Exact transfer-matrix spectrum and closed-form ground-state entropy of the Ising antiferromagnet on the Shastry–Sutherland lattice at $\alpha=1$

The antiferromagnetic Ising model on the Shastry–Sutherland lattice, $H=\sum_{\langle ij\rangle}\sigma_i\sigma_j+2\alpha\sum_{\langle lm\rangle}\sigma_l\sigma_m$, is a paradigmatic frustrated system: for $\alpha>1$ the ground state is exactly $2^{N/2}$-fold degenerate with entropy $\tfrac{1}{2}\log 2$ per site, but at $\alpha=1$ an enlarged manifold appears whose entropy has resisted exact treatment since 1981. A 2026 study by Shastry, Sutherland, Mila and Rufino determined this entropy numerically to high precision ($\Sigma=0.45877772$ per site via corner-transfer-matrix renormalization group, corroborated by exact finite-size transfer-matrix diagonalization and a rigorous variational lower bound $0.45668258$) and explicitly left the analytic problem open: no exact expressions are known for the eigenvalues of the row-to-row transfer matrix $\mathcal{T}_L(r)$ of the generalized model in which parallel-spin diagonal configurations carry weight $r$. The open problem is the exact analytic solution of that spectrum: a closed form for the largest eigenvalue (giving the exact ground-state entropy at $\alpha=1$), an analytic treatment of the second-largest eigenvalue lifting the $(2^L-1)$-fold degeneracy at $r=0$ (giving a closed-form correlation length), and the resulting thermodynamics of the constrained ensemble over $0<r\le1$. Known exact solutions of related Shastry–Sutherland Ising models are restricted to regimes excluding the frustrated $\alpha\ge 1$ case.

Background

The Shastry–Sutherland lattice (SSL) is a square lattice in which every second square, arranged in a checkerboard pattern, carries one diagonal bond; every site then has four neighbors along square edges plus one diagonal neighbor. Stretched so that all bonds have equal length it becomes the Archimedean lattice $(3^2,4,3,4)$ (one of the eleven uniform tilings of the plane, each site surrounded by two triangles, a square, another triangle and another square); with the diagonal bonds shortened it is the orthogonal-dimer lattice relevant to the magnetic insulator $\mathrm{SrCu_2(BO_3)_2}$ and to rare-earth tetraborides $\mathrm{RB_4}$. The antiferromagnetic Ising model on it is $H=\sum_{\langle ij\rangle}\sigma_i\sigma_j+2\alpha\sum_{\langle lm\rangle}\sigma_l\sigma_m$ with $\sigma=\pm1$, where $\langle ij\rangle$ runs over square-edge bonds and $\langle lm\rangle$ over the diagonal bonds (Shastry and Sutherland, Physica 108B, 1069 (1981)). For $\alpha>1$ the diagonal bonds dominate: each must carry antiparallel spins, giving two choices per diagonal and an exact ground-state degeneracy $2^{N/2}$, i.e. a residual (ground-state) entropy $\tfrac{1}{2}\log 2$ per site. At $\alpha=1$ the Hamiltonian decomposes exactly into a sum over the $N$ triangles formed by the diagonals, each contributing $\sigma_a\sigma_b+\sigma_b\sigma_c+\sigma_c\sigma_a$, which takes its minimum $-1$ for every non-monochromatic triangle; the ground-state manifold is therefore the set of all spin configurations with no monochromatic triangle, a strictly larger set whose entropy per site $\Sigma$ was bounded below (with a numerical error, later corrected) in the 1981 paper but never determined exactly. At $\alpha=\tfrac{1}{2}$ all bonds are equivalent (the uniform Archimedean point); Monte Carlo work (Yu, Phys. Rev. E 91, 062121 (2015)) shows that point is only weakly frustrated with a long-range-ordered ground state, so the extensive degeneracy is special to $\alpha\ge 1$. The 2026 source paper (Shastry, Sutherland, Mila, Rufino, arXiv:2605.19831) attacks $\Sigma$ at $\alpha=1$ with two tools. First, a row-to-row transfer matrix: successive lattice rows alternate between two types, (A) and (B), whose diagonal bonds slant in opposite directions, and the matrix elements factorize into two-site operators $V^{(A)}_{ab}$ (their explicit spin-operator form is Eq. (7) of that paper) and mirror images $V^{(B)}_{ab}=V^{(A)}_{ba}$; the full transfer matrix is $\mathcal{T}_L=\mathcal{T}_B\mathcal{T}_A$, a product over interleaved odd and even bonds, and $\Sigma=\lim_{L\to\infty}\frac{1}{2L}\log\lambda^{\max}(L)$ where $\lambda^{\max}$ is the largest eigenvalue. In vertex-model language (each square with its diagonal is a vertex), the model has ten allowed local configurations, and the parameter $0\le r\le 1$ continuously turns on the weight of configurations with parallel spins on a diagonal bond: $r=0$ recovers the $\alpha>1$ manifold, $r=1$ the full $\alpha=1$ manifold. Second, the corner transfer matrix renormalization group (CTMRG; Nishino and Okunishi, J. Phys. Soc. Jpn. 65, 891 (1996)), a tensor-network algorithm that approximates the environment of a site directly in the thermodynamic limit with a truncation dimension $\chi$, giving $\Sigma=0.45877772$, a fraction of ferromagnetic (parallel-spin) diagonal bonds $n_{fmd}=0.20395387$, and a spin-spin correlation length $\xi=0.476859$ lattice constants at $\alpha=1$, converged at $\chi=16$. Exact diagonalization of $\mathcal{T}_L$ for $L=4,6,8,10$ gives $\Sigma(L)=0.456592,0.458905,0.458785,0.458775$, extrapolating to $0.4588\pm0.0002$, and a Rayleigh–Ritz variational ansatz built from the largest eigenvector of the $A$-layer (an exact tensor product with eigenvalue $((3+\sqrt5)/2)^{L/2}$) yields the rigorous lower bound $\Sigma_{\text{bound}}=\frac14\log\left[\frac{1}{40}(27\sqrt5+\sqrt{3390\sqrt5+7590}+65)\right]=0.45668258$, correcting the erroneous 1981 value $0.4812$. At $r=0$ the transfer matrix is 'superstable' (a term from Sutherland and Shastry, J. Stat. Phys. 33, 477 (1983), meaning all configurations enter the leading eigenvector with equal weight): $\lambda^{\max}=2^L$ with a flat eigenvector, and $[\mathcal{T}(0)]^2=2^L\mathcal{T}(0)$, so the entire remaining spectrum is a $(2^L-1)$-fold degenerate zero. First-order perturbation theory gives $\lambda^{\max}(r)\sim 2^L(1+rL/4)$, hence $\Sigma(r)\sim\frac{\log 2}{2}+\frac{r}{8}$ and $n_{fmd}(r)\sim r/4$, while the second eigenvalue could only be estimated numerically, $\lambda^{(2)}(r)\sim\frac{r}{4}2^L(1+rL/4)$, from the CTMRG-inferred limit $\lim_{r\to0}\lambda^{(2)}(r)/(r\lambda^{\max}(r))=\frac14$; this produces the asymptotic correlation length $\xi(r)=\left(\log|\lambda^{\max}(r)/\lambda^{(2)}(r)|\right)^{-1}\sim(2\log 2-\log r)^{-1}$, diverging logarithmically as $r\to 0$. The known exact solutions adjacent to this problem do not reach it: Strečka (Phys. Lett. A 349, 505 (2006)) mapped the spin-$\tfrac12$ Ising model on the SSL to the exactly solved bathroom-tile (4-8) lattice by a star-triangle transformation, but that mapping is valid only for ferromagnetic intra-dimer (diagonal) coupling; Rousochatzakis, Läuchli and Moessner (Phys. Rev. B 85, 104415 (2012)) related the model to the Cairo pentagonal lattice in regimes excluding $\alpha\ge 1$; Dublenych (Phys. Rev. Lett. 109, 167202 (2012)) solved the ground-state selection problem in an applied field, not the zero-field degeneracy counting; and the tensor-network framework of Vanhecke et al. (Phys. Rev. Research 3, 013041 (2021)) was demonstrated on kagome-lattice models. Finally, a footnote of the source paper notes that in a finite-temperature study the factor $r$ 'can be (somewhat artificially) generated by adding energy $-\log r\times T$ for configurations with parallel spins on a diagonal'; the paper's own $r$-dependence results are zero-temperature constrained-ensemble results.

Problem Statement

Derive exact analytic expressions for the eigenvalues of the row-to-row transfer matrix $\mathcal{T}_L(r)$ of the antiferromagnetic Ising model on the Shastry–Sutherland lattice — the staggered product of two-site operators defined by Eqs. (7)–(13) of arXiv:2605.19831, acting on $2^L$ spin configurations with periodic boundary conditions, where $0\le r\le1$ continuously weights configurations with parallel spins on diagonal bonds ($r=0$ recovering the $\alpha>1$ manifold, $r=1$ the full $\alpha=1$ manifold) — together with their thermodynamic limits. Concretely, the problem asks for: (1) a closed form for the largest eigenvalue $\lambda^{\max}(L)$, or directly for $\Sigma=\lim_{L\to\infty}\frac{1}{2L}\log\lambda^{\max}(L)$ at $r=1$, i.e. the exact ground-state entropy per site at $\alpha=1$, replacing the numerical value $0.45877772$ and sharpening the rigorous variational lower bound $0.45668258$; (2) an analytic determination of the second-largest eigenvalue $\lambda^{(2)}(r)$ for small $r$ — lifting the $(2^L-1)$-fold degeneracy of the $r=0$ spectrum, which the source paper could only estimate numerically via CTMRG — yielding a closed-form correlation length $\xi(r)=\left(\log|\lambda^{\max}(r)/\lambda^{(2)}(r)|\right)^{-1}$ and placing the asymptotic law $\xi(r)\sim(2\log2-\log r)^{-1}$ on exact footing; and (3) the thermodynamics of the constrained ensemble over $0<r\le1$ — the entropy $\Sigma(r)$, the density of parallel-spin diagonal bonds $n_{fmd}(r)$, and the correlation length $\xi(r)$ as exact functions of $r$. These three deliverables are spectral consequences of a single object, the exact solution of $\mathcal{T}_L(r)$, and jointly constitute one objective: converting the numerical CTMRG and finite-size characterization of the $\alpha=1$ Shastry–Sutherland Ising antiferromagnet into an exact analytic solution. Partial exact results — for instance a closed form for $\lambda^{\max}$ alone, or a rigorous degenerate-perturbative $\lambda^{(2)}(r)$ alone — constitute progress on the same problem.

References

1. B. Sriram Shastry, Bill Sutherland, Frédéric Mila, Afonso Rufino, 'Ground-state Entropy of the Ising model on a Frustrated lattice', arXiv:2605.19831 (2026), DOI 10.48550/arxiv.2605.19831, https://arxiv.org/abs/2605.19831
2. B. S. Shastry and B. Sutherland, 'Exact ground state of a quantum mechanical antiferromagnet', Physica B+C 108, 1069 (1981), DOI 10.1016/0378-4363(81)90838-x, https://doi.org/10.1016/0378-4363(81)90838-x
3. Jozef Strečka, 'Exact solution of the spin-1/2 Ising model on the Shastry–Sutherland (orthogonal-dimer) lattice', Physics Letters A 349, 505 (2006), DOI 10.1016/j.physleta.2005.09.050, arXiv:cond-mat/0505637, https://arxiv.org/abs/cond-mat/0505637
4. I. Rousochatzakis, A. M. Läuchli, R. Moessner, 'Quantum magnetism on the Cairo pentagonal lattice', Physical Review B 85, 104415 (2012), DOI 10.1103/PhysRevB.85.104415, https://doi.org/10.1103/PhysRevB.85.104415
5. Bram Vanhecke, Jeanne Colbois, Laurens Vanderstraeten, Frank Verstraete, Frédéric Mila, 'Solving frustrated Ising models using tensor networks', Physical Review Research 3, 013041 (2021), DOI 10.1103/PhysRevResearch.3.013041, arXiv:2006.14341, https://arxiv.org/abs/2006.14341
6. Unjong Yu, 'Ising antiferromagnet on the Archimedean lattices', Physical Review E 91, 062121 (2015), DOI 10.1103/PhysRevE.91.062121, arXiv:1606.08191, https://doi.org/10.1103/PhysRevE.91.062121
7. Yuriy Dublenych, 'Ground States of the Ising Model on the Shastry-Sutherland Lattice and the Origin of the Fractional Magnetization Plateaus in Rare-Earth-Metal Tetraborides', Physical Review Letters 109, 167202 (2012), DOI 10.1103/PhysRevLett.109.167202, arXiv:1207.3030, https://doi.org/10.1103/PhysRevLett.109.167202
8. Tomotoshi Nishino and Kouichi Okunishi, 'Corner Transfer Matrix Renormalization Group Method', Journal of the Physical Society of Japan 65, 891 (1996), DOI 10.1143/JPSJ.65.891, https://doi.org/10.1143/JPSJ.65.891
9. Bill Sutherland and B. Sriram Shastry, 'Exact solution of a large class of interacting quantum systems exhibiting ground state singularities', Journal of Statistical Physics 33, 477 (1983), DOI 10.1007/bf01009806, https://doi.org/10.1007/bf01009806