Glassy Potts model: completing the characterization of the finite-dimensional random-permutation Potts glass — critical behavior, replica-symmetry-breaking structure, low-temperature phase, universality, and aging dynamics

The glassy (random-permutation) Potts model is a four-state Potts spin system on a lattice whose bonds carry quenched random permutations of the spin states; an exact local gauge symmetry forbids ferromagnetic order at every temperature, so any ordered phase is necessarily glassy. Marinari, Mossa, and Parisi (Phys. Rev. B 59, 8401 (1999)) introduced it in four dimensions and found, from small lattices, a transition to a glassy phase with a one-step-RSB (Random-Energy-Model-like) overlap distribution and qualitative aging; mean-field theory (Carlucci, Phys. Rev. B 60, 9862 (1999)) predicts a discontinuous one-step-RSB transition for q = 4. Later large-scale FPGA simulations (Fernández et al., Phys. Rev. B 77, 104432 (2008)) confirmed the four-dimensional transition with preliminary exponents and found in three dimensions correlation-length curves that merge rather than cross, but the central questions remain unanswered. The following core is open: (i) the replica-symmetry-breaking structure of the glassy phase in finite dimensions (one-step versus full RSB) and its reconciliation with the apparent continuous transition; (ii) equilibrium properties deep in the low-temperature glassy phase, since equilibrated data stop only slightly below $T_c$; (iii) the three-dimensional transition scenario (Kosterlitz-Thouless-like versus avoided versus transient crossover) and the lower critical dimension; (iv) universality among the two permutation variants and the standard four-state Potts glass; (v) quantitative aging laws and their relation to the static glassy order; (vi) definitive critical exponents with controlled scaling corrections, resolving the present inconsistency between susceptibility-exponent estimates (1.3-1.5 versus 2.5(4)).

Background

The $q$-state Potts model places on each site $i$ of a lattice a discrete 'color' $\sigma_i \in \{0, 1, \ldots, q-1\}$, with an energy that favors equal colors on neighboring sites. If the interactions are quenched, random, and frustrated (pairwise preferences that cannot be simultaneously satisfied), one obtains a Potts spin glass: a disordered system whose low-temperature physics is governed by a rugged energy landscape rather than by conventional order. Spin glasses are diagnosed through the overlap between two independent equilibrium copies ('replicas') of the same disordered sample — here a modified overlap $\omega$ defined below — and the shape of its probability distribution $P(\omega)$ classifies the phase. A single peak means replica symmetry (paramagnet); two peaks separated by a gap mean one-step replica symmetry breaking (1RSB), the pattern of Derrida's Random Energy Model (REM); a continuous band of peaks means full replica symmetry breaking (full RSB) in the Parisi sense. Whether the one-step, mean-field-like scenario of fragile glass formers survives in finite dimensions is a central open issue of glass physics.

The glassy (random-permutation) Potts model was introduced by Marinari, Mossa, and Parisi [1] as a clean finite-dimensional laboratory for glassiness. Its Hamiltonian is

$$H = -\sum_{\langle i,j\rangle} \delta\bigl(\sigma_i, \Pi_{ij}(\sigma_j)\bigr),$$

where $\delta$ is the Kronecker delta, the sum runs over nearest-neighbor bonds, and each bond carries an independent quenched random permutation $\Pi_{ij}$ of the four states, so that a bond is satisfied only if $\sigma_i$ equals the permuted color of $\sigma_j$. The model has an exact local gauge symmetry: relabeling the colors site by site, $\sigma_i \to \Gamma_i(\sigma_i)$, maps the Hamiltonian to the same form with conjugated disorder $\Gamma_i \Pi_{ij} \Gamma_j^{-1}$, and since the disorder average restores the uniform distribution over permutations, gauge-variant quantities such as the magnetization vanish identically at every temperature. Consequently any ordered phase can only be glassy, and a glassy phase, once formed, extends down to $T = 0$ without being preempted by ferromagnetism — the property for which the source paper is titled 'a disordered Potts model without a ferromagnetic phase'. Two variants were studied: $M_{q!}$, with $\Pi_{ij}$ drawn uniformly from all $4! = 24$ permutations, and $M_{co}$, restricted to the eight permutations that commute with the involution $R = (02)(13)$ (which exchanges 0 with 2 and 1 with 3), i.e. the centralizer of $R$ in $S_4$. For $M_{co}$ one uses a modified overlap between two replicas: $\omega_i = +1$ if the two colors at site $i$ are equal, $\omega_i = -1$ if they form one of the cross-coupled pairs $(0,2)$, $(1,3)$, $(2,0)$, $(3,1)$ (pairs related by $R$), and $\omega_i = 0$ otherwise. By construction $P(\omega)$ is then symmetric under $\omega \to -\omega$ at thermal equilibrium, which provides a stringent equilibration check — the analogue of the $\mathbb{Z}_2$ symmetry test in Ising spin glasses, and the main reason $M_{co}$ was introduced. For $M_{q!}$ one instead uses the usual overlap: the site-average of the indicator that the two replicas carry the same color. Standard diagnostics are the Binder parameter $g_\omega = \frac{1}{2}\left(3 - \langle\omega^4\rangle/\langle\omega^2\rangle^2\right)$ (a dimensionless cumulant of the site-averaged overlap $\omega$), the RSB parameter $\rho_\omega$ of Marinari et al. [6] (based on exact identities of Guerra; it compares fluctuations of the overlap within and between real replicas, testing whether the glassy order parameter self-averages, i.e. whether replica symmetry is broken), and the correlation length $\xi$.

The transition is located by finite-size scaling (FSS): near a continuous transition, dimensionless quantities such as $g_\omega$ or the ratio $\xi/L$ become size-independent at the critical temperature, so curves for different lattice sizes $L$ cross there; the drift of the crossings yields the correlation-length exponent $\nu$, the susceptibility scales as $\chi_\omega \sim L^{2-\eta}$, and $\gamma = \nu(2-\eta)$. If instead the transition is of Kosterlitz-Thouless (KT) type — a topological transition, as in the two-dimensional XY model, with an exponentially divergent correlation length — the $\xi/L$ curves merge smoothly rather than crossing, which is difficult to distinguish from a mere crossover or from transient finite-size behavior. Below the transition, equilibration times diverge and simulations age: the two-time autocorrelation $C(t + t_w, t_w)$ depends on the waiting time $t_w$ (the age of the sample), not on $t$ alone.

The mean-field benchmark comes from the fully connected limit: Nishimori and Stephen [3] introduced the gauge-invariant frustrated Potts spin glass, and Carlucci [2] showed that the random-permutation model can be rewritten, in a chiral (root-of-unity) representation, as that model with discrete couplings, whose mean-field free energy for $q = 4$ has a discontinuous one-step-RSB transition from the paramagnet to the glass at least above the upper critical dimension. (The chiral rewriting does not literally realize every element of the full permutation group $S_4$, so the exact reach of this equivalence is itself part of the universality question.) Finite dimensions must then be settled by simulation: [1] studied $q = 4$ in $d = 4$ on lattices $L = 4, 5, 8$ with ten disorder samples, locating $T_c$ between 1.4 and 1.5 and the susceptibility exponent $\gamma$ between 1.3 and 1.5 (the four-dimensional Ising spin glass has $\gamma \approx 2.10$), and observing a 'golf-course' landscape (deep minima with small basins of attraction), a $P(\omega)$ reminiscent of the REM, and aging. The large-scale continuation [4], run on FPGA-based specialized hardware, reached $L \le 16$ in $d = 4$ with $10^3$–$2\times10^3$ disorder samples, giving $\beta_c = 1.41(1)$ (in a convention with a factor of two in the Hamiltonian normalization relative to [1]), $\nu = 1.1(2)$, $\eta = -0.31(3)$, $\gamma = 2.5(4)$, with scaling corrections that could not be resolved at these sizes; it also performed the first $d = 3$ study ($L \le 16$), finding merging $\xi/L$ curves consistent with a KT transition but unable to exclude transient effects from a lower critical dimension slightly below 3. In $d = 3$, the comparison study [5] of the standard four-state Potts glass found no clear-cut transition in the permutation variant and left the universality question explicitly open. On the theory side, the Bethe-approximation analysis of Cammarota et al. [11] suggests that a discontinuous (mean-field-like) transition for $q = 4$ requires coordination $c \ge 12$, i.e. $d \ge 6$ on a hypercubic lattice, while a rigorous result in the same work excludes any phase transition only in the regime $q > 2c$ ($c$ the lattice coordination), which does not cover $q = 4$; and the Migdal-Kadanoff real-space renormalization-group study of Angelini and Biroli [8, 9], whose 'M-class' of disordered glassy models includes the random-permutation Potts glasses, predicts a lower critical dimension above 3 and an avoided three-dimensional transition, but by an uncontrolled approximation.

Problem Statement

Complete the characterization of the glassy phase of the four-state random-permutation (glassy) Potts model on hypercubic lattices — the model introduced in [1], with Hamiltonian

$$H = -\sum_{\langle i,j\rangle} \delta\bigl(\sigma_i, \Pi_{ij}(\sigma_j)\bigr),$$

where $\delta$ is the Kronecker delta, the sum runs over nearest-neighbor bonds, each site carries a 'color' $\sigma_i \in \{0,1,2,3\}$, and each bond carries an independent quenched random permutation $\Pi_{ij}$ of the four colors, so that a bond is satisfied only if $\sigma_i$ equals the permuted color of $\sigma_j$. Study both the full-permutation variant $M_{q!}$ ($\Pi_{ij}$ drawn uniformly from all $4! = 24$ permutations) and the commutative variant $M_{co}$ ($\Pi_{ij}$ drawn uniformly from the eight permutations that commute with the involution $R = (02)(13)$ exchanging color 0 with 2 and 1 with 3, i.e. the centralizer of $R$ in $S_4$). Throughout, $\omega$ denotes the overlap between two independent equilibrium copies ('replicas') of the same disordered sample — for $M_{q!}$ the site-average of the indicator that the two replicas carry the same color; for $M_{co}$ the site-average of $\omega_i = +1$ if the two colors coincide, $-1$ if they are exchanged by $R$, and $0$ otherwise — and $P(\omega)$ is its probability distribution in the thermodynamic limit. Concretely, for the primary case of spatial dimension $d = 4$, and for the secondary case $d = 3$ opened by the direct continuation [4]:

1. (Critical behavior) Determine the transition temperature and the critical exponents — the correlation-length exponent $\nu$, the anomalous-dimension exponent $\eta$, and the susceptibility exponent $\gamma = \nu(2-\eta)$ — with controlled correction-to-scaling, settling the present inconsistency between the warm-phase estimate $\gamma = 1.3$–$1.5$ of [1] and the large-scale estimate $\gamma = 2.5(4)$ of [4] (the two works use Hamiltonian normalizations differing by a factor of two; the comparison must fix the convention explicitly).

2. (RSB structure) Determine the replica-symmetry-breaking pattern of the glassy phase in the thermodynamic limit: is the overlap distribution $P(\omega)$ one-step-RSB-like (two peaks separated by a gap, discontinuous order parameter, as the mean-field solution of [2, 3] predicts for $q = 4$), full-RSB-like (a continuous Parisi-type distribution), or otherwise? Reconcile the answer with the apparently continuous $d = 4$ transition found in [4].

3. (Low-temperature equilibrium) Establish the equilibrium behavior of the glassy phase below the temperatures equilibrated to date (current equilibrated data stop only modestly below the critical temperature $T_c$): the thermodynamic-limit overlap distribution, the replica-symmetry-breaking parameter $\rho_\omega$ of Marinari et al. [6] (a dimensionless estimator, based on exact sum rules, that is nonzero precisely when the overlap fails to self-average), the stiffness and scaling of the order parameter, and the approach to $T = 0$.

4. (Universality) Decide whether $M_{q!}$, $M_{co}$, and the standard four-state Potts glass with binary couplings [5] belong to one universality class, as conjectured in [1] and assumed in [4] but left explicitly open by [5].

5. (Three-dimensional scenario) Determine the nature of the three-dimensional transition: Kosterlitz–Thouless-like (a topological transition with exponentially divergent correlation length, as in the two-dimensional XY model), avoided (no thermodynamic transition, as the Migdal-Kadanoff analysis of [8, 9] predicts for the class of disordered glassy models it treats), or a transient finite-size crossover; and locate the lower critical dimension of the model.

6. (Aging dynamics) Quantify the off-equilibrium dynamics below $T_c$: the scaling form of the two-time autocorrelation $C(t + t_w, t_w)$ (the correlation between the configuration at waiting time $t_w$ and at the later time $t + t_w$, which fails to collapse onto a function of $t$ alone when the system ages), the waiting-time exponents, and the connection between aging and the static replica-symmetry-breaking structure.

These six targets are facets of a single objective — the complete equilibrium and dynamical characterization of the finite-dimensional random-permutation Potts glass — as posed by the source paper [1] and its direct continuation [4, 5], and no facet has been closed by the literature to date. $d = 4$ is the primary setting and $d = 3$ the secondary case opened by [4], and both dimensions are in scope.

References

1. E. Marinari, S. Mossa, G. Parisi, 'Glassy Potts model: A disordered Potts model without a ferromagnetic phase', Physical Review B 59, 8401-8404 (1999). DOI: 10.1103/PhysRevB.59.8401, https://doi.org/10.1103/PhysRevB.59.8401; arXiv: cond-mat/9805300, https://arxiv.org/abs/cond-mat/9805300.
2. D. M. Carlucci, 'Comment on "Glassy Potts model: A disordered Potts model without a ferromagnetic phase"', Physical Review B 60, 9862-9863 (1999). DOI: 10.1103/PhysRevB.60.9862, https://doi.org/10.1103/PhysRevB.60.9862; arXiv: cond-mat/9807271, https://arxiv.org/abs/cond-mat/9807271.
3. H. Nishimori, M. J. Stephen, 'Gauge-invariant frustrated Potts spin-glass', Physical Review B 27, 5644-5652 (1983). DOI: 10.1103/PhysRevB.27.5644, https://doi.org/10.1103/PhysRevB.27.5644.
4. L. A. Fernández, A. Maiorano, E. Marinari, V. Martin-Mayor, D. Navarro, D. Sciretti, A. Tarancón, J. L. Velasco, 'Critical properties of the four-state commutative random permutation glassy Potts model in three and four dimensions', Physical Review B 77, 104432 (2008). DOI: 10.1103/PhysRevB.77.104432, https://doi.org/10.1103/PhysRevB.77.104432; arXiv: 0710.4246, https://arxiv.org/abs/0710.4246.
5. A. Cruz, L. A. Fernandez, A. Gordillo-Guerrero, M. Guidetti, A. Maiorano, F. Mantovani, E. Marinari, V. Martin-Mayor, A. Muñoz Sudupe, D. Navarro, G. Parisi, S. Perez-Gaviro, J. J. Ruiz-Lorenzo, S. F. Schifano, D. Sciretti, A. Tarancon, R. Tripiccione, J. L. Velasco, D. Yllanes, A. P. Young, 'Spin glass phase in the four-state three-dimensional Potts model', Physical Review B 79, 184408 (2009). DOI: 10.1103/PhysRevB.79.184408, https://doi.org/10.1103/PhysRevB.79.184408; arXiv: 0812.1287, https://arxiv.org/abs/0812.1287.
6. E. Marinari, C. Naitza, F. Zuliani, G. Parisi, M. Picco, F. Ritort, 'General Method to Determine Replica Symmetry Breaking Transitions', Physical Review Letters 81, 1698-1701 (1998). DOI: 10.1103/PhysRevLett.81.1698, https://doi.org/10.1103/PhysRevLett.81.1698; arXiv: cond-mat/9802309, https://arxiv.org/abs/cond-mat/9802309.
7. R. Alvarez Baños, A. Cruz, L. A. Fernandez, A. Gordillo-Guerrero, J. M. Gil-Narvion, 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, B. Seoane, S. F. Schifano, A. Tarancon, R. Tripiccione, D. Yllanes, 'Critical behavior of three-dimensional disordered Potts models with many states', Journal of Statistical Mechanics: Theory and Experiment (2010) P05002. DOI: 10.1088/1742-5468/2010/05/P05002, https://doi.org/10.1088/1742-5468/2010/05/P05002; arXiv: 1002.4288, https://arxiv.org/abs/1002.4288.
8. M. C. Angelini, G. Biroli, 'Real space renormalization group theory of disordered models of glasses', Proceedings of the National Academy of Sciences USA 114, 3328-3333 (2017). DOI: 10.1073/pnas.1613126114, https://doi.org/10.1073/pnas.1613126114; arXiv: 1604.03717, https://arxiv.org/abs/1604.03717.
9. M. C. Angelini, G. Biroli, 'Real Space Migdal-Kadanoff Renormalisation of Glassy Systems: Recent Results and a Critical Assessment', Journal of Statistical Physics 167, 476-498 (2017). DOI: 10.1007/s10955-017-1748-4, https://doi.org/10.1007/s10955-017-1748-4; arXiv: 1702.03092, https://arxiv.org/abs/1702.03092.
10. C. Brangian, W. Kob, K. Binder, 'Statics and dynamics of the ten-state mean-field Potts glass model: a Monte Carlo study', Journal of Physics A: Mathematical and General 35, 191-216 (2002). DOI: 10.1088/0305-4470/35/2/302, https://doi.org/10.1088/0305-4470/35/2/302; arXiv: cond-mat/0106314, https://arxiv.org/abs/cond-mat/0106314.
11. C. Cammarota, G. Biroli, M. Tarzia, G. Tarjus, 'Fragility of the mean-field scenario of structural glasses for disordered spin models in finite dimensions', Physical Review B 87, 064202 (2013). DOI: 10.1103/PhysRevB.87.064202, https://doi.org/10.1103/PhysRevB.87.064202; arXiv: 1210.2941, https://arxiv.org/abs/1210.2941. (The arXiv preprint, arXiv:1210.2941, carries the prepublication title 'On the fragility of the mean-field scenario of structural glasses for finite-dimensional disordered spin models'.)
12. C. Yamaguchi, 'A curious relationship between Potts glass models', The European Physical Journal B 88, 195 (2015). DOI: 10.1140/epjb/e2015-60068-7, https://doi.org/10.1140/epjb/e2015-60068-7; arXiv: 1412.8051, https://arxiv.org/abs/1412.8051.