nlin updates on arXiv.org
nlin updates on the arXiv.org e-print archive.
Physics-Inspired Optimizers beyond the Edge of Stability: Period-Two Orbits and Delayed Instability
oai:arXiv.org:2610.27321v1
arXiv:2610.27321v1 Announce Type: new
Abstract: Gradient descent with momentum can be viewed as a particle moving with friction in the loss landscape. In deep learning the learning rate is usually large, so training is a discrete dynamical system: in many full-batch experiments the sharpness grows to the optimizer's stability limit, the edge of stability (EoS), and training continues there. Physics-inspired optimizers change this motion; Velocity-Regularized Adam (VRAdam), for example, shrinks the step at high speed. We study kinetic maps, in which the step is the gradient of a kinetic energy of the momentum, and ask what they do beyond the EoS. (i) In a reduced model, a kinetic map creates a symmetric period-two orbit beyond the edge, strictly linearly stable if and only if the map's elasticity (the log-log slope of step length versus momentum) lies between 0 and 1; on it, training sits at an effective edge of stability. VRAdam's damper fails above twice the critical learning rate, where its kinetic energy is not convex; the Legendre dual of VRAdam's own Lagrangian is convex and passes at every learning rate. (ii) Because the sharpness rises gradually, the loss of stability is delayed, as in Neishtadt's theory of slowly swept bifurcations. In a network with a controlled sweep, the dynamics match this theory without fitted parameters, including the memory of where the sweep started and its erasure by noise. (iii) With momentum, reaching the oscillation also requires elasticity at large momentum: maps that bound the step have competing period-four and chaotic attractors already at beta1 = 0.9, while the cubic-Legendre map reached the oscillation from every sampled start up to beta1 = 0.99. Experiments on an exact quadratic problem, a full-batch MLP and ResNet-32 support the theory; on ResNet-18, much of the apparent benefit of the maps is reproduced by matching the effective learning rate.
Multifractality of human upper-limb velocity tracks motor optimization via hand dominance experiment
oai:arXiv.org:2610.27399v1
arXiv:2610.27399v1 Announce Type: new
Abstract: Theoretical goal-directed movements follow a unique, stereotyped velocity profile, but continuous movement has no unique optimal solution, so it is unclear whether motor optimization makes continuous movement more stereotyped or more diverse. Previous studies indicated higher information content in the dominant hand, suggesting greater diversity. Because continuous velocity is a composition of overlapping joints movement of varying form, timing and amplitude, comparable to the events that compose spike trains and heartbeat series, we asked whether this diversity appears as a wider multifractal spectrum of velocity. We applied multifractal detrended fluctuation analysis to motion-capture velocity of cursive handwriting from right- and left-handed participants writing with both hands; hand dominance is a natural experiment in which task and geometry are fixed while optimization varies. With the dominant hand, spectra were wider, had a lower dominant exponent and were more positively asymmetric in both groups, whereas hands with the same functional role did not differ between groups. A logistic regression on the three spectral features identified the dominant hand with a cross-validated AUC of up to 0.93. Optimized continuous movement is thus not more stereotyped but more diverse: complex, neither stereotyped nor random, with multifractality that tracks hand dominance.
The ABC-equation revisited: missed (quasi) integrable cases. Part II: extensions and actions
oai:arXiv.org:2610.28009v1
arXiv:2610.28009v1 Announce Type: new
Abstract: We continue here the study of the \textsc{ABC}-equation
$Au_xu_{yt} + Bu_yu_{xt} + Cu_tu_{xy} = 0$, $A$, $B$, $C\in\mathbb{R}$ and
using the results obtained in the first part construct new quasi-integrable
structures and compute their actions on symmetries and cosymmetries.
Painlev\'e II connection problem revisited
oai:arXiv.org:2610.28227v1
arXiv:2610.28227v1 Announce Type: new
Abstract: We study the inhomogeneous Painlev\'e II equation for generic monodromy data. Using Kapaev's asymptotic formulas, the parameters of Zak transform representations of the Painlev\'e II tau function on six special rays $\arg t=\frac{\pi p}{3}$ ($p=0,\ldots,5$) in the complex time plane are expressed in terms of Stokes parameters of the associated linear system. We then solve the connection problem for the tau function between these special rays, and provide analytical and numerical checks of our result.
A twisted Marchenko lemma and applications
oai:arXiv.org:2610.28233v1
arXiv:2610.28233v1 Announce Type: new
Abstract: We develop a twisted Marchenko lemma with applications into non-commutative two-component integrable systems. Starting from a twisted differential equation, we introduce the corresponding two-component Wronski matrix $W_{n,m}(\boldsymbol\phi,\boldsymbol\psi)$. Under the Marchenko lemma conditions, the logarithmic derivative $(\partial W_{n,m})W_{n,m}^{-1}$ satisfies nonlinear equations, whose two-component projections yield a non-commutative two-component KP equation, while a single-component reduction recovers the non-commutative Burgers equation, KP equation and DS equation. A direct approach based on two-component quasi-Wronskians is developed, from which a higher-order non-commutative two-component KP equation is generated with matrix-valued solutions. Reductions of the seed functions of the non-commutative two-component KP hierarchy yield the one-component counterpart, while a stepwise reduction of those for the two-component KP equation gives the non-commutative DS and NLS equations.
Toda chain solves large BKP
oai:arXiv.org:2610.28492v1
arXiv:2610.28492v1 Announce Type: new
Abstract: We prove that any solution of the Toda chain hierarchy (the 1D reduction of the 2D Toda lattice) is a solution of the large BKP hierarchy. This is a difference analogue of Alexandrov's result that KdV solves (small) BKP.
Boundary One-Point Functions in the Open XXZ Spin Chain with Generic Integrable Boundaries
oai:arXiv.org:2610.27062v1
arXiv:2610.27062v1 Announce Type: cross
Abstract: We consider the problem of computing observables beyond the spectrum in the open XXZ spin chain with the most generic integrable boundary conditions. For such boundaries, the residual u(1) symmetry is broken and the familiar Baxter TQ equation describing the spectrum becomes inhomogeneous, placing the model outside the realm of existing approaches to correlation functions and form factors. We show that the Functional Separation of Variables method, so far developed mainly for rational, (quasi-)periodic spin chains, is largely insensitive to these modifications, and works equally well for diagonal and generic boundaries. Using this method, we compute exact determinant expressions for a prototypical example of form factors in open-boundary systems, namely the longitudinal and transverse boundary-spin one-point functions in arbitrary Hamiltonian eigenstates at finite system size. No constraint relating the two boundary configurations and no reference state are required. We verify the resulting formulae against direct finite-volume diagonalisation.
Connected $\omega$-limit sets of Cournot maps: periodic orbits, realization and almost every point
oai:arXiv.org:2610.27229v1
arXiv:2610.27229v1 Announce Type: cross
Abstract: Let $F(x,y)=(f_2(y),f_1(x))$ be a Cournot map of the unit square, $\varphi=f_2\circ f_1$ and $\chi=f_1\circ f_2$. Linero Bas and Mu\~noz Guillermo proved that an $\omega$-limit set of $F$ with nonempty interior is connected if and only if \textup{(C1)} its two projections are connected and \textup{(C2)} $\omega_{F^2}(z)\cap\omega_{F^2}(F(z))\neq\emptyset$, and conjectured that this holds for every $\omega$-limit set. We prove the conjecture for every point $z=(x,y)$ such that $\omega_\varphi(x)$ or $\omega_\chi(y)$ is finite, and for Lebesgue almost every point $z$ of the square whenever $\varphi$ is metrically regular (every metric attractor is a periodic orbit, a solenoid, or a cycle of intervals carrying a well-behaved absolutely continuous invariant measure). Over a periodic orbit the $\omega$-limit set is a \emph{comb}, whose sections are rigid; when one projection is connected it is a point, a \emph{lattice} $(P\times J)\cup(K\times Q)$ with $P\subset K$, $Q\subset J$ and $\#P=\#Q$, or an \emph{alternating comb} of $2m$ disjoint segments. Conversely, every lattice and every alternating comb is an $\omega$-limit set of a Cournot map with piecewise linear reaction functions; the construction rests on a lemma producing exact compositions of piecewise linear maps with prescribed values. This answers, for this class, the question of Balibrea, Linero and C\'anovas of which nowhere dense sets are $\omega$-limit sets of Cournot maps. For instance, the letter H is not an $\omega$-limit set at any such point, and for metrically regular $\varphi$ the points whose $\omega$-limit set is connected, infinite, nowhere dense and not a lattice form a Lebesgue null set.
Does a Flux-Defined Markov Blanket Screen Self-Replicating Gray--Scott Spots? A Computational Test of a Blanket Condition from the Free Energy Principle
oai:arXiv.org:2610.27235v1
arXiv:2610.27235v1 Announce Type: cross
Abstract: Friston's free energy principle holds that a system separated from its environment by a Markov blanket---a boundary whose states statistically screen its interior from its exterior---can be treated as performing approximate Bayesian inference, and that such blankets arise wherever coupling is short-range. We test this blanket condition in a concrete nonlinear pattern-forming system: self-replicating spots in a reaction-diffusion model that persist, compete, and reproduce. Directed boundary exchange and timescale separation hold up reasonably well, with qualifications. A flux-defined blanket does not screen interior from exterior. Flux removes only part of the linear dependence, and a nonlinear test shows that conditioning on flux \emph{increases} the dependence, consistent with conditioning on a collider, since flux is a joint function of concentrations on both sides of the boundary. The dependence is concentrated within about one organism radius of the boundary. A band of the surrounding field that thick screens the interior from the exterior beyond it, as any short-range field would, which is not evidence of inference.
Pfaffian solutions to the Ernst equation
oai:arXiv.org:2610.27420v1
arXiv:2610.27420v1 Announce Type: cross
Abstract: We show that the Yamazaki-Hori solution to the Ernst equation can be represented in the Pfaffian form. We also discuss the consequences of this result for Nakamura's conjecture on the relation between the Ernst equation and the 2D Toda model.
Soliton Profiles in 1D: Enhanced Neural-Network Solvers and Constrained Deep Ritz Methods
oai:arXiv.org:2610.27500v1
arXiv:2610.27500v1 Announce Type: cross
Abstract: We develop and compare enhanced neural-network solvers for one-dimensional ground-state profiles, which are solutions of $$ Q^{\prime\prime}=bQ-\gamma Q^{p},\quad Q>0,\quad p>1, \quad b, \gamma >0, $$ the stationary profile equation for the nonlinear Schr\"odinger, generalized Korteweg-de Vries, and nonlinear Klein-Gordon equations. Its explicit sech-type solutions serve as exact reference for error computation and benchmark hyperparameter selection. We improve on the standard Adam-trained physics-informed neural network (PINN) baseline by introducing higher-order optimization via Gauss-Newton/Levenberg-Marquardt and L-BFGS refinements, as well as Fourier-feature PINNs, and constrained Deep Ritz (DR) formulations. In DR approaches we introduce two new neural-network formulations, which implement two classical variational characterizations: a Nehari-constrained Deep Ritz method, in which the height is learned by projecting onto the Nehari/Pokhozhaev manifold, and a Weinstein Deep Ritz method, in which a normalized profile minimizes the Weinstein quotient. The Deep Ritz formulations perform best overall, in particular, the Weinstein variant gives the best error-runtime balance. The neural-network based solvers remain less accurate and efficient than classical Petviashvili iteration in 1D, however, the current study provides a proof of concept: the mesh-free formulation, non-reliance on exact profiles or the related features make them promising in higher dimensions and for more complicated nonlinearities and dispersive operators.
Classical-to-Quantum Transition in Blackbody Radiation Thermodynamics: Planck's Law without Statistical Assumptions
oai:arXiv.org:2610.27858v1
arXiv:2610.27858v1 Announce Type: cross
Abstract: Inspired by Fermi's model of radiation-matter interaction, we investigate the classical dynamics of a closed, finite-energy system, in which matter is coupled to an electromagnetic field. We find that energy is progressively transferred to increasingly high-frequency field modes, so that the effective temperature of both matter and radiation approaches zero in the long-time limit - in sharp contradiction with the finite-temperature equilibrium assumed by classical blackbody thermodynamics. Requiring consistency with the Stefan-Boltzmann and Wien laws identifies a characteristic time $t_L$, at which both laws are simultaneously satisfied. Strikingly, classical spectra obtained for different energies and cavity sizes collapse onto the universal Planck distribution at $t_L$. We conjecture that quantum dynamical localization arrests the classical energy cascade at $t_L$, and that the Planck distribution emerges as a dynamical consequence of this arrest - without invoking any a priori statistical assumption.
Surface defects on lens space and transfer matrices
oai:arXiv.org:2610.28165v1
arXiv:2610.28165v1 Announce Type: cross
Abstract: We study surface defects in four-dimensional $\mathcal{N}=2$ theories of class $\mathcal{S}$ and $\mathcal{N}=1$ theories of class $\mathcal{S}_k$ placed on $L(r,1)\times S^1$. The lens space index depends on the fugacities $p$ and $q$, and the effect of the defects on it is captured by difference operators acting on the fugacities and holonomies of flavor symmetries. For theories of type $A_1$ we determine in closed form the difference operators for the basic defects, labeled by $p^r$, $q^r$ and $pq$, and show that for odd $r$ there exist additional elementary defects, whose operators shift the holonomy by one half modulo $r$. All the operators commute, and the elementary operators generate the others. For even $r$ the same holds once the holonomy is extended to a twisted sector. We show that the elementary operators are traces of $L$-operators whose entries generate a Sklyanin algebra, with nome $q^r$ and shift $pq$. The lens space enters only through the parameters of the algebra and through its representation, which is realized on a doubled holonomy label. The $p$- and $q$-type algebras form a lens space version of the elliptic modular double. For class $\mathcal{S}_k$ theories we propose, and test in several ways, that the defect operators are transfer matrices of inhomogeneous spin chains built from these $L$-operators. Throughout, we use the TQFT structure of the lens space index, which follows from S-duality, both to fix and to check the operators.
Scrambling Separates Superradiance from Quantum Chaos
oai:arXiv.org:2610.28412v1
arXiv:2610.28412v1 Announce Type: cross
Abstract: Early studies on the Dicke model linked superradiance to quantum chaos, whereas later anisotropic Dicke and Rabi studies reached apparently conflicting conclusions. We resolve this tension by distinguishing global chaos from local instability. Along the isotropic Dicke path, a sweep at fixed low excitation energy crosses both the superradiant boundary and a nearby chaotic window, explaining the reported correlation. However, the anisotropic Dicke model can be chaotic in the normal phase and regular in the superradiant phase, making superradiance neither necessary nor sufficient for global chaos. A fidelity out-of-time-order correlator (FOTOC), in contrast, probes the local stability of the stationary point selected by the initial wave packet. The superradiant transition destabilizes the origin, whereas a change in anisotropy can restore its stability and create displaced saddles. Wave packets centered at these saddles exhibit exponential FOTOC growth even on a classically regular Dicke energy shell with Poisson statistics. This local sensitivity also enables quantitative spectroscopy: combined with the topology of stationary points and local density of states relations, FOTOC dynamics reconstructs the absolute amplitudes of spectral singularities in both models without fitting the density of states.
Chaos and scrambling in anisotropic Dicke and Rabi models
oai:arXiv.org:2610.28415v1
arXiv:2610.28415v1 Announce Type: cross
Abstract: Identifying the quantum signatures of classical chaos remains a central question in quantum physics, and quantum scrambling provides a new perspective on this correspondence. We study two distinct limits of the same model to examine how the correspondence between chaos and scrambling changes across different regimes. The thermodynamic-limit anisotropic Dicke model exhibits a close correspondence between chaos and scrambling, whereas this correspondence breaks down in the large-frequency-ratio limit of the anisotropic Rabi model (ARM). The contrasting chaotic behavior in the two limits originates from the different numbers of classical degrees of freedom, whereas their similar scrambling behavior stems from eigenstate delocalization. This implies that, in the large-frequency-ratio limit, eigenstate delocalization alone is insufficient to qualify the ARM as a classically chaotic model. This finding suggests that identifying chaos in quantum systems requires jointly considering the classical degrees of freedom and eigenstate delocalization.
Cumulative locality governs collective quantum charging
oai:arXiv.org:2610.28614v1
arXiv:2610.28614v1 Announce Type: cross
Abstract: Quantum batteries can outperform their classical counterparts through collective quantum effects, yet the conditions enabling such advantage remain unclear. Using a generic class of free-fermionic chargers we prove that global operators are insufficient for scalable superextensive charging and that a cumulative hierarchy of many-body operators is required, with the structure of the battery determining the degree of advantage. This is obtained by relating the charging power to the norm of the commutator between the battery and charger Hamiltonians and deriving an upper bound only capturing noncommuting terms. Moreover, we revisit the concept of participation number and propose an effective definition that accounts for the relative strength of the contributing terms, correlating it with the charging power scaling. Not limited to quadratic Hamiltonians, our results establish cumulative locality as a key resource for quantum charging advantage.
The gauge action on semi-discrete Lax representations, its invariants, and Lax-Miura type constructions
oai:arXiv.org:2604.20830v4
arXiv:2604.20830v4 Announce Type: replace
Abstract: We introduce and study explicit invariants with respect to the action of (local) gauge transformations on the set of semi-discrete (differential-difference) matrix Lax representations (MLRs) for a given (1+1)-dimensional evolutionary differential-difference equation with any number of components. Using these invariants, we obtain the following general results:
- Sufficient conditions for nonremovability of parameters in semi-discrete MLRs. (Nonremovability by means of gauge transformations.)
- Necessary conditions for two semi-discrete MLRs to be gauge equivalent.
Effective use of these results is demonstrated. We present the general theory and apply it to MLRs for equations of Volterra--Yamilov, Cherdantsev--Yamilov, Toda types and Miura-type modifications.
Also, we present a comparison with results of S. Yu. Sakovich and M. Marvan on gauge invariants for (continuous) zero-curvature representations of partial differential equations.
Furthermore, we demonstrate procedures for constructing Miura-type transformations (MTs) for discrete (partial difference) equations (on the two-dimensional integer lattice) by means of gauge simplifications of discrete matrix Lax pairs (DLPs) and reductions of auxiliary systems associated with DLPs. This allows us to construct new MTs for (non-evolutionary) 4-point and 6-point discrete equations from [D. Levi, R.I. Yamilov, J. Phys. A 2009], [S. Konstantinou-Rizos, A.V. Mikhailov, P. Xenitidis, J. Math. Phys. 2015].
Dark solitons in the fractional NLS equation
oai:arXiv.org:2605.21271v2
arXiv:2605.21271v2 Announce Type: replace
Abstract: In the present work we consider the subject of dark fractional solitary waves in the realm of generalized (fractional) forms of the nonlinear Schr\"odinger (NLS) equation. While earlier studies have examined such states in the realm of real field theories, we showcase the existence and stability of individual dark solitary waves in such NLS settings and subsequently turn to two-soliton solutions. We find different branches of such two-soliton solution equilibria and contrary to the real field-theoretic setting all possible branches of two-soliton equilibria are found to be potentially unstable, although with different types of instabilities. Odd branches are potentially subject to oscillatory instabilities, while even branches are always exponentially unstable. The dynamics that results from the instabilities is also examined and is found to potentially feature breathing characteristics. This prompts us to seek and find associated periodic (breathing) orbits that are also unprecedented in this context, to the best of our knowledge. The effective particle-like dynamics of the solitary waves also prompts us to seek ordinary differential equation (ODE) descriptions to the dark soliton interaction dynamics. These are shown to hold promise toward providing us with effective reduced order models, indicating some potential directions for further investigation.
Gliders on Aperiodic Monotilings: Cellular Automata on the Hat and Spectre
oai:arXiv.org:2609.22579v2
arXiv:2609.22579v2 Announce Type: replace
Abstract: The hat and spectre monotiles, discovered in 2023, tile the plane only aperiodically. Cellular automata are studied here on patches generated by finite-state transducers, so that every experiment regenerates deterministically from a small record. Within edge-adjacency semi-totalistic rules, exhaustive and evolutionary searches find only mortal travelers: gliders are absent. Guided by a reproduction of the known Penrose-tiling glider, the rule space is extended to vertex neighborhoods and to priority-table rules whose non-quiescent states are visible to neighbors. Evolutionary search then discovers gliders on both monotilings; tracked by a sliding window that regenerates the patch along the flight, they travel one million rings at constant speed and heading. All headings are quantized, to millidegrees, onto a six-spoke compass: the fast axes of the tiling's graph metric. An ablation shows both rule-space extensions are individually necessary. All results replay exactly in an accompanying interactive essay.
Geometric Transition of Latent Dynamics in Intermittent Chaos: SVD-based Effective Dimension for Detection and Proactive Control
oai:arXiv.org:2610.16495v2
arXiv:2610.16495v2 Announce Type: replace
Abstract: This paper investigates the geometric detection of latent dynamic transitions (where the latent space is defined as a finite-dimensional state space reconstructed via a delay-embedding Hankel matrix) in mutually coupled phase-locked loops (PLLs) exhibiting on-off intermittent chaos. We propose a data-driven approach to identify the transition of the effective dimension in latent dynamics as a precursor to physical bursting events. By applying Singular Value Decomposition (SVD) to a Hankel matrix constructed from high-resolution oversampled data, we extract the effective dimension that reflects the dynamical shift of intrinsic variables - from a synchronized two-dimensional manifold to a higher-dimensional chaotic state - prior to the actual burst. This geometric indicator enables a proactive, event-driven intervention using Pyragas-type delayed feedback control, which is activated only when the detected dimension exceeds a critical threshold. Simulation results demonstrate that this geometric precursor effectively compensates for control latencies and substantially reduces accumulated control energy compared to conventional continuous feedback. The findings establish a methodological strategy for stabilizing infinite-dimensional delayed systems by exploiting early warning signals embedded in dimensional transitions of latent dynamics.
Predator-Prey Model: Driven Hunt for Accelerated Grokking
oai:arXiv.org:2509.10562v2
arXiv:2509.10562v2 Announce Type: replace-cross
Abstract: A machine learning method is proposed using two agents (representing two coupled replicas of the model parameters) that simulate the biological behavior of a predator and a prey. In this method, the two agents interact with each other via a pair of potentials that simulate the dynamics when the predator chases the prey while the prey runs away from the predator - to perform an optimization on the landscape. This method allows, for optimization landscapes with narrow ravines with gentle slope or even flat bottoms, to avoid optimization getting stuck in a ravine. For some examples of grokking (i.e., delayed generalization) problems we show that this method allows for achieving up to a hundred times faster learning in gradient evaluations and up to 60 times in the wall-clock time compared to the standard learning procedure.
Generalized mKdV Equation and Genus Two Jacobi Type Hyperelliptic Differential Equation
oai:arXiv.org:2511.02621v4
arXiv:2511.02621v4 Announce Type: replace-cross
Abstract: We generalized the mKdV equation in order that the static equations include ${\rm sn}$ differential equation. As a result, a good correspondence was obtained between the KdV equation and the mKdV equation.For general genus two hyperelliptic curves, we obtained differential equations for Weierstrass type and Jacobi type hyperelliptic functions. Considering the special case of $\lambda_6=0, \lambda_0=0$, Weierstrass type and Jacobi type hyperelliptic functions are different solutions to the same hyperelliptic differential equations. Then these solutions are connected by the special ${\rm Sp(4, {\bf R})}$ Lie group transformation.
Ising-machine-like Driving Condition Restrictions in SSBM and Additional Pseudo-annealing- Driving Method for SSBM that Achieves Best Cut Value Search -
oai:arXiv.org:2608.12545v3
arXiv:2608.12545v3 Announce Type: replace-cross
Abstract: In our previous work, we proposed and experimentally demonstrated a physics-based simulator for combinatorial optimization problems (SSBM), which exploits a distinctive spontaneous symmetry breaking (SSB) phenomenon identified in a dissipative system exhibiting robust causality described by an iterative equation. We further conducted numerical simulations in which initial fluctuations of distinct patterns were applied to large-scale benchmark problems, and reported that the SSBM exhibits a unique behavior absent in conventional Ising machines: under specific conditions, pseudo-spin patterns progressively converge during the solution-search process and ultimately coalesce into a single pattern, with the cut value of the converged solution reaching 99.7% of the best-known cut value. In the present paper, we elucidate the conditions under which the SSBM can be configured to emulate the behavior of a continuous-variable Ising machine. We further demonstrate that by additionally incorporating an effect analogous to physical annealing into these conditions, the SSBM becomes capable of identifying the best-known cut value for the K_2000 benchmark instance.