The iterated star-triangle (ST) transformation defines an exact real-space renormalization flow on inhomogeneous two-dimensional Ising lattices: each ST step maps a nearest-neighbor Ising Hamiltonian on a triangular (hexagonal) lattice onto another one, so no multi-spin interactions are ever generated. Hilhorst (J. Stat. Mech. 2021) proved that even for frustrated initial couplings, whose iterates become complex, the bond variables S=1/sinh⁡2KS=1/\sinh 2K and C=coth⁡2KC=\coth 2K remain confined to the union of the real and imaginary axes, with an integer 'color' structure controlling the evolution. The paper explicitly leaves open how to exploit this framework: how the distribution of coupling strengths flows under many iterations and whether stationary limits exist; how the multiplicative exp⁡(g)\exp(g) partition-function factors accumulate so that free energies of frustrated lattices can be computed; how to overcome the droplet-type energy-entropy cancellations that limit reliable numerical iteration to roughly 100-200 steps; and, above all, how to extract with these techniques the two zero-temperature exponents of the two-dimensional Ising spin glass - the temperature exponent yy (equal to minus the domain-wall stiffness exponent θ≈−0.279\theta\approx-0.279) and the chaos exponent ζ\zeta - which are currently known only numerically. No analytic derivation of either exponent is currently known. The problem is open.

Background

The Ising model with nearest-neighbor couplings Ki(r⃗)K_i(\vec r) on a triangular lattice (couplings measured in units of J/kBTJ/k_B T, so positive means ferromagnetic and negative antiferromagnetic) admits an exact local identity, the star-triangle (ST) transformation: a triangle of three couplings K1,K2,K3K_1,K_2,K_3 can be replaced, after inserting a central 'decorating' spin σ0\sigma_0, by a three-legged star of couplings p1,p2,p3p_1,p_2,p_3, with the partition function unchanged up to a multiplicative constant. Applying the ST transformation to every up-triangle of a triangular lattice produces a hexagonal (honeycomb) lattice; repartitioning that lattice into down-stars and transforming again produces a new triangular lattice. Iterating yields a sequence T(0)→H(0)→T(1)→⋯T^{(0)}\to H^{(0)}\to T^{(1)}\to\cdots of triangular and hexagonal lattices carrying nearest-neighbor Ising Hamiltonians only - unlike decimation on the square lattice, which generates multi-spin interactions. This 'ST evolution' is therefore an exact real-space renormalization flow in the space of inhomogeneous (position-dependent) coupling sets; it was introduced in the late 1970s by the author of the source paper and coworkers and has been exploited mainly for smoothly inhomogeneous and semi-infinite (surface) Ising systems.

The source paper (Hilhorst, J. Stat. Mech. 2021) extends ST evolution to frustrated lattices, in which positive and negative couplings coexist so that some elementary plaquettes cannot satisfy all bonds simultaneously. For such initial conditions the iterated couplings Ki(n)K_i^{(n)} become complex. Introducing bond variables Si=1/sinh⁡2KiS_i=1/\sinh 2K_i and Ci=coth⁡2KiC_i=\coth 2K_i on triangle sides, and Si=sinh⁡2piS_i=\sinh 2p_i and Ci=cosh⁡2piC_i=\cosh 2p_i on star legs (in both cases Ci2−Si2=1C_i^2-S_i^2=1), one ST step is the rational map Si′=F1/2/(SjSk)S_i'=F^{1/2}/(S_jS_k), Ci′=(Ci+CjCk)/(SjSk)C_i'=(C_i+C_jC_k)/(S_jS_k) with F=2+S12+S22+S32+2C1C2C3F=2+S_1^2+S_2^2+S_3^2+2C_1C_2C_3 (cyclic i,j,ki,j,k), and each step multiplies the partition function by a constant exp⁡(g)\exp(g). The paper's central result is a confinement theorem: although the couplings go complex, the variables Si(n)S_i^{(n)} and Ci(n)C_i^{(n)} remain forever on the union R∪iR\mathbb R\cup i\mathbb R of the real and imaginary axes. The proof assigns each bond one of three 'colors' - black (Si,CiS_i,C_i real), red (SiS_i imaginary, CiC_i real), green (both imaginary) - whose transformation rules encode the parity structure of the frustration. For lattices with boundaries one must distinguish 'receding' from 'advancing' boundaries; at an advancing boundary the ST step has a two-fold sign freedom that must be fixed by a protocol (for example the uniform random protocol, which chooses the two symmetry-related decorations with probability 1/21/2 each), making the boundary evolution irreversible.

The intended flagship application is the two-dimensional Ising spin glass (2D ISG): an Ising model whose couplings are drawn independently from a distribution symmetric about zero, so that frustration is dense. In the droplet theory of Fisher and Huse, the 2D ISG is critical at T=0T=0: spin-glass order is absent at any T>0T>0 because droplet excitations of linear size LL cost an energy of order LθL^\theta with a negative stiffness exponent θ\theta. The correlation length therefore diverges as ξ∼T−1/y\xi\sim T^{-1/y} with y=−θ>0y=-\theta>0 (the temperature exponent), and the equilibrium spin configuration is chaotic under small temperature changes, with an overlap length ℓ∗∼∣ΔT∣−1/ζ\ell^*\sim|\Delta T|^{-1/\zeta} (the chaos exponent ζ\zeta). The theory of the 'critical state' just above the lower critical dimension (Thill and Hilhorst 1996) shows that yy and ζ\zeta are the two nontrivial zero-temperature exponents of the model and explains critical chaos in terms of interacting 'protodroplets'. Both exponents are known only numerically: exact ground-state and domain-wall computations give θ=−0.287(4)\theta=-0.287(4) (Hartmann et al. 2002) and, at the highest precision so far, θ=−0.2793(3)\theta=-0.2793(3) with domain-wall fractal dimension df=1.27319(9)d_f=1.27319(9) for Gaussian couplings (Khoshbakht and Weigel 2018); temperature chaos has been studied by Monte Carlo and scaling analyses (Ney-Nifle and Young 1997, who found the temperature- and bond-chaos exponents to be equal; Huse and Ko 1997). No analytic derivation of either exponent exists.

Hilhorst's numerical pilot study in the source paper shows both the promise and the obstacle: iterating the ST evolution on a 20×2020\times 20 periodic triangular spin glass behaves exactly as the confinement theorem predicts, but in double precision only about 100-200 iterations are reliable (versus 10410^4-10510^5 in the purely ferromagnetic regime), because of droplet-type energy-entropy cancellations in the free energy. The paper's conclusion therefore leaves open, explicitly: (i) how to extract the zero-temperature exponents yy and ζ\zeta of the 2D ISG with the aid of the ST techniques ('a way, still to be devised'); (ii) the structure and cumulative effect of the multiplicative factors exp⁡(g)\exp(g), which enter free-energy calculations; (iii) the quantitative evolution of the distribution of coupling strengths under iteration - whether asymptotic (stationary) limiting distributions exist in the bulk and for bounded lattices, and how they depend on the boundary; and (iv) what can be gained from boundary protocols other than the uniform random one, including protocols that do not confine the bond variables to R∪iR\mathbb R\cup i\mathbb R.

Problem Statement

Can the iterated star-triangle evolution be developed into a controlled exact renormalization scheme for frustrated, disordered two-dimensional Ising lattices, and used to determine the zero-temperature exponents of the two-dimensional Ising spin glass?

With the notation of the background (bond variables Si,CiS_i,C_i with Ci2−Si2=1C_i^2-S_i^2=1; the ST map Si′=F1/2/(SjSk)S_i'=F^{1/2}/(S_jS_k), Ci′=(Ci+CjCk)/(SjSk)C_i'=(C_i+C_jC_k)/(S_jS_k) with F=2+S12+S22+S32+2C1C2C3F=2+S_1^2+S_2^2+S_3^2+2C_1C_2C_3; confinement of all iterates to R∪iR\mathbb R\cup i\mathbb R; bond colors; boundary protocols), the source paper explicitly leaves open the following linked questions.

  1. Flow of the coupling distribution. Given an initial probability distribution of coupling strengths symmetric about zero (e.g. Gaussian or uniform), how do the single-bond and joint distributions of the bond variables evolve under many ST iterations? Does an asymptotic (stationary) limiting distribution exist for the bulk and for bounded lattices, and how does it depend on the boundary conditions and on the choice of boundary protocol (including whether protocols outside the confining class offer any advantage)?

  2. Free-energy bookkeeping. Each ST step multiplies the partition function by exp⁡(g)\exp(g). What is the structure of these factors, their cumulative effect under many iterations and disorder averaging, and how can they be exploited so that free energies and derived thermodynamic quantities of frustrated lattices can actually be computed under the flow?

  3. The target exponents. Using this framework - analytically from the confinement/color/parity structure, or numerically from a stable flow - determine the temperature exponent yy and the chaos exponent ζ\zeta of the two-dimensional Ising spin glass (in droplet notation y=−θy=-\theta with θ\theta the domain-wall stiffness exponent; ξ∼T−1/y\xi\sim T^{-1/y} and ℓ∗∼∣ΔT∣−1/ζ\ell^*\sim|\Delta T|^{-1/\zeta}), with the result accountable against the independent numerical benchmarks (θ=−0.2793(3)\theta=-0.2793(3) for Gaussian couplings). Both exponents must be obtained from the ST evolution itself; the quoted numerical benchmark serves only as an external check on the answer, not as a value to be restated.

A numerical route to (3) requires overcoming the droplet-type energy-entropy cancellations that currently limit reliable double-precision iteration to roughly 100-200 steps in the spin-glass case (versus 10410^4-10510^5 in the ferromagnetic regime), for example by higher-precision arithmetic, a reformulation of the ST variables, or another route into the asymptotic scaling regime.

The common objective of items (1)-(3) is a single one: to convert the proven confinement framework into a quantitatively usable exact renormalization tool for frustrated Ising lattices, of which the zero-temperature spin-glass exponents are the benchmark target. Item (3) is the mandatory deliverable: a complete solution must determine the two exponents. Items (1) and (2) are supporting results that a solution need not resolve separately.

References

  1. H. J. Hilhorst, Iterated star-triangle transformation on inhomogeneous 2D Ising lattices, Journal of Statistical Mechanics: Theory and Experiment 2021, 063210 (2021), DOI: 10.1088/1742-5468/ac06bf, https://doi.org/10.1088/1742-5468/ac06bf (arXiv:2104.00923)
  2. M. J. Thill and H. J. Hilhorst, Theory of the Critical State of Low-Dimensional Spin Glass, Journal de Physique I France 6, 67-95 (1996), DOI: 10.1051/jp1:1996130, https://doi.org/10.1051/jp1:1996130
  3. Daniel S. Fisher and David A. Huse, Ordered Phase of Short-Range Ising Spin-Glasses, Physical Review Letters 56, 1601-1604 (1986), DOI: 10.1103/PhysRevLett.56.1601, https://doi.org/10.1103/PhysRevLett.56.1601
  4. Daniel S. Fisher and David A. Huse, Equilibrium behavior of the spin-glass ordered phase, Physical Review B 38, 386-411 (1988), DOI: 10.1103/PhysRevB.38.386, https://doi.org/10.1103/PhysRevB.38.386
  5. Alexander K. Hartmann, Alan J. Bray, A. C. Carter, M. A. Moore, and A. P. Young, Stiffness exponent of two-dimensional Ising spin glasses for nonperiodic boundary conditions using aspect-ratio scaling, Physical Review B 66, 224401 (2002), DOI: 10.1103/PhysRevB.66.224401, https://doi.org/10.1103/PhysRevB.66.224401
  6. Hamid Khoshbakht and Martin Weigel, Domain-wall excitations in the two-dimensional Ising spin glass, Physical Review B 97, 064410 (2018), DOI: 10.1103/PhysRevB.97.064410, https://doi.org/10.1103/PhysRevB.97.064410
  7. Muriel Ney-Nifle and A. Peter Young, Chaos in a two-dimensional Ising spin glass, Journal of Physics A: Mathematical and General 30, 5311-5317 (1997), DOI: 10.1088/0305-4470/30/15/017, https://doi.org/10.1088/0305-4470/30/15/017
  8. David A. Huse and Lee-Fen Ko, Scaling and chaos in periodic approximations to the two-dimensional Ising spin glass, Physical Review B 56, 14597-14601 (1997), DOI: 10.1103/PhysRevB.56.14597, https://doi.org/10.1103/PhysRevB.56.14597
  9. Laurent Pierre, Bernard Bernu, and Laura Messio, Derivation of free energy, entropy and specific heat for planar Ising models: Application to Archimedean lattices and their duals, SciPost Physics 19, 025 (2025), DOI: 10.21468/SciPostPhys.19.1.025, https://doi.org/10.21468/SciPostPhys.19.1.025 (arXiv:2501.18569)
  10. De-Zhang Li, Xin Wang, and Xiao-Bao Yang, Free-fermion models and the two-dimensional Ising models under the zero field and imaginary field i(π/2)kBT, Entropy 27, 799 (2025), DOI: 10.3390/e27080799, https://doi.org/10.3390/e27080799 (arXiv:2406.00232, https://arxiv.org/abs/2406.00232)
  11. Bram Vanhecke, Jeanne Colbois, Laurens Vanderstraeten, Frank Verstraete, and Frederic Mila, Solving frustrated Ising models using tensor networks, Physical Review Research 3, 013041 (2021), DOI: 10.1103/PhysRevResearch.3.013041, https://doi.org/10.1103/PhysRevResearch.3.013041