Peer Revenue Publish or profit

Papers

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

All of arXiv

Monday 7 September

Differential Geometry

Dual-connection midpoint matrix means and their Gauss composition

Nakamura [J. Comput. Appl. Math. 131 (2001)] proved that the arithmetic–harmonic matrix iteration converges quadratically to the geometric matrix mean, the Riemannian midpoint of the affine-invariant metric on positive-definite matrices. We consider the more general problem of when a Riemannian midpoint is a Gauss compound mean. A Riemannian metric and an affine connection determine three midpoint maps associated with the connection, its metric dual, and the Levi–Civita connection. We show that the Gauss composition of the first two equals the Levi–Civita midpoint if and only if the latter is invariant under one step of the iteration. We give a sufficient condition for this invariance in terms of an isometry which acts as the point reflection about the Levi–Civita midpoint and exchanges the connection with its dual. Under this invariance, the iterations converge quadratically for sufficiently close initial pairs. Nakamura's iteration is an example of this criterion. The criterion also applies to dual pairs of matrix power means whose Gauss composition is the geometric matrix mean. We give a dually flat Hessian example which shows that duality alone does not suffice. We also give non-flat examples with a Euclidean metric and a parallel cubic form in every dimension larger than one. In dimension one, we characterize the invariance completely by using the Matkowski–Sutô equation. For Euclidean dual pairs, we conjecture that the invariance implies that the cubic form is constant. We prove this conjecture in dimension one and for scalar multiples of a constant cubic form.
Read the full abstract

Dynamical Systems

Unique ergodicity, and not, for primitive substitutions on compact alphabets containing isolated points

We consider generalised subshifts generated by continuous substitution on compact Hausdorff alphabets. Although primitivity still implies minimality of the subshift generated by the substitution, Durand, Ormes and Petite showed, in contrast to the finite case, that primitivity no longer implies unique ergodicity, by constructing counterexamples with Cantor alphabet. Here we show that, even for the arguably simplest case of the one-point compactification of the natural numbers, primitivity is still insufficient for unique ergodicity, or even the existence of a natural length function (NLF). In previous work with Mañibo and Rust we showed that, for irreducible substitutions, unique ergodicity and existence of a NLF follow from strong power convergence of the renormalised substitution operator \(T\); criteria were also developed that can sometimes confirm strong power convergence. Here we show the converse, so that strong power convergence of \(T\) is equivalent to having both unique ergodicity and a NLF. We then consider the case of alphabets with only finitely many accumulation points, showing how upper bounds (and usually an exact formula) for the essential spectral radius of \(T\) can be derived from associated finite substitutions, determined by the behaviour of substitution of the accumulation points. For many primitive substitutions this can be used to prove quasi-compactness of \(T\), and thus unique ergodicity. For primitive substitutions on alphabets with at least one isolated point, we show that strong power convergence and quasi-compactness of \(T\) are equivalent and, in fact, that these properties are equivalent to iteration of substitution growing words in length uniformly across all seeds in the alphabet.
Read the full abstract

Computation and Language

Qwen-Audio-3.0-ASR Technical Report

In recent years, automatic speech recognition (ASR) has witnessed transformative advancements driven by three complementary paradigms: data scaling, model scaling, and deep integration with large language models (LLMs). However, bridging the gap between academic benchmark performance and real-world production utility remains a persistent challenge, particularly in handling diverse regional dialects, dynamic entities and hotwords, long-range contextual information, and disfluent spontaneous speech. In this report, we present Qwen-Audio-3.0-ASR, a Mixture-of-Experts (MoE) LLM-based ASR system designed to address these production demands through a unified, instruction-following framework. The model is built upon the Qwen backbone, and is trained on tens of millions of hours of large-scale speech data. Qwen-Audio-3.0-ASR supports transcription across 30 languages and 16 Chinese dialectal varieties spanning eight major dialect regions. Beyond multilingual and dialectal recognition, the model provides production-oriented capabilities including industry-domain entity recognition, hierarchical hotword customization, native single-pass transcription polishing, and long-audio contextual modeling. We further develop a dedicated streaming variant, Qwen-Audio-3.0-ASR-Streaming, for latency-sensitive applications. Extensive evaluations on Chinese, English, multilingual, and real-world industrial test sets demonstrate state-of-the-art or highly competitive recognition performance across a broad range of evaluation conditions, with strong performance relative to leading commercial and proprietary systems including GPT-4o Transcribe and Gemini 3.1 Pro.
Read the full abstract

All authors Chuanmeng Bian, Daren Chen, Peixin Chen, Zhigao Chen, Zhiyun Fan, Zhifu Gao, Bo Gong, Qing Gu, Jiajun He, Yawei Hu, Yunjie Ji, Jingbei Li, Xiangang Li, Xu Li, Zengxi Li, Zheng Li, Chengdong Liang, Baiji Liu, Ying Liu, Bin Ma, Yiping Peng, Yuezhang Peng, Zhendong Peng, Yu Pu, Yang Shi, Xin Shu, Jian Tang, Biao Tian, Peiyao Wang, Tianzi Wang, Wen Wang, Wupeng Wang, Cheng Wen, Yuzhong Wu, Zijian Xia, Yunchong Xiao, Nan Yang, Jianwei Yu, Jixing Yu, Binbin Zhang, Lei Zhang, Sitong Zhao, Guangdong Zhou, Yuan Zhou, Jianheng Zhuo

Materials Science

Switchable Altermagnetism in a Layered van der Waals Metal-Organic Framework Driven by Spin-Crossover

Dynamical control of altermagnetism is a key requirement for translating its unique spin-dependent functionalities into practical spintronic devices, yet effective switching mechanisms remain largely unexplored. Here, we open an unprecedented, versatile and programmable route based on spin-crossover to switch altermagnetism on demand in molecular materials. Using density functional calculations, we demonstrate an altermagnetic ground state in the layered van der Waals MOFs MnX$_2$(tdz)$_2$ (X = Cl, Br; tdz = thiadiazole), stabilized by anisotropic interlayer exchange interactions, which gives rise to a characteristic d-wave momentum-space spin splitting and an associated spin-splitter transport response. Our findings reveal that under hydrostatic pressure, a high-spin to low-spin transition reconfigures the Mn d-orbital occupation, thus modifying the magnetic exchange network, and stabilizing a different antiferromagnetic ground state whose symmetry suppresses the nonrelativistic spin splitting. Crucially, spin-crossover switches altermagnetism not by directly altering the electronic structure, but by changing the symmetry of the magnetic ground state. These results establish molecular spin-crossover altermagnets as a platform for externally reconfigurable spintronic devices.
Read the full abstract

All authors Diego López-Alcalá, Alberto M. Ruiz, Andrei Shumilin, José J. Baldoví

Computer Vision

Re-engineering SORT-based algorithms for low-cost small object tracking from omnidirectional footage

Multi-object tracking (MOT) has advanced rapidly in urban surveillance and autonomous driving, yet many trackers rely on ReID- and transformer-based appearance encoders and are designed for standard FoV cameras. These assumptions break down for low-cost omnidirectional deployments, where equirectangular projection introduces seam discontinuities and targets appear to be small and fast-moving. We address multi-object tracking of flying animals captured in remote environments using omnidirectional cameras. We propose a lightweight framework that re-engineers SORT-based tracking for this geometry, including (i) a Seam-Aware Motion Model that keeps the Kalman state continuous across the seam, (ii) a composite seam-aware association cost that pairs a wrapped Euclidean term with GIoU, and (iii) OmniSmall, a new benchmark of omnidirectional wildlife footage. On our new dataset, with ground-truth detections, our modifications improved over OCSORT by +8.51 HOTA, +9.41 MOTA, and +10.17 IDF1; with YOLOX detections the gain narrows to +1.95 HOTA. Our proposed methods improved tracking performance on OmniSmall and remained competitive on JRDB without adding appearance encoders while keeping the tracking stage CPU-only. Our dataset and source code are available at: https://github.com/Xin-Shu/OmniSORT.git.
Read the full abstract

All authors Xin Shu, Meegan Gower, Yvonne Buckley, Anil Kokaram

Superconductivity

Interference-controlled magnetic bound states and subgap transport in a Square–octagon superconductor

The square–octagon lattice provides a useful setting for studying magnetic adatoms because its multisublattice structure allows different adsorption geometries to probe distinct combinations of the host electronic states. We study magnetic adatoms in an $s$-wave superconducting square–octagon lattice, focusing on the interplay between adsorption geometry, narrow and flat-band states, and subgap transport. We find that the large spectral weight associated with a flat band does not necessarily imply strong impurity–host coupling. For symmetric square-hollow adsorption, destructive interference strongly suppresses the overlap with the flat-band manifold, whereas top and octagon-hollow adsorption retain substantial overlap. This selectivity is reflected in the magnetic bound states, with the square-hollow channel supporting a predominantly atomic state, while the top and octagon-hollow states are strongly hybridized with the superconducting host. The effect develops continuously as the relevant band becomes flatter and remains pronounced in a material-motivated parameter regime. In a normal–superconductor–normal geometry, these differences are also reflected in the nonlocal subgap conductance. At a representative exchange splitting, square-hollow adsorption suppresses the low-bias conductance relative to the clean device, whereas octagon-hollow adsorption enhances it, with the square-hollow response showing a much sharper dependence on exchange splitting. Our results show that the magnetic-adatom response in narrow and flat-band superconductors is governed by the available spectral weight and by the symmetry and wave-function structure of the impurity–host hybridization.
Read the full abstract

Superconductivity

Doping-controlled topological superconducting transition in misfit layer compounds

Achieving topological superconductivity is a key goal in quantum physics, offering a path to fault-tolerant quantum computers. A central challenge in this field is to continuously drive a material through a topological quantum phase transition to directly observe the evolution from trivial to topological superconductivity. However, finding a robust platform that allows such extreme and precise tuning remains a challenge. Here, we demonstrate a doping-controlled phase transition from a conventional to a topological superconducting state in the bulk misfit layer compound (LaxPb1-xSe)1.14(NbSe2)2. We reveal a non-monotonic phase diagram characterized by two distinct superconducting regimes separated by a non-superconducting phase at a precise doping. In the highly doped regime, the superconducting phase becomes remarkably sensitive to non-magnetic disorder, and orientation-selective in-gap modes emerge at atomic step edges. Supported by Bogoliubov-de Gennes calculations, these emergent spatial signatures are consistent with a time-reversal-symmetric crystalline-topological order parameter. Our results establish misfit compounds as a platform to engineering topological superconductivity.
Read the full abstract

All authors Hugo Le Du, Robin Salvatore, Justine Cordiez, Ludovica Zullo, Arindam Mukherjee, Daniel Schmieg, Dominik Volavka, Francois Debontridder, Marie Herve, Tomas Samuely, Shunsuke Sasaki, Florent Pawula, Etienne Janod, Laurent Cario, Tristan Cren

Robotics

Anti-Gravity Walking by a Flying Humanoid Robot via Thrust-Rate Input Whole-Body Model Predictive Control

Flying humanoids are expected to perform tasks in diverse environments, while their existing locomotion is mainly limited to aerial flight and ground walking. The capability to move in complex three-dimensional space can greatly expand their application range. For such walking motion on ceilings and similar anti-gravity environments, whole-body MPC is effective. However, the discontinuous changes in dynamic structure accompanying contact switching during walking can induce thrust spikes, resulting in control instability. Therefore, in this work, we propose and implement a real-time whole-body MPC framework for anti-gravity bipedal walking. First, we formulate whole-body MPC using the time derivative of thrust, namely thrust-rate, as the control input. This formulation guarantees continuity of the thrust trajectory during contact switching while preserving the sparse structure of the optimal control problem for fast computation. Second, we address the lack of natural support forces in anti-gravity environments. We introduce lower bounds on the foot-normal component of the contact force, and smoothly transfer them during the doublesupport phase. Finally, we implement the proposed framework and demonstrate anti-gravity walking by a flying humanoid through simulation and a hardware experiment. To the best of our knowledge, this is the first demonstration of multi-contact whole-body MPC for a transformable aerial robot and walking by a flying humanoid beyond the ground.
Read the full abstract

Fluid Dynamics

Marangoni flow induced by crystallization in saline droplets

In this article, we study the flow induced by crystallization of sodium chloride in sessile droplets. We measure flow velocities on the order of 1 $\mathrm{cm.s^{-1}} during crystallization, two orders of magnitude higher than those reported in previous studies. We show that this discrepancy is due to surface contaminants in earlier experiments. Numerical simulations reveal that in the clean case, the flow is driven by Marangoni stresses, whereas it is density- driven in the contaminated case. Consequently, droplet cleanliness significantly influences the flow which, in turn, determines the growth rate of crystals.
Read the full abstract

Software Engineering

"We Permit the Use of AI, but [...]": The Landscape of AI Policies in Popular Open Source Projects

Open source communities are converging on a new governance artifact: the AI contribution policy. These policies barely existed a few months ago and are now being written and adopted. We analyzed 281 AI contribution policies, and manually classified them along the six dimensions; to study how policies change, we also tracked 92 dedicated AI policy files over time. We answer four research questions on (1) AI usage allowance, (2) AI disclosure practices, (3) AI slop countermeasures, and (4) AI policy evolution. We find that, first, permission is the norm rather than the exception: 83.3% of policies permit or encourage AI in code contributions. But permission comes with conditions, as 67.3% require a high level of human involvement and 43.4% assign accountability. Second, AI disclosure is required by 48.8% of policies, most often in pull request descriptions and commit messages, but what must be disclosed varies widely. Third, we identify ten countermeasures against AI slop, targeting pull requests, users, and autonomous agents. Finally, policies are not static: half of the dedicated AI policy files have already been revised since creation. Our results give maintainers and researchers a baseline and a labeled corpus for studying the impact of AI policies.
Read the full abstract

High Energy Physics (Experiment)

Response of a nitrogen-filled spherical proportional counter to mono-energetic neutrons

Neutron spectroscopy is an invaluable tool for a wide range of scientific and industrial applications, however, current approaches suffer from limitations that restrict their field of applicability. A safe and inexpensive alternative approach to neutron detection and spectroscopy is the use of a nitrogen-filled spherical proportional counter that exploits the $^{14}\rm{N(n,p)}^{14}\rm{C}$ and $^{14}\rm{N(n},\alpha)^{11}\rm{B}$ reactions. The neutron spectroscopy capabilities of the detector are demonstrated using beams of mono-energetic neutrons. A nitrogen-filled spherical proportional counter, operating at a pressure of $1$ bar, is exposed to neutrons with energies from $0.75$ MeV to $2.75$ MeV at the Tandem accelerator of the National Centre for Scientific Research “Demokritos” in Athens. A linear energy response is observed, within the statistical precision of the measurements.
Read the full abstract

All authors T. Avgitas, M. Axiotis, Z. Balmforth, P. Gadow, I. Giomataris, D. Fassouliotis, J. Fock, P. Knights, A. Lagoyannis, I. Manthos, L. Millins, K. Nikolopoulos, I. Oceano, C. Prodromou, L. Schweder, E. Taimpiri, C. Toukmenidis, A. Ziagkova, R. Ward

Computer Vision

TeMo: Temperature Modulation for Multimodal Contrastive Learning

Contrastive learning approaches achieve strong performance by training models to bring similar samples closer while pushing dissimilar samples apart. A crucial component of contrastive learning is the temperature hyperparameter $\tau$, which controls the penalty strength applied to negative samples. However, most existing methods either fix this hyperparameter or learn a global value during training. In this paper, we introduce TeMo, Temperature Modulation framework, a similarity-based modulation approach that adaptively adjusts the temperature for each positive-negative pair according to their similarity, enabling more fine-grained multimodal contrastive learning. Our approach seamlessly integrates temperature-modulated multimodal and unimodal losses with the standard multimodal contrastive loss by gradually transitioning between them. This design allows the model to capture both coarse- and fine-grained semantics at different training stages. Extensive experiments demonstrate that each component of TeMo consistently enhances performance across diverse zero-shot retrieval and classification tasks, establishing new state-of-the-art results.
Read the full abstract

All authors Dhimitrios Duka, Bernt Schiele, Hilde Kuehne, Anna Kukleva

Quantum Physics

Discrete Dirac equation on quantum graphs: A model of a Dirac particle in a branched lattice

We study the one-dimensional time-independent discrete Dirac equation on graphs with discrete edges. Both continuous and discrete formulations of the graphs are considered, with particular focus on star graphs with three edges as a simple example. Exact solutions for eigenstates and spectra are derived and compared with the continuum case, showing good agreement in tabulated results. This approach offers a systematic framework for analyzing relativistic quantum transport on branched structures relevant to nanostructures and quantum device design.
Read the full abstract

Rings and Algebras

Three-Dimensional Geometry in Exceptional Algebra

We review some topics in "exceptional mathematics” from the perspective of 3-dimensional geometry: the octonions $\mathbb{O}$, the split octonions $\mathbb{O}'$, the bioctonions $\mathbb{O}_\mathbb{C} \cong \mathbb{C} \otimes_\mathbb{R} \mathbb{O}$, the complex Albert algebra $\mathfrak{h}_3(\mathbb{O}_\mathbb{C})$, and the complex form of the exceptional Lie algebra $\mathfrak{e}_6$. We show how to functorially build an algebra isomorphic to $\mathbb{O}$ from any 3d complex vector space equipped with an inner product and complex volume form. Similarly, we build one isomorphic to $\mathbb{O}'$ starting from a 3d real vector space equipped with a volume form, and one isomorphic to $\mathbb{O}_\mathbb{C}$ starting from a 3d complex vector space equipped with a complex volume form. We give applications to 3-dimensional real and complex manifolds. Finally, we describe how to build an Jordan algebra isomorphic to $\mathfrak{h}_3(\mathbb{O}_\mathbb{C})$ starting from three 3d complex vector spaces equipped with complex volume forms. This last construction gives a nice explicit description of the complex Lie algebra $\mathfrak{e}_6$ and its subalgebra $\mathfrak{sl}(3,\mathbb{C}) \oplus \mathfrak{sl}(3,\mathbb{C}) \oplus \mathfrak{sl}(3,\mathbb{C})$.
Read the full abstract

Quantum Physics

Toward Fault-Tolerant Variational Optimization: QAOA under [[4,2,2]] Error Detection

We present a partially fault-tolerant implementation of QAOA based on the $[[4,2,2]]$ error-detection code, targeting the Max-Cut problem on a square graph. Our main contribution is a novel ancilla-mediated logical $R_{ZZ}$ gate enabling interactions between qubits in different $[[4,2,2]]$ blocks. We evaluate unencoded and encoded circuits under five noise models, with both all-to-all and grid-routed connectivity, using the Cirq and qsimcirq frameworks with parallel CPU execution. Post-selection on stabilizer measurements consistently improves the probability of sampling optimal bitstrings, with five measurements providing the strongest benefit. These results support error-detection as a practical near-term strategy for improving the quality of variational quantum algorithms.
Read the full abstract

All authors Matteo Robert Child, Emanuele Dri, Giacomo Vitali, Chiara Vercellino, Alberto Leporati

Distributed Computing

Analytical Resource Management for Fine-grained MoE Computation-Communication Overlap

Fine-grained computation–communication overlap in distributed Mixture-of-Experts (MoE) inference allows communication to begin as partial compute results become ready. However, cooperative thread arrays (CTAs) performing computation and communication contend for finite residency capacity on streaming multiprocessors (SMs). Because a resident CTA generally retains its allocated SM resources until completion, CTAs that cannot be co-resident must wait for resources, resulting in wave-like execution. A fixed resource partition cannot adapt to changes in input size, routed expert load, and kernel configuration, potentially causing a communication backlog or reducing expert compute parallelism. We present a wave-quantized analytical model and launch-time resource manager for dependency-coupled overlap pipelines. Using routed-tile counts, kernel occupancy, GPU residency constraints, and split-level readiness dependencies, it selects the communication-CTA count and resource partition before each launch without candidate execution, per-workload profiling, or kernel recompilation. We integrate the method into the public COMET A100 implementation in FLUX. We evaluate three MoE models on four NVIDIA A100 GPUs under several parallelism strategies at the GEMM2+GatherRS operator, complete post-router MoE layer, and complete-model prefill levels. Across 15 real-p90 workloads, the analytical selector achieves 3.22 percent mean regret relative to the measured oracle with a mean solver overhead of 0.157 microseconds. Over COMET, our method achieves geometric-mean speedups of 2.528x at the GEMM2+GatherRS operator, 1.771x at the complete post-router MoE layer, and 1.185x for complete-model prefill, with maxima of 4.218x, 2.584x, and 1.439x, respectively. At every feasible TP=2/EP=2 sequence length of at least 4,096, our implementation outperforms COMET, Megatron core-TE, and FastMoE TP+NCCL.
Read the full abstract

Analysis of PDEs

On Analyticity of Solitons for the Periodic Dispersion-Managed Nonlinear Schrödinger Equation

We explore the regularity of energy maximizers for the Lagrangian of a periodic dispersion managed fiber optic at fixed intensity, on a torus of length $L$ and with vanishing average dispersion. We show that the Fourier coefficients decay at a polynomial rate, and then upgrade this to exponential decay, so that the maximizers are analytic in space for large enough $L$. In addition, by an asymptotic comparison to the optimizers on the real line, we prove that the solutions are non-trivial. We also consider a conjecture that the maximizer necessarily has an underlying symmetry inherent to both the energy functional and the resulting Euler-Lagrange equation. All of our results are supported with illustrative numerical experiments.
Read the full abstract

All authors Sebastian Herr, Dirk Hundertmark, Jeremy L. Marzuola, Tim Van Hoose

Robotics

Proprioception-Anchored Cross-Modal Pretraining for Zero-Shot Sim-to-Real Contact-Rich Assembly

Contact-rich assembly remains challenging because it requires submillimeter spatial accuracy and reliable interpretation of forces during sustained contact. Although simulation-based reinforcement learning offers a scalable training paradigm, discrepancies in visual observations, contact dynamics, and force/torque (F/T) measurements often limit policy transfer. We observe that proprioception is comparatively consistent across domains because joint positions are expressed in a shared calibrated coordinate system and joint velocities are computed consistently in simulation and on hardware. Based on this observation, we present PACE (Proprioception-Anchored Cross-Modal Encoder), which supervises temporal visual and F/T representations by predicting proprioceptive state transitions. Static domain-specific factors, including lighting, texture, and sensor bias, contain little information about joint motion; optimizing the proposed objective therefore suppresses their influence on the learned representation while retaining task-relevant motion cues. Policies trained on frozen PACE features are directly deployed on hardware without real-world fine-tuning or object-pose tracking. Across four contact-rich assembly tasks, PACE attains an average real-world success rate of 93.3% and only a 2.7-percentage-point sim-to-real drop, while remaining robust to perturbations that substantially degrade pose-based and learned-fusion baselines.
Read the full abstract

All authors Yuhan Wang, Yurou Chen, Hongye Jiang, Gaojing Zhang, Wenzhao Lian

Artificial Intelligence

Quantile-Led Feature Extraction for Multi-Horizon Predictive Maintenance in Industrial Manufacturing Systems

In data-driven predictive maintenance (PdM), feature extraction is usually treated as fixed preprocessing: a descriptor set is chosen once and reused while the downstream model or forecasting horizon changes. This paper isolates the representation-learning stage and presents a quantile-led feature-extraction framework based on a dual-stage MLP-QRNN hierarchy. QRNN1 learns a broad ten-quantile conditional distribution for each sensor channel, while skip-connected QRNN2 refines a retained mid-tail quantile set into compact, channel-resolved, distribution-aware features. A fixed thirteen-pipeline ablation spans 1-hour, 70-hour, and 30-day regimes across 72 machines in 9 industrial facilities, with the downstream temporal classifier held fixed within each regime. Increasing the retained mid-tail set from two to four quantiles improves 30- and 60-minute F1-score, reaching 75.92% and 72.44% with attention enabled. The results also show that representations do not transfer reliably beyond their design horizon unless feature capacity, temporal embedding, activation strategy, and sensor breadth are scaled with the forecasting task. The unmodified short-horizon extractor falls to 42.90% F1 at 70 hours, whereas horizon-conditioned extractors reach 60.38% at 70 hours and 79.97% at 30 days. The framework therefore supports treating PdM feature extraction as a horizon-dependent representational stage rather than fixed preprocessing.
Read the full abstract

Robotics

PhysReal: Learning Real-World Deformable Object Physics via Hybrid Constitutive Modeling

Learning physically plausible dynamics from visual observations is essential for interactive world models and embodied agents. However, modeling real-world deformable objects remains challenging because their dynamics often arise from complex, spatially heterogeneous material responses. To address this challenge, we propose PhysReal, a video-driven framework for learning and simulating the underlying physics of real deformable objects. PhysReal integrates a spatially varying hybrid expert-neural constitutive model with a differentiable MPM simulator and 3DGS renderer. Analytical expert models provide interpretable physical priors, while neural constitutive residuals capture material responses beyond predefined formulations. Spatially distributed patches parameterize the constitutive field, enabling a continuous representation of local material variations. To organize the identification of this model from sparse visual observations, we adopt a progressive curriculum that sequentially optimizes global material properties, spatially varying local parameters, and neural constitutive residuals, together with complementary motion and mask supervision. Extensive experiments on diverse deformable-object interactions demonstrate that PhysReal achieves superior performance in dynamic reconstruction and future-state prediction, while showing strong potential for downstream robotic applications.
Read the full abstract

All authors Yinan Deng, Jianqiao Song, Yisi Zhang, Yuhan Wang, Jiahui Wang, Yufeng Yue

Chemical Physics

Simulation-Supervised Foundation Models for Retention Time Prediction in High-Performance Liquid Chromatography beyond Experimental Data Coverage

Accurate prediction of high-performance liquid chromatography (HPLC) retention times (RTs) across diverse molecules and chromatographic methods remains challenging because experimental training data cover only a limited region of chemical and method spaces. Here, we develop FUSE-RT (Foundation model Unifying Simulation and Experimental supervision for Retention Time), a multitask foundation model that integrates RT data from 179 chromatographic methods and adapts to unseen molecules and methods using limited target-domain data. To extend transferability beyond experimental coverage, we introduce simulation-to-real (Sim2Real) transfer learning, in which molecular representations learned from large-scale computational data are transferred to experimental RT prediction. Specifically, we use PolyOmics, comprising 39 properties for approximately 21,400 molecules generated by molecular dynamics and density-functional theory calculations, as auxiliary supervision. We evaluate generalization under molecular, method, and joint molecular–method distribution shifts. Simulation-derived supervision substantially improves transfer beyond the experimental molecular domain, particularly under pronounced coverage gaps and few-shot adaptation. Moreover, RT-prediction error decreases systematically with increasing simulation-data size, following a significant power-law relationship. These results establish Sim2Real transfer as a scalable strategy for extending RT prediction beyond the finite coverage of experimental chromatographic data.
Read the full abstract

All authors Stephen Wu, Yufeng Han, Yasuhiro Mito, Yoshiyuki Watabe, Yoshihiro Hayashi, Hikaru Takaya, Takuya Kubo, Ryo Yoshida

Differential Geometry

Geodesics for generalised Monge-Ampère equations

Xiuxiong Chen proved the existence of $C^{1, \bar 1}$ geodesics in the space of Kähler potentials and the convexity of the $J$-functional along $C^{1, \bar 1}$ geodesics. Analogous results were proved by Collins and Yau for the hypercritical Leung–Yau–Zaslow equation. In this paper, we study a similar picture for generalised Monge–Ampère equations associated with two right-Noetherian polynomials in proper position.
Read the full abstract

Machine Learning

CoER: Defending against Adaptive Indirect Prompt Injection via Adversarial Co-Evolution and Refinement

Language-model agents are vulnerable to indirect prompt injection (IPI) during tool use: adversarial instructions hidden in untrusted tool outputs can covertly redirect legitimate task execution. Existing work often trains and evaluates defenses against fixed attacks that do not adapt to the defender's behavior, so the resulting defenses may struggle against adaptive attacks in real-world settings. We argue that a strong defense against adaptive IPI must adapt during training to a continually evolving attacker. Building on this insight, we propose CoER, a verifier-grounded co-evolution and refinement framework that models interleaved tool calls and adaptive injections within a task as a general-sum Markov game: the defender advances the task through successive tool calls, while the attacker can inject multiple times within the same task and adapt subsequent attacks to the defender's responses and prior execution traces. After initializing the attacker from successful trajectories, bilateral adversarial reinforcement learning (BA-RL) retains historical policies from both roles as opponent populations and mixes current and historical opponents, extending training beyond the latest matchup. Attackers from these populations are then reused to challenge teacher agents, and only demonstrations verified for both safety and task completion are used to fine-tune the co-evolved defender. Across seven domains and three evaluation seeds, CoER reduces adaptive attack success from 41.3% to 0.2% and raises safe task completion from 39.6% to 76.2%; external benchmarks also show improved attack resistance. Further experiments validate the effectiveness of bilateral historical-opponent mixing and population-guided refinement. Attacker analyses show that co-evolution strengthens attack capabilities and that the trained attacker uses execution feedback to adapt subsequent injections.
Read the full abstract

All authors Boyang Zhang, Qingxin Xiao, Lingwei Dang, Qingyao Wu

High Energy Astrophysical Phenomena

Mining the time axis with TRON - III. Discovery of a nulling pulsar towards the SMC with MeerKAT

We report on the discovery of a nulling pulsar, PSR J0052$-$7551, in archival MeerKAT L-band observations of the Small Magellanic Cloud (SMC). The discovery was made using the image-plane transient detection pipeline, TRON, which creates a residual image for each time sample and uses a heuristic pruning strategy to identify promising candidates in the resulting temporal image cube. The TRON pipeline triggered on the approximately minute timescale variability associated with the nulling behaviour of a canonical pulsar. Follow-up MeerKAT UHF-band beamformer observations revealed a pulsar with a spin period of 0.55 s and a measured period derivative of 6.1(6)$\times$10$^{-17}$s$\cdot$s$^{-1}$. The source is seen in projection toward the periphery of the SMC, and has a dispersion measure value of 29.24(2) pc$\cdot$cm$^{-3}$, which is on the order of the expected Milky Way contribution along the line of sight. We use a Gaussian Mixture Model to fit the nulling behaviour which yielded a nulling fraction of $\sim 32$%. We find that the null lengths are consistent with an exponential distribution indicating a stochastic process. Additionally, power spectra of the UHF visibility data reveal an unexplained quasi-periodic feature at ~5$-$7 s that cannot be attributed to the source itself, RFI, or standard instrumental effects. Following on from previous image-plane detections of pulsars by the TRON approach, which relied on either scintillation or eclipsing behaviour, this is the first detection triggered by nulling. This discovery confirms TRON's sensitivity to a wide range of astrophysical phenomena of medium timescale duration.
Read the full abstract

All authors T. Blecher, M. Geyer, O. M. Smirnov, A. Coelen, I. Heywood, I. Pelisoli, A. K. Hughes, J. S. Bright, V. Prayag, T. Myburgh, C. Tasse, B. Hugo, J. S. Kenyon, H. L. Bester, S. J. Perkins, J. Dawson, V. G. Samboco, A. J. T. Ramaila, B. Ngcebetsha, I. Sihlangu, C. Choza, V. S. Dhillon, N. Castro Segura, J. McCormac

Numerical Analysis

Analysis of a Surface Crouzeix-Raviart Element for the Stokes Problem

Recently, increasing attention has been paid to finite element discretizations of vector-valued flow problems posed on curved surfaces. In this work, we study a surface Stokes system defined on a two-dimensional manifold embedded in three-dimensional space. The surface velocity field is approximated using a nonconforming Crouzeix-Raviart finite element on a polyhedral approximation of the surface, while the pressure is discretized by piecewise constant functions. The governing equations involve the symmetric strain-rate tensor, whose approximation with Crouzeix-Raviart finite elements leads to spurious oscillations in the velocity field because the discrete Korn inequality fails on the nonconforming space. We therefore stabilize the momentum equation by adding an edge-jump penalty term, which restores the coervity of the bilinear term as the discrete Korn inequality holds for this stabilized version. Additionally, we establish that this finite element pair of velocity and pressure spaces satisfies the discrete inf-sup compatibility condition. Furthermore, we derive optimal a-priori error estimates in both the energy norm and the $L^2$-norm. These theoretical results are corroborated by numerical experiments.
Read the full abstract

Populations and Evolution

Numerical approximations of population size distributions for multi-type branching processes

Continuous-time multi-type branching processes are fundamental models for expanding and migrating populations with cancer evolution being a prototypical example. Inferring model parameters, like mutation and growth rates, from time-series count data requires efficient computation of population size distributions. Existing methods are mainly based on large-time or large-number asymptotics, which rely on either restricted initial conditions or simplified interactions between cell types. Here, we introduce two numerical approximations of population size distributions for multi-type branching processes on directed graphs with arbitrary initialization. The first approach combines a saddle-point approximation with numerical integration of the probability generating function. We characterize admissibility and establish conditions for saddle-point existence and uniqueness. For directed acyclic graphs, the second approach provides a large-time small-mutation-rate alternative based on closed-form approximate Laplace transforms and efficient numerical inversion. We benchmark the accuracy and speed of both solutions in simulations, showing substantial improvement over the state-of-the-art large-number approximation and orders of magnitude speedup over Gillespie's stochastic simulation algorithm at matching accuracy. We apply our methods to analyze the relapse dynamics of an acute myeloid leukemia patient, where rapid parameter scans over a six-type patient-specific mutation tree quantify how unobserved remission burden and treatment-altered fitness can explain relapse. Our methods provide computational building blocks for future likelihood-based inference in cancer evolution and other expanding populations.
Read the full abstract

Computer Vision

When Semantically Consistent Encoding Meets View-Label Heterogeneity Modeling: A Unified Framework for Incomplete Multi-View Multi-Label Learning

Incomplete multi-view multi-label learning requires not only robust semantic aggregation from partially observed views, but also label-aware exploitation of view-specific evidence. Existing approaches usually emphasize either shared representation learning or decision-level fusion. The former improves robustness against missing views, yet tends to compress label-discriminative view-specific cues into a single latent representation. The latter preserves individual view predictions, but often relies on fixed or globally learned fusion weights, ignoring that different labels of different instances may require different views. To address these limitations, this paper presents V2L, a unified representation-decision framework for incomplete multi-view multi-label classification. On the representation side, V2L constructs semantically consistent variational posteriors from incomplete views through a perturbation-aware encoding mechanism, which provides a stable shared semantic basis. On the decision side, V2L introduces an active view-label relevance modeling strategy that estimates instance-wise and label-wise view contributions, allowing each label prediction to adaptively select useful view-specific evidence. From the perspective of model architecture, these two important strategies are integrated into a unified framework through a hybrid fusion architecture, simultaneously meeting the requirements of cross-view semantic consistency and representational complementarity. Extensive experiments under both incomplete and complete settings show that V2L achieves leading performance on five benchmarks. Code is available at: https://github.com/justsmart/V2L.
Read the full abstract

All authors Chengliang Liu, Bo Li, Bob Zhang, Yanghao Zhou, Jie Wen, Wenwu Wang

Probability

Stationary solution for a fractional stochastic delay differential equations with multiple delays

This paper studies a linear stochastic delay differential equation driven by fractional Brownian motion and involving a finite number of discrete delays. The model combines two sources of memory: delayed feedback in the drift and temporal dependence induced by the Hurst parameter of the fractional Brownian motion. We first introduce the fundamental solution associated with the corresponding deterministic delay equation and use it to obtain an explicit representation of the solution. The stochastic convolution with respect to fractional Brownian motion is defined pathwise as a Riemann–Stieltjes integral, using the finite-variation properties of the fundamental solution. This representation allows us to compute the mean and auto-covariance function of the process. Under a delay-independent stability condition on the drift coefficients, we prove that the fundamental solution decays exponentially and derive the long-time behaviour of the stochastic delay equation. In particular, the solution converges in finite-dimensional distributions to a stationary Gaussian process, whose limiting mean and covariance are given explicitly. When the Hurst parameter satisfies \(H>1/2\), we show that the limiting covariance preserves the long-memory behaviour of the driving fractional Brownian motion and decays asymptotically as \(h^{2H-2}\). Finally, by extending the noise to a double-sided fractional Brownian motion, we construct a stationary solution of the equation and prove its uniqueness. The results provide a tractable framework for modelling stationary systems with both delay effects and fractional memory, with potential applications to stochastic volatility, energy modelling, and other time series exhibiting persistence.
Read the full abstract

High Energy Physics (Experiment)

Reducing simulation-related emissions in rare-event searches through optimised event biasing

Rare-event search experiments continue to extend their sensitivities to unprecedented levels. This requires increasingly small backgrounds, and shielding schemes that can suppress external backgrounds by several orders of magnitude. Moreover, detailed simulations are required to attain a good understanding of these experimental backgrounds. Studies have suggested, however, that simulations and computing-related tasks contribute approximately 10% to the average particle physicist's carbon footprint. With backgrounds frequently below 0.01 counts per kg of target per keV of energy, simulations require more computing resources; growing a rare-event researcher's computing-related carbon footprint. Event biasing is often applied in shielding simulations as a remedy, yet there is a lack of systematic guidance on how to maximally benefit from them. An optimisation study for the importance-splitting biasing technique is discussed, focused on balancing statistical precision with simulation CPU-time, and how this can benefit the average researcher's carbon footprint.
Read the full abstract

Category Theory

Cotorsion pairs and model structures induced by resolution dimensions relative to Frobenius pairs

We explore some methods to obtain cotorsion pairs and model category structures from a Frobenius pair. Specifically, we consider a type of Frobenius pair $(\mathcal{X},\omega)$ in an abelian category $\mathfrak{C}$ for which there exists a nonnegative integer $m$ such that every object in the right orthogonal complement $\mathcal{X}^{\perp_1}$ of $\mathcal{X}$ under $\text{Ext}^1_{\mathfrak{C}}(-,\sim)$ has a special $\omega^\wedge_m$-precover and a special $[\omega^\wedge_m]^{\perp_1}$-preenvelope, where $\omega^\wedge_m$ denotes the class of objects in $\mathfrak{C}$ with resolution dimension relative to $\omega$ at most $m$. The cotorsion pairs and model structures obtained from these Frobenius pairs will involve the classes $\mathcal{X}$, $\omega$, $\mathcal{X}^\wedge_m$, $\omega^\wedge_m$, and their orthogonal complements, in the description of its cofibrant, fibrant and trivial objects. As applications of our results, we obtain several cotorsion pairs and model structures related to relative and absolute Gorenstein homological dimensions.
Read the full abstract