https://arxiv.org/api/UA1YWWm5MAOOVhiRETE6QgPMurc 2026-06-10T14:53:17Z 26559 180 15 http://arxiv.org/abs/2512.11953v2 2026-05-21T03:28:15Z

We find an unexpected phenomenon of coherently synchronized oscillations in a mirror-symmetric many-body localized system. A synchronization transition of the spin oscillations is found by changing the spin-spin interactions. To understand this phenomenon, an effective Ising model based on local integrals of motion is proposed. We find that the synchronization transition can be understood as a paramagnetic-to-ferromagnetic Ising transition. Based on the Ising model, we theoretically estimate the synchronized frequencies and the synchronization transition points, which agree well with numerical results.

2025-12-12T19:00:00Z 5 pages, 4 figures Phys. Rev. B 113, L180204 (2026) Zi-Jian Li Yi-Ting Tu Sankar Das Sarma 10.1103/dlfh-zmw4 http://arxiv.org/abs/2605.21806v1 2026-05-20T23:03:56Z

We investigate the emergence of synchronization in a network of coupled chaotic macroeconomic systems. Each node represents an economy characterized by three key variables savings, gross domestic product (GDP), and foreign capital inflows. These economies interact or are connected through a fitness-based probability that depends on the potential GDP of each node. This formulation allows both structural heterogeneity, arising from uneven network connectivity, and dynamical heterogeneity, due to differences in local parameters, to be explored within a unified framework. Using both numerical simulations and a mean-field approximation, by varying the coupling strength and the degree of heterogeneity of both network topology and dynamical behavior of the nodes, we analyze synchronization transitions. Our results show that the mean-field approach accurately captures the collective dynamics in homogeneous and fully connected networks even with heterogeneity within the intrinsic dynamic of the nodes but fails when strong heterogeneity in the structure of the network is introduced. In heterogeneous networks, the system exhibits partial synchronization and on--off intermittency, where coherent phases of global synchronization alternate with abrupt desynchronization bursts. The distribution of laminar phase durations follows a power-law scaling, consistent with theoretical predictions for intermittent synchronization. From an economic perspective, these results suggest that global business cycle synchronization is inherently fragile: strong integration can promote temporary coordination among economies, but structural and dynamical disparities inevitably lead to intermittent breakdowns of collective behavior.

2026-05-20T23:03:56Z Thierry Njougouo Fernando Fagundes Ferreira Diego Garlaschelli http://arxiv.org/abs/2411.05785v2 2026-05-20T21:55:29Z

The maximum likelihood (ML) decoder in the two-dimensional surface code with generic unitary errors is governed by a statistical mechanics model with complex weights, which can be simulated via (1+1)D transfer matrix contraction. Information loss with an increasing error rate manifests as a ferromagnetic-to-paramagnetic transition in the contraction dynamics. In this work, we establish entanglement as a separate obstruction to decoding; it can undergo a transition from area- to volume-law scaling in the transfer matrix contraction with increasing unitary error rate. In particular, the volume-law entanglement can coexist with ferromagnetic order, giving rise to a phase in which the encoded information is retained yet is effectively undecodable. We numerically simulate the ML decoding in the surface code subject to both single- and two-qubit Pauli-X rotations and obtain a phase diagram that contains a ferromagnetic area-law, a paramagnetic volume-law, and a potential ferromagnetic volume-law phase. We further show that, starting from the paramagnetic volume-law phase, tilting the single-qubit rotation away from the X-axis couples the stat-mech models for X and Z errors and can lead to a ferromagnetic volume-law phase in which, although Z errors remain in principle correctable, the encoded classical information is hard to recover. To perform numerical simulations, we develop an algorithm for syndrome sampling based on an isometric tensor network representation of the surface code.

2024-11-08T18:52:59Z 5 + 13 pages, 3 + 9 figures Yimu Bao Sajant Anand http://arxiv.org/abs/2605.21486v1 2026-05-20T17:59:40Z

Hyperparameter transfer allows extrapolating optimal optimization hyperparameters from small to large scales, making it critical for training large language models (LLMs). This is done either by fitting a scaling law to the hyperparameters or by a judicious choice of parameterization, such as Maximal Update ($μ$P), that renders optimal hyperparameters approximately scale invariant. In this paper, we first develop a framework to quantify hyperparameter transfer through three metrics: (1) the quality of the scaling law fit, (2) the robustness to extrapolation errors, and (3) the asymptotic loss penalty due to choice of parameterization. Next, we investigate through a comprehensive series of ablations why $μ$P appears to offer high-quality learning rate transfer relative to standard parameterization (SP), as existing theory is inadequate. We find that the overwhelming benefit of $μ$P relative to SP when training with AdamW arises simply from maximizing the learning rate of the embedding layer. In SP, the embedding layer learning rate acts as a bottleneck that induces training instabilities; increasing it by a factor of width to match $μ$P dramatically smooths out training while improving hyperparameter transfer. We also find that weight decay improves the scaling law fits, while, in the fixed token-per-parameter setting, it hurts the robustness of the extrapolation.

2026-05-20T17:59:40Z 10+28 pages, 5+17 figures Dayal Singh Kalra Maissam Barkeshli http://arxiv.org/abs/2605.21451v1 2026-05-20T17:42:34Z

Universal approximation theorems provide a mathematical explanation for the expressive power of neural networks. They assert that, under mild conditions on the activation function, feedforward neural networks are dense in broad function classes, such as continuous functions on compact subsets of $\mathbb{R}^d$, $L^p$ spaces, or Sobolev spaces. Over the past four decades, these qualitative universality results have evolved into a rich quantitative theory addressing approximation rates, parameter efficiency, and the role of architectural features such as depth and width. This survey presents several glimpses into this theory. We review classical density results for single-hidden-layer networks, as well as quantitative bounds that relate approximation error to network size and smoothness assumptions on target functions. Particular emphasis is placed on depth--width trade-offs and on results demonstrating that deeper architectures can achieve superior parameter efficiency for structured function classes. In addition to standard feedforward neural networks, we also review recent developments on Kolmogorov--Arnold Networks (KANs), which offer an alternative architectural paradigm and whose approximation-theoretic properties have begun to attract significant theoretical attention.

2026-05-20T17:42:34Z 31 pages, 4 figures Soumendu Sundar Mukherjee Himasish Talukdar http://arxiv.org/abs/2605.21423v1 2026-05-20T17:17:02Z

Quasiperiodic systems are an intermediate class of systems between periodic crystals and disordered systems, famously exhibiting metal-insulator transitions (MITs) even in one dimension. While their transport properties have been studied extensively, a systematic analysis of the finite-frequency optical conductivity near the critical point has been lacking. In this work, we carry out a detailed study of the optical conductivity in the paradigmatic Aubry-André model. We find that the zero-temperature low-frequency optical signal is strongly restructured by the quasiperiodic potential, exhibiting an optical gap that closes discontinuously as the system approaches the MIT. Most strikingly, we uncover a mechanism for a strong enhancement of the low-frequency finite temperature optical conductivity at certain resonant frequencies. This enhancement stems from the thermal activation of Pauli-blocked transitions between strongly resonant van Hove singularities. This mechanism provides new insight into finite-frequency transport in quasiperiodic systems and a new pathway for manipulating optical properties near a localization transition. Furthermore, our findings establish the optical response as a powerful, experimentally accessible tool for probing non-trivial quasiperiodicity effects.

2026-05-20T17:17:02Z 19 pages, 12 figures Raul Liquito Miguel Gonçalves Bruno Amorim Eduardo V. Castro http://arxiv.org/abs/2111.08031v4 2026-05-20T17:04:20Z

The Rosenzweig-Porter random matrix ensemble serves as a qualitative phenomenological model for the level statistics and fractality of eigenstates across the many-body localization transition in static systems. We propose a unitary (circular) analogue of this ensemble, which similarly captures the phenomenology of many-body localization in periodically driven (Floquet) systems. We define this ensemble as the outcome of a Dyson Brownian motion process. We show numerical evidence that this ensemble shares some key statistical properties with the Rosenzweig-Porter ensemble for both the eigenvalues and the eigenstates.

2021-11-15T19:00:10Z 7 pages, 3 figures SciPost Phys. 12, 082 (2022) Wouter Buijsman Yevgeny Bar Lev 10.21468/SciPostPhys.12.3.082 http://arxiv.org/abs/2605.21402v1 2026-05-20T17:00:37Z

Generative neural networks learn how to produce highly realistic images from a large, but finite number of examples - or do they simply memorise their training set? To settle this question, Kadkhodaie, Guth, Simoncelli and Mallat (ICLR '24) trained diffusion models independently on disjoint subsets of a dataset and showed that they converge to nearly the same density when the number of training images is large enough. This result raises two basic questions: how much data do you need for convergence, and what does convergence capture about learning the data distribution? Here, we address these questions by providing an exact analytical characterisation of the transition from memorisation to generalisation in linear generative models. We find that these models memorise at small load, while convergence emerges continuously when the number of samples is linear in the input dimension. Strikingly, we find that convergence is insensitive to recovery of the principal latent factors of the data, which are recovered in a sharp transition. After extending our approach to data with power-law spectra, we find the same distinction between convergence and latent recovery in our experiments with convolutional denoisers and in the data of Kadkhodaie et al. We thus show that generalisation in generative models decomposes into at least two distinct objectives: matching the bulk of the data distribution and recovering the principal latent factors. These objectives correspond to two different distances between true and learnt data distribution, and only the first one is captured by convergence.

2026-05-20T17:00:37Z Antoine Maillard Sebastian Goldt http://arxiv.org/abs/2605.21364v1 2026-05-20T16:28:57Z

Despite the knowledge that social, economical, and ecological networks are often of a small-world nature with inter-nodal distance growing even slower than logarithmically with system size, we often assume theoretical systems to be outside of this regime, to make them easier to treat analytically. Here we derive a framework to apply the powerful dynamical mean-field theory on highly heterogeneous networks that is able to account for more of the degree correlations naturally arising from network constraints, known as structural cut-offs. We apply this framework to the well-studied and understood disordered Lotka-Volterra model, and show typically reported observables such as survival rates and stability for these systems on ultra small-world networks. We find much better agreement for these variables for all ranges of exponents for simulated power-law networks as well as empirically sourced networks.

2026-05-20T16:28:57Z 10 pages, 5 figures Nirbhay Patil Ada Altieri Fabian Aguirre-Lopez http://arxiv.org/abs/2605.12597v2 2026-05-20T15:46:04Z

Computational sampling has been central to the sciences since the mid-20th century. While machine-learning-based approaches have recently enabled major advances, their behavior remains poorly understood, with limited theoretical control over when and why they succeed. Here we provide such insight for diffusion models-a class of generative schemes highly effective in practice-by analyzing their application to the $O(n)$ model of statistical field theory in the Gaussian limit $n \to \infty$. In this analytically tractable setting, we show that training a score model with a one-layer network architecture matching the exact solution exhibits a form of critical slowing down in parameter learning. This slowing down also impacts the generation process, indicating that the well-known difficulties of sampling near criticality persist even for learned generative models. To overcome this bottleneck, we demonstrate the power of combining architectural depth with physical locality. We find that using a two-layer architecture drastically reduces the critical slowing down, with the training time scaling logarithmically rather than quadratically with system size. By introducing a local score approximation we show that this acceleration in training time can be achieved without increasing the number of neural network parameters. Taken together, these results demonstrate that diffusion models can overcome the critical slowing down through appropriate architectural design, and establish a controlled framework for understanding and improving learned sampling methods in statistical physics and beyond.

2026-05-12T18:00:02Z 17 pages, 8 figures Luca Maria Del Bono Giulio Biroli Patrick Charbonneau Marylou Gabrié http://arxiv.org/abs/2605.21254v1 2026-05-20T14:42:07Z

We introduce a random matrix framework for studying statistical-mechanical lattice systems through spectral observables. Equilibrium configurations sampled from a Boltzmann measure are mapped to matrix ensembles whose covariance structure is inherited from the spatial correlations of the underlying model. This construction maps real-space correlation functions to a momentum-space variance profile, providing a direct bridge between statistical-mechanical correlations and correlated random matrix ensembles. We derive this variance profile in finite-correlation-length and critical regimes, and compute spectral moments within a Wick-contraction expansion. A complementary self-consistent description of the bulk density is developed using the resolvent formalism. These analytical methods are benchmarked against Monte Carlo data for the two-dimensional Ising model and three-dimensional Edwards--Anderson spin glasses. In both cases, the spectra evolve from the semicircle law at high temperature to model-dependent critical forms reflecting the structure of correlations. The framework, therefore, provides a quantitative spectral route to probing collective behavior in ordered and disordered statistical systems, while also defining a class of physically motivated correlated random matrix ensembles.

2026-05-20T14:42:07Z 16 pages, 9 figures Yaprak Önder Abbas Ali Saberi Roderich Moessner http://arxiv.org/abs/2502.18004v4 2026-05-20T14:14:02Z

Higher-order interactions have recently emerged as a promising framework for describing new dynamical phenomena in heterogeneous contagion processes. However, a fundamental open question is how to understand their contribution from the perspective of the physics of critical phenomena. Based on a mesoscopic field-theoretic Langevin description, we show that: (i) pairwise mechanisms such as facilitation or thresholding are formally equivalent to higher-order ones, (ii) pairwise interactions at coarse-grained scales govern the higher-order contact process and, (iii) the interplay between noise and topology is determined by the network spectral dimension. In short, we demonstrate that classical field theories, rooted on model symmetries and/or network dimensionality, still capture the nature of the phase transition, also predicting finite-size effects in real and synthetic networks.

2025-02-25T09:09:27Z 6 pages, 4 figures, and Supplementary Material Phys. Rev. E 113, L062301 (2026) Sandro Meloni Andrea Gabrielli Pablo Villegas 10.1103/98rn-9x9q http://arxiv.org/abs/2510.17193v2 2026-05-20T09:16:09Z

Human mobility, enabled by diverse transportation modes, is fundamental to urban functionality. Studying these movements across scales-from microscopic to macroscopic-yields valuable insights into urban dynamics. Local adaptation and (self-)organization in such systems are expected to result in dynamical behaviors that are represented by stationary trajectories of an appropriate effective action. In this study we develop a Lagrangian dynamical model for movement processes, using local population functions as the coordinate variables. An efficient gradient descent algorithm is introduced to estimate the optimal Lagrangian parameters minimizing a local error function of the dynamical process. We show that even a quadratic Lagrangian, incorporating dissipation, effectively captures the dynamics of synthetic and empirical movement data. The inferred models reveal that inertia and dissipation are of comparable importance, while interactions and randomness in the movements induce significant qualitative changes in model parameters. Our results provide an interpretable and generative model for human mobility, with potential applications in movement prediction.

2025-10-20T06:13:59Z 27 pages, 12 figures Phys. Rev. E 113, 054311 (2026) Babak Benam Abolfazl Ramezanpour 10.1103/vztw-pqy4 http://arxiv.org/abs/2507.11957v2 2026-05-20T07:59:52Z

Spin chains with quenched disorder exhibit rich critical behavior, often captured by real-space renormalization group (RSRG) techniques. However, the physics of such systems in the presence of random measurements (i.e., non-Hermitian dephasing) remains largely unexplored. The interplay between measurements and unitary dynamics gives rise to novel phases and phase transitions in monitored quantum systems. In this work, we investigate the disordered XX spin chain subject to stochastic local measurements in the $X$ and $Y$ bases. By mapping the monitored chain to a non-Hermitian spin ladder with complex couplings, we propose an RSRG-for-excited-states (RSRG-X) approach for this open-system setting. Our analysis reveals a new class of strongly disordered fixed points that emerge due to non-unitarity, broadening the landscape of critical phenomena accessible via RSRG.

2025-07-16T06:43:49Z Tiwary, Siddharth, and Joel E. Moore. "Measurement-Induced Phase Transition in a Disordered XX Spin Chain: A Real-Space Renormalization Group Study." Physical Review B, vol. 113, no. 10, Mar. 2026, p. 104203 Siddharth Tiwary Joel E. Moore 10.1103/51bk-vrp4 http://arxiv.org/abs/2605.20847v1 2026-05-20T07:39:07Z

Identifying the regions responsible for plastic flow in amorphous solids remains an open problem, since structural disorder seems to prevent the direct application of concepts such as dislocations, topological defects that successfully describe irreversible deformations in crystalline systems. Here, we introduce fields of dislocation, disclination, and incompatibility densities, that reduce to the standard sources of plasticity in crystals and assess their predictive power in amorphous materials. We find that, in a simulated two-dimensional glass as well in two- and three-dimensional experimental granular systems, these fields exhibit strong spatial correlations with $D^2_{\text{min}}$, the standard measure used to locate plastic events under shear in disordered solids. Unlike $D^2_{\text{min}}$, these fields also allow to disentangle rotational and translational contributions to the plastic events, revealing that rotational defects becoming dominant in three dimensions. Our approach paves the way for a unified description of plasticity in crystalline and amorphous solids.

2026-05-20T07:39:07Z v1: comments welcome Xin Wang Yang Xu Jin Shang Yi Xing Jie Zhang Yujie Wang Walter Kob Matteo Baggioli