Memorisation, convergence and generalisation in generative models

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.

Random Walks Across Dimensions: Exploring Simplicial Complexes

2026-05-20T16:22:47Z

We introduce a novel operator to describe a random walk process on a simplicial complex. Walkers are allowed to wonder across simplices of various dimensions, bridging nodes to edges, and edges to triangles, via a nested organization that hierarchically extends to higher structures of arbitrary large, but finite, dimension. The asymptotic distribution of the walkers provides a natural ranking to gauge the relative importance of higher order simplices. Optimal search strategies in presence of stochastic teleportation are addressed and the peculiar interplay of noise with higher order structures unraveled.

Opinion dynamics: Statistical physics and beyond

2026-05-20T15:47:22Z

Opinion dynamics, the study of how individual beliefs and collective public opinion evolve, is a fertile domain for applying statistical physics to complex social phenomena. Like physical systems, societies exhibit macroscopic regularities from localized interactions, leading to outcomes such as consensus or fragmentation. This field has grown significantly, attracting interdisciplinary methods and driven by a surge in large-scale behavioral data. This review covers its rapid progress, bridging the literature dispersion. We begin with essential concepts and definitions, encompassing the nature of opinions, microscopic and macroscopic dynamics. This foundation leads to an overview of empirical research, from lab experiments to large-scale data analysis, which informs and validates models of opinion dynamics. We then present individual-based models, categorized by their macroscopic phenomena (e.g., consensus, polarization, echo chambers) and microscopic mechanisms (e.g., homophily, assimilation). Furthermore, the review covers common analytical and computational tools, including stochastic processes, treatments, simulations, and optimization. Finally, we explore emerging frontiers, such as connecting empirical data to models and using AI agents as testbeds for novel social phenomena. By systematizing terminology and emphasizing analogies with traditional physics, this review aims to consolidate knowledge, provide a robust theoretical foundation, and shape future research in opinion dynamics.

The critical slowing down in diffusion models

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.

Schrödinger-invariance in non-equilibrium critical dynamics

2026-05-20T15:22:19Z

The scaling functions of single-time and two-time correlators in systems undergoing non-equilibrium critical dynamics with dynamical exponent ${z}=2$ are predicted from a new time-dependent non-equilibrium representation of the Schrödinger algebra. These explicit predictions are tested and confirmed in the ageing of several exactly solvable models.

Schrödinger-invariance in phase-ordering kinetics

2026-05-20T15:16:59Z

The generic shape of the single-time and two-time correlators in non-equilibrium phase-ordering kinetics with ${z}=2$ is obtained from the co-variance of the four-point response functions. Their non-equilibrium scaling forms follow from a new non-equilibrium representation of the Schrödinger algebra.

Random Matrix Spectra from Boltzmann-Weighted Lattice Ensembles

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.

Spinodal-like scaling behavior after a temperature quench across the first-order phase transition in three-dimensional $q$-state Potts models

2026-05-20T14:36:10Z

We study the out-of-equilibrium spinodal-like behavior of three-dimensional (3D) $q$-state Potts models (for $q\ge 3$), observed when the temperature is quenched across the first-order transition (FOT) point $β_{\rm fo}=T_{\rm fo}^{-1}$. We consider a standard quench protocol, in which high-temperature configurations, thermalized at $β_i<β_{\rm fo}$, are driven across the FOT by a purely relaxational dynamics at $β>β_{\rm fo}$. We focus on the emergence of spinodal-like behaviors in the thermodynamic limit, associated with the dynamic phase change. We argue that, if the nucleation of smooth droplets is the relevant mechanism of the post-quench phase change, for sufficiently small $β_{\rm fo}-β_i>0$, the time-dependent energy density should scale in terms of $ρ= (\ln t)^{3/2} δ$, where $δ= β/β_{\rm fo}-1$, with a discontinuity at a particular value $ρ=ρ_s>0$. This implies the emergence of a spinodal-like behavior, whose time scale $τ$ increases exponentially as $\ln τ\approx (ρ_s/δ)^{2/3}$ in the limit $δ\to 0^+$. We present a numerical analysis of the quench protocol in the 3D $q=6$ Potts model, which supports the above spinodal-like scenario.

Higher-order contagion processes in 3.99 dimensions

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.

Non-equilibrium phase transitions in hybrid Voronoi models of cell colonies

2026-05-20T14:06:25Z

Eukaryotic cells are characterized by a stiff nucleus whose effect in modeling the collective behavior of cell aggregates is usually underestimated. However, increasing experimental evidence links nuclear modifications with phenotypic transition, like the one between epithelial and mesenchymal states. In this work, we explore the effect of short-range repulsive forces in the non-equilibrium dynamics of the self-propelled Voronoi model. We show that the competition between steric repulsions (representing nuclear/cellular compressibility) and Vertex interactions (mimicking cell-cell adhesion/interaction and cytoskeleton organization) generate a variety of non-equilibrium phase transitions from Motility-Induced Phase Separation to mesenchymal-like phases up to disordered confluent configurations. Notably, we found that tuning the nucleus's effective size/compressibility provides an additional way to cross the boundary between the different possible phases in line with experimental observations.

Macroscopic Thermodynamic Framework for the Mpemba Effect

2026-05-20T13:16:01Z

The counterintuitive Mpemba effect, wherein a hotter system cools faster, critically lacks a general macroscopic theory. Here, starting from linear irreversible thermodynamics, we formulate a generalized Newton's cooling law, $\mathrm{d}T/\mathrm{d}t = -[γ_0 + \mathcal{M}Q(t)](T-T_r)$, for a system at temperature $T$ relaxing in a thermal reservoir at $T_r$, where the bare relaxation rate $γ_0$ is modified by an initial-state memory term, $Q(t) \propto T(0)-T_r$. Arising from the interplay between heat flux and structural evolution, the coefficient $\mathcal{M}$ governs anomalous relaxation behaviors, where $\mathcal{M} > 0$ ($\mathcal{M} < 0$) induces the (inverse) Mpemba effect. This universal thermodynamic framework maps out phase diagram to provide general criteria for the Mpemba effect in complex systems, offering a macroscopic picture that bridges disparate microscopic approaches.

Approximate FKG inequalities for phase-bound spin systems, with applications to central limit theorems for exponential random graphs

2026-05-20T12:47:14Z

The Fortuin-Kasteleyn-Ginibre (FKG) inequality is an invaluable tool in monotone spin systems satisfying the FKG lattice condition, which provides positive correlations for all coordinate-wise increasing functions of spins. This inequality has numerous applications and plays an integral role in the proof of various central limit theorems (CLTs), including recent work on ferromagnetic exponential random graph models (ERGMs) wherein a Hamiltonian tilt promotes the presence of small subgraphs like triangles. However, the FKG lattice condition fails to hold when confining a spin system to a particular phase in the low-temperature regime of parameters. Thus it is not a priori clear if each phase internally has positive correlations for increasing functions, or if the positive correlations in the overall model (which is a mixture of phases) arise primarily from the global choice of phase. In this article, we show that the individual phases in ERGMs do indeed satisfy an approximate form of the FKG inequality internally. We use this to finish the proof of various CLTs within each individual phase in the phase-coexistence regime, answering a question posed by Bianchi, Collet, and Magnanini. We present the FKG inequality for ERGMs as a consequence of a more general result which holds under certain inputs related to metastable mixing; we expect this general result to be widely applicable, and we devote a section to spelling out the details of its application to a class of generalized higher-order ferromagnetic Curie-Weiss models where the necessary inputs are relatively transparent.

Universal Predictors for Mixing Time more than Liouvillian Gap

2026-05-20T12:46:40Z

We analyze the mixing time of open quantum systems governed by the Lindblad master equation, showing that it is determined not only by the Liouvillian gap, but also by the trace-norm factor of each decaying Liouvillian eigenmode. By utilizing them as universal predictors of mixing time, we establish general conditions for the fast and rapid mixing, respectively. Specifically, we derive rapid mixing conditions for both the strong and weak dissipation regimes, formulated as sparsity constraints on the Hamiltonian and the local Lindblad operators. Our findings provide a general framework for calculating mixing time and offer a guide for designing dissipation to achieve desired mixing speeds, which has significant implications for efficient experimental state preparation.

Thermodynamic and structural behavior of one-dimensional divalent patchy hard rods: Wertheim's first-order thermodynamic perturbation theory versus exact results

2026-05-20T10:41:26Z

We investigate the thermodynamic and structural properties of divalent patchy hard rods confined to a one-dimensional channel by modeling the bonding sites as attractive square-well (SW) patches located at the rod tips. The zero-range sticky limit is recovered by letting the well width vanish while keeping the stickiness parameter finite. While Wertheim's first-order thermodynamic perturbation theory (TPT1) becomes exact in this sticky limit, it fails for finite-range site-site interactions. We show that the theory can be made exact in one dimension by replacing the standard law of mass action with an exact relation between the density and the fraction of unbonded sites, together with an exact bonding free-energy contribution. Finite-range SW sites produce a richer structural behavior than sticky sites, including monotonic and oscillatory asymptotic decay of the pair correlation function, separated by the Fisher--Widom line. In the monotonic regime, the correlation length exhibits an absolute maximum defining the Widom line, while in the oscillatory regime it may display a local maximum and minimum, whose locus defines the ``Extrema of the Correlation length under Oscillatory decay'' (ECO) line. These features disappear in the sticky limit, where the system remains entirely in the oscillatory regime. We also show that the high-pressure behavior of the correlation length changes from $ξ\sim p^2$ for finite-range SW sites to $ξ\sim p^3$ in the sticky limit.

Competing anisotropies and phase transitions in the $q$-state clock model with a $p$-fold crystalline field

2026-05-20T10:26:02Z

We study the two-dimensional $q$-state clock model in the presence of an additional $p$-fold symmetry-breaking crystalline field using Monte Carlo simulations. While the pure clock model exhibits Berezinskii--Kosterlitz--Thouless (BKT) transitions for sufficiently large $q$, the effect of competing discrete anisotropies on this topological phase remains nontrivial. We show that even weak crystalline fields qualitatively modify the phase diagram by suppressing the BKT phase and inducing transitions to states with true long-range order. The resulting behavior depends sensitively on the interplay between the intrinsic $\mathbb{Z}_q$ symmetry and the imposed $\mathbb{Z}_p$ anisotropy. In particular, in the six-state clock model for $p=2$ we observe qualitatively different scenarios depending on the sign of the field: a single transition for $h_2>0$ and a two-step ordering process for $h_2<0$ with an intermediate ordered phase. For $p=3$, the system exhibits a direct transition consistent with three-state Potts criticality. These results demonstrate that the phase structure cannot be inferred from symmetry considerations alone, but is governed by the competition between distinct locking mechanisms. Our findings provide a discrete counterpart to the multi-frequency sine-Gordon description of generalized $XY$ models and illustrate how additional anisotropies reshape topological phase transitions in two dimensions.