D-finiteness and exact solvability of the three-dimensional Ising partition function: the Guttmann–Enting non-solvability conjecture and the joint satisfiability of the six necessary conditions

The exact solution of the nearest-neighbor ferromagnetic Ising model on the simple cubic lattice — the most studied open problem of equilibrium statistical mechanics — is not merely unsolved: even the functional class in which an exact partition function could be expressed is undetermined. Hansel, Maillard, Oitmaa and Velgakis (1987) found that partially resummed anisotropic high-temperature series of the cubic model behave anomalously, and Guttmann and Enting (1996) conjectured from this that the model is 'not solvable' in the sense that its thermodynamic-limit partition function per site is not D-finite (holonomic). Viswanathan, Portillo, Raposo and da Luz (2022) distilled six necessary conditions for any claimed exact solution and posed the attendant open questions. This problem record asks for a rigorous settlement of that status: (A) prove or disprove that the per-site partition function is D-finite/holonomic as a function of its coupling variables; and (B) prove or refute that the six necessary conditions are jointly satisfiable, either by constructing an exact analytic partition function that provably equals the thermodynamic limit and meets all six conditions, or by proving the conditions inconsistent. No rigorous proof in either direction has been found: the problem remains open.

Background

The Ising model is the paradigmatic lattice model of equilibrium statistical mechanics: spins $\sigma_l\in\{-1,+1\}$ occupy the sites of a lattice and interact through a nearest-neighbor Hamiltonian $H=-\sum_{\langle l,l'\rangle}J\,\sigma_l\sigma_{l'}$ with ferromagnetic couplings $J>0$, at inverse temperature $\beta$. All equilibrium thermodynamics is encoded in the partition function $Z_N=\sum_{\sigma}\exp(-\beta H)$, and the thermodynamic limit is taken per site, $Z=\lim_{N\to\infty}Z_N^{1/N}$, along van Hove sequences of finite regions (nested regions whose boundary-to-volume ratio tends to zero); this limit exists and is independent of boundary conditions. On the cubic lattice one allows anisotropic couplings $J_1,J_2,J_3$ along the three axes, with reduced couplings $K_i=\beta J_i$; the isotropic model has $K_1=K_2=K_3=K$. For any finite region the partition function is a polynomial in the variables $\tanh K_i$, so questions of functional class arise only for the thermodynamic-limit function $Z(K_1,K_2,K_3)$.

In 1944 Onsager solved the two-dimensional square-lattice model in zero field, founding the tradition of exact solutions in statistical mechanics. Every attempt to extend this to the three-dimensional simple-cubic model has failed. The isotropic critical coupling is known only numerically, $K_c\approx 0.221654626$, from series and Monte Carlo analyses, and the conformal bootstrap (a program that determines critical exponents from the consistency conditions of scale and conformal invariance) has pinned down the critical behavior to extraordinary precision without producing any closed form for $Z$. Repeated claims of exact three-dimensional solutions — most prominently by Z.-D. Zhang (2007) and Degang Zhang (2021), both refuted in published comments — have violated basic necessary properties of the true partition function.

This raises a sharper question than 'find the solution': in what functional class could a solution live at all? Most classical exact results in lattice statistical mechanics are D-finite (holonomic) functions — analytic functions of one or several variables satisfying systems of linear differential equations with polynomial coefficients. The class contains the algebraic and hypergeometric functions and the complete elliptic integrals pervading two-dimensional exact solutions, and it is a natural formalization of 'closed form' for special functions. Guttmann and Enting proposed to identify 'solvable' with 'D-finite' and, building on the discovery of Hansel, Maillard, Oitmaa and Velgakis that partially resumming the anisotropic high-temperature series of the cubic model (resumming over two of the three coupling variables) yields functions with anomalous singularity behavior — inconsistent with the rigidity of univariate D-finite functions, which can have only finitely many singularities — conjectured that the three-dimensional Ising partition function is not D-finite, that is, that the model is 'not solvable' in this sense. No rigorous proof exists in either direction.

Viswanathan, Portillo, Raposo and da Luz systematized the constraints on any claimed exact $Z$ into six necessary (not sufficient) conditions: (1) boundary-condition independence of the thermodynamic limit; (2) reduction to Onsager's two-dimensional solution whenever any coupling vanishes; (3) analyticity in the isotropic coupling for $0\le K<K_c$ and $K_c<K<\infty$, together with the exact critical point; (4) term-by-term agreement with the known high-temperature series (an expansion in $v=\tanh K$ counting closed subgraphs of the lattice) and low-temperature series (in $u=\exp(-4K)$, counting domain-wall surfaces), whose coefficients are computed exactly by the finite-lattice method, an algorithm deriving infinite-lattice series coefficients from exact enumeration on finite boxes; (5) permutation symmetry $Z(K_1,K_2,K_3)=Z(K_{\pi(1)},K_{\pi(2)},K_{\pi(3)})$ and convexity of $Z$ in the reduced couplings $(K_1,K_2,K_3)$; and (6) that the claimed solution reproduce, under partial resummation over two of the three couplings, the anomalous singularity structure found by Hansel, Maillard, Oitmaa and Velgakis in the anisotropic high-temperature series, thereby accounting for that apparent non-D-finiteness signal. The same authors emphasize that the numerical evidence for non-D-finiteness is strong but unproven, that an exact solution could still be a nonholonomic closed form, that the six conditions might in principle be jointly unsatisfiable (they conjecture they are not), and that analogs of condition (6) for other lattice systems are essentially unexplored.

What is rigorously known nearby delineates the frontier. In two dimensions, Tracy and Widom proved that the diagonal susceptibility of the square-lattice Ising model below the critical temperature — a sum of Toeplitz determinants (determinants whose entries depend only on the difference of their indices) — has the unit circle as a natural boundary, a barrier of densely accumulating singularities, and is therefore provably not D-finite; this shows that non-D-finiteness proofs are possible where determinant or integral structure is available, but it concerns a two-dimensional susceptibility, not the three-dimensional partition function. Whether the full two-dimensional susceptibility is even differentially algebraic (a strictly larger class than D-finite, allowing nonlinear differential equations with polynomial coefficients) remains undecided. Rechnitzer proved non-D-finiteness of the anisotropic generating function of self-avoiding polygons, generalized to hypercubic lattices, confirming a Guttmann–Enting conjecture in a related setting. In three dimensions, Niedziółka and Wojtkiewicz proved that no translationally invariant bilinear action in anticommuting Grassmann variables — the free-fermion reformulation underlying the two-dimensional solution — can reproduce the partition function, so the classical machinery is provably closed. None of these results decides the status of the three-dimensional partition function itself.

Problem Statement

For the nearest-neighbor ferromagnetic Ising model on the simple cubic lattice $\mathbb{Z}^3$ with reduced couplings $(K_1,K_2,K_3)$, let

$Z(K_1,K_2,K_3)=\lim_{k\to\infty}\left(Z^{BC}_{G_k}(K_1,K_2,K_3)\right)^{1/V(G_k)}$

denote the thermodynamic-limit partition function per site, taken along van Hove sequences $(G_k)$ of finite regions of volume $V(G_k)$; the limit exists and is independent of the boundary conditions $BC$. Every finite-volume partition function is a polynomial in the $\tanh K_i$; the questions below concern only the thermodynamic-limit function. Determine, with rigorous proof:

(A) The functional-class status. Prove or disprove that $Z(K_1,K_2,K_3)$ is D-finite (holonomic), i.e., that it satisfies a system of linear differential equations with polynomial coefficients in its variables. The Guttmann–Enting conjecture — that the anomalous singularity structure of the partially resummed anisotropic high-temperature series reflects genuine non-D-finiteness — asserts that it is not. Because the restriction of a multivariate holonomic function to a non-characteristic locus such as the isotropic line is again holonomic, a rigorous proof of non-D-finiteness for the isotropic restriction $Z(K)$ would settle the negative direction; a positive answer must establish holonomy of the full anisotropic function. Numerical evidence, however strong, settles neither direction.

(B) Joint satisfiability of the six necessary conditions. Prove or refute the conjecture of Viswanathan et al. that the six conditions are not mutually inconsistent; that is, either (i) exhibit an exact analytic expression that provably equals $Z(K_1,K_2,K_3)$ and satisfies all six conditions — boundary-condition independence; reduction to Onsager's solution when any coupling vanishes; analyticity on both sides of the critical point with the exact $K_c$; term-by-term agreement with the known high- and low-temperature series; permutation symmetry and convexity of $Z$ in $(K_1,K_2,K_3)$; and reproduction, under partial resummation over two of the three couplings, of the anomalous singularity structure of the partially resummed anisotropic series (Condition 6) — or (ii) prove that no function can jointly satisfy the six conditions as stated.

Questions (A) and (B) are the two faces of one objective: settling the exact-solvability status of the three-dimensional Ising partition function. (A) fixes the functional class in which an exact $Z$ can live; (B) asks whether an exact analytic $Z$ is possible at all. A constructive answer to (B) reduces (A) to determining the functional class of the exhibited exact expression — which, as noted above, could in principle be a nonholonomic closed form — while a negative answer to (A) confines any answer to (B) to nonholonomic closed forms and demands an explanation of how such a function can satisfy the analyticity and series constraints while manifesting the anisotropic resummation anomaly. A further direction raised by the same source — extending the partial-resummation/D-finiteness diagnostic to other three-dimensional lattices and developing general rigorous criteria for non-D-finiteness of lattice partition functions — concerns other systems and lies explicitly outside the completion boundary of this problem, though any method developed for (A) would bear on it.

References

1. Gandhimohan M. Viswanathan, Marco Aurelio G. Portillo, Ernesto P. Raposo, Marcos G. E. da Luz, 'What Does It Take to Solve the 3D Ising Model? Minimal Necessary Conditions for a Valid Solution', Entropy 24(11), 1665 (2022), DOI 10.3390/e24111665, https://doi.org/10.3390/e24111665 (arXiv:2205.12357, https://arxiv.org/abs/2205.12357)
2. Lars Onsager, 'Crystal Statistics. I. A Two-Dimensional Model with an Order-Disorder Transition', Physical Review 65, 117–149 (1944), DOI 10.1103/PhysRev.65.117, https://doi.org/10.1103/PhysRev.65.117
3. A. J. Guttmann and I. G. Enting, 'Series studies of the Potts model. I: The simple cubic Ising model', Journal of Physics A: Mathematical and General 26, 807–821 (1993), DOI 10.1088/0305-4470/26/4/010, https://doi.org/10.1088/0305-4470/26/4/010
4. D. Hansel, J. M. Maillard, J. Oitmaa, M. J. Velgakis, 'Analytical properties of the anisotropic cubic Ising model', Journal of Statistical Physics 48, 69–80 (1987), DOI 10.1007/BF01010400, https://doi.org/10.1007/BF01010400
5. A. J. Guttmann and I. G. Enting, 'Solvability of Some Statistical Mechanical Systems', Physical Review Letters 76, 344–347 (1996), DOI 10.1103/PhysRevLett.76.344, https://doi.org/10.1103/PhysRevLett.76.344
6. Jonathan M. Borwein and Richard E. Crandall, 'Closed Forms: What They Are and Why We Care', Notices of the American Mathematical Society 60(1), 50–65 (2013), DOI 10.1090/noti936, https://doi.org/10.1090/noti936
7. Z.-D. Zhang, 'Conjectures on the exact solution of three-dimensional (3D) simple orthorhombic Ising lattices', Philosophical Magazine 87, 5309–5419 (2007), DOI 10.1080/14786430701646325, https://doi.org/10.1080/14786430701646325
8. F. Y. Wu, B. M. McCoy, M. E. Fisher, L. Chayes, 'Comment on a recent conjectured solution of the three-dimensional Ising model', Philosophical Magazine 88, 3093–3095 (2008), DOI 10.1080/14786430802537738, https://doi.org/10.1080/14786430802537738
9. Degang Zhang, 'Exact Solution for Three-Dimensional Ising Model', Symmetry 13, 1837 (2021), DOI 10.3390/sym13101837, https://doi.org/10.3390/sym13101837
10. Jacques H. H. Perk, 'Comment on Zhang, D. Exact Solution for Three-Dimensional Ising Model. Symmetry 2021, 13, 1837', Symmetry 15(2), 374 (2023), DOI 10.3390/sym15020374, https://doi.org/10.3390/sym15020374 (arXiv:2202.03136, https://arxiv.org/abs/2202.03136)
11. Rong Qiang Wei, 'Limiting free energy per particle for Ising Model by approximating its functional integral', arXiv:2208.10349 (2022), https://arxiv.org/abs/2208.10349
12. Craig A. Tracy and Harold Widom, 'On the diagonal susceptibility of the two-dimensional Ising model', Journal of Mathematical Physics 54, 123302 (2013), DOI 10.1063/1.4836779, https://doi.org/10.1063/1.4836779 (arXiv:1307.5913, https://arxiv.org/abs/1307.5913)
13. A. J. Guttmann, I. Jensen, J-M. Maillard, J. Pantone, 'Is the full susceptibility of the square-lattice Ising model a differentially algebraic function?', Journal of Physics A: Mathematical and Theoretical 49, 504002 (2016), DOI 10.1088/1751-8113/49/50/504002, https://doi.org/10.1088/1751-8113/49/50/504002 (arXiv:1607.04168, https://arxiv.org/abs/1607.04168)
14. Andrew Rechnitzer, 'Haruspicy 2: The anisotropic generating function of self-avoiding polygons is not D-finite', Journal of Combinatorial Theory, Series A 113(3), 520–546 (2006), DOI 10.1016/j.jcta.2005.04.010, https://doi.org/10.1016/j.jcta.2005.04.010 (arXiv:math/0406450, https://arxiv.org/abs/math/0406450)
15. S. Boukraa, S. Hassani, J-M. Maillard, 'Holonomic functions of several complex variables and singularities of anisotropic Ising n-fold integrals', Journal of Physics A: Mathematical and Theoretical 45, 494010 (2012), DOI 10.1088/1751-8113/45/49/494010, https://doi.org/10.1088/1751-8113/45/49/494010 (arXiv:1207.1784, https://arxiv.org/abs/1207.1784)
16. Wojciech Niedziółka and Jacek Wojtkiewicz, 'On nonintegrability of three-dimensional Ising model', Reports on Mathematical Physics 93, 271–285 (2024), DOI 10.1016/S0034-4877(24)00037-5, https://doi.org/10.1016/S0034-4877(24)00037-5 (arXiv:2309.07718, https://arxiv.org/abs/2309.07718)
17. Gandhimohan M. Viswanathan, 'A fermionic path integral for exact enumeration of polygons on the simple cubic lattice', arXiv:2305.17046 (2023), https://arxiv.org/abs/2305.17046
18. Gandhimohan M. Viswanathan, 'Mahler measures, elliptic curves, and L-functions for the free energy of the Ising model', Physical Review E 110, 054134 (2024), DOI 10.1103/PhysRevE.110.054134, https://doi.org/10.1103/PhysRevE.110.054134 (arXiv:2407.19531, https://arxiv.org/abs/2407.19531)
19. Estevão F. Borel, Aldo Procacci, Rémy Sanchis, Roger W. C. Silva, 'Anisotropic Ising Model in d+s Dimensions', Annales Henri Poincaré 26, 1519–1532 (2024), DOI 10.1007/s00023-024-01475-6, https://doi.org/10.1007/s00023-024-01475-6 (arXiv:2404.07079, https://arxiv.org/abs/2404.07079)
20. M. A. G. Portillo and M. G. E. da Luz, 'Correlation trends between the Ising critical temperature and the corresponding spanning tree constant', Journal of Statistical Mechanics: Theory and Experiment 2025, 043202 (2025), DOI 10.1088/1742-5468/adc4cd, https://doi.org/10.1088/1742-5468/adc4cd
21. Sheer El-Showk, Miguel F. Paulos, David Poland, Slava Rychkov, David Simmons-Duffin, Alessandro Vichi, 'Solving the 3D Ising model with the conformal bootstrap', Physical Review D 86, 025022 (2012), DOI 10.1103/PhysRevD.86.025022, https://doi.org/10.1103/PhysRevD.86.025022
22. Slava Rychkov, '3D Ising model: a view from the conformal bootstrap island', Comptes Rendus. Physique 21, 185–198 (2020), DOI 10.5802/crphys.23, https://doi.org/10.5802/crphys.23
23. Hugo Duminil-Copin, '100 Years of the (Critical) Ising Model on the Hypercubic Lattice', arXiv:2208.00864 (2022), https://arxiv.org/abs/2208.00864