Peer Revenue Now accepting positions

Papers

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

All of arXiv

Friday 6 November

Networking

An Empirical Evaluation of a Weighted PALU Model for Internet Traffic Networks

The PALU model is a hybrid power law generative framework that decomposes Internet traffic graphs into three structural components: a preferential attachment core, degree one leaf nodes, and disconnected satellite clusters. Although the original PALU formulation provided a theoretical basis for this decomposition, it was never validated against real traffic data and was restricted to unweighted networks. This work presents the first empirical evaluation of PALU against real Internet traffic, using packet traces from the MAWI WIDE samplepoint F dataset. We introduce a weighted extension of the model incorporating strength degree scaling, and develop a packet-to-graph pipeline for parameter estimation across multiple observation window sizes. Results show that the power law degree exponent $\alpha \approx 2.0$ remains stable across observation scales from $10^5$ to $3 \times 10^6$ packets, consistent with the PALU structural prediction. However, the satellite cluster parameter $\lambda$ and the weighted scaling parameter $\theta$ exhibit systematic dependence on observation scale, and the Poisson star assumption for unattached components is formally rejected at all tested window sizes. These results indicate that PALU captures important aspects of Internet traffic structure but requires an explicit observation model to fully account for finite sampling effects.
Read the full abstract

Numerical Analysis

Bayesian Inverse Problems for EUV reflectometry with regularized JKO scheme

We study a Bayesian inverse problem in EUV reflectometry, where optical constants and layer thicknesses are inferred from angle- and wavelength-resolved reflection measurements. The physical process is modeled and evaluated using an efficient forward model based on the transfer matrix method. As an alternative to computationally expensive sampling-based inference, we employ an Eulerian, entropically regularized Jordan-Kinderlehrer-Otto (JKO) scheme that minimizes the Kullback–Leibler divergence via proximal steps in the Wasserstein space. Using the Benamou–Brenier dynamic formulation and a Hopf-Cole transform, each JKO step is reformulated as a pair of heat equations coupled only through their initial and terminal conditions, enabling an efficient fixed-point solver. The resulting high-dimensional densities are discretized on tensor-product grids and represented in tensor-train format, exploiting Kronecker structure for scalable computation. Preconditioning based on estimated posterior moments further improves tractability. We test the method with numerical experiments on EUV reflectometry data for a ruthenium–silicon ($\ce{Ru}$ on $\ce{Si}$) specimen and compare the learned tensor-train posterior to ensemble Markov chain Monte Carlo (MCMC) reference samples. We demonstrate a use case in which we are able to reuse the tensor-train posterior to obtain samples for shifted target distributions with less computational effort than a comparable MCMC method.
Read the full abstract

Atmospheric and Oceanic Physics

Half-precision inference of AI weather prediction models: negligible loss of skill at a fraction of the cost

AI weather prediction models are released with 32-bit (fp32) weights and run in fp32 by default, although current GPUs compute much faster in 16-bit formats. We run five publicly available models – Aurora, Pangu-Weather, FourCastNet, SFNO and FastNet – on a single 16 GB GPU from 52 initial dates in 2024, in fp32 and with their networks in half precision (fp16), and verify the forecasts against ERA5. The physical state must stay in fp32: converting the whole model rounds it in physical units, which in fp16 overflows and in bf16 discards meaningful information. With the state, its normalisation and denormalisation in fp32, running the network in fp16 changes the global RMSE by at most 1.4%, and by about 0.5% or less for four of the five models, equivalent to less than one hour of lead time up to day 8 and at most two hours at day 10. The differences show no systematic pattern by region, scale or for tropical and extratropical storms, apart from a small tropical penalty in FourCastNet, and fp16 leaves the CRPS, spread and calibration of Aurora's ensemble almost unchanged. No value overflowed fp16's numeric range (+/-65,504); the largest stay 38 to 544 times below it. A 10-day forecast in fp16 is 1.3-3.5 times faster than in fp32 and uses 1.4-3.7 times less energy. Sensitivity experiments show that rules of thumb from other applications of reduced precision carry over only partly: the first and last layers dominate for some models but not for others, and no type of layer is sensitive on its own. For running AI weather models on a single GPU, fp16 with the state in fp32 is practically lossless at a fraction of the cost. bf16, often preferred for its wider range, is as fast but less accurate than fp16 for every model, and simple 8-bit quantisation costs far more skill.
Read the full abstract

Materials Science

Inverse design of superconductors and semiconductors with differentiable foundation models

Designing a crystal for a target property requires choosing its elements, stoichiometry, and atomic structure together, yet gradients exist only for the continuous structural variables. We introduce differentiable Equiformer fast effective-atom theory (DEFEAT), which lets a pretrained EquiformerV3 foundation model accept continuous species weights and site occupancies without retraining on fractional data, thereby enabling gradient-based materials optimization. Gradients from fine-tuned property models and a machine-learned energy penalty then guide updates to composition, atom count, atomic positions, and lattice vectors in a single optimization. In a search for phonon-mediated superconductors, density-functional perturbation theory on 45 structurally novel designs gives isotropic Allen-Dynes critical temperatures above 10 K for 39; the highest is 24.0 K for ZrTiNb2CN3, and H4LiPd3 (23.2 K) lies on the static DFT hull. In searches for photovoltaic absorbers, SCAN calculations confirm ten novel crystal structures with band gaps within 20% of a 1.34 eV target on their respective raw or empirically corrected scales, all within 70 meV/atom of the DFT hull, and show that the systematic gap error of the training functional propagates into the designs unless the objective corrects for it. Quasiharmonic free energies from machine-learned potentials further show that finite-temperature competition with solid phases can reorder candidates relative to the static hull.
Read the full abstract

Numerical Analysis

Symmetric Zassenhaus splittings for bounded and unbounded operators

We establish rigorous local-error bounds for the entire family of symmetric Zassenhaus splittings for possibly unbounded operators, under explicit assumptions on commutator bounds, regularity-preserving flows, and conjugation identities. Truncated symmetric Baker-Campbell-Hausdorff expansions yield an explicit coefficient formula for the Zassenhaus exponents and identify vanishing commutator expressions. In order to extend these cancellations to the realm of unbounded operators, we prove a general theorem on finite expansions of multiple operator conjugations with controlled integral remainders. For splittings of order 2n, our argument gives the sufficient commutator-length bound 4n-1, revealing regularity requirements that are not apparent from the formal order calculation. Using the Schrödinger equation with a quadratic potential as an example, we show that the order-dependent regularity requirements can reduce to one fixed weighted Sobolev space for all splitting orders.
Read the full abstract

Computer Vision

Video Stitching from Multiple Moving Cameras

We address geometric video alignment for the same scene captured by multiple freely moving cameras. Existing methods, which are based on image stitching or video stabilization, are typically limited to two cameras or must assume the restrictive setting of a fixed linear ordering of the cameras. We propose MViS, a lightweight test-time optimization framework for joint multi-view alignment with temporal consistency. MViS uses a spatial-transformer-inspired net to jointly process frames from all views within a short temporal window and predict per-view, per-time homographies. Optimization proceeds in two stages: geometry-guided alignment followed by appearance-based refinement, with both stages enforcing multi-view and temporal consistency across the sequence. The method requires no scene-specific hyperparameter tuning, handles arbitrary and time-varying view-intersection graphs, and remains robust under rapid camera motion. On a standard two-view video benchmark, MViS matches or surpasses the state of the art. To facilitate evaluation in more complex scenarios, we introduce an annotated dataset with more than two freely moving cameras and establish the first benchmark for this setting. Project page: https://bgu-cs-vil.github.io/MViS-project/ .
Read the full abstract

All authors Shira Ifergane, Roy Amoyal, Shahar Benishay, Oren Freifeld

Cryptography and Security

Performance Bottlenecks in Computing ZK Proofs on a GPU

Zero-knowledge proofs allow one party to prove the correctness of a statement or execution of a computation while revealing no more than the statement of correctness itself. Recent use cases, such as AI datasets, suggest the need for faster proof generation over increasingly large constraint sets. This paper explores efforts in optimizing zero-knowledge proof generation, specifically PlonK, using GPUs, for applications that are not amenable to solution via incrementally verifiable computing (IVC). Our intention with this work is to improve ZK performance by introducing fundamental changes in the low-level architectural structure of proof generation. These changes affect performance curves for most current proof generation algorithms. In particular, this paper explores the trade-offs between increased parallelization on GPUs and the resultant memory management problems that occur due to this parallelization. We implement a parallelized PlonK in order to analyze these trade-offs and make arguments concerning the nature of data flows within the proving protocol. Our results lead us to consider multi-GPU environments to maximize performance for large circuit sizes. Such environments introduce issues of communication, efficient hardware management, data partition, and data transfers. This paper further discusses the use cases that motivate multi-GPU implementations, their cost models, and the use of remote direct memory access (RDMA) for data transfer among GPUs.
Read the full abstract

All authors Victor Carolino, Caleb Geren, Michael Kaufman, Maximilian Nussbaum, William Scott, Henry F. Korth

Artificial Intelligence

Alignment by Confident Debate: Preference Learning from Multi-Agent Debate

Post-training language models for reasoning typically rely on gold answers, task-specific verifiers, or learned reward models, limiting scalability when such signals are scarce. Multi-agent debate offers a self-supervised alternative, but converting debate into reliable training supervision remains challenging: majority voting weights agreeing responses equally, lacks a principled tie-breaking rule, and is ill-defined for open-ended tasks without a canonical answer; likelihood-based scoring can conflate fluency with correctness; and homogeneous committees may reinforce correlated errors. We introduce Alignment by Confident Debate (AbCD), which converts debate traces into preference data without gold answer labels. A sampling-diverse committee debates each problem, after which final answers are clustered by task-level semantic equivalence. We use verifiable tasks (mathematical reasoning, multiple-choice quantitative QA, and executable code-generation) as a controlled test bed: their answer keys or executable checkers let us evaluate the quality of debate-derived supervision against ground truth, even though the oracle itself never uses gold labels. Post-training on the resulting preferences outperforms inference-only ensembling and debate, as well as SFT on the same traces, across five model families spanning 1B to 8B parameters and four datasets, improving consensus accuracy, with a maximum gain of $21.5_{pp}$. Ablations show that the approach transfers to a larger model and to code generation, while also revealing limitations of heterogeneous committees and agreement-based weighting. Overall, the results suggest that debate traces can yield useful preference supervision when answer equivalence is combined with confidence-weighted disagreement.
Read the full abstract

All authors Nikhil Verma, Dingxian Wang, Mohammad Reza Rezaei, Andrew Rabinovich, Zhao Chen

Fluid Dynamics

TurtLES: A Large-Scale Benchmark for Turbulent 3D Neural PDE Surrogates

Machine-learning surrogates are increasingly used to accelerate computational fluid dynamics, yet progress is limited by the lack of benchmarks capturing realistic, time-dependent turbulent flows. We introduce TurtLES, a 14 TB dataset of 3D turbulent cylinder wakes in the shear-layer transition regime, computed via high-fidelity implicit large-eddy simulations. The dataset contains 400 trajectories of 400 time steps each, with 3-11 million points per frame, spanning variations in geometry, Reynolds number, and angle of attack. Unlike existing datasets, TurtLES combines (i) physically realistic turbulence with an active energy cascade driven by vortex stretching, (ii) dense irregular meshes, and (iii) long-horizon temporal dynamics. We benchmark state-of-the-art neural operators on autoregressive rollouts and identify key failure modes, including temporal instability, loss of small-scale energy, and challenges in learning on large irregular domains. TurtLES provides a new benchmark for evaluating surrogate models on 3D turbulent wakes at industrially relevant mesh resolution.
Read the full abstract

All authors Jarno Platenburg, Armand Kassaï Koupaï, Paola Cinnella, Patrick Gallinari

Logic

Modal weak Kleene logics: algebraic semantics and Jónsson-Tarski representation

In this paper we continue the work started in [Bonzio & Zamperlin 2025] by investigating the algebraic semantics for modal external weak Kleene logics. In order to do so we study the global modal logics associated with the three-valued Kripke models introduced in the previous article, showing how these systems are algebraizable. We proceed by defining their algebraic counterparts, obtaining the classes of modal Bochvar and modal Halldén algebras, whose structure theory we investigate via Płonka sum decomposition. Of particular interested is the interaction of the modal operator with the homorphisms of the Płonka decomposition. These results allow us to prove representation theorems for these expanded modal algebras in the style of the Jónsson-Tarski representation for Boolean algebras with operators. As a consequence, we also obtain algebraic semnatics for the local modal logics from which the work originated.
Read the full abstract

Statistical Computation

Circular interval-valued regression with application to wind direction: A cylindrical approach

Many real-world datasets are inherently interval-valued, yet standard statistical methods are typically designed for point observations. Furthermore, intervals that are circular in nature may naturally exist. In this paper, we introduce a cylindrical regression framework for interval-valued circular data, addressing the joint challenges of the periodical nature and measurement uncertainty within the observed circular intervals. The proposed model accommodates a circular interval-valued response variable and both circular and linear covariates using a centre-range approach. To estimate the model, a maximum likelihood approach is proposed. The performance of the proposed estimation procedure is demonstrated through a simulation study. Application to two real-world datasets from a wind farm in South Africa and a weather station in Brazil further substantiates the model's practical value.
Read the full abstract

Chaotic Dynamics

Synchronization transitions in Kuramoto oscillators with mixed interactions

The Kuramoto model is a paradigmatic system for emergent synchronization phenomena, and significant efforts have been made to investigate its transition from incoherent to coherent dynamics. However, previous studies have almost exclusively assumed global coupling between the oscillators, even though many real-world systems are organized in specific network structures. Here, we fill this gap by examining finite-size effects in Kuramoto oscillator ensembles on networks with mixed-order interactions. Using numerical simulations, we show how parameters such as the size of the system, the average degree of the nodes and the strength of the interactions alter the behaviour of the transition to synchronization with respect to the known theoretical results that are derived for globally coupled infinite systems.
Read the full abstract

All authors Abhishek Sharma, Ayushi Suman, Mehernaz Anvari, Manish Dev Shrimali, Charo I. del Genio, Sarika Jalan

Quantum Physics

BQP-Complete Susceptibilities Observable at Ballistic Times in Local Quantum Dynamics

Local response measurements track how perturbations spread in quantum simulators and nonequilibrium many-body systems. Locality bounds constrain the propagation speed, but do not determine whether the response can distinguish computational outcomes at fixed measurement precision as the distance increases. Here we construct time-independent dynamics with bounded nearest-neighbor interactions in which a fixed single-site detector responds to a remote bond phase after a time proportional to their separation. The susceptibility has a positive lower bound independent of distance. With product-state preparations, estimating this susceptibility to constant additive accuracy is complete for bounded-error quantum polynomial time (BQP), with a decision gap independent of distance. Two source-phase settings give the susceptibility with a distance-independent shot count at fixed accuracy and confidence. Inverse-polynomially weak local couplings between the computational register and the chain carrying the excitation preserve the gap. The leading correction vanishes when exchanging the two internal states of the excitation reverses the interaction. Varying the source and observation time makes all bond phases locally identifiable near the zero-phase reference setting. We also prove a conditional quantum advantage for distribution-free response prediction.
Read the full abstract

Analysis of PDEs

Regularity of planar Griffith minimizers with connected cracks through a monotonicity formula

We prove a new monotonicity formula for the two-dimensional Neumann elasticity equation with connected boundary. From this we deduce a regularity result for two-dimensional minimizers of the Griffith energy with connected crack set: the crack may be partitioned into regular points, triple junctions, and cracktips. We give a classification of the global minimizers of the Griffith energy with connected crack set. Finally, we discuss some consequences of the monotonicity formula on the evolution of the stress intensity factors along a crack, which may present independent interest.
Read the full abstract

Robotics

Lightweight Path-Aware Frontier Re-ranking for LiDAR-Based UAV Exploration

Autonomous 3D exploration with UAVs requires efficient target selection under limited flight time and onboard computation. In conventional Euclidean-distance-based frontier pipelines, geometric distance may poorly represent the actual collision-free travel cost in cluttered environments, leading to inefficient frontier choices. This paper presents a lightweight path-aware frontier re-ranking method for LiDAR-based UAV exploration. The method first retains a small top-N set of promising frontier candidates using a lightweight utility function and then applies time-bounded RRT* planning to estimate their feasible path costs for re-ranking. The path generated for the selected frontier is further reused for navigation to reduce redundant global planning. Experiments in three warehouse-like environments show that the proposed method consistently produces shorter exploration trajectories than conventional frontier baselines, with traveled-distance reductions of up to 14.1%, while reducing exploration time by up to 9.5% in two environments and remaining competitive in the third. The method maintains map detail comparable to the conventional OctoMap-based baseline, demonstrating an effective trade-off between exploration efficiency and mapping quality.
Read the full abstract

All authors Duc Manh Do, Thanh Nguyen Canh, Ngoc Minh Do, Mai Thanh Thai, Xiem HoangVan

Machine Learning

Solver-Free Decision-Focused Learning via Neural Optimization Layers for Edge Selection

Decision-focused learning trains prediction models to improve downstream decisions, but many methods require costly solver calls or optimal-solution labels. We propose a solver-free decision-focused learning framework for NP-hard edge-selection problems with uncertain parameters in the objective, the constraints, or both. A neural optimization layer is first trained without solution labels, using problem objectives and constraint penalties. Graph-informed edge scoring and iterative refinement encourage concentrated and structurally consistent continuous decisions. The layer is then frozen, and a prediction model is trained through its continuous output using a problem-specific loss evaluated under observed parameters. At deployment, the predictor, optimization layer, and a lightweight decoder produce decisions without solving the original NP-hard problem. Experiments on the asymmetric traveling salesperson and endpoint-free orienteering problems show consistently better decision quality than prediction-focused learning and improvements over most evaluated decision-focused baselines. Our framework requires less training time than the evaluated decision-focused baselines, even when optimization-layer pretraining is included. Neural deployment also produces decisions faster than solver-based deployment, offering a better quality-efficiency trade-off. The layer also reduces training cost in existing decision-focused methods when used as an approximate optimization oracle.
Read the full abstract

Optimization and Control

Can a Language Model Write the Weights? An Empirical Study of LLM-Written Scoring Networks for Column Generation and MILP Branching, with a Bus-Driver Scheduling Application

This paper was generated by an AI model (Claude Opus 5.5, Anthropic) under the author's direction; all results come from computations run on the author's machine, and every reference was checked. Weight-space programming (WSP) is the idea that a large language model (LLM) can write the weights of a small neural network directly, guided only by reading forward-pass outputs, with no gradient training. We test whether WSP can improve exact operations research algorithms where they rely on cheap scoring rules. In four studies, an LLM wrote small multilayer perceptrons that (i) select arcs, labels and ng-neighbourhoods inside column generation for the vehicle routing problem with time windows, and (ii) choose branching variables in SCIP's branch-and-bound. The algorithms stay exact whatever the weights; a bad network only costs time. In column generation, no lever gave a material gain. In branching on synthetic classes, a hand-written network solved independent-set instances in 0.37x the time of SCIP's default rule (18 paired instances, Wilcoxon $p<10^{-5}$) and matched or beat networks trained by imitation of strong branching, but on 39 MIPLIB 2017 instances it solved 19 versus 24. In an application built from the real GTFS feed of Atlanta's transit agency (MARTA), a crew scheduling set-partitioning model solved by a native C plugin with SCIP's factory settings, the LLM-written rules took 0.63x the default's time on 32 held-out instances ($p<10^{-4}$), but SCIP's plain pseudocost rule did exactly as well (ratio 0.99, $p=0.19$). In these settings, the gains came from being cheap, not from what the LLM wrote. All code, logs and the numbers behind every figure are released.
Read the full abstract

Game Theory

Optimal Regret for Profit Maximization in Bilateral Trade

Learning to profit from trade requires learning both sides of a market: what a seller is willing to accept and what a buyer is willing to pay. We study repeated bilateral trade, where an intermediary posts a purchase price to a seller and a sale price to a buyer whose valuations are drawn from an unknown joint distribution. Trade occurs only if both parties accept their respective prices, in which case the intermediary earns the difference between the two prices. The intermediary observes only whether trade occurs, without observing either valuation. We show that learning both sides of the market is no harder, in terms of regret, than learning just one. Specifically, we design an algorithm achieving regret $\widetilde{\mathcal{O}}(T^{2/3})$ against the best fixed pair of prices, matching, up to logarithmic factors, the classical lower bound for one-dimensional dynamic pricing. Our result shows that the classical one-dimensional regret rate survives the transition to two dimensions, raising the question of whether this absence of a curse of dimensionality extends to arbitrarily high dimensions.
Read the full abstract

Algebraic Topology

The splitting of the stable space of equivariant $h$-cobordisms

We prove that the space of equivariant $h$-cobordisms on a smooth $G$-manifold $M$ has a tom Dieck style splitting, after stabilization. As an intermediate step, we introduce a space of isovariant $h$-cobordisms on $M$ and prove that this space also has a splitting, even before stabilizing. This completes the proof of the equivariant version of Waldhausen's stable parametrized $h$-cobordism theorem. Along the way, we establish several foundational results on equivariant smooth manifolds with corners.
Read the full abstract

All authors Thomas Goodwillie, Kiyoshi Igusa, Cary Malkiewich, Mona Merling

High Energy Astrophysical Phenomena

XMM-Newton observations of the highest field polar AR UMa: An extreme case of soft-blobby accretion

(Context) The X-ray spectra of magnetic cataclysmic variables (MCVs) with extreme magnetic fields, the polars, often exhibit a prominent soft X-ray excess attributed to buried shocks. (Aims) Using the first-ever high-state XMM-Newton observation of the highest field polar, AR UMa ($B\approx230$ MG), we established the source's accretion scenario and derived the energy budget at the magnetic extreme of MCVs. (Methods) Using X-ray, UV, and optical spectral and timing analysis, we disentangled the radiation components and located their origin in the binary. (Results) AR UMa showed one active ("southern") pole during our observation. The spectrum of the active pole was dominated by soft, blackbody-like emission with $kT_{\rm BB} \simeq 35$ eV and showed no evidence for an optically thin hard X-ray emission characteristic of a cooling post-shock accretion column. Instead, we find constant, faint, optically thin plasma emission with a temperature of $\approx 1$ keV at all orbital phases, which is uncorrelated with the soft X-ray emission, and we attribute this to shock-heated matter at the stagnation region. Pronounced absorption dips punctuate the phase of active X-ray emission and are associated with puffed-up material in the stagnation region. Using spectro-timing analysis of the absorption dips, the density profile and evolution of the clumpy flow were probed. We find evidence of a cyclotron harmonic from the southern pole (160 MG) near $\sim$2300 Angstrom in the UV spectrum. We report minute-long quasi-periodic oscillations in the optical, UV, and X-ray light curves. (Conclusion) AR UMa represents an extreme case of soft, blobby accretion, with an effectively infinite soft excess, extending the magnetic field range over which magnetic accretion can be studied by a factor of two. It is one of the rare systems in which X-ray emission from a stream-magnetosphere interaction is detected.
Read the full abstract

Representation Theory

Supercuspidal, discrete, and elliptic densities for reductive groups over non-archimedean local fields

Let $G$ be the group of points of a connected reductive group over a non-archimedean local field $F$ of arbitrary characteristic, with residue field of size $q$. For a principal congruence tower $K_n$ attached to a Lie algebra lattice $\mathcal{L}$, the supercuspidal and discrete series proportions of the Plancherel mass of the $K_n$-fixed vectors both converge, with error $O(q^{-n/4})$, to the probability that a Haar-random form in the dual lattice $\Lambda=\mathcal{L}^\vee$ is elliptic. For $\mathrm{GL}_d$, the supercuspidal limit confirms a prediction of Howe. This probability is positive, and we characterize when it is one. More generally, outside a union of congruence cells of total mass $O(q^{-n/4})$, each cell determines a common Levi class for the supercuspidal and discrete series supports of every tempered representation containing its character. The limiting mass of each class is the probability that a Haar-random form in $\Lambda$ lies in the corresponding stratum. The proof combines finite Fourier analysis, quantitative estimates for parabolic incidence maps, and a Jacquet-module argument that detects unitary exponents.
Read the full abstract

Computation and Language

Query-Guided RoPE Modulation for Lost-in-the-Middle Mitigation in Long-Context LLMs

Long-context language models often neglect information in the middle of the input, a phenomenon known as "lost-in-the-middle". Part of this gap may stem from the long-range decay of rotary position embedding (RoPE), which weakens the positional signal between distant tokens. Existing input compression deletes low-importance tokens, and attention reweighting acts only after RoPE has weakened this signal. RoPE-level methods modify positions directly but apply the same remapping to every query. To address these limitations, we propose QGPM (Query-Guided Positional Modulation), a training-free method that reshapes RoPE positions using query relevance from the model's own attention heads. QGPM builds a broad relevance map for lower layers and a sharp one for upper layers. Low-relevance tokens advance by less than one position on fast-rotating planes, pulling relevant evidence toward the query. Slow-rotating planes keep the original layout, with every token in place. Experiments on NoLiMa and RULER with multiple context lengths show that QGPM improves over base models and strong baselines, with the largest gains on retrieval tasks, and generalizes across multiple backbones. Our code is available at https://github.com/iZHpy/QGPM.
Read the full abstract

All authors Peiyu Zhang, Heng Ping, Shinibali Bhattacharyya, Anwesan Pal, Nikos Kanakaris, Kevin Cheang, Shixuan Li, Wei Yang, Paul Bogdan

Statistical Methodology

Risk-set reconstruction balance: zero-targeted diagnostics for artificial censoring in clone-censor-weight analyses

In clone-censor-weight (CCW) analyses, inverse probability of censoring weighting corrects selection from artificial censoring at protocol deviation. Balance diagnostics, such as the end-of-grace between-strategy standardised mean difference (SMD) and censoring balance, compare different populations, and their zero targets under a correct model are not formally stated. We show that, under positivity and a correct censoring model, the weighted retained individuals reconstruct the risk set before each censoring interval and the baseline population, and we propose risk-set reconstruction balance (RRB), which measures how far they depart from these two populations. Interval RRB has a zero target for every history function, including post-baseline covariates, for which between-strategy SMDs have none. Cumulative RRB keeps a zero target for baseline covariates after outcome events, whereas the end-of-grace SMD acquires a depletion term. In simulations, interval RRB responded to an omitted term's effect on censoring whether or not the term was prognostic. Under correct models, the exceedance rate of its studentised maximum over a null reference for chance imbalance was 3.5-6.0% (nominal 5%) when the censoring model pooled nine intervals and mostly 1.0-4.5% under censoring in a single interval. With many early events, cumulative RRB stayed near zero, whereas the end-of-grace SMD reached 0.40; under loss to follow-up that the weights ignored, cumulative RRB departed from zero and interval RRB stayed small. In a case study, interval RRB flagged a dispensing indicator outside the censoring model. Interval RRB checks the censoring model, and cumulative RRB the representation of the baseline population.
Read the full abstract

Distributed Computing

Towards Sustainable Federated Learning via Carbon-Aware Client Selection

The Information/Communication Technology (ICT) industry is severely challenging the global environment due to the sharply rising carbon emissions generated by machine learning (ML) workloads. Federated Learning (FL) is one of the prominent distributed ML systems used in applications with billions of client devices, such as Google Gboard and Apple Siri. Nearly half of the total carbon footprint in FL originates from client-side computation during local training. However, existing client selection strategies in FL primarily focus on minimizing energy consumption rather than carbon emissions, despite the fact that carbon impact depends not only on computational load but also on the underlying carbon intensity of the energy source. In this paper, we address this gap by proposing a carbon-aware fair client selection strategy (CA-FCS) designed to minimize the carbon emissions associated with model training on client devices. In contrast to existing approaches for reducing carbon emissions in FL, we incorporate hourly carbon intensity and a realistic energy consumption model developed using empirical measurements from Google Pixel smartphones to estimate client-side carbon emissions in FL. Since repeatedly selecting low-carbon clients may reduce data diversity and degrade global model performance, particularly under non-independent and identically distributed (non-IID) data settings, CA-FCS incorporates fairness into client selection to mitigate this issue. We conduct extensive experiments under realistic FL settings for mobile devices, where client participation is constrained by device availability conditions, such as charging status and Wi-Fi connectivity. The results demonstrate that our method reduces client-side carbon emissions by up to 46% while maintaining competitive model accuracy.
Read the full abstract

Solar and Stellar Astrophysics

A joint astrometric, eclipse timing and radial velocity orbit for the hierarchical triple V819 Herculis

We present a joint Bayesian orbital analysis of the hierarchical triple system V819 Herculis using ground-based astrometry, radial velocities (RV), and eclipse timing variations (ETV). By coupling these complementary data sets within a physically consistent Keplerian framework, we obtain a self-consistent three-dimensional solution for the outer orbit and the dynamical properties of the system. The outer orbit is found to have a period of $P_3 = 2016.97 \pm 0.27$ d and a high eccentricity of $e_3 = 0.6800 \pm 0.0016$, with the same orbital geometry simultaneously reproducing the sky-plane astrometry, the LiTE signal in the eclipse timings, and the tertiary radial velocities. The joint solution determines the component masses dynamically from these data alone, yielding $M_{12} = 2.721 \pm 0.055\,M_\odot$ for the inner binary and $M_3 = 1.954 \pm 0.070\,M_\odot$ for the tertiary, together with a LiTE amplitude of $A_\mathrm{LiTE} = 0.0104$ d and an RV semi-amplitude of $K_3 = 18.46$ km s$^{-1}$. The joint fit also yields an orbital parallax of $\pi_\mathrm{orb} = 14.08 \pm 0.13$ mas, consistent with Gaia DR3, Gaia DR2 and the HIPPARCOS New Reduction. The derived tertiary mass is higher than expected for a standard G8-type main-sequence star, suggesting that the tertiary component may be in an advanced evolutionary stage. The resulting outer orbit is well constrained, and the system is strongly hierarchical, with an outer period about 900 times the inner one. Finally, we provide predicted astrometric positions for the tertiary component for five years beyond 2026.5 to guide future high-precision observations.
Read the full abstract

Software Engineering

Deliberate Pacing for Human Oversight in Agentic Software Development

Agentic AI increasingly enables software development activities to proceed without continuous human involvement. While this autonomy may accelerate development, it also raises questions about when developers can meaningfully oversee software development done by agents. We propose deliberate pacing: intentionally adjusting when and how autonomous development proceeds to preserve opportunities for meaningful oversight. Drawing on an exploratory multivocal literature review of 20 sources, we identify three factors motivating this vision: changing relative costs of human involvement, the timing of oversight opportunities, and interdependencies between pacing decisions across the software development life cycle. We argue that neither uninterrupted autonomy nor frequent human intervention is universal. Instead, pacing should balance opportunities to understand and influence consequential decisions against costs in human attention, development time, and rework. We outline a research agenda to investigate when deliberate pacing is beneficial, how intervention timing affects oversight, and how pacing decisions interact across development activities.
Read the full abstract

All authors Tobias Reichel, Michèle Stassen, Anamaria Mojica-Hanke, Maria Teresa Baldassarre, Sebastian Baltes, Alexander Serebrenik, Steffen Herbold

Quantum Physics

Relativistic analogue of the Kramers-Henneberger transformation and pair creation

In analogy to the Kramers-Henneberger transformation in non-relativistic quantum mechanics, we propose and investigate a spinor transformation for the Dirac equation which allows us to transform space-time dependent vector potentials $A(t,x)\boldsymbol{e}$_x away. In the mass-less case, we exactly arrive at the free Dirac equation, while for massive Dirac fields, the transformation results in a non-trivial mass matrix containing space-time dependent phases. Special emphasis is placed on the phenomenon of fermion pair creation in the external field $A(t,x)\boldsymbol{e}_x$ where the spinor transformation allows us to conceptually disentangle the different contributions to pair creation. It can also be used to reformulate the WKB approach and the Riccati equation (which can be advantageous in certain cases) and to obtain additional insight into the mechanism of dynamically assisted pair creation.
Read the full abstract

All authors Johannes Marschall, Christian Kohlfürst, Naser Ahmadiniaz, Ralf Schützhold

Artificial Intelligence

CoMoFlow: Controllable Protein Conformational Ensemble Generation via Explicit Collective-Mode Modeling

Protein conformational ensemble generation aims to efficiently characterize the structural variations accessible to a protein beyond a single static structure. Recent flow-based methods efficiently generate conformational ensembles, but they model coordinated structural variation implicitly, which can limit ensemble fidelity and hinder direct control. To address this limitation, we propose CoMoFlow, a controllable framework for generating protein backbone conformational ensembles by explicitly modeling collective variation. CoMoFlow constructs protein-specific collective modes from the equilibrium structure, learns the distribution of their coefficients across the molecular dynamics (MD) ensemble, and uses sampled collective variations to guide SE(3) flow matching through feature conditioning and prior recentering. Experiments on ATLAS and de novo proteins show that CoMoFlow improves reconstruction of MD conformational distributions and coordinated structural variation relative to strong flow-based baselines, while enabling direct control over targeted conformational changes.
Read the full abstract

All authors Yusong Wang, Zheyuan Gu, Ailiang Lin, Minghao Shao, Yicheng Xu, Jiang Cai, Prayag Tiwari, Mingkun Xu

Mesoscale and Nanoscale Physics

Josephson current along the domain walls in Weyl fully compensated ferrimagnet Ti$_2$MnAl

We experimentally investigate Josephson current in planar Nb-Ti$_2$MnAl-Nb proximity devices, which are formed as junctions between superconducting niobium leads and single crystal flakes of the Weyl fully compensated ferrimagnet Ti$_2$MnAl. We demonstrate well developed Josephson effect for 2~$\mu$m distance between the niobium contacts on the Ti$_2$MnAl surface. The measured critical current dependence on the magnetic field $I_c(B)$ shows strong anisotropy for two magnetic field orientations in respect to the Nb-Ti$_2$MnAl-Nb junction plane. For the parallel magnetic field, one can see standard Fraunhofer-like $I_c(B)$ suppression pattern. In contrast, $I_c(B)$ is qualitatively different in normal magnetic field, indicating non-trivial distribution of the Josephson current within the junction plane. From this anisotropy, long-range supercurrent and fast aperiodic fluctuations in $I_c(B)$ we conclude that Josephson current is transferred by the chiral surface states, which emerge along the domain walls in topological compensated ferrimagnet Ti$_2$MnAl. The conclusion on the domain walls is supported by magnetization data for small Ti$_2$MnAl flakes.
Read the full abstract

All authors V.D. Esin, N.N. Orlova, A.R. Banschikova, Yu.S. Barash, D.N. Borisenko, N.N. Kolesnikov, E.V. Deviatov

Cosmology and Nongalactic Astrophysics

Through Ups and Downs: the Transient Lensing Behaviour of SIDM halos undergoing Core Collapse

Self-interacting dark matter (SIDM) haloes can become extremely dense following gravothermal core collapse, motivating their use as explanations for unusually concentrated gravitational lenses. Using an emulator calibrated on gravothermal fluid calculations, we investigate whether increasing central density necessarily strengthens lensing. We follow initially Navarro-Frenk-White (NFW) haloes through core formation and collapse deep into the short-mean-free-path regime, and compute finite-source magnifications, critical curves, and source-plane lensing cross sections for isolated spherical haloes with elastic, velocity-independent self-interactions, including external shear in selected calculations. Lensing strength evolves non-monotonically: it initially decreases during core formation, rises during early collapse, and generally declines at later stages despite continued growth of the central density. This decline accompanies the emergence of a dense inner core whose mass and radius decrease within a more extended, lower-density core. Time-averaged enhancement relative to NFW haloes depends on the lensing statistic, halo concentration, and interaction cross section. Generalized NFW haloes with central cusps $\rho \propto r^{-\gamma}$ with $\gamma \geq 2$ always have larger lensing efficacies than core-collapsed SIDM haloes of equal mass and are therefore poor proxies, including in terms of the demographics (size and radial thickness) of arcs and Einstein rings. The core collapse stage at which SIDM haloes are optimal lenses coincides with the epoch at which the characteristic core surface density is close to critical, which typically occurs during the early stages of core collapse. These results demonstrate that central density alone is insufficient to characterize SIDM lensing and that its evolving mass distribution must be modelled explicitly.
Read the full abstract