Peer Revenue Returns subject to peer review

Papers

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

All of arXiv

Wednesday 2 September

Materials Science

First Principle Analysis of the Magnetism and Electronic Structure of Fe2XSi (X=Ti, V)

The electronic, magnetic, and mechanical properties of Fe2XSi, where X is titanium (Ti) and vanadium (V), are investigated using computational methods. Volume optimization reveals that the ground state of Fe2TiSi is a nonmagnetic narrow-band gap semiconductor, and that of Fe2VSi is a ferrimagnetic metal. The negative formation energies of both materials confirm the stability of their crystal structures. Mechanical properties confirm the static stability of the crystal structure and suggest that the titanium compound is ductile; however, the vanadium material is brittle. Density of states and band structure studies confirm the nonmagnetic semiconducting nature with an indirect band gap at {\Gamma} and X for the titanium material and the ferrimagnetic-metallic nature of the vanadium material. Electronic and magnetic properties of the materials were investigated with applied tension (negative pressure) and compression (positive pressure) to the crystal structure. The pressure study on Fe2TiSi shows a tunable band gap and a possible semiconductor-metal transition, and Fe2VSi shows a tunable magnetic moment and a possible low-spin/high-spin magnetic transition.
Read the full abstract

All authors Hanieh Kachooee, Cindy Kim, Jason Carbajal, K. Hettiarachchilage, N. Haldolaarachchige

Operator Algebras

Noncommutative majorization

We introduce a theory of noncommutative majorization that extends the classical majorization theory introduced by Hardy, Littlewood and Pólya to tuples of self-adjoint matrices that do not necessarily commute. We define and characterize a noncommutative majorization order that extends the classical majorization order. The definition is in terms of convex noncommutative functions, and we utilize the noncommutative convexity theory and noncommutative Choquet theory recently introduced by Davidson and the first author. As an application, we obtain a new necessary and sufficient condition for the existence of a trace-preserving completely positive map, i.e. a quantum channel, that interpolates between two finite sets of matrices. We give examples demonstrating that it is not always possible for this map to be chosen mixed unitary, even locally. We also address the computational difficulty of verifying the noncommutative majorization order. Our results further apply beyond the tracial case, and we obtain more general results characterizing the existence of unital completely positive maps that preserve an arbitrary faithful state or even unital completely positive map.
Read the full abstract

Computer Vision

AutoCompass: Accurate Visual Localization on Public Maps by Learning from Weak Labels

Neural map matchers estimate an image's 3-DoF pose relative to a 2D map. These models are trained on large-scale datasets of geo-referenced images, whose position and heading labels often contain noise that affects the trained models. To address this, we present AutoCompass, a supervision approach for training neural map matchers from inaccurate absolute pose labels. First, we show that heading labels are unnecessary: trained from raw GPS labels, models learn to predict accurate headings, automatically. Second, defining a tolerance region around raw GPS improves positional accuracy. Third, if available, our supervision uses relative poses between training images, obtained via SLAM or SfM, which provide a more accurate training signal. Across driving and egocentric benchmarks, AutoCompass consistently outperforms counterparts trained with the usual strong reliance on absolute pose labels.
Read the full abstract

All authors Javier Tirado-Garín, Alan Savio Paul, Shuai Chen, Axel Barroso-Laguna, Tommaso Cavallari, Daniyar Turmukhambetov, Victor Adrian Prisacariu, Eric Brachmann

General Economics

Dutch Books for Language Models

People increasingly use language models to support life decisions. Many such decisions involve a probabilistic forecast: How likely is a major life event, a natural disaster, or an economic outcome? Users of language models may implicitly trust that these forecasts fall out of a coherent world model. In this paper, we evaluate the coherence of language model probabilistic forecasts through a procedure that builds on a theorem due to de Finetti. We elicit forecasts from language models across events generated from stock returns data. We then use linear programs to compute the largest Dutch-book profit - the profit an arbitrageur could guarantee by betting against model-generated probabilities - which we use as a measure of incoherence. Our procedure does not require outcome labels, so we can evaluate coherence even in settings where outcomes are not observed or have not yet resolved. We find substantial evidence of incoherence in language model forecasts. Such incoherence increases when there are richer logical relationships between events, and irrelevant contextual details can increase incoherence by an order of magnitude. We conclude by discussing how alternative training strategies may improve probabilistic coherence.
Read the full abstract

Computation and Language

DiscoSign: Discourse-Aware Text to Sign Language Gloss Translation

Sign language processing systems have traditionally operated at the sentence level, ignoring critical discourse phenomena fundamental to sign language comprehension. We introduce DiscoSign, a computational approach for discourse-aware text to sign language gloss translation grounded in linguistic research. We address three key phenomena within our modular Large Language Model (LLM)-based translation framework: (i) spatial coreference resolution, where entities maintain consistent spatial locations throughout discourse; (ii) Question-Answer Clauses (QACs), pseudocleft structures serving specific discourse functions; and (iii) concept-gloss consistency, ensuring stable mappings between English concepts and American Sign Language (ASL) signs. Traditional translation metrics fail to capture discourse-level quality, so we introduce a suite of novel evaluation metrics designed to assess each dimension of discourse coherence addressed by our framework. Experiments on sentence-level and discourse-level datasets show that our approach for discourse-aware processing significantly improves spatial consistency and entity tracking relative to sentence-only translation, while maintaining competitive single-sentence gloss translation quality. Our work establishes the first systematic framework for discourse-level text to sign language gloss translation with corresponding evaluation methodology.
Read the full abstract

All authors Vasileios Baltatzis, Mert Inan, Connor Gillis, Raja Kushalnagar, Lorna Quandt, Leah Findlater, Colin Lea

Analysis of PDEs

Global smooth behavior in Kuznetsov and Westervelt type viscous wave equations: A unifying approach covering $W^{1,q}$-small initial data

In a smoothly bounded domain $\Om\subset\R^n$ with $n\geq 1$ and $a>0$, we consider an initial-boundary value problem for the general viscous wave equation \bas h(u,u_t) u_{tt} = \Del u_t + a\Del u + f(u,u_t,\na u,\na u_t) \eas which appears in models of nonlinear acoustics wave propagation; well-established equations of Kuznetsov and Westervelt type form particular examples.\abs % While the existing literature offers extensive results on global solutions for sufficiently small initial data $(u_{0}, u_{0t})=(u, u_{t})|_{t=0}$ in second- and higher-order Sobolev spaces it appears to remain open how far global solvability can be established under smallness conditions involving only first-order Sobolev spaces. The present manuscript addresses this question by proving the existence of global classical solutions together with exponential decay of the pair $(u,u_{t})$ in $ W^{1,r}\times W^{1,p}$-Sobolev spaces whenever the nonlinearities $h$ and $f$ are sufficiently smooth and are such that $h(0,0)>0$ as well as $f(0,0,0,0)=0$ and $\na f(0,0,0,0)=0$.
Read the full abstract

High Energy Physics (Phenomenology)

MadSpin2: automated spin-entangled decays of heavy resonances via the spin-density matrix formalism

The simulation of processes that contain multiple particles from the decay of unstable resonances becomes computationally challenging both at leading order and beyond. This paper presents MADSPIN2, a new version of MADSPIN, a dedicated tool tackling this problem, primarily for events generated with the MG5AMC event generator. While still based on the Frixione-Laenen-Motylinski-Webber formalism, the new version leverages spin-density matrices to lift most of the limitations of the previous implementation. It supports loop-induced processes, N -body decays and samples defined by amplitude-level interference between distinct hard-process amplitudes. It also introduces the multiple-pole approximation and a resonance-helicity decomposition, including off-diagonal helicity-interference contributions. These developments substantially broaden the phenomenological scope of MADSPIN, enabling applications that were previously inaccessible within the framework, several of which are demonstrated in this work.
Read the full abstract

Quantum Physics

Variational preparation of thermofield double states for SYK models via multi-angle QAOA: sequential angle pruning for circuit reduction

Variational preparation of thermofield double (TFD) states can require deep quantum circuits, particularly for interacting many-body systems. Reducing these circuits while retaining high fidelity is therefore crucial for TFD-state preparation on noisy quantum processors. We study this problem by applying the multi-angle quantum approximate optimization algorithm (ma-QAOA) to TFD-state preparation and introducing two top-down sequential angle-pruning algorithms. Starting from the optimized initial ma-QAOA circuit, both algorithms sequentially remove Pauli-string evolutions with small optimized angles and reoptimize the remaining parameters after each removal. We apply these algorithms to Gaussian and binary Sachdev–Ye–Kitaev (SYK) models in both dense and sparse cases. We find that ma-QAOA prepares the target TFD states with high fidelity and that sequential small-angle pruning retains high fidelity while reducing the circuit depth, particularly at low temperature. Moreover, using the post-reoptimization cost in sequential small-angle pruning further improves the fidelity. For the binary sparse $N=10$ SYK model at $\beta=10$, $88.8\%$–$92.1\%$ of the nonlocal Pauli-string evolutions are removed while retaining an average fidelity of approximately $95\%$. Finally, we propose extensions of the sequential pruning algorithms toward quantum–classical hybrid implementation.
Read the full abstract

Quantum Physics

Phonon-limited detection thresholds for genetically encoded fluorescent-protein spin-qubit relaxometry of neural radicals

The demonstration that enhanced yellow fluorescent protein hosts an optically addressable spin-1 qubit in its metastable triplet state raises the prospect of genetically encoded quantum sensing at molecular length scales. We develop a detection-limit theory for using this fluorescent-protein spin qubit (FPSQ) to sense paramagnetic neural signaling radicals by spin relaxometry. We derive the transition-resolved Redfield relaxation matrix of the zero-field-split triplet coupled to a diffusing radical bath, establish the regime in which it collapses to a single exponential, and validate it against Lindblad simulations and nitrogen-vacancy benchmarks. Propagating the effects of photon shot noise, photobleaching-grounded photon budget, and finite measurement bandwidth, we find that the native room-temperature sensor falls short of physiological sensitivity by six to eight orders of magnitude with the bottleneck being the phonon-limited intrinsic $\Tone$. Analyzing the underlying direct and two-phonon Raman processes, we show that room-temperature relaxation is Raman-dominated by $\sim\!720\!:\!1$ and that, because the Raman coefficient scales as $v^{-10}$ with sound velocity, a $\sim\!2\times$ stiffening of the chromophore environment recovers $\Tone\sim\SI{100}{\micro\second}$, sufficient for micromolar sensing. Nanomolar sensing is obstructed by a direct-process ceiling of \SI{79}{\micro\second} that vibronic decoupling alone cannot breach. We obtain quantitative design rules, identify photon yield as a co-equal bottleneck, and propose a frequency-resolved protocol for chemical specificity.
Read the full abstract

All authors Parul Raghuvanshi, Sagnik Ganguly, Sharika E, Mohana Priya T., Vishvendra S. Poonia

High Energy Physics (Phenomenology)

The $H^*H^*V$ couplings from light-cone sum rules

We present a determination of the strong charge couplings $g_{H^*H^*V}$ and magnetic couplings $f_{H^*H^*V}$ (with $H \in \{D, B\}$ and $V \in \{\rho, \omega, K^*, \phi\}$) using the framework of light-cone sum rules (LCSR). The theoretical precision is improved by establishing the leading-power hard-collinear factorization formulas with next-to-leading-order (NLO) $\alpha_s$ corrections, alongside the inclusion of power-suppressed contributions up to the next-to-next-to-leading power (NNLP) for both channels. By parameterizing the $\mathcal{O}(1/m_{H^*})$ power corrections, we extract the universal static couplings $\beta = 0.73 \pm 0.13$ and $\hat{\lambda} = 0.23 \pm 0.06 \text{ GeV}^{-1}$, highlighting their distinct heavy-quark scaling behaviors. While our investigation into SU(3) flavor symmetry breaking shows that these effects are currently overwhelmed by uncertainties in the non-perturbative vector meson distribution amplitudes, our determinations of these spin-conserving and spin-flip parameters provide non-perturbative inputs for modeling the S-wave central potentials, as well as the tensor forces and D-wave mixing effects, in exotic heavy tetraquarks.
Read the full abstract

Machine Learning (Statistics)

Full-Model Optimality for Tunable Linear Generative Priors in Compressed Sensing

Generative models have been studied experimentally and theoretically as priors for inverse problems such as compressed sensing. Recent work by Gunn et al. studied the use of generative priors with tunable complexity, where a family of generative priors with varying complexity is maintained and a specific complexity can be selected at inversion time. They demonstrated that lower reconstruction errors can be experimentally attained for a variety of inverse problems by appropriately tuning the complexity of the generative prior. In the present paper, we establish theory for compressed sensing in the setting of a tunable family of linear generative priors naturally related through their singular value decompositions. We prove that in noiseless Gaussian compressed sensing, the full-dimensional linear prior attains the minimum expected reconstruction error over the entire family of linear priors. Thus, in this idealized linear noiseless setting, tuning to a lower-complexity prior does not improve the expected reconstruction error. This result is in contract to the behavior of denoising, where lower complexity priors attain lower reconstruction errors due to a standard bias-variance tradeoff. This result indicates that the experimental benefits of tunability in compressed sensing with neural network priors arises due to nonlinearities in the generative models.
Read the full abstract

Software Engineering

ShikumiMiner: Mining Recurring Implementation Patterns in AI Codebases

Large language models are paving the way towards innovation by understanding, analyzing, summarizing and generating content in the modern world. Currently there are thousands of LLM projects developed by engineers in open-source repositories. However, whether these LLM projects have underlying patterns or not remains a question. Exploring these underlying patterns will give new dimensions to the developers who aim to develop these LLM projects. In this paper, we propose ShikumiMiner, a static-analysis framework that combines Abstract Syntax Tree (AST) and Control Flow Graph (CFG) features to detect and compare recurring implementation patterns in C++ local LLM codebases. We analyze ten GitHub open-source repositories and classify functions into seven study-specific categories using a multi-label Random Forest model. Studying these patterns can provide useful insights for developers aiming to design LLM applications.
Read the full abstract

All authors Afsana Tasnim, Saif Uddin Mahmud, Md Erfan, Md Rayhanur Rahman

Differential Geometry

Quantized Einstein Metrics on $S^7$ with $SU(3)$-Symmetry

We prove that the standard cohomogeneity-one action of \(SU(3)\) on \(S^7\), with principal orbit the Wallach flag manifold \(SU(3)/T^2\), admits infinitely many invariant Einstein metrics. The proof uses a detection approach to the Einstein boundary-value problem. Starting from one singular orbit, we follow the Einstein solutions only to a canonical hypersurface and measure there how far they are from closing smoothly. When the singular-orbit scale becomes small, this closing problem is governed by a Ricci-flat limiting solution. The linearisattion about its limiting cone has an oscillatory mode. As the scale shrinks, this oscillation repeatedly changes the sign of the closing error, producing infinitely many parameter values for which the metric closes. The closing scales satisfy an asymptotic logarithmic quantization law, with successive ratio tending to \(e^{-2\pi/\sqrt{15}}\). We also determine the geometry of the resulting sequence. Away from the two singular orbits the metrics converge to the singular sine cone over the non-normal Einstein metric on \(SU(3)/T^2\), while after rescaling by the square of the singular-orbit scale at either end they converge to the same complete Ricci-flat threshold metric. Their curvature is of order \(b_n^{-2}\), so the logarithmic phase law induces a corresponding quantization of the focal curvature scale.
Read the full abstract

Statistical Methodology

DOMIC: Provably Calibrated Detection of Dependence-Structure Change Points via Density-Operator Mutual Information

Dependence can change between variable blocks without changing correlation; serial dependence complicates calibration. We propose DOMIC, a rank-based detector using density-operator mutual information (DOMI). Unit-norm random features make sample-level partial traces exact, so prefix sums compute segment statistics at per-split cost independent of segment length; a Gram form uses the exact kernel. Permutation tests are finite-sample exact under pair or block exchangeability for windows and schemes fixed in advance; validity after data-driven scheme selection is unproved. For i.i.d. pairs with feature dimension, frequency draw and bandwidth fixed, we prove a weighted chi-square limit at independence for the rank-based random-feature statistic. An additive entropy cost with Holevo gains drives segmentation. DOMI attains higher localised power than rank, copula, kernel and distance baselines in most non-Gaussian and correlation-free settings; the Gram form leads or ties in all 16 non-Gaussian settings where any statistic exceeds 0.10. On a three-break benchmark, Holevo partitioning returns exactly three breaks in 47.5% of replicates, against at most 19% for calibrated baselines. Block-calibrated primary scans reject in all three pre-specified pairs of a US financial panel, with strongest stage-two evidence for stocks and Treasury yields. One of 27 weather candidates passes the multiplicity-corrected re-test; candidate selection is unadjusted.
Read the full abstract

Artificial Intelligence

SafeEvolve: Harness-Policy Co-Evolution from Agent Experience for Safety Alignment

The performance of LLM-based agents is jointly shaped by the base model and the harness used when interacting with the environment. This exposes them to safety risks in both harmful final responses and multi-step execution trajectories. Existing safety alignment mechanisms often rely on either external harness updates or policy optimization, yet applying either paradigm in isolation fails to bridge runtime control with intrinsic safety. We propose SafeEvolve, an experience-driven self-evolving framework for agent safety alignment. SafeEvolve leverages safety experience from completed on-policy trajectories to drive a continual loop of harness-policy co-evolution. On the harness side, SafeEvolve converts trajectory-level safety evidence into bounded, component-level updates across safety prompt and hierarchical skills, yielding auditable and reversible harness artifacts. On the policy side, SafeEvolve follows a two-stage SFT-RL paradigm, where harness-use SFT bootstraps the policy to actively leverage evolved harness artifacts, and harness-augmented RL further shapes autonomous safety behaviors during multi-step exploration via verifier-decomposed rewards. Through harness-policy co-evolution, SafeEvolve converts safety experience into an evolved runtime harness and improved policy behavior. Experiments on agentic safety benchmarks show that SafeEvolve achieves a stronger safety-utility tradeoff than existing baselines. For Qwen3.5-4B, SafeEvolve achieves a $3\times$ ASR reduction on AgentDojo while improving benign utility from 59.79% to 61.86%.
Read the full abstract

All authors Qinghua Mao, Wanying Qu, Dadi Guo, Leitao Yuan, Qingyu Liu, Yu Li, Guanxu Chen, Yanwei Fu, Xi Lin, Xia Hu, Dongrui Liu

Algebraic Geometry

Characterization of Stanley-Reisner varieties by their automorphism group

We study the automorphism ind-group of a Stanley-Reisner variety $X_\Delta$ through a combinatorial toolkit on the underlying complex: a facet closure operator, its Demazure roots, and the resulting dichotomy between exposed and hidden facets. Our main theorem is that a fully exposed complex, that is, one in which every facet has a private vertex, is recovered from the ind-group: $\mathrm{Aut}(X_\Delta)\cong\mathrm{Aut}(X_{\Delta'})$ forces $\Delta\cong\Delta'$. The hypothesis cannot be dropped, but it holds after one stabilization, so for arbitrary $\Delta,\Delta'$ an isomorphism $\mathrm{Aut}(X_\Delta\times\mathbb{A}^1)\cong\mathrm{Aut}(X_{\Delta'}\times\mathbb{A}^1)$ already forces $\Delta\cong\Delta'$. At the opposite extreme, a fully hidden $\Delta$ gives $\mathrm{Aut}(X_\Delta)=T_0\rtimes S(\Delta)$, never isomorphic to the ind-group of a non-rigid Stanley-Reisner variety. The criterion decides graphs and skeleta, and every complex is homotopy equivalent to a fully exposed one.
Read the full abstract

Analysis of PDEs

Maximizing the fundamental Laplace–Neumann eigenvalue on quadrilaterals

We prove that among convex planar quadrilaterals of equal area, the square uniquely maximizes the first nonzero Laplace–Neumann eigenvalue. This is the quadrilateral case of the Neumann analogue of the long-standing Pólya–Szegő conjecture, which asserts that the regular $n$-gon minimizes the first Laplace–Dirichlet eigenvalue among $n$-gons of equal area. The proof uses a Rayleigh–Ritz lower bound, obtained from the span of the first five nonconstant Neumann modes of the square, on a smooth auxiliary functional whose minimization implies the original maximization. Local maximality at the square follows from a symmetry-reduced Hessian, computed in closed form, together with a certified second-order difference-quotient test that establishes the local inequality on an explicit ball around the square. Global maximality is then established by a certified box covering of the parameter space, certifying on each box a direct upper bound on the first Rayleigh–Ritz eigenvalue. Both certifications operate entirely on integrals over a fixed reference square and avoid a posteriori finite-element bounds on the perturbed quadrilateral.
Read the full abstract

Computation and Language

EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction

Evaluating LLM agents is essential for guiding their development, yet it has grown prohibitively expensive: a single pass of a frontier model over an agentic benchmark can cost hundreds to thousands of dollars, a price paid repeatedly across iterative development cycles. Prior efforts, centered on benchmark distillation, reduce the number of evaluation tasks but leave the cost of executing each retained task untouched. In this work, we introduce early outcome prediction, a complementary axis of efficiency that instead cuts cost within each task. Our key insight is that an agent's final outcome is often evident from its intermediate behavior well before execution completes. We instantiate this idea in EarlyEval, a lightweight framework that trains a pair of LightGBM success and failure classifiers over behavioral, textual, and reference-solution features, and halts an agent run the moment either classifier crosses a calibrated confidence threshold, adding negligible per-step overhead. Across three benchmarks, SWE-bench Verified, TerminalBench, and Toolathlon, EarlyEval can eliminate 13%-26% of agent steps and up to 44.1% input tokens and 29.4% output tokens at 89%-97% prediction accuracy, while perturbing per-agent resolve rates by only one to two percentage points on average.
Read the full abstract

All authors Yuling Shi, Zhensu Sun, Junsen Dong, Chengcheng Wan, David Lo, Xiaodong Gu

Software Engineering

Type Hints in Python Libraries and Frameworks: An Empirical Analysis of Adoption and Maintenance

Context: In Python, type hints allow developers to annotate variables and functions with explicit type information, improving code clarity and reliability. Although type hints are widely available, little is known about how they are adopted and maintained in libraries and frameworks. Objective: We investigate the adoption, usage, maintenance, and rationale of type hints in Python libraries and frameworks. Method: We analyzed 1,000 popular GitHub repositories, identifying libraries and frameworks and extracting their type annotations. We examined annotation coverage, the locations and origins of annotations, their evolution across git histories, and the relationship between developer annotations and types inferred by Pyright. Results: Of the analyzed repositories, 91% of libraries use type hints at least once, although adoption is inconsistent. Among libraries with systematic usage, maintainers prioritize function parameters and return types, with median coverage of 45.8% and 35.9%, respectively, and mainly use built-in types (73.0%). When modified, annotations tend to migrate to more expressive types. Developers annotate members even when Pyright can infer their types, and these annotations often simplify the inferred type. Conclusion: Type hints in Python libraries and frameworks primarily serve as API contracts rather than comprehensive descriptions of implementation details. The findings suggest opportunities for tooling that prioritizes public interfaces, identifies meaningful annotation changes, and supports maintainers in evolving type information.
Read the full abstract

Quantum Physics

Information and reference requirements for auditing collider classifiers

Assertions about collider classifiers require observable evidence and a specified deployment reference. We distinguish the classification threshold that selects events from the claim threshold that defines acceptable performance. For fixed nominal-weighted ratio claims on a frozen finite population, we apply established assorter reductions and bounded-mean confidence sequences under independent sampling with replacement and a predeclared weight bound. The anytime-valid error control is per claim, not automatically simultaneous across claims, and is not a new statistical guarantee. In 19,680 archived paired audit-stream comparisons, changing unweighted correctness and its source reference to nominal-weighted accuracy and its own reference changed 536 supported verdicts to unresolved and 226 in the reverse direction. Scores, labels, weights and exact ratios were available before these simulated audits; their stopping times demonstrate no operational acquisition saving. A post hoc analysis of 30 finite-shot quantum-kernel realizations, with paired RAW and diagonal-loading treatments, describes sensitivity to operating rules, ideal versus realized references, and fixed versus recomposed margin groups. Verdict differences are not auditor error rates, and no quantum-specific mechanism or quantum advantage is established. A CMS open-data example supports restricted aggregate statements while event-level accuracy remains unidentified. This substantial revision withdraws the previous physical-interval coverage conclusions following numerical-validation problems; the available reevaluation does not reestablish them. The retained contribution is mathematical exposition and descriptive evidence, without methodological-priority or calibrated physical-inference claims.
Read the full abstract

All authors Roberto Fernández-Barrios, Iker Pastor-López, Asier González-Santocildes, Pablo García Bringas

2609.02781 · 2 Sep · 14 pages, 0 figures, 2 tables; separate 10-page Supplementary Material (S1–S9, 4 tables) provided as an ancillary PDF. Title changed. Substantial corrective revision: previous physical-interval coverage conclusions withdrawn after numerical-validation problems were identified; methodological attribution and scope revised; descriptive classifier results retained

Computer Vision

ShallowStream: Index Shallow then Answer Deep for Streaming Video Understanding

Streaming video understanding is a critical capability for real-world applications, including embodied intelligence, autonomous driving, industrial monitoring, surveillance and early warning, and wearable assistants. However, processing continuous video streams with multimodal large language models (MLLMs) is computationally expensive. Existing efforts have explored reducing streaming overhead through visual token pruning, token merging, quantization, on-demand frame retrieval, and context offloading. However, most existing methods overlook the dimension of model depth. Repeatedly executing full-depth MLLM prefill over incoming frames is prohibitively expensive, incurring substantial computational overhead and causing the KV cache to grow at a rate directly proportional to the prefill depth. To address these challenges, we propose ShallowStream, a novel framework that leverages the shallow layers of an MLLM to simultaneously perform frame encoding and retrieval index building. During stream processing, ShallowStream maintains an always-on lightweight index using the KV cache of shallow layers. During query-time answering, we leverage the attention scores generated by the shallow layers to score context frames and employ a diversity-aware selection strategy to retrieve precise and comprehensive evidence. ShallowStream achieves performance on par with the strongest existing streaming methods, while reducing per-frame prefill latency and 10-second end-to-end latency by up to 52.1x and 11.9x, respectively. Our code is available at https://github.com/CURRENTF/ShallowStream.
Read the full abstract

All authors Jitai Hao, Ke Yang, Di Yan, Fan Liu, Qiang Huang, Jun Yu

Analysis of PDEs

Sharp convergence rates for the vanishing discount problem with hyperbolic Aubry sets

Let $H\in C^2(T^*M)$ be a Tonelli Hamiltonian on a closed connected manifold and let $u_\lambda$ solve \[ \lambda u_\lambda+H(x,Du_\lambda)=c(H)\qquad\text{in }M. \] We study the convergence rate of $u_\lambda$ to the selected critical solution $u_0$. Assume that the lifted Aubry set is a finite union $\widetilde{A}=\Gamma_1\sqcup\cdots\sqcup\Gamma_N$, where each $\Gamma_i$ is either a hyperbolic equilibrium or a periodic orbit hyperbolic in the critical energy level. We prove \[ -C\lambda\le u_\lambda-u_0\le C\lambda|\log\lambda|. \] Let $\mu_i$ be the projected Mather measure associated with $\Gamma_i$. If \[ \int_M u_0\,\mathrm d\mu_i=0\qquad \text{for every }i, \] then \[ \|u_\lambda-u_0\|_\infty\le C\lambda. \] In particular, if the lifted Aubry set consists of a single hyperbolic equilibrium or a single hyperbolic periodic orbit, the convergence rate is $O(\lambda)$. We give examples showing that both convergence rates $O(\lambda)$ and $O(\lambda|\log\lambda|)$ are optimal. Without hyperbolicity, finite-order degenerate examples give lower bounds of order $\lambda^{1/(2r-1)}$ with $r\ge2$. We also construct infinite-order degenerate examples with arbitrarily slow convergence. Taken together, these results provide, to our knowledge, the first systematic quantitative theory for the vanishing discount problem in the Tonelli setting.
Read the full abstract

Combinatorics

Most $(0,1)$-polytopes are not normal

We prove that the proportion of $0/1$-equivalence classes of $d$-dimensional $(0,1)$-polytopes that are normal tends to zero at least at a double exponential rate as $d\to\infty$. As a consequence, the same holds for any of the following classes given by the type of triangulation possible: (a) quadratic, (b) flag unimodular, (c) regular unimodular, or (d) unimodular, among others. We classify the $0/1$-equivalence classes of $d$-dimensional $(0,1)$-polytopes for $d\leq5$ according to whether they admit a unimodular, flag unimodular, or quadratic triangulation. In dimension five, exactly $175$ out of $1{,}226{,}525$ classes have a flag unimodular triangulation, but no quadratic triangulation. Among them, there are polytopes whose toric rings are not Koszul; thus, we find the first polytopes that have a flag unimodular triangulation, but whose toric ring is not Koszul. In contrast with the matroid case, we exhibit a delta-matroid polytope that is not normal.
Read the full abstract

Signal Processing

A Dry-Contact Ear-EEG System With Continuous Electrode-Skin Impedance Mismatch Monitoring for Motion Artifact Cancellation Using DRL Stimulus

Dry-contact ear-electroencephalography (Ear-EEG) enables wearable neural monitoring. However, motion induced electrode-skin impedance (ESI) mismatches between electrodes can severely degrade signal quality. To the best of our knowledge, this paper presents the first proof-of-concept dry-contact EarEEG system that uses a driven-right-leg (DRL) stimulus for continuous ESI mismatch monitoring, enabling online adaptive motion artifact cancellation. A 1 kHz sinusoidal stimulus is injected through the DRL electrode. The resulting response to the injected carrier is separated from the EEG using bandpass filtering and demodulation, and then used to extract the ESI mismatch information as the reference input for a normalized least-mean-square adaptive filter followed by a Hampel filtering stage. To evaluate artifact suppression and preservation of neural activity, alpha-band EEG activity was analyzed involving four healthy participants performing head nodding, electrode tapping, and jaw clenching. The system achieved artifact power reductions of 6.5, 12.6, and 9.0 dB (77.6%, 92.8%, and 86.4%, respectively) while alpha-band modulation remained clearly observable after processing. This demonstrates the feasibility of DRL-stimulusbased ESI mismatch monitoring for motion artifact cancellation in wearable dry-contact Ear-EEG.
Read the full abstract

All authors Lohan Atapattu, Sajitha Madugalle, Imasha Nethmal, Erandee Jayathilaka, Avishka Herath, Kithmin Wickremasinghe, Simon L. Kappel, Nilan Udayanga, Chamira U. S. Edussooriya

Computer Vision

Video-Based Palm-Vein Authentication under Challenging Conditions

Palm-vein biometrics are increasingly used for secure, contactless authentication. Yet real-world deployment exposes them to surface noise (sweat, dirt), illumination and motion variation, and temperature-driven changes in vascular visibility, which remain underexplored for lack of data captured under such conditions. To study these effects, we introduce the Columbia University Palm-vein (CUP) dataset, to our knowledge the first public video-based palm-vein dataset. CUP records every palm under four surface conditions (a clean baseline, warm, wet, and dirty) and pairs each subject with physiological and demographic metadata. On it we benchmark twenty-one recognizers spanning static, video, and multi-frame aggregation architectures. Models that verify reliably on clean palms lose most of their accuracy on dirty ones, and the mean equal error rate (EER) roughly quadruples. We recover much of that robustness along both axes of the capture. Temporally, a consensus over the few frames the sensor already returns cancels transient corruption; spatially, a test-time matcher that adds no learned parameters fuses the global cosine with a saliency-steered region-level optimal transport that routes the comparison around corrupted regions. The full design leads on every surface of CUP in EER, TAR@FAR=0.01, and Rank-1, at 4.3M parameters and 3.1 GFLOPs, a fraction of the video models' cost. Attached to four frozen state-of-the-art backbones it cuts their mean EER by 29-37% without retraining, and on four public single-image datasets the regional matching alone still helps. A preliminary audit across ten demographic and physiological traits finds two warm-condition gaps, along body water and gender, that survive multiple-comparison correction. CUP will be released for non-commercial research use at https://github.com/MobileX-CU/CUP_v1 upon publication.
Read the full abstract

All authors Xiaofeng Yan, Kechen Liu, Abhilash Venkatesh, Cathy Zhang, Xia Zhou, Salvatore Stolfo

High Energy Physics (Phenomenology)

Strong Constraints on Higgsino Dark Matter from Solar Capture

Higgsino dark matter remains one of the most convincing dark matter candidates. It can exist in a “quasi-Dirac" regime, when the splitting between two Majorana components is on the order of $O(\rm few~\,MeV)$ or less. Owing to the strong gravitational pull of the Sun, the DM particles accelerate to up to $\sim 1400$\,km/sec in the solar core, which allows for inelastic scattering on heavy elements and very effective capture onto solar-bound trajectories. We evaluate the expected flux of the neutrinos from the expected $\chi\chi \to W^+W^-,ZZ$ annihilation for the cosmologically preferred mass, $m_\chi \simeq 1.08$\,TeV while keeping the mass splitting $\delta$ as a free parameter. Confronting it with the non-observation of high-energy neutrinos from the Sun by the IceCube experiment, we obtain a robust limit, $\delta>566$\,keV. These limits exclude the interpretation of the recent LZ event in terms of endothermic inelastic scattering of Higgsino dark matter on xenon nuclei.
Read the full abstract

Cryptography and Security

CodePoisonRAG: Knowledge Poisoning Attacks on Retrieval-Augmented Code Generation

Retrieval-Augmented Code Generation (RACG) improves LLM-based software development by retrieving external code artifacts, documentation, and patches, and incorporating them into the generation context. This reliance on external knowledge introduces a critical trust boundary: poisoned artifacts can influence generated code without modifying the underlying LLM. Prior work shows that selecting existing vulnerable examples can increase the general vulnerability rate of RACG outputs, but leaves open whether a black-box attacker can construct a single task-matched artifact that propagates an attacker-selected weakness. We introduce CodePoisonRAG, a targeted upstream knowledge-poisoning framework that transforms benign fixed-code entries into poisoned artifacts. Its attack chain combines CWE-specific Vulnerability Injection, which embeds a selected source-to-sink flow while retaining task alignment, with Semantic Mislabeling, which adds false safety claims without repairing the vulnerable behavior. The attacker has no access to the victim's deployed knowledge base, retriever, re-ranker, generator, prompt, or defense mechanism and injects at most one artifact per anticipated programming task. We construct 85 poisoned artifacts covering ten CWE classes across Java and C, yielding an aggregate corpus-poisoning ratio of 0.7%. Across three generators, all 85 artifacts appear among the Top-3 results for their corresponding queries, and CodePoisonRAG achieves attack success rates between 0.80 and 0.93. Against CodeGuarder, which injects vulnerability-specific security knowledge into the generation context, the attack retains success rates between 0.40 and 0.71. These results show that RACG poisoning extends beyond the incidental propagation of existing vulnerabilities to the targeted construction and propagation of attacker-selected weaknesses.
Read the full abstract

Theoretical Economics

Sequential Pricing Mechanisms for Surplus Division

Extending the Price-and-Choose (P&C) mechanism of Echenique and Nunez (2025), we propose the Price-Accept-and-Choose (PA&C) mechanism, which preserves efficiency while eliminating P&C's first-mover advantage. We then analyze randomized and bidding variants and show that the resulting equilibrium payoffs correspond to standard solutions in transferable utility games: the Center of the Imputation Set value for P&C and the Shapley value for PA&C. In the randomized variants, these solutions arise in expectation; in the bidding variants, they are implemented on every equilibrium path. We further relate other efficient designs, such as balanced VCG payments, to additional solution concepts, forging an interpretable bridge between implementation theory and cooperative game theory.
Read the full abstract

All authors Yukihiko Funaki, Yukio Koriyama, Matias Nunez, Giacomo Rostagno

Computation and Language

HyperStyler: Low-resource Authorship Style Transfer via Context-aware Style Navigation and Hypernetworks

Low-resource authorship style transfer (LAST) aims to rewrite text into the style of an arbitrary target author using only a few reference examples while preserving the original meaning. Existing methods often struggle to achieve both high style fidelity and semantic preservation because they compress diverse references into a single static author embedding, which averages out context-dependent stylistic variation, and rely on hidden representations for style control, which entangle style with content. We propose HyperStyler, a novel architecture that decouples LAST into style selection and style realization. Stylo-navigator predicts style coordinates by jointly modeling the source context and target-author references, and Stylo-hypernet realizes them via dynamic parameter modulation instead of hidden-state injection. Our experiments on Reddit, Blog, and News datasets demonstrate that HyperStyler consistently outperforms prior methods including LLM-based approaches and generalizes robustly across domains. Notably, HyperStyler achieves superior performance with as few as 2.4% additional parameters over T5-large, while being over 1.8x faster than LLMs at inference.
Read the full abstract

All authors Jongkyung Shin, Minguk Jeon, Chanwoo Park, Chiehyeon Lim

Computation and Language

From Reweighting to Rewriting: Unlocking the Intervention Effects of Influential Samples in Training Data Attribution

Training data attribution (TDA) aims to identify training examples that shape model behavior, but its intervention value depends on both which examples are selected and how they are modified. Influence functions (IF) estimate behavioral changes under infinitesimal reweighting, yet IF-selected examples often show limited advantages over random selection under conventional weight-based interventions. This raises the question of whether influential examples lack intervention value or whether reweighting fails to realize their behavioral leverage.We introduce influence-guided response rewriting, which uses IF to identify intervention targets and replaces their responses with behavior-aligned or behavior-opposed supervision while keeping instructions fixed. Across four open-weight LLMs, we compare rewriting and reweighting on the same influence-selected examples using epistemic abstention as our primary testbed. Response rewriting produces stronger, more persistent, and bidirectional behavioral shifts, while reweighting the same examples yields weak and inconsistent effects. Further analyses show that influence-selected examples provide greater rewriting leverage than alternative selectors, with changes remaining concentrated on target-relevant behaviors. The same qualitative contrast extends to safety refusal. These results distinguish the local reweighting effects captured by influence estimates from the broader intervention leverage of the examples they identify, motivating intervention-aware evaluation of TDA methods.
Read the full abstract

All authors Yuzhang Luo, Chenpeng Wang, Jianhui Chen, Liangming Pan