Correlations among metastable states and the long-time post-jump dynamics in the spherical p-spin spin glass

Between its static transition $T_s$ and its dynamical transition $T_d$, the spherical $p$-spin spin glass is dominated by an exponentially large number of metastable states. In 1997 Cavagna, Giardina and Parisi introduced a three-replica potential $V_3(q_{12},q_{13},q_{23})$ for this model and showed that equilibrium states are not randomly distributed in phase space: local equilibrium (metastable) states surround each equilibrium state up to overlap $\bar q(T)$, the nearest equilibrium state lies at overlap $q^\star(T)$, and $\bar q, q^\star \to q_d$ as $T \to T_d^-$. They left open four issues: the mutual correlations of local equilibrium states and the fate of the long-time dynamics after a jump to such a state (return to the original state or exploration of new states), which they argued requires a many-replica potential; the exact value of the exponent $b$ in $q^\star - q_d \sim (T_d - T)^b$ (numerical fits give $b = 0.24$ or $b = 0.30$ for $p = 3$); validation against numerical simulations; and the extension to the random energy model. Later work has settled pairwise state statistics, basins of attraction, and the activated escape paths by which a state is left, and finite-size simulations now validate the qualitative scenario, but the fate of the long-time dynamics after a jump to a local equilibrium state - whether the system returns to the original state or explores new ones, which the escape-path results do not by themselves determine - along with the higher-order correlations among metastable states, the exact exponent $b$, and a dedicated random-energy-model analysis, remain open.

Background

Spin glasses are disordered systems whose fully connected (mean-field) versions are exactly solvable and capture glassy behavior. A central example is the spherical $p$-spin model, defined by the Hamiltonian $H(\sigma) = -\sum_{i_1 < \dots < i_p} J_{i_1 \dots i_p}\,\sigma_{i_1}\cdots\sigma_{i_p}$ on spin vectors $\sigma$ constrained to the sphere $\sum_i \sigma_i^2 = N$, with independent Gaussian couplings of variance $p!/(2N^{p-1})$. The overlap between two configurations $\sigma, \tau$ is $q = N^{-1}\sum_i \sigma_i \tau_i$; the self-overlap of an equilibrium pure state is the Edwards-Anderson parameter $q_{EA}$. For $p \ge 3$ the model's thermodynamics is described by one-step replica symmetry breaking (1RSB): the replicas (formal copies of the system used in the replica method) have an overlap matrix organized into blocks, which encodes a finite complexity, or configurational entropy $\Sigma$, meaning an exponential number $e^{N\Sigma}$ of pure states. The Thouless-Anderson-Palmer (TAP) approach provides a mean-field free energy whose minima are these states; metastable states are TAP minima that do not dominate the Gibbs measure.

The model has two transitions: a dynamical transition at $T_d$ (of the mode-coupling type, where the equilibrium relaxation time diverges and the paramagnetic state splits into exponentially many states) and a static transition at $T_s < T_d$ where the complexity vanishes. Between them, equilibrium is spread over exponentially many states of vanishing individual weight, while dynamics started from random initial conditions never equilibrates and instead ages (older systems respond increasingly slowly). Threshold states, the marginal states at the edge of stability where the replicon eigenvalue (the soft fluctuation mode of the overlap matrix, analogous to a Goldstone mode) vanishes, are the asymptotic dynamical attractors.

The two-replica (Franz-Parisi) potential $V_2(q_{12})$ is the free-energy cost of constraining a second replica to overlap $q_{12}$ with a replica equilibrated at temperature $T$; its nontrivial minimum at $q_{EA}$ reveals the glassy states but says nothing about how the states are disposed relative to one another. To map that structure, Cavagna, Giardina and Parisi (J. Phys. A 30, 4449 (1997)) introduced the three-replica potential $V_3(q_{12}, q_{13}, q_{23})$: replicas 1 and 2 are mutually constrained at overlap $q_{12}$ (replica 1 in an equilibrium state), and $V_3$ is the free-energy cost for a third replica to sit at overlaps $q_{13}, q_{23}$ from them. For $T_s < T < T_d$ (mainly $p = 3$) they found two nontrivial minima: $M_1$, where the third replica thermalizes in the same equilibrium state as replica 1 ($q_{13} = q_{EA}$), and $M_2$, existing for $0 \le q_{12} \le \bar q(T)$, which describes local equilibrium (metastable TAP) states of higher energy surrounding the constrained pair. Along the $M_2$ branch the state energy equals the equilibrium energy at a special overlap $q^\star$: the nearest equilibrium state of the same kind sits at overlap $q^\star$ from the reference one. As $T \to T_d^-$ all three overlaps merge, $\bar q, q^\star \to q_d$ (with $q_d = 0.5$ for $p = 3$), while $q^\star \to 0$ as $T \to T_s^+$; for $p = 3$ the fits give

$$q^\star(T) \approx -1.2\,(T_d - T)^b + c, \qquad b = 0.24 \text{ with } c = 0.5004, \text{ or } b = 0.30 \text{ with } c = q_d = 0.5,$$

while $q_{EA} - q_d \propto (T_d - T)^{1/2}$. Since the complexity $\Sigma$ stays finite at $T_d$, randomly distributed (uncorrelated) equilibrium states would keep their nearest neighbor at a strictly positive distance at $T_d$; the observed merging therefore implies that equilibrium states are not randomly distributed on the sphere. The random energy model (REM) of Derrida, in which the $2^N$ configuration energies are independent Gaussian variables and which is recovered as the $p \to \infty$ limit of the spherical $p$-spin, is the natural solvable test case for the generality of this picture. The paper's concluding section (Section 6) poses the open questions addressed here.

Problem Statement

The questions below trace to the concluding section of Cavagna, Giardina and Parisi, J. Phys. A 30, 4449 (1997). The common objective is a complete characterization of the organization of metastable states in the spherical $p$-spin model (mainly $p = 3$) in the window $T_s < T < T_d$, and of the long-time dynamics that follows from it. Concretely:

1. (Core) Determine the correlations among local equilibrium (metastable) states around a reference equilibrium state and settle the resulting dynamical question: after the system jumps (on an exponentially long time scale) from an equilibrium state to a nearby local equilibrium state, does it eventually return to the original state, or does it proceed to explore new equilibrium states?

2. (Quantitative component of the same analysis) Determine exactly the asymptotic law with which the characteristic overlaps approach the dynamical transition, i.e. the exponent $b$ in $q^\star(T) = a(T_d - T)^b + q_d$ as $T \to T_d^-$ (numerical fits give $b = 0.24$ with the constant free, or $b = 0.30$ with the constant fixed to $q_d = 0.5$, for $p = 3$).

3. (Generality component) Determine whether the organization found for $p = 3$ - the overlap structure of the local equilibrium states and the exponent $b$ - persists in the random energy model (the $p \to \infty$ limit), where many detailed computations become possible, computing the corresponding overlaps.

The original fourth issue, a comparison of the predicted scenario with numerical simulations, has been largely discharged by later work (Barrat and Franz 1998, J. Phys. A 31, L119; Franz and Parisi 1998, Physica A 261, 317; Charbonneau, Folena, Malatesta, Rizzo and Zamponi 2026, Phys. Rev. E 113, 034107) and is not part of the surviving core; the predicted scenario is consistent with that simulation evidence. These components jointly serve one objective: a self-contained account of how the metastable states of the basic mean-field glass model are organized in phase space and of what the system's long-time dynamics consequently does.

References

1. A. Cavagna, I. Giardina, G. Parisi, 'Structure of metastable states in spin glasses by means of a three replica potential', J. Phys. A: Math. Gen. 30 (1997) 4449-4466. DOI: 10.1088/0305-4470/30/13/004; arXiv:cond-mat/9611068. https://arxiv.org/abs/cond-mat/9611068
2. A. Cavagna, I. Giardina, G. Parisi, 'Barriers between metastable states in the p-spin spherical model', 1997. arXiv:cond-mat/9702069. https://arxiv.org/abs/cond-mat/9702069
3. A. Cavagna, I. Giardina, G. Parisi, 'An investigation of the hidden structure of states in a mean field spin glass model', J. Phys. A: Math. Gen. 30 (1997) 7021-7038. DOI: 10.1088/0305-4470/30/20/009; arXiv:cond-mat/9706027. https://arxiv.org/abs/cond-mat/9706027
4. A. Barrat, S. Franz, G. Parisi, 'Temperature evolution and bifurcations of metastable states in mean-field spin glasses, with connections with structural glasses', J. Phys. A: Math. Gen. 30 (1997) 5593-5612. DOI: 10.1088/0305-4470/30/16/006; arXiv:cond-mat/9703091. https://arxiv.org/abs/cond-mat/9703091
5. A. Barrat, S. Franz, 'Basins of attraction of metastable states of the spherical p-spin model', J. Phys. A: Math. Gen. 31 (1998) L119-L127. DOI: 10.1088/0305-4470/31/6/001; arXiv:cond-mat/9708154. https://arxiv.org/abs/cond-mat/9708154
6. S. Franz, G. Parisi, 'Effective potential in glassy systems: theory and simulations', Physica A 261 (1998) 317-339. DOI: 10.1016/S0378-4371(98)00315-X; arXiv:cond-mat/9711215. https://arxiv.org/abs/cond-mat/9711215
7. A. Cavagna, I. Giardina, G. Parisi, 'Role of saddles in mean-field dynamics above the glass transition', J. Phys. A: Math. Gen. 34 (2001) 5317-5326. DOI: 10.1088/0305-4470/34/26/302; arXiv:cond-mat/0104537. https://arxiv.org/abs/cond-mat/0104537
8. S. Franz, G. Parisi, 'Quasi-equilibrium in glassy dynamics: an algebraic view', J. Stat. Mech. (2013) P02003. DOI: 10.1088/1742-5468/2013/02/P02003; arXiv:1206.4067. https://arxiv.org/abs/1206.4067
9. R. Baviera, M. A. Virasoro, 'A method that reveals the multi-level ultrametric tree hidden in p-spin-glass-like systems', J. Stat. Mech. (2015) P12007. DOI: 10.1088/1742-5468/2015/12/P12007; arXiv:1508.01232. https://arxiv.org/abs/1508.01232
10. H. Mukaida, 'Non-differentiability of the effective potential and the replica symmetry breaking in the random energy model', 2015. arXiv:1510.02613. https://arxiv.org/abs/1510.02613
11. M. Baity-Jesi, G. Biroli, C. Cammarota, 'Activated aging dynamics and effective trap model description in the random energy model', J. Stat. Mech. (2018) 013301. DOI: 10.1088/1742-5468/aa9f43; arXiv:1708.03268. https://arxiv.org/abs/1708.03268
12. V. Ros, G. Biroli, C. Cammarota, 'Dynamical instantons and activated processes in mean-field glass models', SciPost Phys. 10 (2021) 002. DOI: 10.21468/scipostphys.10.1.002; arXiv:2006.08399. https://arxiv.org/abs/2006.08399
13. A. Pacco, G. Biroli, V. Ros, 'Curvature-driven pathways interpolating between stationary points: the case of the pure spherical 3-spin model', J. Phys. A: Math. Theor. 57 (2024) 07LT01. DOI: 10.1088/1751-8121/ad2039; arXiv:2311.18411. https://arxiv.org/abs/2311.18411
14. A. Pacco, A. Rosso, V. Ros, 'Triplets of local minima in a high-dimensional random landscape: correlations, clustering, and memoryless activated jumps', J. Stat. Mech. (2025) 033302. DOI: 10.1088/1742-5468/adbe40; arXiv:2410.18010. https://arxiv.org/abs/2410.18010
15. P. Charbonneau, G. Folena, E. M. Malatesta, T. Rizzo, F. Zamponi, 'Rare trajectories in a prototypical mean-field disordered model: insights into landscape and instantons', Phys. Rev. E 113 (2026) 034107. DOI: 10.1103/f4dj-d2sh; arXiv:2505.00107. https://arxiv.org/abs/2505.00107
16. A. Dembo, E. Subag, 'Dynamics for spherical spin glasses: Gibbs distributed initial conditions', 2025. arXiv:2503.23342. https://arxiv.org/abs/2503.23342
17. S. Franz, G. Parisi, 'Recipes for metastable states in spin glasses', J. Phys. I France 5 (1995) 1401-1415. DOI: 10.1051/jp1:1995201. https://doi.org/10.1051/jp1:1995201
18. J. Kurchan, G. Parisi, M. A. Virasoro, 'Barriers and metastable states as saddle points in the replica approach', J. Phys. I France 3 (1993) 1819-1838. DOI: 10.1051/jp1:1993217. https://doi.org/10.1051/jp1:1993217
19. A. Crisanti, H.-J. Sommers, 'The spherical p-spin interaction spin glass model: the statics', Z. Phys. B 87 (1992) 341-354. DOI: 10.1007/BF01309287. https://doi.org/10.1007/BF01309287
20. L. F. Cugliandolo, J. Kurchan, 'Analytical solution of the off-equilibrium dynamics of a long-range spin-glass model', Phys. Rev. Lett. 71 (1993) 173-176. DOI: 10.1103/PhysRevLett.71.173. https://doi.org/10.1103/PhysRevLett.71.173
21. B. Derrida, 'Random-energy model: An exactly solvable model of disordered systems', Phys. Rev. B 24 (1981) 2613-2626. DOI: 10.1103/PhysRevB.24.2613. https://doi.org/10.1103/PhysRevB.24.2613