Ultrametricity from stochastic stability in short-range spin glasses: the unresolved finite-dimensional core of the Parisi-Ricci-Tersenghi program

In 'On the origin of ultrametricity' (J. Phys. A 33, 113 (2000)), Parisi and Ricci-Tersenghi showed that in mean-field-like spin glasses, stochastic stability (replica equivalence) together with separability/overlap equivalence forces the equilibrium configurations into an ultrametric hierarchy, and they posed five open directions. Later rigorous work has resolved the mean-field core of these directions: Panchenko proved that the Ghirlanda-Guerra identities imply ultrametricity, which discharges the finite-k algebraic exceptions and, for SK-type models, the stochastic-stability-to-ultrametricity implication. The numerical test the authors proposed was carried out in three dimensions with partly contested conclusions. What survives as a precise open problem is the finite-dimensional core: establish (or refute), for short-range Edwards-Anderson spin glasses on $\mathbb{Z}^d$ without any long-range perturbation, the equivalence of separability and overlap equivalence for the physical site overlap and the implication from stochastic stability to ultrametricity, posed as in the source under the assumption of a non-trivial overlap distribution.

Background

A spin glass is a disordered magnet in which randomly placed ferro- and antiferromagnetic couplings frustrate any uniform ordering; the standard lattice model is the Edwards-Anderson (EA) model, Ising spins on a $d$-dimensional lattice with independent random nearest-neighbor couplings, while the fully connected Sherrington-Kirkpatrick (SK) model is its mean-field analogue. The central observable is the overlap $q_{ab}=N^{-1}\sum_i \sigma_i^a\sigma_i^b$ between two independent equilibrium replicas ($a$ and $b$) of the system, and the structure of the probability distribution $P(q)$ of this overlap characterizes the low-temperature phase. In the Parisi replica-symmetry-breaking (RSB) picture of the SK model, $P(q)$ is non-trivial and the organization of equilibrium configurations is ultrametric: for three independent replicas, the two smaller of the three pairwise overlaps coincide, so the distances $d_{ab}=\sqrt{q_{max}-q_{ab}}$ satisfy $d_{ac}\le\max(d_{ab},d_{bc})$ and the configurations form a hierarchy of nested clusters. The competing droplet/scaling picture instead predicts only a trivial $P(q)$ for short-range models. Two structural properties underlie the physics derivation. Stochastic stability (or replica equivalence, introduced by Aizenman and Contucci) states that infinitesimal perturbations of the couplings leave the thermodynamic observables unchanged; in mean-field models the perturbation is normalized as $O(N^{-1/2})$, while for short-range models it is meant in its metastate formulation, as invariance of the metastate law under asymptotically vanishing perturbations of the couplings. It implies the Ghirlanda-Guerra (GG) identities, an infinite hierarchy of constraints on the joint distributions of the overlaps of $k$ replicas. Separability is the replica-space statement that any matrix $M_{ab}$ built covariantly from the replica overlap matrix $Q_{ab}=(q_{ab})$ depends on the replica pair $(a,b)$ only through $Q_{ab}$ - equivalently, that the relevant connected correlation functions do not fluctuate in the thermodynamic limit; its configuration-space counterpart, overlap equivalence, is the stronger statement that every generalized overlap $q_O=N^{-1}\sum_i O_i(\sigma)O_i(\tau)$ becomes a function of the standard overlap $q$ at fixed $q$ in the thermodynamic limit, so that fixing that overlap collapses the Gibbs state onto a single pure state (an extremal, indecomposable equilibrium state). Parisi and Ricci-Tersenghi proved that stochastic stability plus separability/overlap equivalence imply ultrametricity at the level of the $k$-replica overlap statistics, and posed the directions listed below. A further distinction matters in finite dimension: the site overlap sums spin products at the same lattice sites, while the link overlap averages products of the coupling-weighted link variables; mean-field identities can transfer to the lattice for the covariance- or link-type overlap more easily than for the site overlap, and identifying the two is itself part of the open problem. The rigorous finite-volume framework in which such questions are properly posed for short-range models is the metastate (a probability distribution on infinite-volume Gibbs states for fixed disorder, introduced by Newman and Stein), which encodes how the chosen pure state depends on the disorder realization and boundary conditions.

Problem Statement

The source paper left five directions open; after the rigorous resolution of the mean-field parts (refs 13-14), the surviving problem is the finite-dimensional core, stated by Parisi and Ricci-Tersenghi as the question of whether their ultrametricity conclusion and the identities behind it extend to realistic short-range spin glasses. Formally, for the Edwards-Anderson Ising spin glass on $\mathbb{Z}^d$ (and related short-range models) with no added long-range perturbation: (1) Prove or disprove that stochastic stability (which in mean-field-type models implies the Ghirlanda-Guerra identities for the site overlap) implies the ultrametric organization of the low-temperature Gibbs states, without a separate separability or overlap-equivalence assumption, or exhibit a stochastically stable, non-ultrametric short-range model; whether stochastic stability yields the Ghirlanda-Guerra identities for the physical site overlap in short-range models is itself part of what (1) asks. (2) Prove or disprove the equivalence between separability (the replica-space statement that the relevant connected correlation functions do not fluctuate in the thermodynamic limit) and overlap equivalence (the configuration-space statement that fixing one overlap between two replicas collapses their joint state onto a pure state), which for short-range interactions is no longer the clean equivalence it is in the long-range case, including the identification of the link/covariance overlap, for which the existing rigorous results are proved, with the physical site overlap. Both (1) and (2) are required: the finite-dimensional core is resolved only when both are settled, each by a proof or by an explicit counterexample. As in the source paper, these questions are posed under the assumption that the overlap distribution $P(q)$ does not collapse to a delta function in the large-$N$ limit; whether that assumption holds in finite dimension (the RSB-versus-droplet question) is, in the authors' own words, beyond the reach of their arguments and is not part of the present problem. The source paper additionally proposed to check the numerical validity of overlap equivalence through simulations in the fixed-overlap ensemble; the three-dimensional execution of that test remains contested (refs 5, 6, 9-11, 16).

References

1. G. Parisi and F. Ricci-Tersenghi, 'On the origin of ultrametricity', J. Phys. A: Math. Gen. 33, 113 (2000). DOI: 10.1088/0305-4470/33/1/307; arXiv:cond-mat/9905189, https://arxiv.org/abs/cond-mat/9905189 (source paper)
2. S. Ghirlanda and F. Guerra, 'General properties of overlap probability distributions in disordered spin systems. Towards Parisi ultrametricity', J. Phys. A: Math. Gen. 31, 9143 (1998). DOI: 10.1088/0305-4470/31/46/006, https://doi.org/10.1088/0305-4470/31/46/006
3. C. M. Newman and D. L. Stein, 'Metastate approach to thermodynamic chaos', Phys. Rev. E 55, 5194 (1997). DOI: 10.1103/PhysRevE.55.5194, https://doi.org/10.1103/PhysRevE.55.5194
4. M. Aizenman and P. Contucci, 'On the Stability of the Quenched State in Mean-Field Spin-Glass Models', J. Stat. Phys. 92, 765 (1998). DOI: 10.1023/A:1023080223894, https://doi.org/10.1023/A:1023080223894
5. S. Franz and F. Ricci-Tersenghi, 'Ultrametricity in three-dimensional Edwards-Anderson spin glasses', Phys. Rev. E 61, 1121 (2000). DOI: 10.1103/PhysRevE.61.1121, https://doi.org/10.1103/PhysRevE.61.1121
6. G. Hed, A. P. Young and E. Domany, 'Lack of Ultrametricity in the Low-Temperature Phase of Three-Dimensional Ising Spin Glasses', Phys. Rev. Lett. 92, 157201 (2004). DOI: 10.1103/PhysRevLett.92.157201, https://doi.org/10.1103/PhysRevLett.92.157201
7. P. Contucci, C. Giardina, C. Giberti and C. Vernia, 'Overlap Equivalence in the Edwards-Anderson Model', Phys. Rev. Lett. 96, 217204 (2006). DOI: 10.1103/PhysRevLett.96.217204, https://doi.org/10.1103/PhysRevLett.96.217204
8. P. Contucci and C. Giardina, 'The Ghirlanda-Guerra Identities', J. Stat. Phys. 126, 917 (2006). DOI: 10.1007/s10955-006-9091-1, https://doi.org/10.1007/s10955-006-9091-1
9. P. Contucci, C. Giardina, C. Giberti, G. Parisi and C. Vernia, 'Ultrametricity in the Edwards-Anderson Model', Phys. Rev. Lett. 99, 057206 (2007). DOI: 10.1103/PhysRevLett.99.057206, https://doi.org/10.1103/PhysRevLett.99.057206
10. T. Jörg and F. Krzakala, 'Comment on Ultrametricity in the Edwards-Anderson Model', Phys. Rev. Lett. 100, 159701 (2008). DOI: 10.1103/PhysRevLett.100.159701, https://doi.org/10.1103/PhysRevLett.100.159701
11. P. Contucci, C. Giardina, C. Giberti, G. Parisi and C. Vernia, 'Contucci et al. Reply:', Phys. Rev. Lett. 100, 159702 (2008). DOI: 10.1103/PhysRevLett.100.159702, https://doi.org/10.1103/PhysRevLett.100.159702
12. H. G. Katzgraber and A. K. Hartmann, 'Ultrametricity and Clustering of States in Spin Glasses: A One-Dimensional View', Phys. Rev. Lett. 102, 037207 (2009). DOI: 10.1103/PhysRevLett.102.037207, https://doi.org/10.1103/PhysRevLett.102.037207
13. D. Panchenko, 'A connection between the Ghirlanda-Guerra identities and ultrametricity', Ann. Probab. 38, 1416 (2010). DOI: 10.1214/09-AOP484, https://doi.org/10.1214/09-AOP484
14. D. Panchenko, 'The Parisi ultrametricity conjecture', Ann. of Math. (2) 177, 383 (2013). DOI: 10.4007/annals.2013.177.1.8, https://doi.org/10.4007/annals.2013.177.1.8
15. P. Contucci, E. Mingione and S. Starr, 'Factorization Properties in d-Dimensional Spin Glasses. Rigorous Results and Some Perspectives', J. Stat. Phys. 151, 809 (2013). DOI: 10.1007/s10955-013-0730-z, https://doi.org/10.1007/s10955-013-0730-z
16. A. Maiorano, G. Parisi and D. Yllanes, 'Ultrametricity and long-range correlations in the Edwards-Anderson spin glass' (2013). arXiv:1312.2790, https://arxiv.org/abs/1312.2790
17. V. Martin-Mayor, J. J. Ruiz-Lorenzo, B. Seoane and A. P. Young, 'Numerical Simulations and Replica Symmetry Breaking' (2022). arXiv:2205.14089, https://arxiv.org/abs/2205.14089
18. S. Chatterjee, 'Spin glass phase at zero temperature in the Edwards-Anderson model' (2023). arXiv:2301.04112, https://arxiv.org/abs/2301.04112
19. C. M. Newman, N. Read and D. L. Stein, 'Proof of Single-Replica Equivalence in Short-Range Spin Glasses', Phys. Rev. Lett. 130, 077102 (2023). DOI: 10.1103/PhysRevLett.130.077102, https://doi.org/10.1103/PhysRevLett.130.077102
20. N. Read, 'Structural analysis of Gibbs states and metastates in short-range classical spin glasses: indecomposable metastates, dynamically-frozen states, and metasymmetry' (2024). arXiv:2406.13071, https://arxiv.org/abs/2406.13071
21. T. Kim, 'Heavy-Tailed Mixed p-Spin Spherical Model: Breakdown of Ultrametricity and Failure of the Parisi Formula' (2025). arXiv:2506.23987, https://arxiv.org/abs/2506.23987