Peer Revenue Now accepting positions

Papers

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

All of arXiv

Monday 9 November

Signal Processing

Near-Field Channel Estimation via Deep Unfolding for Extremely Large-Scale Antenna Arrays

The Extremely Large-scale Antenna Array (ELAA) has been envisioned as a promising candidate to meet the ambitious demands of 6G. However, this evolution also brings about two primary limitations: high computational complexity caused by large-scale arrays and inherent operation in the near-field region. To address these issues, this work exploits the sparsity of near-field channels in the polar domain and formulates channel estimation as a compressed sensing (CS) problem, thereby enabling estimation with reduced pilot overhead. In particular, we propose a guideline to design a sparsifying codebook for the near-field channel, which is strictly based on a target coherence value. Building on this, we develop a deep unfolding architecture, derived from the iterative solver ISTA, with a hybrid loss, namely Hybrid LISTA. Numerical results demonstrate that our proposed codebook, with comparable complexity, still exhibits improved estimation performance and greater coverage in the near-field region. Moreover, our unfolded model with the hybrid loss consistently delivers better recovery performance over existing estimation schemes.
Read the full abstract

All authors Nguyen The Khang, Do Hai Son, Tran Trong Duy, Trung Thanh Le, Viet-Dung Nguyen, Trong-Minh Hoang, Karim Abed-Meraim, Merouane Debbah, Nguyen Linh Trung

Computation and Language

A generalized transition framework for identifying reduction domains in non-binary constituency parsing

Direct non-binary constituency parsing must identify an ordered sequence of children at each phrase-level reduction. We extend our earlier delimiter-guided bottom-up construction to a common calculus for top-down, in-order, and bottom-up parsing. At reduction, the rightmost delimiter marks the left boundary, the nonterminal marker supplies the mother label, and the completed items in the delimited segment determine the children and arity. We use a stack invariant to prove soundness and completeness and derive the exact length of the static oracle's action sequence. A projection-and-lifting theorem shows that each delimiter-guided system and its canonical counterpart generate the same set of trees under both the structural and grammar-licensed transition relations. In controlled experiments on PTB and CTB, the delimiter-guided parsers achieve similar labeled accuracy to their canonical counterparts across all traversal orders. For bottom-up parsing, delimiter guidance reduces the number of action classes and increases the number of supervised decisions per tree.
Read the full abstract

Software Engineering

Feeling Easier, Working Harder: Cognitive Load in AI-Assisted Code Comprehension

Software developers spend significant time reading and understanding code, increasingly supplemented by generative AI. While these tools aim to reduce the difficulty of reading code, interacting with them introduces a dual-task environment that may impose extraneous cognitive load that is challenging to capture. This study investigates how access to generative AI affects developers' real-time cognitive workload, code comprehension performance, and self-perceived code comprehension strategies. We conducted a 3x2 within-subjects study with 20 novice CS students reviewing Java code snippets of varying lengths with or without AI assistance, followed by an AI-unassisted comprehension quiz. Cognitive workload was continuously captured using a Detection-Response Task (tDRT) during both reading and quiz phases, and compared to post-quiz NASA-TLX and SEQ scores. AI assistance significantly increased objective cognitive workload, raising tDRT reaction times by 11.2%, yet participants' perceived workload remained unchanged. Code comprehension scores were high under both conditions, with no evidence that AI assistance improved comprehension at the difficulty levels evaluated; a ceiling effect limited power to detect benefits on harder tasks. Qualitatively, participants reported trusting the AI conditionally -verifying its output against their own reading for more complex snippets, but directly trusting it for simpler code- and, while many felt their reading process was unchanged, a notable subset shifted toward skimming or reading the AI's explanations in place of the code itself. These findings reveal that developers do not accurately perceive the cognitive resources they expend when managing AI tools. This study provides evidence of added cognitive effort associated with using generative AI, pointing towards a need to account for this dual-task penalty in software engineering tools and education.
Read the full abstract

All authors Maryangela White, Maximilian Rose, Matthew Buckman, Amy McDonnell, Francisco Ortega, Bianca Trinkenreich

Chemical Physics

Extending the reach of real-space real-time TDDFT: new developments in the Octopus code

Octopus is an open-source package implementing time-dependent density functional theory in a real-space, real-time formulation. Here we report on the developments since our previous contribution to the Special Topic Issue on Electronic Structure Software [J. Chem. Phys. 152, 124119 (2020)]. New exchange-correlation functionals extend the range of systems that can be treated, while advances in relativistic and all-electron methods open the door to core-electron spectroscopy. New capabilities for lattice dynamics - stress-tensor calculation, geometry optimization, support for machine-learned force fields, and phonon calculations - enable a fully ab initio description of light-induced lattice dynamics in real time. We also present developments that move beyond the dipole approximation, capturing light-matter interaction more faithfully. Improved methods for exciton dynamics complement a set of algorithmic advances - new propagation schemes, imaginary-time evolution, and a refactored grid handling more than a billion points - that together substantially improve performance and scalability. Combined with technical improvements such as AMD GPU support and a modernized build system, these developments consolidate Octopus as a versatile, high-performance platform for exploring electronic excitations, light-matter interaction, and nonequilibrium dynamics in molecules, nanostructures, and extended systems.
Read the full abstract

All authors Nicolas Tancogne-Dejean, Martin Lüders, Esra Ilke Albar, Heiko Appel, Franco P Bonafé, Alexander Buccheri, Alessandro Casalino, Benshu Fan, Hans Fangohr, Umberto De Giovannini, Henning Glawe, Hannes Hübener, Simon V. B. Jensen, Christian Jöns, Lukas Konecny, Martin Lang, Ruipeng Li, I-Te Lu, Jehferson Mello, Henri Menke, Sumit Naskar, Zhengwei Nie, Sebastian T. Ohlmann, Micael J. T. Oliveira, Subhojit Pal, Jonas Reimann, Rajarshi Sinha-Roy, Jiuyu Sun, Francesco Troisi, Angel Rubio

Robotics

Grasp-Taxonomy Conditioning as a Mode Switch for Dexterous Grasping of Climbing Holds

Dexterous grasping in robotics relies on distinct hand configurations depending on object geometry. We introduce BetaGrasp, a 3D diffusion policy that conditions a DP3-style point-cloud policy on a four-class grasp label (crimp, sloper, pinch, and jug), running on a Franka arm with a 16-DoF LEAP hand. We also introduce a climbing-hold benchmark: a spring testbed whose ratchet gives a slip-force reading in newtons. In our analysis the policy's point-cloud encoder did not separate the four holds, so the scan alone gave it little to choose a grasp from. Without the label (a DP3-style baseline), simulated rollouts from the robot's reset pose all end in a jug grasp, whatever the hold (25% correct grasp type, versus 94% for BetaGrasp). The label works as a switch for which grasp the policy performs. On 80 paired hardware trials, BetaGrasp raises average slip force from 9.6 N to 14.3 N, and grasps slipping fully off the hold drop from 44% of trials to 10%.
Read the full abstract

Machine Learning

Speculative Self-Distillation enables Efficient Knowledge Internalization

Many real-world applications require language models to internalize knowledge that was absent from pretraining, to incorporate continually updated facts, or to learn new domain skills. Self-distillation has emerged as a natural approach for this task. A model conditioned on the source material acts as the teacher, and an unconditioned student is trained to reproduce the teacher's behavior closed-book. We propose Speculative Self-Distillation (SSD), an efficient per-token mixed-policy method for self-distillation. The student generates each rollout by default, while the teacher intervenes only at positions where its next-token distribution substantially diverges from the student's. Unlike off-policy distillation with teacher-generated prefixes, this design keeps training rollouts close to the student's test-time behavior, while avoiding the low-signal states that fully on-policy rollouts drift into. In effect, SSD lets the teacher steer the rollout exactly where the privileged context changes the prediction. On three closed-book QA benchmarks measuring fact recall, knowledge updates, and skill acquisition, SSD nearly matches on-policy distillation while using 38-69% fewer supervised tokens and consistently outperforms off-policy distillation on the held-out test set. On a forgetting stress test, SSD retains the most general capability while also achieving the best in-domain accuracy and using less supervision than the on-policy baseline. Overall, SSD establishes a new accuracy-efficiency-forgetting Pareto frontier for self-distillation.
Read the full abstract

All authors Shayan Talaei, Agam Bhatia, Arshia Soltani Moakhar, Jonas Hübotter, Amin Saberi, Azalia Mirhoseini

Dynamical Systems

A Poisson-bracket operator for Hamiltonian separability: intrinsic geometry and exact defects

Additive separation of the Hamilton-Jacobi equation depends on the Hamiltonian and on the canonical variables, and this paper introduces a five-input Poisson-bracket operator whose values on canonical pairs reproduce the Levi-Civita separability expressions. On a symplectic product structure its off-diagonal values are intrinsic, since they equal a symplectic pairing of the sector Hamiltonian fields and measure the Frobenius obstruction of a Lagrangian distribution, and their vanishing yields a local separation theorem. Because the operator is cubic in the Hamiltonian, its perturbation expansion is exact and defines a linearised separability operator and a defect functional. For Hamiltonians with a cyclic angle, the linearised operator divided by the squared rotor frequency is an angle derivative of the Melnikov integrand, so that the defect integrated along a homoclinic orbit decides the first-order splitting of the separatrix. Closed-form defects are obtained for the polar-nodal normal forms of satellite theory and, through a covariance of the operator on level sets, for Kerr geodesics. Used for coordinate selection, the defect singles out the separating focal distance of the Euler two-centre problem as its unique nondegenerate minimum, and for axisymmetric potentials it shows that the prescriptions of Sanders and Vasiliev are two weightings of the same Stäckel residual. The residual selector and Binney's shell-orbit fit improve both actions over the median prescription for 48 and 50 of 50 orbits in a flattened logarithmic potential, whereas the operator selector lowers the 90th-percentile action spread of 50 orbits in a Miyamoto-Nagai potential from 2.90% to 0.91%.
Read the full abstract

Artificial Intelligence

Benchmarking language models and agents for scientific reasoning in evolving materials laboratories

Large language models and the multi-agent systems built around them could accelerate materials discovery by connecting scientific knowledge with experimental decisions. Existing benchmarks establish capabilities in domain knowledge, multimodal interpretation and scientific tool use. As language models increasingly support materials research workflows, their ability to distinguish closely related material states and reason accurately with quantitative evidence remains insufficiently evaluated. These abilities are essential for comparing material responses, assessing mechanisms and maintaining reliable judgements as experimental evidence accumulates. We introduce MatScope to evaluate three connected capabilities. M1 tests unaided, condition-specific reasoning through 200 tasks spanning ten research functions, using aluminium and high-entropy alloys as representative material systems. M2 tests agent-assisted transfer of related knowledge to unresolved materials questions. MatEvoBench-20 tests initial forecasts across scientific dependency chains and revisions of still-hidden responses as observations accumulate. Across eleven model deployments, including GPT-6 Astra and Claude Fable 5.1, given-input calculations and many physical explanations are handled well. Fine-grained response estimates, comparisons between similar states and quantitatively supported choices remain substantially weaker. Related literature improves selected estimates, while condition and baseline mismatches limit transfer. Progressive observations calibrate some downstream responses and destabilize earlier accurate predictions. MatScope establishes measurable targets for developing reliable scientific reasoning in future self-evolving materials laboratories and materials-aware world models.
Read the full abstract

All authors Xuyu Dong, Yixuan Li, Jianguo Huang, Feng Chen, Yiyun Zhou, Bo An, Zheng Liu, Da Shu

Artificial Intelligence

Enhancing Multi-Step Chemical Mechanism Prediction with Graph-Guided Preference Data

Predicting multi-step chemical reaction mechanisms requires identifying plausible intermediates and correctly ordering the elementary reaction steps. Existing computational methods focus on single-step elementary reaction prediction, limiting their ability to generalize to unseen reaction spaces. We introduce a graph-guided framework for multi-step chemical mechanism prediction that leverages a chemical mechanism knowledge graph (ChemMechKG) to guide large language models (LLMs). To improve pathway prediction, we develop a graph-aligned reaction mechanism predictor (GRMP) that uses graph neural networks to generate synthetic candidate pathways for pathway optimization. During inference, we design a relation retrieval (RR) module to extract relevant graph evidence from the KG for pathway refinement. Experiments show that GRMP achieves the highest pathway accuracy among all compared methods on ChemMechQA and oMe-Gold, reaching 48.1% and 57.9%, respectively. On a fixed sample of 110 out-of-distribution reactions from ChemMechQA-Hard, GRMP achieves 36.4% top-1 pathway accuracy. LLM-as-Judge and human assessment further demonstrate its ability to generalize to unseen reactions.
Read the full abstract

All authors Yu Zhang, Ruijie Yu, Xuelin Zhao, Jinming Yang, Feng Zhu, Xiaokang Yang, Yaohui Jin, Yanyan Xu

Computation and Language

Robust Domain-Mixture Scheduling for Multi-Teacher On-Policy Distillation

Multi-teacher on-policy distillation (MOPD) merges the capabilities of multiple domain-expert teachers, such as mathematical reasoning, code generation, and instruction following, into a single compact student. Since all domains share a limited training budget, how to allocate it across domains directly determines the quality of distillation. Existing methods allocate the budget by each domain's learning progress, but offer no guarantee for lagging domains. We therefore introduce distributionally robust optimization (DRO) into MOPD to prioritize the worst-performing domain. However, the per-domain training signals are estimated from a few rollouts and are inherently noisy, so directly applying DRO mistakes noise for the worst case, lets occasional spikes hijack the budget, and protects inherently hard domains rather than truly regressing ones. To address these issues, we propose RDS-MOPD, which keeps the DRO objective and redesigns its three parts: it shifts the allocation only when differences clearly exceed the noise, bounds the influence of any single anomaly, and defines the worst case as the regression from each domain's own best level, all with theoretical guarantees. Experiments show that RDS-MOPD outperforms recent competing methods and ranks first on most mathematical reasoning, code generation, and instruction-following benchmarks.
Read the full abstract

All authors Yankai Chen, Sida Lin, Yifan Li, Hanjun Luo, Hanrong Zhang, Zhuohan Xie, Bowei He, Wei-Chieh Huang, Meng Ding, Wei Xue, Yike Guo, Philip S. Yu, Xue Liu

Computer Vision

HIST3R: Decomposing and Rectifying State Updates with History for Streaming 3D Reconstruction

Streaming 3D reconstruction with recurrent states enables long-video scene understanding under bounded memory, but turns reconstruction into a state-evolution problem: each incoming frame updates a compact scene memory on which all subsequent predictions depend, so unreliable updates overwrite historical context and accumulate into long-horizon geometric drift. We analyze CUT3R-style recurrent updates through a direction–magnitude decomposition, in which the magnitude sets the update gain, i.e., how strongly an observation is written into the state, and the direction encodes the geometric intent, i.e., where the update drives the state. This decomposition provides a unified view of CUT3R and its variants, revealing that existing training-free extensions perform magnitude-only control: they can attenuate an unreliable update but cannot correct its direction. We propose HIST3R, a training-free framework built on CUT3R that controls both components using history. HIST3R rectifies the update direction using locally consistent historical states, and modulates the update magnitude according to the visual discrepancy between the current observation and its history, suppressing redundant updates while preserving complementary evidence. Across camera pose estimation, 3D reconstruction, and video depth estimation, HIST3R consistently improves long-horizon performance while retaining the memory footprint and inference speed of CUT3R. With 1000 input frames, it reduces ATE by 23.87% on ScanNet and 33.65% on TUM-Dynamics relative to the strongest competing method; with 500 frames, it improves median reconstruction accuracy over the strongest CUT3R-based variant by 56.52% on 7-Scenes and 33.96% on NRGBD.
Read the full abstract

Machine Learning

Scalable Bayesian Optimization for Efficient Exploration of Moire Patterns

Large materials design spaces often contain rare high-performing configurations that are difficult to identify through trial-and-error exploration or purely representative machine-learning (ML) sampling. In this work, we present FruBO, a flexible Bayesian Optimization framework for accelerating the discovery of optimal configurations in computationally demanding materials design problems. As a case study, we investigate multilayer Moiré patterns generated by the superposition of six rotated hexagonal lattices, where the objective is to identify the rotational conformation that maximizes pore-size uniformity, quantified using the Jain fairness index. The resulting benchmark contains 600,000 unique configurations and exhibits a highly skewed objective distribution, making the global optimum difficult to locate by random sampling. We compare random exploration, Gaussian Process-based Bayesian Optimization, and Bayesian Optimization using a lightweight CatBoost surrogate with committee-based uncertainty estimation. Random sampling identifies the global optimum only after 533,012 evaluations, requiring approximately 31 hours, while Gaussian Process-based BO becomes impractical due to GPU memory exhaustion after 290 acquired samples. In contrast, FruBO with CatBoost identifies the global optimum after 1,470 acquired samples and approximately 1.15 hours. These results demonstrate that scalable surrogate models can substantially improve the feasibility and efficiency of BO in large nanomaterials design spaces.
Read the full abstract

All authors Panagiotis Krokidas, Christos Lazaridis, Vassilis Sioros, Efi-Maria Papia, Vassilios Constantoudis, George Giannakopoulos

Group Theory

Integrable Measure Equivalence and Gromov's conjecture on nilpotent groups

In this article we prove a well-known conjecture due to M. Gromov that if two infinite finitely generated torsion free nilpotent groups $\Gamma$ and $\Lambda$ are Quasi-isometric, then their Mal'cev Completions are isomorphic as Lie groups. We prove this result in more general set-up. We prove that if two infinite finitely generated torsion free nilpotent groups $\Gamma$ and $\Lambda$ are Integrable Measure Equivalent, then the Mal'cev completions of $\Gamma$ and $\Lambda$ are isomorphic as Lie groups. Equivalently, using Y. Shalom's result we prove that two infinite countable torsion free nilpotent groups $\Gamma$ and $\Lambda$ are Integrable Measure Equivalent if and only if $\Gamma$ and $\Lambda$ are quasi-isometric if and only if the Mal'cev Completions of $\Gamma$ and $\Lambda$ are isomorphic as Lie groups. Moreover, we discuss some applications of our main theorem to lifting of quasi-isometric embedding from $\Gamma$ to $\Lambda$ to a homo-morphism between their Mal'cev completions, and betti numbers of nilpotent groups.
Read the full abstract

Commutative Algebra

Superlinear regularity of Frobenius powers on a normal hypersurface

We give answers to four questions on linear bounds for Frobenius powers. Let $\mathfrak m$ denote the homogeneous maximal ideal. We construct a height-two prime $I$ in a geometrically normal four-dimensional graded hypersurface $R$ of characteristic two such that, for every power $q\geq16$ of four, the Loewy length of $H^0_{\mathfrak m}(R/I^{[q]})$ is at least $q^{3/2}/8-1$. Thus no fixed integer $C$ gives annihilation by $\mathfrak m^{Cq}$ for all $q$, as required by the Hochster–Huneke condition $(\mathrm{LC})$. The same construction gives superlinear lower bounds for the Castelnuovo–Mumford regularity in Katzman's question, for the exponents of the embedded primary components in the question of Smith and Swanson, and for the degree in the least variable in Katzman's Gröbner basis conjecture. The bounds follow from the degree of the kernel line bundle of a multiplication pencil associated to a family of plane quartics.
Read the full abstract

Fluid Dynamics

Phase-resolved turbulence dynamics in cardiac-cycle-inspired pulsatile pipe flow

Direct numerical simulations examine the phase-dependent turbulence dynamics of pipe flow driven by cardiac-cycle-inspired waveforms. Four cases with and without a low-flow plateau span two Reynolds-number ranges. The response follows four stages of turbulent–turbulent reorganisation. Strong acceleration suppresses transverse fluctuations and vortical activity. During acceleration and early deceleration, streamwise-normal-stress production recovers ahead of pressure–strain redistribution and radial momentum transport, allowing energy to accumulate in elongated streaks. In early deceleration, near-wall high-speed streaks develop predominantly varicose disturbances associated with Q4 sweeps, whereas outer-buffer-layer low-speed streaks develop predominantly sinuous disturbances associated with Q2 ejections. The transition exhibits distinct radial sequences: Q2-based local onset occurs progressively later towards the wall, whereas Q4-based onset is nearly simultaneous across the viscous sublayer and buffer layer, with peaks occurring successively outwards. The two responses overlap between the drag minimum and recovery maximum as redistribution intensifies markedly and multiscale momentum transport recovers rapidly. Turbulence then decays and its scales readjust. Where present, the plateau prolongs this decay and weakens the turbulence inherited by the next acceleration stage. It further delays local instability and lengthens the onset-to-peak intervals of the shear contributions. An unsteady friction decomposition links these phase-dependent dynamics to the drag response, highlighting the role of delayed turbulent shear recovery in sustaining the low-drag interval.
Read the full abstract

Sound

Constraining Evidence Strength in Bayesian Tonal-State Inference: A Posterior-Continuity Diagnostic

Effective pitch-evidence strength governs the concentration of tonal-state posteriors. Expert local-key annotations cannot calibrate this strength because they do not specify the posterior distribution over the 24 candidate keys. This study constrains evidence strength through tonal continuity, quantified by a measure $\lambda$ derived from adjacent-window posterior overlap. Passages with $\lambda$ > 0.50 exhibit relatively strong tonal continuity. At $\lambda$ = 0.50, carrying forward the preceding tonal belief is predictively equivalent to starting afresh with a uniform, key-neutral prior. Across 26 subcorpora of the Distant Listening Corpus, median $\lambda$ remains above 0.50 for changes from major keys to IV, V, and vi throughout the examined range of evidence strengths, whereas changes to i, ii, and iii cross this threshold as evidence strength increases. Because these latter three keys share the departed key's tonic or differ from it by a single key-signature accidental, such changes are expected to retain $\lambda$ > 0.50 in a substantial proportion of cases. An event share G therefore assesses how often this expectation holds as tonal selectivity increases in 4-bar windows. The resulting high-G region identifies evidence-strength settings that may be appropriate for tonal-state inference. When $\lambda$ is extended to cross-scale compatibility, the Presto finale of Mozart's Piano Sonata K. 310 shows that, within this region, a 4-bar C-minor interpretation remains compatible with its 16-bar C-major context. The findings establish tonal continuity as a music-theoretical constraint on evidence strength and illustrate how local and prevailing-key interpretations can coexist when pitch evidence is integrated over nested spans.
Read the full abstract

Statistics Applications

Short-term associations between wild-bird and mammalian high pathogenic avian influenza detection in the United States

High pathogenic avian influenza (HPAI) detections in mammals have raised concern about repeated wildlife spillover and the targeting of mammalian surveillance. We evaluated whether wild-bird detection records contained short-term explanatory and predictive information about mammalian detections in the United States. Public surveillance records from April 2022 to July 2026 comprised 18,523 wild-bird and 815 mammalian detections. Monthly state-level mammalian counts were analysed using negative-binomial mixed models adjusted for seasonality, preceding mammalian detections and state-level heterogeneity. The selected model contained contemporaneous and one-month-lagged same-state wild-bird counts. A doubling of one plus the contemporaneous wild-bird count was associated with a 57.9% increase in the expected mammalian count (95% CI 43.5-73.8%), and a doubling of the preceding-month count with a 15.9% increase (5.2-27.8%). Bordering-state exposure did not improve fit. In a six-month temporal holdout, adding same-state wild-bird exposure reduced mean absolute error by 23.8% and root mean squared error by 27.9%, although total detections remained overpredicted. After exact-row deduplication, the corresponding effects were 59.7% and 13.3%. Wild-bird surveillance records contained short-term, geographically local information about mammalian HPAI detection activity, but the associations do not establish direct transmission and require further calibration before operational forecasting.
Read the full abstract

Statistical Methodology

Conformal Prediction for Multi-Step Forecasting: Conditional Coverage and Sharp Score Calibration Rates

Uncertainty quantification in prediction is a central problem in statistics and machine learning. Conformal prediction provides a flexible approach to constructing prediction sets with valid marginal coverage under exchangeability. In this paper, we investigate conformal prediction for both one-step and multi-step forecasting from a single stationary time series, addressing temporal dependence and the additional challenges of joint prediction across multiple horizons. We propose a kernel-based framework, called Time series Residual Analysis for Calibration with Kernels (TRACK), for estimating conditional quantiles of residual scores and constructing prediction sets from rolling-window forecasters. The framework accommodates heterogeneous uncertainty across horizons and accounts for dependence induced by rolling estimation, while allowing predictor memory and forecast horizons to grow with the sample size. Our finite-sample bounds quantify how temporal dependence and overlapping score windows affect calibration and coverage. We further derive bounds for conditional quantile estimation and pointwise conditional coverage discrepancy. The guarantees extend to more general predictors that admit an accurate rolling-window approximation. We also derive matching minimax bounds over classes of known score maps, clarifying how overlapping score windows affect calibration accuracy. Experimental results from simulations and applications to wind energy generation and bike sharing demand show that our methods improve regional conditional coverage while maintaining competitive interval widths.
Read the full abstract

Probability

Anomalous regularity and scaling limit by rough transport noise for stochastic 2D fluid models on the torus

We consider the stochastic 2D Euler and mSQG equations on the torus, with $L^2$ initial data and driven by rough Kraichnan transport noise, for which we construct weak solutions enjoying anomalous regularity and dissipation properties. Moreover, we show that weak solutions in this regularity class are pathwise unique, thus establishing probabilistically strong well-posedness. We further obtain the scaling limit from these equations to the deterministic Navier–Stokes/mSQG equations, with respect to positive Sobolev norms. The proofs rest on the analysis of energy spectrum of the vorticity and on a uniform weighted discrete Poincaré inequality on the lattice, the latter being crucial for making the positive norms accessible.
Read the full abstract

Chemical Physics

SATE: Transferable Spin-Adapted Neural Wavefunctions

Accurate prediction of molecular reactivity requires both accurate and computationally efficient description of electron correlation and spin across various molecular geometries and electronic states. We introduce SATE, a spin-adapted transferable neural wavefunction architecture with effective core potentials. SATE combines a transferable neural wavefunction, the spin-adapted antisymmetrization method, and effective core potentials into a single foundation model architecture pretrained on compositions containing selected elements from hydrogen to bromine. After pretraining and adaptation, our model is able to predict 3d transition-metal spin-state gaps, bond dissociation energies, and reaction energies by addressing two key issues. First, including the core potentials into the model allows us to reduce the number of explicitly treated electrons. Second, using spin-adapted antisymmetrization enforces the prescribed total spin by construction. By testing our model with 3d transition metal chemistry, we evaluate our model's description of strong electron correlation and competing electronic states in such compounds. We find that pretraining improves most tested component energies at matched adaptation budgets, while improvements in spin-state gaps, dissociation energies, and reaction energies remain system dependent. Ultimately, our work lays the foundation for more accurate and transferable modeling of molecular systems by combining the computational efficiency of effective core potentials with explicit control over total spin.
Read the full abstract

All authors Timothy Sereda, Pallas Chou, Morris Hsieh, Emerson Hsieh

Optimization and Control

Online Digital Twins: Data Assimilation and Control

The online assimilation of data and the application of controls are essential ingredients of the digital twin paradigm. In this paper, we formalize the online interaction between the physical process and its virtual digital representation in the setting of continuously arriving data from the physical process and a receding horizon virtual planning process, which provides the control for the physical process. Our mathematical formulation relies on a mean field representation of the data assimilation process as well as a mean field Pontryagin minimum principle for the optimal control problem. Such mean field formulations are computationally attractive because they allow for simultaneous implementation in terms of Hamiltonian interacting particle systems. A more detailed discussion is included for the ensemble Kalman-Bucy filter in combination with affine approximations of the receding horizon control laws. This paper sets out the mathematical framework, while a computational study is left for future research.
Read the full abstract

Machine Learning

HyPLM Technical Report

Protein language models (PLMs) are standard tools for variant-effect prediction, protein annotation and design, but the compute needed to train and serve them keeps growing. We present HyPLM-Flash, an efficient and open PLM trained by Hyra 2.0, the research agent developed by our team. HyPLM-Flash is a 650-million-parameter masked language model trained on 726 billion tokens, 0.73 times the data of ESM-2 650M and a fifth to an eighth of the data of E1-600M and ESM C 600M. It runs on an inference engine with fused kernels and INT8 matrix multiplication, embedding about 130,000 residues per second on a single NVIDIA H20 GPU, 1.9 to 2.7 times faster than other 600M-class encoders. In a matched evaluation against 28 public checkpoints on 32 tasks, HyPLM-Flash reaches a ProteinGym zero-shot Spearman correlation of 0.409, statistically tied with ESM-2 650M, ESM C 600M and E1-600M, and the highest organismal-fitness score among self-supervised single-sequence models. It gives the best long-range contact precision of all models evaluated, including ESM-2 15B and ESM C 6B, and a lower masked-residue loss than ESM-2 650M, ESM C 600M and E1-600M on held-out sequences and on recently released PDB chains. Matching state-of-the-art accuracy at a fraction of the training and inference cost, HyPLM-Flash sets a new Pareto frontier for protein language models. We have released the model weights and the inference engine at https://github.com/Tencent-Hunyuan/hyra-results/tree/main.
Read the full abstract

Plasma Physics

A Mathematical Multiscale Framework for Nonlinear Modeling of the Heliosphere

The inner heliosphere is a highly nonlinear continuous medium where macroscopic ideal magnetohydrodynamics is inextricably coupled with microscopic, fully developed stochastic turbulence. State-of-the-art global 3D numerical solvers, while proficient at capturing the large-scale propagation of extreme space physics phenomena such as coronal mass ejections and corotating interaction regions, are computationally constrained by grid resolution and fundamental partial differential equation formulations. Consequently, they inherently do not resolve the local small-scale turbulent cascade, magnetic intermittency, and localized thermodynamic anomalies. To bridge this fundamental scaleseparation gap, this paper presents a novel mathematical and numerical framework that harmoniously integrates global heliospheric structures with local turbulent physics. The framework generates a highly realistic synthetic continuous medium by superimposing a stochastically driven nonlinear turbulent cascade onto a bimodal deterministic background flow. This cascade reproduces theoretical constraints for fractional power spectral densities, magnetic compressibility, and small-scale intermittency via a discrete multiplicative binomial cascade (p-model). This multiscale continuous background is dynamically perturbed by analytical models of large-scale nonlinear transients, including kinematic shock steepening, continuous current sheet topological reversals, and solar ejecta formalized as asymmetric force-free magnetic flux ropes. By generating exact spatio-temporal arrays across arbitrary distributed coordinates, this unified mathematical model overcomes the spatial degeneracy of single-point observations and provides a computationally efficient analytical baseline for performance models supporting future in-situ multi-spacecraft heliospheric missions, which are currently in their feasibility study phase.
Read the full abstract

Machine Learning

Reward-Estimated Hypergradient for Bilevel Reinforcement Learning with Black-Box Follower

Hypergradient (HG) methods for bilevel reinforcement learning (RL) typically assume a white-box follower whose reward is known. In practice, however, the leader rarely knows the follower's reward function and can only observe its trajectories. We address this black-box setting and propose Reward-Estimated Hypergradients (RE-HG), which recovers the follower's reward from these trajectories via inverse RL (IRL). Yet IRL identifies the reward only up to potential-based shaping, a transformation that leaves the follower's behavior unchanged but biases the HG. We prove that, for linear rewards, this bias cancels when IRL estimates are aggregated across sufficiently diverse dynamics, which the leader's own updates naturally supply. To handle finite samples, RE-HG truncates small eigenvalues of the aggregated feature covariance, restoring the identifiable subspace and stabilizing the estimation. Experiments show that RE-HG estimates gradients aligned with a white-box Oracle and achieves comparable performance; empirically, the truncation also suppresses HG explosions, letting RE-HG avoid local optima that trap the Oracle in sharp reward landscapes.
Read the full abstract

All authors Shigeki Kusaka, Mikoto Kudo, Takumi Tanabe, Akifumi Wachi, Youhei Akimoto

Machine Learning

HyInv Technical Report

Inverse folding designs amino-acid sequences for a given protein backbone, and screening needs many candidates that are both diverse and well folded. We present HyInv, a 1.96-million-parameter backbone-conditioned graph network with an encoder-decoder architecture, trained by Hyra 2.0, the research agent developed by our team: pretrained with random-order autoregressive sequence modeling, followed by structure-reward reinforcement learning. At inference, a grouped draft-and-rewrite sampler that uses separate draft and rewrite temperatures improves refolding TM-score at matched diversity. An optimized implementation with BF16 numerics designs 128 sequences for a 402-residue backbone in 50.6 ms on a single NVIDIA H20 GPU, about 52 times faster than the reference sampler, and an exact FP32 path that returns the reference sequences is 8.7 times faster. Across five benchmark datasets and seven dataset-predictor combinations, HyInv reaches a higher average scTM than all ten external baselines at comparable or higher sequence diversity, with large margins over ProteinMPNN and ProteinZero V2. The same default setting supports de novo backbones, pocket-constrained libraries, enzyme redesign with the catalytic site kept, and loop redesign. We have released our code and model at https://github.com/Tencent-Hunyuan/Hyra-results/tree/main/AI4Science/hyinv.
Read the full abstract

Robotics

OASIS: Optimization-Aware Subgoal Inference with Safety Constraints for Autonomous Mobile Robot Navigation

Navigation in dynamic environments requires mobile robots to make progress toward their goals while respecting motion limits and avoiding surrounding obstacles. Learned subgoal policies can guide model predictive control beyond its local prediction horizon, but a geometrically suitable subgoal does not necessarily lead to a successfully solved control problem. This paper proposes OASIS, a hierarchical navigation framework that incorporates recent optimization outcomes into reinforcement learning-based subgoal generation. A high-level policy trained using Soft Actor–Critic generates local subgoals, while a nonlinear model predictive controller tracks them subject to motion limits, geometric collision-avoidance constraints, and embedded high-order control barrier function constraints. The outcomes of controller optimizations over each subgoal interval are summarized and provided to the policy at its next decision, without requiring additional optimization calls. The framework is evaluated in five navigation scenarios with static and moving obstacles, including dense interactions and sensing uncertainty. Comparisons with hierarchical and direct-goal control variants show that OASIS achieves the highest aggregate navigation success. Matched comparisons across three training seeds further show a consistent success advantage over the otherwise identical hierarchy without optimization-status input. The selected-policy comparison also shows fewer primary optimization failures with similar mean controller time.
Read the full abstract

All authors Arian Hajizadeh, Javad Pourrafi Arabani, Negar Shenavar, Sana Bashiri, Seyed Majid Esmailzadeh

Artificial Intelligence

Infrathon: Benchmarking Language Agents for Real-World, Long-Horizon AI Infra Optimization

In this work, we study whether frontier agents can continually improve real-world AI infrastructure over long horizons. We introduce Infrathon, a benchmark of diverse, long-horizon AI infrastructure optimization tasks across different levels. We evaluate six models across 150 runs with 24-hour budgets, totaling over 10,000 GPU-hours. Our results show that model rankings vary with the evaluation horizon, while 135 of the 150 runs achieve further improvements during the second half of their budgets. These findings indicate that shorter evaluations can overlook both optimization gains and changes in relative model performance. We also find agents can discover solutions that outperform human-designed baselines: Claude Opus 5.5 increases the mean KV-cache compression ratio from $6.82\times$ with the human-designed Quant-VideoGen codec to $19.86\times$. Analysis of agent trajectories identifies task violations, unreliable testing, and premature completion as barriers to sustained improvement. Together, our Infrathon framework provides resources for constructing long-horizon tasks, evaluating frontier agents, and discovering good infrastructure solutions. Code is available at https://github.com/InfrathonX/Infrathon
Read the full abstract

All authors Zhengyu Hu, Zheyuan Xiao, Yike Wang, Hamish Ivison, Haoyu Wang, Yige Yuan, Rulin Shao, Yuetai Li, Niqi Lyu, Yicheng Ding, Shu Yang, Zhichen Zeng, Banghua Zhu, Jiaxin Pei, Kaize Ding, Pradeep Dasigi, Dan Roth, Hannaneh Hajishirzi, Noah A. Smith, Teng Xiao, Radha Poovendran

Computation and Language

EmoHarness: Boosting Multimodal Emotion Understanding with Meta-Harness Evolving

Small multimodal models can misjudge emotion when words, voice, and visual cues conflict. We present EmoHarness, which runs such a model, the frozen student, inside a harness: the prompts, settings, and verification code around its calls. A human-written starting harness, the scaffold, numbers the student's visual observations, transcript, and prosody descriptions as citable evidence and has the student weigh candidate emotions against it. In Meta-Harness Evolving (MHE), a stronger LLM, the Builder, reads the student's errors and execution traces offline and edits the harness; static checks and validation gates decide which edits are kept, and the Builder never sees evaluation items. On 5,507 MME-Emotion items, EmoHarness raises task-macro Recognition by 7.6-13.4 points over a single video-only call for all four students and both Builders. Transcripts and prosody descriptions supply most of this boost (+5.38 to +10.71); the scaffold changes the audio-augmented call by -0.79 to +2.54, partly through extra multi-label predictions, and evolved edits add at most +0.95, although no evolved increment survives Holm correction. The harness's emotion context also transfers to IntentQA, with small gains for three of four students, none of which survives correction. EmoHarness thus boosts small multimodal students chiefly by supplying and organizing audio-derived evidence.
Read the full abstract

All authors Fenghai Li, Bingxuan Li, Jiawen Zhang, Yifan Shen, Jiateng Liu, Yutong Zhang, Yue Guo

Probability

Pairwise correlation bounds for the random-cluster model with cluster weight at most one

We establish quantitative bounds on pairwise edge correlations for the random-cluster model with cluster weight $q\in[0,1]$ on arbitrary finite connected graphs. We give two explicit bounds that retain the dependence on the activities of the two specified edges and hold uniformly over the activities of all remaining edges. We also obtain a refinement of one of these bounds using connectivity information from the graph with the two specified edges removed. These results sharpen the bound obtained by a direct application of Lorentzian polynomial theory. At $q=0$, the model is supported on connected spanning subgraphs. In particular, for the uniform probability measure on these subgraphs, the joint inclusion probability of two distinct edges is at most $10/9$ times the product of their marginal inclusion probabilities.
Read the full abstract

Optimization and Control

Optimal last-iterate rates for the projected decentralized stochastic subgradient method

We study the projected decentralized stochastic subgradient (PDSG) method for minimizing the average of convex, possibly nonsmooth local objectives over a common closed convex constraint set and a fixed connected undirected network. We focus on the finite-time expected suboptimality of the last local iterate generated at each node, which is the actual output locally available without time averaging or network averaging. Under the standard stepsizes $\alpha_t=\alpha/\sqrt{t}$ and $\alpha_t=2/(\mu t)$, we establish $\mathcal{O}(\log T/\sqrt{T})$ and $\mathcal{O}(\log T/T)$ last-local-iterate rates for PDSG when each local objective is convex and the aggregate objective is strongly convex, respectively. To eliminate the logarithmic loss, we introduce a general horizon-dependent stepsize framework, termed cooldown stepsize, that combines a base stepsize with segmentwise scaling over a multistage partition and includes the dyadic construction (Jain et al, SIAM J. Optim., 2021) as a special case. For adapting to decentralized settings, we propose a long-range variation condition to make the stepsize schedule compatible with network contraction. Under this cooldown stepsize, we obtain $\mathcal{O}(1/\sqrt{T})$ and $\mathcal{O}(1/T)$ last-local-iterate rates for PDSG in both problem classes, matching the optimal centralized dependence on $T$. Experiments on a lower-bound instance and constrained hinge-loss problems illustrate the predicted terminal behavior of local iterates and the control of consensus error under the cooldown stepsize.
Read the full abstract

All authors Hao Wu, Chao Yin, Yuhong Dai, Liping Wang