\chapter{The computer-assisted inputs}\label{s:computer} In this section, we prove \zcref{t:inputs}, according to which Hypotheses (H1), (H2), and (H3) hold at the thresholds $(\cs, \ct) = (961/1000, 187/200)$, and in \zcref{s:computer-proofmain} we complete the proof of \zcref{t:main}. Together with \zcref{t:contraction}, \zcref{t:inputs} gives Hypothesis~(P), and the remaining implications of our argument were proved in \zcref{s:quantum,s:block,s:cut}. For the heuristics behind the three hypotheses, we refer to \zcref{s:overview}. The three hypotheses are verified by computations of three different kinds. Hypothesis~(H1) consists of the polynomial inequalities \eqref{e:H1poly}, one for each of the $7{,}962$ patterns, and we verify them in exact integer arithmetic. Hypothesis~(H2) follows from an inequality for a vector of positive integers indexed by the $916$ histories of six steps defined below, which we also verify in exact arithmetic. Hypothesis~(H3) is a bound on two series with nonnegative terms, which we verify by a floating-point computation with directed rounding \cite{IEEE754}. The proofs of the bounds for (H1) and (H3) are long, and we give them in \zcref{s:local} and \zcref{s:tail}. These appendices rely on the definitions and results of \zcref{s:walks,s:contraction}, which precede them. The programs, the certificates, and the recorded outputs are the ancillary files described in \zcref{s:ancillary}, and we refer to them by their names \texttt{anc/\dots}. \begin{theorem}\label{t:inputs} Let $(\cs, \ct) = (961/1000, 187/200)$. Then Hypothesis~(H1) holds, Hypothesis~(H2) holds with $\gamma_{\mathrm o} = 2499/2500$ and $C_{\mathrm o} = 10$, and Hypothesis~(H3) holds with $K_\star = 0.92320$ and the enumerations $\varsigma_w$ of \zcref{s:tail-enum}, that is, \begin{equation}\label{e:inputsH3} K_{\mathfrak t}(1) \le 0.92320 \qquad \text{for } \mathfrak t \in \{\tcoll, \tdiag\}. \end{equation} \end{theorem} We prove \zcref{t:inputs} at the end of \zcref{s:computer-H1}. Of the three hypotheses, only (H1) becomes harder when the thresholds increase. \begin{remark}\label{r:monotone} The trail weight $\wt(w)$ in (H2) and, for a fixed enumeration $\varsigma_w$, the mass $\mathsf m(\{w\})$ in (H3) are sums of products of the numbers $1/5$, $1/\cs$, $1/\ct$, $\bar\alpha$, $\bar\beta$, $h_\tcoll$, and $h_\tdiag$, with nonnegative coefficients. They are nonincreasing functions of $\cs$ and of $\ct$. For the short trails, the rule of \zcref{d:enum} takes the minimum of the masses over all enumerations, which preserves this monotonicity. The enumeration of a longer trail is independent of the thresholds. Therefore, if (H1) holds at a pair $(\cs', \ct')$ such that $\cs' \ge 961/1000$ and $\ct' \ge 187/200$, then (H1), (H2), and (H3) hold at $(\cs', \ct')$, with the constants of \zcref{t:inputs} and the enumerations of \zcref{d:enum} evaluated at the new thresholds. Thus \zcref{t:contraction} applies at this pair. \end{remark} \section{The trail growth}\label{s:computer-H2} We now prove (H2). Recall from \zcref{s:walks-growth} that, for a trail $w = (v_0, \dots, v_n)$ of $\Z^2$ with an arrival site $v_{-1}$ of $v_0$, the trail weight \eqref{e:trailweight} is the product of the step weights $\wt_j$. Here $\wt_j = 1/5$ if step $j$ is a revisit ($v_{j - 1} \in \{ v_0, \dots, v_{j - 2} \}$), and $\wt_j = 1/(3\cs)$ or $\wt_j = 1/(3\ct)$ if step $j$ is a straight or a turning first visit. Hypothesis~(H2) asks for the bound \eqref{e:H2} on the sum of $\wt(w)$ over the trails of length $n$ with given $v_0$ and $v_{-1}$, subject to $v_1 \neq v_{-1}$ when $n \ge 1$. The weight of a step depends on the whole past of the trail, through the revisits. To bound these sums, we keep only the last $m$ steps of the past. We can then detect a revisit of a site that was visited within the last $m$ steps, and we treat every other step as a first visit. This can only increase the weight of a step, since $1/5 < 1/3 \le \min(1/(3\cs), 1/(3\ct))$. Finite-memory transfer matrices and automata were used in \cite{Alm93,PonitzTittmann00} to bound the connective constant of self-avoiding walks. In \cite{KLT88}, the bond-self-avoiding walks of the square lattice were counted in blocks of four steps, which bounds their number by a constant times $79^{n/4}$. However, $79^{1/4} > 2.98 > 3 \ct$, so this count together with the largest step weight $1/(3\ct)$ bounds the sums in (H2) only by a quantity that grows with $n$. We will see that a memory of six steps suffices at our thresholds. \begin{definition}\label{d:history} Let $m \ge 1$. A \emph{history} of memory $m$ is a sequence $\omega = (\mathsf e_1, \dots, \mathsf e_m)$ of unit vectors of $\Z^2$ such that the path $z_0 = 0$, $z_k = z_{k - 1} + \mathsf e_k$ for $1 \le k \le m$, uses $m$ distinct bonds $\{z_{k - 1}, z_k\}$. We let $\Omega_m$ denote the set of histories of memory $m$. A unit vector $\mathsf e$ is an \emph{admissible step} from $\omega$ if $\{z_m, z_m + \mathsf e\}$ is not one of the bonds $\{z_{k - 1}, z_k\}$, $1 \le k \le m$. For an admissible step $\mathsf e$ from $\omega$, we call \begin{equation*} \omega \mathsf e = (\mathsf e_2, \dots, \mathsf e_m, \mathsf e) \end{equation*} the \emph{successor} of $\omega$ by $\mathsf e$, and we define the \emph{kind} of $\mathsf e$ to be $\mathrm r$ if $z_m \in \{z_0, \dots, z_{m - 1}\}$, to be $\mathrm s$ if $z_m \notin \{z_0, \dots, z_{m - 1}\}$ and $\mathsf e = \mathsf e_m$, and to be $\mathrm t$ otherwise. \end{definition} Note that the successor $\omega \mathsf e$ belongs to $\Omega_m$ (its path is a translate of $(z_1, \dots, z_m, z_m + \mathsf e)$). Suppose that $\omega$ consists of the last $m$ steps of a trail. Then the kind $\mathrm r$ means that the step from the endpoint of the trail leaves a site that the trail visited within its last $m$ steps. The kinds $\mathrm s$ and $\mathrm t$ mean that it leaves another site, by a straight step or by a turn, respectively. We let $\mathsf A_m$ denote the matrix indexed by $\Omega_m \times \Omega_m$ whose entries are, for all $\omega \in \Omega_m$ and all admissible steps $\mathsf e$ from $\omega$, \begin{equation}\label{e:Am} \mathsf A_m(\omega, \omega \mathsf e) = \begin{cases} 1/5 & \text{if } \mathsf e \text{ has kind } \mathrm r, \\ 1/(3\cs) & \text{if } \mathsf e \text{ has kind } \mathrm s, \\ 1/(3\ct) & \text{if } \mathsf e \text{ has kind } \mathrm t, \end{cases} \end{equation} and whose other entries vanish. Since $\mathsf e$ is the last entry of $\omega \mathsf e$, \eqref{e:Am} defines each entry at most once. \begin{proposition}\label{p:H2} Let $(\cs, \ct) = (961/1000, 187/200)$ and $\gamma_{\mathrm o} = 2499/2500$. There exists a vector $\mathsf v = (\mathsf v(\omega))_{\omega \in \Omega_6}$ of positive integers such that \begin{equation}\label{e:H2vector} \mathsf A_6 \mathsf v \le \gamma_{\mathrm o} \mathsf v \quad \text{entrywise}, \qquad \frac{\max_\omega \mathsf v(\omega)}{\min_\omega \mathsf v(\omega)} \, (\ct \gamma_{\mathrm o})^{-6} \le 10 . \end{equation} Further, Hypothesis~(H2) holds with $\gamma_{\mathrm o} = 2499/2500$ and $C_{\mathrm o} = 10$. \end{proposition} \begin{proof} We first verify \eqref{e:H2vector}, and we then deduce (H2) from it. \emph{The vector.} The vector $\mathsf v$ is stored in the file \texttt{anc/h2/h2\_vector.json}. The script \texttt{anc/h2/h2\_check.py} generates the $916$ histories of $\Omega_6$ and their admissible steps from \zcref{d:history}. It then verifies the two inequalities \eqref{e:H2vector} in integer arithmetic, after multiplying the three weights $1/5$, $1/(3\cs) = 1000/2883$, and $1/(3\ct) = 200/561$ by their common denominator $5 \cdot 2883 \cdot 561$. The integer and rational comparisons establish \eqref{e:H2vector}. For reference, the rounded diagnostics in \texttt{anc/h2/h2\_output.log} give \begin{equation*} \max_{\omega \in \Omega_6} \frac{(\mathsf A_6 \mathsf v)(\omega)}{\mathsf v(\omega)} \approx 0.999563271649, \qquad \frac{\max_\omega \mathsf v(\omega)}{\min_\omega \mathsf v(\omega)} \, (\ct \gamma_{\mathrm o})^{-6} \approx 8.049712 . \end{equation*} \emph{Reduction to histories.} We now deduce (H2). Fix $v_0 \in \Z^2$ and a neighbor $v_{-1}$ of $v_0$. For $n \ge 0$, let $Q_n$ denote the sum on the left side of \eqref{e:H2}, so that $Q_0 = 1$. For $n \ge 6$ and $\omega \in \Omega_6$, let $Q_n(\omega)$ denote the part of this sum over the trails whose last six steps $(v_{n - 5} - v_{n - 6}, \dots, v_n - v_{n - 1})$ form $\omega$, so that \begin{equation*} Q_n = \sum_{\omega \in \Omega_6} Q_n(\omega) . \end{equation*} We claim that for all $n \ge 6$ and all $\omega' \in \Omega_6$, \begin{equation}\label{e:H2step} Q_{n + 1}(\omega') \le \sum_{\omega \in \Omega_6} Q_n(\omega) \, \mathsf A_6(\omega, \omega') . \end{equation} To prove \eqref{e:H2step}, let $w = (v_0, \dots, v_{n + 1})$ be a trail counted in $Q_{n + 1}(\omega')$. Its prefix $w_n = (v_0, \dots, v_n)$ is counted in $Q_n(\omega)$, where $\omega$ is formed by the last six steps of $w_n$, and $\omega' = \omega \mathsf e$ with $\mathsf e = v_{n + 1} - v_n$. The step $\mathsf e$ is admissible from $\omega$, since the path of $\omega$ is a translate of $(v_{n - 6}, \dots, v_n)$ and $w$ does not use a bond twice. We have $\wt(w) = \wt(w_n) \wt_{n + 1}$, and we now show that $\wt_{n + 1} \le \mathsf A_6(\omega, \omega \mathsf e)$. We distinguish three cases, and we claim that \begin{equation}\label{e:H2cases} \wt_{n + 1} \begin{cases} = 1/5 = \mathsf A_6(\omega, \omega \mathsf e) & \text{if } \mathsf e \text{ has kind } \mathrm r, \\ = 1/5 < \mathsf A_6(\omega, \omega \mathsf e) & \text{if } \mathsf e \text{ has kind } \mathrm s \text{ or } \mathrm t \text{ and } v_n \in \{v_0, \dots, v_{n - 7}\}, \\ = \mathsf A_6(\omega, \omega \mathsf e) & \text{otherwise}. \end{cases} \end{equation} In the first case, $v_n \in \{v_{n - 6}, \dots, v_{n - 1}\}$, so that step $n + 1$ is a revisit. In the second case, the same holds, and $1/5 < 1/3 \le \min(1/(3\cs), 1/(3\ct))$. In the remaining case, step $n + 1$ is a first visit. Its arrival direction $v_n - v_{n - 1}$ coincides with the last entry $\mathsf e_6$ of $\omega$, so that the step is straight if and only if $\mathsf e$ has kind $\mathrm s$. This proves \eqref{e:H2cases}. Distinct trails $w$ give distinct pairs $(w_n, \mathsf e)$, and summing over $w$ proves \eqref{e:H2step}. \emph{Conclusion.} Iterating \eqref{e:H2step} and using the first inequality of \eqref{e:H2vector}, we find for all $j \ge 0$ that \begin{equation}\label{e:H2iter} \begin{aligned} Q_{6 + j} &\le \frac{1}{\min \mathsf v} \sum_{\omega} Q_{6 + j}(\omega) \mathsf v(\omega) \le \frac{1}{\min \mathsf v} \sum_{\omega} Q_6(\omega) (\mathsf A_6^j \mathsf v)(\omega) \\ &\le \frac{\gamma_{\mathrm o}^j}{\min \mathsf v} \sum_\omega Q_6(\omega) \mathsf v(\omega) \le \frac{\max \mathsf v}{\min \mathsf v} \, \gamma_{\mathrm o}^j Q_6 , \end{aligned} \end{equation} where we used that all entries of $\mathsf A_6$ and all $Q_n(\omega)$ are nonnegative. Next, a trail has at most three continuations at each step, including the first, where the step to $v_{-1}$ is excluded. Every step weight is at most $1/(3\ct)$, since $\ct \le \cs \le 1$. We conclude that \begin{equation}\label{e:H2crude} Q_r \le 3^r (3\ct)^{-r} = \ct^{-r} \qquad \text{for all } r \ge 0 . \end{equation} Since $\ct \gamma_{\mathrm o} < 1$, the function $r \mapsto (\ct \gamma_{\mathrm o})^{-r}$ is increasing. For $n < 6$, we deduce from \eqref{e:H2crude} that \begin{equation*} Q_n \le (\ct \gamma_{\mathrm o})^{-n} \gamma_{\mathrm o}^n \le (\ct \gamma_{\mathrm o})^{-6} \gamma_{\mathrm o}^n . \end{equation*} For $n \ge 6$, \eqref{e:H2iter} with $j = n - 6$ and \eqref{e:H2crude} with $r = 6$ imply that \begin{equation*} Q_n \le \frac{\max \mathsf v}{\min \mathsf v} \, \gamma_{\mathrm o}^{n - 6} \ct^{-6} = \frac{\max \mathsf v}{\min \mathsf v} \, (\ct \gamma_{\mathrm o})^{-6} \gamma_{\mathrm o}^n . \end{equation*} Using the second inequality of \eqref{e:H2vector}, and $\max \mathsf v \ge \min \mathsf v$ for $n < 6$, we obtain $Q_n \le 10 \gamma_{\mathrm o}^n$ for all $n \ge 0$. This is (H2) with $\gamma_{\mathrm o} = 2499/2500$ and $C_{\mathrm o} = 10$. \end{proof} \section{The weighted trail sums}\label{s:computer-H3} Recall that Hypothesis~(H3) concerns the two scalar sums $K_\tcoll(1)$ and $K_\tdiag(1)$ of \eqref{e:kernel}, which are defined with the type weights $h = (1, 5/4)$ and with enumerations $\varsigma_w$ of the sets $\FV(w)$ of first-visit steps. We choose the enumerations in \zcref{s:tail-enum}. For trails of length at most $20$, $\varsigma_w$ minimizes the contribution of $w$ to \eqref{e:mT}, and for longer trails it lists the straight first visits before the turns. Each of these enumerations is a function of $w$ alone, as \zcref{l:Ktheta} requires. We bound the contribution of each type in two parts. The trails of length at most $20$ are enumerated exactly, by depth-first search, and their contributions are summed with upward rounding. A trail of length $n > 20$ is followed exactly for $20$ steps. Its continuation is dominated by a weighted walk on $\Z^2$ whose state consists of a phase, which keeps track of the enumeration of the first-visit steps, the history of the last ten steps, and the position. The main difference from the proof of (H2) is that the walk keeps its position. The walk stops when it reaches $v$. Its steps into $c$ have weight zero, so that it retains the constraint that the trail ends at $v$. A bound by all trails of a given length, closed or not, would be far too weak here, since the growth rate $\gamma_{\mathrm o}$ in (H2) is close to one. By \zcref{l:super}, the total weight of the paths of the walk that end at $v$ is bounded by every nonnegative function $\mathsf F$ of its states such that $\mathsf g + \mathsf T \mathsf F \le \mathsf F$, where $\mathsf T$ is the one-step operator of the walk and $\mathsf g$ its reward at $v$. \zcref[S]{l:box} gives such a function in terms of finitely many numbers, namely its values at the states with position in the box $\mathcal C_{70}$ and a function of the phase and the history, which serves at the positions outside the box (there, the coefficients of the walk no longer depend on the position). A program proposes these numbers. It then verifies the inequalities of \zcref{l:box} using directed rounding, as described in \zcref{s:tail-trust}. The result is \begin{equation}\label{e:computer-H3result} K_\tcoll(1) \le 0.923198, \qquad K_\tdiag(1) \le 0.899348 , \end{equation} which is \zcref{p:tail}. The certified numbers are collected in \zcref{tab:tail}, and the programs are described in \zcref{s:tail-trust}. \section{The local inequalities}\label{s:computer-H1} Recall the class $\Pext$ of extended patterns from \zcref{d:pattern}, the ring functional $\mathcal Z_\pi^{\mathbf a}$ of \eqref{e:ringfunctional} with its twelve exterior parameters $\mathbf a = (\mathbf u, \mathbf m) \in \Ball^{12}$, and the polynomials $I_{\mathrm{cur}}$ and $I_{\mathrm{next}}$ of \eqref{e:IJ}. Hypothesis~(H1) is the family of inequalities \eqref{e:H1poly}. By \zcref{l:local}, it implies the bound $\rho^\tau_j \le 1/c_\pi$ for the step ratio at every first visit whose pattern $\pi$ belongs to $\Pext$. \begin{theorem}\label{t:H1} Let $(\cs, \ct) = (961/1000, 187/200)$. Then for all $\pi \in \Pext$ and all $\mathbf a \in \Ball^{12}$, \begin{equation}\label{e:H1thm} \mathcal Z_\pi^{\mathbf a}[I_{\mathrm{cur}} - c_\pi I_{\mathrm{next}}] \ge 0 , \end{equation} where $c_\pi = \cs$ if the pattern $\pi$ is straight and $c_\pi = \ct$ if it is a turn. In particular, Hypothesis~(H1) holds with $(\cs, \ct) = (961/1000, 187/200)$. \end{theorem} The proof of \zcref{t:H1} is computer-assisted and is carried out in integer and rational arithmetic; we give it in \zcref{s:local}. We now describe its structure. Fix a pattern $\pi$, and let $c = c_\pi$. The middle site and the tips of the template were integrated out in the definition \eqref{e:ringfunctional}, so that $\Phi_c(\mathbf a) = \mathcal Z_\pi^{\mathbf a}[I_{\mathrm{cur}} - c I_{\mathrm{next}}]$ is a polynomial in the twelve vectors of $\mathbf a$ that is affine in each of them. A corner without present ring bonds contributes a positive factor, which we delete. We may also integrate out a corner with exactly one present ring bond, which is then represented by a single vector of norm at most one; see \zcref{l:leaf}. This leads to a reduced polynomial $\hat\Phi_c$ of at most eleven vectors, and \zcref{c:reduce} shows that it suffices to prove that $\hat\Phi_c \ge 0$ when all its vectors are unit vectors. The main step is a representation of $\hat\Phi_c$ by an operator. By \zcref{l:inversion}, $\hat\Phi_c(\mathbf b)$ is a diagonal matrix element of an explicit Hermitian operator, taken at a tensor product of spin-$1/2$ coherent states, after the vectors with indices in an inversion set $\mathcal I$ have been replaced by their negatives. It suffices to show that this operator is positive semidefinite. The inversion is needed, since a polynomial that is nonnegative on spheres can have an operator with a negative eigenvalue. The operator commutes with the action of $\mathrm{SU}(2)$ by \zcref{l:Ostruct}. It therefore suffices to test it on one eigenspace of the total spin component $S^3_{\mathrm{tot}}$, for each choice of exchange symmetry at the corners of status $1$. See \zcref{l:sector}. After multiplication by an integer, each test asks whether a symmetric integer matrix $\mathsf M_\epsilon$ of size at most $245$ is positive definite, which we decide by fraction-free Gaussian elimination \cite{Bareiss68}. See \zcref{p:certificate}. A numerical minimization guides the choice of inversion sets, and \zcref{s:local-computation} gives the bounds needed for exact coefficient arithmetic. The certificates are stored in \texttt{anc/h1/certificates.jsonl}, one for each of the $7{,}962$ patterns of \texttt{anc/h1/patterns.json}. The script \texttt{anc/h1/check\_certificate\_file.py} checks that every pattern has exactly one certificate and that the certificate of $\pi$ carries the threshold $c_\pi$. The checker \texttt{anc/h1/verify\_all.py} verifies every certificate from the stored data alone. Its output is recorded in \texttt{anc/h1/verify\_all\_output.log}. The programs, the further checks, and the code that must be trusted are described in \zcref{s:local}. \begin{proof}[Proof of \zcref{t:inputs}] \zcref[S]{t:H1} gives (H1). \zcref[S]{p:H2} gives (H2) with $\gamma_{\mathrm o} = 2499/2500$ and $C_{\mathrm o} = 10$. We now apply \zcref{p:tail}, with the enumerations $\varsigma_w$ of \zcref{s:tail-enum}, and we obtain the bounds \eqref{e:computer-H3result}. These bounds are at most $0.92320 < 1$, which proves \eqref{e:inputsH3}. The terms of the series \eqref{e:kernel} are nonnegative, so that its partial sums at $t = 1$ are bounded by its sum, and (H3) holds with $K_\star = 0.92320$. \end{proof} \section{Proof of the main theorem}\label{s:computer-proofmain} We now prove \zcref{t:main} by combining \zcref{t:inputs} with the implications proved in \zcref{s:quantum,s:block,s:cut,s:walks,s:contraction}. The explicit values of the constants are verified in \zcref{s:explicit}. \begin{proof}[Proof of \zcref{t:main}] We pass from the computer-assisted inputs to Hypothesis~(P), then to Hypothesis~(W), to the block dynamics, and finally to the quantum gap. \emph{Hypothesis (P).} \zcref[S]{t:inputs} shows that Hypotheses (H1), (H2), and (H3) hold at the pair $(\cs, \ct) = (961/1000, 187/200)$, for which $\min(\cs, \ct) = 187/200$. Then \zcref{t:contraction} implies that Hypothesis~(P) holds with some constants $A \ge 2$ and $\theta \in (0, 1)$. \emph{Hypothesis (W).} Next, \zcref{t:window} shows that Hypothesis~(P) implies Hypothesis~(W), with constants $p_0$ and $M_\star$ that depend only on $A$ and $\theta$. \emph{The gap of $\Lbl$.} Applying \zcref{p:blockgap}, we find that Hypothesis~(W) implies the following. For every integer $M \ge \max(M_\star, M_{\mathrm g})$, with $b = 2M + 1$, and all $L, N \ge b + 4$, the bound \eqref{e:blockgap} holds, with $\Delta_M$ as in \eqref{e:discrepancy}, and its right side is at least $1/2$ for all $M \ge M_{\mathrm{bl}}$. We fix $M \ge \max(M_\star, M_{\mathrm g})$ such that $4 \Delta_M \le b/2$ (for instance, $M \ge M_{\mathrm{bl}}$), so that for all $L, N \ge b + 4$, \begin{equation}\label{e:proofmain-block} \gap(\Lbl) \ge \frac12 . \end{equation} We set $L_0 = b + 4$. \emph{The quantum gap.} Let $L, N \ge L_0$. Using \zcref{t:comparison} and \eqref{e:proofmain-block}, we obtain \begin{equation}\label{e:proofmain-gap} \gap(H_{L, N}) \ge \frac{b}{b + 1} \, g_b \gap(\Lbl) \ge \frac{g_b}{4} , \end{equation} where we used that $b/(b + 1) \ge 1/2$. Recall that $g_b$ is the smallest positive eigenvalue of the patch operator $h_b$, which is the same operator for all $L, N \ge b + 4$. \zcref[S]{l:patchgapexplicit} gives an explicit positive lower bound for $g_b$. \emph{The kernel.} Since $L, N \ge 3$, the kernel of $H_{L, N}$ is spanned by $\VBS$, by \zcref{c:kernel}. \emph{The constants.} In \zcref{s:explicit}, we make the constants of the previous steps explicit. \zcref[S]{l:explicitP} shows that (P) holds with $A = 10^{13}$ and $\theta = 1 - 10^{-20}$, and \zcref{l:explicitW} shows that (W) then holds with $M_\star = 10^{180}$ and $p_0 = \exp(-2 \cdot 10^{36})$. We also show there that $M = \lceil \exp(10^{40}) \rceil$ satisfies $M \ge \max(M_\star, M_{\mathrm g})$ and $4 \Delta_M \le b/2$. We may therefore take this value of $M$ in the third step, which gives $L_0 = 2 \lceil \exp(10^{40}) \rceil + 5$. For this choice, \zcref{l:patchgapexplicit} and \eqref{e:proofmain-gap} imply that $\gap(H_{L, N}) \ge c$ with $c = \exp[-\exp(\exp(10^{41}))]$; see \eqref{e:explicitgap}. This completes the proof. \end{proof}