Materials Science
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 11 September
Helical order can facilitate symmetry breaking and emergent physical responses in crystalline materials, yet how intrinsic chirality manifests beyond atomic length scales remains poorly understood. Here, the direct observation and quantitative characterization of long-range crystallographic helicity in template-grown tellurium (Te) quantum wires on amorphous substrates are reported. Four-dimensional scanning transmission electron microscopy (4D-STEM) enables quantitative mapping of crystallographic orientation with nanometer-scale spatial resolution. The resulting orientation maps establish continuous mesoscale lattice twisting, providing direct evidence of long-range crystallographic helicity. Correlated orientation and strain mapping reveal pronounced lateral strain heterogeneity, with compressive strain concentrated within the wire interior. Systematic analysis across multiple wires suggests that higher twist rates are generally associated with weaker lateral compressive strain, narrower wires, and better atomic chain - template axis alignment. Complementary first-principles calculations on finite Te nanorods further suggest that twisting is intrinsically accessible in nucleus-scale Te clusters and strain can bias the preferred torsional state. Together, these results support a growth-incorporated, strain-biased picture in which nanoscale confinement and anisotropic strain facilitate torsional relaxation and stabilize mesoscale helicity in Te nanostructures highlighting strain and confinement as potential routes for engineering chiral lattice states in van der Waals nanostructures.
Nuclear Experiment
Results on $^{235}$U(n$_{\mathrm{th}}$,f) isotopic fission yields using prompt and delayed gamma rays at the FIPPS spectrometer of the ILL
We present preliminary results from a measurement campaign conducted with the FIPPS $\gamma$-ray spectrometer at the ILL using an active target made of $^{235}$U dissolved in a liquid scintillator. Prompt $\gamma$ rays were used to extract absolute, independent isotopic fission yields. We obtained yields for a selected set of well-produced even-even nuclei from the $^{235}$U(n$_{\mathrm{th}}$,f) reaction and compared them with JEFF-3.3 evaluated data. This included the doubly magic nucleus $^{132}$Sn, for which we observed a pronounced deficit (about a factor 7) with respect to JEFF-3.3. Using the FIFRELIN fission fragment de-excitation code, we interpreted this discrepancy as evidence that $^{132}$Sn is mainly produced in its ground state at scission or after neutron evaporation. Delayed $\gamma$ rays recorded in the same experiment were also analysed. No deviation was found in the beta decay of $^{132}$Sn compared to the nuclear database. However, an excess of about a 40% was found in the $\beta$ decay of $^{90}$Kr, which may originate from an overestimated ground-state feeding of $^{90}$Rb in the evaluated data.
Read the full abstract
Quantum Physics
QMA has perfect completeness
We prove $\mathsf{QMA} = \mathsf{QMA_1}$, i.e., every quantum Merlin-Arthur proof system can be made perfectly complete. Our construction uses only Hadamard, Toffoli, and $X$ gates, yielding a universal gate set for $\mathsf{QMA_1}$. As a consequence, quantum $3$-SAT is $\mathsf{QMA}$-complete. The construction relativizes to classical oracles, so known classical-oracle separations of $\mathsf{QMA}$ from $\mathsf{QCMA}$ extend to $\mathsf{QMA_1}$.
Read the full abstract
Machine Learning
Attention Quantization for Tabular Foundation Models
With the recent rise and adoption of tabular foundation models, optimizing their inference performance becomes an emerging field for efficiency research. While the models are architecturally similar to transformer-based large language models (LLMs), the size and serving patterns differ significantly. We show that the focus should be on the attention calculation and less on weight or KV cache quantization, which are more popular in LLMs. We develop a quantization strategy for queries, keys, and values to FP8 and use explicit FP8 matrix multiplication instructions to speed up the attention calculation. We find that it is crucial to align the quantization error in the test rows with the quantization error in the training rows, as otherwise the accuracy drops drastically. Our Triton kernel achieves a speedup up to 1.7x over regular 16-bit kernels, and we show that on TabPFN-v3 and TabICLv2 there is no relevant accuracy loss across TabArena and BeyondArena.
Read the full abstract
Classical Analysis and ODEs
Equal-Phase Monotonicity for Weighted Jacobi-Radau Functions and Jacobi Lebesgue Constants
For the Jacobi family with parameters $(0,\beta)$, $\beta\geq-1/3$, we prove a continuous comparison theorem for the associated weighted Jacobi–Radau functions. When two consecutive functions are parametrised by the same Prüfer phase, the function of higher degree attains that phase closer to $\theta=0$ and has strictly larger amplitude. In particular, the moduli of all corresponding relative extrema increase strictly with the degree. The lower bound $-1/3$ is sharp for this continuous statement: when $-1<\beta<-1/3$, the amplitude inequality is reversed at sufficiently small positive phases. This local reversal does not determine the optimal range for the discrete extremal inequalities. For $-1/3\leq\beta\leq0$, the positive Jacobi product formula identifies the endpoint values of the Lebesgue functions with the global Lebesgue constants, so $(\Lambda_n^{(0,\beta)})_{n\geq0}$ is strictly increasing. At $\beta=0$ the weighted Radau functions reduce to $P_m^{(0,-1)}$. We thereby recover the theorem of Wong and Zhang and obtain, through an exact total-variation formula, a short proof of the Qu–Wong theorem on Legendre Lebesgue constants. The representation and the termwise comparison together realise the alternative-expression approach proposed by Qu and Wong, without asymptotic expansions, error bounds, or finite numerical verification. The proof is based on the equal-phase Prüfer architecture developed in the first author's earlier preprint arXiv:2608.01404; it applies that architecture to the problem posed by Qu and Wong.
Read the full abstract
Physics and Society
Geometry of Urban Order
Urban street networks are modeled as spatial graphs whose geometry reflects planning, history, and physical constraints. Global orientation measures describe the overall order of a city but do not reveal how street directions form coherent neighborhoods. Here we introduce bearings communities—spatially connected domains whose streets share a common orientation—and analyze them using concepts from the physics of polycrystalline materials. Across $1025$ large cities worldwide, these domains are almost perfectly ordered: their mean coherence is $0.964$ and varies by only $1.4\%$ from one city to the next, far above the value expected for randomly oriented streets. Every city is therefore a mosaic of well-ordered orientational grains, differing only in their number, size, and mutual alignment. Because the grains are equally regular everywhere, a city's global order is set mainly by how many of them it averages over: a city appears ordered because it is built from a few large grains, not because its fabric is locally more regular. A null model in which grain orientations are randomized confirms this grain-counting mechanism. Finally, the interface fraction $\gamma$ measures how strongly domains interlock, yielding a morphology diagram that organizes cities continuously between grid-like, polycrystalline, and fragmented limits.
Read the full abstract
Quantum Algebra
Frobenius functors and one-sided Hopf algebras in braided monoidal categories
Evidence suggests a tight connection between the existence of antipode-like maps on bialgebra-like structures, and the Frobenius property for the associated free Hopf module functor. In this paper, we prove that in any braided monoidal category satisfying few mild assumptions, a bimonoid is a one-sided Hopf monoid and the antipode is a bimonoid anti-homomorphism, if and only if the free Hopf module functor is Frobenius. Our interest in these structures stems from having found genuine examples of one-sided Hopf monoids in contexts where the braiding is non-trivial. In fact, we provide a construction of the free one-sided Hopf monoid over a comonoid in any symmetric monoidal category with some non-restrictive additional assumptions. We present several examples of this construction, and describe the resulting one-sided (sometimes two-sided) Hopf monoids.
Read the full abstract
Number Theory
Kronecker sequences beyond the torus: nearest-neighbour distances and best returns
We study nearest-neighbour distances and strict best returns of cyclic orbits on $(V\times G)/\Lambda$, where $V$ is a finite-dimensional real normed space, $G$ has a bi-invariant ultrametric, and $\Lambda$ is an arbitrary closed subgroup. For the quotient of the maximum product metric, the number of distinct nearest-neighbour distances in every finite orbit segment is at most $P(\overline V)+1$, where $P$ is an open-ball packing constant and $\overline V$ is obtained by factoring out the continuous real part of $\Lambda$. For inner-product norms the bound depends only on the span of the projected subgroup. This gives $2^d+1$ for the maximum norm on $\mathbb R^d$ and strong-kissing-number bounds for inner-product norms, extending the corresponding real and adelic distance bounds. Our second main result is an exact abelianization principle: every such cyclic orbit is isometric, with its integer indexing preserved, to a cyclic orbit on an abelian mixed quotient $(V\times U)/\Lambda_{\mathrm{ab}}$. The construction preserves the reduced and effective real factors. Thus noncommutativity and non-normality enlarge the class of ambient presentations but not the class of cyclic-orbit metrics. The record-count identity and its growth consequences require only an isometric embedding of a metric space. For mixed homogeneous spaces we prove packing recurrences and, using abelianization, the sharp three-step and five-step recurrences in effective Euclidean dimensions one and two, without a normalizer assumption. We also extend the purely real recurrence to noncompact quotients and give counterexamples showing why the stronger real recurrence can fail after an ultrametric factor is added.
Read the full abstract
Econometrics
Identification via Distributional Shifts without Exclusion Restrictions
This paper studies identification and inference in a triangular system with an endogenous regressor when exclusion restrictions are unavailable and the dependence between structural disturbances is modeled through an unrestricted control function. In this setting, standard orthogonality conditions do not deliver point identification, as the unknown control function can rationalize a wide range of structural coefficients. We show that identifying information can be extracted from distributional shifts in the first-stage disturbance induced by an auxiliary variable that may directly affect the outcome. Imposing a local restriction on the log density ratio, together with an explicit bound on the sieve approximation error of the control function, we derive moment inequalities that restrict the structural parameter. We develop a practical inference procedure based on test inversion and multiplier bootstrap that accommodates generated regressors, cross-fitted sieve estimation, and locally estimated density-ratio nuisances. The results clarify how identification can be recovered from weak local distributional structure in the absence of classical instruments.
Read the full abstract
Algebraic Geometry
Positivity properties of Schur classes
Results of Fulton-Lazarsfeld and Ross-Toma show that Schur classes of ample and nef vector bundles have remarkable positivity properties. We generalize these results to a purely algebraic setting. We show that if a projective bundle ring, a ring which resembles the cohomology ring of a projective bundle over a smooth complex projective variety, has the Kähler package with respect to a suitable cone, then the associated Schur classes satisfy a version of the Hodge-Riemann relations. This result is new even for the projectivization of an ample vector bundle over a smooth complex projective variety. We apply this result to prove that Schur coefficients of matroids are nonnegative.
Read the full abstract
Computer Vision
Label-Guided Knowledge Distillation for 3D-CNNs in Action Recognition
As a key model compression technique, knowledge distillation aims to transfer knowledge from a high-capacity teacher model to a lightweight student model for enhancing the latter's performance. In this work, we reviewed the feature knowledge distillation for 3D-CNNs and observed that most feature distillation methods in video analysis are simple adaptations of those used in image analysis, often neglecting the differences of video features in the temporal dimension. To address this issue, we proposed Label-Guided Knowledge Distillation (LGKD) to guide the distillation of student model features using ground truth labels. Our method entails two components: sample-wise distillation and class-wise distillation, enabling the student model to learn feature representation of the teacher model at two levels. Sample-wise distillation utilizes label information and the teacher's probability distribution to guide the learning of features that significantly impact temporal accuracy while mitigating noise. Meanwhile, class-wise feature distillation employs a prototype network to further capture the relational knowledge among samples within the same category, enhancing the student's ability to learn higher-dimensional semantic information and improving model generalization. To demonstrate the effectiveness and superiority of our method, we conducted comprehensive experiments on two benchmark action recognition datasets, UCF101 and HMDB51, achieving competitive results.
Read the full abstract
Classical Analysis and ODEs
Sharp $L^2 \to L^4$ extension inequality for quadratic surfaces in finite fields
In this work, we establish the sharp $L^2 \to L^4$ extension inequality for a large class of quadratic surfaces in $\mathbb{F}^3$, where $\mathbb{F}$ is a finite field of odd characteristic. These surfaces include the cones without the origin and the spheres with nonzero radius $j$ such that $-j$ is not a square in $\mathbb{F}$. We also show that the maximizers of these inequalities must have constant modulus.
Read the full abstract
Quantum Physics
Grover Search with Semiconductor Spin Qubits at Ambient Conditions
Grover's algorithm is executed on a commercial quantum computer based on nitrogen-vacancy centers in diamond operating under ambient conditions, achieving fidelities up to $99.98\,\%$. Within a $N=8$ search space of three solid-state nuclear spin qubits, the measured success probabilities of finding a single or two marked states are $(77.3 \pm 3.4)\,\%$ and $(87.0 \pm 4.2)\,\%$, respectively. These values surpass published results for superconducting qubits or any quantum computer operating at room temperature. In addition, the paper also details the calibrated fidelities of the implemented universal gate set.
Read the full abstract
Number Theory
Projection and fibering in groups of bounded exponent
We develop a projection and fibering method for sets of small combinatorial doubling in (not necessarily abelian) discrete groups. As an application, in the abelian case we prove that, if $A$ is finite, the ambient group has exponent $r$, and $|A+A|\leq K|A|$, then \[ |\langle A\rangle|\leq r^{(2+o(1))K}|A|. \] This answers a question of Ruzsa with an optimal leading coefficient, independent of Fox–Pham. The main ingredient is a discrete version of a fiber spillover argument. For sets in $2$-step nilpotent groups of exponent $r$, we also prove that $|A^3|\leq K|A|$ implies $|\langle A\rangle|\leq r^{(2+o_K(1))K}|A|$. The proof combines the abelian theorem with a weighted averaging of central fibers and commutators.
Read the full abstract
Computational Physics
Physically Consistent 2D to 3D Pore Space Synthesis: 1. Dynamic Grain Packing Under Strict Morphology Constraints
The 3D digital reconstruction of multi-scale, heterogeneous porous media is traditionally limited by the resolution boundaries and artifacts of X-ray computed tomography (XCT). This paper introduces a physically consistent 2D to 3D pore space synthesis framework that supersedes conventional geometric shuffling with dynamic Newtonian gravitational deposition using a high-performance rigid-body mechanics engine. To honor strict physical contact mechanics, statistical 2D targets, and morphological constraints, we develop a two-stage dynamic packing pipeline that combines realistic 3D grain shapes with a virtual sub-particle insertion engine. The framework is validated against synthetic and natural core benchmarks. First, we show that our stereological integral equation systematically converts size distributions; bypassing this inversion step induces a systematic 50% deviation in permeability predictions despite identical total porosities, independently altering flow channel topology. Second, temporary virtual spheres serve as architectural placeholders to stabilize large non-equilibrium dissolution voids, ensuring the reconstructed model satisfies target flow metrics within the prescribed tolerance. Finally, when applied to a fine-grained, weakly consolidated sandstone, the workflow exposes and suppresses systematic XCT artifacts, including phantom internal porosity. Ensembles of two-point correlation functions demonstrate that the hydrodynamically and structurally synthesized digital twins quantitatively exhibit closer structural proximity to the uncorrupted rock geometry than the reference XCT dataset itself. This methodology establishes a scalable foundation for high-fidelity multi-phase rock reconstruction, with potential extensions toward complex amorphous cement phases.
Read the full abstract
Statistics Theory
A class of nonparametric homogeneity tests on the circle
We develop a unified framework for $c$-sample homogeneity testing on the circle. The proposed class of $c$-sample Sobolev tests generalizes the two-sample Sobolev tests based on uniform scores and encompasses the existing multisample tests as particular cases. We further embed this class into a broader aggregation framework, where the samplewise Sobolev components are combined through general merging functions, including average-type, maximum-type, and interpolating families. Within this class, we introduce the first Anderson-Darling-type homogeneity test for circular data, and we further propose two new tests designed to detect multimodal departures from homogeneity, constructed from softmax and Poisson kernels. We derive the asymptotic null distribution of the class, prove its consistency against a broad family of fixed alternatives, and obtain the asymptotic distribution under shift-type local alternatives. The tests are distribution-free and therefore do not require resampling. A comprehensive simulation study demonstrates the strong power of the Anderson-Darling-type test compared with Cramér-von Mises-type competitors across several scenarios, and the effectiveness of the softmax and Poisson tests under several alternatives. The testing toolbox is applied to analyze the nursing patterns of polar bears.
Read the full abstract
Chemical Physics
First application of the transcorrelated method to noncovalent interactions: The A24 dataset
We present the first application of transcorrelated (TC) coupled-cluster (CC) theory to noncovalent interactions within the xTC approximation. The method is assessed for the A24 dataset of hydrogen-bonded, mixed and pure dispersion bound dimers. The xTC interaction energies are computed at the CCSD, DCSD, and CCSD(T) levels in aug-cc-pVDZ (AVDZ), aug-cc-pVTZ (AVTZ) basis sets, and are compared with both canonical and explicitly correlated F12 methods. Because non-covalent interaction energies rely on delicate error cancellation between dimers and monomers, we optimize the TC Jastrow factor for each dimer, and reuse the same parameters for the corresponding monomer calculations. This shared-parameter strategy reduces stochastic optimization noise which would otherwise dominate the interaction energies. The results show that the xTC-CCSD(T)/AVTZ method performs extremely well for the hydrogen-bonded systems, with a mean-absolute error of only 0.007 kcal/mol in the interaction energies, with respect to the benchmark values generated with CCSD(T)/CBS + $\Delta$CCSDT(Q) + core corrections. For pure dispersion bound systems the errors are slightly larger (0.055 kcal/mol), leading to an overall MAE of 0.030 kcal/mol for the entire dataset. Adding a $\Delta$MP2 correction to the xTC-CCSD(T)/AVDZ brings these results close to AVTZ quality and provides a practical route toward larger noncovalent systems. A decomposition of the xTC interaction energy into a mean-field and correlation contribution shows that part of the correlation contribution is systematically shifted by the TC method to the mean-field contribution. This physically appealing feature indicates that the TC method has great potential for the quantitative description of noncovalent interactions, and opens a new route for high-accuracy quantum chemistry to be applied to systems of biological and soft-matter interest.
Read the full abstract
History and Philosophy of Physics
Gauge Matters More Than Curvature
Recent orthodoxy, which we call Curvature Fundamentalism, posits that the field strength $F$ on spacetime exhausts the physically significant content of electromagnetism, at least on prosaic (contractible) spacetimes. We argue that it is gauge invariance that constrains physical significance. As a result, there are physically significant quantities that $F$ does not capture, even on contractible spacetimes. We show that this result is robust: in the presence of charged matter fields, there are interference experiments whose outcome cannot be predicted by $F$ or any other point-local gauge invariant quantities on the basis of initial Cauchy data. We prove this using the example of relative phase of a Klein-Gordon-Maxwell field, and use it to identify a context in which the widely-claimed equivalence between the Maxwell-Faraday formulation $EM_1$ and the gauge formulation $EM_2$ of electromagnetism is untenable, and that new strategies may be needed for the reductionist project.
Read the full abstract
Group Theory
The number of distinct element orders in a finite group
For a finite group $G$ let $\omega(G)$ denote its spectrum, the set of orders of its elements, and set $\eta(G)=|\omega(G)|$, the number of distinct element orders. The set $\omega(G)$ has been studied intensively, but its cardinality has received little systematic attention as an invariant in its own right. We develop the theory of $\eta$. After describing its behaviour under subgroups, quotients, sections, direct products and Frobenius extensions, we prove the two-sided bound $1+\sum_{p\mid|G|}v_p(\exp G)\le\eta(G)\le\tau(\exp G)$, in which the lower equality characterises the CP-groups, those all of whose elements have prime-power order, and the upper equality the groups whose spectrum realises every divisor of the exponent; nilpotent groups lie at the upper end, giving $\eta(G)=\tau(\exp G)$. We also prove $\eta(G)\le\tau(|G|)$ with equality only for cyclic groups, and $\eta(G)\ge\pi(|G|)+1$ with equality only when every nontrivial element has prime order. We show that $\eta(G)\le3$ forces solvability by an argument resting only on Burnside's theorem, and, combining this with the recognition of $A_5$ by its spectrum, that $A_5$ is the unique non-solvable group with $\eta(G)=4$. For the symmetric and alternating groups we record self-contained proofs of closed criteria for membership in the spectrum through a single arithmetic function, placing the associated counting sequences in the present framework, and we determine the extreme values of $\eta$ over all groups of a fixed order.
Read the full abstract
Robotics
Global Path Planner with Multi-Model Switching
This work enhances global path planning via a pure-pursuit controller with multi-model kinematic switching that sustains plan fidelity across diverse terrains. The system includes a traversability graph for terrain analysis, a Heading-Aware A* algorithm for generating feasible paths, and a multi-model Pure Pursuit controller for dynamic tracking. A core innovation is adaptive kinematic modeling, enabling real-time switching between kinematic models based on terrain features and robot states. This adaptability optimizes path efficiency and energy use in challenging scenarios. We validate the approach in simulation on different platforms, namely the Artaban quadruped and the X3 quadrotor drone, showcasing improved performance, robustness, and adaptability over standard baselines.
Read the full abstract
Probability
From skew to reflected SPDEs
In this article, we prove that the solution of the skew stochastic heat equation converges to the solution of the stochastic heat equation with reflection in the limit of infinite skewness. The result holds at equilibrium. This provides an SPDE analog of the link between skew and reflected Brownian motion. One of the intermediate results is the convergence in law of a penalised Brownian bridge to a 3-Bessel bridge.
Read the full abstract
Computer Vision
TileNet: Tile-Based CNN-SVM Architecture for Autonomous Unmanned Aerial Systems Inspection of Flat Roofs
Flat roofs are among the most influential components of the building envelope, governing both structural performance and thermal efficiency, and thereby contributing directly to household energy consumption, carbon emissions, and long-term environmental sustainability. Timely detection of roof defects is essential for reducing heating and cooling losses, preventing moisture-driven degradation such as mold growth, and supporting national climate-change mitigation goals. This paper presents a real-time, Unmanned Aerial System (UAS)-based deep learning framework that autonomously detects defects using live imagery captured during dual-altitude aerial passes. The multi-resolution flight strategy is designed to aid the identification of both small, fine-scale defects and larger structural issues, enabling more comprehensive assessments. To meet the strict computational and power constraints of embedded UAS hardware, the proposed framework integrates a tile-based architecture with a lightweight Convolution Neural Network-Support Vector Machine (CNN-SVM) classifier designed for low-latency onboard inference. The final model-comprising five convolutional layers and four dense layers, the last a linear SVM head, achieved a mean test accuracy of $94.4\%$ ($95\%$ confidence interval $\pm0.4\%$ over three seeds) on a photo-level split ($43,383$ training, $3,869$ validation, and $2,540$ test tiled and augmented images), outperforming GoogLeNet ($89.2\%$) and AlexNet ($79.8\%$). Experimental evaluations using real UAS imagery collected by onsite visits with DJI Matrice 350 RTK drone demonstrate that the system supports rapid, repeatable, and safe roof inspections while reducing human risk, lowering operational costs, and enabling more sustainable building maintenance.
Read the full abstract
Robotics
Comfort by Construction: Adaptive, Comfort-Bounded Action Spaces for Learned Driving Policies
Data-driven driving simulators command accelerations and steering rates from a fixed grid without constraining the realized accelerations and jerks. As a result, reinforcement-learning policies inflate safety metrics through abrupt, last-second maneuvers that lie far outside the range of human driving and would be unacceptable to occupants of a real vehicle, so the metrics measure simulator permissiveness rather than policy quality. Enforcing comfort bounds naively is not enough: lateral limits shrink quadratically with speed, so clamping a static grid saturates it and destroys fine-grained control ("grid collapse"). We propose an adaptive action parameterization that rediscretizes the grid at every step to span exactly the per-step feasible control set, via closed-form inversion of the lateral-jerk constraint. We further present PufferDrive-Editor, a browser-based tool to audit realized kinematics and author kinematically challenging scenes. On the Waymo Open Motion Dataset and a hand-authored slalom, our adaptive model holds comfort violations below 1% while outperforming clipped-grid and direct-jerk baselines in navigability.
Read the full abstract
Machine Learning
Dual-guided Hierarchical Edge Localization for Large-scale Optimal Transport Across Dimensions
Optimal transport (OT) compares distributions and aligns datasets in machine learning, yet unregularized discrete OT requires a linear program with quadratically many transport variables. We propose HELLO, a hierarchical solver that casts large-scale discrete OT as edge localization and uses dual potentials to guide both coarse-to-fine initialization and within-level refinement. Initialization propagates coarse dual potentials across a recursive subsampling hierarchy to assign candidate edges. Refinement then iteratively inserts the largest dual violators in each row and column until the relative KKT residual meets a prescribed tolerance, while budgeted pruning ensures linear memory complexity. For exact-arithmetic refinement, we prove finite termination at a global optimum under a symbolic lexicographic rule. At the million-point scale, HELLO attains lower transport objectives with order-of-magnitude runtime improvements over strong baselines across feature dimensions from single digits to thousands. It further scales to 1.28 million samples per marginal in 8192 dimensions on a single H100, using 41.6 GiB peak GPU memory while satisfying a full relative KKT residual below $10^{-6}$. Beyond standard discrete OT, the framework supports general pairwise costs and serves as a scalable balanced-OT oracle for semi-discrete OT, Gromov–Wasserstein, unbalanced OT, and OT-based Flow Matching.
Read the full abstract
Artificial Intelligence
How Good Are Frontier Models at Physics? Expert Re-Grading Reveals Broken Evaluations and Near-Saturation of Leading Benchmarks
Low reported scores on leading physics benchmarks, including those featured in the Artificial Analysis Intelligence Index (2026), suggest that frontier language models still struggle with advanced physics, a demanding test of their scientific reasoning and quantitative problem-solving abilities. Yet this impression does not always align with domain experts' experiences using these models in their work. We revisit these reported findings by evaluating frontier models on six widely used physics benchmarks and auditing them with experts, focusing on text-only problems with verifiable final answers. For each subfield of physics, faculty and graduate researchers with relevant expertise carefully review problem statements, reference solutions, and model responses to distinguish genuine model errors from grader errors, incorrect reference solutions, and ambiguous or underspecified questions. Most audited cases initially evaluated as incorrect reflect these benchmarking issues rather than errors in the models' physics reasoning. We then ask experts to address these benchmarking issues by correcting erroneous reference solutions and repairing or excluding flawed questions. We find that GPT-5.6-Sol's measured mean@4 rises from 47.3% to 78.7% on HLE-Physics and from 61.0% to 87.2% on CMT-Benchmark, while its corrected pass@4 reaches 94.4% on the 54 retained CritPt challenges. Corrected scores are computed on the retained evaluation subsets following expert review. Scores on the audited subsets of UGPhysics, PRISM-Physics, and PHYBench also rise substantially after correction. These findings suggest that current benchmarks substantially understate frontier models' ability to solve well-posed physics problems. Near-saturation on these closed-ended tasks highlights the need for more demanding, expert-validated evaluations.
Read the full abstract
Computation and Language
Tasks over Application Manuals: Revealing Gaps in Long-Horizon Procedural Reasoning for Language Models
Large language models (LLMs) have achieved strong performance on a wide range of natural language tasks, and recent benchmarks suggest that they are increasingly adept at multi-hop reasoning. However, these benchmarks are typically short-horizon, requiring only a small number of retrieval or inference steps, and provide limited evidence of reliability on real-world tasks that involve following manuals spanning hundreds of pages with complex, interdependent guidelines. In this paper, we introduce Tasks over Application Manuals (TAM), a benchmark for evaluating long-horizon procedural reasoning. We construct TAM by curating real-world tasks from two domains: ICD-10-CM clinical coding (mapping medical conditions to diagnostic codes) and U.S. federal sentencing (computing crime sentencing guideline outcomes, specifically offense levels), with human-validated labels. Each task requires following an authoritative manual with tens of thousands of rules and executing a sequence of interdependent steps across different sections to produce an exact answer. We evaluate general-purpose prompting approaches, including retrieval-augmented generation, ReAct-style prompting, and an agent-harness baseline on GPT-5, and find that the best exact-match performance remains extremely low: 1% on ICD-10-CM coding and 15.5% on sentencing tasks. These results show that current benchmarks may overestimate LLM reasoning ability and miss a key challenge: reliably following long, rule-based procedures. The complete TAM data and code are publicly available.
Read the full abstract
Cryptography and Security
IntentFuzz: A Protocol-Aware Fuzzer for Automated Invariant Violation Detection in Intent-Based Cross-Chain Bridges
Cross-chain bridges move value between blockchains. Intent-based bridges are a variant where a solver fulfills a user's declared outcome and an off-chain settlement layer later reconciles the fill against the deposit. Existing smart-contract fuzzers and static analyzers only flag known-bad code patterns or require protocol-specific hand-written assertions. This work formalizes a taxonomy separating invariant violations, safety properties a contract must enforce locally, from settlement exposures legitimately delegated to the off-chain settlement layer, and proposes IntentFuzz: a protocol-aware fuzzer that recovers a bridge's intent structure and deposit/fill function roles directly from unannotated Solidity source, then synthesizes multi-step fuzz sequences using an LLM-based fallback to help build call arguments. IntentFuzz recovers the correct intent structure in 9/9 benchmark protocols and classifies deposit and fill functions with 100% recall and 82% combined precision; across a corpus of 77 manually labeled contracts, it reaches 79.5% bridge-classification precision and 97.2% recall, and among confirmed bridges, struct selection reaches 88.6% precision and recall while deposit and fill classification each reach 100% recall. On 23 planted-bug mutants, IntentFuzz attains 100% recall and 100% precision, executing 273 templates (507 transactions in a median of 14ms per template). Across 24 real-world deployments, it confirms 17 genuine invariant violations under heuristic-only input generation, rising to 22 with its LLM-assisted tier enabled, spanning eight vulnerable GitHub repositories, each finding reproducible against public, deployed bytecode.
Read the full abstract
Computation and Language
Judging by the Cover: Cleaning LLM Truthfulness Benchmarks to Avoid Surface-Level Feature Leakage
Binary-choice truth benchmarks ask models to choose between a correct and an incorrect answer, but if the two answers differ systematically in surface-level features, models can exceed chance without performing the intended reasoning. We show that this failure mode is detectable and can be exploited by downstream classifiers. In TruthfulQA, a simple six-feature logistic classifier achieves substantial accuracy in separating correct from incorrect answers. We further show that similar surface-level artifacts are present in additional benchmarks. To counteract this, we developed a general mechanism to clean them by removing the most leakage-reinforcing pairs. We release a version of TruthfulQA with surface-feature leakage reduced close to chance and provide a mechanism, Audit-Prune, so that the datasets can be cleaned before release.
Read the full abstract
Combinatorics
Spanning subhypergraphs with degree constraints
An old result of Tutte states that any $d$-regular graph contains a spanning subgraph in which every vertex has degree $k$ or $k+1$, for every $1\leq k\leq d$. We generalize this statement to hypergraphs, showing, for example, that every $3$-uniform $d$-regular hypergraph contains a subgraph in which all degrees are $k, k+1$ or $k+2$, for every $1\leq k\leq d$. This statement is best possible in the sense that the corresponding statement with only two allowed consecutive values is not true. We provide generalizations of this statement to higher uniformities and discuss several open problems.
Read the full abstract
Materials Science
Reliable training of neural hyperelastic models via full-field data
We present a systematic investigation of the robustness and limitations of equilibrium gap-based calibrations for hyperelastic physics-augmented neural networks (PANNs), where we consider the special case of isotropic and polyconvex PANNs. In full-field parameterizations, it is commonly assumed that the displacement field is captured with sufficient spatial resolution for an accurate evaluation of the deformation field, and that the specimen is thin enough for plane stress to hold to a good approximation. Since these assumptions are never ideally satisfied in real experiments, we investigate, using synthetically generated data, how severely an under-resolved surface measurement and a non-negligible specimen thickness can affect the model parameterization. Furthermore, we perform calibration on real experimental data for a set of inhomogeneous specimen geometries. We show that the coverage of the admissible deformation states during calibration governs the ability of a model to generalize to unseen geometries and load cases; this ability can be improved further by appropriate combinations of specimens. Accurately depicting the material behavior underlying this rich data, however, requires a sufficiently flexible constitutive model, for which PANNs are well suited. Yet a rich coverage of deformation states alone is not sufficient: unless the calibration data comprise biaxial-tension-like states, models that include the second deformation invariant extrapolate unphysically towards equi-biaxial tension, whereas restricting the PANN to the first invariant remains reliable.