\chapter{Walk expansions}\label{s:walks} In this section and in \zcref{s:contraction}, we prove that the hypotheses (H1), (H2), and (H3) imply Hypothesis~(P); this implication is \zcref{t:contraction}. The three hypotheses are verified with computer assistance in \zcref{s:computer}, and Hypothesis~(P) is the input of \zcref{t:window}. In this section, we expand a pair correlation of the model with pins of \zcref{s:cut} into a sum over walks that start at one site of the pair and stop at the other site or near the boundary of the base set. Each step of a walk integrates out one spin and changes the domain of integration. We show that the pins change the weight of a step only through the responses of pairs of neighbors of the site that the walk leaves, and these pairs are of the same kind as the pairs in Hypothesis~(P). We also state the local hypothesis (H1), which bounds the weight of a single step, and the growth hypothesis (H2) for the walks. The heuristics behind the argument are explained in \zcref{s:overview}. \section{Domains}\label{s:walks-domains} Throughout this section and \zcref{s:contraction}, $G \subset \Z^2$ denotes a finite set and $\tau$ denotes a pin configuration on $G$. Recall from \zcref{s:cut} the legs of $G$, the depth $\dep_G(x) = \dist(x, \Z^2 \setminus G)$, and the boundary layer $\partial_2 G = \{ z \in G : \dep_G(z) \le 2 \}$. Recall also that the site of every leg has depth one, and in particular belongs to $\partial_2 G$. Each step of our expansion modifies the domain of integration. We record these modifications with \zcref{d:domain} below, which extends the notion of a domain in $G$ of \zcref{s:cut-model}. Before stating it, we define the measures that the sites will carry. We regard the uniform measure $\dd n$ on $\Sph$ as a probability measure on $\Sph \cup \{0\}$ with $\dd n(\{0\}) = 0$, we let $\delta_0$ denote the unit mass at $0 \in \R^3$, and we define the probability measures \begin{equation}\label{e:sitemeasures} \lambda_0 = \dd n, \qquad \lambda_1 = \frac35 \dd n + \frac25 \delta_0 \end{equation} on $\Sph \cup \{0\}$. \begin{definition}\label{d:domain} A \emph{domain} $D$ on $G$ consists of a \emph{status} $\mathrm{st}_D(x) \in \{0, 1, 2\}$ for each site $x \in G$ and a set of bonds $\{x, y\}$ with $x, y \in G$, which we call the \emph{used} bonds of $D$. As in \zcref{s:cut-model}, we call a site of status $2$ removed. We call a bond $\{x, y\}$ with $x, y \in G$ \emph{present} in $D$ if it is not used and neither $x$ nor $y$ is removed. We require that every site of $\partial_2 G$ has status $0$ and that every site of status $1$ has at most two present bonds. We let $\Dcons(G)$ denote the set of domains on $G$ in which every site of status $0$ has no used bond and every site of status $1$ has exactly two used bonds, and we call these domains \emph{consistent}. \end{definition} The statuses count the passages of the walks of our expansion through a site. A site of status $0$ has not been visited, a site of status $1$ has been visited once, and a site of status $2$ has been visited twice or deleted. A walk uses the bonds that it traverses, and a used bond counts at its two ends. When a walk passes once through a site, it uses two bonds at this site, and the measure $\lambda_1$ will replace the factors of these two bonds. (The weight $3/5$ of $\lambda_1$ on $\Sph$ will produce the factor $3/5$ of a revisit in \zcref{p:pairwalk}.) In a domain $D$, each non-removed site $x$ carries a spin $n_x \in \Sph \cup \{0\}$. For a bounded measurable function $f$ of these spins, we define \begin{equation}\label{e:ZD} Z^\tau_D[f] = \int f \prod_{\{x, y\} \text{ present}} (1 + n_x \cdot n_y) \prod_{(x, z) \text{ leg of } G} (1 + \tau_{xz} \cdot n_x) \prod_{x \in G : \, \mathrm{st}_D(x) \neq 2} \lambda_{\mathrm{st}_D(x)}(\mathrm d n_x), \end{equation} and we let $Z^0_D[f]$ denote the same integral without the factors of the legs. The factors of the legs make sense, since the site of every leg belongs to $\partial_2 G$ and therefore has status $0$. We set $Z^\cdot_D = Z^\cdot_D[1]$ and $\E^\cdot_D[f] = Z^\cdot_D[f] / Z^\cdot_D$, where here and below the dot stands for $\tau$ or $0$. Since all spins and pins have norm at most one, the factors in \eqref{e:ZD} are nonnegative. They are positive almost everywhere on the event that all spins belong to $\Sph$, which has positive probability under the product of the site measures. We conclude that $Z^\cdot_D > 0$ and that $\E^\cdot_D$ is a probability measure. Let $Y \subset G \setminus \partial_2 G$. We identify the domain $G \setminus Y$ in $G$ of \zcref{s:cut-model} with the domain on $G$ in which the sites of $Y$ are removed, the other sites have status $0$, and the set of used bonds is empty. This domain is consistent. The present bonds of this domain are the bonds $\{x, y\}$ with $x, y \in G \setminus Y$, and the sites of all legs of $G$ belong to $G \setminus Y$. Therefore, the integrals \eqref{e:ZD} for this domain are those of \eqref{e:Zpin}. For a domain $D$ on $G$ and non-removed sites $u$ and $v$, we call $\E^\tau_D[n_u \cdot n_v] - \E^0_D[n_u \cdot n_v]$ the response of the pair $(u, v)$ in $D$, as in \zcref{s:cut-delete}. We will need the following elementary extension of \zcref{l:elementary,l:threefifths} to domains. \begin{lemma}\label{l:domainfacts} Let $D$ be a domain on $G$, and let the dot stand for $\tau$ or $0$. \begin{enumerate} \item For all $k \ge 1$ and $a_1, \dots, a_k \in \{1, 2, 3\}$, we have \begin{equation*} \int n^{a_1} \cdots n^{a_k} \, \lambda_1(\mathrm d n) = \frac35 \int n^{a_1} \cdots n^{a_k} \dd n. \end{equation*} \item Let $f$ be a finite product of factors $n_x \cdot n_y$, where $x$ and $y$ are non-removed sites. Then $\E^0_D[f] \ge 0$. \item Let $x \neq y$ be non-removed sites such that $\{x, y\}$ is not a present bond. Then, under $\E^\cdot_D$, the spins $n_x$ and $n_y$ are conditionally independent given the spins of all other non-removed sites. \item Let $x$ be a non-removed site with $\dep_G(x) \ge 2$ and at most three present bonds. Then, under $\E^\cdot_D$, the conditional mean of $n_x$ given the spins of all other non-removed sites has norm at most $3/5$. \end{enumerate} \end{lemma} \begin{proof} The measure $\delta_0$ integrates every monomial of positive degree to zero, which proves (1). For (2), we expand the factors of the present bonds in $Z^0_D[f]$ as $\prod (1 + n_x \cdot n_y) = \sum_B \prod_{\{x, y\} \in B} n_x \cdot n_y$, where $B$ runs over the sets of present bonds. Every term of the resulting sum is the integral of a product of scalar products, and we expand each scalar product in coordinates. For each choice of the indices, the integral is a product, over the non-removed sites, of moments of $\lambda_0$ or $\lambda_1$, and these moments are nonnegative by \eqref{e:moments} and part (1). Summing over the indices and over $B$, we find that $Z^0_D[f] \ge 0$, and (2) follows from $Z^0_D > 0$. For (3), observe that every factor of \eqref{e:ZD} involves at most one of the spins $n_x$ and $n_y$. The conditional density of $(n_x, n_y)$ given the other spins is therefore the product of a function of $n_x$ and a function of $n_y$. For (4), let $b_1, \dots, b_k$, with $k \le 3$, denote the spins of the sites joined to $x$ by present bonds, and set $b_i = 0$ for $k < i \le 3$. The site $x$ is not the site of a leg, since $\dep_G(x) \ge 2$. The conditional law of $n_x$ given the other spins is thus proportional to $\prod_{i = 1}^3 (1 + n \cdot b_i) \, \lambda_{\mathrm{st}_D(x)}(\mathrm d n)$, and the vectors $b_i$ belong to $\Ball$. By part (1) of \zcref{l:threefifths}, the normalization $\nu$ of $\prod_{i = 1}^3 (1 + n \cdot b_i) \dd n$ has mean of norm at most $3/5$. This is the claim when $\mathrm{st}_D(x) = 0$. When $\mathrm{st}_D(x) = 1$, the conditional law is a convex combination of $\nu$ and $\delta_0$, since the product is one at $n = 0$. Its mean also has norm at most $3/5$. \end{proof} \section{Trails and traced domains}\label{s:walks-trails} We now define the walks of the expansion and the domains that they leave behind. \begin{definition}\label{d:trail} Let $D$ be a domain on $G$, and let $z \in G$ be a non-removed site. A \emph{trail} in $D$ from $z$ is a sequence $w = (v_0, \dots, v_m)$ of sites, with $m \ge 0$ and $v_0 = z$, such that for all $j \in \{1, \dots, m\}$ the bond $\{v_{j - 1}, v_j\}$ is present in $D[w_{j - 1}]$. Here $w_j = (v_0, \dots, v_j)$, and the pairs $D[w_j]$ of statuses and used bonds are defined recursively by $D[w_0] = D$, with $D[w_j]$ obtained from $D[w_{j - 1}]$ by adding $\{v_{j - 1}, v_j\}$ to the used bonds and raising the status of $v_{j - 1}$ by one. We call $|w| = m$ the \emph{length} of $w$ and $w_j$ its \emph{prefixes}, and we call $D[w] = D[w_m]$ the \emph{traced domain} of $w$. \end{definition} Since a present bond has two non-removed ends, the site $v_{j - 1}$ has status $0$ or $1$ in $D[w_{j - 1}]$ and status $1$ or $2$ in $D[w_j]$. Each step adds a bond to the used bonds. Therefore, the bonds $\{v_{j - 1}, v_j\}$ of a trail are pairwise distinct, and the length of a trail is bounded by the number of bonds $\{x, y\}$ with $x, y \in G$. We will show in \zcref{l:traced} that the pairs $D[w_j]$ are domains for the trails that we use. Let $w = (v_0, \dots, v_m)$ be a trail in $D$. For $1 \le j \le m$, we call the move from $v_{j - 1}$ to $v_j$ the \emph{step} $j$ of $w$, and we call $v_{j - 1}$ its \emph{passage site}. Step $j$ is a \emph{first visit} if $\mathrm{st}_{D[w_{j - 1}]}(v_{j - 1}) = 0$ and a \emph{revisit} if $\mathrm{st}_{D[w_{j - 1}]}(v_{j - 1}) = 1$. We let $\FV(w) \subset \{1, \dots, m\}$ denote the set of first-visit steps of $w$, and we let $\rev_D(w) = m - |\FV(w)|$ denote the number of revisits. Note that a step from a site of status $1$ in $D$ is a revisit even if $w$ has not visited this site before. For $T \subset G$ such that $z \notin T$, we let $\mathcal T_D(z, T)$ denote the set of trails $w = (v_0, \dots, v_m)$ in $D$ from $z$ such that $m \ge 1$, $v_m \in T$, and $v_i \notin T$ for $0 \le i < m$. This set is finite. The following lemma records the properties of traced domains that we use. Its proof is a routine induction along the trail. \begin{lemma}\label{l:traced} Let $D'$ be a domain on $G$, let $T \subset G$ be a set that contains $\partial_2 G$, let $z \in G \setminus T$ be a non-removed site with at most three present bonds in $D'$, and let $w = (v_0, \dots, v_m)$ be a trail in $D'$ from $z$ such that $v_i \notin T$ for $0 \le i < m$. Then $D'[w_j]$ is a domain on $G$ for $0 \le j \le m$. Further, let $2 \le j \le m$. Then the bond $\{v_{j - 2}, v_{j - 1}\}$ is used in $D'[w_{j - 1}]$, and if step $j$ is a revisit, then $\{v_{j - 1}, v_j\}$ is the only present bond at $v_{j - 1}$ in $D'[w_{j - 1}]$. \end{lemma} \begin{proof} Along $w$, the sites that change their status are the passage sites $v_0, \dots, v_{m - 1}$, which do not belong to $T \supset \partial_2 G$. In particular, every site of $\partial_2 G$ has status $0$ in $D'[w_j]$ for all $j$. By the definition of the traced domains, the bond $\{v_{j - 2}, v_{j - 1}\}$ is added to the used bonds at step $j - 1$, which proves the second statement. We show by induction on $j$ that every site of status $1$ in $D'[w_j]$ has at most two present bonds. This holds for $j = 0$, since $D'$ is a domain. Suppose that it holds for $j - 1$. The bonds that are present in $D'[w_j]$ are present in $D'[w_{j - 1}]$, and $v_{j - 1}$ is the only site whose status changes at step $j$. Suppose first that $v_{j - 1}$ has status $0$ in $D'[w_{j - 1}]$. Then it has at most three present bonds there, by assumption for $j = 1$ and by the second statement for $j \ge 2$. The bond $\{v_{j - 1}, v_j\}$ becomes used at step $j$, and $v_{j - 1}$ obtains the status $1$ and at most two present bonds. Suppose next that $v_{j - 1}$ has status $1$ in $D'[w_{j - 1}]$. Then it is removed in $D'[w_j]$. This completes the induction. Now let $2 \le j \le m$ be such that step $j$ is a revisit, and let $y = v_{j - 1}$. The status of $y$ changes only at the steps that leave $y$, and step $j - 1$ leaves $v_{j - 2} \neq y$. We conclude that $y$ has status $1$ in $D'[w_{j - 2}]$ as well. In this domain, $y$ has at most two present bonds, and $\{v_{j - 2}, y\}$ is one of them, since step $j - 1$ uses it. This bond is used in $D'[w_{j - 1}]$. Therefore, $y$ has at most one present bond in $D'[w_{j - 1}]$, and $\{y, v_j\}$ is present there, since step $j$ uses it. \end{proof} For the rest of this section, we let $D$ be a domain on $G$, $u \in G \setminus \partial_2 G$ a non-removed site, $v_{-1}$ a neighbor of $u$ such that the bond $\{v_{-1}, u\}$ is not present in $D$, and $v \neq u$ a non-removed site, and we set $T = \partial_2 G \cup \{v\}$. We assume that $u$ has at most one present bond if $\mathrm{st}_D(u) = 1$. In \zcref{s:contraction}, $v_{-1}$ will be the removed common neighbor of a pair $(u, v)$. For a trail $w = (v_0, \dots, v_m)$ in $D$ from $u$ and $1 \le j \le m$, we call $v_{j - 2}$ the \emph{arrival site} of step $j$ and $\{v_{j - 2}, v_{j - 1}\}$ its \emph{arrival bond}; for $j = 1$, they are $v_{-1}$ and $\{v_{-1}, u\}$. Since the bond $\{v_{-1}, u\}$ is not present in $D$, the site $u$ has at most three present bonds. We may therefore apply \zcref{l:traced} with $D' = D$ and $z = u$, and we find that $D[w_j]$ is a domain for all $w \in \mathcal T_D(u, T)$ and $0 \le j \le |w|$. The arrival bond of each step is not present before that step, by assumption for $j = 1$ and by \zcref{l:traced} for $j \ge 2$. If step $1$ is a revisit, then $u$ has status $1$ and $\{u, v_1\}$ is its only present bond, by the assumption on $u$. \zcref[S]{f:trail} shows an example of a trail and its traced domain. \begin{figure}[ht] \centering \begin{tikzpicture}[scale=0.95, st0/.style={circle, draw, fill=white, inner sep=0pt, minimum size=7pt}, st1/.style={circle, draw, fill=black!35, inner sep=0pt, minimum size=7pt}, st2/.style={circle, draw, fill=black, inner sep=0pt, minimum size=7pt}, trailstep/.style={-{Stealth[length=6pt]}, very thick, shorten <=4pt, shorten >=4pt}] \draw[black!15] (-2.6, -1.6) grid (3.6, 3.6); \foreach \x in {-2, ..., 3} { \foreach \y in {-1, ..., 3} { \node[st0] at (\x, \y) {}; } } \foreach \p in {(-1, 0), (-1, 1), (0, 1), (1, 2), (2, 2), (2, 1)} { \node[st1] at \p {}; } \node[st2] at (0, 0) {}; \node[st2] at (1, 1) {}; \draw[-{Stealth[length=6pt]}, thick, dashed, shorten <=4pt, shorten >=4pt] (0, 0) -- (-1, 0); \draw[trailstep] (-1, 0) -- (-1, 1); \draw[trailstep] (-1, 1) -- (0, 1); \draw[trailstep] (0, 1) -- (1, 1); \draw[trailstep] (1, 1) -- (1, 2); \draw[trailstep] (1, 2) -- (2, 2); \draw[trailstep] (2, 2) -- (2, 1); \draw[trailstep] (2, 1) -- (1, 1); \draw[trailstep] (1, 1) -- (1, 0); \node[below left] at (-1, 0) {$u$}; \node[below] at (0, -0.12) {$c$}; \node[below right] at (1, 0) {$v$}; \node[above left] at (1, 1) {$y$}; \end{tikzpicture} \caption{A trail $w = (v_0, \dots, v_8)$ in $D$ from $u = (-1, 0)$ to $v = (1, 0)$, shown on a portion of the base set $G$. In $D$, the site $c = (0, 0)$ is removed, all other sites shown have status $0$, and no bond is used. The dashed arrow shows the arrival site $v_{-1} = c$ of $u$. The circles show the statuses in the traced domain $D[w]$, with white for status $0$, gray for status $1$, and black for status $2$. The trail leaves the site $y = (1, 1)$ twice, at steps $4$ and $8$. Step $8$ is a revisit and all other steps are first visits.} \label{f:trail} \end{figure} \section{The step identity and the pair walk identity}\label{s:walks-identities} The walk expansion rests on the following identity, which integrates out one spin at the cost of one step of a trail. Its proof is a direct computation with the moments \eqref{e:moments}. \begin{lemma}\label{l:stepid} Let $a, b_1, \dots, b_k \in \R^3$. If $k \le 4$, then \begin{equation}\label{e:stepid1} \int (a \cdot n) \prod_{i = 1}^k (1 + n \cdot b_i) \dd n = \sum_{i = 1}^k \frac{a \cdot b_i}{3} \int \prod_{i' \neq i} (1 + n \cdot b_{i'}) \, \lambda_1(\mathrm d n). \end{equation} If $k \le 2$, then \begin{equation}\label{e:stepid2} \int (a \cdot n) \prod_{i = 1}^k (1 + n \cdot b_i) \, \lambda_1(\mathrm d n) = \frac35 \sum_{i = 1}^k \frac{a \cdot b_i}{3}. \end{equation} \end{lemma} \begin{proof} We expand the products and integrate term by term with \eqref{e:moments}; the terms of odd degree in $n$ vanish. Let $k \le 4$. Only the terms of degree two and four contribute to the left side of \eqref{e:stepid1}, since its terms have degree at most five. We find that the left side of \eqref{e:stepid1} is \begin{equation}\label{e:stepid-lhs} \frac13 \sum_{i = 1}^k a \cdot b_i + \frac1{15} \sum_{i < i' < i''} \bigl[ (a \cdot b_i)(b_{i'} \cdot b_{i''}) + (a \cdot b_{i'})(b_i \cdot b_{i''}) + (a \cdot b_{i''})(b_i \cdot b_{i'}) \bigr]. \end{equation} Each product on the right side of \eqref{e:stepid1} has at most three factors. By part (1) of \zcref{l:domainfacts} and \eqref{e:moments}, we have \begin{equation*} \int \prod_{i' \neq i} (1 + n \cdot b_{i'}) \, \lambda_1(\mathrm d n) = 1 + \frac15 \sum_{\substack{i' < i'' \\ i', i'' \neq i}} b_{i'} \cdot b_{i''}. \end{equation*} Multiplying by $(a \cdot b_i)/3$ and summing over $i$, we obtain \eqref{e:stepid-lhs} again, since each triple $i < i' < i''$ arises from three choices of the index outside the pair. This gives \eqref{e:stepid1}. Next, let $k \le 2$. The integrand on the left side of \eqref{e:stepid2} has degree at most three in $n$, and part (1) of \zcref{l:domainfacts} and \eqref{e:moments} give \begin{equation*} \int (a \cdot n) \prod_{i = 1}^k (1 + n \cdot b_i) \, \lambda_1(\mathrm d n) = \frac35 \sum_{i = 1}^k \int (a \cdot n)(n \cdot b_i) \dd n = \frac35 \sum_{i = 1}^k \frac{a \cdot b_i}{3}. \qedhere \end{equation*} \end{proof} In \eqref{e:stepid1}, the factor $1 + n \cdot b_i$ of one bond is replaced by $(a \cdot b_i)/3$, and the remaining factors are integrated against $\lambda_1$. In \eqref{e:stepid2}, the remaining factors disappear, and the factor $3/5$ appears. Expansions of the correlations of the classical AKLT model into bond-self-avoiding walks with weight $3^{-|w|}$, obtained by integrating out one site at a time with the moment identities of the sphere, go back to Kennedy, Lieb, and Tasaki \cite{KLT88}. They used them to prove exponential decay of the two-point function on bipartite square-lattice tori. For N\'eel boundary conditions they bounded the staggered magnetization by the boundary size times a function that decays exponentially in the distance to the boundary. \zcref[S]{l:stepid} and the following proposition are versions of their expansion in domains with pins and removed sites. \begin{proposition}\label{p:pairwalk} Let $D$, $u$, $v$, and $T$ be as above. Then for $\cdot \in \{\tau, 0\}$, we have \begin{equation}\label{e:pairwalk} \E^\cdot_D[n_u \cdot n_v] = \sum_{w \in \mathcal T_D(u, T)} 3^{-|w|} \Bigl( \frac35 \Bigr)^{\rev_D(w)} \varrho^\cdot_D(w) \, \E^\cdot_{D[w]}[n_{v_{|w|}} \cdot n_v], \qquad \varrho^\cdot_D(w) = \frac{Z^\cdot_{D[w]}}{Z^\cdot_D}. \end{equation} If $w \in \mathcal T_D(u, T)$ ends at $v_{|w|} = v$, then $\E^\cdot_{D[w]}[n_{v_{|w|}} \cdot n_v] = \E^\cdot_{D[w]}[|n_v|^2]$, which equals $1$ if $\mathrm{st}_D(v) = 0$, and $3/5$ if $\mathrm{st}_D(v) = 1$ and $v$ has exactly one present bond in $D$. \end{proposition} \begin{proof} For $\beta \in \{1, 2, 3\}$, we write $n^\beta$ for the $\beta$-th coordinate of $n$. We first prove the following claim. Let $D'$ be a domain on $G$ in which $v$ is non-removed, and let $z \in G \setminus T$ be a non-removed site with at most three present bonds in $D'$. Then for all $\beta$, \begin{equation}\label{e:pairwalk-claim} Z^\cdot_{D'}[n_z^\beta n_v^\beta] = \sum_{w \in \mathcal T_{D'}(z, T)} 3^{-|w|} \Bigl( \frac35 \Bigr)^{\rev_{D'}(w)} Z^\cdot_{D'[w]}[n^\beta_{v_{|w|}} n^\beta_v]. \end{equation} By \zcref{l:traced}, the traced domains on the right side are domains, and so is $D'[(z, y)]$ for every site $y$ such that $\{z, y\}$ is present in $D'$. We prove \eqref{e:pairwalk-claim} by induction on the number of present bonds of $D'$. When $z$ has no present bond, the two sides of \eqref{e:pairwalk-claim} vanish; this covers the base case, in which $D'$ has no present bond. We begin by integrating out $n_z$. Let $y_1, \dots, y_k$ be the sites joined to $z$ by present bonds, so that $k \le 3$. Recall that $z \notin \partial_2 G$ is not the site of a leg, and note that $v \neq z$, since $z \notin T$. The factors of $Z^\cdot_{D'}[n_z^\beta n_v^\beta]$ that involve $n_z$ are therefore $n_z^\beta$ and $1 + n_z \cdot n_{y_i}$ for $1 \le i \le k$. When $\mathrm{st}_{D'}(z) = 0$, we apply \eqref{e:stepid1} with $a$ the $\beta$-th unit vector and $b_i = n_{y_i}$. When $\mathrm{st}_{D'}(z) = 1$, we have $k \le 2$, and we apply \eqref{e:stepid2}. In each case, the $i$-th term of the result, together with the factors that do not involve $n_z$, is the integrand of $Z^\cdot_{D'[(z, y_i)]}[n^\beta_{y_i} n^\beta_v]$, multiplied by $1/3$ in the first case and by $(1/3)(3/5)$ in the second. We obtain \begin{equation}\label{e:pairwalk-onestep} Z^\cdot_{D'}[n_z^\beta n_v^\beta] = \sum_{i = 1}^k \frac13 \Bigl( \frac35 \Bigr)^{\mathbf 1\{\mathrm{st}_{D'}(z) = 1\}} Z^\cdot_{D'[(z, y_i)]}[n^\beta_{y_i} n^\beta_v]. \end{equation} We now treat the terms of \eqref{e:pairwalk-onestep} one by one. For $y_i \in T$, the $i$-th term is the term of the trail $(z, y_i)$ in \eqref{e:pairwalk-claim}. For $y_i \notin T$, we let $D'' = D'[(z, y_i)]$, which is a domain with fewer present bonds than $D'$. The site $v \in T$ remains non-removed because only the status of $z \notin T$ changes. In $D''$, the site $y_i$ is non-removed and has at most three present bonds, since $\{z, y_i\}$ is used. By the induction hypothesis, \eqref{e:pairwalk-claim} holds for $D''$ and $y_i$. The trails in $\mathcal T_{D'}(z, T)$ whose first step goes to $y_i$ are the sequences $(z, v'_0, \dots, v'_{m'})$ with $w'' = (v'_0, \dots, v'_{m'}) \in \mathcal T_{D''}(y_i, T)$. For such a trail $w$, the recursive definition of the traced domains gives \begin{equation*} D'[w] = D''[w''], \qquad |w| = |w''| + 1, \qquad \rev_{D'}(w) = \rev_{D''}(w'') + \mathbf 1\{\mathrm{st}_{D'}(z) = 1\}. \end{equation*} Inserting \eqref{e:pairwalk-claim} for $D''$ into \eqref{e:pairwalk-onestep}, we find that \eqref{e:pairwalk-claim} holds for $D'$. We now apply \eqref{e:pairwalk-claim} with $D' = D$ and $z = u$, which is allowed, since $u \notin T$ and $u$ has at most three present bonds. Summing over $\beta$ and dividing by $Z^\cdot_D$, we obtain \eqref{e:pairwalk}. Finally, let $w \in \mathcal T_D(u, T)$ end at $v$. Since $v$ is not a passage site of $w$, its status in $D[w]$ is $\mathrm{st}_D(v)$. When this status is $0$, we have $|n_v| = 1$ almost surely. Suppose that $\mathrm{st}_D(v) = 1$ and that $v$ has exactly one present bond in $D$. The last step of $w$ uses a bond at $v$ that is present in $D$, and $v$ has no present bond in $D[w]$. Further, $v$ is not the site of a leg, since a site of status $1$ does not belong to $\partial_2 G$. Under $\E^\cdot_{D[w]}$, the spin $n_v$ is therefore independent of the other spins and has law $\lambda_1$, and $\E^\cdot_{D[w]}[|n_v|^2] = 3/5$. \end{proof} \section{Step ratios and responses}\label{s:walks-steps} We now compute how the pins affect the factor $\varrho^\cdot_D(w)$ in \eqref{e:pairwalk}, one step at a time. Let $w = (v_0, \dots, v_m) \in \mathcal T_D(u, T)$. For $1 \le j \le m$, we define the \emph{step ratios} \begin{equation}\label{e:stepratio} \rho^\cdot_j = \frac{Z^\cdot_{D[w_j]}}{Z^\cdot_{D[w_{j - 1}]}}, \qquad \text{so that} \qquad \varrho^\cdot_D(w) = \prod_{j = 1}^m \rho^\cdot_j. \end{equation} We suppress the dependence on $w$ in this notation. Let $1 \le j \le m$, and write $x = v_{j - 1}$ for the passage site and $o = v_j$ for the \emph{out-neighbor} of step $j$. We call the sites $y$ such that $\{x, y\}$ is present in $D[w_{j - 1}]$ the \emph{remaining neighbors} of $x$ at step $j$. The arrival bond of step $j$ is not present in $D[w_{j - 1}]$ (by \zcref{l:traced} and the remark after it), so the remaining neighbors of $x$ belong to the set of the three other neighbors of $x$, and $o$ is one of them. We let $f$ and $g$ denote the two neighbors of $x$ other than $v_{j - 2}$ and $o$, in either order. We let $D^-_j$ denote the domain obtained from $D[w_{j - 1}]$ by changing the status of $x$ to $2$, and we define \begin{equation}\label{e:IJ} I_{\mathrm{cur}} = 1 + \frac13 (n_f \cdot n_g + n_f \cdot n_o + n_g \cdot n_o), \qquad I_{\mathrm{next}} = 1 + \frac15 n_f \cdot n_g, \end{equation} with the convention that $n_f$ is replaced by $0$ if $f$ is not a remaining neighbor of $x$ at step $j$, and likewise for $g$. These are functions of the spins of the remaining neighbors, which are not removed in $D^-_j$. \begin{lemma}\label{l:ratios} Let $w \in \mathcal T_D(u, T)$ and $1 \le j \le |w|$, and let $x$, $o$, $f$, and $g$ be as above. \begin{enumerate} \item If step $j$ is a revisit, then $\rho^\tau_j = \rho^0_j = 1$. \item If step $j$ is a first visit, then $D^-_j$ is a domain on $G$, and for $\cdot \in \{\tau, 0\}$, \begin{equation}\label{e:ratios} Z^\cdot_{D[w_{j - 1}]} = Z^\cdot_{D^-_j} \E^\cdot_{D^-_j}[I_{\mathrm{cur}}], \qquad Z^\cdot_{D[w_j]} = Z^\cdot_{D^-_j} \E^\cdot_{D^-_j}[I_{\mathrm{next}}], \qquad \rho^\cdot_j = \frac{\E^\cdot_{D^-_j}[I_{\mathrm{next}}]}{\E^\cdot_{D^-_j}[I_{\mathrm{cur}}]}. \end{equation} \end{enumerate} \end{lemma} \begin{proof} Recall that the passage site $x$ does not belong to $T \supset \partial_2 G$, so that $x$ is not the site of a leg. Suppose first that step $j$ is a revisit. By \zcref{l:traced} and the remark after it, $x$ has status $1$ in $D[w_{j - 1}]$, and $\{x, o\}$ is its only present bond. In $D[w_j]$, the site $x$ is removed, and all other statuses and present bonds are those of $D[w_{j - 1}]$. Integrating out $n_x$ in $Z^\cdot_{D[w_{j - 1}]}$ and using \begin{equation*} \int (1 + n \cdot n_o) \, \lambda_1(\mathrm d n) = 1, \end{equation*} we obtain $Z^\cdot_{D[w_j]}$, which proves (1). Suppose now that step $j$ is a first visit. The site $x \notin \partial_2 G$ is removed in $D^-_j$, and removing a site can only decrease the set of present bonds, so that $D^-_j$ is a domain. The factors of \eqref{e:ZD} for $D[w_{j - 1}]$ that involve $n_x$ are those of the bonds from $x$ to its remaining neighbors $y$, of which there are at most three. In $D[w_j]$, the site $x$ has status $1$, and its present bonds join it to its remaining neighbors other than $o$. By \eqref{e:moments} and part (1) of \zcref{l:domainfacts}, we have \begin{equation*} \int \prod_{y} (1 + n \cdot n_y) \dd n = I_{\mathrm{cur}}, \qquad \int \prod_{y \neq o} (1 + n \cdot n_y) \, \lambda_1(\mathrm d n) = I_{\mathrm{next}}, \end{equation*} where the products run over the remaining neighbors $y$ of $x$. Here the terms of degree one and three vanish, each term of degree two contributes $(n_y \cdot n_{y'})/3$ in the first integral and $(n_y \cdot n_{y'})/5$ in the second, and there is no term of degree four. The other factors and site measures of $D[w_{j - 1}]$ and of $D[w_j]$ are those of $D^-_j$, since the present bonds of $D^-_j$ are the present bonds of $D[w_{j - 1}]$ that do not contain $x$. Integrating out $n_x$ first, we obtain the first two identities in \eqref{e:ratios}, and the third follows from them. \end{proof} A first-visit step $j$ is \emph{straight} if $o - x = x - v_{j - 2}$ and a \emph{turn} otherwise. For $j = 1$, this refers to the arrival site $v_{-1}$ of $u$. For a first-visit step $j$, we define \begin{equation}\label{e:Gamma} \Gamma_j = \log \rho^0_j - \log \rho^\tau_j, \end{equation} and for remaining neighbors $a \neq b$ of $x$ at step $j$, we let \begin{equation}\label{e:rab} r_{ab} = \E^\tau_{D^-_j}[n_a \cdot n_b] - \E^0_{D^-_j}[n_a \cdot n_b] \end{equation} denote the response of the pair $(a, b)$ in $D^-_j$. The following lemma plays for a site that a trail passes the role that \zcref{l:delete} plays for a deleted site. In the two lemmas, the effect of the pins is expressed through the responses of the pairs formed by neighbors of the site, computed in a domain in which the site is removed. This is the main point of the expansion, since these pairs are again at distance $2$ and have a removed common neighbor. \begin{lemma}\label{l:response} Let $w \in \mathcal T_D(u, T)$, let step $j$ of $w$ be a first visit, and let $x$, $o$, $f$, and $g$ be as above. Then $\rho^0_j \le 1$. Suppose in addition that $\rho^\tau_j \le 200/187$. \begin{enumerate} \item If $f$, $g$, and $o$ are remaining neighbors of $x$ at step $j$, then there exist $\alpha_j \in [0, 1/6]$ and $\beta_j \in (0, 9/25]$ such that \begin{equation}\label{e:response} \Gamma_j = \alpha_j r_{fg} + \beta_j (r_{fo} + r_{go}). \end{equation} \item If $y \in \{f, g\}$ and $o$ are the only remaining neighbors of $x$ at step $j$, then there exists $\beta'_j \in (0, 9/25]$ such that $\Gamma_j = \beta'_j r_{yo}$. \item If $o$ is the only remaining neighbor of $x$ at step $j$, then $\Gamma_j = 0$. \end{enumerate} \end{lemma} \begin{proof} For $\cdot \in \{\tau, 0\}$, we abbreviate $E^\cdot_{\mathrm{cur}} = \E^\cdot_{D^-_j}[I_{\mathrm{cur}}]$ and $E^\cdot_{\mathrm{next}} = \E^\cdot_{D^-_j}[I_{\mathrm{next}}]$, so that $\rho^\cdot_j = E^\cdot_{\mathrm{next}} / E^\cdot_{\mathrm{cur}}$ by \zcref{l:ratios}. \emph{Step 1: the free ratio.} By \eqref{e:IJ}, we have \begin{equation}\label{e:response-free} \begin{gathered} E^0_{\mathrm{cur}} - E^0_{\mathrm{next}} = \frac2{15} \E^0_{D^-_j}[n_f \cdot n_g] + \frac13 \E^0_{D^-_j}[n_f \cdot n_o + n_g \cdot n_o] \ge 0, \\ E^0_{\mathrm{next}} = 1 + \frac15 \E^0_{D^-_j}[n_f \cdot n_g] \ge 1, \end{gathered} \end{equation} where the inequalities follow from part (2) of \zcref{l:domainfacts}. We conclude that \begin{equation*} E^0_{\mathrm{cur}} \ge E^0_{\mathrm{next}} \ge 1, \qquad \rho^0_j = \frac{E^0_{\mathrm{next}}}{E^0_{\mathrm{cur}}} \le 1. \end{equation*} Further, the assumption $\rho^\tau_j \le 200/187$ means that \begin{equation}\label{e:response-assumption} E^\tau_{\mathrm{cur}} \ge \frac{187}{200} E^\tau_{\mathrm{next}}. \end{equation} \emph{Step 2: a lower bound for $E^\tau_{\mathrm{next}}$ in case (1).} In $D^-_j$, the sites $f$ and $g$ are not removed, and each of them has at most three present bonds, since its bond to the removed site $x$ is not present. The depth is $1$-Lipschitz and $\dep_G(x) \ge 3$, so that $\dep_G(f) \ge 2$ and $\dep_G(g) \ge 2$. Further, $f$ and $g$ are not neighbors. We condition on the spins of all non-removed sites other than $f$ and $g$. By part (3) of \zcref{l:domainfacts}, the conditional expectation of $n_f \cdot n_g$ is the scalar product of the conditional means of $n_f$ and $n_g$, and each of these conditional means is also the conditional mean given all spins other than its own. By part (4) of \zcref{l:domainfacts}, these conditional means have norm at most $3/5$. Taking expectations, we find that $\E^\tau_{D^-_j}[n_f \cdot n_g] \ge -(3/5)^2$, and \begin{equation}\label{e:response-116} E^\tau_{\mathrm{next}} \ge 1 - \frac15 \Bigl( \frac35 \Bigr)^2 = \frac{116}{125}. \end{equation} \emph{Step 3: interpolation in case (1).} For $t \in [0, 1]$, we define \begin{equation*} E_{\mathrm{cur}}(t) = (1 - t) E^0_{\mathrm{cur}} + t E^\tau_{\mathrm{cur}}, \qquad E_{\mathrm{next}}(t) = (1 - t) E^0_{\mathrm{next}} + t E^\tau_{\mathrm{next}}. \end{equation*} These numbers are positive, by \eqref{e:response-free}, \eqref{e:response-assumption}, and \eqref{e:response-116}. By \eqref{e:IJ} and \eqref{e:rab}, we have \begin{equation*} E^\tau_{\mathrm{cur}} - E^0_{\mathrm{cur}} = \frac13 (r_{fg} + r_{fo} + r_{go}), \qquad E^\tau_{\mathrm{next}} - E^0_{\mathrm{next}} = \frac15 r_{fg}. \end{equation*} Writing the differences of logarithms as integrals of their derivatives in $t$, we obtain \begin{equation*} \Gamma_j = \bigl( \log E^\tau_{\mathrm{cur}} - \log E^0_{\mathrm{cur}} \bigr) - \bigl( \log E^\tau_{\mathrm{next}} - \log E^0_{\mathrm{next}} \bigr) = \int_0^1 \Bigl( \frac{r_{fg} + r_{fo} + r_{go}}{3 E_{\mathrm{cur}}(t)} - \frac{r_{fg}}{5 E_{\mathrm{next}}(t)} \Bigr) \mathrm d t, \end{equation*} which is \eqref{e:response} with \begin{equation}\label{e:response-coefficients} \alpha_j = \int_0^1 \Bigl( \frac1{3 E_{\mathrm{cur}}(t)} - \frac1{5 E_{\mathrm{next}}(t)} \Bigr) \mathrm d t, \qquad \beta_j = \int_0^1 \frac{\mathrm d t}{3 E_{\mathrm{cur}}(t)} > 0. \end{equation} \emph{Step 4: the bounds in case (1).} Since all spins have norm at most one, we have \begin{equation*} 5 I_{\mathrm{next}} - 3 I_{\mathrm{cur}} = 2 - n_f \cdot n_o - n_g \cdot n_o \ge 0. \end{equation*} Taking expectations, we find that $5 E_{\mathrm{next}}(t) \ge 3 E_{\mathrm{cur}}(t)$ for all $t$, and $\alpha_j \ge 0$. Next, the inequality $E_{\mathrm{cur}}(t) \ge (187/200) E_{\mathrm{next}}(t)$ holds at $t = 0$ by \eqref{e:response-free} and at $t = 1$ by \eqref{e:response-assumption}. Both sides are affine in $t$, and the inequality holds for all $t \in [0, 1]$. Since $y \mapsto 1/y$ is convex on $(0, \infty)$, we also have \begin{equation}\label{e:response-convex} \int_0^1 \frac{\mathrm d t}{(1 - t) a + t a'} \le \frac12 \Bigl( \frac1a + \frac1{a'} \Bigr), \qquad a, a' > 0. \end{equation} Using $E_{\mathrm{cur}}(t) \ge (187/200) E_{\mathrm{next}}(t)$ together with \eqref{e:response-convex}, we obtain \begin{equation*} \alpha_j \le \Bigl( \frac{200}{561} - \frac15 \Bigr) \int_0^1 \frac{\mathrm d t}{E_{\mathrm{next}}(t)} \le \frac{439}{2805} \cdot \frac12 \Bigl( 1 + \frac{125}{116} \Bigr) = \frac{105799}{650760} < \frac16, \end{equation*} where in the second inequality we used $E^0_{\mathrm{next}} \ge 1$ and \eqref{e:response-116}. Similarly, \eqref{e:response-convex} gives \begin{equation*} \beta_j \le \frac16 \Bigl( \frac1{E^0_{\mathrm{cur}}} + \frac1{E^\tau_{\mathrm{cur}}} \Bigr) \le \frac16 \Bigl( 1 + \frac{125 \cdot 200}{116 \cdot 187} \Bigr) = \frac{3891}{10846} < \frac9{25}, \end{equation*} where in the second inequality we used $E^0_{\mathrm{cur}} \ge 1$, \eqref{e:response-assumption}, and \eqref{e:response-116}. \emph{Step 5: cases (2) and (3).} In case (2), our convention gives $I_{\mathrm{cur}} = 1 + (n_y \cdot n_o)/3$ and $I_{\mathrm{next}} = 1$. Then $E^\cdot_{\mathrm{next}} = 1$, and \eqref{e:response-free} and \eqref{e:response-assumption} give $E^0_{\mathrm{cur}} \ge 1$ and $E^\tau_{\mathrm{cur}} \ge 187/200$. With $E_{\mathrm{cur}}(t)$ as above, we find \begin{gather*} \Gamma_j = \log E^\tau_{\mathrm{cur}} - \log E^0_{\mathrm{cur}} = \beta'_j \, r_{yo}, \qquad \beta'_j = \int_0^1 \frac{\mathrm d t}{3 E_{\mathrm{cur}}(t)}, \\ 0 < \beta'_j \le \frac16 \Bigl( 1 + \frac{200}{187} \Bigr) = \frac{129}{374} < \frac9{25}, \end{gather*} where we used \eqref{e:response-convex} in the second line. In case (3), we have $I_{\mathrm{cur}} = I_{\mathrm{next}} = 1$, so that $\rho^\tau_j = \rho^0_j = 1$ and $\Gamma_j = 0$. \end{proof} We now bound the difference of the products \eqref{e:stepratio} with and without pins by the quantities $\Gamma_j$. We call a bijection $\varsigma \colon \{1, \dots, |\FV(w)|\} \to \FV(w)$ an \emph{enumeration} of $\FV(w)$. \begin{lemma}\label{l:exposure} Let $w \in \mathcal T_D(u, T)$, and let $\bar\rho_j \ge 1$, for $j \in \FV(w)$, be numbers such that $\rho^\tau_j \le \bar\rho_j$ for all $j \in \FV(w)$. Then for every enumeration $\varsigma$ of $\FV(w)$, we have \begin{equation}\label{e:exposure} \bigl| \varrho^\tau_D(w) - \varrho^0_D(w) \bigr| \le \sum_{k = 1}^{|\FV(w)|} |\Gamma_{\varsigma(k)}| \prod_{l = 1}^k \bar\rho_{\varsigma(l)}. \end{equation} \end{lemma} \begin{proof} Recall from part (1) of \zcref{l:ratios} that the step ratios of the revisits equal one, with and without pins. Let $q = |\FV(w)|$. We have \begin{equation*} \varrho^\tau_D(w) - \varrho^0_D(w) = \prod_{k = 1}^q \rho^\tau_{\varsigma(k)} - \prod_{k = 1}^q \rho^0_{\varsigma(k)} = \sum_{k = 1}^q \Bigl( \prod_{l < k} \rho^\tau_{\varsigma(l)} \Bigr) \bigl( \rho^\tau_{\varsigma(k)} - \rho^0_{\varsigma(k)} \bigr) \Bigl( \prod_{l > k} \rho^0_{\varsigma(l)} \Bigr). \end{equation*} For $j \in \FV(w)$, we have $\rho^\tau_j \le \bar\rho_j$ by assumption and $\rho^0_j \le 1 \le \bar\rho_j$ by \zcref{l:response}. By the mean value theorem for the exponential function, this implies that \begin{equation*} |\rho^\tau_j - \rho^0_j| \le |\log \rho^\tau_j - \log \rho^0_j| \max(\rho^\tau_j, \rho^0_j) \le |\Gamma_j| \, \bar\rho_j. \end{equation*} Inserting these bounds into the previous identity, we obtain \eqref{e:exposure}. \end{proof} \section{Patterns and the local hypothesis}\label{s:walks-patterns} The hypothesis $\rho^\tau_j \le 200/187$ of \zcref{l:response} is a real restriction, since pins can make a step ratio larger than one. We now reduce it to a finite family of inequalities, which form Hypothesis~(H1). The ratio $\rho^\tau_j$ depends on the whole domain. However, we will show that, after the passage site and four further sites are integrated out, a bound on it follows from an inequality that involves only the sites at distance at most $2$ from the passage site and twelve vectors of norm at most one. This reduction to finitely many local configurations is what makes the step bounds accessible to a computer-assisted proof. For $x \in \Z^2$, we define the \emph{template} $\tmpl(x) = \{ z \in \Z^2 : d(z, x) \le 2 \}$. In the template $\tmpl(0)$, we call $0$ the \emph{middle site} of the template (it corresponds to the passage site), we call the four neighbors $\sS = (0, -1)$, $\sE = (1, 0)$, $\sN = (0, 1)$, and $\sW = (-1, 0)$ of $0$ the \emph{axial sites}, the four sites $(\pm 1, \pm 1)$ the \emph{corners}, and the four sites $(\pm 2, 0)$ and $(0, \pm 2)$ the \emph{tips}. The tip of the axial site $z$ is $2z$, and we call the bond $\{z, 2z\}$ the \emph{tip bond} of $z$. The set $\mathsf R$ of the eight axial sites and corners is the \emph{ring}, and we call its elements ring sites. Each corner is a neighbor of two axial sites, and we call the eight bonds between a corner and an axial site the \emph{ring bonds}; they form a cycle through $\mathsf R$. Each corner $y$ has two neighbors outside $\tmpl(0)$, and we call the two pairs $(y, y')$ formed by $y$ and such a neighbor $y'$ the \emph{slots} of $y$. We enumerate the eight slots as $k = 1, \dots, 8$, in a fixed order, and we let $z(k)$ denote the corner of the slot $k$. \zcref[S]{f:template} shows the template. \begin{figure}[ht] \centering \begin{tikzpicture}[scale=1.25, ctr/.style={circle, draw, fill=black, inner sep=0pt, minimum size=7pt}, ax/.style={circle, draw, fill=black!35, inner sep=0pt, minimum size=7pt}, cor/.style={rectangle, draw, fill=white, inner sep=0pt, minimum size=6.5pt}, tip/.style={rectangle, draw, fill=white, rotate=45, inner sep=0pt, minimum size=5.5pt}] \draw[line width=3.2pt, black!18] (0, -1) -- (1, -1) -- (1, 0) -- (1, 1) -- (0, 1) -- (-1, 1) -- (-1, 0) -- (-1, -1) -- cycle; \draw (0, 0) -- (1, 0) (0, 0) -- (0, 1) (0, 0) -- (-1, 0); \draw[very thick] (0, -1) -- (0, 0); \draw (1, 0) -- (2, 0) (-1, 0) -- (-2, 0) (0, 1) -- (0, 2) (0, -1) -- (0, -2); \draw (0, -1) -- (1, -1) -- (1, 0) -- (1, 1) -- (0, 1) -- (-1, 1) -- (-1, 0) -- (-1, -1) -- cycle; \foreach \x/\y in {1/1, 1/-1, -1/1, -1/-1} { \draw[dashed] (\x, \y) -- (1.45*\x, \y); \draw[dashed] (\x, \y) -- (\x, 1.45*\y); } \node[ctr] at (0, 0) {}; \foreach \p in {(0, -1), (1, 0), (0, 1), (-1, 0)} { \node[ax] at \p {}; } \foreach \p in {(1, 1), (1, -1), (-1, 1), (-1, -1)} { \node[cor] at \p {}; } \foreach \p in {(2, 0), (-2, 0), (0, 2), (0, -2)} { \node[tip] at \p {}; } \node[above right] at (0, 0) {$0$}; \node[below right] at (0, -1) {$\sS$}; \node[below right] at (1, 0) {$\sE$}; \node[above right] at (0, 1) {$\sN$}; \node[below left] at (-1, 0) {$\sW$}; \node[below] at (2, -0.1) {$(2, 0)$}; \node[below] at (-2, -0.1) {$(-2, 0)$}; \node[right] at (0.1, 2) {$(0, 2)$}; \node[right] at (0.1, -2) {$(0, -2)$}; \node[above right] at (1, 1.1) {$(1, 1)$}; \end{tikzpicture} \caption{The template $\tmpl(x)$ after the translation that takes the passage site $x$ to the middle site $0$ (black) and the rotation that takes the arrival site to $\sS = (0, -1)$. The bond $\{\sS, 0\}$, the image of the arrival bond, is drawn thick. Gray circles: the axial sites $\sS$, $\sE$, $\sN$, and $\sW$. Squares: the corners $(\pm 1, \pm 1)$. Diamonds: the tips $(\pm 2, 0)$ and $(0, \pm 2)$, each joined to its axial site by its tip bond. The shaded cycle of ring bonds marks the ring $\mathsf R$. The dashed stubs are the slots, which lead from each corner to its two neighbors $(\pm 2, \pm 1)$ and $(\pm 1, \pm 2)$ outside the template.} \label{f:template} \end{figure} \begin{definition}\label{d:pattern} A \emph{pattern} $\pi$ consists of an out-neighbor $o_\pi \in \{\sE, \sN, \sW\}$, a status $\mathrm{st}_\pi(z) \in \{0, 1, 2\}$ for each ring site $z$, and a set of bonds, which we call the used bonds of $\pi$, each of which has an end at $0$ or at a ring site of status $0$ or $1$. We require that $\mathrm{st}_\pi(o_\pi) \neq 2$, and we regard the middle site $0$ as a site of status $0$. As in \zcref{d:domain}, a ring site of status $2$ is removed, and a ring bond is present in $\pi$ if it is not used and neither of its ends is removed. We call the tip bond of a non-removed axial site \emph{retained} if it is not used. We let $\Pext$ denote the set of patterns $\pi$ such that \begin{enumerate} \item the only used bond of $\pi$ at $0$ is $\{0, \sS\}$; \item the arrival site $\sS$ has status $1$ or $2$; \item every ring site of status $0$ has no used bond; and \item every ring site of status $1$ has exactly two used bonds, \end{enumerate} and we call the elements of $\Pext$ \emph{extended patterns}. We call a pattern straight if $o_\pi = \sN$ and a turn otherwise. \end{definition} The tips have no status in a pattern. Let $w = (v_0, \dots, v_m) \in \mathcal T_D(u, T)$, and let step $j$ of $w$ be a first visit with passage site $x$. Since $x \notin \partial_2 G$, the template $\tmpl(x)$ is contained in $G$. We transport the out-neighbor $o = v_j$, the statuses of the ring sites of $\tmpl(x)$, and the used bonds of $D[w_{j - 1}]$ that have an end at $x$ or at a non-removed ring site of $\tmpl(x)$ to $\tmpl(0)$, by the translation by $-x$ followed by the rotation of $\Z^2$ about $0$ that maps $v_{j - 2} - x$ to $\sS$. The image of $o$ belongs to $\{\sE, \sN, \sW\}$ and is not removed, and the middle site has status $0$, so that these data form a pattern. We call it the \emph{pattern of step $j$}. It is straight if and only if step $j$ is straight. We now define the ring functional. The \emph{exterior parameters} are $\mathbf a = (\mathbf u, \mathbf m) \in \Ball^{12}$, where $\mathbf u = (u_z)$ consists of one vector for each of the four axial sites $z$ and $\mathbf m = (m_k)$ of one vector for each of the eight slots $k$. Let $\pi$ be a pattern. Each non-removed ring site $z$ carries a spin $n_z \in \Sph \cup \{0\}$. For a bounded measurable function $h$ of these spins, we define \begin{multline}\label{e:ringfunctional} \mathcal Z_\pi^{\mathbf a}[h] = \int h \prod_{\{y, z\} \text{ present in } \pi} (1 + n_y \cdot n_z) \prod_{z \text{ axial, tip bond retained}} \Bigl( 1 + \frac35 u_z \cdot n_z \Bigr) \\ \times \prod_{k : \, \mathrm{st}_\pi(z(k)) \neq 2} (1 + m_k \cdot n_{z(k)}) \prod_{z \in \mathsf R : \, \mathrm{st}_\pi(z) \neq 2} \lambda_{\mathrm{st}_\pi(z)}(\mathrm d n_z), \end{multline} where the first product runs over the ring bonds that are present in $\pi$. Every slot of a non-removed corner has a factor, whether or not its bond is used. In the proof of \zcref{l:local}, a missing factor is represented by $m_k = 0$. For a pattern $\pi$, we let $f$ and $g$ denote the two sites of $\{\sE, \sN, \sW\} \setminus \{o_\pi\}$, and we define $I_{\mathrm{cur}}$ and $I_{\mathrm{next}}$ by \eqref{e:IJ} with $o = o_\pi$, where $n_f$ is replaced by $0$ if $f$ is removed, and likewise for $g$. For a first-visit step whose pattern belongs to $\Pext$, this agrees with the convention of \zcref{s:walks-steps}, as we will see in the proof of \zcref{l:local}. \begin{hypothesis}{H1} Let $\cs, \ct \in (0, 1]$. For every $\pi \in \Pext$ and every $\mathbf a \in \Ball^{12}$, we have \begin{equation}\label{e:H1poly} \mathcal Z_\pi^{\mathbf a}[I_{\mathrm{cur}} - c_\pi I_{\mathrm{next}}] \ge 0, \end{equation} where $c_\pi = \cs$ if $\pi$ is straight and $c_\pi = \ct$ if $\pi$ is a turn. \end{hypothesis} For this hypothesis, and for (H2) and (H3) below, we say that the hypothesis holds at $(\cs, \ct)$ when we need to specify the thresholds. Each exterior parameter occurs in at most one factor of \eqref{e:ringfunctional}, so that the left side of \eqref{e:H1poly} is affine in each of the twelve vectors. Since an affine function on $\Ball$ attains its minimum on $\Sph$, it suffices to check \eqref{e:H1poly} for $\mathbf a \in (\Sph)^{12}$. The class $\Pext$ is finite. Modulo the reflection $(z_1, z_2) \mapsto (-z_1, z_2)$ and the duplicate relation defined in \zcref{s:local-computation}, it consists of $7{,}962$ classes, of which $2{,}696$ are straight and $5{,}266$ are turns. \zcref[S]{t:H1} proves (H1) at $(\cs, \ct) = (961/1000, 187/200)$. \begin{lemma}\label{l:local} Let $\cs, \ct \in (0, 1]$, and assume that (H1) holds at $(\cs, \ct)$. Let $w \in \mathcal T_D(u, T)$, and let step $j$ of $w$ be a first visit whose pattern $\pi$ belongs to $\Pext$. Then $\rho^\tau_j \le 1/c_\pi$. \end{lemma} \begin{proof} Let $x = v_{j - 1}$ be the passage site, and write $D' = D[w_{j - 1}]$ and $D'' = D[w_j]$. We may apply a translation and a rotation of $\Z^2$ to $G$, $\tau$, $D$, and $w$, since this does not change the step ratios, and we may therefore assume that $x = 0$ and that the arrival site of step $j$ is $\sS$. By the definition of the pattern of step $j$, the statuses of the ring sites and the used bonds of $D'$ at the middle site and at the non-removed ring sites are then those of $\pi$. \emph{Step 1: splitting the integral.} We divide the factors of \eqref{e:ZD} for $D'$ into those that involve no site of $\tmpl(0)$ and those that involve at least one site of $\tmpl(0)$, and we let $n_{\mathrm{out}}$ denote the spins of the non-removed sites outside $\tmpl(0)$. Integrating first over the spins of the non-removed sites of $\tmpl(0)$, we write $Z^\tau_{D'}$ as the integral, with respect to the site measures of $n_{\mathrm{out}}$, of the product of the factors of the first kind and of a function $Q'(n_{\mathrm{out}})$. We define $Q''(n_{\mathrm{out}})$ in the same way for $D''$. The step changes only the status of $0$ and the bond $\{0, o\}$, and the factors of the first kind and the site measures outside $\tmpl(0)$ are the same for $D'$ and $D''$. These factors are nonnegative. It therefore suffices to show that $Q'(n_{\mathrm{out}}) \ge c_\pi Q''(n_{\mathrm{out}})$ for all values of $n_{\mathrm{out}}$. \emph{Step 2: the middle site and the tips.} By condition (1) of \zcref{d:pattern}, the remaining neighbors of the middle site at step $j$ are the non-removed sites among $\sE$, $\sN$, and $\sW$. In particular, the conventions for $I_{\mathrm{cur}}$ and $I_{\mathrm{next}}$ in \zcref{s:walks-steps} and in the definition of the ring functional agree. As in the proof of \zcref{l:ratios}, integrating out $n_0$ gives the factor $I_{\mathrm{cur}}$ in $Q'$ and $I_{\mathrm{next}}$ in $Q''$. Next, let $t = 2z$ be a non-removed tip, with axial site $z$. Its only neighbor in $\tmpl(0)$ is $z$, and its other factors, at most three, have the form $1 + n_t \cdot b$, where $b$ is a component of $n_{\mathrm{out}}$ or a pin. We let $A_t > 0$ denote the integral of the product of these other factors against $\lambda_{\mathrm{st}(t)}$, and $b_t$ the mean of the corresponding probability measure. Integrating out $n_t$ gives $A_t (1 + b_t \cdot n_z)$ if $\{z, t\}$ is present in $D'$, and $A_t$ otherwise. As in the proof of part (4) of \zcref{l:domainfacts}, we have $|b_t| \le 3/5$. We set $u_z = 5 b_t / 3$ if the tip bond of $z$ is present in $D'$, and $u_z = 0$ otherwise. The tips are not neighbors of each other, of the corners, or of the middle site, so that they are integrated out independently, and they give the same factors in $Q'$ and $Q''$. \emph{Step 3: the ring.} After these integrations, the remaining integrand involves the ring spins through the factors of the present ring bonds, the factors $1 + (3/5) u_z \cdot n_z$ of the axial sites whose tip bond is retained in $\pi$, and the factors of the slots of the non-removed corners. The axial sites carry no legs, since their depth is at least $2$. For the slot $k = (y, y')$ of a non-removed corner $y$, we set $m_k = \tau_{y y'}$ if $y' \notin G$, $m_k = n_{y'}$ if $\{y, y'\}$ is present in $D'$, and $m_k = 0$ otherwise. All these vectors belong to $\Ball$. When the tip bond of $z$ is retained in $\pi$ but the tip is removed, we have $u_z = 0$, and the factor of $z$ is one. Comparing with \eqref{e:ringfunctional}, we find \begin{equation*} Q'(n_{\mathrm{out}}) = K \, \mathcal Z_\pi^{\mathbf a}[I_{\mathrm{cur}}], \qquad Q''(n_{\mathrm{out}}) = K \, \mathcal Z_\pi^{\mathbf a}[I_{\mathrm{next}}], \end{equation*} where $\mathbf a = (\mathbf u, \mathbf m)$ and $K > 0$ is the product of the constants $A_t$ over the non-removed tips. By (H1), we conclude that $Q'(n_{\mathrm{out}}) \ge c_\pi Q''(n_{\mathrm{out}})$. Integrating over $n_{\mathrm{out}}$, we conclude that $Z^\tau_{D'} \ge c_\pi Z^\tau_{D''}$, which is the claim. \end{proof} \section{The trail weight and the growth hypothesis}\label{s:walks-growth} The second hypothesis bounds the number of all trails of $\Z^2$, weighted by the factors that the expansion attaches to their steps. We use the terms of \zcref{s:walks-trails,s:walks-steps} also for sequences of sites of $\Z^2$, as follows. A trail of $\Z^2$ is a sequence $w = (v_0, \dots, v_m)$ of sites of $\Z^2$ such that $v_{j - 1} \sim v_j$ for $1 \le j \le m$ and the bonds $\{v_{j - 1}, v_j\}$ are pairwise distinct, and its length is $|w| = m$. We consider such trails together with a neighbor $v_{-1}$ of $v_0$, the arrival site of $v_0$. For $1 \le j \le m$, step $j$ of $w$ is a revisit if $v_{j - 1} \in \{v_0, \dots, v_{j - 2}\}$ and a first visit otherwise. A first visit is straight if $v_j - v_{j - 1} = v_{j - 1} - v_{j - 2}$ and a turn otherwise. These notions agree with those of \zcref{s:walks-trails,s:walks-steps} when $w$ is a trail in a domain in which all sites of $w$ have status $0$, since the status of a site in $D[w_{j - 1}]$ is then the number of earlier steps that left it. We let $\FV(w)$ denote the set of first-visit steps of $w$ and $\rev(w) = m - |\FV(w)|$ the number of revisits. Let $\cs, \ct \in (0, 1]$. We define the \emph{trail weight} \begin{equation}\label{e:trailweight} \wt(w) = \prod_{j = 1}^m \wt_j, \qquad \wt_j = \begin{cases} 1/5 & \text{if step } j \text{ is a revisit}, \\ 1/(3\cs) & \text{if step } j \text{ is a straight first visit}, \\ 1/(3\ct) & \text{if step } j \text{ is a turning first visit}. \end{cases} \end{equation} The factor $1/5 = (1/3)(3/5)$ of a revisit is its weight in \eqref{e:pairwalk}, with step ratio one. The factor of a first visit is its weight $1/3$ in \eqref{e:pairwalk}, multiplied by the bound $1/c_\pi$ of \zcref{l:local} on its step ratio. \begin{hypothesis}{H2} Let $\cs, \ct \in (0, 1]$. There exist $C_{\mathrm o} < \infty$ and $\gamma_{\mathrm o} \in (0, 1)$ such that for all $n \ge 0$, all $v_0 \in \Z^2$, and all neighbors $v_{-1}$ of $v_0$, we have \begin{equation}\label{e:H2} \sum_w \wt(w) \le C_{\mathrm o} \gamma_{\mathrm o}^n, \end{equation} where the sum runs over the trails $w = (v_0, \dots, v_n)$ of $\Z^2$ of length $n$ with $v_1 \neq v_{-1}$ when $n \ge 1$, and $\wt(w)$ is computed with the arrival site $v_{-1}$. For $n = 0$, the sum consists of the length-zero trail of weight $1$. \end{hypothesis} The condition $v_1 \neq v_{-1}$ means that the first step does not reverse the direction of arrival. \zcref[S]{p:H2} proves (H2) at $(\cs, \ct) = (961/1000, 187/200)$, with $\gamma_{\mathrm o} = 2499/2500$ and $C_{\mathrm o} = 10$.