Low-temperature phase of the D-to-infinity frustrated XY (Josephson-junction) hypercube model: Parisi's moment-matching random-ensemble equivalence conjecture

The source paper (G. Parisi, 'A mean field theory for arrays of Josephson junctions', J. Math. Phys. 37, 5158 (1996)) solves the deterministic frustrated XY model on a single D-dimensional hypercubic cell exactly at D to infinity in the high-temperature phase, via a mapping of the high-temperature-expansion coefficients to vacuum matrix elements of powers of the position operator of the q-deformed harmonic oscillator (equivalently, moments of the Q-Hermite spectral density). Below the transition temperature the spherical approximation fails, and the paper extends the solution only through an unproven conjecture: that the deterministic model is thermodynamically equivalent, in the low-temperature phase, to a random ensemble whose coupling matrix is a Haar-random unitary conjugate of the deterministic spectrum (moment matching), solved within a replica-symmetric ansatz. The spectral problem has since been treated in the double-scaled-SYK/holography literature (refs 10-11). The low-temperature thermodynamics of the deterministic XY model itself remains open: the moment-matching equivalence is neither proved nor refuted, no replica-symmetry-breaking analysis of the moment-matched ensemble exists, the low-temperature 1/D corrections have never been computed, and finite-D numerics (D up to 16) suggest a small genuine discrepancy for positive q that cannot be decided without them.

Background

An array of Josephson junctions in a magnetic field is described, at the level of its superconducting phases, by an XY model with frustrated couplings: each site j carries a phase $\varphi_j$ with $|\varphi_j|=1$ (a point on the unit circle, hence 'XY', i.e. two-component spins), and each nearest-neighbour link carries a complex coupling whose phase reproduces the Aharonov-Bohm flux accumulated around plaquettes of the lattice (a plaquette is an elementary square face). When the flux per plaquette is a fixed angle $\theta$, the model is deterministic but frustrated; when the flux is random one obtains the gauge glass, an XY spin glass with random vector potentials.

In the mid-1990s Parisi and coworkers studied the deterministic frustrated XY model on a single D-dimensional hypercubic cell of side 2, i.e. one hypercube with $2^D$ sites, with Hamiltonian

$$\beta H=-\beta\,\mathrm{Re}\left[(2D)^{-1/2}\sum_{j,k}\varphi_j^{*}\,U_{jk}\,\varphi_k\right],\qquad |\varphi_j|=1,$$

where the couplings $U_{jk}$ are fixed complex phases built from a plaquette-flux tensor $\theta_{\alpha,\beta}=S_{\alpha,\beta}\,\theta$, with $S_{\alpha,\beta}=\pm1$ an antisymmetric sign tensor: for fixed $S$ the couplings are unique up to gauge transformations, but for $D>3$ different choices of $S$ are not equivalent (all choices are equivalent only in $D=2,3$), and the source defines the model with a generic $S$, i.e. with the signs picked at random (the explicit construction is given in the Problem Statement). The frustration parameter is $q=\cos\theta$ ($q=1$ ferromagnet, $q=-1$ fully frustrated, $q=0$ with the same diagrammatics as a spin glass). The normalization $(2D)^{-1/2}$ gives a nontrivial limit $D\to\infty$, which is the mean-field (infinite-dimensional) limit of the array.

In the high-temperature phase this limit is exactly solvable. The coefficients of the high-temperature expansion are traces of powers of the weighted adjacency matrix (the 'Laplacian') $\Delta$, and at $D\to\infty$ one has $2^{-D}\mathrm{Tr}\,\Delta^{2n}\to\langle 0|X^{2n}|0\rangle$, where $X=R_q+L_q$ is the position operator of the q-deformed harmonic oscillator, defined by creation and annihilation operators obeying the deformed commutation relation $L_qR_q-qR_qL_q=1$ with vacuum $L_q|0\rangle=0$. These matrix elements are the moments of the Q-Hermite (q-Gaussian) spectral density; diagrammatically they count closed walks on the hypercube weighted by $q^{\mathrm{area}}$, i.e. q-weighted chord diagrams. The spherical model (replacing the local constraint $|\varphi_j|=1$ by one global constraint on $\sum_j|\varphi_j|^2$) is exact in the high-temperature phase and at the transition temperature $T_c(q)$, but breaks down for $T<T_c(q)$.

The only available handle on the low-temperature phase is a conjecture, stated by Marinari, Parisi and Ritort and consolidated in the source paper: replace the deterministic coupling matrix by a random one, $\tilde U=V^{\dagger}\Lambda V$, where $V$ is a Haar-distributed random unitary matrix (Haar meaning uniform with respect to the invariant measure on the unitary group) and $\Lambda$ is a fixed diagonal matrix whose eigenvalues are sampled so that all spectral moments of $\tilde U$ match those of the deterministic Laplacian. This 'moment-matched' random ensemble is then solved by the replica method - the standard spin-glass technique in which $n$ identical copies of the system are introduced to average the quenched free energy and the limit $n\to 0$ is taken - under a replica-symmetric (RS) ansatz, i.e. assuming all copies are equivalent; replica-symmetry breaking (RSB) is the refinement required when that ansatz is unstable.

Numerical simulations of the deterministic model for $D=3$ to $16$ show excellent agreement with the RS random-model predictions in the low-temperature phase for $q\le 0$ (down to the nearly fully frustrated regime), but for positive $q$ a small and growing discrepancy appears: at $q\approx 0.23$ the Monte Carlo energy at $D=16$ already falls below the random-model spherical value that is expected to bound it from below, which the authors read as evidence that the two models 'do indeed differ, even if only of a small amount'; near $T_c$ the finite-D effects are very strong, so the authors could not decide whether the random-model results are exact with peculiar finite-D corrections or merely a good approximation. At the endpoint $q=-1$ the XY hypercube shows no transition at all (always paramagnetic), while the Ising version on the same geometry is glassy, ages (exhibits history-dependent slow relaxation), and is well described by the random orthogonal model - a random-matrix spin glass - showing that the deterministic-random correspondence is at least partially valid for discrete spins while the continuous-spin (Josephson) case remained undecided. The spherical version of the model was solved exactly shortly afterwards using q-Hermite polynomials: the $D\to\infty$ spectral density is a closed-form (theta-function, q-Gaussian) expression, and the spherical low-temperature phase is not glassy and depends smoothly on $q$, so any glassiness of the XY model must originate in the local compactness constraint $|\varphi_j|=1$. For the neighboring deterministic family whose couplings are quantizations of torus maps, the glassy behaviour of the corresponding random models was later shown to disappear in the deterministic spherical and continuous-XY versions. Finally, since 2020 the same hypercube spectral problem has reappeared, as a quantum single-particle problem, in the literature on the double-scaled Sachdev-Ye-Kitaev (DS-SYK) model - a solvable model of quantum chaos and holography - and on near-$\mathrm{AdS}_2$ holography, where the leading-$1/d$ spectral density is again the Q-Hermite density and the subleading moments are q-weighted chord diagrams. What has never been settled is the classical low-temperature thermodynamics of the deterministic XY model itself.

Problem Statement

Consider the deterministic frustrated XY model on a single D-dimensional hypercubic cell of side 2 ($N=2^D$ sites) with Hamiltonian

$$\beta H=-\beta\,\mathrm{Re}\left[(2D)^{-1/2}\sum_{j,k}\varphi_j^{*}\,U_{jk}\,\varphi_k\right],\qquad |\varphi_j|=1,$$

where the couplings $U_{jk}$ are fixed complex phases constructed as follows. The flux through each plaquette oriented in the $(\alpha,\beta)$ coordinate plane is $\theta_{\alpha,\beta}=S_{\alpha,\beta}\,\theta$, where $S_{\alpha,\beta}=\pm1$ is an antisymmetric sign tensor; for fixed $S$ the couplings are unique up to gauge transformations and are given explicitly by

$$U_1(j)=1,\qquad U_\mu(j)=\exp\!\left(i\theta\sum_{\nu<\mu}S_{\mu,\nu}\,j_\nu\right),\qquad \mu=2,\dots,D,$$

with site coordinates $j_\nu\in\{0,1\}$. For $D>3$ different choices of $S$ are not gauge-equivalent (all choices are equivalent only in $D=2,3$), so the model is defined, as in the source, with a generic $S$ whose $\pm1$ signs are picked at random; the deterministic model is understood as one generic (typical) realization of $S$ (in the source's simulations the fluctuations among $S$ realizations were checked to be small). The frustration parameter is $q=\cos\theta$ with $-1\le q\le 1$ ($q=1$ ferromagnet, $q=-1$ fully frustrated). The high-temperature phase of this model is exactly solvable at $D\to\infty$ through the q-deformed-oscillator mapping described in the Background. The problem is to determine the low-temperature phase $T<T_c(q)$ of this deterministic model in the limit $D\to\infty$, as follows.

1. (Main question) Prove or refute the moment-matching equivalence conjecture of Parisi (1996) and Marinari-Parisi-Ritort (1995): at $D=\infty$ and $T<T_c(q)$, for generic flux values $-1<q<1$, the deterministic model and the random ensemble in which $U$ is replaced by $\tilde U=V^{\dagger}\Lambda V$ ($V$ Haar-distributed unitary; the spectrum of $\Lambda$ fixed by matching all spectral moments of the deterministic Laplacian) have identical thermodynamics - free energy, internal energy, specific heat, and order-parameter structure - including whether the transition at $T_c(q)$ lands the deterministic model in the glassy phase predicted by the replica solution of the random ensemble. Here $T_c(q)$ is the $D\to\infty$ transition temperature of the spherical solution ($\beta_c=(1-q)^{1/2}/2$ in the normalization of Parisi (1994), with critical behavior independent of $q$). The endpoints are source-discussed special cases outside the target: at $q=1$ the model is the exactly known ferromagnet, and at $q=-1$ the $D\to\infty$ limit of $\beta_c$ is discontinuous and the companion simulations report no transition in the XY hypercube. The question is sharpest for not-small positive $q$, where the simulations leave the equivalence undecided, while agreement is excellent for $q\le 0$ and small positive $q$.
2. Determine whether the replica-symmetric solution of the moment-matched random ensemble is stable at $T<T_c(q)$, and the effect of replica-symmetry breaking on its low-temperature predictions for the deterministic model.
3. Compute the $1/D$ corrections to the low-temperature thermodynamics with enough control to decide whether the discrepancies observed in numerical simulations at positive $q$ are finite-$D$ effects. (The high-temperature $1/D$ coefficients are known; the low-temperature expansion has never been computed.)

Item 1 is the objective of the task: a controlled characterization of the low-temperature phase of the infinite-dimensional deterministic frustrated Josephson-junction array through the equivalence conjecture. Items 2 and 3 are supporting facets of the item-1 target. The question is posed for the generic case $T<T_c(q)$ with $-1<q<1$, not for the endpoints $q=\pm1$ or the spherical or Gaussian local-constraint variants of the model.

References

1. G. Parisi, 'A mean field theory for arrays of Josephson junctions', Journal of Mathematical Physics 37, 5158 (1996). DOI: 10.1063/1.531678, https://doi.org/10.1063/1.531678
2. G. Parisi, 'D-dimensional arrays of Josephson junctions, spin glasses and q-deformed harmonic oscillators', Journal of Physics A: Mathematical and General 27, 7555 (1994). DOI: 10.1088/0305-4470/27/23/007, arXiv:cond-mat/9410088, https://arxiv.org/abs/cond-mat/9410088
3. E. Marinari, G. Parisi, F. Ritort, 'Replica theory and large-D Josephson junction hypercubic models', Journal of Physics A: Mathematical and General 28, 4481 (1995). DOI: 10.1088/0305-4470/28/16/008, arXiv:cond-mat/9502067, https://arxiv.org/abs/cond-mat/9502067
4. E. Marinari, G. Parisi, F. Ritort, 'The fully frustrated hypercubic model is glassy and aging at large D', Journal of Physics A: Mathematical and General 28, 327 (1995). DOI: 10.1088/0305-4470/28/2/010, arXiv:cond-mat/9410089, https://arxiv.org/abs/cond-mat/9410089
5. E. Marinari, G. Parisi, F. Ritort, 'Replica field theory for deterministic models: I. Binary sequences with low autocorrelation', Journal of Physics A: Mathematical and General 27, 7615 (1994). DOI: 10.1088/0305-4470/27/23/010, arXiv:hep-th/9405148, https://arxiv.org/abs/hep-th/9405148
6. E. Marinari, G. Parisi, F. Ritort, 'Replica field theory for deterministic models. II. A non-random spin glass with glassy behaviour', Journal of Physics A: Mathematical and General 27, 7647 (1994). DOI: 10.1088/0305-4470/27/23/011, arXiv:cond-mat/9406074, https://arxiv.org/abs/cond-mat/9406074
7. G. Parisi, M. Potters, 'Mean-field equations for spin models with orthogonal interaction matrices', Journal of Physics A: Mathematical and General 28, 5267 (1995). DOI: 10.1088/0305-4470/28/18/016, arXiv:cond-mat/9503009, https://arxiv.org/abs/cond-mat/9503009
8. A. Cappelli, F. Colomo, 'Solving the frustrated spherical model with q-polynomials', Journal of Physics A: Mathematical and General 31, 3141 (1998). DOI: 10.1088/0305-4470/31/14/004, arXiv:hep-th/9710071, https://arxiv.org/abs/hep-th/9710071
9. I. Borsari, F. Camia, S. Graffi, F. Unguendoli, 'Absence of glassy behaviour in the deterministic spherical and XY models', Journal of Physics A: Mathematical and General 31, 1127 (1998). DOI: 10.1088/0305-4470/31/4/004, https://doi.org/10.1088/0305-4470/31/4/004
10. Y. Jia, J. J. M. Verbaarschot, 'Chaos on the hypercube', Journal of High Energy Physics 2020, 154 (2020). DOI: 10.1007/JHEP11(2020)154, arXiv:2005.13017, https://arxiv.org/abs/2005.13017
11. M. Berkooz, Y. Jia, N. Silberstein, 'Parisi's hypercube, Fock-space frustration and near-AdS$_2$/near-CFT$_1$ holography', Physical Review Letters 132, 081601 (2024). DOI: 10.1103/PhysRevLett.132.081601, arXiv:2303.18182, https://arxiv.org/abs/2303.18182
12. A. Mauri, M. I. Katsnelson, 'Frustrated magnets in the limit of infinite dimensions: dynamics and disorder-free glass transition', Physical Review B 109, 144414 (2024). DOI: 10.1103/PhysRevB.109.144414, arXiv:2311.09124, https://arxiv.org/abs/2311.09124