Extending the Pfaffian Sign Theorem for the dimer model to other lattices on the torus and to higher-genus and non-orientable surfaces

The partition function of the dimer model on a graph embedded in a closed surface is a fixed linear combination of Pfaffians of Kasteleyn matrices twisted by homology classes: four Pfaffians on the torus, $2^{2g}$ on an orientable surface of genus $g$, and $2^{2-\chi}$ on a non-orientable surface of Euler characteristic $\chi$. The coefficients of these combinations are fixed by combinatorial topology, but the sign of each Pfaffian is a separate analytic question — each Pfaffian is a polynomial in the positive edge weights with terms of both signs — and these signs are exactly what is needed to convert the decomposition into a tool for rigorous asymptotic analysis. Bleher, Elwood and Petrović (2018) proved the Pfaffian Sign Theorem for the triangular lattice on the torus (the periodic–periodic Pfaffian is negative, the other three positive, for all positive weights) and posed the problem of extending the theorem to other lattices on the torus and to graphs on higher-genus and non-orientable surfaces. Both directions remain open, with delimited settled islands — the square lattice and critical lattices on the torus, and the Klein bottle, where Cimasoni (2023) showed that the partition function of any weighted graph there is a sum of absolute values of two Pfaffian parts. What remains open is a sign theorem for arbitrary doubly-periodic lattices on the torus in noncritical weight regimes, and for the multi-Pfaffian decomposition on orientable surfaces of genus $g\ge 2$ and on general non-orientable surfaces.

Background

The dimer model is a classical statistical-mechanical system of combinatorial origin. On a finite graph $\Gamma=(V,E)$, a dimer configuration (also called a perfect matching) is a set of edges covering every vertex exactly once; a configuration $\sigma$ receives the weight $w(\sigma)=\prod_{e\in\sigma}z_e$, where $z_e>0$ is the weight of edge $e$, and the partition function is $Z=\sum_\sigma w(\sigma)$. Exact computation of $Z$ rests on the Kasteleyn–Temperley–Fisher method: orient the edges so that every face has an odd number of clockwise-oriented boundary edges (a Kasteleyn orientation) and form the signed, weighted, skew-symmetric adjacency matrix $K$ (the Kasteleyn matrix). On a planar graph, $Z$ equals the Pfaffian of $K$, where for a $2N\times 2N$ skew-symmetric matrix $A$ the Pfaffian is the signed sum of products $\mathrm{Pf}\,A=\sum_\sigma \mathrm{sgn}(\sigma)\prod_i a_{i\sigma(i)}$ over perfect matchings of the index set, an object satisfying $(\mathrm{Pf}\,A)^2=\det A$.

On the torus no single Kasteleyn orientation suffices. Kasteleyn (1961) showed that for a graph embedded in the torus the partition function is a fixed linear combination of four Pfaffians,

$$Z=\frac{1}{2}\left(-\mathrm{Pf}\,A_1+\mathrm{Pf}\,A_2+\mathrm{Pf}\,A_3+\mathrm{Pf}\,A_4\right),$$

of Kasteleyn matrices twisted by periodic or antiperiodic boundary conditions along each of the two fundamental cycles ($A_1$ periodic–periodic, $A_2$ periodic–antiperiodic, $A_3$ antiperiodic–periodic, $A_4$ antiperiodic–antiperiodic). The coefficients are fixed a priori, but the sign of each Pfaffian is not: each $\mathrm{Pf}\,A_i$ is a polynomial in the weights whose terms carry both signs, and cancellations depending on the lattice geometry and on the arithmetic of the period lengths $m,n$ decide the outcome. Since $(\mathrm{Pf}\,A)^2=\det A$ determines only absolute values, these signs are precisely the extra information needed to make the formula usable for rigorous asymptotics — for instance to conclude that $Z=\tfrac12\sum_{i=1}^4|\mathrm{Pf}\,A_i|$, the sum of the absolute values of the four Pfaffians with the prefactor $\tfrac12$ of the displayed formula — which underlies full asymptotic expansions of $\log Z$ as $m,n\to\infty$, including the constant-order finite-size correction term that conformal field theory predicts to be universal. Through the Fisher correspondence (an exact mapping of the Ising model on a graph to a dimer model on a decorated graph), the same machinery governs Ising partition functions.

Before the Bleher–Elwood–Petrović paper, the signs were known only in special cases. For the square lattice, Kasteleyn computed $\mathrm{Pf}\,A_1=0$ and assumed the remaining signs. Kenyon, Sun and Wilson (2016) determined the signs for critical dimer models on arbitrary lattices on the torus — critical meaning that the spectral curve, i.e. the zero set of the characteristic polynomial $\det K(z,w)$ of the doubly twisted Kasteleyn matrix with twist phases $z,w$, meets the unit torus $|z|=|w|=1$; in the noncritical (massive) case, where the curve is disjoint from the unit torus, they obtained only the leading term $\log Z=mn f_0+o(1)$. For the noncritical triangular lattice, sign patterns had been conjectured without proof in the physics literature (Fendley–Moessner–Sondhi 2002; Izmailian–Kenna 2011). Bleher, Elwood and Petrović (2018) proved the Pfaffian Sign Theorem for the $m\times n$ triangular lattice on the torus ($m$ even) with arbitrary positive weights $z_h,z_v,z_d$: $\mathrm{Pf}\,A_1<0$ and $\mathrm{Pf}\,A_2,\mathrm{Pf}\,A_3,\mathrm{Pf}\,A_4>0$ — disproving the folklore claim, stated without proof in earlier work, that all four Pfaffians are positive in the noncritical regime — and deduced from it the asymptotic behavior of $Z$ with exponentially small error.

Beyond the torus, the Pfaffian method generalizes at the price of many Pfaffians. For a graph embedded in a closed orientable surface $\Sigma_g$ of genus $g$ (a sphere with $g$ handles), Galluccio–Loebl (1999) and Tesler (2000) expressed $Z$ as a fixed linear combination of $2^{2g}$ Pfaffians of Kasteleyn matrices twisted by the elements of $H_1(\Sigma_g;\mathbb{Z}/2\mathbb{Z})$, the first homology group of the surface with coefficients in the two-element field, whose elements record the parity of the winding around each of the $2g$ fundamental cycles; Cimasoni–Reshetikhin (2007) gave the coefficients a geometric meaning through discrete spin structures (in this discrete setting, a spin structure is a quadratic form on $H_1(\Sigma;\mathbb{Z}/2\mathbb{Z})$ refining the intersection pairing, equivalently the monodromy data of the twist phases; a closed orientable genus-$g$ surface carries $2^{2g}$ of them). For closed non-orientable surfaces (surfaces such as the Klein bottle or the projective plane, which contain a Möbius band and are classified by the Euler characteristic $\chi$), Tesler (2000) and Cimasoni (2009) obtained the analogous formulas with $2^{2-\chi(\Sigma)}$ Pfaffians, the coefficients being organized by pin$^-$ structures, the non-orientable analogues of spin structures. Determining the signs of all these Pfaffians for positive weights — the general Pfaffian Sign Theorem in the sense of Bleher–Elwood–Petrović — is the open problem. One instance is already settled: for the Klein bottle ($\chi=0$, the non-orientable companion of the torus, where the four Pfaffians form two complex-conjugate pairs), Cimasoni showed that for any weighted graph embedded there, $Z=|\mathrm{Im}\,\mathrm{Pf}\,A(1,1)|+|\mathrm{Re}\,\mathrm{Pf}\,A(1,-1)|$ (2023, Proposition 2.6, strengthening Theorem 6.3 of his 2009 paper, which had only the weaker outer-absolute-value form of the identity).

Problem Statement

Bleher–Elwood–Petrović (2018) settled the sign question for the triangular lattice on the torus and posed, in their concluding section, the problem of extending the Pfaffian Sign Theorem to the general surface setting. Concretely, determine the signs of every Pfaffian entering the Galluccio–Loebl/Tesler/Cimasoni–Reshetikhin decomposition of the dimer partition function of a weighted graph with positive edge weights, in the two directions Bleher–Elwood–Petrović name: (i) doubly-periodic lattices on the torus other than the square and triangular lattices, in the noncritical weight regime as well as any residual critical cases not covered by the settled results, for all positive weights and all admissible parities of the period lengths; and (ii) graphs embedded in closed orientable surfaces of genus $g\ge 2$ (the $2^{2g}$-Pfaffian decomposition) and in non-orientable surfaces other than the Klein bottle. The common objective uniting both directions is the same as in the triangular-lattice theorem: reduce $Z$ to an explicitly signed — ideally absolute-value — sum of Pfaffians whose signs are completely determined, so that the decomposition supports rigorous asymptotic analysis (full asymptotic expansions, finite-size corrections) rather than only numerical evaluation. Instances already settled, which an answer must go beyond: the square and triangular lattices on the torus, critical weight systems on arbitrary toric lattices (Kenyon–Sun–Wilson), and the Klein bottle (Cimasoni). Results for single further lattices or single further surfaces constitute partial progress toward the objective. A complete solution requires the general sign theorem in both directions (i) and (ii); a full resolution of either direction alone is a major partial result but does not on its own resolve the task.

References

1. Pavel Bleher, Brad Elwood, Dražen Petrović, 'The Pfaffian Sign Theorem for the Dimer Model on a Triangular Lattice', Journal of Statistical Physics 171(3), 400–426 (2018). DOI: 10.1007/s10955-018-2007-z, https://doi.org/10.1007/s10955-018-2007-z (arXiv:1711.00032, https://arxiv.org/abs/1711.00032)
2. Pavel Bleher, Brad Elwood, Dražen Petrović, 'Dimer model: Full asymptotic expansion of the partition function', Journal of Mathematical Physics 59(9), 091407 (2018). DOI: 10.1063/1.5035271, https://doi.org/10.1063/1.5035271 (arXiv:1806.03742, https://arxiv.org/abs/1806.03742)
3. P. W. Kasteleyn, 'The statistics of dimers on a lattice. I. The number of dimer arrangements on a quadratic lattice', Physica 27(12), 1209–1225 (1961). DOI: 10.1016/0031-8914(61)90063-5, https://doi.org/10.1016/0031-8914(61)90063-5
4. P. Fendley, R. Moessner, S. L. Sondhi, 'Classical dimers on the triangular lattice', Physical Review B 66, 214513 (2002). DOI: 10.1103/PhysRevB.66.214513, https://doi.org/10.1103/PhysRevB.66.214513
5. N. Sh. Izmailian, Ralph Kenna, 'Dimer model on a triangular lattice', Physical Review E 84, 021107 (2011). DOI: 10.1103/PhysRevE.84.021107, https://doi.org/10.1103/PhysRevE.84.021107
6. Richard W. Kenyon, Nike Sun, David B. Wilson, 'On the asymptotics of dimers on tori', Probability Theory and Related Fields 166(3–4), 971–1023 (2016). DOI: 10.1007/s00440-015-0687-8, https://doi.org/10.1007/s00440-015-0687-8 (arXiv:1310.2603, https://arxiv.org/abs/1310.2603)
7. Anna Galluccio, Martin Loebl, 'On the theory of Pfaffian orientations. I. Perfect matchings and permanents', Electronic Journal of Combinatorics 6, Article R6 (1999). DOI: 10.37236/1438, https://doi.org/10.37236/1438
8. Glenn Tesler, 'Matchings in graphs on non-orientable surfaces', Journal of Combinatorial Theory, Series B 78(2), 198–231 (2000). DOI: 10.1006/jctb.1999.1941, https://doi.org/10.1006/jctb.1999.1941
9. David Cimasoni, Nicolai Reshetikhin, 'Dimers on surface graphs and spin structures I', Communications in Mathematical Physics 275(1), 187–208 (2007). DOI: 10.1007/s00220-007-0302-7, https://doi.org/10.1007/s00220-007-0302-7 (arXiv:math-ph/0608070, https://arxiv.org/abs/math-ph/0608070)
10. David Cimasoni, 'Dimers on graphs in non-orientable surfaces', Letters in Mathematical Physics 87(1–2), 149–179 (2009). DOI: 10.1007/s11005-009-0299-2, https://doi.org/10.1007/s11005-009-0299-2 (arXiv:0804.4772, https://arxiv.org/abs/0804.4772)
11. David Cimasoni, Anh Minh Pham, 'Identities between dimer partition functions on different surfaces', Journal of Statistical Mechanics: Theory and Experiment 2016, 103101 (2016). DOI: 10.1088/1742-5468/2016-10-103101, https://doi.org/10.1088/1742-5468/2016-10-103101 (arXiv:1608.00741, https://arxiv.org/abs/1608.00741)
12. David Cimasoni, 'The dimer and Ising models on Klein bottles', Annales de l'Institut Henri Poincaré D, Combinatorics, Physics and their Interactions 11(3), 503–569 (2023). DOI: 10.4171/aihpd/166, https://doi.org/10.4171/aihpd/166 (arXiv:2010.11047, https://arxiv.org/abs/2010.11047)
13. Chi-Ning Chen, Chin-Kun Hu, N. Sh. Izmailian, Ming-Chya Wu, 'Specific heat and partition function zeros for the dimer model on the checkerboard B lattice: Finite-size effects', Physical Review E 99, 012102 (2019). DOI: 10.1103/PhysRevE.99.012102, https://doi.org/10.1103/PhysRevE.99.012102
14. Ruben Costa-Santos, Barry M. McCoy, 'Dimers and the critical Ising model on lattices of genus > 1', Nuclear Physics B 623(3), 439–473 (2002). DOI: 10.1016/S0550-3213(01)00611-3, https://doi.org/10.1016/S0550-3213(01)00611-3 (arXiv:hep-th/0109167, https://arxiv.org/abs/hep-th/0109167)
15. Ruben Costa-Santos, Barry M. McCoy, 'Finite size corrections for the Ising model on higher genus triangular lattices', Journal of Statistical Physics 112(5–6), 889–920 (2003). DOI: 10.1023/A:1024697307618, https://doi.org/10.1023/A:1024697307618
16. Nathanaël Berestycki, Benoit Laslier, Gourab Ray, 'Dimers on Riemann surfaces I: Temperleyan forests', arXiv:1908.00832 (2019), https://arxiv.org/abs/1908.00832
17. Nathanaël Berestycki, Benoit Laslier, Gourab Ray, 'Dimers on Riemann surfaces II: conformal invariance and scaling limit', arXiv:2207.09875 (2022), https://arxiv.org/abs/2207.09875