\documentclass[11pt]{amsart} \usepackage{amssymb,amsmath,amsthm,mathtools} \usepackage[margin=1.15in]{geometry} \usepackage{booktabs} \usepackage{array} \newcolumntype{P}[1]{>{\raggedright\arraybackslash}p{#1}} \raggedbottom \usepackage{hyperref} \hypersetup{colorlinks=true,linkcolor=blue!50!black,citecolor=blue!50!black,urlcolor=blue!50!black} \usepackage{xcolor} \newtheorem{theorem}{Theorem}[section] \newtheorem{proposition}[theorem]{Proposition} \newtheorem{lemma}[theorem]{Lemma} \newtheorem{corollary}[theorem]{Corollary} \newtheorem{conjecture}[theorem]{Conjecture} \theoremstyle{definition} \newtheorem{definition}[theorem]{Definition} \newtheorem{example}[theorem]{Example} \theoremstyle{remark} \newtheorem{remark}[theorem]{Remark} \newcommand{\Q}{\mathbb{Q}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\C}{\mathbb{C}} \newcommand{\F}{\mathbb{F}} \newcommand{\PP}{\mathbb{P}} \newcommand{\Gal}{\operatorname{Gal}} \newcommand{\Tr}{\operatorname{Tr}} \newcommand{\maxord}{\operatorname{maxord}} \title[Free fermions from integer arithmetic]{Free fermions from integer arithmetic:\\ a Galois-theoretic test for free-fermionic spectra} \author{Shankar Sivarajan} \makeatletter \g@addto@macro\@setauthors{% \begin{center}\footnotesize Department of Computational Medicine and Bioinformatics\\ University of Michigan, Ann Arbor, MI, USA\\ \href{mailto:sivarajn@umich.edu}{\texttt{sivarajn@umich.edu}} \end{center}} \makeatother \begin{document} \begin{abstract} The spectrum of a free-fermionic operator is built from a few single-particle quantities. For operators with rational entries this confines, as is known, the Galois group of the characteristic polynomial of each sector to a hyperoctahedral group, up to elementary abelian $2$-groups. We turn this into a test for free-fermionic spectra that uses only integer arithmetic: Frobenius cycle types modulo primes can prove that a spectrum is not free-fermionic, and an exact recovery of the single-particle data, certified by power sums, confirms it when it is. As a worked example, the test recovers Onsager's dispersion relation from the characteristic polynomial of the Ising transfer matrix. As applications, we prove that the periodic Fendley chain with generic couplings admits no free-fermion solution of the Chapman--Elman--Mann form for a range of sizes, bearing on whether a simplicial clique is necessary for free-fermion solvability, and we detect hidden tensor-product and matchgate structure in any basis. The Galois group grades spectral structure: hyperoctahedral for free fermions, symmetric on subsets with particle-number conservation, apparently imprimitive for the partially decoupled Bethe equations of the exclusion process, and alternating within sectors for the interacting models tested. \end{abstract} \maketitle %=============================================================================================================== \section{Introduction} A free-fermionic model is exactly solvable because its many-body spectrum is built from a few single-particle energies. For the two-dimensional Ising model this is the content of Kaufman's spinor solution~\cite{Kaufman} of Onsager's model~\cite{Onsager}: every eigenvalue of the row transfer matrix on a ring of $L$ sites has the form \begin{equation}\label{eq:freeform} \lambda_s=\kappa\prod_{k=1}^{L}e^{s_k\gamma_k/2},\qquad s\in\{\pm1\}^L, \end{equation} with a fixed parity of $s$ in each spin-flip sector. The same structure governs matchgate circuits, which are classically simulable for this reason~\cite{Valiant,TerhalDiVincenzo,JozsaMiyake}, and Hamiltonians that are free only after a nonlocal change of variables, the free fermions in disguise of Fendley~\cite{FendleyDisguise}, characterized graph-theoretically by Chapman, Flammia, Elman, and Mann~\cite{ChapmanFlammia,ElmanChapmanFlammia,ChapmanElmanMann}. Arranged by $s$, the spectrum~\eqref{eq:freeform} is the rank-one tensor $\kappa\,(u_1,u_1^{-1})\otimes\cdots\otimes(u_L,u_L^{-1})$, $u_k=e^{\gamma_k/2}$, a point of the Segre variety $(\PP^1)^L$, and its eigenvalues satisfy $\lambda_s\lambda_t=\lambda_{s'}\lambda_{t'}$ whenever $s+t=s'+t'$. For an operator with rational entries every automorphism of $\overline{\Q}$ preserves these relations, so the Galois group of the characteristic polynomial is confined to the symmetries of the hypercube: for the square of the transfer matrix in a spin-flip sector it is a subgroup of the hyperoctahedral group $W(B_L)$, of order $2^LL!$, and for the transfer matrix itself, whose eigenvalues involve the half-angles $\gamma_k/2$ and a possibly irrational $\kappa$, an extension of such a subgroup by an elementary abelian $2$-group (Proposition~\ref{prop:variants}); a generic matrix of size $2^L$ has the symmetric group $S_{2^L}$. This confinement is a special case of a theorem of Jouve, Kowalski, and Zywina~\cite{JKZ}: for an element $g$ of a reductive group, the Galois group of the characteristic polynomial of $\rho(g)$ lies in the Weyl group extended by the Galois action on characters, and acts through the weights of $\rho$. Free fermions are the case of $\mathrm{Spin}(2L)$ in its spin representation, whose weights are the sign vectors $(\pm\tfrac12,\dots,\pm\tfrac12)$. With particle-number conservation the hidden group is $U(L)$ acting on the $N$-particle space $\Lambda^N\C^L$, and its Weyl group $S_L$ acts on $N$-element subsets. This paper uses the confinement as a test that reads off the hidden group through its Weyl group, with free fermions as the main case. The Galois group of an integer polynomial is accessible through its factorizations modulo primes: by Chebotarev's density theorem~\cite{Chebotarev}, the degrees of the irreducible factors modulo $p$ are cycle types of Frobenius elements, which equidistribute in the group. A cycle length incompatible with the allowed group orders, or a certificate that the group contains an alternating group $A_d$ with $d\ge5$ and $d>L$, rules out free-fermionic structure. The test needs no fermionic representation, works in any basis, and costs a few factorizations of integer polynomials. It concerns the spectrum only: it detects free-fermionic spectral structure, not fermionic dynamics, and operators whose spectra are trivially of free form, such as commuting stabilizer Hamiltonians, pass it. \subsection*{Results} Section~\ref{sec:lemma} proves the hyperoctahedral lemma and derives rigorous criteria for non-freeness: for transfer matrices and their squares, for complex and orthogonal operators, for number-conserving free fermions and free parafermions, and for Pauli Hamiltonians with symmetry sectors. Section~\ref{sec:procedure} gives a three-step procedure (screen, recover, certify) and works it out for the Ising transfer matrix, recovering Onsager's dispersion relation from the integer characteristic polynomial and certifying the Kaufman form exactly for strips of width $3$ to $7$. Section~\ref{sec:applications} contains the applications. The first concerns claw-free frustration graphs without a simplicial clique, which lie outside the free-fermion solutions of Chapman, Elman, and Mann~\cite{ChapmanElmanMann}: we prove that for generic couplings the periodic Fendley chain and further cycle powers admit no free-fermion solution of their form at a range of sizes (Theorem~\ref{thm:fendley}), and we certify $23$ graphs on at most $9$ vertices, $176$ on $10$, and $29$ on $11$. This complements exact free solutions reported for other graphs of this kind, including the periodic Fendley chain on $8$ and $9$ sites~\cite{AnonFrustration}. The second recovers hidden group structure, tensor factors and free-fermion modes, from the multiplicative relations among eigenvalues. The third concerns quantum circuits (Section~\ref{sec:circuits}). Section~\ref{sec:integrable} shows how the Galois group grades spectral structure: interacting integrable models have alternating groups inside their symmetry sectors, in agreement with recent work on Bethe equations~\cite{BellwoodMunro,LiaoRybnikov}, while the totally asymmetric exclusion process, whose Bethe equations partially decouple, appears to have imprimitive groups of intermediate size. Appendix~\ref{app:validation} validates the test on free-fermion models from the literature: in every family the points flagged free are known free-fermion points or have roots-of-unity spectra, except for three-dimensional Ising bars with a Clifford background, which Appendix~\ref{app:validation} explains through the Lie algebra of a Pauli orbit. At one of these points the hidden group is symplectic, which the Galois group cannot distinguish from free fermions; Table~\ref{tab:summary} summarizes the families. Appendix~\ref{app:recoveries} recovers the single-particle spectra of further known models (Table~\ref{tab:recoveries}), and Section~\ref{sec:outlook} lists open questions. The arithmetic of spectra has been used before for particular models, for example the Galois action on the spectra of small Heisenberg rings~\cite{Milewski} and integer factorization as a signature of Hilbert-space fragmentation~\cite{RegnaultLiuBernevig}. The use of hyperoctahedral Galois groups as a test for free-fermionic structure, the certificates of non-freeness, and the three-step procedure appear to be new. %=============================================================================================================== \section{The hyperoctahedral lemma and certificates of non-freeness}\label{sec:lemma} \begin{definition}\label{def:free} Let $\chi\in\Q[\mu]$. We say $\chi$ has \emph{free form of rank $L$} if there are $\kappa\in\Q^\times$, a squarefree reciprocal polynomial $P\in\Q[u]$ of degree $2L$ with roots $u_1^{\pm1},\dots,u_L^{\pm1}$, none equal to $\pm1$, and a set $S$ of sign vectors, either all of $\{\pm1\}^L$ or its even-parity half, such that \[ \chi(\mu)=\prod_{s\in S}\Bigl(\mu-\kappa\prod_{k=1}^{L}u_k^{s_k}\Bigr). \] \end{definition} The arguments below use only the splitting field $N$ of $P$. For uniform couplings $P$ need not be squarefree ($\gamma_q=\gamma_{-q}$ below), and $\chi$ is then not literally of free form of lower rank; but $N$ is the splitting field of the squarefree part of $P$ with the factors $u\mp1$ removed, so parts (1) and (2) of Lemma~\ref{lem:hyper} and Corollary~\ref{cor:cert} hold with the rank $r\le L$ of that part. For the Ising transfer matrix $V$, $\chi$ is the characteristic polynomial of $V^2$ on a spin-flip sector, $P$ is the characteristic polynomial of the single-particle matrix, and $u_k=e^{\gamma_k}$~\cite{Kaufman}. For a free-fermion Hamiltonian the form is additive, $E_s=e_0+\sum_ks_k\varepsilon_k$, and the arguments below apply with sums in place of products. Let $W(B_L)=C_2\wr S_L$ act on $\{\pm1\}^L$ by $(\pi,\epsilon)\cdot s=(\epsilon_{\pi^{-1}(k)}s_{\pi^{-1}(k)})_k$, and let $W(D_L)$ be the subgroup with $\prod_k\epsilon_k=1$. \begin{lemma}\label{lem:hyper} Let $\chi$ have free form of rank $L$ (Definition~\ref{def:free}), with the roots $\mu_s=\kappa\prod_ku_k^{s_k}$, $s\in S$, pairwise distinct, and let $N$ be the splitting field of $P$. Then: \begin{enumerate} \item the splitting field of $\chi$ is contained in $N$, so $\Gal(\chi)$ is a quotient of $\Gal(N/\Q)$; \item $\Gal(N/\Q)$ embeds in $W(B_L)$ via $\sigma(u_k)=u_{\pi(k)}^{\epsilon_k}$; \item $\Gal(N/\Q)$ acts on the roots of $\chi$ through sign vectors: $\sigma(\mu_s)=\mu_{(\pi,\epsilon)\cdot s}$; \item if $S$ is the even half and no $\mu_s$ with $s$ even equals $\kappa\prod_ku_k^{t_k}$ with $t$ odd, the image lies in $W(D_L)$. \end{enumerate} In particular every cycle of a Frobenius element at a prime of good reduction has length at most the largest order $\maxord(B_L)$ of an element of $W(B_L)$; if $S=\{\pm1\}^L$, or under the hypothesis of (4), its cycle type is that of an element of $W(B_L)$, respectively $W(D_L)$, acting on $S$. \end{lemma} \begin{proof} Each $\mu_s$ is a monomial in the roots of $P$, which gives (1). An automorphism $\sigma$ permutes the roots of $P$ and commutes with $u\mapsto u^{-1}$, so it permutes the $L$ pairs $\{u_k,u_k^{-1}\}$, which are distinct because $P$ is squarefree with no root $\pm1$. Recording the permutation $\pi$ of the pairs and whether $u_k$ goes to $u_{\pi(k)}$ or to its inverse gives an injective homomorphism into $W(B_L)$, which is (2). Then $\sigma(\mu_s)=\kappa\prod_ku_{\pi(k)}^{\epsilon_ks_k}=\kappa\prod_ju_j^{\epsilon_{\pi^{-1}(j)}s_{\pi^{-1}(j)}}=\mu_{(\pi,\epsilon)\cdot s}$, which is (3). For (4), $\sigma$ maps roots of $\chi$ to roots of $\chi$, and $(\pi,\epsilon)$ changes the parity of $s$ by $\prod_k\epsilon_k$. \end{proof} Lemma~\ref{lem:hyper} is the elementary core of a general statement. For a reductive group $G$ over $\Q$ with a faithful representation $\rho$, Jouve, Kowalski, and Zywina show that the Galois group of the splitting field of $\det(T-\rho(g))$ is a subquotient of the group generated by the Weyl group and the Galois action on the characters of a maximal torus, that the eigenvalues are the values of the weights of $\rho$ at $g$, and that Galois acts through its action on the weights~\cite[Prop.~2.1, Lemma~2.4]{JKZ}; random elements attain the whole group~\cite[Thm.~1.1]{JKZ} (see also~\cite{GorodnikNevo}). For $\mathrm{Spin}(2L)\times\mathbb{G}_m$ in the spin representation this gives Lemma~\ref{lem:hyper}. Galois groups and multiplicative relations among eigenvalues are related in~\cite{Kowalski,Smyth}. \begin{corollary}[non-freeness certificate]\label{cor:cert} If an irreducible factor of $\chi$ of degree $d\ge5$ has Galois group containing $A_d$ and $d>L$, then $\chi$ divides no polynomial of free form of rank at most $L$. \end{corollary} \begin{proof} Parts (1) and (2) of Lemma~\ref{lem:hyper} do not use that the $\mu_s$ are distinct: the splitting field of every divisor of a polynomial of free form of rank $L$ lies in $N$, so the Galois group of every factor is a quotient of a subgroup of $W(B_L)=C_2^L\rtimes S_L$. Since the factor is irreducible of degree $d$, its Galois group is a subgroup of $S_d$; containing $A_d$, it is $A_d$ or $S_d$, and for $d\ge5$ it has the simple group $A_d$ as a composition factor, so $A_d$ is a composition factor of a subgroup of $W(B_L)$. A subgroup $H$ has the abelian normal subgroup $H\cap C_2^L$ with quotient in $S_L$, so $A_d$ is a composition factor of a subgroup of $S_L$, and $d!/2\le L!$. This forces $d\le L$. \end{proof} The same argument applies to every bound below: in each case the nonabelian composition factors of the relevant Galois group are those of subgroups of $S_L$ (of $S_\alpha$ or $S_t$ in Proposition~\ref{prop:pauli}), so the order bounds stated there are only needed for element orders, and an alternating group $A_d$, $d\ge5$, is excluded as soon as $d>L$ (respectively $d>\alpha$ and $d>t$). In each application the rank is fixed in advance by the size of the system: for a transfer matrix on $L$ sites, or a circuit on $L$ qubits, it is $L$, the number of Majorana pairs, and for a Hamiltonian it is the number of modes; certificates exclude free form of this rank. The bounds below apply in the same way to sectors and factors: they constrain every polynomial whose roots are of the given form, not only the full characteristic polynomial. The operator itself, not its square, is usually given, and its entries may be complex. \begin{proposition}\label{prop:variants} Suppose every eigenvalue of a rational matrix is of the form $\lambda_s=\kappa\prod_kv_k^{s_k}$ for some $s\in\{\pm1\}^L$, with $\kappa^2\in\Q$ and $v_k^{\pm2}$ the roots of a rational reciprocal polynomial of degree $2L$. Then all eigenvalues lie in a Galois extension of $\Q$ of degree at most $G_L=2^{2L+1}L!$, in whose Galois group every element has order dividing $2a$ for the order $a$ of some element of $W(B_L)$. If instead the matrix is the realification of a matrix over $\Q(i)$ whose eigenvalues have this form with $\kappa^2\in\Q(i)$ and $v_k^{\pm2}$ the roots of a reciprocal polynomial over $\Q(i)$, then all eigenvalues lie in a Galois extension of $\Q$ of degree at most $2G_L^2$ in whose Galois group every element has order dividing $4\operatorname{lcm}(a,b)$ for the orders $a,b$ of some elements of $W(B_L)$. Consequently a factor of degree $d\ge5$ with Galois group containing $A_d$ and $d>L$, or a Frobenius cycle whose length divides no such $2a$ (respectively no such $4\operatorname{lcm}(a,b)$), for instance, for a single set, a cycle longer than $2\maxord(B_L)$ in the first case, proves that the eigenvalues are not all of this form. \end{proposition} \begin{proof} Let $N$ be the splitting field of the reciprocal polynomial, of degree at most $2^LL!$. The set $\{v_k^{\pm2}\}$ is Galois stable, so $N(v_1,\dots,v_L,\kappa)$ is Galois over $\Q$ of degree at most $2^{L+1}\cdot2^LL!$, and the kernel of its Galois group onto $\Gal(N/\Q)$ is an elementary abelian $2$-group; if the image of $g$ has order $a$, then $g^{2a}=1$. In the second case the same construction over $\Q(i)$ gives a Galois extension $K$ of $\Q(i)$ of degree at most $G_L$ containing the eigenvalues of the complex matrix. The eigenvalues of its realification are these and their complex conjugates, which lie in $K\bar K$, a Galois extension of $\Q$ of degree at most $2G_L^2$. The group $\Gal(K\bar K/\Q(i))$ embeds in $\Gal(K/\Q(i))\times\Gal(\bar K/\Q(i))$, so its elements have orders dividing $\operatorname{lcm}(2a,2b)$, and every element of $\Gal(K\bar K/\Q)$ has its square in this subgroup. In both cases the nonabelian composition factors are those of subgroups of $S_L$, and $A_d$ with $d\ge5$ is one only if $d\le L$, as in Corollary~\ref{cor:cert}. \end{proof} \begin{remark}[two sectors]\label{rem:twosector} Suppose the eigenvalues form two sets as in Proposition~\ref{prop:variants}, each with its own rational data $\kappa_i^2\in\Q$ and $P_i\in\Q[u]$ (respectively over $\Q(i)$), as for the two parity sectors of a transfer matrix, with no assumption that the sectors are defined over $\Q$. The argument of the proof, applied to the compositum of the two fields, gives element orders dividing $2\operatorname{lcm}(a,b)$, respectively $4\operatorname{lcm}(a,b,c,d)$, for orders $a,b,c,d$ of elements of $W(B_L)$, and the composition-factor criterion is unchanged. Every cycle-length certificate in this paper also satisfies this stronger condition. \end{remark} \begin{remark}[orthogonal operators]\label{rem:selfrecip} If $M=c\,U$ with $U$ real orthogonal and $c^2\in\Q$, then after removal of the factor $x^2-c^2$ (of the factors $x\mp c$ if $c\in\Q$) the characteristic polynomial is $f(x)=x^mh(x+c^2/x)$, and the Galois group of $f$ is imprimitive with blocks $\{x,c^2/x\}$, so it never contains $A_{\deg f}$. Proposition~\ref{prop:variants} applies instead to the half-degree polynomial $h$, whose splitting field lies in that of $f$: a factor of $h$ of degree $d\ge5$ containing $A_d$ with $d>L$ certifies non-freeness. \end{remark} \begin{proposition}[number-conserving free fermions]\label{prop:numcons} Suppose every eigenvalue of a rational matrix is $E_A=\sum_{k\in A}\varepsilon_k$ for an $N$-element subset $A$ of $\{1,\dots,L\}$, where the $\varepsilon_k$ are the distinct roots of a rational polynomial $Q$ of degree $L$. Then the Galois group of the characteristic polynomial is a quotient of $\Gal(Q)\le S_L$. If the eigenvalues are the $E_A$ for all $A$ and are pairwise distinct, it acts on them through the action of $S_L$ on $N$-element subsets. \end{proposition} \begin{proof} Each $E_A$ lies in the splitting field of $Q$. An automorphism $\sigma$ permutes the roots of $Q$ by a permutation $\pi$, and $\sigma(E_A)=E_{\pi(A)}$. \end{proof} This is the case of $U(L)$ acting on $\Lambda^N\C^L$ in~\cite[Lemma~2.4]{JKZ}, whose Weyl group is $S_L$. As in Corollary~\ref{cor:cert}, a factor of degree $d\ge5$ whose Galois group contains $A_d$ with $d>L$ excludes this form; Section~\ref{sec:integrable} confirms the action on subsets for tight-binding chains. \begin{remark}[free parafermions]\label{rem:para} For free parafermions with $\Z_3$ symmetry~\cite{FendleyParafermions} the energies are $E=e_0+\sum_k\omega^{a_k}\varepsilon_k$, $a\in\{0,1,2\}^L$, $\omega=e^{2\pi i/3}$. If $e_0\in\Q(\omega)$ and the $\varepsilon_k^3$ are the roots of a polynomial of degree $L$ over $\Q(\omega)$, then, as in Proposition~\ref{prop:variants}, every energy lies in the field generated over $\Q(\omega)$ by the cube roots of the roots of this polynomial and of its complex conjugate, a Galois extension of $\Q$ of degree at most $2(3^LL!)^2$ whose nonabelian composition factors are those of subgroups of $S_L$; a factor of degree $d\ge5$ whose Galois group contains $A_d$ with $d>L$ excludes this form. \end{remark} \subsection*{Pauli Hamiltonians} Let $H=\sum_jb_jP_j$ be a sum of Pauli strings with frustration graph $G$ (an edge between anticommuting strings). In the free-fermion solutions of Chapman, Elman, and Mann~\cite[Sec.~I.A, Result~1]{ChapmanElmanMann}, each symmetry sector, a joint eigenspace of commuting conserved charges, carries an additive free spectrum $\sum_{j=1}^{\alpha}s_j\varepsilon_j$ with $\alpha=\alpha(G)$ modes, the independence number of $G$. Every Pauli realization of $G$ is a representation of the twisted group algebra of $G$; its irreducible modules, which we call central sectors, are labeled by the eigenvalues of central Pauli products, are defined over $\Q(i)$, and have dimension $D=2^{r/2}$ with $r$ the rank of the anticommutation form. For rational couplings the restriction of $H$ to a central sector is a Hermitian matrix over $\Q(i)$, so its characteristic polynomial lies in $\Q[E]$. We say that a central sector has the \emph{Chapman--Elman--Mann form} if it is the direct sum of the joint eigenspaces $J$ of commuting operators with entries in $\Q(i)$ that lie in the algebra generated by the terms of $H$ and commute with $H$, as the conserved charges of~\cite{ChapmanElmanMann} do, and if on each $J$ the spectrum is the multiset of the numbers $\sum_js_j\varepsilon_{J,j}$, with no constant term, over all $s\in\{\pm1\}^\alpha$ or over a parity half, each $s$ counted at least once, where $\pm\varepsilon_{J,j}$ are the roots of a polynomial whose coefficients lie in the field generated over $\Q(i)$ by the eigenvalues labeling $J$. Allowing parity halves only weakens the criterion below; each $J$ has at least $2^{\alpha-1}$ states. \begin{proposition}\label{prop:pauli} Suppose a central sector of dimension $D$ has the Chapman--Elman--Mann form. If a factor of degree $d\ge5$ of its characteristic polynomial, in $E$ or in $E^2$, has Galois group containing $A_d$, then $d\le\alpha$ or $d\,2^{\alpha-1}\le D$. If $\alpha$ is even, the set of eigenvalues of the sector is symmetric under $E\mapsto-E$. \end{proposition} \begin{proof} Work over $\Q(i)$. The Galois group of the factor over $\Q(i)$ is a subgroup of index at most $2$ of its group over $\Q$, so it still contains $A_d$, and the factor remains irreducible over $\Q(i)$. Let $E$ be an eigenvalue in the subspace $J$ such that $E$ or $E^2$ is a root of the factor, let $F$ be the field generated over $\Q(i)$ by the labels of $J$, and let $t=[F:\Q(i)]$. An automorphism of $\overline{\Q}$ over $\Q(i)$ fixes the central sector and maps $J$ to the joint eigenspace with the conjugate labels; the $t$ embeddings of $F$ give $t$ distinct joint eigenspaces, each of dimension at least $2^{\alpha-1}$, so $t\,2^{\alpha-1}\le D$. The splitting field over $F$ of the polynomial with roots $\pm\varepsilon_{J,j}$ contains $E$ and $E^2$, and as in Lemma~\ref{lem:hyper} its Galois group is isomorphic to a subgroup of $W(B_\alpha)$. (Repeated or vanishing $\varepsilon_{J,j}$ only shrink this group.) Its Galois closure $\Omega$ over $\Q(i)$ is the compositum of its images $\tau(M)$ under the $t$ embeddings $\tau$ of $F$, where $M$ denotes the splitting field. The group $\Gal(\Omega/\Q(i))$ permutes the embeddings by composition, with image in $S_t$; its kernel $K$ fixes every $\tau(F)$ pointwise, so it maps each $\tau(M)$, which is Galois over $\tau(F)$, into itself, and $K$ embeds in $\prod_\tau\Gal(\tau(M)/\tau(F))\cong\Gal(M/F)^t$. The Galois group of the factor over $\Q(i)$ is a quotient of $\Gal(\Omega/\Q(i))$ and contains the simple group $A_d$, so $A_d$ is a composition factor of $K$ or of a subgroup of $S_t$, hence of a subgroup of $W(B_\alpha)$ or of $S_t$. As in Corollary~\ref{cor:cert}, $d\le\alpha$ or $d\le t\le D/2^{\alpha-1}$. For even $\alpha$ the map $s\mapsto-s$ preserves parity, so the set of eigenvalues on each $J$, and hence on the sector, is symmetric. \end{proof} We therefore certify non-freeness of the Chapman--Elman--Mann form when \[ d>\alpha\qquad\text{and}\qquad d\,2^{\alpha-1}>D, \] or, for even $\alpha$, when the set of eigenvalues of a central sector is not symmetric under $E\mapsto-E$, that is, when its squarefree part $\chi_{\mathrm{red}}$ does not satisfy $\chi_{\mathrm{red}}(-E)=\pm\chi_{\mathrm{red}}(E)$. \subsection*{Certifying alternating groups} A primitive permutation group of degree $d$ containing a cycle of prime length $p\le d-3$ contains $A_d$, by a theorem of Jordan~\cite[Thm.~3.3E]{DixonMortimer}. A transitive group containing a $p$-cycle $c$ with $p>d/2$ is primitive. Indeed, $c$ permutes the blocks of a block system in orbits of size $1$ or $p$. A block fixed by $c$ that contains a point moved by $c$ contains all $p$ such points, so it has more than $d/2$ points and is everything; otherwise the moved points lie in at least $p>d/2$ distinct blocks, which are then singletons. A Frobenius element with a single cycle of prime length $p$, $d/21$ the matrix $V_L$ is similar to a positive definite matrix, so its spectrum is determined by that of $V_L^2$. Since the spectrum of $V_L$ determines $\Tr V_L^M$ for all $M$ and $\Tr V_L^M$ is symmetric in $L$ and $M$ on the torus, these certificates give the exact partition function of every $L\times M$ torus with $3\le\min(L,M)\le7$ at these couplings, and for $3\le L\le7$ its restriction to each boundary sector in the direction of length $L$. A proof valid for all $L$ requires an identity uniform in $L$, supplied in the known proofs by the spinor representation~\cite{Kaufman} or a determinant formula for the partition function. %=============================================================================================================== \section{Applications}\label{sec:applications} \subsection{Claw-free frustration graphs without a simplicial clique}\label{sec:frontier} Chapman and Flammia, and Elman, Chapman, and Flammia, showed that claw-free frustration graphs satisfying further conditions give free-fermion solutions~\cite{ChapmanFlammia,ElmanChapmanFlammia}. Chapman, Elman, and Mann proved that a claw-free graph with a simplicial clique, a clique $K$ such that for every $j\in K$ the neighborhood of $j$ induces a clique in $G\setminus K$, admits a free-fermion solution, with energies given by the roots of independence polynomials~\cite[Sec.~I.A, Result~1]{ChapmanElmanMann}; whether the simplicial clique is necessary is open~\cite[Sec.~IX]{ChapmanElmanMann} (see also~\cite{RuhElmanChapman}). The spectrum depends on the realization only through the central sector~\cite[Sec.~VIII]{ChapmanElmanMann}, so the central sectors of the realization with one qubit per vertex cover all realizations. Graphs that become simplicial after twin collapse are free~\cite{RuhElman} and are excluded. Exact free solutions of the Chapman--Elman--Mann form for $107$ graphs outside these classes, among them $C_8^2$ and $C_9^2$, are reported in~\cite{AnonFrustration}. \emph{Small graphs.} There are $14,50,191,881,4494$ connected claw-free graphs on $5,\dots,9$ vertices, of which $0,0,2,13,186$ have no simplicial clique; all of these contain even holes, and $173$ remain after twin collapse. Each was tested in every central sector for three random choices of integer couplings with Proposition~\ref{prop:pauli} (Table~\ref{tab:frontier}). No control with a simplicial clique was ever certified, and none of the $106$ graphs of~\cite{AnonFrustration} with exact free certificates that lie in our sets on at most $10$ vertices is certified. Among the extensions by one vertex of the undecided graphs on $9$ vertices, $176$ of the $2624$ graphs on $10$ vertices are certified, $799$ have a factor containing $A_d$ to which Proposition~\ref{prop:pauli} does not apply, $192$ are consistent with free form, and $1457$ give the test no power; in a sample of $800$ graphs on $11$ vertices with $\alpha\ge3$, $29$ are certified, $582$ are consistent, and $189$ give the test no power. Here a graph is consistent if some factor has degree $d\ge5$ with $d>\alpha$ and $d\,2^{\alpha-1}>D$, so that Proposition~\ref{prop:pauli} could apply, but no certificate is found; otherwise the test has no power beyond the symmetry test for even $\alpha$, which these graphs pass. The classes are assigned in the order certified, $A_d$ uncertified, consistent, no power. A graph counts as certified if some choice of couplings certifies it; of the $192$ consistent graphs, $190$ are consistent at all three choices. Numerically, none of the $190$ graphs on $10$ vertices consistent at all three choices, nor $120$ sampled on $11$, has a sector spectrum that is a full additive hypercube or a union of two full $3$-mode hypercubes; the free forms not excluded for them involve parity halves, which at these sizes impose almost no constraint. \begin{table}[ht] \centering\small \begin{tabular}{@{}lrrrrr@{}} \toprule class & graphs & certified & $A_d$, uncertified & consistent & no power \\ \midrule no simplicial clique, $n=7$ & $1$ & $1$ & $0$ & $0$ & $0$ \\ no simplicial clique, $n=8$ & $9$ & $2$ & $3$ & $0$ & $4$ \\ no simplicial clique, $n=9$ & $163$ & $20$ & $48$ & $6$ & $89$ \\ simplicial clique (controls) & $30$ & $0$ & $0$ & $5$ & $25$ \\ simplicial after twin collapse (controls) & $28$ & $0$ & $0$ & $0$ & $28$ \\ with a claw (controls) & $20$ & $13$ & $0$ & $0$ & $7$ \\ \bottomrule \end{tabular} \caption{Claw-free frustration graphs without a simplicial clique, $n\le9$, and controls. ``$A_d$, uncertified'': some factor has Galois group containing $A_d$, but $d\,2^{\alpha-1}>D$ fails, so Proposition~\ref{prop:pauli} does not apply. ``Consistent'' and ``no power'' are defined in the text.}\label{tab:frontier} \end{table} \emph{Cycle powers.} The square $C_M^2$ of the cycle is the frustration graph of Fendley's chain~\cite{FendleyDisguise} with periodic boundary conditions, whose spectrum Zheng, Pozsgay, and Chen obtain from functional relations of a commuting family of transfer matrices, with free, nested free, or interacting spectra depending on the couplings~\cite{ZhengPozsgayChen} (see also~\cite{FukaiPozsgayVona}). Cycle powers $C_M^k$ are claw-free for every $k$. For each of three random choices of integer couplings, every central sector is certified by Proposition~\ref{prop:pauli}: by a factor whose Galois group contains the alternating group (for $C_{17}^2$, of degree $128$), or, for $C_M^4$ with $M=11,13,14$, by the asymmetry of the spectrum. By the remark on specialization in Section~\ref{sec:procedure}, this proves the following. \begin{theorem}\label{thm:fendley} For generic couplings, in particular for algebraically independent couplings, no Pauli realization of the frustration graph $C_M^2$ of the periodic Fendley chain has a free-fermion solution of the Chapman--Elman--Mann form for $M=7,10,11$ and $13\le M\le19$; the same holds for $C_M^3$ with $13\le M\le18$ and for $C_M^4$ with $M\in\{11,13,14,15,17,18\}$. \end{theorem} For $C_M^2$ with $M=8,9,12$ the sector spectra are degenerate and small, and the test has no power; for $M=8,9$ exact free solutions are reported in~\cite{AnonFrustration}, which also finds numerically that $C_7^2$ and $C_{10}^2$ have none. The undecided graphs are mostly those with $\alpha=2$. For them, if the spectrum is symmetric under $E\mapsto-E$, the factors of the polynomial in $E^2$ have degree at most $D/2$, and a factor in $E$ is either even, so imprimitive, or paired with its reflection, so of degree at most $D/2$. Then either $d\,2^{\alpha-1}>D$ fails or $A_d$ cannot occur, and deciding them requires a test of free form that is not vacuous on parity halves. \subsection{Hidden group structure from multiplicative relations}\label{sec:relations} Galois automorphisms preserve every multiplicative relation $\prod_i\lambda_i^{n_i}=1$ among eigenvalues~\cite{Kowalski,Smyth}. The relation lattice and the configuration of weights it induces identify the algebraic group in which an operator lives: a hypercube or half-cube for free fermions, a grid $\{1,\dots,a\}\times\{1,\dots,b\}$ for a tensor product $A\otimes B$, a cross-polytope for a generic orthogonal or symplectic matrix, and no relations for a generic matrix. We compute the relations by lattice reduction on the logarithms of the eigenvalues at $2000$ to $4000$ bits, keep relations with residual at most $2^{-p/2}$ at a precision of $p$ bits (every accepted relation has residual at most $2^{-1997}$), and identify the configuration against a library of these shapes; the relations are verified numerically, not exactly. Table~\ref{tab:relations} shows that the factor structure is recovered from the characteristic polynomial alone, also after conjugation by a random unimodular matrix: the number of free-fermion modes, the sizes of hidden tensor factors, and mixtures of the two. Spectral methods cannot distinguish free fermions from a product of independent qubits, which give the same cube. \begin{table}[ht] \centering\small \begin{tabular}{@{}P{0.38\textwidth}lll@{}} \toprule operator & expected rank & found & configuration \\ \midrule $A\otimes B$: $2\times4$, $3\times3$, $2\times3$, $4\times4$, $2\times2\times2$ & $\sum_ia_i-r+1$ & $5,5,4,7,4$ & grids (cube) \\ generic, size $6,9,16,32$ & full & full & no relations \\ Ising, zero field, $L=4,\dots,7$ & $L+1$ & $5,6,7,8$ & half-cube $D_L$ \\ Ising, real field, $L=5$ & $32$ & $32$ & no relations \\ matchgate circuits, $n=5,6$ & modes${}+1$ & $5,7$ & cube, $D_6$ \\ XXZ, eight-vertex (integrable), $L=6$ & full & full & no relations \\ Ising ($L=4,5$) $\otimes$ generic $3\times3$ & $L+3$ & $7,8$ & simplex $\times$ half-cube \\ \bottomrule \end{tabular} \caption{Rank of the multiplicative group generated by the distinct eigenvalues and the recovered configuration. For a tensor product of $r$ factors of sizes $a_1,\dots,a_r$ the expected rank is $\sum_ia_i-r+1$; ``full'' means equal to the number of distinct eigenvalues. Circuit matrices are integer multiples of rational orthogonal matrices, and the scale adds one to the rank; for odd $n$, matchgate circuits have a zero mode that lowers it by one. The Ising entries use random couplings.}\label{tab:relations} \end{table} \subsection{Quantum circuits}\label{sec:circuits} Circuits of real orthogonal two-qubit gates with rational entries (rational points on the circle) have rational unitaries. Matchgates $G(A,B)$, acting as $A$ on $\mathrm{span}\{|00\rangle,|11\rangle\}$ and as $B$ on $\mathrm{span}\{|01\rangle,|10\rangle\}$ with $\det A=\det B$, on nearest-neighbor qubits generate free-fermionic circuits~\cite{Valiant,TerhalDiVincenzo,JozsaMiyake}. On $5$, $6$, and $7$ qubits, brickworks of matchgates are free, with longest cycles $4,12,12$, also after conjugation by random local rotations or, on $5$ and $6$ qubits, by a random rational orthogonal matrix of full size, since the test depends only on the spectrum; so are next-nearest-neighbor matchgates routed through fermionic swaps and Floquet matchgate circuits. The screen flags $25$ circuits: on each of $5$, $6$, and $7$ qubits, brickworks of generic gates (longest cycles $30,62,126$), a single generic gate repeated as a Floquet circuit, and matchgate circuits with one generic gate, with one SWAP, with next-nearest-neighbor matchgates without fermionic swaps, with gates $G(A,B)$ having $\det A=-\det B$, with a single-qubit kick, or with alternate layers conjugated by a generic single-qubit rotation; and matchgates on the edges of a $2\times3$ grid. By Remark~\ref{rem:selfrecip} all $25$ are certified not free: each has a factor of the half-degree polynomial whose Galois group contains $A_d$ with $d>n$. For periodic brickworks at even $n$, each parity sector is free with its own modes, as for the Ising sectors of Section~\ref{sec:onsager}, and a hypercube labeling of each sector is recovered. Clifford-type circuits have roots-of-unity spectra and pass the screen; their spectra are highly degenerate, which is why step 1 alone is not a proof of freeness. %=============================================================================================================== \section{What the Galois group sees}\label{sec:integrable} The Galois groups of sector factors grade the algebraic structure of a spectrum (Table~\ref{tab:grade}). The bounds in the first two rows are Lemma~\ref{lem:hyper} and Proposition~\ref{prop:numcons}, and equality with the full group is observed through Frobenius statistics; the last row is certified in every case listed; the third is observed. \begin{table}[ht] \centering\small \begin{tabular}{@{}P{0.3\textwidth}P{0.28\textwidth}P{0.34\textwidth}@{}} \toprule spectral structure & Galois group of a sector factor & examples \\ \midrule free fermions & in $W(B_L)$ or $W(D_L)$ & Ising strips, matchgate circuits, Fendley's chains \\ free fermions, particle number conserved & in $S_L$ on $N$-element subsets & tight-binding chains, Alexander braid representations \\ partially decoupled Bethe equations & imprimitive (observed), blocks over real cyclotomic fields & totally asymmetric exclusion \\ interacting, integrable or not & contains $A_d$ & XXZ, eight-vertex, symmetric exclusion, $M_1$, Baxter--Wu \\ \bottomrule \end{tabular} \caption{Galois groups of factors of sector polynomials.}\label{tab:grade} \end{table} \emph{Number conservation.} For tight-binding chains with random integer hoppings and on-site energies, with $(L,N)=(8,3)$, $(9,4)$, and $(10,3)$, the $N$-particle sector polynomial is irreducible of degree $\binom LN$, the longest Frobenius cycle equals Landau's function $g(L)=15,20,30$, and over $1500$ primes the frequencies of cycle types agree with the class distribution of $S_L$ acting on $N$-element subsets within total variation $0.034$, $0.057$, $0.045$, with no Frobenius element outside the group. The Alexander braid representations of Appendix~\ref{app:validation} show the same signature. \emph{Interacting integrable models.} Integrable models that are not free behave like generic ones inside their symmetry sectors (Table~\ref{tab:8v}). In inhomogeneous eight-vertex models whose columns share Baxter's invariants $\Delta=(a^2+b^2-c^2-d^2)/(2(ab+cd))$ and $\Gamma=(ab-cd)/(ab+cd)$, so that the transfer matrices commute~\cite{FabriciusMcCoy}, and in the six-vertex model, the sector factors have Galois groups containing alternating groups of full degree, while free-fermion columns, $a^2+b^2=c^2+d^2$~\cite{FanWu}, give small groups. \begin{table}[ht] \centering\small \begin{tabular}{@{}lll@{}} \toprule columns & factor degrees & largest factor \\ \midrule free fermion, all different & $4\ (\times4)$, $12\ (\times4)$ & longest cycle $4$ \\ integrable, fixed $(\Delta,\Gamma)$ & $12$ or $16$ & contains $A_{12}$, $A_{16}$ \\ six-vertex, fixed $\Delta$ & $1,6,10,15$ & contains $A_{15}$ \\ generic & $16\ (\times4)$ & contains $A_{16}$ \\ \bottomrule \end{tabular} \caption{Inhomogeneous eight-vertex transfer matrices, $L=6$, $200$ primes per factor, with, in the integrable rows, $\Delta\ne0$ and, except in the six-vertex row ($\Gamma=1$), away from the degenerate values $\Delta=\pm1$, $\Gamma=\pm1$, $\Delta=\pm\Gamma$. Containment of $A_{12}$, $A_{15}$, $A_{16}$ certifies non-freeness by Proposition~\ref{prop:variants}, since $12,15,16>6$.}\label{tab:8v} \end{table} The lattice supersymmetric $M_1$ model~\cite{Nicolai,FendleySchoutensdeBoer}, whose chain is equivalent to the XXZ chain at $\Delta=-1/2$, shows its supersymmetry exactly in the characteristic polynomial: every nonzero factor of the sector with $f$ fermions divides the polynomial of a sector with $f\pm1$ fermions, and the weighted count of zero modes equals the Witten index. Its sectors on chains, ladders, a triangular strip, a $3\times4$ grid, and trees nevertheless contain alternating groups, for example $A_{44}$ and $A_{76}$ on an open chain of $12$ sites. These results agree with the large Galois groups of Bethe equations for the Heisenberg chain~\cite{BellwoodMunro} and with the theorem that the Galois group of the Bethe equations of the $\mathrm{GL}_d$ Gaudin model with one-box weights at all sites contains the alternating group whenever the target shape is neither a hook nor self-transpose~\cite[Thm.~C]{LiaoRybnikov}. The symmetric exclusion process on $14$ sites with $7$ particles gives $A_{32}$, $A_{27}$, and $A_{20}$ in the sector of momentum $0$. \emph{The totally asymmetric exclusion process.} On a ring of $L$ sites with $N$ particles, the Bethe equations of this process decouple, given one collective parameter~\cite{GwaSpohnPRL,GwaSpohnPRA}, and its transition probabilities are determinants~\cite{Schutz}; the generator has integer entries. Over $\Q$ the $N$-particle sector splits into norms of momentum sectors, so we use the rational sectors of momentum $0$ and $\pi$. In every case examined these contain factors whose Galois groups contain alternating groups far above the free bounds, for example $A_{63}$ for $(L,N)=(14,7)$, $A_{80}$ for $(15,5)$, and $A_{131}$ for $(13,6)$, which proves that no sector spectrum is of free form. Other factors are not alternating. For $(L,N)=(14,7)$ the sector of momentum $0$ has a factor of degree $84$ with no Jordan certificate over $600$ primes, whose Frobenius cycle types are compatible only with $3$ blocks of $28$; among $400$ further primes, at each of the $267$ inert in $\Q(\cos2\pi/7)$ every cycle length is divisible by $3$, the pattern of a norm from $\Q(\cos2\pi/7)$. For $(15,5)$ factors of degree $90$ and $30$ are compatible only with $2$ blocks and with norms from $\Q(\sqrt5)$, at the $203$ inert primes among $400$. In both cases the block field is $\Q(\zeta_g)^+$ with $g=\gcd(L,N)$. No blocks occur for $(13,4)$, $(15,3)$, $(16,4)$, and $(12,6)$, where $\Q(\zeta_g)^+=\Q$ and all large factors contain alternating groups, nor for $(10,5)$, where the sector of momentum $0$ has factors of degree $15$, containing $A_{15}$, and of degree $5$. These block structures are observed, not proved. Golinelli and Mallick explain high-order degeneracies of the spectrum of this process by a hidden symmetry of its Bethe ansatz~\cite{GolinelliMallick} (see also~\cite{GolinelliMallickReview}); whether that symmetry accounts for these block structures is open. Integrability organizes the spectrum into sectors and commuting families. Free fermions in addition make every eigenvalue a monomial in a few algebraic numbers, which confines the Galois group to a Weyl group; partial decoupling of Bethe equations appears to leave an intermediate, imprimitive trace; in the other interacting models tested no trace remains. The test is therefore a test for free-fermionic spectral structure, in any basis and in disguise, and not for integrability. Like any test through Weyl groups, it cannot distinguish types $B_L$ and $C_L$, whose Weyl groups coincide: a symplectic hidden group, with eigenvalues of the form $c\,\mu_j^{\pm1}$, can pass the screen. Appendix~\ref{app:validation} gives an example, which the relation lattice of Section~\ref{sec:relations} separates. The analogy with zeros of $L$-functions is close: there Frobenius elements equidistribute in a monodromy group whose type determines the random-matrix statistics~\cite{KatzSarnak}; here the Frobenius statistics of a single characteristic polynomial reflect its Galois group, and in the examples above, at random parameters, they are consistent with the full Weyl group of the hidden group: $W(D_L)$ for free fermions and $S_L$ with particle-number conservation. %=============================================================================================================== \section{Outlook}\label{sec:outlook} The converse of Lemma~\ref{lem:hyper} fails for individual matrices, since a matrix with rational eigenvalues has trivial Galois group. For families one may ask whether a Galois action at generic parameter points through a subgroup of $W(B_L)$ acting on sign vectors, with each sector of size $2^L$ or $2^{L-1}$, forces free form identically in the parameters. The symplectic example of Appendix~\ref{app:validation} shows that abstract embeddability of the Galois groups in $W(B_L)$ does not suffice; whether its Galois action is of this finer form is not settled by our data. More generally the method reads off a hidden reductive group through its Weyl group and weights: $\mathrm{Spin}(2L)$ for free fermions, $U(L)$ with particle-number conservation, the reflection group $C_3\wr S_L$ for free parafermions with $\Z_3$ symmetry, and products of general linear groups for hidden tensor products. The imprimitive groups of the exclusion process suggest a finer invariant that measures partial decoupling of Bethe equations; proving the block structure of Section~\ref{sec:integrable} is open. The limiting factor is size: the screen has no power when all factors have small degree, which leaves part of Section~\ref{sec:frontier} undecided. %=============================================================================================================== \appendix \section{Validation on known models}\label{app:validation} Unless stated otherwise, couplings are integers drawn at random across the row and constant along the transfer direction, which removes translation symmetry. ``Free'' means that the screen finds no certificate under the criteria of Section~\ref{sec:lemma}; ``not free'' means certified by Corollary~\ref{cor:cert}, Proposition~\ref{prop:variants}, Remark~\ref{rem:selfrecip}, or Remark~\ref{rem:para}. The maximal element orders are $\maxord(B_L)=8,12,12,24,30$ for $L=4,\dots,8$. \subsection{Ising strips and fields} In zero field, with couplings random in sign and size so that plaquettes are frustrated, each sector polynomial is irreducible of degree $2^{L-1}$, and the longest Frobenius cycle over $150$ primes equals $\maxord(B_L)$: $12$ at $L=6$ and $24$ at $L=7$. Over $3000$ primes the frequencies of cycle types agree with the class distribution of $W(D_L)$ acting on the $2^{L-1}$ even sign vectors, of order $2^{L-1}L!$ ($23040$ at $L=6$, $322560$ at $L=7$), within total variation $0.02$ to $0.034$, with no Frobenius element outside the group. A real field makes the polynomial irreducible of degree $2^L$, with a full $2^L$-cycle; at $L=6$ its Galois group contains $A_{64}$. At the imaginary field $h=i\pi/2$ the polynomial satisfies $f(x)=g(x^2)$ and $g$ has the same hyperoctahedral image; at $h=i\pi r$ with $r=1/3,1/4,1/6$ the spectrum is not free. The Ising model on the M\"obius strip and the Klein bottle~\cite{LuWu} is free, as is a field $i\pi/2$ on alternate sites; one non-planar chord per row gives $A_{64}$ at $L=7$. \subsection{Eight-vertex models} Deforming free-fermion columns by $d_i\mapsto d_i+t$, $t\in\{-6,\dots,6\}$, the test is free exactly at $t=0$ and certifies non-freeness at the other $12$ values; in a two-parameter deformation $(t,u)\in\{-2,\dots,2\}^2$ it is free exactly at the origin. \subsection{Crossing and four-spin couplings} For the square-lattice Ising model with nearest-neighbor couplings, crossing diagonal couplings $L_1,L_2$, and a four-spin coupling $M$ on each plaquette, the map of Ising models with two- and four-spin couplings to an eight-vertex model~\cite{KadanoffWegner,Wu} and the Fan--Wu condition~\cite{FanWu} give, after the nearest-neighbor weights cancel, the free-fermion condition \begin{equation}\label{eq:plaquette} e^{4M}=\frac{\cosh2(L_1-L_2)}{\cosh2(L_1+L_2)}. \end{equation} We have not found condition~\eqref{eq:plaquette} stated in this form in~\cite{KadanoffWegner,Wu,FanWu}. Without four-spin coupling it reduces to $\sinh2L_1\sinh2L_2=0$, that is, to $L_1L_2=0$ for real couplings. On $15$ integer points of~\eqref{eq:plaquette}, including sign weights (complex couplings), the test reports free; on $18$ nearby points off it, it certifies non-freeness. On grids of $2744$ points at zero field and at field $i\pi/2$ ($L=6$), the $2640$ points off~\eqref{eq:plaquette} in each grid are certified not free; the other $104$ points are degenerate solutions of~\eqref{eq:plaquette}, with a diagonal coupling or the four-spin weight trivial or a pure sign, and are flagged free, so the nontrivial part of the locus is tested only on the $15$ points above. With both diagonal couplings neither trivial nor a pure sign and no four-spin term, all $36$ points of a random-column scan ($L=6$) are certified not free; all $189$ such points with uniform couplings at $L=7$ are certified not free. These nulls are consistent with Barahona's theorem that spin-glass problems become hard off planar lattices or in a field~\cite{Barahona}. \subsection{Three-dimensional models} The three-dimensional model of Wang and Hazzard~\cite{WangHazzard} is solvable at imaginary couplings $K_\perp=\beta h=i\pi/4$. With per-link random weights on its locus the test reports free (factors of degree $8$, longest cycle $6\le\maxord(B_4)$, with $4$ modes per flux sector); off the locus it certifies non-freeness, with a factor of degree $256$ containing $A_{256}$. Its bond graph is non-planar, but the bonds with non-Clifford weights form a perfect matching within each layer, and all three-dimensional connectivity runs through bonds and fields at $i\pi/4$. In a scan of Ising models on the non-planar bars $G\times\Z$ with cross-sections $G=K_4$, $2\times3$, prism, and $2\times4$, with real and Gaussian-integer couplings, all $563$ points are certified not free, by alternating groups or by cycle lengths, such as $5$, $7$, or $9$, that divide no element order allowed by Proposition~\ref{prop:variants}, also with two sectors (Remark~\ref{rem:twosector}). With all couplings equal to $i\pi/4$ the transfer matrix $C$ of the bars with cross-sections $2\times3$ and prism is a Clifford operator, and $C^k$ is a scalar for some $k$; this is a point of the tractable set of~\cite{JVW}. Generic bonds give a transfer matrix $T=CB$ with $B$ a product of exponentials of Pauli strings, so $T^k$ is a scalar times a product of exponentials of the Pauli strings in the orbit of those of $B$ under conjugation by $C$. The Lie algebra generated by these strings is, by~\cite{AguilarCichyEisertBittel}, a direct sum of free-fermionic, orthogonal, symplectic, or unitary algebras or of controlled versions of these (tensored with the group algebra of a group of commuting Pauli strings); $T^k$ lies in the corresponding complexified group, which determines the structure of the spectrum of $T^k$, and hence that of $T$ up to $k$-th roots of unity. With one generic bond per period, all $12$ tests are flagged free, $6$ of them with no factor of degree above $4$; with two bonds, $4$ of $12$ are flagged free, $1$ of them without power, and $8$ are certified not free; with three or four bonds, all $24$ are certified. The dimension of this Lie algebra separates the flagged from the certified points: at every certified point it is at least $240$. At the flagged points it is abelian ($7$ points without power and $2$ others), $\mathfrak{su}(2)^3$ in each sector of the global spin flip ($3$ points), or $\mathfrak{su}(2)$ tensored with the group algebra of a group of $32$ commuting Pauli strings ($3$ points, one with further central elements), all free-fermionic, except at one point with two bonds, where it is $\mathfrak{sp}(8)$ in each sector of the global spin flip. There each sector of $T^k$ is the sum of four copies of the defining representation of $\mathrm{Sp}(8)$, with eigenvalues $c\,\mu_j^{\pm1}$, $j=1,\dots,4$, which are not of free form. The Galois test cannot distinguish this case, since the Weyl groups of types $B_4$ and $C_4$ coincide, but the relation lattice of Section~\ref{sec:relations}, applied to the transfer matrix itself, finds on its two factors of degree $16$ a cross-polytope, as for an orthogonal or symplectic matrix, not a cube. \subsection{Calibrations} Table~\ref{tab:calib} lists further models known to be free or not. \begin{table}[ht] \centering\small \begin{tabular}{@{}P{0.34\textwidth}P{0.17\textwidth}lP{0.25\textwidth}@{}} \toprule model & size & expected & result \\ \midrule transverse-field Ising chain & $N=5,\dots,8$ & free & longest cycles $12,12,24,30$ \\ same, plus longitudinal field & $N=5,\dots,8$ & not free & full cycles, $A_{2^N}$ \\ Fendley's chain $\sum b_mZ_mZ_{m+1}X_{m+2}$~\cite{FendleyDisguise} & $S=2,\dots,5$ modes & free & longest cycles $4,4,6,12$ \\ free parafermions~\cite{FendleyParafermions} & $L=3,4$ & free & $f(x)=g(x^3)$, within $C_3\wr S_L$ \\ parafermions plus $\tau^\dagger$ terms & $L=3,4$ & not free & $A_9$, $A_{27}$ \\ \bottomrule \end{tabular} \caption{Calibrations. For Hamiltonians the energies are additive and the bound is $\maxord(B_S)$ in the number $S$ of modes. Fendley's chain has no Jordan--Wigner map to free fermions and its frustration graph contains no claw; the control with a longitudinal field has a claw~\cite{ChapmanFlammia,ElmanChapmanFlammia}.}\label{tab:calib} \end{table} \subsection{Further models} The three-spin model of Baxter and Wu~\cite{BaxterWu}, integrable and not free, is certified not free on uniform strips of widths $7$ and $8$ (cycles $102$ and $66$, divisible by the primes $17$ and $11$, which divide no allowed element order, and $A_{15}$, $A_{24}$) and with random columns at widths $6$ and $7$ ($A_{16}$, $A_{128}$). The Kitaev--Heisenberg model~\cite{ChaloupkaJackeliKhaliullin} on a $2\times4$ ladder has four factors of degree $64$ containing $A_{64}$; the pure Kitaev ladder has only factors of degree $4$, where the test has no power. Among non-Hermitian chains, the transverse-field Ising chain with an imaginary transverse field ($N=6,7$, realified) satisfies $f(x)=g(x^2)$ with longest cycles $12$ and $24$, consistent with free form, while the Ising chain with an imaginary longitudinal field gives $A_{64}$ and $A_{128}$. A tight-binding chain with a quasiperiodic on-site potential of rational frequency $1/6$ ($L=9$, $N=4$) has a factor of degree $126$ with longest cycle $20=g(9)$ and total variation $0.117$ from $S_9$ on $4$-element subsets over $300$ primes; the chain of Su, Schrieffer, and Heeger~\cite{SuSchriefferHeeger} on $8$ sites has factors too small for the test. For the toric code~\cite{KitaevToric} on a $2\times2$ torus with random couplings every factor is linear, and the spectrum is a constant plus the sums over two even half-cubes of the vertex and plaquette couplings. \subsection{Knot invariants and the Tutte plane} Braid-group representations on $(\C^2)^{\otimes n}$ from the free-fermion six-vertex $R$-matrix, which yields the Alexander--Conway polynomial~\cite{KauffmanSaleur}, decompose into exterior powers of the reduced Burau matrix with factor degrees $\binom{n-1}{k}$, and their Galois groups lie in $S_{n-1}$ (longest cycles $6$ for $n=6,7$). Those from the $U_q(\mathfrak{sl}_2)$ $R$-matrix (Jones polynomial) at rational $q$ contain $A_9$ ($n=6$) and $A_{14}$ ($n=7$) in all $16$ long braid words tested. In the semisimple path representations at $q=e^{i\pi/k}$, all eigenvalues are roots of unity exactly for $k\in\{3,4,6\}$ (and $q=\pm1$), and not for $k=5,7,8,10$; rational $q\ne\pm1$ never collapses. With Hecke parameter $q^2=e^{2\pi i/k}$, these are points at which the Jones polynomial is computable in polynomial time~\cite{JVW}, and for $k\ge3$ they are the orders at which the image of the braid group in the Jones representations is finite for every $n$: it is infinite modulo its center for $k\ge5$, $k\ne6,10$, $n\ge3$, and for $k=10$, $n\ge5$, while the finite images at $k=3,4,6$ (and $k=10$, $n=3,4$) were identified earlier by Jones and others~\cite[Thm.~2.3 and Sec.~4]{FLW}. For Potts transfer matrices on planar strips the test is free for $Q=2$, the Ising hyperbola of the Tutte plane, at widths $W=6,7,8$, with longest cycles $12,24,30$ equal to $\maxord(B_W)$, and certifies non-freeness for $Q=3,4$. \subsection{Supersymmetric models and hard squares} For the hard-core lattice gas at activity $z=-1$, which governs the ground states of the $M_1$ model~\cite{HuijseSchoutens}, the row transfer matrix of hard squares has only roots of unity as eigenvalues for every width and boundary condition tested (periodic widths $6$ to $13$, open widths $2,5,8,11$), as known~\cite{FendleySchoutensvanEerten,Jonsson,Baxter}; at $z=\pm2,-\tfrac12,-3,-\tfrac13,1$ (periodic width $10$) it does not, and the groups contain alternating groups. Hard hexagons, the honeycomb lattice, and the king's graph at $z=-1$ give alternating groups. \begin{table}[ht] \centering\small \begin{tabular}{@{}P{0.3\textwidth}P{0.3\textwidth}P{0.32\textwidth}@{}} \toprule setting & flagged free & known free or roots-of-unity points \\ \midrule Ising strips, real and imaginary fields & zero field; $h=i\pi/2$ & planar zero field~\cite{Onsager,Kaufman}; $i\pi/2$ \\ eight-vertex & $a^2+b^2=c^2+d^2$ & Fan--Wu~\cite{FanWu} \\ crossing and four-spin couplings & \eqref{eq:plaquette} & Fan--Wu via the vertex map \\ three-dimensional Ising & Wang--Hazzard locus & \cite{WangHazzard} \\ braid representations & Alexander; Jones at $k=3,4,6$ & free~\cite{KauffmanSaleur}; roots of unity~\cite{FLW} \\ circuits & matchgates & \cite{Valiant,JozsaMiyake} \\ number-conserving chains & $S_L$ on subsets & tight-binding chains \\ \bottomrule \end{tabular} \caption{Points flagged free in the main families tested; further models are in Table~\ref{tab:calib} and the text. Points with roots-of-unity spectra pass the screen without being free-fermionic in any useful sense; the bars with a Clifford background and generic bonds of Appendix~\ref{app:validation} are explained there by the Lie algebra of a Pauli orbit, and one of them is symplectic, not free.}\label{tab:summary} \end{table} \section{Recoveries of known single-particle spectra}\label{app:recoveries} Section~\ref{sec:onsager} recovers Onsager's dispersion relation; this appendix collects the corresponding recoveries for further free-fermion and free-parafermion models. In each case below, step~2 of the procedure (recovery of the hypercube labeling) was applied to the exact characteristic or minimal polynomial with no input from the known solution, and the recovered single-particle polynomial, recognized over $\Q$ (or over $\Q(\omega)$), was then compared with the published formula. Where stated, step~3 certified the result exactly: the power sums of the free form, computed from the recovered polynomial alone, agree with the exact traces for every exponent up to the dimension, so the characteristic polynomial is the composed product. Transfer matrices use the multiplicative form; Hamiltonians use the additive form $E_s=e_0+\sum_k s_k\varepsilon_k$, for which the recovered object is $\prod_k(y-\varepsilon_k^2)$. \begin{table}[ht] \centering\small \begin{tabular}{@{}P{0.3\textwidth}P{0.17\textwidth}P{0.33\textwidth}l@{}} \toprule model & recovered & agreement with the literature & certificate\\ \midrule dimers, strip width $3\le L\le 8$ & $\prod_k(y-\eta_k^2)$ & $\eta_k=\cos\frac{\pi k}{L+1}$, exact~\cite{Kasteleyn,TemperleyFisher,Lieb} & exact\\ transverse-field Ising chain, open, $4\le L\le 6$ & $\prod_k(y-\varepsilon_k^2)$ & exact~\cite{LiebSchultzMattis,Pfeuty} & exact\\ XY chain, open, $4\le L\le 6$ & $\prod_k(y-\varepsilon_k^2)$ & exact~\cite{LiebSchultzMattis} & exact\\ Kitaev honeycomb, flux-free sector & $\varepsilon_q$ & $|2(J_xe^{iq\cdot n_1}+J_ye^{iq\cdot n_2}+J_z)|$~\cite{Kitaev} & numerical\\ chain in disguise, $8\le N\le 15$ & $\prod_k(y-\varepsilon_k^2)$ & exact~\cite{FendleyDisguise} & exact\\ free parafermions, $n=3$, $L=3,4$ & $\prod_k(y-\varepsilon_k^3)$ & exact~\cite{FendleyParafermions} & exact\\ Fendley--Pozsgay chains, $3\le L\le7$ & $\prod_k(y-\varepsilon_k^2)$ & exact~\cite{FendleyPozsgay} & none\\ Ising on the M\"obius strip, $4\le L\le 7$ & signed hypercube & numerical~\cite{LuWu} & numerical\\ \bottomrule \end{tabular} \caption{Recoveries of known single-particle spectra. ``Exact'' agreement means a rational identity between the recovered polynomial and the published one; an exact certificate means agreement of all power sums with the exact traces.}\label{tab:recoveries} \end{table} \emph{Dimers.} For the dimer transfer matrix on strips of width $L$, the recovered numbers $\eta_k=\sinh w_k$ satisfy $\prod_k(y-\eta_k^2)=\prod_{k=1}^{L}(y-\cos^2\frac{\pi k}{L+1})$ identically, the Chebyshev form of the solution of Kasteleyn, Temperley and Fisher in Lieb's transfer-matrix formulation~\cite{Kasteleyn,TemperleyFisher,Lieb}; the certificate holds with $\kappa^2=1$ for all $m\le 2^L$. \emph{Quantum chains.} For open transverse-field Ising and XY chains with random integer couplings, the recovered polynomial is the characteristic polynomial of the Majorana single-particle matrix of Lieb, Schultz, and Mattis~\cite{LiebSchultzMattis}; for uniform unit couplings, the critical chain with $J=h=1$, it gives $\varepsilon_k=2\cos\frac{\pi k}{2L+1}$ to $140$ digits, as in Pfeuty's solution~\cite{Pfeuty}. \emph{Kitaev's honeycomb model.} On $2\times2$ and $3\times2$ tori, the flux-free sector reproduces the Majorana dispersion of~\cite{Kitaev} to $10^{-14}$, the four loop sectors corresponding to momentum shifts by $(0,0)$, $(0,\tfrac12)$, $(\tfrac12,0)$, $(\tfrac12,\tfrac12)$; only the flux sector with all plaquette operators equal to $+1$ has this form. \emph{Free fermions in disguise.} For the chain of~\cite{FendleyDisguise} with $8\le N\le 15$ sites, the recovered polynomial is $y^SP_M(1/y)$ with $P_M$ given by Fendley's recursion, exactly, in every case, with exact certificates. Here no Jordan--Wigner transformation is available, and the single-particle data are obtained from integer arithmetic alone. The number of levels is $S=\lceil M/3\rceil=\lfloor N/3\rfloor$, where $M=N-2$ is the number of terms, consistent with the example $P_4$ in~\cite{FendleyDisguise}. \emph{Free fermions beyond Jordan--Wigner.} The chains of Fendley and Pozsgay~\cite{FendleyPozsgay} have a frustration graph that contains claws, so the criterion of Chapman, Elman, and Mann does not apply to them directly. For $3\le L\le7$ ($5$ to $9$ qubits) the screen has no power, since all factors have degree at most $4$, but recovery gives $r=\lfloor(L+1)/2\rfloor$ modes and $\prod_k(y-\varepsilon_k^2)=y^rP(1/y)$ with $P$ as in~\cite[eq.~(4.4)]{FendleyPozsgay}, exactly. \emph{Free parafermions.} For the three-state chain of~\cite{FendleyParafermions}, energies are sums $\sum_j\omega^{a_j}\varepsilon_j$ with $\omega=e^{2\pi i/3}$; the recovered $\prod_k(y-\varepsilon_k^3)$ agrees exactly with Fendley's polynomial, and the traces in $\Z[\omega]$ certify the spectrum for all $m\le 3^L$. For uniform couplings and $L=3$, $\varepsilon_k^3=4\cos^2\frac{\pi k}{7}$. \emph{The M\"obius strip.} The transfer matrix composed with the site reflection has the same moduli hypercube as the open strip; its signs are reproduced to $140$ digits by reflection parities $(-1)^{k+1}$ of the modes, ordered by increasing $w_k$, times the reordering sign $(-1)^{\binom{|A|}{2}}$. Since this labeling depends on the ordering of real numbers, it is not defined over $\Q$, and the identification is numerical; compare~\cite{LuWu}. %=============================================================================================================== {\raggedright \begin{thebibliography}{99} \bibitem{AguilarCichyEisertBittel} G.~Aguilar, S.~Cichy, J.~Eisert, and L.~Bittel, \emph{Full classification of Pauli Lie algebras}, preprint (2024), \href{https://arxiv.org/abs/2408.00081}{arXiv:2408.00081}. \bibitem{AnonFrustration} Anonymous, \emph{Which frustration graphs give free fermions? A converse to the even-hole, claw-free theorem, and exact free sectors of the periodic Fendley chain}, preprint, Zenodo (2026), \href{https://doi.org/10.5281/zenodo.22994979}{doi:10.5281/zenodo.22994979}. \bibitem{Barahona} F.~Barahona, \emph{On the computational complexity of Ising spin glass models}, J.\ Phys.\ A \textbf{15} (1982), no.~10, 3241--3253, \href{https://doi.org/10.1088/0305-4470/15/10/028}{doi:10.1088/0305-4470/15/10/028}. \bibitem{Baxter} R.~J.~Baxter, \emph{Hard squares for $z=-1$}, Ann.\ Comb.\ \textbf{15} (2011), no.~2, 185--195, \href{https://doi.org/10.1007/s00026-011-0089-2}{doi:10.1007/s00026-011-0089-2}. \bibitem{BaxterWu} R.~J.~Baxter and F.~Y.~Wu, \emph{Exact solution of an Ising model with three-spin interactions on a triangular lattice}, Phys.\ Rev.\ Lett.\ \textbf{31} (1973), 1294--1297, \href{https://doi.org/10.1103/PhysRevLett.31.1294}{doi:10.1103/PhysRevLett.31.1294}. \bibitem{BellwoodMunro} O.~R.~Bellwood and W.~J.~Munro, \emph{Algebraic structure of finite-size Bethe solutions in the Heisenberg chain}, preprint (2026), \href{https://arxiv.org/abs/2605.05589}{arXiv:2605.05589}. \bibitem{BourbakiCA} N.~Bourbaki, \emph{Commutative Algebra. Chapters 1--7}, Elements of Mathematics, Springer, Berlin, 1989. \bibitem{ChaloupkaJackeliKhaliullin} J.~Chaloupka, G.~Jackeli, and G.~Khaliullin, \emph{Kitaev--Heisenberg model on a honeycomb lattice: possible exotic phases in iridium oxides $A_2\mathrm{IrO}_3$}, Phys.\ Rev.\ Lett.\ \textbf{105} (2010), 027204, \href{https://doi.org/10.1103/PhysRevLett.105.027204}{doi:10.1103/PhysRevLett.105.027204}. \bibitem{ChapmanElmanMann} A.~Chapman, S.~J.~Elman, and R.~L.~Mann, \emph{A unified graph-theoretic framework for free-fermion solvability}, preprint (2023), \href{https://arxiv.org/abs/2305.15625}{arXiv:2305.15625}. \bibitem{ChapmanFlammia} A.~Chapman and S.~T.~Flammia, \emph{Characterization of solvable spin models via graph invariants}, Quantum \textbf{4} (2020), 278, \href{https://doi.org/10.22331/q-2020-06-04-278}{doi:10.22331/q-2020-06-04-278}. \bibitem{DixonMortimer} J.~D.~Dixon and B.~Mortimer, \emph{Permutation Groups}, Graduate Texts in Mathematics, vol.~163, Springer, New York, 1996, \href{https://doi.org/10.1007/978-1-4612-0731-3}{doi:10.1007/978-1-4612-0731-3}. \bibitem{ElmanChapmanFlammia} S.~J.~Elman, A.~Chapman, and S.~T.~Flammia, \emph{Free fermions behind the disguise}, Comm.\ Math.\ Phys.\ \textbf{388} (2021), no.~2, 969--1003, \href{https://doi.org/10.1007/s00220-021-04220-w}{doi:10.1007/s00220-021-04220-w}. \bibitem{FabriciusMcCoy} K.~Fabricius and B.~M.~McCoy, \emph{New developments in the eight vertex model}, J.\ Stat.\ Phys.\ \textbf{111} (2003), no.~1--2, 323--337, \href{https://doi.org/10.1023/A:1022213209641}{doi:10.1023/A:1022213209641}. \bibitem{FanWu} C.~Fan and F.~Y.~Wu, \emph{General lattice model of phase transitions}, Phys.\ Rev.\ B \textbf{2} (1970), no.~3, 723--733, \href{https://doi.org/10.1103/PhysRevB.2.723}{doi:10.1103/PhysRevB.2.723}. \bibitem{FendleyParafermions} P.~Fendley, \emph{Free parafermions}, J.\ Phys.\ A \textbf{47} (2014), no.~7, 075001, \href{https://doi.org/10.1088/1751-8113/47/7/075001}{doi:10.1088/1751-8113/47/7/075001}. \bibitem{FendleyDisguise} P.~Fendley, \emph{Free fermions in disguise}, J.\ Phys.\ A \textbf{52} (2019), 335002, \href{https://doi.org/10.1088/1751-8121/ab305d}{doi:10.1088/1751-8121/ab305d}. \bibitem{FendleyPozsgay} P.~Fendley and B.~Pozsgay, \emph{Free fermions beyond Jordan and Wigner}, SciPost Phys.\ \textbf{16} (2024), 102, \href{https://doi.org/10.21468/SciPostPhys.16.4.102}{doi:10.21468/SciPostPhys.16.4.102}. \bibitem{FendleySchoutensdeBoer} P.~Fendley, K.~Schoutens, and J.~de Boer, \emph{Lattice models with $\mathcal{N}=2$ supersymmetry}, Phys.\ Rev.\ Lett.\ \textbf{90} (2003), 120402, \href{https://doi.org/10.1103/PhysRevLett.90.120402}{doi:10.1103/PhysRevLett.90.120402}. \bibitem{FendleySchoutensvanEerten} P.~Fendley, K.~Schoutens, and B.~van Eerten, \emph{Hard squares with negative activity}, J.\ Phys.\ A \textbf{38} (2005), no.~2, 315--322, \href{https://doi.org/10.1088/0305-4470/38/2/002}{doi:10.1088/0305-4470/38/2/002}. \bibitem{FLW} M.~H.~Freedman, M.~J.~Larsen, and Z.~Wang, \emph{The two-eigenvalue problem and density of Jones representation of braid groups}, Comm.\ Math.\ Phys.\ \textbf{228} (2002), no.~1, 177--199, \href{https://doi.org/10.1007/s002200200636}{doi:10.1007/s002200200636}. \bibitem{FukaiPozsgayVona} K.~Fukai, B.~Pozsgay, and I.~Vona, \emph{Solving models with generalized free fermions II: path-product expansion and conserved charges}, preprint (2026), \href{https://arxiv.org/abs/2605.31453}{arXiv:2605.31453}. \bibitem{GolinelliMallick} O.~Golinelli and K.~Mallick, \emph{Spectral degeneracies in the totally asymmetric exclusion process}, J.\ Stat.\ Phys.\ \textbf{120} (2005), 779--798, \href{https://doi.org/10.1007/s10955-005-6972-7}{doi:10.1007/s10955-005-6972-7}. \bibitem{GolinelliMallickReview} O.~Golinelli and K.~Mallick, \emph{The asymmetric simple exclusion process: an integrable model for non-equilibrium statistical mechanics}, J.\ Phys.\ A \textbf{39} (2006), 12679--12705, \href{https://doi.org/10.1088/0305-4470/39/41/S03}{doi:10.1088/0305-4470/39/41/S03}. \bibitem{GorodnikNevo} A.~Gorodnik and A.~Nevo, \emph{Splitting fields of elements in arithmetic groups}, Math.\ Res.\ Lett.\ \textbf{18} (2011), no.~6, 1281--1288, \href{https://doi.org/10.4310/MRL.2011.v18.n6.a16}{doi:10.4310/MRL.2011.v18.n6.a16}. \bibitem{GwaSpohnPRA} L.-H.~Gwa and H.~Spohn, \emph{Bethe solution for the dynamical-scaling exponent of the noisy Burgers equation}, Phys.\ Rev.\ A \textbf{46} (1992), 844--854, \href{https://doi.org/10.1103/PhysRevA.46.844}{doi:10.1103/PhysRevA.46.844}. \bibitem{GwaSpohnPRL} L.-H.~Gwa and H.~Spohn, \emph{Six-vertex model, roughened surfaces, and an asymmetric spin Hamiltonian}, Phys.\ Rev.\ Lett.\ \textbf{68} (1992), 725--728, \href{https://doi.org/10.1103/PhysRevLett.68.725}{doi:10.1103/PhysRevLett.68.725}. \bibitem{HuijseSchoutens} L.~Huijse and K.~Schoutens, \emph{Supersymmetry, lattice fermions, independence complexes and cohomology theory}, Adv.\ Theor.\ Math.\ Phys.\ \textbf{14} (2010), no.~2, 643--694, \href{https://doi.org/10.4310/ATMP.2010.v14.n2.a8}{doi:10.4310/ATMP.2010.v14.n2.a8}. \bibitem{JVW} F.~Jaeger, D.~L.~Vertigan, and D.~J.~A.~Welsh, \emph{On the computational complexity of the Jones and Tutte polynomials}, Math.\ Proc.\ Cambridge Philos.\ Soc.\ \textbf{108} (1990), no.~1, 35--53, \href{https://doi.org/10.1017/S0305004100068936}{doi:10.1017/S0305004100068936}. \bibitem{Jonsson} J.~Jonsson, \emph{Hard squares with negative activity and rhombus tilings of the plane}, Electron.\ J.\ Combin.\ \textbf{13} (2006), no.~1, R67, \href{https://doi.org/10.37236/1093}{doi:10.37236/1093}. \bibitem{JKZ} F.~Jouve, E.~Kowalski, and D.~Zywina, \emph{Splitting fields of characteristic polynomials of random elements in arithmetic groups}, Israel J.\ Math.\ \textbf{193} (2013), 263--307, \href{https://doi.org/10.1007/s11856-012-0117-x}{doi:10.1007/s11856-012-0117-x}. \bibitem{JozsaMiyake} R.~Jozsa and A.~Miyake, \emph{Matchgates and classical simulation of quantum circuits}, Proc.\ R.\ Soc.\ A \textbf{464} (2008), no.~2100, 3089--3106, \href{https://doi.org/10.1098/rspa.2008.0189}{doi:10.1098/rspa.2008.0189}. \bibitem{KadanoffWegner} L.~P.~Kadanoff and F.~J.~Wegner, \emph{Some critical properties of the eight-vertex model}, Phys.\ Rev.\ B \textbf{4} (1971), no.~11, 3989--3993, \href{https://doi.org/10.1103/PhysRevB.4.3989}{doi:10.1103/PhysRevB.4.3989}. \bibitem{Kasteleyn} P.~W.~Kasteleyn, \emph{The statistics of dimers on a lattice}, Physica \textbf{27} (1961), no.~12, 1209--1225, \href{https://doi.org/10.1016/0031-8914(61)90063-5}{doi:10.1016/0031-8914(61)90063-5}. \bibitem{KatzSarnak} N.~M.~Katz and P.~Sarnak, \emph{Zeroes of zeta functions and symmetry}, Bull.\ Amer.\ Math.\ Soc.\ \textbf{36} (1999), no.~1, 1--26, \href{https://doi.org/10.1090/S0273-0979-99-00766-1}{doi:10.1090/S0273-0979-99-00766-1}. \bibitem{KauffmanSaleur} L.~H.~Kauffman and H.~Saleur, \emph{Free fermions and the Alexander--Conway polynomial}, Comm.\ Math.\ Phys.\ \textbf{141} (1991), 293--327, \href{https://doi.org/10.1007/BF02101508}{doi:10.1007/BF02101508}. \bibitem{Kaufman} B.~Kaufman, \emph{Crystal statistics. II. Partition function evaluated by spinor analysis}, Phys.\ Rev.\ \textbf{76} (1949), 1232--1243, \href{https://doi.org/10.1103/PhysRev.76.1232}{doi:10.1103/PhysRev.76.1232}. \bibitem{KitaevToric} A.~Yu.~Kitaev, \emph{Fault-tolerant quantum computation by anyons}, Ann.\ Physics \textbf{303} (2003), 2--30, \href{https://doi.org/10.1016/S0003-4916(02)00018-0}{doi:10.1016/S0003-4916(02)00018-0}. \bibitem{Kitaev} A.~Kitaev, \emph{Anyons in an exactly solved model and beyond}, Ann.\ Physics \textbf{321} (2006), no.~1, 2--111, \href{https://doi.org/10.1016/j.aop.2005.10.005}{doi:10.1016/j.aop.2005.10.005}. \bibitem{Kowalski} E.~Kowalski, \emph{The large sieve, monodromy, and zeta functions of algebraic curves, 2: independence of the zeros}, Int.\ Math.\ Res.\ Not.\ IMRN \textbf{2008} (2008), rnn091, 57~pp., \href{https://doi.org/10.1093/imrn/rnn091}{doi:10.1093/imrn/rnn091}. \bibitem{LiaoRybnikov} S.~Liao and L.~Rybnikov, \emph{Maximal transitivity of the cactus group on standard Young tableaux}, preprint (2025), \href{https://arxiv.org/abs/2506.16561}{arXiv:2506.16561}. \bibitem{Lieb} E.~H.~Lieb, \emph{Solution of the dimer problem by the transfer matrix method}, J.\ Math.\ Phys.\ \textbf{8} (1967), no.~12, 2339--2341, \href{https://doi.org/10.1063/1.1705163}{doi:10.1063/1.1705163}. \bibitem{LiebSchultzMattis} E.~Lieb, T.~Schultz, and D.~Mattis, \emph{Two soluble models of an antiferromagnetic chain}, Ann.\ Physics \textbf{16} (1961), no.~3, 407--466, \href{https://doi.org/10.1016/0003-4916(61)90115-4}{doi:10.1016/0003-4916(61)90115-4}. \bibitem{LuWu} W.~T.~Lu and F.~Y.~Wu, \emph{Ising model on nonorientable surfaces: exact solution for the M\"obius strip and the Klein bottle}, Phys.\ Rev.\ E \textbf{63} (2001), 026107, \href{https://doi.org/10.1103/PhysRevE.63.026107}{doi:10.1103/PhysRevE.63.026107}. \bibitem{Milewski} J.~Milewski, G.~Banaszak, T.~Lulek, M.~\L{}abuz, and R.~Stagraczy\'nski, \emph{Galois actions on the eigenproblem of the Heisenberg heptagon}, Open Syst.\ Inf.\ Dyn.\ \textbf{19} (2012), no.~2, 1250012, \href{https://doi.org/10.1142/S1230161212500126}{doi:10.1142/S1230161212500126}. \bibitem{Nicolai} H.~Nicolai, \emph{Supersymmetry and spin systems}, J.\ Phys.\ A \textbf{9} (1976), no.~9, 1497--1506, \href{https://doi.org/10.1088/0305-4470/9/9/010}{doi:10.1088/0305-4470/9/9/010}. \bibitem{Onsager} L.~Onsager, \emph{Crystal statistics. I. A two-dimensional model with an order-disorder transition}, Phys.\ Rev.\ \textbf{65} (1944), 117--149, \href{https://doi.org/10.1103/PhysRev.65.117}{doi:10.1103/PhysRev.65.117}. \bibitem{Pfeuty} P.~Pfeuty, \emph{The one-dimensional Ising model with a transverse field}, Ann.\ Physics \textbf{57} (1970), no.~1, 79--90, \href{https://doi.org/10.1016/0003-4916(70)90270-8}{doi:10.1016/0003-4916(70)90270-8}. \bibitem{RegnaultLiuBernevig} N.~Regnault, S.~Liu, and B.~A.~Bernevig, \emph{Integer characteristic polynomial factorization and Hilbert space fragmentation}, preprint (2022), \href{https://arxiv.org/abs/2210.08019}{arXiv:2210.08019}. \bibitem{RuhElman} J.~Ruh and S.~J.~Elman, \emph{Expanding the class of free fermions via twin-collapse methods}, Phys.\ Rev.\ A \textbf{113} (2026), no.~2, 022456, \href{https://doi.org/10.1103/svfv-vdh5}{doi:10.1103/svfv-vdh5}. \bibitem{RuhElmanChapman} J.~Ruh, S.~J.~Elman, and A.~Chapman, \emph{A graph-theoretic analysis of non-generic free-fermion solvability by Krylov decompositions}, preprint (2026), \href{https://arxiv.org/abs/2610.03557}{arXiv:2610.03557}. \bibitem{Schutz} G.~M.~Sch\"utz, \emph{Exact solution of the master equation for the asymmetric exclusion process}, J.\ Stat.\ Phys.\ \textbf{88} (1997), 427--445, \href{https://doi.org/10.1007/BF02508478}{doi:10.1007/BF02508478}. \bibitem{Smyth} C.~J.~Smyth, \emph{Additive and multiplicative relations connecting conjugate algebraic numbers}, J.\ Number Theory \textbf{23} (1986), no.~2, 243--254, \href{https://doi.org/10.1016/0022-314X(86)90094-6}{doi:10.1016/0022-314X(86)90094-6}. \bibitem{Chebotarev} P.~Stevenhagen and H.~W.~Lenstra, Jr., \emph{Chebotar\"ev and his density theorem}, Math.\ Intelligencer \textbf{18} (1996), no.~2, 26--37, \href{https://doi.org/10.1007/BF03027290}{doi:10.1007/BF03027290}. \bibitem{SuSchriefferHeeger} W.~P.~Su, J.~R.~Schrieffer, and A.~J.~Heeger, \emph{Solitons in polyacetylene}, Phys.\ Rev.\ Lett.\ \textbf{42} (1979), 1698--1701, \href{https://doi.org/10.1103/PhysRevLett.42.1698}{doi:10.1103/PhysRevLett.42.1698}. \bibitem{TemperleyFisher} H.~N.~V.~Temperley and M.~E.~Fisher, \emph{Dimer problem in statistical mechanics: an exact result}, Philos.\ Mag.\ \textbf{6} (1961), no.~68, 1061--1063, \href{https://doi.org/10.1080/14786436108243366}{doi:10.1080/14786436108243366}. \bibitem{TerhalDiVincenzo} B.~M.~Terhal and D.~P.~DiVincenzo, \emph{Classical simulation of noninteracting-fermion quantum circuits}, Phys.\ Rev.\ A \textbf{65} (2002), 032325, \href{https://doi.org/10.1103/PhysRevA.65.032325}{doi:10.1103/PhysRevA.65.032325}. \bibitem{Valiant} L.~G.~Valiant, \emph{Quantum circuits that can be simulated classically in polynomial time}, SIAM J.\ Comput.\ \textbf{31} (2002), no.~4, 1229--1254, \href{https://doi.org/10.1137/S0097539700377025}{doi:10.1137/S0097539700377025}. \bibitem{WangHazzard} Z.~Wang and K.~R.~A.~Hazzard, \emph{Topological correlations in three-dimensional classical Ising models: an exact solution with a continuous phase transition}, Phys.\ Rev.\ Research \textbf{5} (2023), 013086, \href{https://doi.org/10.1103/PhysRevResearch.5.013086}{doi:10.1103/PhysRevResearch.5.013086}. \bibitem{Wu} F.~Y.~Wu, \emph{Ising model with four-spin interactions}, Phys.\ Rev.\ B \textbf{4} (1971), no.~7, 2312--2314, \href{https://doi.org/10.1103/PhysRevB.4.2312}{doi:10.1103/PhysRevB.4.2312}. \bibitem{ZhengPozsgayChen} M.~Zheng, B.~Pozsgay, and S.~Chen, \emph{Exact solutions of hidden free fermion models}, preprint (2026), \href{https://arxiv.org/abs/2609.14134}{arXiv:2609.14134}. \end{thebibliography} } \end{document}