A multiscale cavity method for sublinear-rank symmetric matrix factorization

2026-03-19T15:57:49Z

We consider a statistical model for symmetric matrix factorization with additive Gaussian noise in the high-dimensional regime, where the rank of the signal matrix to infer $M$ scales with its size $N$ as $M=\mathrm{o}(\sqrt{\ln N})$. Allowing for an $N$-dependent rank offers new challenges and requires new methods. Working in the Bayes-optimal setting, we show that whenever the signal has i.i.d. entries, the limiting mutual information between signal and data is given by a variational formula involving a rank-one replica symmetric potential. In other words, from the information-theoretic perspective, the case of a (slowly) growing rank is the same as when $M=1$ (namely, the standard spiked Wigner model). The proof is primarily based on a novel multiscale cavity method allowing for growing rank along with some information-theoretic identities on worst noise for the vector Gaussian channel. We believe that the cavity method developed here will play a role in the analysis of a broader class of inference and spin models where the degrees of freedom are large arrays instead of vectors.

A microscopic theory of Anderson localization of electrons in random lattices

2026-03-19T08:40:40Z

The existence of Anderson localization, characterized by vanishing diffusion due to strong randomness, has been demonstrated in numerous ways. A systematic approach based on the Anderson quantum model of the Fermi gas in random lattices that can describe both diffusive and localized regimes has not yet been fully established. We build upon a recent publication \cite{Janis:2025ab} and present a microscopic theory of disordered electrons covering both the metallic phase with extended Bloch waves and the localized phase where the propagating particle forms a quantum bound state with the hole left behind at the origin. The general theory provides a framework for constructing controlled approximations to one-particle and two-particle Green functions that satisfy the necessary conservation laws and causality requirements in the whole range of disorder strength. It is used explicitly to derive a local, mean-field-like approximation for the two-particle irreducible vertices, enabling quantitative analysis of the solution's properties in both metallic and localized phases, including critical behavior at the Anderson localization transition.

Phase Transitions in a Modified Ising Spin Glass Model: A Tensor-Network-based Sampling Approach

2026-03-19T04:43:04Z

Phase transitions in a modified Nishimori model, including the model considered by Kitatani, on a two-dimensional square lattice are investigated using a tensor-network-based sampling scheme. In this model, generating bond configurations is computationally demanding because of the correlated random interactions. The employed sampling method enables hierarchical and independent sampling of both bonds and spins. This approach allows high-precision calculations for system sizes up to $L=256$. The results provide clear numerical evidence that the spin-glass and ferromagnetic transitions are separated on the Nishimori line, supporting the existence of an intermediate Mattis-like spin-glass phase. This finding is consistent with the reentrant transition numerically observed in the two-dimensional Edwards-Anderson (EA) model. Furthermore, critical exponents estimated via finite-size-scaling analysis indicate that the universality class of the transitions differs from that of the standard independent and identically distributed EA model.

Robust Near-Critical Dynamics in Heavy-Tailed Neural Networks

2026-03-19T04:21:42Z

The criticality hypothesis posits that biological neural networks operate near a phase transition, yet within standard Gaussian mean-field theories this regime appears fragile and requires fine tuning. Here we show that heavy-tailed synaptic connectivity provides a robust alternative mechanism. By developing a dynamical mean-field theory for Cauchy-distributed couplings, we reduce the macroscopic dynamics to a one-dimensional gradient flow with a global Lyapunov potential. The resulting theory exhibits a continuous phase transition in which collective activity grows with the square root of the distance to criticality, and static susceptibility diverges only as the square root rather than linearly as in Gaussian mean-field theories. This structure gives rise to an emergent automatic gain control: activity-dependent noise fluctuations suppress the effective gain at high activity levels while preserving high susceptibility near the critical point. Extending this mechanism to general symmetric $α$-stable inputs, we identify heavy-tailed synapses as a key microscopic origin of robust near-critical dynamics in disordered neural circuits.

Asymmetric Energy Landscapes Control Diffusion in Glasses

2026-03-18T21:57:41Z

While diffusion in crystalline solids is quantitatively understood through defect-mediated atomic hops, no comparable quantitative framework exists for glasses. In these systems, the origin of large diffusion activation energies remains puzzling, despite local rearrangements involving low barriers. Using molecular dynamics simulations of metallic glasses, we decompose diffusion into random-walk and correlation contributions and find that back-and-forth correlated motion, not local rearrangement barriers, dominates the activation energy, resolving how low-barrier rearrangements yield large macroscopic activation energies. These correlations arise from asymmetry between forward and reverse barriers, a generic feature of disordered energy landscapes. We find that the correlation-driven mechanism is active beyond metallic glass alloys, including SiO2 and a single-component Lennard-Jones glass. The latter demonstrates that the correlation originates from structural disorder rather than chemical complexity. The framework also explains accelerated surface diffusion, where reduced activation energies arise primarily from weaker correlations rather than changes in local rearrangement barriers. Our results establish a direct, quantitative link between atomic-scale dynamics and macroscopic transport, providing a predictive basis for kinetics in disordered materials.

Anderson transition in disordered Hatano-Nelson systems

2026-03-18T21:16:45Z

We illuminate the fundamental mechanism responsible for the transition between the non-Hermitian skin effect and defect-induced Anderson localization in the bulk via the study of Lyapunov exponents. We obtain a proof that the change of the topological invariant associated with an eigenvalue coincides with the eigenvector crossover from non-Hermitian skin effect to Anderson localization, establishing a universal criterion for localization behavior.

Optimal Decoding with the Worm

2026-03-18T18:44:39Z

We propose a new decoder for "matchable'' qLDPC codes that uses a Markov Chain Monte Carlo algorithm - called the worm algorithm - to approximately compute the probabilities of logical error classes given a syndrome. The algorithm hence performs (approximate) optimal decoding, and we expect it to be computationally efficient in certain settings. The algorithm is applicable to decoding random errors for the surface code, the honeycomb Floquet code, and hyperbolic surface codes with constant rate, in all cases with and without measurement errors. The efficiency of the decoder hinges on the mixing time of the underlying Markov chain. We give a rigorous mixing time guarantee in terms of a quantity that we call the defect susceptibility. We connect this quantity to the notion of disorder operators in statistical mechanics and use this to argue (non-rigorously) that the algorithm is efficient for typical errors in the entire decodable phase. We also demonstrate the effectiveness of the worm decoder numerically by applying it to the surface code with measurement errors as well as a family of hyperbolic surface codes. For most codes, the matchability condition restricts direct application of our decoder to noise models with independent bit-flip, phase-flip, and measurement errors. However, our decoder returns soft information which makes it useful also in heuristic "correlated decoding'' schemes which work beyond this simple setting. We demonstrate this by simulating decoding of the surface code under depolarizing noise, and we find that the threshold for "correlated worm decoding'' is substantially higher than for both minimum-weight perfect matching and for correlated matching.

Statistical Mechanics of Random Hyperbolic Graphs within the Fermionic Maximum-Entropy Framework

2026-03-18T18:09:03Z

The intricate relations between elements in natural and human-made systems sustain the complex processes that shape our world, forming multiscale networks of interactions. These networks can be represented as graphs composed of nodes connected by links and, regardless of their domain, they share a set of fundamental structural properties. The family of network models in hyperbolic space constitutes one of the most advanced frameworks accounting for such properties, including sparsity, the small-world property, heterogeneity and hierarchical organization, high clustering, and scale invariance under network renormalization transformations. These geometric models also exhibit other intriguing phenomena, such as an anomalous, temperature-dependent phase transition between a geometric and a non-geometric phase. In simple graph representations, where network links are unweighted, the model can be derived within a statistical-mechanics framework by maximizing the Gibbs entropy of the graph ensemble subject to constraints imposed by observations, with links effectively behaving as fermionic particles. In this topical review, I revisit these derivations previously scattered across different sources and complement them, in order to properly contextualize and consolidate hyperbolic random graphs within the broad framework of the maximum-entropy principle in the statistical mechanics of complex networks. The approach presented here represents the least-biased prediction of the fundamental set of core network properties and establishes a principled framework for analyzing network structure, offering new perspectives and powerful analytical tools for both theoretical and empirical studies.

Removing nodal and support-mismatch pathologies in Variational Monte Carlo via blurred sampling

2026-03-18T18:00:05Z

Variational Monte Carlo (VMC) is a powerful and fast-growing method for optimizing and evolving parameterized many-body wave functions, especially with modern neural-network quantum states. In practice, however, the stochastic estimators that form the backbone of the method can become unstable or biased due to the presence of nodes, a ubiquitous feature of quantum wave functions. In the continuum, this results in heavy-tailed estimators with potentially divergent variances, while in discrete Hilbert spaces the sampling distribution can miss parts of the support needed to form unbiased estimators. These statistical pathologies lead to unreliable optimization trajectories in stochastic reconfiguration or incorrect variational dynamics in time-dependent Variational Monte Carlo (t-VMC), and severely limit the power of the numerical simulations. We introduce blurred sampling to address these difficulties. The method has a number of rigorous properties that make it well-behaved, effective and efficient. Additionally it is a post-processing approach that can be used without modifying the underlying sampler and incurs only minimal overhead. We demonstrate its effectiveness on several representative examples where standard sampling approaches are known to fail, and apply it to large-scale problems in spin dynamics. This work establishes a broadly applicable framework for robust VMC and t-VMC calculations.

Exponents and front fluctuations in the quenched Kardar-Parisi-Zhang universality class of one and two dimensional interfaces

2026-03-18T16:35:48Z

We have simulated an automaton version of the quenched Kardar-Parisi-Zhang (qKPZ) equation in one and two dimensions in order to study the scaling properties of the interface at the depinning transition. Specifically, the $α$, $β$, $θ$, and $δ$ critical exponents characterizing the surface kinetic roughening and depinning behaviors have been directly computed from the simulations. In addition, by studying the height-difference correlation function in real space, we have also been able to directly compute the dynamic correlation length and its associated dynamic critical exponent $z$. The full sets of scaling exponents are largely compatible with those of the Directed Percolation Depinning universality class for one and two dimensional interfaces. Furthermore, we have computed numerically the probability density function (PDF) of the front fluctuations in the growth regime, finding its asymptotic form in one and two dimensions. While the PDF features strongly non-Gaussian skewness and kurtosis values, it also differs from the PDF of the KPZ equation with time-dependent noise for physical substrate dimensions, both in the central part and at the tails of the distribution.

Unlearnable phases of matter

2026-03-18T15:52:28Z

We identify fundamental limitations in machine learning by demonstrating that non-trivial mixed-state phases of matter are computationally hard to learn. Focusing on unsupervised learning of distributions, we show that autoregressive neural networks fail to learn global properties of distributions characterized by locally indistinguishable (LI) states. We demonstrate that conditional mutual information (CMI) is a useful diagnostic for LI: we show that for classical distributions, long-range CMI of a state implies a spatially LI partner. By introducing a restricted statistical query model, we prove that nontrivial phases with long-range CMI, such as strong-to-weak spontaneous symmetry breaking phases, are hard to learn. We validate our claims by using recurrent, convolutional, and Transformer neural networks to learn the syndrome and physical distributions of toric/surface code under bit flip noise. Our findings suggest hardness of learning as a diagnostic tool for detecting mixed-state phases and transitions and error-correction thresholds, and they suggest CMI and more generally ``non-local Gibbsness'' as metrics for how hard a distribution is to learn.

Low-dimensional model for adaptive networks of spiking neurons

2026-03-18T11:28:29Z

We investigate a large ensemble of Quadratic Integrate-and-Fire (QIF) neurons with heterogeneous input currents and adaptation variables. Our analysis reveals that for a specific class of adaptation, termed quadratic spike-frequency adaptation (QSFA), the high-dimensional system can be exactly reduced to a low-dimensional system of ordinary differential equations, which describes the dynamics of three mean-field variables: the population's firing rate, the mean membrane potential, and a mean adaptation variable. The resulting low-dimensional firing rate equations (FRE) uncover a key generic feature of heterogeneous networks with spike frequency adaptation: Both the center and the width of the distribution of the neurons' firing frequencies are reduced, and this largely promotes the emergence of collective synchronization in the network. Our findings are further supported by the bifurcation analysis of the FRE, which accurately captures the collective dynamics of the spiking neuron network, including phenomena such as collective oscillations, bursting, and macroscopic chaos.

Exploring the role of connectivity in disordered system

2026-03-18T07:38:26Z

We study a minimal model of disordered systems, the random field Ising model (RFIM) on a generalized Petersen Graph, GP(N,k). This graph has a connected inner and outer loop, where both the loops consist of N nodes constituting a total of 2N nodes. The parameter k satisfies the condition 1<=k<=N/2, such that any site i in the inner loop has i-k and i+k as its two nearest neighbours, apart from its connection to a node on the outer loop. Thus, each node in GP(N,k) has coordination number z=3, and by varying k different connections between the nodes in the inner loop can be obtained. The objective is to study whether different connectivity between nodes in these graphs affects the system's response to an external field when the coordination number is fixed. This is of interest because critical behaviour is absent for z<=3 on a random graph which has been solved exactly as well as on the honeycomb lattice in the context of RFIM. Using single-spin-flip Glauber dynamics at zero temperature, we compare the system's response with the known case of a z=3 random graph and the generalized Petersen graph for various connectivity k, albeit for the same z. Our study finds the absence of critical behaviour on GP(N,k) highlighting the importance of coordination number over varying connectivity between the nodes. Additionally, we explore the case of directed GP(N,k) and compare it with the undirected GP(N,k) results.

Anomalous localization and duality in non-Hermitian quasiperiodic models

2026-03-18T06:25:50Z

Boundary conditions can have dramatic impact in non-Hermitian systems, as exemplified by the non-Hermitian skin effect. Focusing on one-dimensional non-Hermitian quasiperioidic lattices, we show that the interplay of quasiperiodicity and the non-Hermitian skin effect leads to counterintuitive localization properties. On the one hand, for Anderson localized states under the periodic boundary condition, we find that their localization features can be boundary-sensitive, which originates from the incompatibility of the periodic boundary condition with quasiperiodicity. On the other hand, for non-localized states, the well-known extended-localized duality relation can break down, as their counterparts in the dual model can also be nonlocal. We discuss how these remarkable phenomena can be engineered and analyzed from the perspective of Lyapunov exponents. Our findings shed new light on localization in non-Hermitian quasiperiodic systems.

Theta-term in Russian Doll Model: phase structure, quantum metric and BPS multifractality

2026-03-17T19:38:16Z

We investigate the phase structure of the deterministic and disordered versions of the Russian Doll Model (RDM), which is a generalization of Richardson model of superconductivity in a finite system with time-reversal symmetry breaking parameter $θ$. It is one of the simplest examples of the cyclic RG where $\log N$ plays the role of the RG time. The deterministic model is integrable and shares the same Bethe Ansatz (BA) equations with the inhomogeneous twisted XXX spin chain. We analyze the quantum metric, the Berry curvature, and the fractal dimension in the sector with a single Cooper pair. A rich phase structure in the $(θ,γ)$ parameter plane is found, where $γ\log N$ quantifies the hopping term. For the deterministic RDM we clearly identify the extended domain of non-ergodic multifractal phase on the $(θ,γ)$ parameter plane supporting the reentrance transitions between the localized, ergodic, and multifractal phases. We find the pattern of phase transitions in the global charge $Q(θ,γ)$, which arises from the BA equation. In particular, in the multifractal phase in the deterministic model $Q(γ)$ exhibits the analogue of "charge concentration" and fortuity phenomena discussed in the context of black hole microstates at finite $N$. The BA equations in RDM exactly coincide with the equations defining the ground states in the theory on the worldvolume of the vortex strings in $N_F=2N_C$ ${\cal N}=2$ SQCD at a strong coupling point $\frac{1}{g_{YM}^2}=0$ with identification $θ_{RDM}= θ_{4D}-π$. We conjecture that the Hamiltonian of the RDM model describes the mixing in particular 2d-4d BPS sector of the Hilbert space. Our findings provide an example of the BPS multifractality regime for the probe operator in the sector of Hilbert space, and we comment on the possible application to dense QCD with $θ$ term.