Peer Revenue Past citations are no guarantee of future returns

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Mathematics

math Watch Mathematics (sign in to use)

Monday 11 May

Statistics Theory

Increasing domain asymptotics for covariate-based nonparametric Bayesian intensity estimation with Gaussian and Besov-Laplace priors

We study the problem of estimating the intensity function of a covariate-driven point process based on observations of the points and covariates over a large window. We consider the nonparametric Bayesian approach, and show that a wide class of Gaussian priors, combined with flexible link functions, achieves minimax-optimal posterior contraction rates in the increasing domain asymptotics and under the assumption that the covariates be ergodic. We also employ Besov-Laplace priors, which are popular in imaging and inverse problems due to their edge-preserving and sparsity-promoting properties. We prove that these yield optimal estimation of spatially inhomogeneous intensities belonging to Besov spaces with low integrability index. These results are based on a general concentration theorem that extends recent findings from the literature. To corroborate the theory, we provide extensive numerical simulations, implementing the considered procedures via suitable posterior sampling schemes. Further, we present two real data analyses motivated by applications in forestry and the environmental sciences.
Read the full abstract

Optimization and Control

Parameterized Complexity of Stationarity Testing for Piecewise-Affine Functions and Shallow CNN Losses

We study the parameterized complexity of testing approximate first-order stationarity at a prescribed point for continuous piecewise-affine (PA) functions, a basic task in nonsmooth optimization. PA functions form a canonical model for nonsmooth stationarity testing and capture the local polyhedral geometry that appears in ReLU-type training losses. Recent work of Tian and So (2025) shows that testing approximate stationarity notions for PA functions is computationally intractable in the worst case, and identifies fixed-dimensional tractability as an open direction. We address this direction from the viewpoint of parameterized complexity, with the ambient dimension $d$ as the parameter. In this paper, we give XP algorithms in fixed dimension for the tractable sides, and prove W[1]-hardness for the complementary sides. Moreover, lower bounds under the parameterized $k$-Clique hypothesis rule out deterministic algorithms running in time $\rho(d)s^{o(d)}$ for any computable function $\rho$, where $s$ denotes the total binary encoding length of the instance. As a further consequence, our results yield the corresponding parameterized complexity picture for testing local minimality of continuous PA functions. We also extend our hardness results to a family of shallow ReLU CNN training losses, tested in the trainable weight space. Thus, the same parameterized-complexity picture also appears for simple CNN training losses. Notably, our lower bounds assume that $k$-Clique on $N$-vertex graphs admits no deterministic $\rho(k)N^{o(k)}$-time algorithm for any computable function $\rho$. The recent refutations of the Exponential Time Hypothesis (ETH) (OpenAI, 2026) and the stronger fine-grained $k$-Clique hypothesis (Vassilevska Williams and Zamir, 2026) do not contradict this assumption.
Read the full abstract

2605.10219 · 11 May · 2:46 Accepted at NeurIPS 2026. 29 pages. This revision states the conditional lower bounds directly under the parameterized k-Clique hypothesis, which is not contradicted by the recent refutations of ETH and the stronger fine-grained k-Clique hypothesis

Statistics Theory

Extended Wasserstein-GAN Approach to Causal Distribution Learning: Density-Free Estimation and Minimax Optimality

Distributional causal inference requires estimating not only average treatment effects but also interventional outcome distributions, including quantiles, tail risks, and policy-dependent uncertainty. As a method for distributional causal inference, generative adversarial network (GAN)-based counterfactual methods are flexible tools for this task. However, these methods have several limitations. First, the objectives of certain techniques do not coincide with the statistical risk of the identifiable causal target, and therefore provide limited theoretical guarantees regarding estimable counterfactual distributions or optimality. Second, implementing target-design risks under observational sampling can require density-ratio estimation. In this paper, we propose GANICE (GAN for Interventional Conditional Estimation) with several advantages: it (i) specifies the conditional interventional distribution for each treatment–covariate state as the causal estimation target; (ii) estimates the conditional distribution such that its averaged Wasserstein risk is minimized; (iii) attains minimax optimality. GANICE achieves these advantages through the extended Wasserstein distance and cell-normalized estimation, with minimax bounds based on Wasserstein GAN oracle inequalities and an anisotropic Assouad construction; Besov theory characterizes the cellwise critics. Our experiments show that GANICE outperforms existing methods on most benchmarks.
Read the full abstract

Functional Analysis · Cross-listed from cs.LO

Just Previsions

Previsions are positively homogeneous functionals, and are generalized forms of integration functionals. We investigate previsions – just previsions, not sublinear or superlinear previsions as in previous work. We show that every prevision can be expressed as an infimum of sublinear previsions, and as a supremum of superlinear previsions under mild conditions. This extends to homeomorphisms between spaces of previsions and certain hyperspaces over spaces of sublinear or superlinear previsions, which can also be characterized in terms of orthogonality relations, making the construction a variant of a double powerspace construction.
Read the full abstract

Numerical Analysis · Cross-listed from cs.LG

jNO: A JAX Library for Neural Operator and Foundation Model Training

jNO (jax Neural Operators) is a JAX-native library for neural operators and foundation models with unified support for both data-driven and physics-informed training. Its core design is a tracing system in which domains, model calls, residuals, supervised losses, and diagnostics are written in one symbolic language and compiled into one optimization pipeline. This allows users to move between operator regression, mesh-aware residual evaluation, and PDE-constrained training without restructuring the surrounding code. jNO also supports multi-model compositions, fine-grained control at parameter level (model, optimizer, and learning rate), hyperparameter tuning, and JAX-native workflows for translated PDE foundation-model families. The source repository is available at https://github.com/FhG-IISB/jNO.
Read the full abstract

All authors Leon Armbruster, Rathan Ramesh, Georg Kruse, Christopher Straub

Optimization and Control · Cross-listed from cs.LG

Learning to Sparsify Stochastic Linear Bandits

This paper addresses the problem of learning to sparsify stochastic linear bandits, where a decision-maker sequentially selects actions from a high-dimensional space subject to a sparsity constraint on the number of nonzero elements in the action vector. The key challenge lies in minimizing cumulative regret while tackling the potential NP-hardness of finding optimal sparse actions due to the inherent combinatorial structure of the problem. We propose an adaptively phased exploration and exploitation algorithmic framework, utilizing ordinary least squares for parameter learning and specialized subroutines for sparse action selection. When the action set is a Euclidean ball, optimal sparse actions can be efficiently computed, enabling us to establish a $\tilde{\mathcal{O}}(d\sqrt{T})$ regret, where $d$ is the dimension of the action vector and $T$ is the time horizon length. For general convex and compact action sets where finding optimal sparse actions is intractable, we employ a greedy subroutine. For general strongly convex action sets, we derive a $\tilde{\mathcal{O}}(d \sqrt{T})$ $\alpha$-regret; for general compact sets lacking strong convexity, we establish a $\tilde{\mathcal{O}}(d T^{2/3})$ $\alpha$-regret, where $\alpha$ pertains to the approximation ratio of the greedy algorithm. Finally, we validate the performance of our algorithms using extensive experiments including an application to recommendation system.
Read the full abstract

All authors Zhengmiao Wang, Ming Chi, Zhi-Wei Liu, Lintao Ye, Carla Fabiana Chiasserini

Probability

Rough path theory and an introduction to rough partial differential equations

The goal of these notes is to provide an introduction to rough partial differential equations. For this purpose, we will present the theory of rough paths to the extend as it is required. Applications to stochastic partial differential equations are presented as well.

Complex Variables

Thompson's groups and Teichmüller modular groups of generalized Cantor sets

Thompson's groups, which are denoted by $F, T$ and $V$, were introduced by R. Thompson. It is known that they are related to various fields in mathematics. In this paper, we discuss Thompson's groups $F, T$ and $V$ in Teichmüller modular groups of Teichmüller spaces of generalized Cantor sets with certain conditions by using a quasiconformal geometric method. Moreover, we show that groups corresponding to $F$ and $T$ act properly discontinuously on such Teichmüller spaces. We also show the same results for the asymptotic Teichmüller space. E. de Faria, F. P. Gardiner and W. J. Harvey show similar results on Thompson's group $F$ and asymptotic Teichmüller spaces. So, our results are improvement of them. We also discuss the actions of Thompson's groups on other spaces which we call \lq\lq pre-Tiechmüller spaces" of generalized Cantor sets. Especially, we show that the action of a group which is isomorphic to Thompson's group $V$ on the pre-Teichüller space is not properly discontinuous while groups isomorphic to $F$ and $T$ act properly discontinuously. Finally, we construct infinitely many distinct Teichmüller spaces of generalized Cantor sets where our results can apply.
Read the full abstract

Complex Variables

The Nitsche–Hopf conjecture for minimal graphs

We prove the Nitsche–Hopf conjecture for non-parametric minimal graphs over disks. If \(S\) is a minimal graph over a disk of radius \(R\), and if \(\xi\) is the point above the center, then \[ W(\xi)^2 |K(\xi)|<\frac{\pi^2}{2R^2}. \] Here \(K\) is the Gaussian curvature and \[ W=\sqrt{1+|\nabla u|^2}=\frac1{n_3} \] is the reciprocal of the vertical component of the upward unit normal. The constant is sharp, as shown by the horizontal tangent-plane extremal sequence of Finn and Osserman. The main difficulty is that the bicentric-quadrilateral comparison theorem for Gaussian curvature controls \(|K|\), but it does not by itself control the normalized quantity \(W^2|K|\): the slope factor \(W\) can be arbitrarily large. We show that the missing information is recovered inside the Scherk-type comparison family from the zero equation for the horizontal harmonic projection. More precisely, in the fixed-arc normalization the point corresponding to the center of the physical disk is a distinguished zero \(z_\circ\) of the harmonic projection. The equation \(f(z_\circ)=0\), written in harmonic-measure coordinates, reduces the sharp Hopf estimate to a scalar derivative inequality at the admissible zero of a monotone function \(G_{A,B}\). We prove this scalar inequality on the full admissible parameter domain by a barrier argument and two explicit Bernstein-polynomial positivity certificates. Combined with the bicentric-quadrilateral comparison theorem of the first author and Melentijević, the Scherk-family estimate gives the sharp normalized Hopf estimate for arbitrary minimal graphs over disks. As a byproduct, we obtain the two-sided bound \[ \frac{\pi^2}{4}\leq W^2|K|\leq \frac{\pi^2}{2} \] throughout the normalized Scherk-type comparison family, evaluated at the distinguished point corresponding to the center.
Read the full abstract

Analysis of PDEs

Global Well-posedness for the Multi-species Boltzmann Equation with Large Amplitude Initial Data

This paper establishes the global well-posedness of the multi-species Boltzmann equation with large-amplitude initial data in the periodic domain $\mathbb{T}^3$. In contrast to the single-species case, the multi-species mixture model lacks structural symmetry in its collision operators due to the distinct masses of different species. This asymmetry makes it difficult to obtain pointwise estimates for the nonlinear collision terms. Although the Carleman representation for the mixture model introduced in \cite{BD2016} provides a useful reduction of the collision integral, it does not directly yield the desired estimate. To overcome this difficulty, we identify an additional algebraic cancellation structure which leads to the pointwise estimates for the nonlinear terms. By applying this refined approach, we derive the necessary velocity-weighted $L^\infty$ estimates for the nonlinear terms. Furthermore, under the smallness assumption on the initial relative entropy, we establish a uniform lower bound for the nonlinear collision frequency and prove that the large-amplitude solutions exist globally in time and decay exponentially to the global equilibrium.
Read the full abstract

Analysis of PDEs

Concentration effects and $\Gamma$-limit for the elastica functional for open and closed curves

We study the $\Gamma$-convergence of a class of elastica-type energies defined on immersed planar curves and depending on a small positive parameter $\epsilon$. As $\epsilon\to 0^+$, sequences with equibounded energy develop concentration phenomena in the curvature, leading to the emergence of singularities described by atomic measures. This naturally gives rise to a limiting framework in terms of pointed curves, consisting of a curve together with a measure encoding curvature concentration. We characterize the first-order $\Gamma$-limit in two settings: for immersed open curves with fixed endpoints and boundary conditions on the tangents, and for immersed closed curves of prescribed length. In both cases, the limiting energy depends only on the number of concentration points and is expressed as a sum of contributions, each given by an integer multiple of $2\pi$. A key feature of the problem is that the rescaled energies exhibit a structure closely related to one-dimensional Modica–Mortola type functionals.
Read the full abstract

All authors Giovanni Bellettini, Virginia Lorenzini, Matteo Novaga, Riccardo Scala

Dynamical Systems

Generalized rescaling limits of a sequence of rational maps

We consider a sequence of complex rational maps (f_n) of a fixed degree d at least 2. Building on the seminal work of Kiwi, we introduce the notion of generalized rescaling limits. These are rational maps possibly defined over a non-Archimedean field obtained by renormalizing at some scale a fixed iterate of the sequence (f_n). We explain that the set of all generalized rescaling limits is naturally organized as a tree, and bound the size of this tree in term of the degree d. We apply our theory to quadratic rational maps. Using Kiwi's classification, we describe all possible trees in this case, and prove a uniform bound on the number of cycles with small multipliers.
Read the full abstract

Information Theory · Cross-listed from cs.NI

GELATO: Generative Entropy- and Lyapunov-based Adaptive Token Offloading for Device-Edge Speculative LLM Inference

The recent growth of on-device Large Language Model (LLM) inference has driven significant interest in device-edge collaborative LLM inference. As a promising architecture, Speculative Decoding (SD) is increasingly adopted where a lightweight draft model rapidly generates candidate tokens to be verified by a powerful target model. However, a fundamental challenge lies in achieving per-token resource scheduling to effectively adapt SD paradigm to resource-constrained edge environment. This paper proposes a Generative Entropy- and Lyapunov-based Adaptive Token Offloading framework, named GELATO, to maximize decoding throughput under energy constraints in a device-edge collaborative SD system. Specifically, an outer drift-plus-penalty loop makes online decisions to establish a reference drafting budget, managing long-term energy-throughput trade-off. Further, a nested entropy-driven generation mechanism executes early exiting to adapt to per-token dynamic generative uncertainty. Theoretical analysis establishes a rigorous performance bound on long-term throughput for GELATO. Extensive evaluations demonstrate that GELATO achieves a globally optimal tradeoff, outperforming state-of-the-art distributed SD architectures by 64.98% in token throughput and reducing energy consumption by 47.47% under resource-constrained environments, while preserving LLM decoding quality.
Read the full abstract

All authors Zengzipeng Tang, Yuxuan Sun, Wei Chen, Jianwen Ding, Bo Ai

Combinatorics

Motzkin paths with two variants of level steps on odd levels – a kernel method approach

The sequence A176677 in the Encyclopedia of Integer Sequences enumerates Motzkin paths where two types of horizontal steps may occur, but only on odd indexed levels. We show how to perform the enumeration, also dealing with partial such Motzkin paths leading to a particular level or to any level (open paths). The method is the kernel method where functional equations are manipulated in a suitable way. The coefficients of sequence A176677 satisfy a holonomic recursion that was recently discussed on the arxiv. We show how this can be established in an (almost) automatic fashion. Eventually we switch the roles of `odd' and `even'. One could also allow more versions of horizontal steps but we leave this to the interested readers.
Read the full abstract

Combinatorics

The Dominating 4-Colour Theorem

A "dominating $K_t$-model" in a graph $G$ is a sequence $(T_1,\dots,T_t)$ of pairwise vertex-disjoint connected subgraphs of $G$, such that whenever $1\leq i<j\leq t$ every vertex in $T_j$ has a neighbour in $T_i$. Replacing "every vertex in $T_j$" by "some vertex in $T_j$" retrieves the standard definition of $K_t$-model, which is equivalent to a $K_t$-minor in $G$. We prove that every graph with no dominating $K_5$-model is $4$-colourable. This generalises and is significantly stronger than the 4-colour theorem for planar graphs or for graphs with no $K_5$-minor. It also makes progress towards Hajós' conjecture on $K_5$-subdivisions in $5$-chromatic graphs.
Read the full abstract

All authors António Girão, Freddie Illingworth, Bojan Mohar, Sergey Norin, Raphael Steiner, Youri Tamitegama, Jane Tan, David R. Wood, Jung Hon Yip

Combinatorics

Randomly Shifted Steinhaus Longimeters and Buffon Discrepancy

Let $\Omega \subset \mathbb{R}^2$ be a bounded convex domain. Steinerberger (2026) introduced the Buffon discrepancy problem: given length $L$, construct a one-dimensional set $S\subset\Omega$ such that the number of intersections of $S$ with a line $\ell$ approximates the Crofton-normalized chord length $$ \frac{2L}{\pi|\Omega|}\cdot\mathcal{H}^1(\ell\cap\Omega).$$ Steinerberger proved a universal upper bound of order $L^{1/3}$ using a Steinhaus longimeter construction, and showed that the disk admits bounded discrepancy. We prove that a randomly shifted Steinhaus construction improves the order of the universal upper bound to $L^{1/5}(\log L)^{2/5}$.
Read the full abstract

Representation Theory

Fractal phenomenon in $c$- and $g$-vectors of the Markov quiver

We study the $C$- and $G$-patterns associated with rank $3$ skew-symmetrizable matrices of $B$-invariant type, including the Markov quiver. Motivated by the self-contained simple mutations in Markov-type cluster algebras, we prove that large classes of subpatterns of modified $c$- and $g$-vectors are linearly isomorphic, yielding a fractal structure of the corresponding $G$-fan. We further derive explicit recursive formulas for all modified $c$- and $g$-vectors in terms of integer pairs satisfying a recursion analogous to the Calkin-Wilf tree, which leads to a parameterization by coprime integers. As an application, we describe all connected components of the complement of the support of the $G$-fan, and show that they are generated recursively by three kinds of linear maps.
Read the full abstract

Combinatorics

Coarse Menger property of quasi-minor excluded graphs and length spaces

Menger's theorem is an important building block of numerous results in the study of graph structure. We consider a variant in terms of coarse geometry. We say that a set of graphs has the weak coarse Menger property if there exist functions $f$ and $g$ such that for any graph $G$ in this set, subsets $X$ and $Y$ of vertices of $G$, and positive integers $k$ and $r$, either there exist $k$ paths between $X$ and $Y$ pairwise at distance at least $r$, or there exists a union of at most $f(k,r)$ balls of radius at most $g(k,r)$ intersecting all paths between $X$ and $Y$. Nguyen, Scott and Seymour proved that the set of all graphs does not have the weak coarse Menger property and asked whether every proper minor-closed family of finite graphs has it. In this paper, we provide a positive answer to this question in a stronger form: it is true for the set of locally finite graphs with an excluded finite minor, and the functions $f$ and $g$ can be chosen so that $f$ only depends on the number $k$ of the paths in the packing and the function $g$ is a linear function of the distance threshold $r$ and is independent of $k$, which is optimal up to a constant factor. Our result extends to every length space quasi-isometric to a locally finite graph or metric graph with an excluded finite minor, such as complete Riemannian surfaces of finite Euler genus, string graphs, and Cayley graphs of finitely generated minor-excluded groups.
Read the full abstract

Mathematical Physics

Generalized i-boson model and boxed BUC plane partitions

This paper is devoted to investigating the relation between the generalized i-boson model and boxed BUC plane partitions. The representation of the generalized i-boson algebra and the actions of the monodromy matrix operators on basis vectors have been studied. We also consider the actions of neutral fermion vertex operators on state vectors in terms of the neutral fermionic Fock space. With the help of the scalar product of the generalized i-boson model, the generating function for boxed BUC plane partitions is derived which can be represented as the products of Schur Q-functions. Moreover, the generating function for BUC plane partitions with the double scaling limit is presented.
Read the full abstract

Probability · Cross-listed from stat.ME

A Statistical Framework for Learning Preferences from the Past

In many real-world settings such as online recommendation or consumer choice modeling, individuals make repeated choices from a fixed set of options. Accurately estimating their underlying preferences is essential for generating personalized future recommendations. Probabilistic models for understanding user choice behavior from past decisions can serve as a valuable addition to existing recommender systems and choice prediction methods. To this end, in this article, we introduce a novel statistical framework for predicting user preferences based on their past choices, under a natural monotonicity assumption: options that were chosen more frequently or more intensely in the past are more likely to be chosen again in the future. Our approach builds on a parametric model proposed by Le Goff and Soulier (2017), originally used to describe how ants in an ant colony select a path among many pre-existing paths. We propose a non-parametric generalization of this model, drawing inspiration from the generalized elephant random walk introduced by Maulik et al. (2024). We develop a method of maximum likelihood estimation of the user preference probabilities under the above-mentioned monotonicity constraint. We also derive theoretical guarantees for our estimator and demonstrate the effectiveness of our method through both simulated experiments and real-world datasets.
Read the full abstract

All authors Tamojit Sadhukhan, Moulinath Banerjee, Krishanu Maulik, Parthanil Roy

Optimization and Control

Stabilization for a Cascaded ODE-Wave Equation with Boundary Nonlinear Disturbances

In this article, we investigate the problem of exponential stabilization via output feedback for a cascaded system composed of an ordinary differential equation (ODE) and a wave partial differential equation (PDE) under boundary control. Four types of boundary interconnections are considered. In the absence of disturbances, a novel transformation is introduced to incorporate the PDE boundary control input into the ODE subsystem. Based on this transformation, a state feedback controller is designed to achieve exponential stability for both the ODE and PDE components. When internal uncertainties and external disturbances that match the control structure are present, a disturbance estimator is constructed. Utilizing this estimator, a Luenberger-type state observer is developed to reject the disturbances and exponentially stabilize the original system via an observer-based control scheme. Furthermore, the boundedness of the entire closed-loop system is rigorously established. Numerical simulations are provided to illustrate the theoretical results.
Read the full abstract

Mathematical Physics · Cross-listed from cond-mat.stat-mech

Families of planar lattices with arbitrarily high $T_{\rm c}$ for the ferromagnetic Ising model

We construct families of periodic tessellations of the plane with arbitrarily high critical temperature, $T_{\rm c}$, for the classical nearest-neighbor ferromagnetic Ising model. Our approach is motivated by recently found exact bounds, which imply that large values of $T_{\rm c}$ require large values of the maximal coordination number of the lattice, $q_{\rm max}$. We create such lattices through iterative triangulation and derive explicit expressions for their $T_{\rm c}$. Furthermore, we show that $T_{\rm c}$ for these families scales asymptotically as $T_{\rm c}/J\sim A \ln q_{\rm max}-2 \ln \ln q_{\rm max}$ with a universal prefactor $A=2/\ln 2$. We introduce a function $T_{\rm c}^*(q_{\rm max})$ that we conjecture to be optimal for all periodic tessellations of the plane. We show that the family of so-called Apollonian lattices, which are derived from the Triangular lattice through iterative triangulation, saturates this bound. The lattices discussed in this work are relevant for theoretical questions of optimality in network systems and may be realized experimentally in Coherent Ising Machines or topoelectric circuits in the future.
Read the full abstract

Combinatorics

Lattice-free Schubitopes

In this paper, we provide a simple criterion for the Schubitope $\mathcal{S}_{D}$ associated to a diagram $D$ to be lattice-free. We further show that $\mathcal{S}_{D}$ is lattice-free if and only if its Ehrhart polynomial is equal to the product of Ehrhart polynomials of the Schubert matroid polytopes corresponding to each column of $D$. As applications, we obtain that the Newton polytopes of the Schubert polynomial $\mathfrak{S}_w(x)$ and the Grothendieck polynomial $\mathfrak{G}_w(x)$ are lattice-free if and only if $w$ avoids the patterns 1423, 1432, 13254, and confirm several conjectures by Mészáros, Setiabrata, and St.Dizier on the support of Grothendieck polynomials for this class of permutations.
Read the full abstract

Analysis of PDEs

Geometry-induced pulse dynamics in a bulk-surface reaction-diffusion system for cell polarization

This paper studies a bulk–surface reaction–diffusion system for cell polarization in two-dimensional domains. The model describes the formation of localized patterns through the wave-pinning mechanism, while explicitly incorporating the effect of cell shape. Using singular perturbation methods, we formally derive reduced ordinary differential equations describing the wave-pinning dynamics on a fast time scale and the subsequent slow drift of pulse solutions induced by domain geometry. The resulting slow dynamics is a gradient flow of a potential function whose geometry-dependent part is expressed in terms of the Neumann Green's function. Moreover, we give a probabilistic interpretation of the potential function using the excursion Poisson kernel. We then analyze the reduced dynamics in several concrete geometries, including dumbbell-shaped domains and perforated disks. In these examples, we characterize stationary pulse positions, their stability, and the bifurcation structures arising from changes in geometric parameters. To evaluate the geometric terms appearing in the reduced dynamics, we use a conformal mapping method to compute the Neumann Green's function for these domains. Our analysis reveals geometry-induced phenomena such as nontrivial stationary pulse locations and both supercritical and subcritical pitchfork bifurcations. Finally, we perform numerical simulations to support the theoretical predictions.
Read the full abstract

Optimization and Control · Cross-listed from cs.AI

Optimizer-Induced Mode Connectivity: From AdamW to Muon

Mode connectivity has been widely studied, yet the role of the optimizer remains underexplored. We revisit it through optimizer-induced implicit regularization, asking how connectivity behaves when restricted to solutions constrained by a given optimizer. For two-layer ReLU networks, we show that solutions from a single optimizer – AdamW, Muon, or others in the Lion-$\mathcal{K}$ family – form a connected set at sufficiently large width, a result not implied by prior work. We then characterize how optimizer-induced regions interact: at large width two different regions can be disjoint or overlap depending on regularization, while in our small-width example AdamW and Muon converge to disconnected zero-loss components separated by a provable loss barrier. Empirically, in GPT-2 pretraining, we observe same-optimizer paths preserve each model's spectrum while cross-optimizer paths traverse a smooth transition. Our results reveal optimizer-dependent structure beyond classical mode connectivity literature.
Read the full abstract

All authors Fangzhao Zhang, Sungyoon Kim, Erica Zhang, Yiqi Jiang, Mert Pilanci

Number Theory

Congruences of first syntomic cohomology groups

Let O_K be the ring of integers of a finite extension K of Q_p. Given two reflexive F-gauges on O_K, we show that for large enough n, the mod p^n-reductions of their first syntomic cohomology groups, which might be regarded as a refinement of local Bloch–Kato Selmer groups, are isomorphic if and only if the mod p^{2n}-reductions of their attached Breuil–Kisin modules with G_K-actions and Nygaard filtrations are isomorphic.
Read the full abstract

Optimization and Control · Cross-listed from cs.LG

Chebyshev Center-Based Direction Selection for Multi-Objective Optimization and Training PINNs

Physics-informed neural networks (PINNs) are a promising approach for solving partial differential equations (PDEs). Their training, however, is often difficult because multiple loss terms induced by PDE residuals and boundary or initial conditions must be optimized simultaneously. To address this difficulty, existing approaches often construct update directions by explicitly enforcing particular desirable properties, such as scale robustness and simultaneous descent. While effective in many cases, such property-by-property designs can make it unclear which conditions are essential, what geometric principle determines the selected update direction, and how different methods are structurally related. In this work, we formulate update-direction selection for PINN training as a Chebyshev-center problem in the dual cone. The proposed formulation selects a normalized direction that maximizes the minimum distance to the cone facets. The resulting formulation admits an efficient dual problem in a much lower-dimensional space and yields a convergence guarantee in the nonconvex setting. It also recovers the key desirable properties targeted by existing approaches without imposing them separately; rather, they follow from the single geometric criterion underlying the formulation. This makes the selected direction interpretable through a single geometric rule and provides a unified basis for systematically comparing related direction-selection methods. Experiments on several PINN benchmarks further demonstrate strong empirical performance of the proposed method.
Read the full abstract

All authors Hoyeol Yoon, Seoungbin Bae, Nam Ho-Nguyen, Dabeen Lee

Mathematical Physics · Cross-listed from cond-mat.mes-hall

Localization phase diagram of the Hexagonal Lattice with irrational magnetic flux

We study the Hofstadter model on a hexagonal lattice with irrational magnetic flux in this work. The Hofstadter model of the square lattice with irrational flux has been solved mathematically by Avila and his collaborators in his Fields medal work. However, this theory is usually not applicable to lattices with internal degrees of freedom, such as spin or sublattice. In this work, we show that for the hexagonal lattice with only nearest neighbor hopping, the system can still be characterized by a two by two transfer matrix and solved exactly by the Avila global theory although this lattice has two sublattices. We obtained the exact localization phase diagram of the hexagonal lattice with irrational flux by this theory, which reveals three pure phases, i.e., the extended, localized and critical states but no mobility edge due to the chiral symmetry. We used the renormalization group (RG) theory to verify these results, which can determine part of the phase diagram. We then computed the fractal dimension of the remaining part numerically. The results from both the RG theory and numerical analysis confirmed the phase diagram we get from the Avila global theory. Our results can be tested in various hexagonal Moire lattices and artificial superlattices in recent experiments.
Read the full abstract