Controlled large-p asymptotics of the Ising p-spin spin glass: finite-p corrections to the exactly solved random-energy limit and the fate of its finite-dimensional fluctuation theory

The infinite-range Ising p-spin spin glass becomes exactly solvable at p = infinity, where it reduces to Derrida's random energy model with a one-step replica-symmetry-breaking solution; this is the result with which Gross and Mezard introduced 'the simplest spin glass' in 1984, leaving three directions open: the structure of finite-p corrections, the incorporation of finite-dimensional fluctuations, and the control of the Onsager reaction terms in the TAP (Thouless-Anderson-Palmer) mean-field equations at large finite p. The p-to-infinity limit of the free energy is rigorously settled (Kouraich, Manai and Warzel, 2025), and corrections exponentially small in p are controlled only in the high-temperature (annealed) regime. The finite-p phase structure is rigorous (Zhou, 2024, 2026): for every p >= 3 the Parisi measure is replica symmetric, then one-step RSB, then full RSB below a Gardner threshold, so the approach to the p = infinity one-step solution is necessarily non-uniform in temperature. The TAP Onsager prong has been superseded by rigorous finite-p TAP theory (Chen and Panchenko, 2018; Chen and Tang, 2024). The open core is precise and nonempty: a systematic large-p expansion of the quenched free energy and of the Parisi order parameter q(x) with controlled convergence and temperature uniformity (including the large-p behavior of the RSB phase boundaries), together with the still-unrealized decision on whether large p makes fluctuations about mean field controllable in finite spatial dimensions (the proposed 1/p^2 expansion) or whether non-perturbative droplet excitations preclude it.

Background

A spin glass is a disordered system whose variables (Ising spins, $\sigma_i = \pm 1$) interact through quenched, i.e. frozen-in-place, random couplings; the infinite-range Sherrington-Kirkpatrick (SK) model, with random pair couplings between every spin pair, is the canonical mean-field example. Gross and Mezard ('The simplest spin glass', 1984) studied its $p$-spin generalization, in which the Hamiltonian couples randomly weighted products of $p$ distinct spins, $\mathcal{H} = -\sum_{i_1 < \cdots < i_p} J_{i_1 \ldots i_p}\,\sigma_{i_1}\cdots\sigma_{i_p}$, with independent Gaussian couplings $J$ scaled so that the energy is extensive in the number of spins $N$. They proved that in the limit $p \to \infty$ (taken after $N \to \infty$) the energies of any finite set of spin configurations with mutual overlaps $q^{(a,b)} = N^{-1}\sum_i \sigma_i^{(a)} \sigma_i^{(b)}$ strictly smaller than 1 become independent Gaussian random variables, so the model reduces to Derrida's random energy model (REM), in which all $2^N$ energy levels are independent. In that limit the model is exactly solvable by three independent routes: a direct microcanonical count of energy levels, the replica method, and the Thouless-Anderson-Palmer (TAP) mean-field equations. The freezing temperature is $T_c = 1/(2\sqrt{\ln 2})$ and the frozen free energy per spin is $F/N = -\sqrt{\ln 2}$ at zero magnetic field.

The replica method computes the disorder-averaged (quenched) free energy $\overline{\ln Z}$ by averaging $Z^n$ over $n$ copies ('replicas') of the system and taking $n \to 0$; its low-temperature solution requires replica symmetry breaking (RSB), encoded in a nondecreasing order-parameter function $q(x)$, $x \in [0,1]$, which determines the distribution of overlaps between thermodynamic pure states. For $p \to \infty$ the exact saddle is one-step RSB: $q(x)$ is a two-plateau step function, and Gross and Mezard showed in their Appendix A that allowing further hierarchical breaking produces no new saddle points. (By contrast, the annealed free energy, i.e. the logarithm of the disorder-averaged partition function $\overline{Z}$, equals the quenched one only at high temperature.) The TAP approach describes pure states instead through self-consistent mean-field equations for the local magnetizations $m_i$; these contain Onsager reaction terms, the feedback correction by which the magnetization of spin $i$ modifies the effective field acting back on it, and those terms vanish in the $p \to \infty$ frozen phase because $m_i = \pm 1$ exactly there.

At finite $p$ the model is no longer solvable in closed form. Gardner (1985) computed its mean-field phase diagram for general $p$ and found two transitions: at $T_1$ the replica-symmetric phase gives way to one-step RSB with a jump of the order parameter, and at a lower temperature $T_2$ (now called the Gardner transition) the one-step solution itself is unstable toward full RSB, in which $q(x)$ varies continuously over an interval. The spherical variant of the model, in which spins are constrained to fixed total length, is exactly solvable for every $p$ (Crisanti and Sommers, 1992) and shows no Gardner instability, so the Ising case is genuinely the harder one. On the rigorous side, Talagrand (2000) obtained low-temperature results for the mean-field $p$-spin model, Panchenko (2014) proved the Parisi variational formula for mixed $p$-spin models, which expresses the quenched free energy as an infimum over order-parameter functions $q(x)$, and the convergence of the $p$-spin free energy to the REM value as $p \to \infty$ is rigorously known (the ingredients are assembled by Kouraich, Manai and Warzel, 2025). What is missing is the quantitative layer between these two fixed points: the concrete form, convergence, and temperature uniformity of a systematic expansion about $p = \infty$, and whether the large-$p$ structure can also control fluctuations about mean field when the spins live in a finite-dimensional geometry.

Problem Statement

Develop the quantitative theory of large but finite $p$ for the infinite-range Ising $p$-spin spin glass of Gross and Mezard, i.e. controlled corrections to the exactly solved $p = \infty$ (random-energy-model) limit, in two linked parts, (a) and (b), which are jointly posed and both bear on this same large-$p$ target.

(a) Mean-field thermodynamics. Construct the systematic large-$p$ expansion of the quenched free energy and of the Parisi order-parameter function $q(x)$ (equivalently the Parisi measure), with explicit leading corrections, whether algebraic in $1/p$ or exponentially small in $p$, stated per temperature (and field) regime, together with controlled error bounds. The expansion must be reconciled with the now-established finite-$p$ phase structure: for every $p \ge 3$ the Parisi measure is replica symmetric for $0 < \beta \le \beta_1^p$ (here $\beta = 1/T$ is the inverse temperature), one-step RSB for $\beta_1^p < \beta \le \beta_2^p$, and full RSB for $\beta > \beta_2^p$. In particular it must determine the large-$p$ behavior of both boundaries $\beta_1^p$ and $\beta_2^p$ (the growth of $\beta_2^p$ being one instance) and the sense in which, at fixed inverse temperature $\beta$, the $p = \infty$ one-step solution is approached as $p \to \infty$.

(b) Fluctuations in finite spatial dimensions. Decide whether fluctuations about the mean-field solution become controlled at large $p$ when the spins are placed in a finite-dimensional geometry. The motivating observation is that the quadratic (propagator) masses of the replica Landau functional, i.e. the curvatures at the saddle that set the correlation lengths of the Gaussian fluctuations about it, grow as $p^2$ at the frozen saddle $Q = 1$ (where $Q$ denotes the overlap between two replicas) while being exponentially small for $Q < 1$, suggesting a perturbation theory in $1/p^2$. The question is whether a consistent and controlled finite-dimensional expansion can be organized (identifying the small parameter, organizing the loop/Ginzburg analysis, in which the Ginzburg criterion sets the width of the critical region where fluctuations dominate the mean-field behaviour, and deriving its consequences for critical behaviour), or whether non-perturbative excitations, at the level of interface and droplet estimates, preclude such an expansion in finite dimensions even at large $p$.

The third direction left open in Gross and Mezard (1984), the controlled size and effect of the Onsager reaction terms in the TAP mean-field equations at large finite $p$, has been substantially resolved by rigorous finite-$p$ TAP theory (Chen and Panchenko, 2018; Chen and Tang, 2024) and is not part of this problem.

References

1. D. J. Gross and M. Mezard, 'The simplest spin glass', Nuclear Physics B 240 (1984) 431-452. DOI: 10.1016/0550-3213(84)90237-2. https://doi.org/10.1016/0550-3213(84)90237-2
2. B. Derrida, 'Random-Energy Model: Limit of a Family of Disordered Models', Physical Review Letters 45 (1980) 79-82. DOI: 10.1103/PhysRevLett.45.79. https://doi.org/10.1103/PhysRevLett.45.79
3. B. Derrida, 'Random-energy model: An exactly solvable model of disordered systems', Physical Review B 24 (1981) 2613-2626. DOI: 10.1103/PhysRevB.24.2613. https://doi.org/10.1103/PhysRevB.24.2613
4. E. Gardner, 'Spin glasses with p-spin interactions', Nuclear Physics B 257 (1985) 747-765. DOI: 10.1016/0550-3213(85)90374-8. https://doi.org/10.1016/0550-3213(85)90374-8
5. V. Dobrosavljevic and D. Thirumalai, '1/p expansion for a p-spin interaction spin-glass model in a transverse field', Journal of Physics A: Mathematical and General 23 (1990) L767-L774. DOI: 10.1088/0305-4470/23/15/013. https://doi.org/10.1088/0305-4470/23/15/013
6. Y. Y. Goldschmidt, 'Solvable model of the quantum spin glass in a transverse field', Physical Review B 41 (1990) 4858-4861. DOI: 10.1103/PhysRevB.41.4858. https://doi.org/10.1103/PhysRevB.41.4858
7. A. Crisanti and H.-J. Sommers, 'The spherical p-spin interaction spin glass model: the statics', Zeitschrift für Physik B Condensed Matter 87 (1992) 341-354. DOI: 10.1007/BF01309287. https://doi.org/10.1007/BF01309287
8. S. Franz and G. Parisi, 'Critical properties of a three-dimensional p-spin model', The European Physical Journal B 8 (1999) 417-422. DOI: 10.1007/s100510050707. https://doi.org/10.1007/s100510050707
9. M. Talagrand, 'Rigorous low-temperature results for the mean field p-spins interaction model', Probability Theory and Related Fields 117 (2000) 303-360. DOI: 10.1007/s004400050009. https://doi.org/10.1007/s004400050009
10. M. A. Moore and B. Drossel, 'p-spin model in finite dimensions and its relation to structural glasses', Physical Review Letters 89 (2002) 217202. DOI: 10.1103/PhysRevLett.89.217202. https://doi.org/10.1103/PhysRevLett.89.217202
11. A. Bovier, I. Kurkova and M. Loewe, 'Fluctuations of the free energy in the REM and the p-spin SK models', The Annals of Probability 30 (2) (2002). DOI: 10.1214/aop/1023481004. https://doi.org/10.1214/aop/1023481004
12. A. Crisanti, L. Leuzzi and T. Rizzo, 'Complexity in mean-field spin-glass models: Ising p-spin', Physical Review B 71 (2005) 094202. DOI: 10.1103/PhysRevB.71.094202. https://doi.org/10.1103/PhysRevB.71.094202
13. M. A. Moore, 'Interface Free Energies in p-Spin Glass Models', Physical Review Letters 96 (2006) 137202. DOI: 10.1103/PhysRevLett.96.137202. https://doi.org/10.1103/PhysRevLett.96.137202
14. M. A. Moore and J. Yeo, 'Thermodynamic Glass Transition in Finite Dimensions', Physical Review Letters 96 (2006) 095701. DOI: 10.1103/PhysRevLett.96.095701. https://doi.org/10.1103/PhysRevLett.96.095701
15. T. Obuchi, H. Nishimori and D. Sherrington, 'Phase diagram of the p-spin-interacting spin glass with ferromagnetic bias and a transverse field in the infinite-p limit', Journal of the Physical Society of Japan 76 (2007) 054002. DOI: 10.1143/JPSJ.76.054002. https://doi.org/10.1143/JPSJ.76.054002
16. D. Panchenko, 'The Parisi formula for mixed p-spin models', The Annals of Probability 42 (3) (2014). DOI: 10.1214/12-AOP800. https://doi.org/10.1214/12-AOP800
17. W.-K. Chen and D. Panchenko, 'On the TAP free energy in the mixed p-spin models', Communications in Mathematical Physics 362 (2018) 219-252. DOI: 10.1007/s00220-018-3143-7. https://doi.org/10.1007/s00220-018-3143-7
18. J. Yeo and M. A. Moore, 'Possible instability of one-step replica symmetry breaking in p-spin Ising models outside mean-field theory', Physical Review E 101 (2020) 032127. DOI: 10.1103/PhysRevE.101.032127. https://doi.org/10.1103/PhysRevE.101.032127
19. W.-K. Chen and S. Tang, 'On the TAP Equations via the Cavity Approach in the Generic Mixed p-Spin Models', Communications in Mathematical Physics 405 (2024), article 87. DOI: 10.1007/s00220-024-04971-2. https://doi.org/10.1007/s00220-024-04971-2
20. A. Bovier and A. Schertzer, 'Fluctuations of the free energy in p-spin SK models on two scales', Probability Theory and Related Fields 189 (2024) 771-810. DOI: 10.1007/s00440-024-01296-y. https://doi.org/10.1007/s00440-024-01296-y
21. Y. Zhou, 'On the Gardner Transition in the Ising Pure p-Spin Glass', arXiv:2408.14630 (2024). https://arxiv.org/abs/2408.14630
22. S. Boettcher and G. E. Lau, 'Ground States of the Mean-Field Spin Glass with 3-Spin Couplings', Physical Review Letters 135 (2025) 037402. DOI: 10.1103/j159-lpfx. https://doi.org/10.1103/j159-lpfx
23. A. Kouraich, C. Manai and S. Warzel, 'The Quantum Random Energy Model is the Limit of Quantum p-Spin Glasses', arXiv:2505.02458 (2025, revised March 2026). https://arxiv.org/abs/2505.02458
24. J. Boursier, 'The Legendre structure of the TAP complexity for the Ising spin glass', arXiv:2604.20660 (2026). https://arxiv.org/abs/2604.20660
25. Y. Zhou, 'On the Gardner Transition in the Ising Pure p-Spin Glass II', arXiv:2608.06523 (2026). https://arxiv.org/abs/2608.06523
26. P. M. Aronow and P. Lopatto, 'Replica symmetry breaking at arbitrary depth', arXiv:2609.05317 (2026). https://arxiv.org/abs/2609.05317