Cryptography and Security
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Saturday 29 August
Large Language Model (LLM) watermarks provide a mechanism for text provenance, enabling model owners to identify machine-generated content and attribute it to a specific watermarked model. However, current LLM watermarking approaches predominantly rely on inference-time sampler methods and focus their analysis on dense models. Inference-time methods are only effective when the text is explicitly generated via the model owner's controlled API; they fail in a post-compromise scenario. An adversary who steals or leaks the model weights gains complete control over inference and can simply run an unmodified sampler, bypassing the watermark and preventing post-theft attribution. In this work, we introduce Watermarking of Experts (WoE), a novel black-box text provenance method that leverages the unique structural properties of sparse Mixture-of-Experts (MoE) models. WoE biases the vocabulary of specific experts and shifts the watermark signal embedding away from unenforceable inference wrappers. This approach ensures the watermark remains intrinsic to the model parameters, enabling defenders to attribute text generated by stolen weights, leaked checkpoints, and secondary dense models distilled from the stolen architecture without needing access to the adversary's deployment or weights. We evaluate WoE across eight MoE models, demonstrating successful watermark detection from suspect text, achieving an average true positive rate of 90.1% at a 1% false positive rate, reaching up to 94.9%, while largely preserving general model utility. Furthermore, WoE remains detectable under adversarial supervised fine-tuning, model extraction, and output-level paraphrasing, forcing malicious actors into a trade-off in which weakening the attribution signal requires additional model adaptation or text-rewriting operations, or compromises the utility of the resulting output.
Analysis of PDEs
Exact Rayleigh Reduction for Direct Detection of Turning Bifurcations
We introduce an exact Rayleigh reduction for the direct construction and detection of turning bifurcations in nonlinear parameter-dependent equations. A generalized Rayleigh functional ordinarily provides only a scalar necessary constraint on solutions. We show that, along a suitable one-parameter transformation orbit $u=\mathcal G_s\phi$, it can instead become the exact parameter law $\lambda=\mathcal R(\mathcal G_s\phi)$ of a genuine solution branch. Turning points are then found from nondegenerate critical points of this one-dimensional Rayleigh profile. The branch tangent automatically yields a kernel direction of the state linearization, and under the usual Fredholm, kernel-simplicity, and parameter-transversality assumptions the detected point is a simple fold. For Kirchhoff equations, amplitude scaling on bounded domains and spatial dilation on $\mathbb R^N$ yield explicit exact branches, global parameter thresholds, and countable hierarchies of turning values. For the generalized Kirchhoff law $M_b(A)=a+bA^\theta$, the branch geometry is governed by the dimension–homogeneity index $\theta(N-2)-2$: an interior turning bifurcation occurs exactly when $\theta(N-2)>2$. This gives a universal bifurcation trichotomy that is independent of the particular Berestycki–Lions nonlinearity.
Read the full abstract
Dynamical Systems
Random attractors and almost-sure stability under discretization of a stochastic autoparametric system
For a stochastic autoparametric block-and-pendulum system, the long-time dynamics exhibit two fundamental features: the almost-sure stability of the single mode solution, characterized by its Lyapunov exponent, and the global asymptotic dynamics when this single mode solution loses stability. This naturally raises the question of whether these dynamical features are preserved under discretization, since such preservation is essential for the resulting discrete system to faithfully capture the qualitative behavior of the continuous system. To address this question, we first establish the existence of a random attractor for the continuous system subject to multiplicative stochastic excitation, providing a rigorous characterization of the global asymptotic dynamics. We then propose a numerical discretization that induces a discrete random dynamical system and prove the convergence of its random attractor to the continuous one as the step size tends to zero. In addition, we show that the numerical Lyapunov exponent of the single mode solution has the same sign as its continuous counterpart for sufficiently small step sizes, thus preserving the corresponding almost-sure stability or instability classification. These results demonstrate that the proposed discretization captures both the global asymptotic dynamics and the stability characteristics of the underlying stochastic autoparametric system.
Read the full abstract
Differential Geometry
Ambient and Poincaré metrics for CR $3$-manifolds associated with the self-dual Einstein ACH metric
We construct new ambient and Poincaré metrics for the Fefferman conformal manifold over $3$-dimensional CR manifolds, starting from the associated self-dual Einstein ACH metric. Unlike the classical ambient and Poincaré metrics arising from approximate solutions to the complex Monge-Ampère equation, these metrics satisfy Einstein-Maxwell-type equations to infinite order rather than the Ricci-flat or Einstein equations. Since these metrics are determined to infinite order without ambiguity, they enable us to construct CR invariant differential operators and local CR invariants involving arbitrarily high order derivatives of the Tanaka-Webster curvature and torsion. As an application, we obtain the ambient metric construction of the CR GJMS (Gover-Graham) operators of all orders in dimension $3$.
Read the full abstract
Geometric Topology
The crossing matrix and the Burau matrix of pure braids
Robotics
Systematic Lightweight Method for Robotics Based on Strain Energy Distribution Optimization
Service robots work with people and are highly expected to be lightweight for safety, agility and energy conservation. As a complex mechanical system, a robot consists of a large number of components and has various working configurations and load environments. An effective method for achieving system-level optimal robot design is a crucial requirement, but it poses significant challenges. In this study, we introduce a novel approach to optimize the distribution of strain energy, which can significantly improve the effectiveness of systematic optimization in a complex system. First, we present and demonstrate that the strain energy per unit mass should be uniformly distributed in an optimal lightweight mechanical system. Based on this criterion, the system-level problem can be decoupled and the design objective of each part can be assigned based on the strain energy. Then, each part can be optimally designed separately according to its specific circumstances by using different approaches, such as size optimization, topological optimization, and material optimization. In this way, the optimization is at the system level, while the computational complexity is at the part level. Weight reduction and improvements in mechanical properties can be obtained simultaneously. As an example, this method is applied to an arbitrarily designed robotic arm, and its effectiveness is further demonstrated in the cases of lightweight with stiffness improvement, considering multiple materials, multiple working conditions, and vibration performance.
Read the full abstract
Computer Vision
STARLINC: Satellite Trail Artifact Removal using Inter-Frame Correlation
The rapid expansion of low Earth orbit satellites such as Starlink is increasingly contaminating astronomical surveys. In practice, contaminated images are often identified through inspection. However, modern surveys generate terabytes of data each night, making manual screening infeasible and necessitating reliable automated methods for satellite trail removal. Unfortunately, existing general-domain line detection methods fail to generalize to astronomical images due to domain mismatch, which are mostly grayscale with sparse bright stars and have a low signal-to-noise ratio. Moreover, training new models from scratch is impractical due to the lack of large-scale annotated astronomical datasets. To address these challenges, we introduce STARLINC, the first ML-based framework for satellite trail removal without requiring tedious pixel-level annotation of astronomical images. STARLINC combines synthetic satellite trail generation for training, inter-frame differential maps from temporally adjacent exposures to highlight transient trails, and heatmaps to provide additional localization cues for pixel-level segmentation. Extensive experiments on real-world data demonstrate substantial improvements over baselines, establishing STARLINC as a scalable solution for next-generation astronomical surveys. Code is available at https://github.com/starioKim/STARLINC.
Read the full abstract
Computation and Language
Quantifying Error Tolerance in Synthetic Data: An Atomic-level Operand vs. Operator Perturbation Study
Synthetic data generation has become a cornerstone for advancing large language models. However, the lack of the quantitative analysis for error tolerance became a critical bottleneck. Consequently, current filtering strategies fluctuate between two extremes: they are either overly aggressive, risking the exclusion of potentially valuable samples, or overly permissive, failing to eliminate erroneous samples effectively. To bridge this gap, this paper introduces Atomic Tree Operation Modeling (ATOM), a framework that decomposes data into functional units ($f(x)\rightarrow y$). ATOM distinguishes benign Operand $x$ perturbations from fatal Operator $f$ perturbations. The former are needlessly discarded by aggressive filtering, while the latter slip through permissive filtering. Our experiments reveal a double dissociation: models are robust to operand perturbations but collapse under operator perturbations. By prioritizing operator over aggressive operand precision, our ATOM-synthesized data outperforms rigorous baselines (e.g., +3.1% gain over LIMA), suggesting that operator diversity matters more than operand precision. Our code is available at https://github.com/Lut-hub/ATOM.
Read the full abstract
Probability
Least Variability in a Polynomial-Square Class of Rational Kernels
We solve an extremal problem for positive randomization kernels that concentrate a random time as tightly as possible around a deterministic target, within the class obtained by exponentially damping the square of a polynomial. The construction, borrowed from the concentrated matrix-exponential literature, automatically guarantees nonnegativity, gives the kernel a Laplace transform with a single repeated real pole, and contains the classical Erlang randomizer as the special case in which the polynomial is a pure power. The half-polynomial need not be real-rooted and so may have nonreal conjugate zeros, yet every optimizer is proved to be real-rooted. After normalizing the kernel to unit mean, the minimum variance at polynomial degree $m$ turns out to equal the smallest relative gap between adjacent zeros of the Laguerre polynomial $L_{m+2}$, with every optimizer obtained by deleting a pair attaining this minimum. This characterization yields the minimal variance, the normalized density, and every optimizer in closed form. The optimizer can be computed from the spectrum of a symmetric tridiagonal Jacobi matrix. In the large-order limit, the minimal variance decays quadratically in the order of the kernel, a marked improvement over the linear decay rate of the Erlang benchmark, and the deleted pair of zeros localizes at normalized location $2$, with limiting absolute separation $2\pi$.
Read the full abstract
Strongly Correlated Electrons
Evolving charge order in the CDW state of AV$_3$Sb$_5$ metals
AV$_3$Sb$_5$ kagome metals are characterized by intertwined electronic and structural orders, which motivated extensive studies in recent years. Yet the details of the electronic state preceding the superconducting phase remain poorly understood. Here we extend our previous investigation [Phys. Rev. Research 7, L032046 (2025)] of RbV$_3$Sb$_5$ using avoided level crossing (ALC) muon-spin spectroscopy to the $A$ = Cs and K systems. Consistent with our previous study, we identify a second transition whose origin cannot be attributed solely to an internal magnetic field, indicating the involvement of an additional electronic mechanism that subtly modifies the charge distribution within the V plane. In particular, the ALC results point towards an additional charge modulation taking place within the charge density wave (CDW) phase and occurring at $T^{*}<T_{CDW}$ for $A$ = Cs and Rb, or in the vicinity of $T_{CDW}$ for $A$ = K.
Read the full abstract
Materials Science
Directional commensurability stabilizes structural superlubricity in patterned mesoscale interfaces
Structural superlubricity, arising from lattice incommensurability, offers a promising route to eliminate friction and associated energy losses in mechanical systems. In real-world systems, roughness and wear currently pose severe limitations on its robustness and especially the contact size. Here, we consider patterned surfaces as a possible route to overcome some of these limitations. We show that the simplest choice of patterning, contacts made up of two incommensurate triangular-triangular patterns, fails at elevated loads because of the small number of load-bearing contacts, causing the maximum local contact pressure to exceed the strength of the superlubric coating. We introduce a square-triangular patterned interface that increases the number of load-bearing contacts and organizes them into continuous contact lines. When sliding along specific directions relative to these lines, superlubricity is maintained at significantly higher loads by reducing pressure-induced coating failure while also remaining somewhat tolerant to surface imperfections. These findings establish a mechanism for stabilizing structural superlubricity against coating failure and a design principle for engineering low-friction interfaces with enhanced load-bearing capacity and defect tolerance.
Read the full abstract
Earth and Planetary Astrophysics
Dynamical Mass Demography of Single S-type Planets Shaped by Stellar Companions
More than 700 S-type planets have been identified from Gaia and TESS observations, including over 150 transiting targets, enabling the precise dynamical orbits and planetary mass characterization in binary environments. Here we present the first systematic N-body dynamical mass determination for 40 single S-type (Satellite-type orbit) planets in close binaries using an improved Markov Chain Monte Carlo (MCMC) framework with Gaussian priors to model stellar companion perturbations. {For binary systems with $a_\mathrm{B}<100$ au, companion-induced perturbations can bias the RV semi-amplitude and consequently the inferred planetary masses. The impact is strongest for short-period sub-Jovian planets, whose masses decrease by up to 20% when transitioning from Keplerian to N-body modeling, while massive Jovian planets with $M_{\mathrm{Jup}} < M_{\mathrm{p}} < 5$ $M_{\mathrm{Jup}}$ show only a 7.2% reduction.} Applying the NAFF stability criterion to TOI-4633, $i_{\mathrm{mut}} = 70^{\circ}$ is the upper limit for the fast rejection of unstable orbital solutions. Our results reveal that intermediate-mass planets exhibit both positive or negative biases in the inferred masses and influence the observed population of the sub-Jovian desert.
Read the full abstract
Artificial Intelligence
More Perspectives, Stronger Signals: Multi-Perspective Enhancement and Progressive Fusion for Multimodal Entity Representation Learning
Learning effective multimodal entity representations is fundamental for reasoning tasks such as multimodal knowledge graph completion (MMKGC). However, existing methods often suffer from semantic over-smoothing within modalities and ineffective noise filtration across modalities, particularly under sparse or ambiguous conditions. To overcome these limitations, we propose PrismF, a unified framework that synergizes multi-perspective enhancement with progressive fusion to extract stronger signals from diverse inputs. PrismF enhances fine-grained intra-modal semantics through a multi-perspective mechanism that decomposes each modality into complementary views and constrains them with a decoupling loss to reduce representation collapse. Furthermore, it improves cross-modal integration through a progressive fusion strategy that dynamically calibrates inter-modal interactions, enabling the model to emphasize informative signals while suppressing noisy or unreliable ones. Extensive experiments on three public benchmarks show that PrismF achieves the strongest overall performance, including relative improvements of 4.04% in MRR and 11.17% in Hits@1 on KVC16K. Our code can be found at https://github.com/HubuKG/PrismF.
Read the full abstract
Analysis of PDEs
uniqueness and asymptotic behavior of the stationary Navier-Stokes equations in a slab
In this paper, we investigate the uniqueness and asymptotic behavior of the stationary Navier-Stokes equations in a slab domain with no-slip boundary conditions. Specifically, under the given external force and the small Poiseuille flow assumptions, we prove that there is an H1 solution, and obtain the pointwise decay around the Poiseuille flow at far field. Furthermore, if the force is small, the solution is shown to be unique. The key point of the proof is the estimate of the Dirichlet integral of the solutions in the truncated domains and the Stokes regularity estimates.
Read the full abstract
Computer Vision
Chat-Edit-3D++: Interactive 3D and 4D Scene Editing via Large Language Models
Recent work on image content manipulation based on vision-language pre-training models has been effectively extended to text-driven 3D scene editing. However, existing schemes for 3D scene editing still have certain shortcomings, hindering their further development as interactive design tools. Such schemes typically adhere to fixed input patterns, limiting flexibility in text input. Furthermore, their editing capabilities are constrained by a single or a few 2D visual models and require intricate pipeline design to integrate these models into 3D reconstruction processes. To address the aforementioned issues, we propose the Hash-Atlas network, which reformulates 3D scene editing as operations on 2D atlas images, thereby achieving a workflow decoupling of the 2D editing and 3D reconstruction processes. Building on this foundation, we introduce a dialogue-based 3D scene editing approach, termed CE3D++, which is centered on a large language model (LLM) that allows arbitrary textual input from users and interprets their intentions, subsequently facilitating the autonomous invocation of the corresponding visual models. Additionally, we extend CE3D++ to monocular 4D scenes by imposing motion constraints on moving objects and further fine-tuning the LLM by creating a trajectory dataset related to editing tasks, which enables the smaller LLM to schedule up to 30 different visual tools accurately. Experimental results demonstrate that CE3D++ effectively integrates multiple visual models to achieve diverse visual editing effects, possessing strong scene comprehension and multi-round dialog capabilities. The source codes and trained models are available at https://github.com/Fangkang515/CE3D.
Read the full abstract
Cryptography and Security
Not the Same Protector: Deployment-Dependent Protective Intervention in LLMs
We ask whether a model protects a user in the same way when that user speaks rather than types. Using a single distress vignette—a physical injury of unstated severity following an interpersonal conflict—we present four frontier models with matched inputs across voice, text, and raw API deployment conditions (n=30 per cell) and code each response along five binary protective indicators, including whether the model issues an explicit medical-care directive. Voice-interface responses are markedly shorter than text-interface responses for three of the four models, and protective behavior contracts alongside that compression: medical directives are at ceiling under both the API and text conditions but decline under voice for every model tested. The contraction is not reducible to length. One model produces voice and text responses of comparable length yet still drops medical directives, and another falls below ceiling between its API and voice conditions, whose responses are of nearly identical length. Under raw API access the pattern is categorical rather than partial: no model asks after the user's safety even once. These results show that protective intervention is sensitive to the surface through which a request arrives, that this sensitivity is detectable using a simple protective coding scheme, and that it is not explained by turn length alone.
Read the full abstract
Dynamical Systems
Decay of correlations and normal approximation for nonstationary heterochaos baker maps
We study statistical properties of nonstationary compositions of a sequence of heterochaos baker maps. For maps whose central direction is mostly expanding, we establish three main results: exponential rate of memory loss for a broad class of measures (Theorem 1.1), a functional correlation bound with stretched exponential decay (Theorem 1.2), and an error bound for the multivariate central limit theorem (CLT) in the Wasserstein-1 distance (Theorem 1.3). The proofs are based on the construction of suitable Gibbs–Markov induced maps for the nonstationary compositions. In a certain special case we obtain a concrete estimate on the rate of convergence in the multivariate CLT (Theorem 1.4).
Read the full abstract
Cryptography and Security
Mechanizing Typed Regulatory Actions for Security Tokens: Semantics, Falsification, and Bounded EVM Evidence
Security-token standards expose privileged controls without identifying the legal effect executed or the evidence and reversal obligations it carries. We formalize in Isabelle/HOL a reference execution semantics for the six ERC-8319 meanings: FREEZE, SEIZE, CONFISCATE, LIQUIDATE, RESTRICT, and RECOVER. It distinguishes applied, rejected, and operational-failure outcomes and mechanizes action-specific reversals, replay and epoch rules, complete frames, case-local terminality, and final receipts; the session builds without unproved placeholders or additional axioms. An indistinguishability theorem shows that bound kernel inputs cannot establish external facts about title, settlement, or entitlement. Constructive witnesses and direct mutations establish reachability and sensitivity for the declared fault set. For a successor ERC-TRUST Solidity/EVM candidate, we report separately scoped Foundry, Certora, Kontrol/KEVM, mutation, deterministic-build, runtime-identity, and independent-reproduction evidence. All twelve named evidence lanes pass with none pending, while the 74-row obligation ledger remains conditional: 70 rows are closed, two runtime-link rows remain successor obligations, and two are inapplicable. The Native runtime is bound separately from an ERC-3643 interoperability reference that explicitly reports Partial and full=false, not Verified Full. These results do not establish complete Isabelle-to-Solidity-to-EVM refinement, compiler correctness, audit completion, production readiness, deployment verification, or external legal truth. They provide a machine-checked domain semantics and an explicit map of proved, bounded, assumed, and open results.
Read the full abstract
Computation and Language
AI Historian: Helping historians organize and verify person-centred temporal clues from dispersed historical narratives
History is not preserved in complete, continuous form. Accounts of a person's activities, relationships and historical contexts are scattered across texts, chapters and narrative perspectives; historians must retrieve, identify and compare these materials to reconstruct temporal sequences and verify them against sources. Here we present AI Historian (AIH), an AI agent system that helps historians organize person-time evidence from dispersed biographical narratives. It takes source sentences as evidence units, identifies people and temporal cues, verifies candidate cross-text associations and infers comparable temporal ranges while preserving traceable source-text evidence. We evaluated AIH on six Shiji cases concerning Liu Bang, Xiang Yu and Xiao He. AIH Agent achieved a temporal-localization MicroIoU of 86.2%, compared with 81.3% for human-only annotation and 17.1% for direct large-language-model prompting; it required about 14 min, versus 1 h 32 min for human-only annotation. We further applied AIH to the Twenty-Four Histories and other ancient Chinese histories, ancient Japanese and Korean histories, and modern and contemporary historical materials, and released the results through Westlake Historian. These results indicate that AIH can reduce the cost of organizing historical materials at scale while turning connections obscured by chapter-based narration into traceable, revisable research questions for collaborative testing.
Read the full abstract
High Energy Physics (Phenomenology)
Higgs boson pair production in SMEFT: shape and truncation studies
We present a comparative study of the influence of Wilson coefficients on Higgs boson pair production cross sections in SMEFT up to dimension-6, also including loop-suppressed operators. Our study accounts for NLO QCD corrections, and is based on updated parametrizations of the cross section as a polynomial in the Wilson coefficients. We focus on differential distributions for the Higgs boson pair invariant mass, as well as the transverse momenta of the leading Higgs bosons, identifying benchmark scenarios exhibiting pronounced shape distortions in the $m_{hh}$ and $p_{T,h}$ distributions. While some of these scenarios remain physically meaningful under a purely linear truncation of the SMEFT expansion, others yield unphysical differential cross sections when quadratic dimension-6 contributions are omitted, exposing the breakdown of the linear approximation for large anomalous couplings.
Read the full abstract
Numerical Analysis
D-RADI: A Low-rank ADI Algorithm for Solving Large-scale Discrete-time Algebraic Riccati Equations
The low-rank alternating direction implicit (ADI) method is an efficient numerical technique for solving several types of large-scale matrix equations that admit low-rank solutions. The discrete-time algebraic Riccati equation (DARE) is an important matrix equation with applications in state estimation, controller design, and filter design. In the literature, the low-rank Cholesky factor ADI method for Stein equations has been used within Newton iterations to solve large-scale DAREs. However, no dedicated low-rank ADI solver is available for such DAREs. To address this gap, this paper presents a low-rank ADI solver for large-scale DAREs. We also propose an efficient approach to generate ADI shifts automatically, which makes the proposed solver fully autonomous for solving DAREs. The effectiveness of the proposed solver is compared with MATLAB's idare on a moderate-order problem. Efficiency and accuracy are further demonstrated on large-scale DAREs of order $10^6$. Numerical results confirm that the solver is efficient, accurate, and fully autonomous.
Read the full abstract
Systems and Control
A Systematic Approach to Mechanism Design with Stochastic Dynamic Stability
We consider a resource allocation problem with strategic agents that have private stochastic satisfaction functions and local constraints. To achieve a global optimal solution, we propose an incentive mechanism that induces a game among the agents. For the payment function of the mechanism, we construct a family of quadratic functions using the linear matrix inequality (LMI) approach that implements the social welfare maximizing outcome on the unique Nash equilibrium (NE) of the induced game while ensuring budget balance and individual rationality. Moreover, we propose a decentralized variable sample-size proximal best-response (VS-PBR) algorithm with Krasnoselskij iteration where only aggregate information is available to the agents. The algorithm is dynamically stable, as it is proven to converge in the mean-square sense to the NE of the game. The efficiency of the mechanism is then investigated on the Sioux Falls City transportation network, where electric vehicle (EV) users jointly select their destination and route.
Read the full abstract
Systems and Control
From Rigid to Adiabatic: Canonical Regularization of AC Networks via Action-Angle Variables
Traditional power system analysis relies on timescale separation and the rigid-network assumption, freezing electromagnetic transients into algebraic power-flow equations via Steinmetz's phasor theory. As grid-forming converter penetration increases, magnetic energy dynamics on transmission lines interact with converter control loops on comparable timescales, challenging this rigid-network assumption. Returning to Faraday's law of electromagnetic induction driven by rotating magnetic fields, this paper models the transmission lines' rotating magnetic fields in action-angle canonical coordinates, regularizes the rigid algebraic constraints of power-flow equations into canonical equations on adiabatic symplectic manifolds, and establishes a port-Hamiltonian standard form for AC power grids. Based on the minimal-counterexample principle and using the equal-area criterion's classical two-machine system, this paper reveals a latitudinal instability channel via Bloch-sphere coordinates: Q-V control releases voltage-amplitude freedom, shifting the stability boundary from the equatorial UEP (unstable equilibrium point) to a saddle point, thereby unifying the analytical frameworks of P-delta angle stability and Q-V voltage stability in power system analysis.
Read the full abstract
Artificial Intelligence
APIFlow-Bench: Measuring Whether Agents Survive Long, Dependent API Workflows
Tool-using agents are commonly evaluated by a single bit: whether an end-to-end workflow completed. This metric fails to distinguish failures that matter in production, such as expired credentials, malformed payloads, or correct execution followed by incorrect final delivery. We introduce APIFlow-Bench, a fully auditable benchmark for long-horizon, dependent REST-API workflows that decomposes performance into seven engineering capabilities and requires agents to produce answers supported by the actual call path. We generate synthetic API worlds forward, subtask by subtask; each subtask is admitted only after a zero-LLM self-test triad verifies its grader and an oracle establishes solvability, and an adversarial audit identified and fixed six grader exploits. Grading is deterministic and provenance-sensitive: a state check traces a mock-minted canary through the API data flow to the response the answer must originate from, and a typed answer card is verified field by field. We release all answer keys and 44,362 unredacted execution transcripts. Across 19 frontier and open-weight models under one neutral scaffold, we find: (1) longer dependency chains degrade success, from 93% on individual subtasks to 74% on clean 20-subtask chains and 61% when including the 8% of chain trials that a model-consensus screen flags as passed by no model; (2) reliability separates models more than best-case capability, with best-of-five spanning seven points but all-five-of-five reliability spanning 44 points; (3) the independent-error account of compounding failure does not fit the data: pass rates on 20-subtask chains are 33 percentage points above the product of subtask-level rates, and on the clean slice 77% of failing runs reached the correct final state and failed only at delivery.
Read the full abstract
Artificial Intelligence
Beyond Correctness: Validity-Oriented Evaluation of Biomedical LLM Judges
We propose a scalable, validity-oriented pipeline for evaluating biomedical LLM judges when high-quality human judgments are scarce. First, we augment existing human-labelled biomedical benchmarks with deterministic, metric-grounded mutations that produce auditable preference pairs. Second, we evaluate judges beyond aggregate correctness using three deployment-relevant dimensions: correctness against metric-derived gold labels, robustness under repeated stochastic sampling, and compliance with the requested output format. We use this pipeline to assess Llama-3.1-8B-Instruct under four regimes: (1) base, using the instruct model as is; (2) SFT, distillation-based supervised fine-tuning only; (3) RL, GRPO-based reinforcement learning only; and (4) SFT$\rightarrow$RL, SFT followed by RL. The base and single-stage regimes struggle on structured medical discrimination such as PICO extraction and clinical calculations, whereas SFT$\rightarrow$RL performs best across correctness, compliance, and robustness; gains concentrate on decomposable tasks (PICO, MedCalc), at times matching or outperforming frontier models.
Read the full abstract
Computer Vision
Efficient Language-to-Vision Feature Injection for Referring Single-Object Tracking
Referring single-object tracking enables language-grounded target initialization and subsequent tracking by jointly leveraging semantic cues and visual templates. The core difficulty is to use language differently across stages: it is indispensable for grounding but can induce semantic drift during tracking when overemphasized. Meanwhile, current methods often require costly vision-language alignment training. We present LVTrack, a pure transformer framework that introduces a mode-conditioned Gated Feature Injector to adaptively regulate textual guidance and alleviate semantic drift. Together with targeted adaptations, it directly harnesses a frozen vision-language pretrained model, greatly reducing training cost and preserving strong language understanding. To further improve temporal localization, LVTrack integrates hybrid relative-absolute positional encodings with a lightweight memory mechanism and optimizes autoregressive box prediction using a Gaussian-smoothed KL loss. Extensive experiments on standard benchmarks demonstrate that LVTrack achieves strong performance.
Read the full abstract
Combinatorics
Cluster Geometry of Universal Schubert Polynomials I: Geometric Bases and Cluster Monomials
We study Fulton's universal Schubert polynomials through the cluster structure on the upper unitriangular group and representations of type $A$. The convex hull of the initial tropical vectors of standard-elementary monomials is unimodularly equivalent to a product of simplices, and right-regular multisegments index three integral geometric bases of the standard-elementary module. A recursive prefix-record condition characterizes those universal Schubert polynomials that are homogeneous under diagonal conjugation, coincide with the corresponding generic and dual canonical basis elements, and are extended cluster monomials. On this locus, cover forests give squarefree factorizations into cluster variables. Outside it, the initial Laurent expansion has a coefficient $-1$.
Read the full abstract
Soft Condensed Matter
A route to the thermodynamics of colloid-polymer mixtures from structural information
Liquid-state theory provides a fundamental connection between microscopic structure and macroscopic thermodynamic behaviour. Here, we develop a framework for predicting thermodynamics and phase behaviour directly from structural correlations, using either radial distribution functions or static structure factors as input. The approach constructs a free-energy functional from structural information obtained at a single thermodynamic state point, without requiring explicit knowledge of the underlying interaction potentials. This circumvents a central difficulty in modelling complex fluids, for which effective interactions are often unknown or rely on approximations. We demonstrate the framework for a range of model fluids, including colloidal and colloid–polymer systems, with predictions in good agreement with molecular simulation data. The results show that structural information at a single state point can provide sufficient information to predict the broader thermodynamic response of a system. This establishes a route toward inferring phase behaviour directly from experimentally measured structure, even when the microscopic interactions are not known a priori.
Read the full abstract
Computer Vision
Dancing Stick Figures: An Introductory Dataset for Training Video Generation Models
Training a video-generation model from scratch is hard for reasons that precede model design. The feedback loop is long: a failure that appears only after a training run can make each attempted fix another run. The data are hard to reach: the corpora and recipes behind strong models are large, heterogeneous, and often unreleased. And scoring is blunt: open-ended generation has no single correct output, and an aggregate score does not by itself establish whether a sample succeeds or which property failed. Dancing Stick Figures is a synthetic video dataset built against these three obstacles. For iteration speed, its 64x64, 64-frame reference task is sized for practical repeated training on a single workstation GPU. For accessibility, the release is a 0.79-GB training tier of 4,020 video clips–1,340 six-second source motions, each rendered from three cameras by a deterministic dataset-generation harness–with checkpoints and a Colab workflow that reruns the reference training pipeline at reduced budget on a 16 GB Tesla T4. For scoring, every frame retains its generating state (ARDY cskel27 joint positions, camera, body parameters, and source motion) and per-pixel depth, surface normals, and part labels. These annotations support dataset-specific metrics for visible topology and part-wise motion; corruptions expose their sensitivities and blind spots.
Read the full abstract
Statistical Mechanics
Light-induced nonconservative static forces in many-body systems
In a quantum many-body system, a periodic drive can often generate effective static forces on the slow collective degrees of freedom. We study the static forces generated by light on electronic order parameters in solid-state systems. We show that the forces can have nonconservative components that originate from dissipation, i.e., optical absorption. This effect is demonstrated in two nontrivial examples. In excitonic insulators, via interband excitations, the light field generates a nonconservative force on the phase of the excitonic order parameter. This force leads to an acceleration of the phase, which manifests as a shift of the photon emission peak from an exciton condensate. In materials with an incommensurate charge density wave, a propagating light field generates a nonconservative force on the phase of its order parameter. It drives the charge density wave into sliding motion, leading to a DC electric current via topological Thouless pumping.