% % \providecommand{\bGM}{\beta_{\mathrm{GM}}} \providecommand{\Floc}{F_1^{\mathrm{loc}}} \providecommand{\cU}{\mathcal U} \providecommand{\sF}{\sigma_F} \providecommand{\epsE}{\varepsilon_E} \providecommand{\Funit}{F_{\mathrm{unit}}} \providecommand{\Ad}{\mathrm{A}} \providecommand{\RNF}{R_{\mathrm{NF}}} \providecommand{\thNF}{\theta_{\mathrm{NF}}} \providecommand{\cNF}{c_3^{\mathrm{NF}}} \providecommand{\fS}{\mathfrak S} \providecommand{\Glong}{\Gamma_{\mathrm{long}}} \providecommand{\Nlong}{\mathcal N^{\mathrm{long}}} \subsection{Frozen flux costs area: group charge and the count} \label{sec:7.8}\label{ssec:7.8} In this subsection we prove that the partition function at fixed class $r$ carries the area weight of $r$ (Proposition~\ref{prop:7.57}). The bounds of this subsection are the only ones that use the handles of the insulating set and (NF$_*$), which holds by (R10) of Theorem~\ref{thm:3.20}. In a configuration of class $r$, a group of wall polymers realizing $r$ is re-expanded as a seed at $X=\emptyset$, and its entropy is controlled by Proposition~\ref{prop:7.56}. \paragraph{Polymers, realizations and the reference realization.} Fix $r\in\Glong\setminus\{0\}$ with $Z_r\neq0$. For a configuration $N$ of $Z_r$ (Definition~\ref{def:7.27}), $\mathcal P(N)$ is the set of its wall polymers, and we put $N_{\mathcal S}:=N\one_{E(\mathcal S)}$ for $\mathcal S\in\mathcal P(N)$ and $N_R:=\sum_{\mathcal S\in R}N_{\mathcal S}$, $q_R:=[N_R]$, $|R|:=\|N_R\|_1$ for $R\subseteq\mathcal P(N)$. By Proposition~\ref{prop:7.9}\,(a), each wall bond lies in exactly one wall polymer. Hence $\sum_{\mathcal S\in\mathcal P(N)}N_{\mathcal S}=N$ and $q_{\mathcal P(N)}=r$, and $q_R\in\Glong$ by Lemma~\ref{lem:7.45}\,(a). A family $R$ \emph{realizes} $r$ if $q_R=r$, and is \emph{irreducible} if no nonempty $T\subseteq R$ has $q_T=0$. An inclusion-minimal realizing subfamily of $\mathcal P(N)$ is irreducible: if $\emptyset\neq T\subseteq R$ had $q_T=0$, then $T\neq R$ as $r\neq0$, and $R\setminus T$ would realize $r$. Let $\fS(r)$ be the set of families $R$, with their wall configurations, that occur as irreducible realizations of $r$ in some configuration of $Z_r$. We fix $R_0\in\fS(r)$ of least area by a deterministic choice, so that $|R|\ge|R_0|\ge\Ad(r)$ for all $R\in\fS(r)$, the last inequality by Lemma~\ref{lem:7.46}. A polymer is determined by its wall interfaces and their values, since closures and links are functions of the walls. Thus $R$ is determined by $N_R$: the wall components of $N_R$ are those of the members of $R$, closures from different members are not linked (they are distinct wall polymers of one configuration), so the wall polymers of $N_R$ are exactly the members of $R$. We also set $\nu(R):=\sum_{I\subseteq\supp N_R}N_I^2$, $c_w:=c'/16$, $\lambda_E:=2c_w\sF=\frac{c'}8\sF$ and $\Theta_4(r):=\sum_{R\in\fS(r)}e^{-\lambda_E|R|}\prod_{\mathcal S\in R}e^{-(c'/2)\Floc(\mathcal S)}$. The notion of c-connectedness is that of Definition~\ref{def:3.17}, and distances between bond sets are Euclidean. \begin{lemma}\label{lem:7.54} Let $R\in\fS(r)$. Then $N_R-N_{R_0}=\sum_k\eps_k\nabla1_{P_k}$ with $\eps_k\in\{-1,1\}$, where (i) each $P_k$ is a nonempty c-connected cell set with c-connected complement in its component $\mathfrak K$ of $\Gp$; (ii) $\partial_{\Gp}P_k\subseteq\supp N_R\cup\supp N_{R_0}$; (iii) $\sum_k|\partial_{\Gp}P_k|=\|N_R-N_{R_0}\|_1\le|R|+|R_0|\le2|R|$; and (iv) every $P_k$ is joined to $\supp N_{R_0}$ by a chain $P_k=P_{k_0},\dots,P_{k_j}$ with $\dist(\partial P_{k_{i-1}},\partial P_{k_i})\le g$ and $\dist(\partial P_{k_j},\supp N_{R_0})\le g$. \end{lemma} \begin{proof} (1) Since $[N_R-N_{R_0}]=q_R-q_{R_0}=0$, Definition~\ref{def:7.27} gives $N_R-N_{R_0}=\nabla f$ with $f:V\to\mathbb Z$ constant on each cell. (2) By the coarea formula for the level sets of $f$, $\nabla f=\sum_t\pm\nabla1_{U_t}$ with $\sum_t|\partial U_t|=\|\nabla f\|_1$. We split each $U_t$ into its c-components $W$ and write $\nabla1_W=-\sum_i\nabla1_{C_i}$ over the c-components $C_i$ of $\mathfrak K\setminus W$, where $\mathfrak K$ is the component of $W$. Since each $C_i$ is c-connected with c-connected complement and the cuts partition, (i) to (iii) hold. (3) For (iv) we use irreducibility, which forbids groups of pieces far from $R_0$, since these would be sub-families of $R$ of class zero. We group the pieces into clusters by the relation ``cuts within distance $g$''. Let $\mathcal V$ be a cluster no cut of which is within $g$ of $\supp N_{R_0}$, and let $B:=\bigcup_{k\in\mathcal V}\partial P_k$. Within distance $g$ of $B$, $N_{R_0}=0$ and the other pieces contribute nothing. It follows that $N_R=N_{\mathcal V}:=\sum_{k\in\mathcal V}\eps_k\nabla1_{P_k}$ in this region, and each wall bond of $N_R$ within distance $g$ of $B$ lies in $B$. Two wall bonds of one polymer are joined by a chain of wall bonds of that polymer with consecutive gaps at most $g$, since walls of one $S_\beta$ have gaps less than $2L'+2D_E$ and linked closures add at most $2D_E+2$. Hence the polymers of $R$ meeting $B$ form a sub-family $T$ with $N_T=N_R1_B=N_{\mathcal V}$, the gradient of a cell-constant function, and therefore $q_T=0$. By irreducibility $T=\emptyset$, so that $N_{\mathcal V}=0$. By (iii) the terms of $N_R-N_{R_0}$ do not cancel, and hence neither do those of the sub-sum: $\|N_{\mathcal V}\|_1=\sum_{k\in\mathcal V}|\partial P_k|$. So $N_{\mathcal V}=0$ forces $\partial P_k=\emptyset$ for every $k\in\mathcal V$. This is impossible, since by (i) each $P_k$ is a nonempty proper cell set of its component $\mathfrak K$. Therefore no such cluster exists: every cluster has a cut within $g$ of $\supp N_{R_0}$, and chaining inside the cluster gives (iv). \end{proof} For a finite set $B$ of interfaces and $s>0$, the $s$-clusters of $B$ and their number $\varphi_s(B)$ are those of Definition~\ref{def:3.17} (for a cut, of its set of interfaces), and $\mathrm{MST}(B)$ denotes a minimal spanning tree of the complete graph on $B$ with edge lengths $\ell_e$ the distances of interfaces. Let $j(\ell):=\#\{j\ge0:2^jL<\ell\}$. \begin{lemma}[Kruskal's identity]\label{lem:7.55} For every $s>0$, $\#\{e\in\mathrm{MST}(B):\ell_e>s\}=\varphi_s(B)-1$, and $\sum_{e\in\mathrm{MST}(B)}j(\ell_e)=\sum_{j\ge0}(\varphi_{2^jL}(B)-1)$. \end{lemma} \begin{proof} The edges of $\mathrm{MST}(B)$ of length $\le s$ form a spanning forest of the threshold graph $\{\ell\le s\}$, since Kruskal's algorithm processes them first. Thus they number $|B|-\varphi_s(B)$. For the second identity we sum $1[\ell_e>2^jL]$ over $j\ge0$ and $e$. \end{proof} Let $\kappa_{\mathrm{NF}}:=4(c_g+4)/c_g^2$, and let $R_{\mathrm{NF}}$, $\thNF$, $\cNF$, $J$ be as in Definition~\ref{def:3.17}. Applying (R10) in the form of Proposition~\ref{prop:3.19} to the pieces $P_k$ of Lemma~\ref{lem:7.54}, we obtain \begin{equation}\label{eq:7.8.NF} J(P_k)\le\thNF\,\Funit\,|\partial_{\Gp}P_k|/L^2 , \end{equation} with the cut measured in fine bonds, of which a coarse face carries at least $r_0^2$. Here $J$ depends only on the cut. \begin{proposition}[The count]\label{prop:7.56} For $\bGM\ge\bGMo$, every $r\in\Glong\setminus\{0\}$ and every $\nu\ge1$, $\#\{R\in\fS(r):\nu(R)=\nu\}\le\exp((c'/2)\Funit\,\nu)$. Consequently $\Theta_4(r)\le\frac14$. \end{proposition} \begin{proof} The idea is that, by Kruskal's identity, specifying the positions of the cut patches costs, per dyadic scale, the number of clusters minus one. Below $R_{\mathrm{NF}}$ this is at most $\log_2(R_{\mathrm{NF}}/L)$ per interface, within the factor $e^{(c'/2)\Funit}$, and above $R_{\mathrm{NF}}$ it is $J$, bounded by the cut \eqref{eq:7.8.NF}. We first bound the count and then prove the second assertion, $\Theta_4\le\frac14$. Fix $R$ with $\nu(R)=\nu$. Let $S$, $S_0$ be the interface sets of $\supp N_R$ and $\supp N_{R_0}$, put $n:=|S|\le\nu$, $n_0:=|S_0|$ and $T:=S\cup S_0$, and let $I_k$ be the interface set of $\partial P_k$. An interface has between $c_g'L^2$ and $4L^2$ bonds by (R8) and since block sides are less than $2L$, and $|R_0|\le|R|\le4L^2\nu$. Hence $n_0\le(4/c_g')\nu$, and by Lemma~\ref{lem:7.54}(iii) $\sum_k|\partial P_k|\le8L^2\nu$, $\sum_k|I_k|\le(8/c_g')\nu\le\kappa_{\mathrm{NF}}\nu$. The last inequality follows from $8/c_g'\le128/(9c_g)\le4(c_g+4)/c_g^2$, which holds since $c_g'\ge\frac9{16}c_g$ by Section~\ref{ssec:D.5}(e) and $c_g\le\frac14$. Since a bond where $N_R\neq N_{R_0}$ lies on some $\partial P_k$, we have $S\subseteq S_0\cup\bigcup_kI_k\subseteq T$. \emph{Step 1 (a spanning forest).} For each $k$ let $t_k$ be the least $j$ for which a chain $P_k=P_{k_0},\dots,P_{k_j}$ as in Lemma~\ref{lem:7.54}(iv) exists. Let $H$ be the graph on $T$ whose edges are the edges of the $\mathrm{MST}(I_k)$ and, for each $k$, one parent edge at distance at most $2g$: from an interface of $I_k$ to one of $S_0$ if $t_k=0$, and otherwise to one of $I_{k'}$, where $P_{k'}$ is the second piece of a chain of length $t_k$, so that $t_{k'}=t_k-1$. Here we use that a bond and its dual coarse face are within distance $\sqrt3\,r_0\le g/2$. Each node of $T\setminus S_0$ lies in some $I_k$, and $t$ decreases along parent edges. Hence the parent edges lead to $S_0$. Consequently $H$ contains a spanning forest $F$ of $T$ with roots in $S_0$, as one sees by adding a virtual node joined to all of $S_0$ and removing it from a spanning tree. Let $B:=\sum_{e\in F}j(\ell_e)\le\sum_{e\in H}j(\ell_e)$. \emph{Step 2 (encoding).} We record the roots (at most $2^{n_0}$ choices), the shape of the plane forest on $|T|\le n+n_0$ nodes (at most $4^{n+n_0}$), and for each child the number $j_e$ and the child among the at most $C_9(2^{j_e}+1)^3\le8C_9\,8^{j_e}$ interfaces within distance $2^{j_e}L$ of its parent. Sequences of $|F|$ nonnegative integers with sum at most $B_{\max}$ (the bound of Step~3, a function of $\nu$ only) number at most $2^{B_{\max}+|F|}$. We then record $S\subseteq T$ (at most $2^{n+n_0}$) and the values $N_I$ with $\sum N_I^2=\nu$ (at most $4^\nu$, counting signs and compositions of $\sum|N_I|\le\nu$). Since $R$ is determined by $N_R$, we obtain $\ln\#\{R:\nu(R)=\nu\}\le C(n+n_0+\nu)+B_{\max}\ln16$ with $C:=\ln(64C_9)+2$. \emph{Step 3 (the bound on $B$).} By Lemma~\ref{lem:7.55}, $\sum_{e\in\mathrm{MST}(I_k)}j(\ell_e)=\sum_{j\ge0}(\varphi_{2^jL}(I_k)-1)$. The scales with $2^jL