Topic overview

cond-mat.dis-nn

1361 works3732 researchers0 institutions

Topic snapshot

What this area looks like now

1361works

3732authors

0experts visible

0communities

Next steps

Move from topic reading into action

The graph preview below keeps the nearby papers, people and communities visible in the same reading flow.

Inspect nearby papers, researchers, institutions and communities without opening a separate graph page.

Building this graph slice

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Finite-size scaling of hetero-associative retrieval in continuous-signal-driven Ising spin systems

Real-world physical signals are continuous and high-dimensional, yet the statistical-mechanics machinery of associative memory operates on discrete Ising spins. We bridge this divide through a multilayer Ising framework that couples a geometry-preserving continuous-to-Ising encoder (PCA whitening composed with SimHash random-hyperplane projection) to Kanter-Sompolinsky pseudo-inverse memory couplings, embedded directly into the local-field equations of a tri-layer hetero-associative system. The pseudo-inverse correction renders the equal-weight mixture state thermodynamically unstable, so that thermal fluctuations break the cross-modal symmetry and select a single global winner. We further establish a dynamical duality: parallel (Little) updates are structurally required to ignite the cross-modal signal avalanche from a single cued layer, whereas sequential (Glauber) sweeps resolve symmetric superpositions. The operational storage capacity obeys the Amit-Gutfreund-Sompolinsky finite-size correction $α_c(N)=α_c(\infty)-c\,N^{-1/2}$, extrapolating to an asymptotic operational limit $α_c(\infty)\approx 0.50$ under macroscopic-basin retrieval. Applied to multi-channel sleep polysomnography (PhysioNet Sleep-EDF), the architecture reconstructs the macroscopic sleep state on parietal EEG and EOG axes from a single noisy frontal-EEG cue, demonstrating cross-modal recall in the presence of quenched biological disorder.

preprint2026arXiv

Parallel Scan Recurrent Neural Quantum States for Scalable Variational Monte Carlo

Neural-network quantum states have emerged as a powerful variational framework for quantum many-body systems, with recent progress often driven by massively parallel architectures such as transformers. Recurrent neural network quantum states, however, are frequently regarded as intrinsically sequential and therefore less scalable. Here we revisit this view by showing that modern recurrent architectures can support fast, accurate, and computationally accessible neural quantum state simulations. Using autoregressive recurrent wave functions together with recent advances in parallelizable recurrence, we develop variational ansätze, called parallel scan recurrent neural quantum states (PSR-NQS), which can be trained efficiently within variational Monte Carlo in one and two spatial dimensions. We demonstrate accurate benchmark results and show that, with iterative retraining, our approach reaches two-dimensional spin lattices as large as $52\times52$ while remaining in agreement with available quantum Monte Carlo data. Our results establish recurrent architectures as a practical and promising route toward scalable neural quantum state simulations with modest computational resources.

preprint2026arXiv

Factual recall in linear associative memories: sharp asymptotics and mechanistic insights

Large language models demonstrate remarkable ability in factual recall, yet the fundamental limits of storing and retrieving input--output associations with neural networks remain unclear. We study these limits in a minimal setting: a linear associative memory that maps $p$ input embeddings in $\mathbb{R}^d$ to their corresponding~$d$-dimensional targets via a single layer, requiring each mapped input to be well separated from all other targets. Unlike in supervised classification, this strict separation induces~$p$ constraints per association and produces strong correlations between constraints that make a direct characterisation of the storage capacity difficult. Here, we provide a precise characterisation of this capacity in the following way. We first introduce a decoupled model in which each input has its own independent set of competing outputs, and provide numerical and analytical evidence that this decoupled model is equivalent to the original model in terms of storage capacity, spectra of the learnt weights, and storage mechanism. Using tools from statistical physics, we show that the decoupled model can store up to $p_c \log p_c / d^2 = 1 / 2$ associations, and generalise the computation of $p_c$ to linear two-layer architectures. Our analysis also gives mechanistic insight into how the optimal solution improves over a naïve Hebbian learning rule: rather than boosting input-output alignments with broad fluctuations, the optimal solution raises the correct scores just above the extreme-value threshold set by the competing outputs. These findings give a sharp statistical-physics characterisation of factual storage in linear networks and provide a baseline for understanding the memory capacity of more realistic neural architectures.

preprint2026arXiv

Deep Learning as Neural Low-Degree Filtering: A Spectral Theory of Hierarchical Feature Learning

Understanding how deep neural networks learn useful internal representations from data remains a central open problem in the theory of deep learning. We introduce Neural Low-Degree Filtering (Neural LoFi), a stylized limit of gradient-based training in which hierarchical feature learning becomes an explicit iterative spectral procedure. In this limit, the dynamics at each layer decouple: given the current representation, the next layer selects directions with maximal accessible low-degree correlation to the label. This yields a tractable surrogate mechanism for deep learning, together with a natural kernel-space interpretation. Neural LoFi provides a mathematically explicit framework for studying multi-layer feature learning beyond the lazy regime. It predicts how representations are selected layer by layer, explains how emergence of concepts arises with given sample complexity,and gives a concrete mechanism by which depth progressively constructs new features from old ones through low-degree compositionality. We complement the theory with mechanistic experiments on fully connected and convolutional architectures, showing that Neural LoFi improves over lazy random-feature baselines, recovers meaningful structured filters, and predicts representations aligned with early gradient-descent feature discovery with real datasets.

preprint2026arXiv

Context-Gated Associative Retrieval: From Theory to Transformers

Hopfield networks and their generalizations have established deep connections among biological associative memories, statistical physics, and transformers. Yet most models treat retrieval as a fixed query-to-memory mapping, ignoring the role of external context in recall. In this work, we propose a two-stage associative memory architecture, wherein a context-gate subcircuit reshapes the retrieval energy landscape before and during recall. We show theoretically that context gating increases inter-memory separation while inducing sparsity, translating into exponential improvements in retrieval. Crucially, we prove that the system admits a unique self-consistent fixed point, revealing that the resulting retrieval state is driven by both a direct contextual bias and a second-order retrieval-gate feedback loop. We then bridge this theory to transformers; specifically, we evaluate a first-order approximation on Llama-3, confirming that in-context learning acts as context-gated retrieval. Native dynamics mirror our theory: context localizes a memory subspace, enabling the zero-shot query to cleanly discriminate. Ultimately, this framework provides a mechanistic link between associative memory theory and LLM phenomenology.

preprint2026arXiv

The critical slowing down in diffusion models

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.

preprint2022arXiv

An Atomistic Model of Field-Induced Resistive Switching in Valence Change Memory

In Valence Change Memory (VCM) cells, the conductance of an insulating switching layer is reversibly modulated by creating and redistributing point defects under an external field. Accurate simulations of the switching dynamics of these devices can be difficult due to their typically disordered atomic structures and inhomogeneous arrangements of defects. To address this, we introduce an atomistic framework for modelling VCM cells. It combines a stochastic Kinetic Monte Carlo approach for atomic rearrangement with a quantum transport scheme, both parameterized at the ab-initio level by using inputs from Density Functional Theory (DFT). Each of these steps operates directly on the underlying atomic structure. The model thus directly relates the energy landscape and electronic structure of the device to its switching characteristics. We apply this model to simulate non-volatile switching between high- and low-resistance states in an TiN/HfO2/Ti/TiN stack, and analyze both the kinetics and stochasticity of the conductance transitions. We also resolve the atomic nature of current flow resulting from the valence change mechanism, finding that conductive paths are formed between the undercoordinated Hf atoms neighboring oxygen vacancies. The model developed here can be applied to different material systems to evaluate their resistive switching potential, both for use as conventional memory cells and as neuromorphic computing primitives.

preprint2022arXiv

The generalized Lyapunov exponent for the one-dimensional Schrödinger equation with Cauchy disorder: some exact results

We consider the one-dimensional Schrödinger equation with a random potential and study the cumulant generating function of the logarithm of the wave function $ψ(x)$, known in the literature as the "generalized Lyapunov exponent"; this is tantamount to studying the statistics of the so-called "finite size Lyapunov exponent". The problem reduces to that of finding the leading eigenvalue of a certain non-random non-self-adjoint linear operator defined on a somewhat unusual space of functions. We focus on the case of Cauchy disorder, for which we derive a secular equation for the generalized Lyapunov exponent. Analytical expressions for the first four cumulants of $\ln|ψ(x)|$ for arbitrary energy and disorder are deduced. In the universal (weak-disorder/high-energy) regime, we obtain simple asymptotic expressions for the generalized Lyapunov exponent and for all the cumulants. The large deviation function controlling the distribution of $\ln|ψ(x)|$ is also obtained in several limits. As an application, we show that, for a disordered region of size $L$, the distribution $\mathcal{W}_L$ of the conductance $g$ exhibits the power law behaviour $\mathcal{W}_L(g)\sim g^{-1/2}$ as $g\to0$.

preprint2022arXiv

Anomalous Floquet-Anderson Insulator with Quasiperiodic Temporal Noise

Time-periodic (Floquet) drive can give rise to novel symmetry breaking and topological phases of matter. Recently, we showed that a quintessential Floquet topological phase known as the anomalous Floquet-Anderson insulator is stable to noise on the timing of its Floquet drive. Here, we perturb the anomalous Floquet-Anderson insulator at a single incommensurate frequency, resulting in a quasiperiodic 2-tone drive. Our numerics indicate that a robust topological phase survives at weak noise with topological pumping that is more stable than the case of white noise. Within the topological phase, we show that particles move subdiffusively, which is directly responsible for stabilizing topological transport. Surprisingly, we discover that when quasiperiodic noise is sufficiently strong to kill topology, the system appears to exhibit diffusive dynamics, suggesting that the correlated structure of the quasiperiodic noise becomes irrelevant.

preprint2026arXiv

Spectral Dynamics in Deep Networks: Feature Learning, Outlier Escape, and Learning Rate Transfer

We study the evolution of hidden-weight spectra in wide neural networks trained by (stochastic) gradient descent. We develop a two-level dynamical mean-field theory (DMFT) that jointly tracks bulk and outlier spectral dynamics for spiked ensembles whose spike directions remain statistically dependent on the random bulk. We apply this framework to two settings: (1) infinite-width nonlinear networks in mean-field/$μ$P scaling and (2) deep linear networks in the proportional high-dimensional limit, where width, input dimension, and sample size diverge with fixed ratios. Our theory predicts how outliers evolve with training time, width, output scale, and initialization variance. In deep linear networks, $μ$P yields width-consistent outlier dynamics and hyperparameter transfer, including width-stable growth of the leading NTK mode toward the edge of stability (EoS). In contrast, NTK parameterization exhibits strongly width-dependent outlier dynamics, despite converging to a stable large-width limit. We show that this bulk+outlier picture is descriptive of simple tasks with small output channels, but that tasks involving large numbers of outputs (ImageNet classification or GPT language modeling) are better described by a restructuring of the spectral bulk. We develop a toy model with extensive output channels that recapitulates this phenomenon and show that edge of the spectrum still converges for sufficiently wide networks.

preprint2022arXiv

Elastic turbulence homogenizes fluid transport in stratified porous media

Many key environmental, industrial, and energy processes rely on controlling fluid transport within subsurface porous media. These media are typically structurally heterogeneous, often with vertically-layered strata of distinct permeabilities -- leading to uneven partitioning of flow across strata, which can be undesirable. Here, using direct in situ visualization, we demonstrate that polymer additives can homogenize this flow by inducing a purely-elastic flow instability that generates random spatiotemporal fluctuations and excess flow resistance in individual strata. In particular, we find that this instability arises at smaller imposed flow rates in higher-permeability strata, diverting flow towards lower-permeability strata and helping to homogenize the flow. Guided by the experiments, we develop a parallel-resistor model that quantitatively predicts the flow rate at which this homogenization is optimized for a given stratified medium. Thus, our work provides a new approach to homogenizing fluid and passive scalar transport in heterogeneous porous media.

preprint2021arXiv

Stark many-body localization: Evidence for Hilbert-space shattering

We study the dynamics of an interacting quantum spin chain under the application of a linearly increasing field. This model exhibits a type of localization known as Stark many-body localization. The dynamics shows a strong dependence on the initial conditions, indicating that the system violates the conventional ("strong") eigenstate thermalization hypothesis at any finite gradient of the field. This is contrary to reports of a numerically observed ergodic phase. Therefore, the localization is crucially distinct from disorder-driven many-body localization, in agreement with recent predictions on the basis of localization via Hilbert-space shattering.

preprint2026arXiv

Exact Fixed-Point Constraints in Neural-ODEs with Provable Universality

We introduce a technique that enables Neural-ODEs to approximate arbitrary velocity fields with a priori planted fixed-points. Specifically, a recipe is given to explicitly accommodate for a finite collection of points in the reference multi-dimensional space of the Neural-ODE where the velocity field is exactly equal to zero. In this way, the gradient-based training is rigorously constrained inside the prescribed hypothesis class while leaving the expressive power of the Neural-ODE unaltered. We rigorously prove the universality of the Neural-ODE under any local constraints in the velocity field and give a computationally convenient way of imposing the fixed points. Our method is then tested on two paradigmatic physical models.

preprint2021arXiv

Hierarchical Short- and Medium-Range Order Structures in Amorphous GexSe1-x for Selectors Applications

In the upcoming process to overcome the limitations of the standard von Neumann architecture, synaptic electronics is gaining a primary role for the development of in-memory computing. In this field, Ge-based compounds have been proposed as switching materials for nonvolatile memory devices and for selectors. By employing the classical molecular dynamics, we study the structural features of both the liquid states at 1500K and the amorphous phase at 300K of Ge-rich and Se-rich chalcogenides binary GexSe1-x systems in the range x 0.4-0.6. The simulations rely on a model of interatomic potentials where ions interact through steric repulsion, as well as Coulomb and charge-dipole interactions given by the large electronic polarizability of Se ions. Our results indicate the formation of temperature-dependent hierarchical structures with short-range local orders and medium-range structures, which vary with the Ge content. Our work demonstrates that nanosecond-long simulations, not accessible via ab initio techniques, are required to obtain a realistic amorphous phase from the melt. Our classical molecular dynamics simulations are able to describe the profound structural differences between the melt and the glassy structures of GeSe chalcogenides. These results open to the understanding of the interplay between chemical composition, atomic structure, and electrical properties in switching materials.

preprint2025arXiv

Decoherence-induced self-dual criticality in topological states of matter

Quantum measurements performed on a subsystem of a quantum many-body state can generate entanglement for its remaining constituents. The whole system including the measurement record is described by a hybrid mixed state, which can exhibit exotic phase transitions and critical phenomena. We demonstrate that generic measurement-induced phase transitions (MIPTs) can be cast as decoherence-induced critical mixed states in one higher dimension, by constructing a projected entangled pair state (PEPS) prior to decoherence or measurement. In this context, a deeper conceptual understanding of such mixed-state criticality is called for, particularly with regard to algebraic symmetry as an advanced organizing principle for such entangled states of matter. Integrating these connections we investigate the role of self-dual symmetry -- a fundamental notion in theoretical physics -- in mixed states, showing that the decoherence of electric (e) and magnetic (m) vortices from the 2D bulk of the toric code, or equivalently, a 2D cluster state with symmetry-protected topological order, can leave a (1+1)D quantum critical mixed state protected by a weak Kramers-Wannier self-dual symmetry. The corresponding self-dual critical bulk is described by the N->1 limit of the 2D Non-linear Sigma Model in symmetry class D with target space SO(2N)/U(N) at $Θ$-angle $π$, and represents a "measurement-version" of the Cho-Fisher network model subjected to Born-rule randomness...

preprint2026arXiv

A Random-Matrix Criterion for Initializing Gated Recurrent Neural Networks

Proper weight initialization prior to training has historically been one of the key factors that helped kick off the deep learning revolution. Initialization is even more crucial in "reservoir computing", where the weights of a readout layer are learned linearly while the reservoir weights are fixed and largely determine the richness, stability and memory of the resulting dynamics. In the infinite-width limit it has been shown that meaningful initializations are those sitting at an effective critical point of the randomly initialized model. The phase transition is controlled by the weight variance $g^2$ and separates an ordered phase from a chaotic one where information progressively degrades. Here we derive a simple criterion to estimate the critical $g_c$ for a broad class of recurrent architectures and we show that it closely tracks the gain at which a gated-RNN reservoir achieves peak performance on a chaotic forecasting task. Finally, we argue that our criterion can serve as a design principle for future initialization schemes.

preprint2022arXiv

Topological Anderson insulators induced by random binary disorders

Different disorders lead to various localization and topological phenomena in condensed matter and artificial systems. Here we study the topological and localization properties in one-dimensional Su-Schrieffer-Heeger model with spatially correlated random binary disorders. It is found that random binary disorders can induce the topological Anderson insulating phase from the trivial insulator in various parameter regions. The topological Anderson insulators are characterized by the disorder-averaged winding number and localized bulk states revealed by the inverse participation ratio in both real and momentum spaces. We show that the topological phase boundaries are consistent with the analytical results of the self-consistent Born approach and the localization length of zero-energy modes, and discuss how the bimodal probability affects the disorder-induced topological phases. The topological characters can be detected from the mean chiral displacement in atomic or photonic systems. Our work provides an extension of the topological Anderson insulators to the case of correlated disorders.

preprint2026arXiv

The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models

Real-world datasets are inherently heterogeneous, yet how per-class structural differences and sampling imbalance shape the training dynamics of diffusion models-and potentially exacerbate disparities-remains poorly understood. While models typically transition from an initial phase of generalization to memorizing the training set, existing theory assumes homogeneous data, leaving open how class imbalance and heterogeneity reshape these dynamics. In this work, we develop a high-dimensional analytical framework to study class-dependent learning in score-based diffusion models. Analyzing a random-features model trained on Gaussian mixtures, we derive the feature-covariance spectrum to characterize per-class generalization and memorization times. We reveal the explicit hierarchy governing these dynamics: class variance is the primary determinant of learning order-consistently favoring higher-variance classes-while centroid geometry plays a secondary role. Sampling imbalance acts as a modulator that can reverse this ordering and, under strong imbalance, forces minority classes to acquire distinct, delayed speciation times during backward diffusion. Together, these results suggest that diffusion models can memorize some classes while others remain insufficiently learned. We validate our theoretical predictions empirically using U-Net models trained on Fashion MNIST.

preprint2024arXiv

The strain gap in a system of weakly and strongly interacting two-level systems

Many disordered lattices exhibit remarkable universality in their low temperature properties, similar to that found in amorphous solids. Recently a two-TLS (two-level system) model was derived based on the microscopic characteristics of disordered lattices. Within the two-TLS model the quantitative universality of phonon attenuation, and the energy scale of $1-3$ K below which universality is observed, are derived as a consequence of the existence of two types of TLSs, differing by their interaction with the phonon field. In this paper we calculate analytically and numerically the densities of states (DOS) of the weakly and strongly interacting TLSs. We find that the DOS of the former can be well described by a Gaussian function, whereas the DOS of the latter have a power law correlation gap at low energies, with an intriguing dependence of the power on the short distance cutoff of the interaction. Both behaviors are markedly different from the logarithmic gap exhibited by a single species of interacting TLSs. Our results support the notion that it is the weakly interacting $τ$-TLSs that dictate the standard low temperature glassy physics. Yet, the power-law DOS we find for the $S$-TLSs enables the prediction of a number of deviations from the universal glassy behavior that can be tested experimentally. Our results carry through to the analogous system of electronic and nuclear spins, implying that electronic spin flip rate is significantly reduced at temperatures smaller than the magnitude of the hyperfine interaction.

preprint2026arXiv

Sharp feature-learning transitions and Bayes-optimal neural scaling laws in extensive-width networks

We study the information-theoretic limits of learning a one-hidden-layer teacher network with hierarchical features from noisy queries, in the context of knowledge transfer to a smaller student model. We work in the high-dimensional regime where the teacher width $k$ scales linearly with the input dimension $d$ -- a setting that captures large-but-finite-width networks and has only recently become analytically tractable. Using a heuristic leave-one-out decoupling argument, validated numerically throughout, we derive asymptotically sharp characterizations of the Bayes-optimal generalization error and individual feature overlaps via a system of closed fixed-point equations. These equations reveal that feature learnability is governed by a sequence of sharp phase transitions: as data grows, teacher features become recoverable sequentially, each through a discontinuous jump in overlap. This sequential acquisition underlies a precise notion of \textit{effective width} $k_c$ -- the number of learnable features at a given data budget $n$ -- which unifies two distinct scaling regimes: a feature-learning regime in which the Bayes-optimal generalization error $\varepsilon^{\rm BO}$ scales as $ n^{1/(2β)-1}$, and a refinement regime in which it scales as $n^{-1}$, where $β>1/2$ is the exponent of the power-law feature hierarchy. Both laws collapse to the single relation $\varepsilon^{\rm BO}=Θ(k_c d/n)$. We further show empirically that a student trained with \textsc{Adam} near the effective width $k_c$ achieves these optimal scaling laws (up to a small algorithmic gap), and provide an information-theoretic account of the associated scaling in model size.

preprint2022arXiv

Properties of the non-Hermitian SSH model: role of PT-symmetry

The present work addresses the distinction between the topological properties of PT symmetric and non-PT symmetric scenarios for the non-Hermitian Su-Schrieffer-Heeger (SSH) model. The non-PT symmetric case is represented by non-reciprocity in both the inter- and the intra-cell hopping amplitudes, while the one with PT symmetry is modeled by a complex on-site staggered potential. In particular, we study the loci of the exceptional points, the winding numbers, band structures, and explore the breakdown of bulk-boundary correspondence (BBC). We further study the interplay of the dimerization strengths on the observables for these cases. The non-PT symmetric case denotes a more familiar situation, where the winding number abruptly changes by a half-integer through tuning of the non-reciprocity parameters, and demonstrates a complete breakdown of BBC, thereby showing the non-Hermitian skin effect. The topological nature of the PT symmetric case appears to follow closely to its Hermitian analogue, except that it shows unbroken (broken) regions with complex (purely real) energy spectra, while another variant of the winding number exhibits a continuous behavior as a function of the strength of the potential, while the conventional BBC is preserved.

preprint2022arXiv

Localisation of Dirac modes in gauge theories and Goldstone's theorem at finite temperature

I discuss the possible effects of a finite density of localised near-zero Dirac modes in the chiral limit of gauge theories with $N_f$ degenerate fermions. I focus in particular on the fate of the massless quasi-particle excitations predicted by the finite-temperature version of Goldstone's theorem, for which I provide an alternative and generalised proof based on a Euclidean ${\rm SU}(N_f)_A$ Ward-Takahashi identity. I show that localised near-zero modes can lead to a divergent pseudoscalar-pseudoscalar correlator that modifies this identity in the chiral limit. As a consequence, massless quasi-particle excitations can disappear from the spectrum of the theory in spite of a non-zero chiral condensate. Three different scenarios are possible, depending on the detailed behaviour in the chiral limit of the ratio of the mobility edge and the fermion mass, which I prove to be a renormalisation-group invariant quantity.

preprint2023arXiv

The structure of heavily doped impurity band in crystalline host

We study the properties of the impurity band in heavily-doped non-magnetic semiconductors using the Jacobi-Davidson algorithm and the supervised deep learning method. The disorder averaged inverse participation ratio (IPR) and thouless number calculation show us the rich structure inside the impurity band. A Convolutional Neural Network(CNN) model, which is trained to distinguish the extended/localized phase of the Anderson model with high accuracy, shows us the results in good agreement with the conventional approach. Together, we find that there are three mobility edges in the impurity band for a specific on-site impurity potential, which means the presence of the extended states while filling the impurity band.

preprint2020arXiv

Nucleation of ergodicity by a single mobile impurity in supercooled insulators

We consider a disordered Hubbard model, and show that, at sufficiently weak disorder, a single spin-down mobile impurity can thermalize an extensive initially localized system of spin-up particles. Thermalization is enabled by resonant processes which involve correlated hops of the impurity and localized particles. This effect indicates that certain localized insulators behave as "supercooled" systems, with mobile impurities acting as ergodic seeds. We provide analytical estimates, supported by numerical exact diagonalization (ED), showing how the critical disorder strength for such mechanism depends on the particle density of the localized system. In the $U\rightarrow\infty$ limit, doublons are stable excitations, and they can thermalize mesoscopic systems by a similar mechanism. The emergence of an additional conservation law leads to an eventual localization of doublons. Our predictions apply to fermionic and bosonic systems and are readily accessible in ongoing experiments simulating synthetic quantum lattices with tunable disorder.