2,028 papers
Stochastic Thermodynamics for Autoregressive Generative Models: A Non-Markovian Perspective
Autoregressive generative models -- including Transformers, recurrent neural networks, classical Kalman filters, state space models, and Mamba -- all generate sequences by sampling each output from a deterministic summary of the past, producing genuinely non-Markovian observed processes. We develop a general theoretical framework based on stochastic thermodynamics for this class of architectures and introduce the entropy production, which can be efficiently estimated from sampled trajectories without exponential sampling cost despite the non-Markovian nature of the observed dynamics. As a proof-of-concept experiment for a large language model (LLM), we evaluate the token-level and sentence-level entropy production for a pre-trained Transformer-based model, GPT-2. We also demonstrate the framework in the linear Gaussian case, where the model reduces to the Kalman innovation representation and the entropy production admits an analytical expression. We further show that the entropy production decomposes exactly into non-negative per-step contributions in terms of retrospective inference, and each of those terms further splits into information-theoretically meaningful terms: a compression loss and a model mismatch. Our results establish a bridge between stochastic thermodynamics and modern generative models, and provide a starting point for quantifying irreversibility in a broad class of highly non-Markovian processes such as LLMs.
2604.07867Apr 2026
View
Boltzmann-Loschmidt dispute reloaded quantum 150 years later
The Boltzmann-Loschmidt dispute of 1876 questioned the possibility of a statistical irreversible description by time reversible classical equations of motion of atoms. Here we show analytically and numerically that the quantum chaos diffusion of cold atoms, or ions, in a harmonic trap and pulsed optical lattice can be inverted back in time with up to 100\% efficiency. This is in sharp contrast to classical evolution where exponentially small errors break time reversibility. We argue that the existing experimental skills allow highlighting the Boltzmann-Loschmidt dispute from a quantum perspective.
2604.04879Apr 2026
View
Analytical approach to subsystem resetting in generalized Kuramoto models
Stochastic resetting has emerged as a powerful mechanism for driving systems into nonequilibrium stationary states with tunable properties. While most existing studies focus on global resetting, where all degrees of freedom are simultaneously reset, recent work has shown that resetting only a subset of degrees of freedom (subsystem resetting) can qualitatively alter collective behavior in interacting many-body systems. In this work, we develop a general theoretical framework for analysing subsystem resetting in Kuramoto-type coupled-oscillator systems. Building on a continued-fraction approach, we derive self-consistent equations for the stationary-state order parameter of the non-reset subsystem, applicable to both noisy and noiseless dynamics and to models with arbitrary interaction harmonics. Using this framework, we systematically investigate how the stationary state and phase transitions depend on the resetting rate, the size of the reset subsystem, and the reset configuration. We show that subsystem resetting can shift or even suppress synchronization transitions, and can give rise to nontrivial features such as re-entrant behavior and restructuring of phase boundaries. In specific cases, including the noiseless Kuramoto model with a Lorentzian frequency distribution, our results recover known analytical predictions and extend them to more general settings. These results establish subsystem resetting as a versatile control protocol for engineering collective dynamics in nonequilibrium interacting systems.
2604.04769Apr 2026
View
Cyclic Heat Engine with the Ising model: role of interactions and criticality
Heat engines that convert thermal energy into work are a cornerstone of classical thermodynamics and remain an active area of contemporary research. Notable examples include microscopic heat engines, trade-off relations between power and efficiency, and the attainability of Carnot efficiency at finite power. We propose a cyclic heat engine based on the Ising model, in which the thermodynamic cycle involves variations of both temperature and magnetic field. We analyze the one-dimensional and mean-field Ising models, which allow for simple analytical results and provide new insight into the role of interactions in cyclic heat engines. In particular, we show that interactions can enhance both power and efficiency. Moreover, a system that does not operate as an engine in the absence of interactions can become an engine upon tuning the interaction strength. The mean-field model enables us to investigate the relevance of the phase transition for the performance of this Ising heat engine. Owing to the emergence of spontaneous magnetization, the mean-field model can still operate as an engine even when one of the magnetic fields is set to zero. Remarkably, when the work is maximized, we find that the optimal parameters are numerically consistent with this regime, in which one magnetic field vanishes and the cycle explores the phase transition. We also consider an alternative cycle for the mean-field model, obtained by varying the interaction strength while keeping both temperatures below the critical temperature and setting the magnetic field to zero throughout the cycle. The power and efficiency of this cycle are analyzed as well. Finally, while our analytical results are valid for the limit of large period we use numerical simulations for finite periods and show that the power decreases monotonically with the period.
2604.04730Apr 2026
View
Effective Bethe Ansatz for Spin-1 Non-integrable Models
This work presents a comprehensive benchmark and validation of a recently proposed method called Effective Bethe Ansatz (EBA). It is a variational method that deforms the exact Bethe wavefunctions of one-dimensional spin chains at integrable points to approximate non-integrable systems. We apply this method to the non-integrable regime of the spin-1 bilinear-biquadratic chain. By performing EBA method starting from the two integrable endpoints, the Takhtajan-Babujian point and the Lai-Sutherland point, we systematically evaluate the accuracy of the EBA for the ground state and first excited state. Our validation is based on a direct comparison with exact diagonalization, assessing energy, fidelity, and entanglement entropy. The results confirm that the EBA provides a physically accurate description near integrability, with fidelity decreasing controllably as the perturbation increases. The method successfully captures key finite-size effects, such as level crossings, manifested as sharp drops in fidelity, and provides a probe to potential phase transitions. This study establishes the EBA as a reliable and efficient semi-analytical tool, clarifying its scope and limitations for studying low-energy physics in non-integrable quantum spin chains.
2604.04640Apr 2026
View
The Roaming Bethe Roots: An Effective Bethe Ansatz Beyond Integrability
We propose an effective Bethe ansatz for solving quantum many-body systems near an integrable point. Our approach retains the functional form of the Bethe wave function while renormalizing the Bethe roots to account for integrability-breaking interactions. These effective roots are determined by minimizing physically motivated cost functions. The resulting off-shell Bethe states serve as approximate eigenstates of the non-integrable models. We assess the quality of the approximation using various physical observables, including the energy eigenvalue, state fidelity, and bipartite entanglement entropy. Our tests show that for models with weak integrability-breaking, the effective Bethe ansatz provides a high-quality approximation to the exact eigenstates over a wide range of deformation parameters. In contrast, for models with strong integrability-breaking interactions, the efficacy of the effective Bethe ansatz degrades relatively quickly as the deformation parameter increases. The efficacy of the method thus offers a useful probe for characterizing the strength of integrability breaking. Within its regime of accuracy, it also provides a new representation of the eigenstates of nearly integrable models, enabling one to exploit the algebraic structure inherited from integrability.
2604.04627Apr 2026
View
Unified geometric formalism for dissipation and its fluctuations in finite-time microscopic heat engines
Microscopic heat engines operate in regimes where thermodynamic quantities fluctuate strongly, making stochastic effects an essential aspect of their performance. However, existing geometric formulations of finite-time thermodynamics primarily characterize average dissipation and do not systematically capture its fluctuations. Here, we develop a unified geometric framework that consistently describes both the mean dissipated availability and its fluctuations. In the linear-response regime, we show that these quantities are governed by metric tensors constructed from equilibrium correlation functions, providing a common geometric structure for dissipation and its fluctuations. This framework yields geometric bounds on both the mean and variance of the dissipated availability, and thereby on the efficiency and its fluctuations. The formalism applies broadly to stochastic systems, including Markov jump processes and overdamped and underdamped Brownian dynamics, establishing a unified geometric description across microscopic heat engines.
2604.04620Apr 2026
View
Semi-Markovian Dynamics of a Self-Propelled Particle in a Confined Environment: A Large-Deviation Study
We study the large deviations of the time-integrated current for a self-propelled particle moving within a confined environment. The dynamics is modeled as a semi-Markovian process, where the transitions between a \textit{normal running phase} (Phase $0$) and a \textit{wall-attached phase} (Phase $1$) are governed by time-dependent reset probabilities. We study two different examples: In the first case, the particle undergoes a biased random walk in Phase $0$, while it intermittently resets and interacts with the container boundaries, remaining stationary in Phase $1$. In this scenario, the reset probabilities for transitions between the two phases follow an ``aging'' logic. In the second case, the particle alternates between two active phases: a Markovian Phase $0$ characterized by memoryless, downstream-biased motion, and a semi-Markovian Phase $1$ with a reversed, upstream bias representing boundary-attached navigation. Here, we assume a time-independent survival probability in Phase $0$ and a time-dependent one in Phase $1$. By analyzing the Scaled Cumulant Generating Function (SCGF) in the long-time limit, we derive the conditions for Dynamical Phase Transition (DPT)s in the fluctuations of the particle velocity. We demonstrate that, depending on the aging strength, the system exhibits either discontinuous (first-order) or continuous (second-order) DPTs. Analytical predictions are validated via computer simulations.
2604.04595Apr 2026
View
Non-Equilibrium Stochastic Dynamics as a Unified Framework for Insight and Repetitive Learning: A Kramers Escape Approach to Continual Learning
Continual learning in artificial neural networks is fundamentally limited by the stability--plasticity dilemma: systems that retain prior knowledge tend to resist acquiring new knowledge, and vice versa. Existing approaches, most notably elastic weight consolidation~(EWC), address this empirically without a physical account of why plasticity eventually collapses as tasks accumulate. Separately, the distinction between sudden insight and gradual skill acquisition through repetitive practice has lacked a unified theoretical description. Here, we show that both problems admit a common resolution within non-equilibrium statistical physics. We model the state of a learning system as a particle evolving under Langevin dynamics on a double-well energy landscape, with the noise amplitude governed by a time-dependent effective temperature $T(t)$. The probability density obeys a Fokker--Planck equation, and transitions between metastable states are governed by the Kramers escape rate $k = (ω_0ω_b/2π)\,e^{-ΔE/T}$. We make two contributions. First, we identify the EWC penalty term as an energy barrier whose height grows linearly with the number of accumulated tasks, yielding an exponential collapse of the transition rate predicted analytically and confirmed numerically. Second, we show that insight and repetitive learning correspond to two qualitatively distinct temperature protocols within the same Fokker--Planck equation: insight events produce transient spikes in $T(t)$ that drive rapid barrier crossing, whereas repetitive practice operates at a modestly elevated but fixed temperature, achieving transitions through sustained stochastic diffusion. These results establish a physically grounded framework for understanding plasticity and its failure in continual learning systems, and suggest principled design criteria for adaptive noise schedules in artificial intelligence.
2604.04154Apr 2026
View
Statistics of Matrix Elements of Operators in a Disorder-Free SYK model
Recently, studies have explored the statistics of matrix elements of local operators in the Lieb-Liniger model. It was found that the probability distribution function for off-diagonal matrix elements $\langle \boldsymbolμ|\mathcal{O}|\boldsymbolλ \rangle$ within the same macro-state is well described by the Fréchet distributions. This represents a significant development for the Eigenstate Thermalization Hypothesis (ETH). In this paper, we investigate a similar phenomenon in another solvable model: the disorder-free Sachdev-Ye-Kitaev (SYK) model. The Hamiltonian of this model consists of 4-body interactions of Majorana fermions. Unlike the conventional SYK model, the coupling strengths in this model are fixed to a constant, earning it the name ``disorder-free.'' We evaluate the matrix elements of operators constructed from products of $n$ Majorana fermions: $\mathcal{O} = χ_{a_1}χ_{a_2}\ldots χ_{a_n}$. For a general choice of indices and $n \geq 4$, we find that the statistics of the off-diagonal matrix elements are well-fitted by a generalized inverse Gaussian distribution rather than Fréchet distributions.
2604.03977Apr 2026
View
A molecular dynamics simulation of thermalization of crystalline lattice with harmonic interaction
Understanding the realization of thermal equilibrium through the thermalization process in a many-body system is a fundamental and complex scientific question, bridging thermodynamics and classical dynamics and connecting to a host of physical phenomena, such as mechanical instabilities in a thermal environment. In this work, based on the harmonic lattice model, we investigate the thermalization process in both velocity and coordinate spaces, by examining microscopic dynamics on the atomic level. We show the distinct relaxation rates of the transverse and longitudinal components of the velocity, reveal the power law governing the nonlinear proliferation of dominant frequencies, and observe the concurrent rapid proliferations of frequencies and topological defects. We also show that the lattice system's persistent out-of-plane deformations exhibit two-stage fluctuation behaviors, characterized by distinct power laws of fractional exponents and associated with the broken up-down symmetry. This work demonstrates the rich dynamics underlying the thermalization process, and advances our understanding on the dynamical adaptations of many-body systems to external disturbances.
2604.03913Apr 2026
View
Cross Spectra Break the Single-Channel Impossibility
Lucente et al. proved that no time-irreversibility measure can detect departure from equilibrium in a scalar Gaussian time series from a linear system. We show that a second observed channel sharing the same hidden driver overcomes this impossibility: the cross-spectral block, structurally inaccessible to any single-channel measure, provides qualitatively new detectability. Under the diagonal null hypothesis, the cross-spectral detectability coefficient $\Scross$ (the leading quartic-order cross contribution) is \emph{exactly} independent of the observed timescales -- a cancellation governed solely by hidden-mode parameters -- and remains strictly positive at exact timescale coalescence, where all single-channel measures vanish. The mechanism is geometric: the cross spectrum occupies the off-diagonal subspace of the spectral matrix, orthogonal to any diagonal null and therefore invisible in any single-channel reduction. For the one-way coupled Ornstein--Uhlenbeck counterpart, the entropy production rate (EPR) satisfies $\EPRtot=α_2λ^2$ exactly; under this coupling geometry, $\Scross>0$ certifies $\EPRtot>0$, linking observable cross-spectral structure to full-system dissipation via $\EPRtot^{\,2}\propto\Scross$. Finite-sample simulations predict a quantitative detection-threshold split testable with dual colloidal probes and multisite climate stations.
2604.03775Apr 2026
View
Geometry- and topology-controlled synchronization phase transition on manifolds
In this work, we explore how the geometry and topology of the underlying manifold shape the synchronization phase transition of a system. To do so, we extend the Kuramoto-Sakaguchi model from spheres to compact, connected, orientable, and homogeneous Riemannian manifolds of arbitrary dimension. Starting from the mean-field kinetic equation on the manifold, we derive a local response equation for the order parameter near the incoherent state and separate the geometric and topological contributions to the phase transition out of the incoherent state. The manifold geometry determines a coefficient $κ\left(M\right)$ to control the critical coupling for the linear loss of stability of the incoherent state. The manifold topology constrains the cubic term of the response equation through the Euler characteristic $χ\left(M\right)$. Under a local sign condition on the cubic term, topology does not allow a generic continuous or tricritical synchronization phase transition to occur when $χ\left(M\right)\neq 0$, and it imposes a non-zero net defect charge on the incipient ordered texture. When an additional local stabilization condition holds in that nonzero-Euler class, topology further selects a discontinuous phase transition. When $χ\left(M\right)=0$, topology does not impose that obstruction, so continuous, discontinuous, and tricritical local branches are all allowed. We verify these findings on representative families including hyperspheres, equal even-sphere products, complex Grassmannians, complex projective spaces, flat tori, real Stiefel manifolds, rotation groups, and unitary groups. Our framework recovers the classical hyperspherical parity law and extends it to a broad class of non-spherical state spaces.
2604.03770Apr 2026
View
Description of KPZ interface growth by stochastic Loewner evolution
In this study, we investigate the relationship between the one-dimensional (1D) Kardar-Parisi-Zhang (KPZ) equation and the stochastic Loewner equation (SLE), which is a one parameter family of the conformal mappings involving stochasticity. The author shows the correspondence between 1D KPZ equation with height function $h(x,t)=(3t^2x+x^3)/6t$ and Loewner equation driven by a nonlinear stochastic process, wherein the 1D dynamics of interface growth is characterized by Loewner entropy $S_{Loew}\simeq-\ln{t/κ}$. These results were numerically verified with discussions in relation to the universality in non-equilibrium statistical physics.
2604.03711Apr 2026
View
Random matrix theory of integrability-to-chaos transition
The statistics of gaps between quantum energy levels is a hallmark criterion in quantum chaos and quantum integrability studies. The relevant distributions corresponding to exactly integrable vs. fully chaotic systems are universal and described by the Poisson vs. Wigner-Dyson curves. In the transitional regime between integrability and chaos, the distributions are much less universal and have not been understood quantitatively until now. We point out that the relevant statistics that controls these distributions is that of the matrix elements of the nonintegrable perturbation Hamiltonian in the energy eigenbasis of the unperturbed integrable system. With this insight, we formulate a simple random matrix ensemble that correctly reproduces the level spacing distributions in a variety of test systems. For the distribution of matrix elements appearing in our construction, we furthermore discover surprising universal features: across a variety of physical systems with diverse degrees of freedom, these distributions are dominated by simple power laws.
2604.03669Apr 2026
View
Non-reciprocal Ising gauge theory
Non-reciprocity and geometric frustration enable many-body systems to avoid crystalline order and instead exhibit complex, liquid-like behavior. Here we show that their interplay is richer than the sum of its parts, leading to surprising structural and dynamical phenomena. In our minimal model, two copies of Ising gauge theory are non-reciprocally coupled in a way that crucially preserves a local $\mathbb{Z}_2$ symmetry. We discover that the combined Wilson loop observable of the two copies exhibits linear asymptotic scaling, with a quasiparticle-pair confinement length tuned by the strength of the non-reciprocal coupling. Key dynamical features are revealed in the behavior of individual deconfined excitations due to strong interactions induced by the non-reciprocity, leading to motion on a critical percolation cluster that follows a self-avoiding trail. Mapping from this quasiparticle dynamics onto the magnetic noise spectrum, we discover that non-reciprocity tunes topological logarithmic contributions and causes long-lived metastable states due to quasiparticle trapping. Our work opens the way for broader investigations of geometrically frustrated non-reciprocity.
2604.03367Apr 2026
View
Testing the Role of Diagonal Interactions in High-Order Hopfield Models via Dynamical Mean-Field Theory
High-order extensions of the Hopfield model are known to exhibit dramatically enhanced storage capacity at equilibrium, while their dynamical retrieval properties remain less well understood. In our previous work, we carried out a dynamical mean-field theory (DMFT) analysis of the Krotov--Hopfield-type dense associative memory and found that the transition between successful and failed retrieval is accompanied by pronounced slow dynamics. As a consequence, the effective basin of attraction observed in numerical simulations extends well beyond that predicted by equilibrium statistical mechanics. A natural hypothesis is that this discrepancy originates from diagonal (self-interaction) contributions in the Krotov--Hopfield model, which generate a large number of lower-order interaction terms and may induce glassy relaxation near the retrieval boundary. To test this hypothesis, we analyze an alternative high-order associative memory model, namely the Abbott--Arian-type $p$-body Hopfield model, in which such diagonal contributions are absent by construction. Using dynamical mean-field theory, we derive an effective single-site process together with closed macroscopic equations governing the retrieval dynamics. Our analysis reveals that both slow dynamics and a substantial enlargement of the apparent basin of attraction persist even in this model. These results indicate that the dynamical slowdown near the retrieval boundary cannot be attributed primarily to diagonal self-interaction effects, but instead originates from intrinsic properties of high-order interactions.
2604.03115Apr 2026
View
Resetting dynamics in a system with quenched disorder
Although resetting has widespread applicability, applying it to the dynamics in the presence of spatial quenched disorder, which is essential in many physical problems, is challenging. In this study, we consider a well-known one-dimensional model of particle hopping on a lattice with quenched disorder in the form of site-dependent hopping probabilities, drawn from a power-law distribution, and apply the resetting formalism. As a physical example, we recast the growth dynamics of microtubules with sudden catastrophic disassembly events as a resetting dynamics. We consider two distinct regimes for growth dynamics: a strongly biased case and a less biased case. Motivated by experimental results, we take a Gamma distribution for the resetting time. Our results show that occasional disassembly events are crucial for the experimentally observed distribution of reset (or catastrophe) lengths. We also analyze steady-state distributions under different resetting protocols-resetting to the initial position versus a random site. We also investigate the distribution of first-passage times to a fixed distance following reset. Finally, by considering other resetting probability distributions, we identify a regime where the mean displacement grows as slowly as $\log^2 t$. We also elucidate the role of disorder in the system properties under the resetting dynamics. Our study paves the way to treat the dynamics of complex physical systems using resetting.
2604.02950Apr 2026
View
Hamiltonian flocks: Time-Reversal Symmetry and its consequences
The fluctuation-dissipation theorem is a hallmark of equilibrium system that stem from their time-reversal symmetry. In many non-equilibrium systems, in particular active ones, extensions and explicit violations of this theorem are used to assess their ''distance'' to equilibrium. In Hamiltonian flocks, conservative yet non-Galilean models of polar liquids, previous work reported collective motion without the activity that usually underlies it. In this paper, we show that this model obeys a generalized time-reversal symmetry that yields a fluctuation-dissipation theorem that mixes position and polarity degrees of freedom. Due to the oddness of spin under time reversal, the system also obeys Onsager-Casimir reciprocity rather than standard Onsager relations. The coupling also induces rich spin orientation dynamics, including a non-trivial diffusion constant at long times. Finally, we show that considering the naïve time-reversal operation rather than the generalized one that leaves the system invariant leads to a spurious entropy production rate, that could be wrongly interpreted as a distance to equilibrium. Our findings suggest looking for possible extensions of time-reversal symmetry in active-looking systems, which may lead to yet unknown generalizations of the fluctuation-dissipation theorem.
2604.02914Apr 2026
View
Structure Functions and Intermittency for Coarsening Systems
In studies of turbulence, there has been extensive use of physical quantities such as {\it energy transfers} and {\it structure functions}. We examine whether these quantities can be useful in understanding problems of domain growth or coarsening, as modeled by the {\it time-dependent Ginzburg-Landau} (TDGL) equation and the {\it Cahn-Hilliard} (CH) equation. This paper has two major themes. First, we review our recent papers on energy transfers in domain growth. Second, we study structure functions and intermittency for coarsening systems. As a consequence of sharp interfaces, the structure functions scale as $S_q \sim r^{ζ_q}$, where $r$ is the distance between two points. For the TDGL and CH models, $ζ_q = 1$, indicating {\it anomalous scaling}
2604.02855Apr 2026
ViewPage 1 of 102
