Ultrametricity of the four-dimensional Edwards-Anderson spin glass in the thermodynamic limit: finite-size scaling of the constrained-Monte-Carlo test

Whether the equilibrium states of a finite-dimensional spin glass are organized ultrametrically, as the replica-symmetry-breaking solution of the mean-field Sherrington-Kirkpatrick model predicts, is a long-standing open question of disordered-systems physics. In 1997 Cacciuto, Marinari and Parisi attacked it with a constrained Monte Carlo method for the four-dimensional $\pm J$ Edwards-Anderson model at $T\simeq0.7T_c$: fixing two of the three mutual overlaps of three replica configurations, they measured the deviation integral $I^L$ (a distance-squared-weighted measure of the third-overlap distribution's weight outside the ultrametrically allowed window) on lattices $L=3$ to $8$ and found $I^L\sim L^{-2.21\pm0.04}$, compared by the authors with the exponent $8/3$ expected from the constrained mean-field analyses of Franz, Parisi and Virasoro. Severe finite-size effects (weight at $q>q_{EA}$, slow convergence of $q_{EA}^{(L)}$) and the difficulty of certifying thermalization of the constrained dynamics prevented a definitive extrapolation, and the authors left explicitly open whether the infinite-volume limit is fully ultrametric and whether the asymptotic approach matches the mean-field prediction. The problem remains open: mean-field ultrametricity is now a rigorous theorem (Panchenko 2013); in finite dimensions ultrametricity is known only conditionally on a non-trivial overlap distribution, which itself remains unproven; three-dimensional numerical evidence is contested; and no later work has revisited the four-dimensional constrained protocol or measured the asymptotic approach exponent.

Background

A spin glass is a magnetic system whose interacting degrees of freedom, Ising spins $\sigma_i=\pm1$, are coupled through quenched (frozen, i.e. held fixed while the spins thermalize) random interactions; the competition between frustrated bonds, which no spin assignment can all satisfy at once, destroys ordinary magnetic order. The Edwards-Anderson (EA) model places random nearest-neighbor couplings $J_{ij}=\pm1$ on a lattice — here the four-dimensional hypercubic lattice of linear size $L$ and volume $V=L^4$ — and is the standard lattice model of realistic spin glasses. Its mean-field idealization is the Sherrington-Kirkpatrick (SK) model, in which all $N$ spins interact through Gaussian random couplings of variance $1/N$. Parisi's replica-symmetry-breaking (RSB) solution of the SK model predicts that below the spin-glass transition the Gibbs probability measure splits into infinitely many pure states — extremal equilibrium components, analogous to a ferromagnet's two ground states but infinite in number and not related by any symmetry — studied through the replica formalism of many copies ('replicas') of the same disordered sample. The natural notion of distance between two configurations $\alpha$ and $\beta$ is built from the overlap $q_{\alpha\beta}=V^{-1}\sum_i\sigma_i^\alpha\sigma_i^\beta$: it takes its maximum value $q_{EA}$ (the Edwards-Anderson order parameter, the self-overlap of a pure state) inside one pure state and typically smaller values across different states. In the RSB solution the pure states, viewed through this distance, form an ultrametric space: every triple of equilibrium configurations satisfies the ultrametric inequality $d_{13}\le\max(d_{12},d_{23})$, stronger than the triangle inequality, so all triangles are isosceles with the two longest sides equal and the states sit at the leaves of a hierarchical random tree. The rival droplet (scaling) picture holds instead that the low-temperature phase contains only two pure states related by the global spin flip $\sigma\to-\sigma$, so the overlap probability distribution $P(q)$ collapses onto two delta functions at $\pm q_{EA}$ and no nontrivial hierarchy exists. Establishing whether the overlap structure of a finite-dimensional EA spin glass is ultrametric therefore separates the two theories directly; the review of Rammal, Toulouse and Virasoro surveys ultrametricity in physics generally.

Cacciuto, Marinari and Parisi (1997) tested this numerically in the four-dimensional $\pm J$ EA model at $T=1.4\simeq0.7T_c$ (the 4D critical temperature is $T_c\simeq2.0$) on lattices $L=3,\dots,8$. Their constrained Monte Carlo method evolves three replicas of the same disorder sample simultaneously, forbidding any update that moves the overlaps $q_{1,2}$ and $q_{2,3}$ outside narrow windows of half-width $\epsilon=0.04$, following the constrained-replica formalism introduced by Franz, Parisi and Virasoro. With $q_{1,2}=q_{2,3}=2q_{EA}/5\simeq0.21$ (taking $q_{EA}=0.54$), the ordinary triangle inequality only forces the third overlap $q\ge-7q_{EA}/5$, whereas ultrametricity forces $q\ge2q_{EA}/5$; the two bounds are far apart, which makes the test sharp. They quantified the deviation by the integral $I^L=\int_{-1}^{q_{1,2}}(q-q_{1,2})^2P_L(q)\,dq+\int_{q_{EA}}^{1}(q-q_{EA})^2P_L(q)\,dq$ (their eq. 6): the probability weight of the measured third-overlap distribution $P_L(q)$ lying outside the ultrametrically allowed window $[2q_{EA}/5,\,q_{EA}]$, weighted by the squared distance from that window. They found $I^L\simeq(0.76\pm0.03)L^{-2.21\pm0.04}$ over $L=3,\dots,8$, and $I^L\simeq(1.95\pm0.08)L^{-1.95\pm0.04}$ for the case $q_{1,2}=4q_{EA}/5$, $q_{2,3}=q_{EA}/5$, where ultrametricity predicts a delta function at $q_{EA}/5$ and the integral runs over the whole range $[-1,1]$ with the squared distance taken from the predicted peak position. The authors compared the exponent $2.21$ with the value $8/3$ expected from the analytic mean-field results of Franz, Parisi and Virasoro. Their own caveats define the open problem: at the largest thermalizable size $L=8$ only about half of the sampled configurations were ultrametric; the dominant violations came from finite-size weight at $q>q_{EA}$, expected to disappear only in the infinite-volume limit; $q_{EA}^{(L)}$ itself converged very slowly; certifying thermalization (equilibration) of the constrained dynamics is difficult because the standard checks — comparing dynamical with equal-time overlaps, or using the $\sigma\to-\sigma$ symmetry of $P(q)$ — are unavailable in the constrained ensemble; and even in the SK model a fully satisfactory numerical determination of the infinite-volume peak of $P(q)$ was lacking. Numerical tests of ultrametricity in finite dimensions have traditionally been benchmarked against the SK model.

Problem Statement

Determine whether the equilibrium overlap (pure-state) organization of the four-dimensional $\pm J$ Edwards-Anderson spin glass in zero field is ultrametric in the thermodynamic limit, with controlled finite-size corrections, as posed by Cacciuto, Marinari and Parisi (1997). Concretely, the thermodynamic-limit question requires controlled evidence at sizes substantially larger than $L=8$ at $T\simeq0.7T_c$, with an explicit, stated thermalization criterion for the constrained dynamics: (i) whether the constrained-ensemble deviation measure $I^L=\int_{-1}^{q_{1,2}}(q-q_{1,2})^2P_L(q)\,dq+\int_{q_{EA}}^{1}(q-q_{EA})^2P_L(q)\,dq$ — the distance-squared-weighted probability weight of the third-overlap distribution outside the ultrametrically allowed window, with two overlaps fixed at $q_{1,2}=q_{2,3}=2q_{EA}/5$ (and analogously in the case $q_{1,2}\ne q_{2,3}$, where the integral runs over $[-1,1]$ around the predicted delta-function position) — converges to zero as $L\to\infty$, and whether its asymptotic decay exponent equals the mean-field value $8/3$ (the measured $2.21\pm0.04$ on $L\le8$ leaves this open) or is altered by systematic finite-size corrections; (ii) whether the finite-volume artifacts — the weight at $q>q_{EA}$ and the slow convergence of $q_{EA}^{(L)}$ — disappear with increasing $L$ consistently with mean-field scaling; and (iii) whether the constrained three-replica overlap distribution converges to the ultrametric form (delta functions at the ultrametric bounds) in the infinite-volume limit. These components jointly support one objective — deciding whether the mean-field hierarchical state organization survives in finite dimension at $d=4$.

References

1. Angelo Cacciuto, Enzo Marinari, Giorgio Parisi (1997). Ultrametric structure of finite dimensional spin glasses. In: Complex Behaviour of Glassy Systems (Lecture Notes in Physics vol. 492), Springer, pp. 220-225. DOI: 10.1007/BFb0104828. https://doi.org/10.1007/bfb0104828
2. Angelo Cacciuto, Enzo Marinari, Giorgio Parisi (1997). A numerical study of ultrametricity in finite-dimensional spin glasses. Journal of Physics A: Mathematical and General 30, L263-L269. DOI: 10.1088/0305-4470/30/9/002; arXiv: cond-mat/9608161. https://arxiv.org/abs/cond-mat/9608161
3. Angelo Cacciuto (1997). Study of the ultrametric structure of Finite Dimensional Spin Glasses throught a constrained Monte Carlo dynamics. arXiv: cond-mat/9704053. https://arxiv.org/abs/cond-mat/9704053
4. S. Franz, G. Parisi, M. A. Virasoro (1992). Ultrametricity in an Inhomogeneous Simplest Spin Glass Model. Europhysics Letters 17, 5-9. DOI: 10.1209/0295-5075/17/1/002. https://doi.org/10.1209/0295-5075/17/1/002
5. S. Franz, G. Parisi, M. A. Virasoro (1992). The replica method on and off equilibrium. Journal de Physique I France 2, 1869-1880. DOI: 10.1051/jp1:1992115. https://doi.org/10.1051/jp1:1992115
6. G. Parisi (1979). Infinite Number of Order Parameters for Spin-Glasses. Physical Review Letters 43, 1754-1756. DOI: 10.1103/PhysRevLett.43.1754. https://doi.org/10.1103/PhysRevLett.43.1754
7. R. Rammal, G. Toulouse, M. A. Virasoro (1986). Ultrametricity for physicists. Reviews of Modern Physics 58, 765-788. DOI: 10.1103/RevModPhys.58.765. https://doi.org/10.1103/RevModPhys.58.765
8. Giorgio Parisi, Federico Ricci-Tersenghi (2000). On the origin of ultrametricity. Journal of Physics A: Mathematical and General 33, 113-129. DOI: 10.1088/0305-4470/33/1/307. https://doi.org/10.1088/0305-4470/33/1/307
9. Silvio Franz, Federico Ricci-Tersenghi (2000). Ultrametricity in three-dimensional Edwards-Anderson spin glasses. Physical Review E 61, 1121-1124. DOI: 10.1103/PhysRevE.61.1121. https://doi.org/10.1103/PhysRevE.61.1121
10. Guy Hed, A. P. Young, Eytan Domany (2004). Lack of Ultrametricity in the Low-Temperature Phase of Three-Dimensional Ising Spin Glasses. Physical Review Letters 92, 157201. DOI: 10.1103/PhysRevLett.92.157201. https://doi.org/10.1103/PhysRevLett.92.157201
11. Stefano Ciliberti, Enzo Marinari (2004). A Quantitative Clustering Approach to Ultrametricity in Spin Glasses. Journal of Statistical Physics 115, 557-580. DOI: 10.1023/B:JOSS.0000019831.87525.db. https://doi.org/10.1023/B:JOSS.0000019831.87525.db
12. Pierluigi Contucci, Cristian Giardinà, Claudio Giberti, Giorgio Parisi, Cecilia Vernia (2007). Ultrametricity in the Edwards-Anderson Model. Physical Review Letters 99, 057206. DOI: 10.1103/PhysRevLett.99.057206. https://doi.org/10.1103/PhysRevLett.99.057206 (arXiv: cond-mat/0607376)
13. Thomas Jörg, Florent Krzakala (2008). Comment on 'Ultrametricity in the Edwards-Anderson Model'. Physical Review Letters 100, 159701. DOI: 10.1103/PhysRevLett.100.159701. https://doi.org/10.1103/PhysRevLett.100.159701 (arXiv: 0709.0894)
14. Pierluigi Contucci, Cristian Giardinà, Claudio Giberti, Giorgio Parisi, Cecilia Vernia (2007). Answer to Comment on "Ultrametricity in the Edwards-Anderson Model" arXiv:0709.0894 (reply to the preceding comment; arXiv-only, not published in the journal). arXiv: 0712.1431. https://arxiv.org/abs/0712.1431
15. Helmut G. Katzgraber, Alexander K. Hartmann (2009). Ultrametricity and Clustering of States in Spin Glasses: A One-Dimensional View. Physical Review Letters 102, 037207. DOI: 10.1103/PhysRevLett.102.037207. https://doi.org/10.1103/PhysRevLett.102.037207
16. Helmut G. Katzgraber, Thomas Jörg, Florent Krzakala, Alexander K. Hartmann (2012). Ultrametric probe of the spin-glass state in a field. Physical Review B 86, 184405. DOI: 10.1103/PhysRevB.86.184405. https://doi.org/10.1103/PhysRevB.86.184405
17. A. Maiorano, G. Parisi, D. Yllanes (2013). Ultrametricity and long-range correlations in the Edwards-Anderson spin glass. arXiv: 1312.2790. https://arxiv.org/abs/1312.2790
18. Dmitry Panchenko (2013). The Parisi ultrametricity conjecture. Annals of Mathematics (2) 177, 383-393. arXiv: 1112.1003. https://arxiv.org/abs/1112.1003
19. Janus Collaboration (R. Alvarez Baños, A. Cruz, L. A. Fernandez, J. M. Gil-Narvion, A. Gordillo-Guerrero, M. Guidetti, A. Maiorano, F. Mantovani, E. Marinari, V. Martin-Mayor, J. Monforte-Garcia, A. Muñoz Sudupe, D. Navarro, G. Parisi, S. Perez-Gaviro, J. J. Ruiz-Lorenzo, S. F. Schifano, B. Seoane, A. Tarancon, R. Tripiccione, D. Yllanes) (2010). Nature of the spin-glass phase at experimental length scales. Journal of Statistical Mechanics: Theory and Experiment, P06026. DOI: 10.1088/1742-5468/2010/06/P06026. https://doi.org/10.1088/1742-5468/2010/06/P06026 (arXiv: 1003.2569)
20. Wenlong Wang, Jonathan Machta, Humberto Munoz-Bauza, Helmut G. Katzgraber (2017). Number of thermodynamic states in the three-dimensional Edwards-Anderson spin glass. Physical Review B 96, 184417. DOI: 10.1103/PhysRevB.96.184417. https://doi.org/10.1103/PhysRevB.96.184417
21. Miguel Aguilar-Janita, Silvio Franz, Victor Martin-Mayor, Javier Moreno-Gordo, Giorgio Parisi, Federico Ricci-Tersenghi, Juan J. Ruiz-Lorenzo (2024). Small field chaos in spin glasses: Universal predictions from the ultrametric tree and comparison with numerical simulations. Proceedings of the National Academy of Sciences 121, e2404973121. DOI: 10.1073/pnas.2404973121. https://doi.org/10.1073/pnas.2404973121 (arXiv: 2403.08503)
22. J. J. Ruiz-Lorenzo (2020). Nature of the Spin Glass Phase in Finite Dimensional (Ising) Spin Glasses. In: Order, Disorder and Criticality: Advanced Problems of Phase Transition Theory, Vol. 6 (ed. Yu. Holovatch), World Scientific. arXiv: 2006.12930. https://arxiv.org/abs/2006.12930
23. M. A. Moore (2021). Droplet-scaling versus replica symmetry breaking debate in spin glasses revisited. Physical Review E 103, 062111. DOI: 10.1103/PhysRevE.103.062111. https://doi.org/10.1103/PhysRevE.103.062111