A general structural theory for dimer models on non-bipartite periodic graphs: weak limits of Boltzmann measures and spectral curves

Boutillier and de Tilière (2010) solved the critical Z-invariant Ising model on periodic isoradial graphs through Fisher's Ising–dimer correspondence and explicitly recorded what their method did not deliver: Sheffield-type convergence theorems for Gibbs measures fail for non-bipartite dimers (they had to argue weak convergence of Boltzmann measures directly for Fisher graphs); the analysis was tied to periodic isoradial graphs at criticality; and the multiplicative constant relating the dimer and Laplacian characteristic polynomials was controlled only up to a unit phase. Later work by the same authors closed most of this: the locality paper (2011) computed the constant, and the massive-Laplacian program (2017; 2019; 2021) extended the full dimer analysis to every elliptic parameter, periodic or not, with explicit constants. What survives as a genuinely open problem is the structural core: a general theory for dimers on $\mathbb{Z}^2$-periodic non-bipartite planar graphs — a Kenyon–Okounkov–Sheffield/Sheffield-analog convergence theorem and a characterization of the spectral curves, which need not be Harnack — covering Fisher graphs of ferromagnetic planar Ising models beyond the isoradial Z-invariant class. A 2026 paper still describes the spectral-curve side of this question as widely open.

Background

The Ising model on a finite graph assigns spins $\sigma_v\in\{\pm1\}$ to vertices with Boltzmann weight proportional to $\exp(\sum_{uv}\!J_{uv}\sigma_u\sigma_v)$. Fisher (1966) showed that the high-temperature (polygon) expansion of the planar Ising model equals, up to explicit factors, the partition function of the dimer model on a decorated graph $G^{\mathrm{F}}$, the Fisher graph: every vertex of degree $k$ of the original graph is replaced by a decoration of $3k$ vertices, with a triangle attached to each incident edge and the triangles linked in a cycle. The dimer model is the study of perfect matchings — sets of edges covering every vertex exactly once — and Fisher graphs are typically non-bipartite, meaning their vertex set cannot be split into two classes with every edge crossing between them. In Fisher's correspondence the dimer edge weights are $\nu_e=1$ on the decoration edges and $\nu_e=\coth(J_e)$ on the edges coming from the Ising graph; $\tanh(J_e)$ is instead the high-temperature polygon weight, carried by the polygon edges that are absent from the dimer configuration.

On planar graphs the dimer partition function is the Pfaffian of a Kasteleyn matrix: a weighted, oriented adjacency matrix whose signs are chosen so that every face boundary carries an odd number of clockwise-oriented edges. If the graph is periodic, with two independent translations as periods (a $\mathbb{Z}^2$-periodic graph with fundamental domain $G_1$), twisting the Kasteleyn matrix by monomials $(z,w)\in\mathbb{C}^2$ along the two periods gives the characteristic polynomial $P(z,w)=\det K(z,w)$; its zero locus, the spectral curve, controls the thermodynamics and fluctuations of the model. Throughout, $P$ is built from one fixed periodic clockwise-odd (Kasteleyn) orientation of the lifted $\mathbb{Z}^2$-periodic graph, induced from the fundamental domain $G_1$: periodic gauge transformations multiply $P$ by a nonzero constant and leave its zero locus unchanged, while the four torus spin structures $K_n^{\theta\tau}$ entering the partition-function formula act only as the twisted evaluation sectors $\det K_n^{\theta\tau}=\prod_{j,k}P(e^{i(2j+\theta)\pi/n},e^{i(2k+\tau)\pi/n})$ of this same $P$; statements about the zeros of $P$ on the unit torus are made for this fixed convention. Boltzmann measures on larger and larger tori $G_n$ are expected to converge weakly to Gibbs measures (probability measures satisfying the DLR — Dobrushin–Lanford–Ruelle — consistency condition, which prescribes the conditional distribution of configurations in every finite window) on dimer configurations of the infinite graph.

For bipartite periodic dimer graphs this picture is complete: Kenyon, Okounkov and Sheffield proved the spectral curve is a Harnack curve — a real algebraic curve whose amoeba, the image of the curve under $(z,w)\mapsto(\log|z|,\log|w|)$, has maximal area among all real algebraic curves with the same Newton polygon — the Newton polygon (the convex hull of exponents of monomials appearing in $P$) parametrizes the ergodic Gibbs measures, and weak convergence of the Boltzmann measures follows from the subsequence argument of Cohn–Kenyon–Propp together with Sheffield's convergence theorem for Gibbs measures. Every one of these pillars uses bipartiteness, essentially through the factorization of the Kasteleyn matrix between black and white vertices.

The Z-invariant Ising model of Baxter (1978) lives on isoradial graphs — planar graphs embedded so that every face is inscribed in a circle of radius one, each edge $e$ carrying the half-angle $\theta_e$ of its rhombus — with coupling constants $J_e=J(\theta_e|k)$ given by Jacobi elliptic functions of an elliptic parameter $k$; $k=0$ corresponds to the critical temperature on every isoradial graph, and as $k^2$ ranges over $(-\infty,1)$ the whole temperature range is covered. Z-invariance means the partition function is invariant, up to a constant, under the star–triangle (Yang–Baxter) move.

Boutillier and de Tilière (2010) analyzed the Fisher dimer model of the critical Z-invariant Ising model on $\mathbb{Z}^2$-periodic isoradial graphs. They proved that the dimer characteristic polynomial equals a constant $c$ times the critical Laplacian characteristic polynomial — hence is a Harnack curve of genus zero whose only zero on the unit torus $\mathbb{T}^2=\{|z|=|w|=1\}$ is a double zero at $(1,1)$ — computed the free energy, and gave an explicit local formula for the Gibbs measure obtained as the weak limit of the Boltzmann measures. Because Fisher graphs are non-bipartite, Sheffield's theorem was unavailable: weak convergence was proved by a direct, model-specific argument, and the authors noted their techniques could be used for other non-bipartite dimer graphs, leaving a general structural theorem open. The constant $c$ was obtained only through a unit-modulus phase factor $\zeta$ appearing in the proof.

The subsequent massive program closed the isoradial Z-invariant questions. The companion locality paper (2011) removed periodicity at $k=0$ and computed the constant explicitly. The Z-invariant massive Laplacian (2017) and the Z-Dirac operator analysis (2021) built the off-critical operator theory, including massive Green functions and random-walk representations. The 2019 paper extended the full dimer analysis to every $k$, periodic or not: an explicit local formula for the inverse Kasteleyn operator, the Gibbs measure (obtained as a weak limit of Boltzmann measures in the periodic case, again by direct arguments since for $k\neq0$ the spectral curve avoids $\mathbb{T}^2$ and $k=0$ was already handled), the local free energy, the shared second-order phase transition with spanning forests, and the explicit constant

$$P^{K}(z,w)=2^{|V_1|+|E_1|}\prod_{e\in E_1}\frac{\mathrm{cn}(\theta_e|k)}{1+\mathrm{cn}(\theta_e|k)}\,P^{\Delta^{m}}(z,w)$$

between the Fisher dimer and massive-Laplacian characteristic polynomials, where $\mathrm{cn}$ denotes the Jacobi elliptic cosine, $\theta_e$ ranges over the rhombus half-angles of the edges of the fundamental domain, and $P^{\Delta^{m}}$ is the characteristic polynomial of the massive Laplacian.

What remains open is precisely the structural generality. For non-bipartite periodic dimers no analogue of the Kenyon–Okounkov–Sheffield theory exists. Known islands: the spectral curve of Fisher graphs of ferromagnetic Ising models on $\mathbb{Z}^2$ is Harnack (Cimasoni–Duminil-Copin; Li), while frustrated Ising models on isoradial graphs were recently shown to produce genuinely non-Harnack genus-one curves (de Tilière–Rey); Newton polygons have been determined only for two special families of non-bipartite torus graphs (Bošković). On the Ising side, Chelkak's s-embeddings — graph embeddings built from a solution of the Kadanoff–Ceva fermion propagation equation, generalizing isoradial embeddings — extend the massive Ising analysis beyond isoradial graphs, but by fermionic methods rather than the dimer route; on the bipartite dimer side, isoradial immersions and elliptic dimers on minimal graphs generalize isoradial embeddings to genus-one elliptic weights. The non-bipartite Fisher-graph case beyond Z-invariant isoradial weights has no general theory.

Problem Statement

Prove a general structural theory — or refute its possibility by an explicit counterexample — for the dimer model with positive edge weights on $\mathbb{Z}^2$-periodic non-bipartite planar graphs whose fundamental domain $G_1$ admits perfect matchings (in particular, has an even number of vertices, so that the Boltzmann measures on the toroidal exhaustions $G_n$ are defined for the $n$ considered), the setting that contains the Fisher graphs of ferromagnetic (positive-coupling) periodic planar Ising models beyond the critical and massive Z-invariant isoradial class (which is now fully understood). Frustrated (sign-changing-coupling) Ising models have non-positive dimer weights $\nu_e=\coth(J_e)$, so their Fisher graphs lie outside this positive-weight class and are invoked below only as motivation, not as instances the theory must characterize. As in the bipartite theory of Kenyon–Okounkov–Sheffield together with Sheffield's convergence theorem, the theory should deliver, from the structure of the characteristic polynomial $P(z,w)=\det K(z,w)$ alone:

1. Convergence: a general theorem that determines, from the zeros of $P$ on the unit torus $\mathbb{T}^2$, the conditions under which the Boltzmann measures on the natural toroidal exhaustions converge weakly — replacing the direct model-specific arguments that are still needed for Fisher graphs — and describes the complementary failure regime.

2. Spectral curves: a characterization of the real algebraic curves that arise as spectral curves of non-bipartite periodic dimer models — which curves occur, under what conditions on the graph and weights they are Harnack, and what zero structure on $\mathbb{T}^2$ is possible, given that frustrated Ising models are known to produce non-Harnack curves while ferromagnetic Fisher graphs on $\mathbb{Z}^2$ are Harnack.

These two components jointly support one overarching objective: a structural theory for the whole class playing the role that the bipartite theory plays for bipartite dimers. A further analysis of one special family or one special lattice (as in the existing Newton-polygon results for triangle-replacement and corner graphs) is genuine progress but is not by itself a solution.

Both components are jointly mandatory: a positive solution must deliver the convergence theorem and the spectral-curve characterization for the class as a whole, and establishing only one component, or only a single sub-claim — the zero-condition on $P$ governing weak convergence of the Boltzmann measures on the toroidal exhaustions; which real algebraic curves occur as spectral curves; when these curves are Harnack; what zero structure on $\mathbb{T}^2$ is possible — does not resolve the task. Because the general theory asserts all of these sub-claims together, the refutation route may instead falsify any one of them by an explicit counterexample for positive edge weights on a graph in the class — for instance a graph whose convergence behaviour contradicts a zero-condition that the theory asserts to govern weak convergence, or a spectral curve whose zero structure on $\mathbb{T}^2$ contradicts a prediction of the characterization. The isolated in-class partial results for the special triangle-replacement and corner-graph families — and likewise the non-Harnack curves that frustrated Ising models produce outside the class — do not by themselves constitute such a refutation.

References

1. Cédric Boutillier and Béatrice de Tilière, The critical Z-invariant Ising model via dimers: the periodic case, Probability Theory and Related Fields 147 (2010), no. 3–4, 379–413. DOI 10.1007/s00440-009-0210-1, https://doi.org/10.1007/s00440-009-0210-1 (arXiv:0812.3848)
2. Cédric Boutillier and Béatrice de Tilière, The critical Z-invariant Ising model via dimers: locality property, Communications in Mathematical Physics 301 (2011), no. 2, 473–516. DOI 10.1007/s00220-010-1151-3, https://doi.org/10.1007/s00220-010-1151-3 (arXiv:0902.1882)
3. Cédric Boutillier, Béatrice de Tilière and Kilian Raschel, The Z-invariant massive Laplacian on isoradial graphs, Inventiones mathematicae 208 (2017), no. 1, 109–189. DOI 10.1007/s00222-016-0687-z, https://doi.org/10.1007/s00222-016-0687-z (arXiv:1504.00792)
4. Cédric Boutillier, Béatrice de Tilière and Kilian Raschel, The Z-invariant Ising model via dimers, Probability Theory and Related Fields 174 (2019), no. 1–2, 235–305. DOI 10.1007/s00440-018-0861-x, https://doi.org/10.1007/s00440-018-0861-x (arXiv:1612.09082)
5. Béatrice de Tilière, The Z-Dirac and massive Laplacian operators in the Z-invariant Ising model, Electronic Journal of Probability 26 (2021), paper no. 53, 86 pp. DOI 10.1214/21-EJP601, https://doi.org/10.1214/21-EJP601 (arXiv:1801.00207)
6. Vladimir Bošković, Newton polygons for the non-bipartite dimer model, arXiv:2607.07503 (2026), https://arxiv.org/abs/2607.07503
7. Béatrice de Tilière and Lucas Rey, Classification of (non)-frustrated 2D Ising models in genus 1 on isoradial graphs, arXiv:2602.13526 (2026), https://arxiv.org/abs/2602.13526
8. Richard Kenyon, Andrei Okounkov and Scott Sheffield, Dimers and amoebae, Annals of Mathematics (2) 163 (2006), no. 3, 1019–1056. DOI 10.4007/annals.2006.163.1019, https://doi.org/10.4007/annals.2006.163.1019
9. Richard Kenyon and Andrei Okounkov, Planar dimers and Harnack curves, Duke Mathematical Journal 131 (2006), no. 3, 499–524. DOI 10.1215/S0012-7094-06-13134-4, https://doi.org/10.1215/S0012-7094-06-13134-4
10. Scott Sheffield, Random surfaces, Astérisque 304 (2005), vi+175 pp. DOI 10.24033/ast.697, https://doi.org/10.24033/ast.697, ISBN 978-2-85629-187-0
11. Henry Cohn, Richard Kenyon and James Propp, A variational principle for domino tilings, Journal of the American Mathematical Society 14 (2001), no. 2, 297–346. DOI 10.1090/S0894-0347-00-00355-6, https://doi.org/10.1090/S0894-0347-00-00355-6
12. Michael E. Fisher, On the dimer solution of planar Ising models, Journal of Mathematical Physics 7 (1966), no. 10, 1776–1781. DOI 10.1063/1.1704825, https://doi.org/10.1063/1.1704825
13. Rodney J. Baxter, Solvable eight-vertex model on an arbitrary planar lattice, Philosophical Transactions of the Royal Society of London. Series A 289 (1978), no. 1359, 315–346. DOI 10.1098/rsta.1978.0062, https://doi.org/10.1098/rsta.1978.0062
14. David Cimasoni and Hugo Duminil-Copin, The critical temperature for the Ising model on planar doubly periodic graphs, Electronic Journal of Probability 18 (2013), paper no. 44, 18 pp. DOI 10.1214/EJP.V18-2352, https://doi.org/10.1214/EJP.V18-2352
15. Zhongyang Li, Spectral curves of periodic Fisher graphs, Journal of Mathematical Physics 55 (2014), no. 12, article 123301. DOI 10.1063/1.4904454, https://doi.org/10.1063/1.4904454
16. Dmitry Chelkak, Ising model and s-embeddings of planar graphs, Annales Scientifiques de l'École Normale Supérieure (4) 57 (2024), no. 5, 1271–1346. DOI 10.24033/asens.2591, https://doi.org/10.24033/asens.2591 (arXiv:2006.14559)
17. Cédric Boutillier, David Cimasoni and Béatrice de Tilière, Isoradial immersions, Journal of Graph Theory 99 (2022), no. 4, 715–757. DOI 10.1002/jgt.22761, https://doi.org/10.1002/jgt.22761
18. Cédric Boutillier, David Cimasoni and Béatrice de Tilière, Elliptic dimers on minimal graphs and genus 1 Harnack curves, Communications in Mathematical Physics 400 (2023), no. 2, 1071–1136. DOI 10.1007/s00220-022-04612-6, https://doi.org/10.1007/s00220-022-04612-6
19. Richard Kenyon, The Laplacian and Dirac operators on critical planar graphs, Inventiones mathematicae 150 (2002), no. 2, 409–439. DOI 10.1007/s00222-002-0249-4, https://doi.org/10.1007/s00222-002-0249-4