space is the union of emission models that have a number of point sources between $N_{min}$ and $N_{max}$, along with parameters characterizing the diffuse emission and the PSF. Therefore the number of point sources in a member model, $N$, itself becomes a discrete parameter of the \emph{metamodel}, i.e., top-level model. We then sample from the posterior of the metamodel given the observed data by implementing the necessary transdimensional proposals to jump across models. This precludes the necessity to run separate MCMC simulations for each model in order to estimate the Bayesian evidence, which is subject to large uncertainties. When marginalized over all other parameters, the posterior distribution of the model indicator, i.e., the number of point sources in the model, can be used to calculate the relative evidence (the Bayes factor) between models. Given that detailed balance is ensured at each across-model proposal, models with too many point sources not justified by the data are less frequently visited, since most of the added parameter space is wasted, i.e., inconsistent with the data. Therefore the resulting Bayes factor penalizes models for the loss of predictivity as well as goodness of fit. Our work inherits elements from and builds on probabilistic cataloging discussed in \citep{Hogg2011}. The resulting statistical model is hierarchical in the sense that the hierarchical priors we place on the point source parameters, e.g., positions, fluxes and colors, are parametrized by a small set of hyperparameters, which, in turn, admit hyperpriors. During the evolution of the MCMC state, the sampler can also propose to change the hyperparameters by respecting the imposed prior. Because they parametrize the prior distribution of the point source parameters, the posterior distribution of the hyperparameters encode our state of knowledge about the population characteristics. This allows one to statistically probe the source count function below the detection threshold of traditional catalogs. In general, there are two distinct questions that can be posed about the underlying distribution of point sources in a given image: \begin{itemize} \item What is the flux distribution of the most significant $N$ point sources? \item What is the number of point sources above a given minimum flux, $f_{min}$? \end{itemize} For small values of $f_{min}$, traditional catalogs, by construction, can only address the first question, whereas a probabilistic catalog can provide an answer to both questions after proper marginalization of the posterior samples. In principle, it is also possible to float the hyperparameter, $f_{min}$, during the generation of a probabilistic catalog. However, in the limit of arbitrarily small $f_{min}$, diffuse origin hypothesis is nearly degenerate with that of a population of unresolved point sources. Therefore, our hierarchical model becomes insensitive to changes in $f_{min}$, when $f_{min}$ becomes much smaller than the typical fluctuations in the image. We pay particular attention to how we choose $f_{min}$, which will be discussed in Appendix \ref{sect:info}. Another approach to probing the source count function at the faint end is the fluctuation analysis where the 1-point probability distribution function of the emission is used to estimate the contribution of unresolved point sources to the total emission \citep{Scheuer1957}. By modeling the tail of the distribution of deviations due to unresolved point sources just below the detection threshold, fluctuation analysis can distinguish truly isotropic emission from unresolved point sources. The method has been considered across the whole electromagnetic spectrum, e.g., in radio \citep{Vernstrom2014, Condon2012}, Far Infrared (FIR) \citep{Friedmann2004}, x-rays \citep{Fabian1992, Worsley2004} and gamma-rays \citep{Faucher-Giguere2010, Zechlin2015, Lee2014}. After introducing the Bayesian framework, we use our formalism to construct a probabilistic catalog of the gamma-ray sky in the North Galactic Polar Cap (NGPC). We then compare our results with the traditional 3FGL catalog published by the Fermi-LAT Collaboration and show that our median catalog agrees with the 3FGL as well as revealing low-significance point sources. Nevertheless, the real benefit of constructing a probabilistic catalog becomes apparent in crowded regions such as the inner regions of the Milky Way. Given the intriguing possibility that the inner galaxy GeV excess could be due to unresolved point sources \citep{Lee2016, Bartels2016}, we leave the application of probabilistic cataloging to the inner galaxy data, to a dedicated future work. The rest of the paper is structured as follows. We begin in Section \ref{sect:modl} by introducing our emission model, leading into a discussion of our hierarchical inference framework. Then, in Section \ref{sect:samp} we present the sampling method we use for probabilistic cataloging. We present our results on mock data and the NGPC region in Sections \ref{sect:mock} and \ref{sect:ngal}, respectively. We then provide a discussion of our results in Section \ref{sect:disc} and conclude in Section \ref{sect:conc}. \section{The hierarchical Bayesian model} \label{sect:modl} \subsection{Modeling photon emission} The emission from a point source can be modeled as a delta function in position space with some parametrized energy spectrum, convolved with the spatial and spectral instrument response, i.e., the point and line spread functions, of the measuring instrument, respectively. Given that the Fermi-LAT energy resolution of $\sim 10\%$ is smaller than our energy bin width of $\sim 100\%$, we assume infinite energy resolution. The delta function at the position of each point source gets convolved with the PSF, $\dd{\mathcal{F}}/\dd{\Omega}$, in units of the fraction of total flux per solid angle. This yields the model point source flux map, $\mathcal{P}_{im}$, in energy bin $i$ and PSF class $m$, with units of 1/cm$^2$/s/sr/GeV, when summed over all point sources. \begin{align} \mathcal{P}_{im} (l, b) = \sum_{n=1}^N & \dv{\mathcal{F}_{im}}{\Omega} (l_n, b_n) f_{ni} \end{align} Here, $f_{ni}$ is the flux of the $n^{th}$ point source in the $i^{th}$ energy bin. We assume a power law for the spectral energy distribution so that \begin{align} f_{ni} = f_n \Bigg(\frac{E_i}{E_0}\Bigg)^{-s_n}, \end{align} where $E_0=1.7$ GeV is the pivot energy, $f_n$ and $s_n$ are the normalization and power law index of the spectral flux of each point source. At $\sim$ GeV energies, simply modeling the spectra of the point sources using a power law is not an accurate description of the data. Galactic gamma-ray emitters such as pulsars are known to exhibit an exponential cutoff in their spectra at $\sim 1-10$ GeV. Nevertheless, since our Region of Interest (ROI) in this work is restricted to the NGPC, most of the sources are expected to be extragalactic. Moreover, extragalactic gamma-ray sources such as blazars and radio galaxies have curved spectra, i.e., a power law with a running index, due to their broadband inverse Compton emission on the external radiation field of the Active Galactic Nucleus (AGN). To address this concern we restrict the energy span to 1.5 decades, i.e., between 0.3 GeV and 10 GeV, where curved spectra can be approximated using a single power law. We do not attempt the full probabilistic regeneration of the 3FGL, which uses data between 100 MeV and 100 GeV by relying on a spectral model with a larger number of free parameters. For a given ROI, there may also be emission from diffuse or extended sources. We therefore include into our model the diffuse emission prediction provided by the Fermi-LAT Collaboration \citep{Acero2016}, which accounts for Inverse Compton Scattering (ICS) emission due to upscattering of star light by relativistic electrons as well as gamma-rays from pion decay and bremsstrahlung due to the interaction of cosmic rays with gas and dust. At high galactic latitudes such as in the NGPC, where the latter process dominates, the model is roughly proportional to the ISM column density \citep{Ackermann2011} as traced by dust \citep{Schlegel1998}. In addition to the spatially varying diffuse emission model, we also add an isotropic template into our background prediction. This component serves two purposes. First, it models isotropic emission, whether of cosmic or instrumental origin. Second, it absorbs potential emission from point sources below the inclusion flux limit. With the addition of the background emission, the total model becomes \begin{align} \mathcal{M}_{im} = \mathcal{P}_{im} + \mathcal{D}_i + \mathcal{I}_i. \end{align} We sample $\mathcal{M}_{im}$ over a \texttt{HealPix} grid of resolution $n_{side} = 256$. The normalization of both templates are allowed to float in all energy and PSF classes and admit log-uniform priors. Ideally, the diffuse model prediction, $\mathcal{D}_i$, should be smoothed to match the PSF kernel of the data in each PSF class, $m$. We did not need to perform this operation, however, given the fact that the diffuse model does not have bright features in the NGPC. In order to be able to marginalize over uncertainties in the background prediction, we allow the normalizations of the background templates to float in each energy bin. We will denote the normalization parameters with the parameter vector $\vec{A}$ in what follows. Finally, in a photon counting experiment, the likelihood function is the Poisson probability of observing $k_{ijm}$ counts given a mean of $\mu_{ijm}$ counts in the $i^{th}$ energy bin, $j^{th}$ pixel and $m^{th}$ PSF class. We then sum the log-likelihood over all pixels, energy bins and PSF classes to construct a joint log-likelihood across different energy bands and PSF classes. \subsection{Modeling the population characteristics of point sources} In general, the set of prior beliefs about the statistical behavior of a model parameter, $\theta$, can be encapsulated in its prior probability distribution, $P(\theta)$. However, in this work we express the prior on the model parameters using a hierarchical structure, which requires a distinction between different levels of prior belief. By using the word \emph{prior}, we refer to the first level in the hierarchy. This includes the prior belief that all model point source fluxes are drawn from a power-law with index $\alpha$ and that the Poisson mean of the number of point sources, $N$, is $\mu$. For this work on the NGPC, we assume that the point source positions are drawn from the uniform distribution and that the colors have a Gaussian distribution. We further assume that the model point sources have vanishing n-point spatial or spectral correlations and are independent and identically distributed realizations of an underlying population. The prior probability distribution of the number of point sources, $N$, is given by \begin{equation} P(N|\mu) = \frac{\mu^N}{N!}e^{-\mu}. \label{equa:numbpnts} \end{equation} The hyperparameter $\mu$ is taken to be log-uniform distributed such that \begin{equation} P(\mu) = \dfrac{1}{\ln \mu_{max} - \ln \mu_{min}} \dfrac{1}{\mu} \end{equation} for $N_{min} < \mu < N_{max}$ and vanishes otherwise. This choice yields a scale-free prior on the number of point sources. Note that we use the same notation to refer to probability densities of continuous variables, e.g., $P(\mu)$, and probabilities of discrete variables, e.g., $P(N)$. When the parameter is discrete, the notation implies probability, whereas it refers to the probability per differential interval in the parameter, if the parameter is continuous. \explain{In our model, there are both continuous parameters and a discrete parameter. Therefore, when referring to the probability densities of continuous variables in the first draft, we used the notation $dP/da$ for some continuous random variable $a$. This was intended to emphasize the fact that we refer to probability per differential interval of parameter $a$. However, we did not maintain this convention throughout the paper to limit notational clutter and sometimes used the simpler $P(a)$ notation, which may be confusing. In this revision, we always use the latter and clearly mention in the text that we refer to probability densities when the parameter is continuous and to probabilities when the parameter is discrete.} Furthermore, we assume that the flux of the $n^{th}$ point source in the pivot energy bin 1 - 3 GeV, $f_n$, is distributed as a power law between some $f_{min}$ and $f_{max}$ with the index $\alpha$ at the central flux bin of $\sim 10^{-9}$/cm$^2$/s/GeV. \begin{align} P(f_n|\alpha) = \begin{cases} \dfrac{1-\alpha}{f_{max}^{1-\alpha} - f_{min}^{1-\alpha}} f_n^{-\alpha} \text{ for } f_{min} < f_n < f_{max} \\ 0 \text{ otherwise} \end{cases}. \label{equa:flux} \end{align} We place a uniform prior on the angle described by this slope, which yields \begin{equation} P(\alpha) = \dfrac{1}{\tan^{-1}(\alpha_{max}) - \tan^{-1}(\alpha_{min})}\dfrac{1}{1+\alpha^2} \end{equation} for $\alpha_{min} < \alpha < \alpha_{max}$ and 0 otherwise. \subsection{PSF modeling} \label{sect:psfn} A particle detector inevitably introduces errors when the arrival direction of a photon event is reconstructed. At small angular deviations, the random scatter around the true arrival direction can be modeled as a Gaussian \begin{align} \mathcal{G}(\theta_0; \sigma) &= \frac{1}{\sqrt{2\pi \sigma^2}} \exp\Bigg(- \frac{\theta^2_0}{2\sigma^2}\Bigg), \end{align} where $\theta_0$ is the angular distance from the true direction. The variance of the Gaussian is the sum of variances due to the resolution of the silicon strips and multiple scattering, which itself scales with energy. At large angular deviations, however, the Fermi-LAT PSF instead follows a power-law. A convenient function that approximates a Gaussian at small deviations while converging to a power law at large deviations is the King function \begin{align} \mathcal{K}(\theta_0; \sigma, \gamma) &= \frac{1}{\sqrt{2\pi \sigma^2}} \Bigg(1-\frac{1}{\gamma}\Bigg) \Bigg(1 + \frac{\theta_0^2}{2\sigma^2\gamma}\Bigg)^{-\gamma}, \end{align} where the Gaussian scale is given by $\sigma$ and the power law slope is set by $\gamma$. The PSF modeling of the Fermi-LAT Collaboration uses a weighted sum of two such functions \citep{Ackermann2013a}. Therefore, there are $5N_eN_{psf}$ independent parameters in the model, where $N_e$ and $N_{psf}$ are the number of energy bins and PSF classes. We use this model when we do not float the PSF parameters. When none of the point sources in the ROI are bright enough, such as Geminga or Vela, to constrain the PSF tails, floating them causes large uncertainties in the inferred background and point source fluxes. This is because the tails of the PSF are nearly degenerate with the background normalization and allowing them to float without strong priors causes a significant bias in the flux predictions. We generalize the algorithm to sample from the PSF parameters, when needed, and float the PSF during our mock runs. However, we fix the PSF during the nominal data run and discuss the results of letting the PSF float in Appendix \ref{sect:psfnvari}. The ability to float the PSF is especially relevant for ground-based optical astronomy, where the PSF is different in each exposure. In the other case, where we do float the PSF, a typical ROI without exceptionally bright point sources cannot constrain a double-King function with a floating scale factor. In particular, a bright source is needed to break the near-degeneracy between the power law slopes of the core and tail components of the radial profile. Therefore, in that case we fix the scale factor to the best-fit value provided by the Fermi-LAT Collaboration and use a linear combination of a King function and a Gaussian with only $4N_eN_{psf}$ free parameters. We place uniform priors on the logarithm of the angular scale, $\sigma$, the relative fraction of the Gaussian and King components, $f$, and the inverse tangent of the slope, $\gamma$. \begin{equation} \mathcal{P} = a_G \mathcal{G}(\theta_0; \sigma_G) + a_K \mathcal{K}(\theta_0; \sigma_K, \gamma) \end{equation} Inference of the point source catalog sensitively depends on the PSF modeling. Due to the power-law tails of the PSF, bright members of the point source population can outshine the faintest point sources even $\sim$ a few degrees away as shown in Figure \ref{figr:eval}. In the rest of the paper, we will collectively denote the set of parameters characterizing the PSF, with the parameter vector $\vec{\eta}$. \subsection{Prior structure} A probabilistic graphical model of our inference framework is presented in Figure \ref{figr:grap}. In this representation, nodes denote random variables, while edges directed into a node denote the set of nodes that hierarchically parameterize the probability distribution of the given node, such as in Equations \ref{equa:numbpnts} and \ref{equa:flux}. The red, blue, green and yellow nodes represent our model parameters, which are assigned prior probability distributions. In particular, the red nodes are the hyperparameters $\mu$ and $\alpha$, which set the normalization and slope of the point source flux distribution, respectively. The blue node indicates the number of point sources in a model, i.e., is the multiplicity of each green node. Likewise, the green nodes are the point source parameters, i.e., longitude, latitude, flux and spectral index from left to right. The yellow nodes are the background normalizations, $\vec{A}$, and PSF parameters, $\vec{\eta}$. $\mathcal{M}$ node is a deterministic function of the above model parameters representing the set of forward modeled photon count maps. Finally, $\mathcal{D}$ represents the observed photon count maps, whose consistency with the former drives the evolution of the MCMC state through the Poisson likelihood. We also color code the directed edges such that black edges denote a probabilistic relation, whereas olive lines show a deterministic relation. Finally, magenta lines imply that the multiplicity of the destination node is set by the origin. Note that we do not use plate notation, since the multiplicity itself is a discrete parameter in our model, which admits a hierarchical prior. Given the parameter and hyperparameters introduced above, the joint prior probability distribution of a model with $N$ point sources becomes \begin{multline} P(\mu, \alpha, N, \vec{A}, \vec{\eta}, \vec{l}, \vec{b}, \vec{f}, \vec{s}) = \\ P(\mu) P(\alpha) P(N|\mu) P(\vec{A}) P(\vec{\eta}) \prod_{n=1}^N P(l_n) P(b_n) P(f_n|\alpha) P(s_n) \end{multline} where we use the vector notation to refer to the union parameter set of $N$ point sources, e.g., $\vec{l} \equiv (l_1, l_2, ..., l_N)$. \explain{This equation replaces Equation 21 in the previous draft.} \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth, trim=4.cm 2cm 2.cm 1.cm, clip]{grap.pdf} \caption{A Probabilistic Graphical Model (PGM) of our transdimensional model. See the text for details.} \label{figr:grap} \end{figure} \subsection{ROI margins} The observed count data in the ROI can potentially be affected by point sources outside the ROI. In order to model such emission from point sources just outside the ROI, we make the spatial prior region slightly larger (1 degree larger on all sides) then the ROI window over which the likelihood is calculated. Therefore the model point sources can move slightly out of the image and probe whether a feature close to the boundary can be fit better by a model point source outside the ROI. As a result the offset provides a smooth transition from a data-informed region well inside the ROI to a prior dominated region outside the ROI, where the posterior asymptotes to the prior. This can be clearly seen in the artificial accumulation of sampled point sources along the boundary of the stacked posterior, as will be shown in Section \ref{sect:ngal}. \section{Sampling Method} \label{sect:samp} In this section, we describe the method used to sample from the probability distribution of catalogs consistent with the given photon count map. The starting point is the Bayesian assumption that there exists an underlying probability distribution that characterizes our knowledge of the model parameters. These parameters are the longitude, $l$, latitude, $b$, flux, $f$, and color, $s$, of each point source. Refer to Section \ref{sect:modl} for details on how we define and place priors on these parameters. The parameter space of the point source metamodel is, then, the union of the parameter spaces of the point source models that contain from $N_{min}$ up to $N_{max}$ point sources. Denoting this space as \emph{the catalog space}, $\mathcal{C}$, we therefore sample from \begin{equation} \mathcal{C} = \bigcup_{N=N_{min}}^{N_{max}} \mathcal{C}_N = \bigcup_{N=N_{min}}^{N_{max}} \mathcal{L}_N \times \mathcal{B}_N \times \mathcal{F}_N \times \mathcal{S}_N, \end{equation} where $\mathcal{C}_N$ is the catalog subspace with $N$ point sources. $\mathcal{C}_N$ can further be written as the product space of $\mathcal{L}_N$, $\mathcal{B}_N$, $\mathcal{F}_N$ and $\mathcal{S}_N$ that denote the longitude, latitude, flux and color spaces of point sources in the model with $N$ point sources. Although sampling from a given $\mathcal{C}_N$ can be performed by constructing an MCMC chain using ordinary Metropolis updates over $\mathcal{L}_N$, $\mathcal{B}_N$, $\mathcal{F}_N$ and $\mathcal{S}_N$, the fact that the catalog space is transdimensional means that it cannot be explored by simple within-model proposals. We assume that the number of point sources in the image is not known a priori. Therefore, the dimensionality of the point source model, $N$, becomes a discrete parameter subject to inference. In addition to the catalog space, the complete hypothesis space includes normalizations of the isotropic and spatially varying background models in each energy bin and PSF class, as well as parameters that characterize the PSF. We will defer the discussion of these degrees of freedom to Section \ref{sect:modl}, since they have fixed dimensionality and can be explored using ordinary within-model updates. \subsection{Trans-dimensional sampling} In the rest of the paper we will generically refer to a parameter in the model by $\theta$. The objective is to sample from the posterior distribution of $\theta$ given the observed count map $D$, $P(\theta|D)$, after updating our prior beliefs about $\theta$, $P(\theta)$. The Bayesian update is accomplished through the likelihood of observing the data given our model, $P(D|\theta)$. For this purpose we construct a Markov chain of states, $\{\theta\}$, whose asymptotic stationary distribution is the desired posterior distribution. Therefore we require that the chain is reversible with respect to the posterior, i.e., satisfies detailed balance condition. If the sampling space was fixed dimensional, this would imply that \begin{multline} \int P(\theta|D) Q(\theta^\prime|\theta) \alpha(\theta^\prime|\theta) \dd{\theta} = \\ \int P(\theta^\prime|D) Q(\theta|\theta^\prime) \alpha(\theta|\theta^\prime) \dd{\theta^\prime}, \end{multline} where $\theta$ and $\theta^\prime$ are the current and the proposed states, $Q(\theta^\prime|\theta)$ is the transition kernel from $\theta$ to $\theta^\prime$ and $\alpha(\theta^\prime|\theta)$ is the appropriate proposal acceptance probability, \begin{equation} \alpha(\theta^\prime|\theta) = \min \Bigg( 1, \frac{P(\theta^\prime|D)}{P(\theta|D)} \times \frac{Q(\theta|\theta^\prime)}{Q(\theta^\prime|\theta)}\Bigg). \label{equa:acppmths} \end{equation} Note that when sampling from the catalog space, $\theta$ and $\theta^\prime$ can have different dimensions. This brings up an interesting issue that the chain can no longer be reversible in the transdimensional case, since the transition is not one to one. The remedy is to draw random auxiliary variables, $u$ and $u^\prime$, using the densities $g(u)$ and $g^\prime(u^\prime)$, to match the dimensions of the initial and final states of the transition such that the mapping $(\theta, u) \leftrightarrows (\theta^\prime, u^\prime)$ is a diffeomorphism, where \begin{align} H(\theta, u) &= (\theta^\prime, u^\prime) \\ H^{-1}(\theta^\prime, u^\prime) &= (\theta, u). \end{align} The dimension matching, \begin{align} \mathcal{D}(\theta) + \mathcal{D}(u) = \mathcal{D}(\theta^\prime) + \mathcal{D}(u^\prime), \end{align} where $\mathcal{D}$ denotes the dimension operator, conceals the transdimensional nature of the proposal and ensures reversibility. In this case we require \begin{multline} \int P(\theta|D)g(u)\alpha(\theta^\prime|\theta) \dd{\theta}\dd{u} = \\ \int P(\theta^\prime|D)g^\prime(u^\prime)\alpha(\theta|\theta^\prime) \dd{\theta^\prime}\dd{u^\prime}. \label{equa:rjmcbala} \end{multline} The transition is accomplished through the transition kernel, $H(\theta, u) = (\theta^\prime, u^\prime)$, which replaces the probabilistic transition kernel, $Q(\theta^\prime|\theta)$, in the fixed-dimension case. For example, in the case where the proposal is to add a new point source, the auxiliary parameters, $u$, are simply the parameters describing the new point source. This method is known as the Reversible Jump MCMC (RJMCMC) \citep{Green1995, Green2008, Hastie2012}, which is a variant of MCMC that allows across-model moves in a pool of models indexed by their dimensionality. We inherit the reversible jump formalism in implementing transdimensional proposals. Given the freedom to jump across models, there are an infinite number of ways to propose such transitions using the current state. However, only some are useful schedules to explore the catalog space. Denoting the probability of the $m^{th}$ type of proposal in state $\theta$ by $j_m(\theta)$, the ratio, \begin{equation} \frac{j_m(\theta^\prime)}{j_m(\theta)}, \end{equation} should be also included in the resulting acceptance rate in order to compensate for any bias in the proposal frequencies. Using the new detailed balance condition, Equation \ref{equa:rjmcbala}, one finally obtains, \begin{gather} \alpha(\theta^\prime|\theta) = \min(1, \alpha_0), \\ \alpha_0 = \frac{P(D|\theta^\prime)}{P(D|\theta)} \frac{P(\theta^\prime)}{P(\theta)} \frac{j_m(\theta^\prime)}{j_m(\theta)} \frac{g(u^\prime)}{g(u)} \Bigg\lvert \frac{\partial(\theta^\prime, u^\prime)}{\partial(\theta, u)} \Bigg\rvert, \label{equa:accp} \end{gather} since the coordinate transformation requires that probability is conserved, i.e., \begin{equation} \dd{\theta^\prime}\dd{u^\prime} = \Bigg\lvert \frac{\partial(\theta^\prime, u^\prime)}{\partial(\theta, u)} \Bigg\rvert \dd{\theta} \dd{u}, \end{equation} for all $\theta$ and $u$. The usual within-model proposals are recovered when both $u$ and $u^\prime$ have the same dimension. Given that the parameter space of the point sources has large covariances in crowded fields, it is likely that the sampler can get stuck in a likelihood island without being able to efficiently visit all high-likelihood regions. In order to prevent this, we take heavy-tailed Gaussian steps when proposing within-model transitions. However, large changes in parameters can also suffer from the prior ratio in Equation \ref{equa:accp}, especially when the prior is a power law such as for the fluxes of point sources. We therefore set the prior ratio to unity by transforming the parameters such that their prior distributions become uniform. This is accomplished through the use of the Cumulative Distribution Function (CDF) and its inverse, which map the parameters to the unit space, $\mathcal{U} = [0, 1]$, and back. Therefore, the actual sampling is performed in the unit space with the prior ratio set to unity by definition at the expense of inverse transforming the parameters back to their genuine values when the likelihood function is to be evaluated or a hyperparameter is updated. This excludes the number of point sources in the model, $N$, which is an integer parameter. Conventional updates to point source positions, fluxes or colors, i.e., within-model proposals, only allow the sampler to explore one catalog subspace, $\mathcal{C}_N$. To explore the full space, $\mathcal{C}$, the sampler must propose updates that change $N$, by adding or removing sources. We use two pairs of such updates, where $u$ and $u^\prime$ have different dimensions. The first type is the elementary operation of adding or deleting a point source to or from the current list of point sources. We denote this pair of proposals birth and death. They are the reverse proposals of each other unlike within-model updates that are manifestly reversible. Therefore both birth and death have to be present in the set of possible proposal types in order for detailed balance to be respected. When a point source is to be added, an auxiliary vector, $u$, is drawn whose elements are distributed uniformly between 0 and 1. We then use the inverse CDF of the position, flux and color distribution functions to transform the uniformly distributed elements to random draws of the position, flux and color. These parameters determine the parameters of the point source to be added. In contrast, when a point source is to be killed, the CDF transform of its parameters define $u$. The second pair of transdimensional proposals are splits and merges, where a point source is split into two, and two point sources are merged into one, respectively. This pair is especially important in the crowded field, where splits and merges make the exploration of the nearly degenerate regions of the parameter space more efficient. We refer the reader to Appendix \ref{sect:accpprob} for the implementation details of the move types. \subsection{Time performance of the sampler} Sampling from the posterior probability distribution of catalogs is a computationally demanding task. The primary reason is the large and variable number of parameters in the sample vector. For a typical MCMC run on a mock image with 200 point sources, the sample vector can have as many as $\sim$ 1000 parameters. Since position and flux changes are proposed one at a time in order to keep the acceptance ratio high, the typical autocorrelation time becomes $\sim$ 10000 MCMC steps. Furthermore the least significant features in the image require the highest number of MCMC steps for convergence, because they are sampled only slightly more frequently compared to a uniform Poisson background. Increasing the performance requirements of the sampler is the fact that hyperparameters parametrize the hierarchical priors on the point source parameters. This means that the convergence of parameters precedes that of the hyperparameters. A typical inference thus requires $\sim 10^8$ MCMC steps. For each proposal in an MCMC run, the current parameter vector is used to compute the model count map, which is then compared to the data count map through the Poisson likelihood. This process typically dominates the time budget and requires careful optimization for the sampler to be scaled up to large ROIs, larger number of point sources or energy bins. We break the optimization into four steps. \begin{itemize} \item Although the sampling is actually performed in the CDF-transformed parameter space, the inverse CDF transforms of the parameters are stored along with those of the CDF-transformed parameters, precluding redundant CDF transformations. \item Flux maps are calculated perturbatively. During each proposal the current flux map is modified by the updated, added or killed point sources. This decreases the time complexity of processing a single sample from $\mathcal{O}(N_{pix}N)$ to $\mathcal{O}(N_{pix})$, where $N$ is the number of point sources and $N_{pix}$ is the number of pixels in the ROI. \item For each point source, the PSF is evaluated over a subset of pixels that lie inside a circle. The list of nearby pixels for each pixel is precomputed and stored as a look-up table. The radius of the circle depends on the flux of the associated point source and is determined such that the largest flux allowed by the prior contaminates the lowest flux at most by one percent. This is illustrated by Figure \ref{figr:eval}, where the horizontal line indicates 0.01$f_{min}$, where $f_{min}$ is the minimum flux allowed by the model. Hence, point sources contribute to the total flux map up to the radius, after which the bias introduced by their neglect is significantly below the faintest possible point source. We monitor the error introduced by this approximation and ensure that the bias is negligible in all pixels, energy bins and PSF classes. As a result of this approximation the time complexity of a single proposal is further reduced to $\mathcal{O}(1)$. This implies that the time complexity of a single proposal (except those that update the PSF or hyperparameters) does not depend on how many sources or pixels there are in the image. \item The leading contribution to the time complexity of the average sample comes from the likelihood evaluation, which requires the computation of the proposed change to the flux over a set of data cubes, e.g., pixels, energy bins and PSF classes. In this operation, the PSF is evaluated by computing the angular distance from the point sources to the pixel centers. In order to accelerate this computation, we precompute unit vectors to the \texttt{HealPix} pixel centers, $\hat{u}_{pix}$. This allows us to compute the angular distance, $\theta_0$, using the dot product \begin{equation} \theta_0 = \arccos(\hat{u}_{PS} \cdot \hat{u}_{pix}), \end{equation} where $\hat{u}_{PS}$ is the unit vector along the point source of interest. Furthermore, we store the radial profile of the PSF in the memory and interpolate it when calculating the flux updates. In overall, the employed acceleration scheme reduces the mean computation time per sample to $\sim 6$ ms. Nevertheless, the evaluation of the PSF takes $\sim 5$ ms per sample on the average and dominates the time budget of a typical sample. \end{itemize} \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{eval.pdf} \caption{The radial profile of emission from a point source as reconstructed by the Fermi-LAT for different flux bins. The tail of the PSF is a power law and is not exponentially suppressed. The horizontal line highlights the minimum flux allowed by the prior multiplied by 0.01. When the position, flux or the color of a point source is changed, its contribution to the total flux map is updated only inside a circle centered at the pixel closest to the point source. The radius of the circle is set such that this approximation causes a negligible error.} \label{figr:eval} \end{figure} In order to scale the algorithm to larger ROIs, multiple energy bins or epochs, the sampler has been parallelized to run over multiple cores. The current time performance is adequate for analyzing full-sky datasets such as that of Fermi-LAT in $\sim 5000$ CPU hours. Although the execution time per sample increases linearly with the typical number of point sources in the sampler, the time it takes to get an independent sample from the catalog space scales roughly as the square of the number of point sources. This is due to the fact that as the number of parameters increases, one either has to take smaller steps in the model space, or make changes in fewer number of parameters in a given proposal, both of which increase the autocorrelation of the resulting chain. Therefore, a slight increase in the ROI size may significantly increase the convergence time. Because this scaling is not desirable for large fields or full sky analyses, it is more feasible to separately sample from the catalog space of patches that are $\sim 10$ PSFs wide. This neglects covariances between sources and background emission in different patches, but allows scaling to much larger fields. However, further speed improvements would be needed for analyzing optical photometric data, where the typical pixel size is less than an arcsecond as opposed to being tens of arcminutes as in Fermi-LAT. In addition to the performance improvement gained by working on a Cartesian grid, this could be achieved using GPUs (Graphics Processing Unit) given the parallelizable nature of the sampling problem. Note that the time complexity of the overall convergence still depends on the minimum flux allowed by the model. This is because it sets the typical number of model point sources and, hence, the size of the parameter space. In a typical run with 3 energy bins, 2 PSF classes, a $40^\circ\times40^\circ$ ROI, $\sim 10^5$ pixels and $\sim 250$ model point sources, the execution time is 250 CPU hours. \subsection{Convergence diagnostics} MCMC formalism allows the exploration of complex posterior distributions with the caveat that convergence to the stationary distribution can require a long simulation time. Given that a typical catalog has thousands of parameters, the finite running time of the Markov chain might raise concerns over whether the sampled distribution is representative of the desired target distribution. One method for evaluating the chain convergence is to inspect the variance of the sampled chain. However the variance of a single MCMC chain can underestimate the true variance, since the realized chain may not have converged to the target distribution despite having a small variance. We therefore run multiple, usually around 20, chains and compare the mean of the chain variances to the variance of the means of the chains \citep{Gelman1992}, ensuring that the initial states of the chains are over-dispersed relative to the target distribution. Having $N_c$ chains and $N_s$ samples in each chain, the resulting test statistic, \begin{equation} \hat{R} = \sqrt{1 + \frac{B}{W} - \frac{1}{N_s}}, \end{equation} is known as the Potential Scale Reduction Factor (PSRF) and can be used to assess whether the chains have converged. Here, $W$ is the within-chain variance, i.e., the mean of the chain variances, \begin{equation} W = \frac{1}{N_c} \sum_c^{N_c} \Bigg( \frac{1}{N_s-1}\sum_s^{N_s}(y_s - \bar{y}_c)^2 \Bigg), \end{equation} $B$ is the between-chain variance, i.e., the variance of the chain means, \begin{equation} B = \frac{1}{N_c-1} \sum_c^{N_c}\Bigg(\bar{y}_c - \frac{1}{N_c} \sum_c^{N_c} \bar{y}_c \Bigg)^2, \end{equation} and the mean of the $c^{th}$ chain is \begin{equation} \bar{y}_c = \frac{1}{N_s} \sum_s^{N_s} y_s. \end{equation} Our definition of $B$ differs from that commonly found in the literature by a factor of $N_s$, which is absorbed into the definition of $\hat{R}$. In a well-mixed chain the PSRF should be close to unity. One caveat of using the PSRF as an estimator of the convergence in this framework is that the sampled chains are transdimensional. Note that the problem of point source inference has a labeling degeneracy. That is to say that there is an N!-fold degeneracy in the likelihood function of a point source model with $N$ point sources, since permuting the parameter labels of these $N$ point sources leaves the likelihood invariant. For any reasonably large $N$, $N!$ is larger than the number of samples that can be drawn from the posterior. Therefore, formal convergence is not possible, but also unnecessary, since well sampling only one of the degenerate likelihood peaks reveals the unique likelihood topology of the problem. In order to probe convergence, we instead monitor the variance of the resulting model emission map. We draw 1000 random voxels (triplets of pixel, energy bin and PSF class) and show the distribution of the PSRF in our data run (Section \ref{sect:ngal}) in Figure \ref{figr:gmrb}, which confirms that the between and within chain variances are similar for most pixels. \begin{figure} \centering \includegraphics[width=0.45\textwidth]{gmrbdist.pdf} \caption{The distribution of the Gelman-Rubin test statistic.} \label{figr:gmrb} \end{figure} \subsection{Autocorrelation of the chain} Given that the sampler updates, adds or kills one point source at a time, nearby samples in the chain are expected to be similar. In order to retain the Markovian property, we therefore thin the chain by a factor equal to the maximum number of parameters allowed by the metamodel. Typically this implies that the chain is thinned by a factor of 1000-10000. In order to ensure that the resulting chain has the Markovian property, we compute the autocorrelation of the chain, where we follow a similar method to that of diagnosing convergence using the predicted emission map. After calculating the autocorrelation for the randomly drawn 1000 voxels, we take the average over chains and parameters. We plot the resulting autocorrelation of the chain in Figure \ref{figr:atcr}, which shows that samples in the diluted chain is memoryless. \begin{figure} \centering \includegraphics[width=0.45\textwidth]{atcr.pdf} \caption{Autocorrelation of the diluted chain normalized by its value at zero lag.} \label{figr:atcr} \end{figure} \subsection{\texttt{PCAT}} We make the resulting software, \texttt{PCAT}, available to the astronomy community. \texttt{PCAT} is a pure \texttt{Python 2.7} implementation of the described sampling algorithm along with extensive routines to customize the sampler for the problem at hand, further process the output, diagnose convergence and visualize probabilistic catalogs. It is designed to sample from the catalog space for a given photon count map, choice of data binning and prior structure, where the level of background and the PSF are potentially unknown. As of version 0.2, it is only intended to process binned count maps from a Poisson process. \section{Mock runs} \label{sect:mock} \begin{figure*}[ht] \centering \begin{minipage}[b]{0.32\linewidth} \includegraphics[width=\linewidth]{mockdatacnts23A.pdf} \end{minipage} \begin{minipage}[b]{0.32\linewidth} \includegraphics[width=\linewidth]{mockmodlcnts23A.pdf} \end{minipage} \begin{minipage}[b]{0.32\linewidth} \includegraphics[width=\linewidth]{mockresicnts23A.pdf} \end{minipage} \caption{A synthetic photon count map of the NGPC, where the mock data is a Poisson realization of emission due to a point source population and the diffuse background. The left and middle panels show the mock photon count map, and a sample model count map, while the right panel is the residual. The color scales indicate the number of photons per pixel in all panels and are arcsinh stretched in order to make faint features more visible. Superimposed with the count maps are the true and sample catalogs, which are shown with green Xs and blue pluses, respectively. If a true point source is missed, it is instead indicated with a green square. Likewise, if the flux of the sample does not agree with that of the true point source despite spatial association, it is shown with a green star. The sizes of the markers are proportional to the logarithm of the fluxes of the corresponding point sources.} \label{figr:sampcnts} \end{figure*} In this section we will first present an ensemble of catalogs sampled using a mock, i.e., simulated, dataset. Hence we will compare the true (input) parameter values with those obtained from the ensemble. The Poisson mean of the mock map is generated by sampling all parameters randomly from the prior and then calculating the resulting emission map due to the background emission and the point sources. Finally mock data is generated as a Poisson realization of the mean emission. In total 300 point sources are generated from a flux distribution with a power law slope set to -1.8. We use the \texttt{Pass 7}, \texttt{source} class exposure map of the Fermi-LAT instrument between weeks 9 and 217, when calculating the mock number of counts from the synthetic emission map. This ensures that the mock point sources are subject to incompleteness at the same flux as the real point sources. Figure \ref{figr:sampcnts} summarizes a fair sample from the probabilistic catalog. The panels show the number of counts in the generated mock map, the sampled model map and the residual, respectively from left to right. Here, the shown sample is only one of many realizations that constitute a fair draw from the underlying probability distribution. The number of counts are given in the 1 GeV - 3 GeV energy bin. As can be seen in the right panel, Poisson fluctuations near bright point sources can be large even if the data is a realization of a generative model, i.e., the model is a good description of the data. In practice, further mismodeling of the PSF can result in large, and possibly, even coherent residuals around bright point sources. Depending on how low the model point source fluxes are allowed to go, this can result in the sampling of spurious point sources around bright ones. Therefore care must be taken to employ a PSF modeling that does not bias the flux distribution of the point source population. Figure \ref{figr:arrydatacnts} tiles together fair samples from the posterior showing the number of data counts in each pixel, i.e., similar to the left panel in Figure \ref{figr:sampcnts}, but showing real NGPC data instead. The grid illustrates the typical evolution of the MCMC state, where bright true point sources have an associated model point source with precise spatial and spectral localization, whereas faint true point sources are only sometimes associated with model sources. \begin{figure*}[ht] \centering \begin{minipage}[b]{0.9\linewidth} \includegraphics[width=\linewidth, trim=0 3cm 0 3cm, clip]{mosa.pdf} \end{minipage} \caption{The NGPC photon count map as measured by the Fermi-LAT (color scale), the 3FGL sources in the ROI (green markers) and six fair samples from the catalog space (blue pluses). The image is centered at the NGP and the axes correspond to a Cartesian projection about the NGP. The size of the markers are proportional to the logarithm of the flux of the point source. The color scale corresponding to the number of photon counts per pixel has been arcsinh stretched in order to emphasize faint features. The 3FGL sources are marked with a green square if the sample catalog does not have a model point source within 0.5 degree. Otherwise they are marked with a green X, indicating a hit. If the fluxes of the associated point sources disagree (sample flux outside the dashed lines in Figure \ref{figr:inptscatspec}), then the green X is replaced with a green star.} \label{figr:arrydatacnts} \end{figure*} Next, Figure \ref{figr:mockhistflux} shows the flux distribution of the mock (green) and median sample (black) with the 68\% credible interval. The 3FGL is also shown with red for reference. The 1$\sigma$ uncertainties cover the truth for most of the flux bins. Note that a flux-incomplete traditional catalog rolls off at the faint end since such sources are fainter than the typical fluctuations of the background. In a probabilistic catalog the flux distribution can be probed even below where a traditional catalog would roll off. This is because even though none of the point sources in the catalog samples is to be taken as true, repeated sampling of the flux distribution can constrain the population characteristics. However, this is only true for point sources more significant than $\sim 1 \sigma$. At yet lower fluxes, the sampled flux distribution function is informed by the prior more than the likelihood, and tends to follow whatever characteristic is imposed by the hierarchical prior, i.e., a power law with a range of indices allowed by the hyperprior. This transition from a likelihood dominated region to a prior dominated one is controlled by the minimum allowed flux of the point sources, $f_{min}$, and indicates where probabilistic cataloging becomes ineffective in constraining the population characteristics. We discuss how we determine the transition region in Appendix \ref{sect:info}. \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{mockhistflux.pdf} \caption{Posterior flux distribution function, showing its median, 68\% and 95\% quantiles. The green histogram shows the distribution of mock point sources. The red histogram highlights the 3FGL for reference. Note that the probabilistic catalog is statistically complete, i.e., has a statistically representative sample of sub-threshold point sources. The lower horizontal axis at the top shows the number of counts for the mean exposure in the ROI that corresponds to the flux axis. The top horizontal axis, then, gives the ratio of this number of counts to the number of background counts inside a Full Width at Half Maximum (FWHM), which is a proxy for the significance of a point source with the given mean number of counts in units of standard deviation.} \label{figr:mockhistflux} \end{figure} The former comparison confirms the statistical agreement between the true and median catalogs. However it is also desirable to perform an element-wise comparison between the two catalogs. Towards this purpose we associate a traditional (in this case, the true) catalog and a probabilistic catalog as follows. For a given sample from the catalog space, we initialize the association algorithm by setting the fluxes of all associations to the true catalog, to zero. Then, for each point source in the sample catalog, we ask if there are any point sources in the true catalog within 0.5 degrees. If so, we take the closest such true point source and add the model point source to the true point source's list of possible associations. We then repeat this for all model point sources in the sample catalog. Along the way it happens that multiple model point sources get matched to the same true point source. In that case, we use the gathered list of possible associations to select the closest model point source as \emph{the} association. After all samples in the probabilistic catalog has been so processed, we take the median flux associated with each point source in the traditional catalog. Figure \ref{figr:mockscatspec} illustrates the resulting correlation. The horizontal axis shows the true flux of the source, while the vertical axis marks the median of the fluxes associated to it, including potential zero fluxes due to missed associations. This procedure is generic to associations between any traditional and probabilistic catalog and not just to the association of the probabilistic catalog to the underlying true catalog. Furthermore, in Figure \ref{figr:mockscatspec} the vertical error bars denote the statistical uncertainty due to the stacking of the ensemble of catalogs. Horizontal error bars are not provided, since the true catalog is a generated catalog without any instrumental uncertainties. At the bright end, model point sources are statistically significant and well localized, yielding small vertical error bars. Moving towards the faint end, the correlation first broadens due to associated features on the image becoming comparable to Poisson fluctuations of the background. In the extreme faint limit, one expects a given true point source to be associated with a random model point source, completely suppressing the correlation. \begin{figure*}[ht] \centering \includegraphics[width=0.99\textwidth]{mockscatspec.pdf} \caption{The median sample flux of spatial associations between the true and the probabilistic catalog. The diagonal line as well as the flux mismatch tolerance lines used to tag associations with an empty green circle in Figures \ref{figr:sampcnts} and \ref{figr:arrydatacnts}, are shown with dashed gray lines. See the text for details on matching the two. In the absence of Poisson noise or systematic uncertainties, associations would lie on the diagonal line. The vertical error bars correspond to statistical uncertainty of the probabilistic catalog and generally increase towards lower fluxes, as the observed image becomes less informative. The outliers (associations far away from a one-to-one correlation) is an artifact of associating an ensemble of sample catalogs with a traditional one in the crowded field limit, where simply associating the closest model point sources with the true catalog causes loss of information and biases the result.} \label{figr:mockscatspec} \end{figure*} We then plot the posterior distribution of the flux distribution normalization and power law slope (Figures \ref{figr:mockfdfnnorm} and \ref{figr:mockfdfnslop}). Similarly, the posterior distribution of the hyperparameters cover the true values. For the mean number point sources, the relevant true value is the imposed number of mock point sources, which is 300. \begin{figure}[ht] \centering \includegraphics[width=0.5\textwidth]{mockfdfnnorm.pdf} \caption{The posterior of the normalization of the flux distribution function obtained for the mock data run.} \label{figr:mockfdfnnorm} \end{figure} \begin{figure}[ht] \centering \includegraphics[width=0.5\textwidth]{mockfdfnslop.pdf} \caption{The posterior of the power law slope of the flux distribution function obtained for the mock data run. The green line shows the parameter value of the mock (true) model.} \label{figr:mockfdfnslop} \end{figure} Lastly, Figure \ref{figr:mocknumbpnts} shows the posterior distribution of the model indicator, i.e., the number of point sources in the sample catalogs. This distribution is formally proportional to the posterior probability of the models assuming that the metamodel is true. Therefore it can be used to calculate the relative evidence (Bayes factor) for any two models. \begin{figure}[ht] \centering \includegraphics[width=0.5\textwidth]{mocknumbpnts.pdf} \caption{The posterior of the number of point sources obtained for the mock data run. The green line shows the true number of point sources.} \label{figr:mocknumbpnts} \end{figure} \section{Application to Fermi-LAT Data} \label{sect:ngal} Next, we show results using the gamma-ray data as measured by the Fermi-LAT instrument. In order to be able to compare with the 3FGL traditional catalog published by the Fermi-LAT Collaboration, we used the reprocessed \texttt{Pass 7}, \texttt{source} class data and generated \texttt{HealPix} sky maps with base resolution 256, in three energy bins between 0.3 - 1 GeV, 1 - 3 GeV and 3 - 10 GeV. Then we constructed a joint-likelihood by fitting the front and back converted sky maps separately and taking the total likelihood. This allows a more precise modeling of the PSF, since photons that convert at the top of the instrument, i.e., front-type events, have better angular reconstruction. The choice of energy binning coincides with that of the 3FGL catalog, in order to facilitate the comparison. Furthermore, since most of the sources at high galactic latitudes are time-variable blazars, we use data in the same time interval as that used to construct the 3FGL catalog, i.e., weeks 9 through 217. A probabilistic catalog, by construction, cannot be reduced to a single list of point sources. Nevertheless its statistical summary can still be compared to a traditional catalog. This provides a means of assessing the performance of our ensemble of catalogs against the well established 3FGL. As in the mock data case, Figures \ref{figr:inpthistflux} and \ref{figr:inptscatspec} show the flux distribution function and associations between the 3FGL and probabilistic catalog. \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{inpthistflux.pdf} \caption{Posterior flux distribution function as in Figure \ref{figr:mockhistflux}. This time the green histogram shows the distribution of the 3FGL points sources.} \label{figr:inpthistflux} \end{figure} The association between the 3FGL and the probabilistic catalog indicates that there is an agreement between the two. The correlation is stronger at larger fluxes, where covariances with the background normalization and PSF affect flux determinations negligibly. The level of agreement decreases towards lower fluxes since covariances with the background level and the radial profile of the PSF as well as the shape and normalization of the flux distribution widens the prediction of the probabilistic catalog. It is also worth noting that the scatter in Figure \ref{figr:inptscatspec} is partially due to the different spectral modeling used by the two catalogs. In other words, even if the flux predictions of the two catalogs were perfectly correlated in the pivot energy bin, 1 GeV - 3 GeV, fluxes in the other energy bins would show some dispersion. Determining the nature of a point source population, i.e., whether it is a pulsar or AGN and what subclass it belongs to, requires the reconstruction of its light curve and color distribution. We show the latter in Figure \ref{figr:inpthistsind}. The distribution around 2.2 implies that most of the point sources are blazars, where the upper tail is dominated by Flat-spectrum Radio Quasars (FSRQs) whereas BL Lacs make up the harder sub-population \citep{Dermer2016}. \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{inpthistsind.pdf} \caption{Posterior distribution of the point source colors.} \label{figr:inpthistsind} \end{figure} \begin{figure*}[ht] \centering \includegraphics[width=\textwidth]{inptscatspec.pdf} \caption{Association of the probabilistic catalog with the 3FGL using the same procedure as in Figure \ref{figr:mockscatspec}. Small departures from a perfect correlation with the 3FGL is partly due to different spectral modeling and partly due to the fundamentally different statistical approaches used to generate the catalogs.} \label{figr:inptscatspec} \end{figure*} By collecting the point source fluxes in the ensemble of catalogs, we also estimate the contribution of the point sources to the total emission in the NGPC. The median spectra of the point sources, isotropic component, and diffuse model are given in Figure \ref{figr:inptcompfracspec}. The diffuse model is observed to be the dominant component, accounting for $57\%\substack{+6 \\ -5}$ of the total emission. The isotropic component and the point sources account for the rest in roughly equal amounts, i.e., $25\%\substack{+4 \\ -3}$ and $18\%\substack{+2 \\ -2}$, respectively. However we note that the partitioning between the point sources and the isotropic component is set by our choice of the minimum allowed point source flux. If we lower the minimum flux allowed for the point sources, then the relative contribution of the point sources would account for some of the isotropic component. The reason for this near-degeneracy is that allowing the sampler to populate the image with point sources much fainter than the level of Poisson fluctuations of the background is equivalent to decreasing the isotropic background emission. In this sense, the question of how much of the emission is accounted for by the point sources should be addressed with reference to a particular minimum point source flux. \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{inptcompfracspec.pdf} \caption{Spectra of emission correlated with the Fermi diffuse model (dotted), isotropic model (dot-dashed) and the total emission from point sources (dotted) averaged over the ROI.} \label{figr:inptcompfracspec} \end{figure} Samples from the catalog space can be stacked together by binning catalog samples in space and flux. This yields a map of the probability of finding a point source at a certain direction and flux, given our model. In general, this should not be interpreted as an unconditional probability, since the model used may not contain the set of all backgrounds and point sources that could be consistent with the observed data. In this work, however, the background is relatively featureless and the allowed flux distribution covers nearly four decades with a floating power law slope. This makes the stacked posterior approximately equal to the probability of finding a point source at a certain direction and flux. We show in Figure \ref{figr:pntsbind}, the catalog samples binned spatially and stacked spectrally. The color scale gives the number of catalog samples, where a model point source lands in the associated pixel. All green stars, which show the locations of the true point sources are associated with model point sources. The hot pixels away from the true point sources show regions that have $\sim1-4\sigma$ count features. \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{pntsbind.pdf} \caption{Catalog samples binned spatially and stacked spectrally, giving the number of point sources in fair samples from the catalog space per pixel. Green stars indicate the locations of the true point sources.} \label{figr:pntsbind} \end{figure} The posterior distribution of the hyperparameters provides another handle on the population characteristics. For example Figures \ref{figr:inptfdfnnorm} and \ref{figr:inptfdfnslop} show the posterior of the normalization and power-law slope of the flux distribution function, respectively. Note that hyperparameters do not directly affect the likelihood. However their posterior distribution is still informed by the data. This is because the hyperparameter updates respect detailed balance with respect to the Poisson probability of observing the sampled model flux distribution and, in turn, model point source updates respect the Poisson probability of observing the data. \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{inptfdfnnorm.pdf} \caption{Posterior of the hyperprior on the normalization of the flux distribution for the real data run. The plot on the left is the MCMC time evolution of the quantity, whereas the right-hand side shows the histogram of the samples.} \label{figr:inptfdfnnorm} \end{figure} \begin{figure}[ht] \centering \includegraphics[width=0.45\textwidth]{inptfdfnslop.pdf} \caption{Posterior of the power-law slope of the flux distribution function for the real data run.} \label{figr:inptfdfnslop} \end{figure} We find the power law slope to be $-1.92\substack{+0.07 \\ -0.05}$. This is smaller than the expectation from a uniform distribution of equally bright blazars, i.e., \begin{equation} \dv{N}{f} = \dv{N}{r}\Bigg(\dv{f}{r}\Bigg)^{-1} \propto r^2 \times \Bigg(\frac{1}{r^3}\Bigg)^{-1} = f^{-5/2}. \end{equation} Previously, \citep{Abdo2010} found that the source count function has a slope of $-2.6 \pm 0.2$ at the bright end, which hardens to $-1.6 \pm 0.1$ at the faint end. Because we use a single power law, the resulting posterior converges to an intermediate value. Moreover, given that we perform sampling only over a $1600$ degree$^2$ patch around the NGP, this analysis is subject to more shot noise. We leave a large ROI, high latitude sampling to future work. As for the mock data, we show in Figures \ref{figr:inptnormisot} and \ref{figr:inptfdfnnorm}, the normalization of the isotropic template and the Fermi diffuse model for each energy bin, respectively. We find the median of the isotropic and diffuse backgrounds to be larger than unity by a factor of $\sim 1.3$ and $\sim 1.1$. Moreover, because the diffuse model is relatively featureless in the NGPC, the two normalizations have a large covariance. \begin{figure}[ht] \centering \includegraphics[width=0.5\textwidth]{inptnormisot2.pdf} \caption{The posterior probability distribution of the normalization of the isotropic template obtained for the real data run.} \label{figr:inptnormisot} \end{figure} \begin{figure}[ht] \centering \includegraphics[width=0.5\textwidth]{inptnormfdfm2.pdf} \caption{The posterior probability distribution of the normalization of the diffuse background template obtained for the real data run.} \label{figr:inptnormisot} \end{figure} Lastly, the Figure \ref{figr:inptnumbpnts} shows the posterior distribution of the number of point sources. We infer that there are $270\substack{+30 \\ -10}$ point sources in the ROI above $3 \times 10^{-11}$/cm$^2$/s/sr/GeV at the $1-3$ GeV energy bin. \begin{figure}[ht] \centering \includegraphics[width=0.5\textwidth]{inptnumbpnts.pdf} \caption{The posterior of the number of point sources obtained for the real data run.} \label{figr:inptnumbpnts} \end{figure} We make our probabilistic catalog available at the Harvard Dataverse \url{https://dataverse.harvard.edu} and refer the reader to Appendix \ref{sect:fits} for details on the data format. \section{Discussion} \label{sect:disc} The Bayesian approach to point source inference allows a more robust treatment of the parameter covariances compared to finding the most likely catalog. Furthermore, information encoded in the subthreshold features in the image is not discarded. An important motivational distinction between conventional and probabilistic cataloging is that the latter aims to find a set of point sources that is free of false positives, whereas member point sources in the former are not guaranteed to exist. It is the repeated sampling of the raw data along with the false positives that make probabilistic cataloging a very useful tool in probing population characteristics. The median flux of spatial associations of our catalog with the 3FGL agrees remarkably well when associated with the 3FGL fluxes. This demonstrates the feasibility and reliability of our probabilistic approach. We further verify the performance of probabilistic sampling in the unresolved regime by using mock data and demonstrate that the truth information is covered by our posterior. With the current implementation several thousand realizations of even full sky gamma-ray catalogs such as the 3FGL or 3LAC \citep{Ackermann2015}, which is an AGN catalog based on the 3FGL, can be generated. In fact, given the potential increase of computational resources available to catalog generation in astrophysics, it may even be feasible to make the next generation standard catalog reduction pipelines, probabilistic. As for the time domain analysis, FAVA \citep{Ackermann2013}, does not use a likelihood approach and therefore cannot be generalized using probabilistic cataloging. Given the time complexity of catalog sampling, it is unlikely that full sky catalogs can be sampled by binning data in time domain. Furthermore, flares or periodic flux variations of individual point sources will likely be washed out in the flux distribution. Since probabilistic sampling is most informative when studying the population characteristics, it is not particularly useful for time domain analyses. It may also be possible to generalize our method to other wavelengths such as optical datasets. However typical PSF size of less than an arcsecond in optical photometry restricts the application of probabilistic cataloging to $\sim$ deg$^2$ sized ROIs due to the large number of pixels. See \citep{Portillo2017} for an example of this technique applied to optical photometry in a crowded stellar field. When multiple samples are drawn from the catalog space, point sources in one sample catalog cannot simply be matched to those in other samples. This is due to the fact that the underlying likelihood function is invariant to permuting the labels of model point sources. This is known as the labeling degeneracy problem \citep{Zhu2014} and is common to all mixture models, where model components are not individually labeled. Therefore taking the ergodic average of a given model parameter without breaking this degeneracy becomes meaningless, since parameters change identity during the evolution of the chain. For instance in the limit of an infinitely long chain, the ergodic average of all flux parameters would be identical and, hence, not useful for inferring the flux posteriors of individual point sources. Nevertheless, inference of population characteristics does not necessitate breaking of the labeling degeneracy, since population characteristics are also invariant under permutations of point source labels. The time we need to explicitly break the labeling degeneracy is when comparing the chain of catalog samples to a traditional catalog. This requires a prescription to associate point sources. Associations at low fluxes inevitably causes loss of information and is the major reason for the increasing spread in Figures \ref{figr:mockscatspec} and \ref{figr:inptscatspec} towards low fluxes. Model choice in the Bayesian framework tries to establish a balance between the goodness of fit of a model and its predictive power. A point source model will fit a given count map at least as well as another with fewer point sources. In fact, in the limit of very large number of point sources, the point source model essentially becomes indistinguishable from a diffuse model with arbitrary number of spatial degrees of freedom. It is thus desirable that the chosen model covers the smallest possible volume when projected onto the data space while still fitting the observed data reasonably well. Known as the Occam's razor, this principle is encoded in the marginal likelihood. This quantity penalizes point source models for wasted parameter space, i.e., for only marginally increasing the goodness of fit at the expense of predicting extraordinary data, which significantly reduces the likelihood for most of the added parameter space. The difficulty in probing faint point sources is the inability of the Poisson-noise count data in constraining point sources fainter than the typical Poisson fluctuations of the background. This sets a fundamental count scale, $\sqrt{C_B}$, where $C_B$ is the mean number of counts expected from the background inside a FWHM for the given exposure. This scale necessitates a careful choice of the minimum point source flux allowed by the model, $f_{min}$. Let us denote the number counts that corresponds to $f_{min}$ by $C_{min}$ for the given exposure. In the limit $C_{min} \gtrsim \sqrt{C_{B}}$, the goodness of fit of the point source model gets reduced due to lack of favorable parameter space. On the other hand, if $C_{min} \lesssim \sqrt{C_{B}}$, then the data cannot constrain the parameter space. In order to assess the regime of a given prior choice, we evaluate the relative information gain in going from prior to the posterior and choose $f_{min}$ such that the posterior is sufficiently informed by the data. See Appendix \ref{sect:info} for a more detailed discussion. A well known problem of Poisson regression of photon count data is that of mismodeling. It is important to note that when a test statistic of maximum log-likelihood difference between the alternative and null models is interpreted as a detection significance, it is implicitly assumed that the underlying model is a good description of the data. Otherwise the unquoted systematic uncertainty can be much larger than the statistical uncertainty. For example, when a flux-incomplete point source model is used, this assumption will not hold true. Therefore predictions for the diffuse templates will be biased so as to minimize the residuals of the flux-incomplete point source model. Probabilistic cataloging addresses this problem by shunting uncertainties due to flux-incompleteness to a pure statistical form. This is accomplished through sampling in the catalog space above a given cutoff flux, which contains all point source configurations that the data is consistent with. By doing so, it allows many false-positives in the model. But these are sampled less frequently compared to the true-positives as long as the sampling is performed in the likelihood-dominated region. Note that probabilistic cataloging can still suffer from mismodeling due to imperfect background models. In this work, we follow Bayesian statistics to perform inference with the motivation that models can be penalized for introducing unnecessary point sources. Strictly speaking a frequentist approach that adds a degree of freedom penalizing term to the test statistic in the form of Akaike Information Criterion (AIC) \citep{Akaike1974} or Bayesian Information Criterion (BIC) \citep{Schwarz1978} could achieve the same goal by iterating the fit over a range of number of point sources. However in the limit of faint point sources, the likelihood topology becomes nearly degenerate and the maximum likelihood becomes a poor indicator of the goodness of fit of a model. In \citep{Brewer2012} a similar sampler was implemented and shown to work on mock data using a single energy bin. The sampling was performed directly on the prior with a hard likelihood threshold that depends on a series of levels whose prior mass decreases successively. Known as the Diffusive Nested Sampling (DNS) \citep{Brewer2009}, this method can sample multimodal distributions with high likelihood contrast. Furthermore, as a by-product, it provides an estimator for the Bayesian evidence. In this work, however, we sample from the posterior, since we do not need the Bayesian evidence for the point source metamodel. Instead, the model choice is based on the relative frequency of visiting different models via transdimensional jumps. In addition, the likelihood topology of the problem, although being highly multi-modal, has shallow islands since most point sources are faint. This argument excludes the well-localized bright point sources. Efficient exploration of these bright members would be impossible by Metropolis-Hastings updates alone. However we employ splits and merges for this reason in order to facilitate chain mixing in the crowded field limit. Similarly, \texttt{HELP: XID+} \citep{Hurley2017} is another Bayesian point source extraction framework that can fold in prior information on the sources. Although both algorithms are tailored for the crowded limit with a rich covariance structure, \texttt{HELP: XID+} differs from \texttt{PCAT} in that it relies on the prior knowledge of a certain number of known source positions in order to infer flux and flux uncertainties of sources consistent with the observed data. \section{Conclusion} \label{sect:conc} In this paper, we implemented a transdimensional sampler to infer point sources in a given photon count map by sampling from the probability distribution of the underlying catalog space. Our approach allows a consistent Bayesian exploration of the background normalization and the PSF, where they are co-sampled along with the model point source positions, fluxes and spectral parameters. The output is an ensemble of catalogs consistent with the count data, which represents the probability distribution of the catalog given the image. Compared to the traditional maximum likelihood solution to point source inference, probabilistic cataloging is computationally expensive. However this is a price paid for superior control over covariances in the catalog space, which becomes critical when a subsequent analysis using the inferred catalog as an input tries to reach a conclusion. Instead of using the maximum likelihood catalog with Gaussian errors around this solution, probabilistic cataloging provides a robust treatment of uncertainties. As a case study we apply our technique first to mock data and then to gamma-ray data in the NGPC. We associate the resulting probabilistic catalog with the 3FGL, and measure the flux distribution function down to $\sim$ 1-sigma sources. \acknowledgments We thank David W. Hogg, Brendon Brewer, Benjamin Lee, Zachary Slepian, Greg Green, Albert Lee, Can G\"{o}kler and Emre Erge\c{c}en for useful discussions during the course of the project. \newpage
ac91c1ed2afc0d0bef05a4fbab6b16f61d9756c9
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} A basic problem in Riemannian geometry is that of describing curvatures on a given manifold. Suppose that $(\Sigma,g)$ is a 2-dimensional compact Riemannian manifold without boundary, and $K$ is the Gaussian curvature on it. Let $\widetilde{g}=e^{2u}g$ be a metric conformal to $g$, where $u\in C^\infty(\Sigma)$. To find a smooth function $\widetilde{K}$ as the Gaussian curvature of $(\Sigma, \widetilde{g})$, one is led to solving the nonlinear elliptic equation \be\label{K-Z}\Delta_g u=K-\widetilde {K}e^{2u},\ee where $\Delta_g$ denotes the Laplacian operator on $(\Sigma,g)$. Let $v$ be a solution to $\Delta_g v=K-\overline{K}$, where $$\overline{K}=\f{1}{{\rm vol}_g(\Sigma)}\int_\Sigma Kdv_g.$$ Set $\psi=2(u-v)$. Then $\psi$ satisfies $$\Delta \psi=2\overline{K}-(2\widetilde{K}e^{2v})e^\psi.$$ If one frees this equation from the geometric situation, then it is a special case of \be\label{equation}\Delta_g u=c-he^u,\ee where $c$ is a constant, and $h$ is some prescribed function, with neither $c$ nor $h$ depends on geometry of $(\Sigma,g)$. Clearly one can consider (\ref{equation}) in any dimensional manifold. Now let $(\Sigma,g)$ be a compact Riemannian manifold of any dimension. Note that the solvability of (\ref{equation}) depends on the sign of $c$. Let us summarize results of Kazdan-Warner \cite{Kazdan-Warner}. For this purpose, think of $(\Sigma,g)$ and $h\in C^\infty(\Sigma)$ as being fixed with ${\rm dim}\, \Sigma\geq 1$.\\ {\it Case 1. $c<0$.} A necessary condition for a solution is that $\overline{h}<0$, in which case there is a critical strictly negative constant $c_ -(h)$ such that (\ref{equation}) is solvable if $c_-(h)<c<0$, but not solvable if $c<c_-(h)$. {\it Case 2. $c=0$.} When ${\rm dim}\, \Sigma\leq 2$, the equation (\ref{equation}) has a solution if and only if both $\overline{h}<0$ and $h$ is positive somewhere. When ${\rm dim}\, \Sigma\geq 3$, the necessary condition still holds. {\it Case 3. $c>0$.} When ${\rm dim}\, \Sigma=1$, so that $\Sigma=S^1$, then (\ref{equation}) has a solution if and only if $h$ is positive somewhere. When ${\rm dim}\, \Sigma=2$, there is a constant $0<c_+(h)\leq +\infty$ such that (\ref{equation}) has a solution if $h$ is positive somewhere and if $0<c<c_+(h)$.\\ There are tremendous work concerning the Kazdan-Warner problem, among those we refer the reader to Chen-Li \cite{Chen-Li1,Chen-Li2}, Ding-Jost-Li-Wang \cite{DJLW1,DJLW2}, and the references therein. \\ In this paper, we consider the Kazdan-Warner equation on a finite graph. In our setting, we shall prove the following: In {\it Case 1}, we have the same conclusion as the manifold case; In {\it Case 2}, the equation (\ref{equation}) has a solution if and only if both $\overline{h}<0$ and $h$ is positive somewhere; While in {\it Case 3}, the equation (\ref{equation}) has a solution if and only if $h$ is positive somewhere. Following the lines of Kazdan-Warner \cite{Kazdan-Warner}, for results of {\it Case 2} and {\it Case 3}, we use the variational method; for results of {\it Case 1}, we use the principle of upper-lower solutions. It is remarkable that Sobolev spaces on a finite graph are all pre-compact. This leads to a very strong conclusion in {\it Case 3} compared with the manifold case.\\ We organized this paper as follows: In Section \ref{sec1}, we introduce some notations on graphs and state our main results. In Section \ref{sec2}, we give two important lemmas, namely, the Sobolev embedding and the Trudinger-Moser embedding. In Sections \ref{sec3}-\ref{sec5}, we prove Theorems \ref{Theorem 1}-\ref{Theorem 4} respectively. In Section \ref{sec6}, we discuss related equations involving higher order derivatives. \section{Settings and main results}\label{sec1} Let $G=(V,E)$ be a finite graph, where $V$ denotes the vertex set and $E$ denotes the edge set. For any edge $xy\in E$, we assume that its weight $w_{xy}>0$ and that $w_{xy}=w_{yx}$. Let $\mu:V\ra \mathbb{R}^+$ be a finite measure. For any function $u:V\ra \mathbb{R}$, the $\mu$-Laplacian (or Laplacian for short) of $u$ is defined by \be\label{lap}\Delta u(x)=\f{1}{\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x)),\ee where $y\sim x$ means $xy\in E$. The associated gradient form reads \be\label{grad-form}\Gamma(u,v)(x)=\f{1}{2\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x))(v(y)-v(x)).\ee Write $\Gamma(u)=\Gamma(u,u)$. We denote the length of its gradient by \be\label{grd}|\nabla u|(x)=\sqrt{\Gamma(u)(x)}=\le(\f{1}{2\mu(x)}\sum_{y\sim x}w_{xy}(u(y)-u(x))^2\ri)^{1/2}.\ee For any function $g:V\ra\mathbb{R}$, an integral of $g$ over $V$ is defined by \be\label{int}\int_V gd\mu=\sum_{x\in V}\mu(x)g(x),\ee and an integral average of $g$ is denoted by $$\overline{g}=\f{1}{{\rm{Vol}(V)}}\int_Vgd\mu=\f{1}{{\rm{Vol}(V)}}\sum_{x\in V}\mu(x)g(x),$$ where ${\rm Vol}(V)=\sum_{x\in V}\mu(x)$ stands for the volume of $V$. The Kazdan-Warner equation on graph reads \be\label{KZ-equ}\Delta u=c-he^u\quad{\rm in}\quad V,\ee where $\Delta$ is defined as in (\ref{lap}), $c\in\mathbb{R}$, and $h:V\ra\mathbb{R}$ is a function. If $c=0$, then (\ref{KZ-equ}) is reduced to \be\label{0equ}\Delta u=-he^u\quad{\rm in}\quad V.\ee Our first result can be stated as following: \begin{theorem}\label{Theorem 1} Let $G=(V,E)$ be a finite graph, and $h (\not\equiv 0)$ be a function on $V$. Then the equation (\ref{0equ}) has a solution if and only if $h$ changes sign and $\int_Vhd\mu<0$. \end{theorem} In cases $c>0$ and $c<0$, we have the following: \begin{theorem}\label{Theorem 2} Let $G=(V,E)$ be a finite graph, $c$ be a positive constant, and $h:V\ra\mathbb{R}$ be a function. Then the equation (\ref{KZ-equ}) has a solution if and only if $h$ is positive somewhere. \end{theorem} \begin{theorem}\label{Theorem 3} Let $G=(V,E)$ be a finite graph, $c$ be a negative constant, and $h:V\ra\mathbb{R}$ be a function.\\ $(i)$ If (\ref{KZ-equ}) has a solution, then $\overline{h}<0$.\\ $(ii)$ If $\overline{h}<0$, then there exists a constant $-\infty\leq c_-(h)<0$ depending on $h$ such that (\ref{KZ-equ}) has a solution for any $c_-(h)<c<0$, but has no solution for any $c<c_-(h)$. \end{theorem} Concerning the constant $c_-(h)$ in Theorem \ref{Theorem 3}, we have the following: \begin{theorem}\label{Theorem 4} Let $G=(V,E)$ be a finite graph, $c$ be a negative constant, and $h:V\ra\mathbb{R}$ be a function. Suppose that $c_-(h)$ is given as in Theorem \ref{Theorem 3}. If $h(x)\leq 0$ for all $x\in V$, but $h\not\equiv 0$, then $c_-(h)=-\infty$. \end{theorem} \section{Preliminaries}\label{sec2} Define a Sobolev space and a norm on it by $$W^{1,2}(V)=\le\{u:V\ra\mathbb{R}: \int_V(|\nabla u|^2+u^2)d\mu<+\infty\ri\},$$ and $$\|u\|_{W^{1,2}(V)}=\le(\int_V(|\nabla u|^2+u^2)d\mu\ri)^{1/2}$$ respectively. If $V$ is a finite graph, then $W^{1,2}(V)$ is exactly the set of all functions on $V$, a finite dimensional linear space. This implies the following Sobolev embedding: \begin{lemma}\label{Sobolv}Let $G=(V,E)$ be a finite graph. The Sobolev space $W^{1,2}(V)$ is pre-compact. Namely, if $\{u_j\}$ is bounded in $W^{1,2}(V)$, then there exists some $u\in W^{1,2}(V)$ such that up to a subsequence, $u_j\ra u$ in $W^{1,2}(V)$. \end{lemma} Also we have the following Trudinger-Moser embedding: \begin{lemma}\label{T-M-lemma} Let $G=(V,E)$ be a finite graph. For any $\beta>1$, there exists a constant $C$ depending only on $\beta$ and $V$ such that for all functions $v$ with $\int_V|\nabla v|^2d\mu\leq 1$ and $\int_Vvd\mu=0$, there holds $$\int_Ve^{\beta v^2}d\mu\leq C.$$ \end{lemma} {\it Proof.} Let $\beta>1$ be fixed. For any function $v$ satisfying $\int_V|\nabla v|^2d\mu\leq 1$ and $\int_Vvd\mu=0$, we have by the Poincare inequality $$\int_Vv^2d\mu\leq C_0\int_V|\nabla v|^2d\mu\leq C_0,$$ where $C_0$ is some constant depending only on $V$. Denote $\mu_{\min}=\min_{x\in V}\mu(x)$. In view of (\ref{int}), the above inequality leads to $\|v\|_{L^\infty(V)}\leq C_0/\mu_{\min}$. Hence $$\int_Ve^{\beta v^2}d\mu\leq e^{\beta C_0^2/\mu_{\min}}{\rm Vol}(V).$$ This gives the desired result. $\hfill\Box$ \section{The case $c=0$ }\label{sec3} In the case $c=0$, our approach comes out from that of Kazdan-Warner \cite{Kazdan-Warner}. {\it Proof of Theorem \ref{Theorem 1}.} {\it Necessary condition.} If (\ref{0equ}) has a solution $u$, then $e^{-u}\Delta u=-h$. Integration by parts gives \bna -\int_Vhd\mu&=&\int_Ve^{-u}\Delta ud\mu\\ &=&-\int_V\Gamma(e^{-u},u)d\mu\\ &=&-\f{1}{2}\sum_{x\in V}\sum_{y\sim x}w_{xy}(e^{-u(y)}-e^{-u(x)})(u(y)-u(x))\\ &>&0, \ena since $(e^{-u(y)}-e^{-u(x)})(u(y)-u(x))\leq 0$ for all $x,y\in V$ and $u$ is not a constant. {\it Sufficient condition.} We use the calculus of variations. Suppose that $h$ changes sign and \be\label{mean-h}\int_Vhd\mu<0.\ee Define a set \be\label{B1}\mathcal{B}_1=\le\{v\in W^{1,2}(V): \int_Vhe^vd\mu=0,\,\,\int_Vvd\mu=0\ri\}.\ee We {\it claim} that \be\label{B-non}\mathcal{B}_1\not=\varnothing.\ee To see this, since $h$ changes sign and (\ref{mean-h}), we can assume $h(x_1)>0$ for some $x_1\in V$. Take a function $v_1$ satisfying $v_1(x_1)=\ell$ and $v_1(x)=0$ for all $x\not= x_1$. Hence \bna \int_Vhe^{v_1}d\mu&=&\sum_{x\in V}\mu(x)h(x)e^{v_1(x)}\\ &=&\mu(x_1) h(x_1)e^\ell+\sum_{x\not=x_1}\mu(x)h(x)\\ &=&(e^\ell-1)\mu(x_1) h(x_1)+\int_Vhd\mu\\ &>&0 \ena for sufficiently large $\ell$. Writing $\phi(t)=\int_Vhe^{tv_1}d\mu$, we have by the above inequality that $\phi(1)>0$. Obviously $\phi(0)=\int_Vhd\mu<0$. Thus there exists a constant $0<t_0<1$ such that $\phi(t_0)=0$. Let $v^\ast=t_0v_1-\f{1}{{\rm vol}(V)}\int_Vt_0v_1d\mu$, where ${\rm vol}(V)=\sum_{x\in V}\mu(x)$ stands for the volume of $V$. Then $v^\ast\in \mathcal{B}_1$. This concludes our claim (\ref{B-non}). We shall minimize the functional $J(v)=\int_V|\nabla v|^2d\mu$. Let $$a=\inf_{v\in \mathcal{B}_1}J(v).$$ Take a sequence of functions $\{v_n\}\subset\mathcal{B}_1$ such that $J(v_n)\ra a$. Clearly $\int_V|\nabla v_n|^2d\mu$ is bounded and $\int_V{v}_nd\mu=0$. Hence $v_n$ is bounded in $W^{1,2}(V)$. Since $V$ is a finite graph, the Sobolev embedding (Lemma \ref{Sobolv}) implies that up to a subsequence, $v_n\ra v_\infty$ in $W^{1,2}(V)$. Hence $\int_V{v_\infty}d\mu=0$, $\int_Vhe^{v_\infty}d\mu=\lim_{n\ra\infty}\int_Vhe^{v_n}d\mu=0$, and thus $v_\infty\in\mathcal{B}_1$. Moreover $$\int_V|\nabla v_\infty|^2d\mu=\lim_{n\ra\infty}\int_V|\nabla v_n|^2d\mu=a.$$ One can calculate the Euler-Lagrange equation of $v_\infty$ as follows: \be\label{E-L1}\Delta v_\infty=-\f{\lambda}{2}he^{v_\infty}-\f{\gamma}{2},\ee where $\lambda$ and $\gamma$ are two constants. Indeed, for any $\phi\in W^{1,2}(V)$, there holds \bea\nonumber 0&=&\le.\f{d}{dt}\ri|_{t=0}\le\{\int_V|\nabla(v_\infty+t\phi)|^2d\mu-\lambda\int_Vhe^{v_\infty+t\phi}d\mu- \gamma\int_V(v_\infty+t\phi)d\mu\ri\}\\\nonumber &=&2\int_V\Gamma(v_\infty,\phi)d\mu-\lambda\int_Vhe^{v_\infty}\phi d\mu-\gamma\int_V\phi d\mu\\ \label{euler}&=&-2\int_V(\Delta v_\infty)\phi d\mu-\lambda\int_Vhe^{v_\infty}\phi d\mu-\gamma\int_V\phi d\mu, \eea which gives (\ref{E-L1}) immediately. Integrating the equation (\ref{E-L1}), we have $\gamma=0$. We {\it claim} that $\lambda\not=0$. For otherwise, we conclude from $\Delta v_\infty=0$ and $\int_Vv_\infty d\mu=0$ that $v_\infty\equiv 0\not\in \mathcal{B}_1$. This is a contradiction. We further {\it claim} that $\lambda>0$. This is true because $\int_Vhd\mu<0$ and $$0<\int_V e^{-v_\infty}\Delta v_\infty d\mu=-\f{\lambda}{2}\int_Vhd\mu.$$ Thus we can write $\f{\lambda}{2}=e^{-\vartheta}$ for some constant $\vartheta$. Then $u=v_\infty+\vartheta$ is a desired solution of (\ref{0equ}). $\hfill\Box$ \section{The case $c>0$}\label{sec4} {\it Proof of Theorem \ref{Theorem 2}.} {\it Necessary condition.} Suppose $c>0$ and $u$ is a solution to (\ref{KZ-equ}). Since $\int_V\Delta ud\mu=0$, we have $$\int_Vhe^ud\mu=c{\rm Vol}(V)>0.$$ Hence $h$ must be positive somewhere on $V$.\\ {\it Sufficient condition.} Suppose $h(x_0)>0$ for some $x_0\in V$. Define a set $$\mathcal{B}_2=\le\{v\in W^{1,2}(V): \int_Vhe^vd\mu=c{\rm Vol}(V)\ri\}.$$ We {\it claim} that $\mathcal{B}_2\not=\varnothing$. To see this, we set $$u_\ell(x)=\le\{\begin{array}{lll} \ell,\quad x=x_0\\[1.2ex] 0,\quad x\not=x_0. \end{array}\ri.$$ It follows that $$\int_Vhe^{u_\ell}d\mu\ra+\infty\quad{\rm as}\quad \ell\ra+\infty.$$ We also set $\widetilde{u}_\ell\equiv -\ell$, which leads to $$\int_Vhe^{\widetilde{u}_\ell}d\mu=e^{-\ell}\int_Vhd\mu\ra 0\quad{\rm as}\quad \ell\ra+\infty.$$ Hence there exists a sufficiently large $\ell$ such that $\int_Vhe^{u_\ell}d\mu>c{\rm Vol}(V)$ and $\int_Vhe^{\widetilde{u}_\ell}d\mu<c{\rm Vol}(V)$. We define a function $\phi:\mathbb{R}\ra\mathbb{R}$ by $$\phi(t)=\int_Vhe^{tu_\ell+(1-t)\widetilde{u}_\ell}d\mu.$$ Then $\phi(0)<c{\rm Vol}(V)<\phi(1)$, and thus there exists a $t_0\in (0,1)$ such that $\phi(t_0)=c{\rm Vol}(V)$. Hence $\mathcal{B}_2\not=\varnothing$ and our claim follows. We shall solve (\ref{KZ-equ}) by minimizing the functional $$J(u)=\f{1}{2}\int_V|\nabla u|^2d\mu+c\int_Vud\mu$$ on $\mathcal{B}_2$. For this purpose, we write $u=v+\overline{u}$, so $\overline{v}=0$. Then for any $u\in\mathcal{B}_2$, we have $$\int_Vhe^vd\mu=c{\rm Vol}(V)e^{-\overline{u}}>0,$$ and thus \be\label{ju}J(u)=\f{1}{2}\int_V|\nabla u|^2d\mu-c{\rm Vol}(V)\log\int_Vhe^vd\mu+c{\rm Vol}(V)\log(c{\rm Vol}(V)).\ee Let $\widetilde{v}=v/\|\nabla v\|_2$. Then $\int_V\widetilde{v}d\mu=0$ and $\|\nabla \widetilde{v}\|_2=1$. By the Poincare inequality, $\|\widetilde{v}\|_2\leq C_0$ for some constant $C_0$ depending only on $V$. By Lemma \ref{T-M-lemma}, for any $\beta>1$, one can find a constant $C$ depending only on $\beta$ and $V$ such that \be\label{T-M}\int_Ve^{\beta {\widetilde{v}}^2}d\mu\leq C(\beta,V).\ee This together with an elementary inequality $ab\leq \epsilon a^2+\f{b^2}{4\epsilon}$ implies that for any $\epsilon>0$, \bna \int_Ve^vd\mu&\leq&\int_Ve^{\epsilon\|\nabla v\|_2^2+\f{v^2}{4\epsilon\|\nabla v\|_2^2}}d\mu\\ &=&e^{\epsilon\|\nabla v\|_2^2}\int_Ve^{\f{v^2}{4\epsilon\|\nabla v\|_2^2}}d\mu\\ &\leq& Ce^{\epsilon\|\nabla v\|_2^2}, \ena where $C$ is a positive constant depending only on $\epsilon$ and $V$. Hence $$\int_Vhe^vd\mu\leq C(\max_{x\in V}h(x))e^{\epsilon\|\nabla v\|_2^2}.$$ In view of (\ref{ju}), the above inequality leads to $$J(u)\geq \f{1}{2}\int_V|\nabla u|^2d\mu-c{\rm Vol}(V)\epsilon\|\nabla v\|_2^2-C_1,$$ where $C_1$ is some constant depending only on $\epsilon$ and $V$. Choosing $\epsilon=\f{1}{4c{\rm Vol}(V)}$, and noting that $\|\nabla v\|_2=\|\nabla u\|_2$, we obtain for all $u\in\mathcal{B}_2$, \be\label{lowerb}J(u)\geq \f{1}{4}\int_V|\nabla u|^2d\mu-C_1.\ee Therefore $J$ has a lower bound on the set $\mathcal{B}_2$. This permits us to consider $$b=\inf_{u\in\mathcal{B}_2}J(u).$$ Take a sequence of functions $\{u_k\}\subset\mathcal{B}_2$ such that $J(u_k)\ra b$. Let $u_k=v_k+\overline{u_k}$. Then $\overline{v_k}=0$, and it follows from (\ref{lowerb}) that $v_k$ is bounded in $W^{1,2}(V)$. This together with the equality $$\int_Vu_kd\mu=\f{1}{c}J(u_k)-\f{1}{2c}\int_V|\nabla v_k|^2d\mu$$ implies that $\{\overline{u_k}\}$ is a bounded sequence. Hence $\{u_k\}$ is also bounded in $W^{1,2}(V)$. By the Sobolev embedding (Lemma \ref{Sobolv}), up to a subsequence, $u_k\ra u$ in $W^{1,2}(V)$. It is easy to see that $u\in\mathcal{B}_2$ and $J(u)=b$. Using the same method of (\ref{euler}), we derive the Euler-Lagrange equation of the minimizer $u$, namely, $\Delta u=c-\lambda he^u$ for some constant $\lambda$. Noting that $\int_V\Delta ud\mu=0$, we have $\lambda=1$. Hence $u$ is a solution of the equation (\ref{KZ-equ}). $\hfill\Box$ \section{The case $c<0$}\label{sec5} In this section, we prove Theorem \ref{Theorem 3} by using a method of upper and lower solutions. In particular, we show that it suffices to construct an upper solution of the equation (\ref{KZ-equ}). This is exactly the graph version of the argument of Kazdan-Warner (\cite{Kazdan-Warner}, Sections $9$ and $10$).\\ We call a function $u_-$ a lower solution of (\ref{KZ-equ}) if for all $x\in V$, there holds $$\Delta u_-(x)-c+he^{u_-(x)}\geq 0.$$ Similarly, $u_+$ is called an upper solution of (\ref{KZ-equ}) if for all $x\in V$, it satisfies $$\Delta u_+(x)-c+he^{u_+(x)}\leq 0.$$ We begin with the following: \begin{lemma}\label{lemma9.3} Let $c<0$. If there exist lower and upper solutions, $u_-$ and $u_+$, of the equation (\ref{KZ-equ}) with $u_-\leq u_+$, then there exists a solution $u$ of (\ref{KZ-equ}) satisfying $u_-\leq u\leq u_+$. \end{lemma} {\it Proof.} We follow the lines of Kazdan-Warner (\cite{Kazdan-Warner}, Lemma 9.3). Set $k_1(x)=\max\{1,-h(x)\}$, so that $k_1\geq 1$ and $k_1\geq -h$. Let $k(x)=k_1(x)e^{u_+(x)}$. We define $L\varphi\equiv\Delta\varphi-k\varphi$ and $f(x,u)\equiv c-he^{u}$. Since $G=(V,E)$ is a finite graph and $\inf_{x\in V}k(x)>0$, we have that $L$ is a compact operator and ${\rm Ker}(L)=\{0\}$. Hence we can define inductively $u_{j+1}$ as the unique solution to \be\label{equ1}Lu_{j+1}=f(x,u_j)-ku_j,\ee where $u_0=u_+$. We {\it claim} that \be\label{monotone}u_-\leq u_{j+1}\leq u_j\leq\cdots\leq u_+.\ee To see this, we estimate $$L(u_1-u_0)=f(x,u_0)-ku_0-\Delta u_0+ku_0\geq 0.$$ Suppose $u_1(x_0)-u_0(x_0)=\max_{x\in V}(u_1(x)-u_0(x))>0$. Then $\Delta(u_1-u_0)(x_0)\leq 0$, and thus $L(u_1-u_0)(x_0)<0$. This is a contradiction. Hence $u_1\leq u_0$ on $V$. Suppose $u_j\leq u_{j-1}$, we calculate \bna L(u_{j+1}-u_j)&=&k(u_{j-1}-u_j)+h(e^{u_{j-1}}-e^{u_j})\\ &\geq& k_1(x)(e^{u_+(x)}-e^{\xi})(u_{j-1}-u_j)\\ &\geq&0, \ena where $u_j\leq\xi\leq u_{j-1}$. Similarly as above, we have $u_{j+1}\leq u_j$ on $V$, and by induction, $u_{j+1}\leq u_j\leq\cdots\leq u_+$ for any $j$. Noting that $$L(u_--u_{j+1})\geq k(u_j-u_-)+h(e^{u_j}-e^{u_-}),$$ we also have by induction $u_-\leq u_j$ on $V$ for all $j$. Therefore (\ref{monotone}) holds. Since $V$ is finite, it is easy to see that up to a subsequence, $u_j\ra u$ uniformly on $V$. Passing to the limit $j\ra+\infty$ in the equation (\ref{equ1}), one concludes that $u$ is a solution of (\ref{KZ-equ}) with $u_-\leq u\leq u_+$. $\hfill\Box$\\ Next we show that the equation (\ref{KZ-equ}) has infinite lower solutions. This reduces the proof of Theorem \ref{Theorem 3} to finding its upper solution. \begin{lemma}\label{lowersolution} There exists a lower solution $u_-$ of (\ref{KZ-equ}) with $c<0$. Thus (\ref{KZ-equ}) has a solution if and only if there exists an upper solution. \end{lemma} {\it Proof.} Let $u_-\equiv-A$ for some constant $A>0$. Since $V$ is finite, we have $$\Delta u_-(x)-c+h(x)e^{u_-(x)}=-c+h(x)e^{-A}\ra -c\quad{\rm as}\quad A\ra+\infty,$$ uniformly with respect to $x\in V$. Noting that $c<0$, we can find sufficiently large $A$ such that $u_-$ is a lower solution of (\ref{KZ-equ}). $\hfill\Box$\\ {\it Proof of Theorem \ref{Theorem 3}}. $(i)$ {\it Necessary condition.} If $u$ is a solution of (\ref{KZ-equ}), then \bna -\int_Vhd\mu&=&\int_Ve^{-u}\Delta ud\mu-c\int_Ve^{-u}d\mu\\ &=&-\int_V\Gamma(e^{-u},u)d\mu-c\int_Ve^{-u}d\mu\\ &>&0. \ena $(ii)$ {\it Sufficient condition.} It follows from Lemmas \ref{lemma9.3} and \ref{lowersolution} that (\ref{KZ-equ}) has a solution if and only if (\ref{KZ-equ}) has an upper solution $u_+$ satisfying $$\Delta u_+\leq c-he^{u_+}.$$ Clearly, if $u_+$ is an upper solution for a given $c<0$, then $u_+$ is also an upper solution for all $\widetilde{c}<0$ with $c\leq \widetilde{c}$. Therefore, there exists a constant $c_-(h)$ with $-\infty\leq c_-(h)\leq 0$ such that (\ref{KZ-equ}) has a solution for any $c>c_-(h)$ but has no solution for any $c<c_-(h)$. We {\it claim} that $c_-(h)<0$ under the assumption $\int_Vhd\mu<0$. To see this, we let $v$ be a solution of $\Delta v=\overline{h}-h$. There exists some constant $a>0$ such that $$|e^{av}-1|\leq \f{-\overline{h}}{2\max_{x\in V} |h(x)|}.$$ Let $e^b=a$. If $c=\f{a\overline{h}}{2}$ and $u_+=av+b$, we have \bna \Delta u_+-c+he^{u_+}&=&ah(e^{av}-1)+\f{a\overline{h}}{2}\\ &\leq&a(\max_{x\in V}|h(x)|)|e^{av}-1|+\f{a\overline{h}}{2}\\ &\leq&\f{a\overline{h}}{2}-\f{a\overline{h}}{2}\\ &=&0. \ena Thus if $c={a\overline{h}}/{2}<0$, then the equation (\ref{KZ-equ}) has an upper solution $u_+$. Therefore, $\overline{h}<0$ implies that $c_-(h)\leq {a\overline{h}}/{2}<0$. $\hfill\Box$\\ {\it Proof of Theorem \ref{Theorem 4}.} We shall show that if $h(x)\leq 0$ for all $x\in V$, but $h\not\equiv 0$, then (\ref{KZ-equ}) is solvable for all $c<0$. For this purpose, we let $v$ be a solution of $\Delta v=\overline{h}-h$. Note that $\overline{h}<0$. Pick constants $a$ and $b$ such that $a\overline{h}<c$ and $e^{av+b}-a>0$. Let $u_+=av+b$. Since ${h}\leq 0$, \bna \Delta u_+-c+he^{u_+}&=&a\Delta v-c+he^{av+b}\\ &=&a\overline{h}-ah-c+he^{av+b}\\ &\leq&h(e^{av+b}-a)\\ &\leq& 0. \ena Hence $u_+$ is an upper solution. Consequently, $c_-(h)=-\infty$ if $h\leq 0$ but $h\not\equiv 0$. $\hfill\Box$ \section{Some extensions}\label{sec6} The equation (\ref{equation}) involving higher order differential operators was also extensively studied on manifolds, see for examples \cite{Djadli-Malchiodi,LLL} and the references therein. In this section, we shall extend Theorems \ref{Theorem 1}-\ref{Theorem 4} to nonlinear elliptic equations involving higher order derivatives. For this purpose, we define the length of $m$-order gradient of $u$ by \be\label{high-deriv}|\nabla^mu|=\le\{\begin{array}{lll} |\nabla\Delta^{\f{m-1}{2}}u|,\,\,{\rm when}\,\,\, m \,\,{\rm is\,\,odd}\\[1.5ex] |\Delta^{\f{m}{2}}u|,\,\,{\rm when}\,\,\, m \,\,{\rm is\,\,even}, \end{array}\ri.\ee where $|\nabla\Delta^{\f{m-1}{2}}u|$ is defined as in (\ref{grd}) for the function $\Delta^{\f{m-1}{2}}u$, and $|\Delta^{\f{m}{2}}u|$ denotes the usual absolute of the function $\Delta^{\f{m}{2}}u$. Define a Sobolev space by $$W^{m,2}(V)=\le\{v:V\ra\mathbb{R}: \int_V(|v|^2+|\nabla^mv|^2)d\mu<+\infty\ri\}$$ and a norm on it by $$\|v\|_{W^{m,2}(V)}=\le(\int_V(|v|^2+|\nabla^mv|^2)d\mu\ri)^{1/2}.$$ Clearly $W^{m,2}(V)$ is the set of all functions on $V$ since $V$ is finite. Moreover, we have the following Sobolev embedding and the Trudinger-Moser embedding: \begin{lemma}\label{m-Sob} Let $G=(V,E)$ be a finite graph. Then for any integer $m>0$, $W^{m,2}(V)$ is pre-compact. \end{lemma} \begin{lemma}\label{m-T-M-lemma} Let $G=(V,E)$ be a finite graph. Let $m$ be a positive integer. Then for any $\beta>1$, there exists a constant $C$ depending only on $m$, $\beta$ and $V$ such that for all functions $v$ with $\int_V|\nabla^m v|^2d\mu\leq 1$ and $\int_Vvd\mu=0$, there holds $$\int_Ve^{\beta v^2}d\mu\leq C.$$ \end{lemma} We consider an analog of (\ref{KZ-equ}), namely \be\label{m-equ}\Delta^m u=c-he^u\quad{\rm in}\quad V,\ee where $m$ is a positive integer, $c$ is a constant, and $h: V\ra\mathbb{R}$ is a function. Obviously (\ref{m-equ}) is reduced to (\ref{KZ-equ}) when $m=1$. Firstly we have the following: \begin{theorem}\label{Theorem 5} Let $G=(V,E)$ be a finite graph, $h(\not\equiv 0)$ be a function on $V$, and $m$ be a positive integer. If $c=0$, $h$ changes sign, and $\int_Vhd\mu<0$, then the equation (\ref{m-equ}) has a solution. \end{theorem} {\it Proof.} We give the outline of the proof. Denote $$\mathcal{B}_3=\le\{v\in W^{m,2}(V):\, \int_Vhe^vd\mu=0,\,\int_Vvd\mu=0\ri\}.$$ In view of (\ref{B1}), we have that $\mathcal{B}_3=\mathcal{B}_1$, since $V$ is finite. Hence $\mathcal{B}_3\not=\varnothing$. Now we minimize the functional $J(v)=\int_V|\nabla^mu|^2d\mu$ on $\mathcal{B}_3$. The remaining part is completely analogous to that of the proof of Theorem \ref{Theorem 1}, except for replacing Lemma \ref{Sobolv} by Lemma \ref{m-Sob}. We omit the details but leave it to interested readers. $\hfill\Box$\\ Secondly, in the case $c>0$, the same conclusion as Theorem \ref{Theorem 2} still holds for the equation (\ref{m-equ}) with $m>1$. Precisely we have the following: \begin{theorem}\label{Theorem 6} Let $G=(V,E)$ be a finite graph, $c$ be a positive constant, $h:V\ra\mathbb{R}$ be a function, and $m$ be a positive integer. Then the equation (\ref{m-equ}) has a solution if and only if $h$ is positive somewhere. \end{theorem} {\it Proof.} Repeating the arguments of the proof of Theorem \ref{Theorem 2} except for replacing Lemmas \ref{Sobolv} and \ref{T-M-lemma} by Lemmas \ref{m-Sob} and $\ref{m-T-M-lemma}$ respectively, we get the desired result. $\hfill\Box$\\ Finally, concerning the case $c<0$, we obtain a result weaker than Theorem \ref{Theorem 3}. \begin{theorem}\label{Theorem 7} Let $G=(V,E)$ be a finite graph, $c$ be a negative constant, $m$ is a positive integer, and $h:V\ra\mathbb{R}$ be a function such that $h(x)<0$ for all $x\in V$. Then the equation (\ref{m-equ}) has a solution. \end{theorem} {\it Proof.} Since the maximum principle can not be available for equations involving poly-harmonic operators, we use the calculus of variations instead of the method of upper and lower solutions. Let $c<0$ be fixed. Consider the functional \be\label{funct}J(u)=\f{1}{2}\int_V|\nabla^mu|^2d\mu+c\int_Vud\mu.\ee Set $$\mathcal{B}_4=\le\{u\in W^{m,2}(V): \int_Vhe^ud\mu=c{\rm Vol}(V)\ri\}.$$ Using the same method of proving (\ref{B-non}) in the proof of Theorem \ref{Theorem 2}, we have $\mathcal{B}_4\not=\varnothing$. We now prove that $J$ has a lower bound on $\mathcal{B}_4$. Let $u\in \mathcal{B}_4$. Write $u=v+\overline{u}$. Then $\overline{v}=0$ and $$\int_Vhe^{v}d\mu=e^{-\overline{u}}c{\rm Vol}(V),$$ which leads to $$\overline{u}=-\log\le(\f{1}{c{\rm Vol}(V)}\int_Vhe^vd\mu\ri).$$ Hence \be\label{lo-b}J(u)=\f{1}{2}\int_V|\nabla^mu|^2d\mu-c{\rm Vol}(V)\log\le(\f{1}{c{\rm Vol}(V)}\int_Vhe^vd\mu\ri).\ee Since $c<0$ and $h(x)<0$ for all $x\in V$, we have $\max_{x\in V}h(x)<0$, and thus \be\label{dl}\f{h}{c{\rm Vol}(V)}\geq \delta=\f{\max_{x\in V}h(x)}{c{\rm Vol}(V)}>0.\ee Inserting (\ref{dl}) into (\ref{lo-b}), we have \be\label{l-3}J(u)\geq \f{1}{2}\int_V|\nabla^mu|^2d\mu-c{\rm Vol}(V)\log\delta-c{\rm Vol}(V)\log\int_Ve^vd\mu.\ee By the Jensen inequality, \be\label{jens}\f{1}{{\rm Vol}(V)}\int_Ve^vd\mu\geq e^{\overline{v}}=1.\ee Inserting (\ref{jens}) into (\ref{l-3}), we obtain \be\label{lll}J(u)\geq \f{1}{2}\int_V|\nabla^mu|^2d\mu-c{\rm Vol}(V)\log\delta-c{\rm Vol}(V)\log{\rm Vol}(V).\ee Therefore $J$ has a lower bound on $\mathcal{B}_4$. Set $$\tau=\inf_{v\in \mathcal{B}_4}J(v).$$ Take a sequence of functions $\{u_k\}\subset\mathcal{B}_4$ such that $J(u_k)\ra \tau$. We have by (\ref{lll}) that \be\label{gra-b}\int_V|\nabla^mu_k|^2d\mu\leq C\ee for some constant $C$ depending only on $c$, $\tau$, $V$ and $h$. By (\ref{funct}), we estimate \be\label{gr-2}\le|\int_Vu_kd\mu\ri|\leq \f{1}{|c|}|J(u_k)|+\f{1}{2c}\int_V|\nabla^mu_k|^2d\mu.\ee The Poincare inequality implies that there exists some constant $C$ depending only on $m$ and $V$ such that \be\label{poinc}\int_V|u_k-\overline{u_k}|^2d\mu\leq C\int_V|\nabla^mu_k|^2d\mu\ee Combining (\ref{gra-b}), (\ref{gr-2}), and (\ref{poinc}), one can see that $\{u_k\}$ is bounded in $W^{m,2}(V)$. Then it follows from Lemma \ref{m-Sob} that there exists some function $u$ such that up to a subsequence, $u_k\ra u$ in $W^{m,2}(V)$. Clearly $u\in\mathcal{B}_4$ and $J(u)=\lim_{k\ra\infty}J(u_k)=\tau$. In other words, $u$ is a minimizer of $J$ on the set $\mathcal{B}_4$. It is not difficult to check that (\ref{m-equ}) is the Euler-Lagrange equation of $u$. This completes the proof of the theorem. $\hfill\Box$\\ {\bf Acknowledgements.} A. Grigor'yan is partly supported by SFB 701 of the German Research Council. Y. Lin is supported by the National Science Foundation of China (Grant No.11271011). Y. Yang is supported by the National Science Foundation of China (Grant No.11171347). \bigskip
ccaf2bdbdea134ed38d947f8a0b77e33783f4a48
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} It is well known that a gas composed of bosonic atoms with repulsive interparticle interaction at appropriate values of density and temperature undergoes Bose-Einstein condensation (BEC), a phase transition which shares similarities to transitions to superfluid and superconductive states. Since the first experimental demonstration of BEC \cite{Anderson-science269,Davis-prl75,Bradley-prl75}, efforts have directed toward in investigating the thermodynamic properties of such a macroscopic quantum system and finding suitable theoretical descriptions of the phase transitions \cite{Shi-physrep304}. Recently there is a revival of experimental interest devoted to the study of the thermodynamics of quantum gases. On one hand, distinguished works have explored the thermodynamics: a Fermi gas with repulsive interactions \cite{Lee-pra85}, a Fermi gas in the limit of very strong interactions, i.e., near the unitary regime \cite{Nascimbene-nature463,Ku-Science335}, a Fermi gas in a three-dimensional optical lattice showing fermionic Mott-insulator transition \cite{Duarte-prl114}, and the Boson gas in a two-dimensional optical lattice showing a bosonic Mott-Insulator transition \cite{Gemelke-nature460}. On the other hand, works on weakly interacting bosonic gases have demonstrated that, even in this simpler system, the understanding and characterization of the thermodynamic behavior, especially across the phase transition, are not yet complete \cite{Hung-nature470, Donner-science315, Olivares-Quiroz-jphysbatomphys43} and that more experimental works is needed to validate the theoretical predictions \cite{Goswami-jlowtempphys172, Floerchinger-pra79, Stanley-revmodphys71, Tarasov-pra90}. New approaches to investigating these systems and new experimental results can therefore contribute, in general, to advance the understanding of the thermodynamics of quantum gases and, in particular, of their phase transitions. In this work, we experimentally determine a global susceptibility from a global thermodynamical variables approach for a harmonically trapped Bose gas \cite{Romero-Rochin-prl94,Romero-Rochin-bjp35,Sandoval-Figueroa-pre78}. We investigate and characterize the behavior of the susceptibility when the gas undergoes a BEC. In standard thermodynamics, the equivalent quantity of the global susceptibility that we define in this work is the isothermal compressibility. This parameter describes the relative variation of the volume $V$ of a system due to a change in the pressure $P$ at constant temperature $T$: $k_T=-\frac{1}{V}\left(\frac{\partial V}{\partial P}\right)_{N,T}$. It is a property associated with density fluctuations and it can also be expressed in terms of a second derivative of the free energy with respect to the pressure. At a second-order phase transition it is therefore expected to show a singularity. Here we provide experimental evidence of such a singular behavior by taking advantage of the global thermodynamic approach. \section{Thermodynamics based on global variables} Global variables have already been successfully employed to obtain the phase diagram \cite{Romero-Rochin-pra85} and measure the heat capacity \cite{Shiozaki-pra90} of a gas in a harmonic potential. The need to review standard thermodynamics when dealing with quantum gases comes naturally from the fact that they are usually trapped in nonhomogeneous (normally harmonic) potentials. In this situation standard definitions of pressure and volume do not apply. In fact, $P$ and $V$ are conjugate variables of thermodynamical systems defined for homogeneous densities. In particular, $P$ is an intensive variable having the same value in every position inside the volume occupied by the gas. The local density approximation (LDA) is often used in non-homogeneous situations to define local variables. A different approach, involving a set of thermodynamic variables with single values for the entire gas, allows a global description of the thermodynamics of an inhomogeneous gas and of its phase transitions. This global approach is particularly suited, compared to the LDA, for the case in which the gas is characterized by abrupt spatial variations of the density, as in the occurrence of a phase transition or in a more exotic situation such as in the presence of vortices or local potential impurities. The use of global variables to describe the thermodynamics of an inhomogeneous system has been extensively described elsewhere \cite{Romero-Rochin-prl94,Romero-Rochin-bjp35,Sandoval-Figueroa-pre78}. In brief, within the basis of thermodynamic and statistical mechanics one can infer a volume parameter and a pressure parameter respectively: \begin{equation} {\cal V}=\frac{1}{\omega_x\omega_y\omega_z}, \label{eq:V} \end{equation} \begin{equation} \Pi=\frac{2}{3{\cal V}}\langle U({\bf r})\rangle=\frac{m}{3{\cal V}}\int d^3r~n({\bf r})(\omega^2_xx^2+\omega^2_yy^2+\omega^2_zz^2),\label{eq:Pi} \end{equation} where $\omega_i$ with $\left(i=x,y,z\right)$ are the harmonic trap frequencies, $ \langle U({\bf r})\rangle $ is the spatial mean of the external potential, and $n({\bf r})$ is the density of the sample. ${\cal V}$ is a natural extensive ``volume'' for the trapped gas and the thermodynamic limit can be achieved by making the density parameter $n_{\cal V}=N/{\cal V}$ constant as $N$ and ${\cal V}$ grow indefinitely. $\Pi$ is its intensive conjugated variable $(\Pi=-\left(\frac{\partial F}{\partial{\cal V}}\right)_{N,T})$, where $F=F(N,{\cal V} ,T)$ is the Helmholtz free energy. A nice proof that $\Pi = \Pi\left( N,{\cal V},T \right)$ and ${\cal V}$ are a good set of variables to describe the system is obtained through the determination of the heat capacity, $C_{\cal V}$ \cite{Shiozaki-pra90}, whose behavior is close that one expected from treatment of a harmonic trapped Bose gas \cite{Grossmann-physletta208, Giorgini-jlowtempphys109}. In this framework, the isothermal compressibility parameter can be obtained from the following relation: \begin{equation} \kappa_T=-\frac{1}{{\cal V}}\left(\frac{\partial{\cal V}}{\partial\Pi}\right)_{N,T}. \end{equation} $\kappa_T$ is a quantity with the same properties of the standard compressibility $k_T$ \cite{Yukalov-pra72, Yukalov-laserphyslett2} and indicates the thermodynamic stability defined by the second derivative of Gibbs free energy. The convexity property of the free energy is maintained with the condition, $0 \leq \kappa_T < \infty$. Therefore, with this susceptibility we characterize a system in thermodynamic equilibrium \cite{Yukalov-laserphys23}. \section{Experimental system and measurement} We performed the measurements to determine $\kappa_T$ across the transition from a thermal cloud to a BEC of $^{87}{\rm Rb}$ atoms with a new experimental setup in which the volume parameter can be easily varied. The system is built in a standard double magneto-optical trap (MOT) configuration \cite{Myatt-oplett21}. In the first vacuum cell we load a MOT of $10^8$ atoms from a dispenser and then we transfer the atoms to the second cell using an on-resonance beam. Here, we recapture the atoms in a second MOT and, after performing a sub-Doppler cooling, we spin polarize the atomic sample in the hyperfine state $F=2,\, m_F=2$. Afterwards, we transfer the atoms at temperatures of about $40~\mu{\rm K}$, in a pure quadrupole magnetic trap where a first radio-frequency evaporation is performed. Simultaneously, we ramp on a far-detuned beam (with wavelength, $\lambda=1064~{\rm nm}$) focused on a waist $w_0=85~\mu{\rm m}$, dislocated by $z_0=300~\mu{\rm m}$ along the gravity direction below the center of the quadrupole trap. When the temperature of the atomic cloud decreases to approximately $10~\mu {\rm K}$, atoms migrate from the quadrupole trap to the center of the beam, which serves as an optical dipole trap (ODT). At that point we reduce the vertical magnetic-field gradient to a value that no longer compensates for the gravity. The atoms are thus confined in a hybrid trap given by the combination of the optical and magnetic confinements \cite{Lin-pra79}. Here we further decrease the temperature of the cloud by a second stage of radio-frequency evaporation followed by optical evaporation obtained by exponentially ramping down the power of the laser beam. We can eventually achieve a pure BEC of $\sim10^5$ atoms at typical temperatures of $100--200~{\rm nK}$. The hybrid potential, including gravity can be described by the following expression: \begin{eqnarray} U({\bf r})=\mu B'_x\sqrt{x^2+\frac{y^2}{2}+\frac{z^2}{2}}-\frac{U_0}{(1+y^2/y^2_R)}\nonumber\\ \exp\left[-\frac{2x^2+2(z-z_0)^2}{w_0^2(1+y^2/y^2_R)}\right]+mg(z-z_0)+E_0 \end{eqnarray} $\mu$ is the atomic magnetic moment, $B'_x$ is the gradient of the quadrupole trap along the $x$ direction, $y_R = w^2_0\pi/\lambda$ is the Rayleigh range of the beam which propagates along direction $y$ and $U_0$ is the optical trap depth. $g$ is the gravitational acceleration, $m$ is the atomic mass and $E_0$ is the energy difference between the zero-field point absent the dipole trap and the total trap minimum, giving the trap minimum $U({\bf r}_{min})=0$ \cite{Lin-pra79}. At low temperatures the effective potential of the HT can be safely approximated by a three dimensional harmonic potential, whose frequencies are \begin{equation} \omega_x\simeq\omega_z=\sqrt{\frac{4U_0}{m w_0^2}},~~\omega_y=\sqrt{\frac{\mu B'_x}{2 m \left| z_0 \right|}}.\label{eq:freq} \end{equation} The trap has a cylindrical symmetry where the radial frequency confinement is due the ODT and the axial weaker confinement is due to magnetic-field gradient. We characterize the atomic cloud by using absorption imaging after a free expansion from the trapping potential with a time of flight of $30~{\rm ms}$. Each image is fitted to a two-dimensional bimodal distribution composed of a Gaussian function and a Thomas-Fermi function, which are known to properly describe the thermal and the condensed component of the gas, respectively. The number of particles and temperature are obtained from the fitted images following conventional procedure. The volume parameter can be easily changed by varying the radial frequencies of the hybrid confinement, which directly depend on the final laser power of the ODT. We consider measurements for seven different sets of frequencies, i.e., for seven different volume parameters. Different temperatures have been obtained by changing the radio-frequency evaporation ramp; in this way the initial conditions for the optical evaporation change, allow us to achieve different final temperatures with the same trapping frequencies since the final power the ODT is the same. For each volume parameter we have performed many experimental runs for temperatures within the range $40--400~{\rm nK}$ and postselected atomic clouds containing $(1\pm 0.1)\times10^5$ atoms to be taken in consideration. In order to calculate the pressure parameter $\Pi$ by performing the integral in Eq. (\ref{eq:Pi}), it is necessary to reconstruct the density profile of the atoms in the trap, $n(\mathbf{r})$, from the measured profiles in the time of flight and the trap frequencies. Toward this aim, for the thermal component we can safely assume a free expansion, whereas for the interacting condensed component we apply the Castin-Dum procedure \cite{Castin-prl77,Kagan-pra55}. In Fig. \ref{PixT} we plot the calculated $\Pi(T)$ for each volume parameter. With a decrease in the temperature the atomic gas undergoes BEC: at high temperatures we observe a linear dependence of the pressure parameter on $T$ until an abrupt change takes place at a critical temperature $T_c$, and the decrease become faster than linear. Above $T_c$, experimental data are well reproduced by the the ideal gas law $\Pi {\cal V} = N k_B T$, plotted in Fig. \ref{PixT} for the known number of particles and the different volumes. Below $T_c$ we perform a proper empirical exponential fit which follows the behavior of the experimental points. These fitting functions, in principle, are not related to any theoretical model. For each volume parameter we can extract the critical pressure for condensation: lower volumes demand higher pressure to condense. The transition line from a thermal atomic cloud to a BEC in the $\Pi{\cal V}$-plane is shown in Fig. \ref{logVxPi} marking the separation between the white (thermal) and the gray (BEC) zone. \begin{figure}[t!] \includegraphics[width=1.0\columnwidth]{Fig1} \caption{(Color online) Pressure parameter vs temperature for a constant number of atoms ($N=1\times10^5$) and different volume parameters: ${\cal V}_{1}=1.9 \times 10^{-7} {\rm s}^3$, ${\cal V}_{2}=6.4\times 10^{-8}~{\rm s}^3$, ${\cal V}_{3}=3.2\times 10^{-8}~{\rm s}^3$, ${\cal V}_{4}=2.1\times 10^{-8}~{\rm s}^3$, ${\cal V}_{5}=1.75\times 10^{-8}~{\rm s}^3$, ${\cal V}_{6}=1.4\times 10^{-8}~{\rm s}^3$, and ${\cal V}_{7}=1\times 10^{-8}~{\rm s}^3$. Solid lines above $T_c$ represent the ideal gas law, whereas below $T_c$ are empirical exponential fits. The dotted black line marks the transition between the thermal and the condensed regimes. Error bars represent the statistical error on the average.} \label{PixT} \end{figure} \section{Isotherms and determination of compressibility parameter} From the measurements shown in Fig. \ref{PixT}, we extract different isotherms relating the volume and pressure parameters, ${\cal V}={\cal V}_T(\Pi)$, which we plot in Fig. \ref{logVxPi}. As the temperature decreases, the overall isothermal lines shift towards a lower pressure. We can clearly identify two different behaviors in the two different regions of the thermal and condensed regimes. In the thermal region, experimental points are well reproduced by the ideal gas law for the known number of atoms and temperatures (plotted as lines on the log-log scale of the figure). When an isotherm crosses the critical line for condensation an abrupt change occurs and it departs from the ideal gas behavior. \begin{figure}[t!] \includegraphics[width=1.0\columnwidth]{Fig2} \caption{(Color online) Isothermal ${\cal V}$ vs $\Pi$. Symbols represent the measured volume parameter vs the pressure parameter for different temperatures. The diagram shows the classical phase in white and the quantum phase in gray, which are separated by the critical line on the ${\cal V} \Pi$ plane. In the thermal region the data obey the ideal gas and its behavior is linear in the log-log scale. On the other hand, in the BEC region the data exhibit a nonlinear behavior, indeed, we implement empirical fitting curves, known in the literature as the extended Langmuir adsorption isotherm equation, to follow each isotherm (these curves are no used in the analysis). The error bars on the ${\cal V}$ axis come from error propagation of the measurement of the frequencies; on the other hand, the error bars on the $\Pi$ axis are associated with the exponential fit in Fig. \ref{PixT}.} \label{logVxPi} \end{figure} We can now extract the isothermal compressibility $\kappa_T$ from derivation of the isotherms in Fig. \ref{logVxPi}. Derivation is performed point by point in correspondence with the experimental data in order not to rely on the arbitrary fitting curves, which do not correspond to any theoretical model. The obtained $\kappa_T$ values for three isothermal curves are shown in Fig. \ref{kTxPi}. We have chosen the curves for $T=150~{\rm nK}$ [Fig. \ref{kTxPi}(a))], $T=80~{\rm nK}$ [Fig. \ref{kTxPi}(b)] and $T=40~{\rm nK}$ [Fig. \ref{kTxPi}(c)] because they demonstrate the three classes of behavior: pure thermal gas, gas undergoing BEC transition, and gas in the single BEC region, respectively. The isothermal curve at $150~{\rm nK}$ shows the decrease in $\kappa_T$ with $1/\Pi$, as expected for an ideal gas. Let us now consider the isotherm at $80~{\rm nK}$: at low pressures the gas is thermal and the compressibility $\kappa_T$ decreases with increasing $\Pi$; when the pressure reaches the region between $20$ and $30~(\times 10^{-19}{\rm J}\cdot{\rm s}^{-3})$, the sudden increase in $\kappa_T$ indicates the transition. The compressibility reaches a maximum value before returning close to the base-line after $40 \times 10^{-19}{\rm J}\cdot{\rm s}^{-3}$. In this pressure range the compressibility acquires values 4 to 8 times higher than the base-line. The behavior of $\kappa_T$ in Fig. \ref{kTxPi} is typical for a second-order phase transition. An investigation of $\kappa_T$ vs $\Pi$ for different isothermal curves, where the transition takes place, reveals that at higher temperatures the transition occurs at a higher pressure and the peak of compressibility is broader for higher temperatures. Contrary to the expectation that quantities involving integration of density over the potential \cite{Ku-Science335} would be weakly sensitive to the phase transition, our data shows a sudden large variation in the compressibility at the thermal-BEC transition. \begin{figure}[h!] \includegraphics[width=1.0\columnwidth]{Fig3} \caption{Isothermal compressibility parameter vs pressure parameter for three temperatures: (a) $T=150~{\rm nK}$, (b) $T=80~{\rm nK}$, and (c) $T=40~{\rm nK}$. The inset in (b) is the $\kappa_T$ calculated from a simple toy model for the density distribution. Lines are guides for the eye. The error bar on the abscissa is not included so as to pollute the behavior of $\kappa_T$. On the other hand, the error bar on the ordinate comes from the extrapolation of the tangent isothermal curve in Fig. \ref{logVxPi}}. \label{kTxPi} \end{figure} \section{Discussion} We performed the data analysis using Castin-Dum procedure to reconstruct the \textit{in situ} density distribution starting with a Thomas-Fermi fit of the condensed component in the time-of-flight images. In order to probe that the general results we found do not depend on the specific model for the analysis, we also tested an alternative, less constrained, model. We fitted our images with two Gaussians for the thermal and condensed components and we reconstructed the \textit{in situ} profiles by applying a variational method \cite{Perez-Garcia-pra56, Teles-pra87, Teles-pra88} which has already proved to be valid to study the ballistic expansion dynamics of a condensate \cite{Teles-pra87, Teles-pra88}. We checked that the $\Pi(T)$ curves, and therefore all the derived thermodynamic quantities, extracted with the two reconstructing methods are quantitatively comparable. A complete theory predicting the exact behavior of the compressibility parameter across the transition does not exist. Nevertheless, the need to make a prediction about the behavior and the shape of the compressibility around the critical point arises naturally. We have therefore attempted a comparison between our findings and the results of a toy model. We calculate $\Pi$ for synthetic density profiles consisting in a Gaussian thermal component and a Thomas-Fermi condensed one with a relative atom number given by the ideal BEC result. This model qualitatively catches the general experimental findings. In particular, the position and the shape of the compressibility peak are reproduced by the model as presented in the inset in Fig. \ref{kTxPi}. Nevertheless this simple model cannot give quantitative predictions, for example, of the absolute value of the compressibility, because it is over-simplified. A fair quantitative comparison would demand a more elaborate model, beyond the scope of this experimental report. The introduction of the global variable approach has proven to be a valid complementary approach to the LDA. Generally speaking, the LDA approach in fact in fact has strong intrinsic limitations in the case where sudden variation of the densities occurs, as at the thermal-condensed interface in a Bose gas. In this situation the LDA would in fact require a very high imaging resolution, which is experimentally challenging. With the global approach we overcome this limitation by describing the system undergoing phase transition as a whole and we can provide evidence of the compressibility peak at the transition. On the other hand, the global variables approach needs many measurements for different volumes with the same atom number to trace a single isothermal curve, and this can be experimentally nontrivial. In this sense the LDA has the advantage of leading to a complete isothermal curve from the analysis of a single image. Due to the lack of experimental points, we cannot precisely measure the compressibility in the close vicinity of the phase transition. Nevertheless, the expected sharp peak in $\kappa_T$ near the critical point is quite clear and shares remarkable similarities to the behavior of the isothermal compressibility for liquid helium as observed across the $\lambda$-point \cite{Boghosian-pr152,Grilly-pr149,Elwell-pr164}. \section{Conclusions} In this article, we have used the concept of global thermodynamic variables to measure, the most appropriate susceptibility to understand the phase transition, the isothermal compressibility parameter of a harmonically confined Bose gas. Once the sample had undergone BEC we characterized this phase transition, from the classical to the quantum regime, indicating a second-order transition likely related to a spontaneous symmetry breaking. The concept of using global variables to determine the global compressibility is quite useful in situations where the LDA cannot be applied. In another more complex physical systems in which there are abrupt changes in the density are of interest for superfluid physics, such as vortices, vortex lattices, solitons, \textit{inter alia}, and, especially, superfluid turbulence, recently demonstrated by our group \cite{Henn-prl103}. In this case the local variables do not make sense and the global behavior in the compressibility may indicate new characteristics of the turbulent regime. Such an investigation is currently in progress. We acknowledge financial support from FAPESP (Brazil), CNPq (Brazil), CAPES (Brazil), and LENS (Italy).
7b579c13dd17bdcca3252d58b00171cbb934decd
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Cell-Free massive multiple-input multiple-output (MIMO) refers to a massive MIMO system \cite{MarzettaNonCooperative} where the base station antennas are geographically distributed \cite{NgoCF,MarzettaCF,Truong}. These antennas, called access points (APs) herein, simultaneously serve many users in the same frequency band. The distinction between cell-free massive MIMO and conventional distributed MIMO \cite{ZhouWCS} is the number of antennas involved in coherently serving a given user. In canonical cell-free massive MIMO, every antenna serves every user. Compared to co-located massive MIMO, cell-free massive MIMO has the potential to improve coverage and energy efficiency, due to increased macro-diversity gain. By operating in time-division duplex (TDD) mode, cell-free massive MIMO exploits the channel reciprocity property, according to which the channel responses are the same in both uplink and downlink. Reciprocity calibration, to the required accuracy, can be achieved in practice using off-the-shelf methods \cite{Lund}. Channel reciprocity allows the APs to acquire channel state information (CSI) from pilot sequences transmitted by the users in the uplink, and this CSI is then automatically valid also for the downlink. By virtue of the law of large numbers, the effective scalar channel gain seen by each user is close to a deterministic constant. This is called \textit{channel hardening}. Thanks to the channel hardening, the users can reliably decode the downlink data using only statistical CSI. This is the reason for why most previous studies on massive MIMO assumed that the users do not acquire CSI and that there are no pilots in the downlink \cite{MarzettaNonCooperative,DebbahULDL,BjornsonHowMany}. In co-located massive MIMO, transmission of downlink pilots and the associated channel estimation by the users yields rather modest performance improvements, owing to the high degree of channel hardening \cite{NgoDlPilots,Khansefid,Zuo}. In contrast, in cell-free massive MIMO, the large number of APs is distributed over a wide area, and many APs are very far from a given user; hence, each user is effectively served by a smaller number of APs. As a result, the channel hardening is less pronounced than in co-located massive MIMO, and potentially the gain from using downlink pilots is larger. \textbf{Contributions:} We propose a downlink training scheme for cell-free massive MIMO, and provide an (approximate) achievable downlink rate for conjugate beamforming processing, valid for finite numbers of APs and users, which takes channel estimation errors and power control into account. This rate expression facilitates a performance comparison between cell-free massive MIMO with downlink pilots, and cell-free massive MIMO without downlink pilots, where only statistical CSI is exploited by the users. The study is restricted to the case of mutually orthogonal pilots, leaving the general case with pilot reuse for future work. \textit{Notation:} Column vectors are denoted by boldface letters. The superscripts $()^*$, $()^T$, and $()^H$ stand for the conjugate, transpose, and conjugate-transpose, respectively. The Euclidean norm and the expectation operators are denoted by $\Vert\cdot\Vert$ and $\mathbb{E}\{\cdot\}$, respectively. Finally, we use $z\sim\mathcal{CN}(0,\sigma^2)$ to denote a circularly symmetric complex Gaussian random variable (RV) $z$ with zero mean and variance $\sigma^2$, and use $z\sim\mathcal{N}(0,\sigma^2)$ to denote a real-valued Gaussian RV. \section{System Model and Notation} Let us consider $M$ single-antenna APs\footnote{We are considering the conjugate beamforming scheme which is implemented in a distributed manner, and hence, an $N$-antenna APs can be treated as $N$ single-antenna APs.}, randomly spread out in a large area without boundaries, which simultaneously serve $K$ single-antenna users, $M>K$, by operating in TDD mode. All APs cooperate via a backhaul network exchanging information with a central processing unit (CPU). Only payload data and power control coefficients are exchanged. Each AP locally acquires CSI and precodes data signals without sharing CSI with the other APs. The time-frequency resources are divided into coherence intervals of length $\tau$ symbols (which are equal to the coherence time times the coherence bandwidth). The channel is assumed to be static within a coherence interval, and it varies independently between every coherence interval. Let $g_{mk}$ denote the channel coefficient between the $k$th user and the $m$th AP, defined as \begin{equation} \label{eq:channelmodel} g_{mk} = \sqrt{\beta_{mk}}h_{mk}, \end{equation} where $h_{mk}$ is the small-scale fading, and $\beta_{mk}$ represents the large-scale fading. Since the APs are not co-located, the large-scale fading coefficients $\{\beta_{mk}\}$ depend on both $m$ and $k$. We assume that $h_{mk}$, $m=1,..., M$, $k=1,..., K$, are i.i.d.\ $\mathcal{CN}(0,1)$ RVs, i.e. Rayleigh fading. Furthermore, $\beta_{mk}$ is constant with respect to frequency and is known, a-priori, whenever required. Lastly, we consider moderate and low user mobility, thus viewing $\{\beta_{mk}\}$ coefficients as constants. The TDD coherence interval is divided into four phases: uplink training, uplink payload data transmission, downlink training, and downlink payload data transmission. In the uplink training phase, users send pilot sequences to the APs and each AP estimates the channels to all users. The channel estimates are used by the APs to perform the uplink signal detection, and to beamform pilots and data during the downlink training and the downlink data transmission phase, respectively. Here, we focus on the the downlink performance. The analysis on the uplink payload data transmission phase is omitted, since it does not affect on the downlink performance. \subsection{Uplink Training} Let $\tau_\textrm{u,p}$ be the uplink training duration per coherence interval such that $\tau_\textrm{u,p}<\tau$. Let $\sqrt{\tau_\textrm{u,p}}\bm{\varphi}_k \in \mathbb{C}^{\tau_\textrm{u,p}\times1}$, be the pilot sequence of length $\tau_\textrm{u,p}$ samples sent by the $k$th user, $k=1,...,K$. We assume that users transmit pilot sequences with full power, and all the uplink pilot sequences are mutually orthonormal, i.e., $\bm{\varphi}_k^H\bm{\varphi}_{k'} = 0$ for $k' \neq k$, and $\Vert\bm{\varphi}_k\Vert^2=1$. This requires that $\tau_\textrm{u,p} \geq K$, i.e., $\tau_\textrm{u,p} = K$ is the smallest number of samples required to generate $K$ orthogonal vectors. The $m$th AP receives a $ \tau_\textrm{u,p}\times1 $ vector of $K$ uplink pilots linearly combined as \begin{equation} \label{eq:receiveduplinkpilot} \textbf{y}_{\textrm{up},m} = \sqrt{\tau_\textrm{u,p}\rho_\textrm{u,p}}\sum^K_{k=1} g_{mk}\bm{\varphi}_k + \textbf{w}_{\textrm{up},m}, \end{equation} where $\rho_\textrm{u,p}$ is the normalized transmit signal-to-noise ratio (SNR) related to the pilot symbol and $\textbf{w}_{\textrm{up},m}$ is the additive noise vector, whose elements are i.i.d. $\mathcal{CN}(0,1)$ RVs. The $m$th AP processes the received pilot signal as follows \begin{equation} \label{eq:receiveduplinkpilotprojection} \check{y}_{\textrm{up},mk} = \bm{\varphi}^H_k\textbf{y}_{\textrm{up},m} = \sqrt{\tau_\textrm{u,p}\rho_\textrm{u,p}} \ g_{mk}+ \bm{\varphi}^H_k\textbf{w}_{\textrm{up},m}, \end{equation} and estimates the channel $g_{mk}$, $k=1,...,K$ by performing MMSE estimation of $g_{mk}$ given $\check{y}_{\textrm{up},mk}$, which is given by \begin{equation} \label{eq:APchannelestimation} \hat{g}_{mk} = \frac{\mathbb{E}\{\check{y}^*_{\textrm{up},mk}g_{mk}\}}{\mathbb{E}\{\vert\check{y}_{\textrm{up},mk}\vert^2\}}\check{y}_{\textrm{up},mk} = c_{mk}\check{y}_{\textrm{up},mk}, \end{equation} where \begin{equation} \label{eq:cmk} c_{mk} \triangleq \frac{\sqrt{\tau_\textrm{u,p}\rho_\textrm{u,p}}\beta_{mk}}{\tau_\textrm{u,p}\rho_\textrm{u,p}\beta_{mk}+1}. \end{equation} The corresponding channel estimation error is denoted by $\tilde{g}_{mk} \triangleq g_{mk} - \hat{g}_{mk}$ which is independent of $\hat{g}_{mk}$. \subsection{Downlink Payload Data Transmission} During the downlink data transmission phase, the APs exploit the estimated CSI to precode the signals to be transmitted to the $K$ users. Assuming conjugate beamforming, the transmitted signal from the $m$th AP is given by \begin{equation} \label{eq:APtransmittedsignal} x_m = \sqrt{\rho_\textrm{d}}\sum^K_{k=1} \sqrt{\eta_{mk}} \ \hat{g}^*_{mk} q_k, \end{equation} where $q_k$ is the data symbol intended for the $k$th user, which satisfies $\mathbb{E}\{\vert q_k \vert^2\}=1$, and $\rho_\textrm{d}$ is the normalized transmit SNR related to the data symbol. Lastly, $\eta_{mk}$, $m=1,...,M$, $k=1,...,K$, are power control coefficients chosen to satisfy the following average power constraint at each AP: \begin{equation} \label{eq:pwConstraint} \mathbb{E}\{|x_m|^2\}\leq\rho_\textrm{d}. \end{equation} Substituting (\ref{eq:APtransmittedsignal}) into (\ref{eq:pwConstraint}), the power constraint above can be rewritten as \begin{equation} \label{eq:pwConstraintGamma} \sum \limits_{k=1}^K \eta_{mk} \gamma_{mk} \leq 1, \ \text{for all}\ m, \end{equation} where \begin{equation} \label{eq:defGamma} \gamma_{mk} \triangleq \mathbb{E}\{{|\hat{g}_{mk}|}^2\} = \sqrt{\tau_\textrm{u,p}\rho_\textrm{u,p}} \beta_{mk} c_{mk} \end{equation} represents the variance of the channel estimate. The $k$th user receives a linear combination of the data signals transmitted by all the APs. It is given by \begin{align} \label{eq:receiveddownlinksignal} r_{\textrm{d},k} &= \sum^M_{m=1} g_{mk} x_m + w_{\textrm{d},k} = \sqrt{\rho_\textrm{d}} \sum^K_{k'=1} a_{kk'} q_{k'} + w_{\textrm{d},k}, \end{align} where \begin{align}\label{eq:akkdef} a_{kk'} \triangleq \sum^M_{m=1} \sqrt{\eta_{mk'}} {g}_{mk} \hat{g}^*_{mk'}, ~ k'=1,...,K, \end{align} and $ w_{\textrm{d},k} $ is additive $\mathcal{CN}(0,1)$ noise at the $k$th user. In order to reliably detect the data symbol $q_k$, the $k$th user must have a sufficient knowledge of the effective channel gain, $a_{kk}$. \begin{figure*}[!t] \normalsize \setcounter{MYtempeqncnt}{\value{equation}} \setcounter{equation}{\ccounter} \begin{equation} \label{eq:genericDLrate} R_k = \mathbb{E}\left\{\log_2\left(1+\frac{\rho_\textrm{d} \left| \mathbb{E}\left\{a_{kk} \mathrel{\big|} \hat{a}_{kk} \right\} \right|^2}{\rho_\textrm{d} \sum\limits^K_{k'=1} \mathbb{E}\left\{{|a_{kk'}|^2 \mathrel{\big|} \hat{a}_{kk}}\right\}-\rho_\textrm{d} \left| \mathbb{E}\left\{a_{kk} \mathrel{\big|} \hat{a}_{kk} \right\} \right|^2 +1} \right)\right\}. \end{equation} \setcounter{equation}{\value{MYtempeqncnt}} \hrulefill \vspace*{4pt} \end{figure*} \subsection{Downlink Training} While the model given so far is identical to that in \cite{NgoCF}, we now depart from that by the introduction of downlink pilots. Specifically, we adopt the Beamforming Training scheme proposed in \cite{NgoDlPilots}, where pilots are beamformed to the users. This scheme is scalable in that it does not require any information exchange among APs, and its channel estimation overhead is independent of $M$. Let $\tau_\textrm{d,p}$ be the length (in symbols) of the downlink training duration per coherence interval such that $\tau_\textrm{d,p}<\tau - \tau_\textrm{u,p}$. The $m$th AP precodes the pilot sequences $\bm{\psi}_{k'} \in \mathbb{C}^{\tau_\textrm{d,p}\times1}$, $k'=1,...,K$, by using the channel estimates $\{\hat{g}_{mk'}\}$, and beamforms it to all the users. The $\tau_\textrm{d,p} \times 1$ pilot vector $\bm{x}_{m,\textrm{p}}$ transmitted from the $m$th AP is given by \begin{equation} \label{eq:DLpilot} \bm{x}_{m,\textrm{p}} = \sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}}\sum^K_{k'=1} \sqrt{\eta_{mk'}} \hat{g}^*_{mk'} \bm\psi_{k'}, \end{equation} where $\rho_\textrm{d,p}$ is the normalized transmit SNR per downlink pilot symbol, and $\{\bm\psi_{k}\}$ are mutually orthonormal, i.e. $\bm{\psi}^H_k\bm{\psi}_{k'} = 0$, for $k' \neq k$, and $\Vert\bm{\psi}_k\Vert^2=1$. This requires that $\tau_\textrm{d,p} \geq K$. The $k$th user receives a corresponding $ \tau_\textrm{d,p}\times 1 $ pilot vector: \begin{equation} \label{eq:receivedDLpilot} \textbf{y}_{\textrm{dp},k} = \sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \sum^K_{k'=1} a_{kk'} \bm\psi_{k'} + \textbf{w}_{\textrm{dp},k}, \end{equation} where $\textbf{w}_{\textrm{dp},k}$ is a vector of additive noise at the $k$th user, whose elements are i.i.d. $\mathcal{CN}(0,1)$ RVs. In order to estimate the effective channel gain $a_{kk}$, $k=1,...,K$, the $k$th user first processes the received pilot as \begin{align} \label{eq:receivedDLpilotprojection} \check{y}_{\textrm{dp},k} &= \bm{\psi}^H_{k} \textbf{y}_{\textrm{dp},k} = \sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \ a_{kk} + \bm\psi^H_{k} \textbf{w}_{\textrm{dp},k} \nonumber \\ &= \sqrt{\tau_\textrm{d,p} \rho_\textrm{d,p}} \ a_{kk} + n_{\textrm{p},k}, \end{align} where $n_{\textrm{p},k} \triangleq \bm{\psi}^H_k\textbf{w}_{\textrm{dp},k} \sim \mathcal{CN}(0,1)$, and then performs linear MMSE estimation of $a_{kk}$ given $\check{y}_{\textrm{dp},k}$, which is, according to \cite{SMKay}, equal to \begin{align} \label{eq:a_kk} \hat{a}_{kk} &= \mathbb{E}\{a_{kk}\} + \nonumber \\ &+ \frac{\sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \ \mathrm{Var}\{a_{kk}\}}{\tau_\textrm{d,p} \rho_\textrm{d,p} \mathrm{Var}\{a_{kk}\} + 1}(\check{y}_{\textrm{dp},k} - \sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \ \mathbb{E}\{a_{kk}\}). \end{align} \textit{Proposition 1:} With conjugate beamforming, the linear MMSE estimate of the effective channel gain formed by the $k$th user, see (\ref{eq:a_kk}), is \begin{align} \label{eq:a_kk2} \hat{a}_{kk} = \frac{\sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}}\ \varsigma_{kk} \ \check{y}_{\textrm{dp},k} + \sum \limits_{m=1}^M \sqrt{\eta_{mk}} \ \gamma_{mk}}{\tau_\textrm{d,p}\rho_\textrm{d,p}\ \varsigma_{kk} + 1}, \end{align} where $\varsigma_{kk} \triangleq \sum_{m=1}^M \eta_{mk} \beta_{mk} \gamma_{mk}$. \begin{IEEEproof} See Appendix A. \end{IEEEproof} \begin{figure*}[!t] \normalsize \setcounter{MYtempeqncnt}{\value{equation}} \setcounter{equation}{23} \begin{align} \label{eq:rateComparison} R_k = \begin{cases} \log_2 \left( 1 + \frac{\rho_\textrm{d}\left(\sum\limits^M_{m=1} \sqrt{\eta_{mk}}\gamma_{mk}\right)^2}{\rho_\textrm{d} \sum\limits^K_{k'=1} \sum\limits^M_{m=1} \eta_{mk'} \beta_{mk} \gamma_{mk'} + 1}\right)&\text{for statistical CSI,} \\ (\ref{eq:DLrateApprox2})&\text{for Beamforming Training,}\\ \mathbb{E} \left\{ \log_2 \left( 1 + \frac{\rho_\textrm{d}|a_{kk}|^2}{\rho_\textrm{d} \sum\limits^K_{k' \neq k} |a_{kk'}|^2 + 1} \right)\right\}&\text{for perfect CSI.} \end{cases} \end{align} \setcounter{equation}{\value{MYtempeqncnt}} \hrulefill \vspace*{4pt} \end{figure*} \section{Achievable Downlink Rate} In this section we derive an achievable downlink rate for conjugate beamforming precoding, using downlink pilots via Beamforming Training. An achievable downlink rate for the $k$th user is obtained by evaluating the mutual information between the observed signal $r_{\textrm{d},k}$ given by (\ref{eq:receiveddownlinksignal}), the known channel estimate $\hat{a}_{kk}$ given by (\ref{eq:a_kk2}) and the unknown transmitted signal $q_k$: $I(q_k;r_{\textrm{d},k},\hat{a}_{kk})$, for a permissible choice of input signal distribution. Letting $\tilde{a}_{kk}$ be the channel estimation error, the effective channel gain $a_{kk}$ can be decomposed as \begin{equation} \label{eq:ChEstErr} a_{kk} = \hat{a}_{kk}+\tilde{a}_{kk}. \end{equation} Note that, since we use the linear MMSE estimation, the estimate $\hat{a}_{kk}$ and the estimation error $\tilde{a}_{kk}$ are uncorrelated, but not independent. The received signal at the $k$th user described in (\ref{eq:receiveddownlinksignal}) can be rewritten as \begin{align} r_{\textrm{d},k} = \sqrt{\rho_\textrm{d}}\ {a}_{kk} q_{k} + \tilde{w}_{\textrm{d},k}, \end{align} where $ \tilde{w}_{\textrm{d},k} \triangleq \sqrt{\rho_\textrm{d}}\ \sum^K_{k' \neq k} a_{kk'} q_{k'} + w_{\textrm{d},k} $ is the effective noise, which satisfies $\mathbb{E}\left\{\tilde{w}_{\textrm{d},k} \mathrel{\big|} \hat{a}_{kk} \right\}=\mathbb{E}\left\{q_k^\ast\tilde{w}_{\textrm{d},k} \mathrel{\big|} \hat{a}_{kk} \right\}=\mathbb{E}\left\{a_{kk}^\ast q_k^\ast\tilde{w}_{\textrm{d},k} \mathrel{\big|} \hat{a}_{kk} \right\} = 0$. Therefore, following a similar methodology as in \cite{Medard}, we obtain an achievable downlink rate of the transmission from the APs to the $k$th user, which is given by (\ref{eq:genericDLrate}) at the top of the page. The expression given in (\ref{eq:genericDLrate}) can be simplified by making the approximation that $a_{kk'}$, $k'=1,...,K$, are Gaussian RVs. Indeed, according to the Cram\'{e}r central limit theorem\footnote{Cram\'{e}r central limit theorem: Let $X_1, X_2, ..., X_n$ are independent circularly symmetric complex RVs. Assume that $X_i$ has zero mean and variance $\sigma^2_i$. If $s^2_n = \sum^n_{i=1} \sigma^2_i \rightarrow \infty$ and $\sigma_i/s_n \rightarrow 0$, as $n\rightarrow \infty$, then $\frac{\sum^n_{i=1} X_i}{s_n} \xrightarrow{d} \mathcal{CN}(0,1), \ \text{as } n \rightarrow \infty$.}, we have \setcounter{equation}{19} \begin{align} \label{eq:approxAk1} &a_{kk'} = \sum^M_{m=1} \sqrt{\eta_{mk'}} \ {g}_{mk} \hat{g}^*_{mk'} \xrightarrow{d} \mathcal{CN}\left(0, \varsigma_{kk'} \right), \text{ as } M \rightarrow \infty, \\ \label{eq:approxAkk} &a_{kk} = \sum^M_{m=1} \sqrt{\eta_{mk}} |\hat{g}_{mk}|^2 + \sum^M_{m=1} \sqrt{\eta_{mk}} \tilde{g}_{mk} \hat{g}^*_{mk} \nonumber \\ &\approx \sum^M_{m=1} \sqrt{\eta_{mk}} |\hat{g}_{mk}|^2 \xrightarrow{d} \mathcal{N}\left(\sum^M_{m=1} \sqrt{\eta_{mk}} \gamma_{mk},\sum^M_{m=1} \eta_{mk} \gamma_{mk}^2 \right), \nonumber \\ &\text{as } M \rightarrow \infty, \end{align} where $\varsigma_{kk'} \triangleq \sum_{m=1}^M \eta_{mk'} \beta_{mk} \gamma_{mk'}$, and $\xrightarrow{d}$ denotes convergence in distribution. The Gaussian approximations (\ref{eq:approxAk1}) and (\ref{eq:approxAkk}) can be verified by numerical results, as shown in Figure \ref{pdfplot}. The pdfs show a close match between the empirical and the Gaussian distribution even for small $M$. Furthermore, with high probability the imaginary part of $a_{kk}$ is much smaller than the real part so it can be reasonably neglected. \begin{figure}[!t] \centering \includegraphics[width=3.3in]{pdfplotlog} \caption{The approximate (Gaussian) and the true (empirical) pdfs of $a_{kk}$ and $a_{kk'}$ for a given $\beta_{mk}$ realization (the large-scale fading model is discussed in detail in Section \ref{NumericalResults}). Here, $M = 20$ and $K = 5$.} \label{pdfplot} \end{figure} Under the assumption that $a_{kk}$ is Gaussian distributed, $\hat{a}_{kk}$ in (\ref{eq:a_kk2}) becomes the MMSE estimate of $a_{kk}$. As a consequence, $\hat{a}_{kk}$ and $\tilde{a}_{kk}$ are independent. In addition, by following a similar methodology as in \eqref{eq:approxAk1} and \eqref{eq:approxAkk}, we can show that any linear combination of $a_{kk}$ and $a_{kk'}$ are asymptotically (for large $M$) Gaussian distributed, and hence $a_{kk}$ and $a_{kk'}$ are asymptotically jointly Gaussian distributed. Furthermore, $a_{kk}$ and $a_{kk'}$ are uncorrelated so they are independent. Hence, the achievable downlink rate (\ref{eq:genericDLrate}) is reduced to\footnote{ A formula similar to \eqref{eq:DLrateApprox} but for co-located massive MIMO systems, was given in \cite{NgoDlPilots,Khansefid} with equality between the left and right hand sides. Those expressions were not rigorously correct capacity lower bounds (although very good approximations), as $a_{kk}$ is non-Gaussian in general. } \begin{equation} \label{eq:DLrateApprox} R_k \!\approx\! \mathbb{E}\! \left\{\! \log_2 \!\left( \!1 \!+\! \frac{\rho_\textrm{d}|\hat{a}_{kk}|^2}{\rho_\textrm{d} \mathbb{E}\{|\tilde{a}_{kk}|^2\} + \rho_\textrm{d} \sum\limits^K_{k' \neq k} \mathbb{E}\{|a_{kk'}|^2\} + 1} \!\right) \!\right\}. \end{equation} \textit{Proposition 2:} With conjugate beamforming, an achievable rate of the transmission from the APs to the $k$th user is \begin{align} \label{eq:DLrateApprox2} R_k \approx \mathbb{E} \left\{ \log_2 \left( 1 + \frac{\rho_\textrm{d}|\hat{a}_{kk}|^2}{\rho_\textrm{d}\frac{\varsigma_{kk}}{\tau_\textrm{d,p}\rho_\textrm{d,p}\varsigma_{kk}+1} + \rho_\textrm{d} \sum\limits^K_{k' \neq k} \varsigma_{kk'} + 1} \right) \right\}. \end{align} \begin{IEEEproof} See Appendix B. \end{IEEEproof} \section{Numerical Results} \label{NumericalResults} We compare the performance of cell-free massive MIMO for three different assumptions on CSI: \textit{(i)} Statistical CSI, without downlink pilots and users exploiting only statistical knowledge of the channel gain \cite{NgoCF}; \textit{(ii)} Beamforming Training, transmitting downlink pilots and users estimating the gain from those pilots; \textit{(iii)} Perfect CSI, where the users know the effective channel gain. The latter represents an upper bound (genie) on performance, and is not realizable in practice. The gross spectral efficiencies for these cases are given by (\ref{eq:rateComparison}) at the top of the page. Taking into account the performance loss due to the downlink and uplink pilots, the \textit{per-user net throughput} (bit/s) is \setcounter{equation}{24} \begin{equation} \mathcal{S}_k = B \frac{1-\tau_{\textrm{oh}}/\tau}{2} R_k, \end{equation} where $B$ is the bandwidth, $\tau$ is the length of the coherence interval in samples, and $\tau_{\textrm{oh}}$ is the pilots overhead, i.e., the number of samples per coherence interval spent for the training phases. We further examine the performance improvement by using the \textit{max-min fairness power control} algorithm in \cite{NgoCF}, which provides equal and hence uniformly good service to all users for the Statistical CSI case. When using this algorithm for the Beamforming Training case (and for the Perfect CSI bound), we use the power control coefficients computed for the Statistical CSI case. This is, strictly speaking, not optimal but was done for computational reasons, as the rate expressions with user CSI are not in closed form. \subsection{Simulation Scenario} Consider $M$ APs and $K$ users uniformly randomly distributed within a square of size $1 \text{ km}^2$. The large-scale fading coefficient $\beta_{mk}$ is modeled as \begin{equation} \label{eq:beta} \beta_{mk} = \text{PL}_{mk} \cdot 10^{\frac{\sigma_{sh}z_{mk}}{10}} \end{equation} where $\text{PL}_{mk}$ represents the path loss, and $10^{\frac{\sigma_{sh}z_{mk}}{10}}$ is the shadowing with standard deviation $\sigma_{sh}$ and $z_{mk}\sim\mathcal{N}(0,1)$. We consider the three-slope model for the path loss as in \cite{NgoCF} and uncorrelated shadowing. We adopt the following parameters: the carrier frequency is 1.9 GHz, the bandwidth is 20 MHz, the shadowing standard deviation is 8 dB, and the noise figure (uplink and downlink) is 9 dB. In all examples (except for Figures~\ref{cdfM50K10hpw} and \ref{cdfM50K10dpw}) the radiated power (data and pilot) is 200 mW for APs and 100 mW for users. The corresponding normalized transmit SNRs can be computed by dividing radiated powers by the noise power, which is given by \begin{align*} \text{noise power} = \text{bandwidth} \times k_B \times T_0 \times \text{noise figure} \text{ (W)}, \end{align*} where $k_B$ is the Boltzmann constant, and $T_0 = 290$ (Kelvin) is the noise temperature. The AP and user antenna height is 15 m, 1.65 m, respectively. The antenna gains are $0$ dBi. Lastly, we take $\tau_{\textrm{d,p}} = \tau_{\textrm{u,p}} = K$, and $\tau = 200$ samples which corresponds to a coherence bandwidth of 200 kHz and a coherence time of 1 ms. To avoid cell-edge effects, and to imitate a network with an infinite area, we performed a wrap-around technique, in which the simulation area is wrapped around such that the nominal area has eight neighbors. \subsection{Performance Evaluation} We focus first on the performance gain, over the conventional scheme, provided by jointly using Beamforming Training scheme and max-min fairness power control in the downlink. We consider two scenarios, with different network densities. \begin{figure}[!t] \centering \includegraphics[width=3.3in]{cdfM50K10} \caption{The cumulative distribution of the per-user downlink net throughput with and without max-min power control (PC), for the case of statistical, imperfect and perfect CSI knowledge at the user, $M = 50$ and $K = 10$.} \label{cdfM50K10} \end{figure} Figure \ref{cdfM50K10} shows the cumulative distribution function (cdf) of the per-user net throughput for the three cases, with $M=50$, $K=10$. In such a low density scenario, the channel hardening is less pronounced and performing the Beamforming Training scheme yields high performance gain over the statistical CSI case. Moreover, the Beamforming Training curve approaches the upper bound. Combining max-min power control with Beamforming Training scheme, gains can be further improved. For instance, Beamforming Training provides a performance improvement of $18\%$ over the statistical CSI case in terms of 95\%-likely per-user net throughput, and $29\%$ in terms of median per-user net throughput. By contrast, for higher network densities the gap between statistical and Beamforming Training tends to be reduced due to two factors: $(i)$ as $M$ increases, the statistical CSI knowledge at the user side is good enough for reliable downlink detection due to the channel hardening; $(ii)$ as $K$ increases, the pilot overhead becomes significant. In Figure \ref{cdfM100K20} the scenario with $M=100$, $K=20$ is illustrated. Here, the 95\%-likely and the median per-user net throughput of the Beamforming Training improves of $4\%$ and $13\%$, respectively, the performance of the statistical CSI case. \begin{figure}[!t] \centering \includegraphics[width=3.3in]{cdfM100K20} \caption{The cumulative distribution of the per-user downlink net throughput with and without max-min power control (PC), for the case of statistical, imperfect and perfect CSI knowledge at the user, $M = 100$ and $K = 20$.} \label{cdfM100K20} \end{figure} Max-min fairness power control maximizes the rate of the worst user. This philosophy leads to two noticeable consequences: $(i)$ the curves describing with power control are more concentrated around their medians; $(ii)$ as $K$ increases, performing power control has less impact on the system performance, since the probability to have users experiencing poor channel conditions increases. Finally, we compare the performance provided by the two schemes by setting different values for the radiated powers. In Figure \ref{cdfM50K10hpw}, the radiated power is set to 50 mW and 20 mW for the downlink and the uplink, respectively, with $M=50$ and $K=10$. In low SNR regime, with max-min fairness power control, Beamforming Training scheme outperforms the statistical CSI case of about $26\%$ in terms of 95\%-likely per-user net throughput, and about $34\%$ in terms of median per-user net throughput. \begin{figure}[!t] \centering \includegraphics[width=3.3in]{cdfM50K10hpw} \caption{The same as Figure \ref{cdfM50K10}, but the radiated power for data and pilot is 50 mW for APs and 20 mW for users.} \label{cdfM50K10hpw} \end{figure} Similar performance gaps are obtained by increasing the radiated power to 400 mW for the downlink and 200 mW for the uplink, as shown in Figure \ref{cdfM50K10dpw}. \begin{figure}[!t] \centering \includegraphics[width=3.3in]{cdfM50K10dpw} \caption{The same as Figure \ref{cdfM50K10}, but the radiated power for data and pilot is 400 mW for APs and 200 mW for users.} \label{cdfM50K10dpw} \end{figure} \section{Conclusion} Co-located massive MIMO systems do not need downlink training since by virtue of channel hardening, the effective channel gain seen by each user fluctuates only slightly around its mean. In contrast, in cell-free massive MIMO, only a small number of APs may substantially contribute, in terms of transmitted power, to serving a given user, resulting in less channel hardening. We showed that by transmitting downlink pilots, and performing Beamforming Training together with max-min fairness power control, performance of cell-free massive MIMO can be substantially improved. We restricted our study to the case of mutually orthogonal pilots. The general case with non-orthogonal pilots may be included in future work. Further work may also include pilot assignment algorithms, optimal power control, and the analysis of zero-forcing precoding technique. \section*{Appendix} \subsection{Proof of Proposition 1} \begin{itemize} \item Compute $\mathbb{E}\{a_{kk'}\}$: From \eqref{eq:akkdef}, and by using $g_{mk}\triangleq\hat{g}_{mk}+\tilde{g}_{mk}$, we have \begin{align} \label{eq:ChEstWithErr} & a_{kk'} = \sum^M_{m=1} \sqrt{\eta_{mk'}} \hat{g}_{mk} \hat{g}^*_{mk'} + \sum^M_{m=1} \sqrt{\eta_{mk'}} \tilde{g}_{mk} \hat{g}^*_{mk'}. \end{align} Owing to the properties of MMSE estimation, $\tilde{g}_{mk}$ and $\hat{g}_{mk}$ are independent, $k=1, \ldots, K$. Therefore, \begin{align} \label{eq:Eakk} \mathbb{E}\{a_{kk'}\} &= \mathbb{E}\left\{\sum^M_{m=1} \sqrt{\eta_{mk'}} \hat{g}_{mk} \hat{g}^*_{mk'}\right\} \nonumber \\ &= \begin{cases} 0 & \quad \text{if } k' \neq k\\ \sum\limits^M_{m=1} \sqrt{\eta_{mk}} \ \gamma_{mk} & \quad \text{if } k' = k.\\ \end{cases} \end{align} \item Compute $\mathrm{Var}\{a_{kk}\}$: \begin{equation} \label{eq:Varakk} \mathrm{Var}\{a_{kk}\} = \mathbb{E}\{{|a_{kk}|}^2\} - {|\mathbb{E}\{a_{kk}\}|}^2. \end{equation} According to (\ref{eq:ChEstWithErr}), we get \begin{align} \label{eq:EakkSq} & \mathbb{E}\{{|a_{kk}|}^2\} = \mathbb{E} \left\{\left| \sum\limits^M_{m=1} \sqrt{\eta_{mk}} |\hat{g}_{mk}|^2 \right|^2\right\} \nonumber \\ &\qquad\qquad\qquad\qquad {} + \mathbb{E} \left\{\left| \sum\limits^M_{m=1} \sqrt{\eta_{mk}} \tilde{g}_{mk} \hat{g}^*_{mk} \right|^2 \right\} \nonumber \\ &\stackrel{(a)}{=} \mathbb{E} \left\{ \sum\limits^M_{m=1} \sum\limits^M_{m'=1} \sqrt{\eta_{mk}} |\hat{g}_{mk}|^2 \sqrt{\eta_{m'k}} |\hat{g}_{m'k}|^2 \right\} \nonumber \\ &\qquad\qquad {} + \sum\limits^M_{m=1} \eta_{mk}(\beta_{mk} - \gamma_{mk})\gamma_{mk} \nonumber \\ &= \sum\limits^M_{m=1} \sum\limits^M_{m'=1} \sqrt{\eta_{mk} \eta_{m'k}} \ \mathbb{E} \left\{ |\hat{g}_{mk}|^2 |\hat{g}_{m'k}|^2 \right\} + \nonumber \\ &\qquad\qquad {} + \sum\limits^M_{m=1} \eta_{mk}(\beta_{mk} - \gamma_{mk})\gamma_{mk} \nonumber \\ &= \sum\limits^M_{m=1} \eta_{mk}(\beta_{mk} - \gamma_{mk})\gamma_{mk} + \sum\limits^M_{m=1} \eta_{mk} \ \mathbb{E} \left\{ |\hat{g}_{mk}|^4 \right\} \nonumber \\ &\qquad\qquad {} + \sum\limits^M_{m=1} \sum\limits^M_{m' \neq m} \sqrt{\eta_{mk} \eta_{m'k}} \ \mathbb{E} \left\{ |\hat{g}_{mk}|^2 |\hat{g}_{m'k}|^2 \right\} \nonumber \\ &\stackrel{(b)}{=} \sum\limits^M_{m=1} \eta_{mk}(\beta_{mk} - \gamma_{mk})\gamma_{mk} + 2 \sum\limits^M_{m=1} \eta_{mk} \gamma_{mk}^2 \nonumber \\ &\qquad\qquad {} + \sum\limits^M_{m=1} \sum\limits^M_{m' \neq m} \sqrt{\eta_{mk} \eta_{m'k}} \ \gamma_{mk} \gamma_{m'k}, \end{align} where $(a)$ follows from the fact that $\mathbb{E}\{{|\tilde{g}_{mk}|}^2\} = \beta_{mk} - \gamma_{mk}$, and $(b)$ from $\mathbb{E}\left\{{\left|\hat{g}_{mk} \right|}^4\right\} = 2 \gamma^2_{mk}$. From (\ref{eq:Eakk}), we have \begin{align} \label{eq:sqEakk} & {|\mathbb{E}\{a_{kk}\}|}^2 = \left| \sum\limits^M_{m=1} \sqrt{\eta_{mk}} \gamma_{mk} \right|^2 \nonumber \\ &= \sum\limits^M_{m=1} \sum\limits^M_{m'=1} \sqrt{\eta_{mk} \eta_{m'k}} \ \gamma_{mk} \gamma_{m'k} \nonumber \\ &= \sum\limits^M_{m=1} \eta_{mk} \gamma_{mk}^2 + \sum\limits^M_{m=1} \sum\limits^M_{m' \neq m} \sqrt{\eta_{mk} \eta_{m'k}} \ \gamma_{mk} \gamma_{m'k} . \end{align} Substituting (\ref{eq:EakkSq}) and (\ref{eq:sqEakk}) into (\ref{eq:Varakk}), we obtain \begin{equation} \label{eq:VarakkVal} \mathrm{Var}\{a_{kk}\} = \sum\limits^M_{m=1} \eta_{mk} \beta_{mk} \gamma_{mk} = \varsigma_{kk}. \end{equation} Substituting (\ref{eq:Eakk}) and (\ref{eq:VarakkVal}) into (\ref{eq:a_kk}), we get (\ref{eq:a_kk2}). \end{itemize} \subsection{Proof of Proposition 2} \begin{itemize} \item Compute $\mathbb{E}\{{|a_{kk'}|}^2\}$ for $k' \neq k$: From (\ref{eq:ChEstWithErr}) and (\ref{eq:Eakk}), we have \begin{align} \label{eq:Varaki} &\mathbb{E}\{{|a_{kk'}|}^2\} = \mathrm{Var}\{a_{kk'}\} \nonumber \\ &= \mathbb{E}\left\{{\left|\sum\limits^M_{m=1} \sqrt{\eta_{mk'}} \hat{g}_{mk} \hat{g}^*_{mk'}\right|}^2\right\} \nonumber \\ &\qquad\qquad\qquad\qquad {} + \mathbb{E}\left\{{\left|\sum\limits^M_{m=1} \sqrt{\eta_{mk'}} \tilde{g}_{mk} \hat{g}^*_{mk'}\right|}^2\right\} \nonumber \\ &= \sum\limits^M_{m=1} \eta_{mk'} \mathbb{E}\left\{{\left|\hat{g}_{mk} \hat{g}^*_{mk'}\right|}^2\right\} + \sum\limits^M_{m=1} \eta_{mk'} \mathbb{E}\left\{{\left|\tilde{g}_{mk} \hat{g}^*_{mk'}\right|}^2\right\} \nonumber \\ &\stackrel{(a)}{=} \sum\limits^M_{m=1} \eta_{mk'} \gamma_{mk} \gamma_{mk'} + \sum\limits^M_{m=1} \eta_{mk'}(\beta_{mk} - \gamma_{mk})\gamma_{mk'} \nonumber \\ &= \sum\limits^M_{m=1} \eta_{mk'} \beta_{mk} \gamma_{mk'} = \varsigma_{kk'}, \end{align} where $(a)$ is obtained by using (\ref{eq:defGamma}) and the fact that $\tilde{g}_{mk}$ has zero mean and is independent of $\hat{g}_{mk}$. Moreover, we have that $\mathbb{E}\{{|\tilde{g}_{mk}|}^2\} = \beta_{mk} - \gamma_{mk}$. \item Compute $\mathbb{E}\{{|\tilde{a}_{kk}|}^2\}$: From (\ref{eq:a_kk2}) and (\ref{eq:ChEstErr}), we have \begin{align} \label{eq:ekk2} & \mathbb{E}\{{|\tilde{a}_{kk}|}^2\} = \mathbb{E}\{{|a_{kk}-\hat{a}_{kk}|}^2\} \nonumber \\ &= \mathbb{E}\left\{\left|a_{kk} - \frac{\sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \varsigma_{kk} \check{y}_{\textrm{dp},k}}{\tau_\textrm{d,p}\rho_\textrm{d,p}\varsigma_{kk} + 1} - \frac{\sum_{m=1}^M \sqrt{\eta_{mk}} \gamma_{mk}}{\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk} + 1} \right|^2\right\} \nonumber \\ &\stackrel{(a)}{=} \mathbb{E}\left\{\left|a_{kk} \left(1- \frac{\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk}}{\tau_\textrm{d,p}\rho_\textrm{d,p}\varsigma_{kk} + 1}\right) - \frac{\sum_{m=1}^M \sqrt{\eta_{mk}} \gamma_{mk}}{\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk} + 1} \right. \right. \nonumber \\ &\quad\quad \left. \left. - \frac{\sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \varsigma_{kk} n_{\textrm{p},k}}{\tau_\textrm{d,p}\rho_\textrm{d,p}\varsigma_{kk} + 1} \right|^2\right\} \nonumber \\ &= \mathbb{E}\left\{\left|\frac{a_{kk} - \sum_{m=1}^M \sqrt{\eta_{mk}} \gamma_{mk} - \sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}} \varsigma_{kk} n_{\textrm{p},k}}{\tau_\textrm{d,p}\rho_\textrm{d,p}\varsigma_{kk} + 1} \right|^2\right\} \nonumber \\ &\stackrel{(b)}{=} \frac{\mathbb{E}\left\{\left|a_{kk} - \mathbb{E}\{a_{kk}\} - \sqrt{\tau_\textrm{d,p}\rho_\textrm{d,p}}\ \varsigma_{kk} n_{\textrm{p},k} \right|^2\right\}}{\left(\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk}+1\right)^2} \nonumber \\ &\stackrel{(c)}{=} \frac{\mathrm{Var}\{a_{kk}\} + \tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma^2_{kk}}{\left(\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk}+1\right)^2} \nonumber \\ &= \frac{\varsigma_{kk} + \tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma^2_{kk}}{\left(\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk}+1\right)^2} \nonumber \\ &= \frac{\varsigma_{kk}}{\tau_\textrm{d,p}\rho_\textrm{d,p} \varsigma_{kk}+1}, \end{align} where $(a)$ is obtained by using (\ref{eq:receivedDLpilotprojection}), and $(b)$ by using (\ref{eq:Eakk}). Instead, $(c)$ follows from the fact that $a_{kk} - \mathbb{E}\left\{a_{kk}\right\}$, $n_{\textrm{p},k}$ are independent and zero-mean RVs. Moreover, $n_{\textrm{p},k}$ has unitary variance. Substituting (\ref{eq:Varaki}) and (\ref{eq:ekk2}) in (\ref{eq:DLrateApprox}), we arrive at the result in Proposition 2. \end{itemize} \bibliographystyle{IEEEtran}
a9b85c0df78f30693c2498d4c57fe32bdf0a3b3a
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section*{Introduction} Carbon nanotubes (CNTs) are promising nanomaterials, which have been extensively studied by many researchers \cite{cnt1,cnt2,cnt3,cnt4,cnt5,cnt6,cnt7,cnt8,cnt9,cnt10,25,27,28,23,16}. Due to different combinations of structural variation, CNTs can exhibit a wide range of electronic and optical properties, which can be of great use in the design of novel techniques \cite{25}. CNTs are also polyfunctional macromolecules, where specific reactions can occur at various sites with different efficiencies \cite{28}. There are three major types of CNTs: armchair CNTs, chiral CNTs, and zigzag CNTs, which are distinguished by the geometrical vector ($n$,$m$), with $n$ and $m$ being integers. CNTs can behave as either metals or semiconductors depending on their chiral angles, diameters, and lengths. Therefore, a further investigation of how these factors affect the properties of CNTs is essential for the comprehensive understanding of these materials \cite{25,23}. In particular, it is useful to study the basic repeating units of CNTs, which still need further fundamental research exploration \cite{16}. The targeting units of the present study, a series of $n$-cyclacenes, consisting of $n$ fused benzene rings forming a closed loop (see \Cref{fig:tube-cyclic10-geometry}), are the shortest ($n$,0) zigzag CNTs with hydrogen passivation, which have attracted considerable interest in the research community due to their fascinating electronic properties \cite{23,16,7,19,31,1,6,11,24,17,10,9,C2015,new1}. As $n$-cyclacenes belong to the category of cata-condensed aromatics (i.e., molecules that have no carbon atoms belonging to more than two rings), each carbon atom is on the periphery of the conjugated system \cite{7}. Before $n$-cyclacenes are intensively connected to zigzag CNTs, they have been studied mainly due to the research curiosity in highly conjugated cyclic systems. The studies of $n$-cyclacenes can also be important for atomic-level structural control in the synthesis of CNTs. In addition, bottom-up approaches to the synthesis of CNTs not only provide a fundamental understanding of the relationship between the design of CNTs and their electronic properties, but also greatly lower the synthetic temperatures \cite{25}. While zigzag CNTs may be synthesized from cycloarylenes by devising the cutout positions of CNTs \cite{27}, it remains important to systematically investigate the properties of $n$-cyclacenes, which can be useful for exploring the possible utility of their cylindrical cavities in host-guest chemistry \cite{31}. The structure of $n$-cyclacene has two types of components: an arenoid belt (composed of fused benzene rings) and two peripheral circuits (the top and bottom peripheral circuits) \cite{1}. The peripheral circuits are of two types: $4k$ and $4k+2$ (where $k$ is an integer), depending on the number of benzene rings in $n$-cyclacene. In previous studies, it has been shown that $n$-cyclacene with even-number benzene rings ($4k$ type) is more stable than that with odd-number benzene rings ($4k+2$ type) \cite{1,7,6,new1}. Therefore, the nature of peripheral circuits (i.e., the cryptoannulenic effect) is expected to be responsible for the properties of $n$-cyclacene. Besides, the structure of $n$-cyclacene can also be regarded as two fused trannulenes (i.e., circular, all-trans cyclic polyene ribbons) \cite{31,11}. From the bond length analysis of $n$-cyclacene, there is bond length alternation in the benzene ring, and the aromaticity is reduced due to the structural strain, which can hence be responsible for the properties of $n$-cyclacene. Even though there has been a keen interest in $n$-cyclacenes, the studies of their electronic properties are scarce. While $n$-cyclacene may be synthesized via an intramolecular cyclization of $n$-acene (a chain-like molecule with $n$ linearly fused benzene rings, e.g., see Figure 1 of Ref.\ \cite{1}), the synthetic procedure has been very challenging, and has not succeeded in producing pure $n$-cyclacene \cite{17,1,10}, possibly due to its highly strained structure and highly reactive nature \cite{16,17}. As the stabilities of annulated polycyclic saturated hydrocarbons decrease rapidly with the number of fused benzene rings \cite{19}, the synthesis of larger $n$-cyclacenes should be even more difficult. To date, the reported properties of $n$-cyclacenes are based on theoretical calculations. Nevertheless, accurate prediction of the electronic properties of larger $n$-cyclacenes has been very challenging for traditional electronic structure methods, due to the presence of strong static correlation effects \cite{9}. Kohn-Sham density functional theory (KS-DFT) \cite{ks2} with conventional (i.e., semilocal \cite{kslda1,kslda2,PBE,M06L}, hybrid \cite{hybrid,B3LYP,LCHirao,wB97X,wB97X-D,wM05-D,LC-D3,LCAC}, and double-hybrid \cite{B2PLYP,wB97X-2,PBE0-2,SCAN0-2}) exchange-correlation (XC) density functionals can yield unreliable results for systems with strong static correlation effects \cite{Cohen2012}. High-level {\it ab initio} multi-reference methods \cite{9,CASPT2,Acene-DMRG,2-RDM,GNRs-DMRG,GNRs-PHF,GNRs-MRAQCC,multi-reference,multi-reference2} are typically required to accurately predict the properties of larger $n$-cyclacenes. However, as the number of electrons in $n$-cyclacene quickly increases with increasing $n$, there have been very few studies on the properties of larger $n$-cyclacenes using multi-reference methods, due to their prohibitively high cost. To circumvent the formidable computational expense of high-level {\it ab initio} multi-reference methods, we have recently developed thermally-assisted-occupation density functional theory (TAO-DFT) \cite{tao1,tao2}, a very efficient electronic structure method for studying the properties of large ground-state systems (e.g., containing up to a few thousand electrons) with strong static correlation effects \cite{z,NK,HSM}. In contrast to KS-DFT, TAO-DFT is a density functional theory with fractional orbital occupations, wherein strong static correlation is explicitly described by the entropy contribution (see Eq.\ (26) of Ref.\ \cite{tao1}), a function of the fictitious temperature and orbital occupation numbers. Note that the entropy contribution is completely missing in KS-DFT. Recently, we have studied the electronic properties of zigzag graphene nanoribbons (ZGNRs) using TAO-DFT \cite{z}. The ground states of ZGNRs are found to be singlets for all the widths and lengths studied. The longer ZGNRs should possess increasing polyradical character in their ground states, with the active orbitals being mainly localized at the zigzag edges. Our results are in good agreement with the available experimental and highly accurate {\it ab initio} data. Besides, on the basis of our TAO-DFT calculations, the active orbital occupation numbers for the ground states of ZGNRs should exhibit a curve crossing behavior in the approach to unity (singly occupied) with increasing ribbon length. Very recently, the curve crossing behavior has been confirmed by highly accurate {\it ab initio} multi-reference methods \cite{multi-reference2}! TAO-DFT has similar computational cost as KS-DFT for single-point energy and analytical nuclear gradient calculations, and reduces to KS-DFT in the absence of strong static correlation effects. Besides, existing XC density functionals in KS-DFT may also be adopted in TAO-DFT. Relative to high-level {\it ab initio} multi-reference methods, TAO-DFT is computationally efficient, and hence very powerful for the study of large polyradical systems. In addition, the orbital occupation numbers from TAO-DFT, which are intended to simulate the natural orbital occupation numbers (NOONs) [i.e., the eigenvalues of one-electron reduced density matrix] \cite{noon}, can be very useful for assessing the possible polyradical character of systems. Recent studies have demonstrated that the orbital occupation numbers from TAO-DFT are qualitatively similar to the NOONs from high-level {\it ab initio} multi-reference methods, giving promise for applying TAO-DFT to large polyradical systems \cite{tao1,z,NK,multi-reference2}. Due to its computational efficiency and reasonable accuracy for large systems with strong static correlation effects, in this work, TAO-DFT is adopted to study the electronic properties of $n$-cyclacenes ($n$ = 4--100). As $n$-cyclacenes have not been successfully synthesized, no experimental data are currently available for comparison. Therefore, our results are compared with the available high-level {\it ab initio} data as well as those obtained from various XC density functionals in KS-DFT. In addition, as $n$-cyclacene can be considered as an interconnection of $n$-acene, the electronic properties of $n$-cyclacene are also compared with those of $n$-acene to assess the role of cyclic topology. \section*{Computational Details} All calculations are performed with a development version of \textsf{Q-Chem 4.0} \cite{qchem}, using the 6-31G(d) basis set with the fine grid EML(75,302), consisting of 75 Euler-Maclaurin radial grid points and 302 Lebedev angular grid points. Results are calculated using KS-LDA (i.e., KS-DFT with the LDA XC density functional \cite{kslda1,kslda2}) and TAO-LDA (i.e., TAO-DFT with the LDA XC density functional and the LDA $\theta$-dependent density functional $E_{\theta}^{\text {LDA}}$ (see Eq.\ (41) of Ref.\ \cite{tao1}) with the fictitious temperature $\theta$ = 7 mhartree (as defined in Ref.\ \cite{tao1}). Note that KS-LDA is simply TAO-LDA with $\theta$ = 0, and hence it is important to assess the performance of KS-LDA here to assess the significance of TAO-LDA. The ground state of $n$-cyclacene/$n$-acene ($n$ = 4--100) is obtained by performing spin-unrestricted KS-LDA and TAO-LDA calculations for the lowest singlet and triplet energies of $n$-cyclacene/$n$-acene on the respective geometries that were fully optimized at the same level of theory. The singlet-triplet energy (ST) gap of $n$-cyclacene/$n$-acene is calculated as $(E_{\text{T}} - E_{\text{S}})$, the energy difference between the lowest triplet (T) and singlet (S) states of $n$-cyclacene/$n$-acene. \section*{Results and Discussion} \subsection*{Singlet-Triplet Energy Gap} \Cref{fig:stgap} shows the ST gap of $n$-cyclacene as a function of the number of benzene rings, calculated using spin-unrestricted KS-LDA and TAO-LDA. The results are compared with the available data \cite{9}, calculated using the complete-active-space second-order perturbation theory (CASPT2) \cite{CASPT2} (a high-level {\it ab initio} multi-reference method) as well as the M06L functional \cite{M06L} (a popular semilocal XC density functional) and the B3LYP functional \cite{hybrid,B3LYP} (a popular hybrid XC density functional) in KS-DFT. As can be seen, the anticipated even-odd oscillations in the ST gaps may be attributed to the cryptoannulenic effects of $n$-cyclacenes \cite{1,7,6,new1}. However, the amplitudes of the even-odd oscillations are considerably larger for KS-DFT with the XC functionals, which are closely related to the degree of spin contamination (as discussed in Ref.\ \cite{9}). In general, the larger fraction of Hartree-Fock (HF) exchange adopted in the XC functional in KS-DFT, the higher the degree of spin contamination for systems with strong static correlation effects. For example, the ST gap obtained with KS-B3LYP is unexpectedly large at $n$ = 10, due to the high degree of spin contamination \cite{9}. On the other hand, as commented in Ref.\ \cite{9}, the ST gaps obtained with CASPT2 are rather sensitive to the choice of active space. Since the complete $\pi$-valence space was not selected as the active space (due to the prohibitively high cost), the CASPT2 results here should be taken with caution. Recent studies have shown that a sufficiently large active space should be adopted in high-level {\it ab initio} multi-reference calculations \cite{Acene-DMRG,GNRs-DMRG,multi-reference2} for accurate prediction of the electronic properties of systems with strong static correlation effects, which can, however, be prohibitively expensive for large systems. Note that the ST gap obtained with CASPT2 unexpectedly increases at $n$ = 12, possibly due to the insufficiently large active space adopted in the calculations \cite{9}. To assess the role of cyclic topology, \Cref{fig:stgapcycace1,fig:stgapcycace2} show the ST gap of $n$-cyclacene/$n$-acene as a function of the number of benzene rings, calculated with spin-unrestricted TAO-LDA. Similar to $n$-acenes, the ground states of $n$-cyclacenes remain singlets for all the cases investigated. In contrast to $n$-acene, the ST gap of $n$-cyclacene, however, displays oscillatory behavior for small $n$, and the oscillation vanishes gradually with increasing $n$. For small $n$, $n$-cyclacene with even-number benzene rings exhibits a larger ST gap (i.e., greater stability) than that with odd-number benzene rings. For sufficiently large $n$ ($n > 30$), the ST gap of $n$-cyclacene converges monotonically from below to the ST gap of $n$-acene (which monotonically decreases with increasing $n$). At the level of TAO-LDA, the ST gaps of the largest $n$-cyclacene and $n$-acene studied (i.e., $n$ = 100) are essentially the same (0.49 kcal/mol). On the basis of the ST gaps obtained with TAO-LDA, the cryptoannulenic effect and structural strain of $n$-cyclacene are more important for the smaller $n$, and less important for the larger $n$. Due to the symmetry constraint, the spin-restricted and spin-unrestricted energies for the lowest singlet state of $n$-cyclacene/$n$-acene, calculated using the exact theory, should be identical \cite{tao1,tao2,z,GNRs-PHF}. Recent studies have shown that KS-DFT with conventional XC density functionals cannot satisfy this condition for the larger $n$-cyclacene/$n$-acene, due to the aforementioned spin contamination \cite{9,Acene-DMRG,GNRs-DMRG,GNRs-PHF,tao1,tao2,z}. To assess the possible symmetry-breaking effects, spin-restricted TAO-LDA calculations are also performed for the lowest singlet energies on the respective optimized geometries. Within the numerical accuracy of our calculations, the spin-restricted and spin-unrestricted TAO-LDA energies for the lowest singlet state of $n$-cyclacene/$n$-acene are essentially the same (i.e., essentially no unphysical symmetry-breaking effects occur in our spin-unrestricted TAO-LDA calculations). \subsection*{Vertical Ionization Potential, Vertical Electron Affinity, and Fundamental Gap} At the lowest singlet state (i.e., the ground-state) geometry of $n$-cyclacene/$n$-acene (containing $N$ electrons), TAO-LDA is adopted to calculate the vertical ionization potential $\text{IP}_{v}={E}_{N-1}-{E}_{N}$, vertical electron affinity $\text{EA}_{v}={E}_{N}-{E}_{N+1}$, and fundamental gap $E_{g}=\text{IP}_{v}-\text{EA}_{v}={E}_{N+1}+{E}_{N-1}-2{E}_{N}$ using multiple energy-difference methods, with ${E}_{N}$ being the total energy of the $N$-electron system. With increasing number of benzene rings in $n$-cyclacene, $\text{IP}_{v}$ oscillatorily decreases (see \Cref{fig:ip}), $\text{EA}_{v}$ oscillatorily increases (see \Cref{fig:ea}), and hence $E_{g}$ oscillatorily decreases (see \Cref{fig:fg}). However, these oscillations are damped and eventually disappear with increasing $n$. For sufficiently large $n$ ($n > 30$), the $\text{IP}_{v}$ and $E_{g}$ values of $n$-cyclacene converge monotonically from above to those of $n$-acene (which monotonically decrease with increasing $n$), while the $\text{EA}_{v}$ value of $n$-cyclacene converges monotonically from below to that of $n$-acene (which monotonically increases with increasing $n$). Note also that the $E_{g}$ value of $n$-cyclacene ($n$ = 13--54) is within the most interesting range (1 to 3 eV), giving promise for applications of $n$-cyclacenes in nanophotonics. \subsection*{Symmetrized von Neumann Entropy} To investigate the possible polyradical character of $n$-cyclacene/$n$-acene, we calculate the symmetrized von Neumann entropy (e.g., see Eq.\ (9) of Ref.\ \cite{GNRs-PHF}) \begin{equation}\label{eq1} S_{\text{vN}} = -\frac{1}{2} \sum_{i=1}^{\infty} \bigg\lbrace f_{i}\ \text{ln}(f_{i}) + (1-f_{i})\ \text{ln}(1-f_{i}) \bigg\rbrace, \end{equation} for the lowest singlet state of $n$-cyclacene/$n$-acene as a function of the number of benzene rings, using TAO-LDA. Here, $f_{i}$ the occupation number of the $i^{\text{th}}$ orbital obtained with TAO-LDA, which ranges from 0 to 1, is approximately the same as the occupation number of the $i^{\text{th}}$ natural orbital \cite{tao1,tao2,z,NK,HSM,multi-reference2}. For a system without strong static correlation ($\{f_{i}\}$ are close to either 0 or 1), $S_{\text{vN}}$ provides insignificant contributions, while for a system with strong static correlation ($\{f_{i}\}$ are fractional for active orbitals and are close to either 0 or 1 for others), $S_{\text{vN}}$ increases with the number of active orbitals. As shown in \Cref{fig:s}, the $S_{\text{vN}}$ value of $n$-cyclacene oscillatorily increases with increasing number of benzene rings. Nonetheless, the oscillation is damped and eventually disappears with the increase of $n$. For sufficiently large $n$ ($n > 30$), the $S_{\text{vN}}$ value of $n$-cyclacene converges monotonically from above to that of $n$-acene (which monotonically increases with increasing $n$). Therefore, similar to $n$-acenes \cite{tao1,tao2,z,NK,HSM,Acene-DMRG,GNRs-DMRG,GNRs-PHF,GNRs-MRAQCC,multi-reference2}, the larger $n$-cyclacenes should possess increasing polyradical character. \subsection*{Active Orbital Occupation Numbers} To illustrate the causes of the increase of $S_{\text{vN}}$ with $n$, we plot the active orbital occupation numbers for the lowest singlet state of $n$-cyclacene as a function of the number of benzene rings, calculated using TAO-LDA. Here, the highest occupied molecular orbital (HOMO) is the ${(N/2)}^{\text{th}}$ orbital, and the lowest unoccupied molecular orbital (LUMO) is the ${(N/2 + 1)}^{\text{th}}$ orbital, where $N$ is the number of electrons in $n$-cyclacene. For brevity, HOMO, HOMO$-$1, ..., and HOMO$-$15, are denoted as H, H$-$1, ..., and H$-$15, respectively, while LUMO, LUMO+1, ..., and LUMO+15, are denoted as L, L+1, ..., and L+15, respectively. As presented in \Cref{fig:occupation}, the number of fractionally occupied orbitals increases with increasing cyclacene size, clearly indicating that the polyradical character of $n$-cyclacene indeed increases with the cyclacene size. Similar to the previously discussed properties, the active orbital occupation numbers of $n$-cyclacene also exhibit oscillatory behavior, showing wave-packet oscillations. \subsection*{Real-Space Representation of Active Orbitals} For the lowest singlet states of some representative $n$-cyclacenes ($n$ = 4--7), we explore the real-space representation of active orbitals (e.g., HOMOs and LUMOs), obtained with TAO-LDA. Similar to previous findings for $n$-acenes \cite{Acene-DMRG,GNRs-DMRG,GNRs-PHF,z}, the HOMOs and LUMOs of $n$-cyclacenes are mainly localized at the peripheral carbon atoms (see \Cref{fig:realspace}). \section*{Conclusions} In conclusion, we have studied the electronic properties of $n$-cyclacenes ($n$ = 4--100), including the ST gaps, vertical ionization potentials, vertical electron affinities, fundamental gaps, symmetrized von Neumann entropy, active orbital occupation numbers, and real-space representation of active orbitals, using our newly developed TAO-DFT, a very efficient electronic structure method for the study of large systems with strong static correlation effects. To assess the effects of cyclic nature, the electronic properties of $n$-cyclacenes have also been compared with those of $n$-acenes. Similar to $n$-acenes, the ground states of $n$-cyclacenes are singlets for all the cases investigated. In contrast to $n$-acenes, the electronic properties of $n$-cyclacenes, however, display oscillatory behavior for small $n$ ($n \le 30$) in the approach to the corresponding properties of $n$-acenes with increasing number of benzene rings, which to the best of our knowledge have never been addressed in the literature. The oscillatory behavior may be related to the cryptoannulenic effect and structural strain of $n$-cyclacene, which have been shown to be important for small $n$, and unimportant for sufficiently large $n$. On the basis of several measures (e.g., the smaller ST gap, the smaller $E_{g}$, and the larger $S_{\text{vN}}$), for small $n$, $n$-cyclacene with odd-number benzene rings should possess stronger radical character than that with even-number benzene rings. In addition, based on the calculated orbitals and their occupation numbers, the larger $n$-cyclacenes are expected to possess increasing polyradical character in their ground states, where the active orbitals are mainly localized at the peripheral carbon atoms. Since TAO-DFT is computationally efficient, it appears to be a promising method for studying the electronic properties of large systems with strong static correlation effects. Nevertheless, as with all approximate electronic structure methods, a few limitations remain. Relative to the exact full configuration interaction (FCI) method \cite{FCI}, TAO-LDA (with $\theta$ = 7 mhartree) is not variationally correct (i.e., overcorrelation can occur), and hence, the orbital occupation numbers from TAO-LDA may not be the same as the NOONs from the FCI method. To assess the accuracy of our TAO-LDA results, as the computational cost of the FCI method is prohibitive, the electronic properties of $n$-cyclacenes from relatively affordable {\it ab initio} multi-reference methods are called for. \section*{Acknowledgements} This work was supported by the Ministry of Science and Technology of Taiwan (Grant No.\ MOST104-2628-M-002-011-MY3), National Taiwan University (Grant No.\ NTU-CDP-105R7818), the Center for Quantum Science and Engineering at NTU (Subproject Nos.:\ NTU-ERP-105R891401 and NTU-ERP-105R891403), and the National Center for Theoretical Sciences of Taiwan. We are grateful to the Computer and Information Networking Center at NTU for the partial support of high-performance computing facilities. \section*{Author Contributions} C.-S.W. and P.-Y.L. contributed equally to this work. J.-D.C. conceived and designed the project. C.-S.W. and J.-D.C. performed the calculations. P.-Y.L. and J.-D.C. wrote the paper. All authors performed the data analysis. \section*{Additional Information} {\bf Competing financial interests:} The authors declare no competing financial interests.
ec2123e1664f35ffb4af3470b792dcf9e5391c4c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{intr} The frameworks of nonstationary nonperiodic and periodic wavelets are introduced and studied separately. The notion of nonstationary wavelet system is introduced independently by M.~Z.~Berkolayko, I.~Y.~Novikov \cite{BN} and by C.~de Boor, R.~DeVore, A.~Ron \cite{BDR}. In \cite{BN}, the nonstationary system (called almost-wavelets) is used to construct an orthonormal shift invariant basis consisting of infinitely differentiable compactly supported functions. It is well known that it is impossible to construct stationary wavelet basis satisfying these properties. Further, nonstationary wavelets are studied in \cite{DR, DKLR, HSh, NPS}. Concerning to periodic case, first, periodic wavelets are generated by periodization of stationary wavelet functions. Later, a general approach to study periodic wavelets directly using a periodic analog of multiresolution analysis (MRA) is developed. A notion of periodic MRA is introduced and discussed in \cite{klt95, NPS, P, PT, S97}. The main purpose of this paper is to compare frameworks of nonstationary nonperiodic wavelets and periodic wavelets. We intend to extrapolate the idea of periodization and to study how to construct one system from another using periodization. Parseval wavelet frames generated by unitary extension principles (UEP) are considered. We have not found a UEP for nonstationary setting in literature, so we formulate this principle in Theorem \ref{nUEP}. The proof is omitted since it can be easily rewritten from Theorem 1.8.9 \cite{NPS} replacing stationary conditions by nonstationary ones. For periodic setting we cite UEP proved by S.S.~Goh, K.M.~Teo in 2008 (Theorem \ref{pUEP}). In Proposition \ref{nst_to_per} it is proved that the periodization of a nonstationary Parseval wavelet frame is a periodic Parseval wavelet frame. The proof is reduced to comparison of the UEPs. The inverse problem, namely to construct a nonstationary system such that its periodization is the initial periodic system, is more interesting and complicated. It has infinitely many solutions. In Lemma \ref{def_mask} we design a family of nonstationary masks $m^K_j$ corresponding to the initial periodic wavelets. A family parameter $K\in\n$ is responsible for the smoothness of a mask and for an order of a zero at the point $\xi=1/2.$ It is well known that the last characteristic is also important since it is necessary condition for the smoothness of scaling and wavelet function, at least, for stationary case. In Lemma \ref{prodL2}, we suggest a nonstationary analogue of a sufficient condition for an infinite product $\prod_{r=1}^{\infty} m_{j+r}(\xi 2^{-r})$ to converge uniformly on a compact and to be in $L_2(\r).$ In Lemma \ref{my_prodL2}, this sufficient condition is specified for the masks $m^K_j$ defined in Lemma \ref{def_mask}. So, we can design nonstationary wavelet system such that its periodization coincides with the given periodic system and its nonstationary masks can be chosen arbitrarily smooth. Now, among these solutions we are looking for wavelet frame with the following additional property. Time-frequency localization of the resulting nonstationary system should be adjusted with angular-frequency localization of the initial periodic system, that is $ \lim_{j\to \infty} UC_B(\psi^P_j) = \lim_{j\to \infty}UC_H(\psi^N_j), $ where $UC_B,$ $UC_H$ are the Breitenberger and the Heisenberg uncertainty constants (UCs). We have a particular motivation to be interested in this issue. Good time-frequency localization (that means finiteness of UC) is a desirable and natural property of wavelet functions. It follows from the uncertainty principle that there is no function with arbitrarily small UC (less than $1/2$ for both UCs). In \cite{Bat} it is proved that there is no wavelet function with $UC_H$ less than $3/2.$ And it is unknown whether there exists a stationary or a nonstationary wavelet (an orthonormal basis or a tight frame) $\psi^0$ such that $UC_H(\psi^0) \leq 3/2 +\varepsilon.$ Results on (un)boundedness of UC for various wavelet families can be found in \cite{GB, GL, L07, L12, L11, N1}. However in the periodic case the situation is little bit better. In \cite{LebPres14} we construct a Parseval periodic wavelet frame satisfying the following property $\lim_{j\to\infty} UC_B(\psi^{P}_j)=3/2$. And in \cite{L15}, we show that $3/2$ is a minimal possible value of $UC_B$ for a wide class of periodic wavelet sequences. So, our particular motivation is to construct nonstationary wavelet systems with good localization starting with well-localized periodic ones, and conversely. In Theorem \ref{nstPQRS} we obtain a sufficient condition for nonstationary and periodic wavelet frames to satisfy equality $ \lim_{j\to \infty} UC_B(\psi^P_j) = \lim_{j\to \infty}UC_H(\psi^N_j). $ In Theorem \ref{adjust_loc} this sufficient condition is formulated in terms of scaling masks of the given periodic wavelet system. \section{Notations and auxiliary results} \label{note} Let $L_2(\mathbb{T})$ be the space of all $1$-periodic square-integrable complex-valued functions, with inner product $(\cdot,\cdot)$ given by $ (f,\,g):= \int_{0}^{1} f(x)\overline{g(x)}\,\mathrm{d}x $ for any $f,g \in L_2(\mathbb{T}),$ and norm $\|\cdot\|:=\sqrt{(\cdot,\,\cdot)}.$ The Fourier series of a function $ f \in L_2(\mathbb{T}) $ is defined by $\sum_{k \in \mathbb{Z}}\widehat{f}(k) \mathrm{e}^{2 \pi \mathrm{i} k x},$ where its Fourier coefficient is defined by $ \widehat{f}(k) = \int_{0}^{1} f(x)\mathrm{e}^{- 2 \pi \mathrm{i} k x}\,\mathrm{d}x. $ Let $L_{2}(\r)$ be the space of all square-integrable complex-valued functions, with inner product $(\cdot,\cdot)$ given by $ (f,\,g):= \int_{\r} f(x)\overline{g(x)}\,\mathrm{d}x $ for any $f,g \in L_2(\r),$ and norm $\|\cdot\|:=\sqrt{(\cdot,\,\cdot)}.$ The Fourier transform of a function $ f \in L_{2}(\r) $ is defined by $ \widehat{f}(\xi):= \int_{\r}f(x) \mathrm{e}^{- 2 \pi \mathrm{i} \xi x}\,\mathrm{d}x. $ We recall the notion of a tight frame. Let $H$ be a separable Hilbert space. If there exists a constant $A>0$ such that for any $f \in H$ the following equality holds $ \sum_{n=1}^{\infty} \left|(f,\,f_n)\right|^2 = A \|f\|^2, $ then the sequence $(f_n)_{n \in \mathbb{N}}$ is called \texttt{a tight frame} for $H.$ In the case $A=1$, a tight frame is called \texttt{ a Parseval frame}. In addition, if $\|f_n\|=1$ for all $n \in \mathbb{N}$, then a Parseval frame forms an orthonormal basis. To keep ideas clear and notations simple we consider one-dimensional wavelet systems with one wavelet generator. We recall the basic definitions. In the sequel, we use the following notation for a shift $ f_{j,k}(x):=f_j(x- 2^{-j} k) $ of a function $f_j \in L_2(\mathbb{T})$ or $f_j \in L_2(\mathbb{R}),$ where $k \in \mathbb{Z}.$ Given $\varphi^N_0,\,\psi^N_j \in L_2(\mathbb{R}),$ $j\in\z_+$, if the set $\Psi^N:=\left\{\varphi^N_0, \psi^N_{j,k} : \ j\in\z_+,\ k \in \mathbb{Z} \right\}$ forms a frame (or a basis) for $L_2(\mathbb{R})$ then $\Psi^N$ is called \texttt{a nonstationary wavelet frame (or a nonstationary wavelet basis)} for $L_2(\mathbb{R}).$ In the stationary case, all the wavelet functions $\psi^N_j$ are generated by a single function $\psi$, namely $\psi^N_j(x)=2^{j/2}\psi(2^j x)$. Analogously, let $\varphi^P_0,\,\psi^P_j \in L_2(\mathbb{T}),$ $j\in\z_+$ be periodic functions. If the set $\Psi^P:=\left\{\varphi^P_0, \psi^P_{j,k} : \ j\in\z_+,\ \right.$ $\left. k=0,\dots,2^j-1 \right\}$ forms a tight frame (or a basis) for $L_2(\mathbb{T})$ then $\Psi^P$ is said to be \texttt{a periodic tight wavelet frame (or a periodic wavelet basis)} for $L_2(\mathbb{T}).$ We compare framework of nonstationary and periodic wavelet systems generated by unitary extension principles (UEP). It is not difficult to revise a proof of stationary UEP (see, for example, Theorem 1.8.9 \cite{NPS}) and to obtain its nonstationary counterpart. The result is formulated in the following theorem. We omit the proof since it can be rewritten from Theorem 1.8.9 \cite{NPS} replacing stationary conditions by nonstationary ones. \begin{teo}[the unitary extension principle for a nonstationary setting] \label{nUEP} Let $\varphi^N_j \in L_{2}(\mathbb{R}),$ $j\in\z_+$ be a sequence of functions such that \begin{equation} \label{ncon1} \lim_{j \to \infty}2^{j/2} \widehat{\varphi^N_j}(\xi) = 1, \ \ \ \ \ \ \ \xi \in \r. \end{equation} Let $m^N_{0,j} \in L_{2}(\mathbb{T})$ be a $1$-periodic function such that \begin{equation} \label{ncon2} \left|m^N_{0,j}(\xi)\right|^2 + \left|m^N_{0,j}(\xi+1/2)\right|^2 =2, \end{equation} and \begin{equation} \label{ncon3} \widehat{\varphi^N_j}(\xi)=m^N_{0,j+1}(\xi/2^{-j-1}) \widehat{\varphi^N_{j+1}}(\xi). \end{equation} Let $\psi^N_j \in L_{2}(\mathbb{R})$, $j\in\z_+$ be a sequence of functions defined by \begin{equation} \label{ncon4} \widehat{\psi^N_j}(\xi):=m^N_{1,j+1}(\xi/2^{-j-1}) \widehat{\varphi^N_{j+1}}(\xi), \end{equation} where $m^N_{1,j}(\xi)= e^{2 \pi i \xi} \overline{m^N_{0,j}(\xi+1/2)}. $ Then the family $\Psi^N =\{\varphi^{N}_{0,k}, \, \psi^{N}_{j,k} : \ j\in \mathbb{Z}_+,\ k \in \mathbb{Z}\} $ forms a Parseval wavelet frame for $L_{2}(\mathbb{R}).$ \end{teo} The functions $\varphi^N_j,$ $\psi^N_j,$ $m^N_{0,j},$ and $m^N_{1,j}$ are called \texttt{a scaling function, a wavelet function, a scaling mask and a wave\-let mask} respectively. \begin{teo}[\cite{GT1}; the unitary extension principle for a periodic setting] \label{pUEP} Let $\varphi^P_j \in L_{2}(\mathbb{T}),$ $j\in\z_+$, be a sequence of $1$-periodic functions such that \begin{equation}\label{con1} \lim_{j \to \infty}2^{j/2} \widehat{\varphi^P_j}(k) = 1 , \ \ \ \ \ \ \ k \in \z. \end{equation} Let $\mu^j_k \in \cn,$ $j\in\z_+,$ $k\in\z$, be a two-parameter sequence such that $\mu^j_{k+2^j}=\mu^j_{k},$ \begin{equation} \label{con2} \left|\mu^j_{k}\right|^2 + \left|\mu^j_{k+2^{j-1}}\right|^2 =2, \end{equation} and \begin{equation} \label{con3} \widehat{\varphi^P_j}(k)=\mu^{j+1}_{k} \widehat{\varphi^P_{j+1}}(k). \end{equation} Let $\psi^P_j \in L_{2}(\mathbb{T})$, $j\in\z_+$ be a sequence of functions defined by \begin{equation} \label{con4} \widehat{\psi^P_j}(k):=\lambda^j_{k} \widehat{\varphi^P_{j+1}}(k), \end{equation} where $\lambda^j_{k}= e^{2 \pi i k/2^{-j}} \overline{\mu^j_{k+2^{j-1}}}. $ Then the family $\Psi^P =\{\varphi^{P}_{0}, \, \psi^{P}_{j,k} : \ j\in \mathbb{Z}_+, \ k=0,\dots,2^j-1 \}$ forms a Parseval wavelet frame for $L_{2}(\mathbb{T}).$ \end{teo} The functions $\varphi^P_j,$ $\psi^P_j$ and sequences $(\mu^j_k)_{k\in\z},$ $(\lambda^j_k)_{k\in\z}$ are called \texttt{a scaling function, a wavelet function, a scaling mask, and a wave\-let mask} respectively. It is convenient to construct both systems starting with scaling masks. Namely, let $\nu^{j}_{k}$ be a sequence such that $\nu^{j}_{k}=\nu^{j}_{k+2^j}$. We define $\widehat{\xi}_{j}(k):=\prod_{r=j+1}^{\infty}\nu^{r}_{k}.$ If the above infinite product converges, then the scaling function, scaling mask, wavelet mask, and wavelet function in a periodic setting are defined respectively as \begin{gather} \widehat{\varphi_{j}}(k):=2^{-j/2}\widehat{\xi}_{j}(k),\qquad\quad \mu^{j}_k:=\sqrt{2} \nu^{j}_k,\ \notag\\ \lambda^{j}_k:=e^{2\pi i 2^{-j}k}\mu^{j}_{k+2^{j-1}},\qquad \quad {\widehat{\psi}}_{j}(k):=\lambda^{j+1}_{k} \widehat{\varphi}_{j+1}(k). \notag \end{gather} Analogously, suppose $m_j$ is a $1$-periodic function, $\widehat{\varphi_j}(\xi):=\prod_{r=1}^{\infty}m_{j+r}(\xi/2^r)$ is well-defined, then the scaling function, scaling mask, and wavelet function in a nonstationary setting are defined respectively as \begin{gather} \label{aux_n} \widehat{\varphi^N_j}(\xi) = 2^{-j/2}\widehat{\varphi_j}(\xi/2^j),\qquad\quad m_j(\xi)=2^{-1/2} m^N_{0,j}(\xi),\ \\ \widehat{\psi^N_j} (\xi) = 2^{-j/2} m_{1,j+1}(\xi/2^{j+1}) \widehat{\varphi_{j+1}}(\xi/2^{j+1})=2^{-j/2} \widehat{\psi_j}(\xi/2^j) \notag \end{gather} Auxiliary scaling and wavelet functions $\varphi_j$ and $\psi_j$ are connected with nonstationary scaling and wavelet functions as $\psi^N_j(x)=2^{j/2}\psi_j(2^j x)$ and $\varphi^N_j(x)=2^{j/2}\varphi_j(2^j x).$ In stationary case $\varphi_j$ and $\psi_j$ are coincided with scaling and wavelet functions $\varphi$ and $\psi$. Comparing these two UEPs we get the following \begin{prop} \label{nst_to_per} If $\Psi^N = \{\varphi^{N}_{0,k}, \, \psi^{N}_{j,k} \}_{j\in \mathbb{Z}_+,\, k \in \mathbb{Z}}$ is a Parseval wavelet frame generated by UEP, $\varphi^{N}_{0}, \psi^{N}_{j} \in L_1(\mathbb{R}),$ and $$ \psi^P_j(x):=\sum_{n\in\mathbb{Z}} \psi^N_j (x-n), $$ then $\Psi^P =\{\varphi^{P}_{0}, \, \psi^{P}_{j,k} \}_{j\in \mathbb{Z}_+,\, k=0,\dots,2^j-1}$ is a Parseval wavelet frame. \end{prop} \textbf{Proof.} Since $\varphi^{N}_{0}, \psi^{N}_{j} \in L_1(\mathbb{R}),$ periodization $ \psi^P_j(x)=\sum_{n\in\mathbb{Z}} \psi^N_j (x-n) $ is well-defined. In the Fourier domain the last equality is rewritten as $\widehat{\psi^N_j}(k)=\widehat{\psi^P_j}(k).$ Therefore, substituting $k\in \z$ or $k 2^{-j},$ where $k \in \z,$ $j\in\z_+$ instead of $\xi\in \r$ in conditions (\ref{ncon1})-(\ref{ncon4}) we immediately get (\ref{con1})-(\ref{con4}). Proposition \ref{nst_to_per} is proved. \hfill $\Diamond$ We recall the definitions of the UCs and the uncertainty principles. \texttt{The Heisenberg UC} of $f \in L_2(\mathbb{R})$ is the functional $UC_H(f):=\Delta(f)\Delta(\widehat{f})$ such that $$ \Delta^2(f):= \frac{1}{\|f\|^{2}} \int_{\mathbb{R}} \bigl(x-c(f)\bigr)^2 \bigl|f(x)\bigr|^2 \, dx , \quad c(f):= \frac{1}{\|f\|^{2}} \int_{\mathbb{R}} x \bigl|f(x)\bigr|^2 \, dx, $$ where $\Delta(f),$ $\Delta(\widehat{f}),$ $c(f),$ and $c(\widehat{f})$ are called \texttt{time variance, frequency variance, time centre,} and \texttt{frequency centre} respectively. The Heisenberg uncertainty principle says that $UC_H(f)\geq 1/2$ for $f \in L_2(\mathbb{R})$, and the equality is attained iff $f$ is the Gaussian function. In \cite[p. 137]{Bat} the following refinement of the Heisenberg uncertainty principle is proved. If $f \in L_2(\r),$ $\ c(\widehat{f})$ $=0$, and $\int_{\r} f = 0,$ then $UC_H(f)\geq 3/2.$ The UC for periodic functions is introduced in \cite{B}. Let $f =\sum_{k \in \mathbb{Z}} c_k \mathrm{e}^{ \mathrm{i} k \cdot}\in L_{2,2\pi}.$ \texttt{ The first trigonometric moment} is defined as $$ \tau(f):=\frac{1}{2 \pi} \int_{-\pi}^{\pi} \mathrm{e}^{ \mathrm{i} x} |f(x)|^2\, \mathrm{d}x = \sum_{k \in \mathbb{Z}} c_{k-1} \overline{c_{k}}. $$ \texttt{The angular variance} of the function $f$ is defined by $$ {\rm var_A }(f):= \frac{\left(\sum_{k \in \mathbb{Z}}|c_k|^2\right)^2}{ \left|\sum_{k \in \mathbb{Z}}c_{k-1} \bar{c}_{k}\right|^2}-1 = \frac{\|f\|^4}{|\tau(f)|^2}-1. $$ \texttt{The frequency variance} of the function $f$ is defined by $$ {\rm var_F }(f):= \frac{\sum_{k \in \mathbb{Z}}k^2 |c_k|^2}{\sum_{k \in \mathbb{Z}}|c_k|^2}- \frac{\left(\sum_{k \in \mathbb{Z}}k|c_k|^2\right)^2}{\left(\sum_{k \in \mathbb{Z}}|c_k|^2\right)^2} = \frac{\|f'\|^2}{\|f\|^2}-\frac{(i f',\, f)^2}{\|f\|^4}. $$ The quantity $ UC_B(\{c_k\}):=UC_B(f):=\sqrt{\mathrm{var_A}(f)\mathrm{var_F}(f)} $ is called \texttt{the Breitenberger (periodic) UC}. The corresponding uncertainty principle \cite{B, PQ} says that if $f \in L_{2,2\pi}$, $f(x)\neq C \mathrm{e}^{ \mathrm{i} k x},$ $C \in \mathbb{R},$ $k \in \mathbb{Z}$, then $UC_B(f) > 1/2$ and there is no function such that $UC_B(f) = 1/2$ \section{Results} \subsection{Nonstationary wavelets generated by periodic wavelets} \begin{lem} \label{def_mask} Let $\nu^j_k\in\mathbb{R},$ $\nu^j_k \geq 0,$ be a two-parametric sequence such that $\nu^j_k=\nu^j_{k+2^j},$ $|\nu^j_k|^2 + |\nu^j_{k+2^{j-1}}|^2=1,$ $\nu^j_k = \nu^j_{-k}.$ By definition, put $\theta^j_k:=\arccos \nu^j_k.$ Let a function $z^K_j$ defined on the interval $[0,\,1/4]$ be a uniform spline of order $K+1$ of minimal defect such that $z^K_j(k 2^{-j}) = \theta^j_k,$ $k=0,\dots,2^{j-2},$ $(z^K_j)^{(l)}(0)=0,$ $l=1,\dots,K-1.$ Finally, let $m^K_j$ be an even $1$-periodic function defined as \begin{equation} \label{mask} m^K_j(\xi)= \left\{ \begin{array}{ll} \cos(z^K_j(\xi)), & \xi \in [0,\,1/4], \\ \sin(z^K_j(1/2-\xi)), & \xi \in (1/4,\,1/2]. \end{array} \right. \end{equation} Then $|m^K_j(\xi)|^2+|m^K_j(\xi+1/2)|^2=1,$ $m^K_j \in C^{K-1}(\mathbb{T}),$ $(m^K_j)^{(l)}(1/2) = 0,$ $l=1,\dots,K-1.$ \end{lem} \textbf{Proof.} The equality $|m^K_j(\xi)|^2+|m^K_j(\xi+1/2)|^2=1$ follows immediately from the detailed definition of $m^K_j$ $$ m^K_j(\xi)= \left\{ \begin{array}{ll} \sin(z^K_j(1/2+\xi-k)), & \xi \in [-1/2+k,\, -1/4+k), \\ \cos(z^K_j(-\xi+k)), & \xi \in [-1/4+k,\, k), \\ \cos(z^K_j(\xi-k)), & \xi \in [k,\, 1/4+k), \\ \sin(z^K_j(1/2-\xi+k)), & \xi \in [1/4+k,\, 1/2+k), k \in \z. \end{array} \right. $$ For instance, for $\xi \in [-1/2+k,\, -1/4+k)$ we get $$ |m^K_j(\xi)|^2+|m^K_j(\xi+1/2)|^2= \sin^2(z^K_j(1/2+\xi-k)) + \cos^2(z^K_j(1/2+\xi-k)) =1. $$ Since $z^K_j$ is a spline of order $K+1$ of minimal defect, we obtain $z^K_j \in C^{K-1}(\mathbb{T}).$ Therefore, $m^K_j$ is $K-1$ times continuously differentiable at all points except $n/4,$ where $n\in\z.$ The smoothness of $m^K_j$ at the origin follows from the fact that $\cos$ is smooth and even. Since $m^K_j$ is periodic and even, it remains to check smoothness at points $\xi = 1/4,$ $\xi=1/2.$ At the point $\xi=1/4$, we get for $n = 0, \dots, K-1$ $$ \left.(m^K_j)^{(n)}(\xi)\right|_{\xi =1/4-0} = \left. \left(\cos \left( z^K_j(\xi) \right)\right)^{(n)} \right|_{\xi =1/4-0} $$ $$ \left.(m^K_j)^{(n)}(\xi)\right|_{\xi =1/4+0} = \left. \left(\cos \left(\pi/2 - z^K_j(1/2 - \xi) \right)\right)^{(n)} \right|_{\xi =1/4+0}. $$ Thus, using $z^K_j(1/4)=\theta^j_{2^{j-2}}=\pi/4$, we get $(m^K_j)^{(n)}(1/4-0) = (m^K_j)^{(n)}(1/4+0)$ for $n = 0, \dots, K-1.$ At the point $\xi=1/2$ we have $$ \left.(m^K_j)^{(n)}(\xi)\right|_{\xi =1/2-0} = \left.(\sin(z^K_j(1/2-\xi)))^{(n)} \right|_{\xi =1/2-0} $$ $$ \left.(m^K_j)^{(n)}(\xi)\right|_{\xi =1/2+0} = \left.(\sin(z^K_j(-1/2+\xi)))^{(n)} \right|_{\xi =1/2+0} $$ So, by the condition $(z^K_j)^{(l)}(0)=0,$ $l=1,\dots,K-1$ we immediately obtain $$ (m^K_j)^{(n)}(1/2-0) = (m^K_j)^{(n)}(1/2+0) = 0. $$ This concludes the proof of Lemma \ref{def_mask}. \hfill $\Diamond$ \begin{lem} \label{prodL2} If $a_j\in L_2(\mathbb{T}),$ $a_j(0)=1$, and $\sum_{j}\|a''_j\|_2/2^j<\infty$, then $\widehat{\varphi_j}(\xi)=\prod_{r=1}^{\infty} a_{j+r}(\xi/2^r)$ uniformly and absolutely converges on any $[a,\,b]\subset \mathbb{R}.$ If additionally $|a_j(\xi)|^2+|a_j(\xi+1/2)|^2=1,$ then $\widehat{\varphi_j} \in L_2(\mathbb{R})$ and $\|\widehat{\varphi_j}\|_2\leq 1.$ \end{lem} \textbf{Proof.} This is a slight modification of the corresponding stationary result (see \cite[Proposition 2.4.1]{NPS}). Suppose $a_j(\xi) = \sum_{k\in \z}c_{j,k} e^{2 \pi i k \xi}$ and $a''_j \in L_2(\mathbb{T})$. Using $a_j(0)=1$, we get $$ |a_j(\xi) - 1|= \left|\sum_{k\in \z} c_{j,k} \left( e^{2 \pi i k \xi} - 1\right)\right| \leq 2\pi \sum_{k\in \z} |c_{j,k}| |k| |\xi| \leq C |\xi| \left(\sum_{k\in \z} | k^2 c_{j,k}|^2 \right)^{1/2} = C_1 |\xi| \|a''_j\|_2. $$ Therefore, $$ \sum_{r=1}^{\infty}|a_{j+r}(\xi/2^r) - 1| \leq C_1 |\xi| \sum_{r=1}^{\infty} \frac{\|a''_{j+r}\|_2}{2^r} = C_1 2^j |\xi| \sum_{n=1}^{\infty} \frac{\|a''_{n}\|_2}{2^n}. $$ Hence, the infinite product $\prod_{r=1}^{\infty} a_{j+r}(\xi/2^r)$ uniformly with respect to $\xi$ and absolutely converges on any $[a,\,b]\subset \mathbb{R}.$ The proof of the facts $\widehat{\varphi_j} \in L_2(\mathbb{R})$ and $\|\widehat{\varphi_j}\|_2\leq 1$ can be rewritten from stationary case \cite[Lemma 4.1.3]{NPS}. Lemma \ref{prodL2} is proved. \hfill $\Diamond$ \begin{lem} \label{my_prodL2} If $m^K_j$ is defined by (\ref{mask}), $\nu^j_0=1,$ and $$ \sum_{j}\frac{1}{2^j}\left(\int_{0}^{1/4}((z^K_j)'(\xi))^4+((z^K_j)''(\xi))^2\,d\xi\right)^{1/2}<\infty, $$ then $\widehat{\varphi_j}(\xi)=\prod_{r=1}^{\infty} m^K_{j+r}(\xi/2^r)$ uniformly and absolutely converges on any $[a,\,b]\subset \mathbb{R},$ $\widehat{\varphi_j} \in L_2(\mathbb{R})$ and $\|\widehat{\varphi_j}\|_2\leq 1.$ \end{lem} \textbf{Proof.} Assumptions of Lemma \ref{my_prodL2} are specifications of conditions of Lemma \ref{prodL2} for the masks $m^K_j$. Indeed, $$ \|(m^K_j)''\|^2_2 = \int_0^1 |(m^K_j)''(\xi)|^2 \, d\xi = 2 \left(\int_0^{1/4} (\cos''(z^K_j(\xi)))^2 \, d\xi + \int_{1/4}^{1/2} (\sin''(z^K_j(1/2 - \xi)))^2 \, d\xi \right) $$ $$ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ = 2\left(\int_0^{1/4} ((z^K_j)'(\xi))^4 + ((z^K_j)''(\xi))^2 \, d\xi \right) \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \Diamond $$ \subsection{Adjustment of localization} The following theorem describes a connection between $UC_H$ and $UC_B$ in nonstationary case. For stationary setup, this theorem is proved in \cite{prqurase03}. \begin{teo} \label{nstPQRS} Let $\Psi^P =\{\varphi^{P}_{0}, \, \psi^{P}_{j,k} \}_{j\in \mathbb{Z}_+,\, k=0,\dots,2^j-1}$ and $\Psi^N = \{\varphi^{N}_{0,k}, \, \psi^{N}_{j,k} \}_{j\in \mathbb{Z}_+,\, k \in \mathbb{Z}}$ be periodic and nonstationary Parseval wavelet frames, and $$ \psi^P_j(x):=\sum_{n\in\mathbb{Z}} \psi^N_j (x-n). $$ If there exist functions $f,\, f_1\in L_2(\mathbb{R})$ such that $|2^{-j/2}\psi^{N}_j(2^{-j} x)|\leq f(x),$ $|(2^{-j/2}\psi^{N}_j(2^{-j} x))'(x)|\leq f_1(x),$ and $f\in AC_{loc}(\mathbb{R})$, $f(x)=O(|x|^{-3/2-\varepsilon}),$ $f_1(x)=O(|x|^{-1-\varepsilon})$ as $x\to \infty,$ $\varepsilon > 0$, then $$ \lim_{j\to \infty} UC_B(\psi^P_j) = \lim_{j\to \infty}UC_H(\psi^N_j). $$ \end{teo} We omit the proof of the theorem since it can be straightforwardly checked that all the steps of the proof of Theorem 3 \cite{prqurase03} holds true for nonstationary case under the assumptions of Theorem \ref{nstPQRS}. It is not convenient to apply Theorem \ref{nstPQRS}. Indeed, our stating point is a periodic wavelet system, however the main condition (existence of majorants $f$, $f_1$) concerns the resulting nonstationary system. The next theorem is free of this drawback and provides sufficient conditions for an adjustment of localization in terms of initial periodic masks. \begin{teo} \label{adjust_loc} Let $\Psi^P =\{\varphi^{P}_{0}, \, \psi^{P}_{j,k} \}_{j\in \mathbb{Z}_+,\, k=0,\dots,2^j-1}$ be a periodic Parseval wavelet frame, $(\mu^j_k)_k=(2^{1/2} \nu^j_k)_k$ be its scaling masks. Let $(\nu^j_k)_k$ satisfy the conditions of Lemma \ref{def_mask}. By definition, put $\theta^j_k:=\arccos \nu^j_k,$ $\overline{\nu}^j_k:=\max\{\nu^j_k,\,\nu^j_{k+1}\}$, $\underline{\nu}^j_k:=\min\{\nu^j_k,\,\nu^j_{k+1}\}$. If \begin{enumerate} \item the series $\sum_{k\in\z}|k b^j_k|$ uniformly converges and uniformly bounded with respect to $j$ , where $b^j_k:=\prod_{r=1}^{\infty} \overline{\nu}^{j+r}_k$, and \item $ |\theta^j_{k+1}-\theta^j_{k}|\leq C 2^{-j}, $ where $C$ is an absolute constant, \end{enumerate} \noindent then the system $\Psi^N = \{\varphi^{N}_{0,k}, \, \psi^{N}_{j,k} \}_{j\in \mathbb{Z}_+,\, k \in \mathbb{Z}}$ with scaling masks $m^1_j(\xi)$ defined in Lemma \ref{def_mask} as $K=1$ forms a nonstationary Parseval wavelet frame, and $ \lim_{j\to \infty} UC_B(\psi^P_j) = \lim_{j\to \infty}UC_H(\psi^N_j). $ \end{teo} \begin{rem} Convergence of the infinite products $a^j_k:=\prod_{r=1}^{\infty} \underline{\nu}^{j+r}_k$ and $b^j_k:=\prod_{r=1}^{\infty} \overline{\nu}^{j+r}_k$ follows from the conditions of Theorem \ref{adjust_loc}. \end{rem} \textbf{Proof.} We check that under assumptions 1. and 2. all the conditions of Theorem \ref{nstPQRS} are fulfilled. First of all, the infinite products $a^j_k=\prod_{r=1}^{\infty} \underline{\nu}^{j+r}_k$ and $b^j_k=\prod_{r=1}^{\infty} \overline{\nu}^{j+r}_k$ converges for any $j\in\n$ and $k\in\z.$ Infinite products are considered convergent also in the case when they are equal to zero. Indeed, it follows from (\ref{con2}) and definition of $\overline{\nu}^{j}_k$ that $1 \geq \overline{\nu}^{j}_k \geq \nu^j_k \geq 0$. Therefore, $\bigl|\log \overline{\nu}^{j}_k \bigr| \leq \bigl|\log \nu^j_k \bigr|$, and we know that the series $\sum_{r=1}^{\infty} \log \nu^j_k$ is absolutely convergent or equal to $-\infty$. So, the series $\sum_{r=1}^{\infty} \log \overline{\nu}^{j}_k$ is absolutely convergent or equal to $-\infty$, hence, the product $\prod_{r=1}^{\infty} \overline{\nu}^{j+r}_k$ is absolutely convergent. Concerning to the infinite product $a^j_k=\prod_{r=1}^{\infty} \underline{\nu}^{j+r}_k$, we note that since the product $\prod_{r=1}^{\infty} \nu^{j+r}_k$ is absolutely convergent, the series $\sum_{r=1}^{\infty} (\nu^{j+r}_k - 1)$ has the same property. Therefore, using 2. we get $$ \sum_{r=1}^{\infty} |\underline{\nu}^{j+r}_k - 1|= \sum_{r=1}^{\infty} (1-\underline{\nu}^{j+r}_k)= \sum_{r=1}^{\infty} (1 - \nu^{j+r}_k) + \sum_{r=1}^{\infty} (\nu^{j+r}_k - \underline{\nu}^{j+r}_k) \leq \sum_{r=1}^{\infty} (1 - \nu^{j+r}_k) + C \sum_{r=1}^{\infty} \frac{1}{2^{j+r}}, $$ so the series $\sum_{r=1}^{\infty} (\underline{\nu}^{j+r}_k - 1)$ and, hence, the product $\prod_{r=1}^{\infty} \underline{\nu}^{j+r}_k$ are absolutely convergent. The next step is to check that the infinite product $\prod_{r=1}^{\infty} m^1_{j+r}(\xi/2^r) = \widehat{\varphi_j}(\xi)$ is uniformly convergent on any interval $[a,\,b]\in\r$ and $\varphi_j \in L_2(\r),$ where the auxiliary mask $m^1_{j}$ is defined in Lemma \ref{def_mask}. Lemma \ref{my_prodL2} can not help us here since it works as $K\geq 2$, and we consider the case $K=1.$ We check aforementioned properties directly. It follows from an elementary property of $\cos$ that the mask $m^1_{j}$ can be rewritten as $ m^1_j(\xi) = \cos (z^1_j(\xi)).$ The function $z^1_j$ is piecewise linear, namely, $ z^1_j(\xi) = 2^j(\theta^j_{k+1}-\theta^j_{k})(\xi - k 2^{-j}) + \theta^j_{k}, $ $\xi \in [k 2^{-j},\, (k+1) 2^{-j}].$ Therefore, $z^1_j(\xi)$ lies between $\theta^j_{k}$ and $\theta^j_{k+1}$ for $\xi \in [k 2^{-j},\, (k+1) 2^{-j}].$ Hence, $\underline{\nu}^{j}_k\leq m^1_j(\xi) \leq \overline{\nu}^{j}_k$ and \begin{equation} \label{est_phi} a^j_k\leq \widehat{\varphi_j}(\xi):= \prod_{r=1}^{\infty} m^1_{j+r}(\xi/2^r) \leq b^j_k \mbox{ for } \xi \in [k 2^{-j},\, (k+1) 2^{-j}]. \end{equation} So, the product $ \prod_{r=1}^{\infty} m^1_{j+r}(\xi/2^r) $ is everywhere finite. Using $0 \leq m^1_j(\xi)\leq 1$, we get $$ \left| \prod_{r=1}^{n} m^1_{j+r}(\xi/2^r) - \widehat{\varphi_j}(\xi) \right| \leq 1- \prod_{r=n+1}^{\infty} m^1_{j+r}(\xi/2^r) \leq 1-\prod_{r=n+1}^{\infty} \underline{\nu}^{j+r}_k \to 0 \mbox{ as } n \to \infty. $$ Since any interval $[a,\,b]$ can be covered by a finite number of the intervals $[k 2^{-j},\,(k+1) 2^{-j}]$, the convergence is uniform on $[a,\,b]$. Using the inequality $0\leq\widehat{\varphi_j}(\xi) \leq b^j_k$ for $\xi \in [k 2^{-j},\, (k+1) 2^{-j}]$ and the following corollary from condition 1. $\sum_{k\in \z} (b^j_k)^2 < \infty $ we immediately get $\widehat{\varphi_j} \in L_2(\r),$ thus $\varphi_j \in L_2(\r).$ Now we claim that the system $\Psi^N = \{\varphi^{N}_{0,k}, \, \psi^{N}_{j,k} \}_{j\in \mathbb{Z}_+,\, k \in \mathbb{Z}}$ with the scaling masks $m^N_{0,j}(\xi) = 2^{1/2} m^1_j(\xi),$ where the auxiliary masks $m^1_j(\xi)$ are defined in Lemma \ref{def_mask} as $K=1$, forms a nonstationary Parseval wavelet frame. It follows from Theorem \ref{nUEP} and (\ref{aux_n}). Indeed, $\varphi_j \in L_2(\r)$ is already checked, conditions (\ref{ncon2}) and (\ref{ncon3}) are provided by Lemma \ref{def_mask} and a definition of $\widehat{\varphi_j}$ by means of the infinite product. Condition (\ref{ncon1}) is equivalent to $\widehat{\varphi_j}(\xi 2^{-j}) \to 1$ as $j \to \infty$ for a fixed $\xi \in \r.$ Taking into account convergence of the infinite products $a^j_k=\prod_{r=1}^{\infty} \underline{\nu}^{j+r}_k$ and $b^j_k=\prod_{r=1}^{\infty} \overline{\nu}^{j+r}_k$, we obtain $a^j_k \to 1$ and $b^j_k \to 1$ as $j \to \infty.$ Then (\ref{ncon1}) follows from inequality (\ref{est_phi}). Switching to property of adjustment of localization, first, we show that for $\xi \neq k 2^{-j}$ \begin{equation} \label{der1} \widehat{\varphi_j}'(\xi)=\sum_{q=1}^{\infty} g_{j,q}(\xi) \left( m^1_{j+q}(\xi/2^q) \right)', \end{equation} where $g_{j,q}(\xi):=\prod_{r=1, r\neq q}^{\infty} m^1_{j+r}(\xi/2^r)$ and \begin{equation} \label{der2} \widehat{\varphi_j}''(\xi)=\sum_{t=1}^{\infty} g_{j,t}'(\xi) \left( m^1_{j+t}(\xi/2^t) \right)'+ g_{j,t}(\xi) \left( m^1_{j+t}(\xi/2^t) \right)'' \end{equation} To prove (\ref{der1}), it is sufficient to check that $$ \sum_{q=1}^{n} \prod_{r=1, r\neq q}^{n} m^1_{j+r}(\xi/2^r) \left( m^1_{j+q}(\xi/2^q) \right)' - \sum_{q=1}^{\infty} g_{j,q}(\xi) \left( m^1_{j+q}(\xi/2^q) \right)' \to 0 $$ as $n \to \infty$ uniformly on any interval $[a,\,b].$ The last difference is equal to $$ \left[\sum_{q=1}^{n} \prod_{r=1, r\neq q}^{n} m^1_{j+r}(\xi/2^r) \left( m^1_{j+q}(\xi/2^q) \right)'\right]\left(1-\prod_{r=n+1}^{\infty} m^1_{j+r}(\xi/2^r)\right) $$ $$ - \sum_{q=n+1}^{\infty} g_{j,q}(\xi) \left( m^1_{j+q}(\xi/2^q) \right)'=: s_1(\xi)-s_2(\xi). $$ To estimate $s_1(\xi)$, we notice that $$ 0 \leq \prod_{r=1, r\neq q}^{n} m^1_{j+r}(\xi/2^r)\leq 1, $$ \begin{equation} \label{der_mask} \left( m^1_{j+q}(\xi/2^q) \right)' = - 2^j (\theta^{j+q}_{k+1}-\theta^{j+q}_{k}) \sin(z^1_{j+q}(\xi/2^q)) \end{equation} $$ 0\leq 1-\prod_{r=n+1}^{\infty} m^1_{j+r}(\xi/2^r)\leq 1-\prod_{r=n+1}^{\infty} \underline{\nu}^{j+r}_k. $$ Combining these estimates and condition 2., we get $$ |s_1(\xi)|\leq \sum_{q=1}^n 2^j \bigl|\theta^{j+q}_{k+1}-\theta^{j+q}_{k}\bigr| \left(1-\prod_{r=n+1}^{\infty} \underline{\nu}^{j+r}_k\right) \leq C \sum_{q=1}^n \frac{2^j}{2^{j+q}} \left(1-\prod_{r=n+1}^{\infty} \underline{\nu}^{j+r}_k\right) \leq C \left(1-\prod_{r=n+1}^{\infty} \underline{\nu}^{j+r}_k\right) $$ The last expression tends to zero as $n \to \infty.$ To estimate $s_2(\xi)$, we use consequently $g_{j,q} \leq 1$, (\ref{der_mask}), and condition 2., so, we have $$ |s_2(\xi)| \leq \sum_{q=n+1}^{\infty} 2^j \bigl|\theta^{j+q}_{k+1}-\theta^{j+q}_{k}\bigr|\leq C\sum_{q=n+1}^{\infty} \frac{2^j}{2^{j+q}} = \frac{C}{2^n}. $$ Thus (\ref{der1}) is proved. Switching to proof of (\ref{der2}) we notice that $$ g'_{j,t}(\xi)=\sum_{q=1}^{\infty}\prod_{r=1, r\neq q, r\neq t}^{n} m^1_{j+r}(\xi/2^r) \left( m^1_{j+q}(\xi/2^q) \right)' $$ (It can be checked analogously to (\ref{der1}).) So, using $0 \leq m^1_j \leq 1$, (\ref{der_mask}), and condition 2., we have $$ |g'_{j,t}(\xi)| \leq \sum_{q=1}^{\infty} \left|\left( m^1_{j+q}(\xi/2^q) \right)'\right| \leq C, $$ therefore, again by (\ref{der_mask}), and condition 2. \begin{equation} \label{est1} \sum_{t=n+1}^{\infty} \left|g_{j,t}'(\xi) \left( m^1_{j+t}(\xi/2^t) \right)'\right| \leq C \sum_{t=n+1}^{\infty} \left| \left( m^1_{j+t}(\xi/2^t) \right)'\right| \leq \frac{C^2}{2^n}. \end{equation} Analogously, $$ \sum_{t=n+1}^{\infty} \left|g_{j,t}(\xi) \left( m^1_{j+t}(\xi/2^t) \right)''\right| \leq \sum_{t=n+1}^{\infty} \left| \left( m^1_{j+t}(\xi/2^t) \right)''\right|\leq \frac{C^2}{2^{2n}} $$ Combining the last estimate and (\ref{est1}) we get (\ref{der2}). Uniform convergence of (\ref{der1}) and (\ref{der2}) implies continuity of $\widehat{\varphi_j}'$ and $\widehat{\varphi_j}''$ on the intervals $(k 2^{-j},\, (k+1) 2^{-j}).$ Now we are ready to provide majorants $f,\, f_1 \in L_2(\r)$ such that $|\psi_j(x)|\leq f(x),$ $|\psi'_j(x)|\leq f_1(x)$, $f(x) = O(|x|^{-2})$, $f_1(x) = O(|x|^{-2})$ as $|x| \to \infty,$ where $\psi_j(x)= 2^{-j/2}\psi^N_j(2^{-j}x)$ are auxiliary wavelet functions. First, we get majorants for scaling functions $\varphi_j.$ Using condition 1. and the inequality $|\widehat{\varphi_j}(\xi)|\leq b^j_k$ for $\xi \in [k 2^{-j},\,(k+1)2^{-j}]$ we get $\widehat{\varphi_j} \in L_1(\r),$ $\xi \widehat{\varphi_j}(\xi) \in L_1(\r).$ The product $\prod_{r=1}^{\infty} m^1_{j+r}(\xi/2^r)$ is uniformly convergent on any $[a,\,b],$ therefore, the function $\widehat{\varphi_j}$ is continuous on $\mathbb{R}$. Thus $$ \varphi_j(x)=\int_{\r}\widehat{\varphi_j}(\xi)e^{2\pi i \xi x}\, d\xi = \left. \frac{1}{2\pi i x} \widehat{\varphi_j}(\xi)e^{2\pi i \xi x}\right|_{\r}- \left. \frac{1}{(2\pi i x)^2} \sum_{k \in \z} \widehat{\varphi_j}'(\xi)e^{2\pi i \xi x}\right|^{\xi=\frac{k+1}{2^j}-0}_{\xi=\frac{k}{2^j}+0} $$ $$ + \frac{1}{(2\pi i x)^2} \int_{\r}\widehat{\varphi_j}''(\xi)e^{2\pi i \xi x}\, d\xi $$ and $$ \varphi'_j(x)=\int_{\r} 2 \pi i \xi \widehat{\varphi_j}(\xi)e^{2\pi i \xi x}\, d\xi = \left. \frac{1}{ x} \xi \widehat{\varphi_j}(\xi)e^{2\pi i \xi x}\right|_{\r}- \left. \frac{1}{2\pi i x^2} \sum_{k \in \z} \left(\xi \widehat{\varphi_j}(\xi) \right)'e^{2\pi i \xi x}\right|^{\xi=\frac{k+1}{2^j}-0}_{\xi=\frac{k}{2^j}+0} $$ $$ + \frac{1}{2\pi i x^2} \int_{\r}\left(\xi\widehat{\varphi_j}(\xi)\right)''e^{2\pi i \xi x}\, d\xi $$ Since $\widehat{\varphi_j} \in L_1(\r)$, $\xi \widehat{\varphi_j}(\xi) \in L_1(\r)$ and $\widehat{\varphi_j}'$ is continuous at any point $\xi \neq k 2^{-j}$, it follows that $\widehat{\varphi_j}(\xi) \to 0$ and $\xi \widehat{\varphi_j}(\xi) \to 0$ as $\xi \to \pm \infty.$ Therefore, $$ \left. \frac{1}{ x} \xi \widehat{\varphi_j}(\xi)e^{2\pi i \xi x}\right|_{\r}=0, \ \ \ \left. \frac{1}{ x} \xi \widehat{\varphi_j}(\xi)e^{2\pi i \xi x}\right|_{\r}=0. $$ Hence, \begin{equation} \label{phi_es} \varphi_j(x)=- \frac{1}{(2\pi i x)^2} \sum_{k \in \z} \left(\widehat{\varphi_j}'\left(\frac{k}{2^j}-0\right)-\widehat{\varphi_j}'\left(\frac{k}{2^j}+0\right)\right)e^{2\pi i \frac{k}{2^j} x} + \frac{1}{(2\pi i x)^2} \int_{\r}\widehat{\varphi_j}''(\xi)e^{2\pi i \xi x}\, d\xi \end{equation} \begin{equation} \label{phi1_es} \varphi'_j(x)= -\frac{1}{2\pi i x^2} \sum_{k \in \z}\frac{k}{2^j} \left(\widehat{\varphi_j}'\left(\frac{k}{2^j}-0\right)-\widehat{\varphi_j}'\left(\frac{k}{2^j}+0\right)\right)e^{2\pi i \frac{k}{2^j} x} \end{equation} $$ + \frac{1}{\pi i x^2} \int_{\r}\widehat{\varphi_j}'(\xi)e^{2\pi i \xi x}\, d\xi +\frac{1}{2\pi i x^2} \int_{\r}\xi\widehat{\varphi_j}''(\xi)e^{2\pi i \xi x}\, d\xi $$ We denote $$ A_1:=\left|\sum_{k \in \z} \left(\widehat{\varphi_j}'\left(\frac{k}{2^j}-0\right)-\widehat{\varphi_j}'\left(\frac{k}{2^j}+0\right)\right)e^{2\pi i \frac{k}{2^j} x}\right|, $$ $$ A_2:= \left|\sum_{k \in \z}\frac{k}{2^j} \left(\widehat{\varphi_j}'\left(\frac{k}{2^j}-0\right)-\widehat{\varphi_j}'\left(\frac{k}{2^j}+0\right)\right)e^{2\pi i \frac{k}{2^j} x}\right|, $$ $$ A_3:=\left|\int_{\r}\widehat{\varphi_j}'(\xi)e^{2\pi i \xi x}\, d\xi\right|, \ \ A_4:= \left|\int_{\r}\widehat{\varphi_j}''(\xi)e^{2\pi i \xi x}\, d\xi\right|, \ \ A_5:=\left|\int_{\r}\xi\widehat{\varphi_j}''(\xi)e^{2\pi i \xi x}\, d\xi\right|. $$ Let us prove that $A_n,$ $n=1,\dots,5$ are uniformly bounded with respect to $j\in\n$ and $x\in\r.$ First, we estimate $\widehat{\varphi_j}'$ and $\widehat{\varphi_j}''$. Let $\xi \in [k 2^{-j},\, (k+1) 2^{-j}].$ Recalling (\ref{der1}), we need to estimate $g_{j,q}(\xi)$ and $(m^1_{j+q}(\xi 2^{-q}))'$. Using $0\leq m^1_j(\xi)\leq 1$, definition of $\widehat{\varphi_j},$ and (\ref{est_phi}) we get $$ 0\leq g_{j,q}(\xi) \leq \prod_{r=q+1}^{\infty} m^1_{j+r}(\xi / 2^{r}) = \widehat{\varphi_{j+q}}(\xi) \leq b^{j+q}_k $$ By (\ref{der_mask}) and condition 2. $$ \left|(m^1_{j+q}(\xi 2^{-q}))'\right| \leq \left|2^j (\theta^{j+q}_{k+1}-\theta^{j+q}_{k})\right| \leq \frac{C}{2^q}. $$ Substituting estimates in (\ref{der1}), we obtain \begin{equation} \label{est_phi'} |\widehat{\varphi_j}'(\xi)|\leq C \sum_{q=1}^{\infty} b^{j+q}_{k} \frac{1}{2^q} \mbox{ for }\xi \in [k 2^{-j},\, (k+1) 2^{-j}]. \end{equation} Now we majorize $\widehat{\varphi_j}''$. We start with (\ref{der2}) and estimate $g'_{j,t}(\xi)$ and $(m^1_{j+t}(\xi 2^{-t}))''$. Analogously to (\ref{est_phi'}) we have $$ |g'_{j,t}(\xi)| \leq \sum_{q=1}^{\infty}\prod_{r=1, r\neq q, r\neq t}^{n} m^1_{j+r}(\xi/2^r) \left|\left( m^1_{j+q}(\xi/2^q) \right)'\right| \leq C \sum_{q=1}^{\infty} b^{j+q+t}_{k} \frac{1}{2^q} $$ By (\ref{der_mask}) and condition 2. $$ \left|(m^1_{j+t}(\xi 2^{-q}))''\right| \leq \left|2^j (\theta^{j+t}_{k+1}-\theta^{j+t}_{k})\right|^2 \leq \frac{C^2}{2^{2t}}. $$ Collecting all inequalities and substituting them in (\ref{der2}) we obtain \begin{equation} \label{est_phi''} \left|\widehat{\varphi_j}''(\xi)\right| \leq C^2 \sum_{t=1}^{\infty} \sum_{q=1}^{\infty} b^{j+q+t}_k \frac{1}{2^{q+t}} + C^2 \sum_{t=1}^{\infty} b^{j+t}_k \frac{1}{2^{2t}} \mbox{ for }\xi \in [k 2^{-j},\, (k+1) 2^{-j}]. \end{equation} Now we return to the proof of boundedness of $A_n$, $n=1,\dots,5.$ Using (\ref{est_phi'}), we get $$ A_1 \leq \sum_{k \in \z} \left|\widehat{\varphi_j}'\left(\frac{k}{2^j}-0\right)-\widehat{\varphi_j}'\left(\frac{k}{2^j}+0\right)\right| \leq \sum_{k \in \z} \sum_{t =1}^{\infty} b^{j+t}_k 2^j \frac{2C}{2^{j+t}}. $$ Since $ \sum_{k \in \z} b^{j}_k \leq b^j_0 + \sum_{k \in \z} |k| b^{j}_k \leq 1 + \sum_{k \in \z} |k| b^{j}_k, $ condition 1. make it possible to change the order of summation, thus we have $$ A_1 \leq 2C \sum_{t =1}^{\infty} \sum_{k \in \z} b^{j+t}_k \frac{1}{2^{t}} \leq 2C^2 \sum_{t =1}^{\infty}\frac{1}{2^{t}} \leq 2 C^2. $$ Analogously, for $A_2$ we get $$ A_2 \leq \sum_{k \in \z} \sum_{t =1}^{\infty} |k| b^{j+t}_k \frac{2C}{2^{j+t}} \leq \frac{2C^2}{2^j}. $$ Again, using (\ref{est_phi'}) we have $$ A_3 \leq \int_{\r} \left|\widehat{\varphi_j}'(\xi)\right|\, d\xi \leq \sum_{k \in \z}\int_{k 2^{-j}}^{(k+1) 2^{-j}} \left|\widehat{\varphi_j}'(\xi)\right|\, d\xi \leq C \sum_{k \in \z}\int_{k 2^{-j}}^{(k+1) 2^{-j}} \sum_{q=1}^{\infty} b^{j+q}_k \frac{1}{2^q} \, d\xi = C \frac{1}{2^j} \sum_{k \in \z} \sum_{q=1}^{\infty} b^{j+q}_k \frac{1}{2^q} $$ $$ = C \frac{1}{2^j} \sum_{q=1}^{\infty} \sum_{k \in \z} b^{j+q}_k \frac{1}{2^q} \leq \frac{C^2}{2^j}. $$ By (\ref{est_phi''}) and condition 1., $$ A_4 \leq \int_{\r} \left|\widehat{\varphi_j}''(\xi)\right|\, d\xi \leq \sum_{k \in \z}\int_{k 2^{-j}}^{(k+1) 2^{-j}} \left|\widehat{\varphi_j}''(\xi)\right|\, d\xi \leq C^2 \frac{1}{2^j} \sum_{k \in \z} \left(\sum_{t=1}^{\infty} \sum_{q=1}^{\infty} b^{j+q+t}_k \frac{1}{2^{q+t}} + \sum_{t=1}^{\infty} b^{j+t}_k \frac{1}{2^{2t}} \right) $$ $$ = C^2 \frac{1}{2^j} \left(\sum_{t=1}^{\infty} \sum_{q=1}^{\infty} \sum_{k \in \z} b^{j+q+t}_k \frac{1}{2^{q+t}} + \sum_{t=1}^{\infty} \sum_{k \in \z} b^{j+t}_k \frac{1}{2^{2t}} \right)\leq \frac{4 C^3}{3} \frac{1}{2^j}, $$ $$ A_5 \leq \int_{\r} \left|\xi \widehat{\varphi_j}''(\xi)\right|\, d\xi \leq \sum_{k \in \z}\int_{k 2^{-j}}^{(k+1) 2^{-j}} \left|\xi\widehat{\varphi_j}''(\xi)\right|\, d\xi \leq \frac{C^2}{2^j} \sum_{k \in \z} (|k|+1)\left(\sum_{t=1}^{\infty} \sum_{q=1}^{\infty} b^{j+q+t}_k \frac{1}{2^{q+t}} + \sum_{t=1}^{\infty} b^{j+t}_k \frac{1}{2^{2t}} \right) $$ $$ = C^2 \frac{1}{2^j} \left(\sum_{t=1}^{\infty} \sum_{q=1}^{\infty} \sum_{k \in \z}(|k|+1) b^{j+q+t}_k \frac{1}{2^{q+t}} + \sum_{t=1}^{\infty} \sum_{k \in \z} (|k|+1) b^{j+t}_k \frac{1}{2^{2t}} \right)\leq \frac{8 C^3}{3} \frac{1}{2^j}. $$ In all above estimates changing the order of summation is justified by condition 1. So, all the expressions $A_n,$ $n=1,\dots,5$ are bounded by an absolute constant. Thus, (\ref{phi_es}), (\ref{phi1_es}) yield $$ \varphi_j(x) \leq C |x|^{-2}, \ \ \ \varphi'_j(x) \leq C|x|^{-2}. $$ On the other hand, $ \varphi_j(x), \varphi'_j(x) $ are bounded by an absolute constant. Indeed, let $v=0$ or $v=1$. Using (\ref{est_phi}) and condition 1., we obtain $$ |\varphi^{(v)}_j(x)| \leq \int_{\r} \left| (2 \pi \xi)^v\widehat{\varphi_j}(\xi)\right| \, d\xi \leq \sum_{k \in \z} \int_{k 2^{-j}}^{(k+1) 2^{-j}}\left|(2 \pi \xi)^v\widehat{\varphi_j}(\xi)\right| \, d\xi \leq \sum_{k \in \z} \frac{\bigl(2 \pi (|k|+1)\bigr)^v}{2^j} b^j_k \leq C. $$ Thus, for functions $\varphi_j$ and $\varphi_j'$ we provide majorants of the form $$ \left\{ \begin{array}{cc} C, & |x|\leq 1; \\ C/|x|^2, & |x|\geq 1. \end{array} \right. $$ Majorants for auxiliary wavelet functions $\psi_j$ and its derivatives $\psi_j'$ can be obtained analogously. Indeed, one can start with the equalities analogous to the case of scaling functions $$ \psi_j(x)=\int_{\r}\widehat{\psi_j}(\xi)e^{2\pi i \xi x}\, d\xi = \left. \frac{1}{2\pi i x} \widehat{\psi_j}(\xi)e^{2\pi i \xi x}\right|_{\r}- \left. \frac{1}{(2\pi i x)^2} \sum_{k \in \z} \widehat{\psi_j}'(\xi)e^{2\pi i \xi x}\right|^{\xi=\frac{k+1}{2^j}-0}_{\xi=\frac{k}{2^j}+0} $$ $$ + \frac{1}{(2\pi i x)^2} \int_{\r}\widehat{\psi_j}''(\xi)e^{2\pi i \xi x}\, d\xi $$ and $$ \psi'_j(x)=\int_{\r} 2 \pi i \xi \widehat{\psi_j}(\xi)e^{2\pi i \xi x}\, d\xi = \left. \frac{1}{ x} \xi \widehat{\psi_j}(\xi)e^{2\pi i \xi x}\right|_{\r}- \left. \frac{1}{2\pi i x^2} \sum_{k \in \z} \left(\xi \widehat{\psi_j}(\xi) \right)'e^{2\pi i \xi x}\right|^{\xi=\frac{k+1}{2^j}-0}_{\xi=\frac{k}{2^j}+0} $$ $$ + \frac{1}{2\pi i x^2} \int_{\r}\left(\xi\widehat{\psi_j}(\xi)\right)''e^{2\pi i \xi x}\, d\xi. $$ Then using the definition of an auxiliary wavelet sequence $ \widehat{\psi_j}(\xi) = e^{\pi i \xi} \overline{m^1_{j+1}(\xi/2+1/2)} \widehat{\varphi_{j+1}}(\xi/2) $ and boundedness of $m^1_{j+1}(\xi/2+1/2),$ $(m^1_{j+1}(\xi/2+1/2))',$ and $(m^1_{j+1}(\xi/2+1/2))''$ we come to the case of scaling sequences. Theorem \ref{adjust_loc} is proved.\hfill $\Diamond$ Analyzing the assumptions of Theorem \ref{adjust_loc}, it is easy to see that condition 2. means the boundedness of the first divided difference $2^j (\theta^j_{k+1} - \theta^j_{k})$ for the data points $(k 2^{-j},\, \theta^j_{k}),$ $k\in \z.$ Unfortunately, Theorem \ref{adjust_loc} is not applicable to the Parseval periodic wavelet frame constructed in \cite{LebPres14}. More precisely, condition 2. is not fulfilled. Indeed, for $k=2^{j-2}$ we get $|\theta^j_{k+1}-\theta^j_{k}| \geq 1/\sqrt{2} - \varepsilon_j,$ where $\varepsilon_j \to 0$ as $j\to \infty.$ Moreover, it follows from the last inequality that for any scaling mask $m^N_j$ of a nonstationary wavelet system that corresponds to the periodic wavelet frame constructed in \cite{LebPres14} there exists a point $\overline{\xi_j}$ such that $|(m^N_j)'(\overline{\xi_j})|\geq C 2^j$. To construct a periodic wavelet sequence $\psi^P_j$ satisfying assumptions of Theorem \ref{adjust_loc} and the equality $\lim_{j\to \infty} UC_B(\psi^P_j) =3/2$ is a task for future investigation. Finally, it is interesting to note that we can always construct a trivial nonstationary wavelet frame starting with a periodic one. Suppose $\Psi^P =\{\varphi^{P}_{0}, \, \psi^{P}_{j,k} \}_{j\in \mathbb{Z}_+,\, k=0,\dots,2^j-1}$ is a periodic Parseval wavelet frame, $\mu^j_k=2^{1/2} \nu^j_k$ is scaling mask of this frame. Let us define a nonstationary scaling mask as a step function such that $$ m^N_j(\xi)=2^{1/2} m_j(\xi)=\mu^j_k, \quad \xi\in[k/2^k,\,(k+1)/2^j). $$ Then a nonstationary scaling function is the following step function $$ \widehat{\varphi^N_j}(\xi) = 2^{-j/2}\prod_{r=1}^{\infty} m_{j+r}(\xi/2^{j+r}) = \widehat{\varphi^P_j}(k), \ \ \ \xi\in[k/2^k,\,(k+1)/2^j). $$ Since $ \|\widehat{\varphi^N_j}\|^2_2 = \|\varphi^P_j\|^2_2$, the scaling function $\varphi^N_j$ is in $L_2(\mathbb{R}).$ It generates a nonstationary Parseval wavelet frame. However, since $\widehat{\varphi^N_j}$ is discontinuous, $\varphi^N_j$ has poor time localization. \section*{Acknowledgments} We great fully acknowledge the funding by the RFBR, grant \#15-01-05796, by Saint Petersburg State University, grant \#9.38.198.2015, and by Volkswagen Foundation.
252379e632a62c2e5e4d2d4b4a5b2d2273214ec3
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} In this paper we consider the problem of finding the number of representations of the following quadratic forms in eight variables given by \begin{equation}\label{quad1} a_1x_1^2 + a_2 x_2^2 + a_3 x_3^2 + a_4 x_4^2 + b_1(x_5^2+x_5x_6 + x_6^2) + b_2(x_7^2+x_7x_8 + x_8^2), \end{equation} where the coefficients $a_i \in \{1,2,3\}$, $1\le i\le 4$ and $b_1,b_2\in \{1,2,4\}$. Without loss of generality we can assume that $a_1\le a_2\le a_3\le a_4$ and $b_1\le b_2$. In \cite{a-a-l-w}, A. Alaca et. \!\!al considered similar types of quadratic forms in four variables, which are either sums of four squares with coefficients $1,2,3,4$ or $6$ ($7$ such forms) or direct sum of the sums of two squares with coefficients $1$ or $3$ and the quadratic form $x^2+xy+y^2$ with coefficients $1, 2$ or $4$ ($6$ such forms). They used theta function identities to determine the representation formulas for these $13$ quadratic forms. In our recent work \cite{r-s-s}, we constructed bases for the space of modular forms of weight $4$ for the group $\Gamma_0(48)$ with character, and used modular forms techniques to determine the number of representations of a natural number $n$ by certain octonary quadratic forms with coefficients $1,2,3,4,6$. Finding formulas for the number of representations for octonary quadratic forms with coefficients $1,2,3$ or $6$ were considered by various authors using several methods (see for example \cite{{a-a-w},{a-a-w1},{a-k-new},{a-k1},{a-k2},{a-w}}). In the present work, we adopt similar (modular forms) techniques to obtain the representation formulas. We show directly that the theta series corresponding to each of the quadratic form considered belongs to the space of modular forms of weight $4$ on $\Gamma_0(24)$ with some character (depending on the coefficients). Now, by constructing a basis for the space of modular forms $M_4(\Gamma_0(24),\chi)$ we find the required formulas. Here $\chi$ is either the trivial Dirichlet character modulo $24$ or one of the primitive Dirichlet characters (modulo $m$) $\chi_m = \left(\frac{m}{\cdot}\right)$, $m=8, 12, 24$. Since $M_4(\Gamma_0(24),\chi) \subseteq M_4(\Gamma_0(48),\chi)$, where $\chi$ is a Dirichlet character modulo $24$, we get the required explict bases from the basis of modular forms $M_4(\Gamma_0(48),\psi)$, where $\psi$ is a Dirichlet character modulo $48$, which was constructed in \cite{r-s-s}. In the second part of the paper, we consider the quadratic forms of eight variables given by: \begin{equation}\label{quad2} (x_1^2+x_1x_2+x_2^2) + c_1(x_3^2+x_3x_4+x_4^2) + c_2(x_5^2+x_5x_6+x_6^2) + c_3(x_7^2+x_7x_8+x_8^2), \end{equation} where $c_1\le c_2\le c_3$ and $c_1,c_2,c_3\in \{1,2,4,8\}$. We note that for the $c_i$'s in the list, each of the quadratic form represents a theta series which belong to the space $M_4(\Gamma_0(24))$. Therefore, using our methods adopted for the earlier case, we also determine explicit formulas for the number of representations of a natural number by these class of quadratic forms. The total number of such quadratic forms given by \eqref{quad1} with coefficients $a_i\in \{1,2,3\}$ \linebreak and $b_i\in \{1,2,4\}$ is $90$. Each quadratic form in this list is denoted as a sextuple \linebreak $(a_1,a_2,a_3,a_4, b_1,b_2)$ and we list them in Table 1. We also put them in four classes corresponding to each of the modular forms space $M_4(\Gamma_0(24), \chi)$. Similarly, we list the quadratic forms (total 19) given by \eqref{quad2} in Table 2. In this case all the corresponding theta series belong to $M_4(\Gamma_0(24))$. We do not consider the case $(1,1,1,1)$ as the formula is already known (see \cite[Theorem 17.4]{williams-book}). It was shown that $s_8(n) = 24 \sigma_3(n) + 216 \sigma_3(n/3)$. In our notation (see \S 3) $s_8(n) = M(1,1,1,1;n)$. Also, the cases $(1,2,2,4)$ and $(1,2,4,8)$ has been proved in \cite{kokluce} by using convolution sums method. The paper is organized as follows. In \S 2 we present the theorems proved in this article and in \ \S 3 we give some preliminary results which are needed in proving the theorems. In \S 4 we give a proof of our theorems using the theory of modular forms. \bigskip \bigskip \begin{center} \textbf{Table 1.\\} List of quadratic forms in 8 variables given in \eqref{quad1}\\ \begin{tabular}{|c|c|} \hline $(a_1,a_2,a_3,a_4,b_1,b_2)$ & space \\ \hline $(1,1,1,1,1,1),(1,1,1,1,1,2),(1,1,1,1,1,4),(1,1,1,1,2,2)$&~~\\ $(1,1,1,1,2,4),(1,1,1,1,4,4),(1,1,2,2,1,1),(1,1,2,2,1,2)$&~~\\ $(1,1,2,2,1,4),(1,1,2,2,2,2),(1,1,2,2,2,4),(1,1,2,2,4,4)$&~~\\ $(1,1,3,3,1,1),(1,1,3,3,1,2),(1,1,3,3,1,4),(1,1,3,3,2,2)$&~~\\ $(1,1,3,3,2,4),(1,1,3,3,4,4),(2,2,2,2,1,1),(2,2,2,2,1,2)$&$M_4(\Gamma_0(24))$\\ $(2,2,2,2,1,4),(2,2,2,2,2,2),(2,2,2,2,2,4),(2,2,2,2,4,4)$&~~\\ $(2,2,3,3,1,1),(2,2,3,3,1,2),(2,2,3,3,1,4),(2,2,3,3,2,2)$&~~\\ $(2,2,3,3,2,4),(2,2,3,3,4,4),(3,3,3,3,1,1),(3,3,3,3,1,2)$&~~\\ $(3,3,3,3,1,4),(3,3,3,3,2,2),(3,3,3,3,2,4),(3,3,3,3,4,4)$&~~\\ \hline $(1,1,1,2,1,1),(1,1,1,2,1,2),(1,1,1,2,1,4),(1,1,1,2,2,2)$&~~\\ $(1,1,1,2,2,4),(1,1,1,2,4,4),(1,2,2,2,1,1,(1,2,2,2,1,2)$&~~\\ $(1,2,2,2,1,4),(1,2,2,2,2,2),(1,2,2,2,2,4),(1,2,2,2,4,4)$&$M_4(\Gamma_0(24),\chi_8)$\\ $(1,2,3,3,1,1),(1,2,3,3,1,2),(1,2,3,3,1,4),(1,2,3,3,2,2)$&~~\\ $(1,2,3,3,2,4),(1,2,3,3,4,4)$&~~\\ \hline $(1,1,1,3,1,1),(1,1,1,3,1,2),(1,1,1,3,1,4),(1,1,1,3,2,2)$&~~\\ $(1,1,1,3,2,4),(1,1,1,3,4,4),(1,2,2,2,1,1,(1,2,2,3,1,2)$&~~\\ $(1,2,2,3,1,4),(1,2,2,3,2,2),(1,2,2,3,2,4),(1,2,2,3,4,4)$&$M_4(\Gamma_0(24),\chi_{12})$\\ $(1,3,3,3,1,1),(1,3,3,3,1,2),(1,3,3,3,1,4),(1,3,3,3,2,2)$&~~\\ $(1,3,3,3,2,4),(1,3,3,3,4,4)$&~~\\ \hline $(1,1,2,3,1,1),(1,1,2,3,1,2),(1,1,2,3,1,4),(1,1,2,3,2,2)$&~~\\ $(1,1,2,3,2,4),(1,1,2,3,4,4),(2,2,2,3,1,1),(2,2,2,3,1,2)$&~~\\ $(2,2,2,3,1,4),(2,2,2,3,2,2),(2,2,2,3,2,4),(2,2,2,3,4,4)$&$M_4(\Gamma_0(24),\chi_{24})$\\ $(2,3,3,3,1,1),(2,3,3,3,1,2),(2,3,3,3,1,4),(2,3,3,3,2,2)$&~~\\ $(2,3,3,3,2,4),(2,3,3,3,4,4)$&~~\\ \hline \end{tabular} \end{center} \bigskip \newpage \begin{center} \textbf{Table 2.} \vskip 0.2 cm List of quadratic forms in \eqref{quad2} indicated by $(1,c_1,c_2,c_3)$.\\ \begin{tabular}{|c|c|} \hline $(1,c_1,c_2,c_3)$& space\\ \hline $(1,1,1,2),(1,1,1,4),(1,1,1,8),(1,1,2,2),(1,1,2,4),(1,1,2,8),(1,1,4,4)$&~~\\ $(1,1,4,8),(1,1,8,8),(1,2,2,2),(1,2,2,4),(1,2,2,8),(1,2,4,4),(1,2,4,8)$&$M_4(\Gamma_0(24))$\\ $(1,2,8,8),(1,4,4,4),(1,4,4,8),(1,4,8,8),(1,8,8,8)$&~~\\ \hline \end{tabular} \end{center} \section{Statement of results} Let ${\mathbb N}, {\mathbb N}_0$ and ${\mathbb Z}$ denote the set of positive integers, non-negative integers and integers respectively. For $(a_1, a_2,a_3, a_4,b_1,b_2)$ as in Table 1, we define \begin{equation*} \begin{split} N(a_1,a_2,a_3,a_4,b_1,b_2;n) := \qquad \hskip 9.7cm &\\ \quad \# \left\{(x_1,\ldots, x_8)\in {\mathbb Z}^8 \big\vert n = \sum_{i=1}^4a_ix_i^2 + b_1(x_5^2+x_5x_6 + x_6^2) + b_2(x_7^2+x_7x_8 + x_8^2) \right\}.\\ \end{split} \end{equation*} to be the number of representations of $n$ by the quadratic form \eqref{quad1}. Note that \linebreak $N(a_1,a_2,a_3,a_4,b_1,b_2;0) =1$. The formulas corresponding to Table 1 are stated in the following theorem. Formulas are divided into four parts each corresponding to one of the four spaces of modular forms $M_4(\Gamma_0(24),\chi)$. \begin{thm}\label{1} Let $n\in {\mathbb N}$. \\ {\rm (i)} For each entry $(a_1,a_2,a_3,a_4,b_1,b_2)$ in Table {\rm 1} corresponding to the space $M_4(\Gamma_0(24))$, we have \begin{equation} N(a_1,a_2,a_3,a_4,b_1,b_2;n) = \sum_{i=1}^{16} \alpha_i A_i(n), \end{equation} where $A_i(n)$ are the Fourier coefficients of the basis elements $f_i$ defined in \S {\rm 4.1} and the values of the constants $\alpha_i$s are given in Table {\rm 3}. \\ {\rm (ii)} For each entry $(a_1,a_2,a_3,a_4,b_1,b_2)$ in Table {\rm 1} corresponding to the space $M_4(\Gamma_0(24),\chi_8)$, we have \begin{equation} N(a_1,a_2,a_3,a_4,b_1,b_2;n) = \sum_{i=1}^{14} \beta_i B_i(n), \end{equation} where $B_i(n)$ are the Fourier coefficients of the basis elements $g_i$ defined in \S {\rm 4.2} and the values of the constants $\beta_i$'s are given in Table {\rm 4}. \\ {\rm (iii)} For each entry $(a_1,a_2,a_3,a_4,b_1,b_2)$ in Table {\rm 1} corresponding to the space $M_4(\Gamma_0(24), \chi_{12})$, we have \begin{equation} N(a_1,a_2,a_3,a_4,b_1,b_2;n) = \sum_{i=1}^{16} \gamma_i C_i(n), \end{equation} where $C_i(n)$ are the Fourier coefficients of the basis elements $h_i$ defined in \S {\rm 4.3} and the values of the constants $\gamma_i$'s are given in Table {\rm 5}. \\ {\rm (iv)} For each entry $(a_1,a_2,a_3,a_4,b_1,b_2)$ in Table {\rm 1} corresponding to the space $M_4(\Gamma_0(24), \chi_{24})$, we have \begin{equation} N(a_1,a_2,a_3,a_4,b_1,b_2;n) = \sum_{i=1}^{14} \delta_i D_i(n), \end{equation} where $D_i(n)$ are the Fourier coefficients of the basis elements $F_i$ defined in \S {\rm 4.4} and the values of the constants $\delta_i$'s are given in Table {\rm 6}. \end{thm} \bigskip \noindent Now we consider the class of quadratic forms given by \eqref{quad2}. For $(1, c_1,c_2,c_3)$ as in Table 2, we define \begin{equation*} \begin{split} M(1, c_1,c_2,c_3;n) := \# \left\{(x_1,\ldots, x_8)\in {\mathbb Z}^8 \big\vert \right. \hskip 8cm &\\ \left. n = \!(x_1^2+x_1x_2+x_2^2) + \!c_1(x_3^2+x_3x_4 + x_4^2) + \!c_2(x_5^2+x_5x_6 + x_6^2) + \!c_3(x_7^2+x_7x_8+x_8^2)\right\}.&\\ \end{split} \end{equation*} to be the number of representations of $n$ by the quadratic form \eqref{quad2}. Note that \linebreak $M(1,c_1,c_2,c_3;0) =1$. The formulas corresponding to Table 2 are stated in the following theorem. \begin{thm}\label{2} Let $n\in {\mathbb N}$. \\ For each entry $(1,c_1, c_2,c_3;n)$ in Table {\rm 2}, we have \begin{equation} M(1, c_1,c_2,c_3;n) = \sum_{i=1}^{16} \nu_i A_i(n), \end{equation} where $A_i(n)$ are the Fourier coefficients of the basis elements $f_i$ defined in \S {\rm 4.1} and the values of the constants $\nu_i$s are given in Table {\rm 7}. \\ \end{thm} \smallskip \noindent {\bf Remark 2.1.} Since one can write down the exact formulas using the explicit Fourier coefficients of the basis elements and using the coefficients tables given in each of the cases, we have not stated explicit formulas in the theorems (due to large number of such formulas). However, in \S 5 (at the end of the Tables), we give some sample formulas corresponding to each case. \smallskip \section{Preliminaries} In this section we present some preliminary facts on modular forms. For $k\in \frac{1}{2}{\mathbb Z}$, let $M_k(\Gamma_0(N),\chi)$ denote the space of modular forms of weight $k$ for the congruence subgroup $\Gamma_0(N)$ with character $\chi$ and $S_k(\Gamma_0(N), \chi)$ be the subspace of cusp forms of weight $k$ for $\Gamma_0(N)$ with character $\chi$. We assume $4\vert N$ when $k$ is not an integer and in that case, the character $\chi$ (which is a Dirichlet character modulo $N$) is an even character. When $\chi$ is the trivial (principal) character modulo $N$, we shall denote the spaces by $M_k(\Gamma_0(N))$ and $S_k(\Gamma_0(N))$ respectively. Further, when $k\ge 4$ is an integer and $N=1$, we shall denote these vector spaces by $M_k$ and $S_k$ respectively. For an integer $k \ge 4,$ let $E_k$ denote the normalized Eisenstein series of weight $k$ in $M_k$ given by $$ E_k(z) = 1 - \frac{2k}{B_k}\sum_{n\ge 1} \sigma_{k-1}(n) q^n, $$ where $q=e^{2 i\pi z}$, $\sigma_r(n)$ is the sum of the $r$th powers of the positive divisors of $n$, and $B_k$ is the $k$-th Bernoulli number defined by $\displaystyle{\frac{x}{e^x-1} = \sum_{m=0}^\infty \frac{B_m}{m!} x^m}$. The classical theta function which is fundamental to the theory of modular forms of half-integral weight is defined by \begin{equation}\label{theta} \Theta(z) = \sum_{n\in {\mathbb Z}} q^{n^2}, \end{equation} and is a modular form in the space $M_{1/2}(\Gamma_0(4))$. Another function which is mainly used in our work is the Dedekind eta function $\eta(z)$ and it is given by \begin{equation}\label{eta} \eta(z)=q^{1/24} \prod_{n\ge1}(1-q^n). \end{equation} An eta-quotient is a finite product of integer powers of $\eta(z)$ and we denote it as follows: \begin{equation}\label{eta-q} \prod_{i=1}^s \eta^{r_i}(d_i z) := d_1^{r_1} d_2^{r_2} \cdots d_s^{r_s}, \end{equation} where $d_i$'s are positive integers and $r_i$'s are non-zero integers. We denote the theta series associated to the quadratic form $x^2+xy+y^2$ by \begin{equation}\label{F1} {\mathcal F}(z) = \sum_{x,y\in {\mathbb Z}} q^{x^2+xy+y^2}. \end{equation} This function is referred to as the Borweins' two dimensional theta function in the literature. By \cite[Theorem 4]{schoeneberg}, it follows that ${\mathcal F}(z)$ is a modular form in $M_1(\Gamma_0(3),\chi_{-3})$. Here and in the sequel, for $m<0$, the character $\chi_m$ is the odd Dirichlet character modulo $|m|$ given by $\left(\frac{-m}{\cdot}\right)$. In the following we shall present some facts about modular forms of integral and half-integral weights, which we shall be using in our proof. We state them as lemmas, whose proofs follow from elementary theory of modular forms (of integral and half-integral weights). \noindent {\bf Lemma 1.} {\em (Duplication of modular forms)\\ If $f$ is a modular form in $M_k(\Gamma_0(N), \chi)$, then for a positive integer $d$, the function $f(dz)$ is a modular form in $M_k(\Gamma_0(dN), \chi)$, if $k$ is an integer and it belongs to the space $M_k(\Gamma_0(dN), \chi \chi_d)$, if $k$ is a half-integer. } \smallskip \noindent {\bf Lemma 2.} {\em For positive integers $r$, $r_1$, $r_2$, $d_1$, $d_2$, we have \begin{equation} \Theta^{r}(d_1z) \in \begin{cases} M_{r/2}(\Gamma_0(4d_1), \chi_{d_1}) & {\rm ~if~} r {\rm ~is~odd}, \\ M_{r/2}(\Gamma_0(4d_1), \chi_{-4}) & {\rm ~if~} r \equiv 2\pmod{4},\\ M_{r/2}(\Gamma_0(4d_1)) & {\rm ~if~} r \equiv 0\pmod{4}. \end{cases} \end{equation} \begin{equation} \Theta^{r_1}(d_1z)\cdot \Theta^{r_2}(d_2z) \in \begin{cases} M_{\frac{r_1+r_2}{2}}(\Gamma_0(4 [d_1, d_2]), \chi_{(-d_1d_2)}) & \!\!{\rm ~if~} r_1 r_2 {\rm ~is~odd~}, r_1+r_2\equiv 2\!\!\!\!\!\pmod{4}, \\ M_{\frac{r_1+r_2}{2}}(\Gamma_0(4 [d_1, d_2]), \chi_{(d_1d_2)}) & \!\!{\rm ~if~} r_1 r_2 {\rm ~is~odd~}, r_1+r_2 \equiv 0\!\!\!\!\!\pmod{4}. \end{cases} \end{equation} } \smallskip \noindent {\bf Lemma 3.} {\em If $f_i \in M_{k_i}(\Gamma_0(M_i), \psi_i)$, $i=1,2$, then the product $f_1\cdot f_2$ is a modular form in $M_{k_1+k_2}(\Gamma_0(M), \psi_1\psi_2)$, where $M= {\rm lcm}(M_1,M_2)$. } \smallskip \noindent {\bf Lemma 4.} {\em The vector space $M_k(\Gamma_1(N))$ is decomposed as a direct sum: \begin{equation} M_k(\Gamma_1(N)) = \oplus_{\chi}M_k(\Gamma_0(N), \chi), \end{equation} where the direct sum varies over all Dirichlet characters modulo $N$ if the weight $k$ is a positive integer and varies over all even Dirichlet characters modulo $N$, $4\vert N$, if the weight $k$ is half-integer. Further, if $k$ is an integer, one has $M_k(\Gamma_0(N),\chi) = \{0\}$, if $\chi(-1) \not= (-1)^k$. We also have the following decomposition of the space into subspaces of Eisenstein series and cusp forms: \begin{equation} M_k(\Gamma_0(N),\chi) = {\mathcal E}_k(\Gamma_0(N),\chi) \oplus S_k(\Gamma_0(N), \chi), \end{equation} where ${\mathcal E}_k(\Gamma_0(N),\chi)$ is the space generated by the Eisenstein series of weight $k$ on $\Gamma_0(N)$ with character $\chi$. } \smallskip \noindent {\bf Lemma 5.} {\em By the Atkin-Lehner theory of newforms, the space $S_k(\Gamma_0(N),\chi)$ can be decomposed into the space of newforms and oldforms: \begin{equation} S_k(\Gamma_0(N),\chi) = S_k^{new}(\Gamma_0(N),\chi) \oplus S_k^{old}(\Gamma_0(N),\chi), \end{equation} where the above is an orthogonal direct sum (with respect to the Petersson scalar product) and \begin{equation} S_k^{old}(\Gamma_0(N), \chi) = \bigoplus_{r\vert N, r<N\atop{rd\vert N}}S_k^{new}(\Gamma_0(r),\chi)\vert B(d). \end{equation} In the above, $S_k^{new}(\Gamma_0(N),\chi)$ is the space of newforms and $S_k^{old}(\Gamma_0(N),\chi)$ is the space of oldforms and the operator $B(d)$ is given by $f(z) \mapsto f(dz)$. } \smallskip \noindent {\bf Lemma 6.} {\em Suppose that $\chi$ and $\psi$ are primitive Dirichlet characters with conductors $M$ and $N$, respectively. For a positive integer $k$, let \begin{equation}\label{eisenstein} E_{k,\chi,\psi}(z) := c_0 + \sum_{n\ge 1}\left(\sum_{d\vert n} \psi(d) \cdot \chi(n/d) d^{k-1}\right) q^n, \end{equation} where $$ c_0 = \begin{cases} 0 &{\rm ~if~} M>1,\\ - \frac{B_{k,\psi}}{2k} & {\rm ~if~} M=1, \end{cases} $$ and $B_{k,\psi}$ denotes generalized Bernoulli number with respect to the character $\psi$. Then, the Eisenstein series $E_{k,\chi,\psi}(z)$ belongs to the space $M_k(\Gamma_0(MN), \chi/\psi)$, provided $\chi(-1)\psi(-1) = (-1)^k$ and $MN\not=1$. When $\chi=\psi =1$ (i.e., when $M=N=1$) and $k\ge 4$, we have $E_{k,\chi,\psi}(z) = - \frac{B_k}{2k} E_k(z)$, where $E_k$ is the normalized Eisenstein series of integer weight $k$ as defined before. We refer to \cite{{miyake}, {stein}} for details. } \smallskip We give a notation to the inner sum in \eqref{eisenstein}: \begin{equation}\label{divisor} \sigma_{k-1;\chi,\psi}(n) := \sum_{d\vert n} \psi(d) \cdot \chi(n/d) d^{k-1}. \end{equation} \smallskip For more details on the theory of modular forms of integral and half-integral weights, we refer to \cite{{a-l}, {koblitz}, {li}, {miyake}, {schoeneberg}, {shimura}}. \smallskip \section{proofs of theorems} In this section, we shall give a proof of our results. As mentioned in the introduction, we shall be using the theory of modular forms. The basic functions for the two types of quadratic forms considered in this paper are $\Theta(z)$ and ${\mathcal F}(z)$. To each quadratic form in \eqref{quad1} with coefficients $(a_1,a_2,a_3,a_4,b_1,b_2)$ as in Table 1, the associated theta series is given by \begin{equation*} \Theta(a_1z) \Theta(a_2z) \Theta(a_3z) \Theta(a_4z) {\mathcal F}(b_1z) {\mathcal F}(b_2z). \end{equation*} Using Lemma 1 and 2 along with the fact that ${\mathcal F}(z) \in M_1(\Gamma_0(3),\chi_{-3})$, it follows that the above product is a modular form in $M_4(\Gamma_0(24), \chi)$, where the character $\chi$ is one of the four characters that appear in Table 1 and it is determined by the coefficients $a_1,a_2,a_3,a_4$. As remarked earlier, the theta series corresponding to the form $x^2+xy+y^2$ is given by \eqref{F1} and it belongs to the space $M_1(\Gamma_0(3),\chi_{-3})$. Therefore, the associated modular form corresponding to the quadratic forms defined by \eqref{quad2} is given explicitly by \begin{equation*} {\mathcal F}(z) {\mathcal F}(c_1z) {\mathcal F}(c_2z) {\mathcal F}(c_3z). \end{equation*} Again by using Lemmas 1, 2 and 3 it follows that the above product is a modular form in $M_4(\Gamma_0(24))$. Therefore, in order to get the required formulae for $N(a_1,a_2,a_3,a_4,b_1,b_2;n)$ and $M(1, c_1,c_2,c_3;n)$ we need a basis for the above spaces of modular forms of level $24$. (We have used the $L$-functions and modular forms database \cite{lfmdb} and \cite{martin} to get some of the cusp forms of weight $4$.) \smallskip \subsection{A basis for $M_4(\Gamma_0(24))$ and proof of \thmref{1}(i).} The vector space $M_4(\Gamma_0(24))$ has dimension $16$ and we have $\dim_{\mathbb C}{\mathcal E}_4(\Gamma_0(24)) = 8$ and $\dim_{\mathbb C}S_4(\Gamma_0(24)) = 8$. For $d=6,8,12$ and $24$, $S_4^{new}(\Gamma_0(d))$ is one-dimensional. Let us define some eta-quotients and use them to give an explicit basis for $S_4(\Gamma_0(24))$. Let\\ \begin{eqnarray} f_{4,6}(z) = 1^{2} 2^{2} 3^{2} 6^2 := \displaystyle{\sum_{n\ge 1}} a_{4,6}(n) q^n, \quad f_{4,8}(z) = 2^4 4^4 := \displaystyle{\sum_{n\ge 1}} a_{4,8}(n) q^n, \\ f_{4,12}(z) = 1^{-1} 2^2 3^3 4^3 6^2 12^{-1} - 1^3 2^2 3^{-1} 4^{-1} 6^2 12^3 := \displaystyle{\sum_{n\ge 1}} a_{4,12}(n) q^n, \\ f_{4,24}(z) = 1^{-4} 2^{11} 3^{-4} 4^{-3} 6^{11} 12^{-3} := \displaystyle{\sum_{n\ge 1}} a_{4,24}(n) q^n. \end{eqnarray} We use the following notation in the sequel. For a Dirichlet character $\chi$ and a function $f$ with Fourier expansion $f(z) =\sum_{n\ge 1} a(n) q^n$, we define the twisted function $f \otimes \chi (z)$ as follows. \begin{equation}\label{twist} f\otimes \chi (z) = \sum_{n\ge 1} \chi(n) a(n)q^n. \end{equation} A basis for the space $M_4(\Gamma_0(24))$ is given in the following proposition. \begin{prop}\label{trivial} A basis for the Eisenstein series space ${\mathcal E}_4(\Gamma_0(24))$ is given by \begin{equation}\left\{E_4(tz), t\vert 24 \right\} \end{equation} and a basis for the space of cusp forms $S_4(\Gamma_0(24))$ is given by \begin{equation} \begin{split} \left\{f_{4,6}(t_1z), t_1\vert 4; f_{4,8}(t_2z), t_2\vert 3; f_{4,12}(t_3z), t_3\vert 2;f_{4,24}\otimes \chi_4(z)\right\}\\ \end{split} \end{equation} Together they form a basis for $M_4(\Gamma_0(24))$. \end{prop} For the sake of simplicity in the formulae, we list these basis elements as $\{f_i(z)\vert 1\le i \le16\}$, where $f_1(z) = E_4(z)$, $ f_2(z) = E_4(2z) $, $ f_3(z) = E_4(3z)$, $ f_4(z) = E_4(4z)$, $f_5(z) = E_4(6z)$, $ f_6(z) = E_4(8z)$, $ f_7(z) = E_4(12z)$,$ f_{8}(z) = E_4(24z)$, $f_{9}(z) = f_{4,6}(z)$, $f_{10}(z) = f_{4,6}(2z)$, $f_{11}(z) = f_{4,6}(4z)$, $f_{12}(z) = f_{4,8}(z)$, $f_{13}(z) = f_{4,8}(3z)$, $f_{14}(z) = f_{4,12}(z)$, $f_{15}(z) = f_{4,12}(2z)$, $f_{16}(z) =f_{4,24}\otimes \chi_4 (z)$\\ \noindent For $1\le i\le 16$, we denote the Fourier coefficients of the basis functions $f_i(z)$ as $$ f_i(z) = \sum_{n\ge 1} A_i(n) q^n. $$ We are now ready to prove the theorem. Noting that all the 36 cases corresponding to the trivial character in Table 1, the resulting functions belong to the space of modular forms of weight $4$ on $\Gamma_0(24)$ with trivial character (using Lemmas 1 to 3). So, we can express these theta functions as a linear combination of the basis given in \propref{trivial} as follows. \begin{equation} \Theta(a_1z) \Theta(a_2z)\Theta(a_3z)\Theta(a_4z){\mathcal F}(b_1z) {\mathcal F}(b_2z) = \sum_{i=1}^{16} \alpha_i f_i(z), \end{equation} where $\alpha_i$'s are some explicit constants. Comparing the $n$-th Fourier coefficients on both the sides, we get \begin{equation*} N(a_1,a_2,a_3,a_4, b_1,b_2;n) = \sum_{i=1}^{16} \alpha_i A_i(n). \end{equation*} Explicit values for the constants \!$\alpha_i$, \!$1\le i \le 16$ corresponding to these 36 cases are given in Table \!3. \smallskip \subsection{A basis for $M_4(\Gamma_0(24),\chi_8)$ and proof of \thmref{1}(ii).} The vector space $M_4(\Gamma_0(24),\chi_8)$ has dimension $14$ and we have $\dim_{\mathbb C}{\mathcal E}_4(\Gamma_0(24),\chi_8)) = 4$ and $\dim_{\mathbb C}S_4(\Gamma_0(24),\chi_8)) = 10$. For $d=6$ and $12$, $S_4^{new}(\Gamma_0(d),\chi_8) = \{0\}$. Also $S_4^{new}(\Gamma_0(8),\chi_8)$ is 2-dimensional and $S_4^{new}(\Gamma_0(24),\chi_8)$ is 6-dimensional.\\ In order to give explicit basis for this space,we define the following \begin{equation}\label{eis:8chi8} E_{4, {\bf 1},\chi_8}(z) ~=~ \frac{11}{2} + \sum_{n\ge 1} \sigma_{3;{\bf 1},\chi_8}(n) q^n, \quad E_{4,\chi_8, {\bf 1}}(z) ~=~ \sum_{n\ge 1} \sigma_{3;\chi_8, {\bf 1}}(n) q^n. \end{equation} \begin{equation}\label{8chi8} f_{4,8,\chi_8;1}(z) = 1^{-2} 2^{11} 4^{-3} 8^2 = \sum_{n\ge 1} a_{4,8,\chi_8;1}(n) q^n ,~~ f_{4,8,\chi_8;2}(z) = 1^{2} 2^{-3} 4^{11} 8^{-2} = \sum_{n\ge 1} a_{4,8,\chi_8;2}(n) q^n. \end{equation} For the space of Eisenstein series we use the basis elements of ${\mathcal E}_4(\Gamma_0(8),\chi_8)$ given in \eqref{eis:8chi8}.A basis for $S_4^{new}(\Gamma_0(8),\chi_8)$ is given in \eqref{8chi8}. The following six eta-quotients span the space $S_4^{new}(\Gamma_0(24), \chi_8)$. \begin{equation} \begin{split}\label{24:chi8} f_{4,24,\chi_8;1}(z) &= 1^2 2^1 3^{-4} 4^1 6^{10} 8^2 12^{-4} := \sum_{n\ge 1} a_{4,24,\chi_8;1}(n) q^n ,~~\\ f_{4,24,\chi_8;2}(z) & = 1^1 2^3 3^{-1} 4^1 6^4 8^{-1} 24^1 := \sum_{n\ge 1} a_{4,24,\chi_8;2}(n) q^n,~~\\ f_{4,24,\chi_8;3}(z) & = 1^{-1} 2^4 3^1 6^3 8^1 12^1 24^{-1} := \sum_{n\ge 1} a_{4,24,\chi_8;3}(n) q^n , ~~\\ f_{4,24,\chi_8;4}(z) & = 1^{-2} 2^4 4^2 6^1 8^2 12^1 := \sum_{n\ge 1} a_{4,24,\chi_8;4}(n) q^n,\\ f_{4,24,\chi_8;5}(z) & = 2^1 3^{-2} 4^1 6^4 12^2 24^2 := \sum_{n\ge 1} a_{4,24,\chi_8;5}(n) q^n , ~~\\ f_{4,24,\chi_8;6}(z) & = 1^{-6} 2^{14} 6^1 8^{-2} 12^1 := \sum_{n\ge 1} a_{4,24,\chi_8;6}(n) q^n \end{split} \end{equation} A basis for the space $M_4(\Gamma_0(24),\chi_{8})$ is given in the following proposition. \begin{prop}\label{chi8} A basis for the space $M_4(\Gamma_0(24),\chi_8)$ is given by \begin{equation} \begin{split} &\left\{E_{4, {\bf 1},\chi_8}(tz),~ E_{4,\chi_8, {\bf 1}}(tz), t\vert3; f_{4,8,\chi_8;1}(t_1z),f_{4,8,\chi_8;2}(t_1z), t_1\vert 3; f_{4,24,\chi_8;1}(z),\right.\\ & \left.f_{4,24,\chi_8;2}(z),f_{4,24,\chi_8;3}(z),f_{4,24,\chi_8;4}(z), f_{4,24,\chi_8;5}(z),f_{4,24,\chi_8;6}(z)\right\}\\ \end{split} \end{equation} where $E_{4,{\bf 1},\chi_8}(z)$ and $E_{4,\chi_8,{\bf 1}}(z)$ are defined in \eqref{eis:8chi8}, $f_{4,8,\chi_8;i}(z)$, $i=1,2$ are defined in \eqref{8chi8} and $f_{4,24,\chi_8;j}(z)$, $1\le j\le 6$ are defined by \eqref{24:chi8}. \end{prop} For the sake of simplifying the notation, we shall list the basis in \propref{chi8} as $$ g_i(z) = \sum_{n\ge 1} B_i(n) q^n, ~1\le i \le 14, $$ where $g_1(z) = E_{4, {\bf 1},\chi_8}(z)$, $g_2(z) = E_{4, {\bf 1},\chi_8}(3z)$, $g_3(z)=E_{4,\chi_8, {\bf 1}}(z)$, $g_4(z) = E_{4,\chi_8, {\bf 1}}(3z)$, $g_5(z) = f_{4,8,\chi_8;1}(z)$, $g_{6}(z) = f_{4,8,\chi_8;1}(3z)$, $g_{7}(z) = f_{4,8,\chi_8;2}(z)$,$g_{8}(z) = f_{4,8,\chi_8;2}(3z)$,$g_{9}(z)=$ $f_{4,24,\chi_8;1}(z)$, $g_{10}(z) = f_{4,24,\chi_8;2}(z)$, $g_{11}(z) = f_{4,24,\chi_8;3}(z)$, $g_{12}(z) = f_{4,24,\chi_8;4}(z)$, $g_{13}(z) = f_{4,24,\chi_8;5}(z)$, $g_{14}(z) = f_{4,24,\chi_8;6}(z)$, \noindent We now prove \thmref{1}(ii). In this case, for all the 18 sextuples corresponding to the $\chi_8$ character space (in Table 1), the resulting products of theta functions are modular forms of weight $4$ on $\Gamma_0(24)$ with character $\chi_8$ (By Lemma 1 to 3). So, we can express these products of theta functions as a linear combination of the basis given in \propref{chi8}: \begin{equation} \Theta(a_1z) \Theta(a_2z)\Theta(a_3z)\Theta(a_4z) {\mathcal F}(b_1z) {\mathcal F}(b_2z) = \sum_{i=1}^{14} \beta_i g_i(z). \end{equation} Comparing the $n$-th Fourier coefficients on both the sides, we get \begin{equation*} N(a_1,a_2,a_3,a_4, b_1,b_2;n) = \sum_{i=1}^{14} \beta_i B_i(n). \end{equation*} Explicit values for the constants $\beta_i$, $1\le i \le 14$ corresponding to these 18 cases are given in Table 4. \smallskip \subsection{A basis for $M_4(\Gamma_0(24), \chi_{12})$ and proof of \thmref{1}(iii).} The dimension of the space in this case is $16$, with $\dim_{\mathbb C}{\mathcal E}_4(\Gamma_0(24), \chi_{12}) = 8$ and $\dim_{\mathbb C}S_4(\Gamma_0(24), \chi_{12}) = 8$. The old class is spanned by the space $S_4^{new}(\Gamma_0(12), \chi_{12})$, which is $4$ dimensional with spanning functions given by the following four eta-quotients: \begin{equation} \begin{split} f_{4,12,\chi_{12};1}(z) ~=~ 2^{-1} 3^4 4^2 6^5 12^{-2}, & \quad f_{4,12,\chi_{12};2}(z) ~=~ 3^4 4^3 6^{-2} 12^3, \\ f_{4,12,\chi_{12};3}(z) ~=~ 2^2 3^4 4^{-1} 6^{-4} 12^7, & \quad f_{4,12,\chi_{12};4}(z) = 1^4 4^{-1} 6^{-2} 12^7. \\ \end{split} \end{equation} We write the Fourier expansions of these forms as $f_{4,12,\chi_{12};j}(z) = \displaystyle{\sum_{n\ge 1}} a_{4,12,\chi_{12};j}(n) q^n$, $1\le j\le 4$.\\ In the following proposition we give a basis for the space $M_4(\Gamma_0(24),\chi_{12})$. \begin{prop}\label{chi12} A basis for the space $M_4(\Gamma_0(24),\chi_{12})$ is given by \begin{equation} \begin{split} \left\{E_{4, {\bf 1},\chi_{12}}(tz),{E_{4,\chi_{12}, {\bf 1}}(tz),E_{4,\chi_{-4},\chi_{-3}}(tz),E_{4,\chi_{-3},\chi_{-4}}(tz)}, t\vert 2; f_{4,12,\chi_{12};j}(t_1z), t_1\vert 2, 1\le j\le 4\right\},&\\ \end{split} \end{equation} where the Eisenstein series in the basis are defined by \eqref{eisenstein}. \end{prop} Let us denote the 16 basis elements in the above proposition as follows. \\ $\left\{h_i(z)\vert 1\le i \le 16\right\}$, where $h_1(z) = E_{4, {\bf 1},\chi_{12}}(z)$, $h_2(z) = E_{4,\chi_{12}, {\bf 1}}(z)$, $h_3(z) = E_{4,\chi_{-4},\chi_{-3}}(z)$, $h_4(z) = E_{4,\chi_{-3},\chi_{-4}}(z)$, $h_5(z) = E_{4, {\bf 1},\chi_{12}}(2z)$, $h_6(z) = E_{4,\chi_{12}, {\bf 1}}(2z)$,$h_7(z) = E_{4,\chi_{-4},\chi_{-3}}(2z)$, $h_8(z) = E_{4,\chi_{-3},\chi_{-4}}(2z)$,$h_{8+j}(z) = f_{4,12,\chi_{12};j}(z)$, $1\le j\le 4$, $h_{12+j}(z) = f_{4,12,\chi_{12};j}(2z)$, $1\le j\le 4$. \smallskip To prove \thmref{1}(iii), we consider the case of 18 sextuples corresponding to the character $\chi_{12}$ in Table 1. The resulting products of theta functions are modular forms of weight $4$ on $\Gamma_0(24)$ with character $\chi_{12}$ (once again we use Lemams 1 to 3 to get this). So, we can express each of these products of theta functions as a linear combination of the basis given in \propref{chi12} as follows. \begin{equation} \Theta(a_1z) \Theta(a_2z)\Theta(a_3z)\Theta(a_4z) {\mathcal F}(b_1z) {\mathcal F}(b_2z) = \sum_{i=1}^{16} \gamma_i g_i(z). \end{equation} Comparing the $n$-th Fourier coefficients on both the sides, we get \begin{equation*} N(a_1,a_2,a_3,a_4, b_1,b_2;n) = \sum_{i=1}^{16} \gamma_i C_i(n). \end{equation*} Explicit values of the constants $\gamma_i$, $1\le i \le 16$ corresponding to these 18 cases are given in Table 5. \smallskip \subsection{A basis for $M_4(\Gamma_0(24), \chi_{24})$ and proof of \thmref{1}(iv).} ~ We have \linebreak $\dim_{\mathbb C} M_4(\Gamma_0(24), \chi_{24}) = 14$ and $\dim_{\mathbb C}{\mathcal E}_4(\Gamma_0(24),\chi_{24}) = 4$. To get the span of the Eisenstein series space ${\mathcal E}_4(\Gamma_0(24),\chi_{24})$, we use the Eisenstein series $E_{4,\chi,\psi}(z)$ defined in \eqref{eisenstein}, where $\chi,\psi \in \{{\bf 1}, \chi_{-8}, \chi_{-12}, \chi_{24}\}$. Note that for d= 6,8 and 12, $S_4^{new}(\Gamma_0(d),\chi_{24}) = \{0\}$ and the space $S_4^{new}(\Gamma_0(24),\chi_{24})$ is spanned by the following ten eta-quotients (notation as in \eqref{eta-q}): \begin{equation} \begin{split} f_{4,24,\chi_{24};1}(z) ~=~ 3^{-2} 6^7 8^3 12^3 24^{-3}, &\quad f_{4,24,\chi_{24};2}(z) ~=~ 3^{2} 4^7 6^{-3} 8^{-2} 12^{4}, \\ ~~f_{4,24,\chi_{24};3}(z) = 3^{2} 4^{-3} 6^{1} 8^{6} 12^{2}, \qquad &\quad f_{4,24,\chi_{24};4}(z) = 3^2 6^{-3} 8^3 12^5 24^1, \\ f_{4,24,\chi_{24};5}(z) = 3^2 4^2 6^{-3} 8^{-1} 12^3 24^5, & \quad f_{4,24,\chi_{24};6}(z) = 3^2 4^1 6^1 8^{-2} 12^{-2} 24^8, \\ f_{4,24,\chi_{24};7}(z) = 3^2 4^1 6^1 8^{-2} 12^{-2} 24^8, & \quad f_{4,24,\chi_{24};8}(z) = 1^1 3^{-1} 6^1 8^{-2} 12^1 24^8, \\ f_{4,24,\chi_{24};9}(z) = 2^2 3^6 4^1 6^{-3} 8^2, &\quad f_{4,24,\chi_{24};10}(z) = 3^2 4^3 6^5 12^{-4} 24^2. \\ \end{split} \end{equation} We write the Fourier expansions as $f_{4,24,\chi_{24};j}(z) = \sum_{n \ge 1} a_{4,24,\chi_{24};j}(n) q^n$. We now give a basis for the space $M_4(\Gamma_0(24),\chi_{24})$ in the following proposition. \begin{prop}\label{chi24} The following functions span the space $M_4(\Gamma_0(24),\chi_{24})$. \begin{equation} \begin{split} & \quad \left\{E_{4, {\bf 1},\chi_{24}}(z),{E_{4,\chi_{24}, {\bf 1}}(z),E_{4,\chi_{-8},\chi_{-3}}(z),E_{4,\chi_{-3},\chi_{-8},}(z)},f_{4,24,\chi_{24};j}(z), 1\le j\le 10;\right\}. \end{split} \end{equation} \end{prop} We list these basis elements as $\{F_i(z)\vert 1\le i \le 14\}$, where $ F_1(z) = E_{4, {\bf 1},\chi_{24}}(z)$, $ F_2(z) = E_{4,\chi_{24}, {\bf 1}}(z)$, $ F_3(z) = E_{4,\chi_{-8},\chi_{-3}}(z)$, $ F_4(z) = E_{4,\chi_{-3},\chi_{-8},}(z)$, $ F_5(z) = f_{4,24,\chi_{24};1}(z)$, $ F_6(z) = f_{4,24,\chi_{24};2}(z)$, $ F_7(z) = f_{4,24,\chi_{24};3}(z)$, $ F_8(z) = f_{4,24,\chi_{24};4}(z)$, $ F_9(z) = f_{4,24,\chi_{24};5}(z)$, $ F_{10}(z) = f_{4,24,\chi_{24};6}(z)$, $ F_{11}(z) = f_{4,24,\chi_{24};7}(z)$, $ F_{12}(z) =f_{4,24,\chi_{24};8}(z)$, $ F_{13}(z) = f_{4,24,\chi_{24};9}(z)$, $ F_{14}(z) = f_{4,24,\chi_{24};10}(z)$,\\ \noindent As in the previous cases, we denote the Fourier coefficients of these basis functions by $$ F_i(z) = \sum_{n\ge 1} D_i(n) q^n, ~1\le i \le 14. $$ To get the formula in \thmref{1}(iv), we note that for all the 18 sextuples corresponding to the character $\chi_{24}$ in Table 1, the resulting functions belong to the space $M_4(\Gamma_0(24), \chi_{24})$, by using Lemmas 1 to 3. So, as before, we express these theta functions as linear combinations of the basis elements: \begin{equation} \Theta(a_1z) \Theta(a_2z)\Theta(a_3z)\Theta(a_4z) {\mathcal F}(b_1z) {\mathcal F}(b_2z) = \sum_{i=1}^{14} \delta_i g_i(z). \end{equation} Comparing the $n$-th Fourier coefficients on both the sides, we get \begin{equation*} N(a_1,a_2,a_3,a_4, b_1,b_2;n) = \sum_{i=1}^{14} \delta_i D_i(n). \end{equation*} Explicit values of the constants $\delta_i$, $1\le i \le 14$ corresponding to these 18 cases corresponding to character $\chi_{24}$ are given in Table 6. \smallskip \subsection{Proof of \thmref{2}} This theorem is corresponding to Table 2 and in this case all the product functions $$ {\mathcal F}(z) {\mathcal F}(c_1z) {\mathcal F}(c_2z) {\mathcal F}(c_3z) $$ belong to the space $M_4(\Gamma_0(24))$. Therefore, proceeding as in the proof of \thmref{1}(i), we express these theta functions as linear combinations of the basis elements: \begin{equation} {\mathcal F}(z) {\mathcal F}(c_1z) {\mathcal F}(c_2z) {\mathcal F}(c_3z) = \sum_{i=1}^{16} \nu_i f_i(z). \end{equation} Comparing the $n$-th Fourier coefficients on both the sides, we get \begin{equation*} M(1,c_1,c_2,c_3;n) = \sum_{i=1}^{16} \nu_i A_i(n). \end{equation*} The constants $\nu_i$, $1\le i \le 16$ corresponding to the 19 cases of table 2 are given in Table 7. \bigskip \section{List of tables and sample formulas } In this section we list the remaining tables mentioned in the theorems and provide explicit sample formulas in some cases. In the first subsection we list the tables and in the second subsection we give the sample formulas. \smallskip \subsection{List of tables} In this section, we list the tables 3, 4, 5, 6 and 7 which give the explicit coefficients that appear in the formulas of \thmref{1} and \thmref{2}. \bigskip \newpage \begin{center} {\tiny \textbf{Table 3.} (Theorem 2.1 (i)) \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|} \hline $(a_1a_2a_3a_4,b_1b_2)$&$ \alpha_1$ &$ \alpha_2$& $ \alpha_3$& $ \alpha_4$&$ \alpha_5$ &$ \alpha_6$& $ \alpha_7$& $ \alpha_8$&$ \alpha_9$ &$ \alpha_{10}$ & $ \alpha_{11}$&$ \alpha_{12}$& $ \alpha_{13}$&$ \alpha_{14}$& $ \alpha_{15}$&$ \alpha_{16}$\\ \hline &&&&&&&&&&&&&&&&\\ $(1111,11)$&$\frac{7}{75}$&$\frac{-7}{100}$&$\frac{-9}{25}$&$\frac{-28}{75}$& $\frac{27}{100}$&0&$\frac{36}{25}$&0&$\frac{-72}{5}$&$\frac{-288}{5}$&0&0&0&12&0&0\\ &&&&&&&&&&&&&&&&\\ $(1111,12)$ &$\frac{13}{300}$&$\frac{-13}{200}$&$\frac{9}{100}$&$\frac{26}{75}$&$\frac{-27}{200}$&0&$\frac{18}{25}$&0&$\frac{48}{5}$&$\frac{96}{5}$&0&0&0&-6&0&0\\ &&&&&&&&&&&&&&&&\\ $(1111,14)$ &$\frac{7}{300}$&0&$\frac{-9}{100}$&$\frac{-28}{75}$&0&0&$\frac{36}{25}$&0&$\frac{-18}{5}$&$\frac{72}{5}$&0&0&0&12&0&0\\ &&&&&&&&&&&&&&&&\\ $(1111,22)$ &$\frac{7}{300}$&0&$\frac{-9}{100}$&$\frac{-28}{75}$&0&0&$\frac{36}{25}$&0&$\frac{12}{5}$&$\frac{-48}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1111,24)$ &$\frac{13}{1200}$&$\frac{-13}{400}$&$\frac{9}{400}$&$\frac{26}{75}$&$\frac{-27}{400}$&0&$\frac{18}{25}$&0&$\frac{12}{5}$&$\frac{96}{5}$&0&0&0&3&0&0\\ &&&&&&&&&&&&&&&&\\ $(1111,44)$ &$\frac{7}{1200}$&$\frac{7}{400}$&$\frac{-9}{400}$&$\frac{-28}{75}$&$\frac{-27}{400}$&0&$\frac{36}{25}$&0&$\frac{18}{5}$&$\frac{72}{5}$&0&0&0&3&0&0\\ &&&&&&&&&&&&&&&&\\ $(1122,11)$ &$\frac{7}{150}$&$\frac{-7}{150}$&$\frac{-9}{50}$&$\frac{7}{300}$&$\frac{9}{50}$&$\frac{-28}{75}$&$\frac{-9}{100}$&$\frac{36}{25}$&$\frac{-36}{5}$&-48&$\frac{-768}{5}$&-3&-81&6&36&9\\ &&&&&&&&&&&&&&&&\\ $(1122,12)$ &$\frac{13}{600}$&$\frac{-13}{600}$&$\frac{9}{200}$&$\frac{-13}{600}$&$\frac{-9}{200}$&$\frac{26}{75}$&$\frac{-9}{200}$&$\frac{18}{25}$&$\frac{24}{5}$&12&$\frac{96}{5}$&$\frac{15}{2}$&$\frac{81}{2}$&-3&-6&$\frac{-9}{2}$\\ &&&&&&&&&&&&&&&&\\ $(1122,14)$ &$\frac{7}{600}$&$\frac{-7}{600}$&$\frac{-9}{200}$&$\frac{7}{300}$&$\frac{9}{200}$&$\frac{-28}{75}$&$\frac{-9}{100}$&$\frac{36}{25}$&$\frac{-9}{5}$&6&$\frac{-48}{5}$&$\frac{-3}{2}$&$\frac{-81}{2}$&6&0&$\frac{9}{2}$\\ &&&&&&&&&&&&&&&&\\ $(1122,22)$ &$\frac{7}{600}$&$\frac{-7}{600}$&$\frac{-9}{200}$&$\frac{7}{300}$&$\frac{9}{200}$&$\frac{-28}{75}$&$\frac{-9}{100}$&$\frac{36}{25}$&$\frac{6}{5}$&-12&$\frac{-288}{5}$&0&0&0&12&0\\ &&&&&&&&&&&&&&&&\\ $(1122,24)$ &$\frac{13}{2400}$&$\frac{-13}{2400}$&$\frac{9}{800}$&$\frac{-13}{600}$&$\frac{-9}{800}$&$\frac{26}{75}$&$\frac{-9}{200}$&$\frac{18}{25}$&$\frac{6}{5}$&12&$\frac{96}{5}$&0&0&$\frac{3}{2}$&-6&0\\ &&&&&&&&&&&&&&&&\\ $(1122,44)$ &$\frac{7}{2400}$&$\frac{-7}{2400}$&$\frac{-9}{800}$&$\frac{7}{300}$&$\frac{9}{800}$&$\frac{-28}{75}$&$\frac{-9}{100}$&$\frac{36}{25}$&$\frac{9}{5}$&6&$\frac{-48}{5}$&0&0&$\frac{3}{2}$&0&0\\ &&&&&&&&&&&&&&&&\\ $(1133,11)$ &$\frac{2}{75}$&$\frac{-1}{30}$&$\frac{6}{25}$&$\frac{8}{75}$&$\frac{-3}{10}$&0&$\frac{24}{25}$&0&$\frac{48}{5}$&$\frac{288}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1133,12)$ &$\frac{1}{60}$&$\frac{-1}{120}$&$\frac{-3}{20}$&$\frac{-2}{15}$&$\frac{3}{40}$&0&$\frac{6}{5}$&0&0&0&0&0&0&6&0&0\\ &&&&&&&&&&&&&&&&\\ $(1133,14)$ &$\frac{1}{150}$&$\frac{-1}{75}$&$\frac{3}{50}$&$\frac{8}{75}$&$\frac{-3}{25}$&0&$\frac{24}{25}$&0&$\frac{42}{5}$&$\frac{168}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1133,22)$ &$\frac{1}{150}$&$\frac{-1}{75}$&$\frac{3}{50}$&$\frac{8}{75}$&$\frac{-3}{25}$&0&$\frac{24}{25}$&0&$\frac{12}{5}$&$\frac{48}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1133,24)$ &$\frac{1}{240}$&$\frac{1}{240}$&$\frac{-3}{80}$&$\frac{-2}{15}$&$\frac{-3}{80}$&0&$\frac{6}{5}$&0&0&0&0&0&0&3&0&0\\ &&&&&&&&&&&&&&&&\\ $(1133,44)$ &$\frac{1}{600}$&$\frac{-1}{120}$&$\frac{3}{200}$&$\frac{8}{75}$&$\frac{-3}{40}$&0&$\frac{24}{25}$&0&$\frac{18}{5}$&$\frac{48}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(2222,11)$ &$\frac{7}{400}$&$\frac{91}{1200}$&$\frac{-27}{400}$&$\frac{-7}{100}$&$\frac{-117}{400}$&$\frac{-28}{75}$&$\frac{27}{100}$&$\frac{36}{25}$&$\frac{-36}{5}$&$\frac{-312}{5}$&$\frac{-768}{5}$&-3&-81&9&36&9\\ &&&&&&&&&&&&&&&&\\ $(2222,12)$ &$\frac{13}{800}$&$\frac{-247}{2400}$&$\frac{27}{800}$&$\frac{13}{200}$&$\frac{-171}{800}$&$\frac{26}{75}$&$\frac{27}{200}$&$\frac{18}{25}$&$\frac{18}{5}$&$\frac{84}{5}$&$\frac{96}{5}$&$\frac{15}{2}$&$\frac{81}{2}$&$\frac{-9}{2}$&-6&$\frac{-9}{2}$\\ &&&&&&&&&&&&&&&&\\ $(2222,14)$ &$\frac{7}{800}$&$\frac{-133}{2400}$&$\frac{-27}{800}$&$\frac{7}{100}$&$\frac{171}{800}$&$\frac{-28}{75}$&$\frac{-27}{100}$&$\frac{36}{25}$&$\frac{-18}{5}$&$\frac{-24}{5}$&$\frac{-48}{5}$&$\frac{-3}{2}$&$\frac{-81}{2}$&$\frac{9}{2}$&0&$\frac{9}{2}$\\ &&&&&&&&&&&&&&&&\\ $(2222,22)$ &0&$\frac{7}{75}$&0&$\frac{-7}{100}$&$\frac{-9}{25}$&$\frac{-28}{75}$&$\frac{27}{100}$&$\frac{36}{25}$&0&$\frac{-72}{5}$&$\frac{-288}{5}$&0&0&0&12&0\\ &&&&&&&&&&&&&&&&\\ $(2222,24)$ &0&$\frac{13}{300}$&0&$\frac{-13}{200}$&$\frac{9}{100}$&$\frac{26}{75}$&$\frac{-27}{200}$&$\frac{18}{25}$&0&$\frac{48}{5}$&$\frac{96}{5}$&0&0&0&-6&0\\ &&&&&&&&&&&&&&&&\\ $(2222,44)$ &0&$\frac{7}{300}$&0&0&$\frac{-9}{100}$&$\frac{-28}{75}$&0&$\frac{36}{25}$&0&$\frac{12}{5}$&$\frac{-48}{5}$&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(2233,11)$ &$\frac{1}{75}$&$\frac{-1}{75}$&$\frac{3}{25}$&$\frac{-1}{150}$&$\frac{-3}{25}$&$\frac{8}{75}$&$\frac{-3}{50}$&$\frac{24}{25}$&$\frac{24}{5}$&48&$\frac{768}{5}$&10&18&0&-24&-6\\ &&&&&&&&&&&&&&&&\\ $(2233,12)$ &$\frac{1}{120}$&$\frac{-1}{120}$&$\frac{-3}{40}$&$\frac{1}{120}$&$\frac{3}{40}$&$\frac{-2}{15}$&$\frac{-3}{40}$&$\frac{6}{5}$&0&-12&-96&$\frac{-7}{2}$&$\frac{-9}{2}$&3&6&$\frac{9}{2}$\\ &&&&&&&&&&&&&&&&\\ $(2233,14)$ &$\frac{1}{300}$&$\frac{-1}{300}$&$\frac{3}{100}$&$\frac{-1}{150}$&$\frac{-3}{100}$&$\frac{8}{75}$&$\frac{-3}{50}$&$\frac{24}{25}$&$\frac{21}{5}$&18&$\frac{48}{5}$&$\frac{5}{2}$&$\frac{63}{2}$&0&-12&$\frac{-3}{2}$\\ &&&&&&&&&&&&&&&&\\ $(2233,22)$ &$\frac{1}{300}$&$\frac{-1}{300}$&$\frac{3}{100}$&$\frac{-1}{150}$&$\frac{-3}{100}$&$\frac{8}{75}$&$\frac{-3}{50}$&$\frac{24}{25}$&$\frac{6}{5}$&12&$\frac{288}{5}$&1&-9&0&0&-3\\ &&&&&&&&&&&&&&&&\\ $(2233,24)$ &$\frac{1}{480}$&$\frac{-1}{480}$&$\frac{-3}{160}$&$\frac{1}{120}$&$\frac{3}{160}$&$\frac{-2}{15}$&$\frac{-3}{40}$&$\frac{6}{5}$&0&0&0&-2&-18&$\frac{3}{2}$&6&0\\ &&&&&&&&&&&&&&&&\\ $(2233,44)$ &$\frac{1}{1200}$&$\frac{-1}{1200}$&$\frac{3}{400}$&$\frac{-1}{150}$&$\frac{-3}{400}$&$\frac{8}{75}$&$\frac{-3}{50}$&$\frac{24}{25}$&$\frac{9}{5}$&6&$\frac{48}{5}$&$\frac{-1}{2}$&$\frac{9}{2}$&0&0&$\frac{-3}{2}$\\ &&&&&&&&&&&&&&&&\\ $(3333,11)$ &$\frac{1}{75}$&$\frac{-1}{100}$&$\frac{-7}{25}$&$\frac{-4}{75}$&$\frac{21}{100}$&0&$\frac{28}{25}$&0&$\frac{24}{5}$&$\frac{96}{5}$&0&0&0&4&0&0\\ &&&&&&&&&&&&&&&&\\ $(3333,12)$ &$\frac{1}{300}$&$\frac{-1}{200}$&$\frac{13}{100}$&$\frac{2}{75}$&$\frac{-39}{200}$&0&$\frac{26}{25}$&0&$\frac{16}{5}$&$\frac{32}{5}$&0&0&0&2&0&0\\ &&&&&&&&&&&&&&&&\\ $(3333,14)$ &$\frac{1}{300}$&0&$\frac{-7}{100}$&$\frac{-4}{75}$&0&0&$\frac{28}{25}$&0&$\frac{6}{5}$&$\frac{-24}{5}$&0&0&0&4&0&0\\ &&&&&&&&&&&&&&&&\\ $(3333,22)$ &$\frac{1}{300}$&0&$\frac{-7}{100}$&$\frac{-4}{75}$&0&0&$\frac{28}{25}$&0&$\frac{-4}{5}$&$\frac{16}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(3333,24)$ &$\frac{1}{1200}$&$\frac{-1}{400}$&$\frac{13}{400}$&$\frac{2}{75}$&$\frac{-39}{400}$&0&$\frac{26}{25}$&0&$\frac{4}{5}$&$\frac{32}{5}$&0&0&0&-1&0&0\\ &&&&&&&&&&&&&&&&\\ $(3333,44)$ &$\frac{1}{1200}$&$\frac{1}{400}$&$\frac{-7}{400}$&$\frac{-4}{75}$&$\frac{-21}{400}$&0&$\frac{28}{25}$&0&$\frac{-6}{5}$&$\frac{-24}{5}$&0&0&0&1&0&0\\ &&&&&&&&&&&&&&&&\\ \hline \end{tabular} } \end{center} \newpage \begin{center} {\tiny \textbf{Table 4.} (Theorem 2.1 (ii)) \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|} \hline &&&&&&&&&&&&&&\\ $a_1a_2a_3$&$ \beta_1$ &$ \beta_2$& $ \beta_3$& $ \beta_4$&$ \beta_5$ &$ \beta_6$& $ \beta_7$& $ \beta_8$&$ \beta_9$ &$ \beta_{10}$ & $ \beta_{11}$&$ \beta_{12}$& $ \beta_{13}$&$ \beta_{14}$\\ $a_4,b_1b_2$&$~~$ &$~~$& $~~$& $~~$&$~~$ &$~~$& $~~$& $~~$&$~~$ &$ ~~$ & $~~$&$~~$& $~~$&$~~$\\ \hline &&&&&&&&&&&&&&\\ $1112,11$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{6656}{451}$&$\frac{27648}{451}$&$\frac{168}{451}$&$\frac{11448}{451}$&$\frac{-2496}{451}$&$\frac{-17280}{451}$&$\frac{24}{41}$&$\frac{936}{41}$&$\frac{144}{41}$&$\frac{-384}{41}$&$\frac{4032}{41}$&$\frac{-48}{41}$\\ &&&&&&&&&&&&&&\\ $1112,12$&$\frac{28}{451}$&$\frac{54}{451}$&$\frac{3584}{451}$&$\frac{-6912}{451}$&$\frac{480}{451}$&$\frac{-2052}{451}$&$\frac{-2688}{451}$&$\frac{1728}{451}$&$\frac{-60}{41}$&$\frac{216}{41}$&$\frac{-108}{41}$&$\frac{-2112}{41}$&$\frac{-1440}{41}$&$\frac{288}{41}$\\ &&&&&&&&&&&&&&\\ $1112,14$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{1664}{451}$&$\frac{6912}{451}$&$\frac{-912}{451}$&$\frac{3672}{451}$&0&$\frac{-6912}{451}$&$\frac{-48}{41}$&$\frac{-54}{41}$&$\frac{702}{41}$&$\frac{1632}{41}$&$\frac{576}{41}$&$\frac{-228}{41}$\\ &&&&&&&&&&&&&&\\ $1112,22$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{1664}{451}$&$\frac{6912}{451}$&$\frac{-912}{451}$&$\frac{3672}{451}$&0&$\frac{-6912}{451}$&$\frac{-48}{41}$&$\frac{684}{41}$&$\frac{-36}{41}$&$\frac{-2304}{41}$&$\frac{576}{41}$&$\frac{264}{41}$\\ &&&&&&&&&&&&&&\\ $1112,24$&$\frac{28}{451}$&$\frac{54}{451}$&$\frac{896}{451}$&$\frac{-1728}{451}$&$\frac{-66}{41}$&$\frac{-108}{451}$&$\frac{-1344}{451}$&$\frac{-864}{451}$&$\frac{-42}{41}$&$\frac{-90}{41}$&$\frac{306}{41}$&$\frac{336}{41}$&$\frac{-576}{41}$&$\frac{-36}{41}$\\ &&&&&&&&&&&&&&\\ $1112,44$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{416}{451}$&$\frac{1728}{451}$&$\frac{-1182}{451}$&$\frac{1728}{451}$&$\frac{624}{451}$&$\frac{-4320}{451}$&$\frac{-66}{41}$&$\frac{252}{41}$&$\frac{288}{41}$&$\frac{-816}{41}$&$\frac{-288}{41}$&$\frac{96}{41}$\\ &&&&&&&&&&&&&&\\ $1222,11$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{3328}{451}$&$\frac{13824}{451}$&$\frac{8468}{451}$&$\frac{6264}{451}$&$\frac{-15264}{451}$&$\frac{-10368}{451}$&$\frac{-352}{41}$&$\frac{-708}{41}$&$\frac{516}{41}$&$\frac{3584}{41}$&$\frac{13536}{41}$&$\frac{-660}{41}$\\ &&&&&&&&&&&&&&\\ $1222,12$&$\frac{28}{451}$&$\frac{54}{451}$&$\frac{1792}{451}$&$\frac{-3456}{451}$&$\frac{9598}{451}$&$\frac{-756}{451}$&$\frac{-16224}{451}$&0&$\frac{-458}{41}$&$\frac{-1956}{41}$&$\frac{168}{41}$&$\frac{2800}{41}$&$\frac{5040}{41}$&$\frac{-420}{41}$\\ &&&&&&&&&&&&&&\\ $1222,14$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{832}{451}$&$\frac{3456}{451}$&$\frac{-6955}{451}$&$\frac{216}{41}$&$\frac{7632}{451}$&$\frac{-5184}{451}$&$\frac{473}{41}$&$\frac{1749}{41}$&$\frac{57}{41}$&-56&-144&$\frac{357}{41}$\\ &&&&&&&&&&&&&&\\ $1222,22$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{832}{451}$&$\frac{3456}{451}$&$\frac{10634}{451}$&$\frac{216}{41}$&$\frac{-14016}{451}$&$\frac{-5184}{451}$&$\frac{-634}{41}$&$\frac{-2310}{41}$&$\frac{426}{41}$&112&288&$\frac{-750}{41}$\\ &&&&&&&&&&&&&&\\ $1222,24$&$\frac{28}{451}$&$\frac{54}{451}$&$\frac{448}{451}$&$\frac{-864}{451}$&$\frac{-3182}{451}$&$\frac{216}{451}$&$\frac{6096}{451}$&$\frac{-1296}{451}$&$\frac{166}{41}$&$\frac{474}{41}$&$\frac{6}{41}$&$\frac{-896}{41}$&$\frac{-3384}{41}$&$\frac{156}{41}$\\ &&&&&&&&&&&&&&\\ $1222,44$&$\frac{-26}{451}$&$\frac{108}{451}$&$\frac{208}{451}$&$\frac{864}{451}$&$\frac{2381}{451}$&$\frac{1404}{451}$&$\frac{-2880}{451}$&$\frac{-3888}{451}$&$\frac{-151}{41}$&$\frac{-681}{41}$&$\frac{219}{41}$&$\frac{1400}{41}$&$\frac{2520}{41}$&$\frac{-219}{41}$\\ &&&&&&&&&&&&&&\\ $1233,11$&$\frac{10}{451}$&$\frac{72}{451}$&$\frac{2560}{451}$&$\frac{-18432}{451}$&$\frac{488}{451}$&$\frac{-6192}{451}$&$\frac{-1600}{451}$&$\frac{6912}{451}$&$\frac{-112}{41}$&$\frac{-480}{41}$&$\frac{432}{41}$&$\frac{704}{41}$&$\frac{-3456}{41}$&$\frac{-24}{41}$\\ &&&&&&&&&&&&&&\\ $1233,12$&$\frac{-8}{451}$&$\frac{90}{451}$&$\frac{1024}{451}$&$\frac{11520}{451}$&$\frac{-1280}{451}$&$\frac{5220}{451}$&$\frac{-256}{451}$&$\frac{-8640}{451}$&$\frac{-20}{41}$&$\frac{312}{41}$&$\frac{348}{41}$&$\frac{-512}{41}$&$\frac{1440}{41}$&$\frac{24}{41}$\\ &&&&&&&&&&&&&&\\ $1233,14$&$\frac{10}{451}$&$\frac{72}{451}$&$\frac{640}{451}$&$\frac{-4608}{451}$&$\frac{-760}{451}$&$\frac{-1008}{451}$&$\frac{-640}{451}$ &0&$\frac{-64}{41}$&$\frac{-66}{41}$&$\frac{306}{41}$&$\frac{-640}{41}$&$\frac{-1152}{41}$&$\frac{96}{41}$\\ &&&&&&&&&&&&&&\\ $1233,22$&$\frac{10}{451}$&$\frac{72}{451}$&$\frac{640}{451}$&$\frac{-4608}{451}$&$\frac{-760}{451}$&$\frac{-1008}{451}$&$\frac{-640}{451}$&0&$\frac{-64}{41}$&$\frac{180}{41}$&$\frac{60}{41}$&$\frac{-640}{41}$&$\frac{-1152}{41}$&$\frac{96}{41}$\\ &&&&&&&&&&&&&&\\ $1233,24$&$\frac{-8}{451}$&$\frac{90}{451}$&$\frac{256}{451}$&$\frac{2880}{451}$&$\frac{-1238}{451}$&$\frac{180}{41}$&$\frac{128}{451}$&$\frac{-4320}{451}$&$\frac{-50}{41}$&$\frac{330}{41}$&$\frac{150}{41}$&-16&0&$\frac{72}{41}$\\ &&&&&&&&&&&&&&\\ $1233,44$&$\frac{10}{451}$&$\frac{72}{451}$&$\frac{160}{451}$&$\frac{-1152}{451}$&$\frac{-1072}{451}$&$\frac{288}{451}$&$\frac{-400}{451}$&$\frac{-1728}{451}$&$\frac{-52}{41}$&$\frac{222}{41}$&$\frac{90}{41}$&$\frac{-976}{41}$&$\frac{-576}{41}$&$\frac{126}{41}$\\ &&&&&&&&&&&&&&\\ \hline \end{tabular} } \end{center} \begin{center} {\tiny \textbf{Table 5.} (Theorem 2.1 (iii)) \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|} \hline &&&&&&&&&&&&&&&&\\ $(a_1a_2a_3a_4,b_1b_2)$&$ \gamma_1$ &$ \gamma_2$& $ \gamma_3$& $ \gamma_4$&$ \gamma_5$ &$ \gamma_6$& $ \gamma_7$& $ \gamma_8$&$ \gamma_9$ &$ \gamma_{10}$ & $ \gamma_{11}$&$ \gamma_{12}$& $ \gamma_{13}$&$ \gamma_{14}$& $ \gamma_{15}$&$ \gamma_{16}$\\ \hline &&&&&&&&&&&&&&&&\\ $(1113,11)$&$\frac{1}{23}$&$\frac{288}{23}$&$\frac{32}{23}$&$\frac{9}{23}$&0&0&0&0&$\frac{84}{23}$&$\frac{720}{23}$&$\frac{336}{23}$&$\frac{864}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1113,12)$&$\frac{1}{23}$&$\frac{144}{23}$&$\frac{-16}{23}$&$\frac{-9}{23}$&0&0&0&0&$\frac{156}{23}$&$\frac{-48}{23}$&$\frac{-168}{23}$&$\frac{-456}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1113,14)$&$\frac{1}{23}$&$\frac{72}{23}$&$\frac{8}{23}$&$\frac{9}{23}$&0&0&0&0&$\frac{186}{23}$&$\frac{600}{23}$&$\frac{-228}{23}$&$\frac{372}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1113,22)$&$\frac{1}{23}$&$\frac{72}{23}$&$\frac{8}{23}$&$\frac{9}{23}$&0&0&0&0&$\frac{48}{23}$&$\frac{48}{23}$&$\frac{48}{23}$&$\frac{96}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1113,24)$&$\frac{1}{23}$&$\frac{36}{23}$&$\frac{-4}{23}$&$\frac{-9}{23}$&0&0&0&0&$\frac{114}{23}$&$\frac{84}{23}$&$\frac{-120}{23}$&$\frac{-156}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1113,44)$&$\frac{1}{23}$&$\frac{18}{23}$&$\frac{2}{23}$&$\frac{9}{23}$&0&0&0&0&$\frac{108}{23}$&$\frac{156}{23}$&$\frac{-162}{23}$&$\frac{42}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1223,11)$&0&$\frac{144}{23}$&$\frac{16}{23}$&0&$\frac{1}{23}$&0&0&$\frac{-9}{23}$&$\frac{162}{23}$&$\frac{264}{23}$&$\frac{-1188}{23}$&$\frac{420}{23}$&$\frac{192}{23}$&$\frac{864}{23}$&$\frac{-4704}{23}$&$\frac{-5760}{23}$\\ &&&&&&&&&&&&&&&&\\ $(1223,12)$&0&$\frac{72}{23}$&$\frac{-8}{23}$&0&$\frac{1}{23}$&0&0&$\frac{9}{23}$&$\frac{120}{23}$&$\frac{96}{23}$&$\frac{336}{23}$&$\frac{-384}{23}$&$\frac{-240}{23}$&$\frac{-912}{23}$&$\frac{4368}{23}$&$\frac{2784}{23}$\\ &&&&&&&&&&&&&&&&\\ $(1223,14)$&0&$\frac{36}{23}$&$\frac{4}{23}$&0&$\frac{1}{23}$&0&0&$\frac{-9}{23}$&$\frac{144}{23}$&$\frac{480}{23}$&$\frac{-504}{23}$&$\frac{312}{23}$&$\frac{-360}{23}$&$\frac{1416}{23}$&$\frac{-1944}{23}$&$\frac{-1344}{23}$\\ &&&&&&&&&&&&&&&&\\ $(1223,22)$&0&$\frac{36}{23}$&$\frac{4}{23}$&0&$\frac{1}{23}$&0&0&$\frac{-9}{23}$&$\frac{6}{23}$&$\frac{-72}{23}$&$\frac{-228}{23}$&$\frac{36}{23}$&$\frac{192}{23}$&$\frac{-240}{23}$&$\frac{-1392}{23}$&$\frac{-1344}{23}$\\ &&&&&&&&&&&&&&&&\\ $(1223,24)$&0&$\frac{18}{23}$&$\frac{-2}{23}$&0&$\frac{1}{23}$&0&0&$\frac{9}{23}$&$\frac{30}{23}$&$\frac{24}{23}$&$\frac{84}{23}$&$\frac{-96}{23}$&$\frac{36}{23}$&$\frac{-360}{23}$&$\frac{504}{23}$&$\frac{576}{23}$\\ &&&&&&&&&&&&&&&&\\ $(1223,44)$&0&$\frac{9}{23}$&$\frac{1}{23}$&0&$\frac{1}{23}$&0&0&$\frac{-9}{23}$&$\frac{36}{23}$&$\frac{120}{23}$&$\frac{-126}{23}$&$\frac{78}{23}$&$\frac{-84}{23}$&$\frac{312}{23}$&$\frac{-840}{23}$&$\frac{-240}{23}$\\ &&&&&&&&&&&&&&&&\\ $(1333,11)$&$\frac{1}{23}$&$\frac{96}{23}$&$\frac{-32}{23}$&$\frac{-3}{23}$&0&0&0&0&$\frac{260}{23}$&$\frac{352}{23}$&$\frac{-544}{23}$&$\frac{-352}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1333,12)$&$\frac{1}{23}$&$\frac{48}{23}$&$\frac{16}{23}$&$\frac{3}{23}$&0&0&0&0&$\frac{116}{23}$&$\frac{160}{23}$&$\frac{-40}{23}$&$\frac{104}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1333,14)$&$\frac{1}{23}$&$\frac{24}{23}$&$\frac{-8}{23}$&$\frac{-3}{23}$&0&0&0&0&$\frac{170}{23}$&$\frac{16}{23}$&$\frac{-292}{23}$&$\frac{-340}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1333,22)$&$\frac{1}{23}$&$\frac{24}{23}$&$\frac{-8}{23}$&$\frac{-3}{23}$&0&0&0&0&$\frac{32}{23}$&$\frac{16}{23}$&$\frac{-16}{23}$&$\frac{-64}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1333,24)$&$\frac{1}{23}$&$\frac{12}{23}$&$\frac{4}{23}$&$\frac{3}{23}$&0&0&0&0&$\frac{26}{23}$&$\frac{76}{23}$&$\frac{32}{23}$&$\frac{116}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1333,44)$&$\frac{1}{23}$&$\frac{6}{23}$&$\frac{-2}{23}$&$\frac{-3}{23}$&0&0&0&0&$\frac{44}{23}$&$\frac{-68}{23}$&$\frac{-22}{23}$&$\frac{-130}{23}$&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ \hline \end{tabular} } \end{center} \begin{center} {\tiny \textbf{Table 6.} (Theorem 2.1 (iv)) \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|} \hline &&&&&&&&&&&&&&\\ $a_1a_2a_3$&$ \delta_1$ &$ \delta_2$& $ \delta_3$& $ \delta_4$&$ \delta_5$ &$ \delta_6$& $ \delta_7$& $ \delta_8$&$ \delta_9$ &$ \delta_{10}$ & $ \delta_{11}$&$ \delta_{12}$& $ \delta_{13}$&$ \delta_{14}$\\ $a_4b_1b_2$&$~~$ &$~~$& $~~$& $~~$&$~~$ &$~~$& $~~$& $~~$&$~~$ &$ ~~$ & $~~$&$~~$& $~~$&$~~$\\ \hline &&&&&&&&&&&&&&\\ $112311$&$\frac{1}{261}$&$\frac{256}{29}$&$\frac{-256}{261}$&$\frac{-1}{29}$&$\frac{1808}{87}$&$\frac{656}{29}$&$\frac{-2056}{87}$&$\frac{-3808}{29}$&$\frac{-4144}{29}$&$\frac{736}{3}$&$\frac{472}{3}$&$\frac{-41984}{87}$&$\frac{-1096}{87}$&$\frac{-968}{87}$\\ &&&&&&&&&&&&&&\\ $112312$&$\frac{1}{261}$&$\frac{128}{29}$&$\frac{128}{261}$&$\frac{1}{29}$&$\frac{208}{87}$&$\frac{-32}{29}$&$\frac{-284}{87}$&$\frac{-368}{29}$&$\frac{1048}{29}$&$\frac{-6224}{87}$&$\frac{-7100}{87}$&$\frac{21248}{87}$&$\frac{8}{3}$&$\frac{500}{87}$\\ &&&&&&&&&&&&&&\\ $112314$&$\frac{1}{261}$&$\frac{64}{29}$&$\frac{-64}{261}$&$\frac{-1}{29}$&$\frac{-84}{29}$&$\frac{-114}{29}$&$\frac{450}{29}$&$\frac{1800}{29}$&$\frac{-324}{29}$&$\frac{-4200}{29}$&$\frac{-2814}{29}$&$\frac{-3072}{29}$&$\frac{318}{29}$&$\frac{414}{29}$\\ &&&&&&&&&&&&&&\\ $112322$&$\frac{1}{261}$&$\frac{64}{29}$&$\frac{-64}{261}$&$\frac{-1}{29}$&$\frac{264}{29}$&$\frac{60}{29}$&$\frac{-420}{29}$&$\frac{-1680}{29}$&$\frac{-1368}{29}$&$\frac{2064}{29}$&$\frac{1884}{29}$&$\frac{-3072}{29}$&$\frac{-204}{29}$&$\frac{-108}{29}$\\ &&&&&&&&&&&&&&\\ $112324$&$\frac{1}{261}$&$\frac{32}{29}$&$\frac{32}{261}$&$\frac{1}{29}$&$\frac{860}{87}$&$\frac{218}{29}$&$\frac{-970}{87}$&$\frac{-2296}{29}$&$\frac{-628}{29}$&$\frac{7976}{87}$&$\frac{-778}{87}$&$\frac{4288}{87}$&$\frac{-622}{87}$&$\frac{-10}{3}$\\ &&&&&&&&&&&&&&\\ $112344$&$\frac{1}{261}$&$\frac{16}{29}$&$\frac{-16}{261}$&$\frac{-1}{29}$&$\frac{16}{87}$&$\frac{-176}{29}$&$\frac{244}{87}$&$\frac{592}{29}$&$\frac{-152}{29}$&$\frac{-6992}{87}$&$\frac{-3404}{87}$&$\frac{-1024}{87}$&$\frac{292}{87}$&$\frac{620}{87}$\\ &&&&&&&&&&&&&&\\ $222311$&$\frac{1}{261}$&$\frac{128}{29}$&$\frac{-128}{261}$&$\frac{-1}{29}$&$\frac{5480}{261}$&$\frac{1472}{87}$&$\frac{-15016}{261}$&$\frac{-2992}{87}$&$\frac{-23584}{87}$&$\frac{79664}{261}$&$\frac{102248}{261}$&$\frac{-194048}{261}$&$\frac{-116}{9}$&$\frac{-3704}{261}$\\ &&&&&&&&&&&&&&\\ $222312$&$\frac{1}{261}$&$\frac{64}{29}$&$\frac{64}{261}$&$\frac{1}{29}$&$\frac{-160}{261}$&$\frac{-640}{87}$&$\frac{7172}{261}$&$\frac{-5656}{87}$&$\frac{12320}{87}$&$\frac{-50824}{261}$&$\frac{-47284}{261}$&$\frac{96640}{261}$&$\frac{1076}{261}$&$\frac{964}{261}$\\ &&&&&&&&&&&&&&\\ $222314$&$\frac{1}{261}$&$\frac{32}{29}$&$\frac{-32}{261}$&$\frac{-1}{29}$&$\frac{824}{261}$&$\frac{-466}{87}$&$\frac{4970}{261}$&$\frac{-1888}{87}$&$\frac{-628}{87}$&$\frac{-24496}{261}$&$\frac{-10030}{261}$&$\frac{-44672}{261}$&$\frac{494}{261}$&$\frac{394}{261}$\\ &&&&&&&&&&&&&&\\ $222322$&$\frac{1}{261}$&$\frac{32}{29}$&$\frac{-32}{261}$&$\frac{-1}{29}$&$\frac{824}{261}$&$\frac{1100}{87}$&$\frac{-9124}{261}$&$\frac{2288}{87}$&$\frac{-10024}{87}$&$\frac{50672}{261}$&$\frac{32252}{261}$&$\frac{-44672}{261}$&$\frac{-1072}{261}$&$\frac{-1172}{261}$\\ &&&&&&&&&&&&&&\\ $222324$&$\frac{1}{261}$&$\frac{16}{29}$&$\frac{16}{261}$&$\frac{1}{29}$&$\frac{-748}{261}$&$\frac{518}{87}$&$\frac{-2470}{261}$&$\frac{2936}{87}$&$\frac{-1156}{87}$&$\frac{11192}{261}$&$\frac{-8830}{261}$&$\frac{21088}{261}$&$\frac{578}{261}$&$\frac{562}{261}$\\ &&&&&&&&&&&&&&\\ $222344$&$\frac{1}{261}$&$\frac{8}{29}$&$\frac{-8}{261}$&$\frac{-1}{29}$&$\frac{-340}{261}$&$\frac{224}{87}$&$\frac{-604}{261}$&$\frac{1520}{87}$&$\frac{-1936}{87}$&$\frac{5840}{261}$&$\frac{-6388}{261}$&$\frac{-7328}{261}$&$\frac{284}{261}$&$\frac{244}{261}$\\ &&&&&&&&&&&&&&\\ $233311$&$\frac{1}{261}$&$\frac{256}{87}$&$\frac{256}{261}$&$\frac{1}{87}$&$\frac{-13360}{261}$&$\frac{-3520}{87}$&$\frac{20816}{261}$&$\frac{12608}{29}$&$\frac{20960}{87}$&$\frac{-219712}{261}$&$\frac{-113968}{261}$&$\frac{133120}{261}$&$\frac{15464}{261}$&$\frac{16384}{261}$\\ &&&&&&&&&&&&&&\\ $233312$&$\frac{1}{261}$&$\frac{128}{87}$&$\frac{-128}{261}$&$\frac{-1}{87}$&$\frac{11168}{261}$&$\frac{2312}{87}$&$\frac{-14212}{261}$&$\frac{-27920}{87}$&$\frac{-14968}{87}$&$\frac{129296}{261}$&$\frac{45212}{261}$&$\frac{-64256}{261}$&$\frac{-9856}{261}$&$\frac{-8948}{261}$\\ &&&&&&&&&&&&&&\\ $233314$&$\frac{1}{261}$&$\frac{64}{87}$&$\frac{64}{261}$&$\frac{1}{87}$&$\frac{3836}{261}$&$\frac{338}{87}$&$\frac{-4126}{261}$&$\frac{-2968}{29}$&$\frac{-2092}{87}$&$\frac{21368}{261}$&$\frac{-10174}{261}$&$\frac{32512}{261}$&$\frac{-2530}{261}$&$\frac{-1514}{261}$\\ &&&&&&&&&&&&&&\\ $233322$&$\frac{1}{261}$&$\frac{64}{87}$&$\frac{64}{261}$&$\frac{1}{87}$&$\frac{-5560}{261}$&$\frac{-1228}{87}$&$\frac{6836}{261}$&$\frac{4688}{29}$&$\frac{7304}{87}$&$\frac{-72592}{261}$&$\frac{-24268}{261}$&$\frac{32512}{261}$&$\frac{5300}{261}$&$\frac{6316}{261}$\\ &&&&&&&&&&&&&&\\ $233324$&$\frac{1}{261}$&$\frac{32}{87}$&$\frac{-32}{261}$&$\frac{-1}{87}$&$\frac{-940}{261}$&$\frac{-274}{87}$&$\frac{842}{261}$&$\frac{2584}{87}$&$\frac{-76}{87}$&$\frac{-12808}{261}$&$\frac{2666}{261}$&$\frac{-14528}{261}$&$\frac{878}{261}$&$\frac{1882}{261}$\\ &&&&&&&&&&&&&&\\ $233344$&$\frac{1}{261}$&$\frac{16}{87}$&$\frac{16}{261}$&$\frac{1}{87}$&$\frac{1088}{261}$&$\frac{128}{87}$&$\frac{-2140}{261}$&$\frac{-1120}{29}$&$\frac{-808}{87}$&$\frac{11168}{261}$&$\frac{5204}{261}$&$\frac{7360}{261}$&$\frac{-1156}{261}$&$\frac{-4}{9}$\\ &&&&&&&&&&&&&&\\ \hline \end{tabular} } \end{center} \begin{center} {\tiny \textbf{Table 7.} (Theorem 2.2) \begin{tabular}{|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|c|} \hline &&&&&&&&&&&&&&&&\\ $(1,c_1,c_2,c_3)$&$ \nu_1$ &$ \nu_2$& $ \nu_3$& $ \nu_4$&$ \nu_5$ &$ \nu_6$& $ \nu_7$& $ \nu_8$&$ \nu_9$ &$ \nu_{10}$ & $ \nu_{11}$&$ \nu_{12}$& $ \nu_{13}$&$ \nu_{14}$& $ \nu_{15}$&$ \nu_{16}$\\ \hline &&&&&&&&&&&&&&&&\\ $(1,1,1,2)$&$\frac{3}{40}$&$\frac{-1}{5}$&$\frac{-27}{40}$&0&$\frac{9}{5}$&0&0&0&0&0&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,1,1,4)$&$\frac{3}{100}$&$\frac{-9}{100}$&$\frac{27}{100}$&$\frac{4}{25}$&$\frac{-81}{100}$&0&$\frac{36}{25}$&0&$\frac{54}{5}$&$\frac{432}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,1,1,8)$&$\frac{3}{160}$&$\frac{-9}{160}$&$\frac{-27}{160}$&$\frac{9}{80}$&$\frac{81}{160}$&$\frac{-1}{5}$&$\frac{-81}{80}$&$\frac{9}{5}$&0&0&0&$\frac{-27}{4}$&$\frac{243}{4}$&0&81&$\frac{81}{4}$\\ &&&&&&&&&&&&&&&&\\ $(1,1,2,2)$&$\frac{1}{50}$&$\frac{2}{25}$&$\frac{9}{50}$&0&$\frac{18}{25}$&0&0&0&$\frac{36}{5}$&0&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,1,2,4)$&$\frac{1}{80}$&$\frac{1}{16}$&$\frac{-9}{80}$&$\frac{-1}{5}$&$\frac{-9}{16}$&0&$\frac{9}{5}$&0&0&0&0&0&0&9&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,1,2,8)$&$\frac{1}{200}$&$\frac{1}{40}$&$\frac{9}{200}$&$\frac{-9}{100}$&$\frac{9}{40}$&$\frac{4}{25}$&$\frac{-81}{100}$&$\frac{36}{25}$&$\frac{9}{5}$&$\frac{144}{5}$&$\frac{1152}{5}$&9&81&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,1,4,4)$&$\frac{1}{200}$&$\frac{3}{200}$&$\frac{9}{200}$&$\frac{2}{25}$&$\frac{27}{200}$&0&$\frac{18}{25}$&0&$\frac{54}{5}$&$\frac{216}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,1,4,8)$&$\frac{1}{320}$&$\frac{3}{320}$&$\frac{-9}{320}$&$\frac{1}{16}$&$\frac{-27}{320}$&$\frac{-1}{5}$&$\frac{-9}{16}$&$\frac{9}{5}$&0&0&0&$\frac{9}{4}$&$\frac{-81}{4}$&$\frac{9}{4}$&27&$\frac{27}{4}$\\ &&&&&&&&&&&&&&&&\\ $(1,1,8,8)$&$\frac{1}{800}$&$\frac{3}{800}$&$\frac{9}{800}$&$\frac{3}{200}$&$\frac{27}{800}$&$\frac{2}{25}$&$\frac{27}{200}$&$\frac{18}{25}$&$\frac{36}{5}$&$\frac{234}{5}$&$\frac{576}{5}$&$\frac{9}{2}$&$\frac{81}{2}$&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,2,2,2)$&$\frac{1}{40}$&$\frac{-3}{20}$&$\frac{-9}{40}$&0&$\frac{27}{20}$&0&0&0&0&0&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,2,2,4)$&$\frac{1}{100}$&$\frac{-7}{100}$&$\frac{9}{100}$&$\frac{4}{25}$&$\frac{-63}{100}$&0&$\frac{36}{25}$&0&$\frac{18}{5}$&$\frac{72}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,2,2,8)$&$\frac{1}{160}$&$\frac{-7}{160}$&$\frac{-9}{160}$&$\frac{9}{80}$&$\frac{63}{160}$&$\frac{-1}{5}$&$\frac{-81}{80}$&$\frac{9}{5}$&0&0&0&$\frac{-9}{4}$&$\frac{81}{4}$&0&9&$\frac{27}{4}$\\ &&&&&&&&&&&&&&&&\\ $(1,2,4,4)$&$\frac{1}{160}$&$\frac{-1}{32}$&$\frac{-9}{160}$&$\frac{-1}{10}$&$\frac{9}{32}$&0&$\frac{9}{10}$&0&0&0&0&0&0&$\frac{9}{2}$&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,2,4,8)$&$\frac{1}{400}$&$\frac{-1}{80}$&$\frac{9}{400}$&$\frac{-1}{20}$&$\frac{-9}{80}$&$\frac{4}{25}$&$\frac{-9}{20}$&$\frac{36}{25}$&$\frac{9}{10}$&$\frac{27}{5}$&$\frac{72}{5}$&$\frac{9}{2}$&$\frac{81}{2}$&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,2,8,8)$&$\frac{1}{640}$&$\frac{-1}{128}$&$\frac{-9}{640}$&$\frac{-3}{160}$&$\frac{9}{128}$&$\frac{-1}{10}$&$\frac{27}{160}$&$\frac{9}{10}$&0&0&0&$\frac{-9}{8}$&$\frac{81}{8}$&$\frac{27}{8}$&$\frac{9}{2}$&$\frac{27}{8}$\\ &&&&&&&&&&&&&&&&\\ $(1,4,4,4)$&$\frac{1}{400}$&$\frac{-9}{400}$&$\frac{9}{400}$&$\frac{3}{25}$&$\frac{-81}{400}$&0&$\frac{27}{25}$&0&$\frac{27}{5}$&$\frac{54}{5}$&0&0&0&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,4,4,8)$&$\frac{1}{640}$&$\frac{-9}{640}$&$\frac{-9}{640}$&$\frac{7}{80}$&$\frac{81}{640}$&$\frac{-1}{5}$&$\frac{-63}{80}$&$\frac{9}{5}$&0&0&0&$\frac{9}{8}$&$\frac{-81}{8}$&$\frac{9}{8}$&0&$\frac{27}{8}$\\ &&&&&&&&&&&&&&&&\\ $(1,4,8,8)$&$\frac{1}{1600}$&$\frac{-9}{1600}$&$\frac{9}{1600}$&$\frac{1}{40}$&$\frac{-81}{1600}$&$\frac{2}{25}$&$\frac{9}{40}$&$\frac{18}{25}$&$\frac{18}{5}$&$\frac{36}{5}$&$\frac{36}{5}$&$\frac{9}{4}$&$\frac{81}{4}$&0&0&0\\ &&&&&&&&&&&&&&&&\\ $(1,8,8,8)$&$\frac{1}{2560}$&$\frac{-9}{2560}$&$\frac{-9}{2560}$&$\frac{9}{320}$&$\frac{81}{2560}$&$\frac{-3}{20}$&$\frac{-81}{320}$&$\frac{27}{20}$&0&0&0&$\frac{27}{32}$&$\frac{-243}{32}$&$\frac{81}{32}$&0&$\frac{81}{32}$\\ &&&&&&&&&&&&&&&&\\ \hline \end{tabular} } \end{center} \vglue -0.7cm \subsection{Sample formulas} In this section we shall give explicit formulas for a few cases from Tables 1 and 2. \noindent {\bf First two formulas of \thmref{1}(i):} \begin{equation*} \begin{split} {\bf N(1,1,1,1,1,1;n)}&=\frac{112}{5}\sigma_{3}(n)-\frac{84}{5} \sigma_{3}(n/2)- \frac{432}{5} \sigma_{3}(n/3)-\frac{448}{5} \sigma_{3}(n/4) +\frac{324}{5}\sigma_{3}(n/6)\\ &+ \frac{1728}{5} \sigma_{3}(n/12)-\frac{72}{5} a_{4,6}(n)-\frac{288}{5}a_{4,6}(n/2)+12 a_{4,12}(n),\\ \end{split} \end{equation*} \begin{equation*} \begin{split} {\bf N(1,1,1,1,1,2;n)}&=\frac{52}{5}\sigma_{3}(n)-\frac{78}{5} \sigma_{3}(n/2)+ \frac{108}{5} \sigma_{3}(n/3)+\frac{416}{5} \sigma_{3}(n/4) -\frac{324}{5}\sigma_{3}(n/6)\\ &+ \frac{864}{5} \sigma_{3}(n/12)+\frac{48}{5} a_{4,6}(n)+\frac{96}{5}a_{4,6}(n/2)-6 a_{4,12}(n).\\ \end{split} \end{equation*} \noindent {\bf First two formulas of \thmref{1}(ii):} \begin{equation*} \begin{split} {\bf N(1,1,1,2,1,1;n)}&=-\frac{26}{451} \sigma_{3;{\bf 1},\chi_{8}}(n)+\frac{108}{451} \sigma_{3;{\bf 1},\chi_{8}}(n/3)+\frac{6656}{451} \sigma_{3;\chi_{8},{\bf 1}}(n)+ \frac{27648}{451} \sigma_{3;\chi_{8},{\bf 1}}(n/3)\\ & + \frac{168}{451} a_{4,8,\chi_8;1}(n)+\frac{11448}{451} a_{4,8,\chi_8;1}(n/3)- \frac{2496}{451} a_{4,8,\chi_8;2}(n)-\frac{17280}{451} a_{4,8,\chi_8;2}(n/3)\\ &+ \frac{24}{41} a_{4,24,\chi_8;1}(n)+\frac{936}{41} a_{4,24,\chi_8;2}(n)+\frac{144}{41} a_{4,24,\chi_8;3}(n)- \frac{384}{41} a_{4,24,\chi_8;4}(n)\\ &+ \frac{4032}{41} a_{4,24,\chi_8;5}(n)-\frac{48}{41} a_{4,24,\chi_8;6}(n),\\ \end{split} \end{equation*} \begin{equation*} \begin{split} {\bf N(1,1,1,2,1,2;n)}&=\frac{28}{451} \sigma_{3;{\bf 1},\chi_{8}}(n)+\frac{54}{451} \sigma_{3;{\bf 1},\chi_{8}}(n/3)+\frac{3584}{451} \sigma_{3;\chi_{8},{\bf 1}}(n)- \frac{6912}{451} \sigma_{3;\chi_{8},{\bf 1}}(n/3)\\ & +\frac{480}{451} a_{4,8,\chi_8;1}(n)-\frac{2052}{451} a_{4,8,\chi_8;1}(n/3)-\frac{2688}{451} a_{4,8,\chi_8;2}(n)+\frac{1728}{451} a_{4,8,\chi_8;2}(n/3)\\ &- \frac{60}{41} a_{4,24,\chi_8;1}(n)+\frac{216}{41} a_{4,24,\chi_8;2}(n)-\frac{108}{41} a_{4,24,\chi_8;3}(n)- \frac{2112}{41} a_{4,24,\chi_8;4}(n)\\ &-\frac{1440}{41} a_{4,24,\chi_8;5}(n)+\frac{288}{41} a_{4,24,\chi_8;6}(n).\\ \end{split} \end{equation*} \smallskip \noindent {\bf First two formulas of \thmref{1}(iii):} \begin{equation*} \begin{split} {\bf N(1,1,1,3,1,1;n)}&=\frac{1}{23} \sigma_{3;{\bf 1},\chi_{12},}(n)+\frac{288}{23} \sigma_{3;\chi_{12},{\bf 1},}(n)+\frac{32}{23} \sigma_{3;\chi_{-4},\chi_{-3}}(n)+\frac{9}{23} \sigma_{3;\chi_{-3},\chi_{-4}}(n)\\ &+ \frac{84}{23}a_{4,12,\chi_{12};1}(n)+\frac{720}{23} a_{4,12,\chi_{12};2}(n)+ \frac{336}{23}a_{4,12,\chi_{12};3}(n)+\frac{864}{23}a_{4,12,\chi_{12};4}(n),\\ \end{split} \end{equation*} \begin{equation*} \begin{split} {\bf N(1,1,1,3,1,2;n)}&=\frac{1}{23} \sigma_{3;{\bf 1},\chi_{12},}(n)+\frac{144}{23} \sigma_{3;\chi_{12},{\bf 1},}(n)-\frac{16}{23} \sigma_{3;\chi_{-4},\chi_{-3}}(n)-\frac{9}{23} \sigma_{3;\chi_{-3},\chi_{-4}}(n)\\ &+ \frac{156}{23}a_{4,12,\chi_{12};1}(n)-\frac{48}{23} a_{4,12,\chi_{12};2}(n)- \frac{168}{23}a_{4,12,\chi_{12};3}(n)-\frac{456}{23}a_{4,12,\chi_{12};4}(n).\\ \end{split} \end{equation*} \smallskip \noindent {\bf First two formulas of \thmref{1}(iv):} \begin{equation*} \begin{split} {\bf N(1,1,2,3,1,1;n)}&=\frac{1}{261} \sigma_{3;{\bf 1},\chi_{24}}(n)+ \frac{256}{29} \sigma_{3;\chi_{24},{\bf 1}}(n)-\frac{256}{261} \sigma_{3;\chi_{-8},\chi_{-3}}(n)-\frac{1}{29} \sigma_{3;\chi_{-3},\chi_{-8}}(n)\\ &+ \frac{1808}{87} a_{4,24,\chi_{24};1}(n)+\frac{656}{29} a_{4,24,\chi_{24};2}(n)-\frac{2056}{87} a_{4,24,\chi_{24};3}(n)-\frac{3808}{29} a_{4,24,\chi_{24};4}(n)\\ & - \frac{4144}{29} a_{4,24,\chi_{24};5}(n)+\frac{736}{3} a_{4,24,\chi_{24};6}(n)+ \frac{472}{3} a_{4,24,\chi_{24};7}(n)-\frac{41984}{87} a_{4,24,\chi_{24};8}(n)\\ &- \frac{1096}{87} a_{4,24,\chi_{24};9}(n)-\frac{968}{87} a_{4,24,\chi_{24};10}(n),\\ \end{split} \end{equation*} \begin{equation*} \begin{split} {\bf N(1,1,2,3,1,2;n)}&=\frac{1}{261} \sigma_{3;{\bf 1},\chi_{24}}(n)+ \frac{128}{29} \sigma_{3;\chi_{24},{\bf 1}}(n)+\frac{128}{261} \sigma_{3;\chi_{-8},\chi_{-3}}(n)+\frac{1}{29} \sigma_{3;\chi_{-3},\chi_{-8}}(n)\\ &+\frac{208}{87} a_{4,24,\chi_{24};1}(n)-\frac{32}{29} a_{4,24,\chi_{24};2}(n)-\frac{284}{87} a_{4,24,\chi_{24};3}(n)-\frac{368}{29} a_{4,24,\chi_{24};4}(n)\\ &+ \frac{1048}{29} a_{4,24,\chi_{24};5}(n)-\frac{6224}{87} a_{4,24,\chi_{24};6}(n)-\frac{7100}{87} a_{4,24,\chi_{24};7}(n)+\frac{21248}{87} a_{4,24,\chi_{24};8}(n)\\ &+\frac{8}{3} a_{4,24,\chi_{24};9}(n)+\frac{500}{87} a_{4,24,\chi_{24};10}(n).\\ \end{split} \end{equation*} \smallskip \noindent {\bf First two formulas of \thmref{2}:} \begin{equation*} \begin{split} {\bf M(1,1,1,2;n)}& = 18 \sigma_{3}(n)- 48 \sigma_{3}(n/2)- 162 \sigma_{3}(n/3) + 432 \sigma_{3}(n/6),\\ {\bf M(1,1,1,4;n)}&=\frac{36}{5}\sigma_{3}(n)- 48 \sigma_{3}(n/2)+\frac{324}{5} \sigma_{3}(n/3)+\frac{192}{5} \sigma_{3}(n/4) -\frac{972}{5}\sigma_{3}(n/6)\\ &+\frac{1728}{5} \sigma_{3}(n/12)+\frac{54}{5} a_{4,6}(n)+\frac{432}{5}a_{4,6}(n/2).\\ \end{split} \end{equation*} \smallskip \bigskip {\small \noindent {\bf Acknowledgements.} ~ We have used the open-source mathematics software SAGE (www.sagemath.org) to perform our calculations. Brundaban Sahu is partially funded by SERB grant SR/FTP/MS-053/2012. Anup Kumar Singh thanks the Department of Mathematics, NISER, Bhubaneswar (where part of this work is done) for their warm hospitality and support. } \smallskip
d0819eb72ed67bc6b70ad623097618d7d680d039
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Artificial neural networks are currently considered the state of the art in applications ranging from image classification, to speech recognition and even machine translation. However, little is understood about the process by which they are trained for supervised learning tasks. The problem of optimizing their parameters is an active area both practical and theoretical research. Despite considerable sensitivity to initialization and choice of hyperparameters, neural networks often achieve compelling results after optimization by gradient descent methods. Due to the nonconvexity and massive parameter space of these functions, it is poorly understood how these sub-optimal methods have proven so successful. Indeed, training a certain kind of neural network is known to be NP-Complete, making it difficult to provide any worst-case training guarantees \cite{Blum:1992:OCT:148433.148441}. Much recent work has attempted to reconcile these differences between theory and practice \cite{Kawaguchi:2016:WithoutPoorLocalMinima,Soudry:2016:NoBadLocalMinima}. This article attempts a modest step towards understanding the dynamics of the training procedure. We establish three main convexity results for a certain class of neural network, which is the current the state of the art. First, that the objective is piecewise convex as a function of the input data, with parameters fixed, which corresponds to the behavior at test time. Second, that the objective is again piecewise convex as a function of the parameters of a single layer, with the input data and all other parameters held constant. Third, that the training objective function, for which all parameters are variable but the input data is fixed, is piecewise multi-convex. That is, it is a continuous function which can be represented by a finite number of multi-convex functions, each active on a multi-convex parameter set. This generalizes the notion of biconvexity found in the optimization literature to piecewise functions and arbitrary index sets \cite{Gorski:2007:BiConvex}. To prove these results, we require two main restrictions on the definition of a neural network: that its layers are piecewise affine functions, and that its objective function is convex and continuously differentiable. Our definition includes many contemporary use cases, such as least squares or logistic regression on a convolutional neural network with rectified linear unit (ReLU) activation functions and either max- or mean-pooling. In recent years these networks have mostly supplanted the classic sigmoid type, except in the case of recurrent networks \cite{Glorot:2011:ReluNetworks}. We make no assumptions about the training data, so our results apply to the current state of the art in many practical scenarios. Piecewise multi-convexity allows us to characterize the extrema of the training objective. As in the case of biconvex functions, stationary points and local minima are guaranteed optimality on larger sets than we would have for general smooth functions. Specifically, these points are partial minima when restricted to the relevant piece. That is, they are points for which no decrease can be made in the training objective without simultaneously varying the parameters across multiple layers, or crossing the boundary into a different piece of the function. Unlike global minima, we show that partial minima are reliably found by the optimization algorithms used in current practice. Finally, we provide some guarantees for solving general multi-convex optimization problems by various algorithms. First we analyze gradient descent, proving necessary convergence conditions. We show that every point to which gradient descent converges is a piecewise partial minimum, excepting some boundary conditions. To prove stronger results, we define a different optimization procedure breaking each parameter update into a number of convex sub-problems. For this procedure, we show both necessary and sufficient conditions for convergence to a piecewise partial minimum. Interestingly, adding regularization to the training objective is all that is needed to prove necessary conditions. Similar results have been independently established for many kinds of optimization problems, including bilinear and biconvex optimization, and in machine learning the special case of linear autoencoders \cite{Wendell:1976:Bilinear,Gorski:2007:BiConvex,Baldi:2012:ComplexValuedAutoencoders}. Our analysis extends existing results on alternating convex optimization to the case of arbitrary index sets, and general multi-convex point sets, which is needed for neural networks. We admit biconvex problems, and therefore linear autoencoders, as a special case. Despite these results, we find that it is difficult to pass from partial to global optimality results. Unlike the encouraging case of linear autoencoders, we show that a single rectifier neuron, under the squared error objective, admits arbitrarily poor local minima. This suggests that much work remains to be done in understanding how sub-optimal methods can succeed with neural networks. Still, piecewise multi-convex functions are in some senses easier to minimize than the general class of smooth functions, for which none of our previous guarantees can be made. We hope that our characterization of neural networks could contribute to a better understanding of these important machine learning systems. \section{Preliminary material} We begin with some preliminary definitions and basic results concerning continuous piecewise functions. \begin{definition}\label{def:piecewise_affine} Let $g_{1},g_{2},...,g_{N}$ be continuous functions from $\mathbb{R}^{n}\rightarrow\mathbb{R}$. A \textbf{continuous piecewise} function $f$ has a finite number of closed, connected sets $S_{1},S_{2},...,S_{M}$ covering $\mathbb{R}^{n}$ such that for each $k$ we have $f(\boldsymbol{x})=g_{k}(\boldsymbol{x})$ for all $\boldsymbol{x}\in S_{k}$. The set $S_{k}$ is called a \textbf{piece }of $f$, and the function $g_{k}$ is called \textbf{active} on $S_{k}$. More specific definitions follow by restricting the functions $g$. A \textbf{continuous piecewise affine }function has $g_{k}(\boldsymbol{x})=\boldsymbol{a}^{T}\boldsymbol{x}+b$ where $\boldsymbol{a}\in\mathbb{R}^{n}$ and $b\in\mathbb{R}$. A \textbf{continuous piecewise convex} function has $g_{k}$ convex, with $S_{k}$ convex as well. \end{definition} Note that this definition of piecewise convexity differs from that found in the convex optimization literature, which focuses on \textit{convex} piecewise convex functions, i.e.~maxima of convex functions \cite{Tsevendorj:2001:PiecewiseConvex}. Note also that we do not claim a unique representation in terms of active functions $g_{k}$ and pieces $S_{k}$, only that there exists at least one such representation. Before proceeding, we shall extend definition \ref{def:piecewise_affine} to functions of multidimensional codomain for the affine case. \begin{definition}\label{def:piecewise_affine_onto_Rn} A function $f:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$, and let $f_{k}:\mathbb{R}^{m}\rightarrow\mathbb{R}$ denote the $k^{th}$ component of $f$. Then $f$ is\textbf{ continuous piecewise affine} if each $f_{k}$ is. Choose some piece $S_{k}$ from each $f_{k}$ and let $S=\cap_{k=1}^{n}S_{k}$, with $S\ne\emptyset$. Then $S$ is a piece of $f$, on which we have $f(\boldsymbol{x})=A\boldsymbol{x}+\boldsymbol{b}$ for some $A\in\mathbb{R}^{n\times m}$ and $\boldsymbol{b}\in\mathbb{R}^{n}$. \end{definition} First, we prove an intuitive statement about the geometry of the pieces of continuous piecewise affine functions. \begin{theorem}\label{theorem:convex_polytope} Let $f:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$ be continuous piecewise affine. Then $f$ admits a representation in which every piece is a convex polytope. \end{theorem} \begin{proof} Let $f_{k}:\mathbb{R}^{m}\rightarrow\mathbb{R}$ denote the $k^{th}$ component of $f$. Now, $f_{k}$ can be written in closed form as a max-min polynomial \cite{Ovchinnikov:2002:pwl_max_min}. That is, $f_{k}$ is the maximum of minima of its active functions. Now, for the minimum of two affine functions we have \[ \min(g_{i},g_{j})=\min(\boldsymbol{a}_{i}^{T}\boldsymbol{x}+b_{i},\boldsymbol{a}_{j}^{T}\boldsymbol{x}+b_{j}). \] This function has two pieces divided by the hyperplane $(\boldsymbol{a}_{i}^{T}-\boldsymbol{a}_{j}^{T})\boldsymbol{x}+b_{i}-b_{j}=0$. The same can be said of $\max(g_{i},g_{j})$. Thus the pieces of $f_{k}$ are intersections of half-spaces, which are just convex polytopes. Since the pieces of $f$ are intersections of the pieces of $f_{k}$, they are convex polytopes as well. \end{proof} See figure \ref{fig:bad_local_minimum} in section \ref{sec:Local-minima} for an example of this result on a specific neural network. Our next result concerns the composition of piecewise functions, which is essential for the later sections. \begin{theorem}\label{theorem:composition_piecewise_affine} Let $g:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$ and $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ be continuous piecewise affine. Then so is $f\circ g$. \end{theorem} \begin{proof} To establish continuity, note that the composition of continuous functions is continuous. Let $S$ be a piece of $g$ and $T$ a piece of $f$ such that $S\cap g^{-1}(T)\ne\emptyset$, where $g^{-1}(T)$ denotes the inverse image of $T$. By theorem \ref{theorem:convex_polytope}, we can choose $S$ and $T$ to be convex polytopes. Since $g$ is affine, $g^{-1}(T)$ is closed and convex \cite{Boyd:2004:ConvexOptimization}. Thus $S\cap g^{-1}(T)$ is a closed, convex set on which we can write \begin{align} f(\boldsymbol{x}) & =\boldsymbol{a}^{T}\boldsymbol{x}+b\label{eq:pwa1}\\ g(\boldsymbol{x}) & =C\boldsymbol{x}+\boldsymbol{d}.\nonumber \end{align} Thus \begin{align} f\circ g(\boldsymbol{x}) & =\boldsymbol{a}^{T}C\boldsymbol{x}+\boldsymbol{a}^{T}\boldsymbol{d}+b\label{eq:pwa2} \end{align} which is an affine function. Now, consider the finite set of all such pieces $S\cap g^{-1}(T)$. The union of $g^{-1}(T)$ over all pieces $T$ is just $\mathbb{R}^{n}$, as is the union of all pieces $S$. Thus we have \begin{align*} \cup_{S}\cup_{T}\left(S\cap g^{-1}(T)\right) & =\cup_{S}\left(S\cap\cup_{T}g^{-1}(T)\right)\\ & =\cup_{S}\left(S\cap\mathbb{R}^{n}\right)\\ & =\mathbb{R}^{n}. \end{align*} Thus $f\circ g$ is piecewise affine on $\mathbb{R}^{n}$. \end{proof} We now turn to continuous piecewise convex functions, of which continuous piecewise affine functions are a subset. \begin{theorem}\label{theorem:covex_piecewise} Let $g:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$ be a continuous piecewise affine function, and $h:\mathbb{R}^{n}\rightarrow\mathbb{R}$ a convex function. Then $f=h\circ g$ is continuous piecewise convex. \end{theorem} \begin{proof} On each piece $S$ of $g$ we can write \[ f(\boldsymbol{x})=h(A\boldsymbol{x}+\boldsymbol{b}). \] This function is convex, as it is the composition of a convex and an affine function \cite{Boyd:2004:ConvexOptimization}. Furthermore, $S$ is convex by theorem \ref{theorem:convex_polytope}. This establishes piecewise convexity by the proof of theorem \ref{theorem:composition_piecewise_affine}. \end{proof} Our final theorem concerns the arithmetic mean of continuous piecewise convex functions, which is essential for the analysis of neural networks. \begin{theorem}\label{theorem:mean_piecewise_convex} Let $f_{1},f_{2},...,f_{N}$ be continuous piecewise convex functions. Then so is their arithmetic mean $(1/N)\sum_{i=1}^{N}f_{i}(\boldsymbol{x})$. \end{theorem} The proof takes the form of two lemmas. \begin{lemma}\label{lemma:sum_piecewise_convex} Let $f_{1}$ and $f_{2}$ be a pair of continuous piecewise convex functions on $\mathbb{R}^{n}$. Then so is $f_{1}+f_{2}$. \end{lemma} \begin{proof} Let $S_{1}$ be a piece of $f_{1}$, and $S_{2}$ a piece of $f_{2}$, with $S_{1}\cap S_{2}\ne\emptyset$. Note that the sum of convex functions is convex \cite{Rockafellar:1970:ConvexAnalysis}. Thus $f_{1}+f_{2}$ is convex on $S_{1}\cap S_{2}$. Furthermore, $S_{1}\cap S_{2}$ is convex because it is an intersection of convex sets \cite{Rockafellar:1970:ConvexAnalysis}. Since this holds for all pieces of $f_{1}$ and $f_{2}$, we have that $f_{1}+f_{2}$ is continuous piecewise convex on $\mathbb{R}^{n}$. \end{proof} \begin{lemma} Let $\alpha>0$, and let $f$ be a continuous piecewise convex function. Then so is $\alpha f$. \end{lemma} \begin{proof} The continuous function $\alpha f$ is convex on every piece of $f$. \end{proof} Having established that continuous piecewise convexity is closed under addition and positive scalar multiplication, we can see that it is closed under the arithmetic mean, which is just the composition of these two operations. \section{Neural networks\label{sec:Neural-networks}} In this work, we define a neural network to be a composition of functions of two kinds: a convex continuously differentiable objective (or loss) function $h$, and continuous piecewise affine functions $g_{1},g_{2},...,g_{N}$, constituting the $N$ layers. Furthermore, the outermost function must be $h$, so that we have \[ f=h\circ g_{N}\circ g_{N-1}\circ...\circ g_{1} \] where $f$ denotes the entire network. This definition is not as restrictive as it may seem upon first glance. For example, it is easily verified that the rectified linear unit (ReLU) neuron is continuous piecewise affine, as we have \[ g(\boldsymbol{x})=\max(\boldsymbol{0},A\boldsymbol{x}+\boldsymbol{b}), \] where the maximum is taken pointwise. It can be shown that maxima and minima of affine functions are piecewise affine \cite{Ovchinnikov:2002:pwl_max_min}. This includes the convolutional variant, in which $A$ is a Toeplitz matrix. Similarly, max pooling is continuous piecewise linear, while mean pooling is simply linear. Furthermore, many of the objective functions commonly seen in machine learning are convex and continuously differentiable, as in least squares and logistic regression. Thus this seemingly restrictive class of neural networks actually encompasses the current state of the art. By theorem \ref{theorem:composition_piecewise_affine}, the composition of all layers $g=g_{N}\circ g_{N-1}\circ...\circ g_{1}$ is continuous piecewise affine. Therefore, a neural network is ultimately the composition of a continuous convex function with a single continuous piecewise affine function. Thus by theorem \ref{theorem:covex_piecewise} the network is continuous piecewise convex. Figure \ref{fig:network} provides a visualization of this result for the example network \begin{equation} f(x,y)=\left(2-\left[\left[x-y\right]_{+}-\left[x+y\right]_{+}+1\right]_{+}\right)^{2},\label{eq:example_network} \end{equation} where $\left[x\right]_{+}=\max(x,0)$. For clarity, this is just the two-layer ReLU network \[ f(x,y,z)=\left(z-\left[a_{5}\left[a_{1}x+a_{2}y\right]_{+}+a_{6}\left[a_{3}x+a_{4}y\right]_{+}+b_{1}\right]_{+}\right)^{2} \] with the squared error objective and a single data point $((x,y),z)$, setting $z=2$ and $a_{2}=a_{6}=-1$, with all other parameters set to $1$. \begin{figure} \begin{centering} \includegraphics[scale=0.5]{network} \par\end{centering} \centering{}\caption{The neural network of equation \ref{eq:example_network}, on the unit square. Although $f$ is not convex on $\mathbb{R}^{2}$, it is convex in each piece, and each piece is a convex set.} \label{fig:network} \end{figure} Before proceeding further, we must define a special kind of differentiability for piecewise continuous functions, and show that this holds for neural networks. \begin{definition} Let $f$ be piecewise continuous. We say that $f$ is \textbf{piecewise continuously differentiable} if each active function $g$ is continuously differentiable. \end{definition} To see that neural networks are piecewise continuously differentiable, note that the objective $h$ is continuously differentiable, as are the affine active functions of the layers. Thus their composition is continuously differentiable. It follows that non-differentiable points are found only on the boundaries between pieces. \section{Network parameters of a single layer\label{sec:Network-parameters-of-single-layer}} In the previous section we have defined neural networks as functions of labeled data. These are the functions relevant during testing, where parameters are constant and data is variable. In this section, we extend these results to the case where data is constant and parameters are variable, which is the function to optimized during training. For example, consider the familiar equation \[ f=(ax+b-y)^{2} \] with parameters $(a,b)$ and data $(x,y$). During testing, we hold $(a,b)$ constant, and consider $f$ as a function of the data $(x,y)$. During training, we hold $(x,y)$ constant and consider $f$ as a function of the parameters $(a,b)$. This is what we mean when we say that a network is being ``considered as a function of its parameters\footnote{This is made rigorous by taking cross-sections of point sets in section \ref{sec:Network-parameters-of-multiple-layers}. }.'' This leads us to an additional stipulation on our definition of a neural network. That is, each layer must be piecewise affine \textit{as a function of its parameters} as well. This is easily verified for all of the layer types previously mentioned. For example, with the ReLU neuron we have \begin{equation} f(A,\boldsymbol{b})=\left[A\boldsymbol{x}+\boldsymbol{b}\right]_{+}\label{eq:ReLU} \end{equation} so for $\left(A\boldsymbol{x}+\boldsymbol{b}\right)_{k}\ge0$ we have that the $k^{th}$ component of $f$ is linear in $(A,\boldsymbol{b})$, while for $\left(A\boldsymbol{x}+\boldsymbol{b}\right)_{k}<0$ it is constant. To see this, we can re-arrange the elements of $A$ into a column vector $\boldsymbol{a}$, in row-major order, so that we have \begin{align} A\boldsymbol{x}+\boldsymbol{b} & =\begin{pmatrix}\boldsymbol{x^{T}} & \boldsymbol{0^{T}} & ... & ... & \boldsymbol{0^{T}} & \boldsymbol{1}^{T}\\ \boldsymbol{0^{T}} & \boldsymbol{x^{T}} & \boldsymbol{0}^{T} & ... & \boldsymbol{0}^{T} & \boldsymbol{1}^{T}\\ ... & ... & ... & ... & ... & ...\\ \boldsymbol{0}^{T} & ... & ... & \boldsymbol{0}^{T} & \boldsymbol{x}^{T} & \boldsymbol{1}^{T} \end{pmatrix}\begin{pmatrix}\boldsymbol{a}\\ \boldsymbol{b} \end{pmatrix}.\label{eq:ReLU_expanded} \end{align} In section \ref{sec:Neural-networks} we have said that a neural network, considered as a function of its input data, is convex and continuously differentiable on each piece. Now, a neural network need \textit{not}\textbf{ }be piecewise convex as a function of the entirety of its parameters\footnote{To see this, consider the following two-layer network: $h(x)=x$, $g_{2}(x)=ax$, and $g_{1}(x)=bx$. For $f=h\circ g_{2}\circ g_{1}$ we have $f(x)=abx$. Now fix the input as $x=1$. Considered as a function of its parameters, this is $f(a,b)=ab$, which is decidedly not convex.}. However, we can regain piecewise convexity by considering it only as a function of the parameters in a single layer, all others held constant. \begin{theorem}\label{theorem:neural_network_single_layer} A neural network $f$ is continuous piecewise convex and piecewise continuously differentiable as a function of the parameters in a single layer. \end{theorem} \begin{proof} For the time being, assume the input data consists of a single point $\boldsymbol{x}$. By definition $f$ is the composition of a convex objective $h$ and layers $g_{1},g_{2},...,g_{N}$, with $g_{1}$ a function of $\boldsymbol{x}$. Let $f_{m}(\boldsymbol{x})$ denote the network $f$ considered as a function of the parameters of layer $g_{m}$, all others held constant. Now, the layers $g_{m-1}\circ g_{m-2}\circ...\circ g_{1}$ are constant with respect to the parameters of $g_{m}$, so we can write $\boldsymbol{y}=g_{m-1}\circ g_{m-2}\circ...\circ g_{1}(\boldsymbol{x})$. Thus on each piece of $g_{m}$ we have \[ g_{m}=A\boldsymbol{y}+\boldsymbol{b}. \] By definition $g_{m}$ is a continuous piecewise affine function of its parameters. Since $\boldsymbol{y}$ is constant, we have that $\tilde{g}_{m}=g_{m}\circ g_{m-1}\circ...\circ g_{1}$ is a continuous piecewise affine function of the parameters of $g_{m}$. Now, by theorem \ref{theorem:composition_piecewise_affine} we have that $g=g_{N}\circ g_{N-1}\circ...\circ\tilde{g}_{m}$ is a continuous piecewise affine function of the parameters of $g_{m}$. Thus by theorem \ref{theorem:covex_piecewise}, $f_{m}$ is continuous piecewise convex. To establish piecewise continuous differentiability, recall that affine functions are continuously differentiable, as is $h$. Having established the theorem for the case of a single data point, consider the case where we have multiple data points, denoted $\{\boldsymbol{x}_{k}\}_{k=1}^{M}$. Now, by theorem \ref{theorem:mean_piecewise_convex} the arithmetic mean $(1/M)\sum_{k=1}^{M}f_{m}(\boldsymbol{x}_{k})$ is continuous piecewise convex. Furthermore, the arithmetic mean preserves piecewise continuous differentiability. Thus these results hold for the mean value of the network over the dataset. \end{proof} We conclude this section with a simple remark which will be useful in later sections. Let $f_{m}$ be a neural network, considered as a function of the parameters of the $m^{th}$ layer, and let $S$ be a piece of $f_{m}$. Then the optimization problem \begin{align} \mbox{minimize } & f_{m}(\boldsymbol{x})\nonumber \\ \mbox{subject to } & \boldsymbol{x}\in S\label{eq:convex_single_layer} \end{align} is convex. \section{Network parameters of multiple layers\label{sec:Network-parameters-of-multiple-layers}} In the previous section we analyzed the convexity properties of neural networks when optimizing the parameters of a single layer, all others held constant. Now we are ready to extend these results to the ultimate goal of simultaneously optimizing all network parameters. Although not convex, the problem has a special convex substructure that we can exploit in proving future results. We begin by defining this substructure for point sets and functions. \begin{definition} Let $S\subseteq\mathbb{R}^{n}$, let $I\subset\{1,2,...,n\}$, and let $\boldsymbol{x}\in S$. The set \[ S_{I}(\boldsymbol{x})=\{\boldsymbol{y}\in S:(\boldsymbol{y}_{k}=\boldsymbol{x}_{k})_{k\notin I}\} \] is the \textbf{cross-section} of $S$ intersecting $\boldsymbol{x}$ with respect to $I$. \end{definition} In other words, $S_{I}(\boldsymbol{x})$ is the subset of $S$ for which every point is equal to $\boldsymbol{x}$ in the components not indexed by $I$. Note that this differs from the typical definition, which is the intersection of a set with a hyperplane. For example, $\mathbb{R}_{\{1\}}^{3}(\boldsymbol{0})$ is the $x$-axis, whereas $\mathbb{R}_{\{1,2\}}^{3}(\boldsymbol{0})$ is the $xy$-plane. Note also that cross-sections are not unique, for example $\mathbb{R}_{\{1,2\}}^{3}(0,0,0)=\mathbb{R}_{\{1,2\}}^{3}(1,2,0)$. In this case the first two components of the cross section are irrelevant, but we will maintain them for notational convenience. We can now apply this concept to functions on $\mathbb{R}^{n}$. \begin{definition} Let $S\subseteq\mathbb{R}^{n}$, let $f:S\rightarrow\mathbb{R}$ and let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$. We say that $f$ is \textbf{multi-convex} with respect to $\mathcal{I}$ if $f$ is convex when restricted to the cross section $S_{I}(\boldsymbol{x})$, for all $\boldsymbol{x}\in S$ and $I\in\mathcal{I}$. \end{definition} This formalizes the notion of restricting a non-convex function to a variable subset on which it is convex, as in section \ref{sec:Network-parameters-of-single-layer} when a neural network was restricted to the parameters of a single layer. For example, let $f(x,y,z)=xy+z$, and let $I_{1}=\{1,3\}$, and $I_{2}=\{2,3\}$. Then $f_{1}(x,y_{0},z)$ is a convex function of $(x,z)$ with $y$ fixed at $y_{0}$. Similarly, $f_{2}(x_{0},y,z)$ is a convex function of $(y,z)$ with $x$ fixed at $x_{0}$. Thus $f$ is multi-convex with respect to $\mathcal{I}=\{I_{1},I_{2}\}$. To fully define a multi-convex optimization problem, we introduce a similar concept for point sets. \begin{definition} Let $S\subseteq\mathbb{R}^{n}$ and let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$. We say that $S$ is \textbf{multi-convex }with respect to $\mathcal{I}$ if the cross-section $S_{I}(\boldsymbol{x})$ is convex for all $\boldsymbol{x}\in S$ and $I\in\mathcal{I}$. \end{definition} This generalizes the notion of biconvexity found in the optimization literature \cite{Gorski:2007:BiConvex}. From here, we can extend definition \ref{def:piecewise_affine} to multi-convex functions. However, we will drop the topological restrictions on the pieces of our function, since multi-convex sets need not be connected. \begin{definition} Let $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ be a continuous function. We say that $f$ is \textbf{continuous piecewise multi-convex} if each there exists a collection of multi-convex functions $g_{1},g_{2},...,g_{N}$ and multi-convex sets $S_{1},S_{2},...,S_{N}$ covering $\mathbb{R}^{n}$ such that for each $k$ we have $f(\boldsymbol{x})=g_{k}(\boldsymbol{x})$ for all $\boldsymbol{x}\in S_{k}$. Next, let $h:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$. Then, $h$ is continuous piecewise multi-convex so long as each component is, as in definition \ref{def:piecewise_affine_onto_Rn}. \end{definition} From this definition, it is easily verified that a continuous piecewise multi-convex function $f:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$ admits a representation where all pieces are multi-convex, as in the proof of theorem \ref{theorem:convex_polytope}. Before we can extend the results of section \ref{sec:Network-parameters-of-single-layer} to multiple layers, we must add one final constraint on the definition of a neural network. That is, each of the layers must be continuous piecewise multi-convex, considered as functions of both the parameters \textit{and} the input. Again, this is easily verified for the all of the layer types previously mentioned. We have already shown they are piecewise convex on each cross-section, taking our index sets to separate the parameters from the input data. It only remains to show that the number of pieces is finite. The only layer which merits consideration is the ReLU, which we can see from equation \ref{eq:ReLU} consists of two pieces for each component: the ``dead'' or constant region, with $(A\boldsymbol{x})_{j}+b_{j}<0$, and its compliment. With $n$ components we have at most $2^{n}$ pieces, corresponding to binary assignments of ``dead'' or ``alive'' for each component. Having said that each layer is continuous piecewise multi-convex, we can extend these results to the whole network. \begin{theorem} Let $f$ be a neural network, and let $\mathcal{I}$ be a collection of index sets, one for the parameters of each layer of $f$. Then $f$ is continuous piecewise multi-convex with respect to $\mathcal{I}$. \end{theorem} We begin the proof with a lemma for more general multi-convex functions. \begin{lemma}\label{lemma:composition_multi_convex} Let $X\subseteq\mathbb{R}^{n}$, $Y\subseteq\mathbb{R}^{m}$, and let $g:X\rightarrow Z$ and $f:Z\times Y\rightarrow\mathbb{R}^{n}$ be continuous piecewise multi-convex, $g$ with respect to a collection of index sets $\mathcal{G}$, and $f$ with respect to $\mathcal{F}=\{I_{Z},I_{Y}\}$, where $I_{Z}$ indexes the variables in $Z$, and $I_{Y}$ the variables in $Y$. Then $h(\boldsymbol{x},\boldsymbol{y})=f(g(\boldsymbol{x}),\boldsymbol{y})$ is continuous piecewise multi-convex with respect to $\mathcal{H}=\mathcal{G}\cup\{I_{Y}\}$. \end{lemma} \begin{proof} Let $G$ be a piece of $g$, let $F$ be a piece of $f$ and let $H=\{(\boldsymbol{x},\boldsymbol{y}):\boldsymbol{x}\in G,\,(g(\boldsymbol{x}),\boldsymbol{y})\in F\}$, with $F$ chosen so that $H\ne\emptyset$. Clearly $h$ is multi-convex on $H$ with respect to $\mathcal{H}$. It remains to show that $H$ is a multi-convex set. Now, let $(\boldsymbol{x},\boldsymbol{y})\in H$ and we shall show that the cross-sections are convex. First, for any $I_{X}\in\mathcal{G}$ we have $H_{I_{X}}(\boldsymbol{x},\boldsymbol{y})=G_{I_{X}}(\boldsymbol{x})\times\{\boldsymbol{y}\}$. Similarly, we have $H_{I_{Y}}(\boldsymbol{x},\boldsymbol{y})=\{\boldsymbol{x}\}\times\{\boldsymbol{y}:(\boldsymbol{z},\boldsymbol{y})\in F_{I_{Y}}(g(\boldsymbol{x}),\boldsymbol{y})\}$. These sets are convex, as they are the Cartesian products of convex sets \cite{Rockafellar:1970:ConvexAnalysis}. Finally, as in the proof of theorem \ref{theorem:composition_piecewise_affine}, we can cover $X\times Y$ with the finite collection of all such pieces $H$, taken over all $G$ and $F$. \end{proof} Our next lemma extends theorem \ref{theorem:mean_piecewise_convex} to multi-convex functions. \begin{lemma}\label{lemma:sum_piecewise_multi_convex} Let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$, and let $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ and $g:\mathbb{R}^{n}\rightarrow\mathbb{R}$ be continuous piecewise multi-convex with respect to $\mathcal{I}$. Then so is $f+g$. \end{lemma} \begin{proof} Let $F$ be a piece of $f$ and $G$ be a piece of $g$ with $\boldsymbol{x}\in F\cap G$. Then for all $I\in\mathcal{I}$, $\left(F\cap G\right)_{I}(\boldsymbol{x})=F_{I}(\boldsymbol{x})\cap G_{I}(\boldsymbol{x})$, a convex set on which $f+g$ is convex. Thus $f+g$ is continuous piecewise multi-convex, where the pieces of $f+g$ are the intersections of pieces of $f$ and $g$. \end{proof} We can now prove the theorem. \begin{proof} For the moment, assume we have only a single data point. Now, let $g_{1}$ and $g_{2}$ denote layers of $f$, with parameters $\boldsymbol{\theta}_{1}\in\mathbb{R}^{m},\,\boldsymbol{\theta}_{2}\in\mathbb{R}^{n}$. Since $g_{1}$ and $g_{2}$ are continuous piecewise multi-convex functions of their parameters and input, we can write the two-layer sub-network as $h=f(g_{1}(\boldsymbol{\theta}_{1}),\boldsymbol{\theta}_{2})$. By repeatedly applying lemma \ref{lemma:composition_multi_convex}, the whole network is multi-convex on a finite number of sets covering the input and parameter space. Now we extend the theorem to the whole dataset, where each data point defines a continuous piecewise multi-convex function $f_{k}$. By lemma \ref{lemma:sum_piecewise_multi_convex}, the arithmetic mean $(1/N)\sum_{k=1}^{N}f_{k}$ is continuous piecewise multi-convex. \end{proof} In the coming sections, we shall see that multi-convexity allows us to give certain guarantees about the convergence of various optimization algorithms. But first, we shall prove some basic results independent of the optimization procedure. These results were summarized by Gorksi et al\@.~for the case of biconvex differentiable functions \cite{Gorski:2007:BiConvex}. Here we extend them to piecewise functions and arbitrary index sets. First we define a special type of minimum relevant for multi-convex functions. \begin{definition} Let $f:S\rightarrow\mathbb{R}$ and let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$. We say that $\boldsymbol{x}_{0}$ is a \textbf{partial minimum} of $f$ with respect to $\mathcal{I}$ if $f(\boldsymbol{x}_{0})\le f(\boldsymbol{x})$ for all $\boldsymbol{x}\in\cup_{I\in\mathcal{I}}S_{I}(\boldsymbol{x}_{0})$. \end{definition} In other words, $\boldsymbol{x}_{0}$ is a partial minimum of $f$ with respect to $\mathcal{I}$ if it minimizes $f$ on every cross-section of $S$ intersecting $\boldsymbol{x}_{0}$, as shown in figure \ref{fig:cross_section}. By convexity, these points are intimately related to the stationary points of $f$. \begin{figure} \begin{centering} \includegraphics[scale=0.7]{cross_section} \par\end{centering} \caption{Cross-sections of a biconvex set.} \label{fig:cross_section} \end{figure} \begin{theorem}\label{theorem:multi_convex_partial_minimum} Let $\mathcal{I}=\{I_{1},I_{2},...,I_{m}\}$ be a collection of sets covering $\{1,2,...,n\}$, let $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ be continuous piecewise multi-convex with respect to $\mathcal{I}$, and let $\nabla f(\boldsymbol{x}_{0})=\boldsymbol{0}$. Then $\boldsymbol{x}_{0}$ is a partial minimum of $f$ on every piece containing $\boldsymbol{x}_{0}$. \end{theorem} \begin{proof} Let $S$ be a piece of $f$ containing \textbf{$\boldsymbol{x}_{0}$}, let $I\in\mathcal{I}$ , and let $S_{I}(\boldsymbol{x}_{0})$ denote the relevant cross-section of $S$. We know $f$ is convex on $S_{I}(\boldsymbol{x}_{0})$, and since $\nabla f(\boldsymbol{x}_{0})=\boldsymbol{0}$, we have that $\boldsymbol{x}_{0}$ minimizes $f$ on this convex set. Since this holds for all $I\in\mathcal{I}$, $\boldsymbol{x}_{0}$ is a partial minimum of $f$ on $S$. \end{proof} It is clear that multi-convexity provides a wealth of results concerning partial minima, while piecewise multi-convexity restricts those results to a subset of the domain. Less obvious is that partial minima of smooth multi-convex functions need not be local minima. An example was pointed out by a reviewer of this work, that the biconvex function $f(x,y)=xy$ has a partial minimum at the origin which is not a local minimum. However, the converse is easily verified, even in the absence of differentiability. \begin{theorem} Let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$, let $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ be continuous piecewise multi-convex with respect to $\mathcal{I}$, and let $\boldsymbol{x}_{0}$ be a local minimum on some piece $S$ of $f$. Then $\boldsymbol{x}_{0}$ is a partial minimum on $S$. \end{theorem} \begin{proof} The proof is essentially the same as that of theorem \ref{theorem:multi_convex_partial_minimum}. \end{proof} We have seen that for multi-convex functions there is a close relationship between stationary points, local minima and partial minima. For these functions, infinitesimal results concerning derivatives and local minima can be extended to larger sets. However, we make no guarantees about global minima. The good news is that, unlike global minima, we shall see that we can easily solve for partial minima. \section{Gradient descent\label{sec:Optimization-and-convergence}} In the realm of non-convex optimization, also called global optimization, methods can be divided into two groups: those which can certifiably find a global minimum, and those which cannot. In the former group we sacrifice speed, in the latter correctness. This work focuses on algorithms of the latter kind, called local or sub-optimal methods, as only this type is used in practice for deep neural networks. In particular, the most common methods are variants of gradient descent, where the gradient of the network with respect its parameters is computed by a procedure called backpropagation. Since its explanation is often obscured by jargon, we shall provide a simple summary here. Backpropagation is nothing but the chain rule applied to the layers of a network. Splitting the network into two functions $f=u\circ v$, where $u:\mathbb{R}^{n}\rightarrow\mathbb{R}$, and $v:\mathbb{R}^{m}\rightarrow\mathbb{R}^{n}$, we have \[ \nabla f=\nabla u\mathcal{D}v \] where $\mathcal{D}$ denotes the Jacobian operator. Note that here the parameters of $u$ are considered fixed, whereas the parameters of $v$ are variable and the input data is fixed. Thus $\nabla f$ is the gradient of $f$ with respect to the parameters of $v$, if it exists. The special observation is that we can proceed from the top layer of the neural network $g_{N}$ to the bottom $g_{1}$, with $u=g_{N}\circ g_{N-1}\circ...\circ g_{m+1}$, and $v=g_{m}$, each time computing the gradient of $f$ with respect to the parameters of $g_{m}$. In this way, we need only store the vector $\nabla u$ and the matrix $\mathcal{D}v$ can be forgotten at each step. This is known as the ``backward pass,'' which allows for efficient computation of the gradient of a neural network with respect to its parameters. A similar algorithm computes the value of $g_{m-1}\circ g_{m-2}\circ...\circ g_{1}$ as a function of the input data, which is often needed to evaluate $\mathcal{D}v$. First we compute and store $g_{1}$ as a function of the input data, then $g_{2}\circ g_{1}$, and so on until we have $f$. This is known as the ``forward pass.'' After one forward and one backward pass, we have computed $\nabla f$ with respect to all the network parameters. Having computed $\nabla f$, we can update the parameters by gradient descent, defined as follows. \begin{definition} Let $S\subset\mathbb{R}^{n}$, and $f:S\rightarrow\mathbb{R}$ be partial differentiable, with $\boldsymbol{x}_{0}\in S$. Then \textbf{gradient descent} on $f$ is the sequence $\{\boldsymbol{x}_{k}\}_{k=0}^{\infty}$ defined by \[ \boldsymbol{x}_{k+1}=\boldsymbol{x}_{k}-\alpha_{k}\nabla f(\boldsymbol{x}_{k}) \] where $\alpha_{k}>0$ is called the \textbf{step size} or ``learning rate.'' In this work we shall make the additional assumption that $\sum_{k=0}^{\infty}a_{k}=\infty$. \end{definition} Variants of this basic procedure are preferred in practice because their computational cost scales well with the number of network parameters. There are many different ways to choose the step size, but our assumption that $\sum_{k=0}^{\infty}a_{k}=\infty$ covers what is usually done with deep neural networks. Note that we have not defined what happens if $\boldsymbol{x}_{k}\notin S$. Since we are ultimately interested in neural networks on $\mathbb{R}^{n}$, we can ignore this case and say that the sequence diverges. Gradient descent is not guaranteed to converge to a global minimum for all differentiable functions. However, it is natural to ask to which points it can converge. This brings us to a basic but important result. \begin{theorem}\label{theorem:gradient_descent} Let $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$, and let $\{\boldsymbol{x}_{k}\}_{k=0}^{\infty}$ result from gradient descent on $f$ with $\lim_{k\rightarrow\infty}\boldsymbol{x}_{k}=\boldsymbol{x}^{*}$, and $f$ continuously differentiable at $\boldsymbol{x}^{*}$. Then $\nabla f(\boldsymbol{x}^{*})=\boldsymbol{0}$. \end{theorem} \begin{proof} First, we have \[ \boldsymbol{x}^{*}=\boldsymbol{x}_{0}-\sum_{k=0}^{\infty}\alpha_{k}\nabla f(\boldsymbol{x}_{k}). \] Assume for the sake of contradiction that for the $j^{th}$ partial derivative we have $|\partial f(\boldsymbol{x}^{*})/\partial(\boldsymbol{x})_{j}|>0$. Now, pick some $\varepsilon$ such that $0<\varepsilon<|\partial f(\boldsymbol{x}^{*})/\partial(\boldsymbol{x})_{j}|$, and by continuous differentiability, there is some $\delta>0$ such that for all $\boldsymbol{x}$, $\|\boldsymbol{x}^{*}-\boldsymbol{x}\|_{2}<\delta$ implies $\|\nabla f(\boldsymbol{x}^{*})-\nabla f(\boldsymbol{x})\|_{2}<\varepsilon$. Now, there must be some $K$ such that for all $k\ge K$ we have $\|\boldsymbol{x}^{*}-\boldsymbol{x}_{k}\|_{2}<\delta$, so that $\partial f(\boldsymbol{x}_{k})/\partial(\boldsymbol{x})_{j}$ does not change sign. Then we can write \begin{align*} \left|\sum_{k=K}^{\infty}\alpha_{k}\frac{\partial f(\boldsymbol{x}_{k})}{\partial\left(\boldsymbol{x}\right)_{j}}\right| & =\sum_{k=K}^{\infty}\alpha_{k}\left|\frac{\partial f(\boldsymbol{x}_{k})}{\partial\left(\boldsymbol{x}\right)_{j}}\right|\\ & \ge\sum_{k=K}^{\infty}\alpha_{k}\left(\left|\frac{\partial f(\boldsymbol{x}^{*})}{\partial\left(\boldsymbol{x}\right)_{j}}\right|-\varepsilon\right)\\ & =\infty. \end{align*} But this contradicts the fact that $\boldsymbol{x}_{k}$ converges. Thus $\nabla f(\boldsymbol{x}^{*})=\boldsymbol{0}$. \end{proof} In the convex optimization literature, this simple result is sometimes stated in connection with Zangwill's much more general convergence theorem \cite{Zangwill:1969:Optimization,Iusem:2003:subgradientConvergence}. Note, however, that unlike Zangwill we state necessary, rather than sufficient conditions for convergence. While many similar results are known, it is difficult to strictly weaken the conditions of theorem \ref{theorem:gradient_descent}. For example, if we relax the condition that $\alpha_{k}$ is not summable, and take $f(x)=x$, then $x_{k}$ will always converge to a non-stationary point. Similarly, if we relax the constraint that $f$ is continuously differentiable, taking $f(x)=|x|$ and $a_{k}$ decreasing monotonically to zero, we will always converge to the origin, which is not differentiable. Furthermore, if we have $f(x)=|x|$ with $\alpha_{k}$ constant, then $x_{k}$ will not converge for almost all $x_{0}$. It is possible to prove much stronger necessary and sufficient conditions for gradient descent, but these results require additional assumptions about the step size policy as well as the function to be minimized, and possibly even the initialization $\boldsymbol{x}_{0}$ \cite{Nesterov:2004:ConvexBook}. It is worth discussing $f(x)=|x|$ in greater detail, since this is a piecewise affine function and thus of interest in our investigation of neural networks. While we have said its only convergence point is not differentiable, it remains subdifferentiable, and convergence results are known for subgradient descent \cite{Iusem:2003:subgradientConvergence}. In this work we shall not make use of subgradients, instead considering descent on a piecewise continuously differentiable function, where the pieces are $x\le0$ and $x\ge0$. Although theorem \ref{theorem:gradient_descent} does not apply to this function, the relevant results hold anyways. That is, $x=0$ is minimal on some piece of $f$, a result which extends to any continuous piecewise convex function, as any saddle point is guaranteed to minimize some piece. Here we should note one way in which this analysis fails in practice. So far we have assumed the gradient $\nabla f$ is precisely known. In practice, it is often prohibitively expensive to compute the average gradient over large datasets. Instead we take random subsamples, in a procedure known as \textit{stochastic }gradient descent. We will not analyze its properties here, as current results on the topic impose additional restrictions on the objective function and step size, or require different definitions of convergence \cite{Bertsekas:2010:IncrementalGradient,Bach:2011:sgd,Ge:2015:sgdSaddle}. Restricting ourselves to the true gradient $\nabla f$ allows us to provide simple proofs applying to an extensive class of neural networks. We are now ready to generalize these results to neural networks. There is a slight ambiguity in that the boundary points between pieces need not be differentiable, nor even sub-differentiable. Since we are interested only in necessary conditions, we will say that gradient descent diverges when $\nabla f(\boldsymbol{x}_{k})$ does not exist. However, our next theorem can at least handle non-differentiable limit points. \begin{theorem}\label{theorem:multi_convex_gradient_descent} Let $\mathcal{I}=\{I_{1},I_{2},...,I_{m}\}$ be a collection of sets covering $\{1,2,...,n\}$, let $f:\mathbb{R}^{n}\rightarrow\mathbb{R}$ be continuous piecewise multi-convex with respect to $\mathcal{I}$, and piecewise continuously differentiable. Then, let $\{\boldsymbol{x}_{k}\}_{k=0}^{\infty}$ result from gradient descent on $f$ , with $\lim_{k\rightarrow\infty}\boldsymbol{x}_{k}=\boldsymbol{x}^{*}$, such that either \begin{enumerate} \item $f$ is continuously differentiable at $\boldsymbol{x}^{*}$, or \item there is some piece $S$ of $f$ and some $K>0$ such that $\boldsymbol{x}_{k}\in\interior S$ for all $k\ge K$. \end{enumerate} Then $\boldsymbol{x}^{*}$ is a partial minimum of $f$ on every piece containing $\boldsymbol{x}^{*}$. \end{theorem} \begin{proof} If the first condition holds, the result follows directly from theorems \ref{theorem:gradient_descent} and \ref{theorem:multi_convex_partial_minimum}. If the second condition holds, then $\{\boldsymbol{x}_{k}\}_{k=K}^{\infty}$ is a convergent gradient descent sequence on $g$, the active function of $f$ on $S$. Since $g$ is continuously differentiable on $\mathbb{R}^{n}$, the first condition holds for $g$. Since $f|_{S}=g|_{S}$, $\boldsymbol{x}^{*}$ is a partial minimum of $f|_{S}$ as well. \end{proof} The first condition of theorem \ref{theorem:multi_convex_gradient_descent} holds for every point in the interior of a piece, and some boundary points. The second condition extends these results to non-differentiable boundary points so long as gradient descent is eventually confined to a single piece of the function\@. For example, consider the continuous piecewise convex function $f(x)=\min(x,x^{4})$ as shown in figure \ref{fig:convergence}. When we converge to $x=0$ from the piece $[0,1]$, it is as if we were converging on the smooth function $g(x)=x^{4}$. This example also illustrates an important caveat regarding boundary points: although $x=0$ is an extremum of $f$ on $[0,1]$, it is not an extremum on $\mathbb{R}$. \begin{figure} \begin{centering} \includegraphics[scale=0.4]{convergence} \par\end{centering} \centering{}\caption{Example of a piecewise convex function. The point $x=0$ minimizes the function on the piece $[0,1]$.} \label{fig:convergence} \end{figure} \section{Iterated convex optimization} Although the previous section contained some powerful results, theorem \ref{theorem:multi_convex_gradient_descent} suffers from two main weaknesses, that it is a necessary condition and that it requires extra care at non-differentiable points. It is difficult to overcome these limitations with gradient descent. Instead, we shall define a different optimization technique, from which necessary and sufficient convergence results follow, regardless of differentiability. Iterated convex optimization splits a non-convex optimization problem into a number of convex sub-problems, solving the sub-problems in each iteration. For a neural network, we have shown that the problem of optimizing the parameters of a single layer, all others held constant, is piecewise convex. Thus, restricting ourselves to a given piece yields a convex optimization problem. In this section, we show that these convex sub-problems can be solved repeatedly, converging to a piecewise partial optimum. \begin{definition} Let $\mathcal{I}=\{I_{1},I_{2},...,I_{m}\}$ be a collection of sets covering $\{1,2,...,n\}$, and let $S\subseteq\mathbb{R}^{n}$ and $f:S\rightarrow\mathbb{R}$ be multi-convex with respect to $\mathcal{I}$. Then \textbf{iterated convex optimization }is any sequence where $\boldsymbol{x}_{k}$ is a solution to the optimization problem \begin{align} \mbox{minimize } & f(\boldsymbol{y})\label{eq:iterated_convex}\\ \mbox{subject to } & \boldsymbol{y}\in\cup_{I\in\mathcal{I}}S_{I}(\boldsymbol{x}_{k-1})\nonumber \end{align} with \textbf{$\boldsymbol{x}_{0}\in S$.} \end{definition} We call this iterated convex optimization because problem \ref{eq:iterated_convex} can be divided into convex sub-problems \begin{align} \mbox{minimize } & f(\boldsymbol{y})\label{eq:convex_subproblem}\\ \mbox{subject to } & \boldsymbol{y}\in S_{I}(\boldsymbol{x}_{k-1}).\nonumber \end{align} for each $I\in\mathcal{I}$. In this work, we assume the convex sub-problems are solvable, without delving into specific solution techniques. Methods for alternating between solvable sub-problems have been studied by many authors, for many different types of sub-problems \cite{Wendell:1976:Bilinear}. In the context of machine learning, the same results have been developed for the special case of linear autoencoders \cite{Baldi:2012:ComplexValuedAutoencoders}. Still, extra care must be taken in extending these results to arbitrary index sets. The key is that $\boldsymbol{x}_{k}$ is not updated until all sub-problems have been solved, so that each iteration consists of solving $m$ convex sub-problems. This is equivalent to the usual alternating convex optimization for biconvex functions, where $\mathcal{I}$ consists of two sets, but not for general multi-convex functions. Some basic convergence results follow immediately from the solvability of problem \ref{eq:iterated_convex}. First, note that $\boldsymbol{x}_{k-1}$ is a feasible point, so we have $f(\boldsymbol{x}_{k})\le f(\boldsymbol{x}_{k-1})$. This implies that $\lim_{k\rightarrow\infty}f(\boldsymbol{x}_{k})$ exists, so long as $f$ is bounded below. However, this does not imply the existence of $\lim_{k\rightarrow\infty}\boldsymbol{x}_{k}$. See Gorski et al.~for an example of a biconvex function on which $\boldsymbol{x}_{k}$ diverges \cite{Gorski:2007:BiConvex}. To prove stronger convergence results, we introduce regularization to the objective. \begin{theorem}\label{theorem:regularization} Let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$, and let $S\subseteq\mathbb{R}^{n}$ and $f:S\rightarrow\mathbb{R}$ be multi-convex with respect to $\mathcal{I}$. Next, let $\inf f>-\infty$, and let $g(\boldsymbol{x})=f(\boldsymbol{x})+\lambda\|\boldsymbol{x}\|$, where $\lambda>0$ and $\|\boldsymbol{x}\|$ is a convex norm. Finally, let $\{\boldsymbol{x}_{k}\}_{k=0}^{\infty}$ result from iterated convex optimization of $g$. Then $\boldsymbol{x}_{k}$ has at least one convergent subsequence, in the topology induced by the metric $d(\boldsymbol{x},\boldsymbol{y})=\|\boldsymbol{x}-\boldsymbol{y}\|$. \end{theorem} \begin{proof} From lemma \ref{lemma:sum_piecewise_convex}, $g$ is multi-convex, so we are allowed iterated convex optimization. Now, if $\inf f+\lambda\|\boldsymbol{x}\|>g(\boldsymbol{x}_{0})$ we have that $g(\boldsymbol{x})>g(\boldsymbol{x}_{0})$. Thus $g(\boldsymbol{x})>g(\boldsymbol{x}_{0})$ whenever $\|\boldsymbol{x}\|>\left(g(\boldsymbol{x}_{0})-\inf f\right)/\lambda$. Since $g(\boldsymbol{x}_{k})$ is a non-increasing sequence, we have that $\|\boldsymbol{x}_{k}\|\le\left(g(\boldsymbol{x}_{0})-\inf f\right)/\lambda$. Equivalently, $\boldsymbol{x}_{k}$ lies in the set $A=\{\boldsymbol{x}:\|\boldsymbol{x}\|\le\left(g(\boldsymbol{x}_{0})-\inf f\right)/\lambda\}$. Since $\|\boldsymbol{x}\|$ is continuous, $A$ is closed and bounded, and thus it is compact. Then, by the Bolzano-Weierstrauss theorem, $\boldsymbol{x}_{k}$ has at least one convergent subsequence \cite{Johnsonbaugh:1970:RealAnalysis}. \end{proof} In theorem \ref{theorem:regularization}, the function $g$ is called the \textbf{regularized} version of $f$. In practice, regularization often makes a non-convex optimization problem easier to solve, and can reduce over-fitting. The theorem shows that iterated convex optimization on a regularized function always has at least one convergent subsequence. Next, we shall establish some rather strong properties of the limits of these subsequences. \begin{theorem}\label{theorem:sequential_convex_optimization_convergence} Let $\mathcal{I}$ be a collection of sets covering $\{1,2,...,n\}$, and let $S\subseteq\mathbb{R}^{n}$ and $f:S\rightarrow\mathbb{R}$ be multi-convex with respect to $\mathcal{I}$. Next, let $\{\boldsymbol{x}_{k}\}_{k=0}^{\infty}$ result from iterated convex optimization of $f$. Then the limit of every convergent subsequence is a partial minimum on $\interior S$ with respect to $\mathcal{I}$, in the topology induced by the metric $d(\boldsymbol{x},\boldsymbol{y})=\|\boldsymbol{x}-\boldsymbol{y}\|$ for some norm $\|\boldsymbol{x}\|$. Furthermore, if $\{\boldsymbol{x}_{m_{k}}\}_{k=1}^{\infty}$ and $\{\boldsymbol{x}_{n_{k}}\}_{k=1}^{\infty}$ are convergent subsequences, then $\lim_{k\rightarrow\infty}f(\boldsymbol{x}_{m_{k}})=\lim_{k\rightarrow\infty}f(\boldsymbol{x}_{n_{k}})$. \end{theorem} \begin{proof} Let $\boldsymbol{x}_{n_{k}}$ denote a subsequence of $\boldsymbol{x}_{k}$ with $\boldsymbol{x}^{*}=\lim_{n\rightarrow\infty}\boldsymbol{x}_{n_{k}}$. Now, assume for the sake of contradiction that $\boldsymbol{x}^{*}$ is not a partial minimum on $\mbox{int}S$ with respect to $\mathcal{I}$. Then there is some $I\in\mathcal{I}$ and some $\boldsymbol{x}^{\prime}\in S_{I}(\boldsymbol{x}^{*})$ with $\boldsymbol{x}^{\prime}\in\interior S$ such that $f(\boldsymbol{x}^{\prime})<f(\boldsymbol{x}^{*})$. Now, $f$ is continuous at $\boldsymbol{x}^{\prime}$, so there must be some $\delta>0$ such that for all $\boldsymbol{x}\in S$, $\|\boldsymbol{x}-\boldsymbol{x}^{\prime}\|<\delta$ implies $|f(\boldsymbol{x})-f(\boldsymbol{x}^{\prime})|<f(\boldsymbol{x}^{*})-f(\boldsymbol{x}^{\prime})$. Furthermore, since $\boldsymbol{x}^{\prime}$ is an interior point, there must be some open ball $B\subset S$ of radius $r$ centered at $\boldsymbol{x}^{\prime}$, as shown in figure \ref{fig:proof_iterated_convex_convergence}. Now, there must be some $K$ such that $\|\boldsymbol{x}_{n_{K}}-\boldsymbol{x}^{*}\|<\min(\delta,r)$. Then, let $\tilde{\boldsymbol{x}}=\boldsymbol{x}_{n_{K}}+\boldsymbol{x}^{\prime}-\boldsymbol{x}^{*}$, and since $\|\tilde{\boldsymbol{x}}-\boldsymbol{x}^{\prime}\|<r$, we know that $\tilde{\boldsymbol{x}}\in B$, and thus $\tilde{\boldsymbol{x}}\in S_{I}(\boldsymbol{x}_{n_{K}})$. Finally, $\|\tilde{\boldsymbol{x}}-\boldsymbol{x}^{\prime}\|<\delta$, so we have $f(\tilde{\boldsymbol{x}})<f(\boldsymbol{x}^{*})\le f(\boldsymbol{x}_{n_{K}+1})$, which contradicts the fact that $\boldsymbol{x}_{n_{K}+1}$ minimizes $g$ over a set containing $\tilde{\boldsymbol{x}}$. Thus $\boldsymbol{x}^{*}$ is a partial minimum on $\interior S$ with respect to $\mathcal{I}$. Finally, let $\{\boldsymbol{x}_{m_{k}}\}_{k=1}^{\infty}$ and $\{\boldsymbol{x}_{n_{k}}\}_{k=1}^{\infty}$ be two convergent subsequences of $\boldsymbol{x}_{k}$, with $\lim_{k\rightarrow\infty}\{\boldsymbol{x}_{m_{k}}\}=\boldsymbol{x}_{m}^{*}$ and $\lim_{k\rightarrow\infty}\{\boldsymbol{x}_{n_{k}}\}=\boldsymbol{x}_{n}^{*}$, and assume for the sake of contradiction that $f(\boldsymbol{x}_{m}^{*})>f(\boldsymbol{x}_{n}^{*})$. Then by continuity, there is some $K$ such that $f(\boldsymbol{x}_{n_{K}})<f(\boldsymbol{x}_{m}^{*})$. But this contradicts the fact that $f(\boldsymbol{x}_{k})$ is non-increasing. Thus $f(\boldsymbol{x}_{m}^{*})=f(\boldsymbol{x}_{n}^{*})$. \end{proof} \begin{figure} \begin{centering} \includegraphics[scale=0.5]{biconvex} \par\end{centering} \caption{Illustration of the proof of theorem \ref{theorem:sequential_convex_optimization_convergence}. Note the cross-sections of the biconvex set $S$.} \label{fig:proof_iterated_convex_convergence} \end{figure} The previous theorem is an extension of results reviewed in Gorski et al.~to arbitrary index sets \cite{Gorski:2007:BiConvex}. While Gorski et al.~explicitly constrain the domain to a compact biconvex set, we show that regularization guarantees $\boldsymbol{x}_{k}$ cannot escape a certain compact set, establishing the necessary condition for convergence. Furthermore, our results hold for general multi-convex sets, while the earlier result is restricted to Cartesian products of compact sets. These results for iterated convex optimization are considerably stronger than what we have shown for gradient descent. While any bounded sequence in $\mathbb{R}^{n}$ has a convergent subsequence, and we can guarantee boundedness for some variants of gradient descent, we cannot normally say much about the limits of subsequences. For iterated convex optimization, we have shown that the limit of any subsequence is a partial minimum, and all limits of subsequences are equal in objective value. For all practical purposes, this is just as good as saying that the original sequence converges to partial minimum. \section{Global optimization\label{sec:Local-minima}} Although we have provided necessary and sufficient conditions for convergence of various optimization algorithms on neural networks, the points of convergence need only minimize cross-sections of pieces of the domain. Of course we would prefer results relating the points of convergence to global minima of the training objective. In this section we illustrate the difficulty of establishing such results, even for the simplest of neural networks. In recent years much work has been devoted to providing theoretical explanations for the empirical success of deep neural networks, a full accounting of which is beyond the scope of this article. In order to simplify the problem, many authors have studied \textit{linear} neural networks, in which the layers have the form $g(\boldsymbol{x})=A\boldsymbol{x}$, where $A$ is the parameter matrix. With multiple layers this is clearly a linear function of the output, but not of the parameters. As a special case of piecewise affine functions, our previous results suffice to show that these networks are multi-convex as functions of their parameters. This was proven for the special case of linear autoencoders by Baldi and Lu \cite{Baldi:2012:ComplexValuedAutoencoders}. Many authors have claimed that linear neural networks contain no ``bad'' local minima, i.e.~every local minimum is a global minimum \cite{Kawaguchi:2016:WithoutPoorLocalMinima,Soudry:2016:NoBadLocalMinima}. This is especially evident in the study of linear autoencoders, which were shown to admit many points of inflection, but only a single strict minimum \cite{Baldi:2012:ComplexValuedAutoencoders}. While powerful, this claim does not apply to the networks seen in practice. To see this, consider the dataset $D=\{(0,1/2),(-1,\alpha),(1,2\alpha)\}$ consisting of three $(x,y)$ pairs, parameterized by $\alpha>1$. Note that the dataset has zero mean and unit variance in the $x$ variable, which is common practice in machine learning. However, we do not take zero mean in the $y$ variable, as the model we shall adopt is non-negative. Next, consider the simple neural network \begin{align} f(a,b) & =\sum_{(x,y)\in D}\left(y-\left[ax+b\right]_{+}\right)^{2}\label{eq:single_layer_objective}\\ & =\left(\frac{1}{2}-\left[b\right]_{+}\right)^{2}+\left(\alpha-\left[b-a\right]_{+}\right)^{2}+\left(2\alpha-\left[b+a\right]_{+}\right)^{2}.\nonumber \end{align} This is the squared error of a single ReLU neuron, parameterized by $(a,b)\in\mathbb{R}^{2}$. We have chosen this simplest of all networks because we can solve for the local minima in closed form, and show they are indeed very bad. First, note that $f$ is a continuous piecewise convex function of six pieces, realized by dividing the plane along the line $ax+b=0$ for each $x\in D$, as shown in figure \ref{fig:bad_local_minimum}. Now, for all but one of the pieces, the ReLU is ``dead'' for at least one of the three data points, i.e.~$ax+b<0$. On these pieces, at least one of the three terms of equation \ref{eq:single_layer_objective} is constant. The remaining terms are minimized when $y=ax+b$, represented by the three dashed lines in figure \ref{fig:bad_local_minimum}. There are exactly three points where two of these lines intersect, and we can easily show that two of them are strict local minima. Specifically, the point $(a_{1},b_{1})=(1/2-\alpha,1/2)$ minimizes the first two terms of equation \ref{eq:single_layer_objective}, while $(a_{2},b_{2})=(2\alpha-1/2,1/2)$ minimizes the first and last term. In each case, the remaining term is constant over the piece containing the point of intersection. Thus these points are strict global minima on their respective pieces, and strict local minima on $\mathbb{R}^{2}$. Furthermore, we can compute $f(a_{1},b_{1})=4\alpha^{2}$ and $f(a_{2},b_{2})=\alpha^{2}$. This gives \begin{align*} \lim_{\alpha\rightarrow\infty}a_{1} & =-\infty,\\ \lim_{\alpha\rightarrow\infty}a_{2} & =+\infty, \end{align*} and \[ \lim_{\alpha\rightarrow\infty}\left(f(a_{1},b_{1})-f(a_{2},b_{2})\right)=\infty. \] Now, it might be objected that we are not permitted to take $\alpha\rightarrow\infty$ if we require that the $y$ variable has unit variance. However, these same limits can be achieved with variance tending to unity by adding $\left\lfloor \alpha\right\rfloor $ instances of the point $(1,2\alpha)$ to our dataset. Thus even under fairly stringent requirements we can construct a dataset yielding arbitrarily bad local minima, both in the parameter space and the objective value. This provides some weak justification for the empirical observation that success in deep learning depends greatly on the data at hand. We have shown that the results concerning local minima in linear networks do not extend to the nonlinear case. Ultimately this should not be a surprise, as with linear networks the problem can be relaxed to linear regression on a convex objective. That is, the composition of all linear layers $g(\boldsymbol{x})=A_{1}A_{2}...A_{n}\boldsymbol{x}$ is equivalent to the function $f(\boldsymbol{x})=A\boldsymbol{x}$ for some matrix $A$, and under our previous assumptions the problem of finding the optimal $A$ is convex. Furthermore, it is easily shown that the number of parameters in the relaxed problem is polynomial in the number of original parameters. Since the relaxed problem fits the data at least as well as the original, it is not surprising that the original problem is computationally tractable. \begin{figure} \begin{centering} \includegraphics[scale=0.5]{single_neuron} \par\end{centering} \caption{Parameter space of the neural network from equation \ref{eq:single_layer_objective}, with pieces divided by the bold black lines. The points $(a_{1},b_{1})$ and $(a_{2},b_{2})$ are local minima, which can be made arbitrarily far apart by varying the dataset.} \label{fig:bad_local_minimum} \end{figure} This simple example was merely meant to illustrate the difficulty of establishing results for \textit{every} local minimum of \textit{every} neural network. Since training a certain kind of network is known to be NP-Complete, it is difficult to give any guarantees about worst-case global behavior \cite{Blum:1992:OCT:148433.148441}. We have made no claims, however, about probabilistic behavior on the average practical dataset, nor have we ruled out the effects of more specialized networks, such as very deep ones. \section{Conclusion} We showed that a common class of neural networks is piecewise convex in each layer, with all other parameters fixed. We extended this to a theory of a piecewise multi-convex functions, showing that the training objective function can be represented by a finite number of multi-convex functions, each active on a multi-convex parameter set. From here we derived various results concerning the extrema and stationary points of piecewise multi-convex functions. We established convergence conditions for both gradient descent and iterated convex optimization on this class of functions, showing they converge to piecewise partial minima. Similar results are likely to hold for a variety of other optimization algorithms, especially those guaranteed to converge at stationary points or local minima. We have witnessed the utility of multi-convexity in proving convergence results for various optimization algorithms. However, this property may be of practical use as well. Better understanding of the training objective could lead to the development of faster or more reliable optimization methods, heuristic or otherwise. These results may provide some insight into the practical success of sub-optimal algorithms on neural networks. However, we have also seen that local optimality results do not extend to global optimality as they do for linear autoencoders. Clearly there is much left to discover about how, or even if we can optimize deep, nonlinear neural networks. \section*{Acknowledgments} The author would like to thank Mihir Mongia for his helpful comments in preparing this manuscript. \section*{Funding} This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. \bibliographystyle{elsarticle-num}
8b991deac18813db505269eef74924c02c11333b
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \subsection{Background and Importance} The minor planets orbiting beyond Neptune provide valuable insight on our solar system's formation and evolution, but they have only been studied since 1992 when the first Trans-Neptunian Object (TNO) after Pluto was discovered \citep{Luu1993}. Almost a quarter century later, $\sim2000$ TNOs and Centaurs are known (see Figure \ref{real}) and they are revealing their properties slowly because of the difficulties involved with detecting the faint, slow-moving members of this distant population. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{real.pdf} \caption{(A) Eccentricity, (B) inclination, and (C) absolute magnitude, versus perihelion distance of known TNOs (including SDOs; orange, magenta, and blue marks), Centaurs (brown), and Inner Oort Cloud objects (IOCs; red dots). Objects previously discovered by PS1 through MOPS and IASC (as mentioned in the text) are depicted as yellow triangles. The limiting absolute magnitude is shown as a dashed line in panel (C) for $V=22.5$.} \label{real} \end{figure} Multiple dedicated TNO surveys have been conducted over the years \citep[e.g.,][]{Larsen2001,Gladman2001, Bernstein2004,Elliot2005,Petit2006,Petit2008,Sheppard2011,Gladman2012,Alexandersen2014,Brown2015} including stellar occultation surveys \citep[e.g.,][]{Schlichting2012} which focussed on the discovery of sub-km objects below the sensitivity limit of optical telescopes. Thanks to these studies, the large $100-1000$~km objects have a well characterized size-frequency distribution \citep[SFD;][] {Petit2008,Fuentes2008} while TNOs smaller than $100$~km have only more recently been studied \citep{Fraser2009,Sheppard2010b,Gladman2012}. However, there is a need for more observational data to confirm the apparent transition from a steep to shallow SFD slope among the Neptune Trojans \citep{Sheppard2010b} and SDOs \citep{Shankman2013} around $D\sim100$~km (corresponding to absolute magnitude $H\sim8.5$) \citep{Alexandersen2014}. If the transition is present within all TNO sub-populations it would suggest the formation scenario in which ``asteroids were born big'' \citep{Morbidelli2009} and imply that objects smaller than $100$~km are dominantly the result of collisional evolution. Most of the known TNOs were discovered in `deep and narrow' observing campaigns using large telescopes with small fields of view. Current Near Earth Object (NEO) surveys \citep{Larson1998, Kaiser2004} have the advantage of continuously monitoring large portions of the sky over several years, but are disadvantaged because they use smaller telescopes with cadences designed to identify NEOs that move more than $10\times$ faster than TNOs. \citet{Brown2015} searched archival data from the Catalina Sky Survey \citep{Larson2003} and Siding Spring Survey \citep{Larson2003} and independently identified the eight brightest known TNOs. Even though they did not discover any new objects they predicted a 32\% chance that an object having magnitude $V<19.1$ remains undiscovered in the unsurveyed region of the sky. Evidence has been mounting in the past few years that there is a large planetary-sized distant object in our solar system whose gravitational perturbations influence the orbits of Scattered-Disc Objects (SDOs), particularly those on orbits similar to the dwarf planet (90377) Sedna \citep[e.g.,][]{Trujillo2014,DeFuMarcos2015,Batygin2016}. These works suggest that all currently known extreme TNOs with semi-major axis greater than $150$~AU (including the only other Sedna-like object: 2012~VP$_{113}$) show a pronounced clustering in their arguments of perihelia ($\omega$) not present in the closer TNO population. \citet{Trujillo2014} suggest this clustering is centred at $\omega\sim0^{\circ}$ and that it is due to the Lidov---Kozai effect, a three-body interaction capable of constraining $\omega$ \citep{Kozai1962}. They propose that a super-Earth mass body located at $\sim250$~AU would be capable of restricting $\omega$ for these objects and be stable for billions of years. However, \citet{Batygin2016} made a similar calculation, but excluded orbits which do not demonstrate long term stability because of Neptune, and found that the distant TNOs cluster around $\omega\sim318^{\circ} \pm 8^{\circ}$ which is inconsistent with the Kozai mechanism. They suggest instead that the clustering can be maintained by a distant ten Earth-mass planet on an eccentric orbit with semi-major axis $700$~AU, nearly co-planar with the distant TNOs, but with $\omega$ shifted by $180^{\circ}$. In addition, such a planet might explain the presence of highly inclined TNOs whose existence has not yet been explained \citep{Gladman2009b}. \citet{Trujillo2014} also state that another plausible explanation for such a peculiar asymmetric $\omega$ configuration would be a strong stellar encounter with the Oort cloud in the past. Increasing the number of known retrograde TNOs and Sedna-like SDOs is needed to further test these hypotheses and to constrain the orbital elements and mass of any potentially undiscovered planet. Towards that end, in this work we report on the discovery and detection of the largest number of TNOs by a single asteroid survey, which due to its long-duration and wide-field coverage, provides an excellent complement to targeted deep-and-narrow surveys, resulting in a relatively unbiased TNO sample. \subsection{Pan-STARRS} The prototype telescope for the Panoramic Survey Telescope and Rapid Response System (Pan-STARRS1, hereafter referred to as PS1) located in the United States on Haleakala, Maui, Hawaii, has been surveying the sky since 2010. Many of the observations by PS1 were taken as a sequence of four exposures, each separated by a Transient Time Interval (TTI) of $\sim20$ minutes. This cadence was selected to optimise detection of Near Earth Objects (NEOs) --- objects which have perihelia $q<1.3$~AU. Observations from each night are rapidly processed by the Image Processing Pipeline \citep{Magnier2006}, and all detected moving objects identified by the Moving Object Processing System \citep[MOPS;][]{Denneau2013} are reported to the Minor Planet Center. PS1 has become the leading discovery telescope for NEOs, discovering almost half of the new Near Earth Asteroids in 2015, and discovering more than half of the new comets in 2015 \citep{Wainscoat2015}. The detection of NEOs is done using subtraction of image pairs which have a TTI of $\sim20$ minutes and are well matched in image quality and telescope pointing. This TTI spacing produces a lower limit on the rate of motion for detection of moving objects, below which moving objects are self-subtracted in their image pairs. The lower limit is typically $\sim0.04^{\circ}$ per day ($=2\mathrm{"}$ in $20$ ~minutes), and is seeing dependent. A substantial number of Centaurs (which we define as having perihelia between Jupiter and Neptune) have been discovered from the pair-subtracted images, but only a few more-distant objects have been reported from PS1 (see Figure \ref{real}), some of which were discovered via the International Astronomical Search Collaboration (IASC\footnote{http://iasc.hsutx.edu/}), an educational outreach program in which images were blinked manually. Now that PS1 has thoroughly surveyed the sky north of $-30^{\circ}$ declination, other methods become viable for object detection that are potentially more sensitive to both fainter and slower moving objects. One method uses subtraction of a high-quality static sky image, derived from the cumulative survey data. The other method uses the historical survey to establish a catalogue of stationary objects, and compares catalogues of new detections in new images to the static sky, to reveal moving objects. Over the course of the PS1 survey, image quality has improved, but the grid structure in the PS1 CCDs requires many dithered images to produce a clean static sky image. And although good images in the \textit{gri} passbands are now available for much of the sky north of $-30^{\circ}$ declination, the coverage in the more sensitive \textit{w} passband is more sparse, because surveying in that band has been more focused on the ecliptic for the purpose of NEO discovery. The PS1 survey has also only recently been extended south to $-49^{\circ}$ declination. For these reasons, we have focussed our initial exploration of methods to extract fainter moving objects on the catalogue based approach. \section{Methodology} To locate moving sources in the PS1 data, a new search method was developed and run on source catalogues previously generated by the IPP \citep{Magnier2006} from PS1 images taken between 2010 Feb 24 and 2015 July 31. These catalogues are generated by a source extraction program which identifies and measures the point spread function for objects in the images. A detection is the information recorded about an object in a single exposure, and the catalogues contain detections of moving objects as well as stationary sources which must be removed. The method, as described in the following sections, links sets of detections (corresponding to the same object) from a single night into a `tracklet'. Tracklets from multiple nights which correspond to the same object are then searched for. Figure \ref{path} shows the apparent path across the celestial sphere over four years for a typical TNO. Our method first searches for two related tracklets which are used to generate an initial orbit from which ephemerides are calculated to identify additional tracklets. All detections identified for an object have their image stamps extracted, which are visually inspected to ensure they are real and do not correspond to image artefacts or stationary sources that were not removed. \begin{figure}[hp] \centering \includegraphics[width=0.9\textwidth]{path.pdf} \caption{The ephemerides of the initial orbit (determined from a tracklet pair) and the final five-tracklet orbit for a sample event. Because the motion cannot be represented with a linear or quadratic projection (the loops are due to the motion of the Earth), a tracklet linking routine based on ephemerides must be used. For this event, the ephemerides from the two orbits differ by $\sim9\arcmin$ after six months, and by $\sim15\arcmin$ after one year, which is less than the $20\arcmin$ search radius described in the text.} \label{path} \end{figure} While the PS1 MOPS pipeline \citep{Denneau2013} was designed to identify TNOs, it requires a survey pattern and cadence that was not implemented. The method presented here is able to link TNOs using a survey pattern intended for NEO detection. \subsection{Stationary Source Identification and Removal} Objects in the outer solar system appear to move slowly across the celestial sphere, with their apparent paths dominated by the motion of Earth. TNOs in a 3:2 mean motion resonance with Neptune, such as Pluto, will have a maximum angular speed at opposition of $0.022^{\circ}$ per day, equal to $3.2\mathrm{"}$ per hour for orbits near zero inclination and eccentricity. An object much farther out at $550$ AU will move $0.25\mathrm{"}$ (the PS1 pixel scale) in one hour. However, even the most distant objects which move less than the astrometric uncertainty in one night will move a noticeable amount over multiple nights, allowing visual verification that they are indeed moving targets. To identify stationary sources, consider a specific telescope pointing (defined by the boresight direction and telescope rotation). All source detections for all exposures overlapping the field-of-view at this boresight direction are loaded. For each detection, the number of neighbouring detections within two magnitudes and $0.6\mathrm{"}$ are counted. Because each PS1 telescope pointing is typically observed four times in a given night (the average TTI is $19$~minutes, giving an average $57$~minute arc), we consider a detection stationary if it is present more than four times, i.e., if there are detections at the same location on the celestial sphere within $0.6\mathrm{"}$ over multiple nights. These criteria were determined empirically and relate to the astrometric and photometric uncertainty in the PS1 images. Once all stationary sources are identified, they are removed from the catalogues. The detections which remain are then used for tracklet creation. Note that while valid detections may be rejected, $100\%$ efficiency is not required as candidate objects are likely to have been observed multiple times at different positions along their orbits. \subsection{Tracklet Creation} Tracklets are formed from detections following a procedure similar to the stationary source removal. The detections from all exposures of a given pointing for a single night are loaded and iterated over. All detections within $0.4$ magnitudes and $16\mathrm{"}$ from each other are assigned a tracklet number. The choice of $16\mathrm{"}$ means many tracklets for inner Solar System objects will be present in the tracklet dataset, as this corresponds to $0.43^ {\circ}$ per day for exposures with a TTI of $15$ minutes. While these objects can be linked with our algorithm, we exclude any identified object having semi-major axis $< 4.8$~AU, as we are only interested in objects at heliocentric distances corresponding to Jupiter's orbit and beyond. While we could use a search distance $<16\mathrm{"}$, this choice does allow us to identify tracklets where the second or third detection occurred within a CCD cell gap. Once formed, each tracklet has its motion along a great circle fit with a constant angular speed model. Only tracklets with RMS residual $< 0.3\mathrm{"}$ and having $\ge3$ detections are kept, and their fitted angular speed and position angle are recorded. We do not create tracklets from detection pairs because we cannot apply an RMS test to judge their astrometric quality. Some fraction of the formed tracklets may be faint stars or image artefacts which exhibit linear motion, and their presence will increase the required computational time since there are more tracklet comparisons to be made. We validate all orbit fits using residual checks. This linking process is different from the kd-tree based algorithm used by MOPS \citep{Denneau2013} described in detail by \citet{Kubica2007}. The method presented here is computationally faster, but is limited to tracklets moving at much slower speed, and it cannot handle the intersection of tracklets which correspond to detections of different objects with similar apparent magnitude. \subsection{Tracklet Pairing} To test if two tracklets correspond to the same object, a brute-force approach is used. All tracklet pairs occurring within up to $6^{\circ}$ (equal to the maximum angular speed of $0.1^{\circ}$ per day with a $60$ day window) have a test orbit fit using FindOrb \footnote{http://www.projectpluto.com/find\_orb.htm}, and we require its reported mean residual $<0.3\mathrm{"}$. The choice of a $60$ day window is a trade off between an improved initial arc length, and the computation time needed to test all tracklet pairs, and is reasonable since the PS1 survey pattern has observed much of the celestial sphere on at least three separate nights within this timespan (see Figure \ref{look}). However, this method is affected by missed tracklets which may have fallen into CCD cell gaps, been obscured by other image artefacts, or are too close to bright stars. We do not require every tracklet pair for an object to be identified, as many TNOs should be detected in more than two tracklets for the timespan considered here (see Figure \ref{look}). It is important to note however, that some orbital geometries will not have any observations: for example, there are no PS1 observations of the Centaur (10199) Chariklo which is located near the galactic centre. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{look.pdf} \caption{Minimum timespan in the PS1 survey to observe the celestial sphere on three separate nights, ignoring observations consisting of only one or two exposures. The block pattern is due to how the survey strategy was implemented. The ecliptic plane is shown as a curved line. Black regions indicate three observations have never been made, while white (mostly corresponding to the galactic plane) means there are more than $120$ days between the first and third nights. The majority of the surveyed area satisfies the tracklet pairing search window of $60$ days.} \label{look} \end{figure} To increase computational efficiency, tracklet pairs are identified using a kd-tree indexed approach, and the position angle of both tracklets must be within $24^{\circ}$ of a line connecting the two tracklets. This value was empirically chosen and considerably reduces the computation time required as it excludes certain unphysical geometries. We consider only tracklets having angular speed between $0.1^{\circ}$ and $0.001^{\circ}$ per day, as our priority in this work is to find TNOs. An extended search for more distant objects will be considered in future work. \subsection{Identifying additional tracklets} The search for additional tracklets to extend an object's arc length requires a starting orbit, for which we use the tracklet pairing results described in the previous section. The metadata for each PS1 exposure is loaded, sorted by the observation time relative to the starting orbit's epoch (where its mean anomaly is defined), and iterated over. For each exposure, an ephemeris is generated, and if located within the field-of-view, the corresponding tracklet database is searched. For each tracklet which contains detections within $20\arcmin$ of the test ephemeris, a new orbit is fit using FindOrb. If the reported mean residual is $<0.2\mathrm{"}$, the tracklet is considered linked, its astrometry appended, and the updated orbit fit kept. The search then continues for additional tracklets. The choice of $0.2\mathrm{"}$ is more strict than the limit used during tracklet pairing, but has identified up to $27$ additional tracklets for the candidate objects. \begin{table}[ht] \centering \caption{Typical sample of a classical TNO (with $H=6.5$) showing how its orbit converges as additional tracklets are found, with five total tracklets being identified. All orbits are integrated to the 20140702.0 epoch used for the initial two-tracklet case. The first column ($N$) in the table is the number of tracklets used to compute the orbit.} \begin{tabular}{cccccccc} \\ [1ex] \toprule $N$ & $a$ [AU] & $e$ & $i$ [$^{\circ}$] & $\omega$ [$^{\circ}$] & $\Omega$ [$^{\circ}$] & $q$ [AU] & $M$ [$^{\circ}$] \\ \midrule 2 & 33.2933 & 0.0726 & 7.374 & 94.233 & 198.516 & 30.8756 & 325.536 \\ 3 & 39.2198 & 0.2176 & 7.283 & 59.306 & 197.561 & 30.6844 & 357.708 \\ 4 & 39.4907 & 0.2247 & 7.288 & 63.932 & 197.506 & 30.6186 & 354.920 \\ 5 & 39.3916 & 0.2212 & 7.279 & 57.769 & 197.518 & 30.6778 & 358.711 \\ \bottomrule \end{tabular} \label{morb} \end{table} Table \ref{morb} shows how the initial test orbit of a sample object converges as additional tracklets were found. The difference in the ephemeris for the initial and final orbit is illustrated in Figure \ref{path}. The choice of the $20\arcmin$ search distance was chosen based on identified TNOs: Figure \ref{slip} shows the angular distance between ephemerides from the initial tracklet-pair based orbit and the final fitted orbit for all identified objects. The majority of events show $<20\arcmin$ difference over $\pm 120$~days, where (based on Figure \ref{look}) most of the celestial sphere has been observed on at least three nights during the PS1 survey. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{slip.pdf} \caption{The difference between the ephemeris from the initial tracklet-pair orbit and the final orbit for all TNO events. Using a tracklet search radius of $20\arcmin$ gives a $\sim250$ day window to find additional tracklets. The two events reaching an ephemeris error of $\sim90\arcmin$ within two weeks are due to particularly short arcs in the initial tracklet pairing.} \label{slip} \end{figure} While we could integrate the test orbits using Mercury6 \citep{Chambers1999} to account for their orbital evolution, this is not required for TNOs as the generated ephemerides will change by much less than the $20\arcmin$ search distance. \subsection{Visual Inspection and Classification} The image stamps for each detection from all identified objects were extracted and visually inspected to confirm they were real and not affected by image artefacts. The orbital elements of the identified objects were used to classify them as potential Trojans, classical or resonant TNOs, Scattered Disc Objects (SDOs), or Centaurs. The orbital elements were also used to identify known objects by comparing the location and position angle of all ephemerides at their orbit epoch to those computed using the Minor Planet Center (MPC) orbit catalogue after integration using Mercury6 \citep{Chambers1999}. This method assumes the orbits from the MPC are more accurate than those identified in this study, which may not be true if our identified objects were independently previously observed by other telescopes over shorter orbital arcs. \section{Results and Discussion} \label{results} The source catalogues used in this study span the time from 2010 Feb 24 to 2015 July 31, and are generated from $529\,609$ exposures in $61\,065$ pointings, with $93\,799\,429\,652$ total detections. The stationary source removal left $7\,655\,731\,998$ detections of which $232\,447\,038$ were linked into $65\,524\,472$ tracklets. Figure \ref{figA} shows eccentricity vs semi-major axis for all identified objects, and we present a breakdown of these in Table \ref{count}, including the number of unknown and known objects, as well as the number expected (both total and to a limiting $V=22.5$) based on the MPC catalogue. We further discuss our identified objects in the following sections. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{fig_A.pdf} \caption{Eccentricity versus semi-major axis for the objects identified in this work, colour coded by absolute magnitude. The detected objects consist of $789$ Jupiter Trojans, $154$ Centaurs, and $465$ total TNOs. The $1:1$ and $3:2$ mean motion resonances with Neptune are noted with vertical dashed lines, while Neptune's aphelion distance of $q=30.4$~AU is noted with a solid curved line.} \label{figA} \end{figure} \begin{table}[ht] \centering \caption{Classification of identified objects, also broken down into known and unknown objects. The fifth column shows the expected number from the MPC catalogue as a fraction of the total for a limiting magnitude of $V=22.5$.} \begin{tabular}{lccccc} \\ [1ex] \toprule orbital type & total & unknown & known & MPC ($V<22.5$) & MPC total \\ \midrule Jupiter Trojans & $789$ & 145 & $644$ & $6169$ & $6207$ \\ Centaurs & $154$ & 78 & $76$ & $211$ & $303$ \\ Classical TNOs & $255$ & 162 & $91$ & $197$ & $1198$ \\ Resonant TNOs & $121$ & 77 & $44$ & $93$ & $319$ \\ Scattered Disc & $89$ & 52 & $37$ & $57$ & $203$ \\ \bottomrule \end{tabular} \label{count} \end{table} \subsection{Bright Objects} \citet{Brown2015} estimate that there is a 32\% chance that a TNO having magnitude $V<19.1$ remains undiscovered after their archival search of the Catalina Sky Survey database. We computed ephemerides for the five year period of data used in this study, for all identified objects, and list the peak brightness of the top eight in Table \ref{bright}. There is one object which reaches $V=18.5$, which is the third brightest TNO over the timespan of data used in this study. \begin{table}[ht] \centering \caption{The brightest eight objects (between 2010 Feb 24 and 2015 July 31) identified in our study. The first column gives the total tracklet count for each object. The magnitude column ($V_{min}$) gives the brightest apparent $V$ magnitude that each object reached during the five year PS1 survey data used based on the absolute magnitude ($H$) determined from FindOrb, and $N$ is the number of tracklets used for the orbit computation.} \begin{tabular}{ccccccl} \\ [1ex] \toprule $N$ & $a$ [AU] & $e$ & $i$ [$^{\circ}$] & $H$ & $V_{min}$ & object \\ \midrule 13 & 43.2588 & 0.1905 & 28.193 & 0.1 & 17.2 & (136108) Haumea \\ 11 & 67.7662 & 0.4410 & 44.057 & -1.6 & 18.2 & (136199) Eris \\ 11 & 36.2522 & 0.1894 & 1.497 & 3.8 & 18.5 & \textbf{unknown} \\ 12 & 39.4661 & 0.2182 & 20.554 & 1.8 & 18.6 & (90482) Orcus \\ 16 & 39.5548 & 0.2786 & 15.464 & 4.1 & 18.6 & (38628) Huya \\ 25 & 39.3913 & 0.2243 & 8.411 & 4.2 & 19.0 & (47171) 1999 TC$_{36}$ \\ 06 & 69.2813 & 0.5289 & 29.424 & 3.3 & 19.0 & 2010 EK$_{139}$ \\ 16 & 41.3601 & 0.0193 & 19.307 & 3.1 & 19.2 & (145452) 2005 RN$_{43}$ \\ \bottomrule \end{tabular} \label{bright} \end{table} \subsection{Large Objects} Figure \ref{figC} plots semi-major axis vs absolute magnitude for all identified objects. The largest identified objects are all known, and are further listed in Table \ref{biggies}. However, (136472) Makemake and (134340) Pluto are not identified. For the first case, its tracklets do not occur within a $60$ day window, and for the latter, the only tracklet pair occurred with a two day arc, and additional tracklets were not linked. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{fig_C.pdf} \caption{Semi-major axis vs absolute magnitude for all identified objects. The cluster along $a\sim5.2$\,AU at the lower right are Jupiter Trojans. The largest objects (left most in the figure) are all known.} \label{figC} \end{figure} There are no unknown distant planetary-sized objects identified in our study. However, our minimum tracklet angular speed cut-off of $0.001^{\circ}$ per day could exclude them if they are present in the source catalogues. \begin{table}[ht] \centering \caption{The most distant objects identified in our study. The column headings are identical to those given in Figure \ref{bright}.} \begin{tabular}{ccccccl} \\ [1ex] \toprule $N$ & $a$ [AU] & $e$ & $i$ [$^{\circ}$] & $H$ & $V_{min}$ & object \\ \midrule 11 & 67.7662 & 0.4410 & 44.057 & -1.60 & 18.2 & (136199) Eris \\ 13 & 43.2588 & 0.1905 & 28.193 & 0.10 & 17.2 & (136108) Haumea \\ 12 & 525.8952 & 0.8554 & 11.929 & 0.80 & 20.1 & (90377) Sedna \\ 19 & 66.8888 & 0.5065 & 30.946 & 1.10 & 20.4 & (225088) 2007 OR$_{10}$ \\ 12 & 39.4661 & 0.2182 & 20.554 & 1.80 & 18.6 & (90482) Orcus \\ 18 & 45.7416 & 0.1402 & 21.497 & 2.70 & 19.4 & (174567) Varda \\ 11 & 47.5112 & 0.1304 & 24.337 & 2.70 & 19.3 & (55565) 2002 AW$_{197}$ \\ 11 & 76.5622 & 0.5186 & 23.344 & 2.80 & 19.2 & (229762) 2007 UK$_{126}$ \\ 13 & 43.1700 & 0.0506 & 17.154 & 3.00 & 19.4 & (20000) Varuna \\ 10 & 112.7967 & 0.6744 & 14.025 & 3.10 & 20.0 & (303775) 2005 QU$_{182}$ \\ 10 & 81.4754 & 0.5860 & 7.565 & 3.10 & 21.2 & 2015 RR$_{245}$ \\ \midrule \bottomrule \end{tabular} \label{biggies} \end{table} \subsection{Distant Objects} Figure \ref{figE} plots semi-major axis vs distance from the Sun (on 2015 July 31). The farthest of these objects are listed in Table \ref{distant}, with the six most distant being known (including the recent 2015 RR$_{245}$ \footnote{http://www.minorplanetcenter.net/mpec/K16/K16N67.html}). In addition to these, we have identified two previously unknown objects which have perihelia well beyond the orbit of Neptune. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{fig_E.pdf} \caption{Semi-major axis vs distance from the Sun (on 2015 July 31) for all identified objects. The colour scheme shows the true anomaly, which gives a direct measure of where each object is along its orbit relative to perihelion. Objects significantly above the dashed line have much larger eccentricity, with comets being located towards the upper left of the plot.} \label{figE} \end{figure} \begin{table}[ht] \centering \caption{The most distant identified objects in our dataset, where $r_{max}$ is their heliocentric distance on 2015 July 31. The first column ($N$) gives the tracklet count for each object. Two of the identified objects are not known.} \begin{tabular}{ccccccl} \\ [1ex] \toprule $N$ & $a$ [AU] & $e$ & $i$ [$^{\circ}$] & $H$ & $r_{max}$ & object \\ \midrule 11 & 67.7662 & 0.4410 & 44.057 & -1.6 & 96.3 & (136199) Eris \\ 19 & 66.8888 & 0.5065 & 30.946 & 1.1 & 87.3 & (225088) 2007 OR$_{10}$ \\ 12 &525.8952 & 0.8554 & 11.929 & 0.8 & 85.9 & (90377) Sedna \\ 10 & 81.4754 & 0.5860 & 7.565 & 3.1 & 64.5 & 2015 RR$_{245}$ \\ 14 & 62.0774 & 0.3355 & 28.814 & 3.4 & 58.9 & 2015 KH$_{162}$ \\ 07 & 57.7016 & 0.1103 & 46.604 & 3.5 & 57.5 & 2004 XR$_{190}$ \\ 19 & 62.1612 & 0.2245 & 10.252 & 3.2 & 56.2 & \textbf{unknown} \\ 08 & 43.9118 & 0.2511 & 14.878 & 4.0 & 53.8 & \textbf{unknown} \\ \bottomrule \end{tabular} \label{distant} \end{table} \subsection{Orbital Clustering} \citet{Batygin2016} suggest the orbital configuration of TNOs decoupled from Neptune's influence support the presence of a planetary-sized perturber in the outer solar system. Due to the sensitivity limit of PS1 ($V\sim22.5$), we can only identify two of these objects, namely (90377) Sedna and 2007 TG$_{422}$, and both were identified in this study. We present the arguments of perihelia of our identified objects in Figure \ref{figB}. Using a Kolmogorov-Smirnov test on all identified objects having $q > 32$~AU and $q > 36$~AU, we do not see clustering in $\omega$. To our sensitivity limit, this is significant as our wide field survey might be expected to have less observational bias than targeted narrow searches. Survey bias might also be expected to induce apparent clustering in an observed population, rather than remove it. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{fig_B.pdf} \caption{Semi-major axis versus argument of perihelion for all identified objects having $q >36$~AU. The location of (90377) Sedna and (136199) Eris are indicated. Sedna is the only object identified in this survey that supports the orbital clustering discussed in \citet{Trujillo2014} and \citet{Batygin2016}.} \label{figB} \end{figure} \subsection{High Inclination Objects} From Figure \ref{figD}, there are several identified objects with high inclination, and we list those having the largest inclination in Table \ref{retro}. We identified in our study the recent 2011 KT$_{19}$ \footnote{Minor Planet Circulars Orbit Supplement 380701}, and also have one unknown object with $i=84^{\circ}$. The orbital evolution of highly-inclined TNOs is not understood, but \citet{Gladman2009b} suggest they could have been perturbed into their current orbits by a planetary-sized perturber in the outer solar system. Another plausible explanation is that they are captured objects. \begin{figure}[p] \centering \includegraphics[width=0.9\textwidth]{fig_D.pdf} \caption{Inclination vs semi-major axis for all identified objects, colour coded by Tisserand's parameter wrt. Neptune. The location of the giant planets are noted with vertical lines, with a large grouping of Jupiter Trojans towards the lower left. The identified object with highest eccentricity is known object 2011 KT$_{19}$.} \label{figD} \end{figure} \begin{table}[ht] \centering \caption{The identified objects (excluding comets) with highest inclination. The magnitude ($V_{min}$) is the brightest each objects reaches over the five year data span used, and is determined from the absolute magnitude ($H$) as fitted by FindOrb. The first column ($N$) gives the number of tracklets used in the orbit computation.} \begin{tabular}{ccccccl} \\ [1ex] \toprule $N$ & $a$ [AU] & $e$ & $i$ [$^{\circ}$] & $H$ & $V_{min}$ & object \\ \midrule 19 & 35.6490 & 0.3330 & 110.206 & 6.6 & 20.5 & 2011 KT$_{19}$ \\ 06 & 259.7509 & 0.9354 & 84.824 & 9.1 & 21.5 & \textbf{unknown} \\ 06 & 349.2179 & 0.9685 & 68.030 & 9.4 & 20.0 & (418993) 2009 MS$_{9}$ \\ 06 & 39.4982 & 0.3899 & 54.265 & 6.5 & 20.6 & \textbf{unknown} \\ 10 & 43.3078 & 0.2160 & 47.507 & 6.0 & 21.3 & \textbf{unknown} \\ 07 & 57.7016 & 0.1103 & 46.604 & 3.5 & 21.1 & 2004 XR$_{190}$ \\ \bottomrule \end{tabular} \label{retro} \end{table} \subsection{Neptune Trojans} We list in Table \ref{trojs} two possible Neptune Trojans present in our identified objects, based solely on their orbital elements. One of these objects was identified in a parallel study using PS1 data \citep{2015DPS....4721117L}, however we did not identify their other candidates. We present a discussion as to why this might be in Section \ref{whyyyy}. \begin{table}[ht] \centering \caption{Two possible Neptune Trojans (based on their orbital elements) identified. The first column ($N$)indicates the number of tracklets used to compute the orbit.} \begin{tabular}{ccccccl} \\ [1ex] \toprule $N$ & $a$ [AU] & $e$ & $i$ [$^{\circ}$] & $H$ & $V_{min}$ & object \\ \midrule 16 & 30.1564 & 0.1192 & 6.661 & 5.90 & 20.7 & 2013 KY$_{18}$ \\ 06 & 30.0280 & 0.0472 & 6.562 & 7.10 & 21.6 & 2010 TS$_{191}$ \\ \bottomrule \end{tabular} \label{trojs} \end{table} \subsection{Centaurs and Comets} We define Centaurs to have perihelia between Jupiter and Neptune, but which are not in a mean motion resonance with Neptune. We identified $136$ Centaurs in our study, as well as $10$ comets which were all known objects. Our identification method does not link objects having $e\ge1$. Although \citet{2013AJ....146...36S} call for a better characterised survey of Centaur objects, we consider them a by-product of the current study and leave this to a future publication --- our primary interest here is identifying TNOs. \subsection{Identification Efficiency} \label{whyyyy} Our discussion would not be complete without mentioning the efficiency of creating and pairing tracklets, as well as identifying additional tracklets. As suggested by Table \ref{count}, our method identifies about half of the expected (i.e., having $V<22.5$) known population of classical and resonant TNOs from the MPC catalogue, and two-thirds of the known SDOs. We do not count Jupiter Trojans or Centaur objects here because our detection parameters were empirically chosen to optimise detection of TNOs. A complete end-to-end measure of the identification efficiency could be made by injecting synthetic detections into our software routines \citep[e.g., as is done by MOPS;][]{Denneau2013}, but ideally this must be done directly into the original PS1 image data which would then be used to produce new source catalogues. This reprocessing by the IPP \citep{Magnier2006} is necessary for a realistic efficiency determination as the fill-factor of the PS1 camera is limited to $\sim70\%$, and our tracklet linking routines have no knowledge of which regions on the CCD are masked. Also, our limiting sensitivity is dependent on the photometric passband as well as local weather conditions. Because of these reasons, determining an accurate efficiency is a significant undertaking, which we will address in subsequent work to obtain an unbiased estimate of the true TNO population. From the current MPC catalogue for all known objects, the average distance between each asteroids within $30^{\circ}$ of the ecliptic and its closest neighbour having apparent magnitude within $0.4$ is $\sim820\mathrm{"}$. This is much greater than the $16\mathrm{"}$ search radius used during the tracklet creation process, suggesting that the majority of tracklets in the source catalogue (after removing stationary sources) will not be contaminated by detections associated with different objects. Our tracklets may be limited in terms of their RMS residual, but even for a distant object at $100$~AU, detections will be spaced $0.45\mathrm{"}$ for a $20$ minute TTI, which is twice the PS1 pixel scale. Because Figure \ref{slip} suggests additional tracklets should be readily found, this implies the efficiency of our search implementation is limited by either the stationary source removal process, or the tracklet pairing stage. We also note that while the PS1 camera fill-factor may lead to missed detections for NEOs, it can result in completely missed tracklets for TNOs due to their much slower speed across the celestial sphere. \section{Conclusions} A search for distant objects was made using the archival PS1 data, with $1420$ objects identified, consisting of $255$ classical TNOs, $121$ resonant TNOs, $89$ SDOs, $154$ Centaurs, and $789$ Jupiter Trojans. Excluding the trojans, $371$ of these are new objects which we could not link to known objects. While our identified objects do not show a clustering in their arguments of perihelia, increasing the number of known retrograde TNOs and Sedna-like SDOs is important in better understanding the distant population in our solar system, especially to constrain the orbital elements and mass of any potentially undiscovered large planetary-sized objects \citep{Trujillo2014,Batygin2016}. Future work will focus on validating the detection efficiency, as well as optimising our detection parameters to work well beyond the classical TNO regime. \begin{comment} \section{Appendix} We list here the heliocentric orbital elements for the detected objects. \begin{landscape} \begin{center} \begin{longtable}{@{}ccccccccc@{}} \caption{blah} \label{list} \\ \toprule \multicolumn{1}{c}{\textbf{name}} & \multicolumn{1}{c}{\textbf{a}} & \multicolumn{1}{c}{\textbf{e}} & \multicolumn{1}{c}{\textbf{i}} & \multicolumn{1}{c}{\textbf{$\omega$}} & \multicolumn{1}{c}{\textbf{$\Omega$}} & \multicolumn{1}{c}{\textbf{M}} & \multicolumn{1}{c}{\textbf{H}} & \multicolumn{1}{c}{\textbf{epoch}} \\ \midrule \endfirsthead \multicolumn{9}{c}{{\tablename} \thetable{} -- continued} \\ \toprule \multicolumn{1}{c}{\textbf{name}} & \multicolumn{1}{c}{\textbf{a}} & \multicolumn{1}{c}{\textbf{e}} & \multicolumn{1}{c}{\textbf{i}} & \multicolumn{1}{c}{\textbf{$\omega$}} & \multicolumn{1}{c}{\textbf{$\Omega$}} & \multicolumn{1}{c}{\textbf{M}} & \multicolumn{1}{c}{\textbf{H}} & \multicolumn{1}{c}{\textbf{epoch}} \\ \midrule \endhead \midrule \multicolumn{9}{l}{{continued on next page \ldots}} \\ \endfoot \bottomrule \endlastfoot \_\_01ThKK3\_14 & 224.296 & 0.847 & 30.761 & 6.899 & 179.307 & 1.233 & 5.5 & 20150622.0 \\ \_\_0KcH78Q\_06 & 259.751 & 0.935 & 84.824 & 38.087 & 246.224 & 359.778 & 9.1 & 20150329.0 \\ \_\_0ROeSVK\_06 & 349.218 & 0.968 & 68.030 & 128.696 & 220.236 & 0.101 & 9.4 & 20141211.0 \\ \_\_0DUX1je\_09 & 516.339 & 0.931 & 18.581 & 285.806 & 112.975 & 0.281 & 5.6 & 20150113.0 \\ \_\_01lN6mI\_12 & 525.895 & 0.855 & 11.929 & 311.351 & 144.552 & 358.172 & 0.8 & 20150113.0 \\ \_\_03dAgtH\_08 & 731.751 & 0.967 & 19.488 & 122.365 & 197.353 & 0.163 & 7.0 & 20150725.0 \\ \_\_06rZJ7X\_12 & 871.600 & 0.991 & 16.533 & 158.104 & 349.213 & 0.005 & 9.5 & 20150410.0 \\ \_\_15HJt7S\_05 & 4292.380 & 0.998 & 25.377 & 145.168 & 153.999 & 359.999 & 10.5 & 20140627.0 \\ \_\_08YkBc9\_05 & 172115.128 & 1.000 & 60.407 & 129.870 & 225.138 & 0.000 & 11.8 & 20150113.0 \\ \end{longtable} \end{center} \end{landscape} \end{comment} \acknowledgements \section*{Acknowledgments} The Pan-STARRS1 Surveys (PS1) have been made possible through contributions of the Institute for Astronomy, the University of Hawaii, the Pan-STARRS Project Office, the Max-Planck Society and its participating institutes, the Max Planck Institute for Astronomy, Heidelberg and the Max Planck Institute for Extraterrestrial Physics, Garching, The Johns Hopkins University, Durham University, the University of Edinburgh, Queen's University Belfast, the Harvard-Smithsonian Center for Astrophysics, the Las Cumbres Observatory Global Telescope Network Incorporated, the National Central University of Taiwan, the Space Telescope Science Institute, the National Aeronautics and Space Administration under Grant Nos. NNX08AR22G, NNX12AR65G, and NNX14AM74G issued through the Planetary Science Division of the NASA Science Mission Directorate, the National Science Foundation under Grant No. AST-1238877, the University of Maryland, and Eotvos Lorand University (ELTE) and the Los Alamos National Laboratory. \bibliographystyle{aasjournal}
09724d78fabb82092f078405fbbf006bb4de8668
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:intro} The extraction of the transport properties of the strongly interacting medium created in heavy ion collision (HIC) experiments is currently a very active topic of research in the HIC community. The methods of relativistic hydrodynamics with minimal viscous correction have been quite successful in describing the time evolution of the hot and dense fireball created in the HIC experiments. These kind of investigations have also concluded that the shear viscosity ($\eta$) to entropy density ($s$) ratio, $\eta/s$, of the medium created in HIC experiments is very close to its quantum lower bound $1/4\pi$~\cite{KSS}. Similar to $\eta$, another transport coefficient is the bulk viscosity, $\zeta$, which is defined as the proportionality constant between the non-zero trace of the viscous stress tensor to the divergence of the fluid velocity, and usually it appears associated with processes accompanied by a change in fluid volume or density. The viscous coefficient $\zeta$ has received much less attention than the $\eta$ in hydrodynamical simulations because its numerical value is assumed to be very small, as it is directly proportional to the trace of the energy-momentum tensor, which generally vanishes for conformally symmetric matter~\cite{Tuchin}. However, according to Lattice Quantum Chromo Dynamics (LQCD) calculations~\cite{Lat1}, the trace of the energy momentum tensor of hot QCD medium might be large near the QCD phase transition, which indicates the possibility of a non-zero and large value of $\zeta$ as well as of $\zeta/s$ near the transition temperature. This indication is confirmed by the Refs.~\cite{LQCD_zeta1,LQCD_zeta2}, related with LQCD estimation, where Ref.~\cite{LQCD_zeta2} exposes the possibility of divergence of $\zeta$ near the transition temperature. In recent times, different phenomenological investigations~\cite{Torrieri,Monnai,Kodama,Rajagopal,Bozek,Heinz,HM,Dusling,Victor,Grassi1,Grassi2,Habich,Gale} demonstrated that bulk viscosity can have a non-negligible effect on heavy ion observables, where the values of $\zeta/s$ in Ref.~\cite{Gale} is assumed to be quite large. On the basis of phenomenological importance, microscopic calculations of $\zeta$ for quark gluon plasma (QGP) and hadronic matter is a matter of contemporary interest in the community of HIC. A list of references are~\cite{{Arnold},{Sasaki},{Cassing},{Defu},{G_IFT},{Kinkar},{Kadam3}, {Pratt},{Dobado},{Purnendu},{Tuchin},{Vinod},{Santosh},{Gavin}, {Sarkar},{Kadam1},{Kadam2},{Sarwar},{Noronha},{Nicola}}, where Ref.~\cite{Arnold} addressed high temperature perturbative QCD calculations of $\zeta$, Refs.~\cite{Sasaki,Cassing,Defu,G_IFT,Kinkar,Kadam3} have gone through Nambu-Jona-Lasinio (NJL) model calculations of $\zeta$ and Refs.~\cite{Pratt,Dobado,Purnendu} provided the discussions on Linear Sigma Model (LSM) estimation of $\zeta$. These effective QCD model calculations ~\cite{Sasaki,Cassing,Defu,G_IFT,Kinkar,Kadam3,Pratt,Dobado,Purnendu} cover both QGP and hadronic phases while hadronic-model calculations of Refs.~\cite{Sarkar,Kadam1,Kadam2,Sarwar,Noronha,Nicola} are restricted within hadronic phase only. The present work is also addressing the estimation of $\zeta$ in the hadronic phase only. At vanishing baryonic chemical potential, most of the microscopic calculations predict that $\zeta(T)$ increases but $\zeta/s(T)$ decreases in the hadronic temperature domain. However, few exceptions are there depending on different scenario. For example, Ref.~\cite{Purnendu} showed that the decreasing function of $\zeta/s(T)$ is transformed to an increasing function in the hadronic temperature domain, when its medium constituents sigma meson becomes heavier. Similar kind of fact is also observed in Ref.~\cite{Dobado} depending on the different nature of phase transition as well as methodological differences of LSM calculations. In the hadronic temperature domain, a decreasing nature of $\zeta(T)$ is observed in Ref.~\cite{Sasaki} while Ref.~\cite{Kadam1} estimated increasing $\zeta/s(T)$. These knowledge from the earlier investigations suggest that the nature of $\zeta(T)$ and $\zeta/s(T)$ are still not very settled issues. Again, the numerical strength of $\zeta$ and $\zeta/s$ from different model calculations exhibit a large band - $\zeta\sim 10^{-5}$ GeV$^3$~\cite{Sarkar} to $10^{-2}$ GeV$^3$~\cite{Sasaki} or, $\zeta/s\sim 10^{-3}$~\cite{Sarkar} to $10^{0}$~\cite{Sasaki}. These uncertainty in nature as well as numerical values of $\zeta(T)$ from the earlier investigations demand for further research on these kind of microscopic calculations. Owing to that motivation, we have gone through a microscopic calculations of $\zeta$ and $\zeta/s$, where equilibrium situations of hadronic matter are controlled by the standard HRG model and non-equilibrium picture of medium constituents is introduced via quantum fluctuation of pion and nucleon in medium. With respect to the earlier HRG calculations of $\zeta$~\cite{Kadam1,Kadam2,Sarwar,Noronha}, the main distinguishable contribution is in the non-equilibrium properties of medium constituents, quantified by their thermal width. Assuming pions and nucleons as most abundant constituents of medium, we have calculated their thermal width, which are coming from their in-medium scattering with different possible mesonic and baryonic resonances. The main formalism for this thermal width calculations of pion and nucleon are explicitly described in the Section~\ref{sec:form}, which is started with a brief description HRG model, handling the equilibrium part. Next, the numerical results are discussed in Section~\ref{sec:num} and lastly, our investigations have been summarized and concluded in Section~\ref{sec:concl}. \section{Formalism} \label{sec:form} The HRG system is an ideal gas of hadrons and their resonances are taken from the Particle Data Book~\cite{pdg}. Here we consider all resonances up to 2 GeV masses. The recent LQCD data at zero baryon chemical potential $(\mu_B)$ show that for temperatures up to the crossover region ($150 - 160$ MeV), HRG provides a reasonably good description of the LQCD thermodynamics~\cite{LQCDHRG1, LQCDHRG2, LQCDHRG3}. All thermodynamic quantities of the HRG can be computed from the logarithm of total partition function \begin{equation} \ln Z_{HRG}\left( T,\mu_B,\mu_Q,\mu_S\right)=\sum_i \ln Z^i_s\left( T,\mu_B, \mu_Q, \mu_S\right)~, \label{eq.ZHRG} \end{equation} where \begin{equation} \ln Z^i_s=\frac{g_i}{2\pi^2}VT^3\sum_{n=1}^\infty \frac{(\mp1)^{(n+1)}}{n^4}\left( \frac{n m_i}{T}\right)^2 K_2 \left( \frac{n m_i}{T} \right)e^{n\beta \mu_i} \label{eq.Zs} \end{equation} is the single particle partition function of the $i$th hadron. In Eq.~(\ref{eq.Zs}), $g_i$ is the degeneracy factor of ith particle with mass $m_i$, $V$ is volume of the medium, and $K_2(..)$ is the modified Bessel function. Under the condition of complete chemical equilibrium, all the hadron chemical potentials can be expressed in terms of only three chemical potentials corresponding to the QCD conserved charges \begin{equation} \mu_i=B_i{\mu_B}_i + Q_i{\mu_Q}_i + S_i{\mu_S}_i \label{eq.mui} \end{equation} where $B_i$, $Q_i$ and $S_i$ are the baryon number, electric charge and strangeness of the $i$th hadron. It is straightforward to compute other thermodynamic quantities from $Z_{HRG}$, such as pressure ($P$), energy density ($\epsilon$), entropy density ($s$): \begin{eqnarray} P &=& -\frac{T}{V}\ln Z_{HRG}\label{eq.PHRG}~,\\ \epsilon &=& \frac{1}{V}\left\{T^2\frac{\partial\ln Z_{HRG}}{\partial T} + \sum_i \mu_i T\frac{\partial \ln Z_{HRG}}{\partial \mu_i}\right\}\label{eq.EHRG}~,\\ s &=& \frac{1}{T}\left\{\epsilon+P-\frac{1}{V}\sum_i \mu_iT\frac{\partial \ln Z_{HRG}} {\partial \mu_i} \right\}\label{eq.sHRG}~. \end{eqnarray} Square of the speed of sound is defined as \begin{equation} c_s^2 = \left(\frac{\partial P}{\partial \epsilon}\right)_{\rho_B}\label{cs2}~, \end{equation} where $\rho_B$ is net baryon density. From the Relaxation Time Approximation (RTA) of kinetic theory approach~\cite{Purnendu,Gavin} or from the one-loop expression of diagrammatic approach based on Kubo formula~\cite{Nicola}, we can get standard expressions of bulk viscosity coefficient for pion and nucleon components~\cite{Purnendu,Gavin,Nicola,Kadam2} : \begin{equation} \zeta_\pi = \left(\frac{g_\pi}{T}\right) \int \frac{d^3\boldsymbol{k}}{(2\pi)^3} \, \frac{ n_\pi \left[1 + n_\pi\right]}{\omega_\pi^2 \, \Gamma_\pi} \left\{\left(\frac{1}{3} - c_s^2\right) \boldsymbol{k}^2 - c_s^2 m_\pi^2 \right\}^2 \label{zeta_pi} \end{equation} and \begin{eqnarray} \zeta_N &=& \left(\frac{g_N}{T} \right) \int \frac{d^3\boldsymbol{k}}{(2\pi)^3} \, \frac{1}{{\omega_N^2 \, \Gamma_N}}\left[ \left\{ \left(\frac{1}{3} - c_s^2\right) \boldsymbol{k}^2 - c_s^2 m_N^2 \right.\right.\nonumber \\ &&\left.\left. - \omega_N\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon} \right\}^2 n^+_N \left(1 - n^+_N\right) + \left\{ \left(\frac{1}{3} - c_s^2\right) \boldsymbol{k}^2 \right.\right.\nonumber\\ && \left.\left. - c_s^2 m_N^2 + \omega_N\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon} \right\}^2 n^-_N \left(1 - n^-_N\right)\right] ~, \label{zeta_N} \end{eqnarray} where $n_\pi=1/\{e^{\omega_\pi/T}-1\}$ is the Bose-Einstein (BE) distribution function of pion with energy $\omega_\pi=\{\boldsymbol{k}^2 + m_\pi^2\}^{1/2}$, $n^{\pm}_N=1/\{e^{(\omega_N \mp \mu_B)/T}+1\}$ are the Fermi-Dirac (FD) distribution functions of nucleon and anti-nucleon respectively with energy $\omega_N=\{\boldsymbol{k}^2 + m_N^2\}^{1/2}$ at finite temperature $T$ and baryon chemical potential $\mu_B$. The degeneracy factors of pion and nucleon components are $g_\pi=3$ and $g_N=2\times 2$ respectively. \begin{figure} \begin{center} \includegraphics[scale=0.52]{Bulk_pi_N.eps} \caption{Pion self-energy diagram with mesonic loops (a) and baryonic loops [(b) and (c) are direct and cross diagrams] and nucleon self-energy diagram (d).} \label{Bulk_pi_N} \end{center} \end{figure} Next, let us come to the important quantities $\Gamma_\pi$ and $\Gamma_N$ of Eq.~(\ref{zeta_pi}) and (\ref{zeta_N}), which are called thermal widths of pion and nucleon respectively. During propagation in the medium, pion and nucleon may go through different on-shell scattering with other mesonic ($M$) and baryonic ($B$) resonances, which can be quantified by their different possible self-energy diagrams. From the imaginary part of their self-energy functions, their respective thermal widths $\Gamma_\pi$ and $\Gamma_N$ can be found. Fig.~\ref{Bulk_pi_N}(a) represents pion self-energy with internal lines of pion ($\pi$) and other mesonic resonances ($M$), which we can shortly call $\pi M$ loop. We will take $M=\sigma$ and $\rho$, as they are dominant resonances of $\pi\pi$ decay channel (within the invariant mass range of 1 GeV). Now, from the retarded self-energy of pion for $\pi M$ loop $\Pi^R_{\pi(\pi M)}(k)$, the corresponding thermal width $\Gamma_{\pi(\pi M)}$ can be obtained as \begin{equation} \Gamma_{\pi(\pi M)}=-{\rm Im}{\Pi}^R_{\pi(\pi M)}(k_0=\omega_\pi,\vec k)/m_\pi ~, \label{pipiM} \end{equation} where subscript notation stands for external (outside the bracket) and internal (inside the bracket) particles for the diagram~\ref{Bulk_pi_N}(a). Following Similar notation, we can define \begin{equation} \Gamma_{\pi(NB)}=-{\rm Im}{\Pi}^R_{\pi(NB)}(k_0=\omega^\pi_k,\vec k)/m_\pi ~, \label{piNB} \end{equation} where intermediate states of pion self-energy are nucleon $N$ and other baryonic resonance $B$ as shown in Fig.~\ref{Bulk_pi_N}(b) along with its cross diagram (c). As a dominant 4-star baryons with spin $J_B=1/2$ and $3/2$, we have taken $B =\Delta(1232)$, $N^*(1440)$, $N^*(1520)$, $N^*(1535)$, $\Delta^*(1600)$, $\Delta^*(1620)$, $N^*(1650)$, $\Delta^*(1700)$, $N^*(1700)$, $N^*(1710)$ and $N^*(1720)$. Adding all these mesonic ($\pi M$) and baryonic ($NB$) loops, the total thermal width of pion $\Gamma_\pi$ can be obtained as \begin{equation} \Gamma_\pi= \Gamma^M_\pi + \Gamma^B_\pi = \sum_M\Gamma_{\pi(\pi M)} + \sum_B\Gamma_{\pi(NB)}~. \label{pi_piM_NB} \end{equation} Similarly, one-loop self-energy of nucleon with pion ($\pi$) and baryon ($B$) intermediate states, which is denoted as $\Sigma^R_{N(\pi B)}$ (retarded part), will be our matter of interest to estimate corresponding nucleon thermal width $\Gamma_{N(\pi B)}$. The diagramatic anatomy of $\Sigma^R_{N(\pi B)}$ is shown in Fig.~\ref{Bulk_pi_N}(d). Here we have taken all the 4-star spin $1/2$ and $3/2$ baryons, mentioned above. Hence, summing all the $\pi B$ loops, we can get our total nucleon thermal width : \begin{equation} \Gamma_N=\sum_B\Gamma_{N(\pi B)}=-\sum_B{\rm Im}\Sigma^R_{N(\pi B)}(k_0=\omega_N,\vec k)~. \label{Gam_N} \end{equation} The imaginary part of self-energies, given in Eqs~(\ref{pipiM}), (\ref{piNB}) and (\ref{Gam_N}), have been derived with help of standard thermal field theoretical techniques. At first, the expression for Im$\Pi^R_{\pi(\pi M)}$ is~\cite{GKS} \begin{eqnarray} {\rm Im}\Pi^R_{\pi(\pi M)}(k_0=\omega_\pi, \vec k) &=& \int \frac{d^3 \vec l}{32\pi^2 \omega_l\omega_u} \nonumber\\ &&L_{\pi\pi M}(k,l)|_{(l_0=-\omega_l,~k_0=\omega_k)} \nonumber\\ &&\left(n_l - n_u\right)~\delta(\omega_\pi +\omega_l - \omega_u) ~, \nonumber\\ \label{G_pi_piM} \end{eqnarray} where $n_l$, $n_u$ are BE distribution functions of $\pi$, $M$ mesons respectively at energies $\omega_l=\{\vec l^2 +m_\pi^2\}^{1/2}$ and $\omega_u=\{(\vec k - \vec l)^2 + m_M^2\}^{1/2}$. The vertex factors $L_{\pi(\pi M)}(k,l)$~\cite{GKS} have been calculated by using the effective Lagrangian density, \begin{equation} {\cal L}_{\pi\pi M} = g_\rho \, {\vec \rho}_\mu \cdot {\vec \pi} \times \partial^\mu {\vec \pi} + \frac{g_\sigma}{2} m_\sigma {\vec \pi}\cdot {\vec\pi}\,\sigma~, \label{Lag_pipiM} \end{equation} where $g_\rho$ and $g_\sigma$ are respectively effective coupling constants of $\rho$ meson field $({\vec \rho}_\mu)$ and $\sigma$ meson field ($\sigma$), which are coupled with the pion field (${\vec \pi}$). Next, the direct and cross diagrams of pion self-energy for $NB$ loop are combinedly expressed as~\cite{G_pi_JPG,G_eta_BJP} \begin{eqnarray} {\rm Im}\Pi^R_{\pi(NB)}(k_0=\omega_\pi, \vec k) &=& \int \frac{d^3 \vec l}{32\pi^2 \omega_l\omega_u} \nonumber\\ &&L_{\pi NB}(k,l)|_{(l_0=-\omega_l,~k_0=\omega_k)} \nonumber\\ &&\{\left(-n^+_l + n^+_u\right)~\delta(\omega_\pi +\omega_l - \omega_u) \nonumber\\ &&+\left(-n^-_l + n^-_u\right)~\delta(\omega_\pi -\omega_l + \omega_u)\} ~, \nonumber\\ \label{G_pi_NB} \end{eqnarray} where $n^{\pm}_l$, $n^{\pm}_u$ are FD distribution functions of $N$ and $B$ ($\pm$ for particle and anti-particle) respectively at energies $\omega_l=\{\vec l^2 +m_N^2\}^{1/2}$ and $\omega_u=\{(\vec k \pm \vec l)^2 + m_B^2\}^{1/2}$ ($\pm$ for diagrams (b) and (c) respectively). With the help of the effective Lagrangian densities for $\pi NB$ interactions~\cite{Leopold}, \begin{eqnarray} {\cal L}_{\pi NB}&=&\frac{f}{m_\pi}{\overline \psi}_B\gamma^\mu \left\{ \begin{array}{c} i\gamma^5 \\ 1\!\!1 \end{array} \right\} \psi_N\partial_\mu\pi + {\rm h.c.}~{\rm for}~J_B^P=\frac{1}{2}^{\pm}, \nonumber\\ &=&\frac{f}{m_\pi}{\overline \psi}^\mu_B \left\{ \begin{array}{c} 1\!\!1 \\ i\gamma^5 \end{array} \right\} \psi_N\partial_\mu\pi + {\rm h.c.}~{\rm for}~J_B^P=\frac{3}{2}^{\pm}~, \nonumber\\ &&(P~{\rm stands~for~parity~of~}B) \label{Lag_BNpi} \end{eqnarray} one can deduced the vertex factors $L_{\pi NB}(k,l)$~\cite{G_pi_JPG,G_eta_BJP}. At last, the expression for Im$\Pi^R_{N(\pi B)}$ is~\cite{G_N,G_NNst_BJP} \begin{eqnarray} {\rm Im}\Pi^R_{N(\pi B)}(k_0=\omega_\pi, \vec k) &=& \int \frac{d^3 \vec l}{32\pi^2 \omega_l\omega_u} \nonumber\\ &&L_{N\pi B}(k,l)|_{(l_0=-\omega_l,~k_0=\omega_k)} \nonumber\\ &&\left(n_l + n^+_u\right)~\delta(\omega_\pi +\omega_l - \omega_u) ~, \nonumber\\ \label{G_N_piB} \end{eqnarray} where $n_l$ is BE distribution functions of $\pi$ at energy $\omega_l=\{\vec l^2 +m_\pi^2\}^{1/2}$ and $n^+_u$ is FD distribution of $B$ at energy $\omega_u=\{(\vec k - \vec l)^2 + m_M^2\}^{1/2}$. With the help of the interaction Lagrangian densities from Eq.~(\ref{Lag_BNpi}), the vertex factors $L_{N\pi B}(k,l)$~\cite{G_N} have been obtained. \section{Results and Discussion} \label{sec:num} \begin{figure} \begin{center} \includegraphics[scale=0.35]{gm_k.eps} \caption{Momentum distribution of pion thermal width for mesonic (dash-dotted line), baryonic loops (dotted line) and their total (solid line) and nucleon thermal width (dashed line) at three different medium parameters: (a) $(T,\mu_B)=(0.130$ GeV, $0)$, (b) $(0.170$ GeV, $0)$ and (c) $(0.130$ GeV, $0.300$ GeV).} \label{gm_k} \end{center} \end{figure} Let us start our numerical discussion with the Fig.~(\ref{gm_k}), where momentum distribution of thermal widths of pion and nucleon have been displayed. With the help of Eqs.~(\ref{pipiM}), (\ref{piNB}), (\ref{pi_piM_NB}), (\ref{G_pi_piM}) and (\ref{G_pi_NB}), $\Gamma_\pi^M$, $\Gamma_\pi^B$ and their total $\Gamma_\pi$ can be found whose momentum distributions are respectively shown by dash-dotted, dotted and solid line in the Fig.~(\ref{gm_k}). Similarly, $\Gamma_N$ can be deduced by using Eqs.~(\ref{G_N_piB}) and (\ref{Gam_N}) and its momentum distribution is represented by dash line. Panels (a), (b) and (c) of Fig.~(\ref{gm_k}) are for different set of temperature $T$ and baryon chemical potential $\mu_B$ of the medium. Though $\Gamma_N$ is approximately constant with nucleon momentum, but $\Gamma_\pi^M$ and $\Gamma_\pi^B$ exhibit a peak structure in some point of $\vec k$-axis, which depends on the medium parameters $T$ and $\mu_B$. These momentum distribution of $\Gamma_\pi$ and $\Gamma_N$ will be integrated out when we will estimate $\zeta_\pi$ and $\zeta_N$ from Eqs.(\ref{zeta_pi}) and (\ref{zeta_N}) respectively. \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_T_ppM.eps} \caption{$\zeta(T)$ due to pion thermal width for $\pi\sigma$ (dotted line), $\pi\rho$ (dashed line) loops and their total (solid line) at $c_S^2=0$ (a), $c_S^2=1/3$ (b) and $c_S^2(T)$ from HRG (c).} \label{z_T_ppM} \end{center} \end{figure} \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_T_mu0.eps} \caption{(a): Temperature dependence of bulk viscosity for pion thermal width with mesonic loops (dash-dotted line), meson + baryon loops (dotted line), for nucleon thermal width (dashed line) and their total $\zeta_T=\zeta_\pi +\zeta_N$ (solid line). (b): $\zeta(T)$ for $c_S^2(T)$ from HRG and two constant values of $c_S^2$ ($c_S^2=0.15$: dotted line and $c_S^2=0.25$: dash line).} \label{z_T_mu0} \end{center} \end{figure} \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_Tmu.eps} \caption{$\zeta(T)$ due to nucleon thermal width (a), pion thermal width for meson loops (b) and meson + baryon loops (c) at $\mu_B=0.250$ GeV (dotted line) and $0.500$ GeV (dash line).} \label{z_Tmu} \end{center} \end{figure} Let us come to the different loop contributions of pion and nucleon thermal width in bulk viscosity coefficient of hadronic matter. Fig.~\ref{z_T_ppM}(c) shows individual contributions of $\pi\sigma$ (dotted line) and $\pi\rho$ (dash line) loops in $\zeta_\pi$, which reveals that they are respectively important in low ($T<0.080$ GeV) and high ($T>0.080$ GeV) temperature domain for getting a non-divergent values of $\zeta_\pi$. These are respectively obtained by putting $\Gamma_{\pi(\pi \sigma)}$ and $\Gamma_{\pi(\pi \rho)}$ in place of $\Gamma_\pi$ of Eq.~(\ref{zeta_pi}). Putting $\Gamma_\pi^M=\Gamma_{\pi(\pi \sigma)}+\Gamma_{\pi(\pi \rho)}$ in place of $\Gamma_\pi$ of Eq.~(\ref{zeta_pi}), we get the solid line, representing total bulk viscosity of pionic component due to meson loops. After a mild decrement in low $T$ ($<0.080$ GeV), it receives an increment nature in high $T$ ($>0.080$ GeV). Along with Fig.~\ref{z_T_ppM}(c), where an explicit temperature dependent $c_s^2$ is taken from HRG model, the results for $c_s^2=0$ and $c_s^2=1/3$ are also displayed in Fig.~\ref{z_T_ppM}(a) and (b), which are little different in nature. Just to show the phase space sensitivity of bulk viscosity via $c_s^2$, these two results are displaying two extreme limits of $c_s^2$. Therefore, we can understand Fig.~\ref{z_T_ppM}(c) as some sort of superposition of \ref{z_T_ppM}(a) and (b). \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_mu_piN.eps} \caption{Same as Fig.~(\ref{z_Tmu}) along $\mu_B$-axis at $T=0.050$ GeV (dotted line), $0.100$ GeV (dash line) and $0.150$ GeV (solid line).} \label{z_mu_piN} \end{center} \end{figure} \begin{figure} \begin{center} \includegraphics[scale=0.35]{cs2_Tmu.eps} \caption{(a): $c_S^2(T)$ at $\mu_B=0$ (solid line), $0.250$ GeV (dotted line) and $0.500$ GeV (dash line), and LQCD results of $c_s^2(T,\mu_B=0)$ (circles)~\cite{Lat1}; (b) : $\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon}$ vs $T$ at $\mu_B=0.250$ GeV (dotted line) and $0.500$ GeV (dash line); (c): $c_S^2(\mu_B)$ at $T=0.050$ GeV (dotted line), $0.100$ GeV (dash line) and $0.150$ GeV (solid line); (d) : $\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon}$ vs $\mu_B$ at $T=0.050$ GeV (dotted line), $0.100$ GeV (dash line) and $0.150$ GeV (solid line). } \label{cs2_Tmu} \end{center} \end{figure} \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_s_Tmu.eps} \caption{$T$ dependence of total bulk viscosities (a), entropy densities from HRG (b) and their ratios $\zeta/s$ (c) at $\mu_B=0$ (solid line), $0.250$ GeV (dotted line) and $0.500$ GeV (dash line).} \label{z_s_Tmu} \end{center} \end{figure} \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_s_mu.eps} \caption{$\mu_B$ dependence of total bulk viscosities (a), entropy densities from HRG (b) and their ratios $\zeta/s$ (c) at $T=0.050$ GeV (dotted line), $0.100$ GeV (dash line) and $0.150$ GeV (solid line).} \label{z_s_mu} \end{center} \end{figure} \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_root_s2.eps} \caption{Center of mass energy ($\sqrt{s}$) dependence of total bulk viscosity (a), entropy density from HRG (b) and their ratio $\zeta/s$ (c).} \label{z_root_s} \end{center} \end{figure} According to Eq.~(\ref{pi_piM_NB}) different baryon loops contribution ($\Gamma_\pi^B$) should have to add with meson loops contribution ($\Gamma_\pi^M$) to get total pion thermal width $\Gamma_\pi$. In Fig.~\ref{z_T_mu0}(a), changing the nature of dash-dotted line to dotted line indicates that inclusion of baryon loops with meson loops becomes the reason for reducing the rate of increment of $\zeta_\pi(T)$ at high temperature region, $T>0.100$ GeV. Putting our calculated nucleon thermal width $\Gamma_N$ in Eq.~(\ref{zeta_N}), we get $\zeta_N$ as shown by dash line in Fig.~\ref{z_T_mu0}(a). Now adding $\zeta_N$ with $\zeta_\pi$ we have total bulk viscosity \begin{equation} \zeta_T=\zeta_\pi + \zeta_N~, \end{equation} as shown by solid line in Fig.~\ref{z_T_mu0}(a). In Fig.~\ref{z_T_mu0}(b), this $\zeta_T$ (solid line) has been compared with the results generated for two constant values of $c_s^2$ ($c_s^2=0.25$: dash line and $c_s^2=0.15$: dotted line), within which $c_s^2(T, \mu_B=0)$ from HRG model more or less varies. At two different values of $\mu_B$, $\zeta(T)$ due to nucleon thermal width ($\Gamma_N$), pion thermal width for meson loops ($\Gamma_\pi^M$) and meson + baryon loops ($\Gamma_\pi$) are shown in Fig.~\ref{z_Tmu}(a), (b) and (c) respectively. Similarly, Fig.~\ref{z_mu_piN}(a), (b) and (c) are displaying different loop contributions in $\zeta(\mu_B)$ at $T=0.050$ GeV (dotted line), $0.100$ GeV (dashed line) and $0.150$ GeV (solid line). From Fig.~\ref{z_Tmu}(a) and \ref{z_mu_piN}(a), we see that $\zeta_N$ increases with $T$ as well as $\mu_B$. From Fig.~\ref{z_Tmu}(b), we see the $\zeta_\pi$ due to $\Gamma_\pi^M$ at finite $\mu_B$ first decreases at low $T$ then increases at high $T$. The nature of these curves are quite similar to the curve of $\zeta_\pi(T)$ at vanishing $\mu_B$ but their minima are only shifted towards lower $T$ as $\mu_B$ increases. Following the same story of vanishing $\mu_B$, inclusion of baryon loops in pion self-energy is again influencing on $\zeta_\pi(T)$ in high temperature domain. The variation with $\mu_B$ of $\zeta_N(\mu_B)$ in Fig.~\ref{z_mu_piN}(a) and $\zeta_\pi(\mu_B)$ in Fig.~\ref{z_mu_piN}(b) and (c) are grossly same as their temperature dependence. For small $T$ and $\mu_B$, $\zeta_N$ and $\zeta_\pi$ are of similar order. However, with increasing $T$ and $\mu_B$, $\zeta_N$ dominates over $\zeta_\pi$. $\zeta_N$ receives additional contribution from $\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon}$. One should keep in mind that the term $\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon}$ goes to zero for $\mu_B=0$. The $T$ and $\mu_B$ dependence of $\left(\frac{\partial P}{\partial \rho_B}\right)_{\epsilon}$ are shown in Fig.~\ref{cs2_Tmu}(b) and (d) respectively while Fig.~\ref{cs2_Tmu}(a) and (c) are displaying the $T$ and $\mu_B$ dependence of $c_s^2$. From Fig.~\ref{cs2_Tmu}(a), we see that our $c_s^2(T, \mu_B=0)$ curve (solid line) is in good agreement with LQCD results~\cite{Lat1} (circles) within the hadronic temperature domain ($T < 0.160$ GeV). Total bulk viscosity $\zeta_T$ (a), entropy density $s$ (b) and their ratio $\zeta/s$ (c) are plotted against $T$ in Fig.~(\ref{z_s_Tmu}) and $\mu_B$ in Fig.~(\ref{z_s_mu}) at three different values $\mu_B$ and $T$ respectively. Since increment of $s(T)$ is larger than the increment of $\zeta(T)$, therefore, $\zeta/s$ is appeared as a decreasing function of $T$. On the other hand, both $\zeta(\mu_B)$ and $s(\mu_B)$ monotonically increase with $\mu_B$ but the ratio $\zeta/s(\mu_B)$ increases first and then decreases at high $\mu_B$ domain. Next, Fig.~\ref{z_root_s}(a), (b) and (c) reveal respectively the variation of total bulk viscosity $\zeta$, entropy density $s$ and their ratio with the variation of center of mass energy $\sqrt{s}$ (Reader are requested to be careful on the same symbol $s$ used for entropy density and square of beam energy). The beam energy dependence of $T$ and $\mu_B$ used in computation are those obtained from fits to hadron yields. We have used the parameterization from Ref.~\cite{HRGKarsch}. We notice in Fig.~\ref{z_root_s} that $\zeta$ (a) as well as $\zeta/s$ (c) are decreasing with $\sqrt{s}$, which is qualitatively agreeing with the results of earlier studies~\cite{Kadam1,Kadam2}. The decreasing trend of $\zeta$ and $\zeta/s$ with $\sqrt{s}$ can be understood from the fact that $\mu_B$ decreases with $\sqrt{s}$ while $T$ remains fairly constant in the range of $\sqrt{s}$ analyzed here and according to Fig.~\ref{z_s_mu}(a) and (c), the $\zeta$ and $\zeta/s$ decreases with decreasing of $\mu_B$. \begin{figure} \begin{center} \includegraphics[scale=0.35]{z_s_TComp.eps} \caption{(Color online) Our results of $\zeta$ (a) and $\zeta/s$ (b) vs $T$ at $\mu_B=0$ are compared with the earlier results of Sasaki et al. (Green triangles down~\cite{Sasaki}), Deb et al. (Pink solid squares~\cite{Kadam3}), Chakraborty et al. (Brown stars~\cite{Purnendu}), Marty et al. (open circles~\cite{Cassing}), Kadam et al. (Violet pluses~\cite{Kadam2}), Fraile et al. (Blue solid circles), Hostler et al. (Open squares~\cite{Noronha})} \label{z_s_TComp} \end{center} \end{figure} Fig.~\ref{z_s_TComp} is dedicated for comparative understanding of our results with respect to the earlier investigations. As most of the works have been done at $\mu_B=0$, so we have plotted $\zeta$ (a) and $\zeta/s$ (b) against $T$ for $\mu_B=0$, where our results for $\pi$- component (red lines) and $(\pi + N)$- components (black lines), using our calculated $\tau(\vec k, T, \mu_B=0)$ (dashed lines) and constant $\tau$ (solid lines), are compared with the results, obtained by Sasaki et al. (Green triangles down~\cite{Sasaki}), Deb et al. (Pink solid squares~\cite{Kadam3}), Chakraborty et al. (Brown stars~\cite{Purnendu}), Marty et al. (open circles~\cite{Cassing}), Kadam et al. (Violet pluses~\cite{Kadam2}), Fraile et al. (Blue solid circles), Hostler et al. (Open squares~\cite{Noronha}). We see a large numerical band for $\zeta$ ($10^{-5}$-$10^{-2}$ GeV$^3$) or $\zeta/s$ ($10^{-3}$-$10^{0}$), within which earlier estimations are located. The results of the present work and Fraile et al.~\cite{Nicola} both show similar kind of temperature dependence of $\zeta$ - it decreases at low $T$ domain ($<0.100$ GeV) and then increases at high $T$ domain ($>0.100$ GeV). Monotonically increasing nature of $\zeta(T)$ for constant value of $\tau$ (solid lines) discloses the fact that the origin of non-monotonic behavior of dashed lines are because of explicit structure of $\tau(\vec k,T,\mu_B=0)$. The $\zeta(T)$ of Ref.~\cite{Kadam2} decreases up to $T\sim 0.150$ GeV after which a mild increment is observed. Most of the earlier works~\cite{Sasaki,Cassing,Kadam3,Dobado,Purnendu,Sarkar,Kadam2,Sarwar,Noronha,Nicola} based on effective QCD model calculations~\cite{Sasaki,Cassing,Kadam3,Dobado,Purnendu} as well as effective hadronic model calculations~\cite{Sarkar,Kadam2,Sarwar,Noronha,Nicola} predicted a decreasing function of $\zeta/s (T)$ in the hadronic temperature domain, which is qualitatively similar with our results (dashed lines). These are not supporting the fact that $\zeta/s$ diverges or becomes large near the transition temperature as indicated by Refs.~\cite{LQCD_zeta1,LQCD_zeta2,Tuchin}, within the temperature domain of quark phase. Some of the effective QCD model calculations~\cite{Defu,G_IFT,Kinkar,Dobado}, which can predict estimations of $\zeta/s$ in both temperature domain, exposed a peak structure near the transition temperature. While some of the HRG model calculations~\cite{Kadam1,Noronha} have supported this behavior by displaying an increasing tendency of $\zeta/s(T)$ as one goes towards the transition temperature from the hadronic temperature domain. This kind of increasing $\zeta/s(T)$ is also observed in our work when we consider the constant value of $\tau$ (solid lines). Regarding this two opposite nature of $\zeta/s(T,\mu_B=0)$ within hadronic temperature domain, Ref.~\cite{Noronha,Dobado} have exposed the possibility of both nature. Ref.~\cite{Noronha} shows that inclusion Hagedorn states (HS) in HRG model can convert $\zeta/s(T)$ from decreasing to increasing function. In this context, our results for explicit $T$, $\mu_B$ dependent $\tau$ and constant value of $\tau$ are also displaying both type of nature. Taking shear viscosity $\eta(T, \mu_B=0)$ from Ref.~\cite{G_eta_BJP}, based on same pion and nucleon thermal fluctuations, we get $\zeta/\{(1/3-c_S^2)^2\eta\}\approx 5-4$ and $\zeta/\{(1/3-c_S^2)\eta\}\approx 0.8-0.7$. This is supporting the estimation of gravity dual theory~\cite{Gravity} instead of the relation $\zeta/\{(1/3-c_S^2)^2\eta\}\approx 15$, followed by photon fields~\cite{Weinberg}, scalar fields~\cite{scalar} or QCD theory~\cite{Arnold}. So our estimation within the hadronic temperature domain is representing the strongly coupled picture instead of weakly coupled scenario~\cite{Arnold}. Again, at high temperature domain, our numerical values of $\zeta/s$ are matching (after extrapolation) with high temperature values of Refs.~\cite{Arnold,Kapusta2}- $\zeta/s(T\approx 0.200-0.400)\approx0.002-0.001$, obtained from the perturbative QCD calculations. In this regard, our estimation is indicating a smooth transformation from the strongly coupled picture of the hadronic temperature domain to a weakly coupled medium of quarks, instead of divergence or peak structure of $\zeta/s$ near transition temperature. \section{Summary} \label{sec:concl} We have gone through a detailed microscopic calculation of bulk viscosity coefficient for hadronic matter, where thermodynamical equilibrium conditions of all hadrons in medium have been treated by standard HRG model, which is very successful to generate LQCD thermodynamics up to the transition temperature. The thermal widths of medium constituents in the bulk viscosity expression inversely determine their numerical strength. Assuming pions and nucleons as most abundant medium constituents, we have concentrated on the bulk viscosity contributions from pion and nucleon components, where their corresponding thermal widths are derived from their in-medium scattering probabilities with dif- ferent mesonic and baryonic resonances in the hadronic matter. Owing to the field theory version of optical theorem, the imaginary part of pion and nucleon self-energy (on-shell) at finite temperature give the estimation of their corresponding thermal widths. In the one-loop diagrams of pion self-energy, we have taken different mesonic and baryonic loops, while pion-baryon intermediate states are considered in the one-loop diagrams of nucleon self-energy. Their thermal widths are basically on-shell values of their corresponding Landau cut contributions, which disappear in the absence of medium and therefore, these are inversely interpreted as their respective relaxation time, which proportionally control the numerical strength of transport coefficients like $\zeta$. Our result show that $\zeta(T)$ at $\mu_B=0$ increases in the high temperature domain ($0.080 <T (\rm{GeV})< 0.175$) but a decreasing nature of $\zeta(T)$ has also been observed at low $T$ ($<0.08$ GeV). The $\pi\sigma$ and $\pi\rho$ loops of pion self-energy are respectively responsible for the decreasing and increasing nature of $\zeta(T)$ at low and high $T$ domain. Addition of baryon loops in pion self-energy mainly make $\zeta(T)$ reduce at high $T$ domain. Bulk viscosity for nucleon component monotonically increases with $T$. At finite $\mu_B$, the nucleon component of bulk viscosity is highly dominating over the pion component. Adding nucleon and pion components, the total $\zeta$ increases with both $T$ and $\mu_B$. However, after dividing by total entropy density, $\zeta/s$ appear as a decreasing function of $T$ and with the variation of $\mu_B$, it increases first at low $\mu_B$ region and then decreases at high $\mu_B$ region. Along the beam energy axis, the $\zeta$ and $\zeta/s$ both decreases, as noticed in some earlier works~\cite{Kadam1,Kadam2,Sarwar}. During comparison with earlier results of $\zeta/s(T)$ at $\mu_B=0$, one can notice that the qualitative as well as quantitative nature is not a very settled issue. Some of them~\cite{LQCD_zeta1,LQCD_zeta2,Tuchin} indicated divergence tendency of $\zeta/s$ near transition temperature, some of effective QCD model calculations~\cite{Defu,G_IFT,Kinkar,Dobado} revealed peak structure near transition temperature, whereas most of the effective QCD model calculations~\cite{Sasaki,Cassing,Kadam3,Dobado,Purnendu} as well as effective hadronic model calculations~\cite{Sarkar,Kadam2,Sarwar,Noronha,Nicola}, including our present work, predict a decreasing function of $\zeta/s (T)$ in the hadronic temperature domain, with few exceptional HRG calculations~\cite{Kadam1,Noronha}. Our decreasing $\zeta/s (T,\mu_B=0)$ is representing a strongly coupled picture in the hadronic temperature domain, whose smooth extrapolation to high temperature domain agrees with a weakly coupled picture~\cite{Arnold}. {\bf Acknowledgment :} During first and major part of this work, SG is financially supported by the DST project with no NISER/R$\&$D-34/DST/PH1002, (with title {\it ``Study of QCD phase Structure through high energy heavy ion collisions''} and principal investigator Prof. B. Mohanty). During the last part of the work, SG is supported from UGC Dr. D. S. Kothari Post Doctoral Fellowship under grant No. F.4-2/2006 (BSR)/PH/15-16/0060. SC acknowledges XIIth plan project no. 12-R$\&D$-NIS-5.11-0300 and CNT project PIC XII-R$\&$D-VECC-5.02.0500 for support. SG thanks to high energy group of NISER (Prof. B. Mohanty, Dr. A. Das, Dr. C. Jena, Dr. R. Singh, R. Haque, V. Bairathi, K. Nayak, V. Lyer, S. Kundu and others) and group of Calcutta University (Prof. A. Bhattacharyya, Prof. G. Gangopadhyay) for getting various academic and non-academic support at NISER and CU during this work and also to Dr. V. Roy and Prof. H. Mishra for some discussion regarding this work.
0096a890f6b8d5115be38da0f72ecb222b6453b3
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Observations have established the presence of magnetic field of various magnitudes and on various spatial scales in our universe. Galaxies such as Milky Wave contain regular magnetic fields of the order of $\mu$G, while coherent fields of the order of $100~\mu$G are detected in distant galaxies \cite{Bernet:2008qp, Wolfe:2008nk}. There is a strong evidence for the presence of magnetic field in intergalactic medium, including voids \cite{Tavecchio:2010mk, Ando:2010rb, Neronov:1900zz, Dolag:2010}, with strengths exceeding $\sim 10^{-15}$~G\@. This supports the idea of cosmological origin of magnetic fields, which are subsequently amplified in galaxies, probably by the dynamo mechanism (see reviews \cite{Widrow:2002ud, Kandus:2010nw, Durrer:2013pga, Subramanian:2015lua}). The origin of cosmological magnetic field is a problem yet to be solved, with several pos\-sible mechanisms under discussion. These can broadly be classified into inflationary and post-inflationary scenarios. Both types still face problems to overcome: inflationary magnetic fields are constrained to be rather weak, while those produced after inflation typically have too small coherence lengths (see \cite{Widrow:2002ud, Kandus:2010nw, Durrer:2013pga, Subramanian:2015lua} for a review of these mechanisms and assessment of these difficul\-ties). It should also be noted that generation of helical hypermagnetic field prior to the elec\-troweak phase transition may explain the observed baryon asymmetry of the universe \cite{Fujita:2016igl, Kamada:2016eeb}. One of the mechanisms of generation of cosmological magnetic fields which is currently under scrutiny is based on the Abelian anomaly of the electroweak interactions \cite{Joyce:1997uy, Frohlich:2000en, Frohlich:2002fg}. If the difference between the number densities of right-handed and left-handed charged fermions in the early hot universe happens to be non-zero (as in the leptogenesis scenario involving physics beyond the standard model; see \cite{Davidson:2008bu, Fong:2013wr} for reviews), then a specific instability arises with respect to generation of helical (hypercharge) magnetic field. The generated helical magnetic field, in turn, is capable of supporting the fermion chiral asymmetry, thus prolonging its own existence to cosmological temperatures as low as tens of MeV \cite{Boyarsky:2011uy}. In this process, magnetic-field power is permanently transferred from small to large spatial scales (the phenomenon known as `inverse cascade'). Further investigation of the general properties of the regime of inverse cascade revealed certain universal scaling laws in its late-time asymptotics \cite{Hirono:2015rla, Xia:2016any, Yamamoto:2016xtu}. In this paper, we study analytically the process of generation of helical magnetic field in the early hot universe by an unspecified external source of lepton chiral asymmetry. Helical magnetic field is produced due to the presence of thermal background, which we extrapolate to all spatial scales, including the super-horizon scales.\footnote{The spectral properties of magnetic fields on superhorizon spatial scales depend on a concrete model of generation of primordial magnetic fields (see \cite{Kandus:2010nw, Durrer:2013pga, Subramanian:2015lua} for recent reviews).} We consider a simple model of generation of magnetic field which assumes that the source of chiral anomaly maintains a constant value of the (conformal) chiral chemical potential of charged leptons. After generation of magnetic field of near maximal helicity, its evolution is traced in the absence of the external source of lepton chiral asymmetry. In this case, the helical magnetic field and the lepton chiral asymmetry are mutually sustained (decaying slowly) by quantum anomaly until temperatures of the order of tens of MeV, with magnetic-field power being permanently transferred from small to large spatial scales in the regime of inverse cascade. We obtain analytic expressions describing the evolution of the lepton chiral chemical potential and magnetic-field energy density. The evolution of both these quantities exhibits certain scaling behavior, effectively depending on a single combined variable. In this case, the late-time asymptotics of the chiral chemical potential reproduces the universal scaling law previously found in the literature for the system under investigation \cite{Hirono:2015rla, Xia:2016any, Yamamoto:2016xtu}. As the temperature drops down because of the cosmological expansion, the processes of lepton scattering with the change of chirality (the so-called chirality-flipping processes) start playing important role, eventually leading to a rapid decay of the lepton chiral asymmetry. We give an analytic expression for the temperature at which this happens, depending on the initially generated values of the magnetic-field energy density and lepton chiral asymmetry. \section{Helical magnetic fields} \label{sec:helical} A spatially flat expanding universe filled by relativistic matter is conveniently described in the comoving conformal coordinate system $(\eta, \bx)$ with the conformal time $\eta$ and scale factor $a (\eta)$ entering the metric line element as $ds^2 = a^2 (\eta) \left( d \eta^2 - d \bx^2 \right)$. By rescaling the conformal coordinates $(\eta, \bx)$, one can suitably normalize the scale factor $a (\eta)$. A divergence-free statistically homogeneous and isotropic cosmological magnetic field has the following general Fourier representation of the two-point correlation function\footnote{The quantities $B_i$ are the components of the so-called comoving magnetic field, which is related to the observable magnetic field strength $\bB_{\rm obs}$ by the equation $\bB = a^2 \bB_{\rm obs}$. The spatial vector indices are treated by using the Kronecker delta-symbol, and their position does not matter.} \cite{Caprini:2003vc}: \beq \label{correl} \ave{B_i^{} (\bk) B_j^* (\bk')} = \left( 2 \pi \right)^3 \delta \left(\bk - \bk' \right) \left[ P_{ij} (\bk) S(k) + i \epsilon_{ijs} \hat k_s A(k) \right] \, , \eeq where $\hat k_i = k_i / k$, $ P_{ij} = \delta_{ij} - \hat{k}_i \hat{k}_j$ is the symmetric projector to the plane orthogonal to $\bk$, and $ \epsilon_{ijk} $ is the normalized totally antisymmetric tensor. It is useful to introduce the helicity components $B_\pm (\bk)$ of the magnetic field via \beq B_i (\bk) = B_+ (\bk) \re^+_i (\bk) + B_- (\bk) \re^-_i (\bk) \, , \eeq where the complex basis $ \re^\pm_i = \frac{1}{\sqrt{2}}\left( \re^1_i \pm i \re^2_i \right)$ is formed from a right-handed (with respect to the orientation $\epsilon_{ijk}$) and orthonormal (with respect to the metric $\delta_{ij}$) basis ${\bf e}^{1} (\bk)$, ${\bf e}^{2} (\bk)$, ${\bf e}^{3} (\bk) = \bk /k$. The coefficients of the symmetric and antisymmetric parts of the correlation function are then expressed through these components as follows: \ber \label{S} \ave{B_-^{} (\bk) B_-^* (\bk') + B_+^{} (\bk) B_+^* (\bk')} &=& 2 \left( 2 \pi \right)^3 \delta \left(\bk - \bk' \right) S (k) \, , \\ \ave{B_-^{} (\bk) B_-^* (\bk') - B_+^{} (\bk) B_+^* (\bk')} &=& 2 \left( 2 \pi \right)^3 \delta \left(\bk - \bk' \right) A (k) \, . \label{A} \eer We note an obvious constraint $|A (k)| \leq S (k)$. The spectrum $A (k)$ of the magnetic-field correlation function characterizes the difference in the power between the left-handed and right-handed magnetic field, i.e., its helicity. The spectrum $S (k)$ characterizes the magnetic field energy density. In the case of so-called maximally helical magnetic field, one has $ |A(k)| = S(k)$, and magnetic field is dominated by its left-handed or right-handed part, depending on the sign of $A (k)$. In this paper, we consider the effects of Abelian anomaly in the presence of spatially homogeneous chiral asymmetry.\footnote{Effects of spatial inhomogeneity in chiral relativistic plasma were under investigation in \cite{Boyarsky:2015faa, Gorbar:2016qfh}.} In this case, the evolution of the comoving magnetic field in conformal coordinates in cosmic plasma with high conductivity $\sigma$ takes the form \beq \label{magev} \frac{\partial \bB}{\partial \eta} = \frac{1}{\sigma_c} \nabla^2 \bB - \frac{\alpha \Delta \mu}{\pi \sigma_c} \rot \bB \, , \eeq where $\Delta \mu \equiv a \left(\mu_\rL - \mu_\rR \right)$ is the spatially homogeneous difference between the (conformal) chemical potentials of the left-handed and right-handed charged leptons, $\sigma_c \equiv a \sigma \approx {\rm const}$ \cite{Turner:1987bw, Baym:1997gq} characterizes the plasma conductivity, and $\alpha \approx 1/137$ is the fine structure constant. The last term in equation (\ref{magev}) is connected with the anomalous current in Maxwell's equations \cite{Joyce:1997uy, Frohlich:2000en, Frohlich:2002fg, Vilenkin:1980fu, Redlich:1984md, Fukushima:2008xe, Tashiro:2012mf}. Using equations (\ref{correl}) and (\ref{magev}), one can obtain the following system of equations for the spectra $S (k, \eta)$ and $A (k, \eta)$ (see \cite{Boyarsky:2011uy}): \ber \frac{\partial S}{\partial \eta} &=& - \frac{2 k^2}{\sigma_c} (S - S_{\rm eq}) + \frac{2 \alpha k}{\pi \sigma_c} \Delta \mu A \, , \label{dotS} \\ \frac{\partial A}{\partial \eta} &=& - \frac{2 k^2}{\sigma_c} A + \frac{2 \alpha k}{\pi \sigma_c} \Delta \mu S \, . \label{dotA} \eer In equation (\ref{dotS}), we have added a term with the thermal equilibrium distribution \beq \label{Seq} S_{\rm eq} (k, \eta) = \frac{k }{e^{k/aT} - 1} \, , \eeq whose role is to ensure relaxation of the spectral energy distribution $S$ to its equilibrium value $S_{\rm eq}$ rather than to zero. This mechanism will not work in the long-wavelength domain $k \lesssim 1/\eta$, which is not causally connected in the expanding hot unverse. This, however, will be of no practical importance, since the anomalous term in equation (\ref{dotS}) will dominate in this spectral region. The initial spectra in the domain of small values of $k$ will also depend on their cosmological origin. We do not consider this issue in the present paper, assuming the initial spectrum to be given by (\ref{Seq}) on all scales. In an early radiation-dominated universe expanding adiabatically with the bulk matter in local thermal equilibrium, the entropy density $(a T)^3 g_*$ remains constant. Here, $g_*$ is the number of relativistic degrees of freedom $g_*$ in thermal equilibrium. In the range of temperatures $80~{\rm GeV} < T < 150~{\rm MeV}$, the value of $g_*$ changes insignificantly from about 86 to 72, and at the quantum-chromodynamical crossover, at $T \simeq 150$~MeV, drops to about 17. The quantity $g_*^{1/3}$ thus drops from about 4.4 to 2.6, and we can see that the product $a T$ remains constant to a great extent. It is then convenient to normalize the scale factor as the inverse of the temperature, $a = 1/ T$. With this choice, we have $a = \eta / M_*$, where $M_* = \left( 45 / 4 \pi^3 g_* \right)^{1/2} M_{\rm P} \simeq 10^{18}$~GeV is the effective Planck mass, and $\sigma_c = \sigma/ T \approx 70$ is constant in time \cite{Boyarsky:2011uy, Turner:1987bw, Baym:1997gq}. The equilibrium spectrum (\ref{Seq}) is independent of time and, with this normalization, reads \beq \label{S0} S_{\rm eq} (k, \eta) \equiv S_0 (k) = \frac{k}{e^k - 1} \, . \eeq The excess $\rho_B (\eta)$ of the energy density of magnetic field over the thermal energy density is then determined by \beq \label{rhoB} \rho_B (\eta) = \frac{T^4}{2 \pi^2} \int_0^\infty P (k, \eta) k^2 d k \, , \eeq where $P (k, \eta) \equiv S (k, \eta) - S_0 (k)$ is the excess over the thermal power spectrum. The system of equations (\ref{dotS}) and (\ref{dotA}) is supplemented by the evolution equation for the chiral chemical potential \cite{Boyarsky:2011uy}: \beq \label{muevo} \frac{d \Delta \mu (\eta)}{d \eta} = - \frac{c_\Delta \alpha}{\pi^2} \int_0^\infty \frac{\partial A (k, \eta)}{\partial \eta} k d k - \Gamma_{\rm f} (\eta) \Delta \mu (\eta) + \beta (\eta) \, . \eeq Here, $c_\Delta$ is a numerical constant of order unity (it would be equal to $3/4$ in pure quantum electrodynamics) that reflects the particle content of the primordial plasma, $\beta (\eta)$ is an external source for the evolution of chiral chemical potential which, for definiteness, we assume to be positive, and $\Gamma_{\rm f} (\eta)$ is the coefficient of the so-called `flipping' term which describes decay of lepton chiral asymmetry caused by chirality flips in the electroweak interactions. \section{Development of helical magnetic fields} Since the conformal conductivity $\sigma_c$ is constant during the period of evolution under investigation, it is more convenient to work in terms of a rescaled conformal time $\tau = \eta / \sigma_c$. With regard of (\ref{S0}), system (\ref{dotS}), (\ref{dotA}) can then be written in the form \ber \dot P &=& - 2 k^2 P + \frac{2 \alpha k}{\pi} \Delta \mu A \, , \label{dotP1} \\ \dot A &=& - 2 k^2 A + \frac{2 \alpha k}{\pi} \Delta \mu \left( P + S_0 \right) \, . \label{dotA1} \eer Here and in what follows, the overdot denotes the derivative with respect to $\tau$. We then set the initial conditions for system (\ref{muevo}), (\ref{dotP1}), (\ref{dotA1}) at the formal cosmological singularity $\tau = 0$ in the form \beq \label{inicon} P (k, 0) \equiv 0 \, , \qquad A (k , 0) \equiv 0 \, , \qquad \gamma (0) = 0 \, . \eeq Hence, we also have $\Delta \mu (0) = 0$. Thus, in the presence of lepton chiral asymmetry ($\Delta \mu \ne 0$), generation of the helicity spectrum $A (k)$ commences, according to (\ref{dotA1}), due to the presence of thermal distribution $S_0 (k)$. Solution of system (\ref{dotP1}) and (\ref{dotA1}) with respect to the spectral functions $P (k, \tau)$ and $A (k, \tau)$ with the initial conditions (\ref{inicon}) is given by \ber P (k, \tau) &=& \frac{2 \alpha k}{\pi} S_0 (k) e^{- 2 k^2 \tau} \int_0^\tau e^{2 k^2 \tau'} \sinh \left( 2 k \left[ \Psi (\tau) - \Psi (\tau') \right] \right) \Delta \mu (\tau') d \tau' \, , \label{solP} \\ A (k, \tau) &=& \frac{2 \alpha k}{\pi} S_0 (k) e^{- 2 k^2 \tau} \int_0^\tau e^{2 k^2 \tau'} \cosh \left( 2 k \left[ \Psi (\tau) - \Psi (\tau') \right] \right) \Delta \mu (\tau') d \tau' \, , \label{solA} \eer where \beq \label{AB} \Psi(\tau) = \frac{\alpha} {\pi} \int^{\tau}_{0} \Delta \mu (\tau') d \tau' \, . \eeq One can see that $A (k, \tau)$ has the same sign as $\Delta \mu (\tau)$, while $P (k, \tau)$ is always positive. \begin{figure}[htp] \centerline{\includegraphics[width=.8\textwidth]{spectrum.eps}} \caption{Spectrum (\ref{spec-P}) is plotted for $\Delta \mu = 5 \times 10^{-6}$ ($k_\mu \simeq 10^{-8}$) and at the cosmological plasma temperatures $T = 10$~GeV (dashed yellow) and $T = 1$~GeV (dashed green). Spectrum (\ref{spec-A}) is also plotted at $T = 10$~GeV (solid blue) and $T = 1$~GeV (solid pink). The spectra increase in magnitude with time. In the region $k / k_\mu \gg 1$, the spectra are saturated to their stationary values $A (k) \simeq S_0 (k) k_\mu /k$ and $P (k) \simeq S_0 (k) \left(k_\mu /k\right)^2$. \label{fig:spec}} \end{figure} To estimate the behavior of the spectral functions $P (k, \tau)$ and $A (k, \tau)$, let us evaluate them under the condition $\Delta \mu \equiv {\rm const} > 0$ [which can be ensured by an appropriate behavior of the source $\beta (\tau)$ in (\ref{muevo})]. Using (\ref{AB}) and taking the elementary integrals in (\ref{solP}) and (\ref{solA}), we have \ber \label{spec-P} P (k, \tau) &=& \frac{S_0 (k)}{(k / k_\mu)^2 - 1}\left(1 - e^{- 2 k^2 \tau} \left[ \cosh \left( 2 k_\mu k \tau \right) + \frac{k}{k_\mu} \sinh \left( 2 k_\mu k \tau \right) \right] \right) \, , \\ A (k, \tau) &=& \frac{S_0 (k)}{(k / k_\mu)^2 - 1}\left(\frac{k}{k_\mu} - e^{- 2 k^2 \tau} \left[ \sinh \left( 2 k_\mu k \tau \right) + \frac{k}{k_\mu} \cosh \left( 2 k_\mu k \tau \right) \right] \right) \, , \label{spec-A} \eer where \beq\label{kmu} k_\mu = \frac{\alpha \Delta \mu}{\pi} \, . \eeq Spectra (\ref{spec-P}) and (\ref{spec-A}) for $\Delta \mu = 5 \times 10^{-6}$ ($k_\mu \simeq 10^{-8}$) are plotted in figure~\ref{fig:spec} for temperatures $T = 10$~GeV and 1~GeV, corresponding to $\tau \approx 1.2 \times 10^{15}$ and $1.2 \times 10^{16}$, respectively. As can be seen from expressions (\ref{spec-P}) and (\ref{spec-A}) and from figure~\ref{fig:spec}, there arise two characteristic regions of wavenumbers: the region of relatively small $k$ (of order $k_\mu$), where the spectra keep growing and approach the property $A (k) \simeq P (k) \gtrsim S_0 (k)$ of maximal helicity, and the region of `tails' of these spectra, where they quickly reach the threshold values $A (k) \simeq S_0 (k) k_\mu /k$ and $P (k) \simeq S_0 (k) \left(k_\mu /k\right)^2$. Indeed, in the region of large wavenumbers $k \gg k_\mu$, equation (\ref{dotA1}) is approximated as \beq \dot A \approx - 2 k^2 A + \frac{2 \alpha k}{\pi} \Delta \mu S_0 = - 2 k^2 A + 2 k_\mu k S_0 \, , \eeq with the solution \beq A (k) = S_0 (k) \frac{k_\mu}{k} \left( 1 - e^{- 2 k^2 \tau} \right) \, , \eeq that exponentially with time approaches the equilibrium $A_{\rm eq} = S_0 k_\mu / k$. Solution of (\ref{dotP1}) is then given by \beq P (k) = S_0 (k) \left( \frac{k_\mu}{k} \right)^2 \left[ 1 - e^{- 2 k^2 \tau} \left( 1 + 2 k^2 \tau \right) \right] \, , \eeq with a rapid exponential convergence to the equilibrium $P_{\rm eq} = S_0 \left(k_\mu /k\right)^2$. For sufficiently slow evolution of $\Delta \mu (\tau)$, these expressions for the spectral `tails' will retain their forms, with $k_\mu (\tau)$ expressed through $\Delta \mu (\tau)$ by (\ref{kmu}). It should be noted that the magnetohydrodynamical description of cosmic plasma cannot be trusted in the domain of large physical wavenumbers $k / a \gtrsim \alpha T$, which, in our system of conformal units, corresponds to values of $k \gtrsim \alpha \sim 10^{-2}$. This limitation is insignificant for the evolution of magnetic instability developing on much larger spatial scales (as is usually the case in the scenarios under consideration). In the physically relevant domain $k \lesssim \alpha$, the thermal spectrum $S_0 (k) \approx 1$. Therefore, we can replace the factor $S_0 (k)$ by unity in all our equations; however, for the sake of better physical clarity, we will retain it. In the region of relatively small wavenumbers, the spectra $P (k, \tau)$ and $A (k, \tau)$ rapidly become equal to each other, as can be seen from figure~\ref{fig:spec}. In the regime $P \gg S_0$, in which the quantity $S_0$ on the right-hand side of equation (\ref{dotA1}) can be neglected, one can multiply equation (\ref{dotP1}) by $A (k, \tau)$, equation (\ref{dotA1}) by $P (k, \tau)$ and subtract them to obtain an equation relating the spectral functions: \beq \left( P^2 \right)^\cdot + 4 k^2 P^2 = \left( A^2 \right)^\cdot + 4 k^2 A^2 \, . \eeq This implies the relation \beq\label{hellim} P^2 (k , \tau) = A^2 (k, \tau) + f_0^2 (k) e^{- 4 k^2 \tau} \, , \eeq where $f_0^2(k)$ is an integration constant. We see that, if the quantity $A^2 (k, \tau)$ is not decaying or decaying slower than the last exponent in (\ref{hellim}), then a maximally helical state develops, with $P \simeq A$. In this regime, system (\ref{dotP1}), (\ref{dotA1}) reduces to a single equation for $A (k, \tau) \simeq P (k, \tau)$: \beq \label{helA} \dot A = \left( - 2 k^2 + \frac{2 \alpha k}{\pi} \Delta \mu \right) A \, . \eeq \section{Evolution of chiral asymmetry and magnetic field} Assuming that a maximally helical configuration quickly develops at some initial time $\tau_\ini$, we are going to establish how it will evolve together with $\Delta \mu (\tau)$ after the source $\beta (\tau)$ in (\ref{muevo}) is switched off. Let us make the notation $\Delta \mu_\ini = \Delta \mu (\tau_\ini)$, $P_\ini (k) = P (k, \tau_\ini)$, $A_\ini (k) = A (k, \tau_\ini)$, and introduce the momentum $k_\ini$ similarly to (\ref{kmu}): \beq\label{kini} k_\ini = \frac{\alpha \Delta \mu_\ini}{\pi} \, . \eeq The initial spectra can be presented as \beq \label{APini} P_\ini (k) = P_0 Z \left(\frac{k}{k_\ini} \right) \, , \qquad A_\ini (k) \approx P_\ini (k) \left( 1 + \frac{k}{k_\ini} \right) \, , \eeq where $Z (x) $ describes the shape of the spectrum, and the factor $(1 + k / k_\ini)$ is introduced to reflect the relation in the `tails' of the spectra. The normalization constant $P_0$ is chosen so that \beq \label{normal} \int_0^\infty Z (x) x^2 d x = 1 \, , \eeq and the initial excess (\ref{rhoB}) of the energy density of magnetic field over the thermal energy density is then equal to \beq \label{helrhoB} \rho_B (\tau_\ini) = \frac{T_\ini^4}{2 \pi^2} \int_0^\infty P (k, \tau_\ini) k^2 d k = \frac{T_\ini^4 k_\ini^3 P_0}{2 \pi^2} \, . \eeq It is convenient to relate this quantity to the total radiation energy density by introducing the dimensionless parameter \beq \label{rB} r_B^\ini \equiv \frac{\rho_B (\tau_\ini)}{\rho_r (\tau_\ini)} = \frac{30 \rho_B (\tau_\ini)}{\pi^2 g_* T_\ini^4} = \frac{15}{2 \pi^4 g_*} P_0 k_\ini^3 \, . \eeq Asymptotically, as $\tau \to \infty$, the maximum of spectrum (\ref{spec-P}) is reached at $k = k_\mu / 2$, and one can derive an approximate asymptotic estimate for $r_B^\ini$: \beq r_B^\ini \simeq \frac{k_\ini^2}{\pi^4} \sqrt{\frac{1}{2 \tau_\ini}}\, e^{k_\ini^2 \tau_\ini / 2} \, , \eeq where $k_\ini$ is given by (\ref{kini}). One can see the exponential dependence of $r_B^\ini$ on the (rescaled) conformal time $\tau_\ini$ (or temperature $T_\ini = M_* / \sigma_c \tau_\ini$) at which the spectrum (\ref{spec-P}) is finally developed by the external source of chiral asymmetry. In what follows, we take $r_B^\ini$ and $\Delta \mu_\ini$ [related to $k_\ini$ by (\ref{kini})] to be our independent parameters. The subsequent evolution of the spectrum in the domain where $P (k, \tau) \approx A (k, \tau) \gg S_0 (k)$ is described by equation (\ref{helA}). Its solution with the initial condition $A (k, \tau_\ini) = A_\ini (k)$ is given by \beq A (k, \tau) = g^2 (k, \tau) A_\ini (k) \, , \eeq where \beq \label{g} g (k, \tau) = e^{- k^2 \Delta \tau + k \Delta \Psi (\tau)} \, , \quad \Delta \tau = \tau - \tau_\ini \, , \quad \Delta \Psi (\tau) = \Psi (\tau) - \Psi (\tau_\ini) \, , \eeq is the growth factor. Solution of (\ref{muevo}) with the zero source $\beta$ and with the flipping term neglected can be written as \beq \label{differ} \Delta \mu_\ini - \Delta \mu (\tau) = \frac{c_\Delta \alpha}{\pi^2} \int_0^\infty \left[ A (k, \tau) - A_\ini (k) \right] k d k \, . \eeq The contribution of the `tail' in the distributions $A (k, \tau)$ and $A_\ini (k)$ to the value of the difference $\Delta \mu_\ini - \Delta \mu (\tau)$ in this expression is negligibly small. Indeed, the integral over the `tail' region is estimated as \ber \frac{c_\Delta \alpha}{\pi^2} \int_{k_{\rm tail}}^\infty \left[ A (k, \tau) - A_\ini (k) \right] k d k &\simeq& \frac{c_\Delta \alpha^2}{\pi^3} \left[ \Delta \mu (\tau) - \Delta \mu_\ini \right] \int_{k_{\rm tail}}^\infty S_0 (k) d k \nonumber \\ &\simeq& \frac{c_\Delta \alpha^2}{\pi^3} \left[ \Delta \mu (\tau) - \Delta \mu_\ini \right] \, , \eer which is much smaller by absolute value than the left-hand side of (\ref{differ}) because of the smallness of $\alpha \approx 1/137$. Thus, we can ignore the presence of power-law tails in the spectra in (\ref{differ}), and write, using (\ref{APini}), \beq \Delta \mu_\ini - \Delta \mu (\tau) = \frac{c_\Delta \alpha k_\ini^2 P_0}{\pi^2} \int_0^\infty \left[ g^2 (k_\ini x, \tau) - 1 \right] Z (x) x d x \, . \eeq Dividing this by $\Delta \mu_\ini$ and using (\ref{kini}) and (\ref{rB}), we obtain the estimate \beq \label{solmu} 1 - \frac{\Delta \mu (\tau)}{\Delta \mu_\ini} = \frac{\pi c_\Delta \alpha^2 g_* r_B^\ini}{15 k_\ini^2} \int_0^\infty \left[ g^2 (k_\ini x, \tau) - 1 \right] Z (x) x d x \, . \eeq For values in a broad typical range of parameters in different cosmological scenarios, the factor in front of the integral in (\ref{solmu}) is much larger than unity. For instance, for $g_* = 75$ and $\Delta \mu_\ini = 5 \times 10^{-6}$, this factor is estimated to be $\sim 10^{13} r_B^\ini$, and will be very large for $r_B^\ini \gg 10^{-13}$. Since the left-hand side of (\ref{solmu}) is always bounded by unity, this implies that the integral on the right-hand side should be extremely small. The relation \beq \label{fincon} \int_0^\infty \left[ g^2 (k_\ini x, \tau) - 1 \right] Z (x) x d x \approx 0 \eeq can then be regarded as an integral equation implicitly expressing the quantity to be found $\Delta \Psi$ through the known quantity $\Delta \Phi$ [both enter the function $g^2 (k_\ini x, \tau)$ under this integral; see (\ref{g})]. It is convenient to introduce the variables \beq \label{phi} \phi = 2 k_\ini^2 \Delta \tau \, , \qquad \psi = 2 k_\ini \Delta \Psi \, . \eeq In terms of these variables, we have \beq \label{g2} g^2 (k_\ini x, \tau) = e^{- \phi x^2 + \psi x} \, , \eeq and equation (\ref{fincon}) establishes the dependence $\psi (\phi)$, which is determined only by the form $Z (x)$ of the initial distribution. To find the dependence $ \psi (\phi)$, we differentiate (\ref{fincon}) with respect to $\phi$. We obtain the Cauchy problem \beq \label{dBA} \frac{d \psi}{d \phi} = \frac{\int_0^\infty e^{- \phi x^2 + \psi x} Z(x) x^3 d x}{\int_0^\infty e^{- \phi x^2 + \psi x} Z(x) x^2 d x} \, , \qquad \psi (0) = 0 \, . \eeq The evolution of the chiral chemical potential can then be calculated by using (\ref{AB}): \beq \label{muev} \frac{\Delta \mu }{\Delta \mu_\ini} = \frac{\pi}{\alpha \Delta \mu_\ini} \dot \Psi = \psi'(\phi) \, . \eeq Remarkably, the evolution of the chiral chemical potential $\Delta \mu$ depends on the initial conditions and on time through a single scaling parameter $\phi$, defined in (\ref{phi}). \begin{figure}[htp] \centerline{\includegraphics[width=.8\textwidth]{muevol.eps}} \caption{Evolution (\ref{muevol}) of the chiral chemical potential plotted in logarithmic scale. \label{fig:muevol}} \end{figure} As an example, let us approximate the initial spectrum $P_\ini (k)$ with subtracted high-frequency `tail' by a corresponding normalized spectral function with exponential cut-off:\footnote{Equation (\ref{zx}) gives a correct linear growth at small $x$, observed in figure~\ref{fig:spec}.} \beq \label{zx} Z (x) = 2 x e^{- x^2} \, . \eeq Then, introducing the variable \beq \label{zeta} \zeta = \frac{\psi}{\sqrt{1 + \phi}} \, , \eeq we can present problem (\ref{dBA}) in the form \beq \label{dzeta} \frac{d \zeta}{d \phi} = \frac{F (\zeta) - \frac12 \zeta}{1 + \phi} \, , \qquad \zeta (0) = 0 \, , \eeq where \beq \label{F} F (\zeta) = \frac{\int_0^\infty e^{- x^2 + \zeta x} x^4 d x}{\int_0^\infty e^{- x^2 + \zeta x} x^3 d x} \, . \eeq Equation (\ref{dzeta}) can, in principle, be integrated, and the function $\zeta (\phi)$ can be found. The evolution (\ref{muev}) of the chiral chemical potential is then given by \beq \label{muevol} \frac{\Delta \mu }{\Delta \mu_\ini} = \psi'(\phi) = \frac{F \left(\zeta (\phi) \right)}{\sqrt{1 + \phi}} \, , \eeq where $F (\zeta)$ is given by (\ref{F}). Solution (\ref{muevol}) is plotted in logarithmic scale in figure~\ref{fig:muevol}. Let us establish the late-time asymptotics of the solution to (\ref{dzeta}), (\ref{F}). In the regime $\zeta \gg 1$, we have $F (\zeta) \approx \zeta/2 + 3/\zeta + {\cal O} \left( \zeta^{-3} \right)$. Solution of (\ref{dzeta}) in this case behaves as \beq \label{large-a} \zeta (\phi) \approx \left[ {\rm const} + 6 \log (1 + \phi) \right]^{1/2} \, , \qquad \phi \gg 1 \, . \eeq This qualitative behavior does not depend on the specific shape (\ref{zx}) of the initial spectrum and is caused by the inverse cascade that transfers the spectral power to small-frequency region. Indeed, for large enough values of $\tau$, function (\ref{g2}) develops a strong Gaussian peak at small values of $x$, where $Z (x)$ behaves rather smoothly (typically, as a power of $x$). Expression (\ref{dBA}) does not then depend on the concrete form of $Z (x)$ in this limit. \begin{figure}[htp] \centerline{\includegraphics[width=.6\textwidth]{interpol.eps}} \caption{Exact solution of system (\ref{dzeta}), (\ref{F}) (solid blue line) versus interpolation (\ref{zeta1}) (dashed pink line). \label{fig:interpol}} \end{figure} With the account of asymptotics (\ref{large-a}), the solution $\zeta (\phi)$ of the differential equation (\ref{dzeta}), (\ref{F}) can be approximated by the expression \beq \label{zeta1} \zeta (\phi) = \left[ \sqrt{1 + 6 \log (1 + \phi)} - 1 \right] \, . \eeq Numerical integration confirms this approximation within about $1\%$ precision (see figure~\ref{fig:interpol}). Then \beq \label{interpol} \psi (\phi) = \sqrt{1 + \phi}\, \zeta (\phi) = \sqrt{1 + \phi} \left[ \sqrt{1 + 6 \log (1 + \phi)} - 1 \right] \, . \eeq Using (\ref{muevol}), we then obtain the universal late-time asymptotics \beq \label{muas} \log \frac{\Delta \mu }{\Delta \mu_\ini} \approx - \frac12 \log (1 + \phi) + \frac12 \log \log (1 + \phi) \, . \eeq This describes very well the almost ideal power-law behavior observed in figure~\ref{fig:muevol} for large values of $\phi$, with the asymptotic power index equal to $- 1/2$. The late-time asymptotics $\Delta \mu \propto \eta^{-1/2}$ in the system under consideration has been previously established in \cite{Hirono:2015rla} (and in \cite{Xia:2016any} with a leading logarithmic correction, $\Delta \mu \propto \eta^{-1/2} \log^{1/2} \eta$), and also noted in \cite{Yamamoto:2016xtu} in the context of chiral magnetohydrodynamics. Let us also determine the behavior of the magnetic-field energy density, described by the parameter $r_B (\tau)$. For our developed chiral distribution, we have $P (k, \tau) \approx g^2 (k, \tau) P_\ini (k)$. Hence, \beq \label{NB} r_B (\tau) = N_B^{} r_B^\ini \, , \qquad N_B (\phi) = \int_0^\infty Z (x) e^{- \phi x^2 + \psi (\phi) x} x^2 d x \, . \eeq Thus, the energy density of magnetic field also depends on time through a single scaling parameter $\phi$, defined in (\ref{phi}). Its behavior at large values of $\phi$ will depend only on the behavior of the function $Z (x)$ at small $x$. For the initial shape (\ref{zx}) of the magnetic-field spectrum, the scaling function $N_B (\phi)$ is determined by approximations (\ref{zeta1}) and (\ref{interpol}). Analytic estimate of integral (\ref{NB}) gives a rather complicated asymptotics at $\phi \gg 1$\,: \beq N_B (\phi) \propto \frac{\log^{3/2} \phi}{\sqrt{\phi}} \, e^{- \frac12 \sqrt{1 + 6 \log \phi}} \, . \eeq However, in a very wide range of the values of argument, $10^2 \lesssim \phi \lesssim 10^{50}$, function (\ref{NB}) is excellently interpolated by a simple power law (see figure~\ref{fig:interpol-1}) \beq \label{interpol-1} N_B (\phi) \approx 9 \phi^{- 5/9} \, . \eeq \begin{figure}[ht] \begin{minipage}[t]{.48\textwidth} \vspace{0pt} \includegraphics[width=\textwidth]{interpol-1-1.eps} \end{minipage} \begin{minipage}[t]{.51\textwidth} \vspace{0pt} \includegraphics[width=\textwidth]{interpol-1-2.eps} \end{minipage} \caption{The function $N_B (\phi)$ in (\ref{NB}) with $Z (x)$ given by (\ref{zx}) (solid blue line) and its interpolation by (\ref{interpol-1}) (dashed pink line) in different ranges of argument. The curves on the right graph practically coincide. \label{fig:interpol-1}} \end{figure} With $\phi$ being asymptotically given by [see (\ref{phi}) and (\ref{kini})] \beq \label{phi-as} \phi = \frac{2 \alpha^2 M_* \Delta \mu_\ini^2}{\pi^2 \sigma_c T} \, , \eeq equations (\ref{muas}) and (\ref{NB}), (\ref{interpol-1}) give the asymptotic behavior of the quantities $\Delta \mu$ and $r_B$ as functions of temperature and of their initial values: \beq \frac{\Delta \mu }{\Delta \mu_\ini} = \left[ \frac{\pi^2 \sigma_c T}{2 \alpha^2 M_* \Delta \mu_\ini^2} \log \frac{2 \alpha^2 M_* \Delta \mu_\ini^2}{\pi^2 \sigma_c T} \right]^{1/2} \, , \qquad \frac{r_B}{r_B^\ini} = 9 \left( \frac{\pi^2 \sigma_c T}{2 \alpha^2 M_* \Delta \mu_\ini^2} \right)^{5/9} \eeq \section{Decay of chiral asymmetry caused by chirality flipping} In the preceding analysis, we totally neglected the flipping term with coefficient $\Gamma_{\rm f} (\eta)$ in (\ref{muevo}), which is justified at high temperatures. However, as the temperature drops down because of cosmological expansion, at some point this term starts dominating over the other terms on the right-hand side of (\ref{muevo}), after which the chiral chemical potential decays exponentially as \beq \label{expodec} \Delta \mu (T) \propto \exp \left( - \frac{\alpha^2 M_* m_e^2}{27\, T^3} \right) \, . \eeq In this section, we estimate the temperature $T_{\rm f}$ at which this decay commences. The contribution to the coefficient $\Gamma_{\rm f} (\tau)$ comes from weak and electromagnetic processes, so that we have $\Gamma_{\rm f} = \Gamma_w + \Gamma_e$, where the weak and electromagnetic contributions are estimated, respectively, as (see \cite{Boyarsky:2011uy}) \beq \label{gammas} \Gamma_w \sim G_{\rm F}^2 T^4 \left( \frac{m_e}{3 T} \right)^2 \, , \qquad \Gamma_e \sim \alpha^2 \left( \frac{m_e}{3 T} \right)^2 \, . \eeq Here, $G_{\rm F}$ is the Fermi constant, $\alpha$ is the fine structure constant, and $m_e$ is the electron mass. The factors in the brackets in (\ref{gammas}) with electron mass $m_e$ describe suppression of chirality-flipping scattering rates with respect to `chirality-preserving' ones. The weak contribution $\Gamma_w$ dominates at temperatures $T > T_{\rm eq} \simeq \sqrt{\alpha/ G_{\rm F}} \approx 25$~GeV, while, at $T < T_{\rm eq}$, chirality flipping is dominated by the electromagnetic processes and is characterized by $\Gamma_e$ in (\ref{gammas}). In view of equation (\ref{dotA1}), the first term on the right-hand side of (\ref{muevo}) is itself a sum of two terms with opposite signs: \beq \label{two-term} - \frac{c_\Delta \alpha}{\pi^2 \sigma_c} \int_0^\infty \dot A (k, \tau) k d k = \frac{2 c_\Delta \alpha}{\pi^2 \sigma_c} \int_0^\infty A (k, \tau) k^3 d k - \frac{2 c_\Delta \alpha^2 \Delta \mu}{\pi^3 \sigma_c} \int_0^\infty \left[ P (k, \tau) + S_0 (k) \right] k^2 d k \, . \eeq In this expression, the thermal `tail' in the spectral function $A (k, \tau)$ compensates the thermal contribution from $S_0 (k)$. Indeed, at the `tail,' we have $A = A_{\rm eq} = S_0 \left(k_\mu/ k \right)$, and taking into account (\ref{kmu}), we observe cancellation of the corresponding integrals in (\ref{two-term}). Therefore, it is the negative term with the integral of the spectral function $P (k, \tau)$ in (\ref{two-term}) that is to be considered. To determine whether the neglect of the flipping term in (\ref{muevo}) is legitimate, we should compare the absolute value of this term with the absolute value $\Gamma_{\rm f} (\tau) \Delta \mu (\tau)$ of the flipping term. It is convenient to divide both quantities by $\Delta \mu (\tau)$. For the first expression, we have \beq \label{regdec} \frac{2 c_\Delta \alpha^2}{\pi^3 \sigma_c} \int_0^\infty P (k, \tau) k^2 d k = \frac{2 c_\Delta \alpha^2}{\pi^3 \sigma_c} g_* r_B (\tau) \int_0^\infty S_0 (k) k^2 d k = \frac{2 \pi c_\Delta \alpha^2}{15\, \sigma_c} g_* r_B (\tau)\, , \eeq Using (\ref{NB}) and (\ref{interpol-1}) to express $r_B (\tau)$ through $r_B^\ini$, we estimate (\ref{regdec}) as \beq \label{regdec1} \frac{2 \pi c_\Delta \alpha^2}{15\, \sigma_c} g_* r_B (\tau) \simeq \frac{\pi \alpha^2 }{\sigma_c \phi^{5/9}} g_* r_B^\ini \, . \eeq This expression should be compared to each of the quantities in (\ref{gammas}). Assuming that chirality flipping comes into play at temperatures $T < T_{\rm eq} \approx 25$~GeV (this will be confirmed by the final estimate), we only need to take into account the electromagnetic part $\Gamma_e$. We then have an equation for the estimate of the temperature of decay caused by chirality flipping: \beq \label{finest} \frac{\pi}{\sigma_c \phi^{5/9}} g_* r_B^\ini \simeq \left( \frac{m_e}{3 T} \right)^2 \, . \eeq The asymptotic value of $\phi \gg 1$ is given by (\ref{phi-as}). Substituting it into (\ref{finest}) and solving the resulting equation with respect to $T$, we obtain \beq \label{Tf} T_{\rm f} = \left( \frac{2^5 \alpha^{10} \sigma_c^4 m_e^{18} M_*^5 \Delta \mu_\ini^{10}}{3^{18} \pi^{19} \left[ g_* r_B^\ini \right]^9} \right)^{1/23} \approx \frac{1.6 \times 10^3}{\sqrt{g_*}} \, \left( \frac{\Delta \mu_\ini^{10}}{\left[ r_B^\ini \right]^9} \right)^{1/23} \, \, \mbox{MeV} \, , \eeq where we have put the numerical values for physical constants. For $g_* = 30$, $\Delta \mu_\ini = 3 \times 10^{-5}$, and $r_B^\ini = 5 \times 10^{-5}$, this equation gives $T_{\rm f} \simeq 150$~MeV (at this time, $\phi \approx 1200$). For $r_B^\ini = 5 \times 10^{-4}$, we obtain $T_{\rm f} \simeq 60$~MeV (with $\phi \approx 3000$). This is in good qualitative agreement with the numerical results of \cite{Boyarsky:2011uy}. Note that the resulting temperature (\ref{Tf}) does not depend on the temperature at which the initial values $\Delta \mu_\ini$ and $r_B^\ini$ are set (and which is assumed to be much higher than $T_{\rm f}$). This is due to the asymptotic scaling $\Delta \mu \propto \phi^{-1/2}$ [see (\ref{muas})], and $r_B \propto \phi^{-5/9}$ [see (\ref{NB}) and (\ref{interpol-1})], ensuring that the ratio $\Delta \mu^{10}/ r_B^9$ remains to be roughly constant in the regime of inverse cascade. \section{Summary} We provided an analytic treatment of the process of generation of helical magnetic field in an early hot universe in the presence of externally induced lepton chiral asymmetry, and of the subsequent mutual evolution of the chiral asymmetry and magnetic field. Helical magnetic field is generated from the thermal initial spectrum (extrapolated to all scales including the super-horizon ones) owing to the effects of quantum chiral anomaly. The thermal bath also serves as a medium of relaxation of magnetic field to its thermal state. The generated helical magnetic field and the lepton chiral asymmetry are capable of supporting each other, thus prolonging their existence to cosmological temperatures as low as tens of MeV, with spectral power being permanently transformed from small to large spatial scales (the so-called `inverse cascade') \cite{Boyarsky:2011uy}. Our main results are summarized as follows. We obtained analytic expressions describing the evolution of the lepton chiral chemical potential and magnetic-field energy density. For a developed maximally helical magnetic field, both the chiral chemical potential $\Delta \mu$ and the relative fraction of magnetic-field energy density $r_B$ depend on their initial values and on time through a single variable $\phi$ introduced in (\ref{phi}). This scaling property is encoded in equations (\ref{dzeta})--(\ref{muevol}) and (\ref{NB})--(\ref{interpol-1}), and depicted in figures \ref{fig:muevol} and \ref{fig:interpol-1}. The late-time asymptotics for $\Delta \mu$ reproduces the scaling law $\Delta \mu \propto \eta^{-1/2} \log^{1/2} \eta$ [see equation (\ref{muas})] previously found in this system in \cite{Hirono:2015rla, Xia:2016any}. By numerical interpolation, we find that the relative fraction $r_B$ of the magnetic-field energy density in this regime decays as $r_B \propto \eta^{-5/9}$ all through the relevant part of the cosmological history. Since the conformal time $\eta$ in our units is related to the temperature $T$ as $\eta = M_* / T$, this also describes the evolution of these quantities with temperature. As the temperature drops to sufficiently low values due to the cosmological expansion, the chirality-flipping lepton scattering processes take control over the evolution of chiral asymmetry, leading to its rapid decay (\ref{expodec}). We derived a simple expression (\ref{Tf}) for the temperature at which this happens, depending on the initially generated values of the energy density of magnetic field and of the lepton chiral asymmetry. The analytic expressions obtained in this paper are sufficiently general and may be used for primary evaluation of scenarios of cosmological magnetogenesis by lepton chiral asymmetry. \acknowledgments We are grateful to Alexey Boyarsky and Oleg Ruchayskiy for valuable comments. M.~S. and O.~T. acknowledge support from the Scientific and Educational Center of the Bogolyubov Institute for Theoretical Physics. The work of O.~T. was supported by the WFS National Scholarship Programme and by the Deutsche Forschungsgemeinschaft (DFG) in part through the Collaborative Research Center ``The Low-Energy Frontier of the Standard Model'' (SFB 1044), in part through the Graduate School ``Symmetry Breaking in Fundamental Interactions'' (DFG/GRK 1581), and in part through the Cluster of Excellence ``Precision Physics, Fundamental Interactions and Structure of Matter'' (PRISMA). The work of Y.~S. was supported by the Swiss National Science Foundation grant SCOPE IZ7370-152581.
5f24b5927a023e8542f9f621e69759e4bb323094
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{s1} In this paper, we consider distributed optimization problems where the goal is to minimize a sum of objective functions over a multi-agent network. Formally, we consider a decision variable,~$\mb{z}\in\mbb{R}^p$, and a strongly-connected network containing~$n$ agents, where each agent,~$i$, only has access to a local objective function,~$f_i:\mbb{R}^p\rightarrow\mbb{R}$. The goal is to have each agent minimize the sum of objectives,~$\sum_{i=1}^nf_i(\mb{z})$, via information exchange with the neighbors. This formulation has gained great interest due to its widespread applications in, e.g., large-scale machine learning,~\cite{distributed_Boyd,ml}, model-predictive control,~\cite{distributed_Necoara}, cognitive networks,~\cite{distributed_Mateos,distributed_Bazerque}, source localization,~\cite{distributed_Rabbit,distributed_Khan}, resource scheduling,~\cite{distributed_Chunlin}, and message routing,~\cite{distributed_Neglia}. Most of the existing algorithms assume information exchange over undirected networks (graphs), where the communication between the agents is bidirectional, i.e., if agent~$i$ sends information to agent~$j$ then agent~$j$ can also send information to agent~$i$. Related work includes Distributed Gradient Descent (DGD),~\cite{uc_Nedic,cc_nedic,cc_Lobel2,cc_Ram2}, which achieves~$O(\frac{\ln k}{\sqrt{k}})$ convergence for arbitrary convex functions, and~$O(\frac{\ln k}{k})$ for strongly-convex functions, where~$k$ is the number of iterations. The convergence rates can be accelerated with an additional Lipschitz-continuity assumption on the associated gradient. For example, see DGD~\cite{DGD_Yuan} that converges at~$O(\frac{1}{k})$ for general convex functions but within a ball around the optimal solution, whereas, it converges linearly to the optimal solution for strongly-convex functions. The distributed Nestrov's method,~\cite{fast_Gradient}, converges at~$O(\frac{\ln k}{k^2})$ for general convex functions. Of significant relevance is EXTRA,~\cite{EXTRA}, which converges to the optimal solution at~$O(\frac{1}{k})$ for general convex functions and is linear for strongly-convex functions. The work in \cite{Augmented_EXTRA} improves EXTRA by relaxing the weight matrices to be asymmetric. Besides the gradient-based methods, the distributed implementation of ADMM,~\cite{ADMM_Mota,ADMM_Wei, ADMM_Shi}, has also been considered over undirected graphs. The aforementioned methods,~\cite{uc_Nedic,cc_nedic,cc_Lobel2,cc_Ram2,DGD_Yuan,fast_Gradient,EXTRA,Augmented_EXTRA,ADMM_Mota,ADMM_Wei, ADMM_Shi}, are applicable to undirected graphs that allow the use of doubly-stochastic weight matrices; row-stochasticity guarantees that all agents reach consensus, while the column-stochasticity ensures that each local gradient contributes equally to the global objective,~\cite{D-DGD}. On the contrary, when the underlying graph is directed, the weight matrix may only be row-stochastic or column-stochastic but not both. In this paper, we provide a distributed optimization algorithm that does not require doubly-stochastic weights and thus is applicable to directed graphs (digraphs). See~\cite{GHARESIFARD2012539,7100912} for work on balancing the weights in strongly-connected digraphs. Optimization in continuous-time over weight-balanced digraphs has been studied earlier in~\cite{weightbalanceddigraphs1,weightbalanceddigraphs2}. Existing discrete-time algorithms include the following: Gradient-Push (GP), \cite{opdirect_Nedic,opdirect_Tsianous,opdirect_Tsianous2,opdirect_Tsianous3}, that combines DGD, \cite{uc_Nedic}, and push-sum consensus,~\cite{ac_directed0,ac_directed}; Directed-Distributed Gradient Descent (D-DGD),~\cite{D-DGD,D-DPS}, which uses Cai and Ishii's work on surplus consensus,~\cite{ac_Cai1}, and combines it with DGD; and~\cite{opdirect_Makhdoumi}, where the authors apply the weight-balancing technique,~\cite{c_Hooi-Tong}, to DGD. These gradient-based methods,~\cite{opdirect_Nedic,opdirect_Tsianous,opdirect_Tsianous2,opdirect_Tsianous3,D-DGD,D-DPS,opdirect_Makhdoumi}, restricted by the diminishing step-size, converge relatively slowly at~$O(\frac{\ln k}{\sqrt{k}})$. When the objective functions are strongly-convex, the convergence rate can be accelerated to~$O(\frac{\ln k}{k})$,~\cite{opdirect_Nedic3}. A recent paper proposed a fast distributed algorithm, termed DEXTRA, \cite{DEXTRA,the_copy_work}, to solve the distributed consensus optimization problem over directed graphs. By combining the push-sum protocol, \cite{ac_directed0,ac_directed}, and EXTRA,~\cite{EXTRA}, DEXTRA achieves a linear convergence rate given that the objective functions are strongly-convex. However, a limitation of DEXTRA is a restrictive step-size range, i.e., the greatest lower bound of DEXTRA's step-size is strictly greater than zero. In particular, DEXTRA requires the step-size,~$\alpha$, to follow~$\alpha\in(\underline{\alpha},\overline{\alpha})$, where~$\underline{\alpha}>0$. Estimating~$\underline{\alpha}$ in a distributed setting is challenging because it may require global knowledge. In contrast if~$\underline{\alpha}=0$, agents can pick a small enough positive constant to ensure the convergence. In this paper, we propose ADD-OPT (Accelerated Distributed Directed Optimization) to address the step-size limitation inherent to DEXTRA. In particular, ADD-OPT's step-size follows~$\alpha\in(0,\overline{\alpha})$, i.e.,~$\underline{\alpha}=0$, ensuring that the lower bound of ADD-OPT's step-size does not require any global knowledge. We show that ADD-OPT converges linearly for strongly-convex functions. The remainder of the paper is organized as follows. Section~\ref{s2} formulates the problem and describes ADD-OPT. We also present appropriate assumptions in Section \ref{s2}. Section~\ref{s3} states the main convergence results. In Section~\ref{s4}, we present some lemmas as the basis of the proof of ADD-OPT's convergence. The proof of main results is provided in Section~\ref{s5}. We show numerical results in Section~\ref{s6} and Section~\ref{s7} contains the concluding remarks. \textbf{Basic Notation:} We use lowercase bold letters to denote vectors and uppercase italic letters to denote matrices. The matrix,~$I_n$, represents the~$n\times n$ identity;~$\mb{1}_n$ and~$\mb{0}_n$ are the~$n$-dimensional column vectors of all~$1$'s and~$0$'s, respectively. We denote by~$A\otimes B$, the Kronecker product of two matrices,~$A$ and~$B$. For any~$f(\mb{x})$,~$\nabla f(\mb{x})$ denotes the gradient of~$f$ at~$\mb{x}$. The spectral radius of a matrix,~$A$, is represented by~$\rho(A)$. \begin{color}{black}For an irreducible, column-stochastic matrix,~$A$, we denote its right and left eigenvectors corresponding to the eigenvalue of~$1$ by $\boldsymbol{\pi}$ and~$\mb{1}_n^\top$, respectively, such that~$\mb{1}_n^\top\boldsymbol{\pi} = 1$. Depending on its argument, we denote~$\|\cdot\|$ as either a particular matrix norm, the choice of which will be clear in Lemma~\ref{lem3}, or a vector norm that is compatible with this particular matrix norm, i.e.,~$\|A\mb{x}\|\leq\|A\|\|\mb{x}\|$ for all matrices,~$A$, and all vectors,~$\mb{x}$. The notation~$\|\cdot\|_2$ denotes the Euclidean norm of vectors and matrices. Since all vector norms on finite-dimensional vector space are equivalent, we have the following:~$c{'}\|\cdot\|\leq\|\cdot\|_2\leq c\|\cdot\|, d{'}\|\cdot\|_2\leq\|\cdot\|\leq d\|\cdot\|_2$, where~$c{'},c,d{'},d$ are some positive constants. \end{color} \section{ADD-OPT Development}\label{s2} In this section, we formulate the optimization problem and describe ADD-OPT. We first derive an informal but intuitive proof showing that ADD-OPT enables the agents to achieve consensus and reach the optimal solution of Problem P1, described below. After propose ADD-OPT, we relate it to DEXTRA and discuss the applicable range of step-sizes. Formal convergence results are deferred to Sections~\ref{s3}. Consider a strongly-connected network of~$n$ agents communicating over a directed graph,~$\mc{G}=(\mc{V},\mc{E})$, where~$\mc{V}$ is the set of agents, and~$\mc{E}$ is the collection of ordered pairs,~$(i,j),i,j\in\mc{V}$, such that agent~$j$ can send information to agent~$i$, $j\rightarrow i$. Define~$\mc{N}_i^{{\scriptsize \mbox{in}}}$ to be the collection of in-neighbors, i.e., the set of agents that can send information to agent~$i$. Similarly,~$\mc{N}_i^{{\scriptsize \mbox{out}}}$ is the set of out-neighbors of agent~$i$. Note that both~$\mc{N}_i^{{\scriptsize \mbox{in}}}$ and~$\mc{N}_i^{{\scriptsize \mbox{out}}}$ include node~$i$. Note that in a directed graph when~$(i,j)\in\mc{E}$, it is not necessary that~$(j,i)\in\mc{E}$. Consequently,~$\mc{N}_i^{{\scriptsize \mbox{in}}}\neq\mc{N}_i^{{\scriptsize \mbox{out}}}$, in general. We assume that each agent~$i$ knows\footnote{Such an assumption is standard in the related literature, see e.g.,~\cite{opdirect_Nedic,opdirect_Tsianous,opdirect_Tsianous2,opdirect_Tsianous3,D-DGD,D-DPS,opdirect_Makhdoumi,DEXTRA}.} its out-degree (the number of out-neighbors), denoted by~$|\mc{N}_i^{{\scriptsize \mbox{out}}}|$; see~\cite{bullo_book} for details. We focus on solving a convex optimization problem that is distributed over the above multi-agent network. In particular, the network of agents cooperatively solves the following optimization problem: \begin{align} \mbox{P1}: \quad\mbox{min }&f(\mb{z})=\sum_{i=1}^nf_i(\mb{z}),\nonumber \end{align} where each local objective function,~$f_i:\mbb{R}^p\rightarrow\mbb{R}$ is known only by agent~$i$. We assume that each local function,~$f_i(z)$, is strongly-convex and differentiable, whereas the optimal solution of Problem P1 exists and is finite. Our goal is to develop a distributed algorithm such that each agent converges to the global solution of Problem P1 via exchanging information with nearby agents over a directed graph. We formalize the set of assumptions as follows. These assumptions are standard in the literature for optimization of smooth convex functions, see e.g.,~\cite{EXTRA,DEXTRA,DGD_Yuan}. \begin{assump}\label{asp2} The communication graph,~$\mc{G}$, is a strongly-connected digraph. Each agent in the network has the knowledge of its out-degree. \end{assump} \begin{assump}[Lipschitz-continuous gradients and strong-convexity]\label{asp1} Each local function,~$f_i$, is differentiable and strongly-convex, and the gradient is globally Lipschitz-continuous, i.e., for any~$i$ and~$\mb{z}_1, \mb{z}_2\in\mbb{R}^p$, {\color{black} \begin{enumerate}[label=(\alph*)] \item there exists a positive constant~$l$ such that $$\|\mb{\nabla} f_i(\mb{z}_1)-\mb{\nabla} f_i(\mb{z}_2)\|_2\leq l\|\mb{z}_1-\mb{z}_2\|_2;$$ \item there exists a positive constant~$s$ such that, $$f_i(\mb{z}_1)-f_i(\mb{z}_2)\leq\mb{\nabla} f_i(\mb{z}_1)^\top(\mb{z}_1-\mb{z}_2)-\frac{s}{2}\|\mb{z}_1-\mb{z}_2\|_2^2.$$ \end{enumerate} Clearly, the Lipschitz-continuity and strongly-convexity constants for the global objective function~$f(\mb{z})$ are~$nl$ and~$ns$, respectively. } \end{assump} \begin{assump}\label{asp3} The optimal solution exists and is bounded and unique. In particular, we denote~$\underline{\mb{z}}^*\in\mbb{R}^p$ the optimal solution, i.e., $$\underline{\mb{z}}^*=\min_{\mb{z}\in\mbb{R}^p}f(\mb{z}).$$ \end{assump} \subsection{ADD-OPT Algorithm} To solve Problem P1, we describe the implementation of ADD-OPT as follows. Each agent,~$j\in\mc{V}$, maintains three vector variables:~$\mb{x}_{k}^j$,~$\mb{z}_{k}^j$,~$\mb{w}_{k}^j,$ all in~$\mbb{R}^p$, as well as a scalar variable,~$\mathsf{y}_{k}^j\in\mbb{R}$, where~$k$ is the discrete-time index. At the~$k$th iteration, agent~$j$ assigns a weight to its states:~$a_{ij}\mb{x}_{k}^j$,~$a_{ij}\mb{w}_{k}^j$, and~$a_{ij}\mathsf{y}_{k}^j$; and sends these to each of its out-neighbors,~$i\in\mc{N}_j^{{\scriptsize \mbox{out}}}$, where the weights,~$a_{ij}$'s are such that: \begin{align} a_{ij}&=\left\{ \begin{array}{rl} >0,&i\in\mc{N}_j^{{\scriptsize \mbox{out}}},\\ 0,&\mbox{otherwise}, \end{array} \right. \quad \sum_{i=1}^na_{ij}=1,\forall j.\label{a} \end{align} With agent~$i$ receiving the information from its in-neighbors, it updates~$\mb{x}_{k+1}^i$,~$\mathsf{y}_{k+1}^i$,~$\mb{z}_{k+1}^i$ and~$\mb{w}_{k+1}^i$ as follows: \begin{subequations}\label{alg1} \begin{align} \mb{x}_{k+1}^i=&\sum_{j\in\mc{N}_i^{{\tiny \mbox{in}}}}a_{ij}\mb{x}_{k}^j-\alpha\mb{w}_{k}^i,\label{alg1a}\\ \mathsf{y}_{k+1}^i=&\sum_{j\in\mc{N}_i^{{\tiny \mbox{in}}}}a_{ij}\mathsf{y}_{k}^j,\label{alg1b}\\ \mb{z}_{k+1}^i=&\frac{\mb{x}_{k+1}^i}{\mathsf{y}_{k+1}^i},\label{alg1c}\\ \mb{w}_{k+1}^i=&\sum_{j\in\mc{N}_i^{{\tiny \mbox{in}}}}a_{ij}\mb{w}_{k}^j+\nabla f_i(\mb{z}_{k+1}^i)-\nabla f_i(\mb{z}_{k}^i).\label{alg1d} \end{align} \end{subequations} In the above,~$\nabla f_i(\mb{z}_{k}^i)$ is the gradient of~$f_i(\mb{z})$ at~$\mb{z}=\mb{z}_{k}^i$. The step-size,~$\alpha$, is a positive number within a certain interval. We will explicitly show the range of~$\alpha$ in Section~\ref{s3}. For any agent~$i$, it is initialized with arbitrary vectors,~$\mb{x}_{0}^i$ and $\mb{z}_0^i$,~$\mb{w}_{0}^i=\nabla f_i(\mb{z}_{0}^i)$, and~$\mathsf{y}_{0}^i=1$. It is worth noting that~$\mathsf{y}_{k}^i\neq0$,~$\forall k$, given its initial condition and Assumption~\ref{asp2}, \cite{eig2}. We note that Eq.~\eqref{a} leads to a column-stochastic weight matrix,~$\underline{A}=\{a_{ij}\}$, by only requiring each agent to know its out-degree. It is indeed possible to construct such weights, e.g., by choosing \begin{align} a_{ij}&=\left\{ \begin{array}{rl} 1/|\mc{N}_j^{{\scriptsize \mbox{out}}}|,&i\in\mc{N}_j^{{\scriptsize \mbox{out}}},\\ 0,&\mbox{otherwise}, \end{array} \right. \quad ,\forall j.\label{aa} \end{align} For analysis purposes, we now write Eq.~\eqref{alg1} in a matrix form. We use the following notation: \begin{eqnarray}\label{not1} \mb{x}_k= \left[ \begin{array}{c} \mb{x}_{k}^1\\ \vdots\\ \mb{x}_{k}^n \end{array} \right],~\mb{w}_k= \left[ \begin{array}{c} \mb{w}_{k}^1\\ \vdots\\ \mb{w}_{k}^n \end{array} \right],~\mb{z}_k= \left[ \begin{array}{c} \mb{z}_{k}^1\\ \vdots\\ \mb{z}_{k}^n \end{array} \right],\\\label{not2} \nabla\mb{f}_k= \left[ \begin{array}{c} \nabla \mb{f}_1(\mb{z}_{k}^1)\\ \vdots\\ \nabla \mb{f}_n(\mb{z}_{k}^n) \end{array} \right],~\mb{y}_k= \left[ \begin{array}{c} \mathsf{y}_{k}^1\\ \vdots\\ \mathsf{y}_{k}^n \end{array} \right]. \end{eqnarray} Let~$\underline{A}\in\mbb{R}^{n\times n}$ be the weighted adjacency matrix, i.e., the collection of weights,~$a_{ij}$; define \begin{eqnarray}\label{not3} A&=&\underline{A}\otimes I_p,\label{A}\\\label{Yk} Y_k&=&\mbox{diag}\left(\mb{y}_k\right)\otimes I_p. \end{eqnarray} where `$\otimes$' is the Kronecker product. Clearly, we have $A,Y_k\in\mbb{R}^{np\times np}$, and~$A$ is a column-stochastic matrix. Given that~$\mb{y}_0=\mb{1}_n$, the graph,~$\mc{G}$, is strongly-connected and the corresponding weight matrix,~$\underline{A}$, is non-negative, $Y_k$ is invertible for any~$k$,~\cite{eig2}. Then, we can write Eq.~\eqref{alg1} in the matrix form, equivalently, as follows: \begin{subequations}\label{alg1_matrix} \begin{align} \mb{x}_{k+1}=&A\mb{x}_{k}-\alpha\mb{w}_{k},\label{alg1_ma}\\ \mb{y}_{k+1}=&\underline{A}\mb{y}_{k},\label{alg1_mb}\\ \mb{z}_{k+1}=&Y_{k+1}^{-1}\mb{x}_{k+1},\label{alg1_mc}\\ \mb{w}_{k+1}=&A\mb{w}_{k}+\nabla \mb{f}_{k+1}-\nabla \mb{f}_{k},\label{alg1_md} \end{align} \end{subequations} where we have the initial condition~$\mb{w}_0=\nabla\mb{f}_0$,~$\mb{y}_0=\mb{1}_n$. \subsection{Interpretation of ADD-OPT} Based on Eq.~\eqref{alg1_matrix}, we now give an intuitive interpretation on the convergence of ADD-OPT to the optimal solution. By combining Eqs. \eqref{alg1_ma} and \eqref{alg1_md}, we obtain that \begin{align} \mb{x}_{k+1}&=A\mb{x}_{k}-\alpha\left[A\mb{w}_{k-1}+\nabla \mb{f}_{k}-\nabla \mb{f}_{k-1}\right],\nonumber\\ &=A\mb{x}_{k}-\alpha A\left[\frac{A\mb{x}_{k-1}-\mb{x}_k}{\alpha}\right]-\alpha\left[\nabla \mb{f}_{k}-\nabla \mb{f}_{k-1}\right],\nonumber\\ &=2A\mb{x}_{k}-A^2\mb{x}_{k-1}-\alpha\left[\nabla \mb{f}_{k}-\nabla \mb{f}_{k-1}\right].\label{alg2} \end{align} Assume that the sequences generated by Eq.~\eqref{alg1_matrix} converge to their limits (note that this is not necessarily true), denoted by~$\mb{x}_\infty$,~$\mb{y}_\infty$,~$\mb{w}_\infty$,~$\mb{z}_\infty$,~$\nabla\mb{f}_\infty$, respectively. It follows from Eq.~\eqref{alg2} that \begin{align}\label{int_2} \mb{x}_{\infty}=2A\mb{x}_{\infty}-A^2\mb{x}_{\infty}-\alpha\left[\nabla \mb{f}_{\infty}-\nabla \mb{f}_{\infty}\right], \end{align} which implies that~$(I_{np}-A)^2\mb{x}_\infty=\mb{0}_{np}$ or~$[(I_n-\underline{A})^2\otimes I_p]\mb{x}_\infty=\mb{0}_{np}$. Considering that~$\mb{y}_\infty=\underline{A}\mb{y}_\infty$, we obtain that~$\mb{x}_\infty\in\mbox{span}\{\mb{y}_\infty\otimes\mb{u}_p\}$ for some arbitrary~$p$-dimensional vector,~$\mb{u}_p$. Therefore, it follows that \begin{align}\label{int_3} \mb{z}_\infty=Y_{\infty}^{-1}\mb{x}_{\infty}\in\mbox{span}\{\mb{1}_n\otimes\mb{u}_p\}, \end{align} where~$\mb{u}_p$ is some arbitrary~$p$-dimensional vector. The consensus is reached. By summing up the updates in Eq.~\eqref{alg2} over~$k$ from~$0$ to~$\infty$, we obtain that \begin{align} \mb{x}_\infty=A\mb{x}_\infty+\sum_{r=1}^\infty(A-I_{np})\mb{x}_r-\sum_{r=0}^\infty(A^2-A)\mb{x}_r-\alpha\nabla\mb{f}_\infty.\nonumber \end{align} Noting that~$\mb{x}_\infty=A\mb{x}_\infty$, it follows \begin{align} \alpha\nabla\mb{f}_\infty=\sum_{r=1}^\infty(A-I_{np})\mb{x}_r-\sum_{r=0}^\infty(A^2-A)\mb{x}_r.\nonumber \end{align} Therefore, we obtain that \begin{align} &\alpha(\mb{1}_n\otimes I_p)^\top\nabla\mb{f}_\infty\nonumber\\ &=\left(\mb{1}_n^\top(\underline{A}-I_{n})\otimes I_p\right)\sum_{r=1}^\infty\mb{x}_r-\left(\mb{1}_n^\top(\underline{A}^2-\underline{A})\otimes I_p\right)\sum_{r=0}^\infty\mb{x}_r,\nonumber\\ &=\mb{0}_p,\nonumber \end{align} which is the optimality condition of Problem P1 considering that~$\mb{z}_\infty\in\mbox{span}\{\mb{1}_n\otimes\mb{u}_p\}$. To summarize, if we assume that the sequences updated in Eq.~\eqref{alg1_matrix} have limits,~$\mb{x}_\infty$,~$\mb{y}_\infty$,~$\mb{w}_\infty$,~$\mb{z}_\infty$,~$\nabla\mb{f}_\infty$, we arrive at a conclusion that~$\mb{z}_\infty$ achieves consensus and reaches the optimal solution of Problem P1. We next discuss the relations between ADD-OPT and DEXTRA. \subsection{ADD-OPT and DEXTRA} Recent papers provide a fast distributed algorithm, termed DEXTRA \cite{DEXTRA,the_copy_work}, to solve Problem P1 over directed graphs. It achieves a linear convergence rate given that the objective functions are strongly-convex. At the~$k$th iteration of DEXTRA, each agent~$i$ keeps and updates three states,~$x_{k,i}$,~$\mathsf{y}_{k,i}$, and~$z_{k,i}$. The iteration, in matrix form, is shown as follows. \begin{subequations}\label{dextra} \begin{align} \mb{x}_{k+1}=&\left(I_{np}+A\right)\mb{x}_k-\widetilde{A}\mb{x}_{k-1}-\alpha\left[\nabla\mb{f}_k-\nabla\mb{f}_{k-1}\right],\label{dextra1a}\\ \mb{y}_{k+1}=&\underline{A}\mb{y}_k,\label{dextra1b}\\ \mb{z}_{k+1}=&Y_{k+1}^{-1}\mb{x}_{k+1},\label{dextra1c} \end{align} \end{subequations} where~$\widetilde{A}$ is a column-stochastic matrix satisfying that~$\widetilde{A}=\theta I_{np}+(1-\theta)A$ with any~$\theta\in(0,\frac{1}{2}]$, and all other notation is the same as from earlier in this paper. By comparing Eqs.~\eqref{alg2} and \eqref{dextra1a},~\eqref{alg1_mb} and \eqref{dextra1b}, and~\eqref{alg1_mc} and \eqref{dextra1c}, it follows that the only difference between ADD-OPT and DEXTRA lies in the weighting matrices used when updating~$\mb{x}_k$. From DEXTRA to ADD-OPT, we change~$(I_{np}+A)$ in \eqref{dextra1a} to~$2A$ in \eqref{alg2}, and~$\widetilde{A}$ to~$A^2$, respectively. Mathematically, if~$A=I_{np}$, (equivalently~$\underline{A}=I_n$), the two algorithms are the same. With this modification, we will show in Section \ref{s3} that ADD-OPT supports a wider range of step-sizes as compared to DEXTRA, i.e., the greatest lower bound,~$\underline{\alpha}$, of ADD-OPT's step-size is zero while that of DEXTRA's is strictly positive. This also reveals the reason why in DEXTRA constructing~$\underline{A}$ to be an extremely diagonally-dominant matrix is preferred, see Assumption A2(c) in \cite{DEXTRA}. The more similar~$\underline{A}$ is to~$I_n$, the closer~$\underline{\alpha}$ approaches zero. However, in DEXTRA,~$\underline{\alpha}$ can never reach zero since~$\underline{A}$ cannot be the identity,~$I_n$, which otherwise means there is no communication between agents. In Section \ref{s5}, we provide a totally different proof, that is further more compact and elegant when compared to DEXTRA's analysis, to show the linear convergence rate of ADD-OPT. \section{Main Result}\label{s3} In this section, we analyze ADD-OPT with the help of the following notation. From Eqs.~\eqref{not1}-\eqref{Yk}, we further define~$\overline{\mb{x}}_k$,~$\overline{\mb{w}}_k$,~$\mb{z}^*$,~$\mb{g}_k$,~$\mb{h}_k\in\mbb{R}^{np}$ as \begin{align}\label{eqxb} \overline{\mb{x}}_k&=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\mb{x}_k,\\\label{eqwb} \overline{\mb{w}}_k&=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\mb{w}_k,\\ \mb{z}^*&=\mb{1}_n\otimes\underline{\mb{z}}^*,\\ \mb{g}_k&=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\nabla\mb{f}_k,\\ \mb{h}_k&=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\nabla\mb{f}(\overline{\mb{x}}_k), \end{align} where \begin{eqnarray*} \nabla\mb{f}(\overline{\mb{x}}_k)= \left[ \begin{array}{c} \nabla f_1(\frac{1}{n}(\mb{1}_n^\top\otimes I_p)\mb{x}_k)\\ \vdots\\ \nabla f_n(\frac{1}{n}(\mb{1}_n^\top\otimes I_p)\mb{x}_k) \end{array} \right], \end{eqnarray*} stacks its components in a column. We denote constants,~$\tau$,~$\epsilon$, and~$\eta$ as {\color{black} \begin{align} \tau&=\left\|A-I_{np}\right\|_2,\label{parameter1}\\ \epsilon&=\left\|I_{np}-A_\infty\right\|_2,\label{parameter2}\\ \eta&=\max\left(\left|1-n\alpha l\right|,\left|1-n\alpha s\right|\right)\label{parameter3}, \end{align} } where~$A$ is the column-stochastic weight matrix used in Eq. \eqref{alg1_matrix},~$A_\infty=\lim_{k\rightarrow\infty}A^k$ represents~$A$'s limit,~$\alpha$ is the step-size, and~$l$ and~$s$ are respectively Lipschitz and strong-convexity constants from Assumption~\ref{asp1}. Let~$Y_\infty$ be the limit of~$Y_k$ in Eq. \eqref{Yk}, \begin{align}\label{yinf_eq} Y_\infty=\lim_{k\rightarrow\infty}Y_k, \end{align} and~$y$ and~$y_-$ be the {\color{black}supremum of~$\|Y_k\|_2$ and~$\|Y_k^{-1}\|_2$} over~$k$, respectively, i.e., {\color{black} \begin{align} y&=\sup_{k}\left\|Y_k\right\|_2,\label{parameter5}\\ y_-&=\sup_{k}\left\|Y_k^{-1}\right\|_2.\label{parameter6} \end{align} }Note that the existence of the limits, $A_{\infty}$ and $Y_{\infty}$, will be clear in the following lemmas. Moreover, we define two constants,~$\sigma$, and,~$\gamma_1$, through the following two lemmas, which are related to the convergence of~$A$ and~$Y_\infty$. \begin{lem}\label{lem2} (Nedic \textit{et al}.~\cite{opdirect_Nedic}) Let Assumption \ref{asp2} hold. Consider~$Y_k$ and its limit~$Y_\infty$ as defined before. There exist~$0<\gamma_1<1$ and~$0<T<\infty$ such that for all~$k$ \begin{align}\label{DkDinfty1} \left\|Y_k-Y_\infty\right\|_2\leq T\gamma_1^{k}. \end{align} \end{lem} {\color{black} \begin{lem}\label{lem3} Let Assumption \ref{asp2} hold. Consider~$Y_\infty$ in Eq.~\eqref{yinf_eq} with~$A$ being the column-stochastic matrix used in Eq. \eqref{alg1_matrix}. For any~$\mb{a}\in\mbb{R}^{np}$, define~$\overline{\mb{a}}=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\mb{a}$. Then, there exists~$0<\sigma<1$ such that for all~$k$ \begin{align}\label{sigma_eq} \left\|A\mb{a}-Y_\infty\overline{\mb{a}}\right\|\leq\sigma\left\|\mb{a}-Y_\infty\overline{\mb{a}}\right\|. \end{align} \end{lem} \begin{proof} First note that~$A=\underline{A}\otimes I_p$. Since~$\underline{A}$ is irreducible, column-stochastic with positive diagonals, from Perron-Frobenius theorem we note that~$\rho(\underline{A})=1$, every eigenvalue of~$\underline{A}$ other than~$1$ is strictly less than~$\rho(\underline{A})$, and~$\boldsymbol{\pi}$ is a strictly positive (right) eigenvector corresponding to the eigenvalue of~$1$ such that~$\mb{1}_n^\top\boldsymbol{\pi} = 1$; thus~$\lim_{k\rightarrow\infty} \underline{A}^k = \boldsymbol{\pi}\mb{1}_n^\top$. Recalling Eq.~$\eqref{A}$, we have $A_{\infty}=\lim_{k\rightarrow\infty}{A^k}=\lim_{k\rightarrow\infty}{(\underline{A}\otimes I_p)^k}=(\lim_{k\rightarrow\infty}{\underline{A}^k})\otimes I_p=(\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p.$ It follows that: \begin{align} AA_{\infty} &= (\underline{A}\otimes I_p)\Big((\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p\Big) = (\underline{A}\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p = A_{\infty}; \nonumber \\ A_{\infty}A_{\infty} &= \Big((\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p\Big)\Big((\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p\Big), \nonumber\\ &=(\boldsymbol{\pi}\mb{1}_n^\top\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p = A_{\infty}. \nonumber \end{align} Thus~$AA_{\infty}-A_{\infty}A_{\infty}$ is a zero matrix. It can also be verified that~$\frac{1}{n}Y_\infty(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)=A_\infty$. Based on the discussion above, we have \begin{eqnarray} A\mb{a}-Y_\infty\overline{\mb{a}}=(A-A_{\infty})(\mb{a}-A_{\infty}\mb{a})=(A-A_{\infty}) (\mb{a}-Y_\infty\overline{\mb{a}}) \nonumber. \end{eqnarray} Next we note that~$$\rho(A-A_{\infty})=\rho\Big((\underline{A}-\boldsymbol{\pi}\mb{1}_n^\top)\otimes I_p\Big)=\rho(\underline{A}-\boldsymbol{\pi}\mb{1}_n^\top)<1,$$ and there exists a matrix norm such that~$\| A-A_{\infty}\| <1$ with a compatible vector norm,~$\|\cdot\|$, see~\cite{hornjohnson:13}: Chapter~5 for details, i.e., \begin{eqnarray} \left\|A\mb{a}-Y_{\infty}\overline{\mb{a}}\right\| \leq \|A-A_{\infty}\| \left\|\mb{a}-Y_{\infty}\overline{\mb{a}}\right\|, \end{eqnarray} and the lemma follows with~$\sigma=\|A-A_{\infty}\|$. \end{proof} } Based on the above notation, we finally denote~$\mb{t}_k$,~$\mb{s}_k\in\mbb{R}^3$, and~$G$,~$H_k\in\mbb{R}^{3\times3}$, for all~$k$ as {\color{black} \begin{align}\label{t} \mb{t}_k&=\left[ \begin{array}{c} \left\|\mb{x}_k-Y_\infty\overline{\mb{x}}_k\right\| \\ \left\|\overline{\mb{x}}_k-\mb{z}^*\right\|_2 \\ \left\|\mb{w}_k-Y_\infty\mb{g}_k\right\| \end{array} \right], \qquad\mb{s}_k=\left[ \begin{array}{cc} \left\|\mb{x}_k\right\|_2 \\ 0\\ 0 \end{array} \right], \notag\\ G&=\left[ \begin{array}{ccc} \sigma & 0 &\alpha \\ \alpha cly_- & \eta & 0\\ cd\epsilon ly_-(\tau+\alpha lyy_-) & \alpha d\epsilon l^2yy_- & \sigma+\alpha cd\epsilon ly_- \end{array} \right], \notag\\ H_k&=\left[ \begin{array}{ccc} 0 & 0 & 0\\ \alpha ly_-T\gamma_1^{k-1} & 0 & 0\\ (\alpha ly+2)d\epsilon ly_-^2T\gamma_1^{k-1} & 0 & 0 \end{array} \right]. \end{align} }We now state a key relation of this paper. \begin{lem}\label{thm1} Let the directed graph be strongly-connected and the optimal solution of Problem P1 exist (Assumption \ref{asp2} and \ref{asp3}). Let~$\mb{t}_k$,~$\mb{s}_k$,~$G$, and~$H_k$ be defined in Eq.~\eqref{t}, in which~$\mb{x}_k$ is the sequence generated by ADD-OPT, Eq.~\eqref{alg1_matrix}, over~$k$. Under the smooth and strong-convexity assumptions (Assumption \ref{asp1}), we have~$\mb{t}_k$,~$\mb{s}_k$,~$G$, and~$H_k$ satisfy the following linear relation, \begin{align}\label{thm1_eq} \mb{t}_k\leq G\mb{t}_{k-1}+H_{k-1}\mb{s}_{k-1}. \end{align} \end{lem} \begin{proof} See Section~\ref{s5}. \end{proof} We leave the complete proof to Section \ref{s5}, with the help of several auxiliary relations in Section \ref{s4}. Note that Eq.~\eqref{thm1_eq} provides a linear iterative relation between~$\mb{t}_k$ and~$\mb{t}_{k-1}$ with matrices,~$G$ and~$H_k$. Thus, the convergence of~$\mb{t}_k$ is fully determined by~$G$ and~$H_k$. More specifically, if we want to prove linear convergence of~$\|\mb{t}_k\|_2$ to zero, it is sufficient to show that~$\rho(G)<1$, where~$\rho(\cdot)$ denotes the spectral radius, as well as the linear decaying of~$H_k$, which is straightforward since~$0<\gamma_1<1$. In Lemma \ref{lem_G}, we first show that with appropriate step-size, the spectral radius of~$G$ is less than~$1$. Afterwards, in Lemma~\ref{lem_GH}, we study the convergence properties of the matrices involving~$G$ and~$H_k$. \begin{lem}\label{lem_G} Consider the matrix~$G$ defined in Eq.~\eqref{t} as a function of the step-size,~$\alpha$, denoted in this lemma as~$G_\alpha$ to motivate this dependence. It follows that~$\rho(G_\alpha)<1$ if the step-size,~$\alpha\in(0,\overline{\alpha})$, where \begin{color}{black} \begin{align}\label{alpha_ub} \overline{\alpha}=\min\left\{\frac{\sqrt{\Delta^2+4ns(1-\sigma)^2cd\epsilon l^2yy_-^2(l+ns)}-\Delta}{2cd\epsilon l^2yy_-^2(l+ns)},\frac{1}{nl}\right\}, \end{align} and~$\Delta=nscd\epsilon ly_-(1-\sigma+\tau)$, where~$c$ and~$d$ are the constants from the equivalence of~$\|\cdot\|$ defined in Lemma~\ref{lem3} and~$\|\cdot\|_2$. \end{color} \end{lem} \begin{proof} First, if~$\alpha<\frac{1}{nl}$ then~$\eta=1-\alpha ns,$ since~$l\geq s$ (see e.g.,~\cite{opt_literature0}: Chapter 3 for details). When~$\alpha=0$, we have that {\color{black} \begin{align} G_0&=\left[ \begin{array}{ccc} \sigma & 0 & 0 \\ 0 & 1 & 0\\ cd\epsilon l\tau y_- & 0 & \sigma \end{array} \right] ,\end{align} } the eigenvalues of which are~$\sigma$,~$\sigma$, and~$1$. Hence,~$\rho(G_0)=1$. We now consider how the eigenvalue of~$1$ is changed if we slightly increase~$\alpha$ from~$0$. Let~$\mc{P_{G_\alpha}}(q)=\mbox{det}(qI_n-G_\alpha)$, i.e., the characteristic polynomial of~$G_\alpha$. Setting~$\mbox{det}(qI_n-G_\alpha)=0$, we get the following equation. {\color{black} \begin{align}\label{det} ((q-\sigma)^2-\alpha cd\epsilon ly_-(q-\sigma))(q-1+n\alpha s)-\alpha^3cd\epsilon l^3 yy_-^2&\nonumber\\ -\alpha(q-1+n\alpha s)(cd\epsilon l\tau y_-+\alpha(cd\epsilon l^2yy_-^2))=0&. \end{align} }Since we have already shown that~$1$ is one of the eigenvalues of~$G_0$, Eq.~\eqref{det} holds when~$q=1$ and~$\alpha=0$. By taking the derivative on both sides of Eq.~\eqref{det}, with~$q=1$ and~$\alpha=0$, we obtain that{\color{black}~$\frac{d q}{d\alpha}|_{\alpha=0,q=1}=-ns<0$.} This leads to the fact that when~$\alpha$ slightly increases from~$0$,~$\rho(G_\alpha)<1$ since the eigenvalues are continuous functions of the parameters of a matrix. We next calculate all possible values of~$\alpha$ for which~$G_\alpha$ has an eigenvalue of~$1$. Let~$q=1$ in Eq.~\eqref{det} and solve for the step-size,~$\alpha$; we obtain three solutions: ~$\alpha_1=0$,~$\alpha_2<0$, and{\color{black}~$$\alpha_3=\frac{\sqrt{\Delta^2+4ns(1-\sigma)^2cd\epsilon l^2yy_-^2(l+ns)}-\Delta}{2cd\epsilon l^2yy_-^2(l+ns)}>0.$$}Since there are no other values of~$\alpha$ with which~$G_\alpha$ has an eigenvalue of $1$, all eigenvalues of~$G_\alpha$ are less than~$1$, i.e.,~$\rho(G_\alpha)<1$, when~$\alpha\in(0,\overline{\alpha})$. \end{proof} \noindent We note that~$\bar{\alpha}$ depends on the global knowledge and it may not be possible to precisely compute it in a distributed fashion. However, this value may be estimated as we will show in Section~\ref{s6}, see e.g.,~\cite{EXTRA}, for a similar approach. \begin{lem}\label{lem_GH} With the step-size,~$\alpha\in(0,\overline{\alpha})$, where~$\overline{\alpha}$ is defined in Eq.~\eqref{alpha_ub}, the following statements hold:~$\forall k$, \begin{enumerate}[label=(\alph*)] \item there exists~$0<\gamma_1<1$ and~$0<\Gamma_1<\infty$, where~$\gamma_1$ is defined in Eq.~\eqref{DkDinfty1}, such that $$\left\|H_k\right\|_2=\Gamma_1\gamma_1^k;$$ \item there exists~$0<\gamma_2<1$ and~$0<\Gamma_2<\infty$, such that $$\left\|G^k\right\|_2\leq\Gamma_2\gamma_2^k;$$ \item let~$\gamma=\max\{\gamma_1,\gamma_2\}$ and~$\Gamma=\Gamma_1\Gamma_2/\gamma$, such that for all~$0\leq r\leq k-1$, $$\left\|G^{k-r-1}H_r\right\|_2\leq\Gamma\gamma^k.$$ \end{enumerate} \end{lem} \begin{proof}~ \begin{enumerate}[label=(\alph*)] \item This can be verified according to Eq.~\eqref{t} and by letting {\color{black} $$\Gamma_1=\frac{1}{\gamma_1}\sqrt{(\alpha ly_-T)^2+(\alpha yl+2)^2(d \epsilon ly_-^2T)^2}.$$ } {\color{black} \item Note that $\rho(G)<1$ when $\alpha\in(0,\overline{\alpha})$. Therefore, the value of some matrix norm of~$G$, denoted by $\gamma_2$, is strictly less than 1. Since all matrix norms are equivalent, we have $\|G^k\|_2\leq\Gamma_2\gamma_2^{k}$, for some positive constant $\Gamma_2$. } \item The proof of (c) is achieved by combining (a) and (b). \end{enumerate} \end{proof} {\color{black} \begin{lem}\label{lem_polyak} (Polyak~\cite{polyak1987introduction}) If nonnegative sequences ~$\{v_k\}$,~$\{u_k\}$,~$\{b_k\}$ and~$\{c_k\}$ are such that ~$\sum_{k=0}^{\infty} b_k < \infty$,~$\sum_{k=0}^{\infty} c_k < \infty$ and $$v_{k+1} \le (1+b_k)v_k - u_k + c_k, \quad \forall t\ge 0,$$ then~$\{v_k\}$ converges and~$\sum_{k=0}^{\infty} u_k < \infty$. \end{lem} } We now present the main result of this paper in Theorem~\ref{main_result}, which shows the linear convergence rate of ADD-OPT. \begin{theorem}\label{main_result} Let the Assumptions \ref{asp2}-\ref{asp3} hold. With the step-size,~$\alpha\in(0,\overline{\alpha})$, where~$\overline{\alpha}$ is defined in Eq.~\eqref{alpha_ub}, the sequence,~$\{\mb{z}_k\}$, generated by ADD-OPT, converges exactly to the unique optimizer,~$\mb{z}^*$, at a linear rate, i.e., {\color{black}there exist some positive constant~$M>0$, such that for any~$k$, \begin{align} \left\|\mb{z}_k-\mb{z}^*\right\|_2\leq M(\gamma+\xi)^k, \end{align} where~$\gamma$ is used in Lemma \ref{lem_GH}(c) and~$\xi$ is a arbitrarily small constant. } \end{theorem} \begin{proof} We write Eq.~\eqref{thm1_eq} recursively, leading to \begin{align}\label{thm2_eq1} \mb{t}_k\leq&G^k\mb{t}_0+\sum_{r=0}^{k-1}G^{k-r-1}H_r\mb{s}_r. \end{align} By taking the norm on both sides of Eq.~\eqref{thm2_eq1} and considering Lemma \ref{lem_GH}, we obtain that \begin{align}\label{thm2_eq2} \left\|\mb{t}_k\right\|_2\leq&\left\|G^k\right\|_2\left\|\mb{t}_0\right\|_2+\sum_{r=0}^{k-1}\left\|G^{k-r-1}H_r\right\|_2\left\|\mb{s}_r\right\|_2,\nonumber\\ \leq&\Gamma_2\gamma_2^k\left\|\mb{t}_0\right\|_2+\sum_{r=0}^{k-1}\Gamma\gamma^k\left\|\mb{s}_r\right\|_2, \end{align} in which we can bound~$\|\mb{s}_r\|_2$ as \begin{align} \left\|\mb{s}_r\right\|_2\leq&\left\|\mb{x}_r-Y_\infty\overline{\mb{x}}_r\right\|_2+\left\|Y_\infty\right\|_2\left\|\overline{\mb{x}}_r-\mb{z}^*\right\|_2+\left\|Y_\infty\right\|_2\left\|\mb{z}^*\right\|_2\nonumber,\\ \leq&(c+y)\left\|\mb{t}_r\right\|_2+y\left\|\mb{z}^*\right\|_2. \end{align} Therefore, we have that for all~$k$ \begin{align}\label{thm2_eq3} \left\|\mb{t}_k\right\|_2\leq&\bigg{(}\Gamma_2\|\mb{t}_0\|_2+\Gamma(c+y)\sum_{r=0}^{k-1}\|\mb{t}_r\|_2+\Gamma yk\|\mb{z}^*\|_2\bigg{)}\gamma^k. \end{align} {\color{black} Denote~$v_k=\sum_{r=0}^{k-1}\|\mathbf{t}_r\|_2$,~$s_k=\Gamma_2\|\mathbf{t}_0\|_2+\Gamma yk\|\mb{z}^*\|_2$, and~$b=\Gamma (c+y)$, then Eq. \eqref{thm2_eq3} can be written as \begin{align}\label{thm2_newEq} \|\mathbf{t}_{k}\|_2 = v_{k+1} - v_k\leq(s_k + bv_k)\gamma^k, \end{align} which implies that~$v_{k+1} \le (1+b\gamma^k)v_k + s_k\gamma^k$. Applying Lemma~\ref{lem_polyak} with~$b_k = b\gamma^k$ and~$c_k = s_k\gamma^k$ (here~$u_k=0$), we have that~$v_k$ converges\footnote{In order to apply Lemma~\ref{lem_polyak}, we need to show that~$\sum_{k=0}^\infty s_k\gamma^k<\infty$, which follows from the fact that~$\lim_{k\rightarrow\infty}\frac{s_{k+1}\gamma^{k+1}}{s_{k}\gamma^{k}}=\gamma<1.$}. and therefore is bounded. By Eq.~\eqref{thm2_newEq},~$\forall \mu \in (\gamma,1)$ we have \begin{equation} \lim_{k\rightarrow\infty}\frac{\|\mathbf{t}_{k}\|_2}{\mu^k} \leq \lim_{k\rightarrow\infty}\frac{(s_k + bv_k)\gamma^k}{\mu^k}=0. \end{equation} Therefore,~$\|\mathbf{t}_{k}\|_2=O(\mu^k)$. In other words, there exists some positive constant~$\Phi$ such that for all~$k$, we have: \begin{align}\label{thm2_eq6} \left\|\mb{t}_k\right\|_2\leq&\Phi(\gamma+\xi)^{k}, \end{align} where~$\xi$ is a arbitrarily small constant. Moreover,~$\left\|\mb{z}_k-\mb{z}^*\right\|_2$ and~$\left\|\mb{t}_k\right\|_2$ satisfy the relation that \begin{align}\label{thm2_eq7} \left\|\mb{z}_k-\mb{z}^*\right\|_2\leq&\left\|Y_k^{-1}\mb{x}_k-Y_k^{-1}Y_\infty\overline{\mb{x}}_k\right\|_2+\left\|Y_k^{-1}Y_\infty\mb{z}^*-\mb{z}^*\right\|_2\nonumber\\ &+\left\|Y_k^{-1}Y_\infty\overline{\mb{x}}_k-Y_k^{-1}Y_\infty\mb{z}^*\right\|_2,\nonumber\\ \leq&y_-(c+y)\left\|\mb{t}_k\right\|_2+y_-T\gamma_1^{k}\left\|\mb{z}^*\right\|_2, \end{align} where in the second inequality we use the relation~$$\|Y_{k}^{-1}Y_\infty-I_{np}\|_2\leq\|Y_{k}^{-1}\|_2\|Y_\infty-Y_{k}\|_2\leq y_-T\gamma_1^{k},$$ achieved from Eq. \eqref{DkDinfty1}. By combining Eqs.~\eqref{thm2_eq6} and \eqref{thm2_eq7}, we obtain that \begin{align} \left\|\mb{z}_k-\mb{z}^*\right\|_2\leq&\Big(y_-(c+y)\Phi+y_-T\|\mb{z}^*\|_2\Big)(\gamma+\xi)^k,\nonumber \end{align} where $\xi$ is a arbitrarily small constant. The proof of theorem is completed by letting~$M=y_-(c+y)\Phi+y_-T\|\mb{z}^*\|_2$. } \end{proof} \noindent Theorem \ref{main_result} shows the linear convergence rate of ADD-OPT. Although ADD-OPT works for a small enough step-size, how small is sufficient may require some estimation of the upper bound, which we discuss this in Section~\ref{s6}. This notion of sufficiently small step-sizes is not uncommon in the literature, see e.g.,~\cite{uc_Nedic, opdirect_Nedic}. Next, each agent must agree on the same value of step-size that may be pre-programmed to avoid implementing an agreement protocol. We now prove Lemma \ref{thm1} in Sections \ref{s4} and \ref{s5}. \section{Auxiliary Relations}\label{s4} We provide several basic relations in this section, which will help the proof of Lemma \ref{thm1}. Lemma \ref{w-x-} derives iterative equations that govern the average sequences,~$\overline{\mb{x}}_k$ and~$\overline{\mb{w}}_k$. Lemma \ref{yy-} gives inequalities that are direct consequences of Eq.~\eqref{DkDinfty1}. Lemma~\ref{gd_approach} can be found in the standard optimization literature, see e.g.,~\cite{opt_literature0}. It states that if we perform a gradient-descent step with a fixed step-size for a smooth, strongly-convex function, then the distance to optimizer shrinks by at least a fixed ratio. \begin{lem}\label{w-x-} Recall $\overline{\mb{x}}_k$ from Eq.~\eqref{eqxb} and~$\overline{\mb{w}}_k$ from Eq.~\eqref{eqwb}. The following equations hold for all~$k$, \begin{enumerate}[label=(\alph*)] \item $\overline{\mb{w}}_k=\mb{g}_k$; \item $\overline{\mb{x}}_{k+1}=\overline{\mb{x}}_k-\alpha\mb{g}_k$. \end{enumerate} \end{lem} \begin{proof} Since~$A$ is column-stochastic, satisfying~$(\mb{1}_n^\top\otimes I_p) A=\mb{1}_n^\top\otimes I_p$, we obtain that \begin{align} \overline{\mb{w}}_k&=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\left(A\mb{w}_{k-1}+\nabla\mb{f}_{k}-\nabla\mb{f}_{k-1}\right),\nonumber\\ &=\overline{\mb{w}}_{k-1}+\mb{g}_k-\mb{g}_{k-1}.\nonumber \end{align} Do this recursively, and we have that \begin{align} \overline{\mb{w}}_k=\overline{\mb{w}}_{0}+\mb{g}_k-\mb{g}_{0}.\nonumber \end{align} Recall that we have the initial condition that~$\mb{w}_0=\nabla\mb{f}_0$, which is equivalent to~$\overline{\mb{w}}_0=\mb{g}_0$. Hence, we achieve the result of (a). The proof of (b) is obtained by the following derivation, \begin{align} \overline{\mb{x}}_{k+1}&=\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\left(A\mb{x}_{k}-\alpha\mb{w}_k\right)\nonumber\\ &=\overline{\mb{x}}_k-\alpha\overline{\mb{w}}_k,\nonumber\\ &=\overline{\mb{x}}_k-\alpha\mb{g}_k,\nonumber \end{align} where the last equation uses the result of (a). \end{proof} \begin{lem}\label{yy-} Recall Lemma~\ref{lem2},~$Y_k$ from Eq.~\eqref{Yk}, and $Y_\infty$ from Eq.~\eqref{yinf_eq}. The following inequalities hold for all~$k\geq 1$, \begin{enumerate}[label=(\alph*)] \item {\color{black}$\left\|Y_{k-1}^{-1}Y_\infty-I_{np}\right\|_2\leq y_-T\gamma_1^{k-1}$;} \item $\left\|Y_{k}^{-1}-Y_{k-1}^{-1}\right\|_2\leq 2y_-^2T\gamma_1^{k-1}$, \end{enumerate} where~$y_-$ is defined in Eq.~\eqref{parameter6}. \end{lem} \begin{proof} By considering Eq.~\eqref{DkDinfty1}, it follows that {\color{black} \begin{align} \left\|Y_{k-1}^{-1}Y_\infty-I_{np}\right\|_2\leq\left\|Y_{k-1}^{-1}\right\|_2\left\|Y_\infty-Y_{k-1}\right\|_2\leq y_-T\gamma_1^{k-1}.\nonumber \end{align} } The proof of (b) follows by \begin{align} \left\|Y_{k}^{-1}-Y_{k-1}^{-1}\right\|_2&\leq\left\|Y_{k-1}^{-1}\right\|_2\left\|Y_{k-1}-Y_{k}\right\|_2\left\|Y_{k}^{-1}\right\|_2,\nonumber\\ &\leq 2y_-^2T\gamma_1^{k-1},\nonumber \end{align} which completes the proof. \end{proof} \begin{lem}\label{gd_approach} (Bubeck~\cite{opt_literature0}) Let Assumption \ref{asp1} hold for the objective functions,~$f_i(\mb{z})$, in Problem P1, and let~$s$ and~$l$ be the strong-convexity and Lipschitz-continuity constants, respectively. For any~$\mb{z}\in\mbb{R}^p$, define~$\mb{z}_+=\mb{z}-\alpha\nabla \mb{f}(\mb{z})$, {\color{black}where~$0<\alpha<\frac{2}{nl}$. Then~$$\left\|\mb{z}_+-\underline{\mb{z}}^*\right\|_2\leq\eta\left\|\mb{z}-\underline{\mb{z}}^*\right\|_2,$$ where~$\eta=\max\left(\left|1-\alpha nl\right|,\left|1-\alpha ns\right|\right)$.} \end{lem} \section{Convergence Analysis}\label{s5} We now provide the proof of Lemma \ref{thm1}. We will bound~$\|\mb{x}_k-Y_\infty\overline{\mb{x}}_k\|$,~$\|\overline{\mb{x}}_k-\mb{z}^*\|_2$, and~$\|\mb{w}_k-Y_\infty\mb{g}_k\|$, linearly in terms of their past values, i.e.,~$\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\|$,~$\|\overline{\mb{x}}_{k-1}-\mb{z}^*\|_2$, and~$\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\|$, as well as~$\|\mb{x}_{k-1}\|_2$. The coefficients are the entries of~$G$ and~$H_{k-1}$. \textbf{Step 1:} Bound~$\|\mb{x}_k-Y_\infty\overline{\mb{x}}_k\|$. \\ According to Eq.~\eqref{alg1_ma} and Lemma \ref{w-x-}(b), we obtain that \begin{align} \left\|\mb{x}_k-Y_\infty\overline{\mb{x}}_k\right\|\leq&\left\|A\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|\nonumber\\ &+\alpha\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|. \end{align} Noticing that~$\|A\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\|\leq\sigma\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\|$ from Eq. \eqref{sigma_eq}, we have \begin{align}\label{step1} \left\|\mb{x}_k-Y_\infty\overline{\mb{x}}_k\right\|\leq&\sigma\left\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|\nonumber\\ &+\alpha\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|. \end{align} \textbf{Step 2:} Bound~$\|\overline{\mb{x}}_k-\mb{z}^*\|_2$. \\ By considering Lemma \ref{w-x-}(b), we obtain that \begin{align} \overline{\mb{x}}_k=\left[\overline{\mb{x}}_{k-1}-\alpha\mb{h}_{k-1}\right]-\alpha\left[\mb{g}_{k-1}-\mb{h}_{k-1}\right]. \end{align} Let~$\mb{x}_+=\overline{\mb{x}}_{k-1}-\alpha\mb{h}_{k-1}$, which is a (centralized) gradient-descent step with respect to the global objective function in Problem P1. Therefore, from Lemma~\ref{gd_approach}, \begin{align} \left\|\mb{x}_+-\mb{z}^*\right\|_2\leq\eta\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2. \end{align} From the Lipschitz-continuity, Assumption \ref{asp1}(a), we obtain {\color{black} \begin{align} \left\|\mb{g}_{k-1}-\mb{h}_{k-1}\right\|_2&\leq\left\|\frac{1}{n}(\mb{1}_n\mb{1}_n^\top)\otimes I_p \right\|_2l\left\|\mb{z}_{k-1}-\overline{\mb{x}}_{k-1}\right\|_2. \end{align} } Therefore, it follows that \begin{align}\label{s2_1} \left\|\overline{\mb{x}}_k-\mb{z}^*\right\|_2&\leq\left\|\mb{x}_+-\mb{z}^*\right\|_2+\alpha\left\|\mb{g}_{k-1}-\mb{h}_{k-1}\right\|_2,\nonumber\\ &\leq\eta\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2+\alpha l\left\|\mb{z}_{k-1}-\overline{\mb{x}}_{k-1}\right\|_2. \end{align} From Eq.~\eqref{alg1_mc} and Lemma \ref{yy-}(a), it follows that {\color{black} \begin{align}\label{s2_2} \left\|\mb{z}_{k-1}-\overline{\mb{x}}_{k-1}\right\|_2 \leq&\left\|Y_{k-1}^{-1}\left(\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right)\right\|_2\nonumber\\ &+\left\|\left(Y_{k-1}^{-1}Y_\infty-I_{np}\right)\overline{\mb{x}}_{k-1}\right\|_2,\nonumber\\ \leq&y_-\left\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|_2\nonumber\\ &+y_-T\gamma_1^{k-1}\left\|\mb{x}_{k-1}\right\|_2, \end{align} }where in the second inequality we also make use of the relation~$\|\overline{\mb{x}}_{k-1}\|_2\leq\|\mb{x}_{k-1}\|_2$. By substituting Eq.~\eqref{s2_2} into Eq.~\eqref{s2_1}, we obtain that \begin{align}\label{step2} \left\|\overline{\mb{x}}_k-\mb{z}^*\right\|_2\leq&\alpha cly_-\left\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|+\eta\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2\nonumber\\ &+\alpha ly_-T\gamma_1^{k-1}\left\|\mb{x}_{k-1}\right\|_2. \end{align} \textbf{Step 3:} Bound~$\|\mb{w}_k-Y_\infty\mb{g}_k\|$. \\ According to Eq.~\eqref{alg1_md}, we have \begin{align} \left\|\mb{w}_k-Y_\infty\mb{g}_k\right\|\leq&\left\|A\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|\nonumber\\ &+\left\|\left(\nabla \mb{f}_k-\nabla \mb{f}_{k-1}\right)-\left(Y_\infty\mb{g}_k-Y_\infty\mb{g}_{k-1}\right)\right\|.\nonumber \end{align} With Lemma \ref{w-x-}(a) and Eq. \eqref{sigma_eq}, we obtain that \begin{align} \left\|A\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|&=\left\|A\mb{w}_{k-1}-Y_\infty\overline{\mb{w}}_{k-1}\right\|,\nonumber\\ &\leq\sigma\left\|\mb{w}_{k-1}-Y_\infty\overline{\mb{w}}_{k-1}\right\|. \end{align} It follows from the definition of~$\mb{g}_k$ that {\color{black} \begin{align} &\left\|\left(\nabla \mb{f}_k-\nabla \mb{f}_{k-1}\right)-\left(Y_\infty\mb{g}_k-Y_\infty\mb{g}_{k-1}\right)\right\|_2\nonumber\\ &=\left\|\left(I_{np}-\frac{1}{n}Y_\infty(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\right)\left(\nabla\mb{f}_k-\nabla\mb{f}_{k-1}\right)\right\|_2. \end{align} } Since~$\frac{1}{n}Y_\infty(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)=A_\infty$, we obtain that \begin{align} \left\|\left(\nabla \mb{f}_k-\nabla \mb{f}_{k-1}\right)-\left(Y_\infty\mb{g}_k-Y_\infty\mb{g}_{k-1}\right)\right\|_2\leq&\epsilon l\left\|\mb{z}_k-\mb{z}_{k-1}\right\|_2,\nonumber \end{align} where we use the Lipschitz-continuity, Assumption \ref{asp1}(a). Therefore, we have \begin{align}\label{s3_1} \left\|\mb{w}_k-Y_\infty\mb{g}_k\right\|\leq&\sigma\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|\nonumber\\ &+d\epsilon l\left\|\mb{z}_k-\mb{z}_{k-1}\right\|_2. \end{align} We now bound~$\|\mb{z}_k-\mb{z}_{k-1}\|_2$. Note that {\color{black} \begin{align} \left\|\mb{h}_{k-1}\right\|_2&=\left\|\frac{1}{n}(\mb{1}_n\otimes I_p)(\mb{1}_n^\top\otimes I_p)\nabla\mb{f}(\overline{\mb{x}}_{k-1})\right\|_2 \nonumber\\ &\leq l\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2. \end{align} } As a result, we have \begin{align} \left\|Y_k^{-1}\mb{w}_{k-1}\right\|_2\leq&\left\|Y_k^{-1}\left(\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right)\right\|_2 \nonumber\\ &+\left\|Y_k^{-1}Y_\infty\mb{h}_{k-1}\right\|_2\nonumber\\ &+\left\|Y_k^{-1}Y_\infty\left(\mb{g}_{k-1}-\mb{h}_{k-1}\right)\right\|_2,\nonumber\\ \leq&y_-\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|_2 \nonumber\\ &+y_-yl\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2\nonumber\\ &+y_-yl\left\|\mb{z}_{k-1}-\overline{\mb{x}}_{k-1}\right\|_2,\nonumber\\ \leq&y_-\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|_2 \nonumber\\ &+y_-yl\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2\nonumber\\ &+y_-^2yl\left\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|_2\nonumber\\ &+y_-^2ylT\gamma_1^{k-1}\left\|\mb{x}_{k-1}\right\|_2, \end{align} where the last inequality holds due to Eq.~\eqref{s2_2}. With the upper bound of~$\|Y_k^{-1}\mb{w}_{k-1}\|_2$ provided in the preceding relation and {\color{black}the equality that~$(A-I_{np})Y_\infty\overline{\mb{x}}_{k-1}=\mb{0}_n$,} we can bound~$\|\mb{z}_k-\mb{z}_{k-1}\|_2$ as follows. \begin{align} \left\|\mb{z}_k-\mb{z}_{k-1}\right\|_2\leq&\left\|Y_k^{-1}\left(\mb{x}_{k}-\mb{x}_{k-1}\right)\right\|_2\nonumber\\ &+\left\|\left(Y_k^{-1}-Y_{k-1}^{-1}\right)\mb{x}_{k-1}\right\|_2,\nonumber\\ \leq&\left\|Y_k^{-1}\left(A-I_{np}\right)\mb{x}_{k-1}\right\|_2+\alpha\left\|Y_k^{-1}\mb{w}_{k-1}\right\|_2\nonumber\\ &+\left\|Y_k^{-1}-Y_{k-1}^{-1}\right\|_2\left\|\mb{x}_{k-1}\right\|_2,\nonumber\\ \leq&(y_-\tau+\alpha y_-^2yl)\left\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|_2\nonumber\\ &+\alpha y_-\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|_2\nonumber\\ &+\alpha y_-yl\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2\nonumber\\ &+(\alpha yl+2)y_-^2T\gamma_1^{k-1}\left\|\mb{x}_{k-1}\right\|_2.\label{s3_2} \end{align} By substituting Eq.~\eqref{s3_2} in Eq.~\eqref{s3_1}, we obtain that \begin{align}\label{step3} \left\|\mb{w}_k-Y_\infty\mb{g}_k\right\|\leq&(cd\epsilon l\tau y_-+\alpha cd\epsilon l^2yy_-^2)\left\|\mb{x}_{k-1}-Y_\infty\overline{\mb{x}}_{k-1}\right\|\nonumber\\ &+\alpha d\epsilon l^2yy_-\left\|\overline{\mb{x}}_{k-1}-\mb{z}^*\right\|_2\nonumber\\ &+(\sigma+\alpha cd\epsilon ly_-)\left\|\mb{w}_{k-1}-Y_\infty\mb{g}_{k-1}\right\|\nonumber\\ &+(\alpha yl+2)d\epsilon ly_-^2T\gamma_1^{k-1}\left\|\mb{x}_{k-1}\right\|_2. \end{align} \textbf{Step 4:} By combining Eqs.~\eqref{step1} in step 1, \eqref{step2} in step 2, and \eqref{step3} in step 3, we complete the proof. \section{Numerical Experiments}\label{s6} In this section, we analyze the performance of ADD-OPT. Our numerical experiments are based on the distributed logistic regression problem over a directed graph: \begin{align} \mb{z}^*=\underset{\mb{z}\in\mbb{R}^p}{\operatorname{argmin}}\left(\frac{\beta}{2}\|\mb{z}\|_2^2+\sum_{i=1}^n\sum_{j=1}^{m_i}\ln\left[1+\exp\left(-b_{ij}\mb{c}_{ij}^\top\mb{z}\right)\right]\right).\nonumber \end{align} Each agent~$i$ has access to~$m_i$ training examples,~$(\mb{c}_{ij},b_{ij})\in\mbb{R}^p\times\{-1,+1\}$, where~$\mb{c}_{ij}$ includes the~$p$ features of the~$j$th training example of agent~$i$ and~$b_{ij}$ is the corresponding label. This problem can be formulated in the form of Problem P1 with the local objective function,~$f_i$, being \begin{align} f_i=\frac{\beta}{2n}\|\mb{z}\|_2^2+\sum_{j=1}^{m_i}\ln\left[1+\exp\left(-\left(\mb{c}_{ij}^\top\mb{z}\right)b_{ij}\right)\right].\nonumber \end{align} In our setting, we have~$n=10$,~$m_i=10$, for all~$i$, and~$p=3$. \subsection{Convergence rate} In our first experiment, we compare the convergence rate of algorithms that solve the above distributed consensus optimization problem over directed graphs, including ADD-OPT, DEXTRA,~\cite{DEXTRA}, Gradient-Push,~\cite{opdirect_Nedic}, Directed-Distributed Gradient Descent,~\cite{D-DGD}, and the Weight Balanced-Distributed Gradient Descent,~\cite{opdirect_Makhdoumi}. The network topology is described in Fig. \ref{fig1}, where we apply the weighting strategy from Eq.~\eqref{aa}. \begin{figure}[!h] \begin{center} \noindent \includegraphics[width=2in]{network.pdf} \caption{A strongly-connected directed network.}\label{fig1} \end{center} \end{figure} The step-size used in Gradient-Push, Directed-Distributed Gradient Descent, and Weight Balanced-Distributed Gradient Descent is~$\alpha_k=1/\sqrt{k}$. The constant step-size used in DEXTRA and ADD-OPT is~$\alpha=0.3$. The convergence rates for these algorithms are shown in Fig.~\ref{fig2}. It shows that ADD-OPT and DEXTRA have a fast linear convergence rate, while other methods are sub-linear. \begin{figure}[!h] \begin{center} \noindent \includegraphics[width=3.5in]{simulation1.pdf} \caption{Convergence rates comparison over directed networks.}\label{fig2} \end{center} \end{figure} \subsection{Step-size range} We now compare ADD-OPT and DEXTRA in terms of their step-size ranges again with the weighting strategy from Eq.~\eqref{aa}. It is shown in Fig. \ref{fig3} that the greatest lower bound of DEXTRA is around~$\underline{\alpha}=0.2$. In contrast, ADD-OPT works for a sufficiently small step-size. In the given setting, we have~$\tau=1.25$,~$\epsilon=1.11$,~$y=1.96$,~$y_-=2.2$,~$l=1$, and~$\sigma<1$; resulting into~$\overline{\alpha}=\frac{\sqrt{8.7}}{9.57}$, where we choose~$c$ and~$d$ to be~$1$. It can be found in Fig. \ref{fig4} that the practical upper bound of step-size is much bigger, i.e.,~$\overline{\alpha}=1.12$. Since the computation of~$\overline{\alpha}$ is related to the global knowledge, e.g., the network topology, and the strong-convexity and Lipschitz-continuity constants, it is preferable to estimate~$\overline{\alpha}$. According to Eq.~\eqref{alpha_ub}, we have that $\overline{\alpha}\approxeq\sqrt{\frac{s(1-\sigma)^2}{\epsilon yy_-^2(l+s)l^2}}$ given that~$\epsilon y(l+s)s(1-\sigma)^2\gg(\epsilon\tau s)^2$. By estimating~$\tau=\epsilon=y=y_-=1$,~$\sigma=0.9$, and noting that~$s\leq l$, we can estimate~$\overline{\alpha}$ as~$\overline{\alpha}\approxeq\frac{1}{10l}$. \begin{figure}[!h] \begin{center} \noindent \includegraphics[width=3.5in]{simulation2.pdf} \caption{Comparison between ADD-OPT and DEXTRA in terms of step-sizes.}\label{fig3} \end{center} \end{figure} \subsection{Convergence rate vs. step-sizes} We note that the convergence rate of ADD-OPT is related to the spectral radius of matrix~$G$, i.e.,~$\rho(G)$, see Eq.~\eqref{thm1_eq}. Therefore, it is possible to achieve the best convergence rate by picking some~$\alpha$ such that the~$\rho(G)$ is minimized. \begin{figure}[!h] \begin{center} \noindent \includegraphics[width=3.5in]{simulation3.pdf} \caption{The range of ADD-OPT 's step-size.}\label{fig4} \end{center} \end{figure} In Fig. \ref{fig5}, we show the relationship between the spectral radius,~$\rho(G_\alpha)$, of~$G$, and the step-size,~$\alpha$, as well as the residual at the $200$-th iteration,~$\frac{\|\mathbf{z}_{200}-\mathbf{z}^*\|}{\|\mathbf{z}_0-\mathbf{z}^*\|}$, and~$\alpha$. We observe that the best convergence rate is achieved when~$\alpha=0.3$, at which~$\rho(G)$ is minimized. Fig.~\ref{fig5} also demonstrates our previous theoretical analysis in Lemma \ref{lem_G}, where we show that~$\rho(G)=1$, when~$\alpha=0$ or~$\alpha=\overline{\alpha}$, and~$\rho(G)<1$ for~$\alpha\in(0,\overline{\alpha})$. We further note that~$\rho(G_\alpha)<1$, when~$\alpha$ lies approximately in~$(0,0.3)$, which is our theoretical bound of the step-size. \begin{figure}[!h] \begin{center} \noindent \includegraphics[width=3.5in]{simulation4.pdf} \caption{Spectral radius,~$\rho(G_\alpha)$ and the residual at the~$200$th iteration versus~$\alpha$.}\label{fig5} \end{center} \end{figure} \subsection{Convergence rate vs graph sparsity} In our last experiment, we observe how does the convergence rate change as a function of the sparsity of the directed graph. We consider three strongly-connected directed graphs as shown in Fig.~\ref{graph}. It can be observed that the residuals decrease faster as the number of edges increases, from~$\mc{G}_a$ to~$\mc{G}_b$ to~$\mc{G}_c$, see Fig.~\ref{fig6}. This indicates faster convergence when there are more communication channels available for information exchange. \begin{figure}[!h] \centering \subfigure{\includegraphics[width=2in]{fig2_a.pdf}} \subfigure{\includegraphics[width=2in]{fig2_b.pdf}} \subfigure{\includegraphics[width=2in]{fig2_c.pdf}} \caption{Three examples of strongly-connected directed graphs.} \label{graph} \end{figure} \section{Conclusions}\label{s7} In this paper, we focus on solving the distributed optimization problem over directed graphs. The proposed algorithm, termed ADD-OPT (Accelerated Distributed Directed Optimization), can be viewed as an improvement of our recent work, DEXTRA. The proposed algorithm, ADD-OPT, achieves the best known rate of convergence for this class of problems,~$O(\mu^{k}),0<\mu<1$, given that the objective functions are strongly-convex with globally Lipschitz-continuous gradients, where~$k$ is the number of iterations. Moreover, ADD-OPT supports a wider and more realistic range of step-sizes in contrast to the existing work. In particular, we show that ADD-OPT converges for arbitrarily small (positive) step-sizes. Simulations further illustrate our results. \begin{figure}[!h] \begin{center} \noindent \includegraphics[width=3.5in]{simulation5.pdf} \caption{The range of ADD-OPT 's step-size.} \label{fig6} \end{center} \end{figure} \bibliographystyle{IEEEbib}
0f25400f40b77edfa427d6c0b2cfd785aca148a2
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{intro} Suitable models for the theory of computation and approximation are certain (quasi-)ordered sets, whose elements represent states of computation, knowledge or information, while the order abstractly describes refinement, improvement or temporal sequence. Let us briefly record the relevant order-theoretical terminology. A {\em quasi-ordered set} or {\em qoset} is a pair $Q = (X,\leq)$ with a reflexive and transitive relation $\leq$ on $X$. The dual order is denoted by $\geq$, and the dual qoset $(X,\geq)$ by $\widetilde{Q}$.\\ If $\,\leq\,$ is antisymmetric, we speak of a {\em (partial) order} and an {\em ordered set} or a {\em poset}.\\ A {\em lower set}, {\em downset} or {\em decreasing set} is a subset $Y$ that coincides with its\,\,{\em down-closure}\,\,${\downarrow\! Y}$, consisting of all $x \!\in\! X$ with $x \!\leq\! y$ for at least one $y \!\in\! Y$. The {\em up-closure} ${\uparrow\! Y}$ and {\em upper sets (upsets, increasing sets)} are defined dually. The upper\,sets form the\,{\em upper Alexandroff topology}\,\,$\alpha Q$, and the lower sets the {\em lower Alexandroff topology}\,\,$\alpha \widetilde{Q}$. A set $D\subseteq Q$ is {\em (up-)directed}, resp.\ {\em filtered} or {\em down-directed}, if every finite subset of $D$ has an upper, resp.\ lower bound in\,$D$; in particular, $D$ cannot be empty. An {\em ideal} of $D$ is a directed lower set, and a {\em filter} is a filtered upper set; for $x \!\in\! X$, the set ${\downarrow\!x} \!=\! {\downarrow\!\{ x\}}$ is the {\em principal ideal}, and ${\uparrow\!x} \!=\! {\uparrow\!\{ x\}}$ is the {\em principal filter} generated by $x$. A poset $P$ is called {\em up-complete}, {\em directed complete}, a {\em dcpo}, or a {\em cpo} if each directed subset $D$ or, equivalently, each ideal has a join, that is, a least upper bound ({\em supremum}), denoted by $\bigvee\! D$. The {\em (ideal) up-completion} of a qoset $Q$ is ${\cal I} Q$, the set of all ideals, ordered by inclusion. The {\em Scott topology} $\sigma P$ of a poset $P$ consists of all upper sets $U$ that meet any directed set having a join in\,\,$U$. Both in the mathematical and in the computer-theoretically oriented literature (see e.g.\ \cite{AJ}, \cite{Edom}, \cite{CLD}, \cite{SLG}, \cite{V}), the word {\em `domain'} represents quite diverse structures, and in order-theoretical contexts, its meaning ranges from rather general notions like dcpos to quite specific kinds of posets like $\omega$-algebraic dcpos, sometimes with additional properties. Here, we adopt the convention to call arbitrary up-complete posets {\em domains} and to speak of a {\em continuous poset} if for each element $x$ there is a least ideal having a join above\,\,$x$. Notice that our {\em continuous domains} are the {\em continuous posets} in \cite{Com} and \cite{Hoff2}, but they are the {\em domains} in \cite{CLD}, whereas our continuous posets and those in \cite{CLD} need not be up-complete. Although continuous domains usually are defined in order-theoretical terms, there exist also topological descriptions of them, for example, as sober locally super\-compact spaces (Ern\'e \cite{EABC, Emin}, Hoffmann \cite{Hoff2}, Lawson \cite{CLD, Lss}). It is one of our main purposes in the subsequent investigations to drop the completeness or sobriety hypotheses without loosing relevant results applicable to domain theory. The term {\em space} always means {\em topological space}, but extensions to arbitrary closure spaces are possible (see \cite{EABC} and \cite{Eclo}). Several classes of spaces may be characterized by certain infinite distribution laws for their lattices of open sets \cite{Eweb}. Recall that a {\em frame} or {\em locale} \cite{Jo} is a complete lattice $L$ satisfying the identity \vspace{-.5ex} $$ \textstyle{ {\rm (d)} \ x\wedge \bigvee Y = \bigvee \{ x\wedge y : y\in Y \} \vspace{-.5ex}} $$ for all $x\in L$ and $Y \subseteq L$; the dual of (d) characterizes {\em coframes}. The identity \vspace{-.5ex} $$ \textstyle{{\rm (D)} \ \bigwedge \,\{ \bigvee Y : Y \!\in {\cal Y}\} = \bigvee\bigcap {\cal Y}} \vspace{-.5ex} $$ for all collections ${\cal Y}$ of lower sets, defining {\em complete distributivity}, is much stronger. However, frames may also be defined by the identity (D) for all {\em finite} collections ${\cal Y}$ of lower sets. An up-complete meet-semilattice satisfying (d) for all directed sets (or ideals) $Y$ is called {\em meet-continuous}. Similarly, the {\em continuous lattices} in the sense of Scott \cite{Com}, \cite{Sco} are the complete lattices enjoying the identity (D) for ideals instead of lower sets. Therefore, completely distributive lattices are also called {\em super\-continuous}; alternative descriptions of complete distributivity by equations involving choice functions are equivalent to the Axiom of Choice\,\,(see\,\,\cite{Herr}). A complete lattice satisfying (D) for all collections of finitely generated lower sets is called {\em ${\cal F}$-distributive} or a {\em wide coframe}, and its dual a {\em quasitopology} or a {\em wide frame} (cf.\ \cite{Eweb}). A lattice is {\em spatial} iff it is isomorphic to a topology. All spatial lattices are wide frames. For any space $(X,{\cal S})$, the frame of open sets is ${\cal S}$, and the coframe of closed sets is denoted by ${\cal S}^c$. The closure of a subset $Y$ is denoted by $cl_{{\cal S}}Y$ or $Y^-\!$, and the interior by $int_{{\cal S}}Y\!$ or $Y^{\circ}$. The {\em specialization order} is given by \vspace{-1ex} $$ x\leq y \ \Leftrightarrow \ x\leq_{{\cal S}} y \ \Leftrightarrow \ x\in \{ y\}^- \ \Leftrightarrow \ \forall\,U\! \in {\cal S}\ (x\in U \,\Rightarrow \, y\in U). $$ It is antisymmetric iff $(X,{\cal S})$ is T$_0$, but we speak of a specialization order also in the non--T$_0$ setting. The {\em saturation} of a subset $Y$ is the intersection of all its neighborhoods, and this is the up-closure of $Y$ relative to the specialization order. In the {\em specialization qoset} $\Sigma^{-\!} (X,{\cal S}) = (X,\leq_{{\cal S}})$, the principal ideals are the point closures, and the principal filters are the cores, where the {\em core} of a point $x$ is the saturation of the singleton $\{ x\}$; the lower sets are the unions of cores, or of arbitrary closed sets, and the upper sets are the saturated sets. A topology ${\cal S}$ on $X$ is {\em compatible} with a quasi-order $\leq$ if $Q = (X,\leq)$ is the specialization qoset of $(X,{\cal S})$ or, equivalently, $\upsilon Q \subseteq {\cal S} \subseteq \alpha Q$, where $\upsilon Q$ is the {\em weak upper topology}, generated by the complements of principal ideals; the {\em weak lower topology} of $Q$ is the weak upper topology $\upsilon \widetilde{Q}$ of the order-dual $\widetilde{Q}$. Of course, in other contexts, compatibility of a topology with an order relation may have a different meaning (cf.\ \cite[VI]{CLD}). In \cite{Eweb}, we have introduced three classes of spaces that might be useful for the mathematical foundation of communication and information theory (order-theoretical notions refer to the specialization order): \begin{itemize} \item[--] {\em web spaces} have neighborhood bases of webs at each point $x$, i.e.\ unions of filtered sets each of which contains $x$, \item[--] {\em wide web spaces} have neighborhood bases of filtered sets at each point, \item[--] {\em worldwide web spaces} or {\em core spaces} have neighborhood bases of principal filters (cores) at each point. \end{itemize} As shown in \cite{Eweb}, each of these three classes of spaces may be described by an infinite distribution law for their topologies: a space is a \vspace{-1ex} \begin{itemize} \item[--] web space iff its topology is a coframe, \item[--] wide web space iff its topology is a wide coframe, \item[--] worldwide web space iff its topology is completely distributive. \end{itemize} In Section \ref{convex}, we briefly review the construction of patch spaces and some applications to web spaces, as developed in \cite{Epatch}. The patch spaces of a given space are obtained by joining its topology with a {\em cotopology} (in \cite{Lss}: \mbox{{\em complementary topology}),} that is, a topology having the dual specialization order. Useful for patch constructions are so-called {\em coselections} $\zeta$, which choose for any topology ${\cal S}$ a subbase $\zeta {\cal S}$ of a cotopology $\tau_{\zeta}{\cal S}$. The topology ${\cal S}^{\zeta}$ generated by ${\cal S} \cup \zeta {\cal S}$ is then a patch topology, and the corresponding (quasi-ordered!) {\em $\zeta$-patch space} is ${\rm P}_{\zeta}(X,{\cal S}) = (X, \leq_{{\cal S}} ,{\cal S}^{\zeta})$. As demonstrated in \cite{Epatch}, web spaces may be characterized by the property that their open sets are exactly the up-closures of the open sets in any patch space. For us, compactness does {\em not} include the Hausdorff separation axiom T$_2$. Locally compact spaces undoubtedly form one of the most important classes of topological spaces. In the non-Hausdorff setting, one has to require whole bases of compact neighborhoods at each point, because one compact neighborhood for each point would not be enough for an efficient theory. In certain concrete cases, one observes that each point of the space under consideration has even a neighborhood base consisting of {\em supercompact} sets, i.e.\ sets each open cover of which has already one member that contains them. Such spaces occur, sometimes unexpectedly, in diverse fields of mathematics\,--\,not only topological but also algebraic ones\,--\,and in theoretical computer science. Section \ref{CDS} is devoted to a closer look at such {\em locally supercompact spaces}; they are nothing but the core spaces, because the supercompact saturated sets are just the cores. These spaces have been introduced in \cite{EABC}, where the name {\em core spaces} referred to the larger class of closure spaces, and discussed further in \cite{Emin} and \cite{Eweb}; core spaces are also called {\em C-spaces} (in \cite{KL}: {\em c-spaces}), but that term has a different meaning in other contexts (e.g.\ in \cite{Kue} and \cite{Pri}). The core spaces are exactly the locally compact wide web spaces, but also the locally hypercompact web spaces, where a set is {\em hyper\-compact} if its saturation is finitely generated; while the interior operator of a web space preserves {\em finite} unions of saturated sets, the interior operator of a core space preserves {\em arbitrary} unions of saturated sets \cite{Eweb}. Moreover, the category of core spaces has a strong order-theoretical feature, being concretely isomorphic to a category of generalized quasi-ordered sets (see \cite{EABC} and Section \ref{CDS} for precise definitions and results). Core spaces share useful properties with the more restricted {\em basic spaces} or {\em B-spaces} (having a least base, which then necessarily consists of all open cores) and with the still more limited {\em Alexandroff-discrete spaces} or {\em A-spaces} (in which all cores are open) \cite{Alex}, \cite{EABC}, \cite{Emin}; but, in contrast to A- and B-spaces, core spaces are general enough to cover important examples of classical analysis. For instance, the Euclidean topology on ${\mathbb R}^n$ (which, ordered componentwise, is a continuous poset but not a domain!) is the weak(est) patch topology of the Scott topology, which makes ${\mathbb R}^n$ a core space. In Section \ref{sectorspaces}, we characterize the patch spaces of core spaces as {\em sector spaces}. These are {\em ${\uparrow}$-stable semi-qospaces} (meaning that up-closures of open sets are open, and principal ideals and principal filters are closed) with neighborhood bases of so-called {\em sectors}, a special kind of webs having least elements. The restriction of the patch functor ${\rm P}_{\zeta}$ to the category of core spaces yields a concrete isomorphism to the category of {\em $\zeta$-sector spaces}, which fulfil strong convexity and separation axioms. In particular, the {\em weak patch functor} ${\rm P}_{\upsilon}$ induces a concrete \mbox{categorical} isomorphism between core spaces and {\em fan spaces}, i.e.\ ${\uparrow}$-stable semi-qospaces in which each point has a neighborhood base of {\em fans} ${\uparrow\!u}\setminus{\uparrow\!F}$ with finite sets\,\,$F$. In Section \ref{fanspaces}, such fan spaces are investigated and characterized by diverse order-topological properties. Our considerations have useful consequences for topological aspects of domain theory, as the continuous domains, equipped with the Scott topology, are nothing but the sober core spaces, and these correspond to fan spaces that carry the {\em Lawson topology}, the weak patch topology of the Scott topology. We find alternative descriptions of such ordered spaces, including convexity properties, separation axioms and conditions on the interior operator. This enables us to generalize the characterization of continuous lattices as meet-continuous lattices whose Lawson topology is Hausdorff \cite[III--2.11]{CLD} and the Fundamental Theorem of Compact Semilattices \cite[VI--3]{CLD} to non-complete situations. Crucial is the fact that a semilattice with a compatible topology is semitopological iff it is a web space, and (locally compact) topological with small semilattices iff it is a (world) wide webspace. The category of T$_0$ core spaces and that of ordered fan spaces are not only equivalent to the category C-ordered sets, but also to the category of {\em based domains}, i.e.\ pairs consisting of a continuous domain and a basis of it (in the sense of \cite{CLD}). In the last section, we study weight and density of the spaces under consideration, using the order-theoretical description of core spaces \cite{EABC}. For example, the weight of a core space is equal to the density of any of its patch spaces, but also to the weight of the lattice of {\em closed} sets. This leads to the conclusion that the weight of a completely distributive lattice is always equal to the weight of the dual lattice. If not otherwise stated, all results are derived in a choice-free set-theoretical environment; i.e., we work in {\sf ZF} or {\sf NBG} (Zermelo--Fraenkel or Neumann--Bernays--G\"odel set theory) but not in {\sf ZFC} (i.e.\ {\sf ZF} plus Axiom of Choice). For basic categorical concepts, in particular, concrete categories, functors and isomorphisms, see Ad\'amek, Herrlich and Strecker \cite{AHS}. For relevant order-theoretical and topological definitions and facts, refer to the monograph {\it Continuous Lattices and Domains} by G.\ Gierz, K.\,H.\ Hofmann, K.\ Keimel, J.\,D.\ Lawson, M.\ Mislove, and D.\,S.\ Scott\,\,\cite{CLD}. \newpage \section{Patch spaces and web spaces} \label{convex} A {\em (quasi-)\-ordered space} is a (quasi-)\-ordered set equipped with a topology. In this elementary definition, no separation properties and no relationship between order and topology are required. However, some classical separation axioms extend to the ordered case as follows. A quasi-ordered space is a {\em lower semi-qospace} if all principal ideals are closed, an {\em upper qospace} if all principal filters are closed, and a {\em semi-qospace} if both conditions hold (these conditions mean that the quasi\-order is {\em lower semiclosed}, {\em upper semiclosed} or {\em semiclosed}, respectively, in the sense of \cite[VI-1]{CLD}). An ordered semi-qospace is a {\em semi-pospace} or {\em T$_1$-ordered\,}. A space equipped with a closed quasi-order $\leq$ (regarded as a subset of the square of the space) is called a {\em qospace}, and a {\em pospace} in case $\leq$ is a (partial) order \cite{CLD}. Alternatively, qospaces may be characterized by the condition that for $x\not\leq y$, there are open $U$ and $V$ with $x \!\in\! U$, $y \!\in\! V$, and ${\uparrow\!U} \mathop{\cap} {\downarrow\! V} = \emptyset$. Similarly, we define {\em T$_2$-ordered spaces} to be ordered spaces in which for $x \not\leq y$ there are an open upper set containing $x$ and a disjoint open lower set containing $y$; some authors call such spaces {\em strongly $T_2$-ordered} and mean by a {\em T$_2$-ordered space} a pospace (cf.\ K\"unzi\,\,\cite{Kue}, McCartan\,\,\cite{McC}). A quasi-ordered space is said to be {\em upper regular} if for each open upper set $O$ containing a point $x$, there is an open upper set $U$ and an closed upper set $B$ such that $x\in U \subseteq B \subseteq O$, or equivalently, for each closed lower set $A$ and each $x$ not in $A$, there is an open upper set $U$ and a disjoint open lower set $V$ with $x\in U$ and $A\subseteq V$. An upper regular T$_1$-ordered space is said to be {\em upper T$_3$-ordered}. {\em Lower regular} spaces are defined dually. Note the following irreversible implications: {\em compact qospace $\Rightarrow$ upper regular semi-qospace $\Rightarrow$ qospace $\Rightarrow$ semi-qospace,}\\ \indent {\em compact pospace $\,\Rightarrow$ upper\,T$_3$-ordered $\Rightarrow$\,T$_2$-ordered $\Rightarrow$ pospace $\Rightarrow$\,T$_1$-ordered$\,+T_2$.} \vspace{.5ex} \noindent For any quasi-ordered space $T = (Q,{\cal T}) = (X,\leq,{\cal T})$, \vspace{.5ex} ${\cal T}^{\,\leq} = {\cal T} \cap \alpha Q\ $ is the topology of all open upper sets (also denoted by ${\cal T}^{\,\sharp}$),\\ \indent ${\cal T}^{\,\geq} = {\cal T} \cap \alpha \widetilde{Q}\ $ is the topology of all open lower sets \hspace{.6ex}(also denoted by ${\cal T}^{\,\flat}$). We call ${\rm U} T = (X,{\cal T}^{\leq})$ the {\em upper space} and ${\rm L} T = (X, {\cal T}^{\geq})$ the {\em lower space} of\,\,$T$. A basic observation is that for lower semi-qospaces, the specialization order of ${\cal T}^{\leq}$ is $\leq$, while for upper semi-qospaces, the specialization order of ${\cal T}^{\geq}$ is $\geq$. Recall that a subset $Y$ of a qoset $Q$ is {\em (order) convex} iff it is the intersection of an upper and a lower set. A quasi-ordered space is {\em locally convex} if the convex open subsets form a base, {\em strongly convex} if its topology ${\cal T}$ is generated by ${\cal T}^{\leq} \cup {\cal T}^{\geq}$, and {\em $\zeta$-convex} if ${\cal T}$ is generated by ${\cal T}^{\leq} \mathop{\cup} \zeta ({\cal T}^{\leq})$, where $\zeta$ is a coselection (see the introduction). Specifically, $\upsilon$-convex quasi-ordered spaces are called {\em hyperconvex}. Thus, hyperconvexity means that the sets $U\setminus {\uparrow\!F}$ with $U\in {\cal T}^{\leq}$ and $F$ finite form a base. Observe that $\zeta$-convexity implies strong convexity, which in turn implies local convexity, but not conversely (counterexamples are given in\,\,\cite{Epatch}). Let $\zeta$ be any coselection. A space $(X,{\cal S})$ is said to be {\em $\zeta$-determined} if $\,{\cal S}^{\,\zeta\leq} = {\cal S}$. A map between spaces is called {\em $\zeta$-proper} if it is continuous and preimages of closed sets relative to the $\zeta$-cotopology are $\zeta$-patch closed (whence such a map is $\zeta$-patch continuous); and a map between quasi-ordered spaces is {\em lower semicontinuous} if preimages of closed lower sets are closed. In \cite{Epatch}, many examples and counterexamples concerning these notions are discussed, and the following facts are established: \vspace{-.5ex} \begin{lemma} \label{pat} The strongly convex semi-qospaces are exactly the patch spaces (of their upper spaces). The patch functor ${\rm P}_{\zeta}$ associated with a coselection $\zeta$ induces a concrete functorial isomorphism between the category of $\zeta$-determined spaces with continuous (resp.\ $\zeta$-proper) maps and that of $\zeta$-convex semi-qospaces with isotone lower semicontinuous (resp.\ continuous) maps; the inverse isomorphism is induced by the concrete upper space functor ${\rm U}$, sending a semi-qospace $(X,\leq, {\cal T})$ to $(X,{\cal T}^{\leq})$. \end{lemma} \vspace{-.5ex} Any {\em upset selection} $\zeta$, assigning to each qoset $Q$ a collection $\zeta Q$ of upper sets such that ${\uparrow\!x} = \bigcap\,\{ V\in \zeta Q : x\in V\}$ for all $x$ in $Q$, gives rise to a coselection by putting $\zeta {\cal S} = \zeta \widetilde {Q}$ for any space $(X,{\cal S} )$ with specialization qoset $Q$. If each $\zeta Q$ is a topology, we call $\zeta$ a {\em topological (upset) selection}; the largest one is $\alpha$, while the smallest one is\,\,$\upsilon$. By Lemma \ref{pat}, the {\em weak patch functor} ${\rm P}_{\upsilon}$ induces an isomorphism between the category of $\upsilon$-determined spaces and that of hyperconvex semi-qospaces. An important intermediate topological selection is $\sigma$, where $\sigma Q$ is the {\em Scott topology}, consisting of all upper sets $U$ that meet every directed subset having a least upper bound that belongs to\,\,$U$ (in arbitrary qosets, $y$ is a least upper bound of $D$ iff $D\subseteq {\downarrow\! z} \Leftrightarrow y\leq z$). The weak patch topology of $\sigma Q$ is the {\em Lawson topology} $\lambda Q = \sigma Q ^{\,\upsilon}$. We denote by $\Sigma Q$ the {\em Scott space} $(X,\sigma Q )$ and by $\Lambda Q$ the (quasi-ordered) {\em Lawson space} $(Q,\lambda Q )$, whose upper space in turn is $\Sigma Q$. Thus, all Scott spaces are $\upsilon$-determined, and all Lawson spaces are hyperconvex semi-qospaces. A quasi-ordered space $(Q,{\cal T})$ is said to be {\em upwards stable} or {\em ${\uparrow}$-stable} if it satisfies the following equivalent conditions: \vspace{-.5ex} \begin{itemize} \item[{\rm (u1)}] $O\in {\cal T}$ implies ${\uparrow\!O}\in {\cal T}$. \item[{\rm (u2)}] ${\cal T}^{\leq} = \{ {\uparrow\! O} : O\in {\cal T}\}$. \item[{\rm (u3)}] The interior of each upper set is an upper set: $int_{{\cal T}} Y\! = int_{{\cal T}^{\leq}}Y$ if $Y \!= {\uparrow\!Y}$. \item[{\rm (u4)}] The closure of each lower set is a lower set: $cl_{{\cal T}} Y\! = cl_{{\cal T}^{\leq}}Y$ if $Y \!= {\downarrow\!Y}$. \end{itemize} A {\em web} around a point $x$ in a qoset is a subset containing $x$ and with each point $y$ a common lower bound of $x$ and $y$; if $\leq$ is a specialization order, that condition means that the closures of $x$ and $y$ have a common point in the web. By a {\em web-(quasi-)ordered space} we mean an ${\uparrow}$-stable (quasi-)ordered space in which every point has a neighborhood base of webs around it. In the case of a space equipped with its specialization order, this is simply the definition of a {\em web space}. Many characteristic properties of web spaces and of web-quasi-ordered spaces are given in \cite{Eweb} and \cite{Epatch}. Note that the {\em meet-continuous dcpos} in the sense of \cite{CLD} are just those domains whose Scott space is a web space \cite{Eweb}. The following result from \cite{Epatch} underscores the relevance of web spaces for patch constructions: \begin{proposition} \label{webpatch} A space $S$ is a web space iff all its patch spaces are web-quasi-or\-dered and have the upper\,space\,\,$S$. Conversely, the strongly convex web-quasi-ordered semi-qospaces are the patch spaces of their upper spaces, and these are web spaces. For any coselection $\zeta$, the patch functor ${\rm P}_{\zeta}$ induces a concrete isomorphism between the category of web spaces and that of $\zeta$-convex, web-quasi-ordered semi-qospaces. \end{proposition} We now are going to derive an analogous result for wide web spaces; the case of worldwide web spaces (core spaces) is deferred to the next section. The notion of wide web spaces is a bit subtle: while in a web space every point has a neighborhood base consisting of {\em open} webs around it \cite{Eweb}, in a wide web space it need not be the case that any point has a neighborhood base consisting of {\em open} filtered sets; the spaces with the latter property are those which have a `dual' (Hoffmann \cite{Hoff1}), that is, whose topology is dually isomorphic to another topology; see \cite{Eweb} for an investigation of such spaces and a separating counterexample. Note the implications \vspace{.5ex} {\em completely distributive $\,\Rightarrow\,\ $ dually spatial $\,\ \ \Rightarrow\ $ wide coframe $\hspace{1.8ex}\Rightarrow\, $ coframe} \vspace{.5ex} \noindent and the corresponding (irreversible) implications for spaces: \vspace{.5ex} {\em worldwide web space $\ \ \ \Rightarrow\ $ space with dual $\,\Rightarrow\, $ wide web space $\,\Rightarrow\, $ web space.} \vspace{.5ex} \noindent It is now obvious to introduce an ordered version of wide web spaces by calling a quasi-ordered space {\em locally filtered} if each point has a base of filtered neighborhoods. \begin{theorem} \label{wideweb} A space $S$ is a wide web space iff all its patch\,spaces\,are\,locally fil\-tered and\,up\-stable with\,upper\,space\,$S$.\,The strongly\,convex, locally filtered\,\,${\uparrow}$-stable semi\-qospaces are the patch spaces of their upper spaces, and these are are wide\,web\,spaces. Each patch functor ${\rm P}_{\zeta}$ induces a concrete isomorphism between the category of wide web spaces and that of $\zeta$-convex, locally filtered, ${\uparrow}$-stable semi-qospaces. \end{theorem} \noindent {\it Proof.} Let $S \!=\! (X,{\cal S} )$ be a wide web space and $T \!=\! (X,\leq, {\cal T} )$ a patch space of\,\,$S$. By Proposition \ref{webpatch}, $T$ is ${\uparrow}$-stable, and $S$ is the upper space of $T$, i.e., ${\cal S} = {\cal T}^{\leq}$. Given $x \!\in\! O \!\in\! {\cal T}$, find $U \!\in\! {\cal S}$ and $V \!\in\! {\cal T}^{\geq}$ with $x \in U \cap V \subseteq O$, and a filtered set $D$ with $x \in W = int_{{\cal S}} D \subseteq D \subseteq U$. Then, $D\cap V$ is filtered (since $V = {\downarrow\!V}$) with $x\in W\cap V \subseteq D\cap V \subseteq U \cap V \subseteq O$, and $W\cap V \in {\cal T}$. Thus, $T$ is locally filtered. Conversely, let $T = (X, \leq, {\cal T} )$ be a strongly convex, locally filtered and ${\uparrow}$-stable semi-qospace. Then $T$ is web-quasi-ordered and, by Proposition \ref{webpatch}, a patch space of the web space $(X,{\cal T}^{\leq})$. For $x \in O \in {\cal T}^{\leq}$, there is a filtered $D\subseteq O$ with $x \in int_{{\cal T}}D$. Then ${\uparrow\!D}$ is filtered, $x \in W = {\uparrow\! int_{{\cal T}} D} \subseteq {\uparrow\!D} \!\subseteq\! O$, and by ${\uparrow}$-stability, $W \in {\cal T}^{\leq}$; thus, $(X,{\cal T}^{\leq})$ is a wide web space. The rest follows from Proposition \ref{pat}. \EP \begin{corollary} \label{webco} The strongly convex, locally filtered and ${\uparrow}$-stable T$_1$-ordered spaces are exactly the patch spaces of T$_0$ wide web spaces. \end{corollary} \section{Core spaces and C-quasi-ordered sets} \label{CDS} Strengthening the notion of compactness, we call a subset $C$ of a space {\em super\-compact} if every open cover of $C$ has a member that contains $C$; and {\em local supercompactness} means the existence of supercompact neighborhood bases at each point. As mentioned in the introduction, an equivalent condition is that each point has a neighborhood base of cores (possibly of different points!)\,--\,in other words, that we have a {\em core space}. Under the assumption of the Ultrafilter Theorem (a consequence of the Axiom of Choice), many properties of locally super\-compact spaces are shared by the more general {\em locally hypercompact spaces}, where a subset is called {\em hypercompact} if its saturation is generated by a finite subset. In view of the next proposition, proven choice-freely in \cite{EABC} and \cite{Eweb}, core spaces may be viewed as an infinitary analogue of web spaces (whence the name {\em `worldwide web spaces'}\,). \begin{proposition} \label{supertop} For a space $S = (X,{\cal S})$, the following conditions are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] $S$ is a core space. \item[{\rm (2)}] $S$ is locally supercompact. \item[{\rm (3)}] The lattice of open sets is supercontinuous (completely distributive). \item[{\rm (4)}] The lattice of closed sets is supercontinuous. \item[{\rm (5)}] The lattice of closed sets is continuous. \item[{\rm (6)}] The interior operator preserves arbitrary unions of upper sets. \item[{\rm (7)}] The closure operator preserves arbitrary intersections of lower sets. \item[{\rm (8)}] $S$ is a locally hypercompact web space. \item[{\rm (9)}] $S$ is a locally compact wide web space. \end{itemize} \end{proposition} Core\,spaces have pleasant properties. For example, on account of (6) resp.\,(7), the interior resp.\ closure operator of a core space induces a complete homomorphism from the completely distributive lattice of upper resp.\,lower sets onto the lattice of open resp.\,closed sets. In \cite{BrE}, it is shown that a nonempty product of spaces is a core space iff all factors are core spaces and all but a finite number of them are supercompact. Similarly, a nonempty product of spaces is a (wide) web space iff all factors are (wide) web spaces and all but a finite number are filtered. Assuming the Principle of Dependent Choices (another consequence of the Axiom of Choice), one can show that all core spaces have a dual (a base of filtered open sets; see \cite{Eweb}). Computationally convenient is the fact that core spaces are in bijective correspondence to so-called {\em idempotent ideal relations} or {\em C-quasi-orders}. These are not really quasi-orders but idempotent relations $R$ (satisfying $x\,R\,z \Leftrightarrow \exists\, y \,(x\,R\,y\,R\,z)$) on a set $X$ so that the sets \vspace{-1ex} $$R y = \{ x \in X : x \,R\, y\} \ \ (y\in X) $$ are ideals with respect to the {\em lower quasi-order} $\leq_{R}$ defined by \vspace{-.5ex} $$ x\leq_{R} y \ \Leftrightarrow R x \subseteq R y. $$ And $R$ is called a {\em C-order} if, moreover, $\leq_{R}$ is an order, i.e., $R y \!=\! R z$ implies $y \!=\! z$. The pair $(X,R )$ is referred to as a {\em C-(quasi-)ordered set} \cite{EABC}. Any C-order $R$ is an {\em approximating auxiliary relation} for the poset $(X,\leq_{R})$ in the sense of \cite{CLD}. For an arbitrary relation $R$ on a set $X$ and any subset $Y$ of $X$, put $$R\, Y = \{ x\in X : \exists\,y\in Y\, (x\,R\,y)\} , \ YR = \{ x\in X : \exists\,y\in Y\, (y\,R\,x)\},\vspace{-.5ex} $$ $$\ _{R}{\cal O} = \{ \,R\,Y : Y\subseteq X\} , \ {\cal O}_{R} = \{ YR : Y\subseteq X\}. \ $$ $_{R}{\cal O} $ and ${\cal O}_{R}$ are closed under arbitrary unions, and in case $R$ is an ideal relation, ${\cal O}_{R}$ is even a topology. On the other hand, if $R$ is idempotent then $_{R}{\cal O}$ consists of all rounded subsets, where a subset $Y$ is said to be {\em round(ed)} if $Y = R Y$ \cite{GK}, \cite{Lri}. Typical examples of C-orders are the {\em way-below relations} $\ll$ of {\em continuous posets} $P$, in which for any element $y$ the set \vspace{-.5ex} $$\textstyle{\ll\! y = \{ x\in P : x \ll y\} = \bigcap \,\{ D\in {\cal I} P : \bigvee\! D \mbox{ exists and } y\leq \bigvee\! D\}} $$ is an ideal (called the {\em way-below ideal of} $y$) with join $y$ (recall that ${\cal I} P$ is the set of all ideals). The next result, borrowed from \cite{EHab}, leans on the {\em interpolation property} of continuous posets, saying that their way-below relation is idempotent. \begin{lemma} \label{conti} For every continuous poset, the way-below relation is a C-order $R$ such that for directed subsets $D$ (relative to $\leq_{R}$), $y = \bigvee\! D$ is equivalent to $R y = R D$; and any C-order with that property is the way-below relation of a continuous poset. \end{lemma} \begin{example} \label{Exnotcon} {\rm The following sublattice $L$ of the plane $\mathbb{R}^2$ has a unique compatible topology $\alpha L = \upsilon L$, and the interval topology $\iota L$ is discrete \cite{Epatch}. \begin{picture}(250,150) \put(-10,120){$L = \{ a_n,\,b_n,\,c_n: n\in \omega\}$} \put(-10,100){with} \put(-10,80){$a_n = (0,-2^{-n}),$} \put(-10,60){$b_n =(2^{-n},0),$} \put(-10,40){$c_n = (2^{-n},2^{-n}).$} \put(110,0) { \begin{picture}(200,150) \put(70,42){$L$} \put(15,10){$a_0$} \put(78,80){$b_0$} \put(78,130){$c_0$} \put(74,77){\line(0,1){59}} \put(10,10){\circle{4}} \put(10,12){\line(0,1){29}} \put(10,43){\circle{4}} \put(10,45){\line(0,1){13}} \put(10,60){\circle{4}} \put(10,66){\circle{1}} \put(10,69){\circle{1}} \put(10,72){\circle{1}} \put(10,75){\circle{1}} \put(13,75){\circle{1}} \put(16,75){\circle{1}} \put(19,75){\circle{1}} \put(13,78){\circle{1}} \put(16,81){\circle{1}} \put(19,84){\circle{1}} \put(24,75){\circle{4}} \put(26,75){\line(1,0){13}} \put(24,77){\line(0,1){9}} \put(41,75){\circle{4}} \put(41,77){\line(0,1){26}} \put(43,75){\line(1,0){29}} \put(24,88){\circle{4}} \put(26,90){\line(1,1){13}} \put(41,105){\circle{4}} \put(43,107){\line(1,1){29}} \put(74,75){\circle{4}} \put(74,138){\circle{4}} \end{picture} } \put(240,0) { \begin{picture}(100,147) \put(70,42){$\widetilde{L}$} \put(55,135){$a_0$} \put(-8,71){$b_0$} \put(-8,12){$c_0$} \put(6,73){\line(0,-1){59}} \put(70,140){\circle{4}} \put(70,138){\line(0,-1){29}} \put(70,107){\circle{4}} \put(70,105){\line(0,-1){13}} \put(70,90){\circle{4}} \put(70,84){\circle{1}} \put(70,81){\circle{1}} \put(70,78){\circle{1}} \put(70,75){\circle{1}} \put(67,75){\circle{1}} \put(64,75){\circle{1}} \put(61,75){\circle{1}} \put(67,72){\circle{1}} \put(64,69){\circle{1}} \put(61,66){\circle{1}} \put(56,75){\circle{4}} \put(54,75){\line(-1,0){13}} \put(56,73){\line(0,-1){9}} \put(39,75){\circle{4}} \put(39,73){\line(0,-1){26}} \put(37,75){\line(-1,0){29}} \put(56,62){\circle{4}} \put(54,60){\line(-1,-1){13}} \put(39,45){\circle{4}} \put(37,43){\line(-1,-1){29}} \put(6,75){\circle{4}} \put(6,12){\circle{4}} \end{picture} } \end{picture} \vspace{-1.5ex} \noindent $L$ and its dual $\widetilde{L}$ are trivially continuous, having no non-principal ideals possessing a join, so that the way-below relation is the order relation in these lattices. But, being incomplete, they are not continuous lattices in the sense of Scott \cite{Com}, \cite{Sco}. The completion of $L$ by one `middle' point $(0,0)$ is a continuous but not super\-continuous frame, while the completion of $\widetilde{L}$ is not even meet-continuous. } \end{example} The aforementioned one-to-one correspondence between core spaces and C-quasi\-orders is based on the following remark and definition: every space $(X,{\cal S})$ carries a transitive (but only for A-spaces reflexive) {\em interior relation} $R_{{\cal S}}$, given by \vspace{-.5ex} $$ x \,R_{{\cal S}}\,y \, \Leftrightarrow \, y \in ({\uparrow\!x})^{\circ} = int_{{\cal S}} ({\uparrow\!x}), \vspace{-.5ex} $$ where ${\uparrow\!x} = \bigcap\, \{ U \in {\cal S} : x \in U\}$ is the core of $x$. Note that for any subset $A$ of $X$,\\ $R_{{\cal S}}A$ is a lower set in $(X, \leq_{{\cal S}})$, as $\,x \!\leq\! y\,R_{{\cal S}} \, z$ entails $z\in ({\uparrow\!y})^{\circ} \!\subseteq\! ({\uparrow\!x})^{\circ}$, hence $x \, R_{{\cal S}} \,z$.\\ A map $f$ between C-quasi-ordered sets $(X,R)$ and $(X',R')$ {\em interpolates} if for all $y \!\in\! X$ and $x' \!\in\! X'$ with $x' R\, f(y)$ there is an $x \in X$ with $x' R' f(x)$ and $x \,R\, y$, and $f$ is {\em isotone} (preserve the lower quasi-orders) iff $x\leq_{R} y$ implies $f(x) \leq_{R '}\! f(y)$. \begin{theorem} \label{Cspace} {\rm (1)} $(X,{\cal S})$ is a core space iff there is a C-quasi-order $R$ with ${\cal S} = {\cal O}_{R}$. In that case, $R$ is the interior relation $R_{{\cal S}}$, and $\leq_{R}\!$ is the specialization order $\leq_{{\cal S}}$. {\rm (2)} By passing from $(X,{\cal S})$ to $(X,R_{{\cal S}})$, and in the opposite direction from $(X,R)$ to $(X,{\cal O}_{R})$, the category of core spaces with continuous maps is concretely isomorphic to the category of C-quasi-ordered sets with interpolating isotone maps. {\rm (3)} For each closed set $A$ in a core space $(X,{\cal S})$, the set $R_{{\cal S}}A$ is the least lower set with closure $A$. The closure operator induces an isomorphism between the super\-continuous lattice $_{R}{\cal O}$ of all rounded sets and that of all closed sets, while ${\cal O}_{R}$ is the supercontinuous lattice of all open sets. {\rm (4)} The irreducible closed subsets of a core space are exactly the closures of directed sets (in the specialization qoset). For an irreducible closed set\,$A$, the set $R_{{\cal S}}A$ is the least ideal with closure $A$.\,The closure operator induces an iso\-morphism between the continuous domain of rounded ideals and that of irreducible closed sets. {\rm (5)} The topology of a core space $(X,{\cal S})$ is always finer than the Scott topology on its specialization qoset. {\rm (6)} The cocompact topology of a core space $(X,{\cal S})$ is the weak lower topology of the specialization qoset. The patch topology ${\cal S}^{\pi}$ agrees with the weak patch topology\,\,${\cal S}^{\upsilon}$. \end{theorem} \noindent {\it Proof.} (1) and (2) have been established in \cite{EABC}. There are several reasonable alternative choices for the morphisms. For example, the {\em quasiopen} maps (for which the saturations of images of open sets are open) between core spaces are the relation preserving isotone maps between the associated C-quasi-ordered sets (cf.\ \cite{EABC}, \cite{HM}). (3) Let $R$ be the interior relation and $\leq$ the specialization order $\leq_{R}$. We prove the identity $(R A)^- = A^-$, using idempotency of $R$ in the equivalence * below: $y\in (R A)^- \ \Leftrightarrow \ \forall\, U\in {\cal O}_{R}\ (y\in U \Rightarrow U\cap R A \neq \emptyset)$ $\Leftrightarrow \ \forall \, x\in X\ (x \,R \,y \,\Rightarrow\, xR \cap R A \neq \emptyset) \ \stackrel{*}{\Leftrightarrow} \ \forall \, x\in X\ (x \,R \,y \Rightarrow xR \cap A \neq \emptyset)$ $\Leftrightarrow \ \forall\, U\in {\cal O}_{R}\ (y\in U \Rightarrow U\cap A \neq \emptyset) \ \Leftrightarrow \ y\in A^-.$ \noindent In particular, $(R A)^- = A$ in case $A$ is closed. On the other hand, any rounded set $Y$ satisfies the identity $Y = R Y = R (Y^-)$: since each $xR$ is an open set, we have $x\in R Y \ \Leftrightarrow \ xR \cap Y \neq \emptyset \ \Leftrightarrow \ xR \cap Y^- \neq \emptyset \ \Leftrightarrow \ x\in R (Y^-).$ \noindent And if $Y$ is any lower set with $Y^-\! = A$ then $R A = R (Y^-) = R Y \subseteq {\downarrow\!Y} = Y.$ (4) Recall that a subset $A$ is irreducible iff it is nonempty and $A \subseteq B\cup C$ implies $A \subseteq B$ or $A \subseteq C$ for any closed sets $B,C$. Directed subsets and their closures are irreducible. The rounded ideals of a C-quasi-ordered set (or a core space) form a domain ${\cal I}_{R} = \{ I\in {\cal I} (X,\leq_{R}): I = R I\}$, being closed under directed unions. It is easy to see that $I \ll J$ holds in ${\cal I}_{R}$ iff there is an $ x\in J$ with $I\subseteq {\downarrow\!x}$, and that the join of the ideal $\ll \!J = {\downarrow_{{\cal I}_{R}}} \{ R x : x\in J\}$ is $J$. Thus, ${\cal I}_{R}$ is continuous (cf.\ \cite{GK},\,\cite{Lss}). For the isomorphism claim, observe that the coprime rounded sets are the rounded ideals, the coprime closed sets are the irreducible ones, and a lattice isomorphism preserves coprimes (an element $q$ is {\em coprime} iff the complement of ${\uparrow\!q}$ is an ideal). (5) By (4), $I = R_{{\cal S}} y$ is an ideal with ${\downarrow\! y} = I^-$, so $y$ is a least upper bound of $I$; now, $y \in U \in \sigma (X,\leq)$ implies $U\cap R_{{\cal S}} y \not = \emptyset$, i.e.\ $y\in UR_{{\cal S}}$. Thus, $U = UR_{{\cal S}}\in {\cal S}$. (6) Let $C$ be a compact saturated set (i.e.\ $C \!\in\! \pi {\cal S}$) and $y \!\in\! X\setminus C$. There is an open neighborhood $U$ of $C$ not containing $y$. By local supercompactness\,of\,\,$(X,{\cal S})$, we have $U = \bigcup\, \{ int_{{\cal S}}({\uparrow\! x}) : x\in U\}$, and by compactness of $C$, we find a finite $F \subseteq U$ with $C \!\subseteq \bigcup\, \{ int_{{\cal S}} ({\uparrow\! x}) : x \!\in\! F\} \!\subseteq\! {\uparrow\! F} \!\subseteq\! U$. For $Q = (X,\leq)$, ${\uparrow\! F}$ is $\upsilon\widetilde{Q}$-closed and does not contain $y$. Thus, $C$ is $\upsilon\widetilde{Q}$-closed, and $\pi {\cal S}$ is contained in $\upsilon \widetilde{Q}$; the reverse inclusion is obvious, since cores (principal filters) are compact and saturated. \EP Recall that a {\em sober} space is a T$_0$-space whose only irreducible closed sets are the point closures; and a {\em monotone convergence space} ({\em mc-space}) or {\em d-space} is a T$_0$ space in which the closure of any directed subset is the closure of a singleton (see \cite{Emin}, \cite{CLD}, \cite{Wy}). Now, from Theorem \ref{Cspace}, one deduces (cf.\ \cite{EABC}, \cite{Hoff2}, \cite{Law}, \cite{Lss}): \begin{corollary} \label{contdom} The following conditions on a space $S$ are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] $S$ is a sober core space. \item[{\rm (2)}] $S$ is a locally supercompact monotone convergence space (d-space). \item[{\rm (3)}] $S$ is the Scott space of a (unique) continuous domain. \end{itemize} \vspace{-.5ex} The category of sober core spaces and the concretely isomorphic category of continuous domains are dual to the category of supercontinuous spatial frames. \end{corollary} \vspace{-.5ex} For the case of continuous posets that are not necessarily domains, see \cite{EScon}, \cite{Emin}. Afficionados of domain theory might remark that continuous frames are automatically spatial (see \cite{CLD}, \cite{HL}). But that `automatism' requires choice principles, which we wanted to avoid in the present discussion. However, it seems to be open whether the spatiality of {\em supercontinuous} frames may be proved in a choice-free manner. Since a T$_0$-space and its sobrification have isomorphic open set frames, it follows from Corollary \ref{contdom} that a T$_0$-space is a core space iff its sobrification is the Scott space of a continuous domain. This completion process is reflected, via Theorem \ref{Cspace}, by the fact that the rounded ideal completion of a C-ordered set is a continuous domain, and the C-order is extended to its completion, meaning that $x\,R\,y$ is equivalent to $R x \ll R y$ in the completion (cf. \cite{Emin}, \cite{CLD}, \cite{GK}, \cite{Lss}). \section{Core stable spaces and sector spaces} \label{sectorspaces} As every core space is a web space, it is equal to the upper spaces of its patch spaces and therefore $\zeta$-determined for any coselection $\zeta$ (see Proposition \ref{webpatch}). We now are going to determine explicitly these patch spaces, which turn out to have very good separation properties, whereas the only T$_1$ core spaces (in fact, the only T$_1$ web spaces) are the discrete ones. For alternative characterizations of such patch spaces, we need further properties of the interior operators. Call a quasi-ordered space $(Q, {\cal T} )$ {\em core stable} or {\em c-stable} if $$ \textstyle{{\uparrow\!O} = \bigcup \,\{ \,int_{{\cal T}}({\uparrow\!u}) : u\in O\} \ \mbox{ for each }O \in {\cal T}, } $$ and {\em d-stable} if for any filtered (i.e.\ down-directed) subset $D$ of $Q$, $$ \textstyle{int_{{\cal T}}D \subseteq \bigcup\,\{ int_{{\cal T}} \, ({\uparrow\!u}): u\in cl_{{\cal T}^{\geq}} D\}.} $$ While core stability is a rather strong property, d-stability is a rather weak one (trivially fulfilled if all filters are principal). The terminology is justified by \begin{lemma} \label{cstable} Let $T = (X,\leq, {\cal T})$ be a semi-qospace. {\rm (1)} $T$ is d-stable whenever its lower space ${\rm L} T$ is a d-space. {\rm (2)} $T$ is core stable iff $\,{\rm U}T = (X,\{ {\uparrow\!O} : O \in {\mathcal T}\})$ is a core space. {\rm (3)} $T$ is core stable iff it is upper regular, locally filtered, ${\uparrow}$-stable and d-stable. \end{lemma} \noindent {\it Proof.} (1) If the lower space $\,{\rm L}T = (X,{\cal T}^{\geq})$ is a d-space then for any filtered set $D$ in $(X,\leq)$ there is a $u$ with ${\uparrow\!u} = cl_{{\cal T}^{\geq}} D$ and $\,int_{{\cal T}} D \subseteq int_{{\cal T}} ({\uparrow\!u})$. Without proof, we note that ${\rm L}T$ is a d-space whenever $T$ is hyperconvex and $(X,\geq)$ is a domain. (2) Clearly, a core stable semi-qospace $(X,\leq ,{\cal T})$ is ${\uparrow}$-stable, whence ${\mathcal T}^{\leq} = \{ {\uparrow\!O} : O \in {\mathcal T}\}$. For $U\in {\cal T}^{\leq}$ and $x\in U$ there exists a $u\in U$ with $x\in int_{{\cal T}} ({\uparrow\!u})$. Then $W = {\uparrow int_{{\cal T}}({\uparrow\!u})}\in {\cal T}^{\leq}$ (by ${\uparrow}$-stability) and $x\in W\subseteq {\uparrow\!u} \subseteq {\uparrow\!U} = U$. Since $\leq$ is the specialization order of ${\cal T}^{\leq}$, this ensures that $(X,{\cal T}^{\leq})$ is a core space. Conversely, suppose $(X,{\cal T}^{\leq}) = (X, \{ {\uparrow\!O} : O \in {\mathcal T}\})$ is a core space. For $O\in {\mathcal T}$ and $y \in {\uparrow\!O}$, there is an $x\in {\uparrow\!O}$ and a $U\in {\cal T}^{\leq} \subseteq {\cal T}$ with $y\in U\subseteq {\uparrow\!x}$. Now, pick a $u\in O$ with $x\in {\uparrow\!u}$; then $y\in U \subseteq {\uparrow\!x} \subseteq {\uparrow\!u}$, whence ${\uparrow\!O} \subseteq \bigcup \,\{ \,int_{{\cal T}}({\uparrow\!u}) : u\in O\}$. (3) Let $T = (X,\leq, {\cal T} )$ be a core stable semi-qospace. For $x\in O\in {\cal T}^{\leq}$, there is a $u\in O$ with $x\in U = int_{{\cal T}} ({\uparrow\!u}) \subseteq {\uparrow\!u} \subseteq O$. By ${\uparrow}$-stability, we have $U\in {\cal T}^{\leq}$, and since $T$ is a semi-qospace, ${\uparrow\!u}$ is a closed upper set. Hence, $T$ is upper regular. In order to check local filteredness, pick for $x\in O\in {\cal T}$ an element $u\in O$ with $x\in int_{{\cal T}} ({\uparrow\!u}) \mathop{\cap} O \subseteq {\uparrow\!u} \mathop{\cap} O\,$; this is a filtered set, possessing the least element\,\,$u$. Moreover, $T$ is not only ${\uparrow}$-stable (see (2)) but also d-stable, since for any subset $D$ and any $x\in O = int_{{\cal T}} D$, there is a $u \in O \subseteq D \subseteq cl_{{\cal T}^{\geq}} D$ with $x\in int_{{\cal T}}({\uparrow\!u})$. Conversely, suppose that $T$ is an upper regular, locally filtered, ${\uparrow}$-stable and d-stable semi-qospace. Then, for $O\in {\cal T}$ and $x\in {\uparrow\!O}$, we have: $x\in {\uparrow\!O} \in {\cal T}^{\leq}$ \hfill (by ${\uparrow}$-stability), there are $U\in {\cal T}^{\leq}$ and $B\in {\cal T}^{\geq c}$ with $x\in U \subseteq B \subseteq {\uparrow\!O}$ \hfill (by upper regularity), a filtered $D$ with $x\in int_{{\cal T}} D \subseteq D \subseteq U$ \hfill (by local filteredness), and an element $y\in cl_{{\cal T}^{\geq}} D$ with $x \in int_{{\cal T}} ({\uparrow\!y})$ \hfill (by d-stability). \noindent It follows that $y \in cl_{{\cal T}^{\geq}} D \subseteq cl_{{\cal T}^{\geq}} U \subseteq B \subseteq {\uparrow\!O}$. Now, choose a $u\in O$ with $u\leq y$, hence ${\uparrow\!y} \subseteq {\uparrow\!u}$, to obtain $x\in int_{{\cal T}} ({\uparrow\!u})$. Thus, ${\uparrow\!O} \subseteq \bigcup\,\{ int_{{\cal T}} ({\uparrow\!u}) : u\in O\}$, showing that $T$ is core stable. \EP By Lemma \ref{cstable}, every core stable semi-qospace is a qospace (being upper regular); in fact, core stability of a semi-qospace splits into the four properties (c1) {\em upper regular} \ \ (c2) {\em locally filtered} \ \ (c3) {\em ${\uparrow}$-stable} \ \ (c4) {\em d-stable}. \noindent These properties are independent: none of them follows from the other three. \begin{example} \label{Ex31} {\rm Let $L$ be a non-continuous wide frame (for example, a T$_2$ topology that is not locally compact, like the Euclidean topology on ${\mathbb Q}$). As shown in \cite{Eweak}, $(L, \upsilon L)$ is a wide web space whose meet operation is continuous in the weak upper topology $\upsilon L$ and in the weak lower topology $\upsilon \widetilde{L}$, hence also in the interval topology $\iota L = \upsilon L\,^{\upsilon}$; and ${\rm I}L = (L,\iota L )$ is a topological meet-semilattice with small subsemilattices (see Section \ref{semi}). Thus, it is locally filtered, ${\uparrow}$-stable and d-stable, since the lower space $(\widetilde{L},\upsilon \widetilde{L})$ is a d-space (for any complete lattice $L$). But the semi-pospace ${\rm I}L$ can be upper regular only if it is a pospace, hence T$_2$, which happens only if $L$ is a continuous lattice \cite{CLD}. Thus, ${\rm I}L$ satisfies (c2), (c3) and (c4), but not (c1). } \end{example} \begin{example} \label{Ex32} {\rm For $0 < s < 1$, consider the non-compact subspace $S = [\,0, s\,[ \, \mathop{\cup}\, \{ 1\}$ of the Euclidean space ${\mathbb R}$, ordered by $x\sqsubseteq y \Leftrightarrow x = y$ or $x = 0$ or $y=1$. As $\{ 1\}$ is clopen, it is easy to check that $S$ satisfies (c1), (c3) and (c4) (all filters are principal), but not (c2): no point except $1$ has a filtered neighborhood not containing 0. } \end{example} \begin{example} \label{Ex33} {\rm In Example 4.1 of \cite{Epatch} it is shown that the set \begin{picture}(300,55) \put(10,40){$X = \{ a,\top\} \cup \{ b_n : n\in \omega\}$, ordered by} \put(10,20){$x\leq y \ \Leftrightarrow x = y \mbox{ or } x = b_0 \mbox{ or } y = \!\top $} \put(81,0){or $x = b_i, y = b_j, i < j,$} \put(260,30){\circle{5}} \put(250,28){$a$} \put(262,32){\line(3,2){40}} \put(262,28){\line(3,-2){40}} \put(305,36){\circle{5}} \put(305,20){\circle{5}} \put(305,33){\line(0,-1){10}} \put(305,18){\line(0,-1){15}} \put(305,39){\line(0,1){5}} \put(305,47){\circle{1}} \put(305,51){\circle{1}} \put(305,55){\circle{1}} \put(305,60){\circle{5}} \put(310,58){$\top$} \put(305,0){\circle{5}} \put(310,31){$b_2$} \put(310,14){$b_1$} \put(310,-4){$b_0$} \end{picture} \vspace{1ex} \noindent is a complete but not meet-continuous lattice and a compact pospace when equipped with the Lawson topology. It satisfies (c1), (c2) and (c4) (because all filters are principal). However, (c3) is violated, since $\{ a \}$ is open, while ${\uparrow\!a} = \{ a, \top\}$ is not. } \end{example} \begin{example} \label{Ex34} {\rm For infinite $I$, the function space ${\mathbb R}^I$ with the coordinatewise order and the Lawson topology satisfies (c1), (c2) and (c3), but its upper space $\Sigma ({\mathbb R}^I)$ is not a core space: otherwise, in view of the fact that bounded directed subsets converge to their suprema, ${\mathbb R}^I$ would have to be a continuous poset (see \cite{Emin}), which holds only for finite $I$ (see \cite{EScon}). Hence, by Lemma \ref{cstable}, (c4) must be violated. } \end{example} Given a quasi-ordered space $(X,\leq,{\cal T})$, we call any nonempty set of the form ${\uparrow\!u}\mathop{\cap} V$ with $V \!\in \!{\cal T}^{\,\geq}\!$ a {\em sector}, and a {\em $\zeta$-sector} if $V\in \zeta ({\cal T}^{\leq})$ for a coselection $\zeta$. Hence, $u$ is the least element of the sector, and every sector is obviously a web around any point it contains. By a ($\zeta$-){\em sector space} we mean an ${\uparrow}$-stable semi-qospace in which each point $x$ has a base of ($\zeta$-)sector neighborhoods ${\uparrow\!u}\cap V$ (but the point $x$ need not be the minimum $u$ of such a sector). \begin{theorem} \label{sectors} The sector spaces are exactly the \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] patch spaces of core spaces, \item[{\rm (2)}] strongly convex, core stable (semi-)qospaces, \item[{\rm (3)}] strongly convex, upper regular, locally filtered, ${\uparrow}$-stable and d-stable qospaces. \end{itemize} In particular, all ordered sector spaces are T$_3$-ordered, a fortiori T$_2$-ordered. \noindent Specifically, for any coselection $\zeta$, the $\zeta$-sector spaces are exactly the \vspace{-.5ex} \begin{itemize} \item[{\rm (1$\zeta$)}] $\zeta$-patch spaces of core spaces, \item[{\rm (2$\zeta$)}] $\zeta$-convex, core stable (semi-)qospaces, \item[{\rm (3$\zeta$)}] $\zeta$-convex, upper regular, locally filtered, ${\uparrow}$-stable and d-stable qospaces. \end{itemize} \end{theorem} \noindent {\it Proof.} (1) and (2): Let $T = (X,\leq,{\cal T})$ be a sector space. For $x\in O\in {\cal T}$, there are $u \!\in\! X$ and $V \!\in\! {\cal T}^{\,\geq}$ such that $x\in int_{{\cal T}} ({\uparrow\!u} \mathop{\cap} V) = int_{{\cal T}} ({\uparrow\!u})\cap V \subseteq {\uparrow\!u} \mathop{\cap} V \subseteq O$. Then $u\in {\uparrow\!u}\cap V \subseteq O$ and $x \in int_{{\cal T}}({\uparrow\!u})$, whence $O \subseteq \bigcup \,\{ int_{{\cal T}}({\uparrow\!u}) : u \!\in\! O\}$. Using ${\uparrow}$-stability and applying the previous argument to ${\uparrow\!O}$ instead of $O$, one concludes that $T$ is core stable. Again by ${\uparrow}$-stability, $U = {\uparrow int_{{\cal T}}}({\uparrow\!u}\,\cap V)$ belongs to ${\cal T}^{\,\leq}$, and the above reasoning yields $x\in U\cap V \subseteq {\uparrow\!u}\cap V\subseteq O$, proving strong convexity. Now, let $(X,\leq,{\cal T})$ be any strongly convex core stable semi-qospace. Then, by Lemma \ref{cstable}, $(X,{\cal T}^{\leq})$ is a core space with specialization order $\leq$, while ${\cal T}^{\geq}$ has the dual specialization order. By strong convexity, ${\cal T}$ is the patch topology ${\cal T}^{\leq} \vee {\cal T}^{\geq}$. On the other hand, by Proposition \ref{pat}, any patch space $(X,\leq, {\cal T} )$ of a core (or web) space $(X,{\cal S} )$ is a web-quasi-ordered space, in particular ${\uparrow}$-stable. For $x\in U \cap V$ with $U\in {\cal S}$ and $V\in {\cal T}^{\geq}$, there are $u\in U$ and $W\in {\cal S}$ with $x\in W \subseteq {\uparrow\!u}$; it follows that $x\in W \mathop{\cap} V \subseteq {\uparrow\!u} \mathop{\cap} V \subseteq U \mathop{\cap} V$. Thus, $(X,\leq, {\cal T} )$ is a sector space. (3) follows from (2) and Lemma \ref{cstable}. Any upper regular semi-qospace is a qospace. \noindent The modified claims involving coselections $\zeta$ are now easily derived from Proposition\,\,\ref{pat} and Lemma \ref{webpatch}, using the fact that core spaces are web spaces. \EP We are ready to establish a categorical equivalence between core spaces and $\zeta$-sector spaces. As explained in \cite{Epatch}, the right choice of morphisms is a bit delicate. Continuous maps would be the obvious morphisms between core spaces. On the other hand, one would like to have as morphisms between quasi-ordered spaces the isotone continuous maps. But, as simple examples in \cite{Epatch} show, a continuous map between two core spaces need not be continuous as a map between the associated $\zeta$-sector spaces, and a continuous isotone map between $\zeta$-sector spaces need not be continuous for the weak lower topologies (consider the map $f$ on the lattice $L$ in Example \ref{Exnotcon} with $f(a_n)=f(b_n) =a_0$, $f(c_n)=c_0$). Therefore, we take the $\zeta$-proper maps (see Section \ref{convex}) as morphisms between core spaces in order to save the isomorphism theorem. Let us summarize the previous results. \begin{proposition} \label{sectorcat} For any coselection $\zeta$, the patch functor ${\rm P}_{\zeta}$ induces a concrete iso\-mor\-phism between the category {\bf CS\hspace{.1ex}} (resp.\ {\bf CS}\hspace{.1ex}{\boldmath $\zeta$}) of core spaces with continuous (resp.\ $\zeta$-proper) maps and the category {\boldmath $\zeta$}{\bf S\hspace{.2ex}l} (resp.\ {\boldmath $\zeta$}{\bf S\hspace{.1ex}c}) of $\zeta$-sector spaces with isotone lower semicontinuous (resp.\ continuous) maps. The inverse isomorphism is induced by the functor ${\rm U}$, sending a sector space $T$ to the core space ${\rm U}T$. \end{proposition} A related class of morphisms is formed by the {\em core continuous maps}, i.e.\ continuous maps for which preimages of cores are cores. In terms of the specialization orders, the latter condition means that these maps are residual (preimages of principal filters are principal filters), or equivalently, that they have lower adjoints (cf.\ \cite{Eclo} and \cite[0--1]{CLD}). Core continuous maps are $\alpha$-, $\sigma$- and $\upsilon$-proper, but not conversely. The following facts are established in \cite{EABC} (see also \cite{HM}): \begin{corollary} \label{qopen} Passing to lower adjoints yields a duality between the category {\bf CSc} of core spaces with core continuous maps and the category {\bf CSq} of core spaces with quasiopen maps (saturations of images of open sets are open) that are residuated (preimages of point closures are point closures). The full subcategory {\bf SCSc} of sober core spaces is equivalent to the category {\bf CDjm} of completely distributive spatial frames with maps preserving arbitrary joins and meets, which in turn is dual (via adjunction) to the category {\bf CDjw} of completely distributive spatial frames with maps preserving joins and the relation $\triangleleft$, where $x \triangleleft\, y \,\Leftrightarrow\, x\in \bigcap\,\{ \,{\downarrow\!A} : y\leq \bigvee\! A\}$. \end{corollary} \begin{picture}(300,70) \put(-2,60){\bf CDjm} \put(-2,2){\bf CDjw} \put(10,50){\vector(0,-1){30}} \put(16,20){\vector(0,1){30}} \put(67,66){\vector(-1,0){30}} \put(37,60){\vector(1,0){30}} \put(67,8){\vector(-1,0){30}} \put(37,2){\vector(1,0){30}} \put(80,60){\bf SCSc} \put(80,2){\bf SCSq} \put(90,50){\vector(0,-1){30}} \put(96,20){\vector(0,1){30}} \put(120,61){$\subset$} \put(123,61){\vector(1,0){30}} \put(120,3){$\subset$} \put(123,3){\vector(1,0){30}} \put(164,60){\bf CSc} \put(164,2){\bf CSq} \put(170,50){\vector(0,-1){30}} \put(176,20){\vector(0,1){30}} \put(197,61){$\subset$} \put(220,61){\vector(1,0){10}} \put(206,60.5){$\dots$} \put(193,47){\small $\zeta = \alpha,\sigma,\upsilon$} \put(241,60){{\bf CS}\hspace{.1ex}{\boldmath $\zeta$}} \put(244,2){{\boldmath $\zeta$}{\bf S\hspace{.1ex}c}} \put(250,50){\vector(0,-1){30}} \put(256,20){\vector(0,1){30}} \put(274,61){$\subset$} \put(277,61){\vector(1,0){30}} \put(274,3){$\subset$} \put(277,3){\vector(1,0){30}} \put(320,60){\bf CS} \put(318,2){{\boldmath $\zeta$}{\bf S\hspace{.2ex}l}} \put(327,50){\vector(0,-1){30}} \put(333,20){\vector(0,1){30}} \end{picture} \vspace{2ex} In contrast to locally supercompact spaces (core spaces), locally hypercompact spaces need not be $\zeta$-determined, and the categorical equivalence between core spaces (i.e.\ locally hypercompact web spaces) and $\zeta$-sector spaces does not extend to locally hypercompact spaces without additional restrictions. The lattice $\widetilde{L}$ in Example \ref{Exnotcon} is locally hypercompact but not locally supercompact in the weak upper topology; in fact, it fails to be a web space: the points $b_n$ have no small web neighborhoods. Since $\widetilde{L}$ is not $\zeta$-determined unless $\zeta \widetilde{L} \!=\! \alpha \widetilde{L}$, Lemma \ref{pat} does not apply to that case. For the complete lattice $M = (X,\leq)$ in Example \ref{Ex33}, the Scott space $\Sigma M$ is locally hypercompact and $\upsilon$-determined, but its weak patch space $\Lambda M = {\rm P}_{\upsilon}\Sigma M$ is not ${\uparrow}$-stable: while $\{ a\}$ is open, ${\uparrow\!a}$ is not. \newpage \section{Fan spaces} \label{fanspaces} We have seen that, by virtue of the weak patch functor ${\rm P}_{\upsilon}$, the core spaces bijectively correspond to the $\upsilon$-sector spaces. We shall now give some effective descriptions of these specific qospaces. By a {\em fan} we mean a nonempty set of the form ${\uparrow\!u}\setminus {\uparrow\!F}$ for some finite $F$. In case $(X,\leq,{\cal T})$ is an upper semi-qospace, each fan is a sector, in fact, a $\upsilon$-sector. An ${\uparrow}$-stable semi-qospace space with a base of fans will be called a {\em fan space}. A nonempty product of quasi-ordered spaces is a fan space iff all factors are fan spaces and all but a finite number have a least element. \begin{example} \label{Ex72} {\rm The Euclidean spaces $\Lambda (\mathbb{Q}^n) = (\Lambda \mathbb{Q})^n$ and $\Lambda (\mathbb{R}^n) = (\Lambda \mathbb{R})^n$ are metrizable fan spaces. In contrast to $\Lambda (\mathbb{R}^n)$, the rational space $\Lambda (\mathbb{Q}^n)$ is not locally compact, whereas the upper spaces $\Sigma (\mathbb{Q}^n)$ and $\Sigma (\mathbb{R}^n)$ are locally super\-compact. But neither $\Sigma (\mathbb{Q}^n)$ nor $\Sigma (\mathbb{R}^n)$ is sober, since $\mathbb{Q}^n$ and $\mathbb{R}^n$ fail to be up-complete. } \end{example} \begin{example} \label{Ex41} {\rm An open web, a sector, and a fan in the Euclidean plane $\Lambda \,{\mathbb R}^2$ \begin{picture}(300,85) \put(10,0){ \begin{picture}(100,85) \multiput(-6,79)(0,-2){22}{.} \multiput(-6,79)(2,0){10}{.} \multiput(-6,35)(2,0){18}{.} \multiput(30,35)(0,-2){18}{.} \multiput(30,-1)(2,0){22}{.} \multiput(74,-1)(0,2){10}{.} \multiput(48,19)(2,0){14}{.} \multiput(48,19)(0,2){17}{.} \multiput(48,53)(-2,0){17}{.} \multiput(14,53)(0,2){14}{.} \put(39.5,44){\circle*{3}} \put(5,44){\line(1,0){34}} \put(4.5,44){\circle{2}} \put(4.5,45){\line(0,1){28}} \put(4.5,74){\circle{2}} \put(39.5,43){\line(0,-1){34}} \put(39.5,8.5){\circle{2}} \put(40,8.5){\line(1,0){30}} \put(70.5,8.5){\circle{2}} \put(-10,-15){\em open web, not sector} \end{picture} } \put(130,0){ \begin{picture}(100,90) \put(5,-15){\em sector, not fan} \put(0,0){\line(0,1){80}} \put(0,0){\line(1,0){80}} \multiput(0,79)(2,0){7}{.} \multiput(14,78)(2,-1){13}{.} \multiput(40,64)(1.5,-1.5){8}{.} \multiput(52,51)(1,-2){13}{.} \multiput(65,25)(1,-1.5){9}{.} \multiput(74,11)(.7,-2){6}{.} \end{picture} } \put(255,0){ \begin{picture}(100,90) \put(30,-15){\em fan} \put(0,0){\line(0,1){78}} \put(0,0){\line(1,6){13}} \put(0,0){\line(1,5){13.5}} \put(0,0){\line(2,5){27}} \put(0,0){\line(1,2){27.5}} \put(0,0){\line(3,4){41}} \put(0,0){\line(1,1){41}} \put(0,0){\line(4,3){55}} \put(0,-.3){\line(2,1){55}} \put(0,-.3){\line(5,2){69}} \put(0,-.3){\line(5,1){69}} \put(0,0){\line(6,1){80}} \put(0,0){\line(1,0){80}} \multiput(0,78)(2,0){7}{.} \multiput(12,77)(0,-2){6}{.} \multiput(12,67)(2,0){7}{.} \multiput(26,67)(0,-2){7}{.} \multiput(26,55)(2,0){8}{.} \multiput(40,55)(0,-2){8}{.} \multiput(40,41)(2,0){8}{.} \multiput(54,41)(0,-2){8}{.} \multiput(54,27)(2,0){8}{.} \multiput(68,27)(0,-2){8}{.} \multiput(68,13)(2,0){6}{.} \multiput(78,13)(0,-2){7}{.} \end{picture} } \end{picture} } \end{example} \vspace{3.5ex} In order to see that hyperconvex upper regular semi-qospaces are regular, it suffices to guarantee the separation of points from subbasic closed sets. For closed lower sets, this is possible by upper regularity. For subbasic closed sets ${\uparrow\!x}$ and points $y\in X\setminus {\uparrow\!x}$, one finds an open upper set $U$ and a closed upper set $B$ with ${\uparrow\!x} \subseteq U \subseteq B \subseteq X\setminus{\downarrow\!y}$; then $X\setminus B$ is an open set disjoint from $U$ and containing\,\,$y$. Now, from Theorem \ref{sectors} and Proposition \ref{sectorcat}, we infer for the case $\zeta = \upsilon$: \begin{theorem} \label{fans} The fan spaces are exactly the \vspace{-1ex} \begin{itemize} \item[{\rm (1$\upsilon$)}] weak patch spaces of core spaces, \item[{\rm (2$\upsilon$)}] hyperconvex core stable (semi-)qospaces, \item[{\rm (3$\upsilon$)}] hyperconvex, upper regular, locally filtered, up- and d-stable qospaces. \end{itemize} The category of core spaces with continuous (resp.\ $\upsilon$-proper) maps is isomorphic to that of fan spaces and lower semicontinuous (resp.\ continuous) isotone maps. All fan spaces are not only upper regular but also regular. Moreover, they are uniformizable, and so completely regular in the presence of the Axiom of Choice. \end{theorem} Let us make the last statement in Theorem \ref{fans} more precise. For the definition of {\em quasi-uniformities} and a study of their relationships to (quasi-)ordered spaces, refer to Fletcher and Lindgren \cite{FL} or Nachbin \cite{Na}. For a quasi-uniformity ${\cal Q}$, the dual ${\cal Q}^{-1}$ is obtained by exchanging first and second coordinate, and ${\cal Q}^*$ is the uniformity generated by ${\cal Q} \cup {\cal Q}^{-1}$. The topology $\tau ({\cal Q})$ generated by ${\cal Q}$ consists of all subsets $O$ such that for $x \!\in\! O$ there is a $\,U\!\in\! {\cal Q}$ with $xU = \{ y : (x,y) \in U\} \!\subseteq\! O$. The following results are due to Br\"ummer and K\"unzi \cite{KB} and Lawson \cite{Lss}. \begin{lemma} \label{localunif} If $(X,{\cal S} )$ is a locally compact space then there is a coarsest quasi-uniformity ${\cal Q}$ with $\tau ({\cal Q}) = {\cal S}$, and ${\cal Q}$ is generated by the sets $\hspace{4ex} U'\!\rightarrow\! U =\{ (x,y) \in X\times X : x\in U' \Rightarrow y\in U\} = (X\setminus U')\!\times\! X \, \cup \, X\!\times\!U$ \noindent with $\,U,U'\in {\cal S}$ and $U\ll U'$ in the continuous frame ${\cal S}$. Specifically, {\rm (1)} If ${\cal B}$ is a subbase for ${\cal S}$ such that $B = \bigcup\,\{ B'\in {\cal B} : B'\ll B\}$ for all $B\in {\cal B}$, \hspace{4ex}then the sets $B'\!\rightarrow\! B$ with $B,B'\in {\cal B}$ and $B\ll B'$ generate ${\cal Q}$, {\rm (2)} $\tau ({\cal Q}^{-1})$ is the cocompact topology $\tau_{\pi} {\cal S}$, {\rm (3)} $\tau({\cal Q}^*)$ is the patch topology ${\cal S}^{\pi}$. \end{lemma} For core spaces, these conclusions may be strengthened as follows. \begin{proposition} \label{superunif} Let $(X,{\cal S} )$ be a core space with specialization qoset $Q = (X,\leq )$ and interior relation $R$.\,Then there is a coarsest quasi-uniformity ${\cal Q}$ with $\tau ({\cal Q}) \!= {\cal S}$,\,and {\rm (1)} ${\cal Q}$ is generated by the sets $x'\!R \!\rightarrow\! y'R = \{ (x,y) : x' R\, x \Rightarrow y'R\, y\}$ with $y' R \, x'$, {\rm (2)} $\tau (Q^{-1})$ is the weak lower topology $\upsilon \widetilde{Q}$, {\rm (3)} the fan space $(X,\leq, {\cal S}^{\upsilon})$ is determined by ${\cal Q}$, i.e.\ $\leq \,= \bigcap {\cal Q}$ and ${\cal S}^{\upsilon} = \tau ({\cal Q}^*)$. \end{proposition} \noindent {\it Proof.} \mbox{We apply Lemma \ref{localunif} to the base ${\cal B} = \{ xR : x\!\in\! X\}$ of ${\cal O}_{R} = {\cal S}$ (Theorem \ref{Cspace}).} By idempotency, $x\,R\, y$ implies $x\,R\, x'R\, y$ for some $x'$, hence $x'R \ll xR$. This yields the equation $xR = \bigcup \,\{ x'\!R \in {\cal B} : x'\!R \ll xR\}$, and then (1) follows from Lemma \ref{localunif}. Now, Theorem \ref{Cspace}\,(6) and Lemma \ref{localunif}\,(2) give $\tau ({\cal Q}^{-1}) = \tau_{\pi}{\cal S} = \upsilon \widetilde{Q}$. Finally, $\bigcap {\cal Q}$ is the specialization order of $\tau ({\cal Q})$, and ${\cal S}^{\upsilon} = {\cal S}^{\pi} = \tau ({\cal Q}^*)$, by Lemma \ref{localunif}\,(3). It is also easy to check the equations ${\cal S} = \tau ({\cal Q})$ and $\upsilon \widetilde{Q} = \tau ({\cal Q}^{-1})$ directly: For $U = x'\!R \!\rightarrow\! y'R$, the set $xU$ is equal to $y'R$ if $x'R\, x$, and to $X$ otherwise, while the set $Uy$ is equal to $X\setminus x'R$ if $y'\!\!\not\!R\, y$, and to $X$ otherwise. \noindent In case $y'R\,x'$ but not $y'R\, y$, the inclusion $y\in X\setminus y'R \subseteq X\setminus {\uparrow\!y'} \subseteq x'R = Uy$ shows that each $Uy$ is a neighborhood of $y$ in the weak lower topology. \EP \vspace{1ex} In \cite{FL} (see also \cite{Kue}, \cite{Lss}, \cite{Na}), it is shown that in {\sf ZFC} a pospace \mbox{$T = (X,\leq,{\cal T})$} is determined by a quasi-uniformity ${\cal Q}$, i.e.\ $\leq \ = \bigcap {\cal Q}$ and ${\cal T} = \tau ({\cal Q}^*)$, iff it has a (greatest) order compactification, which in turn is equivalent to saying that $T$ is {\em completely regular(ly) ordered}, i.e., a pospace such that for $x \!\in\! U \!\in\! {\cal T}$, there are continuous $f,g: X \rightarrow [0,1]$ with $f$ isotone and $g$ antitone, $f(x) = g(x) = 1$, and $f(y) = 0$ or $g(y) = 0$ for $y\in X\setminus U$. Dropping the antisymmetry, we arrive at \begin{corollary} \label{complreg} In {\sf ZFC}, every fan space is completely regularly quasi-ordered. \end{corollary} \newpage \section{Continuous domains as pospaces} \label{domain} We now are going to establish a representation of continuous domains by certain pospaces, generalizing the famous characterization of continuous lattices as meet-continuous lattices whose Lawson topology is Hausdorff (see \cite[III--2.11]{CLD}). In contrast to the situation of complete lattices, we require at most up-completeness. For our purposes, we need a further definition. Given an upset selection $\zeta$ (see Section \ref{convex}), a quasi-ordered space $(Q,{\cal T}) = (X,\leq,{\cal T} )$ or its topology is said to be {\em $\zeta$-stable} if the interior operator $^{\circ}$ of the upper topology ${\cal T}^{\leq}$ satisfies the equation \vspace{-.5ex} $$ \textstyle{ Y^{\circ} = \bigcup\,\{ ({\uparrow\!y})^{\circ} : y\in Y\}} \ \mbox{ for all } Y \!\in \! \zeta Q. $$ If $(Q,{\cal T})$ is ${\uparrow}$-stable, the operator $^\circ$ may be replaced by the interior operator of the original topology ${\cal T}$. For instance, {\em core stable} means {\em $\alpha$-stable plus ${\uparrow}$-stable}.\\ We denote by $\curlyvee Q$ and $\curlywedge Q$ the set of all finite unions (incl.\ $\emptyset \!= \bigcup \emptyset$) resp.\ intersections (incl.\,$X \!= \bigcap \emptyset$) of principal filters, i.e., the unital join- resp.\ meet-sub\-semi\-lattice they generate in the lattice $\alpha Q$ of all upper sets. Further, we denote by $\diamondsuit Q$ the sublattice of $\alpha Q$ generated by the principal filters and containing $\emptyset$ and $X$. \begin{lemma} \label{stable} Let $T = (Q,{\cal T} )$ be any quasi-ordered space. \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] $T$ is $\curlywedge$-stable iff the sets $R y = \{ x : y\in ({\uparrow\!x})^{\circ}\}$ are directed (ideals). \item[{\rm (2)}] $T$ is $\diamondsuit$-stable iff it is $\curlyvee$-stable and $\curlywedge$-stable. \item[{\rm (3)}] $T$ is $\curlyvee$-stable if its upper space ${\rm U}T$ is a web space. \item[{\rm (4)}] $T$ is $\curlywedge$-stable if $Q$ is a join-semilattice with least element. \item[{\rm (5)}] $T$ is $\diamondsuit$-stable if it is a lattice with $0$ and continuous unary $\wedge$-operations. \item[{\rm (6)}] $T$ is a $\curlyvee$-stable lower semi-qospace with locally hypercompact upper space ${\rm U}T$ iff the latter is a core space with specialization qoset $Q$. \end{itemize} \end{lemma} \noindent {\it Proof.} (1) Suppose $T$ is $\curlywedge$-stable. For finite $F\!\subseteq\! R y$, the set \mbox{$F^{\uparrow\!} = \bigcap \, \{{\uparrow\! x} : x\in F\}$} is a member of $\curlywedge Q$, whence $y\in \bigcap \, \{ ({\uparrow\! x})^{\circ} : x\in F\} = (F^{\uparrow})^{\circ} = \bigcup\,\{ ({\uparrow\! u})^{\circ} : u\!\in\! F^{\uparrow\!}\}$, i.e., $F^{\uparrow}\cap R y \neq \emptyset$. In other words, $R y$ is directed (and always a lower set). Conversely, if that is the case then, for each finite subset $F$ of $Q$, one deduces from $y\in (F^{\uparrow})^{\circ}$ that $F$ is contained in $R y$, whence there is an upper bound $u$ of $F$ in $R y$; thus, $y\in \bigcup\,\{ ({\uparrow\!u})^{\circ}\! : u \!\in\! F^{\uparrow}\}$. The reverse inclusion $\bigcup\,\{ ({\uparrow\!u})^{\circ}\! : u \!\in\! F^{\uparrow}\} \!\subseteq\! (F^{\uparrow})^{\circ}$ is clear. (2) Obviously, $\diamondsuit$-stable quasi-ordered spaces are $\curlyvee$- and $\curlywedge$-stable. Conversely, suppose $T$ is $\curlyvee$-and $\curlywedge$-stable. Any $Y\in \diamondsuit Q$ is of the form $Y = \bigcap \,\{ {\uparrow\!F_i} : i \! < \! n\}$ where each of the finitely many sets $F_i$ is finite. Hence, $Y^{\circ} = \bigcap \,\{ ({\uparrow\!F_i})^{\circ} : i < n \} =$ \hfill { ($\curlyvee$-stability)} $\bigcap \,\{ \,\bigcup \,\{ ({\uparrow\!x})^{\circ} : x\in F_i \} : i < n \} =$ \hfill (set distributivity) $\bigcup \,\{ \,\bigcap\,\{ ({\uparrow\!\chi_i})^{\circ}\! : i \!<\! n \} : \chi \in \prod_{i<n}\! F_i \} =$ \hfill { ($\curlywedge$-stability)} $\bigcup\,\{ ({\uparrow\!y})^{\circ}\! : y\in \bigcap\,\{{\uparrow\!\chi_i} : i<n\}, \,\chi \in \prod_{i<n}\! F_i \} =$ \hfill (set distributivity) $\bigcup\,\{ ({\uparrow\!y})^{\circ} : y\in \bigcap\,\{ {\uparrow\!F_i} : i<n\} = Y \}.$ (3) If ${\rm U}T$ is a web space then its interior operator preserves finite unions \cite{Eweb}; in particular, for finite $F$, one has $({\uparrow\!F})^{\circ}\! = \bigcup \,\{ ({\uparrow\!y})^{\circ}\! : y\in F\} = \bigcup \,\{ ({\uparrow\!z})^{\circ}\! : z\in {\uparrow\!F}\}$ (as $y\leq z$ implies ${\uparrow\!z}\subseteq {\uparrow\!y}$). (4) In a join-semilattice $Q$ with least element, $\curlywedge Q$ is the set of all principal filters. (5) If $Q$ is a lattice with least element and ${\cal T}$-continuous unary meet-operations $\wedge_x : y \mapsto x\wedge y$, then ${\rm U}T$ is a web space \cite{Epatch}; hence, $T$ is $\curlyvee$-stable by (3). Furthermore, $T$ is $\curlywedge$-stable by (4), and finally $\diamondsuit$-stable by (2). (6) Let $T =(X,\leq,{\cal T} )$ be a lower semi-qospace. Then $\leq$ is the specialization order of ${\cal T}^{\leq}$. The upper space ${\rm U} T = (X,{\cal T}^{\leq})$ is a core space iff $U \!= \bigcup\,\{ ({\uparrow\!y})^{\circ}\! : y \!\in\! U\}$ for all $U \!\in\! {\cal T}^{\leq}$, which is equivalent to requiring that $T$ is $\curlyvee$-stable and satisfies the equation $U = \bigcup\,\{ ({\uparrow\!F})^{\circ} : F\subseteq U, \, F \mbox{ finite}\}$ for all $U\in {\cal T}^{\leq}$. But the latter condition means that the upper space is locally hypercompact. \EP By an {\em mc-ordered space} we mean an ordered space such that every monotone net in the space or, equivalently, every directed subset of the space, regarded as a net, has a supremum to which it converges. It is shown in \cite{Epatch} that the strongly convex mc-ordered semi-pospaces are exactly the patch spaces of monotone convergence spaces (d-spaces), and that the Lawson spaces of domains are just the hyper\-convex, mc-ordered and upper m-determined semi-pospaces, where an ordered space is {\em upper m-determined} iff every upper set $U$ intersecting all directed sets whose closure meets $U$ is open. We are prepared for the main result in this section, characterizing the Lawson spaces of continuous domains in various ways. \begin{theorem} \label{pospace} For an ordered space $T$, the following conditions are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] $T$ is the Lawson space of a continuous domain. \item[{\rm (2)}] $T$ is a fan space whose upper space is sober (or a d-space). \item[{\rm (3)}] $T$ is the Lawson space of a meet-continuous domain, $\curlywedge$-stable and T$_2$. \item[{\rm (4)}] $T$ is a hyperconvex, mc-ordered and core stable (semi-)pospace. \item[{\rm (5)}] $T$ is hyperconvex, mc-ordered, ${\uparrow}$-stable, $\diamondsuit$-stable, and T$_2$. \end{itemize} \vspace{-.5ex} Moreover, `T$_2$' may be replaced by `T$_2$-ordered', by 'upper T$_3$-ordered', and by `T$_3$'. \end{theorem} \noindent {\it Proof.} (1)$\,\Rightarrow\,$(2): Let $P$ be a continuous domain with $T = \Lambda P$. By Corollary \ref{Cspace}, its upper space $\Sigma P$ is a sober core space, and by Theorem \ref{fans}, $\Lambda P$ is a fan space. (2)$\,\Rightarrow\,$(3): By Theorem\,\ref{fans}, $T =(P,{\cal T} )$ is regular, upper regular and T$_1$, i.e., T$_3$ and upper T$_3$-ordered, {\it a fortiori} T$_2$-ordered; the upper space ${\rm U} T$ is a core space and a d-space. By Corollary\,\,\ref{Cspace}, $P$ is a continuous, hence meet-continuous domain, ${\cal T}^{\leq}$ is its Scott topology $\sigma P$, and ${\cal T} \!=\! {\cal T}^{\leq \upsilon}$ is the Lawson topology $\lambda P$. Since ${\rm U} T$ is a core space, $T$ is $\curlywedge$-stable, by Theorem \ref{Cspace}\,(1) and Lemma\,\ref{stable}\,(1). (2)$\,\Rightarrow\,$(4): By Theorem \ref{fans}, ordered fan spaces are hyperconvex core stable pospaces, and such spaces are mc-ordered if their upper space is a d-space. (3)$\,\Rightarrow\,$(5): If $P$ is a meet-continuous domain then $\Sigma P$ is a web space, so its weak patch space, the Lawson space $\Lambda P$, is web-ordered, hence ${\uparrow}$-stable (Proposition\,\ref{webpatch}); it is a hyperconvex semi-pospace, and mc-ordered since $\Sigma P$ is a d-space. By Lemma \ref{stable}\,(3), $T$ is $\curlyvee$-stable, and if it is $\curlywedge$-stable, it is $\diamondsuit$-stable by Lemma \ref{stable}\,(2). (4)$\,\Rightarrow\,$(5): Core stable semi-pospaces are ${\uparrow}$-stable, $\diamondsuit$-stable and T$_2$ (Lemma\,\ref{cstable}). (5)$\,\Rightarrow\,$(1): By Lemma \ref{stable}\,(1), the sets $R y = \{ x: y\!\in\! ({\uparrow\! x})^{\circ}\}$ are directed. As $T$ is mc-ordered, $R y$ has a join $x \!=\! \bigvee R y$. Assume $x \!<\! y$ and choose disjoint sets $V,W\in {\cal T}$ with $x\in V$ and $y\in W$. By hyperconvexity, there is a $U \in {\cal T}^{\leq}$ and a finite set $F$ with $x\in U\setminus {\uparrow\! F}\subseteq V$. Then $y\in U\cap W\subseteq {\uparrow\! F}$ (as $x < y$ and $V\cap W = \emptyset$), and $U\cap W\in {\cal T}$ yields ${\uparrow\!(U\cap W)}\in {\cal T}^{\leq}$ by ${\uparrow}$-stability. Thus, $y\in ({\uparrow\! F})^{\circ} = \bigcup\,\{ ({\uparrow\! u})^{\circ}\! : u\!\in\! F\}$ by $\curlyvee$-stability, and so $u\in R y$ for some $u\in F$, which leads to the contradiction $u\leq \bigvee \! R y = x\in U\setminus {\uparrow\! F}$. Hence, $y$ is the directed join of\,\,$R y$. Since $T =(P,{\cal T} )$ is mc-ordered, ${\cal T}^{\leq}$ is coarser than $\sigma P$. It follows that $R y$ coincides with the way-below ideal $\ll\! y$ (indeed, $x \ll y$ implies $x\,R\,y$, since $R y$ is an ideal with join $y$; and $x\,R\, y$ implies $x \!\ll\! y$, since for directed $D$, $x \,R \, y = \bigvee\! D$ entails $y\in int_{{\cal T}^{\leq}} {\uparrow\!x} \subseteq int_{\sigma P} {\uparrow\!x} \subseteq x\!\!\ll$). Therefore, $P$ is continuous, ${\cal T}^{\leq} = \sigma P$ (as the base $\{ xR : x\in P\}$ of $\sigma P$ is contained in ${\cal T}^{\leq}$), and finally, ${\cal T} = {\cal T}^{\leq \upsilon} = \lambda P$. \EP Since all topologies on join-semilattices with 0 are $\curlywedge$-stable, we obtain at once: \begin{corollary} \label{complete} A complete\,lattice\,$L$ is continuous iff $\Sigma L$ is a web space and $\Lambda L$\,is\,T$_2$. In that case, $\Lambda L$ is an (upper) regular pospace (and in {\sf ZFC}, it is even compact). \end{corollary} In categorical terminology, parts of Corollary \ref{contdom} and Theorem \ref{pospace} read as follows: \begin{proposition} \label{iso} The category {\bf CD} of continuous domains and maps preserving directed joins is concretely isomorphic, by virtue of the Scott functor $\Sigma$, to the category {\bf SCS} of sober core spaces, and by virtue of the Lawson functor $\Lambda$, to the category {\bf MFS} of mc-ordered fan spaces with isotone lower semicontinuous maps. The inverse isomorphism $\Lambda^-$ is induced by the forgetful functor $(P,{\cal T}) \mapsto P$. \end{proposition} \begin{picture}(300,60)(-80,0) \put(82,51){{\bf CD}} \put(35,10){{\bf SCS}} \put(119,10){{\bf MFS}} \put(78,45){\vector(-1,-1){20}} \put(58,35){$\Sigma$} \put(63,25){\vector(1,1){20}} \put(77,29){$\Sigma^{\!-}$} \put(102,45){\vector(1,-1){20}} \put(117,35){$\Lambda$} \put(117,25){\vector(-1,1){20}} \put(93,29){$\Lambda^{\!-}$} \put(65,11){\vector(1,0){46}} \put(110,14){\vector(-1,0){46}} \put(85,17){${\rm U}$} \put(83,0){${\rm P}_{\upsilon}$} \end{picture} \vspace{1ex} Suitable examples show that the properties in Theorem \ref{pospace}\,(5) are independent; that is, none of them follows from the combination of the other four properties. \begin{example} \label{Ex51} {\rm For ${\mathbb I} = \,]\,0,1\,] \subseteq {\mathbb R}$, the left half-open interval topology ${\cal T} = \upsilon \hspace{.2ex}{\mathbb I}\,^{\alpha}$ makes ${\mathbb I}$ a pospace with all the properties in (5) except hyper- resp.\ strong convexity: for any interval $]\,0,y\,]$ with $0 \!<\! y \!<\! 1$ and any upper set $U$, there is no lower set $V = \,]\,0,\,z\,[$ with $y\in U \cap V \subseteq \ ]\,0,y\,]$, because any $x\in \ ]\,y,z\,[$ belongs to $U\cap V$. } \end{example} \begin{example} \label{Ex52} {\rm All finite products of chains are continuous posets; but they are domains only if all nonempty subsets have joins. The pospaces $\Lambda \,{\mathbb Q}^n$ and $\Lambda \,{\mathbb R}^n$ have all properties in (5) except of being mc-ordered, as neither ${\mathbb Q}$ nor ${\mathbb R}$ has a join. } \end{example} \begin{example} \label{Ex53} {\rm Consider the subset $X = \{ 0\} \cup \, [ \,1,2\, ] \, \cup \{ 3 \}$ of ${\mathbb R}$, equipped with the Euclidean topology inherited from ${\mathbb R}$, and the order $\sqsubseteq $ defined by $$x \sqsubseteq y \ \Leftrightarrow \ x = y \mbox{ or } x = 0 \mbox{ or } y = 1 \mbox{ or } (x = 2 \mbox{ and } y \in \,]\,1,2\,[\,).$$ \vspace{-1.5ex} \begin{picture}(200,70)(-130,0) \put(90,30){\circle{5}} \put(95,27){$3$} \put(87.5,32){\line(-3,2){40}} \put(87.5,28){\line(-3,-2){40}} \put(45,15){\circle{5}} \put(45,13){\line(0,-1){10}} \put(45,60){\circle{5}} \put(35,58){$1$} \put(45,0){\circle{5}} \put(33,34){$]\,1,2\,[$} \put(35,10){$2$} \put(35,-4){$0$} \put(43,17){\line(-2,3){8}} \put(47,17){\line(2,3){8}} \put(43,57.5){\line(-2,-3){8}} \put(47,57.5){\line(2,-3){8}} \end{picture} \vspace{2ex} \noindent This is a compact pospace, hence mc-ordered \cite[VI-1]{CLD} and T$_2$. It is $\curlyvee$-stable, since $({\uparrow\!F})^{\circ} = {\uparrow (F\cap \{ 0,2\})} \cup (F\cap \{ 3\}) = \bigcup \{ ({\uparrow\!x})^{\circ}: x\in F\}$ for all finite subsets\,\,$F$. It is $\curlywedge$-stable (because $(X,\sqsubseteq )$ is a complete lattice), and therefore $\diamondsuit$-stable (Lemma \ref{stable}\,(2)). Furthermore, it is hyperconvex, in view of the equations $\{ 0\} = X\setminus {\uparrow\!\{ 2,3\}}$, $\{ 3 \} = (X\setminus \{ 0\} )\setminus {\uparrow\!2}$, and $U = {\uparrow\!U}\setminus {\uparrow\!1}$ for $U \subseteq \ ]1,2\,]$. \noindent But this pospace fails to be ${\uparrow}$-stable, since $\{ 3\}$ is open, while $\{ 1,3\} = {\uparrow\!3}$ is not. } \end{example} \begin{example} \label{Ex54} {\rm If $(X,{\cal T} )$ is a nonempty T$_2$-space whose finite subsets have empty interior (e.g.\ ${\cal T} = \lambda \,{\mathbb R}$) then the pospace $(X, =, {\cal T} )$ is trivially up- and $\curlyvee$-stable, hyperconvex and mc-ordered, and it is almost $\curlywedge$-stable: for finite subsets $F$ with at least two elements, one has $F^{\uparrow} = \emptyset$, hence $(F^{\uparrow})^{\circ} = \bigcup \,\{ ({\uparrow\!y})^{\circ} :y\in F^{\uparrow}\} = \emptyset$. But, ``by the skin of its teeth'', such a pospace fails to be $\curlywedge$-stable, because for $F = \emptyset$, $F^{\uparrow}$ is the whole ground set $X$, whence $(F^{\uparrow})^{\circ} = X \neq \emptyset = \bigcup \,\{ ({\uparrow\!y})^{\circ} :y\in F^{\uparrow}\! = X\}.$ } \end{example} \begin{example} \label{Ex55} {\rm The completed open real square $S \!= \,]\,0,1\,[^2 \,\mathop{\cup}\, \{ (0,0),(1,1) \}$ endowed with the interval topology is an irreducible, ${\uparrow}$-stable compact semi-pospace, but not T$_2$ (cf.\ Example 3.5 in \cite{Epatch}). However, it is hyperconvex, mc-ordered, $\curlyvee$- and $\curlywedge$-stable, and so $\diamondsuit$-stable (Lemma \ref{stable}\,(2)). But this space heavily fails to be web-ordered, since the only web neighborhood of any point is the whole space. } \end{example} \noindent {\bf Note.} The basic results about core spaces and fan spaces are more than 30 years old; they have been reported by the author at the Annual Meeting of the\,\,DMV, Dortmund 1980\,\cite{Evdt} and at the Summer School on Ordered Sets and Universal Algebra, Donovaly 1985 but have not been published in a systematic treatise until now. A common theory of web spaces and core spaces is possible by introducing so-called {\em $\kappa$-web spaces}, where $\kappa$ is a cardinal parameter; that approach was initiated in \cite{Eweb}. A comprehensive theory of so-called {\em $\zeta$-domains} and their topological manifestation, providing common generalizations of continuous, hypercontinuous, algebraic and many other variants of domains or posets, is presented in \cite{Edom}. \section{Semitopological and topological semilattices} \label{semi} Let us now apply some of the previous results to the situation of {\em semilattices}, by which we always mean {\em meet-semilattices}. A {\em semilattice-ordered space} is an ordered space $S$ whose underlying poset is a semilattice; by slight abuse of language, we call it {\em compatible} if the topology is compatible with the semilattice order, and we call a semilattice- and T$_1$-ordered space a T$_1${\em -semilattice}. A {\em semitopological semilattice} is a semilattice-ordered\,\,space whose unary meet operations \mbox{$\wedge_x \!: y \mapsto x\wedge y$} are continuous, while in a {\em topological semilattice} the binary meet operation is continuous. \vspace{-.5ex} \begin{lemma} \label{top} A semilattice-ordered space $S =(X,\leq,{\cal T})$ is a topological semilattice whenever one of the following conditions is fulfilled: \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] $S$ is compatible and locally filtered (a wide web space). \item[{\rm (2)}] $S$ has a subbase of sets whose complements are filtered. \item[{\rm (3)}] $S$ carries the weak lower topology: ${\cal T} = \upsilon (X,\geq)$. \item[{\rm (4)}] $S$ is a hyperconvex, locally filtered and ${\uparrow}$-stable semi-pospace. \end{itemize} \end{lemma} \vspace{-.5ex} \noindent {\it Proof.} (1) For $x,y\in X$ and a filtered neighborhood $D$ of $x\wedge y$ with $D\subseteq U \in {\cal T} = {\cal T}^{\leq}$, the up-closure $F = {\uparrow\!D}$ is a filter still contained in $U$. For $W = int_{{\cal T}} F$, we obtain $W\wedge W \subseteq F\wedge F \subseteq F \subseteq U$ and $x,y\in W$, as $x\wedge y \in W = {\uparrow\!W}$. (2) Let $V$ be a subbasic open set such that $F = X\setminus V$ is a filter. If $x,y\in X$ satisfy $x\wedge y\in V$ then $x\in V$ or $y\in V$ (otherwise $x\wedge y \in F$). Hence, $(x,y)$ lies in $W = (V \!\times\! X)\cup (X \!\times\! V)$, and that is an open set in $S^2$ with $\wedge\, [W] \subseteq V$ (indeed, $(u,v)\in W$ implies $u\in V$ or $v\in V$, and so $u\wedge v\in V$, since $V$ is a lower set). (3) follows from (2), because $\upsilon (X,\geq )$ has a closed subbase of principal filters. (4) The ordered space $(X,\leq,{\cal T}^{\leq})$ is compatible and locally filtered: for \mbox{$x \!\in\! U \!\in\! {\cal T}^{\leq}$,} find a filtered $D\subseteq U$ with $x\in W = int_{{\cal T}}D$; then ${\uparrow\!D}$ is a filter with $x \!\in\! W \!\subseteq\! {\uparrow\!D} \!\subseteq\! U$, and ${\uparrow\!W}$ is ${\cal T}^{\leq}$-open by ${\uparrow}$-stability. By (1), the operation $\wedge$ is ${\cal T}^{\leq}$-continuous; by (3), it is $\upsilon (X,\geq)$-continuous, and then, by hyperconvexity, it is ${\cal T}$-continuous. \EP We find it convenient to call a topological semilattice {\em s-topological} (resp.\ {\em sc-topological}) if it has small semilattices (resp.\ small convex semilattices), that is, each point has a neighborhood base consisting of (convex) subsemilattices (cf.\ \cite[VI-3]{CLD}). We are ready for the characterization of hyperconvex semitopological, resp.\ s-topological T$_1$-semilattices as certain specific web-ordered spaces. \begin{theorem} \label{topsemi} Let $S$ be a hyperconvex T$_1$-semilattice or, equivalently, the weak patch space of a compatible semilattice-ordered space. \noindent {\rm (1)} The following three conditions are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (w11)}] $S$ is the weak patch space of a web space. \item[{\rm (w12)}] $S$ is a web-ordered space. \item[{\rm (w13)}] $S$ is a semitopological semilattice. \end{itemize} \noindent {\rm (2)} The following three conditions are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (w21)}] $S$ is the weak patch space of a wide web space. \item[{\rm (w22)}] $S$ is a locally filtered and ${\uparrow}$-stable ordered space. \item[{\rm (w23)}] $S$ is an s(c)-topological semilattice. \end{itemize} \noindent {\rm (3)} The following three conditions are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (w31)}] $S$ is the weak patch space of a worldwide web space (a core space). \item[{\rm (w32)}] $S$ is a core stable pospace (a fan space). \item[{\rm (w33)}] $S$ is an s(c)-topological semilattice, and ${\rm U}S$ is locally compact. \end{itemize} \end{theorem} \noindent {\it Proof.} Notice that hyperconvexity ensures closedness of all principal filters, so it suffices to assume that $S$ is a hyperconvex semilattice-ordered lower semi-pospace. (w11)$\,\Leftrightarrow \,$(w12): Apply Proposition \ref{webpatch} to $\zeta = \upsilon$. (w12)$\,\Rightarrow \,$(w13): By Proposition \ref{webpatch}, the upper space $(X,{\cal T}^{\leq})$ of $S = (X,\leq, {\cal T})$ is a web space, and $S$ is its weak patch space. Therefore, as remarked in \cite{Eweb} and \cite{Epatch}, the unary meet operations $\wedge_x$ are ${\cal T}^{\leq}$-continuous; by Lemma \ref{top}\,(3), they are $\upsilon (X,\geq)$-continuous, and by hyperconvexity, they are also ${\cal T}$-continuous. (w13)$\,\Rightarrow \,$(w12): This was shown in \cite{Eweb}. (w21)$\,\Leftrightarrow \,$(w22): Apply Theorem \ref{wideweb} to $\zeta = \upsilon$. (w22)$\,\Rightarrow \,$(w23): By Lemma \ref{top}\,(4), $S$ is a topological semilattice. Given $x \!\in\! O \!\in\! {\cal T}$, use hyperconvexity and local filteredness in order to find a $U\in {\cal T}^{\leq}$, a finite set $F$, and a filtered set $D$ such that $x\in int_{{\cal T}} D \subseteq D \subseteq U \setminus {\uparrow\!F} \subseteq O$. Then ${\uparrow\!D}\setminus {\uparrow\!F}$ is a convex subsemilattice with $x\in int_{{\cal T}} D \subseteq {\uparrow\!D}\setminus {\uparrow\!F} \subseteq U\setminus {\uparrow\!F} \subseteq O$, which shows that $S$ has small convex semilattices. (w23)$\,\Rightarrow \,$(w22): By continuity of the unary operations $\wedge_x$, $S$ is ${\uparrow}$-stable: indeed, $O\in {\cal T}$ implies ${\uparrow\!O} = \bigcup\,\{ \wedge_x^{-1}[O] : x\in O\} \in {\cal T}$. Clearly, subsemilattices are filtered. The equivalence of (w31), (w32) and (w33) is now easily verified with the help of (2), Theorem \ref{fans} and Proposition \ref{supertop}, which says, among other things, that the core spaces are exactly the locally compact web spaces (cf.\ \cite[VI--3.3]{CLD}). \EP \vspace{1ex} Combining Theorem \ref{pospace} with Theorem \ref{topsemi}, we arrive at \begin{corollary} \label{contsemi} For a semilattice-ordered space $S$, the following are equivalent: \vspace{-.5ex} \begin{itemize} \item[{\rm (w41)}] $S$ is the weak patch space of a (unique) sober core space. \item[{\rm (w42)}] $S$ is an mc-ordered fan space. \item[{\rm (w43)}] $S$ is a hyperconvex, mc-ordered, s-topological T$_1$-semilattice with locally compact upper space ${\rm U}S$. \item[{\rm (w44)}] $S$ is the Lawson space of a continuous semilattice. \end{itemize} The weak patch functor ${\rm P}_{\upsilon}$ induces concrete isomorphisms between \vspace{-.5ex} \begin{itemize} \item[{\rm (1)}] the category of compatible semitopological semilattices,\,i.e.\,semilattice-ordered web spaces, and that of hyperconvex semitopological T$_1$-semilattices, \item[{\rm (2)}] the category of compatible s-topological semilattices,\,i.e.\,semilattice-ordered wide web spaces, and that of hyperconvex s-topological T$_1$-semilattices, \item[{\rm (3)}] the category of locally compact compatible s-topological semilattices,\,i.e.\,semi\-lattice-ordered core spaces, and that of topological semilattice fan spaces, \item[{\rm (4)}] the category of Scott\,spaces of continuous\,semilattices,\,i.e.\ semilattice-ordered sober core spaces, and that of Lawson spaces of continuous\,semilattices,\,i.e.\ mc-ordered topological semilattices that are fan spaces. \end{itemize} \end{corollary} \noindent Of course, most elegant results are available for compact pospaces. From \cite[IV--1]{FL} or \cite[VI--1]{CLD}, we learn the following facts: every compact pospace is \vspace{-.5ex} \begin{itemize} \item[--] mc-ordered and dually mc-ordered, \item[--] monotone normal, in particular upper and lower regular, \item[--] strongly convex. \end{itemize} For any closed subset $C$ of a compact pospace $T$, the sets ${\uparrow\!C}$ and ${\downarrow\!C}$ are closed, but $T$ need neither be ${\uparrow}$-stable nor locally filtered: see Examples \ref{Ex33} and \ref{Ex53}. A semilattice is said to be {\em complete} if all directed subsets have suprema and all nonempty subsets have infima (this together with the existence of a top element defines a complete lattice). Recall from \cite[IV--3]{CLD} the famous \vspace{1ex} \noindent {\bf Fundamental Theorem of Compact Semilattices}\\ {\em The Lawson spaces of complete continuous semilattices are exactly the compact T$_2$ s-topological semilattices. } \vspace{1ex} This is now a rather easy consequence of the previous facts, the Ultrafilter Theorem (giving compactness of $\Lambda L$ \cite{EPS}) and the following ``non-complete'' version: \begin{proposition} \label{compact} The compact Lawson spaces of continuous domains are exactly the locally filtered, ${\uparrow}$-stable compact pospaces. \end{proposition} \noindent {\it Proof.} If $P$ is a continuous domain then, by Theorems \ref{fans} and \ref{pospace}, the Lawson space $\Lambda P$ is a fan space, hence a locally filtered ${\uparrow}$-stable pospace. Conversely, if $T = (X,\leq,{\cal T})$ is any locally filtered, ${\uparrow}$-stable compact pospace then $T$ is upper regular, mc-ordered and dually mc-ordered; by Lemma \ref{cstable}, it is d-stable, and its upper space is a core space and a d-space (because $T$ is mc-ordered); hence, by Corollary \ref{contdom}, it is the Scott space of the continuous domain $P =(X,\leq)$. It follows that the Lawson topology $\lambda P = \sigma P \vee \upsilon \widetilde{P}$ is contained in ${\cal T} = {\cal T}^{\leq}\vee {\cal T}^{\geq}$ (as $T$ is strongly convex). Now, $\lambda P$ is T$_2$ and ${\cal T}$ is compact, so both topologies must coincide. \EP In \cite[III--5]{CLD}, one finds a whole collection of various equivalent characterizations of continuous domains that are compact in their Lawson topology. For a detailed study of joint and separate continuity of operations in posets and lattices, see \cite{EG}. \newpage \section{Domain bases and core bases} Following \cite{CLD}, we mean by a {\em basis} of a domain $P$ a subset $B$ such that for each $y\in P$, the set $\{ b \in\! B: b \ll y\}$ is directed with join $y$. The pair $(P,B)$ is then referred to as a {\em based domain}. Note that $P$ is continuous iff it has at least one basis. By a {\em core basis} for a space $(X,{\cal S})$, we mean a subset $B$ of $X$ such that for all $U\in {\cal S}$ and all $y\in U$, the set $R_{{\cal S}}y$ meets $B\cap U$; i.e., $y\in int_{{\cal S}}({\uparrow\!x}) \subseteq {\uparrow\!x} \subseteq U$ for some $x\in B$; in other words, all points have neighborhood bases formed by cores of elements of $B$. (To avoid ambiguities, we use the word {\em basis} for subsets of $X$ and the word {\em base} for subsets of the power set of $X$.) Thus, a space is a core space iff it has a core basis. By a {\em core based space} we mean a pair consisting of a (core) space and a core basis of it. The following extension of Corollary \ref{contdom} is straightforward: \begin{lemma} \label{basis} The bases of a domain are the core bases of its Scott space. Hence, via the Scott functor $\Sigma$, the based domains correspond to the core based sober spaces. \end{lemma} \begin{proposition} \label{Cbasis} The C-ordered sets are exactly the pairs $(B,\ll\!\!|_B)$ where $B$ is a basis of a domain (which is uniquely determind up to isomorphism by $(B,\ll\!\!|_B)$). \end{proposition} \noindent {\it Proof.} That, for any basis $B$ of a domain, the pair $(B,\ll\!\!|_B)$ is a C-ordered set follows easily from the interpolation property of $\ll$ (cf.\ Lemma \ref{conti}). Conversely, any C-ordered set $(X,R )$ is isomorphic to $({\cal B}_{R}, \ll\!\!|_{{\cal B}_{R}})$, where ${\cal B}_{R} = \{ R x : x\in X\}$ is a basis of the continuous domain ${\cal I}_{R}$ of rounded ideals (see Theorem \ref{Cspace}\,(4)). \EP \begin{corollary} \label{Cor} The T$_0$ core spaces are exactly the core bases of sober core spaces, equipped with the induced topology. \end{corollary} On the pointfree side, we define a {\em based supercontinuous lattice} to be a pair consisting of a supercontinuous (i.e.\ completely distributive) lattice $L$ and a {\em (coprime) basis} of $L$, that is, a join-dense subset of coprime elements. We are ready for six different descriptions of C-ordered sets: \begin{theorem} \label{Cequiv} The following six categories are mutually equivalent: \vspace{1ex} \begin{tabular}{|l|l|} \hline objects & morphisms\\ \hline \hline C-ordered sets & interpolating isotone maps\\ \hline T$_0$ core spaces & continuous maps\\ \hline fan ordered spaces & lower semicontinuous isotone maps\\ \hline based domains & maps preserving the bases and directed joins\\ \hline core based sober spaces & continuous maps preserving the core bases\\ \hline based supercontinuous lattices & maps preserving the bases and joins\\ \hline \end{tabular} \end{theorem} \vspace{1ex} \noindent On the object level, these equivalences easily follow from Theorem \ref{Cspace}, Corollary\,\ref{contdom}, Theorem \ref{fans}, Lemma \ref{Cbasis}, Proposition \ref{Cequiv} and Corollary \ref{Cor}. The verification of the claimed correspondences between the morphisms is left as an exercise. \newpage \section{Density and weight} \label{Cbases} We now are looking for characterizations of core bases in terms of the interior relation. Recall that the lower quasi-order $\leq_{R}$ of an arbitrary relation $R$ on a set $X$ is given by $x \leq_{R} y \,\Leftrightarrow\, R x \subseteq R y$. Now, we say a subset $B$ of $X$ is \begin{itemize} \item[--] {\em $R$-dense} if $x\,R\,y$ implies $x\,R\,b$ and $b\,R\,y$ for some $b\in B$, \item[--] {\em $R$-cofinal} if $x\,R\,y$ implies $x\!\leq_{R}\!b$ and $b\,R\,y$ for some $b\in B$. \end{itemize} It is straightforward to see that $B$ is $R$-dense and $R$ is transitive iff $B$ is $R$-cofinal and $R$ is idempotent. Note that the {\em strong patch topology} ${\cal S}^{\alpha} = {\cal S} \vee \alpha (X,\geq)$ of a space $(X,{\cal S} )$ concides with the {\em Skula topology} generated by ${\cal S} \cup {\cal S}^c$ \cite{Sk}. Hence, the sets $C\setminus D$ with $C,D\in {\cal S}^c$ form a base for ${\cal S}^{\alpha}$. \begin{proposition} \label{corebase} Let $(X,{\cal S})$ be a core space and $R$ the corresponding interior relation. For a subset $B$ of $X$, the following conditions are equivalent: \begin{itemize} \item[{\rm (1)}] $B$ is $R$-dense in $X$. \item[{\rm (2)}] $B$ is $R$-cofinal in $X$. \item[{\rm (3)}] $B$ is a core basis for $(X,{\cal S})$. \item[{\rm (4)}] $B$ is dense in $(X,{\cal S}^{\alpha})$ and so in any patch space of $(X,{\cal S} )$. \item[{\rm (5)}] $\{ {\downarrow\!b} : b\in B\}$ is join-dense in the coframe ${\cal S}^c$ of closed sets. \end{itemize} \end{proposition} \noindent {\it Proof.} (1)$\,\Leftrightarrow\,$(2): The above remark applies, as $R$ is idempotent by Theorem \ref{Cspace}\,(1). (1)$\,\Rightarrow\,$(3): $y\in U\in {\cal S}$ means $U = UR$ and $x\,R\,y$ for some $x\in U$. Choose $b\in B$ with $x\,R\, b \, R \, y$. Then, $b\in R y$ and, by transitivity, $b\in xR \subseteq x\!\leq_{R} \ = {\uparrow\!x} \subseteq U$. (3)$\,\Rightarrow\,$(4): If $C,D$ are ${\cal S}$-closed sets with $C \!\not\subseteq\! D$, pick an $x\in C\setminus D$ and find a $b\in B$ with $x\in int_{{\cal S}}({\uparrow\!b}) \subseteq {\uparrow\!b} \subseteq X\setminus D$. It follows that $b\in C$, since $b\leq x \in C = {\downarrow\!C}$. Thus, $b\in B\cap (C\setminus D)$. This proves density of $B$ in ${\cal S}^{\alpha} = {\cal S} \vee {\cal S}^c$. (4)$\,\Leftrightarrow\,$(5) holds for arbitrary spaces. Recall that ${\downarrow\!x}$ is the closure of the singleton $\{ x\}$ in $(X,{\cal S})$. Now, \mbox{$\{ {\downarrow\!b} : b\in B\}$} is join-dense in ${\cal S}^c$ iff for $C,D\in {\cal S}^c$ with $C\not\subseteq D$ there is a $b\in B$ with ${\downarrow\!b}\subseteq C$ but not ${\downarrow\!b}\subseteq D$, i.e.\ $b\in C\setminus D$ (as $C$ and $D$ are lower sets). But the latter means that $B$ meets every nonempty open set in ${\cal S}^{\alpha} = {\cal S} \vee {\cal S}^c$. (4)$\,\Rightarrow\,$(1): Suppose $x\,R\, y$ and choose a $z$ with $x\,R\, z \, R\, y$. Then $xR \in {\cal O}_{R} = {\cal S}$ and therefore $z \in xR \cap {\downarrow\!z} \in {\cal S}^{\alpha}$. Hence, there is a $b \in B \cap xR \cap {\downarrow\!z}$, and it follows that $x \, R \, b \, R \, y$, since $R y$ is a lower set containing $z$ and so $b$. \EP From now on, we assume the validity of the Axiom of Choice. Hence, each set $X$ has a cardinality, represented by the smallest ordinal number equipotent to\,\,$X$. The {\em weight} $w({\cal S})$ resp.\ {\em density} $d({\cal S})$ of a space or its topology ${\cal S}$ is the least possible cardinality of bases resp.\ dense subsets. The weight of a core space is at most the cardinality of any core basis $B$, since $B$ gives rise to a base $\{ bR = int_{{\cal S}}({\uparrow\!b}) : b \in B\}$. \begin{lemma} \label{cardbase} Every core basis of a core space $(X,{\cal S})$ contains a core basis of cardinality $w({\cal S})$. Hence, $w({\cal S})$ is the minimal cardinality of core bases for $(X,{\cal S})$. \end{lemma} \noindent {\it Proof.} Let ${\cal B}$ be an arbitrary base and $B$ a core basis for $(X,{\cal S})$. The Axiom of Choice gives a function picking an element $b_{\,U,V}$\,from each nonempty set of the form $$ B_{\,U,V} = \{ b \in B : U \subseteq {\uparrow\!b} \subseteq V \} \ \ (U,V\in {\cal B} ). $$ Then $ B_0 = \{ b_{\,U,V} : U,V\in {\cal B}, \ B_{\,U,V} \not = \emptyset\} $ is a subset of $B$ and still a core basis; in fact, $x\in V \in {\cal B}$ implies \mbox{$x\in U \subseteq {\uparrow\!b} \subseteq V$} for suitable $b\in B$ and $U\in {\cal B}$, and it follows that $x \in U \subseteq {\uparrow\!b_{\,U,V}} \subseteq V$. If ${\cal S}$ is infinite then so is ${\cal B}$, and consequently $|B_0| \leq |{\cal B}|^2 = |{\cal B}|$. Thus, we get $|B_0| \leq w({\cal S})$, and the remark before yields equality. If ${\cal S}$ is finite then the cores form the least base $\{ {\uparrow\!x} : x\in X\}$, and choosing a set of representatives from these cores, one obtains a core basis of cardinality $w({\cal S})$. \EP \begin{example} \label{Ex62} {\rm Consider the ordinal space $C = \omega \!+\! 2 = \{ 0, 1, ..., \omega, \omega\!+\!1\}$ with the upper (Scott) topology. While $\omega \cup \{ \omega\!+\!1\} = C\setminus \{ \omega \}$ is a core basis, the set $B = \omega\!+\!1 = \omega \cup \{ \omega\}$ is {\em not} a core basis: $U \!= \{ \omega\!+\!1\}$ is $\upsilon$-open but disjoint from $B$. Nevertheless, $\upsilon C \setminus \{ \emptyset\} = \{ bR : b \in B\}$ is a base (note $xR ={\uparrow\!x}$ for $x\neq \omega$). } \end{example} Generalizing the weight of topologies, one defines the {\em weight} $w(L)$ of a lattice $L$ as the least possible cardinality of join-dense subsets of $L$. For any relation $R$ on a set $X$, we define the {\em $R$-cofinality}, denoted by $c(X,R)$, to be the minimal cardinality of $R$-cofinal subsets of $X$. If $R$ is idempotent then $c(X,R)$ is also the {\em $R$-density}, the least cardinality of $R$-dense subsets. From Proposition \ref{corebase} and Lemma \ref{cardbase} we infer: \begin{theorem} \label{w} For a core space $(X,{\cal S})$ with interior relation $R$ and any patch topology ${\cal T}$ of ${\cal S}$, the density of $(X,{\cal T})$ is equal to each of the following cardinal invariants: \vspace{-.5ex} $$ c(X,R) = w({\cal S}) = w({\cal S}^c) = w({\cal S}^{\upsilon}). $$ \end{theorem} Indeed, if $B$ is a core basis for $(X,{\cal S})$ then ${\cal B} = \{ bR \setminus {\uparrow\!F} : b\in B,\, F \!\subseteq\! B, \, F \,\mbox{finite}\}$ is a base for ${\cal S}^{\upsilon}$, and if $B$ is infinite, ${\cal B}$ and $B$ have the same cardinality. If $B$ is finite, \mbox{the base $\{ {\uparrow\!b} : b \in B\}$ of ${\cal S}$ is equipotent to the base $\{ {\uparrow\!b}\cap {\downarrow\!b} : b \in B\}$ of ${\cal S}^{\upsilon} = {\cal S}^{\alpha}$.} Since in {\sf ZCF} any supercontinuous lattice is isomorphic to the topology of a core space \cite{EABC}, Theorem \ref{w} entails a fact that was shown choice-freely in \cite{ECD}: \begin{corollary} The weight of a supercontinuous (i.e.\ completely distributive) lattice is equal to the weight of the dual lattice. \end{corollary} \begin{example} \label{Ex61} {\rm On the real line with the upper (Scott) topology ${\cal S}$, the interior relation is the usual $<\,$. The rationals form a core basis ${\mathbb Q}$, being $<$-dense in ${\mathbb R}$. Hence, \vspace{-1ex} $$\omega = c({\mathbb R},<) = w({\cal S}) = w({\cal S}^c) = w({\cal S}^{\upsilon}) \neq w({\cal S}^{\alpha}) $$ because the half-open interval topology ${\cal S}^{\alpha}$ has no countable base. } \end{example} \newpage
5d70cdcad4f6f6cc3f83289220a137ba17b24018
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{I. \ Objective function used to correct the ZZ-error channel} Thanks to the isomorphism between SU(2) generators, $\sigma_X,\sigma_Y, \sigma_Z$, and the subgroup of SU(4) generators, $\sigma_{ZZ},\sigma_{ZX},\sigma_{IY}$, the right hand side of Eq. (11) can be expressed in a more tractable way as \begin{equation}\label{eq:appendix_1} \left(\prod_{j=4}^{1}\exp\left[i\frac{\psi_j}{2} \sigma_{Z}\right]\left[U\right]^{n_j}\exp\left[-i\frac{\psi_j}{2} \sigma_{Z}\right]\right)U, \end{equation} where $U=\exp\left[-i \frac{5\theta_0}{2}(1+\delta)\sigma_{X}\right]$ and $\theta_0=\arccos\left[\frac{1}{4}\left(\sqrt{13}-1\right)\right]$. The error component of Eq. \eqref{eq:appendix_1} is then isolated by expanding the sequence to first order in $\delta$. The resulting unperturbed matrix and first order error matrix, $A +\delta B $, are expressed in terms of the SU(2) generators as $A=\Lambda_1\sigma_I + i \Lambda_2\sigma_X + i \Lambda_3 \sigma_Y + i \Lambda_4\sigma_Z$ and $B=\Delta_1\sigma_I + i \Delta_2\sigma_X + i \Delta_3 \sigma_Y + i \Delta_4\sigma_Z$, where the $\Delta_i$'s and $\Lambda_i $'s are functions of $\psi_j$'s, $n_j$'s and $\theta_0$. Their closed forms are rather long to include them here, but can be easily obtained with any symbolic computation program.\\ \indent Making use again of the isomorphism between SU(2) and a subgroup of SU(4), we express the local invariants corresponding to $\mathcal{U}^{(6k)}$ in Eq. (11) in terms of the elements of the matrix $A$: \begin{equation} \begin{aligned} G_1(\mathcal{U}^{(6k)})&=(\Lambda_1^2+\Lambda_4^2 - \Lambda_2^2 - \Lambda_3^2)^2\\ G_2(\mathcal{U}^{(6k)})&=3 \Lambda_4^4 + 3 \Lambda_1^4 - 2 \Lambda_1^2 (\Lambda_2^2 + \Lambda_3^2) + 3 (\Lambda_2^2 + \Lambda_3^2)^2 + \Lambda_4^2 (6 \Lambda_1^2 - 2 (\Lambda_2^2 + \Lambda_3^2)). \end{aligned} \end{equation} \indent With the above expressions and the terms that conform the matrix $B$, we construct our objective function such that the error matrix $B$ is canceled and the local invariants of the sequence and target operation are as close as possible. Accordingly, the objective function is given by \begin{equation}\label{eq:objective_function} f=\Delta_1^2+ \Delta_2^2 + \Delta_3^2 + \Delta_4^2 +[G_1(\mathcal{U}^{(6k)})- G_1(\mathfrak{U})]^2 +[G_2(\mathcal{U}^{(6k)})- G_2(\mathfrak{U})]^2, \end{equation} where $G_i(\mathfrak{U})$ are the local invariants of the target operation.\\ \indent The values of the solutions found by numerically minimizing the objective function while targeting a {\sc cnot} operation are: \begin{equation} \begin{aligned} \psi_1=& 1.135268,\\ \psi_2=& -0.405533,\\ \psi_3=& -1.841855,\\ \psi_4=& 0.191753. \end{aligned} \end{equation} Moreover, the angles of the local operations needed to transform $\mathcal{U}^{(6k)}_{\text{\sc cnot}}$ into {\sc cnot}, Eq. (12), are \begin{equation} \begin{aligned} \phi_1=& -1.607820,\\ \phi_2=& 0.234035. \end{aligned} \end{equation} \indent Similarly, the solutions found with the numerical minimization of Eq. \eqref{eq:objective_function} that yield a corrected rotation equivalent to $(5\theta_0/k)_{ZZ}$, for $k=\{5,10,20\}$ respectively, are \begin{equation} \begin{aligned} \psi_1=\{&-0.183589,-0.103032,-0.0522225\},\\ \psi_2=\{&-3.061776,-3.129928,-3.138440\},\\ \psi_3=\{&-2.019322,-2.583841,-2.862841\},\\ \psi_4=\{&1.750803,0.844394,0.418648\}. \end{aligned} \end{equation} \indent Finally, the single-qubit rotation angles in Eq. (13), for $k=\{5,10,20\}$, are \begin{equation} \begin{aligned} \beta_5=&3.111045,\\ \beta_{10}=&2.290846,\\ \beta_{20}=&-1.216184,\\ \gamma_{5}=&-2.117345,\\ \gamma_{10}=&-1.850509,\\ \gamma_{20}=&1.430782. \end{aligned} \end{equation} \section{II. \ Effect of imperfect local gates on the infidelity of the composite pulse sequences} The contour plots in Fig. \ref{fig:CNOT with CK1} present the resulting infidelity of the length-40 (Eq. (10) with $k=20$) and length-120 (Eq. (12) with $k=20$) composite pulse sequences when systematic error in the two-qubit gate and imperfect local gates are taken into account. We apply each sequence to a Hamiltonian formed by an Ising coupling of strength $\alpha$ and random fluctuations only on the SU(4) generators that the particular sequence targets, $H= \alpha\sigma_{ZZ}+\sum \delta_{ij}\sigma_{ij}$ (the length-40 sequence targets all SU(4) generators but $\sigma_{ZZ}$, whereas the length-121 sequence targets all 15 SU(4) generators). As stated in the main text, each local gate of the composite sequence is perturbed by a random local gate of the form $\exp\left[-i \sum \Delta_i\sigma_i\right]\otimes \exp\left[-i \sum \Delta_j \sigma_j\right]$. We analyze separately two types of error that can affect the local gates: systematic and random errors. In order to represent the effect of systematic errors, when the same local gate is invoked multiple times in the sequence, it is invoked with the same perturbation. Whereas for random errors the perturbation is never the same. For each of many realizations of the perturbations, we numerically find the average infidelity of the imperfect local gates invoked as well as the infidelity of the composite pulse sequence, which is formed using those imperfect local gates and it is also perturbed by systematic errors at the two-qubit level. These infidelities are averaged over noise realizations by sampling each stochastic noise variable over a normal distribution of standard deviation $\sigma$, with the average being taken over 500 samples for each value of $\sigma$. \\ \indent As mentioned in the main text and shown in the figures below, the average infidelity of a composite {\sc cnot} caused by errors in the local gates increases with the length of the sequence. For systematic local errors, the {\sc cnot} infidelity increases up to about 80 times the local gate infidelity for the length-120 sequence, which has 121 local gates. On the other hand, random local errors have qualitatively the same effect as systematic ones, but are quantitatively more pernicious, resulting in a {\sc cnot} infidelity increase up to about 480 times the random local gate infidelity for the length-120. Similarly, the {\sc cnot} formed using the length-40 sequence has 41 local gates and about 18 times the local gate infidelity for systematic errors and about 90 times the local gate infidelity for random errors. Fortunately, random errors are typically much smaller to begin with than systematic errors. \section{III. \ Improving the gate fidelity of the cross resonance gate between transmon qubits} In a recent experimental work with the cross resonance (CR) gate between transmon qubits by Sheldon et al. \cite{Sheldon2016}, the two-qubit entangling gate was improved through a detailed Hamiltonian estimation and the application of a cancellation tone that raises the two-qubit fidelity over 99$\%$. The experimental CR Hamiltonian is stated in terms of $\sigma_{ZX}, \ \sigma_{ZY}, \ \sigma_{ZZ}, \ \sigma_{IX}, \ \sigma_{IZ}$, and $ \sigma_{IY}$, of which $\sigma_{ZZ}$ is comparatively small and $\sigma_{IZ}$ is negligible. The authors improve the CR gate by applying a cancellation tone on the target qubit such that unwanted interaction terms of the CR Hamiltonian are eliminated. They choose the cancellation tone phase at which $\sigma_{ZY}$ and $\sigma_{IY}$ are zero, and the amplitude of the cancellation tone is tuned such that $\sigma_{IX}$ and $\sigma_{IY}$ are zero as well. With this technique, the authors report a $\sigma_{ZX}$ gate that is locally equivalent to {\sc cnot} with gate fidelity of 99.1$\%$, an important improvement from previously reported fidelities of 94-96$\%$.\\ \indent Nonetheless, as stated in their work, there are systematic error terms that still remain after the experimental procedure. According to their modeling, the residual error corresponds to a $\sigma_{ZZ}$ term in the Hamiltonian around an order of magnitude smaller than the interaction term, and a $\sigma_{IX}$ term an order of magnitude smaller than $\sigma_{ZZ}$. Following the method presented in the main text, we transform this Hamiltonian into a ZZ coupling by applying a Hadamard transformation to the second qubit. In this context, the dominant error term will give two error channels from the anticommuting set, which can be corrected with the length-5 sequence, Eq. (10) with $k=5$. Using the experimentally reported parameters, and considering that with the length-5 sequence the {\sc cnot} infidelity is about 8 times the average single-qubit gate infidelity, we calculate that our sequence would immediately improve the {\sc cnot} fidelity from the current value of 99.1$\%$ up to 99.6$\%$ with the presently achievable single-qubit gate fidelities of 99.95$\%$ \cite{Sheldon2016a}. All this was calculated with the $\sigma_{IX}$ error present, which, in principle, can be completely corrected from the start by a more precisely tuned cancellation tone. If one were to improve the single-qubit gate fidelities and $T_2$ time, our sequence could further boost the two-qubit fidelity up to 99.98$\%$. \begin{figure}[tbp] \centering \begin{subfigure}[b]{0.45\linewidth} \includegraphics[width=\linewidth]{fig1s_a_Sup_Mat.pdf} \caption{Length-40 sequence, local systematic error }\label{fig:systematiclength40} \end{subfigure} \begin{subfigure}[b]{0.45\linewidth} \includegraphics[width=\linewidth]{fig1s_b_Sup_Mat.pdf} \caption{Length-40 sequence, local random error }\label{fig:randomlength40} \end{subfigure} \begin{subfigure}[b]{0.45\linewidth} \includegraphics[width=\linewidth]{fig1s_c_Sup_Mat.pdf} \caption{Length-120 sequence, local systematic error }\label{fig:systematiclength120} \end{subfigure} \begin{subfigure}[b]{0.45\linewidth} \includegraphics[width=\linewidth]{fig1s_d_Sup_Mat.pdf} \caption{Length-120 sequence, local random error }\label{fig:randomlength120} \end{subfigure} \renewcommand{\baselinestretch}{1} \small\normalsize \begin{quote} \caption{(Color online) Composite {\sc cnot} infidelity vs averaged local gate infidelity vs noise strength ($\sigma/\alpha$). The length-40 sequence is given by Eq. (10) in the main text, with $k=20$. The length-120 sequence is given by Eq. (12) in the main text, with $k=20$.} \label{fig:CNOT with CK1} \end{quote} \end{figure} \renewcommand{\baselinestretch}{2} \small\normalsize \putbib[library] \end{bibunit} \end{document}
e0fc6263b2acf1dcdcec341879b91fa2489ba23b
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The combinatorial structure treated in this paper is a $2 \rightarrow 1$ directed hypergraph defined as follows. \begin{definition} A \emph{$2 \rightarrow 1$ directed hypergraph} is a pair $H = (V,E)$ where $V$ is a finite set of \emph{vertices} and the set of \emph{edges} $E$ is some subset of the set of all pointed $3$-subsets of $V$. That is, each edge is three distinct elements of $V$ with one marked as special. This special vertex can be thought of as the \emph{head} vertex of the edge while the other two make up the \emph{tail set} of the edge. If $H$ is such that every $3$-subset of V contains at most one edge of $E$, then we call $H$ \emph{oriented}. For a given $H$ we will typically write its vertex and edge sets as $V(H)$ and $E(H)$. We will write an edge as $ab \rightarrow c$ when the underlying $3$-set is $\{a,b,c\}$ and the head vertex is $c$. \end{definition} For simplicity from this point on we will always refer to $2 \rightarrow 1$ directed hypergraphs as just \emph{graphs} or sometimes as \emph{$(2 \rightarrow 1)$-graphs} when needed to avoid confusion. This structure comes up as a particular instance of the model used to represent definite Horn formulas in the study of propositional logic and knowledge representation \cite{angluin1992, russell2002}. Some combinatorial properties of this model have been recently studied by Langlois, Mubayi, Sloan, and Gy. Tur\'{a}n in \cite{langlois2009} and \cite{langlois2010}. In particular, they looked at the extremal numbers for a couple of different small graphs. Before we can discuss their results we will need the following definitions. \begin{definition} Given two graphs $H$ and $G$, we call a function $\phi:V(H) \rightarrow V(G)$ a homomorphism if it preserves the edges of $H$: \[ab \rightarrow c \in E(H) \implies \phi(a)\phi(b) \rightarrow \phi(c) \in E(G).\] We will write $\phi:H \rightarrow G$ to indicate that $\phi$ is a homomorphism. \end{definition} \begin{definition} Given a family $\mathcal{F}$ of graphs, we say that a graph $G$ is \emph{$\mathcal{F}$-free} if no injective homomorphism $\phi:F \rightarrow G$ exists for any $F \in \mathcal{F}$. If $\mathcal{F} = \{F\}$ we will write that $G$ is $F$-free. \end{definition} \begin{definition} Given a family $\mathcal{F}$ of graphs, let the \emph{$n$th extremal number} $\text{ex}(n,\mathcal{F})$ denote the maximum number of edges that any $\mathcal{F}$-free graph on $n$ vertices can have. Similarly, let the \emph{$n$th oriented extremal number} $\text{ex}_o(n,\mathcal{F})$ be the maximum number of edges that any $\mathcal{F}$-free oriented graph on $n$ vertices can have. Sometimes we will call the extremal number the \emph{standard} extremal number or refer to the problem of determining the extremal number as the \emph{standard version} of the problem to distinguish these concepts from their oriented counterparts. As before, if $\mathcal{F} = \{F\}$, then we will write $\text{ex}(n,F)$ or $\text{ex}_o(n,F)$ for simplicity. \end{definition} These are often called Tur\'{a}n-type extremal problems after Paul Tur\'{a}n due to his important early results and conjectures concerning forbidden complete $r$-graphs \cite{turan1941, turan1954, turan1961}. Tur\'{a}n problems for uniform hypergraphs make up a large and well-known area of research in combinatorics, and the questions are often surprisingly difficult. Extremal problems like this have also been considered for directed graphs and multigraphs (with bounded multiplicity) in \cite{brown1973} and \cite{brown1969} and for the more general directed multi-hypergraphs in \cite{brown1984}. In \cite{brown1969}, Brown and Harary determined the extremal numbers for several types of specific directed graphs. In \cite{brown1973}, Brown, Erd\H{o}s, and Simonovits determined the general structure of extremal sequences for every forbidden family of digraphs analogous to the Tur\'{a}n graphs for simple graphs. The model of directed hypergraphs studied in \cite{brown1984} have $r$-uniform edges such that the vertices of each edge is given a linear ordering. However, there are many other ways that one could conceivably define a uniform directed hypergraph. The graph theoretic properties of a more general definition of a nonuniform directed hypergraph were studied by Gallo, Longo, Pallottino, and Nguyen in \cite{gallo1993}. There a directed hyperedge was defined to be some subset of vertices with a partition into head vertices and tail vertices. Recently in \cite{cameron2016}, this author tried to capture many of these possible definitions for ``directed hypergraph" into one umbrella class of relational structures called generalized directed hypergraphs. The structures in this class include the uniform and simple versions of undirected hypergraphs, the totally directed hypergraphs studied in \cite{brown1984}, the directed hypergraphs studied in \cite{gallo1993}, and the $2 \rightarrow 1$ model studied here and in \cite{langlois2009,langlois2010}. In \cite{langlois2009, langlois2010}, they study the extremal numbers for two different graphs with two edges each. They refer to these two graphs as the 4-resolvent and the 3-resolvent configurations after their relevance in propositional logic. Here, we will denote these graphs as $R_4$ and $R_3$ respectively and define them formally as \[V(R_4) = \{a,b,c,d,e\} \text{ and } E(R_4) = \{ab \rightarrow c, cd \rightarrow e\}\] and \[V(R_3) = \{a,b,c,d\} \text{ and } E(R_3) = \{ab \rightarrow c, bc \rightarrow d\}.\] In \cite{langlois2010} the authors determined $\text{ex}(n,R_4)$ for sufficiently large $n$, and in \cite{langlois2009} they determined a sequence of numbers asymptotically equivalent to the sequence of numbers $\text{ex}(n,R_3)$ as $n$ increases to infinity. In these papers, the authors discuss a third graph with two edges which they call an Escher configuration because it calls to mind the Escher piece where two hands draw each other. This graph is on four vertices, $\{a,b,c,d\}$ and has edge set $\{ab \rightarrow c,cd \rightarrow b\}$. We will denote it by $E$. These three graphs along with the graph made up of two completely overlapping edges on the same 3-set actually turn out to be the only four nondegenerate graphs with exactly two edges. Their standard and oriented extremal numbers are shown in \cite{cameron2015}. \begin{definition} A graph $H$ is \emph{degenerate} if its vertices can be partitioned into three sets, $V(H) = T_1 \cup T_2 \cup K$ such that every edge of $E(H)$ is of the form $t_1t_2 \rightarrow k$ for some $t_1 \in T_1$, $t_2 \in T_2$, and $k \in K$. \end{definition} An immediate consequence of a result shown in \cite{cameron2016} is that the extremal numbers for a graph $H$ are cubic in $n$ if and only if $H$ is not degenerate. In our model of directed hypergraph, there are nine different graphs with exactly two edges. Of these, five are degenerate. One of these is the graph with two independent edges, $V=\{a,b,c,d,e,f\}$ and $E=\{ab \rightarrow c, de \rightarrow f\}$. In this case the extremal numbers come directly from the known extremal number for two independent edges for undirected $3$-graphs. Therefore, the oriented extremal number is ${n-1 \choose 2}$ and the standard extremal number is $3{n-1 \choose 2}$. We will call the other four degenerate graphs with two edges $I_0$, $I_1$, $H_1$, and $H_2$ and define them as \begin{itemize} \item $V(I_0) = \{a,b,c,d,x\} \text{ and } E(I_0) = \{ab \rightarrow x, cd \rightarrow x\}$ \item $V(I_1) = \{a,b,c,d\} \text{ and } E(I_1) = \{ab \rightarrow c, ad \rightarrow c\}$ \item $V(H_1) = \{a,b,c,d,x\} \text{ and } E(H_1) = \{ax \rightarrow b, cx \rightarrow d\}$ \item $V(H_2) = \{a,b,c,d\} \text{ and } E(H_2) = \{ab \rightarrow c, ab \rightarrow d\}$ \end{itemize} Here, the subscripts indicate the number of tail vertices common to both edges. The $I$ graphs also share a head vertex while the $H$ graphs do not. Some of the proofs that follow rely heavily on the concept of a link graph. For undirected $r$-graphs, the link graph of a vertex is the $(r-1)$-graph induced on the remaining vertices such that each $(r-1)$-set is an $(r-1)$-edge if and only if that set together with the specified vertex makes an $r$-edge in the original $r$-graph \cite{keevash2011}. In the directed hypergraph model here, there are a few ways we could define the link graph of a vertex. We will need the following two. \begin{definition} Let $x \in V(H)$ for some graph $H$. The \emph{tail link graph} of $x$ $T_x$ is the simple undirected 2-graph on the other $n-1$ vertices of $V(H)$ with edge set defined by all pairs of vertices that exist as tails pointing to $x$ in some edge of $H$. That is, $V(T_x) = V(H) \setminus \{x\}$ and \[ E(T_x) = \{yz : yz \rightarrow x \in H\}.\] The size of this set, $|T_x|$ will be called the \emph{tail degree} of $x$. The degree of a particular vertex $y$ in the tail link graph of $x$ will be denoted $d_x(y)$. Similarly, let $D_x$ be the \emph{directed link graph} of $x$ on the remaining $n-1$ vertices of $V(H)$. That is, let $V(D_x) = V(H) \setminus \{x\}$ and \[E(D_x) = \{y \rightarrow z : xy \rightarrow z \in E(H)\}.\] \end{definition} The following notation will also be used when we want to count edges by tail sets. \begin{definition} For any pair of vertices $x,y \in V(H)$ for some graph $H$ let $t(x,y)$ denote the number of edges with tail set $\{x,y\}$. That is \[t(x,y) = |\{v : xy \rightarrow v \in E(H)\}|.\] \end{definition} \section{Forbidden $I_0$} In this section $I_0$ denotes the forbidden graph where two edges intersect in exactly one vertex such that this vertex is a head for both edges. That is $V(I_0) = \{a,b,c,d,x\}$ and $E(I_0) = \{ab \rightarrow x, cd \rightarrow x\}$ (see Figure~\ref{A}). In this section we will prove the following result on the oriented extremal numbers of $I_0$. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (4,0) circle (1pt); \filldraw[black] (2,1) circle (1pt); \filldraw[black] (0,2) circle (1pt); \filldraw[black] (4,2) circle (1pt); \draw[thick] (0,0) -- (0,2); \draw[thick] (4,0) -- (4,2); \draw[thick,->] (0,1) -- (2,1); \draw[thick,->] (4,1) -- (2,1); \node[left] at (0,2) {$a$}; \node[left] at (0,0) {$b$}; \node[above] at (2,1) {$x$}; \node[right] at (4,2) {$c$}; \node[right] at (4,0) {$d$}; \end{tikzpicture} \caption{$I_0$} \label{A} \end{figure} \begin{theorem} \label{exA} For all $n \geq 9$, \[\text{ex}_o(n,I_0) = \begin{cases} n(n-3) + \frac{n}{3} & n \equiv 0 \text{ mod 3}\\ n(n-3) + \frac{n-4}{3} & n \equiv 1 \text{ mod 3} \\ n(n-3) + \frac{n-5}{3} & n \equiv 2 \text{ mod 3} \end{cases} \] with exactly one extremal example up to isomorphism when $3|n$, exactly 18 non-isomorphic extremal constructions when $n \equiv 1 \text{ mod 3}$, and exactly 32 constructions when $n \equiv 2 \text{ mod 3}$. \end{theorem} The proof for this is rather long. However, in the standard version of the problem where each triple of vertices may hold up to three different directed edges, the problem is much simpler so we will begin there. \begin{theorem} For each $n \geq 5$, \[\text{ex}(n,I_0)=n(n-2)\] and for each $n \geq 6$, there are exactly $(n-1)^n$ different labeled $I_0$-free graphs that attain this maximum number of edges. \end{theorem} \begin{proof} Let $H$ be $I_0$-free on $n \geq 5$ vertices. For any $x \in V(H)$, the tail link graph $T_x$ cannot contain two independent edges (see Figure~\ref{Aiff}). Therefore, the edge structure of $T_x$ is either a triangle or a star with $k$ edges all intersecting in a common vertex for some $0 \leq k \leq n-2$. So each vertex $x \in V(H)$ is at the head of at most $n-2$ edges. Hence, \[|E(H)| = \sum_{x \in V(H)} |E(T_x)| \leq n(n-2).\] \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (-4,1) circle [radius=2]; \node at (-6,3) {$T_x$}; \node [below] at (-5,0) {$b$}; \node [below] at (-3,0) {$d$}; \node [above] at (-5,2) {$a$}; \node [above] at (-3,2) {$c$}; \filldraw[black] (-5,0) circle (1pt); \filldraw[black] (-3,0) circle (1pt); \filldraw[black] (-5,2) circle (1pt); \filldraw[black] (-3,2) circle (1pt); \draw[thick] (-5,0) -- (-5,2); \draw[thick] (-3,0) -- (-3,2); \node at (-1,1) {$\iff$}; \node at (0,3) {$H$}; \node [below] at (0,0) {$b$}; \node [below] at (3,0) {$d$}; \node [above] at (0,2) {$a$}; \node [above] at (3,2) {$c$}; \node [above] at (1.5,1) {$x$}; \filldraw[black] (0,0) circle (1pt); \filldraw[black] (3,0) circle (1pt); \filldraw[black] (1.5,1) circle (1pt); \filldraw[black] (0,2) circle (1pt); \filldraw[black] (3,2) circle (1pt); \draw[thick] (0,0) -- (0,2); \draw[thick] (3,0) -- (3,2); \draw[thick,->] (0,1) -- (1.5,1); \draw[thick,->] (3,1) -- (1.5,1); \end{tikzpicture} \caption{$ab, cd \in E(T_x)$ if and only if $ab \rightarrow x, cd \rightarrow x \in H$} \label{Aiff} \end{figure} On the other hand, many different extremal constructions exist that give $n(n-2)$ edges on $n$ vertices without the forbidden intersection. Let \[f:[n] \rightarrow [n]\]be any function such that $f(x) \neq x$ for any $x \in [n]$. Define $H_f$ as the graph with vertex set $V(H_f) = [n]$ and edge set \[E(H_f) = \bigcup_{x \in [n]} \left\{f(x)y \rightarrow x : y \in [n] \setminus \{x,f(x)\}\right\}.\]Certainly each vertex $x$ is at the head of $n-2$ edges and each of its tails contain $f(x)$ which prevents the forbidden subgraph. So $|E(H_f)| = n(n-2)$, and $H_f$ is $I_0$-free for any such function $f$. Moreover, there are $(n-1)^n$ different functions $f$ that will make such a construction on $[n]$. So this gives us $(n-1)^n$ labeledextremal $I_0$-free graphs. Conversely, since any $I_0$-free graph with the maximum number of edges must have $n-2$ edges in $T_x$ for each vertex $x$, then when $n \geq 6$ this implies that all tail link graphs must be $(n-2)$-stars. Therefore, these constructions give all of the extremal examples. \end{proof} The oriented version of this problem is less straight forward, but determining $\text{ex}_o(n,I_0)$ also begins with the observation that every tail link graph of an $I_0$-free graph will either be a triangle, a star, or empty. Broadly speaking, as $n$ gets large, it would make more sense for most, if not all, tail link graphs to be stars in order to fit as many edges into an $I_0$-free graph. This motivates the following auxiliary structure. \subsection{Gates} Let $H$ be some $I_0$-free graph. For each $x \in V(H)$ for which $T_x$ is a star (with at least one edge), let $g(x)$ denote the common vertex for the edges of $T_x$. We will refer to this vertex as the \emph{gatekeeper} of $x$ (in that it is the gatekeeper that any other vertex must pair with in order to ``access" $x$). In the case where $T_x$ contains only a single edge we may choose either of its vertices to serve as the gatekeeper. In this way, we have constructed a partial function, $g: V(H) \nrightarrow V(H)$. Next, construct a directed $2$-graph $G$ on the vertex set $V(H)$ based on this partial function: \[y \rightarrow x \in E(G) \iff y = g(x).\] We'll call this digraph the \emph{gate} of $H$ (or more properly, $G$ is the gate of $H$ \emph{under} $g$ since $g$ isn't necessarily unique). \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (0,0) circle [radius=2]; \node at (0,0) {$C_k$}; \filldraw[black] (xyz polar cs:angle=30, radius=3) circle (1pt); \filldraw[black] (xyz polar cs:angle=40, radius=3) circle (1pt); \filldraw[black] (xyz polar cs:angle=25, radius=4) circle (1pt); \draw[thick,->] (xyz polar cs:angle=35, radius=2) -- (xyz polar cs:angle=30, radius=3); \draw[thick,->] (xyz polar cs:angle=35, radius=2) -- (xyz polar cs:angle=40, radius=3); \draw[thick,->] (xyz polar cs:angle=30, radius=3) -- (xyz polar cs:angle=25, radius=4); \filldraw[black] (xyz polar cs:angle=340, radius=3) circle (1pt); \filldraw[black] (xyz polar cs:angle=345, radius=4) circle (1pt); \draw[thick,->] (xyz polar cs:angle=340, radius=2) -- (xyz polar cs:angle=340, radius=3); \draw[thick,->] (xyz polar cs:angle=340, radius=3) -- (xyz polar cs:angle=345, radius=4); \filldraw[black] (xyz polar cs:angle=190, radius=3) circle (1pt); \filldraw[black] (xyz polar cs:angle=195, radius=4) circle (1pt); \filldraw[black] (xyz polar cs:angle=185, radius=4) circle (1pt); \filldraw[black] (xyz polar cs:angle=180, radius=5) circle (1pt); \filldraw[black] (xyz polar cs:angle=190, radius=5) circle (1pt); \draw[thick,->] (xyz polar cs:angle=190, radius=2) -- (xyz polar cs:angle=190, radius=3); \draw[thick,->] (xyz polar cs:angle=190, radius=3) -- (xyz polar cs:angle=195, radius=4); \draw[thick,->] (xyz polar cs:angle=190, radius=3) -- (xyz polar cs:angle=185, radius=4); \draw[thick,->] (xyz polar cs:angle=185, radius=4) -- (xyz polar cs:angle=190, radius=5); \draw[thick,->] (xyz polar cs:angle=185, radius=4) -- (xyz polar cs:angle=180, radius=5); \filldraw[black] (xyz polar cs:angle=95, radius=3) circle (1pt); \filldraw[black] (xyz polar cs:angle=105, radius=3) circle (1pt); \filldraw[black] (xyz polar cs:angle=115, radius=3) circle (1pt); \draw[thick,->] (xyz polar cs:angle=105, radius=2) -- (xyz polar cs:angle=115, radius=3); \draw[thick,->] (xyz polar cs:angle=105, radius=2) -- (xyz polar cs:angle=105, radius=3); \draw[thick,->] (xyz polar cs:angle=105, radius=2) -- (xyz polar cs:angle=95, radius=3); \end{tikzpicture} \caption{The structure of a connected component of the gate $G$} \label{Ckplus} \end{figure} The edge structure of any gate $G$ is not difficult to determine. Since $g$ is a partial function, then each vertex has in-degree at most one in $G$. Therefore, the structure of any connected component of $G$ can be described as a directed cycle on $k$ vertices, $C_k$, for $1 \leq k$ (where $k=1$ implies a single vertex) unioned with $k$ disjoint directed trees, each with its root vertex on this cycle (see Figure~\ref{Ckplus}). We will refer to this kind of general structure as a \emph{$k$-cycle with branches}. Let \[\mathcal{C} = \bigcup_{k=1}^{n} \mathcal{C}_k\] be the set of maximal connected components of a gate of $H$ where, for each $k$, $\mathcal{C}_k$ is the set of maximal connected components that are $k$-cycles with branches. Note that \[|E(H)| = \sum_{x \in V(H)} |T_x| = \sum_{C \in \mathcal{C}} \left( \sum_{x \in V(C)} |T_x| \right) = \sum_{k=1}^n \left(\sum_{C \in \mathcal{C}_k} \left( \sum_{x \in V(C)} |T_x| \right)\right).\] The next section determines for each $k$ an upper bound on \[\sum_{x \in V(C)} |T_x|\] as a function of the number of vertices, $|V(C)|$, for any $C \in \mathcal{C}_k$. \subsection{Bounding $\sum_{x \in V(C)} |T_x|$ for any connected component $C$ of the gate} Loosely speaking, each gatekeeper edge of a connected component $C$ represents at most $n-2$ edges of $H$. We will arrive at an upper bound on the sum $\sum_{x \in V(C)} |T_x|$ by adding this maximum for each edge of $C$, and then subtracting the number of triples of vertices that such a count has included more than once. This will happen for any triple of vertices which contain two or three gatekeeper edges. We make this observation formal in the following definition and lemma. \begin{definition} Let $G$ be some gate and let $C$ be a maximal connected component of $G$. Let $P(C)$ be the set of $2 \rightarrow 1$ \emph{possible edges} defined by \[P(C) = \bigcup_{a \rightarrow b \in E(C)} \left\{av \rightarrow b : v \in V(H) \setminus \{a,b\} \right\}.\] \end{definition} \begin{lemma} \label{goofy} Let $G$ be a gate, and let $C$ be a maximal connected component of $G$. If a set of three distinct vertices $\{x,y,z\} \subseteq V(C)$ are spanned by two gatekeeper edges of $G$, then $P(C)$ contains at least two edges on these three vertices. \end{lemma} \begin{proof} Without loss of generality, the two spanning edges on $\{x,y,z\}$ are either of the form \[x \rightarrow y \rightarrow z \text{ or } x \leftarrow y \rightarrow z.\] In the former case, $P(C)$ contains the edges $xz \rightarrow y$ and $yx \rightarrow z$. In the latter case, $P(C)$ contains the edges $yz \rightarrow x$ and $yx \rightarrow z$. \end{proof} Now comes the main counting lemma. \begin{lemma} \label{components} Let $H$ be an $I_0$-free graph on $n \geq 8$ vertices. Let $G$ be a gate of $H$. Let $C$ be a maximal connected component of $G$ with $m$ vertices. Then \begin{itemize} \item $\sum_{x \in V(C)} |T_x| \leq m(n-3)$ if $C \in \mathcal{C}_k$ for any $k \neq 3$ with equality possible only if $C = C_k$ for some $k \geq 4$, \item $\sum_{x \in V(C)} |T_x| \leq m(n-3) + 1$ if $C = C_3$, and \item $\sum_{x \in V(C)} |T_x| \leq m(n-3)$ for all other $C \in \mathcal{C}_3$ with equality possible only if $C$ is a $3$-cycle with exactly one nonempty directed path coming off of it. \end{itemize} \end{lemma} \begin{proof} For convenience let \[S = \sum_{x \in V(C)} |T_x|.\] Note that for each $x \in V(C)$ with in-degree one, $ab \in T_x$ implies that $ab \rightarrow x \in P(C)$. Hence, if $C \not \in \mathcal{C}_1$, then every edge counted in the sum $S$ is in $P(C)$. Moreover, $|P(C)| = m(n-2)$. If $C \in \mathcal{C}_k$ for $k \geq 4$, then by Lemma~\ref{goofy}, each intersection of gatekeeper edges of $C$ yields two edges on the same triple of vertices in $P(C)$. Conversely, since $C$ contains no $C_3$, then each distinct triple of vertices contains at most two gatekeeper edges. Therefore, each triple contains at most two edges of $P(C)$. Hence, \[S \leq m(n-2) - \sum_{x \in V(C)} {d_G(x) \choose 2}\] where $d_G(x)$ denotes the total number of vertices incident to $x$ in the gate. Since $C$ has $m$ edges, then $\sum_{x \in V(C)} d_G(x) = 2m$. So \[S \leq m(n-2) - \sum_{x \in V(C)} {d_G(x) \choose 2} \leq m(n-3)\] by Jensen's Inequality. Moreover, equality happens if and only if $d_G(x) = d_G(y)$ for all $x,y \in V(C)$. Therefore, this inequality is strict for all $C \in \mathcal{C}_k$ unless $C=C_k$. Similarly, if $C \in \mathcal{C}_2$, then $P(C)$ contains at least $\sum_{x \in V(C)} {d_G(x) \choose 2}$ multiedges for the same reason as before. But here there are an additional $n-2$ edges counted for each triple containing the $C_2$. Also, \[\sum_{x \in V(C)} d_G(x) = 2(m-1).\] Hence, \[S \leq m(n-2) - (n-2) - \sum_{x \in V(C)} {d_G(x) \choose 2} \leq (m-1)(n-2) - m {\frac{2(m-1)}{m} \choose 2}\] by Jensen's Inequality. This is strictly less than $m(n-3)$. In the acyclic case, Lemma~\ref{goofy} implies that the sum of all $|T_x|$ for each $x \in V(C)$ other than the root vertex is less than or equal to \[(m-1)(n-2) - \sum_{x \in V(C)} {d_G(x) \choose 2}.\] The root vertex itself is the head vertex at most 3 edges in $H$ so Jensen's Inequality gives \[S \leq (m-1)(n-2) - m {\frac{2(m-1)}{m} \choose 2} + 3 < m(n-3)\] for all $n \geq 8$. Finally, if $C \in \mathcal{C}_3$, then each intersection of gatekeeper edges of $C$ yields two edges on the same triple of vertices in $P(C)$. However, exactly one triple of vertices contains three gatekeeper edges and has three edges in $P(C)$. But the rest have at most two since there is only one triangle in $C$. Therefore, $\sum_{x \in V(C)} {d_G(x) \choose 2}$ counts each triple of vertices that contain more than one gatekeeper edge exactly once except for the triple that makes up the $C_3$ which it counts three times. Since we must subtract off 2 edges in $P(C)$ on these three vertices to eliminate repeated triples, then we must subtract $\sum_{x \in V(C)} {d_G(x) \choose 2}-1$ from $|P(C)|$. Therefore, \[S \leq m(n-2) - \sum_{x \in V(C)} {d_G(x) \choose 2} + 1.\] So by Jensen's Inequality, \[S \leq m(n-3) + 1\] with equality possible only if all of the degrees $d_G(x)$ are equal. This can only happen if $C = C_3$. If we want to see for which $C \in \mathcal{C}_3$ the second best bound of $m(n-3)$ could be attained, then we need to set \[\sum_{x \in V(C)} {d_G(x) \choose 2} = m+1.\] Assume that the vertices are $x_1,\ldots,x_m$, and for each $x_i$ let \[d_i = d_G(x_i)-2.\] Then $\sum_{i=1}^m d_i = 0$ and a quick calculation shows that $\sum_{i=1}^m d_i^2 = 2$. Therefore, the only possibility is for some $d_i=1$ and another to equal $-1$ and all the rest must be 0. This corresponds with one vertex degree being 3, another being 1, and all others being 2. The only way that this can happen in a $C_3$ with branches is to have exactly one branch, and that branch must be a directed path. \end{proof} This shows that the best we can hope for in terms of the average number of edges per vertex over any connected component of the gate is $n-3 + \frac{1}{3}$, and this could be attained only in the case where the component is a directed triangle with no branches. Otherwise, the average number of edges of a component is at most $n-3$, and this is attainable only if the component is a directed triangle with a single directed path coming off of one of its vertices or a directed $k$-cycle with no branches for some $k \geq 4$. This is enough for us to establish the upper bound for $\text{ex}_o(n,I_0)$ and to characterize the necessary structure of the gate for any graph attaining this upper bound. \subsection{Upper Bound on $\text{ex}_o(n,I_0)$} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (1,1) circle (1pt); \filldraw[black] (2,0) circle (1pt); \draw[thick,->] (0,0) -- (1,1); \draw[thick,->] (1,1) -- (2,0); \draw[thick,->] (2,0) -- (0,0); \filldraw[black] (3,0) circle (1pt); \filldraw[black] (4,1) circle (1pt); \filldraw[black] (5,0) circle (1pt); \draw[thick,->] (3,0) -- (4,1); \draw[thick,->] (4,1) -- (5,0); \draw[thick,->] (5,0) -- (3,0); \filldraw[black] (7,0) circle (1pt); \filldraw[black] (8,1) circle (1pt); \filldraw[black] (9,0) circle (1pt); \draw[thick,->] (7,0) -- (8,1); \draw[thick,->] (8,1) -- (9,0); \draw[thick,->] (9,0) -- (7,0); \node at (6,0.5) {$\cdots$}; \end{tikzpicture} \caption{Structure of the gate for an extremal $I_0$-free graph when $n \equiv 0 \text{ mod } 3$.} \end{figure} Let $H$ be an $I_0$-free graph on $n \geq 9$ vertices. Let $G$ be a gate of $H$. Let $\mathcal{C}$ be the set of maximal connected components of $G$ and break $\mathcal{C}$ into three disjoint subsets based on the maximum average number of edges attainable for the components in each. That is, let \[\mathcal{C} = \mathcal{D}_1 \cup \mathcal{D}_2 \cup \mathcal{D}_3\] where $\mathcal{D}_1$ are all components with maximum average number of edges per vertex strictly less than $n-3$: those components that are either acyclic, contain a $C_2$, contain a $C_3$ with nonempty branches that are more than just a single path, or contain a $C_k$ for $k \geq 4$ with some nonempty branch; $\mathcal{D}_2$ is the set of all components with maximum number of edges per vertex of $n-3$: those that contain a directed $C_3$ and exactly one directed path or those that are a directed $k$-cycles for any $k \geq 4$ and no branches; and $\mathcal{D}_3$ is the set of components with a maximum average greater than $n-3$: the directed triangles. For each $i$ let $d_i$ be the total number of vertices contained in the components of $\mathcal{D}_i$. Then \[|E(H)| \leq d_3 \left(n-3 + \frac{1}{3} \right) + (n-d_3)(n-3)\] with equality possible only if $d_1=0$. Then this is enough to prove the following. \begin{lemma} \label{mod0} Let $H$ be an $I_0$-free graph on $n \geq 9$ vertices such that $n \equiv 0 \text{ mod } 3$, then \[|E(H)| \leq n(n-3) + \frac{n}{3}.\] Moreover, the only way for $H$ to attain this maximum number of edges is if the gate of $H$ is a disjoint union of directed triangles. \end{lemma} The next two lemmas give the maximum number when $n \equiv 1,2 \text{ mod } 3$. There is only slightly more to consider in these cases. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (1,1) circle (1pt); \filldraw[black] (2,0) circle (1pt); \draw[thick,->] (0,0) -- (1,1); \draw[thick,->] (1,1) -- (2,0); \draw[thick,->] (2,0) -- (0,0); \filldraw[black] (3,0) circle (1pt); \filldraw[black] (4,1) circle (1pt); \filldraw[black] (5,0) circle (1pt); \draw[thick,->] (3,0) -- (4,1); \draw[thick,->] (4,1) -- (5,0); \draw[thick,->] (5,0) -- (3,0); \filldraw[black] (7,0) circle (1pt); \filldraw[black] (8,1) circle (1pt); \filldraw[black] (9,0) circle (1pt); \draw[thick,->] (7,0) -- (8,1); \draw[thick,->] (8,1) -- (9,0); \draw[thick,->] (9,0) -- (7,0); \filldraw[black] (10,0) circle (1pt); \filldraw[black] (11,1) circle (1pt); \filldraw[black] (12,0) circle (1pt); \filldraw[black] (11.5,1.8) circle (1pt); \draw[thick,->] (10,0) -- (11,1); \draw[thick,->] (11,1) -- (12,0); \draw[thick,->] (12,0) -- (10,0); \draw[thick,->] (11,1) -- (11.5,1.8); \node at (6,0.5) {$\cdots$}; \filldraw[black] (0,-3) circle (1pt); \filldraw[black] (1,-2) circle (1pt); \filldraw[black] (2,-3) circle (1pt); \draw[thick,->] (0,-3) -- (1,-2); \draw[thick,->] (1,-2) -- (2,-3); \draw[thick,->] (2,-3) -- (0,-3); \filldraw[black] (3,-3) circle (1pt); \filldraw[black] (4,-2) circle (1pt); \filldraw[black] (5,-3) circle (1pt); \draw[thick,->] (3,-3) -- (4,-2); \draw[thick,->] (4,-2) -- (5,-3); \draw[thick,->] (5,-3) -- (3,-3); \filldraw[black] (7,-3) circle (1pt); \filldraw[black] (8,-2) circle (1pt); \filldraw[black] (9,-3) circle (1pt); \draw[thick,->] (7,-3) -- (8,-2); \draw[thick,->] (8,-2) -- (9,-3); \draw[thick,->] (9,-3) -- (7,-3); \filldraw[black] (10,-3) circle (1pt); \filldraw[black] (10,-2) circle (1pt); \filldraw[black] (12,-2) circle (1pt); \filldraw[black] (12,-3) circle (1pt); \draw[thick,->] (10,-3) -- (10,-2); \draw[thick,->] (10,-2) -- (12,-2); \draw[thick,->] (12,-2) -- (12,-3); \draw[thick,->] (12,-3) -- (10,-3); \node at (6,-1.2) {or}; \node at (6,-2.5) {$\cdots$}; \end{tikzpicture} \caption{The only possible structures of the gate of an extremal $I_0$-free graph when $n \equiv 1 \text{ mod } 3$.} \end{figure} \begin{lemma} \label{mod1} Let $H$ be an $I_0$-free graph on $n \geq 9$ vertices such that $n \equiv 1 \text{ mod } 3$, then \[|E(H)| \leq n(n-3) + \frac{n-4}{3}.\] Moreover, the only way for $H$ to attain this maximum number of edges is if the gate of $H$ is a disjoint union of $\frac{n-4}{3}$ directed triangles together with either a directed $C_4$ or a 3-cycle with an extra edge. \end{lemma} \begin{proof} Since $n \equiv 1 \text{ mod } 3$, then $d_3 \leq n-1$. If $d_3 = n-1$, then the gate consists of $\frac{n-1}{3}$ disjoint directed triangles and one isolated vertex which means that \[|E(H)| \leq (n-1) \left(n-3+\frac{1}{3} \right) + 3.\] If $d_3 \leq n-4$, then we can do better with \[|E(H)| \leq (n-4) \left( n-3+\frac{1}{3} \right) + 4(n-3)\] only in the case of $\frac{n-4}{3}$ disjoint directed triangles and one component from $\mathcal{D}_2$ in the gate. Therefore, \[|E(H)| \leq n(n-3) + \frac{n-4}{3}.\] \end{proof} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (1,1) circle (1pt); \filldraw[black] (2,0) circle (1pt); \draw[thick,->] (0,0) -- (1,1); \draw[thick,->] (1,1) -- (2,0); \draw[thick,->] (2,0) -- (0,0); \filldraw[black] (3,0) circle (1pt); \filldraw[black] (4,1) circle (1pt); \filldraw[black] (5,0) circle (1pt); \draw[thick,->] (3,0) -- (4,1); \draw[thick,->] (4,1) -- (5,0); \draw[thick,->] (5,0) -- (3,0); \filldraw[black] (7,0) circle (1pt); \filldraw[black] (8,1) circle (1pt); \filldraw[black] (9,0) circle (1pt); \draw[thick,->] (7,0) -- (8,1); \draw[thick,->] (8,1) -- (9,0); \draw[thick,->] (9,0) -- (7,0); \filldraw[black] (10,0) circle (1pt); \filldraw[black] (11,1) circle (1pt); \filldraw[black] (12,0) circle (1pt); \filldraw[black] (11.5,1.8) circle (1pt); \filldraw[black] (12,2.6) circle (1pt); \draw[thick,->] (10,0) -- (11,1); \draw[thick,->] (11,1) -- (12,0); \draw[thick,->] (12,0) -- (10,0); \draw[thick,->] (11,1) -- (11.5,1.8); \draw[thick,->] (11.5,1.8) -- (12,2.6); \node at (6,0.5) {$\cdots$}; \filldraw[black] (0,-3) circle (1pt); \filldraw[black] (1,-2) circle (1pt); \filldraw[black] (2,-3) circle (1pt); \draw[thick,->] (0,-3) -- (1,-2); \draw[thick,->] (1,-2) -- (2,-3); \draw[thick,->] (2,-3) -- (0,-3); \filldraw[black] (3,-3) circle (1pt); \filldraw[black] (4,-2) circle (1pt); \filldraw[black] (5,-3) circle (1pt); \draw[thick,->] (3,-3) -- (4,-2); \draw[thick,->] (4,-2) -- (5,-3); \draw[thick,->] (5,-3) -- (3,-3); \filldraw[black] (7,-3) circle (1pt); \filldraw[black] (8,-2) circle (1pt); \filldraw[black] (9,-3) circle (1pt); \draw[thick,->] (7,-3) -- (8,-2); \draw[thick,->] (8,-2) -- (9,-3); \draw[thick,->] (9,-3) -- (7,-3); \filldraw[black] (10,-3) circle (1pt); \filldraw[black] (10,-2) circle (1pt); \filldraw[black] (12,-2) circle (1pt); \filldraw[black] (12,-3) circle (1pt); \filldraw[black] (11,-1.5) circle (1pt); \draw[thick,->] (10,-3) -- (10,-2); \draw[thick,->] (10,-2) -- (11,-1.5); \draw[thick,->] (11,-1.5) -- (12,-2); \draw[thick,->] (12,-2) -- (12,-3); \draw[thick,->] (12,-3) -- (10,-3); \node at (6,-2.5) {$\cdots$}; \node at (6,-1.2) {or}; \end{tikzpicture} \caption{The only possible structures of the gate of an extremal $I_0$-free graph when $n \equiv 2 \text{ mod } 3$.} \label{poss2} \end{figure} \begin{lemma} \label{mod2} Let $H$ be an $I_0$-free graph on $n \geq 11$ vertices such that $n \equiv 2 \text{ mod } 3$, then \[|E(H)| \leq n(n-3) + \frac{n-5}{3}.\] Moreover, the only way for $H$ to attain this maximum number of edges is if the gate of $H$ is a disjoint union of $\frac{n-5}{3}$ directed triangles together with either a directed $C_5$ or a 3-cycle with a directed path of two edges. \end{lemma} \begin{proof} Since $n \equiv 2 \text{ mod } 3$, then $d_3 \leq n-2$ and equality implies that $G$ consists of $\frac{n-2}{3}$ disjoint directed triangles and two additional vertices that are either both isolated, contain one edge, or are a $C_2$ giving $6$, $3+(n-2)$, or $n-2$ additional edges respectively. The best we can do when $d_3=n-2$ is therefore, \[|E(H)| \leq (n-2)\left( n-3+ \frac{1}{3} \right) + (n+1).\] Otherwise, $d_3 \leq n-5$ and the best we can do is \[|E(H)| \leq (n-5)\left( n-3+ \frac{1}{3} \right) + 5(n-3).\] This is better. Moreover, this will happen only when the five non-triangle vertices are in a component (or components) of $G$ that give an average of $n-3$. So they must either make a $C_5$ or a directed triangle with one path. \end{proof} \subsection{Lower bound constructions} The structure of the gates necessary to attain the maximum number of edges for a $I_0$-free graph determined in the previous section are also sufficient. Of these gates, none of them have acyclic components. Therefore, any graph that produces one of these gates has only vertices with stars for tail link graphs. This immediately implies that there is no $I_0$ in any graph that has such a gate. Moreover, if $H$ is a graph with a gate $G$ that is one of these configurations, then \[E(H) \subseteq \bigcup_{C \in \mathcal{C}} P(C)\] where $\mathcal{C}$ is the set of maximal connected components of $G$. All that is left to do in order to construct an extremal example is to pick which edges of each $P(C)$ to delete in order to eliminate triples of vertices with more than one edge. \begin{lemma} Let $H$ be an $I_0$-free graph on $n \geq 9$ vertices such that $n \equiv 0 \text{ mod } 3$, then \[|E(H)| \geq n(n-3) + \frac{n}{3}\] and there is exactly one extremal construction up to isomorphism. \end{lemma} \begin{proof} We know from Lemma~\ref{mod0} that the only way $H$ can possibly attain $n(n-3) + \frac{n}{3}$ edges is if its gate is the disjoint union of $\frac{n}{3}$ directed triangles. Therefore, each $P(C_3)$ contains exactly one vertex triple with all three possible edges. So two of these must be deleted for each component in order to arrive at an extremal construction. The three choices for this deletion on each component are all isomorphic to each other. Therefore, there is exactly one extremal construction up to isomorphism. \end{proof} \begin{lemma} Let $H$ be an $I_0$-free graph on $n \geq 9$ vertices such that $n \equiv 1 \text{ mod } 3$, then \[|E(H)| \geq n(n-3) + \frac{n-4}{3}\] and there are exactly 18 extremal constructions up to isomorphism. \end{lemma} \begin{proof} We know from Lemma~\ref{mod1} that if $H$ has $n(n-3) + \frac{n-4}{3}$ edges, then its gate is the disjoint union of $\frac{n-4}{3}$ directed triangles with either a directed $C_4$ or a $C_3$ plus an edge on the remaining 4 vertices. As in the previous proof, there is only one choice up to isomorphism for which edges to delete from each $P(C_3)$. However, this will not be true of the last component on the remaining four vertices. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (10,0) circle (1pt); \filldraw[black] (11,1) circle (1pt); \filldraw[black] (12,0) circle (1pt); \filldraw[black] (11.5,1.8) circle (1pt); \node [left] at (11,1) {$x$}; \node [left] at (11.5,1.8) {$a$}; \node [left] at (10,0) {$z$}; \node [right] at (12,0) {$y$}; \draw[thick,->] (10,0) -- (11,1); \draw[thick,->] (11,1) -- (12,0); \draw[thick,->] (12,0) -- (10,0); \draw[thick,->] (11,1) -- (11.5,1.8); \end{tikzpicture} \caption{$C_3$ plus an edge} \end{figure} First, let's consider the case where the last component is a $C_3$ plus one edge. Call the vertices $\{x,y,z,a\}$ where $x \rightarrow y \rightarrow z \rightarrow x$ is the $C_3$ and $x \rightarrow a$ is the additional edge. First, note that we have the following three mutually exclusive choices for edges with head vertices in this component: \begin{enumerate} \item $xa \in T_y$ or $xy \in T_a$, \item $za \in T_x$ or $xz \in T_a$, and \item $zx \in T_y$, $yz \in T_x$, or $xy \in T_z$. \end{enumerate} This gives 12 choices, and each choice is unique up to isomorphism. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (10,-3) circle (1pt); \filldraw[black] (10,-2) circle (1pt); \filldraw[black] (12,-2) circle (1pt); \filldraw[black] (12,-3) circle (1pt); \draw[thick,->] (10,-3) -- (10,-2); \draw[thick,->] (10,-2) -- (12,-2); \draw[thick,->] (12,-2) -- (12,-3); \draw[thick,->] (12,-3) -- (10,-3); \node [left] at (10, -2.5) {$2$}; \node [right] at (12, -2.5) {$2$}; \end{tikzpicture} \caption{$C_4$ with 2 additional edges in opposite tail link graphs} \end{figure} Next consider the case of $C_4$. Each $3$-subset of these four vertices holds two edges of $P(C)$ - one that points along the direction of the two gatekeeper edges and one that points the middle vertex of the two gatekeeper edges. For each triple one of these edges must be deleted to arrive at a legal oriented construction. Each tail link graph must have at least $n-4$ edges, and combined they must contain four additional edges. Since each can have up to two more edges, then the distribution of these additional edges must be one of the following integer partitions of 4: \begin{itemize} \item 2, 2, 0, 0 \item 2, 1, 1, 0 \item 1, 1, 1, 1 \end{itemize} There is only one choice up to isomorphism with a distribution of 2, 2, 0, 0. Each of the three ways to place 2, 1, 1, 0 around $C_4$ are possible but each distribution has only one way up to isomorphism. Finally, there are two ways up to isomorphism to put an extra edge into each tail link graph. So all together there are six nonisomorphic ways to distribute these extra edges to the $C_4$ tail link graphs. \end{proof} \begin{lemma} Let $H$ be an $I_n$-free graph on $n \geq 9$ vertices such that $n \equiv 2 \text{ mod } 3$, then \[|E(H)| \geq n(n-3) + \frac{n-5}{3}\] and there are exactly 32 extremal constructions up to isomorphism. \end{lemma} \begin{proof} We can do the same kind of analysis when $n = 3k+2$ as in the previous proof. We know from Lemma~\ref{mod2} that the gate of any extremal construction must be all directed triangles together with either a directed $C_5$ or a directed triangle with a directed path of length two coming off of it (see Figure~\ref{poss2}). First, consider the $C_5$ case. Let the vertices be $\{x_0,\ldots, x_4\}$. For each gatekeeper edge, $x_i \rightarrow x_{i+1}$, every edge of the form $x_iv \rightarrow x_{i+1}$ must be an edge in $H$ for any vertex \[v \neq x_i,x_{i+1},x_{i-1},x_{i+2}.\] Each gatekeeper edge can represent up to two additional edges of $H$, but again, every intersection of gatekeeper edges requires a mutually exclusive choice. Ultimately, we can add 5 additional edges so the extra edges must be distributed in one of the following ways: \begin{itemize} \item 2, 2, 1, 0, 0 \item 2, 1, 1, 1, 0 \item 1, 1, 1, 1, 1 \end{itemize} There are 2 ways to get the first distribution up to isomorphism, 4 ways to get the second, and 2 ways to get the third. Therefore, there are 8 extremal constructions with this gate up to isomorphism. Now consider the case of a directed triangle with a directed two path coming off of it. If we label the vertices as $\{x,y,z,a,b\}$ (see Figure~\ref{c3plusplus}), the mutually exclusive choices are \begin{enumerate} \item $ax \rightarrow y$ or $yx \rightarrow a$, \item $az \rightarrow x$ or $zx \rightarrow a$, \item $zx \rightarrow y$, $yz \rightarrow x$, or $xy \rightarrow z$, and \item $xa \rightarrow b$ or $bx \rightarrow a$ \end{enumerate} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (10,0) circle (1pt); \filldraw[black] (11,1) circle (1pt); \filldraw[black] (12,0) circle (1pt); \filldraw[black] (11.5,1.8) circle (1pt); \filldraw[black] (12,2.6) circle (1pt); \node[left] at (10,0) {$z$}; \node[left] at (11,1) {$x$}; \node[right] at (12,0) {$y$}; \node[left] at (11.5,1.8) {$a$}; \node[left] at (12,2.6) {$b$}; \draw[thick,->] (10,0) -- (11,1); \draw[thick,->] (11,1) -- (12,0); \draw[thick,->] (12,0) -- (10,0); \draw[thick,->] (11,1) -- (11.5,1.8); \draw[thick,->] (11.5,1.8) -- (12,2.6); \end{tikzpicture} \caption{$C_3^{+2}$} \label{c3plusplus} \end{figure} This gives 24 ways of reaching the maximum, and each way is unique up to isomorphism. Therefore, there are 32 total distinct extremal graphs up to isomorphism. \end{proof} This establishes the main result of this section. \section{Forbidden $I_1$} In this section $I_1$ denotes the forbidden graph where two edges intersect in exactly two vertices such that one vertex is a head for both edges and the other is a tail for each edge. That is $V(I_1) = \{a,b,c,d\}$ and $E(I_1) = \{ab \rightarrow c, ad \rightarrow c\}$ (see Figure~\ref{F}). \begin{theorem} For all $n \geq 4$, \[\text{ex}(n,I_1)=\text{ex}_o(n,I_1) = n \left\lfloor \frac{n-1}{2} \right\rfloor\]and there are \[\left(\frac{(n-1)!}{2^{\left\lfloor \frac{n-1}{2} \right\rfloor} \left \lfloor \frac{n-1}{2} \right\rfloor !}\right)^n\] labeled graphs that attain this maximum in the standard case. \end{theorem} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,2) circle (1pt); \filldraw[black] (1,0) circle (1pt); \filldraw[black] (-1,0) circle (1pt); \filldraw[black] (0,0.75) circle (1pt); \draw[thick] (0,2) -- (1,0); \draw[thick] (0,2) -- (-1,0); \draw[thick,->] (0.5,1) -- (0,0.75); \draw[thick,->] (-0.5,1) -- (0,0.75); \node[above] at (0,2) {$a$}; \node[below] at (0,0.75) {$c$}; \node[right] at (1,0) {$d$}; \node[left] at (-1,0) {$b$}; \end{tikzpicture} \caption{$I_1$} \label{F} \end{figure} \begin{proof} Let $H$ be an $I_1$-free graph on $n$ vertices. For any $x \in V(H)$, $T_x$ is a simple undirected $2$-graph on $n-1$ vertices such that no two edges are adjacent (this is true for either version of the problem). Therefore, the edges of $T_x$ are a matching on at most $n-1$ vertices. So there are at most $\left\lfloor \frac{n-1}{2} \right\rfloor$ edges in $T_x$ for every $x \in V(H)$. Thus, \[|E(H)| = \sum_{x \in V(H)} |T_x| \leq n \left\lfloor \frac{n-1}{2} \right\rfloor.\] This shows the upper bound for both versions. Now we want to find lower bound constructions. In the standard version of the problem there are many extremal constructions since for each vertex $x$, we may pick any maximum matching on the remaining $n-1$ vertices to serve as the edges of $T_x$. So \[\text{ex}(n,I_1) = n \left\lfloor \frac{n-1}{2} \right\rfloor.\] Moreover, the number of labeled graphs that attain this maximum equals the number of ways to take a maximum matching to construct each tail link graph. For even $k$, the number of matchings on $k$ vertices is \[M_k = (k-1)M_{k-2}\] since if we fix some vertex, then we can pick any of the remaining $k-1$ vertices to go with it and then take the number of matchings on the remaining $n-2$. Since $M_2=1$, then in general for even $k$, \[M_k = \prod_{i=1}^{\frac{k}{2}}(2i-1).\] If $k$ is odd, then we can first select the vertex left out of the matching to get \[M_k = kM_{k-1} = k \cdot \prod_{i=1}^{\frac{k-1}{2}}(2i-1) = \prod_{i=1}^{\frac{k+1}{2}}(2i-1).\] Therefore, the number of labeled extremal $I_1$-free graphs on $n$ vertices is \[\left(\prod_{i=1}^{\left\lfloor\frac{n}{2}\right\rfloor}(2i-1) \right)^n = \left(\frac{(n-1)!}{2^{\left\lfloor \frac{n-1}{2} \right\rfloor} \left \lfloor \frac{n-1}{2} \right\rfloor !}\right)^n.\] \begin{figure} \centering \begin{tikzpicture} [scale=0.75] \node at (-4,4) {$T_i$}; \filldraw[color=black,fill=blue!5] (0,0) circle [radius=4]; \filldraw[black] (-2.5,0) circle (1pt); \node [above] at (-2.5,0) {$i+1$}; \filldraw[black] (-1.3,2) circle (1pt); \node [above] at (-1.3,2) {$i+2$}; \filldraw[black] (0,2) circle (1pt); \node [above] at (0,2) {$i+4$}; \filldraw[black] (1.7,2) circle (1pt); \node [above] at (1.7,2) {$i-2$}; \filldraw[black] (-1.3,-2) circle (1pt); \node [below] at (-1.3,-2) {$i+1$}; \filldraw[black] (0,-2) circle (1pt); \node [below] at (0,-2) {$i+3$}; \filldraw[black] (1.7,-2) circle (1pt); \node [below] at (1.7,-2) {$i-1$}; \draw[thick] (-1.3,2) -- (-1.3,-2); \draw[thick] (0,2) -- (0,-2); \draw[thick] (1.7,2) -- (1.7,-2); \node at (1,0) {$\cdots$}; \filldraw[black] (5.5,0) circle (1pt); \node [above right] at (5.5,0) {$i$}; \draw[thick,->] (4,0) -- (5.5,0); \end{tikzpicture} \caption{$T_i$ in the oriented extremal construction for even $n$} \end{figure} In the oriented version of the problem we need to be more careful with the construction. First, assume that $n$ is even and define a graph $H$ with vertex set $V(H) = \mathbb{Z}_n$ and edge set \[E(H) = \bigcup_{i=0}^{n-1} \left\{(i+2k)(i+2k+1) \rightarrow i : k=1,2,\ldots,\frac{n-2}{2} \right\}.\]This construction creates a maximum matching for each tail link graph (with $i+1$ as the odd vertex out for each $T_i$). So $H$ has the extremal number of edges and contains no $I_1$. Therefore, all we need to show is that it has no triple with more than edge. \begin{figure} \centering \begin{tikzpicture} [scale=0.75] \node at (-4,4) {$T_i$}; \filldraw[color=black,fill=blue!5] (0,0) circle [radius=4]; \filldraw[black] (-2.5,1) circle (1pt); \node [above] at (-2.5,1) {$i+1$}; \filldraw[black] (-2.5,-1) circle (1pt); \node [below] at (-2.5,-1) {$v$}; \draw[thick] (-2.5,1) -- (-2.5,-1); \filldraw[black] (-1.3,2) circle (1pt); \node [above] at (-1.3,2) {$i+2$}; \filldraw[black] (0,2) circle (1pt); \node [above] at (0,2) {$i+4$}; \filldraw[black] (1.7,2) circle (1pt); \node [above] at (1.7,2) {$i-2$}; \filldraw[black] (-1.3,-2) circle (1pt); \node [below] at (-1.3,-2) {$i+1$}; \filldraw[black] (0,-2) circle (1pt); \node [below] at (0,-2) {$i+3$}; \filldraw[black] (1.7,-2) circle (1pt); \node [below] at (1.7,-2) {$i-1$}; \draw[thick] (-1.3,2) -- (-1.3,-2); \draw[thick] (0,2) -- (0,-2); \draw[thick] (1.7,2) -- (1.7,-2); \node at (1,0) {$\cdots$}; \filldraw[black] (5.5,0) circle (1pt); \node [above right] at (5.5,0) {$i$}; \draw[thick,->] (4,0) -- (5.5,0); \end{tikzpicture} \caption{$T_i$ in the oriented extremal construction on $n+1$ vertices for even $n$} \end{figure} If $H$ does contain such a triple, then there exist three integers in $\mathbb{Z}_n$ that can be represented as both $\{a,a+2k,a+2k+1\}$ and $\{b,b+2i,b+2i+1\}$ with $a \neq b$. Without loss of generality we can assume that $b=0$. If $a+2k = 0$, then $a+2k+1 = 1$, but $1$ is not in any tail that points at $0$. Therefore, it must be the case that $a+ 2k+1 = 0$, but then $a+2k = n-1$. Therefore, the set is equal to $\{0,n-1,n-2\}$, and $a=n-2$, but $n-1$ does not point to $n-2$, a contradiction. Therefore, $H$ can have no such triple. \begin{figure} \centering \begin{tikzpicture} [scale=0.75] \node at (-4,4) {$T_v$}; \filldraw[color=black,fill=blue!5] (0,0) circle [radius=4]; \filldraw[black] (-2.5,1) circle (1pt); \node [above] at (-2.5,1) {$0$}; \filldraw[black] (-2.5,-1) circle (1pt); \node [below] at (-2.5,-1) {$\frac{n}{2}$}; \draw[thick] (-2.5,1) -- (-2.5,-1); \filldraw[black] (-1.3,2) circle (1pt); \node [above] at (-1.3,2) {$1$}; \filldraw[black] (0,2) circle (1pt); \node [above] at (0,2) {$2$}; \filldraw[black] (1.7,2) circle (1pt); \node [above] at (1.7,2) {$\frac{n}{2} - 1$}; \filldraw[black] (-1.3,-2) circle (1pt); \node [below] at (-1.3,-2) {$n-1$}; \filldraw[black] (0,-2) circle (1pt); \node [below] at (0,-2) {$n-2$}; \filldraw[black] (1.7,-2) circle (1pt); \node [below] at (1.7,-2) {$\frac{n}{2}+1$}; \draw[thick] (-1.3,2) -- (-1.3,-2); \draw[thick] (0,2) -- (0,-2); \draw[thick] (1.7,2) -- (1.7,-2); \node at (1,0) {$\cdots$}; \filldraw[black] (5.5,0) circle (1pt); \node [above right] at (5.5,0) {$v$}; \draw[thick,->] (4,0) -- (5.5,0); \end{tikzpicture} \caption{$T_v$ in the oriented extremal construction on $n+1$ vertices for even $n$} \end{figure} Now, we consider odd $n+1$. Here, let $V(H) = \mathbb{Z}_n \cup \{v\}$ where $v$ is a new vertex and use all the edges from the even construction plus some new ones that all contain $v$. So $E(H) = E_{even} \cup E_{new} \cup E_{v}$ where \[E_{even} = \bigcup_{i=0}^{n-1} \left\{(i+2k)(i+2k+1) \rightarrow i : k=1,2,\ldots,\frac{n-2}{2} \right\},\]and \[E_{new}= \left\{v(i+1) \rightarrow i: i=0,1,\ldots,n-1\right\}.\] Certainly, the construction has so far avoided the forbidden subgraph and given each of the first $n$ vertices the maximum number of tails. Now $E_v$ can be constructed as any set of $\frac{n}{2}$ disjoint pairs of elements from $\mathbb{Z}_n$ all pointing at $v$ so that no pair consists of two sequential numbers mod $n$. So any maximum matching of the $n$ elements that observes this condition will do. In particular, we can let \[E_v = \left\{(i)(n-i) \rightarrow v : i = 1,\dots,\frac{n}{2}-1\right\} \cup \left\{(0)\left(\frac{n}{2}\right) \rightarrow v\right\}.\] So \[\text{ex}_o(n,I_1) = n \left\lfloor \frac{n-1}{2} \right\rfloor.\] \end{proof} \section{Forbidden $H_1$} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,2) circle (1pt); \filldraw[black] (-1,0) circle (1pt); \filldraw[black] (1,0) circle (1pt); \filldraw[black] (1.5,1.5) circle (1pt); \filldraw[black] (-1.5,1.5) circle (1pt); \draw[thick] (0,2) -- (-1,0); \draw[thick] (0,2) -- (1,0); \draw[thick,->] (0.5,1) -- (1.5,1.5); \draw[thick,->] (-0.5,1) -- (-1.5,1.5); \node[above] at (0,2) {$x$}; \node[above] at (-1.5,1.5) {$b$}; \node[above] at (1.5,1.5) {$d$}; \node[left] at (-1,0) {$a$}; \node[right] at (1,0) {$c$}; \end{tikzpicture} \caption{$H_1$} \label{B} \end{figure} In this section $H_1$ denotes the forbidden graph where two edges intersect in exactly one vertex such that it is in the tail for each edge. That is $V(H_1) = \{a,b,c,d,x\}$ and $E(H_1) = \{ax \rightarrow b, cx \rightarrow d\}$ (see Figure~\ref{B}). First we will show the following result for the oriented version of the problem. \begin{theorem} \label{exB} For all $n \geq 6$, \[\text{ex}_o(n,H_1)=\left\lfloor \frac{n}{2} \right\rfloor (n-2).\] \end{theorem} We will use this result to solve the standard version of the problem and get the following. \begin{theorem} \label{TypeB} For all $n \geq 8$, \[\text{ex}(n,H_1) = {n+1 \choose 2} - 3.\] \end{theorem} First, note that the proof of Theorem~\ref{exB} is straightforward when $n$ is even. To get a lower bound construction we can take a maximum matching of the $n$ vertices and use each pair of this matching as the tail set to point at all $n-2$ other vertices. That is, let $H$ be the graph with vertex set, \[V(H) = \{x_1,\ldots,x_{\frac{n}{2}}, y_1, \ldots, y_{\frac{n}{2}}\}\] and edge set, \[E(H) = \bigcup_{i=1}^{\frac{n}{2}} \left\{x_iy_i \rightarrow z: z \in V(H) \setminus \{x_i,y_i\}\right\}.\] To show that this is also an upper bound, let $H$ be an $H_1$-free oriented graph on $n$ vertices. Then for any $x \in V(H)$, the directed link graph $D_x$ cannot have two independent edges (see Figure~\ref{Bdirlink}). Therefore, $D_x$ is either empty, a triangle, or a star with at most $n-2$ edges. Since $n \geq 5$, then $|D_x| \leq n-2$ for each $x$. So \[|E(H)| = \frac{1}{2} \sum_{x \in V(H)} |D_x| \leq \frac{1}{2}n(n-2).\] So we are finished for even $n$. However, this proof falls apart when $n$ is odd. We will need a different strategy. \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (0,1) circle [radius=2]; \node at (-2,2.5) {$H$}; \filldraw[black] (0,2) circle (1pt); \node [above] at (0,2) {$x$}; \filldraw[black] (-1,0) circle (1pt); \node [left] at (-1,0) {$y$}; \filldraw[black] (1,0) circle (1pt); \node [right] at (1,0) {$a$}; \filldraw[black] (1.5,1.5) circle (1pt); \node [right] at (1.5,1.5) {$b$}; \filldraw[black] (-1.5,1.5) circle (1pt); \node [left] at (-1.5,1.5) {$z$}; \draw[thick] (0,2) -- (-1,0); \draw[thick] (0,2) -- (1,0); \draw[thick,->] (0.5,1) -- (1.5,1.5); \draw[thick,->] (-0.5,1) -- (-1.5,1.5); \filldraw[color=black,fill=blue!5] (6,1) circle [radius=2]; \node at (4,2.5) {$D_x$}; \filldraw[black] (4.5,1.5) circle (1pt); \node [left] at (4.5,1.5) {$z$}; \node [left] at (5,0) {$y$}; \filldraw[black] (7,0) circle (1pt); \node [right] at (7,0) {$a$}; \filldraw[black] (7.5,1.5) circle (1pt); \node [right] at (7.5,1.5) {$b$}; \draw[thick,->] (7,0) -- (7.5,1.5); \draw[thick,->] (5,0) -- (4.5,1.5); \node at (3,0.75) {$\iff$}; \end{tikzpicture} \caption{$H$ has a copy of $H_1$ with intersection vertex $x$ if and only if the directed link graph $D_x$ has a pair of disjoint directed edges.} \label{Bdirlink} \end{figure} \subsection{Counting edges by possible tail pairs} The basis of our strategy in getting an upper bound on $\text{ex}_o(n,H_1)$ is to count the edges of an $H_1$-free graph $H$ by its tail sets. That is, \[|E(H)| = \sum_{\{x,y\} \in {V(H) \choose 2}} t(x,y)\] It is simple but important to note that if $H$ is $H_1$-free, then any two pairs of vertices that each point to two or more other vertices must necessarily be disjoint. \begin{lemma} Let $H$ be a $H_1$-free oriented graph. If $x_1,x_2,y_1,y_2 \in V(H)$ so that $t(x_1,y_1),t(x_2,y_2) \geq 2$ and $\{x_1,y_1\} \neq \{x_2,y_2\}$, then $\{x_1,y_1\} \cap \{x_2,y_2\} = \emptyset$. \end{lemma} \begin{proof} Suppose, towards a contradiction, that $x_1 = x_2 = x$ but $y_1 \neq y_2$. Since $t(x,y_1) \geq 2$, then there exists some vertex $z_1$ distinct from $x$, $y_1$, and $y_2$ such that \[xy_1 \rightarrow z_1 \in E(H).\] Similarly, since $t(x,y_2) \geq 2$, then there exists some vertex $z_2$ distinct from $x$, $y_1$, and $y_2$ such that \[xy_2 \rightarrow z_2 \in E(H).\] If $z_1 \neq z_2$ this gives a copy of $H_1$. So assume that they are the same vertex, $z_1=z_2=z$. Since $t(x,y_1) \geq 2$, then there is some second vertex that $x$ and $y_1$ point to that is distinct from $z$. The only choice that would not create a copy of $H_1$ with the edge $xy_2 \rightarrow z$ is $y_2$. Similarly, since $t(x,y_2) \geq 2$, then there is some second vertex that $x$ and $y_2$ point to that is distinct from $z$. The only choice that would not create a copy of $H_1$ with the edge $xy_1 \rightarrow z$ is $y_1$. So \[xy_1 \rightarrow y_2, xy_2 \rightarrow y_1 \in E(H)\] which contradicts the fact that $H$ is oriented. \end{proof} Therefore, if we assume that $H$ is $H_1$-free on $n$ vertices, then we can split its vertices up into $k$ disjoint pairs that each serve as tail sets to at least two edges of $H$ plus a set of $n-2k$ vertices that belong to no such pair. That is, \[V(H) = \{x_1,y_1,\ldots,x_k,y_k\} \cup R\] so that $t(x_i,y_i) \geq 2$ for $i=1,\ldots,k$ and $t(w,v) \leq 1$ for all other vertex pairs, $\{w,v\}$ (see Figure~\ref{twopluspairs}). \begin{figure} \centering \begin{tikzpicture} \node at (-1,4) {$H$}; \filldraw[black] (0,3) circle (1pt); \node [left] at (0,3) {$x_1$}; \filldraw[black] (0,2) circle (1pt); \node [left] at (0,2) {$x_2$}; \filldraw[black] (0,0) circle (1pt); \node [left] at (0,0) {$x_k$}; \filldraw[black] (2,3) circle (1pt); \node [right] at (2,3) {$y_1$}; \filldraw[black] (2,2) circle (1pt); \node [right] at (2,2) {$y_2$}; \filldraw[black] (2,0) circle (1pt); \node [right] at (2,0) {$y_k$}; \draw[thick] (0,3) -- (2,3); \draw[thick] (0,2) -- (2,2); \draw[thick] (0,0) -- (2,0); \node at (1,1) {$\vdots$}; \filldraw[color=black,fill=blue!5] (5,1.5) circle [radius=1.5]; \node at (5,1.5) {$R$}; \end{tikzpicture} \caption{An $H_1$-free graph on $n$ vertices breaks down into $k$ disjoint pairs that each point to at least two other vertices plus a remainder set $R$ with $n-2k$ vertices that belong to no such pair.} \label{twopluspairs} \end{figure} We now have two cases to consider. Either there are no such pairs ($k=0$) or there is at least one ($k \geq 1$). \subsection{No pair points to more than one vertex ($k=0$)} Assume that $k=0$. Then $t(x,y) \leq 1$ for every pair $\{x,y\} \in {V(H) \choose 2}$. If $|D_x| \leq n-3$ for all $x \in V(H)$, then \[|E(H)| = \frac{1}{2} \sum_{x \in V(H)} |D_x| \leq \frac{1}{2}n(n-3) < \frac{1}{2}(n-1)(n-2)\] and we are done. Otherwise, there exists some vertex $x$ that belongs to $n-2$ tail sets. Therefore, $D_x$ is a star of directed edges with some focus $y$. Either $t(x,y)=0$ or $t(x,y)=1$. If $t(x,y) = 0$, then all of the $n-2$ directed edges of $D_x$ must point to $y$ (see Figure~\ref{CaseApic}). Such a configuration in $H$ limits the number of edges to ${n-1 \choose 2}$ as proven in Lemma~\ref{CaseA}. \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (0,-2) ellipse (1.5 and 1); \node [left] at (-1.5, -2) {$n-2$}; \filldraw[black] (0,0) circle (1pt); \filldraw[black] (xyz polar cs:angle=240, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=250, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=260, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=290, radius=2) circle (1pt); \filldraw[black] (2.5,-1) circle (1pt); \draw[thick] (0,0)--(xyz polar cs:angle=240, radius=2); \draw[thick] (0,0)--(xyz polar cs:angle=250, radius=2); \draw[thick] (0,0)--(xyz polar cs:angle=260, radius=2); \draw[thick] (0,0)--(xyz polar cs:angle=290, radius=2); \draw[thick,->] (0.35,-1)--(2.5,-1); \node [above] at (0,0) {$x$}; \node at (xyz polar cs:angle=275, radius=1.5) {$\cdots$}; \node [right] at (2.5,-1) {$y$}; \end{tikzpicture} \caption{The special case configuration discussed in Lemma~\ref{CaseA}. Here, vertex $x$ joins with every other element to point to vertex $y$.} \label{CaseApic} \end{figure} On the other hand, if $t(x,y) = 1$, then $xy \rightarrow z \in E(H)$ for some vertex $z$, and $xv \rightarrow y$ for all other vertices $v \neq x,y,z$. Such a configuration in $H$ will limit the number of edges to ${n-1 \choose 2}$ as proven in Lemma~\ref{CaseB}. \begin{lemma} \label{CaseA} Let $H$ be an oriented graph on $n \geq 6$ vertices such that $t(x,y) \leq 1$ for each pair $\{x,y\} \in {V(H) \choose 2}$. If $H$ is $H_1$-free and contains vertices $x$ and $y$ such that $xv \rightarrow y \in E(H)$ for each $v \in V(H) \setminus \{x,y\}$, then \[|E(H)| \leq {n-1 \choose 2}.\] See Figure~\ref{CaseApic}. \end{lemma} \begin{proof} We want to show that there can be no more than ${n-2 \choose 2}$ additional edges in $H$ other than the $n-2$ edges described in the statement of the lemma. This would give an upper bound on the total number of edges of \[{n-2 \choose 2} + (n-2) = {n-1 \choose 2}.\] First, note that every triple of the form $\{x,y,v\}$ already holds an edge. This implies that any additional edge cannot contain both $x$ and $y$ since $H$ is oriented. On the other hand, if we were to add an edge, $vw \rightarrow u$, that excluded both $x$ and $y$ completely, then this new edge would create a copy of $H_1$ with the existing edge, $vx \rightarrow y$. Therefore, every additional edge must be on a triple of the form $\{v,w,x\}$ or $\{v,w,y\}$. However, $x$ is already in the maximum number of tails. So given any pair of non-$\{x,y\}$ vertices, $\{v,w\}$, the only possible additional edges are \[vw \rightarrow x, vw \rightarrow y, yv \rightarrow w, \text{and } yw \rightarrow v.\] The last three all appear on the triple, $\{v,w,y\}$ and are therefore mutually exclusive choices when it comes to adding them to the graph. The first two are also mutually exclusive choices since $t(v,w) \leq 1$. So assume, towards a contradiction, that we could add ${n-2 \choose 2} + 1$ more edges to the existing configuration. Then some pair $\{v,w\}$ of non-$\{x,y\}$ vertices must be used twice. Without loss of generality, this means we must add the edges $vw \rightarrow x$ and $yv \rightarrow w$. Now, let $u$ be any of the remaining $n-4$ vertices. The possible edge $uv \rightarrow y$ would create a copy of $H_1$ with $vw \rightarrow x$, and the possible edge $uv \rightarrow x$ would create a copy of $H_1$ with $vy \rightarrow w$. Therefore, the pair $\{v,u\}$ cannot be a tail set for any edge. We can also view the potential additional edges as two different types: those that have a tail set of two non-$\{x,y\}$ vertices and those that have a tail set that includes $y$. There were originally at most ${n-2 \choose 2}$ of the first type that we are allowed to add in total, one edge for every distinct pair. However, $v$ can now no longer be in a tail set with any of the other $n-4$ vertices. So there are now at most ${n-2 \choose 2} - (n-4)$ edges of this first type left possible to add. Therefore, in order to add ${n-2 \choose 2} + 1$ edges over all, we will need at least $n-3$ of them to be of the second type - those that have $y$ in the tail set. Note that $x$ must be an isolated vertex in the directed link graph $D_y$. Hence, there are at most $n-3$ tails containing $y$ since otherwise the directed graph $D_y$ would have $n-2$ edges among $n-2$ vertices. In this case, $D_y$ would have two independent directed edges and so $H$ would have a copy of $H_1$ with $y$ as its intersection vertex. Moreover, $D_y$ must be a star with a single vertex of intersection. Since $v \rightarrow w \in E(D_y)$, then this vertex of intersection must either be $v$ or $w$. Hence, in order to add ${n-2 \choose 2} + 1$ edges, we'll need to have ${n-2 \choose 2} - (n-4)$ edges that have non-$\{x,y\}$ tail sets. Since the tail set, $\{v,w\}$, already points to $x$, then this implies that all such edges must also point to $x$. Otherwise, we'd have some edge of the form $ab \rightarrow y$. If $a=w$ or $b=w$, then this would create a copy of $H_1$ with $vw \rightarrow x$. If both elements are distinct from $w$, then we would still need to point the pair $wa$ either to $x$ or to $y$. Either choice would create a copy of $H_1$. Let $u$ be one of the remaining vertices. Then $u$ must be adjacent to a directed edge of $D_y$ for there to be $n-3$ edges added with $y$ in the tail set. If $v$ is the vertex of intersection of $D_y$, then this edge must either be $u \rightarrow v$ or $v \rightarrow u$. Either yields a copy of $H_1$. Similarly, if $w$ is the vertex of intersection of $D_y$, then either $wy \rightarrow u \in E(H)$ or $uy \rightarrow w \in E(H)$. Again, either of these yields a copy of $H_1$. Therefore, ${n-2 \choose 2} + 1$ edges cannot be added to the existing configuration. \end{proof} \begin{lemma} \label{CaseB} Let $H$ be an oriented graph on $n \geq 6$ vertices such that for each pair $x,y \in V(H)$, $t(x,y) \leq 1$. If $H$ is $H_1$-free and contains vertices $x$, $y$, and $z$ such that $xy \rightarrow z \in E(H)$ and $xv \rightarrow y \in E(H)$ for each $v \in V(H) \setminus \{x,y,z\}$ (see Figure~\ref{Bpic}), then \[|E(H)| \leq {n-1 \choose 2}.\] \end{lemma} \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (0,-2) ellipse (1.5 and 1); \node [left] at (-1.5, -2) {$n-3$}; \filldraw[black] (0,0) circle (1pt); \filldraw[black] (xyz polar cs:angle=240, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=250, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=260, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=290, radius=2) circle (1pt); \filldraw[black] (2.5,-1) circle (1pt); \filldraw[black] (1.75,0.75) circle (1pt); \draw[thick] (0,0)--(xyz polar cs:angle=240, radius=2); \draw[thick] (0,0)--(xyz polar cs:angle=250, radius=2); \draw[thick] (0,0)--(xyz polar cs:angle=260, radius=2); \draw[thick] (0,0)--(xyz polar cs:angle=290, radius=2); \draw[thick,->] (0.35,-1)--(2.5,-1); \draw[thick] (0,0)--(2.5,-1); \draw[thick,->] (1.25,-0.5)--(1.75,0.75); \node [above] at (0,0) {$x$}; \node at (xyz polar cs:angle=275, radius=1.5) {$\cdots$}; \node [right] at (2.5,-1) {$y$}; \node [right] at (1.75,0.75) {$z$}; \end{tikzpicture} \caption{The special case configuration discussed in Lemma~\ref{CaseB}. Here, $x$ joins with every vertex except $z$ to point to $y$ and then joins with $y$ to point to $z$.} \label{Bpic} \end{figure} \begin{proof} Let $W = \{1,2,\ldots,n-3\}$ be the set of non-$\{x,y,z\}$ vertices. Any additional edge to this graph must have a tail set of the form $\{i,j\}$, $\{i,y\}$, $\{i,z\}$, or $\{y,z\}$ for $i,j \in W$. An $ij$ tail can only point to $x$ or to $y$ and there are ${n-3 \choose 2}$ pairs like this possible. An $iy$ tail cannot point to $x$ because $H$ is oriented. It cannot point to $j$ since that would create a copy of $H_1$ with $xy \rightarrow z$. Therefore, it could only point to $z$. An $iz$ tail could not point to any $j$ since this would create a copy of $H_1$ with the edge $ix \rightarrow y$. Therefore, it could only point to $y$ or to $x$. And a $yz$ tail could not point to $x$ since $H$ is oriented. Therefore, it could only point to some $i$. Assume, towards a contradiction, that we can add \[{n-1 \choose 2} + 1= {n-3 \choose 2} + (n-3)+ 1\] edges to the existing configuration. Since we can add at most ${n-3 \choose 2}$ edges with tail sets made entirely of vertices from $W$, then we must have at least $n-2$ additional edges from the other possibilities. For each $i \in W$ we could have \[iy \rightarrow z, yz \rightarrow i, iz \rightarrow y, \text{and } iz \rightarrow x.\] The first three of these are mutually exclusive choices since they are all on the same triple. Similarly, the last two are mutually exclusive choices since we are only allowing up to one edge per possible tail set. Therefore, in order to add $n-2$ of these types of edges, two must use the same element of $W$. Given the mutually exclusive choices above this implies that there is some vertex $i \in W$ such that either $iz \rightarrow x, yi \rightarrow z \in E(H)$ or $iz \rightarrow x, yz \rightarrow i \in E(H)$. In the first case, $ij$ is no longer a possible tail for any edge for all $n-4$ remaining vertices $j \in W$. This is because $iz \rightarrow x$, $yi \rightarrow z$, and $ix \rightarrow y$ create a triangle in $D_i$. So any additional edge with $i$ in the tail would give two independent edges in $D_i$ and therefore a copy of $H_1$. Hence, we can get at most ${n-3 \choose 2} - (n-4)$ edges with tails in $W$. This means that we will need $2(n-3)$ edges from the other possible edges to make up the difference if we want to add \[{n-3 \choose 2} + (n-3) + 1\] more edges. Since each of the $n-3$ vertices from $W$ can be in up to two of these additional edges, then $iz \rightarrow x$ would need to be an edge for every $i \in W$ and that $\{y,z,i\}$ also needs to hold one edge for every $i \in W$. If $yz \rightarrow i$ is used once, then we get a copy of $H_1$ with $jz \rightarrow x$ for some other $j \in W$. Therefore, for all $i \in W$ we must have the edges $iy \rightarrow z$ and $iz \rightarrow x$. However, any pair $i,j \in W$ can now point to nothing since the only possibilities for such a tail were $x$ or $y$ to begin with and both of these options create copies of $H_1$. So in this case the most that we can add is \[2(n-3) \leq {n-3 \choose 2} + (n-3)\] for all $n \geq 6$. In the other case we have added $iz \rightarrow x$ and $yz \rightarrow i$ for some $i$. Which means that $yz \rightarrow j$ is not allowed for any $j \neq i$ from $W$. Also, $jz \rightarrow y$ would make a copy of $H_1$ with $iz \rightarrow x$ and $jz \rightarrow x$ would make a copy of $H_1$ with $yz \rightarrow i$. Therefore, for all $j \neq i$ we can only add the edge $jy \rightarrow z$. In order to add ${n-3 \choose 2} + n-2$ edges, we'll need all of these as well as all possible edges with tails in $W$. However, since $iz \rightarrow x$, all of these edges with tails completely in $W$ must also point to $x$. Otherwise, some pair $ab$ would point to $y$. If $a=i$ or $b=i$, then this would make a copy of $H_1$ with $iz \rightarrow x$. If $i \neq a,b$, then consider where the pair $ai$ points. It must either point to $x$ or to $y$, but either of these would create a copy of $H_1$. So all pairs of $W$ must point to $x$ and for all $j \in W$ not equal to $i$ we must have the edge $jy \rightarrow z$. But $jy \rightarrow z$ and $ij \rightarrow x$ create a copy of $H_1$, a contradiction. Hence, it is not possible to add more than ${n-3 \choose 2} + (n-3)$ edges to the configuration. Since the configuration already has $n-2$ edges, then there can be no more than ${n-1 \choose 2}$ edges total. \end{proof} Together these two lemmas take care of the cases where all pairs of vertices point to at most one vertex in $H$. \subsection{At least one pair of vertices is the tail set to more than one edge of $H$ ($k > 0$)} So let's return to our description of an $H_1$-free oriented graph as being made up of $k \geq 1$ vertex pairs that each serve as tails to strictly more than one edge plus a set $R$ of the remaining $n-2k$ vertices, \[V(H) = \{x_1,y_1,\ldots,x_k,y_k\} \cup R\] (see Figure~\ref{twopluspairs}). For each pair $\{x_i,y_i\}$ we want to prove the following upper bound, \[t(x_i,y_i) + \sum_{v \neq x_i,y_i} \left(t(x_i,v)+t(y_i,v)\right) \leq n-2.\] That is, the total number of edges that include either $x_i$ or $y_i$ or both in the tail set is at most $n-2$. Now, \[|E(H)| = \sum_{\{x,y\} \in {V(H) \choose 2}} t(x,y) \leq \sum_{i=1}^k \left(t(x_i,y_i) + \sum_{v \neq x_i,y_i} \left(t(x_i,v)+t(y_i,v)\right) \right) + \sum_{\{x,y\} \in {R \choose 2}} t(x,y).\] Note that each pair of vertices in $R$ act as a tail set at most once so \[\sum_{\{x,y\} \in {R \choose 2}} t(x,y) \leq {n-2k \choose 2}.\] Therefore, proving the upper bound for each $\{x_i,y_i\}$ pair would imply that \[|E(H)| \leq k(n-2) + {n-2k \choose 2}.\] Since \[k(n-2) + {n-2k \choose 2} = 2k^2 - (n+1)k + {n \choose 2}\] is a quadratic polynomial with positive leading coefficient in terms of $k$, then it is maximized at the endpoints. Here, that means at $k=1$ and at $k= \left\lfloor\frac{n}{2}\right\rfloor$. When $n$ is odd, both of these values for $k$ give the upper bound, \[|E(H)| \leq {n-1 \choose 2}.\] Only when $n$ is even can we do better and get \[|E(H)| \leq \frac{n(n-2)}{2}\] in the case where $k=\frac{n}{2}$. In either case this give an upper bound of \[|E(H)| \leq \left\lfloor \frac{n}{2} \right\rfloor (n-2).\] So we need only prove that, in general, \[t(x_i,y_i) + \sum_{v \neq x_i,y_i} \left(t(x_i,v)+t(y_i,v)\right) \leq n-2.\] This is straightforward to show if $t(x_i,y_i) \geq 3$. However, when $t(x_i,y_i)=2$ there is a case where it fails to hold. This is taken care of in the following lemma. \begin{lemma} \label{CaseC} Let $H$ be an oriented graph on $n \geq 6$ vertices. If $H$ is $H_1$-free and contains vertices $x$, $y$, $a$, and $b$ such that $\{x,y\}$ is the tail set to exactly 2 edges with \[xy \rightarrow a, xy \rightarrow b, yb \rightarrow a \in E(H),\] and for each $v \in V(H) \setminus \{x,y,a,b\}$, $xv \rightarrow y$ (see Figure~\ref{CaseCpic}), then \[|E(H)| \leq {n-1 \choose 2}.\] \end{lemma} \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (-2,1) ellipse (1 and 2); \filldraw[black] (-2,2) circle (1pt); \filldraw[black] (-2,1.5) circle (1pt); \filldraw[black] (-2,0) circle (1pt); \node [left] at (-3,1) {$n-4$}; \node at (-2,0.8) {$\vdots$}; \draw[thick] (0,2) -- (-2,2); \draw[thick] (0,2) -- (-2,1.5); \draw[thick] (0,2) -- (-2,0); \draw[thick,->] (-1,1) -- (0,0); \filldraw[black] (0,2) circle (1pt); \filldraw[black] (0,0) circle (1pt); \filldraw[black] (1.8,1.5) circle (1pt); \filldraw[black] (1.8,0.5) circle (1pt); \node [above] at (0,2) {$x$}; \node [below] at (0,0) {$y$}; \node [right] at (1.8,1.5) {$a$}; \node [right] at (1.8,0.5) {$b$}; \draw[thick] (0,2) -- (0,0); \draw[thick, ->] (0,1) -- (1.8,1.5); \draw[thick, ->] (0,1) -- (1.8,0.5); \draw[thick] (0,0) -- (1.8,0.5); \draw[thick, ->] (0.9,0.25) -- (1.8,1.5); \end{tikzpicture} \caption{An $H_1$-free graph containing this configuration with have at most ${n-1 \choose 2}$ edges as shown in Lemma~\ref{CaseC}.} \label{CaseCpic} \end{figure} \begin{proof} First consider which pairs of vertices could possibly be a tail set to an edge in this graph. Let $W=\{1,\ldots,n-4\}$ be the set of vertices other than $\{x,y,a,b\}$. Then $\{i,j\}$ can be a tail set to $ij \rightarrow x$ and $ij \rightarrow y$ for any pair $i,j \in W$. Since $xy \rightarrow a$, then $xi$ can point to nothing other than $y$. Similarly, $xa$ and $xb$ could only possibly point to $b$ and $a$ respectively, but either would create a copy of $H_1$ with $xi \rightarrow y$ for any $i \in W$. Also, by assumption $xy$ can point to nothing else. Hence, $x$ is in no additional tail sets. Since $yb \rightarrow a$ and $xy \rightarrow a$, then $ya$ cannot point to $b$ or to $x$. It can also not point to any $i \in W$ since this would create a copy of $H_1$ with $xy \rightarrow b$. So $y$ can be in no additional tails. The pair $ab$ can point to anything aside from $y$ since $H$ is oriented, and $ai$ can point to $x$ or $y$ for any $i \in W$ but not to $b$ or another element of $W$ since either would create a copy of $H_1$ with $xi \rightarrow y$. Similarly, $bi$ can point to $y$ for each $i \in W$ but not to $x$ or to $a$ or to another element of $W$ since these would create a copy of $H_1$ with either $yb \rightarrow a$ or $xi \rightarrow y$. Leaving aside the edges with tail sets completely in $W$ for the moment, this means there are $4(n-4) + 1$ possible edges remaining. There are $n-4$ each of types $ai \rightarrow x$, $ai \rightarrow y$, $bi \rightarrow y$, and $ab \rightarrow i$ plus one extra edge which is $ab \rightarrow x$. Suppose we are able to use at least $2(n-4)+1$ of these edges. First, if one of them is $ab \rightarrow x$, then there could be none of the types $ai \rightarrow y$ or $bi \rightarrow y$. So all of the ones of type $ab \rightarrow i$ and $ai \rightarrow x$ would need to be used. But since $n \geq 6$ there are at least two vertices in $W$. So there would exist edges $ai \rightarrow x$ and $ab \rightarrow j$ with $i \neq j$, a copy of $H_1$. Therefore $ab \rightarrow x$ cannot be used if we want to get more than $2(n-4)$ of these edges. Hence, we need to use at least three types of edges from the four possible types. Since any of the types $ai \rightarrow x$, $ai \rightarrow y$, and $bi \rightarrow y$ eliminate the possibility of using any edge $ab \rightarrow j$ where $j \neq i$, then we can use at most one of this last type of edge. But since $n \geq 6$, then $2(n-4)+1 \geq 5$ which means one of the other types gets used at least twice. Regardless of which one it is, there can be nothing used from the $ab \rightarrow i$ types of edges. Therefore, we must use $2(n-4)+1$ edges from only the first three types. So there must be a vertex from $W$ that belongs to three of these edges, say \[ai \rightarrow y, bi \rightarrow y, \text{and } ai \rightarrow x.\] Then for any $j \neq i$, $aj \rightarrow x$ creates a copy of $H_1$ with $ai \rightarrow y$ and $aj \rightarrow x$ creates a copy of $H_1$ with $ai \rightarrow x$. So at most $2 + (n-4) < 2(n-4)$ edges could be used. Thus, at most $2(n-4)$ of these kinds of edges can be used over all. Now let us look at the edges with tail sets contained in $W$. We have seen that each $ij$ can point to $x$ or to $y$, but nothing so far has kept the pair from pointing to both. However, if some pair does point to both, then no other tail could use either of these vertices since this would create a copy of $H_1$. Therefore, if there are $l$ such pairs, then there are at most $2l + {n - 4 - 2l \choose 2}$ edges with tails from $W$. But since $n \geq 5$, then $\frac{n+5}{2} \leq n$. And since $l \leq \frac{n-4}{2}$, then \[2l + {n-4-2l \choose 2} \leq {n-4 \choose 2}.\] So there are at most ${n-1 \choose 2}$ edges in $H$. \end{proof} \subsection{First main result, $\text{ex}_o(n,H_1) = \left\lfloor \frac{n}{2} \right\rfloor (n-2)$.} Now we can proceed with establishing the upper bound under the assumption that the configuration presented in Lemma~\ref{CaseC} does not occur in our directed hypergraph. As we've seen, all that's necessary to show is that \[t(x_i,y_i) + \sum_{v \neq x_i,y_i} \left(t(x_i,v)+t(y_i,v)\right) \leq n-2\] for any pair of vertices $\{x_i,y_i\}$ that serves as the tail set to at least two edges. So let $\{x,y\}$ be such a pair, and divide the rest of the vertices of $H$ into two groups, those that are a head vertex to some edge with $xy$ as the tail and those that are not. That is, \[V(H) \setminus \{x,y\} = \{h_1,\ldots,h_m\} \cup \{n_1,\ldots,n_t\}\] where for each $i=1,\ldots,m$, there exists an edge, $xy \rightarrow h_i \in E(H)$ and for each $j=1,\ldots,t$, $xy \rightarrow n_j \not \in E(H)$ (note that $t(x,y)=m$ and that $m+t=n-2$). Now, consider an edge that contains either $x$ or $y$ in the tail but not both. Then the other tail vertex is either some $h_i$ or some $n_j$. In the case of $n_j$, this edge must either be of the form $xn_j \rightarrow y$ or $yn_j \rightarrow x$ to avoid a copy of $H_1$ with both $xy \rightarrow h_1$ and $xy \rightarrow h_2$. Moreover, since $H$ is oriented, there can be at most one. Hence, \[\sum_{j=1}^t \left(t(x,n_j)+t(y,n_j)\right) \leq t.\] Now consider a tail set that includes either $x$ or $y$ and some $h_i$. Without loss of generality, assume that $xh_1$ is the tail to some edge. Since $t(x,y) \geq 2$, then there is some other vertex $h_2$ such that $xy \rightarrow h_2 \in E(H)$. In order to avoid a copy of $H_1$ with this edge, $xh_1$ must either point to $y$ or to $h_2$. However, $xh_1 \rightarrow y \not \in E(H)$ since this would give the triple $\{x,y,h_1\}$ more than one edge. Therefore, $xh_1 \rightarrow h_2$ is the only option. However, if $t(x,y) \geq 3$, then this will create a copy of $H_1$ with $xy \rightarrow h_3$. So $xh_i$ and $yh_i$ cannot be tails to any edge. So \[\sum_{i=1}^m \left(t(x,h_i)+t(y,h_i)\right) = 0.\] Therefore, \begin{align*} &t(x,y) + \sum_{v \neq x,y} \left(t(x,v)+t(y,v)\right)\\ &= m + \sum_{j=1}^t \left(t(x,n_j)+t(y,n_j)\right) + \sum_{i=1}^m \left(t(x,h_i)+t(y,h_i)\right)\\ &\leq m+t\\ &=n-2 \end{align*} when $t(x,y) \geq 3$. The only other possibility is that $t(x,y)=2$. So suppose this is the case and that the head vertices to $xy$ are $a$ and $b$. Without loss of generality, assume that $yb \rightarrow a \in E(H)$. Note that this precludes any edges of the form $yn_j \rightarrow x$. Similarly, if we added the edge $xa \rightarrow b$ or the edge $xb \rightarrow a$, then we could not add any edges of the form $xn_j \rightarrow y$ and so \[\sum_{j=1}^t \left(t(x,n_j)+t(y,n_j)\right) = 0.\] Moreover, $ya \rightarrow b$ would lead to more than one edge on the triple $\{y,a,b\}$. So \[\sum_{i=1}^m \left(t(x,h_i)+t(y,h_i)\right) = 2\] and total we would have, \[t(x,y) + \sum_{v \neq x,y} \left(t(x,v)+t(y,v)\right) = 4 \leq n-2.\] On the other hand, if $xa$ and $xb$ are not tails to any edge, then the only way we could get a sum more than $n-2$ is if $xn_j \rightarrow y \in E(H)$ for all $j=1,\ldots, n-4$. But this is exactly the configuration described in Lemma~\ref{CaseC} which we have excluded. Therefore, \[t(x,y) + \sum_{v \neq x,y} \left(t(x,v)+t(y,v)\right) \leq n-2\] for any such pair, and this is enough to establish that \[\text{ex}_o(n,H_1) \leq \left\lfloor \frac{n}{2} \right\rfloor (n-2).\] Conversely, we have already considered an extremal construction in the case where $n$ is even, and this same construction will work when $n$ is odd. That is, take a maximum matching of the vertices (leaving one out) and then use each matched pair as the tail set for all $n-2$ possible edges. Another construction that works for odd $n$ that is not extremal for even $n$ is to designate one vertex as the only head vertex and then make all ${n-1 \choose 2}$ pairs of the rest of the vertices tail sets. Therefore, \[\text{ex}_o(n,H_1) = \left\lfloor \frac{n}{2} \right\rfloor (n-2).\] Also, note that the only way that any construction could have more than ${n-1 \choose 2}$ edges is if $n$ is even \emph{and} the vertices are partitioned into $\frac{n}{2}$ pairs such that each points to at least two other vertices. This fact comes directly from the requirement that $k=\frac{n}{2}$ in the optimization of \[k(n-2) + {n-2k \choose 2}\] in order for the expression to be more than ${n-1 \choose 2}$. \subsection{Intersections of multiedge triples in the standard version} Now, let $H$ be an $H_1$-free graph on $n$ vertices under the standard version of the problem so that any triple of vertices can now have up to all three possible directed edges. If we let $t_H$ be the number of triples of vertices of $H$ that hold at least one edge, and we let $m_H$ be the number of triples that hold at least two, then we have the following simple observation: \[|E(H)| \leq t_H + 2m_H.\] We start our path towards an upper bound on $|E(H)|$ by finding an upper bound on the number of multiedge triples, $m_H$. We will need to prove some facts about the multiedge triples of $H$. First, any triple which holds two edges of $H$ might as well hold three. \begin{lemma} \label{more} Let $H$ be an $H_1$-free graph such that some triple of vertices $\{x,y,z\}$ contains two edges. Define $H'$ by $V(H') = V(H)$ and \[E(H') = E(H) \cup \{xy \rightarrow z,xz \rightarrow y,yz \rightarrow x\}.\]Then $H'$ is also $H_1$-free. \end{lemma} \begin{proof} Suppose $H'$ is not $H_1$-free. Since $H$ is $H_1$-free and the two graphs differ by at most one edge, then they must differ by exactly one edge. Without loss of generality, say \[\{xy \rightarrow z\} = E(H') \setminus E(H).\] This edge must be responsible for creating the copy of $H_1$ in $H'$. So it must intersect another edge in exactly one vertex that is in the tail set of both. Therefore, without loss of generality, there is an edge $xt \rightarrow s \in H$ where $\{s,t\} \cap \{y,z\} = \emptyset$. However, since $\{x,y,z\}$ already contained two edges of $H$, then $xz \rightarrow y \in H$. Since $xt \rightarrow s$ and $xz \rightarrow y$ make a copy of $H_1$, then $H$ cannot be $H_1$-free, a contradiction. \end{proof} Next, we want to show that no two multiedge triples can intersect in exactly one vertex. \begin{lemma} \label{multilimit} Let $H$ be a $H_1$-free graph. If two vertex triples $\{x,y,z\}$ and $\{s,t,r\}$ each contain two or more edges of $H$, then \[|\{x,y,z\} \cap \{s,t,r\}| \neq 1.\] \end{lemma} \begin{proof} Suppose \[|\{x,y,z\} \cap \{s,t,r\}| = 1\]By Lemma~\ref{more}, since $H$ is $H_1$-free, the graph created from $H$ by adding all three possible edges on the triples $\{x,y,z\}$ and $\{s,t,r\}$ is also $H_1$-free. But if $x=r$ and $x$, $y$, $z$, $s$, and $t$ are all distinct, then this graph contains $xy \rightarrow z$ and $xs \rightarrow t$ which is a copy of $H_1$, a contradiction. \end{proof} Therefore, we can use an upper bound on the number of undirected 3-uniform hyperedges such that no two intersect in exactly one vertex as an upper bound on the number of multiedge triples. Moreover, the extremal examples are easy to describe which will be important for finding the upper bound for $\text{ex}(n,H_1)$ as well as for establishing the uniqueness of the lower bound construction. \begin{lemma} \label{undirected} Let $H$ be a 3-uniform undirected hypergraph on $n$ vertices such that no two edges intersect in exactly one vertex, then \[|E(H)| \leq \begin{cases} n & n \equiv 0 \text{ mod } 4\\ n-1 & n \equiv 1 \text{ mod } 4\\ n-2 & n \equiv 2,3 \text{ mod } 4\\ \end{cases}\]and $H$ is the disjoint union of $K_4^{(3)}$s, $K_4^{(3)}$s minus an edge ($K_4^{-}$), and sunflowers with a common intersection of two vertices. \end{lemma} \begin{proof} Two edges of $H$ are either disjoint or they intersect in two vertices. So connected components of $H$ that have 1 or 2 edges are both sunflowers. A third edge can be added to a two-edge sunflower by either using the two common vertices to overlap with both edges in two or by using one common vertex and the two petal vertices. So a connected component of $H$ with 3 edges is either a sunflower or a $K_4^{-}$. The only way to connect a fourth edge to the three-edge sunflower is to make a four-edge sunflower, and this is true for a $k$-edge sunflower to a $(k+1)$-edge sunflower for all $k \geq 3$. The only way to add a fourth edge to the $K_4^{-}$ is to make a $K_4^{(3)}$ and then no new edges may be connected to a $K_4^{(3)}$ without intersecting two of its edges in exactly one vertex each. Therefore, these are the only possible connected components of $H$. A sunflower with $k$ edges uses $k+2$ vertices, and a $K_4^{(3)}$ has four edges on 4 vertices. Therefore, if $n \equiv 0 \text{ mod } 4$ we can get at most $n$ edges with a disjoint collection of $K_4^{(3)}$s. Similarly, the best we can do when $n \equiv 1 \text{ mod } 4$ is $n-1$ edges with a disjoint collection of $K_4^{(3)}$s plus one isolated vertex since any sunflower will automatically limit the number of edges to $n-2$. And if $n \equiv 2 \text{ mod } 4$ or $n \equiv 3 \text{ mod } 4$, then $n-2$ is the best that we can do. \end{proof} In general, the only way to actually have an $H_1$-free graph with $n$ multiedge triples is if the multiedge triples form an undirected 3-uniform hypergraph of $\frac{n}{4}$ disjoint $K_4^{(3)}$ blocks when $n \equiv 0 \text{ mod } 4$. In this case there can be no additional directed edges in $H$ since such an edge would either intersect one of these $K_4^{(3)}$s in one tail vertex which would create a copy of $H_1$ since this means it intersects three of the multiedge triples in exactly one tail vertex (we may assume that each multiedge has all three edges per Lemma~\ref{more}) or it would intersect one of the $K_4^{(3)}$s in two tail vertices which means that it intersects two of the multiedge triples in exactly one tail vertex (see Figure~\ref{blocks}). \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0.4,1.6) circle (1pt); \filldraw[black] (0.4,0.4) circle (1pt); \filldraw[black] (1.6,1.6) circle (1pt); \filldraw[black] (1.6,0.4) circle (1pt); \draw[rounded corners] (-0.1,-0.1) rectangle (2.1,2.1); \node[above left] at (-0.1,2.1) {$K_4^{(3)}$}; \filldraw[black] (2.6,0.6) circle (1pt); \filldraw[black] (2.7,1.7) circle (1pt); \draw[thick] (1.6,1.6) -- (2.6,0.6); \draw[thick, ->] (2.1,1.1) -- (2.7,1.7); \draw[thick, dashed] (1.6,1.6) -- (0.4,0.4); \draw[thick, dashed,->] (1,1) -- (0.4,1.6); \filldraw[black] (5.4,1.6) circle (1pt); \filldraw[black] (5.4,0.4) circle (1pt); \filldraw[black] (6.6,1.6) circle (1pt); \filldraw[black] (6.6,0.4) circle (1pt); \draw[rounded corners] (4.9,-0.1) rectangle (7.1,2.1); \node[above left] at (4.9,2.1) {$K_4^{(3)}$}; \filldraw[black] (7.6,1.1) circle (1pt); \draw[thick] (6.6,1.6) -- (6.6,0.4); \draw[thick, ->] (6.6,1.1) -- (7.6,1.1); \draw[thick, dashed] (6.6,1.6) -- (5.4,0.4); \draw[thick, dashed,->] (6,1) -- (5.4,1.6); \end{tikzpicture} \caption{An edge that intersects a $K_4^{(3)}$ block of multiedge triples in one or two tail vertices will create a copy of $H_1$.} \label{blocks} \end{figure} So in this case, the number of total edges would be bound by \[3n < {n+1 \choose 2} - 3\] for all $n \geq 7$. Next, the only ways to have $n-1$ multiedge triples is to either have $\frac{n-1}{4}$ disjoint $K_4^{(3)}$ blocks when $n \equiv 1 \text{ mod } 4$ or to have $\frac{n}{4} - 1$ disjoint $K_4^{(3)}$ blocks with one $K_4^{-}$ when $n \equiv 0 \text{ mod } 4$. In the first case any additional edge must have at least one and perhaps two of its tail vertices in a single $K_4^{(3)}$ block of multiedge triples which we have already seen will create a copy of $H_1$. So there are at most \[3(n-1) < 3n < {n+1 \choose 2} - 3\] total edges in this case. In the second case, any additional edge that has no tail vertices in a $K_4^{(3)}$ block must have both tail vertices in the $K_4^{-}$. If the head to such an edge were outside of the $K_4^{-}$, then the edge must intersect one of the three multiedge triples of the block in exactly one tail vertex since there are two triples that it intersects in one tail vertex each, one of which must be a multiedge triple. On the other hand, it could have its head vertex inside the $K_4^{-}$. In this case, the additional edge must lie on the triple without multiple edges. This is the only edge that can be added so there are at most \[3(n-1)+1 <3n <{n+1 \choose 2} - 3\] total edges in this case. \subsection{An $H_1$-free graph with $n-2$ multiedge triples} Now, the only ways to have exactly $n-2$ multiedge triples is either to have $\frac{n}{4} - 2$ of the $K_4^{(3)}$ blocks plus two $K_4^{-}$ blocks of multiedge triples when $n \equiv 0 \text{ mod } 4$ or to have $k$ of the $K_4^{(3)}$ blocks of multiedge triples plus a sunflower with $n-4k-2$ petals. The first case is suboptimal for the same reasons already considered. So let's consider the second case. First, assume that $k=0$ and that we have $n-2$ multiedge triples that make a sunflower (see Figure~\ref{mexconst}). How many edges can we add? This structure already has all possible edges with 2 vertices in the core (or so we may assume by Lemma~\ref{more}). On the other hand, if an additional edge has no vertices in the core, then it would intersect two multiedge triples in one tail vertex each which would create a copy of $H_1$. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (0,1.5) circle (1pt); \filldraw[black] (xyz polar cs:angle=200, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=220, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=240, radius=2) circle (1pt); \filldraw[black] (xyz polar cs:angle=290, radius=2) circle (1pt); \draw[thick] (0,1.5) -- (0,0) -- (xyz polar cs:angle=200, radius=2); \draw[thick] (0,1.5) -- (0,0) -- (xyz polar cs:angle=220, radius=2); \draw[thick] (0,1.5) -- (0,0) -- (xyz polar cs:angle=240, radius=2); \draw[thick] (0,1.5) -- (0,0) -- (xyz polar cs:angle=290, radius=2); \draw[rounded corners] (-0.2,-0.2) -- (0.2,-0.2) -- (0.2,1.7) -- (-0.2,1.7) -- cycle; \node at (xyz polar cs:angle=265, radius=1.5) {$\cdots$}; \node [right] at (0.2,0.75) {core vertices}; \node [right] at (1,-1) {$n-2$ petals}; \node [left] at (xyz polar cs:angle=200, radius=2) {${n-2 \choose 2}$ edges pointing back}; \draw[color=blue, thick] (xyz polar cs:angle=200, radius=2) -- (xyz polar cs:angle=220, radius=2); \draw[color=blue, thick,->] (xyz polar cs:angle=210, radius=1.98) -- (0,1.5); \end{tikzpicture} \caption{The unique extremal construction for an $H_1$-free graph has ${n-2 \choose 2} + 3(n-2)$ edges.} \label{mexconst} \end{figure} Therefore, any additional edge must include exactly one vertex from the core. If this vertex is in the tail set to the additional edge and the sunflower has at least three petals, then the additional edge intersects in exactly one tail vertex one of the multiedge triples of the sunflower, a contradiction. Since we assume that $n \geq 6$, then the sunflower has at least three petals. Hence, any additional edge must intersect the core in only its head vertex. If any two additional edges have different core vertices as the head, then either the tails sets are the same or completely disjoint to avoid a copy of $H_1$. Hence, pairs of petal vertices that point to both core vertices must be independent of all other tails sets. And all other petal vertices fall into disjoint sets as to whether they are in additional edges that point to the first core vertex or the second. The number of additional edges will be maximized if every pair of petal vertices point to the same core vertex. Moreover, this will give a total of \[3(n-2) + {n-2 \choose 2} = {n+1 \choose 2} - 3\] edges. We will soon see that this is the best that we can do and that this construction, where the multiedge triples make a sunflower with $n-2$ petals with ${n-2 \choose 2}$ additional edges pointing from pairs of petal vertices to a single core vertex, is unique up to isomorphism. First we will need to see that $k=0$ is the number of $K_4^{(3)}$ multiedge triple blocks that optimizes the total number of edges. So suppose there are $k$ such blocks and that the other $n-4k$ vertices are in a sunflower. Then from prior considerations we know that any additional edge must have both tail vertices in this sunflower. If one of these tail vertices coincides with a petal vertex of the sunflower, then there will be a copy of $H_1$. Therefore, the tail vertices must coincide with the core and the only possibility for such an edge is to point out to a vertex in one of the $k$ blocks. Therefore, there are at most \[3(4k) + 3(n-4k-2) + {n-4k-2 \choose 2} + 4k\] edges in such a construction. Since this expression is quadratic in $k$ with positive leading coefficient, then it must maximize at the endpoints, $k=0$ or $k=\frac{n}{4}$, and we already know that $k=\frac{n}{4}$ is suboptimal. Therefore, if there are exactly $n-2$ multiedge triples, then they must form a sunflower with a two-vertex core and from there the only way to maximize the total number of edges is to add every possible edge with tail set among the petal vertices all pointing to the same head vertex in the core. \subsection{Fewer than $n-2$ multiedge triples} Now suppose that $H$ has fewer than $n-2$ multiedge triples. If $t_H \leq {n-1 \choose 2}$, then \[|E(H)| \leq t_H + 2m_H < {n-1 \choose 2} + 2(n-2) = {n+1 \choose 2} - 3.\] So we must assume that $t_H > {n-1 \choose 2}$. Also, if $m_H=0$, then we know that \[|E(H)| \leq \text{ex}_o(n,H_1) = \left\lfloor \frac{n}{2} \right\rfloor (n-2) < {n+1 \choose 2} - 3.\] So assume that there is at least one multiedge triple, $\{x,y,z\}$. This triple has at least two edges. Assume without loss of generality that they are $xy \rightarrow z$ and $xz \rightarrow y$. Let $H'$ be an oriented graph arrived at by deleting edges from multiedge triples of $H$ until each triple has at most one edge and every triple that had at least one edge in $H$ still has at least one in $H'$. In other words, $H'$ is any subgraph of $H$ such that $t_{H'} = t_{H}$ and $m_{H'}=0$. Without loss of generality, assume that \[xy \rightarrow z \in E(H').\] Since $t_{H'} > {n-1 \choose 2}$, then $n$ must be even, and moreover, there is a matching on the vertices so that every matched pair $\{a,b\}$ points to at least two other vertices. That is, $t(a,b) \geq 2$. Now consider the directed link graphs of the vertices. As stated before, these are either triangles or stars with a common vertex. However, if two or more of these link digraphs have three or fewer edges each (for instance, if they are triangles), then there are fewer edges than we are assuming since \[|E(H')| = \frac{1}{2} \sum_{x \in V(H')} |D_x| \leq \frac{1}{2} \left(6 + (n-3)(n-2) \right) < {n-1 \choose 2}\] for all $n \geq 8$. We will show that it must be the case that here at least two directed link graphs are restricted to at most three directed edges each, contradicting our assumptions about the number of edges in $H$. First, note that $x \rightarrow z \in D_y$ and $y \rightarrow z \in D_x$. To avoid a contradiction, at least one of these two directed link graphs must have four or more edges. Without loss of generality, assume that it is $D_y$. Therefore, $D_y$ is a star and not a triangle. So the additional three directed edges in $D_y$ must either all be incident to $z$ or to $x$. If these directed edges are all incident to $z$, then $y$ and $z$ must be partners under the matching which means that $x$ has another partner $x'$ distinct from $y$ and $z$. Since $t(x,x') \geq 2$ in $H'$, then $x'$ must point to two vertices in $D_x$. Since $D_x$ already has $y \rightarrow z$ and no two edges may be independent in any directed link graph, then $x'$ must point to $y$ and to $z$, forming a triangle. Next, consider $D_{x'}$. We know that \[x \rightarrow y, x \rightarrow z \in D_{x'}.\] If there is an additional edge in $D_{x'}$ that does not complete this triangle then it is either of the form $x \rightarrow t$ or $t \rightarrow x$. If $x \rightarrow t \in D_{x'}$ then $x' \rightarrow t, y \rightarrow z \in D_x$, a contradiction. If $t \rightarrow x \in D_{x'}$, then $x' \rightarrow x \in D_t$. But since $t$ has its own matched vertex, then there exists a distinct $t'$ such that \[t' \rightarrow x, t' \rightarrow x' \in D_{t'}.\] So either $|D_{x'}| \leq 3$ or $|D_{t'}| \leq 3$. Either way, this gives us two directed link graphs that have at most three edges each. So $t_H > {n-1 \choose 2}$. Therefore, we must assume that the three additional edges in $D_y$ are incident to $x$ and that $y$ and $x$ are partners under the matching. So $z$ has some other partner under the matching $z'$ distinct from $x$ and $y$. Now, delete the edge $xy \rightarrow z$ from $H'$ and add $xz \rightarrow y$ to get a new directed hypergraph $H''$. It follows that $H''$ has no multiedge triples and is $H_1$-free since we still have a subgraph of $H$. In adding $xz \rightarrow y$ we have added $x \rightarrow y$ to $D_z$. Since $z'$ must point to two vertices in $D_z$, then this addition means that $D_z$ is a triangle under $H''$. Hence, $|D_z| = 2$ under $H'$. Now, the same argument as above applies to $D_{z'}$. The only way for $|D_{z'}| > 3$ would mean either $z \rightarrow a \in D_{z'}$ or $a \rightarrow z \in D_{z'}$ for some $a$ distinct from $x$, $y$, $z$, and $z'$. The first case would mean that two independent directed edges, $z' \rightarrow a$ and $x \rightarrow y$ are in $D_z$, a contradiction. The second case would mean that $z' \rightarrow z \in D_a$. Since $a$ has its own partner under the matching that must point to two vertices in $D_a$, then in this case, $D_a$ is a triangle. Therefore, $t_H > {n-1 \choose 2}$ and $m_H \geq 1$ cannot both be true in any $H_1$-free graph. This is enough to complete the result, \[\text{ex}(n,H_1) = {n+1 \choose 2} - 3.\] This also exhausts the remaining cases in order to demonstrate that the extremal construction is unique. \section{Forbidden $H_2$} In this section $H_2$ denotes the forbidden graph where two edges intersect in exactly two vertices such that the set of intersection is the tail set to each edge. That is $V(H_2) = \{a,b,c,d\}$ and $E(H_2) = \{ab \rightarrow c, ab \rightarrow d\}$ (see Figure~\ref{D}). \begin{theorem} For all $n \geq 5$, \[\text{ex}(n,H_2) = \text{ex}_o(n,H_2) = {n \choose 2}.\] Moreover, there are ${n \choose 2}^{n-2}$ different labeled $H_2$-free graphs attaining this extremal number when in the standard version of the problem. \end{theorem} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,1) circle (1pt); \filldraw[black] (0,-1) circle (1pt); \filldraw[black] (-2,0) circle (1pt); \filldraw[black] (2,0) circle (1pt); \draw[thick] (0,1) -- (0,-1); \draw[thick, <->] (-2,0) -- (2,0); \node[left] at (-2,0) {$c$}; \node[right] at (2,0) {$d$}; \node[above] at (0,1) {$a$}; \node[below] at (0,-1) {$b$}; \end{tikzpicture} \caption{$H_2$} \label{D} \end{figure} \begin{proof} Let $H$ be $H_2$-free. Regardless of which version of the problem we are considering, each pair of vertices acts as the tail set to at most one directed edge. Therefore, \[\text{ex}(n,H_2), \text{ex}_o(n,H_2) \leq {n \choose 2}.\] In the standard version of the problem any function, $f: {[n] \choose 2} \rightarrow [n]$, that sends each pair of vertices to a distinct third vertex, $f(\{a,b\}) \not \in \{ a,b \}$, has an associated $H_2$-free construction $H_f$ with ${n \choose 2}$ edges. That is, for any such function, $f$, let $V(H_f) = [n]$ and \[E(H_f) = \left\{a,b \rightarrow f(\{a,b\}) : \{a,b\} \in {[n] \choose 2}\right\}.\] Since each pair of vertices acts as the tail set to exactly one directed edge, then $H_f$ is $H_2$-free and has ${n \choose 2}$ edges. So \[\text{ex}(n,H_2) = {n \choose 2}.\] Moreover, there are ${n \choose 2}^{n-2}$ distinct functions from ${[n] \choose 2}$ to $[n]$ such that no pair is mapped to one of its members. Therefore, there are ${n \choose 2}^{n-2}$ labeled graphs that are $H_2$-free with ${n \choose 2}$ edges.\\ In the oriented version of the problem lower bound constructions can be defined inductively on $n$. \begin{figure} \centering \begin{tikzpicture} [scale=0.5] \node at (-3,3) {$H_{n+1}$}; \filldraw[color=black,fill=blue!5] (0,0) circle [radius=3]; \node at (0,0) {$H_n$}; \filldraw[black] (3,3) circle (1pt); \node[above right] at (3,3) {$n$}; \filldraw[black] (0,2.9) circle (1pt); \node[below left] at (0,2.9) {$n-1$}; \filldraw[black] (2.05,2.05) circle (1pt); \node[below left] at (2.05,2.05) {$0$}; \filldraw[black] (2.9,0) circle (1pt); \node[below left] at (2.9,0) {$1$}; \draw[thick] (3,3) -- (0,2.9); \draw[thick] (3,3) -- (2.05,2.05); \draw[thick,->] (1.5,2.95) -- (2.05,2.05); \draw[thick,->] (2.52,2.52) -- (2.9,0); \end{tikzpicture} \caption{Inductive construction of $H_2$-free oriented graphs} \label{D-free} \end{figure} First, let $n=5$ and define $G_5$ as the oriented graph with vertex set \[V(G_5) = \{0,1,2,3,4\}\]and the following edges: $0,1 \rightarrow 2$, $1,3 \rightarrow 0$, $0,4 \rightarrow 1$, $0,2 \rightarrow 3$, $2,4 \rightarrow 0$, $0,3 \rightarrow 4$, $2,3 \rightarrow 1$, $1,2 \rightarrow 4$, $1,4 \rightarrow 3$, and $3,4 \rightarrow 2$. Each pair of vertices of $G_5$ are in exactly one tail set, and each triple of vertices appear together in exactly one edge. Therefore, this construction is $H_2$-free with ${5 \choose 2}$ edges. Now, let $n > 5$, and define $G_n$ by $V(G_n) = \mathbb{Z}_n$ and \[E(G_n) = E(G_{n-1}) \cup \{(n-1)i \rightarrow (i+1) : i = 0,\ldots,n-2\}.\] Then $G_n$ has $n-1$ more edges than $G_{n-1}$. So $|E(H_n)| = {n \choose 2}$. Any two new edges intersect in at most two vertices. Similarly, any new edge and any old edge also intersect in at most two vertices. Hence, at most one edge appears on a given triple of vertices. So each $G_n$ is oriented. Moreover, all tail sets for the new edges are distinct from each other and from any tail sets for the edges of $G_{n-1}$. So $G_n$ is $H_2$-free. Therefore, \[\text{ex}_o(n,H_2) = {n \choose 2}.\] \end{proof} \section{Conclusion} The $2 \rightarrow 1$ version of directed hypergraph originally came to the author's attention as a way to model definite Horn clauses in propositional logic. Definite Horn clauses are more generally modeled by $r \rightarrow 1$ edges for any $r$. Therefore, it seems natural to ask about the extremal numbers for graphs with two $(r \rightarrow 1)$-edges. If we look at every $(r \rightarrow 1)$-graph with exactly two edges we see that these fall into four main types of graph. Let $i$ be the number of vertices that are in the tail set of both edges. Then let $I_r(i)$ denote the graph where both edges point to the same head vertex, let $H_r(i)$ denote the graph where the edges point to different head vertices neither of which are in the tail set of the other, let $R_r(i)$ denote the graph where the first edge points to a head vertex in the tail set of the second edge and the second edge points to a head not in the tail set of the first edge, and let $E_r(i)$ denote the graph where both edges point to heads in the tail sets of each other. This extends the notation used in this paper. The degenerate cases here would be $I_r(i)$ and $H_r(i)$. It would be interesting to find the extremal numbers for these graphs in general. To what extent do the current proofs extend to these graphs? For example, in the standard version of the problem it can easily be seen that \[\text{ex}(n,I_r(0)) = n{n-2 \choose r-1}\] using the same idea as we did for $I_0$. Will the other ideas generalize as well?
2b9759a1c56a44ff4776ae40175fecde3930f8aa
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction and Motivations}\label{sec:intro} The aim of the paper is to study a new notion of two-scale convergence\footnote{A deep bibliographic research, shows that the idea here presented is suggested in an paper (of the late seventies and so well before the introduction of the notion of two-scale convergence) by {\tmname{Papanicolau}} and {\tmname{Varadhan}} {\cite{papanicolaou97boundary}} in the context of stochastic homogenization.} which is very natural and, in our opinion, gives a more straightforward approach to the homogenization process: while removing the bother of the admissibility of test functions {\cite{Allaire1992,lukkassen2002two}}, it nevertheless simplifies the proof of all standard compactness results which made {\tmem{classical}} two-scale convergence (introduced in {\cite{Nguetseng1989,Allaire1992}}) very worthy of interest. Attempts to overcome the question of admissibility of test functions arising in the definition of two-scale convergence have been the subject of various authors {\cite{cioranescu2002periodic,nechvatal2004alternative,valadier1997admissible}}. Among them, the periodic unfolding method is considered one of the most successful. The idea, as well as its nomenclature, is introduced {\cite{cioranescu2002periodic}} where the authors exploit a natural, although purely mathematical, intuition to recover two-scale convergence as a classical functional weak convergence in a suitable larger space. This recovery process is achieved by introducing the so-called unfolding operator which, roughly speaking, turns a sequence of $1$-scale functions into a sequence of $2$-scale functions. On the other hand, as it is simple to show by playing with Lebesgue differentiation theorem, the recovery process is not univocal, and many alternatives are possible. In guessing the one presented below, we did not rely on mathematical intuition only, but we found inspiration from the physics of the homogenization process. That is why we think it is important to dwell on some preliminary considerations before giving definitions, theorems and proofs. The paper is organized as follows: in Section \ref{sec:intro.newapproach} we explain the idea behind the proposed approach which will be formalized in Section \ref{sec:2NA}. In Section \ref{sec:3Compact} we establish compactness results for the new notion of two-scale convergence which play a central role in the homogenization process. In Section \ref{sec:4} we test the effectiveness of our notion of convergence on the {\guillemotleft}classical{\guillemotright} model problem in the theory of homogenization, i.e the one associate to a family of linear second-order elliptic partial differential equation with periodically oscillating coefficients. Section \ref{sec:correctorH1conv} is devoted to the so-called first-order corrector results which aim to improve the convergence of the solution gradients by adding corrector term. In Section \ref{sec:6bl} we introduce the well-known boundary layer terms which aim to compensate the fast oscillation of the family of solutions near the boundary. Eventually, in Section \ref{sec:homharmonicmaps} we test the approach on a nonlinear problem: we prove a weak two-scale compactness result for $\mathbf{S}^2$-valued stationary harmonic map, and make some remarks which point out some possible weaknesses of this alternative notion of two-scale convergence. \section{The cell averaging approach to periodic homogenization}\label{sec:intro.newapproach} \subsection{The classical two-scale convergence approach to periodic homogenization}\label{subsec:sec1classicaltwoscale} Let us focus on the classical model problem in homogenization: a linear second-order partial differential equation with periodically oscillating coefficients. Such an equations models, for example, the stationary heat conduction in a periodic composite medium {\cite{Allaire1992,Donato1999}}. We denote by $\Omega$ the material domain (a bounded open set in $\mathbb{R}^N$) and by $Y := [0, 1]^N$ the unit cell of $\mathbb{R}^N$. Denoting by $f \in L^2 (\Omega)$ the source term and enforcing a Dirichlet boundary condition for the unknown $u_{\varepsilon}$, the model equation reads as \begin{equation} - \tmop{div} (A_{\varepsilon} \nabla u_{\varepsilon}) = f \quad \text{in } \Omega, \quad u_{\varepsilon} = 0 \quad \text{on } \partial \Omega, \label{eq:model-problem} \end{equation} where, for any $\varepsilon > 0$, we have defined $A_{\varepsilon}$ by $A_{\varepsilon} (x) := A (x / \varepsilon)$, with $A$ (the so-called {\tmem{matrix of diffusion coefficients}}) an $L^{\infty}$ and $Y$-periodic matrix valued function, which is uniformly coercive, i.e. such that for two positive constants $0 < \alpha \leqslant \beta$ one has (for a.e. $y \in Y$) $\alpha | \xi |^2 \leqslant A (y) \xi \cdot \xi \leqslant \beta | \xi |^2$ for every $\xi \in \mathbb{R}^N$. Here we have supposed $A$ depending on the periodic variable only although later we will work with the more general case in which $A$ depends on the $x$ variable too. The weak formulation of problem (\ref{eq:model-problem}) reads as: \begin{equation} \int_{\Omega} A_{\varepsilon} \nabla u_{\varepsilon} \cdot \nabla \varphi = \int_{\Omega} f \varphi, \label{eq:model-problemwf} \end{equation} and according to Lax-Milgram theorem for each $\varepsilon > 0$ there exists a unique weak solution $u_{\varepsilon} \in H_0^1 (\Omega)$ of (\ref{eq:model-problemwf}). The family of solutions $(u_{\varepsilon})_{_{\varepsilon \in \mathbb{R}^+}}$ and the family of {\tmem{fluxes}} $(\xi_{\varepsilon})_{\varepsilon \in \mathbb{R}^+} := (A_{\varepsilon} \nabla u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$, constitute bounded subsets respectively of $H_0^1 (\Omega)$ and $L^2 (\Omega)$. Thus there exist subfamilies (that we still denote by $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ and $(\xi_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$) and elements $u_0 \in H_0^1 (\Omega), \xi_0 \in L^2 (\Omega)$ such that $\nabla u_{\varepsilon} \rightharpoonup \nabla u_0$ and $\xi_{\varepsilon} \rightharpoonup \xi_0$ weakly in $L^2 (\Omega)$. Hence, passing to the limit in (\ref{eq:model-problemwf}), we get $(\xi_0, \nabla \varphi)_{L^2 (\Omega)} = (f, \varphi)_{L^2 (\Omega)}$, where the limit flux $\xi_0$ is the weak limit of the product of the weakly convergent sequences $\nabla u_{\varepsilon} \rightharpoonup \nabla u_0$ and $A_{\varepsilon} \rightharpoonup \langle A \rangle_{Y}$. The identification of the limit flux $\xi_0$ in terms of $u_0$ and $A$ is the first aim in the mathematical theory of periodic homogenization. A procedure for the homogenization of problem (\ref{eq:model-problem}) appeared in 1989 by the means of the so-called two-scale convergence. This notion, introduced for the first time by {\tmname{Nguetseng}} in {\cite{Nguetseng1989}}, was later named {\guillemotleft}two-scale convergence{\guillemotright} by {\tmname{Allaire}} {\cite{Allaire1992}} who further developed the notion by giving more direct proofs of the main compactness results. To better understand the idea behind the classical two-scale approach, let us recall the following compactness results {\cite{Allaire1992}}, from which the notion of two-scale convergence originates: \begin{proposition}[Nguetseng {\cite{Nguetseng1989}}, Allaire {\cite{Allaire1992}}] If $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ is a bounded sequence in $L^2 (\Omega)$, there exists $u_0 \in L^2 \left( \Omega \times Y \right)$, such that, up to a subsequence \begin{equation} \lim_{\varepsilon \rightarrow 0} \int_{\Omega} u_{\varepsilon} (x) \varphi (x, x / \varepsilon) \mathrm{d} x = \int_{\Omega \times Y} u_0 (x, y) \varphi (x, y) \mathrm{d} x \; \mathrm{d} y \end{equation} for any test function\footnote{As it is classical in the field, we index by $\sharp$ spaces that consist of periodic functions.} $\varphi \in \mathcal{D} [ \Omega, C^{\infty}_{\sharp} \left( Y \right) ]$. Moreover, if $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ is a bounded sequence in $H^1 (\Omega)$, then there exist functions $u_0 \in H^1 (\Omega)$ and $u_1 \in L^2 [ \Omega, H^1_{\sharp} \left( Y \right) / \mathbb{R} ]$ such that, up to a subsequence \begin{equation} \lim_{\varepsilon \rightarrow 0} \int_{\Omega} \nabla u_{\varepsilon} (x) \cdot \psi (x, x / \varepsilon) \mathrm{d} x = \int_{\Omega \times Y} \left( \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y) \right) \cdot \psi (x, y) \mathrm{d} x \; \mathrm{d} y \end{equation} for any test function $\psi \in \mathcal{D} [ \Omega, C^{\infty}_{\sharp} \left( Y \right) ]^N$. \end{proposition} \begin{figure}[t] \includegraphics{Fig1_2scale_New_fontvec.pdf} \caption{\label{fig:perhom1}If we assume that the heterogeneities are evenly distributed inside the media $\Omega$, we can model the material as periodic. As illustrated in the figure, this means that we can think of the material as being immersed in a grid of small identical cubes $Y_{\varepsilon}$, the side-length of which is $\varepsilon$.} \end{figure} It is then natural to give the following (see {\cite{Allaire1992}}) \begin{definition}[Allaire {\cite{Allaire1992}}] A sequence of functions $u_{\varepsilon}$ in $L^2 (\Omega)$ {\tmem{two-scale}} converges to a limit $u_0 \in L^2 \left( \Omega \times Y \right)$ if, for any function $\varphi \in \mathcal{D} [ \Omega, C^{\infty}_{\sharp} \left( Y \right) ]$ we have \begin{equation} \lim_{\varepsilon \rightarrow 0} \int_{\Omega} u_{\varepsilon} (x) \varphi (x, x / \varepsilon) \mathrm{d} x = \int_{\Omega \times Y} u_0 (x, y) \varphi (x, y) \mathrm{d} x \; \mathrm{d} y . \end{equation} In that case we write $u_{\varepsilon} \xrightarrow{2 s} u_0$. We say that the sequence $(u_{\varepsilon})$ strongly two-scale converges to a limit $u_0 \in L^2 \left( \Omega \times Y \right)$, if $u_{\varepsilon} \xrightarrow{2 s} u_0$ and $\| u_0 \|_{L^2 \left( \Omega \times Y \right)} = \lim_{\varepsilon \rightarrow 0} \| u_{\varepsilon} \|_{L^2 (\Omega)}$. \end{definition} It is now immediate to understand the role played by two-scale convergence in the homogenization process. Indeed, by writing (\ref{eq:model-problemwf}) in the form \begin{equation} \int_{\Omega} \nabla u_{\varepsilon} (x) \cdot A^{\mathsf{T}} \left( \frac{x}{\varepsilon} \right) \nabla \varphi_{\varepsilon} (x) \mathrm{d} x \; = \; \int_{\Omega} f (x) \varphi_{\varepsilon} (x) \mathrm{d} x, \end{equation} and choosing the right shape for the test functions $\varphi_{\varepsilon}$, it is possible to interpret the left-hand side of the previous relation as the product of a strongly two-scale convergent sequence (namely $A^{\mathsf{T}}_{\varepsilon} \nabla \varphi_{\varepsilon} (x)$) with the weakly two-scale convergent sequence $\nabla u_{\varepsilon}$, from which weak two-scale convergence of the product, and hence the homogenized equation, easily follows (cfr. {\cite{Allaire1992,Donato1999}} for details). Unfortunately, for this procedure to be possible it is essential to add a technical hypothesis: the sequence of coefficients $(A_{\varepsilon})$ must be {\tmem{admissible}} in the sense that (cfr. {\cite{Allaire1992}}) \begin{equation} \lim_{\varepsilon \rightarrow 0} \| A_{\varepsilon} \|_{L^2 (\Omega)} \; = \; \| A \|_{L^2 \left( \Omega \times Y \right)} . \end{equation} It turns out that this is a subtle notion. Indeed, for a given function $\psi \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$ there is no reasonable way to give a meaning to the {\guillemotleft}trace{\guillemotright} function $x \mapsto \psi (x, x / \varepsilon)$. The complete space of admissible functions is not known much more precisely. Functions in $L^p [ \Omega, C_{\sharp} \left( Y \right) ]$ as well as $L^p_{\sharp} [ Y, C (\Omega) ]$ are admissible, but it is unclear how much the regularity of $\psi$ can be weakened: we refer to {\cite{Allaire1992}} for an explicit construction of a non admissible function which belongs to $C [ \Omega, L^1_{\sharp} \left( Y \right)]$. \subsection{The cell averaging idea}\label{subsec:intro.newapproach} The {\guillemotleft}classical{\guillemotright} approach to periodic homogenization originates by the modeling assumption that since the heterogeneities are evenly distributed inside the media $\Omega$, we can think of the material as being immersed in a grid of small identical cubes $Y_{\varepsilon}$, the side-length of which is $\varepsilon$ (see Figure \ref{fig:perhom1}). If we denote by $\Omega_a := \Omega + a$, with $a \in \mathbb{R}^N$, a translated copy of $\Omega$ such that $\Omega \cap \Omega_a \neq \emptyset$, this modeling approach assumes that, at scale $\varepsilon$, the contribution of the diffusion coefficients at any $x \in \Omega \cap \Omega_a$, is given by $A (x / \varepsilon)$ both if we focus on the problem $- \tmop{div} (A_{\varepsilon} \nabla u_{\varepsilon}) = f$ in $\Omega$ and on the problem $(f_a := f (x - a))$ $- \tmop{div} (A_{\varepsilon} \nabla u_{\varepsilon}) = f_a$ in $\Omega_a$. Although this assumption is mathematically reasonable when $\varepsilon$ tends to be very small, it is nevertheless the reason why the two-scale convergence produces {\guillemotleft}two-variables{\guillemotright} functions starting from a family of {\guillemotleft}one-variable{\guillemotright} functions. \begin{figure}[t] \includegraphics{Fig2_2scale_New_fontvecjpg.pdf} \caption{\label{fig:perhom2}More realistic is to think of the material as being immersed in a grid of small identical cubes $Y_{\varepsilon}$, up to an unknown translation of size smaller than $\varepsilon$. We thus consider all possible translations, which we take into account by the introduction of a new variable.} \end{figure} On the other hand, it is clear that a more realistic approach consists in taking into account the effects of the diffusion coefficients $A_{\varepsilon} := A (x / \varepsilon)$ via a family of displacement of length at most $\varepsilon$, i.e. via the family of diffusion coefficients $\left( A_{\varepsilon} \left( \, \cdot \, + \varepsilon y \right) \right)_{(\varepsilon, y) \in \mathbb{R}^+ \times Y} = (A (y + \cdot / \varepsilon))_{(\varepsilon, y) \in \mathbb{R}^+ \times Y}$, and hence (see Figure \ref{fig:perhom2}) via the family of boundary value problems depending on the cell-size parameter $\varepsilon \in \mathbb{R}^+$ and on the translation parameter $y \in Y$. The new homogenized problem then goes through the following two steps: for every $\varepsilon \in \mathbb{R}^+$ find (in a suitable sense) a $Y$-periodic solution $u_{\varepsilon} (x, y)$ of the Dirichlet problem \begin{equation} - \tmop{div} (A_{\varepsilon} (x + \varepsilon y) \nabla u_{\varepsilon} (x, y)) = f (x) \quad \text{in } \Omega, \quad u_{\varepsilon} (x, y) = 0 \quad \text{on } \partial \Omega ; \label{eq:model-problemOmeganew} \end{equation} then take the average $\langle u_{\varepsilon} \rangle_{Y}$ as a more realistic modelization of the solution associated, at scale $\varepsilon$, to evenly distributed heterogeneities inside the media $\Omega$. In this framework the homogenization process demands for the computation of the limiting behaviour, as $\varepsilon \rightarrow 0$, of the family of two variable solutions $u_{\varepsilon} (x, y)$, i.e. for an asymptotic expansion of the form \begin{equation} u_{\varepsilon} (x, y) \; = \; u_0 \left( x, y + \frac{x}{\varepsilon} \right) + \varepsilon \; u_1 \left( x, y + \frac{x}{\varepsilon} \right) + \varepsilon^2 \; u_2 \left( x, y + \frac{x}{\varepsilon} \right) + \cdots, \label{eq:asymptexpansion} \end{equation} in which $u_0$ is the solution of the homogenized equation and $u_1$ is the so-called {\tmem{first order corrector}} (cfr. the analogues definitions in {\cite{Allaire1992,Donato1999}}). We are now in position to explain the new approach. To this end, let us introduce the operator \begin{equation} \mathcal{F}_{\varepsilon} : u \in L^2_{\sharp} [ Y, L^2 (\Omega) ] \mapsto u (x, y - x / \varepsilon) \in L^2_{\sharp} [ Y, L^2 (\Omega) ] . \label{eq:cellshiftop1} \end{equation} Due to the $Y$-periodicity of $A$, the variational formulation of (\ref{eq:model-problemOmeganew}) reads as the problem of finding $u_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ such that \begin{equation} \int_{\Omega \times Y} A (y) \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) (x, y) \cdot \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) (x, y) \mathrm{d} x \mathrm{d} y \; = \; \int_{\Omega \times Y} f (x) \mathcal{F}_{\varepsilon} (\psi_{\varepsilon}) (x, y) \mathrm{d} x \mathrm{d} y \label{eq:newweakform} \end{equation} for every $\psi_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$. Therefore, if $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \rightharpoonup \tmmathbf{v}$ weakly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$, then for every couple of {\guillemotleft}test functions{\guillemotright} $\psi, \tmmathbf{\psi} \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$ such that for some family $\psi_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ we have $\mathcal{F}_{\varepsilon} (\psi_{\varepsilon}) \rightarrow \psi$ and $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) \rightarrow \tmmathbf{\psi}$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$, passing to the limit in (\ref{eq:newweakform}), we finish with the {\guillemotleft}homogenized equation{\guillemotright} \begin{equation} \int_{\Omega \times Y} A (y) \tmmathbf{v} (x, y) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \; = \; \int_{\Omega \times Y} f (x) \psi (x, y) \mathrm{d} x \mathrm{d} y. \end{equation} Of course, to find an explicit expression for the homogenized equation, and more generally to build a kind of two-scale calculus, it is important to investigate the interconnections between the convergence of the families $u_{\varepsilon}$ and $\mathcal{F}_{\varepsilon} (u_{\varepsilon})$ in $L^2_{\sharp} [ Y, H^1 (\Omega) ]$, and to understand which are the subspaces of $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ which are reachable by strong convergence of family of the type $\mathcal{F}_{\varepsilon} (\varphi_{\varepsilon})$ in $L^2_{\sharp} [ Y, H^1 (\Omega) ]$. This and many other important aspects of the question are the object of the next two sections. \section{The alternative approach to two-scale convergence}\label{sec:2NA} \subsection{Notation and preliminary definitions} In what follows we denote by $Y = [0, 1]^N$ the unit cell of $\mathbb{R}^N$ and by $\Omega$ an open set of $\mathbb{R}^N$. For any measurable function $u$ defined on $Y$ we denote by $\langle u \rangle_{Y}$ the integral average of $u$. By $C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]$ we mean the vector space of test functions $u : \Omega \times \mathbb{R}^N \rightarrow \mathbb{R}$ such that the section $u (x, \cdot) \in C^{\infty}_{\sharp} \left( Y \right)$ for every $x \in \Omega$, and the section $u (\cdot, y) \in \mathcal{D} (\Omega)$ for every $y \in \mathbb{R}^N$. Similarly we denote by $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ the Hilbert space of $Y$-periodic distributions which are in $L^2 \left( \Omega \times Y \right)$, and by $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ the Hilbert subspace of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ constituted of distributions $u$ such that $\nabla_{\hspace{-1pt} x} u \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$. Next, we denote by $L^2 [ \Omega ; H^1_{\sharp} \left( Y \right) ]$ the Hilbert space of $Y$-periodic distributions $u \in \mathcal{D}' \left( \Omega \times \mathbb{R}^N \right)$ such that $u (\cdot, y) \in L^2 (\Omega)$ for a.e. $y \in Y$ and $u (x, \cdot) \in H^1_{\tmop{loc}} \left( \mathbb{R}^N \right)$ for a.e. $x \in \Omega$. Finally, in the next Proposition \ref{prop:isometricFeps}, we denote by $\mathcal{E}'_{\sharp} [ Y, \mathcal{D}' (\Omega) ]$ the algebraic dual of $C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]$, and for any $u \in \mathcal{E}'_{\sharp} [ Y, \mathcal{D}' (\Omega) ]$ and any $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$ we define the partial gradient $\nabla_{\hspace{-1pt} x} u$ by the position $\left\langle \nabla_{\hspace{-1pt} x} u, \tmmathbf{\psi} \right\rangle := - \left\langle u, \tmop{div}_{\hspace{-1pt} x} \tmmathbf{\psi} \right\rangle$ and the $\varepsilon$-cell shifting of $u$ by the position $\langle u (x, y - x / \varepsilon), \tmmathbf{\psi} (x, y) \rangle := \langle u (x, y), \tmmathbf{\psi} (x, y + x / \varepsilon) \rangle$. \subsection{Cell averaging two-scale convergence} Motivated by the considerations made in subsection \ref{subsec:intro.newapproach} we give the following \begin{definition} Let $\Omega \subseteq \mathbb{R}^N$ be an open set and $Y$ the unit cell of $\mathbb{R}^N$. For any $\varepsilon > 0$, we define the $\varepsilon$-{\mybold{cell shift operator}} $\mathcal{F}_{\varepsilon}$ by the position \begin{equation} u \in L^2_{\sharp} [ Y, L^2 (\Omega) ] \mapsto \mathcal{F}_{\varepsilon} (u) := u (x, y - x / \varepsilon) \in L^2_{\sharp} [ Y, L^2 (\Omega) ], \end{equation} i.e. as the composition of $u$ with the diffeomorphism $(x, y) \in \Omega \times \mathbb{R}^N \mapsto (x, y - x / \varepsilon) \in \Omega \times \mathbb{R}^N$. We then denote by $\mathcal{F}^{\ast}_{\varepsilon}$ the algebraic adjoint operator which maps $u (x, y)$ to $u (x, y + x / \varepsilon)$. \end{definition} \begin{definition} A sequence of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ functions $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ is said to weakly two-scale converges to a function $u_0 \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$, if $\mathcal{F}_{\varepsilon} (u_{\varepsilon}) \rightharpoonup u_0$ weakly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$, i.e if and only if \begin{equation} \lim_{\varepsilon \rightarrow 0^+} \int_{\Omega \times Y} u_{\varepsilon} \left( x, y - \frac{x}{\varepsilon} \right) \psi (x, y) \mathrm{d} x \mathrm{d} y = \int_{\Omega \times Y} u_0 (x, y) \psi (x, y) \mathrm{d} x \mathrm{d} y, \end{equation} for every $\psi \in L^2_{\#} [ Y, L^2 (\Omega) ]$. In that case we write $u_{\varepsilon} \twoheadrightarrow u_0$ weakly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. We say that $u_{\varepsilon} \twoheadrightarrow u_0$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ if $\mathcal{F}_{\varepsilon} (u_{\varepsilon}) \rightarrow u_0$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. \end{definition} \begin{remark} We have stated the definition in the framework of square summable functions. Nevertheless, almost all of what we say here and hereinafter easily extends, with obvious modifications, to the setting of $L^p$ spaces. \end{remark} \begin{remark} Since the notion of two-scale convergence relies on the classical notion of weak convergence in Banach space, we immediately get, among others, boundedness in norm of weakly two-scale convergent sequences. This aspect is not captured by the classical notion of two-scale convergence which, by testing convergence on functions in $\mathcal{D} [ \Omega, C^{\infty}_{\sharp} \left( Y \right) ]$, i.e. having compact support in $\Omega$, may cause loss of information on any concentration of {\guillemotleft}mass{\guillemotright} near the boundary of the sequence $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ (cfr. {\cite{lukkassen2002two}}). \end{remark} We now state some properties of the operator $\mathcal{F}_{\varepsilon}$, which are simple consequence of the definitions, and will be used extensively (and sometime tacitly) in the sequel: \begin{proposition} \label{prop:isometricFeps}Let $\varepsilon > 0$. The operator $\mathcal{F}_{\varepsilon}$ is an isometric isomorphism of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ and the following relations hold: \begin{itemizedot} \item If $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$ then $\mathcal{F}_{\varepsilon} (\tmmathbf{\psi}) \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$ and one has \begin{equation} \tmop{div}_{\hspace{-1pt} x} \mathcal{F}_{\varepsilon} (\tmmathbf{\psi}) = \mathcal{F}_{\varepsilon} \left( \tmop{div}_{\hspace{-1pt} x} \tmmathbf{\psi} \right) - \frac{1}{\varepsilon} \mathcal{F}_{\varepsilon} \left( \tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi} \right), \quad \tmop{div}_{\hspace{-1pt} y} \mathcal{F}_{\varepsilon} (\tmmathbf{\psi}) =\mathcal{F}_{\varepsilon} \left( \tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi} \right) . \label{eqpropdiv2} \end{equation} \end{itemizedot} Next, let us denote by $\mathcal{E}'_{\sharp} [ Y, \mathcal{D}' (\Omega) ]$ the algebraic dual of $C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]$: \begin{itemizedot} \item If $u \in \mathcal{E}'_{\sharp} [ Y, \mathcal{D}' (\Omega) ]$ then $\mathcal{F}_{\varepsilon} (u) \in \mathcal{E}'_{\sharp} [ Y, \mathcal{D}' (\Omega) ]$ and one has \begin{equation} \left\langle \nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (u)], \tmmathbf{\psi} \right\rangle = \left\langle \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u \right) - \frac{1}{\varepsilon} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} y} u \right), \tmmathbf{\psi} \right\rangle, \quad \left\langle \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} y} u \right), \tmmathbf{\psi} \right\rangle = \left\langle \nabla_{\hspace{-1pt} y} [\mathcal{F}_{\varepsilon} (u)], \tmmathbf{\psi} \right\rangle, \label{eq:gradandFeps} \end{equation} for any $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$. \end{itemizedot} \end{proposition} \begin{proof} For every $u \in L^2_{\#} [ Y, L^2 (\Omega) ]$, by the translational invariance of the integral over $Y$ with respect to the section $u (x, \cdot) \in L^2 \left( Y \right)$, we get \begin{equation} \| \mathcal{F}_{\varepsilon} (u) \|_{L^2 \left( \Omega \times Y \right)} \; = \; \left( \int_{\Omega \times Y} | u (x, y - x / \varepsilon) |^2 \right)^{1 / 2} \; = \; \| u \|_{L^2 \left( \Omega \times Y \right)} . \end{equation} Relation (\ref{eqpropdiv2}) is a standard computation. Equation (\ref{eq:gradandFeps}) is a direct consequence of (\ref{eqpropdiv2}). Indeed for any $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$ we have \begin{align} \left\langle \nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (u)], \tmmathbf{\psi} \right\rangle & := - \left\langle u, \; \mathcal{F}_{\varepsilon}^{\ast} \left( \tmop{div}_{\hspace{-1pt} x} \tmmathbf{\psi} \right) \right\rangle \\ &= - \left\langle u, \; \tmop{div}_{\hspace{-1pt} x} \mathcal{F}_{\varepsilon}^{\ast} (\tmmathbf{\psi}) - \frac{1}{\varepsilon} \tmop{div}_{\hspace{-1pt} y} \mathcal{F}_{\varepsilon}^{\ast} (\tmmathbf{\psi}) \right\rangle, \end{align} and this last expression is nothing else than (\ref{eq:gradandFeps}). \end{proof} \section{Compactness results}\label{sec:3Compact} As already pointed out, one of the greatest strengths of the new notion of two-scale convergence is in the simplification we gain in proving compactness results for that notion. In that regard it is important to remark that one of the main contributions given by {\tmname{Allaire}} in {\cite{Allaire1992}} was to give a concise proof of the nowadays classical compactness results associated to two-scale convergence, by the means of Banach-Alaoglu theorem and Riesz representation theorem for Radon measures (cfr. Theorem 1.2 in {\cite{Allaire1992}}). \subsection{Compactness in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$} As as previously announced, the proof of the following compactness result is completely straightforward (cfr. Theorem 1.2 in {\cite{Allaire1992}}). \begin{theorem} \label{Thm:compL2}From every bounded subset $(u_{\varepsilon})_{\varepsilon > 0}$ of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ is possible to extract a weakly two-scale convergent sequence. \end{theorem} \begin{proof} According to Proposition \ref{prop:isometricFeps}, $\mathcal{F}_{\varepsilon}$ is an isometric isomorphism of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ in it, and therefore also $(\mathcal{F}_{\varepsilon} (u_{\varepsilon}))_{\varepsilon > 0}$ is a bounded subset of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. Therefore there exists an $u_0 \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$ and a subsequence extracted from $(u_{\varepsilon})$, still denoted by $(u_{\varepsilon})$, such that $\mathcal{F}_{\varepsilon} (u_{\varepsilon}) \rightharpoonup u_0$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$, i.e. such that $u_{\varepsilon} \twoheadrightarrow u_0$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. \end{proof} \subsection{Compactness in $L^2_{\sharp} [ Y, H^1 (\Omega) ]$} The following compactness results are the counterparts of the well-known corresponding results for the classical notion two-scale convergence (cfr. Proposition 1.14 in {\cite{Allaire1992}}). \begin{proposition} \label{prop:g1classicalctwoscaleL2H1}Let $(u_{\varepsilon})$ be a sequence in $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ such that for some $(u_0, \tmmathbf{v}) \in L^2_{\sharp} [ Y, L^2 (\Omega) ]^{N + 1}$ one has \begin{equation} u_{\varepsilon} \twoheadrightarrow u_0 \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ] \quad, \quad \nabla_{\hspace{-1pt} x} u_{\varepsilon} \twoheadrightarrow \tmmathbf{v} \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ]^N, \end{equation} then $u_0 (x, y) = \langle u_0 (x, \cdot) \rangle_{Y}$, i.e. the two-scale limit $u_0$ does not depends on the $y$ variable. Moreover there exists an element $u_1 \in L^2 [ \Omega ; H^1_{\sharp} \left( Y \right) ]$ such that $\tmmathbf{v}= \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1$. \end{proposition} \begin{proof} The relation $\nabla_{\hspace{-1pt} x} u_{\varepsilon} \twoheadrightarrow \tmmathbf{v}$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$ means, in particular, that for $\varepsilon \rightarrow 0$ one has $\left\langle \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right), \tmmathbf{\psi} \right\rangle \rightarrow \langle \tmmathbf{v}, \tmmathbf{\psi} \rangle$ for any $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$. Moreover, from (\ref{eqpropdiv2}) we get \begin{eqnarray} \int_{\Omega \times Y} \nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (u)] (x, y) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y & \; = \; & - \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} (u_{\varepsilon}) (x, y) \tmop{div}_{\hspace{-1pt} x} \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \nonumber\\ & & \qquad - \frac{1}{\varepsilon} \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} (u_{\varepsilon}) (x, y) \tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \label{eq:templemmaindipendencefromy0L2H10}\\ & \; = \; & \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u \right) (x, y) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \nonumber\\ & & \qquad - \frac{1}{\varepsilon} \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} y} u \right) (x, y) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y, \label{eq:templemmaindipendencefromy0L2H1} \end{eqnarray} for any $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$. Let us investigate the implications of (\ref{eq:templemmaindipendencefromy0L2H10}) and (\ref{eq:templemmaindipendencefromy0L2H1}). Since $\mathcal{F}_{\varepsilon} (u_{\varepsilon}) \rightharpoonup u_0$ and $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \rightharpoonup \tmmathbf{v}$, multiplying both members of relation (\ref{eq:templemmaindipendencefromy0L2H10}) by $\varepsilon$ and then letting $\varepsilon \rightarrow 0$ we get \begin{equation} \int_{\Omega \times Y} u_0 (x, y) \tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \; = \; 0 \quad \forall \tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N, \end{equation} from which the independence of the two-scale limit $u_0$ from the $y$ variable follows. Thus for the limit function we have $u_0 (x, y) = \langle u_0 (x, \cdot) \rangle_{Y}$ for every $y \in Y$. On the other hand, from (\ref{eq:templemmaindipendencefromy0L2H1}), for every $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$ such that $\tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi}= 0$ we have \begin{equation} \int_{\Omega \times Y} \left( \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) (x, y) - \nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (u_{\varepsilon})] (x, y)_{_{_{_{_{}}}}} \right) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \; = \; 0. \end{equation} Since $\mathcal{F}_{\varepsilon} (u_{\varepsilon}) \rightharpoonup u_0$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ one has $\nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (u_{\varepsilon})] \rightarrow \nabla_{\hspace{-1pt} x} u_0$ in the sense of distribution; thus multiplying both members of the previous relation by $\varepsilon$ and then letting $\varepsilon \rightarrow 0$ we get (by hypothesis $\nabla_{\hspace{-1pt} x} u_{\varepsilon} \twoheadrightarrow \tmmathbf{v}$) \begin{equation} \int_{\Omega \times Y} \left( \tmmathbf{v} (x, y) - \nabla_{\hspace{-1pt} x} u_0 (x, y) \right) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \; = \; 0, \end{equation} for every $\tmmathbf{\psi} \in C^{\infty}_{\sharp} [ Y, \mathcal{D} (\Omega) ]^N$ such that $\tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi}= 0$. According to De\,Rham's theorem, which in our context can be easily proved by means of Fourier series on $Y$ (see e.g.$\;${\cite{Jikov1994}} p.6), the orthogonal complement of divergence-free functions are exactly the gradients, and therefore there exists a $u_1 \in L^2 [ \Omega ; H^1_{\sharp} \left( Y \right) ]$ such that $\nabla_{\hspace{-1pt} y} u_1 =\tmmathbf{v}- \nabla_{\hspace{-1pt} x} u_0$. This concludes the proof. \end{proof} \begin{proposition} \label{prop:g1classicalctwoscaleL2H1eps}Let $(u_{\varepsilon})$ be a sequence in $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ such that for some $(u_0, \tmmathbf{v}) \in [ L^2 \left( \Omega \times Y \right) ]^{N + 1}$ one has \begin{eqnarray} u_{\varepsilon} \twoheadrightarrow u_0 \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ] & \lo{\tmop{and}} & \varepsilon \nabla_{\hspace{-1pt} x} u_{\varepsilon} \twoheadrightarrow \tmmathbf{v} \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ]^N, \end{eqnarray} then $\tmmathbf{v}= \nabla_{\hspace{-1pt} y} u_0$. \end{proposition} \begin{proof} As in the proof of Proposition \ref{prop:g1classicalctwoscaleL2H1} we have: \begin{eqnarray} \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \varepsilon \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) (x, y) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y & \; = \; & - \varepsilon \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} (u_{\varepsilon}) (x, y) \tmop{div}_{\hspace{-1pt} x} \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \nonumber\\ & & \qquad - \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} (u_{\varepsilon}) (x, y) \tmop{div}_{\hspace{-1pt} y} \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y. \label{eq:templemmaindipendencefromy0L2H10eps} \end{eqnarray} Let us investigate the implications of (\ref{eq:templemmaindipendencefromy0L2H10eps}). Since $\mathcal{F}_{\varepsilon} \left( \varepsilon \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \rightharpoonup \tmmathbf{v}$ in $[ L^2 \left( \Omega \times Y \right) ]^N$ one has that $\nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (u_{\varepsilon})] \rightharpoonup \nabla_{\hspace{-1pt} x} u_0$ in $[ \mathcal{D}' \left( \Omega \times Y \right) ]^N$. Then taking the limit for $\varepsilon \rightarrow 0$ in relation (\ref{eq:templemmaindipendencefromy0L2H10eps}) and integrating by parts, we get $\left\langle \tmmathbf{v}- \nabla_{\hspace{-1pt} y} u_0, \tmmathbf{\psi} \right\rangle = 0$ in $\mathcal{D}' \left( \Omega \times Y \right)$ and therefore $\tmmathbf{v}= \nabla_{\hspace{-1pt} y} u_0$. \end{proof} \subsection{Test functions reachable by strong two-scale convergence} As pointed out at the end of subsection \ref{subsec:intro.newapproach}, in order to identify the system of homogenized equations it is important to understand the subspaces of $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ which are reachable by strong convergence in $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ (cfr. Lemma 1.13 in {\cite{Allaire1992}}). Although this question become a simple observation in our framework, we will make constantly use of the following result which therefore state as a proposition in order to reference it when used. \begin{proposition} \label{prop:attainedtest}The following statements hold: \begin{enumeratenumeric} \item For every $\varphi \in \mathcal{D} (\Omega)$ there exists a sequence of functions $(\varphi_{\varepsilon})_{\varepsilon > 0}$ of $L^2_{\sharp} [ Y, H^1 (\Omega) ]$ such that $\mathcal{F}_{\varepsilon} (\varphi_{\varepsilon}) = \varphi$ and $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \varphi_{\varepsilon} \right) = \nabla_{\hspace{-1pt} x} \varphi$ for every $\varepsilon > 0$, so that obviously $\varphi_{\varepsilon} \twoheadrightarrow \varphi$ strongly $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ and $\nabla_{\hspace{-1pt} x} \varphi_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} x} \varphi$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$. \item Similarly, for every $\psi \in \mathcal{D} \left( \Omega \times Y \right)$ there exists a sequence of functions $(\psi_{\varepsilon}) \in L^2_{\sharp} [ Y, H^1 (\Omega) ]$ such that $\mathcal{F}_{\varepsilon} (\psi_{\varepsilon}) = \psi$ and $\varepsilon \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) = \nabla_{\hspace{-1pt} y} \psi + \varepsilon \nabla_{\hspace{-1pt} x} \psi$ for every $\varepsilon > 0$. In particular $\psi_{\varepsilon} \twoheadrightarrow \psi$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ and $\varepsilon \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} y} \psi$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$. \end{enumeratenumeric} \end{proposition} \begin{proof} For every $\varphi \in \mathcal{D} (\Omega)$ the constant family of functions defined by the position $\varphi_{\varepsilon} (x, y) := \varphi (x) \otimes 1 (y)$ is in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$, and is such that $\mathcal{F}_{\varepsilon} (\varphi_{\varepsilon}) = \varphi$. Therefore $\mathcal{F}_{\varepsilon} (\varphi_{\varepsilon})$ strongly converges to $\varphi$ in $L^2 \left( \Omega \times Y \right)$ \ and \ $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \varphi_{\varepsilon} \right) = \nabla_{\hspace{-1pt} x} \varphi$ strongly converges to $\nabla_{\hspace{-1pt} x} \varphi$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$. For the second part of the statement we note that for every $\psi \in \mathcal{D} \left( \Omega \times Y \right)$ the family $\psi_{\varepsilon} (x, y) := \psi (x, y + x / \varepsilon)$ is in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$, and is such that $\mathcal{F}_{\varepsilon} (\psi_{\varepsilon}) = \psi$. Hence $\psi_{\varepsilon} \twoheadrightarrow \psi_0$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. Moreover $\varepsilon \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) = \nabla_{\hspace{-1pt} y} [\mathcal{F}_{\varepsilon} (\psi_{\varepsilon})] + \varepsilon \nabla_{\hspace{-1pt} x} [\mathcal{F}_{\varepsilon} (\psi_{\varepsilon})] = \nabla_{\hspace{-1pt} y} \psi + \varepsilon \nabla_{\hspace{-1pt} x} \psi$ so that $\varepsilon \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} y} \psi$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. \end{proof} \section{The {\guillemotleft}classical{\guillemotright} homogenization problem}\label{sec:4} In the mathematical literature, the elliptic equation introduced in subsection \ref{subsec:sec1classicaltwoscale}, Eq.$\;$(\ref{eq:model-problem}), it is nowadays simply referred to as the classical homogenization problem. This classical problem has achieved the role of {\guillemotleft}benchmark problem{\guillemotright} for new methods in periodic homogenization: Whenever a new method for periodic homogenization emerges, it is customary to test it by the ease it allows to solve the classical homogenization problem. This is exactly the aim of this section. Of course, as pointed out in subsection \ref{subsec:intro.newapproach}, our testing problem is slightly different as the matrix of diffusion coefficients is now a function depending on a parameter. Nevertheless, and this is a really important point, the homogenized equations we get are exactly the ones arising from the homogenization of the classical homogenization problem. \subsection{The {\guillemotleft}classical{\guillemotright} homogenization problem} Let $\Omega$ be a bounded open set of $\mathbb{R}^N$. Let $f$ be a given function in $L^2 (\Omega)$. For every $y \in Y$ we consider the following linear second-order elliptic equation \begin{eqnarray} - \tmop{div}_{\hspace{-1pt} x} [ A (x, y + x / \varepsilon) \nabla_{\hspace{-1pt} x} u_{\varepsilon} (x, y) ] & = \hspace{1.2em} f (x) & \hspace{1.2em} \text{in } \Omega \label{eq:homogproblem2dot0feq}\\ u_{\varepsilon} (x, y) & = \hspace{1.2em} 0 & \hspace{1.2em} \text{on } \partial \Omega, \label{eq:homogproblem2dot0} \end{eqnarray} where $A \in [ L^{\infty} \left( \Omega \times \mathbb{R}^N \right) ]^{N^2}$ is a (not necessarily symmetric) matrix valued function defined on $\Omega \times Y$ and $Y$-periodic in the second variable. We also suppose $A$ to be {\mybold{uniformly elliptic}}, i.e. there exists a positive constants $\alpha > 0$ such that $\alpha | \xi |^2 \leqslant A (x, y) \xi \cdot \xi$ for any $\xi \in \mathbb{R}^N$ and every $(x, y) \in \Omega \times Y$. Following {\cite{Allaire1992}} we give the following \begin{definition} \label{def:classhomandcorr}The homogenized equation is defined as \begin{eqnarray} - \tmop{div} [A_{\hom} (x) \nabla u_0 (x)] & = \; \; f (x) & \text{ in } \Omega \label{eq:homogenizedpdej1}\\ u (x) & = \hspace{1.2em} 0 & \text{ on } \partial \Omega \end{eqnarray} where the matrix $A_{\hom}$ is given by \begin{equation} A_{\hom} = \left\langle A (x, \cdot) \left( I_N + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} (x, \cdot) \right) \right\rangle_{Y}, \label{eq:Ahom0} \end{equation} where $\tmmathbf{\chi} := (\chi_1, \chi_2, \ldots, \chi_N)$ is the so-called vector of correctors where for every $i \in \mathbb{N}_N$ the function $\chi_i$ is the unique solution in the space $L^{\infty} [ \Omega, H^1_{\sharp} \left( Y \right) / \mathbb{R} ]$ of the cell problem: \begin{equation} - \tmop{div}_{\hspace{-1pt} y} [ A (x, y) \left( \nabla_{\hspace{-1pt} y} \chi_i (x, y) + e_i \right) ] \; = \; 0. \label{eq:splittedsoldistribcell} \end{equation} \end{definition} We then have \begin{theorem} \label{thm:newhomapproach}For every $\varepsilon \in \mathbb{R}^+$ there exists a unique solution $u_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ of the problem (\ref{eq:homogproblem2dot0feq})-(\ref{eq:homogproblem2dot0}). \begin{enumerate} \item The sequence $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ of $L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ solutions is such that \begin{equation} u_{\varepsilon} \twoheadrightarrow u_0 \quad, \quad \nabla_{\hspace{-1pt} x} u_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ] \end{equation} where $(u_0, u_1)$ is the unique solution in $H_0^1 (\Omega) \times L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$ of the following two-scale homogenized system: \begin{eqnarray} - \tmop{div}_{\hspace{-1pt} y} [ A (x, y) \left( \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y) \right) ] & = \; \; 0 & \text{in } \Omega \times Y, \label{eq:homeq1distrib}\\ - \tmop{div}_{\hspace{-1pt} x} [ \int_{Y} A (x, y) \left( \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y) \right) \mathrm{d} y ] & = \hspace{1.2em} f (x) & \text{in } \Omega \times Y . \label{eq:homeq2} \end{eqnarray} \item Furthermore, the previous system in equivalent to the classical homogenized and cell equations through the relation \begin{equation} u_1 (x, y) = \nabla u_0 (x) \cdot \tmmathbf{\chi} (x, y) . \label{eq:linkclassicaland2scale} \end{equation} \end{enumerate} \end{theorem} \begin{proof} {\tmstrong{1)}} We write the weak formulation of problem (\ref{eq:homogproblem2dot0feq})-(\ref{eq:homogproblem2dot0}) on the space $L^2_{\sharp} [ Y ; H^1_0 (\Omega) ]$: \begin{equation} \int_{Y \times \Omega} A (x, x / \varepsilon + y) \nabla_{\hspace{-1pt} x} u_{\varepsilon} (x, y) \cdot \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y \; = \; \int_{Y \times \Omega} f (x) \psi_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} z, \label{eq:twovarweakform} \end{equation} with $\psi_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$. Once endowed the space $L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ with the equivalent norm $u \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ] \mapsto \left\| \nabla_{\hspace{-1pt} x} u \right\|^2_{\Omega \times Y}$, due to Lax-Milgram theorem, for every $\varepsilon > 0$ there exists a unique solution $u_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ and moreover \begin{equation} \left\| \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right\|_{L^2 \left( \Omega \times Y \right)} \leqslant \frac{c_{\Omega}}{\alpha} \| f \|_{L^2 (\Omega)} \label{eq:stabilityestimateLM} \end{equation} where we have denote by $c_{\Omega}$ the Poincar{\'e} constant for the space $H_0^1 (\Omega)$. As a consequence of the uniform bound (with respect to $\varepsilon$) expressed by (\ref{eq:stabilityestimateLM}), taking into thanks to the reflexivity of the space $L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ and Proposition \ref{prop:g1classicalctwoscaleL2H1}, there exists a subsequence extracted from $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$, and still denoted by $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$, such that \begin{equation} u_{\varepsilon} \twoheadrightarrow u_0 \quad, \quad \nabla_{\hspace{-1pt} x} u_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ], \end{equation} for a suitable $u_0 \in H^1_0 (\Omega)$ and $u_1 \in L^2 [\Omega, H^1_{\sharp} (Y)]$. Next we note that in terms of the operator $\mathcal{F}_{\varepsilon}$, the previous equation (\ref{eq:twovarweakform}) reads as \begin{eqnarray} \int_{Y \times \Omega} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) (x, y) \cdot A^{\mathsf{T}} (x, y) \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) (x, y) \mathrm{d} x \mathrm{d} y & = & \int_{Y \times \Omega} f (x) \psi_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y \nonumber\\ & = & \int_{Y \times \Omega} f (x) \mathcal{F}_{\varepsilon} (\psi_{\varepsilon}) (x, y) \mathrm{d} x \mathrm{d} y . \label{eq:twovarweakformFeps} \end{eqnarray} Now, we already know that $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \rightharpoonup \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. We then observe that (cfr. Proposition \ref{prop:attainedtest}) for every $\varphi \in \mathcal{D} (\Omega)$, there exists a sequence $\psi_{\varepsilon}$ of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ functions such that $\mathcal{F}_{\varepsilon} (\psi_{\varepsilon}) \rightarrow \varphi$ and $\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) \rightarrow \nabla_{\hspace{-1pt} x} \varphi$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]$. Therefore passing to the limit for $\varepsilon \rightarrow 0$ in equation (\ref{eq:twovarweakformFeps}), we get \begin{equation} \int_{Y \times \Omega} A (x, y) \left( \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y)_{_{_{_{_{}}}}} \right) \cdot \nabla_{\hspace{-1pt} x} \varphi (x) \mathrm{d} x \mathrm{d} y \; = \; \int_{\Omega} f (x) \varphi (x) \mathrm{d} x, \label{eq:weakformtemp1} \end{equation} which, due to the arbitrariness of $\varphi \in \mathcal{D} (\Omega)$, in distributional form reads as (\ref{eq:homeq2}). On the other hand, for every $\psi \in \mathcal{D} \left( \Omega \times Y \right)$ there exists (cfr. Proposition \ \ref{prop:attainedtest}) a family $(\psi_{\varepsilon})$ of $L^2_{\sharp} [ Y, L^2 (\Omega) ]$ functions such that $\varepsilon \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} \psi_{\varepsilon} \right) \rightarrow \nabla_{\hspace{-1pt} y} \psi$ strongly in $L^2 \left( \Omega \times Y \right)$ so that, multiplying both members of (\ref{eq:twovarweakformFeps}) for $\varepsilon > 0$ and passing to the limit for $\varepsilon \rightarrow 0$ we get \begin{equation} \int_{Y \times \Omega} A (x, y) \left( \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y) \right) \cdot \nabla_{\hspace{-1pt} y} \psi (x, y) \mathrm{d} x \mathrm{d} y = 0 \label{eq:homeq1} \end{equation} which, due to the arbitrariness of $\psi \in \mathcal{D} (\Omega)$, in distributional form reads as (\ref{eq:homeq1distrib}). We have thus proved that from any extracted subsequence from $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ it is possible to extract a further subsequence which two-scale convergence to the solution of the system of equations (\ref{eq:weakformtemp1}),(\ref{eq:homeq1}). Since the system of equations (\ref{eq:weakformtemp1}),(\ref{eq:homeq1}) has only one solution $(u_0, u_1) \in H_0^1 (\Omega) \times L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$, as it is immediate to check via Lax-Milgram theorem, the entire sequence $(u_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ two-scale convergence to $u_0$. \end{proof} \begin{proof} {\tmstrong{2)}} The homogenization process has led to two partial differential equations, namely (\ref{eq:homeq1distrib}) and (\ref{eq:homeq2}). Let us observe that the distributional equation (\ref{eq:homeq1distrib}) can be equivalently written as \begin{equation} - \tmop{div}_{\hspace{-1pt} y} [ A (x, y) \left. \nabla_{\hspace{-1pt} y} u_1 (x, y) \right) ] \; = \; \tmop{div}_{\hspace{-1pt} y} A (x, y) \cdot \nabla u_0 (x), \label{eq:cellvect} \end{equation} where we have denoted by $\tmop{div}_{\hspace{-1pt} y} A = \left( \tmop{div}_{\hspace{-1pt} y} A_1, \; \tmop{div}_{\hspace{-1pt} y} A_2, \ldots, \, \tmop{div}_{\hspace{-1pt} y} A_N \right)$ the vector whose components are the $\tmop{div}_{\hspace{-1pt} y}$of the columns $A_1, A_2, \ldots, A_N$ of $A$. It is completely standard (see {\cite{papanicolau1978asymptotic}}) to show that there exist a unique solution $u_1 \in L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$ of the cell problem (\ref{eq:cellvect}). Moreover, we observe that (as consequence of Lax-Milgram theorem), for every $i \in \mathbb{N}_N$ and for a.e. $x \in \Omega$ there exists a unique solution $\chi_i (x, \cdot) \in H^1_{\sharp} (Y) / \mathbb{R}$ of the distributional equation \begin{equation} - \tmop{div}_{\hspace{-1pt} y} [ A (x, y) \left. \nabla_{\hspace{-1pt} y} \chi_i (x, y) \right) ] \; = \; \tmop{div}_{\hspace{-1pt} y} A_i (x, y) \label{eq:splittedsoldistrib}, \end{equation} and the stability estimates $\| \chi_i (x, \cdot) \|_{H^1_{\sharp} \left( Y \right)} \leqslant \frac{1}{\alpha} \| A_i \|_{L^{\infty} \left( \Omega \times Y \right)}$ holds a.e. in $\Omega$. Therefore for every $i \in \mathbb{N}_N$ we have $\chi_i \in L^{\infty} [ \Omega, H^1_{\sharp} \left( Y \right) / \mathbb{R} ]$ so that the unique solution of (\ref{eq:splittedsoldistrib}) can be expressed as \begin{equation} u_1 (x, y) = \nabla_{\hspace{-1pt} x} u_0 (x) \cdot \tmmathbf{\chi} (x, y) \label{eq:explicitchi} \end{equation} with $\tmmathbf{\chi} (x, y) := (\chi_1 (x, y), \chi_2 (x, y), \ldots, \chi_N (x, y))$. After that, substituting (\ref{eq:explicitchi}) into (\ref{eq:homeq2}) we get the classical homogenized equation: \begin{eqnarray} f (x) & = & - \tmop{div}_{\hspace{-1pt} x} \left( \left\langle A (x, \cdot) \left( I_N + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} (x, \cdot) \right) \right\rangle_{Y} \nabla_{\hspace{-1pt} x} u_0 (x) \right) \nonumber\\ & = & - \tmop{div}_{\hspace{-1pt} x} \left( A_{\hom} (x) \nabla_{\hspace{-1pt} x} u_0 (x) \right), \label{eq:classicalhomequation} \end{eqnarray} with \begin{equation} A_{\hom} (x) := \int_{Y} A (x, y) \left( I_N + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} (x, y) \right) \mathrm{d} y. \label{eq:Ahom} \end{equation} Note that equation (\ref{eq:classicalhomequation}) is well-posed in $H_0^1 (\Omega)$ since it is easily seen that $A_{\hom}$ is bounded and coercive (see {\cite{papanicolau1978asymptotic}}). The proof is complete. \end{proof} \section{ Strong Convergence in $H^1 (\Omega)$: A corrector result\label{sec:correctorH1conv}} In the classical framework of two-scale convergence, the so-called corrector results aim to improve the convergence of the solution gradients $\nabla_{\hspace{-1pt} x} u_{\varepsilon}$ by adding corrector terms. A typical corrector result has the effect of transforming a weak convergence result into a strong one {\cite{Allaire1992,allaire2012shape,papanicolau1978asymptotic}}. In our context, as we shall see in a moment, the role of the corrector term is replaced by the average over the unit cell $Y$ of the family of solutions $u_{\varepsilon}$ (cfr. Theorem \ref{thm:newhomapproach} for the notations). We thus get a rigorous justification of the two first term in the asymptotic expansion (\ref{eq:asymptexpansion}) of the solution $u_{\varepsilon}$ of the homogenization problem. \begin{theorem} \label{thm:corrH1}For every $\varepsilon \in \mathbb{R}^+$ let $u_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ be the unique solution of the homogenization problem (\ref{eq:homogproblem2dot0feq})-(\ref{eq:homogproblem2dot0}), and $(u_0, u_1) \in H_0^1 (\Omega) \times L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$ the unique solution of the homogenized system of equations (\ref{eq:homeq1distrib})-(\ref{eq:homeq2}). Then for $\varepsilon \rightarrow 0$ we have \begin{equation} \left\| \langle u_{\varepsilon} \rangle_{Y} - u_0 \right\|_{H^1 (\Omega)} \rightarrow 0. \label{eq:strongconvh1} \end{equation} In particular $\left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right\rangle_{Y} - \nabla_{\hspace{-1pt} x} u_0 \rightarrow 0$ strongly in $L^2 (\Omega)$. \end{theorem} \begin{remark} Let us recall that in the classical setting and under some more restrictive assumptions on the matrix $A$ and on the regularity of the homogenized solution $u_0$, it is possible to prove (cfr. {\cite{allaire1999boundary,papanicolau1978asymptotic}} that $\| u_{\varepsilon} (x) - u (x) - \varepsilon u_1 (x, x / \varepsilon) \|_{H^1 (\Omega)} \in \mathcal{O} \left( \sqrt{\varepsilon} \right)$. This estimate, although generically optimal, is considered to be surprising since one could expect to get $\mathcal{O} (\varepsilon)$ if the next order term in the ansatz was truly $\varepsilon^2 u_2 (x, x / \varepsilon)$. As is well known, this worse-than-expected result is due to the appearance of boundary correctors, which must be taken into account to have $\mathcal{O} (\varepsilon)$ estimates. On the other hand, in our framework this this phenomenon disappears because of $\langle u_1 \rangle_{Y} = 0$. Indeed, in the average, the {\guillemotleft}classical{\guillemotright} first order corrector term $u_1$ does not play any role in the asymptotic expansion of $u_{\varepsilon}$ given by (\ref{eq:asymptexpansion}), and as we shall see in the next section, the first order significant (not null average) corrector is the so-called boundary corrector $v_{\varepsilon}$ (cfr. {\cite{allaire1999boundary}} and next section), for which we get the more natural result $\left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} - \nabla_{\hspace{-1pt} x} u_0 - \varepsilon \nabla_{\hspace{-1pt} x} v_{\varepsilon} \right\rangle_{Y} \right\|_{L^2 (\Omega)} \in \mathcal{O} (\varepsilon)$. \end{remark} \begin{proof} Let us observe that using $u_0$ and $u_1$ as test functions in (\ref{eq:weakformtemp1}) and (\ref{eq:homeq1}) we get \begin{eqnarray} \int_{Y \times \Omega} A (x, y) \left( \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y) \right) \cdot \nabla_{\hspace{-1pt} y} u_1 (x, y) \mathrm{d} x \mathrm{d} y & = & 0 \label{eq:tempinv1}\\ \int_{Y \times \Omega} A (x, y) [ \nabla_{\hspace{-1pt} x} u_0 (x) + \nabla_{\hspace{-1pt} y} u_1 (x, y) ] \cdot \nabla_{\hspace{-1pt} x} u_0 (x) \mathrm{d} x \mathrm{d} y & = & \int_{\Omega} f (x) u_0 (x) \mathrm{d} x. \label{eq:tempinv2} \end{eqnarray} We then observe that ($\alpha$ is the ellipticity constant of the matrix $A$) for any $u \in L^2_{\sharp} [ Y, L^2 (\Omega) ]^{}$ one has $\left\| \langle u \rangle_{Y} \right\|_{\Omega} \leqslant \| u \|_{\Omega \times Y}$ and hence, since $\langle u_1 \rangle_{Y} = 0$ we have \begin{eqnarray} \alpha \left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right\rangle_{Y} - \nabla_{\hspace{-1pt} x} u_0 \right\|^2_{L^2 (\Omega)} & \; = \; & \alpha \left\| \left\langle \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) - \nabla_{\hspace{-1pt} y} u_1 - \nabla_{\hspace{-1pt} x} u_0 \right\rangle_{Y} \right\|^2_{L^2 (\Omega)} \\ & \leqslant & \alpha \int_{\Omega \times Y} \left| \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) - \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \right|^2 . \label{eq:corrtemp1} \end{eqnarray} By the uniformly ellipticity of $A$ and (\ref{eq:corrtemp1}) we continue to estimate \begin{eqnarray} \alpha \left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right\rangle_{Y} - \nabla_{\hspace{-1pt} x} u_0 \right\|^2_{L^2 (\Omega)} & \leqslant & \int_{\Omega \times Y} A\mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \cdot \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \nonumber\\ & & \qquad + \int_{\Omega \times Y} A \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \cdot \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \nonumber\\ & & \hspace{4em} - \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \cdot \left( A + A^{\mathsf{T}} \right) \left( \nabla_{\hspace{-1pt} y} u_1 + \nabla_{\hspace{-1pt} x} u_0 \right) \\ & = & \int_{\Omega \times Y} f (x) u_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y + \int_{\Omega \times Y} A \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \cdot \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \nonumber\\ & & \hspace{4em} - \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \cdot \left( A + A^{\mathsf{T}} \right) \left( \nabla_{\hspace{-1pt} y} u_1 + \nabla_{\hspace{-1pt} x} u_0 \right), \end{eqnarray} the second equality being a consequence of the fact that $u_{\varepsilon}$ is the solution of the problem (\ref{eq:homogproblem2dot0feq})-(\ref{eq:homogproblem2dot0}). Taking into account (\ref{eq:tempinv1}) and (\ref{eq:tempinv2}) we then get \begin{eqnarray} \alpha \left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right\rangle_{Y} - \nabla_{\hspace{-1pt} x} u_0 \right\|^2_{L^2 (\Omega)} & \leqslant & \int_{\Omega \times Y} f (x) u_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y + \int_{\Omega} f (x) u_0 (x) \mathrm{d} x \nonumber\\ & & \quad - \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \cdot \left( A + A^{\mathsf{T}} \right) \left( \nabla_{\hspace{-1pt} y} u_1 + \nabla_{\hspace{-1pt} x} u_0 \right) . \end{eqnarray} Since $(A + A^T) \left( \nabla_{\hspace{-1pt} y} u_1 + \nabla_{\hspace{-1pt} x} u_0 \right) \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$, it is a test function for the two-scale convergence, so that (again from (\ref{eq:tempinv1}) and (\ref{eq:tempinv2})) \begin{eqnarray} - \lim_{\varepsilon \rightarrow 0} \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right) \cdot \left( A + A^{\mathsf{T}} \right) \left( \nabla_{\hspace{-1pt} y} u_1 + \nabla_{\hspace{-1pt} x} u_0 \right) & = & - 2 \int_{\Omega \times Y} A \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \cdot \left( \nabla_{\hspace{-1pt} x} u_0 + \nabla_{\hspace{-1pt} y} u_1 \right) \nonumber\\ & = & - 2 \int_{\Omega} f (x) u_0 (x) \mathrm{d} x. \end{eqnarray} Finally, to infer (\ref{eq:strongconvh1}), we simply observe that due to the $Y$ periodicity of $u_{\varepsilon}$ one has \begin{equation} \int_{\Omega \times Y} f (x) u_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y \; = \; \int_{\Omega \times Y} f (x) \mathcal{F} (u_{\varepsilon}) (x, y) \mathrm{d} x \mathrm{d} y \end{equation} with $u_{\varepsilon} \twoheadrightarrow u_0$. The proof is completed. \end{proof} \section{Higher Order Correctors: Boundary Layers}\label{sec:6bl} In what follows assume that the matrix of diffusion coefficients $A$ is symmetric and depends on the {\guillemotleft}periodic variable{\guillemotright} only, i.e. $A \in L^{\infty}_{\sharp} \left( Y \right)$, $A = A^{\mathsf{T}}$ and of course $A$ uniformly elliptic with $\alpha > 0$ as constant of ellipticity. By the uniqueness of the solution of the cell problem (\ref{eq:splittedsoldistribcell}) it is easily seen that in these hypotheses also the vector of correctors (see Definition \ref{def:classhomandcorr}) depends on the {\guillemotleft}periodic variable{\guillemotright} only, i.e. $\tmmathbf{\chi} \in [ H^1_{\sharp} \left( Y \right) ]^N$. In the previous section (see Theorem \ref{thm:corrH1}) we have seen that the sequence of the averaged solutions $\langle u_{\varepsilon} \rangle_{Y}$ strongly converge to $u_0$ in $H^1 (\Omega)$, i.e. that $\left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} - \nabla_{\hspace{-1pt} x} u_0 \right\rangle \right\|_{H^1 (\Omega)} \in \mathcal{O} (1)$. To have higher order estimates, especially near the boundary of $\Omega$, one has to introduce supplementary terms, called {\tmem{boundary layers}} {\cite{lions1981some}}, which roughly speaking aim to compensate the fast oscillation of the family of solutions $u_{\varepsilon}$ near the boundary $\partial \Omega$. More precisely, in this section we show that under suitable hypotheses one has \begin{equation} \left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} - \nabla_{\hspace{-1pt} x} u_0 - \varepsilon \nabla_{\hspace{-1pt} x} v_{\varepsilon} \right\rangle_{Y} \right\|_{L^2 (\Omega)} \in \mathcal{O} (\varepsilon), \label{eq:higherordercorr1} \end{equation} where $v_{\varepsilon}$ is the solution of the boundary layer problem: \begin{eqnarray} \tmop{div}_{\hspace{-1pt} x} \left( A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} v_{\varepsilon} (x, y) \right) & = \hspace{1.2em} \; 0 & \text{ in } \Omega \times Y \label{eq:boundarylayer1D}\\ v_{\varepsilon} (x, y) & = \hspace{1.2em} \; u_1 (x, y + x / \varepsilon) & \text{ on } \partial \Omega \times Y . \label{eq:boundarylayer1Dbc} \end{eqnarray} We also investigate the validity of the following stronger estimate \begin{equation} \left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} - \nabla_{\hspace{-1pt} x} u_0 \right\rangle_{Y} \right\|_{L^2 (\Omega)} \in \mathcal{O} (\varepsilon) . \label{eq:higherordercorr2} \end{equation} Quite remarkably, as we are going to show in the next subsection, in the one-dimensional case the stronger estimate (\ref{eq:higherordercorr2}) holds under the same hypotheses of the weaker estimate (\ref{eq:higherordercorr1}). \subsection{Higher Order Correctors in dimension one} In the one-dimensional setting $Y = [0, 1]$ and $\Omega \subseteq \mathbb{R}$ is an open interval: $\Omega := (0, \omega)$ with $\omega > 0$. We then denote by $a \in L^{\infty}_{\sharp} \left( Y \right)$ the unique coefficient of the matrix valued function $A$. Finally for the generic {\guillemotleft}1D function{\guillemotright} function $u \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ we shall denote by $u' \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$ the weak derivative with respect to the $x$ variable. \begin{theorem} \label{thm:BoundaryLayers1D}Let $(u_0, u_1) \in H_0^1 (\Omega) \times L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$ be the unique solution of the homogenized system of equations (\ref{eq:homeq1distrib})-(\ref{eq:homeq2}). The following estimate holds \begin{equation} \left\| \langle u'_{\varepsilon} \rangle_{Y} - u_0' \right\|_{L^2 (\Omega)} \leqslant 2 \varepsilon \cdot \frac{| a |_{\infty} | \chi_{\infty} |}{\alpha} (| u_0' |_{\infty} + \| u_0'' \|_{L^2 (\Omega)}) . \label{eq:1Dstrongestimate} \end{equation} \end{theorem} We will need the following two lemmas \begin{lemma} For any $\varepsilon > 0$ let $u_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ be the unique solution of the problem (\ref{eq:homogproblem2dot0feq})-(\ref{eq:homogproblem2dot0}). Define the {\mybold{error function}} \begin{equation} e_{\varepsilon} (x, y) := u_{\varepsilon} (x, y) - u_0 (x) - \varepsilon \left. [ u_1 \left( x, y + \frac{x}{\varepsilon} \right) - v_{\varepsilon} (x, y) \right. ], \end{equation} where $v_{\varepsilon} \in L^2_{\sharp} [ Y, H^1 (\Omega) ]$ is the unique solution of the boundary layer problem (\ref{eq:boundarylayer1D})-(\ref{eq:boundarylayer1Dbc}). The following estimate holds: \begin{equation} \| e_{\varepsilon}' \|_{L^2 \left( \Omega \times Y \right)} \leqslant \varepsilon \cdot \frac{| a |_{\infty} | \chi_{\infty} |}{\alpha} \| u_0'' \|_{L^2 (\Omega)} \in \mathcal{O} (\varepsilon) . \label{eq:fundest1derrorfunction1} \end{equation} \end{lemma} \begin{proof} In the 1D setting, the homogenized equation (\ref{eq:homogenizedpdej1}) read as $a_{\hom} u''_0 (x) = - f (x)$ with $a_{\hom} := \langle a^{- 1} (\cdot) \rangle^{- 1}_{Y} > 0$ and therefore $u_0 \in H_0^2 (\Omega)$. Indeed, as a consequence of Theorem \ref{thm:newhomapproach} (see eq. (\ref{eq:linkclassicaland2scale})), the unique solution $u_1 \in L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$ of (\ref{eq:homeq1distrib}) can be expressed in the tensor product form $u_1 (x, y) = \chi (y) u_0' (x)$, where $\chi$ is the unique (null average) solution in $H^1_{\sharp} \left( Y \right) / \mathbb{R}$ of (\ref{eq:splittedsoldistribcell}). A direct integration of the cell equation (\ref{eq:splittedsoldistribcell}) leads to (taking into account the periodicity of $u_1$ and averaging over {Y}) $a (y) (1 + \partial_y \chi (y)) \; = \; a_{\hom}$ with $a_{\hom} := \langle a (y) (1 + \partial_y \chi (y)) \rangle_{Y} = \langle a^{- 1} (\cdot) \rangle^{- 1}_{Y}$. Since $a_{\hom} u''_0 (x) = - f (x)$ from (\ref{eq:homogproblem2dot0feq}) we get $[a_{\hom} u_0' (x)]' = [ a \left( y + \frac{x}{\varepsilon} \right) u_{\varepsilon}' (x, y) ]'$. Hence, taking into account the equation satisfied by $v_{\varepsilon}$, a direct computation shows that for a.e. $y \in Y$ the function $e_{\varepsilon} (\cdot, y)$ satisfies the distributional equation \begin{equation} - \left( a \left( y + \frac{x}{\varepsilon} \right) e_{\varepsilon}' (x, y) \right)' \; = \; \varepsilon F_{\varepsilon}' (x, y) \quad \lo{\tmop{in}} \mathcal{D}' (\Omega), \label{eq:fundequality1Deps} \end{equation} with $F_{\varepsilon} (x, y) := a \left( y + \frac{x}{\varepsilon} \right)_{_{_{_{_{}}}}} \chi \left( y + \frac{x}{\varepsilon} \right) u_0'' (x)$. For every $\varphi_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$, the variational form in $L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ of (\ref{eq:fundequality1Deps}) reads as \begin{equation} \int_{\Omega \times Y} a \left( y + \frac{x}{\varepsilon} \right) e_{\varepsilon}' (x, y) \cdot \varphi_{\varepsilon}' (x, y) \mathrm{d} x \mathrm{d} y \; = \; - \varepsilon \int_{\Omega \times Y} F_{\varepsilon} (x, y) \cdot \varphi_{\varepsilon}' (x, y) \mathrm{d} x \mathrm{d} y. \label{eq:var1Derrorfunction} \end{equation} Since $e_{\varepsilon} \in L^2_{\sharp} [ Y, H^1_0 (\Omega) ]$ evaluating the variational equation (\ref{eq:var1Derrorfunction}) on the test function $\varphi_{\varepsilon} (x, y) := e_{\varepsilon} (x, y)$ and recalling that $a \geqslant \alpha$ we finish with (\ref{eq:fundest1derrorfunction1}). \end{proof} \begin{lemma} \label{lemma:uniformboundveps}Let $v_{\varepsilon} \in L^2_{\sharp} [ Y, H^1 (\Omega) ]$ solve the boundary value problem (\ref{eq:boundarylayer1D})-(\ref{eq:boundarylayer1Dbc}). Then the following uniform estimate (with respect to $\varepsilon$) holds: \begin{equation} \| v_{\varepsilon}' (x, y) \|_{L^2 \left( \Omega \times Y \right)} \leqslant \frac{2}{\alpha} | a |_{\infty} | \chi |_{\infty} | u_0' |_{\infty} \label{eq:fundest1derrorfunction2} . \end{equation} \end{lemma} \begin{proof} Let us integrate (\ref{eq:boundarylayer1D}). We get $v_{\varepsilon}' (x, y) \; = \; c_{\varepsilon} (y) a^{- 1} (y + x / \varepsilon)$ for some measurable real function $c_{\varepsilon}$. Taking into account boundary conditions (\ref{eq:boundarylayer1Dbc}), we compute \begin{equation} c_{\varepsilon} (y) = \frac{\chi (y + \omega / \varepsilon) u_0' (\omega) - \chi (y) u_0' (0)}{| \Omega | \langle a^{- 1} (y + \cdot / \varepsilon) \rangle_{\Omega}} . \end{equation} Next we note that $\left\langle a^{- 1} \left( y + \frac{\cdot}{\varepsilon} \right) \right\rangle_{\Omega}^{- 1} \leqslant | a |_{\infty}$ for a.e. $y \in \mathbb{R}$. Hence, observing that since $f \in L^2 (\Omega)$ one has $u_0 \in W^{1, \infty} (\Omega)$, we finish with the estimate $\alpha | \Omega | | v_{\varepsilon}' (x, y) | \leqslant 2 | a |_{\infty} | \chi |_{\infty} | u_0' |_{\infty}$ from which (\ref{eq:fundest1derrorfunction2}) immediately follows. \end{proof} We can now prove Theorem \ref{thm:BoundaryLayers1D}. \begin{proof}[of Theorem \ref{thm:BoundaryLayers1D}] Observing that $\left\langle [ u_1 \left( x, y + \frac{x}{\varepsilon} \right) ]' \right\rangle_{Y} \; = \; 0$ we compute \begin{eqnarray} \left\| \langle u'_{\varepsilon} \rangle_{Y} - u_0' \right\|_{L^2 (\Omega)} & \; = \; & \left\| \left\langle u'_{\varepsilon} (x, y) - u_0 (x) - \varepsilon [ u_1 \left( x, y + \frac{x}{\varepsilon} \right) ]' \right\rangle_{Y}' \right\|_{L^2 (\Omega)} \\ & \leqslant & \left\| \langle e'_{\varepsilon} (x, y) \rangle_{Y} \right\|_{\Omega} + \varepsilon \left\| \langle v_{\varepsilon}' (x, y) \rangle_{Y} \right\|_{L^2 (\Omega)} \\ & \leqslant & \| e'_{\varepsilon} \|_{L^2 \left( \Omega \times Y \right)} + \varepsilon \| v_{\varepsilon}' \|_{L^2 \left( \Omega \times Y \right)} . \end{eqnarray} Hence taking into account estimates (\ref{eq:fundest1derrorfunction1}) and (\ref{eq:fundest1derrorfunction2}) we get the result. \end{proof} \subsection{Higher Order Correctors in $N$ dimensions} This section is devoted to the proof of estimate (\ref{eq:higherordercorr1}). \begin{theorem} \label{thm:BoundaryLayersND}Let $(u_0, u_1) \in H_0^1 (\Omega) \times L^2 [ \Omega, H^1_{\sharp} (Y) / \mathbb{R} ]$ be the unique solution of the homogenized system of equations (\ref{eq:homeq1distrib})-(\ref{eq:homeq2}). Define the error function by the position \begin{equation} e_{\varepsilon} (x, y) := u_{\varepsilon} (x, y) - u_0 (x) - \varepsilon \left. [ u_1 \left( x, y + \frac{x}{\varepsilon} \right) - v_{\varepsilon} (x, y) \right. ], \end{equation} $v_{\varepsilon} \in L^2_{\sharp} [ Y, H^1 (\Omega) ]$ being the unique solution of the boundary layer problem (\ref{eq:boundarylayer1D})-(\ref{eq:boundarylayer1Dbc}). If $u_0 \in H^2 (\Omega)$ then $\left\| \left\langle \nabla_{\hspace{-1pt} x} e_{\varepsilon} \right\rangle_{Y} \right\|_{\Omega} \in \mathcal{O} (\varepsilon)$. More precisely, the following estimate holds \begin{equation} \left\| \left\langle \nabla_{\hspace{-1pt} x} u_{\varepsilon} \right\rangle_{Y} - \nabla_{\hspace{-1pt} x} u_0 + \varepsilon \left\langle \nabla_{\hspace{-1pt} x} v_{\varepsilon} \right\rangle_{Y} \right\|_{L^2 (\Omega)} \leqslant \varepsilon c_A \| u_0 \|_{H^2 (\Omega)}, \label{eq:NDstrongestimate} \end{equation} for a suitable constant $c_{\alpha} > 0$ depending on the matrix $A$ only. \end{theorem} \begin{proof} Let us set $u_1^{\varepsilon} (x, y) := u_0 (x) + \varepsilon u_1 (x, y)$, where $u_1 (x, y) = \nabla_{\hspace{-1pt} x} u_0 (x) \cdot \tmmathbf{\chi} (y)$ as shown in Theorem \ref{thm:newhomapproach}. We have (let us denote by $\mathcal{H}_x := \nabla_{\hspace{-1pt} x} \nabla_{\hspace{-1pt} x}$ the partial hessian operator) \begin{equation} \nabla_{\hspace{-1pt} x} [ u_1^{\varepsilon} \left( x, y + \frac{x}{\varepsilon} \right) ] \; = \; [ I + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} \left( y + \frac{x}{\varepsilon} \right) ] \nabla_{\hspace{-1pt} x} u_0 + \varepsilon \mathcal{H}_x [u_0] (x) \tmmathbf{\chi} \left( y + \frac{x}{\varepsilon} \right) . \end{equation} Hence \begin{equation} A_{\hom} \nabla_{\hspace{-1pt} x} u_0 (x) - A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} [ u_1^{\varepsilon} \left( x, y + \frac{x}{\varepsilon} \right) ] \; = \; \mathcal{A}_0 \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} u_0 - \varepsilon \tmmathbf{h} \left( x, y + \frac{x}{\varepsilon} \right), \label{eq:jjtemp1} \end{equation} where, for notational convenience, we have introduce the functions \begin{equation} \tmmathbf{h} (x, y) := A (y) \mathcal{H}_x [u_0] (x) \tmmathbf{\chi} (y) \quad, \quad \mathcal{A}_0 (y) := A_{\hom} - a_{\hom} (y) \end{equation} with $a_{\hom} (y) := A (y) [ I + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} (y) ]$. Let us note that $\langle \mathcal{A}_0 \rangle_{Y} = 0$, because $A_{\hom} = \langle a_{\hom} (\cdot) \rangle_{Y}$. By taking the distributional divergence of both members of the previous equation (\ref{eq:jjtemp1}), recalling that $\tmop{div}_{\hspace{-1pt} x} \left( A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} v_{\varepsilon} (x, y) \right) = 0$, that due to (\ref{eq:homogproblem2dot0feq}) and (\ref{eq:homogenizedpdej1}) one has \begin{equation} \tmop{div}_{\hspace{-1pt} x} [ A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} u_{\varepsilon} (x, y) ] \; = \; \tmop{div}_{\hspace{-1pt} x} \left( A_{\hom} (x) \nabla_{\hspace{-1pt} x} u_0 (x) \right), \end{equation} and that $v_{\varepsilon} \in L^2_{\sharp} [ Y, H^1 (\Omega) ]$ is the solution of the boundary layer problem (\ref{eq:boundarylayer1D})-(\ref{eq:boundarylayer1Dbc}), we get \begin{equation} \tmop{div}_{\hspace{-1pt} x} \left( A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} [e_{\varepsilon} (x, y)] \right) \; = \; \tmop{div}_{\hspace{-1pt} x} \left( \mathcal{A}_0 \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} u_0 \right) - \varepsilon \tmop{div}_{\hspace{-1pt} x} [ \tmmathbf{h} \left( x, y + \frac{x}{\varepsilon} \right) ] . \label{eq:fundequality1DepsN} \end{equation} Next, let us recall that in the space $L^2_{\tmop{sol}} \left( Y \right)$ of solenoidal and periodic vector fields, defined by the position $L^2_{\tmop{sol}} \left( Y \right) := \left\{ \tmmathbf{p} \in L^2 \left( Y \right) \; : \; \tmop{div} \tmmathbf{p} (y) = 0 \right\}$ the following Helmholtz-Hodge decomposition holds (cfr. {\cite{Jikov1994}}): if $\tmmathbf{p} \in L^2_{\tmop{sol}} \left( Y \right)$ there exists a skew-symmetric matrix $\tmmathbf{\omega} := \left( \tmmathbf{\omega}^1 \; | \; \tmmathbf{\omega}^2 \; | \; \cdots \; | \; \tmmathbf{\omega}^N \right) \in [ H^1_{\sharp} \left( Y \right) ]^{N \times N}$ such that \begin{equation} \langle \tmmathbf{\omega} \rangle_{Y} =\tmmathbf{0} \quad, \quad \tmmathbf{p} \; = \; \langle \tmmathbf{p} \rangle_{Y} + \sum_{j = 1}^N \partial_j \tmmathbf{\omega}^j = \langle \tmmathbf{p} \rangle_{Y} + \tmmathbf{\tmop{curl}}\hspace{1pt} \tmmathbf{\omega}, \end{equation} with $\tmmathbf{\tmop{curl}}\hspace{1pt} : \tmmathbf{\omega} \mapsto \tmmathbf{\tmop{curl}}\hspace{1pt} \tmmathbf{\omega} := \partial_1 \tmmathbf{\omega}^1 + \cdots + \partial_N \tmmathbf{\omega}^N$. Note that $\tmop{div}_{\hspace{-1pt} y} \mathcal{A}_0 (y) =\tmmathbf{0}$ because $\mathcal{A}_0$ solves the cell equation (\ref{eq:splittedsoldistribcell}). On the other hand, $\langle \mathcal{A}_0 \rangle_{Y} =\tmmathbf{0}$ and therefore due to the Helmholtz-Hodge decomposition there exist skew-symmetric matrices $(\tmmathbf{\omega}_i)_{i \in \mathbb{N}_N} \in [ H^1_{\sharp} \left( Y \right) ]^{N \times N}$ such that $\mathcal{A}_0 (y) \mathbf{e}_i = \tmmathbf{\tmop{curl}}\hspace{1pt} \tmmathbf{\omega}_i (y)$ for every $i \in \mathbb{N}_N$. From the scaling relation \begin{equation} \varepsilon \cdot \tmmathbf{\tmop{curl}}_{\hspace{0.5pt} x} [\tmmathbf{\omega}_i (y + x / \varepsilon)] = \tmmathbf{\tmop{curl}}\hspace{1pt} \tmmathbf{\omega}_i (y + x / \varepsilon), \end{equation} recalling that for any $g \in H^1_{\sharp} \left( Y \right), \tmmathbf{\omega} \in [ H^1_{\sharp} \left( Y \right) ]^{N \times N}$ one has $\tmmathbf{\tmop{curl}}\hspace{1pt} (g\tmmathbf{\omega}) =\tmmathbf{\omega} \nabla g + g \tmmathbf{\tmop{curl}}\hspace{1pt} \tmmathbf{\omega}$ in $\mathcal{D}'$, we have \begin{eqnarray} \mathcal{A}_0 \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} u_0 (x) & \; = \; & \varepsilon \sum_{i \in \mathbb{N}_N} \partial_i u_0 (x) \tmmathbf{\tmop{curl}}_{\hspace{0.5pt} x} [\tmmathbf{\omega}_i (y + x / \varepsilon)] \nonumber\\ & \; = \; & \varepsilon \sum_{i \in \mathbb{N}_N} \tmmathbf{\tmop{curl}}_{\hspace{0.5pt} x} [\partial_i u_0 (x) \tmmathbf{\omega}_i (y + x / \varepsilon)] - \varepsilon \sum_{i \in \mathbb{N}_N} \tmmathbf{\omega}_i (y + x / \varepsilon) \partial_i \nabla_{\hspace{-1pt} x} u_0 (x) \nonumber\\ & \; = \; & \varepsilon \sum_{i \in \mathbb{N}_N} \tmmathbf{\tmop{curl}}_{\hspace{0.5pt} x} [\partial_i u_0 (x) \tmmathbf{\omega}_i (y + x / \varepsilon)] - \varepsilon \tmmathbf{\eta} (x, y + x / \varepsilon), \end{eqnarray} with $\tmmathbf{\eta} (x, y) := \sum_{i \in \mathbb{N}_N} \tmmathbf{\omega}_i (y) \partial_i \nabla_{\hspace{-1pt} x} u_0 (x) \in L^2_{\sharp} [ Y, L^2 (\Omega) ]$ and $\langle \tmmathbf{\eta} \rangle_{Y} =\tmmathbf{0}$. Passing to the divergence in the previous relations, we get $\tmop{div}_{\hspace{-1pt} x} \left( \mathcal{A}_0 \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} u_0 (x) \right) = \varepsilon \tmop{div}_{\hspace{-1pt} x} (\tmmathbf{\eta} (x, y + x / \varepsilon))$. Hence, equation (\ref{eq:fundequality1DepsN}) simplifies to \begin{equation} \tmop{div}_{\hspace{-1pt} x} \left( A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} [e_{\varepsilon} (x, y)] \right) \; = \; - \varepsilon \tmop{div}_{\hspace{-1pt} x} \tmmathbf{F}_{\varepsilon} (x, y) \quad \text{in } \mathcal{D}' \left( \Omega \times Y \right), \label{eq:fundequality1Depscurl} \end{equation} with $\tmmathbf{F}_{\varepsilon} (x, y) := \tmmathbf{\eta} (x, y + x / \varepsilon) +\tmmathbf{h} \left( x, y + \frac{x}{\varepsilon} \right) \in L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$ and $\{ \tmmathbf{F}_{\varepsilon} \}_{\varepsilon \in \mathbb{R}^+}$ a bounded subset of $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$. The previous equation (\ref{eq:fundequality1Depscurl}) reads in variational form as \begin{equation} \int_{\Omega \times Y} A \left( y + \frac{x}{\varepsilon} \right) \nabla_{\hspace{-1pt} x} e_{\varepsilon} (x, y) \cdot \nabla_{\hspace{-1pt} x} \varphi_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y \; = \; - \varepsilon \int_{\Omega \times Y} \tmmathbf{F}_{\varepsilon} (x, y) \cdot \nabla_{\hspace{-1pt} x} \varphi_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y, \label{eq:fundepsestimatenD} \end{equation} for any $\varphi_{\varepsilon} \in L^2_{\sharp} [ Y, H_0^1 (\Omega) ]$. Since $v_{\varepsilon}$ solves the boundary layer problem (\ref{eq:boundarylayer1D})-(\ref{eq:boundarylayer1Dbc}), we have $e_{\varepsilon} \in L^2_{\sharp} [ Y, H_0 (\Omega) ]$ and therefore, testing (\ref{eq:fundepsestimatenD}) on $e_{\varepsilon}$ we finish, for some suitable constant $c_A > 0$ depending on $A$ only, with (\ref{eq:NDstrongestimate}). \end{proof} \section{Weak two-scale compactness for $\mathbf{S}^2$-valuedHarmonic maps}\label{sec:homharmonicmaps} The aim of this section is to prove a weak two-scale compactness result for $\mathbf{S}^2$-valued harmonic maps, and make some remarks which point out possible weaknesses of this alternative notion of two-scale convergence. In what follows \ $\Omega$ is a bounded and Lipschitz domain of $\mathbb{R}^3$ and we shall make use of the following notations: $W (\Omega) := L^{\infty} (\Omega) \cap H^1 (\Omega)$ and $W_0 (\Omega) := L^{\infty} (\Omega) \cap H^1_0 (\Omega)$. \subsection{Harmonic maps equation} We want to focus on the homogenization of the family of harmonic map equations arising as the Euler-Lagrange equations associated to the family of Dirichlet energy functionals \begin{equation} \mathcal{E}_{\varepsilon} (\tmmathbf{u}_{\varepsilon}) := \int_{\Omega \times Y} a_{\varepsilon} (x, y) \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} (x, y) \cdot \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y \quad, \quad a_{\varepsilon} (x, y) := a \left( x, y + \frac{x}{\varepsilon} \right), \label{eq:dirichletenfunc} \end{equation} all defined in $L^{\infty}_{\sharp} [ Y, W \left( \Omega, \mathbf{S}^2 \right) ]^3$. Here, as usual, the coefficient $a \in L^{\infty}_{\sharp} \left( Y, L^{\infty} (\Omega) \right)$ is a positive function bounded from below by some positive constant. The stationary condition on $\mathcal{E}_{\varepsilon}$ with respect to tangential variations in $L^{\infty}_{\sharp} [ Y, W_0 (\Omega) ]^3$ conducts to the equation of harmonic maps \begin{equation} \int_{\Omega \times Y} a_{\varepsilon} (x, y) \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} (x, y) \nabla_{\hspace{-1pt} x} \tmmathbf{\eta}_{\varepsilon} (x, y) \mathrm{d} x \mathrm{d} y = 0 \label{eq:statharmmap} \end{equation} which must be satisfied for every $\tmmathbf{\eta}_{\varepsilon} \in L^{\infty}_{\sharp} [ Y, W_0 (\Omega) ]^3$ such that $\tmmathbf{\eta}_{\varepsilon} (x, y) \in T_{\tmmathbf{u}_{\varepsilon} (x, y)} \mathbf{S}^2$ a.e. in $\Omega \times Y$. \begin{theorem} \label{thm:homogenizationhmaps}For every $\varepsilon \in \mathbb{R}^+$ let $\tmmathbf{u}_{\varepsilon} \in L^{\infty}_{\sharp} [ Y, W \left( \Omega, \mathbf{S}^2 \right) ]^3$ be a solution of the harmonic map equation (\ref{eq:statharmmap}). If $(\tmmathbf{u}_{\varepsilon})_{\varepsilon \in \mathbb{R}^+} \twoheadrightarrow \tmmathbf{u}_0$ weakly in $\mathrm{L^2_{\sharp} [ Y, H^1 (\Omega) ]}^3$ and $\tmmathbf{u}_0$ takes values on $\mathbf{S}^2$, then $\tmmathbf{u}_0$ is still an harmonic map. More precisely, $\tmmathbf{u}_0 \in W \left( \Omega, \mathbf{S}^2 \right)^3$ satisfies the following homogenized harmonic map equation \begin{equation} \int_{\Omega} A_{\hom} (x)_{_{_{_{}}}} \nabla \tmmathbf{u}_0 (x) \nabla \tmmathbf{\varphi} (x) \mathrm{d} x \; = \; 0 \quad \forall \tmmathbf{\varphi} \in T_{\tmmathbf{u}_0} \mathbf{S}^2 \label{eq:homogenizedharmonicmapeq} \end{equation} in which \begin{equation} A_{\hom} (x) := \int_{Y} a (x, y) \left( I + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} (x, y) \right) \mathrm{d} y, \end{equation} and $\tmmathbf{\chi} := (\chi_1, \chi_2, \chi_3) \in L^2 [ \Omega, H^1_{\sharp} \left( Y \right) ]^3$ is the unique null average solution of the cell problems ($i \in \mathbb{N}_3$) \begin{equation} \tmop{div}_{\hspace{-1pt} y} \left( a (x, y) \left( \nabla_{\hspace{-1pt} y} \chi_{i_{}} (x, y) + e_i \right) \right) \; = \; 0. \label{eq:cellharmonic1} \end{equation} \end{theorem} \begin{remark} \label{rm:db1}In stating Theorem \ref{thm:homogenizationhmaps} we have assumed that the weak limit $\tmmathbf{u}_0$ still takes values on the unit sphere of $\mathbb{R}^3$. Indeed, and this is a drawback of the alternative two-scale notion, although the introduction of the $y$ variable in (\ref{eq:statharmmap}) overcomes the problem of the admissibility of the coefficient $a_{\varepsilon}$, it introduces a loss of compactness into the family of energy functionals $\mathcal{E}_{\varepsilon}$ defined in (\ref{eq:dirichletenfunc}). Indeed, in the space $\mathrm{L^2_{\sharp} [ Y, H^1 \left( \Omega, \mathbf{S}^2 \right) ]}^3$, Rellich--Kondrachov theorem does not apply, and therefore any uniform bound on the family $\mathcal{E}_{\varepsilon}$ does not assure compactness of minimizing sequences. \end{remark} \begin{remark} The same result still holds, with minor modifications, if we replace $\mathbf{S}^2$ with $\mathbf{S}^{n - 1}$. Moreover an analogue result holds if one replace the energy density $a_{\varepsilon} \left| \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} \right|^2$ with the energy density $\sum_{i \in \mathbb{N}_3} A_{i, \varepsilon} \nabla_{\hspace{-1pt} x} u_{i, \varepsilon} \cdot \nabla_{\hspace{-1pt} x} u_{i, \varepsilon}$ in which every $A_{\varepsilon, i}$ is a definite positive symmetric matrix. On the other hand, the proof does not work anymore when the image manifold is arbitrary. Indeed, for $\mathbf{S}^{n - 1}$ valued maps, we can exploit a result of {\tmname{Chen}} {\cite{chen1989weak}} which permits to equivalently write the Euler-Lagrange equation (\ref{eq:statharmmap}) as an equation in divergence form. Unfortunately, this conservation law heavily relies on the invariance under rotations of Dirichlet energy for maps into $\mathbf{S}^{n - 1}$. As a matter of fact, when the target manifold is arbitrary, even the less general problem concerning weak compactness for weakly harmonic maps remains open {\cite{Helein2002harmonic}}. \end{remark} We shall make use of the following Lemma which, although more than sufficient for addressing our problem, can still be rephrased to cover more general situations. Note that an equivalent result, in the context of classical two-scale convergence, has already been proved in {\cite{alouges2015homogenization}}. \begin{lemma} \label{Lemma:2scalemanifold}Let $\mathcal{M} \subset \mathbb{R}^N$ be a regular closed orientable hypersurface, and let $(\tmmathbf{u}_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ be a family of $L^2_{\sharp} [ Y, H^1 (\Omega) ]^N$ vector fields such that $\tmmathbf{u}_{\varepsilon} (x, y) \in \mathcal{M}$ a.e. in $\Omega \times Y$. If for some $\tmmathbf{u}_0 \in L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$, $\tmmathbf{\xi} \in L^2_{\sharp} [ Y, L^2 (\Omega) ]^{N \times N}$ one has \begin{equation} \tmmathbf{u}_{\varepsilon} \twoheadrightarrow \tmmathbf{u}_0 \quad \text{strongly in } L^2_{\sharp} [ Y, L^2 (\Omega) ]^N \quad, \quad \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} \twoheadrightarrow \tmmathbf{\xi} \quad \text{in } L^2_{\sharp} [ Y, L^2 (\Omega) ]^{N \times N}, \end{equation} then $\tmmathbf{u}_0 (x, y) = \langle \tmmathbf{u}_0 (x, \cdot) \rangle_{Y}$, i.e. the two-scale limit $\tmmathbf{u}_0$ does not depends on the $y$ variable. Moreover there exists an element $\tmmathbf{u}_1 \in L^2 [ \Omega, H^1_{\sharp} \left( Y \right) ]^N$ such that \begin{equation} \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} \twoheadrightarrow \left( \nabla_{\hspace{-1pt} x} \tmmathbf{u}_0 + \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1 \right) \text{ weakly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$} \label{eq:statementalreadyknown} \end{equation} with $\tmmathbf{u}_0 (x) \in \mathcal{M}$ and $\tmmathbf{u}_1 (x, y) \in T_{\tmmathbf{u}_0 (x)} \mathcal{M}$ for a.e. $(x, y) \in \Omega \times Y$. \end{lemma} \begin{remark} Here, as already observed in Remark \ref{rm:db1}, we have to assume strong two-scale convergence since the boundedness of the family $(\tmmathbf{u}_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ in $L^2_{\sharp} [ Y, H^1 (\Omega) ]^N$ does not imply strong convergence in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$ of a suitable subsequence, which is an essential requirement in order to prove that the limit function $\tmmathbf{u}_0$ takes values on $\mathcal{M}$. \end{remark} \begin{proof} Since $\tmmathbf{u}_{\varepsilon} \twoheadrightarrow \tmmathbf{u}_0$ in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$ the first part of the theorem (namely (\ref{eq:statementalreadyknown})) is nothing else that Proposition \ref{prop:g1classicalctwoscaleL2H1}. It remains to prove the second part. To this end let us recall (cfr. {\cite{do1976differential}}) that since $\mathcal{M}$ is a regular closed orientable surface there exists an open tubular neighbourhood $U \subseteq \mathbb{R}^N$ of $\mathcal{M}$ and a function $g : U \rightarrow \mathbb{R}$ which has zero as a regular value and is such that $\mathcal{M}= g^{- 1} (0)$. Since $\tmmathbf{u}_{\varepsilon} \twoheadrightarrow \tmmathbf{u}_0$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$ we have $0 = g (\tmmathbf{u}_{\varepsilon} (x, y)) \twoheadrightarrow g (\tmmathbf{u}_0 (x))$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^N$ and therefore $g (\tmmathbf{u}_0 (x)) = 0$ a.e. in $\Omega$. Next we observe that for any $\varepsilon \in \mathbb{R}^+$ we have $g (\tmmathbf{u}_{\varepsilon}) = 0$ and hence $\nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} (x, y) .\tmmathbf{n} (\tmmathbf{u}_{\varepsilon} (x, y)) =\tmmathbf{0}$ for a.e. $(x, y) \in \Omega \times Y$. Passing to the two-scale limit we so get \begin{eqnarray} 0 \; = \; \int_{\Omega \times Y} \mathcal{F}_{\varepsilon} \left( [ \nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} ] \tmmathbf{n} (\tmmathbf{u}_{\varepsilon}) \right) (x, y) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \qquad \qquad \qquad \qquad \qquad \qquad \qquad & & \nonumber\\ \xrightarrow{\varepsilon \rightarrow 0} \int_{\Omega \times Y} [ \nabla_{\hspace{-1pt} x} \tmmathbf{u}_0 (x) + \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1 (x, y) ] \tmmathbf{n} (\tmmathbf{u}_0 (x)) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \; = \; 0 & & \label{eq:twoscaleonmanifold} \end{eqnarray} for every $\tmmathbf{\psi} \in L^{\infty}_{\sharp} [ Y, W (\Omega) ]^N$. In particular, by taking $\tmmathbf{\psi} (x, y) := \tmmathbf{\varphi} (x) \otimes 1 (y)$, since $\left\langle \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1 (x, y) \right\rangle_{Y} = 0$ we have $\nabla_{\hspace{-1pt} x} \tmmathbf{u}_0 (x) \tmmathbf{n} (\tmmathbf{u}_0 (x)) =\tmmathbf{0}$ a.e. in $\Omega$. Thus from (\ref{eq:twoscaleonmanifold}) we get \[ \int_{\Omega \times Y} \nabla_{\hspace{-1pt} y} (\tmmathbf{u}_1 (x, y) \cdot \tmmathbf{n} (\tmmathbf{u}_0 (x))) \cdot \tmmathbf{\psi} (x, y) \mathrm{d} x \mathrm{d} y \; = \; 0 \quad \forall \tmmathbf{\psi} \in L^{\infty}_{\sharp} [Y, W_0 (\Omega)]^N \] and hence for some $c \in \mathbb{R}$ we have $\tmmathbf{u}_1 \cdot \tmmathbf{n} (\tmmathbf{u}_0) = c$ a.e. in $\Omega \times Y$. But since $\tmmathbf{u}_1$ is null average on $Y$, so is $\tmmathbf{u}_1 \cdot \tmmathbf{n} (\tmmathbf{u}_0)$ and therefore necessarily $c = 0$. \end{proof} \begin{proof}[of Theorem \ref{thm:homogenizationhmaps}] For any $\tmmathbf{\psi}_{\varepsilon} \in L^{\infty}_{\sharp} [ Y, W_0 (\Omega) ]^3$ we set $\tmmathbf{\eta}_{\varepsilon} := \tmmathbf{u}_{\varepsilon} \times \tmmathbf{\psi}_{\varepsilon}$ in equation (\ref{eq:statharmmap}). We then have $\nabla_{\hspace{-1pt} x} \tmmathbf{u}_{\varepsilon} \nabla_{\hspace{-1pt} x} \tmmathbf{\eta}_{\varepsilon} = \sum_{i \in \mathbb{N}_3} \partial_{x_i} \tmmathbf{\psi}_{\varepsilon} \cdot (\partial_{x_i} \tmmathbf{u}_{\varepsilon} \times \tmmathbf{u}_{\varepsilon})$ and therefore \begin{equation} \sum_{i \in \mathbb{N}_3} \int_{\Omega \times Y} a (x, y) [ \mathcal{F}_{\varepsilon} (\tmmathbf{u}_{\varepsilon}) \times \mathcal{F}_{\varepsilon} \left( \frac{\partial \tmmathbf{u}_{\varepsilon}}{\partial x_i} \right) ] \cdot \frac{\partial \tmmathbf{\psi}_{\varepsilon}}{\partial x_i} \mathrm{d} x \mathrm{d} y = 0 \quad \forall \tmmathbf{\psi}_{\varepsilon} \in L^{\infty}_{\sharp} [ Y, W_0 (\Omega) ]^3 \label{eq:tobetwoscaled} \end{equation} By mimicking the proof of Proposition \ref{prop:attainedtest}, it is simple to get that for every $\tmmathbf{\eta} \in W_0 (\Omega)^3$ there exists a family $(\tmmathbf{\psi}_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ of $L^{\infty}_{\sharp} [ Y, W_0 (\Omega) ]^3$ functions such that $\tmmathbf{\psi}_{\varepsilon} \twoheadrightarrow \tmmathbf{\eta}$ and $\nabla_{\hspace{-1pt} x} \tmmathbf{\psi}_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} x} \tmmathbf{\eta}$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^3$, so that taking into account Proposition \ref{prop:g1classicalctwoscaleL2H1}, passing to the two-scale limit in (\ref{eq:tobetwoscaled}) we get \begin{equation} \sum_{i \in \mathbb{N}_3} \int_{\Omega \times Y} a (x, y) [ \tmmathbf{u}_0 (x) \times \left( \frac{\partial \tmmathbf{u}_0}{\partial x_i} (x) + \frac{\partial \tmmathbf{u}_1}{\partial y_i} (x, y) \right) ] \cdot \frac{\partial \tmmathbf{\eta}}{\partial x_i} (x) \mathrm{d} x \mathrm{d} y \; = \; 0 \quad \forall \tmmathbf{\eta} \in W_0 (\Omega)^3 . \label{eq:quasihomohm} \end{equation} On the other hand, again by by mimicking the proof of Proposition \ref{prop:attainedtest}, we get that for every $\tmmathbf{\psi}_1 \in L^{\infty}_{\sharp} [Y, W_0 (\Omega)]^3$ there exists a family $(\tmmathbf{\psi}_{\varepsilon})_{\varepsilon \in \mathbb{R}^+}$ of $L^{\infty}_{\sharp} [Y, W_0 (\Omega)]^3$ functions such that $\varepsilon \nabla_{\hspace{-1pt} x} \tmmathbf{\psi}_{\varepsilon} \twoheadrightarrow \nabla_{\hspace{-1pt} y} \tmmathbf{\psi}_1$ strongly in $L^2_{\sharp} [ Y, L^2 (\Omega) ]^3$. Hence, from Proposition \ref{prop:g1classicalctwoscaleL2H1}, passing to the two-scale limit in (\ref{eq:tobetwoscaled}) we get \begin{equation} \sum_{i \in \mathbb{N}_3} \int_{\Omega \times Y} [ \tmmathbf{u}_0 (x) \times a (x, y) \left( \frac{\partial \tmmathbf{u}_0}{\partial x_i} (x) + \frac{\partial \tmmathbf{u}_1}{\partial y_i} (x, y) \right) ] \cdot \frac{\partial \tmmathbf{\psi}_1}{\partial y_i} (x, y) \mathrm{d} x \; = \; 0 \end{equation} for every $\tmmathbf{\psi}_1 \in L^{\infty}_{\sharp} [Y, W_0 (\Omega)]^3$. In particular, for any $\tmmathbf{\psi} \in L^{\infty}_{\sharp} [Y, W_0 (\Omega)]^3$, by setting $\tmmathbf{\psi}_1 (x, y) := \tmmathbf{u}_0 (x) \times \tmmathbf{\psi} (x, y)$ and taking into account that due to Lemma \ref{Lemma:2scalemanifold} $\tmmathbf{u}_1 (x, y) \cdot \tmmathbf{u}_0 (x) = 0$ a.e. in $\Omega \times Y$ we finish with the classical cell equation \begin{equation} \sum_{i \in \mathbb{N}_3} \int_{\Omega \times Y} a (x, y) \left( \frac{\partial \tmmathbf{u}_0}{\partial x_i} (x) + \frac{\partial \tmmathbf{u}_1}{\partial y_i} (x, y) \right) \cdot \frac{\partial \tmmathbf{\psi}}{\partial y_1} (x, y) \mathrm{d} x \; = \; 0 \quad \forall \tmmathbf{\psi} \in L^{\infty}_{\sharp} [Y, W_0 (\Omega)]^3 . \label{eq:cellequationshm} \end{equation} The solution of the previous equation is classical. Indeed, due to Lax-Milgram lemma, the cell problem (\ref{eq:cellequationshm}), which in distributional form reads as \begin{equation} - \mathbf{div}_y \left( a (x, y) \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1 (x, y) \right) \; = \; \mathbf{div}_y (a (x, y) \nabla \tmmathbf{u}_0 (x)), \label{eq:cellproblemvarformharmmapdistr} \end{equation} has a unique null average solution in $L^2 [ \Omega, H^1_{\sharp} \left( Y \right) ]^3$. Moreover, if for every $i \in \mathbb{N}_3$ we denote by $\chi_{i_{}}$ the unique null average solution in $L^2 [ \Omega, H^1_{\sharp} \left( Y \right) ]$ of the scalar cell problem (\ref{eq:cellharmonic1}), by the defining the vector valued function $\tmmathbf{\chi} := (\chi_1, \chi_2, \chi_3) \in L^2 [ \Omega, H^1_{\sharp} \left( Y \right) ]^3$ we get that the vector field \begin{equation} \tmmathbf{u}_1 (x, y) := \sum_{j \in \mathbb{N}_3} \left( \tmmathbf{\chi} (x, y) \cdot \nabla_{\hspace{-1pt} x} u_0^j (x) \right) e_j \label{eq:expru1} \end{equation} is the unique null average solution in $L^2 [ \Omega, H^1_{\sharp} \left( Y \right) ]^3$ of the cell problem (\ref{eq:cellproblemvarformharmmapdistr}). Next we note that from (\ref{eq:expru1}) we get $\nabla \tmmathbf{u}_0 (x) + \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1 (x, y) = \left( I + \nabla_{\hspace{-1pt} y} \tmmathbf{\chi} (x, y) \right) \nabla_{\hspace{-1pt} x} \tmmathbf{u}_0 (x)$ and hence, evaluating (\ref{eq:quasihomohm}) on vector fields of the form $\tmmathbf{\eta} (x) := \tmmathbf{u}_0 (x) \times \tmmathbf{\varphi} (x)$ with $\tmmathbf{\varphi} \in W_0 (\Omega)^3$ and $\tmmathbf{\varphi} (x) \in T_{\tmmathbf{u}_0 (x)} \mathbf{S}^2$ we finish with (\ref{eq:homogenizedharmonicmapeq}). \end{proof} \begin{remark} In general, if we do not assume any positivity condition on the coefficient $a$, it is not possible to reduce the domain equation (\ref{eq:quasihomohm}) and the cell equation (\ref{eq:cellequationshm}) to a single homogenized equation (like the one obtained in Theorem \ref{thm:homogenizationhmaps}). Nevertheless the two-scale limit $\tmmathbf{u}_0$ will be a solution of the system of two distributional equations \begin{eqnarray} \mathbf{div}_x \left( \tmmathbf{u}_0 (x) \times \int_{Y} a (x, y) \left( \nabla_{\hspace{-1pt} x} \tmmathbf{u}_0 (x) + \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1 (x, y) \right) \mathrm{d} y \right) & = & \tmmathbf{0} \quad \text{in } \mathcal{D}' (\Omega) \\ \mathbf{div}_y \bigl( a(x, y) \bigl( \nabla \tmmathbf{u}_0 (x) + \nabla_{\hspace{-1pt} y} \tmmathbf{u}_1(x, y) \bigr) \bigr) & = & \tmmathbf{0} \quad \text{in } \mathcal{D}' \left( \Omega \times Y \right) . \end{eqnarray} \end{remark} \section{Conclusion and Acknowledgment} This work was partially supported by the labex LMH through the grant no. ANR-11-LABX-0056-LMH in the {\tmem{Programme des Investissements d'Avenir}}.
7913db4bc34326311360779d4714be05023eb0bf
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The study of hereditary properties of combinatorial structures is an important topic within the field of extremal combinatorics. Out of the many results in this line of research has emerged an pattern for how to prove approximate asymptotic enumeration and structure results. The aim of this paper is to provide a general framework in which to view these results and to formalize this pattern of proof. \subsection{Background} A nonempty class of graphs $\mathcal{P}$ is called a \emph{hereditary graph property} if it is closed under isomorphism and induced subgraphs. Given a hereditary graph property $\mathcal{P}$, let $\mathcal{P}_n$ denote the set of elements of $\mathcal{P}$ with vertex set $[n]$. There has been extensive investigation into the properties of $\mathcal{P}_n$, where $\mathcal{P}$ is a hereditary property of graphs and $n$ is large, see for instance \cite{Alekseev1, Alekseev2, BBW1, BBW2, BBW3, BoTh1, BoTh2,ZitoSch}. The main questions addressed in these papers concern \emph{enumeration} (finding an asymptotic formula for $|\mathcal{P}_n|$) and \emph{structure} (understanding what properties elements of $\mathcal{P}_n$ have with high probability). Given a graph $H$, $\operatorname{Forb}(H)$ (respectively $\operatorname{Forb}_{ind}(H)$) is the class of finite graphs omitting $H$ as a non-induced (respectively induced) subgraph. For any graph $H$, both $\operatorname{Forb}(H)$ and $\operatorname{Forb}_{ind}(H)$ are hereditary graph properties. Therefore, work on hereditary graph properties can be seen as generalizing the many structure and enumeration results about graph properties of the form $\operatorname{Forb}(H)$ and $\operatorname{Forb}_{ind}(H)$, for instance those appearing in \cite{HPS, KPR, ProSte3, ProSte2, ProSte1}. From this perspective, the study of hereditary graph properties has been a central area of research in extremal combinatorics. There are many results which extend the investigation of hereditary graph properties to other combinatorial structures. Examples of this include \cite{BaBoMo1} for tournaments, \cite{BaBoMo} for oriented graphs and posets, \cite{DN, KNR} for $k$-uniform hypergraphs, and \cite{Ishigami} for colored $k$-uniform hypergraphs. The results in \cite{BM, BM2, PersonSchacht, NR} investigate asymptotic enumeration and structure results for specific classes of $H$-free hypergraphs, which are examples of hereditary properties of hypergraphs. Similarly, the results in \cite{OKTZ} concern specific examples of hereditary properties of digraphs. The results in \cite{MT} for metric spaces are similar in flavor, although they have not been studied explicitly as instances of hereditary properties. Thus, extending the investigation of hereditary graph properties to other combinatorial structures has been an active area of research for many years. From this investigation, patterns have emerged for how to prove these kinds of results, along with a set of standard tools, including extremal results, stability\footnote{This use of the word stability refers to a type of theorem from extremal combinatorics and is unrelated to the model theoretic notion of stability.} theorems, regularity lemmas, supersaturation results, and the hypergraph containers theorem. In various combinations with extremal results, stability theorems, and supersaturation results, Szemer\'{e}di's regularity lemma has played a key role in proving many results in this area, especially those extending results for graphs to other settings. A sampling of these are \cite{ABBM, BBS, ProSte2} for graphs, \cite{AlonYuster} for oriented graphs, \cite{BM, BM2, DN, Ishigami, KNR, NR, PersonSchacht} for hypergraphs, and \cite{MT} for metric spaces. The hypergraph containers theorem, independently developed in \cite{Baloghetal1, saxton-thomason}, has been used in many recent papers in place of the regularity lemma. Examples of this include \cite{BBNLMS, BLPS, KKOT, Stegeretal, Baloghetal1, saxton-thomason} for graphs, \cite{OKTZ} for digraphs, and \cite{BW} for metric spaces. In these papers, the commonalities in the proofs are especially clear. Given an extremal result, there is clear outline for how to prove an approximate enumeration theorem. If on top of this, one can characterize the extremal structures and prove a corresponding stability theorem, then there is a clear outline for how to prove an approximate structure theorem. The goal of this paper is to make these proof outlines formal using generalizations of tools, definitions, and theorems from these papers to the setting of structures in finite relational languages. \subsection{Summary of Results} Given a first-order language $\mathcal{L}$, we say a class $\mathcal{H}$ of $\mathcal{L}$-structures has the \emph{hereditary property} if for all $A\in \mathcal{H}$, if $B$ is a model theoretic substructure of $A$, then $B\in \mathcal{H}$. \begin{definition} Suppose $\mathcal{L}$ is a finite relational language. A \emph{hereditary $\mathcal{L}$-property} is a nonempty class of finite $\mathcal{L}$-structures which has the hereditary property and which is closed under isomorphism. \end{definition} This is the natural generalization of existing notions of hereditary properties of various combinatorial structures. Indeed, for appropriately chosen $\mathcal{L}$, almost all of the results cited so far are for hereditary $\mathcal{L}$-properties, including all hereditary properties of graphs, $k$-uniform hypergraphs, colored $k$-uniform hypergraphs, directed graphs, and posets, as well as the the metric spaces from \cite{MT}. Thus hereditary $\mathcal{L}$-properties are the appropriate objects to study in order to generalize many of the results we are interested in. We now give a description of our results. The precise statements require extensive preliminaries and appear in Section \ref{mainresults}. Fix a finite relational language $\mathcal{L}$ with maximum arity $r\geq 2$. Given a hereditary $\mathcal{L}$-property $\mathcal{H}$, we will define an invariant associated to $\mathcal{H}$, called the \emph{asymptotic density of $\mathcal{H}$}, denoted by $\pi(\mathcal{H})$ (see Definition \ref{pidef}). Our first main result, Theorem \ref{enumeration}, gives an asymptotic enumeration of $\mathcal{H}_n$ in terms of $\pi(\mathcal{H})$, where $\mathcal{H}_n$ denotes the set of elements in $\mathcal{H}$ with domain $[n]$. \begin{theorem3}[Enumeration] Suppose $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. Then the following hold. \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $|\mathcal{H}_n|= \pi(\mathcal{H})^{{n\choose r}+o(n^r)}$. \item If $\pi(\mathcal{H})\leq 1$, then $|\mathcal{H}_n|=2^{o(n^r)}$. \end{enumerate} \end{theorem3} \noindent The tools we use to prove this theorem include a general supersaturation theorem for $\mathcal{L}$-structures (Theorem \ref{GENSUPERSAT}) and a new adaptation of the hypergraph containers theorem to the setting of $\mathcal{L}$-structures (Theorems \ref{version1}). \begin{theorem7}[informal] A hereditary $\mathcal{L}$-property always has a supersaturation theorem. \end{theorem7} \noindent The proof of Theorem \ref{GENSUPERSAT} uses our hypergraph containers theorem for $\mathcal{L}$-structures (Theorem \ref{version1}) and a powerful generalization by Aroskar and Cummings \cite{AroskarCummings} of the triangle removal lemma (Theorem \ref{triangleremoval2}). Our proof strategies for these theorems draw on a series of enumeration results for combinatorial structures which employ the hypergraph containers theorem, namely those in \cite{saxton-thomason, Baloghetal1, OKTZ, BW, Stegeretal}. We will also define generalizations of extremal graphs (Definition \ref{genexdef}) and graph stability theorems (Definition \ref{stabdef}). We will prove that the existence of a stability theorem, along with an understanding of extremal structure, always yield an approximate structure theorem. This result, Theorem \ref{stab}, generalizes arguments appearing in many papers, including \cite{OKTZ, BM, BM2, PersonSchacht, MT}. \begin{theorem5}[informal] Stability theorem + Characterization of extremal structures $\Rightarrow$ Approximate Structure. \end{theorem5} \noindent The main tool used to prove Theorem \ref{stab} is a second adaptation of the hypergraph containers theorem to the setting of $\mathcal{L}$-structures, namely Theorem \ref{COROLLARY2}. Our adaptations of the hypergraph containers theorem, Theorems \ref{version1} and \ref{COROLLARY2}, rely on the original hypergraph containers theorem of \cite{Baloghetal1, saxton-thomason}, the general triangle removal lemma in \cite{AroskarCummings}, and the model theoretic tools developed in this paper. In the last main section of the paper, we consider how our results relate to theorems about discrete metric spaces. Given integers $r\geq 3$ and $n\geq 1$, let $M_r(n)$ be the set of metric spaces with distances all in $[r]$ and underlying set $[n]$. We will reprove structure and enumeration theorems from \cite{MT} using the machinery of this paper, along with combinatorial ingredients from \cite{MT}. We include this example because it demonstrates interesting behavior with regards to the existence of a stability theorem. In particular, we will prove that when $r$ is even, the hereditary property associated to $\bigcup_{n\in \mathbb{N}}M_r(n)$ has a stability theorem in the sense of Definition \ref{stabdef}, but when $r$ is odd, this is not the case. This formalizes an intuitive difference between the even and odd cases observed in \cite{MT}. It is important to note that our results apply to languages with relations of arity larger than two, and to structures with non-symmetric relations. To illustrate this we have included appendices explaining how our results apply to examples in the settings of colored hypergraphs (Appendix \ref{coloredhg}), directed graphs (Appendix \ref{dgsec}), and triangle-free hypergraphs (Appendix \ref{trifreesec}). We now clarify what the results in this paper do and what they do not do. Our main theorem, Theorem \ref{enumeration}, gives an enumeration theorem for a hereditary $\mathcal{L}$-property in terms of its asymptotic density. However, determining the asymptotic density of a specific hereditary $\mathcal{L}$-property is often a hard combinatorial problem which this paper does not address. Similarly, while Theorem \ref{stab} shows that a stability theorem and an understanding of extremal structure implies an approximate structure theorem, proving a specific family $\mathcal{H}$ has a stability theorem and understanding its extremal structures are often difficult problems in practice. These difficulties are not addressed by the results in this paper. The role of this paper is to generalize how extremal results and stability theorems give rise to approximate structure and enumeration theorems. While our proofs are inspired by and modeled on those appearing in \cite{saxton-thomason, Baloghetal1, OKTZ, BW, Stegeretal}, our results are more than just straightforward generalizations of existing combinatorial theorems. We use new tools called $\mathcal{L}_{\mathcal{H}}$-templates (see Section \ref{tildeLstructures}) and an application of the hyergraph containers theorem to a hypergraph whose vertices and edges correspond to certain atomic diagrams (see Theorem \ref{VERSION2}). These technical tools and their appearances in our results are non-obvious and of independent interest from a model theoretic perspective. We also provide a simplified version of the generalized triangle removal lemma appearing in \cite{AroskarCummings}, by noting that a simpler notion of the distance between $\mathcal{L}$-structures may be used. \subsection{Conclusion} The work in this paper is significant from the perspective of combinatorics for three main reasons. First, problems in finite combinatorics are most often approached one by one, and techniques developed for specific structures often do not translate well into other contexts. While this style of approach is necessary to solve problems, it creates the impression that generalization is not possible. This work serves as an example that searching to generalize results and proofs within finite combinatorics can be highly fruitful. Second, this work will save researchers time by allowing them to avoid repeating arguments which now appear here in a general context. Third, we believe this paper gives the correct general framework in which to view these questions, which may aid in finding answers to open problems in the area. This work is also of significance from the model theoretic perspective due to the following connection to logical $0$-$1$ laws. Suppose $\mathcal{L}$ is a finite language, and for each $n$, $F(n)$ is a set of $\mathcal{L}$-structures with domain $[n]$. We say $F:=\bigcup_{n\in \mathbb{N}}F(n)$ has a \emph{$0$-$1$ law} if for every first-order $\mathcal{L}$-sentence $\phi$, the limit $$ \mu(\phi):=\lim_{n\rightarrow \infty}\frac{|\{G\in F(n): G\models \phi\}|}{|F(n)|} $$ exists and is equal to $0$ or $1$. If $F$ has a $0$-$1$ law, then $T_{as}(F):=\{\phi: \mu(\phi)=1\}$ is a complete consistent first-order theory. There are many interesting model theoretic questions related to $0$-$1$ laws and almost sure theories. For instance, it is not well understood in general why some classes of finite structure have $0$-$1$ laws and why others do not. One source of known $0$-$1$ laws are asymptotic structure results from extremal combinatorics. For instance, fix $s \geq 3$ and suppose for each $n$, $F(n)$ is the set of all graphs with vertex set $[n]$ omitting the complete graph $K_s$ on $s$ vertices. In \cite{KPR} Kolaitis, Pr\"{o}mel and Rothschild show $F:=\bigcup_{n\in \mathbb{N}}F(n)$ has a $0$-$1$ law. Their proof relies crucially on first proving asymptotic structure and enumeration theorems for $F$. In particular, they show that if $S(n)$ is the set of $(s-1)$-partite graphs on $[n]$, then $S(n)\subseteq F(n)$ for all $n$ and \begin{align}\label{KPRfact} \lim_{n\rightarrow \infty}\frac{|S(n)|}{|F(n)|}=1. \end{align} They then prove a $0$-$1$ law for $S:=\bigcup_{n\in \mathbb{N}}S(n)$ which combines with (\ref{KPRfact}) to imply $F$ has a $0$-$1$ law. Other asymptotic structure results which imply $0$-$1$ laws include \cite{BM2, OKTZ, PersonSchacht, MT} (for details on how these structure results imply $0$-$1$ laws, see \cite{Koponen}). In these papers, \cite{BM2, OKTZ, PersonSchacht, MT}, the precise structure results (which are needed to prove the $0$-$1$ laws) are proven using \emph{approximate} structure and enumeration theorems as stepping stones. This trend suggests that a systematic understanding of precise structure and enumeration will use some version of this ``approximate version" stepping stone. Therefore, understanding approximate structure and enumeration results from a model theoretic perspective is a necessary step in gaining a general understanding of precise structure and enumeration results, and consequently of the logical $0$-$1$ laws which rely on them. \section{Preliminaries}\label{prelims} Our goal here is to include enough preliminaries so that anyone with a rudimentary knowledge of first-order logic will be able to read this paper. Definitions we expect the reader to understand include: first-order languages, constant and relation symbols, formulas, variables, structures, substructures, satisfaction, and consistency. We refer the reader to \cite{Dave} for these definitions. \subsection{Notation and Setup}\label{notation} In this subsection we fix some notational conventions and definitions. We will use the word ``collection'' to denote either a set or a class. Suppose $\ell \geq 1$ is an integer and $X$ is a set. Let $Perm(\ell)$ be the set of permutations of $[\ell]$. We let $\mathcal{P}(X)$ or $2^X$ denote the power set of $X$. Given a finite tuple $\bar{x}=(x_1,\ldots, x_{\ell})$ and $\mu\in Perm(\ell)$, let $\cup \bar{x} =\{x_1,\ldots, x_{\ell}\}$, $|\bar{x}|=\ell$, and $\mu(\bar{x})=(x_{\mu(1)},\ldots, x_{\mu(\ell)})$. An \emph{enumeration of $X$} is a tuple $\bar{x}=(x_1,\ldots, x_{|X|})$ such that $\cup \bar{x} =X$. Given $x\neq y\in X$, we will write $xy$ as shorthand for the set $\{x,y\}$. Set $$ X^{\underline{\ell}}=\{(x_1,\ldots, x_{\ell}) \in X^{\ell}: x_i\neq x_j\text{ for each }i\neq j\}\quad \hbox{ and }\quad {X\choose \ell}=\{Y\subseteq X: |Y|=\ell\}. $$ Suppose $\mathcal{L}$ is a finite relational first-order language. Let $r_{\mathcal{L}}$ denote the maximum arity of any relation symbol in $\mathcal{L}$. Given a formula $\phi$ and a tuple of variables $\bar{x}$, we write $\phi(\bar{x})$ to denote that the free variables in $\phi$ are all in the set $\cup \bar{x}$. Similarly, if $p$ is a set of formulas, we will write $p(\bar{x})$ if every formula in $p$ has free variables in the set $\cup \bar{x}$. We will sometimes abuse notation and write $\bar{x}$ instead of $\cup \bar{x}$ when it is clear from context what is meant. Suppose $M$ is an $\mathcal{L}$-structure. Then $dom(M)$ denotes the underlying set of $M$, and the \emph{size} of $M$ is $|dom(M)|$. If $\mathcal{L'}\subseteq \mathcal{L}$, $M\upharpoonright_{\mathcal{L}'}$ is the $\mathcal{L}'$-structure with underlying set $dom(M)$ such that for all $\ell\geq 1$, if $\bar{a} \in dom(M)^{\ell}$ and $R$ is an $\ell$-ary relation symbol from $\mathcal{L}'$, then $M\upharpoonright_{\mathcal{L}'}\models R(\bar{a})$ if and only if $M\models R(\bar{a})$. We call $M\upharpoonright_{\mathcal{L}'}$ the \emph{reduct of $M$ to $\mathcal{L}'$}. Given $X\subseteq dom(M)$, $M[X]$ is the $\mathcal{L}$-structure with domain $X$ such that for all $\ell\geq 1$, if $\bar{a}\in X^{\ell}$ and $R$ is an $\ell$-ary relation symbol from $\mathcal{L}$, then $M[X]\models R(\bar{a})$ if and only if $M\models R(\bar{a})$. We call $M[X]$ the $\mathcal{L}$-structure induced by $M$ on $X$. Given a tuple $\bar{a}\in dom(M)^{\ell}$, the \emph{quantifier-free type} of $\bar{a}$ is $$ qftp^M(\bar{a})=\{\phi(x_1,\ldots, x_{\ell}): \text{$\phi(x_1,\ldots, x_{\ell})$ is a quantifier-free $\mathcal{L}$-formula and $M\models \phi(\bar{a})\}$}. $$ If $\bar{x}=(x_1,\ldots, x_{\ell})$ and $p(\bar{x})$ is a set of quantifier-free $\mathcal{L}$-formulas, then $p$ is called a \emph{quantifier-free $\ell$-type} if there is some $\mathcal{L}$-structure $N$ and a tuple $\bar{a}\in dom(N)^{\ell}$ such that $N\models \phi(\bar{a})$ for all $\phi(\bar{x})\in p$. In this case we say $\bar{a}$ \emph{realizes $p$ in $N$}. If there is some $\bar{a} \in dom(N)^{\ell}$ realizing $p$ in $N$, we say $p$ \emph{is realized in $N$}. A quantifier-free $\ell$-type $p(\bar{x})$ is \emph{complete} if for every quantifier-free formula $\phi(\bar{x})$, either $\phi(\bar{x})$ or $\neg \phi(\bar{x})$ is in $p(\bar{x})$. Note that any type of the form $qftp^M(\bar{a})$ is complete. All types and formulas we consider will be quantifier-free, so for the rest of the paper, any use of the words type and formula means quantifier-free type and quantifier-free formula. If $X$ and $Y$ are both $\mathcal{L}$-structures, let $X\subseteq_{\mathcal{L}} Y$ denote that $X$ is a $\mathcal{L}$-substructure of $Y$. Given an $\mathcal{L}$-structure $H$, we say that $M$ is $H$-free if there is no $A\subseteq_{\mathcal{L}} M$ such that $A\cong_{\mathcal{L}}H$. Suppose $\mathcal{H}$ is a collection of $\mathcal{L}$-structures. We say $M$ is \emph{$\mathcal{H}$-free} if $M$ is $H$-free for all $H\in \mathcal{H}$. For each positive integer $n$, let $\mathcal{H}(n)$ denote the collection of all elements in $\mathcal{H}$ of size $\ell$, and let $\mathcal{H}_n$ denote the set of elements in $\mathcal{H}$ with domain $[n]$. $\mathcal{H}$ is \emph{trivial} if there is $N$ such that $\mathcal{H}(n)=\emptyset$ for all $n\geq N$. Otherwise $\mathcal{H}$ is \emph{non-trivial}. We now define a modified version of the traditional type space, which is appropriate for working with families of finite structures instead of with complete first-order theories. Given $\bar{x}=(x_1,\ldots, x_{\ell})$, an $\ell$-type $p(\bar{x})$ is \emph{proper} if it contains the formulas $x_i\neq x_j$ for each $i\neq j$. \begin{definition} Suppose $\mathcal{F}$ is a collection of $\mathcal{L}$-structures and $\ell \geq 1$ is an integer. Define $S_{\ell}(\mathcal{F})$ to be the set of all complete, proper, quantifier-free $\ell$-types which are realized in some element of $\mathcal{F}$. Let $S_{\ell}(\mathcal{L})$ denote the set of all complete, proper, quantifier-free $\ell$-types. \end{definition} \noindent We would like to emphasize some important differences between this and the usual type space. First, the elements of these type spaces are proper and contain only quantifier-free formulas. Second, these type spaces are defined relative to families of finite structures instead of complete first-order theories. It will at times be convenient to expand our languages to contain constant symbols naming elements of the structures under consideration. If $V$ is a set, let $C_V$ denote the set of constant symbols $\{c_v: v\in V\}$. Given $\bar{v}=(v_1,\ldots, v_{\ell}) \in V^{\ell}$, let $c_{\bar{v}}=(c_{v_1},\ldots, c_{v_{\ell}})$. Suppose $M$ is an $\mathcal{L}$-structure. The \emph{diagram of $M$}, denoted $Diag(M)$, is the following set of sentences in the language $\mathcal{L}\cup C_{dom(M)}$. $$ Diag(M)=\{\phi(c_{\bar{a}}): \phi(\bar{x}) \text{ is a quantifier-free $\mathcal{L}$-formula, } \cup \bar{a} \subseteq dom(M)\text{, and }M\models \phi(\bar{a})\}. $$ \noindent If $A\subseteq dom(M)$, the \emph{diagram of $A$ in $M$} is the following set of sentences in the language $\mathcal{L}\cup C_{A}$. $$ Diag^M(A)=\{\phi(c_{\bar{a}}): \phi(\bar{x}) \text{ is a quantifier-free $\mathcal{L}$-formula, } \cup \bar{a} \subseteq A\text{, and }M\models \phi(\bar{a})\}. $$ Observe that if $A=\{a_1,\ldots, a_r\}\subseteq dom(M)$ and $p(\bar{x})\in S_r(\mathcal{L})$ is such that $p(\bar{x})=qftp^M(a_1,\ldots, a_r)$, then $Diag^M(A)=p(c_{a_1},\ldots, c_{a_r})$. Given a set of constants $C$, a collection of $\mathcal{L}$-structures $\mathcal{F}$, and $\ell\geq 1$, set $$S_{\ell}(C)=\{p(\bar{c}): p(\bar{x})\in S_{\ell}(\mathcal{L})\text{ and } \bar{c}\in C^{\underline{\ell}}\}\quad \hbox{ and }\quad S_{\ell}(C, \mathcal{F})=\{p(\bar{c}): p(\bar{x})\in S_{\ell}(\mathcal{F})\text{ and } \bar{c}\in C^{\underline{\ell}}\}. $$ We would like to emphasize that if $p(\bar{c})\in S_{\ell}(C)$, then $\bar{c}\in C^{\underline{\ell}}$ is a tuple of $\ell$ \emph{distinct} constants. Note that by this definition, if $A\in {dom(M)\choose \ell}$, then $Diag^M(A)\in S_{\ell}(C_{dom(M)})$. \subsection{Hypergraph Containers Theorem.}\label{containerssec} In this section we state a version of the hypergraph containers theorem, which was independently developed by Balogh-Morris-Samotij in \cite{Baloghetal1} and by Saxton-Thomason in \cite{saxton-thomason}. The particular statement we use, Theorem \ref{containers} below, is a simplified version of Corollary 3.6 in \cite{saxton-thomason}. We will use Theorem \ref{containers} directly in Section \ref{VERSION2pf}. We also think it will be useful for the reader to compare it to the versions for $\mathcal{L}$-structures stated in Section \ref{mainresults} (Theorem \ref{version1} and Corollary \ref{COROLLARY2}). We begin with some definitions. Recall that if $s\geq 2$ is an integer, an \emph{$s$-uniform hypergraph} is a pair $(V,E)$ where $V$ is a set of \emph{vertices} and $E\subseteq {V\choose s}$ is a set of \emph{edges}. Suppose $H$ is an $s$-uniform hypergraph. Then $V(H)$ and $E(H)$ denote the vertex and edge sets of $G$ respectively. We set $v(H)=|V(H)|$ and $e(H)=|E(H)|$. Given $X\subseteq V(H)$, $H[X]$ is the hypergraph $(X,E\cap{V(H)\choose s})$. If $v(H)$ is finite, then the \emph{average degree of $H$} is $d=e(H)s/v(H)$. \begin{definition}\label{containersdef} Suppose $s\geq 2$, $H$ is a finite $s$-uniform hypergraph with $n$ vertices and $\tau>0$. \begin{itemize} \item For every $\sigma \subseteq V(H)$, the \emph{degree of $\sigma$ in $H$} is $d(\sigma) = |\{e\in E(H): \sigma \subseteq e\}|$. \item Given $v\in V(H)$ and $j\in [s]$, set $d^{(j)}(v)=\max \{ d(\sigma): v\in \sigma \subseteq V(H), |\sigma|=j\}$. \item If $d>0$, then for each $j\in [s]$, define $\delta_j=\delta_j(\tau)$ to satisfy the equation $$ \delta_j \tau^{j-1}nd=\sum_{v\in V(H)}d^{(j)}(v), \qquad \hbox{ and set }\qquad \delta(H,\tau)=2^{{s\choose 2}-1}\sum_{j=2}^{s} 2^{-{j-1\choose 2}}\delta_j. $$ If $d=0$, set $\delta(H,\tau)=0$. $\delta(H,\tau)$ is called the \emph{co-degree function}. \end{itemize} \end{definition} \noindent Unless otherwise stated, $n$ is always a positive integer. \begin{theorem}[\bf{Saxton-Thomason \cite{saxton-thomason}}] \label{containers} Let $H$ be an $\ell$-uniform hyptergraph with a vertex set $V$ of size $n$. Suppose $0<\epsilon, \tau<\frac{1}{2}$ and $\tau$ satisfies $\delta(H,\tau)\leq \epsilon/12s!$. Then there exists a constant $c=c(s)$ and a collection $\mathcal{C}\subseteq \mathcal{P}(V)$ such that the following hold. \begin{enumerate}[(i)] \item For every independent set $I$ in $H$, there exists $C\in \mathcal{C}$ such that $I\subseteq C$. \item For all $C\in \mathcal{C}$, we have $e(H[C])\leq \epsilon e(G)$, and \item $\log |\mathcal{C}| \leq c\log(1/\epsilon) n\tau \log (1/\tau)$. \end{enumerate} \end{theorem} \subsection{Distance between first-order structures} In this section we define a notion of distance between finite first-order structures. The following is based on definitions in \cite{AroskarCummings}. \begin{definition} Suppose $\mathcal{L}$ is a first-order language, $B$ is a finite $\mathcal{L}$-structure of size $\ell$, and $M$ is a finite $\mathcal{L}$-structure of size $L$. \begin{itemize} \item The \emph{set of copies of $B$ in $M$} is $\textnormal{cop}(B,M)=\{A: A\subseteq_{\mathcal{L}}M \text{ and }A\cong_{\mathcal{L}}B\}$. \item The \emph{induced structure density of $B$ in $M$} is $\textnormal{prob}(B,M)=|\textnormal{cop}(B,M)|/{L\choose \ell}$ \item If $\mathcal{B}$ is a set of finite $\mathcal{L}$-structures, let $$ \textnormal{cop}(\mathcal{B}, M)=\bigcup_{B\in \mathcal{B}}\textnormal{cop}(B,M)\qquad \hbox{ and }\qquad \textnormal{prob}(\mathcal{B},M)=\max\{p(B,M): B\in \mathcal{B}\}. $$ \end{itemize} If $\mathcal{B}$ is a \emph{class} of finite $\mathcal{L}$-structures, define $\textnormal{cop}(\mathcal{B}, M)=\textnormal{cop}(\mathcal{B}',M)$ and $\textnormal{prob}(\mathcal{B},M)=\textnormal{prob}(\mathcal{B}',M)$, where $\mathcal{B}'$ is any set containing one representative of each isomorphism type in $B$. \end{definition} We now state our definition for the distance between two finite first-order structures. It is a simplified version of the distance notion appearing in \cite{AroskarCummings}. We will discuss the relationship between the two notions in Section \ref{rphrem}. \begin{definition}\label{deltaclosedef1} Let $\mathcal{L}$ be a finite relational first-order language with $r_{\ell}=r\geq 2$. Suppose $M$ and $N$ are two finite $\mathcal{L}$-structures with the same underlying set $V$ of size $n$. Let \begin{align*} \textnormal{diff}(M,N)&=\{A\in {V\choose r}: \text{ for some enumeration $\bar{a}$ of $A$, }qftp^M(\bar{a})\neq qftp^N(\bar{a})\}\text{ and }\\ \textnormal{dist}(M,N)&=\frac{|\textnormal{diff}(M,N)|}{{n\choose r}} \end{align*} We say that $M$ and $N$ are \emph{$\delta$-close} if $\textnormal{dist}(M,N)\leq \delta$. \end{definition} \noindent Observe that in the notation of Definition \ref{deltaclosedef1}, $\textnormal{diff}(M,N)=\{A\in {V\choose r}: Diag^M(A)\neq Diag^N(A)\}$. \subsection{Facts about hereditary properties} Suppose $\mathcal{L}$ is a finite relational language. In this subsection we state some well known facts about hereditary $\mathcal{L}$-properties. First we recall that hereditary $\mathcal{L}$-properties are the same as families of structures with forbidden configurations. This fact will be used throughout the chapter. \begin{definition} If $\mathcal{F}$ is a collection of finite $\mathcal{L}$-structures, let $\operatorname{Forb}(\mathcal{F})$ be the class of all finite $\mathcal{L}$-structures which are $\mathcal{F}$-free. \end{definition} It is easy to check that for any collection $\mathcal{F}$ of finite $\mathcal{L}$-structures, $\operatorname{Forb}(\mathcal{F})$ is a hereditary $\mathcal{L}$-property. The converse to this statement is also true in the sense of Observation \ref{HP} below. This fact is standard, but we include a proof for completeness. \begin{observation}\label{HP} If $\mathcal{H}$ is a hereditary $\mathcal{L}$-property, then there is a class of finite $\mathcal{L}$-structures $\mathcal{F}$ which is closed under isomorphism and such that $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$. \end{observation} \begin{proof} Let $\mathcal{F}$ be the class of all finite $\mathcal{L}$-structures $F$ such that $\textnormal{prob}(F, \mathcal{H})=0$. Clearly $\mathcal{F}$ is closed under $\mathcal{L}$-isomorphism. We show $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$. Suppose $M\in \mathcal{H}$ but $M\notin \operatorname{Forb}(\mathcal{F})$. Then there is some $F'\subseteq_{\mathcal{L}}M$ and $F\in \mathcal{F}$ such that $F\cong_{\mathcal{L}}F'$. Since $\mathcal{F}$ is closed under $\mathcal{L}$-isomorphism, $F'\in \mathcal{F}$. But then $M\in \mathcal{H}$ and $F'\subseteq_{\mathcal{L}}M$ implies $F'\notin \mathcal{F}$ by definition of $\mathcal{F}$, a contradiction. Conversely, suppose $M\in \\operatorname{Forb}(\mathcal{F})$ but $M\notin \mathcal{H}$. Because $M$ is $\mathcal{F}$-free, we must have $M\notin \mathcal{F}$. By definition of $\mathcal{F}$, this implies there is some $M'\in \mathcal{H}$ such that $M\subseteq_{\mathcal{L}}M'$. Because $\mathcal{H}$ has the hereditary property, this implies $M\in \mathcal{H}$, a contradiction. \end{proof} A sentence $\phi$ is universal if it is of the form $\forall \bar{x} \psi(\bar{x})$ where $\psi(\bar{x})$ is quantifier-free. The following well known fact is another reason hereditary $\mathcal{L}$-properties are natural objects of study. \begin{observation} $\mathcal{H}$ is a hereditary $\mathcal{L}$-property if and only if there is a set of universal sentences $\Phi$ such that $\mathcal{H}$ is the class of all finite models of $\Phi$. \end{observation} \begin{comment} \begin{proof} Suppose first $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. Let $\mathcal{F}$ be as in Observation \ref{HP} such that $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$. Let $\{F_i: i\in \omega\}$ contain one representative of each isomorphism class in $\mathcal{F}$, and let $\theta_i$ be the sentence saying $\forall \bar{x} (\bar{x} \ncong_{\mathcal{L}} F_i)$. Then clearly $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$ is the class of all finite models of $\Phi:=\{\theta_i: i\in \omega\}$. Conversely, suppose $\Phi$ is a set of universal sentences and $\mathcal{H}$ is the class of all finite models of $\Phi$. Clearly $\mathcal{H}$ is closed under isomorphism. Suppose now $B\in \mathcal{H}$ and $A\subseteq_{\mathcal{L}}B$. Let $\phi \in \Phi$. Then $B\models \phi$ by assumption. Since $\phi$ is universal, it is preserved under substructures, so $A\models \phi$. Thus $A\models \phi$ for all $\phi \in \Phi$, so $A\in \mathcal{H}$. This shows $\mathcal{H}$ has the hereditary property, and thus is a hereditary $\mathcal{L}$-property. \end{proof} \end{comment} The proof of this is straightforward using Observation \ref{HP}, the fact that $\operatorname{Forb}(\mathcal{F})$ can be axiomatized using universal sentences for any class $\mathcal{F}$ of finite $\mathcal{L}$-structures, and the fact that universal sentences are preserved under taking substructures (see the \L os-Tarski Theorem in \cite{hodges}). \section{$\mathcal{L}_{\mathcal{H}}$-structures}\label{tildeLstructures} {\bf From now on, $\mathcal{L}$ is a fixed finite relational language and $r:=r_{\mathcal{L}}\geq 2$}. For this section, $\mathcal{H}$ is a nonempty collection of finite $\mathcal{L}$-structures. In this section we introduce a language $\mathcal{L}_{\mathcal{H}}$ associated to $\mathcal{L}$ and $\mathcal{H}$. Structures in this new language play key roles in our main theorems. \begin{definition}\label{LHdef} Define $\mathcal{L}_{\mathcal{H}}=\{R_p(\bar{x}): p(\bar{x})\in S_r(\mathcal{H})\}$ to be the relational language with one $r$-ary relation for each $p(\bar{x})$ in $S_r(\mathcal{H})$. \end{definition} The goal of this section is to formalize how an $\mathcal{L}_{\mathcal{H}}$-structure $M$ with the right properties can serve as a ``template'' for building $\mathcal{L}$-structures with the same underlying set as $M$. We now give an example of a hereditary property and its corresponding auxiliary language as in Definition \ref{LHdef}. \begin{example}\label{ex1} To avoid confusion, we will use $\mathcal{P}$ to refer to specific hereditary properties in example settings. Suppose $\mathcal{L}=\{R_1(x,y), R_2(x,y), R_3(x,y)\}$. Let $\mathcal{P}$ be the class of all finite metric spaces with distances in $\{1,2,3\}$, considered as $\mathcal{L}$-structures in the natural way (i.e. $R_i(x,y)$ if and only if $d(x,y)=i$). It is easy to see that $\mathcal{P}$ is a hereditary $\mathcal{L}$-property. Observe that since $r_{\mathcal{L}}=2$, $\mathcal{L}_{\mathcal{P}}=\{R_p(x,y): p(x,y)\in S_2(\mathcal{P})\}$. For each $i\in [3]$, set $$ q_i(x,y):=\{x\neq y\}\cup \{R_i(x,y), R_i(y,x)\}\cup \{\neg R_j(x,y), \neg R_j(y,x): j\neq i\}, $$ and let $p_i(x,y)$ be the unique quantifier-free $2$-type containing $q_i(x,y)$. Informally, the type $p_i(x,y)$ says ``the distance between $x$ and $y$ is equal to $i$.'' We leave it as an exercise to the reader that $S_2(\mathcal{P})=\{p_i(x,y): i\in [3]\}$ (recall $S_2(\mathcal{P})$ consists of \emph{proper} types). Thus $\mathcal{L}_{\mathcal{P}}=\{R_{p_i}(x,y): i\in [3]\}$. \end{example} Observe that in an arbitrary $\mathcal{L}_{\mathcal{H}}$-structure, the relation symbols in $\mathcal{L}_{\mathcal{H}}$ may have nothing to do with the properties of the type space $S_r(\mathcal{H})$. For instance, in the notation of Example \ref{ex1}, we can easily build an $\mathcal{L}_{\mathcal{P}}$-structure $M$ so that for some $a,b\in dom(M)$, $M\models R_{p_1}(a,b) \wedge \neg R_{p_1}(b,a)$, even though $p_1(x,y)=p_1(y,x)$ in $S_2(\mathcal{P})$. This kind of behavior will be undesirable for various technical reasons. We now define the class of $\mathcal{L}_{\mathcal{H}}$-structures which are most nicely behaved for our purposes, and where in particular, this bad behavior does not happen. \begin{definition} An $\mathcal{L}_{\mathcal{H}}$-structure $M$ with domain $V$ is \emph{complete} if for all $A\in {V\choose r}$ there is an enumeration $\bar{a}$ of $A$ and $R_p\in \mathcal{L}_{\mathcal{H}}$ such that $M\models R_p(\bar{a})$. \end{definition} \begin{definition}\label{templatedef} An $\mathcal{L}_{\mathcal{H}}$-structure $M$ with domain $V$ is an \emph{$\mathcal{L}_{\mathcal{H}}$-template} if it is complete and the following hold. \begin{enumerate} \item If $p(\bar{x})\in S_r(\mathcal{H})$ and $\bar{a}\in V^r\setminus V^{\underline{r}}$, then $M\models \neg R_p(\bar{a})$. \item If $p(\bar{x}), p'(\bar{x})\in S_r(\mathcal{H})$ and $\mu\in Perm(r)$ are such that $p(\bar{x})=p'(\mu(\bar{x}))$, then for every $\bar{a}\in V^{\underline{r}}$, $M\models R_p(\bar{a})$ if and only if $M\models R_{p'}(\mu(\bar{a}))$. \end{enumerate} \end{definition} \noindent The idea is that $\mathcal{L}_{\mathcal{H}}$-templates are the $\mathcal{L}_{\mathcal{H}}$-structures which most accurately reflect the properties of $S_r(\mathcal{H})$. \begin{example}\label{ex2} Let $\mathcal{L}$ and $\mathcal{P}$ be as in Example \ref{ex1}. Let $G$ be the $\mathcal{L}_{\mathcal{P}}$-structure with domain $V=\{u,v,w\}$ such that $G\models \bigwedge_{i=1}^3 (R_{p_i}(u,v) \wedge R_{p_i}(v,u))$, \begin{align*} G&\models \neg R_{p_3}(w,v) \wedge \neg R_{p_3}(v,w) \wedge \bigwedge_{i=1}^2 (R_{p_i}(w,v) \wedge R_{p_i}(v,w)),\\ G&\models R_{p_1}(w,u) \wedge R_{p_1}(w,u) \wedge \bigwedge_{i=2}^3 (\neg R_{p_i}(w,u) \wedge \neg R_{p_i}(u,w)), \end{align*} and for all $x\in V$, $G\models \bigwedge_{i=1}^3 \neg R_{p_i}(x,x)$. We leave it to the reader to verify $G$ is a $\mathcal{L}_{\mathcal{P}}$-template. \end{example} While $\mathcal{L}_{\mathcal{H}}$-templates are important for the main results of this paper, many of the definitions and facts in the rest of this section will be presented for $\mathcal{L}_{\mathcal{H}}$-structures with weaker assumptions. \subsection{Choice functions and subpatterns} In this subsection, we give crucial definitions for how we can use $\mathcal{L}_{\mathcal{H}}$-structures to build $\mathcal{L}$-structures. \begin{definition}\label{chdef} Suppose $M$ is an $\mathcal{L}_{\mathcal{H}}$-structure with domain $V$. \begin{enumerate} \item Given $A\in {V\choose r}$, the \emph{set of choices for $A$ in $M$} is $$ Ch_M(A)=\{p(c_{a_1},\ldots, c_{a_r})\in S_r(C_V,\mathcal{H}): \{a_1,\ldots, a_r\}=A \text{ and }M\models R_p(a_1,\ldots, a_r)\}. $$ \item A \emph{choice function for $M$} is a function $\chi: {V\choose r}\rightarrow S_r(C_V,\mathcal{H})$ such that for each $A\in {V\choose r}$, $\chi(A)\in Ch_M(A)$. Let $Ch(M)$ denote the set of all choice functions for $M$. \end{enumerate} \end{definition} \noindent In the notation of Definition \ref{chdef}, note $Ch(M)\neq \emptyset$ if and only if $Ch_M(A)\neq \emptyset$ for all $A\in {V\choose r}$. Observe that $Ch_M(A)\neq \emptyset$ for all $A\in {V\choose r}$ if and only if $M$ is complete. Therefore $Ch(M)\neq \emptyset$ if and only if $M$ is complete. \begin{example}\label{ex3} Recall that if $x$ and $y$ are distinct elements of a set, then $xy$ is shorthand for the set $\{x,y\}$. Let $\mathcal{L}$, $\mathcal{P}$, $V$, and $G$ be as in Example \ref{ex2}. Note that $C_V=\{c_u, c_v,c_w\}$ and $$ S_2(C_V, \mathcal{P})= \{p_i(c_u,c_v): i\in [3]\}\cup \{p_i(c_v,c_w): i\in [3]\}\cup \{p_i(c_u,c_w): i\in [3]\}. $$ By definition of $G$, $Ch_G(uv)=\{p_1(c_u,c_v), p_2(c_u,c_v), p_3(c_u,c_v)\}$, $Ch_G(vw)=\{p_1(c_v,c_w), p_2(c_v,c_w)\}$, and $Ch_G(uw)=\{p_1(c_u,c_w)\}$. Therefore $Ch(G)$ is the set of functions $$ \chi:\{uv, vw, uw\}\rightarrow \{p_i(c_u,c_v): i\in [3]\}\cup \{p_i(c_v,c_w): i\in [3]\}\cup \{p_i(c_u,c_w): i\in [3]\} $$ with the properties that $\chi(uv)\in \{p_1(c_u,c_v), p_2(c_u,c_v), p_3(c_u,c_v)\}$, $\chi(vw)\in \{p_1(c_v,c_w), p_2(c_v,c_w)\}$ and $\chi(uw)=p_1(c_u,c_w)$. Clearly this shows $|Ch(G)|=|Ch_G(uv)||Ch_G(vw)||Ch_G(uw)|=6$. \end{example} The following observation is immediate from the definition of $\mathcal{L}_{\mathcal{H}}$-template. \begin{observation}\label{temob} If $M$ is an $\mathcal{L}_{\mathcal{H}}$-template with domain $V$, then for all $\bar{a}\in V^r$ and $R_p\in \mathcal{L}_{\mathcal{H}}$, $M\models R_p(\bar{a})$ if and only if $|\cup \bar{a}|=r$ and $p(c_{\bar{a}})\in Ch_M(\cup \bar{a})$. \end{observation} \noindent The following fact is one reason why $\mathcal{L}_{\mathcal{H}}$-templates are convenient. \begin{proposition}\label{nice} Suppose $M_1$ and $M_2$ are $\mathcal{L}_{\mathcal{H}}$-templates with domain $V$ such that for all $A\in {V\choose r}$, $Ch_{M_1}(A)=Ch_{M_2}(A)$. Then $M_1$ and $M_2$ are the same $\mathcal{L}_{\mathcal{H}}$-structure. \end{proposition} \begin{proof} We show that for all $\bar{a}\in V^r$ and $R_p\in \mathcal{L}_{\mathcal{H}}$, $M_1\models R_p(\bar{a})$ if and only if $M_2\models R_p(\bar{a})$. Fix $\bar{a} \in V^r$ and $R_p\in \mathcal{L}_{\mathcal{H}}$. Suppose first that $|\cup \bar{a}|<r$. By part (1) of Definition \ref{templatedef}, $M_1\models \neg R_p(\bar{a})$ and $M_2\models \neg R_p(\bar{a})$. So assume $|\cup \bar{a}|=r$. By Observation \ref{temob}, $M_1\models R_p(\bar{a})$ if and only if $p(c_{\bar{a}})\in Ch_{M_1}(\cup \bar{a})$ and $M_2\models R_p(\bar{a})$ if and only if $p(c_{\bar{a}})\in Ch_{M_2}(\cup \bar{a})$. Since $Ch_{M_1}(\cup \bar{a})=Ch_{M_2}(\cup \bar{a})$, this implies $M_1\models R_p(\bar{a})$ if and only if $M_2\models R_p(\bar{a})$. \end{proof} \noindent The next example shows Proposition \ref{nice} can fail when we are not dealing with $\mathcal{L}_{\mathcal{H}}$-templates. \begin{example}\label{ex4} Let $\mathcal{L}$, $\mathcal{P}$, $V$, and $G$ be as in Example \ref{ex3}. Let $G'$ be the $\mathcal{L}_{\mathcal{P}}$-structure with domain $V$ which agrees with $G$ on $V^2\setminus \{(v,u), (w,v), (w,u)\}$ and where $$ G'\models \bigwedge_{i=1}^3 (\neg R_{p_i}(v,u)\wedge \neg R_{p_i}(w,v)\wedge \neg R_{p_i}(w,u)). $$ We leave it to the reader to check that for all $xy\in {V\choose 2}$, $Ch_{G'}(xy)=Ch_G(xy)$. However, $G$ and $G'$ are distinct $\mathcal{L}_{\mathcal{P}}$-structures because, for instance, $G\models R_{p_1}(v,u)$ while $G'\models \neg R_{p_1}(v,u)$. Observe that $G'$ is not an $\mathcal{L}_{\mathcal{P}}$-template because $G'\models R_{p_1}(u,v)\wedge \neg R_{p_1}(v,u)$ while $p_1(x,y)=p_1(y,x)$. \end{example} \noindent The next definition shows how choice functions give rise to $\mathcal{L}$-structures. \begin{definition} Suppose $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure with domain $V$, $N$ is an $\mathcal{L}$-structure such that $dom(N)\subseteq V$, and $\chi\in Ch(M)$. \begin{enumerate} \item $N$ is a \emph{$\chi$-subpattern} of $M$, denoted $N\leq_{\chi}M$, if for every $A\in {dom(N)\choose r}$, $\chi(A)=Diag^N(A)$. \item $N$ is a \emph{full $\chi$-subpattern of $M$}, denoted $N\unlhd_{\chi} M$, if $N\leq_{\chi}M$ and $dom(N)=V$. \end{enumerate} When $N\unlhd_{\chi} M$, we say $\chi$ \emph{chooses $N$}. We say $N$ is a \emph{subpattern of $M$}, denoted $N\leq_p M$, if $N\leq_{\chi}M$ for some choice function $\chi$ for $M$. We say $N$ is a \emph{full subpattern of $M$}, denoted $N\unlhd_p M$, if $N\unlhd_{\chi}M$ for some choice function $\chi$ for $M$. The subscript in $\leq_p$ and $\unlhd_p$ is for ``pattern.'' \end{definition} \begin{observation}\label{ob00} Suppose $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure, $\chi\in Ch(M)$, and $G$ is an $\mathcal{L}$-structure such that $G\unlhd_{\chi}M$. If $G'$ is another $\mathcal{L}$-structure such that $G'\unlhd_{\chi}M$, then $G$ and $G'$ are the same $\mathcal{L}$-structure. If $\chi'\in Ch(M)$ satisfies $G\unlhd_{\chi'}M$, then $\chi=\chi'$. \end{observation} \begin{proof} By definition, $G\unlhd_{\chi}M$ and $G'\unlhd_{\chi}M$ imply that $Diag^G(A)=\chi(A)=Diag^{G'}(A)$ for all $A\in {V\choose r}$. This implies $$ Diag(G)=\bigcup_{A\in {V\choose r}}Diag^G(A)=\bigcup_{A\in {V\choose r}}Diag^{G'}(A)=Diag(G'), $$ which clearly implies $G$ and $G'$ are the same $\mathcal{L}$-structure. Similarly, $G\unlhd_{\chi}M$ and $G\unlhd_{\chi'}M$ imply that for all $A\in {V\choose r}$, $\chi(A)=Diag^G(A)=\chi'(A)$. Thus $\chi=\chi'$. \end{proof} \begin{example}\label{ex3} Let $\mathcal{L}$, $\mathcal{P}$, $V$ and $G$ be as in Example \ref{ex2}. We give two examples of subpatterns of $G$. Let $\chi$ be the function from ${V\choose 2}\rightarrow S_2(C_V,\mathcal{P})$ defined by $\chi(uv)=p_1(c_u,c_v)$, $\chi(vw)=p_2(c_v,c_w)$, and $\chi(uw)=p_1(c_u,c_w)$. Clearly $\chi$ is a choice function for $G$. Let $H$ be the $\mathcal{L}$-structure with domain $V$ such that $H\models p_1(u,v)\cup p_2(v,w)\cup p_1(u,w)$. Then by definition of $H$, $Diag^H(uv)=p_1(c_u,c_v)$, $Diag^H(vw)=p_2(v,w)$ and $Diag^H(uw)=p_1(c_u,c_w)$. In other words, $H\leq_{\chi}G$. Since $dom(H)=dom(G)=V$, $H\unlhd_{\chi} G$. Note that $H$ is a metric space, that is, $H\in \mathcal{P}$. Let $\chi'$ be the function from ${V\choose 2}\rightarrow S_2(C_V,\mathcal{P})$ defined by $\chi'(uv)=p_3(c_u,c_v)$, $\chi'(vw)=p_1(c_v,c_w)$, and $\chi'(uw)=p_1(c_u,c_w)$. Clearly $\chi'$ is a choice function for $G$. Let $H'$ be the $\mathcal{L}$-structure with domain $V$ such that $H'\models p_3(u,v)\cup p_1(v,w)\cup p_1(u,w)$. Then as above, it is easy to see that $H'\leq_{\chi'} G$, and since $dom(H')=V$, $H'\unlhd_{\chi'} G$. However, $H'$ is \emph{not} a metric space, that is, $H'\notin \mathcal{P}$. \end{example} This example demonstrates although $\mathcal{L}_{\mathcal{H}}$-templates are well behaved in certain ways, an $\mathcal{L}_{\mathcal{H}}$-template can have full subpatterns that are not in $\mathcal{H}$. We will give further definitions to address this in Section \ref{temsec}. \subsection{Errors and counting subpatterns} In this subsection we characterize when an $\mathcal{L}_{\mathcal{H}}$-structure has the property that every choice function gives rise to a subpattern. This will be important for counting subpatterns of $\mathcal{L}_{\mathcal{H}}$-structures. \begin{definition} Given $r< \ell < 2r$, an \emph{error of size $\ell$} is a complete $\mathcal{L}_{\mathcal{H}}$-structure $M$ of size $\ell$ with the following properties. There are $\bar{a}_1, \bar{a}_2\in dom(M)^{\underline{r}}$ such that $dom(M)=\cup \bar{a}_1\bigcup \cup \bar{a}_2$ and for some $p_1(\bar{x}), p_2(\bar{x})\in S_r(\mathcal{H})$, $M\models R_{p_1}(\bar{a}_1)\wedge R_{p_2}(\bar{a}_2)$ but $p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2})$ is unsatisfiable. \end{definition} \begin{example}\label{errorex} Let $\mathcal{L}=\{E(x,y,z), R_1(x,y), R_2(x,y), R_3(x,y)\}$ consist of one ternary relation $E$ and three binary relations $R_1,R_2,R_3$. Suppose $\mathcal{P}$ is the class of all finite $\mathcal{L}$-structures $M$ such that the restriction of $M$ to $\{R_1,R_2,R_3\}$ is a metric space with distances in $\{1,2,3\}$ (we put no restrictions on how $E$ must behave). For $i\in [3]$, let $p_i(x,y)$ be the quantifier-free $\{R_1,R_2,R_3\}$-type from Examples \ref{ex1}, \ref{ex2}, and \ref{ex3}. Let $\bar{x}=(x_1,x_2,x_3)$ and set $q_0(\bar{x})=\{E(x_i,x_j,x_k): \{x_i,x_j,x_k\}\subseteq \{x_1,x_2,x_3\}\}$. Then set $q_1(\bar{x})$ and $q_1(\bar{x})$ to be the complete quantifier-free types satisfying the following. \begin{align*} q_0(\bar{x})\cup p_1(x_1,x_2)\cup p_1(x_1,x_3)\cup p_1(x_2,x_3)&\subseteq q_1(\bar{x})\text{ and }\\ q_0(\bar{x})\cup p_2(x_1,x_2)\cup p_1(x_1,x_3)\cup p_1(x_2,x_3)&\subseteq q_2(\bar{x}). \end{align*} It is easy to check that $q_i(\bar{x})\in S_3(\mathcal{P})$ for $i=1,2$. Note $q_1$ and $q_2$ agree about how $E$ behaves, but disagree on how the binary relations in $\mathcal{L}$ behave. Let $V=\{t,u,v,w\}$ be a set of size $4$. Choose $G$ to be the $\mathcal{L}_{\mathcal{P}}$-structure which satisfies $G\models R_{q_1}(x,y,z)$ if and only if $x$, $y$ and $z$ are distinct, $G\models R_{q_2}(x,y,z)$ if and only if $(x,y,z)=(t,u,v)$, and $G\models \neg R_{q}(x,y,z)$ for all $q\in S_3(\mathcal{P})\setminus \{p_1,p_2\}$. By construction, $G$ is a complete $\mathcal{L}_{\mathcal{P}}$-structure. Let $\bar{a}_1=(u,v,w)$ and $\bar{a}_2=(u,v,t)$. Then $dom(G)=\cup \bar{a}_1\bigcup \cup \bar{a}_2$ and $G\models R_{q_1}(\bar{a}_1)\wedge R_{q_2}(\bar{a}_2)$. However, $p_1(c_u,c_v)\subseteq q_1(c_u,c_v,c_w)=q_1(c_{\bar{a}_1})$ implies $q_1(c_{\bar{a}_1})$ contains the formula $R_1(c_u,c_v)$ while $p_2(c_u,c_v)\subseteq q_2(c_u,c_v,c_w)=q_2(c_{\bar{a}_2})$ implies $q_2(c_{\bar{a}_2})$ contains the formula $\neg R_1(c_u,c_v)$ . Therefore $q_1(c_{\bar{a}_1})\cup q_2(c_{\bar{a}_2})$ is unsatisfiable, and $G$ is an error of size $4$. \end{example} \begin{definition} Let $\mathcal{E}$ be the class of $\mathcal{L}_{\mathcal{H}}$-structures which are errors of size $\ell$ for some $r<\ell<2r$. \end{definition} An $\mathcal{L}_{\mathcal{H}}$-structure $M$ is \emph{error-free} if it is $\mathcal{E}$-free. Error-free $\mathcal{L}_{\mathcal{H}}$-structures will be important for the following reason. \begin{proposition}\label{Lrandom} Suppose $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure with domain $V$. Then $M$ is error-free if and only if for every $\chi\in Ch(M)$, there is an $\mathcal{L}$-structure $N$ such that $N\unlhd_{\chi}M$. \end{proposition} \begin{proof} Suppose first that there exists a choice function $\chi: {V\choose r}\rightarrow S_r(C_V,\mathcal{H})$ such that there are no $\chi$-subpatterns of $M$. This means $\Gamma :=\bigcup_{A\in {V\choose r}} \chi(A)$ is not satisfiable. So there is an atomic formula $\psi(\bar{x})$ and a tuple $\bar{c}\subseteq C_V^r$ such that $\psi(\bar{c})\in \Gamma$ and $\neg \psi(\bar{c})\in \Gamma$. For each $A\in {V\choose r}$, because $\chi(A)\in S_r(C_A,\mathcal{H})$, exactly one of $\psi(\bar{c})$ or $\neg\psi(\bar{c})$ is in $\chi(A)$. This implies there must be distinct $A_1$, $A_2 \in {V\choose r}$ such that $\cup \bar{c}\subseteq A_1\cap A_2$ and $\psi(\bar{c})\in \chi(A_1)$ and $\neg \psi(\bar{c})\in \chi(A_2)$. Note $A_1\neq A_2$ and $A_1\cap A_2\neq \emptyset$ imply that if $\ell:=|A_1\cup A_2|$, then $r<\ell<2r$. Let $N$ be the $\mathcal{L}_{\mathcal{H}}$-structure $M[A_1\cup A_2]$. We show $N$ is an error of size $\ell$. By definition of $\chi$ being a choice function, there are $p_1,p_2\in S_r(C_V,\mathcal{H})$ and $\bar{a}_1$, $\bar{a}_2$ such that $\cup \bar{a}_1=A_1$, $\cup \bar{a}_2=A_2$, $p_1(c_{\bar{a}_1})=\chi(A_1)$, $p_2(c_{\bar{a}_2})=\chi(A_2)$, $M\models R_{p_1}(\bar{a}_1)$, and $M\models R_{p_2}(\bar{a}_2)$. By definition, $N\subseteq_{\mathcal{L}_{\mathcal{H}}}M$, thus $N\models R_{p_1}(\bar{a}_1)\wedge R_{p_2}(\bar{a}_2)$. Note $$ \{\psi(\bar{c}),\neg\psi(\bar{c})\}\subseteq \chi(A_1)\cup \chi(A_2)=p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2}) $$ implies $p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2})$ is unsatisfiable. Thus $N\in \mathcal{E}$ and $N\subseteq_{\mathcal{L}_{\mathcal{H}}}M$ implies $M$ is not error-free. Suppose on the other hand that $M$ is not error-free. Say $r< \ell< 2r$ and $N$ is an error of size $\ell$ in $M$. Then $N\subseteq_{\mathcal{L}_{\mathcal{H}}}M$ and there are $\bar{a}_1, \bar{a}_2\in dom(N)^{\underline{r}}$ and types $p_1(\bar{x}), p_2(\bar{x})\in S_r(\mathcal{H})$ such that $dom(N)=\cup \bar{a}_1\bigcup \cup \bar{a}_2$, $N\models R_{p_1}(\bar{a}_1)\wedge R_{p_2}(\bar{a}_2)$, and $p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2})$ is unsatisfiable. We define a function $\chi: {V\choose r}\rightarrow S_r(C_V,\mathcal{H})$ as follows. Set $\chi(\cup \bar{a}_1)=p_1(c_{\bar{a}_1})$ and $\chi(\cup\bar{a}_2)=p_2(c_{\bar{a}_2})$. For every $A'\in {V\choose r}\setminus \{A_1,A_2\}$, choose $\chi(A')$ to be any element of $Ch_M(A')$ (note $Ch_M(A')$ is nonempty since $M$ is complete). By construction, $\chi$ is a choice function for $M$. Suppose there is $G\unlhd_{\chi} M$. Then $p_1(c_{\bar{a}_1})=Diag^G(\cup \bar{a}_1)$ and $p_2(c_{\bar{a}_2})=Diag^G(\cup \bar{a}_2)$ implies $G\models p_1(\bar{a}_1)\cup p_2(\bar{a}_2)$, contradicting that $p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2})$ is unsatisfiable. Thus $\chi\in Ch(M)$ and there are no $\chi$-subpatterns of $M$. This finishes the proof. \end{proof} \begin{definition} Given a finite $\mathcal{L}_{\mathcal{H}}$-structure $M$, let $sub(M)=|\{G: G\unlhd_pM\}|$ be the number of full subpatterns of $M$. \end{definition} This definition and the following observation will be crucial to our enumeration theorem. \begin{observation}\label{ob0} If $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure with finite domain $V$, then \begin{align*} sub(M)\leq \prod_{A\in {V\choose r}}|Ch_M(A)|, \end{align*} and equality holds if and only if $M$ is error-free. \end{observation} \begin{proof} By definition of a choice function, $|Ch(M)|= \prod_{A\in {V\choose r}}|Ch_M(A)|$. By definition of subpattern, for each $G\unlhd_pM$, there is $\chi_G\in Ch(M)$ which chooses $G$. Observation \ref{ob00} implies the map $f:G\mapsto \chi_G$ is a well-defined injection from $\{G: G\unlhd_p M\}$ to $Ch(M)$. Thus $$ sub(M)=|\{G: G\unlhd_p M\}|\leq |Ch(M)|=\prod_{A\in {V\choose r}}|Ch_M(A)|. $$ We now show equality holds if and only if $M$ is error-free. Suppose first $M$ is error-free. We claim $f$ is surjective. Fix $\chi\in Ch(M)$. Since $M$ is error-free, Lemma \ref{Lrandom} implies that there is an $\mathcal{L}$-structure $G_{\chi}$ such that $G_{\chi}\unlhd_{\chi} M$. So $G_{\chi}\in \{G: G\unlhd_pM\}$ implies $f(G_{\chi})$ exists. By Observation \ref{ob00}, we must have $f(G_{\chi})=\chi$. Thus $f$ is surjective, and consequently $sub(M)= |Ch(M)|=\prod_{A\in {V\choose r}}Ch_M(A)$. Conversely, suppose equality holds. Then $f$ is an injective map from a finite set to another finite set of the same size, thus it must be surjective. This implies that for all $\chi\in Ch(M)$, there is an $\mathcal{L}$-structure $G$ such that $G\unlhd_{\chi}M$. By Lemma \ref{Lrandom}, this implies $M$ is error-free. \end{proof} \begin{remark}\label{specialob0} Suppose $\mathcal{L}$ contains no relations of arity less than $r$. If $\mathcal{M}$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure with finite domain $V$, then $sub(M)=\prod_{A\in {V\choose r}}|Ch_M(A)|$. \end{remark} \begin{proof} Our assumption on $\mathcal{L}$ implies $\mathcal{E}=\emptyset$ by definition. Thus, if $\mathcal{M}$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure with finite domain $V$, it is error-free, so Observation \ref{ob0} implies $sub(M)=\prod_{A\in {V\choose r}}|Ch_M(A)|$. \end{proof} \noindent Remark \ref{specialob0} applies to most examples we are interested in, including graphs, (colored) $k$-uniform hypergraphs for any $k\geq 2$, directed graphs, and discrete metric spaces. \subsection{$\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-structures and $\mathcal{L}_{\mathcal{H}}$-templates}\label{temsec} In this subsection we consider $\mathcal{L}_{\mathcal{H}}$-structures with the property that all choice functions give rise to subpatterns in $\mathcal{H}$. \begin{definition}\label{Hrand} An $\mathcal{L}_{\mathcal{H}}$-structure $M$ is \emph{$\mathcal{H}$-random} if it is complete and for every $\chi\in Ch(M)$, there is an $\mathcal{L}$-structure $N\in \mathcal{H}$ such that $N\unlhd_{\chi} M$. \end{definition} Observe that by Proposition \ref{Lrandom}, any $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-structure is error-free. The difference between being error-free and being $\mathcal{H}$-random is as follows. If an $\mathcal{L}_{\mathcal{H}}$-structure is error-free, then it must have at least one full subpattern, however some or all its subpatterns may not be in $\mathcal{H}$. On the other hand, if an $\mathcal{L}_{\mathcal{H}}$-structure is $\mathcal{H}$-random, then it must have at least one full subpattern, and further, all its full subpatterns must also be in $\mathcal{H}$. \begin{example}\label{ex4} Let $\mathcal{L}$, $\mathcal{P}$, $V$, $H'$ and $G$ be as in Example \ref{ex3}. Observe that $G$ is \emph{not} $\mathcal{P}$-random, since $H'\unlhd_pG$, but $H'\notin \mathcal{P}$. We now define an $\mathcal{L}_{\mathcal{P}}$-structure $G''$ which is $\mathcal{P}$-random. Let $G''$ be any $\mathcal{L}_{\mathcal{P}}$-structure with domain $V$ such that for all $(x,y)\in V^{\underline{2}}$, $$ G''\models R_{p_1}(x,y)\wedge R_{p_2}(x,y)\wedge \neg R_{p_3}(x,y)\wedge \bigwedge_{i=1}^3\neg R_{p_i}(x,x). $$ It is easy to check that $G''$ is a $\mathcal{L}_{\mathcal{P}}$-template and for all $xy\in {V\choose 2}$, $Ch_{G''}(xy)=\{p_1(c_x,c_y), p_2(c_x,c_y)\}$. Suppose $\chi''\in Ch(G'')$. Then for all $xy\in {V\choose 2}$, $\chi''(xy)\in \{p_1(c_x,c_y), p_2(c_x,c_y)\}$. Let $M$ be the $\mathcal{L}$-structure such that $M\unlhd_{\chi''}G''$, that is, $dom(M)=V$ and for each $xy\in {V\choose 2}$, $M\models p_i(x,y)$ if and only if $\chi''(xy)=p_i(c_x,c_y)$. Then for all $xy\in {V\choose 2}$, $M\models p_1(x,y)$ or $M\models p_2(x,y)$. Since there is no way to violate the triangle inequality using distances in $\{1,2\}$, $M$ is a metric space. Thus we have shown that for every $\chi''\in Ch(G'')$, there is an $\mathcal{L}$-structure $M\in \mathcal{P}$ such that $M\unlhd_{\chi''} G''$. Thus $G''$ is $\mathcal{P}$-random. \end{example} The most important $\mathcal{L}_{\mathcal{H}}$-structures for the rest of the paper are $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-templates. We now fix notation for these special $\mathcal{L}_{\mathcal{H}}$-structures. \begin{definition} Suppose $V$ is a set, and $n$ is an integer. Then \begin{itemize} \item $\mathcal{R}(V,\mathcal{H})$ is the set of all $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-templates with domain $V$ and \item $\mathcal{R}(n,\mathcal{H})$ is the class of all $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-templates of size $n$. \end{itemize} \end{definition} \noindent In the above notation, $\mathcal{R}$ is for ``random.'' Note that if $\mathcal{H}(n)=\emptyset$ for some $n$, then $\mathcal{R}(n,\mathcal{H})=\emptyset$. \section{Main Results}\label{mainresults} In this section we state the main results of this paper. Recall that $\mathcal{L}$ is a fixed finite relational language of maximum arity $r\geq 2$. We now define our generalization of extremal graphs. By convention, set $\max \emptyset=0$. \begin{definition}\label{genexdef} Suppose $\mathcal{H}$ is a collection of finite $\mathcal{L}$-structures. Given $n$, set $$ ex(n,\mathcal{H})=\max\{sub(M): M\in \mathcal{R}(n,\mathcal{H})\}. $$ We say $M\in \mathcal{R}(n,\mathcal{H})$ is \emph{extremal} if $sub(M)=ex(n,\mathcal{H})$. If $V$ is a set and $n\in \mathbb{N}$, then \begin{itemize} \item $\mathcal{R}_{ex}(V,\mathcal{H})$ is the set of extremal elements of $\mathcal{R}(V,\mathcal{H})$ and \item $\mathcal{R}_{ex}(n,\mathcal{H})$ is the class of extremal elements of $\mathcal{R}(n,\mathcal{H})$. \end{itemize} \end{definition} The main idea is that when $\mathcal{H}$ is a hereditary $\mathcal{L}$-property, $\textnormal{ex}(n,\mathcal{H})$ is the correct generalization of the extremal number of a graph, and elements of $\mathcal{R}_{ex}(n,\mathcal{H})$ are the correct generalizations extremal graphs of size $n$. \begin{definition}\label{pidef} Suppose $\mathcal{H}$ is a nonempty collection of finite $\mathcal{L}$-structures. When it exists, set $$ \pi(\mathcal{H})=\lim_{n\rightarrow \infty}\textnormal{ex}(n,\mathcal{H})^{1/{n\choose r}} $$ \end{definition} \noindent Using techniques similar to those in \cite{BoTh1} we will show the following. \begin{theorem}\label{densityexists} If $\mathcal{H}$ is hereditary $\mathcal{L}$-property, then $\pi(\mathcal{H})$ exists. \end{theorem} We now state our approximate enumeration theorem in terms of the asymptotic density. \begin{theorem}[Enumeration]\label{enumeration} Suppose $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. Then the following hold. \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $|\mathcal{H}_n|= \pi(\mathcal{H})^{{n\choose r}+o(n^r)}$. \item If $\pi(\mathcal{H})\leq 1$, then $|\mathcal{H}_n|=2^{o(n^r)}$. \end{enumerate} \end{theorem} The notion $\pi(\mathcal{H})$ is related to many existing notions of asymptotic density for various combinatorial structures, and Theorem \ref{enumeration} can be seen as generalizing many existing enumeration theorems. Some of these connections will be discussed in Section \ref{end} and Appendices \ref{coloredhg}, \ref{dgsec}, and \ref{trifreesec}. We say a hereditary $\mathcal{L}$-property $\mathcal{H}$ is \emph{fast-growing} if $\pi(\mathcal{H})>1$. In this case, we informally say $M\in \mathcal{R}(n,\mathcal{H})$ is \emph{almost extremal} if $sub(M)\geq\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$ for some small $\epsilon$. Our next theorem shows that almost all elements in a fast-growing hereditary $\mathcal{L}$-property $\mathcal{H}$ are close to subpatterns of almost extremal elements of $\mathcal{R}(n,\mathcal{H})$. Given $\epsilon>0$, $n$, and a collection $\mathcal{H}$ of $\mathcal{L}$-structures, let \begin{align*} E(n,\mathcal{H})&=\{G\in \mathcal{H}_n: G\unlhd_pM\text{ for some $M\in \mathcal{R}_{ex}(n,\mathcal{H})$}\}\text{ and }\\ E(\epsilon, n,\mathcal{H})&=\{G\in \mathcal{H}_n: G\unlhd_pM\text{ for some $M\in \mathcal{R}(n,\mathcal{H})$ with $sub(M)\geq\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$}\}. \end{align*} Given $\delta>0$, let $E^{\delta}(n,\mathcal{H})$ and $E^{\delta}(\epsilon, n,\mathcal{H})$ denote the set of $G\in \mathcal{H}_n$ which are $\delta$-close to any element of $E(n,\mathcal{H})$ and $E(\epsilon, n,\mathcal{H})$, respectively. \begin{theorem}\label{b4stab} Suppose $\mathcal{H}$ is a fast-growing hereditary $\mathcal{L}$-property. For all $\epsilon, \delta>0$ there is $\beta>0$ such that for sufficiently large $n$, $$ \frac{|\mathcal{H}_n\setminus E^{\delta}(\epsilon, n,\mathcal{H})|}{|\mathcal{H}_n|} \leq 2^{-\beta {n\choose r}}. $$ \end{theorem} We now define our generalization of a graph stability theorem. \begin{definition}\label{stabdef} Suppose $\mathcal{H}$ is a nontrivial collection of $\mathcal{L}$-structures. We say $\mathcal{H}$ \emph{has a stability theorem} if for all $\delta>0$ there is $\epsilon>0$ and $N$ such that $n>N$ implies the following. If $M\in \mathcal{R}(n,\mathcal{H})$ satisfies $sub(M)\geq\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$, then $M$ is $\delta$-close to some $M'\in \mathcal{R}_{ex}(n,\mathcal{H})$. \end{definition} Our next result, Theorem \ref{stab} below, shows that if a fast-growing hereditary $\mathcal{L}$-property $\mathcal{H}$ has a stability theorem, we can strengthen Theorem \ref{b4stab} to say that that almost all elements in $\mathcal{H}_n$ are approximately subpatterns of elements of $\mathcal{R}_{ex}(n,\mathcal{H})$. \begin{theorem}\label{stab} Suppose $\mathcal{H}$ is a fast growing hereditary $\mathcal{L}$-property with a stability theorem. Then for all $\delta>0$, there is a $\beta>0$ such that for sufficiently large $n$, $$ \frac{|\mathcal{H}_n\setminus E^{\delta}(n,\mathcal{H})|}{|\mathcal{H}_n|} \leq 2^{-\beta {n\choose r}}. $$ \end{theorem} When one has a good understanding of the structure of elements in $\mathcal{R}_{ex}(n, \mathcal{H})$, Theorem \ref{stab} gives us a good description of the approximate structure of most elements in $\mathcal{H}_n$, when $n$ is large. The main new tool we will use to prove our main theorems is Theorem \ref{version1} below, which is an adaptation of the hypergraph containers theorem to the setting of $\mathcal{L}$-structures. \begin{definition} If $F$ is an $\mathcal{L}$-structure, let $\tilde{F}$ be the set of $\mathcal{L}_{\mathcal{H}}$-structures $M$ such that $F\unlhd_pM$. If $\mathcal{F}$ is a collection of $\mathcal{L}$-structures, let $\tilde{\mathcal{F}}=\bigcup_{F\in \mathcal{F}}\tilde{F}$. \end{definition} \begin{theorem}\label{version1} Suppose $0<\epsilon<1$ and $k\geq r$ is an integer. Then there exist positive constants $c=c(k,r, \mathcal{L},\epsilon)$ and $m=m(k,r)>1$ such that for all sufficiently large $n$ the following holds. Assume $\mathcal{F}$ is a collection of finite $\mathcal{L}$-structures each of size at most $k$ and $\mathcal{B}:=\operatorname{Forb}(\mathcal{F})\neq \emptyset$. For any $n$-element set $W$, there is a collection $\mathcal{C}$ of $\mathcal{L}_{\mathcal{B}}$-templates with domain $W$ such that \begin{enumerate} \item For all $\mathcal{F}$-free $\mathcal{L}$-structures $M$ with domain $W$, there is $C\in \mathcal{C}$ such that $M\unlhd_pC$, \item For all $C\in \mathcal{C}$, $\textnormal{prob}(\tilde{\mathcal{F}},C)\leq \epsilon$ and $\textnormal{prob}(\mathcal{E},C)\leq \epsilon$. \item $\log |\mathcal{C}|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} \end{theorem} We will combine Theorem \ref{version1} with a general version of the graph removal lemma proved by Aroskar and Cummings in \cite{AroskarCummings} to prove a supersaturation theorem for hereditary $\mathcal{L}$-properties (Theorem \ref{GENSUPERSAT} below), and a version of the hypergraph containers theorem for hereditary $\mathcal{L}$-properties (Theorem \ref{COROLLARY2} below). \begin{theorem}[Supersaturation]\label{GENSUPERSAT} Suppose $\mathcal{H}$ is a non-trivial hereditary $\mathcal{L}$-property and $\mathcal{F}$ is as in Observation \ref{HP} so that $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$. Then for all $\delta>0$ there are $\epsilon>0$ and $K$ such that for all sufficiently large $n$, if $M$ is an $\mathcal{L}_{\mathcal{H}}$-template of size $n$ such that $\textnormal{prob}(\tilde{\mathcal{F}}(K)\cup \mathcal{E}(K),M)<\epsilon$, then \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $sub(M)\leq \textnormal{ex}(n,\mathcal{H})^{1+\delta}$. \item If $\pi(\mathcal{H})\leq 1$, then $sub(M)\leq 2^{\delta {n\choose r}}$. \end{enumerate} \end{theorem} \begin{theorem}\label{COROLLARY2} Suppose $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. Then there is $m=m(\mathcal{H}, r_{\mathcal{L}})>1$ such that the following holds. For every $\delta>0$ there is a constant $c=c(\mathcal{H},\mathcal{L},\delta)$ such that for all sufficiently large $n$ there is a set of $\mathcal{L}_{\mathcal{H}}$-templates $\mathcal{C}$ with domain $[n]$ satisfying the following properties. \begin{enumerate} \item For every $H\in \mathcal{H}_n$, there is $C\in \mathcal{C}$ such that $H\unlhd_pC$. \item For every $C\in \mathcal{C}$, there is $C'\in \mathcal{R}([n],\mathcal{H})$ such that $dist(C,C')\leq \delta$. \item $\log |\mathcal{C}|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} \end{theorem} \section{Proofs of Main Theorems} In this section we prove our main results using Theorems \ref{version1}, \ref{GENSUPERSAT}, and \ref{COROLLARY2}. For the rest of the section, $\mathcal{H}$ is a fixed hereditary $\mathcal{L}$-property. \begin{lemma}\label{templatelem2} Suppose $N$ is an $\mathcal{L}$-structure and $\tilde{N}$ is the $\mathcal{L}_{\mathcal{H}}$-structure such that $dom(\tilde{N})=dom(N)$ and for each $\bar{a} \in dom(\tilde{N})^{r}$ and $p(\bar{x})\in S_r(\mathcal{H})$, $\tilde{N}\models R_p(\bar{a})$ if and only if $N\models p(\bar{a})$. Then $\tilde{N}$ is an $\mathcal{L}_{\mathcal{H}}$-template and $N$ is the unique full subpattern of $\tilde{N}$. \end{lemma} \begin{proof} Let $V=dom(N)=dom(\tilde{N})$. We first verify $\tilde{N}$ is an $\mathcal{L}_{\mathcal{H}}$-template. By the definition of $\tilde{N}$, for all $A\in {V\choose r}$, $Ch_{\tilde{N}}(A)=\{Diag^N(A)\}$. Therefore $\tilde{N}$ is complete. If $\bar{a}\in V^r\setminus V^{\underline{r}}$ and $p\in S_r(\mathcal{H})$, then because $p$ is a proper type, $N\nvDash p(\bar{a})$. Thus by definition of $\tilde{N}$, $\tilde{N}\models \neg R_p(\bar{a})$ and $\tilde{N}$ satisfies part (1) of Definition \ref{templatedef}. Suppose $p(\bar{x}), p'(\bar{x})\in S_r(\mathcal{H})$ and $\mu\in Perm(r)$ are such that $p(\bar{x})=p'(\mu(\bar{x}))$. Then for all $\bar{a}\in V^r$, $\tilde{N}\models R_p(\bar{a})$ if and only if $N\models p(\bar{a})$ if and only if $N\models p'(\mu(\bar{a}))$ if and only if $\tilde{N}\models R_{p'}(\mu(\bar{a}))$. Thus $\tilde{N}$ satisfies part (2) of Definition \ref{templatedef}, so $\tilde{N}$ is an $\mathcal{L}_{\mathcal{H}}$-template. Define $\chi: {V\choose r}\rightarrow S_r(C_V,\mathcal{H})$ by setting $\chi(A)=Diag^N(A)$ for each $A\in {V\choose r}$. It is clear that $\chi\in Ch(\tilde{N})$ and $N\unlhd_{\chi}\tilde{N}$. By definition of $\tilde{N}$, $\chi$ is the \emph{only} choice function for $\tilde{N}$, so any full subpattern of $\tilde{N}$ must be chosen by $\chi$. By Observation \ref{ob00}, $\chi$ chooses at most one $\mathcal{L}$-structure, so $N$ is the unique full subpattern of $\tilde{N}$. \end{proof} \noindent We now prove Theorem \ref{enumeration}. The proof is based on the method of proof in \cite{BoTh1}. \vspace{3mm} \noindent {\bf Proof of Theorem \ref{densityexists}.} Let $b_n=ex(n,\mathcal{H})^{1/{n\choose r}}$. If $\mathcal{H}$ is trivial, then for sufficiently large $n$, $\mathcal{R}(n,\mathcal{H})=\emptyset$ so by convention, $\textnormal{ex}(n,\mathcal{H})=0$. Thus, for sufficiently large $n$, $b_n=0$ and $\pi(\mathcal{H})$ exists and is equal to zero. Assume now $\mathcal{H}$ is nontrivial. We show that the sequence $b_n$ is bounded below and non-increasing. Since $\mathcal{H}$ is non-trivial and has the hereditary property, $\mathcal{H}_n\neq \emptyset$ for all $n$. Fix $n\geq 1$ and choose any $N\in \mathcal{H}_n$. Let $\tilde{N}$ be the $\mathcal{L}_{\mathcal{H}}$-structure defined as in Lemma \ref{templatelem2} for $N$. Then $\tilde{N}$ is an $\mathcal{L}_{\mathcal{H}}$-template, and its only full subpattern is $N$. Since $N\in \mathcal{H}$, this implies $\tilde{N}\in \mathcal{R}(n,\mathcal{H})$ and $sub(\tilde{N})=1$. So we have shown $b_n\geq 1$ for all $n\geq 1$. We now show the $b_n$ are non-increasing. Fix $n\geq 2$. Let $M\in \mathcal{R}(n,\mathcal{H})$ be such that $sub(M)\geq 1$ and let $V=dom(M)$. Fix $a\in V$ and set $V_a=V\setminus \{a\}$ and $M_a=M[V_a]$. We claim $M_a\in \mathcal{R}(n-1,\mathcal{H})$. Because $M$ is an $\mathcal{L}_{\mathcal{H}}$-template, the definition of $M_a$ implies $M_a$ is also an $\mathcal{L}_{\mathcal{H}}$-template. Suppose $\chi\in Ch(M_a)$. We want to show there exists $N_a\in \mathcal{H}$ with $N_a\unlhd_{\chi}M_a$. We define a function $\chi':{V\choose r}\rightarrow S_r(C_V,\mathcal{H})$ as follows. For $A\in {V_a\choose r}$, set $\chi'(A)=\chi(A)$, and for $A\in {V\choose r}\setminus A\in {V_a\choose r}$, choose $\chi'(A)$ to be any element of $Ch_{M_a}(A)=Ch_M(A)$ (this is possible since $M$ is complete). Note that for each $A\in {V_a\choose r}$, $\chi(A)\in Ch_M(A)$, so $\chi'\in Ch(M)$. Because $M$ is $\mathcal{H}$-random, there is $N\in \mathcal{H}$ such that $N\unlhd_{\chi'}M$. Let $N_a=N[V_a]$. Because $\mathcal{H}$ has the hereditary property and $N_a\subseteq_{\mathcal{L}}N$, $N_a\in \mathcal{H}$. For each $A\in {V_a\choose r}$, $Diag^{N_a}(A)=Diag^N(A)=\chi'(A)=\chi(A)$, so $N_a\unlhd_{\chi}M_a$. Thus we have verified that $M_a\in \mathcal{R}(n-1,\mathcal{H})$. By definition of $b_{n-1}$, this implies $sub(M_a)^{1/{n-1\choose r}}\leq b_{n-1}$. Because $M_a$ is $\mathcal{H}$-random, Lemma \ref{Lrandom} implies it is error-free, so Observation \ref{ob0} implies $sub(M_a)=\prod_{A\in {V_a\choose r}} |Ch_{M_a}(A)|$. Then observe that $$ sub(M)=\Big(\prod_{a\in V} \prod_{A\in {V_a\choose r}} |Ch_{M_a}(A)|\Big)^{1/(n-r)}= \Big(\prod_{a\in V} sub(M_a)\Big)^{1/(n-r)}. $$ Since $sub(M_a)\leq b_{n-1}^{n-1\choose r}$, this implies $$ sub(M)\leq \Big(\prod_{a\in V} b_{n-1}^{n-1\choose r}\Big)^{1/(n-r)} = b_{n-1}^{n{n-1\choose r}/(n-r)}= b_{n-1}^{n\choose r}. $$ Thus for all $M\in \mathcal{R}(n,\mathcal{H})$, $sub(M)^{1/{n\choose r}}\leq b_{n-1}$. So by definition, $b_n\leq b_{n-1}$. \qed \vspace{3mm} \noindent The following observations follow from the proof of Theorem \ref{densityexists}. \begin{observation}\label{ob5} Assume $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. \begin{enumerate}[(a)] \item For all $n$, $\textnormal{ex}(n,\mathcal{H})^{1/{n\choose r}}\geq \pi(\mathcal{H})$ (since $(b_n)_{n\in \mathbb{N}}$ is non-increasing and converges to $\pi(\mathcal{H})$). \item Either $\mathcal{H}$ is trivial and $\pi(\mathcal{H})=0$ or $\mathcal{H}$ is non-trivial and $\pi(\mathcal{H})\geq 1$. \end{enumerate} \end{observation} \vspace{3mm} \noindent{\bf Proof of Theorem \ref{enumeration}.} Assume $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. Recall we want to show the following. \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $|\mathcal{H}_n|= \pi(\mathcal{H})^{{n\choose r}+o(n^r)}$. \item If $\pi(\mathcal{H})\leq 1$, then $|\mathcal{H}_n|=2^{o(n^r)}$. \end{enumerate} Assume first that $\mathcal{H}$ is trivial. Then by Observation \ref{ob5}(b), $\pi(\mathcal{H})=0\leq 1$, so we are in case 2. Since $|\mathcal{H}_n|=0$ for all sufficiently large $n$, $|\mathcal{H}_n|=2^{o(n^2)}$ holds, as desired. Assume now $\mathcal{H}$ is non-trivial, so $\pi(\mathcal{H})\geq 1$ by Observation \ref{ob5}(b). We show that for all $0<\eta<1$, either $\pi(\mathcal{H})=1$ and $|\mathcal{H}_n|\leq 2^{\eta n^r}$ or $\pi(\mathcal{H})>1$ and $\pi(\mathcal{H})^{n\choose 2}\leq |\mathcal{H}_n|\leq \pi(\mathcal{H})^{{n\choose r}+\eta n^r}$. Fix $0<\eta<1$. Let $\mathcal{F}$ be as in Observation \ref{HP} for $\mathcal{H}$ so that $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$. Choose $\epsilon>0$ and $K$ as in Theorem \ref{GENSUPERSAT} for $\delta=\eta/4$. Replacing $K$ if necessary, assume $K\geq r$. Apply Theorem \ref{version1} to $\epsilon$ and $\mathcal{F}(K)$ to obtain $m=m(K,r)>1$ and $c=c(K,r, \mathcal{L}, \epsilon)$. Assume $n$ is sufficiently large. Theorem \ref{version1} with $W=[n]$ and $\mathcal{B}:=\operatorname{Forb}(\mathcal{F}(K))$ implies there is a collection $\mathcal{C}$ of $\mathcal{L}_{\mathcal{B}}$-templates with domain $[n]$ such that the following hold. \begin{enumerate}[(i)] \item For all $\mathcal{F}(K)$-free $\mathcal{L}$-structures $M$ with domain $[n]$, there is $C\in \mathcal{C}$ such that $M\unlhd_pC$, \item For all $C\in \mathcal{C}$, $\textnormal{prob}(\widetilde{\mathcal{F}(K)},C)\leq \epsilon$ and $\textnormal{prob}(\mathcal{E},C)\leq \epsilon$. \item $\log |\mathcal{C}|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} Note that because $K\geq r$, $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$ and $\mathcal{B}=\operatorname{Forb}(\mathcal{F}(K))$ imply we must have $S_r(\mathcal{H})=S_r(\mathcal{B})$. Consequently all $\mathcal{L}_{\mathcal{B}}$-templates are also $\mathcal{L}_{\mathcal{H}}$-templates. In particular the elements in $\mathcal{C}$ are all $\mathcal{L}_{\mathcal{H}}$-templates. Therefore, (ii) and Theorem \ref{GENSUPERSAT} imply that for all $C\in \mathcal{C}$, either $sub(C)\leq \textnormal{ex}(n,\mathcal{H})^{1+\eta/4}$ (case $\pi(\mathcal{H})>1$) or $sub(C)\leq 2^{\eta{n\choose r}/4}$ (case $\pi(\mathcal{H})=1$). Note every element in $\mathcal{H}_n$ is $\mathcal{F}$-free, so is also $\mathcal{F}(K)$-free. This implies by (i) that every element of $\mathcal{H}_n$ is a full subpattern of some $C\in \mathcal{C}$. Therefore we can construct every element in $\mathcal{H}_n$ as follows. \begin{enumerate}[$\bullet$] \item Choose a $C\in \mathcal{C}$. There are at most $|\mathcal{C}|\leq 2^{cn^{r-\frac{1}{m}} \log n}$ choices. \item Choose a full subpattern of $C$. There are at most $sub(C)\leq \textnormal{ex}(n,\mathcal{H})^{1+\eta/4}$ choices if $\pi(\mathcal{H})>1$ and at most $sub(C)\leq 2^{\eta{n\choose r}/4}$ choices if $\pi(\mathcal{H})=1$. \end{enumerate} This implies \begin{eqnarray}\label{AP} |\mathcal{H}_n|\leq \begin{cases} 2^{cn^{r-\frac{1}{m}}\log n}\textnormal{ex}(n,\mathcal{H})^{1+\eta/4} & \text{ if }\pi(\mathcal{H})>1\\ 2^{cn^{r-\frac{1}{m}}\log n}2^{\eta{n\choose r}/4} & \text{ if }\pi(\mathcal{H})=1. \end{cases}\end{eqnarray} If $\pi(\mathcal{H})>1$, then we may assume $n$ is sufficiently large so that $\textnormal{ex}(n,\mathcal{H})\leq \pi(\mathcal{H})^{(1+\eta/4){n\choose r}}$ (see Observation \ref{ob5}(a)). Combining this with (\ref{AP}), we have that when $\pi(\mathcal{H})>1$, $$ |\mathcal{H}_n|\leq 2^{cn^{r-\frac{1}{m}}\log n}\pi(\mathcal{H})^{(1+\eta/4)^2{n\choose r}}\leq \pi(\mathcal{H})^{{n\choose r}+\eta {n\choose r}}, $$ where the last inequality is because $\pi(\mathcal{H})>1$, $(1+\eta/4)^2<1+\eta$, and $n$ is sufficiently large. If $\pi(\mathcal{H})=1$, then (\ref{AP}) implies $$ |\mathcal{H}_n|\leq 2^{cn^{r-\frac{1}{m}}\log n}2^{\eta{n\choose r}/2}\leq 2^{\eta{n\choose r}}, $$ where the last inequality is because $n$ is sufficiently large. Thus, we have shown $|\mathcal{H}_n|\leq 2^{\eta n^r}$ when $\pi(\mathcal{H})=1$ and $|\mathcal{H}_n|\leq \pi(\mathcal{H})^{{n\choose r}+\eta n^r}$ when $\pi(\mathcal{H})>1$. We just have left to show that when $\pi(\mathcal{H})>1$, then $|\mathcal{H}_n|\geq \pi(\mathcal{H})^{n\choose r}$. This holds because for any $M\in \mathcal{R}_{ex}([n],\mathcal{H})$, all $\textnormal{ex}(n,\mathcal{H})$ many full subpatterns of $M$ are in $\mathcal{H}_n$. Thus $|\mathcal{H}_n|\geq \textnormal{ex}(n,\mathcal{H})\geq \pi(\mathcal{H})^{n\choose r}$, where the second inequality is by Observation \ref{ob5}(a). This finishes the proof. \qed \vspace{3mm} \noindent We now prove a few lemmas needed for Theorems \ref{b4stab} and \ref{stab}. \begin{lemma}\label{templatelem} Suppose $C$ and $C'$ are $\mathcal{L}_{\mathcal{H}}$-templates with the same domain $V$. Then for all $A\in {V\choose r}$, $A\in \textnormal{diff}(C,C')$ if and only if $Ch_C(A)\neq Ch_{C'}(A)$. \end{lemma} \begin{proof} Fix $A\in {V\choose r}$. Suppose first $A\in \textnormal{diff}(C,C')$. Then there is $p\in S_r(\mathcal{H})$ and an enumeration $\bar{a}$ of $A$ such that $C\models R_p(\bar{a})$ and $C'\models \neg R_p(\bar{a})$. This implies $p(c_{\bar{a}})\in Ch_C(A)$. Suppose by contradiction $p(c_{\bar{a}})$ were in $Ch_{C'}(A)$. Then there is $p'(\bar{x})\in S_r(\mathcal{H})$ and $\mu\in Perm(r)$ such that $p'(\mu(\bar{x}))= p(\bar{x})$ and $C'\models R_{p'}(\mu(\bar{a}))$. Because $C'$ is an $\mathcal{L}_{\mathcal{H}}$-template, this implies $C'\models R_p(\bar{a})$, a contradiction. Suppose now $Ch_C(A)\neq Ch_{C'}(A)$. Then there is $p(\bar{x})\in S_r(\mathcal{H})$ and an enumeration $\bar{a}$ of $A$ such that $p(c_{\bar{a}})\in Ch_C(A)$ and $p(c_{\bar{a}})\notin Ch_{C'}(A)$. Since $p(c_{\bar{a}})\in Ch_C(A)$, by definition there is $p'(\bar{x})$ and $\mu\in Perm(r)$ such that $p'(\mu(\bar{x}))= p(\bar{x})$ and $C\models R_{p'}(\mu(\bar{a}))$. Since $p(c_{\bar{a}})\notin Ch_{C'}(A)$ and $C'$ is an $\mathcal{L}_{\mathcal{H}}$-template, $C'\models \neg R_{p'}(\mu(\bar{a}))$. This shows $qftp^C(\bar{a}) \neq qftp^{C'}(\bar{a})$, so $A\in \textnormal{diff}(C,C')$, as desired. \end{proof} \begin{lemma}\label{deltaclose1} Suppose $\mathcal{H}$ is a non-trivial hereditary $\mathcal{L}$-property. Then there is $\gamma=\gamma(\mathcal{H})>0$ such that for all $\delta>0$ and $n\geq r$, if $C$ and $C'$ are $\mathcal{L}_{\mathcal{H}}$-templates with domain $[n]$ such that $C'$ is error-free and $\textnormal{dist}(C,C')\leq \delta$, then the following holds. \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $sub(C)\leq sub(C')\textnormal{ex}(n,\mathcal{H})^{\gamma \delta}$. \item If $\pi(\mathcal{H})=1$, then $sub(C)\leq sub(C')2^{\gamma \delta {n\choose r}}$. \end{enumerate} \end{lemma} \begin{proof} Fix $n\geq r$ and assume $C$ and $C'$ are $\mathcal{L}_{\mathcal{H}}$-templates with domain $[n]$ such that $C'$ is error-free and $\textnormal{dist}(C,C')\leq \delta$. Then by definition of $\textnormal{dist}(C,C')$, $|\textnormal{diff}(C,C')|\leq \delta {n\choose r}$. By Lemma \ref{templatelem}, \begin{align}\label{ch} \textnormal{diff}(C,C')=\{A\in {V\choose r}: Ch_C(A)\neq Ch_{C'}(A)\}. \end{align} Note that for every $A\in {V\choose r}$, $|Ch_C(A)|\leq |S_r(\mathcal{H})|$ (by definition of $Ch_C(A)$) and $1\leq |Ch_{C'}(A)|$ (since $C'$ is complete). Thus $\frac{|Ch_{C}(A)|}{|Ch_{C'}(A)|}\leq |S_r(\mathcal{H})|$. By Observation \ref{ob0} and (\ref{ch}), \begin{align*} sub(C)\leq \prod_{A\in {V\choose r}}|Ch_C(A)|&=\Big(\prod_{A\notin \textnormal{diff}(C,C')}|Ch_{C'}(A)|\Big)\Big(\prod_{A\in \textnormal{diff}(C,C')}|Ch_{C}(A)|\Big)\\ &=\Big(\prod_{A\in {V\choose r}}|Ch_{C'}(A)|\Big)\Big(\prod_{A\in \textnormal{diff}(C,C')}\frac{|Ch_{C}(A)|}{|Ch_{C'}(A)|}\Big). \end{align*} Combining this with $\frac{|Ch_{C}(A)|}{|Ch_{C'}(A)|}\leq |S_r(\mathcal{H})|$ and $|\textnormal{diff}(C,C')|\leq \delta {n\choose r}$ yields \begin{align}\label{ineq1} sub(C)\leq \Big(\prod_{A\in {V\choose r}}|Ch_{C'}(A)|\Big) |S_r(\mathcal{H})|^{\delta{n\choose r}}=sub(C')|S_r(\mathcal{H})|^{\delta {n\choose r}}, \end{align} where the equality is by Observation \ref{ob0} and because $C'$ is error-free. If $\pi(\mathcal{H})>1$, choose $\gamma>0$ such that $|S_r(\mathcal{H})|=\pi(\mathcal{H})^{\gamma}$ (this is possible since $\pi(\mathcal{H})>1$ implies $S_r(\mathcal{H})>1$). Recall from Observation \ref{ob5}(a) that for all $n$, $\textnormal{ex}(n,\mathcal{H})\geq \pi(\mathcal{H})^{n\choose r}$. Combining this with our choice of $\gamma$ and (\ref{ineq1}), we have $$ sub(C)\leq sub(C')|S_r(\mathcal{H})|^{\delta {n\choose r}}=sub(C')\pi(\mathcal{H})^{\gamma \delta {n\choose r}}\leq sub(C')\textnormal{ex}(n,\mathcal{H})^{\gamma \delta}. $$ If $\pi(\mathcal{H})=1$, choose $\gamma>0$ such that $|S_r(\mathcal{H})|\leq 2^{\gamma}$ (this is possible since $\mathcal{H}$ nontrivial implies $|S_r(\mathcal{H})|\geq 1)$. Combining our choice of $\gamma$ with (\ref{ineq1}) implies $$ sub(C)\leq sub(C')|S_r(\mathcal{H})|^{\delta {n\choose r}}\leq sub(C')2^{\gamma \delta {n\choose r}}. $$ \end{proof} \begin{lemma}\label{deltaclose2} Suppose $C$ is an $\mathcal{L}_{\mathcal{H}}$-template with domain $W$ of size $n\geq r$ and $G\unlhd_pC$. If $D\in \mathcal{R}(W,\mathcal{H})$ is such that $\textnormal{dist}(C,D)\leq \delta$, then there is $G'\in \mathcal{H}$ such that $G'\unlhd_pD$ and $\textnormal{dist}(G,G')\leq \delta$. \end{lemma} \begin{proof} Fix $C$ and $D$ satisfying the hypotheses Because $\textnormal{dist}(C,D)\leq \delta$, we have $|\textnormal{diff}(C,D)|\leq \delta{n\choose r}$. By Lemma \ref{templatelem}, \begin{align}\label{ch1} \textnormal{diff}(C,D)=\{A\in {W\choose r}: Ch_C(A)\neq Ch_D(A)\}. \end{align} Define a function $\chi: {W\choose r}\rightarrow S_r(C_W)$ as follows. For $A\in {W\choose r}\setminus \textnormal{diff}(C,D)$, set $\chi(A)=Diag^G(A)$. For each $A\in \textnormal{diff}(C,D)$, choose $\chi(A)$ to be any element of $Ch_D(A)$ (which is nonempty because $D$ is an $\mathcal{L}_{\mathcal{H}}$-template). Since $G\unlhd_pC$, for all $A\in {W\choose r}$, $Diag^G(A)\in Ch_C(A)$. Thus, by definition of $\chi$ and (\ref{ch1}), for all $A\in {W\choose r}\setminus \textnormal{diff}(C,D)$, $\chi(A)=Diag^G(A)\in Ch_C(A)=Ch_D(A)$. For $A\in \textnormal{diff}(C,D)$, $\chi(A)\in Ch_D(A)$ by assumption. Thus $\chi\in Ch(D)$. Because $D$ is $\mathcal{H}$-random, there is $G'\in \mathcal{H}$ such that $G'\unlhd_{\chi}D$. We show $\textnormal{dist}(G,G')\leq \delta$. By definition of $\chi$ and since $G'\unlhd_{\chi}D$, we have that for all $A\in {W\choose r}$, if $A \notin \textnormal{diff}(C,D)$, then $Diag^{G'}(A)=\chi(A)=Diag^G(A)$, which implies $A\notin \textnormal{diff}(G,G')$. Thus $\textnormal{diff}(G,G')\subseteq \textnormal{diff}(C,D)$ so $|\textnormal{diff}(G,G')|\leq \delta {n\choose r}$ and $\textnormal{dist}(G,G')\leq \delta$ by definition. \end{proof} \noindent{\bf Proof of Theorem \ref{b4stab}}. Let $\mathcal{H}$ be a fast-growing hereditary $\mathcal{L}$-property. Fix $\epsilon$ and $\delta>0$. Given $n$, let $A(n, \epsilon, \delta)=\mathcal{H}_n\setminus E^{\delta}(\epsilon, n,\mathcal{H})$. Recall, we want to show there is $\beta>0$ such that for sufficiently large $n$, \begin{align}\label{m} \frac{|A(n,\epsilon, \delta)|}{|\mathcal{H}_n|}\leq 2^{-\beta {n\choose r}}. \end{align} Let $\gamma>0$ be as in Lemma \ref{deltaclose1} for $\mathcal{H}$. Choose $K>2r$ sufficiently large so that $1-\epsilon +\gamma \delta /K<1-\epsilon/2$. Apply Theorem \ref{COROLLARY2} to $\frac{\delta}{K}$ to obtain constants $c$ and $m>1$. Assume $n$ is sufficiently large. Then Theorem \ref{COROLLARY2} implies there is a collection $\mathcal{C}$ of $\mathcal{L}_{\mathcal{H}}$-templates with domain $[n]$ such that the following hold. \begin{enumerate}[(i)] \item For every $H\in \mathcal{H}_n$, there is $C\in \mathcal{C}$ such that $H\unlhd_pC$. \item For every $C\in \mathcal{C}$, there is $C'\in \mathcal{R}([n],\mathcal{H})$ such that $dist(C,C')\leq \delta$. \item $\log |\mathcal{C}|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} Suppose $G\in A(n,\epsilon,\delta)$. By (i), there is $C\in \mathcal{C}_n$ such that $G\unlhd_pC$. By (ii), there is $M_C\in \mathcal{R}([n],\mathcal{H})$ such that $\textnormal{dist}(C,M_C)\leq \frac{\delta}{K}$. By Lemma \ref{deltaclose2}, there is $G'\unlhd_p M_C$ with $\textnormal{dist}(G,G')\leq \frac{\delta}{K}\leq \delta$. If $sub(M_C)\geq \textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$, then by definition of $E^{\delta}(\epsilon, n,\mathcal{H})$, $\textnormal{dist}(G,G')\leq \delta$ and $G'\unlhd_pM_C$ would imply $G\in E^{\delta}(\epsilon, n,\mathcal{H})$, contradicting our assumption that $G\in A(n,\epsilon, \delta)=\mathcal{H}_n\setminus E^{\delta}(\epsilon, n,\mathcal{H})$. Therefore, we must have $sub(M_C)<\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$. Note $M_C\in \mathcal{R}([n],\mathcal{H})$ implies $M_C$ is error-free, so Lemma \ref{deltaclose1} and the fact that $\textnormal{dist}(C,M_C)\leq \delta/K$ imply $sub(C)\leq sub(M_C)\textnormal{ex}(n,\mathcal{H})^{\gamma\delta/K}$. Combining this with the fact that $sub(M_C)<\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$ we have that $$ sub(C)<\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}\textnormal{ex}(n,\mathcal{H})^{\gamma\delta/K}=\textnormal{ex}(n,\mathcal{H})^{1-\epsilon +\gamma \delta/K}\leq\textnormal{ex}(n,\mathcal{H})^{1-\epsilon/2}, $$ where the second inequality is by assumption on $K$. Therefore every $G\in A(n,\epsilon, \delta)$ can be constructed as follows. \begin{itemize} \item Choose $C\in \mathcal{C}_n$ with $sub(C)<\textnormal{ex}(n,\mathcal{H})^{1-\epsilon/2}$. There are at most $|\mathcal{C}_n|\leq 2^{cn^{r-\frac{1}{m}}\log n}$ ways to do this, where the bound is by (iii). Since $n$ is large and $\pi(\mathcal{H})>1$, we may assume $2^{cn^{r-\frac{1}{m}}\log n}\leq \pi(\mathcal{H})^{\epsilon{n\choose r}/4}$. \item Choose a full subpattern of $C$. There are at most $sub(C)<\textnormal{ex}(n,\mathcal{H})^{1-\epsilon/2}$ ways to do this. \end{itemize} Combining these bounds yields $|A(n,\epsilon,\delta)|\leq \pi(\mathcal{H})^{\epsilon{n\choose r}/4}\textnormal{ex}(n,\mathcal{H})^{1-\epsilon/2}$. Recall that $|\mathcal{H}_n|\geq\textnormal{ex}(n,\mathcal{H})$ holds, since for any $M\in \mathcal{R}_{ex}([n],\mathcal{H})$, all $\textnormal{ex}(n,\mathcal{H})$-many full subpatterns of $M$ are all in $\mathcal{H}_n$. Therefore \begin{align}\label{l} \frac{|A(n,\epsilon, \delta)|}{|\mathcal{H}_n|}\leq \frac{\pi(\mathcal{H})^{\epsilon{n\choose r}/4}\textnormal{ex}(n,\mathcal{H})^{1-\epsilon/2}}{\textnormal{ex}(n,\mathcal{H})}=\pi(\mathcal{H})^{\epsilon{n\choose r}/4}\textnormal{ex}(n,\mathcal{H})^{-\epsilon/2}\leq \pi(\mathcal{H})^{-\epsilon {n\choose r}/4}, \end{align} where the last inequality is because $\pi(\mathcal{H})^{n\choose r}\leq\textnormal{ex}(n,\mathcal{H})$. Therefore we have $\frac{|A(n,\epsilon, \delta)|}{|\mathcal{H}_n|}\leq 2^{-\beta{n\choose r}}$, where $\beta = \frac{\epsilon \log \pi(\mathcal{H})}{4\log 2}$. Note $\beta>0$ since $\pi(\mathcal{H})>1$. \qed \vspace{3mm} \noindent {\bf Proof of Theorem \ref{stab}.} Suppose $\mathcal{H}$ is a fast growing hereditary $\mathcal{L}$-property with a stability theorem. Fix $\delta>0$. Recall we want to show there is $\beta>0$ such that for sufficiently large $n$, \begin{align*} \frac{|\mathcal{H}_n\setminus E^{\delta}(n,\mathcal{H})|}{|\mathcal{H}_n|}\leq 2^{-\beta {n\choose r}} \end{align*} By Theorem \ref{b4stab}, it suffices to show that there are $\epsilon_1, \delta_1>0$ such that for all sufficiently large $n$, $E^{\delta_1}(\epsilon_1, n,\mathcal{H})\subseteq E^{\delta}(n,\mathcal{H})$. Because $\mathcal{H}$ has a stability theorem, there is $\epsilon$ such that for all sufficiently large $n$, if $H\in \mathcal{R}([n],\mathcal{H})$ satisfies $sub(H)\geq\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$, then there is $H'\in \mathcal{R}_{ex}([n],\mathcal{H})$ with $\textnormal{dist}(H,H')\leq \frac{\delta}{2}$. Fix $n$ sufficiently large. We claim $E^{\delta/2}(\epsilon, n,\mathcal{H})\subseteq E^{\delta}(n,\mathcal{H})$. Suppose $G\in E^{\delta/2}(\epsilon, n,\mathcal{H})$. Then by definition, $G$ is $\delta/2$-close to some $G'$ such that $G'\unlhd_pH$, for some $H\in \mathcal{R}([n],\mathcal{H})$ satisfying $sub(H)\geq\textnormal{ex}(n,\mathcal{H})^{1-\epsilon}$. By choice of $\epsilon$ and because $n$ is sufficiently large, there is $H'\in \mathcal{R}_{ex}([n],\mathcal{H})$ such that $\textnormal{dist}(H,H')\leq \frac{\delta}{2}$. Lemma \ref{deltaclose2} implies there is some $G''\unlhd_pH'$ such that $\textnormal{dist}(G',G'')\leq \frac{\delta}{2}$. Observe that $G''\in E(n,\mathcal{H})$ and $$ \textnormal{dist}(G, G'')\leq \textnormal{dist}(G,G')+\textnormal{dist}(G', G'') \leq \frac{\delta}{2}+\frac{\delta}{2}=\delta. $$ This implies that $G\in E^{\delta}(n,\mathcal{H})$, as desired. \qed \section{Characterization of $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-templates}\label{Hrandom} In this section we give an equivalent characterization for when an $\mathcal{L}_{\mathcal{H}}$-structure is an $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-template, where $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. The results in this section will be used in the proofs of our remaining results, Theorems \ref{version1}, \ref{GENSUPERSAT}, and \ref{COROLLARY2}. For the rest of this section, $\mathcal{H}$ is a fixed nonempty collection of finite $\mathcal{L}$-structures. \begin{definition} Define $\textnormal{FLAW}$ to be the class of all $\mathcal{L}_{\mathcal{H}}$-structures of size $r$ which are not $\mathcal{L}_{\mathcal{H}}$-templates. Elements of $\textnormal{FLAW}$ are called flaws. \end{definition} \begin{lemma}\label{flaw} An $\mathcal{L}_{\mathcal{H}}$-structure $M$ is an $\mathcal{L}_{\mathcal{H}}$-template if and only if it is $\textnormal{FLAW}$-free. \end{lemma} \begin{proof} Let $dom(M)=V$. It is straightforward from Definition \ref{templatedef} to check that $M$ is an $\mathcal{L}_{\mathcal{H}}$-template if and only if for all $A\in {V\choose r}$, $M[A]$ is an $\mathcal{L}_{\mathcal{H}}$-template. By definition of $\textnormal{FLAW}$, $M$ is $\textnormal{FLAW}$-free if and only if for all $A\in {V\choose r}$, $M[A]$ is an $\mathcal{L}_{\mathcal{H}}$-template. This finishes the proof. \end{proof} \noindent We are now ready to prove the main result of this section. \begin{proposition}\label{random} Suppose $\mathcal{H}$ is a hereditary $\mathcal{L}$-property, and $\mathcal{F}$ is the class of finite $\mathcal{L}$-structures from Observation \ref{HP} such that $\operatorname{Forb}(\mathcal{F})=\mathcal{H}$. Then a complete $\mathcal{L}_{\mathcal{H}}$-structure $M$ is $\mathcal{H}$-random if and only if $M$ is $\tilde{\mathcal{F}}$-free and error-free. \end{proposition} \begin{proof} By Observation \ref{HP}, $\mathcal{F}$ is closed under isomorphism. Fix a complete $\mathcal{L}_{\mathcal{H}}$-structure $M$ and let $V=dom(M)$. Suppose first that $M$ is $\mathcal{H}$-random. Then $M$ is complete and for every $\chi\in Ch(M)$, there is $N\in \mathcal{H}$ such that $N\unlhd_{\chi}M$. This implies by Proposition \ref{Lrandom} that $M$ is error-free. Suppose by contradiction $M$ is not $\tilde{\mathcal{F}}$-free. Combining the assumption that $\mathcal{F}$ is closed under isomorphism and the definition of $\tilde{\mathcal{F}}$, this implies there is $B\subseteq V$ and $F\in \mathcal{F}$ such that $M[B]\in \tilde{F}$. By definition of $\tilde{F}$, there is $\chi_B\in Ch(M[B])$ such that $F\unlhd_{\chi_B}M[B]$. Define a function $\chi:{V\choose r}\rightarrow S_r(C_V,\mathcal{H})$ as follows. For each $A\in {B\choose r}$, set $\chi(A)=\chi_B(A)$. Clearly, $\chi_B\in Ch(M[B])$ implies that for all $A\in {B\choose r}$, $\chi_B(A)\in Ch_M(A)$. For each $A\in {V\choose r}\setminus {B\choose r}$, define $\chi(A)$ to be any element of $Ch_M(A)$ (this is possible since $M$ is complete by assumption). By construction, $\chi\in Ch(M)$. Because $M$ is $\mathcal{H}$-random, there is $D\in \mathcal{H}$ such that $D\unlhd_{\chi}M$. By choice of $\mathcal{F}$, since $D\in \mathcal{H}$, we have that $D$ is $\mathcal{F}$-free, which implies $D$ is $F$-free since $F\in \mathcal{F}$. We claim $D[B]\cong_{\mathcal{L}}F$, a contradiction. For each $A\in {B\choose r}$, $D\unlhd_{\chi}M$, $F\unlhd_{\chi_B}M[B]$, and the definition of $\chi$ imply $$ Diag^D(A)=\chi(A)=\chi_B(A)=Diag^F(A). $$ Thus $Diag(D[B])=\bigcup_{A\in {B\choose r}}Diag^D(A)=\bigcup_{A\in {B\choose r}}Diag^F(A)=Diag(F)$ implies $D[B]\cong_{\mathcal{L}}F$. For the converse, suppose $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure which is $\tilde{\mathcal{F}}$-free and error-free. Suppose by contradiction $M$ is not $\mathcal{H}$-random. Then there is $\chi\in Ch(M)$ such that there is no $N\in \mathcal{H}$ with $N\unlhd_{\chi}M$. Since $M$ is error-free, Proposition \ref{Lrandom} implies there is some $\mathcal{L}$-structure $N$ such that $N\unlhd_{\chi}M$. Thus we must have $N\notin \mathcal{H}$. By choice of $\mathcal{F}$ from Observation \ref{HP}, $N$ is not $\mathcal{F}$-free. This along with the fact that $\mathcal{F}$ is closed under isomorphism implies there is $B\subseteq V$ such that $N[B]\in \mathcal{F}$. But $N\unlhd_{p}M$ implies $N[B]\unlhd_{p}M[B]$ (this is straightforward to check). Since $N[B]\in \mathcal{F}$, this implies $M[B]\in \tilde{\mathcal{F}}$ by definition of $\tilde{\mathcal{F}}$, contradicting that $M$ is $\tilde{\mathcal{F}}$-free. \end{proof} \begin{corollary}\label{charR} Suppose $\mathcal{H}$ is a hereditary $\mathcal{L}$-property, and $\mathcal{F}$ is the class of finite $\mathcal{L}$-structures from Observation \ref{HP} such that $\operatorname{Forb}(\mathcal{F})=\mathcal{H}$. Let $M$ be an $\mathcal{L}_{\mathcal{H}}$-structure. Then $M\in \mathcal{R}(dom(M),\mathcal{H})$ if and only if $M$ is $\tilde{\mathcal{F}}$-free, error-free, and $\textnormal{FLAW}$-free. \end{corollary} \begin{proof} By definition, $M\in \mathcal{R}(dom(M),\mathcal{H})$ if and only if $M$ is an $\mathcal{H}$-random $\mathcal{L}_{\mathcal{H}}$-template. By Lemma \ref{flaw} and Proposition \ref{random}, this holds if and only if $M$ is $\tilde{\mathcal{F}}$-free, error-free, and $FLAW$-free. \end{proof} \section{Graph Removal and Proofs of Theorems \ref{GENSUPERSAT} and \ref{COROLLARY2}.}\label{rphrem} In this section we will use a version of the graph removal lemma from \cite{AroskarCummings} to prove Theorem \ref{GENSUPERSAT} and to prove Theorem \ref{COROLLARY2} from Theorem \ref{version1}. We now state definitions required to quote the graph removal lemma from \cite{AroskarCummings}. Throughout the rest of this section, $\mathcal{L}_0$ is a fixed finite relational language with $r_{\mathcal{L}_0}=r$. Note $\mathcal{L}_0$ is not necessarily the same as $\mathcal{L}$, although we are assuming $r_{\mathcal{L}_0}=r_{\mathcal{L}}=r$. Given a partition $p$ of a finite set $X$, let $||p||$ denote the number of parts in $p$. \begin{definition} Let $Index=\{(R,p): R\in \mathcal{L}_0$ and $p$ is a partition of $[\ell]$ where $\ell$ is the arity of $R\}$. Suppose $(R,p)\in Index$ and $R$ has arity $\ell$. \begin{enumerate} \item $C_p(x_1,\ldots, x_{\ell})$ is the subtuple of $(x_1,\ldots, x_{\ell})$ obtained by replacing each $x_i$ with $x_{p(i)}$ where $p(i)=\min \{j: x_j$ is in the same part of $p$ as $i\}$, then deleting all but the first occurance of each variable in the tuple $(x_{p(1)},\ldots, x_{p(\ell)})$. \item $R_p(C(\bar{x}))$ is the $||p||$-ary relation obtained from $R(x_1, \ldots, x_{\ell})$ by replacing each $x_i$ with $x_{p(i)}$ where $p(i)=\min \{j: x_j$ is in the same part of $p$ as $i\}$. \item If $N$ is an $\mathcal{L}_0$-structure, define $DH^{R}_p(N)=\{\bar{a}\in dom(N)^{\underline{||p||}}: N\models R_p(\bar{a})\}$. \end{enumerate} \end{definition} Now we can define the notion of distance between two $\mathcal{L}_0$-structures from \cite{AroskarCummings}. \begin{definition} Given $(R,p)\in Index$ and $M$, $N$ two finite $\mathcal{L}_0$-structures with the same universe $W$, set $$ d_p^R(M,N)=\frac{|DH_p^R(M)\Delta DH_p^R(N)|}{|W|^{||p||}}\qquad \hbox{ and set }\qquad d(M,N)=\sum_{(R,p)\in Index}d^R_p(M,N). $$ \end{definition} \noindent We will see below in Lemma \ref{distlem} that this notion of distance, $d(M,N)$, is related to our notion of distance, $\textnormal{dist}(M,N)$. We first state the graph removal lemma of Aroskar and Cummings as it appears in their paper (Theorem 2 from \cite{AroskarCummings}). \begin{theorem}[{\bf Aroskar-Cummings \cite{AroskarCummings}}]\label{triangleremoval} Suppose $\mathcal{A}$ is a collection of finite $\mathcal{L}_0$-structures. For every $\delta>0$ there exists $\epsilon>0$ and $K$ such that the following holds. For all sufficiently large finite $\mathcal{L}_0$-structures $M$, if $\textnormal{prob}(\mathcal{A}(K),M)<\epsilon$, then there is an $\mathcal{L}_0$-structure $M'$ with $dom(M')=dom(M)$ such that $d(M',M)<\delta$ and $\textnormal{prob}(\mathcal{A},M')=0$. \end{theorem} The following relationship between $d(M,N)$ and $\textnormal{dist}(M,N)$ will allow us to restate this graph removal lemma. Given a tuple $\bar{x}=(x_1,\ldots, x_{\ell})$, a \emph{subtuple} of $\bar{x}$ is any tuple $\bar{x}'=(x_{i_1},\ldots, x_{i_{\ell'}})$ where $1\leq i_1<\ldots<i_{\ell'}\leq \ell$. If $\ell'<\ell$, we say $\bar{x}'$ is a \emph{proper subtuple} of $\bar{x}$, denoted $\bar{x}'\subsetneq \bar{x}$. \begin{lemma}\label{distlem} If $M$ and $N$ are $\mathcal{L}_0$-structures with the same finite domain $W$ of size at least $2r$, then $$ \textnormal{dist}(M,N)\leq (r!)^22^r d(M,N). $$ \end{lemma} \begin{proof} Let $n=|W|$. Note that $n\geq 2r$ implies for all $1\leq \ell \leq r$, \begin{align}\label{d} \frac{n!}{(n-\ell)!}=n\cdot (n-1)\cdots (n-\ell+1)\geq (n-\ell +1)^{\ell}\geq (n/2)^{\ell}=n^{\ell}/2^{\ell}. \end{align} Given $1\leq \ell\leq r$, define $$ \textnormal{diff}^{\ell}(M,N)=\{\bar{a}\in W^{\underline{\ell}}: qftp^M(\bar{a})\neq qftp^N(\bar{a})\text{ and for all $\bar{a}'\subsetneq \bar{a}$, $qftp^M(\bar{a}')= qftp^N(\bar{a}')$}\}. $$ Observe that elements in $\textnormal{diff}(M,N)$ are \emph{sets} of elements from $W$, while elements in $\textnormal{diff}^{\ell}(M,N)$ are \emph{tuples} of elements of $W$. Clearly if $A\in \textnormal{diff}(M,N)$, there is some $\ell\in [r]$ and a tuple $\bar{a}\in A^{\underline{\ell}}$ such that $\bar{a} \in \textnormal{diff}^{\ell}(M,N)$. Define $\Psi:\textnormal{diff}(M,N)\rightarrow \bigcup_{\ell\in [r]}\textnormal{diff}^{\ell}(M,N)$ to be any map which sends each $A\in \textnormal{diff}(M,N)$ to some such tuple. Given $\ell\in [r]$ and $\bar{a}=(a_1,\ldots, a_{\ell})\in \textnormal{diff}^{\ell}(M,N)$, note that $$ \Psi^{-1}(\bar{a})\subseteq \{A\in {W\choose r}: \cup \bar{a} \subseteq A\}. $$ Since the right hand side has size ${n-\ell \choose r-\ell}$, we have that for all $\bar{a} \in \textnormal{diff}^{\ell}(M,N)$, $|\Psi^{-1}(\bar{a})|\leq {n-\ell \choose r-\ell}$. For each $\ell\in [r]$, we now define a map $f_{\ell}:\textnormal{diff}^{\ell}(M,N)\rightarrow \bigcup_{(R,p)\in Index, ||p||=\ell}DH_p^R(M)\Delta DH_p^R(N)$. Let $\bar{a}\in \textnormal{diff}^{\ell}(M,N)$. Since $\bar{a}\in \textnormal{diff}^{\ell}(M,N)$, there is a relation $R(x_1,\ldots, x_t)\in \mathcal{L}_0$ and a map $h:[\ell]\rightarrow [t]$ such that $M\models R(a_{h(1)},\ldots, a_{h(t)})$ and $N\models \neg R(a_{h(1)},\ldots, a_{h(t)})$ or vice versa. If $h$ is not surjective, then some permutation of $C_p(a_{h(1)},\ldots, a_{h(t)})$ is a proper subtuple $\bar{a}'$ of $\bar{a}$ such that $qftp^M(\bar{a}')\neq qftp^N(\bar{a}')$. But this contradicts that $\bar{a} \in \textnormal{diff}^{\ell}(M,N)$. Thus $h$ is surjective. Let $p$ be the partition of $[t]$ with parts $h^{-1}(\{1\}), \ldots, h^{-1}(\{\ell\})$. Since $h$ is surjective, the parts are all nonempty, so $||p||=\ell$. Then by definition, $C_p(a_{h(1)},\ldots, a_{h(t)}) \in DH^R_p(M)\Delta DH^R_p(N)$. Define $f_{\ell}(\bar{a})=C_p(a_{h(1)},\ldots, a_{h(t)})$. Observe that $\cup C_p(a_{h(1)},\ldots, a_{h(t)})=\cup \bar{a}$ implies $$ f^{-1}_{\ell}(f_{\ell}(\bar{a}))\subseteq \{\bar{b}\in W^{\ell}: \cup \bar{b} = \cup \bar{a}\}, $$ so $|f^{-1}_{\ell}(f_{\ell}(\bar{a}))|\leq \ell !$. Thus $f_{\ell}: \textnormal{diff}^{\ell}(M,N)\rightarrow \bigcup_{(R,p)\in Index, ||p||=\ell}DH_p^R(M)\Delta DH_p^R(N)$ and \begin{align}\label{t} \text{ for all }\bar{c} \in \bigcup_{(R,p)\in Index, ||p||=\ell}DH_p^R(M)\Delta DH_p^R(N),\quad |f_{\ell}^{-1}(\bar{c})|\leq \ell!. \end{align} Define a map $\beta: \textnormal{diff}(M,N)\rightarrow \bigcup_{(R,p)\in Index}DH_p^R(M)\Delta DH_p^R(N)$ as follows. Given $A\in \textnormal{diff}(M,N)$, apply $\Psi$ to obtain $\Psi(A)\in \textnormal{diff}^{\ell}(M,N)$ for some $\ell\in [r]$. Then define $$ \beta(\bar{a}):=f_{\ell}(\Psi(\bar{a}))\in \bigcup_{(R,p)\in Index, ||p||=\ell}DH_p^R(M)\Delta DH_p^R(N). $$ Suppose $\bar{c} \in \bigcup_{(R,p)\in Index}DH_p^R(M)\Delta DH_p^R(N)$ and $\ell:=|\bar{c}|$. Then $\bar{c}\in DH_p^R(M)\Delta DH_p^R(N)$ for some $(R,p)\in Index$ with $||p||=\ell$. By definition of $\beta$, $\beta^{-1}(\bar{c})=\Psi^{-1}(f^{-1}_{\ell}(\bar{c}))$. Combining (\ref{t}) and the fact that $|\Psi^{-1}(\bar{a})|\leq {n-\ell\choose r-\ell}$ for all $\bar{a} \in \textnormal{diff}^{\ell}(M,N)$, we have that $$ |\beta^{-1}(\bar{c})|=|\Psi^{-1}(f^{-1}_{\ell}(\bar{c}))|\leq {n-\ell\choose r-\ell}\ell!. $$ This shows that $|\textnormal{diff}(M,N)|\leq \sum_{\ell\in[r]} \sum_{(R,p)\in Index,||p||=\ell}{n-\ell\choose r-\ell}\ell!|DH_p^R(M)\Delta DH_p^R(N)|$. Dividing both sides of this by ${n\choose r}$, we obtain the following. \begin{align}\label{dd} \textnormal{dist}(M,N)&\leq \sum_{\ell\in[r]} \sum_{(R,p)\in Index,||p||=\ell}\frac{{n-\ell\choose r-\ell}\ell!}{{n\choose r}}|DH_p^R(M)\Delta DH_p^R(N)|. \end{align} Note that for all $1\leq \ell <r$, $$ \frac{{n-\ell\choose r-\ell}\ell!}{{n\choose r}} = \frac{(n-\ell)!}{n!}\frac{\ell! r!}{(r-\ell)!}\leq \frac{2^{\ell}}{n^{\ell}}\frac{\ell! r!}{(r-\ell)!}< \frac{(r!)^22^r}{n^{\ell}}, $$ where the first inequality is by (\ref{d}) and the last is because $\ell<r$. If $\ell=r$, then $$ \frac{{n-\ell\choose r-\ell}\ell!}{{n\choose r}} =\frac{r!}{{n\choose r}}=\frac{(r!)^2(n-r)!}{n!} \leq \frac{(r!)^22^r}{n^r}, $$ where the inequality is by (\ref{d}). Thus for all $\ell\in [r]$, $\frac{{n-\ell\choose r-\ell}\ell!}{{n\choose r}}\leq \frac{(r!)^22^r}{n^{\ell}}$. Combining this with (\ref{dd}) yields $$ \textnormal{dist}(M,N)\leq (r!)^22^r \sum_{\ell\in[r]} \sum_{(R,p)\in Index,||p||=\ell}\frac{|DH_p^R(M)\Delta DH_p^R(N)|}{n^\ell}=(r!)^22^rd(M,N). $$ \end{proof} \noindent We will use the following version of Theorem \ref{triangleremoval}, now adapted to our notation. \begin{theorem}\label{triangleremoval2} Suppose $\mathcal{A}$ is a collection of finite $\mathcal{L}_0$-structures. For every $\delta>0$ there exists $\epsilon>0$ and $K$ such that the following holds. For all sufficiently large finite $\mathcal{L}_0$-structures $M$, if $\textnormal{prob}(\mathcal{A}(K),M)<\epsilon$, then there is an $\mathcal{L}_0$-structure $M'$ with $dom(M')=dom(M)$ such that $\textnormal{dist}(M',M)<\delta$ and $\textnormal{prob}(\mathcal{A},M')=0$. \end{theorem} \begin{proof} Fix $\delta>0$. Let $\delta'=\frac{\delta}{(r!)^22^r }$ and choose $K=K(\delta')$ and $\epsilon=\epsilon(\delta')$ by applying Theorem \ref{triangleremoval} to $\delta'$ and $\mathcal{A}$. Suppose $n$ is sufficiently large so that Theorem \ref{triangleremoval} applies to structures of size $n$. Suppose $M$ is an $\mathcal{L}_0$-structure of size $n$ such that $\textnormal{prob}(\mathcal{A}(K),M)<\epsilon$. Then Theorem \ref{triangleremoval} implies there is an $\mathcal{L}_0$-structure $M'$ with $dom(M')=dom(M)$ such that $d(M',M)<\delta'$ and $\textnormal{prob}(\mathcal{A},M')=0$. Combining this with Lemma \ref{distlem}, we have $\textnormal{dist}(M',M)\leq (r!)^22^r d(M',M)<(r!)^22^r \delta' = \delta$. \end{proof} \noindent{\bf Proof of Theorem \ref{GENSUPERSAT}.} Let $\mathcal{H}$ be a nontrivial hereditary $\mathcal{L}$-property and let $\mathcal{F}$ be as in Observation \ref{HP} so that $\mathcal{H}=\operatorname{Forb}(F)$. Recall we want to show that for all $\delta>0$, there are $\epsilon>0$ and $K$ such that for sufficiently large $n$, for any $\mathcal{L}_{\mathcal{H}}$-template $M$ of size $n$, if $\textnormal{prob}(\tilde{\mathcal{F}}(K) \cup \mathcal{E}(K),M)\leq \epsilon$ then \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $sub(M)\leq \textnormal{ex}(n,\mathcal{H})^{1+\delta}$. \item If $\pi(\mathcal{H})\leq 1$, then $sub(M)\leq 2^{\delta {n\choose r}}$. \end{enumerate} Fix $\delta>0$. Apply Lemma \ref{deltaclose1} to $\mathcal{H}$ to obtain $\gamma>0$. Let $\mathcal{A}=\tilde{\mathcal{F}}\cup \mathcal{E}\cup \textnormal{FLAW}$. Apply Theorem \ref{triangleremoval2} to obtain $K$ and $\epsilon$ for $\delta/2\gamma$ and $\mathcal{A}$. Suppose $n$ is sufficiently large and $M$ is an $\mathcal{L}_{\mathcal{H}}$-template of size $n$ satsifying $\textnormal{prob}(\tilde{\mathcal{F}}(K)\cup \mathcal{E}(K), M)<\epsilon$. Because $M$ is an $\mathcal{L}_{\mathcal{H}}$-template, Lemmas \ref{flaw} implies for all $B\in \textnormal{FLAW}$, $\textnormal{prob}(B,M)=0$. Therefore $\textnormal{prob}(\mathcal{A}(K), M)<\epsilon$, so by Theorem \ref{triangleremoval2}, there is an $\mathcal{L}_{\mathcal{H}}$-structure $M'$ with $dom(M)=dom(M')$ such that $\textnormal{prob}(\mathcal{A},M')=0$ and $\textnormal{dist}(M,M')\leq \delta/2\gamma$. Since $\textnormal{prob}(\mathcal{A},M')=0$, Corollary \ref{charR} implies $M'\in \mathcal{R}(n,\mathcal{H})$. Thus $sub(M')\leq \textnormal{ex}(n,\mathcal{H})$ holds by definition of $\textnormal{ex}(n,\mathcal{H})$. Combining this with Lemma \ref{deltaclose1} (note $M'\in \mathcal{R}(n,\mathcal{H})$ implies $M'$ is error-free), we have the following. \begin{enumerate} \item If $\pi(\mathcal{H})>1$, then $sub(M)\leq sub(M')\textnormal{ex}(n,\mathcal{H})^{\gamma (\delta/2\gamma)}=sub(M')\textnormal{ex}(n,\mathcal{H})^{\delta/2}\leq \textnormal{ex}(n,\mathcal{H})^{1+\delta/2}$. \item If $\pi(\mathcal{H})=1$, then $sub(M)\leq sub(M')2^{\gamma (\delta/2\gamma) {n\choose r}}=sub(M')2^{\delta{n\choose r}/2}\leq \textnormal{ex}(n,\mathcal{H})2^{\delta/2 {n\choose r}}$. \end{enumerate} We are done in the case where $\pi(\mathcal{H})>1$. If $\pi(\mathcal{H})=1$, assume $n$ is sufficiently large so that $\textnormal{ex}(n,\mathcal{H})\leq 2^{\delta/2{n\choose r}}$. Then (2) implies $sub(M)\leq 2^{\delta{n\choose r}}$, as desired. \qed \vspace{3mm} \noindent {\bf Proof of Theorem \ref{COROLLARY2} from Theorem \ref{version1}.} Suppose $\mathcal{H}$ is a hereditary $\mathcal{L}$-property. Let $\mathcal{F}$ be the class of finite $\mathcal{L}$-structures from Observation \ref{HP} so that $\mathcal{H}=\operatorname{Forb}(\mathcal{F})$. Then for each $n$, $\mathcal{H}_n$ is the set of all $\mathcal{F}$-free $\mathcal{L}$-structures with domain $[n]$. Let $\mathcal{A}=\tilde{\mathcal{F}}\cup \mathcal{E}\cup \textnormal{FLAW}$. Fix $\delta>0$ and choose $K$ and $\epsilon$ as in Theorem \ref{triangleremoval2} for $\delta$ and the family $\mathcal{A}$. By replacing $K$ if necessary, assume $K\geq r$. Apply Theorem \ref{version1} to $\mathcal{B}:=\mathcal{F}(K)$ to obtain $c=c(K,r,\mathcal{L},\epsilon)$, $m=m(K,r)$. Observe the choice of $K$ depended on $\mathcal{H}$ and $r=r_{\mathcal{L}}$, so $m=m(\mathcal{H}, r_{\mathcal{L}})$. Since $r_{\mathcal{L}}$ depends on $\mathcal{L}$, $c=c(\mathcal{H},\mathcal{L},\epsilon)$. Let $n$ be sufficiently large. Then Theorem \ref{version1} applied to $W=[n]$ implies there is a collection $\mathcal{C}$ of $\mathcal{L}_{\mathcal{B}}$-templates with domain $[n]$ such that the following hold. \begin{enumerate}[(i)] \item For all $\mathcal{F}(K)$-free $\mathcal{L}$-structures $M$ with domain $[n]$, there is $C\in \mathcal{C}$ such that $M\unlhd_pC$. \item For all $C\in \mathcal{C}$, $\textnormal{prob}(\widetilde{\mathcal{F(K)}},C)\leq \epsilon$ and $\textnormal{prob}(\mathcal{E},C)\leq \epsilon$. \item $\log |\mathcal{C}|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} We show this $\mathcal{C}$ satisfies the conclusions of Theorem \ref{COROLLARY2} with $c$, $m$ and $\delta$. Note that because $K\geq r$, $S_r(\mathcal{H})=S_r(\mathcal{B})$, so all $\mathcal{L}_{\mathcal{B}}$-templates are also $\mathcal{L}_{\mathcal{H}}$-templates. In particular the elements in $\mathcal{C}$ are all $\mathcal{L}_{\mathcal{H}}$-templates. Clearly (iii) implies part (3) of Theorem \ref{COROLLARY2} holds. For part (1), since any $H\in \mathcal{H}_n$ is $\mathcal{F}$-free, it is also $\mathcal{F}(K)$-free, so (i) implies there is $C\in \mathcal{C}$ such that $H\unlhd_pC$. This shows part (1) of Theorem \ref{COROLLARY2} holds. For part (2), fix $C\in \mathcal{C}$. Since $C$ is an $\mathcal{L}_{\mathcal{H}}$-template, Lemma \ref{flaw} implies $\textnormal{prob}(G,C)=0$ for all $G\in \textnormal{FLAW}$. Then (ii) implies that for all $G\in \widetilde{\mathcal{F}(K)}\cup \mathcal{E}$, $\textnormal{prob}(G,C)\leq \epsilon$. Since $\widetilde{\mathcal{F}(K)}=\tilde{\mathcal{F}}(K)$, these facts imply that for all $G\in \mathcal{A}(K)$, $\textnormal{prob}(G,C)\leq \epsilon$. Thus Theorem \ref{triangleremoval2} implies there is an $\mathcal{L}_{\mathcal{H}}$-structure $C'$ with $dom(C)=dom(C')=[n]$ such that $\textnormal{dist}(C,C')\leq \delta$ and $\textnormal{prob}(\mathcal{A},C')=0$. Since $\textnormal{prob}(\mathcal{A},C')=0$, $C'$ is a $\textnormal{FLAW}$-free, $\tilde{\mathcal{F}}$-free, and error-free $\mathcal{L}_{\mathcal{H}}$-structure with domain $[n]$, so by Corollary \ref{charR}, $C'\in \mathcal{R}([n],\mathcal{H})$. This finishes the proof. \qed \section{A Reduction}\label{section2thm1} We have now proved all the results in this paper except Theorem \ref{version1}. In this section we prove Theorem \ref{version1} by reducing it to another result, Theorem \ref{VERSION2} (which is proved in Section \ref{VERSION2pf}). \subsection{Preliminaries}\label{prelims} In this subsection we give preliminaries necessary for the statement of Theorem \ref{VERSION2}. Many of these notions are similar to definitions from Section \ref{tildeLstructures}. However, we will see that our proofs necessitate this more syntactic treatment. \begin{definition}\label{chd2} Suppose $C$ is a set of constants and $\sigma \subseteq S_r(C)$. \begin{enumerate}[$\bullet$] \item $V(\sigma)=\{c\in C: c$ appears in some $p(\bar{c})\in \sigma\}$. \item Given $A\in {V(\sigma)\choose r}$, let $Ch_{\sigma}(A)=\{p(\bar{c}) \in \sigma: \cup \bar{c} =A\}$. Elements of $Ch_{\sigma}(A)$ are \emph{choices for $A$}. \item We say $\sigma$ is \emph{complete} if $Ch_{\sigma}(A)\neq \emptyset$, for all $A\in {V(\sigma)\choose r}$. \end{enumerate} \end{definition} \begin{example}\label{ex6} Let $\mathcal{L}$ and $\mathcal{P}$ be as in Example \ref{ex1} (i.e. metric spaces with distances in $\{1,2,3\}$). Let $W=\{u,v,w\}$ and $\sigma=\{p_1(c_u,c_v), p_2(c_u,c_v), p_2(c_u,c_w)\}\subseteq S_2(C_W,\mathcal{P})$. Then $V(\sigma)=\{c_u,c_v,c_w\}$ and it is easy to check that $Ch_{\sigma}(c_uc_v)=\{p_1(c_u,c_v), p_2(c_u,c_v)\}$, $Ch_{\sigma}(c_uc_w)=\{p_2(c_u,c_w)\}$, and $Ch_{\sigma}(c_vc_w)=\emptyset$. Observe, this $\sigma$ is not complete. \end{example} \begin{definition}\label{syndiag} Suppose $C$ is a set of $n$ constants and $\sigma \subseteq S_r(C)$. Given $m\leq n$, $\sigma$ is a \emph{syntactic $m$-diagram} if $|V(\sigma)|=m$ and for all $A\in {V(\sigma)\choose r}$, $|Ch_{\sigma}(A)|=1$. \end{definition} We will say $\sigma\subseteq S_r(C)$ is a \emph{syntactic type diagram} if it is a syntactic $|V(\sigma)|$-diagram. \begin{example}\label{ex7} Let $\mathcal{L}$ and $\mathcal{P}$ be as in Example \ref{ex6}, and let $W=\{t,u,v,w\}$ be a set of size $4$. Set $\sigma'=\{p_1(c_u,c_v), p_2(c_u,c_w), p_3(c_v,c_w)\}\subseteq S_2(C_W,\mathcal{P})$. Then $V(\sigma')=\{c_u,c_v,c_w\}$ and we have that $Ch_{\sigma}(c_uc_v)=\{p_1(c_u,c_v)\}$, $Ch_{\sigma}(c_uc_w)=\{p_2(c_u,c_w)\}$, and $Ch_{\sigma}(c_vc_w)=\{ p_3(c_v,c_w)\}$. This shows $\sigma'$ is a syntactic $3$-diagram. \end{example} Observe that if $\sigma$ is a syntactic $m$-diagram, then by definition, $|V(\sigma)|=m$ and $|\sigma|={m\choose r}$. Given a tuple of constants $\bar{c}=(c_1,\ldots, c_k)$, a first-order language $\mathcal{L}_0$ containing $\{c_1,\ldots, c_k\}$, and an $\mathcal{L}_0$-structure $M$, let $\bar{c}^M$ denote the tuple $(c_1^M,\ldots, c_k^M)\in dom(M)^k$. \begin{definition}\label{tpd} Suppose $C$ is a set of constants and $\sigma \subseteq S_r(C)$. \begin{enumerate} \item If $M$ is an $\mathcal{L}\cup V(\sigma)$-structure, write $M\models \sigma^M$ if $M\models p(\bar{c}^M)$ for all $p(\bar{c})\in \sigma$. Call $\sigma$ \emph{satisfiable} if there exists an $\mathcal{L}\cup V(\sigma)$-structure $M$ such that $M\models \sigma^M$. \item If $M$ is an $\mathcal{L}\cup C$-structure, the \emph{type-diagram} of $M$ is the set $$ Diag^{tp}(M, C)=\{p(\bar{c})\in S_r(C): M\models p(\bar{c}^M)\}. $$ \end{enumerate} \end{definition} Suppose that $M$ is an $\mathcal{L}$-structure with $dom(M)=W$. The \emph{canonical type-diagram of $M$} is $$ Diag^{tp}(M)=\{p(c_{\bar{a}})\in S_r(C_W): M\models p(\bar{a})\}. $$ In other words, $Diag^{tp}(M)=Diag^{tp}(M,C_W)$ where $M$ is considered with its natural $\mathcal{L}\cup C_W$-structure. Observe that $Diatg^{tp}(M)$ is always a syntactic $|dom(M)|$-diagram. The difference between $Diag^{tp}(M)$ and $Diag(M)$ is that elements of $Diag^{tp}(M)$ are types (with constants plugged in for the variables) while the elements of $Diag(M)$ are formulas (with constants plugged in for the variables). Clearly $Diag(M)$ and $Diag^{tp}(M)$ contain the same information. \begin{example} Let $\mathcal{L}$ and $\mathcal{P}$ be as in Example \ref{ex7}. Let $W=\{u,v,w\}$ and let $M$ be the $\mathcal{L}$-structure with domain $W$ satisfying $M\models p_1(u,v)\cup p_2(u,w)\cup p_3(v,w)$. Then $Diag^{tp}(M)$ is the set $\{p_1(c_u,c_v), p_2(c_u,c_w), p_3(c_v,c_w)\}$, while $Diag(M)$ is the set of all $\mathcal{L}\cup C_W$-sentences implied by $p_1(c_u,c_v)\cup p_2(c_v,c_w)\cup p_3(c_u,c_w)$. \end{example} We now make a few observations which will be used in the remainder of the chapter. \begin{observation}\label{ob1} Suppose $M$ is an $\mathcal{L}$-structure with domain $W$ of size $n$. Then the following hold. \begin{enumerate} \item Suppose $m\leq n$, $\sigma\subseteq S_r(C_W)$ is a syntactic $m$-diagram, and $N$ is an $\mathcal{L}\cup V(\sigma)$-structure of size $m$. Then $N\models \sigma^N$ if and only if $\sigma=Diag^{tp}(N, V(\sigma))$. \item Suppose $N$ is an $\mathcal{L}\cup C_W$-structure of size $n$ and $N\models Diag^{tp}(M)$. Then $M\cong_{\mathcal{L}}N$. \item If $\sigma \subseteq S_r(C_W)$ and $Diag^{tp}(M)\subseteq \sigma$, then $\sigma$ is complete. \end{enumerate} \end{observation} \begin{proof} (1): Suppose first $\sigma=Diag^{tp}(N,V(\sigma))$. Then by Definition \ref{tpd}, $N\models \sigma^{N}$. Converesly, suppose $N\models \sigma^N$. By Definition \ref{tpd}, this implies $\sigma\subseteq Diag^{tp}(N,V(\sigma))$. To show the reverse inclusion, suppose $p(c_{\bar{a}})\in Diag^{tp}(N,V(\sigma))$. By Definition \ref{tpd}, $N\models p(\bar{c}^N)$. Let $A=\cup \bar{c}^N\in {dom(N)\choose r}$ (since $p\in S_r(\mathcal{L})$ is proper and $N\models p(\bar{c}^N)$, the coordinates of $\bar{c}^N$ must all be distinct). Since $\sigma$ is a syntactic $m$-diagram, $|Ch_{\sigma}(A)|=1$, so there is $p'(\bar{x})\in S_r(\mathcal{L})$ and $\mu\in Perm(r)$ such that $p'(\mu(\bar{c}))\in \sigma$. Since $N\models \sigma^N$, this implies $N\models p'(\mu(\bar{c}^N))$. Clearly $N\models p(\bar{c}^N)$ and $N\models p'(\mu(\bar{c}^N))$ implies $p(\bar{c}^N)= p'(\mu(\bar{c}^N))$. So we have $p(\bar{c})=p'(\mu(\bar{c}))\in \sigma$ as desired. (2): Clearly the map $f:W\rightarrow dom(N)$ sending $a\mapsto c_a^N$ is an $\mathcal{L}$-homomorphism of $M$ into $N$. Since by assumption, $M$ and $N$ both have size $n$, it must be a bijection, and thus an $\mathcal{L}$-isomorphism. (3): For each $A\in {C_W\choose r}$, $Diag^M(A)\in Ch_{\sigma}(A)$ implies $Ch_{\sigma}(A)\neq \emptyset$. \end{proof} \begin{definition} Suppose $\mathcal{A}$ is a collection of finite $\mathcal{L}$-structures and $C$ is a set of constant symbols. \begin{enumerate} \item We say $\sigma\subseteq S_r(C)$ is $\mathcal{A}$-satisfiable if $M\models \sigma^M$ and there is an $\mathcal{L}\cup V(\sigma)$-structure $M$ such that $M\upharpoonright_{\mathcal{L}}\in \mathcal{A}$. \item Define $Diag^{tp}(\mathcal{A},C)= \{\sigma \subseteq S_r(C): \sigma $ is a syntactic type diagram which is $\mathcal{A}$-satisfiable$\}$. \item Given $\sigma\subseteq S_r(C)$, set $Span(\sigma)=\{\sigma'\subseteq \sigma: \sigma'\text{ is a syntactic type diagram}\}$. \end{enumerate} \end{definition} \begin{example}\label{ex8} Let $\mathcal{L}$ and $\mathcal{P}$ be as in Example \ref{ex7}. Let $C=\{c_1, c_2, c_3\}$ be a set of three constant symbols. Then $\sigma\subseteq S_2(C, \mathcal{P})$ is a syntactic $3$-diagram if and only if $\sigma=\{p_i(c_1,c_2), p_j(c_1,c_3), p_k(c_2, c_3)\}$ for some $i,j,k\in [3]$. Clearly such a $\sigma$ is $\mathcal{P}$-satisfiable if and only if $|i-j|\leq k\leq i+j$, that is, if and only if the numbers $i,j,k$ do not violate the triangle inequality. Thus $Diag^{tp}(\mathcal{P}, C)$ consists of sets of the form $\sigma=\{p_i(c_1,c_2), p_j(c_1,c_3), p_k(c_2, c_3)\}$ where $i,j,k\in [3]$ satisfy $|i-j|\leq k\leq i+j$. Suppose now that $\sigma=\{p_1(c_1,c_2), p_2(c_1,c_2), p_3(c_1,c_2), p_1(c_2,c_3), p_1(c_1,c_3)\}$. Then $Span(\sigma)$ consists of the following syntactic $3$-diagrams. \begin{enumerate} \item $\{p_1(c_1,c_2), p_1(c_2,c_3), p_1(c_1,c_3)\}$. \item $\{p_2(c_1,c_2), p_1(c_2,c_3), p_1(c_1,c_3)\}$. \item $\{p_3(c_1,c_2), p_1(c_2,c_3), p_1(c_1,c_3)\}$. \end{enumerate} Observe that (1) and (2) are $\mathcal{P}$-satisfiable, while (3) is not. \end{example} \noindent For the rest of this subsection, $\mathcal{H}$ is a fixed collection of finite $\mathcal{L}$-structures. \begin{lemma}\label{lem0} Suppose $X\subseteq W$ are finite sets, $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure with domain $X$, and $\chi\in Ch(M)$. Set $\sigma :=\{\chi(A): A\in {X\choose r}\}\subseteq S_r(C_W, \mathcal{H})$. Then \begin{enumerate} \item $\sigma$ is a syntactic $|X|$-diagram. \item If $F\unlhd_{\chi}M$ then $\sigma=Diag^{tp}(F)$. \end{enumerate} \end{lemma} \begin{proof} Clearly $V(\sigma)=C_X$. Let $m=|C_X|$. Note ${C_X\choose r}=\{C_A: A\in {X\choose r}\}$. By definition of $\sigma$, for each $A\in {X\choose r}$, $\{\chi(A)\}=Ch_{\sigma}(C_A)$. Thus $|Ch_{\sigma}(C_A)|=1$ for all $A\in {X\choose r}$ and $\sigma$ is a syntactic $m$-diagram. This shows 1 holds. For 2, suppose $F\unlhd_{\chi}M$. This means $dom(F)=X$ and for all $A\in {X\choose r}$, $Diag^F(A)=\chi(A)$. Clearly this implies $F\models \sigma^F$, where $F$ is considered with its natural $C_{X}$-structure. Part 1 of Observation \ref{ob1} then implies $\sigma=Diag^{tp}(F)$. \end{proof} \begin{definition} Given an integer $\ell$ and a set of constants $C$, set $$ Err_{\ell}(C)=\{\sigma\subseteq S_r(C): \text{$\sigma$ is an unsatisfiable syntactic $\ell$-diagram}\}. $$ We call the elements of $Err_{\ell}(C)$ \emph{syntactic $C$-errors of size $\ell$}. \end{definition} \begin{example}\label{ex9} Let $\mathcal{L}=\{R_1,R_2,R_3,E\}$ and $\mathcal{P}$ be as in Example \ref{errorex}. Let $C=\{c_1,c_2,c_3,c_4\}$ be a set of constants. Recall from Example \ref{ex9}, that $q_1(c_1,c_2,c_3)\cup q_2(c_1,c_2,c_4)$ is unsatisfiable. Therefore an example of a syntactic $C$-error of size $4$ is the set $\{q_1(c_2,c_3,c_4), q_2(c_1,c_2,c_3), q_1(c_1,c_3,c_4), q_1(c_1,c_2,c_4) \}$. \end{example} \begin{lemma}\label{lem1} Suppose $W$ is finite a set, $r+1\leq \ell<2r$, and $M$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure which is an error of size $\ell$ and with domain $X\subseteq W$. Then there is a choice function $\chi\in Ch(M)$ such that $\{\chi(A): A\in {X\choose r}\}$ is a syntactic $C_W$-error of size $\ell$. \end{lemma} \begin{proof} Since $M$ is an error of size $\ell$ then there are $\bar{a}_1$, $\bar{a}_2 \in X^{\underline{r}}$ such that $\cup \bar{a}_1\bigcup \cup \bar{a}_2=X$ and $p_1(\bar{x}), p_2(\bar{x})\in S_r(\mathcal{H})$ such that $M\models R_{p_1}(\bar{a}_1) \wedge R_{p_2}(\bar{a}_2)$ but $p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2})$ is unsatisfiable. Define a function $\chi:{X\choose r}\rightarrow S_r(C_W,\mathcal{H})$ as follows. Set $\chi(\cup \bar{a}_1)=p(c_{\bar{a}_1})$ and $\chi(\cup \bar{a}_2)=p(c_{\bar{a}_2})$. For all other $A\in {X\choose r}$ choose any $\chi(A)\in Ch_M(A)$ (this is possible because $M$ is a complete). By construction, $\chi\in Ch(M)$. By part 1 of Lemma \ref{lem0}, $\sigma:=\{\chi(A): A\in {X\choose r}\}$ is a syntactic $\ell$-diagram. Because $\sigma$ contains $p_1(c_{\bar{a}_1})\cup p_2(c_{\bar{a}_2})$, it is unsatisfiable. By definition, $\sigma$ is a syntactic $C_W$-error of size $\ell$. \end{proof} \subsection{Proof of Theorem \ref{version1}}\label{section3thm1} In this section we state Theorem \ref{VERSION2} and use it to prove Theorem \ref{version1}. \begin{theorem}\label{VERSION2} Let $0<\epsilon<1$. For all $k\geq r$, there is a positive constant $c=c(k,r,\mathcal{L},\epsilon)$ and $m=m(k,r)>1$ such that for all sufficiently large $n$ the following holds. Suppose $\mathcal{F}$ is a collection of finite $\mathcal{L}$-structures, each of size at most $k$, and $\mathcal{H}:=\operatorname{Forb}(\mathcal{F})\neq\emptyset$. For any set $W$ of size $n$, there is a set $\Sigma \subseteq \mathcal{P}(S_r(C_W,\mathcal{H}))$ such that the following hold. \begin{enumerate} \item For all $\mathcal{F}$-free $\mathcal{L}$-structures $M$ with domain $W$, there is $\sigma\in \Sigma$ such that $Diag^{tp}(M)\subseteq \sigma$. \item For all $\sigma\in \Sigma$ the following hold. For each $1\leq \ell\leq k$, $|Diag^{tp}(\mathcal{F}(\ell), C_W)\cap Span(\sigma)|\leq \epsilon{n\choose \ell}$, and for each $r+1\leq \ell \leq 2r$, $|Err_{\ell}(C_W)\cap Span(\sigma)|\leq \epsilon{n\choose \ell}$. \item $\log |\Sigma|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} \end{theorem} Given a collection $\mathcal{H}$ of finite $\mathcal{L}$-structures, we now define a way of building an $\mathcal{L}_{\mathcal{H}}$-template from a complete subset of $S_r(C_W,\mathcal{H})$. \begin{definition}\label{D_Cdef} Suppose $\mathcal{H}$ is a nonempty collection of $\mathcal{L}$-structures, $W$ is a set, and $\sigma\subseteq S_r(C_W,\mathcal{H})$ is such that $V(\sigma)=C_W$. Define an $\mathcal{L}_{\mathcal{H}}$-structure $D_{\sigma}$ as follows. Set $dom(D_\sigma)=W$ and for each $\bar{a} \in W^{r}$, define $D_{\sigma}\models R_p(\bar{a})$ if and only if $p(c_{\bar{a}})\in Ch_{\sigma}(\cup \bar{a})$. \end{definition} In the notation of Definition \ref{D_Cdef}, note that for all $A\in {W\choose r}$, $Ch_{D_{\sigma}}(A)=Ch_{\sigma}(A)$ (here $Ch_{D_{\sigma}}(A)$ is in the sense of Definition \ref{chdef} and $Ch_{\sigma}(A)$ is in the sense of Definition \ref{chd2}). We now prove two lemmas. \begin{lemma}\label{templateclaim} Suppose $\mathcal{F}$ is collection of finite $\mathcal{L}$-structures and $\mathcal{H}=\operatorname{Forb}(\mathcal{F})\neq \emptyset$. For any set $W$ and complete $\sigma\subseteq S_r(C_W,\mathcal{H})$, $D_{\sigma}$ is an $\mathcal{L}_{\mathcal{H}}$-template. \end{lemma} \begin{proof} First, observe that $D_{\sigma}$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure since for each $A\in {W\choose r}$, $Ch_{D_{\sigma}}(A)=Ch_{\sigma}(C_A)$, and $Ch_{\sigma}(C_A)\neq \emptyset$ because $\sigma$ is complete by assumption (in the sense of Definition \ref{chd2}). Suppose now $\bar{a}\in W^r\setminus W^{\underline{r}}$. Then because $S_r(\mathcal{H})$ contains only proper types, there is no $p(\bar{x})\in S_r(\mathcal{H})$ such that $p(c_{\bar{a}})\in S_r(C_W,\mathcal{H})$. Thus $D_{\sigma}\models \neg R_p(\bar{a})$ for all $p(\bar{x})\in S_r(\mathcal{H})$, so $D_{\sigma}$ satisfies part (1) of Definition \ref{templatedef}. Suppose $p(\bar{x}), p'(\bar{x})\in S_r(\mathcal{H})$ and $\mu\in Perm(r)$ are such that $p(\bar{x})=p'(\mu(\bar{x}))$. Suppose $a\in W^{\underline{r}}$. Then by definition of $D_{\sigma}$, $D_{\sigma}\models R_p(\bar{a})$ if and only if $p(c_{\bar{a}})\in \sigma$. Since $p(c_{\bar{a}})=p'(c_{\mu(\bar{a})})$, $p(c_{\bar{a}})\in \sigma$ if and only if $p'(c_{\mu(\bar{a})})\in \sigma$. By definition of $D_{\sigma}$, $p'(c_{\mu(\bar{a})})\in \sigma$ if and only if $D_{\sigma}\models R_{p'}(\mu(\bar{a}))$. Thus we've shown $D_{\sigma}\models R_p(\bar{a})$ if and only if $D_{\sigma}\models R_{p'}(\mu(\bar{a}))$, so $D_{\sigma}$ satisfies part (2) of Definition \ref{templatedef}. This finishes the verification that $D_{\sigma}$ is an $\mathcal{L}_{\mathcal{H}}$-template. \end{proof} \begin{lemma}\label{boundinglem} Suppose $k\geq r$, $W$ is a finite set, $\mathcal{H}$ is a nonempty collection of finite $\mathcal{L}$-structures, and $\sigma \subseteq S_r(C_W,\mathcal{H})$ is complete. Suppose $\mathcal{F}$ is a collection of finite $\mathcal{L}$-structures, each of size at most $k$. Then for each $1\leq \ell \leq k$, there is an injection $$ \Phi: \textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})\rightarrow Diag^{tp}(\mathcal{F}(\ell), C_W)\cap Span (\sigma). $$ and for each $r+1\leq \ell\leq 2r$, there is an injection $$ \Theta: \textnormal{cop}(\mathcal{E}(\ell), D_{\sigma})\rightarrow Err_{\ell}(C_W)\cap Span (\sigma). $$ \end{lemma} \begin{proof} Without loss of generality, assume $\mathcal{F}$ is closed under isomorphism (we can do this because it does not change either of the sets $\textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})$ or $Diag^{tp}(\mathcal{F}(\ell), C_W)\cap Span (\sigma)$). Suppose $1\leq \ell \leq k$ and $G\in \textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})$. Then $G\subseteq_{\mathcal{L}_{\mathcal{H}}}D_{\sigma}$ and $G\cong_{\mathcal{L}_{\mathcal{H}}}B$, for some $B\in \tilde{\mathcal{F}}(\ell)$. It is straightforward to check that since $\mathcal{F}$ is closed under isomorphism, this implies $G\in \tilde{\mathcal{F}}(\ell)$. So without loss of generality we may assume that $B=G$. Then there is some $F\in \mathcal{F}(\ell)$ such that $F\unlhd_pG$. Choose any such $F$ and let $\chi\in Ch(G)$ be such that $F\unlhd_{\chi}G$. Define $\Phi(G)=\{\chi(A): A\in {dom(G)\choose r}\}$. By part 2 of Lemma \ref{lem0}, $\Phi(G)=Diag^{tp}(F)$. Thus by definition, $\Phi(G)\in Diag^{tp}(\mathcal{F}(\ell), C_W)$. By definition of $D_{\sigma}$ and because $\chi\in Ch(G)$, $G\subseteq_{\mathcal{L}_{\mathcal{H}}} D_{\sigma}$ implies $\Phi(G)\subseteq \sigma$, so $\Phi(G)\in Diag^{tp}(\mathcal{F}(\ell),C_W)\cap Span (\sigma)$, as desired. To see that $\Phi$ is injective, note that for all $G\in \textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})$, $V(\Phi(G))=dom(G)$. Therefore if $G_1\neq G_2\in \textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})$, $dom(G_1)\neq dom(G_2)$ implies $V(\Phi(G_1))\neq V(\Phi(G_2))$, so $\Phi(G_1)\neq \Phi(G_2)$. Suppose now $r+1\leq \ell\leq 2r$ and $G\in \textnormal{cop}(\mathcal{E}(\ell), D_{\sigma})$. Then $G$ is a complete $\mathcal{L}_{\mathcal{H}}$-structure which is an error of size $\ell$. Lemma \ref{lem1} implies there is $\chi\in Ch(G)$ such that $\{\chi(A): A\in {dom(G)\choose r}\}$ is a syntactic $C_W$-error of size $\ell$. Set $\Theta(G)=\{\chi(A): A\in {dom(G)\choose r}\}$. Then this shows $\Theta(G)\in Err_{\ell}(C_W)$. By definition of $D_{\sigma}$ and because $\chi\in Ch(G)$, $G\subseteq_{\mathcal{L}_{\mathcal{H}}}D_{\sigma}$ implies $\Theta(G)\subseteq \sigma$, so $\Theta(G)\in Err_{\ell}(C_W)\cap Span(\sigma)$, as desired. To see that $\Theta$ is injective, note that for all $G\in \textnormal{cop}(\mathcal{E}(\ell), D_{\sigma})$, $V(\Theta(G))=dom(G)$. Therefore if $G_1\neq G_2\in \textnormal{cop}(\mathcal{E}(\ell), D_{\sigma})$, $dom(G_1)\neq dom(G_2)$ implies $V(\Theta(G_1))\neq V(\Theta(G_2))$, so $\Theta(G_1)\neq \Theta(G_2)$.\end{proof} \noindent {\bf Proof of Theorem \ref{version1} from Theorem \ref{VERSION2}.} Let $0<\epsilon<1$ and let $k\geq r$ be an integer. Choose the constants $c=c(k,r,\mathcal{L},\epsilon)$ and $m=m(k,r)$ to be the ones given by Theorem \ref{VERSION2}. Suppose $\mathcal{F}$ is a collection of finite $\mathcal{L}$-structures, each of size at most $k$, and $\mathcal{B}:=\operatorname{Forb}(\mathcal{F})\neq \emptyset$. Suppose $n$ is sufficiently large and $W$ is a set of size $n$. Theorem \ref{VERSION2} applied to $\mathcal{B}$ implies there exists a set $\Sigma\subseteq \mathcal{P}(S_r(C_W,\mathcal{B}))$ such that the following hold. \begin{enumerate}[(i)] \item For all $\mathcal{F}$-free $\mathcal{L}$-structures $M$ with domain $W$, there is $\sigma\in \Sigma$ such that $Diag^{tp}(M)\subseteq \sigma$. \item For all $\sigma\in \Sigma$ the following hold. For each $1\leq \ell\leq k$, $|Diag^{tp}(\mathcal{F}(\ell), C_W)\cap Span(\sigma)|\leq \epsilon{n\choose \ell}$, and for each $r+1\leq \ell \leq 2r$, $|Err_{\ell}(C_W)\cap Span(\sigma)|\leq \epsilon{n\choose \ell}$. \item $\log |\Sigma|\leq cn^{r-\frac{1}{m}}\log n$. \end{enumerate} Set $\mathcal{D}=\{D_{\sigma}: \sigma \in \Sigma\}$, where for each $\sigma\in \Sigma$, $D_{\sigma}$ is the $\mathcal{L}_{\mathcal{B}}$-structure from Definition \ref{D_Cdef}. We claim this $\mathcal{D}$ satisfies conclusions of Theorem \ref{version1}. First note (i) and part 3 of Observation \ref{ob1} imply that every $\sigma\in \Sigma$ is complete in the sense of Definition \ref{chd2}. Therefore Lemma \ref{templateclaim} implies each $D_{\sigma}\in \mathcal{D}$ is an $\mathcal{L}_{\mathcal{B}}$-template. We now verify parts (1)-(3) of Theorem \ref{version1} hold for this $\mathcal{D}$. Clearly $|\mathcal{D}|\leq |\Sigma|$, so (iii) implies part (3) of Theorem \ref{version1} is satisfied. Suppose now $M$ is an $\mathcal{F}$-free $\mathcal{L}$-structure with $dom(M)=W$. By (i), there is $\sigma\in \Sigma$ such that $Diag^{tp}(M)\subseteq \sigma$. We claim that $M\unlhd_pD_{\sigma}$. Let $A \in {W\choose r}$ and suppose $p(\bar{x})\in S_r(\mathcal{H})$ is such that $M\models p(\bar{a})$ for some enumeration $\bar{a}$ of $A$. Then $Diag^M(A)=p(c_{\bar{a}})\in Diag^{tp}(M)\subseteq \sigma$ implies by definition of $D_{\sigma}$, $D_{\sigma}\models R_p(\bar{a})$, so $p(c_{\bar{a}})\in Ch_{D_{\sigma}}(A)$. This shows $M\leq_p D_{\sigma}$. Then $M\unlhd_pD_{\sigma}$ holds because by assumption $dom(M)=dom(D_{\sigma})=W$. Thus part (1) of Theorem \ref{version1} is satisfied. We now verify part (2) of Theorem \ref{version1}. Let $D_{\sigma}\in \mathcal{D}$. We need to show $\textnormal{prob}(\tilde{\mathcal{F}},D_{\sigma})\leq \epsilon$ and $\textnormal{prob}(\mathcal{E}, D_{\sigma})\leq \epsilon$. For each $1\leq \ell \leq k$, we have $$ |\textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})|\leq |Diag^{tp}(\mathcal{F}(\ell),C_W)\cap Span(\sigma)|\leq \epsilon{n\choose \ell}, $$ where the first inequality is because of Lemma \ref{boundinglem} and the second inequality is by (ii). This implies that for all $1\leq \ell \leq k$, $|\textnormal{cop}(\tilde{\mathcal{F}}(\ell), D_{\sigma})|\leq \epsilon {n\choose \ell}$, so $\textnormal{prob}(\tilde{\mathcal{F}}(\ell),D_{\sigma})\leq \epsilon$. Since every element in $\tilde{\mathcal{F}}$ has size at most $k$, we have $\textnormal{prob}(\tilde{\mathcal{F}}, D_{\sigma})\leq \epsilon$. Similarly, for each $r+1\leq \ell \leq 2r$, $$ |\textnormal{cop}(\mathcal{E}(\ell), D_{\sigma})|\leq|Err_{\ell}(C_W)\cap Span (\sigma)|\leq \epsilon {n\choose \ell}, $$ where the first inequality is by Lemma \ref{boundinglem} and the second inequality is by (ii). This implies for all $r+1\leq \ell \leq 2r$, $|\textnormal{cop}(\mathcal{E}(\ell), D_{\sigma})|\leq \epsilon{n\choose \ell}$, so $\textnormal{prob}(\mathcal{E}(\ell), D_{\sigma})\leq \epsilon$. Since every element in $\mathcal{E}$ has size at least $r+1$ and at most $2r$, we have $\textnormal{prob}(\mathcal{E}, D_{\sigma})\leq \epsilon$. This finishes the proof. \qed \section{Applying Hypergraph Containers to Prove Theorem \ref{VERSION2}}\label{VERSION2pf} In this section we prove Theorem \ref{VERSION2}. We will use the hypergraph containers theorem. We begin with a definition. \begin{definition} Suppose $K$ is a positive integer and $\mathcal{A}$ is a collection of finite $\mathcal{L}$-structures each of size at most $K$. Set $$ cl_K(\mathcal{A})=\{M:\text{$M$ is an $\mathcal{L}$-structure of size $K$ such that $prob(\mathcal{A}, M)>0\}$}. $$ \end{definition} Observe that in the above notation, an $\mathcal{L}$-structure of size at least $K$ is $\mathcal{A}$-free if and only if it is $cl_K(\mathcal{A})$-free. We now state a key lemma. \begin{lemma}\label{lemma**} Assume $n\geq k\geq r$ and $\mathcal{F}$ is a nonempty collection of $\mathcal{L}$-structures, each of size at most $k$. Suppose $\mathcal{H}:=\operatorname{Forb}(\mathcal{F})\neq \emptyset$ and $W$ is a set of size $n$. Fix $0<\epsilon<1/2$. Suppose $\sigma\subseteq S_r(C_W,\mathcal{H})$ is complete and satisfies $V(\sigma)=C_W$. If $$ |(Diag^{tp}(cl_k(\mathcal{F}),C_W)\cup Err_k(C_W))\cap Span(\sigma)|\leq \epsilon {n\choose k} $$ holds, then for all $1\leq \ell\leq k$, $|(Diag^{tp}(\mathcal{F}(\ell), C_W) \cup Err_{\ell}(C_W))\cap Span(\sigma)|\leq \epsilon {n\choose \ell}$. \end{lemma} \begin{proof} For $1\leq \ell<k$, set $\Gamma(\ell)=(Diag^{tp}(\mathcal{F}(\ell), C_W) \cup Err_{\ell}(C_W))\cap Span(\sigma)$ and let $$ \Gamma(k)=(Diag^{tp}(cl_k(\mathcal{F}),C_W)\cup Err_k(C_W))\cap Span(\sigma). $$ We want to show that $|\Gamma(k)|\leq \epsilon {n\choose k}$ implies that for all $\ell\in [k]$, $|\Gamma(\ell)|\leq \epsilon {n\choose \ell}$. If $\ell=k$, this is immediate. Fix $1\leq \ell<k$. We claim the following holds. \begin{align}\label{fact} \text{For all $S_0\in \Gamma(\ell)$, $|\{S_1\in \Gamma(k): S_0\subseteq S_1\}|\geq {n-\ell \choose r-\ell}$.} \end{align} Suppose $S_0\in \Gamma(\ell)$. Consider the following procedure for constructing a set $S_1\subseteq S_r(C_W,\mathcal{H})$. \begin{itemize} \item Choose $X\in {C_W\choose k}$ such that $V(S_0)\subseteq X$. There are ${n-\ell\choose k-\ell}$ choices. \item For each $A\in {X\choose r}\setminus {V(S_0)\choose r}$, choose some $p_A\in Ch_{\sigma}(A)$ (this is possible since $\sigma$ is complete). \item Set $S_1=S_0\cup \{p_A: A\in {X\choose r}\setminus {V(S_0)\choose r}\}$. \end{itemize} Suppose $S_1$ is constructed from $S_0$ in this way. We claim $S_1\in \Gamma(k)$. By construction and because $S_0$ is a syntactic $\ell$-diagram, $S_1$ is a syntactic $k$-diagram. Also by construction, $S_1\subseteq \sigma$, so by definition, $S_1 \in Span(\sigma)$. If $S_1$ is unsatisfiable, then by definition $S_1\in Err_k(C_W)\cap Span (\sigma)\subseteq \Gamma(k)$, so we are done. Suppose now $S_1$ is satisfiable and $M$ is an $\mathcal{L}\cup V(S_1)$-structure such that $M\models S_1^M$. Let $N=M[V(S_0)^M]$. Then considered as an $\mathcal{L}\cup V(S_0)$-structure, $N\models S_0^N$, so part 1 of Observation \ref{ob1} implies $Diag^{tp}(N)=S_0$. On the other hand, $S_0\in Diag^{tp}(\mathcal{F}(\ell), C_W)$ implies there is $F\in F(\ell)$ which can be made into an $\mathcal{L}\cup V(S_0)$-structure such that $F\models S_0^F$. Part 2 of Observation \ref{ob1} then implies $N\cong_{\mathcal{L}}F$. Since $\mathcal{F}$ is closed under isomorphism, $N\in \mathcal{F}$. Since $N\subseteq_{\mathcal{L}}M$ and $M$ has size $k$, this implies by definition that $M\in cl_k(\mathcal{F})$. Since $S_1=Diag^{tp}(M,V(S_1))$, we have $S_1\in Diag^{tp}(cl_k(\mathcal{F}),C_W)$ by definition. Thus we have shown that $S_1$ is in $\Gamma(k)$. Observe that every distinct choice of $X$ produces a distinct $S_1$, so this construction produces at least ${n-\ell\choose k-\ell}$ distinct $S_1\in \Gamma(k)$ such that $S_0\subseteq S_1$. We we have proved (\ref{fact}) holds for all $1\leq \ell <k$. Consider the following procedure for constructing element in $S_0\in \Gamma(\ell)$: \begin{itemize} \item Choose $S_1\in \Gamma(k)$. There are $|\Gamma(k)|$ choices. \item Choose $S_0\subseteq S_1$ such that $S_0\in \Gamma(\ell)$ (if one exists). There are at most ${V(S_1)\choose \ell}={k\choose \ell}$ choices. \end{itemize} By (\ref{fact}), this construction produces every element $S_0\in \Gamma(\ell)$ at least ${n-\ell\choose k-\ell}$ times. This shows $$ |\Gamma(\ell)|\leq |\Gamma(k)|{k\choose \ell}\Big/{n-\ell\choose k-\ell}\leq \epsilon {n\choose k}{k\choose \ell}\Big/{n-\ell \choose k-\ell}=\epsilon {n\choose \ell}, $$ where the second inequality is because $|\Gamma(k)|\leq \epsilon {n\choose k}$ by assumption. \end{proof} We now present a computational lemma which will be used in the proof of Theorem \ref{VERSION2}. \begin{lemma}\label{m} For all integers $2\leq x<y$, $m(y,x):=\max\Big\{ \frac{{\ell\choose x}-1}{\ell-x}: x< \ell \leq y \Big\}> 1$. \end{lemma} \begin{proof} We show that for all $2\leq x<y$, ${y\choose x}>y-x+1$. Since by definition, $m(y,x)\geq \frac{{y\choose x}-1}{y-x}$, this will imply $m(y,x)> 1$, as desired. Fix $x\geq 2$. We induct on $t$ where $y=x+t$. Suppose first $y=x+1$. Then ${y\choose x}=\frac{(x+1)!}{x!}=x+1$. By assumption on $x$, $x+1\geq 3>2=y-x+1$. Assume now that $y>x+1$ and suppose by induction the claim holds for $y-1$. Then ${y\choose x}=\frac{(y-1)!y}{x!(y-x-1)!(y-x)}={y-1\choose x}\frac{y}{y-x}$. By our induction hypothesis, $$ {y-1\choose x}\frac{y}{y-x}\geq ((y-1-x)+1)\Big(\frac{y}{y-x}\Big)=(y-x)\frac{y}{y-x}=y>y-x+1, $$ where the last inequality is because $x\geq 2$. Thus ${y\choose x}>y-x+1$, as desired. \end{proof} \vspace{2mm} \noindent {\bf Defining the Hypergraph.} We now give a procedure for defining a special hypergraph given a finite set and a collection of $\mathcal{L}$-structures satisfying certain properties. Assume we are given the following. \begin{enumerate}[$\bullet$] \item A nonempty collection, $\mathcal{F}$, of finite $\mathcal{L}$-structures, each of size at most $k$, where $k\geq r$ is an integer. \item A set $W$ of size $n$, where $n\geq k$ is an integer. \end{enumerate} Let $\mathcal{H}$ be the class of all finite $cl_k(\mathcal{F})$-free $\mathcal{L}$-structures. Define the hypergraph $H=H(\mathcal{F}, W)$ as follows. \begin{align*} V(H)&=S_r(C_W,\mathcal{H})\text{ and }\\ E(H)&=Diag^{tp}(cl_k(\mathcal{F}), C_W)\cup Err_k(C_W). \end{align*} We now make a few observations about $H$. First, note that the edges of $H$ are syntactic $k$-diagrams, so $H$ is a ${k\choose r}$-uniform hypergraph. By definition $|V(H)|={n\choose r}|S_r(\mathcal{H})|$. If $X$ and $X'$ are both in ${C_W\choose k}$, then since relabeling constants does not change the satisfiability properties of a collection of $\mathcal{L}\cup C_W$-sentences, we must have $|Diag^{tp}(Cl_k(\mathcal{F}), X)\cup Err_k(X)|=|Diag^{tp}(Cl_k(\mathcal{F}), X')\cup Err_k(X')|$. Therefore, the following is well defined. \begin{definition}\label{alpha} Let $\alpha=\alpha(\mathcal{F})$ be such that for all $X\in {C_W\choose k}$, $|Diag^{tp}(Cl_k(\mathcal{F}), X)\cup Err_k(X)|=\alpha$. \end{definition} We claim that $|E(H)|=\alpha {n\choose k}$. Indeed, any $\sigma\in E(H)$ can be constructed as follows. \begin{itemize} \item Choose $X\in {C_W\choose k}$. There are ${n\choose k}$ choices. \item Choose an element $\sigma \in Diag^{tp}(Cl_k(\mathcal{F}), C_W)\cup Err_k(C_W)$ such that $V(\sigma)=X$, i.e. choose an element $\sigma \in Diag^{tp}(Cl_k(\mathcal{F}), X)\cup Err_k(X)$. There are $\alpha$ choices. \end{itemize} Each of these choices lead to distinct subsets $\sigma\in E(H)$, so this shows $|E(H)|=\alpha {n\choose k}$. Note that because $\mathcal{F}\neq \emptyset$, $\alpha\geq 1$. On the other hand, there are at most $|S_r(\mathcal{H})|^{k\choose r}$ syntactic $k$-diagrams $\sigma$ with $V(\sigma)=X$, so $\alpha\leq |S_r(\mathcal{H})|^{k\choose r}\leq |S_r(\mathcal{L})|^{k\choose r}$. We now make a key observation about this hypergraph. \begin{proposition}\label{part1prop} For any $\mathcal{F}$-free $\mathcal{L}$-structure $M$ with domain $W$, $Diag^{tp}(M)$ is an independent subset of $V(H)$. \end{proposition} \begin{proof} Suppose towards a contradiction that $Diag^{tp}(M)$ contains an edge $\sigma\in E(H)$. Then $\sigma$ is a syntactic $k$-diagram which is either in $Err_k(C_W)$ or $Diag^{tp}(cl_k(\mathcal{F}), C_W)$. Clearly $\sigma\notin Err_k(C_W)$, since $M\models \sigma^M$ implies $\sigma$ is satisfiable. Thus $\sigma\in Diag^{tp}(cl_k(\mathcal{F}), C_W)$. So there is an $\mathcal{L}\cup V(\sigma)$-structure $B$ such that $B\upharpoonright_{\mathcal{L}}\in cl_k(\mathcal{F})$ and $Diag^{tp}(B, V(\sigma))=\sigma$. Let $A=\{a: c_a\in V(\sigma)\}\subseteq W$ and let $N=M[A]$. Suppose $p(c_{\bar{a}})\in \sigma$. Since $\sigma\subseteq Diag^{tp}(M)$, $M\models p(\bar{a})$. Since $N\subseteq_{\mathcal{L}}M$ and $\cup\bar{a} \subseteq A=dom(N)$, we have $N\models p(\bar{a})$. This shows that with its canonical $\mathcal{L}\cup V(\sigma)$-structure, $N\models \sigma^N$. Since $\sigma$ is a syntactic $k$-diagram and $N$ has size $k$, part 1 of Observation \ref{ob1} implies $\sigma=Diag^{tp}(N)$. Now $Diag^{tp}(N)=\sigma=Diag^{tp}(B, V(\sigma))$ implies by part 2 of Observation \ref{ob1}, that $N\cong_{\mathcal{L}}B$. But now $N$ is an $\mathcal{L}$- substructure of $M$ isomorphic to an element of $cl_k(\mathcal{F})$, contradicting our assumption that $M$ is $\mathcal{F}$-free (recall $|dom(M)|=n\geq k$ implies $M$ is $cl_k(\mathcal{F})$-free if and only if $M$ if $\mathcal{F}$-free). \end{proof} \noindent Observe that by definition of $H$, if $S\subseteq V(H)$, then \begin{align}\label{ob2} E(H[S])= \Big(Diag^{tp}(cl_k(\mathcal{F}), C_W)\cup Err_k(C_W)\Big)\cap Span(S). \end{align} We are now ready to prove Theorem \ref{VERSION2}. At this point the reader may wish to review Theorem \ref{containers} and its corresponding notation in Subsection \ref{containerssec}, as Theorem \ref{VERSION2} is the key tool used in this proof. \vspace{3mm} \noindent {\bf Proof of Theorem \ref{VERSION2}.} Clearly it suffices to show the theorem holds for all $0<\epsilon<1/2$. We claim that further, it suffices to show the theorem holds for any $k\geq 2r$. Indeed, suppose $k<2r$ and Theorem \ref{VERSION2} holds for all $k'\geq 2r$. Suppose $\mathcal{F}$ is a nonempty collection of $\mathcal{L}$-structures, each of size at most $k$ and $\mathcal{H}:=\operatorname{Forb}(\mathcal{F})\neq \emptyset$. Then $\mathcal{F}$ is also a collection of $\mathcal{L}$-structures, each of size at most $k'=2r$. Apply Theorem \ref{VERSION2} to $k'=2r$ to obtain constants $c=c(2r,r,\mathcal{L}, \epsilon)$ and $m=m(2r, r)$. Since $k<2r$, it is clear the conclusions of Theorem \ref{VERSION2} for $\mathcal{H}$ and $2r$ imply the conclusions of Theorem \ref{VERSION2} for $\mathcal{H}$ and $k$. Thus we may take $c(k,r,\epsilon)=c(2r,r,\epsilon)$ and $m(k,r)=m(2r,r)$. We now prove the theorem holds for all $0<\epsilon<1/2$ and $k\geq 2r$. Fix $0<\epsilon<1/2$ and $k\geq 2r$. Apply Theorem \ref{containers} to $s={k\choose r}$ to obtain the constant $c_0=c_0({k\choose r})$ and set $$ m=m(k,r)=\max\Big\{ \frac{{\ell\choose r}-1}{\ell-r}: r< \ell \leq k \Big\}. $$ By Lemma \ref{m}, since $2\leq r<k$, $m> 1$. Set $\epsilon'=\epsilon/|S_r(\mathcal{L})|^{k\choose r}$ and choose $0<\gamma<1$ sufficiently small so that \begin{align}\label{gamma} 2^{{{k\choose r}\choose 2}+1}|S_r(\mathcal{L})|r!(k-r)^{k-r} \gamma &\leq \frac{\epsilon'}{12{k\choose 2}!}. \end{align} Now set $c=c(k,r,\mathcal{L},\epsilon)=(c_0\log(\frac{1}{\epsilon'})|S_r(\mathcal{L})|)/\gamma m$. Observe that $c$ actually depends on $\mathcal{L}$, not just $r_{\mathcal{L}}$. Let $M\geq k$ be such that $n\geq M$ implies $(n-r)^{k-r}\geq n^{k-r}/2$, and $n^{-\frac{1}{m}}\gamma^{-1}<1/2$. We show Theorem \ref{VERSION2} holds for this $c$ and $m$ for all $n\geq M$. Fix $n\geq M$. Suppose $\mathcal{F}$ a nonempty collection of finite $\mathcal{L}$-structures, each of size at most $k$, such that $\mathcal{H}:=\operatorname{Forb}(\mathcal{F})\neq \emptyset$. Let $W$ be a set of size $n$ and let $H=H(\mathcal{F}, W)$ be the ${k\choose r}$-uniform hypergraph described above. Set $\tau=n^{\frac{-1}{m}}\gamma^{-1}$. By our assumptions we have that $0<\epsilon', \tau<\frac{1}{2}$. We show that $\delta(H,\tau)\leq \frac{\epsilon'}{12{k\choose r}!}$ so that we may apply Theorem \ref{containers} to $H$. Let $\alpha=\alpha(\mathcal{F})$ be as in Definition \ref{alpha} so that $E(H)=\alpha{n\choose k}$ and let $N=|V(H)|$. Given $2\leq j\leq {k\choose r}$, set \begin{align}\label{fdef} f(j)=\min\{ \ell: {\ell \choose r}\geq j\}. \end{align} Observe that for each $2\leq j\leq {k\choose r}$, $r<f(j)\leq k$. Indeed, $r<f(j)$ holds since ${f(j)\choose r}\geq j\geq 2$, and $f(j)\leq k$ holds since $k\in \{\ell: {\ell \choose r}\geq j\}$. Thus by definition of $m$, for each $2\leq j\leq {k\choose r}$, \begin{align}\label{minequality} m\geq \frac{{f(j)\choose r}-1}{f(j)-r} \geq \frac{j-1}{f(j)-r}, \end{align} where the inequality is because by (\ref{fdef}), ${f(j)\choose r}\geq j$. We now show that for all $\sigma \subseteq V(H)$ with $2\leq |\sigma|\leq {k\choose r}$, $d(\sigma)\leq \alpha n^{k-f(|\sigma|)}$. Fix $\sigma\subseteq V(H)$ so that $2\leq |\sigma|\leq {k\choose r}$. Observe that if $|V(\sigma)|>k$, then $\{e\in E(H): \sigma\subseteq e\}=\emptyset$ since every $e\in E(H)$ is a syntactic $k$-diagram, so must satisfy $|V(e)|=k$. So in this case $d(\sigma)=0\leq \alpha n^{k-f(|\sigma|)}$. Similarly, if there is $A\in {V(\sigma)\choose r}$ such that $|Ch_{\sigma}(A)|\geq 2$, then $\{e\in E(H): \sigma\subseteq e\}=\emptyset$, since every $e\in E(H)$ is a syntactic $k$-diagram, so must satisfy $|Ch_e(A)|=1$. So in this case, $d(\sigma)=0\leq \alpha n^{k-f(|\sigma|)}$. Suppose now $|V(\sigma)|\leq k$ and $|Ch_{\sigma}(A)|\leq 1$ for all $A\in {V(\sigma)\choose r}$. This implies $|\sigma|\leq {|V(\sigma)|\choose r}$, so by (\ref{fdef}), $f(|\sigma|)\leq |V(\sigma)|$. Every edge in $H$ containing $\sigma$ can be constructed as follows. \begin{itemize} \item Choose a set $X\in {C_W\choose k}$ such that $V(\sigma)\subseteq X$ (this makes sense since $|V(\sigma)|\leq k$). There are ${n-|V(\sigma)|\choose k-|V(\sigma)|}$ ways to do this. \item Choose an element of $Diag^{tp}(cl_k(\mathcal{F}), X) \cup Err_k(X)$ containing $\sigma$. By definition of $\alpha$, there are at most $\alpha$ choices for this. \end{itemize} Therefore, $d(\sigma)=|\{ e\in E(H): \sigma\subseteq e\}| \leq \alpha {n-|V(\sigma)|\choose k-|V(\sigma)|}\leq \alpha n^{k-|V(\sigma)|}\leq \alpha n^{k-f(|\sigma|)}$, where the last inequality is because $f(|\sigma|)\leq |V(\sigma)|$. Thus we have shown that for any $2\leq j\leq {k\choose r}$ and $\sigma\subseteq V(H)$ with $|\sigma|=j$, $d(\sigma)\leq \alpha n^{k-f(j)}$. Thus given $2\leq j\leq {k\choose r}$ and a vertex $v\in V(H)$, $$ d^{(j)}(v)=\max \{ d(\sigma): v\in \sigma, |\sigma|=j\}\leq \alpha n^{k-f(j)}. $$ Note the average degree of $H$ is $$ d={k\choose r}|E(H)|/|V(H)| = \frac{{k\choose r}\alpha {n\choose k}}{|S_r(\mathcal{H})|{n\choose r}}=\frac{\alpha}{|S_r(\mathcal{H})|}{n-r\choose k-r}\geq \frac{\alpha}{|S_r(\mathcal{H})|}\Big(\frac{n-r}{k-r}\Big)^{k-r}. $$ Combining this with our assumption $n$, we obtain the following inequality. \begin{align}\label{dbound} d\geq \frac{\alpha}{|S_r(\mathcal{H})|}\Big(\frac{n-r}{k-r}\Big)^{k-r}=\frac{\alpha}{|S_r(\mathcal{H})|(k-r)^{k-r}} (n-r)^{k-r}\geq \frac{\alpha}{2|S_r(\mathcal{H})|(k-r)^{k-r}} n^{k-r}. \end{align} Combining all of these computations we have the following. $$ \delta_j=\frac{\sum_{v\in V(H)} d^{(j)}(v)}{Nd\tau^{j-1}}\leq \frac{Nn^{k-f(j)}}{Nd\tau^{j-1}}= \frac{n^{k-f(j)+(j-1)\frac{1}{m}}\gamma^{j-1}}{d}. $$ Using our lower bound for $d$ from (\ref{dbound}), this implies \begin{align*} \delta_j \leq 2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma^{j-1}\alpha^{-1} n^{k-f(j)+\frac{j-1}{m}-k+r}=2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma^{j-1}\alpha^{-1} n^{r-f(j)+\frac{j-1}{m}}. \end{align*} By (\ref{minequality}), $\frac{j-1}{m}\leq f(j)-r$, so this implies $\delta_j$ is at most $$ 2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma^{j-1}\alpha^{-1} n^{r-f(j)+f(j)-r}= 2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma^{j-1}\alpha^{-1}\leq 2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma, $$ where the last inequality is because $j\geq 2$ and $\gamma<1$ implies $\gamma^{j-1}<\gamma$ and $\mathcal{F}\neq \emptyset$ implies $\alpha^{-1}\leq 1$. Therefore \begin{align}\label{delta1} \delta(H,\tau)&=2^{{{k\choose r}\choose 2}-1}\sum_{j=2}^{{k\choose r}}2^{-{j-1\choose 2}} \delta_j\leq 2^{{{k\choose r}\choose 2}-1}2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma \sum_{j=2}^{{k\choose r}}2^{-{j-1\choose 2}}. \end{align} If $t={{k\choose r}\choose 2}$, then $\sum_{j=2}^{{k\choose r}}2^{-{j-1\choose 2}}\leq\sum_{j=0}^t2^{-t}$. Using the formula for summing finite geometric series, $\sum_{j=0}^t2^{-t}=\frac{1-2^{-t-1}}{1-2^{-1}}=2(1-2^{-t-1})<2$. Plugging this into (\ref{delta1}) yields \begin{align*} \delta(H,\tau)&\leq 2^{{{k\choose r}\choose 2}-1}2|S_r(\mathcal{H})|(k-r)^{k-r} \gamma 2=2^{{{k\choose r}\choose 2}+1}|S_r(\mathcal{H})|(k-r)^{k-r} \gamma\leq 2^{{{k\choose r}\choose 2}+1}|S_r(\mathcal{L})|(k-r)^{k-r} \gamma. \end{align*} By (\ref{gamma}), the right hand side above is at most $\frac{\epsilon'}{12{k\choose r}!}$, so we have shown that $\delta(H,\tau) \leq \frac{\epsilon'}{12{k\choose r}!}$. Thus Theorem \ref{containers} implies there is $\Sigma_0\subseteq \mathcal{P}(V(H))$ with the following properties. \begin{enumerate}[(i)] \item For every independent set $I\subseteq V(H)$, there is $\sigma \in \Sigma_0$ such that $I \subseteq \sigma$. \item For every $\sigma \in \Sigma_0$, $e(H[\sigma])\leq \epsilon' e(H)$. \item $\log |\Sigma_0| \leq c_0\log(1/\epsilon') N\tau \log (1/\tau)$. \end{enumerate} Define $\Sigma=\{\sigma \in \Sigma_0: \exists \text{ an $\mathcal{F}$-free $\mathcal{L}$-structure $M$ with domain $W$ such that $Diag^{tp}(M)\subseteq \sigma \}$}$. Observe that every $\sigma \in \Sigma$ is complete by part 3 of Observation \ref{ob1}. We show $\Sigma$ satisfies the conclusions (1)-(3) of Theorem \ref{VERSION2}. Suppose $M$ is an $\mathcal{F}$-free $\mathcal{L}$-structure with domain $W$. Proposition \ref{part1prop} implies $Diag^{tp}(M)$ is an independent subset of $V(H)$, so by (i) and definition of $\Sigma$, there is $\sigma\in \Sigma$ such that $Diag^{tp}(M)\subseteq \sigma$. Thus part (1) of Theorem \ref{VERSION2} holds. We now show part (2) holds. Fix $\sigma\in \Sigma$. By (\ref{ob2}), $(Diag^{tp}(cl_k(\mathcal{F}), C_W)\cup Err_k(C_W))\cap Span(\sigma)=E(H[\sigma])$. So (ii) implies $$ e(H[\sigma])=|(Diag^{tp}(cl_k(\mathcal{F}), C_W)\cup Err_k(C_W))\cap Span(\sigma)|\leq \epsilon' e(H). $$ By definition of $\epsilon'$ and because $\alpha \leq |S_r(\mathcal{L})|^{k\choose r}$, $$ \epsilon'e(H)=\epsilon'\alpha{n\choose k}=\frac{\epsilon}{|S_r(\mathcal{L})|^{k\choose r}}\alpha {n\choose k}\leq \epsilon {n\choose k}. $$ Thus $|(Diag^{tp}(cl_k(\mathcal{F}), C_W)\cup Err_k(C_W))\cap Span(\sigma)|\leq \epsilon{n\choose k}$. By Lemma \ref{lemma**}, for all $1\leq \ell \leq k$, $$ |(Diag^{tp}(\mathcal{F}(\ell),C_W)\cup Err_{\ell}(C_W))\cap Span (\sigma)|\leq \epsilon {n\choose \ell}. $$ Since $k\geq 2r$, this immediately implies part (2) of Theorem \ref{VERSION2} holds. By (iii), definition of $c$, and because $\Sigma\subseteq \Sigma_0$ we have that \begin{align*} |\Sigma|\leq |\Sigma_0|\leq c_0\log(1/\epsilon') N\tau \log (1/\tau)&=c_0\log(1/\epsilon') |S_r(\mathcal{H})|{n\choose r}\tau \log (1/\tau) \\ &\leq c_0\log(1/\epsilon') |S_r(\mathcal{L})|{n\choose r}\tau \log (1/\tau)=c\gamma m{n\choose r}\tau \log\Big(\frac{1}{\tau}\Big). \end{align*} This shows $|\Sigma|\leq c\gamma mn^r\tau \log(\frac{1}{\tau})$. By definition of $\tau$, $$ c\gamma mn^r\tau \log\Big(\frac{1}{\tau}\Big)=c mn^{r-\frac{1}{m}}\Big(\frac{\log n}{m}+\log\gamma\Big)= c n^{r-\frac{1}{m}}\Big(\log n+m\log\gamma\Big)\leq cn^{r-\frac{1}{m}}\log n, $$ where the last inequality is because $\gamma \leq 1\leq m$ implies $m\log \gamma \leq 0$. This shows part (3) of Theorem \ref{VERSION2} holds, so we are done. \qed \section{Metric Spaces}\label{ms} Given integers $r\geq 3$ and $n\geq 1$, let $M_r(n)$ be the set of metric spaces with distances all in $[r]$ and underlying set $[n]$. In this section we will reprove certain structure and enumeration theorems about $M_r(n)$ from \cite{MT} using the machinery of this paper, along with combinatorial ingredients from \cite{MT}. We include this example because it demonstrates interesting behavior with regards to the existence of a stability theorem. In particular, we will prove that when $r$ is even, the hereditary property associated to $\bigcup_{n\in \mathbb{N}}M_r(n)$ has a stability theorem in the sense of Definition \ref{stabdef}, but when $r$ is odd, this is not the case. We would like to point out that in order to be consistent with \cite{MT}, $r$ will always denote the largest distance appearing in our metric spaces. This $r$ has nothing to do with our use of the letter $r$ as shorthand for $r_{\mathcal{L}}$ throughout the rest of the paper. In this section all languages will binary (i.e. $r_{\mathcal{L}}=2$), so we do not think any confusion will arise. \subsection{Results from \cite{MT}.} In this subsection we state results from \cite{MT}. {\bf For the rest of this section, $r\geq 3$ is a fixed integer}. We require some notation and definitions in order to state the relevant theorems from \cite{MT}. An \emph{$[r]$-graph} (respectively an $2^{[r]}$-graph) is a pair $(V,c)$ such that $V$ is a set of vertices and $c:{V\choose 2}\rightarrow [r]$ (respectively $c:{V\choose 2}\rightarrow 2^{[r]}$) is a function. \begin{definition} Given an $[r]$-graph $G=(V,d)$ and a $2^{[r]}$-graph $R=(V',c)$, we say $G$ is a \emph{sub-$[r]$-graph} of $R$, denoted $G\subseteq_{[r]} R$, if $V'\subseteq V$ and for each $xy\in {V'\choose r}$, $d(xy)\in c(xy)$. We say $G$ is a \emph{full sub-$[r]$-graph} of $R$, denote $G\subseteqq_{[r]}R$, if moreover, $V'=V$. \end{definition} A $2^{[r]}$-graph $(V,c)$ is \emph{complete} if for all $xy\in {V\choose 2}$, $|c(xy)|\geq 1$. Two $[r]$-graphs (respectively $2^{[r]}$-graphs) $(V,d)$ and $(V',d')$ are \emph{isomorphic} if there is a bijection $f:V\rightarrow V'$ such that for all $xy\in {V\choose 2}$, $d(xy)=d'(f(x)f(y))$. A \emph{violating triple} is a tuple $(i,j,k)\in \mathbb{N}^3$ such that $|i-j|\leq k\leq i+j$ is false. An $[r]$-graph $G=(V,d)$ is a \emph{metric $[r]$-graph} if for all $\{x,y,z\}\in {V\choose 3}$, $(d(xy), d(yz),d(xz))$ is not a violating triple. A $2^{[r]}$-graph $G=(V,c)$ is a \emph{metric $2^{[r]}$-graph} for all $\{x,y,z\}\in {V\choose 3}$, no $(i,j,k)\in c(xy)\times c(yz)\times c(xz)$ is a violating triple. Observe that if $R$ is a complete $2^{[r]}$-graph, then $R$ is a metric $2^{[r]}$-graph if and only if all its full sub-$[r]$-graphs are metric $[r]$-graphs. Given integers $i<j$, set $[i,j]=\{i,i+1, \ldots, j\}$. Set $$ m(r)=\Bigg\lceil \frac{r+1}{2}\Bigg\rceil. $$ If $r$ is odd, let $L_r=[ \frac{r-1}{2}, r-1]$ and $U_r=[\frac{r+1}{2}, r]$. Observe that if $r$ is odd, then $m(r)=|U_r|=|L_r|$ and if $r$ is even, then $m(r)=|[\frac{r}{2},r]|$. We now define a special subfamily of $M_r(n)$. \begin{definition}\label{Crdefeven} Suppose $n\geq 3$ is an integer. \begin{enumerate} \item If $r\geq 4$ is even, define $C_r(n)$ be the set of $[r]$-graphs $G=([n],d)$ such that $d(e)\in [\frac{r}{2}, r ]$ for all $e\in {[n]\choose 2}$. \item If $r\geq 3$ is odd, define $C_r(n)$ to be the set of $[r]$-graphs $G=([n],d)$ such that there is a partition $V_1\cup \ldots\cup V_t$ of $[n]$ so that for every $xy\in {[n]\choose 2}$, \[ d(xy) \in \begin{cases} L_r& \text{if } xy\in {V_i\choose 2}\textnormal{ for some }i\in [t] \\ U_r & \text{if } x\in V_i, y\in V_j \textnormal{ for some }i\neq j\in [t]. \end{cases} \] \end{enumerate} \end{definition} It is straightforward to check that in both even and odd cases, $C_r(n)\subseteq M_r(n)$. \begin{definition}\label{newdeltaclose} Let $G=(V,c)$ and $G=(V,c')$ be finite $C$-graphs where $C=[r]$ or $C=2^{[r]}$. Set $$ \Delta(G,G')=\{xy\in {V\choose 2}: c(xy)\neq c'(xy)\}. $$ Given $\delta>0$, we say $G$ and $G'$ are \emph{$\delta$-close}\footnote{In \cite{MT}, $\delta$-closeness is instead defines as $|\Delta(G,G')|\leq \delta n^2$. For large $n$, this is basically the same as Definition \ref{newdeltaclose} up to a factor of $2$, and therefore doesn't change the statements of the results from \cite{MT}. We have chosen to use Definition \ref{newdeltaclose} for convenience.} if $|\Delta(G,G')|\leq \delta{|V|\choose 2}$. \end{definition} Set $C_r^{\delta}(n)=\{G\in M_r(n): \textnormal{there is some }G' \in C_r(n)\textnormal{ such that }|\Delta(G,G')|\leq \delta{|V|\choose 2} \}$. We can now state the approximate structure theorem from \cite{MT}. Informally, it states that for all $\delta>0$, most members of $M_r(n)$ are in $C_r^{\delta}(n)$ when $n$ is sufficiently large depending on $\delta$. \begin{theorem}[{\bf Mubayi-Terry \cite{MT}}]\label{DELTACLOSETHM} Let $r\geq 3$ be an integer. Then for all $\delta>0$, there exists an $M$ and $\beta >0$ such that $n>M$ implies $$ \frac{|M_r(n)\setminus C_r^{\delta}(n)|}{|M_r(n)|}\leq \frac{|M_r(n)\setminus C_r^{\delta}(n)|}{m(r)^{n\choose 2}}\leq 2^{-\beta {n\choose 2}}. $$ \end{theorem} \noindent We now state the approximate enumeration theorem from \cite{MT}. \begin{corollary}[{\bf Mubayi-Terry \cite{MT}}]\label{oddcor} Let $r\geq3$ be an integer. Then $|M_r(n)|=m(r)^{{n\choose 2}+o(n^2)}$. \end{corollary} In fact, \cite{MT} contains much more precise structural and enumerative results for $M_r(n)$ in the case when \emph{$r$ is even}. Finding similar refinements of Theorem \ref{DELTACLOSETHM} and Corollary \ref{oddcor} in the case when $r$ is odd is still open. This suggests there is something ``nicer'' about the even case than the odd case. We will show in this section that one candidate for what makes the even case ``nice'' is that when $r$ is even, the hereditary property corresponding to $M_r(n)$ has a stability theorem in the sense of Definition \ref{stabdef}, while in the odd case it does not. \subsection{Translation to hereditary $\mathcal{L}$-property.} In this subsection we translate some of the combinatorial notions used in \cite{MT} to the setup used in this paper. Recall $r\geq 3$ is a fixed integer. Let $\mathcal{L}_r=\{R_1(x,y),\ldots, R_r(x,y)\}$ consist of $r$ binary relation symbols. We consider elements $(V,d)$ of $M_r(n)$ as $\mathcal{L}_r$-structures by interpreting $R_i(x,y)$ if and only if $d(xy)=i$, for each $(x,y)\in V^2$. Let $\mathcal{M}_r$ denote the class of $\mathcal{L}_r$-structures obtained by closing $M_r=\bigcup_{n\in \mathbb{N}}M_r(n)$ under isomorphism. Clearly $\mathcal{M}_r$ is a hereditary $\mathcal{L}_r$-property, and $(\mathcal{M}_r)_n=M_r(n)$. For the rest of the section, $\mathcal{P}=\mathcal{M}_r$, and $\mathcal{L}=\mathcal{L}_r$. Observe that since $r_{\mathcal{L}}=2$, $\mathcal{L}_{\mathcal{P}}=\{R_p(x,y): p(x,y)\in S_2(\mathcal{P})\}$. For each $i\in [r]$, set $$ q_i(x,y):=\{x\neq y\}\cup \{R_i(x,y), R_i(y,x)\}\cup \{\neg R_j(x,y), \neg R_j(y,x): j\in [r]\setminus \{i\}\}, $$ and let $p_i(x,y)$ be the unique quantifier-free $2$-type containing $q_i(x,y)$. In other words, $p_i(x,y)$ is the type saying the distance between $x$ and $y$ is $i$. Clearly $S_2(\mathcal{P})=\{p_i(x,y): i\in [r]\}$, so $\mathcal{L}_{\mathcal{P}}=\{R_{p_i}(x,y): i\in [r]\}$. The following observation will be useful and is obvious from the definition of a choice function. \begin{observation}\label{ob7} If $G$ is an $\mathcal{L}_{\cal}$-template with domain $V$ and $\chi\in Ch(G)$, then for all $uv\in {V\choose 2}$, $Ch_G(uv)=\{p_j(c_u,c_v): G\models R_{p_j}(u,v)\vee R_{p_j}(v,u)\}$. \end{observation} \begin{definition} Suppose $G$ is an $\mathcal{L}_{\mathcal{P}}$-structure with underlying set $V$. The \emph{$2^{[r]}$-graph associated to $G$} is $\Psi(G):=(V, c)$, where for each $xy\in {V\choose 2}$, $c(xy)=\{i: G\models R_{p_i}(x,y)\vee R_{p_i}(y,x)\}$ (in other words, $i\in c(x,y)$ if and only if $p_i(c_x,c_y)\in Ch_G(xy)$). \end{definition} Given a $2^{[r]}$-graph $(V,c)$, define $\Psi^{-1}(V,c)$ to be the $\mathcal{L}_{\mathcal{P}}$-structure $G$ with domain $V$ such that for each $(x,y)\in V^2$ and $i\in [r]$, $G\models R_{p_i}(x,y)$ if and only if $x\neq y$ and $i\in c(xy)$. We leave the following observations to the reader. \begin{observation}\label{obms} Suppose $(V,c)$ is a complete $2^{[r]}$-graph and $G$ is an $\mathcal{L}_{\mathcal{P}}$-template. Then \begin{enumerate}[(a)] \item $\Psi^{-1}(V,c)$ is an $\mathcal{L}_{\mathcal{P}}$-template and $\Psi(\Psi^{-1}(V,c))=(V,c)$. \item $\Psi(G)$ is a complete $2^{[r]}$-graph and $\Psi^{-1}(\Psi(G))=G$. \end{enumerate} \end{observation} Suppose $G\in M_r(n)$ is the $[r]$-graph $(V,d)$ considered as an $\mathcal{L}$-structure. We will often abuse notation by conflating $G$ and $(V,d)$. For instance if $R$ is a $2^{[r]}$-graph, we will write $G\subseteqq_{[r]}R$ to mean $(V,d)\subseteqq_{[r]}R$. Similarly if $R$ is an $\mathcal{L}_{\mathcal{P}}$-template, we will write $(V,d)\unlhd_pR$ to mean $G\unlhd_pR$. \begin{proposition}\label{translation} Suppose $G$ is an $\mathcal{L}_{\mathcal{P}}$-template with domain $V$ and $\Psi(G)=(V,c)$. Then $G'\unlhd_pG$ if and only if $G'\subseteqq_{[r]}\Psi(G)$. \end{proposition} \begin{proof} Suppose first $G'\unlhd_pG$. Let $\chi\in Ch(G)$ such that $G'\unlhd_{\chi}G$. Let $(V,d)$ be the $[r]$-graph such that for each $uv\in {V\choose 2}$, $d(uv)=i$ is the unique element of $[r]$ such that $\chi(uv)= p_i(c_u,c_v)$. Observe that $G'=(V,d)$, considered as an $\mathcal{L}$-structure. We show $(V,d)\subseteqq_{[r]}\Psi(G)$. Fix $uv\in {V\choose 2}$. We want to show $d(uv)\in c(uv)$. Since $\chi\in Ch(G)$, by definition of $(V,d)$, and by Observation \ref{ob7}, if $i=d(uv)$, then $$ p_i(c_u,c_v)=\chi(uv)\in Ch_G(uv)=\{p_j(c_u,c_v): G\models R_{p_j}(u,v)\vee R_{p_j}(v,u)\}. $$ This implies $d(uv)=i\in \{j: G\models R_{p_j}(u,v)\vee R_{p_j}(v,u)\}=c(uv)$, where the last equality is by definition of $\Psi(G)=(V,c)$. Thus $d(uv)\in c(uv)$ for all $uv\in {V\choose 2}$, so $(V,d)\subseteqq_{[r]}(V,c)$, as desired. Suppose on the other hand that $G'=(V,d)\subseteqq_{[r]}\Psi(G)$. We want to show that considered as an $\mathcal{L}$-structure, $G'\unlhd_pG$. Define a function $\chi:{V\choose 2}\rightarrow S_2(C_V,\mathcal{P})$ as follows. For each $uv\in {V\choose 2}$, if $d(uv)=i$, set $\chi(uv)=p_i(c_u,c_v)$. Since $(V,d)\subseteqq_{[r]}(V,c)$ and by definition of $\Psi(G)=(V,c)$, we have $$ d(uv)\in c(u,v)=\{j: G\models R_{p_j}(x,y)\vee R_{p_j}(y,x)\}. $$ Thus $\chi(uv)=p_i(c_u,c_v)\in \{p_j(c_u,c_v): G\models R_{p_j}(u,v)\vee R_{p_j}(v,u)\}=Ch_G(uv)$, where the last equality is by Observation \ref{ob7}. This verifies that $\chi\in Ch(G)$. By definition of $\chi$, for each $uv\in {V\choose 2}$, $\chi(uv)=Diag^{G'}(uv)$, so $G'\unlhd_{\chi}G$. \end{proof} Given $n$, let $\tilde{M}_r(n)$ be the set of complete metric $2^{[r]}$-graphs on $[n]$. \begin{corollary}\label{cor1} If $([n],c)\in \tilde{M}_r(n)$, then $\Psi^{-1}(V,c)\in \mathcal{R}([n],\mathcal{P})$. If $G\in \mathcal{R}([n],\mathcal{P})$, then $\Psi(G)\in \tilde{M}_r(n)$. \end{corollary} \begin{proof} Suppose $([n],c)\in \tilde{M}_r(n)$. By Observation \ref{obms}(a), $\Psi^{-1}([n],c)$ is an $\mathcal{L}_{\mathcal{P}}$-template with domain $[n]$. Since $([n],c)$ is a complete metric $2^{[r]}$-graph, all full sub-$[r]$-graphs of $([n],c)$ are metric spaces. By Proposition \ref{translation}, this implies all full subpatterns of $\Psi^{-1}([n],c)$ are metric spaces, which implies $\Psi^{-1}([n],c)$ is $\mathcal{P}$-random by definition. Suppose now $G\in \mathcal{R}([n],\mathcal{P})$. Let $\Psi(G)=([n],c)$. By Observation \ref{obms}(b), $([n],c)$ is a complete $2^{[r]}$-graph. To show $([n],c)$ is a metric $2^{[r]}$-graph, let $([n],d)\subseteqq_{[r]}([n],c)$. By Proposition \ref{translation}, $([n],d)\unlhd_pG$, so since $G$ is $\mathcal{P}$-random, $([n],d)$ is a metric space. Thus all full sub-$[r]$-graphs of $([n],c)$ are metric $[r]$-graphs. This implies $([n],c)$ is a metric $2^{[r]}$-graph. Thus we have shown $([n],c)\in \tilde{M}_r(n)$. \end{proof} We now recall a definition from \cite{MT}. If $G=(V,c)$ is a $2^{[r]}$-graph, then $W(R)=\prod_{xy\in {V\choose 2}}|c(xy)|$. \begin{proposition}\label{0} Suppose $G$ is a finite $\mathcal{L}_{\mathcal{P}}$-template. Then $sub(G)=W(\Psi(G))$. \end{proposition} \begin{proof} Let $\Psi(G)=(V,c)$. By Proposition \ref{translation}, the full subpatterns of $G$ are exactly the full sub-$[r]$-graphs of $(V,c)$, considered as $\mathcal{L}$-structures. Clearly the number of full sub-$[r]$-graphs of $(V,c)$ is $\prod_{uv\in {V\choose 2}}|c(uv)|$. This shows $sub(G)=\prod_{uv\in {V\choose 2}}|c(uv)|=W(\Psi(G))$. \end{proof} If $G\in \tilde{M}_r(n)$, say that $G$ is \emph{product-extremal} if $$ W(G)=\max\{W(G'): G'\in \tilde{M}_r(n)\}. $$ \begin{proposition}\label{1} Suppose $G$ is an $\mathcal{L}_{\mathcal{P}}$-template with domain $[n]$. Then $G\in \mathcal{R}_{ex}([n],\mathcal{P})$ if and only if $\Psi(G)$ is a product-extremal element of $\tilde{M}_r(n)$. \end{proposition} \begin{proof} Suppose first $G\in \mathcal{R}_{ex}([n],\mathcal{P})$. By Corollary \ref{cor1}, and definition of $\tilde{M}_r(n)$, $\Psi(G)\in \tilde{M}_r(n)$. Suppose towards a contradiction that $\Psi(G)$ is not product-extremal. Then there is $H\in \tilde{M}_r(n)$ such that $W(H)>W(\Psi(G))$. Corollary \ref{cor1} implies $\Psi^{-1}(H)\in \mathcal{R}([n],\mathcal{P})$ and Proposition \ref{0} and Observation \ref{obms}(a) implies $sub(\Psi^{-1}(H))=W(\Psi(\Psi^{-1}(H)))=W(H)>W(\Psi(G))=sub(G)$, contradicting that $G\in \mathcal{R}_{ex}([n],\mathcal{P})$. Conversely, suppose $\Psi(G)$ is a product-extremal element of $\tilde{M}_r(n)$. By Corollary \ref{cor1} and Observation \ref{obms}(b), $\Psi^{-1}(\Psi(G))=G\in \mathcal{R}([n],\mathcal{P})$. Suppose towards a contradiction $G\notin \mathcal{R}_{ex}([n],\mathcal{P})$. Then there is $G'\in \mathcal{R}([n],\mathcal{P})$ such that $sub(G')>sub(G)$. Corollary \ref{cor1} implies $\Psi(G')\in \tilde{M}_r(n)$ and Proposition \ref{0} implies $W(\Psi(G'))=sub(G')>sub(G)=W(\Psi(G))$, contradicting that $\Psi(G)$ is product-extremal. \end{proof} \subsection{Characterizing extremal structures and computing $\pi(\mathcal{P})$.} In this subsection, we characterize product-extremal elements of $\tilde{M}_r(n)$. These results, Lemmas \ref{extlemeven} and \ref{extlemodd}, are new results. We will then use the correspondence between product-extremal elements of $\tilde{M}_r(n)$ and elements of $\mathcal{R}_{ex}([n],\mathcal{P})$ from the preceding section to compute $\pi(\mathcal{P})$. We begin by defining a special subfamily of $\tilde{M}_r(n)$ corresponding to the special subfamily $C_r(n)$ of $M_r(n)$. \begin{definition}\label{tildeCrdef} Let $\tilde{C}_r(n)$ be the the set of complete $2^{[r]}$-graphs $([n],c)$ such that \begin{enumerate}[(i)] \item if $r$ is even, then for all $xy\in {[n]\choose 2}$, $c(xy)=[\frac{r}{2}, r]$. \item if $r$ is odd, then there is a partition $V_1\cup \ldots\cup V_t$ of $[n]$ such that for all $xy\in {[n]\choose 2}$, \[ c(xy) = \begin{cases} L_r & \text{if } xy\in {V_i\choose 2}\textnormal{ for some }i\in [t] \\ U_r & \text{if } xy\in E(V_i,V_j)\textnormal{ for some }i\neq j\in [t]. \end{cases} \] \end{enumerate} \end{definition} Note that for all $n$, $\tilde{C}_r(n)\subseteq \tilde{M}_r(n)$ and when $r$ is even, $\tilde{C}_r(n)$ consists of a single $2^{[r]}$-graph. We will need a few results about multigraphs. A \emph{multigraph} is a pair $(V,w)$ where $V$ is a set of vertices and $w:{V\choose 2}\rightarrow \mathbb{N}$ is a function. Given integers $s\geq 2$ and $q\geq 0$, an \emph{$(s,q)$-graph} is a multigraph $(V,w)$ such that for all $X\in {V\choose s}$, $\sum_{xy\in {X\choose 2}}w(xy)\leq q$. Given a multigraph $G=(V,w)$, set $P(G)=\prod_{xy\in {V\choose 2}}w(xy)$. \begin{definition} Suppose $a\geq 1$ is an integer. \begin{enumerate}[(a)] \item $U_{1,a}(n)$ is the set of multigraphs $([n],w)$ such that $w(xy)= a$ for all $xy\in {[n]\choose 2}$. \item $U_{2,a}(n)$ is the set of multigraphs $([n],w)$ such that there is a set $\{e_1,\ldots, e_{\lfloor n/2 \rfloor}\}$ of pairwise disjoint elements in ${[n]\choose 2}$ such that $w(e_i)=a+1$ for each $e_i$ and for all $xy\in {[n]\choose 2}\setminus \{e_1,\ldots, e_{\lfloor n/2 \rfloor}\}$, $w(xy)=a$. \end{enumerate} \end{definition} Observe that if $G\in U_{1,a}(n)$, then $P(G)=a^{n\choose 2}$ and if $G\in U_{2,a}(n)$, then $P(G)=a^{n\choose 2}(\frac{a+1}{a})^{\lfloor n/2\rfloor }$. \begin{theorem}[{\bf Mubayi-Terry, Theorem 5.5.6 of \cite{mythesis}}]\label{mgthm} Suppose $a\geq 1$, $n\geq 3$ and $G=([n],w)$ is a multigraph. \begin{enumerate}[(a)] \item If $G$ is a $(3,3a)$-graph, then $P(G)\leq a^{n\choose 2}$ with equality holding if and only if $G\in U_{1,a}(n)$. \item If $G$ is a $(3,3a+1)$-graph, then $P(G)\leq a^{n\choose 2}(\frac{a+1}{a})^{\lfloor \frac{n}{2}\rfloor}$ with equality holding if and only if $G\in U_{2,a}(n)$. \end{enumerate} \end{theorem} In the following lemma we use a result of Balogh and Wagner from \cite{BW}. In particular, we use Lemma 3.2 of \cite{BW}, which is a corollary of combinatorial results in \cite{MT}. \begin{lemma}\label{extlemeven} Suppose $r\geq 4$ is even, $n\geq 3$, and $G\in \tilde{M}_r(n)$ is product-extremal. Then $W(G)=m(r)^{n\choose 2}$ and $G$ is the unique $2^{[r]}$-graph in $\tilde{C}_r(n)$. \end{lemma} \begin{proof} Suppose $G=(V,c)\in \tilde{M}_r(n)$ is product-extremal. Let $G_0$ be the unique element in $\tilde{C}_r(n)$. Then by definition, $W(G_0)=|[\frac{r}{2},r]|^{n\choose 2}=m(r)^{n\choose 2}$. Since $G$ is product-extremal, this implies $W(G)\geq W(G_0)=m(r)^{n\choose 2}$. Now let $H=([n],w)$ be the multigraph defined by $w(xy)=|c(xy)|$ for all $xy\in {[n]\choose 2}$. Observe $P(H)=W(G)$. We claim that $H$ is a $(3, 3m(r))$-graph. Suppose towards a contradiction there were three is $\{x,y,z\}\in {[n]\choose 3}$ such that $w(xy)+w(yz)+w(xz)>3m(r)$. Then $c(xy), c(yz), c(xz)$ are nonempty subsets of $[r]$ satisfying $$ w(xy)+w(yz)+w(xz)=|c(xy)|+|c(yz)|+|c(xz)|>3m(r). $$ By Lemma 3.2 of \cite{BW}, this implies there is $(i,j,k)\in c(xy)\times c(yz)\times c(xz)$ which is a violating triple, contradicting that $G\in \tilde{M}_r(n)$. Thus $H$ is a $(3,3m(r))$-graph. Theorem \ref{mgthm} implies that $W(G)=P(H)\leq m(r)^{n\choose r}$ and equality holds if and only if $w(xy)=m(r)$ for all $xy\in {[n]\choose 2}$. Since we already showed $W(G)\geq m(r)^{n\choose 2}$, equality must hold. Thus $P(H)=W(G)=m(r)^{n\choose 2}$ and $w(xy)=m(r)=|c(xy)|$ for all $xy\in {[n]\choose 2}$. Part (1) of Corollary 4.15 in \cite{MT} implies that the only metric $2^{[r]}$-graph $([n],c')$ satisfying $|c'(x,y)|=m(r)$ for all $xy\in {[n]\choose 2}$ is $G_0$. Thus $G=G_0$. \end{proof} If $r\geq 3$ is odd and $n\geq 3$, define $\tilde{E}_r(n)$ to be the set of $2^{[r]}$-graphs $([n],c)$ such that there is a set $\{e_1,\ldots, e_{\lfloor n/2\rfloor}\}$ of pairwise disjoint elements in ${[n]\choose 2}$ such that $c(e_i)=U_r\cup L_r$ and if $xy\in {[n]\choose 2}\setminus \{e_1,\ldots, e_{\lfloor n/2\rfloor}\}$ then $c(xy)=U_r$. Observe that $\tilde{E}_r(n)\subseteq \tilde{C}_r(n)$ and \begin{align}\label{Efact} \text{ for any $G\in \tilde{E}_r(n)$, }W(G)=m(r)^{n\choose r}\Big(\frac{m(r)+1}{m(r)}\Big)^{\lfloor n/2 \rfloor}. \end{align} \begin{lemma}\label{extlemodd} Suppose $r\geq 3$ is odd, $n\geq 3$, and $G$ is a product-extremal element of $\tilde{M}_r(n)$. Then $P(G)=m(r)^{n\choose r}(\frac{m(r)+1}{m(r)})^{\lfloor n/2 \rfloor}$ and $G\in \tilde{E}_r(n)$. \end{lemma} \begin{proof} Suppose $G=([n],c)\in \tilde{M}_r(n)$ is product-extremal. Let $H$ be the multigraph $([n],w)$ where $w(xy)=|c(xy)|$ for all $xy\in {[n]\choose 2}$. Observe $P(H)=W(G)$. We claim $H$ is a $(3,3m(r)+1)$-graph. Suppose by contradiction there is $\{x,y,z\} \in {[n]\choose 3}$ such that $w(xy)+w(yz)+w(xz)>3m(r)+1$. Then $c(xy), c(yz), c(xz)$ are nonempty subsets of $[r]$ satisfying $$ w(xy)+w(yz)+w(xz)=|c(xy)|+|c(yz)|+|c(xz)|>3m(r)+1. $$ By Lemma 3.2 of \cite{BW}, this implies there is $(i,j,k)\in c(xy)\times c(yz)\times c(xz)$ which is a violating triple, contradicting that $G\in \tilde{M}_r(n)$. Thus $H$ is a $(3,3m(r)+1)$-graph, so by Theorem \ref{mgthm}, $W(G)=P(H)\leq m(r)^{n\choose r}(\frac{m(r)+1}{m(r)})^{\lfloor n/2 \rfloor}$ and equality holds if and only if $H\in U_{1,m(r)}(n)$. Since $G$ is product-extremal, we must have $W(G)\geq W(G')$ for all $G'\in \tilde{E}_r(n)$. Combining these facts with (\ref{Efact}), we must have $W(G)=P(H)=m(r)^{n\choose r}(\frac{m(r)+1}{m(r)})^{\lfloor n/2 \rfloor}$ and $H\in U_{1,m(r)}(n)$. So there is a set $\{e_1,\ldots, e_{\lfloor n/2 \rfloor}\}$ of pairwise disjoint elements in ${[n]\choose 2}$ such that $w(e_i)=|c(e_i)|=m(r)+1$ for each $e_i$ and for all $xy\in {[n]\choose 2}\setminus \{e_1,\ldots, e_{\lfloor n/2 \rfloor}\}$, $w(xy)=|c(xy)|=m(r)$. These facts along with part (2) of Corollary 4.15 in \cite{MT}, imply that for each $e_i$, $c(e_i)=U_r\cup L_r$ and for all $xy\in {[n]\choose 2}\setminus \{e_1,\ldots, e_{\lfloor n/2 \rfloor}\}$, $c(xy)=U_r$. In other words, $G$ is in $\tilde{E}_r(n)$. \end{proof} \begin{corollary}\label{mspi} Let $n\geq 2$. If $r\geq 2$ is even, then $\textnormal{ex}(n,\mathcal{P})=m(r)^{n\choose 2}$. If $r\geq 3$ is odd, then $\textnormal{ex}(n,\mathcal{P})=m(r)^{n\choose r}(\frac{m(r)+1}{m(r)})^{\lfloor n/2 \rfloor}$. Consequently, $\pi(\mathcal{P})=m(r)$. \end{corollary} \begin{proof} That $\textnormal{ex}(n,\mathcal{P})=\max\{W(G): G\in \tilde{M}_r(n)\}$ follows from Propositions \ref{0} and \ref{1}. Thus if $r$ is even, Lemma \ref{extlemeven} implies $\textnormal{ex}(n,\mathcal{P})=m(r)^{n\choose 2}$. Similarly, if $r$ is odd, Lemma \ref{extlemodd} implies $\textnormal{ex}(n,\mathcal{P})=m(r)^{n\choose r}(\frac{m(r)+1}{m(r)})^{\lfloor n/2 \rfloor}$. That $\pi(\mathcal{P})=m(r)$ holds in both cases now follows from the definition of $\pi(\mathcal{P})=\lim_{n\rightarrow \infty}\textnormal{ex}(n,\mathcal{P})^{1/{n\choose 2}}$. \end{proof} \subsection{Proofs of Results} In this subsection, we prove Corollary \ref{oddcor} and Theorem \ref{DELTACLOSETHM}. We also prove that if $r$ is even then $\mathcal{P}$ has a stability theorem, and if $r$ is odd, then $\mathcal{P}$ does \emph{not} have a stability theorem. These results, Lemma \ref{msstabeven} and Corollary \ref{msstabodd}, are new. We begin by proving Corollary \ref{oddcor}. \vspace{3mm} \noindent {\bf Proof of Corollary \ref{oddcor}.} Theorem \ref{enumeration} implies $|\mathcal{P}_n|=\pi(\mathcal{P})^{(1+o(1)){n\choose 2}}$. By definition of $\mathcal{P}$, $M_r(n)=\mathcal{P}_n$, and Corollary \ref{mspi} implies $\pi(\mathcal{P})=m(r)$. Thus $|\mathcal{P}_n|=|M_r(n)|=m(r)^{(1+o(1)){n\choose 2}}$. \qed \vspace{3mm} \noindent We now state a stability style result, Theorem 4.13 from \cite{MT}. \begin{theorem}[{\bf Mubayi-Terry \cite{MT}}]\label{stabthm} Fix an integer $r\geq 3$. For all $\delta>0$ there is $0<\epsilon<1$ and $M$ such that for all $n>M$ the following holds. If $R\in \tilde{M}_r(n)$ and $W(R)>m(r)^{(1-\epsilon){n\choose 2}}$, then $R$ is $\delta$-close to an element in $\tilde{C}_r(n)$ (in the sense of Definition \ref{newdeltaclose}). \end{theorem} The following is straightforward from the definitions. \begin{observation}\label{ob8} If $G=(V,[n])$ and $G=(V,[n])$ are in $M_r(n)$, then $\textnormal{dist}(G,G')\leq \delta$ (in the sense of Definition \ref{deltaclosedef1}, considered as $\mathcal{L}$-structures) if and only if $G$ and $G'$ are $\delta$-close in the sense of Definition \ref{newdeltaclose} (i.e. if and only if $|\Delta(G,G')|\leq \delta {n\choose 2}$). \end{observation} We can now prove Theorem \ref{DELTACLOSETHM}. \vspace{3mm} \noindent {\bf Proof of Theorem \ref{DELTACLOSETHM}.} Fix $\delta>0$. Choose $\epsilon>0$ and $M_1$ from Theorem \ref{stabthm} such that if $n>M_1$ and $H\in \tilde{M}_r(n)$ satisfies $W(H)>m(r)^{(1-\epsilon){n\choose 2}}$ then $H$ is $\delta/2$-close in the sense of Definition \ref{newdeltaclose} to an element of $\tilde{C}_r(n)$. Now let $\beta>0$ and $M_2$ be as in Theorem \ref{b4stab} applied to $\delta/2$ and $\epsilon$. Let $N=\max\{M_1,M_2\}$. We show for all $n>N$, $$ \frac{|M_r(n)\setminus C^{\delta}_r(n)|}{|M_r(n)}|\leq 2^{-\beta n^2}. $$ By Theorem \ref{b4stab}, it suffices to show that for all $n\geq N$, $E^{\delta/2}(\epsilon, n, \mathcal{P})\subseteq C^{\delta}_r(n)$. Fix $n\geq N$ and suppose $H=(V,d)\in E^{\delta/2}(\epsilon, n, \mathcal{P})$. We want to show $H\in C^{\delta}_r(n)$. By definition of $E^{\delta/2}(\epsilon, n, \mathcal{P})$, there is some $H'=(V,d')\in E(\epsilon, n, \mathcal{P})$ such that $\textnormal{dist}(H,H')\leq \delta/2$. By Observation \ref{ob8}, this implies $|\Delta(H,H')|\leq \delta {n\choose 2}$. By definition of $E(\epsilon, n,\delta)$, there is $G'\in \mathcal{R}([n],\mathcal{P})$ such that $H'\unlhd_pG'$ and $sub(G')>ex(n,\mathcal{P})^{1-\epsilon}$. Recall that by Observation \ref{ob5}(a), for all $n$, $\textnormal{ex}(n,\mathcal{P})\geq \pi(\mathcal{P})^{{n\choose 2}}$, so this implies $sub(G')>\pi(\mathcal{P})^{(1-\epsilon){n\choose 2}}$. Corollary \ref{cor1} implies $\Psi(G'):=(V,c')\in \tilde{M}_r(n)$, and Proposition \ref{0} implies $W(\Psi(G'))=sub(G')>\pi(\mathcal{P})^{(1-\epsilon){n\choose 2}}$. Consequently, Theorem \ref{stabthm} implies $\Psi(G')$ is $\delta/2$-close in the sense of Definition \ref{newdeltaclose} to some $M=([n],c)\in \tilde{C}_r(n)$. Define $H''=([n],d'')$ as follows. \begin{enumerate}[$\bullet$] \item If $xy \notin \Delta(\Psi(G'), M) \cup \Delta (H,H')$, let $d''(xy)=d'(xy)=d(xy)$. \item If $xy\in \Delta(\Psi(G'), M)\cup \Delta (H,H')$, let $d''(xy)$ be any element of $c(xy)$. \end{enumerate} We claim $H''\subseteqq_{[r]}M$. Fix $xy\in {[n]\choose 2}$. We want to show $d''(xy)\in c(xy)$. This holds by definition of $H''$ when $xy\in \Delta(\Psi(G'), G'')\cup \Delta (H,H')$. If $xy\not\in \Delta(\Psi(G'), M)\cup \Delta (H,H')$, then $d''(xy)=d'(xy)$ and $c(xy)=c'(xy)$. Since, $H'\unlhd_pG'$, Proposition \ref{translation} implies $H'\subseteqq_{[r]}\Psi(G')$, thus $d'(xy)\in c'(xy)$. Since $d''(xy)=d'(xy)$ and $c(xy)=c'(xy)$, this implies $d''(xy)\in c(xy)$, as desired. Therefore, $H''\subseteqq_{[r]}M$ and $M\in \tilde{C}_r(n)$ implies $H''\in C_r(n)$. We claim $|\Delta(H,H'')|\leq \delta {n\choose 2}$. By definition of $H''$, $$ \Delta(H,H'')\subseteq \Delta(H, H')\cup \Delta(\Psi(G'), M), $$ so $|\Delta(H,H'')|\leq |\Delta(H,H')|+|\Delta(\Psi(G'),M)|\leq \delta {n\choose 2}$, where the inequality is by our assumptions. This shows $H''\in C_r(n)$ and $|\Delta(H,H'')|\leq \delta{n\choose 2}$, i.e. $H\in C_r^{\delta}(n)$. \qed \vspace{3mm} We leave the following lemma to the reader. \begin{lemma}\label{msuse} If $M$ and $N$ are complete $2^{[r]}$-graphs with the same vertex set $V$, then $\Delta(M,N)=\textnormal{diff}(\Psi^{-1}(M),\Psi^{-1}(N))$. \end{lemma} We now show that when $r$ is even, $\mathcal{P}$ has a stability theorem in the sense of Definition \ref{stabdef}, but when $r$ is odd, this is not the case. \begin{theorem}\label{msstabeven} If $r\geq 2$ is even, then $\mathcal{P}$ has a stability theorem. \end{theorem} \begin{proof} Fix $\delta>0$. By Theorem \ref{stabthm}, there is $\epsilon>0$ and $M$ such that for all $n>M$ if $H\in \tilde{M}_r(n)$ satisfies $P(H)>m(r)^{(1-\epsilon){n\choose 2}}$, then $H$ is $\delta$-close in the sense of Definition \ref{newdeltaclose} to the unique $2^{[r]}$-graph $H_0\in \tilde{C}_r(n)$. Suppose now that $n>M$ and $G\in \mathcal{R}([n],\mathcal{P})$ satisfies $sub(G)\geq \textnormal{ex}(n,\mathcal{P})^{1-\epsilon}$. We want to show there is $G'\in \mathcal{R}_{ex}([n],\mathcal{P})$ such that $\textnormal{dist}(G,G')\leq \delta$. Recall that by part (a) of Observation \ref{ob5}, for all $n$, $\pi(\mathcal{P})^{n\choose 2}\leq \textnormal{ex}(n,\mathcal{P})$. Thus our assumptions imply $sub(G)\geq \textnormal{ex}(n,\mathcal{P})^{1-\epsilon}\geq \pi(\mathcal{P})^{(1-\epsilon){n\choose 2}}$. Proposition \ref{mspi} implies $\pi(\mathcal{P})=m(r)$ and Corollary \ref{cor1} implies $\Psi(G)\in \tilde{M}_r(n)$, so by Proposition \ref{0}, $$ W(\Psi(G))=sub(G)\geq \pi(\mathcal{P})^{(1-\epsilon){n\choose 2}}=m(r)^{(1-\epsilon){n\choose 2}}. $$ Thus Theorem \ref{stabthm} implies $\Psi(G)$ is $\delta$-close in the sense of Definition \ref{newdeltaclose} to the unique $2^{[r]}$-graph $H_0\in \tilde{C}_r(n)$. By Lemma \ref{extlemeven}, $H_0$ is a product-extremal element of $\tilde{M}_r(n)$. By Corollary \ref{cor1}, $\Psi^{-1}(H_0)\in \mathcal{R}([n],\mathcal{P})$. Since by Observation \ref{ob5}(a), $\Psi(\Psi^{-1}(H_0))=H_0$, Proposition \ref{1} and the fact that $H_0$ is product-extremal imply $\Psi^{-1}(H_0)\in \mathcal{R}_{ex}([n],\mathcal{P})$. By Lemma \ref{msuse} and Observation \ref{ob5}(b), \begin{align}\label{labeldiff} \Delta(H_0,\Psi(G))=\textnormal{diff}(\Psi^{-1}(H_0),\Psi^{-1}(\Psi(G)))=\textnormal{diff}(\Psi^{-1}(H_0), G). \end{align} Since $H_0$ and $\Psi(G)$ are $\delta$-close in the sense of Definition \ref{newdeltaclose}, $|\Delta(H_0, \Psi(G))|\leq \delta {n\choose 2}$. Combining this with (\ref{labeldiff}), we have that $\textnormal{dist}(\Psi^{-1}(H_0),G)\leq \delta$. Since $\Psi^{-1}(H_0)\in \mathcal{R}_{ex}([n],\mathcal{P})$, we are done. \end{proof} \begin{corollary}\label{msstabodd} When $r\geq 3$ is odd, $\mathcal{P}$ does not have a stability theorem. \end{corollary} \begin{proof} Let $A=([n],c)$ be such that for all $xy\in {[n]\choose 2}$, $c(xy)=L_r$. Then by definition, $A\in \tilde{M}_r(n)$ and $W(A)=m(r)^{n\choose 2}$. By Corollary \ref{cor1}, $\Psi^{-1}(A)\in \mathcal{R}([n],\mathcal{P})$, and by Proposition \ref{0}, $$ sub(\Psi^{-1}(A))=W(A)=m(r)^{n\choose 2}. $$ Let $B\in \mathcal{R}_{ex}([n],\mathcal{P})$. By Proposition \ref{extlemodd}, $\Psi(B)\in \tilde{E}_r(n)$. By definition of $A$ and $\tilde{E}_r(n)$, $\Delta(A,\Psi(B))={[n]\choose 2}$. By Lemma \ref{msuse} and Observation \ref{ob5}(b), $$ \Delta(A,\Psi(B))=\textnormal{diff}(\Psi^{-1}(A),\Psi^{-1}(\Psi(B))=\textnormal{diff}(\Psi^{-1}(A),B). $$ Therefore, $\textnormal{dist}(\Psi^{-1}(A),B)=|\Delta(A,\Psi(B))|/{n\choose 2}=1$. So for all $\delta<1$, $\Psi^{-1}(A)$ is not $\delta$-close to any element of $\mathcal{R}_{ex}([n],\mathcal{P})$. However, for all $\epsilon>0$, $\pi(\mathcal{P})=m(r)$ implies that if $n$ is sufficiently large, $sub(\Psi^{-1}(A))=m(r)^{n\choose 2}\geq ex(n,\mathcal{P})^{1-\epsilon}$, so $\Psi^{-1}(A)\in E(\epsilon, n, \mathcal{P})$. This shows that $\mathcal{P}$ does not have a stability theorem in the sense of Definition \ref{stabdef}. \end{proof} \section{Concluding Remarks}\label{end} We end with some questions and conjectures. Returning to the metric spaces of the previous section, it was shown in \cite{MT} the following is true. \begin{theorem}[{\bf Mubayi-Terry \cite{MT}}] If $r\geq 2$ is even, then $M_r=\bigcup_{n\in \mathbb{N}}M_r(n)$ has a $0$-$1$ law in the language $\mathcal{L}_r$. \end{theorem} It was then conjectured in \cite{MT} that this theorem is false in the case when $r$ is odd (to our knowledge, this is still open). In the previous section, we showed that in the case when $r$ is even, $\mathcal{M}_r$ has a stability theorem in the sense of Definition \ref{stabdef}, while when $r$ is odd, this is false. These facts lead us to make the following conjecture. \begin{conjecture} Suppose $\mathcal{L}$ is a finite relational language with $r_{\mathcal{L}}\geq 2$, and $\mathcal{H}$ is a fast-growing hereditary $\mathcal{L}$-property such that $\bigcup_{n\in \mathbb{N}}\mathcal{H}_n$ has a $0$-$1$ law. Then $\mathcal{H}$ has a stability theorem. \end{conjecture} The idea behind this conjecture is that if $\mathcal{H}$ has nice asymptotic structure, it should reflect the structure of $\mathcal{R}_{ex}([n],\mathcal{H})$. Another phenomenon which can be observed from known examples is that the structures in $\mathcal{R}_{ex}(n,\mathcal{H})$ are not very complicated. The following questions are various ways of asking if this is always the case. \begin{question} Suppose $\mathcal{L}$ is a finite relational language with $r_{\mathcal{L}}\geq 2$, and $\mathcal{H}$ is a fast-growing hereditary $\mathcal{L}$-property. For each $n$, let $\mathcal{P}_n=\mathcal{R}_{ex}([n],\mathcal{H})$. Does $\bigcup_{n\in \mathbb{N}} \mathcal{P}_n$ always have a $0$-$1$ law? \end{question} We direct the reader to \cite{classification} for the definition of a formula having the $k$-order property. \begin{question} Suppose $\mathcal{L}$ is a finite relational language with $r_{\mathcal{L}}\geq 2$, and $\mathcal{H}$ is a fast-growing hereditary $\mathcal{L}$-property. Is there a finite $k=k(\mathcal{H})$ such that for all $n$ and $M\in \mathcal{R}_{ex}(n,\mathcal{H})$, every atomic formula $\phi(x;y)\in \mathcal{L}_{\mathcal{H}}$ does not have the $k$-order property in $M$? \end{question} A weaker version of this question is the following. We direct the reader to \cite{simon2015guide} for the definition of the $VC$-dimension of a formula. \begin{question} Suppose $\mathcal{L}$ is a finite relational language with $r_{\mathcal{L}}\geq 2$, and $\mathcal{H}$ is a fast-growing hereditary $\mathcal{L}$-property. Is there a finite $k=k(\mathcal{H})$ such that for all $n$ and $M\in \mathcal{R}_{ex}(n,\mathcal{H})$, every atomic formula $\phi(x;y)\in \mathcal{L}_{\mathcal{H}}$ has $VC$-dimension bounded by $k$ in $M$? \end{question}
a19e7cc772382c45e130c49c1e5e55a9877042a0
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Let $(\Omega,{\cal F},\mathbb{F},\mathbb{P})$ be a complete filtered probability space on which a standard one-dimensional Brownian motion $\{W(t),t\ges0\}$ is defined such that $\mathbb{F}=\{{\cal F}_t\}_{t\ges0}$ is the natural filtration of $W(\cdot)$ augmented by all the $\mathbb{P}$-null sets in ${\cal F}$. Consider the following controlled linear (forward) stochastic differential equation (FSDE, for short) on $[t,T]$: \begin{equation}\label{state}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big[A(s)X(s)+B_1(s)u_1(s)+B_2(s)u_2(s)+b(s)\big]ds\\ \noalign{\smallskip}\displaystyle\qquad\quad~~~+\big[C(s)X(s)+D_1(s)u_1(s)+D_2(s)u_2(s)+\sigma(s)\big]dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)= x.\end{array}\right.\end{equation} In the above, $X(\cdot)$ is called the {\it state process} taking values in the $n$-dimensional Euclidean space $\mathbb{R}^n$ with the {\it initial pair} $(t,x)\in[0,T)\times\mathbb{R}^n$; for $i=1,2$, $u_i(\cdot)$ is called the {\it control process} of Player $i$ taking values in $\mathbb{R}^{m_i}$. We assume that the {\it coefficients} $A(\cdot)$, $B_1(\cdot)$, $B_2(\cdot)$, $C(\cdot)$, $D_1(\cdot)$, and $D_2(\cdot)$ are deterministic matrix-valued functions of proper dimensions, and that $b(\cdot)$ and $\sigma(\cdot)$ are $\mathbb{F}$-progressively measurable processes taking values in $\mathbb{R}^n$. For $i=1,2$ and $t\in[0,T)$, we define $${\cal U}_i[t,T]=\Big\{u_i:[t,T]\times\Omega\to\mathbb{R}^{m_i}\bigm|u_i(\cdot) \hbox{ is $\mathbb{F}$-progressively measurable, }\mathbb{E}\int_t^T|u_i(s)|^2ds<\infty\Big\}.$$ Any element $u_i(\cdot)\in{\cal U}_i[t,T]$ is called an {\it admissible control} of Player $i$ on $[t,T]$. Under some mild conditions on the coefficients, for any initial pair $(t,x)\in[0,T)\times\mathbb{R}^n$ and controls $u_i(\cdot)\in{\cal U}_i[t,T]$, $i=1,2$, the state equation \eqref{state} admits a unique solution $X(\cdot)\equiv X(\cdot\,;t,x,u_1(\cdot),u_2(\cdot))$. The cost functional for Player $i$ is defined by the following: \begin{equation}\label{cost}\begin{array}{ll} \displaystyle J^i(t,x;u_1(\cdot),u_2(\cdot))\triangleq\mathbb{E}\Big\{\langle G^iX(T),X(T)\rangle+2\langle g^i,X(T)\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\qq\quad+\int_t^T\[\langle {\scriptstyle\begin{pmatrix}\scriptstyle Q^i(s) &\negthinspace \scriptstyle S^i_1(s)^\top &\negthinspace \scriptstyle S^i_2(s)^\top\\ \scriptstyle S^i_1(s) &\negthinspace \scriptstyle R^i_{11}(s) &\negthinspace \scriptstyle R^i_{12}(s) \\ \scriptstyle S^i_2(s) &\negthinspace \scriptstyle R^i_{21}(s) &\negthinspace \scriptstyle R^i_{22}(s) \end{pmatrix}} {\scriptstyle\begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u_1(s) \\ \scriptstyle u_2(s)\end{pmatrix}}, {\scriptstyle\begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u_1(s) \\ \scriptstyle u_2(s)\end{pmatrix}}\rangle +2\langle {\scriptstyle\begin{pmatrix}\scriptstyle q^i(s) \\ \scriptstyle\rho^i_1(s) \\ \scriptstyle\rho^i_2(s)\end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u_1(s) \\ \scriptstyle u_2(s) \end{pmatrix}}\rangle\]ds\Big\},\end{array}\end{equation} where $Q^i(\cdot)$, $S^i_1(\cdot)$, $S^i_2(\cdot)$, $R^i_{11}(\cdot)$, $R^i_{12}(\cdot)$, $R^i_{21}(\cdot)$, and $R^i_{22}(\cdot)$ are deterministic matrix-valued functions of proper dimensions with $$Q^i(\cdot)^\top=Q^i(\cdot),\quad R^i_{jj}(\cdot)^\top=R^i_{jj}(\cdot), \quad R^i_{12}(\cdot)^\top=R^i_{21}(\cdot), \qquad i,j=1,2,$$ where the superscript $^\top$ denotes the transpose of matrices, and $G^i$ is a symmetric matrix; $q^i(\cdot)$, $\rho^i_1(\cdot)$, and $\rho^i_2(\cdot)$ are allowed to be vector-valued $\mathbb{F}$-progressively measurable processes, and $g^i$ is allowed to be an ${\cal F}_T$-measurable random vector. Then we can formally pose the following problem. \medskip \bf Problem (SDG). \rm For any initial pair $(t,x)\in[0,T)\times\mathbb{R}^n$ and $i=1,2$, Player $i$ wants to find a control $u^*_i(\cdot)\in{\cal U}_i[t,T]$ such that the cost functional $J^i(t,x;u_1(\cdot),u_2(\cdot))$ is minimized. \medskip The above posed problem is referred to as a linear quadratic (LQ, for short) stochastic {\it two-person differential game}. In the case \begin{equation}\label{16Apr2-zero}\begin{array}{ll} \displaystyle J^1(t,x;u_1(\cdot),u_2(\cdot))+J^2(t,x;u_1(\cdot),u_2(\cdot))=0,\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\qq\qquad\qq\forall\,(t,x)\in[0,T]\times\mathbb{R}^n,~\forall\,u_i(\cdot)\in{\cal U}_i[t,T],\quad i=1,2, \end{array}\end{equation} the corresponding Problem (SDG) is called an LQ stochastic two-person {\it zero-sum} differential game. To guarantee \eqref{16Apr2-zero}, one usually assumes that \begin{equation}\label{16Apr2-1+2=0}\begin{array}{lllll} \displaystyle G^1+G^2=0, &~ g^1+g^2=0, &~ Q^1(\cdot)+Q^2(\cdot)=0, &~ q^1(\cdot)+q^2(\cdot)=0,\\ \noalign{\smallskip}\displaystyle S_j^1(\cdot)+S^2_j(\cdot)=0, &~ R^1_{jk}(\cdot)+R^2_{jk}(\cdot)=0, &~ \rho^1_j(\cdot)+\rho^2_j(\cdot)=0, &~ j,k=1,2. \end{array}\end{equation} We refer the readers to \cite{Sun-Yong 2014} (and the references cited therein) for the case of LQ stochastic two-person zero-sum differential games. Recall that in \cite{Sun-Yong 2014}, open-loop and closed-loop saddle points were introduced and it was established that the existence of an open-loop saddle point for the problem is equivalent to the solvability of a forward-backward stochastic differential equation (FBSDE, for short), and the existence of a closed-loop saddle point for the problem is equivalent to the solvability of a (differential) Riccati equation. In this paper, we will not assume \eqref{16Apr2-1+2=0} so that \eqref{16Apr2-zero} is not necessarily true. Such a Problem (SDG) is usually referred to as an LQ stochastic two-person {\it nonzero-sum} differential game, emphasizing that \eqref{16Apr2-zero} is not assumed. We have two main goals in this paper: Establish a theory for Problem (SDG) parallel to that of \cite{Sun-Yong 2014} (for zero-sum case); and study the difference between the closed-loop representation of open-loop Nash equilibria and the outcome of closed-loop Nash equilibria. It turns out that the above-mentioned difference for the non-zero sum case is indicated through the symmetry of the corresponding Riccati equations: One is symmetric and the other is not. On the other hand, we found that the situation in the zero-sum case, which was not discussed in \cite{Sun-Yong 2014}, is totally different: The closed-loop representation of open-loop saddle points coincides with the outcome of the corresponding closed-loop saddle point, when both exist. In particular, for stochastic linear quadratic optimal control problem, the closed-loop representation of open-loop optimal controls is the outcome of the corresponding closed-loop optimal strategy (\cite{Sun-Li-Yong 2016}). \medskip Mathematically, posing condition \eqref{16Apr2-1+2=0} makes the structure of the problem much simpler, since with such a condition, only one performance index is needed, for which one player is the minimizer and the other player is the maximizer. However, as we know that in the real life, each player should have his/her own cost functional, and even for the totally hostile situation, the objectives of the opponents might not necessarily be exactly the opposite (zero-sum). Therefore, realistically, it is more meaningful to investigate Problem (SDG) without assuming \eqref{16Apr2-1+2=0}. By the way, although we will not discuss such a situation in the current paper, we still would like to point out that sometimes, certain cooperations between the players might result in both players rewarded more. \medskip Static version of nonzero-sum differential games could be regarded as a kind of non-cooperative games for which one can trace back to the work of Nash \cite{Nash 1951}. For some early works on nonzero-sum differential games, we would like to mention Lukes--Russell \cite{Lukes-Russell 1971}, Friedman \cite{Friedman 1972}, and Bensoussan \cite{Bensoussan 1974}. In the past two decays, due to the appearance of backward stochastic differential equations (BSDEs, for short), some new and interesting works published; Among them, we would like to mention \cite{Hamadene 1998, Hamadene 1999,El Karoui-Hamadene 2003, Buckdahn-Cardaliaguet-Rainer 2004, Rainer 2007, Hamadene-Mu 2015}. \medskip The rest of the paper is organized as follows. Section 2 will collect some preliminaries. Among other things, we will recall some known results on LQ optimal control problems. In Section 3, we will introduce open-loop and closed-loop Nash equilibria. A characterization of the existence of open-loop Nash equilibria in terms of solvability of two coupled FBSDEs will be presented in Section 4. Section 5 is devoted to the discussion on the closed-loop Nash equilibria whose existence is characterized by the solvability of two coupled symmetric Riccati equations. In Section 6, we will present two examples showing the difference between open-loop and closed-loop Nash equilibria. In Section 7, closed-loop representation of open-loop Nash equilibria will be studied, and comparison between the closed-loop representation of open-loop Nash equilibria and the outcome of closed-loop Nash equilibria will be carried out. Finally, we will take a deeper look at the situation for LQ zero-sum games in Section 8. \section{Preliminaries} Let $\mathbb{R}^{n\times m}$ be the space of all $(n\times m)$ matrices and $\mathbb{S}^n\subseteq\mathbb{R}^{n\times n}$ be the set of all $(n\times n)$ symmetric matrices. The inner product $\langle\cdot\,,\cdot\rangle$ on $\mathbb{R}^{n\times m}$ is given by $\langle M,N\rangle\mapsto\hbox{\rm tr$\,$}(M^\top N)$, and the induced norm is given by $|M|=\sqrt{\hbox{\rm tr$\,$}(M^\top M)}$. We denote by $\mathscr{R}(M)$ the range of a matrix $M$, and for $M,N\in\mathbb{S}^n$ we use the notation $M\geqslant N$ (respectively, $M>N$) to indicate that $M-N$ is positive semi-definite (respectively, positive definite). Recall that any $M\in\mathbb{R}^{n\times m}$ admits a unique (Moore--Penrose) {\it pseudo-inverse} $M^\dag\in\mathbb{R}^{m\times n}$ having the following properties (\cite{Penrose 1955}): $$MM^\dag M=M, \quad M^\dag MM^\dag=M^\dag, \quad (MM^\dag)^\top=MM^\dag, \quad (M^\dag M)^\top=M^\dag M.$$ Further, if $M\in\mathbb{R}^{n\times m}$ and $\Psi\in\mathbb{R}^{n\times\ell}$ such that $$\mathscr{R}(\Psi)\subseteq\mathscr{R}(M),$$ then all the solutions $\Theta$ to the linear equation $$M\Theta=\Psi$$ are given by the following: $$\Theta=M^\dag\Psi+(I-M^\dag M)\Gamma,\qquad\Gamma\in\mathbb{R}^{m\times\ell}.$$ In addition, if $M=M^\top\in\mathbb{S}^n$, then $$M^\dag=(M^\dag)^\top, \quad MM^\dag=M^\dag M; \quad\hbox{and}\quad M\ges0\iff M^\dag\ges0.$$ Next, let $T>0$ be a fixed time horizon. For any $t\in[0,T]$ and Euclidean space $\mathbb{H}$, we introduce the following spaces of deterministic functions: $$\begin{array}{ll} \noalign{\smallskip}\displaystyle L^p(t,T;\mathbb{H})=\Big\{\varphi:[t,T]\to\mathbb{H}\bigm|\int_t^T|\varphi(s)|^pds<\infty\Big\},\quad 1\leqslant p<\infty,\end{array}$$ $$\begin{array}{ll} \noalign{\smallskip}\displaystyle L^\infty(t,T;\mathbb{H})=\Big\{\varphi:[t,T]\to\mathbb{H}\bigm|\mathop{\rm esssup}_{s\in[t,T]}|\varphi(s)|<\infty\Big\},\\ \noalign{\smallskip}\displaystyle C([t,T];\mathbb{H})=\Big\{\varphi:[t,T]\to\mathbb{H}\bigm|\varphi(\cdot)\hbox{ is continuous}\Big\}.\end{array}$$ Further, we introduce the following spaces of random variables and stochastic processes: For any $t\in[0,T]$, $$\begin{array}{ll} \displaystyle L^2_{{\cal F}_t}(\Omega;\mathbb{H})=\Big\{\xi:\Omega\to\mathbb{H}\bigm|\xi\hbox{ is ${\cal F}_t$-measurable, }\mathbb{E}|\xi|^2<\infty\Big\},\\ \noalign{\smallskip}\displaystyle L_\mathbb{F}^2(t,T;\mathbb{H})=\Big\{\varphi:[t,T]\times\Omega\to\mathbb{H}\bigm|\varphi(\cdot)\hbox{ is $\mathbb{F}$-progressively measurable, } \mathbb{E}\int^T_t|\varphi(s)|^2ds<\infty\Big\},\\ \noalign{\smallskip}\displaystyle L_\mathbb{F}^2(\Omega;C([t,T];\mathbb{H}))=\Big\{\varphi:[t,T]\times\Omega\to\mathbb{H}\bigm|\varphi(\cdot)\hbox{ is $\mathbb{F}$-adapted, continuous, } \mathbb{E}\(\sup_{t\leqslant s\leqslant T}|\varphi(s)|^2\)<\infty\Big\},\\ \noalign{\smallskip}\displaystyle L^2_\mathbb{F}(\Omega;L^1(t,T;\mathbb{H}))=\Big\{\varphi:[t,T]\times\Omega\to\mathbb{H}\bigm|\varphi(\cdot)\hbox{ is $\mathbb{F}$-progressively measurable, } \mathbb{E}\(\int_t^T|\varphi(s)|ds\)^2<\infty\Big\}.\end{array}$$ \smallskip We now recall some results on stochastic LQ optimal control problems. Consider the state equation \begin{equation}\label{LQ-state}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big[A(s)X(s)+B(s)u(s)+b(s)\big]ds\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big[C(s)X(s)+D(s)u(s)+\sigma(s)\big]dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x. \end{array}\right.\end{equation} The cost functional takes the following form: \begin{equation}\label{LQ-cost}\begin{array}{ll} \displaystyle J(t,x;u(\cdot))\triangleq\mathbb{E}\Big\{\langle GX(T),X(T)\rangle+2\langle g,X(T)\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\qq +\int_t^T\[\langle{\scriptstyle\begin{pmatrix}\scriptstyle Q(s) & \scriptstyle S(s)^\top \\ \scriptstyle S(s) & \scriptstyle R(s) \end{pmatrix} \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u(s)\end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u(s)\end{pmatrix}}\rangle +2\langle{\scriptstyle\begin{pmatrix}\scriptstyle q(s) \\ \scriptstyle\rho(s)\end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u(s) \end{pmatrix}}\rangle\]ds\Big\}. \end{array}\end{equation} We adopt the following assumptions. \medskip {\bf(S1)} The coefficients of the state equation satisfy the following: $$\left\{\negthinspace \negthinspace \begin{array}{llll} A(\cdot)\in L^1(0,T;\mathbb{R}^{n\times n}), &~ B(\cdot)\in L^2(0,T;\mathbb{R}^{n\times m}), &~ b(\cdot)\in L^2_\mathbb{F}(\Omega;L^1(0,T;\mathbb{R}^n)),\\ \noalign{\smallskip} C(\cdot)\in L^2(0,T;\mathbb{R}^{n\times n}), &~ D(\cdot)\in L^\infty(0,T;\mathbb{R}^{n\times m}), &~ \sigma(\cdot)\in L_\mathbb{F}^2(0,T;\mathbb{R}^n). \end{array}\right.$$ {\bf(S2)} The weighting coefficients in the cost functional satisfy the following: $$\left\{\negthinspace \negthinspace \begin{array}{llll} Q(\cdot)\in L^1(0,T;\mathbb{S}^n), & S(\cdot)\in L^2(0,T;\mathbb{R}^{m\times n}), & R(\cdot)\in L^\infty(0,T;\mathbb{S}^m),\\ \noalign{\smallskip} q(\cdot)\in L^2_\mathbb{F}(\Omega;L^1(0,T;\mathbb{R}^n)), & \rho(\cdot)\in L_\mathbb{F}^2(0,T;\mathbb{R}^m), & g\in L^2_{{\cal F}_T}(\Omega;\mathbb{R}^n), \quad G\in\mathbb{S}^n. \end{array}\right.$$ Note that under (S1), for any $(t,x)\in[0,T)\times\mathbb{R}^n$ and $u(\cdot)\in{\cal U}[t,T]\equiv L^2_\mathbb{F}(t,T;\mathbb{R}^m)$, the state equation \eqref{LQ-state} admits a unique strong solution $X(\cdot)\equiv X(\cdot\,;t,x,u(\cdot))$. Further, if (S2) is also assumed, then the cost functional \eqref{LQ-cost} is well-defined for every $(t,x)\in[0,T)\times\mathbb{R}^n$ and $u(\cdot)\in{\cal U}[t,T]$. Therefore, the following problem is meaningful. \medskip \bf Problem (SLQ). \rm For any given initial pair $(t,x)\in[0,T)\times\mathbb{R}^n$, find a $\bar u(\cdot)\in{\cal U}[t,T]$ such that \begin{equation}\label{J(u)=inf}J(t,x;\bar u(\cdot))=\inf_{u(\cdot)\in{\cal U}[t,T]}J(t,x;u(\cdot)).\end{equation} Any $\bar u(\cdot)\in{\cal U}[t,T]$ satisfying \eqref{J(u)=inf} is called an {\it open-loop optimal control} of Problem (SLQ) for $(t,x)$; the corresponding $\bar X(\cdot)\equiv X(\cdot\,;t,x,\bar u(\cdot))$ is called an {\it open-loop optimal state process} and $(\bar X(\cdot),\bar u(\cdot))$ is called an {\it open-loop optimal pair}. \begin{definition}\label{open-loop solvable} \rm Let $(t,x)\in[0,T)\times\mathbb{R}^n$. If there exists a (unique) $\bar u(\cdot)\in{\cal U}[t,T]$ such that \eqref{J(u)=inf} holds, then we say that Problem (SLQ) is ({\it uniquely}) {\it open-loop solvable at} $(t,x)$. If Problem (SLQ) is (uniquely) open-loop solvable for any $(t,x)\in[0,T)\times\mathbb{R}^n$, then we say that Problem (SLQ) is ({\it uniquely}) {\it open-loop solvable on} $[0,T)\times\mathbb{R}^n$. \end{definition} The following result is concerned with open-loop optimal controls of Problem (SLQ) for a given initial pair, whose proof can be found in \cite{Sun-Yong 2014} (see also \cite{Sun-Li-Yong 2016}). \begin{theorem}\label{bt-16Apr2-17:10}\sl Let {\rm(S1)--(S2)} hold. For a given initial pair $(t,x)\in[0,T)\times\mathbb{R}^n$, a state-control pair $(\bar X(\cdot),\bar u(\cdot))$ is an open-loop optimal pair of Problem {\rm(SLQ)} if and only if the following hold: \medskip {\rm(i)} The {\it stationarity condition} holds: $$B(s)^\top\bar Y(s)+D(s)^\top\bar Z(s)+S(s)\bar X(s)+R(s)\bar u(s)+\rho(s)=0, \quad \hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}$$ where $(\bar Y(\cdot),\bar Z(\cdot))$ is the {\it adapted solution} to the following BSDE: $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\bar Y(s)=-\big[A(s)^\top\bar Y(s)+C(s)^\top\bar Z(s)+Q(s)\bar X(s)+S(s)^\top\bar u(s)+q(s)\big]ds+\bar Z(s)dW(s),\quad s\in[t,T],\\ \noalign{\smallskip}\displaystyle\bar Y(T)=G\bar X(T)+g. \end{array}\right.$$ {\rm(ii)} The map $u(\cdot)\mapsto J(t,0;u(\cdot))$ is convex. \end{theorem} Next, for any given $t\in[0,T)$, take $\Theta(\cdot)\in L^2(t,T;\mathbb{R}^{m\times n})\equiv{\cal Q}[t,T]$ and $v(\cdot)\in{\cal U}[t,T]$. For any $x\in\mathbb{R}^n$, let us consider the following equation: \begin{equation}\label{closed-loop1}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big\{[A(s)+B(s)\Theta(s)]X(s)+B(s)v(s)+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\{[C(s)+D(s)\Theta(s)]X(s)+D(s)v(s)+\sigma(s)\big\}dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x, \end{array}\right.\end{equation} which admits a unique solution $X(\cdot)\equiv X(\cdot\,;t,x,\Theta(\cdot),v(\cdot))$, depending on $\Theta(\cdot)$ and $v(\cdot)$. The above is called a {\it closed-loop system} of the original state equation \eqref{LQ-state} under {\it closed-loop strategy} $(\Theta(\cdot),v(\cdot))$. We point out that $(\Theta(\cdot),v(\cdot))$ is independent of the initial state $x$. With the above corresponding solution $X(\cdot)$, we define $$\begin{array}{ll} \displaystyle J(t,x;\Theta(\cdot)X(\cdot)+v(\cdot))=\mathbb{E}\Big\{\langle GX(T),X(T)\rangle+2\langle g,X(T)\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad~ +\int_t^T\[\langle{\scriptstyle\begin{pmatrix}\scriptstyle Q(s) &\negthinspace \negthinspace \scriptstyle S(s)^\top \\ \scriptstyle S(s) &\negthinspace \negthinspace \scriptstyle R(s) \end{pmatrix} \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle\Theta(s)X(s)+v(s) \end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle\Theta(s)X(s)+v(s) \end{pmatrix}}\rangle % +2\langle{\scriptstyle\begin{pmatrix}\scriptstyle q(s) \\ \scriptstyle\rho(s) \end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle\Theta(s)X(s)+v(s)\end{pmatrix}}\rangle\]ds\Big\}. \end{array}$$ We now recall the following definition. \begin{definition}\label{bde-16Apr2-21:00}\rm A pair $(\bar\Theta(\cdot),\bar v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$ is called a {\it closed-loop optimal strategy} of Problem (SLQ) on $[t,T]$ if \begin{equation}\label{16Apr2-21:30}\begin{array}{ll} \displaystyle J(t,x;\bar\Theta(\cdot)\bar X(\cdot)+\bar v(\cdot))\leqslant J(t,x;\Theta(\cdot)X(\cdot)+v(\cdot)),\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\qq\qquad\qq\quad\forall\, x\in\mathbb{R}^n,~\forall\,(\Theta(\cdot),v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T], \end{array}\end{equation} where $\bar X(\cdot)=X(\cdot\,;t,x,\bar\Theta(\cdot),\bar v(\cdot))$, and $X(\cdot)=X(\cdot\,;t,x,\Theta(\cdot),v(\cdot))$. \end{definition} We emphasize that the pair $(\bar\Theta(\cdot),\bar v(\cdot))$ is required to be independent of the initial state $x\in\mathbb{R}^n$. It is interesting that the following equivalent theorem holds. \begin{proposition}\label{strategy equivalence}\sl Let {\rm(S1)--(S2)} hold and let $(\bar\Theta(\cdot),\bar v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$. Then the following statements are equivalent: \medskip {\rm(i)} $(\bar\Theta(\cdot),\bar v(\cdot))$ is a closed-loop optimal strategy of Problem {\rm(SLQ)} on $[t,T]$. \medskip {\rm(ii)} For any $x\in\mathbb{R}^n$ and $v(\cdot)\in{\cal U}[t,T]$, $$J(t,x;\bar\Theta(\cdot)\bar X(\cdot)+\bar v(\cdot))\leqslant J(t,x;\bar\Theta(\cdot)X(\cdot)+v(\cdot)),$$ where $\bar X(\cdot)=X(\cdot\,;t,x,\bar\Theta(\cdot),\bar v(\cdot))$ and $X(\cdot)=X(\cdot\,;t,x,\bar\Theta(\cdot),v(\cdot))$. \medskip {\rm(iii)} For any $x\in\mathbb{R}^n$ and $u(\cdot)\in{\cal U}[t,T]$, \begin{equation}\label{bar v<u}J(t,x;\bar\Theta(\cdot)\bar X(\cdot)+\bar v(\cdot))\leqslant J(t,x;u(\cdot)),\end{equation} where $\bar X(\cdot)=X(\cdot\,;t,x,\bar\Theta(\cdot),\bar v(\cdot))$. \end{proposition} \it Proof. \rm The implication (i) $\mathop{\Rightarrow}$ (ii) follows by taking $\Theta(\cdot)=\bar\Theta(\cdot)$ in \eqref{16Apr2-21:30}. \medskip For the implication (ii) $\mathop{\Rightarrow}$ (iii), take any $u(\cdot)\in{\cal U}[t,T]$ and let $X(\cdot)=X(\cdot\,;t,x,u(\cdot))$. Then $$\begin{array}{ll} \displaystyle dX(s)=\big\{[A(s)+B(s)\bar\Theta(s)]X(s)+B(s)[u(s)-\bar\Theta(s)X(s)]+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq\quad +\big\{[C(s)+D(s)\bar\Theta(s)]X(s)+D(s)[u(s)-\bar\Theta(s)X(s)]+\sigma(s)\big\}dW(s),\end{array}$$ with $X(t)=x$. Thus, if let $$v(\cdot)=u(\cdot)-\bar\Theta(\cdot)X(\cdot),$$ we have $$J(t,x;\bar\Theta(\cdot)\bar X(\cdot)+\bar v(\cdot))\leqslant J(t,x;\bar\Theta(\cdot)X(\cdot)+v(\cdot))=J(t,x;u(\cdot)),$$ which proves (iii). \medskip For the implication (iii) $\mathop{\Rightarrow}$ (i), take any $(\Theta(\cdot),v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$ and let $X(\cdot)$ be the solution to \eqref{closed-loop1}. Let $u(\cdot)=\Theta(\cdot)X(\cdot)+v(\cdot)$, Then by (iii), we have $$J(t,x;\bar\Theta(\cdot)\bar X(\cdot)+\bar v(\cdot))\leqslant J(t,x;u(\cdot))=J(t,x;\Theta(\cdot)X(\cdot)+v(\cdot)).$$ This completes the proof. \signed {$\sqr69$} \medskip From the above result, we see that if $(\bar\Theta(\cdot),\bar v(\cdot))$ is a closed-loop optimal strategy of Problem (SLQ) on $[t,T]$, then for any fixed initial state $x\in\mathbb{R}^n$, with $\bar X(\cdot)$ denoting the state process corresponding to $(t,x)$ and $(\bar\Theta(\cdot),\bar v(\cdot))$, \eqref{bar v<u} implies that the outcome $$\bar u(\cdot)\equiv\bar\Theta(\cdot)\bar X(\cdot)+\bar v(\cdot)\in{\cal U}[t,T]$$ is an open-loop optimal control of Problem (SLQ) for $(t,x)$. Therefore, for Problem (SLQ), the existence of closed-loop strategies on $[t,T]$ implies the existence of open-loop optimal controls for initial pair $(t,x)$ for any $x\in\mathbb{R}^n$. We point out that the situation will be different for two-person differential games. Details will be carried out later. \medskip For closed-loop optimal strategies, we have the following characterization (\cite{Sun-Yong 2014,Sun-Li-Yong 2016}). \begin{theorem}\label{closed-loop strategy}\sl Let {\rm(S1)--(S2)} hold. Then Problem {\rm(SLQ)} admits a closed-loop optimal strategy on $[t,T]$ if and only if the following Riccati equation admits a solution $P(\cdot)\in C([t,T];\mathbb{S}^n)$: $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P+PA+A^\top P+C^\top PC+Q\\ \noalign{\smallskip}\displaystyle\,~~-(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag(B^\top P+D^\top PC+S)=0,\qquad\hbox{\rm a.e.}~\hbox{on }[t,T],\\ \noalign{\smallskip}\displaystyle\mathscr{R}(B^\top P+D^\top PC+S)\subseteq\mathscr{R}(R+D^\top PD),\qquad\hbox{\rm a.e.}~\hbox{on }[t,T],\\ \noalign{\smallskip}\displaystyle R+D^\top PD\ges0,\qquad\hbox{\rm a.e.}~\hbox{on }[t,T],\\ \noalign{\smallskip}\displaystyle P(T)=G, \end{array}\right.$$ such that $$(R+D^\top PD)^\dag(B^\top P+D^\top PC+S)\in L^2(t,T;\mathbb{R}^{m\times n}),$$ and the adapted solution $(\eta(\cdot),\zeta(\cdot))$ to the BSDE $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\eta=-\,\Big\{\big[A-B(R+D^\top PD)^\dag(B^\top P+D^\top PC+S)\big]^\top\eta\\ \noalign{\smallskip}\displaystyle\qquad\quad~~+\big[C-D(R+D^\top PD)^\dag(B^\top P+D^\top PC+S)\big]^\top\zeta\\ \noalign{\smallskip}\displaystyle\qquad\quad~~+\big[C-D(R+D^\top PD)^\dag(B^\top P+D^\top PC+S)\big]^\top P\sigma\\ \noalign{\smallskip}\displaystyle\qquad\quad~~-(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag\rho+Pb+q\Big\}ds+\zeta dW,\quad~ s\in[t,T],\\ \noalign{\smallskip}\displaystyle\eta(T)=g, \end{array}\right.$$ satisfies $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle B^\top\eta+D^\top\zeta+D^\top P\sigma+\rho\in\mathscr{R}(R+D^\top PD),\quad \hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\\ \noalign{\smallskip}\displaystyle (R+D^\top PD)^\dag(B^\top\eta+D^\top\zeta+D^\top P\sigma+\rho)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m). \end{array}\right.$$ In this case, any closed-loop optimal strategy $(\bar\Theta(\cdot),\bar v(\cdot))$ of Problem {\rm(SLQ)} admits the following representation: $$\left\{\negthinspace \negthinspace \begin{array}{cll} \bar\Theta \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \negthinspace \negthinspace \negthinspace ~ -(R+D^\top PD)^\dag(B^\top P+D^\top PC+S) +\big[I-(R+D^\top PD)^\dag(R+D^\top PD)\big]\theta,\\ \noalign{\smallskip}\bar v \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \negthinspace \negthinspace \negthinspace ~ -(R+D^\top PD)^\dag(B^\top\eta+D^\top\zeta+D^\top P\sigma+\rho) +\big[I-(R+D^\top PD)^\dag(R+D^\top PD)\big]\nu, \end{array}\right.$$ for some $\theta(\cdot)\in L^2(t,T;\mathbb{R}^{m\times n})$ and $\nu(\cdot)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m)$. Further, the value function is given by $$\begin{array}{ll} \noalign{\smallskip}\displaystyle V(t,x)=\mathbb{E}\Big\{\langle P(t)x,x\rangle+2\langle\eta(t),x\rangle +\int_t^T\[\langle P\sigma,\sigma\rangle+2\langle\eta,b\rangle+2\langle\zeta,\sigma\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\quad~-\big\langle(R+D^\top PD)^\dag(B^\top\eta+D^\top\zeta+D^\top P\sigma+\rho), B^\top\eta+D^\top\zeta+D^\top P\sigma+\rho\big\rangle\]ds\Big\}. \end{array}$$ \end{theorem} \section{Stochastic Differential Games} We return to our Problem (SDG). Recall the sets ${\cal U}_i[t,T]=L^2_\mathbb{F}(t,T;\mathbb{R}^{m_i})$ of all open-loop controls of Player $i$ ($i=1,2$). For notational simplicity, we let $m=m_1+m_2$ and denote $$\begin{array}{ll} \displaystyle B(\cdot)=(B_1(\cdot),B_2(\cdot)),\quad D(\cdot)=(D_1(\cdot),D_2(\cdot)),\\ \noalign{\smallskip}\displaystyle S^i(\cdot)\negthinspace =\negthinspace \begin{pmatrix}S^i_1(\cdot) \\ S^i_2(\cdot)\end{pmatrix}, \quad R^i(\cdot)\negthinspace =\negthinspace \begin{pmatrix}R^i_{11}(\cdot) & R^i_{12}(\cdot)\\ R^i_{21}(\cdot) & R^i_{22}(\cdot)\end{pmatrix} \negthinspace \equiv\negthinspace \begin{pmatrix}R^i_1(\cdot) \\ R^i_2(\cdot)\end{pmatrix}, \quad \rho^i(\cdot)\negthinspace =\negthinspace \begin{pmatrix}\rho^i_1(\cdot) \\ \rho^i_2(\cdot)\end{pmatrix}, \quad u(\cdot)\negthinspace =\negthinspace \begin{pmatrix}u_1(\cdot) \\ u_2(\cdot)\end{pmatrix}. \end{array}$$ Naturally, we identify ${\cal U}[t,T]={\cal U}_1[t,T]\times{\cal U}_2[t,T]$. With such notations, the state equation becomes \begin{equation}\label{16Apr3-state}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big[A(s)X(s)+B(s)u(s)+b(s)\big]ds\\ \noalign{\smallskip}\displaystyle\qquad\qq\negthinspace \negthinspace ~+\big[C(s)X(s)+D(s)u(s)+\sigma(s)\big]dW(s), \qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x, \end{array}\right.\end{equation} and the cost functionals become ($i=1,2$) $$\begin{array}{ll} \displaystyle J^i(t,x;u(\cdot))=\mathbb{E}\Big\{\langle G^iX(T),X(T)\rangle+2\langle g^i,X(T)\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\qq+\int_t^T\[\big\langle \scriptstyle{\begin{pmatrix}\scriptstyle Q^i(s) &\scriptstyle S^i(s)^\top \\ \scriptstyle S^i(s) &\scriptstyle R^i(s)\end{pmatrix} \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u(s) \end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u(s) \end{pmatrix}}\big\rangle+2\big\langle \begin{pmatrix}\scriptstyle q^i(s) \\ \scriptstyle\rho^i(s) \end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u(s) \end{pmatrix}\big\rangle\]ds\Big\}. \end{array}$$ Now let us introduce the following standard assumptions: \medskip {\bf(G1)} The coefficients of the state equation satisfy the following: $$\left\{\negthinspace \negthinspace \begin{array}{lll} \displaystyle A(\cdot)\in L^1(0,T;\mathbb{R}^{n\times n}), & B(\cdot)\in L^2(0,T;\mathbb{R}^{n\times m}), & b(\cdot)\in L^2_\mathbb{F}(\Omega;L^1(0,T;\mathbb{R}^n)),\\ \noalign{\smallskip}\displaystyle C(\cdot)\in L^2(0,T;\mathbb{R}^{n\times n}), & D(\cdot)\in L^\infty(0,T;\mathbb{R}^{n\times m}), & \sigma(\cdot)\in L_\mathbb{F}^2(0,T;\mathbb{R}^n). \end{array}\right.$$ {\bf(G2)} The weighting coefficients in the cost functionals satisfy the following: For $i=1,2$, $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle Q^i(\cdot)\in L^1(0,T;\mathbb{S}^n),\quad S^i(\cdot)\in L^2(0,T;\mathbb{R}^{m\times n}), \quad R^i(\cdot)\in L^\infty(0,T;\mathbb{S}^m),\\ \noalign{\smallskip}\displaystyle q^i(\cdot)\in L^2_\mathbb{F}(\Omega;L^1(0,T;\mathbb{R}^n)),\quad\rho^i(\cdot)\in L_\mathbb{F}^2(0,T;\mathbb{R}^m),\quad g^i\in L^2_{{\cal F}_T}(\Omega;\mathbb{R}^n),\quad G^i\in\mathbb{S}^n.\end{array}\right.$$ Under (G1), for any $(t,x)\in[0,T)\times\mathbb{R}^n$ and $u(\cdot)=(u_1(\cdot)^\top,u_2(\cdot)^\top)^\top\in{\cal U}[t,T]$, equation \eqref{16Apr3-state} admits a unique solution (\cite{Yong-Zhou 1999}) $$X(\cdot)\triangleq X(\cdot\,;t,x,u_1(\cdot),u_2(\cdot))\equiv X(\cdot\,;t,x,u(\cdot)) \in L^2_\mathbb{F}(\Omega;C([t,T];\mathbb{R}^n)).$$ Moreover, the following estimate holds: $$\mathbb{E}\(\sup_{t\leqslant s\leqslant T}|X(s)|^2\)\leqslant K\mathbb{E}\Big\{|x|^2+\(\int_t^T|b(s)|ds\)^2+\int_t^T|\sigma(s)|^2ds+\int^T_t|u(s)|^2ds\Big\},$$ where $K>0$ represents a generic constant. Therefore, under (G1)--(G2), the cost functionals $J^i(t,x;u(\cdot))\equiv J^i(t,x;u_1(\cdot),u_2(\cdot))$ are well-defined for all $(t,x)\in[0,T)\times\mathbb{R}^n$ and all $(u_1(\cdot),u_2(\cdot))\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$. Having the above, we now introduce the following definition. \begin{definition}\label{open-Nash-equilibrium} \rm A pair $(u^*_1(\cdot),u^*_2(\cdot))\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$ is called an {\it open-loop Nash equilibrium} of Problem (SDG) for the initial pair $(t,x)\in[0,T)\times\mathbb{R}^n$ if \begin{equation}\label{Nash-open*}\begin{array}{ll} \noalign{\smallskip}\displaystyle J^1(t,x;u^*_1(\cdot),u^*_2(\cdot))\leqslant J^1(t,x;u_1(\cdot),u^*_2(\cdot)),\qquad\forall\, u_1(\cdot)\in{\cal U}_1[t,T],\\ \noalign{\smallskip}\displaystyle J^2(t,x;u^*_1(\cdot),u^*_2(\cdot))\leqslant J^2(t,x;u^*_1(\cdot),u_2(\cdot)), \qquad\forall\, u_2(\cdot)\in{\cal U}_2[t,T].\end{array}\end{equation} \end{definition} Next, we denote $${\cal Q}_i[t,T]=L^2(t,T;\mathbb{R}^{m_i\times n}),\qquad i=1,2.$$ For any initial pair $(t,x)\in[0,T)\times\mathbb{R}^n$, $\Theta(\cdot)\equiv(\Theta_1(\cdot)^\top,\Theta_2(\cdot)^\top)^\top\in{\cal Q}_1[t,T]\times{\cal Q}_2[t,T]$ and any $v(\cdot)\equiv(v_1(\cdot)^\top$, $v_2(\cdot)^\top)^\top\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$, consider the following system: \begin{equation}\label{16Apr4-state-closed}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big\{[A(s)+B(s)\Theta(s)]X(s)+B(s)v(s)+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq\negthinspace ~+\big\{[C(s)+D(s)\Theta(s)]X(s)+D(s)v(s)+\sigma(s)\big\}dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)= x. \end{array}\right.\end{equation} Under (G1), the above admits a unique solution $X(\cdot)\equiv X(\cdot\,;t,x,\Theta_1(\cdot),v_1(\cdot),\Theta_2(\cdot),v_2(\cdot))$. If we denote \begin{equation}\label{ui}u_i(\cdot)=\Theta_i(\cdot)X(\cdot)+v_i(\cdot),\qquad i=1,2,\end{equation} then the above \eqref{16Apr4-state-closed} coincides with the original state equation \eqref{state}. We call $(\Theta_i(\cdot),v_i(\cdot))$ a {\it closed-loop strategy} of Player $i$, and call \eqref{16Apr4-state-closed} the {\it closed-loop system} of the original system under closed-loop strategies $(\Theta_1(\cdot),v_1(\cdot))$ and $(\Theta_2(\cdot),v_2(\cdot))$ of Players 1 and 2. Also, we call $u(\cdot)\equiv(u_1(\cdot)^\top,u_2(\cdot)^\top)^\top$ with $u_i(\cdot)$ defined by \eqref{ui} the outcome of the closed-loop strategy $(\Theta(\cdot),v(\cdot))$. With the solution $X(\cdot)$ to \eqref{16Apr4-state-closed}, we denote \begin{equation}\label{16Apr7-22:00}\begin{array}{ll} \noindent\medskip\displaystyle J^i(t,x;\Theta(\cdot)X(\cdot)+v(\cdot))\equiv J^i(t,x;\Theta_1(\cdot)X(\cdot)+v_1(\cdot),\Theta_2(\cdot)X(\cdot)+v_2(\cdot))\\ \noindent\medskip\displaystyle=\mathbb{E}\Big\{\langle G^iX(T),X(T)\rangle+2\langle g^i,X(T)\rangle\\ \noindent\medskip\displaystyle\quad~+\int_t^T\[\big\langle{\scriptstyle \begin{pmatrix}\scriptstyle Q^i & \scriptstyle (S^i)^\top\\ \scriptstyle S^i & \scriptstyle R^i \end{pmatrix} \begin{pmatrix}\scriptstyle X \\ \scriptstyle\Theta X+v \end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle\Theta X+v \end{pmatrix}}\big\rangle +2\big\langle{\scriptstyle \begin{pmatrix}\scriptstyle q^i \\ \scriptstyle\rho^i \end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle\Theta X+v\end{pmatrix}}\big\rangle\]ds\Big\}\\ \noindent\medskip\displaystyle=\mathbb{E}\Big\{\langle G^iX(T),X(T)\rangle+2\langle g^i,X(T)\rangle\\ \displaystyle\quad~+\int_t^T\[\big\langle{\scriptstyle \begin{pmatrix}\scriptstyle Q^i+\Theta^\top S^i+(S^i)^\top\Theta+\Theta^\top R^i\Theta &\scriptstyle (S^i)^\top+\Theta^\top R^i\\ \scriptstyle S^i+R^i\Theta &\scriptstyle R^i \end{pmatrix} \begin{pmatrix}\scriptstyle X \\ \scriptstyle v\end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle v\end{pmatrix}}\big\rangle +2\big\langle{\scriptstyle \begin{pmatrix}\scriptstyle q^i+\Theta^\top\rho^i \\ \scriptstyle\rho^i\end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle v \end{pmatrix}}\big\rangle\]ds\Big\}.\end{array}\end{equation} Similarly, one can define $J^i(t,x;\Theta_1(\cdot)X(\cdot)+v_1(\cdot),u_2(\cdot))$ and $J^i(t,x;u_1(\cdot),\Theta_2(\cdot)X(\cdot)+v_2(\cdot))$. We now introduce the following definition. \begin{definition}\label{bde-16Apr4-17:00}\rm A 4-tuple $(\Theta_1^*(\cdot),v_1^*(\cdot);\Theta_2^*(\cdot),v_2^*(\cdot)) \in{\cal Q}_1[t,T]\times{\cal U}_1[t,T]\times{\cal Q}_2[t,T]\times{\cal U}_2[t,T]$ is called a {\it closed-loop Nash equilibrium} of Problem (SDG) on $[t,T]$ if for any $x\in\mathbb{R}^n$ and any 4-tuple $(\Theta_1(\cdot),v_1(\cdot);$ $\Theta_2(\cdot),v_2(\cdot))\in {\cal Q}_1[t,T]\times{\cal U}_1[t,T]\times{\cal Q}_2[t,T]\times{\cal U}_2[t,T]$, the following hold: \begin{eqnarray} &&\label{Nash-closed1}\begin{array}{lll} \displaystyle J^1(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta_2^*(\cdot)X^*(\cdot)+v_2^*(\cdot))\leqslant J^1(t,x;\Theta_1(\cdot)X(\cdot)+v_1(\cdot),\Theta_2^*(\cdot)X(\cdot)+v_2^*(\cdot)),\end{array}\\ \noalign{\smallskip}&&\label{Nash-closed2}\begin{array}{ll} \displaystyle J^2(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta_2^*(\cdot)X^*(\cdot)+v_2^*(\cdot))\leqslant J^2(t,x;\Theta^*_1(\cdot)X(\cdot)+v^*_1(\cdot),\Theta_2(\cdot)X(\cdot)+v_2(\cdot)).\end{array} \end{eqnarray} \end{definition} Note that in both \eqref{Nash-closed1} and \eqref{Nash-closed2}, $$X^*(\cdot)=X(\cdot\,;t,x,\Theta_1^*(\cdot),v_1^*(\cdot),\Theta_2^*(\cdot),v_2^*(\cdot)),$$ whereas, in \eqref{Nash-closed1}, $$X(\cdot)=X(\cdot\,;t,x,\Theta_1(\cdot),v_1(\cdot),\Theta_2^*(\cdot),v_2^*(\cdot)),$$ and in \eqref{Nash-closed2}, $$X(\cdot)=X(\cdot\,;t,x,\Theta_1^*(\cdot),v_1^*(\cdot),\Theta_2(\cdot),v_2(\cdot)).$$ Thus, $X(\cdot)$ appeared in \eqref{Nash-closed1} and \eqref{Nash-closed2} are different in general. We emphasize that the closed-loop Nash equilibrium $(\Theta^*_1(\cdot),v^*_1(\cdot);\Theta^*_2(\cdot),v^*_2(\cdot))$ is independent of the initial state $x$. The following result provides some equivalent definitions of closed-loop Nash equilibrium. \begin{proposition}\label{bp-16Apr4-17:30}\sl Let {\rm(G1)--(G2)} hold and let $(\Theta^*_1(\cdot),v_1^*(\cdot); \Theta^*_2(\cdot),v^*_2(\cdot))\in{\cal Q}_1[t,T]\times{\cal U}_1[t,T]\times{\cal Q}_2[t,T]\times{\cal U}_2[t,T]$. Then the following are equivalent: \medskip {\rm(i)} $(\Theta^*_1(\cdot),v_1^*(\cdot);\Theta^*_2(\cdot),v^*_2(\cdot))$ is a closed-loop Nash equilibrium of Problem {\rm(SDG)} on $[t,T]$. \medskip {\rm(ii)} For any $(v_1(\cdot),v_2(\cdot))\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$, $$\begin{array}{ll} \noindent\smallskip\displaystyle J^1(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta_2^*(\cdot)X^*(\cdot)+v_2^*(\cdot))\leqslant J^1(t,x;\Theta_1^*(\cdot)X(\cdot)+v_1(\cdot),\Theta_2^*(\cdot)X(\cdot)+v_2^*(\cdot)),\\ \noindent\smallskip\displaystyle J^2(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta_2^*(\cdot)X^*(\cdot)+v_2^*(\cdot))\leqslant J^2(t,x;\Theta^*_1(\cdot)X(\cdot)+v^*_1(\cdot),\Theta_2^*(\cdot)X(\cdot)+v_2(\cdot)).\end{array}$$ {\rm(iii)} For any $(u_1(\cdot),u_2(\cdot))\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$, \begin{eqnarray} &&\label{16Apr4-Nash-closed1*}\begin{array}{ll} \displaystyle J^1(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta_2^*(\cdot)X^*(\cdot)+v_2^*(\cdot))\leqslant J^1(t,x;u_1(\cdot),\Theta_2^*(\cdot)X(\cdot)+v_2^*(\cdot)),\end{array}\\ \noalign{\smallskip}&&\label{16Apr4-Nash-closed2*}\begin{array}{ll} \displaystyle J^2(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta_2^*(\cdot)X^*(\cdot)+v_2^*(\cdot))\leqslant J^2(t,x;\Theta^*_1(\cdot)X(\cdot)+v^*_1(\cdot),u_2(\cdot)).\end{array} \end{eqnarray} \end{proposition} \it Proof. \rm The proof is similar to that of Proposition \ref{strategy equivalence}. \signed {$\sqr69$} \medskip If we denote \begin{equation}\label{16Apr4-bar u}\bar u_i(\cdot)=\Theta^*_i(\cdot)X^*(\cdot)+v^*_i(\cdot),\qquad i=1,2,\end{equation} then \eqref{16Apr4-Nash-closed1*}--\eqref{16Apr4-Nash-closed2*} become \begin{eqnarray} &&\label{16Apr4-20:00-1}J^1(t,x;\bar u_1(\cdot),\bar u_2(\cdot))\leqslant J^1(t,x;u_1(\cdot),\Theta_2^*(\cdot)X(\cdot)+v_2^*(\cdot)),\\ \noalign{\smallskip}&&\label{16Apr4-20:00-2}J^2(t,x;\bar u_1(\cdot),\bar u_2(\cdot))\leqslant J^2(t,x;\Theta^*_1(\cdot)X(\cdot)+v^*_1(\cdot),u_2(\cdot)). \end{eqnarray} Since in \eqref{16Apr4-20:00-1}, $X(\cdot)$ corresponds to $u_1(\cdot)$ and $(\Theta_2^*(\cdot),v_2^*(\cdot))$, one might not have $$\bar u_2(\cdot)=\Theta^*_2(\cdot)X(\cdot)+v_2^*(\cdot).$$ Likewise, one might not have the following either: $$\bar u_1(\cdot)=\Theta_1^*(\cdot)X(\cdot)+v_1^*(\cdot).$$ Hence, comparing this with \eqref{Nash-open*}, we see that the outcome $(\bar u_1(\cdot),\bar u_2(\cdot))$ of the closed-loop Nash equilibrium $(\Theta^*_1(\cdot),v_1^*(\cdot);\Theta^*_2(\cdot),v_2^*(\cdot))$ defined by \eqref{16Apr4-bar u} is not an open-loop Nash equilibrium of Problem (SDG) for $(t,X^*(t))$ in general. \medskip On the other hand, if $(\Theta_1^*(\cdot),v_1^*(\cdot);\Theta^*_2(\cdot),v_2^*(\cdot))$ is a closed-loop Nash equilibrium of Problem (SDG) on $[t,T]$, we may consider the following state equation (denoting $\Theta^*(\cdot)=(\Theta_1^*(\cdot)^\top\negthinspace ,\Theta_2^*(\cdot)^\top)^\top$) \begin{equation}\label{16Apr4-20:30}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)\negthinspace =\negthinspace \big[(A\negthinspace +\negthinspace B\Theta^*)X\negthinspace +\negthinspace B_1v_1\negthinspace +\negthinspace B_2v_2\negthinspace +\negthinspace b\big]ds\negthinspace +\negthinspace \big[(C\negthinspace +\negthinspace D\Theta^*)X\negthinspace +\negthinspace D_1v_1\negthinspace +\negthinspace D_2v_2\negthinspace +\negthinspace \sigma\big]dW(s),\quad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x, \end{array}\right.\end{equation} with cost functionals \begin{equation}\label{16Apr4-20:40} \widetilde J^i(t,x;v_1(\cdot),v_2(\cdot)) =J^i(t,x;\Theta^*_1(\cdot)X(\cdot)+v_1(\cdot),\Theta^*_2(\cdot)X(\cdot)+v_2(\cdot)),\quad~ i=1,2.\end{equation} Then by (ii) of Proposition \ref{bp-16Apr4-17:30}, $(v_1^*(\cdot),v^*_2(\cdot))$ is an open-loop Nash equilibrium of the corresponding (nonzero-sum differential) problem. Such an observation will be very useful below. \section{Open-Loop Nash Equilibria and FBSDEs} In this section, we discuss the open-loop Nash equilibria for Problem (SDG) in terms of FBSDEs. The main result of this section can be stated as follows. \begin{theorem}\label{Theorem 4.1}\sl Let {\rm(G1)--(G2)} hold and let $(t,x)\in[0,T)\times\mathbb{R}^n$ be given. Then $u^*(\cdot)\equiv(u_1^*(\cdot)^\top\negthinspace ,u_2^*(\cdot)^\top)^\top\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$ is an open-loop Nash equilibrium of Problem {\rm(SDG)} for $(t,x)$ if and only if the following two conditions hold: \medskip {\rm(i)} For $i=1,2$, the adapted solution $(X^*(\cdot),Y_i^*(\cdot),Z_i^*(\cdot))$ to the FBSDE on $[t,T]$ \begin{equation}\label{FBSDEi}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX^*(s)=\big[A(s)X^*(s)+B(s)u^*(s)+b(s)\big]ds+\big[C(s)X^*(s)+D(s)u^*(s)+\sigma(s)\big]dW(s),\\ \noalign{\smallskip}\displaystyle dY_i^*(s)=-\big[A(s)^\top Y_i^*(s)+C(s)^\top Z_i^*(s)+Q^i(s)X^*(s)+S^i(s)^\top u^*(s)+q^i(s)\big]ds+Z_i^*(s)dW(s),\\ \noalign{\smallskip}\displaystyle X^*(t)=x,\qquad Y_i^*(T)=G^iX^*(T)+g^i, \end{array}\right.\end{equation} satisfies the following stationarity condition: \begin{equation}\label{stationary}\begin{array}{rl} \displaystyle B_i(s)^\top Y_i^*(s)+D_i(s)^\top Z_i^*(s)+S_i^i(s)X^*(s)+R^i_i(s)u^*(s)+\rho_i^i(s)=0,\quad\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{array}\end{equation} {\rm(ii)} For $i=1,2$, the following convexity condition holds: \begin{equation}\label{convexity}\begin{array}{ll} \displaystyle\mathbb{E}\Big\{\int_t^T\[\big\langle Q^i(s)X_i(s),X_i(s)\big\rangle+2\big\langle S^i_i(s)X_i(s),u_i(s)\big\rangle +\big\langle R^i_{ii}(s)u_i(s),u_i(s)\big\rangle\]ds\\ \noalign{\smallskip}\displaystyle\qquad\qq\quad+\,\big\langle G^iX_i(T),X_i(T)\big\rangle\Big\}\ges0,\qquad\forall\,u_i(\cdot)\in{\cal U}_i[t,T],\end{array}\end{equation} where $X_i(\cdot)$ is the solution to the following FSDE: \begin{equation}\label{homogeneous}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX_i(s)=\big[A(s)X_i(s)+B_i(s)u_i(s)\big]ds+\big[C(s)X_i(s)+D_i(s)u_i(s)\big]dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X_i(t)=0.\end{array}\right.\end{equation} Or, equivalently, the map $u_i(\cdot)\mapsto J^i(t,x;u(\cdot))$ is convex (for $i=1,2$). \end{theorem} \it Proof. \rm For a given $(t,x)\in[0,T)\times\mathbb{R}^n$ and $u^*(\cdot)\in{\cal U}[t,T]$, let $(X^*(\cdot),Y_1^*(\cdot),Z_1^*(\cdot))$ be the adapted solution to FBSDE \eqref{FBSDEi} with $i=1$. For any $u_1(\cdot)\in{\cal U}_1[t,T]$ and $\varepsilon\in\mathbb{R}$, let $X^\varepsilon(\cdot)$ be the solution to the following perturbed state equation on $[t,T]$: $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX^\varepsilon(s)=\big\{A(s)X^\varepsilon(s)+B_1(s)[u_1^*(s)+\varepsilon u_1(s)]+B_2(s)u_2^*(s)+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\{C(s)X^\varepsilon(s)+D_1(s)[u_1^*(s)+\varepsilon u_1(s)]+D_2(s)u_2^*(s)+\sigma(s)\big\}dW(s),\\ \noalign{\smallskip}\displaystyle X^\varepsilon(t)=x.\end{array}\right.$$ Then denoting $X_1(\cdot)$ the solution of \eqref{homogeneous} with $i=1$, we have $X^\varepsilon(\cdot)=X^*(\cdot)+\varepsilon X_1(\cdot)$ and $$\begin{array}{ll} \displaystyle J^1(t,x;u^*_1(\cdot)+\varepsilon u_1(\cdot),u_2^*(\cdot))-J^1(t,x;u_1^*(\cdot),u_2^*(\cdot))\\ \noalign{\smallskip}\displaystyle=\varepsilon\mathbb{E}\Big\{\big\langle G^1[2X^*(T)+\varepsilon X_1(T)],X_1(T)\rangle+2\langle g^1,X_1(T)\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\quad~+\int_t^T\[\big\langle{\scriptstyle \begin{pmatrix}\scriptstyle Q^1 &\scriptstyle (S^1_1)^\top &\scriptstyle (S^1_2)^\top\\ \scriptstyle S^1_1 &\scriptstyle R^1_{11} &\scriptstyle R^1_{12} \\ \scriptstyle S^1_2 &\scriptstyle R^1_{21} &\scriptstyle R^1_{22} \end{pmatrix} \begin{pmatrix}\scriptstyle 2X^*+\varepsilon X_1 \\ \scriptstyle 2u_1^*+\varepsilon u_1 \\ \scriptstyle 2u_2^* \end{pmatrix}, \begin{pmatrix}\scriptstyle X_1 \\ \scriptstyle u_1 \\ \scriptstyle 0 \end{pmatrix}}\big\rangle +2\big\langle{\scriptstyle\begin{pmatrix}\scriptstyle q^1 \\ \scriptstyle\rho^1_1\end{pmatrix}, \begin{pmatrix}\scriptstyle X_1 \\ \scriptstyle u_1 \end{pmatrix}}\big\rangle\]ds\Big\}\\ \noalign{\smallskip}\displaystyle=2\varepsilon\mathbb{E}\Big\{\big\langle G^1X^*(T)+g^1,X_1(T)\big\rangle+\int_t^T\[\big\langle Q^1X^*+(S^1)^\top u^*+q^1,X_1\big\rangle +\big\langle S^1_1X^*+R^1_1u^*+\rho^1_1,u_1\big\rangle\]ds\Big\}\\ \noalign{\smallskip}\displaystyle~~+\varepsilon^2\mathbb{E}\Big\{\big\langle G^1X_1(T),X_1(T)\big\rangle+\int_t^T\[\big\langle Q^1X_1,X_1\big\rangle +2\big\langle S^1_1X_1,u_1\big\rangle+\big\langle R^1_{11}u_1,u_1\big\rangle\]ds\Big\}. \end{array}$$ On the other hand, applying It\^o's formula to $s\mapsto\langle Y_1^*(s),X_1(s)\rangle$, we obtain $$\begin{array}{ll} \displaystyle\mathbb{E}\Big\{\big\langle G^1X^*(T)+g^1,X_1(T)\big\rangle +\int_t^T\[\big\langle Q^1X^*+(S^1)^\top u^*+q^1,X_1\big\rangle +\big\langle S^1_1X^*+R^1_1u^*+\rho^1_1,u_1\big\rangle\]ds\Big\}\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle-\big[A^\top Y_1^*+C^\top Z_1^*+Q^1X^*+(S^1)^\top u^*+q^1\big],X_1\big\rangle +\big\langle Y_1^*,AX_1+B_1u_1\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\langle Z_1^*,CX_1+D_1u_1\big\rangle+\big\langle Q^1X^*+(S^1)^\top u^*+q^1,X_1\big\rangle +\big\langle S^1_1X^*+R^1_1u^*+\rho^1_1,u_1\big\rangle\Big\}ds\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\big\langle B_1^\top Y_1^*+D_1^\top Z_1^*+S^1_1X^*+R^1_1u^*+\rho^1_1,u_1\big\rangle ds. \end{array}$$ Hence, $$\begin{array}{ll} \displaystyle J^1(t,x;u^*_1(\cdot)+\varepsilon u_1(\cdot),u_2^*(\cdot))-J^1(t,x;u_1^*(\cdot),u_2^*(\cdot))\\ \noalign{\smallskip}\displaystyle=2\varepsilon\mathbb{E}\int_t^T\big\langle B_1^\top Y_1^*+D_1^\top Z_1^*+S^1_1X^*+R^1_1u^*+\rho^1_1,u_1\big\rangle ds\\ \noalign{\smallskip}\displaystyle~~+\varepsilon^2\mathbb{E}\Big\{\big\langle G^1X_1(T),X_1(T)\big\rangle+\int_t^T\[\big\langle Q^1X_1,X_1\big\rangle +2\big\langle S^1_1X_1,u_1\big\rangle+\big\langle R^1_{11}u_1,u_1\big\rangle\]ds\Big\}.\end{array}$$ It follows that $$J^1(t,x;u_1^*(\cdot),u_2^*(\cdot))\leqslant J^1(t,x;u_1^*(\cdot)+\varepsilon u_1(\cdot),u_2^*(\cdot)), \qquad\forall\,u_1(\cdot)\in{\cal U}_1[t,T],~\forall\,\varepsilon\in\mathbb{R},$$ if and only if \eqref{convexity} holds for $i=1$, and \begin{equation}\label{stationary1}B_1^\top Y_1^*+D_1^\top Z_1^*+S^1_1X^*+R^1_1u^*+\rho^1_1=0, \qquad\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{equation} Similarly, $$J^2(t,x;u_1^*(\cdot),u_2^*(\cdot))\leqslant J^2(t,x;u_1^*(\cdot),u_2^*(\cdot)+\varepsilon u_2(\cdot)), \qquad\forall\,u_2(\cdot)\in{\cal U}_2[t,T],~\forall\,\varepsilon\in\mathbb{R},$$ if and only if \eqref{convexity} holds for $i=2$, and \begin{equation}\label{stationary2}B_2^\top Y_2^*+D_2^\top Z_2^*+S^2_2X^*+R^2_2u^*+\rho^2_2=0, \qquad\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{equation} Combining \eqref{stationary1}--\eqref{stationary2}, we obtain \eqref{stationary}. \signed {$\sqr69$} \medskip Note that \eqref{FBSDEi} for $i=1,2$ are two coupled FBSDEs, and these two FBSDEs are coupled through the relation \eqref{stationary}. In fact, from \eqref{stationary}, we see that $$\begin{pmatrix}R^1_{11}&R^1_{12}\\ R^2_{21}&R^2_{22}\end{pmatrix}\begin{pmatrix}u^*_1\\ u^*_2\end{pmatrix}=-\begin{pmatrix}B_1^\top Y_1^*+D_1^\top Z^*_1+S^1_1X^*+\rho^1_1\\ B_2^\top Y_2^*+D_2^\top Z^*_2+S^2_2X^*+\rho^2_2\end{pmatrix}.$$ Thus, say, in the case that the coefficient matrix of $u^*$ is invertible, one has $$\begin{pmatrix}u^*_1\\ u^*_2\end{pmatrix}=-\begin{pmatrix}R^1_{11}&R^1_{12}\\ R^2_{21}&R^2_{22}\end{pmatrix}^{-1}\begin{pmatrix}B_1^\top Y_1^*+D_1^\top Z^*_1+S^1_1X^*+\rho^1_1\\ B_2^\top Y_2^*+D_2^\top Z^*_2+S^2_2X^*+\rho^2_2\end{pmatrix}.$$ Plugging the above into \eqref{FBSDEi}, we see the coupling between the two coupled FBSDEs (with $i=1,2$). \medskip To conclude this section, let us write FBSDE \eqref{FBSDEi} and stationarity condition \eqref{stationary} more compactly. For this, we introduce the following: $$\begin{array}{ll} \noindent\medskip\displaystyle{\bf A}(\cdot)=\begin{pmatrix}A(\cdot)&0 \\ 0&A(\cdot)\end{pmatrix}, \quad{\bf B}(\cdot)=\begin{pmatrix}B(\cdot)&0 \\ 0&B(\cdot)\end{pmatrix} \equiv\begin{pmatrix}B_1(\cdot)&B_2(\cdot)&0&0\\ 0&0&B_1(\cdot)&B_2(\cdot)\end{pmatrix},\\ \noindent\medskip\displaystyle{\bf C}(\cdot)=\begin{pmatrix}C(\cdot)&0 \\ 0&C(\cdot)\end{pmatrix}, \quad{\bf D}(\cdot)=\begin{pmatrix}D(\cdot)&0 \\ 0&D(\cdot)\end{pmatrix} \equiv\begin{pmatrix}D_1(\cdot)&D_2(\cdot)&0&0\\ 0&0&D_1(\cdot)&D_2(\cdot)\end{pmatrix},\\ \noindent\medskip\displaystyle{\bf Q}(\cdot)=\begin{pmatrix}Q^1(\cdot)&0 \\ 0&Q^2(\cdot)\end{pmatrix}, \quad{\bf S}(\cdot)=\begin{pmatrix}S^1(\cdot)&0 \\ 0&S^2(\cdot)\end{pmatrix}, \quad{\bf R}(\cdot)=\begin{pmatrix}R^1(\cdot)&0 \\ 0&R^2(\cdot)\end{pmatrix},\\ \displaystyle{\bf q}(\cdot)=\begin{pmatrix} q^1(\cdot) \\ q^2(\cdot) \end{pmatrix},\quad {\rho\3n\rho\3n\rho\3n\rho\3n\rho}(\cdot)=\begin{pmatrix}\rho^1(\cdot) \\ \rho^2(\cdot)\end{pmatrix},\quad {\bf G}=\begin{pmatrix}G^1&0 \\ 0&G^2\end{pmatrix},\quad {\bf g}=\begin{pmatrix} g^1 \\ g^2 \end{pmatrix}.\end{array}$$ Then $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle{\bf A}(\cdot)\in L^1(0,T;\mathbb{R}^{2n\times2n}),\quad{\bf B}(\cdot)\in L^2(0,T;\mathbb{R}^{2n\times2m}),\\ \noalign{\smallskip}\displaystyle{\bf C}(\cdot)\in L^2(0,T;\mathbb{R}^{2n\times2n}),\quad{\bf D}(\cdot)\in L^\infty(0,T;\mathbb{R}^{2n\times2m}),\\ \noalign{\smallskip}\displaystyle{\bf Q}(\cdot)\in L^1(0,T;\mathbb{S}^{2n}),\quad{\bf S}(\cdot)\in L^2(0,T;\mathbb{R}^{2m\times2n}), \quad{\bf R}(\cdot)\in L^\infty(0,T;\mathbb{S}^{2m}),\\ \noalign{\smallskip}\displaystyle{\bf q}(\cdot)\in L^2_\mathbb{F}(\Omega;L^1(0,T;\mathbb{R}^{2n})),\quad{\rho\3n\rho\3n\rho\3n\rho\3n\rho}(\cdot)\in L_\mathbb{F}^2(0,T;\mathbb{R}^{2m}), \quad{\bf G}\in\mathbb{S}^{2n},\quad{\bf g}\in L^2_{{\cal F}_T}(\Omega;\mathbb{R}^{2n}).\end{array}\right.$$ Further, let $${\bf J}=\begin{pmatrix}I_{m_1}&0 \\ 0&0 \\ 0&0 \\ 0&I_{m_2}\end{pmatrix} \equiv\begin{pmatrix}I_{m_1} & 0_{m_1\times m_2} \\ 0_{m_2\times m_1} & 0_{m_2\times m_2} \\ 0_{m_1\times m_1} & 0_{m_1\times m_2} \\ 0_{m_2\times m_1} & I_{m_2} \end{pmatrix}\in\mathbb{R}^{2m\times m}, \qquad{\bf I}_k=\begin{pmatrix}I_k \\ I_k\end{pmatrix}\in\mathbb{R}^{2k\times k}.$$ Clearly, one has $$\begin{array}{ll} \noindent\medskip\displaystyle{\bf B}(\cdot){\bf J} \equiv\begin{pmatrix}B_1(\cdot)&B_2(\cdot)&0&0 \\ 0&0&B_1(\cdot)&B_2(\cdot)\end{pmatrix} \begin{pmatrix}I_{m_1}&0 \\ 0&0 \\ 0&0 \\ 0&I_{m_2}\end{pmatrix} =\begin{pmatrix}B_1(\cdot)&0 \\ 0&B_2(\cdot)\end{pmatrix},\\ \noindent\medskip\displaystyle{\bf D}(\cdot){\bf J} \equiv\begin{pmatrix}D_1(\cdot)&D_2(\cdot)&0&0 \\ 0&0&D_1(\cdot)&D_2(\cdot)\end{pmatrix} \begin{pmatrix}I_{m_1}&0 \\ 0&0 \\ 0&0 \\ 0&I_{m_2}\end{pmatrix} =\begin{pmatrix}D_1(\cdot)&0 \\ 0&D_2(\cdot)\end{pmatrix},\\ \noindent\medskip\displaystyle{\bf J}^\top{\bf S}(\cdot) \equiv\begin{pmatrix}I_{m_1}&0&0&0 \\ 0&0&0&I_{m_2}\end{pmatrix} \begin{pmatrix}S^1_1(\cdot) & 0 \\ S^1_2(\cdot) & 0\\ 0 & S^2_1(\cdot) \\ 0 & S^2_2(\cdot)\end{pmatrix} =\begin{pmatrix}S^1_1(\cdot)&0 \\ 0&S^2_2(\cdot)\end{pmatrix},\\ \noindent\medskip\displaystyle{\bf J}^\top{\bf R}(\cdot) \equiv\begin{pmatrix}I_{m_1}&0&0&0 \\ 0&0&0&I_{m_2}\end{pmatrix} \begin{pmatrix}R^1_1(\cdot)&0 \\ R^1_2(\cdot)&0\\ 0&R^2_1(\cdot) \\ 0&R^2_2(\cdot)\end{pmatrix} =\begin{pmatrix}R^1_1(\cdot)&0 \\ 0&R^2_2(\cdot)\end{pmatrix},\\ \displaystyle{\bf J}^\top{\rho\3n\rho\3n\rho\3n\rho\3n\rho}(\cdot) \equiv\begin{pmatrix}I_{m_1}&0&0&0 \\ 0&0&0&I_{m_2}\end{pmatrix} \begin{pmatrix}\rho^1_1(\cdot) \\ \rho^1_2(\cdot)\\ \rho^2_1(\cdot) \\ \rho^2_2(\cdot)\end{pmatrix} =\begin{pmatrix}\rho^1_1(\cdot) \\ \rho^2_2(\cdot)\end{pmatrix}.\end{array}$$ With the above notation, FBSDE \eqref{FBSDEi} can be written as (suppressing $s$ and dropping $*$) \begin{equation}\label{FBSDE**}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX=\big(AX+Bu+b\big)ds+\big(CX+Du+\sigma\big)dW,\\ \noalign{\smallskip}\displaystyle d{\bf Y}=-\big({\bf A}^\top{\bf Y}+{\bf C}^\top{\bf Z}+{\bf Q}{\bf I}_nX+{\bf S}^\top{\bf I}_mu+{\bf q}\big)ds+{\bf Z} dW,\\ \noalign{\smallskip}\displaystyle X(t)=x,\qquad {\bf Y}(T)={\bf G}{\bf I}_nX(T)+{\bf g},\end{array}\right.\end{equation} where $${\bf Y}(\cdot)=\begin{pmatrix}Y_1(\cdot)\\ Y_2(\cdot)\end{pmatrix},\qquad{\bf Z}(\cdot)=\begin{pmatrix}Z_1(\cdot)\\ Z_2(\cdot)\end{pmatrix}.$$ and the stationarity condition \eqref{stationary} can be written as \begin{equation}\label{stationary**}{\bf J}^\top\big({\bf B}^\top{\bf Y}+{\bf D}^\top{\bf Z}+{\bf S}{\bf I}_nX+{\bf R}{\bf I}_m u+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)=0,\qquad\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{equation} Keep in mind that \eqref{FBSDE**} is a coupled FBSDE with the coupling given through \eqref{stationary**}. \section{Closed-Loop Nash Equilibria and Riccati Equations} We now look at closed-loop Nash equilibria for Problem (SDG). Again, for simplicity of notation, we will suppress the time variable $s$ as long as no confusion arises. First, we present the following result which is a consequence of Theorem \ref{Theorem 4.1}. \begin{proposition}\label{prop 5.1}\sl Let {\rm(G1)--(G2)} hold. Suppose that $(\Theta^*_1(\cdot),v_1^*(\cdot);\Theta^*_2(\cdot),v^*_2(\cdot)) \in{\cal Q}_1[t,T]\times{\cal U}_1[t,T]\times{\cal Q}_2[t,T]\times{\cal U}_2[t,T]$ is a closed-loop Nash equilibrium of Problem {\rm(SDG)} on $[t,T]$. Denote $\Theta^*(\cdot)\negthinspace \equiv\negthinspace (\Theta^*_1(\cdot)\negthinspace ^\top\negthinspace ,\Theta^*_2(\cdot)\negthinspace ^\top)\negthinspace ^\top$ and let $\mathbb{X}(\cdot)$ be the solution to the $\mathbb{R}^{n\times n}$-valued SDE \begin{equation}\label{16Apr13-dbX}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\mathbb{X}=(A+B\Theta^*)\mathbb{X} ds+(C+D\Theta^*)\mathbb{X} dW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle \mathbb{X}(t)=I. \end{array}\right.\end{equation} Then for $i=1,2$, the adapted solution $(\mathbb{Y}_i(\cdot),\mathbb{Z}_i(\cdot))$ to the $\mathbb{R}^{n\times n}$-valued BSDE \begin{equation}\label{16Apr13-dbYi}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\mathbb{Y}_i=-\Big\{(A+B\Theta^*)^\top\mathbb{Y}_i+(C+D\Theta^*)^\top\mathbb{Z}_i\\ \noalign{\smallskip}\displaystyle\qquad\qq~ +\big[Q^i+(\Theta^*)^\top S^i+(S^i)^\top\Theta^*+(\Theta^*)^\top R^i\Theta^*\big]\mathbb{X}\Big\}ds +\mathbb{Z}_idW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle \mathbb{Y}_i(T)=G^i\mathbb{X}(T),\end{array}\right.\end{equation} satisfies \begin{equation}\label{16Apr13-yushu}B_i^\top\mathbb{Y}_i+D_i^\top\mathbb{Z}_i+(S_i^i+R_i^i\Theta^*)\mathbb{X}=0,\quad~\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{equation} \end{proposition} \it Proof. \rm Let us consider state equation \eqref{16Apr4-20:30} with the cost functionals defined by \eqref{16Apr4-20:40}. Denoting $v(\cdot)=(v_1(\cdot)^\top,v_2(\cdot)^\top)^\top$, by an argument similar to \eqref{16Apr7-22:00}, we have: $$\begin{array}{ll} \displaystyle\widetilde J^i(t,x;v(\cdot))\equiv J^i(t,x;\Theta^*(\cdot)X(\cdot)+v(\cdot))\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\Big\{\langle G^iX(T),X(T)\rangle+2\langle g^i,X(T)\rangle\\ \noalign{\smallskip}\displaystyle\quad+\negthinspace \int_t^T\negthinspace \[\langle{\scriptstyle \begin{pmatrix}\scriptstyle Q^i+(\Theta^*)^\top\negthinspace S^i+(S^i)^\top\negthinspace \Theta^*+(\Theta^*)^\top\negthinspace R^i\Theta^* &\negthinspace \scriptstyle (S^i)^\top\negthinspace +(\Theta^*)^\top\negthinspace R^i \\ \scriptstyle S^i+R^i\Theta^* &\negthinspace \scriptstyle R^i \end{pmatrix} \begin{pmatrix}\scriptstyle X \\ \scriptstyle v\end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle v\end{pmatrix}}\rangle\negthinspace +\1n2\langle{\scriptstyle\begin{pmatrix}\scriptstyle q^i+(\Theta^*)^\top\negthinspace \rho^i \\ \scriptstyle\rho^i\end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle v \end{pmatrix}}\rangle\]ds\Big\}.\end{array}$$ We know by (ii) of Proposition \ref{bp-16Apr4-17:30} that $v^*(\cdot)\equiv(v_1^*(\cdot)^\top,v^*_2(\cdot)^\top)^\top$ is an open-loop Nash equilibrium for the problem with the state equation \eqref{16Apr4-20:30} and with the cost functionals $\widetilde J^i(t,x;v(\cdot))$ for any initial pair $(t,x)$. Thus, according to Theorem \ref{Theorem 4.1}, we have for $i=1,2$, \begin{equation}\label{16Apr13-14:00}B_i^\top Y_i^*+D_i^\top Z_i^*+(S_i^i+R_i^i\Theta^*)X^*+R^i_iv^*+\rho_i^i=0, \quad~\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{equation} with $X^*(\cdot)$ being the solution to the closed-loop system: \begin{equation}\label{16Apr13-14:25}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX^*=\big[(A+B\Theta^*)X^*+Bv^*+b\big]ds+\big[(C+D\Theta^*)X^*+Dv^*+\sigma\big]dW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle X^*(t)=x, \end{array}\right.\end{equation} and $(Y_i^*(\cdot),Z_i^*(\cdot))$ being the adapted solution to the following BSDE: \begin{equation}\label{16Apr13-15:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dY_i^*=-\Big\{(A\negthinspace +\1nB\Theta^*)^\top\1nY_i^*+(C\negthinspace +\1nD\Theta^*)^\top\negthinspace Z_i^* +\big[Q^i\negthinspace +\negthinspace (\Theta^*)^\top\negthinspace S^i\negthinspace +\negthinspace (S^i)^\top\negthinspace \Theta^*\negthinspace +\negthinspace (\Theta^*)^\top\negthinspace R^i\Theta^*\big]X^*\\ \noalign{\smallskip}\displaystyle\qquad\qq~+(S^i+R^i\Theta^*)^\top v^*+q^i+(\Theta^*)^\top\rho^i\Big\}ds+Z_i^*dW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle Y_i^*(T)=G^iX^*(T)+g^i.\end{array}\right.\end{equation} Since $(\Theta^*(\cdot),v^*(\cdot))$ is independent of $x$ and \eqref{16Apr13-14:00}--\eqref{16Apr13-15:00} hold for all $x\in\mathbb{R}^n$, by subtracting solutions corresponding to $x$ and $0$, the latter from the former, we see that for any $x\in\mathbb{R}^n$, the adapted solution $(X(\cdot),Y_i(\cdot),Z_i(\cdot))$ $(i=1,2)$ to the following FBSDE: $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX=(A+B\Theta^*)Xds+(C+D\Theta^*)XdW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle dY_i=-\Big\{(A+B\Theta^*)^\top Y_i+(C+D\Theta^*)^\top Z_i\\ \noalign{\smallskip}\displaystyle\qquad\qq~ +\big[Q^i+(\Theta^*)^\top S^i+(S^i)^\top\Theta^*+(\Theta^*)^\top R^i\Theta^*\big]X\Big\}ds+Z_idW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x,\qquad Y_i(T)=G^iX(T),\end{array}\right.$$ satisfies $$B_i^\top Y_i+D_i^\top Z_i+(S_i^i+R_i^i\Theta^*)X=0,\quad~\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}$$ The desired result then follows easily. \signed {$\sqr69$} \medskip Now we are ready to present the main result of this section, which characterizes the closed-loop Nash equilibrium of Problem (SDG). \begin{theorem}\label{Theorem 5.2}\sl Let {\rm(G1)--(G2)} hold. Then $(\Theta^*(\cdot),v^*(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$ is a closed-loop Nash equilibrium of Problem {\rm(SDG)} on $[t,T]$ if and only if the following hold: \medskip {\rm(i)} For $i=1,2$, the solution $P_i(\cdot)\in C([t,T];\mathbb{S}^n)$ to the Lyapunov type equation \begin{equation}\label{Riccati-i}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_i+P_iA+A^\top P_i+C^\top P_iC+Q^i+(\Theta^*)^\top(R^i+D^\top P_iD)\Theta^*\\ \noalign{\smallskip}\displaystyle~~~+\big[P_iB+C^\top P_iD+(S^i)^\top\big]\Theta^*+(\Theta^*)^\top\big[B^\top P_i+D^\top P_iC+S^i\big]=0, \quad~\hbox{\rm a.e.}~s\in[t,T],\\ \noalign{\smallskip}\displaystyle P_i(T)=G^i,\end{array}\right.\end{equation} satisfies the following two conditions: \begin{eqnarray} &\label{R+DPD>0}R_{ii}^i+D_i^\top P_iD_i\ges0,\quad~\hbox{\rm a.e.}~s\in[t,T],&\\ &\label{stationary-closed-i}B_i^\top P_i+D_i^\top P_iC+S_i^i+(R_i^i+D_i^\top P_iD)\Theta^*=0,\quad~\hbox{\rm a.e.}~s\in[t,T].& \end{eqnarray} \medskip {\rm(ii)} For $i=1,2$, the adapted solution $(\eta_i(\cdot),\zeta_i(\cdot))$ to the BSDE \begin{equation}\label{BSDE-closed-i}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\eta_i=-\Big\{A^\top\eta_i+C^\top\zeta_i +(\Theta^*)^\top\big[B^\top\eta_i+D^\top\zeta_i+D^\top\negthinspace P_i\sigma+\rho^i+(R^i+D^\top\negthinspace P_iD)v^*\big]\\ \noalign{\smallskip}\displaystyle\qquad\quad~~+\big[P_iB+C^\top P_iD+(S^i)^\top\big]v^*+C^\top P_i\sigma+P_ib+q^i\Big\}ds+\zeta_i dW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle \eta_i(T)=g^i, \end{array}\right.\end{equation} satisfies \begin{equation}\label{eta-i}B_i^\top\eta_i+D_i^\top\zeta_i+D_i^\top P_i\sigma+\rho_i^i+(R_i^i+D_i^\top P_iD)v^*=0, \quad~\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\end{equation} \end{theorem} \it Proof. \rm We first prove the necessity. Suppose that $(\Theta^*(\cdot),v^*(\cdot))$ is a closed-loop Nash equilibrium of Problem (SDG) on $[t,T]$, where $\Theta^*(\cdot)\equiv(\Theta^*_1(\cdot)^\top,\Theta^*_2(\cdot)^\top)^\top$ and $v^*(\cdot)\equiv(v_1^*(\cdot)^\top,v_2^*(\cdot)^\top)^\top$. Let $\mathbb{X}(\cdot)$ and $\mathbb{Y}_i(\cdot)$ $(i=1,2)$ be the solutions of \eqref{16Apr13-dbX} and \eqref{16Apr13-dbYi}, respectively. Consider the following linear ordinary differential equation (ODE, for short) which is equivalent to \eqref{Riccati-i}: \begin{equation}\label{Riccati-i*}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_i+P_i(A+B\Theta^*)+(A+B\Theta^*)^\top P_i+(C+D\Theta^*)^\top P_i(C+D\Theta^*)\\ \noalign{\smallskip}\displaystyle~~~+Q^i+(\Theta^*)^\top S^i+(S^i)^\top\Theta^*+(\Theta^*)^\top R^i\Theta^*=0,\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle P_i(T)=G^i.\end{array}\right.\end{equation} Such an equation admits a unique solution $P_i(\cdot)\in C([t,T];\mathbb{S}^n)$. By It\^o's formula, we have $$\begin{array}{lll} \displaystyle d(P_i\mathbb{X})\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \dot P_i\mathbb{X} ds+P_i(A+B\Theta^*)\mathbb{X} ds+P_i(C+D\Theta^*)\mathbb{X} dW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace -\,\big\{(A+B\Theta^*)^\top P_i\mathbb{X} +(C+D\Theta^*)^\top P_i(C+D\Theta^*)\mathbb{X}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \quad~+\big[Q^i+(\Theta^*)^\top S^i+(S^i)^\top\Theta^*+(\Theta^*)^\top R^i\Theta^*\big]\mathbb{X}\big\}ds +P_i(C+D\Theta^*)\mathbb{X} dW. \end{array}$$ Comparing the above with \eqref{16Apr13-dbYi}, by the uniqueness of adapted solutions to BSDEs, one has $$\mathbb{Y}_i=P_i\mathbb{X},\quad \mathbb{Z}_i=P_i(C+D\Theta^*)\mathbb{X};\qquad i=1,2.$$ From \eqref{16Apr13-dbX}, we see that the process $\mathbb{X}(\cdot)$ is invertible almost surely. Then, the above together with \eqref{16Apr13-yushu} leads to \eqref{stationary-closed-i}. Now let $X^*(\cdot)$ be the solution to \eqref{16Apr13-14:25}, and for $i=1,2$, let $(Y_i^*(\cdot),Z_i^*(\cdot))$ be the adapted solution to \eqref{16Apr13-15:00}. Define \begin{equation}\label{etaz}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\eta_i=Y_i^*-P_iX^*,\\ \noalign{\smallskip}\zeta_i=Z_i^*-P_i(C+D\Theta^*)X^*-P_i(Dv^*+\sigma).\end{array}\right.\end{equation} Then $\eta_i(T)=g^i$, and $$\begin{array}{lll} \displaystyle d\eta_i \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace dY_i^*-\dot P_iX^*ds-P_i dX^*\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace -\,\Big\{(A+B\Theta^*)^\top Y_i^*+(C+D\Theta^*)^\top Z_i^* +(S^i+R^i\Theta^*)^\top v^*+P_i(Bv^*+b)+q^i+(\Theta^*)^\top\rho^i\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \quad~ +\big[\dot P_i+P_i(A+B\Theta^*)+Q^i+(\Theta^*)^\top S^i+(S^i)^\top\Theta^*+(\Theta^*)^\top R^i\Theta^*\big]X^*\Big\}ds\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace +\,\Big\{Z_i^*-P_i\big[(C+D\Theta^*)X^*+Dv^*+\sigma\big]\Big\}dW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace -\,\Big\{(A+B\Theta^*)^\top Y_i^*+(C+D\Theta^*)^\top Z_i^* +(S^i+R^i\Theta^*)^\top v^*+P_i(Bv^*+b)+q^i+(\Theta^*)^\top\rho^i\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \quad~ -(A+B\Theta^*)^\top P_iX^*-(C+D\Theta^*)^\top P_i(C+D\Theta^*)X^*\Big\}ds+\zeta_idW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace -\,\Big\{(A+B\Theta^*)^\top\eta_i+(C+D\Theta^*)^\top\zeta_i+(C+D\Theta^*)^\top P_i(Dv^*+\sigma)\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \quad~+(S^i+R^i\Theta^*)^\top v^*+P_i(Bv^*+b)+q^i+(\Theta^*)^\top\rho^i\Big\}ds+\zeta_idW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace -\,\Big\{A^\top\eta_i+C^\top\zeta_i +(\Theta^*)^\top\big[B^\top\eta_i+D^\top\zeta_i+D^\top P_i\sigma+\rho^i+(R^i+D^\top P_iD)v^*\big]\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \quad~+\big[P_iB+C^\top P_iD+(S^i)^\top\big]v^*+C^\top P_i\sigma+P_ib+q^i\Big\}ds+\zeta_idW.\end{array}$$ Thus, $(\eta_i,\zeta_i)$ is the adapted solution to BSDE \eqref{BSDE-closed-i}. Next, from the proof of Proposition \ref{prop 5.1} we know that \eqref{16Apr13-14:00} holds. Thus (noting \eqref{stationary-closed-i} and \eqref{etaz}), $$\begin{array}{lll} \ds0 \negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle B_i^\top Y_i^*+D_i^\top Z_i^*+(S_i^i+R_i^i\Theta^*)X^*+R^i_iv^*+\rho_i^i\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle B_i^\top\eta_i\negthinspace +\negthinspace D_i^\top\zeta_i\negthinspace +\negthinspace D_i^\top P_i\sigma\negthinspace +\negthinspace \rho_i^i\negthinspace +\negthinspace (R_i^i\negthinspace +\negthinspace D_i^\top P_iD)v^*\negthinspace +\negthinspace \big[B_i^\top P_i\negthinspace +\negthinspace D_i^\top P_iC\negthinspace +\negthinspace S_i^i\negthinspace +\negthinspace (R_i^i\negthinspace +\negthinspace D_i^\top P_iD)\Theta^*\big]X^*\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle B_i^\top\eta_i+D_i^\top\zeta_i+D_i^\top P_i\sigma+\rho_i^i+(R_i^i+D_i^\top P_iD)v^*,\end{array}$$ which is \eqref{eta-i}. The proof of \eqref{R+DPD>0} will be included in the proof of sufficiency. \medskip To prove the sufficiency, we take any $v(\cdot)=(v_1(\cdot)^\top,v_2(\cdot)^\top)^\top\in{\cal U}_1[t,T]\times{\cal U}_2[t,T]$. Denote $w=(v_1^\top$, $(v_2^*)^\top)^\top$, and let $$X(\cdot)=X(\cdot\,;t,x,\Theta_1^*(\cdot),v_1(\cdot),\Theta_2^*(\cdot),v_2^*(\cdot))$$ be the state process corresponding to $(t,x)$ and $(\Theta_1^*(\cdot),v_1(\cdot),\Theta_2^*(\cdot),v_2^*(\cdot))$. By It\^{o}'s formula, we have $$\begin{array}{lll} \displaystyle\mathbb{E}\[\big\langle G^1X(T),X(T)\big\rangle+2\big\langle g^1,X(T)\big\rangle\]-\mathbb{E}\[\langle P_1(t)x,x\rangle+2\langle \eta_1(t),x\rangle\]\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle\dot P_1X,X\big\rangle+2\langle P_1X,(A+B\Theta^*)X+Bw+b\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\langle P_1\big[(C+D\Theta^*)X+Dw+\sigma\big],(C+D\Theta^*)X+Dw+\sigma\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~-2\big\langle A^\top\eta_1+C^\top\zeta_1 +(\Theta^*)^\top\big[B^\top\eta_1+D^\top\zeta_1+D^\top\negthinspace P_1\sigma+\rho^1+(R^1+D^\top\negthinspace P_1D)v^*\big],X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~-2\big\langle\big[P_1B+C^\top P_1D+(S^1)^\top\big]v^*+C^\top P_1\sigma+P_1b+q^1,X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\langle\eta_1,(A+B\Theta^*)X+Bw+b\rangle+2\langle\zeta_1,(C+D\Theta^*)X+Dw+\sigma\rangle\Big\}ds\end{array}$$ $$\begin{array}{ll} \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle\big[\dot P_1+P_1(A+B\Theta^*)+(A+B\Theta^*)^\top P_1 +(C+D\Theta^*)^\top P_1(C+D\Theta^*)\big]X,X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\langle P_1X,Bw+b\rangle+2\langle P_1(C+D\Theta^*)X,Dw+\sigma\rangle+\langle P_1(Dw+\sigma),Dw+\sigma\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~-2\big\langle(\Theta^*)^\top\big[D^\top\negthinspace P_1\sigma+\rho^1+(R^1+D^\top\negthinspace P_1D)v^*\big],X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~-2\big\langle\big[P_1B+C^\top P_1D+(S^1)^\top\big]v^*+C^\top P_1\sigma+P_1b+q^1,X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\langle\eta_1,Bw+b\rangle+2\langle\zeta_1,Dw+\sigma\rangle\Big\}ds\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle\big[\dot P_1+P_1(A+B\Theta^*)+(A+B\Theta^*)^\top P_1 +(C+D\Theta^*)^\top P_1(C+D\Theta^*)\big]X,X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle(P_1B+C^\top P_1D)w-\big[P_1B+C^\top P_1D+(S^1)^\top\big]v^*-q^1,X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle D^\top P_1Dw-(R^1+D^\top P_1D)v^*-\rho^1,\Theta^*X\big\rangle+\big\langle D^\top P_1Dw,w\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle B^\top\eta_1+D^\top\zeta_1+D^\top P_1\sigma,w\big\rangle +\langle P_1\sigma,\sigma\rangle+2\langle\eta_1,b\rangle+2\langle\zeta_1,\sigma\rangle\Big\}ds.\end{array}$$ On the other hand, we have $$\begin{array}{ll} \displaystyle J^1(t,x;\Theta^*X(\cdot)+w(\cdot))-\mathbb{E}\[\big\langle G^1X(T),X(T)\big\rangle+2\big\langle g^1,X(T)\big\rangle\]\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\[\big\langle\scriptstyle{ \begin{pmatrix}\scriptstyle Q^1+(\Theta^*)^\top S^1+(S^1)^\top\Theta^*+(\Theta^*)^\top R^1\Theta^* &\scriptstyle (S^1)^\top+(\Theta^*)^\top R^1 \\ \scriptstyle S^1+R^1\Theta^* & \scriptstyle R^1 \end{pmatrix} \begin{pmatrix}\scriptstyle X \\ \scriptstyle w\end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle w\end{pmatrix}}\big\rangle+2\big\langle{\scriptstyle \begin{pmatrix}\scriptstyle q^1+(\Theta^*)^\top\rho^1 \\ \scriptstyle\rho^1\end{pmatrix}, \begin{pmatrix}\scriptstyle X \\ \scriptstyle w \end{pmatrix}}\big\rangle\]ds\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle\big[Q^1+(\Theta^*)^\top S^1+(S^1)^\top\Theta^* +(\Theta^*)^\top R^1\Theta^*\big]X,X\big\rangle+2\big\langle(S^1)^\top w+q^1,X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\quad~+2\big\langle R^1w+\rho^1,\Theta^*X\big\rangle+\big\langle R^1w,w\big\rangle+2\big\langle\rho^1,w\big\rangle\Big\}ds.\end{array}$$ Combining the above two equations, together with equation \eqref{Riccati-i*} (which is equivalent to \eqref{Riccati-i}) and conditions \eqref{stationary-closed-i} and \eqref{eta-i}, one obtains $$\begin{array}{ll} \displaystyle J^1(t,x;\Theta^*X(\cdot)+w(\cdot))-\mathbb{E}\[\langle P_1(t)x,x\rangle+2\langle \eta_1(t),x\rangle\]\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{2\big\langle\big[P_1B+C^\top P_1D+(S^1)^\top\big](w-v^*),X\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle(R^1+D^\top P_1D)(w-v^*),\Theta^*X\big\rangle+\big\langle(R^1+D^\top P_1D)w,w\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle B^\top\eta_1+D^\top\zeta_1+D^\top P_1\sigma+\rho^1,w\big\rangle +\langle P_1\sigma,\sigma\rangle+2\langle\eta_1,b\rangle+2\langle\zeta_1,\sigma\rangle\Big\}ds\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{2\big\langle\big[P_1B_1+C^\top P_1D_1+(S^1_1)^\top\big](v_1-v_1^*),X\big\rangle +2\big\langle(R^1_1+D_1^\top P_1D)^\top(v_1-v_1^*),\Theta^*X\big\rangle \\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\langle(R^1_{11}+D_1^\top P_1D_1)v_1,v_1\big\rangle +2\big\langle(R^1_{12}+D_1^\top P_1D_2)v_2^*,v_1\big\rangle+\big\langle(R^1_{22}+D_2^\top P_1D_2)v_2^*,v_2^*\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle B_1^\top\eta_1+D_1^\top\zeta_1+D_1^\top P_1\sigma+\rho_1^1,v_1\big\rangle +2\big\langle B_2^\top\eta_1+D_2^\top\zeta_1+D_2^\top P_1\sigma+\rho_2^1,v_2^*\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\langle P_1\sigma,\sigma\rangle+2\langle\eta_1,b\rangle+2\langle\zeta_1,\sigma\rangle\Big\}ds\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle(R^1_{11}+D_1^\top P_1D_1)v_1,v_1\big\rangle -2\big\langle(R^1_{11}+D_1^\top P_1D_1)v_1^*,v_1\big\rangle+\big\langle(R^1_{22}+D_2^\top P_1D_2)v_2^*,v_2^*\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+2\big\langle B_2^\top\eta_1+D_2^\top\zeta_1+D_2^\top P_1\sigma+\rho_2^1,v_2^*\big\rangle +\langle P_1\sigma,\sigma\rangle+2\langle\eta_1,b\rangle+2\langle\zeta_1,\sigma\rangle\Big\}ds\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\int_t^T\Big\{\big\langle(R^1_{11}+D_1^\top P_1D_1)(v_1-v_1^*),v_1-v_1^*\big\rangle -\big\langle(R^1_{11}+D_1^\top P_1D_1)v_1^*,v_1^*\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\langle(R^1_{22}+D_2^\top P_1D_2)v_2^*,v_2^*\big\rangle +2\big\langle B_2^\top\eta_1+D_2^\top\zeta_1+D_2^\top P_1\sigma+\rho_2^1,v_2^*\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\langle P_1\sigma,\sigma\rangle+2\langle\eta_1,b\rangle+2\langle\zeta_1,\sigma\rangle\Big\}ds.\end{array}$$ Consequently, $$ J^1(t,x;\Theta^*X(\cdot)+w(\cdot))-J^1(t,x;\Theta^*X^*(\cdot)+v^*(\cdot)) =\mathbb{E}\int_t^T\big\langle(R^1_{11}+D_1^\top P_1D_1)(v_1-v_1^*),v_1-v_1^*\big\rangle ds.$$ It follows that for any $ v_1(\cdot)\in{\cal U}_1[t,T]$, $$ J^1(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta^*_2(\cdot)X(\cdot)+v_2^*(\cdot)) \leqslant J^1(t,x;\Theta_1^*(\cdot)X(\cdot)+v_1(\cdot),\Theta^*_2(\cdot)X(\cdot)+v_2^*(\cdot)),$$ if and only if $$R^1_{11}+D_1^\top P_1D_1\ges0,\qquad\hbox{\rm a.e.}~s\in[t,T].$$ Similarly, for any $ v_2(\cdot)\in{\cal U}_2[t,T]$, $$ J^2(t,x;\Theta_1^*(\cdot)X^*(\cdot)+v_1^*(\cdot),\Theta^*_2(\cdot)X(\cdot)+v_2^*(\cdot)) \leqslant J^2(t,x;\Theta_1^*(\cdot)X(\cdot)+v_1^*(\cdot),\Theta^*_2(\cdot)X(\cdot)+v_2(\cdot)),$$ if and only if $$R^2_{22}+D_2^\top P_2D_2\ges0,\qquad\hbox{\rm a.e.}~s\in[t,T].$$ This proves the sufficiency, as well as the necessity of \eqref{R+DPD>0}. \signed {$\sqr69$} \medskip Note that condition \eqref{stationary-closed-i} is equivalent to the following: $$\begin{pmatrix}B_1^\top P_1+D_1^\top P_1C+S_1^1 \\ B_2^\top P_2+D_2^\top P_2C+S_2^2\end{pmatrix} +\begin{pmatrix}R_1^1+D_1^\top P_1D \\ R_2^2+D_2^\top P_2D\end{pmatrix}\Theta^*=0.$$ Therefore, \begin{equation}\label{Th*}\Theta^*=-\begin{pmatrix}R_1^1+D_1^\top P_1D \\ R_2^2+D_2^\top P_2D\end{pmatrix}^{-1} \begin{pmatrix}B_1^\top P_1+D_1^\top P_1C+S_1^1 \\ B_2^\top P_2+D_2^\top P_2C+S_2^2\end{pmatrix},\end{equation} provided the involved inverse (which is an $\mathbb{R}^{m\times m}$-valued function) exists. By plugging such a $\Theta^*(\cdot)$ into \eqref{Riccati-i}, we see that the equations for $P_1(\cdot)$ and $P_2(\cdot)$ are coupled, symmetric, and of Riccati type. \medskip Now, let us try to rewrite the Riccati equation in a more compact form. Note that (recalling the notation we introduced in the previous section) $$\begin{array}{lll} \ds0 \negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \begin{pmatrix}B_1^\top P_1+D_1^\top P_1C+S_1^1 \\ B_2^\top P_2+D_2^\top P_2C+S_2^2\end{pmatrix} +\begin{pmatrix}R_1^1+D_1^\top P_1D \\ R_2^2+D_2^\top P_2D\end{pmatrix}\Theta^*\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \begin{pmatrix}B_1^\top&0 \\ 0&B_2^\top\end{pmatrix} \begin{pmatrix}P_1&0 \\ 0&P_2\end{pmatrix} \begin{pmatrix}I_n \\ I_n\end{pmatrix} +\begin{pmatrix}D_1^\top&0 \\ 0&D_2^\top\end{pmatrix} \begin{pmatrix}P_1&0 \\ 0&P_2\end{pmatrix} \begin{pmatrix} C&0 \\ 0&C \end{pmatrix} \begin{pmatrix} I_n \\ I_n \end{pmatrix}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle +\begin{pmatrix}I_{m_1}&0&0&0 \\ 0&0&0&I_{m_2}\end{pmatrix} \begin{pmatrix}S^1&0 \\ 0&S^2\end{pmatrix} \begin{pmatrix} I_n \\ I_n \end{pmatrix}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle +\left[\begin{pmatrix}I_{m_1}&0&0&0 \\ 0&0&0&I_{m_2}\end{pmatrix} \begin{pmatrix}R^1&0 \\ 0&R^2\end{pmatrix} \begin{pmatrix} I_m \\ I_m\end{pmatrix} +\begin{pmatrix}D_1^\top&0 \\ 0&D_2^\top\end{pmatrix} \begin{pmatrix} P_1&0 \\ 0&P_2 \end{pmatrix} \begin{pmatrix} D&0 \\ 0&D \end{pmatrix} \begin{pmatrix} I_m \\ I_m \end{pmatrix}\right]\Theta^*\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &\equiv&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle {\bf J}^\top\big({\bf B}^\top{\bf P}+{\bf D}^\top{\bf P}{\bf C}+{\bf S}\big){\bf I}_n +\big[{\bf J}^\top\big({\bf R}+{\bf D}^\top{\bf P}{\bf D}\big){\bf I}_m\big]\Theta^*,\end{array}$$ with $${\bf P}(\cdot)\equiv\begin{pmatrix}P_1(\cdot)&0\\ 0&P_2(\cdot)\end{pmatrix}.$$ Hence, in the case that $\big[{\bf J}^\top\big({\bf R}+{\bf D}^\top{\bf P}{\bf D}\big){\bf I}_m\big]^{-1} \equiv\begin{pmatrix}R_1^1+D_1^\top P_1D\\ R_2^2+D_2^\top P_2D\end{pmatrix}^{-1}$ exists and is bounded, we have \begin{equation}\label{Th**}\Theta^*=-\big[{\bf J}^\top\big({\bf R}+{\bf D}^\top{\bf P}{\bf D}\big){\bf I}_m\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\bf P}+{\bf D}^\top{\bf P}{\bf C}+{\bf S}\big){\bf I}_n,\end{equation} which is the same as \eqref{Th*}. On the other hand, \eqref{Riccati-i} can be written as $$\begin{array}{ll} \ds0=\begin{pmatrix}\dot P_1&0 \\ 0&\dot P_2\end{pmatrix} +\begin{pmatrix}P_1&0 \\ 0&P_2\end{pmatrix} \begin{pmatrix}A&0 \\ 0&A\end{pmatrix} +\begin{pmatrix}A&0 \\ 0&A\end{pmatrix}^\top \begin{pmatrix}P_1&0 \\ 0&P_2\end{pmatrix}\end{array}$$ $$\begin{array}{ll} \noalign{\smallskip}\displaystyle\qquad+\begin{pmatrix} C&0 \\ 0&C\end{pmatrix}^\top \begin{pmatrix}P_1&0 \\ 0&P_2\end{pmatrix} \begin{pmatrix}C&0 \\ 0&C\end{pmatrix} +\begin{pmatrix}Q_1&0 \\ 0&Q_2\end{pmatrix}\\ \noalign{\smallskip}\displaystyle\qquad+\begin{pmatrix}\Theta^*&0 \\ 0&\Theta^*\end{pmatrix}^\top \begin{pmatrix}R^1+D^\top P_1D&0 \\ 0&R^2+D^\top P_2D\end{pmatrix} \begin{pmatrix}\Theta^*&0 \\ 0&\Theta^*\end{pmatrix}\\ \noalign{\smallskip}\displaystyle\qquad+\begin{pmatrix}P_1B+C^\top P_1D+(S^1)^\top&0 \\ 0&P_2B+C^\top P_2D+(S^2)^\top\end{pmatrix} \begin{pmatrix}\Theta^*&0 \\ 0&\Theta^*\end{pmatrix}\\ \noalign{\smallskip}\displaystyle\qquad+\begin{pmatrix}\Theta^*&0 \\ 0&\Theta^*\end{pmatrix}^\top \begin{pmatrix}B^\top P_1+D^\top P_1C+S^1&0 \\ 0&B^\top P_2+D^\top P_2C+S^2\end{pmatrix}.\end{array}$$ Consequently, one sees that the following holds: \begin{equation}\label{Riccati**}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot{\bf P}+{\bf P}{\bf A}+{\bf A}^\top{\bf P}+{\bf C}^\top{\bf P}{\bf C}+{\bf Q}+\,{\Theta\3n\2n\Theta\3n\2n\Theta}^\top\big({\bf R}+{\bf D}^\top{\bf P}{\bf D}\big)\,{\Theta\3n\2n\Theta\3n\2n\Theta}\\ \noalign{\smallskip}\displaystyle\,~~+\big({\bf P}{\bf B}+{\bf C}^\top{\bf P}{\bf D}+{\bf S}^\top\big)\,{\Theta\3n\2n\Theta\3n\2n\Theta}+\,{\Theta\3n\2n\Theta\3n\2n\Theta}^\top\big({\bf B}^\top{\bf P}+{\bf D}^\top{\bf P}{\bf C}+{\bf S}\big)=0, \quad~\hbox{\rm a.e.}~s\in[t,T],\\ \noalign{\smallskip}\displaystyle{\bf P}(T)={\bf G},\end{array}\right.\end{equation} where $$\,{\Theta\3n\2n\Theta\3n\2n\Theta}(\cdot)=\begin{pmatrix}\Theta^*(\cdot)&0\\ 0&\Theta^*(\cdot)\end{pmatrix},$$ and $\Theta^*$ is given by \eqref{Th**}. Clearly, \eqref{Riccati**} is symmetric. \section{Two Examples} From the previous sections, we see that the existence of an open-loop Nash equilibrium is equivalent to the solvability of a coupled system of two FBSDEs, together with the convexity condition for the cost functionals (see \eqref{convexity}); and that the existence of a closed-loop Nash equilibrium is equivalent to the solvability of a coupled system of two symmetric Riccati equations satisfying certain type of non-negativity condition (see \eqref{R+DPD>0}). Then a natural question is: Are open-loop and closed-loop Nash equilibria really different? In this section, we will present two examples showing that they are indeed different. \medskip The following example shows that Problem (SDG) may have only open-loop Nash equilibria. \begin{example}\label{Example 6.1}\rm Consider the following Problem (SDG) with one-dimensional state equation $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big[u_1(s)+u_2(s)\big]ds+\big[u_1(s)-u_2(s)\big]dW(s),\qquad s\in[t,1], \\ \noalign{\smallskip}\displaystyle X(t)=x,\end{array}\right.$$ and cost functionals $$ J^1(t,x;u_1(\cdot),u_2(\cdot))=J^2(t,x;u_1(\cdot),u_2(\cdot))=\mathbb{E} X(1)^2\equiv J(t,x;u_1(\cdot),u_2(\cdot)).$$ Let $\beta\geqslant{1\over 1-t}$. We claim that $$\left(u_1^\beta(s),u_2^\beta(s)\right)=-\left({\beta x\over2}{\bf1}_{[t,t+{1\over\beta}]}(s), {\beta x\over2}{\bf1}_{[t,t+{1\over\beta}]}(s)\right), \qquad s\in[t,1],$$ is an open-loop Nash equilibrium of the problem for the initial pair $(t,x)$. Indeed, it is clear that for any $u_1(\cdot)\in L^2_\mathbb{F}(t,1;\mathbb{R})$, $$J(t,x;u_1(\cdot),u_2^\beta(\cdot))\ges0.$$ On the other hand, the state process $X^\beta(\cdot)$ corresponding to $\big(u_1^\beta(s),u_2^\beta(s)\big)$ and $(t,x)$ satisfies $X^\beta(1)=0$. Hence, $$J(t,x;u_1^\beta(\cdot),u_2^\beta(\cdot))=0\leqslant J(t,x;u_1(\cdot),u_2^\beta(\cdot)),\qquad \forall\, u_1(\cdot)\in L^2_\mathbb{F}(t,1;\mathbb{R}).$$ Likewise, $$J(t,x;u_1^\beta(\cdot),u_2^\beta(\cdot))=0\leqslant J(t,x;u_1^\beta(\cdot),u_2(\cdot)),\qquad \forall\, u_2(\cdot)\in L^2_\mathbb{F}(t,1;\mathbb{R}).$$ This establishes the claim. \medskip However, this problem does {\it not} admit a closed-loop Nash equilibrium. We now show this by contradiction. Suppose $(\Theta^*_1(\cdot),v_1^*(\cdot);\Theta^*_2(\cdot),v^*_2(\cdot))$ is a closed-loop Nash equilibrium. Consider the corresponding ODEs in Theorem \ref{Theorem 5.2}, which now become \begin{equation}\label{16Apr27-20:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_i+P_i(\Theta_1^*-\Theta^*_2)^2+2P_i(\Theta_1^*+\Theta^*_2)=0,\\ \noalign{\smallskip}\displaystyle P_i(1)=1,\end{array}\right.\qquad i=1,2.\end{equation} The corresponding constraints read \begin{equation}\label{6.4}P_1,P_2\ges0,\qquad P_1+P_1(\Theta_1^*-\Theta^*_2)=0,\qquad P_2-P_2(\Theta_1^*-\Theta^*_2)=0.\end{equation} Since $P_1(\cdot)$ and $P_2(\cdot)$ satisfy the same ODE \eqref{16Apr27-20:00}, we have $P_1(\cdot)=P_2(\cdot)$. Then \eqref{6.4} implies $P_1(\cdot)=0$, which contradicts the terminal condition $P_1(1)=1$. \end{example} The following example shows that Problem (SDG) may have only closed-loop Nash equilibria. \begin{example}\label{Example 6.2}\rm Consider the following Problem (SDG) with one-dimensional state equation $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=u_1(s)ds+u_2(s)dW(s),\qquad s\in[t,1], \\ \noalign{\smallskip}\displaystyle X(t)=x,\end{array}\right.$$ and cost functionals \begin{eqnarray} J^1(t,x;u_1(\cdot),u_2(\cdot))\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \mathbb{E}\Big\{|X(1)|^2+\int_t^1|u_1(s)|^2 ds\Big\},\nonumber\\ J^2(t,x;u_1(\cdot),u_2(\cdot))\negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \mathbb{E}\Big\{-|X(1)|^2+\int_t^1\[-|X(s)|^2+|u_2(s)|^2\] ds\Big\}.\nonumber \end{eqnarray} We claim that the problem admits a closed-loop Nash equilibrium of form $(\Theta_1(\cdot),0;$ $\Theta_2(\cdot),0)$. In fact, by Theorem \ref{Theorem 5.2}, we need to solve the following Riccati equations for $P_1(\cdot)$ and $P_2(\cdot)$: \begin{eqnarray} &&\label{16Apr26-17:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \dot P_1(s)+P_1(s)\Theta_2(s)^2+2P_1(s)\Theta_1(s)+\Theta_1(s)^2=0, \\ \noalign{\smallskip}\displaystyle P_1(1)=1,\\ \noalign{\smallskip}\displaystyle P_1(s)+\Theta_1(s)=0, \end{array}\right.\\ \noalign{\smallskip}&&\label{16Apr26-17:30}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \dot P_2(s)+P_2(s)\Theta_2(s)^2+2P_2(s)\Theta_1(s)+\Theta_2(s)^2-1=0, \\ \noalign{\smallskip}\displaystyle P_2(1)=-1,\\ \noalign{\smallskip}\displaystyle 1+P_2(s)\ges0,\\ \noalign{\smallskip}\displaystyle [1+P_2(s)]\Theta_2(s)=0.\end{array}\right. \end{eqnarray} By the fourth equation in \eqref{16Apr26-17:30}, we may assume $\Theta_2(\cdot)=0$. Then \eqref{16Apr26-17:00}--\eqref{16Apr26-17:30} become (taking into account $\Theta_1(\cdot)=-P_1(\cdot)$ from the third equation in \eqref{16Apr26-17:00}) $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \dot P_1(s)=P_1(s)^2, \\ \noalign{\smallskip}\displaystyle P_1(1)=1,\end{array}\right.\qquad \left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \dot P_2(s)=2P_1(s)P_2(s)+1, \\ \noalign{\smallskip}\displaystyle P_2(1)=-1,\quad 1+P_2(s)\ges0.\end{array}\right. $$ A straightforward calculation leads to $$P_1(s)={1\over 2-s},\qquad P_2(s)={-(2-s)^3-2\over 3(2-s)^2}.$$ Therefore, $((2-s)^{-1},0;0,0)$ is a closed-loop Nash equilibrium of the problem. \medskip Next, we claim that the problem does {\it not} have open-loop Nash equilibria. Indeed, suppose $(u_1^*(\cdot),u_2^*(\cdot))$ is an open-loop Nash equilibrium for some initial pair $(t,x)$. Then $u_2^*(\cdot)$ is an open-loop optimal control of the following Problem (SLQ) with state equation \begin{equation}\label{16Apr26-21:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=u_1^*(s)ds+u_2(s)dW(s),\qquad s\in[t,1], \\ \noalign{\smallskip}\displaystyle X(t)=x,\end{array}\right.\end{equation} and cost functional \begin{equation}\label{16Apr26-21:30}\widetilde J(t,x;u_2(\cdot))=\mathbb{E}\Big\{-|X(1)|^2+\int_t^1\[-|X(s)|^2+|u_2(s)|^2\] ds\Big\}.\end{equation} For any $u_2(\cdot)\in L^2_\mathbb{F}(t,1;\mathbb{R})$, the corresponding solution to \eqref{16Apr26-21:00} is given by \begin{equation}\label{16Apr26-22:00}X(s)=x+\int_t^s u_1^*(r)dr+\int_t^s u_2(r)dW(r).\end{equation} Let $\varepsilon>0$ be undetermined. Substituting \eqref{16Apr26-22:00} into \eqref{16Apr26-21:30} and using the inequality $(a+b)^2\geqslant(1-{1\over\varepsilon})a^2+(1-\varepsilon)b^2$, we see $$\begin{array}{lll} \displaystyle\widetilde J(t,x;u_2(\cdot))\negthinspace \negthinspace \negthinspace \negthinspace &\leqslant&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\({1\over\varepsilon}-1\)\mathbb{E}\(x+\int_t^1u_1^*(s)ds\)^2 +(\varepsilon-1)\mathbb{E}\(\int_t^1u_2(s)dW(s)\)^2\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,\({1\over\varepsilon}-1\)\mathbb{E}\int_t^1\(x+\int_t^s u_1^*(r)dr\)^2ds\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,(\varepsilon-1)\mathbb{E}\int_t^1\(\int_t^s u_2(r)dW(r)\)^2ds+\mathbb{E}\int_t^1|u_2(s)|^2 ds\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\({1\over\varepsilon}-1\)\mathbb{E}\[\(x+\int_t^1u_1^*(s)ds\)^2 +\int_t^1\(x+\int_t^s u_1^*(r)dr\)^2ds\]\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,\varepsilon\mathbb{E}\int_t^1|u_2(s)|^2 ds+(\varepsilon-1)\mathbb{E}\int_t^1\int_t^s|u_2(r)|^2drds. \end{array}$$ Now, by taking $u_2(s)=\lambda$, $\lambda\in\mathbb{R}$, we have $$\begin{array}{lll} \displaystyle\widetilde J(t,x;\lambda)\negthinspace \negthinspace \negthinspace \negthinspace &\leqslant&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\({1\over\varepsilon}-1\)\mathbb{E}\[\(x+\int_t^1u_1^*(s)ds\)^2 +\int_t^1\(x+\int_t^s u_1^*(r)dr\)^2ds\]\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,{\lambda^2(1-t)\over2}\big[2\varepsilon+(\varepsilon-1)(1-t)\big].\end{array}$$ Choosing $\varepsilon>0$ small enough so that $2\varepsilon+(\varepsilon-1)(1-t)<0$ and then letting $\lambda\to\infty$, we see that $$\inf_{u_2(\cdot)\in L^2_\mathbb{F}(t,1;\mathbb{R})}\widetilde J(t,x;u_2(\cdot))=-\infty,$$ which contradicts the fact that $u_2^*(\cdot)$ is an open-loop optimal control of the associated LQ problem. \end{example} \section{Closed-Loop Representation of Open-Loop Nash Equilibria} Inspired by the decoupling technique introduced in \cite{Ma-Protter-Yong 1994,Ma-Yong 1999,Yong 1999, Yong 2006}, we now look at the solvability of FBSDE \eqref{FBSDEi}--\eqref{stationary}. Recall that with the notation introduced in Section 4, \eqref{FBSDEi} and \eqref{stationary} are equivalent to \eqref{FBSDE**} and \eqref{stationary**}, respectively. To solve FBSDE \eqref{FBSDE**}--\eqref{stationary**}, let $({\eta\3n\eta\3n\eta}(\cdot),{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot))$ be the adapted solution to the following BSDE for some undetermined $\alpha:[t,T]\times\Omega\to\mathbb{R}^{2n}$: $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d{\eta\3n\eta\3n\eta}(s)=\alpha(s)ds+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(s)dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle{\eta\3n\eta\3n\eta}(T)={\bf g},\end{array}\right.$$ where $${\eta\3n\eta\3n\eta}(\cdot)=\begin{pmatrix}\eta_1(\cdot)\\ \eta_2(\cdot)\end{pmatrix}, \qquad{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot)=\begin{pmatrix}\zeta_1(\cdot)\\ \zeta_2(\cdot)\end{pmatrix}.$$ Let $(X(\cdot),{\bf Y}(\cdot),{\bf Z}(\cdot))$ be an adapted solution to FBSDE \eqref{FBSDE**}. Suppose the following holds: \begin{equation}\label{BY}{\bf Y}(\cdot)=\begin{pmatrix}\Pi_1(\cdot)X(\cdot)+\eta_1(\cdot)\\ \Pi_2(\cdot)X(\cdot)+\eta_2(\cdot)\end{pmatrix}\equiv{\bf\Pi}(\cdot) X(\cdot)+{\eta\3n\eta\3n\eta}(\cdot),\qquad{\bf\Pi}(\cdot)\triangleq\begin{pmatrix}\Pi_1(\cdot)\\ \Pi_2(\cdot)\end{pmatrix},\end{equation} for some differentiable maps $\Pi_i:[t,T]\to\mathbb{R}^{n\times n}$ with $\Pi_i(T)=G^i$. By It\^o's formula, we have $$\begin{array}{ll} \displaystyle-\big({\bf A}^\top{\bf Y}+{\bf C}^\top{\bf Z}+{\bf Q}{\bf I}_nX+{\bf S}^\top{\bf I}_mu+{\bf q}\big)ds+{\bf Z} dW(s)=d{\bf Y}\\ \noalign{\smallskip}\displaystyle=\big[\dot{\bf\Pi} X+{\bf\Pi}(AX+Bu+b)+\alpha\big]ds+\big[{\bf\Pi}(CX+Du+\sigma)+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\big]dW(s)\\ \noalign{\smallskip}\displaystyle=\big[(\dot{\bf\Pi}+{\bf\Pi} A)X+{\bf\Pi} Bu+{\bf\Pi} b+\alpha\big]ds+\big[{\bf\Pi} CX+{\bf\Pi} Du+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\big]dW(s).\end{array}$$ Hence, one should have \begin{equation}\label{BZ}{\bf Z}={\bf\Pi} CX+{\bf\Pi} Du+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}.\end{equation} Then the stationarity condition \eqref{stationary**} becomes $$\begin{array}{lll} 0\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle{\bf J}^\top\big({\bf B}^\top{\bf Y}+{\bf D}^\top{\bf Z}+{\bf S}{\bf I}_nX+{\bf R}{\bf I}_mu+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle{\bf J}^\top\big[{\bf B}^\top({\bf\Pi} X+{\eta\3n\eta\3n\eta})+{\bf D}^\top({\bf\Pi} CX+{\bf\Pi} Du+{\bf\Pi}\sigma +{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta})+{\bf S}{\bf I}_nX+{\bf R}{\bf I}_mu+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big]\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle{\bf J}^\top\big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)X+{\bf J}^\top\big({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D\big)u+{\bf J}^\top\big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big).\end{array}$$ Note that $$ {\bf J}^\top\big({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D\big) =\begin{pmatrix} I_{m_1}&0&0&0 \\ 0&0&0&I_{m_2} \end{pmatrix} \begin{pmatrix}R^1+D^\top\Pi_1D \\ R^2+D^\top\Pi_2D\end{pmatrix} =\begin{pmatrix}R^1_{11}+D_1^\top\Pi_1D_1 & R^1_{12}+D_1^\top\Pi_1D_2\\ R^2_{21}+D_2^\top\Pi_2D_1 & R^2_{22}+D_2^\top\Pi_2D_2\end{pmatrix}.$$ This is an $\mathbb{R}^{m\times m}$-valued function which is not symmetric in general, even $\Pi_1$ and $\Pi_2$ are symmetric. We now assume that the above is invertible. Then one has \begin{equation}\label{open-u}\begin{array}{lll} \displaystyle u\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)X\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big),\end{array}\end{equation} and $$\begin{array}{lll} 0\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\big(\dot{\bf\Pi}+{\bf\Pi} A\big)X+{\bf\Pi} Bu+{\bf\Pi} b+\alpha+{\bf A}^\top({\bf\Pi} X+{\eta\3n\eta\3n\eta})\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,{\bf C}^\top({\bf\Pi} CX+{\bf\Pi} Du+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta})+{\bf Q}{\bf I}_nX+{\bf S}^\top{\bf I}_mu+{\bf q}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\big(\dot{\bf\Pi}+{\bf\Pi} A+{\bf A}^\top{\bf\Pi}+{\bf C}^\top{\bf\Pi} C+{\bf Q}{\bf I}_n\big)X +\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)u\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,\alpha+{\bf A}^\top{\eta\3n\eta\3n\eta}+{\bf C}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf\Pi} b+{\bf C}^\top{\bf\Pi}\sigma+{\bf q}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\big(\dot{\bf\Pi}+{\bf\Pi} A+{\bf A}^\top{\bf\Pi}+{\bf C}^\top{\bf\Pi} C+{\bf Q}{\bf I}_n\big)X\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\,\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)X\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\,\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top\big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,\alpha+{\bf A}^\top{\eta\3n\eta\3n\eta}+{\bf C}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf\Pi} b+{\bf C}^\top{\bf\Pi}\sigma+{\bf q} \\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\Big\{\dot{\bf\Pi}+{\bf\Pi} A+{\bf A}^\top{\bf\Pi}+{\bf C}^\top{\bf\Pi} C+{\bf Q}{\bf I}_n\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\,\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)\Big\}X\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\,\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,\alpha+{\bf A}^\top{\eta\3n\eta\3n\eta}+{\bf C}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf\Pi} b+{\bf C}^\top{\bf\Pi}\sigma+{\bf q}.\end{array}$$ Now, let ${\bf\Pi}(\cdot)$ be the solution to the following Riccati equation: \begin{equation}\label{Riccati-Pi}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot{\bf\Pi}+{\bf\Pi} A+{\bf A}^\top{\bf\Pi}+{\bf C}^\top{\bf\Pi} C+{\bf Q}{\bf I}_n\\ \noalign{\smallskip}\displaystyle-\,\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)\big[{\bf J}^\top({\bf R}{\bf I}_m\negthinspace +{\bf D}^\top{\bf\Pi} D)\big]^{-1} {\bf J}^\top\big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)=0,\quad s\in[t,T],\\ \noalign{\smallskip}\displaystyle{\bf\Pi}(T)={\bf G}{\bf I}_n.\end{array}\right.\end{equation} Then the above leads to the BSDE for $({\eta\3n\eta\3n\eta}(\cdot),{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot))$ of the following form: \begin{equation}\label{BSDE-open}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d{\eta\3n\eta\3n\eta}=-\Big\{\({\bf A}^\top-\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big) \big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top{\bf B}^\top\){\eta\3n\eta\3n\eta}\\ \noalign{\smallskip}\displaystyle\qquad\quad+\,\({\bf C}^\top-\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big) \big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top{\bf D}^\top\){\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\\ \noalign{\smallskip}\displaystyle\qquad\quad+\,\({\bf C}^\top-\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big) \big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top{\bf D}^\top\){\bf\Pi}\sigma\\ \noalign{\smallskip}\displaystyle\qquad\quad+\,{\bf\Pi} b+{\bf q}-\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big) \big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\Big\}ds+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta} dW,\quad s\in[t,T],\\ \noalign{\smallskip}\displaystyle{\eta\3n\eta\3n\eta}(T)={\bf g}.\end{array}\right.\end{equation} Hence, we have the following result. \begin{theorem}\label{Theorem 4.2} \sl Let {\rm(G1)--(G2)} hold and let $t\in[0,T)$ be given. Suppose that the convexity condition \eqref{convexity} holds for $i=1,2$, and that the Riccati equation \eqref{Riccati-Pi} admits a solution ${\bf\Pi}(\cdot)$. Let $({\eta\3n\eta\3n\eta}(\cdot),{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot))$ be the adapted solution to BSDE \eqref{BSDE-open} and let $X(\cdot)$ be the solution to the following FSDE with an arbitrary initial state $x$: \begin{equation}\label{open-closed}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX=\Big\{\(A-B\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top \big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)\) X\\ \noalign{\smallskip}\displaystyle\qquad\qq~-B\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top \big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)+b\Big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\quad+\,\Big\{\(C-D\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top \big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)\)X\\ \noalign{\smallskip}\displaystyle\qquad\qq~-D\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]^{-1}{\bf J}^\top\big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta} +{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)+\sigma\Big\}dW,\quad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x.\end{array}\right.\end{equation} Then the process $u(\cdot)$ defined by \eqref{open-u} is an open-loop Nash equilibrium of Problem {\rm(SDG)} for $(t,x)$. \end{theorem} \it Proof. \rm From the above procedure, we see that with $u(\cdot)$ defined by \eqref{open-u}, the triple $(X(\cdot),{\bf Y}(\cdot)$, ${\bf Z}(\cdot))$ defined through FSDE \eqref{open-closed}, \eqref{BY} and \eqref{BZ}, is an adapted solution to FBSDE \eqref{FBSDE**}, and that the stationarity condition \eqref{stationary**} holds. Hence, together with the convexity condition \eqref{convexity}, making use of Theorem \ref{Theorem 4.1}, we see that $u(\cdot)$ is an open-loop Nash equilibrium of Problem (SDG) for $(t,x)$. \signed {$\sqr69$} \medskip Under the assumptions of Theorem \ref{Theorem 4.2}, Problem (SDG) admits an open-loop Nash equilibrium for every initial state $x$, and the open-loop Nash equilibria take the following form: \begin{equation}\label{16July9-21:00}u(\cdot)=\Theta(\cdot)X(\cdot)+v(\cdot),\end{equation} for some $(\Theta(\cdot),v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$ which is independent of $x$. The above \eqref{16July9-21:00} is called a {\it closed-loop representation} of the open-loop Nash equilibria of Problem (SDG). More precisely, we have the following definition. \begin{definition}\label{bde-16June29-17:00} \rm We say that open-loop Nash equilibria of Problem {\rm(SDG)} on $[t,T]$ admit a {\it closed-loop representation}, if there exists a pair $(\Theta(\cdot),v(\cdot))\in {\cal Q}[t,T]\times{\cal U}[t,T]$ such that for any initial state $x\in\mathbb{R}^n$, the process \begin{equation}\label{16June29-17:40}u(s)\triangleq \Theta(s)X(s)+v(s),\qquad s\in[t,T]\end{equation} is an open-loop Nash equilibrium of Problem {\rm(SDG)} for $(t,x)$, where $X(\cdot)$ is the solution to the following closed-loop system: \begin{equation}\label{16June29-X}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big\{[A(s)+B(s)\Theta(s)]X(s)+B(s)v(s)+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq\negthinspace ~+\big\{[C(s)+D(s)\Theta(s)]X(s)+D(s)v(s)+\sigma(s)\big\}dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)= x. \end{array}\right.\end{equation} \end{definition} Comparing Definitions \ref{bde-16Apr4-17:00} and \ref{bde-16June29-17:00}, it is natural to ask whether the closed-loop representation of open-loop Nash equilibria is the outcome of some closed-loop Nash equilibrium. The following example shows that this is not the case in general. \begin{example}\label{Example 6.3} \rm Consider the following state equation: $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big[u_1(s)+u_2(s)\big]ds+X(s)dW(s),\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle X(t)=x,\end{array}\right.$$ with cost functionals $$\begin{array}{ll} \displaystyle J^1(t,x;u_1(\cdot),u_2(\cdot))=\mathbb{E}\[X(T)^2+\int_t^Tu_1(s)^2ds\],\\ \noalign{\smallskip}\displaystyle J^2(t,x;u_1(\cdot),u_2(\cdot))=\mathbb{E}\[X(T)^2+\int_t^Tu_2(s)^2ds\].\end{array}$$ For this case, we have $$\left\{\negthinspace \negthinspace \begin{array}{llll} \displaystyle A=0,~C=1, & B_1=B_2=1, & D_1=D_2=0, & b=\sigma=0,\\ \noalign{\smallskip} Q^1=Q^2=0, & S^1=S^2=0, & R^1=\begin{pmatrix}1&0 \\ 0&0\end{pmatrix}, & R^2=\begin{pmatrix}0&0 \\ 0&1\end{pmatrix},\\ \noalign{\smallskip} G^1=G^2=1, & q^1=q^2=0, & \rho^1=\rho^2=0, & g^1=g^2=0.\end{array}\right.$$ Clearly, the convexity condition \eqref{convexity} holds for $i=1,2$. In this example, the Riccati equation \eqref{Riccati-Pi} can be written componentwise as follows: \begin{eqnarray} &&\label{Pi1-1*}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi_1(s)+\Pi_1(s)-\Pi_1(s)\big[\Pi_1(s)+\Pi_2(s)\big]=0,\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle\Pi_1(T)=1,\end{array}\right.\\ \noalign{\smallskip}&&\label{Pi2-1*}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi_2(s)+\Pi_2(s)-\Pi_2(s)\big[\Pi_1(s)+\Pi_2(s)\big]=0,\qquad s\in[t,T],\\ \noalign{\smallskip}\displaystyle\Pi_2(T)=1.\end{array}\right. \end{eqnarray} It is easy to see that $$\Pi_1(s)=\Pi_2(s)={e^{T-s}\over 2e^{T-s}-1}$$ are solutions to \eqref{Pi1-1*} and \eqref{Pi2-1*}, respectively. Note that in this case the adapted solution $({\eta\3n\eta\3n\eta}(\cdot),{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot))$ to BSDE \eqref{BSDE-open} is $(0,0)$. Then by Theorem \ref{Theorem 4.2}, the open-loop Nash equilibria of this Problem (SDG) on $[t,T]$ admit a closed-loop representation given by \begin{equation}\label{16June26_22:30-1}u_1(s)=u_2(s)=-{e^{T-s}\over 2e^{T-s}-1}X(s),\qquad s\in[t,T].\end{equation} \smallskip Next we verify that the problem admits a closed-loop Nash equilibrium of form $(\Theta_1(\cdot),0;$ $\Theta_2(\cdot),0)$. In light of Theorem \ref{Theorem 5.2}, we need to solve the following Riccati equations for $P_1(\cdot)$ and $P_2(\cdot)$: \begin{equation}\label{16June26-21:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_1(s)+P_1(s)+\Theta_1(s)^2+2P_1(s)\big[\Theta_1(s)+\Theta_2(s)\big]=0,\\ \noalign{\smallskip}\displaystyle P_1(T)=1,\\ \noalign{\smallskip}\displaystyle P_1(s)+\Theta_1(s)=0, \end{array}\right.\end{equation} and \begin{equation}\label{16June26-21:10}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_2(s)+P_2(s)+\Theta_2(s)^2+2P_2(s)\big[\Theta_1(s)+\Theta_2(s)\big]=0,\\ \noalign{\smallskip}\displaystyle P_2(T)=1,\\ \noalign{\smallskip}\displaystyle P_2(s)+\Theta_2(s)=0. \end{array}\right.\end{equation} Noting the third equations in \eqref{16June26-21:00} and \eqref{16June26-21:10}, we can further write \eqref{16June26-21:00}-\eqref{16June26-21:10} as follows: \begin{eqnarray} &&\label{16June26-22:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_1(s)=P_1(s)^2+2P_1(s)P_2(s)-P_1(s),\\ \noalign{\smallskip}\displaystyle P_1(T)=1, \end{array}\right.\\ \noalign{\smallskip}&&\label{16June26-22:10}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot P_2(s)=P_2(s)^2+2P_2(s)P_1(s)-P_2(s),\\ \noalign{\smallskip}\displaystyle P_2(T)=1. \end{array}\right. \end{eqnarray} Now it is easily seen that $$P_1(s)=P_2(s)={e^{T-s}\over3e^{T-s}-2}.$$ Hence, \begin{equation}\label{16June26_22:30-2}\Theta_1(s)=\Theta_2(s)=-P_1(s)=-{e^{T-s}\over3e^{T-s}-2}.\end{equation} Comparing \eqref{16June26_22:30-1} with \eqref{16June26_22:30-2}, we see that the closed-loop representation of open-loop Nash equilibria is different from the outcome of closed-loop Nash equilibria. \end{example} Now we give a characterization of the closed-loop representation of open-loop Nash equilibria. \begin{theorem}\label{bt-16June29-17:30}\sl Let {\rm(G1)--(G2)} hold and let $(\Theta(\cdot),v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$. Then open-loop Nash equilibria of Problem {\rm(SDG)} on $[t,T]$ admit the closed-loop representation \eqref{16June29-17:40} if and only if the following hold: \medskip {\rm(i)} The convexity condition \eqref{convexity} holds for $i=1,2$. \medskip {\rm(ii)} The solution ${\bf\Pi}(\cdot)\in C([t,T];\mathbb{R}^{n\times 2n})$ to the ODE on $[t,T]$ \begin{equation}\label{16June29-Pi}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot{\bf\Pi}+{\bf\Pi} A+{\bf A}^\top{\bf\Pi}+{\bf C}^\top{\bf\Pi} C+{\bf Q}{\bf I}_n +\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)\Theta=0,\\ \noalign{\smallskip}\displaystyle{\bf\Pi}(T)={\bf G}{\bf I}_n,\end{array}\right.\end{equation} satisfies \begin{equation}\label{16June29-21:40-1}\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]\Theta +{\bf J}^\top\big({\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n\big)=0,\end{equation} and the adapted solution $({\eta\3n\eta\3n\eta}(\cdot),{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot))$ to the BSDE on $[t,T]$ \begin{equation}\label{16June29-BBeta}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d{\eta\3n\eta\3n\eta}=-\big[{\bf A}^\top{\eta\3n\eta\3n\eta}+{\bf C}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)v +{\bf C}^\top{\bf\Pi}\sigma+{\bf\Pi} b+{\bf q}\big]ds+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta} dW,\\ \noalign{\smallskip}\displaystyle{\eta\3n\eta\3n\eta}(T)={\bf g},\end{array}\right.\end{equation} satisfies \begin{equation}\label{16June29-21:40-2}\big[{\bf J}^\top({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\big]v +{\bf J}^\top\big({\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)=0.\end{equation} \end{theorem} \it Proof. \rm For any $x\in\mathbb{R}^n$, let $X(\cdot)$, ${\bf\Pi}(\cdot)$, and $({\eta\3n\eta\3n\eta}(\cdot),{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}(\cdot))$ be the solutions to \eqref{16June29-X}, \eqref{16June29-Pi}, and \eqref{16June29-BBeta}, respectively. Let $u(\cdot)$ be defined by \eqref{16June29-17:40} and set $${\bf Y}={\bf\Pi} X+{\eta\3n\eta\3n\eta},\qquad {\bf Z}={\bf\Pi}(C+D\Theta)X+{\bf\Pi} Dv+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}.$$ Then ${\bf Y}(T)={\bf G}{\bf I}_nX(T)+{\bf g}$, and $$\begin{array}{lll} \displaystyle d{\bf Y}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \dot{\bf\Pi} Xds+{\bf\Pi} dX+d{\eta\3n\eta\3n\eta}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \big[\dot{\bf\Pi} X+{\bf\Pi}(A+B\Theta)X+{\bf\Pi} Bv+{\bf\Pi} b-{\bf A}^\top{\eta\3n\eta\3n\eta}-{\bf C}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\qquad~-\big({\bf\Pi} B+{\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)v-{\bf C}^\top{\bf\Pi}\sigma-{\bf\Pi} b-{\bf q}\big]ds\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle+\,\big[{\bf\Pi}(C+D\Theta)X+{\bf\Pi} Dv+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\big]dW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \big[-\big({\bf A}^\top{\bf\Pi}+{\bf C}^\top{\bf\Pi} C+{\bf Q}{\bf I}_n+{\bf C}^\top{\bf\Pi} D\Theta+{\bf S}^\top{\bf I}_m\Theta\big)X -{\bf A}^\top{\eta\3n\eta\3n\eta}-{\bf C}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\qquad~-\big({\bf C}^\top{\bf\Pi} D+{\bf S}^\top{\bf I}_m\big)v-{\bf C}^\top{\bf\Pi}\sigma-{\bf q}\big]ds+{\bf Z} dW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \big\{-{\bf A}^\top({\bf\Pi} X+{\eta\3n\eta\3n\eta})-{\bf Q}{\bf I}_n X-{\bf C}^\top\big[{\bf\Pi}(C+D\Theta)X+{\bf\Pi} Dv+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}\big]\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\qquad~-{\bf S}^\top{\bf I}_m(\Theta X+v)-{\bf q}\big\}ds+{\bf Z} dW\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \big(-{\bf A}^\top{\bf Y}-{\bf Q}{\bf I}_n X-{\bf C}^\top{\bf Z}-{\bf S}^\top{\bf I}_m u-{\bf q}\big)ds+{\bf Z} dW. \end{array}$$ This shows that $(X(\cdot),{\bf Y}(\cdot),{\bf Z}(\cdot),u(\cdot))$ satisfies the FBSDE \eqref{FBSDE**}. According to Theorem \ref{Theorem 4.1}, the process $u(\cdot)$ defined by \eqref{16June29-17:40} is an open-loop Nash equilibrium for $(t,x)$ if and only if (i) holds and $$\begin{array}{lll} \ds0\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle {\bf J}^\top\big({\bf B}^\top{\bf Y}+{\bf D}^\top{\bf Z}+{\bf S}{\bf I}_nX+{\bf R}{\bf I}_m u+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big)\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle {\bf J}^\top\big\{{\bf B}^\top({\bf\Pi} X+{\eta\3n\eta\3n\eta})+{\bf D}^\top[{\bf\Pi}(C+D\Theta)X+{\bf\Pi} Dv+{\bf\Pi}\sigma+{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}]+{\bf S}{\bf I}_nX+{\bf R}{\bf I}_m(\Theta X+v)+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}\big\}\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle {\bf J}^\top\big[{\bf B}^\top{\bf\Pi}+{\bf D}^\top{\bf\Pi} C+{\bf S}{\bf I}_n+({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)\Theta\big]X\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle\qquad~ +{\bf J}^\top\big[{\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}+({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)v\big]. \end{array}$$ Since the initial state $x$ is arbitrary and ${\bf J}^\top[{\bf B}^\top{\eta\3n\eta\3n\eta}+{\bf D}^\top{\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta\3n\zeta}+{\bf D}^\top{\bf\Pi}\sigma+{\rho\3n\rho\3n\rho\3n\rho\3n\rho}+({\bf R}{\bf I}_m+{\bf D}^\top{\bf\Pi} D)v]$ is independent of $x$, the above leads to \eqref{16June29-21:40-1} and \eqref{16June29-21:40-2}. \signed {$\sqr69$} \medskip Let us write \eqref{16June29-Pi}--\eqref{16June29-21:40-2} componentwise as follows: For $i=1,2$, \begin{eqnarray} &\label{16June30-Pi}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi_i+\Pi_iA+A^\top\Pi_i+C^\top\Pi_iC+Q^i+\big[\Pi_iB+C^\top\Pi_iD+(S^i)^\top\big]\Theta=0,\\ \noalign{\smallskip}\displaystyle\Pi_i(T)=G^i,\end{array}\right.&\\ \noalign{\smallskip}&\label{16June30-Pi-yueshu}\begin{pmatrix}R_1^1+D_1^\top\Pi_1D \\ R_2^2+D_2^\top\Pi_2D\end{pmatrix}\Theta +\begin{pmatrix}B_1^\top\Pi_1+D_1^\top\Pi_1C+S_1^1 \\B_2^\top\Pi_2+D_2^\top\Pi_2C+S_2^2\end{pmatrix}=0,&\\ \noalign{\smallskip}&\label{16June30-eta}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\eta_i=-\Big\{A^\top\eta_i+C^\top\zeta_i+\big[\Pi_iB+C^\top\Pi_iD+(S^i)^\top\big]v +C^\top\Pi_i\sigma+\Pi_ib+q^i\Big\}ds+\zeta_i dW,\\ \noalign{\smallskip}\displaystyle\eta_i(T)=g^i,\end{array}\right.&\\ \noalign{\smallskip}&\label{16June30-eta-yueshu}\begin{pmatrix}R_1^1+D_1^\top\Pi_1D \\ R_2^2+D_2^\top\Pi_2D\end{pmatrix}v +\begin{pmatrix}B_1^\top\eta_1+D_1^\top\zeta_1+D_1^\top\Pi_1\sigma+\rho_1^1\\ B_2^\top\eta_2+D_2^\top\zeta_2+D_2^\top\Pi_2\sigma+\rho_2^2\end{pmatrix}=0.& \end{eqnarray} Noting the relation \eqref{16June30-Pi-yueshu}, one sees the equations for $\Pi_1(\cdot)$ and $\Pi_2(\cdot)$ are coupled and none of them is symmetric. Consequently, $\Pi_1(\cdot)$ and $\Pi_2(\cdot)$ are not symmetric in general. Whereas the Riccati equations \eqref{Riccati-i} for $P_i(\cdot)$ $(i=1,2)$ are symmetric. This is the main reason that the closed-loop representation of open-loop Nash equilibria is different from the outcome of closed-loop Nash equilibria. \section{Zero-Sum Cases} In the previous section, we have seen that for Problem (SDG), the closed-loop representation of open-loop Nash equilibria is different from the outcome of closed-loop Nash equilibria in general. Now we would like to take a look at the situation for LQ stochastic two-person zero-sum differential games. In this case, Nash equilibria are usually called saddle points. According to \eqref{16Apr2-1+2=0}, we have \begin{equation}\label{16July1-16:00}\begin{array}{ll} \displaystyle G^1=-G^2\equiv G, \quad g^1=-g^2\equiv g, \quad Q^1(\cdot)=-Q^2(\cdot)\equiv Q(\cdot), \quad q^1(\cdot)=-q^2(\cdot)\equiv q(\cdot),\\ \noalign{\smallskip}\displaystyle\begin{pmatrix}R^1_{11}(\cdot)&R^1_{12}(\cdot) \\ R^1_{21}(\cdot)&R^1_{22}(\cdot)\end{pmatrix} \equiv-\begin{pmatrix}R^2_{11}(\cdot)&R^2_{12}(\cdot) \\ R^2_{21}(\cdot)&R^2_{22}(\cdot)\end{pmatrix} \equiv\begin{pmatrix}R_{11}(\cdot)&R_{12}(\cdot) \\ R_{21}(\cdot)&R_{22}(\cdot)\end{pmatrix} \equiv\begin{pmatrix}R_1(\cdot) \\ R_2(\cdot)\end{pmatrix}\equiv R(\cdot),\\ \noalign{\smallskip}\displaystyle\begin{pmatrix}S^1_1(\cdot) \\ S^1_2(\cdot)\end{pmatrix}=-\begin{pmatrix}S^2_1(\cdot) \\ S^2_2(\cdot)\end{pmatrix} \equiv\begin{pmatrix}S_1(\cdot) \\ S_2(\cdot)\end{pmatrix}\equiv S(\cdot), \quad\begin{pmatrix}\rho^1_1(\cdot) \\ \rho^1_2(\cdot)\end{pmatrix} =-\begin{pmatrix}\rho^2_1(\cdot) \\ \rho^2_2(\cdot)\end{pmatrix} \equiv\begin{pmatrix}\rho_1(\cdot) \\ \rho_2(\cdot)\end{pmatrix}\equiv\rho(\cdot), \end{array}\end{equation} and $$\begin{array}{ll} \displaystyle J^1(t,x;u_1(\cdot),u_2(\cdot))=-J^2(t,x;u_1(\cdot),u_2(\cdot))\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\Big\{\langle GX(T),X(T)\rangle+2\langle g,X(T)\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+\int_t^T\[\langle {\scriptstyle\begin{pmatrix}\scriptstyle Q(s) &\negthinspace \scriptstyle S_1(s)^\top &\negthinspace \scriptstyle S_2(s)^\top\\ \scriptstyle S_1(s) &\negthinspace \scriptstyle R_{11}(s) &\negthinspace \scriptstyle R_{12}(s) \\ \scriptstyle S_2(s) &\negthinspace \scriptstyle R_{21}(s) &\negthinspace \scriptstyle R_{22}(s) \end{pmatrix}} {\scriptstyle\begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u_1(s) \\ \scriptstyle u_2(s)\end{pmatrix}}, {\scriptstyle\begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u_1(s) \\ \scriptstyle u_2(s)\end{pmatrix}}\rangle +2\langle {\scriptstyle\begin{pmatrix}\scriptstyle q(s) \\ \scriptstyle\rho_1(s) \\ \scriptstyle\rho_2(s)\end{pmatrix}, \begin{pmatrix}\scriptstyle X(s) \\ \scriptstyle u_1(s) \\ \scriptstyle u_2(s) \end{pmatrix}}\rangle\]ds\Big\}\\ \noalign{\smallskip}\displaystyle\equiv J(t,x;u_1(\cdot),u_2(\cdot)). \end{array}$$ Let $(\Theta(\cdot),v(\cdot))\in{\cal Q}[t,T]\times{\cal U}[t,T]$ and assume the open-loop saddle points of Problem {\rm(SDG)} on $[t,T]$ admit the closed-loop representation \eqref{16June29-17:40}. The equations \eqref{16June30-Pi} $(i=1,2)$ for $\Pi_1(\cdot)$ and $\Pi_2(\cdot)$ now become $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi_1+\Pi_1 A+A^\top\Pi_1+C^\top\Pi_1 C+Q+\big(\Pi_1 B+C^\top\Pi_1 D+S^\top\big)\Theta=0,\\ \noalign{\smallskip}\displaystyle\Pi_1(T)=G,\end{array}\right.$$ and $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi_2+\Pi_2 A+A^\top\Pi_2+C^\top\Pi_2 C-Q+\big(\Pi_2 B+C^\top\Pi_2 D-S^\top\big)\Theta=0,\\ \noalign{\smallskip}\displaystyle\Pi_2(T)=-G,\end{array}\right.$$ respectively. Obviously, both $\Pi_1(\cdot)$ and $-\Pi_2(\cdot)$ satisfy \begin{equation}\label{16June30-17:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi+\Pi A+A^\top\Pi+C^\top\Pi C+Q+\big(\Pi B+C^\top\Pi D+S^\top\big)\Theta=0,\\ \noalign{\smallskip}\displaystyle\Pi(T)=G.\end{array}\right.\end{equation} Thus, $\Pi_1(\cdot)=-\Pi_2(\cdot)\equiv\Pi(\cdot)$, and \eqref{16June30-Pi-yueshu} becomes $$\begin{pmatrix}R_1+D_1^\top\Pi D \\ -R_2-D_2^\top\Pi D\end{pmatrix}\Theta +\begin{pmatrix}B_1^\top\Pi+D_1^\top\Pi C+S_1 \\-B_2^\top\Pi-D_2^\top\Pi C-S_2\end{pmatrix}=0,$$ or equivalently, $$(R+D^\top\Pi D)\Theta+B^\top\Pi+D^\top\Pi C+S=0.$$ This is also equivalent to \begin{equation}\label{16June30-16:43}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \mathscr{R}(B^\top\Pi+D^\top\Pi C+S)\subseteq\mathscr{R}(R+D^\top\Pi D),\qquad\hbox{\rm a.e.}~s\in[t,T],\\ \noalign{\smallskip}\displaystyle (R+D^\top\Pi D)^\dag(B^\top\Pi+D^\top\Pi C+S)\in L^2(t,T;\mathbb{R}^{m\times n}), \end{array}\right.\end{equation} and \begin{equation}\label{16June30-16:40}\Theta=-(R+D^\top\Pi D)^\dag(B^\top\Pi+D^\top\Pi C+S) +[I-(R+D^\top\Pi D)^\dag(R+D^\top\Pi D)]\theta,\end{equation} for some $\theta(\cdot)\in L^2(t,T;\mathbb{R}^{m\times n})$. Upon substitution of \eqref{16June30-16:40} into \eqref{16June30-17:00}, the latter becomes \begin{equation}\label{16June30-20:00}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle\dot\Pi+\Pi A+A^\top\Pi+C^\top\Pi C+Q\\ \noalign{\smallskip}\displaystyle\quad-\,(\Pi B+C^\top\Pi D+S^\top)(R+D^\top\Pi D)^\dag(B^\top\Pi+D^\top\Pi C+S)=0,\quad~ s\in[t,T],\\ \noalign{\smallskip}\displaystyle\Pi(T)=G,\end{array}\right.\end{equation} with constraints \eqref{16June30-16:43}. Note that equation \eqref{16June30-20:00} is symmetric. Likewise, we have $(\eta_1(\cdot),\zeta_1(\cdot))=-(\eta_2(\cdot),\zeta_2(\cdot))\equiv(\eta_\Pi(\cdot),\zeta_\Pi(\cdot))$ satisfying \begin{equation}\label{16June30-20:14}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\eta_\Pi=-\Big\{\big[A^\top\negthinspace -(\Pi B+C^\top\Pi D+S^\top)(R+D^\top\Pi D)^\dag B^\top\big]\eta_\Pi\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big[C^\top\negthinspace -(\Pi B+C^\top\Pi D+S^\top)(R+D^\top\Pi D)^\dag D^\top\big]\zeta_\Pi\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big[C^\top\negthinspace -(\Pi B+C^\top\Pi D+S^\top)(R+D^\top\Pi D)^\dag D^\top\big]\Pi\sigma\\ \noalign{\smallskip}\displaystyle\qquad\qq~-(\Pi B+C^\top\Pi D+S^\top)(R+D^\top\Pi D)^\dag\rho+\Pi b+q\Big\}ds+\zeta_\Pi dW,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle\eta_\Pi(T)=g,\end{array}\right.\end{equation} with constraints \begin{equation}\label{16June30-20:16}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle B^\top\eta_\Pi+D^\top\zeta_\Pi+D^\top\Pi\sigma+\rho\in\mathscr{R}(R+D^\top\Pi D),\qquad\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\\ \noalign{\smallskip}\displaystyle (R+D^\top\Pi D)^\dag(B^\top\eta_\Pi+D^\top\zeta_\Pi+D^\top\Pi\sigma+\rho)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m), \end{array}\right.\end{equation} and in this case, $$v=-(R+D^\top\Pi D)^\dag(B^\top\eta_\Pi+D^\top\zeta_\Pi+D^\top\Pi\sigma+\rho) +\big[I-(R+D^\top\Pi D)^\dag(R+D^\top\Pi D)\big]\nu,$$ for some $\nu(\cdot)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m)$. To summarize, we have the following result for LQ stochastic two-person zero-sum differential games. \begin{theorem}\label{bt-16June30-18:00}\sl Let {\rm(G1)--(G2)} and \eqref{16July1-16:00} hold. Then the open-loop saddle points of Problem {\rm(SDG)} on $[t,T]$ admit a closed-loop representation if and only if the following hold: \medskip {\rm(i)} The following convexity-concavity condition holds: For $i=1,2$, \begin{equation}\label{16July1-17:28}\begin{array}{ll} \displaystyle(-1)^{i-1}\mathbb{E}\Big\{\int_t^T\[\langle Q(s)X_i(s),X_i(s)\rangle+2\langle S_i(s)X_i(s),u_i(s)\rangle +\langle R_{ii}(s)u_i(s),u_i(s)\rangle\]ds\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\qq+\,\langle GX_i(T),X_i(T)\rangle\Big\}\ges0,\qquad\forall\,u_i(\cdot)\in{\cal U}_i[t,T],\end{array}\end{equation} where $X_i(\cdot)$ is the solution to FSDE \eqref{homogeneous}. \medskip {\rm(ii)} The Riccati equation \eqref{16June30-20:00} admits a solution $\Pi(\cdot)\in C([t,T];\mathbb{S}^n)$ such that \eqref{16June30-16:43} holds, and the adapted solution of \eqref{16June30-20:14} satisfies \eqref{16June30-20:16}. \medskip In the above case, all the closed-loop representations of open-loop saddle points are given by $$\begin{array}{lll} \displaystyle u\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle \Big\{\negthinspace -(R+D^\top\Pi D)^\dag(B^\top\Pi+D^\top\Pi C+S) +\big[I-(R+D^\top\Pi D)^\dag(R+D^\top\Pi D)\big]\theta\Big\}X\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &~&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle-\,(R+D^\top\Pi D)^\dag(B^\top\eta_\Pi+D^\top\zeta_\Pi+D^\top\Pi\sigma+\rho) +\big[I-(R+D^\top\Pi D)^\dag(R+D^\top\Pi D)\big]\nu, \end{array}$$ where $\theta(\cdot)\in L^2(t,T;\mathbb{R}^{m\times n})$ and $\nu(\cdot)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m)$. \end{theorem} \it Proof. \rm The result can be proved by combining Theorem \ref{bt-16June29-17:30} and the previous argument. We leave the details to the interested reader. \signed {$\sqr69$} \medskip Now let us recall from \cite{Sun-Yong 2014} the characterization of closed-loop saddle points of LQ stochastic two-person zero-sum differential games. \begin{theorem}\label{bt-16July1-15:00} \sl Let {\rm(G1)--(G2)} and \eqref{16July1-16:00} hold. Then Problem {\rm(SDG)} admits a closed-loop saddle point on $[t,T]$ if and only if the following hold: \medskip {\rm(i)} The Riccati equation \begin{equation}\label{16July1-Ric}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \dot P+PA+A^\top P+C^\top PC+Q\\ \noalign{\smallskip}\displaystyle\quad-\,(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag(B^\top P+D^\top PC+S)=0,\quad~s\in[t,T],\\ \noalign{\smallskip}\displaystyle P(T)=G,\end{array}\right.\end{equation} admits a solution $P(\cdot)\in C([t,T];\mathbb{S}^n)$ such that the following hold: \begin{eqnarray} &\label{16July1-yushu*}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle \mathscr{R}(B^\top P+D^\top PC+S)\subseteq\mathscr{R}(R+D^\top PD),\qquad\hbox{\rm a.e.}~s\in[t,T],\\ \noalign{\smallskip}\displaystyle (R+D^\top PD)^\dag(B^\top P+D^\top PC+S)\in L^2(t,T;\mathbb{R}^{m\times n}), \end{array}\right.&\\ \noalign{\smallskip}&\label{16July1-yushu**}R_{11}+D_1^\top PD_1\ges0,\qquad R_{22}+D_2^\top PD_2\les0,\qquad\hbox{\rm a.e.}~s\in[t,T].& \end{eqnarray} {\rm(ii)} The adapted solution $(\eta_P(\cdot),\zeta_P(\cdot))$ of the BSDE on $[t,T]$ \begin{equation}\label{16July2-eta_P}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\eta_P=-\Big\{\big[A^\top\negthinspace -(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag B^\top\big]\eta_P\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big[C^\top\negthinspace -(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag D^\top\big]\zeta_P\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big[C^\top\negthinspace -(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag D^\top\big]P\sigma\\ \noalign{\smallskip}\displaystyle\qquad\qq~-(PB+C^\top PD+S^\top)(R+D^\top PD)^\dag\rho+Pb+q\Big\}ds+\zeta_P dW,\\ \noalign{\smallskip}\displaystyle\eta_P(T)=g,\end{array}\right.\end{equation} satisfies \begin{equation}\label{}\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle B^\top\eta_P+D^\top\zeta_P+D^\top P\sigma+\rho\in\mathscr{R}(R+D^\top PD),\qquad\hbox{\rm a.e.}~s\in[t,T],~\hbox{\rm a.s.}\\ \noalign{\smallskip}\displaystyle (R+D^\top PD)^\dag(B^\top\eta_P+D^\top\zeta_P+D^\top P\sigma+\rho)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m). \end{array}\right.\end{equation} In this case, the closed-loop saddle point $(\Theta^*(\cdot),v^*(\cdot))$ admits the following representation: \begin{equation}\label{16July2-15:30}\left\{\negthinspace \negthinspace \begin{array}{cll} \displaystyle\Theta^*\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle -(R+D^\top PD)^\dag(B^\top P+D^\top PC+S) +\big[I-(R+D^\top PD)^\dag(R+D^\top PD)\big]\theta,\\ \noalign{\smallskip} v^*\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle -(R+D^\top PD)^\dag(B^\top\eta_P+D^\top\zeta_P+D^\top P\sigma+\rho) +\big[I-(R+D^\top PD)^\dag(R+D^\top PD)\big]\nu, \end{array}\right.\end{equation} where $\theta(\cdot)\in L^2(t,T;\mathbb{R}^{m\times n})$ and $\nu(\cdot)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m)$. \end{theorem} Comparing Theorems \ref{bt-16June30-18:00} and \ref{bt-16July1-15:00}, one may ask: For LQ stochastic two-person zero-sum differential games, when both the closed-loop representation of open-loop saddle points and the closed-loop saddle point exist, does the closed-loop representation coincide with the outcome of the closed-loop saddle point? The answer to this question is affirmative, as shown by the following result. \begin{theorem}\label{bt-16July1-17:00}\sl Let {\rm(G1)--(G2)} and \eqref{16July1-16:00} hold. If both the closed-loop representation of open-loop saddle points and the closed-loop saddle point exist on $[t,T]$, then the closed-loop representation coincides with the outcome of the closed-loop saddle point. \end{theorem} \it Proof. \rm The proof is immediate from Theorems \ref{bt-16June30-18:00} and \ref{bt-16July1-15:00}, once we show that the solution $\Pi(\cdot)$ to the Riccati equation \eqref{16June30-20:00} with constraints \eqref{16June30-16:43} coincides with the solution $P(\cdot)$ to \eqref{16July1-Ric} with constraints \eqref{16July1-yushu*}--\eqref{16July1-yushu**}. \medskip First, we note that if the convexity-concavity condition \eqref{16July1-17:28} holds for initial time $t$, it also holds for any $t^\prime\in[t,T]$. Indeed, for any $t^\prime\in[t,T]$, and any $u_1(\cdot)\in{\cal U}_1[t^\prime,T]$, let $X_1(\cdot)$ be the solution to $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX_1(s)=\big[A(s)X_1(s)+B_1(s)u_1(s)\big]ds+\big[C(s)X_1(s)+D_1(s)u_1(s)\big]dW(s),\quad~s\in[t^\prime,T], \\ \noalign{\smallskip}\displaystyle X_1(t^\prime)=0,\end{array}\right.$$ and define the {\it zero-extension} of $u_1(\cdot)$ as follows: $$[\,0I_{[t,t^\prime)}\oplus u_1](s)=\left\{\negthinspace \negthinspace \begin{array}{ll}0,& s\in[t,t^\prime),\\ \noalign{\smallskip}\displaystyle u_1(s),& s\in[t^\prime,T].\end{array}\right.$$ Then $\tilde{u}_1(\cdot)\equiv[\,0I_{[t,t^\prime)}\oplus u_1](\cdot)\in{\cal U}_1[t,T]$, and due to the initial state being 0, the solution $\widetilde X_1(s)$ of $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle d\widetilde X_1(s)=\big[A(s)\widetilde X_1(s)+B_1(s)\tilde{u}_1(s)\big]ds+\big[C(s)\widetilde X_1(s)+D_1(s)\tilde{u}_1(s)\big]dW(s),\quad s\in[t,T],\\ \noalign{\smallskip}\displaystyle \widetilde X_1(t)=0,\end{array}\right.$$ satisfies $$\widetilde X_1(s)=\left\{\negthinspace \negthinspace \begin{array}{ll}0,& s\in[t,t^\prime),\\ \noalign{\smallskip}\displaystyle X_1(s),& s\in[t^\prime,T].\end{array}\right.$$ Hence, $$\begin{array}{ll} \displaystyle\mathbb{E}\Big\{\int_{t^\prime}^T\[\langle QX_1,X_1\rangle+2\langle S_1X_1,u_1\rangle+\langle R_{11}u_1,u_1\rangle\]ds +\langle GX_1(T),X_1(T)\rangle\Big\}\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\Big\{\int_t^T\[\big\langle Q\widetilde X_1,\widetilde X_1\big\rangle+2\big\langle S_1\widetilde X_1,\tilde{u}_1\big\rangle +\big\langle R_{11}\tilde{u}_1,\tilde{u}_1\big\rangle\]ds+\big\langle G\widetilde X_1(T),\widetilde X_1(T)\big\rangle\Big\}\ges0. \end{array}$$ This proves the case $i=1$. The case $i=2$ can be treated similarly. \smallskip Now let $(\Theta^*(\cdot),v^*(\cdot))$ be a closed-loop saddle point of Problem (SDG) on $[t,T]$. Under the assumption of the theorem, it is clear from Theorem \ref{bt-16June30-18:00} that for any initial pair $(t^\prime,x)$ with $t^\prime\in[t,T]$, the outcome $$u^*(s)=\Theta^*(s)X^*(s)+v^*(s),\qquad s\in[t^\prime,T]$$ of $(\Theta^*(\cdot),v^*(\cdot))$ is an open-loop saddle point for $(t^\prime,x)$, where $X^*(\cdot)$ is the solution to $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX^*(s)=\big\{[A(s)+B(s)\Theta^*(s)]X^*(s)+B(s)v^*(s)+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\{[C(s)+D(s)\Theta^*(s)]X^*(s)+D(s)v^*(s)+\sigma(s)\big\}dW(s),\qquad s\in[t^\prime,T],\\ \noalign{\smallskip}\displaystyle X^*(t^\prime)= x. \end{array}\right.$$ By Theorem \ref{bt-16July1-15:00}, $(\Theta^*(\cdot),v^*(\cdot))$ admits the representation \eqref{16July2-15:30}, and a straightforward calculation shows that $$\begin{array}{ll} \displaystyle\dot P+P(A+B\Theta^*)+(A+B\Theta^*)^\top\negthinspace P+(C+D\Theta^*)^\top\negthinspace P(C+D\Theta^*)\\ \noalign{\smallskip}\displaystyle\quad+\,(\Theta^*)^\top R\Theta^*+S^\top\Theta^*+(\Theta^*)^\top S+Q=0,\end{array}$$ and that the adapted solution $(\eta_P(\cdot),\zeta_P(\cdot))$ of \eqref{16July2-eta_P} satisfies $$ d\eta_P=-\big[(A+B\Theta^*)^\top\eta_P+(C+D\Theta^*)^\top\zeta_P+(C+D\Theta^*)^\top P\sigma +(\Theta^*)^\top\rho+Pb+q\big]ds+\zeta_P dW.$$ Then applying It\^o's formula to $s\mapsto\langle P(s)X^*(s),X^*(s)\rangle+2\langle \eta_P(s),X^*(s)\rangle$ and noting that $$(R+D^\top PD)\Theta^*+B^\top P+D^\top PC+S=0,$$ we have \begin{equation}\label{16July2-17:44}\begin{array}{ll} \displaystyle J(t^\prime,x;u^*(\cdot))=J(t^\prime,x;\Theta^*(\cdot)X^*(\cdot)+v^*(\cdot))\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\Big\{\langle GX^*(T),X^*(T)\rangle+2\langle g,X^*(T)\rangle+\int_{t^\prime}^T\[\langle QX^*,X^*\rangle+2\langle SX^*,\Theta^*X^*+v^*\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+\langle R(\Theta^*X^*+v^*),\Theta^*X^*+v^*\rangle+2\langle q,X^*\rangle+2\langle\rho,\Theta^*X^*+v^*\rangle\]ds\Big\}\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\Big\{\langle P(t^\prime)x,x\rangle+2\langle\eta_P(t^\prime),x\rangle+\int_{t^\prime}^T\[\langle\dot PX^*,X^*\rangle+2\langle PX^*,(A+B\Theta^*)X^*+Bv^*+b\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+\langle P[(C+D\Theta^*)X^*+Dv^*+\sigma],(C+D\Theta^*)X^*+Dv^*+\sigma\rangle\\ \noalign{\smallskip}\displaystyle\qquad~-2\big\langle(A+B\Theta^*)^\top\eta_P+(C+D\Theta^*)^\top\zeta_P+(C+D\Theta^*)^\top P\sigma+(\Theta^*)^\top\rho+Pb+q,X^*\big\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+2\langle\eta_P,(A+B\Theta^*)X^*+Bv^*+b\rangle+2\langle\zeta_P,(C+D\Theta^*)X^*+Dv^*+\sigma\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+\big\langle\big[Q+S^\top\Theta^*+(\Theta^*)^\top S+(\Theta^*)^\top R\Theta^*\big]X^*,X^*\big\rangle+2\langle (R\Theta^*+S)X^*,v^*\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+2\big\langle q+(\Theta^*)^\top\rho,X^*\big\rangle+\langle Rv^*,v^*\rangle+2\langle\rho,v^*\rangle\]ds\Big\}\\ \noalign{\smallskip}\displaystyle=\mathbb{E}\Big\{\langle P(t^\prime)x,x\rangle+2\langle\eta_P(t^\prime),x\rangle+\int_{t^\prime}^T\[\langle P\sigma,\sigma\rangle+2\langle\eta_P,b\rangle+2\langle\zeta_P,\sigma\rangle\\ \noalign{\smallskip}\displaystyle\qquad~+\big\langle(R+D^\top PD)v^*,v^*\big\rangle+2\big\langle B^\top\eta_P+D^\top\zeta_P+D^\top P\sigma+\rho,v^*\big\rangle\]ds\Big\}. \end{array}\end{equation} \smallskip Next, let $\theta(\cdot)\in L^2(t,T;\mathbb{R}^{m\times n}), \nu(\cdot)\in L_\mathbb{F}^2(t,T;\mathbb{R}^m)$ and denote $$\left\{\negthinspace \negthinspace \begin{array}{cll} \displaystyle \Theta\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle -(R+D^\top\Pi D)^\dag(B^\top\Pi+D^\top\Pi C+S) +[I-(R+D^\top\Pi D)^\dag(R+D^\top\Pi D)]\theta,\\ \noalign{\smallskip} v\negthinspace \negthinspace \negthinspace \negthinspace &=&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle -(R+D^\top\Pi D)^\dag(B^\top\eta_\Pi+D^\top\zeta_\Pi+D^\top\Pi\sigma+\rho) +[I-(R+D^\top\Pi D)^\dag(R+D^\top\Pi D)]\nu. \end{array}\right.$$ For any initial pair $(t^\prime,x)$ with $t^\prime\in[t,T]$, define $u(\cdot)\in{\cal U}[t^\prime,T]$ by $$u(s)=\Theta(s)X(s)+v(s),\qquad s\in[t^\prime,T],$$ with $X(\cdot)$ being the solution to $$\left\{\negthinspace \negthinspace \begin{array}{ll} \displaystyle dX(s)=\big\{[A(s)+B(s)\Theta(s)]X(s)+B(s)v(s)+b(s)\big\}ds\\ \noalign{\smallskip}\displaystyle\qquad\qq~+\big\{[C(s)+D(s)\Theta(s)]X(s)+D(s)v(s)+\sigma(s)\big\}dW(s),\qquad s\in[t^\prime,T],\\ \noalign{\smallskip}\displaystyle X(t^\prime)= x. \end{array}\right.$$ By Theorem \ref{bt-16June30-18:00}, $u(\cdot)$ is an open-loop saddle point for $(t^\prime,x)$, and by a computation similar to \eqref{16July2-17:44}, we obtain \begin{equation}\label{16July2-18:15}\begin{array}{ll} \displaystyle J(t^\prime,x;u(\cdot))=\mathbb{E}\Big\{\langle\Pi(t^\prime)x,x\rangle+2\langle\eta_\Pi(t^\prime),x\rangle +\int_{t^\prime}^T\[\langle\Pi\sigma,\sigma\rangle+2\langle\eta_\Pi,b\rangle+2\langle\zeta_\Pi,\sigma\rangle\\ \noalign{\smallskip}\displaystyle\qquad\qq\qquad\quad~~+\big\langle(R+D^\top\Pi D)v,v\big\rangle+2\big\langle B^\top\eta_\Pi+D^\top\zeta_\Pi+D^\top\Pi\sigma+\rho,v\big\rangle\]ds\Big\}. \end{array}\end{equation} Since both $u^*(\cdot)\equiv(u_1^*(\cdot)^\top,u_2^*(\cdot)^\top)^\top$ and $u(\cdot)\equiv(u_1(\cdot)^\top,u_2(\cdot)^\top)^\top$ are open-loop saddle points for $(t^\prime,x)$, we have $$\begin{array}{lll} \displaystyle J(t^\prime,x;u_1^*(\cdot),u_2^*(\cdot))\negthinspace \negthinspace \negthinspace \negthinspace &\leqslant&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle J(t^\prime,x;u_1(\cdot),u^*_2(\cdot))\leqslant J(t^\prime,x;u_1(\cdot),u_2(\cdot))\\ \noalign{\smallskip}\negthinspace \negthinspace \negthinspace \negthinspace &\leqslant&\negthinspace \negthinspace \negthinspace \negthinspace \displaystyle J(t^\prime,x;u^*_1(\cdot),u_2(\cdot))\leqslant J(t^\prime,x;u_1^*(\cdot),u_2^*(\cdot)).\end{array}$$ Therefore, $J(t^\prime,x;u^*(\cdot))=J(t^\prime,x;u(\cdot))$ for all $(t^\prime,x)$ with $t^\prime\in[t,T]$, which, together with \eqref{16July2-17:44} and \eqref{16July2-18:15}, yields $\Pi(\cdot)=P(\cdot)$. \signed {$\sqr69$} \medskip Finally, we have the following corollary for Problem (SLQ), which should be but has not been stated in \cite{Sun-Li-Yong 2016}. \begin{corollary}\label{bc-16July11-15:30}\sl For Problem {\rm(SLQ)}, if the open-loop optimal controls admit a closed-loop representation, then every open-loop optimal control must be an outcome of a closed-loop optimal strategy. \end{corollary}
8a749c78a0fac3cfd39a257c74f9c55d3c1664b5
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:introduction} Let $G$ be a simple graph without isolated vertices. If the edge set of a simple graph $K$ can be partitioned into edge sets of graphs each isomorphic to $G$, we say that there exists a {\em decomposition} of $K$ into $G$. In the case where $K$ is the complete graph $K_n$ we refer to the decomposition as a $G$ {\em design} of order $n$. The {\em spectrum} of $G$ is the set of positive integers $n$ for which there exists a $G$ design of order $n$. We refer the reader to the survey article of Adams, Bryant and Buchanan, \cite{ABB} and, for more up to date results, the Web site maintained by Bryant and McCourt, \cite{BM}. A {\em snark} is a connected, bridgeless 3-regular graph with chromatic index 4. However, a snark is usually regarded as trivial (or reducible) if it has girth less than 5 or if it has three edges the deletion of which results in a disconnected graph each of whose components is non-trivial (as a graph). The spectra for non-trivial snarks of up to 22 vertices have been successfully determined, as follows. \begin{enumerate} \item{The smallest non-trivial snark is the 10-vertex Petersen graph for which designs of order $n$ exist if and only if $n \equiv 1$ or 10 (mod 15), \cite{AB-pet}.} \item{There are no non-trivial snarks on 12, 14 or 16 vertices.} \item{There are two non-trivial snarks on 18 vertices, namely the $(1,1)$- and $(1,2)$-Blanu\v{s}a snarks. For each of them, designs of order $n$ exist if and only if $n \equiv 1$ (mod 27), \cite{For-snark}.} \item{For each of the six 20-vertex non-trivial snarks (including the flower snark J5), designs of order $n$ exist if and only if $n \equiv 1,$ 16, 25, 40 (mod 60), $n \neq 16$, \cite{For-snark}.} \item{For each of the twenty 22-vertex non-trivial snarks (including the two Loupekine snarks), designs of order $n$ exist if and only if $n \equiv 1$ or 22 (mod 33), \cite{For-snark}.} \end{enumerate} The purpose of this paper is to extend these results to 24-vertex non-trivial snarks of which there are precisely 38, \cite{BGHM}. We prove the following. {\theorem \label{thm:snark24} Designs of order $n$ exist for each of the thirty-eight non-trivial snarks on $24$ vertices if and only if $n \equiv 1$ or $64~(\mathrm{mod}~72)$.}\\ \noindent The methods use to obtain results like Theorems~\ref{thm:snark24} are explained in \cite{For-snark}, to which the reader should regard this paper as a sequel. The main tool is \cite[Proposition 1.8]{For-snark}, repeated here for convenience as Proposition~\ref{prop:d=3, v=24}. {\proposition \label{prop:d=3, v=24} Let $G$ be a 3-regular graph on $24$ vertices. Suppose there exist $G$ designs of order $64$, $73$, $136$ and $145$. Suppose also that there exist decompositions into $G$ of the complete multipartite graphs $K_{12,12,12}$, $K_{24,24,15}$, $K_{72,72,63}$, $K_{24,24,24,24}$ and $K_{24,24,24,21}$. Then there exists a $G$ design of order $n$ if and only if $n \equiv 1$ or $64~(\textup{mod}~ 72)$.}\\ \noindent\textbf{Proof.} See \cite{For-snark}.\eproof\\ To prove Theorem~\ref{thm:snark24} it is sufficient, therefore, to construct the four designs and five multipartite graph decompositions required by Proposition~\ref{prop:d=3, v=24} for each of the 38 non-trivial snarks. To save space the details of the constructions are given only for designs of order 136, which, in the author's opinion, were by far the most difficult to obtain. The details of the remaining constructions are given in the Appendix, which is present only in the full version of this paper, available by request from the author. The decompositions were obtained and checked by computer programs in the same manner as those of \cite{For-snark}. \section{The 38 non-trivial snarks on 24 vertices} \label{sec:Snark24} \newcommand{\adfGa}{\mathrm{G1}} \newcommand{\adfGb}{\mathrm{G2}} \newcommand{\adfGc}{\mathrm{G3}} \newcommand{\adfGd}{\mathrm{G4}} \newcommand{\adfGe}{\mathrm{G5}} \newcommand{\adfGf}{\mathrm{G6}} \newcommand{\adfGg}{\mathrm{G7}} \newcommand{\adfGh}{\mathrm{G8}} \newcommand{\adfGi}{\mathrm{G9}} \newcommand{\adfGj}{\mathrm{G10}} \newcommand{\adfGk}{\mathrm{G11}} \newcommand{\adfGl}{\mathrm{G12}} \newcommand{\adfGm}{\mathrm{G13}} \newcommand{\adfGn}{\mathrm{G14}} \newcommand{\adfGo}{\mathrm{G15}} \newcommand{\adfGp}{\mathrm{G16}} \newcommand{\adfGq}{\mathrm{G17}} \newcommand{\adfGr}{\mathrm{G18}} \newcommand{\adfGs}{\mathrm{G19}} \newcommand{\adfGt}{\mathrm{G20}} \newcommand{\adfGu}{\mathrm{G21}} \newcommand{\adfGv}{\mathrm{G22}} \newcommand{\adfGw}{\mathrm{G23}} \newcommand{\adfGx}{\mathrm{G24}} \newcommand{\adfGy}{\mathrm{G25}} \newcommand{\adfGz}{\mathrm{G26}} \newcommand{\adfGA}{\mathrm{G27}} \newcommand{\adfGB}{\mathrm{G28}} \newcommand{\adfGC}{\mathrm{G29}} \newcommand{\adfGD}{\mathrm{G30}} \newcommand{\adfGE}{\mathrm{G31}} \newcommand{\adfGF}{\mathrm{G32}} \newcommand{\adfGG}{\mathrm{G33}} \newcommand{\adfGH}{\mathrm{G34}} \newcommand{\adfGI}{\mathrm{G35}} \newcommand{\adfGJ}{\mathrm{G36}} \newcommand{\adfGK}{\mathrm{G37}} \newcommand{\adfGL}{\mathrm{G38}} The thirty-eight non-trivial snarks on 24 vertices are represented by ordered 24-tuples of vertices: (1, 2, \dots, 24)${}_{\adfGa}$, (1, 2, \dots, 24)${}_{\adfGb}$, \dots, (1, 2, \dots, 24)${}_{\adfGL}$. Their edge sets, as appearing in Royale's list, \cite{Roy}, are respectively $\adfGa$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,11\}$, $\{10,12\}$, $\{11,15\}$, $\{12,16\}$, $\{13,14\}$, $\{13,17\}$, $\{14,18\}$, $\{15,16\}$, $\{15,19\}$, $\{16,20\}$, $\{17,21\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGb$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,11\}$, $\{10,15\}$, $\{11,16\}$, $\{12,15\}$, $\{12,16\}$, $\{13,14\}$, $\{13,17\}$, $\{14,18\}$, $\{15,19\}$, $\{16,20\}$, $\{17,21\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGc$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,11\}$, $\{10,15\}$, $\{11,16\}$, $\{12,15\}$, $\{12,16\}$, $\{13,17\}$, $\{13,18\}$, $\{14,19\}$, $\{14,20\}$, $\{15,21\}$, $\{16,22\}$, $\{17,19\}$, $\{17,23\}$, $\{18,20\}$, $\{18,24\}$, $\{19,24\}$, $\{20,23\}$, $\{21,22\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGd$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,11\}$, $\{10,15\}$, $\{11,16\}$, $\{12,15\}$, $\{12,17\}$, $\{13,18\}$, $\{13,19\}$, $\{14,20\}$, $\{14,21\}$, $\{15,22\}$, $\{16,17\}$, $\{16,22\}$, $\{17,23\}$, $\{18,20\}$, $\{18,23\}$, $\{19,21\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGe$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,12\}$, $\{10,15\}$, $\{11,12\}$, $\{11,16\}$, $\{13,14\}$, $\{13,17\}$, $\{14,18\}$, $\{15,16\}$, $\{15,19\}$, $\{16,20\}$, $\{17,21\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGf$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,12\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,16\}$, $\{13,14\}$, $\{13,17\}$, $\{14,18\}$, $\{15,19\}$, $\{16,20\}$, $\{17,21\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGg$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,12\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,16\}$, $\{13,17\}$, $\{13,18\}$, $\{14,19\}$, $\{14,20\}$, $\{15,21\}$, $\{16,22\}$, $\{17,19\}$, $\{17,23\}$, $\{18,20\}$, $\{18,24\}$, $\{19,24\}$, $\{20,23\}$, $\{21,22\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGh$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,12\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,17\}$, $\{13,18\}$, $\{13,19\}$, $\{14,20\}$, $\{14,21\}$, $\{15,22\}$, $\{16,17\}$, $\{16,23\}$, $\{17,22\}$, $\{18,20\}$, $\{18,23\}$, $\{19,21\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGi$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,12\}$, $\{10,15\}$, $\{11,16\}$, $\{11,17\}$, $\{12,16\}$, $\{13,18\}$, $\{13,19\}$, $\{14,20\}$, $\{14,21\}$, $\{15,17\}$, $\{15,22\}$, $\{16,22\}$, $\{17,23\}$, $\{18,20\}$, $\{18,23\}$, $\{19,21\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGj$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{10,16\}$, $\{11,15\}$, $\{11,17\}$, $\{12,15\}$, $\{12,18\}$, $\{13,19\}$, $\{13,20\}$, $\{14,21\}$, $\{14,22\}$, $\{16,17\}$, $\{16,18\}$, $\{17,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGk$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{10,16\}$, $\{11,15\}$, $\{11,17\}$, $\{12,15\}$, $\{12,18\}$, $\{13,19\}$, $\{13,20\}$, $\{14,21\}$, $\{14,22\}$, $\{16,18\}$, $\{16,23\}$, $\{17,18\}$, $\{17,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGl$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{10,16\}$, $\{11,15\}$, $\{11,17\}$, $\{12,16\}$, $\{12,18\}$, $\{13,19\}$, $\{13,20\}$, $\{14,21\}$, $\{14,22\}$, $\{15,18\}$, $\{16,17\}$, $\{17,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGm$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,8\}$, $\{6,11\}$, $\{7,12\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{10,16\}$, $\{11,17\}$, $\{11,18\}$, $\{12,15\}$, $\{12,17\}$, $\{13,19\}$, $\{13,20\}$, $\{14,21\}$, $\{14,22\}$, $\{15,18\}$, $\{16,17\}$, $\{16,23\}$, $\{18,24\}$, $\{19,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGn$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,11\}$, $\{8,13\}$, $\{9,13\}$, $\{10,14\}$, $\{11,15\}$, $\{12,14\}$, $\{12,16\}$, $\{13,17\}$, $\{14,18\}$, $\{15,17\}$, $\{15,19\}$, $\{16,20\}$, $\{16,21\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,20\}$, $\{19,23\}$, $\{20,24\}$, $\{21,22\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGo$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,11\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{11,14\}$, $\{12,15\}$, $\{12,16\}$, $\{13,17\}$, $\{13,18\}$, $\{14,17\}$, $\{15,19\}$, $\{16,20\}$, $\{16,21\}$, $\{17,22\}$, $\{18,20\}$, $\{18,23\}$, $\{19,23\}$, $\{19,24\}$, $\{20,24\}$, $\{21,22\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGp$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,11\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{11,16\}$, $\{12,15\}$, $\{12,17\}$, $\{13,14\}$, $\{13,18\}$, $\{14,16\}$, $\{15,19\}$, $\{16,20\}$, $\{17,21\}$, $\{17,22\}$, $\{18,21\}$, $\{18,23\}$, $\{19,23\}$, $\{19,24\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGq$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,11\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{11,16\}$, $\{12,15\}$, $\{12,17\}$, $\{13,18\}$, $\{13,19\}$, $\{14,20\}$, $\{14,21\}$, $\{15,16\}$, $\{16,22\}$, $\{17,22\}$, $\{17,23\}$, $\{18,20\}$, $\{18,23\}$, $\{19,21\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGr$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,11\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{11,16\}$, $\{12,15\}$, $\{12,17\}$, $\{13,18\}$, $\{13,19\}$, $\{14,20\}$, $\{14,21\}$, $\{15,22\}$, $\{16,17\}$, $\{16,22\}$, $\{17,23\}$, $\{18,20\}$, $\{18,23\}$, $\{19,21\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGs$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,11\}$, $\{8,13\}$, $\{9,14\}$, $\{10,15\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{13,20\}$, $\{14,16\}$, $\{14,19\}$, $\{15,21\}$, $\{15,22\}$, $\{16,20\}$, $\{17,21\}$, $\{17,23\}$, $\{18,22\}$, $\{18,24\}$, $\{19,23\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGt$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,14\}$, $\{11,15\}$, $\{12,16\}$, $\{12,17\}$, $\{13,18\}$, $\{14,19\}$, $\{15,16\}$, $\{16,20\}$, $\{17,21\}$, $\{17,22\}$, $\{18,21\}$, $\{18,23\}$, $\{19,22\}$, $\{19,24\}$, $\{20,23\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGu$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,14\}$, $\{11,15\}$, $\{12,16\}$, $\{12,17\}$, $\{13,18\}$, $\{14,19\}$, $\{15,20\}$, $\{16,21\}$, $\{16,22\}$, $\{17,23\}$, $\{17,24\}$, $\{18,19\}$, $\{18,21\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGv$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,14\}$, $\{11,16\}$, $\{12,15\}$, $\{12,17\}$, $\{13,18\}$, $\{14,19\}$, $\{15,20\}$, $\{16,21\}$, $\{16,22\}$, $\{17,21\}$, $\{17,23\}$, $\{18,19\}$, $\{18,23\}$, $\{19,24\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGw$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,14\}$, $\{11,16\}$, $\{12,15\}$, $\{12,17\}$, $\{13,18\}$, $\{14,19\}$, $\{15,20\}$, $\{16,21\}$, $\{16,22\}$, $\{17,21\}$, $\{17,23\}$, $\{18,19\}$, $\{18,24\}$, $\{19,23\}$, $\{20,22\}$, $\{20,24\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGx$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,14\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,16\}$, $\{14,19\}$, $\{15,20\}$, $\{15,21\}$, $\{16,22\}$, $\{17,20\}$, $\{17,23\}$, $\{18,21\}$, $\{18,24\}$, $\{19,22\}$, $\{19,23\}$, $\{20,24\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGy$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{14,16\}$, $\{14,20\}$, $\{15,21\}$, $\{16,19\}$, $\{17,20\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,23\}$, $\{20,24\}$, $\{21,22\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGz$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{14,16\}$, $\{14,20\}$, $\{15,21\}$, $\{16,22\}$, $\{17,19\}$, $\{17,23\}$, $\{18,22\}$, $\{18,24\}$, $\{19,20\}$, $\{20,24\}$, $\{21,23\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGA$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{14,17\}$, $\{14,20\}$, $\{15,21\}$, $\{16,20\}$, $\{16,22\}$, $\{17,19\}$, $\{18,22\}$, $\{18,23\}$, $\{19,24\}$, $\{20,23\}$, $\{21,23\}$, $\{21,24\}$, $\{22,24\}$\}, $\adfGB$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,13\}$, $\{10,15\}$, $\{11,15\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{14,17\}$, $\{14,20\}$, $\{15,21\}$, $\{16,20\}$, $\{16,22\}$, $\{17,23\}$, $\{18,22\}$, $\{18,24\}$, $\{19,20\}$, $\{19,24\}$, $\{21,23\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGC$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,13\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,17\}$, $\{14,19\}$, $\{14,20\}$, $\{15,21\}$, $\{15,22\}$, $\{16,23\}$, $\{17,23\}$, $\{18,19\}$, $\{18,21\}$, $\{19,22\}$, $\{20,21\}$, $\{20,24\}$, $\{22,24\}$, $\{23,24\}$\}, $\adfGD$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,13\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{14,15\}$, $\{14,20\}$, $\{15,19\}$, $\{16,21\}$, $\{17,20\}$, $\{17,22\}$, $\{18,23\}$, $\{18,24\}$, $\{19,23\}$, $\{20,24\}$, $\{21,22\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGE$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,13\}$, $\{11,16\}$, $\{12,17\}$, $\{12,18\}$, $\{13,19\}$, $\{14,15\}$, $\{14,20\}$, $\{15,21\}$, $\{16,17\}$, $\{17,22\}$, $\{18,20\}$, $\{18,23\}$, $\{19,23\}$, $\{19,24\}$, $\{20,24\}$, $\{21,22\}$, $\{21,23\}$, $\{22,24\}$\}, $\adfGF$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,13\}$, $\{11,17\}$, $\{12,16\}$, $\{12,18\}$, $\{13,19\}$, $\{14,15\}$, $\{14,20\}$, $\{15,19\}$, $\{16,21\}$, $\{17,22\}$, $\{17,23\}$, $\{18,20\}$, $\{18,22\}$, $\{19,24\}$, $\{20,23\}$, $\{21,23\}$, $\{21,24\}$, $\{22,24\}$\}, $\adfGG$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,13\}$, $\{11,17\}$, $\{12,16\}$, $\{12,18\}$, $\{13,19\}$, $\{14,15\}$, $\{14,20\}$, $\{15,19\}$, $\{16,21\}$, $\{17,22\}$, $\{17,23\}$, $\{18,22\}$, $\{18,24\}$, $\{19,24\}$, $\{20,21\}$, $\{20,22\}$, $\{21,23\}$, $\{23,24\}$\}, $\adfGH$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,13\}$, $\{11,17\}$, $\{12,18\}$, $\{12,19\}$, $\{13,18\}$, $\{14,20\}$, $\{14,21\}$, $\{15,22\}$, $\{15,23\}$, $\{16,17\}$, $\{16,18\}$, $\{17,24\}$, $\{19,20\}$, $\{19,22\}$, $\{20,23\}$, $\{21,22\}$, $\{21,24\}$, $\{23,24\}$\}, $\adfGI$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,16\}$, $\{11,17\}$, $\{12,18\}$, $\{12,19\}$, $\{13,15\}$, $\{13,17\}$, $\{14,18\}$, $\{14,20\}$, $\{15,20\}$, $\{16,21\}$, $\{17,22\}$, $\{18,23\}$, $\{19,22\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGJ$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,10\}$, $\{8,13\}$, $\{8,14\}$, $\{9,15\}$, $\{10,16\}$, $\{11,16\}$, $\{11,17\}$, $\{12,18\}$, $\{12,19\}$, $\{13,15\}$, $\{13,18\}$, $\{14,17\}$, $\{14,20\}$, $\{15,20\}$, $\{16,21\}$, $\{17,22\}$, $\{18,23\}$, $\{19,22\}$, $\{19,24\}$, $\{20,24\}$, $\{21,23\}$, $\{21,24\}$, $\{22,23\}$\}, $\adfGK$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,13\}$, $\{8,14\}$, $\{8,15\}$, $\{9,14\}$, $\{10,16\}$, $\{10,17\}$, $\{11,16\}$, $\{11,18\}$, $\{12,19\}$, $\{12,20\}$, $\{13,18\}$, $\{13,21\}$, $\{14,22\}$, $\{15,19\}$, $\{15,23\}$, $\{16,21\}$, $\{17,18\}$, $\{17,23\}$, $\{19,24\}$, $\{20,22\}$, $\{20,23\}$, $\{21,24\}$, $\{22,24\}$\} and $\adfGL$: \{$\{1,2\}$, $\{1,3\}$, $\{1,4\}$, $\{2,5\}$, $\{2,6\}$, $\{3,7\}$, $\{3,8\}$, $\{4,9\}$, $\{4,10\}$, $\{5,7\}$, $\{5,9\}$, $\{6,11\}$, $\{6,12\}$, $\{7,13\}$, $\{8,14\}$, $\{8,15\}$, $\{9,14\}$, $\{10,16\}$, $\{10,17\}$, $\{11,16\}$, $\{11,18\}$, $\{12,19\}$, $\{12,20\}$, $\{13,18\}$, $\{13,21\}$, $\{14,22\}$, $\{15,19\}$, $\{15,23\}$, $\{16,21\}$, $\{17,18\}$, $\{17,24\}$, $\{19,24\}$, $\{20,22\}$, $\{20,23\}$, $\{21,23\}$, $\{22,24\}$\}. {\lemma \label{lem:Snark24 136} There exist a design of order $136$ for each of the thirty-eight non-trivial 24-vertex snarks.}\\ \noindent\textbf{Proof.} Let the vertex set of $K_{136}$ be $Z_{135} \cup \{\infty\}$. The decompositions consist of $(\infty,1,96,44,95,99,21,54,102,3,90,11,\adfsplit 130,94,5,133,120,31,2,25,77,116,98,9)_{\adfGa}$, $(33,35,88,18,101,9,132,34,58,126,42,114,\adfsplit 120,39,72,119,107,28,70,105,12,133,56,106)_{\adfGa}$, $(29,13,28,12,27,84,31,37,56,95,46,133,\adfsplit 122,10,114,121,130,60,3,4,85,14,41,119)_{\adfGa}$, $(120,47,86,95,25,14,80,29,104,58,26,7,\adfsplit 76,28,97,121,41,33,18,46,70,63,133,4)_{\adfGa}$, $(132,70,3,96,55,5,24,60,59,130,26,56,\adfsplit 37,76,98,16,111,102,109,22,79,47,63,74)_{\adfGa}$, \adfSgap $(97,84,54,95,83,35,119,77,41,9,12,65,\adfsplit 57,120,55,42,126,50,53,62,123,63,98,51)_{\adfGa}$, $(2,102,26,48,105,20,56,50,59,114,22,45,\adfsplit 80,24,9,98,116,77,29,11,74,62,113,122)_{\adfGa}$, \adfLgap $(\infty,71,112,54,34,92,74,91,27,13,99,46,\adfsplit 107,115,123,121,97,25,19,47,31,51,106,14)_{\adfGb}$, $(116,84,89,26,70,52,113,75,125,123,122,82,\adfsplit 132,81,95,43,78,32,97,13,62,20,83,115)_{\adfGb}$, $(30,93,125,4,59,85,79,15,78,23,0,81,\adfsplit 126,106,46,36,82,72,99,103,100,77,43,41)_{\adfGb}$, $(89,126,98,31,28,105,95,10,22,16,110,77,\adfsplit 61,86,107,106,133,60,14,96,101,3,68,79)_{\adfGb}$, $(98,49,63,78,25,76,67,60,117,30,19,70,\adfsplit 22,58,20,6,105,120,59,126,111,83,3,38)_{\adfGb}$, \adfSgap $(131,57,125,42,87,59,96,119,54,29,123,21,\adfsplit 51,84,12,62,120,38,45,128,56,53,36,50)_{\adfGb}$, $(44,90,56,122,26,77,41,27,111,62,60,117,\adfsplit 87,98,50,48,2,83,21,108,80,78,9,14)_{\adfGb}$, \adfLgap $(\infty,31,33,86,82,70,32,55,90,95,1,21,\adfsplit 115,68,16,2,13,91,19,73,39,111,45,134)_{\adfGc}$, $(107,50,31,72,78,116,33,48,42,12,91,30,\adfsplit 90,112,59,18,13,80,122,32,17,98,11,58)_{\adfGc}$, $(38,82,85,67,95,128,78,112,50,97,116,96,\adfsplit 132,13,101,55,36,16,93,83,104,12,76,66)_{\adfGc}$, $(92,68,119,3,76,104,105,60,44,85,54,32,\adfsplit 125,78,72,45,8,86,71,12,34,65,33,102)_{\adfGc}$, $(29,59,126,91,44,94,128,42,77,80,103,70,\adfsplit 119,24,9,114,111,33,10,0,92,130,32,58)_{\adfGc}$, \adfSgap $(35,88,118,1,21,24,97,73,69,100,112,93,\adfsplit 36,117,22,45,82,85,70,127,34,60,124,106)_{\adfGc}$, $(1,60,36,79,97,64,40,10,131,25,111,30,\adfsplit 31,49,15,76,67,90,101,75,115,22,112,19)_{\adfGc}$, \adfLgap $(\infty,75,70,68,60,122,6,33,35,109,71,48,\adfsplit 91,111,13,65,22,63,121,118,42,7,47,126)_{\adfGd}$, $(7,83,59,128,76,43,22,102,14,71,21,124,\adfsplit 8,129,72,98,5,132,52,125,39,110,95,97)_{\adfGd}$, $(79,81,129,16,52,40,29,109,69,98,118,56,\adfsplit 67,47,37,87,48,63,116,44,21,26,50,131)_{\adfGd}$, $(108,81,122,131,105,11,91,50,95,67,20,82,\adfsplit 117,26,17,75,128,73,89,107,106,66,96,45)_{\adfGd}$, $(62,91,38,64,95,83,20,73,126,65,117,3,\adfsplit 17,90,70,30,33,85,130,109,96,81,58,69)_{\adfGd}$, \adfSgap $(127,95,52,91,24,121,6,70,94,79,4,66,\adfsplit 38,51,115,19,1,102,64,84,18,67,0,21)_{\adfGd}$, $(7,4,121,39,124,64,109,85,127,99,13,25,\adfsplit 97,78,24,45,117,118,71,15,0,102,70,103)_{\adfGd}$, \adfLgap $(\infty,93,53,76,43,0,113,49,24,2,34,54,\adfsplit 64,77,6,82,28,21,125,117,96,122,87,71)_{\adfGe}$, $(2,105,34,106,112,78,114,68,10,24,79,63,\adfsplit 23,72,52,82,45,32,64,39,116,1,18,43)_{\adfGe}$, $(125,24,118,77,86,16,92,6,84,78,131,35,\adfsplit 76,23,18,4,117,52,95,60,130,80,29,128)_{\adfGe}$, $(46,112,67,127,26,106,114,131,128,110,61,101,\adfsplit 59,43,91,14,16,19,40,84,111,7,86,93)_{\adfGe}$, $(120,57,101,16,0,52,106,115,109,133,107,8,\adfsplit 55,87,76,104,82,75,39,80,93,38,119,72)_{\adfGe}$, \adfSgap $(76,98,51,38,71,83,45,68,50,65,104,93,\adfsplit 106,81,44,82,128,117,21,57,111,105,6,87)_{\adfGe}$, $(5,17,33,99,32,29,59,123,9,71,11,41,\adfsplit 129,20,53,42,18,125,36,90,51,120,21,84)_{\adfGe}$, \adfLgap $(\infty,64,33,83,116,34,85,45,111,82,28,109,\adfsplit 37,0,50,35,86,110,70,68,115,40,69,56)_{\adfGf}$, $(11,89,129,50,90,60,94,3,115,122,39,40,\adfsplit 88,92,102,133,33,101,66,32,46,100,112,99)_{\adfGf}$, $(55,11,50,92,35,24,53,101,132,118,64,47,\adfsplit 43,122,67,96,108,23,111,103,134,4,37,15)_{\adfGf}$, $(80,125,14,101,50,36,69,64,132,8,13,129,\adfsplit 130,115,15,68,112,106,74,58,34,71,44,96)_{\adfGf}$, $(8,25,72,60,97,39,1,115,63,76,74,27,\adfsplit 23,7,18,9,96,55,14,31,51,128,130,36)_{\adfGf}$, \adfSgap $(67,46,92,32,38,25,126,4,123,80,50,30,\adfsplit 131,84,65,0,39,78,21,66,69,72,105,117)_{\adfGf}$, $(5,52,92,129,17,44,127,0,60,95,33,80,\adfsplit 108,54,117,114,78,96,30,45,111,56,8,3)_{\adfGf}$, \adfLgap $(\infty,123,16,92,100,30,29,69,132,57,27,67,\adfsplit 121,18,95,3,77,60,90,23,61,2,66,12)_{\adfGg}$, $(131,5,95,72,13,10,56,49,57,3,119,48,\adfsplit 66,11,89,32,86,16,46,74,22,114,88,4)_{\adfGg}$, $(80,6,101,23,62,35,91,82,0,134,68,92,\adfsplit 48,26,119,123,61,76,4,53,50,130,75,20)_{\adfGg}$, $(34,95,67,83,91,51,85,88,42,89,65,39,\adfsplit 115,61,82,5,118,53,133,57,71,69,19,60)_{\adfGg}$, $(89,24,101,121,103,30,83,11,15,91,62,1,\adfsplit 65,98,31,25,93,37,28,21,22,102,33,132)_{\adfGg}$, \adfSgap $(98,101,95,4,84,104,48,10,3,91,87,132,\adfsplit 61,111,69,51,109,115,9,93,36,0,58,117)_{\adfGg}$, $(1,55,27,3,54,75,121,7,130,105,49,70,\adfsplit 66,80,103,69,25,117,33,63,16,42,110,109)_{\adfGg}$, \adfLgap $(\infty,40,32,105,117,59,76,93,77,68,104,37,\adfsplit 5,85,128,18,4,34,79,110,65,64,51,126)_{\adfGh}$, $(57,36,114,81,4,55,19,66,112,10,46,17,\adfsplit 40,31,113,84,125,134,86,117,129,95,9,7)_{\adfGh}$, $(20,37,120,112,122,129,56,3,74,0,6,28,\adfsplit 75,30,52,73,72,79,71,10,83,53,63,110)_{\adfGh}$, $(54,18,123,88,74,128,81,69,4,8,51,110,\adfsplit 17,40,62,5,28,103,120,85,50,57,27,119)_{\adfGh}$, $(20,31,82,26,105,7,103,87,52,111,0,3,\adfsplit 36,39,6,70,48,121,28,42,41,86,118,50)_{\adfGh}$, \adfSgap $(90,129,51,77,116,33,119,20,74,25,59,82,\adfsplit 83,89,19,22,94,67,88,73,104,124,62,31)_{\adfGh}$, $(2,85,95,11,28,89,100,98,32,7,110,130,\adfsplit 35,41,70,26,124,9,66,113,92,82,77,134)_{\adfGh}$, \adfLgap $(\infty,23,123,79,87,12,80,58,38,90,134,109,\adfsplit 42,15,50,122,29,37,132,121,82,104,99,1)_{\adfGi}$, $(85,50,125,73,47,78,16,96,133,120,25,48,\adfsplit 59,71,60,83,12,23,84,69,75,22,40,91)_{\adfGi}$, $(3,128,52,36,83,27,58,7,50,80,61,60,\adfsplit 53,18,6,59,120,133,11,35,69,126,12,107)_{\adfGi}$, $(84,48,90,134,87,45,16,21,127,46,53,7,\adfsplit 59,64,56,31,129,119,70,8,43,0,6,52)_{\adfGi}$, $(36,83,90,94,78,35,133,34,37,3,117,108,\adfsplit 24,40,45,109,86,107,104,116,110,22,43,38)_{\adfGi}$, \adfSgap $(86,67,60,58,119,71,34,44,22,38,19,53,\adfsplit 18,56,65,104,14,2,127,109,107,70,10,28)_{\adfGi}$, $(1,103,34,31,50,80,64,133,73,57,52,83,\adfsplit 125,107,41,106,56,40,130,121,122,68,70,88)_{\adfGi}$, \adfLgap $(\infty,17,37,72,38,19,104,108,74,117,122,48,\adfsplit 119,93,106,92,64,42,68,15,107,88,44,84)_{\adfGj}$, $(31,55,121,113,26,51,22,108,27,117,48,53,\adfsplit 95,10,88,41,103,40,46,47,24,3,36,101)_{\adfGj}$, $(132,71,12,108,9,85,106,120,98,134,93,45,\adfsplit 61,89,63,6,59,94,56,127,119,100,126,104)_{\adfGj}$, $(16,101,92,23,72,68,35,60,103,66,65,57,\adfsplit 32,69,117,78,30,95,58,122,106,4,111,76)_{\adfGj}$, $(133,21,48,16,0,93,9,49,36,112,53,25,\adfsplit 105,20,28,114,59,119,125,124,113,91,37,72)_{\adfGj}$, \adfSgap $(3,106,28,46,11,81,19,49,103,58,124,94,\adfsplit 112,109,115,67,61,29,97,52,37,14,131,89)_{\adfGj}$, $(1,43,16,44,35,31,73,68,62,4,83,52,\adfsplit 25,127,46,38,56,94,104,37,112,71,131,69)_{\adfGj}$, \adfLgap $(\infty,8,99,25,127,54,124,24,9,38,74,122,\adfsplit 2,62,53,66,101,47,79,44,5,109,51,4)_{\adfGk}$, $(20,54,93,44,43,48,125,32,21,124,100,33,\adfsplit 7,112,88,39,131,83,92,41,1,130,109,82)_{\adfGk}$, $(69,65,31,28,129,128,106,38,57,97,13,132,\adfsplit 29,113,35,107,55,119,58,34,110,4,60,102)_{\adfGk}$, $(45,91,72,62,19,105,64,13,76,48,66,8,\adfsplit 89,113,102,36,67,100,97,70,1,9,46,63)_{\adfGk}$, $(97,57,49,86,62,52,34,58,80,82,79,69,\adfsplit 78,105,48,63,132,87,26,127,131,114,66,95)_{\adfGk}$, \adfSgap $(130,27,95,62,29,60,105,28,97,33,11,116,\adfsplit 128,129,117,84,44,77,93,63,51,74,35,120)_{\adfGk}$, $(0,59,77,78,126,108,26,66,71,80,39,96,\adfsplit 131,81,29,12,6,41,47,92,48,15,14,99)_{\adfGk}$, \adfLgap $(\infty,6,19,47,82,37,55,86,93,98,40,88,\adfsplit 26,10,85,73,56,35,42,90,71,87,22,21)_{\adfGl}$, $(130,134,50,119,3,103,76,98,7,28,84,133,\adfsplit 71,104,60,58,116,99,36,111,23,124,17,57)_{\adfGl}$, $(58,29,80,131,59,88,47,72,87,107,24,120,\adfsplit 90,11,122,67,81,130,127,8,56,43,55,14)_{\adfGl}$, $(119,23,109,42,120,95,84,25,126,35,113,125,\adfsplit 64,38,96,127,0,116,102,16,15,39,22,105)_{\adfGl}$, $(52,98,87,59,101,32,44,115,22,106,42,96,\adfsplit 27,51,15,65,9,41,99,57,111,72,74,62)_{\adfGl}$, \adfSgap $(8,75,7,22,84,70,16,9,133,64,56,15,\adfsplit 18,19,55,82,123,76,61,86,21,85,67,100)_{\adfGl}$, $(1,6,129,96,132,13,130,79,46,4,34,49,\adfsplit 25,115,90,9,16,40,66,67,109,73,127,108)_{\adfGl}$, \adfLgap $(\infty,125,103,69,70,64,82,106,24,20,81,123,\adfsplit 47,66,77,33,15,49,115,40,45,16,61,75)_{\adfGm}$, $(109,91,70,124,77,55,118,117,71,20,129,49,\adfsplit 45,110,112,121,98,35,99,68,114,130,0,43)_{\adfGm}$, $(8,48,10,64,80,128,14,78,53,116,37,90,\adfsplit 85,38,51,106,66,39,79,84,60,41,1,32)_{\adfGm}$, $(14,31,13,51,116,54,69,122,42,113,68,103,\adfsplit 58,119,12,96,93,23,118,7,89,129,134,87)_{\adfGm}$, $(54,102,48,1,122,127,52,105,103,93,29,3,\adfsplit 104,114,34,73,38,7,60,51,76,35,86,84)_{\adfGm}$, \adfSgap $(58,21,104,49,72,96,123,125,132,11,12,25,\adfsplit 117,5,113,92,128,20,34,98,43,122,80,38)_{\adfGm}$, $(128,14,41,40,116,62,53,122,35,8,20,17,\adfsplit 59,83,119,46,134,95,23,91,11,54,129,100)_{\adfGm}$, \adfLgap $(\infty,124,62,102,110,112,34,70,15,93,86,103,\adfsplit 89,12,7,10,98,90,8,75,37,117,104,97)_{\adfGn}$, $(54,13,19,133,108,124,93,88,85,118,38,28,\adfsplit 8,128,34,5,33,94,4,77,114,16,134,42)_{\adfGn}$, $(54,120,1,72,56,18,76,22,45,78,85,77,\adfsplit 53,64,49,87,36,67,113,106,74,28,24,117)_{\adfGn}$, $(85,44,52,75,127,132,95,74,24,13,9,134,\adfsplit 96,64,62,1,59,26,33,91,48,89,69,116)_{\adfGn}$, $(56,67,0,24,13,96,132,4,120,101,10,127,\adfsplit 65,63,28,70,66,11,134,53,36,126,98,102)_{\adfGn}$, \adfSgap $(10,98,47,15,131,29,71,68,8,38,45,24,\adfsplit 125,20,3,66,110,134,26,32,59,22,41,27)_{\adfGn}$, $(5,29,26,52,41,104,83,77,57,89,20,108,\adfsplit 53,101,71,78,11,68,66,36,62,131,96,2)_{\adfGn}$, \adfLgap $(\infty,112,90,29,49,24,116,74,127,134,5,25,\adfsplit 34,124,1,77,2,119,56,125,89,13,78,132)_{\adfGo}$, $(11,19,4,118,5,131,119,69,60,21,38,1,\adfsplit 9,51,54,92,71,61,66,13,126,48,134,94)_{\adfGo}$, $(47,4,121,20,10,12,78,17,35,24,88,1,\adfsplit 8,114,98,18,108,96,28,118,90,80,40,77)_{\adfGo}$, $(28,134,24,132,88,102,63,34,127,12,79,120,\adfsplit 49,2,76,78,133,16,117,58,112,103,43,77)_{\adfGo}$, $(121,18,15,68,91,110,92,13,125,54,127,88,\adfsplit 130,108,53,51,83,9,48,96,81,11,12,75)_{\adfGo}$, \adfSgap $(97,61,48,42,121,1,72,88,66,125,81,76,\adfsplit 75,110,36,40,26,29,80,27,33,62,119,113)_{\adfGo}$, $(1,100,123,96,93,51,41,77,11,5,27,53,\adfsplit 23,110,3,107,56,47,32,131,83,116,122,71)_{\adfGo}$, \adfLgap $(\infty,77,30,61,108,51,14,37,58,60,27,102,\adfsplit 0,117,46,126,91,28,103,74,17,70,15,118)_{\adfGp}$, $(40,73,110,11,61,4,16,88,92,116,134,59,\adfsplit 69,21,106,71,120,36,35,103,14,28,52,83)_{\adfGp}$, $(68,11,16,17,53,8,98,12,112,2,9,63,\adfsplit 124,110,88,15,93,30,61,77,23,95,62,105)_{\adfGp}$, $(83,87,107,39,82,66,34,30,95,51,93,80,\adfsplit 17,27,118,44,74,8,126,28,16,124,106,111)_{\adfGp}$, $(105,60,62,116,0,41,54,24,57,15,29,20,\adfsplit 106,80,76,117,120,36,53,48,82,16,78,46)_{\adfGp}$, \adfSgap $(29,128,56,1,85,68,36,116,27,7,41,32,\adfsplit 73,127,100,21,13,92,109,55,67,126,25,4)_{\adfGp}$, $(5,97,112,115,48,43,106,121,22,134,21,73,\adfsplit 16,51,98,85,15,7,125,127,79,64,58,105)_{\adfGp}$, \adfLgap $(\infty,87,65,55,68,108,114,25,7,132,34,110,\adfsplit 120,100,128,75,31,89,117,78,2,99,4,48)_{\adfGq}$, $(12,1,111,20,129,57,31,79,3,65,13,61,\adfsplit 15,52,81,97,23,91,124,37,47,74,120,99)_{\adfGq}$, $(57,98,91,14,105,59,63,34,35,0,83,92,\adfsplit 87,93,120,97,18,37,32,39,1,8,17,72)_{\adfGq}$, $(117,17,90,50,16,97,76,15,110,86,116,96,\adfsplit 85,78,13,38,109,120,66,83,122,80,130,92)_{\adfGq}$, $(129,6,25,90,84,89,130,63,113,42,41,35,\adfsplit 18,46,48,107,120,13,65,1,109,103,112,71)_{\adfGq}$, \adfSgap $(23,16,40,106,59,127,122,58,130,95,110,68,\adfsplit 52,28,98,125,115,134,26,10,56,128,82,4)_{\adfGq}$, $(5,112,31,20,77,1,74,107,47,133,44,101,\adfsplit 7,110,94,61,37,35,124,22,28,125,50,118)_{\adfGq}$, \adfLgap $(\infty,41,28,45,96,58,76,32,134,65,21,13,\adfsplit 48,0,57,61,62,25,67,125,6,15,101,16)_{\adfGr}$, $(32,0,9,119,17,84,53,107,12,131,70,69,\adfsplit 40,79,56,104,42,35,47,50,112,59,124,126)_{\adfGr}$, $(29,100,87,20,115,24,52,128,76,62,93,121,\adfsplit 81,66,132,48,79,83,16,126,45,1,57,123)_{\adfGr}$, $(77,91,126,110,50,8,122,85,97,130,88,35,\adfsplit 117,114,13,24,58,7,108,106,52,96,118,74)_{\adfGr}$, $(74,8,20,0,32,112,2,73,57,109,82,27,\adfsplit 16,96,51,39,75,37,9,10,63,69,56,64)_{\adfGr}$, \adfSgap $(18,34,32,71,77,36,16,7,115,10,65,119,\adfsplit 47,53,35,96,94,49,109,55,59,112,110,17)_{\adfGr}$, $(1,41,3,17,118,20,74,56,102,68,6,26,\adfsplit 13,71,132,59,90,7,86,92,57,80,5,130)_{\adfGr}$, \adfLgap $(\infty,121,122,120,7,30,84,123,21,129,2,133,\adfsplit 124,44,11,67,82,116,113,58,48,91,4,63)_{\adfGs}$, $(79,18,124,51,86,47,36,89,8,43,7,91,\adfsplit 16,60,132,2,134,117,10,115,67,53,108,68)_{\adfGs}$, $(33,99,82,15,75,92,117,79,5,123,96,71,\adfsplit 89,7,98,101,104,34,30,27,63,62,46,47)_{\adfGs}$, $(99,27,107,112,6,8,121,101,53,102,37,46,\adfsplit 22,14,49,97,61,7,103,34,29,31,9,63)_{\adfGs}$, $(49,56,120,6,41,12,122,16,101,17,109,69,\adfsplit 124,11,89,8,47,73,20,50,71,77,28,23)_{\adfGs}$, \adfSgap $(17,68,30,63,81,119,0,126,33,32,87,132,\adfsplit 51,88,78,48,121,120,21,102,16,45,96,18)_{\adfGs}$, $(4,99,106,72,40,96,73,81,10,75,93,15,\adfsplit 129,78,123,47,48,55,19,98,121,0,124,66)_{\adfGs}$, \adfLgap $(\infty,2,52,0,66,95,87,17,63,118,80,48,\adfsplit 53,31,78,89,21,115,75,132,125,74,85,42)_{\adfGt}$, $(78,90,27,37,84,95,106,3,9,49,50,36,\adfsplit 17,29,0,10,61,30,75,121,94,113,8,2)_{\adfGt}$, $(0,56,124,32,110,90,44,65,93,102,81,25,\adfsplit 40,26,1,128,91,88,6,99,119,41,97,92)_{\adfGt}$, $(43,0,63,62,116,16,107,47,25,49,75,122,\adfsplit 88,52,67,13,58,65,15,64,103,132,4,45)_{\adfGt}$, $(54,121,0,123,24,36,131,87,81,91,75,21,\adfsplit 82,50,39,25,44,109,49,23,92,111,124,95)_{\adfGt}$, \adfSgap $(96,67,56,7,88,103,49,74,95,28,83,100,\adfsplit 62,23,8,32,106,102,19,20,13,112,73,16)_{\adfGt}$, $(5,45,134,62,91,16,52,116,55,119,86,115,\adfsplit 49,92,35,95,73,131,22,77,8,26,20,110)_{\adfGt}$, \adfLgap $(\infty,126,41,121,109,56,78,82,13,2,114,34,\adfsplit 127,124,113,54,49,33,3,9,91,55,42,95)_{\adfGu}$, $(43,56,119,122,55,19,11,85,117,79,52,101,\adfsplit 60,133,32,23,115,36,45,75,77,89,72,123)_{\adfGu}$, $(127,108,132,42,2,55,30,15,13,62,37,51,\adfsplit 85,69,75,112,74,20,9,14,35,76,24,56)_{\adfGu}$, $(48,127,131,64,76,44,87,56,9,90,14,73,\adfsplit 37,78,46,35,75,15,68,20,60,107,12,39)_{\adfGu}$, $(53,69,55,108,27,18,15,5,35,49,24,11,\adfsplit 44,93,73,14,45,13,70,43,88,114,47,83)_{\adfGu}$, \adfSgap $(90,86,55,116,61,22,67,65,71,133,29,57,\adfsplit 16,37,33,53,83,19,25,59,2,4,32,47)_{\adfGu}$, $(2,112,35,131,109,95,43,10,29,31,46,50,\adfsplit 17,110,37,62,86,88,116,103,97,113,80,94)_{\adfGu}$, \adfLgap $(\infty,61,119,105,104,129,46,118,24,10,1,60,\adfsplit 42,123,84,124,50,93,125,109,26,4,17,39)_{\adfGv}$, $(33,67,62,43,61,36,69,12,118,131,11,7,\adfsplit 89,80,90,37,55,6,52,101,59,87,88,98)_{\adfGv}$, $(54,48,34,109,126,26,30,52,29,9,67,74,\adfsplit 133,93,32,43,125,61,105,47,127,68,108,41)_{\adfGv}$, $(60,33,63,37,38,80,128,61,116,133,73,126,\adfsplit 72,70,88,52,0,121,42,24,20,69,99,50)_{\adfGv}$, $(37,17,79,51,21,28,5,63,20,104,128,60,\adfsplit 121,30,123,93,133,55,131,3,50,45,119,40)_{\adfGv}$, \adfSgap $(48,108,104,91,16,33,132,131,83,53,93,76,\adfsplit 56,112,116,65,49,106,89,134,119,74,11,7)_{\adfGv}$, $(2,14,119,125,32,22,100,1,44,50,38,104,\adfsplit 71,117,88,91,41,4,89,61,64,131,122,80)_{\adfGv}$, \adfLgap $(\infty,94,105,59,27,134,123,19,114,117,46,118,\adfsplit 47,8,121,64,22,131,14,65,36,29,6,1)_{\adfGw}$, $(25,35,88,54,2,94,107,56,91,120,48,86,\adfsplit 129,39,30,123,57,106,93,83,82,79,60,112)_{\adfGw}$, $(63,84,127,13,86,104,101,66,90,88,67,8,\adfsplit 73,118,112,69,68,72,98,22,111,103,129,64)_{\adfGw}$, $(112,95,25,121,29,71,105,116,78,1,39,22,\adfsplit 73,104,134,77,49,40,52,51,119,98,108,61)_{\adfGw}$, $(55,66,121,116,46,13,80,81,71,105,14,49,\adfsplit 62,86,127,82,56,112,134,113,29,114,84,32)_{\adfGw}$, \adfSgap $(127,57,54,65,114,120,3,66,30,111,84,58,\adfsplit 44,50,123,20,131,101,87,107,42,36,60,6)_{\adfGw}$, $(3,101,47,131,60,0,104,54,53,90,57,15,\adfsplit 18,17,27,81,99,96,111,97,59,35,126,24)_{\adfGw}$, \adfLgap $(\infty,95,16,90,68,19,79,22,85,66,10,106,\adfsplit 62,53,49,101,32,15,30,52,84,47,120,12)_{\adfGx}$, $(31,81,0,97,103,14,4,26,101,29,76,40,\adfsplit 59,78,48,46,67,125,134,112,91,83,105,3)_{\adfGx}$, $(38,44,24,101,16,37,91,74,100,115,25,78,\adfsplit 108,99,59,7,90,29,127,132,87,39,57,41)_{\adfGx}$, $(124,28,0,114,110,48,20,128,105,25,38,127,\adfsplit 106,98,6,81,70,2,130,5,43,97,123,26)_{\adfGx}$, $(113,110,129,94,13,64,67,65,42,18,43,35,\adfsplit 100,101,56,85,19,83,48,74,87,2,11,17)_{\adfGx}$, \adfSgap $(13,62,36,101,53,102,21,120,123,60,81,79,\adfsplit 131,33,114,122,128,32,111,96,54,57,119,0)_{\adfGx}$, $(0,6,11,108,80,57,72,51,102,126,9,30,\adfsplit 123,24,105,48,90,93,1,27,111,89,50,78)_{\adfGx}$, \adfLgap $(\infty,68,108,127,24,55,134,11,115,126,0,96,\adfsplit 45,33,114,118,121,28,73,2,72,21,97,6)_{\adfGy}$, $(103,106,68,54,93,47,117,108,95,35,19,24,\adfsplit 3,126,100,5,44,30,133,73,134,97,4,88)_{\adfGy}$, $(35,0,6,99,112,9,60,51,131,96,129,133,\adfsplit 12,115,83,81,72,53,11,5,120,44,50,62)_{\adfGy}$, $(91,20,19,64,3,27,124,94,59,113,54,58,\adfsplit 133,47,134,101,51,17,18,108,49,67,16,6)_{\adfGy}$, $(54,133,129,116,8,27,4,34,76,25,17,90,\adfsplit 30,86,134,2,43,109,22,81,26,74,31,131)_{\adfGy}$, \adfSgap $(96,34,64,133,131,66,28,97,71,101,4,103,\adfsplit 95,10,77,2,86,1,62,128,11,44,134,49)_{\adfGy}$, $(2,79,92,77,31,71,134,80,133,122,47,13,\adfsplit 97,44,59,110,45,49,74,82,95,118,41,26)_{\adfGy}$, \adfLgap $(\infty,58,111,29,43,22,106,26,54,80,66,61,\adfsplit 1,74,28,118,30,57,119,50,16,59,46,65)_{\adfGz}$, $(96,63,55,4,107,104,121,32,26,94,62,48,\adfsplit 67,87,17,36,13,38,69,60,28,39,23,75)_{\adfGz}$, $(55,132,45,63,115,8,93,58,40,122,100,96,\adfsplit 128,86,47,15,90,114,60,106,43,80,61,49)_{\adfGz}$, $(46,108,45,100,56,84,33,0,50,26,4,98,\adfsplit 49,63,130,88,36,3,70,103,101,95,122,23)_{\adfGz}$, $(124,3,53,82,118,78,115,13,50,6,10,7,\adfsplit 59,61,34,42,41,44,43,5,108,80,113,9)_{\adfGz}$, \adfSgap $(106,69,99,98,38,122,41,71,66,59,18,130,\adfsplit 83,101,72,131,93,123,80,62,114,57,0,45)_{\adfGz}$, $(3,84,99,42,26,63,131,38,23,119,44,24,\adfsplit 6,77,66,105,78,102,120,89,73,83,65,36)_{\adfGz}$, \adfLgap $(\infty,68,0,109,87,60,20,118,48,112,65,66,\adfsplit 127,25,37,9,92,98,126,21,26,114,14,32)_{\adfGA}$, $(95,53,38,92,19,83,90,129,54,22,25,118,\adfsplit 79,80,86,9,66,21,126,56,23,111,49,10)_{\adfGA}$, $(86,107,90,91,67,61,112,65,130,102,32,17,\adfsplit 13,28,45,64,76,40,29,123,21,50,69,70)_{\adfGA}$, $(48,66,113,16,39,83,111,122,70,69,56,34,\adfsplit 24,103,25,1,126,107,61,9,52,78,22,85)_{\adfGA}$, $(23,22,45,76,16,43,95,5,96,133,94,112,\adfsplit 105,129,116,130,124,97,126,128,14,122,24,83)_{\adfGA}$, \adfSgap $(115,119,6,129,101,60,21,57,53,83,2,122,\adfsplit 126,31,47,63,45,32,42,35,62,78,17,14)_{\adfGA}$, $(2,56,66,30,33,41,132,81,107,129,128,23,\adfsplit 84,35,68,53,99,77,0,134,123,113,54,109)_{\adfGA}$, \adfLgap $(\infty,54,7,8,112,66,103,72,134,131,96,53,\adfsplit 64,57,117,1,61,113,118,13,35,30,125,16)_{\adfGB}$, $(27,121,75,106,125,12,68,9,13,6,72,59,\adfsplit 74,126,26,8,84,112,25,10,117,88,5,20)_{\adfGB}$, $(109,66,21,10,69,120,19,56,94,26,97,82,\adfsplit 53,67,102,50,32,6,51,112,35,86,5,61)_{\adfGB}$, $(38,57,39,42,41,88,85,44,69,3,20,68,\adfsplit 33,81,35,52,10,58,27,49,74,79,96,40)_{\adfGB}$, $(124,14,87,64,108,99,88,19,75,101,115,9,\adfsplit 82,67,59,121,33,130,30,107,110,52,24,131)_{\adfGB}$, \adfSgap $(91,122,62,74,132,46,54,31,48,117,92,108,\adfsplit 77,93,86,61,30,125,8,32,20,107,81,51)_{\adfGB}$, $(2,71,129,61,77,38,51,98,44,123,62,56,\adfsplit 92,26,72,14,85,39,5,132,101,131,68,29)_{\adfGB}$, \adfLgap $(\infty,103,120,11,101,56,111,46,67,93,134,70,\adfsplit 57,115,35,118,28,77,50,7,131,36,119,126)_{\adfGC}$, $(2,76,77,78,31,47,93,80,50,104,64,95,\adfsplit 99,111,18,28,86,58,38,17,35,133,37,123)_{\adfGC}$, $(4,84,25,104,47,76,68,51,109,46,42,70,\adfsplit 134,118,38,98,121,44,132,36,129,127,14,69)_{\adfGC}$, $(117,77,73,20,118,57,88,5,42,90,105,11,\adfsplit 71,109,126,1,3,35,134,55,65,129,66,93)_{\adfGC}$, $(113,71,11,28,79,116,26,74,40,32,45,104,\adfsplit 51,72,2,57,108,18,132,3,82,102,76,4)_{\adfGC}$, \adfSgap $(25,48,28,12,1,7,123,45,132,27,114,57,\adfsplit 97,84,19,3,40,109,91,112,52,94,43,0)_{\adfGC}$, $(0,15,54,88,127,69,94,67,84,70,121,112,\adfsplit 34,45,3,10,27,130,73,49,42,25,51,1)_{\adfGC}$, \adfLgap $(\infty,116,40,57,7,90,43,53,59,73,42,130,\adfsplit 102,3,107,16,105,112,89,26,76,117,92,49)_{\adfGD}$, $(101,29,18,120,91,97,107,19,12,115,1,95,\adfsplit 69,126,2,98,34,76,8,103,131,92,61,47)_{\adfGD}$, $(64,50,113,107,78,84,83,112,119,73,102,21,\adfsplit 63,103,111,18,86,67,120,71,22,77,118,43)_{\adfGD}$, $(127,83,122,14,47,39,58,46,71,96,86,121,\adfsplit 133,75,54,93,51,67,18,45,38,88,99,114)_{\adfGD}$, $(34,59,93,22,27,32,49,112,12,16,25,28,\adfsplit 60,67,36,87,21,74,131,54,129,55,132,37)_{\adfGD}$, \adfSgap $(115,22,20,50,62,64,0,110,8,29,134,106,\adfsplit 96,56,98,66,120,71,23,116,44,30,92,87)_{\adfGD}$, $(1,15,101,41,11,35,33,31,86,78,0,72,\adfsplit 45,131,17,113,68,5,83,39,75,110,126,59)_{\adfGD}$, \adfLgap $(\infty,119,43,93,115,17,83,26,55,88,128,56,\adfsplit 132,63,120,74,33,129,82,98,113,36,20,84)_{\adfGE}$, $(27,117,61,49,83,50,128,45,93,82,132,40,\adfsplit 109,76,121,71,10,56,8,77,72,64,58,115)_{\adfGE}$, $(16,34,74,96,129,15,131,55,67,8,2,95,\adfsplit 124,46,70,105,114,51,99,120,23,108,72,45)_{\adfGE}$, $(122,100,28,66,85,126,81,60,26,105,80,108,\adfsplit 73,17,35,83,52,91,72,45,43,0,102,20)_{\adfGE}$, $(134,50,75,52,45,120,33,108,129,104,21,0,\adfsplit 115,118,61,5,11,37,23,31,96,76,49,131)_{\adfGE}$, \adfSgap $(45,53,44,62,125,5,88,67,110,2,123,68,\adfsplit 131,4,46,122,97,20,71,118,109,70,86,91)_{\adfGE}$, $(2,65,31,112,134,57,70,52,68,14,74,56,\adfsplit 94,124,88,44,100,131,121,101,46,71,82,22)_{\adfGE}$, \adfLgap $(\infty,70,51,113,45,67,106,81,101,82,97,85,\adfsplit 10,123,64,31,9,48,2,77,40,88,7,86)_{\adfGF}$, $(38,21,11,66,103,56,128,7,44,25,47,79,\adfsplit 115,96,33,107,52,92,41,87,85,95,68,121)_{\adfGF}$, $(28,67,70,38,129,102,27,1,63,133,81,12,\adfsplit 80,18,49,132,43,121,8,50,100,58,93,23)_{\adfGF}$, $(12,129,70,0,59,30,20,3,104,119,10,105,\adfsplit 44,101,52,48,16,132,117,13,40,35,2,123)_{\adfGF}$, $(31,121,43,88,72,9,23,103,67,75,37,19,\adfsplit 81,99,38,20,44,0,5,3,132,122,84,98)_{\adfGF}$, \adfSgap $(87,63,83,36,116,52,90,81,56,86,18,8,\adfsplit 95,14,35,38,62,2,89,68,82,69,48,93)_{\adfGF}$, $(2,89,83,77,87,41,113,93,121,8,51,116,\adfsplit 95,11,132,66,101,126,74,102,15,75,35,68)_{\adfGF}$, \adfLgap $(\infty,96,44,49,13,35,120,45,27,122,4,21,\adfsplit 98,3,127,88,89,28,129,85,11,22,23,67)_{\adfGG}$, $(126,120,92,15,53,54,8,99,111,55,11,75,\adfsplit 66,40,95,52,110,129,56,10,67,116,21,134)_{\adfGG}$, $(15,105,78,132,41,14,114,7,129,27,128,11,\adfsplit 16,26,3,64,72,37,130,34,84,80,110,61)_{\adfGG}$, $(22,123,3,21,31,43,64,7,112,53,75,27,\adfsplit 124,35,28,78,90,23,93,24,107,41,20,118)_{\adfGG}$, $(113,133,86,18,15,24,121,45,28,13,112,132,\adfsplit 33,120,92,54,76,32,91,5,35,64,128,104)_{\adfGG}$, \adfSgap $(117,79,32,80,37,101,41,17,97,58,22,134,\adfsplit 8,50,52,118,47,124,68,28,76,14,31,128)_{\adfGG}$, $(6,104,56,103,34,119,121,116,113,61,21,38,\adfsplit 118,37,32,47,71,55,70,127,62,17,67,65)_{\adfGG}$, \adfLgap $(\infty,103,69,134,98,133,46,114,16,104,38,83,\adfsplit 120,9,84,122,81,71,6,63,25,72,80,108)_{\adfGH}$, $(89,92,13,128,101,102,30,127,100,112,22,80,\adfsplit 83,52,6,54,133,76,94,46,97,123,48,35)_{\adfGH}$, $(66,126,68,86,112,94,8,53,115,117,109,28,\adfsplit 60,121,39,33,120,79,37,50,13,2,98,48)_{\adfGH}$, $(74,94,41,9,106,40,131,102,124,104,1,2,\adfsplit 134,10,46,16,35,65,56,111,120,82,92,13)_{\adfGH}$, $(109,68,7,44,76,26,119,129,34,8,60,106,\adfsplit 64,85,39,9,70,33,43,93,21,6,14,30)_{\adfGH}$, \adfSgap $(109,77,108,102,56,69,35,115,113,123,72,76,\adfsplit 83,114,119,23,111,75,44,0,128,90,96,134)_{\adfGH}$, $(35,87,102,21,92,14,3,6,24,65,71,126,\adfsplit 42,17,68,117,50,131,27,74,9,30,39,44)_{\adfGH}$, \adfLgap $(\infty,11,133,99,111,59,74,16,125,95,66,5,\adfsplit 94,76,51,79,89,30,82,93,13,128,12,113)_{\adfGI}$, $(57,134,10,133,132,126,118,4,31,131,100,28,\adfsplit 45,87,67,122,69,33,21,75,42,24,97,8)_{\adfGI}$, $(107,16,79,32,23,65,76,29,20,131,6,97,\adfsplit 47,52,115,88,93,77,41,54,72,55,87,112)_{\adfGI}$, $(10,78,55,98,58,13,97,17,93,35,22,63,\adfsplit 81,74,24,41,32,23,82,45,4,111,88,34)_{\adfGI}$, $(101,116,36,7,82,55,61,115,19,51,109,40,\adfsplit 4,66,8,83,110,23,92,35,59,48,111,125)_{\adfGI}$, \adfSgap $(5,110,47,129,80,118,69,6,30,120,90,114,\adfsplit 45,105,29,81,96,75,7,111,54,134,92,3)_{\adfGI}$, $(76,48,68,72,123,108,96,98,122,36,119,3,\adfsplit 114,45,105,69,12,44,9,23,99,50,66,116)_{\adfGI}$, \adfLgap $(\infty,72,73,113,28,115,61,130,3,117,129,66,\adfsplit 26,132,97,111,1,32,5,80,98,105,114,107)_{\adfGJ}$, $(89,24,112,4,61,131,115,16,13,20,118,34,\adfsplit 26,31,80,3,54,14,33,32,85,73,63,71)_{\adfGJ}$, $(84,60,19,74,87,126,25,62,48,12,50,5,\adfsplit 34,47,0,130,105,104,133,55,30,72,75,128)_{\adfGJ}$, $(87,40,128,102,90,59,23,84,13,3,54,131,\adfsplit 112,63,71,126,25,110,106,117,2,1,76,34)_{\adfGJ}$, $(40,124,22,133,41,108,110,44,98,106,49,100,\adfsplit 62,7,38,20,75,11,64,97,84,43,54,53)_{\adfGJ}$, \adfSgap $(52,118,131,121,92,49,36,38,95,68,54,128,\adfsplit 91,123,96,12,101,35,125,45,62,93,88,102)_{\adfGJ}$, $(41,129,60,126,78,107,18,92,125,134,51,105,\adfsplit 99,69,122,44,111,48,65,77,86,110,108,84)_{\adfGJ}$, \adfLgap $(\infty,112,122,54,86,109,60,81,51,9,116,97,\adfsplit 13,90,26,19,36,103,78,98,47,37,87,80)_{\adfGK}$, $(24,38,68,119,33,18,132,73,49,51,75,111,\adfsplit 127,43,94,83,76,32,112,92,34,74,19,71)_{\adfGK}$, $(130,51,49,103,29,39,0,12,128,92,26,57,\adfsplit 63,78,133,40,20,5,64,11,127,13,90,72)_{\adfGK}$, $(27,40,133,6,128,64,49,34,53,30,60,44,\adfsplit 125,57,96,0,70,108,124,69,119,37,62,15)_{\adfGK}$, $(75,47,26,81,25,84,27,3,114,70,122,40,\adfsplit 112,97,64,80,18,77,55,17,131,34,99,119)_{\adfGK}$, \adfSgap $(24,82,39,9,29,48,97,54,134,67,38,79,\adfsplit 58,88,85,34,14,125,118,47,116,104,101,50)_{\adfGK}$, $(96,130,127,86,134,41,31,8,56,110,32,118,\adfsplit 2,47,35,64,14,71,89,16,68,74,122,98)_{\adfGK}$ \noindent and $(\infty,12,46,92,123,2,67,77,91,116,33,76,\adfsplit 3,58,113,70,125,7,100,114,63,20,131,22)_{\adfGL}$, $(108,111,17,69,5,82,80,62,46,124,53,13,\adfsplit 58,96,76,133,114,19,8,38,105,3,121,132)_{\adfGL}$, $(70,134,24,34,129,108,96,85,104,42,120,32,\adfsplit 83,74,12,109,23,76,88,113,55,133,90,118)_{\adfGL}$, $(131,47,54,89,113,108,81,13,105,96,22,23,\adfsplit 27,100,17,130,116,36,29,31,7,74,119,78)_{\adfGL}$, $(133,83,130,8,118,98,55,111,56,47,49,7,\adfsplit 72,19,65,31,124,102,117,1,73,123,3,40)_{\adfGL}$, \adfSgap $(57,93,56,128,5,23,83,134,126,36,21,80,\adfsplit 104,42,39,0,35,105,24,9,107,78,129,38)_{\adfGL}$, $(131,39,75,60,126,90,47,123,38,113,45,42,\adfsplit 84,108,95,5,0,98,132,89,2,87,116,66)_{\adfGL}$ \noindent under the action of the mapping $\infty \mapsto \infty$, $x \mapsto x + 3$ (mod 135) for the first five graphs in each design, and $x \mapsto x + 9$ (mod 135) for the last two. \eproof \adfVfy{136, \{\{135,15,9\},\{1,1,1\}\}, 135, -1, \{5,\{\{135,3,3\},\{1,1,1\}\}\}} {\lemma \label{lem:Snark24 designs} There exist designs of order $64$, $73$ and $145$ for each of the thirty-eight 24-vertex non-trivial snarks.}\\ \noindent\textbf{Proof.} The decompositions are given in the Appendix. \eproof {\lemma \label{lem:Snark24 multipartite} There exist decompositions of the complete multipartite graphs $K_{12,12,12}$, $K_{24,24,15}$, $K_{72,72,63}$, $K_{24,24,24,24}$ and $K_{24,24,24,21}$ for each of the thirty-eight 24-vertex non-trivial snarks.}\\ \noindent\textbf{Proof.} The decompositions are given in the Appendix. \eproof \vskip 2mm Theorem~\ref{thm:snark24} follows from Lemmas~\ref{lem:Snark24 136}, \ref{lem:Snark24 designs} and \ref{lem:Snark24 multipartite}, and Proposition~\ref{prop:d=3, v=24}.
cbc85a15fa71c0f6eb113890e1e1f047d4cdcfee
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Simple bounds for the needed embedding constants} \def\Alph{section}{\Alph{section}} The following theorem provides the best constant in the classical Sobolev inequality with critical exponents. \begin{theo}[T.~Aubin \cite{aubin1976} and G.~Talenti \cite{talenti1976}]\label{talentitheo} Let $u$ be any function in $W^{1,q}\left(\mathbb{R}^{n}\right)\ (n\geq 2)$, where $q$ is any real number such that $1<q<n$. Moreover, set $p=nq/\left(n-q\right)$. Then, $u \in L^{p}\left(\mathbb{R}^{n}\right)$ and \begin{align*} \left(\int_{\mathbb{R}^{n}}\left|u(x)\right|^{p}dx\right)^{\frac{1}{p}}\leq T_{p}\left(\int_{\mathbb{R}^{n}}\left|\nabla u(x)\right|_{2}^{q}dx\right)^{\frac{1}{q}} \end{align*} holds for \begin{align} T_{p}=\pi^{-\frac{1}{2}}n^{-\frac{1}{q}}\left(\frac{q-1}{n-q}\right)^{1-\frac{1}{q}}\left\{\frac{\Gamma\left(1+\frac{n}{2}\right)\Gamma\left(n\right)}{\Gamma\left(\frac{n}{q}\right)\Gamma\left(1+n-\frac{n}{q}\right)}\right\}^{\frac{1}{n}}\label{talenticonst}, \end{align} where $\left|\nabla u\right|_{2}=\left((\partial u/\partial x_{1})^{2}+(\partial u/\partial x_{2})^{2}+\cdots+(\partial u/\partial x_{n})^{2}\right)^{1/2}$, and $\Gamma$ denotes the gamma function. \end{theo} The following corollary, obtained from Theorem \ref{talentitheo}, provides a simple bound for the embedding constant from $H_{0}^{1}\left(\Omega\right)$ to $L^{p}(\Omega)$ for a bounded domain $\Omega$. \begin{coro}\label{roughboundtheo} Let $\Omega\subset \mathbb{R}^{n}\,(n\geq 2)$ be a bounded domain. Let $p$ be a real number such that $p\in(n/(n-1),2n/(n-2)]$ if $n\geq 3$ and $p\in(n/(n-1),\infty)$ if $n=2$. Moreover, set $q=np/(n+p).$ Then, $(\ref{embedding})$ holds for \begin{align*} C_{p}=\left|\Omega\right|^{\frac{2-q}{2q}}T_{p}, \end{align*} where $T_{p}$ is the constant in {\rm (\ref{talenticonst})}. \end{coro} \begin{proof} By zero extension outside $\Omega$, we may regard $u\in H_{0}^{1}\left(\Omega\right)$ as an element $u\in W^{1,q}\left(\mathbb{R}^{n}\right)$; note that $q\leq2$, and $q<2$ if $n=2$. Therefore, from Theorem \ref{talentitheo}, \begin{align} \left\|u\right\|_{L^{p}\left(\Omega\right)} \leq T_{p}\left(\int_{\Omega}\left|\nabla u\left(x\right)\right|_{2}^{q}dx\right)^{\frac{1}{q}}.\label{embedding/theo/1} \end{align} H\"{o}lder's inequality gives \begin{align} \int_{\Omega}\left|\nabla u\left(x\right)\right|_{2}^{q}dx& \leq\left(\int_{\Omega} \left| \nabla u\left(x\right) \right|_{2}^{q\cdot\frac{2}{q}}dx\right)^{\frac{q}{2}}\left(\int_{\Omega}1^{\frac{2}{2-q}}dx\right)^{\frac{2-q}{2}}\nonumber\\ &=\left|\Omega\right|^{\frac{2-q}{2}} \left(\int_{\Omega}\left| \nabla u\left(x\right) \right|_{2}^{2}dx\right)^{\frac{q}{2}},\nonumber \end{align} that is, \begin{align} \left(\int_{\mathbb{R}^{n}}\left|\nabla u\left(x\right)\right|_{2}^{q}dx\right)^{\frac{1}{q}} \leq\left|\Omega\right|^{\frac{2-q}{2q}}\left\|\nabla u\right\|_{L^{2}\left(\Omega\right)},\label{embedding/theo/2} \end{align} where $\left|\Omega\right|$ is the measure of $\Omega$. From (\ref{embedding/theo/1}) and (\ref{embedding/theo/2}), it follows that \begin{align*} \left\|u\right\|_{L^{p}\left(\Omega\right)}& \leq\left|\Omega\right|^{\frac{2-q}{2q}}T_{p}\left\|\nabla u\right\|_{L^{2}\left(\Omega\right)}. \end{align*} \end{proof} \section{Numerical example}\label{sec/example} In this section, we present a numerical example where a solution to \eqref{positiveproblem} is numerically verified. All computations were carried out on a computer with Intel Xeon E7-4830 at 2.20 GHz$\times$40, 2 TB RAM, CentOS 6.6, and MATLAB 2012b. All rounding errors were strictly estimated using toolboxes---the INTLAB version 9 \cite{rump1999book} and KV library version 0.4.16 \cite{kashiwagikv}---for verified numerical computations. Therefore, the correctness of all results was mathematically guaranteed. We consider the case in which $p=3/2$ and $\Omega=\Omega_{s}(:=(0,1)^{2})$. By selecting $q=r=4$ and $s=2$ in Theorem \ref{selectiong}, we may select \begin{align} g\displaystyle \left(t\right)=\frac{3}{2}C_{2}^{\frac{3}{2}}C_{4}t^{\frac{1}{2}}\label{gfor3/2} \end{align} to satisfy \eqref{lip} and \eqref{gconv} in Theorem \ref{plum2001}. Moreover, by selecting $q=4$ and $r=2$ in Corollary \ref{Linfcoro}, we have \begin{align*} \left\|u-\hat{u}\right\|_{L^{\infty}(\Omega)}\leq c_{0}C_{2}\varepsilon+c_{1}\varepsilon+c_{2}\left\{\frac{3}{2}\varepsilon C_{4}\sqrt{\left\|\hat{u}\right\|_{L^{2}\left(\Omega\right)}+\frac{\varepsilon}{2}C_{2}}+\left\|\Delta\hat{u}+\left|\hat{u}\right|^{\frac{1}{2}}\hat{u}\right\|_{L^{2}(\Omega)}\right\}. \end{align*} On the square $\Omega_{s}$, it is well known that $C_{2}=(\sqrt{2}\pi)^{-1}$; moreover, using Corollary \ref{roughboundtheo}, we computed $C_{4}\leq 0.318309887$. We select a finite-dimensional subspace $V_{N}$ of $V$ as \begin{align*} V_{N}:=\left\{\sum_{(i,j)\in\Lambda^{1,1}_{N}}a_{i,j}\varphi_{i,j}\ :\ a_{i,j}\in \mathbb{R}\right\}, \end{align*} where $\varphi_{i,j}(x,y)=\sin\left(i\pi x\right)\sin\left(j\pi y\right)$. For this $V_{N}$, we may select $C_{N}=(N+1)^{-1}\pi^{-1}$ to satisfy \eqref{CN}, because \begin{align*} \left\|u-P_{N}u\right\|_{V}^{2}&=\left\|\left(u-P_{N}u\right)_{x}\right\|_{L^{2}(\Omega)}^{2}+\|\left(u-P_{N}u\right)_{y}\|_{L^{2}(\Omega)}^{2}\\ &=\displaystyle \sum_{(n,m)\in\Lambda^{1,1}_{\infty}\backslash\Lambda^{1,1}_{N}}a_{m,n}^{2}(m^{2}\pi^{2}+n^{2}\pi^{2})\left\|\varphi_{m,n}\right\|_{L^{2}(\Omega)}^{2}\\ &\displaystyle \leq\sum_{(n,m)\in\Lambda^{1,1}_{\infty}\backslash\Lambda^{1,1}_{N}}a_{m,n}^{2}\frac{(m^{2}\pi^{2}+n^{2}\pi^{2})^{2}}{(N+1)^{2}\pi^{2}}\left\|\varphi_{m,n}\right\|_{L^{2}(\Omega)}^{2}\\ &\displaystyle \leq\frac{1}{\left(N+1\right)^{2}\pi^{2}}\left\|-\Delta u\right\|_{L^{2}(\Omega)}^{2}. \end{align*} We are interested in finding a reflection symmetric solution, and hence restricted the solution space to the following subspace of $V$: \begin{align*} \left\{u\in V\ :\ u{\rm~is~symmetric~with~respect~to~}x=\frac{1}{2}{\rm~and~}y=\frac{1}{2}\right\} \end{align*} endowed with the same topology as $V$. This restriction helped us to somewhat reduce the calculation quantity. Moreover, since eigenfunctions of \eqref{eiglam} are now also restricted to symmetric functions, eigenvalues associated with anti-symmetric eigenfunction drop out of the minimization in \eqref{mu0}, and so the constant $K$ is possibly reduced. The other constants required in the process of the verification (i.e., $C_{p},\ \delta$) are not affected by the restriction. We computed an approximate solution $\hat{u}$ to \eqref{positiveproblem}, which is displayed in Fig.~\ref{pic}, with the Fourier-Galerkin method, i.e., $\hat{u}$ was put up in the form \begin{align*} \displaystyle \hat{u}\left(x,y\right)=\sum_{\substack{1\leq i,j\leq N_{u}\\i,j{\rm~are~odd}}}a_{i,j}\varphi_{i,j}(x,y), \end{align*} where $N_{u}=60.$ Using Theorem \ref{plum2001} and Corollary \ref{Linfcoro}, we proved the existence of a solution $u$ to \eqref{absproblem} in an $H_{0}^{1}$-ball $\overline{B}(\hat{u},r_{1};\ \left\|\cdot\right\|_{V})$ and an $L^{\infty}$-ball $\overline{B}(\hat{u},r_{2};\ \left\|\cdot\right\|_{L^{\infty}(\Omega)})$, where $\overline{B}(x,r;\ \left\|\cdot\right\|)$ denotes the closed ball whose center is $x$, and whose radius is $r\geq 0$ with respect to the norm $\left\|\cdot\right\|$. Table \ref{veri/result} presents the verification result, which ensures positiveness of the verified solution $u$ owing to the condition $\sup\{\sqrt{u_{-}\left(x\right)}\,|\,x\in\Omega_{s}\}\leq\lambda_{1}(=2\pi^2)$, and therefore, it is also a (strong) solution to \eqref{positiveproblem}. Here, the upper bound of $\sup\{\sqrt{u_{-}\left(x\right)}\,|\,x\in\Omega_{s}\}$ was calculated by $\displaystyle [|\min\{\hat{u}(x)\ :\ x\in\Omega_{s}\}|+r_{2}]^{p-1}$ with verification. Note also that $u\in C^{2}(\Omega)$ by local regularity, and $u\in C(\overline{\Omega})$ due to the embedding $H^{2}(\Omega) \hookrightarrow C(\overline{\Omega})$, which indeed allows application of Theorem \ref{positive/theo}. \begin{figure}[h] \begin{center} \vspace{-20mm}\includegraphics[height=120mm]{appsol1.eps} \end{center} \vspace{5mm}\caption{Approximate solution to \eqref{positiveproblem} on $\Omega_{s}(:=(0,1)^{2})$, the amplitude of which is proved to be in the interval [575.15, 575.61]. }\label{pic} \end{figure} \begin{table}[h] \caption{Verification result for \eqref{positiveproblem}.} \label{veri/result} \begin{center} \renewcommand\arraystretch{1.3} \footnotesize \begin{tabular}{cccccl} \hline $\|\Delta\hat{u}+\left|\hat{u}\right|^{\frac{1}{2}}\hat{u}\|_{L^{2}}$& $\delta$& $K\ (N=14)$& $r_{1}$& $r_{2}$& $\displaystyle \sup\,\sqrt{u_{-}(x)}$\\ \hline \hline [0.8311281,~0.8314938]& 0.1871519& 2.0000005& 0.3909193& 1.1462326& 1.0706226\\ \hline \end{tabular} \end{center} \vspace{-2mm}\hspace*{2cm}\small{$\delta,\ K$: the constants required in Theorem \ref{plum2001}.\\ \hspace*{2cm}The value of $\|\Delta\hat{u}+\left|\hat{u}\right|^{\frac{1}{2}}\hat{u}\|_{L^{2}}$ is proved to be in the displayed interval. The other numerical\\ \hspace*{2cm}values represent upper bounds of the corresponding constants. } \end{table} \section{Verified numerical integration}\label{integration} To apply Theorem \ref{plum2001} to problem \eqref{absproblem}, one has to construct a ``good'' approximation $\hat{u}\in V$ of a solution to \eqref{absproblem} such that $\delta$ in \eqref{zansa} is sufficiently small. In this paper, we assume that such an approximation $\hat{u}$ is constructed by a finite linear combination of basis functions $\left\{\phi_{i}\right\}_{i=1}^{\infty}$ that span $V$, where each $\phi_{i}$ is in $C^{\infty}(\overline{\Omega})$ (and therefore, $\hat{u}\in C^{\infty}(\overline{\Omega})$). To obtain concrete bounds for $\delta$ and $K$ required in Theorem \ref{plum2001}, one has to compute, in particular, $(\Delta\hat{u},\left|\hat{u}\right|^{p-1}\hat{u})_{L^{2}(\Omega)}$ and $(\phi_{i},\left|\hat{u}\right|^{p-1}\phi_{j})_{L^{2}(\Omega)}$ with verification (recall that $1<p<2$ which makes this integration non-trivial). In this section, for the square $\Omega_{s}=(0,1)^{2}\subset \mathbb{R}^{2}$, $0<q<1$, and $\eta,$\,$\xi\in C^{\infty}(\overline{\Omega}_{s})$, we propose a method for computing the integral \begin{align*} I=\displaystyle \int_{\Omega_{s}}\left\{\eta(x,y)\right\}^{q}\xi(x,y)dxdy, \end{align*} in verified form, i.e., for computing an enclosure for this integral, where we assume that $\eta>0$ in $\Omega_{s}$ and $\eta=0$ on $\partial\Omega_{s}$; indeed, we later select $\eta=\hat{u}$, an approximate solution to \eqref{positiveproblem}, which has these properties. We prove the positivity of $\hat{u}$ in $\Omega$ using the procedures described in Subsections \ref{intS11} to \ref{intS00}. There are some verified integration methods that can be applied to such an integration, under the assumption that $\eta>0$ on the whole closure of a domain $\Omega\subset \mathbb{R}^{2}$ (see, e.g, \cite{storck1993numerical}). However, since here the derivative of $\left\{\eta(\cdot,\cdot)\right\}^{q}:\Omega\rightarrow \mathbb{R}$ is in general not bounded near the boundary $\partial\Omega$, where $\eta$ vanishes, previous methods cannot be applied in our situation. To overcome this difficultly, we employ a Taylor expansion based method as follows: We first divide $\Omega_{s}$ into four sub-squares, and consider the integration over $\Omega_{s/4}:=(0,1/2)^{2}$; integration over the three other parts can be carried out similarly, after translation and rotation such that $\eta=0$ on both the left and the lower edge. Moreover, we divide $\overline{\Omega_{s/4}}$ into closed rectangles that are grouped into four types ($S_{1,1},\ S_{1,0},\ S_{0,1}$, and $S_{0,0}$) as in Fig.~\ref{figrectangle}. \begin{figure}[h] \begin{center} \includegraphics[height=100mm]{square1.eps} \end{center} \caption{Division of the domain.} \label{figrectangle} \end{figure} These types of rectangles have the following properties: \begin{itemize} \setlength{\leftskip}{0.5cm} \item[$S_{1,1}$:]a square where $\eta$ is zero on both the left and the lower edge; \item[$S_{0,1}$:]rectangles where $\eta$ is zero only on the lower edge; \item[$S_{1,0}$:]rectangles where $\eta$ is zero only on the left edge; \item[$S_{0,0}$:]squares where $\eta>0$. \end{itemize} Then, the integration over $\Omega_{s/4}$ can be expressed by summation of integrations over the above four types of rectangles. Therefore, we discuss an integration method over the four types of domains. Hereafter, we employ the notation $\Lambda_{n}^{1,1}=\{(i,j)\in \mathbb{N}^{2}\ :\ i\leq n,\ j\leq n\}$, $\Lambda_{n}^{0,1}=\{(i,j)\in \mathbb{N}_{0}\times \mathbb{N}\ :\ i\leq n,\ j\leq n\}$, $\Lambda_{n}^{1,0}=\{(i,j)\in \mathbb{N}\times \mathbb{N}_{0}\ :\ i\leq n,\ j\leq n\}$, and $\Lambda_{n}=\{(i,j)\in \mathbb{N}_{0}^{2}\ :\ i\leq n,\ j\leq n\}$, where $\mathbb{N}=\{1,2,3\cdots\}$ and $\mathbb{N}_{0}=\{0,1,2\cdots\}$. \subsection{Integration over $S_{1,1}$}\label{intS11} On the basis of the Taylor expansion around the lower left corner $(0,0)$, we enclose $\eta(x,y)$ as \begin{align} \displaystyle \eta(x,y)\in\sum_{(i.j)\in\Lambda_{n-1}^{1,1}}a_{i,j}x^{i}y^{j}+\sum_{(i,j)\in\Lambda_{n}^{1,1}\backslash\Lambda_{n-1}^{1,1}}[\underline{a}_{i,j},\ \overline{a}_{i,j}]x^{i}y^{j},\label{taylorexpansionS11} \end{align} for $(x,y)\in S_{1,1}$, where $a_{i,j},\,\underline{a}_{i,j},\,\overline{a}_{i,j}\in \mathbb{R},\ \underline{a}_{i,j}\leq\overline{a}_{i,j}$. In Section \ref{psa}, we will introduce a numerical method (Type-II PSA) for deriving such an enclosure. We then denote \begin{align*} [\displaystyle \eta_{1,1}(x,y)]:=\sum_{(i,j)\in\Lambda_{n-1}^{1,1}}a_{i,j}x^{i-1}y^{j-1}+\sum_{(i,j)\in\Lambda_{n}^{1,1}\backslash\Lambda_{n-1}^{1,1}}[\underline{a}_{i,j},\ \overline{a}_{i,j}]x^{i-1}y^{j-1}, \end{align*} which more precisely means the set of all continuous functions $w$ over $S_{1,1}$ such that $w(x,y)\in[\eta_{1,1}(x,y)]$ for all $(x,y)\in S_{1,1}$. Therefore $\eta(x,y)\in xy[\eta_{1,1}(x,y)]$. We moreover assume that $[\eta_{1,1}(x,y)]$ is positive in $S_{1,1}$ (i.e., $z>0$ for all $z\in[\eta_{1,1}(x,y)],\ (x,y)$ $\in S_{1,1}$); if $S_{1,1}$ is sufficiently small and $n$ is sufficiently large, this positivity condition is expected to hold for $\eta=\hat{u}$ (in the actual computation, this condition will be numerically checked by suitable interval arithmetic techniques \cite{moore2009introduction, rump1999book}). Under this assumption, we use Type-II PSA first to enclose $[\eta_{1,1}(x,y)]^{q}$, and then, in a second step, to enclose $[\eta_{1,1}(x,y)]^{q}\xi(x,y)$ as \begin{align*} [\displaystyle \eta_{1,1}(x,y)]^{q}\xi(x,y)\in\sum_{(i,j)\in\Lambda_{n-1}}b_{i,j}x^{i}y^{j}+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}[\underline{b}_{i,j},\ \overline{b}_{i,j}]x^{i}y^{j}, \end{align*} where $b_{i,j},\,\underline{b}_{i,j},\,\overline{b}_{i,j}\in \mathbb{R},\ \underline{b}_{i,j}\leq\overline{b}_{i,j}$. Hence, the integration over $S_{1,1}$ is enclosed as \begin{align} &\displaystyle \int_{S_{1,1}}\left\{\eta(x,y)\right\}^{q}\xi(x,y)dxdy\nonumber\\ &\displaystyle \in\sum_{(i,j)\in\Lambda_{n-1}}\int_{S_{1,1}}b_{i,j}x^{i+q}y^{j+q}dxdy+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}\int_{S_{1,1}}[\underline{b}_{i,j},\ \overline{b}_{i,j}]x^{i+q}y^{j+q}dxdy.\label{exintS11} \end{align} \begin{rem} We wish to make a remark about integration of a set of continuous functions as in \eqref{exintS11}, i.e., we provide an explanation of verified integration of \begin{align*} \displaystyle \int_{y_{1}}^{y_{2}}\int_{x_{1}}^{x_{2}}ax^{p}y^{q}dxdy, \end{align*} where generally, $a,\ p,\ q,\ x_{1},\ x_{2},\ y_{1},$ and $y_{2}$ are real intervals. Note that $ax^{p}y^{q}$ precisely means the set of all continuous functions $w$ over $(\underline{x_{1}},\overline{x_{2}})\times(\underline{y_{1}},\overline{y_{2}})$ such that $w(x,y)\in ax^{p}y^{q}$ for all $(x,y)\in(\underline{x_{1}},\overline{x_{2}})\times(\underline{y_{1}},\overline{y_{2}})$, where we denote $\displaystyle \underline{z}=\inf z$ and $\displaystyle \overline{z}=\sup z$ for an interval $z$. While formally the integral is simply computed as \begin{align*} \displaystyle \int_{y_{1}}^{y_{2}}\int_{x_{1}}^{x_{2}}ax^{p}y^{q}dxdy=&\displaystyle \frac{a}{(p+1)(q+1)}y_{2}^{q+1}x_{2}^{p+1}-\frac{a}{(p+1)(q+1)}y_{2}^{q+1}x_{1}^{p+1}\\ &-\left(\frac{a}{(p+1)(q+1)}y_{1}^{q+1}x_{2}^{p+1}-\frac{a}{(p+1)(q+1)}y_{1}^{q+1}x_{1}^{p+1}\right), \end{align*} one has to compute the above formula in correct order using suitable interval arithmetic techniques, because the distributive law does not hold in interval arithmetics. For example, $\int_{-1}^{1}[0.8,1]$ $xdx$ is not zero, but is correctly computed as \begin{align*} \displaystyle \int_{-1}^{1}[0.8,1]xdx=\left[[0.4,0.5]x^{2}\right]_{-1}^{1}=[0.4,0.5]-[0.4,0.5]=[-0.1,0.1]. \end{align*} \end{rem} \subsection{Integration over $S_{0,1}$ and $S_{1,0}$}\label{intS10} Let $(x_{0},0)$ be the midpoint of the lower edge of $S_{0,1}$. We denote $\eta^{*}(x,y):=\eta(x+x_{0},y),\ \xi^{*}(x,y):=\xi(x+x_{0},y)$, and $S_{0,1}^{*}:=S_{0,1}-(x_{0},0)$. Since we have \begin{align*} \displaystyle \int_{S_{0,1}}\left\{\eta(x,y)\right\}^{q}\xi(x,y)dxdy=\int_{S_{0,1}^{*}}\left\{\eta^{*}(x,y)\right\}^{q}\xi^{*}(x,y)dxdy, \end{align*} we consider the right integral in this subsection. By Taylor expanding $\eta^{*}(x,y)$ around the midpoint $(0,0)$ of the lower edge of $S_{0,1}^{*}$, we enclose $\eta^{*}(x,y)$ as \begin{align*} \displaystyle \eta^{*}(x,y)\in\sum_{(i.j)\in\Lambda_{n-1}^{0,1}}a_{i,j}x^{i}y^{j}+\sum_{(i,j)\in\Lambda_{n}^{0,1}\backslash\Lambda_{n-1}^{0,1}}[\underline{a}_{i,j},\ \overline{a}_{i,j}]x^{i}y^{j}, \end{align*} for $(x,y)\in S_{0,1}^{*}$, where $a_{i,j},\,\underline{a}_{i,j},\,\overline{a}_{i,j}\in \mathbb{R},\ \underline{a}_{i,j}\leq\overline{a}_{i,j}$. We then denote \begin{align*} [\displaystyle \eta_{0,1}^{*}(x,y)]:=\sum_{(i,j)\in\Lambda_{n-1}^{0,1}}a_{i,j}x^{i}y^{j-1}+\sum_{(i,j)\in\Lambda_{n}^{0,1}\backslash\Lambda_{n-1}^{0,1}}[\underline{a}_{i,j},\ \overline{a}_{i,j}]x^{i}y^{j-1} \end{align*} (therefore, $\eta^{*}(x,y)\in y[\eta_{0,1}^{*}(x,y)]$), and again assume that $[\eta_{0,1}^{*}(x,y)]$ is positive in $S_{0,1}^{*}$. We then again enclose $[\eta_{0,1}^{*}(x,y)]^{q}\xi^{*}(x,y)$ as \begin{align*} [\displaystyle \eta_{0,1}^{*}(x,y)]^{q}\xi^{*}(x,y)\in\sum_{(i,j)\in\Lambda_{n-1}}b_{i,j}x^{i}y^{j}+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}[\underline{b}_{i,j},\ \overline{b}_{i,j}]x^{i}y^{j}, \end{align*} where $b_{i,j},\,\underline{b}_{i,j},\,\overline{b}_{i,j}\in \mathbb{R},\ \underline{b}_{i,j}\leq\overline{b}_{i,j}$. Thus, we can enclose the integral over $S_{0,1}^{*}$ as \begin{align*} &\displaystyle \int_{S_{0,1}^{*}}\left\{\eta^{*}(x,y)\right\}^{q}\xi^{*}(x,y)dxdy\\ &\displaystyle \in\sum_{(i,j)\in\Lambda_{n-1}}\int_{S_{0,1}^{*}}b_{i,j}x^{i}y^{j+q}dxdy+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}\int_{S_{0,1}^{*}}[\underline{b}_{i,j},\ \overline{b}_{i,j}]x^{i}y^{j+q}dxdy. \end{align*} The integration over $S_{1,0}$ is carried out similarly by exchanging the roles of the variables $x$ and $y$. \subsection{Integration over $S_{0,0}$}\label{intS00} Let $(x_{0},y_{0})$ be the center of $S_{0,0}$, and we re-define $\eta^{*}(x,y):=\eta(x+x_{0},y+y_{0}),\ \xi^{*}(x,y):=\xi(x+x_{0},y+y_{0})$, and $S_{0,0}^{*}:=S_{0,0}-(x_{0},y_{0})$. Since we have \begin{align*} \displaystyle \int_{S_{0,0}}\left\{\eta(x,y)\right\}^{q}\xi(x,y)dxdy=\int_{S_{0,0}^{*}}\left\{\eta^{*}(x,y)\right\}^{q}\xi^{*}(x,y)dxdy, \end{align*} we consider the right integral in this subsection. By Taylor expanding $\eta^{*}(x,y)$ around the center $(0,0)$ of $S_{0,0}^{*}$, we have \begin{align*} \displaystyle \eta^{*}(x,y)\in[\eta_{0,0}^{*}(x,y)]:=\sum_{(i.j)\in\Lambda_{n-1}}a_{i,j}x^{i}y^{j}+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}[\underline{a}_{i,j},\ \overline{a}_{i,j}]x^{i}y^{j}, \end{align*} for $(x,y)\in S_{0,0}^{*}$, where $a_{i,j},\,\underline{a}_{i,j},\,\overline{a}_{i,j}\in \mathbb{R},\ \underline{a}_{i,j}\leq\overline{a}_{i,j}$. Assuming that $[\eta_{0,0}^{*}(x,y)]$ is positive on $S_{0,0}^{*}$ (since $\eta>0$ on $S_{0,0}^{*}$, this property is expected to hold if $n$ is sufficiently large), we have \begin{align*} [\displaystyle \eta_{0,0}^{*}(x,y)]^{q}\xi^{*}(x,y)\in\sum_{(i,j)\in\Lambda_{n-1}}b_{i,j}x^{i}y^{j}+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}[\underline{b}_{i,j},\ \overline{b}_{i,j}]x^{i}y^{j}, \end{align*} where $b_{i,j},\,\underline{b}_{i,j},\,\overline{b}_{i,j}\in \mathbb{R},\ \underline{b}_{i,j}\leq\overline{b}_{i,j}$. Thus, we have \begin{align*} &\displaystyle \int_{S_{0,0}^{*}}\left\{\eta^{*}(x,y)\right\}^{q}\xi^{*}(x,y)dxdy\\ &\displaystyle \in\sum_{(i,j)\in\Lambda_{n-1}}\int_{S_{0,0}^{*}}b_{i,j}x^{i}y^{j}dxdy+\sum_{(i,j)\in\Lambda_{n}\backslash\Lambda_{n-1}}\int_{S_{0,0}^{*}}[\underline{b}_{i,j},\ \overline{b}_{i,j}]x^{i}y^{j}dxdy. \end{align*} \begin{rem} Integration over $S_{0,0}$ can also be carried out using common methods $($see, e.g., {\rm \cite{storck1993numerical}}$)$. \end{rem} \section{Introduction}\label{intro} We are concerned with a verified numerical computation method for the following elliptic problem:\begin{align} \left\{\begin{array}{l l} -\Delta u=f\left(u\right) &\mathrm{i}\mathrm{n}\ \Omega,\\ u=0 &\mathrm{on}\ \partial\Omega, \end{array}\right.\label{gproblem} \end{align} where $\Omega\subset \mathbb{R}^{n}$ ($n=1,2,3$) is a given domain (i.e., an open connected set) and $f:H_{0}^{1}\left(\Omega\right)\rightarrow L^{2}\left(\Omega\right)$ is a given nonlinear operator. Here, assuming that $H^{1}\left(\Omega\right)$ denotes the first order $L^{2}$-Sobolev space on $\Omega$, we define $ H_{0}^{1}\left(\Omega\right):=\{u\in H^{1}\left(\Omega\right)\ :\ u=0\ \mathrm{on}\ \partial\Omega$ in the trace sense$\}$ with inner product $\left(\cdot,\cdot\right)_{H_{0}^{1}\left(\Omega\right)}:=\left(\nabla\cdot,\nabla\cdot\right)_{L^{2}\left(\Omega\right)}$ and norm $\left\|\cdot\right\|_{H_{0}^{1}\left(\Omega\right)}:=\left\|\nabla\cdot\right\|_{L^{2}\left(\Omega\right)}$. Hereafter, we denote $V=H_{0}^{1}\left(\Omega\right)$, and $V^{*}=H^{-1}\left(\Omega\right)$($:=$(dual of $V$)) with the usual sup-norm. Moreover, the $L^{2}$-inner product is simply denoted by $\left(\cdot,\cdot\right)$ if no confusion arises. Verified numerical computation methods for differential equations originate from Nakao's \cite{nakao1988numerical} and Plum's work \cite{plum1991computer}, and have been further developed by many researchers. Moreover, the applicability of such methods to semilinear elliptic boundary value problems has been investigated (see, e.g., \cite{nakao2001numerical, nakao2011numerical, plum2001computer, plum2008, takayasu2014remarks}). In their frameworks, \eqref{gproblem} is transformed into a suitable operator equation for proving the existence of a solution close to a computed numerical approximation. In this paper, by defining $\mathcal{F}:V\rightarrow V^{*}$ as \begin{align*} \left\langle\mathcal{F}(u),v\right\rangle:=\left(\nabla u,\nabla v\right)-\left(f\left(u\right),v\right)~~~{\rm for}~u,v\in V, \end{align*} we first re-write \eqref{gproblem} as \begin{align} \mathcal{F}(u)=0~{\rm in}~V^{*},\label{gFproblem} \end{align} and discuss the verified numerical computation for \eqref{gFproblem}. In other words, we first consider the existence of a weak solution to \eqref{gproblem} (a solution to \eqref{gFproblem} in $V$), and then we discuss its $H^{2}$-regularity if necessary. In particular, we select $f\left(u\right)=u^{p} \left(1<p<2\right)$ as a Fr\'echet differentiable operator, the Fr\'echet derivative of which is, however, not Lipschitz continuous. We are looking for positive solutions to \eqref{gproblem}, that is, we consider the following problem: \begin{align} \left\{\begin{array}{l l} -\Delta u=u^{p} &\mathrm{i}\mathrm{n}\ \Omega,\\ u>0 &\mathrm{i}\mathrm{n}\ \Omega,\\ u=0 &\mathrm{on}\ \partial\Omega. \end{array}\right.\label{positiveproblem} \end{align} It is well known that the Fr\'echet differentiability of $f$ (or of $\mathcal{F}$) is essential for the existing theories of verified numerical computations for solutions to \eqref{gproblem} (see again, e.g., \cite{nakao2001numerical, nakao2011numerical, plum2001computer, plum2008, takayasu2014remarks}). Moreover, the Lipschitz continuity of the Fr\'echet derivative of $f$ has been required in real examples, mainly for obtaining convenient mapping properties of the fixed point operator to be constructed, but also to avoid some technical difficulties in computing integrals (often they are needed with verification), the integrands of which contain $f'\left(v_{N}\right)$ and $f\left(v_{N}\right)$ for $v\in V_{N}$; $V_{N}$ is a finite dimensional subspace of $V$, e.g., spanned by a finite element basis or a Fourier-Galerkin basis. For example, when we set $f\left(u\right)=u^{p} \left(1<p<2\right)$ as mentioned above, such integrations are difficult to calculate with high-precision as well as difficult to estimate with verification, since, even for smooth functions $u$, the second derivative of $f\left(u(\cdot)\right)$ is not bounded near points $x\in \mathbb{R}^{n}$ such that $u\left(x\right)=0$. Such integrations are required at many points in the verification process, e.g., when we estimate the norm of the residual $\left\|\mathcal{F}\left(\hat{u}\right)\right\|_{V^{*}}$ for some approximation $\hat{u}\in V$ with verification, and when we compute verified bounds for the operator norm of the inverse of $\mathcal{F}_{\hat{u}}':V\rightarrow V^{*}$, where $\mathcal{F}_{\hat{u}}'$ is the Fr\'echet derivative of $\mathcal{F}$ at $\hat{u}\in V$. In this paper, we apply Plum's theorem \cite{plum2001computer} (see Theorem \ref{plum2001}) to the verified numerical computation for a solution to \eqref{positiveproblem} with $p\in(1,2)$. To be precise, we prove the existence of a solution to \eqref{positiveproblem}, on the basis of Theorem \ref{plum2001}, in balls centered around a numerically computed approximate solution, in the sense of both norms $\left\|\nabla\cdot\right\|_{L^{2}\left(\Omega\right)}$ and $\left\|\cdot\right\|_{L^{\infty}\left(\Omega\right)}$. For this purpose, we first try to obtain a numerical inclusion of a solution to \eqref{gproblem} with $f\left(u\right)=\left|u\right|^{p-1}u$, i.e., a solution to \begin{align} \left\{\begin{array}{l l} -\Delta u=\left|u\right|^{p-1}u &\mathrm{i}\mathrm{n}\ \Omega,\\ u=0 &\mathrm{on}\ \partial\Omega. \end{array}\right.\label{absproblem} \end{align} After that, we confirm its positiveness with a simple computation in order to verify a solution to \eqref{positiveproblem} (see Section \ref{positiveness} for a method for confirming the positiveness). We remark that, when the Fr\'echet derivative of $f$ is not Lipschitz continuous, a function that replaces the Lipschitz constant of the Fr\'echet derivative of $f$ is required to be concretely constructed; such a function will be denoted by $g$ in Theorem \ref{plum2001}. We also propose a concrete construction of such a function for the $u^{p}$-nonlinearity in Section \ref{verificationtheory}. A numerical integration method for integrands arising from the $u^{p}$-nonlinearity with $p\in(1,2)$ will be proposed in Section \ref{integration}. \section{Verification of positiveness}\label{positiveness} One can prove positiveness of a (strong) solution to \eqref{absproblem} using the following theorem. \begin{theo}\label{positive/theo} Let $\Omega$ be a bounded domain in $\mathbb{R}^{n} (n=1,2,3,\cdots)$. If a solution $u\in C^{2}\left(\Omega\right)\cap C\left(\overline{\Omega}\right)$ to \eqref{absproblem} is positive in a nonempty subdomain $\Omega'\subset\Omega$ and $\displaystyle \sup\{(u_{-}(x))^{p-1}\,|\, x\in\Omega\}<\lambda_{1}(\Omega)$, then $u>0$ in the original domain $\Omega$; that is, $u$ is also a solution to \eqref{positiveproblem}. Here, $\lambda_{1}(\Omega)>0$ is the first eigenvalue of the problem \begin{align*} \left(\nabla u,\nabla v\right)_{L^{2}\left(\Omega\right)}=\lambda\left(u,v\right)_{L^{2}\left(\Omega\right)},~~\forall v\in V. \end{align*} and $u_{-}$ is defined by \begin{align*} u_{-}\left(x\right):= \left\{\begin{array}{l l} -u\left(x\right), &u\left(x\right)<0,\\ 0, &u\left(x\right)\geq 0. \end{array}\right. \end{align*} \end{theo} A proof can be found in \cite{tanaka2016sharp,tanaka2015numerical}. \section{Power series arithmetic}\label{psa} Two types of Power Series Arithmetic (called Type-I PSA and Type-II PSA) were proposed by Kashiwagi \cite{kashiwagi1995proc, kashiwagi1995power}, and have been packaged in \cite{kashiwagikv}. Both PSAs were originally designed to perform operations for sets of continuous functions defined on a closed interval $D=[\underline{d},\overline{d}]$ with $\underline{d},\,\overline{d}\in \mathbb{R}$, written in the form \begin{align} [u(x)]=\displaystyle \sum_{i=0}^{n}u_{i}x^{i}:=\left\{v\in C(D)\ :\ v(x)\in\sum_{i=0}^{n}u_{i}x^{i}~~\forall x\in D\right\}.\label{psaform} \end{align} where each $u_{i}$ ($i=0,1,2,\cdots,n$) is a real number or a real interval $[\underline{u_{i}},\overline{u_{i}}],\ \underline{u_{i}}\leq\overline{u_{i}}$. Type-I PSA performs such operations with neglecting terms of degree higher than $n$. Therefore, Type-I PSA gives approximate results of the operations. On the other hand, Type-II PSA gives a verified result of such operations, that is, the operation result from Type-II PSA includes the correct operation result in a strict mathematical sense. In this section, we introduce the original Type-II PSA in the one-dimensional case together with some operation examples. Subsequently, we present a generalization of Type-II PSA to the higher-dimensional cases in order to obtain a verified inclusion such as \eqref{taylorexpansionS11}. \subsection{Type-II PSA in the one-dimensional case} We consider a verified operation method for a set of continuous functions, written in the form \eqref{psaform}. The addition operation and the subtraction operation are respectively performed as \begin{align*} [u(x)]+[v(x)]=\displaystyle \sum_{i=0}^{n}(u_{i}+v_{i})x^{i}, \shortintertext{and} [u(x)]-[v(x)]=\displaystyle \sum_{i=0}^{n}(u_{i}-v_{i})x^{i}. \end{align*} The multiplication operation is performed as follows. We first multiply $[u(x)]$ and $[v(x)]$ without degree omissions: \begin{align*} [u(x)]\displaystyle \times[v(x)]=\sum_{i=0}^{2n}w_{i}x^{i},~~w_{k}=\displaystyle \sum_{i=\max(0,k-n)}^{\min(k,n)}u_{i}v_{k-i}. \end{align*} Then, we reduce its degree from $2n$ to $n$ on the basis of the degree reduction defined as follows. \begin{defi}[Degree reduction]\label{degred} For a power series $[u(x)]=u_{0}+u_{1}x+\cdots+u_{m}x^{m}$ over $D$, the degree reduction $[v(x)]$ to $n$~$(n<m)$ is defined by \begin{align*} [v(x)]=\displaystyle \sum_{i=0}^{n}v_{i}x^{i}, \end{align*} where \begin{align*} v_{i}=u_{i}~(i=0,1,\cdots,n-1)~~{\rm and}~~v_{n}=\left\{\sum_{i=n}^{m}u_{i}x^{i-n}\ |\ x\in D\right\}. \end{align*} \end{defi} Thus, the terms of degree more than $n$ are resorbed in the term of degree $n$. Therefore, the result of the multiplication by Type-II PSA includes the correct multiplication result. \begin{rem}\label{horner} When computing \begin{align*} \left\{\sum_{i=n}^{m}u_{i}x^{i-n}\ |\ x\in D\right\}, \end{align*} one has to evaluate the range of the polynomial $u_{n}+u_{n+1}x+\cdots+u_{m}x^{m-n}$. Since the common interval arithmetic occasionally over-estimates the range, one should use a method that gives the range more accurately, e.g., the Horner scheme, in order to obtain a precise multiplication result. \end{rem} We then apply Type-II PSA to $C^{\infty}$-functions (e.g, $\log(\cdot)$ and $\sin(\cdot)$) on the basis of the Taylor expansion with a remainder term. For a $C^{\infty}$-function $f,\ f(u_{0}+u_{1}x+\cdots+u_{n}x^{n})$ is computed as \begin{align} &f(u_{0}+u_{1}x+\cdots+u_{n}x^{n})\nonumber\\ \subset&f(u_{0})+\displaystyle \sum_{i=1}^{n-1}\frac{1}{i!}f^{(i)}(u_{0})(u_{1}x+\cdots+u_{n}x^{n})^{i}\nonumber\\ &~~~~~~~~~~~~~~+\displaystyle \frac{1}{n!}f^{(n)}\left({\rm hull}_{\,}\left( u_0, \left\{\sum_{i=0}^{n}u_{i}x^{i}\ |\ x\in D\right\}\right)\right)(u_{1}x+\cdots+u_{n}x^{n})^{n},\label{eq:hull} \end{align} by Taylor expanding $f$ around $u_{0}$, where hull\,($a,b$) denotes the convex hull of real numbers or real intervals $a$ and $b$. Here, additions, subtractions, and multiplications in the above process are operated by Type-II PSA defined so far, and the expression \begin{align*} \left\{\sum_{i=0}^{n}u_{i}x^{i}\ |\ x\in D\right\} \end{align*} is similarly computed as mentioned in Remark \ref{horner}. The division can be operated as $[u]/[v]:=[u]\times f([v])$ with $f(x)=1/x$, using the above method. \begin{rem} In our examples, the interval $D$ in \eqref{eq:hull} contains zero in all cases; indeed, in the integration procedures described in Section \ref{integration}, the domains $S_{0,1}$, $S_{1,0}$, and $S_{0,0}$ of integrations are translated to contain $(0,0)$ $($Type-II PSA in the two-dimensional case will be introduced in \ref{subsec:higherdimPSA} using the one-dimensional method$)$. Hence, in our examples, \begin{align*} {\rm hull}_{\,}\left( u_0, \left\{\sum_{i=0}^{n}u_{i}x^{i}\ |\ x\in D\right\}\right) = \left\{\sum_{i=0}^{n}u_{i}x^{i}\ |\ x\in D\right\} \end{align*} always holds in \eqref{eq:hull}. \end{rem} \begin{rem} Basically, Type II-PSA is designed to ensure that the coefficients of degree less than $n$ are points $($real numbers$)$, and the coefficient of degree $n$ is a real interval. However, in an actual computation, in order to strictly verify all results from Type II-PSA, the coefficients of degree less than $n$ are often intervals that arise only from rounding error enclosures. \end{rem} \subsection{Examples of Type-II PSA} Here, we present simple examples of Type-II PSA where $($degree of PSA $n)=2$ and $D=[0,0.1],\ [u(x)]=1+2x-3x^{2}$ and $[v(x)]=1-x+x^{2}$. The addition operation and the subtraction operation are respectively performed as $[u(x)]+[v(x)]=2+x-2x^{2}$ and $[u(x)]-[v(x)]=0+3x-4x^{2}$. The multiplication operation is performed as follows. We first multiply them as \begin{align*} [u(x)]\times[v(x)]&=1+x-4x^{2}+5x^{3}-3x^{4}\\ &=1+x+(-4+5x-3x^{2})x^{2} \end{align*} without degree omissions. Since interval arithmetic gives \begin{align*} -4+(5-3\times[0,0.1])\times[0,0.1]\subset-4+[0,0.5]\subset[-4,-3.5], \end{align*} we determine the multiplication result as \begin{align*} [u(x)]\times[v(x)]=1+x-[-4,-3.5]x^{2}, \end{align*} on the basis of the the degree reduction in Definition \ref{degred}. The computation $\log([u(x)])$ is performed as follows. The range of $[u(x)]$ is computed as \begin{align*} 1+(2-3\times[0,0.1])\times[0,0.1]\subset 1+[0,0.2]\subset[1,1.2]. \end{align*} We then compute the second degree Taylor expansion, with a remainder term, of $\log(t)$ around $1$ (the constant term of $[u(x)]$) on $[1,1.2]$ as \begin{align*} 0+(t-1)-\displaystyle \frac{1}{2[1,1.2]^{2}}(t-1)^{2}. \end{align*} By substituting $[u(x)]$ for $t$ in this expansion, we have \begin{align*} 0+(2x-3x^{2})-\displaystyle \frac{1}{2[1,1.2]^{2}}(2x-3x^{2})^{2}. \end{align*} Consequently, by reducing this expression using Type-II PSA, we have \begin{align*} \log([u(x)])=0+2x+\left[-5,-\frac{143}{36}\right]x^{2}. \end{align*} \subsection{Type-II PSA in the higher-dimensional cases}\label{subsec:higherdimPSA} One-dimensional Type-II PSA is designed for power series that have real or real interval coefficients. In fact, the set of coefficients in Type-II PSA can be generalized to any set equipped with the four arithmetic operations. Moreover, the set of power series is endowed with the four arithmetic operations by Type-II PSA. Therefore, Type-II PSA can be generalized to two-dimensional cases by replacing its coefficients with one-dimensional power series. To be precise, by replacing each coefficient $u_{i}$ in \begin{align} [u(x)]=\displaystyle \sum_{i=0}^{n}u_{i}x^{i},~~x\in D_{u}\label{psau} \end{align} with one-dimensional power series \begin{align*} [v_{i}(y)]=\displaystyle \sum_{j=0}^{n}v_{i,j}y^{j},~~y\in D_{v}, \end{align*} we can regard $[u]$ as a two-dimensional power series \begin{align} [u(x,y)]=\displaystyle \sum_{i=0}^{n}[v_{i}(y)]x^{i}=\sum_{i=0}^{n}\sum_{j=0}^{n}v_{i,j}x^{i}y^{j},~~(x,y)\in D_{u}\times D_{v}.\label{w} \end{align} Thus, Type-II PSA for the one-dimensional case is naturally carried over to the two-dimensional case. In the same way, Type-II PSA can be also applied to higher-dimensional cases, that is, by replacing each coefficient $u_{i}$ in \eqref{psau} with $n$-dimensional power series, $(n+1)$-dimensional power series with the four arithmetic operations are defined. \section{Verification theorem for elliptic problems}\label{verificationtheory} In this section, we apply the method summarized in \cite{plum1992explicit, plum2001computer, plum2008} to a verified numerical computation for solutions to \eqref{gproblem}. Throughout this paper, the norm bound for the embedding $V\hookrightarrow L^{p}\left(\Omega\right)$ is denoted by $C_{p}$, i.e., $C_{p}$ is a positive number that satisfies \begin{align} \left\|u\right\|_{L^{p}(\Omega)}\leq C_{p}\left\|u\right\|_{V}~~~{\rm for~all}~u\in V.\label{embedding} \end{align} Since a concrete upper bound for $C_{p}$ is important for the verification theory, a formula that gives such an upper bound for a given bounded domain $\Omega$ is provided in Corollary \ref{roughboundtheo}. \subsection{$H_{0}^{1}$ error estimation} We use the following verification theorem for obtaining $H_{0}^{1}$ error estimations for solutions to \eqref{gFproblem}, i.e., weak solutions to \eqref{gproblem}. \begin{theo}[\cite{plum2001computer, plum2008}]\label{plum2001} Let $\mathcal{F}:V\rightarrow V^{*}$ be a Fr\'echet differentiable operator. Suppose that $\hat{u}\in V$, and that there exist $\delta>0,\ K>0$, and a non-decreasing function $g$ satisfying \begin{align} &\left\|\mathcal{F}\left(\hat{u}\right)\right\|_{V^{*}}\leq\delta,\label{zansa}\\ &\left\|u\right\|_{V}\leq K\left\|\mathcal{F}_{\hat{u}}'u\right\|_{V^{*}}~~~{\rm for~all}~u\in V,\label{inverse}\\ &\left\|\mathcal{F}_{\hat{u}+u}'-\mathcal{F}_{\hat{u}}'\right\|_{\mathcal{B}(V,V^{*})}\leq g\left(\left\|u\right\|_{V}\right)~~~{\rm for~all}~u\in V,\label{lip} \shortintertext{and} &g(t)\rightarrow 0~~{\rm as}~~t\rightarrow 0\label{gconv}. \end{align} Moreover, suppose that some $\alpha>0$ exists such that \begin{align*} \displaystyle \delta\leq\frac{\alpha}{K}-G\left(\alpha\right)~~and~~Kg\left(\alpha\right)<1, \end{align*} where $G(t):=\displaystyle \int_{0}^{t}g(s)ds$. Then, there exists a solution $u\in V$ to the equation $\mathcal{F}(u)=0$ satisfying \begin{align} \left\|u-\hat{u}\right\|_{V}\leq\alpha.\label{al} \end{align} The solution is moreover unique under the side condition \eqref{al}. \end{theo} In the rest of this section, we consider the application of Theorem \ref{plum2001} to \eqref{absproblem}. Note that, in this case, the Fr\'echet derivative $\mathcal{F}_{\hat{u}}'$ of $\mathcal{F}$ at $\hat{u}\in V$ is given by \begin{align*} \left\langle\mathcal{F}_{\hat{u}}'u,v\right\rangle=\left(\nabla u,\nabla v\right)-p\left(\left|\hat{u}\right|^{p-1}u,v\right)~~~{\rm for}~u,v\in V. \end{align*} \subsubsection*{Residual bound $\delta$} For $\hat{u}\in V$ satisfying $\Delta\hat{u}\in L^{2}\left(\Omega\right)$, the residual bound $\delta$ is computed as \begin{align*} C_{2}\left\|\Delta\hat{u}+\left|\hat{u}\right|^{p-1}\hat{u}\right\|_{L^{2}\left(\Omega\right)}; \end{align*} the $L^{2}$-norm can be computed by a numerical integration method with verification (see Section \ref{integration} for details). \subsubsection*{Bound $K$ for the operator norm of $\mathcal{F}_{\hat{u}}^{\prime-1}$} In addition, we compute a bound $K$ for the operator norm of $\mathcal{F}_{\hat{u}}^{\prime-1}$ by the following theorem, proving simultaneously that this inverse operator exists and is defined on the whole of $V^{*}$. \begin{theo}[\cite{plum2009computer}]\label{invtheo} Let $\Phi:V\rightarrow V^{*}$ be the canonical isometric isomorphism, i.e., $\Phi$ is given by \begin{align*} \left\langle\Phi u,v\right\rangle:=\left(u,v\right)_{V} =\left(\nabla u,\nabla v\right ~~~{\rm for}~u,v\in V. \end{align*} If \begin{align} \displaystyle \mu_{0} :=\min\left\{|\mu|\ :\ \mu\in\sigma_{p}\left(\Phi^{-1}\mathcal{F}_{\hat{u}}'\right)\cup\{1\}\right\} >0, \label{mu0} \end{align} with $\sigma_{p}$ denoting the point spectrum, then the inverse of $\mathcal{F}_{\hat{u}}'$ exists and \begin{align} \left\|\mathcal{F}_{\hat{u}}^{\prime-1}\right\|_{B(V^{*},V)}\leq\mu_{0}^{-1}.\label{Ktheo} \end{align} \end{theo} \begin{proof} We prove this theorem by adapting a theory of Fredholm operators, i.e., we have recourse to the fact that the injectivity and the surjectivity of a Fredholm operator are equivalent. The operator $N:=\Phi-\mathcal{F}_{\hat{u}}'$ from $V$ to $V^{*}$ is given by $\left\langle Nu,v\right\rangle=p(\left|\hat{u}\right|^{p-1}u,v)$ for all $u,v\in V$. Thus, actually $N$ maps $V$ into $L^{2}(\Omega)$; note that $p \leq 2$ and $n \leq 3$. Hence $N:V\rightarrow V^{*}$ is compact, owing to the compactness of the embedding $L^{2}(\Omega)\hookrightarrow V^{*}$. Therefore, $\mathcal{F}_{\hat{u}}'$ is a Fredholm operator, and the spectrum $\sigma\left(\Phi^{-1}\mathcal{F}_{\hat{u}}'\right)$ of $\Phi^{-1}\mathcal{F}_{\hat{u}}'$ is given by \begin{align*} \sigma\left(\Phi^{-1}\mathcal{F}_{\hat{u}}'\right)=1-\sigma\left(\Phi^{-1}N\right)=1-\left\{\sigma_{p}\left(\Phi^{-1}N\right)\cup\{0\}\right\}=\sigma_{p}\left(\Phi^{-1}\mathcal{F}_{\hat{u}}'\right)\cup\{1\}. \end{align*} Since $\Phi^{-1}\mathcal{F}_{\hat{u}}'$ is self-adjoint, we have, for all $u\in V$, \begin{align*} \displaystyle \left\|\mathcal{F}_{\hat{u}}'u\right\|_{V^{*}}^{2}=\left\|\Phi^{-1}\mathcal{F}_{\hat{u}}'\right\|_{V}^{2}=\int_{-\infty}^{\infty}\mu^{2}d\left(E_{\mu}u,u\right)_{V}\geq\mu_{0}^{2}\int_{-\infty}^{\infty}d\left(E_{\mu}u,u\right)_{V}=\mu_{0}^{2}\left\|u\right\|_{V}^{2}, \end{align*} where $E_{\mu}$ is the resolution of the identity of $\Phi^{-1}\mathcal{F}_{\hat{u}}'$. Hence, $\mathcal{F}_{\hat{u}}'$ is one to one, and therefore is also onto. This implies \eqref{Ktheo}. \end{proof} \begin{rem} The property of the spectrum of $\Phi^{-1}\mathcal{F}_{\hat{u}}'$ is more precisely discussed in {\rm \cite[Section 3.3]{plum2009computer}}. \end{rem} The eigenvalue problem $\Phi^{-1}\mathcal{F}_{\hat{u}}'u=\mu u$ in $V$ is equivalent to \begin{align*} \left(\nabla u,\nabla v\right)-p\left(\left|\hat{u}\right|^{p-1}u,v\right)=\mu\left(\nabla u,\nabla v\right)~~{\rm for~all}~v\in V. \end{align*} Since $\mu=1$ is already known to be in $\sigma\left(\Phi^{-1}\mathcal{F}_{\hat{u}}'\right)$, it suffices to look for eigenvalues $\mu\neq 1$. By setting $\lambda=(1-\mu)^{-1}$, we further transform this eigenvalue problem into \begin{align} {\rm Find}~u\in V~{\rm and}~\lambda\in \mathbb{R}~{\rm s.t.}~\left(\nabla u,\nabla v\right)=\lambda\left(p\left|\hat{u}\right|^{p-1}u,v\right)~~{\rm for~all}~v\in V.\label{eiglam} \end{align} When we assume that $\hat{u}(x)\neq 0$ for almost all $ x\in\Omega$ (which will be true by numerical construction in our example), i.e., we have $\left|\hat{u}\right|^{p-1}>0$ a.e.~in $\Omega$, then \eqref{eiglam} is a regular eigenvalue problem, the spectrum of which consists of a sequence $\{\lambda_{k}\}_{k=1}^{\infty}$ of eigenvalues converging to $+\infty$. In order to compute $K$ on the basis of Theorem \ref{invtheo}, we concretely enclose the eigenvalue $\lambda$ of \eqref{eiglam} that minimizes the corresponding absolute value of $|\mu|\left(=|1-\lambda^{-1}|\right)$, by considering the following approximate eigenvalue problem \begin{align} {\rm Find}~u\in V_{N}~{\rm and}~\lambda^{N}\in \mathbb{R}~{\rm s.t.}~\left(\nabla u_{N},\nabla v_{N}\right)=\lambda^{N}\left(p\left|\hat{u}\right|^{p-1}u_{N},v_{N}\right)~~{\rm for~all}~v_{N}\in V_{N},\label{applam} \end{align} where $V_{N}$ is a finite-dimensional subspace of $V$. Note that \eqref{applam} amounts to a matrix eigenvalue problem, the eigenvalues of which can easily be enclosed by verified numerical linear algebra (see, e.g., \cite{behnke1991calculation,rump1999book,miyajima2012numerical}). To estimate the error between the $k$th eigenvalue $\lambda_{k}$ of \eqref{eiglam} and the $k$th eigenvalue $\lambda_{k}^{N}$ of \eqref{applam}, we consider the weak formulation of the Poisson equation \begin{align} \left(\nabla u,\nabla v\right)=\left(g,v\right)~~~{\rm for~all}~v\in V\label{poisson} \end{align} for given $g\in L^{2}\left(\Omega\right)$; it is well known that this equation has a unique solution $u\in V$ for each $g\in L^{2}\left(\Omega\right)$. Moreover, we introduce the orthogonal projection $P_{N}:V\rightarrow V_{N}$ defined by \begin{align*} \left(P_{N}u-u,v_{N}\right)_{V}=0~~~{\rm for~all}~u\in V{\rm~and~}v_{N}\in V_{N}. \end{align*} The following theorem enables us to estimate the error between $\lambda_{k}$ and $\lambda_{k}^{N}$. \begin{theo}[\cite{tanaka2014verified, liu2015framework}]\label{eigtheo} Suppose that $\hat{u} \in L^{\infty}(\Omega)$, and let $C_{N}$ denote a positive number such that \begin{align} \left\|u_{g}-P_{N}u_{g}\right\|_{V}\leq C_{N}\left\|g\right\|_{L^{2}(\Omega)}\label{CN} \end{align} for any $g\in L^{2}\left(\Omega\right)$ and the corresponding solution $u_{g}\in V$ to \eqref{poisson}. Then, \begin{align*} \displaystyle \frac{\lambda_{k}^{N}}{\lambda_{k}^{N}C_{N}^{2}\|p\left|\hat{u}\right|^{p-1}\|_{L^{\infty}(\Omega)}+1}\leq\lambda_{k}\leq\lambda_{k}^{N}. \end{align*} \end{theo} The right inequality is well known as Rayleigh-Ritz bound, which is derived from the min-max principle: \begin{align*} \displaystyle \lambda_{k}=\min_{H_{k}\subset V}\left(\max_{v\in H_{k}\backslash\{0\}}\frac{\left\|\nabla v\right\|_{L^{2}(\Omega)}^{2}}{\left\|av\right\|_{L^{2}(\Omega)}^{2}}\right)\leq\lambda_{k}^{N}, \end{align*} where we set $a=\sqrt{p\left|\hat{u}\right|^{p-1}}$ and the minimum is taken over all $k$-dimensional subspaces $H_{k}$ of $V$. Moreover, proofs of the left inequality can be found in \cite{tanaka2014verified, liu2015framework}. Assuming the $H^{2}$-regularity of solutions to \eqref{poisson} (e.g., when $\Omega$ is convex {\rm \cite[Section 3.3]{grisvard2011elliptic}}), \cite[Theorem 4]{tanaka2014verified} ensures the left inequality. A more general statement, that does not require the $H^{2}$-regularity, can be found in \cite[Theorem 2.1]{liu2015framework}. \begin{rem} When the $H^{2}$-regularity of solutions to \eqref{poisson} is confirmed a priori, e.g., when $\Omega$ is convex {\rm \cite[Section 3.3]{grisvard2011elliptic}}, \eqref{CN} can be replaced by \begin{align} \left\|u-P_{N}u\right\|_{V}\leq C_{N}\left\|-\Delta u\right\|_{L^{2}(\Omega)}~~~{\rm for~all}~u\in H^{2}(\Omega)\cap V. \end{align} The computation of a concrete value of $C_{N}$ for a given subspace $V_{N}$ will be discussed in Section $\ref{sec/example}$. \end{rem} \subsubsection*{Lipschitz bound $g$ for $\mathcal{F}_{\hat{u}}'$} Furthermore, a concrete construction of a function $g$ satisfying \eqref{lip} and \eqref{gconv} is important for our verification process. The following lemma is required for the construction. \begin{lem}\label{ablem} For $a,b\in \mathbb{R}$ and $q\in\left(0,1\right)$, \begin{align*} \left|\left|a+b\right|^{q}-\left|a\right|^{q}\right|\leq\left|b\right|^{q}. \end{align*} \end{lem} \begin{proof} For $\alpha,\beta\in[0,\infty)$ we have \begin{align*} (\displaystyle \alpha+\beta)^{q}-\alpha^{q}=q\int_{0}^{\beta}(\alpha+t)^{q-1}dt\leq q\int_{0}^{\beta}t^{q-1}dt=\beta^{q}, \end{align*} which readily gives \begin{align*} \left|a+b\right|^{q}-\left|a\right|^{q}\leq\left|b\right|^{q}~~~{\rm for}~a,b\in \mathbb{R}. \end{align*} Redefining terms, this inequality also implies \begin{align*} \left|a\right|^{q}-\left|a+b\right|^{q}=\left|(a+b)+(-b)\right|^{q}-\left|a+b\right|^{q}\leq\left|-b\right|^{q}=\left|b\right|^{q}~~~{\rm for}~a,b\in \mathbb{R} \end{align*} and hence the assertion. \end{proof} The following theorem gives us a concrete construction of the function $g$ in Theorem \ref{plum2001} for the nonlinearity $f\left(u\right)=\left|u\right|^{p-1}u~(1<p<2)$. \begin{theo}\label{selectiong} For $1<p<2$, we may select \begin{align} g\left(t\right)=pC_{r}C_{s}C_{q(p-1)}^{p-1}t^{p-1}\label{g} \end{align} to satisfy \eqref{lip} and \eqref{gconv} in Theorem $\ref{plum2001}$, where $q,r,s$ are positive numbers that satisfy $q^{-1}+r^{-1}+s^{-1}=1$ and $q\left(p-1\right)\geq 1$. \end{theo} \begin{proof} For fixed $\hat{u}\in V$, the left hand side of \eqref{lip} is written as \begin{align*} \displaystyle \left\|\mathcal{F}_{\hat{u}+u}'-\mathcal{F}_{\hat{u}}'\right\|_{\mathcal{B}(V,V^{*})}=p\sup_{v,\phi\in V\backslash\{0\}}\frac{\left|\left((\left|\hat{u}+u\right|^{p-1}-\left|\hat{u}\right|^{p-1})v,\phi\right)\right|}{\left\|v\right\|_{V}\left\|\phi\right\|_{V}}. \end{align*} Moreover, we have \begin{align*} \left|\left((\left|\hat{u}+u\right|^{p-1}-\left|\hat{u}\right|^{p-1})v,\phi\right)\right|&\leq\left\|\left|\hat{u}+u\right|^{p-1}-\left|\hat{u}\right|^{p-1}\right\|_{L^{q}\left(\Omega\right)}\left\|v\right\|_{L^{r}\left(\Omega\right)}\left\|\phi\right\|_{L^{s}\left(\Omega\right)}\\ &\leq C_{r}C_{s}\left\|\left|\hat{u}+u\right|^{p-1}-\left|\hat{u}\right|^{p-1}\right\|_{L^{q}\left(\Omega\right)}\left\|v\right\|_{V}\left\|\phi\right\|_{V}, \end{align*} and, owing to Lemma \ref{ablem}, \begin{align*} \left\|\left|\hat{u}+u\right|^{p-1}-\left|\hat{u}\right|^{p-1}\right\|_{L^{q}\left(\Omega\right)}&=\left(\int_{\Omega}\left|\left|\hat{u}(x)+u(x)\right|^{p-1}-\left|\hat{u}(x)\right|^{p-1}\right|^{q}dx\right)^{1/q}\\ &\leq\left(\int_{\Omega}\left|u(x)\right|^{q(p-1)}dx\right)^{1/q}=\left\|u\right\|_{L^{q(p-1)}\left(\Omega\right)}^{p-1}. \end{align*} Therefore, it follows that \begin{align*} \left\|\mathcal{F}_{\hat{u}+u}'-\mathcal{F}_{\hat{u}}'\right\|_{\mathcal{B}(V,V^{*})}\leq pC_{r}C_{s}C_{q(p-1)}^{p-1}\left\|u\right\|_{V}^{p-1}=g\left(\left\|u\right\|_{V}\right). \end{align*} \end{proof} \subsection{$L^{\infty}$ error estimation} In this subsection, we discuss a method that gives an $L^{\infty}$ error bound for a solution to \eqref{absproblem} from a known $H_{0}^{1}$ error bound, that is, we compute a concrete bound for $\left\|u-\hat{u}\right\|_{L^{\infty}\left(\Omega\right)}$ for a solution $u\in V$ to \eqref{absproblem} satisfying \begin{align} \left\|u-\hat{u}\right\|_{V}\leq\varepsilon\label{h10error} \end{align} with $\varepsilon>0$ and $\hat{u}\in V$. To obtain such an error estimation, we assume that $\Omega$ is convex and polygonal; this condition gives the $H^{2}$-regularity of solutions to \eqref{absproblem} (and therefore, ensures their boundedness) a priori. To be precise, when $\Omega$ is a convex polygonal domain, a weak solution $u\in V$ to \eqref{poisson} with $g\in L^{2}\left(\Omega\right)$ is $H^{2}$-regular (see, e.g., \cite[Section 3.3]{grisvard2011elliptic}). A solution $u$ satisfying \eqref{h10error} can be written in the form $ u=\hat{u}+\varepsilon\omega$ with some $\omega\in V,\ \left\|\omega\right\|_{V} \leq1$. Moreover, $\omega$ satisfies \begin{align*} \left\{\begin{array}{l l} -\Delta\varepsilon\omega=\left|\hat{u}+\varepsilon\omega\right|^{p-1}\left(\hat{u}+r\omega\right)+\Delta\hat{u} &\mathrm{i}\mathrm{n}\ \Omega,\\ \omega=0 &\mathrm{on}\ \partial\Omega, \end{array}\right. \end{align*} and therefore is also $H^{2}$-regular if $\Delta\hat{u}\in L^{2}(\Omega)$. We then use the following theorem to obtain an $L^{\infty}$ error estimation. \begin{theo}[\cite{plum1992explicit}]\label{linf} For all $u\in H^{2}\left(\Omega\right)$, \begin{align*} \|u\|_{L^{\infty}(\Omega)}\le c_{0}\|u\|_{L^{2}(\Omega)}+c_{1}\|\nabla u\|_{L^{2}(\Omega)}+c_{2}\|u_{xx}\|_{L^{2}(\Omega)} \end{align*} with \begin{align*} c_{j}=\displaystyle \frac{\gamma_{j}}{\left|\overline{\Omega}\right|}\left[\max_{x_{0}\in\overline{\Omega}}\int_{\overline{\Omega}}|x-x_{0}|^{2j}dx\right]^{1/2},~(j=0,1,2), \end{align*} where $u_{xx}$ denotes the Hesse matrix of $u,\ \left|\overline{\Omega}\right|$ is the measure of $\overline{\Omega}$, and \begin{align*} \gamma_{0}=1,~\gamma_{1}=1.1548,~\gamma_{2}=0.22361~~{\it if}~n=2. \end{align*} For $n=3$, other values of $\gamma_{0},~\gamma_{1},$ and $\gamma_{2}$ have to be chosen $($see {\rm \cite{plum1992explicit}}$)$. \end{theo} \begin{rem} The norm of the Hesse matrix of $u$ is precisely defined by \begin{align*} \|u_{xx}\|_{L^{2}(\Omega)}=\sqrt{\sum_{i,j=1}^{2}\left\|\frac{\partial^{2}u}{\partial x_{i}\partial x_{j}}\right\|_{L^{2}(\Omega)}^{2}}. \end{align*} Moreover, since $\Omega$ is polygonal, $\left\|u_{xx}\right\|_{L^{2}(\Omega)}=\left\|\Delta u\right\|_{L^{2}(\Omega)}$ for all $u\in H^{2}(\Omega)\cap V$~$($see, e.g., {\rm \cite{grisvard2011elliptic}}$)$. \end{rem} \begin{rem} Concrete values of each $c_{j}$ are provided for some special domains $\Omega$ in {\rm \cite{plum1992explicit,plum2001computer}}. According to these papers, one can choose, for $\Omega=(0,1)^{2}$, \begin{align*} c_{0}=\displaystyle \gamma_{0},\ c_{1}=\sqrt{\frac{2}{3}}\gamma_{1},{\rm~and~}c_{2}=\frac{\gamma_{3}}{3}\sqrt{\frac{28}{5}} . \end{align*} \end{rem} Applying Theorem \ref{linf}, we obtain the following corollary. \begin{coro}\label{Linfcoro} Let $u$ be a solution to \eqref{absproblem} satisfying \eqref{h10error} with $\hat{u}\in V$ such that $\Delta \hat{u} \in L^{2}(\Omega)$. Moreover, let $c_{0},\ c_{1}$, and $c_{2}$ be as in Theorem {\rm \ref{linf}}, and $p':=2(p-1)$. Then, \begin{align} &\left\|u-\hat{u}\right\|_{L^{\infty}(\Omega)}\nonumber\\ \leq&c_{0}C_{2}\varepsilon+c_{1}\varepsilon+c_{2}\left\{\max\{1,2^{\frac{p'-1}{2}}\}p\varepsilon C_{q}\sqrt{\left\|\hat{u}\right\|_{L^{rp'}\left(\Omega\right)}^{p'}+\frac{\varepsilon^{p'}}{p'+1}C_{rp'}^{p'}}+\left\|\Delta\hat{u}+\left|\hat{u}\right|^{p-1}\hat{u}\right\|_{L^{2}(\Omega)}\right\}\label{coroinequ} \end{align} holds for any $q$ and $r$ satisfying $q\geq 2,\ r\geq(p-1)^{-1}$, and $2q^{-1}+r^{-1}=1$. \end{coro} \begin{proof} Due to Theorem \ref{linf}, we have \begin{align*} \left\|u-\hat{u}\right\|_{L^{\infty}(\Omega)}&=\varepsilon\left\|\omega\right\|_{L^{\infty}(\Omega)}\\ &\leq\varepsilon\left(c_{0}\left\|\omega\right\|_{L^{2}(\Omega)}+c_{1}\left\|\omega\right\|_{V}+c_{2}\left\|\Delta\omega\right\|_{L^{2}(\Omega)}\right)\\ &\leq\varepsilon\left(c_{0}C_{2}+c_{1}+c_{2}\left\|\Delta\omega\right\|_{L^{2}(\Omega)}\right). \end{align*} The last term $\left\|\Delta\omega\right\|_{L^{2}(\Omega)}$ is estimated by \begin{align*} \varepsilon\left\|\Delta\omega\right\|_{L^{2}(\Omega)}&=\left\|\left|\hat{u}+\varepsilon\omega\right|^{p-1}\left(\hat{u}+\varepsilon\omega\right)+\Delta\hat{u}\right\|_{L^{2}(\Omega)}\\ &=\left\|\left|\hat{u}+\varepsilon\omega\right|^{p-1}\left(\hat{u}+\varepsilon\omega\right)-\left|\hat{u}\right|^{p-1}\hat{u}+\left|\hat{u}\right|^{p-1}\hat{u}+\Delta\hat{u}\right\|_{L^{2}(\Omega)}\\ &\leq\left\|\left|\hat{u}+\varepsilon\omega\right|^{p-1}\left(\hat{u}+\varepsilon\omega\right)-\left|\hat{u}\right|^{p-1}\hat{u}\right\|_{L^{2}(\Omega)}+\left\|\Delta\hat{u}+\left|\hat{u}\right|^{p-1}\hat{u}\right\|_{L^{2}(\Omega)} \end{align*} Since the mean value theorem ensures that \begin{align*} &\displaystyle \int_{\Omega}\left(\left|\hat{u}(x)+\varepsilon\omega(x)\right|^{p-1}\left(\hat{u}(x)+\varepsilon\omega(x)\right)-\left|\hat{u}(x)\right|^{p-1}\hat{u}(x)\right)^{2}dx\\ &=\displaystyle \int_{\Omega}\left(\varepsilon p\omega(x)\int_{0}^{1}\left|\hat{u}(x)+\varepsilon t\omega(x)\right|^{p-1}dt\right)^{2}dx\\ &\displaystyle \leq p^{2}\varepsilon^{2}\int_{\Omega}\omega(x)^{2}\int_{0}^{1}\left|\hat{u}(x)+\varepsilon t\omega(x)\right|^{p'}dtdx\\ &=p^{2}\displaystyle \varepsilon^{2}\int_{0}^{1}\int_{\Omega}\omega(x)^{2}\left|\hat{u}(x)+\varepsilon t\omega(x)\right|^{p'}dxdt\\ &\displaystyle \leq p^{2}\varepsilon^{2}\left\|\omega\right\|_{L^{q}\left(\Omega\right)}^{2}\int_{0}^{1}\left\|\left|\hat{u}+\varepsilon\omega t\right|^{p'}\right\|_{L^{r}\left(\Omega\right)}dt\\ &=p^{2}\displaystyle \varepsilon^{2}\left\|\omega\right\|_{L^{q}\left(\Omega\right)}^{2}\int_{0}^{1}\left\|\hat{u}+\varepsilon\omega t\right\|_{L^{rp'}\left(\Omega\right)}^{p'}dt\\ &\displaystyle \leq p^{2}\varepsilon^{2}\left\|\omega\right\|_{L^{q}\left(\Omega\right)}^{2}\int_{0}^{1}\left(\left\|\hat{u}\right\|_{L^{rp'}\left(\Omega\right)}+t\varepsilon\left\|\omega\right\|_{L^{rp'}\left(\Omega\right)}\right)^{p'}dt\\ &\displaystyle \leq\max\{1,2^{p'-1}\}p^{2}\varepsilon^{2}\left\|\omega\right\|_{L^{q}\left(\Omega\right)}^{2}\left\{\left\|\hat{u}\right\|_{L^{rp'}\left(\Omega\right)}^{p'}+\int_{0}^{1}\left(t\varepsilon\left\|\omega\right\|_{L^{rp'}\left(\Omega\right)}\right)^{p'}dt\right\}\\ &=\displaystyle \max\{1,2^{p'-1}\}p^{2}\varepsilon^{2}\left\|\omega\right\|_{L^{q}\left(\Omega\right)}^{2}\left(\left\|\hat{u}\right\|_{L^{rp'}\left(\Omega\right)}^{p'}+\frac{\varepsilon^{p'}}{p'+1}\left\|\omega\right\|_{L^{rp'}\left(\Omega\right)}^{p'}\right)\\ &\displaystyle \leq\max\{1,2^{p'-1}\}p^{2}\varepsilon^{2}C_{q}^{2}\left(\left\|\hat{u}\right\|_{L^{rp'}\left(\Omega\right)}^{p'}+\frac{\varepsilon^{p'}}{p'+1}C_{rp'}^{p'}\right), \end{align*} it follows that \begin{align*} \displaystyle \varepsilon\left\|\Delta\omega\right\|_{L^{2}(\Omega)}\leq\max\{1,2^{\frac{p'-1}{2}}\}p\varepsilon C_{q}\sqrt{\left\|\hat{u}\right\|_{L^{rp'}\left(\Omega\right)}^{p'}+\frac{\varepsilon^{p'}}{p'+1}C_{rp'}^{p'}}+\left\|\Delta\hat{u}+\left|\hat{u}\right|^{p-1}\hat{u}\right\|_{L^{2}(\Omega)}. \end{align*} Consequently, the $L^{\infty}$ error of $u$ is estimated as asserted in \eqref{coroinequ}. \end{proof}
c35deb6e25c2d4726086571871c372586524a4e9
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:introduction} Bioinformaticians define the $k$th-order de Bruijn graph for a string or set of strings to be the directed graph whose nodes are the distinct $k$-tuples in those strings and in which there is an edge from $u$ to $v$ if there is a \((k + 1)\)-tuple somewhere in those strings whose prefix of length $k$ is $u$ and whose suffix of length $k$ is $v$.\footnote{An alternative definition, which our data structure can be made to handle but which we do not consider in this paper, has an edge from $u$ to $v$ whenever both nodes are in the graph.} These graphs have many uses in bioinformatics, including {\it de novo\/} assembly~\cite{zerbino2008velvet}, read correction~\cite{DBLP:journals/bioinformatics/SalmelaR14} and pan-genomics~\cite{siren2014indexing}. The datasets in these applications are massive and the graphs can be even larger, however, so pointer-based implementations are impractical. Researchers have suggested several approaches to representing de Bruijn graphs compactly, the two most popular of which are based on Bloom filters~\cite{wabi,cascading} and the Burrows-Wheeler Transform~\cite{bowe2012succinct,boucher2015variable,belazzougui2016bidirectional}, respectively. In this paper we describe a new approach, based on minimal perfect hash functions~\cite{mehlhorn1982program}, that is similar to that using Bloom filters but has better theoretical bounds when the number of connected components in the graph is small, and is fully dynamic: i.e., we can both insert and delete nodes and edges efficiently, whereas implementations based on Bloom filters are usually semi-dynamic and support only insertions. We also show how to modify our implementation to support, e.g., jumbled pattern matching~\cite{BCFL12} with fixed-length patterns. Our data structure is based on a combination of Karp-Rabin hashing~\cite{KR87} and minimal perfect hashing, which we will describe in the full version of this paper and which we summarize for now with the following technical lemmas: \begin{lemma} \label{lem:static} Given a static set $N$ of $n$ $k$-tuples over an alphabet $\Sigma$ of size $\sigma$, with high probability in $O(kn)$ expected time we can build a function \(f : \Sigma^k \rightarrow \{0, \ldots, n - 1\}\) with the following properties: \begin{itemize} \item when its domain is restricted to $N$, $f$ is bijective; \item we can store $f$ in $O(n + \log k+\log\sigma)$ bits; \item given a $k$-tuple $v$, we can compute \(f (v)\) in $\Oh{k}$ time; \item given $u$ and $v$ such that the suffix of $u$ of length \(k - 1\) is the prefix of $v$ of length \(k - 1\), or vice versa, if we have already computed \(f (u)\) then we can compute \(f (v)\) in $\Oh{1}$ time. \end{itemize} \end{lemma} \begin{lemma} \label{lem:dynamic} If $N$ is dynamic then we can maintain a function $f$ as described in Lemma~\ref{lem:static} except that:\ \begin{itemize} \item the range of $f$ becomes \(\{0, \ldots, 3 n - 1\}\); \item when its domain is restricted to $N$, $f$ is injective; \item our space bound for $f$ is $\Oh{n (\log \log n + \log \log \sigma)}$ bits with high probability; \item insertions and deletions take $\Oh{k}$ amortized expected time. \item the data structure may work incorrectly with very low probability (inversely polynomial in $n$). \end{itemize} \end{lemma} Suppose $N$ is the node-set of a de Bruijn graph. In Section~\ref{sec:static} we show how we can store $\Oh{n \sigma}$ more bits than Lemma~\ref{lem:static} such that, given a pair of $k$-tuples $u$ and $v$ of which at least one is in $N$, we can check whether the edge \((u, v)\) is in the graph. This means that, if we start with a $k$-tuple in $N$, then we can explore the entire connected component containing that $k$-tuple in the underlying undirected graph. On the other hand, if we start with a $k$-tuple not in $N$, then we will learn that fact as soon as we try to cross an edge to a $k$-tuple that is in $N$. To deal with the possibility that we never try to cross such an edge, however --- i.e.,\@\xspace that our encoding as described so far is consistent with a graph containing a connected component disjoint from $N$ --- we cover the vertices with a forest of shallow rooted trees. We store each root as a $k$-tuple, and for each other node we store \(1 + \lg \sigma\) bits indicating which of its incident edges leads to its parent. To verify that a $k$-tuple we are considering is indeed in the graph, we ascend to the root of the tree that contains it and check that $k$-tuple is what we expect. The main challenge for making our representation dynamic with Lemma~\ref{lem:dynamic} is updating the covering forest. In Section~\ref{sec:dynamic} how we can do this efficiently while maintaining our depth and size invariants. Finally, in Section~\ref{sec:jumbled} we observe that our representation can be easily modified for other applications by replacing the Karp-Rabin hash function by other kinds of hash functions. To support jumbled pattern matching with fixed-length patterns, for example, we hash the histograms indicating the characters' frequencies in the $k$-tuples. \section{Static de Bruijn Graphs} \label{sec:static} Let \(G\) be a de Bruijn graph of order \(k\), let \(N = \{v_0, \ldots, v_{n-1}\}\) be the set of its nodes, and let \(E = \{a_0, \ldots, a_{e-1}\}\) be the set of its edges. We call each \(v_i\) either a node or a \(k\)-tuple, using interchangeably the two terms since there is a one-to-one correspondence between nodes and labels. We maintain the structure of \(G\) by storing two binary matrices, \IN and \OUT, of size \(n \times \sigma\). For each node, the former represents its incoming edges whereas the latter represents its outgoing edges. In particular, for each \(k\)-tuple \(v_x = c_1 c_2 \ldots c_{k-1} a\), the former stores a row of length \(\sigma\) such that, if there exists another \(k\)-tuple \(v_y = b c_1 c_2 \ldots c_{k-1}\) and an edge from \(v_y\) to \(v_x\), then the position indexed by \(b\) of such row is set to \TRUE. Similarly, \OUT contains a row for \(v_y\) and the position indexed by \(a\) is set to \TRUE. As previously stated, each \(k\)-tuple is uniquely mapped to a value between \(0\) and \(n-1\) by \(f\), where $f$ is as defined in Lemma~\ref{lem:static}, and therefore we can use these values as indices for the rows of the matrices \IN and \OUT, i.e.,\@\xspace in the previous example the values of \(\IN[f(v_x)][b]\) and \(\OUT[f(v_y)][a]\) are set to \TRUE. We note that, e.g., the SPAdes assembler~\cite{Ban12} also uses such matrices. Suppose we want to check whether there is an edge from \(b X\) to \(X a\). Letting \(f(b X) = i\) and \(f(X a) = j\), we first assume \(b X\) is in \(G\) and check the values of \(\OUT [i] [a] \) and \( \IN [j] [b]\). If both values are \TRUE, we report that the edge is present and we say that the edge is \emph{confirmed} by \IN and \OUT; otherwise, if any of the two values is \FALSE, we report that the edge is absent. Moreover, note that if \(b X\) is in \(G\) and \(\OUT [i] [a] = \TRUE\), then \(X a\) is in \(G\) as well. Symmetrically, if \(X a\) is in \(G\) and \(\IN [j] [b] = \TRUE\), then \(b X\) is in \(G\) as well. Therefore, if \(\OUT [i] [a] = \IN [j] [b] = \TRUE\), then \(b X\) is in \(G\) if and only if \(X a\) is. This means that, if we have a path \(P\) and if all the edges in \(P\) are confirmed by \IN and \OUT, then either all the nodes touched by \(P\) are in \(G\) or none of them is. We now focus on detecting false positives in our data structure maintaining a reasonable memory usage. Our strategy is to sample a subset of nodes for which we store the plain-text \(k\)-tuple and connect all the unsampled nodes to the sampled ones. More precisely, we partition nodes in the undirected graph \(G^\prime\) underlying \(G\) into a forest of rooted trees of height at least \(k \lg \sigma \) and at most \(3 k \lg \sigma\). For each node we store a pointer to its parent in the tree, which takes \(1 + \lg \sigma\) bits per node, and we sample the \(k\)-mer at the root of such tree. We allow a tree to have height smaller than \(k \lg \sigma\) when necessary, e.g., if it covers a connected component. Figure~\ref{fig:trees} shows an illustration of this idea. \begin{figure}[t!] \begin{center} \includegraphics[width=\textwidth]{trees.pdf} \caption{Given a de Bruijn graph (left), we cover the underlying undirected graph with a forest of rooted trees of height at most \(3 k \lg \sigma\) (center). The roots are shown as filled nodes, and parent pointers are shown as arrows; notice that the directions of the arrows in our forest are not related to the edges' directions in the original de Bruijn graph. We sample the $k$-tuples at the roots so that, starting at a node we think is in the graph, we can verify its presence by finding the root of its tree and checking its label in $\Oh{k \log \sigma}$ time. The most complicated kind of update (right) is adding an edge between a node $u$ in a small connected component to a node $v$ in a large one, $v$'s depth is more than \(2 k \lg \sigma\) in its tree. We re-orient the parent pointers in $u$'s tree to make $u$ the temporary root, then make $u$ point to $v$. We ascend \(k \lg \sigma\) steps from $v$, then delete the parent pointer $e$ of the node $w$ we reach, making $w$ a new root. (To keep this figure reasonably small, some distances in this example are smaller than prescribed by our formulas.)} \label{fig:trees} \end{center} \end{figure} We can therefore check whether a given node \(v_x\) is in \(G\) by first computing \(f(v_x)\) and then checking and ascending at most \(3 k \lg \sigma\) edges, updating \(v_x\) and \(f(v_x)\) as we go. Once we reach the root of the tree we can compare the resulting \(k\)-tuple with the one sampled to check if \(v_x\) is in the graph. This procedure requires \Oh{k \lg \sigma} time since computing the first value of \(f(v_x)\) requires \Oh{k}, ascending the tree requires constant time per edge, and comparing the \(k\)-tuples requires \Oh{k}. We now describe a Las Vegas algorithm for the construction of this data structure that requires, with high probability, \Oh{kn + n\sigma} expected time. We recall that \(N\) is the set of input nodes of size \(n\). We first select a function \(f\) and construct bitvector \(B\) of size \(n\) initialized with all its elements set to \FALSE. For each elements \(v_x\) of \(N\) we compute \(f(v_x) = i\) and check the value of \(B[i]\). If this value is \FALSE we set it to \TRUE and proceed with the next element in \(N\), if it is already set to \TRUE, we reset \(B\), select a different function \(f\), and restart the procedure from the first element in \(N\). Once we finish this procedure --- i.e.,\@\xspace we found that \(f\) do not produces collisions when applied to \(N\) --- we store \(f\) and proceed to initialize \IN and \OUT correctly. This procedure requires with high probability \Oh{kn} expected time for constructing \(f\) and \Oh{n\sigma} time for computing \IN and \OUT. Notice that if \(N\) is the set of \(k\)-tuples of a single text sorted by their starting position in the text, each \(f(v_x)\) can be computed in constant time from \(f(v_{x-1})\) except for \(f(v_0)\) that still requires \Oh{k}. More generally, if \(N\) is the set of \(k\)-tuples of \(t\) texts sorted by their initial position, we can compute \(n - t\) values of the function \(f(v_x)\) in constant time from \(f(v_{x-1})\) and the remaining in \Oh{k}. We will explain how to build the forest in the full version of this paper. In this case the construction requires, with high probability, \(\Oh{kt + n + n\sigma} = \Oh{kt + n\sigma}\) expected time. Combining our forest with Lemma~\ref{lem:static}, we can summarize our static data structure in the following theorem: \begin{theorem} \label{thm:static} Given a static $\sigma$-ary $k$th-order de Bruijn graph $G$ with $n$ nodes, with high probability in $\Oh{k n + n \sigma}$ expected time we can store $G$ in $\Oh{\sigma n}$ bits plus $\Oh{k \log \sigma}$ bits for each connected component in the underlying undirected graph, such that checking whether a node is in $G$ takes $\Oh{k \log \sigma}$ time, listing the edges incident to a node we are visiting takes $\Oh{\sigma}$ time, and crossing an edge takes $\Oh{1}$ time. \end{theorem} In the full version we will show how to use monotone minimal perfect hashing~\cite{BBPV09} to reduce the space to $(2+\epsilon)n\sigma$ bits of space (for any constant $\epsilon>0$). We will also show how to reduce the time to list the edges incident to a node of degree $d$ to $O(d)$, and the time to check whether a node is in $G$ to $\Oh{k}$. We note that the obtained space and query times are both optimal up to constant factors, which is unlike previous methods which have additional factor(s) depending on $k$ and/or $\sigma$ in space and/or time. \section{Dynamic de Bruijn Graphs} \label{sec:dynamic} In the previous section we presented a static representation of de Buijn graphs, we now present how we can make this data structure dynamic. In particular, we will show how we can insert and remove edges and nodes and that updating the graph reduces to managing the covering forest over \(G\). In this section, when we refer to $f$ we mean the function defined in Lemma~\ref{lem:dynamic}. We first show how to add or remove an edge in the graph and will later describe how to add or remove a node in it. The updates must maintain the following invariant: any tree must have size at least $k\log\sigma$ and height at most $3k\log\sigma$ except when the tree covers (all nodes in) a connected component of size at most $k\log\sigma$. Let \(v_x\) and \(v_y\) be two nodes in \(G\), \(e = (v_x, v_y)\) be an edge in \(G\), and let \(f(v_x) = i\) and \(f(v_y) = j\). Suppose we want to add \(e\) to \(G\). First, we set to \TRUE the values of \(\OUT[i][a]\) and \(\IN[j][b]\) in constant time. We then check whether \(v_x\) or \(v_y\) are in different components of size less than \(k \lg \sigma\) in \Oh{k \lg \sigma} time for each node. If both components have size greater than \(k \lg \sigma\) we do not have to proceed further since the trees will not change. If both connected components have size less than \(k \lg \sigma\) we merge their trees in \Oh{k \lg \sigma} time by traversing both trees and switching the orientation of the edges in them, discarding the samples at the roots of the old trees and sampling the new root in \Oh{k} time. If only one of the two connected components has size greater than \(k \lg \sigma\) we select it and perform a tree traversal to check whether the depth of the node is less than \(2 k \lg \sigma\). If it is, we connect the two trees as in the previous case. If it is not, we traverse the tree in the bigger components upwards for \(k \lg \sigma\) steps, we delete the edge pointing to the parent of the node we reached creating a new tree, and merge it with the smaller one. This procedure requires \Oh{k \lg \sigma} time since deleting the edge pointing to the parent in the tree requires \Oh{1} time, i.e.,\@\xspace we have to reset the pointer to the parent in only one node. Suppose now that we want to remove \(e\) from \(G\). First we set to \FALSE the values of \(\OUT[i][a]\) and \(\IN[j][b]\) in constant time. Then, we check in \Oh{k} time whether \(e\) is an edge in some tree by computing \(f(v_x)\) and \(f(v_y)\) checking for each node if that edge is the one that points to their parent. If \(e\) is not in any tree we do not have to proceed further whereas if it is we check the size of each tree in which \(v_x\) and \(v_y\) are. If any of the two trees is small (i.e.,\@\xspace if it has fewer than \(k \lg \sigma\) elements) we search any outgoing edge from the tree that connects it to some other tree. If such an edge is not found we conclude that we are in a small connected component that is covered by the current tree and we sample a node in the tree as a root and switch directions of some edges if necessary. If such an edge is found, we merge the small tree with the bigger one by adding the edge and switch the direction of some edges originating from the small tree if necessary. Finally if the height of the new tree exceeds $3k\log\sigma$, we traverse the tree upwards from the deepest node in the tree (which was necessarily a node in the smaller tree before the merger) for \(2k \lg \sigma\) steps, delete the edge pointing to the parent of the reached node, creating a new tree. This procedure requires $\Oh{k \lg \sigma}$ since the number of nodes traversed is at most \(O(k \lg \sigma)\) and the number of changes to the data structures is also at most \(O(k \lg \sigma)\) with each change taking expected constant time. It is clear that the insertion and deletion algorithms will maintain the invariant on the tree sizes. It is also clear that the invariant implies that the number of sampled nodes is $O(n/(k\log\sigma))$ plus the number of connected components. We now show how to add and remove a node from the graph. Adding a node is trivial since it will not have any edge connecting it to any other node. Therefore adding a node reduces to modify the function \(f\) and requires \Oh{k} amortized expected time. When we want to remove a node, we first remove all its edges one by one and, once the node is isolated from the graph, we remove it by updating the function \(f\). Since a node will have at most \(\sigma\) edges and updating \(f\) requires \Oh{k} amortized expected time, the amortized expected time complexity of this procedure is $\Oh{\sigma k\lg \sigma+ k}$. Combining these techniques for updating our forest with Lemma~\ref{lem:dynamic}, we can summarize our dynamic data structure in the following theorem: \begin{theorem} \label{thm:dynamic} We can maintain a $\sigma$-ary $k$th-order de Bruijn graph $G$ with $n$ nodes that is fully dynamic (i.e., supporting node and edge insertions and deletions) in $\Oh{n (\log \log n + \sigma)}$ bits (plus $\Oh{k \log \sigma}$ bits for each connected component) with high probability, such that we can add or remove an edge in expected \Oh{k\lg\sigma} time, add a node in expected \Oh{k+\sigma} time, and remove a node in expected \Oh{\sigma k\lg \sigma} time, and queries have the same time bounds as in Theorem~\ref{thm:static}. The data structure may work incorrectly with very low probability (inversely polynomial in $n$). \end{theorem} \section{Jumbled Pattern Matching} \label{sec:jumbled} Karp-Rabin hash functions implicitly divide their domain into equivalence classes --- i.e., subsets in which the elements hash to the same value. In this paper we have chosen Karp-Rabin hash functions such that each equivalence class contains only one $k$-tuple in the graph. Most of our efforts have gone into being able, given a $k$-tuple and a hash value, to determine whether that $k$-tuple is the unique element of its equivalence class in the graph. In some sense, therefore, we have treated the equivalence relation induced by our hash functions as a necessary evil, useful for space-efficiency but otherwise an obstacle to be overcome. For some applications, however --- e.g., parameterized pattern matching, circular pattern matching or jumbled pattern matching --- we are given an interesting equivalence relation on strings and asked to preprocess a text such that later, given a pattern, we can determine whether any substrings of the text are in the same equivalence class as the pattern. We can modify our data structure for some of these applications by replacing the Karp-Rabin hash function by other kinds of hash functions. For indexed jumbled pattern matching~\cite{BCFL12,KRR13,ACLL14} we are asked to pre-process a text such that later, given a pattern, we can determine quickly whether any substring of the text consists of exactly the same multiset of characters in the pattern. Consider fixed-length jumbled pattern matching, when the length of the patterns is fixed at pre-processing time. If we modify Lemmas~\ref{lem:static} and~\ref{lem:dynamic} so that, instead of using Karp-Rabin hashes in the definition of the function $f$, we use a hash function on the histograms of characters' frequencies in $k$-tuples, our function $f$ will map all permutations of a $k$-tuple to the same value. The rest of our implementation stays the same, but now the nodes of our graph are multisets of characters of size $k$ and there is an edge between two nodes $u$ and $v$ if it is possible to replace an element of $u$ and obtain $v$. If we build our graph for the multisets of characters in $k$-tuples in a string $S$, then our process for checking whether a node is in the graph tells us whether there is a jumbled match in $S$ for a pattern of length $k$. If we build a tree in which the root is a graph for all of $S$, the left and right children of the root are graphs for the first and second halves of $S$, etc., as described by Gagie et al.~\cite{GHLW15}, then we increase the space by a logarithmic factor but we can return the locations of all matches quickly. \begin{theorem} \label{thm:jumbled} Given a string \(S [1..n]\) over an alphabet of size $\sigma$ and a length $k \ll n$, with high probability in $\Oh{k n + n \sigma}$ expected time we can store \((2n \log \sigma)(1+o(1))\) bits such that later we can determine in $\Oh{k \log \sigma}$ time if a pattern of length $k$ has a jumbled match in $S$. \end{theorem} \section*{Acknowledgements} Many thanks to Rayan Chikhi and the anonymous reviewers for their comments. \bibliographystyle{splncs03}
f11ef27323222b08c16b381c698b58d670f5722f
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\subsection*{Acknowledgements} A large number of people have supplied advice and encouragement on this project over a period of many years. The undergraduates from my Introductory VIGRE Research Group held in Fall 2010 at the University of Georgia and my student Darcy Chanin performed many calculations for genus 4, 5, and 6 surfaces. I am grateful to the computational algebra group at the University of Sydney, especially John Cannon and Mark Watkins, for hosting me for a visit in June 2011 where I began programming the main algorithm in \texttt{Magma} \cite{Magma}. I have had many helpful conversations with my classmates and colleagues at Columbia University, the University of Georgia, and Fordham University. Valery Alexeev and James McKernan suggested the algorithm for matrix generators of representations outlined in Section \ref{matrix generators section}. Finally, I am grateful to Jennifer Paulhus, Tony Shaska, and John Voight, whose encouragement was essential in completing this project. This work was partially supported by the University of Georgia's NSF VIGRE grant DMS-03040000, a Simons Foundation Travel Grant, and a Fordham Faculty Research Grant. \subsection*{Online material} My webpage for this project is \cite{mywebpage}. This page contains links to the latest version of my \texttt{Magma} code, files detailing the calculations for specific examples, and many equations that are omitted in the tables in Section \ref{results section}. In future work, Jennifer Paulhus and I plan to include much of the data described in this paper and on the website \cite{mywebpage} in the L-Functions and Modular Forms Database at \texttt{lmfdb.org}. \section{The main algorithm} \label{algorithm section} We begin by stating the main algorithm. Then, in the following subsections, we discuss each step in more detail, including precise definitions and references for terms and facts that are not commonly known. \begin{algorithm} \label{main algorithm} \mbox{} \\ \textsc{Inputs:} \begin{enumerate} \item A finite group $G$; \item an integer $g \geq 2$; \item a set of surface kernel generators $(a_1,\ldots,a_{g_0}; b_1,\ldots,b_{g_0}; g_1,\ldots,g_r)$ determining a family of nonhyperelliptic Riemann surfaces $X$ of genus $g$ with $G \subset \operatorname{Aut}(X)$\\ \end{enumerate} \textsc{Output:} A locally closed set $B \subset \mathbb{A}^{n}$ and a family of smooth curves $\mathcal{X} \subset \mathbb{P}^{g-1} \times B$ such that for each closed point $b \in B$, the fiber $\mathcal{X}_b$ is a smooth genus $g$ canonically embedded curve with $G \subset \operatorname{Aut}(\mathcal{X}_b)$.\\ \begin{enumerate} \item[Step 1.] Compute the conjugacy classes and character table of $G$. \item[Step 2.] Use the Eichler trace formula to compute the character of the action on differentials and on cubics in the canonical ideal. \item[Step 3.] Obtain matrix generators for the action on holomorphic differentials. \item[Step 4.] Use the projection formula to obtain candidate cubics. \item[Step 5.] Compute a flattening stratification and select the locus yielding smooth algebraic curves with degree $2g-2$ and genus $g$. \end{enumerate} \end{algorithm} \subsection{Step 1: conjugacy classes and character table of $G$} This step is purely for bookkeeping. It is customary to list the conjugacy classes of $G$ in increasing order, and to list the rows in a character table by increasing degree. However, there is no canonical order to either the conjugacy classes or the irreducible characters. Given two different descriptions of a finite group $G$, modern software such as \texttt{Magma} may order the classes or the irreducible characters of $G$ differently. Hence, we compute and fix these at the beginning of the calculation. \subsection{Step 2: Counting fixed points and the Eichler trace formula} Here we define surface kernel generators for the automorphism group of a Riemann surface. These generators determine the Riemann surface as a branched cover of $\mathbb{P}^1$ and are used in a key formula (see Theorem \ref{count fixed points} below) for counting the number of fixed points of an automorphism. \begin{definition}[cf.~\cite{Breuer} Theorem 3.2, Theorem 3.14] \label{surface kernel definition} A \emph{signature} is a list of integers $(g_0; e_1,\ldots,e_r)$ with $g_0\geq 0$, $r \geq 0$, and $e_i \geq 2$. \\ A set of \emph{surface kernel generators} for a finite group $G$ and signature $(g_0; e_1,\ldots,e_r)$ is a sequence of elements $a_1,\ldots, a_{g_0}, b_{1},\ldots, b_{g_0}, g_1,\ldots, g_r \in G$ such that \begin{enumerate} \item $\langle a_1,\ldots, a_{g_0}, b_{1},\ldots, b_{g_0}, g_1,\ldots, g_r \rangle = G$; \item $\operatorname{Order}(g_i)=e_i$; and \item $\prod_{j=1}^{g_0} [a_j,b_j] \prod_{i=1}^{r} g_i = \operatorname{Id}_G$. \end{enumerate} \end{definition} Surface kernel generators have many other names in other papers; they are called \emph{ramification types} in \cite{MSSV} and \emph{generating vectors} in \cite{Paulhus}. As explained in \cite{Breuer}*{Section 3.11}, surface kernel generators describe the quotient morphism $X \rightarrow X/G$ as a branched cover. Here $X$ is a Riemann surface of genus $g$, $G$ is a subgroup of $\operatorname{Aut}(X)$, the quotient $X/G$ has genus $g_0$, the quotient morphism branches over $r$ points, and the integers $e_i$ describe the ramification over the branch points. In the sequel we will be primarily interested in large automorphism groups, that is, $|\operatorname{Aut}(X)| > 4(g_X-1)$. In this case, the Riemann-Hurwitz formula implies that $g_0 = 0$ and $3 \leq r \leq 4$. Surface kernel generators are used in the following formula for the number of fixed points of an automorphism: \begin{theorem}[\cite{Breuer}*{Lemma 11.5}] \label{count fixed points} Let $\sigma$ be an automorphism of order $h>1$ of a Riemann surface $X$ of genus $g\geq 2$. Let $(g_1,\ldots,g_r)$ be part of a set of surface kernel generators for $X$, and let $(m_1,\ldots, m_r)$ be the orders of these elements. Let $\mbox{Fix}_{X,u}(\sigma)$ be the set of fixed points of $X$ where $\sigma$ acts on a neighborhood of the fixed point by $z \mapsto \exp(2 \pi i u/h) z$. Then \begin{displaymath} |\operatorname{Fix}_{X,u}(\sigma)| = |C_{G}(\sigma)| \sum_{\substack{g_i \, s.t. \\ h|m_i \\ \sigma \sim g_i^{m_i u/h} }} \frac{1}{m_i} \end{displaymath} Here $C_{G}(\sigma)$ is the centralizer of $\sigma$ in $G$, and $\sim$ denotes conjugacy. \end{theorem} Next we recall the Eichler Trace Formula. For a Riemann surface $X$, let $\Omega_X$ be the holomorphic cotangent bundle, and let $\omega_X = \bigwedge \Omega_X$ be the sheaf of holomorphic differentials. The Eichler Trace Formula gives the character of the action of $\operatorname{Aut}(X)$ on $\Gamma(\omega_X^{\otimes d})$. \begin{theorem}[Eichler Trace Formula \cite{FarkasKra}*{Theorem V.2.9}] \label{Eichler Trace Formula} Suppose $g_{X} \geq 2$, and let $\sigma$ be a nontrivial automorphism of $X$ of order $h$. Write $\chi_{d}$ for the character of the representation of $\operatorname{Aut}(X)$ on $\Gamma(\omega_X^{\otimes d})$. Then \begin{displaymath} \chi_d(\sigma) = \left\{ \begin{array}{ll} 1 + \displaystyle \sum_{\substack{1 \leq u < h \\ (u,h)=1}} |\operatorname{Fix}_{X,u}(\sigma) | \frac{\zeta_{h}^{u}}{1-\zeta_{h}^{u}} & \mbox{ if $d=1$} \\ \displaystyle \sum_{\substack{1 \leq u < h \\ (u,h)=1}} |\operatorname{Fix}_{X,u}(\sigma) | \frac{\zeta_{h}^{u (d\%h)}}{1-\zeta_{h}^{u}} & \mbox{ if $d \geq 2$} \end{array} \right. \end{displaymath} \end{theorem} Together, the previous two results give a group-theoretic method for computing the character of the $\operatorname{Aut}(X)$ action on $\Gamma(\omega_X^{\otimes d})$ starting from a set of surface kernel generators. We can use the character of $\operatorname{Aut}(X)$ on $\Gamma(\omega_X^{\otimes d})$ to obtain the character of $\operatorname{Aut}(X)$ on quadric and cubics in the canonical ideal as follows. Let $S$ be the coordinate ring of $\mathbb{P}^{g-1}$, let $I \subset S$ be the canonical ideal, and let $S_d$ and $I_d$ denote the degree $d$ subspaces of $S$ and $I$. By Noether's Theorem, the sequence \[ 0 \rightarrow I_d \rightarrow S_d \rightarrow \Gamma(\omega_X^{\otimes d}) \rightarrow 0 \] is exact for each $d \geq 2$, and by Petri's Theorem, the canonical ideal is generated either by quadrics or by quadrics and cubics. Thus, beginning with the character of the action on $\Gamma(\omega_X) \cong S_1$, we may compute the characters of the actions on $S_2 = \operatorname{Sym}^2 S_1$ and $S_3 = \operatorname{Sym}^3 S_1$ and $\Gamma(\omega_X^{\otimes 2})$ and $\Gamma(\omega_X^{\otimes 3})$, and then obtain the characters of the actions on $I_2$ and $I_3$. \subsection{Step 3: matrix generators for a specified irreducible character} \label{matrix generators section} From Step 2 we have the character of the action on $\Gamma(\omega_X)$. We seek matrix generators for this action. It suffices to find matrix generators for each irreducible $G$-module appearing in $\Gamma(\omega_X)$. Given a finite group $G$ and an irreducible character $\chi$ of $G$, software such as \texttt{GAP} \cite{GAP} and \texttt{Magma} contain commands for producing matrix generators of a representation $V$ of $G$ with character $\chi$. Finding efficient algorithms to produce matrix generators with good properties (for instance, sparse matrices, or matrices whose entries have small height, or matrices whose entries belong to a low degree extension of $\mathbb{Q}$) is a subject of ongoing research \cites{Dabbaghian,DabbaghianDixon}. It seems that computer algebra systems implement several different algorithms that cover many special cases. I do not know a reference for a general algorithm. Hence, I briefly present an algorithm that was suggested to me by Valery Alexeev and James McKernan. This algorithm is not expected to perform efficiently; it is included merely to establish that Step 3 in Algorithm \ref{main algorithm} can be performed algorithmically. \begin{algorithm} \mbox{} \\ \textsc{Inputs:} \begin{enumerate} \item a finite group $G$ with generators $g_1,\ldots,g_r$; \item an irreducible character $\chi: G \rightarrow \mathbb{C}$ of degree $n$. \end{enumerate} \textsc{Output:} matrices $M_{1},\ldots,M_{r} \in \operatorname{GL}(n,\mathbb{C})$ such that the homomorphism $g_i \mapsto M_i$ is a representation with character $\chi$ \begin{enumerate} \item[Step 1.] Compute matrix generators for the regular representation $V$ of $G$. These matrices are permutation matrices, and hence their entries are in $\{0,1\}$. \item[Step 2.] Use the projection formula (see Theorem \ref{projection formula theorem} below) to compute matrix generators $\rho_W(g)$ for a representation $W$ with character $n\chi$. Let $K $ be the smallest field containing $\{ \chi(g) : g \in G\}$. Note that $\mathbb{Q} \subseteq K \subseteq \mathbb{Q}[\zeta_{|G|}]$. Then the matrix generators $\rho_W(g)$ lie in $\operatorname{GL}(n^2,K)$. \item[Step 3.] Let $x_1,\ldots,x_{n^2}$ be indeterminates. Let $M$ be the $|G| \times n^2$ matrix over $K$ whose rows are given by the vectors $\rho_W(g).(x_1,\ldots,x_{n^2})$. Let $X \subset \mathbb{P}^{n^2-1}_{K}$ be the determinantal variety $\operatorname{rank} M \leq n$. Since representations of finite groups are completely reducible in characteristic zero, the representation $W$ is isomorphic over $K$ to the direct sum $V_{\chi}^{\oplus n}$, and therefore $X(K)$ is non-empty. \item[Step 4.] Intersect $X$ with generic hyperplanes with coefficients in $K$ to obtain a zero-dimensional variety $Y$. \item[Step 5.] If necessary, pass to a finite field extension $L$ of $K$ to obtain a reduced closed point $y \in Y(L)$. \item[Step 6.] The point $y$ (regarded as a vector in $W \otimes L$) generates the desired representation. \end{enumerate} \end{algorithm} An example where this algorithm is used to produce matrix generators for the degree two irreducible representation of the symmetric group $S_3$ is available at my webpage \cite{mywebpage}. Finally, we note that in \cite{Streit}, Streit describes a method for producing matrix generators for the action of $\operatorname{Aut}(X)$ on $\Gamma(\omega_X)$ for some Bely\u{i} curves. \subsection{Step 4: the projection formula} Recall the projection formula for representations of finite groups. (See for instance \cite{FultonHarris} formula (2.31)). \begin{theorem}[Projection formula] \label{projection formula theorem} Let $V$ be a finite-dimensional representation of a finite group $G$ over $\mathbb{C}$. Let $V_1,\ldots,V_k$ be the irreducible representations of $G$, let $\chi_i$ be their characters, and let $V \cong \bigoplus_{i=1}^{k} V_{i}^{\oplus m_i}$. Let $\pi: V \rightarrow V_{i}^{\oplus m_i}$ be the projection onto the $i^{th}$ isotypical component of $V$. Then \[ \pi_i = \frac{\dim(V_i)}{|G|} \sum_{g \in G} \overline{\chi_{i}(g)} g. \] \end{theorem} From Step 3, we have matrix generators for the $G$ action on $\Gamma(\omega_X) = S_1$. Thus, we can compute matrix generators for the actions on $S_2$ and $S_3$, and use the projection formula to compute the isotypical subspace $S_{d,p}$ of degree $d$ polynomials on which $G$ acts with character $\chi_p$. In some a few examples, we have $I_{d,p} = S_{d,p}$, but more commonly, we have strict containment $I_{d,p} \subset S_{d,p}$. In this case we write elements of $I_{d,p} $ as generic linear combinations of the basis elements of $S_{d,p}$ and then seek coefficients that yield a smooth algebraic curve with the correct degree and genus. The coefficients used to form these generic linear combinations form the base space $\mathbb{A}^n$ of the family $\mathcal{X}$ produced by the main algorithm. \subsection{Step 5: Flattening stratifications} \label{flattening stratification section} \begin{theorem}[\cite{MumfordCurves}*{Lecture 8}] Let $f: X \rightarrow S$ be a projective morphism with $S$ a reduced Noetherian scheme. Then there exist locally closed subsets $S_1,\ldots, S_n$ such that $S = \sqcup_{i=1}^{n} S_i$ and $f|_{f^{-1}(S_i)} $ is flat. \end{theorem} The stratification $S = \sqcup_{i=1}^{n} S_i$ is called a flattening stratification for the map $f$. Since $S$ is reduced, flatness implies that over each stratum, the Hilbert polynomial of the fibers is constant. We find the stratum with Hilbert polynomial $P(t) = (2g-2)t -g+1$, then intersect this stratum with the locus where the fibers are smooth. This completes the algorithm. Flattening stratifications have been an important tool in theoretical algebraic geometry for over 50 years. There exist Gr\"obner basis techniques for computing flattening stratifications; in the computational literature, these are typically called \emph{comprehensive} or \emph{parametric Gr\"obner bases}, or \emph{Gr\"obner systems}. The foundational work on this problem was begun by Weispfenning, and many authors, including Manubens and Montes, Suzuki and Sato, Nabeshima, and Kapur, Sun, and Wang, have made important improvements on the original algorithm \cites{Weispfenning,Nabeshima,KapurSunWang}. The size of a Gr\"obner basis can grow very quickly with the number of variables and generators of an ideal, and unfortunately, even the most recent software cannot compute flattening stratifications for the examples we consider. Thus, in section \ref{partial flattening stratification section} below, we discuss a strategy for circumventing this obstacle. \section{Heuristic improvements} \label{heuristics section} Many steps of Algorithm \ref{main algorithm} can be run using a computer algebra system, but even for modest examples, the flattening stratification required in the final step is intractable. Therefore we discuss various heuristics that can be employed to speed the computation. \subsection{Tests for gonality and reduction to quadrics} Given a set of surface kernel generators, it is useful to discover as early as possible whether the corresponding Riemann surface is hyperelliptic, trigonal, a plane quintic, or none of these. We discuss these properties in turn. \textit{Hyperelliptic Riemann surfaces}. Algorithm \ref{main algorithm} supposes that one begins with surface kernel generators corresponding to a nonhyperelliptic curve. However, we can easily test for hyperellipticity if this property is not known in advance. A Riemann surface $X$ is hyperelliptic if and only if $\operatorname{Aut}(X)$ contains a central involution with $2g_X+2$ fixed points. Thus, given a set of surface kernel generators, we can search for a central involution and count its fixed points using Theorem \ref{count fixed points} (or even better, using \cite{Breuer}*{Lemma 10.4}). In \cite{Shaska}, Shaska gives equations of the form $y^2=f(x)$ for hyperelliptic curves with automorphisms. Additionally, we can use the algorithm described in \cite{StevensDeformations} to get the equations of $C$ under a linear series such as the transcanonical embedding or bicanonical embedding. So suppose the Riemann surface is not hyperelliptic. By Petri's Theorem, the canonical ideal is generated by quadrics if $X$ is not hyperelliptic, not trigonal, and not a plane quintic. Thus, ruling out these possibilities allows us to work with quadrics instead of cubics, which significantly speeds up the algorithm. This leads us to consider trigonal Riemann surfaces and plane quintics. \textit{Trigonal Riemann surfaces.} Trigonal Riemann surfaces may be divided into two types: cyclic trigonal and general trigonal \cite{CostaIzquierdo}. Cyclic trigonal curves can be detected by searching for degree three elements fixing $g+2$ points. Their automorphism groups have been classified \cite{BCGCyclicTrigonal}*{Theorem 2.1}, and one may hope for a paper treating equations of cyclic trigonal Riemann surfaces as the paper \cite{Shaska} treats equations of hyperelliptic Riemann surfaces. Less is known about general trigonal curves. We have Arakawa's bounds \cite{Arakawa}*{Remark 5} and a few additional necessary conditions \cite{CostaIzquierdo}*{Prop. 4 and Lemma 5}. We will not say more about general trigonal Riemann surfaces here because after studying the Riemann surfaces with large automorphism groups with genus $ 4 \leq g \leq 7$, we learn \textit{a posteriori} that very few of them are general trigonal. \textit{Plane quintics.} Plane quintics only occur in genus 6, and the canonical model of a plane quintic lies on the Veronese surface in $\mathbb{P}^5$. Thus, we have a necessary condition: $X$ is a plane quintic only if $\Gamma(\omega_X) \cong \operatorname{Sym}^2 V$ for some (possibly reducible) three-dimensional representation $V$ of $G$. In practice, it is generally quite fast to discover whether a nonhyperelliptic non-cyclic trigonal genus 6 Riemann surface is a plane quintic. \subsection{Partial flattening stratifications} \label{partial flattening stratification section} In this section we use several notions from the theory of Gr\"obner bases. We will not recall all the definitions here, and instead refer to \cite{Eisenbud}*{Chapter 15} for the details. The algorithms for comprehensive Gr\"obner bases described in Section \ref{flattening stratification section} all begin with the same observation. Let $S=K[x_0,\ldots,x_m]$ be a polynomial ring over a field. Let $\preceq $ be a multiplicative term order on $S$. Then a theorem of Macaulay states that the Hilbert function of $I$ is the same as the Hilbert function of its initial ideal with respect to this term order (see \cite{Eisenbud}*{Theorem 15.26}). Therefore, whenever two ideals in $S$ have Gr\"obner bases with the same leading monomials with respect to some term order, they will have the same initial ideal for that term order, hence they must have the same Hilbert function and Hilbert polynomial, and therefore they will lie in the same stratum of a flattening stratification. To reach a different stratum of the flattening stratification, it is necessary to alter the leading terms of the Gr\"obner basis --- for instance, by restricting to the locus where that coefficient vanishes. Here is a brief example to illustrate this idea. Let $\mathbb{A}^2$ have coordinates $c_1,c_2$, and let $\mathbb{P}^3$ have coordinates $x_0$,$x_1$,$x_2$,$x_3$. The ideal \[ I = \langle c_1 x_0x_2-c_2 x_1^2, c_1 x_0x_3-c_2 x_1x_2, c_1 x_1x_3-c_2 x_2^2 \rangle \] defines a 2-parameter family of subschemes of $\mathbb{P}^3$. A Gr\"obner basis for $I$ in $\mathbb{C}[c_1,c_2][x_0,x_1,x_2,x_3]$ with respect to the lexicographic term order is \[ \begin{array}{cc} c_1 x_0 x_2-c_2 x_1^2, c_1 x_0 x_3-c_2 x_1 x_2, c_1 x_1 x_3-c_2 x_2^2,\\ (c_1 c_2-c_2^2) x_1 x_2^2, c_2 x_1^2 x_3-c_2 x_1 x_2^2, (c_1 c_2-c_2^2) x_1^2 x_2, c_2 x_0 x_2^2-c_2 x_1^2 x_2, \\ (c_1 c_2^2-c_2^3) x_1^4, (c_1 c_2^2-c_2^3) x_2^4, c_2^2 x_1 x_2^2 x_3-c_2^2 x_2^4, c_2^2 x_0 x_1^2 x_2-c_2^2 x_1^4. \end{array} \] Over the locus where $c_1$, $c_2$, and $c_1-c_2$ are invertible, the initial ideal is \\$\langle x_0 x_2, x_0 x_3, x_1 x_3,x_1 x_2^2,x_1^2 x_2,x_1^4,x_2^4\rangle$ with Hilbert polynomial $P(t) = 8$. On the other hand, when $c_1=0$, or $c_2=0$, or $c_1-c_2=0$, we get a different initial ideal and Hilbert polynomial. For example, the locus $c_1=c_2 \neq 0$ yields the twisted cubic with $P(t) = 3t+1$. Note that to discover this locus, it is not necessary to compute the entire Gr\"obner basis; it would suffice for instance to compute the S-pair reduction for the first two generators, which yields $(c_1 c_2-c_2^2) x_1 x_2^2$. Modern software packages by Nabeshima, Montes, and Kapur, Sun, and Wang can completely analyze this example. However, these packages did not yield answers on the problems that arose in this work. Therefore, I used the strategy outlined above. I partially computed a Gr\"obner basis in \texttt{Macaulay2} \cite{Macaulay2}, and set some coefficients to zero. Remarkably, this was sufficient to obtain the equations of the genus $4 \leq g \leq 7$ Riemann surfaces with large automorphism groups. Some of the families analyzed in this manner had as many as six coefficients $c_1,\ldots,c_6$. \section{Example: a genus 7 Riemann surface with 64 automorphisms} \label{example section} Magaard, Shaska, Shpectorov, and V\"olklein's tables show that there exists a smooth, compact genus 7 Riemann surface with automorphism group $G$ given by the group labeled $(64,41)$ in the GAP library of small finite groups. It has $X/G \cong \mathbb{P}^1$. The quotient morphism is branched over 3 points of $\mathbb{P}^1$, and the ramification indices over these points are 2, 4, and 16. A naive search for a set of surface kernel generators in this group yields elements $g_1$ and $g_2$ with orders 2 and 4 such that $(g_1g_2)^{-1}$ has order 16. There are four relations among these generators: \begin{displaymath} g_1^2, \, g_2^4, \, (g_2^{-1} g_1)^2 g_2^2 g_1 g_2 g_1 g_2^{-1}, \, (g_2 g_1)^2 g_2^{-1} (g_1 g_2)^2 g_1 g_2^{-1} (g_1 g_2)^2 g_1 \end{displaymath} \textit{Step 1.} We use \texttt{Magma} to compute the conjugacy classes and character table of $G$. There are 16 conjugacy classes. For convenience, write $g_3 = g_1^{-1} g_2^{-1} g_1g_2$. Then a list of representatives of the conjugacy classes is \[ \begin{array}{c} \operatorname{Id}, \, g_3^4, \, g_2^2,\, g_1,\, g_3^2,\, g_2^2 g_3^2,\, g_2^3,g_2,\\ g_1 g_2^2, \, g_3 g_2^2 g_3^4, \, g_3 g_2^2, \, g_3,g_2 g_1, \, g_2 g_1 g_2^2 g_3^2, \, g_2 g_1 g_3^2, \, g_2 g_1 g_2^2 \end{array} \] Next we compute the character table. The irreducible characters are given below by their values on the sixteen conjugacy classes. \begin{align*} \chi_{1} & = ( 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1 , 1)\\ \chi_{2} & = (1 , 1 , 1 , -1 , 1 , 1 , -1 , -1 , -1 , 1 , 1 , 1 , 1 , 1 , 1 , 1)\\ \chi_{3} & = (1 , 1 , 1 , 1 , 1 , 1 , -1 , -1 , 1 , 1 , 1 , 1 , -1 , -1 , -1 , -1)\\ \chi_{4} & = (1 , 1 , 1 , -1 , 1 , 1 , 1 , 1 , -1 , 1 , 1 , 1 , -1 , -1 , -1 , -1)\\ \chi_{5} & = (1 , 1 , -1 , 1 , 1 , -1 , -i , i , -1 , -1 , -1 , 1 , i , -i , i , -i)\\ \chi_{6} & = (1 , 1 , -1 , 1 , 1 , -1 , i , -i , -1 , -1 , -1 , 1 , -i , i , -i , i)\\ \chi_{7} & = (1 , 1 , -1 , -1 , 1 , -1 , i , -i , 1 , -1 , -1 , 1 , i , -i , i , -i)\\ \chi_{8} & = (1 , 1 , -1 , -1 , 1 , -1 , -i , i , 1 , -1 , -1 , 1 , -i , i , -i , i)\\ \end{align*} \begin{align*} \chi_{9} & = (2 , 2 , -2 , 0 , 2 , -2 , 0 , 0 , 0 , 2 , 2 , -2 , 0 , 0 , 0 , 0)\\ \chi_{10} & = (2 , 2 , 2 , 0 , 2 , 2 , 0 , 0 , 0 , -2 , -2 , -2 , 0 , 0 , 0 , 0)\\ \chi_{11} & = (2 , 2 , 2 , 0 , -2 , -2 , 0 , 0 , 0 , 0 , 0 , 0 ,-\sqrt{2} , \sqrt{2}, \sqrt{2}, -\sqrt{2})\\ \chi_{12} & = (2 , 2 , -2 , 0 , -2 , 2 , 0 , 0 , 0 , 0 , 0 , 0 , \sqrt{2}i , \sqrt{2}i , -\sqrt{2}i , -\sqrt{2}i)\\ \chi_{13} & = (2 , 2 , 2 , 0 , -2 , -2 , 0 , 0 , 0 , 0 , 0 , 0 , \sqrt{2}, -\sqrt{2}, -\sqrt{2}, \sqrt{2})\\ \chi_{14} & = (2 , 2 , -2 , 0 , -2 , 2 , 0 , 0 , 0 , 0 , 0 , 0 , -\sqrt{2}i , -\sqrt{2}i, \sqrt{2}i , \sqrt{2}i)\\ \chi_{15} & = (4 , -4 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , -\sqrt{8}i , \sqrt{8}i , 0 , 0 , 0 , 0 , 0)\\ \chi_{16} & = (4 , -4 , 0 , 0 , 0 , 0 , 0 , 0 , 0 , \sqrt{8}i , -\sqrt{8}i , 0 , 0 , 0 , 0 , 0) \end{align*} \textit{Step 2.} Let $V_i$ be the irreducible $G$-module with character $\chi_i$ given by the table above. For any $G$-module $V$, let $V\cong \bigoplus_{i=1}^{r} V_{i}^{\oplus m_i}$ be its decomposition into irreducible $G$-modules. We use the Eichler trace formula in \texttt{Magma} to compute these multiplicities $m_i$ for several relevant $G$-modules. Let $S=\mathbb{C}[x_0,x_1,x_2,x_3,x_4,x_5,x_6]$, and let $S_d$ denote polynomials of degree $d$. Let $I_d$ be the kernel defined by \begin{displaymath} 0 \rightarrow I_d \rightarrow S_d \rightarrow \Gamma(\omega_X^{\otimes d}) \rightarrow 0. \end{displaymath} Then we have \begin{align*} S_1 \cong \Gamma(\omega_X) & \cong V_8 \oplus V_{14} \oplus V_{15} \\ I_2 & \cong V_{3} \oplus V_{5} \oplus V_{10} \oplus V_{11} \oplus V_{16} \\ S_2 & \cong V_{3}^{\oplus 2} \oplus V_{5} \oplus V_{6} \oplus V_{10}^{\oplus 2} \oplus V_{11}^{\oplus 2} \oplus V_{13} \oplus V_{14} \oplus V_{15} \oplus V_{16}^{\oplus 2} \\ \Gamma(\omega_X^{\otimes 2})& \cong V_{3} \oplus V_{6} \oplus V_{10} \oplus V_{11} \oplus V_{13} \oplus V_{14} \oplus V_{15} \oplus V_{16} \end{align*} We use \texttt{GAP} to obtain matrix representatives of a $G$ action with character equal to the character of the $G$ action on $S_1$. Such a representation is obtained by mapping the generators $g_1$ and $g_2$ to the matrices below. \begin{displaymath} \left[ \begin{array}{rrrrrrr} -1 & 0 & 0 & 0 & 0 & 0 & 0\\ 0 & 0 & 1 & 0 & 0 & 0 & 0\\ 0 & 1 & 0 & 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 & 1 & 0 & 0\\ 0 & 0 & 0 & 1 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 & 0 & 0 & 1\\ 0 & 0 & 0 & 0 & 0 & 1 & 0 \end{array} \right], \qquad \left[ \begin{array}{rrrrrrr} \zeta_{8}^2& 0 & 0 & 0 & 0 & 0 & 0\\ 0 & 0 & -\zeta_{8} & 0 & 0 & 0 & 0\\ 0 & -\zeta_{8}^3 & 0 & 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 & 0 & 0 & 1\\ 0 & 0 & 0 & 0 & 0 & -\zeta_{8} & 0\\ 0 & 0 & 0 & 0 &-\zeta_{8}^3 & 0 & 0\\ 0 & 0 & 0 & 1 & 0 & 0 & 0 \end{array} \right] \end{displaymath} The decomposition of $S_1$ as a sum of three irreducible $G$-modules gives rise to the block diagonal form of these matrices. \textit{Step 4.} We use the projection formula in \texttt{Magma} to decompose the $G$-module of quadrics $S_2$ into its isotypical components. When an isotypical component has multiplicity greater than 1, we (noncanonically) choose ordered bases so that the $G$ action is given by the same matrices on each ordered basis. \[ \begin{array}{lllll} S_{2,3} & \cong & V_3^{\oplus 2} & = & \langle x_0^2\rangle \oplus \langle x_1x_2 \rangle \\ S_{2,5} & \cong & V_{5} & = & \langle x_3x_4 -\zeta_{8}x_5x_6 \rangle \\ S_{2,10} & \cong & V_{10}^{\oplus 2} &= & \langle x_1^2,x_2^2\rangle \oplus \langle x_3x_6+ix_4x_5, ix_3x_6+x_4x_5\rangle \\ S_{2,11} & \cong & V_{11}^{\oplus 2} & = & \langle x_0x_1, x_0x_2\rangle \oplus \langle x_3^2 +\zeta_{8}^3 x_5^2, -x_4^2 -\zeta_{8}^3x_6^2\rangle \\ S_{2,16} & \cong & V_{16}^{\oplus 2} & = &\langle x_0x_3,x_0x_4, x_0x_5,x_0x_6\rangle \oplus \langle -\zeta_{8} x_2 x_6,\zeta_{8} x_1 x_5,-x_2 x_4,x_1 x_3\rangle \end{array} \] The first isotypical subspace yields a polynomial of the form $c_1 x_0^2 + c_2 x_1x_2$. We may assume that $c_1$ and $c_2$ are nonzero, scale $x_0$ to make $c_1=c_2$, and then divide by $c_1$ to obtain the polynomial $x_0^2+x_1 x_2$. The second isotypical subspace yields the polynomial $x_3 x_4-\zeta_{8} x_5 x_6$. The third isotypical subspace yields polynomials of the form $c_3 x_0 x_1+c_4 (x_3^2+\zeta_{8}^3 x_5^2)$ and $c_3x_0 x_2+c_4 (-x_4^2-\zeta_{8}^3 x_6^2)$. We assume that $c_3$ and $c_4$ are nonzero, scale $x_1,x_2$ to make $c_3=c_4$, then divide by $c_3$ and $c_4$. In the remaining isotypical subspaces no further scaling is possible, and hence we are left with two undetermined coefficients $c_6$ and $c_8$. Thus, a Riemann surface in this family has an ideal of the form \begin{displaymath} \begin{array}{l} x_0^2+x_1 x_2 \\ x_3 x_4-\zeta_{8} x_5 x_6\\ x_1^2+x_3 x_6+i x_4 x_5\\ x_2^2+i x_3 x_6+x_4 x_5 \\ x_0 x_1+c_6 (x_3^2+\zeta_{8}^3 x_5^2)\\ x_0 x_2+c_6 (-x_4^2-\zeta_{8}^3 x_6^2)\\ x_0 x_3+c_8 (-\zeta_{8} x_2 x_6)\\ x_0 x_4+c_8 (\zeta_{8} x_1 x_5)\\ x_0 x_5+c_8 (-x_2 x_4)\\ x_0 x_6+c_8 (x_1 x_3) \end{array} \end{displaymath} \textit{Step 5.} To find values of the coefficients $c_6,c_6$ that yield a smooth curve, we partially compute a flattening stratification. Begin Buchberger's algorithm. We compute the S-pair reductions between the generators and find that \begin{align*} S(f_1,f_6) & \rightarrow (c_6 c_8 + \zeta_8^{-1})x_1x_4x_5+ \cdots \\ S(f_1,f_9) & \rightarrow ( c_8^2 - \zeta_8^{-1} ) x_1 x_2x_5 + \cdots \end{align*} Therefore, in Buchberger's algorithm, these polynomials will be added to the Gr\"obner basis. This suggests that we study the locus given by the equations $c_8^2 -\zeta_8^{-1}=0$ and $c_6 c_8 + \zeta_8^{-1} = 0$ as an interesting stratum in the flattening stratification. We check in \texttt{Magma} that the values $c_6 = \zeta_{16}^{7}$ and $c_8 = \zeta_{16}^{-1}$ yield a smooth genus 7 curve in $\mathbb{P}^6$ with the desired automorphism group. From these equations, we can compute the Betti table of this ideal: \begin{center} \begin{tabular}{|c|c|c|c|c|c|} \hline 1 & & & & & \\ \hline & 10 & 16 & 3 & & \\ \hline && 3 & 16 & 10 & \\ \hline &&&&& 1 \\ \hline \end{tabular} \end{center} Schreyer has classified Betti tables of genus 7 canonical curves in \cite{Schreyer1986}. This Betti table implies that the curve is tetragonal (there exists a degree 4 morphism $C \rightarrow \mathbb{P}^1$) but not trigonal or hyperelliptic, and it has no degree 6 morphism $C \rightarrow \mathbb{P}^2$. \section{Results} \label{results section} This project had two goals. The first goal was to establish that the heuristics described in Section \ref{heuristics section} allow us to run a variant of the main algorithm to completion for genus $4 \leq g \leq 7$ Riemann surfaces with large automorphism groups. To this end, for each Riemann surface from Table 4 of \cite{MSSV}, the website \cite{mywebpage} contains a link to a calculation where a variant of the main algorithm is used to produce equations. The surface kernel generators needed to begin the algorithm were generally obtained by a naive search through the triples or quadruples in the groups listed in Table 4 of \cite{MSSV}. However, my \texttt{Magma} code also includes functions allowing the user to input surface kernel generators from any type of group, or to put in matrix surface kernel generators with the desired representation on $\Gamma(\omega_X)$. We note that Breuer's data has been recently extended and republished by Paulhus \cite{Paulhus}, and Conder's data is available online \cite{Conder}, so these sources could be used instead. The equations obtained depend strongly on the matrix generators of the representation $\operatorname{Aut}(X)$ on $\Gamma(\omega_X)$. I generally obtained these matrices from \texttt{Magma}, \texttt{GAP}, the papers \cites{KKg34,KKg5}, or \cite{Breuer}*{Appendix B}, and thus had little control over this step. Indeed, in a few cases, the resulting equations are almost comically bad; for an example of this, compare my equations at \cite{mywebpage} for the genus 7 curve with 504 automorphisms to Macbeath's equations for this curve. Given this, it is perhaps surprising that in most cases, the algorithm produces reasonable equations (i.e., polynomials supported on a small number of monomials with small coefficients). The second goal of this project was to create a reference that would contain the most useful information about the equations and automorphisms of these curves. Thus, in this section, I print the best equations and automorphisms that I know, whether these were found in the literature or by the main algorithm. Many of the equations for genus $4 \leq g \leq 6$ are classical, and references are given whenever possible. However, the matrix surface kernel generators are not always equally easy to find. The equations for the genus 7 curves are almost all new, as are most of the 1-parameter families on the website \cite{mywebpage}. \subsection{Description of the tables} In the following tables I give equations for the Riemann surfaces of genus $4 \leq g \leq 7$ with large automorphism groups that are unique in moduli ($\delta=0$ in the notation of Table 4 of \cite{MSSV}). The 1-parameter families ($\delta=1$) are not printed here but can be found on the website \cite{mywebpage}. I order the examples the same way they appear in \cite{MSSV}. For hyperelliptic Riemann surfaces, I give an equation of the form $y^2 = f(x)$. Many of these are classically known, and all of them can be found in \cite{Shaska}. For plane quintics in genus 6, we give the plane quintic and surface kernel generators in $\operatorname{GL}(3,\mathbb{C})$. The canonical ideal and $G$ action can be easily computed from this data. For nonhyperelliptic curves that are not plane quintics, we print equations of the canonical ideals and surface kernel generators as elements of $\operatorname{GL}(g,\mathbb{C})$. Whenever such a matrix $M \in \operatorname{GL}(g,\mathbb{C})$ is sufficiently sparse, I frequently write the product $ M [x_0,\ldots, x_{g-1}]^{t}$ to save space. For the cyclic trigonal equations, I also print a cyclic trigonal equation, that is, one of the form $y^3 = \prod_{i=1}^{d_1} (x-\alpha_i) \prod_{i=1}^{d_2} (x-\beta_i)$, following the notation of \cite{AchterPries}*{Section 2.5} (where cyclic trigonal curves are also called \emph{trielliptic}). Throughout the tables below, canonical ideals are shown in the polynomial ring $\mathbb{C}[x_0,\ldots,x_{g-1}]$. The symbol $\zeta_n$ denotes $e^{2 \pi i/n}$, and we write $i$ for $\zeta_4$. \subsection{Genus 4} In genus 4, every Riemann surface is either hyperelliptic or trigonal. Of the nine entries in Table 4 of \cite{MSSV}, four are hyperelliptic, four are cyclic trigonal, and one is general trigonal. Note: the Riemann surface with automorphism group $(120,34) = S_5$ is known as Bring's curve. Its best-known embedding is in $\mathbb{P}^5$, with equations $\sum_{i=0}^{4} x_i$, $\sum_{i=0}^{4} x_i^2$, $\sum_{i=0}^{4} x_i^3$. \noindent Genus 4, Locus 1: Group (120,34) = $S_5$, signature (2,4,5), general trigonal\\ \begin{tabular}{ll} Ideal: & $x_0^2 + x_0 x_1 + x_1^2 - x_1x_2+ x_2^2 - x_2 x_3 + x_3^2$, \\ & $x_0^2 x_1 + x_0 x_1^2 + x_1^2 x_2 - x_1x_2^2+ x_2^2 x_3 - x_2x_3^2$\end{tabular}\\ \begin{tabular}{ll} Maps: &$(x_0,x_1,x_2,x_3) \mapsto (-x_0,-x_1,-x_2,-x_2+x_3)$,\\ & $(x_0,x_1,x_2,x_3) \mapsto (x_0+x_1,-x_0-x_2,-x_0-x_3,-x_3)$\end{tabular}\\ \mbox{} \\ Genus 4, Locus 2: Group (72,42), signature (2,3,12), cyclic trigonal\\ \begin{tabular}{l} Trigonal equation: $y^3 = x(x^4-1)$\end{tabular}\\ \begin{tabular}{ll} Ideal: & $x_1x_3-x_2^2,$ \\ & $x_0^3 - x_1^2 x_2 +x_2 x_3^2 $\end{tabular} \\ \begin{tabular}{l} Maps: $\left[ \begin{array}{rrrr} -1 & 0 & 0 & 0 \\ 0 & 0 & 0 & -i \\ 0 & 0 & -1 & 0 \\ 0 & i & 0 & 0 \end{array} \right]$, \qquad $\left[ \begin{array}{rrrr} -\zeta_6 & 0 & 0 & 0 \\ 0 & -\frac{1}{2}\zeta_{12} & \frac{1}{2}\zeta_{3}& \frac{1}{2}\zeta_{12} \\ 0 & \zeta_{12} & 0 & \zeta_{12} \\ 0 & -\frac{1}{2}\zeta_{12}& -\frac{1}{2}\zeta_{3} & \frac{1}{2}\zeta_{12} \end{array} \right]$, \end{tabular}\\ \mbox{} \\ Genus 4, Locus 3: Group (72,40), signature (2,4,6), cyclic trigonal\\ \begin{tabular}{l} Trigonal equation: $y^3 = (x^3-1)^2(x^3+1)$\end{tabular}\\ \begin{tabular}{ll} Ideal: &$x_0 x_3-x_1 x_2,$ \\ & $x_1^3-x_0^3-x_3^3-x_2^3 $\end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,x_1,x_2,x_3) \mapsto(-x_0,x_2,x_1,-x_3)$\\ & $(x_0,x_1,x_2,x_3) \mapsto (-x_2,\zeta_6^2x_0,\zeta_6x_3,x_1)$\end{tabular}\\ \mbox{} \\ Genus 4, Locus 4: Group (40,8), signature (2,4,10), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{10}-1$ \end{tabular} \\ Genus 4, Locus 5: Group (36,12), signature (2,6,6), cyclic trigonal\\ \begin{tabular}{l} Trigonal equation: $y^3 = (x^3-1)(x^3+1)$\end{tabular}\\ \begin{tabular}{ll} Ideal: & $x_1 x_3-x_2^2,$ \\ & $x_0^3-x_3^3+x_1^3 $\end{tabular}\\ \begin{tabular}{ll} Maps: &$(x_0,x_1,x_2,x_3) \mapsto(-x_0,\zeta_3 x_3,-x_2,-\zeta_6 x_1)$\\ & $(x_0,x_1,x_2,x_3) \mapsto (\zeta_3 x_0,-\zeta_3 x_3, \zeta_6 x_2, -x_1)$\end{tabular}\\ \newpage \noindent Genus 4, Locus 6: Group (32,19), signature (2,4,16), hyperelliptic\\ \begin{tabular}{l} $y^2=x^9-x$ \end{tabular} \\ \mbox{} \\ Genus 4, Locus 7: Group (24,3), signature (3,4,6), hyperelliptic\\ \begin{tabular}{l} $y^2=x(x^4-1)(x^4+2i\sqrt{3}+1)$ \end{tabular} \\ \mbox{} \\ Genus 4, Locus 8: Group (18,2), signature (2,9,18), hyperelliptic\\ \begin{tabular}{l} $y^2=x^9-1$ \end{tabular} \\ \mbox{} \\ Genus 4, Locus 9: Group (15,1), signature (3,5,15), cyclic trigonal\\ \begin{tabular}{l} Trigonal equation: $y^3 = x^5-1$ \end{tabular}\\ \begin{tabular}{ll} Ideal: & $x_1 x_3 -x_2^2,$ \\ & $x_0^3 - x_1^2 x_2 +x_3^3 $\end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,x_1,x_2,x_3) \mapsto (\zeta_{3}^{2} x_0,\zeta_{3} x_1, \zeta_{3}x_2, \zeta_{3} x_3)$\\ & $(x_0,x_1,x_2,x_3) \mapsto (\zeta_{5} x_0,\zeta_{5}^3 x_1,\zeta_{5}^2 x_2,\zeta_{5} x_3)$ \end{tabular} \subsection{Genus 5} Of the ten entries in Table 4 of \cite{MSSV}, five are hyperelliptic, and one is cyclic trigonal. The remaining four are general, hence their canonical models are complete intersections of three quadrics. \noindent Genus 5, Locus 1: Group (192,181), signature (2,3,8) \\ \begin{tabular}{ll} Ideal: & Wiman, \cite{Wiman}:\\ & $x_0^2+x_3^2+x_4^2$, \\ & $x_1^2+x_3^2-x_4^2$\\ & $x_2^2+x_3 x_4$\end{tabular}\\ \begin{tabular}{ll} Maps: & $\left[\begin{array}{rrrrr} 0 & 0 & \frac{1}{2}(i+1) & 0 & 0 \\ 0 & -1 & 0 & 0 & 0\\ 1-i & 0 & 0 & 0 & 0\\ 0 & 0 & 0 & -\frac{1}{\sqrt{2}} & -\frac{i}{\sqrt{2}}\\ 0 & 0 & 0 & \frac{i}{\sqrt{2}} & \frac{1}{\sqrt{2}} \end{array} \right]$,\\ & $\quad \left[\begin{array}{rrrrr} 0 & \zeta_8^{-1} & 0 & 0 & 0 \\ 0 & 0 & -\frac{1}{\sqrt{2}} & 0 & 0\\ -1-i & 0 & 0 & 0 & 0\\ 0 & 0 & 0 & \frac{1}{2}(i-1) & -\frac{1}{2}(i+1)\\ 0 & 0 & 0 & -\frac{1}{2}(i-1) & -\frac{1}{2}(i+1) \end{array} \right]$ \end{tabular}\\ \mbox{} \\ Genus 5, Locus 2: Group (160,234), signature (2,4,5) \\ \begin{tabular}{ll} Ideal: & Wiman, \cite{Wiman}:\\ & $x_0^2+x_1^2+x_2^2+x_3^2+x_4^2$, \\ & $x_0^2+\zeta_5 x_1^2+\zeta_5^2 x_2^2+\zeta_5^3 x_3^2+\zeta_5^4 x_4^2,$\\ & $\zeta_5^4 x_0^2+\zeta_5^3 x_1^2+\zeta_5^2 x_2^2+\zeta_5 x_3^2+x_4^2$\end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,x_1,x_2,x_3,x_4) \mapsto (-x_3,x_2,x_1,-x_0,-x_4)$,\\ & $(x_0,x_1,x_2,x_3,x_4) \mapsto (-x_0,x_4,-x_3,x_2,-x_1)$\end{tabular}\\ \mbox{} \\ Genus 5, Locus 3: Group (120,35), signature (2,3,10), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{11}+11x^6-x$\end{tabular}\\ \mbox{}\\ Genus 5, Locus 4: Group (96,195), signature (2,4,6) \\ \begin{tabular}{ll} Ideal: & Wiman, \cite{Wiman}:\\ & $x_0^2 + x_3^2 + x_4^2$, \\ & $x_1^2+\zeta_3 x_3^2+\zeta_3^2 x_4^2,$\\ & $x_2^2+\zeta_3^2 x_3^2+\zeta_3 x_4^2$\end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,x_1,x_2,x_3,x_4) \mapsto (-x_2,-x_1,-x_0,\zeta_3^2 x_4,\zeta_3 x_3)$,\\ & $(x_0,x_1,x_2,x_3,x_4) \mapsto (-x_0,x_2,-x_1,-x_4,x_3)$\end{tabular}\\ \newpage \noindent Genus 5, Locus 5: Group (64,32), signature (2,4,8) \\ \begin{tabular}{ll} Ideal: & Wiman, \cite{Wiman}:\\ & $x_0^2+x_1^2+x_2^2+x_3^2+x_4^2$, \\ & $x_0^2+i x_1^2-x_2^2-i x_3^2,$\\ & $x_0^2-x_1^2+x_2^2-x_3^2$\end{tabular} \\ \begin{tabular}{ll} Maps: &$(x_0,x_1,x_2,x_3,x_4) \mapsto (-x_0,x_1,-x_2,-x_3,-x_4)$,\\ & $(x_0,x_1,x_2,x_3,x_4) \mapsto (ix_1,-ix_2,ix_3,-ix_0,ix_4)$\end{tabular}\\ \mbox{} \\ Genus 5, Locus 6: Group (48,14), signature (2,4,12), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{12}-1$\end{tabular}\\ \mbox{}\\ Genus 5, Locus 7: Group (48,30), signature (3,4,4), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{12}-33x^8-33x^{4}+1$\end{tabular}\\ \mbox{}\\ Genus 5, Locus 8: Group (40,5), signature (2,4,20), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{11}-x$\end{tabular}\\ \mbox{}\\ Genus 5, Locus 9: Group (30,2), signature (2,6,15), cyclic trigonal \\ \begin{tabular}{l} Trigonal equation: $y^3=(x^5-1)x^2$ \end{tabular}\\ \begin{tabular}{ll} Ideal: & $x_0 x_3-x_1 x_2$, \, $x_0 x_4-x_1 x_3$, \, $x_2 x_4-x_3^2,$\\ & $x_0^2 x_1-x_3 x_4^2+x_2^3,$\\ & $x_0 x_1^2-x_4^3+x_2^2 x_3$ \end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,x_1,x_2,x_3,x_4) \mapsto (\zeta_5 x_1,\zeta_5^4 x_0,-\zeta_5^2 x_4,-x_3,-\zeta_5^3 x_2)$,\\ & $(x_0,x_1,x_2,x_3,x_4) \mapsto (\zeta_{15}^{14} x_1,\zeta_{15}^{11} x_0,-\zeta_{15}^{13} x_4,-\zeta_{15}^{10} x_3,-\zeta_{15}^7x_2)$\end{tabular}\\ \mbox{} \\ Genus 5, Locus 10: Group (22,2), signature (2,11,22), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{11}-1$\end{tabular} \subsection{Genus 6} Table 4 in \cite{MSSV} contains eleven entries for genus 6 Riemann surfaces with large automorphism groups and no moduli ($\delta=0$). Of these, four are hyperelliptic, three are cyclic trigonal, and three are plane quintics; only one is general. For the plane quintics, we give the plane quintic equation in the variables $y_0,y_1,y_2$, and surface kernel generators acting on the plane. The canonical model of a plane quintic lies on the Veronese surface, and the multiples of the quintic by $y_0,y_1,y_2$ may be encoded as cubics in $x_0,\ldots,x_5$. \\ \noindent Genus 6, Locus 1: Group (150,5), signature (2,3,10), plane quintic \\ \begin{tabular}{l} Plane quintic equation: $y_0^5+y_1^5+y_2^5$\end{tabular}\\ \begin{tabular}{ll} Maps: & $(y_0,y_1,y_2) \mapsto (-\zeta_5^3 y_1, -\zeta_5^2 y_0, -y_2)$,\\ & $(y_0,y_1,y_2) \mapsto (-\zeta_5^3 y_1, -\zeta_5^2 y_0, -y_2)$ \end{tabular}\\ \mbox{} \\ \noindent Genus 6, Locus 2: Group $(120,34) = S_5$, signature (2,4,6) \\ \begin{tabular}{ll} Ideal: & Inoue and Kato,\cite{InoueKato}:\\ &$-x_0x_2 + x_1x_2 - x_0x_3 + x_1x_4,$\\ & $ -x_0x_1 + x_1x_2 - x_0x_3 + x_2x_5,$ \\ & $ -x_0x_1 - x_0x_2 - 2x_0x_3 - x_3x_4 - x_3x_5,$ \\ & $ -x_0x_1 - x_0x_2 - x_0x_3 - x_1x_4 - x_3x_4 - x_4x_5,$ \\ & $ -x_0x_1 - x_0x_2 - x_0x_3 - x_2x_5 - x_3x_5 - x_4x_5$, \\ & $2 (\sum_{i=1}^{6} x_i^2) + x_0 x_1 + x_0 x_2 + x_1 x_2 + 2 x_1 x_3 + 2 x_2 x_3 $\\ & $\mbox{ }+2 x_0 x_4 + 2 x_2 x_4 + x_3 x_4 + 2 x_0 x_5 + 2 x_1 x_5 + x_3 x_5 + x_2 x_5 $ \end{tabular}\\ \begin{tabular}{l} Maps: $\left[\begin{array}{rrrrrr} 0 & 0 & 0 & 0 & -1 & 0 \\ 0 & -1 & 1 & 0 & -1 & 1 \\ -1 & 0 & 0 & 0 & 0 & 1 \\ -1 & 0 & 1 & -1 & 0 & 1 \\ -1 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 & -1 & 0 \end{array} \right]$, \quad $\left[\begin{array}{rrrrrr} 0 & 0 & 1 & 0 & -1 & 0 \\ -1 & 0 & 1 & -1 & 0 & 1\\ -1 & 0 & 0 & 0 & 0 & 0 \\ -1 & 0 & 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 & -1 & 0 \\ 0 & -1 & 1 & 0 & -1 & 0 \end{array} \right]$ \end{tabular}\\ \mbox{} \\ Genus 6, Locus 3: Group (72,15), signature (2,4,9), cyclic trigonal \\ \begin{tabular}{l} Trigonal equation: $y^3 = (x^4-2 \sqrt{3} i x^2+1)(x^4+2 \sqrt{3}ix^2 + 1)^2$\end{tabular}\\ \begin{tabular}{ll}Ideal: & $x_0 x_2 - x_1^2$, $x_0 x_4 - x_1 x_3 $, $x_0 x_5-x_1 x_4 $,\\ & $ x_1 x_4 - x_2 x_3$, $x_1x_5-x_2x_4 $, $x_3x_5-x_4^2 $,\\ & $x_0^3+(4 \zeta_6-2) x_0^2 x_2+x_0 x_2^2+x_3^3+(-4 \zeta_6+2) x_3^2 x_5+x_3 x_5^2$\\ & $x_0^2 x_1+(4 \zeta_6-2) x_0 x_1 x_2+x_1 x_2^2+x_3^2 x_4+(-4 \zeta_6+2) x_3 x_4 x_5+x_4 x_5^2,$\\ & $x_0^2 x_2+(4 \zeta_6-2) x_0 x_2^2+x_2^3+x_3^2 x_5+(-4 \zeta_6+2) x_3 x_5^2+x_5^3$\end{tabular} \\ \begin{tabular}{ll} Maps: & $\left[\begin{array}{rrrrrr} 0 & 0 & 0 & -\frac{1}{2}\zeta_9 & \frac{1}{2}\zeta_{36}^{13} & \frac{1}{2}\zeta_9\\ 0 & 0 & 0 & -\zeta_{36}^{13} & 0 & -\zeta_{36}^{13}\\ 0 & 0 & 0 & \frac{1}{2}\zeta_9 & \frac{1}{2}\zeta_{36}^{13} & -\frac{1}{2}\zeta_9\\ \frac{1}{2}\zeta_{36}^{14} & \frac{1}{2}\zeta_{36}^5 & -\frac{1}{2}\zeta_{36}^{14} & 0 & 0 & 0\\ -\zeta_{36}^5 & 0 & -\zeta_{36}^5 & 0 & 0 & 0\\ -\frac{1}{2}\zeta_{36}^{14} & \frac{1}{2}\zeta_{36}^5 & \frac{1}{2}\zeta_{36}^{14} & 0 & 0 & 0 \end{array} \right],$\\ & $(x_0,\ldots,x_5) \mapsto (\zeta_{12}x_3,\zeta_3 x_4, -\zeta_{12} x_5, \zeta_{12}^5x_0,\zeta_3^2 x_1,-\zeta_{12}^5x_2) $\end{tabular}\\ \mbox{} \\ Genus 6, Locus 4: Group (56,7), signature (2,4,14), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{14}-1$ \end{tabular}\\ \mbox{}\\ Genus 6, Locus 5: Group (48,6), signature (2,4,24), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{13}-x$ \end{tabular}\\ \mbox{} \\ Genus 6, Locus 6: Group (48,29), signature (2,6,8), hyperelliptic \\ \begin{tabular}{l} $y^2=x(x^4-1)(x^{8}+14x^4+1)$ \end{tabular}\\ \mbox{} \\ Genus 6, Locus 7: Group (48,15), signature (2,6,8), cyclic trigonal \\ \begin{tabular}{l} Trigonal equation: $y^3 = (x^4-1)^2(x^4+1)$\end{tabular}\\ \begin{tabular}{ll} Ideal: & $x_0 x_2 - x_1^2$, $x_0 x_4 - x_1 x_3 $, $x_0 x_5-x_1 x_4 $,\\ & $ x_1 x_4 - x_2 x_3$, $x_1x_5-x_2x_4 $, $x_3x_5-x_4^2 $,\\ & $x_0 x_1^2 - x_2^3 - x_3 x_4^2 - x_5^3, $\\ & $x_0^2 x_1 - x_1 x_2^2 - x_3^2 x_4 - x_4 x_5^2,$\\ & $x_0^3 - x_1^2 x_2 - x_3^3 - x_4^2 x_5$\end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,\ldots,x_5) \mapsto (\zeta_8^3 x_5, -i x_4, \zeta_8 x_3, -\zeta_8^3 x_2, i x_1, -\zeta_8 x_0)$,\\ & $(x_0,\ldots,x_5)\mapsto (-\zeta_6 x_2,\zeta_6 x_1, -\zeta_6 x_0, -\zeta_3 x_5,\zeta_3 x_4,-\zeta_3 x_3)$\end{tabular}\\ \mbox{} \\ Genus 6, Locus 8: Group (39,1), signature (3,3,13), plane quintic \\ \begin{tabular}{l} Plane quintic equation: $y_0^4y_1 + y_1^4y_2 + y_2^4y_0$\end{tabular}\\ \begin{tabular}{ll} Maps: & $(y_0,y_1,y_2) \mapsto (\zeta_{13}^4 y_1, \zeta_{13}^{10} y_2, \zeta_{13}^{12} y_0)$,\\ & $(y_0,y_1,y_2) \mapsto (\zeta_{13}^8 y_2, \zeta_{13}^{7} y_0, \zeta_{13}^{11} y_1)$\end{tabular}\\ \mbox{} \\ Genus 6, Locus 9: Group (30,1), signature (2,10,15), plane quintic \\ \begin{tabular}{l} Plane quintic equation: $y_0^5+y_1^4 y_2+2 \zeta_5 y_1^3 y_2^2+2 \zeta_5^2 y_1^2 y_2^3+\zeta_5^3 y_1 y_2^4$\end{tabular}\\ \begin{tabular}{ll} Maps: & $(y_0,y_1,y_2) \mapsto (y_0,\zeta_5 y_2, \zeta_5^4 y_1)$,\\ & $(y_0,y_1,y_2) \mapsto (\zeta_5^3 y_0, -\zeta_5y_1-\zeta_5^2y_2,\zeta_5)$\end{tabular}\\ \mbox{} \\ Genus 6, Locus 10: Group (26,2), signature (2,13,26), hyperelliptic\\ \begin{tabular}{l} $y^2=x^{13}-1$ \end{tabular}\\ Genus 6, Locus 11: Group (21,2), signature (3,7,21), cyclic trigonal \\ \begin{tabular}{l} Trigonal equation: $y^3 = x^7-1$\end{tabular} \\ \begin{tabular}{ll} Ideal: &$x_0 x_3-x_1 x_2 $, $x_0 x_4-x_1 x_3 $, $x_0 x_5-x_1 x_4 $, \\ & $x_2 x_4-x_3^2 $, $x_2 x_5-x_3 x_4 $, $x_3 x_5-x_4^2 $, \\ & $x_0^3-x_3 x_5^2 +x_2^3 $,\\ & $x_0^2 x_1-x_4 x_5^2 + x_2^2 x_3$,\\ & $x_0 x_1^2-x_5^3+x_2^2 x_4 $\end{tabular} \\ \begin{tabular}{ll} Maps:& $(x_0,\ldots,x_5) \mapsto (x_0,\zeta_7 x_1, x_2, \zeta_7 x_3,\zeta_7^2 x_4 \zeta_7^3 x_5)$,\\ & $(x_0,\ldots,x_5)\mapsto (\zeta_3^2 x_0, \zeta_3^2 x_1, \zeta_3 x_2, \zeta_3 x_3, \zeta_3 x_4, \zeta_3 x_5)$ \end{tabular} \subsection{Genus 7} Of the thirteen entries in Table 4 of \cite{MSSV} for genus 7 curves, three are hyperelliptic and two are cyclic trigonal. After computing the canonical equations of the nonhyperelliptic Riemann surfaces, we can compute the Betti tables of these ideals and use the results of \cite{Schreyer1986} to classify the curve as having a $g^{1}_{4}$, $g^{2}_{6}$, $g^{1}_{3}$, or none of these. \noindent Genus 7, Locus 1: Group $(504,156)$, signature (2,3,7) \\ \begin{tabular}{ll} Ideal: & Macbeath,\cite{Macbeath}: \\ & $ x_0^2+x_1^2+x_2^2+x_3^2+x_4^2+x_5^2+x_6^2,$ \\ & $ x_0^2+\zeta_{7} x_1^2+\zeta_{7}^2 x_2^2+\zeta_{7}^3 x_3^2+\zeta_{7}^4 x_4^2+\zeta_{7}^5 x_5^2+\zeta_{7}^6 x_6^2,$ \\ & $x_0^2+\zeta_{7}^{-1} x_1^2+\zeta_{7}^{-2} x_2^2+\zeta_{7}^{-3} x_3^2+\zeta_{7}^{-4} x_4^2+\zeta_{7}^{-5} x_5^2+\zeta_{7}^{-6} x_6^2,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_0 x_6-(\zeta_{7}^{-2}-\zeta_{7}^2) x_1 x_4+(\zeta_{7}-\zeta_{7}^{-1}) x_3 x_5,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_1 x_0-(\zeta_{7}^{-2}-\zeta_{7}^2) x_2 x_5+(\zeta_{7}-\zeta_{7}^{-1}) x_4 x_6,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_2 x_1-(\zeta_{7}^{-2}-\zeta_{7}^2) x_3 x_6+(\zeta_{7}-\zeta_{7}^{-1}) x_5 x_0,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_3 x_2-(\zeta_{7}^{-2}-\zeta_{7}^2) x_4 x_0+(\zeta_{7}-\zeta_{7}^{-1}) x_6 x_1,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_4 x_3-(\zeta_{7}^{-2}-\zeta_{7}^2) x_5 x_1+(\zeta_{7}-\zeta_{7}^{-1}) x_0 x_2,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_5 x_4-(\zeta_{7}^{-2}-\zeta_{7}^2) x_6 x_2+(\zeta_{7}-\zeta_{7}^{-1}) x_1 x_3,$ \\ & $(\zeta_{7}^{-3}-\zeta_{7}^3) x_6 x_5-(\zeta_{7}^{-2}-\zeta_{7}^2) x_0 x_3+(\zeta_{7}-\zeta_{7}^{-1}) x_2 x_4$ \end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,\ldots,x_6)\mapsto (x_0,-x_1,-x_2,-x_3,x_4,x_5,-x_6 )$,\\ & $\left[\begin{array}{rrrrrrr} 0 & \frac{1}{2} & \frac{1}{2} & -\frac{1}{2} & 0 & -\frac{1}{2} & 0 \\ -\frac{1}{2} & -\frac{1}{2} & \frac{1}{2} & 0 & -\frac{1}{2} & 0 & 0 \\ \frac{1}{2} & -\frac{1}{2} & 0 & -\frac{1}{2} & 0 & 0 & -\frac{1}{2}\\ -\frac{1}{2} & 0 & -\frac{1}{2} & 0 & 0 & -\frac{1}{2} & -\frac{1}{2} \\ 0 & \frac{1}{2} & 0 & 0 & -\frac{1}{2} & \frac{1}{2} & -\frac{1}{2} \\ \frac{1}{2} & 0 & 0 & \frac{1}{2} & -\frac{1}{2} & -\frac{1}{2} & 0 \\ 0 & 0 & \frac{1}{2} & \frac{1}{2} &\frac{1}{2} & 0 & -\frac{1}{2} \end{array} \right]$ \end{tabular}\\ \mbox{} \\ Genus 7, Locus 2: Group $(144,127)$, signature (2,3,12). Has $g^{2}_{6}$ \\ \begin{tabular}{ll} Ideal: &$x_0^2 + x_3x_4 - \zeta_6x_3x_5 - \zeta_6x_5x_6,$ \\ &$2ix_1^2 + x_3x_4 + \zeta_6x_3x_5 + 2x_4x_6 - \zeta_6x_5x_6,$ \\ &$2ix_1x_2 + (-2\zeta_6 + 1)x_3x_4 + \zeta_6x_3x_5 - 2\zeta_6x_4x_6 + \zeta_6x_5x_6,$ \\ &$2ix_2^2 -x_3x_4 + (-\zeta_6 + 2)x_3x_5 + (2\zeta_6 - 2)x_4x_6 + (-\zeta_6 + 2)x_5x_6,$ \\ & $x_1x_3 - \zeta_6x_2x_6 + \zeta_{12}x_4^2 + (\zeta_{12}^3 - 2\zeta_{12})x_4x_5 + \zeta_{12}x_5^2,$ \\ & $x_1x_4 -\zeta_3x_2x_5 -x_3x_6 - x_6^2,$ \\ & $x_1x_5 - x_2x_5 + x_3^2 + (-\zeta_6 + 2)x_3x_6,$ \\ & $x_1x_6 + x_2x_6 -\zeta_{12}x_4^2 + \zeta_{12}x_4x_5 - \zeta_{12}x_5^2,$ \\ & $x_2x_3 - \zeta_6x_2x_6 + \zeta_{12}x_4^2,$ \\ & $x_2x_4 + (-\zeta_6 - 1)x_2x_5 + \zeta_6x_3^2 + 2\zeta_6x_3x_6 + \zeta_6 x_6^2$\end{tabular}\\ \begin{tabular}{ll}Maps: &$\left[\begin{array}{rrrrrrr} -1 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & \zeta_{12}^{-1} & -\zeta_{12} & 0 & 0 & 0 & 0 \\ 0 & -\zeta_{12} & -\zeta_{12}^{-1} & 0 & 0 & 0 & 0\\ 0 & 0 & 0 & 0 & \zeta_6 & \zeta_6 & 0 \\ 0 & 0 & 0 & -\zeta_3 & 0 & 0 & \zeta_3\\ 0 & 0 & 0 & 0 & 0 & 0 & -\zeta_3 \\ 0 & 0 & 0 & 0 & 0 & \zeta_6 & 0 \end{array} \right]$, \\ & $\left[\begin{array}{rrrrrrr} \zeta_3 & 0 & 0 & 0 & 0 & 0 & 0 \\ 0 & -1 & \zeta_3 & 0 & 0 & 0 & 0 \\ 0 & \zeta_6 & 0 & 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & \zeta_3 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 & \zeta_3 & 0 & 0 \\ 0 & 0 & 0 & 0 & -\zeta_3 & -\zeta_6 & 0 \\ 0 & 0 & 0 & 0 & 0 & 0 & 1 \end{array} \right]$\end{tabular}\\ \newpage \noindent Genus 7, Locus 3: Group $(64,41)$, signature (2,3,16), tetragonal \\ \begin{tabular}{ll}Ideal: & $x_0^2+x_1 x_2,$ \\ & $x_3 x_4-\zeta_8 x_5 x_6,$ \\ & $x_1^2+x_3 x_6+i x_4 x_5,$ \\ & $x_2^2+i x_3 x_6+x_4 x_5,$ \\ & $x_0 x_1+\zeta_{16}^7 x_3^2-\zeta_{16}^5 x_5^2,$ \\ & $x_0 x_2-\zeta_{16}^7 x_4^2+\zeta_{16}^5 x_6^2,$ \\ & $x_0 x_3-\zeta_{16} x_2 x_6,$ \\ & $x_0 x_4+\zeta_{16} x_1 x_5,$ \\ & $x_0 x_5+\zeta_{16}^7 x_2 x_4,$ \\ & $x_0 x_6-\zeta_{16}^7 x_1 x_3$\end{tabular} \\ \begin{tabular}{ll}Maps:& $(x_0,\ldots,x_6) \mapsto (-x_0,x_2,x_1,x_4,x_3,x_6,x_5)$, \\ & $(x_0,\ldots,x_6) \mapsto (ix_0, -\zeta_{8}^3 x_2,-\zeta_{8}x_1,x_6,-\zeta_{8}^3x_5,-\zeta_{8} x_4,x_3)$\end{tabular}\\ \mbox{} \\ Genus 7, Locus 4: Group $(64,38)$, signature (2,4,16), hyperelliptic \\ \begin{tabular}{l} $y^2=x^{16}-1$ \end{tabular} \\ \mbox{}\\ Genus 7, Locus 5: Group $(56,4)$, signature (2,4,28), hyperelliptic \\ \begin{tabular}{l} $y^2=x^{15}-x$ \end{tabular} \\ \mbox{}\\ Genus 7, Locus 6: Group $(54,6)$, signature (2,6,9) \\ \begin{tabular}{ll}Ideal: & $x_1 x_6 + x_2 x_4 + x_3 x_5,$ \\ & $x_0^2-x_1 x_6 +\zeta_6 x_2 x_4 -\zeta_3 x_3 x_5,$ \\ & $x_1 x_4 + \zeta_3 x_2 x_5 - \zeta_6 x_3 x_6,$ \\ & $x_1 x_5 + \zeta_3 x_2 x_6 - \zeta_6 x_3 x_4,$ \\ & $x_0 x_1-\zeta_6 x_5^2-x_4 x_6,$ \\ & $x_0 x_2+x_6^2-\zeta_3 x_4 x_5,$ \\ & $x_0 x_3+\zeta_3 x_4^2+\zeta_6 x_5 x_6,$ \\ & $x_0 x_4-x_1^2+\zeta_3 x_2 x_3,$ \\ & $x_0 x_5-\zeta_3 x_2^2-\zeta_6 x_1 x_3,$ \\ & $x_0 x_6+\zeta_6 x_3^2+x_1 x_2$\end{tabular}\\ \begin{tabular}{ll}Maps: & $(x_0,\ldots,x_6) \mapsto (-x_0,\zeta_9^5 x_6,\zeta_9^8 x_4,\zeta_9^2 x_5,\zeta_9 x_2,\zeta_9^7 x_3,\zeta_9^4 x_1 )$, \\ & $(x_0,\ldots,x_6) \mapsto (\zeta_6 x_0, \zeta_3^2 x_4, \zeta_3^2 x_5, \zeta_3^2 x_6, \zeta_3 x_3, \zeta_3 x_1, \zeta_3 x_2)$\end{tabular}\\ \mbox{} \\ Genus 7, Locus 7: Group $(54,6)$, signature (2,6,9) \\ \begin{tabular}{l}Complex conjugate of the previous curve \end{tabular}\\ \mbox{} \\ Genus 7, Locus 8: Group $(54,3)$, signature (2,6,9) cyclic trigonal\\ \begin{tabular}{l} Trigonal equation: $y^3 = x^9-1$\end{tabular}\\ \begin{tabular}{ll} Ideal: & $2\times 2$ minors of $\left[ \begin{array}{rrrrr} x_0 & x_2 & x_3 & x_4 & x_5 \\ x_1 & x_3 & x_4 & x_5 & x_6 \end{array} \right]$, and \\ & $ x_0^3-x_6^2+x_2^3$, \\ & $x_0^2 x_1-x_6^2 x_4+x_2^2 x_3,$ \\ & $x_0 x_1^2-x_6^2 x_5+x_2^2 x_4,$ \\ & $x_1^3-x_6^3+x_2^2 x_5$ \end{tabular}\\ \begin{tabular}{ll} Maps: & $(x_0,\ldots,x_6) \mapsto (x_1,x_0,-x_6,-x_5,-x_4,-x_3,-x_2)$, \\ & $(x_0,\ldots,x_6) \mapsto (\zeta_9 x_1, \zeta_9^2 x_0,-\zeta_9 x_6, -\zeta_9^2 x_5, -\zeta_9^3 x_4, -\zeta_9^4 x_3, -\zeta_9^5 x_2 )$\end{tabular}\\ \setlength{\textheight}{9.25in} \newpage \noindent Genus 7, Locus 9: Group $(48,32)$, signature (3,4,6). Has $g^{2}_{6}$ \\ \begin{tabular}{l}Ideal: \\ $x_0^2+x_3 x_5 + \zeta_6 x_3 x_6 -\zeta_3 x_4 x_6,$ \\ $\sqrt{3}i (x_1 x_3 - x_2 x_4+ x_1 x_5) - x_1 x_6 + x_2 x_5 - x_2 x_6,$ \\ $\sqrt{3}i (2 x_1 x_4 - x_2 x_5 + x_2 x_6)+ x_1 x_5 - x_1 x_6,$ \\ $\sqrt{3}i (2 x_2 x_3 + x_1 x_6 - x_2 x_5) + 3 x_1 x_5 + x_2 x_6,$ \\ $-3 (x_1^2+x_3 x_5 - x_3 x_6) + \sqrt{3}i (x_4 x_5 - x_4 x_6)+2 x_5^2 + 2 (\zeta_6 - 1) x_5 x_6 -2 \zeta_6 x_6^2,$ \\ $-3 (x_1 x_2 - x_4 x_5)+\sqrt{3}i (x_3 x_5 - x_3 x_6 + x_4 x_6-x_5^2) + 2 \zeta_6 x_5 x_6 - x_6^2,$ \\ $-3 (x_2^2-x_3 x_5 - x_4 x_6)+ \sqrt{3}i (x_3 x_6 -3 x_4 x_5) +2 (\zeta_6 + 1) x_5 x_6 + 2 (\zeta_6 - 1) x_6^2,$ \\ $-3 (2 x_4^2-x_3 x_5 + x_3 x_6) +\sqrt{3}i (- x_4 x_5 + x_4 x_6)+2 x_5^2 + 2 (\zeta_6 - 1) x_5 x_6 - 2 \zeta_6 x_6^2,$ \\ $-3 (-2 x_3 x_4 + x_4 x_5) + \sqrt{3}i (- x_3 x_5 + x_3 x_6 - x_4 x_6-x_5^2) + 2 \zeta_6 x_5 x_6 - x_6^2,$ \\ $-3 (2 x_3^2 + x_3 x_5 +x_4 x_6) +\sqrt{3}i (- x_3 x_6 + 3 x_4 x_5) +2 (\zeta_6 + 1) x_5 x_6 + 2 (\zeta_6 - 1) x_6^2$ \end{tabular}\\ \begin{tabular}{l}Maps: \\ $(x_0,\ldots,x_6) \mapsto (\zeta_3x_0,-\zeta_6x_2,-\zeta_3x_1-x_2,-x_3+\zeta_3x_4,\zeta_6x_3,\zeta_3 x_5,-\zeta_6 x_5+x_6)$, \\ $(x_0,\ldots,x_6) \mapsto (-x_0,-x_2,x_1,x_4,-x-3,\zeta_3x_5-\zeta_6x_6,-\zeta_6x_5-\zeta_3x_6)$\end{tabular}\\ \mbox{} \\ Genus 7, Locus 10: Group $(42,4)$, signature (2,6,21) cyclic trigonal\\ \begin{tabular}{l}Trigonal equation: $y^3 = x^8-x$\end{tabular}\\ \begin{tabular}{ll}Ideal: & $2\times 2$ minors of $\left[ \begin{array}{rrrrr} x_0 & x_2 & x_3 & x_4 & x_5 \\ x_1 & x_3 & x_4 & x_5 & x_6 \end{array} \right]$, and \\ & $ x_0^3-x_6^2 x_2+x_2^2 x_3$ \\ & $x_0^2 x_1-x_6^2 x_3+x_2^2 x_4,$ \\ & $x_0 x_1^2-x_6^2 x_4+x_2^2 x_5,$ \\ & $x_1^3-x_6^2 x_5+x_2^2 x_6$ \end{tabular}\\ \begin{tabular}{l}Maps:\\ $(x_0,\ldots,x_6) \mapsto (\zeta_7 x_1,\zeta_7^6 x_0,-\zeta_7^4 x_6,-\zeta_7^2x_5,-x_4,-\zeta_7^5 x_3,-\zeta_7^3 x_2)$, \\ $(x_0,\ldots,x_6) \mapsto (\zeta_{21}^{-2} x_1, \zeta_{21}^{-5} x_0,-\zeta_{21}^{-1} x_6, -\zeta_{21}^{-4} x_5, (\zeta_3+1) x_4, -\zeta_{21}^{11} x_3, -\zeta_{21}^8 x_2 )$\end{tabular}\\ \mbox{} \\ Genus 7, Group 11: Group $(32,11)$, signature (4,4,8). Has $g^{2}_{6}$ \\ \begin{tabular}{ll} Ideal: & $x_3 x_5+x_4 x_6,$ \\ & $ x_0^2+x_1 x_5+i x_2 x_6,$ \\ & $x_1 x_4+i x_2 x_3+x_5 x_6,$ \\ & $x_1 x_2+x_3 x_4,$ \\ & $x_1 x_6+\zeta_8^3 x_4 x_5,$ \\ & $x_2 x_5+\zeta_8 x_3 x_6,$ \\ & $x_1^2-i x_3^2-\zeta_8^3 x_5^2,$ \\ & $x_2^2+i x_4^2+\zeta_8^3 x_6^2,$ \\ & $-i x_2 x_4+\zeta_8^3 x_3^2,$ \\ & $x_1 x_3-\zeta_8^3 x_4^2$ \end{tabular}\\ \begin{tabular}{ll} Maps:& $(x_0,\ldots,x_6) \mapsto (-x_0,-x_2,x_1,-ix_4,-ix_3,ix_6,ix_5)$, \\ & $(x_0,\ldots,x_6) \mapsto (ix_0,x_1,ix_2,-x_3,-ix_4,-x_5,ix_6)$\end{tabular}\\ \mbox{} \\ Genus 7, Locus 12: Group $(32,10)$, signature (4,4,8) \\ \begin{tabular}{ll}Ideal: & $ x_1 x_6+\zeta_{16}^6 x_2 x_5+x_3 x_4, $ \\ & $x_1 x_2+x_5 x_6,$ \\ & $x_0^2+x_1 x_6-\zeta_{16}^6 x_2 x_5,$ \\ & $x_3 x_6-\zeta_{16}^4 x_4 x_5,$ \\ & $x_1^2-\zeta_{16}^7 x_4^2-\zeta_{16}^6 x_5^2,$ \\ & $x_2^2+\zeta_{16}^3 x_3^2-\zeta_{16}^{10} x_6^2,$ \\ & $-\zeta_{16}^2 x_2 x_6+(\zeta_{16}^{16}+\zeta_{16}^8) x_4^2-\zeta_{16}^7 x_5^2,$ \\ & $x_1 x_5+(-\zeta_{16}^{12}-\zeta_{16}^4) x_3^2-\zeta_{16}^{11} x_6^2,$ \\ & $x_1 x_3+\zeta_{16}^7 x_4 x_6,$ \\ & $x_2 x_4+\zeta_{16} x_3 x_5$ \end{tabular}\\ \begin{tabular}{ll} Maps: &$(x_0,\ldots,x_6) \mapsto (-x_0,-x_2,x_1,-\zeta_{16}^2 x_4,-\zeta_{16}^6 x_3, -\zeta_{16}^6 x_6, -\zeta_{16}^2 x_5)$, \\ & $(x_0,\ldots,x_6) \mapsto (i x_0, -\zeta_{16}^6 x_2,-\zeta_{16}^2 x_1, -i x_4, i x_3, -i x_6, -i x_5)$\end{tabular}\\ Genus 7, Locus 13: Group $(30,4)$, signature (2,15,30), hyperelliptic, $y^2=x^{15}-1$ \section*{References} \bibliographystyle{amsplain} \begin{biblist} \bibselect{autcfinal} \end{biblist} \end{document}
685c5c179c71703dff362282218196092868cfde
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Preliminaries and Problem Formulation} \label{sec:prelims} \label{sec2} \paragraph{Bipartite Graphs} A \textbf{bipartite graph} $G=(\mathcal{M},\mathcal{B},E)$ is a graph such that the vertices $\mathcal{M}\cup \mathcal{B}$ can be divided into two disjoint subsets, $\mathcal{M}$ and $\mathcal{B}$, and there are no edges connecting vertices in the same subset, $E\subseteq \mathcal{M}\times \mathcal{B}$. Such a graph is \textbf{balanced} if $|\mathcal{M}| = |\mathcal{B}|$, i.e., if the two subsets have the same cardinality. A \textbf{perfect matching} in a balanced bipartite graph $G=(\mathcal{M},\mathcal{B},E)$ is a subset of edges $E_{pm} \subseteq E$ such that every vertex in $G$ is incident upon exactly one edge of the matching. We denote the neighbors of a set of vertices $S$ by $N(S)$, where $N(S)\triangleq \{j\in \mathcal{B}: \exists i \in S~\text{s.t.}~(i,j) \in E\}$ when $ S \subseteq \mathcal{M}$ and $N(S) \triangleq \{j\in \mathcal{M}: \exists i \in S~\text{s.t.}~(j,i) \in E\}$ when $ S \subseteq \mathcal{B}$. \begin{dfn} A set $S \subseteq \mathcal{M}$ or $S \subseteq \mathcal{B}$ in a bipartite graph $G=(\mathcal{M},\mathcal{B},E)$ is called a \textbf{constricted set} if $|S|>|N(S)|$. More precisely, we call $S$ a \textbf{constricted good set} if $S \subset \mathcal{M}$ or a \textbf{constricted buyer set} if $S \subset \mathcal{B}$. \end{dfn} \begin{thm*}[\textbf{Hall's marriage theorem}~\cite{hall}] For a balanced bipartite graph $G=(\mathcal{M},\mathcal{B},E)$, $G$ contains no perfect matching if and only if $G$ contains a constricted set. \end{thm*} \paragraph{Matching Market} We consider a matching market with a set $\mathcal{B}$ of buyers, and a set $\mathcal{M}$ heterogeneous merchandise with exactly one copy of each type of good. Each buyer $i\in \mathcal{B}$ has a non-negative valuation $v_{ij}\geq 0$ for good $j\in M$, and desires at most one good (e.g. they are unit-demand buyers). We denote the $|\mathcal{B}|\times |\mathcal{M}|$ valuation matrix by $\mathbf{V}$. Our assumption that $|\mathcal{B}|=|\mathcal{M}|=m$ is without loss of generality because we can always add dummy goods or dummy buyers for balance. Given a price vector $\mathbf{P}=[P_1 \, P_2 \, ...\, P_m]$, we assume a quasi-linear utilities for the buyers, i.e., buyer $i$ receiving good $j$ has utility $U_{i,j}=v_{i,j}-P_j$. Since each buyer is unit-demand, we define $U^*_i$ be the maximum (non-negative) payoff of buyer $i \in \mathcal{B}$, i.e., $U^*_i=\max \big\{0,\underset{j\in \mathcal{M}}{\max}~v_{i,j}-P_j\big\}$. Since buyers can opt out of the market and obtain zero, we insist on the payoff being non-negative. \begin{dfn} \label{def:pgset} Under a price vector $\mathbf{P}$, the \textbf{preferred-good set} of buyer $i\in \mathcal{B}$ is a set of goods $L_i\subseteq \mathcal{M}$ such that getting each good in $L_i$ maximizes buyer $i$'s payoff, $L_i=\{j\in \mathcal{M}|v_{i,j}-P_j=U^*_i\}$. \end{dfn} \noindent Note that the preferred goods set of a buyer is empty if its payoff for all the goods is negative. \begin{dfn} By connecting each buyer with its preferred goods and recalling the assumption of $|M|=|B|$, we can construct a balanced bipartite graph which we call the \textbf{preference graph}, i.e., $G_{\mathrm{pref}}=(\mathcal{M},\mathcal{B},E_{\mathrm{pref}})$ where $E_{\mathrm{pref}}=\{(j,i): i \in \mathcal{B}$ and $j\in L_i\}$. \end{dfn} \noindent To avoid any confusion, we always place goods on the left-hand side and buyers on the right-hand side of the preference graphs. \begin{dfn} The set of goods $M$ is over-demanded in $G_{\mathrm{pref}}$ if it's a union of preferred-good sets of a set of buyers $B$, where $|B|>|M|$. Given a particular preference graph $G_{\mathrm{pref}}$ that doesn't contain a perfect matching and a constricted buyer set $B$, an over-demanded set of goods coincides with the neighbor set of $B$, i.e., $N(B)$, where the neighbor set is determined in $G_{\mathrm{pref}}$. Similarly, the under-demanded set of goods $M$ coincides with a constricted good set. \end{dfn} Given a specific price vector, if the preference graph $G_{\mathrm{pref}}$ contains a perfect matching $E_{pm}\subseteq E_{\mathrm{pref}}$, then we can allocate to each buyer exactly one of the goods it prefers and also sell all the goods. A price vector that leads to a perfect matching in the realized preference graph is called a \textbf{market-clearing price (MCP)} (also called a Walrasian price). Given any valuation $V$, it is well known that the set of MCPs is non-empty and bounded \cite{core}. Boundedness is obvious from the finiteness of the valuations. Non-emptiness is established either using the characterization in \cite{core}, using a constructive ascending price algorithm \cite{constructMCP} that starts from all the prices being $0$, or by using the VCG mechanism price (see Chap 15 in \cite{textbook}). Furthermore, the set of MCPs has a lattice structure \cite{core}, so that given any two different MCP vectors, the element-wise maximum of the vectors and the element-wise minimum of the vectors are also MCPs. This guarantees the existence of the maximum and the minimum MCPs. \paragraph{Complexity of Algorithms} An algorithm runs in \textbf{strongly polynomial time} if the number of operations and the space used are bounded by a polynomial in the number of input parameters, i.e., $O(\text{polynomial of }|M|)$, but both do not depend on the size of the parameters (assuming unit time for basic mathematical operations). If this does not hold but the number of operations is still bounded by a polynomial in the number of input parameters where the coefficients depend on the size of the parameters, then we say that the algorithm runs in \textbf{weakly polynomial time}. \section{Related Work} \label{sec:related} While sponsored search auctions are a recent motivation to study matching markets, there is a vast history to the problem. The term ``matching market" can be traced back to the seminal paper ``College Admissions and the Stability of Marriage" work by Gale and Shapley \cite{gale1962}. In matching markets, the necessary and sufficient condition for the existence of an efficient matching using Hall's marriage theorem~\cite{hall} has been proved \cite{gale1960theory} and a widely used mathematical model of two-sided matching markets was introduced in ``The Assignment Game I: The Core" \cite{core} by Shapley and Shubik. In \cite{core}, the set of MCPs is further shown to be solutions of a linear programming (LP) problem and the lattice property is also established. Despite this the study of this problem goes back at least to the well-known strongly polynomial-time Hungarian algorithm \cite{kuhn1955} for finding the maximum weight matching in a weighted bipartite graph, which in fact can also be used to find the minimum MCP. Furthermore, several auction algorithms enhancing the run-time efficiency in markets with specific properties have been presented in \cite{bertsekas1992forward,bertsekas1993reverse}. Leonard \cite{Truth-LP} considered mechanisms with sealed-bids and proved that charging the minimum competitive equilibrium price from bidders will result in an incentive compatible mechanism, and also that MCP coincides with the VCG price. Soon after, an ascending-price-based auction~\cite{krishna} algorithm was presented by Demange, Gale, and Sotomayor (DGS) in \cite{constructMCP}, which starts at the zero-price vector and then increases the posted price for any of the minimal over-demanded sets \cite{gale1960theory} of goods to obtain the minimum MCP. Thereafter, plenty of ascending-price-based auction mechanisms have been studied under different assumptions in \cite{bikhchandani2006, gul2000english, ausubel2004}. We pause here to remind the reader that the DGS ascending price algorithm is only known to be weakly polynomial-time. On the other hand, there has only been a limited study of descending-price auction algorithms to obtain the maximum MCP. Mishra and Parkes present a descending price auction called the Vickrey-Dutch auction to generate the VCG price in equilibrium \cite{mishramulti}. To aim for a higher revenue for sellers, Mishra and Garg generalized the Dutch auction to provide a descending-price-based approximation algorithm in \cite{mishragarg}. As mentioned in Section \ref{sec:intro}, Mishra and Garg's algorithm yields an approximation to the maximum MCP via a weak polynomial-time algorithm, and furthermore, there is no analysis of the strategic bidding in their work. We remark again that the sequential LP approach in \cite{Truth-LP} can be used to obtain the maximum MCP via a weakly polynomial-time algorithm. Finally, there is a body of literature that attempts to raise the revenue of sellers in equilibrium in related problems, such as sponsored search auctions and combinatorial auctions. In sponsored search markets, Ghosh and Sayedi considered a two-dimensional bid on advertisers' valuations according to exclusive and nonexclusive display \cite{ghosh2010expressive}, then run a GSP-like auction to determine the allocation that maximizes the search engine's revenue. With this small variation, efficiency does not hold for GSP, and hence the revenue will be different from the VCG mechanism. Additionally, in combinatorial auctions, it is well-known that designing a revenue maximizing auction mechanism is still an open problem. To achieve a higher expected revenue of sellers, Likhodedov and Sandholm presented a class of auctions, called virtual valuations combinatorial auctions \cite{likhodedov2004methods}, to maximize the sum of a pre-determined weighted valuation and an evaluation function of allocation rather than maximizing the total valuations as in the VCG mechanism to get a higher revenue. \section{Design of Descending Price Algorithm} \label{sec3} The problem considered in our work, as mentioned earlier, is to find the generalization of the Dutch auction\footnote{Despite a similar sounding name, what we seek to implement is completely different from the generalized first price (GFP) auction as in \cite{hoy2013dynamic, Expressiveness}.} to matching markets. Specifically, we seek a descending price auction that always converges to the maximum MCP. Like the DGS mechanism, our mechanism will choose a particular constricted good set to ensure the convergence. Specifically, we will define a dual to the ``minimal over-demanded set" which we call the \textit{maximally skewed set}. Unlike minimal over-demanded sets, the maximally skewed set is unique, and an example of failure to achieve the maximum MCP if this set is not chosen will be discussed in Section \ref{sec6.1}. \subsection{Framework of Descending Price Algorithms}~\label{sec:framework} We design a descending auction, which is the analogue of the ascending auction, in a straightforward framework. We start from a high enough initial price, iteratively pick a constricted good set to decrement prices, and terminate the algorithm when there exists a perfect matching. Clearly, this framework does not guarantee the termination in finite time, let alone strongly polynomial time. In order to make the algorithm run in strongly polynomial time, we will exploit the combinatorial structure of the preference graph, and make the evolution of the preference graph in the run of the algorithm be such that any specific bipartite graph appears at most once. To achieve this goal, we will specify a particular initial configuration, and a particular price reduction to be carried out in each step of the algorithm. \paragraph{Initial Price Choice:} A perfect matching requires every good be preferred by some buyers. Then a reasonable starting point should guarantee that the preferred-buyer set of every good is non-empty, otherwise it cannot be an MCP for any valuation matrix. Thus, the natural candidate for the initial price is $P_j =\max_{i \in \mathcal{M}}v_{i,j}$ for good $j$, which is (element-wise) greater than or equal to any MCP but ensures that every good is preferred by at least one buyer from the very outset. \paragraph{Price Reduction:} In computing the price reduction for a given constricted set $S$, we need to reduce the price by a large enough amount to trigger a change in the preference graph (otherwise we still have a constricting set and the same set of goods can be chosen again), but we should also avoid reducing the price of any good below its price in the maximum MCP. In other words, we want to find the minimum value to compensate the buyers not in the $N(S)$ to make at least one buyer indifferent between one of the goods in $S$ and the good(s) she prefers initially; the buyer in question may have an empty preferred goods set, in which case it is sufficient to ensure that one of the goods in $S$ has a non-negative utility with the price reduction. Lemma \ref{lemA} formally states the price reduction to be used in the proposed family of descending price algorithms. \begin{lem} \label{lemA} Given a constricted good set $S$ and a price vector $\mathbf{P}$, the minimum price reduction of all goods in this set $S$ guaranteeing to add at least a new buyer to the set $N(S)$ is \begin{equation} \underset{i \in B \setminus N(S),l\in S}{\min} \{\underset{k \in M \setminus S}{\max}(v_{i,k}-P_k)- (v_{i,l}-P_l) \}. \end{equation} \end{lem} \begin{figure}[t] \centering \includegraphics[width=0.8\textwidth,height=1in]{./Figures/skewness.pdf} \vspace{-12pt}\caption{Criteria for choosing constricted good sets} \label{fig:skewness} \end{figure} \subsection{Choice of Constricted Good Sets and The Skewness Function} \label{skewfunc} Since different choices of constricted good sets could generate different MCPs when the algorithm terminates, pinpointing the right constricted good sets iteratively has a pivotal role when designing the algorithm for finding the maximum MCP\footnote{Appendix \ref{sec6.1} provides an example where a different choice fails to obtain the maximum MCP.}. Before detailing the selection criterion, we use Figure~\ref{fig:skewness} to provide some quick intuition. On one hand, we prefer choosing the constricted good set in LL to LR because we want to choose the largest good set given the same set of neighbors (buyers). On the other hand, we prefer RL to RR because we do not include any subgraph (set of good-buyer pairs) that already has a perfect matching. With this intuition in mind, we present the following formal criteria for choosing constricted-good sets: \begin{enumerate} \item Pick the constricted goods sets $S$ with the largest difference $|S|-|N(S)|$. \item If there are multiple sets with the same $|S|-|N(S)|$, choose the one with the smallest size. \end{enumerate} The first criterion ensures that at each step the algorithm (simultaneously) reduces the price of the most critical set of goods. The proof that our algorithm returns the maximum MCP will not hold without this property. As an added bonus it also positively impacts the speed of convergence. The second criterion excludes any subset of goods $S'\subset S$ which is already perfectly matched to a subset of buyers, i.e., $|N(S')\setminus N(S)|\geq |S'|$. Jointly the criteria imply that we are searching for the most ``skewed" constricted good set in the preference graph. To formulate this mathematically, we define a function to measure the skewness of a set. \begin{dfn} \label{dfn:skew} The skewness of a set of goods $S$ is defined by function $f:2^\mathcal{M}\setminus {\emptyset} \mapsto R$ with $f(S)=|S|-|N(S)|+\frac{1}{|S|}$ for all $S\subseteq \mathcal{M}$ with $S\neq \emptyset$, where $2^\mathcal{M}$ is the power set of $\mathcal{M}$. \end{dfn} With this skewness function, the criteria described earlier are equivalent to choosing the constricted goods set with the maximal skewness. To formally make this statement we need to show two properties. The first one is the uniqueness of the maximally skewed set when the preference graph has no perfect matching; and the second one is that the maximally skewed set is a constricted goods set when the preference graph has no perfect matching. Lemma~\ref{lem2} proves these. \begin{lem} \label{lem2} Given a bipartite graph with no perfect matching, the maximally skewed set is unique and coincides with the constricted goods set with the maximal skewness. \end{lem} With Lemma \ref{lem2} in place, it easily follows that the two rules we imposed are equivalent to finding the maximally skewed set at every iteration (as we already know that a perfect matching doesn't exist). With the proper initial price vector choice, specified price reduction per round, and the unique choice of the maximally skewed set, the complete algorithm is described in Algorithm \ref{alg:alg1}. Note that the DGS algorithm, which searches for over-demanded sets to increase the price, has a dual structure to our algorithm. Thus, it is not surprising that the minimally over-demanded sets of items in the DGS algorithm, denoted as DGS sets below, have a relationship with the skewness function $f(\cdot)$. They are ones that obtain the minimum positive value of the function $|N(S)|-|S|-\frac{1}{|N(S)|}$ when the algorithm starts with initial price 0. We highlight the fact that the DGS sets may not be unique as there can be multiple sets of goods that yield the same minimum positive value for the function $|N(S)|-|S|-\frac{1}{|N(S)|}$. In contrast to our algorithm, the lack of uniqueness in the DGS algorithm is not as critical because different choices of DGS sets lead to the same minimum MCP. Understanding this contrast better is for future work. \begin{algorithm}[h] \caption{Skewed-set Aided Descending Price Auction} \begin{algorithmic}[1] \Require A $|\mathcal{B}| \times |\mathcal{M}|$ valuation matrix $\mathbf{V}$. \Ensure MCP $\mathbf{P}$. \State Initialization, set the price of good $j$, $P_j=\max_{i \in \mathcal{B}} v_{i,j}$. \State Construct the preference graph. \While{There exists a constricted good set} \State Find the maximally skewed set $\mathbf{S}$. \State For all $j \in \mathbf{S}$, reduce $P_j$ by $\min_{i \in \mathcal{B} \setminus N(\mathbf{S}),l\in \mathbf{S}} \{\max_{k \in \mathcal{M} \setminus \mathbf{S}}(v_{i,k}-P_k)- (v_{i,l}-P_l) \}$. \State Construct the preference graph. \EndWhile \State Return $\mathbf{P}$. \end{algorithmic} \label{alg:alg1} \end{algorithm} \vspace{-12pt} \section{Price Attained, Convergence Rate, and Complexity}\label{correctness} First, we demonstrate that the proposed skewed-set aided descending price auction algorithm returns the maximum MCP. We achieve this by performing a check by adding a fictitious dummy good to the preference bipartite graph at termination. Second, we use the potential function to prove the finite time convergence of the algorithm. Finally, we analyze the complexity of the algorithm by presenting algorithms to find the maximally skewed set. \subsection{Attaining Maximum Market-Clearing Price} In advance of analyzing the relationship between the skew-aided algorithm and the maximum MCPs, we have to precisely characterize the extremal nature of the maximum MCP. Wearing an optimization hat and using the idea of feasible directions, one would expect that checking whether the MCP of any good can be increased or not is straightforward\footnote{There is a history of such variational characterizations in the stable matching literature \cite{immorlica2005marriage,hatfield2005matching} where agents are assumed to have ordinal preferences.}. However, this logic misses the underlying matching problem and the Marriage theorem. Additionally, since the skewed-set aided algorithm is built on the combinatorial structure of the problem, to bridge the maximum MCP to our algorithm requires a combinatorial characterization of the maximum MCP. The combinatorial characterization requires adding a fictitious dummy good to preserve the property. Hence, we have to provide the following definition before stating the variational and combinatorial characterizations of the maximum MCP in Theorem \ref{lem4.1}. \begin{dfn}Given a bipartite graph $G$, let $N_G^D(B)$ be the neighbor set of the buyer set $B$ after adding a dummy good---a good for which every buyer has value 0. If the graph $G$ is clear from the context, we will also simplify the notation further to $N^D(B)$ after adding a dummy good. \end{dfn} \begin{thm} \label{lem4.1} An MCP $P^*$ is the maximum if and only if for any subset of goods, increasing the price of all goods in the set will change the preference graph such that no perfect matching exists. Equivalently, by adding a dummy good, $P^*$ is the maximum if and only if any subset of buyers $B$ has a cardinality strictly less than the cardinality of the set of buyers' neighbors $N^D(B)$. \end{thm} For further clarification, any buyer who has zero surplus at the maximum MCP (by definition of the maximum there will exist at least one such buyer) will be indifferent between the matched good and the dummy good. Hence, for every buyer set $B$ containing a zero-surplus buyer, the dummy good $D$ will be in the neighbor set of this set, $D\in N^D(B)$. We also show a dual property to VCG prices of the maximum MCP in Section~\ref{sec:externality}. With Theorem \ref{lem4.1} in hand, we will now establish the correctness of the algorithm, assuming that it halts (in finite-time). Since the skew-aided algorithm continually changes the preference graph, it is necessary to label the bipartite graph in each round of our algorithm before starting any analysis. Let $G_0$ be the initial bipartite graph, in the running of our algorithm, we obtain a bipartite graph $G_t$ at $t^{th}$ round. Then, we'll need to check whether the terminal condition holds. To avoid cumbersome notation, we will use $N_t^D(B)$ instead of $N_{G_t}^D(B)$. With Theorem \ref{lem4.1}, the proof of Theorem \ref{thm2} followings by checking that the preference graphs at termination coincides has the combinatorial characterization outlined above. \begin{thm} \label{thm2} The skewed-set aided descending-price algorithm always returns the maximum MCP. \end{thm} \subsection{Preference Graphs Converge Quadratically in the Number of Goods} \label{sec4.2} The Algorithm \ref{alg:alg1} changes the preference graph in each round to obtain the bipartite graph with combinatorial structure of MCP at termination. We will now show that the algorithm terminates in at most $m^2$ rounds. Given a specific preference graph $G$, we can define the skewness of the graph $W(G)$ to equal the skewness of the maximally skewed set. Therefore, by defining a sequence $W(G_t)=\max_{S \in \mathcal{M}, S \neq \emptyset} f_t(S)$, where $G_t$ is the preference graph obtained at the $t^{\mathrm{th}}$ iteration of Algorithm \ref{alg:alg1}, we show the convergence of the algorithm in finite rounds by proving that $W(G_t)$ strictly decreases with the decrease at least some positive constant. Thus, $W(\cdot)$ is a potential function that will be shown to strictly decrease in every iteration of the algorithm in the proof of Lemma~\ref{lem:conv}. \begin{lem} \label{lem:conv} For any unit demand matching market with $m>1$ the sequence $\{W(G_t)\}_{t\geq 0}$ of the skewness value of the maximally skewed set in each round of Algorithm \ref{alg:alg1} is strictly decreasing with minimum decrement $\tfrac{1}{m^2-m}$. \end{lem} Given the minimum decrement in Lemma \ref{lem:conv}, it is straightforward that the preference graphs converge to the bipartite graph with combinatorial structure of MCP in time upper bounded by $m^3$ because $W(G)<m$. However, as there are only $m^2$ positive distinct feasible values of $W(G)$ \footnote{Since there are only $m$ possible values of $|S|-|N(S)|$ and $m$ possible values of $\frac{1}{|S|}$.}, we are ensured convergence in time at most $m^2$. \subsection{Complexity of the Algorithm} Based on the results thus far determining the complexity of our algorithm depends only on the run-time of finding the maximally skewed set. We now discuss two approaches for this. \subsubsection{Algorithm design in search of the maximally skewed set} Given the uniqueness, we can always perform a brute-force search to get the maximally skewed set. Since there are $2^m-1$ non-empty subsets of $\mathcal{M}$, the complexity is $O(2^m)$, which doesn't meet our goal. We will exploit the combinatorial structure of the preference graph to scale down the complexity of finding the maximally skewed set. For this we design a graph coloring algorithm to color the preference graphs in Algorithm \ref{alg:alg3}. \begin{dfn} A colored preference graph $G(\mathcal{M},\mathcal{B}, E)$ is an undirected graph that colors each vertex (goods and buyers) in three colors either red, green, or blue; and each edge is colored red or blue. Denote $X_c, X=\{\mathcal{M},\mathcal{B}\}, c=\{r,g,b\}$ to be the set of goods/buyers colored red, green or blue. $E^{gb}_{rb}$ denotes edges connecting goods in $\mathcal{M}_g \cup \mathcal{M}_b$ and buyers in $\mathcal{B}_r \cup \mathcal{B}_b$. \end{dfn} In any colored preference graphs, we want red edges to represent edges connecting matched pairs of good-buyer in a maximum matching, and blue edges to represent the rest of the edges. Hence, each vertex has at most one red edge. Additionally, we want the set of red goods $\mathcal{M}_r$ to represent the set of goods not in the maximally skewed set, the set of blue goods $\mathcal{M}_b$ to be goods in the maximally skewed set but ones that do not have matched pairs to buyers in this maximum matching (because of the nature of constricted good set), and the set of green goods $\mathcal{M}_g$ are the rest of the goods. On the buyer side, the buyers that are neighbors of the maximally skewed set should be colored green, the buyers that are not the neighbors of the maximally skewed set but have a matched good should be colored red, and the rest of the buyers should be colored blue. Given the object we seek, we now present an algorithm to color vertices/edges properly in strongly polynomial-time complexity. The steps will include an initial coloring and followed by an update of the preference graph. Before detailing the initial coloring, we define various depth-first search and breadth-first search procedures relevant to the algorithm. \begin{dfn} A rb-DFS in $G(\mathcal{M},\mathcal{B},E)$ is a depth-first search (DFS) only using red edges from $\mathcal{M}$ to $\mathcal{B}$ and only using blue edges from $\mathcal{B}$ to $\mathcal{M}$. Similarly, a br-BFS in $G(\mathcal{M},\mathcal{B},E)$ is a breadth-first search (BFS) only using blue edges from $\mathcal{M}$ to $\mathcal{B}$ and only using red edges from $\mathcal{B}$ to $\mathcal{M}$. The set of nodes obtained at the end of the procedure will be called the reachable set $Rch(\cdot)$. \end{dfn} \paragraph{Initial Coloring} First, we find a maximum matching using the Hopcroft-Karp algorithm and color edges linked matched pairs red, and other edges blue. After that, we start from the set of good without matched buyer in this maximum matching, color them blue, run the br-BFS algorithm starting from the set of blue goods. When the br-BFS algorithm terminates, color the set of reachable goods $Rch(\mathcal{M}_b)\cap \mathcal{M}$ with matched buyers green, color the rest set of goods red. Then, color the matched buyers of red goods red, color the buyers in the $Rch(\mathcal{M}_b)\cap \mathcal{B}$ green (they are the neighbors of the most skewed set), and color the rest of buyer blue. Finally, the following lemma states that we get the maximally skewed set from the initial coloring. \begin{lem} \label{lem:colorMSS} After the initial coloring, $\{\mathcal{M}_g \cup \mathcal{M}_b\}$ is the maximally skewed set. \end{lem} Given that the Hopcroft-Karp algorithm has complexity $O(m^{2.5})$ and the br-BFS has complexity upper-bounded by $O(m^2)$, we learn the initial coloring has the complexity $O(m^{2.5})$. Note that the initial coloring does not rely on the initial price, our first algorithm, say \textbf{\textit{initial coloring based decreasing price auction}}, will use this in every iteration to get the maximally skewed set. This algorithm has complexity $O(m^2\times m^{2.5})=O(m^{4.5})$, which is already strongly polynomial. \begin{lem} \label{lem:alg2} Given a bipartite graph with no perfect matching, initial coloring returns the maximally skewed set of this bipartite graph in a strongly polynomial run time of $O(m^{2.5})$. \end{lem} \paragraph{Update the preference graph} We further scale down the complexity of the \textbf{\textit{initial coloring based decreasing price auction}} algorithm by exploiting and updating the colored preference graph colored in previous round without completely coloring the whole graph. This is detailed in Algorithm \ref{alg:alg3}. Since the procedure in Algorithm \ref{alg:alg3} is elaborate, we will highlight some facts of perfect matching and give the sketch of how we use them in the Algorithm \ref{alg:alg3}. First, we know that if there is a perfect matching, no vertex should be colored blue otherwise we fail to get a maximum matching. Furthermore, the following Lemma \ref{lem:color} states that we will never need to change a vertex from red or green to blue. \begin{lem} \label{lem:color} If we have to change the color of a vertex from red or green to blue based on interpretation given to the colors, one of the following is true: a) The coloring of previous preference graph is incorrect; b) The price reduction is not optimal; c) One of the updates from the previous preference graph to current graph is wrong. \end{lem} Given the property of a proper coloring stated in Lemma \ref{lem:color}, we know that the set of blue buyers in round $t$, denoted as $\mathcal{B}_b^t$, is a decreasing set. Therefore, the key idea of designing Algorithm \ref{alg:alg3} is to restrict operations irrelevant to reducing the set of blue buyers to some constant number of $O(m^2)$ operations, and to allow the complexity of operations reducing the set of blue buyers to be upper-bounded by $O(m^3)$. To achieve this, we observe that the set of buyers at round $t$ that are willing to get goods in previous maximally skewed set, denoted by $A_t$, only contains red and blue buyers, i.e., $A_t\subseteq \mathcal{B}_r^t \cup\mathcal{B}_b^t$. Then, we know that if $A_t \subseteq \mathcal{B}_r$ and we cannot reach any blue buyers from $A_t$ without passing green or blue goods, then the current matching is maximized, and all we need to do for updating colors is to run rb-BFS from the set of $\mathcal{M}_b$ with complexity $O(m^2)$ as we did in initial coloring. In other cases, we need to find the new maximum matching with number of matched pairs increased by at least 1, which can be achieved by at least $O(m^{2.5})$ using the Hopcroft-Karp algorithm, and at least one blue buyer will be recolored. Since the maximum number of matched pairs is $m$, the complexity of the whole algorithm attaining the maximum MCP will be upper-bounded by the \{(complexity of updating process not recoloring blue buyers)$+$(complexity of computing price reduction)\}$\times$ (convergence rate of the preference graph)$+$(complexity of updating process increasing maximum matched pairs)$\times$ (maximum number of blue buyers)$=O((m^2+m^2)\times m^2+m^{2.5}\times m)=O(m^4)$. \begin{thm} \label{thm1} The skewed-aided descending price algorithm has a strongly polynomial run time of $O(m^4)$ by using Algorithm \ref{alg:alg3} to search the maximally skewed set. \end{thm} \section{Introduction} \label{sec:intro} Online Advertising is an over \$70 billion busines with double-digit growth in consecutive years over a period of many years. Since nearly all of the ads are sold via auction mechanisms, auction-based algorithm design, which focuses on the online advertising, has become an important class of mechanism design to study. Among all online advertising auctions, the sponsored search auction, also known as a keyword auction, is the one that most captures researchers' attention. In a typical sponsored search auction, the auctioneer has a set of web slots to sell and every advertiser has different valuations on different web slots. Problems in sponsored search auctions are usually modeled as problems in (cardinal-preference) matching markets, and prices are used to clear the market. The key assumption of sponsored search auctions is that every advertiser shares the identical ordinal preference on web slots. Under this assumption, the celebrated Vickrey-Clarke-Groves (VCG) mechanism \cite{VCG_V,VCG_C,VCG_G}, which makes truthful bidding by the advertisers as (weakly) dominant strategies but yields low revenue to the auctioneer, is adopted by some web giants such as Facebook\footnote{Facebook Ad Auction: See https://www.facebook.com/business/help/163066663757985.}, and is also a robust option in scenarios where the revenue equivalence theorem~\cite{krishna} holds. However, as the revenue equivalence theorem does not hold in multi-good auctions~\cite{bayes}, auctioneers can look for greater expected revenue than the value obtained by VCG mechanism by using even different efficient and market-clearing auction mechanisms. The most popular auction among these mechanisms is the Generalized Second Price (GSP) auction employed by Google. Since the GSP is not incentive compatible, its equilibrium behavior needs to be analyzed \cite{bayes,caragiannis2015bounding,edelmanstrategic}, and there are some Bayesian Nash equilibria (BNE) \cite{krishna} that have greater expected revenue than the expected revenue of the VCG mechanism. It should also be noted that designing (revenue) optimal mechanisms \cite{myerson1981optimal} is intractable~\cite{cai2012optimal,daskalakis2014complexity} even in the context of matching markets when there is more than one good. Thus, the possibility of higher expected revenue coupled with the ease of implementing the GSP auction and the intractability of optimal mechanisms has lead to the popularity of the GSP mechanism. Unlike a decade ago where there were only statically-listed ads, websites now serve a variety of ads simultaneously, including sidebar images, pop-ups, embedded animations, product recommendations, etc. With this in mind, and the growing heterogeneity in both advertisers and consumers, it is clear that the ``shared ordinal preference'' assumption is untenable in the context of market design. Search engines and ad-serving platforms will be faced with a growing need to implement general unit-demand matching markets~\cite{gale1962}, and such market settings are the focus of our work. We refer to the prices that efficiently allocate the set of goods to the bidders according to their private valuations as a vector of \emph{market-clearing prices} (MCP). An ascending price auction algorithm that generalizes the English auction was presented by Demange, Gale and Sotomayor \cite{constructMCP}. This ascending price algorithm (DGS algorithm) obtains the element-wise minimum MCP, that coincides with the VCG price. DGS is thus incentive compatible yet obtains low expected revenue for the mechanism. Of course, simultaneously maximizing revenue and maintaining incentive compatibility is computationally intractable once we have more than one good for sale, but we should still hope to obtain better than the \emph{minimum} MCP within efficient mechanisms. In the present paper we design a family of mechanisms that seek to elicit the \emph{maximum} MCP from the participants without sacrificing computational efficiency. Here we focus explicitly on how we can efficiently compute the maximum MCP given some representation of the bidder utilities, and defer the general\footnote{Several illustrative instances of Bayesian equilibrium of strategic buyers are discussed in the full version \cite{fullver}. One particular instance explicitly demonstrates an example where our mechanism yields greater expected revenue when compared to the expected revenue of the VCG mechanism.} analysis of strategic behavior to future work. Critical to our paper would be answering whether there exists a strongly polynomial-time algorithm to obtain the maximum MCP exactly. Before discussing the literature for the maximum MCP, we discuss the state of the art for the minimum MCP. The intuitively appealing DGS ascending price algorithm that attains the minimum MCP is only weakly polynomial time: the potential function used to show convergence depends on the valuations; and its value decrements by at least a constant independent of the valuations in each step. In fact, it is the well-known strongly polynomial-time Hungarian algorithm \cite{kuhn1955} for finding the maximum weight matching in a weighted bipartite graph, that yields a strongly polynomial time algorithm for finding the minimum MCP, $O(m^4)$ in the original implementation that can then be reduced to $O(m^3)$~\cite{edmonds1972theoretical}. This will be the aspirational goal of this work. Using the method outlined in \cite{Truth-LP}, where one computes the solution of two linear programs, it is possible to determine the maximum MCP. Note, however, that this is at best a weakly polynomial-time algorithm, and is neither a combinatorial nor an auction algorithm. Given that the DGS ascending price mechanism returns the minimum MCP, it is also intuitive to study descending price mechanisms to obtain the maximum MCP, i.e., generalize the Dutch auction to multiple goods. The first attempt to obtain the maximum MCP through descending price auction is in the work by Mishra and Garg \cite{mishragarg}, where they provide a descending-price-based auction algorithm that yields an approximation algorithm. The algorithm doesn't require agents to bid their whole valuation but still yields a price-vector in weakly polynomial-time\footnote{Again, as the number of iterations depends on both $\epsilon$ and the input valuation matrix.} that is within $\epsilon$ in $l_\infty$ norm of the maximum MCP\footnote{Even though the final price may not be market clearing, decreasing it further by $\epsilon$ and then running the DGS algorithm, it is possible to obtain a market-clearing price that is within $2\epsilon$ of the maximum MCP.}. Therefore, in this work, one of main goals is to develop a strongly polynomial-time combinatorial/auction algorithm using descending prices for the \textit{exact} computation of the maximum MCP. Note that based on the analysis in \cite{position} choosing the maximum MCP in sponsored search markets has exactly the same complexity as the VCG, GSP and Generalized first-price (GFP) mechanisms: The web-slots are sold from best to the worst and in decreasing order of the bids of the agents, with the only difference being the price that's ascribed to each good. Once the computational problem is solved, setting the maximum MCP is a viable option for general unit-demand markets, and is an alternate efficient mechanism. \subsection{Our contribution} By judiciously exploiting the combinatorial structure in matching markets, we propose a strongly polynomial-time\footnote{See Section~\ref{sec:prelims} for a definition of strongly and weakly polynomial-time complexity.} descending price auction algorithm that obtains the maximum MCPs in time $O(m^4)$ with $m$ goods (and bidders). Critical to the algorithm is determining the set of under-demanded goods (to be defined precisely later on) for which we reduce the prices simultaneously in each step of the algorithm. This we accomplish by choosing the subset of goods that maximize a skewness function, which is obtained by proposed graph coloring algorithm a simple combinatorial algorithm to keep updating the bipartite graph and the collection under-demanded goods set. We start by discussing an intuitively appealing algorithm to solve this problem that uses the Hopcraft-Karp~\cite{hopcroft} algorithm and Breadth-First-Search (BFS). This procedure will only yield a complexity of $O(m^{4.5})$. We will then present a refinement that cleverly exploits past computations and the structure of the problem to reduce the complexity to $O(m^4)$. \section{Appendix-Details of Algorithms} \begin{algorithm}[H] \caption{Algorithm in search of the maximally skewed set by coloring preference graph} \begin{algorithmic}[1] \Require A colored preference graph, a set of buyers $A$ would be added to the neighbor of the previous maximally skewed set \Ensure An updated colored preference graph \If {Input preference graph is not colored} \State Run the algorithm for initial coloring \Else \State Update the colored preference graph by removing all edges connecting red goods and green buyers and adding corresponding blue edges connecting goods and buyers in $A$ \While {$\{A\cap\mathcal{B}_b\} \neq \emptyset$} \State Pick an element $a$ in $\{A\cap\mathcal{B}_b\}$ \If {$|N(a) \cap \mathcal{M}_b|>1$} \State Pick arbitrary $x \in \{N(a)\cap \mathcal{M}_b\}$, color $(a,x) \in E$ red and color $a,x$ green. \ElsIf {$|\{N(a) \cap \mathcal{M}_b\}|=1$} \State Let $x$ be the unique good in $\{N(a)\cap \mathcal{M}_b\}$, color $(a,x) \in E$ red and color $a,x$ green. \If{$\{N(a)\cap \mathcal{M}_g\}\neq \emptyset$} \State Run the rb-DFS starting from $x$ in $G(\mathcal{M}_g \cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{gb})$ to get a reachable set $Rch(x)$, then color vertice in $Rch(S)$ red if $Rch(x)\cap \mathcal{M}_b = \{x\}$ \Else \State Run the br-DFS starting from $a$ in $G(\mathcal{M}_g \cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{gb})$ to get a reachable set $Rch(a)$, then color vertice in $Rch(S)$ red if $Rch(a)\cap \mathcal{M}_b = \{x\}$ \EndIf \ElsIf {$\{N(a) \cap \mathcal{M}_g\} \neq \emptyset$} \State Run the rb-DFS starting from $a$ in $G(\mathcal{M}_g\cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{gb})$ till find the first $x \in \mathcal{M}_b$ \State Color $a,x$ green and switch the color of every edge used in a path from $a$ to $x$. \State Start a br-BFS from $\mathcal{M}_b$ in $G(\mathcal{M}_g\cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{g})$ to get $Rch(\mathcal{M}_b)$. \State Color every vertex in $\{\mathcal{M}_g \cup \mathcal{B}_g\}\setminus Rch(\mathcal{M}_b)$ red \EndIf \State Remove $a$ from $A$ \EndWhile \State Run the br-BFS starting from $\{\mathcal{M}_g\}$ in $G(\mathcal{M},\mathcal{B}, E)$ to get a reachable set $Rch(S^*)$ \If {$\{Rch(S^*) \cap \mathcal{B}_b\}=\emptyset$} \State Color all vertice in $Rch(S)\setminus \mathcal{M}_b$ green \Else \While{$\{Rch(S^*) \cap \mathcal{B}_b\}\neq\emptyset$} \State Pick $a \in Rch(S^*) \cap \mathcal{B}_b$ and run rb-DFS starting from $a$ to get $Rch(a)$ \If {$\exists x\in Rch(a), x \in \mathcal{M}_b$} \State Color $a,x$ green; switch the color of every edge used in a path from $a$ to $x$ \State Start a br-BFS from $\mathcal{M}_b$ in $G(\mathcal{M}_g\cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{g})$ to get $Rch(\mathcal{M}_b)$. \State Color every vertex in $\{\mathcal{M}_g \cup \mathcal{B}_g\}\setminus Rch(\mathcal{M}_b)$ red \EndIf \EndWhile \State Run the br-BFS starting from $\{\mathcal{M}_b\}$ in $G(\mathcal{M},\mathcal{B}_g, E)$ to get $Rch(\mathcal{M}_b)$ \State Color all vertice in $\{\mathcal{M}_g\cup \mathcal{B}_g\}\setminus Rch(\mathcal{M}_b)$ red \EndIf \EndIf \end{algorithmic} \label{alg:alg3} \end{algorithm} \begin{algorithm}[h] \caption{Algorithm for initial coloring} \begin{algorithmic}[1] \Require A preference graph with no perfect matching \Ensure Colored preference graph, the maximally skewed set \State Find a maximum matching by Hopcroft-Karp algorithm \State Color every edges connecting a matched pair red and all other edges blue, Color every vertex in a matched pair red and all other vertice blue \State Starting from $\mathcal{M}_b$, run the rb-BFS to get a reachable set $Rch(\mathcal{M}_b)$ \State Color $Rch(\mathcal{M}_b)\setminus \mathcal{M}_b$ green \end{algorithmic} \label{alg:alg2} \end{algorithm} \section{Appendix-Discussion} \subsection{Importance of Maximally Skewed Set} \label{sec6.1} To highlight the importance of choosing the maximally skewed set in our algorithm, we are starting to ask a natural question: if we run the algorithm twice but choose different constricted good sets at some iterations such that the preference graph produced in every round of these two executions are the same, will we get the same MCP vector? Unfortunately, the answer is no. The choice of the same initial price vector, the same price reduction rule, and the emergence of the same bipartite graph in every round are not enough to guarantee the same returned MCPs. A counterexample is provided in Fig. \ref{fig4}. \begin{figure}[h] \centering \includegraphics[width=0.4\textwidth]{./Figures/constricted.pdf} \caption{Counterexample of same bipartite graph but different set MCPs} \label{fig4} \end{figure} In Fig. \ref{fig4}, the bipartite graphs A-1, B-1 have the same preference graph. Though the chosen constricted good sets in A, B are different, they add the same buyer to the constricted graph. Therefore, the preference graphs in A-2, B-2 are still the same. However, the updated price vector of A-2, B-2 must be different. If A, B choose the same constricted goods set in every round, the returned sets of MCPs of A, B must be different. This example shows that just the bipartite graphs in every round cannot uniquely determine the MCP vector obtained at the termination of the algorithm. Then, we state a stronger claim in Lemma \ref{lem:multicon} that if an algorithm wants to get the maximum MCP, the law of choosing the constricted good set must pick the maximally skewed set at the round before termination. \begin{lem} \label{lem:multicon} Given a descending price algorithm with a specific law of choosing the constricted good set, and assuming this algorithm terminates at round $T_V$ when giving valuation matrix $V$m, if this law may not pick the most skewed set at round $T_V-1$, this algorithm not always return the maximum MCP. \end{lem} However, if a law of choosing the constricted good set pick the maximally skewed on in the last round before termination, the answer of whether it returns the maximum MCP will be case by case. Since Fig. \ref{fig4} provide an example of not achieving the maximum MCP, we now provide an example which still gets the maximum MCP in Fig.~\ref{fig5}. \begin{figure}[h] \centering \includegraphics[width=0.6\textwidth]{./Figures/counterexample.pdf} \caption{Choosing non-maximally skewed set in the middle but get maximum MCP} \label{fig5} \end{figure} In Fig.~\ref{fig5}, the algorithm in upper flow runs the algorithm we proposed and choose maximally skewed set in each round. We pick $\{A,B\}$ in the first iteration and then pick $\{A,B,C\}$ in the second iteration; and the algorithm terminates at $P=[1,1,2]$. The algorithm in lower flow does not choose the maximally skewed set in the first round, but the algorithm is forced to choose the maximally skewed set in the second round because it is the only constricted good set. It picked $\{A,B,C\}$ and lower the price to $[2,2,2]$; and pick $\{A,B\}$ to get a MCP at $[1,1,2]$. As shown in the figure, these two algorithm both return the maximal MCP. Therefore, the space of sets we can choose which guaranteeing to get the maximum MCP is still an open problem. \subsection{Interpretation of the Maximum MCP Using Buyers' Externalities,} Since the minimum posted price derived from DGS algorithm corresponds to the personalized VCG price given by the Clarke pivot rule \cite{VCG_C}, it is not surprising that there is an analogous structure between the posted price and the personalized price also for the maximum MCP. The Clarke pivot rule determines the VCG payment of buyer $i$ using the externality that a buyer imposes on the others by her/his presence, i.e., payment of buyer $i=$ (social welfare\footnote{Social welfare is defined as the sum of buyers' surplus plus the sum of goods' prices, which is the same as the sum of each buyer's value on her matched good.} of others if buyer $i$ were absent) - (social welfare of others when buyer $i$ is present). Using the combinatorial characterization of the maximum MCP, we obtain an exact analogue of the Clarke pivot rule when viewing the maximum MCP as the personalized price in Theorem~\ref{thm:externality}. In the theorem below we can use any perfect matching to determine the good matched to buyer $i$. \begin{thm} \label{thm:externality} Under the maximum MCP, the price that buyer $i$ pays is (social welfare of the current market adding a duplicate pair of buyer $i$ and its matched good) - (social welfare of the current market adding a duplicate buyer $i$). This price is the same as the decrease in social welfare that results by removing the good matched to buyer $i$. \end{thm} \section{Appendix-Preliminary Analysis of Strategic Buyers and\\ Bayesian-Nash Equilibrium} \label{sec5} As mentioned in the introduction, the proposed descending price algorithm is not incentive compatible so that we need to explore Bayesian Nash equilibrium (BNE) to predict buyers' strategic behaviors. From the messaging viewpoint we will assume that we have a direct mechanism where the buyers bid their valuation: a scalar in sponsored search markets, and a vector in the general matching market. Algorithm \ref{alg:alg1} is then applied as a black-box to the inputs to produce a price on each good and a perfect matching. Although the strategic behavior of buyers is not the main subject of this work, with expected revenue being an important concern, we provide an instance achieving higher expected revenue than VCG mechanism in a BNE with asymmetric distributions of buyers' valuations. Furthermore, we analyze the BNE in two simple cases with symmetric buyers and the associated bidding strategy of the buyers. \subsection{An Instance Achieving higher expected revenue than VCG}\label{sec:greatRev} Given the proposed algorithm, it is important to check if there exists a scenario that our algorithm can achieve a higher revenue than the VCG mechanism. It is well-known that the revenue equivalence theorem holds with an assumption that valuation of every player is drawn from a path-connected space in Chap. 9.6.3 of \cite{SSMbook}\footnote{Two other assumptions are (1) both mechanisms implement the same social welfare function; (2) for every player $i$, there exists a type that the expected payments are the same in both mechanisms}. Furthermore, there's a large body of literature that has discussed the failure of getting the VCG revenue under asymmetric distribution of buyers' valuations, e.g., see \cite{VCGfailure}. With this knowledge, we demonstrate a $3\times 3$ matching market where our mechanism achieves higher than VCG revenue, where the buyers have asymmetric distributions of their valuations. Consider three advertisers named Alice, Bob, and Carol, and three different types of ads called listing ads, sidebar ads, and pop-ups. The realized valuation is only known to the advertiser (equivalently buyer), but the distribution of an advertisers' valuation is known to other advertisers but not the auctioneer. In other words, the auctioneer can only calculate the price according to the bids submitted by the advertisers. The minimum increment of the submitted bids is $\epsilon$, which is a positive infinitesimal, and the valuation matrix of advertisers is displayed in Table \ref{tbl:1}. \begin{table}[h] \centering \caption{Valuation matrix of advertisers} \label{tbl:1} \begin{tabular}{|l|l|l|l|} \hline & Listing & Sidebar & Pop-ups \\ \hline Alice & w & 0 & 0 \\ \hline Bob & x & 0 & 0.5 \\ \hline Carol & y & z & 2 \\ \hline \end{tabular} \\ \begin{tabular}{|ll|} \hline Probability density function of $w,x,y,z$ &\\ \hline $f_w(w)=\begin{cases} \frac{2}{3} &~w\in [0,1) \\ \frac{1}{3} &~w\in (2,3] \\ 0 &~\text{o/w} \end{cases} $ & $f_x(x)=\begin{cases} \frac{2}{3} &~x\in [1.5,2.5] \\ \frac{1}{3} &~x\in (2.5,3.5] \\ 0 &~\text{o/w} \end{cases}$ \\ $f_y(y)=\begin{cases} 2 &~y\in [3.5,4] \\ 0 &~\text{o/w} \end{cases}$ &$f_z(z)=\begin{cases} 1 &~z\in [3,4] \\ 0 &~\text{o/w} \end{cases} $ \\ \hline \end{tabular} \end{table} Now, we give an asymmetric BNE\footnote{There need not be a unique equilibrium.} in this matching market and provide a detailed verification in the Appendix \ref{asymBNE}. First, consider Alice always bids 0 on sidebar ads and pop-ups, and bids $\max \{1-\epsilon, \frac{w}{2}\}$ on listing ads. The best response of Bob is to bid 0 on both sidebar ads and pop-ups, and to bid $\max \{1, \frac{x-0.5}{2}\}$ on listing ads for any realized $x$. Now, given Bob's bidding function as above, one of Alice's best responses is to follow her original bidding function. Last, consider the bidding function of Alice and Bob mentioned above, a best response of Carol is to bid 0 on listing ads and pop-ups, and to bid $\epsilon$ on sidebar ads regardless of the outcomes of $y$ and $z$. \footnote{The $\epsilon$ is designed to avoid complex tie-breaking rules.}, even if $y>z$. This is because Carol will never win listing ads for any bids less than $1$ as the probability $\text{Pr}\{y-z \geq 1\}=0$. Now, given Carol's bidding strategy, Alice and Bob will not change their bidding functions. Therefore, the strategy $\{\beta_{Alice}(w,0,0), \beta_{Bob}(x,0,0.5),\beta_{Carol}(y,z,2)\}=\{(\max \{1-\epsilon, \frac{w}{2}\},0,0), (\max \{1, \frac{2x-1}{4}\},0,0),(0,\epsilon,0)\}$ can be verified as an asymmetric BNE. With the asymmetric BNE in hand, we want to calculate the expected revenue of the auctioneers and compare it with the expected revenue of the VCG mechanism. Since Carol always wins the sidebar ads and both Alice and Bob bid 0 on that, Carol will pay $\epsilon$. Additionally, in the asymmetric BNE, Alice and Bob compete on the listing ads and both bid 0 on sidebar ads and pop-ups, resulting in the payment of listing ads to be the same as the payment in the first price auction. Next, let's calculate the expected revenue of auctioneers, which is given by \begin{eqnarray} &&\epsilon+\frac{4}{9}\int_0^1\int_{1.5}^{2.5} 1 dxdw+\frac{2}{9}\int_2^3\int_{1.5}^{2.5} \frac{w}{2} dxdw+\frac{2}{9}\int_1^2\int_{2.5}^{3.5} \frac{2x-1}{4} dxdw \nonumber \\&+&\frac{1}{9}\int_2^3\int_{2.5}^{w+0.5} \frac{w}{2} dxdw+\frac{1}{9}\int_{2.5}^{3.5}\int_2^{x-0.5} \frac{2x-1}{4} dwdx =\dfrac{31}{27}+\epsilon \end{eqnarray} The last step is to calculate the expected revenue of the VCG mechanism, which is given by \begin{eqnarray} &&\frac{2}{3}\int_0^1\int_{1.5}^{2.5} w dxdw+\frac{2}{9}\int_2^3\int_{1.5}^{2.5} (x-0.5) dxdw+\frac{1}{9}\int_{2.5}^{3.5}\int_2^{x-0.5} w dwdx \nonumber \\&+&\frac{1}{9}\int_2^3\int_{2.5}^{w+0.5} (x-0.5) dxdw =\dfrac{1}{3}+\dfrac{2}{9}\times \dfrac{3}{2}+\dfrac{1}{9}\times \dfrac{7}{6}++\dfrac{1}{9}\times \dfrac{7}{6}=\dfrac{25}{27} \end{eqnarray} Even if we set $\epsilon$ to 0, it is obvious that the expected revenue derived under our descending price auction algorithm is strictly greater than the expected revenue of the VCG mechanism. This shows that in some instances the proposed descending price algorithm is preferred to the (DGS) ascending price algorithm, even taking the strategic behavior of buyers into account. \subsection{Symmetric Bayesian Nash Equilibrium in $2\times 2$ Sponsored Search Market} Now we focus on the analysis of symmetric distributions of buyers. Next we will detail the analysis of two cases for a market with two goods and two buyers, one in the following paragraphs and another one in the next subsection. Since the primary application of our algorithm is in online advertising auctions, it is useful for us to analyze the strategic behavior under the conventional assumptions made in the sponsored search market setting. The sponsored search market assumes every buyer's (advertiser's) value on goods (web slots) can be determined by a product of the buyer's private weight and a common click-through rate. Now, we consider a $2 \times 2$ case in sponsored search market with settings detailed below. There are two web slots with click-through rates $c_1$ and $c_2$, with $c_1\geq c_2$, and two advertisers with private weights $w_1, w_2$. We assume that $w_j$ is an \emph{i.i.d.} non-negative random variable with PDF $f_{w_j}(\cdot)$, and the private value for getting web slots $j$ is $w_ic_j$. Each advertiser knows his/her true weight but only knows the distribution of another advertiser's value, and both know that it is a sponsored search market. Under our descending price auction algorithm, they each have to effectively place a one-dimensional bid, denoted by $b_i$, according to their bidding function $\beta_j(w_j,f_{w_{-j}}(\cdot))$, where ${-j}$ denotes the other advertiser(s). To simplify the analysis, we assume weights $w_1, w_2$ are uniformly distributed in $[0,1]$. Extensions to other symmetric distributions, asymmetric distribution/knowledge space and more web slots are all for future work. \subsubsection{Payment of advertisers} If $c_2=0$, the descending price algorithm terminates at the initial point. At this point, the advertiser $j$ wins the first web slots pays $c_1w_j$ and another advertiser pays $0$. This is, equivalent, to the single good case, which has been carefully studied before. When $c_2 \neq 0$, the descending price algorithm makes the advertiser indifferent between two slots if he/she bid truthfully. Therefore, if advertiser $i$ wins the first slot, the payment will be $(c_1-c_2)b_i+c_2b_{-i}$, if he/she gets the second slot, the payment will be $c_2b_i$. \subsubsection{Analysis of strategic behavior} Before deriving the bidding function, we show that the bidding function is a monotonic increasing function of the weight. Since the outcome of this auction is exactly the same as what the VCG mechanism provides, this monotonicity leads us to expect revenue equivalence to hold in this case. Exploring the generality of this result is for future work. \begin{lem} \label{lem:mono} The bidding function is monotonically increasing in the $2\times 2$ sponsored search market. \end{lem} \begin{proof} See Appendix \ref{pf:lem:mono}. \end{proof} This monotonicity property of bidding functions can be generalized to all sponsored search markets when every advertiser's private weight follows the same uniform distributions. \begin{cor} \label{cor:mono} In sponsored search markets with a symmetric uniform distribution of every advertiser's weight, the bidding function is monotonic and the allocation is always efficient. \end{cor} \begin{proof} See Appendix \ref{pf:cor:mono}. \end{proof} Without loss of generality, we now derive the bidding function of advertiser 1. Since the optimal bidding function is monotonic in the private value, the surplus function of advertiser 1 can be written using an integral form. Advertiser 1 wants to maximize the following surplus function\footnote{In this two advertisers case, for advertiser 1, $f_{w_{-1}}(\cdot)=f(w_2)(\cdot)$ and $\beta_{-1}(\cdot)=\beta_{2}(\cdot)$.}: \begin{eqnarray} \int_0^{\beta_2^{-1}(b_1)}(c_1w_1-[(c_1-c_2)b_1 +c_2\beta_2(x)])f_{w_2}(x)dx +c_2(w_1-b_1)[1-\int_0^{\beta_2^{-1}(b_1)}f_{w_2}(x)dx] \end{eqnarray} With detailed analysis presented in Appendix \ref{apdx:analysis}, the bidding function $\beta_1(w_1)$ is \begin{eqnarray} \begin{cases} e^{-w_1}+w_1-1 \qquad \qquad \qquad \qquad \qquad \qquad ~~~c_1=2c_2 \\ w_1-(2-w_1)\ln(1-0.5w_1) \qquad \qquad \qquad ~~~~c_1=1.5c_2 \\ \dfrac{c_2}{2c_1-3c_2}\Big[(c_1-c_2)w_1+\big(\dfrac{c_2}{c_2+(c_1-2c_2)w_1}\big)^{\tfrac{c_1-c_2}{c_1-2c_2}}-1\Big] \\ \nonumber \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \text{otherwise} \nonumber \end{cases} \end{eqnarray} Without loss of generality, we can let $c_1=1$ and $c_2 \in [0,1]$, Figure \ref{fig5} displays the optimal bidding bid corresponds to the buyer's private weight and the ratio of click-through rates. \begin{figure}[h] \centering \includegraphics[width=0.6\textwidth]{./Figures/weight-bid.jpg} \caption{Optimal bidding function of the symmetric BNE in $2 \times 2$ sponsored search markets} \label{fig5} \end{figure} Then, the expected revenue of the search engine is \begin{eqnarray} &&2\bigg\{\int_0^1\int_0^{w_1}([(c_1-c_2) \beta_1(w_1)+c_2 \beta_1(w_2)])dw_2dw_1 +\int_0^1\int_0^{w_2}c_2\beta_1(w_2)dw_1dw_2 \bigg\} \nonumber\\ &=& 2\bigg\{(c_1-c_2)\int\limits_0^1 w_1\beta_1(w_1)dw_1+2c_2\int\limits_0^1\int\limits_0^{w_1}\beta_1(x)dxdw_1 \bigg\} =\begin{cases} \frac{c_2}{3}&~c_1=2c_2 \\ \frac{c_2}{6}&~c_1=1.5c_2 \\ \frac{c_1-c_2}{3} &~\text{otherwise} \nonumber \end{cases} \end{eqnarray} It is well known that in the VCG mechanism, the expected VCG revenue is $\dfrac{c_1-c_2}{3}$. Therefore, in this $2\times 2$ case, the proposed algorithm has the same expected revenue as the VCG mechanism after taking the strategic behavior into consideration. \subsection{Symmetric Bayesian Nash Equilibrium in $2\times 2$ General Unit-demand Matching Markets} Theoretically, the same method we used to studied the $2\times 2$ case of a sponsored search market can be used to solve the $2\times 2$ case for a general matching market when valuations are assumed to be \emph{i.i.d.}. However, that method requires us to consider two variables simultaneously, and we cannot avoid solving the resulting partial differential equations. Hence, before directly solving the general $2\times 2$ cases, we present the following lemmas to simplify the analysis afterward. \begin{lem} \label{lem_2g1} In a general $2\times 2$ matching market, any strategy placing non-zero bids on both goods is a weakly-dominated strategy of a rational buyer with non-zero valuation on goods. \end{lem} \begin{proof} Please see Appendix \ref{sec:lem_2g1}. \end{proof} \begin{lem} \label{lem_2g2} If $v_{i1}>v_{i2}$, any bidding strategy of buyer $i$ with $b_{i1}<b_{i2}$ is a weakly dominated strategy. \end{lem} \begin{proof} Please see Appendix \ref{sec:lem_2g2}. \end{proof} With the above two lemmas, the following corollary is straightforward. \begin{cor} Consider a rational buyer $i$, any bidding strategy putting non-zero bid on the good with private value lower than another good is a weakly-dominated strategy. \end{cor} Now, we can get rid of all weakly-dominated strategies to analyze the buyer's strategic behavior to find a symmetric Bayesian Nash Equilibrium. It is worth to note that weakly-dominated strategies can still be rationalizable strategies, our approach (analyzing Bayesian Nash equilibrium(BNE) without taking weakly-dominated strategies into consideration) may miss some BNEs. Our current goal is not to find all BNEs but just one. Therefore, we can continue to find a BNE with restricted strategy space. First, assume $v_{i1}\geq v_{i2}$ without loss of generality, a rational buyer $i$ will bid 0 on the good 2, and bid no more than $v_{11}-v_{12}$ on good 1. Then, we can compute the equilibrium in symmetric bidding strategies. Assume the symmetric bidding function is $\beta(\cdot,\cdot)$ for the good with higher value and 0 for the other one, buyer $i$ bids $b$ on good 1 and the CDF of buyer $-i$'s valuation on good $j$ is $F_{-ij}(\cdot)$, the objective function is \begin{eqnarray} \max_b E_{(v_{-i1},v_{-i2})} [u_i(b,0,\beta\mathbf{1}_{\{v_{-i1}>v_{-i2}\}},\beta\mathbf{1}_{\{v_{-i1}\leq v_{-i2}\}})], \end{eqnarray} where $\beta$ denotes $\beta(v_{-i1},v_{-i2})$. Now, we need Lemma \ref{pro1} to simplify our analysis. \begin{lem} \label{pro1} Denote $v_{ih}$ to be the buyer $i$'s value of the good that has higher value and $v_{il}$ be the buyer $i$'s value of good that has lower value, an optimal bidding function of the higher good $\beta(v_{ih},v_{il})$ is monotonic in $v_{ih}-v_{il}$. \end{lem} \begin{proof} Please see Appendix \ref{sec:pro1}. \end{proof} With Lemma \ref{pro1}, we can revise and simplify the bidding function $\beta(v_{ih},v_{il})$ to be $\beta_{r}(v_{ih}-v_{il})$. Define the joint CDF of $v_{-i1}-v_{-i2}$ to be $F_{-i}(\cdot)$. Then the expected surplus of buyer $i$ for a specific bid $b$ of good 1 is: \begin{eqnarray} (v_{i1}-b)F_{-i}(\beta_r^{-1}(b))+v_{i2}(1-F_{-i}(\beta_r^{-1}(b))) =v_{i2}+(v_{i1}-v_{i2}-b_{i1})F_{-i}(\beta_r^{-1}(b)) \end{eqnarray} Therefore, determining the objective function is equivalent to solving the following optimization problem: \begin{eqnarray} \max_b ~(v_{i1}-v_{i2}-b)F_{-i}(\beta_r^{-1}(b)). \end{eqnarray} To simplify our problem, let's assume the distributions of private value on goods of two buyers are i.i.d. uniformly distributed on $[0,1]$. Then assume that the bidding function is differentiable and denote $x=v_{i1}-v_{i2}$. Using the same technique we used to solve the symmetric BNE in sponsored search market. With the analysis in Appendix \ref{apdx:analysis2}, the bidding function $\beta_r(x)$ is \begin{eqnarray} \beta_r(x)= \dfrac{1}{2-(1-x)^2}\int_0^{x} 2\tau(1-\tau)d\tau =\dfrac{x^2(1-\tfrac{2x}{3})}{2-(1-x)^2} \end{eqnarray} Then, the expected revenue is \begin{eqnarray} 4\int_0^1 \beta_r(x)(1-x)(\frac{1}{2}+x-\frac{x^2}{2}) dx =2\int_0^1 x^2(1-\tfrac{2x}{3})(1-x) dx = 2(\dfrac{1}{3} - \dfrac{5}{12} + \dfrac{2}{15})= \dfrac{1}{10} \end{eqnarray} Finally, computing the expected revenue with VCG price yields: \begin{eqnarray} 4\int_0^1 (1-x)\int_0^x y(1-y) dydx = 4\int_0^1 (1-x)(\dfrac{x^2}{2}-\dfrac{x^3}{3}) dx = 4 ( \dfrac{1}{6}- \dfrac{5}{24}+\dfrac{1}{15}) = \dfrac{1}{10} \end{eqnarray} It is shown here that the revenue equivalence theorem continues to hold in this case. \section{Appendix-Proofs} \subsection{Proof of Lemma \ref{lemA}}\label{pf:lem1} First, we will show that given a constricted good set $S$, reducing price of every good in $S$ with the amount specified in the statement will increase the size of its neighbor, i.e., $N(S)$. Given a constricted good set $S$ under a specific price vector $\mathbf{P}$. Consider another price vector $\mathbf{P'}$, $P'_j=\begin{cases} P_j , & j \notin S \\ P_j-c, & j \in S \end{cases} $ , where $c:=\min_{i \in \mathcal{B} \setminus N(S),l\in S} \{\max_{k \in \mathcal{M} \setminus S}(v_{i,k}-P_k)- (v_{i,l}-P_l) \}$. For some $j \in S$, there must exists an $i\in \mathcal{B} \setminus N(S)$ satisfying $v_{i,j}-P_j'= \max_{k \in \mathcal{M} \setminus S} v_{i,k}-P_k$. Now, by an abuse of notation to denote $N'(S)$ as the neighbor of $S$ under $P'$, $i \in N'(S)$. For those $l \in N(S)$, $\max_{j \in S}v_{l,j}-P'_j>\max_{j \in S}v_{l,j}-P_j\geq \max_{j \in \mathcal{M}\setminus S}v_{l,j}-P_j$ implies $l \in N'(S)$. Therefore, $|N(S)|<|N'(S)|$. Then, we need to prove that $c$ is the minimum decrement. Consider another price vector Consider another price vector $\mathbf{P''}$, $P'_j=\begin{cases} P_j , & j \notin S \\ P_j-d, & j \in S \end{cases} $, where $d<c$. It is straightforward that for all $i \in \mathcal{B} \setminus N(S)$, $v_{i,j}-P''_j< v_{i,j}-P_j+c \le \max_k (v_{i,k}-P_k)$. Hence, no buyers will be added to the $N(S)$. Now, it is clear that $\min_{i \in \mathcal{B} \setminus N(S),l\in S} \{\max_{k \in \mathcal{M} \setminus S}(v_{i,k}-P_k)- (v_{i,l}-P_l) \}$ is the minimum price reduction which guarantees to add at least a new buyer to the $N(S)$. \subsection{Proof of Lemma \ref{lem2}}\label{pf:lem2} In order to prove the statement, we start from showing the most skewed set is always a constricted good set when there is no perfect matching. Then, we prove the uniqueness of the most skewed set by contradiction. When there is no perfect matching, there must exists a constricted good set. By definition, the constricted good set $S$ has the property $|S|>|N(S)|$. Since $|S|$, $|N(S)|$ are integers, the skewness of a constricted good set \begin{equation} \label{eq1} f(S)=|S|-|N(S)|+\frac{1}{|S|} \geq 1+\frac{1}{|S|} >1. \end{equation} Then, for any non-constricted good set $S'$, $|S'| \leq |N(S')|$. The skewness of $S'$ is \begin{equation} \label{eq2} f(S')=|S'|-|N(S')|+\frac{1}{|S'|} \leq 0+\frac{1}{|S'|} \leq 1. \end{equation} With equation (\ref{eq1}), (\ref{eq2}), the skewness of a constricted good set is always greater than any non-constricted good sets. Therefore, if a preference graph exists a constricted good set, the most skewed set is always a constricted good set. Then, we start to prove the uniqueness of the most skewed set. Suppose there exists two disjoint sets $S_1$, $S_2$ and both sets are the most skewed sets, i.e., $f(S_1)=f(S_2)= \max_{S \subset \mathcal{M}, S \neq \emptyset} f(S)$. Consider the skewness of the union of $S_1$ and $S_2$. \begin{eqnarray} &&f(S_1 \cup S_2) \\ &=& |S_1 \cup S_2|- |N(S_1 \cup S_2)|+ \frac{1}{|S_1 \cup S_2|} \\ &=& |S_1|+ |S_2| - |N(S_1) \cup N(S_2)|+ \frac{1}{|S_1 \cup S_2|} \\ &\geq & |S_1|+ |S_2| - |N(S_1)|- |N(S_2)|+ \frac{1}{|S_1 \cup S_2|} \\ &=& f(S_1)+ |S_2|- |N(S_2)| -\frac{1}{|S_1|}+\frac{1}{|S_1 \cup S_2|} \label{eq3}\\ &\geq & f(S_1) +1 -\frac{1}{|S_1|}+\frac{1}{|S_1 \cup S_2|} > f(S_1) \label{eq4} \end{eqnarray} From (\ref{eq3}) to (\ref{eq4}) is true because $S_2$ is a constricted good set. (\ref{eq4}) contradicts our assumption that $S_1$, $S_2$ are the most skewed set. Therefore, we know that if there exists multiple sets share the same highest skewness value, these sets are not disjoint. Now, suppose there exists two sets $S_1$, $S_2$ satisfying that $S_1 \cup S_2 \neq \emptyset$ and both sets are the most skewed sets, the following two inequalities must hold: \begin{eqnarray} f(S_1)-f(S_1 \cup S_2) \geq 0 \\ f(S_2)-f(S_1 \cap S_2) \geq 0 \end{eqnarray} Let's sum up the two inequalities and represent the formula in twelve terms. \begin{eqnarray} &&f(S_1)+f(S_2) - f(S_1 \cup S_2)-f(S_1 \cap S_2) \\ &=& |S_1|+|S_2|-|S_1 \cup S_2|-|S_1 \cap S_2| \nonumber \\ && +|N(S_1 \cup S_2)|+ |N(S_1 \cap S_2)|- |N(S_1)|-|N(S_2)| \nonumber \\ && +\dfrac{1}{|S_1|}+\dfrac{1}{|S_2|}-\dfrac{1}{|S_1 \cup S_2|}-\dfrac{1}{|S_1 \cap S_2|} \end{eqnarray} The first four terms $|S_1|+|S_2|-|S_1 \cup S_2|-|S_1 \cap S_2|=0$. Using the similar argument, $|N(S_1)|+|N(S_2)|=|N(S_1) \cap N(S_2)|+|N(S_1) \cup N(S_2)|$. \begin{eqnarray} |N(S_1 \cup S_2)|=|N(S_1) \cup N(S_2)| \\ |N(S_1 \cap S_2)| \leq |N(S_1) \cap N(S_2)| \label{eq5} \end{eqnarray} Equation (\ref{eq5}) is true because there may exist some elements in $S_1 \setminus S_2$ and $S_2 \setminus S_1$ but have common neighbors. Thus, the second four terms are smaller than or equal to $0$. To check the last four terms, let $|S_1|=a$, $|S_2|=b$, and $|S_1 \cap S_2|=c$, where $c<\min\{a,b\}$ because $S_1,S_2$ are not disjoint. The last four terms are \begin{eqnarray} &&\dfrac{1}{|S_1|}+\dfrac{1}{|S_2|}-\dfrac{1}{|S_1 \cup S_2|}-\dfrac{1}{|S_1 \cap S_2|} \\ &=& \dfrac{1}{a}+\dfrac{1}{b}-\dfrac{1}{a+b-c}-\dfrac{1}{c} \\ &=& \dfrac{a+b}{ab}-\dfrac{a+b}{(a+b-c)c} \\ &=& \dfrac{a+b}{abc(a+b-c)}(ac+bc-c^2-ab) \\ &=&-\dfrac{(a+b)(a-c)(b-c)}{abc(a+b-c)}<0 \end{eqnarray} To conclude, the first four terms are $0$, the second four terms are smaller than or equal to $0$, and the last four terms are strictly negative make \begin{eqnarray} f(S_1)+f(S_2) - f(S_1 \cup S_2)-f(S_1 \cap S_2)<0. \end{eqnarray} Therefore, at least one of set $S_1 \cup S_2$, $S_1 \cap S_2$ has the skewness value greater than $f(S_1)=f(S_2)$, which leads to a contradiction that $S_1$ and $S_2$ are the most skewed set. Finally, we can claim that the most skewed set is unique when there is no perfect matching. \subsection{Proof of Lemma \ref{lem:conv}}\label{pf:lem:conv} First, we prove that the algorithm terminates in finite (at most $|\mathcal{M}|^3$) rounds by investigating the relationship of the most skewed sets in the consecutive rounds, $S^*_t$, $S^*_{t+1}$. The relationship of $S^*_t$, $S^*_{t+1}$ has four cases. \begin{enumerate} \item $S^*_t=S^*_{t+1}$ \item $S^*_t \subset S^*_{t+1}$ \item $S^*_t \supset S^*_{t+1}$ \item $S^*_t \nsubseteq S^*_{t+1}$ and $S^*_t \nsupseteq S^*_{t+1}$ \end{enumerate} Recall that $W(G)$ is the skewness of the graph $G$ and $N_t(S)$ is the neighbor of $S$ based on the preference graph at round $k$. \\ In case 1, $W(G_t)-W(G_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_t)\geq 1$. \\ In case 2, define $S'=S^*_{t+1}\setminus S^*_t$, and it is trivial that $S^*_t \subset S^*_{t+1}$ implies $1>\frac{1}{|S^*_t|}-\frac{1}{|S^*_{t+1}|}>0$. \begin{eqnarray} &&W(G_t)-W(G_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_{t+1})\\ &=&|S^*_t|-|N_t(S^*_t)|-|S^*_{t+1}|+|N_{t+1}(S^*_{t+1})| +\dfrac{1}{|S^*_t|}-\dfrac{1}{|S^*_{t+1}|} \label{eq6}\\ &\geq &|S^*_{t+1}|-|N_t(S^*_{t+1})|-|S^*_{t+1}|+|N_{t+1}(S^*_{t+1})|+\dfrac{1}{|S^*_t|}-\dfrac{1}{|S^*_{t+1}|} \label{eq7}\\ &=&|N_{t+1}(S^*_t \cup S')|-|N_t(S^*_{t+1})|+\frac{|S^*_{t+1}|-|S^*_t|}{|S^*_t||S^*_{t+1}|} \nonumber \\ &\geq &|N_{t+1}(S^*_t \cup S')|-|N_t(S^*_{t+1})| + \dfrac{1}{|\mathcal{M}|(|\mathcal{M}|-1)} \label{eq8}\\ &=&|N_{t+1}(S^*_t)|+|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c| -|N_t(S^*_{t+1})|+ \dfrac{1}{|\mathcal{M}|^2-|\mathcal{M}|} \label{eq9}\\ &=&|N_t(S^*_t)|+|N_{t+1}(S_t)\setminus N_t(S^*_t)|- |N_t(S^*_{t+1})| +|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c|+ \tfrac{1}{|\mathcal{M}|^2-|\mathcal{M}|} \label{eq10} \end{eqnarray} Before going to further steps, we have to briefly explain the logic behind the above equations. (\ref{eq6}) to (\ref{eq7}) is true because $N_t(S^*_t) \subseteq N_t(S^*_{t+1})$. (\ref{eq8}) to (\ref{eq9}) is to expand $|N_{t+1}(S^*_t \cup S')|$ to $|N_{t+1}(S^*_t)|+|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c|$. (\ref{eq9}) to (\ref{eq10}) is to expand $|N_{t+1}(S^*_t)|$ to $|N_t(S^*_t)|+|N_{t+1}(S_t)\setminus N_t(S^*_t)|$. Since $|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c|=|N_{t+1}(S')\cap N_{k}(S^*_t)^c|-|N_{t+1}(S')\cap (N_{t+1}(S_t)\setminus N_t(S^*_t))|$, and $|N_{t+1}(S_t)\setminus N_t(S^*_t)|\geq |N_{t+1}(S')\cap (N_{t+1}(S_t)\setminus N_t(S^*_t))|$, we can further summarize the first four terms in (\ref{eq10}). \begin{eqnarray} &&|N_t(S^*_t)|+|N_{t+1}(S_t)\setminus N_t(S^*_t)| +|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c| - |N_t(S^*_{t+1})| \\ &\geq & |N_t(S^*_t)|+|N_{t+1}(S')\cap N_{k}(S^*_t)^c| -|N_t(S^*_{t+1})|\\ &=& |N_t(S^*_t)|+|N_{t+1}(S')\cap N_t(S^*_t)^c| -|N_t(S^*_t)|-|N_t(S')\cap N_t(S^*_t)^c| \\ &=& |N_{t+1}(S')\cap N_t(S^*_t)^c|-|N_t(S')\cap N_t(S^*_t)^c| \label{eq32} \\ &=& |N_t(S')\cap N_t(S^*_t)^c|-|N_t(S')\cap N_t(S^*_t)^c|=0 \label{eq33} \end{eqnarray} (\ref{eq32}) to (\ref{eq33}) is true because $S'$ is not in $S^*_t$, the neighbor of $S'$ not in the neighbor of $S^*_t$ remains the same from $k$ to $t+1$ round. With equation (\ref{eq10}) and (\ref{eq33}), we can conclude that $W(G_t)-W(G_{t+1})\geq \frac{1}{|\mathcal{M}|^2-|\mathcal{M}|}$ in case 2.\\ \\ In case 3, since every elements in $S^*_t$ belongs to $S^*_t$, $f_t(S^*_{t+1}) \geq f_{t+1}(S^*_{t+1})$. Given that $\dfrac{1}{|S^*_t|}<\dfrac{1}{|S^*_{t+1}|}$ and $f_t(S^*_t)-f_t(S^*_{t+1})>0$ by definition, $|S^*_t|-|S^*_{t+1}|+|N_t(S^*_t)|-|N_t(S^*_{t+1})|\geq 1$. With the knowledge that $S^*_t$ and $S^*_t$ are non-empty set, it is obvious that $f_t(S^*_t)-f_t(S^*_{t+1})$ is lower-bounded by $\tfrac{1}{2}$. Therefore, we can conclude that $W(G)_t-W(G_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_{t+1}) \geq f_t(S^*_t)-f_t(S^*_{t+1}) \geq \frac{1}{2}$ in case 3.\\ \\ In case 4, define $S'=S^*_{t+1} \setminus S^*_t$, $S''=S^*_t \setminus S^*_{t+1}$, and $T=S^*_t \cap S^*_{t+1}$. \begin{eqnarray} && W(G_t)-W(G_{t+1}) \\ &=&f_t(S^*_t)-f_{t+1}(S^*_{t+1}) \\ &=& |T|+|S''|-|N_t(T)|-|N_t(S'')\setminus N_t(T)|+\frac{1}{|S^*_t|} - |T|-|S'|+|N_{t+1}(S^*_{t+1})|-\frac{1}{|S^*_{t+1}|} \\ &=& |S''|-|N_t(S'')\setminus N_t(T)|+\frac{1}{|S^*_t|}-\frac{1}{|S^*_{t+1}|} +|N_{t+1}(S^*_{t+1})|-|N_t(T)|-|S'| \\ &\geq & 1+\frac{1}{|S^*_t|}-\frac{1}{|S^*_{t+1}|}+|N_{t+1}(S^*_{t+1})| -|N_t(T)|-|S'| \\ &\geq & 1+\frac{1}{|\mathcal{M}|}-1+|N_{t+1}(T))|-|N_t(T)|+|N_{t+1}(S')\setminus N_{t+1}(T)|-|S'| \\ &= & \frac{1}{|\mathcal{M}|}+|N_{t+1}(T))|-|N_t(T)| +|N_{t+1}(S')\setminus N_{t+1}(T)|-|S'| \label{eq11}\\ &\geq & \frac{1}{|\mathcal{M}|}+|N_{t+1}(S')\setminus N_{k}(T)|-|S'| \geq \frac{1}{|\mathcal{M}|} \label{eq12}\\ \end{eqnarray} Most of the equations in case 4 are straight-forward except from (\ref{eq11}) to (\ref{eq12}). (\ref{eq11}) to (\ref{eq12}) is true because of the following inequalities: \begin{eqnarray} &&|N_{t+1}(S')\setminus N_{t+1}(T)|+|N_{t+1}(T))|-|N_t(T)| \nonumber \\ &=&|N_{t+1}(S')\setminus N_t(T)|+|N_{t+1}(T))|-|N_t(T)|-| \{N_{t+1}(S') \cap N_{t+1}(T) \}\setminus N_t(T)| \\ &\geq & |N_{t+1}(S')\setminus N_t(T)|-| N_{t+1}(T)\setminus N_t(T)| +|N_{t+1}(T))|-|N_t(T)| \\ &=& |N_{t+1}(S')\setminus N_t(T)| \end{eqnarray} Combine these four cases, we know that \begin{eqnarray} W(G_t)-W(G_{t+1}) \geq \dfrac{1}{ |\mathcal{M}|^2-|\mathcal{M}|}. \end{eqnarray} Therefore, we can conclude that the proposed algorithm terminates in finite rounds. (At most $|\mathcal{M}|^3$ rounds because $W(G)<|\mathcal{M}|$ and minimum decrement is greater than $\tfrac{1}{|\mathcal{M}|^3}$). \subsection{Proof of Lemma \ref{lem:colorMSS}} First, we want to show that suppose $\mathcal{M}_g \cup\mathcal{M}_b$ is a constricted good set, we can not include any good colored red to increase the skewness of the set. For any set of good colored red $S_r$ being added to $\mathcal{M}_g \cup\mathcal{M}_b$, at least the same size of buyers being matched pairs of those red goods are join to $N(\mathcal{M}_g \cup\mathcal{M}_b \cup S_r)$. Since we know that $|N(S_r)\setminus N(\mathcal{M}_g \cup\mathcal{M}_b)|\geq |S_r|$, including any set of red goods will decrease the skewness of the set $\mathcal{M}_g \cup\mathcal{M}_b$. Then, we want to show remove any subset of goods $S\subseteq \mathcal{M}_g \cup\mathcal{M}_b$ will also reduce the skewness of the set. Clearly, removing any subset of blue good will not reduce $N(\mathcal{M}_g \cup\mathcal{M}_b)$, hence blue goods are definitely included in the maximally skewed set. Therefore, we only need to consider the impact of removing set of green goods $S \subseteq \mathcal{M}_g$. Since for any $S \subseteq \mathcal{M}_g$, there has has at least one red edge connecting $S$ and $N(\mathcal{M}_g \cup\mathcal{M}_b\setminus S)$ according to the algorithm, Hence, $|N(\mathcal{M}_g \cup\mathcal{M}_b)|-|N(\mathcal{M}_g \cup\mathcal{M}_b\setminus S)|<|S|$, which implies that removing any $S \subset \mathcal{M}_g$ will only increase the skewness of the set. With these facts and the uniqueness property of the maximally skewed set, we can conclude that $\mathcal{M}_g \cup\mathcal{M}_b$ is the maximally skewed set. \subsection{Proof of Lemma \ref{lem:alg2}}\label{pf:lemalg2} We will prove Algorithm \ref{alg:alg2} always return the most skewed set by contradiction.\\ Since every untraversed good upon termination can be matched with an untraversed buyer without repetition. Therefore, adding any set of runtraversed goods $S_U$ to the set $S$ will always reduce the skewness of $S$ (because the increase of cardinality in neighbor of $S$, $|N(S_U)|-|N(S)|$ is always greater than or equal to the increase of cardinality of $S$, $|S_U|-|S|$). Hence, the most skewed set will never contain any untraversed good. Suppose there exists a set $S'$ is the most skewed set, with a higher skewness than the set $S$ return by Algorithm \ref{alg:alg2}, $S'$ must be a subset of $S$ because there's no untraversed good in the most skewed set and $f(S')>f(S)$. Let $S^*=S\setminus S'$, $f(S')>f(S)$ and $S' \subset S$ implies $|N(S^*)\setminus N(S')|-|S^*|>0$. If this happens, there must be a non-empty set of matching pairs match nodes from $S^*$ to $N(S^*)\setminus N(S')$ without repetition. Since $N(S^*)\setminus N(S')$ is not in $N(S')$ under the directed graph, nodes in $S^*$ will not be traversed in the algorithm contradicts that $S^* \subset S$. \subsection{Proof of Theorem \ref{lem4.1}}\label{pf:lem4.1} First, let's begin with the proof of the first half statement of the theorem, which is a variational characterization of MCPs. ($\Rightarrow$) This direction is obvious, otherwise the MCP is not the maximum by definition. ($\Leftarrow$) Recall that $\mathcal{M}$ is the set of goods. Suppose there exists an MCP $P^1$ satisfying the conditions but it is not the maximum MCP. Then there must exist a set of goods $S_1$ such that for all $i \in S_1$, $P^1_i<P^*_i$, and for all $i \in \mathcal{M}-S_1$, $P^1_i \geq P^*_i$; $\mathcal{M}-S_1$ can be an empty set. Let $P^2_i= P^1_i$ for all $i \in \mathcal{M}-S_1$, and $P^2_i= P^*_i$ for all $i \in S_1$. We will verify that $P^2$ is an MCP. WLOG, we can assume $P^1$ and $P^*$ have the same allocation; this is true as every MCP supports all efficient matchings. Then, consider any buyer who is assigned a good in $S_1$ under $P^*$. When the price vector changes from $P^*$ to $P^2$, the buyer has no profitable deviation from his/her assigned good because $P^2_i \geq P^*_i$ for all $i$ and $P^2_j= P^*_j$ for all $j \in S_1$. Similarly, when the price vector changes from $P^1$ to $P^2$, the buyer who is assigned a good in $\mathcal{M} \setminus S_1$ under $P^1$ has no profitable deviation because $P^2_i=P^1_i$ for all $i \in \mathcal{M} \setminus S_1$ and $P^2_j>P^1_j$ for all $j \in S_1$. Finally, since $P^1$ and $P^*$ have the same allocation, no buyer will deviate if we assign this allocation to buyers under $P^2$. Since all buyers have non-negative surpluses under $P^2$, it follows that $P^2$ is an MCP. Given $P^1$, there exists a set of goods $S_1$ whose price we can increase and still get market clearing because both $P^1$ and $P^2$ are MCPs. This contradicts the assumption that $P^1$ satisfies the stated conditions, and the proof of the variational characterization follows. For the second half part, which is a combinatorial characterization of MCPs, let's try to prove the statement using the result of the variational characterization. Given that we cannot increase the price for any subset of goods, it implies that for any subset of goods $S$, the set of corresponding matched buyer, either there exists at least one buyer has an edge connected to a good not in this subset or there exists at least one buyer with surplus zero. In the former case, it is obvious that $B<N(B)$ for such corresponding buyer set $B$. In the later case, $B\leq N(B)$ and $D\in N^D(B)$ guarantees $B < N^D(B)$. For the opposite direction, if every set of buyer with $B < N^D(B)$ under the current MCP, increasing the price for any set of good will make at least one corresponding buyer deviate from the matched good and cause no perfect matching. Therefore, the condition in variational characterization holds if and only if the condition in the combinatorial characterization holds. \subsection{Proof of Theorem \ref{thm2}}\label{pf:thm2} As mentioned earlier, showing the algorithm returns the maximum MCP is equivalent to show that when the algorithm terminates, the bipartite structure guarantees that we can not increase the price of any subset of good with the result in Theorem \ref{lem4.1}. It implies that after adding a dummy good with zero price, the support of any subset of goods has a size less than the size of the subset of goods, i.e., $|B|< |N^D_T(B)|$. Therefore, the proof of the theorem can be transformed to prove that the algorithm satisfies $|B|< |N^D_T(B)|$ for any non-empty set of buyer $B$ on termination. Let's start the proof with several claims. \begin{claim} \label{cl1} For any subset of buyer $B$, $B\neq \emptyset$, $|B|\leq |N^D_T(B)|$, where $T$ is the terminating time of our algorithm. \end{claim} Since our algorithm returns an MCP vector, if $|B|> |N^D_T(B)|$, there does not exist a perfect matching because of the existence of constricted buyer set. \begin{claim} \label{cl2} There does not exists a subset of buyer $B\subseteq \mathcal{B}$, $B\neq \emptyset$, $|B|= |N^D_T(B)|$ and $D\in N^D_T(B)$, where $T$ is the terminating time of our algorithm. \end{claim} $|B|= |N^D_T(B)|$ and $D\in N^D_T(B)$, imply $|B|> |N_T(B)|$. The preference graph has constricted sets and has no perfect matching. \begin{claim} \label{cl3} There does not exists a subset of buyer $B\subseteq \mathcal{B}$, $B\neq \emptyset$, $|B|= |N_T(B)|$, where $T$ is the terminating time of our algorithm. \end{claim} \begin{proof} With Claim \ref{cl1}, \ref{cl2}, it is equivalent to show $|B|<|N_T(B)|$. At time $t$, we denote the maximum non-negative surplus of buyer $b$ by $u^*_t(b)$ and the most skewed set by $S^*_t$; and $\mathscr{B}^s_t$ is the set of buyers with positive surplus at time $t$, i.e., $\mathscr{B}^s_t=\{b|u^*_t(b)>0, b\in\mathcal{B}\}$. It is obvious that $\mathscr{B}^s_i \subseteq \mathscr{B}^s_j$ for all $i<j$. Then, we want to prove the claim by mathematical induction. Prior to the proof, we need to introduce another claim. \begin{claim} \label{cl4} $\forall B\subseteq N_t(S^*_t), 0\leq t<T$, $|B|<|N_t(B) \cap S^*_t)|=|N_{t+1}(B)|$ \end{claim} \begin{proof} If the left inequality does not hold, we can remove all the goods contained in $N_t(B) \cap S^*_t$ from $S^*_t$ to get a more skewed set, which violates that $S^*_t$ is the most skewed one. Since we reduce the price in $S^*_t$, buyers in $B$ will not prefer any good outside of $S^*_t$. Therefore, $N_t(B) \cap S^*_t$ and $N_{t+1}(B)$ are identical. Hence $|N_t(B) \cap S^*_t|=|N_{t+1}(B)|$ is absolutely true. \end{proof} Now, we can prove Claim \ref{cl3} by induction. At t=0, $\mathscr{B}^s_0=\emptyset$. At t=1, $\mathscr{B}^s_1=\mathcal{N_0}(S^*_0)$. With Claim \ref{cl4}, $|B|<|N_1(B)|~~\forall B\subseteq \mathscr{B}^s_1$ and $B \neq \emptyset$. At a finite time $t$, suppose for all $B\subseteq \mathscr{B}^s_t$, $B \neq \emptyset$ satisfy $|B|<|N_t(B)|$, consider at time $t+1$: Since $\mathscr{B}^s_t \subseteq \mathscr{B}^s_{t+1}$, $\mathscr{B}^s_{t+1}$ contains three disjoint components: \begin{equation} \mathscr{B}^s_{t+1}= \{\mathscr{B}^s_t \cap {N_t(S^*_t)}^c \} \cup \{\mathscr{B}^s_t \cap N_t(S^*_t) \} \cup \{{\mathscr{B}^s_t}^c \cap N_t(S^*_t) \} \end{equation} Buyers in the first two parts are originally with positive utilities. Buyers in the last part have zero utilities at time $t$ but have positive utilities at time $t+1$. Consider the subset of buyers $B_{\alpha} \subseteq \{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$. Since every $b \in B_{\alpha}$ does not prefer any good in $S^*_t$, the price reduction in $S^*_t$ will never remove any edges between $\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$ and $N_t(\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \})$. Therefore, for any non-empty set of buyers $B_{\alpha}$, $|B_{\alpha}|< |N_t(B_{\alpha})|\leq |N_{t+1}(B_{\alpha})|$. Then, consider the second and the third parts. Since $\{\mathscr{B}^s_t \cap N_t(S^*_t) \} \cup \{{\mathscr{B}^s_t}^c \cap N_t(S^*_t) \}=N_t(S^*_t)$, every non-empty set of buyers $B_{\beta} \subseteq N_t(S^*_t)$ satisfies $|B_{\beta}|<|N_{t+1}(B_{\beta})|$ by Claim \ref{cl4}. At the last step, consider $B_{\gamma}=B_{\gamma_1} \cup B_{\gamma_2}$, where $B_{\gamma_1} \subseteq \{\mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$, $B_{\gamma_2} \subseteq N_t(S^*_t)$ and $B_{\gamma_1},B_{\gamma_2} \neq \emptyset$. \begin{eqnarray} |N_{t+1}(B_{\gamma})| &=&|N_{t+1}(B_{\gamma_1}) \cup N_{t+1}(B_{\gamma_2})| \\ &=& |N_{t+1}(B_{\gamma_1})|+|N_{t+1} (B_{\gamma_2})|-|N_{t+1}(B_{\gamma_1}) \cap N_{t+1}(B_{\gamma_2})| \\ &=& |N_{t+1}(B_{\gamma_1})|-|N_{t+1}(B_{\gamma_1}) \cap N_{t+1}(B_{\gamma_2})|+|N_{t+1} (B_{\gamma_2})| \\ &=& |N_{t+1}(B_{\gamma_1}) \cap {N_{t+1}(B_{\gamma_2})}^c|+|N_{t+1} (B_{\gamma_2})| \\ &\geq& |N_{t+1}(B_{\gamma_1}) \cap {N_{t+1}(N_t(S^*_t))}^c|+|N_{t+1} (B_{\gamma_2})| \label{eqx1} \\ &=& |N_{t+1}(B_{\gamma_1}) \cap {S^*_t}^c|+|N_{t+1} (B_{\gamma_2})| \label{eqx2}\\ &=& |N_t(B_{\gamma_1})|+|N_{t+1} (B_{\gamma_2})| \label{eqx3}\\ &>& |B_{\gamma_1})|+|B_{\gamma_2}|=|B_{\gamma}| \end{eqnarray} From (\ref{eqx1}) to (\ref{eqx2}) is true because after price reduction at time $t$, buyer belongs to the neighbor of constricted good set $S^*_t$ will only prefer goods in $S^*_t$, therefore $N_{t+1}(N_t(S^*_t))=S^*_t$. From (\ref{eqx2}) to (\ref{eqx3}) is true because $N_t(B_{\gamma_1}) \cap S^*_t = \emptyset$ (by definition). Then, we have discussed before that the price reduction in $S^*_t$ will never remove any edges between $\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$ and $N_t(\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \})$. Since goods not in the most skewed set at time $t$ will not add new buyers to their neighbor, the equivalence between $N_{t+1}(B_{\gamma_1}) \cap {S^*_t}^c$ and $N_t(B_{\gamma_1})$ holds. Therefore, for every non empty set of buyers $B\subseteq \mathscr{B}^s_{t+1}$, $|B|<|N_{t+1}(B)|$. The mathematical induction works for all $k \in \mathbb{N}$. Since our algorithm terminates in finite round, $|B|< |N_T(B)|$, Q.E.D. \end{proof} With Claim \ref{cl1}, \ref{cl2}, \ref{cl3}, the skew-aided algorithm satisfies $|B|<|N^D_t(B)|$ for any non-empty set of buyer $B$ on termination. \subsection{Proof of Theorem \ref{thm1}}\label{pf:thm1} First, we know from Section 5.2 that the preference graph changes at most $m^2$ times in Algorithm \ref{alg:alg1}. Second, we analyze the initial step first. In Algorithm \ref{alg:alg2}, the Hopcroft-Karp algorithm runs in time $O(|\mathcal{M}|^{2.5})$, and the complexity of the BFS algorithm is $O(m+|\text{number of edges}|)$. Since the number of directed edges is upper-bounded by $m^2$, the complexity of Algorithm \ref{alg:alg2} is $O(m^{2.5}+m^2)=O(m^{2.5})$. Then, for each $a \in A$ colored blue, we run at most twice of BFS/DFS algorithm, which has complexity upper-bounded by $O(m+|\text{number of edges}|)\geq O(m^2)$. After that, we do things similar to find the maximum matching if we can add at least one blue buyer to the maximum matching, and run at most twice BFS/DFS. Therefore, the complexity should be upper-bounded by $O(m^{2.5}+2\times m^2)=O(m^{2.5})$. If we know that there is no blue buyers going to be added in this round, we do not need to find a new maximum matching. Therefore, the total complexity of the algorithm attaining the maximum MCP will be upper-bounded by the \{(complexity of updating process not recoloring blue buyers)$+$(complexity of computing price reduction)\}$\times$ (convergence rate of the preference graph)$+$(complexity of updating process increasing maximum matched pairs)$\times$ (maximum number of blue buyers)$=O((m^2+m^2)\times m^2+m^{2.5}\times m)=O(m^4)$ \subsection{Proof of Lemma \ref{lem:multicon}} Suppose the picked set which is not the most skewed one, this set must contain a pair of good-buyer $(x,y)$ not in the maximally skewed set. Since we know that algorithm terminates at round $T$, reducing the price of the good $x$ make the price vector lower than the maximum MCP at round $T$. (Because if we pick another set excluding good $x$, reducing the same amount of price as we did in this algorithm will also terminate at $T$.) Therefore, we know that the constricted good set we pick should not include any good not in the maximally skewed set. Now, we want to claim that the picked set is a subset of the maximally skewed set will never happen. If such a set $S$ exists and can terminate the algorithm at round $T$, then the most skewed set $S^*$ contains a set of goods $S^*\setminus S$ is already perfectly matched to a set of buyers outside $N(S)$, and the skewness of $S$ must be greater than $S^*$, which contradicts that $S^*$ is the most skewed set. Because of the above claims, we can conclude that any choice of constricted good set which is not the maximally skewed set at round $T$ will never return the maximum MCP at round $T$. \subsection{Proof of Theorem \ref{thm:externality}}\label{pf:thm:externality} First, it is obvious that the current market adding a duplicate pair of buyer buyer $i$ and its matched good $j$ is still market clearing and the social welfare will be the $\text{(the current social welfare)}+U^*_i+P_j$. Hence, what we need to show is the current market adding a duplicate buyer buyer $i$ has the social welfare: $\text{(the current social welfare)}+U^*_i$. Since adding a dummy good will not change the social welfare, we can transfer our problem to prove that the current market adding a duplicate buyer buyer $i$ and a dummy good has the social welfare $\text{(the current social welfare)}+U^*_i+0$. It is equivalent to show that the current market adding a duplicate buyer buyer $i$ and a dummy good is still market clearing. Using the combinatorial characterization of the maximum MCP, we know that $|B|<|N^D(B)|$ for any subset of B in the current market. Denote the buyer set of current market as $\mathcal{B}$ and the duplicate buyer $i$ as $\hat{i}$ , what we want to show is that $|B|\leq |N^D(B)|$ for any $B \in \{\mathcal{B}\cup\hat{i}\}$. If at least one of $i, \hat{i}$ not in $B$, it is straightforward that $|B|<|N^D(B)|$. If both $i, \hat{i}\in B$, $|B|=|B\setminus \hat{i}|+1 \leq |N^D(B\setminus \hat{i})-1|+1=|N^D(B\setminus \hat{i})|\leq |N^D(B)|)$. Using the Hall's marriage theorem, the inequality guaranteed that the current market adding a duplicate buyer buyer $i$ and a dummy good is market clearing, and the proof stands here. \subsection{Verification of Asymmetric BNE in a $3\times 3$ matching market}\label{asymBNE} Given the valuation matrix described in Table \ref{tbl:1}, we want to verify that if the strategy profile $\{\beta_{Alice}(w,0,0), \beta_{Bob}(x,0,0.5),\beta_{Carol}(y,z,2)\}=\{(\max \{1-\epsilon, \frac{w}{2}\},0,0), (\max \{1, \frac{2x-1}{4}\},0,0),(0,\epsilon,0)\}$, where $\epsilon$ is an infinitesimal is a BNE. First, consider Alice's best response according to Bob's strategy $(\max \{1, \frac{2x-1}{4}\},0,0)$ and Carol's strategy $(0,\epsilon,0)$. When Alice has valuation between $[0,1]$ on the listing ads, any bidding strategy $(b,0,0)$ is a best response for all $b<1$ because Alice will always get the Pop-ups at price zero. Then, when Alice has valuation $w$ between $(2,3]$ on the listing ads. The bidding function maximizes Alice's expected payoff is \begin{eqnarray} &&\max_b \int_{1.5}^{3.5} (w-b)\mathbf{1}_{\{b>\max \{1, (2x-1)/4\}\}} f_x(x) dx \\ &=& \max_b \Big\{ \frac{2}{3}(w-b)+ \frac{1}{3}\int_{2.5}^{3.5} (w-b)\mathbf{1}_{\{b>\max \{1, (2x-1)/4\}\}} dx \Big\} \\ &=& \max_b \frac{2}{3}(w-b)+ \frac{1}{3}(w-b)\times 2(b-1) \end{eqnarray} Now, it is easy to solve the optimal bid of Alice is \begin{eqnarray} &&\text{arg}\max_b \frac{2}{3}(w-b)+ \frac{1}{3}(w-b)\times 2(b-1) \\ &=& \text{arg}\max_b (w-b)+(w-b)(b-1)=\frac{w}{2} \end{eqnarray} Since $\frac{w}{2}<1$ for all $w \in [0,1)$ and $\frac{w}{2}>1$ for all $w \in (2,3]$ , we can conclude that $(\max \{1-\epsilon, \frac{w}{2}\},0,0)$ is a best response of Alice under $\{\beta_{Bob}(x,0,0.5),\beta_{Carol}(y,z,2)\}=\{(\max \{1, \frac{2x-1}{4}\},0,0),(0,\epsilon,0)\}$. Second, we use the similar technique to get Bob's best response. Given the strategy of Alice and Carol as mentioned, Bob can always get the Pop-ups with price $0$. Therefore, the bidding function maximizes Bob's expected payoff is \begin{eqnarray} &&\max_b 0.5+\int_{0}^{3} (w-b-0.5)\mathbf{1}_{\{b>\max \{1, \frac{w}{2}\}\}} f_w(w) dw \\ &=& 0.5+\max_b \Big\{ \frac{2}{3}(w-b-0.5)\mathbf{1}_{\{b\geq 1 \}}+ \frac{1}{3}\int_{2}^{3} (w-b-0.5)\mathbf{1}_{\{b>\max \{1, \frac{w}{2}\}\}} dw \Big\} \\ &=& 0.5+\max_b \frac{2}{3}(w-b-0.5)\mathbf{1}_{\{b\geq 1 \}}+ \frac{2}{3}(w-b-0.5)(b-1) \end{eqnarray} The optimal bid of Bob is \begin{eqnarray} &&\text{arg}\max_b \frac{2}{3}(w-b-0.5)\mathbf{1}_{\{b\geq 1 \}}+ \frac{2}{3}(w-b-0.5)(b-1) \\ &=& \text{arg}\max_b (w-b-0.5)(b-1+\mathbf{1}_{\{b\geq 1 \}})=\max \{1, \frac{2x-1}{4}\} \end{eqnarray} Last, we have to verify the Carol's best response given $\{\beta_{Alice}(w,0,0), \beta_{Bob}(x,0,0.5)\}=\{(\max \{1-\epsilon, \frac{w}{2}\},0,0), (\max \{1, \frac{2x-1}{4}\},0,0)\}$. To against any tie-breaking rule not in favor of Carol, the strategy $(0,\epsilon,0)$ is the minimum bid to ensure getting the sidebar ads. Now, we complete the verification that $\{\beta_{Alice}(w,0,0), \beta_{Bob}(x,0,0.5),\beta_{Carol}(y,z,2)\}=\{(\max \{1-\epsilon, \frac{w}{2}\},0,0), (\max \{1, \frac{2x-1}{4}\},0,0),(0,\epsilon,0)\}$ is an (asymmetric) BNE in this $3 \times 3$ matching market. \subsection{Proof of Lemma \ref{lem:mono}}\label{pf:lem:mono} We prove the monotonicity by contradiction. Suppose there exists two weight $x>y$ such that $\beta(x)<\beta(y)$, where $\beta (\cdot)$ is the optimal bidding function. It guarantees that the expected surplus of bidding $\beta(x)$ is never worse than bidding $\beta(y)$ given the private weight $x$. (Otherwise it is not the optimal bidding function.) Similarly, it also has to satisfy that the expected surplus of bidding $\beta(y)$ will be never worse than bidding $\beta(x)$ given the private weight $y$. Mathematically, the following two inequality should hold. \begin{eqnarray} \label{eqn:x>y} \int_0^1(c_1x-[(c_1-c_2)\beta(x) +c_2\beta(u)])f_{w}(u)\mathbf{1}_{\{\beta(x)>\beta(u)\}} du +\int_0^1 c_2(x-\beta(x))f_{w}(u)\mathbf{1}_{\{\beta(x)\leq \beta(u)\}}du \nonumber \\ \geq \int_0^1(c_1x-[(c_1-c_2)\beta(y) +c_2\beta(u)])f_{w}(u)\mathbf{1}_{\{\beta(y)>\beta(u)\}} du +\int_0^1 c_2(x-\beta(y))f_{w}(u)\mathbf{1}_{\{\beta(y)\leq \beta(u)\}}du \end{eqnarray} \begin{eqnarray} \label{eqn:y>x} \int_0^1(c_1y-[(c_1-c_2)\beta(y) +c_2\beta(u)])f_{w}(u)\mathbf{1}_{\{\beta(y)>\beta(u)\}} du +\int_0^1 c_2(y-\beta(y))f_{w}(u)\mathbf{1}_{\{\beta(y)\leq \beta(u)\}}du \nonumber \\ \geq \int_0^1(c_1y-[(c_1-c_2)\beta(x) +c_2\beta(u)])f_{w}(u)\mathbf{1}_{\{\beta(x)>\beta(u)\}} du +\int_0^1 c_2(y-\beta(x))f_{w}(u)\mathbf{1}_{\{\beta(x)\leq \beta(u)\}}du \end{eqnarray} Now, let us sum up and summarize the inequalities (\ref{eqn:x>y}), (\ref{eqn:y>x}). \begin{eqnarray} \int_0^1c_1(x-y)f_{w}(u)\mathbf{1}_{\{\beta(x)>\beta(u)\}} du +\int_0^1 c_2(x-y)f_{w}(u)\mathbf{1}_{\{\beta(x)\leq \beta(u)\}}du \nonumber \\ \geq \int_0^1c_1(x-y)f_{w}(u)\mathbf{1}_{\{\beta(y)>\beta(u)\}} du +\int_0^1 c_2(x-y)f_{w}(u)\mathbf{1}_{\{\beta(y)\leq \beta(u)\}}du \end{eqnarray} Then, the inequality can be further simplified to the following: \begin{eqnarray} \label{eqn:xy>yx} (x-y) \int_0^1(c_1-c_2)f_{w}(u)(\mathbf{1}_{\{\beta(x)>\beta(u)\}}-\mathbf{1}_{\{\beta(y)>\beta(u)\}})du\geq 0 \end{eqnarray} Since $x>y$ and $\beta(x)<\beta(y)$, $(x-y) \int_0^1(c_1-c_2)f_{w}(u)(\mathbf{1}_{\{\beta(x)>\beta(u)\}}-\mathbf{1}_{\{\beta(y)>\beta(u)\}})du<0$ holds given that the private weight $w$ is uniformly distributed from [0,1], which contradicts inequality \ref{eqn:xy>yx} and the proof stands here. \subsection{Proof of Corollary \ref{cor:mono}}\label{pf:cor:mono} Using the same technique as we used in the proof of Lemma \ref{lem:mono}, we can use induction to generalized to a finite-slots sponsored search markets. Consider the advertisers' private weight are symmetrical distributed from a distribution $D$. Suppose there exists two weight $x>y$ such that $\beta(x)<\beta(y)$, where $\beta (\cdot)$ is the optimal bidding function. Suppose there are $n$ slots and let $u_i$ be other advertiser $i$'s private weight, assuming $\beta(u_i)\geq \beta(u_j)~\forall i>j$ without loss of generality. Define the space $A=\{(u_1,u_2,...,u_{n-1})|\beta(u_i)\geq \beta(u_j)~\forall i<j\}$, $\beta(u_0)=+\infty$, and $c_{n+1}=0$, the two inequalities that the optimal bidding function have to satisfy are as follows: \begin{eqnarray} &&\sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A \bigg\{c_ix-\beta(x)(c_i-c_{i+1})-\sum_{j=i+1}^n \beta(u_{j-1})(c_j-c_{j+1})\bigg\}\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(x)>\beta(u_i)\}} \prod_{k=i}^{n-1}f_D(u_k)du_k \nonumber \\ &\geq& \sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A \bigg\{c_ix-\beta(y)(c_i-c_{i+1})-\sum_{j=i+1}^n \beta(u_{j-1})(c_j-c_{j+1})\bigg\}\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(y)>\beta(u_i)\}} \prod_{k=i}^{n-1}f_D(u_k)du_k \nonumber \end{eqnarray} \begin{eqnarray} &&\sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A \bigg\{c_iy-\beta(y)(c_i-c_{i+1})-\sum_{j=i+1}^n \beta(u_{j-1})(c_j-c_{j+1})\bigg\}\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(y)>\beta(u_i)\}} \prod_{k=i}^{n-1}f_D(u_k)du_k \nonumber \\ &\geq& \sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A \bigg\{c_iy-\beta(x)(c_i-c_{i+1})-\sum_{j=i+1}^n \beta(u_{j-1})(c_j-c_{j+1})\bigg\}\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(x)>\beta(u_i)\}} \prod_{k=i}^{n-1}f_D(u_k)du_k \nonumber \end{eqnarray} Now, we have to sum up the two inequalities above and cancel terms exist in both sides. \begin{eqnarray} \label{eqn:mono} &&\sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A c_i(x-y)\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(x)>\beta(u_i)\}} \prod_{k=i}^{n-1}f_D(u_k)du_k \nonumber \\ &\geq& \sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A c_i(y-x)\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(y)>\beta(u_i)\}} \prod_{k=i}^{n-1}f_D(u_k)du_k \end{eqnarray} We can further simplify the inequality (\ref{eqn:mono}) to be the following inequality: \begin{eqnarray} (x-y)\sum_{i=1}^{n-1} \tbinom{n-1}{k} \oint_A c_i\{\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(x)>\beta(u_i)\}}-\mathbf{1}_{\{\beta(u_{i-1})\geq \beta(y)>\beta(u_i)\}}\}\prod_{k=i}^{n-1}f_D(u_k)du_k \geq 0 \end{eqnarray} Given that $c_i\geq c_j$ for all $i<j$, $x>y$ and $\beta(x)<\beta(y)$, the above inequality is always $\leq 0$ and the equality holds when $F_D(x)-F_D(y)=0$. Therefore, we can claim that the optimal bidding function in sponsored search markets with symmetric advertisers are monotonic increasing with the advertiser's private weight. \subsection{Proof of Lemma \ref{lem_2g1}}\label{sec:lem_2g1} Consider buyer i's bid $b_{i1}$, $b_{i2}$ on good 1 and 2. WLOG, suppose $b_{11},b_{12}\neq 0$. If $b_{11}+b_{22}\geq b_{12}+b_{21}$ (the scenario that player 1 will win good 1). The price of good 1 is $P_1=b_{11}-(b_{12}-b_{22})\mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}}$. Similarly, if $b_{11}+b_{22}< b_{12}+b_{21}$ (the scenario that player 1 will win good 2). The price of good 2 at this time is $P_1=b_{12}-(b_{11}-b_{21})\mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}}$. The buyer 1's surplus is \begin{eqnarray} &&u_1(b_{11},b_{12},b_{21},b_{22})\nonumber \\ &=&(v_{11}-b_{11})\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}} + (v_{12}-b_{12})(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})\nonumber\\ &&+\mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}}\big[(b_{12}-b_{22})\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}}+(b_{11}-b_{21})(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})\big] \nonumber \end{eqnarray} Now, define $c=\min \{b_{11},b_{12}\}$ and consider another bidding strategy $(b^*_{11} , b^*_{12})$ that $(b_{11}^*=b_{11}-c$, $(b_{12}^*=b_{12}-c$. Since reducing same amount of bids on both goods reduces the same amount on $b_{11}+b_{22}$ and $b_{12}+b_{21}$, this new bidding strategy will not change the probability of buyer 1 to win good 1 or 2. Therefore, we can calculate the difference of surplus between these two bidding strategy: \begin{eqnarray} &&u_1(b^*_{11},b^*_{12},b_{21},b_{22})-u_1(b_{11},b_{12},b_{21},b_{22}) \\ &=&(b_{11}-b^*_{11})\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}} + (b_{12}-b^*_{12})(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})\nonumber\\ &&-\mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}}\big[(b_{12}-b_{22})\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}}+(b_{11}-b_{21})(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})\big] \nonumber \\ &=& c[\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}} +(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})] \nonumber \\ &&-\mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}}\big[(b_{12}-b_{22})\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}}+(b_{11}-b_{21})(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})\big] \label{eqn12} \\ &\geq & c- \mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}}\big[c\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}}+c(1-\mathbf{1}_{\{b_{11}+b_{22}\geq b_{12}+b_{21}\}})\big] \label{eqn13} \\ &=& c- c\mathbf{1}_{\{b_{11}\geq b_{21},b_{12}\geq b_{22}\}} \geq 0 \label{eqn14} \end{eqnarray} The most critical part is from (\ref{eqn12}) to (\ref{eqn13}). When $b_{11}+b_{22}\geq b_{12}+b_{21}, b_{11}\geq b_{21}$, and $b_{12}\geq b_{22}$, they imply that $b_{12}\geq b_{22}\leq \min\{b_{11},b_{12}\}=c$. Similarly, $b_{11}+b_{22}\leq b_{12}+b_{21}, b_{11}\geq b_{21}$, and $b_{12}\geq b_{22}$ imply $b_{11}\geq b_{21}\leq \min\{b_{11},b_{12}\}=c$. With Equation (\ref{eqn14}), we can conclude that any strategy placing non-zero bids on both goods is a weakly-dominated strategy. \subsection{Proof of Lemma \ref{lem_2g2}}\label{sec:lem_2g2} If a strategy bid $b_{i1}, b_{i2}$ satisfying $b_{i1}<b_{i2}$ when $v_{i1}>v_{i2}$, consider another strategy bid $b^*_{i1}, b^*_{i2}$ as follows: \begin{eqnarray} &&b^*_{i1}=b_{i1} \nonumber\\ &&b^*_{i2}= \begin{cases} b_{i1}~~~~ \text{if~} v_{i1}>v_{i2} \\ b_{i2}~~~~ \text{otherwise} \end{cases} \end{eqnarray} With Lemma \ref{lem_2g1}, we can assume $b_{i1}=0$. Used the similar technique to compare the buyer's surplus when $v_{i1}>v_{i2}$. \begin{eqnarray} &&u_i(b^*_{i1},b^*_{i2},b_{-i1},b_{-i2})-u_i(b_{i1},b_{i2},b_{-i1},b_{-i2}) = u_i(0,0,b_{-i1},b_{-i2})-u_i(0,b_{i2},b_{-i1},b_{-i2}) \nonumber \\ &=&v_{i1}\mathbf{1}_{\{b_{-i2}\geq b_{-i1}\}} + v_{i2}(1-\mathbf{1}_{\{b_{-i2}\geq b_{-i1}\}})- v_{i1}\mathbf{1}_{\{b_{-i2}\geq b_{i2}+ b_{-i1}\}} - (v_{i2}-b_{i2})(1-\mathbf{1}_{\{b_{-i2}\geq b_{i2}+b_{-i1}\}})\nonumber \\ &=& b_{i2}(1-\mathbf{1}_{\{b_{-i2}\geq b_{i2}+b_{-i1}\}})+(v_{i1}-v_{i2})\mathbf{1}_{\{b_{-i1}+b_{i2}\geq b_{-i2}\geq b_{-i1}\}} \geq 0 \label{eqn16} \end{eqnarray} Since equation (\ref{eqn16}) is non-negative. The lemma is proved. \subsection{Proof of Lemma \ref{pro1}}\label{sec:pro1} Suppose $\beta(v_{ih},v_{il})=b^*>b'=\beta(v_{ih}',v_{il}')$ but $v_{ih}-v_{il}<v_{ih}'-v_{il}'$, consider another bidding function $\beta'$ exchanges the bid between these two pair of valuations and bids the same as $\beta$ otherwise. Suppose the original bidding function is optimal, the following two inequality holds: \begin{eqnarray} &&[(v_{ih}-\beta(v_{ih},v_{il}))\text{Pr(win higher valued good with bid $\beta(v_{ih},v_{il}$))} \nonumber \\ &+&v_{il}\text{Pr(win lower valued good with bid $\beta(v_{ih},v_{il}$))}] \nonumber \\ &-&(v_{ih}-\beta'(v_{ih},v_{il}))\text{Pr(win higher valued good with bid $\beta'(v_{ih},v_{il}$))} \nonumber \\ &-&v_{il}\text{Pr(win lower valued good with bid $\beta'(v_{ih},v_{il}$))} \geq 0 \end{eqnarray} \begin{eqnarray} &&[(v_{ih}'-\beta(v_{ih}',v_{il}'))\text{Pr(win higher valued good with bid $\beta(v_{ih}',v_{il}'$))} \nonumber \\ &+&v_{il}'\text{Pr(win lower valued good with bid $\beta(v_{ih}',v_{il}'$))}] \nonumber \\ &-&(v_{ih}'-\beta'(v_{ih}',v_{il}'))\text{Pr(win higher valued good with bid $\beta'(v_{ih}',v_{il}'$))} \nonumber \\ &-&v_{il}'\text{Pr(win lower valued good with bid $\beta'(v_{ih}',v_{il}'$))} \geq 0 \end{eqnarray} Sum up the above two inequality and we know that $\text{Pr(win higher valued good with bid $\beta'(v_{ih}',v_{il}'$))}=\text{Pr(win higher valued good with bid $\beta(v_{ih},v_{il}$))}$ We will get \begin{eqnarray} && [(v_{ih}-v_{il})-(v_{ih}'-v_{il}')][\text{Pr(win higher valued good with bid $b^*$)} \nonumber\\ &-&\text{Pr(win higher valued good with bid b')}]\geq 0 \end{eqnarray} However, we know $v_{ih}-v_{il})<(v_{ih}'-v_{il}')$ and $b^*>b'$. there's a contradiction. Hence, the optimal bidding function of the higher good $\beta(v_{ih},v_{il})$ is monotonic to $v_{ih}-v_{il}$. \section{Appendix} \subsection{Choice of Constricted Goods Set} \label{sec6.1} Suppose we do not choose the most skewed set in our algorithm, then it follows that we may obtain different MCP vectors. However, a natural question to ask is the following: if we run the algorithm two times and choose different constricted good sets at some iterations such that the bipartite graph produced in every round of these two executions are the same, will we get the same MCP vector? Unfortunately, the choice of the same initial price vector, the same price reduction rule, and the emergence of the same bipartite graph in every round are not enough to guarantee the same returned MCPs. A counterexample is provided in Fig. \ref{fig4}. \begin{figure}[h] \centering \includegraphics[width=0.4\textwidth]{./Figures/constricted.pdf} \caption{Counterexample of same bipartite graph but different set MCPs} \label{fig4} \end{figure} In Fig. \ref{fig4}, the bipartite graphs A-1, B-1 have the same preference graph. Though the chosen constricted good sets in A, B are different, they add the same buyer to the constricted graph. Therefore, the preference graphs in A-2, B-2 are still the same. However, the updated price vector of A-2, B-2 must be different. If A, B choose the same constricted goods set in every round, the returned sets of MCPs of A, B must be different. This example shows that just the bipartite graphs in every round cannot uniquely determine the MCP vector obtained at the termination of the algorithm. \subsection{Proof of Lemma \ref{lemA}}\label{pf:lem1} First, we will show that given a constricted good set $S$, reducing price of every good in $S$ with the amount specified in the statement will increase the size of its neighbor, i.e., $N(S)$. Given a constricted good set $S$ under a specific price vector $\mathbf{P}$. Consider another price vector $\mathbf{P'}$, $P'_j=\begin{cases} P_j , & j \notin S \\ P_j-c, & j \in S \end{cases} $ , where $c:=\min_{i \in \mathcal{B} \setminus N(S),l\in S} \{\max_{k \in \mathcal{M} \setminus S}(v_{i,k}-P_k)- (v_{i,l}-P_l) \}$. For some $j \in S$, there must exists an $i\in \mathcal{B} \setminus N(S)$ satisfying $v_{i,j}-P_j'= \max_{k \in \mathcal{M} \setminus S} v_{i,k}-P_k$. Now, by an abuse of notation to denote $N'(S)$ as the neighbor of $S$ under $P'$, $i \in N'(S)$. For those $l \in N(S)$, $\max_{j \in S}v_{l,j}-P'_j>\max_{j \in S}v_{l,j}-P_j\geq \max_{j \in \mathcal{M}\setminus S}v_{l,j}-P_j$ implies $l \in N'(S)$. Therefore, $|N(S)|<|N'(S)|$. Then, we need to prove that $c$ is the minimum decrement. Consider another price vector Consider another price vector $\mathbf{P''}$, $P'_j=\begin{cases} P_j , & j \notin S \\ P_j-d, & j \in S \end{cases} $, where $d<c$. It is straightforward that for all $i \in \mathcal{B} \setminus N(S)$, $v_{i,j}-P''_j< v_{i,j}-P_j+c \le \max_k (v_{i,k}-P_k)$. Hence, no buyers will be added to the $N(S)$. Now, it is clear that $\min_{i \in \mathcal{B} \setminus N(S),l\in S} \{\max_{k \in \mathcal{M} \setminus S}(v_{i,k}-P_k)- (v_{i,l}-P_l) \}$ is the minimum price reduction which guarantees to add at least a new buyer to the $N(S)$. \subsection{Proof of Lemma \ref{lem2}}\label{pf:lem2} In order to prove the statement, we start from showing the most skewed set is always a constricted good set when there is no perfect matching. Then, we prove the uniqueness of the most skewed set by contradiction. When there is no perfect matching, there must exists a constricted good set. By definition, the constricted good set $S$ has the property $|S|>|N(S)|$. Since $|S|$, $|N(S)|$ are integers, the skewness of a constricted good set \begin{equation} \label{eq1} f(S)=|S|-|N(S)|+\frac{1}{|S|} \geq 1+\frac{1}{|S|} >1. \end{equation} Then, for any non-constricted good set $S'$, $|S'| \leq |N(S')|$. The skewness of $S'$ is \begin{equation} \label{eq2} f(S')=|S'|-|N(S')|+\frac{1}{|S'|} \leq 0+\frac{1}{|S'|} \leq 1. \end{equation} With equation (\ref{eq1}), (\ref{eq2}), the skewness of a constricted good set is always greater than any non-constricted good sets. Therefore, if a preference graph exists a constricted good set, the most skewed set is always a constricted good set. Then, we start to prove the uniqueness of the most skewed set. Suppose there exists two disjoint sets $S_1$, $S_2$ and both sets are the most skewed sets, i.e., $f(S_1)=f(S_2)= \max_{S \subset \mathcal{M}, S \neq \emptyset} f(S)$. Consider the skewness of the union of $S_1$ and $S_2$. \begin{eqnarray} &&f(S_1 \cup S_2) \\ &=& |S_1 \cup S_2|- |N(S_1 \cup S_2)|+ \frac{1}{|S_1 \cup S_2|} \\ &=& |S_1|+ |S_2| - |N(S_1) \cup N(S_2)|+ \frac{1}{|S_1 \cup S_2|} \\ &\geq & |S_1|+ |S_2| - |N(S_1)|- |N(S_2)|+ \frac{1}{|S_1 \cup S_2|} \\ &=& f(S_1)+ |S_2|- |N(S_2)| -\frac{1}{|S_1|}+\frac{1}{|S_1 \cup S_2|} \label{eq3}\\ &\geq & f(S_1) +1 -\frac{1}{|S_1|}+\frac{1}{|S_1 \cup S_2|} > f(S_1) \label{eq4} \end{eqnarray} From (\ref{eq3}) to (\ref{eq4}) is true because $S_2$ is a constricted good set. (\ref{eq4}) contradicts our assumption that $S_1$, $S_2$ are the most skewed set. Therefore, we know that if there exists multiple sets share the same highest skewness value, these sets are not disjoint. Now, suppose there exists two sets $S_1$, $S_2$ satisfying that $S_1 \cup S_2 \neq \emptyset$ and both sets are the most skewed sets, the following two inequalities must hold: \begin{eqnarray} f(S_1)-f(S_1 \cup S_2) \geq 0 \\ f(S_2)-f(S_1 \cap S_2) \geq 0 \end{eqnarray} Let's sum up the two inequalities and represent the formula in twelve terms. \begin{eqnarray} &&f(S_1)+f(S_2) - f(S_1 \cup S_2)-f(S_1 \cap S_2) \\ &=& |S_1|+|S_2|-|S_1 \cup S_2|-|S_1 \cap S_2| \nonumber \\ && +|N(S_1 \cup S_2)|+ |N(S_1 \cap S_2)|- |N(S_1)|-|N(S_2)| \nonumber \\ && +\dfrac{1}{|S_1|}+\dfrac{1}{|S_2|}-\dfrac{1}{|S_1 \cup S_2|}-\dfrac{1}{|S_1 \cap S_2|} \end{eqnarray} The first four terms $|S_1|+|S_2|-|S_1 \cup S_2|-|S_1 \cap S_2|=0$. Using the similar argument, $|N(S_1)|+|N(S_2)|=|N(S_1) \cap N(S_2)|+|N(S_1) \cup N(S_2)|$. \begin{eqnarray} |N(S_1 \cup S_2)|=|N(S_1) \cup N(S_2)| \\ |N(S_1 \cap S_2)| \leq |N(S_1) \cap N(S_2)| \label{eq5} \end{eqnarray} Equation (\ref{eq5}) is true because there may exist some elements in $S_1 \setminus S_2$ and $S_2 \setminus S_1$ but have common neighbors. Thus, the second four terms are smaller than or equal to $0$. To check the last four terms, let $|S_1|=a$, $|S_2|=b$, and $|S_1 \cap S_2|=c$, where $c<\min\{a,b\}$ because $S_1,S_2$ are not disjoint. The last four terms are \begin{eqnarray} &&\dfrac{1}{|S_1|}+\dfrac{1}{|S_2|}-\dfrac{1}{|S_1 \cup S_2|}-\dfrac{1}{|S_1 \cap S_2|} \\ &=& \dfrac{1}{a}+\dfrac{1}{b}-\dfrac{1}{a+b-c}-\dfrac{1}{c} \\ &=& \dfrac{a+b}{ab}-\dfrac{a+b}{(a+b-c)c} \\ &=& \dfrac{a+b}{abc(a+b-c)}(ac+bc-c^2-ab) \\ &=&-\dfrac{(a+b)(a-c)(b-c)}{abc(a+b-c)}<0 \end{eqnarray} To conclude, the first four terms are $0$, the second four terms are smaller than or equal to $0$, and the last four terms are strictly negative make \begin{eqnarray} f(S_1)+f(S_2) - f(S_1 \cup S_2)-f(S_1 \cap S_2)<0. \end{eqnarray} Therefore, at least one of set $S_1 \cup S_2$, $S_1 \cap S_2$ has the skewness value greater than $f(S_1)=f(S_2)$, which leads to a contradiction that $S_1$ and $S_2$ are the most skewed set. Finally, we can claim that the most skewed set is unique when there is no perfect matching. \subsection{Proof of Theorem \ref{lem4.1}}\label{pf:lem4.1} First, let's begin with the proof of the first half statement of the theorem, which is a variational characterization of MCPs. ($\Rightarrow$) This direction is obvious, otherwise the MCP is not the maximum by definition. ($\Leftarrow$) Recall that $\mathcal{M}$ is the set of goods. Suppose there exists an MCP $P^1$ satisfying the conditions but it is not the maximum MCP. Then there must exist a set of goods $S_1$ such that for all $i \in S_1$, $P^1_i<P^*_i$, and for all $i \in \mathcal{M}-S_1$, $P^1_i \geq P^*_i$; $\mathcal{M}-S_1$ can be an empty set. Let $P^2_i= P^1_i$ for all $i \in \mathcal{M}-S_1$, and $P^2_i= P^*_i$ for all $i \in S_1$. We will verify that $P^2$ is an MCP. WLOG, we can assume $P^1$ and $P^*$ have the same allocation; this is true as every MCP supports all efficient matchings. Then, consider any buyer who is assigned a good in $S_1$ under $P^*$. When the price vector changes from $P^*$ to $P^2$, the buyer has no profitable deviation from his/her assigned good because $P^2_i \geq P^*_i$ for all $i$ and $P^2_j= P^*_j$ for all $j \in S_1$. Similarly, when the price vector changes from $P^1$ to $P^2$, the buyer who is assigned a good in $\mathcal{M} \setminus S_1$ under $P^1$ has no profitable deviation because $P^2_i=P^1_i$ for all $i \in \mathcal{M} \setminus S_1$ and $P^2_j>P^1_j$ for all $j \in S_1$. Finally, since $P^1$ and $P^*$ have the same allocation, no buyer will deviate if we assign this allocation to buyers under $P^2$. Since all buyers have non-negative surpluses under $P^2$, it follows that $P^2$ is an MCP. Given $P^1$, there exists a set of goods $S_1$ whose price we can increase and still get market clearing because both $P^1$ and $P^2$ are MCPs. This contradicts the assumption that $P^1$ satisfies the stated conditions, and the proof of the variational characterization follows. For the second half part, which is a combinatorial characterization of MCPs, let's try to prove the statement using the result of the variational characterization. Given that we cannot increase the price for any subset of goods, it implies that for any subset of goods $S$, the set of corresponding matched buyer, either there exists at least one buyer has an edge connected to a good not in this subset or there exists at least one buyer with surplus zero. In the former case, it is obvious that $B<N(B)$ for such corresponding buyer set $B$. In the later case, $B\leq N(B)$ and $D\in N^D(B)$ guarantees $B < N^D(B)$. For the opposite direction, if every set of buyer with $B < N^D(B)$ under the current MCP, increasing the price for any set of good will make at least one corresponding buyer deviate from the matched good and cause no perfect matching. Therefore, the condition in variational characterization holds if and only if the condition in the combinatorial characterization holds. \subsection{Proof of Theorem \ref{thm:externality}}\label{pf:thm:externality} First, it is obvious that the current market adding a duplicate pair of buyer buyer $i$ and its matched good $j$ is still market clearing and the social welfare will be the $\text{(the current social welfare)}+U^*_i+P_j$. Hence, what we need to show is the current market adding a duplicate buyer buyer $i$ has the social welfare: $\text{(the current social welfare)}+U^*_i$. Since adding a dummy good will not change the social welfare, we can transfer our problem to prove that the current market adding a duplicate buyer buyer $i$ and a dummy good has the social welfare $\text{(the current social welfare)}+U^*_i+0$. It is equivalent to show that the current market adding a duplicate buyer buyer $i$ and a dummy good is still market clearing. Using the combinatorial characterization of the maximum MCP, we know that $|B|<|N^D(B)|$ for any subset of B in the current market. Denote the buyer set of current market as $\mathcal{B}$ and the duplicate buyer $i$ as $\hat{i}$ , what we want to show is that $|B|\leq |N^D(B)|$ for any $B \in \{\mathcal{B}\cup\hat{i}\}$. If at least one of $i, \hat{i}$ not in $B$, it is straightforward that $|B|<|N^D(B)|$. If both $i, \hat{i}\in B$, $|B|=|B\setminus \hat{i}|+1 \leq |N^D(B\setminus \hat{i})-1|+1=|N^D(B\setminus \hat{i})|\leq |N^D(B)|)$. Using the Hall's marriage theorem, the inequality guaranteed that the current market adding a duplicate buyer buyer $i$ and a dummy good is market clearing, and the proof stands here. \subsection{Proof of Theorem \ref{thm2}}\label{pf:thm2} As mentioned earlier, showing the algorithm returns the maximum MCP is equivalent to show that when the algorithm terminates, the bipartite structure guarantees that we can not increase the price of any subset of good with the result in Theorem \ref{lem4.1}. It implies that after adding a dummy good with zero price, the support of any subset of goods has a size less than the size of the subset of goods, i.e., $|B|< |N^D_T(B)|$. Therefore, the proof of the theorem can be transformed to prove that the algorithm satisfies $|B|< |N^D_T(B)|$ for any non-empty set of buyer $B$ on termination. Let's start the proof with several claims. \begin{claim} \label{cl1} For any subset of buyer $B$, $B\neq \emptyset$, $|B|\leq |N^D_T(B)|$, where $T$ is the terminating time of our algorithm. \end{claim} Since our algorithm returns an MCP vector, if $|B|> |N^D_T(B)|$, there does not exist a perfect matching because of the existence of constricted buyer set. \begin{claim} \label{cl2} There does not exists a subset of buyer $B\subseteq \mathcal{B}$, $B\neq \emptyset$, $|B|= |N^D_T(B)|$ and $D\in N^D_T(B)$, where $T$ is the terminating time of our algorithm. \end{claim} $|B|= |N^D_T(B)|$ and $D\in N^D_T(B)$, imply $|B|> |N_T(B)|$. The preference graph has constricted sets and has no perfect matching. \begin{claim} \label{cl3} There does not exists a subset of buyer $B\subseteq \mathcal{B}$, $B\neq \emptyset$, $|B|= |N_T(B)|$, where $T$ is the terminating time of our algorithm. \end{claim} \begin{proof} With Claim \ref{cl1}, \ref{cl2}, it is equivalent to show $|B|<|N_T(B)|$. At time $t$, we denote the maximum non-negative surplus of buyer $b$ by $u^*_t(b)$ and the most skewed set by $S^*_t$; and $\mathscr{B}^s_t$ is the set of buyers with positive surplus at time $t$, i.e., $\mathscr{B}^s_t=\{b|u^*_t(b)>0, b\in\mathcal{B}\}$. It is obvious that $\mathscr{B}^s_i \subseteq \mathscr{B}^s_j$ for all $i<j$. Then, we want to prove the claim by mathematical induction. Prior to the proof, we need to introduce another claim. \begin{claim} \label{cl4} $\forall B\subseteq N_t(S^*_t), 0\leq t<T$, $|B|<|N_t(B) \cap S^*_t)|=|N_{t+1}(B)|$ \end{claim} \begin{proof} If the left inequality does not hold, we can remove all the goods contained in $N_t(B) \cap S^*_t$ from $S^*_t$ to get a more skewed set, which violates that $S^*_t$ is the most skewed one. Since we reduce the price in $S^*_t$, buyers in $B$ will not prefer any good outside of $S^*_t$. Therefore, $N_t(B) \cap S^*_t$ and $N_{t+1}(B)$ are identical. Hence $|N_t(B) \cap S^*_t|=|N_{t+1}(B)|$ is absolutely true. \end{proof} Now, we can prove Claim \ref{cl3} by induction. At t=0, $\mathscr{B}^s_0=\emptyset$. At t=1, $\mathscr{B}^s_1=\mathcal{N_0}(S^*_0)$. With Claim \ref{cl4}, $|B|<|N_1(B)|~~\forall B\subseteq \mathscr{B}^s_1$ and $B \neq \emptyset$. At a finite time $t$, suppose for all $B\subseteq \mathscr{B}^s_t$, $B \neq \emptyset$ satisfy $|B|<|N_t(B)|$, consider at time $t+1$: Since $\mathscr{B}^s_t \subseteq \mathscr{B}^s_{t+1}$, $\mathscr{B}^s_{t+1}$ contains three disjoint components: \begin{equation} \mathscr{B}^s_{t+1}= \{\mathscr{B}^s_t \cap {N_t(S^*_t)}^c \} \cup \{\mathscr{B}^s_t \cap N_t(S^*_t) \} \cup \{{\mathscr{B}^s_t}^c \cap N_t(S^*_t) \} \end{equation} Buyers in the first two parts are originally with positive utilities. Buyers in the last part have zero utilities at time $t$ but have positive utilities at time $t+1$. Consider the subset of buyers $B_{\alpha} \subseteq \{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$. Since every $b \in B_{\alpha}$ does not prefer any good in $S^*_t$, the price reduction in $S^*_t$ will never remove any edges between $\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$ and $N_t(\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \})$. Therefore, for any non-empty set of buyers $B_{\alpha}$, $|B_{\alpha}|< |N_t(B_{\alpha})|\leq |N_{t+1}(B_{\alpha})|$. Then, consider the second and the third parts. Since $\{\mathscr{B}^s_t \cap N_t(S^*_t) \} \cup \{{\mathscr{B}^s_t}^c \cap N_t(S^*_t) \}=N_t(S^*_t)$, every non-empty set of buyers $B_{\beta} \subseteq N_t(S^*_t)$ satisfies $|B_{\beta}|<|N_{t+1}(B_{\beta})|$ by Claim \ref{cl4}. At the last step, consider $B_{\gamma}=B_{\gamma_1} \cup B_{\gamma_2}$, where $B_{\gamma_1} \subseteq \{\mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$, $B_{\gamma_2} \subseteq N_t(S^*_t)$ and $B_{\gamma_1},B_{\gamma_2} \neq \emptyset$. \begin{eqnarray} |N_{t+1}(B_{\gamma})| &=&|N_{t+1}(B_{\gamma_1}) \cup N_{t+1}(B_{\gamma_2})| \\ &=& |N_{t+1}(B_{\gamma_1})|+|N_{t+1} (B_{\gamma_2})|-|N_{t+1}(B_{\gamma_1}) \cap N_{t+1}(B_{\gamma_2})| \\ &=& |N_{t+1}(B_{\gamma_1})|-|N_{t+1}(B_{\gamma_1}) \cap N_{t+1}(B_{\gamma_2})|+|N_{t+1} (B_{\gamma_2})| \\ &=& |N_{t+1}(B_{\gamma_1}) \cap {N_{t+1}(B_{\gamma_2})}^c|+|N_{t+1} (B_{\gamma_2})| \\ &\geq& |N_{t+1}(B_{\gamma_1}) \cap {N_{t+1}(N_t(S^*_t))}^c|+|N_{t+1} (B_{\gamma_2})| \label{eqx1} \\ &=& |N_{t+1}(B_{\gamma_1}) \cap {S^*_t}^c|+|N_{t+1} (B_{\gamma_2})| \label{eqx2}\\ &=& |N_t(B_{\gamma_1})|+|N_{t+1} (B_{\gamma_2})| \label{eqx3}\\ &>& |B_{\gamma_1})|+|B_{\gamma_2}|=|B_{\gamma}| \end{eqnarray} From (\ref{eqx1}) to (\ref{eqx2}) is true because after price reduction at time $t$, buyer belongs to the neighbor of constricted good set $S^*_t$ will only prefer goods in $S^*_t$, therefore $N_{t+1}(N_t(S^*_t))=S^*_t$. From (\ref{eqx2}) to (\ref{eqx3}) is true because $N_t(B_{\gamma_1}) \cap S^*_t = \emptyset$ (by definition). Then, we have discussed before that the price reduction in $S^*_t$ will never remove any edges between $\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \}$ and $N_t(\{ \mathscr{B}^s_t \cap {N_t(S^*_t)}^c \})$. Since goods not in the most skewed set at time $t$ will not add new buyers to their neighbor, the equivalence between $N_{t+1}(B_{\gamma_1}) \cap {S^*_t}^c$ and $N_t(B_{\gamma_1})$ holds. Therefore, for every non empty set of buyers $B\subseteq \mathscr{B}^s_{t+1}$, $|B|<|N_{t+1}(B)|$. The mathematical induction works for all $k \in \mathbb{N}$. Since our algorithm terminates in finite round, $|B|< |N_T(B)|$, Q.E.D. \end{proof} With Claim \ref{cl1}, \ref{cl2}, \ref{cl3}, the skew-aided algorithm satisfies $|B|<|N^D_t(B)|$ for any non-empty set of buyer $B$ on termination. \subsection{Proof of Lemma \ref{lem:conv}}\label{pf:lem:conv} First, we prove that the algorithm terminates in finite (at most $|\mathcal{M}|^3$) rounds by investigating the relationship of the most skewed sets in the consecutive rounds, $S^*_t$, $S^*_{t+1}$. The relationship of $S^*_t$, $S^*_{t+1}$ has four cases. \begin{enumerate} \item $S^*_t=S^*_{t+1}$ \item $S^*_t \subset S^*_{t+1}$ \item $S^*_t \supset S^*_{t+1}$ \item $S^*_t \nsubseteq S^*_{t+1}$ and $S^*_t \nsupseteq S^*_{t+1}$ \end{enumerate} Recall that $W(G)$ is the skewness of the graph $G$ and $N_t(S)$ is the neighbor of $S$ based on the preference graph at round $k$. \\ In case 1, $W(G_t)-W(G_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_t)\geq 1$. \\ In case 2, define $S'=S^*_{t+1}\setminus S^*_t$, and it is trivial that $S^*_t \subset S^*_{t+1}$ implies $1>\frac{1}{|S^*_t|}-\frac{1}{|S^*_{t+1}|}>0$. \begin{eqnarray} &&W(G_t)-W(G_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_{t+1})\\ &=&|S^*_t|-|N_t(S^*_t)|-|S^*_{t+1}|+|N_{t+1}(S^*_{t+1})| +\dfrac{1}{|S^*_t|}-\dfrac{1}{|S^*_{t+1}|} \label{eq6}\\ &\geq &|S^*_{t+1}|-|N_t(S^*_{t+1})|-|S^*_{t+1}|+|N_{t+1}(S^*_{t+1})|+\dfrac{1}{|S^*_t|}-\dfrac{1}{|S^*_{t+1}|} \label{eq7}\\ &=&|N_{t+1}(S^*_t \cup S')|-|N_t(S^*_{t+1})|+\frac{|S^*_{t+1}|-|S^*_t|}{|S^*_t||S^*_{t+1}|} \nonumber \\ &\geq &|N_{t+1}(S^*_t \cup S')|-|N_t(S^*_{t+1})| + \dfrac{1}{|\mathcal{M}|(|\mathcal{M}|-1)} \label{eq8}\\ &=&|N_{t+1}(S^*_t)|+|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c| -|N_t(S^*_{t+1})|+ \dfrac{1}{|\mathcal{M}|^2-|\mathcal{M}|} \label{eq9}\\ &=&|N_t(S^*_t)|+|N_{t+1}(S_t)\setminus N_t(S^*_t)|- |N_t(S^*_{t+1})| +|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c|+ \tfrac{1}{|\mathcal{M}|^2-|\mathcal{M}|} \label{eq10} \end{eqnarray} Before going to further steps, we have to briefly explain the logic behind the above equations. (\ref{eq6}) to (\ref{eq7}) is true because $N_t(S^*_t) \subseteq N_t(S^*_{t+1})$. (\ref{eq8}) to (\ref{eq9}) is to expand $|N_{t+1}(S^*_t \cup S')|$ to $|N_{t+1}(S^*_t)|+|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c|$. (\ref{eq9}) to (\ref{eq10}) is to expand $|N_{t+1}(S^*_t)|$ to $|N_t(S^*_t)|+|N_{t+1}(S_t)\setminus N_t(S^*_t)|$. Since $|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c|=|N_{t+1}(S')\cap N_{k}(S^*_t)^c|-|N_{t+1}(S')\cap (N_{t+1}(S_t)\setminus N_t(S^*_t))|$, and $|N_{t+1}(S_t)\setminus N_t(S^*_t)|\geq |N_{t+1}(S')\cap (N_{t+1}(S_t)\setminus N_t(S^*_t))|$, we can further summarize the first four terms in (\ref{eq10}). \begin{eqnarray} &&|N_t(S^*_t)|+|N_{t+1}(S_t)\setminus N_t(S^*_t)| +|N_{t+1}(S')\cap N_{t+1}(S^*_t)^c| - |N_t(S^*_{t+1})| \\ &\geq & |N_t(S^*_t)|+|N_{t+1}(S')\cap N_{k}(S^*_t)^c| -|N_t(S^*_{t+1})|\\ &=& |N_t(S^*_t)|+|N_{t+1}(S')\cap N_t(S^*_t)^c| -|N_t(S^*_t)|-|N_t(S')\cap N_t(S^*_t)^c| \\ &=& |N_{t+1}(S')\cap N_t(S^*_t)^c|-|N_t(S')\cap N_t(S^*_t)^c| \label{eq32} \\ &=& |N_t(S')\cap N_t(S^*_t)^c|-|N_t(S')\cap N_t(S^*_t)^c|=0 \label{eq33} \end{eqnarray} (\ref{eq32}) to (\ref{eq33}) is true because $S'$ is not in $S^*_t$, the neighbor of $S'$ not in the neighbor of $S^*_t$ remains the same from $k$ to $t+1$ round. With equation (\ref{eq10}) and (\ref{eq33}), we can conclude that $W(G_t)-W(G_{t+1})\geq \frac{1}{|\mathcal{M}|^2-|\mathcal{M}|}$ in case 2.\\ \\ In case 3, since every elements in $S^*_t$ belongs to $S^*_t$, $f_t(S^*_{t+1}) \geq f_{t+1}(S^*_{t+1})$. Given that $\dfrac{1}{|S^*_t|}<\dfrac{1}{|S^*_{t+1}|}$ and $f_t(S^*_t)-f_t(S^*_{t+1})>0$ by definition, $|S^*_t|-|S^*_{t+1}|+|N_t(S^*_t)|-|N_t(S^*_{t+1})|\geq 1$. With the knowledge that $S^*_t$ and $S^*_t$ are non-empty set, it is obvious that $f_t(S^*_t)-f_t(S^*_{t+1})$ is lower-bounded by $\tfrac{1}{2}$. Therefore, we can conclude that $W(G)_t-W(G_{t+1})=f_t(S^*_t)-f_{t+1}(S^*_{t+1}) \geq f_t(S^*_t)-f_t(S^*_{t+1}) \geq \frac{1}{2}$ in case 3.\\ \\ In case 4, define $S'=S^*_{t+1} \setminus S^*_t$, $S''=S^*_t \setminus S^*_{t+1}$, and $T=S^*_t \cap S^*_{t+1}$. \begin{eqnarray} && W(G_t)-W(G_{t+1}) \\ &=&f_t(S^*_t)-f_{t+1}(S^*_{t+1}) \\ &=& |T|+|S''|-|N_t(T)|-|N_t(S'')\setminus N_t(T)|+\frac{1}{|S^*_t|} - |T|-|S'|+|N_{t+1}(S^*_{t+1})|-\frac{1}{|S^*_{t+1}|} \\ &=& |S''|-|N_t(S'')\setminus N_t(T)|+\frac{1}{|S^*_t|}-\frac{1}{|S^*_{t+1}|} +|N_{t+1}(S^*_{t+1})|-|N_t(T)|-|S'| \\ &\geq & 1+\frac{1}{|S^*_t|}-\frac{1}{|S^*_{t+1}|}+|N_{t+1}(S^*_{t+1})| -|N_t(T)|-|S'| \\ &\geq & 1+\frac{1}{|\mathcal{M}|}-1+|N_{t+1}(T))|-|N_t(T)|+|N_{t+1}(S')\setminus N_{t+1}(T)|-|S'| \\ &= & \frac{1}{|\mathcal{M}|}+|N_{t+1}(T))|-|N_t(T)| +|N_{t+1}(S')\setminus N_{t+1}(T)|-|S'| \label{eq11}\\ &\geq & \frac{1}{|\mathcal{M}|}+|N_{t+1}(S')\setminus N_{k}(T)|-|S'| \geq \frac{1}{|\mathcal{M}|} \label{eq12}\\ \end{eqnarray} Most of the equations in case 4 are straight-forward except from (\ref{eq11}) to (\ref{eq12}). (\ref{eq11}) to (\ref{eq12}) is true because of the following inequalities: \begin{eqnarray} &&|N_{t+1}(S')\setminus N_{t+1}(T)|+|N_{t+1}(T))|-|N_t(T)| \nonumber \\ &=&|N_{t+1}(S')\setminus N_t(T)|+|N_{t+1}(T))|-|N_t(T)|-| \{N_{t+1}(S') \cap N_{t+1}(T) \}\setminus N_t(T)| \\ &\geq & |N_{t+1}(S')\setminus N_t(T)|-| N_{t+1}(T)\setminus N_t(T)| +|N_{t+1}(T))|-|N_t(T)| \\ &=& |N_{t+1}(S')\setminus N_t(T)| \end{eqnarray} Combine these four cases, we know that \begin{eqnarray} W(G_t)-W(G_{t+1}) \geq \dfrac{1}{ |\mathcal{M}|^2-|\mathcal{M}|}. \end{eqnarray} Therefore, we can conclude that the proposed algorithm terminates in finite rounds. (At most $|\mathcal{M}|^3$ rounds because $W(G)<|\mathcal{M}|$ and minimum decrement is greater than $\tfrac{1}{|\mathcal{M}|^3}$). \subsection{Proof of Lemma \ref{lem:alg2}}\label{pf:lemalg2} We will prove Algorithm \ref{alg:alg2} always return the most skewed set by contradiction.\\ Since every untraversed good upon termination can be matched with an untraversed buyer without repetition. Therefore, adding any set of runtraversed goods $S_U$ to the set $S$ will always reduce the skewness of $S$ (because the increase of cardinality in neighbor of $S$, $|N(S_U)|-|N(S)|$ is always greater than or equal to the increase of cardinality of $S$, $|S_U|-|S|$). Hence, the most skewed set will never contain any untraversed good. Suppose there exists a set $S'$ is the most skewed set, with a higher skewness than the set $S$ return by Algorithm \ref{alg:alg2}, $S'$ must be a subset of $S$ because there's no untraversed good in the most skewed set and $f(S')>f(S)$. Let $S^*=S\setminus S'$, $f(S')>f(S)$ and $S' \subset S$ implies $|N(S^*)\setminus N(S')|-|S^*|>0$. If this happens, there must be a non-empty set of matching pairs match nodes from $S^*$ to $N(S^*)\setminus N(S')$ without repetition. Since $N(S^*)\setminus N(S')$ is not in $N(S')$ under the directed graph, nodes in $S^*$ will not be traversed in the algorithm contradicts that $S^* \subset S$. \subsection{Proof of Theorem \ref{thm1}}\label{pf:thm1} First, it is straightforward that the algorithm terminates in at most $|\mathcal{M}|^3$ rounds (because $W(G)<|\mathcal{M}|$ and minimum decrement is greater than $\tfrac{1}{|\mathcal{M}|^3}$). However, only $|\mathcal{M}|^2$ positive distinct values of $W(G)$ are feasible because there are only $|\mathcal{M}|$ possible values on $|S^*_t|-|N(S^*_t)|$ and $|\mathcal{M}|$ possible values on $\frac{1}{|S^*_t|}$. Given that the sequence $W(G_t)$ is strictly decreasing w.r.t. $t$, there are at most $|\mathcal{M}|^2$ rounds of iteration in Algorithm \ref{alg:alg1}. Second, in Algorithm \ref{alg:alg2}, the Hopcroft-Karp algorithm runs in time $O(|\mathcal{M}|^{2.5})$, and the complexity of the BFS algorithm is $O(|\mathcal{M}|+|\text{number of directed edges}|)$. Since the number of directed edges is upper-bounded by $|\mathcal{M}|^2$, the complexity of Algorithm \ref{alg:alg2} is $O(|\mathcal{M}|^{2.5}+|\mathcal{M}|^2)=O(|\mathcal{M}|^{2.5})$. Therefore, Algorithm \ref{alg:alg1} has complexity upper-bounded by $O(|\mathcal{M}|^{2.5}\times|\mathcal{M}|^2)=O(|\mathcal{M}|^{4.5})$. \begin{algorithm}[H] \caption{Algorithm in search of the maximally skewed set by coloring preference graph} \begin{algorithmic}[1] \Require A colored preference graph, a set of buyers $A$ would be added to the neighbor of the previous maximally skewed set \Ensure An updated colored preference graph \If {Input preference graph is not colored} \State Run Hopcroft-Karp algorithm to find a maximum matching \State Color every edges connecting a matched pair red and all other edges blue, Color every vertex in a matched pair red and all other vertice blue \State Starting from $\mathcal{M}_b$, run the rb-BFS to get a reachable set $R(\mathcal{M}_b)$ \State Color $R(\mathcal{M}_b)\setminus \mathcal{M}_b$ green \Else \State Update the colored preference graph by removing all edges connecting red goods and green buyers and adding corresponding blue edges connecting goods and buyers in $A$ \While {$\{A\cap\mathcal{B}_b\} \neq \emptyset$} \State Pick an element $a$ in $\{A\cap\mathcal{B}_b\}$ \If {$|N(a) \cap \mathcal{M}_b|>1$} \State Pick arbitrary $x \in \{N(a)\cap \mathcal{M}_b\}$, color $(a,x) \in E$ red and color $a,x$ green. \ElsIf {$|\{N(a) \cap \mathcal{M}_b\}|=1$} \State Let $x$ be the unique good in $\{N(a)\cap \mathcal{M}_b\}$, color $(a,x) \in E$ red and color $a,x$ green. \If{$\{N(a)\cap \mathcal{M}_g\}\neq \emptyset$} \State Run the rb-DFS starting from $x$ in $G(\mathcal{M}_g \cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{gb})$ to get a reachable set $R(x)$, then color vertice in $R(S)$ red if $R(x)\cap \mathcal{M}_b = \{x\}$ \Else \State Run the br-DFS starting from $a$ in $G(\mathcal{M}_g \cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{gb})$ to get a reachable set $R(x)$, then color vertice in $R(S)$ red if $R(x)\cap \mathcal{M}_b = \{x\}$ \EndIf \ElsIf {$\{N(a) \cap \mathcal{M}_g\} \neq \emptyset$} \State Run the rb-DFS starting from $a$ in $G(\mathcal{M}_g\cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{gb})$ till find the first $x \in \mathcal{M}_b$ \State Color $a,x$ green and switch the color of every edge used in a path from $a$ to $x$. \State Start a br-BFS from $\mathcal{M}_b$ in $G(\mathcal{M}_g\cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{g})$ to get a reachable set $R(\mathcal{M}_b)$. \State Color every vertex in $\{\mathcal{M}_g \cup \mathcal{B}_g\}\setminus R(\mathcal{M}_b)$ red \EndIf \State Remove $a$ from $A$ \EndWhile \If {$A \neq \emptyset$} \State Run the br-BFS starting from $\{\mathcal{M}_g\}$ in $G(\mathcal{M},\mathcal{B}, E)$ to get a reachable set $R(S^*)$ \If {$\{R(S^*) \cap \mathcal{B}_b\}=\emptyset$} \State Color all vertice in $R(S)\setminus \mathcal{M}_b$ green \Else \While{$\{R(S^*) \cap \mathcal{B}_b\}\neq\emptyset$} \State Pick $a \in R(S^*) \cap \mathcal{B}_b$ and run rb-DFS starting from $a$ to get the reachable set $R(a)$ \If {$\exists x\in R(a), x \in \mathcal{M}_b$} \State Color $a,x$ green; switch the color of every edge used in a path from $a$ to $x$ \State Start a br-BFS from $\mathcal{M}_b$ in $G(\mathcal{M}_g\cup \mathcal{M}_b,\mathcal{B}_g, E^{gb}_{g})$ to get a reachable set $R(\mathcal{M}_b)$. \State Color every vertex in $\{\mathcal{M}_g \cup \mathcal{B}_g\}\setminus R(\mathcal{M}_b)$ red \EndIf \EndWhile \State Run the br-BFS starting from $\{\mathcal{M}_b\}$ in $G(\mathcal{M},\mathcal{B}_g, E)$ to get a reachable set $R(\mathcal{M}_b)$ \State Color all vertice in $\{\mathcal{M}_g\cup \mathcal{B}_g\}\setminus R(\mathcal{M}_b)$ red \EndIf \EndIf \EndIf \end{algorithmic} \label{alg:alg3} \end{algorithm} \section{Conclusions and Future Work} \label{sec7} In this paper, we proposed a descending price algorithm in search of sets of the maximum MCPs by exploiting the combinatorial structure of bipartite graphs in matching markets. The algorithm terminates in at most $m^2$ rounds for any non-negative valuation matrix with runtime $O(m^4)$. There are three main avenues for future work. First, we would like to determine whether one can reduce the complexity further to $O(m^3)$ mirroring the Hungarian algorithm. Second, as incentive compatibility does not hold with the maximum MCP, we would like to determine the equilibrium bidding strategy in a Bayesian Nash equilibrium given the proposed mechanism. This will be necessary for expected revenue computation, and for a comparisons with the VCG mechanism, GSP and laddered auction proposed in \cite{GSP,ladder}, and also other mechanisms such as the GFP auction. Finally, many real-world applications of matching markets outside of the online advertising setting are not unit-demand~\cite{oviedo2005theory} and obtaining a combinatorial version of the descending price auction returning the maximum MCP is a challenging open problem for future work. \section{Discussions} \label{sec6} We now discuss some interesting points of the choices made in our algorithm, some nuances in terms of the final bipartite graphs, and a few comments on the non-monotonicity of bidding with strategic concerns. \subsection{Choice of Constricted Goods Set} \label{sec6.1} Suppose we do not choose the most skewed set in our algorithm, then it follows that we may obtain different MCP vectors. However, a natural question to ask is the following: if we run the algorithm two times and choose different constricted good sets at some iterations such that the bipartite graph produced in every round of these two executions are the same, will we get the same MCP vector? Unfortunately, the choice of the same initial price vector, the same price reduction rule, and the emergence of the same bipartite graph in every round are not enough to guarantee the same returned MCPs. A counterexample is provided in Fig. \ref{fig4}. \begin{figure}[h] \centering \includegraphics[width=0.4\textwidth]{./Figures/constricted.pdf} \caption{Counterexample of same bipartite graph but different set MCPs} \label{fig4} \end{figure} In Fig. \ref{fig4}, the bipartite graphs A-1, B-1 have the same preference graph. Though the chosen constricted good sets in A, B are different, they add the same buyer to the constricted graph. Therefore, the preference graphs in A-2, B-2 are still the same. However, the updated price vector of A-2, B-2 must be different. If A, B choose the same constricted goods set in every round, the returned sets of MCPs of A, B must be different. This example shows that just the bipartite graphs in every round cannot uniquely determine the MCP vector obtained at the termination of the algorithm. \subsection{Failure of Connectivity in Matching Markets} \label{sec6.3} In general matching markets, the set of goods in the bipartite graph is not always connected, even at the maximum MCP. A disconnected structure in bipartite graph implies a set of buyers' bids/values may play no role on another set of buyers' payment in the MCP. In other words, a set of buyers will be indifferent when another sets of buyers bids/values are within specific regions; this phenomenon happens when sets of buyers have greater affinity for some pre-determined types of goods, i.e., valuation of men and women on ties and high heels would satisfy such a property. In sponsored search markets, however, the set of goods in the initial bipartite graph is connected\footnote{Every good is preferred by the buyer with the highest bid.}. Furthermore, the following lemma shows that our skewed-set aided descending price algorithm will preserve the connectivity property in all iterations, and therefore also at termination. \begin{lem} \label{lem8} In sponsored search markets with the skewed-set aided descending price algorithm, any set of goods in every bipartite graph that is encountered during the execution of Algorithm \ref{alg:alg1} is connected. \end{lem} \begin{proof} Please see Appendix~\ref{sec:lem8}. \end{proof} In Lemma~\ref{lem8}, we addressed the connectivity property in sponsored search markets during the execution of our algorithm. Unfortunately, the connectivity does not always hold in general matching markets. Here we demonstrate this by a $4$-by-$4$ counter example below. To make this case as simple as possible, we just consider a realized valuation matrix \begin{eqnarray*} V= \begin{bmatrix} 5 & 4 & 1 & 1 \\ 3 & 3 & 2 & 2 \\ 2 & 2 & 3 & 3 \\ 1 & 1 & 4 & 5 \end{bmatrix}. \end{eqnarray*} Initially, the first two goods are preferred by buyer 1, and the last two goods are preferred by buyer buyer 4. In our algorithm, the most skewed set contains all goods, and price is reduced by one in the first round. After that, the MCP vector $\mathbf{P}=[4,3,3,4]$ and the bipartite graphs of initial and final steps are in Fig. \ref{fig:connectivity}. \begin{figure}[h] \centering \includegraphics[width=0.4\textwidth]{./Figures/connectivity.pdf} \caption{An example where both initial and final bipartite graphs are disconnected} \label{fig:connectivity} \end{figure} Fig. \ref{fig:connectivity} demonstrates that both the initial and final bipartite graphs are disconnected in our algorithm as there are two components. Therefore, preserving the connectivity property in the running of our algorithm holds in sponsored search markets, and may not otherwise. \subsection{Further Comments on Strategic Bidding} Unlike most of the non-incentive compatible auction mechanisms that overbidding above buyers' actual value is always irrational, a buyer may benefit from bidding higher than her actual value to enjoy a lower price under the proposed algorithm in some scenarios. We can show this by a simple example below. First, we present two valuations matrices that will yield the same matching but different prices with the difference in the matrices only being the valuations of one buyer. These are \begin{eqnarray*} V_1= \begin{bmatrix} 4 & 5 & 5 & 6 \\ 2 & 4 & 5 & 5 \\ 1 & 2 & 4 & 5 \\ 0 & 1 & 2 & 4 \end{bmatrix}, \quad V_2 = \begin{bmatrix} 4 & 5.9 & 5.9 & 6.9 \\ 2 & 4 & 5 & 5 \\ 1 & 2 & 4 & 5 \\ 0 & 1 & 2 & 4 \end{bmatrix}. \end{eqnarray*} The maximum MCPs are $P_1=[1, 2, 3, 4]$ and $P_2=[0.1,2,3,4]$, respectively, and in both cases buyer $i$ gets matched to good $i$ for all $i\in \{1,\dotsc,4\}$. Now if we assume that the true valuations of all the buyers is given by matrix $V_1$, then by unilaterally deviating to a higher bid, buyer $1$ is able to dramatically increase her payoff without impacting any other buyers. While this is only an example, we believe that this will occur with non-zero probability and will necessitate an alternate means of determining the equilibrium bidding behavior. \section{Preliminary Analysis of Strategic Buyers and Bayesian-Nash Equilibrium} \label{sec5} As mentioned in the introduction, the proposed descending price algorithm is not incentive compatible, and so we need to explore Bayesian Nash equilibrium (BNE) to predict buyers' strategic behaviors. From the messaging viewpoint we will assume that we have a direct mechanism where the buyers bid their valuation: a scalar in sponsored search markets, and a vector in the general matching market. Algorithm \ref{alg:alg1} is then applied as a black-box to the inputs to produce a price on each good and a perfect matching. With expected revenue being an important concern, we provide an instance achieving higher expected revenue than VCG mechanism in an asymmetric BNE. Furthermore, a case provided to show the possibility of overbidding. Additional analyses of BNE in two simple cases with symmetric buyers and the associated bidding strategy of the buyers are provided in Appendices \ref{apdx:analysis} and \ref{apdx:analysis2}. \subsection{An Instance Achieving higher expected revenue than VCG}\label{sec:greatRev} Given the proposed algorithm, it is important to check if there exists a scenario that our algorithm obtain a higher expected revenue than the VCG mechanism. There is a large body of literature discussing the failure of getting the VCG revenue under asymmetric distribution of buyers' valuations, e.g., see \cite{VCGfailure}. Additionally, the well-known revenue equivalence theorem holds with an assumption that valuation of every player is drawn from a path-connected space in Chap. 9.6.3 of \cite{SSMbook}\footnote{Two other assumptions are (1) both mechanisms implement the same social welfare function; (2) for every player $i$, there exists a type that the expected payments are the same in both mechanisms}. With this knowledge, we demonstrate a $3\times 3$ matching market, where the buyers have asymmetric distributions of their valuations, for which our mechanism achieves higher than VCG revenue. Consider three advertisers named Alice, Bob, and Carol, and three different types of ads called listing ads, sidebar ads, and pop-ups. The realized valuation is only known to the advertiser (equivalently buyer), but the distribution of an advertisers' valuation is known to other advertisers but not the auctioneer. In other words, the auctioneer can only calculate the price according to the bids submitted by the advertisers. The minimum increment of the submitted bids is $\epsilon$, which is a positive infinitesimal, and the valuation matrix of advertisers is displayed in Table \ref{tbl:1}. \begin{table}[h] \centering \caption{Valuation matrix of advertisers} \label{tbl:1} \begin{tabular}{|l|l|l|l|} \hline & Listing & Sidebar & Pop-ups \\ \hline Alice & W & 0 & 0 \\ \hline Bob & X & 0 & 0.5 \\ \hline Carol & Y & Z & 2 \\ \hline \end{tabular} \\ \begin{tabular}{|ll|} \hline Probability density function of $W,X,Y,Z$ &\\ \hline $f_W(w)=\begin{cases} \frac{2}{3} &~w\in [0,1) \\ \frac{1}{3} &~w\in (2,3] \\ 0 &~\text{o/w} \end{cases} $ & $f_X(x)=\begin{cases} \frac{2}{3} &~x\in [1.5,2.5] \\ \frac{1}{3} &~x\in (2.5,3.5] \\ 0 &~\text{o/w} \end{cases}$ \\ $f_Y(y)=\begin{cases} 2 &~y\in [3.5,4] \\ 0 &~\text{o/w} \end{cases}$ &$f_Z(z)=\begin{cases} 1 &~z\in [3,4] \\ 0 &~\text{o/w} \end{cases} $ \\ \hline \end{tabular} \end{table} Now, we give an asymmetric BNE\footnote{Multiple equilibria may exist.} in this matching market, with a detailed verification in the Appendix \ref{asymBNE}. First, assume that Alice always bids 0 on sidebar ads and pop-ups, and bids $\max \{1-\epsilon, \frac{w}{2}\}$ on listing ads. The best response of Bob is to bid 0 on both sidebar ads and pop-ups, and to bid $\max \{1, \frac{x-0.5}{2}\}$ on listing ads for any realized $x$. Now, given Bob's bidding function as above, one of Alice's best responses is to follow her original bidding function. Last, consider the bidding function of Alice and Bob mentioned above, a best response of Carol is to bid 0 on listing ads and pop-ups, and to bid $\epsilon$ on sidebar ads regardless of the outcomes of $y$ and $z$. \footnote{The $\epsilon$ is designed to avoid complex tie-breaking rules.}, even if $y>z$. This is because Carol will never win listing ads for any bids less than $1$ as the probability $\text{Pr}\{y-z \geq 1\}=0$. Now, given Carol's bidding strategy, Alice and Bob will not change their bidding functions. Therefore, the strategy $\{\beta_{\mathrm{Alice}}(w,0,0), \beta_{\mathrm{Bob}}(x,0,0.5),\beta_{\mathrm{Carol}}(y,z,2)\}=\{(\max \{1-\epsilon, \frac{w}{2}\},0,0), (\max \{1, \frac{2x-1}{4}\},0,0),(0,\epsilon,0)\}$ can be verified to be an asymmetric BNE. With the asymmetric BNE in hand, we want to calculate the expected revenue of the auctioneer and compare it with the expected revenue of the VCG mechanism. Since Carol always wins the sidebar ads and both Alice and Bob bid 0 on that, Carol will pay $\epsilon$. Additionally, in the asymmetric BNE, Alice and Bob compete on the listing ads and both bid 0 on sidebar ads and pop-ups, resulting in the payment of listing ads to be the same as the payment in the first price auction. Using this the expected revenue of the auctioneers is given by \begin{eqnarray} &&\epsilon+\frac{4}{9}\int_0^1\int_{1.5}^{2.5} 1 dxdw+\frac{2}{9}\int_2^3\int_{1.5}^{2.5} \frac{w}{2} dxdw+\frac{2}{9}\int_1^2\int_{2.5}^{3.5} \frac{2x-1}{4} dxdw \nonumber \\&+&\frac{1}{9}\int_2^3\int_{2.5}^{w+0.5} \frac{w}{2} dxdw+\frac{1}{9}\int_{2.5}^{3.5}\int_2^{x-0.5} \frac{2x-1}{4} dwdx =\dfrac{31}{27}+\epsilon \end{eqnarray} The last step is to calculate the expected revenue of the VCG mechanism, which is given by \begin{eqnarray} &&\frac{2}{3}\int_0^1\int_{1.5}^{2.5} w dxdw+\frac{2}{9}\int_2^3\int_{1.5}^{2.5} (x-0.5) dxdw+\frac{1}{9}\int_{2.5}^{3.5}\int_2^{x-0.5} w dwdx \nonumber \\&+&\frac{1}{9}\int_2^3\int_{2.5}^{w+0.5} (x-0.5) dxdw =\dfrac{1}{3}+\dfrac{2}{9}\times \dfrac{3}{2}+\dfrac{1}{9}\times \dfrac{7}{6}++\dfrac{1}{9}\times \dfrac{7}{6}=\dfrac{25}{27} \end{eqnarray} Even if we set $\epsilon$ to 0, it is obvious that the expected revenue derived under our descending price auction algorithm is strictly greater than the expected revenue of the VCG mechanism. This shows that in some instances the proposed descending price algorithm is preferred to the (DGS) ascending price algorithm, even taking the strategic behavior of buyers into account. \subsection{Further Comments on Strategic Bidding} Unlike most of the non-incentive compatible auction mechanisms where overbidding above buyers' actual value is always irrational, in some scenarios a buyer may benefit from bidding higher than her actual value to enjoy a lower price in the proposed algorithm. We show this by a simple example below. First, we present two valuations matrices that will yield the same matching but different prices with the difference in the matrices only being the valuations of one buyer. These are \begin{eqnarray*} V_1= \begin{bmatrix} 4 & 5 & 5 & 6 \\ 2 & 4 & 5 & 5 \\ 1 & 2 & 4 & 5 \\ 0 & 1 & 2 & 4 \end{bmatrix}, \quad V_2 = \begin{bmatrix} 4 & 5.9 & 5.9 & 6.9 \\ 2 & 4 & 5 & 5 \\ 1 & 2 & 4 & 5 \\ 0 & 1 & 2 & 4 \end{bmatrix}. \end{eqnarray*} The maximum MCPs are $P_1=[1, 2, 3, 4]$ and $P_2=[0.1,2,3,4]$, respectively, and in both cases buyer $i$ gets matched to good $i$ for all $i\in \{1,\dotsc,4\}$. Now if we assume that the true valuations of all the buyers is given by matrix $V_1$, then by unilaterally deviating to a higher bid on three goods, buyer $1$ is able to dramatically increase her payoff without impacting any other buyer. While this is only an example, we believe that this will occur with non-zero probability and will necessitate an alternate means of determining the equilibrium bidding behavior.
37b1241402dc2d669283f6b66eb4e2f5c59730bf
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{section1} The majority of stars are born in large clusters, and so, these sites are key to understanding the stellar contents of galaxies. Until recently the Milky Way was believed to be devoid of large young clusters. The realisation that our Galaxy harbours many young clusters with masses (M\,$>$\,10$^3$\,M$_{\odot}$) like Westerlund~1 (Wd~1), Arches, Quintuplet, RSG~1, RSG~3, Stephenson~2, Mercer~81, NGC3603, h+$\chi$ Persei, Trumpler~14, Cygnus~OB2, [DBS2003] and VdBH~222 \citep[see summary by][]{Negueruela2014}, reveals a different scenario than previously thought. Although Milky Way offers the opportunity to resolve the cluster members, unlike other galaxies beyound the Local Group, accurate determination of the fundamental properties of these recently discovered clusters: distance, mass, age, initial mass function (IMF), and binary fraction is still lacking in many cases. The two fundamental parameters upon which all others depend are the interstellar extinction and the distance. In the present work we use accurate techniques to derive the interstellar extinction towards Wd~1 and its surroundings, what was not accurately done in previous works. The extinction is related to the observed magnitudes by the fundamental relation (distance modulus): \begin{equation} m_{\lambda} = M_{\lambda} + 5\log_{10} \left(\frac{d}{10}\right) + A_{\lambda}. \label{eq1} \end{equation} A set of different filters can be combined to define colour excess indices: $\displaystyle E_{\lambda1}-E_{\lambda2}=(m_{\lambda1}-m_{\lambda2})_{\rm{obs}}-(m_{\lambda1}-m_{\lambda2})_0$, where the zero index indicates the intrinsic colour of the star. Some authors -- like the classical work of \citet{Indebetouw+2005} -- attempted to derive $\displaystyle A_\lambda$ directly from the above relations, using a minimization procedure to a large number of observations from 2MASS survey \citep{Stru06}. Actually the number of variables is greater than the degrees of freedom of the system of equations we have to compute. All possible colour excess relations are linear combinations of the same parameters. In the specific case of the NIR, after dividing by the K$_s$--band extinction, there will be an infinite number of pairs $\displaystyle A_J/A_{Ks}$ and $\displaystyle A_H/A_{Ks}$ satisfaying the relations. Many minimization programs (like the downhill technique used in the {\it amoeba} program) just pick up a local minimum close to the first guess as a solution, hiding the existence of other possibilities. Derivation of the extinction can only be accomplished on the basis of a specific extinction law as a function of wavelength, which ultimately reflects the expected properties of dust grains. Interstellar reddening is caused by scattering and refraction of light by dust grains, and the amount of light subtracted from the incoming beam (extinction) is governed by ratio between its wavelength ($\lambda$) and the size of the grains (d). For $\lambda<<d$ all photons are scattered/absorbed (gray extinction).For $\lambda>d$ the fraction of photons that escape being scattered/absorbed increases. The interstelar dust is a mixture of grains of different sizes and refactory indices, leading to a picture a little more complicated than described above. This was first modelled by \citet{Hulst1946} for different dust grain mixtures. All subsequent observational works resulted in optical and NIR extinction laws similar to that of van de Hulst model (in particular his models \#15 and \#16). A remarkable feature is that they are well represented by a power low ($A_\lambda$\,$\propto$\,$\lambda^{-\alpha}$) in the range 0.8\,$<$\,$\lambda$\,$<$\,2.4\,$\mu$m \citep[see e.g.][]{F99}. The $\alpha$ exponent of the extinction power law: $\displaystyle A_{\lambda}/A_{Ks} = (\lambda_{Ks}/\lambda)^\alpha$ is related to the observed colour excess through: \begin{equation} \frac{A_{\lambda_1}-A_{\lambda_2}}{A_{\lambda_2}-A_{\lambda_{Ks}}} = \frac{\left(\frac{\lambda_2}{\lambda_1}\right)^{\alpha}-1}{1-{\left(\frac{\lambda_{Ks}}{\lambda2}\right)^{\alpha}}}. \label{eq2} \end{equation} The value of the $\alpha$ exponent is driven by: a) the specific wavelength range covered by the data; b) the effective wavelengths of the filter set, which may differ from one to another photometric system, especially the $R$ and $I$ bands; and c) the fact that the effective wavelength depends on the spectral energy distribution (SED) of the star, the transmission curve of the filter and the amount of reddening of the interstellar medium (ISM). The power law exponent {\large $\alpha$} in the range 1.65\,$<$\,$\alpha$\,$<$\,2.52 has been reported \citep[see e.g.,][]{CCM89, Berdnikov+1996, Indebetouw+2005, Stead+2009, RL85, FM09, Nishiyama+2006, GF14}, but it is not clear how much spread in the value of the exponent is due to real physical differences in the dust along different lines-of-sight and how much comes from the method used on the determination of the exponent. As shown by \citet[their Fig. 5]{GF14} using the 2MASS survey, the ratio of colour excess $E_(H-K)/E_(J-K)$ grows continuously from 0.615 to 0.66 as the distance modulus grows from 6 to 12 towards the inner Galactic Plane (GP, their Fig. 5). This corresponds to a change in $\alpha$ from 1.6 to 2.2 which translates into $A_J/A_{Ks}$ from 2.4 to 3.4. \citet{Zas09} also used 2MASS data to show that colour excess ratios varies as a function of Galactic longitude, indicating increasing proportions of smaller dust grains towards the inner GP. Reddening laws steeper than the''canonical” ones have been suggested for a long time, but their reality is now clearly evident from deep imaging surveys. The large progress reached in recent years revealed that there is no ``universal reddening law'', as believed in the past. Moreover, the extinction coefficients are quite variable from one line--of--sight to another, even on scales as small as arcminutes. At optical bands ($UBV$) it was already well established long ago that the extinction law for particular lines-of-sight had large discrepancies from the average \citep{FM09,He+1995,Popowski2000,Sumi2004,Racca+2002}. In recent years this has been proved to occur also for the NIR wavelength range \citep{Larson+2005, Nishiyama+2006, Froebrich+2007, Gosling+2009}. The patchy tapestry of extinction indices is particularly impressive in the large area work in the NIR$/$optical done by \citet{Nataf16,Nataf13} for the Galactic Bulge and by \citet{Scha16} for the GP. Although we cannot use directly the extinction coefficients from these works for our particular target (line--of--sight), their derived reddening relations help for checking the consistency of our results. Targets to derive the extinction must be selected between stars with well known intrinsic colour indices, in order to measure accurate colour excesses. Wd~1 cluster members with known spectral types are ideal for this, especially because the majority of them are hot stars, for which the intrinsic colours are close to zero. There are $\approx$ 92 stars in this group; the statistics can be improved by using stars in the field around the cluster. \citet{Wozniak+1996} proposed using Red Clump (RC) stars, to derive the ratio between the total to selective extinction. These Horizontal Branch stars are the most abundant type of luminous stars in the Galaxy and have a relatively narrow range in absolute colours and magnitudes. RCs form a compact group in the colour-magnitude diagram (CMD) as shown by \citet[and references therein]{Stanek+2000}. This is due to the low dispersion -- a few tenths of magnitude -- in intrinsic colours and luminosities of RC stars \citep{Stanek+1997,Paczynski+1998}. This technique, initially designed for using filters $V$ and $I$ in OGLE survey for microlensing events, and filters $V$ and $R$ in MACHO, was adapted for the $JHKs$ bands \citep{Flaherty+2007,Indebetouw+2005,Nishiyama+2006,Nishiyama+2009}. As shown by \citet{Indebetouw+2005} and by \citet{Nishiyama+2006,Nishiyama+2009}, RC stars in the CMD (e.g. $J-Ks$ {\it versus} $Ks$) may appear as an over--density ``strip''. That strip in the CMD contains interlopers which mimic RC star colours but have different luminosities (like nearby red dwarfs and distant supergiants). This does not allow the application of the relation~\eqref{eq1} to derive the absolute extinction from each particular star in the strip, but still works for the colour excess ratios in the relation~\eqref{eq2}. From the measured colour excess ratio (e.g. $E_{J-H}/E_{J-Ks}$) the value of the exponent $\alpha$ can be calculated and, therefore, the ratios $A_J/A_{Ks}$ and $A_H/A_{Ks}$. \citet{Nishiyama+2006} reported $\alpha$\,=\,1.99 and $A_J$\,$\approx$\,3.02 and $A_H$\,$\approx$\,1.73 in a study of the Galactic Bulge, which is much higher then all previous results. \citet{Fritz+2011} also derived a large value $\alpha$\,=\,2.11 for the Galactic Centre (GC), using a completely different technique. \citet{Stead+2009} reported $\alpha$\,=\,2.14\,$\pm$\,0.05 from UKIDSS data and similar high exponents from 2MASS data. They did not derive $A_\lambda/A_{Ks}$, since in their approach those quantities vary because of shifts in the effective wavelengths as the extinction increases (see Section 2.2). However, at a first approximation, using the isophotal wavelengths, we can calculate from their extinction law: $A_J/A_{Ks}$\,$\approx$\,3.25 and $A_H/A_{Ks}$\,$\approx$\,1.78. The dream of using interstellar DIBs to evaluate the exinction has been hampered by saturation effects in the strength of the features and on the behaviour of the carriers which differ from the hot/diffuse ISM as compared to cold/dense clouds -- however, see \citet{Maiz+2015}. The 8620\,\AA\,DIB correlates linearly with the dust extinction \citep{Munari+2008}, at least for low reddening, and is relatively insensitive to the ISM characteristics. Since this spectral region will be observed by GAIA for a large number of stars up to relatively large extinction, we used our data to extend \citet{Munari+2008} relation, which was derived for low reddening. This work is organised as follows. In Section~\ref{section2} we describe the photometric and spectroscopic observations and data reduction. In Section~\ref{section3} we describe the colour excess ratios relations, the ratio between the absolute magnitudes and a suggested extinction law for the inner GP. In Section~\ref{section4} we compare our results with others reported in the literature. In Section~\ref{section5} we perform $J-{Ks}$ extinction maps around Wd~1 field for a series of colour slices and evaluate the 3D position of obscuring clouds. In Section~\ref{section6} we analyse the relation between the interstellar extinction and the Equivalent width (EW) of the 8620~\AA\ DIB. In Section~\ref{section7} we present our conclusions. \section{Observations and data reduction}\label{section2} \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.7cm 0cm 2cm, clip]{./RCstrip_dens.png}} \caption{The $J-Ks$ vs. $Ks$ Hess diagram for Wd~1 field showing the RC strip. The two lower continuous lines correspond to extinction laws by \citet{Indebetouw+2005} (blue) - with extinction modules $\mu_{Ks}$\,=\,0.15\,mag\,kpc$^{-1}$ - and by \citet{Nishiyama+2006} (red) - with $\mu_{Ks}$\,=\,0.1\,mag\,kpc$^{-1}$. The two dashed red linear are empirical limits containing the RC strip, using Nishiyama's law with lower limit $\mu_{Ks}$\,=\,0.11 and upper limit 0.23 mag\,kpc$^{-1}$.} \label{fig:RCstrip} \end{figure} \subsection{Photometry}\label{section2.1} The main set of images in the $JHKs$ filters was taken on 2006 June 06 with the ISPI camera at the 4m Blanco Telescope (CTIO), with 10\arcmin\,$\times$\,10\arcmin\ FOV \citep{Bliek+2004}. The image quality, after combination of the dithered sub-images was 1.0\arcsec in the $Ks$ filter and 1.8\arcsec in $J$ and $H$. The number of stars with measurements in all the three filters was 23834 with errors less than 0.5\,mag. The limiting magnitudes were: $J$\,$<$\,21.4, $H$\,$<$\,19.2 and $Ks$\,$<$\,18.2. The basic calibration (flatfield, linearity correction) was done with {\tt IRAF}\footnote{http://www.iraf.noao.edu} and the photometry extraction was performed with \texttt{STARFINDER}\footnote{http://www.bo.astro.it/StarFinder/paper6.htm}. In order to cover the brighter stars, we took $JHKs$ imaging at the 1.6-m and 0.6-m OPD/LNA Brazilian telescopes with the CAMIV camera. Many images were taken in several subsequent years, covering a FOV larger than that of ISPI. For each star, we always use the three $JHKs$ magnitudes taken in a single night, to preserve the stellar colours against variability. In addition to traditional imaging, we used different strategies to record the very bright stars, for example, deploying a 5\,mag neutral density filter or masks with four 5\,cm diameter holes in front of the telescope. We compared the results obtained with the different setups and found good agreement. Instead of the $Ks$ filter, we used a narrow filter with central wavelength 2.14\,$\mu$m (FWHM\,=\,0.02\,$\mu$m) and cross calibrated the photometry with $Ks$ magnitudes from the ISPI camera. The uncertainty on our $JHKs$ magnitudes reached at most 0.1\,mag for a few bright stars like W26, when comparing magnitudes at different epochs, which could be due to intrinsic variability. Data processing was made in the same way as with ISPI images. Our catalogue in $JHKs$ from these two cameras is $\approx$\,100\% complete for the blue and red giants and blue supergiants/hypergiants down to the bottom of the Main Sequence (MS, $Ks$\,$\sim$\,15). The peak of the luminosity function in the $Ks$ filter (completeness $>$\,85\%) is at $Ks$\,=\,16.0). We call this set of data the ISPI+CAMIV photometric sample. The calibration of ISPI photometry was performed against {\tt 2MASS}\footnote{http://irsa.ipac.caltech.edu/cgi-bin/Gator/}. Then, we used the CTIO/ISPI catalogue and stars in common with OPD/CAMIV photometry performed at smaller telescopes to extend 2MASS calibrations up to $Ks$\,$\approx$\,0. The central 2\arcmin\,$\times$\,2\arcmin field of Wd~1 is so crowded that some stars were resolved only in nights with excellent seeing. The Spartan camera at the SOAR telescope was used for this purpose on 2013 September 03 when the seeing was approximately 0.5\arcsec. We used the VVV survey for $ZY$ magnitudes of the RC candidates we detected with ISPI and CAMIV cameras. The magnitudes were extracted using aperture photometry from the database of the survey\footnote{http://horus.roe.ac.uk/vsa/}. The date of those images was 2010 July 19th. For those specific stars, we used our ISPI JHKs magnitudes (calibrated in 2MASS system) to transform from VISTA magnitdes to 2MASS. However, even if the $JHKs$ magnitudes obtained with VVV survey were in excellent agreement with our ISPI+CAMIV+Spartan catalogue, we used them for two specific aims: a) to perform the colour index density maps and b) for checking the correct identification of the ISPI/CAMIV sources. We used magnitudes for the W1 and W2 MIR bands of the WISE survey \citep{wright10}. We used also images from Herschel Hi--GAL survey to look for cold and warm dust \citep{molinari10}. For the Wd~1 cluster members, we used $BVI$ photometry from \citet{Lim+2013}. A few stars without $B$ or $V$ magnitudes from those authors were taken from \citet[]{Clark+2005}, who kindly made a machine readable copy of the photometry available to us. We recalibrated those magnitudes to \citet{Lim+2013} system. Additional magnitudes were taken from \citet{Bonanos2007} and also transformed to the \citet{Lim+2013} system. For the $R$ filter, we used the data from \citet{Clark+2005} and \citet{Bonanos2007} without any transformation. We used 104 stars with spectral classification done by \citet{Clark+2005}, \citet{Negueruela+2010} and \citet{Ritchie+2009}, which have accurate intrinsic colour indices. For reddening studies, we included also the Main Sequence OB eclipsing binary Wddeb \citep{Bonanos2007}. The majority of the bright cluster members are OB Supergiants, and we included also cooler giants and hypergiants. We excluded W8a, W9, W12a, W20, W26, W27, W42a, W75, W265 and the WC9 WRds which have circumstellar hot dust emission (in the $Ks$ filter) or absorption by dust (in $B$ filter). \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.65cm 0cm 2cm, clip]{./EJHXEJK.pdf}} \caption{The excess colour ratio is $E_{J-H}/E_{J-Ks} = 0.655 \pm 0.001$. RCs are in red points (averaged in groups of 10) and Wd~1 cluster members are blue crosses and were used together RCs in the linear fit. It is clear that there is a differential extinction among the Wd~1 cluster members and their average value is intermediate as compared to the RC sample. See other plots in Appendix B.} \label{fig:EJHXEJK} \end{figure} There have been discussions about the accuracy of the optical data by the cited authors, since the colour indices of the program stars were larger than that of photometric standards. Errors could amount up to a few tenths of magnitudes in the B filter and a little less at longer wavelengths. This has an impact when comparing magnitudes from different authors, but has a negligeable effect for other purposes, like study of variability in time series taken with the same instrument. The impact is minor regarding the study of reddening due to the fact that the extinction is orders of magnitudes larger than photometric errors. Similar systematic errors due to the lack of very red photometric standards is a common place in large deep surveys, and, although they are present at all wavelength windows, they have not been discussed in connection to deep surveys. We used extensively the R\footnote{https://www.R-project.org/} programming packages to perform statistics, fitting and plots. \subsection{Effective filter wavelengths}\label{section2.2} Since JHKs filters are close in wavelengths, special care must be taken, using accurate effective wavelengths for this particular range. As a matter of fact every single star is measured at different wavelength, defined by the intrinsic SED of the star, modified by its particular amount of reddening and convolved with the filter passband. This explains why the colour--colour diagram (CCD) produces curved bands. The effective wavelength of the filters shifts continuously to longer wavelengths as the reddening increases, such that the ratios vary continuously. The shifts in the effective wavelengths are approximately straight lines (see Fig.~2a in \citet{Stead+2009}), in order that ratios between effective wavelengths are reasonably constant for large extinctions. Linear approximations for the effective wavelengths of 2MASS filters taken from that plot are presented in Apendix\,\ref{appendixA}. Although a colour plot $J-$H\,{\it vs.}\,$H-Ks$ does not produce a straight reddening line, the curvature of the line is small for large reddening and the ratio converges to a constant value. In this way, we could use a template star SED for evaluating the effective filter wavelengths. An approach simpler to the one we discussed in the last paragraph was also used by \citet{Stead+2009}, who convolved the 2MASS filter passbands with a \citet{Castelli+2004} K2III giant star spectrum, obtaining $\lambda_J$\,=\,1.244, $\lambda_H$\,=\,1.651 and $\lambda_{Ks}$\,=\,2.159. Since our data are from deep imaging towards the inner Galaxy and are calibrated in the 2MASS system, we use the above effective wavelengths for $JHKs$ filters (Table\,\ref{table1}, column~2). Other filter wavelengths were taken as they are repported in the parent papers and not shifted, as we do not use them for critical calculations. Care must be taken because the same filter names have different wavelengths in different papers, especially $R$, $I$, $Z$ filters. For example, the $R$ filter can have $\lambda_{\rm eff} = 0.664$ or 0.70 $\micron$. For the $I$ filter, the most used in recent works is $\lambda_{\rm eff}$\,=\,0.805\,$\mu$m, but \citet{CCM89} uses 0.90\,$\mu$m, which in reality matches better the $Z$ band. To avoid confusion, we state in Table\,\ref{table1} the effective wavelength of each filter we are using. \subsection{Spectroscopy}\label{section2.3} For the 8620 \AA DIB, we observed 11 bright members from the Wd~1 cluster and 12 along other Galactic directions at the Coud\'e focus of the 1.6-m telescope. The OPD/LNA spectra have $R$\,=\,15000. In addition, we downloaded 31 spectra from the ESO database (see Table\,\ref{table2}) also with $R$\,$\sim$\,15000. OPD data were reduced with IRAF; the ESO prog. ID 073D-0327 data with ESO Reflex and ESO 081D-0324 with Gasgano. The spectral resolution was measured on telluric absorption lines. When there were stars in both samples, we used the OPD/LNA because its Deep Depletion CCD presented no fringes, which are prominent in some ESO spectra, especially to the red of the 8620\,DIB. EWs were measured by Gaussian fitting after the spectra were normalised to the stellar continuum. Typical uncertainty is $\sim$\,10\% for the bright and blue stars, but errors are difficult to asses for fainter stars. EWs were also difficult to measure in spectral types later than F5, because of blends with stellar lines. Our results are presented in Table\,\ref{table2} column~2. One specific star, the eclipsing binary Wddeb was measured with Gemini South/GMOS with spectral resolution $R$\,$\sim$8000. \section{The reddening law using colour excesses }\label{section3} When the intrinsic colours of a star are known, its colour excesses can be derived from the observed colours, and the reddening law is calculated from the ratios between colour excesses. This is the case for the Wd~1 cluster members with spectral classification reported in \citet{Clark+2005}, \citet{Negueruela+2010} and \citet{Ritchie+2009}. The intrinsic colours were derived using calibrations by \citet{Wegner1994}. For the Wolf-Rayet stars, we used \citet{Crowther+2006}. Since the number of Wd~1 cluster members are not large, additional stars were taken from the surrounding field. They do not have spectral classification, but for RC stars there are reliable intrinsic colour indices (see below). \subsection{Selection of Red Clump stars}\label{section3.1} A similar procedure to that of \citet{Wozniak+1996} was adopted for selecting RC candidates as they can easily be identified as an overdensity in the CMD (see Fig.\,\ref{fig:RCstrip}). We represented the photometry by a Hess diagram in order to see more clearly the overdensity structures. The lower left overdensity corresponds to foreground (low luminosity) red stars. The one that goes up vertically, around $J-Ks \approx 1.5$ is the Main Sequence (MS) of the Wd~1 cluster and the large region which curves to the red below $Ks\approx 14$ is the cluster Pre-Main Sequence (PMS). The overdensity region running brighter/bluer to fainter/redder in the middle of the CMD is the expected locus containing RC stars observed at different distances and amounts of extinction. For a preliminary study, we plotted in our observed $Ks$\,$\times$\,$J-Ks$ CMD the position of a typical RC star by varying its distance and using different extinction laws like those of \citet{Indebetouw+2005}, who reported a ``extinction modulus'' $\mu_{Ks}$\,=\,$A_{Ks}/d$\,=\,0.15\,mag\,kpc$^{-1}$ - blue line. We followed the same procedure using the reddening law of \citet{Nishiyama+2006}, who do not report the ``extinction modulus'', but which can be derived from the RC peak in their Fig.\,2, corresponding to $\mu_{Ks}$\,=\,0.1\,mag\,kpc$^{-1}$. Although we do not use the extinction modulus for any particular measurement (since it varies from star to star) Fig.\,\ref{fig:RCstrip} indicates that typical extinction in the direction of Wd~1 cluster is much larger than for directions explored by \citet{Indebetouw+2005} - large Galactic longitudes, or by \citet{Nishiyama+2006} - the Bulge. \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.5cm 0.5cm 1.5cm, clip]{./allaws_dif.pdf}} \caption{Differences between $A_\lambda/A_{Ks}$ for a set of reddening laws and that of \citet{Hulst1946}, taken as a zero point.The vertical magenta line is the range reported by \citet{Nataf16}. Data assigned as Schlafly16 \citep{Scha16} in reality were derived by us, using their colour excess ratios. The letters at the top of the plot indicate the approximate wavelengths of the filters - see Table\,\ref{table1}.} \label{fig:allaws_dif} \end{figure} The empirical limits of the RC strip in the CMD could be defined by eye, as done in many works, but we have a more objective way to do it. We used Nishiyama's law with $\mu_{Ks}$\,=\,0.11 and 0.23\,mag\,kpc$^{-1}$ to encompass the RC overdensity - dashed lines in Fig.\,\ref{fig:RCstrip}. The fact that the RC strip crosses the Wd~1 MS, indicates that there are intruders in the overdensity strip. Many of those intruders can be excluded on the basis of their colours in the CCD. We used Nishyiama's law to evaluate the colour excesses and got the average $E_{J-Ks}/E_{H-Ks}$\,=\,2.9\,$\pm$\,0.07 for the RC candidates inside the strip. We used this average and excluded objects with measurements departing more than 3 $\sigma$, keeping the ones in the range 2.7\,$<$\,$E_{J-Ks}/E_{H-Ks}$\,$<$\,3.1, which resulted in 8463 stars with colours compatible with RCs, with $JHKs$ magnitudes. Of course, stars with similar colours but different absolute magnitudes (red dwarfs, K giants) remain in the RC sample. However, they work exactly in the same way as RC stars regarding the measurement of colour excess and the reddening law. For the RC candidates, we obtained $BVIZYW1W2$ magnitudes from the published catalogues described in Section\,\ref{section2.1}. The final number of RCs is: 26 in $B$, 64 in $V$, 205 in $I$, 1439 in $Z$, 1395 in $Y$, 8463 in $JHKs$ and 256 in $W1,W2$. We combine this RC sample with the 105 Wd~1 cluster members which have published spectral types and $BVRIJHKs$ magnitudes. Intrinsic colours for RC stars were taken from Table 1 in \citet{Nataf16} for the optical and NIR and from \citet{GF14} for WISE - their Table\,1 - interpolating the W1 and W2 effective wavelengths reported by \citet{Scha16} - their Table\,2. Absolute magnitudes were calculated by adopting $M_{I,RC}$\,=\,$-0.12$ from \citet{Nataf13}. \subsection{Colour excess ratios combining Red Clump stars and Wd~1 cluster members}\label{section3.2} The colour excess ratios $E_{J-H}\,/\,E_{J-Ks}$ are presented in Fig.\,\ref{fig:EJHXEJK} displays the colour excess ratio for RCs (red crosses) and Wd~1 cluster members (blue crosses) and the linear fit relating these quantities. Similar fit for other colour excess ratios are presented in Appendix\,\ref{AppendixC}. We present below the set of colour excess ratios, combining data from RCs and the Wd~1 cluster members. \begin{eqnarray} E_{B-J} & = & (8.167\pm0.263) ~E_{J-Ks}, \label{eq3} \\ E_{V-J} & = & (5.257\pm0.181) ~E_{J-Ks}, \label{eq4} \\ E_{R-J} & = & (3.597\pm0.185) ~E_{J-Ks}, \label{eq5} \\ E_{I-J} & = & (2.465\pm0.032) ~E_{J-Ks}, \label{eq6} \\ E_{Z-J} & = & (1.797\pm0.012) ~E_{J-Ks}, \label{eq7} \\ E_{Y-J} & = & (0.841\pm0.004) ~E_{J-Ks}, \label{eq8} \\ E_{J-H} & = & (0.655\pm0.001) ~E_{J-Ks}, \label{eq9} \\ E_{H-Ks} & = & (0.345\pm 0.001) ~E_{J-Ks}, \label{eq10} \\ E_{J-H} & = & (1.891\pm0.001) ~E_{H-Ks}, \label{eq11} \\ E_{W1-J} & = & (-1.274\pm 0.030) ~E_{J-Ks}, \label{eq12} \\ E_{W2-J} & = & (-1.194\pm 0.055) ~E_{J-Ks}. \label{eq13} \end{eqnarray} The set of colour excess ratios shows that our results are very accurate for the $JHKs$ bands, since we have much more data for these wavelengths than for shorter ones, and the relation for this set of three indices is completely dominated by RCs. The relation $E_{J-H} \times E_{J-Ks}$ is the tightest one. The plot even reveals the spread of extinctions among the Wd~1 cluster members and because of this we are anchoring our results on this particular colour ratio. The ratio $E_{J-H}/E_{J-Ks}$\,=\,0.655\,$\pm$\,0.001 corresponds to the more usual form $E_{J-H}/E_{H-Ks}$\,=\,1.891\,$\pm$\,0.001. In the case of $Z$ and $Y$ filters, we have only data for RCs (taken from the VISTA archive), not for the cluster members. Regarding the $R$ filter, on the other hand, we have data only for Wd~1. The corresponding relation $E_{R-J}$\,$\times$\,$E_{J-Ks}$ does not enable an accurate linear fit and we derived the angular coefficient from the average ratio between these colour excesses. The number of measurements from RCs in the $B$ and $V$ filters is much smaller than for longer wavelengths - because RCs are faint in that spectral range - and just the closer RCs have reliable photometry in those filters. As seen in the corresponding plots, the deep imaging of Wd~1 (reaching stars under large extinction) was crucial to warranty a good fit in colour relations involving filters in the optical window. \subsection{Absolute extinctions derived from colour excess ratios}\label{section3.3} \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.7cm 0cm 2cm, clip]{./planelaw.pdf}} \caption{Extinction law for the Galactic Plane (GP), combining our results (filters with capital letters at the top and empty squares) with colour excess ratios from \citet{Scha16} - filled triangles $g,r,i,z$ filters - for which we derived the extinctions. The dotted red line is the fit with Eq. \ref{eq19} and black asterisks are residuals from the fit minus observations. The black {\it dashed} line is the power law with exponent $\alpha$\,=\,2.17, which is a very good representation of the extinction law for the inner GP in the range 0.8-4.0\,$\mu$m.} \label{fig:planelaw} \end{figure} We selected the colour excess ratio $E_{J-H}/E_{J-Ks}$ to derive the $\alpha$ exponent of the reddening law from Eq.\,\eqref{eq2}. This choice is based on the high accuracy of that index and on the fact that this is the best wavelength range to fit the reddening curve by a power law. We performed an initial study to derive alpha based on the dynamical effective wavelengths, by following the procedure designed by Stead and Hoare (2009), which evolve as the colour indices increase. We approximate their relations from their Fig.\,2a by linear relations in Appendix~\ref{appendixA} to calculate the effective wavelengths as a function of $(H-K)$ for $JHKs$ filters. We then extracted the photometry from 2MASS catalogue inside a circle of 1$^\circ$ diameter centered on the Wd~1 cluster from which we selected RCs as we did for our photometry. We evolved the effective wavelengths using $H-Ks$ for each star, derived their $\alpha$ and took the median value: $\alpha$\,=\,2.25\,$\pm$\,0.15. We used an alternative and simpler procedure, as described in Section\,\ref{section2}, taking the wavelengths of each filter fixed, as derived by \citet{Stead+2009} using a K2\,III star. We found $\alpha$\,=\,2.14\,$\pm$\,0.10, in reasonable agreement with the more complex method and in excellent agreement with \citet{Stead+2009} who derived $\alpha$\,=\,2.03\,$\pm$\,0.18 from 2MASS data and $\alpha$\,=\,2.17\,$\pm$\,0.07 from UKIDSS data. Since our data are calibrated in the 2MASS system, we adopted the fixed effective wavelengths, as just mentioned, for our photometric data, which have an observational setup very close to that of 2MASS filters. Now, using Eq.\,\eqref{eq2} and the measured ratio $E_{J-H}/E_{J-Ks}$\,=\,0.655\,$\pm$\,0.001 we derived $\alpha$\,=\,2.126\,$\pm$\,0.080 which translates into $A_J/A_{Ks}$\,=\,3.229 and $A_H/A_{Ks}$\,=\,1.769. Using this value of $A_J/A_{Ks}$ and the colour excess ratios in the previous sub--section (Equations\,\ref{eq3} to \ref{eq13}), we derive the remaining $A_\lambda/A_{Ks}$ listed below: \begin{eqnarray} A_B = 21.43:A_V=14.95:A_R=11.25:A_I=8.72: \nonumber \\ A_Z = 7.23:A_Y=5.10:A_J=3.23:A_H=1.77: \nonumber \\ A_{Ks} = 1:A_{W1}= 0.39:A_{W2} = 0.26. \label{eq14} \end{eqnarray} The above set of $A_\lambda/A_{Ks}$ defines an extinction law which is much steeper than typical ones published up to $\sim$10 years ago, but which are still in use \citep{CCM89,Indebetouw+2005,RL85}. From these relations, we can obtain the absolute extinction in all filters for a star, provided its $A_{Ks}$ is known. $A_{Ks}$ is easy to obtain from the colour excesses, for example: \begin{eqnarray} A_{Ks} & = & 0.449 ~E_{J-Ks}, \label{eq15} \\ A_{Ks} & = & 0.685 ~E_{J-H}, \label{eq16} \\ A_{Ks} & = & 1.300 ~E_{H-Ks}. \label{eq17} \end{eqnarray} In order to minimise the photometric errors, we use all these three relations and then perform the average to get a more robust value for $A_{Ks}$. We did this to obtain the extinction towards the Wd~1 cluster members, presented in Table\,\ref{table2} (column\,3) for all filters. The average value for this cluster based on 92 stars, after excluding those with obvious circumstellar contamination is: \begin{equation} <A_{Ks}~Wd1> = 0.736 \pm 0.079. \label{eq18} \end{equation} The 18 WR stars which are not affected by dust emission resulted in $<A_{Ks} ~Wd1-WRs>$\,=\,0.736\,$\pm$\,0.082, which is indistinguishable from the less evolved Wd~1 cluster members. This value is substantially lower than those derived by previous authors for this cluster, and this has an impact on the derived distance. \subsection{A reddening law for the Galactic Plane in the range 0.4--4.8 $\mu$m}\label{section3.4} Our target is located in the GP and it would be interesting to compare our results with others in the same region, since most of the recent studies based on large area surveys are focused on the Galactic Bulge (GB). A particularly important work in the GP is that reported by \citet{Scha16} for many thousands of stars, using the APOGEE spectroscopic survey and photometry from the ten--band Pan-STARRS1 survey.The majority of the targets are at -5$^\circ<$\,b\,$<$+5$^\circ$ and 0$^\circ<$\,l\,$<$250$^\circ$. \citet{Scha16} reported colour excess ratios in the optical, NIR and MIR windows, and their results are in excellent agreement with ours. They did not derive the extinction relations ($A_\lambda/A_{Ks}$), but this can be obtained in the same way as we described in the previous sub--section, especially because their ratios are anchored on 2MASS colours excesses. From their $E_{J-H}/E_{H-Ks} = 1.943$ we obtain $\alpha$\,=\,2.209, which translates into $A_J/A_{Ks}$\,=\, 3.380 and $A_H/A_{Ks}$\,=\,1.809. The set of $A_\lambda/A_{Ks}$ corresponding to their colour excess ratios is: $\displaystyle A_g\,=\,16.61\,:\,A_r\,=\,12.24\,:\,A_i\,=\,9.10\,:\,A_z\,=\,7.05\,:\,A_J\,=\,3.38\,:\,A_H\,=\,1.81\,:\,A_{Ks}\,\,=\,\,1\,:\,A_{W1}\,=\,0.43\,:\,A_{W2}\,\,=\,\,0.21$. Due to the excellent agreement of their results with ours, one can infer an average extinction law for the inner GP by performing a polynomial fit on both data sets in the range 0.4-4.8\,$\mu$m (see Eq.\,\eqref{eq19}, where $x = log(2.159/\lambda)$ in units of $1/\mu$m, and the corresponding plot is in Fig.\,\ref{fig:planelaw}). \begin{equation} \log\frac{A_\lambda}{A_{Ks}} = -0.015 + 2.330 x + 0.522 x^2 - 3.001 x^3 + 2.034 x^4. \label{eq19} \end{equation} The standard deviation of the observed minus calculated fit (O--C) is very small and is driven by the residuals at wavelengths shorter than 1\,$\mu$m, specially in the $B$ filter. It is surprising to see that a power law with $\alpha$\,=\,2.17 is an almost perfect representation of the data in the range 0.8-4\,$\mu$m with r.m.s.\,=\,0.09\,magnitudes. \section{Comparison between different extinction laws}\label{section4} \subsection{The $\alpha$ exponent and JHKs colour ratios}\label{section4.1} The comparison between reddening laws from different authors should be straightforward in the NIR, since $JHKs$ filters are very similar in modern times. Most of the authors report a narrow range of $E_{J-H}/E_{H-Ks}$ values (1.8--2.0). Surprisingly, there is a large spread in the published reddening laws, even when derived in the narrow NIR window and also when using the same database (2MASS): $\alpha$\,$\approx$\,1.65-2.64, which translates into $A_J/A_{Ks}$\,$\approx$\,2.5-4.2. The different procedures used to derive the extinction law from the colour excess ratios have different biases. In most cases, this is due to the difficulty in transforming broad-band measurements into monochromatic wavelengths \citep{Sale15}, but there are other additional issues, some of which we describe here. \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.7cm 0cm 2cm, clip]{./RJKIJxRAIVI.pdf}} \caption{Comparison between our results with other authors, showing a lack of correlation between $E_{J-Ks}/E_{I-J}$ and $A_I/E_{V-I}$. Our result for RCs (red crosses) compares well with that by \citet{Nataf16} - dashed rectangle - for the Galactic Bulge and samples the low density, higher ionization ISM. Wd~1 cluster members (blue crosses) have a contribution from a slightly different grain distribution from a denser medium. Older results by \citet{FM07} and by \citet{CCM89} are represented by pink and green circles, respectively.} \label{fig:RJKIJxRAIVI} \end{figure} \begin{table*} \scriptsize \label{table1} \begin{minipage}{170mm} \caption{Comparison between some reddening laws. Authors - 1st row: vdH\#16 = \citet{Hulst1946}, Nat16 = \citet{Nataf16}, Sch16 = \citet{Scha16}, SH09 = \citet{Stead+2009}, Ind05 = \citet{Indebetouw+2005}, CCM89 = \citet{CCM89}, FM09 = \citet{FM09}, Nishi06 = \citet{Nishiyama+2006}, Nishi08 = \citet{Nishiyama+2008}. Values marked with * were derived in this work from reddening and effective wavelengths reported by the corresponding authors. The second row indicates the photometry source. For FM09, we used only the star BD+45 973.} \begin{tabular}{lcccccccccccc} \hline \hline Filter &$\lambda_{\rm eff}$& Wd~1 & Wd1 + RCs & vdH\#16&Nat16&Sch16& SH09& Ind05& CCM89& FM09& Nishi06,08 \\ \hline & $\mu$m &$A_{\lambda}$&$A_{\lambda}/A_{Ks}$&&VVV & Pan-STARRS1&UKIDSS &2MASS &&HST/SNAP & SIRIUS \\ \hline $B$ & 0.442 &15.49$\pm$0.10& 21.43& 14.36 & - & - &- & - & 11.49& 17.52& -\\ $V$ & 0.537 &11.26$\pm$0.07& 14.95& 11.69& 13-15& 16.61g*&- & - & 8.77&12.76 &16.13\\ $R$ & 0.664 &8.44$\pm$0.10 & 11.25& 8.63& - & 12.24r*&- & - & 6.59& 9.48& -\\ $I$ & 0.805 &5.71$\pm$0.04 & 8.72& 6.32& 7.26 & 9.10i* & & - & - & 6.52& -\\ $Z$ & 0.878 & - & 7.23& 5.30& - & 7.05z* & - & - & 4.20& 5.81& -\\ $Y$ & 1.021 & - & 5.10& 3.97& - & - & - &- & - & - & - \\ $J$ & 1.244 &2.34$\pm$0.03 & 3.23& 2.70& 2.85 & 3.56* &3.25*& 2.5& 2.47& 2.90& 3.02 \\ $H$ & 1.651 &1.29$\pm$0.02 & 1.77& - & - & 1.84*&1.78*&1.61 & 1.54 & 1.67&1.73 \\ $Ks$ & 2.159 &0.74$\pm$0.01 & 1 & 1 & 1 & 1 & 1 & 1& 1 & 1 & 1 \\ $W1$ & 3.295 &0.29$\pm$0.03 & 0.39& - & - &0.43* & - & - & - & - &0.40 \\ $W2$ & 4.4809 & 0.19$\pm$0.05& 0.26& - & - & 0.21* & - & - & - & - &0.20 \\ \hline {\bf$\alpha$} & - & - & 2.13& - &1.88* & 2.21*& 2.14 &1.66*&1.68*& 2.00& 1.99 \\ \hline \end{tabular} \end{minipage} \end{table*} An example indicating differences related to the methodology is by comparing three extreme results based on 2MASS data. We adopted the effective wavelengths based on \citet{Stead+2009}, reported in Section\,\ref{section2}, and the observed colour excess ratio $E_{H-Ks}/E_{J-Ks}$\,=\,0.345 to derive $\alpha$\,=\,2.13 and $A_J/A_{Ks}$\,=\,3.23. \citet{Indebetouw+2005} measured $E_{H-Ks}/E_{J-Ks}$\,=\,0.36 and derived $A_J/A_{Ks}$\,=\,2.5 directly by minimization of the colour excesses. It is surprising to see that \citet{GF14} using 2MASS data obtained a very high $\alpha$\,=\,2.64 ($A_J/A_{Ks}$\,=\,4.2), although their colour excess ratio $E_{J-H}/E_{H-Ks}$\,=\,1.934 is very similar to others. Their method is complex to follow what exactly drove that power law exponent, but it corresponds to the the extreme values reported by \citet{FM09}, from which they adopted the reddening law. In some way their procedure converged to the higher and not to the average exponent of \citet{FM09}. Using our procedure (anchoring the reddening law in the $\alpha$ exponent) with the colour excess ratios reported by those authors, we derive $\alpha$\,=\,1.89 with $A_J/A_{Ks}$\,=\,2.84 and $\alpha$\,=\,2.15 with $ 3.27$, for \citet{Indebetouw+2005} and \citet{GF14} respectively. This agreements indicates that small differences on the effective wavelengths are not the main cause of the discrepancies found in the reddening laws. Beyond discrepant methodolgies, genuine differences due to dust properties do exist and are clearly shown in \citet{FM09}, based on spectrophotometric studies and not affected by problems defining the effective wavelengths. Large area surveys based on broad--band filters, such as reported by \citet{Nataf16} for the Galactic Bulge and by \citet{Scha16} for the GP, also show real variations and surprisingly, they exist on very small scales (arcminutes), on top of a general pattern related to the angular distances from the Galactic Centre. \citet{Nishiyama+2006} used their SIRIUS survey of the Bulge to measure $E_{H-Ks}/E_{J-Ks}$\,=\,0.343 and derived $\alpha=1.99$ with $A_J/A_{Ks}$\,=\,3.02, in reasonable agreement with our results for the Wd~1 direction. Our extinction law is in excellent agreement with \citet{Stead+2009}, who obtained $\alpha$\,=\,2.14. By translating their exponential law into extinction ratios (not presented in their original work), we computed $A_J/A_{Ks}$\,=\,3.25. \citet{Nataf16} reported an $<A_J/A_{Ks}>$\,=\,2.85 from the large and deep VVV survey towards the Bulge, as in \citet{Nishiyama+2006}, but covering more diverse environments close to the GP. Using the effective wavelengths from VVV, this corresponds to $\alpha$\,=\,1.88. \citet{Scha16} measured $E_{J-H}/E_{H-Ks}$\,=\,1.943 for a 2MASS sample in the inner GP, which corresponds to $E_{H-Ks}/E_{J-Ks}$\,=\,0.340. Using the wavelengths they adopted for 2MASS effective wavelengths ($\lambda_J$\,=\,1.2377, $\lambda_H$\,=\,1.6382 and $\lambda_{Ks}$\,=\,2.151) this implies $\alpha$\,=\,2.3 with $A_J/A_{Ks}$\,=\,3.56 and $A_H/A_{Ks}$\,=\,1.84. This is a little higher than our value, and using our adopted effective wavelengths this is reduced to $\alpha$\,=\,2.21 with $A_J/A_{Ks}$\,=\,3.38 plus $A_H/A_{Ks}$\,=\,1.81. We will use these last values in Fig.\,\ref{fig:allaws_dif} for consistency with our procedure, although differences are very small. As a summary we see that in general, extinction laws in $JHKs$ bands derived after 2005 are steeper than previous ones and this is driven by the deep surveys in the inner Galaxy. The foreground extinction we obtained in this work for the Wd~1 cluster, $A_{Ks}$\,=\,0.736, is smaller than $\approx$\,0.9-1.0 claimed in previous papers \citep{Andersen16, Gennaro+2011, Crowther+2006, Brandner+2008}. We derived a separate $A_{Ks}$ extinction for 18 Wolf--Rayet stars (excluding the dusty WC9) and obtained $A_{Ks}$\,=\,0.743\,$\pm$\,0.08 in excellent agreement with that based on non--WR cluster members. For comparison, using the same set of WRs, \citet{Crowther+2006} reported $A_{Ks}$\,=\,0.96\,$\pm$\,0.14. Later, those authors revised their value to $A_{Ks}$\,=\,1.01\,$\pm$\,0.14 \citep{Crowther+2008}. A recent study on the Wd~1 low mass contents by \citet{Andersen16} reports $A_{Ks}$\,=\,0.87\,$\pm$\,0.01 in better agreement with ours than previous works, but still not compatible. This is because they used the freddening line from \citet{Nishiyama+2006}, which is appropriate for the Bulge and our target in in the GP. \citet{Andersen16} found some evidence for the extinction gradient to increase towards N--NE, based on a large population of PMS, in agreement with our result, base on fewer evolved stars. Those authors decided to derive the extinction for every single PMS star, instead of using an average extinction. Our absolute extinction for Wd~1 is not compatible with previous ones and has an impact on the cluster distance, age and the absolute magnitudes of WRs. After measuring the cluster distance based on an eclipsing binary, we will tackle this question in a future paper. \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0]{./VVV_map.pdf}} \caption{Colour coded map from VVV survey from filter $Z$ (blue), $J$ (green) and $Ks$ (red). The labels indicate the directions in Fig.\,8 for which we measured the density of stars as a function of $J-Ks$ colour. The circles have $r$\,=\,2\arcmin. The {\it Centre} contains the Wd~1 cluster and the {\it ring} has width\,=\,1\arcmin. The yellow rectangle indicates the ISPI FOV.} \label{fig:VVV_map} \end{figure} \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 2.2cm 0cm 3.6cm 0cm, clip]{./dustmap.pdf}} \caption{ Extinction map. a): $A_{Ks}$ for the cluster members: {\it white crosses} are for values departing less than 0.5$\sigma$ from the average ($0.70<A_{Ks}<0.77$); {\it orange triangles} for slightly higher reddening ($0.78<A_{Ks}<0.79$); {\it red triangles} for the highest valus ($0.81<A_{Ks}<1.17$); {\it green squares} for $0.66<A_{Ks}<0.70$; and {\it blue squares} for $0.60<A_{Ks}<0.65$. b): {\it red clouds} for cold dust (160~$\mu$m); {\it blue clouds} for warm dust (70~$\mu$m) from Herschel survey; and {\it contour lines} for hot dust (24~$\mu$m) from WISE survey. Black regions in the Figure indicate the absence of dust emission. For a complete identification of the features in this Figure, we refer to the papers \citet{Negueruela+2010}, \citet{Ritchie+2009} and \citep{dougherty10}. } \label{fig:dustmap} \end{figure} \subsection{Relation between the extinction in the NIR, optical and MIR windows}\label{section4.2} As seen in Fig.\,\ref{fig:allaws_dif} all recent laws are much steeper than that of \citet{Hulst1946}, taken as a zero point, in contrast with \citet{CCM89}, which is much shallower. The frequently used rule of thumb $A_V/A_{Ks}$\,$\approx$\,10 based on that law and other similar laws, is no longer acceptable as representative for the Galaxy . The derived ratio in this work is $A_V/A_{Ks}$\,=\,14.95 and that of \citet{Scha16}, at a little shorter wavelength is $A_g/A_{Ks}$\,=\,16.61, both done along the GP. It is surprising that these values are not far from $A_V/A_{Ks}$\,$\approx$\,13-16 we derived from the colour excess ratios reported by \citet{Scha16} and $A_V/A_{Ks}$\,=\,16.13 reported by \citet {Nishiyama+2008}, both for the Bulge. The absolute extinction towards Wd~1 obtained in the present work $A_V$\,=\,11.26\,$\pm$\,0.07 is in excellent agreement with $A_V$\,=\,11.6 derived by \citet{Clark+2005} for Yellow Hypergiants in Wd~1. After the works by \citet{FM09} and \citet{Nataf16}, it became clear that all families of extinction laws cannot be represented by a single parameter, $R_V$, for example. \citet{Nataf16} showed that the the relation the ratios $R_{JKIJ}$\,=\,$E_{J-Ks}/E_{I-J}$ and $R_I$\,=\,$A_I/E(V-I)$ are not correlated. This is not unexpected, since the grain properties might be different in the general field (RCs) and inside the denser environment of the cluster. In Fig.\,\ref{fig:RJKIJxRAIVI} we compare our results with those from previous works. Wd~1 cluster members are seen under a range of extinction $A_V$\,$\approx$\,9-15 magnitudes, indicating the presence gas/dust condensations in the intra-cluster medium. Results based on alternative ratios in the optical (e.g. $A_V/E_{B-V}$ and $A_V/E_{V-I}$ against $R_{JKIJ}$) point to the same scenario of a lower $A_V/E_{B-V}$ in the inner Galaxy as shown by \citet{Nataf13}.The $R_V$\,=\,$A_V/E_{B-V}$ ratio have been reported to disagree with the ``universal'' $R_V$\,=\,3.1 for the Wd~1 cluster \citep{Clark+2005,Negueruela+2010,Lim+2013}. Using 38 members of the Wd~1 cluster we obtained $\displaystyle R_V = {A_V} / {E_{B-V}} = 2.50\pm0.04$. For the MIR, using data from WISE for our measured colour excess ratios, we obtained $A_{W1}$\,=\,0.39\,$\pm$\,0.03 and $A_{W2}$\,=\,0.26\,$\pm$\,0.02. Such values are in very good agreement with those derived from \citet{Scha16} colour excess ratios. \section{Extinction map}\label{section5} The extinction map towards the Wd~1 cluster can be tackled in several diffrent ways: a) by the distrubution of the cluster members extinction; b) from maps of the colours of stars; and c) from the correlation between these two maps and the distribution of dust clouds in the FOV. \subsection{Extiction map of the cluster members}\label{section5.1} In Fig.\,\ref{fig:dustmap} we present a map of extinction towards Wd~1. Points represent our measurements of $A_{Ks}$ for cluster members. White crosses represent the average extinction ($0.66<A_{Ks}<0.82$) or $10.1<A_{Ks}<12.6$), orange and red triangles higher than 1$\sigma$ and green and blue points are for values lower than 1$\sigma$ from the average. Lower extinction dominates at W/SW regions and higher extinction to N/NE from the cluster centre and in the central region there are all ranges of values. The mixture of values in the central region indicates the existence of intra--cluster dust, with some members displaced to our side and others to the back. It seems that at W/SW the dust is in the back side of the cluster, but crossing to the front at N/NE. The reality of this gradient can be checked in the next sub--section, when we discuss the far--infrared (FIR) images. Typical cluster members are seen through extinction in the range $0.66<A_{Ks}<0.82$ or $10.1<A_V<12.6$, represented by the 1$\sigma$ spread around the median value (78 stars). They are affected by interstellar plus local extinction. The $A_V = 2.5$ magnitudes range is due to local dust in front or internal to the cluster, and is surpringly large for $\sim 3.5 \arcmin$ FOV. The interstelar component can be measured from the (five) stars with lower extinction: $A_{Ks}=0.63\pm 0.02$ or $A_V=9.7\pm 0.30$. The nature of the reddening for the highest (seven) values $0.83<A_{Ks}<1.17$ or $12.7<A_V<17.9$ is difficult to assess: some stars may have really high extinction, but others might be affected by hot circumstellar dust emission, which mimics large reddening. \subsection{Cold and warm dust probed by FIR imaging }\label{section5.2} The red clouds in Fig.\,\ref{fig:dustmap} represent cold dust as measured by Herschel survey as measured by Herschel survey, using the PACS instrument \citep{Poglitsch2010}. It is notable that there are areas clear of dust (black regions) at E and W, probably caused by winds from massive stars and/or supernovae explosions. Just three "fingers" of cold dust reach the central regions of the cluster at NW, S and N and their tips are crowned by bright blue spots, tracing arm dust. The ''elephant trunk" at NW was detected in radio emission by \citet{dougherty10}. Warm dust impacts the cluster center, but dominates the East side, in rough agreement with higher extinction measured for stars. Four noticeable spots of warm dust coincide with the Red Supergiants (RSGs) W237, W20, W20 and the W9 B[e] star. However, the reddening of the RSGs stars is not larger than the average, indicating that the heated dust there is in the back of the star cluster. \subsection{Extinction and colours in the field around Wd~1}\label{section5.3} We can use the spatial density of the colour indices $(J-Ks)$ in our whole catalogue of sources to study the 3D distributon of dust. To study the surface density of stars projected inside a particular FOV, we need to keep in mind the many effects involved in the observed star counts. The most important parameter is the increase in the projected density with the distance, which grows as the volume of the spherical sector in the FOV (for uniform density). At a given magnitude limit this is counterbalanced by the inverse square law for the brightness and the extinction. Finally, in some regions and at a given image quality, stellar crowding drops the stellar counts. These considerations can provide guidelines to interpret the stellar counts as a function of magnitude in a particular image, but most of the effects can not be corrected. We binned the $J-Ks$ colour indices for ISPI+CAMIV photometry and normalised each slice of colours by the average colour index of the entire slice. After some initial trials, we selected a few slices which are representative of the features we found in the image, as shown in Fig.\,\ref{sb1:colour_density}.The upper left panel of in Fig. \ref{sb1:colour_density} displays a remarkable concentration of blue stars at W--SW of the Wd~1 which is probably due to a real cluster and not a result of a window of low extinction on the image. The upper right panel (1.251\,$<$\,$(J-Ks)$\,$<$\,1.788) shows the Wd~1 cluster at the centre of the image. The cluster size is larger than 6\arcmin and is elongated in the way described by \citet{Gennaro+2011} for the inner regions. The lower left panel shows a slice without strong concentrations in the stellar density. The regions of enhanced density at E and W of the cluster centre seems to be a halo of red pre main sequence (PMS) stars belonging to the cluster. The lower density at the S is probably due to a high extinction patch which shifts stars from that particular slice of colours to higher values since on average that slice represents the stellar field farthest from the Wd~1 cluster ($D$\,$>$\,4.5\,kpc). The depression at the Centre is due to photometric incompleteness caused by crowding of cluster members. The lower right panel (3.757\,$<$\,$(J-Ks)$\,$<$\,4.365) represents a slice of very red colour indices. In addition to distant stars (reddened by distance), it also represents stars shifted to redder colours by clouds both closer and farther than Wd~1. This is the case for the enhanced density at S of Wd~1. We suggest that the enhanced density zone ranging from SE to NE is also due to an obscuring cloud more distant than the central cluster. \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 2.6cm 0.3cm 2.6cm 0.5cm, clip]{./colour_density.pdf}} \caption{Stellar density maps:redder zones represents higher stellar densities. {\it Top left}: 0.350\,$>$\,$J-Ks$\,$>$\,0 showing a foreground concentration of stars with low extinction at SW of Wd~1 cluster. {\it Top right}: 1.788\,$>$\,$J-Ks$\,$>$\,1.251 showing the Wd~1 cluster with an extended halo. {\it Lower left}: 3.041\,$>$\,$J-Ks$\,$>$\,2.145 located farther than Wd~1. The low density region coincident with the cluster position is due to photometric incompleteness caused by crowding. The similar one to the S of the cluster is due to a cloud of higher extinction. {\it Lower right}: 4.365\,$>$\,$J-Ks$\,$>$\,3.757: the higher densities at the E and SE of the cluster are due to a zone of enhanced extinction farther than Wd~1.} \label{sb1:colour_density} \end{figure} \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.5cm 0cm 2cm, clip]{./J-Kdensity.pdf}} \caption{Density of stars for a range of bins in $J-Ks$ colour in some key directions defined in Fig.\,\ref{fig:VVV_map}. The first maximum in the {\it blue} line indicates spatial overdensity of stars at the SW foreground. The peak of the black {\it solid} and {\it dotted} lines ($J-Ks$\,$\approx$\,1.6) indicates the cluster centre. The green line (S) shows an increasing density towards redder colours up to the cluster colour/distance, then a decline followed by a recovery for large colour indices. For the NE and SE directions (red and brown lines) there is a similar behaviour, but the enhanced extinction occurs at the cluster distance/colour.} \label{fig:J-Kdensity} \end{figure} In order to explore in more detail these results, we made a plot with VVV images of the region, as seen in Fig.\,\ref{fig:VVV_map}. In fact there is a zone with blue stars (foreground young stars) at W-SW and a dark patch at the S and E of the cluster, indicating zones of higher extinction. The other structures revealed in Fig.\,\ref{fig:J-Kdensity} are not seen in the colour image of Fig.\,\ref{fig:VVV_map}. We then selected a few spatial directions, shown in the VVV map (Fig.\,\ref{fig:VVV_map}), to measure the stellar density as a function of the colour index $J-Ks$ along the cylinders with 2$'$ circular sections. In Fig.\,\ref{fig:J-Kdensity} we plot the density profiles with the same colours as the circles in Fig.\,\ref{fig:VVV_map}. The solid black line in Fig.\,\ref{fig:J-Kdensity} shows the density profile in which the most remarkable feature is the cluster in the colour range 1\,$<$\,$J-Ks$\,$<$\,2. The red extension of the peak is due to the PMS stars, which are intrinsically redder than the MS members in the peak. PMS stars also show up in the ring around the cluster area ({\it black dotted line}), with colour up to $J-Ks$\,$\approx$\,2.5. The {\it blue line} in both Figs.\,\ref{fig:VVV_map} and \ref{fig:J-Kdensity} shows the concentration of stars with $J-Ks$\,$<$\,1 -- bluer than the rest of the sources -- indicating that they are foreground stars, as discussed above. The {\it green line} explores the density profile to the S of the cluster. The density profile in this direction is normal for colours bluer than those of the Wd~1 cluster, but than it decreases faster than normal up to $J-Ks$\,$\approx$\,3 when the stellar density grows again up to $J-Ks$\,$\approx$\,4. The minimum around $J-Ks$\,$\approx$\,3 is caused by enhanced extinction, which shifts the counts to larger reddening. The {\it red line} is for the circle at NE of the cluster. Its behaviour is normal for $J-Ks<1$ after which there is a local minimum in coincidence with the cluster colour, followed by a constant increase up to $J-Ks$\,$\approx$\,3.8. Our interpretation is that there is an enhanced extinction at the distance of the cluster in that direction. The {\it brown line} corresponds to the circle at the SE and behaves similarly to that at NE, except it starts with lower counts than at the NE and presents two minima around $J-Ks$\,$\approx$\,1.5 and 3.5. Our interpretation is that there are two clouds at different distances in this direction. The extinction toward the SE looks to be higher than at any other direction. Just for a sake of spatial scale, the colour excess grows in proportion to $E_{J-Ks}$\,$\approx$\,0.37\,mag\,kpc$^{-1}$, in manner that the deepest minimum at $J-Ks$\,$\approx$\,3.7 corresponds to a distance {$D$\,$\approx$\,10\,kpc. As a summary, the most obscuring clouds are located to the S--SE border of our image at $D$\,$\approx$\,10\,kpc and there is a lighter obscuring cloud at SE--NE border at a distance comparable to that of Wd~1, but which does not impact the borders of the ISPI image at S--SE. The question is if this particular cloud has any impact on the properties of the stellar cluster. \section{The 8620\,\AA\,DIB tracing the extinction beyond $E_{B-V} > 1$}\label{section6} \citet{Munari+2008} and \citet[][]{Wallerstein+2007} showed that the 8620\,\AA\,DIB has a tight correlation with $E_{B-V}$. \citet{Maiz+2015} also showed the linear correlation of the equivalent width of this DIB (EW8620\,\AA) as a function of the extinction holds up to $A_V$\,$\approx$\,6. These authors also showed that the diffuse low density interstellar medium exposed to UV radiation has a different relation to the extinction as compared to the dense cold ISM. Recently, \citet{Munari+2008} presented 68 measurements of EW8620\,\AA~in the spectra from the RAVE survey ($R \sim 7500$) obtaining the relation $\displaystyle E_{B-V} = (2.72 \pm 0.03) \times EW8620$\,\AA~for $E_{B-V} < 1.2$. In this work we extend that relation to higher extinction values, which is relevant for the innermost Galactic region of the GAIA survey. \begin{figure} \centering \resizebox{\hsize}{!}{\includegraphics[width=20.0cm,angle=0, trim= 0cm 0.5cm 0.5cm 2cm, clip]{./AKEW8620.pdf}} \caption{$A_{Ks}$ extinction $versus$ EW8620\,\AA. Red symbols: this work -- triangles Wd~1 members, diamonds bright field stars. Black crosses \citep{Munari+2008}. Blue squares: \citet{Wallerstein+2007}. For EW8620\,$<$\,0.6\,\AA which translates to $E_{B-V}$\,$<$\,1 or $A_V$\,$<$\,7, the relation is linear and coincides with that of \citet{Munari+2008}. The size of the black cross indicates the 1-$\sigma$ standard deviation for the Wd~1 cluster members.} \label{fig:AKEW8620} \end{figure} Spectra were measured as described in Sect.\,\ref{section2.3}. For Wd~1 cluster members, $A_{Ks}$ were measured in the present work. For the Munari and Wallerstein stars, we used the 2MASS photometry and spectral type taken from {\tt SIMBAD}\footnote{http://simbad.u-strasbg.fr/} to derive the extinction. Table 2 displays our results of EW8620\,\textup{\AA} (column 2), $A_{Ks}$ (column 3) and the source for the 8620\,\AA\ data (column 4). We adopt $A_{Ks}$ instead of $A_V$ or $E_{B-V}$ because it is much less sensitive to dust size than optical wavelengths, and because our photometry was done in $JHKs$ bands. In this way, we use the relation we obtained for the Wd~1 cluster members: $A_{Ks}$\,=\,0.29\,$E_{B-V}$ to transform the extinction reported by \citet{Munari+2008}, \citet[][]{Wallerstein+2007} and for the nine field OB stars without $A_{Ks}$ measurements. The linear relation obtained by \citet[][]{Munari+2008} was thus transformed to $A_{Ks}$\,=\,0.691\,$\times$\,$EW8620$\,\AA. Combining these values with those derived here, we fit the set with a polynomial function. The relation we derive from the combined dataset is: \begin{table} \label{table2} \tiny \centering \caption{EW8620\,\AA\,, $A_{Ks}$ and $A_V$ extinction. Sources identified as Schulte\# are reported in \citet{Wallerstein+2007}. First 10 rows of the table are presented. A full version containing 103 entries is available online at the CDS.} \begin{tabular}{lllll} \hline Identification &EW8620 (\AA)& $A_{Ks}$& $A_{V} $ & Spec. source \\ \hline \hline W2a & 0.885 & 0.686 &10.068& OPD\\ W6a & 1.067 & 0.649 &10.985& OPD \\ W6b & 0.817 & 0.719 &10.970& 081.D-0324 \\ W7 & 0.844 & 0.772 &11.773& OPD \\ W8a & - & 0.641 &11.033& OPD \\ W8b & 0.830 & 0.760 &10.411& 081.D-0324 \\ W11 & - & 0.734 &10.765& \\ W12a & 0.870 & 0.800 &12.543& \\ W13 & 1.05 & 0.784 &10.818& 081.D-0324 \\ W15 & 0.948 & 0.733 &12.258& 081.D-0324 \\ \hline \end{tabular} \end{table} \begin{equation} A_{Ks} = (0.612 \pm 0.013)~EW + (0.191 \pm 0.025)~EW^2 \label{eq25} \end{equation} This equation is in excellent agreement with \citet{Munari+2008} which is linear for $EW$\,$<$\,0.6\,\AA (equivalent to $A_V$\,$<$\,9 for the inner GP extinction law of this paper). \section{Discussion and Conclusions}\label{section7} We present a study of the interstellar extinction in a FOV 10$'$ $\times$ 10$'$ in direction of the young cluster Westerlund~1 in $JHKs$ with photometric completeness $>$\,90\% at $\,$Ks$\,$=$\,15$. Using data publicly available, we extended the wavelength coverage to shorter and longer wavelengths from the optical to the MIR (although with less complete photometry). Colour excess ratios were derived by combining (92) Wd~1 cluster members with published spectral classification with (8463) RC stars inside the FOV. Our result for the NIR: $E_{J-H}/E_{H-Ks} = 1.891\pm0.001$ is typical of recent deep imaging towards the inner Galaxy. Using the procedure designed by \citet{Stead+2009} to obtain effective wavelengths of 2MASS survey and Eq.\,\eqref{eq2} we derived a power law exponent $\alpha = 2.126 \pm 0.080$ that implies $A_J/A_{Ks} = 3.23$. This extinction law is steeper than the older ones \citep{Indebetouw+2005,CCM89,RL85}, based on less deep imaging and is in line with recent results based on deep imaging surveys \citep{Stead+2009, Nishiyama+2006, Nataf16}. In the NIR, this implies in smaller $A_{Ks}$ and larger distances than laws based on more shallow photometry, which has a large impact on inner Galaxy studies. Using our measured $A_{Ks} / E_{J-Ks} = 0.449$ (plus combinations between other filters) we obtained the extinction to Wd~1, $<A_{Ks}> = 0.736 \pm 0.056$. This is $0.2-0.3$ magnitudes smaller than previous work, based on older (shallower) extinction laws \citep{Gennaro+2011, Negueruela+2010, Lim+2013, Piatti+1998}. On the other hand our $A_V = 11.26 \pm 0.07$ is in excellent agreement with $A_V = 11.6$ derived by \citet{Clark+2005} based on a completely different method: the OI~7774\,\AA\,EW $\times$ $M_V$ of six Yellow Hypergiants. The cluster extinction encompass the range $A_{Ks} = 0.55-1.17$ (which translates into $A_{V} \approx 8.5-17$). Cluster members have typical extinction $A_{Ks}=0.74\pm 0.08$ which translates into $A_V=11.4\pm 1.2$. The foreground interstellar component is $A_{Ks} = 0.63\pm 0.02$ or $A_V = 9.66\pm 0.30$. The extinction spread of $A_V \sim 2.5$ magnitudes inside a FOV 3.5$\arcmin$ indicates that it is produced by dust connected to the cluster region. In fact Fig\ref{fig:dustmap} shows a patchy distribution of warm dust. There are indications for a gradient in $A_{Ks}$ increasing from SW to NE, which is in line with the map of warm dust and with the colour density maps in the surrounding field. However, the effect is not very clear, suggesting a patchy intra-cluster extinction. The $J-Ks$ colour density maps unveiled the existence of a group of blue foreground stars, which may or may not be a real cluster. Since those stars partially overlap the Wd~1 cluster, they must be taken into account when subtracting the field population in the usual procedures to isolate Wd~1 cluster members. We measured the EW8620\,\AA\,DIB for 43 Wd~1 cluster members and combined them with additional filed stars and results collected from the literature, showing a good correlation with $A_{Ks}$. Although the linear relation reported by \citet{Munari+2008} was recovered for $E_{B-V}$\,$\approx 1$, it deviates for larger values and we present a polynomial fit extending the relation. The moderately large scatter in the Wd~1 measurements seems to reflect the uncertainties in our procedures to measure the extinction and the EW. Unfortunately our sample does not probe the range 0.4\,$<$\,EW8620\,$<$\,0.8, see Fig.\,\ref{fig:AKEW8620}. In order to improve the situation aiming for GAIA, the above relation should be re--done incorporating measurements from stars with 1$\,<$\,$E_{B-V}$\,$<$\,2.5 (equivalent to 0.3\,$<$\,$A_{Ks}$\,$<$\,0.7). As a matter of fact, it is expected that such relation will be different for the general ISM as compared to the denser ambients prevalent inside regions of recent star forming regions. We examined our result $R_V\,=\,A_V /E_{B-V}\,=\,2.50 \pm 0.04$ with care, since previous works found suspicious that this ratio for Wd~1 was much smaller than the usual $R_V$\,=\,3.1 \citet{Clark+2005,Lim+2013, Negueruela+2010}, although similar values have been reported by \citet{FM09} for a couple of stars like star BD+45~973. Moreover, this value is in excellent agreement with \citet{Nataf13}, as deduced from $R_I/E_{V-I}$ based on OGLE~III fields in the Galactic Bulge, close to the position of Wd~1. However, even if there was minor systematic errors in the photometric calibrations in the B--band, it would not be larger than a few tenths of magnitudes, which are dwarfed by the large extinctions: $A_B \approx 15$. An interesting result is the lack of correlation between the reddening law in the optical, as compared with the NIR (see Fig.\ref{fig:RJKIJxRAIVI}). This is in agreement with the result found by \citet{Nataf16} for a much larger field in the inner Galaxy. Looking to the position of \citet{CCM89} in that figure, we confirm other diagnostics, showing that dust grain properties in the inner Galaxy are different from those sampled by shallower imaging. Even in our small field (12\arcmin$\times$12\arcmin), the colour ratio diagram of Fig.\,\ref{fig:RJKIJxRAIVI} shows that the Wd~1 cluster members spread to a different zone in the diagram, as compared to RCs. This suggests that intra--cluster dust grains have properties different from the lower density ISM where RCs are located. The large spread of indices between Wd~1 members indicates the existence of clumps of dust grains with a variety of size properties. We derived the extinction law for the range 0.4-4.8\,$\mu$m which is in very good agreement with the colour excess ratios obtained by \citet{Scha16} from large photometric and spectroscopic surveys in the inner GP. We propose our law presented in Eq.\,\eqref{eq19} to be representative of the average inner GP. A striking feature of this law is its close coincidence with a power law with exponent $\alpha = 2.17$ for the entire range 0.8-4\,$\mu$m. We call the reader's attention to the fact that this is an average law, usefull for general purposes, since the reddening law varies from place to place inside the narrow zone of the GP. \section*{Acknowledgements} We thank an anonymous referee for the very productive questions/remarks which has improved our manuscript. We also thank M. Gennaro and A. Bonanos for a critical comments on earlier versions of this paper. AD thanks to Funda\c{c}\~{a}o de Amparo {\'a} Pesquisa do Estado de S\~{a}o Paulo - FAPESP for support through proc. 2011/51680-6. LAA acknowledges support from the FAPESP (2013/18245-0 and 2012/09716-6). FN acknowledges support from the FAPESP (2013/11680-2). This research has made use of the VizieR catalogue access tool, CDS, Strasbourg, France.
1c07eb1032a3137098d691a20059a4bec8845aa0
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{sc_intro} In this paper we discuss the application of a skewed-$t$ model suitable to capture skewness and kurtosis commonly present in insurance risks data. In particular, we introduce a Bayesian approach based on minimal informative prior distributions \citep{Villa:Walker:2014b} to estimate the parameters of the asymmetric Student-$t$ distribution (AST) introduced in \cite{Fernandez:Steel:1998} and re-proposed in \cite{Zhu:Galb:2010}. Insurance risks data, for example insurance losses, tend to present skewed behaviour \citep{Lane:2000, Vernic:2006} in almost any circumstance. Furthermore, losses related to catastrophes (e.g. earthquakes, exceptional floods, etc) are also better modelled by distributions with relatively heavy tails, as they quite often include extreme events. It is therefore appropriate to use a statistical model that can simultaneously account for data that is skewed and with values that may be extreme in behaviour. The generalised AST distribution, presented in Section \ref{sc_prelim}, includes a skewness parameter and allows to control the heaviness of both the left and right tails by means of a parameter representing the number of degrees of freedom. An obvious advantage of the AST, as it will be discussed, is that it allows to \textit{adjust} to the observed data, in the sense that can consider both skewed and non-skewed data and heavy-tailed data as well as observations that are suitable to be modelled by a distribution with tails like the one of the normal density. In other words, by estimating the parameter controlling the skewness of the distribution and the number of degrees of freedom, translates into a model selection scenario, where the competing models are: the normal distribution, the $t$ distribution, the skew normal and the skew $t$. It is often important to consider cases where the prior information about the true parameter values is minimal. There are circumstances where the prior information is not available or, for some reasons, it is not reliable or practical to be used. As such, we consider an inferential scenario where the prior distributions for the parameters of the model are set up according to objective Bayesian criteria. Note that here, the term ``objective'' is not intended to represent an actually objective set up, but simply to categorise the procedure followed to derive the prior distributions as one that provides an output in a sort of an automated fashion \citep{Berger:2006}. The choice of the objective prior for the skewness parameter is the Jeffreys' prior, while for the location and the scale parameters we will adopt the reference prior. For the number of degrees of freedom we adopt the truncated discrete prior proposed in \cite{Villa:Walker:2014a}. Although we do not discuss the possibility of using prior distributions elicited on the basis of reliable prior information, we still believe that it would be the most sensible way to proceed. Of course, the fact that inferential results can be effectively obtained by means of an objective Bayesian approach, it just strengthens the proposed model, as it allows to remove the (not always easy) task of translating prior information into prior probability distributions. \\ The paper is organised as follows. In Section \ref{sc_prelim} we introduce the asymmetric Student-$t$ distribution and discuss some of its properties. We also present the prior distributions used for the parameters and, in particular, the prior for the number of degrees of freedom, which we argue to have a support that is discrete and truncated. In Section \ref{sc_priors} we explicit the prior for the number of degrees of freedom and prove its independence from the skewness parameter. This result is key in motivating the prior for $\nu$ proposed by \cite{Villa:Walker:2014a}. Section \ref{sc_simul} is dedicated to present the results of a simulation study aimed to analyse the frequentist properties of the posterior distribution for the number of degrees of freedom. To complete the discussion of the proposed approach to model insurance loss data, in Section \ref{sc_real} we analyse two well known data sets. We are able to show the versatility of the model in a scenario of skewed data with extreme events and in a scenario where a symmetric normal distribution is sufficient to be used as a model. Final discussion points and conclusions are presented in Section \ref{sc_disc}. \section{Preliminaries}\label{sc_prelim} In this paper, we consider the asymmetric Student-$t$ (AST) distribution whose general density has the form \begin{equation}\label{eq_ast1} f(x|\alpha,\nu,\mu,\sigma) = \left\{ \begin{array}{l l} \dfrac{K(\nu)}{\sigma}\left[1+\dfrac{1}{\nu}\left(\dfrac{x-\mu}{2\alpha\sigma}\right)^2\right]^{-\frac{\nu+1}{2}} & \quad x\leq\mu\\ \dfrac{K(\nu)}{\sigma}\left[1+\dfrac{1}{\nu}\left(\dfrac{x-\mu}{2(1-\alpha)\sigma}\right)^2\right]^{-\frac{\nu+1}{2}} & \quad x>\mu \end{array} \right. \end{equation} where $\alpha\in(0,1)$ represents the skewness parameter, $\mu$ is the location parameter, $\sigma$ is the scale parameter, and $\nu$ represents the degrees of freedom, with $$K(\nu)\equiv\Gamma((\nu+1)/2)/[\sqrt{\pi\nu}\Gamma(\nu/2)],$$ see \cite{Fernandez:Steel:1998} and \cite{Zhu:Galb:2010} for a detailed discussion of the distribution and its properties. The usual Student-$t$ distribution can be recovered by setting $\alpha=1/2$ in \eqref{eq_ast1}, while the skewed Cauchy and the skewed normal distributions are special cases obtained when $\nu=1$ and $\nu=+\infty$, respectively. In the Bayesian framework, the inference about the parameter of a model is accomplished by combining the prior uncertainty about the true value of the parameters, expressed in the form of probability distributions, and the information contained in the observed sample. The latter is expressed by the likelihood function which, for the model in \eqref{eq_ast1}, has the form \begin{eqnarray}\label{eq_like1} L(\alpha,\nu,\mu,\sigma|x) &=& \prod_{i=1}^n f(x_i|\alpha,\nu,\mu,\sigma) \nonumber \\ &=& \prod_{x_i\leq\mu}\dfrac{K(\nu)}{\sigma}\left[1+\dfrac{1}{\nu}\left(\dfrac{x_i-\mu}{2\alpha\sigma}\right)^2\right]^{-\frac{\nu+1}{2}} \times \prod_{x_i>\mu}\dfrac{K(\nu)}{\sigma}\left[1+\dfrac{1}{\nu}\left(\dfrac{x_i-\mu}{2(1-\alpha)\sigma}\right)^2\right]^{-\frac{\nu+1}{2}}. \end{eqnarray} Thus, if we indicate the joint prior distribution for the parameters by $\pi(\alpha,\nu,\mu,\sigma)$, the posterior distribution is given by \begin{equation}\label{eq_prior_1} \pi(\alpha,\nu,\mu,\sigma) \propto L(\alpha,\nu,\mu,\sigma|x)\times\pi(\alpha,\nu,\mu,\sigma). \end{equation} The prior distributions will be discussed in Section \ref{sc_priors}. However, we believe it is useful to give an overview of the overall approach in this section. In this paper we adopt an objective Bayesian approach to make inference on the unknown parameters of the density in \eqref{eq_ast1}. To proceed, we assume that the tail parameter $\nu$ is discrete. That is, $\nu=1,2,\ldots$. The reason of choosing a discrete parameter space for $\nu$ is due to the close proximity of models with consecutive number of degrees of freedom. In line with what discussed for the $t$ density by \cite{Jacquier:2004} and \cite{Villa:Walker:2014a}, the amount of information from the observations (i.e. the sample size) will rarely be sufficient to discern between consecutive densities with a difference in the number of degrees of freedom smaller than one. As such, the choice of a discrete support for $\nu$, with intervals of size one, is sensible. It is still possible to define a discrete prior (with the same approach) over a more dense support; however, the choice has to be motivated by some prior evidence, such as the certainty of dealing with a very large sample size, and the resulting prior distribution will obviously be different as the amount of information would have changed. \section{The prior distributions for the parameters of the AST model}\label{sc_priors} In this section we outline the approach to derive the prior distributions for the parameters of the AST model. We dedicate most of this section to the prior for the number of degrees of freedom because, being a discrete parameter, presents some non-trivial challenges. We start by making the assumption that the parameters present, a priori, some degree of independence and, therefore, the prior has the form \begin{equation}\label{eq_prior.1} \pi(\alpha,\nu,\mu,\sigma) \propto \pi(\nu|\alpha,\mu,\sigma)\pi(\mu,\sigma)\pi(\alpha). \end{equation} As we will show in Section \ref{sc_nuprior}, the prior for the number of degrees of freedom does not depend on the skewness parameter, therefore $\pi(\nu|\alpha,\mu,\sigma)=\pi(\nu|\mu,\sigma)$. It has to be noted that, although objective methods aim to obtain prior distributions depending only on the chosen model, in practice there is always some degree of subjectivity involved. In this case, we make the assumption that the parameters are independent \emph{a priori}, noting that this is a common practice in objective Bayesian analysis. \subsection{Objective Bayes for discrete parameter spaces}\label{sc_intro_obj} One of the assumptions on which the model presented in this paper is based upon, is that the parameter representing the number of degrees of freedom ($\nu$) is considered as discrete. The specific reasons for this choice are presented in Section \ref{sc_nuprior}, but it is worthwhile to give an overview of the general idea behind the prior specific for $\nu$ here. Whilst there are several approaches to deal with continuous parameter spaces, such as Jeffreys' prior \citep{Jeffreys:1961} and reference prior \citep{BBS:2009}, discrete parameter spaces have always been dealt with on a case-by-case basis. Only recently general solutions for the discrete case have been put forward, with different degree of success. Among these, possibly the most noteworthy include \cite{Barger:Bunge:2008} and \cite{BBS:2012}. However, the former approach can be applied to a limited set of models, and the latter approach shows some deficiencies in terms of objectivity and generality. In this paper, we consider the method discussed in \cite{Villa:Walker:2014b}, which, as far as we know, is the sole objective approach that can be applied to any discrete parameter space without the necessity of being ``adjusted'' to the chosen model. The prior proposed in \cite{Villa:Walker:2014b} is based on the idea of assigning a \emph{worth} to each element $\theta$ of the discrete parameter space $\Theta$. The \emph{worth} is objectively measured by assessing what is lost if that parameter value is removed from $\Theta$, and it is the true one. Once the \emph{worth} has been determined, this will be linked to the prior probability by means of the self-information loss function \citep{Merhav:Feder:1998} $-\log\pi(\theta)$. A detailed illustration of the idea can be found in \cite{Villa:Walker:2014b}, but here is an overview. Let us indicate by $f(x;\theta)$ a distribution (either a mass function or a density) characterised by the unknown discrete parameter(s) $\theta$ and let $$D(f(x;\theta)\|f(x;\theta^\prime))=\int f(x;\theta)\log\left\{\frac{f(x;\theta)}{f(x;\theta')}\right\}\,dx$$ be the Kullback--Leibler divergence \citep{Kull:1951}. The utility (i.e. \emph{worth)} to be assigned to $f(x;\theta)$ is a function of the Kullback--Leibler divergence measured from the model to the nearest one; where the nearest model is the one defined by $\theta^\prime\neq \theta$ such that $D(f(x;\theta)\|f(x;\theta^\prime))$ is minimised. In fact (see \cite{Berk:1966}) $\theta^\prime$ is where the posterior asymptotically accumulates if the true value $\theta$ is excluded from $\Theta$. The objectivity of how the utility of $f(x;\theta)$ is measured is obvious, as it depends on the choice of the model only. Let us now write $u_1(\theta)=\log\pi(\theta)$ and let the minimum divergence from $f(x;\theta)$ be represented by $u_2(\theta)$. Note that $u_1(\theta)$ is the utility associated with the prior probability for model $f(x;\theta)$, and $u_2(\theta)$ is the utility in keeping $\theta$ in $\Theta$. We want $u_1(\theta)$ and $u_2(\theta)$ to be matching utility functions, as they are two different ways to measure the same utility in $\theta$. As it stands, $-\infty<u_1\leq0$ and $0\leq u_2<\infty$, while we actually want $u_1=-\infty$ when $u_2=0$. The scales are matched by taking exponential transformations; so $\exp(u_1)$ and $\exp(u_2)-1$ are on the same scale. Hence, we have \begin{equation}\label{eq1} e^{u_1(\theta)} = \pi(\theta) \propto e^{g\{u_2(\theta)\}}, \end{equation} where \begin{equation}\label{eq2} g(u) = \log(e^u-1). \end{equation} By setting the functional form of $g$ in \eqref{eq1}, as it is defined in \eqref{eq2}, we derive the proposed objective prior for the discrete parameter $\theta$ as follows \begin{equation}\label{eq3} \pi(\theta) \propto \exp \left\{ \min_{\theta\neq \theta^\prime \in\Theta} D(f(x;\theta)\|f(x;\theta^\prime)) \right\} - 1. \end{equation} We note that in this way the Bayesian approach is conceptually consistent, as we update a prior utility assigned to $\theta$, through the application of Bayes theorem, to obtain the resulting posterior utility expressed by $\log\pi(\theta\mid x)$. Indeed, there is an elegant procedure akin to the Bayes Theorem which works from a utility point of view, namely that $$\log\pi(\theta\mid x) = K + \log f(x\mid\theta) + \log\pi(\theta),$$ which has the interpretation of $$\text{Utility}(\theta\mid x,\pi) = K + \text{Utility}(\theta\mid x) + \text{Utility}(\theta\mid\pi),$$ where $K$ is a constant which does not depend on $\theta$. There is then a retention of meaning between the prior and the posterior information (here represented as utilities). This property is not shared by the usual interpretation of Bayes theorem when priors are objectively obtained; in fact, the prior would usually be improper, hence not representing probabilities, whilst the posterior is (and has to be) a proper probability distribution. \subsection{The prior for $\nu$}\label{sc_nuprior} In this Section we will show that the prior for $\nu$ introduced in \cite{Villa:Walker:2014a} can be used as prior for $\nu$ for the skewed Student-$t$ distribution. The prior for $\nu$ is based on similar arguments as the ones discussed in \cite{Villa:Walker:2014a}. In particular, we consider the following. First, the parameter is treated as discrete; that is, $\nu=1,2,\ldots$. The reason is that the amount of information provided from the data will rarely be sufficient to discern densities with $\nu$ separated by an interval smaller than one. In other words, in general, the sample size is not going to be sufficient to allow estimates more precise than size one. Whilst it is possible, in principle, to consider any continuous parameter as discrete, for this model we deem appropriate to limit the discretisation to $\nu$. Besides the above motivation, we note that considering $\nu$ as discrete has connection with an interpretation of a $t$-distributed random variable. In fact, a $t$ with $\nu$ degrees of freedom can be seen as the ratio of two independent random variables: a standard normal and the square root of a chi-square divided by its number of degrees of freedom $\nu$. While in principle it is possible to discretise any continuous parameter, such as $\mu$, $\sigma$ and $\alpha$, the procedure would carries a strong degree of subjectivity, namely the discretisation density. In fact, the above considerations made for $\nu$ cannot be applied to the remaining three parameters of the model. Hence, the choice to consider $\nu$ only as discrete. Second, the parameter space of $\nu$ has to be truncated. In \cite{Villa:Walker:2014a} this argument is motivated by the fact that, as the number of degrees of freedom ($\nu$) of a Student-$t$ goes to infinity, the density converges to a normal in distribution. As such, after a certain value of $\nu$, the model can be consider normal for any value of the parameter. A sensible choice is to set $\nu_{\max}=30$ to represent the normal model. Therefore, the inference problem reduces in choosing among $t$ densities with $\nu=1,\ldots,29$ and the normal density. As recalled in the introduction, a similar result holds for the skewed Student-$t$ distribution as well. Indeed, as $\nu$ goes to infinity, the model converges to a skewed normal distribution. This allows to apply the same truncation argument to the skewed Student-$t$ distribution. To derive the prior for $\nu$, we apply the approach introduced in Section \ref{sc_intro_obj}. Let us, at first, assume that $\nu=1,\ldots,30$ and $\alpha=0.5$. In this case, the density in \eqref{eq_ast1} represents a symmetrical $t$ distribution with $\nu$ degrees of freedom, and the prior for the parameter $\nu$ has the form as in \cite{Villa:Walker:2014a}. To simplify the notation, we indicate the $t$-density with parameters $\nu$, $\alpha$, $\mu$ and $\sigma$ by $f_{\nu}^{\alpha}$; therefore, the prior for $\nu$ is given by $$\pi(\nu|\alpha,\mu,\sigma) \propto \exp\left\{D(f_\nu^{\alpha}\|f_{\nu+1}^{\alpha})\right\}-1,$$ for $\nu<29$ and, for $\nu\geq29$ $$\pi(\nu|\alpha,\mu,\sigma) \propto \exp\left\{D(f_\nu^{\alpha}\|f_{\nu-1}^{\alpha})\right\}-1,$$ where $f_{30}^{\alpha}\sim \phi^{\alpha}$ is the skewed normal distribution with mean $\mu$ and standard deviation $\sigma$, i.e. \begin{equation}\label{eq_ast2} \phi^{\alpha}(x|\mu,\sigma) = \left\{ \begin{array}{l l} \dfrac{1}{\sigma\sqrt{2\pi}}\exp\left\lbrace -\left(\dfrac{x-\mu}{2\alpha\sigma}\right)^2\right\rbrace & \quad x\leq\mu\\ \dfrac{1}{\sigma\sqrt{2\pi}}\exp\left\lbrace -\left(\dfrac{x-\mu}{2(1-\alpha)\sigma}\right)^2\right\rbrace & \quad x>\mu \end{array} \right. \end{equation As recalled in the introduction, the choice of $\alpha=0.5$ corresponds to the usual Student-$t$ distribution and the prior above is the one introduced in \cite{Villa:Walker:2014a}. The following Theorem states a crucial result to set $\pi(\nu|\alpha,\mu,\sigma)$. \begin{theorem} Let $f_{\nu}^{\alpha}$ be the skewed Student-$t$ distribution with parameters $\mu$ and $\sigma$. Then $$D(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})=D(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5}),$$ for every $\nu\geq 1$. \end{theorem} \begin{proof} Without loss of generality, we can consider $\mu=0$ and $\sigma=1$. Note that $$D(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})=D_{\leq}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})+D_{>}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha}),$$ where \begin{equation*} \begin{split} D_{\leq}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})&=\int_{-\infty}^{0}f_{\nu}^{\alpha}(y)\log\left\{\frac{f_{\nu}^{\alpha}(y)}{f_{\nu+1}^{\alpha}(y)}\right\}\,dy,\\ \end{split} \end{equation*} and \begin{equation*} \begin{split} D_{>}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})&=\int_{0}^{+\infty}f_{\nu}^{\alpha}(y)\log\left\{\frac{f_{\nu}^{\alpha}(y)}{f_{\nu+1}^{\alpha}(y)}\right\}\,dy .\\ \end{split} \end{equation*} We focus on the first term \begin{equation*} \begin{split} D_{\leq}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})&=\int_{-\infty}^0K(\nu)\left[1+\frac{1}{\nu}\left(\frac{y}{2\alpha}\right)^2\right]^{-\frac{\nu+1}{2}}\log\left\{\frac{K(\nu)\left[1+\frac{1}{\nu}\left(\frac{y}{2\alpha}\right)^2\right]^{-\frac{\nu+1}{2}}}{K(\nu+1)\left[1+\frac{1}{\nu+1}\left(\frac{y}{2\alpha}\right)^2\right]^{-\frac{\nu+2}{2}}}\right\}\,dy.\\ \end{split} \end{equation*} The change of variable $z={y}/{2\alpha}$ yields \begin{equation*} \begin{split} D_{\leq}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha}) &=2\alpha\int_{-\infty}^0 K(\nu)\left[1+\frac{z^2}{\nu}\right]^{-\frac{\nu+1}{2}}\log\left\{\frac{K(\nu)\left[1+\frac{z^2}{\nu}\right]^{-\frac{\nu+1}{2}}}{K(\nu+1)\left[1+\frac{z^2}{\nu+1}\right]^{-\frac{\nu+2}{2}}}\right\}\,dz\\ &=2\alpha D_{\leq}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5}). \end{split} \end{equation*} In a similar fashion $$D_{>}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})=2(1-\alpha)D_{>}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5}).$$ The symmetry of the standard Student-$t$ distribution ensures that $$D_{\leq}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})=D_{>}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5}).$$ Therefore $$2D_{\leq}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})=2D_{>}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})=D(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5}),$$ and we can easily conclude \begin{equation*} \begin{split} D(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})&=D_{\leq}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})+D_{>}(f_{\nu}^{\alpha}\parallel f_{\nu+1}^{\alpha})\\ &=2\alpha D_{\leq}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})+2(1-\alpha)D_{>}(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})\\ &=\alpha D(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})+(1-\alpha)D(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5})\\ &=D(f_{\nu}^{0.5}\parallel f_{\nu+1}^{0.5}). \end{split} \end{equation*} \end{proof} \noindent In a similar way, it can be proved that $$D(f_\nu^{\alpha}\|f_{\nu-1}^{\alpha})=D(f_\nu^{0.5}\|f_{\nu-1}^{0.5})$$ for every $\nu\geq 2$. The above result also holds when we assume that for $\nu=30$, $f_{\nu}^{\alpha}$ is the skewed normal distribution. These results lead to the following important considerations: \begin{enumerate} \item The objective prior distribution for $\nu$ doesn't depend by the skewness parameter $\alpha$; \item The objective prior distribution for $\nu$ for the skewed model is exactly the prior introduced in \cite{Villa:Walker:2014a}, i.e. $\pi(\nu|\mu,\sigma,\alpha)=\pi(\nu|0.5,\mu,\sigma)$. \end{enumerate} \subsection{Prior distributions for $\alpha$ and $(\mu,\sigma)$}\label{sc_prior_others} The derivation of non-informative priors for the remaining parameters of the AST is straightforward. A common assumption is that the parameters are independent a priori. Although this assumption can be relaxed, in the sense of limiting the independence to the one between the skewness parameter on one side and the location and scale parameters of the other side, the resulting overall prior is the same. In fact, if we consider $\mu$ independent from $\sigma$, the Jeffreys' independent prior will have the form $\pi(\mu,\sigma)=\pi(\mu)\pi(\sigma)$. Given that the Jeffreys' prior for a location parameter is proportional to 1, and the Jeffreys' prior for a scale parameter is proportional to the inverse of the parameter, we would have $\pi(\mu,\sigma)\propto 1/\sigma$ \citep{Jeffreys:1961}. However, the above prior coincides with the reference prior for the pair $(\mu,\sigma)$ \citep{BBS:2009}. Therefore, assuming or not assuming prior independence between the location parameter and the scale parameter does not make any practical difference from an inferential point of view. For the skewness parameter $\alpha$, we will use the Jeffreys' prior discussed in \cite{Rubio:Steel:2014}, which is a Beta distribution with both parameters equal to 1/2; that is $\pi(\alpha)\sim\mbox{Beta}(1/2,1/2)$. In fact, it can be seen in Theorem 3 and Corollary 3 of \cite{Rubio:Steel:2014} that the (independence) Jeffreys prior for $\alpha$, under the parameterisation in \eqref{eq_ast1}, is precisely a Beta distribution with both parameters equal to $1/2$. \section{Simulation study}\label{sc_simul} In this section we present a simulation study of the prior for $\nu$. Due to the objective nature of the prior considered it is appropriate to present the frequentist properties of the yielded marginal posterior for the number of degrees of freedom. In particular, we analyse the frequentist mean squared error (MSE) and the frequentist coverage of the $95\%$ credible intervals. The former represents a measure of the precision of the estimate, while the latter reports the proportion of times the true value is contained in the interval defined by the $2.5\%$ and the $97.5\%$ quantiles of the posterior in a repeated sampling scenario. The posterior distribution for $\nu$ tends to be skewed, as for example is shown in Figure \ref{fig:single_samp}; as such, an appropriate index of the posterior which summarises its centrality is the median. Furthermore, considering $\nu$ as discrete calls for an index which is discrete as well, i.e. the median. Therefore, the MSE is computed with respect to the median, and the precision of the estimate is defined by the relative square root of the mean squared error from the median: $\sqrt{\mbox{MSE}(\nu)}/\nu$. Given that the model converges to the normal distribution for $\nu\rightarrow\infty$, it is more difficult to discern between AST densities with large values of $\nu$. By considering the relative MSE, we somehow counterbalance an otherwise naturally increasing MSE and give a more interpretable information about the performance of the prior. The posterior distribution for $\nu$ is obtained by marginalising the full posterior $\pi(\alpha,\nu,\mu,\sigma)\propto L(\alpha,\nu,\mu,\sigma|x)\pi(\alpha,\nu,\mu,\sigma)$, where $\pi(\alpha,\nu,\mu,\sigma)$ is the prior defined in \eqref{eq_prior.1} and $L(\alpha,\nu,\mu,\sigma|x)$ is the likelihood function defined in \eqref{eq_like1}. As the posterior distribution is analytically intractable, we use Monte Carlo methods to obtain the marginal posterior distributions for each parameter. The algorithm employed is outlined in the Appendix. In this simulation study we consider the following scenarios. We noted that the location parameter and the scale parameter do not have any effect on the inferential results for the number of degrees of freedom, as such we have considered, without loss of generality, $\mu=0$ and $\sigma=1$. We have considered three different values of the skewness parameter, that is $\alpha=0.3$, $\alpha=0.5$ and $\alpha=0.8$. For each of the above three values we have performed repeated sampling with $\nu=1,\ldots,20$. We have run 100000 iterations of the MCMC algorithm for each case, and for a sequence of sample sizes $n=50$, $n=100$ and $n=1000$. The simulation has been done in R language and it took about 5.4 hours per case in a cluster of computers with i7 processors. Although the main fields of application of the asymmetrical $t$ density discussed in this paper call for relatively large sample sizes, we have also considered the possibility of applying the model to small numbers of observations. It is in fact for relatively small sample size that Bayesian analysis tends to give better pay-off compared to the frequentist approach. Therefore, beside considering $n=1000$, we have also analysed the frequentist properties for $n=100$ and $n=50$. The prior used for the parameters $\alpha$, $\mu$ and $\sigma$ are the objective priors outlined in Section \ref{sc_prelim} which, as shown in \cite{Rubio:Steel:2014} and the references therein, yield proper posterior distributions. \begin{figure}[h!] \centering \subfigure[]{% \includegraphics[scale=0.25]{MSE_n50.pdf} \label{fig:subfigure1}} \quad \subfigure[]{% \includegraphics[scale=0.25]{Coverage_n50.pdf} \label{fig:subfigure2}} \subfigure[]{% \includegraphics[scale=0.25]{MSE_n100.pdf} \label{fig:subfigure3}} \quad \subfigure[]{% \includegraphics[scale=0.25]{Coverage_n100.pdf} \label{fig:subfigure4}} \subfigure[]{% \includegraphics[scale=0.25]{MSE_n1000.pdf} \label{fig:subfigure5}} \quad \subfigure[]{% \includegraphics[scale=0.25]{Coverage_n1000.pdf} \label{fig:subfigure6}} \caption{Frequentist coverage of the $95\%$ credible intervals for $\nu$ (right) and square root of relative mean squared error of the estimator of $\nu$ (left). The simulations are for $\alpha=0.3$ (solid), $\alpha=0.5$ (dashed) and $\alpha=0.8$ (dotted), and for $n=50$ (top), $n=100$ (middle) and $n=1000$ (bottom).} \label{fig:figure1} \end{figure} The simulation results are summarised in Figure \ref{fig:figure1}. The left column, plots (a), (c) and (e), shows the square root of the relative mean squared error for the posterior medians of $\nu$. While there is a dependence on the accuracy of the estimate from the sample size, we do not appreciate any effect from the value of the skewness parameter $\alpha$. In fact, within each plot on the left-hand-side, the mean squared error curves have a similar behaviour, with a higher value towards the region of the parameter space where contiguous number of degrees of freedom characterise distributions relatively different. As $\nu$ increases, leading to $t$ densities which are more and more similar to each other, the relative mean squared error decreases. As expected, the mean squared error in higher for small sample sizes than for larger sample sizes, given that more information is carried by the observations via the likelihood function. This is easily seen by moving from the top to the bottom of the left column of Figure \ref{fig:figure1}, corresponding to sample sizes of, respectively, $n=50$, $n=100$ and $n=1000$. The right column of Figure \ref{fig:figure1} shows the frequentist coverage of the $95\%$ credible intervals for $n=50$ (top), $n=100$ (middle) and $n=1000$ (bottom). Although there is no apparent impact from different values of $\alpha$, we note more variable coverage values around the region of the parameter space associated with higher relative mean squared errors, compared to areas where the relative mean squared error is smaller. Although this behaviour is common to any sample size, we see a shifting of the more uncertain area towards higher number of degrees of freedom. This last result is common to other Bayesian objective methods to estimate $\nu$, such as in \cite{Fonseca:Ferreira:Migon:2008}, \cite{Villa:Walker:2014a} and the references therein. The above conclusion can also be drawn by inspecting the posterior median credible intervals for $\nu$, shown in Table \ref{tab:median_intervals}. In particular, the higher the value of the number of degrees of freedom, keeping the sample size fixed, the larger the interval. Alternatively, the higher the sample size, for the same value of $\nu$, the smaller the median credible interval. As expected, there is no appreciable difference in the median interval for different values of $\alpha$.\\ \begin{table}[] \centering \begin{tabular}{c|ccc|ccc|ccc} \hline & \multicolumn{3}{c|}{$n=50$} & \multicolumn{3}{c|}{$n=100$} & \multicolumn{3}{c}{$n=1000$} \\ \hline $\nu$ & $\alpha=0.3$ & $\alpha=0.5$ & $\alpha=0.8$ & $\alpha=0.3$ & $\alpha=0.5$ & $\alpha=0.8$ & $\alpha=0.3$ & $\alpha=0.5$ & $\alpha=0.8$ \\ \hline 1 & (1,1) & (1,1) & (1,1) & (1,1) & (1,1) & (1,1) & (1,1) & (1,1) & (1,1) \\ 2 & (2,4) & (1,5) & (2,3) & (1,2) & (2,3) & (2,3) & (2,2) & (2,2) & (2,2) \\ 3 & (2,6) & (2,8) & (2,7) & (2,3) & (2,4) & (2,6) & (3,3) & (3,3) & (3,3) \\ 4 & (2,7) & (2,7) & (2,6) & (3,9) & (2,5) & (3,7) & (4,5) & (4,5) & (4,5) \\ 5 & (4,8) & (3,8) & (2,8) & (4,14) & (3,10) & (4,14) & (4,5) & (4,5) & (4,5) \\ 6 & (2,25) & (3,7) & (3,20) & (4,15) & (4,12) & (4,17) & (5,6) & (5,6) & (5,7) \\ 7 & (3,29) & (5,29) & (5,30) & (3,10) & (4,15) & (4,14) & (5,7) & (6,8) & (5,8) \\ 8 & (2,26) & (3,29) & (3,26) & (3,28) & (5,27) & (5,29) & (6,10) & (6,9) & (6,10) \\ 9 & (5,30) & (3,30) & (5,30) & (3,22) & (4,24) & (4,23) & (7,11) & (7,11) & (7,12) \\ 10 & (3,26) & (4,30) & (4,29) & (7,30) & (5,28) & (5,28) & (7,12) & (9,16) & (7,12) \\ 11 & (5,30) & (5,30) & (3,27) & (5,29) & (5,28) & (4,27) & (9,18) & (8,13) & (8,14) \\ 12 & (4,27) & (6,29) & (6,30) & (8,30) & (6,29) & (4,24) & (8,13) & (9,17) & (9,16) \\ 13 & (5,30) & (3,30) & (6,30) & (7,30) & (5,28) & (7,30) & (9,15) & (10,18) & (10,20) \\ 14 & (4,30) & (8,27) & (6,30) & (5,27) & (6,29) & (6,30) & (12,29) & (9,19) & (9,19) \\ 15 & (6,30) & (3,29) & (4,29) & (5,29) & (6,30) & (7,30) & (9,18) & (9,18) & (9,17) \\ 16 & (5,30) & (3,30) & (6,30) & (5,28) & (8,30) & (7,30) & (13,29) & (11,25) & (11,22) \\ 17 & (4,29) & (7,29) & (5,30) & (6,30) & (5,27) & (5,26) & (9,19) & (10,21) & (11,24) \\ 18 & (5,30) & (4,30) & (7,30) & (7,30) & (6,29) & (7,30) & (13,30) & (11,24) & (13,29) \\ 19 & (5,30) & (4,30) & (4,30) & (7,30) & (6,30) & (6,29) & (11,23) & (12,29) & (13,29) \\ 20 & (4,29) & (2,29) & (5,29) & (7,30) & (6,30) & (7,30) & (15,30) & (13,29) & (13,29) \\ \hline \end{tabular} \caption{Median $95\%$ credible intervals of the posterior of $\nu$, for simulations with $\nu=1,\ldots,20$, $\alpha=0.3,0.5,0.8$ and $n=50,100,1000$.} \label{tab:median_intervals} \end{table} To have a feeling for the overall estimation procedure, we illustrate in detail the analysis of a single independent and identically distributed sample from a known model. We draw a sample of size $n=200$ from an AST distribution with parameters $\alpha=0.35$, $\mu=2$, $\sigma=1.5$ and $\nu=6$. The posterior distributions are obtained via Monte Carlo methods with 100000 iterations and a burn-in period of 5000 iterations and by considering the objective priors described in Section \ref{sc_priors}. In Figure \ref{fig:single_samp} we have reported, for each parameter, the chain samples and the histogram of the posterior distribution, \begin{figure}[h!] \centering \subfigure[]{% \includegraphics[scale=0.25]{SS_alpha_chain.pdf} \label{fig:sf1}} \quad \subfigure[]{% \includegraphics[scale=0.25]{SS_alpha_hist.pdf} \label{fig:sf2}} \subfigure[]{% \includegraphics[scale=0.25]{SS_mu_chain.pdf} \label{fig:sf3}} \quad \subfigure[]{% \includegraphics[scale=0.25]{SS_mu_hist.pdf} \label{fig:sf4}} \subfigure[]{% \includegraphics[scale=0.25]{SS_sigma_chain.pdf} \label{fig:sf5}} \quad \subfigure[]{% \includegraphics[scale=0.25]{SS_sigma_hist.pdf} \label{fig:sf6}} \subfigure[]{% \includegraphics[scale=0.25]{SS_nu_chain.pdf} \label{fig:sf7}} \quad \subfigure[]{% \includegraphics[scale=0.25]{SS_nu_hist.pdf} \label{fig:sf8}} \caption{Sample chains (left graphs) and histograms of the posterior distributions (right graphs) of the parameters for the simulated data from the AST with $\alpha=0.35$, $\mu=2$, $\sigma=1.5$ and $\nu=6$.} \label{fig:single_samp} \end{figure} while in Table \ref{tab:singlesample} we have the summary statistics of each posterior. In particular, we have computed the posterior mean, the posterior median and the $95\%$ credible interval of the posterior distribution. \begin{table} \centering \begin{tabular}{cccc} Parameter & Mean & Median & $95\%$ C.I. \\ \hline $\alpha$ & 0.36 & 0.36 & (0.25,0.49) \\ $\mu$ & 1.76 & 1.76 & (1.02,2.53) \\ $\sigma$ & 1.79 & 1.78 & (1.33,2.37) \\ $\nu$ & 6.24 & 6 & (3,13) \\ \hline \end{tabular} \caption{Summary statistics of the posterior distributions for the parameters of the simulated data from an AST with $\alpha=0.35$, $\mu=2$, $\sigma=1.5$ and $\nu=6$.} \label{tab:singlesample} \end{table} By inspecting the histograms and the summary statistics of the posterior distributions we can assess on the appropriateness of the inferential process. In particular, the mean (or the median for $\nu$) of the posteriors are very close to the true parameter values, which are well within the limits of the corresponding credible intervals. \section{Real data analysis}\label{sc_real} To show how the discussed model works in practice, we have chosen two well known data sets, both related to insurance loss. The first data set contains 2,167 individual losses each with a value of one million Danish Krone (DKK) or above, collected from January 1980 to December 1990 \citep{Mcneil:1997}. The second data set relates to 1,500 indemnity payments, in thousand of US dollars \citep{FreVal:1998}. \subsection{Danish fire losses}\label{sc_danish} The Danish fire loss data set contains losses due to fire with a single value of at least DKK 1 million. Table \ref{tab:danish_loss1} reports some descriptive statistics of the data, both in the nominal scale and in the log-scale. It is common practice, when analysing insurance data (and not only), to consider the logarithm of the data for modelling purposes as this results in a reduction of skewness \citep{Bolance:2008}. Although the skewness index is drastically reduced by the log-transform of the Danish loss data, as shown in Table \ref{tab:danish_loss1}, its value still indicates a significantly positive skewness. The above result is easily noticeable by inspecting the histograms of the data in Figure \ref{fig:Danish_hist}. \begin{table} \centering \begin{tabular}{lcc} & Danish & Danish (log-scale) \\ \hline Mean & 3.39 & 0.79 \\ Standard Deviation & 8.51 & 0.72 \\ Skewness & 18.75 & 1.76 \\ \hline \end{tabular} \caption{Descriptive statistics of the Danish loss data set in millions of Danish Krone (left) and in the log-scale (right).} \label{tab:danish_loss1} \end{table} Suitable statistical tests can be performed to support the conclusion of departure for normality, such as the Jarque--Bera test for normality \citep{JarBer:1980}, the D'agostino test for skewness \citep{Dago:1970} or the Anscombe-Glynn test of kurtosis \citep{AnsGly:1983}, for example (results not reported here). \begin{figure}[h!] \centering \subfigure[]{% \includegraphics[scale=0.30]{Danish_loss.pdf} \label{fig:sub1}} \quad \subfigure[]{% \includegraphics[scale=0.30]{Danish_loss_log.pdf} \label{fig:sub2}} \caption{Histograms of the Danish fire loss data (left) and of the same data set in the log-scale (right).} \label{fig:Danish_hist} \end{figure} The prior distributions used to analyse the Danish fire loss data set where in line with the overall objective approach discussed in the paper. In particular, we used the Jeffreys' prior for the skewness parameter $\alpha$, that is $\pi(\alpha)\sim \mbox{Beta}(1/2,1/2)$, the discrete truncated prior for the number of degrees of freedom $\nu$, and the reference prior for the pair location-scale parameters $(\mu,\sigma)$, that is $\pi(\mu,\sigma)\propto 1/\sigma$. \begin{figure}[h!] \centering \includegraphics[scale=0.30]{PredictiveDanishData.png} \caption{Real Data (dashed line) vs Posterior Predictive Distribution (dotted line) of the Danish Loss data. \label{fig:predDan}} \end{figure} The marginal posterior distributions for the parameters, as they are analytically intractable, have been obtained via Monte Carlo methods by applying the algorithm described in the Appendix. We have run multiple chains for each parameter, with different sparse starting points. In particular, we have run 500000 iterations and considered a burn in of 100000. The convergence has been assessed by computing the Gelman and Rubin's statistics \citep{BroGel:1998,GelRub:1992} and monitoring the posterior running means. \begin{figure}[h] \centering \subfigure[]{% \includegraphics[scale=0.30]{Danish_alpha_hist.pdf} \label{fig:dan_hist_sub1}} \quad \subfigure[]{% \includegraphics[scale=0.30]{Danish_mu_hist.pdf} \label{fig:dan_hist_sub2}} \subfigure[]{% \includegraphics[scale=0.30]{Danish_sigma_hist.pdf} \label{fig:dan_hist_sub3}} \quad \subfigure[]{% \includegraphics[scale=0.30]{Danish_nu_hist.pdf} \label{fig:dan_hist_sub4}} \caption{Histograms of the posterior distributions for $\alpha$, $\mu$, $\sigma$ and $\nu$ for the Danish fire loss data set (in the log-scale).} \label{fig:danish_post_hist} \end{figure} We have reported the histograms of the posterior marginal distributions for the parameters of the model in Figure \ref{fig:danish_post_hist}, and the corresponding summary statistics in Table \ref{tab:danish_post_summary}. \begin{table} \centering \begin{tabular}{cccc} \hline Parameter & Mean & Median & $95\%$ C.I. \\ \hline $\alpha$ & 0.0005 & 0.0003 & (0.0001,0.0020) \\ $\mu$ & 0.0042 & 0.0004 & (0.0000,0.0378) \\ $\sigma$ & 0.4142 & 0.4123 & (0.3906,0.4359) \\ $\nu$ & 9.4900 & 9 & (8,12) \\ \hline \end{tabular} \caption{Summary statistics of the posterior distribution of the Danish fire loss data set in the log-scale.} \label{tab:danish_post_summary} \end{table} As expected, the value of the skewness parameter is very close to zero. In fact, both from the data histogram and summary statistics, it is possible to deduce that the data has a strong positive skewness. The median of the posterior of the number of degrees of freedom $\nu$ indicates a heavy-tailed behaviour in the observations. Both the skewness and heavy-tail results are consistent with the expected behaviour of insurance loss data, even after the data has been log-transformed, in this case. Figure \ref{fig:predDan} shows the posterior predictive density against the real data. We note no substantial difference in the two curves. Finally, to compare the data statistics of Table \ref{tab:danish_loss1} with the MCMC estimation, we have computed the Monte Carlo estimates (in the log-scale) of the mean, the standard deviation and the skewness index, obtaining, respectively, the values 0.79, 0.72, 1.77. \subsection{US indemnity loss}\label{sc_US} The second data set we analyse is widely used in the literature and it contains US indemnity losses publicly available \citep{FreVal:1998}. It contains 1,500 liability claims each of which with an associated indemnity payment in thousands of US dollars (USD). \begin{table} \centering \begin{tabular}{lcc} & US & US (log-scale) \\ \hline Mean & 41.21 & 2.46 \\ Standard Deviation & 102.75 & 1.64 \\ Skewness & 9.16 & -0.15 \\ \hline \end{tabular} \caption{Descriptive statistics of the US loss data set in thousands of USD (left) and in the log-scale (right).} \label{tab:usloss1} \end{table} Table \ref{tab:usloss1} shows the descriptive statistics of the US loss data set, both in thousands of USD and in the log-scale. As we did for the Danish fire loss data, we perform the analysis on the log-transformation of the observed values. Figure \ref{fig:US_hist} shows the histogram of the US loss data (left plot) and of the same data in the log-scale (right plot). From the first histogram on the left it is possible to see the typical behaviour of this type of data, that is a relatively high number of losses with a small value and a few losses of large value. \begin{figure}[ht] \centering \subfigure[]{% \includegraphics[scale=0.30]{USloss_hist.pdf} \label{fig:subfigur1}} \quad \subfigure[]{% \includegraphics[scale=0.30]{USlogloss_hist.pdf} \label{fig:subfigur2}} \caption{Histograms of the US loss data (left) and of the same data set in the log-scale (right).} \label{fig:US_hist} \end{figure} The procedure followed to analyse the US loss data, in the log-scale, is analogous to the one employed to analyse the Danish fire loss data set in Section \ref{sc_danish}. The histograms of the posterior distributions of the parameters of the model are in Figure \ref{fig:US_post_hist}, with the corresponding summary statistics in Table \ref{tab:US_post_summary}. \begin{figure}[h] \centering \subfigure[]{% \includegraphics[scale=0.30]{US_alpha_hist.pdf} \label{fig:US_hist_sub1}} \quad \subfigure[]{% \includegraphics[scale=0.30]{US_mu_hist.pdf} \label{fig:US_hist_sub2}} \subfigure[]{% \includegraphics[scale=0.30]{US_sigma_hist.pdf} \label{fig:US_hist_sub3}} \quad \subfigure[]{% \includegraphics[scale=0.30]{US_nu_hist.pdf} \label{fig:US_hist_sub4}} \caption{Histograms of the posterior distributions for $\alpha$, $\mu$, $\sigma$ and $\nu$ for the US loss data set (in the log-scale).} \label{fig:US_post_hist} \end{figure} \begin{table} \centering \begin{tabular}{cccc} \hline Parameters & Mean & Median & $95\%$ C.I. \\ \hline $\alpha$ & 0.52 & 0.52 & (0.48,0.56) \\ $\mu$ & 2.47 & 2.48 & (2.01,2.65) \\ $\sigma$ & 1.52 & 1.52 & (1.44,1.59) \\ $\nu$ & 27.16 & 28 & (21,30) \\ \hline \end{tabular} \caption{Summary statistics of the posterior distribution of the US loss data set (in the log-scale).} \label{tab:US_post_summary} \end{table} We note the following two important results. First, the skewness parameter is estimated to be close to 0.5. This indicates that a symmetric model would be suitable for this data set. Second, the estimated number of degrees of freedom is very closed to the upper bound of the parameter space. This is a clear indication that the data could be represented by a $t$ density with a high number of degrees of freedom or, which is equivalent, by a normal density. For what it concerns the inferential results, we see that the credible intervals are relatively narrow, indicating a relatively strong posterior beliefs about the obtained estimates. Figure \ref{fig:predUS} shows the posterior predictive density against the real data, where again we do not see any substantial difference between the two curves. As done in Section \ref{sc_danish}, we have performed Monte Carlo estimates of the data statistics fro the US Loss data set. In particular, we have computed the mean, the standard deviation and the skewness index obtaining, respectively, 2.45, 1.64, -0.15. These values can be compared with the ones in Table \ref{tab:usloss1}. \begin{figure}[h!] \centering \includegraphics[scale=0.30]{PredictiveUSData.png} \caption{Real Data (dashed line) vs Posterior Predictive Distribution (dotted line) of the US Loss data. \label{fig:predUS}} \end{figure} \section{Discussion}\label{sc_disc} An important aspect in analysis insurance loss data is their tendency to follow a skewed distribution and, because extreme events are not uncommon, to exhibit heavy tails. However, in some circumstances, symmetrical models with either non-heavy tails, like for example the normal distribution, or heavy tails, such as the $t$ density, may be effectively employed. Usually, this type of modelling can be achieved by considering the data in the log-scale. Although competing models could be estimated and assessed for their inferential and predictive performances, it is appealing to be able to consider a single model which, on the basis of the estimated values of some of the parameters, ``adjusts'' itself, in a sort of an automated fashion, to the problem under consideration. The asymmetrical Student-$t$ considered in this paper to represent insurance loss data has the above flavour. In fact, the estimated value of the skewness parameter $\alpha$ would suggest a symmetrical or non-symmetrical scenario, and the number of degrees of freedom would indicate the fatness of the tails of the data. When dealing with a $t$ distribution, whether the usual symmetrical one or the one considered here, the estimation of the number of degrees of freedom has always been a challenge. It is not uncommon that the problem is somehow eluded by either setting $\nu$ on the basis of some appropriate theoretical results, or by consider different values and chose the most appropriate on the basis of some criteria. A Bayesian approach, and in particular an objective Bayesian approach, allows to obtain reliable estimates of the number of degrees of freedom with minimal initial input. We have here consider the number of degrees of freedom discrete and bounded above on the basis of the well known property of the $t$ density to converge to the normal distribution for sufficiently large $\nu$. With this consideration, we have been able to consider the objective prior for $\nu$ presented in \cite{Villa:Walker:2014a}. In addition, we prove here that the prior is not dependent on the value of the skewness parameter $\alpha$, therefore directly applicable to the model here considered. We have studied the frequentist properties of the posterior for $\nu$ yielded by the truncated objective prior. As expected from the analytical result discussed in Section \ref{sc_priors}, different values of the skewness parameter do not affect the performance of the prior, except regular small variations due to the randomness of the repeating sampling. As expected, the performance of the prior distribution, in terms of MSE, improves when the sample size increases. The above results are in line with the ones obtained in \cite{Villa:Walker:2014a}. We have then employed the discussed model to analyse real insurance loss data. The work has been carried out by using objective priors for each parameter of the model, to ensure a minimal information approach to the problem. In the first illustration we look to a well-known data set of losses due to fire in Denmark. The peculiarity of the data is to show strong (positive) skewness and to have extreme values. Even by considering the logarithmic transformation of the observations skewness and extremeness cannot be removed. The inferential procedure shows the model adjustment to the scenario by resulting in an $\alpha$ very close to zero and a number of degrees of freedom equal to 9. The same Bayesian model, i.e. sampling distribution and prior distribution, is applied to a different data set. This data set, which contains indemnity losses in the US market, shows skewness and extreme values as well, but these appear to be removed once the logarithm of the observations is considered. A posterior mean of $\alpha=0.5$ indicates a symmetrical distribution, and a posterior median of $\nu=28$ indicates a distribution that is not very different from a normal density. The above two results show how the same model can be applied to insurance loss data sets, without the need to change neither any of of its components nor the prior distributions for the parameters. To conclude, it is obvious that the same approach here illustrated can be adopted to other types of data, which exhibit similar characteristics of skewness. That is, the objective Bayesian analysis of data by means of the AST model, including both the distribution and the priors for the parameters, can be generalised and employed in other disciplines, such as finance, environmental sciences and engineering, for example. \section*{Acknowledgements} The authors are thankful to the Associate Editor and the anonymous reviewer for their useful comments which significantly improved the quality of the paper. The authors are very grateful to F. J. Rubio for all the stimulating discussions and suggestions. Fabrizio Leisen's research has been supported by the European Community's Seventh Framework Programme[FP7/2007-2013] under grant agreement no: 630677.
10c719ed75e2abe1b9cd25954114801efe8c331d
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{introduction} For future fusion devices such as ITER, operating at least in a partially detached state is important for reducing the heat flux incident on the divertor to below engineering limits (10 $\text{MW/m}^2$) \cite{LOARTE2007NF}. Modelling for ITER demonstrates reduction of the peak heat flux near the separatrix by factors of up to 100 due to a number of atomic physics processes including line radiation, charge exchange and recombination \cite{KUKUSHKIN2013JNM}. To address the need for further power removal before exhaust heat reaches the targets, which is needed for a DEMO fusion reactor and beyond, an enhanced understanding of the detachment process would be beneficial, which will enable better models for predicting ITER and DEMO performance and potentially provides insight in enhancing both detachment power and particle loss as well as control of detachment. There has been considerable work utilizing spectroscopic measurements for understanding detachment, where the characteristics of the recombining region are extracted from the Balmer series emission \cite{LIPSCHULTZ1999POP,MCCRACKEN1998NF,MEIGS2013JNM,LOMANOWSKI2015NF,POTZEL2014NF}. Typically a high density recombination front forms at the target and moves rapidly towards the x-point as the core plasma density is increased. The aim of this study is to develop a detailed understanding of the detachment process at TCV (medium-sized tokamak ($R=0.89 \text{ m}$, $a = 0.25 \text{ m}$, $B_t = 1.4 \text{ T}$)) where low densities should give us insight into how the role of recombination changes as a function of plasma density in the divertor. In addition, this provides an understanding of how detachment on TCV relates to the general experience of detachment. This is required to interpret recent experiments on TCV, which have been performed to investigate how magnetic divertor geometry influences detachment \cite{THEILER2016NF, REIMERDES2016FEC}. For this investigation a new spectroscopic diagnostic has been developed for the TCV divertor and improvements for extracting information on recombination and electron temperature from Balmer series spectra have been made. Using spectroscopic measurements we show that the observed high density recombination front at TCV during a density ramp likely stays near the target even after the target ion current drops. \section{Experimental setup} \label{ExpSetup} \subsection{TCV's Divertor Spectroscopy System (DSS)} \label{DSS} The primary measurements of the recombination characteristics are made using a new spectrometer with views of the divertor, which we refer to as the DSS. The viewing optics provide a poloidal, line-integrated, view of the divertor, yielding 32 lines of sight (figure \ref{fig:SpectraFigure}B). The fibres of each system are coupled to a Princeton Instruments Isoplane SCT 320 spectrometer coupled to an Andor iXon Ultra 888 EMCCD camera with a 1024 x 1024 pixel sensor. A 1800 l/mm grating was used to allow $n_e$ measurements through Stark broadening of the n=7 Balmer series line with a measured FWHM resolution of 0.06 nm. The system has been absolutely calibrated in intensity ($\sim$15\% inaccuracy) and wavelength ($<0.1$ nm), taking stray light contributions into account. A dark frame is acquired before and after the plasma discharge, which is subtracted from the measurements. Due to the long frame transfer time (1.2 ms) with respect to the acquisition time (5 - 10 ms), the measurements are susceptible to read-out smear of the CCD \cite{DORRINGTON2005}. At least 90\% of the smearing is removed by post-processing using a numerical matrix-based algorithm. For the results analysed in this work, the measured spectra have been re-sampled by averaging frames and/or multiple chordal signals over the entire discharge, improving S/N ratio by up to a factor 40 which leads to improved determination of $n_e$ from line fitting (section \ref{NeStark}). Figure \ref{fig:SpectraFigure}A shows that the observed intensity of medium-n Balmer lines ($n=6,7$) increases strongly during the density ramp. The observed spectra corresponds to the view line close to the target highlighted in figure \ref{fig:SpectraFigure}B (red), where the locations of the primary diagnostics used in this work are shown. \begin{figure}[H] \centering \includegraphics[scale=1.4]{SpectraFigure2} \caption{A) Example Balmer line spectra \# 52065, averaged over 100 ms, measured by along the DSS chord closest to the target at two different core densities. B) Primary diagnostic viewing chords used in this work.} \label{fig:SpectraFigure} \end{figure} \subsection{Extracting information on recombination from Balmer lines using a collisional-radiative model} \label{CollRadMod} The brightness ($B_{n\rightarrow2}$ in $[\text{photons } \text{m}^{-2} \text{s}^{-1}]$) of a hydrogen Balmer line with quantum number $n$ can be modelled using the Photon Emissivity Coefficients ($PEC_{n\rightarrow2}^{rec,exc}$) [$\text{photons } \text{m }^{3} \text{s }^{-1}$] obtained from the ADAS collisional-radiative model \cite{ADAS} for recombination and excitation, as indicated in equation \ref{eq:EmissEQ}. $B_{n\rightarrow2}$ consists of recombination and excitation parts: $B_{n\rightarrow2}^{rec,exc}$. It is assumed that all line emission comes from a plasma slab with spatially constant electron density $n_e$, electron temperature $T_e$, neutral density $n_o$ and width $\Delta L$. Additional assumptions are that hydrogen collisional radiative model results are valid for deuterium and that the contribution of charge exchange and molecular reactions (molecular reactions might be significant for detachment in low density plasmas \cite{KUKUSHKIN2016PSI}) to the emission of a certain Balmer line are negligible. For simplicity we have assumed all electrons come from hydrogen ($Z_{eff}=1$), which is discussed in section \ref{impeffect}. For further discussion we define $F_{rec}$ as the fraction of total Balmer line radiation due to recombination ($F_{rec} (n) = B_{n \rightarrow 2}^{rec} / B_{n \rightarrow 2}$). We also define $F_{76}$ as the ratio of brightness of the $7\rightarrow2$ and $6\rightarrow2$ Balmer lines ($F_{76} = B_{7\rightarrow2}/B_{6\rightarrow2}$). We define $R_L$ $[\text{rec } / \text{ s} \text{ m}^2]$ as the volumetric recombination rate ($R$ $[\text{rec } / \text{ s} \text{ m}^3]$) line integrated along the line of sight through the plasma for a length $\Delta L$. Although the analysis in this section is mainly focused on the $n = 6,7$ Balmer lines, the analysis strategy is general and can be applied to other Balmer lines. \begin{equation} B_{n \rightarrow 2} = \underbrace{\Delta L n_e^2 PEC_{n->2}^{rec} (n_e, T_e)}_{B_{n \rightarrow 2}^{rec}} + \underbrace{\Delta L n_o n_e PEC_{n->2}^{exc} (n_e, T_e)}_{B_{n \rightarrow 2}^{exc}} \label{eq:EmissEQ} \end{equation} \subsubsection{Using Balmer line ratios to obtain the fraction of Balmer line emission due to recombination} \label{Frec} We have developed a method for inferring the recombination contribution to the Balmer line emission, which is important for determining several characteristics of the local plasma. For a fixed $n_e$ and $n_o$, both $F_{rec} (n)$ and $F_{76}$ only depend on $T_e$. In figure \ref{fig:LineRat} the relation between $F_{rec} (n)$ and $F_{76}$ is shown, where $T_e$ is varied between 0.2 and 1000 eV for each curve, while $n_o/n_e = [10^{-3}, 1]$ and $n_e = 10^{20} \text{ m}^{-3}$ are fixed. Figure \ref{fig:LineRat} indicates the ratio of two Balmer lines (e.g. $F_{76}$) changes as function of $F_{rec}$ and is thus useful to infer the dominance of recombination in the total emission of a particular Balmer line. The relation between $F_{rec} (n)$ and $F_{76}$ depends only weakly on $n_e$ and $n_o/n_e$. Divertor pressure measurements with an absolutely calibrated baratron gauge have been used to estimate $n_o$ and indicate $n_o/n_e$ rises from order $10^{-3}$ to order $10^{-1}$ as $T_e$ drops, which is supported by OSM-Eirene modelling \cite{HARRISON2016PSI} and SOLPS-Eirene modelling \cite{WISCHMEIER2005EPFL} of the TCV divertor. Based on the above $n_o/n_e$ estimates, we utilize a $n_o/n_e$ range between 0.01 and 0.25. In this $n_o/n_e$ range and the typical TCV divertor density range (between $10^{19} m^{-3}$ and $10^{20} m^{-3}$) $F_{rec}$ changes by $<0.1$. When inferring $T_e$ and $R_L$ from either $6,7\rightarrow2$ lines (using $F_{rec} (n=6,7)$) the result differs by $<3\%$. Line integration effects (section \ref{ProfEffect}) are negligible to the determination of $F_{rec}$. Although $n_o/n_e$ only weakly influences the relation between $F_{rec} (n)$ and $F_{76}$, it strongly affects the temperature dependence of both $F_{rec} (n)$ and $F_{76}$, as shown in figure \ref{fig:LineRat}. Therefore, to determine $T_e$ from $F_{76}$ an accurate $n_o/n_e$ determination is needed, but is not currently possible. In section \ref{Recom}, we develop another method to derive $T_e$. Note that, even if $n_o/n_e$ would be accurately known, the $T_e$ obtained would be line averaged and weighted over both the excitation and recombination part of the Balmer line emission profile along the line of sight. This is in contrast to the $T_e$ determination described in section \ref{Recom}, where only the recombination part of the Balmer line emission is taken into account. \begin{figure}[H] \centering \includegraphics[scale=1.4]{LineRatTe2} \caption{Relation between $F_{76}$ and $F_{rec} (n=6,7)$, in which $T_e$ is varied between 0.2 and 1000 eV, assuming a fixed $n_e = 10^{20} m^{-3}$ and $n_o/n_e = [10^{-3}, 1]$. The value of $T_e$ at $F_{rec} (n=7) = 0.3$ and $F_{rec} (n=7) = 0.8$ is shown for both values of $n_o/n_e$ at $n_e = 10^{20} m^{-3}$.} \label{fig:LineRat} \end{figure} \subsubsection{Obtaining $R_L$ and $T_e^{avg}$ from absolute Balmer line intensities} \label{Recom} We have developed a method for calculating $R_L$, which has the advantage over previous work \cite{TERRY1998POP} that no direct temperature estimate is required in the calculation. The first step is to determine the number of recombinations per photon as in \cite{TERRY1998POP} for a particular Balmer line, which is (assuming the plasma is optically thin) the ratio of the ADAS effective recombination rate coefficient ($ACD (n_e, T_e)$), which takes into account both radiative and three body recombination, and the ADAS $PEC_{n \rightarrow 2}^{rec} (n_e, T_e)$. By multiplying the number of recombinations per emitted photon with $B_{n\rightarrow2}^{rec}$, we obtain $R_L (n_e, T_e, \Delta L)$ [$\text{rec } / \text{m}^2 \text{s}$]. Once $F_{rec}$ is determined from figure \ref{fig:LineRat}, we can obtain $B_{n\rightarrow2}^{rec} = F_{rec} (n) \times B_{n\rightarrow2}$, from which we can derive other important characteristics of the plasma along each chord. With fixed $n_e$ and $\Delta L$, both $R_L$ and $B_{n\rightarrow2}^{rec}$ only depend on $T_e$ and a one-to-one relationship between $R_L$ and $B_{n\rightarrow2}^{rec} $ is obtained (figure \ref{fig:RecRate}). In addition, as $T_e$ varies along each curve in figure \ref{fig:RecRate} from 0.2 to 1000 eV, $T_e$ is also obtained when determining $R_L$. We refer to this as $T_e^{avg}$ as it is line averaged and weighted by the recombination part of the Balmer line emission profile along the line of sight. Using $n_e$ (Stark broadening - section \ref{NeStark}), $\Delta L$ and $B_{n\rightarrow2}^{rec}$ both $R_L$ and $T_e^{avg}$ can be determined. As shown in figure \ref{fig:RecRate}, determining $R_L$ through this method is only weakly affected by $n_e$ and $\Delta L$. The measurement inaccuracy of $R_L$ is generally $\sim$ 40 \% when $F_{rec} \sim 1$ and is mostly due to the inaccuracy in $B_{n\rightarrow2}^{rec}$, which is affected by inaccuracies in both the absolute Balmer line intensity and the Balmer line ratio used to obtain $F_{rec}$. Line integration effects influence $R_L$ by $< 5 \%$, except for cases with a strongly hollow $n_e$ and peaked $T_e$ profile, where $R_L$ can be underestimated by up to 30 \% (section \ref{ProfEffect}). A similar approach as described here could be used to obtain excitation rates and track the excitation region, but with larger uncertainties. Obtaining $T_e^{avg}$ through the method above has the advantage that less spectral information is needed to obtain $T_e$ than for other methods \cite{LIPSCHULTZ1999POP,TERRY1998POP}. However, this method is sensitive to inaccuracies in $\Delta L$ and is strongly affected by line-integration effects. Assuming peaked $n_e, T_e$ profiles along the line of sight $T_e^{avg}$ is in between 50-100 \% of the peak $T_e$ if $F_{rec} \sim 1$ (section \ref{ProfEffect}). $T_e^{avg}$ should not be used as an absolute $T_e$ measurement, but as an indicator for trends in $T_e$ which shows the role $T_e$ plays in the increase of $R_L$ during a density ramp discharge. We define $\Delta L$ as the full-width $1/e$ fall-off length of the $n_e$ profile at the target measured by Langmuir probes, which is mapped along the flux surfaces to determine $\Delta L$ for each point where the DSS view line intersects with the separatrix at multiple time points. $\Delta L$ for TCV is generally between 2.5 and 10 cm, depending on the magnetic equilibrium used for that pulse and time. During a density ramp the density profile in the divertor broadens. Together with a constant magnetic equilibrium, $\Delta L$ can increase by up to 70 \%. As a trend in the density profile at the target is not necessarily representable for trends in the density profile across the divertor leg above the target, both the Langmuir probe spatial resolution and experimental variations in $\Delta L$ during a single discharge with constant magnetic equilibrium are used to estimate the uncertainty of $\Delta L$, which makes up at most $ 25 \%$ of the measurement uncertainty in $R_L$. \begin{figure}[H] \centering \includegraphics[scale=1.4]{RecRate3} \caption{Modelled relation between the $R_L$ and $B_{7\rightarrow2}^{rec}$ for a range of different $n_e$ and $\Delta L$.} \label{fig:RecRate} \end{figure} \subsection{Obtaining $n_e$ from Stark broadening} \label{NeStark} The spectrally resolved line profile is affected by Stark broadening. Our chordal measurement provides a density weighted integral of contributions to the line shape and thus of the electron density \cite{LOMANOWSKI2015NF} $(n_e^{Stark})$. The Stark broadened line shape of a Balmer line can be expressed as a modified Lorentzian \cite{LOMANOWSKI2015NF} as function of $n_e$ and $T_e$, which is a parametrisation of the Microfield Model Method \cite{STEHLE1999AASS}. The spectrometer induces additional instrumental broadening to the emitted spectral line, which is parametrized using a modified asymmetric Lorentzian whose parameters are obtained as function of wavelength and viewing chord. The experimentally observed Balmer line shape is fitted using a numerical algorithm based on the Gradient Expansion Algorithm \cite{BEVINGTON2003}. The fitting function used is the convolution of Stark broadening, Doppler broadening (depends on $T_i$) \cite{KUNZE2009} and the instrumental line shape. Magnetic effects are neglected. To lower the amount of fitting parameters it is assumed $T_e = T_i = 3 \text{ eV}$. For $T_e$ between 0.6 and 15 eV, the variations in $n_e^{Stark}$ are $<7\%$. For $T_i$ between 0.2 and 15 eV the variations in $n_e^{Stark}$ are $<10\%$. Assuming peaked $n_e$ profiles, $n_e^{\text{Stark}}$ is in between 65 - 100 \% of the peak $n_e$ (section \ref{ProfEffect}). The main parameter leading to measurement uncertainty in $n_e^{Stark}$ is the signal/noise level. By fitting synthetic spectra with a level of random noise, we have determined the measurement uncertainty of $n_e^{Stark}$ as function of $n_e$, S/N level and viewing chord. We utilize the $7\rightarrow2$ line for determining $n_e^{Stark}$ since, for the same $n_e$, higher-n Balmer lines lead to wider line shapes, which are more accurately analysed. \subsection{Investigating line-integration effects on $n_e^{Stark}$, $T_e$ and recombination measurements} \label{ProfEffect} The sensitivity of the $n_e^{Stark}$, $T_e^{avg}$, $F_{rec}$ and $R_L$ inferences to line-integration effects have been discussed in sections \ref{Frec}, \ref{Recom} and \ref{NeStark}. These sensitivities have been determined using the methods described in this section. Line integration effects have been studied by assuming various a priori peaked and hollow $n_e$, $T_e$ profiles along the integration chord. For peaked profiles Gaussian profile shapes have been assumed with widths varying from 0.5 to 7 cm using peak densities: $n_{e,0} = [3, 5, 10] . 10^{19} \text{ m}^{-3}$ and corresponding peak temperatures: $T_{e,0} = [15, 3, 1] \text{ eV}$. A flat neutral density profile using $n_o = [10^{18}, 10^{19}] \text{ m}^{-3}$ has been assumed. Using these profiles, the Balmer line emission is modelled at every point of the profile and the corresponding Stark line shape is calculated. The Stark line shapes, weighted by the Balmer line emission, are summed over all points of the profile to obtain a synthetic Balmer line spectrum. $n_e^{Stark}$, $R_L$, $F_{rec}$ and $T_e$ are inferred from the synthetic spectrum using the methods described in sections \ref{CollRadMod} and \ref{NeStark}. \subsection{The role of impurity concentration on inferred results} \label{impeffect} For simplicity in section \ref{CollRadMod} it has been assumed that the hydrogen ion density equals the electron density ($n_H^{+} = n_e$). However, a portion of the electrons can originate from plasma impurities. This can be taken into account by replacing the $n_e^2$ term in $B_{n\rightarrow2}^{rec}$ (equation \ref{eq:EmissEQ}) with $n_e n_H^{+}$, which can be written as $f n_e^2$, where $f=n_H^{+}/n_e$. By including $f$ in equation \ref{eq:EmissEQ} and propagating the effect of $f$ towards the inference of $R_L$, $F_{rec}$ and $T_e^{avg}$ the role of the impurity concentration has been investigated. Based on $Z_{eff}$ measurements and fractional abundance modelling through ADAS (using carbon and boron as the main plasma impurity species) we estimate that $f$ is in between 0.6 and 1.0. For this range $F_{rec}$ differs by 0.01, $R_L$ differs by 10 \% and $T_e^{avg}$ differs by 20 \%. Therefore, the impurity concentration is expected to have an effect on the inferred results which is small compared to the estimated uncertainty margins. \section{Experimental results} \label{ExpResults} In this section we will use the DSS data and analysis techniques described in section \ref{ExpSetup} to illustrate how divertor conditions vary as detachment proceeds in TCV. Connections will be made to other diagnostic measurements to form a more complete picture of the detachment process. Observations of the Balmer line intensity ($B_{n\rightarrow2}$) and the inferred $F_{rec}$ from $F_{76}$ presented in this section correspond to the $n=7$ Balmer line. Our observation is that some of the characteristics of detachment on TCV are similar to that found at other, higher density, tokamaks. However, detachment in TCV does not lead to a large movement of the recombination region. \subsection{Onset, evolution and dynamics of detachment} \label{DetachDyn} A reference plasma discharge is utilized for illustrating the process of detachment in TCV (\#52065). It has a single null magnetic divertor geometry with a plasma current of 340 kA and a reversed toroidal field direction ($\nabla B$ away from the x-point). The spectroscopic data has been acquired at 200 Hz and has been averaged over a number of time frames to improve S/N level, as indicated in the legends in figure \ref{fig:Overview50648}. The line colour and line style shown in figure \ref{fig:Overview50648}A-H correspond to the diagnostic locations shown in figure \ref{fig:SpectraFigure}B. Similar detachment characteristics as observed for \#52065 have been found for $\sim 20$ other density ramp discharges, with slight variation in timing of changes (e.g. drop in target density as determined by Langmuir probes) and magnitude (e.g. the total recombination). The vertical error bars shown in figures \ref{fig:Overview50648}A-H represent 95 \% confidence intervals. Measurement uncertainties have been determined by propagating measurement uncertainties in the absolute calibration; in fit parameters (determined through the Gradient Expansion Method \cite{BEVINGTON2003}); in $\Delta L$ and assuming $n_o/n_e$ is in between 0.01 and 0.25. Bolometry and spectral features consistently indicate an expansion of a cold plasma region from the target towards the x-point during a density ramp. During a considerable increase in Greenwald fraction from $\overline{n_e}/n_G = 0.3$ to $\overline{n_e}/n_G = 0.5$ (figure \ref{fig:Overview50648}A, $F_{76}$ increases resulting in an increase in $F_{rec}$ from $<0.35$ to $\sim 1$ (figure \ref{fig:Overview50648}C). Significant increases in $F_{rec}$ first occur near the target and later the region of enhanced $F_{rec}$ expands towards the x-point. The radiation front as measured by bolometry (figure \ref{fig:Overview50648}E), which is representative of higher temperatures than those at which recombination occurs \cite{HARRISON2016PSI,STANGEBY2000}, also moves from the target towards the x-point and is correlated with the increase in $F_{rec}$. The above spectral features are consistent with a strong recombining region near the target. Those features include a strong increase in $B_{n\rightarrow2}$ (figure \ref{fig:Overview50648}B) which, combined with a rising $F_{rec} (n=7)$, implies that $R_L$ (figure \ref{fig:Overview50648}G) is strongly increasing. Similar to trends in $F_{rec}$, the onset of this non-linear increase starts first close to the target and later increases closer to the x-point. The increase in $B_{n \rightarrow 2}$ during the density ramp is both due to the $n_e$ increase (figure \ref{fig:Overview50648}D) and $T_e^{avg}$ decrease (figure \ref{fig:Overview50648}H). Our results suggest that recombination is insufficient to effectively reduce the particle flux at the time of the particle flux roll-over. Furthermore, the (Stark) density close to the target does not decrease. After $F_{rec} \rightarrow 1$, $B_{n \rightarrow 2}$, $R_L$ and $n_e^{Stark}$ keep increasing until the end of the discharge while remaining highest at the lowest DSS chord 5 cm above the target. At first glance this and bolometric measurements (figure \ref{fig:Overview50648}E) would seem to indicate that while ionization and impurity radiation have detached from the target, the high density region has not. However, Langmuir probe data (taken from the probe closest to the separatrix) suggests the density has dropped at the target (figure \ref{fig:Overview50648}) as discussed in section \ref{RecLP}. It is possible that the inferred $R_L$ is an underestimate, since the closest target DSS view line intersects the separatrix 5 cm above the strike point. If the $R_L$ spatial profile is extrapolated to the target, $R_L$ at the target is three times higher than at the DSS chord closest to the target. However, target probe measurements (figure \ref{fig:Overview50648}D) indicate $n_e$ drops in this non-observed region, which would lower $R_L$. Combining LP data and spectroscopic data (section \ref{RecLP}) suggests that either the electron temperature at the target is very low ($< 0.06$ eV) or the high-density recombination front has moved off-target and is located in the region between the target and the lowest DSS chord. Detachment in TCV has so far never reached the level where the density and recombination region peak moves to points above the lowest DSS chord. The total recombination rate in the divertor $R_V$ [rec./s] is determined by integrating $R_L$ toroidally and poloidally across the chords. $R_V$ increases strongly during the last phase of the discharge (figure \ref{fig:Overview50648}F), and reaches values of up to $R_V = (6 \pm 2) \cdot 10^{21}$ rec/s, which is similar to the total particle flux measured by Langmuir probes at that time, indicating that $R_V$ contributes significantly to the particle flux drop at this time. However, the particle flux measured by the Langmuir probes drops at 1.0 s. $R_V$ at that time is relatively low, which indicates that recombination losses are not the main contributor to the initial particle flux drop. \begin{figure} \centering \includegraphics[scale=1.1]{52065_NewData} \caption{Temporal evolution of several quantities measured by DSS and inferred from DSS measurements for three view lines during a single null density ramp shot (52065). In addition, data obtained from Langmuir probes (LP) and bolometry is shown. The colours of the plot indicate the measurement locations shown in figure \ref{fig:SpectraFigure}B.} \label{fig:Overview50648} \end{figure} \subsection{Recombination signatures compared with Langmuir probe data} \label{RecLP} Combining data from the DSS and divertor target Langmuir probe data is informative about the development of detachment. The peak target electron density determined from Langmuir probe (LP) I-V characteristics is in agreement with $n_e^{Stark}$ near the target until 0.9 s (figure \ref{fig:Overview50648}D), which is close to the time when $B_{n\rightarrow2}$ starts increasing strongly. Across many tokamaks it has been found that the temperature derived from Langmuir probes is overestimated for $T_e < 5 \text{ eV}$, \cite{BATISHCHEV1997POP,BATISHCHEV1996POP}. Assuming this is also true for TCV, we utilize $J_{sat}$ and $n_e^{Stark}$ (5 cm from the target) to calculate $T_e^{\text{mod}}$ (figure \ref{fig:Overview50648}H). $T_e^{\text{mod}}$ decreases during the density ramp in agreement with the Balmer line derived $T_e^{avg}$ up until 1.1 s when both the target particle flux and target density (LP) have started dropping. Near the end of the discharge, $T_e^{mod}$ reaches temperatures below 0.06 eV, much lower than $T_e^{avg}$ obtained from Balmer line analysis ($\sim 0.5$ eV). In addition it should be noted that the peak density in the density profile along the line of sight is likely higher (up to 35 \%) than the density inferred from Stark broadening due to the weighted average along the chord, which would lead to an even lower $T_e^{mod}$. Therefore, if $T_e$ at the target would be higher than 0.06 eV, it would imply the target density would be lower than $n_e^{Stark}$. Hence, the density front would have moved between the target and the first DSS chord. \section{Discussion} \label{Disc} The onset of detachment observed spectroscopically at TCV is generally similar to the dynamics previously observed at higher density machines, but there are also significant differences. As the core density is increased in L-mode plasmas, the target density increases and the temperature decreases, which are general characteristics of a high-recycling divertor. However, the ion current to the target does not increase $\propto <n_e>^2$ as expected from the two point model (assuming $n_{e, up} \propto <n_e>$) \cite{STANGEBY2000} (figure \ref{fig:Overview50648}F). This difference to other, high density machines and the two-point model may be due to the fact that the ionization mean free path in TCV ($\lambda_{ioniz} \sim 5-10 \text{ cm}$) is larger compared to the width of the divertor plasma ($d_{fan} \sim$ a few cm) near the target \cite{WISCHMEIER2005EPFL}. Together with the open divertor geometry neutrals are not well-confined, which likely leads to less ionization and a slower rise in divertor density. That could reduce the amount of charge exchange and recombination, thus slowing down the detachment process. Once the detachment process starts with the drop in divertor target density and the rise in recombination signatures ($F_{rec}$ and $R_L$, figures \ref{fig:Overview50648}C and \ref{fig:Overview50648}G) the process of detachment proceeds slowly in TCV. Instead of a swift movement of the recombination and high-density regions observed at other higher density machines \cite{LIPSCHULTZ1999POP, POTZEL2014NF}, the recombining region and peak density stays near the divertor target at TCV while recombination signatures extend towards the x-point. At the highest core and divertor densities in the TCV plasmas studied so far, the drop in target density (figure \ref{fig:Overview50648}D) concurrent with the continued increase in the DSS-inferred density is consistent with the detachment region (in the sense of both low density and low temperature) moving off the divertor target slightly, less than the 5 cm corresponding to the lowest DSS chord, as discussed in section \ref{RecLP}. However, such a movement is very slow given that the Stark-derived density continues to rise throughout the remainder of the discharge. The inference of recombination rates through the DSS data analysis also provide some insight into the role of recombination in removing ions from the plasma and causing a density drop. As discussed earlier, recombination remains highest near the target throughout the discharge, with the total amount of recombination rising rapidly to levels at the end of the discharge comparable to the target ion flux. Given that the target density drops earlier in the pulse and that the total recombination rate is less than 1\% of the particle flux at the point the particle flux starts dropping, the question is whether recombination is playing an important role at this time. The two possibilities are that the ion source upstream could start dropping at the same time as the target density falls. Or, that the recombination local to the flux tube of the peak ion flux is removing significant ion flux. We do not have enough spatial information at this time to comment further on the relative important of the two effects. \section{Summary} The physics of the TCV divertor, including the detached regime, has been investigated at TCV, using a newly developed divertor spectroscopy system (DSS), together with advancements in techniques for extracting information from the Balmer spectrum. Analysis of the DSS data has been instrumental in characterizing the behaviour of detachment in TCV. We find that the detachment process develops slowly: the radiation first peaks near the divertor and then moves towards the x-point. The rise in the dominance of recombination signatures over excitation signatures follows the movement of the radiation peak, while the strongest level of density and recombination remains close to the target. Even as the plasma density above the target continues to increase the ion current to the target drops, which may imply that the detached (low pressure and density) region has moved off the target. But within the density range studied on TCV, the detachment front moves no further. The role of recombination in ion loss has been investigated. We find that there is no clear connection between the initial roll-over in the target ion current and the level of recombination. However, later in time, $R_V$ approaches the integral ion current and it may be that the recombination front moves further off the target if higher densities could be achieved. Further studies are needed. \section{Acknowledgements} This work has been carried out within the framework of the EUROfusion Consortium and has received funding from the Euratom research and training programme 2014-2018 under grant agreement No 633053. The views and opinions expressed herein do not necessarily reflect those of the European Commission. \bibliographystyle{plainnat}
5fff69caf411dbd51f9742206d76b22d147dec26
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction: classical configuration theorems} \label{intro} Projective configuration theorems are among the oldest and best known mathematical results. The next figures depict the famous theorems of Pappus, Desargues, Pascal, Brianchon, and Poncelet. \begin{figure}[hbtp] \centering \includegraphics[height=1.7in]{Pappusfig} \caption{The Pappus theorem: if $A_1,A_2,A_3$ and $B_1,B_2,B_3$ are two collinear triples of points, then $C_1,C_2,C_3$ is also a collinear triple.} \label{Pappusfig} \end{figure} \begin{figure}[hbtp] \centering \includegraphics[height=2.6in]{Desarguesfig1} \caption{The Desargues theorem: if the lines $A_1 B_1, A_2 B_2$ and $A_3 B_3$ are concurrent, then the points $C_1,C_2,C_3$ are collinear.} \label{Desarguesfig} \end{figure} \begin{figure}[hbtp] \centering \includegraphics[width=2.8in]{Pascalfig} \ \includegraphics[width=2.5in]{Brianchon1} \caption{The Pascal theorem, a generalization of the Pappus theorem: the points $A_1, \ldots, A_6$ lie on a conic, rather than the union of two lines. The Brianchon theorem is projectively dual to Pascal's.} \label{Pascalfig} \end{figure} \begin{figure}[hbtp] \centering \includegraphics[height=3in]{Ponceletfig} \caption{The Poncelet Porism, case $n=5$: if the polygonal line $A_1 A_2 A_3 A_4 A_5$, inscribed into a conic and circumscribed about a conic, closes up after five steps, then so does any other polygonal line $B_1 B_2 B_3 B_4 B_5$.} \label{Ponceletfig} \end{figure} The literature on configuration theorems is vast; the reader interested in a panoramic view of the subject is recommended \cite{Be,RG}. Configuration theorems continue to be an area of active research. To a great extent, this is due to the advent of computer as a tool of experimental research in mathematics. In particular, interactive geometry software is a convenient tool for the study of geometric configurations. The illustration in this article are made using such a software, Cinderella 2 \cite{Cin}. Another reason for the popularity of configuration theorems is that they play an important role in the emerging field of discrete differential geometry and the theory of completely integrable systems \cite{BS}. The goal of this survey is to present some recent results motivated and inspired by the classical configuration theorems; these results make the old theorems fresh again. The selection of topics reflects this author's taste; no attempt was made to present a comprehensive description of the area. In the cases when proofs are discussed, they are only outlined; the reader interested in details is referred to the original papers. We assume that the reader is familiar with the basics of projective, Euclidean, spherical, and hyperbolic geometries. One of the standard references is \cite{Be1}, and \cite{HC} is as indispensable as ever. Now let us specify what we mean by configuration theorems in this article. The point of view is dynamic, well adapted for using interactive geometry software. An initial data for a configuration theorem is a collection of labelled points $A_i$ and lines $b_j$ in the projective plane, such that, for some pairs of indices $(i,j)$, one has the incidence $A_i \in b_j$. If, in addition, a polarity is given, then one can associate the dual line to a point, and the dual point to a line. In presence of polarity, the initial data includes information that, for some pairs of indices $(k,l)$, the point $A_k$ is polar dual to the line $b_l$. One also has an ordered list of instructions consisting of two operations: draw a line through a pair of points, or intersect a pair of lines at a point. These new lines and points also receive labels. If polarity is involved, one also has the operation of taking the polar dual object, point to line, or line to point. The statement of a configuration theorem is that, among so constructed points and lines, certain incidence relations hold, that is, certain points lie on certain lines. It is assumed that the conclusion of a configuration theorem holds for almost every initial set of points and lines satisfying the initial conditions, that is, holds for a Zariski open set of such initial configurations. This is different from what is meant by a configuration of points and lines in chapter 3 of \cite{HC} or in \cite{Gr}: the focus there is on whether a combinatorial incidence is realizable by points and lines in the projective plane. \bigskip {\bf Acknowledgements}. I am grateful to R. Schwartz for numerous stimulating discussions and to P. Hooper for an explanation of his work. I was supported by NSF grant DMS-1510055. This article was written during my stay at ICERM; it is a pleasure to thank the Institute for its inspiring and friendly atmosphere. \section{Iterated Pappus theorem and the modular group} \label{itPapp} The Pappus theorem can be viewed as a construction in ${\mathbb {RP}}^2$ that inputs two ordered triples of collinear points $A_1,A_2,A_3$ and $B_1,B_2,B_3$, and outputs a new collinear triple of points $C_1,C_2,C_3$, see Figure \ref{Pappusfig}. One is tempted to iterate: say, take $A_1,A_2,A_3$ and $C_1,C_2,C_3$ as an input. Alas, this takes one back to the triple $B_1,B_2,B_3$. To remedy the situation, swap points $C_1$ and $C_3$. Then the input $A_1,A_2,A_3$ and $C_1,C_2,C_3$ yields a new collinear triple of points, and so does the input $C_1,C_2,C_3$ and $B_1,B_2,B_3$. And one can continue in the same way indefinitely, see Figure \ref{Pappusiter}. The study of these iterations was the topic of R. Schwartz's paper \cite{Sch93}. \begin{figure}[hbtp] \centering \includegraphics[height=2in]{Pappusbw} \caption{Iterations of the Pappus construction produced by R. Schwartz's applet \cite{App}.} \label{Pappusiter} \end{figure} Return to Figure \ref{Pappusfig}. The input of the Pappus construction is the {\it marked box} $(A_1,A_3, B_3, B_1; A_2, B_2)$, a quadrilateral $A_1 A_3 B_3 B_1$ with the top distinguished point $A_2$ and the bottom distinguished point $B_2$. The marked box is assumed to satisfy the {\it convexity condition}: the points $A_1$ and $A_3$ are separated by the points $A_2$ and $O$ on the projective line $a$, and likewise for the pairs of points $B_1, B_3$ and $B_2, O$ on the line $b$. Marked boxes that differ by the involution $$ (A_1, A_3, B_3, B_1; A_2, B_2) \leftrightarrow (A_3, A_1, B_1, B_3; A_2, B_2) $$ are considered to be the same. A convex set in ${\mathbb {RP}}^2$ is a set that is disjoint from some line and that is convex in the complement to this line, the affine plane. Two points in ${\mathbb {RP}}^2$ can be connected by a segment in two ways. The four points $A_1, A_3, B_3, B_1$, in this cyclic order, define 16 closed polygonal lines, but only one of them is the boundary of a convex quadrilateral, called the interior of the convex marked box. Recall that the points of the dual projective plane are the lines of the initial plane. Let $\Theta=(A_1,A_3, B_3, B_1; A_2, B_2)$ be a convex marked box in ${\mathbb {RP}}^2$. Its dual, $\Theta^*$, is a marked box in the dual projective plane whose points are the lines $$ (A_2B_1,A_2B_3,A_1B_2,A_3B_2; a,b). $$ The dual marked box is also convex. The moduli space of projective equivalence classes of marked boxes in 2-dimensional. One can send the points $A_1, A_3, B_3, B_1$ to the vertices of a unit square; then the projective equivalence class of a convex marked box is determined by the positions of the points $A_2$ and $B_2$ on the horizontal sides of the square. Namely, let $x=|A_1 A_2|, y=|B_1 B_2|$. Then the equivalence class \begin{equation} \label{invol} (x,y) \sim (1-x,1-y), \end{equation} where $0<x,y<1$, determines the the projective equivalence class of a convex marked box. We denote this equivalence class by $[x,y]$. The Pappus construction defines two operations on convex marked boxes, see Figure \ref{complement}: \begin{equation*} \begin{split} \tau_1: (A_1,A_3, B_3, B_1; A_2, B_2) \mapsto (A_1,A_3, C_3, C_1; A_2, C_2),\\ \tau_2: (A_1,A_3, B_3, B_1; A_2, B_2) \mapsto (C_1,C_3, B_3, B_1; C_2, B_2). \end{split} \end{equation*} \begin{figure}[hbtp] \centering \includegraphics[height=1.8in]{complement1} \caption{The interior of the convex marked box $i(\Theta)$ is bounded by the segments $A_1 A_3, A_3 B_1, B_1 B_3$ and $B_3 A_1$. Two of these segments cross the line at infinity.} \label{complement} \end{figure} Add to it a third operation $$ i: (A_1,A_3, B_3, B_1; A_2, B_2) \mapsto (B_1,B_3,A_1,A_3; B_2,A_2), $$ also shown in Figure \ref{complement}. The three operations form a semigroup $G$. The operations satisfy the following identities, proved by inspection. \begin{lemma} \label{groupG} One has: $$ i^2=1,\ \tau_1i\tau_2=\tau_2 i \tau_1 = i,\ \tau_1 i \tau_1 = \tau_2,\ \tau_2 i \tau_2 = \tau_1. $$ \end{lemma} As a consequence, $G$ is a group; for example, $\tau_1^{-1}= i \tau_2 i$. Recall that the modular group $M$ is the group of fractional-linear transformations with integral coefficients and determinant one, that is, the group $PSL(2,{\mathbb Z})$. Since $PGL(2,{\mathbb R})$ is the group of orientation preserving isometries of the hyperbolic plane, the modular group $M$ is realized as a group of isometries of $H^2$. \begin{figure}[hbtp] \centering \includegraphics[height=2.3in]{Farey} \caption{A tiling of the hyperbolic plane, in the Poincar\'e disk model, by ideal triangles.} \label{Farey} \end{figure} Consider the tiling of $H^2$ by ideal triangles obtained from one such triangle by consecutive reflections in the sides, see Figure \ref{Farey} for the beginning of this construction. The modular group is generated by two symmetries of the tiling: the order three rotation about point $A$ and the order two rotation (central symmetry) about point $B$. Algebraically, $M$ is a free product of ${\mathbb Z}_3$ and ${\mathbb Z}_2$. Return to the group $G$. It is generated by the elements $\alpha=i\tau_1$ and $\beta = i$. Lemma \ref{groupG} implies that $\alpha^3=\beta^2=1$. One can prove that there are no other relations, and hence $G={\mathbb Z}_3 * {\mathbb Z}_2$ is identified with the modular group. Given a convex marked box $\Theta$, consider its orbit $\Omega = G(\Theta)$ under the action of the group $G$. The orbit can be described by its oriented incidence graph $\Gamma$. The edges of $\Gamma$ correspond to the marked boxes of $\Omega$, oriented from top to bottom, and the vertices correspond to the tops and the bottoms of the boxes. One can embed $\Gamma$ in the hyperbolic plane as in Figure \ref{Farey} (the orientations of the edges are not shown). The group $G$ acts by permutations of the edges of $\Gamma$. The operation $i$ reverses the orientations of the edges. The operation $\tau_1$ rotates each edge counterclockwise one `click' about its tail, and $\tau_2$ rotates the edges one `click' clockwise about their heads. (This is a different action from the one generated by rotations about points $A$ and $B$ in Figure \ref{Farey}). Denote by $G'$ the index two subgroup of $G$ that consists of the transformations that preserve the orientations of the edges. The orbit $\Omega$ of a convex marked box $\Theta$ has a large group of projective symmetries, namely, an index two subgroup $M'$ of the modular group $M$. This is one of the main results of \cite{Sch93}. Specifically, one has \begin{proposition} \label{projsym} Given a convex marked box $\Theta$, there is an order three projective transformation with the cycle $$ i(\Theta) \mapsto \tau_1(\Theta) \mapsto \tau_2(\Theta). $$ In addition, there exists a polarity that identifies $i(\Theta)$ with the dual box $\Theta^*$. \end{proposition} \paragraph{Proof.} For the proof of the first statement, one can realize the box $\Theta$ in such a way that the three-fold rotational symmetry is manifestly present, see Figure \ref{threefold}. Namely, \begin{equation*} \begin{split} &\Theta = (B_3,B_1,A_1,A_3;B_2,A_2),\ i(\Theta) = (A_1,A_3,B_1,B_3;A_2,B_2),\\ &\tau_1(\Theta) = (B_1,B_3,C_1,C_3;B_2,C_2),\ \tau_2(\Theta) = (C_1,C_3,A_1,A_3;C_2,A_2). \end{split} \end{equation*} \begin{figure}[hbtp] \centering \includegraphics[height=2.5in]{threefold} \caption{A symmetric realization of the marked boxes $i(\Theta), \tau_1(\Theta), \tau_2(\Theta)$.} \label{threefold} \end{figure} In terms of the marked box coordinates $(x,y)$, described in (\ref{invol}), the three operations, $i, \tau_1$, and $\tau_2$, act in the same way: $[x,y] \mapsto [1-y,x].$ \begin{figure}[hbtp] \centering \includegraphics[height=2in]{inversion} \caption{Projective equivalence of $i(\Theta)$ and $\Theta^*$.} \label{inversion} \end{figure} For the second statement, consider another realization depicted in Figure \ref{inversion}. The marked points $A_2$ and $B_2$ are at infinity, and $|OA_1| |OA_3| = |OB_1| |OB_3| = 1$. Then the polarity with respect to the unit circle centered at point $O$ acts as follows: $$ A_1 \mapsto A_3B_2,\ A_3 \mapsto A_1 B_2,\ B_1 \mapsto B_3 A_2,\ B_3 \mapsto B_1 A_2, $$ providing the desired projective equivalence. $\Box$\bigskip If one identifies the projective plane with its dual by a polarity, then the above discussion describes a faithful representation of the modular group $M$ as the group of projective symmetries of the $G$-orbit $\Omega$ of a convex marked box. A marked box $\Theta$ determines a natural map $f$ of the set of vertices of the graph $\Gamma$ to the set of the marked points of the orbit $\Omega$. The map $f$ conjugates the actions of the group $G'$ on the graph $\Gamma$ and the group $M'$ of projective symmetries of the orbit $\Omega$. The set of vertices of $\Gamma$ is dense on the circle at infinity of the hyperbolic plane $S^1$, see Figure \ref{Farey}. Using the nested properties of the interiors of the boxes in $\Omega$ and estimates on their sizes (in the elliptic plane metric), Schwartz proves the following theorem. \begin{theorem} \label{curve} The map $f$ extends to a homeomorphism of $S^1$ to its image. \end{theorem} The image $\Lambda = f(S^1)$ is called the {\it Pappus curve}; see Figure \ref{Pappusiter} that provides an approximation of this curve. The above discussion shows that the Pappus curve is projectively self-similar. In the exceptional case of $x=y=1/2$, the curve $\Lambda$ is a straight line. Otherwise, it is not an algebraic curve, see \cite{Ha}. The tops and bottoms of the marked boxes form a countable collection of lines that also extends to a continuous family, a curve $L$ in the dual projective plane. Define a transverse line field along $\Lambda$ as a continuous family of lines such that each line from the family intersects the curve at exactly one point and every point of $\Lambda$ is contained in some line. \begin{theorem} \label{transverse} If the Pappus curve $\Lambda$ is not a straight line, then $L$ is a unique transverse line field along $\Lambda$. \end{theorem} This theorem, the fact that the Pappus curve is projectively self-similar, and computer experiments suggest that $\Lambda$ is a true fractal (unless it is a straight line). The thesis \cite{Ki} contains some preliminary numerical results on the box dimension of the Pappus curve and its dependence on the coordinates $[x,y]$ of the initial convex marked box. According to these experiments, the maximal possible box dimension of $\Lambda$ is about 1.25. Finding the fractal dimensions of the Pappus curves as a function of $[x,y]$ or, at least, proving that this dimension is greater than one in all non-exceptional cases $[x,y]\neq [1/2,1/2]$, is an outstanding open problem. \section{Steiner theorem and the twisted cubic} \label{itSteiner} This section is based on another recent ramification of the Pappus theorem, the work of J. F. Rigby \cite{Ri} and P. Hooper \cite{Ho}. \begin{figure}[hbtp] \centering \includegraphics[height=2in]{dualPap} \caption{Dual Pappus theorem.} \label{dualPap} \end{figure} Let us start with the dual Pappus theorem, see Figure \ref{dualPap} where the objects dual to the ones in Figure \ref{Pappusfig} are denoted by the same letters, with the upper and lower cases swapped (the Pappus theorem is equivalent to its dual). As an aside, let us mention that the dual Pappus theorem has an interpretation in the theory of webs: the 3-web, made of three pencils of lines, is flat, see \cite{FT}, lecture 18. Now consider Pascal's theorem, Figure \ref{Pascalfig}. The six permutations of the points on the conic yield 60 Pascal lines. These lines and their intersection points, connected by further lines, form a intricate configuration of 95 points and 95 lines, the {\it hexagrammum mysticum}. There is a number of theorems describing this configuration, due to Steiner, Pl\"ucker, Kirkman, Cayley, and Salmon. See \cite{CR1,CR2} for a contemporary account of this subject. The Pappus theorem is a particular case of Pascal's theorem, and in this case, the number of lines that result from permuting the initial points (say, points $B_1,B_2,B_3$ in Figure \ref{Pappusfig}) reduces to six, shown in Figure \ref{return2}. Let us introduce notations. Consider Figure \ref{Pappusfig} and denote the triples of points: $$ {\cal A} = (A_1,A_2,A_3),\ {\cal B} = (B_1,B_2,B_3). $$ The Pappus theorem produces a new triple, ${\cal C} = (C_1,C_2,C_3)$. The lines containing these triples are denoted by $a,b,c$, respectively. We write: $c = \ell ({\cal A},{\cal B}).$ We use a similar notation for the dual Pappus theorem: if $$ \alpha=(a_1,a_2,a_3),\ \beta=(b_1,b_2,b_3) $$ are two triples of concurrent lines, then $\ell^* (\alpha,\beta)$ is the point of intersection of the triple of lines $(c_1,c_2,c_3)$, see Figure \ref{dualPap}. The permutation group $S_3$ acts on triples by the formula $$ s({\cal B}) = (B_{s^{-1}(1)},B_{s^{-1}(2)},B_{s^{-1}(3)}). $$ Let $\sigma\in S_3$ be a cyclic permutation, and $\tau\in S_3$ be a transposition of two elements. The following result, depicted in Figure \ref{return2}, is due to J. Steiner. \begin{theorem} \label{St1} The three Pappus lines $\ell ({\cal A},s({\cal B}))$ where $s\in S_3$ is an even permutation, are concurrent, and so are the three lines corresponding to the odd permutations. \end{theorem} \begin{figure}[hbtp] \centering \includegraphics[height=3.2in]{return2} \caption{Two Steiner points, corresponding to even and odd permutations, are labelled. One of the points $\ell^*(\varphi, s(\psi))$ is shown.} \label{return2} \end{figure} Thus we obtain two triples of concurrent lines; denote them by $$ \varphi = (\ell ({\cal A},{\cal B}), \ell ({\cal A},\sigma({\cal B}), \ell ({\cal A},\sigma^2({\cal B})), \psi = (\ell ({\cal A},\tau({\cal B})), \ell ({\cal A},\tau\sigma({\cal B}), \ell ({\cal A},\tau\sigma^2({\cal B})). $$ Apply the dual Pappus theorem to the permutations of these triples of lines. By the dual Steiner theorem, the six points $\ell^*(\varphi, s(\psi)),\ s\in S_3$, are collinear in threes. More is true. The next two theorems are due to Rigby \cite{Ri}. \begin{theorem} \label{Ri1} The points $\ell^*(\varphi, s(\psi))$ lie on line $a$ when $s$ is an even permutation, and on line $b$ when $s$ is odd. \end{theorem} Let ${\cal B}'$ be another collinear triple of points such that the line $b'$ still passes through point $O = a\cap b$. Applying the above constructions to ${\cal A}, {\cal B}'$, we obtain new triples of lines $\varphi', \psi'$, and a new triple of points $\ell^*(\varphi', s(\psi'))$ on line $a$ where $s$ is an even permutation. \begin{theorem} \label{Ri2} The new triple of points coincides with the old one: for even permutations $s$, one has $\ell^*(\varphi', s(\psi'))= \ell^*(\varphi, s(\psi))$. \end{theorem} Theorems \ref{Ri1} and \ref{Ri2} are stated by Rigby without proof; to quote, \begin{quote} The theorems in this section have been verified in a long and tedious manner using coordinates. There seems little point in publishing the calculations; it is to be hoped that shorter and more elegant proofs will be found. \end{quote} Conceptual proofs are given in \cite{Ho}; the reader is referred to this paper and is encouraged to find an alternative approach to these results. The above theorems make it possible to define the {\it Steiner map} $$ S_O: (A_1,A_2,A_3) \mapsto (\ell^*(\varphi, \psi), \ell^*(\varphi, \sigma^2(\psi)), \ell^*(\varphi, \sigma(\psi))). $$ This map depends on the point $O$, but not on the choice of the triple ${\cal B}$. The Steiner map commutes with permutations of the points involved, and hence it induces a map of the space of unordered triples of points of the projective line. Abusing notation, we denote this induced map by the same symbol. Hooper \cite{Ho} gives a complete description of the Steiner map. Assume that the ground field is complex. The space of unordered triples of points of ${\mathbb {CP}}^1$, that is, the symmetric cube $S^3({\mathbb {CP}}^1)$, is identified with ${\mathbb {CP}}^3$. This is a particular case of the Fundamental Theorem of Algebra, one of whose formulations is that $S^n({\mathbb {CP}}^1)={\mathbb {CP}}^n$ (given by projectivizing the Vieta formulas that relate the coefficients of a polynomial to its roots). Thus $S_O$ is a self-map of ${\mathbb {CP}}^3$. The set of cubic polynomials with a triple root corresponds to a curve $\Gamma \subset {\mathbb {CP}}^3$, the twisted cubic (the moment curve). The secant variety of the twisted cubic, that is, the union of its tangent and secant lines, covers ${\mathbb {CP}}^3$, and the lines are pairwise disjoint, except at the points of $\Gamma$. The set of cubic polynomials with a zero root corresponds to a plane in ${\mathbb {CP}}^3$. Denote this plane by $\Pi$. The Steiner map $S_O: {\mathbb {CP}}^3 \to {\mathbb {CP}}^3$ is described in the next theorem. \begin{theorem} \label{Stmap} (i) The map $S_O$ preserves the secants of the twisted cubic $\Gamma$ that do not pass through the origin (the image of the cubic polynomial $z^3$). \\ (ii) One can choose projective coordinates on these secant lines so that the map is given by the formula $x \mapsto x^2$. \\ (iii) The choice of coordinates is as follows: the two points of intersection of the secant line with $\Gamma$ have coordinates $0$ and $\infty$, and the intersection point of the secant with the plane $\Pi$ has coordinate $-1$. \end{theorem} In homogeneous coordinates of ${\mathbb {CP}}^3$, the map $S_O$ is polynomial of degree 6; see \cite{Ho} for an explicit formula for a particular choice $O=(0:1)$. In the real case, the secant lines are identified with the circle ${\mathbb R}/{\mathbb Z}$ , and the Steiner map becomes the doubling map $t \mapsto 2t$ mod 1, a well known measure preserving ergodic transformation. \section{Pentagram-like maps on inscribed polygons} \label{pentalike} This section, based on \cite{ST1}, concerns eight configuration theorems of projective geometry that were discovered in the study of the pentagram map. The pentagram map, whose study was put forward by R. Schwartz \cite{Sch92}, is a transformation of the moduli space of projective equivalence classes of polygons in the projective plane depicted in Figure \ref{penta}. The pentagram map has become a popular object of study: it is a discrete completely integrable system, closely related with the theory of cluster algebras. See \cite{GSTV,Gl,GP,OST1,OST2,So} for a sampler of the current literature on this subject. \begin{figure}[hbtp] \centering \includegraphics[height=1.7in]{penta} \caption{The pentagram map takes an $n$-gon $P$ to the polygon made by the intersection points of the short (skip one) diagonals of $P$. Here $n=7$.} \label{penta} \end{figure} To formulate the results, let us introduce some notations. By a polygon in the projective plane we mean a cyclically ordered collection of its vertices (that also determines the cyclically ordered collection of lines, the sides of the polygon). Let ${\cal C}_n$ and ${\cal C}_n^*$ be the spaces $n$-gons in the projective plane ${\mathbb {RP}}^2$ and its dual $({\mathbb {RP}}^2)^*$. Define the $k$-diagonal map $T_k: {\cal C}_n \to {\cal C}_n^*$: for $P=\{p_1,...,p_n\}$, $$T_k(P)=\{(p_1p_{k+1}),(p_2p_{k+2}),\ldots, (p_np_{k+n})\}.$$ Each map $T_k$ is an involution; the map $T_1$ is the projective duality that sends a polygon to the cyclically ordered collection of its sides. Extend the notation to muti-indices: $T_{ab}=T_a \circ T_b, T_{abc}=T_a \circ T_b\circ T_c$, etc. For example, the pentagram map is $T_{12}$. If $P$ is a polygon in ${\mathbb {RP}}^2$ and $Q$ a polygon in $({\mathbb {RP}}^2)^*$, and there exists a projective transformation ${\mathbb {RP}}^2 \to ({\mathbb {RP}}^2)^*$ that takes $P$ to $Q$, we write: $P \sim Q$. Now we are ready to formulate our results; they concern polygons inscribed into a conic or circumscribed about a conic. \begin{theorem} \label{thm12} (i) If $P$ is an inscribed $6$-gon, then $P \sim T_2(P)$.\\ (ii) If $P$ is an inscribed $7$-gon, then $P \sim T_{212}(P)$.\\ (iii) If $P$ is an inscribed $8$-gon, then $P \sim T_{21212}(P)$. \end{theorem} Surprisingly, this sequence doesn't continue! Theorem \ref{thm12} (iii) is depicted in Figure \ref{octagon}. See also Schwartz's applet \cite{App1} for illustration of this and other results of this section. \begin{figure}[hbtp] \centering \includegraphics[height=2in]{octagon} \caption{The third iteration of the pentagram map on an inscribed octagon yields a projectively dual octagon.} \label{octagon} \end{figure} \begin{theorem} \label{thm313} If $P$ is a circumscribed $9$-gon, then $P \sim T_{313}(P)$. \end{theorem} See Figure \ref{nonagon}. \begin{figure}[hbtp] \centering \includegraphics[height=2.3in]{nonagon} \caption{Theorem \ref{thm313}.} \label{nonagon} \end{figure} \begin{theorem} \label{thm34} If $P$ is an inscribed $12$-gon, then $P \sim T_{3434343}(P)$. \end{theorem} The next results have a somewhat different flavor: one does not claim anymore that the final polygon is projectively related to the initial one. \begin{theorem} \label{thm13} (i) If $P$ is an inscribed $8$-gon, then $T_{3}(P)$ is circumscribed.\\ (ii) If $P$ is an inscribed $10$-gon, then $T_{313}(P)$ is circumscribed.\\ (iii) If $P$ is an inscribed $12$-gon, then $T_{31313}(P)$ is circumscribed. \end{theorem} Again, in spite of one's expectation, this sequence does not continue. Theorem \ref{thm13} (iii) is illustrated in Figure \ref{dodecagon}. \begin{figure}[hbtp] \centering \includegraphics[height=2.5in]{dodecagon} \caption{Theorem \ref{thm13} (iii).} \label{dodecagon} \end{figure} Now about the discovery of these results and their proofs. Theorems \ref{thm12} (i) and (ii) were discovered in our study of the pentagram map. Then V. Zakharevich, a participant of the 2009 Penn State REU (Research Experience for Undergraduates) program, discovered Theorem \ref{thm313}. Inspired by this discovery, we did an extensive computer search for this kind of configuration theorems; the results are the above eight theorems. We think that the list above is exhaustive, but this remains a conjecture. Note that one may cyclically relabel the vertices of a polygon to deduce seemingly new theorems. Let us illustrate this by an example. Rephrase the statement of Theorem \ref{thm13} (iii) as follows: {\it If $P$ is an inscribed dodecagon then $T_{131313}(P)$ is also inscribed}. Now relabel the vertices by $\sigma(i)=5i$ mod 12. The map $T_3$ is conjugated by $\sigma$ as follows: $$ i\mapsto 5i\mapsto 5i+3\mapsto 5(5i+3)=i+3\ \ {\rm mod}\ 12, $$ that is, it is the map is $T_3$ again, and the map $T_1$ becomes $$ i\mapsto 5i\mapsto 5i+1\mapsto 5(5i+1)=i+5\ \ {\rm mod}\ 12, $$ that is, the map is $T_5$. One arrives at the statement: {\it If $P$ is an inscribed dodecagon then $T_{535353}(P)$ is also inscribed}. We proved all of the above theorems, except Theorem \ref{thm13} (iii), by uninspiring computer calculations (the symbolic manipulation required for a proof of Theorem \ref{thm13} (iii) was beyond what we could manage in Mathematica). Of course, one wishes for elegant geometric proofs. Stephen Wang found proofs of Theorems \ref{thm13} (i) and (ii) which are presented below, and Maria Nastasescu, a 2010 Penn State REU participant, found algebraic geometry proofs of the same two theorems. Fedor Nilov proved Theorem \ref{thm13} (iii) using a planar projection of hyperboloid of one sheet. Unfortunately, none of these proofs were published. Here is Wang's proof of Theorem \ref{thm13} (i). Consider Figure \ref{Wang2}. We need to prove that the points $B_1,\ldots, B_8$ lie on a conic. \begin{figure}[hbtp] \centering \includegraphics[height=2.2in]{Wang2} \caption{Proof of Theorem \ref{thm13} (i).} \label{Wang2} \end{figure} The hexagon $A_6 A_1 A_4 A_7 A_2 A_5$ is inscribed so, by Pascal's theorem, the points $B_1, B_6$ and $C$ are collinear. That is, the intersection points of the opposite sides of the hexagon $B_1 B_2 B_3 B_4 B_5 B_6$ are collinear. By the converse Pascal theorem, this hexagon is inscribed. A similar argument shows that the hexagon $B_2 B_3 B_4 B_5 B_6 B_7$ is inscribed. But the two hexagons share five vertices, hence they are inscribed in the same conic. Likewise, $B_8$ lies on this conic as well. Now, to the proof of Theorem \ref{thm13} (ii), see Figure \ref{Wang3}. \begin{figure}[hbtp] \centering \includegraphics[height=3.4in]{Wang3} \caption{Proof of Theorem \ref{thm13} (ii).} \label{Wang3} \end{figure} Consider the inscribed hexagon $A_3A_6A_9A_{10}A_7A_4$. By Pascal's theorem, the points $$ (A_3A_6)\cap (A_{10}A_7),\ (A_6A_9)\cap (A_7A_4),\ (A_9A_{10})\cap (A_4A_3) $$ are collinear. Hence the triangles $A_3B_3A_4$ and $A_{10}A_9B_6$ are perspective. By the Desargues theorem, the points $ A_4, A_9, (B_3B_6)\cap (A_3A_{10})$ are collinear. It follows that the triangles $B_9B_{10}Y$ and $B_3B_6X$ are perspective. By the Desargues theorem, the points $X,Y, (B_6B_9)\cap (B_3B_{10})$ are collinear. The same argument, with all indices shifted by five, implies that the points $X,Y, (B_1B_4)\cap (B_8B_5)$ are collinear as well. Hence the points $$ (B_6B_9)\cap (B_3B_{10}), (B_1B_4)\cap (B_8B_5),\ {\rm and}\ X $$ are collinear. Reinterpret this as the collinearity of $$ (C_{10}B_{10})\cap (C_5B_9),\ (C_{10}B_4)\cap (C_5B_5),\ (B_3B_4)\cap (B_5B_6). $$ It follows that the triangles $B_3B_4C_{10}$ and $B_5B_6C_5$ are perspective. By the Desargues theorem, the points $B_4,B_5$ and $(C_{10}C_5)\cap (C_2C_3)$ are collinear. That is, the points $$ (C_{10}C_5)\cap (C_2C_3), (C_{10}C_1)\cap (C_3C_4),\ (C_1C_2)\cap (C_4C_5) $$ are collinear, and by the converse Pascal theorem, the points $$C_{10},C_1,C_2,C_3,C_4,C_5$$ lie on a conic. The rest is the same as in the previous proof. $\Box$\bigskip One can add to Theorems \ref{thm12} -- \ref{thm13} a statement about pentagons. Consider the following facts:\\ (i) every pentagon is inscribed in a conic and circumscribed about a conic;\\ (ii) every pentagon is projectively equivalent to its dual;\\ (iii) the pentagram map sends every pentagon to a projectively-equivalent one.\\ Therefore one may add the following theorem to our list: {\it for a pentagon $P$, one has $P\sim T_2(P)$}. The following result of R. Schwartz \cite{Sch01,Sch08,Gl} also has a similar flavor. \begin{theorem} \label{degen} If $P$ is a $4n$-gon inscribed into a degenerate conic (that is, a pair of lines) then $$(T_1T_2T_1T_2\dots T_1)(P)\qquad (4n-3\ {\rm terms)}$$ is also inscribed into a degenerate conic. \end{theorem} One wonders whether there is a unifying theme here. A possibly relevant reference is \cite{GP}. \section{Poncelet grid, string construction, and billiards in ellipses} \label{grid} A Poncelet polygon is a polygon that is inscribed into an ellipse $\Gamma$ and circumscribed about an ellipse $\gamma$. Let $L_1,\ldots, L_n$ be the lines containing the sides of a Poncelet $n$-gon, enumerated in such a way that their tangency points with $\gamma$ are in the cyclic order. The {\it Poncelet grid} is the collection of $n(n+1)/2$ points $L_i \cap L_j$, where $L_i \cap L_i$ is the tangency point of the line $L_i$ with $\gamma$. To simplify the exposition, assume that $n$ is odd (for even $n$, the formulations are slightly different). One can partition the Poncelet grid in two ways. Define the sets $$ P_k = \cup_{i-j=k} \ell_i \cap \ell_j,\quad Q_k = \cup_{i+j=k} \ell_i \cap \ell_j, $$ where the indices are understood mod $n$. There are $(n + 1)/2$ sets $P_k$ , each containing $n$ points, and $n$ sets $Q_k$ , each containing $(n + 1)/2$ points. The sets $P_k$ are called concentric, and the sets $Q_k$ are called radial, see Figure \ref{Grid}. \begin{figure}[hbtp] \centering \includegraphics[height=2.2in]{Grid} \caption{Poncelet grid, $n=9$: shown are the concentric sets $P_0, P_2, P_3$, and $P_4$ that lie on four ellipses.} \label{Grid} \end{figure} The following theorem is proved in \cite{Sch07}. \begin{theorem} \label{Pgrid} (i) The concentric sets lie on nested ellipses, and the radial sets lie on disjoint hyperbolas. \\ (ii) The complexified versions of these conics have four common tangent lines. \\ (iii) All the concentric sets are projectively equivalent to each other, and so are all the radial sets. \end{theorem} In this section, following \cite{LT}, we prove this projective theorem using Euclidean geometry, namely, the billiard properties of conics. As a by-product of this approach, we establish the Poncelet theorem and prove the theorem of Reye and Chasles on inscribed circles. See \cite{DR,Fl} for general information about the Poncelet theorem, and \cite{KT,Tab95,Tab05} for the theory of billiards. The reduction to billiards goes as follows. Any pair of nested ellipses $\gamma \subset \Gamma$ can be taken to a pair of confocal ellipses by a suitable projective transformation. This transformation takes a Poncelet polygon to a periodic billiard trajectory in $\Gamma$. The billiard inside a convex domain with smooth boundary is a transformation of the space of oriented lines (rays of light) that intersect the domain: an incoming billiard trajectory hits the boundary (a mirror) and optically reflects so that the angle of incidence equals the angle of reflection. The space of oriented lines has an area form, preserved by the optical reflections (independently of the shape of the mirror). Choose an origin, and introduce coordinates $(\alpha,p)$ on the space of rays: $\alpha$ is the direction of the ray, and $p$ is its signed distance to the origin. Then the invariant area form is $\omega=d\alpha\wedge d p$. A {\it caustic} of a billiard is a curve $\gamma$ with the property that if a segment of a billiard trajectory is tangent to $\gamma$, then so is each reflected segment. There is no general method of describing caustics of a given billiard curve,\footnote{The existence of caustics for strictly convex and sufficiently smooth billiard curves is proved in the framework of the KAM theory.} but the converse problem, to reconstruct a billiard table $\Gamma$ from its caustic $\gamma$, has a simple solution given by the following {\it string construction}: wrap a non-stretchable closed string around $\gamma$, pull it tight, and move the farthest point around $\gamma$; the trajectory of this point is the billiard curve $\Gamma$. This construction yields a 1-parameter family of billiard tables sharing the caustic $\gamma$: the parameter is the length of the string. The reason is as follows, see Figure \ref{string}. For a point $X$ outside of the oval $\gamma$, consider two functions: $$ f(X)=|XA|+\stackrel{\smile}{|AO|},\ g(X)=|XB|+\stackrel{\smile}{|BO|}. $$ \begin{figure}[hbtp] \centering \includegraphics[height=1.7in]{string.pdf} \caption{String construction} \label{string} \end{figure} The gradients of these functions are the unit vectors along the lines $AX$ and $BX$, respectively. It follows that these two lines make equal angles with the level curves of the functions $f+g$ and $f-g$, and that these level curves are orthogonal to each other. In particular, the level curves of $f+g$ are the billiard tables for which $\gamma$ is a caustic. Note that the function $f+g$ does not depend on the choice of the auxiliary point $O$, whereas the function $f-g$ is defined up to an additive constant, so its level curves are well defined. Here is a summary of the billiard properties of conics. The interior of an ellipse is foliated by confocal ellipses: these are the caustics of the billiard inside an ellipse. Thus one has Graves's theorem: {\it wrapping a closed non-stretchable string around an ellipse yields a confocal ellipse}. The space of rays $A$ that intersect an ellipse is topologically a cylinder, and the billiard system inside the ellipse is an area preserving transformation $T: A\to A$. The cylinder is foliated by the invariant curves of the map $T$ consisting of the rays tangent to confocal conics, see Figure \ref{portrait}. \begin{figure}[hbtp] \centering \includegraphics[height=1.7in]{portrait.png} \caption{Phase portrait of the billiard map in an ellipse} \label{portrait} \end{figure} The curves that go around the cylinder correspond to the rays that are tangent to confocal ellipses, and the curves that form `the eyes' to the rays that are tangent to confocal hyperbolas. A singular curve consists of the rays through the foci, and the two dots to the 2-periodic back and forth orbit along the minor axis of the ellipse. One can choose a cyclic parameter, say, $x$ modulo 1, on each invariant circle, such that the map $T$ becomes a shift $x \mapsto x+c$, where the constant $c$ depends on the invariant curve. Here is this construction (a particular case of the Arnold-Liouville theorem in the theory of integrable systems). Choose a function $H$ whose level curves are the invariant curves that foliate $A$, and consider its Hamiltonian vector field sgrad $H$ with respect to the area form $\omega$. This vector field is tangent to the invariant curves, and the desired coordinate $x$ on these curves is the one in which sgrad $H$ is a constant vector field $d/dx$. Changing $H$ scales the coordinate $x$ on each invariant curve and, normalizing the `length' of the invariant curves to 1, fixes $x$ uniquely up to an additive constant. In other words, the 1-form $dx$ is well defined on each invariant curve. The billiard map $T$ preserves the area form and the invariant curves, therefore its restriction to each curve preserves the measure $dx$, hence, is a shift $x \mapsto x+c$. An immediate consequence is the Poncelet Porism: if a billiard trajectory in an ellipse closes up after $n$ reflections, then $nc \equiv 0$ mod 1, and hence all trajectories with the same caustic close up after $n$ reflections. Note that the invariant measure $dx$ on the invariant curves does not depend on the choice of the billiard ellipse from a confocal family: the confocal ellipses share their caustics. This implies that the billiard transformations with respect to two confocal ellipses commute: restricted to a common caustic, both are shifts in the same coordinate system. This statement can be considered as a configuration theorem; see Figure \ref{commute}. \begin{figure}[hbtp] \centering \includegraphics[height=1.8in]{commute}\quad \includegraphics[height=1.8in]{commute1} \caption{Left: the billiard reflections of the rays from a focus in two confocal ellipses commute. Right: the general case.} \label{commute} \end{figure} To summarize, an ellipse is a billiard caustic for the confocal family of ellipses. It carries a coordinate $x$, defined up to an additive constant, in which the billiard reflection in confocal ellipses is given by $x \mapsto x+c$. We refer to the coordinate $x$ as the canonical coordinate. Consider an ellipse $\gamma$, and let $x$ be the canonical coordinate on it. Define coordinates in the exterior of the ellipse: the coordinates of a point $X$ outside of $\gamma$ are the coordinates $x_1$ and $x_2$ of the tangency points of the tangent lines from $X$ to $\gamma$. Let us call $(x_1,x_2)$ the string coordinates of point $X$. The confocal ellipses are given by the equations $x_2-x_1=$ const. \begin{lemma} \label{hyp} The confocal hyperbolas have the equations $x_2+x_1=$ const. \end{lemma} \paragraph{Proof.} Consider Figure \ref{circlequad}. Let the canonical coordinates of the tangency points on the inner ellipse, from left to right, be $x_1,x_2,x_3,x_4$, so that the string coordinates are as follows: $$ A(x_1,x_3),\ B(x_2,x_4),\ C(x_2,x_3),\ D(x_1,x_4). $$ Since $A$ and $B$ are on a confocal ellipse, $x_4-x_2=x_3-x_1$, and hence $x_2+x_3=x_1+x_4$. By the billiard property, the arc of an ellipse $AB$ bisects the angles $CAD$ and $CBD$. Therefore, in the limit $A\to B$, the infinitesimal quadrilateral $ABCD$ becomes a kite: the diagonal $AB$ is its axis of symmetry. Hence $AB \perp CD$, and the locus of points given by the equation $x_1+x_4=$ const and containing points $C$ and $D$ is orthogonal to the ellipse through points $A$ and $B$. Therefore this locus is a confocal hyperbola. $\Box$\bigskip \begin{figure}[hbtp] \centering \includegraphics[height=2in]{circlequad} \caption{Two pairs of tangents from an ellipse to a confocal ellipse.} \label{circlequad} \end{figure} The next result is due to Reye and Chasles. \begin{theorem} \label{inscribed} Let $A$ and $B$ be two points on an ellipse. Consider the quadrilateral $ABCD$, made by the pairs of tangent lines from $A$ and $B$ to a confocal ellipse. Then its other vertices, $C$ and $D$, lie on a confocal hyperbola, and the quadrilateral is circumscribed about a circle, see Figure \ref{circlequad}. \end{theorem} \paragraph{Proof.} In the notation of the proof of the preceding lemma, $x_2+x_3=x_1+x_4$, hence points $C$ and $D$ lie on a confocal hyperbola. Furthermore, in terms of the string construction, $$ f(A)+g(A)=f(B)+g(B),\ \ f(C)-g(C)=f(D)-g(D), $$ hence $$ f(D)-f(A) - g(A)+g(C)+f(B)-f(C)-g(D)+g(B)=0, $$ or $|AD|-|AC|+|BC|-|BD|=0$. This is necessary and sufficient for the quadrilateral $ABCD$ to be circumscribed. $\Box$\bigskip Now, consider a Poncelet $n$-gon, an $n$-periodic billiard trajectory in the ellipse $\Gamma$. Oner can choose the canonical coordinates of the tangency points of the sides of the polygon with the confocal ellipse $\gamma$ to be $$ 0,\ \frac{1}{n},\ \frac{2}{n},\ \dots,\ \frac{n-1}{n}. $$ Then the string coordinates of the points of the concentric set $P_k$ are $$ \left(0, \frac{k}{n}\right), \left(\frac{1}{n}, \frac{k+1}{n}\right), \left(\frac{2}{n}, \frac{k+2}{n}\right), \ldots , $$ that is, their difference equals $k/n$, a constant. It follows that $P_k$ lies on a confocal ellipse. Likewise for the radial sets $Q_k$, proving the first claim of Theorem \ref{Pgrid}. Theorem \ref{inscribed} implies that each quadrilateral of the Poncelet grid is circumscribed, see Figure \ref{circles}. We refer to \cite{AB} for circle patterns related to conics. \begin{figure}[hbtp] \centering \includegraphics[height=2.5in]{circles} \caption{Poncelet grid of circles.} \label{circles} \end{figure} Next, we prove the second claim of Theorem \ref{Pgrid}. The confocal family of conics is given by the equation $$ \frac{x_1^2}{a_1^2+\lambda}+\frac{x_2^2}{a_2^2+\lambda}=1, $$ where $\lambda$ is a parameter. Its dual family is the pencil $$ (a_1^2+\lambda) x_1^2+(a_2^2+\lambda) x_2^2=1 $$ that consists of the conics that share four points, possibly, complex. Hence the confocal family consists of the conics that share four tangent lines, also possibly complex. To prove the last claim of Theorem \ref{Pgrid}, we need the following classical result. Let $\gamma$ and $\Gamma$ be confocal ellipses, centered at the origin and symmetric with respect to the coordinate axes, and let $A$ be the diagonal matrix with positive entries that takes $\gamma$ to $\Gamma$. \begin{lemma}[Ivory] \label{Ivory} For every point $P \in \gamma$, the points $P$ and $A(P)$ lie on a confocal hyperbola. \end{lemma} Let us show that the linear map $A$ takes $P_k$ to $P_m$ or to its centrally symmetric set; the argument for the radial sets is similar. It is convenient to change the string coordinates $(x,y)$ to $u=(x+y)/2, v= (y-x)/2$. The $(u,v)$-coordinates of the points of the sets $P_k$ and $P_m$ are $$ \left(\frac{k}{2n}+\frac{j}{n}, \frac{k}{2n}\right),\ \left(\frac{m}{2n}+\frac{j}{n}, \frac{m}{2n}\right)\ (j=0,1,\dots,n-1). $$ We know that $P_k$ and $P_m$ lie on confocal ellipses $\gamma$ and $\Gamma$. According to Lemma \ref{Ivory}, the map $A$ preserves the $u$-coordinate. Therefore the coordinates of the points of the set $A (P_k)$ are $$ \left(\frac{k}{2n}+\frac{j}{n}, \frac{m}{2n}\right)\ (j=0,1,\dots,n-1). $$ If $m$ has the same parity as $k$, this coincides with the set $P_m$, and if the parity of $m$ is opposite to that of $k$, then this set is centrally symmetric to the set $P_m$. This completes the proof. \section{Identities in the Lie algebras of motions} \label{Lie} It is well known that the altitudes of a Euclidean triangle are concurrent. It is a lesser known fact that an analogous theorem holds in the spherical and hyperbolic geometries. In this section, we describe V. Arnold's observation \cite{Ar} that these results have interpretations as the Jacobi identity in the Lie algebras of motions of the respective geometries of constant positive or negative curvatures; see also \cite{Iv,Sk}. Following \cite{Ai,To}, we shall discuss the relation of othr classical configuration theorems with identities in these Lie algebras. In spherical geometry, one has the duality between points and lines that assigns the pole to an equator. There are two poles of a great circle; one can make the choice of the pole unique by considering oriented great circle, or by factorizing by the antipodal involution, that is, by replacing the sphere by the elliptic plane. This spherical duality can be expressed in terms of the cross-product: if $A$ and $B$ are two vectors in ${\mathbb R}^3$ representing points in the elliptic plane, then the vector $A \times B$ represents the point dual to the line $AB$. In the following argument, we do not distinguish between points and their dual lines. \begin{figure}[hbtp] \centering \includegraphics[height=2.1in]{sphere} \caption{An altitude of a spherical triangle.} \label{sphere} \end{figure} Given a spherical triangle $ABC$, the altitude dropped from $C$ to $AB$ is the great circle connecting the pole of the great circle $AB$ and point $C$. Using the identification of points and lines, and cross-product, this altitude is represented by the vector $(A\times B) \times C$, see Figure \ref{sphere}. Two other altitudes are given by similar cross-products, and the statement that the three great circles are concurrent is equivalent to linear dependence of the these three cross-products. But $$ (A\times B) \times C + (B\times C)\times A + (C\times A)\times B =0, $$ the Jacobi identity for cross-product, hence the three altitudes are concurrent. Note that the Lie algebra $({\mathbb R}^3,\times)$ is $so(3)$, the algebra of motions of the unit sphere. Thus the Jacobi identity in $so(3)$ implies the the existence of the spherical orthocenter. A similar, albeit somewhat more involved, argument works in the hyperbolic plane, with the Lie algebra of motions $sl(2,{\mathbb R})$ replacing $so(3)$. Note that these algebras are real forms of the complex Lie algebra $sl(2,{\mathbb C})$. Interestingly, the Euclidean theorem on concurrence of the three altitudes of a triangle does not seem to admit an interpretation as the Jacobi identity of the Lie algebra of motions of the plane. Developing these ideas, T. Tomihisa \cite{To} discovered the following identity. \begin{theorem} \label{Tid} For every quintuple of elements of the Lie algebra $sl(2)$ (with real or complex coefficients), one has $$ [F_1,[[F_2,F_3],[F_4,F_5]]] + [F_3,[[F_2,F_5],[F_4,F_1]]] + [F_5,[[F_2,F_1],[F_4,F_3]]] =0. $$ \end{theorem} Note that the indices $1,3,5$ permute cyclically, while $2$ and $4$ are frozen. \begin{figure}[hbtp] \centering \includegraphics[height=1.8in]{Tomihisa} \caption{The Tomihisa identity as the dual Pappus theorem: the lines $AF_1, BF_3$, and $CF_5$ are concurrent.} \label{Tomihisa} \end{figure} As above, the Tomihisa identity can be interpreted as a configuration theorem: the Lie bracket corresponds to one of the two basic operations: connecting a pair of points by a line or intersecting a pair of lines at a point. See Figure \ref{Tomihisa} for such an interpretation. \section{Skewers} \label{skewers} This section is based upon the recent paper \cite{Tab}. The main idea is that planar projective configuration theorems have space analogs where points and lines in the projective plane are replaced by lines in space, and the two operations, connecting two points by a line and intersecting two lines at a point, are replaced by taking the common perpendicular of two lines. The {\it skewer} of two lines in 3-dimensional space is their common perpendicular. We denote the skewer of lines $a$ and $b$ by $S(a,b)$. In Euclidean and hyperbolic spaces, a generic pair of lines has a unique skewer; in the spherical geometry, a generic pair of lines (great circles) has two skewers, similarly to a great circle on $S^2$ having two poles. We always assume that the lines involved in the formulations of the theorems are in general position. Here is the `skewer translation' of the Pappus theorem, as depicted in Figure \ref{Pappusfig}: \begin{theorem} \label{skPappus} Let $a_1,a_2,a_3$ be a triple of lines with a common skewer, and let $b_1,b_2,b_3$ be another triple of lines with a common skewer. Then the lines $$ S(S(a_1,b_2),S(a_2,b_1)),\ S(S(a_1,b_3),S(a_3,b_1)),\ {\rm and}\ \ S(S(a_2,b_3),S(a_3,b_2)) $$ share a skewer. \end{theorem} This theorem, as well as in the following ones, holds in ${\mathbb R}^3, S^3$ and $H^3$. And here is the skewer version of the Desargues theorem, as depicted in Figure \ref{Desarguesfig}: \begin{theorem}\label{skDesargues} Let $a_1,a_2,a_3$ and $b_1,b_2,b_3$ be two triples of lines such that the lines $S(a_1,b_1), S(a_2,b_2)$ and $S(a_3,b_3)$ share a skewer. Then the lines $$ S(S(a_1,a_2),S(b_1,b_2)),\ S(S(a_1,a_3),S(b_1,b_3)),\ {\rm and}\ \ S(S(a_2,a_3),S(b_2,b_3)) $$ also share a skewer. \end{theorem} The `rules of translation' should be clear from these examples. As a third example, consider a configuration theorem that involves polarity, namely, the theorem that the three altitudes of a triangle are concurrent that was discussed in Section \ref{Lie}. In its skewer version, one does not distinguish between polar dual objects, such as a great circle and its pole. This yields \begin{theorem} \label{skAlt} Given three lines $a,b,c$, the lines $$ S(S(a,b),c),\ S(S(b,c),a),\ \ {\rm and}\ \ S(S(c,a),b) $$ share a skewer. \end{theorem} This is the Petersen-Morley, also known as Hjelmslev-Morley, theorem \cite{Mo}. An equivalent formulation: {\it the common normals of the opposite sides of a rectangular hexagon have a common normal}. See Figure \ref{ten}, borrowed from \cite{Mo2}. \begin{figure}[hbtp] \centering \includegraphics[height=1.8in]{ten} \caption{Petersen-Morley configuration of ten lines.} \label{ten} \end{figure} Denote the 2-parameter family of lines that meet a given line $\ell$ at right angle by ${\cal N}_{\ell}$. The sets ${\cal N}_{\ell}$ plays the role of lines in the skewer versions of configuration theorems. Two-parameter families of lines in 3-space are called congruences. Next we describe line analogs of circles. Let $\ell$ be an oriented line in 3-space (elliptic, Euclidean, or hyperbolic). Let $G_\ell$ be the 2-dimensional subgroup of the group of orientation preserving isometries that preserve $\ell$. The orbit $G_\ell(m)$ of an oriented line $m$ is called the {\it axial congruence} with $\ell$ as axis (an analog of the center of a circle). In ${\mathbb R}^3$, the lines of an axial congruence with axis $\ell$ are at equal distances from $\ell$ and make equal angles with it. In the hyperbolic space, one can define the so-called complex distance between oriented lines, see \cite{Ma}. The complex distance between the lines of an axial congruence and its axis is constant. Axial congruences share the basic properties of circles: if two generic axial congruences share a line, then they share a unique other line; and three generic oriented lines belong to a unique axial congruence. The next result is a skewer analog of the Pascal theorem, see Figure \ref{Pascalfig}, in the particular case when the conic is a circle. \begin{theorem} \label{skPascal} Let $A_1,\ldots,A_6$ be lines from an axial congruence. Then $$ S(S(A_1,A_2),S(A_4,A_5)),\ S(S(A_2,A_3),S(A_5,A_6)), \ {\rm and}\ S(S(A_3,A_4),S(A_6,A_1)) $$ share a skewer. \end{theorem} As another, lesser known, example, consider the Clifford's Chain of Circles. This chain of theorems starts with a collection of concurrent circles labelled $1,2,3,\ldots, n$. The intersection point of the circles $i$ and $j$ is labelled $ij$. The circle through points $ij, jk$ and $ki$ is labelled $ijk$. The first statement of the theorem is that the circles $ijk, jkl, kli$ and $lij$ share a point; this point is labelled $ijkl$. The next statement is that the points $ijkl, jklm, klmi, lmij$ and $mijk$ are concyclic; this circle is labelled $ijklm$. And so on, with the alternating claims of being concurrent and concyclic; see \cite{Coo,Mo2}, and Figure \ref{Clifford} where the initial circles are represented by lines (circles of infinite radius sharing a point at infinity). \begin{figure}[hbtp] \centering \includegraphics[height=3.2in]{Clifford} \caption{Clifford's Chain of Circles ($n=5$).} \label{Clifford} \end{figure} The next theorem, in the case of ${\mathbb R}^3$, is due to Richmond \cite{Ri}. \begin{theorem} \label{skClifford} 1) Consider axial congruences ${\cal C}_i,\ i=1,2,3,4$, sharing a line. For each pair of indices $i,j \in \{1,2,3,4\}$, denote by $\ell_{ij}$ the line shared by ${\cal C}_i$ and ${\cal C}_j$. For each triple of indices $i,j,k \in \{1,2,3,4\}$, denote by ${\cal C}_{ijk}$ the axial congruence containing the lines $\ell_{ij},\ell_{jk},\ell_{ki}$. Then the congruences ${\cal C}_{123}, {\cal C}_{234}, {\cal C}_{341}$ and ${\cal C}_{412}$ share a line. \\ 2) Consider axial congruences ${\cal C}_i,\ i=1,2,3,4,5$, sharing a line. Each four of the indices determine a line, as described in the previous statement of the theorem. One obtains five lines, and they all belong to an axial congruence.\\ 3) Consider axial congruences ${\cal C}_i,\ i=1,2,3,4,5,6$, sharing a line. Each five of them determine an axial congruence, as described in the previous statement of the theorem. One obtains six axial congruences, and they all share a line. And so on... \end{theorem} Next one would like to define line analogs of conics. A first step in this direction is made in \cite{Tab}, but much more work is needed. In particular, one would like to have skewer analogs of various configuration theorems involving conics, including the Pascal theorem and the whole hexagrammum mysticum, the Poncelet Porism, and the theorems described in Section \ref{pentalike}. As of now, this is an open problem. Now we outline two approaches to proofs of the above theorems and the skewer versions of other planar configuration theorems. The first approach is by way of the spherical geometry, and the second via the hyperbolic geometry. Either approach implies the results in all three classical geometries by `analytic continuation'. This analytic continuation principle is well known in geometry; see, e.g., \cite{AP1,Pa} where it is discussed in detail. \paragraph{Elliptic approach.} The space of oriented great circles in $S^3$, or lines in the elliptic space ${\mathbb {RP}}^3$, is he Grassmannian $G(2,4)$ of oriented 2-dimensional subspaces in ${\mathbb R}^4$. Below we collect pertinent facts concerning this Grassmannian. To every oriented line $\ell$ in ${\mathbb {RP}}^3$ there corresponds its dual oriented line $\ell^*$: the respective oriented planes in ${\mathbb R}^4$ are the orthogonal complements of each other. The dual lines are equidistant and they have infinitely many skewers. The Grassmannian is a product of two spheres: $G(2,4)=S^2_-\times S^2_+$. This provides an identification of an oriented line in ${\mathbb {RP}}^3$ with a pair of points of the unit sphere $S^2$: $\ell\leftrightarrow (\ell_-,\ell_+)$. The antipodal involutions of the spheres $S^2_-$ and $S^2_+$ generate the action of the Klein group ${\mathbb Z}_2\times{\mathbb Z}_2$ on the space of oriented lines generated by reversing the orientation of a line and by taking the dual line. Two lines $\ell$ and $m$ intersect at right angle if and only if $d(\ell_-,m_-)=d(\ell_+,m_+)=\pi/2$, where $d$ denotes the spherical distance in $S^2$. A line $n$ is a skewer of lines $\ell$ and $m$ if and only if $n_-$ is a pole of the great circle $\ell_- m_-$, and $n_+$ is a pole of the great circle $\ell_+ m_+$. The set of lines that intersect $\ell$ at right angle coincides with the set of lines that intersect $\ell$ and $\ell^*$. A generic pair of lines has exactly two skewers (four, if orientation is taken into account), and they are dual to each other. It follows that a configuration involving lines in elliptic space and their skewers can be identified with a pair of configurations on the spheres $S^2_-$ and $S^2_+$. Under this identification, the great circles of these spheres are not distinguished from their poles, just like in the proof described in Section \ref{Lie}. That is, the operation of taking the skewer of two lines is represented, on both spheres, by the cross-product. In this way, a configuration of lines in space becomes the direct product of the corresponding planar configurations. For example, the Petersen-Morley Theorem \ref{skAlt} splits into two statements that the altitudes of triangles, on the spheres $S^2_-$ and $S^2_+$, are concurrent. \paragraph{Hyperbolic approach.} In a nutshell, a skewer configuration theorem in 3-dimensional hyperbolic space is a complexification of a configuration theorem in the hyperbolic plane. We follow the ideas of F. Morley \cite{Mo1,Mo2}, Coxeter \cite{Co}, and V. Arnold \cite{Ar}. Consider the hyperbolic space in the upper halfspace model. The isometry group is $SL(2,{\mathbb C})$, and the sphere at infinity (the celestial sphere of \cite{Mo1}) is the Riemann sphere ${\mathbb {CP}}^1$. A line in $H^3$ intersects the sphere at infinity at two points, hence the space of (non-oriented) lines is the configuration space of unordered pairs of points. As we mentioned in Section \ref{itSteiner}, $S^2({\mathbb {CP}}^1)={\mathbb {CP}}^2$, namely, to a pair of points in the projective line one assigns the binary quadratic form having zeros at these points: $$ (a_1:b_1,a_2:b_2) \longmapsto (a_1y-b_1x)(a_2y-b_2x). $$ Thus a line in $H^3$ can be though of as a complex binary quadratic form, up to a factor. The space of binary quadratic forms $ax^2+2bxy+cy^2$ has the discriminant quadratic form $\Delta=ac-b^2$ and the respective bilinear form. The equation $\Delta=0$ defines the diagonal of $S^2({\mathbb {CP}}^1)$; this is a conic in ${\mathbb {CP}}^2$ that does not correspond to lines in $H^3$. The next result is contained in \S 52 of \cite{Mo2}. \begin{lemma} \label{Jacobian} Two lines in $H^3$ intersect at right angle if and only if the respective binary quadratic forms $f_i=a_i x^2 + 2 b_i xy + c_i y^2,\ i=1,2$, are orthogonal with respect to $\Delta$: \begin{equation} \label{ort} a_1c_2-2b_1b_2+a_2c_1=0. \end{equation} If two lines correspond to binary quadratic forms $f_i=a_i x^2 + 2 b_i xy + c_i y^2,\ i=1,2$, then their skewer corresponds to the Poisson bracket (the Jacobian) $$ \{f_1,f_2\} = (a_1b_2-a_2b_1)x^2 + (a_1c_2-a_2c_1) xy + (b_1c_2-b_2c_1) y^2. $$ \end{lemma} If $(a_1:b_1:c_1)$ and $(a_2:b_2:c_2)$ are homogeneous coordinates in the projective plane and the dual projective plane, then (\ref{ort}) describes the incidence relation between points and lines. In particular, the set of lines in $H^3$ that meet a fixed line at right angle corresponds to a line in ${\mathbb {CP}}^2$. Suppose a configuration theorem involving polarity is given in ${\mathbb {RP}}^2$. The projective plane with a conic provides the projective model of the hyperbolic plane, so the configuration is realized in $H^2$. Consider the complexification, the respective configuration theorem in ${\mathbb {CP}}^2$ with the polarity induced by $\Delta$. According to Lemma \ref{Jacobian}, this yields a configuration of lines in $H^3$ such that the pairs of incident points and lines correspond to pairs of lines intersecting at right angle. \begin{remark}[On Lie algebras] \label{Liealgebraic} {\rm From the point of view of the identities in Lie algebras, discussed in Section \ref{Lie}, the relation between configuration theorems in the hyperbolic plane and the hyperbolic space is the relation between $sl(2,{\mathbb R})$ and $sl(2,{\mathbb C})$: an identity in the former implies the same identity in the latter. As to the Lie algebras in space, in the elliptic case, the Lie algebra of motions is $so(4)=so(3)\oplus so(3)$, and in the hyperbolic case, it is $sl(2,{\mathbb C})$. Accordingly, an elliptic skewer configuration splits into two configurations in $S^2$, and a hyperbolic skewer configuration is obtained from a configuration in $H^2$ by complexification. } \end{remark} We finish the section by discussing two results concerning lines in 3-space that do not follow the above described general pattern. The first of them is the skewer version of the Sylvester Problem. Given a finite set $S$ of points in the plane, assume that the line through every pair of points in $S$ contains at least one other point of $S$. J.J. Sylvester asked in 1893 whether $S$ necessarily consists of collinear points. See \cite{BM} for the history of this problem and its generalizations. In ${\mathbb {RP}}^2$, the Sylvester Problem, along with its dual, has an affirmative answer (the Sylvester-Galai theorem), but in ${\mathbb {CP}}^2$ one has a counter-example: the 9 inflection points of a cubic curve (of which at most three can be real, according to a theorem of Klein), connected by 12 lines. The skewer version of the Sylvester Problem concerns a finite collection of pairwise skew lines in space such that the skewer of any pair intersects at least one other line at right angle. The question is whether a collection of lines with this skewer Sylvester property necessarily consists of the lines that intersect some line at right angle. \begin{theorem} \label{Sylvth} The skewer version of the Sylvester-Galai theorem holds in the elliptic and Euclidean geometries, but fails in the hyperbolic geometry. \end{theorem} \paragraph{Proof.} In the elliptic case, we argue as in the above described elliptic proof. A collection of lines becomes two collections of points, in ${\mathbb {RP}}^2_-$ and in ${\mathbb {RP}}^2_+$, and the skewer Sylvester property implies that each of these sets has the property that the line through every pair of points contains another point, so one applies the Sylvester-Galai theorem on each sphere. In the hyperbolic case, we argue as in the hyperbolic proof. Let $a_1,\ldots,a_9$ be the nine inflection points of a cubic curve in ${\mathbb {CP}}^2$, and let $b_1,\ldots,b_{12}$ be the respective lines. Let $b_1^*,\ldots,b_{12}^*$ be the polar dual points. Then the points $a_i$ correspond to nine lines in $H^3$, and the points $b_j^*$ to their skewers. We obtain a collection of nine lines that has the skewer Sylvester property but does not possess a common skewer. In the intermediate case of ${\mathbb R}^3$, the argument is due to V. Timorin (private communication). Let us add to ${\mathbb R}^3$ the plane at infinity $H$; the points of $H$ are the directions of lines in space. One has a polarity in $H$ that assigns to a direction the set of the orthogonal directions, a line in $H$. Therefore, if three lines in ${\mathbb R}^3$ share a skewer, then their intersections with the plane $H$ are collinear. Let $L_1,\ldots, L_n$ be a collection of lines with the skewer Sylvester property. Then, by the Sylvester-Galai theorem in $H$, the points $L_1 \cap H,\ldots, L_n \cap H$ are collinear. This means that the lines $L_1,\ldots, L_n$ lie in parallel planes, say, the horizontal ones. Consider the vertical projection of these lines. We obtain a finite collection of non-parallel lines in the plane such that through the intersection point of any two there passes at least one other line. By the dual Sylvester-Galai theorem, all these lines are concurrent. Therefore the respective horizontal lines in ${\mathbb R}^3$ share a vertical skewer. $\Box$\bigskip The second result is a different skewer version of the Pappus theorem. \begin{theorem} \label{othPapp} Let $\ell$ and $m$ be a pair of skew lines. Choose a triple of points $A_1,A_2,A_3$ on $\ell$ and a triple of points $B_1,B_2,B_3$ on $m$. Then the lines $$ S((A_1 B_2), (A_2 B_1)), \ S((A_2 B_3), (A_3 B_2)),\ {\rm and}\ \ S((A_3 B_1), (A_1 B_3)) $$ share a skewer. \end{theorem} We proved this result, in the hyperbolic case, by a brute force calculation using the approach to hyperbolic geometry, developed in \cite{Fe}; see \cite{Tab} for details. It is not clear whether this theorem is a part of a general pattern. Let us close with inviting the reader to mull over the skewer versions of other constructions of planar projective geometry. For example, one can define the skewer pentagram map that acts on cyclically ordered tuples of lines in space: $$ \{L_1, L_2,\ldots \} \mapsto \{S(S(L_1,L_3),S(L_2,L_4)), S(S(L_2,L_4),S(L_3,L_5)), \ldots \} $$ Is this map completely integrable?
5468cfb4c7702025bd5eb65395f5e9e81f8a552e
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec.intro} Random linear network coding (RNLC) is the process of constructing coded packets, which are random linear combinations of source packets over a finite field \cite{Ho06}. If $k$ source packets are considered, decoding at a receiving node starts after $k$ linearly independent coded packets have been collected. The probability of recovering all of the $k$ source packets when at least $k$ coded packets have been received has been derived in \cite{Trullols-Cruces11}. However, the requirement for a large number of received coded packets before decoding can introduce undesirable delays at the receiving nodes. In an effort to alleviate this problem, \textit{rank-deficient} decoding was proposed in \cite{Yan2013} for the recovery of a subset of source packets when fewer than $k$ coded packets have been obtained. Whereas the literature on network coding defines \textit{decoding success} as the recovery of 100\% of the source packets with a certain probability, the authors of \cite{Yan2013} presented simulation results that measured the \textit{fraction of decoding success}, that is, the recovery of a percentage of the source packets with a certain probability. The fundamental problem that has motivated our work is the characterization of the probability of recovering some of the $k$ source packets when $n$ coded packets have been retrieved, where $n$ can be smaller than, equal to or greater than $k$. This idea was considered in~\cite{Gadouleau2011} for random network communications over a matroid framework. The authors show that partial decoding is highly unlikely. This problem has also been explored in the context of secure network coding, e.g., \cite{Bhattad2005,Lima2007}. Strict information-theoretic security can be achieved if and only if the mutual information between the packets available to an eavesdropper and the source packets is zero \cite{Cai2002}. When network coding is used, \textit{weak} security can be achieved if the eavesdropper cannot obtain $k$ linearly independent coded packets and, hence, cannot recover any meaningful information about the $k$ source packets \cite{Bhattad2005}. The authors of \cite{Bhattad2005} obtained bounds on the probability of RLNC being weakly secure and showed that the adoption of large finite fields improves security. A different setting but a similar problem was investigated in \cite{Lima2007}. Intermediate relay nodes between transmitting and receiving nodes were treated as potentially malicious, and criteria for characterizing the algebraic security of RLNC were defined. The authors demonstrated that the probability of an intermediate node recovering a strictly positive number of source packets tends to zero as the field size and the number of source packets go to infinity. This paper revisits the aforementioned problem and obtains an exact expression for the probability that a receiving node will recover at least $x$ of the $k$ source packets if $n$ coded packets are collected, for $x\leq n$. The derived expression can be seen as a generalization of \cite[eq. (7)]{Trullols-Cruces11}. The \mbox{paper also looks} at the impact of transmitting source packets along with coded packets, known as \textit{systematic} RLNC, as opposed to transmitting only coded packets, referred to as \textit{non-systematic} RLNC. In the remainder of the paper, Section~\ref{sec.System} formulates the problem, Section~\ref{sec.Analysis} obtains the probability of recovering a fraction of a network-coded message, Section~\ref{sec.results} presents results and Section~\ref{sec.conclusions} summarizes the conclusions of this work. \section{System model and problem formulation} \label{sec.System} We consider a receiving network node, which collects $n$ packets and attempts to reconstruct a message that consists of $k$ source packets. The $n$ packets could have been broadcast by a single transmitting node or could have been originated from multiple nodes that possess the same message. In the case of \textit{non-systematic} communication, transmitted packets are generated from the $k$ source packets using RLNC over $\mathbb{F}_q$ \cite{Ho06}, where $q$ is a prime power and $\mathbb{F}_q$ denotes the finite field of $q$ elements. In the case of \textit{systematic} RLNC, a sequence of $n_\mathrm{T}$ transmitted packets consists of the $k$ source packets and $n_\mathrm{T}-k$ coded packets that have been generated as in the non-systematic case. In both cases, a coding vector of length $k$, which contains the weighting coefficients used in the generation of a packet, is transmitted along with each packet. At the receiving node, the coding vectors of the $n$ successfully retrieved packets form the rows of a matrix $\mathbf{M}\in\mathbb{F}^{n\times k}_q$, where $\mathbb{F}_q^{n \times k}$ denotes the set of all $n \times k$ matrices over $\mathbb{F}_q$. The $k$ source packets can be recovered from the $n$ received packets if and only if $k$ of the $n$ coding vectors are linearly independent, implying that $\mathrm{rank}(\mathbf{M})=k$ for $n\geq k$. The probability that the $n\times k$ random matrix $\mathbf{M}$ has rank $k$ and, thus, the receiving node can reconstruct the entire message is given in \cite{Trullols-Cruces11} for non-systematic RLNC and \cite{Shrader09} for systematic RLNC. The objective of this paper is to derive the probability that a receiving node will reconstruct at least $x\leq k$ source packets upon reception of $n$ network-coded packets. To formulate this problem, let $\mathbf{e}_i$ denote the $i$-th unit vector of length $k$. A coding vector, or a row of $\mathbf{M}$, equal to $\mathbf{e}_i$ represents the $i$-th source packet. Let $X$ be the set of indices corresponding to the unit vectors contained in the rowspace of $\mathbf{M}$, denoted by $\Row(\mathbf{M})$, so that $X=\{i:\mathbf{e}_i \in \Row(\mathbf{M})\}$. We write $\lvert X\rvert$ to denote the cardinality of random variable $X$. Furthermore, we define random variables $R$ and $N$ to give the rank of $\mathbf{M}$ and the number of rows in $\mathbf{M}$, respectively. The considered problem has been decomposed into the following two tasks: \begin{enumerate} \item Obtain the probability of recovering at least $x$ source packets, provided that $r$ out of the $n$ received packets are linearly independent, for $x\leq r\leq k$. This is equivalent to finding the probability of $\Row(\mathbf{M})$ containing at least $x$ unit vectors, given $\mathbf{M}$ has $n$ rows and rank $r$. We denote this probability by $P(\lvert X\rvert\geq x\,\vert\,R=r,\,N=n)$. \item {Obtain the probability of recovering at least $x$ source packets, provided that $n\geq x$ packets have been collected.} We write $P(\lvert X\rvert\geq x\,\vert\,N=n)$ to refer to this probability. \end{enumerate} Derivation of $P(\lvert X\rvert\geq x\,\vert\,R=r,\,N=n)$ and \mbox{$P(\lvert X\rvert\geq x\,\vert\,N=n)$} is the focus of the following section. \section{Probability analysis} \label{sec.Analysis} The analysis presented in this section relies on the well- known Principle of Inclusion and Exclusion \cite[Prop. 5.2.2]{Cameron1994}, which is repeated below for clarity. \begin{lemma} \label{lemma.PIE} \textbf{Principle of inclusion and exclusion.} Given a set $A$, let $f$ be a real valued function defined for all sets $S,J\subseteq A$. If $g(S) = \sum_{J:J\supseteq S} f(J)$ then $f(S) = \sum_{J:J\supseteq S} (-1)^{\lvert J\setminus S\rvert} g(J)$. \end{lemma} For non-negative integers $m$ and $d$, we denote by $\binom{m}{d}$ the binomial coefficient, which gives the number of $d$-element sets of an $m$-element set. The $q$-analog of the binomial coefficient, known as the \textit{Gaussian binomial coefficient} and denoted by $\qbinom{m}{d}$, enumerates all $d$-dimensional subspaces of an $m$-dimensional space over $\mathbb{F}_q$ \cite[p. 125]{Cameron1994}. Given $\mathbf{M}$ has rank $r$, let $P(\lvert X\rvert=x\,\vert\,R=r,\,N=n)$ denote the probability of recovering \textit{exactly} $x \leq r$ source packets or, equivalently, the probability of $\Row(\mathbf{M})$ containing \textit{exactly} $x \leq r$ unit vectors. The following theorem obtains an expression for $P(\lvert X\rvert= x\,\vert\,R=r,\,N=n)$, which is then used in the derivation of $P(\lvert X\rvert\geq x\,\vert\,R=r,\,N=n)$. \begin{theorem} \label{thm.cond_exactly_x} Given a random $n \times k$ matrix $\mathbf{M}$ of rank $r$, the probability that the rowspace of $\mathbf{M}$ contains exactly $x \leq r$ unit vectors is given by \begin{equation} \label{eq.cond_exactly_x} P(\lvert X\rvert=x\,\vert\,R=r,\,N=n)=\frac{\binom{k}{x}}{\qbinom{\phantom{\widetilde k}\!\!\!k}{r}}% \displaystyle\sum_{j=0}^{k-x}(-1)^{j}\binom{k-x}{j}\!\genfrac{[}{]}{0pt}{}{\,k-x-j\,}{\,r-x-j\,}_{\!q}. \end{equation} \end{theorem} \begin{proof} For $S\!\subseteq\!J\subseteq\!\{1, \dots k \}$, let $g(S)$~be~the~probability that $\{ \mathbf{e}_i : i \in S \} \subseteq \Row(\mathbf{M})$, that is, the probability that $S \subseteq X$. This is just the probability that $\Row( \mathbf{M} )$ contains a fixed $\lvert S\rvert$-dimensional subspace, namely the space \mbox{$V=\Span\{\mathbf{e}_i:i \in S\}$}. We see that, by considering the quotient space $\mathbb{F}_q^k/V$, there is a direct correspondence between \mbox{$r$-dimensional} subspaces of $\mathbb{F}_q^k$ containing $V$, and $(r-\lvert S\rvert)$-dimensional subspaces of a \mbox{$(k-\lvert S\rvert)$-dimensional} space. Hence, there are $\qbinom{k-\lvert S \rvert}{r-\lvert S \rvert}$ $r$-dimensional subspaces of $\mathbb{F}_q^k$ containing $V$. The probability that $\Row(\mathbf{M})$ contains the space $V$ is equal to \begin{equation} \label{g(S)} g(S) = \frac{\qbinom{k-\lvert S \rvert}{r-\lvert S \rvert}}{\qbinom{\phantom{\widetilde k}\!\!\!k}{r}}. \end{equation} where the denominator in \eqref{g(S)} enumerates the $r$-dimensional subspaces of $\mathbb{F}_q^k$. Now, let $f(S)$ be the probability that $S=X$, that is, the probability that $\{ \mathbf{e}_i : i \in S \} \subseteq \Row( \mathbf{M})$ and $\mathbf{e}_i \notin \Row(\mathbf{M})$ for $i\notin S$. It follows that $g(S) = \sum_{J \supseteq S} f(J)$. Invoking the Principle of Inclusion and Exclusion (Lemma~\ref{lemma.PIE}) and using \eqref{g(S)}, we can write $f(S) = \sum_{J \supseteq S} (-1)^{|J\setminus S|} \cdot g(J)$ and expand it to \begin{align} f(S) \notag&= \sum_{J \supseteq S} (-1)^{|J\setminus S|} \cdot \frac{\qbinom{k-\lvert J \rvert}{r-\lvert J \rvert}}{\qbinom{k}{r}} \notag \\ &= \frac{1}{\qbinom{k}{r}} \sum_{J' \subseteq \{ 1, \dots , k \}\setminus S} (-1)^{\lvert J' \rvert} \qbinom{k-\lvert S\rvert-\lvert J'\rvert}{r-\lvert S\rvert-\lvert J'\rvert} \label{f(S):J'} \\ &= \frac{1}{\qbinom{k}{r}} \sum_{j=0}^{k-\lvert S\rvert} (-1)^{j} \binom{k-\lvert S\rvert}{j} \qbinom{k-\lvert S\rvert-j}{r-\lvert S\rvert-j} \label{f(S):j} \end{align} where \eqref{f(S):J'} follows by setting $J'=J\setminus S$, and \eqref{f(S):j} follows since there are $\binom{k-\lvert S\rvert}{j}$ sets $J'$ of size $j$. Considering that $f(S)$ is the probability that $X=S$, we can write \begin{equation} P(\lvert X\rvert=x\,\vert\,R=r,\,N=n) =\!\!\sum_{S: \lvert S \rvert =x}\!\!\!f(S) = \binom{k}{x} f(S') \label{eq:P(|X|=x)} \end{equation} where $S'$ is any subset of $\{1, \dots, k \}$ of size $x$. The second equality in \eqref{eq:P(|X|=x)} holds since there are $\binom{k}{x}$ sets \mbox{$S \subseteq \{1, \dots, k\}$} of size $x$. Substituting \eqref{f(S):j} in \eqref{eq:P(|X|=x)} gives the result. \end{proof} \begin{remark} Theorem~\ref{thm.cond_exactly_x} can be seen as a special case of~\cite[Proposition 6]{Gadouleau2011}. Whereas the proof in \cite{Gadouleau2011} uses elements of matroid theory, our paper proposes an alternative and more intuitive proof strategy. \end{remark} \begin{corollary} \label{cor.cond_atleast_x} Given a random $n \times k$ matrix $\mathbf{M}$ of rank $r$, the probability that the rowspace of $\mathbf{M}$ contains at least $x \leq r$ unit vectors is given by \begin{equation} \label{eq.cond_atleast_x} P(\lvert X\rvert\!\geq\! x\,\vert\,R\!=\!r,\,N\!=\!n) =\frac{1}{\qbinom{\phantom{\widetilde k}\!\!\!k}{r}}% \!\sum_{i=x}^{r}\!\binom{k}{i}\!\sum_{j=0}^{k-i}(-1)^{j}\binom{k-i}{j}\!\qbinom{k-i-j}{r-i-j}. \end{equation} \end{corollary} \begin{proof} By definition, the probability $P(\lvert X\rvert\geq x\,\vert\,R=r,\,N=n)$ is equal to $\sum_{i=x}^{r} P(\lvert X\rvert=i\,\vert\,R=r,\,N=n)$. Substituting in \eqref{eq.cond_exactly_x} gives the result. \end{proof} Note that, although $\mathbf{M}$ is an $n \times k$ matrix, the probabilities in \eqref{eq.cond_exactly_x} and \eqref{eq.cond_atleast_x} hold for any value of $n\!\geq\!r$. Having obtained an expression for the probability $P(\lvert X\rvert\!\geq\!x\,\vert\,R\!=\!r,\,N\!=\!n)$, we now proceed to the derivation of $P(\lvert X\rvert\!\geq\!x\,\vert\,N\!=\!n)$. This probability is denoted by $P_\mathrm{ns}(\lvert X\rvert\!\geq\!x\,\vert\,N\!=\!n)$ and $P_\mathrm{s}(\lvert X\rvert\!\geq\! x\,\vert\,N\!=\!n)$ for non-systematic and systematic RLNC, respectively. Expressions for each case are derived in the following two propositions. \begin{proposition} \label{prop.nonsys} If a receiving node collects $n$ random linear combinations of $k$ source packets, the probability that at least $x\leq k$ source packets will be recovered is \begin{align} \label{eq.prob_ns} &P_\mathrm{ns}(\lvert X\rvert \geq x \,\vert\,N=n)\nonumber\\% &=\frac{1}{q^{nk}} \displaystyle\sum_{r=x}^{\min(n,k)}\!% \Biggl(% \displaystyle\sum_{i=x}^{r}% \binom{k}{i}% \displaystyle\sum_{j=0}^{k-i}% \left(-1\right)^{j}% \binom{k-i}{j}\!% \genfrac{[}{]}{0pt}{}{\,k-i-j\,}{\,r-i-j\,}_{\!q\!}% \Biggr)\!% \displaystyle\prod_{\ell=0}^{r-1}(q^n-q^\ell).% \end{align} \end{proposition} \begin{proof} Let $P(R\!=\!r\,\vert\,N\!=\!n)$ denote the probability that the $n\times k$ matrix $\mathbf{M}$ has rank $r$. This is equivalent to the probability that $r$ out of the $n$ collected packets are linearly independent. The probability that at least $x$ of the $k$ source packets will be recovered can be obtained from \begin{align} \label{eq.ns_step1} &P_\mathrm{ns}(\lvert X\rvert\geq x \,\vert\,N=n)\nonumber\\ &=\!\!\!\sum_{r=x}^{\min(n,k)}\!\!P(R=r\,\vert\,N=n)\,P(\lvert X\rvert\geq x\,\vert\,R=r,N=n). \end{align} The probability $P(R=r\,\vert\,N\!=\!n)$ is equal to \cite[Sec. II.A]{Gadouleau2010} \begin{equation} \label{eq.ns_step3} P(R=r\,\vert\,N\!=\!n) = \frac{1}{q^{nk}}\qbinom{n}{r}% \prod_{\ell=0}^{r-1}\left(q^{k}-q^{\ell}\right).% \end{equation} Substituting \eqref{eq.cond_atleast_x} and \eqref{eq.ns_step3} into \eqref{eq.ns_step1} and taking into account that \begin{equation} \label{eq.ns_step4} \frac{\qbinom{n}{r}}{\qbinom{k}{r}}\prod_{\ell=0}^{r-1}(q^k-q^\ell)=\prod_{\ell=0}^{r-1}(q^n-q^\ell) \end{equation} leads to \eqref{eq.prob_ns}. \end{proof} \newpage \begin{proposition} \label{prop.sys} If $k$ source packets and $n_\mathrm{T}-k$ random linear combinations of those $k$ source packets are transmitted over single-hop links, the probability that a receiving node will recover at least $x\leq k$ source packets from $n\leq n_\mathrm{T}$ received packets is \begin{align} \label{eq.prob_s} &P_\mathrm{s}(\lvert X\rvert\geq x\,\vert\, N=n )% \nonumber% \\ &=\frac{1}{\binom{n_\mathrm{T}}{n}} \cdot\!\!\displaystyle\sum_{r=x}^{\min(n,k)}\!\!\!% \sum_{h=h_{\min}}^{r}\!\!\Biggl(% \binom{k}{h}\binom{n_\mathrm{T}-k}{n-h}% q^{-(n-h)(k-h)}% \!\!\prod_{\ell=0}^{r-h-1}% (q^{n-h}-q^\ell)% \cdot\nonumber% \\ &\quad\cdot\!\!\displaystyle\sum_{i=x_{\min}}^{r-h}% \!\!\!\binom{k-h}{i}% \!\sum_{j=0}^{k-h-i}% \left(-1\right)^{j}% \binom{k-h-i}{j}% \qbinom{k-h-i-j}{r-h-i-j}% \Biggr)% \end{align} where $h_{\min}\!=\!\max{(0,n\!\,-\!\,n_\mathrm{T}\!\,+\!\,k)}$ and $x_{\min}\!=\!\max(0,x\!-\!h)$. \end{proposition} \begin{proof} Let us assume that some or none of the $k$ transmitted source packets have been received and let $X^{\prime}\subseteq X$ be the set of indices of the remaining source packets that can be recovered from the received coded packets. If $n^{\prime}$ of the $n_\mathrm{T}-k$ coded packets have been received and $k^{\prime}$ source packets remain to be recovered, the respective coding vectors will form an $n^{\prime}\times k^{\prime}$ random matrix $\mathbf{M}^{\prime}$. The probability that $r^{\prime}\leq\min(k^{\prime}, n^{\prime})$ coding vectors are linearly independent and at least $x^{\prime}\leq r^{\prime}$ source packets can be recovered is given by \begin{multline} \label{eq.s_step1} P(\lvert X^{\prime}\rvert\geq x^{\prime}, R^{\prime}=r^{\prime}\,\vert\,N^{\prime}=n^{\prime})= \\ P(R^{\prime}=r^{\prime}\,\vert\,N^{\prime}=n^{\prime})\,% P(\lvert X^{\prime}\rvert\geq x^{\prime}\,\vert\, R^{\prime}=r^{\prime},\,N^{\prime}=n^{\prime}) \nonumber \end{multline} where the two terms of the product can be obtained from \eqref{eq.ns_step3} and \eqref{eq.cond_atleast_x}, respectively. The random variables $N'$ and $R'$ denote the number of received coded packets and the rank of matrix $\mathbf{M}^{\prime}$, respectively. If $n$ of the $n_\mathrm{T}$ transmitted packets are received, the probability that $h$ of them are source packets and the remaining $n-h$ are coded packets is \begin{equation} \label{eq.s_step2} P(N^{\prime}=n-h\,\vert\, N=n)=\frac{\binom{k}{h}\binom{n_\mathrm{T}-k}{n-h}}{\binom{n_\mathrm{T}}{n}}.% \end{equation} The coding vectors of the $n$ received packets compose a matrix of rank $r$, based on which $x$ or more source packets can be recovered when $h$ of the $n$ received packets are source packets. Parameters $x^{\prime}$, $r^\prime$, $k^{\prime}$ and $n^{\prime}$, which are concerned with the received \textit{coded} packets only, can be written as $x-h$, $r-h$, $k-h$ and $n-h$, respectively. The probability of recovering at least $x$ source packets for all valid values of $r$ and $h$ is \begin{align} \label{eq.s_step3} P&_\mathrm{s}(\lvert X\rvert\geq x \,\vert\, N=n)\nonumber \\ =&% \sum_{r=x}^{\min(n,k)}\!\!% \sum_{h=h_{\min}}^{r}\!\!% P(N^{\prime}=n-h\,\vert\, N=n)\,\cdot\nonumber% \\ &\cdot P\bigl(\lvert X^{\prime}\rvert \geq\max(0,\,x\!-\!h),R^{\prime}=r-h\,\vert\,N^{\prime}=n\!-\!h\bigr) \end{align} which expands into \eqref{eq.prob_s}. Note that $\max(0,\,x-h)$ ensures that the value of $\lvert X^{\prime}\rvert$ is a non-negative integer when $h>x$. \end{proof} \begin{remark} In systematic RLNC, if the receiving node attempts to recover source packets as soon as the transmission is initiated, i.e., $n_\mathrm{T}\!\leq\! k$, at least $x$ source packets will certainly be recovered when $n\!\geq\! x$ source packets are received, that is, \begin{equation} \label{eq.prob_s_less_than_k} P_\mathrm{s}(\lvert X\rvert\geq x\,\vert\, N=n)=\left\{% \begin{array}{ll} 1,&\mathrm{if}\;\; n_\mathrm{T}\leq k\;\;\mathrm{and}\;\;x \leq n\\ 0,&\mathrm{if}\;\; n_\mathrm{T}\leq k\;\;\mathrm{and}\;\;x > n.% \end{array}\right. \end{equation} \end{remark} \section{Results and discussion} \label{sec.results} \begin{figure}[t] \centering \includegraphics[width=0.95\columnwidth]{clari1.eps} \caption{{Simulation results and theoretical values for (a) non-systematic RLNC and (b) systematic RLNC. The probability of recovering at least $x$ source packets has been plotted for $q=2$, $k=20$, $x=1,5,10,20$ and $n_\mathrm{T}=30$.}} \label{fig.theory_vs_sims} \end{figure} In order to demonstrate the exactness of the derived expressions, simulations that generated 60000 realisations of an $n\times k$ random matrix $\mathbf{M}$ over $\mathbb{F}_2$ were carried out for $n=1,\dots,30$ and $k\!=\!20$. In each case, matrix $\mathbf{M}$ was converted into reduced row echelon form using Gaussian elimination. Then, the rows that correspond to unit vectors $\mathbf{e}_i$, which represent recoverable source packets, were counted and averaged over all realisations. Fig.~\ref{fig.theory_vs_sims}(a) and Fig.~\ref{fig.theory_vs_sims}(b) show that measurements obtained through simulations match the calculations obtained from \eqref{eq.prob_ns} and \eqref{eq.prob_s} for non-systematic RLNC and systematic RLNC, respectively. In general, simulation results match analytical predictions for any finite field $\mathbb{F}_q$ of order $q\geq 2$. Fig.~\ref{fig.study_both} considers the simple case of RLNC transmission over a broadcast erasure channel. If the transmission of $n_\mathrm{T}$ packets is modeled as a sequence of $n_\mathrm{T}$ Bernoulli trials whereby $\varepsilon$ signifies the probability that a transmitted packet will be erased, the probability that a receiving node shall recover \textit{at least} $x$ of the $k$ source packets can be expressed as \begin{equation} \label{eq.prob_BEC} P(\lvert X\rvert\!\geq\! x)\!=\!\displaystyle\sum_{n=x}^{n_\mathrm{T}}\!\displaystyle\binom{n_\mathrm{T}}{n}\!\left(1-\varepsilon\right)^{n}\varepsilon^{n_\mathrm{T}-n}\,P(\lvert X\rvert\!\geq\! x\,\vert\, N\!=\!n). \end{equation} The probability $P(\lvert X\rvert\geq x\,\vert\, N=n)$ is equal to \eqref{eq.prob_ns} for non-systematic RLNC {and \eqref{eq.prob_s} or \eqref{eq.prob_s_less_than_k}}, depending on the value of $n_\mathrm{T}$, for systematic RLNC. Fig.~\ref{fig.study_both}(a) focuses on non-systematic RLNC and depicts $P(\lvert X\rvert\geq x)$ in terms of $n_\mathrm{T}$ for $x\in\{2,4,10,16,20\}$ when $k\!=\!20$, and for $x\in\{3,6,15,24,30\}$ when $k\!=\!30$. Results have been obtained for $q\in\{2,8\}$ and $\varepsilon\!=\!0.2$. For $q\!=\!2$, the transmission of only a few additional coded packets can increase the fraction of the recovered message from at least $x/k=0.1$ to $x/k=1$. However, for $q$ as low as $8$, the range of $n_\mathrm{T}$ values for which a receiving node will proceed from recovering a small portion of the transmitted message to recovering the whole message gets very narrow. Furthermore, for $q\!=\!2$, segmentation of the message into $k\!=\!20$ source packets permits a receiving node to recover the same fraction ($x/k$) of the message with a higher probability than dividing the same message into $k\!=\!30$ source packets. Systematic RLNC is considered in Fig.~\ref{fig.study_both}(b). Besides the reduced decoding complexity \cite{Lucani12}, we observe that systematic RLNC enables a receiving node to gradually reveal an increasingly larger portion of the message as more packets are transmitted. However, a large number of source packets or a high order finite field impairs the progressive recovery of the message for $n_\mathrm{T}\!>\!k$. This is because source packets are transmitted for $n_\mathrm{T}\leq k$ but coded packets are sent for $n_\mathrm{T}\!>\!k$; the decoding behaviour of a receiving node changes at $n_\mathrm{T}\!=\!k$ and causes a change in the slope of $P(\lvert X\rvert\geq x)$ for $x/k\!=\!0.8$. The results show that, if information-theoretic security is required, non-systematic RLNC over finite fields of size 8 or larger can be used to segment each message into a large number of source packets. The number of transmitted packets can then be adjusted to the channel conditions to achieve a balance between the probability of legitimate nodes reconstructing the message and the probability of eavesdroppers being unable to decode even a portion of the message. If the objective of the system is to maximize the number of nodes that will recover at least a large part of a message, systematic RLNC over small finite fields can be used to divide data into source packets. If the receiving nodes do not suffer from limited computational capabilities, the size of the finite field can be increased to improve the probability of recovering the entire message. \begin{figure}[t] \captionsetup[subfigure]{labelformat=empty} \begin{subfigure}{0.492\columnwidth} \centering \includegraphics[height=2.4cm]{clari2_left_top.eps} \vspace{-3mm} \caption{\scriptsize $k=20$, $q=2$} \end{subfigure} \begin{subfigure}{0.492\columnwidth} \centering \includegraphics[height=2.4cm]{clari2_right_top.eps} \vspace{-3mm} \caption{\scriptsize $k=20$, $q=2$} \end{subfigure} \\ \begin{subfigure}{0.492\columnwidth} \vspace{1mm} \centering \includegraphics[height=2.4cm]{clari2_left_middle.eps} \vspace{-3mm} \caption{\scriptsize $k=30$, $q=2$} \end{subfigure} \begin{subfigure}{0.492\columnwidth} \vspace{1mm} \centering \includegraphics[height=2.4cm]{clari2_right_middle.eps} \vspace{-3mm} \caption{\scriptsize $k=30$, $q=2$} \end{subfigure} \\ \begin{subfigure}{0.492\columnwidth} \vspace{1mm} \centering \includegraphics[height=2.4cm]{clari2_left_bottom.eps} \vspace{-3mm} \caption{\scriptsize $k=30$, $q=8$} \end{subfigure} \begin{subfigure}{0.492\columnwidth} \vspace{1mm} \centering \includegraphics[height=2.4cm]{clari2_right_bottom.eps} \vspace{-3mm} \caption{\scriptsize $k=30$, $q=8$} \end{subfigure} \\ \begin{subfigure}{0.492\columnwidth} \vspace{2mm} \centering \caption{\scriptsize (a) Non-systematic RLNC} \end{subfigure} \begin{subfigure}{0.492\columnwidth} \vspace{2mm} \centering \caption{\scriptsize (b) Systematic RLNC} \end{subfigure} \vspace{-3mm} \caption{Depiction of the probability of recovering at least $x$ source packets when $n_\mathrm{T}$ packets have been transmitted over a packet erasure channel with $\varepsilon\!=\!0.2$ using (a) non-systematic RLNC and (b) systematic RLNC.} \label{fig.study_both} \vspace{-4mm} \end{figure} \section{Conclusions} \label{sec.conclusions} This paper derived exact expressions for the probability of decoding a fraction of a source message upon reception of an arbitrary number of network-coded packets. Results unveiled the potential of non-systematic network coding in offering weak information-theoretic security, even when operations are over small finite fields. On the other hand, systematic network coding allows for the progressive recovery of the source message as the number of received packets increases, especially when the size of the finite field is small. \section{Acknowledgments} \label{sec.ack} Jessica Claridge has been supported by an EPSRC PhD studentship. Both authors appreciate the support of the COST Action IC1104 and thank Simon R. Blackburn for his advice. \bibliographystyle{IEEEtran}
7ab653d1273e1535ea384433140b5fb7976c2689
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{Fsec.1} In this paper we will study the probability of the hitting time to a moving boundary. To state the main result, we need some notations on a Brownian motion. Let us call $P_{r,s}, r\in\mathbb{R}, s\geq 0$, the law on $\displaystyle C([s,\infty))$ of the Brownian motion $B_t$, $t\geq s$, which starts from $r$ at time $s$, i.e. $B_s=r$. For each $t>s$ the law of $B_t$ is absolutely continuous with respect to the Lebesgue measure and has a density $G_{s,t}(r,\cdot)$ which is the Gaussian $\displaystyle G(\cdot,t;r,s)=\frac{1}{\sqrt{2\pi (t-s)}}\exp{\left(-\frac{(\cdot-r)^{2}}{2(t-s)}\right)}$. We denote by $E_{r,s}$ the expectation under $P_{r,s}$. For a given curve $\displaystyle X=\{t\rightarrow X_t\}$, $s\geq 0$ and $r<X_s$, we define \begin{eqnarray} \displaystyle \tau_{r,s}^{X}=\inf\{t \geq s: B_t\geq X_t \},\ and=\infty\ \textrm{if the set is empty}, \end{eqnarray} where $B_s=r$ and denote by $F_{r,s}^{X}(dt)$ the distribution of $\displaystyle \tau_{r,s}^{X}$ induced by $P_{r,s}$. For s=0, we use abbreviated forms $P_{r}$, $E_{r}$, $\displaystyle \tau_{r}^{X}$, $F_{r}^{X}(dx)$ instead of $P_{r,0}$, $E_{r,0}$, $\displaystyle \tau_{r,0}^{X}$, $F_{r,0}^{X}(dx)$ respectively whenever it is needed. In addition, for $t>0$, let us call $d\mu_{r_0}(\cdot,t)$ the positive measure on $(-\infty,X_t)$ such that \begin{eqnarray} \label{1.0} \displaystyle \int_{(-\infty,X_t)} d\mu_{r_0}(x,t)f(x)=E_{r_0}[f(B_t);\tau_{r_0}^{X}> t] \end{eqnarray} for all $\displaystyle f\in C_c^{\infty}(\mathbb{R})$ with $\displaystyle\textrm{supp}\hspace{0.5mm}f\Subset (-\infty,X_t)$. The main result in the paper is; \begin{thm} \label{mainthm} If $X$ is locally H\"{o}lder continuous in $[0,\infty)$ with exponent $\displaystyle\gamma\in\left(1/2,1\right]$ and $r_0<X_0$, then \begin{enumerate} \item\label{1.1} $d\mu_{r_0}(x,t)=G_{0,t}^{X}(r_0,x)dx$ where for all $x<X_t$, \begin{eqnarray} \label{12345} \displaystyle G_{0,t}^{X}(r_0,x)=G_{0,t}(r_0,x)-\int_{[0,t)}F_{r_0}^{X}(ds)G_{s,t}(X_{s},x). \end{eqnarray} \item\label{1.2} $\displaystyle F_{r_0}^{X}(ds)$ has a density function $p$ on $[0,\infty)$, namely $F_{r_0}^{X}(ds)=p(s)ds$.\\ \item\label{1.3} $\displaystyle p(t)=-\frac{1}{2}\frac{\partial}{\partial x}G_{0,t}^{X}(r_0,x)\Big|_{x=X_t^{-}}$ for all\ $t>0$.\\ \item\label{1.4} $\displaystyle G_{0,t}^{X}(r_0,x)$ solves \begin{eqnarray} \label{1}\displaystyle v_t=\frac{1}{2}v_{xx},\ \ -\infty<x<X_t,\ t>0, \\ \label{2}\displaystyle \lim_{(x,t)\rightarrow (X_s,s)}v(x,t)=0,\ s>0,\\ \label{3} \displaystyle \lim_{(x,t)\rightarrow (y,0)}v(x,t)=\delta_{r_0}(y),\ -\infty<y<X_0. \end{eqnarray} \end{enumerate} \end{thm} \textbf{Remarks} \vskip 0.1cm Item $\ref{1.4}$ of $\textbf{Theorem~\ref{mainthm}}$ states that for any $r_0<X_0$ the function $\displaystyle G_{0,\cdot}^{X}(r_0,\cdot)$ given by $(\ref{12345})$ is the Green function of the heat equation with Dirichlet boundary conditions at the moving boundary $X$. In $\cite{AP}$, when a moving boundary is infinitely differentiable, it is showed that the space derivative of the Green function of the heat equation at the boundary is proportional to the hitting time density function. Likewise by items $\ref{1.2}$ and $\ref{1.3}$ of $\textbf{Theorem~\ref{mainthm}}$, the space derivative of $\displaystyle G_{0,\cdot}^{X}(r_0,\cdot)$ at the moving boundary is proportional to the hitting time density function $p$.\\ For the case when $X_t=a+bt$, it is well known(see for instance $\cite{KS}$) that for $r<a$, $\displaystyle\tau_r^{X}$ has a probability density function given by $\displaystyle f(t)=\frac{a-r}{\sqrt{2\pi t^3}}\exp{\left(-\frac{(a+bt-r)^2}{2t}\right)}\mathbf{1}_{t>0}$. In addition, there is another result in $\cite{RSS}$ when $X_t=a+bt^{\frac{1}{p}}$ for $p\geq 2$, $r<a$, then $\displaystyle\tau_r^{X}$ has a probability density function. \\ In $\cite{PS}$, it is proved that for any continuous curve $X_t$ and $r<X_0$, there is a distribution $\displaystyle F_r^{X}$ of $\displaystyle \tau_{r}^{X}$ which satisfies the following integral equation(called the Master Equation): \begin{eqnarray} \label{0000} \displaystyle\Psi\left(\frac{z-r}{\sqrt{t}}\right)=\int_{0}^{t}\Psi\left(\frac{z-X_s}{\sqrt{t-s}}\right)F_r^{X}(ds), \end{eqnarray} where $z\geq X_t$, $t>0$ and $\displaystyle\Psi(z)=\int_z^{\infty}\frac{1}{\sqrt{2\pi}}\exp{\left(-\frac{x^{2}}{2}\right)}dx$. This can be proved intuitively as follows; the left hand side of $(\ref{0000})$ is the probability that a Brownian motion starts at $r$ at time 0 and reaches $z$ greater or equal to $X_t$ at time $t$. Then it should hit the boundary at least once which implies the right hand side of of $(\ref{0000})$. Moreover, it is showed that if $X_t$ is $\displaystyle C^{1}$, then there exists a continuous density function $f$ of $\displaystyle F_r^{X}$. There is an extension, in $\cite{RSS}$, to curves $X$ which are differentiable with $\displaystyle \left|\frac{dX_t}{dt}\right|\leq Ct^{-\alpha}$ for some constant $C>0$ and $\alpha<1/2$, then $F_r^{X}$ has a probability density function. \\ By $(\ref{0000})$, for any continuous curve $X_t$ and $r<X_0$, we have \begin{eqnarray} \label{2.3} \displaystyle\Psi\left(\frac{X_t-r}{\sqrt{t}}\right)=\int_{0}^{t}\Psi\left(\frac{X_t-X_s}{\sqrt{t-s}}\right)F_r^{X}(ds). \end{eqnarray} which can be regarded as an integral equation for $\displaystyle F_r^{X}(ds)$. In $\cite{TM}$, it is studied the equation $(\ref{2.3})$ when $X$ is H\"{o}lder continuous with exponent greater than $1/2$, it is proved that there exists a unique continuous function $q$ such that \begin{eqnarray} \displaystyle \Psi\left(\frac{X_t-r}{\sqrt{t}}\right)=\int_{0}^{t}\Psi\left(\frac{X_t-X_s}{\sqrt{t-s}}\right)q(s)ds. \end{eqnarray} To conclude that $F_r^{X}(ds)=q(s)ds$, one still needs that $\displaystyle F_r^{X}(ds)$ is absolutely continuous with respect to Lebesgue measure. The analysis in $\cite{TM}$ as well as the proof of $\textbf{Theorem~\ref{mainthm}}$ uses extensively the work by Cannon $\cite{Cannon}$ on the heat equation with the moving boundary. \vskip1cm \setcounter{equation}{0} \section{Outline of the proof of \textbf{Theorem~\ref{mainthm}}} \label{Fsec.2} Let $X$ be a continuous curve defined on $[0,\infty)$ and let $r_0<X_0$. Using the strong Markov property, we have, for $\displaystyle f\in C_c^{\infty}(\mathbb{R})$ with $\displaystyle\textrm{supp}\hspace{0.5mm}f\Subset (-\infty,X_t)$ and $0\leq s\leq t$, \begin{eqnarray} \displaystyle E_{r_0}[f(B_t)|\tau_{r_0}^{X}= s]=E_{X_{s},s}[f(B_{t})]. \end{eqnarray} Thus we get \begin{eqnarray} \displaystyle E_{r_0}[f(B_t);\tau_{r_0}^{X}\leq t]=\int_{[0,t]}F_{r_0}^{X}(ds)E_{X_{s},s}[f(B_{t})]. \end{eqnarray} It follows that item $\ref{1.1}$ of $\textbf{Theorem~\ref{mainthm}}$ holds with $\displaystyle G_{0,t}^{X}(r_0,x)$ given by $(\ref{12345})$.\\ To show the other items of $\textbf{Theorem~\ref{mainthm}}$, we define $D$ as $$\displaystyle D:=\{(x,t):-\infty<x<X_t,\ t>0\}$$ and \begin{eqnarray} \label{2345} \displaystyle u(x,t):=\int_{-\infty}^{0}h(\xi)G_{0,t}^{X}(\xi,x)d\xi \end{eqnarray} for given $\displaystyle h\in C_c^{\infty}((-\infty,X_0);\mathbb{R}_{+})$ and all $(x,t)\in D$.\\ If $X$ is locally H\"{o}lder continuous in $[0,\infty)$ with exponent $\gamma\in[1/2,1]$, then $u$ solves the heat equation in $D$ with the initial condition $h$ and Dirichlet boundary condition.(See $\textbf{Theorem~\ref{thm1}}$.) However, in general, $u$ is not a unique solution of the heat equation with the initial condition $h$ and Dirichlet boundary condition.(See $\textbf{Remark~\ref{rem1} and \ref{rem2}} $ below)\\ To have the uniqueness, first of all, we restrict time variable of $D$ in a finite interval. Thus we fix $T>0$ and define the parabolic cylinder $D_T$ which is a subset of $D$ as $$\displaystyle D_T:=\{(x,t):-\infty<x<X_t,\ 0<t\leq T\}.$$ Consider the following initial-boundary value problem \begin{equation} \left\{ \, \begin{IEEEeqnarraybox}[][c]{l?s} \IEEEstrut \displaystyle v\in C(\overline{D_T})\cap C^{2,1}(D_T), \\ \displaystyle v_t=\frac{1}{2}v_{xx},\ \ (x,t)\in D_T, \\ \displaystyle v(X_t,t)=0,\ 0 < t \leq T,\\ \displaystyle v(x,0)=h(x),\ -\infty<x<X_0,\\ \displaystyle \lim_{x\rightarrow -\infty}\sup_{0< t< T}|v(x,t)|=0. \IEEEstrut \end{IEEEeqnarraybox} \right. \label{4242} \end{equation} We prove the following weaker form of $\textbf{Theorem~\ref{mainthm}}$ in $\textbf{Section~\ref{Fsec.2}}$. \begin{thm} \label{thm1} Let $X$ be a locally H\"{o}lder continuous curve in $[0,\infty)$ with exponent $\gamma\in[1/2,1]$ and let $X_0=0$. \begin{enumerate} \item\label{thm1.1}The function $u$ defined in $(\ref{2345})$ is the unique solution of $(\ref{4242})$.\\ \item\label{thm1.2}If $\gamma\in(1/2,1]$, then $u$ has the left hand derivative at the boundary $u_x({X_t^{-}},t)$ which is continuous on $(0,\infty)$. Moreover, $\displaystyle p_{h}(t):=-\frac{1}{2}{u_x}(X_{t}^{-},t)$ satisfies \begin{eqnarray} \label{888} \displaystyle p_{h}(t)=-\int_{-\infty}^{0}h(\xi)G_x(X_t,t;\xi,0)d\xi+\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)p_{h}(\tau)d\tau\ for\ all\ t>0. \end{eqnarray} \end{enumerate} \end{thm} \begin{rem} \label{rem1} The uniqueness for $(\ref{4242})$ is not guaranteed if we do not assume $\displaystyle v\in C(\overline{D_T})$. When $X_t=0$ for all $t$ and $h$ is identically $0$, if $v(x,t)$ is given by $\displaystyle \frac{1}{\sqrt{2\pi t}}\left\{-\frac{x}{t}\right\}\exp\left(-\frac{x^{2}}{2t}\right)$, then this satisfies the heat equation with the initial data $0$ and is also $0$ on the boundary, but this is not continuous at $(0,0)$. \end{rem} \begin{rem} \label{rem2} There is another nontrivial solution when $X_t=0$ for all $t\geq 0$, the initial data and the boundary condition are all $0$, then the function $$\displaystyle v(x,t)=\sum_{n=0}^{\infty}f^{(n)}\left(\frac{t}{2}\right)\frac{x^{2n+1}}{(2n+1)!},$$ where $$ \displaystyle f(t)= \begin{cases} \displaystyle \exp{\left(-\frac{1}{t^{2}}\right)},\ \ t>0, \\ 0,\hspace{7.5mm}t\leq0. \end{cases} $$ satisfies the heat equation with the initial data $0$ and is also $0$ on the boundary. Furthermore, $\displaystyle v\in C(\overline{D_T})\cap C^{2,1}(D_T)$. Thus we need the condition $\displaystyle \lim_{x\rightarrow -\infty}\sup_{0< t< T}|v(x,t)|=0$ to have the uniqueness. \end{rem} \begin{rem} \label{rem.3} It can be shown $\displaystyle\lim_{t\rightarrow 0}p_h(t)=0$ by $(\ref{888})$ and $\textbf{Lemma}~\ref{lem1}$ so that $u_x({X_t^{-}},t)$ is continuous on $[0,\infty)$. \end{rem} By approximating the initial delta measure of $\textbf{Theorem~\ref{mainthm}}$, in $\textbf{Section~\ref{Fsec.3}}$, we prove the following proposition. \begin{prop} \label{890} Let $X$ be a locally H\"{o}lder continuous curve in $[0,\infty)$ with exponent $\gamma\in(1/2,1]$ and let $X_0=0$. We fix $r_0<X_0=0$ and choose a sequence $\displaystyle\{h_n\}\subset C_c^{\infty}(\mathbb{R};\mathbb{R}_{+})$ with $\textrm{supp}\hspace{0.1cm}{h_n}=[r_0-\frac{1}{n},r_0+\frac{1}{n}]\Subset (-\infty,0)$ and $\displaystyle \lVert h_n \rVert_{1}=1$. For each $h_n$, there exist a corresponding solution $\displaystyle u_{n}$ with $\displaystyle -\frac{1}{2}\frac{\partial u_n}{\partial x}\Big|_{(X_{t}^{-},t)}=:p_{n}(t)$ in the sense of $\textbf{Theorem~\ref{thm1}}$. Then we have the following statements: \begin{enumerate} \label{100} \item There is a unique $p\in C([0,\infty))$ with $\displaystyle p(0)=0$ such that \begin{eqnarray} \label{111} \displaystyle p(t)=-G_x(X_t,t;r_0,0)+\int_0^{t}G_x(X_t,t;X_\tau,\tau)p(\tau)d\tau\ for\ all\ t>0. \end{eqnarray} \label{222} \item $p_n$ converges to $p$ in $\displaystyle C_{(\eta)}^{0}((0,T])$ for all $0<\eta<1/2$. \end{enumerate} \end{prop} \vskip 0.3cm Finally, we show that $p$ is the density function $\textbf{Section~\ref{Fsec.3}}$ and conclude $\textbf{Theorem~\ref{mainthm}}$ in $\textbf{Section~\ref{Fsec.4}}$. \vskip1cm \section{Proof of \textbf{of Theorem~\ref{thm1}}} \label{Fsec.2} Following $\cite{Cannon}$, we introduce $\displaystyle C_{(\nu)}^{0}\left((0,T]\right)$, $0<\nu \leq 1$, as the subspace of $C((0,T])$ that consists of those functions $\varphi$ such that $$\displaystyle \lVert \varphi\rVert_{T}^{(\nu)}=\sup_{0<t\leq T}t^{1-\nu}|\varphi(t)|<\infty.$$ Then $\displaystyle C_{(\nu)}^{0}\left((0,T]\right)$ is a Banach space under the norm $\displaystyle \lVert \cdot \rVert_{T}^{(\nu)}$. Moreover, we also introduce the following lemmas from $\cite{Cannon}$ which plays an essential role in our analysis. \begin{lem}[\emph{jump relation}] For $\displaystyle\varphi\in C_{(\nu)}^{0}\left((0,T]\right)$, we have \begin{eqnarray} \lim_{x\rightarrow X_t^{\pm}}\frac{\partial w_\varphi}{\partial x}(x,t)=\mp\varphi(t)+\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)\varphi(\tau)d\tau, \end{eqnarray} where $\displaystyle w_\varphi(x,t)=\int_{0}^{t}G(x,t;X_\tau,\tau)\varphi(\tau)d\tau$. \end{lem} For two continuous curves $s_1$, $s_2$ such that $s_1(t)<s_2(t), t\in[0,T]$, let us set $\displaystyle E_T:=\{(x,t): s_{1}(t)<x<s_{2}(t),\ 0<t\leq T\}$ and $\displaystyle B_T:=\{(s_{i}(t),t): 0\leq t\leq T,\ i\in\{1,2\}\}\cup\{(x,0): s_{1}(0)< x< s_{2}(0)\}$. \begin{lem}[\emph{The Weak Maximum(Minimum) Principle}] For a solution $u$ of $\displaystyle u_t=\frac{1}{2}u_{xx}$ in $E_T$, which is continuous in $E_T\cup B_T$, \begin{eqnarray} \displaystyle \max_{E_T\cup B_T}u= \max_{B_T}u.\ \ \left(\min_{E_T\cup B_T}u= \min_{B_T}u.\right) \end{eqnarray} \end{lem} Before going to the proof of $\textbf{Theorem~\ref{thm1}}$, we need the following proposition. \begin{prop} \label{3456} Let $X$ be a locally H\"{o}lder continuous curve in $[0,\infty)$ with exponent $\gamma\in[1/2,1]$. If the starting point of the Brownian motion is close to $X$, the first hitting time converges to 0. Precisely, $\displaystyle\lim_{\xi\rightarrow X_{0}}P_{\xi}\left[\tau_{\xi}^{X}> s\right]=0$ for all $s>0$. \end{prop} \begin{proof} Without loss of generality, we may reduce this problem as the case for Brownian motion starting at $0$ and $X_0=\epsilon>0$ and let $\epsilon$ go to $0$. For $s>0$, let $\displaystyle m:=\sup_{0\leq t_1<t_2\leq s}\frac{|X_{t_2}-X_{t_1}|}{|t_2-t_1|^{\gamma}}$. Since $\displaystyle\limsup_{t\downarrow 0}\frac{B_t}{\sqrt{t}}=\infty$ a.e., for $M>m$, we have a sequence ${t_k}\downarrow0$ such that $M\sqrt{t_k}\leq B_{t_k}$ a.e. and $M\sqrt{t_k}-mt_k^{\gamma}\downarrow 0$ for all $k$. Thus, for $0<t\leq s$, we deduce that a.e. $$\displaystyle \sup_{0\leq l\leq t}{\left\{B_l-X_l\right\}}\geq \sup_{0\leq l\leq t}{\left\{B_l-(\epsilon+ml^{\gamma})\right\}} \geq\sup_{t_k\leq t}{\left\{B_{t_k}-(\epsilon+mt_k^{\gamma})\right\}}\geq\sup_{t_k\leq t}{\left\{M\sqrt{t_k}-(\epsilon+mt_k^{\gamma})\right\}}.$$ Therefore, $$\displaystyle P_{0}\left(\sup_{0\leq l\leq t}{\left\{B_l-X_l\right\}}<0\right)\leq P_{0}\left(\sup_{t_k\leq t}{\left\{M\sqrt{t_k}-mt_k^{\gamma}\right\}<\epsilon}\right).$$ For each sufficently small $\epsilon>0$, there is a greatest $k(\epsilon)$ such that $\displaystyle M\sqrt{t_{k(\epsilon)}}-mt_{k(\epsilon)}^{\gamma}\geq\epsilon$. Then $k(\epsilon)$ is an increasing function as $\epsilon$ decreases and $\displaystyle \lim_{\epsilon\rightarrow 0}k(\epsilon)=\infty$, so we obtain that $\tau_0^{X}\leq t_{k(\epsilon)}\leq s$ a.e. for all sufficiently small $\epsilon>0$. \end{proof} \begin{proof}\hspace{-1mm} \textbf{of Theorem~\ref{thm1}}\\ Let us define $X_{\tau}^{\prime}:=X_{t-\tau}$ for all $0\leq\tau\leq t$. Using the invariance of the law of the Brownian motion under time reversal, we have \begin{eqnarray} \displaystyle u(x,t)=\int_{-\infty}^{0} h(r^{\prime})G_{0,t}^{X}(r^{\prime},x)dr^{\prime}=E_{x}[h(B_t);\tau_x^{X^{\prime}}> t]. \end{eqnarray} Using this equality, we also have \begin{eqnarray} \displaystyle\left|u(x,t)\right|=\left|E_{x}[h(B_t);\tau_x^{X^{\prime}}> t]\right|\leq \lVert h\rVert_{\infty}P_{x}[\tau_x^{X^{\prime}}> t]. \end{eqnarray} For $s>0$, let us choose $\displaystyle 0<s^{*}<s$. Then for all $(x,t)$ sufficiently close to $(X_s,s)$, we obtain \begin{eqnarray} P_{x}[\tau_x^{X^{\prime}}> t]\leq P_{x}[\tau_x^{X^{\prime}}> s^{*}] \end{eqnarray} which vanishes when $\displaystyle (x,t)\rightarrow (X_{s},s)$ by $\textbf{Proposition~\ref{3456}}$ so that $\displaystyle\lim_{(x,t)\rightarrow(X_s,s)}u(x,t)=0$.\\ In addition, we have \begin{eqnarray} \displaystyle\left|u(x,t)\right|=\left|E_{x}[h(B_t);\tau_x^{X^{\prime}}> t]\right| \leq E_{x}[ | h(B_t)|]=\int_{-\infty}^{0} |h(\xi)|G_{0,t}(x,\xi)d\xi \end{eqnarray} which also vanishes when $\displaystyle (x,t)\rightarrow (0,0)$, since the support of $h$ is strictly away from $0$. To prove that $u$ satisfies the initial data $h$, we write $\displaystyle y_t=\min_{s\in[0,t]}X_s^{\prime}$. For $x<0$ and any positive $\lambda>0$, \begin{eqnarray} \label{3.13} P_{x}\left[\tau_x^{X^{\prime}}\leq t\right]\leq P_{x}\left[\max_{s\in[0,t]}B_s\geq y_t\right]= P_{x}\left[\max_{s\in[0,t]}\exp{(\lambda B_s)}\geq \exp{(\lambda y_t)}\right]. \end{eqnarray} Since the exponential of Brownian motion is a positive submartingale, we can apply Doob's inequality, then \begin{eqnarray} \label{3.14} P_{x}\left[\max_{s\in[0,t]}\exp{(\lambda B_s)}\geq \exp{(\lambda y_t)}\right]\leq \frac{E_{x}[\exp{(\lambda B_t)}]}{\exp{(\lambda y_t)}}=\exp{\left(\frac{1}{2}\lambda^{2}t-\lambda(y_t-x) \right)}. \end{eqnarray} From $(\ref{3.13})$ and $(\ref{3.14})$, we get \begin{eqnarray} \displaystyle\lim_{(x,t)\rightarrow (y,0)}P_{x}\left[\tau_x^{X^{\prime}}\leq t\right] \leq \exp{(-\lambda(X_0-y))}= \exp{(\lambda y)} \end{eqnarray} so that the left hand side vanishes since $\lambda$ is arbitrary. Thus we obtain that \begin{eqnarray} \displaystyle \lim_{(x,t)\rightarrow (y,0)}\int_{-\infty}^{0}h(\xi)G_{0,t}^{X}(\xi,x)d\xi=\lim_{(x,t)\rightarrow (y,0)}E_{x}[h(B_t)]=h(y). \end{eqnarray} By the properties of the Gaussian kernel, we deduce that $u$ solves $(\ref{4242})$. We now prove uniqueness. If $v_1$, $v_2$ satisfy all the above conditions, then $\displaystyle v_1-v_2\in C(\overline{D_T})\cap C^{2,1}(D_T)$ satisfies the heat equation with the initial data $0$ and is also $0$ on the boundary. Moreover, $\displaystyle \lim_{x\rightarrow -\infty}\sup_{0< t< T}|v_1(x,t)-v_2(x,t)|=0$ so that by the weak maximum(minimum) principle, $v_1-v_2$ is all $0$ in $D_T$. Therefore, $v_1=v_2$ so that item $\ref{thm1.1}$ is proved.\\ To prove item $\ref{thm1.2}$, assuming that $\gamma\in(1/2,1]$, we define, $(x,t)\in D_T$, \begin{eqnarray} \label{435} \displaystyle v(x,t):=\int_{-\infty}^{0}h(\xi)G(x,t;\xi,0)d\xi+\int_{0}^{t}G(x,t;X_\tau,\tau)\varphi(\tau)d\tau, \end{eqnarray} where $\displaystyle \varphi\in C_{(\gamma)}^{0}\left((0,T]\right)$ satisfies \begin{eqnarray} \label{11} 0=\int_{-\infty}^{0}h(\xi)G(X_t,t;\xi,0)d\xi+\int_{0}^{t}G(X_t,t;X_\tau,\tau)\varphi(\tau)d\tau. \end{eqnarray} If we apply the Abel inverse operator $A$ defined by $$\displaystyle (AF)(t)=\frac{1}{\pi}\frac{d}{dt}\int_{0}^{t}\frac{F(\eta)}{(t-\eta)^{\frac{1}{2}}}d\eta$$ on both sides of (\ref{11}), then from Chapter 14 in $\cite{Cannon}$ we have a equivalent Volterra integral equation of the second kind: \begin{eqnarray} \label{434} \varphi(t)=\psi(t) + \int_{0}^{t}H(t,\tau)\varphi(\tau)d\tau, \end{eqnarray} where $\displaystyle \psi\in C_{(\gamma)}^{0}\left((0,T]\right)$ and $|H(t,\tau)|\leq C(t-\tau)^{2\gamma-2}$. The existence of $\displaystyle \varphi\in C_{(\gamma)}^{0}\left((0,T]\right)$ which satisfies $(\ref{434})$ can be proved similarly to the proof of $\textbf{Proposition~\ref{100}}$ that we will show later. Then $v$ is well-defined and solves $(\ref{4242})$ so that $v=u$ since $u$ is the unique solution of $(\ref{4242})$. By the jump relation, we have \begin{eqnarray} \displaystyle u_x(X_t^{-},t)=\int_{-\infty}^{0}h(\xi)G_x(X_t,t;\xi,0)d\xi+\varphi(t)+\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)\varphi(\tau)d\tau \end{eqnarray} so that $\displaystyle u_x(X_t^{-},t)\in C_{(\gamma)}^{0}((0,T])$. Since $T$ is arbitrary, we deduce $\displaystyle u_x(X_t^{-},t)$ is continuous on $(0,\infty)$.\\ To show $(\ref{888})$, let us fix $(x,t)\in D_T$ and let us define $\displaystyle D_{\epsilon,r}^{(t)}:=\{(\xi,\tau): r<\xi<X_\tau-\epsilon,\ \epsilon<\tau<t-\epsilon\}$ for each $\epsilon>0$ and $r\in\mathbb{R}$. By the Green's identity, we have \begin{eqnarray} \displaystyle \frac{1}{2}(u_\xi G-uG_\xi)_\xi-(uG)_{\tau}=0\ \Longrightarrow\ \ \oint_{\partial D_{\epsilon,r}^{(t)}} \frac{1}{2}(u_\xi G-uG_\xi)d\tau+(uG)d\xi=0. \end{eqnarray} It can be also shown that $\displaystyle \lim_{x\rightarrow -\infty}\sup_{0< t< T}|u_x(x,t)|=0$ by the properties of the Gaussian kernel. Hence we obtain another representation of $u$ by letting $\epsilon\rightarrow 0$, $r\rightarrow -\infty$, \begin{eqnarray} \label{777} \displaystyle u(x,t)=\int_{-\infty}^{0}h(\xi)G(x,t;\xi,0)d\xi+\frac{1}{2}\int_{0}^{t}G(x,t;X_\tau,\tau)u_x(X_\tau^{-},\tau)d\tau. \end{eqnarray} Differentiating both sides of $(\ref{777})$ with respect to x and applying the jump relation, we get \begin{eqnarray} \displaystyle \frac{1}{2}u_x(X_t^{-},t)=\int_{-\infty}^{0}h(\xi)G_x(X_t,t;\xi,0)d\xi+\frac{1}{2}\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)u_x(X_\tau^{-},\tau)d\tau. \end{eqnarray} which implies $(\ref{888})$. \end{proof} \vskip 1cm \section{Existence of a density function} \label{Fsec.3} From now on, $X$ is locally H\"{o}lder continuous in $[0,\infty)$ with exponent $\gamma\in(1/2,1]$ and $X_0=0$. Comparing the definition of $u$ and $(\ref{777})$, we see the following equality: \begin{eqnarray} \displaystyle \int_{[0,t)}G_{\tau,t}(X_\tau,x)\int_{-\infty}^{0}h(\xi)F_{\xi}^{X}(d\tau)d\xi=-\frac{1}{2}\int_{0}^{t}G(x,t;X_\tau,\tau)u_x(X_\tau^{-},\tau)d\tau. \end{eqnarray} Set $\displaystyle F_{h}^{X}(d\tau):=\int_{-\infty}^{0} h(\xi) F_{\xi}^{X}(d\tau)d\xi$. To conclude $\displaystyle F_{h}^{X}(d\tau)=-\frac{1}{2}u_x(X_\tau^{-},\tau)d\tau$, we introduce the mass lost $\displaystyle \Delta_{I}^{X}(u)$, $I=[t_1,t_2]\subset [0,T]$, $t_1\leq t_2$, is defined by \begin{eqnarray} \displaystyle \Delta_{I}^{X}(u)=\int_{-\infty}^{X_{t_1}}u(r,t_1)dr-\int_{-\infty}^{X_{t_2}}u(r,t_2)dr. \end{eqnarray} If we see the right hand side of $(\ref{777})$, we can extend $u$ to $\bar{u}$ defined in $\{(x,t): x\in\mathbb{R},\ 0<t\leq T\}$ as \begin{eqnarray} \displaystyle \bar{u}(x,t)=\int_{-\infty}^{0}h(\xi)G(x,t;\xi,0)d\xi+\frac{1}{2}\int_{0}^{t}G(x,t;X_\tau,\tau)u_x(X_\tau^{-},\tau)d\tau. \end{eqnarray} Then this satisfies the heat equation with $\displaystyle \lim_{(x,t)\rightarrow (y,0)}\bar{u}(x,t)=0$ for all $y\geq 0$ and also satisfies $\bar{u}(X_t,t)=0$ for all $0<t\leq T$. Moreover, by the properties of Gaussian kernel, we have \begin{eqnarray} \displaystyle \lim_{x\rightarrow \infty}\sup_{0< t< T}|\bar{u}(x,t)|=0. \end{eqnarray} It follows that $\bar{u}(x,t)=0$ in $\{(x,t): x\geq X_t, 0< t\leq T\}$ by the weak maximum(minimum) principle. Thus we assume that $u$ is defined $\{(x,t): x\in\mathbb{R},\ 0\leq t\leq T\}$ such that it is $0$ in $\{(x,t): x\geq X_t,\ 0\leq t\leq T\}$. \\ Heuristically \begin{eqnarray*} \displaystyle \Delta_{I}^{X}(u)= -\int \int_{t_1}^{t_2} u_t(x,t)dtdx= -\int \int_{t_1}^{t_2}\frac{1}{2}u_{xx}(x,t)dxdt= -\frac{1}{2}\int_{t_1}^{t_2} u_x(X_t^{-},t)dt. \end{eqnarray*} Since we do not control $u_{xx}$ at the moving boundary, we cannot make this argument rigorously. Thus we use a different approach. \begin{prop} $\displaystyle -\frac{1}{2}\int_{I}u_x(X_t^{-},t)dt=\Delta_{I}^{X}(u)=F_h^{X}(I).$ \end{prop} \begin{proof} If we integrate both sides of $(\ref{777})$, then \begin{eqnarray*} \displaystyle \int_{-\infty}^{\infty}u(x,t)dx= \int_{-\infty}^{\infty}\int_{-\infty}^{0}h(\xi)G(x,t;\xi,0)d\xi dx+ \int_{-\infty}^{\infty}\frac{1}{2}\int_{0}^{t}G(x,t;X_\tau,\tau)u_x(X_\tau^{-},\tau)d\tau dx. \end{eqnarray*} Applying Fubini's theorem, we get \begin{eqnarray} \displaystyle \int_{-\infty}^{X_t}u(x,t)dx=\int_{-\infty}^{\infty}u(x,t)dx= \int_{-\infty}^{0}h(\xi)d\xi+\frac{1}{2}\int_{0}^{t}u_x(X_\tau^{-},\tau)d\tau . \end{eqnarray} Thus we get the first equality of the proposition, \begin{eqnarray} \displaystyle \Delta_{[t_1,t_2]}^{X}(u)=-\frac{1}{2}\int_{t_1}^{t_2}u_x(X_\tau^{-},\tau)d\tau. \end{eqnarray} From $(\ref{1.0})$ and item $\ref{1.1}$ of $\textbf{Theorem~\ref{mainthm}}$, we get \begin{eqnarray} \displaystyle P_{\xi}[\tau_\xi^{X} > t]=\int_{-\infty}^{X_t}G_{0,t}^{X}(\xi,x)dx. \end{eqnarray} For $0=t_1<t_2$, using Fubini's theorem again, we get \begin{eqnarray*} \displaystyle \Delta_{I}^{X}(u)&=&\int_{-\infty}^{0}h(\xi)d\xi-\int_{-\infty}^{X_{t_2}}\int_{-\infty}^{0} h(\xi)G_{0,t_2}^{X}(\xi,x)d\xi dx\\ &=&\int_{-\infty}^{0}h(\xi)d\xi-\int_{-\infty}^{0}h(\xi)P_{\xi}[\tau_\xi^{X} > t_{2}]d\xi=\int_{-\infty}^{0}h(\xi)P_{\xi}[0\leq\tau_\xi^{X} \leq t_{2}]d\xi.\\ \end{eqnarray*} For $0<t_1<t_2$, similarly, \begin{eqnarray*} \displaystyle \Delta_{I}^{X}(u)&=&\int_{-\infty}^{X_{t_1}}\int_{-\infty}^{0} h(\xi)G_{0,t_1}^{X}(\xi,x)d\xi dx-\int_{-\infty}^{X_{t_2}}\int_{-\infty}^{0} h(\xi)G_{0,t_2}^{X}(\xi,x)d\xi dx\\ &=&\int_{-\infty}^{0}h(\xi)P_{\xi}[t_1< \tau_\xi^{X} \leq t_{2}]d\xi. \end{eqnarray*} Then for $I_\epsilon=[t_1-\epsilon, t_1]$, we get \begin{eqnarray*} \displaystyle \lim_{\epsilon\rightarrow 0}\Delta_{I_\epsilon}^{X}(u)&=&\lim_{\epsilon\rightarrow 0}-\frac{1}{2}\int_{I_\epsilon}u_x(X_t^{-},t)dt=0=\lim_{\epsilon\rightarrow 0}\int_{-\infty}^{0}h(\xi)P_{\xi}[t_1-\epsilon< \tau_\xi^{X} \leq t_1]d\xi\\ &=&\int_{-\infty}^{0}h(\xi)P_{\xi}[\tau_\xi^{X} = t_{1}]d\xi. \end{eqnarray*} Finally we conclude that \begin{eqnarray} \displaystyle \Delta_{I}^{X}(u)=\int_{-\infty}^{0}h(\xi)P_{\xi}[t_1\leq \tau_\xi^{X}\leq t_2]d\xi=F_h^{X}(I). \end{eqnarray} \end{proof} We use the following lemma extensively to prove $\textbf{Proposition~\ref{890}}$: \begin{lem} \label{lem1} $\displaystyle \int_0^{t}\tau^{\alpha_1}(t-\tau)^{\alpha_2}d\tau=\frac{\Gamma(1+\alpha_1)\Gamma(1+\alpha_2)}{\Gamma(2+\alpha_1+\alpha_2)}t^{1+\alpha_1+\alpha_2}$ for $\alpha_1, \alpha_2>-1$, where $\Gamma$ is the gamma function. \end{lem} \vskip 0.5cm \begin{proof}\hspace{-1mm} \textbf{of Proposition~\ref{890}}\\ Let $T_s>0$. Since $\displaystyle |G_x(X_t,t;X_\tau,\tau)|=\left|\frac{1}{\sqrt{2\pi (t-\tau)}}\left\{-\frac{X_{t}-X_{\tau}}{t-\tau}\right\}\exp{\left(-\frac{(X_t-X_\tau)^{2}}{2(t-\tau)}\right)}\right|\leq\frac{C_0}{(t-\tau)^{\frac{3}{2}-\gamma }}$, we deduce that for $\displaystyle q_1, q_2\in C([0,T_s])$, $0<t\leq T_s$, \begin{eqnarray*} \displaystyle \left|\int_0^{t}G_x(X_t,t;X_\tau,\tau)(q_1(\tau)-q_2(\tau))d\tau\right|\leq \int_0^{t}\frac{C_0\lVert q_1-q_2\rVert_{T_s}}{(t-\tau)^{\frac{3}{2}-\gamma}}d\tau\\ =C_1t^{\gamma-\frac{1}{2}}\lVert q_1-q_2\rVert_{T_s}\leq C_1T_s^{\gamma-\frac{1}{2}}\lVert q_1-q_2\rVert_{T_s}. \end{eqnarray*} We define $\displaystyle F:C([0,T_s])\rightarrow C([0,T_s])$ as, for $q\in C([0,T_s])$, \begin{eqnarray} \displaystyle (Fq)(t)=-G_x(X_t,t;r_0,0)+\int_0^{t}G_x(X_t,t;X_\tau,\tau)q(\tau)d\tau\ for\ t>0, \end{eqnarray} and $(Fq)(0)=0$. Since $\displaystyle\lim_{t\rightarrow 0}G_x(X_t,t;r_0,0)=0$, it is well-defined. If we choose $T_s$ such that $\displaystyle C_1T_s^{\gamma-\frac{1}{2}}<1$, then $F$ is a contraction mapping so that $F$ has a unique fixed point since $C([0,T_s])$ is a Banach space. Let's call this $\displaystyle p_{T_s}$.\\ Now we assume that we have $p_{T_s}$ for some $T_s>0$. For $\displaystyle T^{\star}>T_s$, we define $\displaystyle H:C([T_s,T^{\star}])\rightarrow C([T_s,T^{\star}])$ as, for $q\in C([T_s,T^{\star}])$, \begin{eqnarray*} \displaystyle (Hq)(t)=-G_x(X_t,t;r_0,0)+\int_0^{T_s}G_x(X_t,t;X_\tau,\tau)p_{T_s}(\tau)d\tau+ \int_{T_s}^{t}G_x(X_t,t;X_\tau,\tau)q(\tau)d\tau. \end{eqnarray*} Then for $\displaystyle q_1, q_2\in C([T_s,T^{\star}])$, we have \begin{eqnarray} \displaystyle \lVert Hq_1-Hq_2\rVert_{\infty}\leq C_2(t-T_s)^{\gamma-\frac{1}{2}}\lVert q_1-q_2\rVert_{\infty}\leq C_2(T^{\star}-T_s)^{\gamma-\frac{1}{2}}\lVert q_1-q_2\rVert_{\infty}. \end{eqnarray} Similarly, if we choose $\displaystyle T^{\star}$ such that $\displaystyle C_2(T^{\star}-T_s)^{\gamma-\frac{1}{2}}<1$, then $H$ is a contraction mapping so that $H$ has a unique fixed point since $\displaystyle C([T_s,T^{\star}])$ is a Banach space.\\ Therefore, if we have $p$ defined $[0,T_s]$, $p_{T_s}$, then we can extend this to time $T_s+C_3$ where $C_3$ is an independent constant. Thus if we repeat this step inductively, we have $p$ defined on $[0,\infty)$ which satisfies $(\ref{111})$. \\ We now prove that $p_n$ converges to $p$ in $\displaystyle C_{(\eta)}^{0}((0,T_s])$ for all sufficiently small $T_s>0$. By $(\ref{888})$, \begin{eqnarray} \label{999} \displaystyle p_n(t)=-\int_{-\infty}^{0}h_n(\xi)G_x(X_t,t;\xi,0)d\xi+\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)p_n(\tau)d\tau. \end{eqnarray} For $0<t\leq T_s$, taking the difference between $(\ref{111})$ and $(\ref{999})$, we get \begin{eqnarray*} t^{1-\eta}|p_n(t)-p(t)|&\leq& t^{1-\eta}\left|\int_{-\infty}^{0}h_n(\xi)[G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)]d\xi\right|\\ &+&t^{1-\eta}\left|\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)(p_n(\tau)-p(\tau))d\tau\right|. \end{eqnarray*} For the second term of the right hand side, we have \begin{eqnarray*} t^{1-\eta}\left|\int_{0}^{t}G_x(X_t,t;X_\tau,\tau)(p_n(\tau)-p(\tau))d\tau\right|\leq C_0t^{1-\eta}\int_{0}^{t}\frac{\lVert p_n-p\rVert_{T_s}^{(\eta)}}{(t-\tau)^{\frac{3}{2}-\gamma}\tau^{1-\eta}}d\tau\\ =C_1t^{\gamma-\frac{1}{2}}\lVert p_n-p\rVert_{T_s}^{(\eta)}\leq C_1T_s^{\gamma-\frac{1}{2}}\lVert p_n-p\rVert_{T_s}^{(\eta)}. \end{eqnarray*} Let us choose $T_s>0$ such that $C_1T_s^{\gamma-\frac{1}{2}}<1$. Then \begin{eqnarray*} \displaystyle (1-C_1T_s^{\gamma-\frac{1}{2}})\lVert p_n-p\rVert_{T_s}^{(\eta)}\leq t^{1-\eta}\left|\int_{-\infty}^{0}h_n(\xi)[G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)]d\xi\right|.\\ \displaystyle\leq\int_{-\infty}^{0}h_n(\xi)\sup_{\substack{0<t\leq T_s\\ \xi\in[r_0-\frac{1}{n},r_0+\frac{1}{n}]}}t^{1-\eta}\left|G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)\right|d\xi. \end{eqnarray*} For all sufficiently large $n$, $0<t\leq\frac{1}{n}$ and $\displaystyle r_0-\frac{1}{n}\leq\xi\leq r_0+\frac{1}{n}$, there exists $C_2$ such that \begin{eqnarray*} \left|G_x(X_t,t;\xi,0)\right|= \left|\frac{1}{\sqrt{2\pi}}\left\{-\frac{X_t-\xi}{t^{\frac{3}{2}}}\right\}\exp{\left(-\frac{(X_t-\xi)^{2}}{2t}\right)}\right|\leq C_{2}, \end{eqnarray*} Thus \begin{eqnarray*} \sup_{\substack{0<t\leq \frac{1}{n}\\ \xi\in[r_0-\frac{1}{n},r_0+\frac{1}{n}]}}t^{1-\eta}\left|G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)\right|\leq \sup_{0<t\leq \frac{1}{n}}2t^{1-\eta}C_{2}\leq \frac{2C_2}{n^{1-\eta}}. \end{eqnarray*} For all sufficiently large $n$, $\frac{1}{n}<t\leq T_s$ and $\displaystyle r_0-\frac{1}{n}\leq\xi\leq r_0+\frac{1}{n}$, since $\displaystyle|G_{xx}(\cdot,t;\cdot, 0)|\leq\frac{C_3}{t^{\frac{3}{2}}}$, we have \begin{eqnarray*} \left|G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)\right| \leq \frac{C_3|\xi-r_0|}{t^{\frac{3}{2}}}\leq\frac{C_3}{nt^{\frac{3}{2}}}, \end{eqnarray*} Thus \begin{eqnarray*} \sup_{\substack{\frac{1}{n}<t\leq T_s\\ \xi\in[r_0-\frac{1}{n},r_0+\frac{1}{n}]}}t^{1-\eta}\left|G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)\right|\leq \sup_{\frac{1}{n}<t\leq T_s}\frac{C_3}{nt^{\frac{1}{2}+\eta}}\leq\frac{C_3}{n^{\frac{1}{2}-\eta}}. \end{eqnarray*} Therefore, we conclude that $p_n$ converges to $p$ in $\displaystyle C_{(\eta)}^{0}((0,T_s])$ for all sufficiently small $T_s>0$.\\ To extend from $T_s$ to $T^{\star}$, assuming that $p_n$ converges to $p$ in $\displaystyle C_{(\eta)}^{0}((0,T_s])$ for some $T_s>0$ and writing $\displaystyle \lVert p_n-p\rVert_{[T_s,T^{\star}]}=\displaystyle\sup_{\tau\in[T_s,T^{\star}]} \left|p_n(\tau)-p(\tau)\right|$, for $\displaystyle T_s\leq t\leq T^{\star}$, we deduce that \begin{eqnarray*} |p_n(t)-p(t)|&\leq& \left|\int_{-\infty}^{0}h_n(\xi)[G_x(X_t,t;\xi,0)-G_x(X_t,t;r_0,0)]d\xi\right|\\ &+&\left|\int_{0}^{T_s}G_x(X_t,t;X_\tau,\tau)(p_n(\tau)-p(\tau))d\tau\right|+\left|\int_{T_s}^{t}G_x(X_t,t;X_\tau,\tau)(p_n(\tau)-p(\tau))d\tau\right|\\ &\leq& C_3\int_{-\infty}^{0}h_n(\xi)\frac{|\xi-r_0|}{t^{\frac{3}{2}}}d\xi+C_0\int_{0}^{T_s}\frac{\lVert p_n-p\rVert_{T_s}^{(\eta)}}{(t-\tau)^{\frac{3}{2}-\gamma}\tau^{1-\eta}}d\tau+C_0\int_{T_s}^{t}\frac{\lVert p_n-p\rVert_{[T_s,T^{\star}]}}{(t-\tau)^{\frac{3}{2}-\gamma}}d\tau\\ &\leq& \frac{C_3}{nt^{\frac{3}{2}}}+C_0\int_{0}^{T_s}\frac{\lVert p_n-p\rVert_{T_s}^{(\eta)}}{(T_s-\tau)^{\frac{3}{2}-\gamma}\tau^{1-\eta}}d\tau+C_4\lVert p_n-p\rVert_{[T_s,T^{\star}]}(t-T_s)^{\gamma-\frac{1}{2}}\\ &\leq&\frac{C_3}{nT_s^{\frac{3}{2}}} +C_1T_s^{\gamma-\frac{3}{2}+\eta}\lVert p_n-p\rVert_{T_s}^{(\eta)}+C_4\lVert p_n-p\rVert_{[T_s,T^{\star}]}(T^{\star}-T_s)^{\gamma-\frac{1}{2}}. \end{eqnarray*} Let us choose $T^{\star}>T_s$ such that $C_4(T^{\star}-T_s)^{\gamma-\frac{1}{2}}<1$, then we have \begin{eqnarray*} (1-C_4(T^{\star}-T_s)^{\gamma-\frac{1}{2}})\lVert p_n-p\rVert_{[T_s,T^{\star}]}\leq \frac{C_3}{nT_s^{\frac{3}{2}}}+C_1T_s^{\gamma-\frac{3}{2}+\eta}\lVert p_n-p\rVert_{T_s}^{(\eta)}. \end{eqnarray*} The right term vanishes when $n$ goes to $\infty$ so that $p_n$ converges to $p$ in $\displaystyle C_{(\eta)}^{0}((0,T_s+C_5])$ for some independent constant $C_5>0$. By repeating this argument inductively, it follows that $p_n$ converges to $p$ in $\displaystyle C_{(\eta)}^{0}((0,T])$. \end{proof} \vskip 0.5cm Hence we have \begin{eqnarray*} \displaystyle \lim_{n\rightarrow \infty}\int_{-\infty}^{0}h_{n}(\xi)P_{\xi}(\tau_\xi^{X}\in I)d\xi=\lim_{n\rightarrow \infty}\int_{I}p_n(t)dt=\int_{I}p(t)dt. \end{eqnarray*} For $I=[0,t]\subset[0,T]$, $\displaystyle P_{\xi}(\tau_\xi^{X}\in I)$ is an increasing function of $\xi$, so if we choose $h_n$ so that $\displaystyle\lim_{n\rightarrow\infty}\int_{r_0}^{r_0+\frac{1}{n}}h_n(\xi)d\xi=1$, we get \begin{eqnarray} \displaystyle \lim_{\xi\rightarrow r_0^{+}}P_{\xi}(\tau_\xi^{X}\in I)\leq \lim_{n\rightarrow \infty}\int_{r_0}^{r_0+\frac{1}{n}}h_{n}(\xi)P_{\xi}(\tau_\xi^{X}\in I)d\xi=\lim_{n\rightarrow \infty}\int_{-\infty}^{0}h_{n}(\xi)P_{\xi}(\tau_\xi^{X}\in I)d\xi. \end{eqnarray} Similarly, if we choose $h_n$ so that $\displaystyle\lim_{n\rightarrow\infty}\int_{r_0-\frac{1}{n}}^{r_0}h_n(\xi)d\xi=1$, we get \begin{eqnarray} \displaystyle \lim_{n\rightarrow \infty}\int_{-\infty}^{0}h_{n}(\xi)P_{\xi}(\tau_\xi^{X}\in I)d\xi=\lim_{n\rightarrow \infty}\int_{r_0-\frac{1}{n}}^{r_0}h_{n}(\xi)P_{\xi}(\tau_\xi^{X}\in I)d\xi\leq \lim_{\xi\rightarrow r_0^{-}}P_{\xi}(\tau_\xi^{X}\in I). \end{eqnarray} Therefore, we obtain that \begin{eqnarray} \displaystyle F_{r_0}^{X}([0,t])=P_{r_0}(\tau_{r_0}^{X}\leq t)=\int_{0}^{t}p(\tau)d\tau, \end{eqnarray} which implies that $p$ is the density function of $\displaystyle F_{r_0}^{X}$. \vskip 1cm \section{Proof of $\textbf{Theorem~\ref{mainthm}}$} \label{Fsec.4} Items $\ref{1.1}$ and $\ref{1.2}$ of $\textbf{Theorem~\ref{mainthm}}$ were proved in the previous sections. By $\textbf{Theorem~\ref{thm1}}$ and the properties of the Gaussian kernel, $\displaystyle G_{0,t}^{X}(r_0,x)$ solves $(\ref{1})$, $(\ref{2})$ and $(\ref{3})$. Hence $G^{X}$ is the Green function of the heat equation with Dirichlet boundary condition which implies item $\ref{1.4}$ of $\textbf{Theorem~\ref{mainthm}}$. Furthermore, $\displaystyle G_{0,t}^{X}(r_0,x)$ can be written as \begin{eqnarray} \displaystyle G_{0,t}^{X}(r_0,x)=G_{0,t}(r_0,x)-\int_{0}^{t}G_{\tau,t}(X_\tau,x)p(\tau)d\tau. \end{eqnarray} Differentiating with respect to x, applying the jump relation and $(\ref{111})$, we have \begin{eqnarray} \displaystyle \frac{\partial}{\partial x}G_{0,t}^{X}(r_0,x)\Big|_{x=X_t^{-}}&=&\frac{\partial}{\partial x}G_{0,t}(r_0,X_t)-p(t)-\int_{0}^{t} \frac{\partial}{\partial x}G_{\tau,t}(X_\tau,X_t)p(\tau)d\tau\\ &=&-2p(t). \end{eqnarray} Thus $\displaystyle p(t)=-\frac{1}{2}\frac{\partial}{\partial x}G_{0,t}^{X}(r_0,x)\Big|_{x=X_t^{-}}$, so it proves item $\ref{1.3}$ of $\textbf{Theorem~\ref{mainthm}}$. \vskip 1cm
9f30788dd903452d27e1d45966994dbec6ccc838
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Representations $\rho:\pi_1(M)\rar G$ of the fundamental group of of a manifold $M$ inside a Lie group $G$ naturally arise as monodromies of $(G,G/H)$-geometric structures \`a la Ehresmann \cite{ehresmann:infinitesimales} \cite{ehresmann:locales} on $M$ (see also \cite{goldman:geometric-structures} and \cite{goldman:locally-homogeneous-manifolds}). From a differential-geometric point of view, the datum of such a representation is equivalent to that of a flat principal $G$-bundle on $M$, or of a vector bundle of rank $N$ endowed with a flat connection and with monodromy in $G$, in case $G\subset\GL_N$ is a linear group: flatness is somehow the counterpart of the homegeneity of $G/H$. Conversely, a way of ``understanding'' such a representation $\rho$ is to {\it{geometrize}} it, namely to find a geometric structure on $M$ with monodromy $\rho$. \subsection{Closed surfaces} Let $S$ be a compact connected oriented surface of genus $g(S)\geq 2$. \subsubsection{Hyperbolic structures} A remarkable example of geometric structure on $S$ is given by hyperbolic structures, that is hyperbolic metrics on $S$ up to isotopy. Indeed, a hyperbolic metric is locally isometric to the upper half-plane $\HH^2$, and so it induces a $(\PSL_2(\RR),\HH^2)$-structure on $S$ with monodromy representation $\rho:\pi_1(S)\rar\PSL_2(\RR)\cong\mathrm{Isom}_+(\HH^2)$. A result credited to Fricke-Klein \cite{fricke-klein:automorphen} (see also Vogt \cite{vogt:invariants}) states that hyperbolic structures on $S$ are in bijective correspondence with a connected component of the space $\Rep(S,G)$ of conjugacy classes of representations $\pi_1(S)\rar G$ with $G=\PSL_2(\RR)$. \subsubsection{Euler number of a representation in $\PSL_2(\RR)$} The connected components of the whole space $\Rep(S,\PSL_2(\RR))$ can be classified according to a topological invariant of the $\RR\PP^1$-bundle over $S$ associated to each such representation $\rho$: the Euler number $\Eu(\rho)\in\ZZ$. The bound $|\Eu(\rho)|\leq -\chi(S)$ was proven by Milnor \cite{milnor:inequality} and Wood \cite{wood:milnor}; then Goldman \cite{goldman:components} showed that each admissible value corresponds exactly to a connected component of the representation space and that monodromies of hyperbolic structures correspond to the component with $\Eu=-\chi(S)$. It is easy to see that $\rho:\pi_1(S)\rar\PSL_2(\RR)$ can be lifted to $\SL_2(\RR)$ if and only if $\Eu(\rho)$ is even. \begin{remark} More refined invariants of a representation are given by {\it{bounded}} characteristic classes. The bounded Euler class for topological $S^1$-bundles was investigated by Matsumoto \cite{matsumoto:foliated} and the analogous Toledo invariant for $G/H$ of Hermitian type by Toledo \cite{toledo:complex}. Bounded Euler and Toledo classes were used by Burger-Iozzi-Wienhard \cite{burger-iozzi-wienhard:maximal} to characterize maximal representations. \end{remark} \subsubsection{Local study of the representation spaces} Traces of a local study of $\Rep(S,G)$ are already in Weil \cite{weil:discrete-1} \cite{weil:discrete-2}. A more general treatment of the tangent space at a point $[\rho]$ and the determination of the smooth locus of $\Rep(S,G)$ can be found in Goldman \cite{goldman:symplectic}, Lubotzky-Magid \cite{lubotzky-magid:representations} and in the lectures notes \cite{goldman:representations-notes} by Goldman and \cite{labourie:book} by Labourie. A deeper analysis of the singularities of such moduli space can be found in Goldman-Millson \cite{goldman-millson:deformation}. \subsubsection{Symplectic structure on the representation space} When $G$ is reductive, a natural symplectic structure on the smooth locus of $\Rep(S,G)$ is defined by Atiyah-Bott \cite{atiyah-bott:yang-mills} by using the equivalence between representations of the fundamental group of $S$ in $G$ and flat $G$-bundles on $S$. In the case of the Fricke-Klein component of $\Rep(S,\PSL_2(\RR))$, such a symplectic structure was seen by Goldman \cite{goldman:symplectic} to agree with the Hermitian pairing defined by Weil \cite{weil:modules} using Petersson's work \cite{petersson:pairing} on automorphic forms. Ahlfors \cite{ahlfors:kaehler} showed that such a Weil-Petersson pairing defines a K\"ahler form, which is rather ubiquitous when dealing with deformations of hyperbolic structures (see for instance \cite{wolpert:symplectic}, \cite{sozen-bonahon:weil-petersson}, \cite{bonsante-mondello-schlenker:cyclic2}). \subsubsection{Flat unitary bundles and holomorphic bundles} Consider the case $G=\U_N$ and fix a complex structure $I$ on $S$. Representations of $\pi_1(S)$ in the unitary group $\U_N$ were object of a classical theorem by Narasimhan-Seshadri \cite{narasimhan-seshadri:unitary}, in which a real-analytic correspondence is established between irreducible representations $\pi_1(S)\rar\U_N$ and stable holomorphic vector bundles of rank $N$ and degree $0$ on the Riemann surface $(S,I)$. One direction is easy, since every flat complex bundle is $I$-holomorphic; for the other direction, the authors show that stable bundles of degree $0$ admit a flat Hermitian metric: their argument works by continuity method; an analytic proof of this statement was found later by Donaldson \cite{donaldson:narasimhan-seshadri} proving the convergence of the hermitian Yang-Mills flow, as suggested in the fundamental work of Atiyah-Bott \cite{atiyah-bott:yang-mills}. \subsubsection{Flat bundles and Higgs bundles} The celebrated paper \cite{hitchin:self-duality} by Hitchin treated the case of representations in $G=\SL_2$ and established a real-analytic correspondence between irreducible $\rho:\pi_1(S)\rar \SL_2(\CC)$ and stable Higgs bundles $(E,\Phi)$, namely holomorphic vector bundles $E\rar (S,I)$ of rank $N$ and trivial determinant endowed with a holomorphic $\End_0(E)$-valued $(1,0)$-form $\Phi$ on $(S,I)$ and subject to a suitable stability condition. Compared to Narasimhan-Seshadri's, the correspondence is less intuitive, since the holomorphic structure on $E$ does not agree with the underlying holomorphic structure on the flat complex vector bundle $V\rar S$ determined by the representation $\rho$ (coming from the fact that locally constant functions are holomorphic) but it is twisted: the exact amount of such twisting is determined by the aid of the harmonic metric on $V$, whose existence was shown by Donaldson \cite{donaldson:twisted}. For $G=\GL_N$ or $G=\SL_N$, the existence of the harmonic metric was proven (on any compact manifold) by Corlette \cite{corlette:harmonic} (and later Labourie \cite{labourie:harmonic}) and the correspondence (in any dimension) was proven by Simpson \cite{simpson:yang-mills}, who also clarified the general picture by showing \cite{simpson:higgs1} \cite{simpson:higgs2} that the fundamental objects to consider are local systems (classified by a ``Betti'' moduli space), vector bundles with a flat connection (classified by a ``de Rham'' moduli space) and holomorphic Higgs bundles (classified by a ``Dolbeault'' moduli space) and by constructing their moduli spaces. \subsubsection{Correspondence for $\SL_2(\RR)$} Back to the rank $2$ case, among the many results contained in \cite{hitchin:self-duality}, Hitchin could determine which Higgs bundles correspond to monodromies of hyperbolic metrics, thus parametrizing Teichm\"uller space by holomorphic quadratic differentials on $(S,I)$ and making connection with Wolf's result \cite{wolf:teichmuller} (namely, the Higgs field in Hitchin's work identifies to the Hopf differential of the harmonic map in Wolf's parametrization). Moreover, the space of isomorphism classes of Higgs bundles $(E,\Phi)$ carries a natural $S^1$-action $u\cdot(E,\Phi)=(E,u\Phi)$, which is also rather ubiquitous when dealing with harmonic maps with a two-dimensional domain (for instance \cite{bonsante-mondello-schlenker:cyclic1}); in rank $2$, the locus fixed by the $(-1)$-involution $[(E,\Phi)]\leftrightarrow [(E,-\Phi)]$ is identified to the locus of unitary (if $\Phi=0$) or real (if $\Phi\neq 0$) representations. This allows Hitchin to fully determine the topology of the connected components of $\Rep(S,\PSL_2(\RR))$ with non-zero Euler number as that of a complex vector bundle over a symmetric product of copies of $S$. The real component with Euler number zero seems slightly subtler to treat, since it contains classes of reducible representations (or, equivalently, of strictly semi-stable Higgs bundles) for which the correspondence does not hold. \subsection{Surfaces with punctures} Let $S$ be a compact connected oriented surface and let $P=\{p_1,\dots,p_n\}\subset S$ be a subset of $n$ distinct marked points. Denote by $\dot{S}$ the punctured surface $S\setminus P$ and assume $\chi(\dot{S})<0$. \subsubsection{Absolute and relative representation space} The space $\Rep(\dot{S},G)$ of conjugacy classes of representations $\rho:\pi_1(\dot{S})\rar G$ can be partitioned according to the boundary behavior of $\rho$. More explicitly, fix an $n$-uple $\bCo=(\bco_1,\dots,\bco_n)$ of conjugacy classes in $G$ and define $\Rep(\dot{S},G,\bCo)$ as the space of conjugacy classes of representations $\rho:\pi_1(\dot{S})\rar G$ that send a loop positively winding about the puncture $p_i$ to an element of $\bco_i\subset G$. \subsubsection{Spherical and hyperbolic structures} Similarly to the case of a closed surface, isotopy classes of metrics of constant curvature $K$ are the easiest examples of geometric structures on $\dot{S}$; a standard requirement is to ask that the completion of such metrics has either conical singularities or geodesic boundary of finite length (or cusps, if $K<0$) at the punctures. Monodromies of spherical structures ($K=1$) naturally take values in $\mathrm{PSU}_2\cong\SO_3(\RR)$ but can be lifted to $\SU_2$: if $S$ has genus $0$, such liftability imposes restrictions on the angles of the conical points of a spherical metric \cite{mondello-panov:spherical}. Monodromies of hyperbolic structures ($K=-1$) determine conjugacy classes of representations $\rho:\pi_1(\dot{S})\rar\PSL_2(\RR)$, which are also liftable to $\SL_2(\RR)$. \subsubsection{Euler number of a representation in $\PSL_2(\RR)$} The Euler number of a representation $\rho:\pi_1(\dot{S})\rar\PSL_2(\RR)$ and a generalized Milnor-Wood inequality $|\Eu(\rho)|\leq -\chi(\dot{S})$ are treated by Burger-Iozzi-Wienhard in \cite{burger-iozzi-wienhard:maximal}, who also show that all values in the interval $[\chi(\dot{S}),-\chi(\dot{S})]$ are attained and that representations $\rho$ with $\Eu(\rho)=-\chi(\dot{S})$ correspond to monodromies of complete hyperbolic metrics on $\dot{S}$. Since $\Eu:\Rep(\dot{S},\PSL_2(\RR))\rar\RR$ is continuous and its restriction to the locus $\Rep(\dot{S},\PSL_2(\RR),\bCo)$ is locally constant, it is an invariant of the connected components of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$. \subsubsection{Local structure and Poisson structure} Similarly to the closed case, representations $\pi_1(\dot{S})\rar G$ (possibly with prescribed boundary values) correspond to flat $G$-bundles with the same boundary monodromy; the deformation theory is also analogous. If $G$ is reductive, a natural Poisson structure \cite{GHJW:poisson} can be defined on the smooth locus of $\Rep(\dot{S},G)$, which restricts to a symplectic structure on the smooth locus of the spaces $\Rep(\dot{S},G,\bCo)$ (see for instance \cite{mondello:poisson} \cite{mondello:weil-petersson} for its link with Weil-Petersson structure when $G=\PSL_2(\RR)$ and for explicit formulae in the case of surfaces with conical points or with boundary geodesics). \subsubsection{Flat unitary bundles and holomorphic parabolic bundles} Unitary representations $\pi_1(\dot{S})\rar\U_N$ determine a complex vector bundle $\dot{V}\rar\dot{S}$ of rank $N$ endowed with a flat connection and a parallel Hermitian metric. Such a vector bundle admits a canonical extension $V\rar S$ (Deligne \cite{deligne:equations}), in such a way that the connection may have at worst simple poles at $P$ with eigenvalues of the residues in $[0,1)$ and the natural parallel Hermitian metric $H$ vanishes at $P$ of order in $[0,1)$. Mehta-Seshadri \cite{mehta-seshadri:parabolic} introduced the important notion of a parabolic structure on $V$ at $P$, namely a filtration of the fibers of $V$ over $P$ by order of growth with respect to $H$, and established the analogue of Narasimhan-Seshadri's result: for every complex structure $I$ on $S$, there is a correspondence between irreducible flat unitary bundles on $\dot{S}$ of rank $N$ with prescribed monodromy at the punctures and stable holomorphic bundles of rank $N$ and (parabolic) degree $0$ on $(S,I)$ with parabolic structure at $P$ of prescribed type. As in the closed case, going from a flat bundle to a holomorphic parabolic bundle is easy; conversely, the existence of a flat Hermitian metric on a stable holomorphic bundle of degree $0$ with prescribed polynomial growth at the parabolic points $p_i$ was achieved in \cite{mehta-seshadri:parabolic} by continuity method, and then proved by Biquard \cite{biquard:paraboliques} by analytic techniques. \subsubsection{Flat bundles and parabolic Higgs bundles} In a fundamental article \cite{simpson:harmonic} Simpson established the correspondence between representations $\rho:\pi_1(\dot{S})\rar \GL_N(\CC)$ with Zariski-dense image and parabolic $I$-holomorphic vector bundles $E_\bullet$ of rank $N$ and degree $0$ endowed with a Higgs field $\Phi\in H^0(S,K_S\otimes \End(E_\bullet))$ subject to a suitable stability condition, the weights of $E_\bullet$ and the residues of $\Phi$ at $P$ being determined by the values of $\rho$ on peripheral loops. The real-analytic nature of Simpson's correspondence was proven by Konno \cite{konno:construction} and Biquard-Boalch \cite{biquard-boalch:wild}. The case of a general algebraic reductive group $G$ was recently treated by Biquard, Garcia-Prada and Mundet i Riera \cite{biquard:higher}. \subsubsection{Topological study of moduli spaces of Higgs bundles} Following Hitchin's ideas, Boden-Yokogawa \cite{boden-yokogawa:parabolic} analyzed some aspects of the case of $G=\SL_2(\CC)$ and in particular the Betti numbers of the moduli space using Morse theory. Their result was then extended by Logares \cite{logares:U(21)} to the case of $\mathrm{U}(2,1)$-Higgs bundles and by Garc{\'i}a-Prada, Gothen and Mu{\~n}oz \cite{garcia-prada-gothen-munoz:rank3} to the $\SL_3(\CC)$ and $\GL_3(\CC)$ cases; on the other hand, Garc{\'i}a-Prada, Logares and Mu{\~n}oz \cite{garcia-prada-logares-munoz:U(pq)} established the a Milnor-Wood inequality and determined the connected components of the moduli space of $\mathrm{U}(p,q)$-Higgs bundles. A different approach via orbifold structures was taken by Nasatyr-Steer \cite{nasatyr-steer:orbifold}, who implemented Hitchin's ideas in rank $2$ and also determined the topology of the relative $\SL_2(\RR)$-representation space in the case of positive Euler number and elliptic boundary monodromy of finite order. \subsubsection{Content of the paper} After giving an overview of the subject and of the fundamental results of the theory mentioned above, we focus on the topology of the real locus of the moduli space of parabolic $\SL_2$-Higgs bundles, and in particular on what happens as the parabolic structure degenerates, or equivalently on the topology of $\SL_2(\RR)$-representation spaces as some of the boundary monodromies cease to be strictly elliptic.\\ Fix a complex structure $I$ on $S$. Given conjugacy classes $\bcoa_i\subset\psl_2(\CC)$, numbers $w_1(p_i)\in [0,\frac{1}{2}]$ and a line bundle $\Det$ on $S$, we consider the moduli space $\Higgs^s(S,\bm{w},2,\Det,\bCoa)$ of stable parabolic Higgs bundles $(E_\bullet,\Phi)$ on $S$ of rank $2$ with parabolic weights $w_1(p_1),\,1-w_1(p_i)$ (briefly, of type $\bm{w}$), $\det(E_\bullet)\cong\Det$ and residue $\Res_{p_i}(\Phi)$ in $\bcoa_i$. In order to avoid to introduce too much notation at this point, we prefer not to give here complete statements of the main results contained in this paper but rather to list them in an informal way: \begin{itemize} \item[(1)] stable parabolic Higgs bundles in $\Higgs^s(S,\bm{w},2,\Det,\bCoa)$ that correspond to representations $\rho:\pi_1(\dot{S})\rar \SL_2(\RR)$ with $\Eu(\rho)>0$ are characterized in Lemma \ref{lemma:real} and Theorem \ref{thm:correspondence}; \item[(2)] a classification of the connected and the irreducible components of the locus of $\Higgs^s(S,\bm{w},2,\Det,\bCoa)$ mentioned in (1) and of its closure and the determination of their topology is obtained in Proposition \ref{prop:real} and Proposition \ref{prop:topology}; in particular, these results combine in Theorem \ref{thm:correspondence} to show that the closure (for the classical topology) of each connected component of $\Rep(\dot{S},\SL_2(\RR),\bCo)$ with $\Eu>0$ is homeomorphic either to a complex vector space or to an $H^1(S;\ZZ/2\ZZ)$-cover of a complex vector bundle over a symmetric product of $S\setminus P_{hyp}$ with $P_{hyp}=\{p_i\in P\,|\,\text{$c_i$ hyperbolic}\}$; \item[(3)] again by Proposition \ref{prop:topology} and Theorem \ref{thm:correspondence}, the closure of each connected component of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$ with $\Eu>0$ is homeomorphic to a complex vector bundle over a symmetric product of $S\setminus P_{hyp}$; \item[(4)] the connected components of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$ are classified by their Euler number by Corollary \ref{cor:euler-components}; \item[(5)] the connected components of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$ that can host monodromies of hyperbolic structures are determined in Proposition \ref{prop:uniformization} and their topology is deduced in Corollary \ref{cor:uniformization}. \end{itemize} The precise topological description of such representation spaces in the punctured case can be compared to the one in the closed case by looking at Theorem \ref{thm:rep-closed} and Theorem \ref{thm:rep-punctured}. \begin{remark} In the special case of all elliptic boundary monodromies, the above results (1-4) are already in \cite{boden-yokogawa:parabolic} and, for elliptic boundary monodromies of finite order, the same topological description was obtained in \cite{nasatyr-steer:orbifold}. The case of monodromy representations of hyperbolic structures with cusps (and so maximal Euler number) was analyzed by Biswas, Ar{\'e}s-Gastesi and Govindarajan \cite{biswas:parabolic}. Similarly to what happens with closed surfaces, the work of Wolf \cite{wolf:infinite-energy} on harmonic maps that ``open the node'' from nodal Riemann surfaces to smooth hyperbolic surfaces relates to the case of an $\SL_2(\RR)$-Higgs bundle with imaginary residues at the punctures. \end{remark} As an example of a by-product of our analysis, we describe the topology of components of the representation space that can contain monodromies of hyperbolic structure (see also Corollary \ref{cor:uniformization}). \begin{corollary}[Topology of uniformization irreducible components]\label{cor:intro} Let $\rho$ be the monodromy representation of a hyperbolic metric on $\dot{S}$ of area $2\pi e>0$ whose completion has conical singularities of angles $\th_1,\dots,\th_k>0$ at $p_1,\dots,p_k$, cusps at $p_{k+1},\dots,p_r$ and geodesic boundaries of lengths $\ell_{r+1},\dots,\ell_n> 0$. Let $\bco_i$ be the conjugacy class of the monodromy $\rho$ about the $i$-th end of $\dot{S}$ and let $s_0=\#\{i\in\{1,\dots,k\}\,|\,\th_i\in 2\pi\NN_+\}$.\\ Then $[\rho]$ belongs to the irreducible component of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$ with Euler number $e=-\chi(\dot{S})-\sum_{i=1}^k \frac{\th_i}{2\pi}>0$, and such irreducible component is real-analytically diffeomorphic to a holomorphic vector bundle of rank $3g-3+n-m$ over $\mathrm{Sym}^{m-s_0}(S\setminus\{p_{k+1},\dots,p_n\})$, where $m=\sum_{i=1}^k\floor{\frac{\th_i}{2\pi}}$. Moreover, its closure $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})$ is homeomorphic to a holomorphic vector bundle of rank $3g-3+n-m$ over $\mathrm{Sym}^{m-s_0}(S\setminus\{p_{r+1},\dots,p_n\})$ \end{corollary} \subsection{Structure of the paper} In Section \ref{sec:representations} we review the definition of representation space of the fundamental group of a punctured surface $\dot{S}$ and the basic smoothness results. Then we recall the (Riemann-Hilbert) correspondence between representations of the fundamental group of a surface in an algebraic group $G$ and flat principal $G$-bundles. In particular, we review the case of a linear group $G\subset\GL_N$ and the flat vector bundle of rank $N$ attached to a representation. Then we analyze the case of a $\PSL_2(\RR)$-representation $\rho$ and we discuss the Euler number of $\rho$, some well-known fundamental results on the topology of the $\PSL_2(\RR)$-representation space and we compare our results with them. Finally, we briefly mention monodromy representations coming from hyperbolic structures possibly with cusps, conical singularities and geodesic boundaries. In Section \ref{sec:parabolic} we first recall the notion of parabolic bundle following Simpson and we show examples over the disk $\Delta$, originating (\`a la Mehta-Seshadri) from representations $\pi_1(\dot{\Delta})\rar \U_1$ and $\pi_1(\dot{\Delta})\rar\SU_2$. Similarly, we introduce the definition of Higgs bundles and show examples coming from $\pi_1(\dot{\Delta})\rar \GL_N(\CC)$ with $N=1,2$. Then we recall the notion of slope stability and of moduli space of stable parabolic Higgs bundles. Finally, we specialize to the case of $G=\SL_2$ and we study the topology of the locus fixed by the involution $[(E_\bullet,\Phi)]\leftrightarrow [(E_\bullet,-\Phi)]$ and that corresponds to representations in $\SL_2(\RR)$ with $\Eu>0$. In the final Section \ref{sec:correspondence} we recall first the main correspondence results in the theory of representations of fundamental groups of surfaces and holomorphic (parabolic) (Higgs) bundles. Then we illustrate how the correspondence works for $\SL_2(\RR)$ and for $\PSL_2(\RR)$ and we describe the topology of the components of the representation space. We finally conclude with two corollaries about uniformization components. \subsection{Acknowledgements} The motivation for this work originated in conversations about monodromies of hyperbolic structures with conical points with Roberto Frigerio, whom I wish to thank. I am grateful to Nicolas Tholozan (reporting a conversation with Olivier Biquard) for observing that the case of compact components with non-degenerate parabolic structure as in Corollary \ref{cor:supermaximal} corresponds to the class of ``super-maximal'' representations studied by Deroin-Tholozan in \cite{deroin-tholozan:super-maximal} and to Bertrand Deroin for explaining me that the analysis carried out in this paper leads to Corollary \ref{cor:geometric}. I also thank Paul Seidel for pointing out a former discrepancy (now corrected) in the statement of Corollary \ref{cor:intro}, Johannes Huebschmann for precise remarks on the locally semialgebraic nature of real representation spaces and on their tangent spaces, Peter Gothen and an anonymous referee for drawing my attention to some relevant works in the field, and both referees for useful comments. The author's research was partially supported by FIRB 2010 national grant ``Low-dimensional geometry and topology'' (code: \texttt{RBFR10GHHH\_003}) and by GNSAGA INdAM group. \subsection{Notation} Let $S$ be compact, connected, oriented surface of genus $g(S)$ and $P=\{p_1,\dots,p_n\}$ be a subset of distinct points of $S$. Denote by $\dot{S}$ the punctured surface $S\setminus P$ and assume $\chi(\dot{S})<0$. Let $\pi$ be the fundamental group $\pi_1(\dot{S},b)$, where $b\in \dot{S}$ is a base point, and fix a universal cover $(\wti{\dot{S}},\ti{b})\rar (\dot{S},b)$ on which $\pi$ then acts by deck transformations. Fix a standard set $\{\a_1,\b_1,\dots,\a_{g(S)},\b_{g(S)},\gamma_1,\dots,\gamma_n\}$ of generators of $\pi$, that satisfy the unique relation $[\a_1,\b_1]\cdots[\a_{g(S)},\b_{g(S)}]\gamma_1\cdots\gamma_n=\id$. In particular, $\gamma_i$ is freely homotopic to a small loop that simply winds about the puncture $p_i$ counterclockwise. We will also write $\pa_i$ for the conjugacy class of $\gamma_i$ in $\pi$ (or, equivalently, for the free homotopy class of $\gamma_i$ on $\dot{S}$). We will denote by $G$ a reductive real or complex algebraic group with finite center $Z=Z(G)$ and by $\gfrak$ its Lie algebra. If $\bco_i$ is a conjugacy class or a union of conjugacy classes in $G$, then we denote by $\ol{\bco}_i$ its closure. Similarly, if $\bcoa_i$ is a conjugacy class or a union of conjugacy classes in $\gfrak$, we denote by $\ol{\bcoa}_i$ its closure. We use the symbol $\bCo$ for an $n$-uple $(\bco_1,\dots,\bco_n)$ and $\bCoa$ for an $n$-uple $(\bcoa_1,\dots,\bcoa_n)$, and similarly for their closures $\ol{\bCo}$ and $\ol{\bCoa}$. If $\bm{r}=(r_1,\dots,r_n)$ is a string of $n$ non-negative real numbers, then $\NO{\bm{r}}$ will denote their sum $r_1+r_2+\dots+r_n$. \subsubsection{Convention} We identify $\HH\subset\CC\PP^1$ via $z\mapsto [1:z]$, so that a matrix in $\PSL_2(\RR)$ acts on $\HH$ as \[ \left( \begin{array}{cc} a & b\\ c & d \end{array}\right)\cdot z= \frac{c+dz}{a+bz} \] Consider the transformations $R_\theta,T\in\PSL_2(\RR)$ defined as \[ R_\theta=\left( \begin{array}{cc} \cos(\theta/2) & -\sin(\theta/2)\\ \sin(\theta/2) & \cos(\theta/2) \end{array}\right), \quad T=\left( \begin{array}{cc} 1 & 0\\ 1 & 1 \end{array}\right) \] Isometries of $\HH$ conjugate to $R_\theta$ (resp. to $T$, or to $T^{-1}$) in $\PSL_2(\RR)$ are called {\it{rotations of angle $\theta$}} (resp. {\it{positive unipotents}}, or {\it{negative unipotents}}). \section{Representations of the fundamental group and flat bundles}\label{sec:representations} \subsection{Representation spaces} The set $\Hom(\pi,G)$ of homomorphisms $\pi\rar G$ is denoted by $\Rep_b(\dot{S},G)$ and it can be identified to the locus \[ \left\{ (A_1,B_1,\dots,A_{g(S)},B_{g(S)},C_1,\dots,C_n)\in G^{2{g(S)}+n}\,|\, [A_1,B_1]\cdots[A_{g(S)},B_{g(S)}]C_1\cdots C_n=\id \right\} \] inside $G^{2{g(S)}+n}$. The algebraic structure on $\Rep_b(\dot{S},G)$ induced as a hypersurface in $G^{2{g(S)}+n}$ is independent of the choice of the generators. The group $G$ acts on $\Rep_b(\dot{S},G)$ by conjugation, that is sending a homomorphism $\rho\in\Rep_b(\dot{S},G)$ to $\Ad_g\circ \rho$. Given any other base-point $b'\in\dot{S}$, the isomorphism $\Rep_b(\dot{S},G)\cong\Rep_{b'}(\dot{S},G)$ depends on the choice of a path between $b$ and $b'$, but it becomes canonical after factoring out the action of $G$. \begin{definition} The {\it{representation space}} $\Rep(\dot{S},G)$ is the Hausdorffization of the topological quotient $\RRep(\dot{S},G):=\Rep_b(\dot{S},G)/G$. \end{definition} If $G$ is compact, then all $G$-orbits in $\Rep_b(\dot{S},G)$ are closed and so the map $\RRep(\dot{S},G)\rar\Rep(\dot{S},G)$ is a homeomorphism. \begin{remark} If $G$ is a complex group, then $\Rep(\dot{S},G)$ agrees with the GIT quotient $\Rep_b(\dot{S},G)/\!/G$. If $G$ is real and $G_\CC$ is its complexification, then $\Rep(\dot{S},G)$ is a locally semialgebraic subset (i.e. locally defined by real algebraic equalities and inequalities) inside the real locus $\Rep(\dot{S},G_\CC)(\RR)$, see for instance \cite{huebschmann:singularities}. \end{remark} \subsubsection{Closed case} For closed surfaces ($n=0$ and so $\dot{S}=S$), the following result was proven by Rapinchuk, Benyash-Krivetz and Chernousov \cite{RBC:representation}. \begin{theorem}[Irreducibility for representations in $\SL_N(\CC)$ and $\PSL_N(\CC)$] The algebraic varieties $\Rep_b(S,G)$ are irreducible for $G=\SL_N(\CC),\, \PSL_N(\CC)$. \end{theorem} \begin{problem}[Topology of representation spaces] Assuming $n=0$, determine the topology of $\Rep_b(S,G)$ and $\Rep(S,G)$: in particular, enumerate the connected components (for the classical topology) of $\Rep(S,G)$. \end{problem} We will see below that the above problem was almost completely solved by Hitchin \cite{hitchin:self-duality} for $G=\SL_2(\RR),\,\PSL_2(\RR)$. \subsubsection{Punctured case} Assume now $n>0$. Since $\pi_1(\dot{S})$ is free on $2g(S)+n-1$ generators, $\Rep_b(\dot{S},G)$ is isomorphic to $G^{2{g(S)}+n-1}$ and $G$ acts diagonally by conjugation on each factor of $G^{2{g(S)}+n-1}$. The situation becomes more interesting if we consider the relative situation. \begin{definition} Let $\bCo=(\bco_1,\dots,\bco_n)$, where $\bco_i$ is a conjugacy class in $G$ (or possibly a finite union of conjugacy classes). The {\it{relative homomorphism space}} $\Rep_b(\dot{S},G,\bCo)$ is the locus in $\Rep_b(\dot{S},G)$ of homomorphisms $\rho:\pi\rar G$ such that $\rho(\pa_i)\in\bco_i$. \end{definition} Equivalently, $\Rep_b(\dot{S},G,\bCo)$ is the preimage of $\bCo\subset G^n$ under the evaluation map $\ev:\Rep_b(\dot{S},G)\rar G^n$ that sends $\rho$ to $\big(\rho(\gamma_1),\dots,\rho(\gamma_n)\big)$. If $G$ is a complex group, then conjugacy classes and their closures are algebraic subvarieties and so $\Rep_b(\dot{S},G,\bCo)$ is an algebraic subvariety of $\Rep_b(\dot{S},G)$. If $G$ is real, then conjugacy classes and their closures are in general semialgebraic subsets and so $\Rep_b(\dot{S},G,\bCo)$ is semialgebraic inside $\Rep_b(\dot{S},G)$. \begin{example} In the special case of $G=\PSL_2(\RR)$ conjugacy classes consisting of hyperbolic or elliptic elements are algebraic and closed; on the other hand, the class $\bco\subset\PSL_2(\RR)$ consisting of positive unipotent elements (or negative unipotent elements) is semi-algebraic and not closed: indeed, its classical closure $\ol{\bco}$ consists of $\bco\cup\{\id\}$ and its Zariski closure is $\bco\cup\{\id\}\cup(-\bco)$. \end{example} The definition of $\Rep_b(\dot{S},G,\bCo)$ as well as the induced (semi-)algebraic structure are independent of the choice of the loops $\gamma_1,\dots,\gamma_n$. \begin{definition} The {\it{relative representation space}} $\Rep(\dot{S},G,\bCo)$ is the Hausdorffization of the topological quotient $\RRep(\dot{S},G,\bCo):=\Rep_b(\dot{S},G,\bCo)/G$. \end{definition} As in the absolute case, if $G$ is a complex group, then $\Rep(\dot{S},G,\bCo)$ agrees with the GIT quotient $\Rep_b(\dot{S},G,\bCo)/\!/G$ and it is an algebraic subvariety of $\Rep(\dot{S},G)$. If $G$ is real, then $\Rep(\dot{S},G,\bCo)$ is a locally semialgebraic subset of $\Rep(\dot{S},G)$. \begin{problem}[Topology of relative representation spaces] Assuming $n>0$, determine the topology of $\Rep_b(\dot{S},G,\bCo)$ and $\Rep(\dot{S},G,\bCo)$: in particular, enumerate the connected components (for the classical topology) of $\Rep(\dot{S},G,\bCo)$. \end{problem} Same remarks and questions hold for representations with boundary values in $\ol{\bCo}$. \begin{remark}\label{rmk:center} In light of the short exact sequence $0\rar Z\rar G\rar G/Z\rar 0$, we can view $\Rep_b(\dot{S},G)$ as an $H^1(\dot{S},Z)$-bundle over $\Rep_b(\dot{S},G/Z)$. It is easy to see that, if $\tilde{c}_i\subset G$ is the preimage of the conjugacy class $c_i\subset G/Z$ under the projection $G\rar G/Z$, then $\Rep_b(\dot{S},G,\bm{\wti{\bCo}})$ is an $H^1(S,Z)$-bundle over $\Rep_b(\dot{S},G/Z,\bCo)$. \end{remark} \subsection{Flat $G$-bundles} Let $\Gcal$ be the sheaf of smooth functions with values in $G$ and $\ul{G}$ the subsheaf of locally constant functions. It is well-known that there is a bijective correspondence between $G$-local systems on $\dot{S}$ and principal $G$-bundles $\xi\rar\dot{S}$ endowed with a flat connection $\nabla\in \Omega^1(\xi,\gfrak)^G$. Indeed, for every flat $G$-bundle $(\xi,\nabla)$, the sheaf $\ul{\xi}$ of parallel sections of $\xi$ is a local system; vice versa, given a $G$-local system $\ul{\xi}$, the $G$-bundle whose sheaf of smooth sections is $\xi=\Gcal\times_{\ul{G}}\ul{\xi}$ is endowed with a flat connection induced by the exterior differential $d:\Ocal\rar\Omega^1$. Such a construction also establishes a correspondence between {\it{framed flat principal $G$-bundles}} $(\xi,\nabla,\tau)$ consisting of a flat $G$-bundle $(\xi,\nabla)$ on $\dot{S}$ with a trivialization $\tau:\xi_b\arr{\sim} G$ at the base point $b\in\dot{S}$ and {\it{framed $G$-local systems}} $(\ul{\xi},\tau')$ consisting of a $G$-local system $\ul{\xi}$ with a trivialization $\tau':\ul{\xi}_b\arr{\sim} G$ at $b$. The isomorphisms $\tau$ and $\tau'$ are called {\it{framings}}. \begin{notation} We will denote by $\Flat(\dot{S},G)$ the set isomorphism classes of flat principal $G$-bundles on $\dot{S}$ and by $\Flat_b(\dot{S},G)$ the set of isomorphism classes of $b$-framed flat principal $G$-bundles on $\dot{S}$. \end{notation} We wish to recall the well-known correspondence between flat $G$-bundles and $G$-representations of $\pi$ and to adapt it to the framed case. Consider the trivial $G$-bundle $\ti{\xi}:=\wti{\dot{S}}\times G\rar\wti{\dot{S}}$ with the framing $\ti{\tau}: \ti{\xi}_{\ti{b}}=\{\ti{b}\}\times G\arr{\sim}G$ given by the projection onto the second factor. The flat connection $\wti{\nabla}$ on $\ti{\xi}$ is simply given by the de Rham differential. Given a representation $\rho\in\Rep_b(\dot{S},G)$, the fundamental group $\pi$ acts on $\ti{\xi}$, and more precisely via deck transformations on the factor $\wti{\dot{S}}$ and via $m_L\circ \rho$ on the factor $G$, where $m_L$ is the action of $G$ on $G$ by left multiplication. As a consequence, $\wti{\nabla}$ descends to a flat connection $\nabla$ on the $G$-bundle $\xi_\rho:=\ti{\xi}/\pi$ on $\dot{S}$. Moreover, $\ti{\tau}$ induces a framing $\tau:\xi_b\arr{\sim}G$ through the isomorphism $\ti{\xi}_{\ti{b}}\arr{\sim}\xi_b$. This construction determines an application \[ \Xi_b:\Rep_b(\dot{S},G)\lra \Flat_b(\dot{S},G). \] Vice versa, given a framed flat $G$-bundle $(\xi,\nabla,\tau)$, the holonomy representation based at $b$ descends to a homomorphism $\pi\rar \Aut(\xi_b)\cong G$ by the flatness of $\nabla$, and so via $\tau$ to a homomorphism $\rho:=\hol_b(\xi):\pi\rar G$. This construction determines an application \[ \hol_b:\Flat_b(\dot{S},G)\lra \Rep_b(\dot{S},G). \] It is easy to check that $\Xi_b$ and $\hol_b$ are set-theoreticallly inverse of each other. Any two trivializations $\tau_1,\tau_2: \xi_b\rar G$ at $b$ are related by a unique element $g\in G$, namely $\tau_2=m_L(g)\circ\tau_1$. Hence, factoring out the action of $G$ one obtains the applications $\Xi: \RRep(\pi,G)\rar \Flat(\dot{S},G)$ and $\Rhol: \Flat(\dot{S},G)\rar \RRep(\pi,G)$ which are set-theoretically inverse of each other. \begin{remark} Viewing $\Flat_b(\dot{S},G)$ as a quotient of the space of flat connections on $G\times \dot{S}\rar \dot{S}$, it is possible to endow $\Flat_b(\dot{S},G)$ with the structure of (real or complex) analytic variety that makes $\hol_b$ and $\Xi_b$ analytic isomorphisms. Since we do not want to go deeper in this direction, we can alternatively just put on $\Flat_b(\dot{S},G)$ the analytic structure induced by $\hol_b$ and $\Xi_b$. \end{remark} The correspondence in the relative case is dealt with analogously. \begin{notation} Denote by $\Flat_b(\dot{S},G,\bCo)$ the set of isomorphism classes of $b$-framed flat $G$-bundles on $\dot{S}$ with holonomy along the path $\pa_i$ in $\co_i$, and by $\Flat(\dot{S},G,\bCo)$ the set of isomorphisms classes of $G$-bundles on $\dot{S}$ with holonomy along the $i$-th end in $\co_i$. Analogous notation for flat bundles with boundary monodromy in $\ol{\bCo}$. \end{notation} The same construction as above works in the relative case and we summarize the discussion in the following statement. \begin{lemma}[Equivalence between representations of $\pi_1$ and flat $G$-bundles] The applications \[ \xymatrix@R=0in{ \Rep_b(\dot{S},G,\bCo) \ar@/^1pc/[rr]^{\hol_b} && \ar@/^1pc/[ll]^{\Xi_b} \Flat_b(\dot{S},G,\bCo) } \] are set-theoretically inverse of each other. By factoring the $G$-action by conjugation on the representation and by left multiplication on the $b$-framing, we recover the correspondence \[ \xymatrix@R=0in{ \RRep(\dot{S},G,\bCo)\ar@/^1pc/[rr]^{\Rhol} && \ar@/^1pc/[ll]^{\Xi} \Flat(\dot{S},G,\bCo). } \] \end{lemma} We call the composition $\Flat(\dot{S},G,\bCo)\arr{\Rhol}\RRep(\dot{S},G,\bCo)\rar \Rep(\dot{S},G,\bCo)$ simply $\hol$. \subsection{Smoothness} Given a representation $\rho:\pi\rar G$, we denote by $\gfrak^\rho\subset\gfrak$ (resp. $G^\rho\subset G$) the subset of elements which are invariant under the adjoint action of $\pi$ though $\rho$. We can naturally identify $\Aut(\xi_\rho)$ with the centralizer $Z_G(\rho(\pi))=G^{\rho}$ of $\rho(\pi)$ in $G$ and the space $T_\id \Aut(\xi_\rho)$ of first-order automorphisms with $\gfrak^\rho\cong H^0(\dot{S},\Ad(\xi_\rho))$, where $\Ad(\xi_\rho)$ is the flat $\gfrak$-bundle with monodromy $\Ad\circ\rho$. \begin{definition} A representation $\rho$ is {\it{regular}} if $\gfrak^\rho=\{0\}$; it is {\it{Zariski-dense}} if its image is Zariski-dense. We denote by $\Rep_b^r(\dot{S},G)$ (resp. by $\Rep_b^{Zd}(\dot{S},G)$) the subsets of $\Rep_b(\dot{S},G)$ of regular (resp. Zariski-dense) representations, and similarly by $\Rep^r(\dot{S},G)$ (resp. by $\Rep^{Zd}(\dot{S},G)$) the corresponding locus in $\Rep(\dot{S},G)$. \end{definition} Since we are assuming $G$ algebraic, $\rho$ is regular if and only if $Z(\rho(\pi))$ is finite. A proof of the following statement can be found in \cite{johnson-millson:deformation} and in Section 5.3.4 of \cite{labourie:book}. \begin{lemma}[Proper action of $G$ on $\Rep_b^r$]\label{lemma:properness} The action on the subset $\Rep_b^{Zd}(\dot{S},G)\subseteq\Rep_b^r(\dot{S},G)$ is proper and with stabilizer $Z$. \end{lemma} A standard deformation theory argument shows that first-order deformations of $\xi$ are parametrized by $H^1(\dot{S},\Ad(\xi))$. The proof of the following result can be found for instance in \cite{goldman:symplectic}, \cite{johnson-millson:deformation}, \cite{labourie:book}. \begin{proposition}[Tangent space to representation spaces]\label{prop:smoothness} For every representation $\rho\in\Rep_b(\dot{S},G)$, we have \[ H^0(\dot{S};\Ad(\xi_\rho)) \cong \gfrak^\rho \qquad\text{and}\qquad H^2(\dot{S};\Ad(\xi_\rho)) \cong \begin{cases} (\gfrak^\vee)^\rho & \text{if $n=0$}\\ 0 & \text{otherwise} \end{cases} \] by Poincar\'e duality. The centralizer $G^\rho$ acts on $H^1(\dot{S};\Ad(\xi_\rho))$ by adjunction and the quotient can be identified to $T_{[\rho]}\Rep(\dot{S},G)$. Hence, the regular locus is an open dense orbifold in the representation spaces and \begin{align*} \def\arraystretch{1.3}\begin{array}{|c|c|c|} \hline & \Rep^r_b(\dot{S},G) & \Rep^r(\dot{S},G)\\ \hline \dim & (-\chi(\dot{S})+1)\dim(G) & -\chi(\dot{S})\dim(G) \\ \hline \end{array} \end{align*} for any $n\geq 0$. \end{proposition} The tangent space $T_{[\rho]}\Rep(\dot{S},G,\bCo)$ in the relative case can be analyzed by means of the following exact sequence associated to the couple $(\dot{S},\dot{\Delta})$ \begin{align*} 0=H^0(\dot{S},\dot{\Delta};\Ad(\xi))\rar H^0(\dot{S};\Ad(\xi)) \rar H^0(\dot{\Delta};\Ad(\xi))\rar H^1(\dot{S},\dot{\Delta};\Ad(\xi))\rar\\ \rar H^1(\dot{S};\Ad(\xi))\rar H^1(\dot{\Delta};\Ad(\xi)) \rar H^2(\dot{S},\dot{\Delta};\Ad(\xi))\rar H^2(\dot{S};\Ad(\xi))=0 \end{align*} where the $\Delta_i$'s are disjoint open contractible neighbourhoods of the $p_i$'s and $\dot{\Delta}=\bigcup_{i=1}^n \dot{\Delta}_i$ is the union of the punctured disks $\dot{\Delta}_i=\Delta_i\setminus\{p_i\}$. In fact, since $G$ is reductive, the Lie algebra $\gfrak$ has a non-degenerate invariant symmetric bilinear form, which induces an $\Ad$-invariant isomorphism $\gfrak\cong\gfrak^\vee$. By Lefschetz duality, $H^2(\dot{S},\dot{\Delta};\Ad(\xi))\cong H^0(\dot{S};\Ad(\xi))^\vee$ and $H^1(\dot{S},\dot{\Delta};\Ad(\xi))\cong H^1(\dot{S};\Ad(\xi))^\vee$. Thus, we obtain the $G^\rho$-equivariant exact sequence \[ 0\rar \gfrak^\rho\rar H^0(\dot{\Delta};\Ad(\xi))\arr{\epsilon^\vee} H^1(\dot{S};\Ad(\xi))^\vee \rar H^1(\dot{S};\Ad(\xi)) \arr{\epsilon} H^1(\dot{\Delta};\Ad(\xi))\rar (\gfrak^\vee)^\rho \rar 0 \] Cocycles in $\ker(\epsilon)$ induce first-order deformations $\rho_t$ of $\rho$ such that for every $i$ the boundary value $\rho_t(\gamma_i)$ is conjugate to $\rho(\gamma_i)$ for all $t$. Thus, taking $G^\rho$-coinvariants, we obtain $T_{[\rho]}\Rep(\dot{S},G,\bCo)\cong \ker(\epsilon)_{G^\rho}$. Combining the above computation with the properness in Lemma \ref{lemma:properness}, we can conclude as follows. \begin{corollary}[Tangent space to relative representation spaces]\label{cor:relative-smoothness} For $n>0$, the locus $\Rep^r(\dot{S},G,\bCo)$ is a smooth orbifold of dimension \[ \def\arraystretch{1.3}\begin{array}{|c|c|c|} \hline & \Rep^r_b(\dot{S},G,\bCo) & \Rep^r(\dot{S},G,\bCo)\\ \hline \dim & (-\chi(S)+1)\dim(G)+ \dim(\bCo) & -\chi(S)\dim(G) +\dim(\bCo)\\ \hline \end{array} \] Moreover, the singular locus of $\Rep^r(\dot{S},G,\ol{\bCo})$ consists of those $[\rho]$ with boundary values in the singular locus of $\ol{\bCo}$. \end{corollary} \subsection{Flat vector bundles} Let $\KK=\RR,\CC$ and consider first $G=\GL_N(\KK)$. To every flat principal $G$-bundle $\xi$ on $\dot{S}$ we can associate a vector bundle $V=\xi\times_G \KK^N$ of rank $N$ endowed with a natural flat connection $\nabla$ in such a way that the monodromy of $(V,\nabla)$ coincides with that of $\xi$. Vice versa, given a flat vector bundle $(V,\nabla)$, we can construct the associate flat principal $G$-bundle $\xi$ using the same locally constant transition functions as $V$, so that $\xi$ has the same monodromy as $V$. This establishes a correspondence \[ \xymatrix@R=0in{ \Flat(\dot{S},\KK^N)\ar@/^0.5pc/[rr] && \ar@/^0.5pc/[ll] \Flat(\dot{S},\GL_N(\KK)). } \] where $\Flat(\dot{S},\KK^N)$ is the set of isomorphism classes of flat vector bundles of rank $N$. An analogous correspondence holds for framed flat bundle, or for flat bundles with monodromy at the punctures in prescribed conjugacy classes. For $G=\SL_N(\KK)$, the correspondence is between flat principal $\SL_N$-bundles and flat vector bundles $V$ of rank $N$ endowed with a trivialization of their determinant $\det(V)=\Lambda^N V$. Similarly, flat $\PGL_N$-bundles correspond to flat $\KK\PP^{N-1}$-bundles. Such a $\KK\PP^{N-1}$-bundle $\PP$ need not be a projectivization of a flat vector bundle, since its monodromy need not lift to $\GL_N(\KK)$. In the real case, $\PSL_{2N+1}(\RR)=\PGL_{2N+1}(\RR)$; whereas flat $\PSL_{2N}(\RR)$-bundles correspond to flat orientable $\RR\PP^{2N-1}$-bundles. \subsection{The case of $\PSL_2$} Assume now that $G/Z=\PSL_2(\CC)$, namely that $G=\SL_2(\CC),\PSL_2(\CC)$, and so $\gfrak=\mathfrak{sl}_2(\CC)$. Then to each flat principal $G$-bundle $\xi\rar\dot{S}$ we can associate a flat $\CC\PP^1$-bundle $\PP:=\xi\times_G \CC\PP^1$ on $\dot{S}$. \begin{remark} The image of $\rho$ is Zariski-dense if and only if $\rho$ is irreducible, namely if and only if no point of $\CC\PP^1$ is fixed by $\rho(\pi)$. The same holds for $G/Z=\PSU_2\subset \PSL_2(\CC)$. In case $G/Z=\PSL_2(\RR)$, such Zariski-density can be expressed in terms of non-existence of fixed points in $\ol{\HH}$. \end{remark} An easy consequence of Corollary \ref{cor:relative-smoothness} is the following. \begin{corollary}[Smoothness for $\PSL_2$]\label{cor:smoothness-psl2} Let $G/Z$ be $\PSL_2(\CC),\ \PSU_2,\ \PSL_2(\RR)$ and let $\bCo=(\co_1,\dots,\co_n)$ an $n$-uple of conjugacy classes of elements of $G$. \begin{itemize} \item[(a)] If $G/Z=\PSU_2$, then $\Rep_b^{r}(\dot{S},G,\bCo)$ and $\Rep^{r}(\dot{S},G,\bCo)$ are smooth orbifolds. \item[(b)] If $G/Z=\PSL_2(\CC),\PSL_2(\RR)$, then $\rho\in \Rep_b^{r}(\dot{S},G,\ol{\bCo})$ and $[\rho]\in\Rep^{r}(\dot{S},G,\ol{\bCo})$ are singular points if and only if there exists $i$ such that $\rho(\pa_i)\in Z$ and $\co_i$ consists of non-central unipotent elements. \end{itemize} \end{corollary} Condition (b) in the above lemma can be easily understood by remembering that the only non-closed conjugacy class in $\PSL_2(\CC)$ is the class $\co$ of non-trivial unipotent elements, whose closure contains the identity as a singular point of $\ol{\co}$. \subsection{Euler number of $\PSL_2(\RR)$-representations} Let $\wti{\PSL}_2(\RR)$ be the universal cover of $\PSL_2(\RR)$ and let $\ZZ \cdot\zeta\subset\wti{\PSL}_2(\RR)$ be its center, where $\zeta=\exp(R)$ and \[ R=\pi\left( \begin{array}{cc} 0 & -1\\ 1 & 0 \end{array} \right)\in\psl_2(\RR) \] is an infinitesimal generator of the subgroup of (counterclockwise) rotations that fix $i\in\HH$. We define the {\it{rotation number}} $\rot:\wti{\PSL}_2(\RR)\rar\RR$ as follows. If $\ti{g}\in\wti{\PSL}_2(\RR)$ is elliptic, then $\ti{g}$ is conjugate to $\exp(r\cdot R)$ for a unique $r\in\RR$. In this case, we define $\rot(\ti{g}):=r$, so that $\rot(\ti{g})\in \ZZ \iff \ti{g}\in \ZZ\cdot\zeta$. If $\ti{g}\in\wti{\PSL}_2(\RR)$ is not elliptic, then there exists a unique $r\in\ZZ$ such that $\ti{g}$ can be connected to $r\cdot\zeta$ through a continuous path of non-elliptic elements. In this case, we define $\rot(\ti{g}):=r$. Define also a ``fractional'' rotation number as $\{\rot\}:\PSL_2(\RR)\rar [0,1)$ by requiring that $\rot(\ti{g})-\{\rot\}(g)\in\ZZ$, where $\ti{g}$ is any lift of $g$ to $\wti{\PSL}_2(\RR)$. It can be easily seen that $\rot$ and $\{\rot\}$ are invariant under conjugation by elements of $\PSL_2(\RR)$ and that the rotation number $\rot$ is continuous but $\{\rot\}$ is not (see also \cite{burger-iozzi-wienhard:maximal} for more properties of the rotation number). Here we adopt a result by Burger-Iozzi-Wienhard \cite{burger-iozzi-wienhard:maximal} as a definition of Euler number for a representation $\rho:\pi_1(\dot{S})\rar\PSL_2(\RR)$. \begin{definition} Assume $n>0$. The {\it{Euler number of $\rho:\pi\rar\PSL_2(\RR)$}} is \[ \Eu(\rho):=-\sum_{i=1}^n r_i\in\RR \] where $\ti{\rho}:\pi\rar\wti{\PSL}_2(\RR)$ is any lift of $\rho$ and $r_i:=\rot(\ti{\rho}(\gamma_i))\in \RR$. \end{definition} Notice that $\Eu(\rho)+\NO{\bm{\{r\}}}\in \ZZ$, where $\{r_i\}:=\{\rot\}(\rho(\gamma_i))\in[0,1)$. \begin{remark}\label{rmk:identity} If a representation is obtained as a composition $\rho':\pi_1(S\setminus\{p_1,\dots,p_n\},b)\rar\pi_1(S\setminus\{p_1,\dots,p_k\},b)\arr{\rho}\PSL_2(\RR)$ with $0\leq k<n$, then $\{r_{k+1}\}=\dots=\{r_n\}=0$ and $\Eu(\rho')=\Eu(\rho)$. This allows to coherently define the Euler number in the case of an unpunctured surface. \end{remark} \begin{remark} Suppose that $[\rho]$ is a singular point of $\Rep(\dot{S},\PSL_2(\RR))$. Then $\rho$ must fix a point of $\RR\PP^1$ or it must be Abelian. In the former case, there exists a lift $\ti{\rho}$ whose action on the universal cover $\wti{\RR\PP^1}$ fixes a point. Hence, $\rot\circ\ti{\rho}=0$ and so $\Eu(\rho)=0$ and $\{r_j\}=0$ for all $j$. In the latter case, $\rho$ Abelian implies $\Eu(\rho)=0$ and so $\NO{\bm{\{r\}}}\in\ZZ$. \end{remark} Being invariant under conjugation by elements of $\PSL_2(\RR)$, the Euler number descends to a continuous map \[ \Eu:\Rep(\dot{S},\PSL_2(\RR))\lra \RR \] and we call $\Rep(\dot{S},\PSL_2(\RR))_e$ the preimage $\Eu^{-1}(e)$. If $n=0$, then $\Eu$ is integral and so it is constant on connected components of $\Rep(S,\PSL_2(\RR))$. Moreover, the conjugation by an element in $\PGL_2(\RR)\setminus\PSL_2(\RR)$ induces the isomorphism $\Rep(S,\PSL_2(\RR))_e\cong \Rep(S,\PSL_2(\RR))_{-e}$. For closed surfaces the topology of $\Rep(S,\PSL_2(\RR))_e$ with $e\neq 0$ is completely determined. \begin{theorem}[Topology of representation spaces of closed surfaces in $\PSL_2(\RR)$]\label{thm:rep-closed} Assume $n=0$ and $g(S)\geq 2$. \begin{itemize} \item[(a)] Every $\rho\in\Rep_b(S,\PSL_2(\RR))$ satisfies $|\Eu(\rho)|\leq -\chi(S)$ (Milnor \cite{milnor:inequality}, Wood \cite{wood:milnor}). If $\Eu(\rho)=-\chi(S)$, then $\rho$ is the monodromy of a hyperbolic metric (Goldman \cite{goldman:components}). \item[(b)] $\Rep(S,\PSL_2(\RR))_e\neq\emptyset$ if and only if \[ e\in \ZZ\cap \Big[ \chi(S),\,-\chi(S)\,\Big] \] and, in this case, it is also connected (Goldman \cite{goldman:components}). \item[(c)] If $e\neq 0$, then $\Rep(S,\PSL_2(\RR))_e$ is smooth.\\ For $e>0$ the manifold $\Rep(S,\PSL_2(\RR))_e$ is real-analytically diffeomorphic to a complex vector bundle of rank $-\frac{3}{2}\chi(S)-m$ over $\Sym^m(S)$, where $m=-\chi(S)-e$ (Hitchin \cite{hitchin:self-duality}). \end{itemize} \end{theorem} Let now $n>0$. Given an $n$-uple $\bCo=(\bco_1,\dots,\bco_n)$ of conjugacy classes in $\PSL_2(\RR)$, we call $\{r_i\}$ the rotation number of any element in $\bco_i$. Moreover, we denote by $P_{hyp}$ (resp. $P_{ell}$, $P_+$, $P_-$, $P_0$) the subset of points $p_i\in P$ such that $\bco_i$ is hyperbolic (resp. elliptic, positive unipotent, negative unipotent, the identity) and we let $s_-=\# P_-$ (resp. $s_0=\# P_0$). The restriction \[ \Eu_{\bCo}:\Rep(\dot{S},\PSL_2(\RR),\bCo)\lra\RR \] of $\Eu$ has the property that $\Eu_{\bCo}+\NO{\bm{\{r\}}}\in \ZZ$, where $\NO{\bm{\{r\}}}$ only depends on $\bCo$. Thus, $\Eu_{\bCo}$ is also constant on connected components of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$. As above, we denote by $\Rep(\dot{S},\PSL_2(\RR),\bCo)_e$ the preimage $\Eu_{\bCo}^{-1}(e)$ and we observe that the conjugation by an element of $\PGL_2(\RR)\setminus \PSL_2(\RR)$ induces the isomorphism $\Rep(\dot{S},\PSL_2(\RR),\bCo)_e \cong \Rep(\dot{S},\PSL_2(\RR),\bCo^{-1})_{-e}$, where $\bco_i^{-1}=\{g^{-1}\in\PSL_2(\RR)\,|\,g\in\bco_i\}$. In analogy with Theorem \ref{thm:rep-closed} in the closed case, the following result holds in the punctured case. \begin{theorem}[Topology of relative representation spaces of punctured surfaces in $\PSL_2(\RR)$]\label{thm:rep-punctured} Assume $n>0$ and let $\bCo$ and $\bm{\{r\}}$ be as above. \begin{itemize} \item[(a)] The image of $\Eu:\Rep(\dot{S},\PSL_2(\RR))\rar\RR$ is the interval $\big[\chi(\dot{S}),-\chi(\dot{S})\big]$. If $\Eu(\rho)=-\chi(\dot{S})$, then all $\rho(\pa_i)$'s are hyperbolic or positive unipotent elements for all $i$ and $\rho$ is the monodromy of a hyperbolic metric with geodesic boundary components and cusps (Burger-Iozzi-Wienhard \cite{burger-iozzi-wienhard:maximal}). \item[(b)] Assume $e>0$ and fix conjugacy classes $\bco_1,\dots,\bco_n\subset\PSL_2(\RR)$.\\ Then $\Rep(\dot{S},\PSL_2(\RR),\bCo)_e\neq\emptyset$ if and only if \[ e+\NO{\bm{\{r\}}} +s_0+s_-\in \ZZ\cap \left( 0,-\chi(\dot{S})\right] \] and, in this case, it is smooth and connected. \end{itemize} Assume now that $e$ satisfies the hypotheses in (b). \begin{itemize} \item[(c)] The component $\Rep(\dot{S},\PSL_2(\RR),\bCo)_e$ is real-analytically diffeomorphic to the complement of $s_-$ affine subbundles of codimension $1$ inside a holomorphic affine bundle of rank $-\frac{3}{2}\chi(S)+n-m+s_-$ over $\Sym^{m-(s_0+s_-)}\Big(S\setminus (P_{hyp}\cup P_+)\Big)$, where $m=-\chi(\dot{S})-e-\NO{\bm{\{r\}}}$. \item[(d)] The locus $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e$ is homeomorphic to an affine holomorphic bundle of rank $-\frac{3}{2}\chi(S)+n-m+s_-$ over $\Sym^{m-(s_0+s_-)}(S\setminus P_{hyp})$. \item[(e)] If $c_i$ is the class of positive unipotents and $\bCo^0$ is obtained from $\bCo$ by replacing $c_i$ with $\{\id\}$, then $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo}^0)_e$ includes in $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e$ as the preimage over $p_i+\Sym^{m-(s_0+s_-)-1}(S\setminus P_{hyp})\subset \Sym^{m-(s_0+s_-)}(S\setminus P_{hyp})$.\\ If $c_i$ is the class of negative unipotents and $\bCo^0$ is obtained from $\bCo$ by replacing $c_i$ with $\{\id\}$, then $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo}^0)_e$ includes in $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e$ as an affine subbundle over $\Sym^{m-(s_0+s_-)}(S\setminus P_{hyp})$ of codimension $1$. \end{itemize} \end{theorem} Claims (c-d-e) are consequence of Theorem \ref{thm:correspondence}, which in turn relies on Proposition \ref{prop:topology} and Corollary \ref{cor:quotient-topology}. Claim (b) easily follows from (c-d-e). The case of some $\bco_i=\{\id\}$ can be also dealt with as in Remark \ref{rmk:identity}. \subsection{Hyperbolic metrics} Let $\bm{\ell}=(\ell_1,\dots,\ell_n)$ with $\ell_i=\sqrt{-1}\th_i$ and $\th_i>0$ for $i=1,\dots,k$ and $\ell_i\geq 0$ for $i=k+1,\dots,n$. We are interested in isotopy classes of {\it{hyperbolic metrics of boundary type $\bm{\ell}$}} on $\dot{S}$, i.e. metrics on curvature $-1$ on $\dot{S}$, whose completion has a conical singularity of angle $\th_i$ at $p_i$ for $i=1,\dots,k$ and a boundary component of length $\ell_i$ (resp. a cusp if $\ell_i=0$) instead of the puncture $p_i$ for $i=k+1,\dots,n$. We assume that the quantity \[ e_{\bm{\ell}}:=-\chi(\dot{S})-\sum_{i=1}^k\frac{\th_i}{2\pi}= -\chi(S)-\sum_{i=1}^k\left(\frac{\th_i}{2\pi}-1\right) \] is positive, since hyperbolic metrics of boundary type $\bm{\ell}$ have total area $e_{\bm{\ell}}>0$ by Gauss-Bonnet. In this case, denote by $\Y(\dot{S},\bm{\ell})$ the space of isotopy classes of hyperbolic metrics on $\dot{S}$ of boundary type $\bm{\ell}$. Surfaces of curvature $-1$ are locally isometric to portions of the hyperbolic plane $\HH$. Given a metric $h$ of curvature $-1$ on $\dot{S}$, consider the pull-back $\ti{h}$ on $\ti{\dot{S}}$. Since $\ti{\dot{S}}$ is simply-connected, local isometries into $\HH$ glue to give a global {\it{developing map}} $\dev_h:\ti{\dot{S}}\rar\HH$, which is a local isometry. Moreover, $\pi$ acts on $\HH$ via a {\it{monodromy}} homomorphism $\rho_h:\pi\rar\Iso_+(\HH)\cong\PSL_2(\RR)$ and $\dev_h$ is $\pi$-equivariant. Notice that $\dev_h$ is well-defined up to post-composition with an isometry of $\HH$, and so also $\rho_h$ is well-defined only as an element of $\Rep(\dot{S},\PSL_2(\RR))$. We also observe that $\rho_h$ arises as a monodromy of a flat principal $\PSL_2(\RR)$-bundle as follows. Pull the trivial $\PSL_2(\RR)$-bundle over $\HH$ back via $\dev_h$ to a (trivializable) $\ti{\xi}_h\rar\ti{\dot{S}}$. By $\rho_h$-equivariance, it descends to a flat principal $\PSL_2(\RR)$-bundle $\xi_h\rar\dot{S}$ with $\hol_{\xi_h}=\rho_h$. Then we have a composition of real-analytic maps \[ \xymatrix@R=0in{ \Y(\dot{S},\bm{\ell})\ar[r]^{\Xi_{\bm{\ell}}\qquad} & \Flat(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}}) \ar[r]^{\hol\quad} & \Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})\\ h \ar@{|->}[r] & \xi_h \ar@{|->}[r] & \rho_h } \] where the correspondence between $\ell_i$ and $\bco_i$ is dictated by the following table \[ \def\arraystretch{1.3}\begin{array}{|c|c|} \hline \text{conjugacy class $\bco_\ell$} & \ell\\ \hline \id & 2\pi\sqrt{-1}\cdot\NN_+\\ \text{positive unipotents} & 0\\ \text{hyperbolics $g$ with $|\tr(g)|=2|\cosh(\ell/2)|$} & \RR_+\\ \text{elliptics $g$ with $\{\rot\}(g)=\{\ell/(2\pi\sqrt{-1})\}$} & 2\pi\sqrt{-1}\cdot (\RR_+\setminus\NN_+)\\ \hline \end{array} \] In fact, a cusp at $p_i$ corresponds to positive unipotent monodromy along $\gamma_i$. \begin{proposition}[Uniformization components]\label{prop:uniformization} Let $\th_1,\dots,\th_k>0$ and $\ell_{k+1},\dots,\ell_n\geq 0$ and call $\bm{\ell}=(\sqrt{-1}\th_1,\dots,\sqrt{-1}\th_k,\ell_{k+1},\dots,\ell_n)$. If $e_{\bm{\ell}}>0$, then the image of $\hol\circ\Xi_{\bm{\ell}}$ is contained inside $\Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}}$. \end{proposition} \begin{proof} After the above discussion, it is enough to notice that the Euler number $\Eu$ can be identified to the Toledo invariant (see for instance \cite{burger-iozzi-wienhard:maximal}) and so $\Eu(\hol_h)=\frac{1}{2\pi}\mathrm{Area}(h)$. \end{proof} \subsubsection{Questions} Still little is known in general about the image of \[ \hol\circ\Xi_{\bm{\ell}}:\Ycal(\dot{S},\bm{\ell})\lra \Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}} \] beside the fact that it is open for the classical topology and so its complement $\mathcal{C}_{\bm{\ell}}$ is closed. \begin{question}\label{q:surjective} For which $\bm{\ell}$ is the image of $\hol\circ\Xi_{\bm{\ell}}$ the whole $\Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}}$? \end{question} \begin{question}\label{q:zero} For which $\bm{\ell}$ is $\mathcal{C}_{\bm{\ell}}$ of zero measure? \end{question} \begin{question} For which $\bm{\ell}$ is $\mathcal{C}_{\bm{\ell}}$ a countable union of proper (semi-)algebraic subsets with no internal part? \end{question} \begin{question}\label{q:ergodic} For which $\bm{\ell}$ does the mapping class group of $(S,P)$ act ergodically on $\Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}}$? \end{question} Question \ref{q:surjective} has affirmative answer if all the angles are smaller than $\pi$. Roughly speaking, a possible argument is as follows.\\ Since the angles are smaller than $\pi$, simple closed geodesics on $\dot{S}$ avoid the conical points and so their lengths are detected by the traces of the monodromy. Moreover, any pair of pants decomposition of $\dot{S}$ gives rise to Fenchel-Nielsen coordinates (and so $\Ycal(\dot{S},\bm{\ell})$ is diffeomorphic to $(\RR_+\times\RR)^{3g-3+n}$) and this quickly leads to the injectivity of $\hol\circ\Xi_{\bm{\ell}}$. The collar lemma allows to prove properness, and one concludes that $\hol\circ\Xi_{\bm{\ell}}$ is indeed a diffeomorphism. In this case, the action of the mapping class group is properly discontinuous and with finite stabilizers, and so very far from being ergodic. Concerning Question \ref{q:ergodic}, Goldman has conjectured that the mapping class group of a closed surface $S$ acts ergodically on the components $\Rep(S,\PSL_2(\RR))_e$ with $e\neq 0,\,\pm\chi(S)$. A positive answer to Question \ref{q:ergodic} would immediately imply a positive answer to Question \ref{q:zero}. \section{Parabolic Higgs bundles}\label{sec:parabolic} In this section, $S$ will denote a compact connected Riemann surface endowed with complex structure $I$. \subsection{Parabolic structures} Let $E$ be a holomorphic vector bundle on $S$, which we constantly identify with the locally-free sheaf of its sections. \begin{definition} Let $E$ be a holomorphic vector bundle on $S$. A {\it{parabolic structure}} on $E$ over $(S,P)$ is a filtration $\RR\ni w\mapsto E_w \subset E(\infty \cdot P)$ of the sheaf $E(\infty\cdot P)$ of sections which are meromorphic at $P$ such that \begin{itemize} \item[(a)] $E_{w}\supseteq E_{w'}$ if $w\leq w'$ {\it{(decreasing)}} \item[(b)] for every $w\in\RR$ there exists $\e>0$ such that $E_{w-\e}=E_{w}$ ({\it{left-continuous}}) \item[(c)] $E_{0}=E$ and $E_{w+1}=E_{w}(-P)$ ({\it{normalized}}). \end{itemize} We will denote by $E_\bullet$ the datum of the bundle $E$ and the given parabolic structure. \end{definition} \begin{notation} Denote by $E_{p_i}$ the space of regular germs at $p_i$ of sections of $E$ and by $E_{p_i}(\infty\cdot p_i)$ the space of germs at $p_i$ of sections of $E$ which are meromorphic at $p_i$. Given a parabolic structure on $E$, the induced filtration on $E_{p_i}(\infty\cdot p_i)$ is denoted by $w\mapsto E_{p_i,w}$. \end{notation} By definition, the {\it{jumps}} in the filtration at $p_i$ occur at those weights $w$ such that $E_{p_i,w}\supsetneq E_{p_i,w+\e}$ for all $\e>0$. Thus, a parabolic structure is equivalent to the datum of a weights \[ 0\leq w_1(p_i)<w_2(p_i)<\dots<w_{b_i}(p_i)< w_{b_i+1}(p_i)=1 \] for some $b_i\in[1,\rk(E)+1]$ and a filtration \[ E_{p_i}=E_{p_i,w_1(p_i)} \supsetneq\dots\supsetneq E_{p_i,w_{b_i}(p_i)}\supsetneq E_{p_i,w_{b_i+1}(p_i)}=E_{p_i}(-p_i) \] for each $p_i\in P$. \begin{notation} We use the symbol $\bm{w}$ to denote the collection of $(w_k(p_i),m_k(p_i))_{i=1}^n$, where $m_k(p_i)=\dim(E_{p_i,w_k(p_i)}/E_{p_i,w_{k+1}(p_i)})$ and we will say that the parabolic bundle $E_\bullet$ is of {\it{type $\bm{w}$}}. We will also write $\NO{\bm{w}}=(\NO{w(p_1)},\dots,\NO{w(p_n)})$, where $\NO{w(p_i)}=\sum_{k=1}^{b_i} m_k(p_i)w_k(p_i)$, and we will say that $\bm{w}$ is {\it{integral}} if $\NO{\bm{w}}\in\NN^n$. \end{notation} The {\it{(parabolic) degree}} of $E_\bullet$ is defined as $\dis \deg(E_\bullet):=\deg(E)+\sum_{p_i\in P}\NO{w(p_i)} $.\\ Every holomorphic bundle $E$ can be endowed with a {\it{trivial parabolic structure}}, by choosing $b_i=1$ and $w_1(p_i)=0$ for all $i=1,\dots,n$. This provides an embedding on the category of holomorphic bundles on $S$ inside the category of parabolic bundles on $S$.\\ Direct sums, homomorphisms and tensor products of parabolic bundles are defined as \begin{align*} (E\oplus E')_{p_i,w} & :=E_{p_i,w}\oplus E'_{p_i,w}\\ \Hom(E_\bullet,E'_\bullet) & :=\left\{ f\in\Hom(E,E') \ \Big| \ w_j(p_i)>w'_k(p_i)\implies f(E_{p_i,w_j(p_i)})\subseteq E'_{p_i,w'_{k+1}(p_i)} \right\}\\ (E\otimes E')_{p_i,w''}& :=\left(\bigcup_{w+w'=w''}E_{p_i,w}\otimes E'_{p_i,w'}\right)\subset (E\otimes E')_{p_i}(\infty \cdot p_i). \end{align*} and we will write $\bm{w}\otimes\bm{w'}$ for the type of a parabolic bundle $E_\bullet\otimes E'_\bullet$ obtained by tensoring $E_\bullet$ of type $\bm{w}$ with $E'_\bullet$ of type $\bm{w'}$. It is also possible to define a $\mathcal{H}om$-sheaf just by letting $\mathcal{H}om(E_\bullet,E'_\bullet)(U):=\Hom\left(E_\bullet|_U,E'_{\bullet}|_U\right)$ with $\mathcal{H}om(E_\bullet,E'_\bullet)_{p_i,w}=\{\text{germs at $p_i$ of morphisms $E_\bullet\rar E'_{\bullet+w}$}\}$ and also a dual $E_\bullet^\vee:=\mathcal{H}om(E_\bullet,\Ocal_S)$. We will say that a homomorphism is {\it{injective}} if it is so as a morphism of sheaves, namely if it is injective at the general point of $S$, and {\it{properly injective}} if it is injective but not an isomorphism. A {\it{parabolic sub-bundle}} of $E_\bullet$ is just a sub-bundle $F\subseteq E$, endowed with the induced filtration $F_w:=F\cap E_w$; the quotient bundle $E/F$ can be also endowed with a natural parabolic structure by letting $(E/F)_w$ be the image of $E_w$ under the natural projection $E\rar E/F$.\\ Since $H^0(U,E_\bullet)=\Hom(\Ocal_U,E_\bullet|_U)=\Hom(\Ocal_U,E|_U)=H^0(U,E)$, sections of $E_\bullet$ are sections of $E$ and so the same holds for higher cohomology groups.\\ In order to understand parabolic structures, it is enough to localize the analysis and consider bundles on a disk with parabolic structure at the origin. The typical setting is the following. \begin{example}[Flat vector bundles on a punctured disk]\label{example:disk} Let $N>0$ be an integer and let $\dot{\Delta}=\Delta\setminus\{p\}$ with $p=0$. Let $\HH\rar\dot{\Delta}$ be the universal cover, defined as $u\mapsto z=\exp(2\pi \sqrt{-1}\cdot u)$, and let $b\in\dot{\Delta}$ be a base-point. Call $\wti{V}:=\HH\times\CC^N\rar \HH$ the trivial vector bundle and endow it with the natural connection $\wti{\nabla}$ that can be expressed as $[\wti{\nabla}]_{\wti{\Vcal}}=d$ with respect to the canonical basis $\wti{\Vcal}=\{\ti{v}_1,\dots,\ti{v}_n\}$ of sections of $\wti{V}$ and a natural holomorphic structure $\ol{\pa}^{\wti{V}}$.\\ Given $T=\exp(-2\pi\sqrt{-1}\cdot M)\in\GL_N(\CC)$, one can lift the natural action of $\pi_1(\dot{\Delta},b)=\langle\gamma\rangle$ on $\HH$ to an action on $\wti{V}$ by letting $\gamma\cdot(u,v)=(u+1,T(v))$. The induced bundle $\dot{V}:=\wti{V}/\pi_1(\dot{\Delta},b)$ inherits a flat connection $\nabla$ that can be written as $[\nabla]_{\Vcal}=d$ with respect to the basis $\Vcal=\{v_1,\dots,v_N\}$ of flat $\ol{\pa}^V\!-$holomorphic multi-sections of $\dot{V}$. Moreover, chosen the standard determination of $\log(z)$ on $\dot{\Delta}$, \[ \left( \begin{array}{c} v'_1\\ \vdots\\ v'_N\end{array} \right):= \exp\Big(\log(z)\cdot M\Big) \left( \begin{array}{c} v_1\\ \vdots\\ v_N\end{array} \right) \] defines a basis $\Vcal'=\{v'_1,\dots,v'_N\}$ of univalent $\ol{\pa}^V\!-$holomorphic sections of $\dot{V}$ such that $[\nabla]_{\Vcal'}=d+M\frac{dz}{z}$. Notice that $\nabla$ and so $\Res_p(\nabla)\in\End(V|_p)$ are not uniquely defined by $T$, since $\exp(-2\pi\sqrt{-1}\,\bullet):\gl_N(\CC)\rar \GL_N(\CC)$ is not injective: in particular, the eigenvalues of $\Res_p(\nabla)$ are only well-defined in $\CC/\ZZ$. \end{example} \subsubsection{Rank $1$} A parabolic structure at $P$ on a line bundle $L\rar S$ is just the datum of a weight $w(p_i)\in [0,1)$ for each $p_i\in P$, so that it makes sense to write $L_\bullet=L(\sum_i w(p_i)p_i)$. The parabolic degree of $L_\bullet$ is simply $\deg(L_\bullet)=\deg(L)+\sum_i w(p_i)$. \begin{notation} If $L_\bullet=L(\sum_i w(p_i)p_i)$ is a parabolic line bundle, we denote by $\floor{L_\bullet}$ its {\it{integral part}}, namely the underlying line bundle $L$ with trivial parabolic structure and by $\{L_\bullet\}:=L_\bullet \otimes \floor{L_\bullet}^\vee=\Ocal_S(\sum_i w(p_i)p_i)$ its {\it{fractional part}}. \end{notation} We incidentally remark that integral parabolic structures in rank $1$ are trivial. \begin{example}[Unitary line bundles on a punctured disk]\label{example:u(1)} Keep the notation as in Example \ref{example:disk} and let $N=1$, $v=v_1$ and $T=\exp(-2\pi\sqrt{-1}\cdot\lambda)\in \U_1$ with $\lambda\in[0,1)$. With respect to $\Vcal'=\{v'\}$, the flat connection \[ [\nabla]_{\Vcal'}=d+\lambda\frac{dz}{z} \] on $\dot{V}$ with $[\Res_p(\nabla)]_{\Vcal'}=\lambda$ has monodromy $\langle T\rangle$ and $v=z^{-\lambda}v'$ is a flat holomorphic multi-section of $\dot{V}$. Let $V\rar\Delta$ be the extension of $\dot{V}$ defined by requiring that $v'$ is a generator. We can put on $V$ a $\nabla$-invariant metric $H$ by prescribing that $\|v\|_H=c>0$ is constant, namely \[ \|v'\|_H:=c|z|^{\lambda} \] and so such an invariant norm has a zero of order $\lambda\in[0,1)$ at $0$. We can then put on $E:=V$ the same holomorphic structure as $V$ and take $e=v'$ as a holomorphic generator of $E$, and define the filtration $E_w$ by \[ H^0(U,E_{w}) = \left\{s\in H^0(U,E)\ \Big|\ \begin{array}{c} \text{on every neighbourhood $U'\Subset U$ of $p$}\\ \text{$|z|^{\e-w}\|s\|_H$ is bounded for all $\e>0$}\end{array} \right\} \] so that $H^0(U,E)=H^0(U,E_w)$ for every $w\in(-1+\lambda,\lambda]$ and the jumps occur at weights in $\lambda+\ZZ$. \end{example} The determinant $\det(E_\bullet)$ of a parabolic bundle $E_\bullet$ of rank $N$ is the parabolic line bundle defined in the standard way as a quotient of $E_\bullet^{\otimes N}$ by the alternating action of $\mathfrak{S}_N$. If $E_\bullet$ is of type $\bm{w}$, then \[ \det(E_\bullet)\cong\det(E)\otimes\left(\bigotimes_{i=1}^n \Ocal_S\big(\NO{w(p_i)}p_i\big)\right) \] In particular, if $\bm{w}$ is integral, then $\det(E_\bullet)$ has trivial parabolic structure. \subsubsection{Rank $2$ of integral type} Among all parabolic bundles of rank $2$ we focus on those of integral type because of their relation with $\SL_2$-bundles. \begin{remark} A parabolic structure $E_\bullet$ on a bundle $E$ of rank $2$ on $(S,P)$ is of integral type $\bm{w}$ (or, briefly, just {\it{integral}}) if and only if the following condition holds for every $p_i\in P$: \begin{itemize} \item[(1)] either $b_i=1$ ({\it{degenerate case}}) and $w_1(p_i)\in\left\{0,\frac{1}{2}\right\}$; \item[(2)] or $b_i=2$ ({\it{non-degenerate case}}) and $\left(0,\frac{1}{2}\right)\ni w_1(p_i)<w_2(p_i)=1-w_1(p_i)\in \left(\frac{1}{2},1\right)$. \end{itemize} If $E_\bullet$ is integral of rank $2$, then $\deg(E_\bullet)=\deg(E)+\#\{p_i\in P\,|\, w_1(p_i)>0\}$. \end{remark} We remark that, if $E_\bullet$ is non-degenerate at $p_i$ (i.e. $b_i=2$), then giving $E_{p_i,w_2(p_i)}$ is equivalent to giving a line $L_i\subset E|_{p_i}$. Indeed, knowing $E_{p_i,w_2(p_i)}$, the line $L_i$ can be recovered as the kernel of $E|_{p_i}\rar \left(E_{p_i}/E_{p_i,w_2(p_i)}\right)$. Vice versa, given $L_i$, the germ $E_{p_i,w_2(p_i)}$ is the kernel of $E_{p_i}\rar E|_{p_i}/L_i$. \begin{example}[Type of parabolic line sub-bundle]\label{example:sub-bundle} Let $E_\bullet$ be an integral parabolic bundle of rank $2$ and let $F\subset E$ be a sub-bundle of rank $1$. Then the jump of the induced parabolic structure on $F$ at $p_i$ occurs at $w_F(p_i)$, where \[ \begin{array}{|c|c|c|} \hline & \text{$E_\bullet$ degenerate at $p_i$} & \text{$E_\bullet$ non-degenerate at $p_i$}\\ \hline w_F(p_i) & w_1(p_i) & \begin{array}{cl} w_1(p_i) & \text{if $F|_{p_i}\neq L_i$}\\ 1-w_1(p_i) & \text{if $F|_{p_i}=L_i$} \end{array} \\ \hline \end{array} \] \end{example} The following example illustrates how parabolic structures on holomorphic vector bundles arise from unitary representations: since the argument is local, we will only deal with the case of a bundle over a disk. A complete treatment can be found in \cite{mehta-seshadri:parabolic}. \begin{example}[Rank $2$ special unitary vector bundles on a punctured disk]\label{example:disk-rank2} Keep the notation as in Example \ref{example:disk} and let $N=2$ and $T\in \SU_2$. Up to conjugation, we can assume that $T$ is diagonal and that $T(v_1)=\exp(-2\pi\sqrt{-1}\cdot \lambda)v_1$ and $T(v_2)=\exp(2\pi \sqrt{-1}\cdot \lambda)v_2$ with $\lambda\in[0,1)$, and so $T$ acts on $\CC\PP^1$ as a positive rotation of angle $4\pi\lambda$ that fixes $[1:0]$. The connection $\nabla$ on $\dot{V}$ defined as \[ [\nabla]_{\Vcal'}=d+\left(\begin{array}{cc}\lambda & 0\\ 0 & -\lambda\end{array}\right)\frac{dz}{z} \quad \text{with} \ [\Res_p(\nabla)]_{\Vcal'}=\left(\begin{array}{cc}\lambda & 0\\ 0 & -\lambda\end{array}\right) \] with respect to the basis $\Vcal'=\{v'_1=z^\lambda v_1,\ v'_2=z^{-\lambda}v_2\}$ has monodromy $\langle T\rangle$. Put on $\dot{V}$ a $\nabla$-invariant metric $H$ by prescribing that $v_1,v_2$ are orthogonal with $\|v_i\|_H=c_i>0$ and define the univalent sections $e_1,e_2$ of $\dot{V}$ as \[ e_1=v'_1, \qquad e_2= \begin{cases} v'_2 & \text{if $\lambda=0$}\\ z\cdot v'_2 & \text{if $\lambda>0$} \end{cases} \] and extend $\dot{V}$ to a vector bundle $V\rar\Delta$ with generators $e_1,\ e_2$. The holomorphic vector bundle $E:=V$ endowed with the same holomorphic structure as $V$ has a pointwise orthogonal basis $\Ecal=\{e_1,e_2\}$ of holomorphic sections, that satisfy \[ \|e_1\|_H=c_1|z|^{\lambda}, \qquad \|e_2\|_H= \begin{cases} c_2 & \text{if $\lambda=0$}\\ c_2|z|^{1-\lambda} & \text{if $\lambda>0$} \end{cases} \] If $\lambda=0$, then $\|e_i\|_H=c_i$ and we have a degenerate parabolic structure with $w_1=0$. Similarly, if $\lambda=\frac{1}{2}$, then $\|e_i\|_H=c_i |z|^{\frac{1}{2}}$ and $w_1=\frac{1}{2}$. Assume now that $\lambda>0$ but $\lambda\neq\frac{1}{2}$.\\ If $\lambda\in \left(0,\frac{1}{2}\right)$, then we let $w_1=\lambda<w_2=1-\lambda$ and $L=\CC e_2\times\{p\}\subset E|_{p}$. Since $|z|^\lambda>|z|^{1-\lambda}$, a section $s=f_1(z)e_1+f_2(z)e_2$ satisfies \[ \ord_{p}\|s\|_H= \ord_{p}\left(|f_1 z^\lambda|+|f_2 z^{1-\lambda}|\right)= \begin{cases} w_1=\lambda & \text{if $f_1(p)\neq 0$, i.e. if $s(p)\notin L$}\\ w_2=1-\lambda & \text{if $f_1(p)=0$, i.e. if $s(p)\in L$.} \end{cases} \] If $\lambda\in \left(\frac{1}{2},1\right)$, then we let $w_1=1-\lambda<w_2=\lambda$ and $L=\CC e_1\times\{p\}\subset E_p$. In both cases, $E_{w}$ is defined by \[ H^0(U,E_{w}) = \{s\in H^0(U,E)\,|\,\|s\|_H\cdot |z|^{\e-w}\ \text{bounded near $p$ for all $\e>0$}\} \] and in particular, $H^0(U,E_{w_2})=\{s\in H^0(U,E)\,|\,s(p)\in L\}$ and \[ H^0(U,E_w)= \begin{cases} H^0(U,E) & \text{for $w\in[0,w_1]$}\\ H^0(U,E_{w_2}) & \text{for $w\in (w_1,w_2]$}\\ H^0(U,E(-p)) & \text{for $w\in(w_2,1)$}. \end{cases} \] Observe that such parabolic structure is integral. \end{example} \subsection{Higgs bundles} \begin{definition} A {\it{parabolic Higgs bundle}} on $(S,P)$ is a couple $(E_\bullet,\Phi)$, where $E_\bullet$ is a holomorphic parabolic vector bundle of rank $N$ and $\Phi\in H^0(S,K(P)\otimes\End(E_\bullet))$. The {\it{residue}} of $\Phi$ at $p_i$ is the induced endomorphism $\Res_{p_i}(\Phi)\in \End(E_\bullet|_{p_i})$ of the filtered vector space $E_\bullet|_{p_i}$. \end{definition} Here is the motivating example in rank $N=1$ on the punctured disk. \begin{example}[Line bundles on a punctured disk]\label{example:rank1-higgs} Keep the notation as in Example \ref{example:disk} and let $N=1$ and $T=\exp[-2\pi \sqrt{-1}(\lambda+i\nu)]\in \GL_1(\CC)=\CC^*$ with $\lambda\in [0,1)$ and $\nu\in\RR$. We can assume that $\nu\neq 0$, as the case $\nu=0$ has already been discussed in Example \ref{example:u(1)}. Such a monodromy $T$ is induced by a flat connection $\nabla$ that can be written as \[ [\nabla]_{\Vcal'}=d+(\lambda+i\nu)\frac{dz}{z} \quad \text{with} \ [\Res_p(\nabla)]_{\Vcal'}=\lambda+i\nu \] with respect to $\Vcal'=\{v'=v'_1\}$. The metric $H$ on $\dot{V}$ defined by \[ \|v'\|_H:=c|z|^\lambda \] is harmonic with respect to $\nabla$, since $i\pa\ol{\pa}\log\|v'\|^2_H=0$. Thus, \begin{align*} [\nabla]_{\Vcal'} =[\nabla^H]_{\Vcal'} +\Phi +\ol{\Phi}= \left(d+\lambda\frac{dz}{z}+i\frac{\nu}{2}\frac{dz}{z}+i\frac{\nu}{2}\frac{d\ol{z}}{\ol{z}}\right) + \left(i\frac{\nu}{2}\frac{dz}{z}\right) +\left(-i\frac{\nu}{2}\frac{d\ol{z}}{\ol{z}}\right) \end{align*} where $\nabla^H$ is a connection on $\dot{V}$ compatible with the metric $H$ and $\Res_p(\Phi)=i\nu/2$. Extend $\dot{V}$ to the bundle $V=\CC v'\times\Delta\rar\Delta$ and put on the complex line bundle $E:=V$ the holomorphic structure given by $\ol{\pa}^E:=\ol{\pa}^V-\ol{\Phi}$, so that $\nabla^H$ is a Chern connection on $(E,\ol{\pa}^E,H)$. Holomorphic sections of $E$ are generated then by $e=\exp(-i\nu\log|z|)v'$. Since $\|e\|_H=\|v'\|_H=c|z|^{\lambda}$, the filtration $E_\bullet$ is as in Example \ref{example:u(1)}, and so that the jumps occur at $\lambda+\ZZ$. \end{example} The following computation is borrowed from \cite{simpson:harmonic}. \begin{example}[Flat $\SL_2$-vector bundles on a punctured disk]\label{example:log} Keep the notation as in Example \ref{example:disk} and let $N=2$ and $T\in \SL_2(\CC)$. If $T$ is diagonalizable, then the bundle $\dot{E}$ splits and we are reduced to the rank $1$ case of Example \ref{example:rank1-higgs}. Thus, up to conjugation, we can assume that \[ \left[T\right]_{\Vcal'}=\frac{1}{2}\left( \begin{array}{cc} 1 & -1\\ 0 & 1 \end{array} \right), \qquad \left[\nabla\right]_{\Vcal'} =d+ \frac{1}{2}\left( \begin{array}{cc} 0 &1\\ 0 & 0 \end{array} \right)\frac{dz}{z} \] with respect to $\Vcal'=\{v'_1,v'_2\}$ and so $[\Res_p(\nabla)]_{\Vcal'}=\frac{1}{2}\left( \begin{array}{cc} 0 &1\\ 0 & 0 \end{array} \right)$. Put on $\dot{V}$ the metric $H$ defined by \[ [H]_{\Vcal'}= \left( \begin{array}{cc} 2|\log|z||^{-1} & -1\\ -1 & |\log|z|| \end{array} \right) \] which is harmonic for $\nabla$ and extend $\dot{V}$ to a complex bundle $V\rar S$ with basis $\Vcal'$. Thus, $\nabla=\nabla^H+\Phi+\ol{\Phi}^H$, where \[ \left[\Phi\right]_{\Vcal'}= \frac{1}{2} \left( \begin{array}{cc} -|\log|z||^{-1} & -1\\ -|\log|z||^{-2} & |\log|z||^{-1} \end{array} \right)\frac{dz}{z}, \quad \left[\ol{\Phi}^H\right]_{\Vcal'}= \frac{1}{2} \left( \begin{array}{cc} 0 & 0\\ |\log|z||^{-2} & 0 \end{array} \right)\frac{d\ol{z}}{\ol{z}}. \] and $\nabla^H$ is compatible with $H$. Let now $\dot{E}:=\dot{V}$ as a complex vector bundle and notice that $\Ecal=\{e_1:=v'_1+\frac{v'_2}{|\log|z||},\ e_2:=v'_2\}$ is a set of generators for $\dot{E}$, which are holomorphic with respect to the operator $\ol{\pa}^E:=\ol{\pa}^V-\ol{\Phi}^H$. Thus, we can extend $\dot{E}$ to a $E\rar\Delta$ by requiring that $e_1,e_2$ are generators. A quick calculation gives \[ [\Phi]_{\Ecal}=\frac{1}{2}\left(\begin{array}{cc} 0 & 1\\ 0 & 0 \end{array}\right)\frac{dz}{z}, \quad [\ol{\Phi}^H]_{\Ecal}=\frac{1}{2}\left(\begin{array}{cc} 0 & 0\\ |\log|z||^{-2} & 0 \end{array}\right)\frac{d\ol{z}}{\ol{z}} \] and so $\Phi\in H^0(\Delta,K(p)\otimes\End_0(\dot{E}))$ is a traceless Higgs field with nonzero nilpotent residue at $p$ and $\ol{\Phi}^H$ is its $H$-adjoint. From \[ \frac{c}{|\!\log|z||^{\frac{1}{2}}}\leq \|e_j\|_H\leq c'|\!\log|z||^{\frac{1}{2}} \] it follows that $w_1=0$, the parabolic structure is integral degenerate and the jumps occur at $\ZZ$. A similar computation shows that, for $T'=-T$, the norm satisfies \[ \frac{c|z|^{\frac{1}{2}}}{|\!\log|z||^{\frac{1}{2}}}\leq \|e_j\|_H\leq c' |\!\log|z||^{\frac{1}{2}}|z|^{\frac{1}{2}} \] and so we would still obtain an integral degenerate parabolic structure with $w_1=\frac{1}{2}$ and a traceless Higgs field with nonzero nilpotent residue, but the jumps would occur at $\frac{1}{2}+\ZZ$. \end{example} \medskip A morphism of parabolic Higgs bundles $f:(E_\bullet,\Phi)\rar (E'_\bullet,\Phi')$ is a map $f:E_\bullet\rar E'_\bullet$ of parabolic bundle that makes the following diagram \[ \xymatrix{ E \ar[rr]^{\Phi} \ar[d]^f && E\otimes K(P)\ar[d]^{f\otimes 1}\\ E' \ar[rr] ^{\Phi'} && E'\otimes K(P) } \] commutative. We will say that $f$ is injective (resp. properly injective) if $f:E\rar E'$ is. A parabolic Higgs sub-bundle of $(E_\bullet,\Phi)$ is a sub-bundle $F_\bullet\subseteq E_\bullet$ such that $\Phi(F)\subseteq F\otimes K(P)$; the map $\Psi:(E/F)_\bullet\rar (E/F)_\bullet\otimes K(P)$ induced by $\Phi$ makes $\left((E/F)_\bullet,\Psi\right)$ into a parabolic Higgs quotient bundle.\\ \subsection{Stability} The {\it{slope}} of the parabolic bundle $E_\bullet$ on $(S,P)$ is defined as $\mu(E_\bullet):=\frac{\deg(E_\bullet)}{\rk(E)}$. \begin{definition} A parabolic Higgs bundle $(E_\bullet,\Phi)$ on $(S,P)$ is {\it{stable}} (resp. {\it{semi-stable}}) if $\mu(F_\bullet)<\mu(E_\bullet)$ (resp. $\mu(F_\bullet)\leq \mu(E_\bullet)$) for every properly injective $(F_\bullet,\Psi)\rar (E_\bullet,\Phi)$. A direct sum of stable parabolic Higgs bundles with the same $\mu$ is said {\it{polystable}}. \end{definition} \begin{remark}\label{rmk:simple} It is well-known that stable bundles are {\it{simple}}, i.e. their endomorphisms are multiples of the identity, and so the group of their automorphism is $\CC^*$. The same argument works for parabolic Higgs bundles. Indeed, if $f:(E_\bullet,\Phi)\rar (E_\bullet,\Phi)$ is a non-zero homomorphism, then $\mu(E_\bullet)\leq \mu(\IM(f))\leq \mu(E_\bullet)$ by semistability of $(E_\bullet,\Phi)$. This forces $\IM(f)=E_\bullet$ because $(E_\bullet,\Phi)$ is stable. Now pick a point $q\in S$ and let $\lambda\in\CC$ be an eigenvalue of $f_q:E|_q\rar E|_q$. The endomorphism $(f-\lambda \cdot \id)\in \End(E_\bullet,\Phi)$ is not surjective and so it vanishes by the above argument. It follows that $f=\lambda \cdot \id$. \end{remark} Semi-stable parabolic Higgs bundles have the Jordan-H\"older property: if $(E_\bullet,\Phi)$ is semi-stable, then there exists a filtration \[ \{0\}=E^0_\bullet\subsetneq E^1_\bullet\subsetneq E^2_\bullet\subsetneq\dots\subsetneq E_\bullet \] by parabolic sub-Higgs-bundles such that the Higgs bundle structure $\Gr^s(E_\bullet,\Phi)$ induced on the quotient $E^s_\bullet/E^{s-1}_\bullet$ is stable and with slope $\mu(\Gr^s(E_\bullet))=\mu(E_\bullet)$. It can be checked that, though the filtration is not canonical, the associated graded object \[ \Gr(E_\bullet,\Phi)=\bigoplus_{s} \Gr^s(E_\bullet,\Phi) \] is. As for vector bundles, two parabolic Higgs bundles $(E_\bullet,\Phi)$, $(E'_\bullet,\Phi')$ are called {\it{$S$-equivalent}} if $\Gr(E_\bullet,\Phi)\cong \Gr(E'_\bullet,\Phi')$. Thus, every semistable object is $S$-equivalent to a unique polystable one, up to isomorphism. \subsection{Moduli spaces of parabolic $\SL_n$-Higgs bundles} Fix a type $\bm{w}$, an $N$-uple $\bCoa$ of conjugacy classes in $\psl_N(\CC)$ and a base-point $b\in\dot{S}$. Fix also a holomorphic parabolic line bundle $\Det_\bullet$ of type $\NO{\bm{w}}$ \begin{definition} A {\it{parabolic Higgs bundle of rank $N$ with determinant $\Det_\bullet$}} on $(S,P)$ {\it{of type $\bm{w}$}} is a triple $(E_\bullet,\eta,\Phi)$, where $E_\bullet$ is a holomorphic parabolic vector bundle of rank $N$ and type $\bm{w}$, endowed with an isomorphism $\eta:\det(E_\bullet)\arr{\sim}\Det_\bullet$ and a {\it{Higgs field}} $\Phi\in H^0(S,K(P)\otimes\End_0(E_\bullet))$. An isomorphism $(E_\bullet,\eta,\Phi)\rar (E'_\bullet,\eta',\Phi')$ of parabolic Higgs bundles of rank $N$ with determinant $\Det_\bullet$ is a map $f:E_\bullet\rar E'_\bullet$ which is an isomorphism of parabolic Higgs bundles and such that $\eta=\eta'\circ\det(f)$. \end{definition} By Remark \ref{rmk:simple}, an automorphism $f$ of $(E_\bullet,\eta,\Phi)$ must satisfy $\det(f_x)=1$ at all $x\in S$. Hence, if $(E_\bullet,\Phi)$ is simple, then $\Aut(E_\bullet,\eta,\Phi)=\mu_N\cdot\id$, where $\mu_N\subset\CC^*$ is the cyclic subgroup of $N$-th roots of unity.\\ Denote by $\Higgs^{ss}_b(S,N,\bm{w},\Det_\bullet,\bCoa)$ the set of isomorphism classes of quadruples $(E_\bullet,\eta,\Phi,\tau)$ such that \begin{itemize} \item $(E_\bullet,\eta,\Phi)$ is a semistable parabolic Higgs bundle on $(S,P)$ of rank $N$, type $\bm{w}$ and with determinant $\Det_\bullet$ \item $\Res_{p_i}(\Phi)\in\coa_i$ for all $i=1,\dots,n$ \item $\tau:E|_b\arr{\sim}\CC^N$ is a framing at $b$. \end{itemize} We denote by $\Higgs^{s}_b\subseteq\Higgs^{ps}_b\subseteq\Higgs^{ss}_b$ the stable and polystable loci. The following two results are due to Simpson \cite{simpson:harmonic}, Konno \cite{konno:construction} and Yokogawa \cite{yokogawa:compactification}. \begin{theorem}[Moduli space of framed semi-stable parabolic Higgs bundles] The space $\Higgs_b^{ss}(S,N,\bm{w},\Det_\bullet,\bCoa)$ is a normal quasi-projective variety and a fine moduli space of $b$-framed semi-stable Higgs bundles on $(S,P)$ of rank $N$, type $\bm{w}$ and with determinant $\Det_\bullet$. Moreover, the stable locus $\Higgs_b^s(S,N,\bm{w},\Det_\bullet,\bCoa)$ is smooth. \end{theorem} The group $\GL_N(\CC)$ acts by post-composition on the $b$-framing, and so it acts on $\Higgs_b^{ss}(S,N,\bm{w},\Det_\bullet,\bCoa)$: we denote by $\RRHiggs(S,N,\bm{w},\Det_\bullet;\bCoa)$ the set-theoretic quotient and by $\Higgs(S,N,\bm{w},\Det_\bullet,\bCoa)$ its Hausdorffization. \begin{theorem}[Moduli space of stable parabolic Higgs bundles] The Hausdorff quotient $\Higgs(S,N,\bm{w},\Det_\bullet,\bCoa)$ is a normal quasi-projective variety, whose points are in bijection with $S$-equivalence classes of semi-stable Higgs bundles on $(S,P)$ of rank $N$, type $\bm{w}$ and with determinant $\Det_\bullet$. The open locus $\Higgs^s(S,N,\bm{w},\Det_\bullet,\bCoa)$ is an orbifold and a fine moduli space of stable objects. \end{theorem} Let $F_\bullet$ be a line bundle on $S$ with parabolic structure of type $\bm{w_F}$ (which is trivial at $b$) and fix a trivialization of $F_\bullet$ at the basepoint $b\in\dot{S}$. Then $E_\bullet\mapsto F_\bullet\otimes E_\bullet$ induces a $\GL_N(\CC)$-equivariant isomorphism \[ \Higgs^{ss}_b(S,N,\bm{w},\Det_\bullet,\bCoa)\arr{\sim} \Higgs^{ss}_b(S,N,\bm{w_F}\otimes\bm{w},F_\bullet^{\otimes N}\otimes\Det_\bullet,\bCoa) \] that preserves polystable and stable locus. Thus, it induces an isomorphism \[ \Higgs(S,N,\bm{w},\Det_\bullet,\bCoa)\cong \Higgs(S,N,\bm{w_F}\otimes \bm{w},F_\bullet^{\otimes N}\otimes\Det_\bullet,\bCoa) \] that preserves the stable locus. \begin{remark} Let $L_\bullet=L(\sum_{i=1}^n w(p_i)p_i)$ be a parabolic line bundle on $(S,P)$ and let $N\geq 2$. Chosen $p_0\in S$, there exist an integer $r$ and a line bundle $Q$ such that $L\cong Q^{\otimes N}\otimes \Ocal_S(r\cdot p_0)$ with $0\leq r<N-1$. Thus, $L_\bullet\cong Q(\frac{r}{N}p_0+\sum_{i=1}^n\frac{w(p_i)}{N}p_i)^{\otimes N}$. Hence, if $n>0$, then we can choose $p_0=p_1$ for instance, and so $L_\bullet$ admits an $N$-root which is a line bundle on $S$ with parabolic structure at $P$. If $n=0$, then $L_\bullet$ admits an $N$-th root which is a line bundle (possibly) with parabolic structure at $p_0$. \end{remark} By the above remark, we can choose an $N$-th root $F_\bullet$ of $\Det^\vee_\bullet$ and so we have established an isomorphism $\Higgs^s(S,N,\bm{w},\Det_\bullet,\bCoa)\rar\Higgs^s(S,N,\bm{w_F}\otimes \bm{w},\Ocal_S,\bCoa)$. For rank $2$ integral parabolic structures the remark specializes to the following. \begin{corollary}[Odd and even rank $2$ integral parabolic structures]\label{cor:det} Let $N=2$ and fix an integral parabolic type $\bm{w}$ and a line bundle $\Det$ with trivial parabolic structure. Also fix an auxiliary point $p_0\in\dot{S}$ different from $b$ and let $\bm{w_0}$ be the parabolic type of $\Ocal_S(\frac{1}{2}p_0)$.\\ Then the moduli space $\Higgs^{s}(S,2,\bm{w},\Det,\bCoa)$ is isomorphic to either of the following: \begin{itemize} \item[(1)] $\Higgs^{s}(S,2,\bm{w}_0\otimes\bm{w},\Ocal_S,\bCoa)$, if $\deg(\Det)$ is odd; \item[(2)] $\Higgs^{s}(S,2,\bm{w},\Ocal_S,\bCoa)$, if $\deg(\Det)$ is even. \end{itemize} \end{corollary} \subsection{Involution and fixed locus} We now restrict to the case of rank $N=2$, and we remark that every element $X\in\psl_2(\CC)$ is in the same $\Ad_{\SL_2(\CC)}$-orbit as $-X$. Fix integral $\bm{w}$, an $n$-uple of conjugacy classes $\bCoa$ in $\psl_2(\CC)$ and a line bundle $\Det$ with trivial parabolic structure and let $d_0=\deg(\Det)$. Following Hitchin, consider the involution $\sigma_b:\Higgs^{ss}_b(S,2,\bm{w},\Det,\bCoa)\rar\Higgs^{ss}_b(S,2,\bm{w},\Det,\bCoa)$ defined as $\sigma(E_\bullet,\eta,\Phi,\tau):=(E_\bullet,\eta,-\Phi,\tau)$, and let $\sigma$ be the induced map on $\Higgs(S,2,\bm{w},\Det,\bCoa)$. \begin{lemma}[$\sigma$-fixed locus]\label{lemma:real} Let $(E_\bullet,\eta,\Phi)$ be a polystable parabolic Higgs bundle of rank $2$, type $\bm{w}$, determinant $\Det$ and with residues at $p_i$ in $\coa_i$. The point $[E_\bullet,\eta,\Phi]$ is fixed under $\sigma$ if and only if there exists an isomorphism $\iota:(E_\bullet,\eta,\Phi)\arr{\sim} (E_\bullet,\eta,-\Phi)$, which happens if and only if one of the following conditions is satisfied: \begin{itemize} \item[(a)] $\Phi=0$ and $E_\bullet$ is polystable; \item[(b)] $E_\bullet\cong (L^\vee_\bullet\otimes\Det)\oplus L_\bullet$ with $\deg(\Det)\leq 2\deg(L_\bullet)$ and \[ 0\neq\Phi=\left(\begin{array}{cc} 0 & \phi\\ \psi & 0 \end{array}\right), \qquad \iota=\pm\left(\begin{array}{cc} i & 0\\ 0 & -i \end{array}\right) \] with respect to this decomposition, where \begin{align*} 0\neq & \phi \in \Hom(L_\bullet,L^\vee_\bullet\otimes\Det)\otimes K(P)\\ &\psi \in \Hom(L^\vee_\bullet\otimes\Det,L_\bullet)\otimes K(P). \end{align*} If $\deg(\Det)<2\deg(L_\bullet)$, then $(E_\bullet,\Phi)$ is necessarily stable. If $\deg(\Det)=2\deg(L_\bullet)$, then $(E_\bullet,\Phi)$ is polystable if and only if $\psi\neq 0$ too. \end{itemize} Furthermore, if $(E_\bullet,\eta,\Phi)$ is stable with $\Phi\neq 0$ and $[E_\bullet,\eta,\Phi]$ is fixed by $\sigma$, then \begin{itemize} \item[(b1)] the isomorphism $\iota$ is unique up to $\{\pm 1\}$; \item[(b2)] if $\Det\not\cong L^{\otimes 2}_\bullet$, then the decomposition $E_\bullet\cong (L^\vee_\bullet\otimes\Det)\oplus L_\bullet$ is unique; \item[(b3)] if $\Det\cong L^{\otimes 2}_\bullet$ and so $E_\bullet\cong L_\bullet\oplus L_\bullet$, then $\phi,\psi$ are not proportional. \end{itemize} Finally, if $(E_\bullet,\eta,\Phi)$ is strictly polystable with $\Phi\neq 0$ and $[E_\bullet,\eta,\Phi]$ is fixed by $\sigma$, then $(E_\bullet,\Phi)\cong \left(L_\bullet\oplus L_\bullet,\left(\begin{array}{cc} \phi & 0\\ 0 & -\phi \end{array}\right)\right)$ with $0\neq \phi\in H^0(S,K(P))$ and $\iota=\pm\left(\begin{array}{cc} 0 & -1\\ 1 & 0 \end{array}\right)$. \end{lemma} \begin{proof} The argument is essentially the same as in \cite{hitchin:self-duality}, Sec.~10.\\ In case (a), the Higgs bundle $(E_\bullet,0)$ is polystable. So we assume $\Phi\neq 0$ and we want to show that (b) holds. Since $\iota\circ \Phi\circ \iota^{-1}=-\Phi$, it is easy to check that $\Phi$ must vanish at all points $Q$ where $\iota$ is a unipotent automorphism. Moreover, a quick computation shows that $\iota$ must have eigenvalues $\pm i$ at $\dot{S}\setminus Q$. Since $\Phi\neq 0$ except at a finite number of points, $\iota$ has everywhere eigenvalues $\pm i$ with eigenbundles $L_\bullet$ and $L^\vee_\bullet\otimes\Det$, and so $\iota=\pm \left(\begin{array}{cc} i & 0\\ 0 & -i \end{array}\right)$ and $\Phi=\left(\begin{array}{cc} 0 & \phi\\ \psi & 0 \end{array}\right)$.\\ Suppose first $\deg(\Det)<2\deg(L_\bullet)$.\\ Then $L_\bullet$ cannot be preserved by $\Phi$ by semi-stability, and so $\phi\neq 0$. Moreover, $\deg(\Det)/2>\deg(L^\vee_\bullet\otimes\Det)$ implies a line sub-bundle $L'_\bullet\subset E_\bullet$ with $\deg(L'_\bullet)>\mu(E_\bullet)=\deg(\Det)/2$ must necessarily be $L'_\bullet=L_\bullet$. It follows that the projection $L'_\bullet\rar L^\vee_\bullet\otimes\Det$ necessarily vanishes and so $(E_\bullet,\Phi)$ is stable. Suppose now $\deg(\Det)=2\deg(L_\bullet)$.\\ Then $\phi=0$ would imply $\psi=0$ by polystability (and vice versa): hence, we must have $\phi,\psi\neq 0$. About the second part of the statement, given $\iota,\iota':(E_\bullet,\eta,\Phi)\rar (E_\bullet,\eta,-\Phi)$ isomorphisms of stable Higgs bundles with determinant $\Det$, the composition $(\iota^{-1}\circ \iota')\in\Aut(E_\bullet,\eta,\Phi)$ and so $\iota^{-1}\circ \iota'=\pm\id$ because $(E_\bullet,\eta,\Phi)$ is simple. Property (b2) easily follows from (b1) and (b). As for (b3), a destabilizing sub-bundle (necessarily isomorphic to $L_\bullet$) exists if and only if $0\neq \phi,\psi\in H^0(S,K(P))$ are proportional. Concerning the final claim, it is enough to observe that $\phi$ vanishes only on finitely many points of $S$ and the involution $\iota$ must exchange the two eigenspaces of $\Phi/\phi$ away from those finitely many points. \end{proof} Denote by $\Higgs^{ps}_b(S,2,\bm{w},\Det,\bCoa)(\RR)$ the set of $(E_\bullet,\eta,\Phi,\tau,\{\pm\iota\})$ such that $(E_\bullet,\eta,\Phi,\tau)$ is a polystable parabolic Higgs bundle of rank $2$ of type $\bm{w}$ with determinant $\Det$ and residues in $\bCoa$, and $\iota:(E_\bullet,\eta,\Phi)\rar (E_\bullet,\eta,-\Phi)$ is an isomorphism. We denote by $\Higgs^s_b(S,2,\bm{w},\Det,\bCoa)(\RR)$ the locus of stable objects and by $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)$ its quotient by $\SL_2(\CC)$. By the above lemma, $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)$ can be identified to the locus in $\Higgs^s(S,2,\bm{w},\Det,\bCoa)$ fixed by $\sigma$, and so we will denote a point just by $(E_\bullet,\eta,\Phi)$. If $\ol{\bCoa}\supseteq \bm{\mathfrak{0}}=\{0\}^n$, then $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)$ contains the locus of $\Phi=0$, namely \[ \Bun^s(S,2,\bm{w},\Det) := \big\{\text{$(E_\bullet,\eta)$ stable parabolic rank $2$ bundle of type $\bm{w}$ with $\eta:\det(E_\bullet)\arr{\sim}\Det$}\big\} \] \begin{definition} We say that the couple $(\bm{w},\bCoa)$ is {\it{compatible (with the $\sigma$-involution)}} if \begin{itemize} \item $0<w_1(p_i)<1/2\ \implies\ \bCoa_i=\{0\}$ \item $w_1(p_i)=0;\ 1/2\ \implies\ \det(\bCoa_i)\geq 0$ \item $s_0+\chi(\dot{S})<0$. \end{itemize} \end{definition} \begin{notation} Given $\bm{w}$, denote by $J\even_{deg}$ (resp. $J\odd_{deg}$) the set of indices $j\in\{1,2,\dots,n\}$ for which $w_1(p_j)=0$ (resp. $w_1(p_j)=\frac{1}{2}$) and let $J_{deg}=J\even_{deg}\cup J\odd_{deg}$.\\ Given $\bCoa$, we define $J_{0} =\{j\in J_{deg}\,|\,\bcoa_j=\{0\}\}$, $J_{nil} =\{j\in J_{deg}\,|\,\text{$\bcoa_j$ nilpotent}\}$ and $J_{inv} =\{j\in J_{deg}\,|\,\det(\bcoa_j)\neq 0\}$.\\ Finally, call $s=\# J_{deg}$, $s_{0}=\# J_0$ and $s_{inv}=\# J_{inv}$, and also $s\even=\# J\even_{deg}$ and $s\odd=\# J\odd_{deg}$. \end{notation} \begin{definition} Given $\Det$ and compatible $(\bm{w},\bCoa)$, we say that the couple $(d,\bm{a})\in\ZZ\times\{0,1\}^n$ is {\it{admissible}} if \begin{itemize} \item[(a)] $a_j=0$ for all $j\in J_{deg}$; \item[(b)] $e(d,\bm{a},\bm{w}):=2d-d_0+2\sum_{i=1}^n \left(a_i+(-1)^{a_i}w_1(p_i)\right)\geq 0$ \item[(c)] $2d\leq d_0-\chi(S)-\NO{\bm{a}}+s\even-s_0$. \end{itemize} \end{definition} For every admissible $(d,\bm{a})$, we define the following locus in $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)$ \[ \Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)_{d,\bm{a}} : = \left\{ \begin{array}{l} \text{$E_\bullet\cong(L^\vee_\bullet\otimes\Det)\oplus L_\bullet$ with $\Phi=\left(\begin{array}{cc} 0 & \phi\\ \psi & 0\end{array}\right)$ and $\Res_{p_i}(\Phi)\in\bcoa_i$}\\ 0\neq \phi\in H^0(S,\Det K L_\bullet^{-2}(P)), \quad \psi\in H^0(S,\Det^\vee K L_\bullet^2(P))\\ \deg(L)=d, \quad w_L(p_i)=\begin{cases} w_1(p_i) & \text{if $a_i=0$}\\ 1-w_1(p_i) &\text{if $a_i=1$}\end{cases} \end{array} \right\} \] Thus, $e(d,\bm{a},\bm{w})$ can be rewritten as $e(d,\bm{a},\bm{w})=2d-d_0+2\sum_{i=1}^n w_L(p_i)$ and the admissibility constraints (b-c) read \[ -2\sum_{i=1}^n w_L(p_i)\leq 2d-d_0\leq -\chi(S)-\NO{\bm{a}}+s\even-s_0. \] Moreover, the condition $e(d,\bm{a},\bm{w})\geq 0$ is equivalent to $\deg(L_\bullet)\geq \deg(\Det\otimes L_\bullet^\vee)$; thus, in view of Lemma \ref{lemma:real}, it is understood that we also require $\psi\neq 0$, if $e(d,\bm{a},\bm{w})=0$. We can rephrase our analysis as follows. \begin{proposition}[Partition of the $\sigma$-fixed locus]\label{prop:real} The space $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})$ can be decomposed into the disjoint union of the following loci \[ \begin{array}{ll} \Bun^s(S,2,\bm{w},\Det) & \text{if $\bm{\mathfrak{0}}\in\ol{\bCoa}$}\\ \Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)_{d,\bm{a}} & \text{for admissible $(d,\bm{a})$.} \end{array} \] \end{proposition} \subsection{Topology of the $\sigma$-fixed locus}\label{sec:topology} Let $\bcoa_i$ be conjugacy classes and $\ol{\bcoa}_i$ be their closures in $\psl_2(\CC)$. Throughout this section, we will assume that $(\bm{w},\bCoa)$ are compatible, that $(d,\bm{a})$ is admissible and that $e(d,\bm{a},\bm{w})>0$. We begin with some simple observations. \begin{lemma} Let $(E_\bullet,\eta,\Phi,\{\pm\iota\})\in \Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)_{d,\bm{a}}$ and $E_\bullet=(L_\bullet^\vee\otimes\Det)\oplus L_\bullet$. Then \begin{itemize} \item[(i)] for $i\notin J_{deg}$ ($\bm{w}$ non-degenerate at $p_i$), $L^{-2}_\bullet(P+\bm{a}P)$ has parabolic twist $1+a_i-2w_L(p_i)\in [0,1)$ and $L^2_\bullet(-\bm{a}P)$ has twist $2w_L(p_i)-a_i\in [0,1)$ at $p_i$; \item[(ii)] for $j\in J_{deg}$ ($\bm{w}$ degenerate at $p_j$), $L^2$ and $L^{-2}$ have trivial parabolic structure at $p_j$. \end{itemize} Thus, $\deg\floor{\Det L_\bullet^{-2}(P)}=d_0-2d-\NO{\bm{a}}+s\even$ and $\deg\floor{\Det^\vee L_\bullet^2}=2d-d_0+\NO{\bm{a}}+s\odd \geq 1-n$. As a consequence, \[ \begin{array}{rcl} m:=& \deg\floor{\Det L^{-2}_\bullet K(P)}&=d_0+2g-2-2d-\NO{\bm{a}}+s\even\geq 0\\ & \deg\floor{\Det^\vee L^{2}_\bullet K(P)}&=-d_0+2g-2+n+2d+\NO{\bm{a}}+s\odd\geq 0\\ m':= & h^0(S,\Det^\vee L^2_\bullet K(P)) &=-d_0+g-1+n+2d+\NO{\bm{a}}+s\odd\geq 0 \end{array} \] since $\deg(\floor{\Det^\vee L^2_\bullet(P)})>0$. Thus, $m+m'=3g-3+n+s$. \end{lemma} \begin{proof} Parts (i) and (ii) are elementary computations. The bound for $\deg\floor{\Det^\vee L_\bullet^2}$ follows by observing that $0<e(d,\bm{a},\bm{w})=2d-d_0+2\sum_{i=1}^n w_L(p_i)\leq 2d-d_0+\NO{\bm{a}}+n$ because $w_L(p_i)\leq a_i+1$. Thus, $m'$ can be calculated by Riemann-Roch. \end{proof} Since we are assuming $e(d,\bm{a},\bm{w})>0$, a point of $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)_{d,\bm{a}}$ can be identified with a triple $(L_\bullet,\phi,\psi)$ that satisfies certain conditions, up to isomorphism. In fact, the map $L_\bullet\rar L_\bullet$ of multiplication by $\lambda\in\CC^*$ induces an isomorphism of $(L_\bullet,\phi,\psi)$ with $(L_\bullet,\lambda^{-2}\phi,\lambda^2\psi)$. Hence, such a point can be identified with the triple $(L_\bullet,Q,-\phi\psi)$, where $Q$ is an effective divisor in the linear system $|\Det \floor{L_\bullet^{-2}} K(P)|$ and $-\phi\psi=\det(\Phi)\in H^0(S,K^2(2P))$. Moreover, $L^{-2}_\bullet$ can be reconstructed up to isomorphism from $Q$: so, given $(Q,\det(\Phi))$, there are exactly $2^{2g}$ choices for $L_\bullet$ and the set of such choices is a $\Pic^0(S)[2]$-torsor. Consider then the residues of $\det(\Phi)$. \begin{itemize} \item Suppose that $i\notin J_{deg}$ and so $\bm{w}$ is non-degenerate at $p_i$.\\ Then necessarily $\Res_{p_i}(\Phi)=0$. Thus, we will assume that $\bcoa_i=\{0\}$ for all $i\notin J_{deg}$. \item Suppose that $j\in J_{deg}$ and so $\bm{w}$ is degenerate at $p_j$.\\ Then the condition on the residue at $p_j$ is not automatically satisfied. For $j$ in $J_{nil}$ or $J_{inv}$, the elements in $\ol{\bcoa}_j$ are detected by their determinant $\det(\bcoa_j)$: thus it is enough to require that $(-\phi\psi)(p_j)=\det(\coa_j)$. For $j\in J_0$, we must require that $\phi(p_j)=0$ and that $\ord_{p_j}(\phi\psi)>\ord_{p_j}(\phi)$. \end{itemize} Consider the space $\Xcal=\Sym^{m}(S\setminus P_{inv})\times H^0(S,K^2(P+P_{deg}))$, where $P_{deg}=\sum_{j\in J_{deg}} p_j$ and $P_{inv}=\sum_{j\in J_{inv}} p_j$, and the loci \begin{align*} \Qcal & =\{(Q,q)\in\Xcal\,|\,Q\leq \mathrm{div}(q)\}\\ \ol{\Rcal}_i & =\{(Q,q)\in\Xcal\,|\, \Res_{p_i}(q)=\det(\coa_i) \} & \text{for all $i=1,\dots,n$.} \end{align*} Moreover, for $j\in J_0\cup J_{nil}$, the locus $\ol{\Rcal}_j$ can be split into \begin{align*} \ol{\Rcal}^-_j & = \{(Q,q)\in\Xcal\,|\, p_j\in Q\} & \text{corresponding to $\phi(p_j)=0$}\\ \ol{\Rcal}^+_j & = \{(Q,q)\in\Xcal\,|\, \ord_{p_j}(q)>\mathrm{mult}_{p_j}(Q)\} & \text{corresponding to $\psi(p_j)=0$} \end{align*} and we call $\Rcal^{0}_j:=\ol{\Rcal}^+_j\cap\ol{\Rcal}^-_j$ for all $j\in J_0$. Finally, for every $\bm{\e}:J_{nil}\rar \{+,-\}$ we denote by $P_\pm(\bm{\e})$ the subsets of points $p_j$ such that $\bm{\e}(j)=\pm$ and let $s_\pm(\bm{\e})=\# P_\pm(\bm{\e})$. Then we define \[ \ol{\Rcal}^{\bm{\e}}:= \left(\bigcap_{j\in J_{inv}}\ol{\Rcal}_j\right) \cap \left( \bigcap_{j\in J_{nil}} \ol{\Rcal}_j^{\e_j} \right) \cap \left( \bigcap_{j\in J_0} \Rcal_j^0 \right) \quad \text{and} \quad \ol{\Rcal}=\bigcup_{\bm{\e}} \ol{\Rcal}^{\bm{\e}}. \] and we call $\Rcal^{\bm{\e}}:=\ol{\Rcal}^{\bm{\e}}\setminus \bigcup_{j\in J_{nil}} \Rcal_j^0$ and $\Rcal:=\bigcup_{\bm{\e}}\Rcal^{\bm{\e}}$. The locally closed subvariety $\Qcal$ has codimension $m$ in $\Xcal$ and it is isomorphic to a holomorphic vector bundle of rank $m'$ over $\Sym^{m}(S\setminus P_{inv})$ and so the above discussion leads to the following conclusion. \begin{proposition}[Topology of $\sigma$-fixed components]\label{prop:topology} The loci in $\Xcal$ defined above satisfy the following properties. \begin{itemize} \item[(a)] The locus $\Qcal\cap\ol{\Rcal}^{\bm{\e}}$ is a holomorphic affine bundle of rank $m'-[s_{inv}+s_0+s_+(\bm{\e})]$ over $\mathrm{Sym}^{m-[s_0+s_-(\bm{\e})]}(S\setminus P_{inv})$. The locus $\Qcal\cap\Rcal^{\bm{\e}}$ is obtained from $\Qcal\cap\ol{\Rcal}^{\bm{\e}}$ by first restricting the affine bundle over $\mathrm{Sym}^{m-[s_0+s_-(\bm{\e})]}\left(S\setminus (P_+(\bm{\e})\cup P_{inv})\right)$ and then removing $s_-(\bm{\e})$ affine subbundles of codimension $1$. \item[(b)] The locus $\Qcal\cap\ol{\Rcal}$ is connected, has pure codimension $s+s_0+m$ in $\Xcal$ and consists of the irreducible components $\Qcal\cap \ol{\Rcal}^{\bm{\e}}$ of dimension $3g-3+n-s_0$. The locus $\Qcal\cap\Rcal$ is the disjoint union of all $\Qcal\cap\Rcal^{\bm{\e}}$. \item[(c)] The morphism \[ \xymatrix@R=0in{ \Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)_{d,\bm{a}}\ar[rr]&& \Xcal\\ (E_\bullet,\eta,\Phi) \ar@{|->}[rr] && (\mathrm{div}(\phi),\det(\Phi)) } \] is a $\Pic^0(S)[2]$-torsor over $\Qcal\cap\ol{\Rcal}$ and $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)_{d,\bm{a}}$ is a $\Pic^0(S)[2]$-torsor over $\Qcal\cap\Rcal$. \item[(d)] The restriction $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ of the $\Pic^0(S)[2]$-torsor in (c) over the component $\Qcal\cap \Rcal^{\bm{\e}}$ is connected, unless $\Qcal\cap\ol{\Rcal}^{\bm{\e}}$ is an affine space (i.e. $m-[s_0+s_-(\bm{\e})]=0$): in this case it is necessarily trivial. \end{itemize} \end{proposition} We stress that, by definition, $\Qcal\cap \Rcal^{\bm{\e}}=\emptyset$ if $m-[s_0+s_-(\bm{\e})]<0$ or $m'-[s_{inv}+s_0+s_+(\bm{\e})]<0$. \begin{corollary}[Topology of {$\Pic^0(S)[2]$}-quotient of $\sigma$-fixed irreducible components]\label{cor:quotient-topology} The quotient $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)_{d,\bm{a}}^{\bm{\e}}/\Pic^0(S)[2]$ is isomorphic to a holomorphic affine bundle of rank $m'-[s-s_-(\bm{\e})]$ over $\Sym^{m-[s_0+s_-(\bm{\e})]}(S\setminus P_{inv})$ and $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)_{d,\bm{a}}^{\bm{\e}}/\Pic^0(S)[2]$ is isomorphic to the complement of $s_-(\bm{\e})$ affine codimension $1$ subbundles inside a holomorphic affine bundle of rank $m'-[s-s_-(\bm{\e})]$ over $\Sym^{m-[s_0+s_-(\bm{\e})]}\left(S\setminus (P_+(\bm{\e})\cup P_{inv})\right)$. \end{corollary} \begin{proof}[Proof of Proposition \ref{prop:topology}] Considering the above discussion, we are only left to prove (d). The action of $\Pic^0(S)[2]$ on $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ is given by $A\cdot (E_\bullet,\eta,\Phi)\mapsto (E_\bullet\otimes A,\eta,\Phi)$ for $A\in\Pic^0(S)[2]$. If $m-[s_0+s_-(\bm{\e})]=0$, then $\Qcal\cap\ol{\Rcal}^{\bm{\e}}$ is an affine space and so the torsor is trivial. Assume now $m-[s_0+s_-(\bm{\e})]>0$ and fix $A\in\Pic^0(S)[2]$. We want to show that every $(E_\bullet,\eta,\Phi)$ in $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ can be connected to $(E_\bullet\otimes A,\eta,\Phi)$ by a continuous path. Consider the map $f:\Higgs^s(S,2,\bm{w},\Det;\bCoa)(\RR)^{\bm{\e}}_{d,\bm{a}} \rar \Pic^d(S)\times \Sym^{m-s_0(\bm{\e})-s_-(\bm{\e})}(S\setminus P_{inv})$ defined by $f(E_\bullet,\eta,\Phi)=(L,[\phi])$, where $E_\bullet=(L^\vee_\bullet\otimes\Det)\oplus L_\bullet$ and $\Phi=\left(\begin{array}{cc} 0 & \phi \\ \psi & 0\end{array}\right)$. Since the knowledge of $\bm{w}$ and $\bm{a}$ allows to reconstruct the parabolic bundle $L_\bullet$ out of $L$, the image $\widetilde{\mathcal{S}}$ of $f$ can be identified to the locus of couples $(L,Q)$ such that $Q_{fix}+Q\in|\Det L^{-2}_\bullet K(P)|$, where $Q_{fix}:=\sum_{j\in J_0} p_j+\sum_{\e_j=-}p_j$. Note that $\widetilde{\mathcal{S}}$ is an \'etale cover over $\Sym^{m-s_0(\bm{\e})-s_-(\bm{\e})}(S\setminus P_{inv})$ of degree $2^{2g}$ and $f$ is a fibration with fiber $\CC^{m'-[s-s_-(\bm{\e})]}$. Thus, it is enough to show that $\widetilde{\mathcal{S}}$ is connected. Now fix $(L,Q)\in\widetilde{\mathcal{S}}$ and let $B:[0,1]\rar \Pic^0(S)$ be a continuous path from $\Ocal_S$ to $A$, so that $B^{-2}:[0,1]\rar\Pic^0(S)$ is a closed path. Since $m-[s_0+s_-(\bm{\e})]>0$, we can choose a point $x\in Q$ and we can consider the map \[ \xymatrix@R=0in{ S\setminus P_{inv} \ar[r] & \Pic^0(S)\\ y \ar@{|->}[r] & \Ocal_S(y-x) } \] which induces a surjection $\pi_1(S\setminus P_{inv})\twoheadrightarrow \pi_1(\Pic^0(S))$. Thus, there exists a path $Y:[0,1]\rar S\setminus P_{inv}$ based at $Y(0)=Y(1)=x$ which is mapped to a path homotopic to $B^{-2}$. Define $Q(t):=Q-x+Y(t)$ and let $A(t)$ be the unique continuous path in $\Pic^0(S)$ such that $A(0)=\Ocal_S$ and $A(t)^2=\Ocal_S(x-Y(t))$. By definition, $Q_{fix}+Q(t)\in |\Det (A(t)\otimes L_\bullet)^{-2}K(P)|$. Since the path $A^{-2}$ is homotopic to $B^{-2}$, we have $A(1)=B(1)=A$ and so the path $t\mapsto (L\otimes A(t),Q(t))$ joins $(L,Q)$ and $(L\otimes A,Q)$. \end{proof} Provided $e(d,\bm{a},\bm{w})$ remains positive, Proposition \ref{prop:topology} shows that the isomorphism class of the moduli space $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ remains constant as a parabolic weight $w_1(p_i)$ is varied within the interval $(0,1/2)$ and $\bcoa_i$ is kept equal to $\{0\}$. Moreover, if $w_1(p_i)\in (0,1/2)$ is pushed to $w'_1(p_i)=0;\ 1/2$ and the class $\bcoa_i$ is switched to $\bcoa'_i=\{\text{nilpotents}\}$, then the moduli space is isomorphic to some $\Higgs^s(S,2,\bm{w'},\Det,\ol{\bCoa}')(\RR)^{\bm{\e'}}_{d',\bm{a'}}$, where $\e'_i=+$ if $e(d',\bm{a'},\bm{w'})>e(d,\bm{a},\bm{w})>0$, and $\e'_i=-$ if $0<e(d',\bm{a'},\bm{w'})<e(d,\bm{a},\bm{w})$. More precisely, we have the following. \begin{corollary}[Varying the parabolic weights]\label{cor:moving} Fix $\bCoa$, $d$, $\bm{a}$ and $\bm{w}$ such that $e=e(d,\bm{a},\bm{w})>0$, and assume that $\bm{w}$ is non-degenerate at $p_i$ and $\bcoa_i=\{0\}$. Then $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ is isomorphic to \begin{itemize} \item[(a)] $\Higgs^s(S,2,\bm{w'},\Det,\ol{\bCoa})(\RR)^{\bm{\e'}}_{d',\bm{a'}}$ with $d'=d$, $\bm{a'}=\bm{a}$, $\bm{\e'}=\bm{\e}$ and for every $\bm{w'}$ that differs from $\bm{w}$ only on the $i$-th entrance and such that $0<w'_1(p_i)<1/2$ (see also Nakajima \cite{nakajima:hyperkaehler}); \item[(b)] $\Higgs^s(S,2,\bm{w'},\Det,\ol{\bCoa}')(\RR)^{\bm{\e'}}_{d',\bm{a'}}$ where $\bm{w'},\bCoa',\bm{a'}$ differ from $\bm{w},\bCoa,\bm{a}$ only on the $i$-th entrance, $\bcoa'_i=\{\text{nilpotents}\}$, $a'_i=0$ and either of the following hold: \[ \begin{array}{|c|c|c|c|c|} \hline a_i & w'_1(p_i) & \e'_i & d' & e'\\ \hline\hline 0 & 0 & - & d & e-2w_1(p_i)\\ \hline 0 & 1/2 & + & d & e+(1-2w_1(p_i))\\ \hline 1 & 0 & + & d+1 & e+2w_1(p_i)\\ \hline 1 & 1/2 & - & d & e-(1-2w_1(p_i))\\ \hline \end{array} \] \end{itemize} as long as $e'=e(d',\bm{a'},\bm{w'})>0$. \end{corollary} Similarly, since $\Res_{p_i}(q)=\det(\bcoa_i)$ is an affine equation in $H^0(S,K^2(P+P_{deg}))$, we also have the following result. \begin{corollary}[Varying the quadratic residue of $\det(\Phi)$] Let $(\bm{w},\bCoa),d,\bm{a}$ be such that $e(d,\bm{a},\bm{w})>0$ and assume that $\det(\bcoa_i)>0$. Then $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})^{\bm{\e}}_{d,\bm{\a}}$ is isomorphic to $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa}')^{\bm{\e}}_{d,\bm{\a}}$, where $\bCoa'$ differs from $\bCoa$ only on the $i$-th entrance and $\det(\bcoa'_{i})>0$. \end{corollary} Notice that compact components may also occur, but only in a few limited cases. If $g=0$ and $n=3+s_0$, then $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ consists of a single point (if it is nonempty) and so it is compact. Biquard-Tholozan remarked that the other cases of compact components with non-degenerate parabolic structure correspond to representations that Deroin-Tholozan \cite{deroin-tholozan:super-maximal} call ``super-maximal'' via Theorem \ref{thm:correspondence}. I would like to thank Nicolas Tholozan for drawing my attention to this point. \begin{corollary}[Compact components]\label{cor:supermaximal} Assume $e=e(d,\bm{a},\bm{w})>0$ and $(g,n)\neq (0,3+s_0)$. The locus $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ is compact if and only if \[ (\star) \begin{cases} g=0\\ s_-(\bm{\e})=s_{inv}=0\\ \dis e(d,\bm{a},\bm{w})=1-\sum_{\substack{i\notin J_{deg} \\ a_i=0}} (1-2w_1(p_i)) - \sum_{\substack{i\notin J_{deg}\\ a_i=1}} 2w_1(p_i)\in (0,1]. \end{cases} \] In this case, Higgs bundles $[E,\Phi]\in \Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ have $\psi=0$ and so nilpotent $\Phi$, and the whole component is isomorphic to $\CC\PP^{n-3-s_0}$.\\ The locus $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)^{\bm{\e}}_{d,\bm{a}}$ is compact if and only if the condition $(\star)$ is satisfied and $s_+(\bm{\e})=0$. Moreover, in this case such component is again isomorphic to $\CC\PP^{n-3-s_0}$. \end{corollary} \begin{proof} Since $e=e(d,\bm{a},\bm{w})=-d_0+2d+\sum_{i=1}^n 2w_L(p_i)$, we have \begin{align*} m'-s+s_-(\bm{\e}) &=(-d_0+2d)+g-1+n+\NO{\bm{a}}+s\odd-s+s_-(\bm{\e})=\\ &= e-\sum_{i=1}^n 2w_L(p_i)+g-1+n+\NO{\bm{a}}-s\even+s_-(\bm{\e})=\\ & = g-1+e+\sum_{i\notin J_{deg}}\left(1+a_i-2w_L(p_i)\right)+s_-(\bm{\e}) \geq e+g-1 \end{align*} In Corollary \ref{cor:quotient-topology}, $\Higgs^s(S,2,\bm{w},\Det,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ is presented as a fibration over a symmetric product and the fiber is an open subset of an affine space. Thus, it is compact if and only if the fiber is $0$-dimensional and the base is a symmetric product of a compact surface (i.e. $s_{inv}=0$). For the fiber to be $0$-dimensional, we must have $m'\leq s-s_-(\bm{\e})$. Since $1+a_i-2w_L(p_i)> 0$ for $i\notin J_{deg}$ and $e>0$, this implies that $g=0$, $s_-(\bm{\e})=0$ and \[ 0<e=1-\sum_{i\notin J_{deg}}\left(1+a_i-2w_L(p_i)\right)\leq 1. \] Vice versa, if the above numerical conditions are satisfied, it is immediate to check that the fiber is $0$-dimensional and indeed it consists of a single point. In this case, such component is isomorphic to $\Sym^{n-3-s_0}(\CC\PP^1)\cong \CC\PP^{n-3-s_0}$. Similarly, the open component $\Higgs^s(S,2,\bm{w},\Det,\bCoa)(\RR)^{\bm{\e}}_{d,\bm{a}}$, fibers over $\Sym^{n-3-s_0}(S\setminus (P_{inv}\cup P_+))$. The numerical conditions for the $0$-dimensionality of the fiber are the same; for the base to be compact we need $s_{inv}=s_+(\bm{\e})=0$. Again the component will be isomorphic to $\CC\PP^{n-3-s_0}$. \end{proof} \section{Hitchin-Simpson correspondence and topology}\label{sec:correspondence} Fix a complex structure $I$ on the compact surface $S$ and let $\Ocal_S$ be the sheaf of $I$-holomorphic functions on $S$. \subsection{Closed case} Let $S$ be compact and unpunctured ($n=0$). Since locally constant functions on $S$ are $I$-holomorphic, a flat $\CC$-vector bundle $V$ on $S$ can be naturally given an $I$-holomorphic structure $\ol{\pa}^V$. In particular, if $(\xi,\nabla)$ is a flat $\GL_N$-bundle, then $V:=\xi\times_{\GL_N}\CC^N$ is a $I$-holomorphic bundle endowed with a flat connection, which we will still denote by $\nabla$ by a little abuse. The point of departure is then the following classical result. \begin{theorem}[Narasimhan-Seshadri \cite{narasimhan-seshadri:unitary}]\label{thm:narasimhan-seshadri} The map that sends a flat $\U_N$-bundle $(\xi,\nabla)$ on $S$ to the $I$-holomorphic vector bundle $E:=\xi\times_{\U_N}\Ocal_S^{\oplus N}$ induces a real-analytic homeomorphism \[ \xymatrix@R=0in{ \Flat^{irr}(S,\U_N) \ar[rr]^{\sim}&& \Bun^s(S,N)_0 } \] between the moduli space of irreducible flat $\U_N$-principal bundles on $S$ and the moduli space of stable $I$-holomorphic vector bundles of rank $N$ and degree $0$ on $(S,I)$. Such homeomorphism restricts to \[ \xymatrix@R=0in{ \Flat^{irr}(S,\SU_N)\ar[rr]^{\sim}&&\Bun^s(S,N,\Ocal_S)\\ (\xi,\nabla)\ar@{|->}[rr] && (E,\eta) } \] where $E=\xi\times_{\SU_N}\Ocal_S^{\otimes N}$ and $\eta:\det(E)\arr{\sim}\Ocal_S$ sends the unit volume element to $1$. \end{theorem} Such correspondence can be lifted to bundles endowed with a trivialization $\tau$ at the base-point $b$; moreover, just by taking direct sums it can be extended to a correspondence between completely decomposable $b$-framed flat bundles and polystable $b$-framed $I$-holomorphic bundles. \begin{corollary} There is a real-analytic homeomorphism \[ \xymatrix@R=0in{ \Flat_b^{dec}(S,\U_N)\ar[rr]^{\sim} && \Bun^{ps}_b(S,N)_0\\ (\xi,\nabla,\tau) \ar@{|->}[rr] && (E=\xi\times_{\U_N} \Ocal_S^{\oplus N},\tau'=\tau\otimes_{\U_N}\CC^{N}) } \] which restricts to \[ \xymatrix@R=0in{ \Flat_b^{dec}(S,\SU_N)\ar[rr]^{\sim} && \Bun^{ps}_b(S,N,\Ocal_S)\\ (\xi,\nabla,\tau) \ar@{|->}[rr] && (E=\xi\times_{\SU_N} \Ocal_S^{\oplus N},\eta,\tau'). } \] \end{corollary} In the proof by Narasiman-Seshadri, surjectivity is achieved by continuity method. In particular, it does not provide a way to construct $(\xi,\nabla)$ starting from a stable $E$. The following important result fills such a gap. \begin{theorem}[Donaldson \cite{donaldson:narasimhan-seshadri}]\label{thm:donaldson-NS} A holomorphic vector bundle $E$ of rank $N$ and degree $0$ on $(S,I)$ admits a flat invariant metric if and only if $E$ is polystable. Moreover, such a metric is unique up to automorphisms of $E$. \end{theorem} In the case of bundles with monodromy not contained in $\U_N$, no invariant Hermitian metric is available. In order to codify all information in term of holomorphic structures on $(S,I)$, the idea is to replace invariant metrics by harmonic metrics. \begin{definition} A {\it{harmonic metric}} on a flat $\GL_N$-bundle $(\xi,\nabla)$ on $(S,I)$ is the Hermitian metric $H=h\,h^T$ on the flat bundle $V=\xi\times_{\GL_N}\CC^N$ associated to a section $h:S\rar \xi\times_{\GL_N}(\GL_N/\U_N)$ that minimizes the energy with respect to the natural metric on the symmetric space $\GL_N/\U_N$. \end{definition} Existence and uniqueness of harmonic metrics and was first proven by Donaldson \cite{donaldson:twisted} in the rank $2$ case. A more general existence theorem is due to Corlette: here we recall the statement for Riemann surfaces only. \begin{theorem}[Corlette \cite{corlette:harmonic}]\label{thm:corlette} A flat $\GL_N$-bundle $(\xi,\nabla)$ on $(S,I)$ has a harmonic metric $H$ if and only if $\xi$ is reductive (i.e. the closure of the image of $\hol_\xi$ in $\GL_N(\CC)$ is reductive). \\ Moreover, such an $H$ is unique up to automorphisms of $\xi$. \end{theorem} Given a hermitian metric $H$ on the flat vector bundle $V=\xi\times_{\GL_N}\CC^N$, the connection $\nabla$ decomposes as \[ \nabla=\nabla^H+\Phi+\ol{\Phi}^H \] where $\Phi$ is an $\End(V)$-valued $(1,0)$-form, $\ol{\Phi}^H$ is its $H$-adjoint and $\nabla^H$ is compatible with $H$. We can define a holomorphic structure on $E=V$ by letting $\ol{\pa}^E=\ol{\pa}^V-\ol{\Phi}^H$, so that $\nabla^H$ is a Chern connection for the holomorphic Hermitian bundle $(E,H)$. Harmonicity of the metric $H$ is then equivalent to the $\ol{\pa}^E\!-$holomorphicity of $\Phi\in C^\infty(S,K\otimes \End(E))$. Thus, once we find a harmonic metric $H$ on $(V,\nabla)$, we can produce a Higgs bundle $(E,\Phi)$ of degree $\deg(E)=\deg(V)=0$.\\ Conversely, given a Higgs bundle $(E,\Phi)$ on $(S,I)$ and a Hermitian metric $H$ on $E$, we can consider the underlying complex vector bundle $V$ endowed with the connection $\nabla=\nabla^H+\Phi+\ol{\Phi}^H$, where $\nabla^H$ is the Chern connection on $(E,H)$ and $\ol{\Phi}^H$ is the $H$-adjoint of $\Phi$. Harmonicity of the metric $H$ on $(V,\nabla)$ is equivalent to the flatness of $\nabla$ and the following theorem provides the wished counterpart to Theorem \ref{thm:corlette} (proven before by Hitchin \cite{hitchin:self-duality} in the rank $2$ case). \begin{theorem}[Simpson \cite{simpson:yang-mills}] A $\GL_N$-Higgs bundle $(E,\Phi)$ on $(S,I)$ supports a metric $H$ such that the induced $(V,\nabla)$ is flat if and only if $(E,\Phi)$ is polystable. Moreover, such a metric is unique up to automorphisms of $(E,\Phi)$. \end{theorem} Considering the $\SL_N$-bundles correspond to vector bundles with trivializable determinant, we summarize the above results as follows. \begin{corollary}[Simpson \cite{simpson:higgs1} \cite{simpson:higgs2}] There are smooth diffeomorphisms \[ \xymatrix@R=0in{ \Flat_b^{red}(S,\GL_N)\ar[rr] && \Higgs^{ps}_b(S,N)_0\\ \Flat_b^{red}(S,\SL_N)\ar[rr] && \Higgs^{ps}_b(S,N,\Ocal_S) } \] which induce a correspondence \[ \xymatrix@R=0in{ \Flat^{Zd}(S,\GL_N)\ar[rr] && \Higgs^{s}(S,N)_0\\ \Flat^{Zd}(S,\SL_N)\ar[rr] && \Higgs^{s}(S,N,\Ocal_S) } \] between the space of flat bundles with Zariski-dense monodromy and the space of stable $I$-holomorphic Higgs bundles. \end{corollary} Again, the case $N=2$ of above corollary is due to Hitchin \cite{hitchin:self-duality}, whose construction also implies that the involved diffeomorphisms are real-analytic. Simpson showed that in general the above correspondence does not continuously extend over the whole semi-stable locus (see \cite{simpson:higgs2}, pp.38--39).\\ We remind that flat $\U_N$-bundles are always decomposable and so reductive; in this case, Zariski-density of the monodromy is equivalent to irreducibility. \subsection{Punctured case}\label{sec:correspondence-punctured} Given a flat $\U_N$-bundle $(\xi,\nabla)$ on the punctured surface $\dot{S}$, we can as before produce a $I$-holomorphic vector bundle $\dot{E}=\xi\times_{\U_N}\Ocal_{\dot{S}}^{\oplus N}$ on $\dot{S}$ which carries an invariant Hermitian metric $H$. Moreover, such $\dot{E}$ admits a unique extension $E\rar S$ {\it{(Deligne extension \cite{deligne:equations})}} such that the induced $\nabla$ has real residues at $p_i$ with eigenvalues $0\leq w_1(p_i)<w_2(p_i)<\dots<w_{b_i}(p_i)<1$ for all $p_i\in P$ and algebraic multiplicities $m_1(p_i),\dots,m_{b_i}(p_i)$. \begin{notation} Given an endomorphism $f$ of a complex vector space and a $w\in\RR$, we denote by $\Eig_w(f)$ the direct sum of generalized eigenspaces of $f$ corresponding to eigenvalues with real part $w$. We also denote by $\Eig_{\geq w}(f)$ the direct sum of all $\Eig_{w'}(f)$ with $w'\geq w$. \end{notation} Equipping each vector space $E|_{p_i}$ with the flag \[ E|_{p_i}=\Eig_{\geq w_1(p_i)}(\Res_{p_i}(\nabla))\supsetneq \Eig_{\geq w_2(p_i)}(\Res_{p_i}(\nabla))\supsetneq\dots\supsetneq \Eig_{\geq w_{b_i}(p_i)}(\Res_{p_i}(\nabla))\supsetneq\{0\} \] defines a parabolic structure on $E$ at $P$ of type $\bm{w}$. If a local section $s$ near $p_i$ satisfies $0\neq s(p_i)\in\Eig_w(\Res_{p_i}(\nabla))$, then $\ord_{p_i}\|s\|_H=w$ and so the parabolic structure just defined corresponds to the filtration \[ E_w = \left\{s\ |\ \|s\|_H\cdot |z_i|^{\e-w}\ \text{bounded near $p_i$ for all $\e>0$ and all $p_i\in P$} \right\}\subset E(\infty\cdot P) \] where $z_i$ is a local holomorphic coordinate on $S$ centered at $p_i$. Thus, $\Res_{p_i}(\nabla)$ belongs to the conjugacy class $\CL{M_i}\subset \u_N$ of \[ M_i=\left( \begin{array}{c|c|c} w_1(M_i)\id_{m_1(M_i)} & 0 & 0\\ \hline 0 & \ddots & 0\\ \hline 0 & 0 & w_{b_i}(M_i)\id_{m_{b_i}(M_i)} \end{array} \right) \] where $w_k(M_i)=w_k(p_i)$ and $m_k(M_i)=m_k(p_i)$, and the monodromy of $\nabla$ along $\pa_i$ belongs to the conjugacy class $\bco_i=\CL{\exp[-2\pi\sqrt{-1}M_i]}\subset\U_N$ of $\exp(-2\pi \sqrt{-1}M_i)$. \begin{notation} If $\bco_i=\CL{C_i}\subset \U_N$, then there exists a unique matrix $M_i\in \sqrt{-1}\cdot\u_N$ with real eigenvalues in $[0,1)$ such that $\exp(-2\pi\sqrt{-1}M_i)=C_i$. We will write $w_k(\bco_i):=w_k(M_i)$ and $m_k(\bco_i):=m_k(M_i)$ and $\bm{w}(\bCo)$ for the collection of all $w_k(M_i)$ and $m_k(M_i)$. \end{notation} We can now state the analogue of Theorem \ref{thm:narasimhan-seshadri} for punctured surfaces, whose proof is again by continuity method. \begin{theorem}[Mehta-Seshadri \cite{mehta-seshadri:parabolic}]\label{thm:mehta-seshadri} The map that sends a flat $\U_N$-bundle $(\xi,\nabla)$ on $\dot{S}$ to the Deligne extension $E_\bullet$ of the $I$-holomorphic vector bundle $\dot{E}=\xi\times_{\U_N}\Ocal_S^{\oplus N}$ induces a real-analytic homeomorphism \[ \xymatrix@R=0in{ \Flat^{irr}(\dot{S},\U_N,\bCo) \ar[rr]^{\sim}&& \Bun^s(S,\bm{w}(\bCo),N)_0 } \] between the moduli space of irreducible flat $\U_N$-principal bundles on $\dot{S}$ with monodromy along $\pa_i$ in $\bco_i$ and the moduli space of stable $I$-holomorphic parabolic vector bundles of rank $N$, type $\bm{w}(\bCo)$ and degree $0$ on $(S,I)$. Such homeomorphism restricts to \[ \xymatrix@R=0in{ \Flat^{irr}(\dot{S},\SU_N,\bCo)\ar[rr]^{\sim}&&\Bun^s(S,\bm{w}(\bCo),N,\Ocal_S)\\ (\xi,\nabla)\ar@{|->}[rr] && (E_\bullet,\eta) } \] where $\eta:\det(E_\bullet)\rar\Ocal_S$ sends the unit volume element to $1$. \end{theorem} As seen before, the statement extends to $b$-framed polystable parabolic bundles. \begin{corollary} There are real-analytic homeomorphisms \[ \xymatrix@R=0in{ \Flat_b^{dec}(\dot{S},\U_N,\bCo)\ar[rr]^{\sim} && \Bun^{ps}_b(S,\bm{w}(\bCo),N)_0\\ \Flat_b^{dec}(\dot{S},\SU_N,\bCo)\ar[rr]^{\sim} && \Bun^{ps}_b(S,\bm{w}(\bCo),N,\Ocal_S). } \] In both cases, irreducible flat bundles correspond to stable $I$-holomorphic bundles. \end{corollary} The counterpart to Theorem \ref{thm:donaldson-NS} for punctured surfaces was proven by Biquard using analytic methods. \begin{theorem}[Biquard \cite{biquard:paraboliques}] A holomorphic vector bundle $E_\bullet$ of rank $N$ and degree $0$ on $(S,I)$ with parabolic structure at $P$ of type $\bm{w}$ admits a flat invariant metric if and only if $E_\bullet$ is polystable. Moreover, such a metric is unique up to automorphisms of $E_\bullet$. \end{theorem} The above achievements (both in the closed and punctured case) culminate in the more general correspondence between flat $\GL_N$-bundles on punctured surfaces and parabolic Higgs bundles proven by Simpson \cite{simpson:harmonic}. Here we describe how one direction works, namely how to go from flat bundles to parabolic Higgs bundles. Given a $\GL_N$-bundle $\xi$ on $\dot{S}$, the induced flat vector bundle $\dot{V}=\xi\times_{\GL_N}\CC^N$ can be extended to $V\rar S$ in such a way that the real parts of the eigenvalues $\lambda_k(p_i)+i\nu_k(p_i)$ of $\Res_{p_i}(\nabla)$ satisfy $0\leq \lambda_1(p_i)<\dots<\lambda_{b_i}(p_i)<1$. Moreover, if $\xi$ has reductive monodromy, an adaptation of Corlette's theorem in the noncompact case (see also Labourie \cite{labourie:harmonic}) ensure the existence (and uniqueness up to isomorphism) of a {\it{tame}} harmonic metric $H$ on the flat vector bundle $\dot{V}=\xi\times_{\GL_N}\CC^N$. This means that $\nabla$ on $E:=V$ can be decomposed as $\nabla=\nabla^H+\Phi+\ol{\Phi}^H$, where $\nabla^H$ is compatible with $H$ as before, $\Phi$ and its $H$-adjoint $\ol{\Phi}^H$ have {\it{at worst simple poles at $P$}} and the {\it{Higgs field $\Phi\in C^\infty(S,K(P)\otimes\End(E_\bullet))$}} is holomorphic with respect to $\ol{\pa}^E=\ol{\pa}^V-\ol{\Phi}^H$. Furthermore, $E$ can be endowed with a parabolic structure at $P$ of type $\bm{w}$ defined by the filtration \[ E|_{p_i}=\Eig_{\geq w_1(p_i)}(\Res_{p_i}(\nabla))\supsetneq \Eig_{\geq w_2(p_i)}(\Res_{p_i}(\lambda))\supsetneq\dots\supsetneq \Eig_{\geq w_{b_i}(p_i)}(\Res_{p_i}(\nabla))\supsetneq\{0\} \] where $w_k(p_i)=\lambda_k(p_i)$ and $m_k(p_i)=\dim\,\Eig_{\lambda_k(p_i)}(\Res_{p_i}(\nabla))$. \begin{notation} Given a matrix $M=D+2M^0\in\gl_N(\CC)$ in Jordan form, with $D$ diagonal and $M^0$ nilpotent, we call $M'=\mathrm{Re}(D)+M^0$ and $M''=\sqrt{-1}\,\mathrm{Im}(D)+M^0$. \end{notation} It can be checked that, if $\Res_{p_i}(\nabla)$ belongs to $\CL{M_i}\subset \psl_N$ with $M_i$ in Jordan form and so the monodromy along $\pa_i$ belongs to $\bco_i=\CL{\exp(-2\pi\sqrt{-1}M_i)}$, then $\Res_{p_i}(\Phi)$ belongs to $\bcoa_i=\CL{M''_i/2}\subset\psl_N$. As before, we will denote by $\bm{w}(\bCo)$ the collection of all $w_k(M_i)$ and $m_k(M_i)$.\\ Summarizing our discussion, the correspondence preserves generalized eigenspaces of $\hol_\xi(\gamma_i)$, of $\Res_{p_i}(\nabla)$ and of $\Res_{p_i}(\Phi)$; inside a single generalized eigenspace it works as illustrated in the table below (borrowed from \cite{simpson:harmonic}), where $\varsigma$ is a local holomorphic section of $E$ that does not vanish at $p_i$. \[ \begin{array}{|c|c|c|} \hline & (E_\bullet,\Phi) & (V,\nabla) \\ \hline \text{jump at $p_i$} & \lambda & \lambda \\ \hline \text{residue eigenvalue at $p_i$} & \sqrt{-1}\,\nu/2 & \lambda+\sqrt{-1}\,\nu \\ \hline \end{array} \begin{array}{|c|c|} \hline \text{monodromy} & \ord_{p_i} \|\varsigma\|_H \\ \hline & \lambda\\ \exp\big[-2\pi \sqrt{-1}(\lambda+\sqrt{-1}\,\nu)\big] & \\ \hline \end{array} \] \begin{remark} The order of growth of $\|\varsigma\|_H$ near $p_i$ may have logarithmic factors as in Example \ref{example:log} (see \cite{simpson:harmonic}). More precisely, if $\varsigma(p_i)$ takes values in a subspace of $V|_{p_i}$ corresponding to a Jordan block of $\Res_{p_i}(\nabla)$ of size $m$ and eigenvalue $\lambda+\sqrt{-1}\,\nu$, then \[ \|\varsigma\|_H\sim |z_i|^{\lambda}|\log|z_i||^{l-\frac{m+1}{2}} \] where $z_i$ is a local coordinate on $S$ at $p_i$ and $l>0$ is the smallest integer such that $\left[\Res_{p_i}(\nabla)-\left(\lambda+\sqrt{-1}\,\nu\right)\id\right]^l s(p_i)=0$. \end{remark} The above set-theoretic correspondence can be promoted to a real-analytic one. Similarly to the closed case treated by Hitchin and Simpson, the moduli spaces of flat bundles and of Higgs bundles are different holomorphic manifestations of the same hyperk\"ahler manifold: the moduli space of harmonic bundles. This is proven by Konno \cite{konno:construction} for Higgs fields with nilpotent residues and by Biquard-Boalch \cite{biquard-boalch:wild} in general. \begin{theorem}[Simpson \cite{simpson:harmonic}, Konno \cite{konno:construction}, Biquard-Boalch \cite{biquard-boalch:wild}]\label{thm:simpson-konno} Let $M_i\in\gl_N$ be a matrix in Jordan form and let $\bco_i=\CL{\exp(-2\pi\sqrt{-1}M_i)}$ be a conjugacy class in $\GL_N$ and $\bcoa_i=\CL{M''_i/2}\subset\gl_N$ for $i=1,\dots,n$.\\ There is a real-analytic diffeomorphism \[ \xymatrix@R=0in{ \Flat_b^{red}(\dot{S},\GL_N,\bCo)\ar[rr]^{\sim} && \Higgs^{ps}_b(S,\bm{w},N,\bCoa)_0\\ } \] between the moduli space of $b$-framed flat $\GL_N$-bundles $(\xi,\nabla,\tau)$ on $\dot{S}$ with reductive monodromy and $\hol_\xi(\pa_i)\in\bco_i$ and the moduli space of $b$-framed polystable Higgs bundles $(E_\bullet,\eta,\Phi,\tau)$ on $(S,I)$ of parabolic type $\bm{w}=\bm{w}(\bCo)$ at $P$ and degree $0$ with $\Res_{p_i}(\Phi)\in\bcoa_i$.\\ Under such diffeomorphism, flat bundles with Zariski-dense monodromy correspond to stable parabolic Higgs bundles, and so the induced \[ \xymatrix@R=0in{ \Flat^{Zd}(\dot{S},\GL_N,\bCo)\ar[rr]^{\sim} && \Higgs^{s}(S,\bm{w},N,\bCoa)_0\\ } \] is a real-analytic diffeomorphism too. Similarly, via \[ \xymatrix@R=0in{ \Flat_b^{red}(\dot{S},\SL_N,\bCo)\ar[rr]^{\sim} && \Higgs^{ps}_b(S,\bm{w},N,\Ocal_S,\bCoa)\\ \Flat^{Zd}(\dot{S},\SL_N,\bCo)\ar[rr]^{\sim} && \Higgs^{s}(S,\bm{w},N,\Ocal_S,\bCoa)\\ } \] $\SL_N$-bundles correspond to parabolic Higgs bundles $(E_\bullet,\Phi)$ endowed with a trivialization $\eta:\det(E_\bullet)\arr{\sim}\Ocal_S$. \end{theorem} \subsection{Correspondence and the real locus in rank $2$} Following Hitchin \cite{hitchin:self-duality}, consider a point $(E_\bullet,\eta,\Phi)$ in $\Higgs^s(S,\bm{w},2,\Ocal_S,\bCoa)$ fixed by the involution $\sigma$ that sends $\Phi$ to $-\Phi$. By Lemma \ref{lemma:real}, $(E_\bullet,\eta,\Phi)$ may belong to $\Bun^s(S,\bm{w},2,\Ocal_S)$ or to some $\Higgs^s(S,\bm{w},2,\Ocal_S,\bCoa)(\RR)_{d,\bm{a}}$. In the former case, $\Phi=0$ and so the point corresponds to a flat $\SU_2$-bundle by Theorem \ref{thm:mehta-seshadri}. In the latter case, $\Phi\neq 0$ and we assume $e(d,\bm{a},\bm{w})>0$. Then Lemma \ref{lemma:real} provides a splitting $E_\bullet=L^\vee_\bullet\oplus L_\bullet$ and an automorphism $\iota$ of $E_\bullet$ that preserves the splitting and that sends $\Phi$ to $-\Phi$. Moreover, such splitting and $\iota$ are essentially unique, since $(E_\bullet,\Phi)$ is stable. Let $\dot{V}:=\dot{E}$ and let $\nabla$ (resp. $\nabla'$) be the flat connection on $\dot{V}$ associated to $(E_\bullet,\Phi)$ (resp. $(E_\bullet,-\Phi)$). It is easy to see that the flat vector bundles $(\dot{V},\nabla)$ and $(\dot{V},\nabla')$ support the same tame harmonic metric $H$, which means that $\nabla=\nabla^H+\Phi+\ol{\Phi}^H$ and $\nabla'=\nabla^H-\Phi-\ol{\Phi}^H$. Since $\iota^*(-\Phi)=\Phi$ and $\iota^*(\nabla')=\nabla$, the automorphism $\iota$ preserves $\nabla^H$ and so is an $H$-isometry; as a consequence, $L_\bullet$ and $L^\vee_\bullet$ are $H$-orthogonal and $H$ induces the identification $\ol{L}_\bullet \cong L^\vee_\bullet$. Since the anti-linear involution \[ T:\xymatrix@R=0in{ L^\vee_\bullet\oplus L_\bullet \ar[r] & L^\vee_\bullet\oplus L_\bullet\\ (\alpha,\beta)\ar@{|->}[r] & (\ol{\beta},\ol{\alpha}) } \] satisfies $T\circ\Phi=\ol{\Phi}^H\circ T$ and so commutes with $\nabla$, the monodromy of $\nabla$ preserves $\dot{V}(\RR):=\mathrm{Fix}(T)\subset\dot{V}$ and so it defines a representation $\rho_{_\RR}$ that takes values in $\SL_2(\RR)$. Moreover, $\dot{V}(\RR)\hra \dot{V}\cong \dot{L}\oplus \dot{\ol{L}}\cong \dot{V}(\RR)\otimes\CC$ and we identify $\dot{V}(\RR)$ to $\dot{L}$. As a consequence, if $\rho_{_\RR}(\gamma_i)$ is elliptic and $\{\rot\}(\rho_{_\RR}(\gamma_i))=\{r_i\}$, then $\{r_i\}=\deg_{p_i}\{L^{-2}_\bullet\}$. Notice that the power $2$ appears because $\SL_2(\RR)\rar\PSL_2(\RR)$ is a cover of degree $2$. A version of the following can be found for instance in Section 3.6 of \cite{burger-iozzi-wienhard:higher}. \begin{lemma}[Euler number as a first Chern class]\label{lemma:chern} The parabolic degree of $L_\bullet$ and the Euler number of $\rho$ satisfy $\Eu(\rho) =2\deg(L_\bullet)=2\deg(L)+2\NO{\bm{w_L}}$. \end{lemma} The above discussion then leads to the following result, the bound on $d$ being a consequence of Proposition \ref{prop:topology}(a) and Lemma \ref{lemma:chern}. \begin{theorem}[Correspondence for $\SL_2(\RR)$]\label{thm:correspondence} For every $i=1,\dots,n$, let \begin{itemize} \item $\bco_i$ be a conjugacy class in $\SL_2(\RR)$ \item $\bcoa_i$ be a conjugacy class in $\psl_2(\CC)$ \item $w_1(p_i)\in[0,1/2]$ and $\e:J_{nil}=\{j\,|\,\bcoa_j\ \text{nilpotent}\}\rar \{+,-\}$ \item $a_i\in\{0,1\}$ \end{itemize} that match according to the following table \begin{center} \[ \hspace{-1cm}\begin{array}{|c|c|c|c|c|} \hline & a & \bCoa & \bCo & w_L\\ \hline\hline \text{deg.} & 0 & 0 & (-1)^{2w_1}\id & w_1=0;\frac{1}{2}\\ \hline \begin{array}{c}\text{deg.}\\ \e=\pm 1\end{array} & 0 & \left(\begin{array}{cc}0 & 1\\0 & 0\end{array}\right) & (-1)^{2w_1}\left(\begin{array}{cc}1 & 0\\ \e& 1\end{array}\right) & w_1=0;\frac{1}{2}\\ \hline \begin{array}{c}\text{deg.}\\ \ell>0\end{array} & 0 & \left(\begin{array}{cc}\sqrt{-1}\,\ell/8\pi & 0\\0 & -\sqrt{-1}\,\ell/8\pi\end{array}\right) & (-1)^{2w_1}\left(\begin{array}{cc}\exp(\ell/2) & 0\\0 & \exp(-\ell/2)\end{array}\right) & w_1=0;\frac{1}{2}\\ \hline \text{non-deg.} & \{0,1\} & 0 & \left(\begin{array}{cc}\cos(2\pi w_L) & \sin(2\pi w_L)\\-\sin(2\pi w_L) & \cos(2\pi w_L)\end{array}\right) & a+(-1)^a w_1\\ \hline \end{array} \] \captionof{table}{} \end{center} Let $d\in\ZZ$ such that \[ -\NO{\bm{w_L}}<d\leq g-1+\frac{s\even-\NO{\bm{a}}-s_0-s_-(\bm{\e})}{2} \] where $s\even=\#\{j\,|\,w_1(p_j)=0\}$, $s_0=\#\{i\,|\,\bco_i=\{\id\}\}$ and $s_-(\bm{\e})=\#\{j\in J_{nil}\,|\,\e_j=-\}$.\\ Then there are real-analytic diffeomorphisms \[ \xymatrix@R=0in{ \Rep(\dot{S},\SL_2(\RR),\bCo)_e\ar[rr]^{\sim} && \Higgs(S,\bm{w},2,\Ocal_S,\bCoa)(\RR)^{\bm{\e}}_{d,\bm{a}}\\ \Rep(\dot{S},\PSL_2(\RR),\bCo)_e\ar[rr]^{\sim\qquad} && \Higgs(S,\bm{w},2,\Ocal_S,\bCoa)(\RR)^{\bm{\e}}_{d,\bm{a}}/\Pic^0(S)[2]\\ } \] where $e=2d+2\sum_{i=1}^n w_L(p_i)>0$ is the Euler number of the associated oriented $\RR\PP^1$-bundle. Moreover, the above maps extend to homeomorphisms \[ \xymatrix@R=0in{ \Rep(\dot{S},\SL_2(\RR),\ol{\bCo})_e\ar[rr]^{\sim} && \Higgs(S,\bm{w},2,\Ocal_S,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}\\ \Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e\ar[rr]^{\sim\qquad} && \Higgs(S,\bm{w},2,\Ocal_S,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}/\Pic^0(S)[2]\\ } \] for the closures $\ol{\bCo}$ of $\bCo$ and $\ol{\bCoa}$ of $\bCoa$. \end{theorem} \begin{proof} We are only left to prove the last assertion, namely that the map $\Rep(\dot{S},\SL_2(\RR),\ol{\bCo})_e\rar \Higgs(S,\bm{w},2,\Ocal_S,\ol{\bCoa})(\RR)^{\bm{\e}}_{d,\bm{a}}$ is a homeomorphism. We already know that it is bijective and we want to show that it is continuous and proper. Endow $\dot{S}$ with the unique $I$-conformal hyperbolic metric of finite area and its universal cover $\widetilde{\dot{S}}$ with the pull-back metric. Let $S^\circ$ be the compact subsurface obtained from $S$ by removing small open disk neighborhoods of the marked points and let $\widetilde{S}^\circ$ be its preimage inside the universal cover $\widetilde{\dot{S}}$ and $F\subset\widetilde{\dot{S}}$ be a fundamental domain for the action of $\pi_1(\dot{S})$. We recall that, for every equivariant harmonic map $\tilde{h}:\widetilde{\dot{S}}\rar \HH^2\cong \SL_2(\RR)/\SO_2(\RR)$, we have \begin{equation}\tag{$\star$}\label{eq} \left|\nabla \tilde{h}\right|^2\leq C\cdot \mathcal{E}_{S^\circ}(\tilde{h}) \end{equation} at every point of $\widetilde{S}^\circ\cap F$, where $C$ depends only on the diameter of $\widetilde{S}^\circ\cap F$ and $\mathcal{E}_{S^\circ}(\tilde{h})$ is the energy of the restriction of $\tilde{h}$ to $\widetilde{S}^\circ\cap F$. Consider a sequence $\rho^{(k)}$ of representations in $\Rep(\dot{S},\SL_2(\RR),\ol{\bCo})_e$ and let $(E^{(k)}_\bullet,\eta^{(k)},\Phi^{(k)})$ be the corresponding Higgs bundle. By Lemma \ref{lemma:real}, the bundle $E^{(k)}$ is isomorphic to $(L^{(k)}_\bullet)^\vee\oplus L^{(k)}_\bullet$, with $L^{(k)}$ of fixed degree $d$. Suppose now that $\rho^{(k)}\rar\rho$ and let $(E_\bullet=L^\vee_\bullet\oplus L_\bullet,\eta,\Phi)$ the Higgs bundle determined $\rho$. By Proposition 2.6.1 of \cite{korevaar-schoen:1}, there are $\rho^{(k)}$-equivariant harmonic maps $\tilde{h}^{(k)}:\widetilde{\dot{S}}\rar \HH^2$ that are locally equi-bounded and equi-Lipschitz. Thus, $\tilde{h}^{(k)}$ locally Lipschitz converges to the unique $\rho$-equivariant harmonic map $\tilde{h}$. Thus, $\ol{\pa}^{L^{(k)}}\rar \ol{\pa}^{L}$ and $\Phi^{(k)}\rar \Phi$ uniformly on the compact subsets of $\dot{S}$. Since the parabolic structure is fixed, this implies that $L^{(k)}\rar L$ and so $\Phi^{(k)}\rar \Phi$. This proves continuity of the correspondence. Suppose finally that $\rho^{(k)}$ is divergent. We want to show that $\|\Phi^{(k)}|_{S^\circ}\|^2$ is divergent. By contradiction, up to extracting a subsequence, $\mathcal{E}_{S^\circ}(\tilde{h}^{(k)}) =2\|\Phi^{(k)}|_{S^\circ}\|^2$ would be bounded. By the locally uniform bound \eqref{eq}, the $\tilde{h}^{(k)}|_{\widetilde{S}^\circ\cap F}$ would be equi-Lipschitz and so $\rho^{(k)}$ would have a convergent subsequence. \end{proof} Notice that $\bco_i$ determines $a_i$, $w_1(p_i)$ and $\bcoa_i$ (and the sign $\e_i$, if $\bcoa_i$ is nilpotent) and vice versa. Thus we can also draw the following conclusion. \begin{corollary}[Components of $\PSL_2(\RR)$-representations]\label{cor:euler-components} Connected components of $\Rep(\dot{S},\PSL_2(\RR),\bCo)$ with Euler number $e=2d+2\NO{\bm{w_L}}>0$ are classified by the integers $d$ such that $-\NO{\bm{w_L}}<d\leq g-1+\frac{1}{2}\left(s\even-\NO{\bm{a}}-s_0-s_-(\bm{\e})\right)$. \end{corollary} The topology of $\Higgs(S,\bm{w},2,\Ocal_S,\bCoa)(\RR)_{d,\bm{a}}$ and of its quotient by $\Pic^0(S)[2]$ is described in Proposition \ref{prop:topology}. \subsection{Uniformization components} Let $\bm{\ell}=(\ell_1,\dots,\ell_n)$ with $\ell_i=\sqrt{-1}\th_i$ and $\th_i>0$ for $i=1,\dots,k$ and $\ell_i\geq 0$ for $i=k+1,\dots,n$. Call $s_{0}=\#\{ i\in\{1,\dots,k\}\,|\, \th_i\in 2\pi\NN_+\}$. A consequence of the above work is the following result stated in the introduction. \begin{corollary}[Topology of uniformization components]\label{cor:uniformization} Assume $e_{\bm{\ell}}>0$ and consider the monodromy map \[ \hol\circ\Xi_{\bm{\ell}}:\Ycal(\dot{S},\bm{\ell})\lra \Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}}. \] The space $\Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}}$ is real-analytically diffeomorphic to a holomorphic affine bundle of rank $3g-3+n-m$ over $\mathrm{Sym}^{m-s_0}(S\setminus \{p_{k+1},\dots,p_n\})$, with $m=\sum_{1\leq i\leq k}\floor{\frac{\th_i}{2\pi}}$. \end{corollary} \begin{proof} The monodromy map takes values in $\Rep(\dot{S},\PSL_2(\RR),\bCo_{\bm{\ell}})_{e_{\bm{\ell}}}$ by Proposition \ref{prop:uniformization}. The result then follows from Theorem \ref{thm:correspondence} and Proposition \ref{prop:topology}, remembering that cusps correspond to positive unipotents. \end{proof} In \cite{deroin-tholozan:super-maximal} Deroin-Tholozan show that ``super-maximal'' components of $\PSL_2(\RR)$-representations consist of monodromies of hyperbolic metrics. Deroin told me that this result can be recovered using Corollary \ref{cor:supermaximal} and the analysis carried out in Section \ref{sec:topology} as follows. \begin{corollary}\label{cor:geometric} Fix $\bm{\bCo}=(\bco_1,\dots,\bco_n)$ and $e>0$ and assume $n>3+s_0$. Then $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e$ is compact if and only if \[ \begin{cases} g=0\\ \text{no $\bco_i$ is hyperbolic or negative unipotent}\\ e=1-\NO{\{\bm{r}\}}\in (0,1]. \end{cases} \] Moreover, in this case every representation in $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e$ is the monodromy of some hyperbolic metric on $\dot{S}$. \end{corollary} \begin{proof} By Corollary \ref{cor:supermaximal}, representations lying in compact components cannot have hyperbolic or negative unipotent boundary monodromy. Fix a complex structure $I$ on $S$. Let $\Det$ be a line bundle of degree $d_0=0$ if $n-1$ is even, or of degree $d_0=1$ if $n-1$ is odd. Fix $d=(d_0+1-n)/2$. Let $a_1=\dots=a_n=0$ and let $w_1(p_i)=\frac{1}{2}\left(1-\{r_i\}\right)$. Finally, pick the conjugacy classes $\bCoa_i$ to be $\{0\}$ if $c_i$ is elliptic, and to be nilpotent if $c_i$ is positive unipotent (and, in this case, we set $\e_i=+$). By Theorem \ref{thm:correspondence}, the component $\Rep(\dot{S},\PSL_2(\RR),\ol{\bCo})_e$ is homeomorphic to $\Higgs(S,\bm{w},2,\Ocal_S,\ol{\bCoa})_{d,\bm{a}}^{\bm{\e}}$ and so the first claim follows from Corollary \ref{cor:supermaximal}. As for the last claim, remember that all Higgs bundles $[E,\Phi]$ in the compact $\Higgs(S,\bm{w},2,\Ocal_S,\ol{\bCoa})_{d,\bm{a}}^{\bm{\e}}$ have $\psi=0$ and so $\Phi$ is nilpotent. The harmonic section $h$ of $\xi_\rho\times_{\SL_2(\RR)}(\SL_2(\RR)/\SO_2(\RR))$ constructed by Donaldson corresponds to a $\rho$-equivariant harmonic map $\tilde{h}:\widetilde{\dot{S}}\rar \HH^2\cong \SL_2(\RR)/\SO_2(\RR)$. Since the quadratic differential $\det(\Phi)$ pulls back to $\widetilde{\dot{S}}$ to the Hopf differential of $\tilde{h}$, the vanishing $\det(\Phi)=0$ implies that $\tilde{h}$ is conformal. Hence, the pull-back via $\tilde{h}$ of the hyperbolic metric on $\HH^2$ descends to a conformal hyperbolic metric on $(\dot{S},I)$, possibly with extra conical points of angles in $2\pi\NN_+$, and with monodromy $\rho$. \end{proof} The case $(g,n)=(0,3+s_0)$ of the pair of pants can be done by hands.\\ The following result by McOwen \cite{mcowen:metric} and Troyanov \cite{troyanov:prescribing} is a version of Koebe's uniformization theorem \cite{koebe:first} \cite{koebe:second} for hyperbolic surfaces with conical singularities. \begin{theorem}[Uniformization with conical singularities]\label{thm:troyanov} For every $\th_1,\dots,\th_n\geq 0$ such that $e_{\sqrt{-1}\bm{\th}}>0$, there exists exactly one metric on $\dot{S}$ with conical singularity of angle $2\pi\th_i$ at $p_i$ (or with a cusp at $p_i$, if $\th_i=0$) and which is $I$-conformal. \end{theorem} Mimicking Hitchin's computation \cite{hitchin:self-duality} in the case of a closed surface, we then have the following expected consequence. \begin{corollary}[Uniformization Higgs bundles] Assume $k=n$ and let $\delta_i=\frac{\th_i}{2\pi}-1$. Then the monodromy representation $\hol_h:\pi\rar\PSL_2(\RR)$ of the unique hyperbolic metric $h$ in $\Ycal(\dot{S},\bm{\ell})$ with conical singularities of angle $2\pi\th_i$ at $p_i$ which is {\it{$I$-conformal}} corresponds to the $\Pic^0(S)[2]$-equivalence class of the parabolic Higgs bundle $(E_\bullet,\Phi)$, with \[ E_\bullet=L_\bullet^\vee\oplus L_\bullet, \quad \Phi=\left( \begin{array}{cc} 0 & 1\\ 0 & 0 \end{array} \right) \] where $L_\bullet=B(-\frac{1}{2}\bm{\delta}\cdot P)$ and $B^{2}\cong K_S$, and so $L_\bullet^{-2}K(P)=T_S(\bm{\delta}\cdot P)\otimes K_S(P)$. \end{corollary} \begin{proof} It is enough to notice that the harmonic metric the $\RR^2$-bundle $\dot{V}\rar\dot{S}$ with monodromy $\hol_h$ is provided by the (equivariant) developing map $(\wti{\dot{S}},\wti{h})\rar \HH^2=\SL_2(\RR)/\SO_2(\RR)$ itself, and then follow Hitchin's computation. \end{proof} \bibliographystyle{amsplain}
7c7871cedea49617171941f63cecc55b2bb183f0
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Magneto-hydrodynamics (MHD) forms the basis of large-scale description of physics of dense plasma in compact stars. A key quantity in the dissipative formulations of MHD is the conductivity of matter. It determines, for example, the dissipation of currents and therefore the decay of magnetic fields, the dispersion of plasma waves, etc. In turn, magnetic field decay affects the rotational and thermal evolutions of neutron stars and consequently a broad array of their observational manifestations. Transport in compact star plasma was studied traditionally in the cold (essentially zero-tem\-pe\-rature) and dense regime where the constituents form degenerate quantum liquids. This regime is relevant for mature isolated or accreting neutron stars as well as interiors of white dwarfs. The dilute and warm (non-zero temperature) regime is of interest in the context of transient, short-lived states of neutron stars, such as proto-neutron stars newly born in supernova explosions or hypermassive remnants formed in the aftermath of neutron star binary mergers. We start this article with an overview of the transport calculations of electrical conductivity of compact star matter in the density regime corresponding to their outer crusts ($\rho \le 10^{11}$ g cm$^{-3}$). Then we go on to describe our recent effort to calculate the electrical conductivity of non-zero temperature crustal plasma. We focus on sufficiently high temperatures where nuclei form a liquid coexisting with electronic background of arbitrary degeneracy. We close this review with a summary and outlook. Below we use the natural (Gaussian) units with $\hbar= c = k_B = k_e = 1$, $e=\sqrt{\alpha}$, $\alpha=1/137$ and the metric signature $(1,-1,-1,-1)$. \section{Overview} At densities relevant to interiors of white dwarfs and neutron star crusts the electron-ion system is in a plasma state - the ions are fully ionized while free electrons are the most mobile carriers of charge. By charge conservation electron density is related to the ion charge $Z$ by $n_e=Zn_i$, where $n_i$ is the number density of nuclei. Electrons to a good accuracy form non-interacting gas which becomes degenerate below the Fermi temperature $T_F = \varepsilon_F-m = (p_F^2+m^2)^{1/2}-m$, where the electron Fermi momentum is given by $p_F = (3\pi^2n_e)^{1/3}$ and $m$ is the electron mass. The state of ions (mass number $A$ and charge $Z$) is controlled by the value of the Coulomb plasma parameter $\Gamma$ \begin{eqnarray}\label{eq:Gamma} \Gamma=\frac{e^2 Z^2}{Ta_i}\approx 22.73 \frac{Z^2}{T_6}\bigg(\frac{\rho_6}{A}\bigg)^{1/3}, \end{eqnarray} where $e$ is the elementary charge, $T$ is the temperature, $a_i=(4\pi n_i/3)^{-1/3}$ is the radius of the spherical volume per ion, $T_6$ is the temperature in units $10^6$ K and $\rho_6$ is the density in units of $10^6$ g cm$^{-3}$. If $\Gamma\ll 1$ or, equivalently $T\gg T_{\rm C}\equiv Z^2e^2/a_i$, ions form weakly coupled Boltzmann gas. In the regime $\Gamma\ge 1$ ions are strongly coupled and form a liquid for low values of $\Gamma\leq\Gamma_m\simeq 160$ and a lattice for $\Gamma>\Gamma_m$. The melting temperature of the lattice associated with $\Gamma_m$ is defined as $T_m=(Ze)^2/\Gamma_ma_i$. For temperatures below the ion plasma temperature \begin{eqnarray} T_p = \biggl(\frac{4\pi Z^2e^2n_i}{M }\biggl)^{1/2} , \end{eqnarray} where $M $ is the ion mass, the quantization of oscillations of the lattice becomes important. Figure \ref{fig:PhaseDiagram} shows the temperature-density phase diagram of the crustal material in the cases where it is composed of iron $\isotope[56]{Fe}$ (left panel) or carbon $\isotope[12]{C}$ (right panel). \begin{figure}[t] \begin{center} \includegraphics[width=8.0cm,keepaspectratio]{phase_diagram2.eps} \caption{ The temperature-density phase diagram of dense plasma composed of iron $\isotope[56]{Fe}$ (upper panel) and carbon $\isotope[12]{C}$ (lower panel). The electron gas degeneracy sets in below the Fermi temperature $T_F$ (short dashed lines). The ionic component solidifies below the melting temperature $T_m$ (solid lines), while quantum effects become important below the plasma temperature (dash-dotted lines). For temperatures above $T_{\rm C}$ (long dashed lines) the ionic component forms a Boltzmann gas. Note that for $\isotope[12]{C}$ the quantum effects become important in the portion of the phase diagram lying between the lines $T_p(\rho)$ and $T_m(\rho)$. The present study does not cover the shaded portion of the phase diagram. } \label{fig:PhaseDiagram} \end{center} \end{figure} \begin{figure}[t] \begin{center} \includegraphics[width=8.0cm,keepaspectratio]{phase_diagram_comp_pearson.eps} \caption{ Same as in Fig.~1 but for density dependent composition; for details and references see the discussion in Ref.~[12]. } \label{fig:PhaseDiagram2} \end{center} \end{figure} \hskip 0.5cm While the structure of the phase diagrams for $\isotope[56]{Fe}$ and $\isotope[12]{C}$ are similar there is an important difference as well: as the temperature is lowered the quantum effects become important for carbon prior to solidification, whereas iron solidifies close to the temperature where ionic quantum effects become important. Except of hydrogen and perhaps helium which may not solidify because of quantum zero point motions all heavier elements $Z> 2$ solidify at low enough temperature. The phase diagram in the case of density dependent composition is shown in Fig.~\ref{fig:PhaseDiagram2}. The earliest studies of transport in dense matter go back to the work by Mestel, Hoyle \cite{1950PCPS...46..331M} and Lee~\cite{1950ApJ...111..625L} in 1950s, who obtained the ``conductive opacity'', or equivalently the thermal conductivity of the electron-ion plasma in non-relativistic electron regime in the context of radiative and thermal transport in white dwarfs. Above densities of the order of $10^6$ g cm$^{-3}$ electrons are relativistic. Following the initial qualitative estimates of the conductivity of highly compressed matter by Abrikosov in 1963~\cite{1964Abrikosov} more detailed calculations were carried out in the 1970s by many authors. In particular the transport in neutron star crusts in the relativistic electron regime was studied in much detail by Flowers and Itoh~\cite{1976ApJ...206..218F} both in the solid and in the liquid regime using a variational method. They were able to cover a broad range of densities and low-temperature regime including multiple channels of scattering, relatively accurate description of collective modes (phonons) which contribute to the transport coefficients in the solid phase. Their discussion also extended to the neutron drip region where free neutrons contribute to the thermal conductivity and shear viscosity of matter. A critical analysis of the numerical values for the transport coefficients found by different authors was given by Yakovlev and Urpin~\cite{1980SvA....24..303Y}, who also provided useful and simple approximations for the transport coefficients in the degenerate electron regime in terms of the Coulomb logarithm. Nandkumar and Pethick~\cite{1984MNRAS.209..511N} studied the temperature regime above the melting temperature, i.e., where ions form a liquid, showing that the screening of electron-ion interactions can lead to substantial corrections in this case. These calculations agree with those of Itoh et al.~\cite{1983ApJ...273..774I} who also provide useful fitting formulae for the transport coefficient. Subsequent refinements of the results quoted above included, among other things, multi-phonon process and Debye-Waller factor~\cite{1999A&A...346..345P} in the solid phase and improved correlation functions in the liquid phase~\cite{2008ApJ...677..495I}. The implementation of the transport coefficients of dense matter in the dissipative MHD equations in the case of cold neutron crust plasma in the presence of magnetic fields were discussed by a number of authors \cite{1979ApJ...227..995E,1987Ap.....26..295S}. We do not discuss here the physics in ultra-strong fields and confine our attention to non-quantizing fields, i.e., fields below the critical field $B\simeq 10^{14}$~G above which the Landau quantization of electron trajectories becomes important. \section{Formalism} The Boltzmann equation for electron distribution function is given by \begin{eqnarray}\label{eq:Boltzmann} \frac{\partial f}{\partial t}+ {\bm V}\frac{\partial f}{\partial\bm r}- e(\bm E+{\bm V} \times \bm H)\frac{\partial f} {\partial{\bm p}}=I[f], \end{eqnarray} where ${\bm V}$ is the electron velocity, ${\bm E}$ and ${\bm H}$ are the electric and magnetic fields and $I$ is the collision intergal. We are interested in the regime where the collision integral describes electron-ion scattering and, therefore, has the form \begin{eqnarray}\label{eq:collision} I[f]=-(2\pi)^4\sum\limits_{234}|M_{12\to 34}|^2 \delta^{(4)}(p+p_2-p_3-p_4) [f(1-f_3)g_2-f_3(1-f)g_4], \end{eqnarray} where $p_i$ are four-momenta of particles, $g$ is the equilibrium distribution function of ions, which to a good accuracy can be described by the Maxwell-Boltzmann distribution with energy spectrum $\varepsilon=p^2/2M$, where $M$ is the ion mass. The sum Eq.~\eqref{eq:collision} stands symbolically for the integrals over the phase-space of scattering particles, $M_{12\to 34}$ is the transition matrix element for scattering of relativistic electrons off correlated ions and is given by \begin{eqnarray}\label{eq:matrix_element} M_{12\to 34}=\frac{J_0J'_0}{q^2+\Pi_l(\omega,{\bm q})}- \frac{\bm J_t\bm J'_t}{q^2-\omega^2+\Pi_t(\omega,{\bm q})}, \end{eqnarray} where the electron and ion four-currents are given, respectively, \begin{eqnarray}\label{eq:currents} J^{\mu}=-e^*\bar{u}^{s_3}(p_3)\gamma^\mu u^s(p),\quad J'^{\mu}=Ze^*V'^{\mu}=Ze^*(1,{\bm p}'/M), \end{eqnarray} $e^* = \sqrt{4\pi}e$, and $J_t, J'_t$ are the components of the currents transversal to the moment transfer ${\bm q}$, $\Pi_l(\omega,{\bm q})$ and $\Pi_t(\omega,{\bm q})$ are the longitudinal and transverse components of the polarization tensor, which describe respectively, the (irreducible) self-energies of longitudinal and transverse photons in the medium (plasma). The form of the matrix element \eqref{eq:matrix_element} includes thus the dynamical screening of the electron-ion interaction due to the exchange of transverse photons. Such separation has been employed in the treatment of transport in unpaired \cite{1993PhRvD..48.2916H} and superconducting quark matter~\cite{2014PhRvC..90e5205A} and we adopt an analogous approach here. We linearize the Boltzmann equation \eqref{eq:Boltzmann} by writing \begin{eqnarray}\label{eq:perturbation} f= f^0+\delta f,\quad \delta f=-\Phi \frac{\partial f^0}{\partial\varepsilon}, \end{eqnarray} where $f^0$ is the equilibrium Fermi-Dirac distribution function, $\delta f \ll f^0,$ and $\Phi$ is the perturbation. The electric field appears in the drift term of linearized Boltzmann equation at $O(1)$ in perturbation, whereas the term involving magnetic field at order $O(\Phi)$, because $ [{\bm V}\times {\bm H}]({\partial f^0}/{\partial{\bm p}})\propto [{\bm V}\times {\bm H}]{\bm V}=0. $ We next specify the form of the function $\Phi$ in the case of conduction as $ \Phi={\bm p}\cdot {\bm \Xi}(\varepsilon), $ which after substitution in the linearized Boltzmann equation gives \begin{eqnarray}\label{eq:boltzmann2} e{\bm V}\cdot \left[{\bm E}+({\bm \Xi}\times{\bm H})\right]=- {\bm \Xi}\cdot {\bm p}~\tau^{-1}(\varepsilon), \end{eqnarray} where the relaxation time, which depends on electron energy $\varepsilon$, is defined by \begin{eqnarray}\label{eq:t_relax} \tau^{-1}(\varepsilon)=(2\pi)^{-5} \int d\omega d\bm q\int d\bm p_2|{M}_{12\to 34}|^2 \frac{\bm q\cdot \bm p}{p^2} \delta(\varepsilon-\varepsilon_3-\omega)\delta(\varepsilon_2-\varepsilon_4+\omega) g_2\frac{1-f^0_3}{1-f^0}. \end{eqnarray} (Here and below the indices 2 and 4 are reserved for ions, the index 3 corresponds to the outgoing electron). In transforming the linearized collision integral we introduced a dummy integration over energy and momentum transfers, i.e., $\omega = \varepsilon-\varepsilon_3$ and ${\bm q} = {\bm p}-{\bm p}_3$. It remains to express the vector ${\bm \Xi}$ describing the perturbation in terms of physical fields; its most general decomposition is given by \begin{eqnarray}\label{eq:Xi} {\bm \Xi}=\alpha{\bm e}+\beta{\bm h}+\gamma[{\bm e}\times{\bm h}], \end{eqnarray} where ${\bm h} \equiv {\bm H}/H$ and ${\bm e} \equiv {\bm E}/E$ and the coefficients $\alpha$, $\beta$, $\gamma$ are functions of the electron energy. Substituting Eq.~\eqref{eq:Xi} in Eq.~\eqref{eq:boltzmann2} one finds that $\alpha=-eE\tau /\varepsilon (1+\omega^2_c\tau^2)$, $\beta/\alpha=(\omega_c\tau)^2({\bm e}\cdot {\bm h})$ and $\gamma/\alpha=-\omega_c\tau$, where $\omega_c=eH\varepsilon^{-1}$ is the cyclotron frequency. As a result, the most general form of the perturbation is given by \begin{eqnarray}\label{eq:phi} \Phi=-\frac{e\tau}{1+(\omega_c\tau)^2} V_i\left[\delta_{ij}-\omega_c\tau\varepsilon_{ijk} h_k+(\omega_c\tau)^2h_ih_j\right]E_j. \end{eqnarray} Using the standard expression for the electrical current in terms of the perturbation $\Phi $ we arrive at the conductivity tensor $ \sigma_{ij}=\delta_{ij}\sigma_0-\varepsilon_{ijm}h_m \sigma_1 +h_ih_j\sigma_2, $ where the components of the tensor are defined as \begin{eqnarray}\label{eq:sigma2} \sigma_n=\frac{e^2}{3\pi^2 T}\int_m^\infty\!\! d\varepsilon \frac{p^3}{\varepsilon}\frac{\tau(\omega_c\tau)^n} {1+(\omega_c\tau)^2}f^0(1-f^0),\quad n=0,1,2, \end{eqnarray} where $T$ is the temperature and the lower bound of the integral is given by the mass of the electron, which vanishes in the ultra-relativistic limit. The conductivity tensor has a particularly simple form if the magnetic field is along the $z$-direction \begin{eqnarray}\label{eq:sigma3} \hat{\sigma}= \begin{pmatrix} \sigma_0 & -\sigma_1 & 0 \\ \sigma_1 & \sigma_0 & 0 \\ 0 & 0 & \sigma \end{pmatrix}. \end{eqnarray} For zero magnetic field the current is along the electric field and we find the scalar conductivity \begin{eqnarray}\label{eq:sigma} &&\sigma=\frac{e^2}{3\pi^2 T}\int_m^\infty d\varepsilon \frac{p^3}{\varepsilon}\tau f^0(1-f^0)=\sigma_0+\sigma_2. \end{eqnarray} Thus, the components of the conductivity tensor are fully determined if the relaxation time $\tau$ is known. We evaluate the square of the scattering matrix using the standard QFT methods and then average over the positions of correlated ions, which effectively multiplies the transition probability by the structure function of ions. After some computations we find for the relaxation time \begin{eqnarray}\label{eq:relax_time2} \tau^{-1}(\varepsilon) &=&\frac{\pi Z^2e^4n_i}{\varepsilon p^3} \int_{-\infty}^{\varepsilon-m} d\omega e^{-\omega/2T} \frac{f^0(\varepsilon-\omega)}{f^0(\varepsilon)} \int_{q_-}^{q_+} dq(q^2-\omega^2+2\varepsilon\omega)S(q)F^2(q) \frac{1}{\sqrt{2\pi}\theta}\nonumber\\ &\times& e^{-\omega^2/2q^2\theta^2}e^{-q^2/8MT}\bigg\{ \frac{(2\varepsilon-\omega)^2 -q^2}{|q^2+\Pi_l|^2}+\theta^2 \frac{(q^2-\omega^2)[(2\varepsilon-\omega)^2 +q^2]-4m^2q^2}{q^2|q^2-\omega^2+\Pi_t|^2}\bigg\},\nonumber\\ \end{eqnarray} where $S(q)$ is the ionic structure function, $\theta \equiv \sqrt{{T}/{M}}$, $q_{\pm} = \vert\pm p+ \sqrt{p^2-(2\omega\epsilon- \omega^2)}\vert$ and $\varepsilon = \sqrt{p^2+m^2}$ for non-interacting electrons. The contribution of longitudinal and transverse photons in \eqref{eq:relax_time2} separate. The dynamical screening effects contained in the transverse contribution are parametrically suppressed by the factor $T/M$ at low temperatures and for heavy nuclei. This contribution is clearly important in the cases where electron-electron ($e$-$e$) scattering contributes to the collision intergral. This is the case, for example, when ions form a solid lattice and, therefore, Umklapp $e$-$e$ processes are allowed, or in the case of thermal conduction and shear stresses when the $e$-$e$ collisions contribute to the dissipation. Finally, we note that in order to account for the finite size of the nuclei we have multiplied the transition probability in Eq.~\eqref{eq:relax_time2} by the standard expression for the nuclear formfactor~\cite{1984ApJ...285..758I} \begin{eqnarray}\label{formfactor} F(q)=-3\frac{qr_c\cos(qr_c)-\sin(qr_c)}{(qr_c)^3}, \end{eqnarray} where $r_c$ is the charge radius of the nucleus given by $r_c=1.15\, A^{1/3}$ fm. \begin{figure}[t] \begin{center} \includegraphics[width=9cm,height=9cm]{structure.eps} \caption{ Dependence of the structure function of one-component plasma on the magnitude of momentum transfer $q$ in units of inverse $a_i$. For $\Gamma\ge 2$ the structure function is taken from Monte-Carlo calculations of Galam and Hansen \cite{1976PhRvA..14..816G}. For $\Gamma<2$ we obtain the structure function from the analytical expressions provided by Tamashiro et al.~\cite{1999PhyA..268...24T}. } \label{fig:Sq} \end{center} \end{figure} \section{Results} For the numerical computations we need to specify the ion structure function $S(q)$. We assume that only one sort of ions exists at a given density, so that the structure functions of one-component plasma (OCP) can be used. These has been extensively computed using various numerical methods. We adopt the Monte-Carlo results of Galam and Hansen~\cite{1976PhRvA..14..816G} for Coulomb OCP provided in tabular form and set a two-dimension spline function in the space spanned by the magnitude of the momentum transfer $q$ and the plasma parameter $\Gamma$. In the low-$\Gamma$ regime ($\Gamma\le 2$) we used the analytical (leading order) expressions derived by Tamashiro et al.~\cite{1999PhyA..268...24T} for Coulomb OCP derived using density functional methods. The resulting structure functions for various values of the plasma parameter $\Gamma$ are shown in Fig.~\ref{fig:Sq} as a function of the dimensionless parameter $a_iq$, where $a_i$ is the ion-radius as defined after Eq.~\eqref{eq:Gamma}. It is seen that the structure factor universally suppresses the contribution from small-$q$ scattering. The suppression sets in for larger $q$ at larger values of $\Gamma$. The large-$q$ asymptotics is independent of $\Gamma$ as $S(q)\to 1.$ The major difference arises for intermediate values of $q$ where the structure factor oscillates and the amplitude of oscillations increases with the value of $\Gamma$ parameter. The screening of longitudinal and transverse interactions is determined by the corresponding components of the polarization tensor. While expression \eqref{eq:relax_time2} is exact with respect to the form of the polarization tensor, in the numerical calculations we use the hard-thermal-loop approximation and next-to-leading expansion in $x = \omega/q$. For the real and imaginary parts of the polarization tensor we find \begin{eqnarray} \Pi_l (q,\omega) = q_D^2\chi_l, \qquad \Pi_t (q,\omega) = q_D^2\chi_t, \end{eqnarray} where $q_D$ is the Debye wave-length and the susceptibilities to order $O(x^2)$ are given by \begin{eqnarray} \label{eq:chi_l} &&{\rm Re}\chi_l (q,\omega) = 1-\frac{x^2}{\bar{v}^2}, \quad {\rm Im}\chi_l (q,\omega) =-\frac{\pi x}{2\bar{v}},\\ \label{eq:chi_t} &&{\rm Re}\chi_t (q,\omega) = x^2, \qquad {\rm Im}\chi_t (q,\omega) = \frac{\pi}{4}x\bar{v}, \end{eqnarray} where $\bar{v}$ is the electrons average velocity. Because the terms containing $\bar{v}$ are small as well as electrons are ultra-relativistic in the most of the regime of interest we approximate $\bar{v} = 1$ in our numerical calculations. For the longitudinal piece of the polarization tensor the screening is finite in the static case $x = 0 $, while it vanishes for the transverse piece as $\Pi_t (q,\omega)\propto x^2$, hence the purely dynamical nature of the transverse screening. \begin{figure}[t] \begin{center} \includegraphics[width=11cm,height=11cm]{sigma_b12_temp_Fe_C.eps} \caption{ Dependence of the electrical scalar conductivity $\sigma$ (upper panel) and tensor components $\sigma_0$ (middle panel) and $\sigma_1$ (lower panel) on temperature for three densities log$_{10}\rho = 10$ (solid lines), log$_{10}\rho = 8$ (dashed lines) and log$_{10}\rho = 6$ (dash-dotted lines). The left column contains results for $^{56}{\rm Fe}$, the right one for $^{12}{\rm C}$. The dotted lines (symbols in the two lower panels) associated with each line show the same, but are evaluate from the zero-temperature Drude formula. The magnetic field is fixed at $B_{12}=1.$} \label{fig:sigma} \end{center} \end{figure} \begin{figure}[t] \begin{center} \includegraphics[width=11cm,height=11cm]{sigma_b12_rho_Fe_C.eps} \caption{Dependence of the electrical scalar conductivity $\sigma$ (upper panel) and tensor components $\sigma_0$ (middle panel) and $\sigma_1$ (lower panel) on density for three temperatures $T= 10$ MeV (solid lines), $T= 1$ MeV (dashed lines) and $T= 0.1$ MeV (dash-dotted lines) in the case of $^{56}{\rm Fe}$ (left) and $C^{12}{\rm C}$ (right) nuclei and magnetic field $B_{12}=1$. } \label{fig:sigma2} \end{center} \end{figure} In the zero-temperature limit Eq.~\eqref{eq:sigma} simplifies via the substitution $T\partial f^0/\partial\varepsilon=- f^0(1-f^0) \to -T\delta(\varepsilon-\varepsilon_F)$, i.e., \begin{eqnarray}\label{sigma_fermi} \sigma=\frac{e^2}{3\pi^2}\int_m^\infty d\varepsilon \frac{p^3}{\varepsilon}\tau(\varepsilon)\delta (\varepsilon-\varepsilon_F)= \frac{n_ee^2\tau_F}{\varepsilon_F}, \end{eqnarray} where $\tau_F$ is the relaxation time (\ref{eq:relax_time2}) taken on the Fermi surface in the $T=0$ limit \begin{eqnarray}\label{eq:relax_fermi} \tau^{-1}_F\equiv\tau^{-1}(\varepsilon_F) =\frac{4}{3\pi}Ze^4 \varepsilon_F\int_{0}^{2p_F}dq\frac{q^3}{|q^2+\Pi_l|^2}\bigg(1-\frac{q^2}{4\varepsilon^2_F}\bigg)S(q)F^2(q), \end{eqnarray} where employed the charge neutrality condition $n_e=Zn_i$. Neglecting the screening ($\Pi_l\to 0$) and the nuclear formfactor [$F(q)\to 1$] we obtain from \eqref{eq:relax_fermi} \begin{eqnarray}\label{eq:relax_Pethick} \tau^{-1}_F =\frac{4Ze^4 \varepsilon_F}{3\pi}\int_{0}^{2p_F}\frac{dq}{q}\bigg(1-\frac{q^2}{4\varepsilon^2_F}\bigg)S(q), \end{eqnarray} which coincides with Eqs. (9) and (11) of Ref.~\cite{1984MNRAS.209..511N}. With the input described above we have evaluated the relaxation time for electron scattering off the ions using Eq.~\eqref{eq:relax_time2} and then the components of the conductivity tensor according to Eq.~\eqref{eq:sigma2}. Here we demonstrate selected results, while our complete results are discussed elsewhere~\cite{2016arXiv160507612H}. The conductivity as a function of temperature is shown in Fig.~\ref{fig:sigma} for carbon $\isotope[12]{C}$ and iron $\isotope[56]{Fe}$ nuclei. The magnitude of the magnetic field is fixed to $B_{12} = 1$, where $B_{12}$ is the magnetic field in units of $10^{12}$~G. The full results are compared to the case where the conductivity is evaluated from the Drude formula \eqref{sigma_fermi}, which is shown by dotted lines. The deviation from the zero temperature result are visible for temperatures in the range 0.1$-$1 MeV (MeV = $1.16\times 10^{10}$ K) when the density is varied from $10^6$ to $10^{10}$ g cm$^{-3}$. It is seen that the $\sigma$ component of {\it conductivity has a minimum as a function of temperature:} the low-temperature decrease is replace by a power-law increase with increasing temperature. This increase can be understood in terms of the smearing of the Fermi surface by temperature which makes more electrons available for conduction. The minimum of the conductivity is one of the key findings of our work. The same as in Fig.~\ref{fig:sigma} but as a function of density for fixed temperature values is shown in Fig.~\ref{fig:sigma2}. The scalar conductivity and $\sigma_0$ component are increasing functions of density and depend strongly on the temperature in the low-density limit, which is associated with lifting of the degeneracy as the temperature is increased. The behaviour of $\sigma_1$ is reversed: it is almost independent of temperature and has a maximum. This is the consequence of the different scaling of the components of the conductivity tensor with $\omega_c\tau$ parameter, which describes the effects of magnetic field. Our results in the cases of matter composed of $\isotope[12]{C}$ or matter composed of series of nuclei (when the composition varies with density) show the same general trends as for $\isotope[56]{Fe}$. The differences between these cases are quantitative and are discussed in detailed in Ref.~\cite{2016arXiv160507612H}. \section{Conclusions} In this contribution we gave an overview of our current work on the conductivity of dense matter in the envelopes of neutron stars at non-zero temperature. One ingredient of our effort is the formulation of the transport in a manner which allows us to include the dynamical screening exactly, provided that the polarization tensor of electrons (or equivalently the self-energies of QED photons) in plasma can be computed to desired accuracy. Here we employed the results based on the hard-thermal-loop approximation and low-frequency expansion appropriate at not very high temperatures. We have shown that for electron-ion scattering the dynamical screening is suppressed parametrically by a factor $M/T$, but we anticipate that its effect would be substantial in the cases (a) of high temperatures and presence of light clusters, (b) of low temperatures, where the Umklapp processes with $e$-$e$ scattering are important, (c) of transport processes where $e$-$e$ scattering may be dominant from the outset, such as the thermal conductivity and shear viscosity. Our numerical results show that the scalar conductivity (no anisotropy due to the $B$-field) has a minimum as a function of temperature, with a power-law decrease at low-temperatures and a power-law increase at higher temperatures. The range of validity of the zero-temperature Drude formula extends from low temperatures up to 0.1$-$1 MeV ($10^9-10^{10}$ K), where the lower of these bounds corresponds to density $\rho \sim 10^{6}$ g cm$^{-3}$ and the upper one to $\rho \sim 10^{10}$ g cm$^{-3}$. The behaviour of the off-diagonal $\sigma_1$ component of the conductivity tensor is similar to the one described above except at low densities, where it remains almost constant. Finally the $\sigma_0$ component shows strongly density dependent behaviour: for large densities (high degeneracy) it behaves analogous to $\sigma$, but shows the inverse trend for low-densities which is associated with the transition from the regime $\omega_c\tau <1 $ to $\omega_c\tau >1 $. \section*{Acknowledgments} We thank M. Alford, H. Nishimura, L. Rezzolla and D. Rischke for discussions. This work was supported by the HGS-HIRe graduate program at Frankfurt University (A. H.), by the Deutsche Forschungsgemeinschaft's Grant No. SE 1836/3-1 (A. S.) and by the NewCompStar COST Action MP1304. We thank the Volkswagen Stiftung for the support of the 2015 edition of the conference series {\it ``The Modern Physics of Compact Stars and Relativistic Gravity''}. \bibliographystyle{JHEP} \providecommand{\href}[2]{#2}\begingroup\raggedright \section*{Supplemental material} Below we present numerical tables for the conductivities log$_{10}\sigma$, log$_{10}\sigma_0$ and log$_{10}\sigma_1$ (in units of s$^{-1}$) for various values of magnetic field (in units of $10^{12}$ G) for sets of values of density [g cm$^{-3}$] and temperature [MeV]. The tables are provided for three types of composition of matter: $\isotope[12]{C}$ nuclei, $\isotope[56]{Fe}$ nuclei, and density-dependent composition as indicated in Fig.~2. Analytical fits to these results with relative error $\le 10\%$ can be found in Ref.~\cite{2016arXiv160507612H}. \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma$ for $\isotope[12]{C}$} \label{tab:sigma_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 20.745 & 20.853 & 20.973 & 21.100 & 21.222 & 21.356 & 21.498 & 21.646 & 21.798 & 21.957 & 22.123 \\ 6.2 & 20.816 & 20.905 & 21.013 & 21.131 & 21.251 & 21.378 & 21.518 & 21.663 & 21.814 & 21.970 & 22.136 \\ 6.4 & 20.898 & 20.965 & 21.057 & 21.165 & 21.281 & 21.401 & 21.538 & 21.681 & 21.830 & 21.985 & 22.149 \\ 6.6 & 20.990 & 21.033 & 21.106 & 21.202 & 21.313 & 21.426 & 21.559 & 21.699 & 21.846 & 22.000 & 22.162 \\ 6.8 & 21.089 & 21.110 & 21.163 & 21.244 & 21.346 & 21.452 & 21.581 & 21.719 & 21.864 & 22.015 & 22.176 \\ 7.0 & 21.192 & 21.193 & 21.226 & 21.290 & 21.381 & 21.482 & 21.605 & 21.739 & 21.882 & 22.031 & 22.190 \\ 7.2 & 21.296 & 21.283 & 21.296 & 21.342 & 21.420 & 21.518 & 21.630 & 21.760 & 21.900 & 22.048 & 22.205 \\ 7.4 & 21.398 & 21.375 & 21.372 & 21.400 & 21.462 & 21.553 & 21.656 & 21.783 & 21.920 & 22.065 & 22.220 \\ 7.6 & 21.498 & 21.467 & 21.452 & 21.463 & 21.510 & 21.589 & 21.684 & 21.806 & 21.940 & 22.083 & 22.236 \\ 7.8 & 21.595 & 21.560 & 21.535 & 21.532 & 21.562 & 21.629 & 21.719 & 21.831 & 21.961 & 22.102 & 22.253 \\ 8.0 & 21.688 & 21.650 & 21.620 & 21.606 & 21.620 & 21.672 & 21.756 & 21.858 & 21.984 & 22.121 & 22.270 \\ 8.2 & 21.779 & 21.740 & 21.705 & 21.682 & 21.683 & 21.720 & 21.794 & 21.886 & 22.008 & 22.142 & 22.288 \\ 8.4 & 21.867 & 21.827 & 21.789 & 21.761 & 21.751 & 21.773 & 21.833 & 21.920 & 22.033 & 22.163 & 22.306 \\ 8.6 & 21.979 & 21.913 & 21.874 & 21.841 & 21.822 & 21.831 & 21.877 & 21.958 & 22.059 & 22.186 & 22.326 \\ 8.8 & 22.065 & 21.998 & 21.958 & 21.921 & 21.896 & 21.893 & 21.925 & 21.996 & 22.088 & 22.209 & 22.346 \\ 9.0 & 22.151 & 22.082 & 22.041 & 22.002 & 21.972 & 21.959 & 21.978 & 22.036 & 22.121 & 22.234 & 22.368 \\ 9.2 & 22.237 & 22.191 & 22.124 & 22.084 & 22.050 & 22.029 & 22.035 & 22.080 & 22.159 & 22.261 & 22.390 \\ 9.4 & 22.323 & 22.274 & 22.207 & 22.166 & 22.129 & 22.102 & 22.097 & 22.128 & 22.198 & 22.290 & 22.414 \\ 9.6 & 22.410 & 22.358 & 22.289 & 22.247 & 22.208 & 22.177 & 22.163 & 22.181 & 22.238 & 22.323 & 22.439 \\ 9.8 & 22.496 & 22.443 & 22.396 & 22.329 & 22.289 & 22.254 & 22.233 & 22.238 & 22.282 & 22.362 & 22.466 \\ 10.0 & 22.581 & 22.529 & 22.479 & 22.411 & 22.369 & 22.332 & 22.305 & 22.300 & 22.330 & 22.401 & 22.495 \\ 10.2 & 22.665 & 22.615 & 22.563 & 22.492 & 22.451 & 22.411 & 22.380 & 22.366 & 22.383 & 22.441 & 22.528 \\ 10.4 & 22.747 & 22.701 & 22.648 & 22.600 & 22.532 & 22.492 & 22.457 & 22.435 & 22.441 & 22.486 & 22.568 \\ 10.6 & 22.830 & 22.788 & 22.734 & 22.683 & 22.614 & 22.573 & 22.535 & 22.508 & 22.503 & 22.534 & 22.608 \\ 10.8 & 22.915 & 22.873 & 22.821 & 22.768 & 22.697 & 22.654 & 22.614 & 22.583 & 22.569 & 22.588 & 22.649 \\ 11.0 & 23.002 & 22.959 & 22.910 & 22.854 & 22.805 & 22.737 & 22.695 & 22.660 & 22.639 & 22.646 & 22.694 \\ \hline \end{tabular} } \caption {$\log\sigma_0$ for $\isotope[12]{C}$ at $B_{12}=1$} \label{tab:sigma0_b12_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 16.648 & 16.551 & 16.423 & 16.279 & 16.132 & 15.987 & 15.843 & 15.696 & 15.546 & 15.390 & 15.227 \\ 6.2 & 16.947 & 16.878 & 16.771 & 16.641 & 16.503 & 16.363 & 16.222 & 16.078 & 15.929 & 15.776 & 15.613 \\ 6.4 & 17.238 & 17.195 & 17.112 & 16.999 & 16.871 & 16.736 & 16.599 & 16.458 & 16.312 & 16.160 & 15.999 \\ 6.6 & 17.522 & 17.504 & 17.445 & 17.352 & 17.236 & 17.108 & 16.976 & 16.838 & 16.694 & 16.544 & 16.385 \\ 6.8 & 17.805 & 17.806 & 17.771 & 17.699 & 17.597 & 17.478 & 17.351 & 17.216 & 17.075 & 16.928 & 16.770 \\ 7.0 & 18.090 & 18.104 & 18.091 & 18.040 & 17.955 & 17.845 & 17.724 & 17.594 & 17.455 & 17.310 & 17.155 \\ 7.2 & 18.378 & 18.402 & 18.405 & 18.375 & 18.307 & 18.211 & 18.096 & 17.970 & 17.835 & 17.692 & 17.539 \\ 7.4 & 18.672 & 18.702 & 18.716 & 18.704 & 18.655 & 18.572 & 18.465 & 18.345 & 18.213 & 18.073 & 17.923 \\ 7.6 & 18.969 & 19.003 & 19.027 & 19.028 & 18.997 & 18.929 & 18.832 & 18.718 & 18.590 & 18.453 & 18.306 \\ 7.8 & 19.271 & 19.307 & 19.336 & 19.349 & 19.333 & 19.281 & 19.196 & 19.089 & 18.966 & 18.832 & 18.688 \\ 8.0 & 19.575 & 19.613 & 19.646 & 19.667 & 19.664 & 19.628 & 19.556 & 19.457 & 19.340 & 19.210 & 19.069 \\ 8.2 & 19.882 & 19.921 & 19.956 & 19.982 & 19.990 & 19.968 & 19.910 & 19.821 & 19.711 & 19.586 & 19.448 \\ 8.4 & 20.190 & 20.229 & 20.265 & 20.295 & 20.310 & 20.300 & 20.257 & 20.179 & 20.078 & 19.959 & 19.826 \\ 8.6 & 20.473 & 20.535 & 20.571 & 20.602 & 20.622 & 20.622 & 20.592 & 20.530 & 20.438 & 20.328 & 20.201 \\ 8.8 & 20.778 & 20.837 & 20.872 & 20.903 & 20.925 & 20.931 & 20.913 & 20.865 & 20.787 & 20.688 & 20.570 \\ 9.0 & 21.077 & 21.133 & 21.164 & 21.191 & 21.213 & 21.222 & 21.213 & 21.180 & 21.119 & 21.035 & 20.929 \\ 9.2 & 21.367 & 21.399 & 21.441 & 21.464 & 21.481 & 21.489 & 21.486 & 21.466 & 21.425 & 21.360 & 21.272 \\ 9.4 & 21.642 & 21.669 & 21.699 & 21.712 & 21.722 & 21.725 & 21.723 & 21.713 & 21.692 & 21.652 & 21.589 \\ 9.6 & 21.897 & 21.915 & 21.929 & 21.932 & 21.931 & 21.927 & 21.922 & 21.918 & 21.915 & 21.901 & 21.869 \\ 9.8 & 22.128 & 22.132 & 22.131 & 22.120 & 22.108 & 22.095 & 22.084 & 22.081 & 22.090 & 22.101 & 22.101 \\ 10.0 & 22.329 & 22.320 & 22.305 & 22.278 & 22.257 & 22.235 & 22.217 & 22.210 & 22.223 & 22.252 & 22.280 \\ 10.2 & 22.500 & 22.480 & 22.453 & 22.410 & 22.382 & 22.353 & 22.328 & 22.316 & 22.325 & 22.362 & 22.410 \\ 10.4 & 22.644 & 22.616 & 22.580 & 22.545 & 22.491 & 22.457 & 22.427 & 22.408 & 22.410 & 22.445 & 22.506 \\ 10.6 & 22.767 & 22.735 & 22.693 & 22.650 & 22.590 & 22.552 & 22.518 & 22.492 & 22.487 & 22.514 & 22.577 \\ 10.8 & 22.876 & 22.842 & 22.796 & 22.748 & 22.682 & 22.642 & 22.605 & 22.574 & 22.561 & 22.577 & 22.634 \\ 11.0 & 22.979 & 22.940 & 22.894 & 22.842 & 22.796 & 22.730 & 22.690 & 22.655 & 22.635 & 22.641 & 22.687 \\ \hline \end{tabular} } \caption {$\log\sigma_0$ for $\isotope[12]{C}$ at $B_{12}=10$} \label{tab:sigma0_b13_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 14.648 & 14.551 & 14.423 & 14.280 & 14.133 & 13.988 & 13.842 & 13.696 & 13.546 & 13.390 & 13.227 \\ 6.2 & 14.948 & 14.878 & 14.771 & 14.642 & 14.503 & 14.363 & 14.221 & 14.078 & 13.929 & 13.776 & 13.613 \\ 6.4 & 15.238 & 15.195 & 15.112 & 14.999 & 14.871 & 14.736 & 14.599 & 14.458 & 14.312 & 14.160 & 13.999 \\ 6.6 & 15.523 & 15.504 & 15.446 & 15.352 & 15.236 & 15.108 & 14.975 & 14.838 & 14.694 & 14.544 & 14.385 \\ 6.8 & 15.806 & 15.806 & 15.772 & 15.700 & 15.598 & 15.478 & 15.351 & 15.216 & 15.075 & 14.928 & 14.770 \\ 7.0 & 16.090 & 16.105 & 16.092 & 16.041 & 15.955 & 15.846 & 15.724 & 15.594 & 15.455 & 15.310 & 15.155 \\ 7.2 & 16.379 & 16.403 & 16.406 & 16.376 & 16.308 & 16.212 & 16.096 & 15.970 & 15.835 & 15.692 & 15.539 \\ 7.4 & 16.672 & 16.703 & 16.718 & 16.705 & 16.656 & 16.573 & 16.466 & 16.345 & 16.213 & 16.073 & 15.923 \\ 7.6 & 16.970 & 17.005 & 17.028 & 17.031 & 16.999 & 16.931 & 16.833 & 16.719 & 16.591 & 16.454 & 16.306 \\ 7.8 & 17.273 & 17.310 & 17.339 & 17.353 & 17.337 & 17.285 & 17.199 & 17.090 & 16.967 & 16.833 & 16.688 \\ 8.0 & 17.579 & 17.617 & 17.651 & 17.673 & 17.671 & 17.634 & 17.561 & 17.460 & 17.342 & 17.212 & 17.069 \\ 8.2 & 17.888 & 17.928 & 17.964 & 17.992 & 18.000 & 17.978 & 17.920 & 17.828 & 17.716 & 17.589 & 17.450 \\ 8.4 & 18.199 & 18.240 & 18.278 & 18.310 & 18.327 & 18.318 & 18.274 & 18.193 & 18.088 & 17.965 & 17.830 \\ 8.6 & 18.487 & 18.553 & 18.593 & 18.628 & 18.652 & 18.654 & 18.623 & 18.556 & 18.457 & 18.340 & 18.208 \\ 8.8 & 18.801 & 18.868 & 18.909 & 18.946 & 18.975 & 18.986 & 18.969 & 18.914 & 18.824 & 18.713 & 18.586 \\ 9.0 & 19.115 & 19.184 & 19.225 & 19.264 & 19.297 & 19.315 & 19.309 & 19.268 & 19.189 & 19.085 & 18.962 \\ 9.2 & 19.429 & 19.475 & 19.542 & 19.582 & 19.617 & 19.642 & 19.645 & 19.617 & 19.551 & 19.454 & 19.337 \\ 9.4 & 19.742 & 19.791 & 19.858 & 19.899 & 19.936 & 19.965 & 19.977 & 19.961 & 19.908 & 19.820 & 19.709 \\ 9.6 & 20.055 & 20.106 & 20.175 & 20.215 & 20.254 & 20.287 & 20.305 & 20.300 & 20.260 & 20.182 & 20.079 \\ 9.8 & 20.368 & 20.419 & 20.465 & 20.531 & 20.570 & 20.605 & 20.628 & 20.632 & 20.604 & 20.540 & 20.444 \\ 10.0 & 20.680 & 20.731 & 20.779 & 20.844 & 20.883 & 20.918 & 20.945 & 20.956 & 20.940 & 20.889 & 20.804 \\ 10.2 & 20.993 & 21.040 & 21.089 & 21.154 & 21.191 & 21.226 & 21.255 & 21.270 & 21.263 & 21.226 & 21.153 \\ 10.4 & 21.304 & 21.346 & 21.394 & 21.436 & 21.493 & 21.526 & 21.553 & 21.570 & 21.570 & 21.545 & 21.488 \\ 10.6 & 21.610 & 21.647 & 21.691 & 21.732 & 21.783 & 21.811 & 21.835 & 21.851 & 21.855 & 21.839 & 21.799 \\ 10.8 & 21.907 & 21.940 & 21.978 & 22.015 & 22.057 & 22.078 & 22.095 & 22.107 & 22.110 & 22.101 & 22.077 \\ 11.0 & 22.192 & 22.221 & 22.250 & 22.278 & 22.299 & 22.320 & 22.328 & 22.331 & 22.330 & 22.325 & 22.315 \\ \hline \end{tabular} } \end{center} \end{table} \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma_0$ for $\isotope[12]{C}$ at $B_{12}=100$} \label{tab:sigma0_b14_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 12.648 & 12.551 & 12.423 & 12.280 & 12.133 & 11.988 & 11.843 & 11.696 & 11.546 & 11.390 & 11.227 \\ 6.2 & 12.948 & 12.878 & 12.771 & 12.642 & 12.503 & 12.363 & 12.222 & 12.078 & 11.929 & 11.776 & 11.613 \\ 6.4 & 13.238 & 13.195 & 13.112 & 12.999 & 12.871 & 12.736 & 12.599 & 12.458 & 12.312 & 12.160 & 11.999 \\ 6.6 & 13.523 & 13.504 & 13.446 & 13.352 & 13.236 & 13.108 & 12.976 & 12.838 & 12.694 & 12.544 & 12.385 \\ 6.8 & 13.806 & 13.806 & 13.772 & 13.700 & 13.598 & 13.478 & 13.351 & 13.216 & 13.075 & 12.928 & 12.770 \\ 7.0 & 14.090 & 14.105 & 14.092 & 14.041 & 13.955 & 13.846 & 13.724 & 13.594 & 13.455 & 13.310 & 13.155 \\ 7.2 & 14.379 & 14.403 & 14.406 & 14.376 & 14.308 & 14.212 & 14.096 & 13.970 & 13.835 & 13.692 & 13.539 \\ 7.4 & 14.672 & 14.703 & 14.718 & 14.705 & 14.656 & 14.573 & 14.466 & 14.345 & 14.213 & 14.073 & 13.923 \\ 7.6 & 14.970 & 15.005 & 15.028 & 15.031 & 14.999 & 14.931 & 14.833 & 14.719 & 14.591 & 14.454 & 14.306 \\ 7.8 & 15.273 & 15.310 & 15.339 & 15.353 & 15.337 & 15.285 & 15.199 & 15.090 & 14.967 & 14.833 & 14.688 \\ 8.0 & 15.579 & 15.618 & 15.651 & 15.673 & 15.671 & 15.634 & 15.561 & 15.460 & 15.342 & 15.212 & 15.069 \\ 8.2 & 15.888 & 15.928 & 15.964 & 15.992 & 16.000 & 15.979 & 15.920 & 15.828 & 15.716 & 15.589 & 15.450 \\ 8.4 & 16.200 & 16.240 & 16.278 & 16.310 & 16.327 & 16.319 & 16.274 & 16.193 & 16.088 & 15.965 & 15.830 \\ 8.6 & 16.487 & 16.553 & 16.593 & 16.628 & 16.652 & 16.655 & 16.624 & 16.556 & 16.457 & 16.340 & 16.208 \\ 8.8 & 16.801 & 16.869 & 16.909 & 16.947 & 16.976 & 16.987 & 16.969 & 16.914 & 16.825 & 16.714 & 16.586 \\ 9.0 & 17.115 & 17.185 & 17.226 & 17.265 & 17.298 & 17.316 & 17.310 & 17.269 & 17.190 & 17.085 & 16.962 \\ 9.2 & 17.429 & 17.476 & 17.543 & 17.583 & 17.619 & 17.644 & 17.647 & 17.619 & 17.553 & 17.455 & 17.337 \\ 9.4 & 17.743 & 17.792 & 17.860 & 17.901 & 17.939 & 17.969 & 17.981 & 17.965 & 17.911 & 17.823 & 17.711 \\ 9.6 & 18.057 & 18.108 & 18.178 & 18.219 & 18.259 & 18.292 & 18.312 & 18.306 & 18.266 & 18.188 & 18.082 \\ 9.8 & 18.371 & 18.423 & 18.470 & 18.538 & 18.578 & 18.614 & 18.640 & 18.644 & 18.616 & 18.550 & 18.452 \\ 10.0 & 18.686 & 18.738 & 18.788 & 18.856 & 18.897 & 18.935 & 18.965 & 18.978 & 18.962 & 18.909 & 18.819 \\ 10.2 & 19.002 & 19.052 & 19.104 & 19.174 & 19.216 & 19.256 & 19.289 & 19.308 & 19.303 & 19.263 & 19.183 \\ 10.4 & 19.319 & 19.365 & 19.419 & 19.466 & 19.534 & 19.575 & 19.611 & 19.636 & 19.641 & 19.612 & 19.545 \\ 10.6 & 19.637 & 19.679 & 19.732 & 19.783 & 19.852 & 19.893 & 19.932 & 19.961 & 19.974 & 19.957 & 19.903 \\ 10.8 & 19.952 & 19.993 & 20.045 & 20.098 & 20.169 & 20.211 & 20.251 & 20.284 & 20.303 & 20.297 & 20.255 \\ 11.0 & 20.264 & 20.307 & 20.356 & 20.411 & 20.459 & 20.527 & 20.568 & 20.604 & 20.629 & 20.632 & 20.602 \\ \hline \end{tabular} } \caption {$\log\sigma_1$ for $\isotope[12]{C}$ at $B_{12}=1$} \label{tab:sigma1_b12_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 18.633 & 18.633 & 18.633 & 18.633 & 18.629 & 18.629 & 18.629 & 18.629 & 18.629 & 18.629 & 18.629 \\ 6.2 & 18.833 & 18.833 & 18.833 & 18.833 & 18.831 & 18.829 & 18.829 & 18.829 & 18.829 & 18.829 & 18.829 \\ 6.4 & 19.033 & 19.033 & 19.033 & 19.033 & 19.032 & 19.029 & 19.029 & 19.029 & 19.029 & 19.029 & 19.029 \\ 6.6 & 19.233 & 19.233 & 19.233 & 19.233 & 19.233 & 19.229 & 19.229 & 19.229 & 19.229 & 19.229 & 19.229 \\ 6.8 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.429 & 19.429 & 19.429 & 19.429 & 19.429 & 19.429 \\ 7.0 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.629 & 19.629 & 19.629 & 19.629 & 19.629 & 19.629 \\ 7.2 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.831 & 19.829 & 19.829 & 19.829 & 19.829 & 19.829 \\ 7.4 & 20.032 & 20.032 & 20.032 & 20.032 & 20.032 & 20.032 & 20.028 & 20.029 & 20.029 & 20.029 & 20.029 \\ 7.6 & 20.232 & 20.232 & 20.232 & 20.232 & 20.232 & 20.232 & 20.228 & 20.228 & 20.229 & 20.229 & 20.229 \\ 7.8 & 20.431 & 20.431 & 20.431 & 20.430 & 20.430 & 20.431 & 20.429 & 20.428 & 20.428 & 20.429 & 20.429 \\ 8.0 & 20.630 & 20.629 & 20.629 & 20.628 & 20.628 & 20.629 & 20.629 & 20.626 & 20.627 & 20.628 & 20.628 \\ 8.2 & 20.828 & 20.827 & 20.825 & 20.824 & 20.824 & 20.824 & 20.826 & 20.824 & 20.826 & 20.827 & 20.828 \\ 8.4 & 21.024 & 21.022 & 21.020 & 21.018 & 21.017 & 21.017 & 21.019 & 21.020 & 21.022 & 21.025 & 21.027 \\ 8.6 & 21.220 & 21.215 & 21.211 & 21.207 & 21.204 & 21.203 & 21.206 & 21.212 & 21.215 & 21.220 & 21.224 \\ 8.8 & 21.410 & 21.402 & 21.396 & 21.389 & 21.383 & 21.380 & 21.383 & 21.393 & 21.400 & 21.411 & 21.418 \\ 9.0 & 21.595 & 21.582 & 21.571 & 21.560 & 21.549 & 21.542 & 21.544 & 21.557 & 21.573 & 21.592 & 21.606 \\ 9.2 & 21.771 & 21.757 & 21.732 & 21.714 & 21.696 & 21.682 & 21.680 & 21.697 & 21.726 & 21.755 & 21.782 \\ 9.4 & 21.932 & 21.910 & 21.872 & 21.845 & 21.817 & 21.793 & 21.785 & 21.802 & 21.844 & 21.890 & 21.938 \\ 9.6 & 22.074 & 22.040 & 21.984 & 21.946 & 21.906 & 21.871 & 21.853 & 21.867 & 21.919 & 21.987 & 22.060 \\ 9.8 & 22.190 & 22.142 & 22.094 & 22.016 & 21.964 & 21.916 & 21.886 & 21.892 & 21.947 & 22.039 & 22.137 \\ 10.0 & 22.277 & 22.215 & 22.151 & 22.055 & 21.993 & 21.934 & 21.891 & 21.884 & 21.934 & 22.038 & 22.162 \\ 10.2 & 22.332 & 22.261 & 22.182 & 22.069 & 21.999 & 21.931 & 21.877 & 21.855 & 21.891 & 21.995 & 22.137 \\ 10.4 & 22.358 & 22.284 & 22.194 & 22.112 & 21.990 & 21.916 & 21.851 & 21.815 & 21.832 & 21.924 & 22.078 \\ 10.6 & 22.364 & 22.289 & 22.193 & 22.100 & 21.971 & 21.892 & 21.820 & 21.771 & 21.768 & 21.840 & 21.990 \\ 10.8 & 22.358 & 22.281 & 22.184 & 22.082 & 21.945 & 21.863 & 21.786 & 21.726 & 21.705 & 21.753 & 21.887 \\ 11.0 & 22.348 & 22.265 & 22.171 & 22.063 & 21.968 & 21.833 & 21.752 & 21.684 & 21.647 & 21.670 & 21.781 \\ \hline \end{tabular} } \caption {$\log\sigma_1$ for $\isotope[12]{C}$ at $B_{12}=10$} \label{tab:sigma1_b13_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 17.633 & 17.633 & 17.633 & 17.633 & 17.629 & 17.629 & 17.629 & 17.629 & 17.629 & 17.629 & 17.629 \\ 6.2 & 17.833 & 17.833 & 17.833 & 17.833 & 17.831 & 17.829 & 17.829 & 17.829 & 17.829 & 17.829 & 17.829 \\ 6.4 & 18.033 & 18.033 & 18.033 & 18.033 & 18.032 & 18.029 & 18.029 & 18.029 & 18.029 & 18.029 & 18.029 \\ 6.6 & 18.233 & 18.233 & 18.233 & 18.233 & 18.233 & 18.229 & 18.229 & 18.229 & 18.229 & 18.229 & 18.229 \\ 6.8 & 18.433 & 18.433 & 18.433 & 18.433 & 18.433 & 18.429 & 18.429 & 18.429 & 18.429 & 18.429 & 18.429 \\ 7.0 & 18.633 & 18.633 & 18.633 & 18.633 & 18.633 & 18.629 & 18.629 & 18.629 & 18.629 & 18.629 & 18.629 \\ 7.2 & 18.833 & 18.833 & 18.833 & 18.833 & 18.833 & 18.832 & 18.829 & 18.829 & 18.829 & 18.829 & 18.829 \\ 7.4 & 19.033 & 19.033 & 19.033 & 19.033 & 19.033 & 19.033 & 19.029 & 19.029 & 19.029 & 19.029 & 19.029 \\ 7.6 & 19.233 & 19.233 & 19.233 & 19.233 & 19.233 & 19.233 & 19.229 & 19.229 & 19.229 & 19.229 & 19.229 \\ 7.8 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.431 & 19.429 & 19.429 & 19.429 & 19.429 \\ 8.0 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.629 & 19.629 & 19.629 & 19.629 \\ 8.2 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.829 & 19.829 & 19.829 & 19.829 \\ 8.4 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.030 & 20.029 & 20.029 & 20.029 \\ 8.6 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.232 & 20.229 & 20.229 & 20.229 \\ 8.8 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.428 & 20.429 & 20.429 \\ 9.0 & 20.633 & 20.633 & 20.633 & 20.632 & 20.632 & 20.632 & 20.632 & 20.632 & 20.629 & 20.628 & 20.629 \\ 9.2 & 20.833 & 20.833 & 20.832 & 20.832 & 20.832 & 20.832 & 20.831 & 20.832 & 20.831 & 20.828 & 20.828 \\ 9.4 & 21.032 & 21.032 & 21.031 & 21.031 & 21.031 & 21.030 & 21.030 & 21.030 & 21.030 & 21.027 & 21.028 \\ 9.6 & 21.231 & 21.231 & 21.230 & 21.229 & 21.228 & 21.228 & 21.227 & 21.227 & 21.228 & 21.226 & 21.226 \\ 9.8 & 21.430 & 21.429 & 21.428 & 21.426 & 21.425 & 21.423 & 21.422 & 21.422 & 21.423 & 21.424 & 21.423 \\ 10.0 & 21.628 & 21.626 & 21.625 & 21.621 & 21.619 & 21.616 & 21.614 & 21.612 & 21.613 & 21.617 & 21.617 \\ 10.2 & 21.824 & 21.822 & 21.819 & 21.813 & 21.809 & 21.804 & 21.799 & 21.796 & 21.796 & 21.801 & 21.806 \\ 10.4 & 22.017 & 22.014 & 22.008 & 22.003 & 21.992 & 21.984 & 21.975 & 21.968 & 21.966 & 21.973 & 21.986 \\ 10.6 & 22.206 & 22.201 & 22.192 & 22.182 & 22.164 & 22.151 & 22.136 & 22.124 & 22.118 & 22.125 & 22.145 \\ 10.8 & 22.388 & 22.380 & 22.366 & 22.349 & 22.320 & 22.299 & 22.276 & 22.255 & 22.243 & 22.248 & 22.276 \\ 11.0 & 22.560 & 22.546 & 22.526 & 22.499 & 22.471 & 22.423 & 22.390 & 22.358 & 22.335 & 22.336 & 22.370 \\ \hline \end{tabular} } \end{center} \end{table} \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma_1$ for $\isotope[12]{C}$ at $B_{12}=100$} \label{tab:sigma1_b14_C} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 16.633 & 16.633 & 16.633 & 16.633 & 16.629 & 16.629 & 16.629 & 16.629 & 16.629 & 16.629 & 16.629 \\ 6.2 & 16.833 & 16.833 & 16.833 & 16.833 & 16.831 & 16.829 & 16.829 & 16.829 & 16.829 & 16.829 & 16.829 \\ 6.4 & 17.033 & 17.033 & 17.033 & 17.033 & 17.032 & 17.029 & 17.029 & 17.029 & 17.029 & 17.029 & 17.029 \\ 6.6 & 17.233 & 17.233 & 17.233 & 17.233 & 17.233 & 17.229 & 17.229 & 17.229 & 17.229 & 17.229 & 17.229 \\ 6.8 & 17.433 & 17.433 & 17.433 & 17.433 & 17.433 & 17.429 & 17.429 & 17.429 & 17.429 & 17.429 & 17.429 \\ 7.0 & 17.633 & 17.633 & 17.633 & 17.633 & 17.633 & 17.629 & 17.629 & 17.629 & 17.629 & 17.629 & 17.629 \\ 7.2 & 17.833 & 17.833 & 17.833 & 17.833 & 17.833 & 17.832 & 17.829 & 17.829 & 17.829 & 17.829 & 17.829 \\ 7.4 & 18.033 & 18.033 & 18.033 & 18.033 & 18.033 & 18.033 & 18.029 & 18.029 & 18.029 & 18.029 & 18.029 \\ 7.6 & 18.233 & 18.233 & 18.233 & 18.233 & 18.233 & 18.233 & 18.229 & 18.229 & 18.229 & 18.229 & 18.229 \\ 7.8 & 18.433 & 18.433 & 18.433 & 18.433 & 18.433 & 18.433 & 18.431 & 18.429 & 18.429 & 18.429 & 18.429 \\ 8.0 & 18.633 & 18.633 & 18.633 & 18.633 & 18.633 & 18.633 & 18.633 & 18.629 & 18.629 & 18.629 & 18.629 \\ 8.2 & 18.833 & 18.833 & 18.833 & 18.833 & 18.833 & 18.833 & 18.833 & 18.829 & 18.829 & 18.829 & 18.829 \\ 8.4 & 19.033 & 19.033 & 19.033 & 19.033 & 19.033 & 19.033 & 19.033 & 19.030 & 19.029 & 19.029 & 19.029 \\ 8.6 & 19.233 & 19.233 & 19.233 & 19.233 & 19.233 & 19.233 & 19.233 & 19.232 & 19.229 & 19.229 & 19.229 \\ 8.8 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.433 & 19.429 & 19.429 & 19.429 \\ 9.0 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.633 & 19.630 & 19.629 & 19.629 \\ 9.2 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.833 & 19.832 & 19.829 & 19.829 \\ 9.4 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.033 & 20.029 & 20.029 \\ 9.6 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.233 & 20.230 & 20.229 \\ 9.8 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.433 & 20.432 & 20.429 \\ 10.0 & 20.633 & 20.633 & 20.633 & 20.633 & 20.633 & 20.633 & 20.633 & 20.633 & 20.633 & 20.633 & 20.629 \\ 10.2 & 20.833 & 20.833 & 20.833 & 20.833 & 20.833 & 20.833 & 20.833 & 20.833 & 20.833 & 20.833 & 20.829 \\ 10.4 & 21.033 & 21.033 & 21.033 & 21.033 & 21.033 & 21.033 & 21.033 & 21.033 & 21.033 & 21.033 & 21.032 \\ 10.6 & 21.233 & 21.233 & 21.233 & 21.233 & 21.233 & 21.232 & 21.232 & 21.232 & 21.232 & 21.232 & 21.232 \\ 10.8 & 21.433 & 21.433 & 21.433 & 21.432 & 21.432 & 21.432 & 21.431 & 21.431 & 21.431 & 21.431 & 21.431 \\ 11.0 & 21.633 & 21.632 & 21.632 & 21.632 & 21.631 & 21.631 & 21.630 & 21.630 & 21.629 & 21.629 & 21.629 \\ \hline \end{tabular} } \caption {$\log\sigma$ for $\isotope[56]{Fe}$} \label{tab:sigma_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 20.144 & 20.260 & 20.375 & 20.506 & 20.628 & 20.762 & 20.902 & 21.048 & 21.200 & 21.357 & 21.522 \\ 6.2 & 20.213 & 20.310 & 20.413 & 20.537 & 20.658 & 20.785 & 20.923 & 21.066 & 21.216 & 21.372 & 21.536 \\ 6.4 & 20.293 & 20.368 & 20.456 & 20.570 & 20.689 & 20.809 & 20.944 & 21.085 & 21.233 & 21.388 & 21.550 \\ 6.6 & 20.382 & 20.435 & 20.517 & 20.607 & 20.720 & 20.834 & 20.966 & 21.104 & 21.250 & 21.403 & 21.565 \\ 6.8 & 20.477 & 20.510 & 20.572 & 20.647 & 20.753 & 20.861 & 20.989 & 21.125 & 21.268 & 21.420 & 21.580 \\ 7.0 & 20.576 & 20.592 & 20.634 & 20.692 & 20.788 & 20.889 & 21.013 & 21.146 & 21.287 & 21.437 & 21.595 \\ 7.2 & 20.675 & 20.680 & 20.702 & 20.756 & 20.826 & 20.925 & 21.038 & 21.168 & 21.307 & 21.454 & 21.611 \\ 7.4 & 20.772 & 20.769 & 20.777 & 20.813 & 20.867 & 20.961 & 21.065 & 21.191 & 21.327 & 21.473 & 21.628 \\ 7.6 & 20.865 & 20.858 & 20.856 & 20.875 & 20.913 & 20.997 & 21.094 & 21.215 & 21.348 & 21.492 & 21.645 \\ 7.8 & 20.954 & 20.944 & 20.937 & 20.943 & 20.980 & 21.036 & 21.127 & 21.241 & 21.370 & 21.511 & 21.663 \\ 8.0 & 21.040 & 21.028 & 21.019 & 21.015 & 21.036 & 21.079 & 21.165 & 21.268 & 21.394 & 21.532 & 21.682 \\ 8.2 & 21.123 & 21.110 & 21.099 & 21.090 & 21.098 & 21.125 & 21.202 & 21.297 & 21.418 & 21.553 & 21.701 \\ 8.4 & 21.203 & 21.189 & 21.177 & 21.166 & 21.165 & 21.192 & 21.242 & 21.329 & 21.444 & 21.576 & 21.721 \\ 8.6 & 21.280 & 21.267 & 21.254 & 21.242 & 21.234 & 21.249 & 21.285 & 21.368 & 21.471 & 21.599 & 21.742 \\ 8.8 & 21.356 & 21.343 & 21.329 & 21.318 & 21.306 & 21.310 & 21.331 & 21.406 & 21.500 & 21.624 & 21.764 \\ 9.0 & 21.429 & 21.418 & 21.404 & 21.392 & 21.379 & 21.375 & 21.399 & 21.446 & 21.532 & 21.650 & 21.787 \\ 9.2 & 21.502 & 21.492 & 21.479 & 21.465 & 21.453 & 21.443 & 21.455 & 21.489 & 21.572 & 21.678 & 21.811 \\ 9.4 & 21.574 & 21.565 & 21.552 & 21.538 & 21.526 & 21.514 & 21.516 & 21.536 & 21.610 & 21.707 & 21.837 \\ 9.6 & 21.645 & 21.637 & 21.625 & 21.611 & 21.599 & 21.586 & 21.580 & 21.603 & 21.651 & 21.739 & 21.864 \\ 9.8 & 21.716 & 21.708 & 21.698 & 21.684 & 21.671 & 21.658 & 21.648 & 21.660 & 21.694 & 21.780 & 21.892 \\ 10.0 & 21.787 & 21.779 & 21.770 & 21.758 & 21.743 & 21.731 & 21.719 & 21.721 & 21.742 & 21.819 & 21.922 \\ 10.2 & 21.858 & 21.851 & 21.842 & 21.831 & 21.816 & 21.804 & 21.791 & 21.786 & 21.810 & 21.860 & 21.955 \\ 10.4 & 21.929 & 21.922 & 21.914 & 21.904 & 21.890 & 21.876 & 21.864 & 21.854 & 21.867 & 21.905 & 21.998 \\ 10.6 & 22.001 & 21.994 & 21.986 & 21.977 & 21.963 & 21.949 & 21.937 & 21.925 & 21.929 & 21.953 & 22.038 \\ 10.8 & 22.073 & 22.066 & 22.058 & 22.050 & 22.038 & 22.023 & 22.011 & 21.998 & 21.995 & 22.023 & 22.081 \\ 11.0 & 22.143 & 22.140 & 22.132 & 22.123 & 22.113 & 22.099 & 22.085 & 22.073 & 22.065 & 22.082 & 22.127 \\ \hline \end{tabular} } \caption {$\log\sigma_0$ for $\isotope[56]{Fe}$ at $B_{12}=1$} \label{tab:sigma0_b12_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 17.169 & 17.065 & 16.945 & 16.800 & 16.654 & 16.512 & 16.370 & 16.226 & 16.078 & 15.925 & 15.766 \\ 6.2 & 17.472 & 17.395 & 17.295 & 17.161 & 17.023 & 16.886 & 16.748 & 16.607 & 16.460 & 16.309 & 16.151 \\ 6.4 & 17.768 & 17.715 & 17.637 & 17.519 & 17.391 & 17.258 & 17.124 & 16.986 & 16.842 & 16.692 & 16.536 \\ 6.6 & 18.057 & 18.026 & 17.957 & 17.872 & 17.755 & 17.629 & 17.499 & 17.364 & 17.223 & 17.075 & 16.920 \\ 6.8 & 18.345 & 18.331 & 18.285 & 18.219 & 18.115 & 17.997 & 17.873 & 17.742 & 17.603 & 17.457 & 17.304 \\ 7.0 & 18.634 & 18.631 & 18.605 & 18.560 & 18.471 & 18.362 & 18.245 & 18.117 & 17.982 & 17.839 & 17.687 \\ 7.2 & 18.926 & 18.930 & 18.919 & 18.879 & 18.821 & 18.725 & 18.614 & 18.492 & 18.360 & 18.219 & 18.070 \\ 7.4 & 19.221 & 19.228 & 19.228 & 19.205 & 19.165 & 19.082 & 18.979 & 18.863 & 18.736 & 18.598 & 18.452 \\ 7.6 & 19.519 & 19.527 & 19.533 & 19.524 & 19.500 & 19.432 & 19.339 & 19.231 & 19.109 & 18.976 & 18.832 \\ 7.8 & 19.816 & 19.825 & 19.833 & 19.833 & 19.809 & 19.771 & 19.692 & 19.593 & 19.479 & 19.351 & 19.210 \\ 8.0 & 20.108 & 20.118 & 20.126 & 20.130 & 20.119 & 20.093 & 20.031 & 19.944 & 19.840 & 19.720 & 19.586 \\ 8.2 & 20.391 & 20.399 & 20.405 & 20.411 & 20.407 & 20.394 & 20.348 & 20.279 & 20.189 & 20.081 & 19.955 \\ 8.4 & 20.655 & 20.661 & 20.665 & 20.669 & 20.669 & 20.657 & 20.635 & 20.587 & 20.518 & 20.426 & 20.314 \\ 8.6 & 20.893 & 20.895 & 20.897 & 20.898 & 20.897 & 20.892 & 20.882 & 20.858 & 20.814 & 20.747 & 20.655 \\ 8.8 & 21.098 & 21.096 & 21.094 & 21.092 & 21.088 & 21.086 & 21.083 & 21.082 & 21.066 & 21.031 & 20.968 \\ 9.0 & 21.267 & 21.263 & 21.257 & 21.251 & 21.245 & 21.241 & 21.246 & 21.256 & 21.268 & 21.268 & 21.241 \\ 9.2 & 21.404 & 21.398 & 21.390 & 21.381 & 21.373 & 21.365 & 21.370 & 21.385 & 21.421 & 21.452 & 21.465 \\ 9.4 & 21.517 & 21.510 & 21.501 & 21.489 & 21.480 & 21.469 & 21.469 & 21.481 & 21.531 & 21.584 & 21.635 \\ 9.6 & 21.613 & 21.606 & 21.596 & 21.584 & 21.572 & 21.561 & 21.555 & 21.574 & 21.610 & 21.676 & 21.755 \\ 9.8 & 21.698 & 21.691 & 21.681 & 21.669 & 21.656 & 21.644 & 21.635 & 21.645 & 21.674 & 21.748 & 21.838 \\ 10.0 & 21.777 & 21.769 & 21.761 & 21.749 & 21.735 & 21.723 & 21.711 & 21.713 & 21.732 & 21.804 & 21.896 \\ 10.2 & 21.853 & 21.845 & 21.837 & 21.826 & 21.812 & 21.799 & 21.786 & 21.781 & 21.805 & 21.853 & 21.943 \\ 10.4 & 21.926 & 21.919 & 21.911 & 21.901 & 21.887 & 21.873 & 21.861 & 21.852 & 21.864 & 21.901 & 21.992 \\ 10.6 & 21.999 & 21.992 & 21.984 & 21.975 & 21.962 & 21.948 & 21.936 & 21.924 & 21.927 & 21.951 & 22.036 \\ 10.8 & 22.072 & 22.066 & 22.057 & 22.049 & 22.037 & 22.023 & 22.010 & 21.998 & 21.994 & 22.022 & 22.080 \\ 11.0 & 22.142 & 22.139 & 22.132 & 22.123 & 22.112 & 22.098 & 22.085 & 22.073 & 22.064 & 22.081 & 22.127 \\ \hline \end{tabular} } \end{center} \end{table} \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma_0$ for $\isotope[56]{Fe}$ at $B_{12}=10$} \label{tab:sigma0_b13_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 15.171 & 15.068 & 14.948 & 14.801 & 14.655 & 14.512 & 14.370 & 14.226 & 14.078 & 13.925 & 13.766 \\ 6.2 & 15.475 & 15.398 & 15.298 & 15.163 & 15.024 & 14.886 & 14.748 & 14.607 & 14.460 & 14.309 & 14.151 \\ 6.4 & 15.771 & 15.718 & 15.641 & 15.522 & 15.392 & 15.259 & 15.125 & 14.986 & 14.842 & 14.692 & 14.536 \\ 6.6 & 16.061 & 16.031 & 15.961 & 15.875 & 15.757 & 15.630 & 15.500 & 15.365 & 15.223 & 15.075 & 14.920 \\ 6.8 & 16.350 & 16.336 & 16.290 & 16.224 & 16.118 & 15.999 & 15.874 & 15.742 & 15.603 & 15.457 & 15.304 \\ 7.0 & 16.640 & 16.638 & 16.612 & 16.568 & 16.476 & 16.366 & 16.246 & 16.119 & 15.983 & 15.839 & 15.688 \\ 7.2 & 16.934 & 16.939 & 16.929 & 16.889 & 16.830 & 16.731 & 16.617 & 16.494 & 16.361 & 16.220 & 16.070 \\ 7.4 & 17.234 & 17.242 & 17.243 & 17.220 & 17.180 & 17.093 & 16.986 & 16.868 & 16.739 & 16.600 & 16.452 \\ 7.6 & 17.539 & 17.549 & 17.556 & 17.548 & 17.524 & 17.452 & 17.353 & 17.240 & 17.115 & 16.979 & 16.834 \\ 7.8 & 17.848 & 17.859 & 17.870 & 17.872 & 17.848 & 17.806 & 17.718 & 17.611 & 17.490 & 17.357 & 17.214 \\ 8.0 & 18.162 & 18.174 & 18.186 & 18.194 & 18.183 & 18.156 & 18.080 & 17.980 & 17.864 & 17.735 & 17.594 \\ 8.2 & 18.479 & 18.492 & 18.504 & 18.516 & 18.515 & 18.502 & 18.438 & 18.346 & 18.236 & 18.111 & 17.973 \\ 8.4 & 18.798 & 18.812 & 18.824 & 18.837 & 18.843 & 18.826 & 18.792 & 18.710 & 18.606 & 18.485 & 18.351 \\ 8.6 & 19.119 & 19.132 & 19.145 & 19.158 & 19.169 & 19.163 & 19.141 & 19.071 & 18.974 & 18.858 & 18.727 \\ 8.8 & 19.441 & 19.454 & 19.467 & 19.479 & 19.492 & 19.494 & 19.485 & 19.427 & 19.338 & 19.229 & 19.102 \\ 9.0 & 19.763 & 19.774 & 19.787 & 19.799 & 19.813 & 19.820 & 19.806 & 19.776 & 19.697 & 19.596 & 19.474 \\ 9.2 & 20.084 & 20.092 & 20.105 & 20.117 & 20.129 & 20.140 & 20.135 & 20.116 & 20.051 & 19.957 & 19.843 \\ 9.4 & 20.399 & 20.406 & 20.417 & 20.429 & 20.440 & 20.451 & 20.453 & 20.444 & 20.392 & 20.309 & 20.204 \\ 9.6 & 20.704 & 20.711 & 20.719 & 20.730 & 20.739 & 20.749 & 20.754 & 20.742 & 20.715 & 20.647 & 20.555 \\ 9.8 & 20.995 & 21.000 & 21.006 & 21.014 & 21.021 & 21.028 & 21.034 & 21.028 & 21.012 & 20.963 & 20.887 \\ 10.0 & 21.261 & 21.265 & 21.268 & 21.273 & 21.277 & 21.281 & 21.284 & 21.282 & 21.273 & 21.244 & 21.190 \\ 10.2 & 21.496 & 21.497 & 21.498 & 21.499 & 21.500 & 21.500 & 21.499 & 21.497 & 21.493 & 21.482 & 21.455 \\ 10.4 & 21.694 & 21.693 & 21.691 & 21.689 & 21.686 & 21.682 & 21.678 & 21.673 & 21.673 & 21.673 & 21.675 \\ 10.6 & 21.855 & 21.853 & 21.849 & 21.844 & 21.838 & 21.831 & 21.824 & 21.816 & 21.815 & 21.820 & 21.846 \\ 10.8 & 21.986 & 21.982 & 21.977 & 21.971 & 21.963 & 21.953 & 21.944 & 21.934 & 21.930 & 21.945 & 21.973 \\ 11.0 & 22.091 & 22.091 & 22.085 & 22.078 & 22.069 & 22.058 & 22.046 & 22.036 & 22.028 & 22.040 & 22.070 \\ \hline \end{tabular} } \caption {$\log\sigma_0$ for $\isotope[56]{Fe}$ at $B_{12}=100$} \label{tab:sigma0_b14_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 13.171 & 13.068 & 12.948 & 12.801 & 12.655 & 12.512 & 12.370 & 12.226 & 12.078 & 11.925 & 11.766 \\ 6.2 & 13.475 & 13.398 & 13.298 & 13.163 & 13.025 & 12.886 & 12.748 & 12.607 & 12.460 & 12.309 & 12.151 \\ 6.4 & 13.771 & 13.718 & 13.641 & 13.522 & 13.392 & 13.259 & 13.125 & 12.986 & 12.842 & 12.692 & 12.536 \\ 6.6 & 14.061 & 14.031 & 13.961 & 13.876 & 13.757 & 13.630 & 13.500 & 13.365 & 13.223 & 13.075 & 12.920 \\ 6.8 & 14.350 & 14.336 & 14.290 & 14.224 & 14.118 & 13.999 & 13.874 & 13.742 & 13.603 & 13.457 & 13.304 \\ 7.0 & 14.640 & 14.638 & 14.612 & 14.568 & 14.476 & 14.366 & 14.246 & 14.119 & 13.983 & 13.839 & 13.688 \\ 7.2 & 14.934 & 14.939 & 14.929 & 14.889 & 14.830 & 14.731 & 14.617 & 14.494 & 14.361 & 14.220 & 14.070 \\ 7.4 & 15.234 & 15.242 & 15.243 & 15.221 & 15.180 & 15.093 & 14.986 & 14.868 & 14.739 & 14.600 & 14.452 \\ 7.6 & 15.539 & 15.549 & 15.557 & 15.548 & 15.525 & 15.452 & 15.353 & 15.240 & 15.115 & 14.979 & 14.834 \\ 7.8 & 15.849 & 15.860 & 15.871 & 15.872 & 15.848 & 15.806 & 15.718 & 15.611 & 15.490 & 15.358 & 15.214 \\ 8.0 & 16.163 & 16.175 & 16.186 & 16.195 & 16.184 & 16.157 & 16.081 & 15.980 & 15.864 & 15.735 & 15.594 \\ 8.2 & 16.480 & 16.493 & 16.505 & 16.517 & 16.516 & 16.503 & 16.440 & 16.347 & 16.236 & 16.111 & 15.973 \\ 8.4 & 16.800 & 16.813 & 16.826 & 16.839 & 16.845 & 16.828 & 16.794 & 16.712 & 16.607 & 16.486 & 16.351 \\ 8.6 & 17.122 & 17.135 & 17.149 & 17.161 & 17.173 & 17.166 & 17.145 & 17.074 & 16.976 & 16.860 & 16.728 \\ 8.8 & 17.447 & 17.459 & 17.473 & 17.485 & 17.499 & 17.501 & 17.493 & 17.433 & 17.343 & 17.232 & 17.104 \\ 9.0 & 17.773 & 17.784 & 17.798 & 17.811 & 17.825 & 17.833 & 17.819 & 17.788 & 17.707 & 17.603 & 17.479 \\ 9.2 & 18.100 & 18.110 & 18.124 & 18.137 & 18.150 & 18.163 & 18.158 & 18.139 & 18.070 & 17.971 & 17.852 \\ 9.4 & 18.428 & 18.437 & 18.450 & 18.464 & 18.476 & 18.491 & 18.494 & 18.487 & 18.428 & 18.337 & 18.224 \\ 9.6 & 18.757 & 18.765 & 18.776 & 18.790 & 18.803 & 18.817 & 18.826 & 18.813 & 18.783 & 18.701 & 18.593 \\ 9.8 & 19.085 & 19.093 & 19.103 & 19.117 & 19.130 & 19.144 & 19.156 & 19.152 & 19.133 & 19.063 & 18.961 \\ 10.0 & 19.413 & 19.421 & 19.430 & 19.443 & 19.457 & 19.469 & 19.484 & 19.487 & 19.479 & 19.420 & 19.326 \\ 10.2 & 19.741 & 19.748 & 19.757 & 19.768 & 19.782 & 19.795 & 19.809 & 19.817 & 19.803 & 19.772 & 19.687 \\ 10.4 & 20.067 & 20.074 & 20.082 & 20.092 & 20.105 & 20.119 & 20.132 & 20.143 & 20.138 & 20.118 & 20.044 \\ 10.6 & 20.391 & 20.397 & 20.405 & 20.414 & 20.426 & 20.440 & 20.451 & 20.464 & 20.466 & 20.456 & 20.394 \\ 10.8 & 20.710 & 20.716 & 20.723 & 20.731 & 20.742 & 20.755 & 20.766 & 20.779 & 20.785 & 20.769 & 20.734 \\ 11.0 & 21.007 & 21.028 & 21.034 & 21.042 & 21.050 & 21.062 & 21.073 & 21.083 & 21.092 & 21.084 & 21.059 \\ \hline \end{tabular} } \caption {$\log\sigma_1$ for $\isotope[56]{Fe}$ at $B_{12}=1$} \label{tab:sigma0_b12_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 18.600 & 18.601 & 18.601 & 18.601 & 18.597 & 18.596 & 18.597 & 18.597 & 18.597 & 18.597 & 18.597 \\ 6.2 & 18.800 & 18.800 & 18.801 & 18.801 & 18.798 & 18.796 & 18.797 & 18.797 & 18.797 & 18.797 & 18.797 \\ 6.4 & 18.999 & 19.000 & 19.000 & 19.000 & 18.999 & 18.996 & 18.996 & 18.997 & 18.997 & 18.997 & 18.997 \\ 6.6 & 19.199 & 19.199 & 19.199 & 19.200 & 19.200 & 19.196 & 19.196 & 19.197 & 19.197 & 19.197 & 19.197 \\ 6.8 & 19.398 & 19.398 & 19.398 & 19.399 & 19.399 & 19.396 & 19.396 & 19.396 & 19.397 & 19.397 & 19.397 \\ 7.0 & 19.596 & 19.596 & 19.596 & 19.597 & 19.598 & 19.595 & 19.596 & 19.596 & 19.596 & 19.597 & 19.597 \\ 7.2 & 19.793 & 19.793 & 19.793 & 19.794 & 19.795 & 19.796 & 19.794 & 19.795 & 19.796 & 19.796 & 19.796 \\ 7.4 & 19.988 & 19.988 & 19.988 & 19.989 & 19.990 & 19.993 & 19.992 & 19.994 & 19.995 & 19.996 & 19.996 \\ 7.6 & 20.181 & 20.180 & 20.179 & 20.179 & 20.181 & 20.186 & 20.186 & 20.190 & 20.193 & 20.195 & 20.196 \\ 7.8 & 20.368 & 20.366 & 20.365 & 20.364 & 20.367 & 20.372 & 20.378 & 20.384 & 20.389 & 20.392 & 20.394 \\ 8.0 & 20.547 & 20.544 & 20.541 & 20.539 & 20.541 & 20.547 & 20.560 & 20.570 & 20.580 & 20.587 & 20.591 \\ 8.2 & 20.712 & 20.707 & 20.702 & 20.697 & 20.697 & 20.702 & 20.724 & 20.742 & 20.762 & 20.776 & 20.785 \\ 8.4 & 20.856 & 20.848 & 20.841 & 20.833 & 20.829 & 20.841 & 20.861 & 20.893 & 20.927 & 20.954 & 20.972 \\ 8.6 & 20.972 & 20.961 & 20.949 & 20.939 & 20.929 & 20.938 & 20.961 & 21.014 & 21.065 & 21.111 & 21.145 \\ 8.8 & 21.052 & 21.038 & 21.022 & 21.008 & 20.994 & 20.996 & 21.016 & 21.087 & 21.162 & 21.237 & 21.296 \\ 9.0 & 21.095 & 21.080 & 21.060 & 21.042 & 21.024 & 21.017 & 21.050 & 21.110 & 21.210 & 21.319 & 21.412 \\ 9.2 & 21.105 & 21.089 & 21.067 & 21.045 & 21.025 & 21.010 & 21.031 & 21.088 & 21.211 & 21.347 & 21.481 \\ 9.4 & 21.089 & 21.074 & 21.052 & 21.027 & 21.005 & 20.984 & 20.991 & 21.033 & 21.165 & 21.321 & 21.495 \\ 9.6 & 21.057 & 21.041 & 21.020 & 20.994 & 20.970 & 20.947 & 20.940 & 20.989 & 21.087 & 21.252 & 21.457 \\ 9.8 & 21.014 & 20.997 & 20.979 & 20.952 & 20.926 & 20.903 & 20.885 & 20.914 & 20.993 & 21.166 & 21.376 \\ 10.0 & 20.964 & 20.947 & 20.930 & 20.905 & 20.877 & 20.854 & 20.831 & 20.841 & 20.894 & 21.060 & 21.269 \\ 10.2 & 20.910 & 20.895 & 20.878 & 20.855 & 20.827 & 20.802 & 20.778 & 20.772 & 20.830 & 20.947 & 21.148 \\ 10.4 & 20.854 & 20.840 & 20.823 & 20.803 & 20.776 & 20.749 & 20.725 & 20.709 & 20.743 & 20.836 & 21.038 \\ 10.6 & 20.798 & 20.785 & 20.768 & 20.750 & 20.724 & 20.696 & 20.673 & 20.651 & 20.665 & 20.730 & 20.920 \\ 10.8 & 20.744 & 20.731 & 20.715 & 20.697 & 20.674 & 20.645 & 20.620 & 20.597 & 20.595 & 20.664 & 20.803 \\ 11.0 & 20.693 & 20.678 & 20.663 & 20.645 & 20.624 & 20.596 & 20.569 & 20.546 & 20.534 & 20.579 & 20.692 \\ \hline \end{tabular} } \end{center} \end{table} \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma_1$ for $\isotope[56]{Fe}$ at $B_{12}=10$} \label{tab:sigma0_b13_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 17.601 & 17.601 & 17.601 & 17.601 & 17.597 & 17.596 & 17.597 & 17.597 & 17.597 & 17.597 & 17.597 \\ 6.2 & 17.801 & 17.801 & 17.801 & 17.801 & 17.799 & 17.796 & 17.797 & 17.797 & 17.797 & 17.797 & 17.797 \\ 6.4 & 18.001 & 18.001 & 18.001 & 18.001 & 18.000 & 17.996 & 17.997 & 17.997 & 17.997 & 17.997 & 17.997 \\ 6.6 & 18.201 & 18.201 & 18.201 & 18.201 & 18.201 & 18.196 & 18.197 & 18.197 & 18.197 & 18.197 & 18.197 \\ 6.8 & 18.401 & 18.401 & 18.401 & 18.401 & 18.401 & 18.396 & 18.397 & 18.397 & 18.397 & 18.397 & 18.397 \\ 7.0 & 18.601 & 18.601 & 18.601 & 18.601 & 18.601 & 18.596 & 18.597 & 18.597 & 18.597 & 18.597 & 18.597 \\ 7.2 & 18.801 & 18.801 & 18.801 & 18.801 & 18.801 & 18.799 & 18.797 & 18.797 & 18.797 & 18.797 & 18.797 \\ 7.4 & 19.001 & 19.001 & 19.001 & 19.001 & 19.001 & 19.001 & 18.997 & 18.997 & 18.997 & 18.997 & 18.997 \\ 7.6 & 19.201 & 19.201 & 19.201 & 19.201 & 19.201 & 19.201 & 19.196 & 19.197 & 19.197 & 19.197 & 19.197 \\ 7.8 & 19.401 & 19.401 & 19.401 & 19.401 & 19.401 & 19.401 & 19.398 & 19.396 & 19.397 & 19.397 & 19.397 \\ 8.0 & 19.601 & 19.600 & 19.601 & 19.601 & 19.601 & 19.601 & 19.600 & 19.596 & 19.596 & 19.597 & 19.597 \\ 8.2 & 19.800 & 19.800 & 19.800 & 19.800 & 19.800 & 19.800 & 19.800 & 19.796 & 19.796 & 19.796 & 19.797 \\ 8.4 & 19.999 & 19.999 & 19.999 & 19.999 & 19.999 & 19.999 & 19.999 & 19.996 & 19.996 & 19.996 & 19.996 \\ 8.6 & 20.198 & 20.198 & 20.198 & 20.198 & 20.197 & 20.197 & 20.198 & 20.197 & 20.195 & 20.196 & 20.196 \\ 8.8 & 20.396 & 20.396 & 20.395 & 20.395 & 20.394 & 20.394 & 20.394 & 20.396 & 20.393 & 20.394 & 20.395 \\ 9.0 & 20.592 & 20.591 & 20.591 & 20.590 & 20.589 & 20.589 & 20.589 & 20.590 & 20.589 & 20.592 & 20.594 \\ 9.2 & 20.784 & 20.784 & 20.782 & 20.781 & 20.780 & 20.779 & 20.779 & 20.780 & 20.784 & 20.786 & 20.790 \\ 9.4 & 20.971 & 20.970 & 20.968 & 20.966 & 20.964 & 20.962 & 20.961 & 20.961 & 20.969 & 20.974 & 20.983 \\ 9.6 & 21.148 & 21.146 & 21.144 & 21.140 & 21.136 & 21.132 & 21.130 & 21.133 & 21.140 & 21.152 & 21.168 \\ 9.8 & 21.310 & 21.307 & 21.303 & 21.297 & 21.291 & 21.285 & 21.279 & 21.281 & 21.289 & 21.314 & 21.339 \\ 10.0 & 21.448 & 21.443 & 21.437 & 21.429 & 21.419 & 21.411 & 21.401 & 21.399 & 21.406 & 21.445 & 21.486 \\ 10.2 & 21.553 & 21.547 & 21.539 & 21.529 & 21.515 & 21.502 & 21.489 & 21.481 & 21.500 & 21.536 & 21.599 \\ 10.4 & 21.622 & 21.614 & 21.604 & 21.592 & 21.574 & 21.557 & 21.541 & 21.527 & 21.541 & 21.582 & 21.673 \\ 10.6 & 21.655 & 21.646 & 21.633 & 21.620 & 21.600 & 21.579 & 21.560 & 21.542 & 21.547 & 21.584 & 21.697 \\ 10.8 & 21.658 & 21.648 & 21.634 & 21.620 & 21.600 & 21.575 & 21.554 & 21.533 & 21.529 & 21.580 & 21.677 \\ 11.0 & 21.642 & 21.629 & 21.616 & 21.600 & 21.581 & 21.555 & 21.530 & 21.509 & 21.496 & 21.533 & 21.624 \\ \hline \end{tabular} } \caption {$\log\sigma_1$ for $\isotope[56]{Fe}$ at $B_{12}=100$} \label{tab:sigma0_b14_Fe} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 16.601 & 16.601 & 16.601 & 16.601 & 16.597 & 16.596 & 16.597 & 16.597 & 16.597 & 16.597 & 16.597 \\ 6.2 & 16.801 & 16.801 & 16.801 & 16.801 & 16.799 & 16.796 & 16.797 & 16.797 & 16.797 & 16.797 & 16.797 \\ 6.4 & 17.001 & 17.001 & 17.001 & 17.001 & 17.000 & 16.996 & 16.997 & 16.997 & 16.997 & 16.997 & 16.997 \\ 6.6 & 17.201 & 17.201 & 17.201 & 17.201 & 17.201 & 17.196 & 17.197 & 17.197 & 17.197 & 17.197 & 17.197 \\ 6.8 & 17.401 & 17.401 & 17.401 & 17.401 & 17.401 & 17.396 & 17.397 & 17.397 & 17.397 & 17.397 & 17.397 \\ 7.0 & 17.601 & 17.601 & 17.601 & 17.601 & 17.601 & 17.596 & 17.597 & 17.597 & 17.597 & 17.597 & 17.597 \\ 7.2 & 17.801 & 17.801 & 17.801 & 17.801 & 17.801 & 17.799 & 17.797 & 17.797 & 17.797 & 17.797 & 17.797 \\ 7.4 & 18.001 & 18.001 & 18.001 & 18.001 & 18.001 & 18.001 & 17.997 & 17.997 & 17.997 & 17.997 & 17.997 \\ 7.6 & 18.201 & 18.201 & 18.201 & 18.201 & 18.201 & 18.201 & 18.197 & 18.197 & 18.197 & 18.197 & 18.197 \\ 7.8 & 18.401 & 18.401 & 18.401 & 18.401 & 18.401 & 18.401 & 18.398 & 18.397 & 18.397 & 18.397 & 18.397 \\ 8.0 & 18.601 & 18.601 & 18.601 & 18.601 & 18.601 & 18.601 & 18.600 & 18.597 & 18.597 & 18.597 & 18.597 \\ 8.2 & 18.801 & 18.801 & 18.801 & 18.801 & 18.801 & 18.801 & 18.801 & 18.797 & 18.797 & 18.797 & 18.797 \\ 8.4 & 19.001 & 19.001 & 19.001 & 19.001 & 19.001 & 19.001 & 19.001 & 18.997 & 18.997 & 18.997 & 18.997 \\ 8.6 & 19.201 & 19.201 & 19.201 & 19.201 & 19.201 & 19.201 & 19.201 & 19.200 & 19.197 & 19.197 & 19.197 \\ 8.8 & 19.401 & 19.401 & 19.401 & 19.401 & 19.401 & 19.401 & 19.401 & 19.401 & 19.397 & 19.397 & 19.397 \\ 9.0 & 19.601 & 19.601 & 19.601 & 19.601 & 19.601 & 19.601 & 19.601 & 19.601 & 19.597 & 19.597 & 19.597 \\ 9.2 & 19.801 & 19.801 & 19.801 & 19.801 & 19.801 & 19.801 & 19.801 & 19.801 & 19.800 & 19.797 & 19.797 \\ 9.4 & 20.001 & 20.001 & 20.001 & 20.001 & 20.001 & 20.001 & 20.001 & 20.001 & 20.001 & 19.996 & 19.997 \\ 9.6 & 20.201 & 20.201 & 20.201 & 20.200 & 20.200 & 20.200 & 20.200 & 20.200 & 20.200 & 20.196 & 20.196 \\ 9.8 & 20.400 & 20.400 & 20.400 & 20.400 & 20.400 & 20.400 & 20.400 & 20.400 & 20.400 & 20.399 & 20.396 \\ 10.0 & 20.599 & 20.599 & 20.599 & 20.599 & 20.599 & 20.599 & 20.599 & 20.599 & 20.599 & 20.599 & 20.595 \\ 10.2 & 20.798 & 20.798 & 20.798 & 20.797 & 20.797 & 20.797 & 20.797 & 20.796 & 20.797 & 20.797 & 20.794 \\ 10.4 & 20.995 & 20.995 & 20.995 & 20.994 & 20.994 & 20.993 & 20.993 & 20.993 & 20.993 & 20.993 & 20.994 \\ 10.6 & 21.190 & 21.190 & 21.190 & 21.189 & 21.188 & 21.187 & 21.187 & 21.186 & 21.185 & 21.186 & 21.189 \\ 10.8 & 21.382 & 21.381 & 21.381 & 21.380 & 21.379 & 21.377 & 21.376 & 21.374 & 21.373 & 21.374 & 21.377 \\ 11.0 & 21.558 & 21.566 & 21.565 & 21.564 & 21.562 & 21.559 & 21.557 & 21.554 & 21.551 & 21.553 & 21.556 \\ \hline \end{tabular} } \caption {$\log\sigma$ for density-dependent composition} \label{tab:sigma_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 20.144 & 20.260 & 20.376 & 20.506 & 20.628 & 20.762 & 20.902 & 21.048 & 21.200 & 21.357 & 21.522 \\ 6.2 & 20.213 & 20.310 & 20.413 & 20.537 & 20.658 & 20.785 & 20.923 & 21.066 & 21.216 & 21.372 & 21.536 \\ 6.4 & 20.293 & 20.368 & 20.456 & 20.570 & 20.689 & 20.809 & 20.944 & 21.085 & 21.233 & 21.388 & 21.550 \\ 6.6 & 20.382 & 20.435 & 20.517 & 20.607 & 20.720 & 20.834 & 20.966 & 21.104 & 21.250 & 21.403 & 21.565 \\ 6.8 & 20.477 & 20.510 & 20.572 & 20.647 & 20.753 & 20.861 & 20.989 & 21.125 & 21.268 & 21.420 & 21.580 \\ 7.0 & 20.536 & 20.555 & 20.598 & 20.671 & 20.755 & 20.857 & 20.981 & 21.114 & 21.256 & 21.406 & 21.565 \\ 7.2 & 20.633 & 20.641 & 20.666 & 20.721 & 20.792 & 20.893 & 21.006 & 21.136 & 21.276 & 21.424 & 21.581 \\ 7.4 & 20.730 & 20.730 & 20.740 & 20.777 & 20.833 & 20.928 & 21.033 & 21.160 & 21.296 & 21.442 & 21.598 \\ 7.6 & 20.823 & 20.817 & 20.819 & 20.839 & 20.893 & 20.964 & 21.062 & 21.184 & 21.317 & 21.461 & 21.615 \\ 7.8 & 20.912 & 20.903 & 20.899 & 20.907 & 20.944 & 21.003 & 21.094 & 21.209 & 21.339 & 21.481 & 21.633 \\ 8.0 & 20.998 & 20.987 & 20.980 & 20.979 & 21.001 & 21.044 & 21.132 & 21.236 & 21.363 & 21.501 & 21.652 \\ 8.2 & 21.080 & 21.068 & 21.059 & 21.053 & 21.063 & 21.105 & 21.169 & 21.265 & 21.387 & 21.523 & 21.671 \\ 8.4 & 21.160 & 21.147 & 21.136 & 21.128 & 21.129 & 21.157 & 21.208 & 21.296 & 21.412 & 21.545 & 21.691 \\ 8.6 & 21.232 & 21.220 & 21.207 & 21.198 & 21.193 & 21.209 & 21.248 & 21.332 & 21.438 & 21.567 & 21.711 \\ 8.8 & 21.307 & 21.296 & 21.283 & 21.273 & 21.264 & 21.270 & 21.308 & 21.370 & 21.466 & 21.592 & 21.733 \\ 9.0 & 21.380 & 21.371 & 21.358 & 21.346 & 21.337 & 21.334 & 21.360 & 21.409 & 21.497 & 21.617 & 21.756 \\ 9.2 & 21.311 & 21.304 & 21.297 & 21.288 & 21.282 & 21.284 & 21.302 & 21.353 & 21.428 & 21.540 & 21.679 \\ 9.4 & 21.382 & 21.375 & 21.369 & 21.360 & 21.352 & 21.351 & 21.362 & 21.400 & 21.465 & 21.569 & 21.704 \\ 9.6 & 21.478 & 21.472 & 21.465 & 21.455 & 21.446 & 21.442 & 21.445 & 21.472 & 21.539 & 21.621 & 21.751 \\ 9.8 & 21.549 & 21.543 & 21.536 & 21.527 & 21.517 & 21.511 & 21.511 & 21.528 & 21.581 & 21.659 & 21.779 \\ 10.0 & 21.619 & 21.614 & 21.606 & 21.599 & 21.589 & 21.580 & 21.579 & 21.588 & 21.628 & 21.698 & 21.809 \\ 10.2 & 21.717 & 21.712 & 21.704 & 21.696 & 21.686 & 21.675 & 21.670 & 21.673 & 21.702 & 21.761 & 21.863 \\ 10.4 & 21.787 & 21.782 & 21.776 & 21.768 & 21.758 & 21.747 & 21.741 & 21.739 & 21.758 & 21.817 & 21.903 \\ 10.6 & 21.888 & 21.883 & 21.876 & 21.868 & 21.858 & 21.846 & 21.837 & 21.831 & 21.841 & 21.888 & 21.967 \\ 10.8 & 21.976 & 21.970 & 21.964 & 21.955 & 21.946 & 21.933 & 21.922 & 21.915 & 21.917 & 21.952 & 22.021 \\ 11.0 & - & 22.055 & 22.049 & 22.040 & 22.031 & 22.018 & 22.006 & 21.997 & 21.994 & 22.019 & 22.076 \\ \hline \end{tabular} } \end{center} \end{table} \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma_0$ for density-dependent composition at $B_{12}=1$} \label{tab:sigma0_b12_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 17.169 & 17.065 & 16.945 & 16.800 & 16.654 & 16.512 & 16.370 & 16.226 & 16.078 & 15.925 & 15.765 \\ 6.2 & 17.472 & 17.395 & 17.294 & 17.161 & 17.023 & 16.885 & 16.748 & 16.607 & 16.460 & 16.309 & 16.151 \\ 6.4 & 17.768 & 17.715 & 17.637 & 17.519 & 17.390 & 17.258 & 17.124 & 16.986 & 16.842 & 16.692 & 16.536 \\ 6.6 & 18.057 & 18.026 & 17.957 & 17.872 & 17.754 & 17.628 & 17.499 & 17.364 & 17.223 & 17.075 & 16.920 \\ 6.8 & 18.345 & 18.331 & 18.285 & 18.219 & 18.115 & 17.997 & 17.873 & 17.741 & 17.603 & 17.457 & 17.304 \\ 7.0 & 18.650 & 18.645 & 18.617 & 18.555 & 18.480 & 18.370 & 18.252 & 18.124 & 17.989 & 17.845 & 17.694 \\ 7.2 & 18.943 & 18.943 & 18.931 & 18.889 & 18.830 & 18.733 & 18.621 & 18.499 & 18.367 & 18.226 & 18.076 \\ 7.4 & 19.238 & 19.242 & 19.240 & 19.216 & 19.174 & 19.090 & 18.986 & 18.870 & 18.743 & 18.605 & 18.458 \\ 7.6 & 19.534 & 19.541 & 19.544 & 19.533 & 19.494 & 19.439 & 19.346 & 19.238 & 19.116 & 18.982 & 18.838 \\ 7.8 & 19.829 & 19.837 & 19.843 & 19.841 & 19.817 & 19.777 & 19.697 & 19.599 & 19.485 & 19.357 & 19.217 \\ 8.0 & 20.119 & 20.127 & 20.133 & 20.136 & 20.124 & 20.098 & 20.035 & 19.949 & 19.846 & 19.726 & 19.591 \\ 8.2 & 20.397 & 20.404 & 20.409 & 20.413 & 20.408 & 20.386 & 20.350 & 20.281 & 20.193 & 20.085 & 19.960 \\ 8.4 & 20.655 & 20.660 & 20.663 & 20.666 & 20.664 & 20.653 & 20.632 & 20.585 & 20.518 & 20.429 & 20.317 \\ 8.6 & 20.871 & 20.872 & 20.873 & 20.873 & 20.872 & 20.867 & 20.858 & 20.835 & 20.791 & 20.725 & 20.633 \\ 8.8 & 21.069 & 21.067 & 21.064 & 21.062 & 21.059 & 21.056 & 21.058 & 21.056 & 21.042 & 21.008 & 20.945 \\ 9.0 & 21.232 & 21.228 & 21.222 & 21.216 & 21.211 & 21.208 & 21.214 & 21.226 & 21.240 & 21.243 & 21.217 \\ 9.2 & 21.258 & 21.253 & 21.247 & 21.240 & 21.235 & 21.236 & 21.247 & 21.281 & 21.323 & 21.374 & 21.410 \\ 9.4 & 21.352 & 21.346 & 21.341 & 21.333 & 21.326 & 21.325 & 21.333 & 21.363 & 21.412 & 21.482 & 21.553 \\ 9.6 & 21.458 & 21.452 & 21.446 & 21.437 & 21.428 & 21.425 & 21.427 & 21.450 & 21.507 & 21.571 & 21.662 \\ 9.8 & 21.537 & 21.532 & 21.525 & 21.517 & 21.507 & 21.501 & 21.501 & 21.517 & 21.566 & 21.635 & 21.735 \\ 10.0 & 21.613 & 21.608 & 21.601 & 21.593 & 21.583 & 21.575 & 21.573 & 21.582 & 21.621 & 21.687 & 21.788 \\ 10.2 & 21.713 & 21.708 & 21.700 & 21.693 & 21.682 & 21.672 & 21.667 & 21.669 & 21.697 & 21.754 & 21.851 \\ 10.4 & 21.785 & 21.780 & 21.773 & 21.766 & 21.756 & 21.745 & 21.739 & 21.737 & 21.755 & 21.814 & 21.897 \\ 10.6 & 21.887 & 21.882 & 21.874 & 21.867 & 21.857 & 21.844 & 21.835 & 21.830 & 21.839 & 21.886 & 21.964 \\ 10.8 & 21.975 & 21.969 & 21.963 & 21.955 & 21.945 & 21.932 & 21.921 & 21.914 & 21.916 & 21.951 & 22.019 \\ 11.0 & - & 22.055 & 22.048 & 22.040 & 22.031 & 22.018 & 22.005 & 21.996 & 21.993 & 22.018 & 22.075 \\ \hline \end{tabular} } \caption {$\log\sigma_0$ for density-dependent composition at $B_{12}=10$} \label{tab:sigma0_b13_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 15.171 & 15.068 & 14.948 & 14.801 & 14.654 & 14.512 & 14.370 & 14.226 & 14.078 & 13.925 & 13.765 \\ 6.2 & 15.475 & 15.398 & 15.297 & 15.163 & 15.024 & 14.886 & 14.748 & 14.607 & 14.461 & 14.309 & 14.151 \\ 6.4 & 15.771 & 15.718 & 15.641 & 15.521 & 15.392 & 15.259 & 15.124 & 14.986 & 14.842 & 14.692 & 14.536 \\ 6.6 & 16.061 & 16.031 & 15.961 & 15.875 & 15.757 & 15.630 & 15.500 & 15.364 & 15.223 & 15.075 & 14.920 \\ 6.8 & 16.350 & 16.336 & 16.290 & 16.224 & 16.118 & 15.999 & 15.874 & 15.742 & 15.603 & 15.457 & 15.304 \\ 7.0 & 16.657 & 16.652 & 16.625 & 16.563 & 16.486 & 16.374 & 16.254 & 16.126 & 15.990 & 15.846 & 15.694 \\ 7.2 & 16.952 & 16.954 & 16.942 & 16.901 & 16.840 & 16.740 & 16.625 & 16.501 & 16.368 & 16.227 & 16.077 \\ 7.4 & 17.252 & 17.258 & 17.257 & 17.233 & 17.190 & 17.102 & 16.994 & 16.875 & 16.745 & 16.607 & 16.459 \\ 7.6 & 17.557 & 17.565 & 17.570 & 17.560 & 17.519 & 17.461 & 17.361 & 17.248 & 17.122 & 16.986 & 16.840 \\ 7.8 & 17.867 & 17.877 & 17.884 & 17.884 & 17.859 & 17.816 & 17.726 & 17.619 & 17.497 & 17.364 & 17.221 \\ 8.0 & 18.180 & 18.192 & 18.201 & 18.207 & 18.195 & 18.166 & 18.089 & 17.988 & 17.871 & 17.741 & 17.601 \\ 8.2 & 18.497 & 18.510 & 18.520 & 18.529 & 18.526 & 18.496 & 18.448 & 18.354 & 18.243 & 18.117 & 17.979 \\ 8.4 & 18.817 & 18.829 & 18.840 & 18.850 & 18.855 & 18.838 & 18.802 & 18.718 & 18.613 & 18.492 & 18.357 \\ 8.6 & 19.116 & 19.128 & 19.140 & 19.150 & 19.158 & 19.151 & 19.127 & 19.055 & 18.956 & 18.839 & 18.708 \\ 8.8 & 19.438 & 19.449 & 19.461 & 19.472 & 19.482 & 19.482 & 19.456 & 19.411 & 19.320 & 19.210 & 19.082 \\ 9.0 & 19.760 & 19.769 & 19.781 & 19.793 & 19.803 & 19.809 & 19.793 & 19.761 & 19.680 & 19.577 & 19.455 \\ 9.2 & 20.169 & 20.175 & 20.181 & 20.189 & 20.195 & 20.194 & 20.185 & 20.150 & 20.094 & 19.999 & 19.883 \\ 9.4 & 20.473 & 20.478 & 20.483 & 20.490 & 20.496 & 20.497 & 20.492 & 20.469 & 20.427 & 20.345 & 20.240 \\ 9.6 & 20.721 & 20.726 & 20.730 & 20.736 & 20.742 & 20.745 & 20.743 & 20.729 & 20.691 & 20.632 & 20.540 \\ 9.8 & 20.993 & 20.996 & 20.999 & 21.002 & 21.006 & 21.008 & 21.008 & 21.001 & 20.978 & 20.937 & 20.864 \\ 10.0 & 21.234 & 21.235 & 21.236 & 21.237 & 21.238 & 21.239 & 21.238 & 21.235 & 21.225 & 21.203 & 21.156 \\ 10.2 & 21.425 & 21.424 & 21.424 & 21.424 & 21.423 & 21.422 & 21.420 & 21.418 & 21.415 & 21.407 & 21.383 \\ 10.4 & 21.603 & 21.601 & 21.599 & 21.597 & 21.593 & 21.589 & 21.586 & 21.583 & 21.585 & 21.594 & 21.596 \\ 10.6 & 21.754 & 21.751 & 21.748 & 21.744 & 21.739 & 21.732 & 21.727 & 21.723 & 21.724 & 21.740 & 21.758 \\ 10.8 & 21.887 & 21.884 & 21.880 & 21.874 & 21.868 & 21.859 & 21.851 & 21.845 & 21.845 & 21.863 & 21.893 \\ 11.0 & - & 21.998 & 21.993 & 21.986 & 21.979 & 21.969 & 21.959 & 21.951 & 21.947 & 21.963 & 21.997 \\ \hline \end{tabular} } \caption {$\log\sigma_0$ for density-dependent composition at $B_{12}=100$} \label{tab:sigma0_b13_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 13.171 & 13.068 & 12.948 & 12.801 & 12.654 & 12.512 & 12.370 & 12.226 & 12.078 & 11.925 & 11.765 \\ 6.2 & 13.475 & 13.398 & 13.297 & 13.163 & 13.024 & 12.886 & 12.748 & 12.607 & 12.461 & 12.309 & 12.151 \\ 6.4 & 13.771 & 13.718 & 13.641 & 13.521 & 13.392 & 13.259 & 13.124 & 12.986 & 12.842 & 12.692 & 12.536 \\ 6.6 & 14.061 & 14.031 & 13.961 & 13.875 & 13.757 & 13.630 & 13.500 & 13.364 & 13.223 & 13.075 & 12.920 \\ 6.8 & 14.350 & 14.336 & 14.290 & 14.224 & 14.118 & 13.999 & 13.874 & 13.742 & 13.603 & 13.457 & 13.304 \\ 7.0 & 14.657 & 14.652 & 14.625 & 14.563 & 14.486 & 14.374 & 14.254 & 14.126 & 13.990 & 13.846 & 13.694 \\ 7.2 & 14.952 & 14.954 & 14.942 & 14.901 & 14.840 & 14.740 & 14.625 & 14.501 & 14.368 & 14.227 & 14.077 \\ 7.4 & 15.252 & 15.258 & 15.257 & 15.233 & 15.190 & 15.102 & 14.994 & 14.875 & 14.745 & 14.607 & 14.459 \\ 7.6 & 15.557 & 15.565 & 15.570 & 15.560 & 15.520 & 15.461 & 15.361 & 15.248 & 15.122 & 14.986 & 14.840 \\ 7.8 & 15.867 & 15.877 & 15.885 & 15.885 & 15.860 & 15.816 & 15.727 & 15.619 & 15.497 & 15.364 & 15.221 \\ 8.0 & 16.181 & 16.193 & 16.201 & 16.208 & 16.196 & 16.167 & 16.090 & 15.988 & 15.871 & 15.742 & 15.601 \\ 8.2 & 16.498 & 16.511 & 16.521 & 16.530 & 16.528 & 16.498 & 16.449 & 16.355 & 16.244 & 16.118 & 15.980 \\ 8.4 & 16.819 & 16.831 & 16.843 & 16.852 & 16.857 & 16.840 & 16.804 & 16.720 & 16.615 & 16.493 & 16.358 \\ 8.6 & 17.119 & 17.131 & 17.144 & 17.154 & 17.163 & 17.155 & 17.132 & 17.058 & 16.958 & 16.841 & 16.708 \\ 8.8 & 17.444 & 17.455 & 17.468 & 17.479 & 17.489 & 17.490 & 17.463 & 17.418 & 17.325 & 17.213 & 17.084 \\ 9.0 & 17.771 & 17.780 & 17.793 & 17.805 & 17.816 & 17.822 & 17.807 & 17.773 & 17.690 & 17.584 & 17.459 \\ 9.2 & 18.201 & 18.208 & 18.215 & 18.225 & 18.232 & 18.232 & 18.222 & 18.185 & 18.124 & 18.020 & 17.897 \\ 9.4 & 18.530 & 18.536 & 18.543 & 18.552 & 18.561 & 18.563 & 18.559 & 18.532 & 18.484 & 18.387 & 18.269 \\ 9.6 & 18.804 & 18.811 & 18.817 & 18.827 & 18.837 & 18.842 & 18.843 & 18.826 & 18.775 & 18.701 & 18.589 \\ 9.8 & 19.133 & 19.139 & 19.146 & 19.154 & 19.164 & 19.172 & 19.174 & 19.165 & 19.127 & 19.064 & 18.957 \\ 10.0 & 19.461 & 19.467 & 19.474 & 19.481 & 19.491 & 19.500 & 19.504 & 19.500 & 19.473 & 19.423 & 19.322 \\ 10.2 & 19.730 & 19.735 & 19.743 & 19.750 & 19.761 & 19.771 & 19.777 & 19.779 & 19.762 & 19.723 & 19.631 \\ 10.4 & 20.056 & 20.061 & 20.067 & 20.074 & 20.084 & 20.095 & 20.102 & 20.106 & 20.097 & 20.055 & 19.989 \\ 10.6 & 20.317 & 20.322 & 20.328 & 20.336 & 20.345 & 20.357 & 20.366 & 20.373 & 20.370 & 20.339 & 20.284 \\ 10.8 & 20.604 & 20.609 & 20.615 & 20.622 & 20.631 & 20.642 & 20.653 & 20.661 & 20.663 & 20.642 & 20.598 \\ 11.0 & - & 20.872 & 20.878 & 20.885 & 20.893 & 20.904 & 20.915 & 20.923 & 20.928 & 20.915 & 20.882 \\ \hline \end{tabular} } \end{center} \end{table} \begin{table} \begin{center} \vskip -1cm \caption {$\log\sigma_1$ for density-dependent composition at $B_{12}=1$} \label{tab:sigma0_b12_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 18.600 & 18.601 & 18.601 & 18.601 & 18.597 & 18.596 & 18.597 & 18.597 & 18.597 & 18.597 & 18.597 \\ 6.2 & 18.800 & 18.800 & 18.801 & 18.801 & 18.798 & 18.796 & 18.797 & 18.797 & 18.797 & 18.797 & 18.797 \\ 6.4 & 18.999 & 19.000 & 19.000 & 19.000 & 18.999 & 18.996 & 18.996 & 18.997 & 18.997 & 18.997 & 18.997 \\ 6.6 & 19.199 & 19.199 & 19.199 & 19.200 & 19.200 & 19.196 & 19.196 & 19.197 & 19.197 & 19.197 & 19.197 \\ 6.8 & 19.398 & 19.398 & 19.398 & 19.399 & 19.399 & 19.396 & 19.396 & 19.396 & 19.397 & 19.397 & 19.397 \\ 7.0 & 19.583 & 19.583 & 19.583 & 19.585 & 19.586 & 19.582 & 19.583 & 19.584 & 19.584 & 19.584 & 19.585 \\ 7.2 & 19.780 & 19.780 & 19.780 & 19.781 & 19.783 & 19.783 & 19.782 & 19.783 & 19.784 & 19.784 & 19.784 \\ 7.4 & 19.975 & 19.974 & 19.974 & 19.975 & 19.977 & 19.980 & 19.979 & 19.982 & 19.983 & 19.984 & 19.984 \\ 7.6 & 20.166 & 20.165 & 20.164 & 20.165 & 20.168 & 20.173 & 20.174 & 20.178 & 20.181 & 20.183 & 20.184 \\ 7.8 & 20.352 & 20.350 & 20.348 & 20.348 & 20.351 & 20.357 & 20.363 & 20.370 & 20.376 & 20.380 & 20.382 \\ 8.0 & 20.527 & 20.524 & 20.522 & 20.520 & 20.522 & 20.529 & 20.544 & 20.555 & 20.567 & 20.574 & 20.579 \\ 8.2 & 20.688 & 20.683 & 20.678 & 20.674 & 20.675 & 20.688 & 20.705 & 20.726 & 20.747 & 20.762 & 20.772 \\ 8.4 & 20.826 & 20.818 & 20.810 & 20.804 & 20.801 & 20.814 & 20.837 & 20.873 & 20.910 & 20.938 & 20.958 \\ 8.6 & 20.927 & 20.916 & 20.905 & 20.896 & 20.889 & 20.899 & 20.925 & 20.981 & 21.035 & 21.083 & 21.118 \\ 8.8 & 21.000 & 20.987 & 20.972 & 20.960 & 20.948 & 20.953 & 20.993 & 21.052 & 21.130 & 21.208 & 21.268 \\ 9.0 & 21.037 & 21.023 & 21.004 & 20.987 & 20.973 & 20.969 & 21.005 & 21.071 & 21.174 & 21.287 & 21.383 \\ 9.2 & 20.813 & 20.801 & 20.788 & 20.772 & 20.762 & 20.768 & 20.803 & 20.897 & 21.021 & 21.185 & 21.349 \\ 9.4 & 20.778 & 20.765 & 20.753 & 20.736 & 20.722 & 20.723 & 20.746 & 20.825 & 20.952 & 21.132 & 21.335 \\ 9.6 & 20.795 & 20.782 & 20.769 & 20.751 & 20.733 & 20.727 & 20.737 & 20.797 & 20.933 & 21.092 & 21.316 \\ 9.8 & 20.744 & 20.733 & 20.719 & 20.703 & 20.683 & 20.672 & 20.675 & 20.716 & 20.832 & 20.995 & 21.225 \\ 10.0 & 20.690 & 20.680 & 20.665 & 20.651 & 20.631 & 20.615 & 20.614 & 20.638 & 20.731 & 20.884 & 21.110 \\ 10.2 & 20.704 & 20.693 & 20.679 & 20.663 & 20.642 & 20.622 & 20.614 & 20.623 & 20.692 & 20.827 & 21.043 \\ 10.4 & 20.647 & 20.637 & 20.623 & 20.608 & 20.589 & 20.567 & 20.555 & 20.556 & 20.603 & 20.738 & 20.927 \\ 10.6 & 20.667 & 20.656 & 20.642 & 20.626 & 20.606 & 20.582 & 20.565 & 20.557 & 20.584 & 20.695 & 20.873 \\ 10.8 & 20.652 & 20.641 & 20.627 & 20.611 & 20.592 & 20.566 & 20.545 & 20.532 & 20.543 & 20.629 & 20.789 \\ 11.0 & - & 20.632 & 20.619 & 20.602 & 20.583 & 20.557 & 20.533 & 20.516 & 20.515 & 20.579 & 20.717 \\ \hline \end{tabular} } \caption {$\log\sigma_1$ for density-dependent composition at $B_{12}=10$} \label{tab:sigma0_b13_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 17.601 & 17.601 & 17.601 & 17.601 & 17.597 & 17.596 & 17.597 & 17.597 & 17.597 & 17.597 & 17.597 \\ 6.2 & 17.801 & 17.801 & 17.801 & 17.801 & 17.799 & 17.796 & 17.797 & 17.797 & 17.797 & 17.797 & 17.797 \\ 6.4 & 18.001 & 18.001 & 18.001 & 18.001 & 18.000 & 17.996 & 17.997 & 17.997 & 17.997 & 17.997 & 17.997 \\ 6.6 & 18.201 & 18.201 & 18.201 & 18.201 & 18.201 & 18.196 & 18.197 & 18.197 & 18.197 & 18.197 & 18.197 \\ 6.8 & 18.401 & 18.401 & 18.401 & 18.401 & 18.401 & 18.396 & 18.397 & 18.397 & 18.397 & 18.397 & 18.397 \\ 7.0 & 18.589 & 18.589 & 18.589 & 18.589 & 18.589 & 18.584 & 18.585 & 18.585 & 18.585 & 18.585 & 18.585 \\ 7.2 & 18.789 & 18.789 & 18.789 & 18.789 & 18.789 & 18.787 & 18.785 & 18.785 & 18.785 & 18.785 & 18.785 \\ 7.4 & 18.989 & 18.989 & 18.989 & 18.989 & 18.989 & 18.988 & 18.984 & 18.985 & 18.985 & 18.985 & 18.985 \\ 7.6 & 19.189 & 19.189 & 19.189 & 19.189 & 19.189 & 19.189 & 19.184 & 19.185 & 19.185 & 19.185 & 19.185 \\ 7.8 & 19.389 & 19.389 & 19.389 & 19.389 & 19.389 & 19.389 & 19.385 & 19.384 & 19.385 & 19.385 & 19.385 \\ 8.0 & 19.588 & 19.588 & 19.588 & 19.588 & 19.588 & 19.589 & 19.588 & 19.584 & 19.584 & 19.585 & 19.585 \\ 8.2 & 19.788 & 19.788 & 19.788 & 19.788 & 19.788 & 19.788 & 19.788 & 19.784 & 19.784 & 19.784 & 19.785 \\ 8.4 & 19.987 & 19.987 & 19.987 & 19.987 & 19.987 & 19.987 & 19.987 & 19.983 & 19.984 & 19.984 & 19.984 \\ 8.6 & 20.172 & 20.172 & 20.172 & 20.171 & 20.171 & 20.171 & 20.172 & 20.171 & 20.169 & 20.170 & 20.170 \\ 8.8 & 20.369 & 20.369 & 20.369 & 20.368 & 20.368 & 20.368 & 20.369 & 20.369 & 20.367 & 20.368 & 20.369 \\ 9.0 & 20.565 & 20.564 & 20.564 & 20.563 & 20.562 & 20.562 & 20.563 & 20.564 & 20.563 & 20.566 & 20.568 \\ 9.2 & 20.724 & 20.723 & 20.722 & 20.720 & 20.719 & 20.719 & 20.720 & 20.725 & 20.730 & 20.735 & 20.742 \\ 9.4 & 20.899 & 20.897 & 20.896 & 20.893 & 20.891 & 20.890 & 20.891 & 20.897 & 20.907 & 20.918 & 20.931 \\ 9.6 & 21.058 & 21.056 & 21.053 & 21.050 & 21.046 & 21.044 & 21.043 & 21.049 & 21.066 & 21.079 & 21.100 \\ 9.8 & 21.200 & 21.197 & 21.193 & 21.188 & 21.181 & 21.177 & 21.175 & 21.181 & 21.204 & 21.231 & 21.264 \\ 10.0 & 21.311 & 21.307 & 21.300 & 21.294 & 21.285 & 21.277 & 21.274 & 21.279 & 21.306 & 21.347 & 21.401 \\ 10.2 & 21.416 & 21.410 & 21.403 & 21.394 & 21.383 & 21.371 & 21.365 & 21.365 & 21.391 & 21.438 & 21.511 \\ 10.4 & 21.465 & 21.458 & 21.449 & 21.439 & 21.426 & 21.410 & 21.401 & 21.398 & 21.421 & 21.491 & 21.578 \\ 10.6 & 21.534 & 21.526 & 21.515 & 21.504 & 21.488 & 21.470 & 21.456 & 21.447 & 21.462 & 21.530 & 21.630 \\ 10.8 & 21.564 & 21.555 & 21.544 & 21.531 & 21.514 & 21.493 & 21.475 & 21.462 & 21.468 & 21.530 & 21.637 \\ 11.0 & - & 21.575 & 21.563 & 21.548 & 21.532 & 21.508 & 21.486 & 21.471 & 21.468 & 21.518 & 21.623 \\ \hline \end{tabular} } \caption {$\log\sigma_1$ for density-dependent composition at $B_{12}=100$} \label{tab:sigma0_b14_comp} {\tiny \begin{tabular}{ |c||c|c|c|c|c|c|c|c|c|c|c| } \hline $\log\rho$ & $\log T$=-1.0 & $\log T$=-0.8 & $\log T$=-0.6 & $\log T$=-0.4 & $\log T$=-0.2 & $\log T$=0.0 & $\log T$=0.2 & $\log T$=0.4 & $\log T$=0.6 & $\log T$=0.8 & $\log T$=1.0 \\ \hline 6.0 & 16.601 & 16.601 & 16.601 & 16.601 & 16.597 & 16.596 & 16.597 & 16.597 & 16.597 & 16.597 & 16.597 \\ 6.2 & 16.801 & 16.801 & 16.801 & 16.801 & 16.799 & 16.796 & 16.797 & 16.797 & 16.797 & 16.797 & 16.797 \\ 6.4 & 17.001 & 17.001 & 17.001 & 17.001 & 17.000 & 16.996 & 16.997 & 16.997 & 16.997 & 16.997 & 16.997 \\ 6.6 & 17.201 & 17.201 & 17.201 & 17.201 & 17.201 & 17.196 & 17.197 & 17.197 & 17.197 & 17.197 & 17.197 \\ 6.8 & 17.401 & 17.401 & 17.401 & 17.401 & 17.401 & 17.396 & 17.397 & 17.397 & 17.397 & 17.397 & 17.397 \\ 7.0 & 17.589 & 17.589 & 17.589 & 17.589 & 17.589 & 17.584 & 17.585 & 17.585 & 17.585 & 17.585 & 17.585 \\ 7.2 & 17.789 & 17.789 & 17.789 & 17.789 & 17.789 & 17.787 & 17.785 & 17.785 & 17.785 & 17.785 & 17.785 \\ 7.4 & 17.989 & 17.989 & 17.989 & 17.989 & 17.989 & 17.989 & 17.985 & 17.985 & 17.985 & 17.985 & 17.985 \\ 7.6 & 18.189 & 18.189 & 18.189 & 18.189 & 18.189 & 18.189 & 18.185 & 18.185 & 18.185 & 18.185 & 18.185 \\ 7.8 & 18.389 & 18.389 & 18.389 & 18.389 & 18.389 & 18.389 & 18.386 & 18.385 & 18.385 & 18.385 & 18.385 \\ 8.0 & 18.589 & 18.589 & 18.589 & 18.589 & 18.589 & 18.589 & 18.588 & 18.585 & 18.585 & 18.585 & 18.585 \\ 8.2 & 18.789 & 18.789 & 18.789 & 18.789 & 18.789 & 18.789 & 18.789 & 18.785 & 18.785 & 18.785 & 18.785 \\ 8.4 & 18.989 & 18.989 & 18.989 & 18.989 & 18.989 & 18.989 & 18.989 & 18.985 & 18.985 & 18.985 & 18.985 \\ 8.6 & 19.175 & 19.175 & 19.175 & 19.175 & 19.175 & 19.175 & 19.175 & 19.174 & 19.171 & 19.171 & 19.171 \\ 8.8 & 19.375 & 19.375 & 19.375 & 19.375 & 19.375 & 19.375 & 19.375 & 19.375 & 19.371 & 19.371 & 19.371 \\ 9.0 & 19.575 & 19.575 & 19.575 & 19.575 & 19.575 & 19.575 & 19.575 & 19.575 & 19.571 & 19.571 & 19.571 \\ 9.2 & 19.756 & 19.756 & 19.756 & 19.756 & 19.756 & 19.756 & 19.756 & 19.756 & 19.754 & 19.752 & 19.752 \\ 9.4 & 19.956 & 19.956 & 19.956 & 19.955 & 19.956 & 19.956 & 19.956 & 19.956 & 19.955 & 19.951 & 19.951 \\ 9.6 & 20.141 & 20.141 & 20.141 & 20.141 & 20.141 & 20.141 & 20.141 & 20.141 & 20.141 & 20.136 & 20.137 \\ 9.8 & 20.340 & 20.340 & 20.340 & 20.340 & 20.340 & 20.340 & 20.340 & 20.340 & 20.340 & 20.338 & 20.336 \\ 10.0 & 20.539 & 20.538 & 20.538 & 20.538 & 20.538 & 20.538 & 20.538 & 20.538 & 20.538 & 20.538 & 20.535 \\ 10.2 & 20.721 & 20.721 & 20.721 & 20.721 & 20.720 & 20.720 & 20.720 & 20.720 & 20.720 & 20.721 & 20.718 \\ 10.4 & 20.918 & 20.917 & 20.917 & 20.917 & 20.916 & 20.916 & 20.916 & 20.915 & 20.916 & 20.917 & 20.917 \\ 10.6 & 21.097 & 21.096 & 21.096 & 21.095 & 21.095 & 21.094 & 21.093 & 21.093 & 21.093 & 21.095 & 21.097 \\ 10.8 & 21.280 & 21.280 & 21.279 & 21.279 & 21.278 & 21.276 & 21.275 & 21.274 & 21.273 & 21.275 & 21.279 \\ 11.0 & - & 21.449 & 21.448 & 21.447 & 21.446 & 21.444 & 21.442 & 21.440 & 21.438 & 21.440 & 21.444 \\ \hline \end{tabular} } \end{center} \end{table}
7da1a44fb55455b4defef399ac008ce28f41d342
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} A popular version of the third law of thermodynamics is that the entropy density of a physical system tends to zero in the $T \to 0$ limit\cite{mandl}. However, there is a class of theoretical models that violate this law\cite{fowler33,pauling,nagle66,lieb67,chow87,bramwell01,castelnovo08}:\ models in this class exhibit a ground-state degeneracy which grows exponentially with the system size, leading to a non-zero entropy density even at $T=0$. Nor can these be easily dismissed as theorists' abstractions, since one also sees ample evidence in experiment\cite{giauque36,harris97,ramirez99,higashinaka03} that there are systems in which the entropy plateaus at a non-zero value over a large range of temperature. In many such cases it is suspected that it eventually falls to zero at a much lower temperature scale, though recent theoretical work on skyrmion magnets suggests that this intuition may not always be reliable \cite{moessner}. Whatever the ultimate low-temperature fate of these materials, it is clear that over a broad range of temperatures they exhibit physics which is well captured by models with a non-zero residual entropy density. One important class of these are so-called ice models, in which the ground-state manifold consists of all configurations which satisfy a certain local `ice rule' constraint\cite{siddharthan99,denhertog00,isakov05}. The first such model was Pauling's model for the residual configurational entropy of water ice\cite{pauling}. Here the local constraint is that two of the four hydrogens neighboring any given oxygen should be chemically bonded to it to form a water molecule. Similar models were subsequently discovered to apply to the orientations of spins along local Ising axes in certain rare-earth pyrochlores\cite{siddharthan99,bramwell01}, which by analogy were dubbed `spin ice' compounds. Such models develop power-law spin-spin correlations at low temperatures, with characteristic `pinch points' in the momentum representation of the spin-spin correlation function\cite{bramwell01a}, but they do not order. Their low-temperature state is often referred to as a `co-operative paramagnet' \cite{villain79}. One interesting feature of such co-operative paramagnets is their response to an applied magnetic field. The configurations that make up the ice-rule manifold usually have different magnetizations; thus an applied field, depending on its direction, may either reduce\cite{higashinaka03,hiroi03,tabata06} or entirely eliminate\cite{fukazawa02,jaubert08} the degeneracy. In the latter case, further interesting physics may arise when the system is heated, especially if the ice-rule constraints do not permit the thermal excitation of individual flipped spins. In such cases the lowest-free-energy excitation may be a {\it string\/} of flipped spins extending from one side of the system to the other. A demagnetization transition mediated by such excitations is known as a {\it Kasteleyn transition}\cite{kasteleyn,fennell07,jaubert08}. In spin ice research to date, insight has often been gained from the study of simplified models where the dimensionality is reduced or the geometry simplified while retaining the essential physics\cite{mengotti11,chern12,wan}. In that spirit, we present in this paper a two-dimensional ice model which exhibits a Kasteleyn transition in an applied magnetic field. The model is especially interesting since, unlike its three-dimensional counterparts, it has the same Ising quantization axis for every spin. This raises the possibility that it could be extended to include a transverse magnetic field, thereby allowing the exploration of quantum Kasteleyn physics. The remainder of this paper is structured as follows. In section \ref{sec:model}, we present our spin ice model, along with some analytical and numerical results on its thermodynamic properties in the absence of an applied magnetic field. In section \ref{sec:kasteleyn}, we analyse the model in the presence of a magnetic field:\ we show that it has a Kasteleyn transition, and we characterize it. In section \ref{sec:entropy}, we use an alternative representation of the ice-rule states --- the `string representation' --- to determine the model's entropy as a function of its magnetization. Finally, in section \ref{sec:summary}, we summarize our findings and discuss possible future lines of work. \section{The model} \label{sec:model} The model that we shall consider has the following Hamiltonian: \be H = \sum_{ij} J_{ij} \sigma_i \sigma_j - h \sum_i \sigma_i. \label{ham} \ee Here $i$ and $j$ label the sites of a two-dimensional square lattice, $\sigma_i = \pm 1$ is an Ising variable on lattice site $i$, and $h$ is an externally applied (longitudinal) magnetic field. The exchange interaction $J_{ij}$ is given by: \be J_{ij} = \left\{ \begin{array}{lll} \phantom{-}J & \qquad & {\bf r}_j = {\bf r}_i + {\hat {\bf x}}; \\ -J & \qquad & {\bf r}_j = {\bf r}_i + {\hat {\bf y}}; \\ -J & \qquad & {\bf r}_i = n {\hat {\bf x}} + m {\hat {\bf y}} \,\,\,\,(n+m\,\,\mbox{odd}) \\ & & \quad \mbox{and}\,{\bf r}_j = {\bf r}_i + {\hat {\bf x}} + {\hat {\bf y}}; \\ -J & \qquad & {\bf r}_i = n {\hat {\bf x}} + m {\hat {\bf y}} \,\,\,\,(n+m\,\,\mbox{even}) \\ & & \quad \mbox{and}\,{\bf r}_j = {\bf r}_i - {\hat {\bf x}} + {\hat {\bf y}}; \\ \phantom{-}0 & & \mbox{otherwise,} \end{array} \right. \label{exchanges} \ee where ${\bf r}_i$ is the position vector of site $i$, ${\hat {\bf x}}$ and ${\hat {\bf y}}$ are the unit vectors of a Cartesian system in the two-dimensional plane, and $J$ is a positive constant. In this paper, we shall always work in the limit $J \gg \vert h \vert, k_B T$. Furthermore, where necessary we shall take the number of sites in the lattice to be $N$, always assuming $N$ to be large enough that edge effects can be neglected. When we refer to the density of something (e.g.\ the entropy density), we shall always mean that quantity divided by the number of spins --- not, for example, by the number of plaquettes. The lattice described by (\ref{exchanges}) is shown in the upper-left inset of Fig.~\ref{defects}, with ferromagnetic bonds represented by solid lines and antiferromagnetic bonds represented by dotted lines. One may view this lattice as made of corner-sharing plaquettes, one of which is shown in the lower-right inset of Fig.~\ref{defects}. It is easy to see that the bonds on this plaquette cannnot all be satisfied at once:\ the model (\ref{ham}) is therefore magnetically frustrated. The sixteen spin configurations of the elementary plaquette, together with their energies, are shown in Table \ref{plaqconf}. \begin{table} \begin{center} \begin{tabular}{c|c|c|c|c|c|c} Configuration & $\ua\ua\ua\ua$ & $\ua\da\da\ua$ & $\ua\da\ua\da$ & $\da\ua\da\ua$ & $\da\ua\ua\da$ & $\da\da\da\da$ \\ \hline Energy & $-2J-4h$ & $-2J$ & $-2J$ & $-2J$ & $-2J$ & $-2J+4h$ \end{tabular} \vspace*{3mm} \begin{tabular}{c|c|c|c|c|c|c|c|c} Configuration & $\ua\ua\ua\da$ & $\ua\ua\da\ua$ & $\ua\da\ua\ua$ & $\da\ua\ua\ua$ & $\da\da\da\ua$ & $\da\da\ua\da$ & $\da\ua\da\da$ & $\ua\da\da\da$ \\ \hline Energy & $-2h$ & $-2h$ & $-2h$ & $-2h$ & $2h$ & $2h$ & $2h$ & $2h$ \end{tabular} \vspace*{3mm} \begin{tabular}{c|c|c} Configuration & $\ua\ua\da\da$ & $\da\da\ua\ua$ \\ \hline Energy & $6J$ & $6J$ \end{tabular} \end{center} \caption{The energies of the sixteen spin configurations of the elementary plaquette. Each configuration is specified by listing the orientations of the four plaquette spins in the order corresponding to the numbering in Fig.~\ref{defects}. The first six configurations listed are those that, in the absence of an external magnetic field, constitute the sixfold-degenerate ground-state (or `ice rule') manifold.} \label{plaqconf} \end{table} When $h=0$, i.e.\ in the absence of an external magnetic field, there are six degenerate ground-state configurations. They are shown in the left-hand inset of Fig.~\ref{C}:\ we shall call them the `ice-rule configurations,' and the manifold spanned by them the `ice-rule manifold.' Since this Ising model is magnetically frustrated, we do not expect it to show an ordering transition as the temperature is reduced. Rather, we expect a smooth crossover into a co-operative paramagnetic state in which every plaquette is in one of the ice-rule configurations. The density of defects (a measure of how many plaquettes are not in an ice-rule configuration) should vanish smoothly as the temperature tends to zero, and the specific heat will show a corresponding Schottky-like peak at temperatures $T \sim J/k_B$ but no sharp features. Because the ground-state degeneracy is exponential in the system size, the model will have a non-zero entropy density even at zero temperature. A na{\"\i}ve estimate would suggest a value of $k_B \ln 6$ per plaquette, i.e.\ $\frac{1}{2} k_B \ln 6 \approx 0.896\,k_B$ per spin, due to the six-fold ground-state degeneracy. This estimate, however, is too na{\"\i}ve, since it ignores the important constraint that the ice-rule configurations chosen for two neighboring plaquettes must agree on the orientation of the spin at their shared corner. We may easily improve our estimate of the zero-temperature entropy density by taking this constraint into account at a local level. Imagine `growing' a spin configuration of the lattice from top to bottom. Each time a new row is added, the orientations of spins 1 and 2 of each plaquette of the row being added ($j$) will be fixed by the (already chosen) configuration of the row above ($j-1$). The ice rules for this model do not favor any particular direction for any single site on the plaquette; hence the probabilities of the four configurations of this pair of spins are simply $P_{\ua\ua} = P_{\ua\da} = P_{\da\ua} = P_{\da\da} = 1/4$. The number of ice-rule configurations consistent with these constraints is (see Fig.~\ref{C}) $N_{\ua\ua} = N_{\da\da} = 1$; $N_{\ua\da} = N_{\da\ua} = 2$. Thus half the plaquettes in the new row have no choice of configuration, while the other half may choose between two. This gives an average entropy per plaquette of $\frac{1}{2} k_B \ln 2$, which corresponds to an entropy density of $\frac{1}{4} k_B \ln 2 \approx 0.173\,k_B$ per spin. This estimate is still rather crude, since it neglects correlations between the configurations of neighboring plaquettes in row $j-1$, which will be induced by their connections to a common plaquette in row $j-2$. However, it was shown by Lieb \cite{lieb67} that such correlation corrections may be resummed to yield an exact result for the ground-state entropy density of such `square ice' models:\ $s_0 \equiv S_0/N = \frac{3}{4} k_B \ln \left( \frac{4}{3} \right) \approx 0.216\,k_B$. We shall call this value the `Lieb entropy density,' and denote it $s_0^{\rm Lieb}$. All of the above expectations are borne out by Monte Carlo simulations of the model, the results of which are shown in Figs.~\ref{defects}--\ref{C}. First, we demonstrate the increasing predominance of ice-rule configurations as the temperature is lowered. For this it is useful to define the number of defects on a plaquette as the number of single spin-flips by which the spin configuration deviates from the closest ice-rule configuration. By this measure, the states in the top line of Table \ref{plaqconf} have zero defects, those in the second line have one, and those in the third line have two. Fig.~\ref{defects} shows the density of defects as a function of temperature. \begin{figure} \centerline{\includegraphics[width=0.95\columnwidth]{Fig1.eps}} \caption{The density of defects, $\rho_{\rm defects}$, as a function of scaled temperature, $k_B T/J$, for a lattice of 8192 spins and in the absence of an applied magnetic field. The number of defects on a plaquette is defined as the number of single spin-flips by which it differs from the nearest ice-rule configuration. Thus each state in the ground-state manifold of the system has $\rho_{\rm defects}=0$. The dotted line marks the high-temperature asymptotic value of $3/8$ (see text). Inset (top left): A portion of the lattice, with ferromagnetic bonds represented by solid lines and antiferromagnetic bonds by dotted lines. Inset (bottom right): The unit cell of the lattice, including the numbering convention we use for the spins on a single plaquette.} \label{defects} \end{figure} The asymptotic high-temperature value of this quantity can be easily calculated. In the infinite-temperature limit all configurations of a plaquette are equally probable, i.e.\ each has a probability $\frac{1}{16}$. From Table \ref{plaqconf}, we see that there are six configurations with no defects, eight configurations with one, and two configurations with two. Hence the average number of defects per plaquette at infinite temperature is $0 \times \frac{6}{16} + 1 \times \frac{8}{16} + 2 \times \frac{2}{16} = \frac{3}{4}$. Since there are twice as many spins as plaquettes, the defect density is simply half of this, i.e.\ $\rho_{\rm defects} \to \frac{3}{8} = 0.375$ as $k_B T/J \to \infty$. Second, we calculate the entropy density of the system as a function of temperature, using the Wang-Landau method\cite{wang2001}. The results are shown in Fig.~\ref{Entropy}. At high temperatures the entropy density tends to $k_B \ln 2$, the Ising paramagnetic value. At low temperatures it tends to a non-zero constant value which is in good agreement with the Lieb entropy density $s_0^{\rm Lieb}$ given above. In between there are no sharp features, confirming that the model exhibits only a crossover from high-temperature paramagnetic to low-temperature cooperative-paramagnetic behavior. \begin{figure} \centerline{\includegraphics[width=\columnwidth]{Fig2.eps}} \caption{The dimensionless entropy density of the system, $s/k_B$, as a function of scaled temperature, $k_B T/J$, for a lattice of 8192 spins and in the absence of an applied magnetic field, calculated using the Wang-Landau method. At high temperatures the entropy density is that of an Ising paramagnet, $k_B \ln 2$ per spin. The zero-temperature residual entropy density is consistent with Lieb's exact result for two-dimensional ice models, $s_0^{\rm Lieb} = \frac{3}{4} k_B \ln \left( \frac{4}{3} \right) \approx 0.216\,k_B$.} \label{Entropy} \end{figure} Third, we obtain the specific heat capacity as a function of temperature, also using the Wang-Landau method. The results are shown in Fig.~\ref{C}. In keeping with our results for the entropy density in Fig.~\ref{Entropy}, we see that although there is a broad Schottky-like peak at temperatures of order $J/k_B$ there are no sharp features, supporting our expectation that this model would not exhibit a phase transition. \begin{figure} \centerline{\includegraphics[width=0.85\columnwidth]{Fig3.eps}} \caption{The dimensionless heat capacity per spin, $C/k_B$, as a function of scaled temperature, $k_B T/J$, in the absence of an applied magnetic field, calculated using the Wang-Landau method. Inset (left):\ the six degenerate zero-field ground states for a single plaquette. Inset (right):\ the same states in the string representation.} \label{C} \end{figure} \section{Kasteleyn transition} \label{sec:kasteleyn} Our real interest in this model, however, is in its unusual response to an externally applied longitudinal magnetic field. We call this field $h$, and in the following we shall take it to be positive. As shown in the first line of Table \ref{plaqconf}, the degeneracy between the six ice-rule configurations is lifted as soon as the field $h$ is applied. Indeed, for any non-zero $h$ (and remembering that we always work in the $h \ll J$ limit) the ground state of a plaquette is the unique `all up' configuration. It follows that, at $T=0$, the entire lattice simply has $\sigma_i = +1$ for all sites $i$. Now let us consider what happens to this fully magnetized state as the temperature is increased. One might expect the appearance of a dilute set of `down' spins. However, a feature of this model is that a single spin-flip takes the system out of the ice-rule manifold, and at $h,k_B T \ll J$ this will not occur. To understand what will happen instead, let us introduce a representation of the states in the ice-rule manifold in terms of strings. We begin with a single plaquette. If we take as our reference state the one in which all the spins are up, we may represent the six ice-rule configurations in terms of lines joining the spins that are down. This is shown in the right-hand inset of Fig.~\ref{C}. Representing the `all down' configuration as two vertical lines rather than two horizontal ones is in principle arbitrary, but it has the advantage of yielding a model in which these lines of down spins can neither cross each other nor form closed loops. To make an ice-rule-obeying configuration of the entire lattice, we must put these plaquettes together in such a way that any string that leaves one plaquette enters its neighbor. Thus there is a one-to-one mapping between ice-rule-obeying configurations of the spins $\sigma_i$ and configurations of these strings. Each string must extend all the way across the lattice. To proceed further, let us suppose that the lattice consists of $L_x$ sites in the horizontal direction and $L_y$ sites in the vertical direction, so that $N = L_x L_y$. Each string, irrespective of its configuration, contains precisely $L_y$ spins, so that a configuration with $N_s$ strings has $N_s L_y$ down spins and thus an energy of $2 h N_s L_y$ relative to the fully magnetized state (or `string vacuum'). Such a string is the {\it minimal\/} demagnetizing excitation of the system that is consistent with the ice rule. Since a single string has an energy cost proportional to the linear size of the system, it might appear that such strings cannot be thermally excited. This is not true, however, because a single string also has two choices about which way to go every time it enters a new plaquette, meaning that its entropy of $k_B L_y \ln 2$ is also proportional to $L_y$. Thus the free-energy cost of introducing a single string into the fully magnetized state is \be F = E - TS = \left( 2h - k_B T \ln 2 \right) L_y. \ee When the temperature reaches the critical value $T_c = 2h/(k_B \ln 2)$, this free-energy cost flips sign, and the system becomes unstable to the proliferation of strings. (This is somewhat similar to what happens in a Berezinskii-Kosterlitz-Thouless transition\cite{b,kt}, except that in our model we do not have `positive' and `negative' strings, so the physics of screening plays no r{\^o}le.) In fact the increase in the string density from zero for $T > T_c$ --- which corresponds directly to the decrease in the magnetization from its saturated value --- is continuous. This is because the above argument applies strictly only to a single string introduced into the fully magnetized state. Once a finite density of strings has been created the entropy associated with new ones is reduced, and thus the temperature at which it becomes free-energetically favorable to create them goes up. This kind of transition, in which the elementary thermal excitations are system-spanning strings, is called a {\it Kasteleyn transition\/}. It was first described by Kasteleyn in the context of dimer models\cite{kasteleyn}. The above predictions are again borne out by our Monte Carlo simulations, the results of which are shown in Figs.~\ref{MvsHT}--\ref{TkvsH}. Fig.~\ref{MvsHT} shows a three-dimensional plot of the equilibrium value of the magnetization, $M$, as a function of the temperature and the applied magnetic field. At all temperatures below $T_c(h)$ the magnetization takes its saturated value; above $T_c(h)$ it decreases smoothly with increasing temperature, tending to zero only as $T \to \infty$. This may be understood in the string representation of the problem. As more and more strings are introduced, the entropy density of each new one decreases; in the limit where half the lattice sites are populated by strings it tends to zero, meaning that this will occur only in the infinite-temperature limit. \begin{figure} \centerline{\includegraphics[width=\columnwidth]{Fig4.eps}} \caption{The ratio of the magnetization to its saturated value, $M/M_{\rm sat}$, as a function of scaled temperature, $k_B T/J$, and scaled longitudinal field, $h/J$. The solid black line shows the theoretical prediction for the Kasteleyn transition temperature, $T_c = 2h/(k_B \ln 2)$.} \label{MvsHT} \end{figure} Fig.~\ref{chivsT} shows the magnetic susceptibility, determined at three different values of the applied field. In each case, one sees at $T=T_c(h)$ the asymmetric peak characteristic of a Kasteleyn transition. This highlights an intriguing consequence of the physics of the Kasteleyn strings:\ below $T_c(h)$ the linear susceptibility is strictly zero, while as $T_c(h)$ is approached from above the susceptibility diverges. For a two-dimensional Kasteleyn transition one expects to find $\beta= 1/2$ on the high-temperature side\cite{nagle1975,moessner2003}, that is, \be \mu \sim t^{1/2}, \ee where $\mu \equiv (M_{\rm sat}-M)/M_{\rm sat}$ is the reduced magnetization and $t \equiv (T - T_c)/T_c$ is the reduced temperature. This is indeed the case in our simulations:\ the inset of Fig.~\ref{chivsT} is a logarithmic plot of $\mu$ as a function of $t$, calculated for a system of 8192 spins and with an applied field of $h/J = 0.017$ (grey filled circles), compared with the expected $t^{1/2}$ behavior (solid red line). Similar behavior is found for all simulated fields below $0.1 h/J$. \begin{figure} \centerline{\includegraphics[width=\columnwidth]{Fig5.eps}} \caption{The magnetic susceptibility, $\chi$, as a function of scaled temperature, $k_B T/J$, for a lattice of 8192 spins with three different values of the scaled magnetic field $h/J$: 0.017 (black symbols, leftmost peak), 0.034 (red symbols, middle peak), and 0.051 (blue symbols, rightmost peak). The inset shows the reduced magnetization, $\mu$, as a function of the reduced temperature, $t$, for an applied field $h/J=0.017$ (grey filled circles). The solid red line corresponds to $\mu \sim t^{1/2}$.} \label{chivsT} \end{figure} In Fig.~\ref{TkvsH} we collect our data into a phase diagram. The filled red circles show the temperature of the Kasteleyn transition, determined from the data in Fig.~\ref{MvsHT} as the temperature at which the magnetization departs from its saturated value. The thick black line is the prediction $T_c(h) = 2h/(k_B \ln 2)$ derived above. The departure of the red points from this line at larger fields and temperatures is due to the violation of the condition $h, k_B T \ll J$. In the pink region the thermal excitations are not full strings, but instead string fragments extending from one ice-rule-violating plaquette to another. The physics of such string fragments, and their signatures in neutron scattering, were discussed by Wan and Tchernyshyov\cite{wan}. \begin{figure} \centerline{\includegraphics[width=\columnwidth]{Fig6.eps}} \caption{The phase diagram of the model as a function of scaled temperature, $k_B T/J$, and scaled magnetic field, $h/J$. The red dots show the Kasteleyn temperature as determined from the magnetization curves, i.e.\ the temperature at which the magnetization first departs from its saturated value. The black line is the theoretical prediction $T_c(h)= 2h/(k_B \ln 2)$. As expected, the simulation results depart from the theoretical prediction at temperatures where the condition that the spin configuration remain strictly in the ice-rule manifold, $h, k_B T\ll J$, is no longer fulfilled (pink area).} \label{TkvsH} \end{figure} \section{Entropy as a function of magnetization} \label{sec:entropy} Finally, let us demonstrate the usefulness of the string representation by using it to calculate the entropy density of the system, $s$, at a fixed value of the magnetization density, $m \equiv M/M_{\rm sat}$. Clearly $s(m)$ is an even function of $m$, so we may restrict our calculation to the case $m \geqslant 0$. The magnetization density may equivalently be expressed as the density of strings, $\eta_s$, via the formula $\eta_s = (1-m)/2$. To determine the entropy density corresponding to a given value of $\eta_s$, consider propagating the string configuration downwards from the top of the lattice. We shall assume that this propagation has reached a certain row $j$, and concentrate on a single string in that row. As it enters a new plaquette in row $j+1$, this string has in principle two choices:\ to continue vertically downwards, or to cross the plaquette diagonally. However, if another string is entering the same plaquette, it has only one choice, since the strings cannot cross (see Fig.~\ref{C}). The probability that a second string enters the same plaquette in row $j+1$ as the first is simply $\eta_s$. Thus the average number of choices available to the first string upon entering the new plaquette is $\eta_s \times 1 + (1-\eta_s) \times 2 = 2 - \eta_s$. This means that each string has a total entropy $S_s \approx k_B L_y \ln \left( 2 - \eta_s \right)$; with a total number of strings $\eta_s L_x$, it follows that the total entropy is $S \approx k_B L_x L_y \eta_s \ln \left( 2 - \eta_s \right)$. Dividing by the number of spins $N=L_x L_y$, and using $\eta_s = (1-m)/2$, we obtain \be s_0(m) \approx {\tilde s}_0(m) \equiv k_B \left( \frac{1-m}{2} \right) \ln \left( \frac{3+m}{2} \right). \label{sofm} \ee In Fig.~\ref{S0vsM} we compare this approximation with numerical results for the entropy density obtained using the Wang-Landau method. The filled black circles are the numerical results, while the dashed red curve is our analytical approximation (\ref{sofm}). It is clear that these were never going to coincide, since the $m \to 0$ limit of ${\tilde s}_0(m)$ is the Pauling entropy density, $\frac{1}{2} k_B \ln \left( \frac{3}{2} \right)$, while the $m \to 0$ limit of the actual entropy density is the Lieb entropy density, $\frac{3}{4} k_B \ln \left( \frac{4}{3} \right)$. \begin{figure} \centerline{\includegraphics[width=\columnwidth]{Fig7.eps}} \caption{The residual dimensionless entropy per site, $s_0/k_B$, as a function of the scaled magnetization per site, $m \equiv M/M_{\rm sat}$. The black filled circles show the values obtained numerically using the Wang-Landau method for a lattice of 512 spins. The dashed red line is the free-string result ${\tilde s}_0(m)$ (see text); note that it tends to Pauling's entropy at $m=0$ (filled red square). The solid blue line is the curve obtained by multiplying the number of microstates by a constant factor, chosen to rescale ${\tilde s}_0(0)$ to match Lieb's exact result (open blue circle).} \label{S0vsM} \end{figure} The origin of the difference between Lieb's exact result and Pauling's approximation lies in positive correlation of closed loops\cite{nagle66,lieb67}, which increases by a small factor the number of possible configurations obeying the ice rule. If one makes the crude assumption that this factor is independent of $m$, this results in a constant additive change to the logarithm in (\ref{sofm}): \be {\tilde s}_0(m) \longrightarrow k_B \left( \frac{1-m}{2} \right) \left[ \ln \left( \frac{3+m}{2} \right) + \alpha \right]. \ee If we choose the constant $\alpha$ to match the known result at $m=0$, the resulting curve (shown in blue) gives a very reasonable fit to the numerical data points over the whole range $0 \leqslant m \leqslant 1$. \vspace{5mm} \section{Summary and future work} \label{sec:summary} In this paper, we have presented a new spin-ice model defined on a two-dimensional lattice of mixed ferro- and antiferromagnetic bonds. We have analysed its thermodynamic properties in zero applied magnetic field, and we have also characterized the Kasteleyn transition that it exhibits when a magnetic field is applied. Finally, we have shown that its entropy when the magnetization is non-zero is well captured by the string representation. One appealing feature of this model is that, unlike full three-dimensional spin ices, the Ising quantization axis is the same on each lattice site. This makes it natural to consider adding to the model a spatially uniform transverse magnetic field, $\Gamma$. The results of this should be particularly interesting in the $h,\Gamma,k_B T \ll J$ regime, where the applied field is expected to stabilize the string phase at low temperatures, leading to a line of quantum Kasteleyn transitions in the zero-temperature $(h,\Gamma)$ plane. This extension of the model (\ref{ham}) is the subject of a forthcoming work \cite{dhgb}. \vspace{1mm} \begin{acknowledgments} We are pleased to acknowledge useful discussions with Rodolfo Borzi, Daniel Darroch, and Peter Holdsworth. This research was supported in part by the National Science Foundation under Grant No.\ NSF PHY11-25915, and CAH is grateful to the Kavli Institute for Theoretical Physics in Santa Barbara for their hospitality. CAH also gratefully acknowledges financial support from the EPSRC (UK) via grant number EP/I031014/1. SAG would like to acknowledge financial support from CONICET, and ANPCYT (Argentina) via grant PICT-2013-2004. \end{acknowledgments}
0de96d7b73ce49baf4120f35a6fc772776e9cc0f
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The calculation of the transport properties of strongly coupled quantum theories is a challenging puzzle of interest to theorists working on a wide range of systems including ultra-cold Fermi gases at unitarity~\cite{Adams:2012th,Bulgac:2010dg}, heavy ion collisions~\cite{Adams:2012th,Bhalerao:2010wf}, and neutron stars~\cite{Page:2006ud,Alford:2014doa}. At strong coupling, perturbative expansions fail to give reliable results. Sophisticated Monte-Carlo techniques which are used to study such theories non-perturbatively by evaluating path-integrals in imaginary time, while very successful for calculating equilibrium properties (in the Fermi gas context see Ref.~\cite{Gezerlis:2007fs} and Refs therein; for heavy ion collisions see Ref.~\cite{Gavai:2005dy} and Refs therein) can not be easily generalized to study transport (in the Fermi gas context see Ref.~\cite{Wlazlowski:2012jb,Wlazlowski:2015yga}; for heavy ion collisions see Ref.~\cite{Meyer:2009} and Refs therein). A class of strongly interacting quantum field theories in $d$ dimensions in some limits can be related to weakly coupled theories of gravity (called their dual) in $(d+1)$ dimensions. This correspondence~\cite{Maldacena:1999} allows us to compute dynamical properties of these theories. These computations have provided many insights into the transport properties of strongly coupled field theories. In certain limits (large t'Hooft coupling $\lambda$ and large number of ``colors'' $N_c$) one can show that for all isotropic theories in $3+1$ dimensions which admit gravity duals, the ratio of shear viscosity $\eta$ to entropy density $s$ is ${\eta \over s} = \frac{1}{4\pi}$~\cite{Son:2002sd,Kovtun:2004de} (we are working in units with $\hbar=1$ and $k_B=1$). Since weakly coupled theories typically have much larger ${\eta \over s}$, it was conjectured by Kovtun, Son and Starinets (KSS) that ${\eta \over s}$ is bounded from below by $1/(4\pi)$. Subsequently it was found that finite $\lambda$ corrections can drive ${\eta \over s}$ below the KSS bound~\cite{Brigante:2007nu,Brigante:2008gz,Kats:2007mq,Buchel:2008vz,Sinha:2009ev,Cremonini:2011iq}. While the theories describing ultra-cold Fermi gases and heavy ion collisions do not have known gravitational duals and controlled calculations are difficult, beautiful experiments have managed to measure the value of $\eta/s$ in the two systems. The value of $\eta/s$ of the quark gluon plasma created in heavy ion collisions, required for hydrodynamic simulations to be consistent with the experimentally measured spectrum of low energy particles (see Ref.~\cite{Heinz:2013th} for a review), seems to be close to $1/(4\pi)$. Remarkably, $\eta/s$ has been measured for ultra-cold fermions at unitarity for a wide range of temperatures and the minimum value (see Refs.~\cite{Schafer:2007pr,Cao58,Thomas:2015}) is about six times the KSS bound. On the other hand the shear viscosity tensor for many interesting systems is often anisotropic. For example, it has been suggested that the highly anisotropic initial states in heavy ion collisions (the direction parallel to the collision axes is fundamentally different from the transverse directions) may give rise to anisotropic transport properties~\cite{Martinez:2012tu}. Furthermore, many interesting states of matter, eg. spin density waves and spatially modulated phases, are anisotropic. Another possibility, that we shall explore in detail in this paper, is that an externally applied field can pick a particular direction and give rise to anisotropies in the shear viscosity. This possibility has been explored extensively for the case of weakly coupled theories in the presence of a background magnetic field. (See Ref.~\cite{Landau1987Fluid} for a classic treatment, Ref.~\cite{Tuchin:2011jw} for applications to heavy ion collisions and Ref.~\cite{Ofengeim:2015qxz} for applications to neutron stars.) The behavior of strongly coupled theories in the presence of an external field is less well explored. With this in mind, anisotropic gravitational backgrounds in field theory have been recently studied using the AdS/CFT correspondence (see \cite{Landsteiner:2007bd,Azeyanagi:2009pr,Natsuume:2010ky,Erdmenger:2010xm, Basu:2011tt,Erdmenger:2011tj,Mateos:2011ix,Mateos:2011tv,Iizuka:2012wt}) and the behavior of the viscosity in some of these anisotropic phases has also been analyzed (see \cite{Rebhan:2011vd, Polchinski:2012nh} and \cite{Giataganas:2012zy,Mamo:2012sy, Bhattacharyya:2014wfa, Jain:2014vka,Critelli:2014kra,Ge:2014aza}). The results of Ref.~\cite{Jain:2014vka} and Ref.~\cite{Jain:2015txa} for example, indicate that one may obtain parametric violations of the KSS bound in such anisotropic scenarios. This feature arises in a wide variety of examples and seems to be quite general. In particular, for a spatially constant driving force which breaks rotational invariance, it was found that by increasing the strength of the driving force compared to the temperature, the ratio for appropriate components of the shear viscosity to entropy density can be made arbitrarily small, violating the KSS bound. \\ If this phenomenon also carries over to the unitary Fermi gases, it may be possible to measure these small viscosities in experiments with trapped ultra-cold Fermi gases. For this purpose, it is helpful to consider traps which share the essential features of the systems in Ref.~\cite{Jain:2014vka,Jain:2015txa} listed at the end of Sec.~\ref{gravity} of this paper. The goal of this paper is to give a concrete proposal for the trap geometry and parameters where this effect is likely to be seen. While typical trap potentials are harmonic, [quadratic (Eq.~\ref{eq:harmonic_potential}) rather than linear in the distance] by using existing results for the thermodynamics of unitary Fermi gases, we show that for a range of temperatures the dominant contribution to the damping of collective modes due to viscosity arises from a narrow region in the trap not near the center, where the trapping potential can be approximately considered as linear. In analogy with Ref.~\cite{Jain:2014vka,Jain:2015txa} it is desirable to have traps that are highly anisotropic, which can be simulated by taking the trapping frequencies~\cite{Ketterle:2008} in one of the directions (say $\omega_z$) to be much larger than the frequencies in the other directions. \\ We describe two hydrodynamic modes whose dissipation is governed by the components of viscosity which are expected to become small in the anisotropic situation considered here. One of them is known in the literature as the scissor mode which has been well studied for bosonic superfluids at $T=0$ theoretically~\cite{PhysRevLett.83.4452} and has also been experimentally excited in both bosonic~\cite{PhysRevLett.84.2056} and fermionic~\cite{PhysRevLett.99.150403} superfluids. The second mode is a new quasi-stationary solution to the hydrodynamic equations. Especially for the scissor mode, we show that for experimentally reasonable values of trap parameters, the damping rate of the mode lies within an experimentally accessible range. It should therefore be possible to study this mode, measure the relevant component of the viscosity and its possible suppression. To gain some additional understanding of how the anisotropic system might behave, we also make a rough estimate of the viscosity components in the presence of an anisotropic trapping potential using the Boltzmann equation. We find that as the anisotropy increases, due to an increase in the trapping frequency $\omega_z$ in one of the directions, some components of the viscosity tensor decrease, compared to their value in the isotropic case. The outline for the paper is as follows. We review the relevant results~\cite{Jain:2014vka,Jain:2015txa} for anisotropic theories with gravitational dual in Sec.~\ref{gravity} and summarize the essential features required in a system to exhibit the suppression of $\eta/s$. Further details on the gravity results is also provided in Appendix \ref{grdetails}. Next, we consider the unitary Fermi gas in an anisotropic harmonic trapping potential and describe the two hydrodynamic modes which couple to the small components of the shear viscosity tensor in Sec.~\ref{vprofile}. In Appendix~\ref{sci} and \ref{stat} we show that these two hydrodynamic modes satisfy the equations of superfluid hydrodynamics. Sec.~\ref{edisp} discusses the energy dissipation due to shear viscosity in these two modes we have studied. In Sec.~\ref{validity} we examine the constraints on the mode amplitudes by demanding validity of fluid mechanics and in Sec.~\ref{oc} we discuss the damping in the outer regions of the cloud. Next we review the thermodynamics of the system in Sec.\ref{thermo}. In Sec.~\ref{trap} we give parameter values for traps (the trapping potential, the temperature and the chemical potential at the center of the trap) which are tuned such that the system possesses the required essential features, and show that by measuring the damping rate of fluid modes (described in Sec.~\ref{vprofile}) one can measure the shear viscosity. This section contains some of the key results in the paper. Sec.~\ref{local} discusses an analysis in a weakly coupled anisotropic theory using the Boltzmann equation. We conclude our discussion in Sec.~\ref{res}. The solution of the Boltzmann equation used to estimate the values of the trap potentials for which we expect the corrections to the viscosity to be substantial is given in Appendix~\ref{microscopic}. In Appendix~\ref{modes} we compare the modes (discussed in Sec.~\ref{vprofile}) with the well known breathing modes. \section{Results of shear viscosity from gravity} \label{gravity} We briefly review results of computations of shear viscosity in the gravity picture obtained by studying anisotropic blackbranes~\cite{Jain:2014vka} where the breaking of isotropy is due to an externally applied force which is translationally invariant. The simplest system discussed in Ref.~\cite{Jain:2014vka} consists of a massless dilaton minimally coupled to gravity, and a cosmological constant. The action is \begin{equation} S = \frac{1}{16\pi G} \int d^5 x \sqrt{g}~ [R+12\Lambda - \frac{1}{2}\partial_\mu \phi\partial^\mu\phi]\;,~\label{eq:5dlag} \end{equation} where $G$ is Newton's constant in $5$ dimensions and $\Lambda$ is a cosmological constant. The dual field theory in the absence of anisotropy is a $3+1$ dimensional conformal field theory. The dilaton profile, linear in the spatial co-ordinate $z$ \begin{equation} \label{anisoparam} \phi=\rho z\;, \end{equation} explicitly breaks the symmetry to $2+1$.\\ The conservation equation for the stress tensor gets modified as \begin{equation} \partial_\mu T^{\mu\nu}=\langle O \rangle \partial^\nu \phi~\label{eq:hydrodynamics}\;, \end{equation} where $O$ is the operator dual to the field $\phi$. The right hand side arises because the varying dilaton results in a driving force on the system. We see that a linear profile results in a constant value for $\partial^\nu \phi$ and thus a constant driving force. Using AdS/CFT one finds~\cite{Jain:2014vka} that for a system at temperature $T$, (using the compact notation $\eta_{ijij}=\eta_{ij}$) $\eta_{xz}=\eta_{yz}$ (which are spin 1 with respect to the surviving Lorentz symmetry) is affected by the background dilaton. In the low anisotropy regime ($\rho/T \ll 1$): \begin{equation} \frac{\eta_{xz}}{s}= \frac{1}{4\pi}-\frac{\rho^2 \log 2}{16 \pi^3 T^2} + \frac{(6-\pi^2+54 (\log 2)^2)\rho^4}{2304\pi^5 T^4} + {\cal{O}}\bigg[\bigg(\frac{\rho}{T}\bigg)^6\bigg]~\label{eq:eta_low}\;. \end{equation} The correction to the zero anisotropy result, the KSS bound $\frac{\eta_{xz}}{s}=\frac{1}{4\pi}$, is proportional to $\frac{(\nabla \phi)^2}{T^2}$ where $\nabla \phi=\rho\hat{z}$ is the driving force and $1/T$ is the microscopic length scale in the system. \\ In extreme anisotropy ($\rho/T \gg 1$), \begin{equation} {\eta_{xz}}/{s}\rightarrow ({1}/{4\pi}) ({32\pi^2 T^2}/{3\rho^2})\, \end{equation} and hence becomes parametrically small~\cite{Jain:2014vka}. But this domain will not be physically accessible in the cold atom systems.\\ In contrast the $\eta_{xy}$ component (which couples to a spin $2$ metric perturbation) was found to be unchanged from its value in the isotropic case, $\frac{\eta_{xy}} {s}= \frac{1}{4\pi}$.\\ Parametric reduction of the spin $1$ components of $\eta/s$ has been found for a variety of strongly coupled theories with a gravitational dual~\cite{Jain:2015txa,Rebhan:2011vd}. Motivated by the generality of the above results, (see \cite{ Jain:2015txa}) in the gravity side, we may hope to find parametrically suppressed viscosities compared to the KSS bound in systems where the following basic requirements are met. \label{cond} \begin{enumerate} \item{The system is strongly interacting and in the absence of anisotropy have a viscosity close to the KSS bound.} \item{The equations of hydrodynamics for the system admits modes sensitive to the spin one viscosity components as described above and in Ref.~\cite{Jain:2014vka,Jain:2015txa}.} \item{Sufficient anisotropy needs to be introduced in the system (say in the $z$ direction with rotational symmetry preserved along the $x-y$ plane), such that these spin one components of the viscosity, when measured in units of the entropy density, show an experimentally measurable decreasing tendency from its lowest value observed so far in ultracold Fermi gases.} \item{The force responsible for breaking of isotropy is approximately spatially constant.} \item{The velocity gradients are small enough (compared to say the inverse mean free path) ensuring that hydrodynamics is the appropriate effective theory to describe the system.} \end{enumerate} In the next section (Sec.~\ref{Fermi}) we explore a system of trapped ultra-cold Fermi gases, chosen so as to explore anisotropic fluid dynamics. While some of the details of this system are different from the systems with dual gravitational theories discussed above, it is possible to choose a set of parameters such that the system has the five features listed above. It can therefore be used to explore the behavior of the viscosity in the anisotropic regime. While gravitational duals for the ultra-cold Fermi gases are not yet known and hence we can not calculate the anisotropic viscosity coefficients in this strongly coupled system, if the main feature that $\eta_{xz}$ is smaller than the KSS bound holds true for these, one could potentially measure this phenomenon in experiments. \section{Anisotropic viscosity in trapped anisotropic Fermi gases} \label{Fermi} Trapped ultra-cold Fermi gas with their scattering length tuned to be near the unitarity limit~\cite{Stringari:2008,Ketterle:2008}, are strongly interacting systems for which $\eta/s$~\cite{Schafer:2007pr,Cao58,Thomas:2015}, was measured to be close to the KSS bound $1/(4\pi)$. In this section we shall explore the properties of this system, when it is placed in an anisotropic trap. We identify suitable hydrodynamic modes which probe the viscosity component expected to be suppressed due to the potential in a highly anisotropic harmonic trap and find that for reasonable choices of parameters the five criterion referred to above, (see Sec.\ref{cond}), can be met in these modes. This leads us to suggest that an anisotropic shear viscosity can arise in such systems and appropriate components of the viscosity may show a reduction from the isotropic values in an experimentally accessible way. One method~\cite{Thomas:2015} to measure the viscosity is by starting with an initial state where the fluid is trapped in an anisotropic harmonic trap. On removing the trapping potential, the fluid experiences elliptic flow and the extent of the flow is related to the initial anisotropy and the viscosity. The relevant bulk viscosity of the system vanishes~\cite{Son:2007,Taylor:2010}, which allows one to cleanly extract the shear viscosity. Note that even though the initial state of the fluid is anisotropic, the experiment does not probe anisotropic shear viscosities: after the trap potential is removed, the viscosity tensor at any point is isotropic. An alternative technique is to measure the damping rate of breathing modes~\cite{Schafer:2007pr,Cao58} which is related to the loss of energy due to the viscosity. The experiments we propose in this paper use this alternative technique and propose to measure the relevant component of the shear viscosity by measuring the damping of appropriate hydrodynamic modes. \begin{figure} \begin{tabular}{cc} \includegraphics[width=8.5cm]{rotation.png} &\includegraphics[width=8.5cm]{1flow.png}\\ \end{tabular} \caption{(Arbitrary units for coordinates) The flow profile in the $x-z$ plane for {the Elliptic mode}, ie. ${\bf{v}}=z~\hat{x}-x~\hat{z}$ (left panel, corresponding to $\omega_x/\omega_z=1$ in Eq.~\ref{modea}) and ${\bf{v}}=z~\hat{x}-0.001~x~\hat{z}$ (right panel, corresponding to $\omega_x/\omega_z=0.03$ in Eq.~\ref{modea}).} \label{flow} \end{figure} The unitary Fermi gas system we consider here shares important features with the gravitational system described in Sec.~\ref{gravity}. The role of a linear potential was emphasized in Sec.~\ref{gravity}. While such a linear potential cannot arise in the trapped fermion system we consider, we shall see below that if we choose the velocity profile and the trap parameters carefully, the dominant contribution to shear viscosity comes from a region of the trap where the confining force is approximately constant: satisfying the fourth criterion listed in Sec.~\ref{gravity}. The system we consider consists of an ultra-cold Fermi gas under harmonic confinement described by the potential \begin{equation} \phi({\bf{r}}) = \sum_i \frac{1}{2} m \omega_{i}^{2} x_{i}^{2} ~\label{eq:harmonic_potential} \end{equation} where $i$ runs over $x,y,z$ and $m$ denotes the mass of the fermionic species. The trap is anisotropic if $\omega_i$'s are unequal. For example, $\omega_z\gg\omega_x, \omega_y$ gives rise to a pancake like trap: thin in the $z$ direction. This can lead to an anisotropic shear viscosity tensor as described in Sec.~\ref{local}. The potential gradient in the $x$ and $y$ directions is small in most of the trap. This section is organized as follows. After a general discussion we describe the two modes of interest (referred to as the Elliptic mode and the Scissor mode) in subsection \ref{vprofile}. The equations of superfluid hydrodynamics are described in Appendix \ref{superfl}, following which, in Appendix \ref{sci} and \ref{stat} respectively we show that the Scissor mode and the Elliptic mode satisfy these equations. The fluid flow profile in the Elliptic mode is similar to that shown in Fig.~\ref{plates}: a velocity in the $x$ direction with a gradient in the $z$ direction. The scissor mode is well known in the literature. In subsection \ref{edisp} we show that the dissipation of energy in the two modes of interest is determined by the relevant components of the viscosity tensor (the spin $1$ components described in the previous section). In Subsection \ref{validity} we find a constraint on the magnitude of the velocity for the two modes by demanding the validity of fluid mechanics. The thermodynamics of the system is discussed in subsection \ref{thermo}. Finally in subsection \ref{trap} we bring this understanding together and show that for reasonable values of parameters the required criterion listed in Sec.~\ref{cond} can indeed be met. \begin{figure} \includegraphics[width=8.5cm]{scissor.png} \caption{(Arbitrary units for coordinates) The flow profile in the $x-z$ plane at time $t=0$ for the Scissor mode, ie. $v=z~\hat{x}+x~\hat{z}$ (Eq.~\ref{scissor})} \label{scimode} \end{figure} \subsection{Choice of Velocity Profile} \label{vprofile} Here we first describe the two modes of interest which arise as solutions to the equations of ideal superfluid hydrodynamics. Each of these modes is characterized by the superfluid and the normal components, which we denote by $\textbf{v}_s$ and $\textbf{v}_n$ respectively.\\ The first mode, which we call the Elliptic mode has $\textbf{v}_s=0$ and $\textbf{v}_n=\textbf{v}$ given by \begin{equation} \label{eq:vprofile} {\bf{v}} ~=e^{i \omega t} (\alpha_x z ~\hat{x} + \alpha_z x ~\hat{z}) \end{equation} with the following relations: \begin{equation} \label{modea} {\textbf{Elliptic mode}}:\;\;\omega=0,~ \alpha_{z}=- \frac{\omega_{x}^{2}}{\omega_{z}^{2}} \alpha_{x} \end{equation} The other mode of interest, denoted by the Scissor mode, has $\textbf{v}_s=\textbf{v}_n=\textbf{v}$ given by Eq.~\ref{eq:vprofile} with \begin{equation} \label{scissor} {\textbf{Scissor mode}}:\;\; \omega=\sqrt{\omega_x^{2}+\omega_z^{2}},~ \alpha_{z}= \alpha_{x}. \end{equation} From the right panel in Fig.~\ref{flow} we see that in the high anisotropy limit $\omega_{z}\gg\omega_{x}$, $\alpha_{z}\to 0$ for the Elliptic mode, and hence we recover a flow profile similar to that considered in~\cite{Jain:2014vka} (shown in Fig.~\ref{plates}); namely a time independent (in the limit of small viscosity) velocity (${\bf{v}}\propto z\hat{x}$) linearly increasing with the coordinate in the direction of the gradient of the external potential ($z$), pointing ($\hat{x}$) in the direction perpendicular to the gradient of the external potential (neglecting $\omega_x$, $\omega_y$. The gradient is in the $\hat{z}$ direction). To the best of our knowledge, the Elliptic mode has not been studied in ultra-cold gas experiments. The scissors mode which has been studied extensively (for example see Refs.~\cite{PhysRevLett.83.4452,PhysRevLett.84.2056, PhysRevLett.99.150403}). \subsection{Energy dissipation due to viscosity} \label{edisp} The energy dissipated due to viscosity is given by \begin{equation} \begin{split} \dot{E}_{{\rm{kinetic}}} =& -\frac{1}{2} \int d^3{{\bf{r}}}\, \eta_{ijij}({{\bf{r}}})\, \left(\partial_iv_j+\partial_jv_i-\frac{2}{3}\delta_{ij} \partial_k v_k \right)^2 -\int d^3{{\bf{r}}} \, \zeta({{\bf{r}}})\, \big( \partial_iv_i\big)^2 ~\label{eq:dissipation} \end{split} \end{equation} where $\eta_{ijij} \equiv \eta_{ij}$ is the relevant component of the shear viscosity and $\zeta$ is the bulk viscosity. We note that for our chosen velocity profiles, the bulk viscosity contribution vanishes. Also in the traps we will consider, the temperature T is constant throughout the trap. Hence we also ignored contributions from thermal conductivity. Thus, \begin{equation} \begin{split} \dot{E}_{{\rm{kinetic}}} &= - \int d^3{{\bf{r}}}\, \eta_{xz}({{\bf{r}}}) ~\alpha_{x}^{2} (1- \frac{\omega_{x}^{2}}{\omega_{z}^{2}})^2 \label{etaexp} \end{split} \end{equation} is the energy dissipation rate for the Elliptic mode, where we have simply written $\eta_{xzxz}$ as $\eta_{xz}$. The energy dissipated per unit cycle for the oscillatory time dependent scissor mode is \begin{equation} \begin{split} \dot{E}_{{\rm{kinetic}}} &= -2\int d^3{\bf{r}}\, \eta_{xz}({\bf{r}}) ~\alpha_{x}^{2} . \label{etaosc} \end{split} \end{equation} \subsection{Validity of hydrodynamics} \label{validity} One expects that hydrodynamics is a valid description of the system as long as the viscous correction to the stress tensor is small compared to its value in an ideal fluid (for eg. see Ref.~\cite{LandauPhysical} or Sec.~$10.3.4$ in Ref.~\cite{zwergerbook}). For {the Elliptic mode} the contribution to the stress energy tensor from viscosity is \begin{equation} \eta_{xz}\frac{1}{2}(\alpha_x+\alpha_z)\approx\eta_{xz}\frac{1}{2}(\alpha_x) \end{equation} where we have assumed $\omega_{z}\gg \omega_{x,\;y}$ and neglected the contribution from $\alpha_z$ (see Eq.~\ref{modea}). For {the Scissor mode} the magnitude of the contribution to the stress energy tensor from viscosity is \begin{equation} \eta_{xz}\frac{1}{2}(\alpha_x+\alpha_z)=\eta_{xz}(\alpha_x) \end{equation} where we have $\alpha_z=\alpha_x$ for {the Scissor mode}. At any point ${\bf{r}}$, hydrodynamics is expected to be valid if the viscosity contribution is smaller than the pressure $P({\bf{r}})$, \begin{equation} \alpha_{x} \eta_{xz}({{\bf{r}}}) \ll P ({{\bf{r}}}) ~\label{eq:hydro_condition1}\;. \end{equation} In the outer edges of the trap the pressure becomes small while $\eta$ tends to a constant~\cite{PhysRevA.72.043605,Bruun:2006kj,Bluhm:2015raa,PhysRevA.71.033607} and Eq.~\ref{eq:hydro_condition1} is necessarily violated regardless of how small $\alpha_x$ is chosen. The contribution of this region to the total energy loss is typically small however. (Note that the expression Eq.~\ref{eq:dissipation} can not be used to evaluate the energy loss if Eq.~\ref{eq:hydro_condition1} is not satisfied~\cite{Bluhm:2015raa}.) What we desire is that hydrodynamics should be a good theory in the region where the energy loss is substantial. When we consider specific numerical values for the parameters of the trap in Subsection \ref{trap}, we will identify a point $\bf {r}_{max}$ close to the edge of the trap, such that the integral Eq.~\ref{eq:dissipation} receives most of its contribution for $r<\bf{r}_{max}$. We can then define $\alpha_{x}^{\rm{max}} $ by the condition that for this amplitude the viscosity contribution to the stress energy tensor is equal to the pressure at the point ${\bf{r}}_{{\rm{max}}}$ \begin{equation} \alpha_{x}^{\rm{max}} = \frac{P ({{\bf{r}}_{{\rm{max}}}})}{\eta_{xz}({{\bf{r}}_{{\rm{max}}}})}. ~\label{eq:hydro_condition}\; \end{equation} For $\alpha_{x}<\alpha_{x}^{\rm{max}}$ hydrodynamics is valid in the region of interest. This constraint limits how large $\alpha_x$ and consequently $\dot{E}_{{\rm{kinetic}}}$ can be. As long as this dominates over other processes of energy loss (interaction with the environment) this damping can be measured. In Table.~\ref{ptrap2} in Sec.~\ref{trap} we show this numerical limit for the traps described in that Section.\\ \subsection{ The outer core} \label{oc} It has been noted that a naive application of hydrodynamics at the outer region of the trap where the density of the atoms is very low leads to an unphysical result. Since the shear viscosity in the ultra-dilute regime has the form $\eta\sim (mkT)^{3/2}$, ($m$ is the mass, $k$ is the Boltzmann's constant and $T$ is the temperature) the contribution from the tail (or the outer cloud) is independent of the density, and hence is divergent [\cite{Bruun:2006kj,Bluhm:2015raa,PhysRevA.72.043605, PhysRevA.76.045602,PhysRevA.94.043644, PhysRevA.78.053609,PhysRevLett.116.115301,PhysRevA.71.033607}]. The unphysical result arises because in the outer part of the trap collisions are rare and hydrodynamics breaks down. In fact the better approximation in this region is assuming that atom dynamics in this ultra-dilute region is collisionless and hence does not contribute significantly to damping. Here we use a simple procedure to take this physics into account. We only consider traps where the chemical potential at the center is positive and cutoff the damping contribution from the outer cloud by integrating the viscosity contribution only from the center of the trap up to ${\bf{r}}_{\rm{max}}$ which is defined as the surface where $\mu-V({\bf{r}}_{\rm{max}})=T$. We have checked that changing $r_{max}$ by a little (for example by choosing a slightly larger $r_{max}^0$ by using the condition $\mu-V(r_{max}^0)=0$) gives similar results for the damping rates. Similar prescriptions have been followed previously by \cite{Cao58,Thomas:2015} (see~\cite{PhysRevLett.116.115301} for an overview). One can also perform a more careful estimate of the contribution from the outer cloud. To be concrete, let us consider the scissor mode. We follow the procedure described in Ref.~\cite{PhysRevA.76.045602} which solves the Boltzmann equation in the dilute regime, rather than assuming that hydrodynamics is accurate in this region. Their important result is that for the scissor mode~\footnote{ Let us also note that the scissor mode is excited in the $x-y$ plane in Ref.~\cite{PhysRevA.76.045602}. We have taken care of this fact in our calculations and comparisons.} the energy loss rate in the dilute regime can be written as the integral over $\eta$ divided by a suppression factor that increases exponentially as a function of the trapping potential. More precisely, \begin{equation} \langle\dot{E}_{\rm kinetic}\rangle|_{{\rm{oc}}}= -2\alpha^2\int_{{\bf{r}}>{\bf{r}}_{\rm{max}}} d^3{ {\bf{r}}} \frac{\eta}{1+\omega^2\tau_\eta^2({\bf r})}, \label{edotoc} \end{equation} where in the dilute regime (or the ``classical limit'') \begin{equation} \tau_{\eta}({\bf r})= \frac{4.17}{N\bar{\omega}}\left(\frac{kT}{\hbar\bar{\omega}}\right)^{2} e^{V({\bf r})/kT}\;, \end{equation} and the viscosity $\eta$ is given by \begin{equation} \eta=\frac{15}{32\sqrt{\pi}}\frac{(mkT)^{3/2}}{\hbar^2}\;. \end{equation} The scissor mode frequency is given by, \begin{equation} \omega =\sqrt{\omega_x^2 +\omega_z^2}\;,\label{omega} \end{equation} and the geometric mean $\bar {\omega} = ( \omega_x \omega_y \omega_z)^{1 \over 3}$. The integral Eq.~\ref{edotoc} is convergent because of the exponential increase in the relaxation time $\tau_{\eta}({\bf{r}})$ even if we take the upper limit of the integral to $\infty$ but for the numerical evaluation we take the upper limit of the $x$-integration to be $x_{\rm{max}}+L$, for the $y$-integration to be $y_{\rm{max}}+L$, and $z$-integration to be $z_{\rm{max}}+L$ with $L\gg |{\bf{r}}_{{\rm{max}}}|$. At the core of the trap hydrodynamics is a good approximation (unless $T\ll T_c$ where the superfluid phonons can move out of equilibrium). This is a crucial point because Boltzmann transport is not a valid approximation at the core where the density of atoms is high. As we explained in the last section, as long as $\alpha_x=\alpha_z=\alpha<\alpha_x^{\rm{max}}$, hydrodynamics is a good approximation and the local contribution from the viscosity to the stress energy tensor \begin{equation} \begin{split} \alpha ~ \eta ({\bf{r}}) \end{split} \end{equation} is smaller than the pressure \begin{equation} \begin{split} P({\bf{r}}) \end{split} \end{equation} for ${\bf{r}}<{\bf{r}}_{{\rm{max}}}$. Therefore, using hydrodynamics to evaluate the damping contribution from the core, we get \begin{equation} \langle\dot{E}_{\rm kinetic}\rangle|_{{\rm{c}}}= -2 \alpha^2\int_{{\bf{r}}<{\bf{r}}_{\rm{max}}} d^3{ {\bf{r}}}~{\eta({\bf{r}})}\;, \label{edotc} \end{equation} where the local value of $\eta({\bf{r}})$ is calculated using the data for $\eta$ from \cite{Thomas:2015}. The integration is performed over $x<x_{\rm{max}}$, $y<y_{\rm{max}}$ and $z<z_{\rm{max}}$. This approximates the actual ellipsoidal region with a rectangular shape, but we see that this will not change the results substantially since the contribution from the outer cloud is small. \begin{table} \begin{center} \begin{tabular}{|c|c|c|c|} \hline $T$ & $\Gamma_{{\rm{c}}}$ ($s^{-1}$) & $\Gamma_{{\rm{oc}}}$ ($s^{-1}$)\\ \hline $4 T_c/5$ &$23.03$ &$0.0044$ \\ $2 T_c/3$ &$18.32$ &$0.00009$ \\ $4 T_c/7$ &$14.6$ &$2.14\times 10^{-6}$\\ $T_c/2$ &$11.86$ &$4.69\times10^{-8}$ \\ \hline \end{tabular} \end{center} \caption{Comparison of contributions to the damping rates for the scissor mode from the core [$\Gamma$(c) Eq.~\ref{gammac}], and the outer core [$\Gamma$(oc) Eq.~\ref{gammaoc}] for the trap parameters we will explore in our paper.} \label{table1} \end{table} The amplitude decay rate is given by \begin{equation} \Gamma = \frac{|\langle\dot{E}_{\rm kinetic}\rangle|}{2\langle E\rangle} ~\label{eq:defGamma} \end{equation} $\langle E\rangle$ is the total mechanical energy averaged over a cycle, \begin{equation} \begin{split} \langle E\rangle&=\frac{1}{2} \int d^3{r}mn({\bf r}) |v|^2({\bf r})\\ &=\frac{1}{2}m\alpha^2 \int d^3{r}mn({\bf r})(z^2 +x^2)\;, \label{eng} \end{split} \end{equation} where $v=\alpha e^{i\sqrt{\omega_x^2+\omega_z^2}t}(z\hat{x}+x\hat{z})$. In Eq.~\ref{gammac}, $\alpha^2$ cancels out and we only need $n({\bf{r}})$ which is obtained from experiments as explained in detail in Sec.\ref{thermo}. The damping rate contribution from the outer cloud is given by \begin{equation} \begin{split} \Gamma_{{\rm{oc}}}&= \frac{|\langle\dot{E}_{\rm kinetic}\rangle|_{{\rm{oc}}}}{2\langle E\rangle} \label{gammaoc} \end{split} \end{equation} and the contribution from the core is given by \begin{equation} \begin{split} \Gamma_{{\rm{c}}}&= \frac{|\langle\dot{E}_{\rm kinetic}\rangle|_{{\rm{c}}}}{2\langle E\rangle}\;, \label{gammac} \end{split} \end{equation} and the total damping rate Eq.~\ref{eq:defGamma} is the sum of the two. In Table.~\ref{table1}, for the representative trap parameters which we will be considering later ( $\omega_z =2 \pi \times 10^4$ rads/s, $\omega_x=\omega_y= 2\pi \times 385$~rads/s and $\mu=10\mu$K and $T/T_c$ values as given in the table), we present the comparison of the contribution to damping from the outer cloud and the core in Table.~\ref{table1}. We see that the damping contribution from the outer cloud is small, especially for the low temperatures, justifying our approach. A direct comparison using our technique (where we cut off the integral for $\dot{E}_{\rm{kinetic}}$ at the point of the trap where hydrodynamics breaks down) can only be made for the lowest temperature ($T/T_F=0.1$) of Ref.~\cite{PhysRevLett.99.150403}. Our calculations (using the trap parameters of \cite{PhysRevLett.99.150403})give a damping rate of 250 $s^{-1}$ which agrees with experiments (255 $\pm$ 40 $s^{-1}$, \cite{PhysRevLett.99.150403}). This is a non-trivial check of our methodology and gives us confidence in our approach in this regime.\\ \begin{figure} \begin{tabular}{cc} \includegraphics[width=8.5cm]{sovernfg.png}&\includegraphics[width=8.5cm]{muvst.png}\\ \end{tabular} \caption{(Color online) Data of $\frac{S}{N}$ as a function of $T/T_F$ (left panel) and $\mu/E_F$ versus $T/T_F$ (right panel) from Ref.~\cite{Ku563}. The central curves (blue online) correspond to the central values and the band gives an error estimate (Ref.~\cite{Ku563}). The band denoted by the dashed vertical lines corresponds to the phase transition between the normal and the superfluid phase. The error bands represent the maximum error chosen from a set of representative error bars given in Ref.~\cite{Ku563}. } \label{sovern} \end{figure} \begin{figure} \begin{tabular}{cc} \includegraphics[width=8.5cm]{GSchafer.png}& \includegraphics[width=8.5cm]{GprimeSchafer.png}\\ \end{tabular} \begin{center} \includegraphics[width=8.5cm]{ffSchafer.png} \end{center} \caption{(Color online) The thermodynamic function ${\cal{G}}$ (top left panel) and its derivative (top right panel) as a function of $T\over \mu$. The lower panel shows ${\cal{F}}$. These dimensionless functions are defined in Eq.~\ref{nands}. The error bands follow from the error bands in Fig.~\ref{sovern}.} \label{gfun} \end{figure} \subsection{Thermodynamics} \label{thermo} The evaluation of the energy loss from Eq.~\ref{etaexp} and Eq.~\ref{etaosc} requires the viscosity $\eta$ as a function of the position ${{\bf{r}}}$ in the trap. In the highly anisotropic traps we are considering the viscosity is actually a tensor and the different components of the shear viscosity can acquire different values, in contrast with the isotropic case. For the modes of interest, Eq.~\ref{eq:vprofile} we need to determine the behavior of the component ($\eta_{xz}$). To get a first estimate of the region of the trap which gives a dominant contribution to the integral in Eq.~\ref{eq:dissipation}, we use the local density approximation (LDA) and estimate the resulting viscosity. More specifically, we assume in this approximation that thermodynamic variables like the number density $n$, the entropy density $s$ depend only on the local value of $T$ and $\mu$. The viscosity is also then taken to be given by these local values of $T,\mu$, neglecting any effects of anisotropy which could make the different components of the tensor take different values. The effect of anisotropy on the viscosity tensor are estimated using Eq.~\ref{eq:anisotropy_corrections}, in a following section (Sec.~\ref{local}). While we cannot reliably compute them, the key point of our analysis here is that they may be experimentally measured and could lie below the KSS bound. To apply the LDA approximation mentioned above, we start first by considering a homogeneous system characterized by temperature $T,\;\mu$ and review the behavior of the thermodynamical parameters and the viscosity as a function of these parameters. This is covered in this subsection. In the presence of the trap $\mu$ varies in the equilibrium configuration. The effects of the trap, in this approximation, are then incorporated by using the resulting local value for $\mu$ and $T$ in the behavior for the homogeneous case. The next subsection will then incorporate the effects of the trap. In certain thermodynamic regimes, the viscosity of a uniform unitary Fermi gas can be computed in a controlled manner. At temperatures much smaller than the chemical potential, transport is dominated by the Goldstone mode associated with superfluidity and the viscosity can be computed by solving the Boltzmann transport equations~\cite{PhysRevA.76.053607}. At temperatures large compared to the chemical potential, the density of fermions is small and a kinetic estimate of the viscosity, $\eta={\rm{const.}}\times(mT)^{3/2}$, is adequate~\cite{PhysRevA.72.043605,Bruun:2006kj,Bluhm:2015raa}. But we shall see that the largest contribution to damping arises from the regime where $T$ and $\mu$ are comparable, and a theoretical evaluation of the viscosity is difficult. Monte Carlo~\cite{Wlazlowski:2012jb,Wlazlowski:2015yga} methods, microscopic approaches~\cite{Guo:2010dc}, and $T-$matrix techniques~\cite{Enss:2010qh} have been used to calculate the viscosity in this regime but presently the best estimate for the viscosity in this intermediate regime comes from experiments. \begin{figure} \begin{center} \includegraphics[width=0.45\textwidth]{etan.png} \includegraphics[width=0.45\textwidth]{etas.png} \caption{(Color online) The left panel shows $\eta \over n$ versus $T/T_F$ from Figs. $2$ and $3$ of~\cite{Thomas:2015}. The right panel shows $\eta \over s$ versus $T/T_F$ from Fig. $5$ of~\cite{Thomas:2015}.} \label{etaovern} \end{center} \end{figure} In Refs.~\cite{Schafer:2007pr,Cao58}, $\eta/s$ was measured for the first time. Recently, this measurement was refined in Ref.~\cite{Thomas:2015} and the result for the dimensionless ratio $\eta/n$ was measured for a wide range of $T/\mu$, which we show in Fig.~\ref{etaovern}. Therefore, to obtain the LDA value of the viscosity, we just need $n(\mu, T)$. In the next few paragraphs we describe how to obtain $n(\mu,T)$ using the scaling properties of the unitary Fermi gas. With that understanding at hand we will then return to a discussion of how to obtain the viscosity in the approximation described above. In the unitary Fermi gas, the chemical potential $\mu$ and the temperature $T$ are the only energy scales in the problem. Therefore, we can express various thermodynamic quantities as a function of the dimensionless quantity $y=T/\mu$ multiplied by an appropriate dimensionful function of only one of the two variables. Following~\cite{Schafer:2007pr} we write, \begin{equation} \begin{split} n(\mu,\;T)=&~ n_{f}(\mu)\mathcal{F}(y), \\ s(\mu,\;T)=&~ \frac{2}{5}n_{f}(\mu)\mathcal{G}^{\prime}(y)~\label{nands}, \end{split} \end{equation} where $n$ is the number density, $s$ is the entropy density, and $\mathcal{F}(y)=\mathcal{G}(y)-2~y~\mathcal{G}^{\prime}(y)/5$, $n_{f}(\mu)={1\over3 \pi^2}(2 m \mu)^{3 \over 2}$ is the number density of a free Fermi gas. Therefore one can compute the desired thermodynamic quantities if the function $\mathcal{G}(y)$ is known. For example, one can write the pressure as \begin{equation} \begin{split} P(\mu,\;T)=&\frac{2}{5} \mu~n_{f}(\mu)~\mathcal{G}(y). \label{prs} \end{split} \end{equation} In the following discussion, we use the usual definitions \begin{equation} k_F = (3\pi^2 n)^{1/3},\; E_F = \frac{k_F^2}{2m},\;T_F = E_F/k_B,\;v_F=\frac{k_F}{m}\;. \end{equation} At low temperatures (${T\over T_{F}} \lesssim 0.6$) we use the $S \over N$ data from Fig.~$3(b)$ of Ref.~\cite{Ku563} to obtain $\mathcal{G}(y)$. Data from two graphs obtained from Ref.~\cite{Ku563} are shown here in the two panels of Fig.~\ref{sovern} for convenience. The left panel shows $S/N=s/n$ as a function of $T/T_F$ and the right panel shows $\mu/E_F$ as a function of $T/T_F$. In order to solve Eq.~\ref{nands} we need to get $S \over N$ as a function of $y$. We use Fig. $3(a)$ of Ref.~\cite{Ku563} to convert the $S \over N$ data in terms of $y={T\over \mu}$ rather than $T\over T_{F}$. We obtain the function $\mathcal{G}(y)$ by numerically solving Eq.~\ref{nands}, subject to the boundary condition $\mathcal{G}(0)=1/\xi^{3/2}$ at $T=0$. We use $\xi=0.376\pm 0.0075$. (The value of $\xi$ quoted here is from~\cite{Ku563}. Various theoretical calculations can be found in~\cite{Gezerlis:2007fs,Bulgac:2007,Forbes:2010gt,PhysRevB.49.12975, Goulko:2015rsa, PhysRevLett.96.160402, PhysRevLett.96.090404}.) Fig.~\ref{gfun} shows the numerically extracted function $\mathcal{G}$ , its first derivative and the function $\mathcal{F}$. In Fig.~\ref{gfun} and the rest of the figures, the band denoted by the dashed vertical lines corresponds to the phase transition between the normal and the superfluid phase. The data in Ref.~\cite{Ku563} stops at $T/T_F\approx 0.6$. For higher temperatures the density is small and as far as thermodynamics is concerned, we can model the system as a gas of weakly interacting fermions with a self energy correction in the chemical potential associated with self interactions in the normal phase. Therefore $n$ and $s$ have the same form as in a Fermi gas, (Ref.~\cite{PhysRevA.75.023610}) \begin{equation} \begin{split} n_{\rm{norm}} &= - g~ (m T)^{3 \over 2}\frac{{\rm{PolyLog}} \left(\frac{3}{2},-e^{\mu\over T }\right)}{2 \sqrt{2} \pi ^{3/2}}\\ s_{\rm{norm}} &= \frac{\sqrt{T} \left(2~\mu~ {\rm{PolyLog}} \left(\frac{3}{2},-e^{\mu \over T }\right)-5~T ~{\rm{PolyLog}}\left(\frac{5}{2},-e^{\mu \over T }\right)\right)}{2 \sqrt{2}\pi ^{3/2}} \;,~\label{eq:highTthermo} \end{split} \end{equation} where $n_{\rm{norm}}$, $s_{\rm{norm}}$ denote the number density and entropy in the normal phase, $g=2$ is the energy level degeneracy, and $\mu$ with self energy corrections is replaced by $\mu-\frac{3^{2/3} n^{2/3} \pi ^{4/3} (\xi _n -1)}{2 m}$. Fitting to high temperature data gives $\xi_n \approx 0.45$~\cite{Ku563}. This description works well all the way down to temperatures $T/T_F\gtrsim0.5$ or equivalently ${T \over \mu} \gtrsim 3.2$ as one can check by comparing the values of $S/N$ as a function of $T/T_F$ in this approximation with the results from~\cite{PhysRevA.75.023610}. These results match smoothly to the low temperature measurements in Ref.~\cite{Ku563}. Therefore for ${T \over \mu} >3.2$ we use Eq.~\ref{eq:highTthermo} to compute the thermodynamics. Now that we have understood how to obtain $n(T,\mu)$ we can return to our discussion of the viscosity. To evaluate $\eta$ at a given $\mu$ and $T$ we simply multiply $\eta \over n$ from Fig.~$3$ of Ref.~\cite{Thomas:2015} (shown here in the left panel of Fig.~\ref{etaovern}) with the number density that can be found using Eq.~\ref{nands}. One could alternatively multiply $\eta \over s$ from Fig. $5$ of Ref.~\cite{Thomas:2015} (shown here in the right panel of Fig.~\ref{etaovern}) with the entropy that can be found using Eq.~\ref{nands}. The former works better because of the smaller error bars. As we shall see in the next section when we describe the fermions in a trap, the dominant contribution to the energy loss arises from the region in the trap where $T/\mu$ is about $0.54$. This is just above the critical temperature $T_c$ given by the relation \begin{equation} T_c/T_F = 0.167\pm 0.013\;,~\label{T_cbyT_F} \end{equation} or equivalently \begin{equation} {T_{c} \over \mu }= 0.4\pm 0.03\;.~\label{T_cbymu} \end{equation} From the right panel of Fig.~\ref{etaovern} we see that just above ${T_{c} \over \mu }\approx 0.4$, $\eta/s\approx 0.7 \approx 8(\frac{1}{4\pi})$. This fact will be relevant in the next section. \begin{figure} \begin{center} \includegraphics[width=0.45\textwidth]{num.png} \includegraphics[width=0.45\textwidth]{entropy.png} \caption{(Color online) Variation of number density (left panel) and the entropy density (right panel) with respect to $z$ for $T={2 T_{c} \over 3}$ at $\omega_z=2\pi\times10^4$ rads/s with chemical potential at the trap center 10$\mu$K. The vertical lines denote the band in $z$ where $T=(0.4\pm 0.03)(\mu-\phi(z))$ (Eq.~\ref{T_cbymu}). } \label{num} \end{center} \end{figure} \subsection{Results for the trap} \label{trap} Having understood the thermodynamics in the absence of the trap, we now turn to incorporating the trap potential in the discussion. We first use the LDA approximation to calculate how thermodynamic quantities like $s, n$ etc. vary along the trap. It turns out that on starting at the center of the trap at a sufficiently low temperature, the entropy density has a peak, $z_0$, close to the point where the superfluid-normal transition occurs. In turn, this leads to the viscosity and damping effects for the fluid modes of interest receiving their contribution from a region close to the peak and with a width, $\delta z$ that can be made narrow, $\delta z/z_0<1$. Finally, in this subsection we examine the resulting behavior of the system for a range of reasonable values of parameters and show that the five conditions listed at the end of Section \ref{cond} can be met. It turns out that both the time scales for energy loss, and the magnitude of the total energy, lie in the range of experimentally accessible values. Before we start let us note that there are three energy scales, $T,\mu, \omega_z$ in the system ($\mu$ without an argument refers to the chemical potential at the center of the trap, and we are neglecting $\omega_x, \omega_y$ here). These give rise to two dimensionless ratios, $T/\mu, \omega_z/\mu$. Length scales can be obtained from these energy scales using the mass, via the relation, $L = {1\over \sqrt{2m E}}$. {\it Thermodynamics in the Trap: } As discussed in Subsection \ref{superfl} in the presence of a trap the equations for superfluid dynamics can be solved at equilibrium by taking the chemical potential to have a local value which varies along the trap, as given by \footnote{From now on $\mu$ without the argument ${\bf{r}}$ refers to the chemical potential at the center of the trap and $\mu({\bf{r}})=\mu - \phi({\bf{r}})$.} Eq.~\ref{muans}. The temperature $T$ in equilibrium is a constant. Once we have the function $\mathcal{G}$ as discussed in Sec.~\ref{thermo}, one can then use LDA to express all quantities of interest as a function of the displacement from the trap center (which we denote by ${{\bf{r}}}$). Thus, within LDA, we can write the number density as \begin{equation} \label{lda} n({{\bf{r}}}) = n\left(\mu({\bf{r}}),\;T\right). \end{equation} We can also express energy and entropy density in the same fashion as a function of the distance from the trap center. Some comments on the conditions for the violation of LDA will be made in the end of the section. To set the scales we show (see Fig.~\ref{num}) the number density and the entropy density as a function of the distance $z$ from the trap center at $x=0, y=0$, for a typical trap configuration that we consider. In all the examples we consider, we will take Li$_{6}$ as the fermionic species. In making Fig.~\ref{num}, the chemical potential at the center of the trap is chosen to be $10\mu$K which is typical for experiments performed with fermionic cold atoms~\cite{Cao58,Caoth}. The potential is taken to be harmonic (Eq.~\ref{eq:harmonic_potential}), with the confinement frequency along $z$ direction, $\omega_z= 2 \pi \times 10^4$ rads/s which is about 10 times that chosen in Ref.~\cite{Caoth}.~\footnote{For conversions to energy units, we use $1$ eV$^{-1}=1.97 \times 10^{-7}$ m, $1$ eV$=1.78 \times 10^{-36}$ kg, $1$ eV$^{-1}=6.58 \times 10^{-16 }$ s, $1$ eV$= 1.16 \times 10^{4}$ K. The mass of Li$_6$ in natural units is $5.6\times 10^{9}$ eV.} Since we are taking $x=y=0$, $\omega_x$ and $\omega_y$ do not matter in drawing Fig.~\ref{num}. However, since we will be exploring anisotropic traps we keep in mind the condition that $\omega_x=\omega_y\ll\omega_z$. The temperature throughout the trap is taken to be $T=\frac{2 T_{c}}{3}$, where $T_{c}$ is the critical temperature (Eq.~\ref{T_cbymu}) associated with the chemical potential ($\mu$) at the center of the trap defined by \begin{equation} \label{tcdef} T_c \equiv 0.4~\mu \;. \end{equation} To avoid confusion we note that $T_c$ is the temperature at which the superfluid to normal phase transition would have occurred at the center of the trap. In the system under consideration with $T=\frac{2T_c}{3}$, since $T$ at the center of the trap is below the local critical temperature at the center of the trap, the transition actually occurs away from the center of the trap, at a location $z=z_c$, where the local chemical potential $\mu(z_c)=\frac{T }{(0.4)}$ [where we have abbreviated $\mu((0,0,z_c))$ as $\mu(z_c)$] corresponding to the phase transition to the normal phase. In Fig.~\ref{num} we have denoted it by dashed (gray online) vertical lines corresponding to the central value and the error bands. The error bands to the densities (marked by red curves online surrounding the blue central curve) are associated with the errors in ${\cal{G}}$ (Fig.~\ref{gfun}). They are discontinued from $z=27.5 \times 10^{-5}$ cm corresponding to the point where we switch to Eq.~\ref{eq:highTthermo} to calculate the thermodynamics. In the other trap geometries we consider below, we will keep the chemical potential at the center, $\mu$, unchanged as it will set the overall scale of the problem, and only change the temperature of the trap and the confining frequency $\omega_z$, in order to explore traps which satisfy criteria listed in Sec.~\ref{gravity}. The strategy we follow is given below. As explained in the last section, we estimate the $\eta$ at a given location ${{\bf{r}}}$ corresponding to the local chemical potential $\mu({\bf{r}})$ and temperature $T$ by simply multiplying the local number density $n$ we find using Eq.~\ref{nands} with $\eta \over n$ from Fig.~$3$ of Ref.~\cite{Thomas:2015}. (We have reproduced it here in Fig.~\ref{etaovern} for convenience.) This estimate assumes that not only thermodynamic but also the transport quantities are determined by the local chemical potential and the temperature. This estimate necessarily implies that the viscosity is isotropic. Nonetheless this will help us identify the values of $T/\mu$ for which the energy loss of the hydrodynamic shear modes is dominated by a region where the potential can be approximated as a linear potential. Having done that, we will increase $\omega_z$ to induce anisotropy in the transport coefficients. \begin{figure} \begin{tabular}{cc} \includegraphics[width=9cm]{eta1_25.png}&\includegraphics[width=9cm]{eta1_50.png}\\ \includegraphics[width=9cm]{eta1_75.png}&\includegraphics[width=9cm]{eta2.png}\\ \end{tabular} \caption{(Color online) Local shear viscosity with respect to $z$ for $T={4T_{c} \over 5}$ (top left), $T={2T_{c} \over 3}$ (top right) $T={4 T_{c} \over 7}$ (bottom left) and $T={T_{c} \over 2}$ (bottom right) at $\omega_z=2 \pi \times 10^4$ rads/s and $\mu=10\mu$K. The red curves around the central blue curve denote the error estimate which include errors in the measurement of $\eta/n$~\cite{Thomas:2015} as well as errors in ${\cal{G}}$ due to errors in the measurements of thermodynamics~\cite{Ku563}. The black dashed vertical line to the left is at $z_c$ while the one to the right is at $z_{trap}=\sqrt{\frac{2 \mu}{m \omega_z^2}}$. We do not extend the viscosity curves in the dilute regime as discussed in Section \ref {validity}. The dashed orange horizontal line corresponds to $\eta/n_f$ in the $\mu \rightarrow - \infty$ limit (\cite{Bruun:2006kj}).} \label{localvis} \end{figure} Let us consider the four panels in Fig.~\ref{localvis}. They show the local shear viscosity (in units of $(2m\mu)^{3/2}/(3\pi^2)$ where $\mu$ is the central chemical potential) as a function of $z$ for $x=0, y=0$ for four different temperatures at $\omega_z=2 \pi \times 10^4$ rads/s. The chemical potential at the center is taken to be $10$$\mu$K. The temperatures are $T={4T_{c} \over 5}$ (top left panel), $T={2T_{c} \over 3}$ (top right panel) and $T={4 T_{c} \over 7}$ (bottom left panel) and $T={T_{c} \over 2}$ (bottom right panel). Like Fig.~\ref{num}, the vertical line (gray online) corresponds to $z_c$ where $T=0.4\mu(z_c)$. The error bands of the curves are associated with the errors in ${\cal{G}}$ --- which impact $n$ --- as well as the errors in the measured $\eta/n$. The $x$-axes of the plots is the $z$ coordinate scaled by the trap size \begin{equation} z_{\rm{trap}} = \sqrt{\frac{2 \mu}{m \omega_z^2}}\;. \end{equation} One can also define a characteristic distance $z_{{\rm{max}}}$ where $T/\mu(z)=1$ given by \begin{equation} \label{rmax} z_{{\rm{max}}}= \sqrt{\frac{2(\mu-T)}{m \omega_z^2}}\,. \end{equation} For $\mu =10\mu$K at the center of the trap and $\omega_z=2 \pi \times 10^4$ rads/s, $z_{\rm{trap}}$ and $z_{\rm{max}}$ are $\sim 10^{-4}$ cm. Beyond the distance $z_{\rm{trap}}$, we assume the viscosity to behave like $\frac{15}{32 \sqrt {\pi}}(m T)^\frac{3}{2}$ as predicted by the two-body Boltzmann equation~\cite{Bruun:2006kj}. Note that within LDA the plots in Fig.~\ref{localvis} are independent of $\omega_z$ if we keep $T/T_c$ fixed. This is because scaling $\omega_z$ by a factor $f$ can be undone by scaling $z$ by a factor $1/{f}$. Since $z_{\rm{trap}}$ is scaled by the same factor, $z/z_{\rm{trap}}$ at any point on the curve remains unchanged. To understand the behavior of viscosity along the trap, first consider the central values in Fig.~\ref{localvis} (blue curve online). For all temperatures given above (notice that they are all below $T_c$ meaning that the centre of the trap is superfluid), we find the presence of a peak in the middle region of the trap length. Qualitatively we understand this from the fact that the local entropy (see Eq.~\ref{nands}) is the product of $n_f(\mu({\bf{r}}))$ which decreases along the length of the trap, while the function $\mathcal{G}^{\prime}$ increases along the length of the trap, hence it is natural to expect a peak for the entropy density somewhere along the length of the trap. It is clearly seen in the right panel of Fig.~\ref{num}. Since the local shear viscosity over entropy density is relatively slowly varying in this region (the peak location is just above the critical region), it is not surprising that the local shear viscosity shows a similar behavior. Henceforth, we will denote the position of this peak by $z_{0}$. We also denote the full width at half maximum of the peak by $\delta z$. The existence of the peak allows us to construct a system where the dominant contribution comes from a region where the potential approximately varies linearly, modeling the theories (Sec.~\ref{gravity}) where the force that breaks rotational invariance is spatially constant. Here, the trap potential is harmonic, but the dominant contribution to the integral in Eq.~\ref{etaexp} and Eq.~\ref{etaosc} comes from an interval $\delta z$ near $z_0$. If we expand the confinement potential as a Taylor series around $z_{0}$ as \begin{equation} \phi(z_{0}) + \phi^{'}(z_{0}) (\delta z) +{1 \over 2} \phi^{''}(z_{0})(\delta z)^2 + ...... \end{equation} The linearity approximation will hold as long as the confinement potential satisfies \begin{equation} \frac{\phi^{''}(z)}{\phi^{'}(z)} ~\delta z \ll 1 \Rightarrow l \equiv \frac{ \delta z}{z_{0}} \ll 1\;. \label{eq:lin} \end{equation} Since we are using a harmonic trap, there are no higher order terms. Our criterion for constant driving force is therefore straightforward. We desire that the dimensionless ratio $l\equiv{\delta z \over z_{0}}$ be less than $1$. There are other motivations to choose the dominant contribution to shear viscosity to arise from such a localized region. We are interested in extracting the value of $\eta/s$, for suitable components of the viscosity tensor, for particular values of $T, \mu$ (in particular, close to the critical temperature $T_c$ where $\eta/s$ is known to be close to the KSS bound). Due to the varying trap potential, $\mu(z)$ and therefore the entropy density at equilibrium also vary along the trap. The change resulting in the viscosity due to anisotropy should be bigger than the effect due to the variation of the trap potential on $s$, thereby giving rise to the condition, \begin{equation} \label{condaa} {\delta \eta\over \eta}> {\partial s\over \partial z}{\delta z \over s}\,. \end{equation} As we saw in Sec.~\ref{gravity} after Eq.~\ref{eq:eta_low} the corrections to the viscosity due to anisotropy go like square of the force that generates the anisotropy. For the system at hand this leads to the expectation \begin{equation} \frac{\delta \eta}{\eta} \sim \frac{(\nabla \phi )^2 }{\left(\mu(z)^2 k_F(z)^2\right)}.\label{eq:anisotropy_corrections_rough} \end{equation} This estimate agrees with the analysis based on the Boltzmann equation as discussed later in Sec.\ref{local} (see Eq.~\ref{eq:anisotropy_corrections}). The RHS in Eq.~\ref{condaa} goes like ${\partial s\over \partial z}{\delta z \over s}\sim \delta z /z_0=l$, and this gives rise to the condition \begin{equation} \label{conab} \kappa_{{\rm{LDA}}}^2 >l \end{equation} where we have introduced the notation \begin{equation} \label{defkappalda} \kappa_{{\rm{LDA}}}= \frac{(\nabla \phi ) }{\left(\mu(z_0)~ k_F(z_0)\right)} \,. \end{equation} It is easy to see that $\kappa_{{\rm{LDA}}}$ roughly scales as \begin{equation} \label{condac} \kappa_{{\rm{LDA}}}\sim\frac{\omega_z}{ \mu} \end{equation} so that Eq.~\ref{conab} leads to the condition \begin{equation} \label{conddd} {\omega_z^2 \over \mu^2}>l. \end{equation} For fixed $T,\mu$ one can show that $l$ does not change as $\omega_z$ changes. Thus the left hand side is independent of the ratio ${\omega_z \over \mu}$ for fixed $T/\mu$, and the inequality can be met for sufficiently large $\frac{\omega_z}{\mu}$. Let us also mention that the gravity results apply to situations with only linearly varying potential (Eq.~\ref{anisoparam}) leading to only $|\nabla\phi|^2 $ corrections due to the anisotropy. In general we would expect that there are additional corrections proportional to $\nabla^2\phi$. There is little guidance on what these corrections do, for the kind of strongly coupled system we are dealing with here. Thus, to the extent we are trying to stay close to situations where gravitational systems give at least some guidance, it is desirable to choose the dominant contribution to shear viscosity to arise from a narrow localized region. {\it Viscosity and Other Properties For Varying Trap Parameters: Table \ref{ptrap1} }\\ We now turn to examining the behavior of $\eta$, $\eta/s$, and $l=\frac{\delta z}{z_0}$ as trap parameters are varied. In Table \ref{ptrap1} we keep $\omega, \mu$ fixed to take the values $\omega_z =2 \pi \times 10^4$ rads/s, $\mu=10\mu$K and vary $T$. As mentioned at the beginning of Subsection \ref{trap} there are two dimensionless ratios that characterize the energy scales in this system. The different rows corresponding to different values of $T$ in units of $T_c$ show how various quantities vary with $T/\mu$. The scaling of these quantities with $\omega_z/\mu$ is given in the first line on top of the Table.~\ref{ptrap1}. Thus $\kappa_{{\rm{LDA}}}$ scales like $\omega_z/\mu$. $z_0, z_{\rm{trap}}$ and $ \delta z$ scale like $1/\omega_z$ for fixed $T,\mu$, as was discussed above after Eq.~\ref{rmax}. Thus their ratios, ${z_o\over z_{\rm{trap}}}, l= {\delta z \over z_0}$ etc. are independent of $\omega_z/\mu$. The third column of the Table.~\ref{ptrap1} tests the linearity of the potential, which is a good approximation near the peak if $l=\delta z/z_0 \ll 1$. The ratio $l$ is governed by the temperature of the trap divided by the chemical potential or equivalently $T_c$ at the center. As we decrease $T/T_c$, $z_0$ increases and $\delta z$ decreases. This consideration would suggest that to obtain ${\delta z \over z_{0}}$ as small as possible we should consider as small a temperature as possible. But this conclusion is not correct as is clear from the upper error band in Fig.~\ref{localvis} (red online). \begin{table} \begin{center} \begin{tabular}{|c|c|c|c|c|c|c|c|} \hline $T$ & $z_{\rm{trap}}\sqrt{\frac{\mu}{10 \mu {\rm{K}}}}\frac{2 \pi \times 10^4}{\omega}$ cm & $\frac{z_{0}}{z_{\rm{trap}}} $ & $l$ & $\frac{T}{\mu(z)}|_{z_0}$ & $\frac{\eta}{n}|_{z_0}$ & $\frac{\eta}{s}|_{z_0}$ & $\kappa_{{\rm{LDA}}}\frac{10 \mu {\rm{K}}}{\mu}\frac{\omega_z}{2 \pi \times 10^4 rad/s}$\\ \hline $4T_{c}/5$ &$27 \times 10^{-5}$ &0.63 &0.98 &0.54 & 0.89 &0.85 &0.05 \\ $2T_{c}/3$ &$27 \times 10^{-5}$ &0.71 &0.62 &0.54 & 0.89 &0.85 &0.08 \\ $4 T_{c}/7$ &$27 \times 10^{-5}$ &0.76 &0.46 &0.54 & 0.89 &0.85 &0.11 \\ $T_{c}/2$ &$27 \times 10^{-5}$ &0.8 &0.37 &0.55 & 0.91 &0.85 & 0.13\\ \hline \end{tabular} \end{center} \caption{Trap characteristics for various $T/T_c$. The scaling behavior of various quantities with $\omega_z$ are also shown. The entries were calculated for $\mu=10\mu$K, $T_c=0.4\mu$. $l=\frac{\delta z }{z_0}$ (Eq.~\ref{eq:lin}) tests how well the potential can be approximated as a linear potential in the regime of interest. $\kappa_{\rm{LDA}}$ (Eq.~\ref{tf}) tests how well LDA is expected to work at $z_0$.} \label{ptrap1} \end{table} \begin{table} \begin{center} \begin{tabular}{|c|c|c|c|c|c|c|c|} \hline $T$ & $\alpha_x^{\rm{max}}$($10^{-10}$eV) & ${\dot{E}_{{\rm{kinetic}}}}$(j/s)(\textbf{a}) & $E$(j) (\textbf{a}) & $\tau_0(s)$(\textbf{a}) & ${\dot{E}_{{\rm{kinetic}}}}$(j/s)(\textbf{b}) & $E$(j) (\textbf{b}) & $\tau_0(s)$(\textbf{b})\\ \hline $ 4T_{c}/5$ &$2.83$ &$2.37 \times 10^{-16}$ &$3\times 10^{-20}$ &$0.0002$ & $4.7 \times 10^{-16}$ & $10^{-17}$ &0.04 \\ $2T_{c}/3$ &$2.35$ &$1.25 \times 10^{-16}$ &$2\times10^{-20}$ &0.0003 & $2.5\times 10^{-16}$ &6.8 $\times 10^{-18}$ &0.05 \\ $4 T_{c}/7$ &$2.02$ &$7.12 \times 10^{-17}$ &$1.4 \times 10^{-20}$ &0.0004 & $1.4\times 10^{-16}$ &4.8 $\times 10^{-18}$ &0.07 \\ $T_{c}/2$ &$1.77$ &$4.33\times10^{-17}$ & $1.1\times 10^{-20}$ &0.0005 &$8.65\times 10^{-17}$ &3.6 $\times 10^{-18}$ &0.08\\ \hline \end{tabular} \end{center} \caption{Additional trap characteristics for various $T/T_c$ at $\omega_z =2 \pi \times 10^4$ rads/s, $\omega_x=\omega_y= 2\pi \times 385$ rads/s and $\mu=10\mu$K. The energy is given in joules abbreviated as `j' and energy loss rate in joules per second, (j/s). For a fixed $ T/ \mu$, the energy of the Elliptic mode scales as $ \sim \frac{1}{\omega_x \omega_y \omega_z^3} $ and that of the Scissor mode scales as $\sim \frac{1} {\omega_x^3 \omega_y \omega_z}$. The characteristic time $\tau_0$ ( given in seconds `s' in the table and defined in Eq.\ref{taudef}) of the Elliptic mode scales as $ \sim \frac{\mu}{ \omega_z^2} $ and that of the Scissor mode scales as $\sim \frac{\mu} {\omega_x^2 }$. For the Elliptic mode to account for the fact that only the normal component of the velocity is non-zero near the trap centre, we assume that the normal component density in this region is $\frac{T}{T_c}$ times the total density in this region. For the Scissor mode we have the full number density.} \label{ptrap2} \end{table} The errors bands on $\eta$ are fairly narrow in the region near $z_0$. However, the errors grow near $z\rightarrow 0$, in particular for smaller $T/T_c$ (Fig.~\ref{localvis}). The reason is the large errors in the measured $\eta/n$ in the superfluid regime (see the region $T/T_F\lesssim0.16$ in Fig.~\ref{etaovern}). Indeed, we expect that for $T\ll T_F$, the viscosity is dominated by superfluid phonons whose contribution diverges as $T\rightarrow0$ as $\eta\approx (9.3\times 10^{-6})~\xi^5~(T_F^8/v^3T^5)$ where $v$ is the speed of superfluid phonons \cite{PhysRevA.76.053607}. Numerically, $\eta/n\approx 2.5 \times 10^{-5}~ \frac{T_F^5}{T^5}$. Therefore, to avoid a large contribution from the center of the trap rather than from near $z_0$, we do not consider temperatures below $T_c/2$. Within this constrained temperature regime between $T_c/2$ and $T_c$ we find that the linearity condition $\delta z/z_0<1$ is satisfied, although it is not possible to generate traps where $\delta z/z_0$ is parametrically small. In the narrow range of temperatures, it turns out that the location of $z_0$ is such that $T/\mu(z_0)\approx0.54$, just off to the right of the phase transition at $T/\mu(z_c)\approx0.4$. Note that, as explained in the discussion above, a few paragraphs after Eq.~\ref{tcdef}, the value for the viscosity $\eta/s$ which appears in the Table \ref{ptrap1} is an approximate one, obtained by taking the value in the isotropic situation corresponding to the local value for $\mu$, $T$ at the location $z_0$. By a similar argument as before, this value is independent of the ratio $\omega_z/\mu$ for a fixed $T/T_c$. We note that the values of $\eta/s$ in the Table \ref{ptrap1} are about $10$ times the KSS bound. One would expect that various components of the viscosity tensor deviate from this rough value by a fraction of order $\kappa_{{\rm{LDA}}}^2$. The parameter $\kappa_{{\rm{LDA}}}$ which was introduced in Eq.~\ref{defkappalda} above, when computed at the location of the peak $z_0$, has the more exact form \begin{equation} \label{kappaexact} \kappa_{{\rm{LDA}}} =\frac{ m\omega_z^2 z_0}{(3 \pi^2 n(z_0))^{1\over3} \mu(z_0) }= \frac{\sqrt{m \over 2} \omega_z^{2} z_0} {[{\cal{F}}(\frac{T}{\mu(z_0)})]^{1/3}[\mu(z_0)]^{\frac{3}{2}}} \end{equation} as one can easily check by using Eq.~\ref{nands}. {\it Energy Damping For Varying Values of Trap Parameters: Table \ref{ptrap2}} We now turn to considering the effects of varying the trap parameters on various quantities like the total energy $E_{{\rm{kinetic}}}$, the damping rate of this energy $\dot{E}_{{\rm{kinetic}}}$, etc. In Table \ref{ptrap2} we again keep $\mu, \omega_z$ fixed to take values $\omega_z =2 \pi \times 10^4$ rads/s, $\mu=10\mu$K and consider the effects of varying $T$. In addition, we also need to consider the effects of the harmonic trap in the $x,y$ directions. We keep $\omega_x,\omega_y$ to be fixed to take values $\omega_x=\omega_y= 2\pi \times 385$ rads/s. The different rows then give how various quantities vary as $T/\mu$ changes. We note that for the range of temperatures considered the total number of atoms in the trap is approximately, $\sim 10^6$. The energy which appears in this Table is the total mechanical energy $E$ given by \begin{eqnarray}}\def\eeq{\end{eqnarray} E = 2 E_{\rm{kinetic}} \eeq where \begin{equation} \begin{split} {E}_{{\rm{kinetic}}} &= \langle \frac{1}{2} \int d^3{{\bf{r}}}\, m n({{\bf{r}}}) {\bf{v}}^2\rangle\;, \label{energy} \end{split} \end{equation} where ${\bf{v}}$ is the velocity of either mode and the average is taken over one cycle for the scissor mode (the elliptic mode is non-oscillatory). For the Elliptic mode and the Scissor mode with amplitude $\alpha_x^{\max}$, the kinetic energy is given as follows: \begin{equation} \label{ekin} \begin{split} {\rm{For }}\; {\bf{ Elliptic}},\; E_{{\rm{kinetic}}}(\textbf{a}) &= \int d^3{\bf{r}} \; \frac{1}{2}m\;n_{normal}\;(\alpha_x^{max})^2[\frac{\omega_x^4}{\omega_z^4}x^2+z^2]\\ {\rm{For }}\; {\bf{Scissor}},\; E_{{\rm{kinetic}}}(\textbf{b}) &= \int d^3{\bf{r}} \; \frac{1}{4}m\;n\;(\alpha_x^{max})^2[x^2+z^2]\,. \end{split} \end{equation} $\dot{E}_{{\rm{kinetic}}}$ is the rate of energy loss due to viscosity induced dissipation, Eq.~\ref{eq:dissipation}. The energy loss, $\dot{E}_{{\rm{kinetic}}}$ in these modes is given by Eqns.~\ref{etaexp}, \ref{etaosc}. Note that for the Scissor mode the expression corresponds to the kinetic energy averaged over an oscillation cycle. Also, for the Elliptic mode, $v_s=0$, Eq.~\ref{modea}, and only the normal component contributes to the kinetic energy. The density in the normal phase is estimated in the region close to the centre, where both the superfluid and normal components are present, as being $\frac{T}{T_c}$ times the total density in this region and we have denoted it by $n_{normal}$ in Eq.~\ref{ekin}. For the Scissor mode we have the full number density denoted by $n$ in the above formulas. The validity of hydrodynamics imposes a condition on how big $\alpha_x$ can become, the resulting maximum value, $\alpha_x^{max}$ was estimated in Eq.~\ref{eq:hydro_condition}. The quantities $E_{{\rm{kinetic}}}$, $\dot{E}_{{\rm{kinetic}}}$ which appear in Table \ref{ptrap2} are obtained from Eq.~\ref{eq:dissipation}, Eq.~\ref{ekin} by setting $\alpha_x=\alpha_x^{max}$. A convenient quantity with which to compare $\alpha_x^{\rm{max}}$ is the ratio of the speed of sound at the centre $c_s=\sqrt{\frac{2\mu}{3m}}$ to a measure of the trap size $z_{\rm{trap}}$. For comparison, let us note that for $\omega_z=2 \pi \times 10^4$ rads/s we obtain $\frac{c_s}{z_{\rm{trap}}}=\frac{\omega_z}{\sqrt{3}}=3.63\times 10^{-11}$ eV. The (amplitude) damping time $\tau_0$, which appears in Table \ref{ptrap2}, is defined as \begin{equation} \label{taudef} \tau_0 = {2E/\dot{E}_{kinetic}} \end{equation} As mentioned above, the table considers the effects of varying the temperature while keeping $\mu, \omega_z, \omega_x,\omega_y$ fixed. For fixed $T/ \mu$ one can also consider what happens as the angular frequencies are varied. In the highly anisotropic situations $\omega_z\gg \omega_x, \omega_y$, one finds that the total energy $E_{{\rm{kinetic}}}$ for the Elliptic mode approximately scales like \begin{equation} \label{ea} E_{{\rm{kinetic}}}(a) \sim \mu \frac{\mu}{\omega_x}\frac{\mu}{\omega_y}\left(\frac{\mu}{\omega_z}\right)^3 \end{equation} and the damping time $\tau_0$ for the Elliptic mode approximately scales like \begin{equation} \label{ta} \tau_0(a)\sim \frac{\mu}{\omega_z^2}\,. \end{equation} Similarly for the Scissor mode we get \begin{equation} \label{eb} E_{{\rm{kinetic}}}(b) \sim \mu \frac{\mu}{\omega_y}\frac{\mu}{\omega_z}\left(\frac{\mu}{\omega_x}\right)^3, \end{equation} \begin{equation} \label{tb} \tau_0(b) \sim \frac{\mu }{\omega_x^2} \,. \end{equation} These scalings are obtained by noting that $\alpha_x^{\max}\sim \mu $ for fixed $T/\mu$, and also that the trap potential is unchanged under a rescaling $\omega_z \rightarrow \lambda~ \omega_z, z \rightarrow z/\lambda$ and similarly for $x,y$. We have also assumed that $\omega_z\gg \omega_x, \omega_y$. Some of these scalings are summarized in the caption below Table \ref{ptrap2}. For example, the scalings of the scissor mode, can be derived as follows: $ E \sim \int dx dy dz [m n v^2] \sim L_x L_y L_z [m n \alpha^2 L_x^2] \sim \frac{\mu^6} {\omega_x^3 \omega_y \omega_z} $, where we have assumed that at the center of the trap $\mu>0$ and $L_i=\sqrt{2\mu/(m\omega_i^2)}$.) In a similar manner, one can derive the approximate scalings for energy dissipation rates: $\dot{E} \sim \frac{\mu^5}{\omega_x \omega_y \omega_z}$ for both the modes (assuming $\eta$ scales the same way as $n$ ie. $\sim ( m \mu)^{3 \over 2}$. The approximate value of $T, \mu, \omega_z$ we consider here are of the same order as those considered in~\cite{Cao58} where the viscosity of a unitary Fermi gas was measured, using a radial breathing mode. The Scissor mode has been considered in the literature before. The damping rate has been measured for cold atoms system in this mode in superfluid bosonic (see Ref.~\cite{PhysRevLett.84.2056} and Refs. therein) and in fermionic systems~\cite{PhysRevLett.99.150403}. In particular \cite{PhysRevLett.99.150403} carries out these measurements in the unitary Fermi gas. The values for trap parameters we consider are similar to those considered for example in~\cite{Cao58} and not very different from those considered in~\cite{PhysRevLett.99.150403}. The maximum angular amplitude of the the scissor mode is determined by the velocity amplitude $\alpha_x$ (Eqs.~\ref{scissor},~\ref{eq:vprofile}) which is bounded above by $\alpha_x^{\rm{max}}$ in Table \ref{ptrap2}. One can show that the angular amplitude (in radians) of the oscillation executed by the deformed cloud in the scissor mode is given by \begin{equation} \begin{split} \theta = \tan^{-1}\left(\frac{e^{\frac{2\alpha_x }{\omega} }-1}{e^{\frac{2\alpha_x }{\omega}}+1}\right)\;, \end{split} \end{equation} where $\omega=\sqrt{\omega_x^2+\omega_z^2}$. Taking $\alpha_x$ to be the maximum value $\alpha_x^{\rm{max}}\sim 10^{-10}$ eV and $\omega$ to be $2\pi\times10^{4}$ rads/s $\equiv4.16\times10^{-11}$ eV, we find $\theta_{\rm{max}}\sim \tan^{-1}[1]$ $\equiv45^\circ$. For a frequency $10$ times larger, $\theta_{\rm{max}}\sim \tan^{-1}[0.4]$ $\equiv24^\circ$. It is satisfying that these amplitudes are larger than those measured in~\cite{PhysRevLett.99.150403} for the scissor mode and hence the condition for hydrodynamics (Eq.~\ref{eq:hydro_condition}) does not force the amplitudes to be so small as to preclude observation using existing techniques. For $\mu=10\mu K$, $\omega_x=\omega_y=2\pi\times 385$ rads$/$s and $\omega_z=2\pi\times 10^4$ rads$/$s, $\tau_0$ ranges from roughly $0.04$ sec to $0.08$ sec. The damping of the scissor mode has been observed for slightly different parameters values, $\mu\approx 1\mu K$, $\omega_x=2\pi\times 830$ Hz, $\omega_y=2 \pi \times 415$ Hz and $\omega_z=2\pi \times 22$ Hz in Ref.~\cite{PhysRevLett.99.150403} where the damping time scales measured are of the order of milliseconds.\\ {\it Summary: } Now we come to the punch line of this section. The effects of anisotropy can cause a fractional change in components of the viscosity tensor, potentially lowering some of them. This effect is expected to go like, $\delta \eta/\eta\sim \kappa_{{\rm{LDA}}}^2$, as mentioned in Eq.~\ref{eq:anisotropy_corrections_rough}. We see from Table \ref{ptrap1} that, for fixed $\omega_z/\mu$, $\kappa_{{\rm{LDA}}}$ increases as $T$ decreases (i.e. $T/\mu$ decreases), with the maximum value, within the range of allowed temperatures, being of order $\kappa_{{\rm{LDA}}}\sim 10\%$. This would lead, one expects, to a fractional change in components of the viscosity of order $\delta \eta /\eta \sim (\rm{few}) \times 1 \%$, which is quite small. However note that increasing $\omega_z$ will increase $\kappa_{{\rm{LDA}}}$ with a linear dependence $\kappa_{{\rm{LDA}}}\sim \omega_z/\mu$ as noted in Eq.~\ref{condac} and also in the first row of Table \ref{ptrap1}. In turn this should lead to a quadratic fractional change in $\delta \eta /\eta\sim ({\omega_z\over \mu})^2$ . We can carry out this change while keeping $\omega_x, \omega_y$ fixed thereby increasing the anisotropy. Note that this change of $\omega_z$ will decrease the total energy of this mode $E_{{\rm{kinetic}}}(b)\sim 1/\omega_z$, Eq.\ref{eb}, but it does not change $\tau_0$ significantly, since $\tau_0$ depends to a good approximation on $\omega_x$ and not $\omega_z$ as seen from Eq.~\ref{tb}. Also note that changing $\omega_z$ while keeping $T/\mu$ fixed will not change $l$ and thus the localized nature of the region from which the damping arise. In fact it will make it easier to meet the condition Eq.~\ref{conddd}. Also it is worth commenting that it is easy to see from Eq.~\ref{condac}, Eq.~\ref{eb} and Eq.~\ref{tb} that if one want to keep $\tau_0$ and $E_{{\rm{kinetic}}}$ for the scissor mode both fixed and increase $\kappa_{{\rm{LDA}}} \rightarrow \lambda~ \kappa_{{\rm{LDA}}}$ one could do this (while keeping $\omega_x=\omega_y$) by scaling \begin{equation} \label{scal} \omega_x \rightarrow \lambda ^{\frac{1}{6}}~ \omega_x , ~\omega_y \rightarrow \lambda ^{\frac{1}{6}}~ \omega_y , ~ \omega_z \rightarrow \lambda ^{\frac{4}{3}}~ \omega_z,~ \mu \rightarrow \lambda ^{\frac{1}{3}} ~\mu,~T\rightarrow \lambda ^{\frac{1}{3}} ~T. \end{equation} This keeps $\frac{T}{\mu}$, $\tau_0$ and $E_{{\rm{kinetic}}}$ fixed, increases the overall magnitude of $\mu$, increases $\omega_z$ and also $\omega_x,\omega_y$. The discussion of the previous two paragraphs suggests that one can quite plausibly keep the damping time scale and the total energy in the experimentally accessible range, while gradually increasing $\omega_z$ making $\kappa_{{\rm{LDA}}}\sim \mathcal{O}(1)$ and the effects of anisotropy significant. While some of the theoretical approximations made will break down in this limit it is possible that the effects of anisotropy would get more pronounced, and potentially even dramatic, driving the spin one components of the viscosity to be much smaller than their values in the isotropic case, and potentially even violating the KSS bound. We have not discussed the Elliptic mode in as much detail. One reason is that unlike the scissor mode, this mode has not been experimentally realized in cold atom systems yet.\footnote{One possible way to set up the elliptic mode is to start with a more circular trap and exciting a rotational mode by using rotating lasers using a set up similar to Ref.~\cite{2005Natur.435.1047Z}. If the rotational frequency is small enough, vortices will not be excited and only the normal fluid will rotate like a rigid body. On adiabatically deforming the trap one would then get the elliptic mode because during adiabatic deformations, hydrodynamics is satisfied at each time and we expect that the normal fluid will go smoothly from circular rotation to the elliptic mode. } Also we see from Table \ref{ptrap2} that the damping time $\tau_0$ in this case is about two orders of magnitude smaller, and this too might be an issue of some experimental concern. It may of course turn out that this mode is experimentally accessible. It will then be certainly interesting to explore its properties, especially since this mode in a very direct way measures the resistance to shear in the resulting fluid flow. Finally we note that all the five conditions which were listed at the end of Sec.~\ref{gravity} for observing the suppression of viscosity can be met in the system being analyzed here. Conditions $1$ and $2$ are met by the two modes discussed above in the unitary Fermi gas. We have ensured that $l <1$ (Table \ref{ptrap1}) so that the contribution arises from a localized region where the potential is approximately linear, meeting condition $4$. As argued above, for the scissors mode the anisotropy can be made large enough while staying within the fluid mechanics approximation ($\alpha_x<\alpha_x^{max}$) thereby meeting conditions $3$ and $5$. The resulting values for the total energy and the damping time we find lie within the experimentally accessible range. To summarize, we have seen in this section that for experimentally reasonable values of parameters one can increase the anisotropy of the trapping potential and probe the viscosity tensor by measuring the energy loss and related damping time in the scissor mode. As the anisotropy is increased, its effects could well become quite significant driving some components of the viscosity (spin $1$ in our notation) to become very small, and potentially making them even smaller than the KSS bound. \subsection{ Discussion on $\kappa_{{\rm{LDA}}}$} \label{ kappalda} In this subsection, we present a detailed discussion on $\kappa_{{\rm{LDA}}}$ given in the last column of Table.~\ref{ptrap1}. The results discussed so far assume LDA is valid. LDA rests on the assumption that the trap potential varies slowly on the scale of the local Fermi wavelength $k_{F}^{-1}({{\bf{r}}}) = \left(3 \pi^2 n({{\bf{r}}})\right)^{\frac{1}{3}}$ ie. at any local point ${{\bf{r}}}$ along the length of the trap, the following condition holds true - \begin{equation} \begin{split} \nonumber \left|{\nabla_{{{\bf{r}}}}(\mu({\bf{r}}))} {1 \over k_{F}({{\bf{r}}})}\right|_{{{\bf{r}}}} \ll \mu({\bf{r}}) \label{tfgen} \end{split} \end{equation} Since we desire $\omega_x, \omega_y\ll \omega_z$, the gradient is strongest in the $z$ direction and hence taking $x,\;y=0$ and moving along the harmonic trap in the $z$ direction, $\frac{d (\mu(z))}{dz} = -m \omega_z^{2}{{z}}$, we note that LDA violations will be significant if \begin{equation} \begin{split} &m\omega_z^2 z\frac{1}{(3 \pi^2 n(z))^{1\over3}}\sim \mu(z)\;. \label{tf2} \end{split} \end{equation} For any trap geometry at the outer edges of the trap when the density becomes small enough, LDA will be violated ($\mu(z)<0$ for $z>z_{\rm{trap}}$). These regions typically do not contribute significantly to the trap energy loss. But focusing on the region near $z_0$, LDA is a good approximation if \begin{equation} \begin{split} \kappa_{\rm{LDA}} = \frac{\sqrt{m \over 2} \omega_z^{2} z_0} {[{\cal{F}}(\frac{T}{\mu(z_0)})]^{1/3}[\mu(z_0)]^{\frac{3}{2}}} \ll 1\;, \label{tf0} \end{split} \end{equation} Approximating $ {\cal{F}}(\frac{T}{\mu(z_0)})]^{1/3}\approx \frac{1}{\sqrt{\xi}}$ [Since ${\cal{F}}(0)=1/\xi^{3/2}$, and the deviations from ${\cal{F}}(0)$ are small for $T/\mu\lesssim 1$], we find \begin{equation} \begin{split} \kappa_{\rm{LDA}} =\frac{\sqrt{m \over 2} \omega_z^{2} z_0} {[\mu(z_0)]^{\frac{3}{2}}}\sqrt{\xi} \ll 1\;, \label{tf} \end{split} \end{equation} Since $z_0$ scales as $1/\omega_z$ for fixed $\mu$ and $T$, LDA will be violated at $z_0$ if $\omega_z$ is large enough. From Table~\ref{ptrap1} one can see that for $\mu=10\mu$K and $T=T_c/2$, $\kappa_{\rm{LDA}}>1$ for $\omega_z> 2 \pi \times 77000$ rads/s. Alternatively, taking $\omega_z=2\pi\times10^4$ rads/s and $T=T_c/2$, $\kappa_{\rm{LDA}}$ can become larger than $1$ if $\mu < 1.3 ~\mu$K. For $T\rightarrow0$ the corrections to LDA have been previously studied in Refs.~\cite{Son:2005rv,Forbes:2012yp}. One can write \begin{equation} n({\bf{r}}) = n_{\rm{LDA}}\bigl(1 -\frac{c_\chi}{64} \frac{({\bf{\nabla}} \phi({\bf{r}}))^2+4(\mu-\phi({\bf{r}})){\bf{\nabla}}^2\phi({\bf{r}})} {m(\mu-\phi({\bf{r}}))^3} +{{\cal{O}}}{({\bf{\nabla}}^3\phi({\bf{r}}))} \bigr)\;, \label{eq:LDAviolation0} \end{equation} where $c_\chi$ is related to the response of the density to a periodic fluctuation in the potential. The low energy constant $c_\chi$ has not been calculated using {\it{ab-initio}} techniques so far. In all model calculations $c_\chi\sim 1$, including in a sophisticated analysis using SLDA (Ref.~\cite{Forbes:2012yp}). For finite $T$ for an isothermal system, the deviations from LDA are not related to the density response but for $T\lesssim (\mu-\phi({\bf{r}}))$ we can write corrections to LDA in analogy with Eq.~\ref{eq:LDAviolation0} \begin{equation} n({\bf{r}}) = n_{\rm{LDA}}\bigl(1 -\frac{c_1}{64} \frac{({\bf{\nabla}} \phi({\bf{r}}))^2} {m(\mu-\phi({\bf{r}}))^3} -\frac{c_2}{16} \frac{{\bf{\nabla}}^2\phi({\bf{r}})} {m(\mu-\phi({\bf{r}}))^2} + {\cal{O}}((\nabla V)^3) \bigr)\;, \label{eq:LDAviolation} \end{equation} where $c_{1,\;2}$ are functions of $(T/\mu)$ and tend to $1$ as $T/\mu\rightarrow 0$. In particular, for the interesting region the term proportional to $c_1$ is dominant (the exception is near the center of the trap). Therefore, the corrections to LDA near $z_0$ can be written as \begin{equation} n(z) = n_{\rm{LDA}}\bigl(1 -\frac{c_1}{64}\frac{2}{\xi} \kappa^2_{\rm{LDA}} +\cdot\cdot\bigr)\;, \label{eq:LDAviolationvskappa} \end{equation} where we have used the low temperature expression \begin{equation} m\mu({\bf{r}}) = \frac{\xi}{2} k_F^2({\bf{r}})\;, \end{equation} to write the correction in terms of $\kappa_{\rm{LDA}}$. In the absence of further information about $c_1$ at finite $T$ it is difficult to make precise statements about the relevance of LDA corrections for the traps with large values of $\omega_z$ that we show in the next Section are needed to make the shear viscosity tensor locally anisotropic. Therefore, we simply use $\kappa_{\rm{LDA}}\gtrsim 1$ as a marker for significant LDA violation. However, it is important to keep in mind that if $c_1(\frac{T}{\mu(z_0)})\sim c_1(0.54) \sim1$ (since $\frac{T}{\mu(z_0)}\sim 0.54$ for the cases we consider), then the pre-factor of $1/(32\xi)$ implies that the corrections to LDA can be small even for $\kappa_{\rm{LDA}}\approx 1$. \section{Local anisotropy} \label{local} Hydrodynamics is an effective theory: The conserved currents are written as a series of terms ordered by the number of derivatives acting on the local fluid velocity. The lowest order terms are simply given by the Galilean (for non-relativistic systems) or Lorentz (for relativistic systems) transforms of the local thermodynamic properties like the density and the pressure, from the local rest frame of the fluid to the laboratory frame. The first order terms are given by the local gradients of the velocity $(\partial_i u_j +\partial_j u_i)/2$ multiplied by proportionality constants given by the transport coefficients --- for example viscosities --- of the system. We will not consider higher derivative terms in this paper, instead restricting ourselves to situations (see Eq.~\ref{eq:hydro_condition1}) where the first order correction is smaller than the lowest order terms. In the presence of external fields, the law of conservation of energy features a source term proportional to the driving force, $\nabla \phi({\bf{r}})$. If $\nabla \phi({\bf{r}})$ is ``small'' (which we shall define in a moment), its effect on the thermodynamics and transport can be neglected, and hydrodynamics describes a locally isotropic fluid (with isotropic thermodynamic functions and isotropic transport coefficients)~\footnote{This assumes that microscopically the fluid is isotropic. For example it is not a crystal~\cite{LandauPhysical} or a fluid phase with an anisotropic order parameter.} moving in a space dependent potential. The key realization therefore is that to observe an anisotropy in thermal or transport properties it is not sufficient for $\omega_x, \omega_y\ll \omega_z$. Corrections to isotropy will start becoming significant as we increase $\omega_z$, if $\omega_z$ starts becoming comparable to some microscopic scale of the system. The criterion for the thermodynamic quantities to exhibit the effect of $\nabla \phi({\bf{r}})$ is clear from the previous section. If the potential varies on length scales comparable to the inter-particle separation --- the Thomas-Fermi approximation, or LDA breaks down --- the pressure of the fluid in the direction of the gradient will be different from the pressure in the perpendicular directions. In this case, clearly the transport coefficients will also be anisotropic. To explore an analogous system to the one described in Sec.~\ref{gravity}, this argument prompts us to consider $\omega_z$ large enough that LDA is broken (see Table~\ref{ptrap1}). For such systems, the estimates for the density Fig.~\ref{num} and viscosities Fig.~\ref{localvis} using LDA will be only rough guiding values, but if the analogy with the system in Sec.~\ref{gravity} holds true, the viscosity values relevant for the modes described in Sec.~\ref{vprofile} will be lower than the LDA values, and could be lower than $1/(4\pi)$ in suitable quantum units. To estimate the order of the correction to the shear viscosity due to potential gradients we note that the first order correction to transport due to $\nabla \phi({\bf{r}})$ simply appear as the source term, and hence assuming that the next order corrections will be analytic in $\nabla \phi({\bf{r}})$, we expect \begin{equation} \eta_{ijkl} = \eta \frac{1}{2} [(\delta_{ik}\delta_{jl}+ \delta_{il}\delta_{jk}- \frac{2}{3}\delta_{ij}\delta_{kl}) + \bigl(\frac{\lambda^2(\nabla \phi({\bf{r}}))(\nabla \phi({\bf{r}}))}{[\mu({\bf{r}})]^2}\bigr)\sum_{\alpha=0}^{4} c_{(\alpha)}M_{\alpha\,ijkl}]+{\cal{O}}(\nabla^2\phi, (\nabla\phi)^4)\;, ~\label{eq:anisotropy_corrections0} \end{equation} where $\lambda$ is a microscopic length scale of the system, $c_{(\alpha)}$ are dimensional constants of order $1$ which depend on the microscopic details of the system, and $M_i$ are $5$ orthonormal projection operators that arise in a system with one special direction (for eg. see Ref.~\cite{Goossens2010}). We have given these projection operators in Appendix.~\ref{microscopic} (Eq.~\ref{eq:projections}). $\lambda$ is a length scale that determines transport behavior. In a system admitting a quasi-particle description we expect $\lambda$ to be of the order of the mean free path. (We show this explicitly in Appendix.~\ref{microscopic}.) The other length scale in the system is the inter-particle separation $1/k_F$. In terms of $k_F$ we can write the corrections as \begin{equation} \begin{split} \eta_{ijkl} &\approx \eta \frac{1}{2} [(\delta_{ik}\delta_{jl}+ \delta_{il}\delta_{jk}- \frac{2}{3}\delta_{ij}\delta_{kl}) + (\lambda k_F)^2\bigl(\frac{(\nabla \phi({\bf{r}}))(\nabla \phi({\bf{r}}))}{k_F^2[\mu({\bf{r}})]^2}\bigr)\sum_{\alpha=0}^{4} c_{(\alpha)}M_{\alpha\,ijkl}]\\ &= \eta \frac{1}{2} [(\delta_{ik}\delta_{jl}+ \delta_{il}\delta_{jk}- \frac{2}{3}\delta_{ij}\delta_{kl}) + (\lambda k_F)^2\bigl(\kappa^2_{\rm{LDA}}\bigr)\sum_{\alpha=0}^{4} c_{(\alpha)}M_{\alpha\,ijkl}]\;, ~\label{eq:anisotropy_corrections} \end{split} \end{equation} For weakly interacting quasi-particles, the $\lambda k_F\gg1$. But for a strongly interacting system in the absence of more information about $\lambda k_F$ and $c_{(\alpha)}$ it is not possible to make a more concrete statement about the corrections to viscosity. We can only state that the corrections are important if $\kappa_{\rm{LDA}}\sim1$ as we did in Eq.~\ref{eq:anisotropy_corrections_rough}. As discussed in Sec.~\ref{gravity}, for the theories considered in Sec.~\ref{gravity}, there is no quasi-particle description. The only relevant length scale is $1/T$ and the field $\phi$ changes by order $1$ on a length scale $1/\rho$. Using AdS/CFT it has been shown~\cite{Jain:2015txa} that the corrections to isotropy go as Eq.~\ref{eq:eta_low}. For the unitary Fermi gas there is no known gravitational dual~\cite{Bekaert:2011cu} and we will need to resort to a rough calculation to estimate $c_{(\alpha)}$ and $\lambda k_F$. We solve the Boltzmann transport equation in the relaxation time approximation. We hope this will give semi-quantitative results. We leave the challenging calculation of the viscosity for temperatures in the strongly coupled regime just above the critical temperature in the presence of a background potential for future work. As we show in Appendix.~\ref{microscopic}, the corrections to $\eta$ for a weakly interacting, normal (unpaired) Fermi gas at low temperatures ($T<\mu$) are given by (Eq.~\ref{eq:etaoflambda}) \begin{equation} \begin{split} \eta_0 &= \eta(0)[1-\frac{31}{84}(\lambda k_F)^2 \frac{(\E)^2}{k_F^2\mu^2}+{\cal{O}}((\tau\E)^4)] =\eta(0)[1-\frac{31}{84}(\lambda k_F)^2\kappa_{\rm{LDA}}^2+{\cal{O}}((\tau\E)^4)]\\ \eta_1 &= \eta(0)[1-\frac{13}{28}(\lambda k_F)^2 \frac{(\E)^2}{k_F^2\mu^2}+{\cal{O}}((\tau\E)^4)] =\eta(0)[1-\frac{13}{28}(\lambda k_F)^2\kappa_{\rm{LDA}}^2+{\cal{O}}((\tau\E)^4)]\\ \eta_2 &= \eta(0)[1-\frac{11}{28}(\lambda k_F)^2 \frac{(\E)^2}{k_F^2\mu^2}+{\cal{O}}((\tau \E)^4)] =\eta(0)[1-\frac{11}{28}(\lambda k_F)^2\kappa_{\rm{LDA}}^2+{\cal{O}}((\tau\E)^4)]\\ \eta_3 &= 0, ~\eta_4 = 0\;,~\label{eq:etaoflambda2} \end{split} \end{equation} where $\tau$ is the effective relaxation time.\\ For the Elliptic mode $\frac{1}{2}(\partial_i u_j + \partial_j u_i)=\frac{1}{2} \alpha_x(1-\frac{\omega_x^2}{\omega_z^2})= V_{xz}$ which probes the viscosity contribution to the stress energy tensor \begin{equation} \sigma_{2_{ \alpha \beta}} = 2 ~\eta_2~(V_{\alpha \gamma} b_{\beta} b_{\gamma} +b_{\alpha}V_{\beta \gamma} b_{\gamma}- 2 b_{\alpha} b_ \beta b_\gamma b_\delta V_{\gamma \delta} ) \;, \end{equation} where $b$ is a unit vector along the gradient of the potential. For the Scissor mode, $\frac{1}{2}(\partial_i u_j + \partial_j u_i)=\alpha_x = V_{xz}$ which also probes $\eta_2$. ($\eta_2$ is the coefficient that corresponds to the projection operator $M_2$ in Eq.~\ref{eq:projections}.) In both cases (see Appendix.~\ref{microscopic}) , $\eta$ is reduced from its value in the absence of the potential, $\eta(0)$, for $\frac{\tau^2}{k_F^2}(\E)^2\lesssim 1$. To estimate the value of $\tau$ near $z=z_0$, we note that for $z\sim z_0$, $T(z_0)\sim 0.54~ \mu(z_0)$. At this $T$, $\eta(0)/n|_{z_0}\sim 1$. Using the relaxation time approximation and thermodynamic expressions for a weakly interacting Fermi gas to estimate $\lambda$ near $z_0$, we obtain (Eq.~\ref{eq:eta0degenerate}) \begin{equation} \begin{split} {\eta(0)}(z_0) &= \frac{(2m\mu(z_0))^\frac{5}{2}\tau(z_0)}{15\pi^2 m}\\ &=\frac{2}{5}n(z_0)\mu(z_0)\tau(z_0)\;. \end{split} \end{equation} Therefore near $z_0$, $\tau(z_0)\sim \frac{5}{2\mu(z_0)}\frac{\eta(0)}{n}|_{z_0}$, or, \begin{equation} \begin{split} \lambda(z_0) &= v_F(z_0) \tau(z_0)\\ &\sim \frac{k_F(z_0)}{m} \frac{5}{2\mu(z_0)}\frac{\eta(0)}{n}|_{z_0}\\ &=\frac{5}{4k_F(z_0)}\frac{\eta(0)}{n}|_{z_0}\;. \end{split} \end{equation} (We have just kept the pre-factors of the order of $1$ to serve as mnemonics of the derivation of $\lambda$. They have no quantitative significance.) Therefore, (since $\frac{\eta(0)}{n}|_{z_0} \sim 1$ from $\frac{\eta}{n}$ data) \begin{equation} \begin{split} \lambda(z_0)k_{F}(z_0) =\frac{5}{4}\frac{\eta(0)}{n}|_{z_0}\sim 1\;. \end{split} \end{equation} The fact that $k_{F}(z_0) \lambda(z_0)\sim 1$ means that the Boltzmann transport calculation shown in Appendix.~\ref{microscopic} is not quantitatively trustworthy near $z_0$. But we hope that two the main qualitative consequences of Eq.~\ref{eq:etaoflambda2} survive a more controlled calculation. \begin{enumerate} \item{First, the coefficient of $\kappa_{\rm{LDA}}^2$ in Eq.~\ref{eq:etaoflambda2} is of the order of $1$.} \item{Second, the sign of the correction term is negative} \end{enumerate} If true, this would imply that the shear viscosity component $\eta_{xzxz}$ measured using the Elliptic mode or the Scissor mode will reduced by order $1$ from its value in isotropic traps, if $\omega_z\gtrsim 2 \pi \times 77000$ rads/s (Table.~\ref{ptrap1}). One might be concerned that for $\omega_z\sim 2 \pi \times 77000$ rads/s, our conclusions in the previous section about $\delta z/z_0$ will be violated because of the violation of LDA. In the absence of more concrete information on these coefficients we can not assure this will not happen. We simply note that if the coefficient $c_1$ in Eq.~\ref{eq:LDAviolationvskappa} is of the order of $1$ (which it is at $T\ll\mu$, but may be larger for $T\sim0.54~\mu(z_0)$) then there is a regime where the corrections to the thermodynamics due to LDA is small, but the reduction in transport coefficients is substantial. \section{Conclusions} \label{res} We present a concrete realization of a system of ultra-cold Fermi gases at unitarity, in an anisotropic trap, which may show significant reduction in the viscosity compared to its value in isotropic traps. Given that the value of the isotropic viscosity has been measured to be few times the KSS bound in this system, it presents a candidate setup to observe a shear viscosity smaller than the KSS bound when it is subjected to an anisotropic driving force. The anisotropic force is obtained by placing the system in an anisotropic trap. The trapping potential is harmonic, Eq.~\ref{eq:harmonic_potential}, and characterized by three angular frequencies, $\omega_x,\omega_y,\omega_z$. We consider an anisotropic situation where $\omega_z\gg \omega_x, \omega_y$, so that the trapping potential is much stronger in the $z$ direction. For simplicity, we also take $\omega_x=\omega_y$ so that the system preserves rotational invariance in the $x-y$ plane. For some of the discussion below we can neglect the effects of the trapping potential in the $x,y$ directions characterized by $\omega_x,\omega_y$. We work in conventions where $k_B=\hbar=1$. There are three energy scales $T, \mu, \omega_z$ and two dimensionless ratios $T/\mu$ and $\omega_z/\mu$ which then characterize the system. The Li$_6$ atoms have a mass $m$, using this parameter, any of the energy scales can be converted to a length scale, $L={1\over \sqrt{2 m E}}$. Based on the behavior seen quite generically in gravity systems we identify five criterion (Sec.~\ref{cond}) which when met could plausibly lead to a decrease in the value of some components of the viscosity tensor (the spin one components). These are summarized towards the end of Sec.~\ref{gravity} . On studying the superfluid equations we identify two modes which are sensitive to these components of the viscosity tensor. One of these is the scissor mode which has already been studied experimentally in some detail. By taking reasonable values for the parameters- $T$, $\mu$, $\omega_z, \omega_x, \omega_y$, which are in the experimentally accessible range, Ref.~\cite{Cao58}, we find that all the five criteria can be met. Furthermore, we find that the resulting energy and damping rate of this energy, from which the viscosity can be extracted, lie within the range of values which are measured by experiments currently being done on cold atom systems, in particular on Li$_6$ unitary Fermi gas systems, Ref.~\cite{PhysRevLett.99.150403}. For example, for $\mu=10\mu$K, $\omega_z \sim 2 \pi \times 77000$ rads/s, and $T= \frac{T_c}{2}$ ($T_c=0.4\mu$) we find that the anisotropy, as measured by the parameter $\kappa_{LDA} $ , Eq.~\ref{defkappalda}, is of order unity and therefore significant. At these extreme values of anisotropy our theoretical calculation, strictly speaking, do not apply, but a reasonable extrapolation suggests that the maximum total energy is of the order of $10^{-17}$ joules which corresponds to the angular amplitude of the scissor mode of about $24^{\circ}$ which is within the experimental range of \cite{PhysRevLett.99.150403}. The damping time $\tau_{0}$ is of the order of $10^{-2} $ seconds, which is roughly ten times longer than the observed amplitude damping time that has been accurately measured in the experiments on ultracold Fermi gases \cite{PhysRevLett.99.150403}. \\ While the system is certainly close to being two-dimensional when $\kappa_{LDA} \sim1$ and $z_{trap} \sim 5.4~ k_F^{-1}$ (this corresponds to $\mu/\omega_z \sim 2.7$) is on the small side, the effect of small viscosity can already set in when $\kappa_{LDA}$ is somewhat smaller than unity. We illustrate this with concrete quantitative examples below. \\ For concreteness, let us consider traps where we fix $T/T_c=1/2$ ($T_c=0.4\mu$, where $\mu$ is the chemical potential at the center of the trap) and change $\omega_z$. Further, for concreteness, we set the overall scale by $\mu=10\mu$K. Considering first a representative trap geometry where the shear viscosity tensor is locally isotropic to a large accuracy, we take $\omega_z=0.048\mu$ (corresponding to $\omega_z=2\pi\times 10^4$Hz which is typical), for which $\kappa_{{\rm{LDA}}}=0.13$. The fractional reduction in the shear viscosity for this value of $\omega_z$, taking $c_2$ to be its Boltzmann transport value $11/28$ is \begin{equation} \begin{split} \frac{\Delta\eta}{\eta}\approx -\frac{11}{28} (\kappa_{{\rm{LDA}}})^2 = -0.7\%\;, \end{split} \end{equation} which is a small reduction in the shear viscosity and may not be even measurable above measurement errors. At the other extreme we considered, $\omega_z= {\mu \over 2.7 }$ (corresponding to $\omega_z=2 \pi \times 77.16$~kHz), for which $\kappa_{{\rm{LDA}}}=1$ and the fractional reduction is \begin{equation} \begin{split} \frac{\Delta\eta}{\eta}\approx -\frac{11}{28} (\kappa_{{\rm{LDA}}})^2 = -39\%\;, \end{split} \end{equation} which is very large. However, in this extreme limit ($\omega_z={ \mu \over 2.7 }$) only the lowest $2-3$ Landau levels are occupied and the dynamics may be approximately two dimensional. Now consider an intermediate value, say $\omega_z=0.9T = 0.18 \mu$ for which $\kappa_{LDA}= 0.48 <1$. This gives a correction \begin{equation} \begin{split} \frac{\Delta\eta}{\eta}\approx -9\% \end{split} \end{equation} which --- while not large --- is still substantial. More generally, the criterion for confinement in the $z$ direction is \begin{equation} \begin{split} \omega_z\gtrsim {\rm{max}}(\Delta, T)\;, \end{split} \end{equation} since both $T$ and pairing allow for excitations between the harmonic oscillator levels. At these extreme values, where the inequality above is met, our approximations do break down, (shell effects become important as $\omega_z\gtrsim T$, which is another way of saying that confinement in the $z$ direction becomes strong). For $\omega_z={ \mu \over 2.7 }$, $\omega_z=1.85~ T$ and indeed confinement in the $z$ direction is too strong. But, as illustrated by the cases above, by taking $\omega_z$ a factor of $2$ or $3$ smaller ( say $\omega_z=0.9~ T$ that was chosen above for illustration\footnote{The deviations from LDA due to shell effects for unpaired fermions was calculated in Ref.~\cite{PhysRevA.67.053601}. A naive application of the results of Ref.~\cite{PhysRevA.67.053601} suggests that for our trap with $\omega_z =0.18\mu$, the corrections to the number density is about $15\%$ at $T=4 T_c/5$ near the region relevant for our purposes. Note however, that pairing suppresses LDA violations (\cite{Forbes:2012yp,Bulgac:2010dg}) and we expect the corrections to be much smaller in the relevant region.}) than the extreme limit, one can measure the tendency of the spin one component of the viscosity to decrease from its lowest value observed in ultra-cold Fermi gases. In an optimistic scenario where $c_2$ is larger in magnitude than the approximate value of $11/28$ in the Boltzmann transport approximation, the reduction will be even more substantial. Let us also point out that comparing with Ref.\cite{PhysRevLett.108.070404} the typical values of $\omega_z/E_F$ in the paper is about $80$ and the value of $\omega_z/T$ is $120$. In that case, the trap is truly 2 dimensional as opposed to when $\omega_z/T\sim 0.9$.\\ Thus, for smaller values of anisotropy, the theoretical estimates are more reliable and suggest that the different viscosity tensor components should have a fractional difference given in terms of $\kappa_{LDA}$ by Eq.~\ref{eq:etaoflambda2}. This tendency of the viscosity to decrease should already be measurable at more moderate values of the anisotropy. Our proposal is the first proposal to measure parametrically suppressed anisotropic viscosity components in ultra-cold Fermi gases. Our proposal is different from the discussion of anisotropic hydrodynamics in Ref.~\cite{Bluhm:2015raa} since we are demanding that hydrodynamics be a good description (in the sense of Eq.~\ref{eq:hydro_condition1}) in the regime which dominantly contributes to the dissipation of the fluid dynamics modes. Future theoretical work can improve upon our proposal in several ways. First, our estimate of the corrections to the shear viscosity components due to the potential (Eq.~\ref{eq:etaoflambda2}) was based on a relaxation time treatment of the Boltzmann equation. For strongly interacting fermions, this is not a good approximation and a more rigorous calculation of the anisotropy corrections is desirable. This will require calculating transport properties in a strongly coupled theory without a gravitational dual, in the presence of a background potential: a formidable challenge. Second, we have focused on the region that dominantly contributes to the dissipation. In particular we have neglected the contributions from the tail of the cloud. While this is presumably small, it would be nice to establish this by solving the Boltzmann transport equations in this dilute regime. It is also worth noting that while the cold-atom system proposed here shares many features with those discussed in Ref.~\cite{Jain:2014vka,Jain:2015txa}, it also has some differences. First, in equilibrium the stress energy tensor is not invariant under translations even for a linear potential. Rather the density decreases with increasing $z$, but the driving force is proportional to the gradient of the potential $\phi({\bf{r}})$ (see Eq.~\ref{eq:hydrodynamics}) as in Ref.~\cite{Jain:2014vka,Jain:2015txa}. Second, in addition to energy-momentum, the cold-atom system features another conserved quantity: the particle number. Consequently the system is locally characterized by two thermodynamic variables $T$ and $\mu$ rather than just $T$. It would also be interesting to further study the behavior of viscosity in gravitational systems which correspond to anisotropy driven strongly coupled systems with a finite chemical potential. The examples in Ref.~\cite{Jain:2014vka,Jain:2015txa} did not have a finite chemical potential, for some discussion of anisotropic gravity systems with a chemical potential see Ref.~\cite{Ge:2014aza,Chakraborty:2017msh}. As a first step, we have analyzed a weakly coupled system with a linear varying potential in Appendix.~\ref{microscopic} and find that the viscosity does become anisotropic in this case.\\ However, there is no reason to wait for these theoretical advances. The central point of this paper is that there is already enough motivation, based on the behavior quite generically seen in gravitational systems, to suggest that some components of the viscosity tensor in anisotropic strongly coupled systems might well become small, making $\eta/s$ for these components potentially even smaller than the KSS bound, $1/ 4\pi$. Such a decrease in the viscosity might well happen in cold atom systems, for example the unitary fermi gas, which are experimentally well studied. As argued above, the range of values involved for temperature, chemical potential and angular frequencies are well within the experimental regime for such a system, and the scissor mode which is sensitive to the relevant components of the viscosity has already been realised experimentally in them. Further, the resulting values for the energy and the damping time from which the viscosity can be extracted lie in the experimentally accessible range which has already been achieved. We hope our experimental colleagues in the cold atoms community will take note of these results, and implore them to carry out a study of viscosity in anisotropic traps. \section{Acknowledgments} We thank D. D. Ofengeim and in particular D. G. Yakovlev for sharing their notes on the calculation of the various components of viscosities in the presence of the magnetic field. We especially thank M. Randeria for sharing his valuable comments and insights. We also acknowledge conversations with K. Damle, S. Gupta, S. Jain, N. Kundu, G. Mandal, S. Minwalla, T. Sch{\"a}fer, R. SenSarma and N. Trivedi. SPT acknowledges support from the J. C. Bose fellowship DST, Government of India. We acknowledge support from the DAE, Government of India. Most of all we thank the people of India for supporting our research.
b555261b90153dda8c7183c0b453562be50a19f8
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction}\label{sec:intro} The precise manipulation of nano and sub-nanoscale physical systems lies at the heart of the ongoing quantum revolution, by which new communication and information technologies are expected to emerge \cite{bib:natphot2009,bib:nature2010}. In this context, an amazing progress has been made in the study of non-equilibrium dynamics of many-body quantum systems, both theoretically and experimentally \cite{bib:rev_mbody2008,bib:rev_neq2011}. A wide range of different phenomena has been closely studied in recent years, such as many-body localization \cite{bib:mbl2010,bib:mbl2015}, relaxation \cite{bib:rigol2007,bib:eisert2008,bib:wisn2015}, thermalization \cite{bib:rigol2008,bib:santos2011,bib:fazio2011}, quantum phase transitions \cite{bib:qpt2011}, among others.\\ Understanding the dynamics of such complex quantum systems is the first step towards the ultimate goal: the ability to engineer its complete time-evolution using a small number of properly tailored control fields. To tackle this problem, optimal control theory (OCT) \cite{bib:tannor1993,bib:rabitz1998} emerges as the natural tool. Routinely used in various branches of science \cite{bib:krotov1996}, optimization techniques allows to derive the required shape for a control field $\epsilon(t)$ that optimizes a particular dynamical process for a quantum system described by a Hamiltonian $H(\epsilon)$. For example, a typical goal in quantum control is to connect a given initial $\Ket{\psi_0}$ and target states $\Ket{\psi_f}$ in some evolution time $T$. In recent years, optimal control has been applied with great success in systems of increasing complexity, with applications including state control of many-boson dynamics \cite{bib:sherson2013,bib:calarco2015}, the crossing of quantum-phase transitions \cite{bib:doria2011}, generation of many-body entangled states \cite{bib:mintert2010,bib:caneva2012} and optimization of quantum thermodynamic cycles \cite{bib:montangero2016}. A lot of attention has also been devoted to investigate the fundamental limitations of OCT, most of all in connection with the study of the so-called quantum speed limit \cite{bib:caneva2009,bib:murphy2010,bib:hegerfeldt2013,bib:nos_qsl2013,bib:nos_qsl2015}. In a recent work, OCT has even been used in a citizen-science scenario allowing to investigate the power of gamification techniques in solving quantum control problems \cite{bib:sherson2016}.\\ In this work, we investigate the connection between the complexity of a quantum system and its controllability. To this end, we study optimal control protocols on a spin-1/2 chain with short-range interactions, both in the few- and many-body regimes. By using this model, we are able to tune the physical complexity of the system in two different ways: (a) by adding excitations to the chain, we can increase the system space dimension; (b) by tuning the interparticle coupling, we can drive the system through a transition from a regular energy spectrum to a chaotic one. We perform an unconstrained optimization in order to obtain the control fields needed to drive various physical processes, and define two figures of merit based on the frequency spectrum of the fields: the spectral bandwidth, associated with the maximum frequency present in the field and the spectral inverse participation ratio (sIPR), related to the signal complexity. We find that the spectral bandwidth is strongly connected to the structure of the control Hamiltonian. In the common scenario where the control is applied locally on any site of the chain, we find that the bandwidth is independent of the state space dimension, for various processes. On the other hand, the complexity of the signal grows with the dimension, due to the increase of energy levels. Inspired by this, we asses the role of quantum chaos in the control of quantum many-body dynamics. We find that the transition between regular and chaotic energy spectrum does not affect the complexity of the control problem in a significant way.\\ We point out that previous works have studied the relationship between optimal control and the integrable or non-integrable nature of the quantum system under analysis \cite{bib:caneva2014,bib:lloyd2014}. There, a suitable measure for the control field complexity was defined, related with the number of frequencies in the field, as allowed by the optimization procedure. Then, it was shown that the complexity required to achieve control scaled exponentially with the dimension of the manifold supporting the dynamics. Here, we focus on studying the complexity of the control fields regardless of the details of the optimization method itself. We do this by deliberately allowing many frequency components in the control fields, and then analysing which of those components are required to effectively drive the system. \\ This article is organized as follows. In Sec. \ref{sec:model} we present the model of a spin-1/2 chain and discuss its symmetries. Also, we propose a few control protocols and put forward the main elements of QOC theory. In Sec. \ref{sec:result} we define two measures of complexity for the optimal control fields, and present a systematic study on how they vary with both the system state dimension and chaos parameter. In. Sec. \ref{sec:discuss} we analyse in detail the reasons why the chaotic regime does not affect the controllability of the system. Finally, Sec. \ref{sec:conclu} contains some concluding remarks. \section{Model and methods} \label{sec:model} \subsection{Spin chain model} Let us consider a one-dimensional system of $L$ spin-1/2 particles that interact through nearest-neighbor (NN) and next-to-nearest-neighbor (NNN) homogenous couplings with open boundary conditions. The Hamiltonian for this models reads \begin{eqnarray} H_{01} &=& H_0 + \Gamma H_1, \label{ec:h01}\\ H_0 &=& \frac{J}{2}\sum_{i=1}^{L-1}\sigma_i^x\sigma_{i+1}^x+\sigma_i^y\sigma_{i+1}^y+\alpha_z\sigma_i^z\sigma_{i+1}^z,\\ H_1 &=& \frac{J}{2}\sum_{i=1}^{L-2}\sigma_i^x\sigma_{i+2}^x+\sigma_i^y\sigma_{i+2}^y+\alpha_z\sigma_i^z\sigma_{i+2}^z, \end{eqnarray} \noindent where $\sigma_i^{x,y,z}$ are the Pauli matrices for the i-th particle and we have taken $\hbar=1$. The Hamiltonian $H_0$, which has only NN couplings, is the usual $XXZ$ Heisenberg model, which can be exactly solved via the Bethe ansatz \cite{bib:bethe}. The parameter $\Gamma$ measures the ratio between the NNN exchange and the NN couplings. This model has been extensively studied in the literature in many contexts \cite{bib:vedral2001,bib:fazio2002}, in particular when investigating quantum chaos \cite{bib:wisn2015,bib:santos2012}, i.e., the study of the quantum mechanical properties of systems which classical analogs display a chaotic behavior. Albeit lacking a semiclassical counterpart, this spin model displays a transition in its level spacing distribution as $\Gamma$ changes. For $\Gamma\lesssim0.5$, the energy spectrum is regular, and its level spacings follow a Poisson distribution, while for $\Gamma\gtrsim0.5$, the distribution follows Wigner-Dyson statistics, and the spectrum is deemed "chaotic" \cite{bib:stockman,bib:bohigas} (see the Appendix for more details).\\ While the full Hilbert space $\mathcal{H}$ of this model has a dimension of $2^L$, we can identify two constant of motions such that $\mathcal{H}$ is decomposed into smaller subspaces. First, the total magnetization the in $z$ direction $\sigma_z=\sum_{i=1}^L \sigma_i^z$ is conserved, which defines $L+1$ subspaces with fixed $\sigma_z$. Each subspace can then be characterized by this quantum number, which can be interpreted as the number of sites with spins pointing ``up'' or ``excitations'' in the chain. The dimension of the subspace with $K$ excitations is given by \begin{equation} D_K=\frac{L!}{K!(L-K)!}. \end{equation} We consider also conservation of parity. The parity operator $\Pi$ acts as permutation between mirrored sites of the chain, and commutes with the Hamiltonian $H_{01}$ for all values of $J$ and $\Gamma$. As a consequence, each of the above mentioned subspaces break up into two (positive and negative) parity subspaces, each of which of dimension $D_{K,\Pi}\simeq D_K/2$ \cite{bib:santos2012}. Lastly, we avoid conservation of $\sigma_z^2$ by chosing $\alpha_z=0.5$, and choose an odd value of the chain length $L=15$. \subsection{Control protocols} We now describe an scenario to perform control operations on the spin chain. First, we define the control Hamiltonian by means of which we intend to steer the ``free'' chain Hamiltonian $H_{01}$. Several different proposals have been studied in the literature. For example, in Ref. \cite{bib:bose2007}, the author proposed using a global parabolic magnetic potential to control the transfer of excitations from one end of the chain to the other. Later, the same configuration was used to study the optimal evolution time for such processes \cite{bib:caneva2009,bib:murphy2010}. Other control configurations have also been proposed in scenarios where only one \cite{bib:spin1} or two \cite{bib:spin2,bib:spin2b,bib:spin2c} sites are locally addressed by external fields. Here, we will consider time-dependent magnetic fields in the $z$ direction which are locally applied at each site of the chain. In order to comply with the system symmetries, we will consider the situation where the first and last spin of the chain are affected by the same field $\epsilon(t)$, whereas all the other spins do not interact with any external field. Consequently, the full Hamiltonian can be written as \begin{equation} H(t) = H_{01} + \epsilon(t)H_c,\ \mathrm{where}\ H_c=\frac{J}{2}\left(\sigma_1^z+\sigma_L^z\right). \label{ec:hcontrol} \end{equation} We point out that $H(t)$ preserves the same symmetries as $H_{01}$ for any choice of $\epsilon(t)$. In addition to this, we checked that for any fixed value of the field, $H$ still shows a transition between a regular and a chaotic spectrum for $\Gamma\simeq0.5$. For more details about this issue, see the Appendix.\\ The next step is to define the control processes we aim to perform. We will consider two different protocols (A and B) in order to obtain general results about the systems controllability. In both cases, we define initial and target states which we denote $\Ket{\psi_0^\alpha}$ and $\Ket{\psi_f^\alpha}$, where $\alpha=A,B$. These states are deliberately defined to allow the system evolve within a particular subspace with fixed (positive) parity and number of excitations $K$ of the complete Hilbert space, as discussed in the previous section. First, process ``A'' involves the system initially prepared in a state with all excitations in the middle sites of the chain (in this scheme, the central site has no excitations if K is even). We then intend to drive this configuration into a coherent superposition as defined by \begin{eqnarray} \Ket{\psi_0^A}&=&\Ket{\downarrow\ldots\downarrow\uparrow\uparrow\uparrow\downarrow\ldots\downarrow}\\ \Ket{\psi_f^A}&=&\frac{1}{\sqrt{2}}\left(\Ket{\uparrow\uparrow\uparrow\downarrow\ldots\downarrow}+\Ket{\downarrow\ldots\downarrow\uparrow\uparrow\uparrow}\right). \end{eqnarray} Process A then represents and orderered control process in which entanglement is generated between both ends of the chain. On the other hand, we define a disordered process B, where the system starts from the ground state of $H_0$ and its steered into a random superposition of excited states (with positive parity). \begin{eqnarray} \Ket{\psi_0^B}&=&\Ket{\mathrm{g.s.}_0}\\ \Ket{\psi_f^B}&=&\sum_{n=1}^{D_{K,+}-1}a_n\Ket{n_0} \end{eqnarray} \noindent where $\left\{\Ket{n_0}\right\}$, $n=0,\ldots,D_{K,+}$ are the positive eigenvectors of $H_0$ in the subspace of $K$ excitations and $\Ket{\mathrm{g.s.}_0}\equiv\Ket{0_0}$. The coefficientes $\left\{a_n\right\}$ are a set of random complex numbers so that $\sum_i |a_n|^2=1$.\\ \subsection{Optimal control methods} In order to obtain the control fields $\epsilon(t)$ which drive the system for both processes, we use optimal control theory. Here we briefly sketch the Krotov optimization algorithm, as described in many previous works (see for example Refs. \cite{bib:krotov1996,bib:sklarz2002,bib:montangero2007,bib:werschnik2007,bib:nos_oc2015}). This procedure takes as an input a fixed evolution time $T$, an initial guess for the control field $\epsilon^{(0)}(t)$, and both the initial and final states $\Ket{\psi_0^\alpha}$ and $\Ket{\psi_f^\alpha}$, where $\alpha=A,B$. The procedure starts by evolving $\Ket{\psi_0^\alpha}$ according to the Hamiltonian $H(\epsilon^{(0)}(t))$, from $t=0$ to $t=T$. The final state $\Ket{\psi(T)}$ is then projected to $\Ket{\psi_f^\alpha}$ in order to obtain an auxiliary state $\Ket{\chi(T)}=\BraKet{\psi_f^\alpha}{\psi(T)}\Ket{\psi_f^\alpha}$. This state is finally evolved backwards with the same Hamiltonian, from $t=T$ to $t=0$. The process is then repeated, but the control field is updated following the recipe \begin{equation} \epsilon^{(k+1)}(t)\rightarrow\epsilon^{(k)}(t) + \frac{1}{\lambda(t)}\mathrm{Im}\left\{\Bra{\chi(t)}\frac{\partial H}{\partial \epsilon}\Ket{\psi(t)}\right\}, \end{equation} \noindent where $\lambda(t)$ is an weight function. Note that, for the model considered here, the operator $\partial H/\partial \epsilon$ is fully defined by equation (\ref{ec:hcontrol}) and equals simply to $H_c$. The iterative procedure stops when a certain target fidelity $\mathcal{F}=|\BraKet{\psi(T)}{\psi_f}|^2$ has been achieved.\\ \begin{figure}[!t] \includegraphics[width=\linewidth]{fig1.pdf} \caption{\label{fig:fig1} (color online) (a) A typical optimized control field $\epsilon(t)$ obtained for process A and $\Gamma=1$ and (b) its Fourier transform. (c) Spectral bandwidth of the optimized control fields as a function of the adimensional NNN coupling $\Gamma$, for various values of $K$ (the number of excitations in the spin chain). Data shown is for both control processes A and B (see text for details). State space dimension ranges from 15 (K=1) to 1365 (K=4). The dashed curves show the energy spread of the free Hamiltonian $H_{01}$ (\ref{ec:h01}) for different values of $K$. The dash-dotted vertical line indicates the critical value of $\Gamma=0.5$ for which the regular-chaotic transition ocurrs in the energy spectrum of $H_{01}$. Inset: same as in the main figure, but plotted as a function of the NN coupling J, for a fixed value of $\Gamma=1$.} \end{figure} As we intend to compare the optimal control fields obtained by this optimization procedure, we fix the input parameters of the optimization as follows. For the total evolution time $T$ we set $T=15\times T_L$ where $T_L=(L-1)\frac{\pi}{J}$ can be regarded as the typical evolution time required for transferring a single excitation from one end of the chain to the other \cite{bib:murphy2010}. We have checked that using this value we are operating well beyond the quantum speed limit \cite{bib:caneva2009}, and so that fidelities up to 0.99 or greater can be achieved, for both control processes and every value of the number of excitations $K$ and the NNN coupling $\Gamma$ considered. Also, we used a constant initial guess $\epsilon^{(0)}(t)=0.1$ in all cases. We have checked that the results we present in the next section hold for other choices of this function. \section{Analysis of the optimal control fields} \label{sec:result} In Fig. \ref{fig:fig1} (a) we show a typical example of the control field $\epsilon(t)$ obtained by the optimization procedure outlined in the previous section, together with its Fourier spectrum. The time signal shown can be seen to be complex and to have many spectral components up to certain frequency threshold. In order to characterize quantitively this features, we define two measures of complexity for the control fields: the frequency bandwidth and the spectral inverse participation ratio (sIPR). In this section we investigate these quantites. \subsection{Frequency bandwidth} \begin{figure}[!b] \includegraphics[width=\linewidth]{fig2.pdf} \caption{\label{fig:fig2} (color online) (a) Absolute values of the matrix elements for the control Hamiltonian in the computational basis (left panels) and in the energy eigenbasis with positive parity (right panels). Top panels: local control with $H_c$ as in eq. \ref{ec:hcontrol}. Bottom panels: long-range control with $H'_c$ described in text. (b) Frequency bandwidth of the optimized control fields as a function of the adimensional NNN coupling $\Gamma$, for $K=1,\:2$ (the number of excitations in the spin chain). Data shown is for process A (see text for details), using the long range control Hamiltonian $H'_c$. } \end{figure} Given a time-varying signal $\epsilon(t)$ and its Fourier transform $\hat{\epsilon}(\omega)$, we first define its frequency bandwidth as the value $\omega_{bw}$ such that \begin{equation} \int_0^{\omega_{bw}}d\omega\:|\hat{\epsilon}(\omega)|^2=1-\beta, \label{ec:bw} \end{equation} \noindent where $0<\beta<1$ and the frequency distribution is normalized such that $\int_0^{\infty}d\omega\:|\hat{\epsilon}(\omega)|^2=1$. By this definition, the frequency interval $[0,\omega_{bw}]$ concentrates the $[(1-\beta)\times100]\%$ of the power spectrum (here, we use $\beta=10^{-2}$). In other words, $\omega_{bw}$ is a measure of the maximum frequency present in $\epsilon(t)$. \\ In Fig. \ref{fig:fig1} (b) we show the frequency bandwidth $\omega_{bw}$ as a function of the NNN or chaos parameter $\Gamma$, for different number $K$ of excitations in the chain. Results obtained for both processes A and B are shown in the same plot. Remarkably, we find that all data roughly coincides in the same curve. This result indicates that the bandwidth is independent not only of the control processes considered, but also of the state space dimension. Note that, in each case, $\omega_{bw}$ is approximately constant for $\Gamma<0.5$ and then increases steadly for $\Gamma>0.5$. Although this behavior correlates with the onset of chaos in the system (as discussed in Sec. \ref{sec:intro}), we must first consider that increasing the interparticle coupling $\Gamma$ necessarily increases the energy of the chain. As previously discussed in the context of QOC \cite{bib:rabitz1998_2,bib:tesch2002}, we expect that the frequency distribution of the control fields presents peaks located at the resonances of the free Hamiltonian $H_{01}$. Following this criterion, the maximum frequency is bounded by the energy spread $\Delta E$ of $H_{01}$, defined as \begin{equation} \Delta E = E_{max}-E_{0}, \end{equation} \noindent where $E_{max}$ and $E_0$ are the maximum and minimum (ground state) energies of the Hamiltonian. Note that $\Delta E$ is a function of the interparticle interaction parameters $J$ and $\Gamma$ and of the number of excitations $K$. We show such functions as dashed lines in Fig. \ref{fig:fig1} (c). It is clear that the dependence of the bandwidth with $\Gamma$ closely resembles the energy spread with $K=1$. The same observation can be drawn by studying both quantities as a function the NN coupling $J$ (for fixed $\Gamma$). There, the behaviour is obviously linear, as shown in the inset of Fig. 1 (c).\\ Note, however, that the striking independence of the bandwith with the state dimension cannot be explained by its relation with the energy spread of the free Hamiltonian, since $\omega_{bw}$ is in every case significantly smaller than $\Delta E$ for $K>1$. In order to gain a deeper insight about this result, we turn to investigate the role of the control Hamiltonian $H_c$, defined in eq. (\ref{ec:hcontrol}). We first study the structure of the matrix $H_c$ written in the basis of (positive) eigenvectors of the free chain Hamiltonian $H_{01}$. In the top row of Fig. \ref{fig:fig2} (a) we plot the absolute value of such matrix elements for fixed values of $\Gamma$ and $K$. From this plots we can see that $H_c$ does not connect eigenstates which distant energies: for example, the ground state is not connected with excited states beyond the middle of the spectrum. This explains the absence of such high transition frequencies in the spectrum of the control fields. In order to provide numerical proof about this feature, we studied the implementation of one the control processes with a different choice of control operator $H'_c$ which presents a higher connectivity \cite{bib:santos2012} between distant states in the spectrum. Such Hamiltonian matrix is shown in the bottom-right panel of Fig. \ref{fig:fig2} (a). Results for the new optimized fields are shown in Fig. \ref{fig:fig2} (b), where we show the frequency bandwidth as a function of $\Gamma$ for this case. It can be seen that $\omega_{bw}$ is greater for $K=2$ than for $K=1$, for all values of $\Gamma$ considered. We point out that, by looking at the representation of $H'_c$ in the computational basis (bottom left panel), we can see that this alternative control procedure would involve tuning a complex combination of multi-spin interactions, in clear contrast with the simple structure of $H_c$. \\ The results shown so far allows us to assert that the control bandwidth, which measures the range of frequencies present in the fields is determined exclusively by the energy spread of the free Hamiltonian and the structure of the control Hamiltonian. This gives us a measure of the physical complexity of the control field which turns to be independent of the number of particles in the system. We point out here that we are not interested in analyzing the complexity of the optimization itself, as has been done in previous works which have obtained interesting results \cite{bib:caneva2014,bib:lloyd2014}. We work our way around this issue by fixing the time step of our numerical implementation to very small values, $J\Delta t = 10^{-2}$. This determines that the maximum allowed frequency in the fields is en every case at least on order of magnitud higher than the actual physical frequencies found by Fourier analysis in the control fields. \subsection{Spectral localization} \begin{figure}[!b] \includegraphics[width=\linewidth]{fig3.pdf} \caption{\label{fig:fig3} (color online) (a) Various spectral densities of the optimal control fields found for process A. Values for the excitation number $K$ and the adimensional next to nearest-neighbour coupling $\Gamma$ are shown for each case I-IV. (b) Normalized spectral inverse participation ratio (sIPRn), which measures the control field complexity, as a function of $\Gamma$, for various values of $K$ (the number of excitations in the spin chain). Data shown is for process A (left) and B (right). The points indicated by labels I-IV correspond to the spectrums shown in (a). State space dimension ranges from 15 ($K=1$) to 1365 ($K=4$). The dash-dotted vertical line shows the critical value of $\Gamma$ for the regular-chaotic transition in the energy spectrum.} \end{figure} We now turn our attention to another measure of the control field complexity. In this case, we to study how the number of frequencies which appear in the signal spectrum varies as the systems complexity is increased. For this purpose, we define the following quantity \begin{equation} \mathrm{sIPR} = \left(\int_0^\infty d\omega\:|\hat{\epsilon}(\omega)|^4\right)^{-1}, \end{equation} \noindent which we call ``spectral inverse participation ratio'' (sIPR) as it is inspired in the commonly known IPR \cite{bib:wisn2015,bib:ipr1,bib:ipr2}. The sIPR quantifies the localization in the Fourier transform of a time signal, and thus allows us to asses how complex the control field is inside its bandwidth. Note that localized frequency spectrums give sIPR$\rightarrow0$, and complex signals with delocalized spectrum tend to higher sIPR. As an example, take a completely random signal with frequency components up to $\omega_{bw}$. We expect such a signal to have a flat Fourier transform $\hat{\epsilon}(\omega)=1/\omega_{bw}$ for $0<\omega<\omega_{bw}$ and $\hat{\epsilon}(\omega)=0$ for $\omega>\omega_{bw}$. Calculating the sIPR in that case is straightforward and gives $\omega_{bw}$. We point out that here we intend to quantify the optimal control field complexity regardless of the frequency distribution width. For this purpose, we evaluate the normalized sIPR \begin{equation} \mathrm{sIPRn} = \frac{\mathrm{sIPR}}{\omega_{bw}}. \end{equation} Following the discussion on the previous paragraph, we expect sIPRn to range between 0 and 1, and we can interpret it as a measure of resamblence between the signal under study and a completely random time field.\\ In Fig. \ref{fig:fig3} we plot the normalized spectral IPR as a function of the NNN exchange $\Gamma$ for different values of the number $K$ of excitations in the spin chain, and for both control processes A and B. We show also some examples of the frequency spectrums we obtained, and it can be corroborated that sIPRn effectively measures how localized the spectrum is in Fourier space. More generally, it can be seen from the figure that sIPRn takes small values for $K=1$ and then grows with $K$, and thus with state space dimension of the system. This is in sharp contrast with the behaviour of the frequency bandwidth $\omega_{bw}$, which was found to be independent of $K$. We point out that this behaviour is common to both control processes. It is interesting to note that the high-dimensional cases ($K=3,4$) roughly converge to the same value of sIPRn, indicating that there maybe an upper bound for this measure which is below its maximal theoretical sIPRn$=1$, which is achieved when the frequency spectrum is flat. Physically, the existence of an upper bound $<1$ means that optimal control fields can always be distinguished from completely random, white noise-type fields. We leave this issue for future investigation.\\ Despite the dependance of normalized spectral IPR with the space dimension, it can be seen also that this indicator does not exhibit any clear trend with the NNN parameter $\Gamma$. We observe that, for small values of $K$, this parameter shows large fluctuations which tend to attenuate when for larger space dimensions. We recall that, for $K\geq 3$, the system exhibits a clear transition from a regular energy spectrum to a more complex (chaotic) one at $\Gamma=0.5$. As can be seen from Fig. \ref{fig:fig3}, there is no evidence of such leap in complexity in our numerical study. In this way, we can assert that the optimal fields required to control the dynamics of regular or chaotic Hamiltonians display a similar spectral complexity.\\ As a final remark, we point out that it would not be correct to claim that the spectral properties analized in this section are completely independent of the choice of initial and final state. This can be easily seen by considering a processes where we intend to connect the ground state of the free Hamiltonian $H_{01}$ and one of its excited states $\Ket{n(\Gamma)}$. If the control Hamiltonian $H_c$ connects both states, we expect that the bandwidth of the control field will be given by the energy difference between both levels $E_n(\Gamma)-E_0(\Gamma)$, which can be signficantly lower than the obtained $\omega_{bw}$ for processes A and B if $\Ket{n}$ lies in the low-energy region. Nevertheless, our results do apply to general linear combinations of energy eigenstates, which is the more common scenario. \section{Discussion} \label{sec:discuss} \begin{figure}[b] \includegraphics[width=\linewidth]{fig4.pdf} \caption{\label{fig:fig4} (color online) Distribution of energy differences $\delta E_{n,m}$ defined in eq. (\ref{ec:levelspm}) for $\Gamma=0$ (regular) and $\Gamma=1$ (chaotic), where $\Gamma$ is the NNN coupling parameter. All cases are normalized such that their mean value is 1 (in units of $J$). In the top left plot, we show the $M=1$ case corresponding to the standard level spacing distribution. Poisson and Wigner-Dyson distributions are shown on top of the histograms. In all cases, we consider the positive subspace with $K=4$ excitations in the chain ($D_{4,+}\simeq 700$). } \end{figure} We will now look more closely at the connection between the spectral features of the optimal control fields and the structural properties of the system spectrum. We have already pointed out in Sec. \ref{sec:model} that the free chain Hamiltonian $H_{01}$ \ref{ec:h01} shows a transition in its level spacing distribution $\{\delta E_n\}$ as the NNN coupling parameter $\Gamma$ changes, where \begin{equation} \delta E_n = E_{n+1}-E_n, \label{ec:levelsp1} \end{equation} \noindent and $E_n$ is the $n$th ordered eigenvalue of $H_{01}$. If the space dimension is high enough ($K\geq 3$), the level spacings statistics show a Poisson distribution for $\Gamma\lesssim0.5$, and a Wigner-Dyson distribution $\Gamma\gtrsim0.5$ (see the Appendix for more details). We have also discussed in Sec. \ref{sec:result} that we observed a connection between the frequency components present in the optimized control field and the resonances of the free Hamiltonian $H_{01}$. Thus, an interesting point arises: if the energy spectrum of $H_{01}$ changes its structure with $\Gamma$, why is there no evidence about those changes in the frequency distribution of the optimal control fields?\\ The key point here is to note that the resonances of $H_{01}$, which feed the frequency distribution of the control field, are not only formed by the difference of two consecutive energies $\delta E_n$ (\ref{ec:levelsp1}). If connected by the control Hamiltonian, every energy difference present in the spectrum is also a suitable candidate for appearing the control field frequency spectrum. Following this discussion, we studied the distribution of the energy differences defined as \begin{equation} \delta E_{n,m} = E_{n+m}-E_n\ \mathrm{with}\ 0<m\leq M, \label{ec:levelspm} \end{equation} \noindent such that $\delta E_{n,1}\equiv \delta E_n$. Note that, for every $n$, the value of $M$ indicates how many levels above $E_n$ are considered, and is thus bounded by the space dimension $D_{K,+}$. In Fig. \ref{fig:fig4} we show the distributions of normalized energy differences for different values of $M$, using $\Gamma=0$ and $\Gamma=1$. There, it can be seen that both distributions show the expected Poisson and Wigner-Dyson shapes when $M=1$ (as discussed in the previous paragraph), but start to converge to a common form when $M$ grows. As an example, for $K=4$, we have that dimension of the positive subspace is $D_{4,+}\simeq700$, and already taking $M\simeq D_{4,+}/10$ already gives near perfectly matching distributions for both values of the chaos parameter $\Gamma$. This analysis indicates that, while level spacing distributions are quite different for regular and chaotic spectra, the overall energy difference distributions converge to a same shape. This interesting behavior determines that the frequency spectrum of the optimal field which control both type of systems have the same complexity.\\ \section{Final remarks} \label{sec:conclu} In this work we studied control processes in a chain of spin-1/2 particles and investigated how the complexity of the physical system relates to the complexity of the control field. We studied a Heisenberg chain model, which allowed us to consider separately different space dimensions (ranging from $\sim10$ to $\sim700$ states) by adding excitations to the system. By allowing next-to-nearest neighbour interactions, we were also able to parametrically tune the system from regular to chaotic. We find the time-dependent control fields required to drive different processes using optimal control theory and defined two measures of complexity based on the Fourier spectrum of those fields. By doing so, we could identify which aspects of the systems complexity affect the control fields. For instance, we found that the spectral bandwidth, which measures the maximimum frequency present in the field, is quite generally independent of the system space dimension. However, we showed that exceptions to this rule ocurr if we choose highly non-local control fields. Also, we investigated how many frequencies present inside the signal bandwidth, by defining a measure of localization: the spectral inverse participation ratio (sIPR). We found that this measure of field complexity does increase when excitations are added to system. Finally, we assesed the role of quantum chaos in the control of the system by studying the fields as a function of the chaos parameter $\Gamma$. We found no evidence of the regular - chaotic transition in the field spectral measures, allowing us to assert that the fields required to control chaotic and integrable systems display the same complexity. Concerning the role of quantum chaos in the dynamics of many-body systems, it is interesting to point out that a previous work \cite{bib:wisn2015} studied relaxation processes in such systems. Although working in an opposite scenario to coherent control, the authors also found no trace of the chaoticity of the system in the relaxation dynamics. In our case, we present further evidence about the irrelevance of quantum chaos in the coherent dynamics of many-body system. \\
1e3ffacba3892591a49facee33fb5eefba8dc145
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The coupling of electronic and nuclear motions plays a significant role in many fascinating scientific phenomena, e.g., photoreactions \cite{photo1,photo2,photo3}, molecular electronics \cite{molel1,molel2,molel3}, and strong-field processes \cite{SFEN1,SFEN2,SFEN3,SFEN4,SFEN5}. In particular, the study of strong-field processes has been one of the most dynamic research areas in the past few decades with the advent of femtosecond and attosecond technology \cite{atto0,atto1,atto2,atto3,atto4,atto5}. Irradiation of atoms and molecules by intense laser pulses gives rise to highly nonlinear effects \cite{SFP1,SFP2,SFP3,SFP4,SFP5,SFP6,SFP7,SFP8,SFP9} such as above-threshold ionization or dissociation, Coulomb explosion, or high-harmonic generation. To understand the mechanisms of any nonlinear molecular phenomenon and non-adiabatic reactions, and to move the technology further forward, it is essential to be able to correctly describe any type of coupled electron-nuclear dynamics. Developing such a theoretical tool has been one of the biggest issues in theoretical physics and chemistry. Numerous studies have been conducted and many sophisticated methods have been developed \cite{enpaper0, enpaper01, enpaper1, enpaper2, enpaper3, enpaper4, enpaper5, enpaper6, enpaper7,enpaper8,enpaper9,enpaper10,enpaper11}, among them, multiple-spawning method \cite{spawn1,spawn2}, multiconfiguraton time-dependent Hartree method \cite{MCTDH1,MCTDH2,MCTDH3,MCTDH4}, and nonadiabatic Bohmian dynamics method \cite{NBD1,NBD2,Conditional} are the methods that retain a quantum description of the nuclei and simulate nonadiabatic electron-nuclear dynamics very accurately. However, they incur huge computational cost when applied to systems with many atoms. Moreover, inclusion of a large number of electronic states is required when higher-intensity fields exist, and ionization processes are very difficult to treat within these approaches. Therefore, alternative approaches have also been developed extensively, which have significantly reduced the computational cost. One of the most widely used approaches is mixed quantum-classical (MQC) approximation, where the nuclei are treated as classical particles, while the electrons are treated quantum mechanically. Among these, the Ehrenfest \cite{Ehrenfest1, Ehrenfest2, Ehrenfest3} and trajectory surface hopping (TSH) \cite{TSH1,TSH2,TSH3,TSH4,TSH5,TSH6,TSH7,TSH8} methods are the most widely used, and have been employed in many studies. However, both Ehrenfest and TSH have certain discrepancies that arise from the fact that, in both methods, the forces acting on classical nuclei and the potential acting on electrons are derived with approximations. There are ongoing intensive efforts to improve these approaches \cite{dec1,dec2,dec3,dec4,dec5,dec6,dec7}. Recently, a new approach to the coupled electron-nuclear motion, the so-called exact factorization of the electron-nuclear wavefunction \cite{AMG,AMG2,*AMG2C,*AMG2R,SAMYG}, has been proposed. This method provides a new route to go beyond the Born-Oppenheimer (BO) approximation \cite{bBO1,bBO2,bBO3,bBO4} and to study the force acting on the classical nuclei \cite{AAYG,AAYG2,AAYMMG}, and then to develop a rigorous MQC method \cite{MQC1,MQC2,MQC3,CT1,CT2}. In this framework, the full wavefunction is written as the product of a nuclear wavefunction and conditional electronic wavefunctions, which parametrically depend on the nuclear configuration. The coupled equations drive the dynamics of these two components, and the motion of the nuclear wavefunction is governed by a single time-dependent Schr\"{o}dinger equation (TDSE), which contains a time-dependent potential-energy surface (TDPES) and a time-dependent vector potential. Since this nuclear wavefunction provides the exact nuclear and current densities, the TDSE that it satisfies has been identified as the exact nuclear TDSE. The presence of a single exact nuclear TDSE has been found to be very useful in developing the MQC approach systematically \cite{AAYG,AAYG2,AAYMMG,MQC1,MQC2,MQC3,CT1,CT2}. In previous studies, the features of the TDPES in a one-dimensional nonadiabatic electron-transfer model system have been fully analyzed \cite{AAYG,AAYG2,Interference}. Indeed, it has been shown that evolving an ensemble of classical nuclear trajectories using the force determined from the gradient of the TDPES reproduces the nuclear wavepacket dynamics very well \cite{AAYMMG}. These led to the idea of developing the MQC method based on the TDPES and multiple trajectories. Recently, a novel MQC algorithm - the coupled-trajectory (CT) MQC algorithm \cite{CT1, CT2} - has been proposed and shown to be able to accurately simulate the coupled electron-nuclear dynamics in a one-dimensional field-free process. On the other hand, the features of the TDPES under external fields have also been studied \cite{AMG,AMG2,*AMG2C,*AMG2R,SAMYG,localization,EAM}. We proposed the {\it reverse} factorization \cite{SAMYG}, which allows us to define the exact electronic TDSE and the exact electronic TDPES. These are found to be very useful for exploring the mechanism of electron dynamics under an external field \cite{SAMYG, EAM}. Furthermore, we have recently studied the nuclear TDPES in laser-induced electron localization in the H$_2^+$ molecule \cite{localization}, and showed that the propagation of an ensemble of classical trajectories using the gradient of the TDPES yields nuclear density dynamics that are very similar to the exact quantum nuclear ones. This result encourages the idea of developing the MQC dynamics method for the strong-field processes as well. This would be useful since none of the methods that presently exist can accurately simulate the coupled electron-nuclear dynamics of medium- and large-sized systems under a strong field. However, it is still not clear whether the gradient of the TDPES can reproduce the quantum nuclear dynamics in strong-field processes such as strong-field ionization and dissociation, where the quantum effects of the nuclei are significant. In fact, the gradient of the TDPES is not exactly the same as the force that appears in the quantum Hamilton-Jacobi equation derived by applying the Bohmian mechanics approach to the exact nuclear TDSE \cite{AAYMMG}; it lacks the force from the so-called Bohmian quantum potential. In previous studies \cite{AMG,AMG2,*AMG2C,*AMG2R}, it was shown that a single classical trajectory evolved by the gradient of the TDPES does not yield the correct time evolution of the mean nuclear distance in strong-field dissociation of the one-dimensional H$_2^+$ molecular model. The question then arises as to whether multiple classical trajectories evolved by the quantum Hamilton-Jacobi equation derived from the exact nuclear TDSE give the correct quantum nuclear dynamics in strong-field processes. In this paper, we show that multiple classical trajectories propagated by the gradient of the TDPES {\it plus} Bohmian quantum potential can reproduce quantum nuclear dynamics in the strong-field processes: it produces the correct dissociation dynamics and splitting of nuclear probability density in the one-dimensional H$_2^+$ molecular model. The Bohmian quantum potential is found to play a non-negligible role in giving the correct nuclear dynamics for the present strong-field processes, where ionization and/or splitting of the nuclear wavepacket occur/s. Our findings provide a useful basis toward the development of the MQC method for strong-field processes. The rest of the paper is organized as follows. In section II, we briefly review the concepts of the exact factorization of the full electron-nuclear wavefunction and the quantum Hamilton-Jacobi equation derived from the exact nuclear TDSE. There, we show the exact force acting on the classical nuclei and its relationship with the gradient of the TDPES. In section III, we first describe our model system of strong-field dissociation of H$_2^+$, and then show the quantum potential in this system together with the exact TDPES. We then propagate multiple classical trajectories with the force from the gradient of the TDPES plus quantum potential and demonstrate that it perfectly reproduces the quantum nuclear dynamics in strong-field processes. We also show the role of the quantum potential by showing the dynamics propagated only by the gradient of the TDPES. In section IV, we summarize the results and speculate on future directions. \section{THEORY} In Ref. \cite{AMG,AMG2,*AMG2C,*AMG2R}, it was shown that the full electron-nuclear wavefunction $\Psi({\underline{\underline{\bf R}}},{\underline{\underline{\bf r}}}, t)$ that solves the TDSE $ \hat{H}\Psi({\underline{\underline{\bf R}}},{\underline{\underline{\bf r}}},t)=i\partial_t\Psi({\underline{\underline{\bf R}}},{\underline{\underline{\bf r}}}, t) $ can be factorized exactly to the single product \begin{equation} \Psi({\underline{\underline{\bf R}}},{\underline{\underline{\bf r}}}, t)=\chi({\underline{\underline{\bf R}}},t)\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t) \label{eqn: factorization} \end{equation} of the nuclear wavefunction $\chi({\underline{\underline{\bf R}}},t)$ and the electronic wavefunction $\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)$ that parametrically depends on the nuclear positions ${\underline{\underline{\bf R}}}$ and satisfies the partial normalization condition \begin{equation} \int d{\underline{\underline{\bf r}}} |\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)|^2=1 \end{equation} for any ${\underline{\underline{\bf R}}}$ and $t$. Throughout this paper, ${\underline{\underline{\bf R}}}$ and ${\underline{\underline{\bf r}}}$ collectively represent the sets of nuclear and electronic coordinates, respectively (i.e., ${\underline{\underline{\bf R}}} \equiv \{ {\bf R}_1,{\bf R}_2,\cdots,{\bf R}_{N_n} \}$ and ${\underline{\underline{\bf r}}} \equiv \{ {\bf r}_1,{\bf r}_2,\cdots,{\bf r}_{N_e} \}$) , and atomic units are used unless stated otherwise. The complete molecular Hamiltonian is \begin{equation} \hat{H} = \hat{T}_{\rm n}({\underline{\underline{\bf R}}})+ \hat{V}^{\rm n}_{\rm ext}({\underline{\underline{\bf R}}},t) +\hat{H}_{\rm BO}({\underline{\underline{\bf r}}},{\underline{\underline{\bf R}}}) +\hat{v}^{\rm e}_{\rm ext}({\underline{\underline{\bf r}}},t), \end{equation} and $\hat{H}_{\rm BO}({\underline{\underline{\bf r}}},{\underline{\underline{\bf R}}})$ is the BO electronic Hamiltonian, \begin{equation} \hat{H}_{\rm BO} = \hat{T}_{\rm e}({\underline{\underline{\bf r}}})+ \hat{W}_{\rm ee}({\underline{\underline{\bf r}}}) +\hat{W}_{\rm en}({\underline{\underline{\bf r}}},{\underline{\underline{\bf R}}})+\hat{W}_{\rm nn}({\underline{\underline{\bf R}}}), \end{equation} where $\hat{T}_{\rm n}=-\sum_{\alpha=1}^{N_{\rm n}}\frac{\nabla^2_\alpha}{2M_\alpha}$ and $\hat{T}_{\rm e}=-\sum_{j=1}^{N_{\rm e}}\frac{\nabla^2_j}{2}$ are the nuclear and electronic kinetic energy operators, $\hat{W}_{\rm ee}$, $\hat{W}_{\rm en}$ and $\hat{W}_{\rm nn}$ are the electron-electron, electron-nuclear and nuclear-nuclear interactions, and $\hat{V}^{\rm n}_{\rm ext}({\underline{\underline{\bf R}}},t)$ and $\hat{v}^{\rm e}_{\rm ext}({\underline{\underline{\bf r}}},t)$ are time-dependent (TD) external potentials acting on the nuclei and electrons, respectively. The stationary variations of the quantum mechanical action with respect to $\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)$ and $\chi({\underline{\underline{\bf R}}},t)$ under the normalization condition of $\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)$ lead to the following equations of motion for $\chi({\underline{\underline{\bf R}}},t)$ and $\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)$ \cite{AMG,AMG2,*AMG2C,*AMG2R}: \begin{equation} \begin{split} \left(\hat{H}_{\rm BO}({\underline{\underline{\bf r}}},{\underline{\underline{\bf R}}})+\hat{v}^{\rm e}_{\rm ext}({\underline{\underline{\bf r}}},t) +\hat U_{\rm en}^{\rm coup}[\Phi_{\underline{\underline{\bf R}}},\chi]-\epsilon({\underline{\underline{\bf R}}},t)\right) \Phi_{{\underline{\underline{\bf R}}}}({\underline{\underline{\bf r}}},t)\\ =i\partial_t \Phi_{{\underline{\underline{\bf R}}}}({\underline{\underline{\bf r}}},t) \end{split}\label{eqn: exact electronic eqn} \end{equation} \begin{equation} \begin{split} \left[\sum_{\alpha=1}^{N_{\rm n}} \frac{\left[-i\nabla_\alpha+{\bf A}_\alpha({\underline{\underline{\bf R}}},t)\right]^2}{2M_\alpha} +\hat{V}^{\rm n}_{\rm ext}({\underline{\underline{\bf R}}},t) + \epsilon({\underline{\underline{\bf R}}},t)\right]\chi({\underline{\underline{\bf R}}},t)\\ =i\partial_t \chi({\underline{\underline{\bf R}}},t) \label{eqn: exact nuclear eqn}. \end{split} \end{equation} Here, $\epsilon({\underline{\underline{\bf R}}},t)$ is the exact nuclear TDPES \begin{equation}\label{eqn: tdpes} \epsilon({\underline{\underline{\bf R}}},t)=\left\langle\Phi_{\underline{\underline{\bf R}}}(t)\right|\hat{H}_{\rm BO}+\hat{v}^{\rm e}_{\rm ext}({\underline{\underline{\bf r}}},t) +\hat U_{\rm en}^{\rm coup}-i\partial_t\left| \Phi_{\underline{\underline{\bf R}}}(t)\right\rangle_{\underline{\underline{\bf r}}}, \end{equation} $\hat U_{\rm en}^{\rm coup}[\Phi_{\underline{\underline{\bf R}}},\chi]$ is the electron-nuclear coupling operator, \begin{align} \hat U_{\rm en}^{\rm coup}&[\Phi_{\underline{\underline{\bf R}}},\chi]=\sum_{\alpha=1}^{N_{\rm n}}\frac{1}{M_\alpha}\left[ \frac{\left[-i\nabla_\alpha-{\bf A}_\alpha({\underline{\underline{\bf R}}},t)\right]^2}{2} \right.\label{eqn: enco} \\ & \left.+\left(\frac{-i\nabla_\alpha\chi}{\chi}+{\bf A}_\alpha({\underline{\underline{\bf R}}},t)\right) \left(-i\nabla_\alpha-{\bf A}_{\alpha}({\underline{\underline{\bf R}}},t)\right)\right],\nonumber \end{align} and ${\bf A}_{\alpha}\left({\underline{\underline{\bf R}}},t\right)$ is the TD vector potential, \begin{equation}\label{eqn: vector potential} {\bf A}_{\alpha}\left({\underline{\underline{\bf R}}},t\right) = \left\langle\Phi_{\underline{\underline{\bf R}}}(t)\right|-i\nabla_\alpha\left.\Phi_{\underline{\underline{\bf R}}}(t) \right\rangle_{\underline{\underline{\bf r}}}. \end{equation} The symbol $\left\langle\,\,\cdot\,\,\right\rangle_{\underline{\underline{\bf r}}}$ indicates an integration over electronic coordinates only. The partial normalization condition of $\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)$ makes the factorization~(\ref{eqn: factorization}) unique up to within a $({\underline{\underline{\bf R}}},t)$-dependent gauge transformation, $ \chi({\underline{\underline{\bf R}}},t)\rightarrow\tilde\chi({\underline{\underline{\bf R}}},t)=e^{-i\theta({\underline{\underline{\bf R}}},t)}\chi({\underline{\underline{\bf R}}},t)$ and $ \Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)\rightarrow\tilde\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)=e^{i\theta({\underline{\underline{\bf R}}},t)}\Phi_{\underline{\underline{\bf R}}}({\underline{\underline{\bf r}}},t)$. Eqs. (\ref{eqn: exact electronic eqn}) and (\ref{eqn: exact nuclear eqn}) are form invariant under this transformation while the scalar potential and the vector potential transform as $ \tilde{\epsilon}({\underline{\underline{\bf R}}},t) = \epsilon({\underline{\underline{\bf R}}},t)+\partial_t\theta({\underline{\underline{\bf R}}},t)$ and $\tilde{\bf A}_{\alpha}({\underline{\underline{\bf R}}},t) = {\bf A}_{\alpha}({\underline{\underline{\bf R}}},t)+\nabla_\alpha\theta({\underline{\underline{\bf R}}},t)$. The equation for $\chi({\underline{\underline{\bf R}}},t)$, Eq.~(\ref{eqn: exact nuclear eqn}), has the form of a Schr\"odinger equation. Note that $\chi({\underline{\underline{\bf R}}},t)$ can be interpreted as the exact nuclear wave-function since it leads to an $N$-body nuclear density, $\Gamma({\underline{\underline{\bf R}}},t)=\vert\chi({\underline{\underline{\bf R}}},t)\vert^2,$ and an $N$-body current density, ${\bf J}_\alpha({\underline{\underline{\bf R}}},t)=\frac{1}{M_\alpha}\Big[\mbox{Im}(\chi^*({\underline{\underline{\bf R}}},t)\nabla_\alpha\chi({\underline{\underline{\bf R}}},t))+ \Gamma({\underline{\underline{\bf R}}},t){\bf A}_\alpha({\underline{\underline{\bf R}}},t)\Big],$ which yields the true nuclear $N$-body density and current density obtained from the full wavefunction $\Psi({\underline{\underline{\bf R}}},{\underline{\underline{\bf r}}},t)$ \cite{AMG2,*AMG2C,*AMG2R}. Therefore, the equation~(\ref{eqn: exact nuclear eqn}) can be regarded the {\it exact} nuclear TDSE. Having a single exact TDSE for the nuclear subsystem, it is possible to consider its hydrodynamic reformulation~\cite{AAYMMG} using the approach of Bohmian mechanics \cite{Bohm0, Bohm1, Bohm2}, to study how the exact force acting on the classical nuclei can be defined. To this end, the polar forms of the wavefunction $\chi({\underline{\underline{\bf R}}},t)=\vert\chi({\underline{\underline{\bf R}}},t)\vert e^{iS({\underline{\underline{\bf R}}},t)}$ ($\vert\chi({\underline{\underline{\bf R}}},t)\vert$ and $S({\underline{\underline{\bf R}}},t)$ are real-valued amplitude and action functions, respectively) are substituted into the exact nuclear TDSE~(\ref{eqn: exact nuclear eqn}). Here, to easily find the exact force acting on the classical nuclei, we set the gauge of the wavefunction such that the vector potential ${\bf A}_\alpha({\underline{\underline{\bf R}}},t)$ is always zero. Note that whenever the vector potential is curl-free ($\nabla_\alpha\times{\bf A}_\alpha({\underline{\underline{\bf R}}},t)=0$), the gauge can be chosen such that ${\bf A}_\alpha({\underline{\underline{\bf R}}},t)$ is zero. Whether and under which conditions $\nabla_\alpha\times{\bf A}_\alpha({\underline{\underline{\bf R}}},t)=0$ is currently under investigation \cite{Gphase1, Gphase2}. Under this choice of the gauge, the exact nuclear TDSE~(\ref{eqn: exact nuclear eqn}) can be written as \begin{equation} \left[-\sum_{\alpha=1}^{N_n} \frac{\nabla_\alpha^2}{2M_\alpha} + \epsilon({\underline{\underline{\bf R}}},t)\right]\chi({\underline{\underline{\bf R}}},t) =i\partial_t \chi({\underline{\underline{\bf R}}},t) \label{eqn: exact nuclear eqn2}. \end{equation} Here we include $\hat{V}^{\rm n}_{\rm ext}({\underline{\underline{\bf R}}},t)$ in the exact TDPES $\epsilon({\underline{\underline{\bf R}}},t)$. Substituting $\chi({\underline{\underline{\bf R}}},t)=\vert\chi({\underline{\underline{\bf R}}},t)\vert e^{iS({\underline{\underline{\bf R}}},t)}$ into Eq.~(\ref{eqn: exact nuclear eqn2}), the following two coupled equations are obtained \cite{AAYMMG}: \begin{equation} \begin{split} \sum_{\alpha=1}^{N_{\rm n}}\frac{(\nabla_\alpha S({\underline{\underline{\bf R}}},t))^2}{2M_\alpha}+\epsilon({\underline{\underline{\bf R}}},t) -\sum_{\alpha=1}^{N_{\rm n}}\frac{1}{2M_\alpha}\frac{\nabla_\alpha^2 \vert\chi({\underline{\underline{\bf R}}},t)\vert}{\vert\chi({\underline{\underline{\bf R}}},t)\vert}\\ =-\frac{\partial S({\underline{\underline{\bf R}}},t)}{\partial t} \label{eqn: QHJeqn} \end{split} \end{equation} and \begin{equation} \begin{split} -\sum_{\alpha=1}^{N_{\rm n}}\frac{\nabla_\alpha S({\underline{\underline{\bf R}}},t)\cdot\nabla_\alpha \vert\chi({\underline{\underline{\bf R}}},t)\vert}{M_\alpha} -\sum_{\alpha=1}^{N_{\rm n}}\frac{\vert\chi({\underline{\underline{\bf R}}},t)\vert\nabla_\alpha^2 S({\underline{\underline{\bf R}}},t)}{2M_\alpha}\\ =\frac{\partial \vert\chi({\underline{\underline{\bf R}}},t)\vert}{\partial t}. \label{eqn: continuity} \end{split} \end{equation} Equation (\ref{eqn: QHJeqn}) and (\ref{eqn: continuity}) are the hydrodynamic formulation of the exact nuclear TDSE~(\ref{eqn: exact nuclear eqn2}). Equation~(\ref{eqn: QHJeqn}) can be regarded as the {\it exact} quantum Hamilton-Jacobi equation, while~(\ref{eqn: continuity}) produces the continuity equation. Identifying $\nabla_\alpha S({\underline{\underline{\bf R}}},t)$ as a momentum ${\bf P}_\alpha$ of a classical trajectory, (\ref{eqn: QHJeqn}) can be solved by propagating an ensemble of classical trajectories that obey the following Newton's equations: \begin{equation} \begin{split} \frac{d {\bf P}_\alpha}{dt}&=-\nabla_\alpha \left[ \epsilon({\underline{\underline{\bf R}}},t) - \sum_{\alpha=1}^{N_{\rm n}}\frac{1}{2M_\alpha}\frac{\nabla_\alpha^2 \vert\chi({\underline{\underline{\bf R}}},t)\vert}{\vert\chi({\underline{\underline{\bf R}}},t)\vert}\right]\\ &=-\nabla_\alpha \left[\epsilon({\underline{\underline{\bf R}}},t)+\epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)\right]. \label{eqn: Newton} \end{split} \end{equation} The right-hand side of (\ref{eqn: Newton}) can now be considered as the exact force acting on the classical nuclei, since it is derived from the exact nuclear TDSE. It is a gradient of the sum of the exact TDPES $\epsilon({\underline{\underline{\bf R}}},t)$ and the additional time-dependent potential \begin{equation} \epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)=-\sum_{\alpha=1}^{N_{\rm n}}\frac{1}{2M_\alpha} \frac{\nabla_\alpha^2 \vert\chi({\underline{\underline{\bf R}}},t)\vert}{\vert\chi({\underline{\underline{\bf R}}},t)\vert}, \label{eqn: eQP} \end{equation} which is referred to as the quantum potential in Bohmian mechanics. In the previous studies, we propagated multiple classical trajectories according to (\ref{eqn: Newton}) {\it without} this Bohmian quantum potential $\epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)$, i.e., \begin{equation} \begin{split} \frac{d {\bf P}_\alpha}{dt}=-\nabla_\alpha \epsilon({\underline{\underline{\bf R}}},t) \label{eqn: Newton2} \end{split} \end{equation} for the field-free nonadiabatic charge-transfer process \cite{AAYMMG} and the laser-induced electron localization processes in the H$_2^+$ molecule \cite{localization}. We found that an ensemble of independent classical nuclear trajectories on $\epsilon({\underline{\underline{\bf R}}},t)$ provides dynamics that accurately reproduce the exact nuclear wavepacket dynamics. Here, we study whether the same multiple classical trajectory approach can also reproduce strong-field processes in which ionization and/or splitting of nuclear density occur/s. In such strong-field processes, nuclear quantum effects are significant; in fact, previous studies \cite{AMG,AMG2,*AMG2C,*AMG2R} have shown that a single classical trajectory cannot yield the molecular dissociation via tunneling that occurs under the strong field, even though it is propagated by the force of the gradient of the exact TDPES, $\epsilon({\underline{\underline{\bf R}}},t)$. In the next section, we will propagate multiple classical trajectories by $\epsilon({\underline{\underline{\bf R}}},t)$ according to~(\ref{eqn: Newton2}). We will also calculate the exact quantum potential $\epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)$ and propagate multiple classical trajectories by $\epsilon({\underline{\underline{\bf R}}},t)+\epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)$, i.e., (\ref{eqn: Newton}), to study the importance of the force from $\epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)$. Note this study demonstrates the role of $\epsilon^{\rm QP}({\underline{\underline{\bf R}}},t)$ in multiple classical-trajectories dynamics for the first time. \section{Results and discussion} \subsection{Theoretical model} To study whether the propagation of multiple classical trajectories can reproduce the exact quantum nuclear dynamics in strong laser fields, we employ a simplified model of the H$_2^+$ molecule, which is the same as that used in previous studies \cite{SFP4,AMG,AMG2,*AMG2C,*AMG2R,SAMYG,localization,EAM}. In this model, the dimensionality of the problem is reduced by restricting the motion of the nuclei and the electron to the direction of the polarization axis of the laser field \cite{1DM0, 1DM1, 1dM2, 1DM3}. In the center-of-mass system, the dynamics of this one-dimensional model of H$_2^+$ is governed by the full Hamiltonian $\hat{H}(R,r,t) =\hat{T}_{\rm n}(R)+\hat{T}_{\rm e}(r)+\hat{W}_{\rm nn}(R)+\hat{W}_{\rm en}(R,r)+\hat{v}_{\rm laser}(r,t) $, where $R$ is the internuclear distance and $r$ is the electronic coordinate as measured from the nuclear center of mass. The kinetic energy terms are $\hat{T}_{\rm n}(R) = -\frac{1}{2\mu_n}\frac{\partial^2}{\partial R^2}$ and, $\hat{T}_{\rm e}(r) = -\frac{1}{2\mu_{\rm e}}\frac{\partial^2}{\partial r^2}$, respectively, where the reduced mass of the nuclei is given by $\mu_{\rm n}=M_{\rm H}/2$, and reduced electronic mass is given by $\mu_{\rm e}=\frac{2M_{\rm H}}{2M_{\rm H}+1}$ ($M_{\rm H}$ is the proton mass). The interactions are soft-Coulomb: $\hat{W}_{\rm nn}(R) = \frac{1}{\sqrt{0.03+R^2}}$, and $\hat{W}_{\rm en}(R,r) = -\frac{1}{\sqrt{1.0+(r-\frac{R}{2})^2}} -\frac{1}{\sqrt{1.0+(r+\frac{R}{2})^2}}$ (and $\hat{W}_{\rm ee} = 0$). The field is described within the dipole approximation and length gauge, as $\hat{v}_{\rm laser}(r,t) = E(t)q_{\rm e}r$, where the reduced charge $q_{\rm e}=\frac{2M_{\rm H}+2}{2M_{\rm H}+1}$. This reduced-dimensional model has proven useful since it allows numerically exact solutions to the TDSE while capturing the essential physics in strong-field processes such as multiphoton ionization, above-threshold ionization and dissociation, enhanced ionization, non-sequential double ionization, and high-harmonic generation \cite{SFP1,SFP2,SFP3,SFP4,SFP5,SFP6,SFP7,SFP8,SFP9}. In this study, we investigate the dynamics of the model H$_2^+$ system under a $\lambda=228$ nm ($5.4$ eV) UV-laser pulse, which is represented by $ E(t)=E_0f(t)\sin(\omega t), $ with two peak intensities, $I_1 =\vert E_0\vert^2=10^{14}$ W/cm$^2$ and $I_2 =\vert E_0\vert^2=2.5\times10^{13}$ W/cm$^2$. This frequency provides an energy that is about twice as much as the dissociation energy of the model molecule ($2.88$ eV); thus, dissociation is expected. The envelope function $f(t)$ is chosen such that the field is linearly ramped from zero to its maximum strength at $t=7.6$ fs (over 10 optical cycles) and thereafter, held constant for an additional 15 laser cycles, corresponding to a total simulation time of about 19 fs. The same system and parameters were employed in previous studies \cite{SFP4,AMG,AMG2,*AMG2C,*AMG2R}, where the important role of the complex coupling between the electronic and nuclear motions in these strong-field systems was revealed. In \cite{AMG,AMG2,*AMG2C,*AMG2R}, in particular, the exact TDPES~(\ref{eqn: tdpes}) in these systems was calculated and shown to be a very useful tool for analyzing and interpreting the complicated quantum nuclear dynamics in the strong-field processes. Here, we will study the possibility of the Bohmian mechanics being established for these strong-field processes by using the concept of exact factorization, and whether multiple classical trajectories can give the correct quantum nuclear dynamics. We first calculated the full molecular wavefunction $\Psi(R,r,t)$ by propagating the full TDSE \begin{equation} \hat{H}(R,r,t)\Psi(R,r,t)=i\partial_t\Psi(R,r,t) \label{eqn: fullTDSE} \end{equation} numerically exactly using the second-order split-operator method \cite{SPO}. As the initial state of the time propagation, $\Psi(R,r,t)$ was prepared in its ground state by imaginary-time propagation. \begin{figure}[h] \centering \includegraphics*[width=1.0\columnwidth]{Figure1.pdf} \caption{(color online). Electron-nuclear density $|\Psi(R,r,t)|^2$ for the model H$_2^+$ molecule after the 24th optical cycle ($t=18.6$ fs) in a $\lambda=228$ nm laser field. (Upper panel: higher-intensity case ($I_1 =10^{14}$ W/cm$^2$). Lower panel: lower-intensity case ($I_2 =2.5\times10^{13}$ W/cm$^2$.) In atomic units.} \label{fig:Fig1} \end{figure} In Fig.~\ref{fig:Fig1}, the electron-nuclear density $|\Psi(R,r,t)|^2$ at $t=18.6$ fs (after the 24th optical cycle) is shown for both the higher-intensity case ($I_1 =10^{14}$ W/cm$^2$) (upper panel) and the lower-intensity case ($I_2 =2.5\times10^{13}$ W/cm$^2$) (lower panel). These indicate the probability of finding an electron at position $r$ and the nuclear separation at position $R$ at $t=18.6$ fs for each case. In the upper panel, it is observed that $|\Psi(R,r,t)|^2$ at $t=18.6$ fs exists at larger $R$ compared to the expectation value at the ground state $\langle R \rangle(t=0)=2.65$ a.u., indicating that dissociation occurred. We also observe large streaks of $|\Psi(R,r,t)|^2$ in both negative and positive $r$ directions, which shows that a considerable ionization occurred in this higher-intensity case. Therefore, dissociation occurs here via the Coulomb-explosion mechanism, as already discussed in previous studies \cite{SFP4,AMG,AMG2,*AMG2C,*AMG2R}. However, in the lower-intensity case (lower panel in Fig.~\ref{fig:Fig1}), different dynamics occurred: a small amount of $|\Psi(R,r,t)|^2$ exists in the region larger than $\langle R \rangle(t=0)=2.65$ a.u., but a large part of it remains in the ground-state position around $\langle R \rangle(t=0)$. Therefore, a splitting of probability density occurred here and a small amount of $|\Psi(R,r,t)|^2$ went to dissociation. It is also seen that the probability of ionization is very low, and hence, dissociation occurred predominantly via the photodissociation channel (H$^+_2\rightarrow$H$+$H$^+$) \cite{SFP4,AMG,AMG2,*AMG2C,*AMG2R}. Previous studies \cite{SFP4,AMG,AMG2,*AMG2C,*AMG2R} have shown that this lower-intensity case ($I_2 =2.5\times10^{13}$ W/cm$^2$) represents a particularly challenging system when we consider simulating it by using the approximated method. The Ehrenfest and time-dependent Hartree methods could not reproduce the probability of dissociation of this system. A more sophisticated correlated time-dependent variational approach \cite{SFP4} succeeded in giving the dissociation probability to some degree, but still could not reproduce the nuclear density dynamics well. The exact TDPES in this system \cite{AMG,AMG2,*AMG2C,*AMG2R} provided a clear picture that explains the difficulty in the simulation of these nuclear dynamics: it was shown that the quantum tunneling through the TDPES occurs, which is difficult to reproduce by the approximated methods. The question arises as to whether we get the force acting on classical nuclei that gives the correct dynamics in this system if we formulate Bohmian mechanics in the exact-factorization framework. \subsection{Multiple classical trajectory dynamics on the exact TDPES + the exact quantum potential} \begin{figure}[] \centering \includegraphics*[width=1.0\columnwidth]{Figure2.pdf} \caption{(color online). (a) Top panel: Snapshots of the exact TDPES $\epsilon(R,t)$ (blue or dark gray solid lines) and nuclear density (black solid lines) at times indicated, for H$^+_2$ subjected to the laser-field ($\lambda=228$ nm) with the peak intensity $I_1 =10^{14}$ W/cm$^2$. For reference, the ground-state BO potential-energy surface (black dashed lines) is also shown. Bottom panel: Snapshots of the exact quantum potential $\epsilon^{\rm QP}$ (pink or light gray solid lines) for the same system as in the top panels at the same times. Nuclear density is again shown (black solid lines). (b) Same as (a) but for the case of lower intensity ($I_2 =2.5\times10^{13}$ W/cm$^2$).} \label{fig:Fig2} \end{figure} To answer the questions raised in the previous sections, we now present our results. We begin with the calculation of the exact TDPES $\epsilon(R,t)$~(\ref{eqn: tdpes}) and the quantum potential $\epsilon^{\rm QP}(R,t)$~(\ref{eqn: eQP}) that appear in the exact quantum Hamilton-Jacobi equation~(\ref{eqn: QHJeqn}). Since we already have numerically exact $\Psi(R,r,t)$ at each time step obtained by propagating the full TDSE~(\ref{eqn: fullTDSE}), we can easily calculate the TDPES $\epsilon(R,t)$ by fixing the gauge \cite{AMG2,*AMG2C,*AMG2R} and $\epsilon^{\rm QP}(R,t)$ with $|\chi(R,t)|=\sqrt{\int{|\Psi(R,r,t)|^2 dr}}$. In Fig.~\ref{fig:Fig2} (a) and (b), we show snapshots of the exact TDPES $\epsilon(R,t)$ (blue or dark gray solid lines) and the quantum potential $\epsilon^{\rm QP}(R,t)$ (pink or light gray solid lines) at times indicated for the system with peak intensities $I_1 =10^{14}$ W/cm$^2$ (Fig.~\ref{fig:Fig2} (a)) and $I_2 =2.5\times10^{13}$ W/cm$^2$ (Fig.~\ref{fig:Fig2} (b)), respectively. The nuclear density at each time is also shown (black solid lines). Note that the TDPES have already been reported in previous studies \cite{AMG,AMG2,*AMG2C,*AMG2R}, and here we show the quantum potential for the first time. The TDPES in each system shows the characteristic feature of the time-dependent potentials that the nuclear wavepacket experiences. In the Coulomb explosion case (Fig.~\ref{fig:Fig2}(a)), the wells in the TDPES $\epsilon(R,t)$ that confine the wavefunction in the ground state flatten out as the laser is switched on, causing the nuclear density to spill out to larger separations. However, in the photodissociation (without ionization) case (Fig.~\ref{fig:Fig2} (b)), the well in the $\epsilon(R,t)$ exists at all times, and thus, the nuclear density can leak out from it only by tunneling. \begin{figure*}[] \centering \includegraphics*[width=2.0\columnwidth]{Figure3.pdf} \caption{(color online). (a) Snapshots of the nuclear density reconstructed as a histogram from the distribution of the classical positions obtained by solving Newton's equation~(\ref{eqn: Newton}) for H$^+_2$ subjected to the laser field ($\lambda=228$ nm) with peak intensity $I_1 =10^{14}$ W/cm$^2$ at indicated times (red or gray linepoints). Exact nuclear density is also shown (black solid line). Green or light gray dashed line indicates the sum of the exact TDPES and quantum potential ($\epsilon(R,t) + \epsilon^{\rm QP}(R,t)$) whose gradient is the force used to obtain the red or gray linepoints. Blue or dark gray solid line is $\epsilon(R,t)$ and black dashed line is the ground-state BO potential-energy surface. (b) Same as (a), but for the case of lower intensity ($I_2 =2.5\times10^{13}$ W/cm$^2$). } \label{fig:Fig3} \end{figure*} By comparing the upper and lower panels in both laser cases, we find that $\epsilon^{\rm QP}(R,t)$ has non-negligible structures, especially at earlier simulation times. It is found that $\epsilon^{\rm QP}(R,t)$ actually has an effect that flattens the well in $\epsilon(R,t)$, thus helping the nuclei delocalize against the force from the confining potential at the initial time and leak out from it. Especially at the initial time when the wavefunction is in its ground state, the shape of $\epsilon^{\rm QP}(R,t=0)$ is exactly opposite to that of TDPES $\epsilon(R,t=0)$ as seen in the left-hand panels of Figs.~\ref{fig:Fig2} (a) and (b). This is understood by the fact that the ground state of a one-dimensional wavefunction is expressed by a real function multiplied by a complex constant, i.e., $\chi(R)=|\chi(R)|e^{ia}$ where $a$ is a real constant. Since the exact nuclear Schr\"{o}dinger equation in the ground state is written as \cite{Gphase1,staticEF} \begin{equation} \left[- \frac{1}{2\mu_n}\frac{d^2}{dR^2} + \epsilon(R)\right]\chi(R) =E \chi(R) \label{eqn: static nuclear eqn} \end{equation} (where $E$ is the total energy of the system in the ground-state), the quantum potential at the initial time can be written as \begin{equation} \epsilon^{\rm QP}(R)=\frac{-\frac{1}{2\mu_{\rm n}}\frac{d^2}{dR^2}|\chi(R)|}{|\chi(R)|}=E-\epsilon(R). \label{eqn: static eQP} \end{equation} Thus, $\epsilon^{\rm QP}(R)$ shows the opposite curvature to the exact TDPES $\epsilon(R)$. We now demonstrate that the effect of $\epsilon^{\rm QP}$ on the dynamics is significant, especially for the low-intensity case, as it causes the tunneling of the classical nuclei. Having the exact TDPES $\epsilon(R,t)$ and the quantum potential $\epsilon^{\rm QP}(r,t)$ at each time step, we now solve the {\it exact} quantum Hamilton-Jacobi equation~(\ref{eqn: QHJeqn}) for these two systems by propagating multiple classical trajectories according to Newton's equation ~(\ref{eqn: Newton}). We first show the results for the higher-intensity ($I_1 =10^{14}$ W/cm$^2$) case. We propagate 1000 trajectories according to Eq.~(\ref{eqn: Newton}), where the initial positions $R$ are sampled from the initial (ground-state) nuclear density $N(R,t=0)=\int{|\Psi(R,r,t=0)|^2 dr}$, which was obtained in the previous TDSE calculation, and the initial momentum is set to zero since $\frac{d}{dR}S(R,t)$ is zero at the initial time. In Fig.~\ref{fig:Fig3} (a), the nuclear density reconstructed as a histogram from the distributions of classical positions evolving on $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ for the system with the peak intensity $I_1 =10^{14}$ W/cm$^2$ is shown as red (or gray) linepoints. The exact nuclear density obtained from the full TDSE~(\ref{eqn: fullTDSE}) is also shown as a black solid line. By comparing the red (or gray) linepoints and black line, we find that the nuclear density obtained from the multiple classical trajectories yields the exact quantum nuclear dynamics in this strong-field process: the characteristic Coulomb-explosion dynamics of the quantum nuclei are reproduced by the ensemble of classical trajectories. This result indicates that even strong-field processes in which ionization occurs can be simulated, in principle, by the MQC approximation method. We also show $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ (green or light gray dashed line) and $\epsilon(R,t)$ (blue or dark gray solid line). By comparing these, the discussion above is confirmed: $\epsilon^{\rm QP}$ plays a role to flatten the well in $\epsilon(R,t)$, which initially confines the wavepacket to the equilibrium position, thus enhancing dissociation. Here, in the higher-intensity ($I_1 =10^{14}$ W/cm$^2$) case, together with the TDPES that gives a strong repulsive force that reflects ionization, almost the entire nuclear density leaks out from the well and moves to the dissociation. We observe that, when the nuclear density moves outside the well, the effect of $\epsilon^{\rm QP}$ becomes very small. Thus, in this higher-intensity case, the quantum potential affects the dynamics only during the earlier time of the propagation when the nuclear density is about to leak out from the well. Next, we turn to the lower-intensity case ($I_2 =2.5\times10^{13}$ W/cm$^2$), where tunneling of the nuclear wavepacket occurs and causes the difficulty to be simulated by the approximated method \cite{SFP4,AMG,AMG2,*AMG2C,*AMG2R}. We propagate 2,000 trajectories according to ~(\ref{eqn: Newton}) with $\epsilon(R,t)$ plus $\epsilon^{\rm QP}(R,t)$ calculated above. Snapshots of the nuclear density reconstructed as a histogram from the distributions of classical positions are plotted as red (or gray) linepoints in Fig.~\ref{fig:Fig3} (b). Comparison with the exact nuclear density (black solid line in the same figure) again shows excellent agreement. Multiple classical trajectories evolving on $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ perfectly reproduce the splitting of the exact nuclear density, i.e., long-reaching tails of the exact nuclear density are correctly reproduced. Thus, it is found that the strong-field photodissociation dynamics can also be simulated, in principle, by multiple classical trajectories when their motion is driven by the correct force, i.e., the gradient of $(\epsilon(R,t)+\epsilon^{\rm QP}(R,t))$. In the same figure, $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ and $\epsilon(R,t)$ are also plotted. Similar to the higher-intensity case, $\epsilon^{\rm QP}(R,t)$ flattens the well in $\epsilon(R,t)$; thus, it allows the nuclear density to escape from the confining potential. This explains how quantum tunneling is reproduced by classical-trajectory dynamics: the quantum potential is responsible for it. Here, a large part of the nuclear density remains in the ground-state position and the exact TDPES $\epsilon(R,t)$ has a confining well at all times. However, the quantum potential $\epsilon^{\rm QP}(R,t)$ has the opposite curve to $\epsilon(R,t)$ around the equilibrium position. Thus, in total, $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ always shows a somewhat flattened structure and some classical nuclei can experience the force to escape the well and go to the dissociation. Note that once the nuclei leak outside the well, the quantum potential $\epsilon^{\rm QP}(R,t)$ has an almost negligible effect on their dynamics. \begin{figure}[] \centering \includegraphics*[width=1.0\columnwidth]{Figure4.pdf} \caption{(color online). (a) Snapshots of the nuclear density reconstructed as a histogram from the distribution of the classical positions at indicated times for the case where H$^+_2$ is subjected to the higher-intensity ($I_1 =10^{14}$ W/cm$^2$) laser field ($\lambda=228$ nm). The red or light gray linepoints represent the density obtained by propagating classical trajectories on $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ , while the blue or dark gray linepoints indicate the results obtained by propagating classical trajectories only on $\epsilon(R,t)$. Exact nuclear density is also shown as a black solid line. (b) Time evolution of the mean inter-nuclear distance $\langle R\rangle(t)$ obtained for the same system as in (a) from three different simulations: propagation of classical trajectories on $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ (red or light gray solid line), propagation of classical trajectories only on $\epsilon(R,t)$ (blue or dark gray solid line), and the exact TDSE solution (black solid line). (c/d) Same as (a/b) but for the case of lower intensity ($I_2 =2.5\times10^{13}$ W/cm$^2$). } \label{fig:Fig4} \end{figure} Finally, to confirm if the above discussion about the role of the quantum potential $\epsilon^{\rm QP}(R,t)$ is correct, we compute the multiple classical trajectory dynamics without taking account of the quantum potential, i.e., propagated according to ~(\ref{eqn: Newton2}). In Fig.~\ref{fig:Fig4} (a) and (c), we show snapshots of the calculated dynamics: the blue (or dark gray) linepoints indicate the nuclear density reconstructed as a histogram from the distribution of the classical positions evolving only on $\epsilon(R,t)$. ((a) shows the results for the higher-intensity case and (c) is for the lower-intensity one). As a reference, the densities obtained from the multiple classical trajectories on $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ (red or light gray linepoints) and the exact nuclear density (black solid line) are also shown. To see the difference between these results more clearly, in Figs.~\ref{fig:Fig4} (b) and (d), we also plot the time evolution of the mean inter-nuclear distance $\langle R\rangle(t)$ obtained from each calculation ((b) is for the higher-intensity case and (d) is for the lower-intensity one). Exact results (black solid line) are obtained by $\langle R\rangle(t)=\langle\Psi(R,r,t)|\hat{R}|\Psi(R,r,t)\rangle$ while $\langle R\rangle(t)$ from classical trajectories (red or light gray solid line indicates the dynamics on $\epsilon(R,t)+\epsilon^{\rm QP}(R,t)$ and blue or dark gray solid line indicates the dynamics only on $\epsilon(R,t)$) are obtained as $\langle R\rangle(t)=\frac{1}{N_{\rm traj}}\sum^{N_{\rm traj}}_{I=1}R_I(t)$, where $N_{\rm traj}$ is the total number of trajectories and $R_I(t)$ is the distance at time $t$ of each trajectory. Comparison between the blue (or dark gray) and red (or light gray) lines clearly reveals the failure in the dynamics propagated only by the force from $\epsilon(R,t)$. In the higher-intensity case, the dynamics driven only by $\epsilon(R,t)$ succeeded in reproducing the dissociation ones, but their speed is lower than the exact result; the shape of the nuclear density (blue or dark gray) is more localized in the smaller $R$ region than the exact density. This is easily understood because the quantum potential has the effect of flattening the well of the ground-state potential, as shown above, and its absence makes the nuclear density likely to be trapped at the equilibrium position, leading to slower dissociation. In the lower-intensity case, failure in the dynamics propagated only by the force from $\epsilon(R,t)$ is more noticeable: the majority of the nuclear density shown as the blue (or dark gray) curve is trapped by the confining potential well at the equilibrium position, and only a very small fraction leaks out to the dissociation, even in the final snapshot ($t=18.6$ fs). The time evolution of $\langle R\rangle(t)$ in Fig.~\ref{fig:Fig4} (d) clearly reflects this: the speed of the dissociation in the dynamics propagated only by $\epsilon(R,t)$ is much lower than that in the exact dynamics. This result is also supported by the above analysis, which shows that the quantum potential would have flattened the well and caused tunneling if it had been included in the force. From these results, we can conclude that the quantum potential plays a non-negligible role in reproducing the exact quantum nuclear dynamics in the strong-field processes studied here by propagating an ensemble of classical trajectories according to the Newton's equation~(\ref{eqn: Newton}) derived from Bohmian mechanics in the exact nuclear TDSE~(\ref{eqn: exact nuclear eqn}). This suggests that careful assessment of its effect is required when developing the MQC method based on the exact nuclear TDSE~(\ref{eqn: exact nuclear eqn}), especially when we aim to develop the method for strong-field processes. \section{Conclusions} In this paper, we have demonstrated that the propagation of multiple classical trajectories can reproduce quantum nuclear dynamics even in strong-field processes when they are propagated by Newton's equation with the force determined from the gradient of the exact TDPES {\it plus} the exact quantum potential, which is defined in Bohmian mechanics in the exact-factorization framework. We employed a one-dimensional H$_2^+$ model system subject to two different-intensity laser fields, which give rise to different types of dissociation dynamics, and showed that both processes can be simulated by multiple classical trajectory dynamics. We found that the exact quantum potential has a non-negligible effect on the classical dynamics: its force accelerates the classical nuclei to overcome the confining well of the ground-state potential-energy surface, and causes tunneling of the nuclear density in the lower-intensity case. The results here offer important knowledge for developing the MQC algorithm for coupled electron-nuclear dynamics based on the exact-factorization approach, especially when one wants to develop it for strong-field processes. Although it is a challenging task to develop suitable approximations to the quantum Hamilton-Jacobi equation (\ref{eqn: QHJeqn}) and make it solvable on-the-fly, our results are encouraging since they show that quantum nuclear dynamics can be simulated, in principle, by classical trajectories if the force is properly prepared, and the existence of such exact forces is proved. In this study, we explored the case where the gauge is fixed such that the vector potential in the exact nuclear TDSE vanishes. It is also desirable to formulate Bohmian mechanics in a more general gauge, i.e., where the vector potential exists and the gauge-dependent part of the TDPES vanishes. This would provide a more general basis for developing an MQC method that can yield the correct quantum nuclear-density dynamics in various non-adiabatic situations including strong-field processes. {\it Acknowledgments:} This study was supported by JSPS KAKENHI Grant No. 16K17768.
bb081fabeebb0fdb8e3d1f6ede6f2fa306b9e013
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} In robotics motion planning and control, one major challenge is to specify a high level task for a robot without specifying how to solve this task. For legged locomotion such a task include a goal position to reach or to manipulate an object without specifying gaits, contacts, balancing or other behaviors. Trajectory Optimization recently gained a lot of attention in robotics research since it promises to solve some of these problems. It could potentially solve complex motion planning tasks for robots with many degrees of freedom, leveraging the full dynamics of the system. Yet, there are two challenges of optimization optimization. Firstly, Trajectory Optimization problems are hard problems to solve, especially for robots with many degrees of freedom and for robots that make or break contact. Therefore, many approaches add heuristics or pre-specify contact points or sequences. However, this then defines again how the robot is supposed to solve the task, affecting optimality and generality. Secondly, Trajectory Optimization cannot be blindly applied to hardware but requires an accurate model as well as a good control and estimation framework. In this work, we are addressing both issues. In our Trajectory Optimization framework, we only specify high level tasks, allowing the solver to find the optimal solution to the problem, optimizing over the entire dynamics and automatically discovering the contact sequences and timings. Second, we also demonstrate how such a dynamic task can be generated online. This work does not present a general tracking controller suitable for all trajectories. However, we show the successful integration of our Trajectory Optimization with our estimation and control framework allowing for executing a selection of tasks even under disturbances. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{images/hardware.png} \caption{Sequence of images during execution of the rough manipulation task in hardware. The time series starts at the top left corner and time progresses line-wise. The motion shows how the shift of the main body, lift-off, push-over and return to default stance.} \label{fig:hw_sequence} \end{figure} Trajectory Optimization tries to solve a general, time-varying, non-linear optimal control problem. There are various forms of defining and solving the control problem. An overview can be found in \cite{betts1998survey}. With the increase of computational power, Trajectory Optimization can be applied to higher dimensional systems like legged robots. Thus, it has gained a lot of attention in recent years and impressive results have been demonstrated in simulation \cite{posa2014direct,mordatch2012discovery,tassa2012synthesis}. One of the conceptually closest related work is \cite{tassa2014control}. However, no gait discovery or very dynamic motions are shown and no hardware results are presented. Later work \cite{koenemann2015whole} includes hardware results but the planning horizon is short, the motions are quasi-static and contact changes are slow. In \cite{mastalli} Trajectory Optimization through contacts is demonstrated on hardware. While the results are promising, the approach is only tested on a low-dimensional, single leg platform and for very simple tasks. In general, work that demonstrates Trajectory Optimization through contacts applied to physical, legged systems is very rare. Automatically discovered gaits and dynamic motions demonstrated on quadruped hardware are presented in \cite{gehring2016}. The presented results are very convincing and they are also considering actuator dynamics. However, their approach differs in key areas to the presented work: Contact sequences are pre-specified and they optimize over control parameters for a fixed control structure rather than whole body trajectories and control inputs. Additionally, they are using black box optimization. However, the tracking controllers employed in both approaches are very similar. \subsection{Contributions} In this work, we apply Trajectory Optimization (TO) in form of Sequential Liner-Quadratic (SLQ) control \cite{slq} to the quadruped robot HyQ \cite{hyq}, both in simulation and on hardware. Our work is one of very few examples, where TO is used to plan dynamic whole body motions through contact changes and where some of these trajectories are applied to hardware. Furthermore, this is possibly the first time that a whole body TO approach is shown on quadruped hardware. In contrast to state-of-the-art approaches, neither contact switching times nor contact events nor contact points are defined a priory. Instead they are a direct outcome of the optimization. This allows us to generate a diverse set of motions and gaits using a single approach. By using an efficient formulation based on Differential Dynamic Programming and a customized, high-performance solver, optimization times for these tasks usually do not exceed a minute, even for complex trajectories. Thus, while the Trajectory Optimization is not performed in a control loop, it is still run online. Hence, this work is also one of the earliest examples of whole body TO fully integrated with a control, state estimation and ground estimation pipeline of a legged robotic system. \section{Trajectory Optimization} \subsection{Optimal Control Problem} In this work, we consider a general non-linear system of the following form \begin{align} \dot{\myvec{x}}(t) = f(\myvec{x}(t),\myvec{u}(t)) \label{eq:system_model} \end{align} where $x(t)$ and $u(t)$ denote state and input trajectories respectively. In Trajectory Optimization, we indirectly optimize these trajectories by altering a linear, time-varying feedback and feedforward controller of the form \begin{align} \myvec{u}(\myvec{x},t) = \myvec{u}_{ff}(t) + \mathcal{K}(t) \myvec{x}(t) \label{control_law} \end{align} where $\mathcal{K}(t)$ is a time-varying control gain matrix and $u_{ff}(t)$ a time-varying feedforward control action. Our Trajectory Optimization approach then tries to solve a finite-horizon optimal control problem which minimizes a given cost function of the following form \begin{align} J(\myvec{x},\myvec{u}) = h\left( \myvec{x}(t_f) \right) + \int_{t=0}^{t_f}l \left( \myvec{x}(t),\myvec{u}(t) \right) dt \label{eq:cost_function} \end{align} \subsection{Sequential Linear Quadratic Control} Sequential Linear Quadratic Control is an iterative optimal-control algorithm. SLQ first rolls out the system dynamics. Then, the non-linear system dynamics are linearized around the trajectory and a quadratic approximation of the cost function is computed. The resulting Linear-Quadratic Optimal Control problem is solved backwards. Since the solution to the Linear-Quadratic Optimal Control problem can be computed in closed form with Ricatti-like equations, SLQ is very efficient. Algorithm \ref{alg:slq} summarizes the algorithm. SLQ computes both, a feedforward control action as well as a time-varying linear-quadratic regulator that stabilizes the trajectory. \begin{algorithm}[tpb] \caption{SLQ Algorithm} \label{alg:slq} \begin{algorithmic} \scriptsize \STATE \textbf{Given} \STATE - System dynamics: $\myvec{x}(t+1)=\myvec{f}\left(\myvec{x}(t),\myvec{u}(t)\right)$ \STATE - Cost function $J = h\left( \myvec{x}(t_f) \right) + \sum\limits_{t=0}^{t_f-1}{l \left( \myvec{x}(t),\myvec{u}(t) \right)}$ \STATE - Initial stable control law, $\mathbf{\myvec{u}}(\myvec{x},t)$ \REPEAT \STATE - t(0) simulate the system dynamics: \STATE $\tau: \myvec{x}_n(0),\myvec{u}_n(0),\myvec{x}_n(1),\myvec{u}_n(1),\dots,\myvec{x}_n(t_f-1),\myvec{u}_n(t_f-1),\myvec{x}_n(t_f)$ \STATE - Linearize the system dynamics along the trajectory $\tau$: \STATE $\delta\myvec{x}(t+1) = \myvec{A}(t)\delta\myvec{x}(t)+\myvec{B}(t)\delta\myvec{u}(t)$ \STATE \qquad $\{ \myvec{A}(t) \}_{0}^{t_f-1}$, $\myvec{A}(t)=\frac{\partial\myvec{f}}{\partial \myvec{x}}|_{\myvec{x}(t),\myvec{u}(t)}$ \STATE \qquad $\{ \myvec{B}(t) \}_{0}^{t_f-1}$, $\myvec{B}(t)=\frac{\partial\myvec{f}}{\partial \myvec{u}}|_{\myvec{x}(t),\myvec{u}(t)}$ \STATE - Quadratize cost function along the trajectory $\tau$: \STATE $\tilde{J} \approx p(t)+\delta\myvec{x}^T(t_f)\myvec{p}(t_f)+\frac{1}{2}\delta\myvec{x}^T(t_f)\myvec{P}(t_f)\delta\myvec{x}(t_f) $ \STATE \quad \hspace{2ex} $+ \sum\limits_{t=0}^{t_f-1}q(t)+\delta\myvec{x}^T\myvec{q}(t)+\delta\myvec{u}^T\myvec{r}(t) $ \STATE \quad \hspace{2ex} $+ \frac{1}{2}\delta\myvec{x}^T\myvec{Q}(t)\delta\myvec{x}+\frac{1}{2}\delta\myvec{u}^T\myvec{R}(t)\delta\myvec{u}$ \STATE - Backwards solve the Riccati-like difference equations: \STATE \qquad $\myvec{P}(t) = \myvec{Q}(t)+\myvec{A}^T(t)\myvec{P}(t+1)\myvec{A}(t)+ $ \STATE \hspace{12ex} $\myvec{K}^T(t)\myvec{H}\myvec{K}(t) + \myvec{K}^T(t)\myvec{G}+\myvec{G}^T\myvec{K}(t)$ \STATE \qquad $\myvec{p}(t) = \myvec{q}+\myvec{A}^T(t)\myvec{p}(t+1)+\myvec{K}^T(t)\myvec{H}\myvec{l}(t)+\myvec{K}^T(t)\myvec{g}+\myvec{G}^T\myvec{l}(t)$ \STATE \qquad $\myvec{H} = \myvec{R}(t)+\myvec{B}^T(t)\myvec{P}(t+1)\myvec{B}(t)$ \STATE \qquad $\myvec{G} = \myvec{B}^T(t)\myvec{P}(t+1)\myvec{A}(t)$ \STATE \qquad $\myvec{g} = \myvec{r}(t)+\myvec{B}^T(t)\myvec{p}(t+1)$ \STATE \qquad $\myvec{K}(t)= -\myvec{H}^{-1}\myvec{G}$ \% feedback update \STATE \qquad $\myvec{l}(t)= -\myvec{H}^{-1}\myvec{g}$ \% feedforward increment \STATE \textbf{Line search} \STATE - Initialize $\alpha = 1$ \REPEAT \STATE - Update the control: \STATE $\mathbf{\myvec{u}}(\myvec{x},t) = \myvec{u}_n(t) + \alpha \myvec{l}(t) + \myvec{K}(t) \left(\myvec{x}(t)-\myvec{x}_n(t)\right)$ \STATE - Forward simulate the system dynamics: \STATE $\tau: \myvec{x}_n(0),\myvec{u}_n(0),\myvec{x}_n(1),\dots,\myvec{x}_n(t_f-1),\myvec{u}_n(t_f-1),\myvec{x}_n(t_f)$ \STATE - Compute new cost: \STATE $J = h\left(\myvec{x}(t_f) \right) + \sum_{t=0}^{t_f-1}{l \left( \myvec{x}(t),\myvec{u}(t) \right) dt}$ \STATE - decrease $\alpha$ by a constant $\alpha_{d}$: \STATE $\alpha = \alpha / \alpha_{d}$ \UNTIL{found lower cost or number of maximum line search steps reached} \UNTIL{maximum number of iterations or converged ($\myvec{l}(t) < l_t$)} \end{algorithmic} \end{algorithm} One drawback of SLQ is that it cannot handle hard state constraints. While one can add state constraints as soft constraints into the cost function, they can result in bad numerical behavior. Handling input constraints in these type of algorithms is possible, as shown in \cite{ilqg}, we are not considering them as of now. Even though we are not considering constraints in this work, most tasks result in constraint satisfactory trajectories within the joint position and torque limits. \subsection{Cost Function} While SLQ can handle non-quadratic cost functions, pure quadratic cost functions increase convergence and are often sufficient to obtain the desired behaviour. Therefore, in this work we assume our cost function to be of quadratic form \begin{align} J &= \bar{\myvec{x}}(t_f)^T \myvec{H} \bar{\myvec{x}}(t_f) + \int\limits_{t=0}^{t_f}{ \bar{\myvec{x}}(t)^T \myvec{Q} \bar{\myvec{x}}(t) + \bar{\myvec{u}}(t)^T \myvec{R} \bar{\myvec{u}}(t) } \nonumber \\ &+ W(\myvec{x},t) \, dt \label{eq:cost_function_experiments} \end{align} where $\bar{x}(t)$ and $\bar{u}(t)$ represent deviations of state and input from a desired state and input respectively. $H$, $Q$ and $R$ are the weightings for final cost, intermediate cost and input cost respectively. Additionally, we add intermediate state waypoints to guide the trajectory. These are weighted by the waypoint cost matrix $W(x,t)$ defined as \begin{align} W(\myvec{x},t) = \sum\limits_{n=0}^{N}{ \hat{\myvec{x}}(t)^T \myvec{W}_p \hat{\myvec{x}}(t) \sqrt{\frac{\rho_p}{2 \pi}} \exp{\left( - \frac{\rho_p}{2} (t - t_p)^2 \right)} } \label{eq:waypoint} \end{align} \section{System Model and Robot Description} For the experiments in this paper, we use the hydraulically actuated, quadrupedal robot HyQ \cite{hyq}. Each of the four legs on HyQ has three degrees of freedom, namely hip abduction/adduction (HAA), hip flexion/extension (HFE) and knee flexion/extension (KFE). Each of these joints is driven by a hydraulic actuator which is controlled by a hydraulic valve. The joint torque and position is measured via load cells and encoders respectively. A hydraulic force control loop is closed at joint level, providing joint torque reference tracking. \subsection{Rigid Body Dynamics} While it is a simplification, torque tracking performance on HyQ is sufficient for modeling it as a perfectly torque controlled robot. Thus, we assume HyQ to behave like a rigid body system defined as \begin{equation} \myvec{M}(\myvec{q})\ddot{\myvec{q}} + \myvec{C}(\myvec{q},\dot{\myvec{q}}) + \myvec{G}(\myvec{q}) = \myvec{J}^T_c \lambda(\myvec{q}, \dot{\myvec{q}}) + \myvec{S}^T \tau \label{eq:rbd} \end{equation} where $M$ denotes the inertia matrix, $C$ the centripetal and coriolis forces and $G$ gravity terms. $q$ is the state vector containing the 6 DoF base state as well as joint positions and velocities. External and contact forces $\lambda$ act on the system via the contact Jacobian $J_c$. Torques created by the actuation system $\tau$ are mapped to the states via the selection matrix $S$. To formulate our dynamics according to Equation \ref{eq:system_model}, we define our state as \begin{equation} \myvec{x} = [{}_{\mathcal{W}}\myvec{q}~{}_{\mathcal{L}}\dot{\myvec{q}}]^T = [{}_{\mathcal{W}}\myvec{q}_B~{}_{\mathcal{W}}\myvec{x}_B~\myvec{q}_J~{}_{\mathcal{L}}\omega_B~{}_{\mathcal{L}}\dot{\myvec{x}}_B~\dot{\myvec{q}}_J]^T \label{eq:coordinates} \end{equation} where ${}_{\mathcal{W}}\myvec{q}_B$ and ${}_{\mathcal{W}}\myvec{x}_B$ define base orientation and position respectively, which are expressed in a global inertial ``world'' frame $\mathcal{W}$. The base orientation is expressed in euler angles (roll-pitch-yaw). The base's angular and linear velocity are denoted as $\omega_B$ and $\dot{\myvec{x}_B}$ respectively and both quantities are expressed in a local body frame $\mathcal{L}$. Joint angles and velocities are expressed as $\myvec{q}_J$ and $\dot{\myvec{q}}_J$ respectively. Expressing base pose and twist in different frames allows for more intuitive tuning of the cost function weights. Using Equations \ref{eq:rbd} and \ref{eq:coordinates} our system dynamics in Equation \ref{eq:system_model} become \begin{align} \dot{\myvec{x}}(t) &= \begin{bmatrix} {}_{\mathcal{W}}\dot{\myvec{q}} \\ {}_{\mathcal{L}}\ddot{\myvec{q}} \end{bmatrix} \\ &= \begin{bmatrix} \myvec{R}_{\mathcal{W}\mathcal{L}}~{}_{\mathcal{L}}\dot{\myvec{q}} \\ \myvec{M}^{-1}(\myvec{q})(\myvec{S}^T\tau + \myvec{J}_c \lambda(\myvec{q},\dot{\myvec{q}}) - \myvec{C}(\myvec{q}, \dot{\myvec{q}}) - \myvec{G}(\myvec{q})) \end{bmatrix} \nonumber \label{eq:system_model} \end{align} where $\myvec{R}_{\mathcal{W}\mathcal{L}}$ defines the rotation between the local body frame $\mathcal{L}$ and the inertial ``world'' frame $\mathcal{W}$. While dropped in our notation for readability, $\myvec{R}_{\mathcal{W}\mathcal{L}}$ is a function of ${}_{\mathcal{W}}\myvec{q}$ which needs to be considered during linearization. For SLQ we need to linearize the system given in in Equation \ref{eq:system_model}. For the derivatives of the upper row with respect to the state $\myvec{x}$ as well as all derivatives with respect to $\tau$ we compute analytical derivatives. For the derivatives of the lower row in Equation \ref{eq:system_model} with respect to the state $\myvec{x}$, we use numerical differentiation. \subsection{Contact Model} Choosing or designing an appropriate contact model is critical for the performance of Trajectory Optimization. There are two main criteria defining the performance of a contact model: Physical accuracy and numeric stability. For Trajectory Optimization it is beneficial to use a smooth contact model which provides good gradients of the dynamics. Yet, this can lead to undesirable, unphysical defects like ground penetration. Therefore, there is a trade-off to be made. To mitigate this trade-off we are using a non-linear spring-damper contact model extending the model proposed in \cite{slq}. We consider two contact models: One collinear and one orthogonal to the surface normal. The normal contact model is defined as \begin{equation} \lambda_N = \begin{cases} 0 & p_n \leq 0 \\ (k_n + d_n \dot{p}_n) \frac{p_n^2}{2\alpha_c} \myvec{n}_s & 0 < p_n < \alpha_c \\ (k_n+d_n \dot{p}_n) (p_n - \frac{\alpha_c}{2}) \myvec{n}_s & p_n \geq \alpha_c \end{cases} \end{equation} while the tangential model is defined as \begin{equation} \lambda_t = \begin{cases} 0 & p_n \leq 0 \\ (k_t p_t \myvec{n}_{d} + d_t \dot{p}_t \myvec{n}_{v}) \frac{p_n^2}{2\alpha_c} & 0 < p_n < \alpha_c \\ (k_t p_t \myvec{n}_{d} + d_n \dot{p}_n \myvec{n}_{v}) (p_n - \frac{\alpha_c}{2}) & p_n \geq \alpha_c \end{cases} \end{equation} Both models include a proportional and derivative terms which can be imagined as springs with stiffnesses $k_n$ and $k_t$ and dampers with damping ratios $d_n$ and $d_t$ respectively. For the normal direction the spring displacement $p_n$ is defined as the ground penetration orthogonal to the surface normal $\myvec{n}_s$. In tangential direction, the offset $p_t$ is computed between the current contact location and the location where the contact has been established. This effectively creates additional states in our system. However, we treat these states as hidden states, such that the size of the SLQ problem remains constant. In case of the tangential model, the force vector $\lambda_t$ is composed of the spring force in tangential displacement direction $\myvec{n}_{d}$ and a damping force in displacement velocity direction $\myvec{n}_{v}$. Both, the normal and tangential contact models are visualized in Figure \ref{fig:contact_models}. \begin{table}[htbp] \centering \caption{Typical Values for Contact Model Parameters} \label{tbl:contact_parameters} \begin{tabular}{|c|c|c|c|c|} \hline $\alpha_c$ & $k_n$ & $d_n$ & $k_t$ & $d_t$ \\ \hline 0.01 & 8000 ... 90000 & 2000 ... 50000 & 0 ... 5000000 & 2000 ... 5000 \\ \hline \end{tabular} \end{table} To avoid discontinuities during contact changes, the models are non-linear towards zero ground penetration, i.e. for $p_n < \alpha$, where $\alpha_c$ is a smoothing coefficient. Typical values for the contact model are given in Table \ref{tbl:contact_parameters}. The parameters are task dependent to trade-off physical correctness and convergence. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/contact_models.pdf} \caption{Visualization of the contact model. The left plot shows the contact force in surface normal direction as a function of penetration and its derivative. The right plot shows the force in tangential direction as a function of penetration and tangential displacement. Both contact models are smooth at impact, i.e. at penetration 0.} \label{fig:contact_models} \end{figure} Friction and friction limits are considered via friction cones. This allows the Trajectory Optimization algorithm to reason about possible slippage and contact force saturation. In this work, we assume that static and dynamic friction coefficients are the same. Therefore, the friction cone can be expressed as a contact force saturation \begin{equation} \myvec{F}_t = max(||\myvec{F}_t||, \mu F_n) \myvec{n}_s \end{equation} where $F_t$ is the contact force parallel to the surface, $F_n$ is the force normal to the surface, $\mu$ is the friction coefficient and $\myvec{n}_s$ is the surface normal. \section{State Estimation and Tracking Control} Our TO approach is fully integrated into our estimation and control framework, illustrated in Figure \ref{fig:control_overview}. Since SLQ assumes full state-feedback, joint position and velocity measurements are fused with IMU data to obtain a base estimate\cite{bloesch2013state}. Furthermore, since SLQ reasons about contacts with the environment, a ground estimator is added which estimates elevation and inclination of the ground. SLQ then optimizes a trajectory which is then fed to a task-space base controller and the joint controller. These control loops are closed at a rate of 200 Hz. The corrective output of both controllers is then added to the feedforward control signal as optimized by SLQ and sent to the robot. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{images/control_overview.pdf} \caption{Overview of the control and estimation pipeline. Base, stance and ground estimators provide information about the location and contact configuration of the robot. A joint controller and a base controller jointly control the robot's state.} \label{fig:control_overview} \end{figure} \subsection{State Estimation} SLQ assumes full state-feedback, i.e. all states are assumed to be known. While we can directly measure joint positions and velocities with the encoders on HyQ, the base state cannot be directly measured. Therefore, we use \cite{bloesch2013state} to estimate the base's pose and twist. The contact state is estimated by mapping joint torques to estimated ground reaction forces. If the normal component of the ground reaction force estimate is above a fixed threshold we assume the leg being in contact. In order to reason about the contact forces $\lambda$, we also have to know the elevation and contact surface normal at the contact points. In order for the algorithm to work online we cannot assume a perfectly leveled ground. Instead, we assume a ground plane which is fitted to the last contact point of each feet. This implementation allows for incorporating a ground map in the future. \subsection{Tracking Controller} SLQ does not only optimize feed-forward control action but also a feedback controller. While we have successfully applied theses feedback gains to robotic hardware \cite{mpc_quad}, we are not using the optimized feedback gains in this work for two reasons. First, the optimized feedback gains correspond to a time-varying, linear-qaudratic regulator (TVLQR). Therefore, the gains significantly depend on the linearization point. Especially due to the non-linearity of the contacts, the robot's state will drift from the linearization point during trajectory execution. Secondly, due to the usage of a single cost function, any regularization in the cost function will also effect the feedback gains which is undesirable. One solution is to recompute the gains online. However, we have found that the following tracking controller shows great performance at reduced complexity. Assuming a known, fixed stance configuration with no slippage and at least three stance legs, the stance legs as well as the torso can be described by a 6 degrees of freedom state. This means, the base state, consisting of pose and twist, is coupled with joint positions and velocities of the stance legs. Therefore, controlling either set of states, the joint states of the stance legs or the base state, will subsequently also track the other set. A base controller allows us to directly track the base state and tune feedback gains on these states intuitively. Yet, for swing legs, we still require a joint controller. Hence, to get best of both worlds, we are using a combination of a task space base controller and a joint space controller. The joint space controller tracks the desired position and desired velocity of all joints independently. Additionally, the base controller regulates the base state with a PD controller. This task space control can be imagined as virtual springs and dampers attached to the robot's trunk on one side and the optimized body trajectory on the other \cite{pratt2001virtual}. \begin{equation} \myvec{F}_{cog} = \myvec{P}_x (\myvec{x}_{cog}^* - \myvec{x}_{cog}) + \myvec{D}_x (\dot{\myvec{x}}_{cog}^* - \dot{\myvec{x}}_{cog}) \label{eq:virtual_model} \end{equation} The desired body wrench $\mathbf{F}_{cog}$ is applied to the robot by converting it to forces in the contact points $\boldsymbol{\lambda}_{c}$ and then mapping these to the joint torques through $\boldsymbol{\tau}_{fb} = \mathbf{J}_{c}^T \boldsymbol{\lambda}_{c}$. These torques are then added to the feedforward control action obtained from SLQ. This feedforward control action already includes torques that counteract gravity and thus, no gravity compensation term is added in Equation \ref{eq:virtual_model}. \section{Experiments} To evaluate the approach, we first show, how a galloping and a trotting gait with optimized contact switching sequences and timings can be obtained in simulation. Furthermore, we demonstrate a manipulation motion that includes contact switches and more complex base motions. To verify the applicability of our approach, we demonstrate hardware experiments for the last task. Finally, we show a highly under-actuated task where HyQ walks on its hind legs like a humanoid. For each task, the cost function is shown as a color code below the heading. The colors indicate the individual, relative weightings of the diagonal entries of each weighting matrix, ranging from lowest (green) to highest (red) on a logarithmic scale. No color means that the according value is zero and all off-diagonal elements are zero as well. The input cost matrix $\myvec{R}$ is set to identity in all experiments. All tasks are initialized with a simple PD feedback controller on all joints and no feedforward control. This leads to HyQ maintaining its initial state over the entire length of each trajectory. \subsection{Simulation} \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{images/all_color.png} \caption{Time series of the optimized trajectories for each task. Poses are shown as a color gradient over time ranging from red (initial pose) to green (final pose). Intermediate poses are indicated in transparent white. All displayed motions contain contact switches during dynamic maneuvers. These contact switches result from optimization and are not pre-specified.} \label{fig:strips} \end{figure} \subsubsection{Galloping} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/gallop.png} \end{center} The first gait we demonstrate is galloping. The galloping gait is a direct outcome of the optimization with a very simple cost function. The cost function consists of final costs on the base position and leg positions. Additionally, we add some regularization on the base and leg motion to prevent excessive motions of the body or the limbs. However, the cost function does not include any terms related to contact sequences or timings and no priors on a galloping gait. Also, we are not using any intermediate cost terms here. HyQ starts in its default stance and is supposed to reach its final position 2 m in front. The chosen time horizon is 3 seconds. As for all tasks, the initial controller is a simple joint PD controller which results in HyQ staying in place and maintaining its intial configuration. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/gallop_base.pdf} \caption{Plots of base pose and base twist during galloping as optimized by SLQ. The robot takes in total 9 galloping steps. As expected, there is significant pitch motion during galloping. The desired final position at $x = 2.0$ m is reached with good accuracy. } \label{fig:gallop_base} \end{figure} \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/gallop_torques.pdf} \caption{Torque profiles of the different joints during a galloping motion. As the lower plots show, the hind legs and especially the HFE joints, contribute greatly to the acceleration. The front leg torques seem to contribute fairly evenly to the galloping motion throughout the trajectory.} \label{fig:gallop_torques} \end{figure} As the results in Figure \ref{fig:gallop_base} illustrates, we obtain a gallop motion with 9 steps which includes acceleration and deceleration. Finally, the robot reaches the desired position at $x=2.0$ m. As expected, we see significant pitch motion of the upper body. From Figure \ref{fig:gallop_torques} we can tell that the hind legs, especially the HFE joints, are used for acceleration. While our findings are not directly comparable to biology, we can see similar effects: In fast running animals, like leopards and horses, the rear legs also bear a larger load during running than the front legs, which explains the different sizing of legs and muscles between rear and front. Snapshots of the gallop are shown in Figure \ref{fig:strips} and the full motion is shown in the video\footnote{\label{video}\url{https://youtu.be/sILuqJBsyKs}}. \subsubsection{Trotting} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/trot.png} \end{center} One hypothesis for the previous task resulting in a galloping behavior is the short time horizon and no penalty on the orientation. Now if we increase the time horizon to 8 seconds and penalyze the base motion, i.e. we give HyQ more time and encourage smoother base motions, we see that the optimization starts to prefer a trotting motion instead of the galloping. Interestingly enough, the weightings allow to blend between galloping and trotting. As a result, HyQ first gallops and then smoothly transitions to a trotting gait. Again, neither the contact sequence, nor the gait or the transition is given. The parameters influencing the gait are the diagonal elements of $\myvec{Q}$ and $\myvec{R}$ as in Equation \ref{eq:cost_function_experiments}. By increasing the base motion penalty, the trajectory results in a pure trot. The trot consists of 4 steps per diagonal leg pair with almost constant stride length. By setting the desired position to the side instead of the front, the resulting trajectory is a sidestepping motion. If only a desired yaw angle is set, the robot turns on the spot. Both trajectories are trotting variants where the diagonal leg pairs are moved together. Both motions are illustrated in Figure \ref{fig:strips} and included in the video\textsuperscript{\ref{video}}. \subsubsection{Squat Jump} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/squat.png} \end{center} Next, we test if our Trajectory Optimization approach can leverage and reason about the dynamics of our system. Thus, we define a task with intermediate waypoint states that are not statically reachable. First, we add an intermediate waypoint cost term for the base position and orientation. For this waypoint, we choose a strong weighting for the base height and set the desired value of the base height to $z=0.8 m$. Furthermore, the waypoint costs contain low weighting on the base orientation to keep the base level. By adding a cost on the deviation from default joint position, we ensure HyQ cannot just try to straighten its legs to try to reach the base pose, but that it has to jump. After running our optimization, we obtain a near symmetric squat jump. The overall optimization spans the entire motion, e.g. preparation for lift-off from default pose, the lift-off itself, going to default pose in the air, landing and returning to the default pose. The apex waypoint is localized in time but contact switches and timings are optimized. Also, we are not directly specifying a squat jump. Therefore, simultaneous lift-off and landing of the legs is an optimization outcome rather than pre-specified. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/squat_base.pdf} \caption{Plots of base pose and base twist during a squat jump as optimized by SLQ. The graphs show that the robot keeps its base stable while jumping. Also unnecessary base motion in x- and y-direction are minimized. Furthermore, we can see that the desired apex height of 0.8 m is reached (measured value 0.8004).} \label{fig:squat_base} \end{figure} Figure \ref{fig:squat_base} shows the optimized squat jump motion. We can see that unnecessary motions such as rotations or position changes in x- or y-direction are avoided. Also, the desired apex height of 0.8 m, despite being a soft constraint, is reached with sub-millimeter accuracy (measured 0.8004 m). The small pitch velocity results from an off-center mass and the optimization goal of low torques which promotes equal distribution of torques between legs as seen in Figure \ref{fig:squat_torques}. The same plot also shows that the largest torques appear in the KFE and HFE joints. Furthermore, the lower leg of HyQ weighs less than 1 kg compared to a total mass of about 80 kg. Therefore, the mass that the lower leg drives changes rapidly during contact changes occurring at around $t=0.55 s$ and $t=0.9 s$. Hence, the torques also change quickly. While the knee joint is still moved in the air to take the default angle and prepare for landing, the required torque is minimal. Again, both motions are shown as a snapshot series in Figure \ref{fig:strips} as well as in the video\textsuperscript{\ref{video}}. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/squat_torques.pdf} \caption{Plots of the torque profiles during a squat jump. During stance phases the profiles are fairly flat. There are two distinct spikes which are the torques produced during take-off and landing. By including the torques in the optimization criterion of the squat jump task, the distribution is even between all legs.} \label{fig:squat_torques} \end{figure} \subsubsection{Rearing} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/rear.png} \end{center} For the next task, we are using a similar cost function as the squat jump. Instead of penalizing the deviation from a neutral base pose, we penalize deviations from a 30$^\circ$ pitched base orientation and we lower the desired apex height to 0.7 m. Yet again, we do not specify contact/stance configurations. Therefore, the optimization algorithm is free to optimize the trajectory. The final trajectory represents a rearing motion, where HyQ lifts off with the front legs, reaches the apex position and finally returns to full contact as well as its default pose. A screenshot of the apex position is shown in Figure \ref{fig:strips}. The full motion is shown in the accompanying video\textsuperscript{\ref{video}}. \subsubsection{Diagonal Balance} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/balance.png} \end{center} In order to demonstrate that the SLQ can also find statically unstable trajectories, we are demonstrating a diagonal balance task. Here, we use a waypoint term in our cost function again. This term penalizes the orientation and height of the base. Furthermore, it encourages the robot to pull up its legs by bending HFE and KFE of the left front and right hind leg. The final trajectory shows the expected balancing behavior. Again a screenshot at apex is shown in Figure \ref{fig:strips} and the video\textsuperscript{\ref{video}} shows the full motion. Interestingly, while we are using a single intermediate term with a single time point and absolutely symmetric costs, the lift-off and touch-down of the swing legs is not synchronous but the front left leg lifts-off later and touches down earlier. This asymmetry most likely stems from the asymmetric location of the CoG and assymmetric inertia of the robot's main body. \subsubsection{Humanoid Walk} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/humanoid.png} \end{center} As a final test, we evaluate if our algorithm is capable of optimizing a behavior that allows HyQ to stand on its hind legs, similar to a humanoid robot, and then go to a target point in front. While standing on its hind legs exceeds the capabilities of the hardware, such a motion is physically possible. Compared to a humanoid with ankles, HyQ has point feet. This increases the level of difficulty since the absence of ankles makes HyQ highly under-actuated. This means that during a standing or walking task no torques can be applied to the ground but balance can only be maintained by moving the links. In this task, we again initialize HyQ in its default stance, requiring it to stand up in place first and then reach a goal point where it should stabilize for a second. The first waypoint has a very wide spread in time and penalizes deviations in base orientation and height. This cost ensures that HyQ stays upright during the entire task after getting up. We do not add this cost to the general intermediate cost, as it would encourage HyQ to stand up as fast as possible, rather than getting up in a controlled, efficient manner. The stand up motion is encouraged by the second waypoint cost penalizing base orientation, height and changes in forward position. We add a third waypoint one second before the end of the time horizon, defining the target pose and orientation. While the last waypoint and the final cost specify the same base pose, we separate them to demonstrate that HyQ can stay upright and stabilize in place for longer times without showing signs of falling. The resultant motion shows interesting, non-trivial behavior. Before getting up HyQ pulls its hind left leg in, moving the contact point more closely below the center of gravity. Also, it uses the front left leg (``left arm'') to get up, resulting in a very natural, coordinated, asymmetric motion. After getting into a two-leg standing phase, a forward motion is initiated by a short symmetric hopping but quickly changes to a coordinated stepping/walking pattern. While we penalize joint velocity, the front limbs are moved to support balance during the motion until they are retracted to a target pose at the end of the trajectory. The entire motion underlines once more the capabilities of the approach. Classic controllers and motion planners might not find such a complex motion. Additionally, by allowing optimization over contact points and sequences, we allow the Trajectory Optimization to find the best solution for the task. While a rearing motion, as demonstrated above, is feasible and can be found using our problem formulation, the optimizer prefers a less dynamic, coordinated motion here. \subsection{Hardware} \begin{center} \includegraphics[width=\columnwidth]{plots/costs/hardware.png} \end{center} To verify that the optimized trajectories can be applied to hardware, we run an example task on hardware. This task is a rough manipulation task where HyQ pushes over a pallet with its front left leg. The trajectory requires a sequence of motions such as shifting its CoG in the support polygon of the three stance legs, lifting off of the front left leg, executing the push motion, putting the foot back down and shifting the CoG back. While in classic approaches these motions would possibly all handcoded, they result from a single cost function with only one intermediate cost function term for the front left leg in our Trajectory Optimization approach. While this trajectory works perfectly fine in simulation, we have to slightly alter it for the hardware. Since our TO approach is deterministic and we penalize control input, the algorithm tries to minimize the shift of the CoG, leading to a very risky trajectory. Thus, small model inaccuracies or disturbances make the robot loose its balance. This is a general problem of all deterministic TO approaches and can only be fundamentally solved by using risk aware or stochastic approaches. While there is a risk-aware SLQ variant \cite{ileg2015}, here we apply a work-around by simply encouraging a larger CoG shift in our cost function. For visual purposes only, we add additional intermediate waypoints for the front left leg only, leading to a slower, more appealing push motion. While we could easily add the push contact to our optimization, we leave it unmodelled on purpose such that it becomes a disturbance to our controller, underlining its robustness. The pallet is a 1200 by 800 mm Euro pallet that weighs approximately 24 kg and thus requires a force of about 70 N at the point of contact of the leg to be pushed over. \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/hardware/base_tracking.pdf} \caption{Base pose and twist trajectories (solid) and their respective references (dashed) for the hardware test where HyQ is executing a rough manipulating task. The plots show that the planned and executed base trajectories are very similar. While the task space controller is tracking the base trajectory, it also gets indirectly tracked through the joint space controller.} \label{fig:hw_base_tracking} \end{figure} \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/hardware/joint_tracking.pdf} \caption{Planned (dashed) and executed (solid) joint trajectories during the hardware test where HyQ is executing a rough manipulation task. The joint tracking controller follows the reference very well and thus also contributes to tracking the base trajectory.} \label{fig:hw_joint_tracking} \end{figure} \begin{figure}[htbp] \centering \includegraphics[width=\columnwidth]{plots/hardware/torques.pdf} \caption{Joint torques during the hardware test where HyQ is executing a rough manipulation task. These joint torques are the sum of the feedforward torques of the TO algorithm and the task space controller output. These torques are given to the joint space controller which adds additional tracking torques.} \label{fig:hw_torques} \end{figure} To demonstrate that the algorithm is being run online we approach the pallet with a separate walking controller. Thus, the robot is faced with different initial conditions every time. Figure \ref{fig:strips} shows a sequence of images of the optimized motion while Figure \ref{fig:hw_sequence} shows a sequence of images taken during execution. The video\textsuperscript{\ref{video}} shows the entire motion as well as examples from our physics simulator, where initial conditions differ even more significantly. As can be seen in all examples, the motion is dynamic and lift-off and touch-down events of the front left leg partially overlap in time with the CoG shift. This underlines that static stability is not required in this task but the algorithm finds a dynamically stable trajectory. The base pose/twist tracking for this task during hardware experiments is shown in Figure \ref{fig:hw_base_tracking} and the joint position/velocity tracking in Figure \ref{fig:hw_joint_tracking}. We can see that the joint tracking is better than base tracking due to a more aggressive joint controller. However, the base pose and twist do not significantly deviate from their planned trajectories either. Figure \ref{fig:hw_torques} shows the combination of feedforward torques and task space control input. In the front left leg, we can nicely see the lift off and touch down events in between which joint torques are almost zero. This happens since this leg does not further contribute to sustaining the weight of the body. Subsequently, the other legs have to bear a higher load. Especially in the neighboring legs, i.e. the right front and left hind leg, we see an increase in torques. Yet, this increase is not very significant since the robot has already shifted its CoG towards the support polygon of the stance legs. \subsection{Runtime and Convergence} When running Trajectory Optimization online, runtime and convergence become a major concern since these measures define how long the robot ``thinks'' before executing a task. Especially in a dynamic environment or in presence of drifting estimates, we want to keep the optimization procedures short to be able to react to a given situation quickly. Even better than pre-optimizing a trajectory before execution, we reoptimize and adjust them during execution, forming a model-predictive control scheme. In this section, we will look at both the number of iterations for each task as well as the runtime of each iteration. This gives us an indicator of the complexity of a task and tells us how far we are from running our approach in an MPC fashion. \begin{figure}[!htbp] \centering \includegraphics[width=\columnwidth]{plots/convergence.pdf} \caption{Convergence rates for different tasks. The convergence rate seems to be influenced by the length and complexity of the task. The fastest convergence can be observed in the manipulation tasks since the complexity and duration is lower than in other tasks. The trotting behavior converges the slowest.} \label{fig:convergence} \end{figure} As a first test, we take a look at convergence rates across tasks. To obtain comparable results between different tasks, we initialize all tasks with a standing controller in form of a pure joint position controller and normalize the costs with the initial cost. The results of this test are shown in Figure \ref{fig:convergence}. The curves suggest that there is a relation between the time horizon and complexity of a task and the corresponding convergence rate. The trotting behavior is one of the most complex behaviors and also has the longest time horizon. The rearing task is relatively simple and well guided by the waypoint costs which is a possible explanation of the fast convergence rate. Most tasks converge within 10 and 40 iterations, which given the runtime per iteration, shown in Figure \ref{fig:runtime}, usually means an overall optimization time of less than 1 minute. \begin{figure}[!htbp] \centering \includegraphics[width=\columnwidth]{plots/runtime.pdf} \caption{Runtime per iteration for different tasks. From a theoretical point of view, the relation between number of time steps and the runtime per iteration should be linear. This plot nicely supports this hypothesis. As a result, SLQ scales well also for problems with large time horizons.} \label{fig:runtime} \end{figure} When it comes to runtime, SLQ has a major advantage over many other Trajectory Optimization approaches: It scales linear with discretization steps and thus with time horizon. Each SLQ step is fixed runtime which means the fewer steps we use, the faster the optimizer runs. Therefore, one would ideally increase step size but also shorten the trajectory length. Yet, there is a limit to both. Figure \ref{fig:runtime} underlines the linear relation between the number of trajectory points (or steps) and the runtime per iteration. The timings are measured on a standard quadcore laptop and averaged over multiple iterations. To achieve these runtimes we use a custom multi-threaded solver. Additionally, we use highly optimized, auto-generated code for the computation of forward dynamics, used for rollouts/line-search and linearization. This code is generated using RobCoGen \cite{robcogen}, a code-generator framework for rigid body kinematics and dynamics. \section{Discussion, Conclusion and Outlook} In this paper, we have presented a fully dynamic, whole-body Trajectory Optimization framework that is able to create motions which involve multiple contact changes. The approach does not require any priors or initial guesses on contact points, sequences or timings. We demonstrate the capabilities on various tasks including squat jumps, rearing, balancing and rough manipulation. Furthermore, our approach is able to discover gaits such as galloping, trotting and two legged walking. First hardware results show that the optimized trajectories can be transferred to hardware. While some motions might not be directly applicable to hardware, they can serve as an initialization for other algorithms and they could potentially provide insight into optimal gait parameters such as stepping frequencies or stride lengths. As with most optimal control approaches, a certain amount of cost function tuning is required. However, we use only one cost function per task and tried to stay away from complex terms that tune for numerical properties. Instead, we tuned for functionality and behavior and provide insight into the weightings used. While tuning cost functions is a manual procedure, most of the obtained motion would require a complex combination of different planners and controllers if implemented with state-of-the-art approaches, e.g. \cite{victor_trot,coros2011locomotion}. In contrast to these methods, whole body Trajectory Optimization is versatile and the structure of the approach is not task dependent. Also, the optimized trajectories leverage the full dynamics of the system and do not rely on heuristics or simplified models. Using our formulation and solvers, the optimized trajectory is usually available in less than one minute, even for complex tasks and despite not pre-specifying contacts. This compares favorable to reported results in literature where solving time can be several minutes or even hours. However, there are also some short comings of the approach. The solution space of the approach is huge, e.g. we can apply our approach to a broad variety of tasks despite limiting ourselves to a single cost function structure. While this generality is a strength of the approach, it also requires the user to ``choose'' a solution by modifying cost function weights or adding additional term. While this is a more or less intuitive approach, one would wish to further reduce the number of open parameters by applying some form of meta algorithm. Another drawback is that the optimized motions might not be very robust to model inaccuracies, disturbances or noise. Thus, alterations of the cost functions might be required to successfully execute these motions on hardware. This is a known issue of deterministic Trajectory Optimization approaches and can be tackled only by considering the stochasticity of the problem. Lastly, SLQ cannot handle state constraints, which can be a limiting factor for complex Trajectory Optimization problems. We believe that SLQ can be especially useful when applied as a shorter horizon MPC controller where its fast run-times are leveraged and complex state constraints are taken care of by a higher level planner/optimizer. While the run-time of SLQ seems far off from an MPC scenario, warm starting is an efficient measure to make the algorithm converge in only a few iterations. Also, run-time can be further reduced by limiting the time horizon. Thus, as a next step, we are planning to apply our approach in MPC fashion to HyQ, extending our work in \cite{mpc_quad}. Hopefully, SLQ-MPC will be able to act as a general stabilization and tracking controller as well as short horizon re-planner, allowing us to execute most tasks demonstrated in this work on hardware. One important step in this direction will be to include input constraints. \section*{Acknowledgment} This research has been funded through a Swiss National Science Foundation Professorship award to Jonas Buchli and the NCCR Robotics. \bibliographystyle{IEEEtran}
929a6a71f937242c2ee06aa7ed3e9864fff7437a
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The combinatorial structure treated in this paper is a $2 \rightarrow 1$ directed hypergraph defined as follows. \begin{definition} A \emph{$2 \rightarrow 1$ directed hypergraph} is a pair $H = (V,E)$ where $V$ is a finite set of \emph{vertices} and the set of \emph{edges} $E$ is some subset of the set of all pointed $3$-subsets of $V$. That is, each edge is three distinct elements of $V$ with one marked as special. This special vertex can be thought of as the \emph{head} vertex of the edge while the other two make up the \emph{tail set} of the edge. If $H$ is such that every $3$-subset of V contains at most one edge of $E$, then we call $H$ \emph{oriented}. For a given $H$ we will typically write its vertex and edge sets as $V(H)$ and $E(H)$. We will write an edge as $ab \rightarrow c$ when the underlying $3$-set is $\{a,b,c\}$ and the head vertex is $c$. \end{definition} For simplicity from this point on we will always refer to $2 \rightarrow 1$ directed hypergraphs as just \emph{graphs} or sometimes as \emph{$(2 \rightarrow 1)$-graphs} when needed to avoid confusion. This structure comes up as a particular instance of the model used to represent definite Horn formulas in the study of propositional logic and knowledge representation \cite{angluin1992, russell2002}. Some combinatorial properties of this model have been recently studied by Langlois, Mubayi, Sloan, and Gy. Tur\'{a}n in \cite{langlois2009} and \cite{langlois2010}. In particular, they looked at the extremal numbers for a couple of different small graphs. Before we can discuss their results we will need the following definitions. \begin{definition} Given two graphs $H$ and $G$, we call a function $\phi:V(H) \rightarrow V(G)$ a homomorphism if it preserves the edges of $H$: \[ab \rightarrow c \in E(H) \implies \phi(a)\phi(b) \rightarrow \phi(c) \in E(G).\] We will write $\phi:H \rightarrow G$ to indicate that $\phi$ is a homomorphism. \end{definition} \begin{definition} Given a family $\mathcal{F}$ of graphs, we say that a graph $G$ is \emph{$\mathcal{F}$-free} if no injective homomorphism $\phi:F \rightarrow G$ exists for any $F \in \mathcal{F}$. If $\mathcal{F} = \{F\}$ we will write that $G$ is $F$-free. \end{definition} \begin{definition} Given a family $\mathcal{F}$ of graphs, let the \emph{$n$th extremal number} $\text{ex}(n,\mathcal{F})$ denote the maximum number of edges that any $\mathcal{F}$-free graph on $n$ vertices can have. Similarly, let the \emph{$n$th oriented extremal number} $\text{ex}_o(n,\mathcal{F})$ be the maximum number of edges that any $\mathcal{F}$-free oriented graph on $n$ vertices can have. Sometimes we will call the extremal number the \emph{standard} extremal number or refer to the problem of determining the extremal number as the \emph{standard version} of the problem to distinguish these concepts from their oriented counterparts. As before, if $\mathcal{F} = \{F\}$, then we will write $\text{ex}(n,F)$ or $\text{ex}_o(n,F)$ for simplicity. \end{definition} These are often called Tur\'{a}n-type extremal problems after Paul Tur\'{a}n due to his important early results and conjectures concerning forbidden complete $r$-graphs \cite{turan1941, turan1954, turan1961}. Tur\'{a}n problems for uniform hypergraphs make up a large and well-known area of research in combinatorics, and the questions are often surprisingly difficult. Extremal problems like this have also been considered for directed graphs and multigraphs (with bounded multiplicity) in \cite{brown1973} and \cite{brown1969} and for the more general directed multi-hypergraphs in \cite{brown1984}. In \cite{brown1969}, Brown and Harary determined the extremal numbers for several types of specific directed graphs. In \cite{brown1973}, Brown, Erd\H{o}s, and Simonovits determined the general structure of extremal sequences for every forbidden family of digraphs analogous to the Tur\'{a}n graphs for simple graphs. The model of directed hypergraphs studied in \cite{brown1984} have $r$-uniform edges such that the vertices of each edge is given a linear ordering. However, there are many other ways that one could conceivably define a uniform directed hypergraph. The graph theoretic properties of a more general definition of a nonuniform directed hypergraph were studied by Gallo, Longo, Pallottino, and Nguyen in \cite{gallo1993}. There a directed hyperedge was defined to be some subset of vertices with a partition into head vertices and tail vertices. Recently in \cite{cameron2016}, this author tried to capture many of these possible definitions for ``directed hypergraph" into one umbrella class of relational structures called generalized directed hypergraphs. The structures in this class include the uniform and simple versions of undirected hypergraphs, the totally directed hypergraphs studied in \cite{brown1984}, the directed hypergraphs studied in \cite{gallo1993}, and the $2 \rightarrow 1$ model studied here and in \cite{langlois2009,langlois2010}. In \cite{langlois2009, langlois2010}, they study the extremal numbers for two different graphs with two edges each. They refer to these two graphs as the 4-resolvent and the 3-resolvent configurations after their relevance in propositional logic. Here, we will denote these graphs as $R_4$ and $R_3$ respectively and define them formally as \[V(R_4) = \{a,b,c,d,e\} \text{ and } E(R_4) = \{ab \rightarrow c, cd \rightarrow e\}\] and \[V(R_3) = \{a,b,c,d\} \text{ and } E(R_3) = \{ab \rightarrow c, bc \rightarrow d\}.\] In \cite{langlois2010} the authors determined $\text{ex}(n,R_4)$ for sufficiently large $n$, and in \cite{langlois2009} they determined a sequence of numbers asymptotically equivalent to the sequence of numbers $\text{ex}(n,R_3)$ as $n$ increases to infinity. In these papers, the authors discuss a third graph with two edges which they call an Escher configuration because it calls to mind the Escher piece where two hands draw each other. This graph is on four vertices, $\{a,b,c,d\}$ and has edge set $\{ab \rightarrow c,cd \rightarrow b\}$. We will denote it by $E$. These three graphs actually turn out to be the only three nondegenerate graphs with exactly two edges on more than three vertices. \begin{definition} A graph $H$ is \emph{degenerate} if its vertices can be partitioned into three sets, $V(H) = T_1 \cup T_2 \cup K$ such that every edge of $E(H)$ is of the form $t_1t_2 \rightarrow k$ for some $t_1 \in T_1$, $t_2 \in T_2$, and $k \in K$. \end{definition} An immediate consequence of a result shown in \cite{cameron2016} is that the extremal numbers for a graph $H$ are cubic in $n$ if and only if $H$ is not degenerate. In our model of directed hypergraph, there are nine different graphs with exactly two edges. Of these, four are not degenerate. One of these is the graph on three vertices with exactly two edges, $V=\{a,b,c\}$ and $E=\{ab \rightarrow c, ac \rightarrow b\}$. It is trivial to see that both the standard and oriented extremal numbers for this graph are ${n \choose 3}$. The other three nondegenerate graphs are $R_4$, $R_3$, and $E$. We will determine both the standard and oriented extremal numbers for each of these graphs in Sections 2, 3, and 4 respectively. The extremal numbers for the five degenerate cases are determined in a companion paper \cite{cameron2015deg}. The proofs that follow rely heavily on the concept of a link graph. For undirected $r$-graphs, the link graph of a vertex is the $(r-1)$-graph induced on the remaining vertices such that each $(r-1)$-set is an $(r-1)$-edge if and only if that set together with the specified vertex makes an $r$-edge in the original $r$-graph \cite{keevash2011}. In the directed hypergraph model here, there are a few ways we could define the link graph of a vertex. We will need the following three. \begin{definition} Let $x \in V(H)$ for some graph $H$. The \emph{tail link graph} of $x$ $T_x$ is the simple undirected 2-graph on the other $n-1$ vertices of $V(H)$ with edge set defined by all pairs of vertices that exist as tails pointing to $x$ in some edge of $H$. That is, $V(T_x) = V(H) \setminus \{x\}$ and \[ E(T_x) = \{yz : yz \rightarrow x \in H\}.\] The size of this set, $|T_x|$ will be called the \emph{tail degree} of $x$. The degree of a particular vertex $y$ in the tail link graph of $x$ will be denoted $d_x(y)$. Similarly, let $D_x$ be the \emph{directed link graph} of $x$ on the remaining $n-1$ vertices of $V(H)$. That is, let $V(D_x) = V(H) \setminus \{x\}$ and \[E(D_x) = \{y \rightarrow z : xy \rightarrow z \in E(H)\}.\] And let $L_x$ denote the \emph{total link graph} of $x$ on the remaining $n-1$ vertices: $V(L_x) = V(H) \setminus \{x\}$ and \[E(L_x) = E(T_x) \cup E(D_x).\] So $L_x$ is a partially directed 2-graph. \end{definition} \section{The 4-resolvent graph $R_4$} \begin{figure} \centering \begin{tikzpicture} \filldraw [black] (-2,2) circle (1pt); \filldraw [black] (-2,0) circle (1pt); \draw[thick] (-2,2) -- (-2,0); \filldraw [black] (0,1) circle (1pt); \draw[thick, ->] (-2,1) -- (0,1); \filldraw [black] (0,-1) circle (1pt); \draw[thick] (0,1) -- (0,-1); \draw[thick,->] (0,0) -- (2,0); \filldraw [black] (2,0) circle (1pt); \end{tikzpicture} \caption{$R_4$} \label{C} \end{figure} In \cite{langlois2009}, the authors gave a simple construction for an $R_4$-free graph. Partition the vertices into sets $T$ and $K$ and take all possible edges with tail sets in $T$ and head vertex in $K$. When there are $n$ vertices, this construction gives ${t \choose 2}(n-t)$ edges where $t=|T|$. This is optimized when $t = \left\lceil \frac{2n}{3} \right\rceil$. In \cite{langlois2010}, they showed that this number of edges is maximum for $R_4$-free graphs for sufficiently large $n$ and that the construction is the unique extremal $R_4$-free graph. \begin{figure} \centering \begin{tikzpicture} [scale=0.5] \filldraw[color=black,fill=blue!5] (0,0) ellipse (2 and 3); \node at (-2,3) {$T$}; \filldraw[color=black,fill=blue!5] (6,0) ellipse (1 and 2); \node at (7,2) {$K$}; \node [left] at (-2,0) {${n-k \choose 2}$ tail pairs}; \node [right] at (7,0) {$k$ heads}; \filldraw[black] (0,-1) circle (1pt); \filldraw[black] (0,1) circle (1pt); \filldraw[black] (6,0.5) circle (1pt); \draw[thick] (0,-1) -- (0,1); \draw[thick,->] (0,0) -- (6,0.5); \end{tikzpicture} \caption{The lower bound construction for a graph with no $R_4$.} \end{figure} We now give an alternate shorter proof that $\left\lfloor \frac{n}{3} \right\rfloor{\left\lceil \frac{2n}{3} \right\rceil \choose 2}$ is an upper bound on the extremal number for $R_4$ for sufficiently large $n$ in both the standard and oriented versions of the problem. The proof also establishes the uniqueness of the construction. \begin{theorem} \label{TypeC} For all $n \geq 29$, \[ex_o(n,R_4) = \left\lfloor \frac{n}{3} \right\rfloor{\left\lceil \frac{2n}{3} \right\rceil \choose 2}\]and for all $n \geq 70$, \[ex(n,R_4) = \left\lfloor \frac{n}{3} \right\rfloor{\left\lceil \frac{2n}{3} \right\rceil \choose 2}.\]Moreover, in each case there is one unique extremal construction up to isomorphism when $n \equiv 0,1 \text{ mod } 3$ and exactly two when $n \equiv 2 \text{ mod } 3$. \end{theorem} \begin{proof} In either the standard or the oriented model, let $H$ be an $R_4$-free graph on $n$ vertices. Partition $V(H)$ into sets $T \cup K \cup B$ where $T$ is the set of vertices that appear in tail sets of edges but never appear as the head of any edge, $K$ is the set of vertices that do not belong to any tail set, and $B$ is the set that appear as both heads and tails. If $B$ is empty, then $H$ is a subgraph of the lower bound construction and we are done. So assume that there exists some $v \in B$. The link graph $L_v$ must contain at least one undirected edge and at least one directed edge. If any undirected edge is independent from any directed edge in $L_v$, then $v$ would be the intersection vertex for an $R_4$ in $H$. Therefore, every directed edge in $L_v$ is incident to every undirected edge. \begin{figure} \centering \begin{tikzpicture} [scale=0.5] \filldraw[color=black, fill=blue!5] (0,1) circle [radius=3]; \node at (-3,4) {$H$}; \filldraw [black] (-2,2) circle (1pt); \filldraw [black] (-2,0) circle (1pt); \draw[thick] (-2,2) -- (-2,0); \filldraw [black] (0,1) circle (1pt); \draw[thick, ->] (-2,1) -- (0,1); \filldraw [black] (0,-1) circle (1pt); \draw[thick] (0,1) -- (0,-1); \draw[thick,->] (0,0) -- (2,0); \filldraw [black] (2,0) circle (1pt); \node [above right] at (0,1) {$v$}; \node [right] at (2,0) {$z$}; \node [below right] at (0,-1) {$y$}; \node [above right] at (-2,2) {$a$}; \node [below right] at (-2,0) {$b$}; \node at (4.5,1) {$\iff$}; \filldraw[color=black, fill=blue!5] (9,1) circle [radius=3]; \node at (6,4) {$L_v$}; \filldraw [black] (7,2) circle (1pt); \node [above right] at (7,2) {$a$}; \filldraw [black] (7,0) circle (1pt); \node [below right] at (7,0) {$b$}; \draw[thick] (7,2) -- (7,0); \filldraw[black] (9,-1) circle (1pt); \node [below right] at (9,-1) {y}; \filldraw[black] (11,0) circle (1pt); \node [right] at (11,0) {z}; \draw[thick,->] (9,-1) -- (11,0); \end{tikzpicture} \caption{$H$ has an $R_4$ if and only if the link graph of some vertex $v$ contains a directed edges and an undirected edge that do not intersect.} \end{figure} We want to show that if $v \in B$, then $|E(L_v)| = O(n)$. Determining an upper bound on the number of edges in $L_v$ is equivalent to determining an upper bound on the number of red and blue edges on $n-1$ vertices such that each red edge is incident to each blue edge and there is at least one edge of each color. If we are working in the oriented model where multiple edges on the same triple are not allowed then no pair of vertices in $L_x$ can hold more than one edge. If we are working in the standard model, then two vertices in this graph may have up to three edges between them, say two red and one blue. First, let's consider the oriented version. In this case we have at least one edge of each color and they must be incident. So let $xy$ be blue and let $yz$ be red. Then all other edges must be incident to $x$, $y$, or $z$. Moreover, any edge from $x$ to the remaining $n-4$ vertices must be red since it is independent from $yz$ and any edge from $z$ to the remaining $n-4$ must be blue. Therefore, there are at most $2(n-4)$ edges from $\{x,y,z\}$ to the remaining $n-4$ vertices. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,6) circle (1pt); \node [above] at (0,6) {$y$}; \filldraw[black] (-2,5) circle (1pt); \node [left] at (-2,5) {$x$}; \filldraw[black] (2,5) circle (1pt); \node [right] at (2,5) {$z$}; \draw[blue, thick] (0,6)--(-2,5); \draw[red, thick] (0,6)--(2,5); \filldraw[color=black, fill=blue!5] (0,2) ellipse (4 and 2); \node at (0,1) {$n-4$}; \filldraw[black] (-3,2) circle (1pt); \filldraw[black] (-2,2) circle (1pt); \filldraw[black] (-1,2) circle (1pt); \node at (0.5,2) {$\cdots$}; \filldraw[black] (2,2) circle (1pt); \filldraw[black] (3,2) circle (1pt); \draw[red, thick,dashed] (0,6)--(-1,2); \draw[blue, thick,dashed] (0,6)--(2,2); \draw[red, thick,dashed] (-2,5)--(-1,2); \draw[red, thick,dashed] (-2,5)--(-3,2); \draw[blue, thick,dashed] (2,5)--(3,2); \draw[blue, thick,dashed] (2,5)--(-2,2); \end{tikzpicture} \caption{A simple graph on $n-1$ vertices with red and blue edges such that each red edge is incident to each blue edge and there is at least one blue edge, $xy$, and at least one red edge, $yz$, can have no edge contained in the remaining $n-4$ vertices. Moreover, only red edges can go from $x$ to the remaining vertices and only blue edges can go from $z$ to the remaining vertices.} \end{figure} In the standard case our initial two red and blue edges may either be incident as before with $xy$ blue and $yz$ red or they might be incident in two vertices so that $xy$ holds both a red and a blue edge. If none of the first type of incidence exists, then there are at most 3 edges, all on $xy$. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,6) circle (1pt); \node [above] at (0,6) {$y$}; \filldraw[black] (-1,4) circle (1pt); \node [left] at (-1,4) {$x$}; \filldraw[black] (1,4) circle (1pt); \node [right] at (1,4) {$z$}; \draw[blue, thick] (0,6)--(-1,4); \draw[red, thick] (0,6)--(1,4); \filldraw[black] (5,6) circle (1pt); \node [right] at (5,6) {$y$}; \filldraw[black] (5,4) circle (1pt); \node [right] at (5,4) {$x$}; \draw[blue, thick] (5,6) .. controls (4.5,5) .. (5,4); \draw[red, thick] (5,6) .. controls (5.5,5) .. (5,4); \end{tikzpicture} \caption{When two vertices are allowed to have up to two red edges and one blue edge, then an adjacent red and blue edge pair is either incident in one or two vertices.} \end{figure} So let's assume that the first type of incidence exists - $xy$ is a blue edge and $yz$ is a red edge. As before, all other edges must be incident to these three vertices, any edge from $x$ to the remaining $n-4$ vertices must be red, and any edge from $z$ to these vertices must be blue. Since each pair can have up to two reds and one blue, then there are at most $2(n-4)$ edges from $x$, $3(n-4)$ edges from $y$, and $n-4$ from $z$ to the remaining vertices. Therefore, there are at most $5(n-4)$ additional edges. In either the standard or oriented versions of the problem, edges that do not contain vertices of $B$ must have their tails in $T$ and their heads in $K$. So there are at most \[\left\lfloor \frac{n-b}{3} \right\rfloor{\left\lceil \frac{2(n-b)}{3} \right\rceil \choose 2}\] edges that do not intersect $B$. Hence, \[|E(H)| < \left\lfloor \frac{n-b}{3} \right\rfloor{\left\lceil \frac{2(n-b)}{3} \right\rceil \choose 2} + cnb\]where $c=2$ in the oriented case and $c=5$ in the standard case. This expression is maximum on $b \in [0,n]$ only at the endpoint $b=0$ for all $n \geq 29$ when $c=2$ and for all $n \geq 70$ when $c=5$. Therefore, we can never do better than the lower bound construction. Moreover, since $B$ must be empty to reach this bound, then the construction is unique when $n \equiv 0,1 \text{ mod } 3$. When $n \equiv 2 \text{ mod } 3$, then \[\left\lfloor \frac{n}{3} \right\rfloor{\left\lceil \frac{2n}{3} \right\rceil \choose 2}=\left\lceil \frac{n}{3} \right\rceil{\left\lfloor \frac{2n}{3} \right\rfloor \choose 2}\]so there are exactly two extremal constructions in that case. \end{proof} \section{The 3-resolvent graph $R_3$} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (-1,0) circle (1pt); \filldraw[black] (1,0) circle (1pt); \filldraw[black] (0,2) circle (1pt); \filldraw[black] (2,1.75) circle (1pt); \draw[thick] (-1,0)--(1,0); \draw[thick,->] (0,0)--(0,2); \draw[thick] (1,0)--(0,2); \draw[thick,->] (0.5,1)--(2,1.75); \end{tikzpicture} \caption{$R_3$} \end{figure} In \cite{langlois2009}, the authors gave a simple construction for an $R_3$-free graph. Partition the vertices into sets $A$ and $B$ and take all possible edges with a tail set in $A$ and head vertex in $B$ plus all possible edges with a tail set in $B$ and a head in $A$. When there are $n$ vertices, this construction gives $(n-a){a \choose 2} + a{n-a \choose 2}$ edges where $a=|A|$. This is optimized when $a = \left\lceil \frac{n}{2} \right\rceil$. The authors showed that this number of edges is asymptotically optimal for $R_3$-free graphs. \begin{figure} \centering \begin{tikzpicture} [scale=0.5] \filldraw[color=black,fill=blue!5] (0,0) ellipse (2 and 3); \node at (-2,3) {$A$}; \filldraw[color=black,fill=blue!5] (6,0) ellipse (2 and 3); \node at (8,3) {$B$}; \filldraw[black] (0,-1) circle (1pt); \filldraw[black] (0,1) circle (1pt); \filldraw[black] (6,0.5) circle (1pt); \draw[thick] (0,-1) -- (0,1); \draw[thick,->] (0,0) -- (6,0.5); \filldraw[black] (6,-0.5) circle (1pt); \filldraw[black] (6,-2) circle (1pt); \filldraw[black] (1,-1.5) circle (1pt); \draw[thick] (6,-0.5) -- (6,-2); \draw[thick,->] (6,-1.25) -- (1,-1.5); \end{tikzpicture} \caption{The unique $R_3$-free extremal construction.} \end{figure} We show that in both the standard and the oriented versions of this problem that this construction is in fact the best that we can do. We will start with the oriented case since it is less technical. \subsection{The oriented version} \begin{theorem} \label{exE} For all $n$, \[ex_o(n,R_3) = \left\lfloor \frac{n}{2} \right\rfloor \left\lceil \frac{n}{2} \right\rceil \frac{n-2}{2}.\]Moreover, there is one unique extremal $R_3$-free construction up to isomorphism for each $n$. \end{theorem} \begin{proof} Let $H$ be an $R_3$-free graph on $n$ vertices. Consider the total link graph, $L_x$, for some $x \in V(H)$. If \[yz, z \rightarrow t \in E(L_x)\] or if \[y \rightarrow z, z \rightarrow t \in E(L_x)\] then $H$ is not $R_3$-free (See Figure~\ref{forLx}). \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (1,1) circle (1pt); \filldraw[black] (2,0) circle (1pt); \draw[thick] (0,0) -- (1,1); \draw[thick,->] (1,1) -- (2,0); \filldraw[black] (5,0) circle (1pt); \filldraw[black] (6,1) circle (1pt); \filldraw[black] (7,0) circle (1pt); \draw[thick,->] (5,0) -- (6,1); \draw[thick,->] (6,1) -- (7,0); \end{tikzpicture} \caption{Forbidden intersection types in $L_x$ for any vertex $x$ in an $R_3$-free graph.} \label{forLx} \end{figure} Let $U_x \subseteq V(L_x)$ be the set of vertices that appear as the tail vertex of some directed edge in $L_x$. Then no edges of $L_x$ can be contained entirely inside $U_x$ - it is an independent set with respect to both directed and undirected edges. Moreover, all undirected edges of $L_x$ must appear entirely within the complement, $C_x := V(L_x) \setminus U_x$. Hence, if we let $u_x = |U_x|$, then \[2|E(H)| = \sum_{x \in V(H)} |D_x| \leq \sum_{x \in V(H)} u_x(n-1-u_x).\] \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (0,0) ellipse (3 and 1); \filldraw[color=black,fill=blue!5] (0,3) ellipse (3 and 1); \node [right] at (3,0) {$C_x$}; \node [right] at (3,3) {$U_x$}; \filldraw[black] (0,3.1) circle (1pt); \filldraw[black] (-1,2.9) circle (1pt); \filldraw[black] (1,3) circle (1pt); \filldraw[black] (2,3.2) circle (1pt); \filldraw[black] (-2,3) circle (1pt); \filldraw[black] (0,0) circle (1pt); \filldraw[black] (-1,-0.4) circle (1pt); \filldraw[black] (1,-0.2) circle (1pt); \filldraw[black] (2,0.1) circle (1pt); \filldraw[black] (-2,-0.1) circle (1pt); \draw[thick,->] (0,3.1)--(1,-0.2); \draw[thick,->] (0,3.1)--(-2,-0.1); \draw[thick,->] (1,3)--(1,-0.2); \draw[thick,->] (2,3.2)--(2,0.1); \draw[thick,->] (2,3.2)--(1,-0.2); \draw[thick,->] (2,3.2)--(0,0); \draw[thick,->] (-1,2.9)--(-1,-0.4); \draw[thick,->] (-2,3)--(-1,-0.4); \draw[thick] (-2,-0.1)--(-1,-0.4); \draw[thick] (-1,-0.4)--(0,0); \draw[thick] (0,0)--(1,-0.2); \draw[thick] (1,-0.2)--(2,0.1); \draw[thick] (-2,-0.1)--(0,0); \end{tikzpicture} \caption{The structure of $L_x$ for any $x$ in an $R_3$-free graph} \label{Lxstructure} \end{figure} Each term of this sum is maximized when $u_x \in \left\{\left\lfloor \frac{n-1}{2} \right\rfloor, \left\lceil \frac{n-1}{2} \right\rceil\right\}$. Therefore, the result is immediate if $n$ is even. The situation is slightly more complicated for odd $n$. In this case, \[u_x(n-1-u_x) \leq \left(\frac{n-1}{2}\right)^2\] for each $x$. However, we need $u_x = \frac{n-1}{2}$ in order to attain this maximum value. This means that there are $\frac{n-1}{2}$ vertices in $C_x$, and so there are at most ${\frac{n-1}{2} \choose 2}$ edges in $T_x$. Therefore, if every $x \in V(H)$ gave $u_x = \frac{n-1}{2}$, then \[|E(H)| = \sum_{x \in V(H)} |T_x| < \frac{(n-2)(n-1)(n+1)}{8},\] which is given by the construction we want to show is optimal. For each $x$ let $i_x \in \{0,\ldots,\frac{n-1}{2}\}$ be the integer such that \[u_x(n-1-u_x) = \left(\frac{n-1}{2} - i_x\right)\left(\frac{n-1}{2}+i_x\right).\]Then, \[|E(H)| \leq \frac{1}{2} \sum_{x \in V(H)} \left(\frac{n-1}{2} - i_x\right)\left(\frac{n-1}{2}+i_x\right)= \frac{n(n-1)^2}{8} - \frac{1}{2} \sum_{j = 0}^{\frac{n-1}{2}} k_j j^2\] where $k_j$ is the number of vertices $x \in V(H)$ for which $i_x=j$. Since the construction gives $\frac{(n-2)(n-1)(n+1)}{8}$ for odd $n$, then we are only interested in beating this. So set \[\frac{(n-2)(n-1)(n+1)}{8} \leq \frac{n(n-1)^2}{8} - \frac{1}{2} \sum_{j = 0}^{\frac{n-1}{2}} k_j j^2.\] This gives \[\sum_{j = 0}^{\frac{n-1}{2}} k_j j^2 \leq \frac{n-1}{2}.\] Since we can also find $|E(H)|$ by counting the number of undirected edges over the $L_x$, then we can upper bound the number of these by assuming $u_x = \frac{n-1}{2} - i_x$ for each $x$ since this increases the size of $C_x$. This gives \[|E(H)| \leq \sum_{x \in V(H)} {\frac{n-1}{2} + i_x \choose 2}= \frac{n^3 - 4n^2 + 3n}{8} + \frac{1}{2} \sum_{j = 0}^{\frac{n-1}{2}} j(n+j-2)k_j.\] We can also set this greater than or equal to the known lower bound: \[\frac{(n-2)(n-1)(n+1)}{8} \leq \frac{n^3 - 4n^2 + 3n}{8} + \frac{1}{2} \sum_{j = 0}^{\frac{n-1}{2}} j(n+j-2)k_j\] to get \[\frac{(n-1)^2}{2} \leq \sum_{j = 0}^{\frac{n-1}{2}} k_j j^2 + (n-2) \sum_{j = 0}^{\frac{n-1}{2}} k_j j.\] Combining the inequalities gives \[0 \leq \sum_{j = 0}^{\frac{n-1}{2}} k_j (j - j^2).\] Since $j - j^2 < 0$ for any $j \geq 2$ and $j - j^2 = 0$ when $j=0,1$, then $k_j = 0$ for all $j \geq 2$. Moreover, once all these are set to zero we get that \[k_1 \leq \frac{n-1}{2} \leq k_1.\] Therefore, $k_1 = \frac{n-1}{2}$ and so $k_0 = \frac{n+1}{2}$ since $\sum k_j = n$. This gives the desired upper bound. Now we can show that the lower bound construction is the unique extremal example up to isomorphism. Let $H$ be an extremal example on $n$ vertices, and define a relation, $\sim$, on the vertices such that $x \sim y$ if and only if either $x = y$ or $y \in U_x$. This defines an equivalence relation on $V(H)$. Reflexivity and symmetry are both immediate. For transitivity note that the proof of the upper bound requires that every possible directed edge be taken from $U_x$ to $C_x$ for each $x \in V(H)$. Therefore, if we assume towards a contradiction that $y \in U_x$ and $z \in U_y$ but $z \not \in U_x$, then $z \in C_x$. So $xy \rightarrow z \in E(H)$ which means $z \in C_y$, a contradiction. When $n$ is even there must be exactly two equivalence classes each of size $\frac{n}{2}$. Similarly, when $n$ is odd there must be two equivalence classes of sizes $\frac{n-1}{2}$ and $\frac{n+1}{2}$. Therefore, the lower bound construction must be unique. \end{proof} \subsection{The standard version} \begin{theorem} \label{TypeE} For all $n \geq 6$, \[ex(n,R_3) = \left\lfloor \frac{n}{2} \right\rfloor \left\lceil \frac{n}{2} \right\rceil \frac{n-2}{2}.\]Moreover, there is one unique extremal $R_3$-free construction up to isomorphism for each $n$. \end{theorem} \begin{proof} Let $H$ be an $R_3$-free graph on $n$ vertices. Let $x \in V(H)$, and call any pair of vertices in $L_x$ a multiedge if they contain more than one edge. Let $V(L_x) = U_x \cup C_x \cup M_x$ where $M_x$ is the set of vertices that are incident to multiedges (that is, the minimal subset of vertices that contains all multiedges) and $U_x$ and $C_x$ are defined on the rest of the vertices as in Theorem~\ref{exE}. The goal is to show that if $M_x$ is nonempty for any vertex $x$, then $H$ has strictly fewer than the number of edges in the unique oriented construction given in Theorem~\ref{exE}. Therefore, that construction must be the unique extremal $R_3$ example for the standard problem as well. There are three possibilities for multiedges in $M_x$: two oppositely directed edges, one directed edge and one undirected edge, and one undirected edge with two oppositely directed edges. If $y,z \in M_x$ have two directed edges between them, then neither $y$ nor $z$ is incident to any other edge in $L_x$ since any incidence would create one of the two forbidden edge incidences of $L_x$ as discussed in the previous theorem. If $y$ and $z$ have only one directed edge (assume it is $y \rightarrow z$) and one undirected edge between them, then $y$ cannot be incident to any more edges for the same reason as before, but $z$ can be incident to undirected edges as well as directed edges with $z$ at the head. This means that $z$ may be the vertex of intersection of a star of these types of multiedges within $M_x$, and between any two such stars, the vertices of intersection may have an undirected edge between them, but no directed. \begin{figure} \centering \begin{tikzpicture} \filldraw[color=black,fill=blue!5] (1.5,-0.5) circle [radius=3]; \node at (-1.5,2.5) {$M_x$}; \filldraw[black] (0,0) circle (1pt); \filldraw[black] (1,0) circle (1pt); \filldraw[black] (0,-1) circle (1pt); \filldraw[black] (-0.707,-0.707) circle (1pt); \filldraw[black] (0,1) circle (1pt); \draw[thick, ->] (1,0) -- (0,0); \draw[thick, ->] (0,-1) -- (0,0); \draw[thick, ->] (-0.707,-0.707) -- (0,0); \draw[thick, ->] (0,1) -- (0,0); \filldraw[black] (2,2) circle (1pt); \filldraw[black] (1,2) circle (1pt); \draw[thick, ->] (1,2) -- (2,2); \filldraw[black] (2,1) circle (1pt); \filldraw[black] (2,0) circle (1pt); \filldraw[black] (3,0) circle (1pt); \draw[thick, ->] (3,0) -- (2,0); \draw[thick, ->] (2,1) -- (2,0); \filldraw[black] (0,-1.7) circle (1pt); \filldraw[black] (0,-2.7) circle (1pt); \draw[thick,<->] (0,-1.7) -- (0,-2.7); \filldraw[black] (1,-1.7) circle (1pt); \filldraw[black] (1,-2.7) circle (1pt); \draw[thick,<->] (1,-1.7) -- (1,-2.7); \filldraw[black] (2,-1.7) circle (1pt); \filldraw[black] (2,-2.7) circle (1pt); \draw[thick,<->] (3,-1.7) -- (3,-2.7); \filldraw[black] (3,-1.7) circle (1pt); \filldraw[black] (3,-2.7) circle (1pt); \draw[thick,<->] (2,-1.7) -- (2,-2.7); \end{tikzpicture} \caption{Example structure of $M_x$ with 3 single directed edge stars and 4 double directed pairs} \label{Mx} \end{figure} Therefore, the structure of the internal directed edges of $M_x$ looks like Figure~\ref{Mx} with only the vertices of intersection of the single directed edge stars able to accept more edges from the rest of $L_x$. Directed edges from the rest of the graph to $M_x$ must originate in $U_x$. Therefore, if $M_x$ consists of $d$ double directed edge pairs of vertices and $k$ single directed stars with the $i$th star containing $s_i$ vertices, then the total number of directed edges incident to vertices of $M_x$ is at most \[2d + \sum_{i=1}^k (s_i - 1 + u)\]where $u$ is the number of vertices in $U_x$. If we assume that $M_x$ is nonempty, then $|M_x| = m \geq 2$. The number of directed edges incident to or inside of $M_x$ is at most $m+k(u-1)$. Therefore, for $u \geq 2$, the number of directed edges incident to vertices of $M_x$ is maximized when the number of single directed edge stars is maximized. This is $\left\lfloor \frac{m}{2} \right\rfloor$ stars. Therefore, there are at most \[ \frac{m}{2} (u+1) \]directed edges incident to vertices of $M_x$. Thus, if $|C_x|=c$, then $L_x$ can have at most $uc + \frac{m}{2} (u+1)$ directed edges. And since $u \geq 2$, then \[uc + \frac{m}{2}(u+1) < u(c+m).\] So $L_x$ has strictly less directed edges than a complete bipartite graph on the same number of vertices would. In Theorem~\ref{exE} every $L_x$ needed to be a complete bipartite graph in terms of the directed edges in order for the maximum number of edges to be obtained, and only in the case of odd $n$ could some of these bipartitions be less than equal or almost equal. In those cases the parts could only have $\frac{n-1}{2} - 1$ and $\frac{n-1}{2} + 1$ vertices. Therefore, the only way that $u(c+m)$ could have more than this is if $u = c+m$ and so $u = \frac{n-1}{2}$. We assume that $m \geq 2$ and $u \geq 2$, but if both are equal to 2, then $c = u-m =0$ and $n=4$, a contradiction since $n$ is odd. Therefore, one of them must be strictly greater. So \[uc + \frac{m}{2}(u+1) < (u-1)(u+1) = \left(\frac{n-1}{2} - 1\right)\left(\frac{n-1}{2} + 1\right).\] This leaves only the cases where $u=0$ and $u=1$ which are trivial. So every link graph of $H$ that contains a multiedge has strictly less than $(\frac{n-1}{2})^2-1$ directed edges. This is enough to prove that an extremal $R_3$-free graph on an even number of vertices must be oriented. However, if there are an odd number of vertices it is possible that there could be enough directed link graphs with the maximum $\left(\frac{n-1}{2}\right)^2$ directed edges to make up the deficit for the directed link graphs with strictly less than $\left(\frac{(n-3)(n+1)}{4}\right)$ due to multiedges. In this case there would need to be at least $\frac{n+3}{2}$ vertices with directed link graphs that are complete bipartite graphs with parts of size $\frac{n-1}{2}$ each. Let $S$ be the set of these vertices. For any $x,y \in S$ define the relation $x \sim y$ if and only if $y \in U_x$. As in the proof of Theorem~\ref{exE}, this turns out to be an equivalence relation. By the definition of $S$ one equivalence class can hold at most $\frac{n+1}{2}$ vertices. So there must be two nonempty classes. Let these classes be $A$ and $B$ with $a$ and $b$ vertices respectively. Let $C$ be the set of vertices that are in every $U_x$ for $x \in A$ but not in $A$ itself. Similarly, let $D$ be the set of vertices that are in every $U_x$ for $x \in B$ but are not in $B$ itself. The sets $A$, $B$, $C$, and $D$ are disjoint. Let $c=|C|$ and $d=|D|$. Then \[a+c = b+d = \frac{n+1}{2},\] but $a+b+c+d \leq n$, a contradiction. This is enough to show the result. \end{proof} \section{The Escher graph $E$} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (0,2) circle (1pt); \filldraw[black] (2,1) circle (1pt); \filldraw[black] (2,-1) circle (1pt); \draw[thick] (0,0) -- (0,2); \draw[thick] (2,-1) -- (2,1); \draw[thick,->] (0,1) -- (2,1); \draw[thick,->] (2,0) -- (0,0); \end{tikzpicture} \caption{$E$} \label{Gpic} \end{figure} In this section, we will prove the following result on the maximum number of edges of a $E$-free. \begin{theorem} \label{mexG} For all $n$, \[ex(n,E) = {n \choose 3} + 2\]and there are exactly two extremal construction up to isomorphism for each $n \geq 4$. \end{theorem} But first we will prove the easier oriented version of the problem. This result will be needed to prove Theorem~\ref{mexG}. \subsection{The oriented version} \begin{theorem} \label{exG} For all $n$, \[ex_o(n,E) = {n \choose 3}\]and there is exactly one extremal construction up to isomorphism. \end{theorem} \begin{proof} The upper bound here is trivial so we need only come up with an $E$-free construction that uses ${n \choose 3}$ edges. Let $H$ be the directed hypergraph defined on vertex set $V(H) = [n]$ and edge set, \[E(H) = \left \{ ab \rightarrow c : a <b<c \right \}.\] That is take some linear ordering on the $n$ vertices and for each triple direct the edge to the largest vertex. Then every triple has an edge and $H$ contains no copy of $E$. Now we will show that this construction is unique. Let $H$ be an $E$-free graph on $n$ vertices and ${n \choose 3}$ edges. Define a relation on the vertices, $\prec$, where $x \prec y$ if and only if there exists an edge in $E(H)$ with $x$ in the tail and $y$ as the head vertex. Then $\prec$ is a partial ordering of the vertices that is almost linear in that every pair of vertices are comparable except for the two smallest elements (see Figure ~\ref{almostlin}). \end{proof} \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0.5) circle (1pt); \filldraw[black] (0,1.5) circle (1pt); \filldraw[black] (2,1) circle (1pt); \filldraw[black] (4,1) circle (1pt); \filldraw[black] (8,1) circle (1pt); \node at (1,1) {$<$}; \node at (3,1) {$<$}; \node at (5,1) {$<$}; \node at (6,1) {$\cdots$}; \node at (7,1) {$<$}; \node [left] at (0,1.5) {$1$}; \node [left] at (0,0.5) {$2$}; \node [above] at (2,1) {$3$}; \node [above] at (4,1) {$4$}; \node [above] at (8,1) {$n$}; \end{tikzpicture} \caption{An ``almost" linear ordering on the vertices of an $E$-free directed hypergraph.} \label{almostlin} \end{figure} We now shift our attention to the the standard version of the problem where a triple of vertices can have more than one edge. Here, both of the lower bound constructions are similar to the unique extremal construction in the oriented version. \subsection{Two lower bound constructions for $\text{ex}(n,E)$} The first construction is the same as the extremal construction in the oriented case but with two additional edges placed on the ``smallest" triple. That is, let $H_1 = ([n],E_1)$ where \[E_1 = \left \{ ab \rightarrow c : a <b<c \right \} \cup \{13 \rightarrow 2, 23 \rightarrow 1\}.\] See Figure~\ref{mexG1}. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (0,2) circle (1pt); \filldraw[black] (2,1) circle (1pt); \filldraw[black] (4,1) circle (1pt); \filldraw[black] (8,1) circle (1pt); \draw[thick] (0,0) -- (0,2); \draw[thick] (0,0) -- (2,1); \draw[thick] (0,2) -- (2,1); \draw[thick,->] (0,1) -- (2,1); \draw[thick,->] (1,0.5) -- (0,2); \draw[thick,->] (1,1.5) -- (0,0); \node at (3,1) {$\rightarrow$}; \node at (5,1) {$\rightarrow$}; \node at (6,1) {$\cdots$}; \node at (7,1) {$\rightarrow$}; \node [left] at (0,2) {$1$}; \node [left] at (0,0) {$2$}; \node [right] at (2,1) {$3$}; \node [above] at (4,1) {$4$}; \node [above] at (8,1) {$n$}; \end{tikzpicture} \caption{The first extremal construction, $H_1$, for an $E$-free directed hypergraph on $n$ vertices.} \label{mexG1} \end{figure} Moreover, it is important to note that if an $E$-free graph with ${n \choose 3} + 2$ edges has at least one edge on every vertex triple, then it must be isomorphic to $H_1$. This is because we can remove two edges to get an $E$-free subgraph where each triple has exactly one edge. Therefore, this must be the unique extremal construction established in Theorem~\ref{exG}. The only way to add two edges to this construction and avoid creating an Escher graph is to add the additional edges to the smallest triple under the ordering. The second construction is also based on the oriented extremal construction. Let $H_2=([n],E_2)$ where \[E_2 = \left( E_1 \setminus \{23 \rightarrow 4, 23 \rightarrow 1\} \right) \cup \{14 \rightarrow 2, 14 \rightarrow 3\}.\] See Figure~\ref{mexG2}. \begin{figure} \centering \begin{tikzpicture} \filldraw[black] (0,0) circle (1pt); \filldraw[black] (2,0) circle (1pt); \filldraw[black] (2,2) circle (1pt); \filldraw[black] (2,-2) circle (1pt); \filldraw[black] (4,0) circle (1pt); \filldraw[black] (8,0) circle (1pt); \draw[thick] (0,0) -- (2,0); \draw[thick] (0,0) -- (2,2); \draw[thick] (0,0) -- (2,-2); \draw[thick,->] (1,1) -- (2,0); \draw[thick,->] (1,1) -- (2,-2); \draw[thick,->] (1,0) -- (2,2); \draw[thick,->] (1,0) -- (2,-2); \draw[thick,->] (1,-1) -- (2,2); \draw[thick,->] (1,-1) -- (2,0); \node at (3,0) {$\rightarrow$}; \node at (5,0) {$\rightarrow$}; \node at (6,0) {$\cdots$}; \node at (7,0) {$\rightarrow$}; \node [left] at (0,0) {$1$}; \node [above] at (2,2) {$2$}; \node [below right] at (2,0) {$3$}; \node [below] at (2,-2) {$4$}; \node [above] at (4,0) {$5$}; \node [above] at (8,0) {$n$}; \draw[rounded corners, dashed] (1.6,2.5) -- (1.6,-2.5) -- (2.4,-2.5) -- (2.4,2.5) -- (1.6,2.5); \end{tikzpicture} \caption{The second extremal construction, $H_2$, for an $E$-free graph on $n$ vertices.} \label{mexG2} \end{figure} For the rest of this section we will show that any $E$-free graph is either isomorphic to one of these two constructions or has fewer than ${n \choose 3} + 2$ edges. Roughly speaking, the strategy is to take any $E$-free graph and show that we can add and remove edges to it so that we preserve $E$-freeness, remove most multiple edges from triples that had more than one, and never decrease the overall number of edges. \subsection{Add and Remove Edges} Let $H$ be an $E$-free graph and represent its vertices as the disjoint union of three sets: \[V(H) = D \cup R \cup T\] where $D$ (for `Done') is the set of all vertices that have complete graphs on three or more vertices as tail link graphs, $R$ (for `Ready to change') is the set of vertices not in $D$ that have at least three edges in their tail link graphs, and $T$ is the set of all other vertices (those with `Two or fewer edges in their tail link graphs'). The plan is now to remove and add edges in order make a new graph $H'$ which is also $E$-free, has at least as many edges as $H$, and whose vertices make a disjoint union, \[V(H') = D' \cup T'\] where $D'$ and $T'$ are defined exactly the same as $D$ and $T$ except in terms of the vertices of $H'$. That is, for each vertex $x \in R$, we will add all possible edges to complete $T_x$. This moves $x$ from $R$ to $D$. The edges removed will be all those that pointed from $x$ to a vertex that points to $x$. This will destroy triples with more than one edge as we go. The following observation will ensure that this procedure only ever moves vertices from $R$ to $D$, from $R$ to $T$, from $R$ to $R$, and from $T$ to $T$. Since each step moves one vertex from $R$ to $D$ and ends when $R$ is empty, then the procedure is finite. Here is the observation: \begin{lemma} \label{observation} Let $H$ be an $E$-free graph, and let $x,y \in V(H)$. If $d_x(y),d_y(x) > 0$, then $d_x(y) = d_y(x) = 1$. In other words, for any two vertices, $x$ and $y$, if $d_y(x) \geq 2$, then $d_x(y) = 0$. \end{lemma} \begin{proof} Suppose not. Let $d_x(y),d_y(x) > 0$ and suppose $d_x(y) \geq 2$. Then there exist two distinct vertices, $a$ and $b$ such that \[ay \rightarrow x, by \rightarrow x \in E(H).\] There also exists a vertex $c$ such that $xc \rightarrow y \in E(H)$. Since $c$ must be distinct from either $a$ or $b$ if not both, then this yields an Escher graph. \end{proof} Now, let us make the procedure slightly more formal: While there exist vertices in $R$, pick one, $x \in R$, and for each pair $a,b \in V(T_x)$, add the edge $ab \rightarrow x$ to $E(H)$ if it is not already an edge. Then, for each $a \in V(T_x)$, remove all edges of $E(H)$ of the form $xs \rightarrow a$ for any third vertex $s$. Since there were at least three edges in $T_x$, then the added edges will move $x$ from $R$ to $D$. The removed edges, if any, will only affect vertices in $R$ or in $T$ since if $xs$ is removed from $T_a$, then this implies that $a \in T_x$ and that $x \in T_a$ and so both had degree one in the other's tail link graph. Hence, $a \not \in D$. Moreover, an affected vertex in $R$ will either stay in $R$ or move to $T$ while an affected vertex in $T$ will stay in $T$ since it is only losing edges from its tail link graph. Moreover, at the end of this process $D'$ will contain no triple of vertices with more than one edge. Therefore, the only such triples of vertices of $H'$ will be entirely in $T'$ or will consist of vertices from both $T'$ and $D'$. We will show later that there cannot be too many of these triples. First, we need to show that after each step of this procedure, no Escher graph is created and at least as many edges are added to the graph as removed. \subsection{No copy of $E$ is created and the number of edges can only increase} Fix a particular vertex $x \in R$ to move to $D$. Add and remove all of the designated edges. Suppose that we have created an Escher graph. Since the only edges added point to $x$, then the configuration must be of the form, $ab \rightarrow x, xc \rightarrow a$ for some distinct vertices, $a$, $b$, and $c$. Therefore, $a \in V(T_x)$ and so $xc \rightarrow a$ would have been removed in the process. Now we will show that at least as many edges have been added to $H$ as removed by induction on the number of independent edges in $T_x$. Start by assuming there are 0 independent edges in $T_x$ and assume that there are $k$ vertices in $T_x$ that have degree one. Then at most $k$ edges will be removed. If $k=0$, then no edges are removed and there is a strict increase in the number of edges. If $k=1$, then let $y_1$ be the degree one vertex and let $y_2$ be the vertex it is incident to. Since $d_x(y_2) \neq 1$ and $d_x(y_2) \geq 1$, then $d_x(y_2) \geq 2$. So there exists a third vertex, $y_3$, and similarly, $d_x(y_3) \geq 2$ but $y_2$ is not adjacent to $y_1$. Hence, there exists a fourth vertex, $y_4$. So at most one edge is removed and at least two edges are added, $y_1y_3 \rightarrow x$ and $y_1y_4 \rightarrow x$. Therefore, there is a strict increase in the number of edges. If $k = 2$, then the fact that $T_x$ has at least three edges means that there must be at least two additional vertices in $T_x$. Hence, at most two edges are removed but at least three are added. If $k \geq 3$, then at most $k$ are removed but ${k \choose 2}$ are added which nets \[{k \choose 2} - k = \frac{k(k-3)}{2} \geq 0\] edges added. Now, for the induction step, assume that $T_x$ has $m>0$ independent edges and that the process on a $T_x$ with $m-1$ independent edges adds just as many edges as it removes. Let $yz$ be an independent edge in $T_x$ and let $A$ be the set of vertices of $T_x$ that are not $y$ or $z$. Since $T_x$ has at least three edges, then $A$ contains at least three vertices. Therefore, the number of added edges is at least 6 between $A$ and $\{y,z\}$. The number of edges removed from $T_y$ and $T_z$ together is at most 2. By assumption, the number of edges removed from the other tail link graphs of vertices in $A$ is offset by the number of edges added inside $A$. Therefore, there is a strict increase in the number of edges. To summarize, we have shown that $H'$ is an $E$-free graph such that \[|E(H)| \leq |E(H')|\] and \[V(H') = D' \cup T'\] such that any triple of vertices of $H'$ with more than one edge must intersect the set $T'$. We will now consider what is happening in $T'$ by cases. \subsection{Case 1: $|T'| \geq 5$} Let $T' = \{x_1,x_2,\ldots,x_t\}$ for $t \geq 5$. For each $x_i$ remove all edges of $H'$ that have $x_i$ as a head. By the definition of $T'$ this will remove at most $2t$ edges from $H'$. Next, add all edges to $T'$ that follow the index ordering. That is, for each triple $\{x_i,x_j,x_k\}$ add the edge that points to the largest index, $x_ix_j \rightarrow x_k$ where $i<j<k$. This will add ${t \choose 3}$ edges. The new graph has \[{t \choose 3} - 2t \geq 0\] more edges than $H'$. Moreover, it is $E$-free and oriented. Therefore, $|E(H)| < {n \choose 3}$. \subsection{Case 2: $|T'| \leq 4$ and there exists an $x \in T'$ such that $T_x$ is two independent edges} Assume that some $x \in T'$ has a tail link graph $T_x$ such that $ab, cd \in E(T_x)$ for four distinct vertices, $\{a,b,c,d\}$. If \[d_a(x) = d_b(x) = d_c(x) = d_d(x) = 1,\] then $a,b,c,d,x \in T'$, a contradiction of the assumption that $|T'| \leq 4$. Therefore, we can add the edges \[ac \rightarrow x, ad \rightarrow x, bc \rightarrow x, bd \rightarrow x\] and remove any edges that point to a vertex from $\{a,b,c,d\}$ with $x$ in the tail set. Because $x$ has zero degree in at least one of those tail link graphs, then we have removed at most three edges and added four, a strict increase. We have also not created any triples of vertices with more than one edge or any Escher graphs. We may now assume that $|T'| \leq 4$ and that the tail link graphs of vertices in $T'$ are never two independent edges. \subsection{Case 3: $|T'| =0,1,2$} First, note that if $H'$ has a triple with more than one edge $\{x,y,z\}$ then at least two of its vertices must be in $T'$ as a consequence of Lemma~\ref{observation}. Therefore, if $|T'|=0,1$, then $H'$ is oriented and so \[|E(H)| \leq |E(H')| < {n \choose 3}.\] Moreover, if $T'=\{x,y\}$ and $H'$ is not oriented, then any vertex triple with more than one edge must have two edges of the form, \[zx \rightarrow y,zy \rightarrow x\] for some third vertex $z$. If there exist two such vertices $z_1 \neq z_2$ that satisfy this, then there would be an Escher graph. Hence, there is at most one vertex triple with more than one edge and it would have at most two edges. Therefore, \[|E(H)| \leq |E(H')| \leq {n \choose 3}+1.\] \subsection{Case 4: $|T'| =3$} First, suppose that there exists a triple, $\{x,y,z\}$ with all three possible edges. Then $T' = \{x,y,z\}$. Since any triple with multiple edges must intersect $T'$ in at least two vertices, then any additional such triple would make an Escher graph with one of the edges in $T'$. Therefore, $H'$ has exactly one triple of vertices with all three edges on it and no others. So \[|E(H) \leq |E(H')| \leq {n \choose 3} + 2.\] Moreover, to attain this number of edges, no triple of vertices can be empty of edges. In this case,$H'$ must be isomorphic to the first construction $H_1$. Next, assume that no triple of vertices has all three edges and let $T' = \{x,y,z\}$. Therefore, $H'$ needs at least two triples of vertices that each hold two edges or else \[|E(H)| \leq |E(H')| \leq {n \choose 3} + 1\] automatically. Suppose one of the multiedges is $\{x,y,z\}$ itself. Then without loss of generality let the edges be $xy \rightarrow z$ and $xz \rightarrow y$. The second triple with two edges must have its third vertex in $D'$. Call this vertex $v$.The vertex $x$ cannot be in this second triple of vertices without creating an Escher graph. So the edges must be $vy \rightarrow z$ and $vz \rightarrow y$. But this also creates an Escher graph. Therefore, neither of the two triples that hold two edges are contained entirely within $T'$. So without loss of generality they must be $vx \rightarrow y, vy \rightarrow x$ and $wy \rightarrow z, wz \rightarrow y$. If $v \neq w$, then $vx, wz \in T_y$, a contradiction to our assumption that $T'$ contains no vertices with tail link graphs that are two independent edges. Hence, $v=w$. Since $v \in D$, then $T_v$ has at least three vertices. Moreover, since $v$ is in the tail link graphs of each vertex of $T'$, then none of these vertices can be in $T_v$. Remove all edges pointing to the vertices of $T'$. This is at most 6 edges. Add all possible edges with $v$ as the head and a tail set among the set $V(T_v) \cup \{x,y,z\}$. This adds at least 12 new edges. The new graph is oriented and $E$-free. Therefore, $|E(H)| < {n \choose 3}$. \subsection{Case 5: $|T'| =4$} First, assume that there is some triple $\{x,y,z\}$ that contains all three possible edges. As before, there are no additional triples with more than one edge. So \[|E(H)| \leq |E(H')| \leq {n \choose 3} + 2.\] The first construction $H_1$ is the unique extremal construction under this condition since all triples must be used at least once. So assume that all triples with more than one edge have two edges each. Then we must have at least two. Assume that one of them is contained within $T' = \{a,b,c,d\}$. Without loss of generality let it be $ab \rightarrow c, ac \rightarrow b$. Since the second such triple intersects $T'$ in at least two vertices, then it must intersect $\{a,b,c\}$ in at least one vertex. If it intersects $\{a,b,c\}$ in two vertices, then without loss of generality (to avoid a copy of $E$) the second triple must be of the form $ab \rightarrow x, ax \rightarrow b$. Hence, $x \in T'$ so $x=d$. But now there is no edge possible on $\{b,c,d\}$. Therefore, there must be a third such triple for $H'$ to have ${n \choose 3} +2$ edges. This triple must be $ac \rightarrow d, ad \rightarrow c$. And the only way to actually make it to the maximum number of edges now must be to have an edge on every other triple. Every triple of the form $\{b,c,s\}$ for $s \in D$ must have the edge $bc \rightarrow s$ since the other two options would create an Escher graph. Similarly, $bd \rightarrow s$ and $cd \rightarrow s$ are the only options for triples of the form $\{b,d,s\}$ and $\{c,d,s\}$ respectively. Next, any triple of the form $\{a,b,s\}$ must hold the edge $ab \rightarrow s$ since the other two edges create Escher graphs. Similarly, every triple of the forms $\{a,c,s\}$ and $\{a,d,s\}$ must hold the edges $ac \rightarrow s$ and $ad \rightarrow s$ respectively. Since each triple contained in $D$ holds exactly one edge, then the induced subgraph on $D$ must be isomorphic to the oriented extremal example of an $E$-free graph on $n-4$ vertices. Therefore, the entire graph $H'$ must be isomorphic to the second extremal construction $H_2$ in order to attain ${n \choose 3} + 2$ edges. So assume that the second triple with two edges intersects $\{a,b,c\}$ in only one vertex. Then these edges must be $xa \rightarrow d, xd \rightarrow a$. This can be the only additional triple with two edges. So to make it to ${n \choose 3} +2$ edges we need each triple to have an edge. However, the edge for $\{a,b,d\}$ is forced to be $ad \rightarrow b$ and the edge for $\{b,c,d\}$ is forced to be $bc \rightarrow d$. This makes an Escher graph. So \[|E(H)| \leq |E(H')| \leq {n \choose 3} + 1.\] Now assume that no vertex triple with multiple edges is contained entirely within $T'$, but assume that there are at least two such triples in $H'$. The only way that two triples could have distinct vertices in $D'$ is if they were of the forms (without loss of generality), $xa \rightarrow b, xb \rightarrow a$, and $yc \rightarrow d, yd \rightarrow c$. Otherwise, the pairs of the two triples that are in $T'$ would intersect resulting in either a copy of $E$ (if both triples use the same pair) or a vertex in $T'$ with two independent edges as a tail link graph. So there must be exactly two such triples. Therefore, all other triples of vertices must contain exactly one edge in order to reach ${n \choose 3}+2$ edges overall. To avoid the forbidden subgraph this edge must be $ab \rightarrow c$ for the triple $\{a,b,c\}$ and $cd \rightarrow a$ for the triple $\{a,c,d\}$. But this is an Escher graph. Hence, not all triples may be used and so \[|E(H)| \leq |E(H')| \leq {n \choose 3} + 1.\] Therefore, we may now assume for each multiedge triple that the vertex from $D'$ is always $x$. First, assume that there are only two such triples. As before, if we assume that the only two such triples are $xa \rightarrow b, xb \rightarrow a$ and $xc \rightarrow d, xd \rightarrow c$, then there can be not be an edge on both $\{a,b,c\}$ and $\{a,c,d\}$. Hence, there would be a suboptimal number of edges overall. On the other hand, if the only two such triples are adjacent in $T'$, then they are, without loss of generality, $xa \rightarrow b, xb \rightarrow a$ and $xb \rightarrow c, xc \rightarrow a$. In this case, no edge can go on the triple $\{a,b,c\}$ at all and so there are at most ${n \choose 3} + 1$ edges overall. Therefore, we must assume there are at least three such triples that meet at $x$. If these three triples make a triangle in $T'$, then they are $xa \rightarrow b, xb \rightarrow a$, $xb \rightarrow c, xc \rightarrow b$, and $xc \rightarrow a, xa \rightarrow c$. Again, there can be no edges on the triple $\{a,b,c\}$. Hence, every other triple must hold an edge to attain ${n \choose 3} + 2$ edges overall. On the triple $\{a,b,d\}$ this edge must be $ab \rightarrow d$ to avoid making a copy of $E$. Similarly, we must have the edges $ac \rightarrow d$ and $bc \rightarrow d$. But this means that $d \not \in T'$, a contradiction. On the other hand, if there are three triples of vertices with more than one edge on each that do not make a triangle in $T'$ or if there are four or more such triples, then $x$ is in the tail link graphs for each vertex in $T'$. Hence, none of these vertices may be in the tail link graph, $T_x$. However, $x \in D'$ so its tail link graph has at least three vertices. Remove all edges pointing to vertices of $T'$ (at most 8). Add all edges pointing to $x$ with tail sets in $T'$ (6 new edges) and between $T'$ and $V(T_x)$ (at least 12 new edges). So this adds at least ten edges to $H'$ to create $H''$. $H''$ is oriented so \[|E(H)| < |E(H'')| \leq {n \choose 3}.\] This exhausts all of the cases and establishes that \[\text{ex}(n,E) = {n \choose 3} +2\] with exactly two extremal examples up to isomorphism. \section{Conclusion} In \cite{brown1969}, Brown and Harary started studying extremal problems for directed $2$-graphs by determining the extremal numbers for many ``small" digraphs and for some specific types of digraphs such as tournaments - a digraph where every pair of vertices has exactly one directed edge. We could follow their plan of attack in studying this $2 \rightarrow 1$ model and also look for the extremal numbers of tournaments. Here, a tournament would be a graph with exactly one directed edge on every three vertices. In particular, a transitive tournament might be an interesting place to begin. A transitive tournament is a tournament where the direction of each edge is based on an underlying linear ordering of the vertices as in the oriented lower bound construction in Theorem~\ref{exG}. Denote the $2 \rightarrow 1$ transitive tournament on $k$ vertices by $TT_k$. Since the ``winning" vertex of the tournament will have a complete $K_{k-1}$ as its tail link graph, then any $H$ on $n$ vertices for which each $T_{x}$ is $K_{k-1}$-free must be $TT_k$-free. Therefore, \[n \left(\frac{n-1}{k-2}\right)^2 {k-2 \choose 2} \leq \text{ex}(n,TT_k), \text{ex}_o(n,TT_k).\] This also immediately shows that the transitive tournament on four vertices with the ``bottom" edge removed has this extremal number exactly. \begin{theorem} Let $TT_4^-$ denote the graph with vertex set $V(TT_4^-) = \{a,b,c,d\}$ and edge set \[E(TT_4^-)=\{ab \rightarrow d, bc \rightarrow d, ac \rightarrow d\}.\] Then \[\text{ex}(n,TT_4^-) = n \left\lfloor \frac{n-1}{2} \right\rfloor \left\lceil \frac{n-1}{2} \right\rceil.\] \end{theorem} Is it still true if we add an edge to $\{a,b,c\}$? \begin{conjecture} Let $TT_4$ denote the graph with vertex set $V(TT_4) = \{a,b,c,d\}$ and edge set \[E(TT_4)=\{ab \rightarrow d, bc \rightarrow d, ac \rightarrow d, ab \rightarrow c\}.\] Then \[\text{ex}(n,TT_4) = n \left\lfloor \frac{n-1}{2} \right\rfloor \left\lceil \frac{n-1}{2} \right\rceil.\] \end{conjecture} Another way of generalizing the extremal questions asked in this paper is to ask about $r \rightarrow 1$ models of directed hypergraphs. If we look at every $(r \rightarrow 1)$-graph with exactly two edges we see that these fall into four main types of graph. Let $i$ be the number of vertices that are in the tail set of both edges. Then let $T_r(i)$ denote the graph where both edges point to the same head vertex, let $H_r(i)$ denote the graph where the edges point to different head vertices neither of which are in the tail set of the other, let $R_r(i)$ denote the graph where the first edge points to a head vertex in the tail set of the second edge and the second edge points to a head not in the tail set of the first edge, and let $E_r(i)$ denote the graph where both edges point to heads in the tail sets of each other. So in terms of the graphs discussed in this paper, the 3-resolvent would be a $R_2(1)$, the 4-resolvent would be a $R_2(0)$, the Escher graph would be a $E_2(0)$, and two edges on the same triple of vertices would be an $E_2(1)$. The nondegenerate cases here would be $R_r(i)$ and $E_r(i)$. It would be interesting to find the extremal numbers for these graphs in general. To what extent do the current proofs extend to these graphs? For example, any $r \rightarrow 1$ transitive tournament on $n$ vertices would be $E_r(i)$-free. This solves the oriented version and gives a lower bound for the standard version: \[\text{ex}_o(n,E_r(i)) = {n \choose r+1}.\] For the generalized resolvent configurations, the lower bound extremal constructions for $R_3$ and $R_4$ generalize easily to the $r \rightarrow 1$ setting, but are they ever tight? Can we generalize these constructions to get extremal numbers for all $R_r(i)$?
4c4bb3a9f4dad4f9fcd9559dc34711ca86db1f7a
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} \label{sec:Intro} Wireless communication, by its inherent broadcast nature, is vulnerable to eavesdropping by illegitimate receivers within communication range of the source. Wyner in \cite{075wyner}, for the first time, information-theoretically addressed the problem of secure communication in the presence of an eavesdropper and showed that secure communication is possible if the eavesdropper channel is a degraded version of the destination channel. The rate at which information can be transferred from the source to the intended destination while ensuring complete equivocation at the eavesdropper is termed as \textit{secrecy rate} and its maximum over all input probability distributions is defined as the \textit{secrecy capacity} of channel. Later, \cite{078cheongHellman} extended Wyner's result to Gaussian channels. These results are further extended to various models such as multi-antenna systems \cite{105paradaBlahut, 110khistiWornell}, multiuser scenarios \cite{108liuMaric, 108khisti_thesis}, fading channels \cite{108liangPoorShamai, 108gopalaLaiGamal}. An interesting direction of work on secure communication in the presence of eavesdropper(s) is one in which the source communicates with the destination via relay nodes \cite{108laiGamal, 109dongHanPetropuluPoor, 110zhangGursoy, 113yangLiMaChing, 115sarmaAgnihotriKuri, 116sarmaAgnihotriKuri}. Such work has considered various scenarios such as different relaying schemes (\textit{amplify-and-forward} and \textit{decode-and-forward}), constraints on total or individual relay power consumption, one or more eavesdroppers. However, except for a few specific scenarios, such work does not provide tight characterization of secrecy capacity or even optimal secrecy rate achievable with the given relaying scheme. Further, all previous work only considered secure communication scenarios where the source communicates with the legitimate destination(s) in two hops, over so called \textit{diamond network} \cite{101schein}. We consider a multihop unicast communication over \textit{layered} network of relays in the presence of a single eavesdropper. The relays nodes are arranged in layers where all relays in a particular layer can communicate only with the relays in the next layer. The relay nodes, operating under individual power constraints, amplify-and-forward the signals received at their input. In this scenario, multiple relay nodes in each layer can cooperate to enhance the end-to-end achievable rate. Also, the signals transmitted simultaneously by the relays add in the air, thus providing an opportunity for the relays in the second layer onward to perform Analog Network Coding (ANC) on the received \textit{noisy sum} of these signals, where each relay merely amplifies and forwards this noisy sum \cite{107kattiGollakottaKatabi, 110maricGoldsmithMedard}. The eavesdropper can overhear the transmissions from the relay nodes of any of the layers depending on its location. The objective is to maximize the rate of secure transmission from the source to the destination by choosing the optimal set of scaling factors for the ANC-relays, irrespective of the relays that the eavesdropper listens to. However, so far, there exists no closed-form expression or polynomial time algorithm to exactly characterize the optimal AF secrecy rate even for general two-hop (diamond) relay networks, except for a few specific cases where eavesdropper's channel is a degraded or scaled version of destination channel \cite{116sarmaAgnihotriKuri} and characterizing the optimal AF secrecy rate for general layered network is an even harder problem than general diamond network. Thus, to get some insights into the nature of the optimal solution for such networks, we consider symmetric layered networks, where all channel gains between the nodes in two adjacent layers are equal, thus the nomenclature of these networks as \textit{``Equal Channel Gains between Adjacent Layers (ECGAL)''} networks \cite{112agnihotriJaggiChen}. We provide closed-form solutions for the optimal secure AF rate for such networks. We envision that these results may help us gain insight into the nature of the optimal solution and develop techniques which may further help in construction of low-complexity optimal schemes for general relay networks. The eavesdropper being a passive entity, a realistic eavesdropper scenario is the one where nothing about the eavesdropper's channel is known, neither its existence, nor its channel state information (CSI). However, the existing work on secrecy rate characterization assumes one of the following: (1) the transmitter has prefect knowledge of the eavesdropper channel states, (2) \textit{compound channel:} the transmitter knows that the eavesdropper channel can take values from a finite set \cite{109liangKramerPoorShamai, 109kobayashiLiangShamaiDebbah, 109ekremUlukus}, and (3) \textit{fading channel:} the transmitter only knows distribution of the eavesdropper channel \cite{108liangPoorShamai, 108gopalaLaiGamal}. In this paper, we assume that the CSI of the eavesdropper channel is known perfectly for the following two reasons. First, this provides an upper bound to the achievable secure ANC rate for the scenarios where we have imperfect knowledge of the eavesdropper channel. For example, the lower (upper) bound on the compound channel problem can be computed by solving the perfect CSI problem with the worst (best) channel gain from the corresponding finite set. Further, this also provides a benchmark to evaluate the performance of achievability schemes in such imperfect knowledge scenarios. Second, this assumption allows us to focus on the nature of the optimal solution and information flow, instead of on complexities arising out of imperfect channel models. The key contribution of this work is the computation of the globally optimal set of scaling factors for the nodes that maximizes the end-to-end secrecy rate for a class of layered networks. We also show that in the high-SNR regime, ANC achieves secrecy rates within an explicitly computed constant gap of the cutset upper bound on the secrecy capacity. To the best of our knowledge, this work offers the first characterization of the performance of secure ANC in multi-layered networks in the presence of an eavesdropper. \textit{Organization:} In Section~\ref{sec:sysMdl} we introduce the system model and formulate the problem of maximum secure ANC rate achievable in the proposed system model. In section~\ref{sec:OptBeta} we compute the optimal vector of scaling factors of the nodes of an ECGAL network when eavesdropper snoops on the transmissions of the nodes in any one of the $L$ layers. Then, in Section~\ref{sec:highSNRanalysis} we analyze the high-SNR behavior of the achievable secure ANC rate and show that it lies within a constant gap from the corresponding cutset upper bound on the secrecy capacity and Section~\ref{sec:numSim} we numerically validate these results. Finally, Section~\ref{sec:cnclsn} concludes the paper. \section{System Model} \label{sec:sysMdl} Consider a $(L+2)$-layer wireless network with directed links. The source $s$ is at layer `$0$', the destination $t$ is at layer `$L+1$' and the relays from the set $R$ are arranged in $L$ layers between them. The $l^{th}$ layer contains $n_l$ relay nodes, $\sum _{l-1}^{L} n_l = |R|$. The source $s$ transmits message signals to the destination $t$ via $L$ relay layers. However, the signals transmitted by the relays in a layer are also overheard by the eavesdropper $e$. An instance of such a network is given in Figure~\ref{fig:layrdNetExa}. Each node is assumed to have a single antenna and operate in full-duplex mode, \textit{e.g.} as in \cite{112agnihotriJaggiChen, 105gastparVetterli}. \begin{figure}[!t] \centering \includegraphics[width=3.5in]{layrdNetExa} \caption{An ECGAL network with 3 relay layers between the source $s$ and the destination $t$. Each layer contains two relay nodes. The eavesdropper overhears the transmissions from the relays in layer $2$.} \label{fig:layrdNetExa} \end{figure} At instant $n$, the channel output at node $i, i \in R \cup \{t, e\}$, is \begin{equation} \label{eqn:channelOut} y_i[n] = \sum_{j \in {\mathcal N}(i)} h_{ji} x_j[n] + z_i[n], \quad - \infty < n < \infty, \end{equation} where $x_j[n]$ is the channel input of node $j$ in neighbor set ${\mathcal N}(i)$ of node $i$. In \eqref{eqn:channelOut}, $h_{ji}$ is a real number representing the channel gain along the link from the node $j$ to the node $i$ and constant over time (as in \cite{112agnihotriJaggiChen}, for example) and known (even for the eavesdropper channels) throughout the network \cite{109dongHanPetropuluPoor,110zhangGursoy}. All channel gains between the nodes in two adjacent layers are assumed to be equal, thus the nomenclature of these networks as \textit{``Equal Channel Gains between Adjacent Layers (ECGAL)''} networks \cite{112agnihotriJaggiChen}. The source symbols $x_s[n], - \infty < n < \infty$, are i.i.d. Gaussian random variables with zero mean and variance $P_s$ that satisfy an average source power constraint, $x_s[n] \sim {\cal N}(0, P_s)$. Further, $\{z_i[n]\}$ is a sequence (in $n$) of i.i.d. Gaussian random variables with $z_i[n] \sim {\cal N}(0, \sigma^2)$. We assume that $z_i$ are independent of the input signal and of each other. We also assume that each relay's transmit power is constrained as: \begin{equation} \label{eqn:pwrConstraint} E[x_i^2[n]] \le P, \quad i \in R, - \infty < n < \infty \end{equation} In ANC, each relay node amplifies and forwards the noisy signal sum received at its input. More precisely, relay node $i, i \in R$, at instant $n+1$ transmits the scaled version of $y_i[n]$, its input at time instant $n$, as follows \begin{equation} \label{eqn:AFdef} x_i[n+1] = \beta_i y_i[n], \quad 0 \le \beta_i^2 \le \beta_{i,max}^2 = P/P_{R,i}, \end{equation} where $P_{R,i}$ is the received power at node $i$ and choice of scaling factor $\beta_i$ satisfies the power constraint \eqref{eqn:pwrConstraint}. Assuming equal delay along each path, for the network in Figure \ref{fig:layrdNetExa}, the copies of the source signal ($x_s[.]$) and noise signals ($z_i[.]$), respectively, arrive at the destination and the eavesdropper along multiple paths of the same delay. Therefore, the signals received at the destination and eavesdropper are free from intersymbol interference (ISI). Thus, we can omit the time indices and use equations \eqref{eqn:channelOut} and \eqref{eqn:AFdef} to write the input-output channel between the source $s$ and the destination $t$ as \begin{equation} \label{eqn:destchnl} y_t = \left[\sum\limits_{(i_1,...,i_L) \in K_{st}}\!\!\!\!\!\!\!\!\! h_{s,i_1}\beta_{i_1}h_{i_1, i_2}...h_{i_{L-1}, i_L}\beta_{i_L} h_{i_L, t}\right] x_s + \sum\limits_{l=1}^L \sum\limits_{j-1}^{n_l}\left[\sum\limits_{(i_1,...,i_{L-l}) \in K_{lj,t}} \!\!\!\!\!\!\!\!\!\beta_{lj} h_{lj,i_1}...\beta_{i_{L-l}} h_{i_{L-l},t}\right] z_{lj} + z_t \end{equation} where $K_{st}$ is the set of $L$-tuples of node indices corresponding to all paths from the source $s$ to the destination $t$ with path delay $L$. Similarly, $K_{lj,t}$ is the set of $L - l$- tuples of node indices corresponding to all paths from the $j^{th}$ relay of the $l^{th}$ layer to the destination with path delay $L - l + 1$. We introduce modified channel gains as follows. For all the paths between the source and the destination: \begin{equation} h_{st} = \sum\limits_{(i_1,...,i_L) \in K_s} h_{s,i_1}\beta_{i_1}h_{i_1, i_2}...h_{i_{L-1}, i_L}\beta_{i_L} h_{i_L, t} \end{equation} For all the paths between the $j^{th}$ relay of the $l^{th}$ layer to the destination $t$ with path delay $L - l + 1$: \begin{equation} h_{lj,t} = \sum\limits_{(i_1,...,i_{L-l}) \in K_{lj}} \beta_{lj} h_{lj,i_1}...\beta_{i_{L-l}} h_{i_{L-l},t} \end{equation} In terms of these modified channel gains, the source-destination channel in \eqref{eqn:destchnl} can be written as: \begin{equation} \label{eqn:destchnlmod} y_t = h_{st} x_s + \sum_{l=1}^{L} \sum_{j=1}^{n_l} h_{lj,t} z_{lj} + z_t, \end{equation} Similarly, assuming that the eavesdropper is overhearing the transmissions of the relays in the layer $E, 1\le E \le L$, the input-output channel between the source and the eavesdropper can be written as \begin{equation} \label{eqn:evechnlmod} y_e = h_{se} x_s + \sum_{l=1}^{E} \sum_{j=1}^{n_l} h_{lj,e} z_{lj} + z_t, \end{equation} The secrecy rate at the destination for such a network model can be written as \cite{075wyner}, $R_s(P_s)= [I(x_s;y_t)-I(x_s;y_e)]^+$, where $I(x_s;y)$ represents the mutual information between random variable $x_s$ and $y$ and $[u]^+=\max\{u,0\}$. The secrecy capacity is attained for the Gaussian channels with the Gaussian input $x_s \sim \mathcal{N}(0,P_s)$, where $\mathbf{E}[x_s^2] = P_s$, \cite{078cheongHellman}. Therefore, for a given network-wide scaling vector $\bm{\beta} = (\beta_{li})_{1 \le l \le L, 1 \le i \le n_l}$, the optimal secure ANC rate for the channels in \eqref{eqn:destchnlmod} and \eqref{eqn:evechnlmod} can be written as the following optimization problem. \begin{subequations} \label{eq:optSecrate} \begin{align} R_s(P_s) &= \max_{\bm{\beta}} \left[R_t(P_s,\bm{\beta}) - R_e(P_s,\boldsymbol{\beta})\right]\\ &= \max_{\boldsymbol{\beta}} \left[\frac{1}{2}\log\frac{1+SNR_t(P_s,\bm{\beta})}{1+SNR_e(P_s,\bm{\beta})}\right], \end{align} \end{subequations} where $SNR_t(P_s,\bm{\beta})$, the signal-to-noise ratio at the destination $t$ is: \begin{equation} \label{eqn:snrt} SNR_t(P_s,\bm{\beta}) = \frac{P_s}{\sigma^2}\frac{h_{st}^2}{1 + \sum_{l=1}^{L} \sum_{j=1}^{n_l} h_{lj,t}^2} \end{equation} and similarly, $SNR_e(P_s,\bm{\beta})$ is \begin{equation} \label{eqn:snre} SNR_e(P_s,\bm{\beta}) = \frac{P_s}{\sigma^2}\frac{h_{se}^2}{1 + \sum_{l=1}^{E} \sum_{j=1}^{n_l} h_{lj,e}^2} \end{equation} Given the monotonicity of the $\log(\cdot)$ function, we have \begin{align} \bm{\beta}_{opt} &= \operatornamewithlimits{argmax}_{\bm{\beta}} \left[R_t(P_s,\bm{\beta}) - R_e(P_s,\boldsymbol{\beta})\right] \nonumber\\ &= \operatornamewithlimits{argmax}_{\bm{\beta}} \frac{1+SNR_t(P_s,\bm{\beta})}{1+SNR_e(P_s,\bm{\beta})} \label{eqn:eqProb} \end{align} \section{The Optimal Secure ANC Rate Analysis} \label{sec:OptBeta} In this section, we analyze the optimal secure ANC rate problem in \eqref{eq:optSecrate} or \eqref{eqn:eqProb} first for diamond networks and then for ECGAL networks. \begin{figure}[!t] \centering \includegraphics[width=3.0in]{diamondNet} \caption{A symmetric $N$ relay diamond network with an eavesdropper.} \label{fig:diamondNet} \end{figure} \subsection{Symmetric Diamond Networks} \label{subsec:diamondNetOptBeta} Consider a symmetric diamond with $N$ relay nodes arranged in a layer between the source and the destination as shown in Figure~\ref{fig:diamondNet}. Using \eqref{eqn:snrt} and \eqref{eqn:snre}, the $SNR_t$ and $SNR_e$ in this case are: \begin{align*} SNR_t = \frac{P_s h_s^2 }{\sigma^2} \frac{(\sum_{i=1}^{N} \beta_i )^2 h_t^2}{1 + \left(\sum_{i=1}^{N} \beta_i^2 \right) h_t^2 } &&\mbox{ and }&& SNR_e = \frac{P_s h_s^2 }{\sigma^2} \frac{(\sum_{i=1}^{N} \beta_i )^2 h_e^2}{1 + \left(\sum_{i=1}^{N} \beta_i^2 \right) h_e^2 } \end{align*} \begin{pavikl} \label{lemma:diamondNetReducedBeta1} For symmetric diamond network, $\bm{\beta}_{opt}$ in \eqref{eqn:eqProb} is: \begin{align*} \beta_{1,opt},\cdots,\beta_{N,opt}=\beta_{opt}= \begin{cases} \min(\beta_{max}^2, \beta_{glb}^2), \quad \mbox{if } h_{t}>h_e,\\ 0, \quad \mbox{otherwise} \end{cases} \end{align*} where \begin{align*} \beta_{glb}^2 &= \sqrt{\frac{1}{N^2 h_t^2 h_e^2 \left(1+N\frac{P_s h_s^2}{\sigma^2} \right)}} \end{align*} \end{pavikl} \begin{IEEEproof} Please refer to Appendix~\ref{appndx:lemma0}. \end{IEEEproof} Here, it is assumed that the eavesdropper chooses to snoop on all the nodes of the layer which is an optimal strategy in the symmetric layered networks for the eavesdropper as we prove later. However, in general this may not be the case. The eavesdropper can choose to snoop on fewer nodes and still get higher rate compared to the case when it snoops on all the nodes of a layer as illustrated in the following example. \begin{pavike} Consider the relay network shown in Figure~\ref{fig:counterex}. Let $h_s=0.6,\ h_t=0.3,\ h_{1e}=0.2,\ h_{2e}=0.6,\ h_{3e}=0.4$. Let $P_s=P_1=P_2=P_3=5$ and noise variance $\sigma^2=1.0$ at each node. \begin{figure}[!t] \centering \includegraphics[width=2.5in]{counterexample.pdf} \caption{3 relay diamond network with eavesdropper overhearing the transmissions of all the relay nodes.} \label{fig:counterex} \end{figure} \textit{Case 1:} Eavesdropper chooses to snoop on all the relay nodes.\\ In this case we have the following secrecy rate maximization problem: \begin{align*} R_s &= \max_{\beta_1, \beta_2, \beta_3} \left\{ \frac{1}{2} \log\left(\!1\!\!+\!\!\frac{P_s h_s^2}{\sigma^2} \frac{\left(\beta_1\!+\!\beta_2\!+\!\beta_3\right)^2 h_t^2}{1\!+\!\left(\beta_1^2\!+\!\beta_2^2\!+\!\beta_3^2\right)h_t^2}\!\right) - \frac{1}{2} \log\left(1\!\!+\!\!\frac{P_s h_s^2}{\sigma^2} \frac{\left(\beta_1 h_{1e} \!+\! \beta_2 h_{2e} \!+\! \beta_3 h_{3e}\right)^2}{1\!+\!\beta_1^2 h_{1e}^2 \!+\! \beta_2^2 h_{2e}^2 \!+\! \beta_3^2 h_{3e}^2}\right)\right\}\\ &= \max_{\beta_1, \beta_2, \beta_3} \left\{ \frac{1}{2} \log\left(1\!+\! \frac{0.162\left(\beta_1\!+\!\beta_2\!+\!\beta_3\right)^2 }{1\!+\!\left(\beta_1^2\!+\!\beta_2^2\!+\!\beta_3^2\right)0.09}\right) - \frac{1}{2} \log\left(1\!+\! \frac{1.8\left(0.2 \beta_1 \!+\! 0.6 \beta_2 \!+\! 0.4 \beta_3\right)^2}{1\!+\!0.04 \beta_1^2 \!+\! 0.36 \beta_2^2 \!+\! 0.16 \beta_3^2 }\right)\right\} \end{align*} The optimal solution of this problem is $\beta_1=\beta_{1,max} = 1.3363, \ \beta_2 = 0.0, \ \beta_3 = 0.0$. For these optimum values of $\beta$'s, the rate achievable at the eavesdropper is: \begin{align*} R_e &= \frac{1}{2} \log\left(1+ \frac{1.8\left(0.2 \beta_1 + 0.6 \beta_2 + 0.4 \beta_3\right)^2}{1+0.04 \beta_1^2 + 0.36 \beta_2^2 + 0.16 \beta_3^2 }\right) = 0.081749 \mbox{ bits/s/Hz} \end{align*} \textit{Case 2:} Eavesdropper chooses to snoop on relay nodes $2$ and $3$. \\ In this case we have the following secrecy rate maximization problem: \begin{align*} R_s &= \max_{\beta_1, \beta_2, \beta_3} \left\{ \frac{1}{2} \log\left(1+\frac{P_s h_s^2}{\sigma^2} \frac{\left(\beta_1+\beta_2+\beta_3\right)^2 h_t^2}{1+\left(\beta_1^2+\beta_2^2+\beta_3^2\right)h_t^2}\right) - \frac{1}{2} \log\left(1+\frac{P_s h_s^2}{\sigma^2} \frac{\left(\beta_2 h_{2e} + \beta_3 h_{3e}\right)^2}{1+\beta_2^2 h_{2e}^2 + \beta_3^2 h_{3e}^2}\right)\right\}\\ &= \max_{\beta_1, \beta_2, \beta_3} \left\{ \frac{1}{2} \log\left(1+ \frac{0.162\left(\beta_1+\beta_2+\beta_3\right)^2 }{1+\left(\beta_1^2+\beta_2^2+\beta_3^2\right)0.09}\right)- \frac{1}{2} \log\left(1+ \frac{1.8\left( 0.6 \beta_2 + 0.4 \beta_3\right)^2}{1+ 0.36 \beta_2^2 + 0.16 \beta_3^2 }\right)\right\} \end{align*} The optimal solution of this problem is $\beta_1=\beta_{1,max} = 1.3363, \ \beta_2 = 0.0, \ \beta_3 = 0.7298$. For these optimum values of $\beta$'s, the rate achievable at the eavesdropper is: \begin{align*} R_e &= \frac{1}{2} \log\left(1+ \frac{1.8\left( 0.6 \beta_2 + 0.4 \beta_3\right)^2}{1+ 0.36 \beta_2^2 + 0.16 \beta_3^2 }\right) = 0.095368 \mbox{ bits/s/Hz} \end{align*} From above it is clear that in the asymmetric diamond networks, the eavesdropper achieves a higher rate when it chooses to snoop on two nodes ($2$ and $3$) compared to the case where eavesdropper snoops on all the three nodes.\hfill\IEEEQEDclosed \end{pavike} Although, for general layered networks it is very difficult to find which subset of relay nodes the eavesdropper chooses to snoop on so as to maximize its rate; for symmetric networks it can be easily verified that the rate at the eavesdropper is maximized when it snoops on all the nodes of a layer. For instance, consider the scenario where all the nodes transmitting at maximum power is optimum with respect to secrecy rate maximization. The SNR at the eavesdropper when it chooses to snoop on $k$ nodes is given as \begin{align} SNR_e^k & = \frac{P_s h_s^2}{\sigma^2} \frac{k^2 \beta_{max}^2 h_e^2}{1+ k \beta_{max}^2 h_e^2} \end{align} Similarly, SNR at the eavesdropper when it chooses to snoop on $k+1$ nodes is \begin{align} SNR_e^{k+1} & = \frac{P_s h_s^2}{\sigma^2} \frac{(k+1)^2 \beta_{max}^2 h_e^2}{1+ (k+1) \beta_{max}^2 h_e^2} \end{align} Clearly, \begin{align*} SNR_e^{k+1}-SNR_e^k & = \frac{P_s h_s^2}{\sigma^2} \frac{{\beta}^{2} h_e^2 \left( {\beta}^{2} h_e^2 {k}^{2}+{\beta}^{2} h_e^2 k+2 k+1\right)}{\left( {\beta}^{2} h_e^2 k+1\right) \left( {\beta}^{2} h_e^2 k+{\beta}^{2} h_e^2+1\right) } \geq 0, \end{align*} i.e., $SNR_e^{k+1} \geq SNR_e^k$. Thus, for such scenarios eavesdropper achieves a higher rate when it chooses to snoop on more number of nodes and eventually it will snoop on maximum possible number of nodes to maximize its rate. \begin{figure}[!t] \centering \includegraphics[width=4.5in]{3relay_diamond_net_Re_plot.pdf} \caption{Achievable rate at the eavesdropper when it snoops on different number of nodes of a symmetric diamond network of three relay nodes with $P=10.0,\ \sigma^2=1.0, \ h_s=0.278, \ h_t=0.379, \ h_e=0.073$.} \label{fig:3relayDiamondNetRePlot} \end{figure} Figure~\ref{fig:3relayDiamondNetRePlot} shows the rates achievable at the eavesdropper $(R_{e,i})$ when it chooses to snoop on $i,\ i \in \{1,2,3\}$ number of nodes of a 3 relay symmetric diamond network of the specified parameters with all the nodes transmitting at their corresponding optimum values so as to maximize the secrecy rate in each case. Here, again it can be seen that rate at the eavesdropper is increases when it snoops on all the nodes. \subsection{ECGAL Layered Networks} \label{subsec:lyrNetOptBeta} In this subsection, we consider the optimal secure ANC rate problem in \eqref{eq:optSecrate} for ECGAL networks where the source communicates with the destination via $L$ intermediate relay layers with all channel gains between two adjacent layers being equal. For the sake of ease of representation, let there be $N$ relays in each layer. The eavesdropper overhears the transmission from the nodes in relay layer $M, 1 \le M \le L$. An instance of such a network is given in Figure~\ref{fig:layrdNetExa}. Using \eqref{eqn:snrt} and \eqref{eqn:snre}, the $SNR_t$ and $SNR_e$ in this case are: \begin{align*} SNR_t &= \frac{P_s}{\sigma^2} \frac{h_s^2 H_{1,M-1}^2 \left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_M^2 H_{M+1,L}^2}{\left[\left(\sum_{i=1}^{M-1} G_{i,M-1}^2\right)\left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_M^2 +\left(\sum_{n=1}^N \beta_{M,n}^2\right)h_M^2\right] H_{M+1,L}^2 +\sum_{i=M+1}^L G_{i,L}^2+1}\\ SNR_e &= \frac{P_s}{\sigma^2} \frac{h_s^2 H_{1,M-1}^2 \left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_e^2}{\left(\sum_{i=1}^{M-1} G_{i,M-1}^2\right)\left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_e^2+\left(\sum_{n=1}^N \beta_{M,n}^2\right)\,h_e^2+1} \end{align*} where \begin{align*} H_{i,j}^2 &= \prod_{k=i}^j \left(\sum_{n=1}^N \beta_{k,n}\right)^2 h_k^2\\ G_{i,j}^2 &= \left(\sum_{n=1}^N \beta_{i,n}^2\right) h_i^2 \prod_{k=i+1}^{j} \left(\sum_{n=1}^N \beta_{k,n}\right)^2 h_k^2 \end{align*} \begin{pavikl} \label{lemma:lyrdNetReducedBeta1} For the ECGAL layered networks, $\bm{\beta}_{opt}$ in \eqref{eqn:eqProb} is: \begin{equation*} \bm{\beta}_{opt} = (\bm{\beta}_{1,opt}, \ldots, \bm{\beta}_{M,opt}, \bm{\beta}_{M+1,max}, \ldots, \bm{\beta}_{L,max}) \end{equation*} \end{pavikl} \begin{IEEEproof} Please refer to Appendix~\ref{appndx:lemma4}. \end{IEEEproof} Introduce the following parameters \begin{align*} E &= h_s^2 H_{1,M-1}^2 \\ F &= \sum_{i=1}^{M-1} G_{i,M-1}^2 \\ \alpha &= \prod_{i=M+1}^{L}\left(\sum_{n=1}^N\sqrt{P_{i,n}}\right)^2 h_i^2 \\ \lambda &= \sum_{i=M+1}^L \lambda_{i} \left(\sum_{n=1}^N P_{i+1,n}\right) \prod_{j=i+2}^{L}\left(\sum_{n=1}^N\sqrt{P_{i,n}}\right)^2 h_j^2 \\ \mu &= \prod_{i=M+1}^{L}\left(\sum_{n=1}^N\sqrt{P_{i,n}}\right)^2 h_i^2 + \sum_{i=M+1}^L \mu_{i} \left(\sum_{n=1}^N P_{i+1,n}\right) \prod_{j=i+2}^{L}\left(\sum_{n=1}^N\sqrt{P_{i,n}}\right)^2 h_j^2\\ &= \alpha + \sum_{i=M+1}^L \mu_{i} \left(\sum_{n=1}^N P_{i+1,n}\right) \prod_{j=i+2}^{L}\left(\sum_{n=1}^N\sqrt{P_{i,n}}\right)^2 h_j^2 \\ \nu &= \sum_{i=M}^L \nu_{i} \left(\sum_{n=1}^N P_{i+1,n}\right) \prod_{j=i+2}^{L}\left(\sum_{n=1}^N\sqrt{P_{i,n}}\right)^2 h_j^2, \quad \nu_M = 1, \end{align*} where \begin{align*} \lambda_{M+1} &= P_s (s_{M+1} + \sigma^2 n_{M+1}), \quad s_{M+1} = 1, n_{M+1} = 0 &\\ \lambda_{i} &= P_s\left[s_{i-1} \left(\left(\sum_{n=1}^N\!\!\sqrt{P_{i-1,n}}\right)^2\!\! h_{i-1}^2\!\!+\!\sigma^2\right)\!\! +\! \sigma^2 n_{i-1} \left(\left(\sum_{n=1}^N P_{i-1,n}\right) h_{i-1}^2 \!\!+\!\sigma^2 \right)\right], & i \in \{M+2, \ldots, L\}\\ \mu_{M+1} &= \sigma^2 (s_{M+1} + \sigma^2 n_{M+1}), \quad s_{M+1} = 1, n_{M+1} = 0 &\\ \mu_{i} &= \sigma^2 \left[s_{i-1} \left(\left(\sum_{n=1}^N\!\!\sqrt{P_{i-1,n}}\right)^2\!\! h_{i-1}^2\!\!+\!\sigma^2\right)\!\! +\! \sigma^2 n_{i-1} \left(\left(\sum_{n=1}^N P_{i-1,n}\right) h_{i-1}^2 \!\!+\!\sigma^2 \right)\right], & i \in \{M+2, \ldots, L\}\\ \nu_{M+1} &= (s_{M+1} + \sigma^2 n_{M+1}), \quad s_{M+1} = 0, n_{M+1} = 1&\\ \nu_{i} &= \sigma^2 \left[s_{i-1} \left(\left(\sum_{n=1}^N\!\!\sqrt{P_{i-1,n}}\right)^2\!\! h_{i-1}^2\!\!+\!\sigma^2\right)\!\! +\! \sigma^2 n_{i-1} \left(\left(\sum_{n=1}^N P_{i-1,n}\right) h_{i-1}^2 \!\!+\!\sigma^2 \right)\right], & i \in \{M+2, \ldots, L\} \end{align*} Using the preceding lemma and the above parameters, the problem \begin{equation*} \bm{\beta}_{opt} = \operatornamewithlimits{argmax}_{\bm{\beta}} \frac{1+SNR_t}{1+SNR_e} \end{equation*} is reduced to the following subproblem \begin{equation} \label{eqn:secRate4lemma2} (\bm{\beta}_{1,opt}, \ldots, \bm{\beta}_{M,opt}) = \operatornamewithlimits{argmax}_{(\bm{\beta}_{1}, \ldots, \bm{\beta}_{M})} \frac{1+SNR_t|_{\bm{\beta}_{M+1:L,max}}}{1+SNR_e}, \end{equation} where for a given network-wide vector of scaling factors $(\bm{\beta}_1, \ldots, \bm{\beta}_M, \bm{\beta}_{M+1,max}, \ldots, \bm{\beta}_{L,max})$, the received SNRs at the destination and the eavesdropper are, respectively \begin{align*} SNR_t|_{\bm{\beta}_{M+1:L,max}} &= \frac{P_s}{\sigma^2} \frac{A \left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_M^2}{B \left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_M^2 + C\left(\sum_{n=1}^N \beta_{M,n}^2\right) h_M^2 + D}\\ SNR_e &= \frac{P_s}{\sigma^2} \frac{E \left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_e^2}{F \left(\sum_{n=1}^N \beta_{M,n}\right)^2 h_M^2 + \left(\sum_{n=1}^N \beta_{M,n}^2\right) h_M^2 + 1} \end{align*} with \begin{align*} A &= \alpha E\\ B &= \lambda E + \mu F\\ C &= \mu\\ D &= \nu \end{align*} \begin{pavikl} \label{lemma:lyrdNetReducedBeta2} For ECGAL layered networks, the subvector $(\beta_{M,1,opt}, \ldots, \beta_{M,N,opt})$ of the optimum scaling vectors for the nodes in the $M^{th}$ layers for given sub-vector $(\bm{\beta}_1, \ldots, \bm{\beta}_{M-1})$ is: \begin{equation*} \beta_{M,1,opt}^2 = \ldots = \beta_{M,N,opt}^2 = \beta_{M,opt}^2 = \begin{cases} \min(\beta_{M,max}^2, \beta_{M,glb}^2), \quad \mbox{if } (h_M^{2}\,\alpha-h_e^{2}\,\nu) > 0,\\ 0, \quad \mbox{otherwise} \end{cases} \end{equation*} where \begin{equation*} \beta_{M,glb}^2 = \frac{|{\mathcal B}|}{2\,|{\mathcal A}|}\left(\sqrt{1 + \frac{4\,|{\mathcal A}|\,{\mathcal C}}{{\mathcal B}^2}} - 1\right) \end{equation*} with \begin{align*} {\mathcal A} & = 4\,h_M^{2}\,h_e^{2}\bigg\{h_e^{2}\,\alpha\,\nu\,(2\,F+1)\,\left(2\,F+1+2\,\frac{P_s}{\sigma^2}\,E\right)\\ &\quad -h_M^{2}\,[2\,\lambda E +(2\,F+1)\mu]\,\left[(2\,F+1)\mu+2\,\left(\lambda +\alpha\,\frac{P_s}{\sigma^2}\right) E\,\right]\bigg\}\\ {\mathcal B} & = 4\,h_M^{2}\,h_e^{2}\,\nu\,[(\alpha-\mu)\,(2\,F+1)-2\,\lambda E]\\ {\mathcal C} & = \nu\,(h_M^{2}\,\alpha-h_e^{2}\,\nu) \end{align*} \end{pavikl} \begin{IEEEproof} Please refer to Appendix~\ref{appndx:lemma5}. \end{IEEEproof} \begin{pavikl} \label{lemma:lyrdNetReducedBeta3} For ECGAL layered networks, \begin{equation*} (\bm{\beta_{1,opt}}, \ldots, \bm{\beta_{M-1,opt}}) = (\bm{\beta_{1,max}}, \ldots, \bm{\beta_{M-1,max}}), \end{equation*} where \begin{align*} \beta_{1,n,max}^2 &= \frac{P_{1,n}}{P_{s} h_s^2 + \sigma^2}, n \in \{1, \ldots, N\}\\ \beta_{i,n,max}^2 &= \frac{P_{i,n}}{P_{R_x,i}}, \quad i \in \{2, \ldots, M-1\}, n \in \{1, \ldots, N\} \end{align*} with \begin{align*} P_{Rx,i} &= P_s h_s^2 H_{1,i-1}^2 +\left[\sum_{j=1}^{i-1} G_{j,i-1}^2+1\right] \sigma^2 \end{align*} \end{pavikl} \begin{IEEEproof} Please refer to Appendix~\ref{appndx:lemma6}. \end{IEEEproof} In short, Lemma~\ref{lemma:lyrdNetReducedBeta1}-\ref{lemma:lyrdNetReducedBeta3} together establish that for the \textit{ECGAL} layered networks with $L$ relays, the optimal vector of the scaling factors that maximizes the secure ANC rate is $\bm{\beta}_{opt} = (\bm{\beta}_{1,max}, \ldots, \bm{\beta}_{M,opt}, \bm{\beta}_{M+1,max}, \ldots, \bm{\beta}_{L,max})$, where $\beta_{M,opt}$ is given by Lemma~\ref{lemma:lyrdNetReducedBeta2}. \section{High-SNR Analysis of Achievable ANC Secrecy Rate in ECGAL Networks} \label{sec:highSNRanalysis} We define a wireless layered network to be in high SNR regime if \begin{equation*} \min_{i \in \{1, \ldots, L\}} SNR_i \ge \frac{1}{\delta} \end{equation*} for some small $\delta \ge 0$. Here, $SNR_i$ is the signal-to-noise ratio at the input of any of the relay nodes in the $i^{th}$ layer. Assume that each relay node in layer $i$ uses the amplification factor \begin{equation*} \beta_i^2 = \frac{P}{(1+\delta) P_{R_i,max}}, i \in \{1, \ldots, L\}, \end{equation*} where $P_{R_i,max}$ is the maximum received signal power at any relay node in the $i^{th}$ layer which in this case is equal to $N^2 P h_{i-1}^2$ as each relay node receives the transmissions from $N$ relay nodes in the previous layer with maximum transmit power constrained by $P$. It should be noted that $\beta_i$ is such that the maximum power constraint \eqref{eqn:pwrConstraint} is satisfied at each node as \begin{align*} \beta_{i,max}^2 = \frac{P}{P_{R_i}+P_{z_i}+\sigma^2} = \frac{P}{\left(1+\frac{1}{SNR_i}\right)P_{R_i}} \geq \frac{P}{(1+\delta)P_{R_i,max}} \quad [\mbox{Since }1/SNR_i \leq \delta, P_{R_i} \leq P_{R_i,max}] \end{align*} Here, $P_{R_i}$ and $P_{z_i}$ are the received signal and noise powers at the input of the node $i$, respectively. Note that as $\delta \rightarrow 0, \ \beta_i^2 \rightarrow \beta_{i,max}^2$. We now analyze the secrecy rate achievable with these scaling factors. However, before discussing the achievable secrecy rate, we discuss an upper bound on the secrecy capacity of such a network. For the source-destination(eavesdropper) path, an upper bound on the capacity is given by the capacity of the Gaussian multiple-access channel between the relays in the $L^{th}$ layer and the destination (the eavesdropper). This results in the following upper bound on the secrecy capacity of such networks: \begin{equation*} C_{cut} = \frac{1}{2}\log\left[\frac{1 + P_t/\sigma^2}{1 + P_e/\sigma^2}\right], \end{equation*} where $P_t = N^2 P_L h_t^2$ and $P_e = N^2 P_L h_e^2$. The power of the source signal reaching the destination $t$ is: \begin{align} P_{s,t} &= P_s h_s^2 \left(\prod_{i=1}^{L-1} N^2\beta_i^2 h_i^2\right) N^2 \beta_L^2 h_t^2 = \frac{N^2 P_L h_t^2}{(1+\delta)^L} \label{eqn:source_power_at_dstntn} \end{align} The total power of noise reaching the destination $t$ from all relay nodes: \begin{equation} P_{z,t} = \sum_{i=1}^L P_{z,t}^i = \sigma^2\left(\sum\limits_{i=1}^{L-1}N \beta_i^2 h_i^2 \left(\prod\limits_{j=i+1}^{L-1}(N \beta_j h_j)^2\right)N^2 \beta_L^2 h_t^2 + N \beta_L^2 h_t^2\right) \label{eqn:noise_power_at_dstntn} \end{equation} where $P_{z,t}^i$ is the noise power reaching the destination form nodes in $i^{th}$ layer. Now, \begin{align*} &&P_{z,t}^1 &= \sigma^2 N \beta_1^2 h_1^2 \left(\prod\limits_{j=2}^{L-1}N^2\beta_j^2 h_j^2\right) N^2 \beta_L^2 h_t^2 = \frac{\sigma^2}{P_s h_s^2} \frac{N P h_t^2}{(1+\delta)^L} \leq \frac{\delta}{(1+\delta)^L}N P h_t^2 &&\\ &&P_{z,t}^2 &= \sigma^2 N \beta_2^2 h_1^2 \left(\prod\limits_{j=3}^{L-1}N^2\beta_j^2 h_j^2\right) N^2 \beta_L^2 h_t^2 = \frac{\sigma^2}{N^2 P h_1^2} \frac{N P h_t^2}{(1+\delta)^{L-1}} \leq \frac{\delta}{(1+\delta)^{L-1}}N P h_t^2 &&\\ && \vdots & &&\\ &&P_{z,t}^i &= \sigma^2 N \beta_i^2 h_i^2 \left(\prod\limits_{j=i+1}^{L-1}N^2\beta_j^2 h_j^2\right) N^2 \beta_L^2 h_t^2 = \frac{\sigma^2}{N^2 P h_{i-1}^2} \frac{N P h_t^2}{(1+\delta)^{L-i+1}} \leq \frac{\delta}{(1+\delta)^{L-1}}N P h_t^2 &&\\ && \vdots & &&\\ &&P_{z,t}^L &= \sigma^2 N \beta_L^2 h_t^2 = \frac{\sigma^2}{N^2 P h_{L-1}^2} \frac{N P h_t^2}{(1+\delta)} \leq \frac{\delta}{(1+\delta)}N P h_t^2 && \end{align*} Therefore, \begin{align*} P_{z,t} &= \sum_{i=1}^L P_{z,t}^i \leq \sum_{i=1}^L \frac{\delta}{(1+\delta)^{L-i+1} NP h_t^2} \end{align*} or \begin{align} P_{z,t} & \leq NP h_t^2\left[1 - \frac{1}{(1+\delta)^L}\right] \label{eqn:uppr_bnd_noise} \end{align} The results in \eqref{eqn:source_power_at_dstntn} and \eqref{eqn:noise_power_at_dstntn} imply that we have the following for the achievable rate at the destination \begin{align*} R_t &= \frac{1}{2} \log\left[1 + \frac{1}{(1+\delta)^L} \frac{N^2 P_L h_t^2}{P_{z,t} + \sigma^2} \right] \end{align*} Similarly, the source and noise power reaching the eavesdropper $e$ respectively are: \begin{align*} P_{s,e} &= P_s h_s^2 \left(\prod_{i=1}^{L-1} N^2\beta_i^2 h_i^2\right) N^2 \beta_L^2 h_e^2= \frac{N^2 P_L h_e^2}{(1+\delta)^L}\\ P_{z,e} &= \sum_{i=1}^L P_{z,t}^i = \sum\limits_{i=1}^{L-1}N \beta_i^2 h_i^2 \left(\prod\limits_{j=i+1}^{L-1}(N \beta_j h_j)^2\right)N^2 \beta_L^2 h_e^2 + N \beta_L^2 h_e^2 = P_{z,t}\ h_e^2/h_t^2 \end{align*} and the achievable rate at the eavesdropper is: \begin{align*} R_e &= \frac{1}{2} \log\left[1 + \frac{1}{(1+\delta)^L} \frac{N^2 P_L h_e^2}{P_{z,t} h_e^2/h_t^2 + \sigma^2} \right] \end{align*} Therefore, the achievable secrecy rate is \begin{equation} \label{eqn:hghSNRsecRate} R_s = R_t - R_e = \frac{1}{2}\log\left[\frac{1 + \frac{1}{(1+\delta)^L} \frac{N^2 P_L h_t^2}{P_{z,t} + \sigma^2}}{1 + \frac{1}{(1+\delta)^L} \frac{N^2 P_L h_e^2}{P_{z,t} h_e^2/h_t^2 + \sigma^2}}\right] \end{equation} Thus, we have the following for the gap between the cutset upper-bound and the achievable secrecy rate \begin{align} C_{cut} - R_s &= \frac{1}{2}\log\left[\frac{1 + P_t/\sigma^2}{1 + P_e/\sigma^2}\right] - \frac{1}{2}\log\left[\frac{1 + \frac{1}{(1+\delta)^L} \frac{N^2 P_L h_t^2}{P_{z,t} + \sigma^2}}{1 + \frac{1}{(1+\delta)^L} \frac{N^2 P_L h_e^2}{P_{z,t} h_e^2/h_t^2 + \sigma^2}}\right] \label{eqn:actual_gap} \end{align} Here, R.H.S. is an increasing function of $P_{z,t}$. Thus, from \eqref{eqn:uppr_bnd_noise} and \eqref{eqn:actual_gap}, we have \begin{align} C_{cut} - R_s &\leq \frac{1}{2}\log\left[\frac{\left(1+\left(1-\frac{1}{(1+\delta)^L}\right) \frac{NP h_t^2}{\sigma^2}\right) \left(1+\frac{N^2 P h_t^2}{\sigma^2}\right) \left(1+\left(1-\frac{1}{(1+\delta)^L}\right)\frac{N P h_e^2}{\sigma^2} + \frac{N^2 P h_e^2}{(1+\delta)^L \sigma^2}\right)}{\left(1+\left(1-\frac{1}{(1+\delta)^L}\right) \frac{NP h_e^2}{\sigma^2}\right) \left(1+\frac{N^2 P h_e^2}{\sigma^2}\right) \left(1+\left(1-\frac{1}{(1+\delta)^L}\right)\frac{N P h_t^2}{\sigma^2} + \frac{N^2 P h_t^2}{(1+\delta)^L \sigma^2}\right)}\right] \nonumber\\ & \leq \frac{1}{2}\log\left[\!\frac{\left(1+L \delta\frac{N P h_t^2}{\sigma^2}\right)\left(1+\frac{N^2 P h_t^2}{\sigma^2}\right)\left(1+ L \delta\frac{N P h_e^2}{\sigma^2} + (1-L\delta) \frac{N^2 P h_e^2}{\sigma^2}\right)}{\left(1+L \delta\frac{N P h_e^2}{\sigma^2}\right)\left(1+\frac{N^2 P h_e^2}{\sigma^2}\right)\left(1+ L \delta\frac{N P h_t^2}{\sigma^2} + (1-L\delta) \frac{N^2 P h_t^2}{\sigma^2}\right)}\right] \quad\left[\mbox{as }{1}/{(1+\delta)^L} \ge 1-L \delta\right]\nonumber\\ &= \frac{1}{2}\log\left[\left.\frac{1+{N^2 P h_t^2}/{\sigma^2}}{1+\frac{(1-L\delta) NPh_t^2/\sigma^2}{1+L\delta P h_t^2/\sigma^2}}\middle/ \right. \frac{1+{N^2 P h_e^2}/{\sigma^2}}{1+\frac{(1-L\delta) NPh_e^2/\sigma^2}{1+L\delta P h_e^2/\sigma^2}}\right]\nonumber\\ &\leq \frac{1}{2}\log\left[\left(\left. \frac{1+L\delta NP h_t^2/\sigma^2}{(1-L\delta)}\right)\middle/ \right.\left(1+L\delta \frac{NPh_e^2}{\sigma^2}\right)\right]\nonumber\\ &=\frac{1}{2}\log\left[\frac{1}{(1-L\delta)} \frac{1+L\delta NP h_t^2/\sigma^2}{1+L\delta NP h_e^2/\sigma^2}\right] \label{eqn:gap_bnd} \end{align} Since, $R_{s,opt} \geq R_s, \ C_{cut}-R_{s,opt} \leq C_{cut}-R_{s}$. Note that as $\delta \rightarrow 0, \ C_{cut}-R_s \rightarrow 0$, i.e., secrecy rate approaches the cut-set bound. \section{Numerical Simulations} \label{sec:numSim} In this section, we present numerical results to evaluate the performance of the proposed high SNR approximation scheme. We consider a 2-layer network with two nodes in each layer and eavesdropper snooping on the transmissions of the nodes in the last layer. In Figure~\ref{fig:2lyrdNetSecRateHighSNRplot}, we plot the achievable secrecy rate when all the nodes transmit at their maximum power along with the corresponding cut-set as a function of source power for the specified system parameters. \begin{figure}[!t] \begin{subfigure}[b]{0.5\textwidth} \includegraphics[width=\textwidth]{2lyrHighSNRfinalPlot1} \caption{} \label{fig:2lyrdNetSecRateHighSNRplot1} \end{subfigure} \begin{subfigure}[b]{0.5\textwidth} \includegraphics[width=\textwidth]{2lyrHighSNRfinalPlot3} \caption{} \label{fig:2lyrdNetSecRateHighSNRplot2} \end{subfigure} \caption{Plot of achievable secrecy rate with all relays transmitting at maximum power against varying source power along with the cutset bound for $2$-layer network with $2$ nodes in each layer. $P=500,\ \sigma^2=1.0, \ \delta=0.005,\ $ (a)~$ h_s=0.689, \ h_1=0.603, \ h_t=0.203, \ h_e= 0.031,$ and (b)~$ h_s=0.260, \ h_1 = 0.925,\ h_t=0.113, \ h_e=0.012.$} \label{fig:2lyrdNetSecRateHighSNRplot} \end{figure} From the figure, it can be observed that at high SNR the achievable rate lies within a constant gap from the cut-set bound. In Figure~\ref{fig:2lyrdNetSecRateHighSNRplot1}, the actual gap between the cut-set bound and achievable rate is $0.05$ bits/sec/Hz while the upper bound on this gap as given by \eqref{eqn:gap_bnd} is $0.25$ bits/sec/Hz. Similarly, in Figure~\ref{fig:2lyrdNetSecRateHighSNRplot2}, the actual gap is $0.03$ bits/sec/Hz while the upper bound on this gap is $0.09$ bits/sec/Hz. Thus, it can be seen that \eqref{eqn:gap_bnd} tightly approximates the gap between achievable secrecy rate and the cut-set bound. \section{Conclusion and Future Work} \label{sec:cnclsn} We consider the problem of secure ANC rate maximization over a class of Gaussian layered networks where a source communicates with a destination through $L$ intermediate relay layers with $N$ nodes in each layer in the presence of a single eavesdropper which can overhear the transmissions of the nodes in any one layer. The key contribution of is the computation of the globally optimal set of scaling factors for the nodes that maximizes the end-to-end secrecy rate for a class of layered networks. We also show that in the high-SNR regime, ANC achieves secrecy rates within a constant gap of the cutset upper bound on the secrecy capacity and numerically validate this. In future, we plan to extend this work for more general layered networks.
45dc3a242155cda7490ef1e2ab7c9d1e213c6b7c
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The process of particle creation from quantum vacuum because of moving boundaries or time-dependent properties of materials, commonly referred as the dynamical Casimir effect (DCE)\cite{1,2}, has been investigated since the pioneering works of Moore in 1970 \cite{moor}, who showed that photons would be created in a Fabry-Perot cavity if one of the ends of the cavity walls moved periodically, \cite{rev,rev1}. The dynamical Casimir effect is frequently used nowadays for phenomena connected with the photon creation from vacuum due to fast changes of the geometry or material properties of the medium. Moving bodies experience quantum friction \cite{Ramin} and so energy damping \cite{en,en1} and decoherence \cite{dec} due to the scattering of vacuum field fluctuations. The damping is accompanied by the emission of photons \cite{moor}, thus conserving the total energy of the combined system \cite{con}. An explicit connection between quantum fluctuations and the motion of boundaries was made in \cite{v}, where the name non-stationary Casimir effect was introduced, and in \cite{mir,mir1}, where the names Mirror Induced Radiation and Motion-Induced Radiation (with the same abbreviation MIR) were proposed. The frequency of created Photons in a mechanically moving boundary are bounded by the mechanical frequency of the moving body and to observe a detectable number of created photons the oscillatory frequency must be of the order of GHz which arise technical problems. Therefore, recent experimental schemes focus on simulating moving boundaries by considering material bodies with time-dependent electromagnetic properties \cite{sim, sim1}. In this scheme, for example for two semi-infinite dielectrics, the boundary is not moving mechanically but its moving is simulated or modelled by changing the electromagnetic properties of one of the dielectrics in a small slab periodically. An important factor in detecting the created photons is keeping the sample at a low temperature of $\sim$ 100 mK to suppress the number of thermal black body photons to less than unity. Particularly, the problem has been considered with mirrors (single mirror and cavities), where the input field reflected completely from the surface. Recently the Robin boundary condition (RBC) has been used as a helpful approach to consider the dynamical boundary condition for this kind of problem. The well known Drichlet and Neuwmann boundary conditions can be obtained as the limiting cases of Robin boundary condition \cite{hector,mintz}. The aim of the present work is to use a perturbative approach to study the effect of transition trough the interface on the spectral distribution of created photons. The interface between two semi-infinite dielectrics is modelled to simulate the oscillatory motion of the moving boundary. For this purpose, the electromagnetic field quantization in the presence of a dielectric medium \cite{matloob,matloob1} is reviewed briefly then a general approach to investigate the dynamical Casimir effect for simulated motion of some part of a dielectric medium is introduced and finally, the spectral distribution of created photons are derived and the effect of small transitions trough the interface has been discussed . \section{ The electromagnetic field quantization in absorbing dielectrics} In this section we review briefly the electromagnetic field quantization in the presence of two adjacent semi-infinite absorbing media with different homogeneous and isotropic dielectric functions \cite{matloob1}. Therefore, The dielectric function is defined by \begin{equation}\label{e} \varepsilon(x,\omega)=\left\{ \begin{array}{ll} \varepsilon_1(\omega)=n_1^2(\omega)=[\eta_1(\omega)+i\kappa_1(\omega)]^2, & x<0 \\ \varepsilon_2(\omega)=n_2^2(\omega)=[\eta_2(\omega)+i\kappa_2(\omega)]^2, & x>0 \end{array} \right. \end{equation} where the subscript indices 1 and 2 correspond to the regions $ x>0$ and $x<0$, respectively. The inhomogeneous nature of the problem requires the imposition of boundary conditions on the spatial mode functions on the interface. The vector potential in frequency space satisfies the familiar equation \cite{matloob1} \begin{equation}\label{A} (\frac{\partial^2}{{\partial x}^2}+\varepsilon(x,\omega)\frac{\omega ^2}{c^2})\,A(x,\omega)=-\frac{1}{\varepsilon_0 c^2 S}\,J(x,\omega). \end{equation} We can decompose any field to its positive and negative frequency parts then the positive frequency part of the vector potential is given by \begin{equation}\label{a} \hat{A}^+(x,\omega)=S\int_{-\infty}^\infty dx' G(x,x',\omega)\hat{J}^+(x',\omega), \end{equation} where $S$ is the interface area and the Green's function fulfills the equation \begin{equation}\label{g} (\frac{\partial^2}{{\partial x}^2}+\varepsilon(x,\omega)\frac{\omega ^2}{c^2})G(x,x',\omega)=-\frac{1}{\varepsilon_0 c^2 S}\delta(x-x'). \end{equation} The Green's function is obtained explicitly as \begin{figure} \includegraphics[scale=0.7]{ex2.pdf}\\ \caption{Representation of the notation for the annihilation operators used in the definition of the vector potential operator for two adjacent dielectrics. }\label{ex1} \end{figure} \begin{align}\label{g1} G(x,x',\omega)=\frac{i}{2\varepsilon_0c\omega n_1(\omega)S}\{R_L(\omega) \exp(\frac{i\omega n_1(\omega)(x+x')}{c})\qquad\qquad\qquad \\ \nonumber +\exp(\frac{i\omega n_1(\omega)|x-x'|}{c})\}, \qquad\qquad\qquad\qquad\qquad\qquad x>0,\,\,x'>0 \\ \nonumber=\frac{i}{2\varepsilon_0c\omega n_2(\omega)S}T_R(\omega)\exp(\frac{i\omega[n_1(\omega)x-n_2(\omega)x']}{c}), \qquad x>0,\,\,x'<0 \end{align} \begin{align}\label{g2} G(x,x',\omega)=\frac{i}{2\varepsilon_0c\omega n_2(\omega)S}\{R_R(\omega) \exp(-\frac{i\omega n2(\omega)(x+x')}{c})\qquad\qquad\quad\\ \nonumber+\exp(\frac{i\omega n_2(\omega)|x+x'|}{c})\}, \qquad\qquad\qquad\qquad\qquad\qquad x<0,\,\,x'<0 \\ \nonumber =\frac{i}{2\varepsilon_0c\omega n_1(\omega)S}T_L(\omega)\exp(-\frac{i\omega[n_2(\omega)x-n_1(\omega)x']}{c}), \quad x<0,\,\,x'>0 \end{align} where $T(\omega)$ and $R(\omega)$ are the usual transmission and reflection coefficients respectively, and the subscript indices $R$ and $L$ refer to the light incident on the interface from the right or left. These coefficients are given by \begin{equation}\label{R} R_L(\omega)=-R_R(\omega)=\frac{n_1(\omega)-n_2(\omega)}{n_1(\omega)+n_2(\omega)} \end{equation} \begin{equation}\label{T} \frac{T_L(\omega)}{n_1(\omega)}=\frac{T_R(\omega)}{n_2(\omega)}=\frac{2}{n_1(\omega)+n_2(\omega)} \end{equation} One can show that the vector potential in space-time can be written as \cite{matloob1} \begin{align}\label{A1} \hat{A}^+(x,t)=\int_0^{+\infty}d\omega(\frac{\hbar\eta(\omega)}{4\pi\varepsilon_0c\omega\varepsilon(\omega)S})^\frac{1}{2} \{\hat{c}_R(x,\omega)+\hat{c}_L(x,\omega)\}e^{-i\omega t}. \end{align} The complete expressions for the operators $\hat{c}_{1R}(x,\omega)$ and $\hat{c}_{1L}(x,\omega)$ in the positive $x$ domain are determined using (\ref{a}), (\ref{A1}) and insertion of (\ref{g1}), (\ref{g2}) into (\ref{a}) as \begin{align}\label{aaa1} \hat{c}_{1L}(x,\omega)=i(\frac{2\omega\kappa_1(\omega)}{c})^{1/2} \int_x^{+\infty}dx' \exp(-\frac{i\omega n_1(\omega)}{c}(x-x')) \hat{f}(x',\omega) \end{align} \begin{eqnarray}\label{aaa2} && \hat{c}_{1R}(x,\omega)=\exp(\frac{i\omega n_1(\omega)x}{c})\times\nonumber\\ && \bigg\{ i(\frac{2\eta_2(\omega)\omega\kappa_1(\omega)}{\eta_1(\omega)c})^{1/2}\frac{n_1(\omega)}{n_2(\omega)}T_R(\omega) \int_{-\infty}^0 dx'\, \exp(-\frac{i\omega n_2(\omega)x'}{c}) \hat{f}(x',\omega)\nonumber\\ && +i(\frac{2\omega\kappa(\omega)}{c})^{1/2}\bigg[R_L(\omega)\int_0^{+\infty}dx'\,\exp(\frac{i\omega n_1(\omega)x'}{c})\hat{f}(x',\omega)\nonumber\\ && +\int_0^xdx'\exp(-\frac{i\omega n_1(\omega)x'}{c})\hat{f}(x',\omega)\bigg]\bigg\} \end{eqnarray} where $\hat{f}(x,\omega)=\hat{J}^+(x,\omega)\sqrt{S/2\varepsilon_0\hbar\omega^2\varepsilon_i(\omega)}$. The expressions for $\hat{c}_{1L}(x,\omega)$ and $\hat{c}_{1R}(x,\omega)$ contain $e^{(\frac{-i\omega n_1(\omega)x}{c})}$ and $e^{(\frac{i\omega n_1(\omega)x}{c})}$ respectively, which shows the direction of propagation of the field operators. This property can be used to determine the terms in (\ref{a}) which correspond to $\hat{c}_{1L}(x,\omega)$ and $ \hat{c}_{1R}(x,\omega)$ easily. \section{Simulating the moving boundary} Motivated by experiments in which moving boundaries are simulated by time dependent properties of static systems including, changing the effective inductance of the SQUID by a time-dependent magnetic flux \cite{ex1,ex2} or MIR experiment \cite{ex3,ex} and also \cite{ex4},\cite{ex4},\cite{ex5},\cite{ex6}, we discuss here a model to change the dielectric function of a slab dielectric with thickness $\delta q$ which is placed at the interface of semi infinite absorbing dielectrics and its dielectric function oscillates between $\varepsilon_1(\omega)$ and $\varepsilon_2(\omega)$ with the frequencies $\omega_0$. This consumption equals to the oscillation of boundary with the mechanical frequency $\omega_0$. (see figure\ref{ex}) \begin{figure} \includegraphics[scale=0.5]{ex.pdf}\\ \caption{In the left figure the dielectric function of slab equals to $\varepsilon_1(\omega)$ while in the right one it equals to $\varepsilon_2(\omega)$ easily this oscillation causes moving boundary. }\label{ex} \end{figure} \par To solve the problem through a perturbative approach, we consider the dielectric function as: \begin{equation}\label{e2} \varepsilon(x,t,t')=\varepsilon(x,t-t')+F(x,t,t') \end{equation} Where \begin{equation}\label{f} \lim_{\delta q\rightarrow 0}F(x,t,t')=0 \end{equation} $F(x,t,t')$ simulates the motion of boundary and is taken into account as the perturbation term, which is given by \begin{align}\label{f1} \nonumber F(x,t,t')=\{\varepsilon_1(t-t')-\varepsilon_2(t-t')\}f(x,t') \end{align} \begin{equation}\label{e1} f(x,t') = \left\{ \begin{array}{cc} 0 \quad\quad\quad\quad\quad\quad\quad\quad\quad x<0 \\ \sin^2(\frac{\omega_0t'}{2})e^{\frac{-x}{\delta q}}\quad\quad\quad\quad x>0 \end{array} \right. \end{equation} where $e^{\frac{-x}{\delta q}}$ limits the thickness of slab to $\delta q$. \par We start from inhomogeneous Helmholtz differential equation \begin{equation}\label{a2} \frac{\partial^2\hat{A}(x,t)}{{\partial x}^2}-\frac{1}{c^2}\frac{\partial}{\partial t}\int dt' \varepsilon (t-t') \frac{\partial}{\partial t'}\hat{A}(x,t')=\frac{1}{\varepsilon_0c^2}\hat{J}(x,t) \end{equation} where the transverse operator $\hat{J}(x,t)$ plays the role of a Langevin force associated with the noise reservoir. The field operators are separated into positive and negative frequency components in usual way, \begin{equation}\label{a+} \hat{A}(x,t)=\hat{A}^+(x,t)+\hat{A}^-(x,t) \end{equation} and the frequency space Fourier transform operators are defined according to \begin{equation}\label{a+w} \hat{A}^+(x,t)=\frac{1}{\sqrt{2\pi}}\int_0^\infty d\omega \hat{A}^+(x,\omega)e^{-i\omega t} \end{equation} With similar separations and transforms for noise current operators. The negative frequency component are provided bye hermitian conjugates of the positive frequency operators. \par We consider the effect of motion as a small perturbation \begin{equation}\label{del a} \hat{A}(x,t)=\hat{A}_0(x,t)+\delta\hat{A}(x,t) \end{equation} where the unperturbed field $\hat{A}_0(x,t)$ corresponds to a solution with a static boundary at $x=0$. The first order field $\delta\hat{A}(x,t)$ then satisfies the following equation \begin{align}\label{del a1} \frac{\partial^2\delta\hat{A}(x,t)}{{\partial x}^2}-\frac{1}{c^2}\frac{\partial}{\partial t}\int dt' \varepsilon (x,t-t') \frac{\partial}{\partial t'}\delta\hat{A}(x,t')\\ \nonumber =\frac{1}{c^2}\frac{\partial}{\partial t}\int dt' F(x,t,t') \frac{\partial}{\partial t'}\hat{A}_0(x,t') \end{align} After transforming the above equation to Fourier space and (\ref{e1}), we find \begin{align}\label{del a2} \nonumber \frac{\partial^2\delta\hat{A}(x,\omega)}{{\partial x}^2}+\frac{\omega^2}{c^2}\varepsilon(x,\omega)\delta\hat{A}(x,\omega) =-\int_{-\infty}^\infty d\omega' \omega \\ \times(\omega-\omega')f(x,\omega')\{\varepsilon_1(\omega)-\varepsilon_2(\omega)\}\hat{A}_0(x,\omega-\omega') \end{align} where $f(x,\omega)$ is the Fourier transform of $f(x,t)$ \par To solve (\ref{del a2}) for $\delta\hat{A}(x,\omega)$ in terms of $\hat{A}_0(x,\omega)$, we consider the right hand side of that as a source and use (\ref{g1}) and (\ref{g2}), then \begin{align}\label{del a3} \nonumber \delta \hat{A}(x,\omega)=-\int_{-\infty}^\infty dx'' G(x,x'',\omega)\int_{-\infty}^\infty d\omega'\omega\qquad \\ \times(\omega-\omega')f(x'',\omega')\{\varepsilon_1(\omega)-\varepsilon_2(\omega)\}\hat{A}_0(x'',\omega-\omega') \end{align} from (\ref{del a3}) and (\ref{e1}) we find \begin{align}\label{del a4} \nonumber \delta\hat{A}(x,\omega)=-\int_0^{\delta q}dx''G(x,x'',\omega)[\varepsilon_1(\omega)-\varepsilon_2(\omega)]\\ \nonumber \times \{\omega^2 \hat{A}_0(x'',\omega)\quad\qquad\qquad\\ \nonumber -\frac{1}{2}\omega(\omega-\omega_0)\hat{A_0}(\omega-\omega_0)\quad\\ -\frac{1}{2}\omega(\omega+\omega_0)\hat{A_0}(\omega+\omega_0)\} \end{align} \par From (\ref{del a}) $\delta\hat{A}(x,\omega)$ is the first order of field correction and we can separate that for negative and positive frequencies. \par If in (\ref{del a4}) we consider $\omega>0$ or positive frequencies, which correspond to annihilation operators, the final field $\hat{A}^+(x,\omega)$ contains the negative frequencies, because of $\hat{A_0}(\omega-\omega_0)$ term which contains the creation operators for $0<\omega<\omega_0$ (negative frequencies) and we easily can show, the vacuum state for static field $\hat{A}_0(x,\omega)$ is not a vacuum state with respect to dynamical field $\hat{A}(x,\omega)$ with moving boundary condition. In the other word particles are created here by frequency $\omega$ which is less than the mechanical frequency $\omega_0$. \begin{align}\label{del a4} \delta\hat{A}_1(x,\omega)=-\int_0^{\delta q} dx''\frac{i(\varepsilon_1(\omega)-\varepsilon_2(\omega))}{2\varepsilon_0c\omega n_1(\omega)S} \{R_L \exp{(\frac{i\omega n_1(\omega)(x+x'')}{c})} \\ \nonumber +\exp{(\frac{i\omega n_1(\omega)(x-x'')}{c})}\} \{\omega^2 \hat{A}_0(x'',\omega)-\frac{1}{2}\omega(\omega-\omega_0) \hat{A}_0(x'',\omega-\omega_0)\\ \nonumber -\frac{1}{2}\omega(\omega+\omega_0)\hat{A}_0(x'',\omega+\omega_0)\} \end{align} Where we have $\hat{A}_0$ from equation (\ref{A1}). We calculate the perturbation of the field for positive $x$ domain. Further physical insight is gained if we drive the perturbation term of creation and annihilation operators. Obviously in (\ref{del a4}) $\delta\hat{A}_1(x,\omega)$ just contain a perturbation on the rightward operator $\hat{c}_{1R}$, because of $e^{(\frac{i\omega n_1(\omega)x}{c})}$ term, which shows the right ward propagation. We expected this kind of operators correction. \begin{align}\label{c1} \delta \hat{c}_{1R}(x,\omega)=-\int_0^{\delta q} dx''\frac{i(\varepsilon_1(\omega)-\varepsilon_2(\omega))}{2\varepsilon_0c\omega n_1(\omega)S} \{R_L \exp{(\frac{i\omega n_1(\omega)(x+x'')}{c})} \\ \nonumber +\exp{(\frac{i\omega n_1(\omega)(x-x'')}{c})}\} \times \{\omega^2[\hat{c_0}_{1R}(x'',\omega)+\hat{c_0}_{1L}(x'',\omega)]\\ \nonumber -\frac{1}{2}(\omega-\omega_0)\omega\Theta(\omega-\omega_0) [\hat{c_0}_{1R}(x'',\omega-\omega_0)+\hat{c_0}_{1L}(x'',\omega-\omega_0)] \\ \nonumber -\frac{1}{2}(\omega_0-\omega)\omega\Theta(\omega_0-\omega) [\hat{c_0}_{1R}^\dag(x'',\omega_0-\omega)+\hat{c_0}_{1L}^\dag(x'',\omega_0-\omega)]\\ \nonumber -\frac{1}{2}(\omega+\omega_0)\omega [\hat{c_0}_{1R}(x'',\omega+\omega_0)+\hat{c_0}_{1L}(x'',\omega+\omega_0)]\} \end{align} Where $\hat{c_0}_{1R}$ and $\hat{c_0}_{1L}$ are the unperturbed operators which are calculated in\cite{matloob,matloob1}. Easily we can drive $\delta \hat{c}_{1R}^\dag(x,\omega)$ by complex conjugating (\ref{c1}) or by using (\ref{del a4}) and the negative frequency domain. Both give us the same result. \begin{equation}\label{&c} \hat{c}=\hat{c_0}+\delta\hat{c} \end{equation} \par Now we consider the lossless dielectrics where $\kappa \rightarrow 0$ . In this case the commutator of the operators $\hat{c_0}_{1R}(x,\omega)$ and $\hat{c_0}_{1R}^\dag(x,\omega)$ is obtained in \cite{matloob1} \begin{equation}\label{cc+} [\hat{c_0}_{1R}(x,\omega),\hat{c_0}_{1R}^\dag(x',\omega')]=\delta(\omega-\omega')\exp(\frac{i\omega n_1(\omega)(x-x')}{c}) \end{equation} The commutation relations between the leftwards and rightwards annihilation and creation operators are also \begin{align}\label{cc+1} \nonumber [\hat{c_0}_{1R}(x,\omega),\hat{c_0}_{1L}^\dag(x',\omega')]=[\hat{c_0}_{1L}(x,\omega),\hat{c_0}_{1R}^\dag(x',\omega')]^*\qquad \qquad\qquad \\ = \delta(\omega-\omega')R_L(\omega)\exp(\frac{i\omega n_1(\omega)(x+x')}{c}) \end{align} For $x\geq0$ domain, we can consider only rightwards operators, because the leftwards operators are leaved unchanged by the perturbation. \begin{equation}\label{c1l} \hat{c}_{1L}=\hat{c_0}_{1L} \end{equation} With \begin{equation}\label{n0} <0_0|\hat{c_0}_{1R}^\dag(\omega)\hat{c_0}_{1R}(\omega)|0_0>=<0_0|\hat{c}_{1L}^\dag(\omega)\hat{c}_{1L}(\omega)|0_0>=0 \end{equation} \par Since the rightwards annihilation operator is contaminated by leftwards and rightwards creation operators, the static vacuum state $|0_0>$ is not a vacuum state with respect to the dynamic operators. \section{Frequency spectrum} The number of particles created with frequencies between $\omega$ and $\omega+d\omega$ $(\omega\geq 0)$ is \begin{figure} \includegraphics[scale=1.2]{large.pdf}\\ \caption{Spectral distribution of the emitted particles $\frac{dN}{d\omega}(\frac{\delta q\omega_0}{c}\backsimeq 0.1\pi)$.Dashed line : Spectral distribution for $R_L=-0.988$.Dotted line : Spectral distribution for $R_L=-0.980$. Solid line : Spectral distribution for $R_L=-0.967 $.}\label{nem} \end{figure} \begin{equation}\label{n} \frac{dN}{d\omega}(\omega)d\omega=<0_0|\hat{c}_{1R}^\dag(x,\omega)\hat{c}_{1R}(x,\omega)|0_0>\frac{d\omega}{2\pi} \end{equation} The spectrum is obtained by inserting (\ref{c1}) , (\ref{cc+}) and (\ref{cc+1}) into (\ref{n}) \begin{align}\label{n1} \frac{dN}{d\omega}(\omega)=\int_0^{\delta q}dx'\int_0^{\delta q}dx'' \frac{(n_1(\omega)-n_2(\omega))^2}{2\pi(4\varepsilon_0c n_1(\omega)S)^2}(\omega-\omega_0)^2\Theta({\omega_0-\omega})\\ \nonumber\{R_L^2(\omega)\exp(\frac{-i\omega n_1(\omega)(x''-x')}{c})+2R_L(\omega)\cos[\frac{\omega n_1(\omega)}{c}(x''+x')]\\ \nonumber+\exp(\frac{i\omega n_1(\omega)(x''-x')}{c})\}\times2\{\cos[\frac{\omega n_1}{c}(x''-x')]\qquad\qquad\quad\\ \nonumber+R_L(\omega)\cos[\frac{\omega n_1}{c}(x''+x')]\}\qquad\qquad\qquad\qquad\qquad\qquad\qquad \end{align} We define dimensionless parameter $a=\frac{x'\omega_0}{c}$ and $a'=\frac{x''\omega_0}{c}$ and also $y=\frac{\omega}{\omega_0}$ which is always smaller than unity and rewrite (\ref{n1}) again. \begin{align}\label{n11} \frac{dN}{dy}(y)=\int_0^{\frac{\delta q\omega_0}{c}}da\int_0^{\frac{\delta q\omega_0}{c}}da' \frac{(n_1(y)-n_2(y))^2}{2\pi(4\varepsilon_0 n_1(y)S)^2}(y-1)^2\Theta({1-y})\\ \nonumber\{R_L^2(y)\exp(-iy n_1(y)(a'-a))+2R_L(y)\cos[y n_1(y)(a'+a)]\\ \nonumber+\exp(iy n_1(y)(a'-a))\}\times2\{\cos[y n_1(y)(a'-a)]\qquad\qquad\quad\\ \nonumber+R_L(y)\cos[y n_1(y)(a'+a)]\}\qquad\qquad\qquad\qquad\qquad\qquad\qquad \end{align} \begin{figure} \includegraphics[scale=1.2]{small.pdf}\\ \caption{Spectral distribution of the emitted particles $\frac{dN}{d\omega}(\frac{\delta q\omega_0}{c}\backsimeq 0.001\pi)$.Dashed line : Spectral distribution for $R_L=-0.99998$.Dotted line : Spectral distribution for $R_L=-0.99997$. Solid line : Spectral distribution for $R_L=-0.99994 $.}\label{nem1} \end{figure} \par Now we are going to plot the spectrum as a function of $ y $. In this paper we work with the non relativistic approximation and as the previous work on the simulation of motion of the bound \cite{ex1}, the mechanical speed of bound can be considered about $\%10$ of the speed of light. In this limit $\frac{\delta q\omega_0}{c}\backsimeq 0.1\pi$, so it is not small enough to expand (\ref{n11}) in the first order of $\frac{\delta q\omega_0}{c}$. Figure (\ref{nem}) shows the spectrum in this case. This spectrum doesn't contain symmetry around $y=\frac{1}{2}(\omega=\frac{\omega_0}{2})$ and doesn't vanish too fast with respect to $R_L$ less than unity. So we have valuable content for spectrum even in case of about $\%4$ transition of the incidental fields. Another meaningful choice for the mechanical speed of bound would be about $\%0.1 $ of the speed of light where in this case $\frac{\delta q\omega_0}{c}\backsimeq 0.001\pi$ and so it would be small enough to expand (\ref{n11}) and we find \begin{align}\label{n2} \frac{dN}{dy}(y)= \frac{(n_2(y)-n_1(y))^2}{2\pi(4\varepsilon_0S)^2}(y-1)^2\Theta({1-y})\{(1-R_L)^3-2R_L^2(\frac{\delta q\omega_0 y}{c})^2\} \end{align} If we consider $ R_L\rightarrow -1 $, which represent the case of complete reflection of the leftward field from the bound, we find \begin{equation}\label{n3} \frac{dN}{dy}(y)=-2 \frac{(n_2(y)-n_1(y))^2(\delta q\omega_0)^2}{2\pi(4\varepsilon_0 cS)^2}(y-1)^2y^2\Theta({1-y}) \end{equation} Figure (\ref{nem1}) shows the spectrum with these considerations. As we see from figure (\ref{nem}) and (\ref{nem1}) the spectrum vanishes for $ y\leq 1 $ or in the other word $\omega \leq \omega_0$ and so no particle is created with frequency greater than the mechanical frequency of the bound. But here the spectrum (\ref{n3}) is the symmetry around $y=1/2$ where the spectrum has a peak over there (figure (\ref{nem1})), and in this case the spectrum is valuable just for $R_L$ too close to unity or in case of complete reflection. \section{Conclusion} \par As a result of the figure (\ref{nem}) and (\ref{nem1}) , spectrum decrease rapidly by the decrease in value of $ R_L $ and actually for a small variation from $-1$, it vanishes. But in figure (\ref{nem}) the decrease in spectrum with respect to $R_L$ is less than the figure (\ref{nem1}). So we would have valuable content of spectrum, even in case of a little transition of the incidental fields. But at all, if we are going to detect the created particles, we would increase our chance by considering one of the medium as a conductor. \par In the case $ R_L\rightarrow -1 $ and $ \delta q\rightarrow 0 $ the spectrum was the same as the spectrum of dynamical casimir effect which has been studied by a variety of methods \cite{ex1},\cite{ex3},\cite{ex5},\cite{ex6},\cite{mintz} such as Robin boundary condition\cite{rbc,rbc1}.
072bc831d00d069997282016e47c9ad44dc7dc1e
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\subsection{Logarithmic neural scales} Receptive fields that are evenly-spaced and of equal width on a logarithmic scale (Fig.~\ref{fig:FreeSimo}a, top) lead naturally to the Weber-Fechner perceptual law. A logarithmic scale implies several properties of the receptive fields (Fig.~\ref{fig:FreeSimo}a bottom). A logarithmic scale implies that there should be fewer receptive fields centered at high values of $x$, and receptors coding for higher values of $x$ should have wider receptive fields. These qualitative impressions can be refined to several quantitatively precise relationships. If the receptive fields are evenly-spaced on a logarithmic axis then the width of receptive fields centered on a value $x$ should go up proportional to $x$. More generally, if the $i$th receptor has a receptive field centered on $x_i$, then the shape of the receptive field should be constant for all receptors, but the width of the receptive field should scale with the value of $x_i$. If we denote the center of the $i$th cells receptive field as $x_i$ and define $\Delta_i \equiv x_i - x_{i-1}$, then a logarithmic scale implies that the ratio of adjacent receptors, $\Delta_{i+i}/\Delta_i$, is a constant for all values of $i$.\footnote{Note that while logarithmic scaling implies a constant ratio of adjacent receptor spacings, the converse is not true. For instance, if the ratio of receptor spacings was one, this would not lead to logarithmic scaling. Similarly, if we took a set of receptors on a logarithmic scale and added a constant to each one, the ratio $\Delta_{i+1}/\Delta_i$ would be unchanged and remain constant with respect to $i$ whereas $x_{i+1}/x_i$ would no longer be constant with respect to $i$.} Finally we note that because $\log 0$ is not finite, a logarithmic scale with a finite number of receptors cannot continue to $x=0$. In practice, this means that if the brain makes use of logarithmic scales, there must be some other consideration for values near zero. In fact, as we will see below, the brain appears to solve this problem in the visual system by including a region of approximately constant receptor spacing for small $x$ (note the flat region around zero in Fig.~\ref{fig:FreeSimo}b). \subsection{Logarithmic neural scales in the mammalian brain} There is strong evidence that the visual system in mammals obeys logarithmic scaling in representing \emph{extrafoveal} retinal position. In monkeys, both the spacing of receptive fields (the cortical magnification factor) \cite{DaniWhit61,HubeWies74,VanEEtal84,GattEtal88} and the width of receptive fields \cite{HubeWies74,GattEtal88,Schw77} obey the quantitative relationships specified by logarithmic scaling to a good degree of approximation outside of the fovea. Figure~\ref{fig:FreeSimo}b (reproduced from \citeNP{FreeSimo11}) shows receptive field width in the macaque across several cortical regions. The fovea is visible as a region of constant receptive field width, followed by a much larger region showing a linear relationship between field width and eccentricity, consistent with logarithmic scaling. If the curves obeyed logarithmic scaling precisely, the lines would continue to the origin rather than flattening out in the fovea. However, because the spacing between receptors would go to zero, this would require an infinite number of receptors. In the case of vision, logarithmic scaling can ultimately be attributed to the distributions of receptors along the surface of the retina. However, this logarithmic mapping may be more general, extending to other sensory \cite{MerzEtal73} and motor maps \cite{Schw77}. In addition, evidence suggests that logarithmic scaling holds for variables that are not associated with any sensory organ. Neural evidence from monkeys and humans \cite{NiedMill03,HarvEtal13} suggests that neural representations of non-verbal numerosity are arranged on a logarithmic scale, consistent with a broad range of behavioral studies of non-verbal mathematical cognition \cite{GallGelm92,FeigEtal04}. Behavioral and theoretical work \cite{BalsGall09,HowaEtal15} suggests that a similar form of logarithmic scaling could also apply to representations of time, which is qualitatively consistent with growing body of neurophysiological evidence from ``time cells'' in the hippocampus, entorhinal cortex, and striatum \cite{SalzEtal16,KrauEtal15,MellEtal15}. The neurophysiological evidence that these ``cognitive'' variables use logarithmic scaling is not nearly as well-quantified as visual receptive fields. \subsection{Overview} If a set of receptors was optimized under the expectation of a particular form of statistics in the world, it would be suboptimal, perhaps disastrous, if the organism encountered a state of the world with very different statistics. The next section (``Optimal receptor distribution\ldots'') considers the problem of optimally placing receptors to represent an unknown arbitrary function from a local perspective. The result of this section is that for all receptors to convey the same amount of information as their neighbors, receptor spacings should be in a constant ratio. The following section~(``Global function representation'') considers the amount of information conveyed by a set of receptors about a function controlled by some fixed, but unknown, scale. The global organization that leads to equivalent information for a wide range of scales has a region of constant spacing (ratio~1) followed by a region with constant ratio spacing (ratio~$>1$) which resembles the organization of the visual system with a closely-packed fovea surrounded by a logarithmic scale. \section{Optimal receptor distribution for representing arbitrary unknown functions} \label{sec:local} We want to represent a function in the world over some continuous real value $x$. Receptors sample the function in some neighborhood such that each receptor is ``centered'' on a particular value of $x$. The goal is to distribute the receptors to enable them to represent an arbitrary and unknown function $f(x)$. In order for this problem to be meaningful we must assume that there is a finite number of receptors. We assume here that the lower and upper bounds of the scale $x$ are physically constrained. At the lower bound, perhaps there is some minimal degree of resolution that can be achieved with the receptors. In some cases the upper bound may be given by the properties of the world or anatomy. For instance, in vision, ecccentricity cannot possibly be larger than $\pi$. Our question, then is how to distribute the location of the receptors within the range of $x$ values to be represented. Each additional receptor provides an additional benefit to the organism to the extent that it provides additional information about the function. Intuitively, if we clumped all of the receptors close together this would be suboptimal, because the receptors would all be conveying the same information and there would be no receptors to communicate information about other regions of the function. Naively, we might expect that the solution is trivial---perhaps receptors should be evenly-spaced along the $x$-axis. It will turn out that this is not in general the optimal solution. Rather, the spacing of each pair of receptors should be in a constant ratio to the spacing of the previous pair of receptors. Constant spacing corresponds to a ratio of~1; it will turn out that other ratios are admissable as well. In order to do the calculation we need to measure the redundancy between a pair of receptors. There are a number of measures of redundancy one might use. For instance, we might measure the mutual information between the output of two receptors after observing the world for some period of time. There are other measures that one might use. All things equal, we would expect the redundancy between two receptors to be higher for receptors that are placed close together rather than receptors that are placed far apart. The development below applies to any measure of redundancy that obeys some basic properties.\footnote{After the general derivation we include a worked example with a particularly tractable measure of redundancy.} The distribution of receptors is optimal if each receptor is expected to be as redundant with its predecessor as it is with its successor. If that was not the case, then we could move one of the receptors and get more non-redundant information out about the function. \subsection{Formulation of the problem} \begin{figure} \centering \begin{tabular}{lc} \textbf{a}\\ & \includegraphics[width=0.3\columnwidth]{xzeroxprexpostwide}\\ ~\\ \textbf{b}&\\ & \includegraphics[width=0.55\columnwidth]{discretize} \end{tabular} \caption{ \textbf{a.} Discretizing the function. After picking some small resolution (``jnd'' in the figure), we discretize the signal. This defines a variable $s(x)$ describing the distance (or scale) between $x$ and the next location where the function takes on a different discretized value. \textbf{b.} Schematic for notation. The derivation assumes that there are receptors at $\ensuremath{{x_{i-1}}}$ and $\ensuremath{{x_i}}$ separated by $\ensuremath{\Delta_{i}}$. The goal is to choose $\ensuremath{\Delta_{i+1}}$ (and thus $\ensuremath{{x_{i+1}}}$) such that the redundancy between the pairs of receptors is equated. \label{fig:xzeroschematic} } \end{figure} Suppose we have set of receptors with receptive fields centered on positions $x_1$, $x_2$, \ldots $x_N$. As before, let us denote the distance between the locations of receptor $\ensuremath{{x_{i-1}}}$ and $\ensuremath{{x_i}}$ as $\ensuremath{\Delta_{i}}$. Constant receptor spacing would imply that $\ensuremath{\Delta_{i}} = \ensuremath{\Delta_{i+1}}$ for all $i$. It will turn out that this is not the general solution to our problem. Rather, the solution is to space the receptors such that the ratio of adjacent receptor spacings is a constant across the scale. That is, the ratio $\ensuremath{\Delta_{i+1}}/\ensuremath{\Delta_{i}}$ takes the same value for all $i$. We assume that in the neighborhood of each $x_i$, the function is controlled by some scale $s_\ensuremath{{x_i}}$. To quantify this in an unambiguous way, assume that our receptors can only respond over a finite range with some non-zero resolution. Accordingly, we discretize the function to be represented to some degree of resolution (Figure~\ref{fig:xzeroschematic}a) so that at each value of $x$, $s_x$ gives the distance to the next value of $x$ where the discretized function would take a different value than the discretized $f(x)$. That is, for each $x$, the discretized function $f(x) = f(x')$ for each $x < x' < x + s_x$. Note that if we knew the scale $s_\ensuremath{{x_i}}$ at $\ensuremath{{x_i}}$, it would be straightforward to estimate the redundancy between the receptor at $\ensuremath{{x_i}}$ and the receptor at $\ensuremath{{x_{i+1}}}$. If we knew that the spacing between receptors $\ensuremath{\Delta_{i+1}}$ were much smaller than $s_\ensuremath{{x_i}}$, the receptors would measure the same value and we would expect the redundancy to be high. In contrast, if we knew that the spacing between the receptors were much smaller than $s_\ensuremath{{x_i}}$ then the value of the function would have changed at least once between $\ensuremath{{x_i}}$ and $\ensuremath{{x_{i+1}}}$ and we would expect the redundancy to be low. Let us denote the function relating the redundancy we would observe between receptors spaced by a particular value $\ensuremath{\Delta_{i}}$ if we knew the scale took a particular value $s$ as $\alphaspre$. We assume $\alphaspre$ to be a monotonically decreasing function of its argument. Note that we could choose any number of different ways to measure redundancy. For instance, we could treat mutual information as our measure of redundancy and this, coupled with knowledge of the properties of the receptors, would specify a specific form for $\alphaspre$. But we could just as well measure redundancy in other ways or assume different properties of our receptors, which would result in a different form for $\alphaspre$. The arguments below about optimal receptor spacing apply to any measure of redundancy that obeys some basic properties described below and does not depend critically on how one chooses to measure redundancy. Observing the redundancy between the receptors at $\ensuremath{{x_{i-1}}}$ and $\ensuremath{{x_i}}$ spaced by $\ensuremath{\Delta_{i}}$ allows us to infer the value of the scale at the first receptor $s_\ensuremath{{x_{i-1}}}$. For instance, if the redundancy is maximal, that implies that the function did not change between $\ensuremath{{x_{i-1}}}$ and $\ensuremath{{x_i}}$. In contrast, if the redundancy between the receptors at $\ensuremath{{x_{i-1}}}$ and $\ensuremath{{x_i}}$ is very small, that implies that the scale of the function at $\ensuremath{{x_{i-1}}}$, $s_\ensuremath{{x_{i-1}}}$ was smaller than $\ensuremath{\Delta_{i}}$. Knowing the scale at $\ensuremath{{x_{i-1}}}$ places constraints on the value at $\ensuremath{{x_i}}$. For instance, if we knew with certainty that $s_\ensuremath{{x_{i-1}}} = \ensuremath{\Delta_{i}} + C$, then we would know with certainty that $s_\ensuremath{{x_i}} = C$. So, if we knew the probability of each value of $s_\ensuremath{{x_{i-1}}}$ this constrains the probability distribution for $s_\ensuremath{{x_i}}$. Let us fix the spacing $\ensuremath{\Delta_{i}}$ between the receptors at $\ensuremath{{x_{i-1}}}$ and $\ensuremath{{x_i}}$, sample the world for some time (observing many values of the the receptor outputs in response to many samples from the function) and denote the average value of redundancy we observe as $\alphaobs$. Denoting the (unknown) probability of observing each possible value of $s_\ensuremath{{x_{i-1}}}$ as $\ppre{s}$, then the observed value of redundancy, $\alphaobs$, resulted from an integral: \begin{equation} \alphaobs = \int \ppre{s}\ \alphaspre\ ds \label{eq:alphaobs} \end{equation} This can be understood as an inference problem; observation of $\alphaobs$ leads us to some belief about the distribution $\ppre{s}$. Informed by this knowledge, we then place the third receptor at a spacing $\ensuremath{\Delta_{i+1}}$. If we knew the distribution of scales at $\ensuremath{{x_i}}$, $\ppost{s}$, then we would expect to observe a redundancy of \begin{equation} \alphapred = \int \ppost{s}\ \alphaspost\ ds \label{eq:alphapred} \end{equation} between the second pair of receptors. Our problem is to choose $\ensuremath{\Delta_{i+1}}$ such that we would expect $\alphapred = \alphaobs$. \subsection{Minimal assumptions about the statistical properties of the world} The actual value of $\ensuremath{\Delta_{i+1}}$ that makes $\alphapred=\alphaobs$ of course depends on the detailed properties of the receptors, the function $\alphas$ and the statistics of the world. We show, however, that for minimal assumptions about the world, this problem results in the solution that, whatever the value of $\alphaobs$ one finds that for some spacing $\ensuremath{\Delta_{i}}$, the choice of $\ensuremath{\Delta_{i+1}}$ is such that only $\ensuremath{\Delta_{i+1}}/\ensuremath{\Delta_{i}}$ is affected by the observed value $\alphaobs$. First, we make minimal assumptions about the function to be estimated. We assume an uninformative prior for $\ppre{s}$. We make the minimal assumptions that successive values of $s$ are independent, and that the value of the function past the $s_x$ is independent of the value at $x$. That is we assume that $f(x)$ and $f(x+s_x)$ are independent of one another, as are $s_x$ and $s_{x+s_x}$. \footnote{The critical point of these assumptions is that at no point to we introduce an assumption about the statistics of the world that would fix a scale for our receptors. For instance, the result of receptor spacings in a constant ratio would also hold if we assumed a power law prior for $\ppre{s}$ rather than a uniform prior because the power law is a scale-free distribution. Similarly, it is acceptable to relax the assumption of independence between $f(x)$ and $f(x+s_x)$ as long as doing so does not introduce a scale. \label{foot:power} } Second, we require that the receptors do not introduce a scale \emph{via} $\alphas$. In general, the function $\alphas$ will depend on how we choose to quantify redundancy and the properties of the receptors. However, as long as $\alphas$ rescales, such that it can be rewritten in a canonical form \begin{equation} \alphas = \alphahat(s/\Delta), \label{eq:alphahat} \end{equation} the conclusions in this section will hold. One can readily imagine idealized settings where Eq.~\ref{eq:alphahat} will hold. For instance, we could assume we have perfect receptors that sample the function at only one point and take our measure of redundancy to be one if the receptors observe the same value up to the resolution used to specify $s$ (Fig.~\ref{fig:xzeroschematic}b) and zero otherwise. More generally, for imperfect receptors that sample a range of values, Eq.~\ref{eq:alphahat} requires that the receptive fields scale up with $\Delta$. This assumption is necessary, but not sufficient to equate redundancy between receptors. Our third requirement is referred to as the Copernican principle. Because it is critical to the argument in this section, we discuss it in some detail in the next subsection. \subsection{The Copernican Principle} In addition to minimal assumptions about the statistics of the world and the properties of the receptors, we assume that the world's choice of $s_\ensuremath{{x_{i-1}}}$ is unaffected by our choice of $\ensuremath{\Delta_{i}}$. The belief that there is nothing privileged about one's point of observation is referred to as the Copernican principle. The Copernican principle was employed by \citeA{Gott93} to estimate a probability distribution for the duration of human civilization. Because this provides a concrete illustration of an important concept, it is worth explaining Gott's logic in some detail. Suppose we observe Fenway Park in 2017 and learn it has been standing for 103 years. Knowing nothing about construction techniques or the economics of baseball we want to estimate how much longer Fenway Park will stand. Because there is nothing special about our current viewpoint in 2017, our observation of Fenway should be uniformly distributed across its lifetime. That is, the observation at \ensuremath{t_\textnormal{now}}{} in 2017 ought to be uniformly distributed between $\ensuremath{t_\textnormal{start}}$ when Fenway Park was constructed (in 1914) and the unknown time of its destruction $\ensuremath{t_\textnormal{end}}$. Gott argued that we should expect $\ensuremath{t_\textnormal{now}}$ to be uniformly distributed between $\ensuremath{t_\textnormal{start}}$ and $\ensuremath{t_\textnormal{end}}$, such that we should expect $\ensuremath{t_\textnormal{end}} - \ensuremath{t_\textnormal{now}}$ to be longer than $\ensuremath{t_\textnormal{now}}-\ensuremath{t_\textnormal{start}}$ half the time. This means that the age of the structure (102 years) fixes the units of the distribution of $\ensuremath{t_\textnormal{end}} - \ensuremath{t_\textnormal{start}}$. If we observed an object that has been existing for 1030 years, or for 103 seconds, then our inference about its expected duration would have the same shape, but only differ in the choice of units. The Copernican argument also applies to our inference problem. Here the scale at the first receptor $s_\ensuremath{{x_{i-1}}}$ plays a role analogous to the duration of the object $\ensuremath{t_\textnormal{end}} - \ensuremath{t_\textnormal{start}}$. Suppose that we have a set of receptors with a known function $\alphas$ that obeys Eq.~\ref{eq:alphahat} and we encounter a world with some unknown statistics. In this first world we choose some $\ensuremath{\Delta_{i}}$, observe some measure of redundancy $\alphaobs$, use some method of inference to estimate $\ppre{s}$ and $\ppost{s}$ and then select the value of $\ensuremath{\Delta_{i+1}}$ to yield the expectation that $\alphapred=\alphaobs$. Let us refer to the distribution of $\ppost{s}$ that we inferred from fixing $\ensuremath{\Delta_{i}}$ to its particular value and observing $\alphaobs$ as $\palpha{s;\ensuremath{\Delta_{i}}}$. Now, suppose that we encounter another world with the same receptors. Assume further that we choose a different $\ensuremath{\Delta_{i}}$ but observe the \emph{same} value of $\alphaobs$. How should our inference about the distribution of $s_\ensuremath{{x_i}}$ be related to our inference from the first world? If Eq.~\ref{eq:alphahat} holds, the Copernican Principle requires that all of the distributions that can be inferred for a particular value of $\alphaobs$ must be related to one another \emph{via} a canonical form: \begin{equation} \palpha{s;\ensuremath{\Delta_{i}}} = \frac{1}{\ensuremath{\Delta_{i}}} \hat{p}_\alpha(s/\ensuremath{\Delta_{i}}). \label{eq:pcanon} \end{equation} If Equation~\ref{eq:pcanon} did not hold, it would imply that we can infer something about the world's choice of $s_\ensuremath{{x_i}}$ from our choice of $\ensuremath{\Delta_{i+1}}$ \emph{beyond} that communicated by the value of $\alphaobs$. As long as the receptors and our measure of redundancy scale with $\ensuremath{\Delta_{i}}$ as in Eq.~\ref{eq:alphahat} then any such effect would imply that the world's choice of $s$ and our choice of $\ensuremath{\Delta_{i}}$ are not independent and thus violate the Copernican Principle. \subsection{Ratio scaling results from minimal assumptions about the statistics of the world and the Copernican Principle} Equations~\ref{eq:alphahat}~and~\ref{eq:pcanon} imply that redundancy is equated for receptor spacings in some ratio. To see this, let us rewrite Eq.~\ref{eq:alphapred} and find \begin{eqnarray} \alphapred(\ensuremath{\Delta_{i+1}}) & = & \frac{1}{\ensuremath{\Delta_{i}}} \int_0^\infty \hat{p}_\alpha\left(s/\ensuremath{\Delta_{i}}\right) \ \alphahat(\ensuremath{\Delta_{i+1}}/s) \ ds \nonumber\\ & = & \int_0^\infty \hat{p}_\alpha(s') \ \alphahat(r/s') \ ds' \label{eq:canon} \end{eqnarray} In the second line $s' \equiv s/\ensuremath{\Delta_{i}}$ and we define the ratio $r \equiv \ensuremath{\Delta_{i+1}}/\ensuremath{\Delta_{i}}$. The right hand side of Equation~\ref{eq:canon} clearly does not depend on $\ensuremath{\Delta_{i+1}}$ directly, but only on the ratio. The finding, then, is that a principle of minimal assumptions about the statistics of the world coupled with the Copernican Principle implies that the optimal distribution of receptor locations is such that the ratio of successive receptor spacings is constant. Logarithmic neural scales also imply that the ratio of successive receptor spacings are constant. In this sense, logarithmic scales can be understood as a response to the demand that the receptor layout are expected to equalize redundancy across receptors in a world with unknown statistics. Because logarithmic scales do not uniquely predict constant ratio spacing this cannot be the entire story. We pursue additional constraints that imply logarithmic scales, with some important exceptions when $x$ approaches zero, in the next section. Before that, for concreteness we include a worked example for a idealized set of receptors and a specific choice for $\alphas$. \subsection{A worked example} The general development above provides a set of conditions that result in a constant ratio of spacing between receptors. In order to make this more concrete, we work out a specific example with specific choices for the properties of the receptors, the measure of redundancy and the method of inference. This example assumes that the receptors are perfect and the $i$th receptor samples the function only at the location $x_i$. For simplicity we define a measure of redundancy that gives $\alpha = 1$ if the two receptors observe the same value of $f$ and $\alpha=0$ otherwise. This lets us write out \[ \alphaspre = \begin{cases} 1, & \mbox{if} \ \ensuremath{\Delta_{i}} < s_\ensuremath{{x_{i-1}}}\\ 0, & \mbox{if}\ \ensuremath{\Delta_{i}} \ge s_\ensuremath{{x_{i-1}}} \end{cases} \] The same holds for $\alphaspost$, only comparing $\ensuremath{\Delta_{i+1}}$ to $s_\ensuremath{{x_i}}$. With our simplified definition of redundancy, any value of $\alphaobs$ that is not zero or one must have resulted from a mixture of those two cases with probability $\alphaobs$ and $1-\alphaobs$ respectively. Let us first consider the case where $\ensuremath{\Delta_{i}} < s_\ensuremath{{x_{i-1}}}$, as it aligns perfectly with the Gott argument. If $s_\ensuremath{{x_{i-1}}} > \ensuremath{\Delta_{i}}$, then $s_\ensuremath{{x_{i-1}}}$ plays the role of the unknown duration of the lifetime of an object such as Fenway Park, $\ensuremath{t_\textnormal{end}}-\ensuremath{t_\textnormal{start}}$. The difference between the two receptors $\ensuremath{\Delta_{i}}$ plays the role of the time of the current observation $\ensuremath{t_\textnormal{now}} - \ensuremath{t_\textnormal{start}}$ and the unknown $s_\ensuremath{{x_{i+1}}}$ is analogous to $\ensuremath{t_\textnormal{end}} - \ensuremath{t_\textnormal{now}}$. Gott's calculation defines the ratio $r=\left(\ensuremath{t_\textnormal{end}} - \ensuremath{t_\textnormal{now}}\right)/\left(\ensuremath{t_\textnormal{now}}-\ensuremath{t_\textnormal{start}}\right)$. The Copernican Principle leads to the belief that $\ensuremath{t_\textnormal{now}}$ ought to be uniformly distributed between $\ensuremath{t_\textnormal{start}}$ and $\ensuremath{t_\textnormal{end}}$. Thus the cumulative distribution of $r$ obeys: \begin{equation} P\left(r > Y\right) = \frac{1}{1+Y} \label{eq:Gott} \end{equation} Note that the elapsed duration $\ensuremath{t_\textnormal{now}} - \ensuremath{t_\textnormal{start}}$ only enters this expression \emph{via} the ratio. Similar arguments apply to the spacing of receptors. If we observe $\alphaobs=1$, then the posterior $\ppost{s}$ is controlled by Eq.~\ref{eq:Gott}, with $\ensuremath{\Delta_{i}}$ in place of $\ensuremath{t_\textnormal{now}} - \ensuremath{t_\textnormal{start}}$ and $\ensuremath{\Delta_{i+1}}$ in place of $\ensuremath{t_\textnormal{end}} - \ensuremath{t_\textnormal{now}}$.\footnote{That is, Eq.~\ref{eq:Gott} gives the cumulative of the posterior distribution.} In this toy problem, the value of $\alphapred$ is given by Eq.~\ref{eq:Gott} if $\alphaobs=1$. If instead of $\alphaobs = 1$, we observed $\alphaobs=0$, we would infer a uniform distribution of $s(\ensuremath{{x_i}})$.\footnote{One can argue for other ways to make the inference in this toy problem when $\alphaobs=0$. As long as those depend only on the ratio $\ensuremath{\Delta_{i+1}}/\ensuremath{\Delta_{i}}$ and not explictly on $\ensuremath{\Delta_{i+1}}$, the conditions of the general argument still hold.} If our prior on the distribution of $s_\ensuremath{{x_i}}$ is uniform, then $\alphapred$ will be one for any finite value of $\ensuremath{\Delta_{i+1}}$. After sampling the world for some period of time, if we observe $\alphaobs$ as a number between zero and one, then our posterior distribution of scales should be a mixture of the inference from the two cases and simplify Eq.~\ref{eq:alphapred} as: \begin{eqnarray} \alphapred(\ensuremath{\Delta_{i+1}}) = \alphaobs \frac{1}{1+\ensuremath{\Delta_{i+1}}/\ensuremath{\Delta_{i}}} + \left(1-\alphaobs\right) \label{eq:simpleequation} \end{eqnarray} It is clear that the right hand side is only a function of the ratio $r \equiv \ensuremath{\Delta_{i+1}}/\ensuremath{\Delta_{i}}$ so that the value of $\ensuremath{\Delta_{i+1}}$ that makes $\alphapred=\alphaobs$ depends only on the ratio.\footnote{In this simple example there is a solution for values of $\alphaobs > 1/2$. } For any value of $\alphaobs$, the same value of $r$ satisfies this equation for any choice of $\ensuremath{\Delta_{i}}$. In this simple problem we see that the choice of $\ensuremath{\Delta_{i}}$ can only affect the answer by fixing its units. This naturally results in the ratio of adjacent receptor spacings being constant, as implied by logarithmic receptor scales. \section{Global function representation} \label{sec:global} \begin{figure} \begin{tabular}{lclc} \textbf{a} & & \textbf{b}\\ &\includegraphics[width=0.4\columnwidth]{scalesbars-1-1.pdf} &&\includegraphics[width=0.4\columnwidth]{scalesbars-2-1.pdf}\\ \textbf{c} & & \textbf{d}\\ &\includegraphics[width=0.4\columnwidth]{scalesbars-2-3.pdf} &&\includegraphics[width=0.4\columnwidth]{scalesbars-3-3.pdf} \end{tabular} \caption{ { \bf Cartoon illustrating the interaction of receptor spacing and function scale.} In each panel, the top curve shows a specific function $f_s(x)$; the functions in the four panels differ only in their scale $s$. On the bottom of each panel, the vertical lines display receptor locations. {\bf a-b.} Constant spacing. When $c=0$, the receptors are evenly-spaced, fixing a scale to the receptors. If the function has the same scale as the receptors, as in \textbf{a}, each receptor captures a different value of the function, the veridical region runs the entire length of the receptor array and the information transmitted is maximal. However, if the scale of the function is larger than the constant spacing, as in \textbf{b}, the array of receptors conveys less information because there is less information in the function over that range. \textbf{c-d.} Ratio spacing. When $c > 0$, the receptors can extend over a much wider range of $x$ values than with constant spacing. The receptors carry information about the function over a veridical region where $\Delta_i < s$; the border of the veridical regions is shown (in cartoon form) as a vertical red line. As $s$ increases (from \textbf{c} to \textbf{d}), the size of the veridical region in $x$ changes, but as long as the border is covered by the receptors, the amount of nonredundant information in the veridical region is constant as a function of $s$. This property does not hold if the scale of the function approaches the minimum receptor spacing. \label{fig:adaptive}} \end{figure} The foregoing analysis conducted at the level of pairs of receptors showed that the optimal spacing in order to represent arbitrary functions places the receptors in constant ratio, but does not specify the value of the ratio. It is convenient to parameterize the ratio between adjacent receptors by a parameter $c$ such that $r \equiv \frac{\ensuremath{\Delta_{i+1}}}{\ensuremath{\Delta_{i}}} = 1+c$. If $c=0$, receptor spacing is constant; if $c>0$ the ratio is constant, as in a logarithmic scale. We consider the global coding properties of these two schemes as well as a hybrid scheme in which the first part of the axis has constant spacing ($c=0$) followed by a region with logarithmic spacing ($c>0$), analogous to the organization of the visual system, with the region of constant spacing corresponding roughly to the fovea (Fig.~\ref{fig:FreeSimo}b). \subsection{Formulating the problem} For simplicity, we assume that each receptor perfectly samples the value of a function in the world at a single perfectly-specified location and consider a simple class of functions $f_s(x)$, which consist of independently chosen values over the range 0~to~1 at a spacing of $s$ such that $f_s(x)$ and $f_s(x+s)$ are independent and $f_s(x)$ and $f_s(x+s - \epsilon)$ are identical for $\epsilon < s$. The information about a function contained in a given range by a particular instantiation of the function is just the number of entries specifying the values of the function over that range. Note that $s$ controls the density of information conveyed by $f_s$ over a given range of $x$. We assume that $f_s(x)$ is specified over the entire range of $x$ from zero to infinity so that the total amount of information that could be extracted from a function is infinite for every finite scale. Figure~\ref{fig:adaptive} shows several functions and choices of receptor scaling in cartoon form. A set of receptors can do a ``good'' job in representing a function $f_s(x)$ as long as the spacing between the receptors is less than or equal to $s$. With the simplifying assumptions used here, each of the non-redundant values of the function is captured by at least one receptor allowing reconstruction of the function with error in the $x$ location no worse than $s$. More generally, even with coarse-graining and noisy receptors it is clear that there is a qualitative difference between the ability of the receptors to measure the function when $\Delta < s$ compared to when $\Delta > s$. Let us define the veridical region of the function as the range of $x$ over which $\ensuremath{\Delta_{i}} < s$. We also assume there is some minimum possible receptor spacing $\Deltamin$. We measure the amount of veridical information $I$ conveyed by the set of receptors as the number of unique function values that the function has within the veridical region. \subsection{Constant receptor spacing, $c=0$} First, consider the implications of constant receptor spacing. If we knew the value of $s$ controlling $f_s(x)$, placing our receptors with constant separation $\Delta=s$ would be sufficient to convey all of the information in the function over the entire range of $x$ values covered by the receptors. $N$ receptors would be able to accurately represent the function over a veridical region of width $N\Delta $ and the amount of information conveyed about the function in that region is just $N$. However, our constant choice of spacing would have poor consequences if our choice of $\Delta$ did not correspond to the world's choice of $s$. If $s > \Delta$, the veridical region is still of length $N \Delta$, but the amount of information contained in that region is only $N\frac{\Delta}{s}$, decreasing dramatically as $s$ increases (curve labeled $c=0$ in Figure~\ref{fig:informationc}). With constant spacing, $c=0$, the amount of information conveyed by the receptors depends dramatically on $s$. \begin{figure} \centering \includegraphics[width=0.8\columnwidth]{IVfig4.pdf} \caption{\textbf{Information in the veridical region as a function of scale for three forms of receptor spacing.} The curve labeled $c=0$ shows the results for constant spacing. This falls off rapidly for scales larger than the scale set by the receptor spacing. Solid curves show results for ratio spacing ($c>0$) with a fovea; dashed curves show results for ratio spacing without a fovea. $N=100$ for all curves. See Figure~\ref{fig:schematicfovea}a for an intuitive explanation of why the information decreases for small values of $s$. The fovea has an additional $1/c$ receptors. Note that there is a region where the amount of information conveyed is approximately constant as a function of scale before falling off as the veridical region exhausts the number of receptors $N$. For $c=.05$, this region is off the scale of this figure. Note also the error at small scales for $c>0$ without a fovea. \label{fig:informationc} } \end{figure} \subsection{Constant receptor spacing ratio, $c>0$} If we set $c>0$ such that all receptors are placed in constant ratio the responsiveness to functions of different scales is very different. Let us place the zeroth receptor $x_0$ at $\xmin$, and the first receptor at $x_1 = \xmin + \Deltamin$. Continuing with $\Delta_i = \left(1+c\right) \Delta_{i-1}$, we find that the spacing of the $i$th receptor is given by \begin{equation} \Delta_i = \Deltamin \left(1+c\right)^{i-1} \label{eq:deltai} \end{equation} The position of the $i$th receptor is thus given by the geometric series \begin{eqnarray} x_i &=& \xmin + \frac{\Deltamin}{c}\left[\left(1+c\right)^{i} -1 \right] \label{eq:xi} \end{eqnarray} Equations~\ref{eq:deltai}~and~\ref{eq:xi} show that when $c>0$ the range of $x$ values and scales that can be represented with $N$ receptors goes up exponentially like $(1+c)^N$. For a function of scale $s$, the set of ratio-scaled receptors has a veridical region ranging from $\xmin$ to \begin{equation} \xcrit = \left\{ \begin{array}{lr} s\frac{1+c}{c} - \frac{\Deltamin}{c} + \xmin & s \leq \Delta_N \\ \\ \xmax & s > \Delta_N \end{array} \right. \end{equation} The amount of information present in the function in the veridical region is just $(\xcrit-\xmin)/s$, which is given by \begin{equation} I_{c>0} = \left\{ \begin{array}{lr} \frac{1+c}{c} - \frac{\Deltamin}{sc}& s \leq \Delta_N \\ \\ \frac{\xmax-\xmin}{s} & s > \Delta_N \end{array} \right. \label{eq:Icgreater0} \end{equation} The second expression, which happens when $s$ is larger than the largest spacing among the set of receptors, is closely analogous to the case with constant spacing ($c=0$), decreasing like $s^{-1}$. However, the first expression with $s \leq \Delta_N$ includes a term $(1+c)/c$ which is independent of $s$. The second term is small when $s$ is large, so that for large values of $s$ the information conveyed is approximately constant. However, when $s$ is small the second term is substantial and constant ratio spacing ($c>0$) fails to convey much information about the function in the veridical region. When $s=\Deltamin$ the veridical region includes only one value of the function (see Figure~\ref{fig:schematicfovea}a). The dashed line in Figure~\ref{fig:informationc} shows the analytic result when $c > 0$ and is constant across all receptors. \begin{figure} \begin{center} \begin{tabular}{lclc} \textbf{a} && \textbf{b}\\ &\includegraphics[width=0.4\columnwidth]{smallscalebars-2.pdf} &&\includegraphics[width=0.4\columnwidth]{smallscalebars-1.pdf}\\ \textbf{c}\\ &\multicolumn{3}{c}{\includegraphics[width=0.9\columnwidth]{cartoon.png}} \end{tabular} \end{center} \caption{ \textbf{Rationale for inclusion of a fovea.} \textbf{a-b.} Functions are shown with different scales, as in Figure~\ref{fig:adaptive} for logarithmically-spaced receptor locations. \textbf{a.} With a function scale that is much larger than $\Deltamin$, several values of the function can be represented (here $I=5$). \textbf{b.} When the scale goes to $\Deltamin$, then much less total information can be represented. Here only one function value fits into the veridical region. Note that the information conveyed by the same set of receptors decreases as $s$ decreases, as in the dashed lines in Figure~\ref{fig:informationc}. \textbf{c.} Schematic for notation for hybrid neural scaling. The first part of the scale, analogous to a fovea, has $n$ evenly spaced receptors from $0$ to $\xmin$. The second part of the scale region, from $\xmin$ to $\xmax$ has $N$ receptors spaced such that the ratio of adjacent differences is $1+c$. The derivation suggests $n$ should be fixed at $1/c$, whereas $N$ can in principle grow without bound. If $\xmin=\Deltamin/c$, the scale starts precisely at $x=0$. \label{fig:schematicfovea} } \end{figure} \subsection{A ``fovea'' surrounded by a logarithmic scale} Note that if the $\frac{\Deltamin}{sc}$ term in Eq.~\ref{eq:Icgreater0} were canceled out, the amount of information in the veridical region conveyed by the set of receptors would be precisely invariant with respect to $s$ over the central range. This can be accomplished by preceding the region of ratio spacing ($c>0$) with a region of constant spacing ($c=0$). In order to equalize the information carried at small scales, the region of constant spacing should have $n = \frac{1}{c}$ receptors each spaced by $\Deltamin$, as in Figure~\ref{fig:schematicfovea}b. Adding a set of constantly spaced receptors enables constant information transfer over a range of scales starting exactly at $\Deltamin$ (flat solid curves in Figure~\ref{fig:informationc}). When the region with ratio spacing starts at $\xmin=\frac{\Deltamin}{c}$, then the region of constant spacing starts at $x=0$. Note that the number of receptors in the fovea is controlled only by $c$ and is independent of the number of receptors in the region with ratio spacing. The ratio of adjacent receptor spacings is constant within each region and only fails to hold exactly at the transition between regions. \section{Discussion} Current technology is rapidly leading to a situation where we can design intelligent agents. The results in this paper point to a design choice that may have played out on an evolutionary time scale. To the extent neural representations of very different one-dimensional quantities all obey the same scaling laws, it suggests convergent evolution optimizing some design principle. One strategy that could be taken in designing receptor systems is to optimally adapt the organism to the behaviorally-relevant statistics of the environment. Evolution has certainly made this choice in a number of cases---the putative ``fly detectors'' in the frog's visual system are a famous example of this approach (\citeNP{LettEtal59,MatuEtal60}, see also \citeNP{HerzBarl92}). However, optimizing receptor arrays to a specific configuration of the world comes with a cost in terms of flexibility in responding to changes in the world. \subsection{Logarithmic receptor scales and natural statistics} In this paper, we have pursued the implications of a neural uncertainty principle---maintaining maximum ignorance about the statistics of the world---for the design of sensory receptors. Operationally, this means we have made only minimal assumptions about the statistics of the functions to be represented, limiting ourselves to uniform priors and the assumption of independence. Had we assumed a non-uniform distribution for the prior of $s$, this would have required us to estimate at least one parameter. Similarly, if the successive values of $s_x$ were not independent, we would have to have estimated at least one parameter to characterize the nature of the dependence. It can be shown that the arguments developed in the first section of this paper (``Optimal receptor distribution \ldots'') for uniform priors also generalize to scale-free (power law) priors (see footnote~\ref{foot:power}). Much empirical work has characterized statistical regularities in the world. For instance, power spectra in natural images tend to be distributed as a power law with exponent near $-2$ \cite{Fiel87,RudeBial94,SimoOlsh01}. Similarly, it has been argued that power spectra for auditory stimuli are distributed as a power law with exponent $-1$ \cite{VossClar75}. Perhaps it would be adaptive to design a set of receptors that is optimized for these naturally occurring statistics \cite<e.g.,>{WeiStoc12,Pian16}. Results showing power law spectra with similar exponents for natural images mask variability across orientations and categories of images \cite{TorrOliv03}. For instance, the exponent observed, and especially second-order statistics, vary widely across pictures of landscapes \emph{vs} pictures of office environments or pictures of roads. To the extent these statistics differ, if receptors were optimized for any one of these categories of images, they would be suboptimally configured for other environments \cite{WeiStoc12}. Moreover, because the eyes are constantly in motion, the statistics of natural images are not necessarily a good proxy for the statistics of light landing on the retina averaged by synaptic time constants. More concretely, the temporal variation due to fixational eye movements has the effect of whitening images with conventional power law spectra \cite{KuanEtal12,Rucc08}. In the time domain, it can be shown that in the presence of long-range correlated signals, logarithmic receptor spacing is optimal for predicting the next stimulus that will be presented \cite{ShanHowa13}. A derivation based on a uniform prior is more general than a belief that the statistics of the world should be power law. The former would apply across a range of environments and equally well apply to any world with some unknown statistics. Moreover, the development in the next section (entitled ``Global function representation''), which predicts the existence of a fovea under some circumstances, would be quite different if we had a strong prior belief about the probability of observing a particular scale $s$. If the goal was to maximize the information across states of the world, and if there was a non-uniform prior about the distribution of scales, we would not have obtained asymptotically logarithmic receptor spacing. \subsection{Specific predictions deriving from the approach in this paper} There have been a great many other approaches to understanding the ubiquity of logarithmic psychological scales and the more general problem of constructing psychological scales with well-behaved mathematical properties from continua in the world. These are briefly reviewed in the next subsection (entitled ``Placing this work in historical context'') with special attention to drawing contrasts with the present paper. However, to our knowledge, the predictions about the scaling of the foveal region are unique to the present approach. Asymptotically, a logarithmic neural scale enables a set of receptors to provide equivalent information about functions of a wide range of intrinsic scales, implementing the principle of neural equanimity. However, logarithmic neural scales are untenable as $\Delta$ goes to zero---the number of receptors necessary tends to infinity and the set of receptors conveys less information about functions with scale near the smallest receptor spacing (corresponding to the maximum resolution). In the absence of a fovea, there is a small-scale correction (dashed lines Fig.~\ref{fig:informationc}). A region of constant receptor spacing near zero---a fovea---solves this problem allowing the set of receptors to carry the same amount of non-redundant information about functions with every possible scale ranging from $\Deltamin$ to $\Delta_N$. Equalizing the information across scales results in a fixed number of receptors within the fovea. Critically, the number of receptors along a radius of the fovea, measured in units of $\Deltamin$, depends only on $c$. The value of $c$ can be estimated from noting the slope of the line relating receptor size to the center of the receptive field, as in Figure~\ref{fig:FreeSimo}b. \footnote{This prediction is at least roughly consistent with the trend of the hinged linear functions generated by \cite{FreeSimo11} shown in Fig.~\ref{fig:FreeSimo}b, but this visual impression should not be taken as strong quantitative evidence.} Note that the value of $c$ is estimated from receptive fields outside of the fovea, whereas $\Deltamin$ and the number of receptors along a radius are estimated from information about the fovea. As such, measurement of the two quantities ought to be completely independent. This quantitative relationship constitutes a specific prediction of this approach and can be evaluated across brain regions within the same modality, and even across modalities. \subsection{Placing this work in historical context} The present paper provides a rational basis for logarithmic neural scales and the presence of a ``fovea'' for values of $x$ near zero. It also adds to a long tradition of work in mathematical psychology organized along several themes. \subsubsection{Measurement theory} Researchers in mathematical psychology have long considered the form of psychological spaces. This work has concluded, \emph{contra} the present approach, that there is not a privileged status for logarithmic psychological scales. Fechner's \citeyear{Fech60} reasoning for logarithmic psychological scales started with the empirical observation of a constant Weber fraction and then made what was apparently a straightforward conclusion: integration of the Weber law results in logarithmic psychological scale. That is, the Weber law states that the change in the psychological discriminability $\Delta p$ due to a change in the magnitude of a physical stimulus $\Delta x$ goes like $\Delta p \propto \frac{\Delta x}{x}$. Taking the limit as $\Delta x$ goes to zero and integrating gives a logarithmic scale $p = \log x + C$. On its face this seems reasonable; surely the psychological distance between two stimuli should be the sum of the JNDs along the path between them. \citeA{LuceEdwa58} noted that Fechner's reasoning is not in general sound; the integration procedure is only valid if the empirical Weber fraction result holds. \citeA{Luce59} showed that a logarithmic psychological scale is one of a class of relationships that can map a physical scale with a natural zero (a ratio scale) onto an interval scale that is translation-invariant \cite{Stev46}. \citeA{DzhaColo99,DzhaColo01} developed a much more general framework for constructing multidimensional psychological spaces from local discriminability functions (see \citeNP{LuceSupp02} for an accessible introduction to the history of these questions) that also does not find any privileged status for logarithmic functions. And none of these approaches provides a natural account for the existence of a ``fovea''---a region of heightened discriminabiligy near $x=0$. To the extent that logarithmic neural scales are a general property of the brain, the neural considerations described in this paper provide a potentially important complement to measurement theory. \subsubsection{Recent approaches to Weber-Fechner scaling} More recently, investigators in mathematical psychology and related fields have also considered the rationale underlying the apparent ubiquity of the Weber-Fechner law. These approaches have in general taken a more restrictive approach than the quite general considerations in this paper. Moreover, they do not lead to the specific predictions regarding the fovea derived here. Some recent approaches have noted that if the psychological scale is designed to minimize a relative error measure \cite{SunEtal12,PortSvai11}, logarithmic scales naturally result. For instance, in stimulus quantization, the goal is to choose a discrete quantization of the stimulus space in order to minimize the expected value of some measure of reconstruction error between the true stimulus and the quantized stimulus. If one minimizes mean squared error, the optimal quantization is uniform. However, if one attempts to minimize relative error, and if the quantization is constrained to have a fixed entropy, it can be shown that the optimal quantization is on a logarithmic scale independent of the input stimulus statistics \cite{SunGoya11,SunEtal12}. \citeA{Wilk15} compared the suitability of various relative error measures from an evolutionary perspective. These approaches depend critically on the assumption of relative error measures, without explaining why those might be desirable (other than that they result in Weber-Fechner scales). Other recent approaches that result in Weber-Fechner spacing make specific assumption about the statistics of the world. For instance \citeA{ShouEtal13} described the variability in perception as due to the variability in spike count statistics from a neuron with some I/O function and Poisson variability. They concluded that the Weber law is optimal if the world has power law statistics. Similarly, \citeA{Pian16} derived a Weber-Fechner scale for numerosity under the assumption that the probability that each possible number will be utilized goes down like a power law. This builds on earlier work deriving a rational basis for power law forgetting on the probability of use of a memory a certain time in the past \cite{AndeScho91}. \citeA{WeiStoc12} argued that receptor spacings should be constructed so that each receptor carries the same amount of information about the world. This depends on the statistics of the world and a logarithmic neural scale results only if the world has power law statistics with exponent $-1$. To the extent the world has different statistics in different modalities and different environments, these approaches are limited in accounting for the ubiquity of Weber-Fechner neural scales. The approach in the present paper is more general in that it does not make any strong prior assumptions about the statistics of the world. \subsubsection{Universal exponential generalization} On its face, logarithmic neural scales seem to be closely related to Shepard's work on universal exponential generalization \cite{Shep87}. If the mapping between the physical world and the neural scale is logarithmic, one would expect that the mapping between the neural scale and the physical world is exponential. Indeed, the approach in the section ``Optimal receptor distribution \ldots'' was very much inspired by Shepard's pioneering approach to the structure of an abstract psychological space \cite{Shep87}. However, the results are actually quite distinct. \citeA{Shep87} studied the problem of generalization; if a particular stimulus $x$ leads to some response, what is the optimal way to generalize the response to other stimuli in the neighborhood of $x$? Shepard's derivation hypothesizes a consequential region of unknown size. Although the results depend on one's prior belief about the distribution of sizes, if one uses a Copernican principle like Gott to set the prior, one obtains an exponential generalization gradient. The size of the consequential region in Shepard's work plays a role analogous to the scale of the function in this paper. And an exponential gradient is analogous to a logarithmic scale. The value $1+c$ serves the role of the base of the logarithm in the scales derived here. This corresponds (roughly) to an exponential generalization gradient that goes like $(1+c)^x$. In Shepard's framework, the ``space constant'' of the exponential gradient is controlled by the expected size of the consequential region, which functions like a prior choice of scale. Note that the neural scales developed here would not lead to an exponential generalization gradient around any particular point $x$. The width of receptive fields (which is proportional to $\Delta$) provides a natural lower limit to generalization. We would not expect this generalization gradient to be symmetric in $x$ (see Fig.~\ref{fig:FreeSimo}a). Of course psychological generalization need not be solely a function of neural similarity and is more difficult to observe than receptive fields. \subsection{Neural scales for cognitive dimensions} The quantitative evidence for spacing of visual receptive fields is quite strong, due to intense empirical study for decades \cite{DaniWhit61,HubeWies74,VanEEtal84}. The derivation in this paper applies equally well to any one-dimensional quantity over which we want to represent a function. If the neural uncertainty principle is a design goal for the nervous system, it should be possible to observe similar scaling laws for representations of other dimensions. As discussed in the introduction, there is evidence that the brain maintains receptive fields for dimensions that do not correspond to sensory continua. In the visual system, a neuron's spatial receptive field is the contiguous region over visual space that causes the neuron to be activated. Cognitive continua can also show the same coding scheme. For instance, in a working memory task that required macaques to remember the number of stimuli presented over the delay, neurons in lateral prefrontal cortex responded to a circumscribed set of numbers to be remembered. That is, one neuron might be activated when the number of stimuli to be remembered is 3-5, but another neuron might be activated when the number to be remembered is 4-7. Time cells fire during a circumscribed period of time within a delay interval. Each time cell can be thought of as having a receptive field over past time. In much the same way that a neuron with a visually-sensitive receptive field fires when a stimulus is in the appropriate part of the visual field, so too the time cell fires when a relevant stimulus, here the beginning of the delay interval, enters its temporal receptive field. Because different time cells have different receptive fields, as the stimulus recedes into the past with the passage of time, a sequence of time cells fire. In this way, receptive fields tile time, a continuous dimension in much the same way that eccentricity or numerosity would be coded. Time and numerosity do not correspond to sensory receptors in the same way that, say, eccentricity in the visual system does. Nonetheless, the logic of the arguments in this paper apply equally well to time and numerosity. Do these ``cognitive'' receptive fields show logarithmic spacing? Although this is still an open empirical question, qualitative findings are consistent with logarithmic spacing. Logarithmic spacing would imply that the width of time fields should increase linearly with the time of peak firing within the delay. Although a precisely linear relationship has not been established, it is certain that the width of time fields increases with time of peak firing. This is true in the hippocampus \cite{MacDEtal11,SalzEtal16}, entorhinal cortex \cite{KrauEtal15}, striatum \cite{AkhlEtal16,MellEtal15,JinEtal09}, and medial prefrontal cortex \cite{TigaEtal16}. Similarly, logarithmic spacing implies that the number density of cells with a time field centered on time $\tau$ should go down like $\tau^{-1}$. While this quantitative relationship has not been established thus far, it is certain that the number density goes down with $\tau$. This qualitative pattern at least holds in all of the regions where time cells have been identified thus far. In addition to logarithmic spacing, the logic motivating the existence of a fovea would apply equally well to cognitive dimensions. This does not imply that foveal organization ought to be ubiquitous. We would expect the size of the fovea to be large when $c$ is small. The cost of choosing a small value of $c$ is that the range of scales that can be represented with a fixed $N$ goes down dramatically (Eq.~\ref{eq:deltai}). In domains where there is no natural upper limit to $x$, such as numerosity or (arguably) time, this concern may be quite serious. However, the cost of failing to represent arbitrarily large scales is perhaps not so high in cases where there is a natural upper bound on the function to be represented. In the case of vision, there is a natural upper bound to the value of eccentricity that could be observed. In the case of functions of numerosity, the subitizing range, the finding that small integers are represented with little error, may be analogous to the fovea. In the case of functions of time, the analog of a fovea would correspond to something like traditional notions of short-term memory. In any case, direct measurement of $c$, which can be estimated from the width of receptive fields as a function of their peak, would provide strong constraints on whether or not there ought to be a ``fovea'' for numerosity and/or time. \subsection{Constructing ratio scales for cognitive dimensions} Ultimately, one can understand the spacing of the retinal coordinate system, or other sensory domains, as the result of a developmental process that aligns receptors along the sensory organ. However, if these arguments apply as well to non-sensory dimensions such as number and time, then this naturally raises the question of how the brain constructs ``receptors'' for these entities. One recent hypothesis for constructing representations of time, space and number describes time cells---which have receptive fields for particular events in the past---as extracted from exponentially-decaying neurons with long time constants. The requirement for logarithmic neural scales amounts to the requirement that the time constants of exponentially-decaying cells should be organized such that ``adjacent'' cells, along some gradient, should have time constants in a constant ratio. There is now good evidence for long time constants in cortex both \emph{in vitro} \cite{EgorEtal02} and \emph{in vivo} \cite{LeitEtal16}. The long time constants can be implemented at the single cell level using known biophysical properties of cortical neurons \cite{TigaEtal15}. Noting that a set of exponentially-decaying cells encode the Laplace transform of the history, it has been proposed that time cells result from an approximate inversion of the Laplace transform \cite{ShanHowa13}. The inversion can be accomplished with, essentially, feedforward on-center/off-surround receptive fields. The mathematical basis of this approach is extremely powerful. It is straightfoward to show that the computational framework for a representation of time can be generalized to space and number \cite{HowaEtal14} or any variable for which the time derivative can be computed. Moreover, access to the Laplace domain means one can implement various computations, for instance implementing translation \cite{ShanEtal16} or comparison operators \cite{HowaEtal15}. To the extent that representations from different domains utilize the same scaling relationships, access to a set of canonical computations could lead to a general computational framework for cognition.
b219d0e4676240f880f1ec8d83cff1941ab1d7b6
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} Relativity theory, quantum theory, and information theory are fundamental blocks of theoretical physics \cite{peres2004}. The goal of theoretical physics is to describe and, to a certain extent, understand natural phenomena. Unfortunately, complex difficulties arise when one attempts to merge general relativity theory and quantum theory. For example, in classical mechanics it is often said that gravity is a purely geometric theory since the mass does not appear in the usual Newtonian equation of a particle trajectory \begin{equation} m\frac{d^{2}\vec{x}}{dt^{2}}=-m\vec{\nabla}_{\vec{x}}\Phi_{\text{gravity }\Leftrightarrow\frac{d^{2}\vec{x}}{dt^{2}}=\vec{g}\text{. \end{equation} This is a direct consequence of the equality of the gravitational and inertial masses. In quantum mechanics, the situation is rather different. As a matter of fact, the Schrodinger quantum-mechanical wave equation is given by \cite{sakurai} \begin{equation} \left[ -\frac{\hbar^{2}}{2m}\vec{\nabla}_{\vec{x}}^{2}+m\Phi_{\text{gravity }\right] \psi\left( \vec{x}\text{, }t\right) =i\hbar\frac{\partial }{\partial t}\psi\left( \vec{x}\text{, }t\right) \text{. \end{equation} The mass $m$ no longer cancels and, instead, it appears in the combination $\hbar/m$ (where $\hbar\overset{\text{def}}{=}h/2\pi$ and $h$ denotes the Planck constant). Therefore, in an instance where $\hbar$ appears, $m$ is also expected to appear \cite{colella1975}. It seems evident that there is the possibility that such difficulties are not simply technical and mathematical, but rather conceptual and fundamental. This viewpoint was recently presented in \cite{brukner14}, where the idea was advanced that \emph{quantum causality} might shed some light on foundational issues related to the general relativity-quantum mechanics problem. When describing natural phenomena at the quantum scale, the interaction between the mechanical object under investigation and the observer (or, observing equipment) is not negligible and cannot be predicted \cite{bohr35, bohr37, bohr50}. This fact leads to the impossibility of unambiguously distinguishing between the object and the measuring instruments. This, in turn, is logically incompatible with the classical notion of causality; the possibility of sharply distinguishing between the subject and the object is essential to the ideal of causality. In his attempt to bring consistency in science, Bohr proposed to replace the classical ideal of causality with a more general viewpoint termed \emph{complementarity}. Roughly speaking, anyone can understand that one cannot bow in front of somebody without showing one's back to somebody else. This oversimplified statement is behind one of the most revolutionary scientific concepts of the twentieth century, namely Bohr's complementarity principle \cite{wheeler63}. This principle is a key feature of quantum physics and represents the dichotomy between the corpuscular (particle) and ondulatory (wave) nature of mechanical objects (matter and light). Within this descriptive framework, particle and wave properties are symbolized by well-defined position and momentum, respectively \cite{vaccaro10}. In 1935, using the complementarity principle, Bohr criticized the Einstein, Podolsky, and Rosen conclusion according to which the quantum mechanical description of physical reality given by wave functions was not complete based on a line of reasoning relying on a thought experiment (gedankenexperiment) \cite{epr}. In addition to criticizing the Einstein-Podolsky-Rosen argumentation (EPR paradox), Bohr provided a different interpretation of the concept of locality \cite{bohr35}. The solution to the EPR paradox is due to John Bell's 1964 theorem \cite{bell64}. The notion of causality played a key-role in the EPR paradox, Bohr's complementarity principle, and Bell's theorem. In this article, we reexamine Gisin's 1991 original proof concerning the violation of Bell's inequality for any pure entangled state of two-particle systems \cite{gisin91}. Our investigation is motivated by didactic reasons and permits to straighten a few mathematical points in the original proof that in no way modify the physical content provided by Gisin's work. \section{Background} Einstein, Podolsky, and Rosen considered a composite quantum system consisting of two distant particles, with an entangled wave function $\psi$ given by \cite{epr}, \begin{equation} \psi=\delta\left( x_{1}-x_{2}-L\right) \delta\left( p_{1}+p_{2}\right) \text{.}\label{wave \end{equation} The following background discussion follows very closely the presentation presented by Asher Peres in Ref. \cite{peres95}. The quantity $\delta$ denotes a normalizable function with an arbitrarily high and narrow peak. The quantity $L$ is a large distance, much larger than the range of mutual interaction of particles $1$ and $2$. The physical meaning of the wave function in Eq. (\ref{wave}) is that the two particles have been prepared in such a way that their relative distance is arbitrarily close to $L$, and their total momentum is arbitrarily close to zero. Note that the operators $x_{1}-x_{2}$ and $p_{1}+p_{2}$ commute. In the state $\psi$, one knows nothing about the positions of the individual particles (we only know their distance from each other); and one knows nothing of their individual momenta (one only knows the total momentum). However, if one measures $x_{1}$, one shall be able to predict with certainty the value of $x_{2}$, without having in any way disturbed particle $2$. At this point, EPR argue that \textit{since at the time of measurement the two systems no longer interact, no real change can take place in the second system in consequence of anything that may be done to the first system}. Therefore, $x_{2}$ corresponds to an element of physical reality, as defined by EPR. On the other hand, if one prefers to perform a measurement of $p_{1}$ rather than $x_{1}$, one shall then be able to predict with certainty the value of $p_{2}$, again without having in any way disturbed particle $2$. Therefore, by the same argument as above, $p_{2}$ also corresponds to an element of reality. However, quantum mechanics precludes the simultaneous assignment of precise values to both $x_{2}$ and $p_{2}$, since these operators do not commute, and thus EPR are forced to conclude that the quantum mechanical description of physical reality given by wave functions is not complete. However, they prudently leave open the question of whether or not a complete description exists. Reality according to EPR can be described as follows: If, without in any way disturbing a system, one can predict with certainty the value of a physical quantity, then there exists an element of reality corresponding to this physical quantity. The EPR article was not wrong, but it had been written too early \cite{peres04}. The EPR argument did not take into account that the observer's information was localized, like any other physical object. Information is not just an abstract notion. It requires a physical carrier, and the latter is approximately localized. Let us recall that Einstein's locality principle asserts that events occurring in a given spacetime region are independent of external parameters that may be controlled, at the same moment, by agents located in distant spacetime regions. In quantum mechanics, one has to accept that a measurement on what seems to be a part of the system is to be considered as a measurement on the whole system. If one persists with keeping Einstein's locality principle, alternative theories incorporating such a principle lead to a testable inequality (Bell's inequality, \cite{bell64}) relation among suitable observables that are not in agreement with the predictions of quantum mechanics. Several violations of Bell's inequality have been experimentally verified \cite{peres95}. For this reason, despite the psychological uncomfortable situation, quantum mechanics has prevailed over alternative theories.\emph{ Dura lex sed lex}: this is the experimental verdict. Quantum mechanics predictions are incompatible with Bell's inequality. There is an experimentally verifiable difference between quantum mechanics and the alternative theories satisfying Einstein's locality principle. It is somewhat of ironic that Bell's theorem is the most profound discovery of science because it is not obeyed by experimental facts \cite{stapp75}. We remark that Bell's paper is not about quantum mechanics. Rather, it is a general proof, independent of any specific physical theory, that there is an upper limit to the correlation of distant events, if one solely assumes the validity of Einstein's principle of local causes. Specifically, Bell showed that in a theory in which parameters are added to determine the results of individual measurements, without changing the statistical predictions, there must be a mechanism whereby the setting of one measuring device can influence the reading of another instrument, however remote. Moreover, the signal involved must propagate instantaneously, so that such a theory could not be Lorenz invariant. In particular, for any nonfactorable quantum state $\psi$, it is possible to find pairs of observables whose correlations violate Bell's inequality (see Eq. (\ref{bell})). This means that, for such a state, quantum theory makes statistical predictions which are incompatible with the demand that the outcomes of experiments performed at a given location in space be independent of the arbitrary choice of other experiments that can be performed, simultaneously, at distant locations (this apparently reasonable demand, as stated earlier, is the principle of local causes, also called Einstein locality). Bell's theorem implies that quantum mechanics is incompatible with the view that physical observables possess pre-existing values independent of the measurement context. A hidden variable theory which would predict individual events must violate the canons of special relativity: there would be no covariant distinction between cause and effect. The EPR paradox is resolved in the way which Einstein would have liked least. \section{Reexamination} Here, we choose to present Bell's Theorem as presented by Gisin in \cite{gisin91} (for a later and stronger derivation by Gisin and Peres, we refer to \cite{gisin92}). In what follows, whenever helpful to the discussion, we shall be using the conventional Dirac bra-ket notation where a wave vector $\psi$ will be denoted as $\left\vert \psi\right\rangle $. Gisin's Theorem can be stated as follows \cite{gisin91}:\ Let $\psi \in\mathcal{H}_{1}\otimes\mathcal{H}_{2}$. If $\psi$ is entangled (i.e. $\psi$ is not a product), then $\psi$ violates Bell's inequality. In other words, there are projectors $a$, $a^{\prime}$, $b$, $b^{\prime}$, such tha \begin{equation} \left\vert P\left( a\text{, }b\right) -P\left( a\text{, }b^{\prime}\right) \right\vert +P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime}\right) >2\text{,} \label{bell \end{equation} wher \begin{equation} P\left( a\text{, }b\right) \overset{\text{def}}{=}\left\langle \left( 2a-1\right) \otimes\left( 2b-1\right) \right\rangle _{\psi}\text{.} \label{quantity \end{equation} The proof proceeds as follows. \begin{itemize} \item First, from the Schmidt Decomposition Theorem \cite{mosca}, if $\psi$ is a vector in a tensor product space $\mathcal{H}_{1}\otimes\mathcal{H}_{2}$, then there exists an orthonormal basis $\left\{ \varphi_{i}\right\} $ with $1\leq i\leq n_{\mathcal{H}_{1}}$ for $\mathcal{H}_{1}$ where $n_{\mathcal{H _{1}}\overset{\text{def}}{=}\dim_ \mathbb{C} }\mathcal{H}_{1}$, and an orthonormal basis $\left\{ \theta_{j}\right\} $ with $1\leq j\leq n_{\mathcal{H}_{2}}$ for $\mathcal{H}_{2}$ where $n_{\mathcal{H}_{2}}\overset{\text{def}}{=}\dim_ \mathbb{C} }\mathcal{H}_{2}$, and non-negative \emph{real} numbers $\left\{ c_{i}\right\} $ so tha \begin{equation} \psi=\sum_{k}c_{k}\varphi_{k}\otimes\theta_{k}\text{.} \label{state \end{equation} The coefficients $c_{k}$ are called Schmidt coefficients, and the number of terms in the sum in Eq. (\ref{state}) will be at most the minimum of $n_{\mathcal{H}_{1}}$ and $n_{\mathcal{H}_{2}}$. Without loss of generality, in order to have a non-product state of two-particle systems, assume that at least two coefficients are nonzero, $c_{1}\neq0\neq c_{2}$ while $c_{k}=0$ for any $k>2$. We point out that, in principle, coefficients $c_{k}$ with $k>2$ could also be nonzero. However, since the original Clauser-Horne-Shimony-Holt (CHSH, \cite{chsh}) version of Bell's inequality can be violated in any two dimensional Hilbert subspace with nonzero Schmidt coefficient \cite{bell64, prlgisin}, without loss of generality, one can restrict the discussion to the case in which $c_{k}=0$ for any $k>2$. For the sake of completeness, we point out that the projection of the joint state of n pairs of particles onto a subspace spanned by states having a common Schmidt coefficients is a key step in the so-called Schmidt projection method, a technique used to study entanglement concentration in any pure state of a bipartite system \cite{fuck01}. Furthermore, for a practical demonstration of the violation of the CHSH version of Bell's inequality in $2D$ subspaces of an higher-dimensional orbital angular momentum Hilbert space, we refer to \cite{fuck02}. \item Second, recall that for any separable quantum state, the measure of entanglement is zero. Furthermore, the behavior of entanglement remains unchanged under simple local transformations, i.e. local unitary transformations \cite{vedral}. A local unitary transformation simply represents a change of basis in which we consider the given entangled state. A change of basis should not change the amount of entanglement that is accessible to us, because at any given time we could just reverse the basis change. Therefore in both bases the entanglement should be the same. Given these remarks, for the sake of convenience and without changing the entanglement behavior of the state $\psi$, let us apply a local (product) unitary transformation $U_{1}\otimes U_{2}$ on $\psi$ such that \begin{equation} \psi\rightarrow\left( U_{1}\otimes U_{2}\right) \psi=\left( U_{1}\otimes U_{2}\right) \left( c_{1}\varphi_{1}\otimes\theta_{1}+c_{2}\varphi _{2}\otimes\theta_{2}\right) \text{, \end{equation} where \begin{equation} \left( U_{1}\otimes U_{2}\right) \left( c_{1}\varphi_{1}\otimes\theta _{1}+c_{2}\varphi_{2}\otimes\theta_{2}\right) =c_{1}U_{1}\varphi_{1}\otimes U_{2}\theta_{1}+c_{2}U_{1}\varphi_{2}\otimes U_{2}\theta_{2}=c_{1}\left\vert +-\right\rangle +c_{2}\left\vert -+\right\rangle \text{.} \label{sopra \end{equation} In the particular case of spin $1/2$ systems, we have that $\left\vert +\right\rangle $ and $\left\vert -\right\rangle $ in\ Eq. (\ref{sopra}) are given by, \begin{equation} \left\vert +\right\rangle \overset{\text{def}}{=}\binom{1}{0}\text{, and }\left\vert -\right\rangle \overset{\text{def}}{=}\binom{0}{1}\text{, \end{equation} respectively. \item Third, an arbitrary density matrix $\rho$ for a mixed state qubit can be written as \cite{nielsen2000, cafaro2012} \begin{equation} \rho\overset{\text{def}}{=}\frac{I+\vec{P}\cdot\vec{\sigma}}{2}\text{, \end{equation} where $\vec{P}\i \mathbb{R} ^{3}$ with $\left\Vert \vec{P}\right\Vert \leq1$ is the Bloch vector for the state $\rho$, $I$ is the $2\times2$ identity matrix and $\vec{\sigma}$ is the Pauli matrix vector given by \cite{nielsen2000, cafaro2010, cafaro2014} \begin{equation} \vec{\sigma}=\left[ \sigma_{x}\text{, }\sigma_{y}\text{, }\sigma_{z}\right] \overset{\text{def}}{\mathbf{=}}\left[ \left( \begin{array} [c]{cc 0 & 1\\ 1 & 0 \end{array} \right) \text{, }\left( \begin{array} [c]{cc 0 & -i\\ i & 0 \end{array} \right) \text{, }\left( \begin{array} [c]{cc 1 & 0\\ 0 & -1 \end{array} \right) \right] \text{.} \label{pauli \end{equation} A state $\rho$ is pure if and only if $\left\Vert \vec{P}\right\Vert =1$. A density matrix for a pure state can be described as $\rho=\left\vert \psi\right\rangle \left\langle \psi\right\vert $ where $\left\vert \psi\right\rangle \left\langle \psi\right\vert $ is an orthogonal projector since $\rho^{2}=\rho$ and $\rho^{\dagger}=\rho$ (the symbol $\dagger$ denotes the usual Hermitian conjugation operation in quantum mechanics). \end{itemize} Having considered the above mentioned three points, the idea is to check the correctness of the inequality in Eq. (\ref{bell}) for entangled states $\psi$ in Eq. (\ref{state}) and for a suitable choice of projectors $a$, $a^{\prime $, $b$, and $b^{\prime}$. Consider projectors $a$, $a^{\prime}$, $b$, $b^{\prime}$ defined as \begin{equation} a\overset{\text{def}}{=}\frac{1+\vec{a}\cdot\vec{\sigma}}{2}\text{, a^{\prime}\overset{\text{def}}{=}\frac{1+\vec{a}^{\prime}\cdot\vec{\sigma} {2}\text{, }b\overset{\text{def}}{=}\frac{1+\vec{b}\cdot\vec{\sigma} {2}\text{, }b^{\prime}\overset{\text{def}}{=}\frac{1+\vec{b}^{\prime}\cdot \vec{\sigma}}{2}\text{,} \label{projectors \end{equation} with $\vec{a}$, $\vec{a}^{\prime}$, $\vec{b}$, $\vec{b}^{\prime}\i \mathbb{R} ^{3}$, and \begin{equation} \left\Vert \vec{a}\right\Vert =\left\Vert \vec{a}^{\prime}\right\Vert =\left\Vert \vec{b}\right\Vert =\left\Vert \vec{b}^{\prime}\right\Vert =1\text{, \end{equation} so that $a^{2}=a$, $a^{\prime2}=a^{\prime}$, $b^{2}=b$, $b^{\prime2 =b^{\prime}$. From Eq. (\ref{projectors}), $P\left( a\text{, }b\right) $ in Eq. (\ref{quantity}) become \begin{equation} P\left( a\text{, }b\right) \overset{\text{def}}{=}\left\langle \left( 2a-1\right) \otimes\left( 2b-1\right) \right\rangle _{\psi}=\left\langle \left( \vec{a}\cdot\vec{\sigma}\right) \otimes\left( \vec{b}\cdot \vec{\sigma}\right) \right\rangle _{\psi}\text{, \end{equation} with $\left\vert \psi\right\rangle \overset{\text{def}}{=}c_{1}\left\vert +-\right\rangle +c_{2}\left\vert -+\right\rangle $ and $c_{1}$, $c_{2}\i \mathbb{R} \backslash\left\{ 0\right\} $. Let us compute the explicit expression of $P\left( a\text{, }b\right) $. Using Eq. (\ref{pauli}), we\textbf{ }obtai \begin{align} \left( \vec{a}\cdot\vec{\sigma}\right) \otimes\left( \vec{b}\cdot \vec{\sigma}\right) & =\left( \begin{array} [c]{cc a_{z} & a_{x}-ia_{y}\\ a_{x}+ia_{y} & -a_{z \end{array} \right) \otimes\left( \begin{array} [c]{cc \ b_{z} & b_{x}-ib_{y}\\ b_{x}+ib_{y} & -b_{z \end{array} \right) \nonumber\\ & \nonumber\\ & =\left( \begin{array} [c]{cccc a_{z}b_{z} & a_{z}\left( b_{x}-ib_{y}\right) & b_{z}\left( a_{x -ia_{y}\right) & \left( a_{x}-ia_{y}\right) \left( b_{x}-ib_{y}\right) \\ a_{z}\left( b_{x}+ib_{y}\right) & -a_{z}b_{z} & \left( a_{x}-ia_{y}\right) \left( b_{x}+ib_{y}\right) & -b_{z}\left( a_{x}-ia_{y}\right) \\ b_{z}\left( a_{x}+ia_{y}\right) & \left( a_{x}+ia_{y}\right) \left( b_{x}-ib_{y}\right) & -a_{z}b_{z} & -a_{z}\left( b_{x}-ib_{y}\right) \\ \left( a_{x}+ia_{y}\right) \left( b_{x}+ib_{y}\right) & -b_{z}\left( a_{x}+ia_{y}\right) & -a_{z}\left( b_{x}+ib_{y}\right) & a_{z}b_{z \end{array} \right) \text{. \end{align} Therefore the quantity $P\left( a\text{, }b\right) $ becomes \begin{align} P\left( a\text{, }b\right) & =\left\langle \left( \vec{a}\cdot\vec {\sigma}\right) \otimes\left( \vec{b}\cdot\vec{\sigma}\right) \right\rangle _{\psi}=\left\langle \psi\left\vert \left( \vec{a}\cdot\vec{\sigma}\right) \otimes\left( \vec{b}\cdot\vec{\sigma}\right) \right\vert \psi\right\rangle \nonumber\\ & \nonumber\\ & =\left( \begin{array} [c]{cccc 0 & c_{1} & c_{2} & 0 \end{array} \right) \cdot\nonumber\\ & \nonumber\\ & \cdot\left( \begin{array} [c]{cccc a_{z}b_{z} & a_{z}\left( b_{x}-ib_{y}\right) & b_{z}\left( a_{x -ia_{y}\right) & \left( a_{x}-ia_{y}\right) \left( b_{x}-ib_{y}\right) \\ a_{z}\left( b_{x}+ib_{y}\right) & -a_{z}b_{z} & \left( a_{x}-ia_{y}\right) \left( b_{x}+ib_{y}\right) & -b_{z}\left( a_{x}-ia_{y}\right) \\ b_{z}\left( a_{x}+ia_{y}\right) & \left( a_{x}+ia_{y}\right) \left( b_{x}-ib_{y}\right) & -a_{z}b_{z} & -a_{z}\left( b_{x}-ib_{y}\right) \\ \left( a_{x}+ia_{y}\right) \left( b_{x}+ib_{y}\right) & -b_{z}\left( a_{x}+ia_{y}\right) & -a_{z}\left( b_{x}+ib_{y}\right) & a_{z}b_{z \end{array} \right) \cdot\nonumber\\ & \nonumber\\ & \cdot\left( \begin{array} [c]{c 0\\ c_{1}\\ c_{2}\\ 0 \end{array} \right) \nonumber\\ & \nonumber\\ & =-\left( c_{1}^{2}+c_{2}^{2}\right) a_{z}b_{z}+c_{1}c_{2}\left( 2a_{x}b_{x}+2a_{y}b_{y}\right) \text{, \end{align} that is \begin{equation} P\left( a\text{, }b\right) =-\left( c_{1}^{2}+c_{2}^{2}\right) a_{z b_{z}+c_{1}c_{2}\left( 2a_{x}b_{x}+2a_{y}b_{y}\right) \text{. \end{equation} Normalization of the wave function demands $\left\langle \psi\left\vert \psi\right. \right\rangle =1$. Therefore, we have $c_{1}^{2}+c_{2}^{2}=1$ and \begin{equation} P\left( a\text{, }b\right) =+2c_{1}c_{2}\left( a_{x}b_{x}+a_{y b_{y}\right) -a_{z}b_{z}\text{.} \label{first correction \end{equation} This simple mathematical computation leading to Eq. (\ref{first correction}) leads to rectify the incorrect sign that appears in \cite{gisin91}. Following Gisin, we assume that convenient expressions for the Bloch vectors are given by \begin{align} \vec{a} & =\left( a_{x}\text{, }a_{y}\text{, }a_{z}\right) =\left( \sin\alpha\text{, }0\text{, }\cos\alpha\right) \text{,}\nonumber\\ \vec{b} & =\left( b_{x}\text{, }b_{y}\text{, }b_{z}\right) =\left( \sin\beta\text{, }0\text{, }\cos\beta\right) \text{,} \label{vector \end{align} with $\alpha=0$, $\alpha^{\prime}=\pm\pi/2$ where the sign is the \emph{same} as that of $c_{1}c_{2}$ (note that due to the previous sign mistake, in \cite{gisin91} it is reported that $\alpha^{\prime}=\pm\pi/2$ where the sign is the \emph{opposite} of that of $c_{1}c_{2}$). For possible alternative parametrizations of the Bloch vectors, we refer to Appendix A. Let us now consider the expression given by $\left\vert P\left( a\text{, }b\right) -P\left( a\text{, }b^{\prime}\right) \right\vert +P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime}\right) $. Using Eqs. (\ref{quantity}), (\ref{projectors}), and (\ref{vector}), we obtai \begin{equation} \left\vert P\left( a\text{, }b\right) -P\left( a\text{, }b^{\prime}\right) \right\vert =\left\vert -\cos\beta+\cos\beta^{\prime}\right\vert =\left\vert -\left( \cos\beta-\cos\beta^{\prime}\right) \right\vert =\left\vert \cos\beta-\cos\beta^{\prime}\right\vert \text{. \end{equation} Furthermore, using the same line of reasoning, $P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime}\right) $ become \begin{align} P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime }\right) & =\left( 2c_{1}c_{2}\sin\alpha^{\prime}\sin\beta-\cos \alpha^{\prime}\cos\beta\right) +\left( 2c_{1}c_{2}\sin\alpha^{\prime \sin\beta^{\prime}-\cos\alpha^{\prime}\cos\beta^{\prime}\right) \nonumber\\ & \nonumber\\ & =2c_{1}c_{2}\sin\alpha^{\prime}\left( \sin\beta+\sin\beta^{\prime}\right) -\cos\alpha^{\prime}\left( \cos\beta+\cos\beta^{\prime}\right) \text{, \end{align} that is \begin{equation} P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime }\right) =2c_{1}c_{2}\sin\alpha^{\prime}\left( \sin\beta+\sin\beta^{\prime }\right) -\cos\alpha^{\prime}\left( \cos\beta+\cos\beta^{\prime}\right) \text{. \end{equation} Assuming $c_{1}c_{2}>0$ and $\alpha^{\prime}=+\pi/2$, yield \begin{equation} P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime }\right) =2\left\vert c_{1}c_{2}\right\vert \left( \sin\beta+\sin \beta^{\prime}\right) \text{. \end{equation} Moreover, assuming $c_{1}c_{2}<0$ and $\alpha^{\prime}=-\pi/2$, we ge \begin{equation} P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime }\right) =-2c_{1}c_{2}\left( \sin\beta+\sin\beta^{\prime}\right) \text{, \end{equation} with $-2c_{1}c_{2}>0$. Therefore $2\left\vert c_{1}c_{2}\right\vert =-2c_{1}c_{2}>0$, an \begin{equation} P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime }\right) =2\left\vert c_{1}c_{2}\right\vert \left( \sin\beta+\sin \beta^{\prime}\right) \text{. \end{equation} In summary, the expression for $\left\vert P\left( a\text{, }b\right) -P\left( a\text{, }b^{\prime}\right) \right\vert +P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime}\right) $ is given by \begin{equation} \left\vert P\left( a\text{, }b\right) -P\left( a\text{, }b^{\prime}\right) \right\vert +P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime}\right) =\left\vert \cos\beta-\cos\beta^{\prime}\right\vert +2\left\vert c_{1}c_{2}\right\vert \left( \sin\beta+\sin\beta^{\prime }\right) \text{. \end{equation} At this point, choose $\beta$ and $\beta^{\prime}$ such that $\sin\beta>0$ and $\sin\beta^{\prime}>0$ so that $2\left\vert c_{1}c_{2}\right\vert \left( \sin\beta+\sin\beta^{\prime}\right) >0$. Take also $-\cos\beta^{\prime =\cos\beta=\left( 1+4\left\vert c_{1}c_{2}\right\vert \right) ^{-1/2}$. We finally obtain \begin{equation} \left\vert P\left( a\text{, }b\right) -P\left( a\text{, }b^{\prime}\right) \right\vert +P\left( a^{\prime}\text{, }b\right) +P\left( a^{\prime}\text{, }b^{\prime}\right) \geq2\left( 1+4\left\vert c_{1}c_{2}\right\vert \right) ^{-1/2}>2\text{. \end{equation} This concludes the proof. \section{Conclusions} In this article, we presented a simple and explicit reexamination of Gisin's 1991 original proof concerning the violation of Bell's inequality for any entangled state of two-particle systems. Gisin's original work on Bell's inequality was presented in a very synthetic and intuitive manner. After so many years, Gisin's work continues to be highly regarded also in pure research \cite{rudolph}. For this reason, we have considered it especially helpful for didactic purposes. For a different alternative pedagogical presentation of Bell's inequality, we refer to \cite{selleri, maccone13}. We remark that we have simply expanded the original proof and in doing so, we have rectified with Eq. (\ref{first correction}) an incorrect mathematical sign in the expression of $P\left( a\text{, }b\right) $ in \cite{gisin91} that now becomes \begin{equation} P\left( a\text{, }b\right) =+2c_{1}c_{2}\left( a_{x}b_{x}+a_{y b_{y}\right) -a_{z}b_{z}\text{.} \label{pab \end{equation} The incorrectness of the sign in $P\left( a\text{, }b\right) $ as reported in \cite{gisin91} generates and propagates an incorrect statement. As a consequence of Eq. (\ref{pab}), we have also corrected such a subsequent statement. Namely, the correct statement become \begin{equation} \alpha^{\prime}=\pm\pi/2\text{ where the sign is the \emph{same }as that of }c_{1}c_{2}\text{.} \label{fu \end{equation} Notwithstanding these two corrections presented in Eqs. (\ref{pab}) and (\ref{fu}), the physical conclusions provided by Gisin remain unaltered. \begin{acknowledgments} C. Cafaro thanks Nicolas Gisin for his kindness in providing useful correspondence and for pointing out reference \cite{gisin92}. C. Cafaro is also grateful to Ariel Caticha for helpful discussions. \end{acknowledgments} \bigskip
cdd2136e2fe55783e95f9757d2e68d2dd56b0897
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }
\section{Introduction} The handwritten signature is a widely accepted means of authentication in government, legal, and commercial transactions. Signature verification systems aim to confirm the identity of a person based on their signature \cite{jain_introduction_2004}, that is, they classify signature samples as ``genuine'' (created by the claimed individual) or ``forgery'' (created by an impostor). In offline (static) signature verification, the signatures are acquired after the signature writing process is completed, by scanning a document containing the signature. This is in contrast with online (dynamic) signature verification, where the signature is captured directly on a device (such as a pen tablet), and therefore the dynamic information of the signature is available, such as the velocity of the pen movements. The lack of dynamic information in the offline case makes it a challenging problem, and much of the effort in this field has been devoted to obtaining a good feature representation for signatures \cite{hafemann_offline_2015}. There are two main approaches for the problem in the literature: in a Writer-Dependent approach, for each user, a training set of genuine signatures of the user (and, often, genuine signatures from other users as negative samples) is used to train a binary classifier. In a Writer-Independent approach, a single global classifier is trained using a dissimilarity approach, by using a training set consisted of positive samples (a difference vector between two genuine signatures from the same author), and negative samples (a difference vector between a genuine signature and a forgery). At test time, a distance vector is calculated between the query signature and one or more reference signatures (known to be genuine signatures of the claimed individual), and then classified using the Writer-Independent classifier. For a comprehensive review on the problem, refer to \cite{impedovo_automatic_2008}, and for a recent review, see \cite{hafemann_offline_2015}. Recent work on the area explore a variety of different feature descriptors: Extended Shadow Code (ESC) and Directional-Probabilistic Density Function (DPDF) \cite{rivard_multi-feature_2013}, \cite{eskander_hybrid_2013}; Local Binary Patterns (LBP), Gray-Level Co-occurrence Matrix (GLCM) and Histogram of Oriented Gradients (HOG) \cite{yilmaz_offline_2015}, \cite{hu_offline_2013}; Curvelet transform \cite{guerbai_effective_2015}, among others. Instead of relying on hand-engineered feature extractors, we investigate feature learning algorithms applied to this task. In previous research \cite{hafemann_ijcnn_2016}, we have shown that we can learn useful features for offline signature verification, by learning Writer-Independent features from a development dataset (a set of users not enrolled in the system). Using this formulation, we obtained results close to the state-of-the-art in the GPDS dataset, when compared against models that rely on a single feature extractor technique. In the same vein, Ribeiro et al. \cite{ribeiro_deep_2011} used Restricted Boltzmann Machines (RBMs) for learning features from signature images. However, the authors considered only a small set of users (10), and did not use the features to actually classify signatures, only reporting a visual representation of the learned weights. Khalajzadeh \cite{khalajzadeh_persian_2012} used Convolutional Neural Networks for Persian signature verification, but did not considered skilled forgeries. In this paper, we further analyze the method introduced in \cite{hafemann_ijcnn_2016}, investigating the impact of the depth (number of layers) of the Deep Convolutional Neural Network (CNN), and the size of the embedding layer on learning good representations for signatures, as measured by the classification performance on a different set of users. Using better training techniques (in particular, using Batch Normalization \cite{ioffe2015batch} for training the network), we can improve performance significantly, achieving a state-of-the-art performance of 2.74\% Equal Error Rate (EER) on the GPDS-160 dataset, which surpasses all results in the literature, even comparing to results where model ensembles are used. We also perform an analysis on the errors committed by the model, and visualize the quality of the learned features with a 2D projection of the embedding space. Visual analysis suggests that the learned features capture the overall aspect of the signature, which is sufficient to perform well in separating genuine signatures from skilled forgeries in some cases: for users that have complex signatures, or that maintain very stable signatures (i.e. different genuine samples are very similar). We also notice that the learned features are particularly vulnerable to slowly-traced forgeries, where the overall signature shape is similar, but the line quality is poor. \section{Methodology} The central idea is to learn a feature representation (a function $\phi(.)$) for offline signature verification in a Writer-Independent format, use this function to extract features from signatures $\textbf{X}$ of the users enrolled in the system ($\phi(\textbf{X})$), and use the resulting feature vectors to train a binary classifier for each user. The rationale for learning the feature representation in a Writer-Independent format is two-fold: 1) learning feature representations directly for each user is impractical, given the low number of samples available for training (around 5-10 signatures); and 2) having a fixed representation useful for any user makes it straightforward to add new users to the system, by simply using the model to extract features for the new user's genuine signatures, and training a Writer-Dependent classifier. In order to learn features in a Writer-Independent format, we consider two separate sets of signatures: a development set $\mathcal{D}$, that contains signatures from users not enrolled in the system, and an exploitation set $\mathcal{E}$ that contains a disjoint set of users. The signatures from set $\mathcal{D}$ are only used for learning a feature representation for signatures, with the hypothesis that the features learned for this set of users will be relevant to discriminate signatures from other users. On the other hand, the set $\mathcal{E}$ represents the users enrolled to the system. We use the model trained on the set $\mathcal{D}$ to ``extract features'' for the signatures of these users, and train a binary classifier for each user. It is worth noting that we do not use skilled forgeries during training, since it is not practical to require forgeries for each new user enrolled in the system. Overall, the method consists in the following steps: \begin{itemize} \item Training a deep neural network on a Development set \item Using this network to obtain a new representation for signatures on $\mathcal{E}$ (i.e. obtain $\phi(X)$ for all signatures $X$) \item Training Writer-Dependent classifiers in the exploitation set, using the learned representation \end{itemize} Details of these steps are presented below. \subsection{Convolutional Neural Network training} As in \cite{hafemann_ijcnn_2016}, we learn a function $\phi(.)$ by training a Deep Convolutional Neural Network on a Development set, by learning to discriminate between different users. That is, we model the network to output $M$ units, that estimate $P(y|X)$ where $y$ is one of the $M$ users in the set $\mathcal{D}$, and $X$ is a signature image. In this work we investigate different architectures for learning feature representations. In particular, we evaluate the impact of depth, and the impact of the size of the embedding layer (the layer from which we obtain the representation of the signature). Multiple studies suggest that depth is important to address complex learning problems, with both theoretical arguments \cite{bengio_learning_2009}, and as shown empirically (e.g. \cite{simonyan2014very}). The size of the embedding layer is an important factor for practical considerations, since this is the size of the feature vectors that will be used for training classifiers for each new user of the system, as well as performing the final classification. In our experiments, we had difficulty to train deep networks, with more than 4 convolutional layers and 2 fully-connected layers, even using good initialization strategies such as recommended in \cite{glorot_understanding_2010}. Surprisingly, the issue was not that the network overfit the training set, but rather both the training and validation losses remained high, suggesting issues in the optimization problem. To address this issue, we used Batch Normalization \cite{ioffe2015batch}. This technique consists in normalizing the outputs of a layer, for each individual neuron, so that the values have zero mean and unit variance (in a mini-batch of training data), and showed to be fundamental in our experiments in obtaining good performance. Details of this technique can be found in \cite{ioffe2015batch}. In the experiments for this paper, we only report the results using Batch Normalization, since for most of the proposed architectures, we could not train the network without this technique (performance on a validation set remained the same as random chance). \begin{table} \ra{1.3} \centering \caption{CNN architectures evaluated in this paper} \label{tbl:architectures} \begin{tabular}{|c|c|c|c|} \hline AlexNet\_\textbf{N}\textsubscript{reduced}&AlexNet\_\textbf{N}& VGG\_\textbf{N}\textsubscript{reduced}&VGG\_\textbf{N} \\ \hline conv11-96-s4-p0 & conv11-96-s4-p0 & conv3-64 & conv3-64 \\ & & conv3-64 & conv3-64 \\ \hline pool3-s2-p0 & pool3-s2-p0 & pool3 & pool3 \\ \hline conv5-256-p2 & conv5-256-p2 & conv3-128 & conv3-128\\ & & conv3-128 & conv3-128 \\ \hline pool3-s2-p0 & pool3-s2-p0 & pool4 & pool3 \\ \hline conv3-384 & conv3-384 & conv3-256 & conv3-256 \\ conv3-256 & conv3-384 & conv3-256 & conv3-256 \\ & conv3-256 & conv3-256 & conv3-256 \\ & & conv3-256 & conv3-256 \\ \hline pool3-s2-p0 & pool3-s2-p0 & pool4 & pool3 \\ \hline & & & conv3-256 \\ & & & conv3-256 \\ & & & conv3-256 \\ & & & conv3-256 \\ \cline{4-4} & & & pool2 \\ \cline{4-4} & & & conv3-256 \\ & & & conv3-256 \\ & & & conv3-256 \\ & & & conv3-256 \\ \cline{4-4} & & & pool2 \\ \cline{4-4} & \textbf{FC1-N} & & \textbf{FC1-N} \\ \cline{2-2} \cline{4-4} \textbf{FC1-N} &\textbf{FC2-N} &\textbf{FC1-N} &\textbf{FC2-N} \\ \hline \multicolumn{4}{|c|}{FC-531 + softmax} \\ \hline \end{tabular} \end{table} Table \ref{tbl:architectures} shows the architectures we evaluated in this research. The models are based on AlexNet \cite{krizhevsky_imagenet_2012} and VGG \cite{simonyan2014very}, which are two architectures that perform remarkably well in Computer Vision problems. Each row specifies a layer of the network. For convolutional layers, we include first the filter-size, followed by number of convolutional filters. For pooling layers, we include the size of the pooling region. Unless otherwise indicated, we use stride 1 and padding 1. For instance, conv11-96-s4-p0 refers to a convolutional layer with 96 filters of size 11x11, with stride 4 and no padding. For the Fully Connected (FC) layers, we just indicate the name of the layer (that we refer later in the text), and the number of output units. For instance, FC1-N refers to a Fully connected layer named ``FC1'', with $N$ output units. The network AlexNet\_N\textsubscript{reduced} contains 9 layers: 6 layers with learnable parameters (convolutions and fully connected layers), plus 3 pooling layers, where the VGG\_N network contains 24 layers: 19 layers with learnable parameters plus 5 pooling layers. We used Batch Normalization for each learnable layer in the network, before applying the non-linearity (ReLU, except for the last layer, which uses softmax). We considered 4 types of networks: AlexNet: a network similar to the network proposed in \cite{krizhevsky_imagenet_2012}, but without Local Response Normalization and Dropout (which we found unnecessary when using Batch Normalization); AlexNet\textsubscript{reduced}: a similar network but with reduced number of layers; VGG: An architecture similar to \cite{simonyan2014very}; VGG\textsubscript{reduced}: a similar network but with reduced number of layers. For each network, we consider a different size $N$ for the embedding layer, $N \in \{512, 1024, 2048, 4096\}$. This is the size of the feature vectors that will be used for training the Writer-Dependent classifiers, and therefore they impact the cost of training classifiers for new users that are enrolled to the system, as well as the classification cost. Considering the different architectures, and size of the embedding layer, we tested in total 16 architectures. The images were pre-processed by: centralizing the signatures in a 840x1360 image according to their center of mass; removed the background using OTSU's algorithm; inverted the images so that the background pixels were zero valued; and finally resizing them to 170x242 pixels. More details on these preprocessing steps can be found in \cite{hafemann_ijcnn_2016}. \subsection{Training the Writer-Dependent classifiers} After training a Deep CNN on the development set $\mathcal{D}$, we use the network to extract feature representations for signatures from another set of users ($\mathcal{E}$) and train Writer-Dependent classifiers. To do so, we resize the images to 170x242 pixels, perform feedforward propagation until a fully-connect layer, and used the activations at that layer as the feature vector for the image. For the architectures marked as ``reduced'' (see table \ref{tbl:architectures}), that contained only two fully-connected layers, we consider only the last layer before softmax as the embedding layer (FC1). For the architectures with three fully-connected layers, we consider both layers FC1 and FC2, that is, the last two fully-connected layers before the softmax layer. For each user in the set $\mathcal{E}$, we build a training set consisted of $r$ genuine signatures from the user as positive samples, and 14 genuine signatures from users in $\mathcal{D}$ as negative samples. We train a Support Vector Machine (SVM) classifier on this dataset, considering both linear SVM, and SVM with an RBF kernel. Similarly to \cite{eskander_hybrid_2013}, we use different weights for the positive and negative class to account for the imbalance of having many more negative samples than positive. We do so by duplicating the positive samples, so that the dataset is roughly balanced. For testing, we use 10 genuine signatures from the user (not used for training), 10 random forgeries (signatures from other users in $\mathcal{E}$), and the 30 skilled forgeries made targeting the user's signature. \section{Experimental Protocol} We conducted experiments in the GPDS-960 dataset \cite{vargas_off-line_2007}, which is the largest publicly available dataset for offline signature verification. The dataset consists of 881 users, with 24 genuine samples per user, and 30 skilled forgeries. We divide the dataset into an exploitation set $\mathcal{E}$ consisting of the first 300 users, and a development set consisting of the remaining 581 users. Since in this work we evaluate many different architectures, we split the Development set into a training and validation set, using a disjoint set of users for each set. We use a subset of 531 users for training the CNN models, and a subset of 50 users for validation. We use the same protocol above to train Writer-Dependent classifiers for the validation set, to obtain an estimate of how well the features learned by the models generalize to other users. Finally, we use the best models to train WD classifiers for users in the exploitation set $\mathcal{E}$. We performed 10 runs, each time training a network with a different initialization, and using a random split of the genuine signatures used for training/testing, when training the Writer-Dependent classifiers. We report the mean and standard deviation of the errors across the 10 runs. To compare with previous work, we use as exploitation set the first 160 users (to compare with work that used the dataset GPDS-160), and the first 300 users (to compare with work that used GPDS-300). To assess the impact of the number of genuine signatures for each user in training, we ran experiments with 5 and 14 genuine signatures for training ($r \in \{5,14\}$). We evaluated the results in terms of False Rejection Rate (FRR - the fraction of genuine signatures misclassified as forgery), False Acceptance Rate (FAR - the fraction of forgeries misclassified as genuine signatures). For completeness, we show the FAR results for both random and skilled forgeries. We also report Equal Error Rates (EER), which is the error when FAR = FRR. In this case, we use only genuine signatures and skilled forgeries, to enable comparison with other work in the literature. We considered two forms of calculating the EER: EER\textsubscript{user-thresholds}: using user-specific decision thresholds; and EER\textsubscript{global threshold}: using a global decision threshold. Lastly, we also report the Mean Area Under the Curve (Mean AUC), by averaging the AUCs of Receiving Operating Curves (ROC curves) calculated for each user. For calculating FAR and FRR in the exploitation set, we used a decision threshold selected from the validation set (the threshold that achieved EER using a global decision threshold). \section{Results} \begin{figure} \centering \includegraphics[width=\columnwidth]{images/varying_n_linear} \caption{Equal Error Rates on the validation set, for WD models trained with a \textbf{Linear SVM}, using the representation space learned by different architectures (at the layer indicated in parenthesis), and different representation sizes (N)} \label{fig:varying_N_linear} \end{figure} \begin{figure} \centering \includegraphics[width=\columnwidth]{images/varying_n_rbf} \caption{Equal Error Rates on the validation set, for WD models trained with an SVM with \textbf{RBF kernel}, using the representation space learned by different architectures (at the layer indicated in parenthesis), and different representation sizes (N)} \label{fig:varying_N_rbf} \end{figure} We first consider the results of the experiments in the validation set, varying the depth of the networks, and the size of the embedding layer. In these experiments, we trained the CNN architectures defined in table \ref{tbl:architectures}, used them to extract features for the users in the validation set, and trained Writer-Dependent classifiers for these users, using 14 reference signatures. We then analyzed the impact in classification performance of the different architectures/sizes of the embedding layer, measuring the average Equal Error Rate of the classifiers. Figures \ref{fig:varying_N_linear} and \ref{fig:varying_N_rbf} show the classification results on the validation set using Linear SVMs and SVMs with an RBF kernel, respectively. The first thing we notice is that, contrary to empirical results in object recognition (ImageNet), we did not observe improved performance with very deep architectures. The best performing models were the AlexNet architecture (with 8 trainable layers) and the AlexNet\textsubscript{reduced} (with 6 trainable layers) when using the features to training linear SVMs and SVMs with RBF kernel, respectively. We also notice that the performance with a linear classifier is already quite good, demonstrating that the feature representations learned in a writer-independent way seems to generalize well to new users. \begin{figure} \centering \includegraphics[width=\columnwidth]{images/comparison_freq_euclid} \caption{Cumulative frequencies of pairs of signatures within a given distance. For the genuine-forgery pairs, the y axis shows the fraction of pairs that are \textit{closer} than a given euclidean distance. For genuine-genuine pairs the y axis shows the fraction of pairs that are \textit{further} than a given distance, to show the overlap with the genuine-forgery pairs.} \label{fig:distances} \end{figure} \begin{figure} \centering \includegraphics[scale=0.35]{images/50_genuine} \caption{2D t-SNE projection of the embedding space $\phi(.)$ for the genuines signatures of the 50 authors in the validation set. Each point refers to one signature, and different users are shown in different colors} \label{fig:tsne_genuine} \end{figure} \begin{figure} \centering \includegraphics[width=\columnwidth]{images/50_genuine_and_forg} \caption{2D t-SNE projection of the embedding space $\phi(.)$ for the genuine signatures and skilled forgeries of the 50 authors in the validation set. Points that refer to skilled forgeries are shown with a black border. The zoomed area on the left shows a region where three skilled forgeries (on the top-left) are close to four genuine signatures for user 394 in the GPDS dataset. The zoomed area on the right shows three skilled forgeries for the same user that are far from the genuine signatures in the embedding space} \label{fig:tsne_genuine_fogery} \end{figure} Another aspect that we investigated is how the signatures from the users in the validation set are represented in the embedding layer of the CNN. For this analysis, we took one of the models (AlexNet\_2048, at layer FC2), obtained the representations for signatures of all 50 users in the validation set, and analyzed this representation. We started by analyzing how well separated are the signatures from each user in this feature space, as well as compared to skilled forgeries targeted to the users. To do so, we measured the euclidean distance between genuine signatures of the same user, genuine signatures from different users (i.e. a genuine signature vs. a random forgery), and the distance between pairs of a genuine signature and a skilled forgery made for this user. Figure \ref{fig:distances} plots the cumulative frequency of pairs of signatures within a given distance, that is, the fraction of pairs that are closer than a given euclidean distance in the embedding space. For the genuine-genuine pairs, we show the inverse of the cumulative frequency (i.e. the fraction of pairs further than a given distance), to better show the overlap between the curves. For a particular distance, the y axis is the expected False Rejection Rate (on the genuine-genuine pairs) and False Acceptance Rate (on the genuine-forgery pairs) if we used a distance-based classifier, with a single genuine signature as reference. We can see that in the space projected by the CNN, the genuine signatures from different users very well separated: there is a very small overlap between the curves of genuine-genuine pairs and genuine-random forgery pairs. A similar behavior can be seem with the genuine-skilled forgeries pairs, although the curves overlap more in this case. To further analyze how the signatures are dispersed in this representation space, we used the t-SNE algorithm \cite{van2008visualizing} to project the samples from $\mathbb{R}^{2048}$ to $\mathbb{R}^{2}$. This allows us to inspect the local structure present in this higher dimensionality representation. Figure \ref{fig:tsne_genuine} shows the result of this projection of the genuine signatures of the 50 users in the validation set. Each point is a signature, and each color represent a different user. We can see that the signatures from different users are very well separated in this representation, even though no samples from these authors were used to train the CNN, which suggests that the learned representation is not specific to the writers in the training set, but generalize to other users. Figure \ref{fig:tsne_genuine_fogery} shows the projection of both the genuine signatures and skilled forgeries for these 50 users. We can see that the skilled forgeries are more scattered around, and for several users, the skilled forgeries are close to the representation of the genuine signatures. \begin{table*} \centering \caption{Detailed performance of the WD classifiers on the GPDS dataset (Errors and Standard deviations in \%)} \label{table:results} \resizebox{\textwidth}{!}{% \begin{tabular}{llllrrrrrr} \toprule Dataset& \begin{tabular}[x]{@{}c@{}}\#samples\\per user\end{tabular} & Model & Classifier & FRR & FAR\textsubscript{random} & FAR\textsubscript{skilled} & EER\textsubscript{global threshold} & EER\textsubscript{user-thresholds} & Mean AUC \\ \midrule GPDS-160 & 5 & AlexNet\_2048 & Linear SVM & 32.12 (+-1.21) & 0.01 (+-0.02) & 0.89 (+-0.13) & 8.24 (+-0.54) & 4.25 (+-0.37) & 0.9845 (+-0.0018) \\ & & AlexNet\_2048\textsubscript{reduced} & SVM (RBF) & 26.56 (+-1.01) & 0.00 (+-0.00) & 1.03 (+-0.14)& 6.99 (+-0.26) & 3.83 (+-0.33) & 0.9861 (+-0.0015) \\ & 14 & AlexNet\_2048 & Linear SVM & 10.41 (+-0.72) & 0.01 (+-0.02) & 2.77 (+-0.17) & 5.66 (+-0.22) & 3.37 (+-0.13) & 0.9894 (+-0.0008) \\ & & \textbf{AlexNet\_2048\textsubscript{reduced}} & \textbf{SVM (RBF)} & \textbf{6.75 (+-0.42)} & \textbf{0.00 (+-0.00)} & \textbf{3.46 (+-0.12)} & \textbf{4.85 (+-0.11)} & \textbf{2.74 (+-0.18)} & \textbf{0.9913 (+-0.0009)} \\ GPDS-300 & 5 & AlexNet\_2048 & Linear SVM & 31.48 (+-1.32) & 0.00 (+-0.00) & 1.87 (+-0.15) & 9.39 (+-0.44) & 5.41 (+-0.40) & 0.9760 (+-0.0017) \\ & & AlexNet\_2048\textsubscript{reduced} &SVM (RBF) & 26.33 (+-0.99) & 0.00 (+-0.00) & 1.74 (+-0.11) & 8.04 (+-0.20) & 4.53 (+-0.14) & 0.9817 (+-0.0008) \\ & 14 & AlexNet\_2048 &Linear SVM & 10.42 (+-0.58) & 0.00 (+-0.01) & 4.73 (+-0.19) & 7.00 (+-0.15) & 4.17 (+-0.25) & 0.9823 (+-0.0012) \\ & & \textbf{AlexNet\_2048\textsubscript{reduced}} &\textbf{SVM (RBF)} & \textbf{6.55 (+-0.25)} & \textbf{0.00 (+-0.01)} & \textbf{5.13 (+-0.20)} & \textbf{5.75 (+-0.19)}& \textbf{3.47 (+-0.16)} & \textbf{0.9871 (+-0.0007)}\\ \bottomrule \end{tabular} } \end{table*} \begin{table} \centering \caption{Comparison with state-of-the art on the GPDS dataset (errors and standard deviations in \%)} \label{table:soa_gpds} \resizebox{\columnwidth}{!}{% \begin{tabular}{llllr} \toprule Reference & Dataset& \begin{tabular}[x]{@{}c@{}}\#samples\\per user\end{tabular} &Features \& (Classifier) & EER\\ \midrule Vargas et al \cite{vargas_off-line_2010} &GPDS-100 & 5 &Wavelets (SVM) & 14.22 \\ Vargas et al \cite{vargas_off-line_2011} & GPDS-100 &10 &LBP, GLCM (SVM)& 9.02 \\ Hu and Chen \cite{hu_offline_2013}& GPDS-150 &10 &LBP, GLCM, HOG (Adaboost) & 7.66\\ Yilmaz \cite{yilmaz_offline_2015} & GPDS-160 &12&LBP (SVM) & 9.64\\ Yilmaz \cite{yilmaz_offline_2015} &GPDS-160 &12&LBP, HOG (Ensemble of SVMs)& 6.97\\ Hafemann et al \cite{hafemann_ijcnn_2016} & GPDS-160 & 14 & WI-learned with a CNN (SVM) & 10.70 \\ \midrule \textbf{Present work} &\textbf{GPDS-160} & \textbf{5} &\textbf{WI-learned with a CNN (SVM)}& \textbf{3.83 (+- 0.33)}\\ \textbf{Present work} &\textbf{GPDS-160} & \textbf{14} &\textbf{WI-learned with a CNN (SVM)}& \textbf{2.74 (+- 0.18)}\\ \textbf{Present work} &\textbf{GPDS-300} & \textbf{5} &\textbf{WI-learned with a CNN (SVM)}& \textbf{4.53 (+- 0.14)}\\ \textbf{Present work} &\textbf{GPDS-300} & \textbf{14} &\textbf{WI-learned with a CNN (SVM)}& \textbf{3.47 (+- 0.16)}\\ \bottomrule \end{tabular} } \end{table} \begin{figure} \centering \includegraphics[width=\columnwidth]{images/fine_comparison} \caption{Details of a genuine signature (top), and a skilled forgery (bottom) that are mapped close in the embedding space $\phi(.)$} \label{fig:closeup} \end{figure} The zoomed-in area in figure \ref{fig:tsne_genuine_fogery} shows skilled forgeries made for a particular user that were represented close to genuine signatures from this user. We can see that these forgeries have a close resemblance to the genuine signatures of the user in its overall shape. This is further examined in figure \ref{fig:closeup}, where we take a genuine signature and a skilled forgery from this user, that were close in the embedding space. We can notice that the overall shape of the skilled forgery is similar to the genuine signature, but looking in the details of the strokes we can see that the line quality of the skilled forgery is much worse. We have noticed the same behavior with other users / signatures. This suggests that the features learned by the network are useful in distinguishing the signatures in an ``overall shape'', but do not capture important properties of forgeries, such as the quality of the pen strokes, and therefore are particularly not discriminative to slowly-traced forgeries. Table \ref{table:results} shows the detailed result of the experiments in the exploitation set. We noticed that the performance was very good in terms of equal error rates, but that the global decision thresholds from the validation set are only effective in a few situations - in particular, when using the same number of samples as used in the validation set, where the threshold was selected. We also notice that the performance using a global threshold is significantly worse than using per-user thresholds. Note that the values of EER refer to the scenario where the decision threshold is optimally selected, which in itself is a hard task (given the low number of samples per user), and which we did not explore in this paper. Finally, table \ref{table:soa_gpds} compares our best results (using EER\textsubscript{user-thresholds}) with the state-of-the-art. We noticed a big drop in classification errors (as measured by Equal Error Rate) compared to the literature, even when using only 5 samples per user for the Writer-Dependent training. \section{Conclusions} In this work, we presented a detailed analysis of different architectures for learning representations for offline signature verification. We showed that features learned in a writer-independent format can be very effective for signature verification, achieving a large improvement of state-of-the-art performance in the GPDS-160 dataset, with 2.74\% Equal Error Rate compared to 6.97\% reported in the literature. We also showed that writer-dependent classifiers trained with these features can perform very well even with limited number of samples per user (e.g. 5 samples) and linear classifiers. Our analysis of the signature representations showed that the learned features are mostly useful to distinguish signatures on a ``general appearance'' instead of finer details. This make these features relevant for distinguishing random forgeries and skilled forgeries that are made with quick motion (but do not perfectly capture the overall aspect of the genuine signature). On the other hand, it makes these features less discriminant to slowly-traced skilled forgeries, where the forgery looks like a genuine signature in the overall shape, but has a poor line quality. Future work can investigate the combination of such features with features particularly targeted to discriminate the quality of pen strokes. \section*{Acknowledgment} This research has been supported by the CNPq grant \#206318/2014-6. \bibliographystyle{IEEEtran}
490da31f8763b0a623c5c49a47d8963a66578df6
{ "file_path": "/home/ubuntu/dolma-v1_7/arxiv-0000.json.gz" }