arXiv Science⌕ Search

SEARCH · arXiv Science

Search arXiv Science

Search indexed arXiv papers on artificial intelligence, large language models, computer vision and robotics. Read source abstracts and follow links to arXiv.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,531 records · Page 85Linked to original sources
arXiv2026

Softmax Attention on Gaussian Mixtures: Linear When It Can, Selective When It Must

Softmax attention, at the heart of Transformers, has demonstrated remarkable capabilities. Yet its underlying mechanisms remain only partially understood. Recent theoretical work studies Gaussian prompts, where the infinite-prompt limit reduces softmax attention to a linear map, but also removes the query-dependent selection that distinguishes it from linear attention. This work studies the infinite-prompt limit of softmax attention on Gaussian mixtures, which retain the tractability of Gaussian data while introducing latent structure, multimodality, and nonlinear dependencies. We show that softmax attention can represent and learn, via gradient-based methods, optimal solutions to a range of statistical tasks, including supervised classification and denoising. Our results highlight two complementary capabilities of softmax attention: it can recover linear tasks as effectively as its simpler linear counterpart, while also exploiting query-dependent context selection to solve nonlinear tasks beyond the reach of linear attention.

stat.ML↗
arXiv2026

What powers the most energetic flares in the Kepler field? Revisiting the superflaring star KIC 2852961

Flare activity in red giants is quite common when continuously monitored with high-precision space photometry. In fact, the most powerful flares, also known as "superflares" have been observed on red giant stars. KIC 2852961 is a red giant star producing the highest-energy flares in the Kepler field, and is also known for its large-amplitude starspot-induced light variations. Our goal is to find out what causes the extreme flare activity and flare energies in this star. For our investigations, we use ground-based spectroscopic measurements and space photometric data, which we process using standard tools. We also collect relevant data from databases of large surveys such as Gaia. Our new radial velocity data confirm that the red giant star is part of a close binary system, in which tidal forces play a crucial role in controlling the evolution of the binary system. Our results show that tidal evolution has driven the system close to orbital circularization and rotational synchronization, but it did not completely constrain the differential rotation of the red giant, in which the magnetic dynamo can operate extremely efficiently. We find that high-energy (super)flares are primarily controlled by the internal magnetic activity of the active giant component, rather than phase-dependent interactions with the secondary star. The large spot-induced rotational light modulations indicate extensive starspots that can store enormous amounts of magnetic energy. This provides the conditions for the formation of extremely large flares, confirming the close relationship between large spot coverage and high flare energies and/or frequency. Behind all this, differential rotation may play a dual role in the flare activity of KIC 2852961: it contributes both to the formation of large starspots and to the shearing and twisting of coronal magnetic fields that can power energetic superflares.

astro-ph.SR↗
arXiv2026

Sharp commutator bounds on complex simple Lie algebras: the Böttcher--Wenzel inequality and the comass of the Cartan $3$-form

For a complex simple Lie algebra $\mathfrak{g}$ with compact conjugation $τ$, Killing form $B$, and Hermitian form $H(x, y) = -B(x, τy)$, it is shown that $H([x,y],[x,y]) \leq (h^{\vee})^{-1}H(x,x)H(y,y)$, where $h^{\vee}$ is the dual Coxeter number, with equality exactly for $H$-orthogonal pairs in $τ$-stable long-root subalgebras isomorphic to $\mathfrak{sl}(2, \mathbb{C})$. Equivalently, the comass of the Cartan $3$-form with respect to $H$ equals its comass on the compact real form: the spectral norm of the Cartan $3$-form does not increase under complexification. For the Frobenius norm on the image of a representation of Dynkin index $\ell$ the optimal constant is $2/\ell$; this contains the Böttcher--Wenzel and Bloch--Iserles inequalities, extends the latter to complex skew-symmetric matrices, and improves them substantially for the exceptional Lie algebras of types $F_{4}$, $E_{6}$, $E_{7}$, and $E_{8}$. The proof uses the Nahm algebra of $\mathfrak{g}$ with the conjugation induced by $τ$: critical points of the relevant function correspond to $τ$-twisted idempotents, and its Hessian is governed by $τ$-twisted multiplication operators, whose anticommutation with multiplication by $i$ turns the one-sided second order condition at a maximum into a two-sided spectral bound. Subalgebras isomorphic to $\mathfrak{sl}(2, \mathbb{C})$ of Dynkin index $j$ give critical points with critical value $1/(jh^{\vee})$, whose Morse indices are computed, and the set of maximum points is a nondegenerate critical manifold. Geometrically, the maximal complex sectional curvature of a compact simple Lie group equals its maximal sectional curvature.

math.RT↗
arXiv2026

ObliVul: Alert-Conditioned Safety Obligation Modeling and Bidirectional Counterfactual Validation for Code Vulnerability Detection

In real-world software development, the primary challenge in vulnerability detection is often not finding suspicious code, but identifying which alerts among the large number of candidate alerts produced by static analysis truly warrant attention. Existing learning-based methods mainly identify suspicious patterns at the function or line level, making it difficult to extract complete program evidence centered on an individual alert. Although large language models can infer risk sources, dangerous operations, protection conditions, and state preconditions from local program facts, such semantic information cannot be reliably aligned with specific program nodes, dependency relations, and propagation paths, and is therefore insufficient to verify whether the corresponding safety obligations truly affect the current alert. To address this problem, we propose ObliVul, an alert-conditioned safety obligation modeling and bidirectional counterfactual validation framework for vulnerability detection. For each candidate alert, ObliVul first extracts a Local Evidence Pack (LEP) from the Code Property Graph (CPG) and uses a large language model to recover candidate safety obligations. It then aligns the safety obligations with program nodes, dependency edges, and path scopes to construct a Local Safety Obligation Graph (LSOG). Finally, VAFI aggregates complementary verified alert evidence while suppressing redundant or weaker evidence to produce a function-level vulnerability prediction. Experimental results show that ObliVul effectively distinguishes vulnerable versions from fixed versions and reduces persistent false positives on fixed code. Ablation studies further confirm the necessity of each component for safety obligation recovery, risk response validation, and function-level vulnerability inference.

cs.SE↗
arXiv2026

Loyalty of the affine line in motivic homotopy theory

We show that integers act invertibly on a derived scheme $S$ if and only if they act invertibly on the reduced motivic suspension spectrum of the affine line in $\mathrm{MS}_S$. As a consequence, over regular locally noetherian schemes and derived schemes on which some nonzero integer acts as zero, the motivic sphere spectrum is $\mathbf{A}^1$-invariant after inverting primes that are not invertible in the base scheme. In particular, over regular locally noetherian $\mathbf{Q}$-schemes, the motivic sphere spectrum is $\mathbf{A}^1$-invariant.

math.AG↗
arXiv2026

PRAXIS: Learning Dynamics of Self-Improving Models with Symbolic Archives

Self-improving learning systems adapt data selection, optimization, and auxiliary symbolic components, inducing nonstationary objectives outside standard learning assumptions. We introduce \textsc{PRAXIS}, a co-evolutionary framework that models generators, learners, and symbolic archives as interacting dynamical processes. We prove that KL-constrained generator updates and controlled archive-weight movement bound one-step objective drift, that archive updates suppress a program relative to any fixed comparator with a persistent cumulative utility advantage under sub-Gaussian noise, and that stochastic gradient descent achieves an average-stationarity guarantee whose degradation is governed by cumulative objective drift. Experiments across visual robustness, relational graph reasoning, and algorithmic graph reasoning exhibit generator stabilization, decreasing learner loss, and archive concentration consistent with these theoretical mechanisms.

cs.LG↗
arXiv2026

A generalization of the sum of element orders function

In this paper, for a finite group $G$ and a positive integer $k$, we introduce the function $$ψ''_k(G)=\dfrac{ψ_k(G)}{|G|^{k+1}}, \quad \text{where} \quad ψ_k(G)=\sum_{g \in G}o(g)^k,$$ as the $k$-th generalization of the function $ψ''$, introduced by M. T{ă}rn{ă}uceanu. Different thresholds for cyclic, abelian, nilpotent and (super)solvable groups in terms of $k$ are obtained. It is shown that these thresholds often outperform similar thresholds derived from other parameters like $ψ'$ and $ψ''$. Moreover, considering $ψ''_k$ as a $(0,1]$-valued function, we study its injectivity, surjectivity and density.

math.GR↗
arXiv2026

New Contraction Bounds for Multidimensional Asymptotic Consensus in Dynamic Networks

This work studies the multidimensional asymptotic agreement problem in dynamic networks. We assume that nodes have $d$-dimensional inputs and need to converge arbitrarily close to each other's inputs while their vectors stay inside the convex hull of all inputs. We present a new upper bound of $\sqrt{2d/(3d+1)}$ on the contraction rate of asymptotic agreement in the non-split network model, where $d$ is the dimension of the input and $\sqrt{2d/(3d+1)}\rightarrow \sqrt{2/3}\approx 0.816$ for $d\rightarrow\infty$. To this end, we adapt the BallMidpoint algorithm (Melnyk, 2026) recently introduced for the all-to-all communication model. This algorithm lets the nodes choose the midpoint of the smallest enclosing ball of the received vectors. This bound is strictly worse than the $\sqrt{1/2}\approx 0.707$ contraction rate for fault-tolerant approximate agreement in all-to-all communication networks because the convex hulls of the nodes do not have a common intersection, and only intersect pairwise. Our bound improves over the previously best-known contraction rate of $\sqrt{7/8}\approx 0.935$ for dynamic networks via the MidExtremes algorithm (Függer and Nowak, 2018). We present the first multi-dimensional lower bound of $2/3$ for the contraction of coordinate-free memoryless anonymous deterministic algorithms in non-split networks, for $d\ge 8$. This result shows that the previously best-known lower bound of $1/2$ on the contraction rate for $n\ge 3$ and $d=1$ is not tight in higher dimensions. If $n$ is infinite, we extend this lower bound to $\sqrt{1/2\cdot d/(d+1)}$, which converges to $\sqrt{1/2}$ for $d\rightarrow \infty$. We further show that for a single round, the one-round contraction is at least $\sqrt{\frac{2d-2}{3d+7}}$, which converges to $\sqrt{2/3}$ for $d\rightarrow\infty$, showing that the BallMidpoint strategy is asymptotically optimal for one round.

cs.DC↗
arXiv2026

Grey Random Fields on the Sphere

We introduce a new and tractable class of non-Gaussian isotropic random fields on the sphere, which we call, for brevity, Grey Spherical Random Fields (GSRFs). The proposed framework provides a flexible extension of the classical Gaussian theory, while preserving strong isotropy and embedding Gaussian random fields as a limiting special case. The construction is based on random spherical harmonic coefficients obtained as Gaussian mixtures driven by multivariate subordinators evaluated at independent random times (with Mittag-Leffler or more general distribution). This representation allows us to derive explicit characteristic functions, mixed moments and cumulants of arbitrary order. We also investigate asymptotic high-frequency behaviour and identify general conditions under which suitably normalized spherical harmonic coefficients converge to Gaussian scale mixtures, recovering several important models, including stable, Gamma and inverse Gaussian cases, as particular examples. Proving that GSRFs emerge as scaling limits of multivariate continuous-time random walks, with heavy-tailed waiting times, provides a microscopic foundation for the proposed models. We further obtain quantitative bounds measuring the deviation from Gaussianity of the spherical harmonic coefficients. The resulting framework substantially broadens the class of analytically tractable isotropic random fields on the sphere, while preserving explicit probabilistic representations and flexible dependence structures.

math.PR↗
arXiv2026

Photoinduced Odd- and Mixed-Parity Magnetism in Collinear Antiferromagnets

Altermagnetism, characterized by compensated collinear magnetic moments in real space and momentum-dependent spin splitting with even-parity symmetry, has recently emerged as a distinct form of magnetic order beyond the conventional ferro- and antiferromagnetic paradigms. However, realizing odd- and mixed-parity magnetism in fully compensated collinear magnets remains challenging. Here, we demonstrate that light irradiation can induce odd- and mixed-parity magnetism in conventional collinear antiferromagnets. Taking monolayer CoS and CoSe as representative systems, and combining Floquet theory with first-principles calculations, we show that circularly polarized light (CPL) induces odd-parity $f$-wave spin splitting, whereas elliptically polarized light (EPL) induces odd-parity $p$-wave spin splitting. Uniaxial strain combined with CPL provides an alternative route to realizing the $p$-wave state. More remarkably, applying an electric field together with CPL or EPL transforms the spin splitting into mixed-parity forms. The light-induced breaking of $\mathcal{PT}$ symmetry, together with spin-orbit coupling, further gives rise to a finite anomalous Hall response. Our work establishes a versatile strategy for engineering odd- and mixed-parity spin splitting in compensated antiferromagnets, offering a route toward light-controlled electronic and magnetic phenomena.

cond-mat.mtrl-sci↗
arXiv2026

Mixing time and isoperimetry of long-range percolation in the diffusive regime

We consider long-range percolation (LRP) on $\mathbb{Z}^d, d \ge 1$ with connection probabilities $\mathbf{p}(x, y) \approx \fracβ{\|x-y\|^s}$ for $β>0$ and $s > \min(2d, d+2)$. Our main result is that the mixing time of the random walk on the giant component of supercritical LRP inside a box of side length $n$ is of order $n^2$. To establish the upper bound, we show that a $d$-dimensional isoperimetric inequality holds almost surely simultaneously for all connected sets inside the giant component that are large enough.

math.PR↗
arXiv2026

Seek-and-View Reasoning for Multi-View Spatial Understanding

Existing approaches to multi-view spatial reasoning operate largely on sparse input views. Vision-language models (VLMs) are thus restricted to understand a scene and infer spatial relations within these fixed views, leading to fragile cross-view alignment and geometry-to-language bottleneck. To address these issues, we formulate a novel Seek-and-View reasoning approach to find implicit cross-view spatial evidence by locating a question-relevant view to support the spatial reasoning. To realize this approach, we propose Vantage, a training-free model-agnostic reasoning framework that pairs a VLM with a 3D foundation model: a viewpoint-grounded reasoning stage for question analysis and view planning, followed by a geometry-grounded evidence augmentation stage to effectively synthesize and incorporate visual evidence into the final reasoning. Comprehensive experiments on six VLMs demonstrate consistent improvements on five benchmarks without fine-tuning. Overall, by revealing spatial evidence through view-grounded reasoning, Vantage can largely reduce reliance on language-based cross-view alignment and improve multi-view spatial understanding. Our code is available at https://github.com/q1xiangchen/Vantage.

cs.CV↗
arXiv2026

Idempotence criteria for lazy cellular automata

A lazy cellular automaton $τ:A^G\to A^G$ is determined by a finite neighborhood $S\subseteq G$ containing the group identity, a pattern $p\in A^S$, and a writing symbol $a\in A\setminus\{p(e)\}$: its local rule changes the central symbol to $a$ precisely when the neighborhood pattern equals $p$. For a nonempty set $T$ of positions where $p$ takes the value $a$, we introduce two compatibility conditions on $p$ along the right translates $St$, $t\in T$. We show that the existence of a compatible set is necessary for $τ$ to be non-idempotent over every group, and that it is also sufficient whenever every position where $p$ takes the value $a$ commutes with every position where $p$ takes the value $p(e)$. In particular, compatibility characterizes non-idempotence over abelian groups, replacing a search over patterns on $SS=\{st:s,t\in S\}$ by a search over sets of positions where $p$ takes the value $a$. Finally, for every dihedral group $D_n$ with $n\geq 4$, we construct an idempotent lazy cellular automaton that admits a compatible set, showing that the commutation hypothesis cannot simply be dropped.

math.GR↗
arXiv2026

Deterministic Approximation of the Total Variation Distance Between Spin Systems

We study deterministic relative approximation of the total variation distance between two Gibbs distributions induced by spin systems on the same bounded-degree graph. For the hard-core model, we give a deterministic polynomial-time $\varepsilon$-relative-error approximation when both external-field vectors lie in $[b,(1-η)λ_{\mathrm c}(Δ)]^V$, where $b>0$ and $η\in(0,1)$ are fixed and $λ_{\mathrm c}(Δ)$ is the hard-core uniqueness threshold. For the Ising model, we obtain deterministic polynomial-time algorithms in two settings: the ferromagnetic Lee--Yang regime and the antiferromagnetic correlation-decay regime. We develop a new deterministic framework that reduces this task to estimating suitably chosen partition functions.

cs.DS↗
arXiv2026

The Hand of God Revisited: Revealing the Shell Structure and Jet-Supernova Remnant Interaction in MSH 15-52

We present an X-ray analysis of the supernova remnant (SNR) MSH 15-52 associated with the young rotation-powered pulsar PSR B1509-58, using Chandra, XMM-Newton, and eROSITA, complemented by MeerKAT radio data. We report the discovery of the missing southern counterpart to the bright northern RCW 89 region and identify a foreground X-ray source that may be a previously unidentified SNR candidate. Imaging and spectroscopy with XMM-Newton and eROSITA reveal the shell properties of MSH 15-52 and suggest that both the southern and northern shell-like structures are consistent with shocked interstellar medium associated with the forward shock, while their low ionization timescales and possible hard power-law components indicate possible influence from particle injection or additional heating by the pulsar wind nebula (PWN) jets. New Chandra spectral maps of the southern jet reveal a region of hard non-thermal emission, with a photon index of approximately 1.6, suggesting interaction between the PWN and the reverse shock and possible re-acceleration of the pulsar outflow. Spatially resolved spectroscopy of the RCW 89 northern clumps reveals high-velocity, metal-rich ejecta with enhanced Ne and Mg abundances. Comparison with core-collapse nucleosynthesis models favors a massive progenitor with an initial mass of approximately 29 solar masses in one-dimensional yield models, while asymmetric explosion models show that lower-mass progenitors can also reproduce the observed abundance ratios. Our study supports a picture in which RCW 89 is part of a larger asymmetric SNR shell interacting with energetic pulsar jets, making the system a valuable laboratory for probing pulsar-PWN-SNR interaction.

astro-ph.HE↗
arXiv2026

Gauss--Chebyshev Quadrature, $\mathbb S^1$-Designs, and the Planar Minkowski Inverse Problem: A One-Dimensional Model for Design-Induced Convex Geometry

We study the relation between equal-weight circle quadrature, Gauss--Chebyshev quadrature, and planar Minkowski reconstruction. The regular $N$-point grid is an $\mathbb S^1$-design of strength $N-1$; a half-step rotation of the $2M$-point grid projects to the classical $M$-point Gauss--Chebyshev rule. A common family of quadrature-error functionals gives exact identities for integration error, mixed-area error, and support-function error. For $N\ge3$, the centered Minkowski polygon $P_N$ with perimeter $2π$ has the exact Hausdorff distance from the unit disk $B^2$ \[ d_H(P_N,B^2)=1-\fracπ{N}\cot\fracπ{N} =\frac{π^2}{3N^2}+O(N^{-4}). \] With Steiner centering, we obtain $d_H(K_μ,K_ν)\leπW_1(μ,ν)$ for balanced, nondegenerate probability measures, where $W_1$ uses geodesic distance. Its exponent is sharp even for smooth positive densities near the uniform measure. The identity $W_1(ν_N,σ)=π/(2N)$ yields a general $O(N^{-1})$ bound for the regular grid. The sharper $O(N^{-2})$ rate follows from the sparse discrepancy spectrum and the inverse multiplier $(1-k^2)^{-1}$. Finally, a perturbation estimate gives sufficient conditions for preserving the quadratic rate and its leading constant.

math.NA↗
arXiv2026

Fast Pose Tracking of Rigid Objects with Compact Pose Graph Optimization

Tracking a novel object's 6D pose over long horizons currently requires either expensive onboarding or a reconstruction maintained throughout the sequence. This makes current trackers impractical for robotic manipulation and augmented reality, which need trackers that are ready to use and run in real time. We show that a lightweight tracking module can be applied on top of a wide range of correspondence estimation methods to keep drifts bounded while maintaining fast runtime. Our key idea is to avoid point-based optimization in the pose graph and operate only on relative pose constraints, which we weight by a derived uncertainty from the geometric alignment. This makes optimization independent of the number of correspondences while avoiding the direct inclusion of noisy point measurements, leading to fast and robust long-term tracking. Across four real-world benchmarks, our approach achieves tracking accuracy comparable to reconstruction-based trackers with a fraction of the optimization cost. Overall, these results suggest that a compact and reliable pose graph optimization can provide long-horizon consistency at substantially lower computational cost.

cs.CV↗
arXiv2026

Chaos in the Text: Revealing the Modality Preference in Mixed-Modality Retrievers

Dense retrievers have made significant progress on text and image corpora, but whether these capabilities extend reliably to mixed corpora containing text, image, and fused text-image documents remains unclear. In this paper, we systematically examine retrievers across architectures and find that their performance is highly sensitive to modality composition. As image documents are progressively replaced with semantically corresponding text representations, retrieval performance follows a pronounced V-shaped curve, remaining strong on single-modality corpora but degrading substantially when modalities coexist. In particular, irrelevant text causes more severe degradation than an equal number of irrelevant images, a phenomenon we term Chaos in the Text. Further analysis reveals modality preference, whereby text representations receive systematically higher similarity scores, allowing irrelevant text to outrank relevant images. To mitigate this bias, we introduce Trident, which constructs text, image, and fused text-image views of each document as co-equal positives and jointly optimizes relevance discrimination and positive-view balance through Multi-Positive View InfoNCE. Experiments across visual document and natural image benchmarks show that trident improves mixed-modality retrieval on both CLIP-based and VLM-based architectures, reduces sensitivity to modality composition and text distractors, and increases average single-modality retrieval performance.

cs.IR↗