https://arxiv.org/api/qQofJonDmC2WGvYS+V5PYjU4/L0 2026-06-10T08:58:10Z 26559 90 15 http://arxiv.org/abs/2606.02239v1 2026-06-01T13:32:57Z

How does a persistent random walker perceive a complex energy landscape? We address this question by studying the persistent Langevin dynamics of the p=3 spherical spin glass, a paradigmatic mean-field model with a scale-free hierarchical landscape. By tuning the persistence time tau_p, which controls the walker's inertia and effectively sets its energy resolution delta E proportional to 1/tau_p, we measure the energy correlation time tau_corr. At temperature T=1.0, we find tau_corr scales as tau_p to the power alpha with alpha = 0.337 plus or minus 0.035 for tau_p in the range [2,32] (for N=64), in excellent agreement with the Kardar-Parisi-Zhang (KPZ) universality class prediction alpha = 1/3. Finite-size scaling using N = 16,32,48,64,128 yields the thermodynamic limit alpha(infinity) = 0.3333 plus or minus 0.0134, fully consistent with 1/3. Thus, tau_p acts as a tunable probe that reveals the predicted scale-free hierarchy of the landscape. Moreover, the temperature dependence alpha(T) for T = 0.5,1.0,1.5,2.0 exhibits a clear U-shaped curve, identifying three dynamical regimes: ballistic/inertial, KPZ, and noise-dominated. Our results establish persistent Langevin dynamics as a powerful tool for uncovering hidden landscape topology and demonstrate that the p=3 spherical spin glass belongs to the KPZ universality class.

2026-06-01T13:32:57Z 7 pages, 2 figures Li Zhenpeng http://arxiv.org/abs/2606.02207v1 2026-06-01T13:07:50Z

Information scrambling is a hallmark of quantum chaos and thermalization in isolated quantum many-body systems. We investigate scrambling dynamics in the all-to-all interacting spin Sachdev-Ye-Kitaev (SYK)-$q$ model using both pure- and mixed-state entanglement measures. We show that von-Neumann and Rényi entropies exhibit rapid growth followed by saturation near Haar-random values, signaling efficient scrambling. The scrambling rate reveals a nontrivial dependence on the interaction order, system size, and Hamiltonian scaling. We further employ mixed-state entanglement as a powerful probe of information scrambling. We numerically find a universal relation between the Rényi-1/2 mutual information and entanglement negativity for minimal interaction order in the early growth regime. Furthermore, entanglement negativity displays a Page-curve-like behavior under unequal subsystem partitioning, characterized by the birth, spread, and eventual death of quantum correlations. Our results provide a generic description of information scrambling using entanglement dynamics in all-to-all interacting spin systems with multi-body interactions.

2026-06-01T13:07:50Z 8+16 pages, 4+15 figures Abhik Kumar Saha Tanay Pathak Masaki Tezuka http://arxiv.org/abs/2510.27507v2 2026-06-01T12:39:30Z

We study the stability of ground states in the Edwards-Anderson Ising spin glass in dimensions two and higher against perturbations of a single coupling. After reviewing the concepts of critical droplets, flexibilities and metastates, we show that, in any dimension, a certain kind of critical droplet with space-filling (i.e., positive spatial density) boundary does not exist in ground states generated by coupling-independent boundary conditions. Using this we show that if incongruent ground states exist in any dimension, the variance of their energy difference restricted to finite volumes scales proportionally to the volume. This in turn is used to prove that a metastate generated by (e.g.) periodic boundary conditions is unique and supported on a single pair of spin-reversed ground states in two dimensions. We further show that any excitation above an infinite-volume ground state with a positive-density interface must have the property that its energy difference with the ground state in a restricted volume diverges as the square root of the restricted volume.

2025-10-31T14:29:11Z 48 pages, 2 figures. Changes from v1: references added, Section 2 expanded, Section 3 in v1 reorganized and split into 2 sections (now Sections 3 and 4), Section 6.2 (Section 5.2 in v1) revised and expanded C. M. Newman D. L. Stein http://arxiv.org/abs/2606.02098v1 2026-06-01T11:29:30Z

We propose a quantum probability-flow principle for deriving local learning rules in associative memory. A transverse field defines leakage channels from data states, and minimizing the measured survival loss gives stability-driven updates. For imaginary-time, dephased dynamics, the local leakage free energy is the log-sum-exp of energy gaps; its gradient is a softmax-weighted Hebbian rule. Real-time stability instead yields a power-law weighting. D-Wave standard- and fast-anneal tests of a one-hot attention forward map are better fitted by an effective softmax than by a Lorentzian power law.

2026-06-01T11:29:30Z 4 pages Masayuki Ohzeki http://arxiv.org/abs/2606.02052v1 2026-06-01T10:40:15Z

We apply the single-equation duality criterion to the square-octagonal random-bond Ising model recently obtained from an exact error-edge map for a surface code with nearest-neighbor correlated errors. The calculation is performed for the minimal cell after the error-edge reduction. For the symmetric case \(p_1=p_2=p_3=p\), this gives \(p_c=0.0288427147\), in close agreement with the reported numerical threshold of about \(3\%\).

2026-06-01T10:40:15Z 2 pages, 1 figure Masayuki Ohzeki http://arxiv.org/abs/2602.23039v2 2026-06-01T09:29:09Z

Training large neural networks exposes neural scaling laws for the generalization error, which points to a universal behavior across network architectures of learning in high dimensions. It was also shown that this effect persists in the limit of highly overparametrized networks as well as the Neural network Gaussian process limit. We here develop a principled understanding of the typical behavior of generalization in Neural Network Gaussian process regression dynamics. We derive a dynamical mean-field theory that captures the typical case learning dynamics: This allows us to unify multiple existing regimes of learning studied in the current literature, namely Bayesian inference on Gaussian processes, gradient flow with or without weight-decay, and stochastic Langevin training dynamics. Employing tools from statistical physics, the unified framework we derive in either of these cases yields an effective description of the high-dimensional microscopic behavior of networks dynamics in terms of lower dimensional order parameters. We show that collective training dynamics may be separated into the dynamics of N independent eigenmodes, those evolution equations are only coupled through collective response functions and a common statistics of an effective, independent noise. Our approach allows us to quantitatively explain the dynamics of the generalization error by linking spectral and dynamical properties of learning on data with power law spectra, including phenomena such as neural scaling laws and the effect of early stopping.

2026-02-26T14:24:11Z Jakob Kramp Javed Lindner Moritz Helias http://arxiv.org/abs/2606.01593v1 2026-06-01T02:42:45Z

Recent experiments and simulations provide compelling evidence for the emergence of a non-phononic flat band in the dynamical structure factor of two- and three-dimensional amorphous solids. This feature has been suggested to be connected to the excess in the reduced vibrational density of states of glasses, commonly known as the boson peak, and displays several apparently universal characteristics. First, it is nearly dispersionless, with an energy close to the boson-peak frequency. Second, its intensity is negligible below a critical wave vector of the order of the first diffraction peak. Third, its reduced intensity exhibits a strong correlation with the static structure factor. Here, we revisit the resonant-coupling model, a single-mode harmonic realization of the soft-potential scenario in which acoustic phonons interact with single frequency quasi-localized vibrations. We show that this minimal framework naturally reproduces the main features of the observed flat band and clarifies its connection to the boson peak.

2026-06-01T02:42:45Z v1: comments welcome Matteo Baggioli Bingyu Cui http://arxiv.org/abs/2606.01501v1 2026-05-31T23:50:13Z

The Longest Increasing Subsequence problem - a classic combinatorial challenge with deep connections to statistical mechanics- exhibits a rich thermodynamic landscape. Introducing a temperature we identify two distinct energy scales: A Schottky-like crossover at T_cross and a condensation transition at T_cond, below which the number of maximum-length configurations becomes sub-exponential in system size. We also show that despite the existence of polynomial-time dynamic programming algorithms for the ground state, local Monte Carlo dynamics, after sudden quenches at low temperatures, become trapped in metastable state configurations displaying characteristic glassy signatures: two-step relaxation, persistent dynamical overlaps and aging. On the other hand, logarithmic annealing tracks equilibrium down to the ground state. These results establish that thermodynamic sparsity - not energetic barriers - can render local search dynamically intractable, positioning the LIS problem as a bridge between exactly solvable optimization and glassy spin-glass phenomenology.

2026-05-31T23:50:13Z 8 pages and 9 figures Silvio Franz Roberto Mulet http://arxiv.org/abs/2601.09037v2 2026-05-31T02:27:56Z

Probabilistic computers built from p-bits offer a promising path for combinatorial optimization, but the dense connectivity required by real-world problems scales poorly in hardware. Here, we address this through graph sparsification with auxiliary copy variables and demonstrate two fully on-chip parallel tempering solvers on an FPGA. Targeting MIMO detection, a dense, NP-hard problem central to wireless communications, we first fit 11 temperature replicas of a 128-node sparsified system (1,408 p-bits) on-chip and achieve bit error rates significantly below conventional linear detectors on $64 \times 64$ BPSK MIMO. We report complete end-to-end solution times of 3~ms per instance, including all loading, sampling, readout, and verification overheads. ASIC projections in 7~nm technology indicate 103~MHz operation at 285.8~mW, suggesting that massive parallelism across multiple chips could approach the throughput demands of next-generation wireless systems. Sparsification, however, introduces a sharp sensitivity to the copy-constraint strength $P$ that requires manual tuning. To eliminate this bottleneck, we utilize Two-Dimensional Parallel Tempering (2D-PT), which exchanges replicas across both temperature ($β$) and constraint ($P$) dimensions. On Sherrington--Kirkpatrick spin glasses, 2D-PT converges roughly $250\times$ faster than optimally tuned 1D-PT, and on $128 \times 128$ MIMO it reaches zero bit errors at high SNR where 1D-PT exhibits an error floor. We further validate 2D-PT entirely on-chip with 54 replicas (1,728 p-bits) on a $16 \times 16$ MIMO instance, where it tracks the maximum-likelihood bound in just 50 Monte Carlo steps -- $10\times$ fewer than 1D-PT -- at projected 111~MHz and 124~mW in 7~nm. Together, these results establish an on-chip p-bit architecture and a scalable, tuning-free algorithmic framework for dense combinatorial optimization.

2026-01-14T00:01:58Z MMHS and KC-C are equally contributing first authors M Mahmudul Hasan Sajeeb Kevin Callahan-Coray Corentin Delacour Sanjay Seshan Tathagata Srimani Kerem Y. Camsari http://arxiv.org/abs/2605.30185v2 2026-05-30T09:28:32Z

We investigate the random close packing density, $φ_\textrm{RCP}$, of polydisperse hard sphere systems using a theoretical framework based on the equilibrium model of crowding. We derive a closed-form solution for $φ_\textrm{RCP}$ in terms of the moments of the diameter distribution, enabling an analytical exploration of the effects of polydispersity ($δ$) and skewness ($S$) on packing density. For a binary mixture, it is possible to explore a broader range of dependence of $φ_\textrm{RCP}$ on $δ$ for a given $S$ or on $S$ for a given $δ$. We show that the dependencies of $φ_\textrm{RCP}$ on skewness for a variety of continuous distributions collapse onto a theoretical master curve obtained for the binary mixture case. By correcting the theory so that it obeys known exact limiting behaviours for extreme size asymmetry, our analytical predictions not only agree with previously obtained numerical results, but also predict previously unexplored regions of the $φ_\textrm{RCP}$ parameter space.

2026-05-28T16:30:20Z Missing refs. added Vinay Vaibhav Carmine Anzivino Alessio Zaccone http://arxiv.org/abs/2606.00326v1 2026-05-29T20:09:09Z

The spectral behaviour of the synaptic matrix, representing the neuronal connection strengths, is an important tool to analyze the stability and transient dynamics of a typical brain as well as its learning process and memory capacity. The complexity of the brain due to large number of neurons as well as underlying transient mechanisms e.g. homeostasis, seizure or synaptic plasticity can lead to networks with time-varying degree and type of sparsity. This renders an exact determination of the synaptic matrix not only technically difficult but also meaningless, leaving its statistical analysis as the best available theoretical approach. This motivates us to pursue a spectral analysis of the synaptic matrix models with different type of sparsity and thereby analyze latter's role on various aspects of network dynamics and stability. Our results have potential relevance for detemining the type of synaptic sparsity required to induce a specific brain function or desired transient mechanism e.g for pharmacological effects or physiological modulators.

2026-05-29T20:09:09Z 46 pages, 12 figures Mohd. Gayas Ansari Pragya Shukla http://arxiv.org/abs/2606.00286v1 2026-05-29T19:19:05Z

Quenched disorder in the environment can fundamentally alter transport dynamics in both active and passive systems. We explore how disordered arrays of filaments govern the distribution of intermittently moving particles which switch between diffusive and processive transport. Motivated by the mixed-polarity arrangements of parallel microtubules observed in mammalian dendrites, we show that such arrays tend to result in localization of particles at regions of convergent filament orientation. In the rapid attachment-detachment limit, the disordered system can be described by a noisy one-dimensional effective energy landscape, whose structure is approximated by a random walk. The depth and width of wells on this landscape are expressed as a function of the transport kinetics and system geometry. Localization is shown to be strongest at intermediate run-lengths, where biased transport persists long enough to sense the quenched filament polarity but not so long as to facilitate escape from local traps. These results demonstrate robust localization of particles moving on random filament networks, highlighting the emergent spatial organization that arises from an interplay of active transport and quenched disorder.

2026-05-29T19:19:05Z Owen Santoso Elena Koslover http://arxiv.org/abs/2510.21955v2 2026-05-29T18:55:41Z

In light of recent progress in the study of amorphous topological phases, we investigate the effects of structural disorder on the topological properties of a two-dimensional quantum spin Hall insulator modeled by the Bernevig-Hughes-Zhang Hamiltonian. Using a real-space formulation of the Z2 invariant for Dirac-type Hamiltonian, we map out the phase diagram as a function of disorder strength and the mass parameter. Our results reveal that under the influence of structural disorder, a system can either undergo a phase transition from a topologically non-trivial to a topologically trivial phase or from a trivial to non-trivial phase. Remarkably, in certain parameter regimes, the system exhibits a re-entrant behaviour: a topologically non-trivial phase in the perfect lattice undergoes a transition to a trivial state under the influence of weak disorder but re-emerges as the disorder strength is further increased. We corroborate these findings through analysis of the bulk-boundary correspondence and transport calculations.

2025-10-24T18:28:14Z Revised version, accepted for publication in Phys. Rev. B Ranadeep Roy Yuan-Ming Lu http://arxiv.org/abs/2605.21648v2 2026-05-29T18:47:12Z

We develop a mean-field theory of dropout as a perturbation of critical signal propagation at the edge of chaos, and show that it predicts a simple, no-cost change to standard practice: \emph{front-loaded} dropout schedules cut test loss by \(18\)--\(35\%\) over constant dropout in MLPs and Vision Transformers at fixed budget. The theoretical mechanism is that dropout shifts the perfect-alignment fixed point, making the depth scale for information propagation finite even at critical initialization. We derive critical and crossover scaling laws for correlation decay and establish that smooth activations and kinked, \relu{}-like activations constitute distinct universality classes, with different critical exponents and a universal two-parameter scaling collapse in detuning and dropout strength. The distinction traces to the analytic structure of the correlation map: smooth activations admit a Taylor expansion near perfect alignment, while kinked activations develop a branch point with universal non-analyticity. As a corollary, the framework yields saturated dropout profiles under fixed budget; a regularization-reach argument then selects front-loaded schedules, with accuracy gains as a consistent secondary effect. We also discuss how the same Gaussian-kernel structure extends the theory beyond MLPs toward CNNs and residual architectures.

2026-05-20T19:00:02Z Accepted at the 43rd International Conference on Machine Learning (ICML 2026). 36 pages, 11 figures. Camera-ready version Lucas Fernandez Sarmiento http://arxiv.org/abs/2601.20931v2 2026-05-29T16:46:20Z

Anderson localization is a phase transition between a "metallic phase", where wavefunctions are extended and delocalized in space, and an "insulating phase", where wavefunctions are completely localized. These transitions are driven by uncorrelated or quasiperiodic disorder, e.g., in the case of the Aubry-André model. Here, I consider a family of Hamiltonians that generalizes the Aubry-André model, obtained by replacing the position and momentum operators with an arbitrary pair of canonically conjugate operators. These models exhibit a hidden localization transition. The system transitions between phases where wavefunctions are either localized or delocalized with respect to the new canonically conjugate operators, acting as an insulator or metal in this rotated space. These canonically conjugate operators can be taken as a linear combination of position and momentum, corresponding to a "rotation" in the abstract space of canonical operators. In this case, the hidden localization transition is signaled by the simultaneous vanishing of both the inverse participation ratio (IPR) and the normalized participation ratio (NPR) in the position and momentum space in the thermodynamic limit. This identifies the emergence of multifractal states that are neither fully extensive nor localized on the lattice. Hence, the states exhibit a multifractal dimension at the hidden phase transition, while remaining extended (i.e., one-dimensional) in both momentum and position everywhere else in the parameter space. Surprisingly, I found that at the phase transition, this model Hamiltonian coincides with the lattice Hamiltonian of a massless Dirac fermion in a curved spacetime background, indicating an unexpected relation between localization transitions and analog gravity.

2026-01-28T19:00:01Z 16 pages, 12 figures, additional calculations and references Pasquale Marra