Peer Revenue Play credits. Real papers.

Papers

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

All of arXiv

Wednesday 9 September

Cosmology and Nongalactic Astrophysics

A universal connection between lens density profiles and low-frequency wave optics in gravitational-wave lensing

We investigate the low-frequency behavior of the amplification factor in gravitational lensing and explore how it encodes information about the density profile of the lensing object. We derive the low-frequency expansion of the amplification factor under the Born approximation for a broad class of projected density profiles. For spherically symmetric profiles that decay faster than any power law at large distances, we derive a systematic expansion of the amplification factor in powers of frequency, with the logarithmic dependence appearing only in the leading term, and show that each expansion coefficient is determined by a finite set of moments of the density profile. We then extend the analysis to profiles with power-law tails and demonstrate that such profiles induce additional non-analytic frequency dependences, including fractional powers and logarithmic terms, which directly reflect the asymptotic behavior of the density distribution. Furthermore, we investigate the effects of non-sphericity and show that contributions from the quadrupole moment appear only as higher-order corrections relative to the spherically symmetric component in the low-frequency regime. Finally, we investigate the validity of the Born approximation in the low-frequency expansion. We derive a criterion for the maximum order of the low-frequency expansion up to which the Born approximation remains dominant over the post-Born corrections.
Read the full abstract

Information Retrieval

Should I Be Polite to My LLM Relevance Judge? Tone as a Severity Operating-Point Shift

Large language models are increasingly used as relevance judges, yet their labels can shift with prompt surface form. We study one such feature – tone – on 3,498 TREC DL19/DL20 query-passage pairs, across eight judge models, five classifier-calibrated politeness levels, and three paraphrases per level. Effects are strongly model-dependent: one judge shows a structured U-shaped response, whereas most show only small changes. Where tone changes agreement, the results are more consistent with a shift in the judge's severity operating point – its overall scoring leniency – than with improved judgment. Agreement rises or falls as this shift moves the judge toward or away from human annotators' strictness. A query-disjoint cross-fit retains the expected association (Spearman $\rho = -0.683$; exact model-block permutation $p = 0.019$). Tone affects calibration-based agreement more than ranking outcomes: across 32 model-tone contrasts, the largest absolute mean change in NDCG@10 is 0.011, although Kendall's $\tau$ as low as 0.743 shows that reordering is reduced, not absent. The account reconciles prior contradictory findings and identifies prompt tone as a potential validity threat when absolute relevance labels matter.
Read the full abstract

Artificial Intelligence

Decision Shifts, Lost Label Functionality, and an Inconclusive Grounding Audit in Correctness-Gated Multi-Teacher Distillation

Candidate decision correctness and rationale grounding are different objectives. We examine correctness-gated multi-teacher distillation in a fixed experiment. Eight arms share 4,330 sources, a 63.9M-parameter student, 12,990 optimization rows, 406 updates, evidence inputs, and a decoder; seven teacher-based arms use one fixed three-response pool. Three seeds are evaluated on 267 held-out examples. Relative to unfiltered distillation, the correctness-weighted arm differed in accuracy by +0.1660 (95% observed-matrix interval [0.0670, 0.2455]), five-label macro-F1 by +0.1323 ([0.0916, 0.1731]), and task-defined conditional unsafe-action rate by -0.4979 ([-0.5926, -0.3686]). These shifts do not imply uniformly better behavior. Source-label SFT had the highest mean macro-F1 (0.586). The weighted arm had zero Refuted recall in every seed, and two seeds assigned NotEnoughInfo to all 167 claim examples. In an availability-amended audit at one reference seed, weighted and unfiltered outputs had 0/20 versus 1/20 evidence-supported positives and 20/20 versus 19/20 positives containing unsupported material. Samples were non-paired, source overlap was not serialized, and the amendment followed automatic summarization but preceded annotation. The audit therefore cannot estimate a common-source grounding effect and is inconclusive about system-level improvement or harm. Hard filtering already achieved 0.660 accuracy, 0.530 macro-F1, and 0.135 conditional unsafe rate. The implemented weighted arm showed no demonstrated incremental decision benefit over hard filtering. This fixed-matrix failure analysis shows decision redistribution with lost label functionality; the available human audit does not establish a grounding gain.
Read the full abstract

Data Structures and Algorithms

Fast Algorithms for Sparse PCA and Robust Sparse Estimation

We study fast algorithms for sparse-PCA certification. Given a positive semidefinite matrix $M$, the problem asks either to rule out a large $k$-sparse quadratic form or to return a high-value (relaxed) witness. The standard semidefinite relaxation provides such certificates, but existing general-purpose solvers require $\Omega(d^4)$ time. We give a bicriteria algorithm running in $O(d^2+d k^{O(\log k)})$ time: if some $k$-sparse unit vector has quadratic form greater than $2$, it returns either an $O(k^2)$-sparse unit vector or an SDP-feasible matrix of value at least $1$. For $k\leq\exp(O(\sqrt{\log d}))$, this running time is $O(d^2)$. We also go below the quadratic barrier in the sample-access model: Given $n=d^{o(1)}$ samples, our algorithm obtains a related one-sided certificate in $d^{2 - \Omega(1)}$ time for $k=\mathrm{polylog}(d)$, without forming the empirical covariance matrix. As an application, these certificate routines yield the first quadratic and subquadratic-time algorithms for robust sparse estimation for broad families of distributions. Our sparse-PCA algorithm reduces a high-value sparse direction to a bounded-radius set in the graph of large correlations and searches the resulting candidate supports. The subquadratic implementation constructs this graph using fast correlation detection.
Read the full abstract

Human-Computer Interaction

AppetiteCheck: Feasibility of Momentary Vagus Nerve Stimulation as an Implicit Intervention for Eating Behavior

Overeating and emotional eating are common health issues that affect people even without an eating disorder. The vagus nerve plays a critical role in the gut-brain axis, and implanted vagus nerve stimulators have been associated with reduced appetite. In this paper, we propose transcutaneous cervical vagus nerve stimulation (tcVNS) as a ubiquitous system to provide immediate, low-effort intervention during an eating episode. In a study with 24 participants, we evaluated a mobile, handheld tcVNS device during a single episode of distracted snacking. We found that participants ate 9.6% less and 23.6% more slowly during vagus nerve stimulation than during sham stimulation. Post-snacking satiety was the same in both conditions, while heart rate was lower during vagus nerve stimulation. The stimulation was described as subtle and barely noticeable. Overall, these results provide evidence for the feasibility of non-invasive vagus nerve stimulation as a low-attention intervention for managing eating behavior – one that can be packaged inside ubiquitous interactive systems. As such, we extend the design space of implicit interfaces toward physiological intervention, motivating future ubiquitous systems that pair eating-related sensing with low-attention interventions.
Read the full abstract

All authors Tan Gemicioglu, Jas Brooks, Pedro Lopes, Tanzeem Choudhury

Strongly Correlated Electrons

Anomalous Hall Response Induced by Correlated Disorder in the Breathing Kagome Lattice Mn$_{3}$Sn

Macroscopic transport tensors are generally constrained by the average crystallographic and magnetic symmetries of a material. In the kagome antiferromagnetic Weyl semimetals Mn$_{3+\delta}X$ ($X=$~Sn or Ge), previous studies showed that the anomalous Hall conductivity $\sigma_{yx}$ is forbidden by the average \hexsg{} structure and coplanar inverse-triangular magnetic order. Here we report that nearly stoichiometric Mn$_3$Sn nevertheless exhibits a finite $\sigma_{yx}$ with large hysteresis, together with enhanced $\sigma_{zx}$ and $\sigma_{yz}$, in the inverse-triangular phase below $T_{\mathrm{N1}}\approx 440~\mathrm{K}$, whereas all AHE components vanish in the amplitude-modulated conical phase below $T_{\mathrm{N2}}\approx 280~\mathrm{K}$. Total scattering and magnetic pair distribution function analysis reveal correlated orthorhombic distortions and noncoplanar Mn moments. First-principles calculations show that this coupled lattice-spin distortion activates the average symmetry forbidden $\sigma_{yx}$ within the inverse-triangular phase. Its disappearance below $T_{\mathrm{N2}}$ indicates that the correlated disorder must cooperate with a long-range inverse-triangular antiferromagnetic order capable of supporting Berry curvature. Our results establish correlated disorder as an active symmetry-breaking degree of freedom that enables topological transport inaccessible from the Bragg-average structure alone.
Read the full abstract

All authors Tsung-Han Yang, Seng Huat Lee, Hengxin Tan, Yuanpeng Zhang, Benjamin A. Frandsen, Václav Petříček, Huibo Cao, Daniel Olds, Matthew G. Tucker, Jiaqiang Yan, Binghai Yan, Zhiqiang Mao, Qiang Zhang

Machine Learning

Kernel-Complexity Edge Sanitization for Training-Free Defense against Structural Graph Attacks

Graph Neural Networks (GNNs) have achieved remarkable success across diverse applications, yet they remain highly vulnerable to adversarial attacks that maliciously perturb graph structure. Existing defenses often lack rigorous theoretical grounding, rely on attack-specific heuristics, or require costly retraining procedures such as adversarial training. To address these limitations, we propose Kernel-Complexity Edge Sanitization (KCES), a training-free and model-agnostic framework for defending against structural attacks. KCES is built upon Graph Kernel Complexity (GKC), a principled metric derived from the graph Gram matrix that appears in a generalization upper bound on the GNN test error. From this bound, we define an edge-specific KC score that quantifies each edge's structural influence via its induced change in GKC. KCES then identifies and prunes high-KC edges, which are empirically enriched with adversarial perturbations under structural attacks, to mitigate their harmful impact. Computationally efficient and scalable, KCES operates as a lightweight preprocessing step without retraining and can be seamlessly integrated with existing defenses. Extensive experiments demonstrate that KCES consistently outperforms representative robust baselines across diverse attack settings and scales effectively to large graphs. Supported by theoretical analysis and extensive empirical validation, KCES provides a principled and efficient framework for securing GNNs. Our code is available at https://github.com/karpning/KCScore.
Read the full abstract

All authors Yaning Jia, Shenyang Deng, Yaoqing Yang, Chiyu Ma, Wenxuan Xu, Soroush Vosoughi

Cryptography and Security

PrivAudit: A Dual-Lens Auditing Framework for Website Privacy Practices under the CCPA

Five years after the enforcement of the California Consumer Privacy Act (CCPA), understanding how website privacy practices evolve at scale in response to regulation remains a key challenge for both researchers and regulators. Prior work and regulatory efforts have focused on manual and case-specific enforcement, but there remain no scalable approaches to systematically audit two key user-facing facets of websites that are crucial signals for the CCPA: privacy disclosures and front-end user tracking behavior. In this paper, we present PrivAudit, an automated auditing framework that adopts a dual-lens approach to capture: (1) privacy disclosures through large language model-based analysis of privacy policies grounded in CCPA provisions, and (2) user-observable data collection behavior through automated browser measurements of cookie writes under diverse privacy configurations. We apply PrivAudit to 998 websites and report two broad findings. The law is associated with stronger privacy disclosures: CCPA-subject policies are more likely to disclose opt-out mechanisms, data-sharing practices, and user rights. On the other hand, cookie-based tracking remains pervasive, with both CCPA-subject and not-subject websites setting a total of 6,392 targeting cookies, 49% of which are third-party writes. Moreover, cookies show limited-to-moderate responsiveness to privacy signals and consent choices, even when websites claim to honor them in their disclosures. Our results highlight the need for multi-layered and scalable auditing approaches that combine policy analysis with behavioral evidence. PrivAudit can support these auditing workflows at scale by generating actionable signals and patterns for further manual review. We open-source PrivAudit and are engaging with regulators to support auditing in practice.
Read the full abstract

All authors Mohamed Moustafa Dawoud, Riya Aggarwal, Likith Rahul Krishnamurthy, Ram Sundara Raman

Computation and Language

When Auditors Fabricate: Batch-Size Degradation and Confident Hallucination in LLM Detection of Planted Document Contamination

Large language models are increasingly proposed as automated auditors of document quality, yet their reliability as detectors of planted errors is poorly characterised. We construct a contaminated corpus of 150 academic papers spanning supply chain management and medical research, injecting 450 known contaminants of three types: typographical corruption, semantic reversal, and absurd out-of-context insertion. We then evaluate Google Gemini 3.0 Pro's ability to recover a 180-contaminant answer-key subset across 60 documents under three prompting regimes of increasing scale: single document, small batch, and large batch. Detection is unreliable even at small scale and collapses entirely at large scale: 50% recovery on single documents and 60% on small batches, a difference this sample cannot resolve, against 2.8% on large batches. The failure mode at scale is not abstention but fabrication. Rather than reporting incomplete processing, the model produced confident findings including invented contaminants of its own, absurdities such as "telepathic squirrel" and "quantum-powered toaster" that mimic the style of the planted material but do not appear in any document. Detection also varies by contamination type: absurd insertions were recovered at 75% in completed evaluations, while semantic reversals and typographical corruptions were each recovered at only 50%. The corruptions most likely to occur in the wild, plausible ones, are the ones most often missed. We conclude that LLM document auditing degrades not gracefully but deceptively, and outline the harness such systems require: bounded batch sizes, direct content injection, and mechanical verification of every reported finding against source text.
Read the full abstract

All authors Karan Parekh, Sanjana Pendyala Ravinder, Sana Mhapsekar, Medina Maloku

General Relativity and Quantum Cosmology

Ovcharenko-Podolsky Einstein-Maxwell gravitational instantons

We present an explicit three-parameter family of toric Einstein-Maxwell gravitational instantons. These are complete Riemannian manifolds with vanishing scalar curvature which satisfy the Riemannian Einstein-Maxwell equations. We argue that the metric can be extended to a space with two asymptotic ends diffeomorphic to $\mathbb{H}^2 \times S^2$ with a non-collapsing two-cycle (bolt) in the bulk. The global metric is produced by suitably extending an analytic continuation of a local family of Lorentzian black hole geometries constructed by Ovcharenko-Podolsky.
Read the full abstract

Theoretical Economics

The Attention Cost of Stable Matching

In large markets, scarce attention limits partner evaluation and creates allocation loss, which stability magnifies. In an independent random market with average executable degree $d$, unmatched shares fall at rates $e^{-\sqrt d}$ under stability and $e^{-d}$ under maximum matching on the same graph. Changing consideration can make applications rejected in a provisional active-screen computation relevant again. Exact query-neutral implementation must retain allocation-relevant off-screen authorizations; otherwise, missing authorization must be reacquired. Limited-attention deferred acceptance (LA-DA) preserves valid authorizations and reengages eligible pairs. Conditional on exact next-best information and persistent execution rights, adaptive discovery saves a logarithmic factor in reached proposals relative to independent exposure. In an application to speed dating, bilateral reports let us compare stable and maximum matching on restricted graphs, separating missed opportunities from same-graph stability loss. In Chilean school choice, we document 9,502 applicants accepting higher-ranked or new placements through retained rankings.
Read the full abstract

Robotics

CT-SAFR: Safe and Interpretable Chain-of-Thought Reasoning for Autonomous Robots: A Multi-Layered Verification Framework for Trustworthy AI-Driven Robotic Decision Making

Chain-of-Thought (CoT) prompting enables LLMs to perform explicit, step-by-step reasoning, creating opportunities for sophisticated autonomous robots. However, recent research reveals that reasoning models verbalize their actual decision processes only 25-39% of the time, with faithfulness degrading 44% on complex tasks. This paper presents CT-SAFR (Chain-of-Thought Safety and Faithfulness for Robotics), a multi-layered verification framework achieving 94.2% hallucination detection (n = 500, 95% CI: 91.8-95.9%) with sub-500ms latency. Through a warehouse robot case study, this work demonstrates 87% reduction in unsafe reasoning outputs (p < 0.001) and provides recommendations for responsible deployment of reasoning-capable autonomous robots.
Read the full abstract

Computation and Language

Looped GPT-BERT: Trading Parameters for Computation in Small Language Modeling

When training data are limited, increasing parameter count is not the only way to improve language-model performance. A small parameter set, when repeatedly applied, can also deliver comparable performance. We study Looped GPT-BERT in the BabyLM 2026 Strict-small setting, combining GPT-BERT's masked next-token and causal language-modeling objectives with depth-wise parameter sharing. We train on a preprocessed 7.48M-word English corpus and compare objective ratios, non-looped and looped architectures, and loop counts. Our final $4\times12$ model uses four physical layers for twelve recurrent traversals and contains 12.18M parameters. The BabyLM 2026 leaderboard reports an Overall Average of 35.42 and an NLP Average of 48.48. Compared with public BabyLM 10M Strict-small GPT-2 and GPT-BERT baselines, it achieves comparable performance on selected linguistic and downstream metrics, including BLiMP and GLUE, with fewer parameters. The loop ablations show that additional recurrent computation can improve training and preserve strong performance on selected linguistic tasks, whereas poorer performance on other tasks may reveal an inherent limitation of the looped design: using only a few physical layers restricts the model's representational space.
Read the full abstract

All authors Tingshuo Fan, Hongtao Mu, Tianyu Zhou, Hansen Liu, Tao Ji

Quantum Physics

Cascade-induced high-performance nonreciprocal quantum batteries

Nonreciprocal quantum batteries harness reservoir engineering for controlled energy transfer, heralding a paradigm shift for quantum energy storage. A prior study utilizing the Metelmann-Clerk formalism reported a fourfold enhancement in nonreciprocal energy accumulation over reciprocal counterparts. This fourfold enhancement, however, appears to stem from a Lindblad master equation that exhibits fundamental inconsistencies, which may systematically understate the actual nonreciprocal advantage. Our approach is rooted in the rigorously established cascaded open quantum systems formalism, which ensures both mathematical consistency and physical fidelity. Remarkably, we surpass the previously reported fourfold benchmark, attaining a regime-independent sixteenfold steady-state nonreciprocal energy advantage over reciprocal systems. We further uncover a dissipation-dependent battery-to-charger efficiency, fourfold under symmetric damping, surpassing this benchmark when the battery is less dissipative, and lower otherwise. This work establishes the cascaded formalism as a mathematically rigorous and experimentally viable foundation for high-performance quantum energy storage.
Read the full abstract

Probability

A Self-Contained Proof of the Portmanteau Theorem

The Portmanteau Theorem gives several equivalent characterizations of weak convergence of probability measures (equivalently, convergence in distribution of random variables). This note presents a unified proof of the theorem in a seven-statement formulation, emphasizing the structure of the implication cycle connecting expectations of bounded continuous functions with probability bounds on open and closed sets. The argument highlights standard approximation techniques and clarifies the role of intermediate function classes such as bounded Lipschitz functions.
Read the full abstract

Combinatorics

Borsuk-Ulam type theorem for the orthogonal group and orthogonal four-partitions

Makeev [2] stated that every finite Borel measure in $\mathbb R^d$ assigning zero mass to hyperplanes can be cut by $d$ mutually orthogonal hyperplanes so that every pair divides the measure into four equal parts, and outlined a proof strategy, but the key steps were left incomplete. We give a direct and elementary proof. The main ingredient is a Borsuk–Ulam theorem for $O(k)$: every $B_k$-equivariant map from $O(k)$ to a natural representation of the hyperoctahedral group $B_k$ has a zero. An explicit model map has one free orbit of zeros, consisting of the signed eigenbases of a diagonal operator with simple spectrum. A derivative computation and mod-$2$ equivariant degree complete the proof. This is a self-contained proof of a special case of the general Stiefel-manifold theorem in [5].
Read the full abstract

Social and Information Networks

Chance, Persistent Advantage, and the Generative-AI Era in Open-Source Package Careers

Studies of careers in science, film, music, and books report a common pattern. When a person's most successful work arrives is close to a random draw over the works they produce. How large their successes tend to be, in contrast, follows a stable, person-specific factor. We test whether this pattern holds for open-source software careers and whether it changed when generative AI coding tools arrived. From the complete public record of GitHub push events (2015-2025), we reconstruct 102.2M career works by 6.15M contributors, and for the 908k contributors whose repositories publish packages, we measure each work's impact by how many downstream packages come to depend on it. First, we find that the timing of a career's biggest hit is close to a lottery over their works, as in science and the arts, with a small, replicable lean toward early career that grows as careers get longer. Second, some coders reliably produce higher-impact work than others, but this lasting personal factor accounts for only part of why impact persists (about a fifth in our primary specification); the rest behaves like momentum, success feeding on itself for a period of time. Third, within the same contributors, this structure did not change after ChatGPT's release. The stable factor's weight grew by about as much as it grew for an earlier cohort that simply aged, and subtracting the effect of aging from the effect of generative AI puts the shift at +0.03 (95% CI [-0.22, +0.23]), indistinguishable from zero. The success pattern documented in science and the arts therefore describes open-source careers too, and it shows no detectable break across the arrival of generative AI. These results have implications for how track records on open platforms should be read and on what to expect from generative AI for the careers built on them.
Read the full abstract

Optimization and Control

Counterexamples for BFGS-type methods under arbitrary strong Wolfe constants

Whether the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method and its variants can fail to converge on smooth nonconvex functions under realistic line-search parameters has remained a central open problem in quasi-Newton theory since the landmark counterexample of Dai (2002), which was confined to small Armijo parameters ($c_1 \le 1/84 \approx 0.0119$ on Powell's geometry, or $c_1 \le 69/7480 \approx 0.0092$ on his six-point cycle) and an objective function unbounded below. In this paper, we resolve the long-standing open question, posed by Dai (2002) following a discussion with J. C. Gilbert, of whether such counterexamples exist in theory for every Armijo parameter $c_1 \in (0, 1)$. Specifically, for every prescribed pair of line-search parameters $0 < c_1 < c_2 < 1$, we construct an objective function $f \in C^\infty(\mathbb{R}^2)$, bounded below and with Lipschitz continuous gradient, on which every method in a broad conjugacy class $\mathcal{C}$ equipped with the first-local-minimizer line search generates an infinite sequence of iterates with $\|\nabla f(x_k)\| = 1$ for all $k \ge 0$. The class $\mathcal{C}$ encompasses the classical full-memory BFGS method, limited-memory BFGS (L-BFGS) with arbitrary memory $m \ge 1$, the Broyden positive family, and the Hestenes-Stiefel conjugate gradient method. The steps are the standard first local minimizers along the search rays and simultaneously satisfy the strong Wolfe, weak Wolfe, Armijo, and Goldstein conditions with constants $(c_1, c_2)$. The construction operates in the minimal possible dimension $n = 2$, exploiting a non-decaying conjugate descent orbit in the plane coupled with an explicit tubular interpolation whose two-bump axial curvature profile places the Armijo ratio anywhere in $(0, 1)$.
Read the full abstract

Machine Learning

ALIGN-HOLD: Experience Alignment for Real-Time Hold Control in Large-Scale Ride-Hailing Matching at DiDi

Real-time hold control is a high-leverage mechanism in large-scale ride-hailing systems: by selectively deferring driver-order pairs, the platform can wait for better matching opportunities and improve end-to-end passenger-driver experience. Existing production systems such as EXHOLD learn bandit-based hold policies from handcrafted combinations of trip completion, cancellations, waiting time, and driver effort. However, designing such rewards becomes increasingly difficult as marketplace preferences are heterogeneous and observed passenger-driver behavior can be sparse, noisy, and affected by dynamic supply-demand conditions. We present ALIGN-HOLD, a production-scale experience alignment framework that learns hold policy from implicit marketplace preferences. ALIGN-HOLD constructs complementary preference pairs from order trajectories, driver trajectories, and contemporaneous local matching graphs, and trains an experience Reward Model (RM) using balanced multi-view sampling and model-adaptive hard preference sampling. During simulator-based policy learning, the frozen RM provides a dense, context-dependent reward and supports label-free filtering of low-identifiability interactions whose behavioral feedback is difficult to attribute to matching quality. We deploy ALIGN-HOLD on DiDi's ride-hailing platform and evaluate it in a 28-day randomized A/B experiment, covering approximately 100,000 passenger requests per day. Compared with the deployed production policy, ALIGN-HOLD achieves statistically significant improvements in trip completion rate and driver income, while significantly reducing passenger cancellations before and after driver acceptance. Complementary ablations, RM diagnostics, and behavioral analyses validate the contributions of the proposed components. ALIGN-HOLD has been fully ramped up and is currently serving DiDi's Brazil marketplace.
Read the full abstract

All authors Zuhao Zhang, Xu Liu, Kai Wan, Zihao Lu, Li Ma, Shuai Li

Computation and Language

An Empirical Study of Development-Tuned Representation Sensitivity for Medical QA Data Selection

We study medical question-answering (QA) data selection when a developer already has questions, labels, and auxiliary explanatory texts but can retain only $B$ examples for a final LoRA update. Representation-Sensitivity Prioritization (RSP) first restricts the candidate bank to its top-$4B$ predictive-entropy examples, then combines the frequency of answer flips with non-flipping contraction of the top-two answer margin under hidden-state perturbations. We evaluate this operational score on five medical QA datasets with google/medgemma-4b-it, $B=128$, and three seed branches. After removing four MedExpQA test questions that also occur in its acquisition pool, dataset-specific development-tuned RSP attains the highest observed five-dataset mean accuracy (53.44%) and Macro-F1 (48.21%) among the evaluated selectors. Relative to random selection within the same entropy-restricted pool, the descriptive differences are +1.49 and +2.10 percentage points, respectively; paired 95% bootstrap intervals include zero. The audit mixes benchmarks with different relationships to documented backbone post-training sources, and the same test cohorts had been used in earlier analyses before the expanded tuning grid was designed. Results are therefore exploratory, modest, and dataset-dependent rather than evidence of statistical superiority. RSP point estimates also exceed an archived no-update reference, but unpinned model revisions and incomplete historical input identifiers limit attribution of that difference to fine-tuning. Because every selected training example includes auxiliary text and no matched answer-only control is evaluated, the experiments also do not identify a causal benefit from auxiliary-text conditioning itself.
Read the full abstract

Optimization and Control

Primal-Dual Inexact Newton-MR for Nonconvex Optimization with Equality Constraints

Optimization problems with nonlinear equality constraints arise throughout science, engineering, and increasingly in machine learning. Prominent methods for solving such problems include sequential quadratic programming and, more broadly, primal-dual Newton methods. Classical analyses of these methods typically rely on strong assumptions, perhaps most notably positive definiteness of the Lagrangian Hessian on the null space of the constraint Jacobian. In practice, this assumption often necessitates strong regularization or the use of a positive definite Hessian surrogate. Moreover, in large-scale settings, solving the primal-dual Newton subproblem exactly is often computationally infeasible. To address these issues, we propose an inexact primal-dual Newton method with an inner solver based on the conjugate residual (CR) method. Exploiting recently established properties of CR, including negative-curvature detection, iterate monotonicity, and descent guarantees, our method handles indefiniteness in the subproblem directly as it arises. Our method thereby avoids detrimental regularization of the Lagrangian Hessian while naturally accommodating inexact solves. We establish worst-case global convergence guarantees and demonstrate strong empirical performance on large-scale, nonconvex problems.
Read the full abstract

Machine Learning

Settling: Equilibrium Inference for Non-Convex Validity Sets

Many learning systems return a single point estimate even when admissible outputs form disconnected or non-convex sets. Under squared loss, an ambiguous conditional distribution can therefore have a Bayes-optimal conditional mean that is invalid. We formalize this failure as conditional mean collapse and introduce Settling, an equilibrium-based inference operator that separates proposal generation, consistency evaluation, and test-time equilibrium selection. The operator treats a mean-seeking proposal as an initialization and refines it toward a locally stable configuration; conditional on initialization, refinement is deterministic. We establish exact-gradient descent, local convergence, and an inexact-gradient robustness condition relevant to learned consistency critics. In a reproducible 100-context geometric diagnostic, the mean-seeking baseline succeeds in 0/100 contexts, stochastic denoising in 100/100, and Settling in 99/100 while producing substantially lower trajectory roughness. A 1,200-run sensitivity study yields 97-100% success across obstacle-jitter ranges up to 0.20 and 94-100% across one-time initialization perturbations from 0.05 to 0.50. Cross-domain panels remain mechanism illustrations; learned high-dimensional validation remains an open empirical test.
Read the full abstract

Sound

Beyond Accuracy: ARIA-Rubrics for Evaluating Audio Reasoning in Large Audio Language Models

Large Audio Language Models (LALMs) have shown strong performance on audio reasoning benchmarks, but accuracy alone cannot distinguish true reasoning from superficial pattern matching, often overestimating reasoning ability since high scores may result from guessing rather than genuine audio understanding. Evaluating the reasoning process itself is essential for improving LALMs' reasoning ability, yet remains challenging. Existing methods either rely on costly human annotation or opaque LLM-as-judge approaches, making them impractical, biased, and lacking transparency. Moreover, audio reasoning introduces unique challenges absent in text-based settings, perceptual hallucination and cross-modal alignment between audio understanding and textual inference, hence text-based evaluation frameworks cannot be directly applied. Therefore, we propose ARIA-Rubrics (Audio Reasoning Integrity Assessment), a lightweight, annotation-free gold reasoning chains, automatic and transparent framework comprising six complementary metrics that evaluate audio reasoning quality across perceptual grounding, reasoning coherence, and answer consistency. We use Chain-of-Thought prompting as an externalization mechanism to make the reasoning process observable. Experiments on 9 models across 2 benchmarks identify three reasoning modes of current LALMs with actionable directions for future development, with ARIA-Rubrics achieving high correlation with human judgments. The code is available at the Github Repository.
Read the full abstract

All authors Yupei Li, Qiyang Sun, Mohamed Mady, Chenxi Wang, Zhengwei Gong, Berrak Sisman, Björn Schller

Mathematical Physics

Wasserstein Stability, Couplings Across Volumes, and the $1+1/d$ Moment Thresholds in the Edwards–Anderson Model

We study the quenched pressure of the nearest-neighbor Edwards–Anderson Ising model with free and periodic boundary conditions. First, we prove that the infinite-volume pressure is $\beta d$-Lipschitz in the coupling law for the $1$-Wasserstein distance, yielding quantitative thermodynamic limits for spatially inhomogeneous disorder. Second, for periodic volumes, we prove that almost-sure convergence under every joint coupling of the finite-volume disorder arrays is equivalent to complete convergence of the one-volume pressure laws. Third, we prove that $\mathbb{E}\vert{}J\vert{}^{1+1/d}<\infty$ guarantees this universal-coupling conclusion. We further show that this exponent is optimal among uniform power-moment assumptions: for every $1\leq q<1+1/d$, there is a centered symmetric law with finite $q$-th moment for which canonical nested volumes converge almost surely, whereas independently resampled volumes with the same fixed-volume marginals converge in probability but not almost surely. Finally, in dimension one, we prove that the first-moment condition is also necessary for a finite limiting pressure.
Read the full abstract

Materials Science

Automated Generation of Commensurate Magnetic Structures based on Spin Space Groups and Graph Theory

Magnetic structures with symmetry constraints are candidates for energetically favorable configurations. Enumerating these structures is essential for identifying experimental observations and provides unbiased, linearly stable, and optimally sampled reference configurations for energy fitting when extracting spin interactions. We present SpinGraph, an automated workflow generating symmetry-distinct magnetic configurations. SpinGraph is not only compatible with the more general spin space groups, but also allows precise control of the prescribed single- or multi-Q states superposition. For finite groups, we enumerate compatible subgroups directly. For spin space groups with a continuous or special spin-only part, the finite component is enumerated first and then combined with the compatible exact spin-only constraints. These actions are translated into graph constraints in real or Fourier space. The symmetry of every generated structure is re-evaluated after construction. Finally, we integrate SpinGraph with the magnetic analysis code AMATIS to perform a thorough calculation for the spin interactions of the insulating monolayer CrI3. SpinGraph complements the last part of AMATIS and realizes the fully automated workflow for the spin Hamiltonian construction, which is essential for studying phase transitions, spin textures, magnons, and spin dynamics.
Read the full abstract

All authors Yifan Wang, Boyang Deng, John Robertson, Weisheng Zhao, Stefan Blügel, Haichang Lu

Artificial Intelligence

Safe to Stop? Risk-Constrained Stopping for Sequential Clinical Diagnosis Agents

Clinical diagnosis agents must decide not only what test to request next, but also when to diagnose or defer. Existing agent benchmarks largely evaluate accuracy after fixed or unconstrained interaction, leaving autonomous stopping reliability implicit. We present Cros, a risk-constrained stopping layer combining state-wise error ranking, policy design on disjoint development splits, and LTT-style exact tests of selective diagnostic error and minimum autonomous coverage for complete sequential policies. Its finite-sample guarantee requires the candidate family, testing rule, and any randomization to be frozen before calibration labels are accessed. On a 1,834-episode MIMIC-derived abdominal-pain benchmark, the full ranker achieves exploratory state-error AUROC 0.853, compared with 0.715 for maximum class probability and 0.552 for the backbone's native stop score. On the previously viewed 367-episode evaluation split, analytically averaging over the frozen Cros weights yields 16.9% selective error at 78.8% coverage, cost 5.57, and 0.68 tests, versus 30.8% error at 100% coverage, cost 8.14, and 1.53 tests under native stopping. Forced continuation is non-monotone: error is 28.3% with HPI alone and 34.3% after full workup. However, the uniform-weight mixture ablation is cheaper on this viewed split despite missing the locked development margins, and Cros nominally satisfies the joint criterion in only 6 of 20 development resplits. Because evaluation labels were inspected during earlier development, these findings provide exploratory feasibility and audit evidence, not a confirmatory safety certificate.
Read the full abstract

Computation and Language

X2-NativeCursor: Native-Token Text Progress Tracking for Incremental-Text Streaming Codec TTS

Incremental-text streaming text-to-speech (TTS) needs online text progress tracking for synchronized highlighting, interruption handling, and dialogue-history updates. Input text arrives before it is spoken, so text arrival alone cannot indicate speech progress. Existing waveform-based alignment requires complete audio or adds acoustic processing during streaming. We propose X2-NativeCursor, a lightweight observer that tracks progress from native speech tokens before waveform decoding without changing the TTS generator. Its normalization plan links spoken labels to their original-text spans. Text and native-token encoders feed a local matcher that estimates the current label position. A separate output rule converts revisable position estimates into a cursor that never moves backward. Mean absolute error against an automatic reference is 0.151 Chinese characters with 80-ms lookahead, versus 1.253 characters with 320-ms lookahead for an online waveform baseline. Alignment real-time factor also decreases from 0.3598 to 0.0180 relative to this baseline. Lower tracking error is retained under a second automatic alignment reference. We evaluate X2-NativeCursor on Qwen3-TTS and validate its adaptation to CosyVoice2 by training a separate observer for each backbone. Code is publicly available at https://github.com/X-Square-Robot/X2Streaming-TTS.
Read the full abstract

All authors Zehan Liu, Carl Chen, Rime Wen, Kaiqi Fu, Altman Lin, Shawn Qin, Lights Shi, Roy Gan, Hao Wang, Qian Wang

Machine Learning

Muon-C: Operator-Aligned Muon for Convolutional Kernels

Muon replaces matrix momentum with an approximately orthogonal polar direction, but its geometry depends on the matrix representation. For convolution, standard unfolding describes a local patch map rather than the convolution operator. We introduce Muon-C, an operator-aligned optimizer that represents kernel momentum as frequency-wise channel-transfer matrices, polarizes these blocks independently, and uses a critical Fourier grid to return updates exactly to the original finite kernel support. We show that the new geometry arises from combining the block partition and Fourier coordinates. The exact-polar direction is a linear minimization oracle under the critically sampled convolution norm. Its worst-case guarantee relative to the continuous convolution-operator norm is never weaker than unfolding and is strictly stronger for $3\times3$ kernels. On CIFAR-10 flow matching with matched applied-update RMS, Muon-C reaches 9.87 FID at 40k iterations, compared with 22.26 for unfolded Muon and 51.31 for Adam. It reaches their final quality using $0.62\times$ and $0.64\times$ their model FLOPs, respectively. Under equal tuning budgets, Muon-C achieves 3.42 FID. Gains persist across data scales and transfer to classification across convolutional architectures.
Read the full abstract

Algebraic Geometry

Determinantal Kernel Schemes of Matrix Nets and Applications to Positive Maps

For a linear map on a matrix space, we study the projective schemes obtained by intersecting its projectivized kernel with determinantal rank loci. For matrix nets, that is, three dimensional matrix spaces, we classify every positive dimensional intersection with the rank one Segre variety in arbitrary rectangular size. The possibilities are a ruling plane, a smooth conic, two Segre lines from opposite rulings, a reduced Segre line, or a Segre line with one reduced or embedded residual point. The smooth conic case is contained in a $2\times2$ compression. For $a,b\ge3$, the corresponding locus in $\mbox{Gr}(3,M_{a,b})$ has exactly three irreducible components, whose geometry and intersections are determined explicitly. For nets in $M_3(\mathbb C)$, every finite rank one scheme has length at most three; the adjugate identity gives an intrinsic determinantal obstruction to the length four case allowed for general systems of plane quadrics. As an application, a four parameter family arising from the merging construction admits exact positivity and decomposability criteria. After normalization, positivity is the unit square and decomposability is the quarter disk. The remaining region is atomic and carries explicit PPT entangled states of birank $(5,5)$ and Schmidt number two. The point of this application is that the exact phase boundary is realized on a smooth conic determinantal kernel stratum selected independently by the projective classification.
Read the full abstract

All authors Trung Hoa Dinh, Minh Toan Ho, Cong Trinh Le, Trung Dung Vuong