Completing the eigenvalue-amplitude theory of the Potts model on a torus: non-generic lattices, width-dependent level degeneracies, and the Beraha-number zero structure

For the $Q$-state Potts model on an $L\times N$ torus, Richard and Jacobsen proved that the partition function decomposes over traces of transfer-matrix blocks at level $l$ projected onto irreducible representations of the cyclic group $C_l$, with eigenvalue amplitudes given by the Read–Saleur formula — but only for 'generic' lattices (e.g. triangular) on which every admissible permutation of the winding clusters is realizable. Four linked gaps were left open and remain open: (i) an analytic characterization of the toroidal partition-function zero locus, in particular the role of the Beraha numbers $B_p=4\cos^2(\pi/p)$ that govern the cyclic geometry; (ii) an algebraic interpretation of the degeneracies between eigenvalues of different levels, which depend on the width $L$; (iii) the extension (or precise modification) of the amplitude formula to non-generic lattices such as the square and honeycomb lattices, where some restricted partition functions $Z_{j,n_1,P}$ vanish; and (iv) a deeper periodic Temperley–Lieb / quantum-group description systematizing these structures. Later work settled pieces numerically on the chromatic line ($v=-1$, widths $L\le7$) and structurally at the critical point (affine Temperley–Lieb plus algebraic geometry: closed forms to $L=6$, exact Potts torus partition functions to $L=10$, and loop-model zero-condensation curves to $L=14$), but none of the four gaps is closed at the level of proof for general temperature.

Background

The $Q$-state Potts model assigns to each vertex of a graph a spin taking $Q$ values and weights each edge by a Boltzmann factor depending on whether its endpoint spins agree. In the Fortuin–Kasteleyn (FK) random-cluster representation the partition function becomes a sum over subgraphs (clusters), each weighted by $Q^{\#\text{clusters}}v^{\#\text{edges}}$, where $v=e^J-1$ is the temperature variable and $J$ the coupling on an edge. On a finite $L\times N$ strip with periodic boundary conditions in both directions (a torus) some clusters are non-contractible: they wind around a periodic direction. Because of this non-locality, $Z(L,N)$ is not simply the trace of a power of the row-to-row transfer matrix $T_L$ acting on connectivity states (states recording which sites of a row are connected through the part of the lattice already built).

Richard and Jacobsen resolved this (following empirical work of Chang and Shrock and a continuum-limit formula of Read and Saleur) as follows. Configurations are classified by $j$, the number of non-trivial-homotopy clusters winding $n_1$ times in the transfer direction and realizing a permutation $P$ of the winding branches; the admissible permutations at a given level form the cyclic group $C_l$. Projecting onto irreducible representations $D_k$ of $C_l$ produces generalized characters $K_{l,D_k}$ which are traces of powers of the transfer-matrix block at level $l$ (the level counts the bridges, i.e. the clusters connecting a row to itself across the periodic transfer direction). For any finite torus of a generic lattice — one on which every admissible cyclic permutation is realizable, the triangular lattice being the standard example — and for arbitrary inhomogeneous couplings, one obtains

$Z(L,N)=\sum_{l,D_k} b^{l,D_k}\,K_{l,D_k},$

where the amplitudes of the transfer-matrix eigenvalues are exactly the Read–Saleur values

$b^{(l,m)}=\Lambda(l,m;e_0)+(Q-1)\,\Lambda(l,m;\tfrac{1}{2}),\qquad \Lambda(l,m;e)=2\sum_{d\mid l}\frac{\mu\big(\tfrac{m}{m\wedge d}\big)\varphi\big(\tfrac{l}{d}\big)}{l\,\varphi\big(\tfrac{m}{m\wedge d}\big)}\cos(2\pi d e),$

with $m$ running over the divisors of $l$ (the number of distinct amplitudes at level $l$ equals the number of divisors of $l$), $\mu$ and $\varphi$ the Möbius and Euler-totient functions, $m\wedge d$ the greatest common divisor, and $\sqrt{Q}=2\cos(\pi e_0)$.

Two features distinguish the torus from the cyclic geometry (only one periodic direction). First, in the cyclic case all eigenvalues at a given level share a single amplitude $\alpha^{(l)}=U_{2l}(\sqrt{Q}/2)$ (a Chebyshev polynomial of the second kind), while on the torus the amplitude splits over the divisors of $l$. Second, the zero structure of $Z$ is controlled by these amplitudes through the Beraha–Kahane–Weiss (BKW) theorem: for $Z=\sum_i a_i\lambda_i^N$, zeros accumulate as $N\to\infty$ at parameter values where the amplitude of a dominant (largest-modulus) eigenvalue vanishes or where two dominant eigenvalues are equimodular. In the cyclic geometry this singles out exactly the Beraha numbers $B_p=4\cos^2(\pi/p)$, independently of $L$; determining the amplitudes matters physically because in the antiferromagnetic regime the Potts model has a massless Berker–Kadanoff phase — the critical antiferromagnetic phase, obtained by analytic continuation of the ferromagnetic transition, in which the temperature is irrelevant in the renormalization-group sense — that is destroyed precisely at BKW-relevant values of $Q$, producing $Q$-driven phase transitions. On the torus, an eigenvalue can belong to several levels simultaneously, and the resulting degeneracies between different levels depend on the width $L$ (they were observed first by Chang and Shrock); within the present framework they have, in the authors' words, 'no simple algebraic interpretation'. The source notes that the toroidal transfer matrix no longer commutes with the generators of the quantum group $U_q(sl_2)$ (a one-parameter deformation of the Lie algebra $sl_2$ of $2\times2$ traceless matrices) — the commutant mechanism that makes the cyclic amplitudes simple — and that only partial algebraic progress exists through representations of the periodic (affine) Temperley–Lieb algebra, the diagram algebra of local reconnection moves underlying the transfer matrix in the loop representation (the rewriting of Fortuin–Kasteleyn cluster configurations as non-crossing loops on the medial lattice).

Finally, the combinatorial proof requires every restricted partition function $Z_{j,n_1,P}$ to be non-zero. This fails on non-generic lattices such as the square and honeycomb lattices, where for $n_1 j=L$ with $n_1>1$ there is not enough 'space' for all winding cluster branches to percolate while realizing a non-trivial permutation, so some $Z_{j,n_1,P}$ vanish identically. Such lattices, the authors state, introduce additional difficulties that must be treated on a case-to-case basis; the empirical evidence that the same amplitudes nevertheless appear there has not been turned into a theorem.

Problem Statement

Complete the exact finite-size amplitude theory of the $Q$-state Potts model on $L\times N$ tori, in the framework established by Read–Saleur and Richard–Jacobsen, by resolving the four linked gaps (i)–(iv) the source paper left open, addressed jointly through the three parts below: the deeper periodic Temperley–Lieb / quantum-group description sought in gap (iv) is not a separate deliverable but the algebraic framework within which part 3 requests an explanation of the width-dependent degeneracies, and which — as the closing paragraph states — is the expected systematization of parts 1 and 2. A full answer should, with proofs at the level of rigor of the original combinatorial derivation and without restricting the temperature variable $v$ (inhomogeneous couplings included) or the width $L$:

1. Non-generic lattices. Either prove that the eigenvalue-amplitude decomposition with the amplitudes $b^{(l,m)}$ (equivalently $b^{l,D_k}$) holds verbatim on non-generic lattices — at minimum the square and honeycomb lattices, where some $Z_{j,n_1,P}$ with $n_1 j=L$ and $n_1>1$ vanish — or determine precisely how the amplitudes are modified there, explaining the fate of the cyclic-group ($C_l$) projection argument when not every admissible permutation of the winding clusters is realizable. The existing empirical agreement (Chang–Shrock at small widths; Jacobsen–Salas up to $L=6$ at $v=-1$, partially at $L=7$; Böhm et al. at the critical point up to $L\sim10$) must be proved or refuted.

2. Zero locus and the Beraha numbers. Characterize analytically the accumulation locus of the toroidal partition-function zeros in the $N\to\infty$ limit via the Beraha–Kahane–Weiss mechanism — the set of $Q$ (as a function of $v$, and of the lattice type) at which a dominant amplitude vanishes or two dominant eigenvalues are equimodular. In particular, determine the role of the Beraha numbers $B_p=4\cos^2(\pi/p)$ in toroidal geometry: in cyclic geometry the BKW-relevant values are exactly the $B_p$, independent of $L$, while on the torus the numerics at $v=-1$ show only integer $Q$ playing that role for the dominant eigenvalues, with the $B_p$ expected only as level-crossing loci of subdominant eigenvalues. A solution states and proves the general statement, including its dependence on $L$.

3. Algebraic interpretation of the width-dependent degeneracies. Explain, within an algebraic structure — expected to be the representation theory of the periodic (affine) Temperley–Lieb algebra or a related quantum-group structure — why eigenvalues of the toroidal transfer matrix occur simultaneously at several levels $l$ and why the degeneracies between different levels depend arithmetically on the width $L$, thereby systematizing what is currently a case-by-case observation.

These three parts form one objective: the amplitudes and their degeneracies are exactly the input to the BKW characterization of the zeros, and the algebraic structure of part 3 is the expected systematization of parts 1 and 2. Numerical phase diagrams on the chromatic line $v=-1$ alone (Jacobsen–Salas, $L\le7$) or results restricted to the critical line via loop-model integrability (Böhm et al.) are progress but not a solution: the question concerns arbitrary $v$ and requires proof.

References

1. J.-F. Richard and J. L. Jacobsen, 'Eigenvalue amplitudes of the Potts model on a torus', Nucl. Phys. B 769 (2007) 256–274. DOI: 10.1016/j.nuclphysb.2007.01.028; arXiv:math-ph/0608055. https://arxiv.org/abs/math-ph/0608055
2. N. Read and H. Saleur, 'Exact spectra of conformal supersymmetric nonlinear sigma models in two dimensions', Nucl. Phys. B 613 (2001) 409–444. DOI: 10.1016/S0550-3213(01)00395-9; arXiv:hep-th/0106124. https://arxiv.org/abs/hep-th/0106124
3. S.-C. Chang and R. Shrock, 'Transfer matrices for the partition function of the Potts model on lattice strips with toroidal and Klein-bottle boundary conditions', Physica A 364 (2006) 231–262. DOI: 10.1016/j.physa.2005.08.076; arXiv:cond-mat/0506274. https://arxiv.org/abs/cond-mat/0506274
4. A. Nichols, 'The Temperley-Lieb algebra and its generalizations in the Potts and XXZ models', J. Stat. Mech. (2006) P01003. DOI: 10.1088/1742-5468/2006/01/P01003; arXiv:hep-th/0509069. https://arxiv.org/abs/hep-th/0509069
5. J. L. Jacobsen and J. Salas, 'Phase diagram of the chromatic polynomial on a torus', Nucl. Phys. B 783 (2007) 238–296. DOI: 10.1016/j.nuclphysb.2007.04.023; arXiv:cond-mat/0703228. https://arxiv.org/abs/cond-mat/0703228
6. J. L. Jacobsen and J. Salas, 'A generalized Beraha conjecture for non-planar graphs', Nucl. Phys. B 875 (2013) 678–718. DOI: 10.1016/j.nuclphysb.2013.07.012; arXiv:1303.5210. https://arxiv.org/abs/1303.5210
7. J. L. Jacobsen, 'Critical points of Potts and O(N) models from eigenvalue identities in periodic Temperley-Lieb algebras', J. Phys. A 48 (2015) 454003. DOI: 10.1088/1751-8113/48/45/454003; arXiv:1507.03027. https://arxiv.org/abs/1507.03027
8. A. Morin-Duchesne and Y. Saint-Aubin, 'The Jordan Structure of Two Dimensional Loop Models', J. Stat. Mech. (2011) P04007. DOI: 10.1088/1742-5468/2011/04/P04007; arXiv:1101.2885. https://arxiv.org/abs/1101.2885
9. L. Grans-Samuelsson, L. Liu, Y. He, J. L. Jacobsen and H. Saleur, 'The action of the Virasoro algebra in the two-dimensional Potts and loop models at generic Q', JHEP 10 (2020) 109. DOI: 10.1007/JHEP10(2020)109; arXiv:2007.11539. https://arxiv.org/abs/2007.11539
10. J. Böhm, J. L. Jacobsen, Y. Jiang and Y. Zhang, 'Geometric Algebra and Algebraic Geometry of Loop and Potts Models', JHEP 05 (2022) 068. DOI: 10.1007/JHEP05(2022)068; arXiv:2202.02986. https://arxiv.org/abs/2202.02986
11. A. Morin-Duchesne, A. Klümper and P. A. Pearce, 'Modular covariant torus partition functions of dense $A_1^{(1)}$ and dilute $A_2^{(2)}$ loop models', J. Stat. Mech. (2025) 073103. DOI: 10.1088/1742-5468/ade133; arXiv:2501.19288. https://arxiv.org/abs/2501.19288