Physical realization of the Parisi spin-glass order parameter q(x): pure-state decompositions of the Gibbs measure in short-range spin glasses and non-random glass formers

Parisi's 1983 paper 'Order Parameter for Spin-Glasses' gave the replica order-parameter function $q(x)$ a proposed physical meaning: the Gibbs state of a spin glass should be a mixture of many pure thermodynamic states, $q(x)$ being the inverse cumulative distribution of the overlaps between those states, accessible in principle through the two-replica generating function $g(y)$. The paper left the construction itself open: no method to decompose the Gibbs state into pure states, clustering assumptions taken heuristically, the measurement program only sketched, and no justification of the framework outside disordered spin Hamiltonians. The measurement program and the field-limit statement were subsequently developed, and the mean-field framework became rigorous (Parisi formula, ultrametricity), but the central open core survives: whether the low-temperature Gibbs state of realistic short-range spin glasses really decomposes into many pure states with a non-trivial overlap distribution, and whether non-random amorphous systems (structural glasses, hard-sphere packings) possess many pure thermodynamic states in the infinite-volume limit at all, so that an order parameter of the $q(x)$ type is physically realized there. The problem is to settle this pure-state structure, constructively or by refutation, outside the mean-field Sherrington-Kirkpatrick model.

Background

A spin glass is a magnet whose exchange couplings $J_{ij}$ between Ising spins $\sigma_i=\pm 1$ are random in sign, so that no spin configuration can satisfy all bonds simultaneously (frustration). The Edwards-Anderson (EA) model places such random nearest-neighbor couplings on a lattice in dimension $d$; the Sherrington-Kirkpatrick (SK) model makes every pair of the $N$ spins interact through a Gaussian coupling of variance $1/N$. Below a freezing temperature these models enter a glassy phase whose correct description became one of the central unsolved problems of statistical mechanics.

The replica trick computes the disorder-averaged free energy by introducing $n$ identical copies ('replicas') of the system and analytically continuing $n\to 0$. Parisi's replica-symmetry-breaking (RSB) solution of the SK model found that the order parameter is not a single number but a monotone function $Q(x)$ on the interval $x\in[0,1]$ (Parisi 1979, 1980).

Parisi's 1983 paper, 'Order Parameter for Spin-Glasses', Phys. Rev. Lett. 50, 1946 (1983), proposed a physical interpretation of $Q(x)$. Writing the infinite-volume Gibbs equilibrium state as a convex combination ('mixture') of pure thermodynamic states $\langle\cdot\rangle_\alpha$ with weights $P_\alpha$ — a pure state being an extremal Gibbs state whose connected correlation functions decay to zero at large separation (the 'clustering' property) — it defined the overlap of two pure states, $q_{\alpha\beta}=N^{-1}\sum_i m_i^\alpha m_i^\beta$ with $m_i^\alpha=\langle\sigma_i\rangle_\alpha$, the overlap distribution $P_{\mathrm{pure}}(q)=\sum_{\alpha,\beta}P_\alpha P_\beta\delta(q-q_{\alpha\beta})$, and $q(x)$ as the inverse of its cumulative distribution function. It identified $q(x)$ with the replica function $Q(x)$: the endpoint $q(1)$ equals the Edwards-Anderson parameter $q_{\rm EA}$ (the squared local magnetization, common to all pure states), $\int_0^1 q(x)dx=N^{-1}\sum_i\langle\sigma_i\rangle^2$, and the linear-response and total susceptibilities are $\chi_{\rm LR}=\beta[1-q(1)]$ and $\chi=\beta[1-\int_0^1 q(x)dx]$. It further introduced the two-replica generating function $g(y)=\langle\exp[yN^{-1}\sum_i\sigma_i^1\sigma_i^2]\rangle_2$ (two independent replicas sampled from the same Gibbs state), for which $g(y)=\int_0^1dx\,e^{yq(x)}$ under the clustering assumptions; a constant $q(x)$ would signal a single pure state, a non-constant $q(x)$ many coexisting glassy 'valleys'. The paper did not, however, provide a way to obtain the pure-state decomposition itself ('the explicit decomposition of a state into its pure components is not a simple operation'), treated clustering heuristically, stated without full derivation that the overlap of magnetization patterns at two infinitesimally different magnetic fields $h_1,h_2$ obeys $\lim_{h_2\to h_1}q(h_1,h_2)=q(0)$, and left the $g(y)$ measurement program schematic.

Later work split into several strands. On the mean-field side the framework became rigorous: the Ghirlanda-Guerra identities (exact relations that the overlap distribution of any disordered spin system must satisfy) constrain overlap distributions generally; Talagrand proved the Parisi formula, the variational expression for the SK free energy predicted by the RSB solution; Panchenko proved the Parisi ultrametricity conjecture for mixed $p$-spin models (mean-field spin glasses whose Hamiltonian combines random interaction terms of several orders $p$), showing that the overlaps of any three pure states form an ultrametric, i.e. the two smallest of the three pairwise overlap distances coincide. These results give $q(x)$ firm mathematical content in mean field, but say nothing directly about finite-dimensional short-range models. On the short-range side, Newman and Stein showed that the naive reading of the 1983 picture — limiting pure states within a single sample — conflicts with 'chaotic size dependence' (the finite-volume Gibbs state of a disordered system changes chaotically as the system grows), and introduced the metastate, a probability distribution over infinite-volume Gibbs states within a single coupling realization, as the framework in which any many-state (RSB-like) picture must be formulated. Four competing scenarios for the spin-glass phase are currently debated: replica symmetry breaking with a non-trivial metastate over many pure states; the droplet/scaling picture, with only a single pair of pure states related by the global spin flip; the 'TNT' picture, with a trivial two-state ground-state structure but Non-Trivial overlap fluctuations among low-lying excitations; and chaotic pairs, with uncountably many ground-state pairs whose selected member varies chaotically with system size. In parallel, the RSB formalism was extended to glass formers without quenched disorder — structural glasses and dense hard-sphere packings — where there is no random Hamiltonian and where the very existence of many pure thermodynamic states in the infinite-volume limit has not been established. Recent work still proposes competing order parameters for non-mean-field spin glasses and for real glasses, indicating that the question the 1983 construction opened — what, physically, is the order parameter of a glassy phase in a realistic system — has not been closed.

Problem Statement

The 1983 Parisi construction equates the spin-glass order parameter $q(x)$ with the inverse cumulative distribution of overlaps among pure thermodynamic states of the Gibbs measure, but outside the mean-field Sherrington-Kirkpatrick model this identification has never been given a physical realization: the decomposition of the Gibbs state into pure states is not constructive, its clustering assumptions are unjustified, and for systems without quenched disorder the existence of many pure states is unproven. The problem is to settle this state structure, and it has two linked facets sharing one objective — establishing whether the many-pure-state structure encoded by $q(x)$ exists as real thermodynamic structure in systems that are not the mean-field SK model:

(1) Short-range spin glasses. For the Edwards-Anderson Ising spin glass in spatial dimension $d\ge 3$ below its glass transition, determine whether the equilibrium description requires many pure thermodynamic states: whether the infinite-volume Gibbs equilibrium state decomposes into pure states whose overlap distribution is non-trivial and described by a non-constant $q(x)$ — with the clustering assumptions underlying the integral representations of $g(y)$ and of the susceptibility formulas rigorously justified, and the decomposition formulated within the metastate framework wherever chaotic size dependence makes that framework necessary (the many-state structure may then reside in the states drawn from the metastate rather than in the metastate-averaged state) — or whether the overlap distribution among the pure states is trivial — a single pair of pure states related by the global spin flip (as in the droplet/scaling scenario, and state by state in the chaotic-pairs scenario, whose uncountably many pairs are dispersed over the metastate rather than coexisting within a single Gibbs state), or a TNT-type structure in which non-trivial overlap fluctuations arise from low-lying excitations rather than from distinct pure states — so that $q(x)$ is not physically realized as a pure-state order parameter for short-range spin glasses. Throughout, the overlap distribution among pure thermodynamic states (the object of the 1983 construction) must be distinguished from the replica overlap distribution $P(q)$ measured in simulations, which can remain non-trivial even when the pure-state structure is trivial. The competing scenarios for the spin-glass phase differ in the pure-state and overlap structure they predict, so settling this facet distinguishes, or supersedes, them.

(2) Non-random amorphous systems. Beyond random spin Hamiltonians — for non-random amorphous systems of the kind the 1983 paper itself named, such as real glasses and hard spheres at high density — establish whether many distinct pure thermodynamic states exist in the infinite-volume limit at all, and if so identify and characterize the object that plays the role of $q(x)$, thereby placing replica-theoretic descriptions of these systems on the physical footing the 1983 construction claimed for spin glasses. A demonstration that such systems do not possess many distinct pure states in the infinite-volume limit, so that no $q(x)$-type order parameter is definable there, equally resolves this facet. Settling this for hard spheres at high density (or for any other single well-defined non-random amorphous system) resolves this facet; the named cases illustrate the class and do not each require a separate settlement.

The two facets are one problem because both ask the same question of the 1983 construction: does the many-pure-state structure that $q(x)$ encodes exist as actual infinite-volume thermodynamic structure outside mean field? That question is settled for each of the two classes of systems named above — the short-range Edwards-Anderson spin glass, and non-random amorphous systems such as real glasses and hard spheres at high density — by an infinite-volume result that establishes the pure-state structure, constructively with characterized overlap statistics or by refutation; a formal manipulation, a mean-field derivation, or a finite-size numerical observation alone does not settle it.

References

1. G. Parisi, Infinite Number of Order Parameters for Spin-Glasses, Phys. Rev. Lett. 43, 1754 (1979). DOI: 10.1103/PhysRevLett.43.1754, https://doi.org/10.1103/PhysRevLett.43.1754
2. G. Parisi, Order Parameter for Spin-Glasses, Phys. Rev. Lett. 50, 1946 (1983). DOI: 10.1103/PhysRevLett.50.1946, https://doi.org/10.1103/PhysRevLett.50.1946
3. G. Parisi, The order parameter for spin glasses: a function on the interval 0-1, J. Phys. A: Math. Gen. 13, 1101 (1980). DOI: 10.1088/0305-4470/13/3/042, https://doi.org/10.1088/0305-4470/13/3/042
4. A. P. Young, Direct Determination of the Probability Distribution for the Spin-Glass Order Parameter, Phys. Rev. Lett. 51, 1206 (1983). DOI: 10.1103/PhysRevLett.51.1206, https://doi.org/10.1103/PhysRevLett.51.1206
5. N. Parga, G. Parisi, M. A. Virasoro, A numerical investigation of the overlap distribution among pure states in the spin glass phase, J. Phys. Lett. 45, 1063 (1984). DOI: 10.1051/jphyslet:0198400450220106300, https://doi.org/10.1051/jphyslet:0198400450220106300
6. J. D. Reger, R. N. Bhatt, A. P. Young, Monte Carlo study of the order-parameter distribution in the four-dimensional Ising spin glass, Phys. Rev. Lett. 64, 1859 (1990). DOI: 10.1103/PhysRevLett.64.1859, https://doi.org/10.1103/PhysRevLett.64.1859
7. Stefano Ghirlanda, Francesco Guerra, General properties of overlap probability distributions in disordered spin systems. Towards Parisi ultrametricity, J. Phys. A: Math. Gen. 31, 9149 (1998). DOI: 10.1088/0305-4470/31/46/006, https://doi.org/10.1088/0305-4470/31/46/006
8. Michel Talagrand, The Parisi formula, Annals of Mathematics 163, 221 (2006). DOI: 10.4007/annals.2006.163.221, https://doi.org/10.4007/annals.2006.163.221
9. Dmitry Panchenko, The Parisi ultrametricity conjecture, Annals of Mathematics 177, 383 (2013). DOI: 10.4007/annals.2013.177.1.8, https://doi.org/10.4007/annals.2013.177.1.8
10. C. M. Newman, D. L. Stein, Distribution of pure states in short-range spin glasses, Int. J. Mod. Phys. B 24, 2091 (2010). DOI: 10.1142/S0217979210055779, https://doi.org/10.1142/S0217979210055779
11. N. Read, Short-range Ising spin glasses: The metastate interpretation of replica symmetry breaking, Phys. Rev. E 90, 032142 (2014). DOI: 10.1103/PhysRevE.90.032142, arXiv:1407.4136, https://arxiv.org/abs/1407.4136
12. A. Billoire, L. A. Fernandez, A. Maiorano, E. Marinari, V. Martin-Mayor, J. Moreno-Gordo, G. Parisi, F. Ricci-Tersenghi, J. J. Ruiz-Lorenzo, Numerical construction of the Aizenman-Wehr metastate, Phys. Rev. Lett. 119, 037203 (2017). DOI: 10.1103/PhysRevLett.119.037203, https://doi.org/10.1103/PhysRevLett.119.037203
13. C. M. Newman, D. L. Stein, The State(s) of Replica Symmetry Breaking: Mean Field Theories vs. Short-Ranged Spin Glasses (2001). arXiv:cond-mat/0105282, https://arxiv.org/abs/cond-mat/0105282
14. C. M. Newman, N. Read, D. L. Stein, Proof of single-replica equivalence in short-range spin glasses (2022). arXiv:2207.10208, https://arxiv.org/abs/2207.10208
15. C. M. Newman, D. L. Stein, Ground State Stability and the Nature of the Spin Glass Phase (2021). arXiv:2110.11229, https://arxiv.org/abs/2110.11229
16. C. M. Newman, D. L. Stein, Overlap Structure and Free Energy Fluctuations in Short-Range Spin Glasses (2023). arXiv:2306.07132, https://arxiv.org/abs/2306.07132
17. Giorgio Parisi, Francesco Zamponi, Mean-field theory of hard sphere glasses and jamming, Rev. Mod. Phys. 82, 789 (2010). DOI: 10.1103/RevModPhys.82.789, https://doi.org/10.1103/RevModPhys.82.789
18. B. Nachtergaele, L. Slegers, The overlap distribution for a non-random frustrated Ising model, J. Phys. A: Math. Gen. 21, 529 (1988). DOI: 10.1088/0305-4470/21/2/030, https://doi.org/10.1088/0305-4470/21/2/030
19. A. C. D. van Enter, A. Hof, J. Miekisz, Overlap distributions for deterministic systems with many pure states, J. Phys. A: Math. Gen. 25, L1133 (1992). DOI: 10.1088/0305-4470/25/18/007, https://doi.org/10.1088/0305-4470/25/18/007
20. Miguel Aguilar-Janita, Silvio Franz, Victor Martin-Mayor, Javier Moreno-Gordo, Giorgio Parisi, Federico Ricci-Tersenghi, Juan J. Ruiz-Lorenzo, Small field chaos in spin glasses: Universal predictions from the ultrametric tree and comparison with numerical simulations, Proc. Natl. Acad. Sci. USA 121, e2404973121 (2024). DOI: 10.1073/pnas.2404973121, https://doi.org/10.1073/pnas.2404973121
21. Michele Castellana, Order parameter for non-mean-field spin glasses, Phys. Rev. B 112, 224206 (2025). DOI: 10.1103/7lm2-ljrd, arXiv:2512.08691, https://arxiv.org/abs/2512.08691
22. Koun Shirai, Nature of the Order Parameters of Glass, Foundations 5, 9 (2025). DOI: 10.3390/foundations5010009, https://doi.org/10.3390/foundations5010009