\chapter{Contraction of the boundary response}\label{s:contraction} The main result of this section is \zcref{t:contraction}, which states that Hypothesis~(P) holds if (H1), (H2), and (H3) hold at the same pair of thresholds $(\cs, \ct)$ with $\min(\cs, \ct) \ge 187/200$. Together with \zcref{t:inputs}, it gives Hypothesis~(P), the input of \zcref{t:window}. We work on one finite set $G$ and apply the pair walk identity of \zcref{p:pairwalk} to a response. The trails that return to the second site of the pair produce responses of new pairs in new domains. By \zcref{l:closure}, these pairs belong to the same class as the original pair. By (H3), their total weighted contribution is bounded by a constant smaller than one times the supremum of all such responses. The trails that end near the boundary contribute an amount that decays exponentially in the distance of the pair from $\partial_2 G$, by (H2). The boundary layer $\partial_2 G$ is the same for all domains on $G$, and this gives a bound that is uniform in $G$. The heuristics are explained in \zcref{s:overview}. As in \zcref{s:walks}, $G \subset \Z^2$ denotes a finite set and $\tau$ a pin configuration on $G$. \section{Pair classes}\label{s:contraction-pairs} We begin by defining the class of pairs whose responses we bound. \begin{definition}\label{d:pairclass} Let $D \in \Dcons(G)$ and $\mathfrak t \in \{\tcoll, \tdiag\}$. We let $\mathcal J_{\mathfrak t}(D)$ denote the set of ordered pairs $p = (u, v)$ of non-removed sites such that \begin{enumerate} \item $v - u \in \{(\pm 2, 0), (0, \pm 2)\}$ if $\mathfrak t = \tcoll$, and $v - u \in \{(\pm 1, \pm 1)\}$ if $\mathfrak t = \tdiag$; \item $u$ and $v$ have a common neighbor of status $2$; and \item each of $u$ and $v$ has status $0$, or has status $1$ and at most one present bond. \end{enumerate} We call the pairs in $\mathcal J_\tcoll(D)$ \emph{collinear} and the pairs in $\mathcal J_\tdiag(D)$ \emph{diagonal}, and we call $\mathfrak t$ the \emph{type} of $p$. For each pair $p = (u, v)$ in $\mathcal J_\tcoll(D) \cup \mathcal J_\tdiag(D)$, we fix a common neighbor $c$ of $u$ and $v$ of status $2$, and we call it the \emph{center} of $p$. \end{definition} A collinear pair has one common neighbor and a diagonal pair has two. Either common neighbor can serve as the center of a diagonal pair when both are removed. For $p = (u, v) \in \mathcal J_\tcoll(D) \cup \mathcal J_\tdiag(D)$, we define the response of $p$ and its distance from the boundary layer by \begin{equation}\label{e:pairresponse} r^\tau_D(p) = \E^\tau_D[n_u \cdot n_v] - \E^0_D[n_u \cdot n_v], \qquad d(p) = \min \bigl( \dist(u, \partial_2 G), \dist(v, \partial_2 G) \bigr). \end{equation} Since all spins have norm at most one, we have $|r^\tau_D(p)| \le 2$. The pairs of Hypothesis~(P) belong to these classes. Indeed, let $Y \subset G \setminus \partial_2 G$. Let $u, v \in G \setminus Y$ satisfy $d(u, v) = 2$ and have a common neighbor in $Y$. Recall from \zcref{s:walks-domains} that the domain $G \setminus Y$ is consistent and that its integrals are those of \eqref{e:Zpin}. The sites $u$ and $v$ have status $0$ in it, so $(u, v) \in \mathcal J_\tcoll(G \setminus Y) \cup \mathcal J_\tdiag(G \setminus Y)$. In this notation, Hypothesis~(P) asks for the bound $|r^\tau_{G \setminus Y}((u, v))| \le A \theta^{d((u, v))}$. Therefore, a bound $|r^\tau_D(p)| \le A \theta^{d(p)}$ for all $D \in \Dcons(G)$ and all $p \in \mathcal J_\tcoll(D) \cup \mathcal J_\tdiag(D)$, for all finite $G$ and all $\tau$, implies Hypothesis~(P) with the same constants. \section{Closure of the pair classes}\label{s:contraction-closure} Let $D \in \Dcons(G)$ and $\mathfrak t \in \{\tcoll, \tdiag\}$. Let $p = (u, v) \in \mathcal J_{\mathfrak t}(D)$ with center $c$, and suppose that $u \notin \partial_2 G$. We let $\hat D$ denote the domain obtained from $D$ by adding the bond $\{c, u\}$ to the used bonds. Since $c$ is removed, this bond is not present in $D$, so $\hat D$ and $D$ have the same statuses and the same present bonds. In particular, $\hat D$ is a domain, $Z^\cdot_{\hat D}[f] = Z^\cdot_D[f]$ for all $f$, and the trails in $\hat D$ from $u$ are the trails in $D$ from $u$, with the same statuses along them. The additional used bond only serves to count the used bonds at $u$. We set $T = \partial_2 G \cup \{v\}$. The assumptions of \zcref{s:walks-trails} hold for the domain $\hat D$, the site $u$, the arrival site $v_{-1} = c$ of $u$, and the site $v$. We have $u \notin T$ by construction, and \zcref{d:pairclass} makes $u$ and $v$ non-removed. The removed center $c$ makes $\{c,u\}$ absent from the present bonds. If $u$ has status $1$, then it has at most one present bond by condition (3) of \zcref{d:pairclass}. In this section, the traced domains, steps, arrival sites, remaining neighbors, domains $D^-_j$, and patterns of steps of \zcref{s:walks} refer to these choices. In \zcref{f:trail}, the pair $(u, v)$ is collinear, with center $c$. \begin{lemma}\label{l:closure} Let $D$, $p = (u, v)$, $c$, $\hat D$, and $T$ be as above, and let $w = (v_0, \dots, v_m) \in \mathcal T_{\hat D}(u, T)$. Then for $0 \le j \le m$, the following hold, where $v_{-1} = c$. \begin{enumerate} \item In $\hat D[w_j]$, every site of status $0$ other than $v_j$ has no used bond, and every site of status $1$ other than $v_j$ has exactly two used bonds. \item If $v_j$ has status $0$ in $\hat D[w_j]$, then its only used bond in $\hat D[w_j]$ is $\{v_{j - 1}, v_j\}$, and $v_{j - 1}$ has status $1$ or $2$ in $\hat D[w_j]$. \item If $j < m$, then the domain $D^-_{j + 1}$ is consistent. \item If $j < m$ and step $j + 1$ is a first visit, then the pattern of step $j + 1$ belongs to $\Pext$. Further, every pair $(a, b)$ of distinct remaining neighbors of $v_j$ at step $j + 1$ belongs to $\mathcal J_\tcoll(D^-_{j + 1}) \cup \mathcal J_\tdiag(D^-_{j + 1})$, and $v_j$ is a common neighbor of $a$ and $b$ of status $2$ in $D^-_{j + 1}$. \end{enumerate} \end{lemma} \begin{proof} Along $w$, the status of a site changes only when the site is left, from $0$ to $1$ at a first visit and from $1$ to $2$ at a revisit. Step $j$ adds the bond $\{v_{j - 1}, v_j\}$ to the used bonds, which changes the number of used bonds only at $v_{j - 1}$ and $v_j$. We prove (1) and (2) by induction on $j$. Recall that $\hat D$ differs from $D \in \Dcons(G)$ only by the used bond $\{c, u\}$, and that $c$ is removed. Every site other than $u$ therefore satisfies the conditions in (1) for $j = 0$. Suppose that $u$ has status $0$. Since $D$ is consistent, $u$ has no used bond in $D$, and its only used bond in $\hat D$ is $\{c, u\}$, whose end $c$ has status $2$. We conclude that (1) and (2) hold for $j = 0$. Suppose that (1) and (2) hold for some $j < m$. We first consider the site $v_j$. When $v_j$ has status $0$ in $\hat D[w_j]$, part (2) shows that it has status $1$ and exactly two used bonds in $\hat D[w_{j + 1}]$, namely $\{v_{j - 1}, v_j\}$ and $\{v_j, v_{j + 1}\}$. When $v_j$ has status $1$ in $\hat D[w_j]$, it is removed in $\hat D[w_{j + 1}]$. We next consider the site $v_{j + 1} \neq v_j$, which is not removed, since $\{v_j, v_{j + 1}\}$ is present in $\hat D[w_j]$. When $v_{j + 1}$ has status $0$, part (1) shows that it has no used bond in $\hat D[w_j]$. Its only used bond in $\hat D[w_{j + 1}]$ is then $\{v_j, v_{j + 1}\}$, and $v_j$ has status $1$ or $2$ in $\hat D[w_{j + 1}]$. The statuses and used bonds of the other sites stay the same at step $j + 1$, and (1) and (2) hold for $j + 1$. We now prove (3). Let $j < m$. The site $v_j$ does not belong to $T \supset \partial_2 G$, and every bond that is present in $D^-_{j + 1}$ is present in $\hat D[w_j]$. Therefore, $D^-_{j + 1}$ is a domain. In $D^-_{j + 1}$, the site $v_j$ is removed, and the statuses and used bonds of all other sites are those of $\hat D[w_j]$. By (1), $D^-_{j + 1}$ is consistent. We now prove (4). Let $j < m$ be such that step $j + 1$ is a first visit. Its passage site $x = v_j$ has status $0$ in $\hat D[w_j]$. By (2), the only used bond at $x$ is the arrival bond $\{v_{j - 1}, x\}$, and the arrival site $v_{j - 1}$ has status $1$ or $2$. The ring sites of $\tmpl(x)$ are different from $x$. By (1), every ring site of status $0$ has no used bond and every ring site of status $1$ has exactly two used bonds. Transported to $\tmpl(0)$, these are the conditions (1)--(4) of \zcref{d:pattern}, and the pattern of step $j + 1$ belongs to $\Pext$. Since the only used bond at $x$ leads to $v_{j - 1}$, the remaining neighbors of $x$ at step $j + 1$ are the non-removed neighbors of $x$ other than $v_{j - 1}$. Let $a \neq b$ be two of them. Then $d(a, b) = 2$, the sites $a$ and $b$ are not removed in $D^-_{j + 1}$, and their common neighbor $x$ is removed in $D^-_{j + 1}$. Suppose that $a$ has status $1$. By (1), it has exactly two used bonds in $\hat D[w_j]$, and $\{a, x\}$ is not one of them, since the only used bond at $x$ leads to $v_{j - 1} \neq a$. The site $a$ does not belong to $\partial_2 G$, so that its four bonds join it to sites of $G$, and the bond $\{a, x\}$ is not present in $D^-_{j + 1}$, since $x$ is removed. We conclude that $a$ has at most one present bond in $D^-_{j + 1}$. The same holds for $b$. Since $D^-_{j + 1}$ is consistent by (3), it follows that $(a, b) \in \mathcal J_\tcoll(D^-_{j + 1}) \cup \mathcal J_\tdiag(D^-_{j + 1})$. \end{proof} We record the types of the pairs in part (4). Let step $j$ be a first visit with passage site $x$ and out-neighbor $o$. Let $f$ and $g$ be the two neighbors of $x$ other than the arrival site and $o$. At a straight first visit, $f$ and $g$ lie on opposite sides of $x$, so $(f, g)$ is collinear, and $(f, o)$ and $(g, o)$ are diagonal. At a turn, one of $f$ and $g$, say $f$, is opposite to the arrival site, and $g$ is opposite to $o$; then $(f, g)$ and $(f, o)$ are diagonal, and $(g, o)$ is collinear. By \zcref{l:response}, the pair that does not contain $o$ carries the coefficient $\alpha_j \le 1/6$, and the pairs that contain $o$ carry coefficients at most $9/25$. \section{Weighted trail sums}\label{s:contraction-sums} We now define the weighted sums over trails of $\Z^2$ that will dominate the contribution of the trails that end at the second site of a pair. Let $\mathfrak t \in \{\tcoll, \tdiag\}$. In this subsection, we write $c = (0, 0)$, $u = (-1, 0)$, and $v = v^{\mathfrak t}$, where $v^\tcoll = (1, 0)$ and $v^\tdiag = (0, 1)$, so that $(u, v)$ is a pair of type $\mathfrak t$ with center $c$. We let $\mathcal T^{\mathfrak t}_n$ denote the set of trails $w = (v_0, \dots, v_n)$ of $\Z^2$ of length $n$ that do not visit $c$, with $v_0 = u$, $v_n = v$, and $v_i \neq v$ for $0 \le i < n$. We use the notions of \zcref{s:walks-growth} with the arrival site $v_{-1} = c$ of $u$. Let $w \in \mathcal T^{\mathfrak t}_n$. Recall from \zcref{s:walks-growth} that step $j$ of $w$, for $1 \le j \le n$, is the move from $v_{j - 1}$ to $v_j$, that its arrival site is $v_{j - 2}$, and that it is a revisit if $v_{j - 1} \in \{v_0, \dots, v_{j - 2}\}$. There are $|\FV(w)|$ first-visit steps and $\rev(w)$ revisits. We set $\bar\alpha = 1/6$ and $\bar\beta = 9/25$, the bounds on the coefficients in \zcref{l:response}. Let $j \in \FV(w)$. The three neighbors of $v_{j - 1}$ other than the arrival site $v_{j - 2}$ form three unordered pairs, which we call the \emph{pairs at step} $j$. We assign the coefficient $\bar\alpha$ to the pair at step $j$ that does not contain $v_j$, and the coefficient $\bar\beta$ to the two pairs that contain $v_j$. We call a pair at step $j$ \emph{dead} if it contains $c$ or a site that $w$ has left twice before step $j$ (a site that occurs twice among $v_0, \dots, v_{j - 2}$), and \emph{live} otherwise. We use the \emph{type weights} \begin{equation}\label{e:typeweights} h = (h_\tcoll, h_\tdiag) = (1, 5/4), \end{equation} we let $h(p) = h_{\mathfrak t'}$ denote the type weight of a pair $p$ of type $\mathfrak t'$, and we set $h_{\min} = 1$ and $h_{\max} = 5/4$. For $j \in \FV(w)$, we define \begin{equation}\label{e:kappa} \kappa(j) = \sum_{p \text{ live at step } j} (\text{coefficient of } p) \, h(p), \end{equation} and we set $\bar\rho_j = 1/\cs$ if step $j$ is straight and $\bar\rho_j = 1/\ct$ if it is a turn. For each $w \in \mathcal T^{\mathfrak t}_n$, let $\varsigma_w$ be an enumeration of $\FV(w)$, as in \zcref{s:walks-steps}. For $\mathcal T \subset \mathcal T^{\mathfrak t}_n$, we define \begin{equation}\label{e:mT} \mathsf m(\mathcal T) = \sum_{w \in \mathcal T} 3^{-n} \Bigl( \frac35 \Bigr)^{\rev(w)} \sum_{k = 1}^{|\FV(w)|} \kappa(\varsigma_w(k)) \prod_{l = 1}^k \bar\rho_{\varsigma_w(l)}, \qquad \mathsf m_{\mathfrak t}(n) = \mathsf m(\mathcal T^{\mathfrak t}_n), \end{equation} and for $0 < t \le 1$, we define the normalized weighted sums \begin{equation}\label{e:kernel} K_{\mathfrak t}(t) = \frac1{h_{\mathfrak t}} \sum_{n \ge 1} \mathsf m_{\mathfrak t}(n) \, t^{-n - 1} \in [0, \infty]. \end{equation} These quantities depend on $\cs$ and $\ct$ through the numbers $\bar\rho_j$, and on the family of enumerations $(\varsigma_w)$. The term of a trail $w$ in \eqref{e:mT} is the bound that \zcref{l:exposure} and \zcref{l:response} give for a trail that returns to the second site of a pair, with the response coefficients replaced by $\bar\alpha$ and $\bar\beta$ and each response replaced by its type weight. The dead pairs are left out, since in the expansion they contain a removed site and have no response. The factor $t^{-n - 1}$ in \eqref{e:kernel} accounts for the distance to $\partial_2 G$ that a pair can lose along a trail of length $n$. \begin{hypothesis}{H3} Let $\cs, \ct \in (0, 1]$, and let the sums \eqref{e:mT} be formed with $\bar\alpha = 1/6$, $\bar\beta = 9/25$, the type weights $h = (1, 5/4)$, and a fixed family of enumerations $\varsigma_w$ of the sets $\FV(w)$ of first-visit steps, chosen as in \zcref{s:tail}. There exists $K_\star < 1$ such that $K_\tcoll(1) \le K_\star$ and $K_\tdiag(1) \le K_\star$. \end{hypothesis} Since the terms of \eqref{e:kernel} are nonnegative, the bound $K_{\mathfrak t}(1) \le K_\star$ is equivalent to the bounds $\sum_{n = 1}^N \mathsf m_{\mathfrak t}(n) \le K_\star h_{\mathfrak t}$ for all $N \ge 1$, and it can be verified without knowing in advance that the series converges. Our arguments hold for every family of enumerations; the family of \zcref{s:tail} is chosen to make the sums small. \begin{lemma}\label{l:Ktheta} Let $\cs, \ct \in (0, 1]$, and assume that (H2) and (H3) hold at $(\cs, \ct)$. Then for $\mathfrak t \in \{\tcoll, \tdiag\}$, the function $K_{\mathfrak t}$ is finite and continuous on $(\gamma_{\mathrm o}, 1]$. In particular, there exists $\theta \in (\gamma_{\mathrm o}, 1)$ such that \begin{equation}\label{e:Ktheta} K_\theta = \max \bigl( K_\tcoll(\theta), K_\tdiag(\theta) \bigr) < 1. \end{equation} \end{lemma} \begin{proof} The finiteness of $K_{\mathfrak t}$ is a routine consequence of (H2). Let $w \in \mathcal T^{\mathfrak t}_n$. Every prefix product in \eqref{e:mT} is bounded by the product of $\bar\rho_j$ over all $j \in \FV(w)$, since $\bar\rho_j \ge 1$. Further, $\kappa(j) \le h_{\max} (\bar\alpha + 2 \bar\beta)$ for all $j$, and $|\FV(w)| \le n$. The term of $w$ in \eqref{e:mT} is therefore bounded by \begin{equation*} n \, h_{\max} (\bar\alpha + 2 \bar\beta) \, 3^{-n} \Bigl( \frac35 \Bigr)^{\rev(w)} \prod_{j \in \FV(w)} \bar\rho_j = n \, h_{\max} (\bar\alpha + 2 \bar\beta) \wt(w), \end{equation*} where $\wt(w)$ is the trail weight \eqref{e:trailweight} with the arrival site $c$; here we used $\wt_j = \bar\rho_j / 3$ for $j \in \FV(w)$ and $\wt_j = (1/3)(3/5)$ for a revisit. Every $w \in \mathcal T^{\mathfrak t}_n$ is a trail of $\Z^2$ of length $n$ from $u$, and its first step does not go to the arrival site $c$. Applying (H2) with $v_0 = u$ and $v_{-1} = c$, and using $h_{\mathfrak t} \ge h_{\min}$, we obtain \begin{equation}\label{e:massmajorant} \frac{\mathsf m_{\mathfrak t}(n)}{h_{\mathfrak t}} \le \frac{h_{\max}}{h_{\min}} (\bar\alpha + 2 \bar\beta) \, n \, C_{\mathrm o} \gamma_{\mathrm o}^n \le 2 n C_{\mathrm o} \gamma_{\mathrm o}^n, \end{equation} where in the last inequality we used $(5/4)(1/6 + 18/25) = 133/120 < 2$. Let $t_0 \in (\gamma_{\mathrm o}, 1]$. For $t \in [t_0, 1]$, the terms of \eqref{e:kernel} are continuous in $t$ and bounded by $2 n C_{\mathrm o} \gamma_{\mathrm o}^n t_0^{-n - 1}$, which is summable in $n$. Hence the series \eqref{e:kernel} converges uniformly on $[t_0, 1]$, and $K_{\mathfrak t}$ is finite and continuous on $(\gamma_{\mathrm o}, 1]$. By (H3), $K_{\mathfrak t}(1) \le K_\star < 1$ for $\mathfrak t \in \{\tcoll, \tdiag\}$, and the existence of $\theta$ follows from the continuity at $t = 1$. \end{proof} \section{The contraction}\label{s:contraction-proof} We now state the main result of this section. \begin{theorem}\label{t:contraction} Let $\cs, \ct \in [187/200, 1]$, and assume that (H1), (H2), and (H3) hold at $(\cs, \ct)$. Then there exist $A \ge 2$ and $\theta \in (0, 1)$ such that for every finite $G \subset \Z^2$, every pin configuration $\tau$ on $G$, every $D \in \Dcons(G)$, and every $p \in \mathcal J_\tcoll(D) \cup \mathcal J_\tdiag(D)$, we have \begin{equation}\label{e:responsebound} |r^\tau_D(p)| \le A \theta^{d(p)}. \end{equation} In particular, Hypothesis~(P) holds with these constants. \end{theorem} The proof occupies the rest of this section. We first split the response of a pair $(u, v)$ into the contributions of the trails that end at $v$ and of the trails that end in $\partial_2 G$. For a domain $D'$ and a trail $w$ in $D'$, we set $a_{D'}(w) = 3^{-|w|} (3/5)^{\rev_{D'}(w)}$, the coefficient of $w$ in \eqref{e:pairwalk}. \begin{lemma}\label{l:split} Let $D \in \Dcons(G)$, let $\mathfrak t \in \{\tcoll, \tdiag\}$, and let $p = (u, v) \in \mathcal J_{\mathfrak t}(D)$ with center $c$ and $d(p) \ge 1$. Let $\hat D$ and $T$ be as in \zcref{s:contraction-closure}, let $\mathcal T_{\mathrm{ret}}$ denote the set of trails in $\mathcal T_{\hat D}(u, T)$ that end at $v$, and let $\mathcal T_{\mathrm{pin}} = \mathcal T_{\hat D}(u, T) \setminus \mathcal T_{\mathrm{ret}}$. Then $r^\tau_D(p) = \Upsilon_{\mathrm{ret}} + \Upsilon_{\mathrm{pin}}$, where \begin{equation}\label{e:split} \begin{aligned} \Upsilon_{\mathrm{ret}} &= \sum_{w \in \mathcal T_{\mathrm{ret}}} a_{\hat D}(w) \bigl( \varrho^\tau_{\hat D}(w) - \varrho^0_{\hat D}(w) \bigr) \E^0_{\hat D[w]}[|n_v|^2], \\ \Upsilon_{\mathrm{pin}} &= \sum_{w \in \mathcal T_{\mathrm{pin}}} a_{\hat D}(w) \Bigl( \varrho^\tau_{\hat D}(w) \, \E^\tau_{\hat D[w]}[n_{v_{|w|}} \cdot n_v] - \varrho^0_{\hat D}(w) \, \E^0_{\hat D[w]}[n_{v_{|w|}} \cdot n_v] \Bigr). \end{aligned} \end{equation} Moreover, $\E^0_{\hat D[w]}[|n_v|^2] \le 1$ for all $w \in \mathcal T_{\mathrm{ret}}$, and every trail in $\mathcal T_{\mathrm{pin}}$ ends in $\partial_2 G$. \end{lemma} \begin{proof} Since $d(p) \ge 1$, the sites $u$ and $v$ do not belong to $\partial_2 G$, and the construction of \zcref{s:contraction-closure} applies. The domains $\hat D$ and $D$ define the same integrals, and we have \begin{equation*} r^\tau_D(p) = \E^\tau_{\hat D}[n_u \cdot n_v] - \E^0_{\hat D}[n_u \cdot n_v]. \end{equation*} We expand the two terms with \zcref{p:pairwalk}, applied to $\hat D$, $u$, $v$, and $T$. The two expansions run over the same set $\mathcal T_{\hat D}(u, T)$, with the same coefficients $a_{\hat D}(w)$. Let $w \in \mathcal T_{\mathrm{ret}}$. By condition (3) of \zcref{d:pairclass}, the site $v$ has status $0$, or status $1$ and at most one present bond. In the first case, and in the second case when $v$ has exactly one present bond, \zcref{p:pairwalk} shows that the last factor of the term of $w$ is $\E^\cdot_{\hat D[w]}[|n_v|^2] \in \{1, 3/5\}$, with the same value for $\cdot = \tau$ and $\cdot = 0$. When $v$ has status $1$ and no present bond, the set $\mathcal T_{\mathrm{ret}}$ is empty, since the last step of a trail that ends at $v$ uses a bond at $v$ that is present in $\hat D$. Subtracting the two expansions gives $r^\tau_D(p) = \Upsilon_{\mathrm{ret}} + \Upsilon_{\mathrm{pin}}$. Finally, a trail in $\mathcal T_{\mathrm{pin}}$ ends in $T \setminus \{v\} \subset \partial_2 G$. \end{proof} The next lemma compares the returning trails in $\hat D$ with the trails of $\mathcal T^{\mathfrak t}_n$ that enter the sums \eqref{e:mT}. This comparison is the subtle point of the argument. A step of a returning trail can be a revisit in $\hat D$ and a first visit in $\Z^2$, a pair can be absent from the expansion and live in \eqref{e:mT}, and the coefficients of \zcref{l:response} must be compared with $\bar\alpha$ and $\bar\beta$; in each case, the trail sums \eqref{e:mT} give the larger contribution. \begin{lemma}\label{l:transport} Let $D$, $\mathfrak t$, $p = (u, v)$, $c$, $\hat D$, and $\mathcal T_{\mathrm{ret}}$ be as in \zcref{l:split}. Then there exists an isometry of $\Z^2$ that maps $c$ to $(0, 0)$, $u$ to $(-1, 0)$, and $v$ to $v^{\mathfrak t}$. Fix such an isometry, and for $w \in \mathcal T_{\mathrm{ret}}$, let $\tilde w = (\tilde v_0, \dots, \tilde v_n)$ denote the image of $w = (v_0, \dots, v_n)$ under it. Then the following hold. \begin{enumerate} \item We have $\tilde w \in \mathcal T^{\mathfrak t}_n$, and the map $w \mapsto \tilde w$ is injective on $\mathcal T_{\mathrm{ret}}$. \item The set $\FV(w)$ of first-visit steps of $w$ in $\hat D$ is contained in the set $\FV(\tilde w)$ of first-visit steps of $\tilde w$ in the sense of \zcref{s:walks-growth}, and $\rev_{\hat D}(w) \ge \rev(\tilde w)$. A step $j \in \FV(w)$ is straight for $w$ if and only if it is straight for $\tilde w$. \item Let $j \in \FV(w)$. The image of every pair of distinct remaining neighbors of $v_{j - 1}$ at step $j$ is a live pair at step $j$ of $\tilde w$, of the same type. It contains $\tilde v_j$ if and only if the original pair contains $v_j$. \end{enumerate} \end{lemma} \begin{proof} The vectors $c - u$ and $v - c$ are unit vectors, which are equal if $\mathfrak t = \tcoll$ and orthogonal if $\mathfrak t = \tdiag$. Hence an isometry with the stated properties exists. It preserves neighbors, the distinction between straight steps and turns, and the types of pairs. Further, it maps the arrival site $c$ of $u$ to the arrival site $(0, 0)$ of $(-1, 0)$. For (1), observe that every site of $w$ is non-removed in $\hat D$, since $v_0 = u$ and $v_j$ is an end of a bond that is present in $\hat D[w_{j - 1}]$ for $j \ge 1$. In particular, $w$ does not visit $c$. The bonds of $w$ are pairwise distinct, $v_n = v$, and $v_i \neq v$ for $i < n$. Hence $\tilde w \in \mathcal T^{\mathfrak t}_n$. An isometry is injective, and so is the map $w \mapsto \tilde w$. For (2), let $1 \le j \le n$ be such that step $j$ of $\tilde w$ is a revisit, so that $v_{j - 1} = v_i$ for some $i \le j - 2$. Then step $i + 1$ of $w$ left $v_{j - 1}$ and raised its status, so $v_{j - 1}$ has status $1$ in $\hat D[w_{j - 1}]$, and step $j$ is a revisit of $w$ in $\hat D$. This proves the inclusion $\FV(w) \subset \FV(\tilde w)$, and the inequality follows from $\rev_{\hat D}(w) = n - |\FV(w)|$ and $\rev(\tilde w) = n - |\FV(\tilde w)|$. The steps in $\FV(\tilde w) \setminus \FV(w)$ leave sites that have status $1$ in $\hat D$ and that $w$ has not left before. The statement on straight steps holds since the isometry preserves them. For (3), let $j \in \FV(w)$. Let $a \neq b$ be remaining neighbors of $x = v_{j - 1}$ at step $j$. They are neighbors of $x$ other than the arrival site $v_{j - 2}$, so their images form a pair at step $j$ of $\tilde w$. This pair does not contain $(0, 0)$, since $c$ is removed and remaining neighbors are not. Suppose that it contains a site that $\tilde w$ has left twice before step $j$. The corresponding site of $w$ has then been left twice by $w$ before step $j$, so it has status $2$ in $\hat D[w_{j - 1}]$ and is not a remaining neighbor. This contradiction shows that the pair is live. The remaining statements hold since the isometry preserves types and maps $v_j$ to $\tilde v_j$. \end{proof} We now bound the contribution of the returning trails. \begin{lemma}\label{l:compare} Let $\cs, \ct \in [187/200, 1]$, and assume that (H1) holds at $(\cs, \ct)$. Let $\mathcal M \ge 0$ and $\theta \in (0, 1]$ be such that \begin{equation}\label{e:compare-hyp} |r^\tau_{D'}(p')| \le \mathcal M h(p') \theta^{d(p')} \quad \text{for all } D' \in \Dcons(G) \text{ and all } p' \in \mathcal J_\tcoll(D') \cup \mathcal J_\tdiag(D'). \end{equation} Let $D$, $\mathfrak t$, $p = (u, v)$, $\hat D$, $\mathcal T_{\mathrm{ret}}$, and $\Upsilon_{\mathrm{ret}}$ be as in \zcref{l:split}, and let $d = d(p)$. Then for all $n \ge 1$, \begin{equation}\label{e:compare} \sum_{\substack{w \in \mathcal T_{\mathrm{ret}} \\ |w| = n}} a_{\hat D}(w) \bigl| \varrho^\tau_{\hat D}(w) - \varrho^0_{\hat D}(w) \bigr| \le \mathcal M \theta^{d - n} \mathsf m_{\mathfrak t}(n). \end{equation} If $K_{\mathfrak t}(\theta) < \infty$, then $|\Upsilon_{\mathrm{ret}}| \le \mathcal M h_{\mathfrak t} \theta^{d + 1} K_{\mathfrak t}(\theta)$. \end{lemma} \begin{proof} Let $w = (v_0, \dots, v_n) \in \mathcal T_{\mathrm{ret}}$, and let $\tilde w$ be as in \zcref{l:transport}. We let $\FV(w)$ denote the set of first-visit steps of $w$ in $\hat D$. For $j \in \FV(\tilde w)$, we write $\kappa(j)$ and $\bar\rho_j$ for the quantities of \zcref{s:contraction-sums}, computed for $\tilde w$. \emph{Step 1: the step bounds.} Let $j \in \FV(w)$. By part (2) of \zcref{l:transport}, step $j$ of $\tilde w$ is a first visit of the same type as step $j$ of $w$. By part (4) of \zcref{l:closure}, the pattern $\pi$ of step $j$ belongs to $\Pext$. Using \zcref{l:local} and the assumption that $\cs, \ct \in [187/200, 1]$, we obtain \begin{equation*} \rho^\tau_j \le \frac1{c_\pi} = \bar\rho_j, \qquad 1 \le \bar\rho_j \le \frac{200}{187}. \end{equation*} Therefore, \zcref{l:response} applies at step $j$, and \zcref{l:exposure} applies with these numbers $\bar\rho_j$, $j \in \FV(w)$. \emph{Step 2: the responses at a step.} Let $j \in \FV(w)$, let $x = v_{j - 1}$, and let $o = v_j$. We define $\kappa'(j)$ as the sum, over the unordered pairs $\{a, b\}$ of distinct remaining neighbors of $x$ at step $j$, of $\bar\alpha \, h((a, b))$ if $o \notin \{a, b\}$ and of $\bar\beta \, h((a, b))$ if $o \in \{a, b\}$. By part (4) of \zcref{l:closure}, $D^-_j \in \Dcons(G)$, and each such pair belongs to $\mathcal J_\tcoll(D^-_j) \cup \mathcal J_\tdiag(D^-_j)$. By \eqref{e:compare-hyp}, we have $|r_{ab}| \le \mathcal M h((a, b)) \theta^{d((a, b))}$. The sites $a$ and $b$ are neighbors of $x$, and the prefix $w_{j - 1}$ joins $u$ to $x$ in $j - 1$ steps. Thus both $d(u, a)$ and $d(u, b)$ are at most $j \le n$. Since $z \mapsto \dist(z, \partial_2 G)$ is $1$-Lipschitz, we obtain \begin{equation*} d((a, b)) \ge \dist(u, \partial_2 G) - n \ge d - n, \qquad \theta^{d((a, b))} \le \theta^{d - n}. \end{equation*} \zcref[S]{l:response} bounds $|\Gamma_j|$ by the sum of $|r_{ab}|$ times $\alpha_j \le \bar\alpha$ for the pair without $o$, and times $\beta_j \le \bar\beta$ or $\beta'_j \le \bar\beta$ for the pairs with $o$. Therefore, \begin{equation}\label{e:compare-Gamma} |\Gamma_j| \le \mathcal M \theta^{d - n} \kappa'(j) \le \mathcal M \theta^{d - n} \kappa(j), \end{equation} where the second inequality follows from part (3) of \zcref{l:transport}, since $\kappa'(j)$ is a sum of some of the nonnegative terms of $\kappa(j)$. \emph{Step 3: the enumeration.} Let $q = |\FV(\tilde w)|$, and let $\varsigma_{\tilde w}$ be the enumeration of $\FV(\tilde w)$ in \eqref{e:mT}. We list the elements of $\FV(w) \subset \FV(\tilde w)$ in the order in which they occur in $\varsigma_{\tilde w}(1), \dots, \varsigma_{\tilde w}(q)$, and we let $\varsigma$ denote the resulting enumeration of $\FV(w)$. Let $1 \le k \le |\FV(w)|$, and let $k'$ be such that $\varsigma_{\tilde w}(k') = \varsigma(k)$. Then $\{\varsigma(1), \dots, \varsigma(k)\} \subset \{\varsigma_{\tilde w}(1), \dots, \varsigma_{\tilde w}(k')\}$, and since $\bar\rho_j \ge 1$ for all $j \in \FV(\tilde w)$, we have \begin{equation}\label{e:compare-prefix} \prod_{l = 1}^k \bar\rho_{\varsigma(l)} \le \prod_{l = 1}^{k'} \bar\rho_{\varsigma_{\tilde w}(l)}. \end{equation} \emph{Step 4: conclusion.} By \zcref{l:exposure} with the enumeration $\varsigma$, \eqref{e:compare-Gamma}, and \eqref{e:compare-prefix}, we have \begin{equation*} \bigl| \varrho^\tau_{\hat D}(w) - \varrho^0_{\hat D}(w) \bigr| \le \mathcal M \theta^{d - n} \sum_{k = 1}^{|\FV(w)|} \kappa(\varsigma(k)) \prod_{l = 1}^k \bar\rho_{\varsigma(l)} \le \mathcal M \theta^{d - n} \sum_{k' = 1}^q \kappa(\varsigma_{\tilde w}(k')) \prod_{l = 1}^{k'} \bar\rho_{\varsigma_{\tilde w}(l)}. \end{equation*} In the second inequality, the term $k$ on the left is bounded by the term $k'$ with $\varsigma_{\tilde w}(k') = \varsigma(k)$ on the right, and the other terms on the right are nonnegative. By part (2) of \zcref{l:transport}, we also have $a_{\hat D}(w) \le 3^{-n} (3/5)^{\rev(\tilde w)}$. Multiplying the two bounds, we find that $a_{\hat D}(w) |\varrho^\tau_{\hat D}(w) - \varrho^0_{\hat D}(w)|$ is bounded by $\mathcal M \theta^{d - n}$ times the term of $\tilde w$ in \eqref{e:mT}. Summing over $w \in \mathcal T_{\mathrm{ret}}$ with $|w| = n$ and using part (1) of \zcref{l:transport}, we obtain \eqref{e:compare}. Finally, if $K_{\mathfrak t}(\theta) < \infty$, \zcref{l:split} and \eqref{e:compare} imply that \begin{equation*} |\Upsilon_{\mathrm{ret}}| \le \sum_{n \ge 1} \mathcal M \theta^{d - n} \mathsf m_{\mathfrak t}(n) = \mathcal M \theta^{d + 1} \sum_{n \ge 1} \mathsf m_{\mathfrak t}(n) \theta^{-n - 1} = \mathcal M h_{\mathfrak t} \theta^{d + 1} K_{\mathfrak t}(\theta), \end{equation*} where we used $\E^0_{\hat D[w]}[|n_v|^2] \le 1$ and the definition \eqref{e:kernel}. \end{proof} We now bound the contribution of the trails that end in $\partial_2 G$. \begin{lemma}\label{l:pinpart} Let $\cs, \ct \in (0, 1]$, and assume that (H1) and (H2) hold at $(\cs, \ct)$. Let $D$, $\mathfrak t$, $p = (u, v)$, $c$, $\hat D$, $\mathcal T_{\mathrm{pin}}$, and $\Upsilon_{\mathrm{pin}}$ be as in \zcref{l:split}, and let $d = d(p)$. Then \begin{equation}\label{e:pinpart} |\Upsilon_{\mathrm{pin}}| \le C_{\mathrm{pin}} \gamma_{\mathrm o}^d, \qquad C_{\mathrm{pin}} = \frac{2 C_{\mathrm o}}{1 - \gamma_{\mathrm o}}. \end{equation} \end{lemma} \begin{proof} Since all spins have norm at most one, we have $|\E^\cdot_{\hat D[w]}[n_{v_{|w|}} \cdot n_v]| \le 1$, and \eqref{e:split} gives \begin{equation}\label{e:pinpart-start} |\Upsilon_{\mathrm{pin}}| \le \sum_{w \in \mathcal T_{\mathrm{pin}}} a_{\hat D}(w) \bigl( \varrho^\tau_{\hat D}(w) + \varrho^0_{\hat D}(w) \bigr). \end{equation} Let $w = (v_0, \dots, v_n) \in \mathcal T_{\mathrm{pin}}$. We regard $w$ as a trail of $\Z^2$ with the arrival site $c$ of $u$, and we let $\wt(w)$ be its trail weight \eqref{e:trailweight}. We claim that \begin{equation}\label{e:pinpart-claim} a_{\hat D}(w) \, \varrho^\cdot_{\hat D}(w) \le \wt(w), \qquad \cdot \in \{\tau, 0\}. \end{equation} To prove \eqref{e:pinpart-claim}, recall from part (1) of \zcref{l:ratios} that the step ratios of the revisits in $\hat D$ equal one. Let $j$ be a first-visit step of $w$ in $\hat D$. By part (4) of \zcref{l:closure} and \zcref{l:local}, we have $\rho^\tau_j \le 1/c_\pi$, where $\pi$ is the pattern of step $j$, and \zcref{l:response} gives $\rho^0_j \le 1 \le 1/c_\pi$. We conclude that $a_{\hat D}(w) \varrho^\cdot_{\hat D}(w)$ is bounded by the product over the steps $j$ of $w$ of the factor $1/5$ if step $j$ is a revisit in $\hat D$, and $1/(3c_\pi)$ if it is a first visit in $\hat D$. We now compare these factors with $\wt_j$. When step $j$ leaves a site that $w$ has left before, it is a revisit in $\hat D$, and the two factors are $1/5$. Otherwise, $\wt_j = 1/(3\cs)$ or $1/(3\ct)$, according to the type of step $j$, and the factor above is either $1/(3c_\pi) = \wt_j$ or $1/5 < 1/3 \le \wt_j$. This gives \eqref{e:pinpart-claim}. The trails in $\mathcal T_{\mathrm{pin}}$ are distinct trails of $\Z^2$ from $u$. Their first step avoids the removed site $c$, and they end in $\partial_2 G$. Their length is therefore at least $\dist(u, \partial_2 G) \ge d$. Inserting \eqref{e:pinpart-claim} into \eqref{e:pinpart-start} and applying (H2) with $v_0 = u$ and $v_{-1} = c$, we obtain \begin{equation*} |\Upsilon_{\mathrm{pin}}| \le 2 \sum_{n \ge d} C_{\mathrm o} \gamma_{\mathrm o}^n = C_{\mathrm{pin}} \gamma_{\mathrm o}^d, \end{equation*} which proves \eqref{e:pinpart}. \end{proof} \begin{proof}[Proof of \zcref{t:contraction}] Choose $\theta \in (\gamma_{\mathrm o}, 1)$ as in \zcref{l:Ktheta}, so that $K_\theta < 1$. This choice uses only the functions $K_\tcoll$ and $K_\tdiag$ and the constant $\gamma_{\mathrm o}$. Let $G \subset \Z^2$ be finite. We define the weighted supremum \begin{equation}\label{e:MG} \mathcal M_G = \sup_{\tau, D, p} \frac{|r^\tau_D(p)|}{h(p) \theta^{d(p)}}, \end{equation} where the supremum runs over all pin configurations $\tau$ on $G$, all $D \in \Dcons(G)$, and all $p \in \mathcal J_\tcoll(D) \cup \mathcal J_\tdiag(D)$, and where the supremum of the empty set is $0$. Since $|r^\tau_D(p)| \le 2$, $h(p) \ge 1$, and $d(p)$ is bounded by the diameter of $G$, the supremum $\mathcal M_G$ is finite. We work with a single set $G$ because all responses that occur below have the same boundary layer $\partial_2 G$, which makes the following bound uniform. Let $\tau$, $D$, and $p = (u, v) \in \mathcal J_{\mathfrak t}(D)$ be as in \eqref{e:MG}, and let $d = d(p)$. When $d = 0$, we have \begin{equation*} \frac{|r^\tau_D(p)|}{h(p) \theta^{d(p)}} \le \frac2{h_{\min}}. \end{equation*} Suppose now that $d \ge 1$. By the definition of $\mathcal M_G$, the bound \eqref{e:compare-hyp} holds with $\mathcal M = \mathcal M_G$, for the pin configuration $\tau$. Since $\cs, \ct \in [187/200, 1]$, we may apply \zcref{l:split}, \zcref{l:compare}, and \zcref{l:pinpart}, and we obtain \begin{equation*} |r^\tau_D(p)| \le |\Upsilon_{\mathrm{ret}}| + |\Upsilon_{\mathrm{pin}}| \le \mathcal M_G h_{\mathfrak t} \theta^{d + 1} K_{\mathfrak t}(\theta) + C_{\mathrm{pin}} \gamma_{\mathrm o}^d \le K_\theta \mathcal M_G h_{\mathfrak t} \theta^d + C_{\mathrm{pin}} \theta^d, \end{equation*} where in the last inequality we used $\theta \le 1$, \eqref{e:Ktheta}, and $\gamma_{\mathrm o} < \theta$. Dividing by $h(p) \theta^d = h_{\mathfrak t} \theta^d \ge h_{\min} \theta^d$, we obtain \begin{equation*} \frac{|r^\tau_D(p)|}{h(p) \theta^{d(p)}} \le K_\theta \mathcal M_G + \frac{C_{\mathrm{pin}}}{h_{\min}}. \end{equation*} Taking the supremum over $\tau$, $D$, and $p$, we find that \begin{equation}\label{e:Mcontract} \mathcal M_G \le \max \Bigl( \frac2{h_{\min}}, \, K_\theta \mathcal M_G + \frac{C_{\mathrm{pin}}}{h_{\min}} \Bigr). \end{equation} Since $\mathcal M_G < \infty$ and $K_\theta < 1$, the bound \eqref{e:Mcontract} implies that \begin{equation*} \mathcal M_G \le \frac1{h_{\min}} \max \Bigl( 2, \frac{C_{\mathrm{pin}}}{1 - K_\theta} \Bigr). \end{equation*} Using $h(p) \le h_{\max}$ and the definition \eqref{e:MG}, we obtain \eqref{e:responsebound} with \begin{equation*} A = \frac{h_{\max}}{h_{\min}} \max \Bigl( 2, \frac{C_{\mathrm{pin}}}{1 - K_\theta} \Bigr). \end{equation*} The constant $A$ is independent of $G$ and $\tau$, and $A \ge 2$. Finally, \zcref{s:contraction-pairs} shows that \eqref{e:responsebound} for all finite $G$ implies Hypothesis~(P) with the same constants $A$ and $\theta$. This completes the proof. \end{proof}