Peer Revenue Publish or profit

Papers

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

All of arXiv

Saturday 5 September

Game Theory

Games Over Observation Spaces in Multi-Agent Capture the Flag

We consider a multi-agent Capture the Flag (CtF) scenario in a graph-based environment, where a team of attackers seeks to reach designated flag vertices while a defending team attempts to intercept them. In our setting, both teams operate using decentralized heuristic policies. While the attacking team may choose its heuristic from a diverse library of policies, the defense is restricted to playing a single fixed policy. To overcome this limitation, a centralized defense oracle strategically restricts the portion of the graph visible to each of its agents in order to elicit a wider range of emergent behaviors from its fixed policy. We formalize this interaction as a two-player zero-sum game, where the attacker reasons over its library of heuristics and the defense reasons over the combinatorial space of visibility profiles. To solve this intractably large game, we propose a Double Oracle algorithm to find approximate empirical equilibria, and we empirically validate that observation manipulation can improve the defense's performance.
Read the full abstract

Quantum Physics

Quantifying Nonstabilizerness of Quantum Codes by Removing the Inert Background

Nonstabilizerness (magic) is the quantum resource that, together with stabilizer operations, makes universal quantum computation possible. However, quantifying nonstabilizerness is difficult. The standard measure sums over exponentially many Pauli operators, and for quantum codes no quantitative theory has been available. A structural observation removes this obstacle: the Clifford sector carries no magic and can be removed by a Clifford transformation, leaving an order-three object whose magic is an exactly evaluable fourth moment. Removing this inert background yields closed forms for several code families: a formula for all Dicke states, reducing a 100-qubit case from $4^{100}$ terms to a short binomial sum; a bound for cubic-phase codes (twisted quantum doubles and non-Abelian topological order), saturated only by the $D_4$ code; and a cyclic/zero criterion for group multiplication states. For codeword-stabilized (CWS) codes, the companion paper \cite{Liu26arXiv} carries this reduction to its conclusion: the most magical codes are exactly the Sidon sets, and the nonstabilizerness bounds the code's transversal non-Clifford power. Together these results reduce the computation of nonstabilizerness for a broad class of codes to finite classical counting problems, and identify the maximally magical codes among them.
Read the full abstract

Combinatorics

Long induced cycles in pseudorandom graphs

We show that, for some absolute constants $c_1,c_2>0$, every $(n,d,\lambda)$-graph with $\lambda\le c_1 d$ contains an induced cycle of length at least $c_2n\log(d/\lambda)/d$. This is best possible up to the values of $c_1,c_2$. Our techniques include a multi-scale algorithmic analysis, an adapted depth-first exploration procedure, a link to percolation theory, and estimates for random row-and-column extraction in symmetric matrices.
Read the full abstract

All authors Sahar Diskin, Lyuben Lichev, Michael Krivelevich, Itay Markbreit

Combinatorics

Sunflower-Free Uniform Families: Recursive Constructions and Explicit Bounds

Let $f(w,k)$ be the maximum size of a $w$-uniform family containing no sunflower with $k$ petals. We introduce a recursive construction for sunflower-free families and use it to obtain a general lower bound on the exponential growth rate of $f(w,k)$. We also prove a general upper bound for $3$-uniform families with at least four petals. Our results give $39\le f(3,4)\le49$, $f(3,5)\le146$, $153\le f(3,6)\le255$, $259\le f(3,7)\le474$, and $54\le f(4,3)\le83$. In addition, we prove that the maximum size of an intersecting $4$-uniform family containing no sunflower with three petals is $27$. The upper bounds $49$ and $83$ are computer-assisted. The finite lower bounds come from explicit constructions.
Read the full abstract

All authors Edward Axante, Cristian Budala, David Chitic, Bogdan Dumitru, Mihai Nacu

Signal Processing

Exploiting LLM Agents for Trustworthy AutoResearch in Wireless Communications

Large language model (LLM) agent-enabled AutoResearch is attracting growing interest across scientific disciplines, in which LLMs are leveraged for knowledge synthesis, multistep planning, code generation, tool invocation, and iterative refinement, thus automating the research lifecycle, from hypothesis generation and experimentation to analysis and manuscript preparation. Wireless communications is particularly suitable to this paradigm. This is due to the fact that advances in this field often rely on fundamental-limit analysis, system optimization, and protocol design, all supported by mature mathematical, simulation, and optimization toolchains, as well as standards, measurement, and digital-twin platforms. However, integrating autonomous agents into rigorous wireless research workflows requires traceable processes and verifiable evidence. This article presents a general trustworthy wireless AutoResearch framework. This framework employs typed research contracts, version-controlled artifacts, independent validators, and bounded agent authority to connect hypothesis generation with system modeling, mathematical formu lation, algorithm design, code generation, simulation, reproducible claims, and manuscript preparation. We conduct a case study on integrated sensing and communication (ISAC) for unmanned aerial vehicles (UAVs). It is shown that the proposed framework can generate innovative research ideas and, with appropriate expert intervention, produce a manuscript whose evaluation score is comparable to or higher than those of related IEEE conference and letter papers. These findings demonstrate the effectiveness of LLM agents in orchestrating authoritative scientific tools and versioned research artifacts, while underscoring the importance of human-AI collaboration.
Read the full abstract

All authors Yuan Guo Zixiang Ren, Jie Xu, Liang Hong, Fan Liu, Rui Zhang

Machine Learning

FMMO: Detecting the Divergence Between Local Attribution and Global Drift

Post-deployment drift poses a critical risk to algorithmic accountability, particularly when ground truth labels are delayed and performance degradation becomes a "silent failure". While Explainable AI (XAI) is often relied upon to audit these shifts, we demonstrate that popular local attribution methods (e.g., TreeSHAP) can exhibit misleading stability even as model reliability collapses. In this paper, we propose a Framework for Model Monitoring and Observability (FMMO) designed to expose the divergence between local explanation stability and global distribution shifts. Using benchmark, synthetic, and real-world datasets, we show that local XAI methods fail to flag drift-induced disparate impact, specifically where False Positive Rates spike for protected groups while feature attributions remain unchanged. By integrating global surrogate models with model utilization measurements, FMMO mitigates this fairness blind spot, ensuring that stakeholders can detect discriminatory deterioration that standard local XAI tools overlook.
Read the full abstract

Operating Systems

AutoUVM: Automated Prefetching Framework for LLMs under UVM Oversubscription

Large language models (LLMs) increasingly exceed the memory capacity of commodity GPUs, making memory oversubscription common in practical deployments. NVIDIA Unified Virtual Memory (UVM) provides transparent access to host memory, but its page-fault-driven migrations introduce severe performance overhead. While UVM exposes primitives (e.g., prefetching and placement hints) to mitigate these costs, they require low-level CUDA modifications, limiting their applicability for most LLM users. Meanwhile, existing UVM optimizations operate at coarse managed-object granularity and fail to capture deep learning frameworks' internal tensor-level memory behavior, leading to excessive data movement and CPU-GPU interconnect bottlenecks. We propose AutoUVM, an automated, framework-aware UVM prefetching system for efficient LLM execution under memory oversubscription. AutoUVM bridges the semantic gap between deep learning frameworks and UVM by exposing tensor-level access information and enabling policy-driven prefetching at fine granularity. Implemented as a transparent extension, AutoUVM requires no changes to model code and dynamically adapts to runtime memory pressure. We instantiate AutoUVM with a roofline-inspired policy to identify performance-critical data transfers. Across ten LLMs, AutoUVM achieves an average 3.1x speedup over baseline UVM and consistently surpasses the best-performing prior UVM prefetcher by 1.9x, with improvements of up to 4.7x over object-level prefetchers, while significantly reducing page faults.
Read the full abstract

All authors Mao Lin, Hui Feng, Xianzhong Ding, Guilherme Cox, Qian Wang, Hyeran Jeon

High Energy Physics (Phenomenology)

Inelastic dark matter and baryon flavor symmetry in light of LUX-ZEPLIN (LZ) experiment

We present a novel model for inelastic dark matter in a simple extension of the Standard Model (SM) with a local $U(1)'$ baryon flavor symmetry, such as $U(1)_{B_i-B_j}$ with $i\neq j$. In this framework, we realize a realistic flavor structure for the mixing Yukawa couplings due to higher dimensional operators or renormalizable couplings between the SM quarks and extra vector-like quarks that are anomaly-free by themselves. We regard a Dirac singlet fermion as being dark matter and obtaining a small mass splitting for inelastic dark matter, only after the $U(1)'$ symmetry is broken spontaneously. We show that there is a consistent parameter space for explaining both the LZ nuclear recoil event and the correct relic density for dark matter and the existing collider and indirect detection bounds are satisfied at the same time. In this case, we predict a relatively light $Z'$ boson with $10-20\,{\rm GeV}$ mass and a weak gauge coupling, suggesting a dedicated search at colliders. The future data with higher recoil energies at direct detection experiments will play a decisive role of discriminating between the scenarios of light and heavy dark matter.
Read the full abstract

Materials Science

A Transport Framework for Evaluating Nanoscale Interconnect Materials

As CMOS technology continues to scale, metallic interconnects increasingly limit circuit performance through rising resistivity, self-heating, and reliability degradation. Although several alternative metals have been proposed, a quantitative framework for evaluating their transport performance under nanoscale confinement remains unavailable. Here, we establish a transport framework by combining independently measured thermal and electrical conductivities with normalized and unnormalized transport figures of merit. We apply this framework to Cu, Ru, W, Co, Ir, and Mo thin films using new steady-state thermoreflectance measurements of Mo, Co, and Ir together with previously reported Cu, Ru, and W data. While Cu exhibits the highest intrinsic transport performance, its effective performance is substantially reduced by thickness scaling and Ta liner resistance. In contrast, Ru and Mo maintain favorable transport properties while enabling barrierless integration, identifying them as promising candidates for next-generation CMOS interconnects
Read the full abstract

Machine Learning

Decision-Aware Suffix Prediction and Reasoning of Business Processes

Suffix prediction forecasts the remaining sequence of events of a running case until completion. Most approaches rely on neural networks trained on event logs, which, on average, perform well but struggle with short prefixes or targets belonging to a rare process variant. In such scenarios, the correct path may cross multiple branching decisions, determined primarily by case- and event-level attributes, a signal that NN-based suffix prediction models tend to underweight because they may heavily weight (dense) event labels. Decision mining extracts rules for such decisions from the event log, but has so far been applied only to post-hoc and what-if analysis, not suffix prediction. We therefore extend suffix prediction with decision mining, introducing a decision-aware suffix prediction framework, a neuro-symbolic approach that enables reasoning about predicted events via mined decision rules. Experiments on three of four event logs and three suffix predictors show that the framework can improve suffix prediction, especially for short prefixes but also for rare process variants, and adds intrinsic interpretability.
Read the full abstract

Classical Analysis and ODEs

Terminating Zero-Balanced Hypergeometric Series Using Divided Differences

There is a close relationship between divided differences and hypergeometric series, and recent studies have shown that divided differences can be used effectively to derive terminating hypergeometric identities. In this paper, we apply this approach to terminating zero-balanced hypergeometric series. Using explicit product evaluations arising from divided differences, we give alternative proofs of zero-balanced ${}_3F_2(1)$ and ${}_4F_3(1)$ summation formulas and then extend the argument to the general terminating ${}_{r+1}F_r(1)$ case. We also combine the Lagrange representation with the Leibniz rule for divided differences to derive a finite convolution transformation for a terminating ${}_{r+2}F_{r+1}(1)$ series. Thus, the divided-difference approach not only reproduces the known zero-balanced summation formulas but also yields a finite convolution transformation for terminating hypergeometric series, together with its zero-balanced specialization.
Read the full abstract

Number Theory

Central non-vanishing of Dirichlet $L$-functions

We prove that, for every sufficiently large modulus $q\not\equiv2\bmod4$, at least $3/8-o(1)$ of the primitive Dirichlet characters $\chi$ modulo $q$ have $L(1/2,\chi)\ne0$. We also establish stronger proportions for moduli $q$ satisfying certain arithmetic conditions.

Number Theory

The entropy of coclass one and of SmallGroup(729,45)

Investigating the descendant tree T(Delta) of 3-groups G of maximal nilpotency-class, with root Delta=SmallGroup(27,3), we show for the first time that an entropy-value H(P_r)=sum P_r(G)|log(P_r(G))| can also be determined for the real probability-distribution P_r(G) of all terminal vertices outside of the mainline on this infinite digraph. These vertices arise as automorphism groups Gal(F(3,infinity,K)/K) of maximal unramified pro-3-extensions F(3,infinity,K) of real-quadratic number fields K=Q(d^1/2), d>0. For the groups G of another descendant tree T(N), with root N=SmallGroup(729,45) and unbounded coclass, which has been studied by L. Bartholdi and M. R. Bush, we compare the entropy H(P_r) of the real measure P_r(G) with the entropy H(P_i) of the imaginary measure P_i(G).
Read the full abstract

Computer Vision

Multiple Myeloma Lesion Segmentation on Whole-Body Diffusion-Weighted Imaging via Efficient Anatomical Anticipation and Multimodal Confirmation

Whole-body diffusion-weighted imaging (WB-DWI) is widely used for multiple myeloma (MM) assessment, yet automated lesion segmentation remains challenging due to limited anatomical delineation and the low specificity of marrow hyperintensity. Existing studies have introduced bone region-of-interest (ROI) information and apparent diffusion coefficient (ADC) maps to mitigate these ambiguities, but practical limitations remain. Bone ROI construction often relies on costly manual annotation, image registration, or dedicated bone models, while ADC is usually incorporated only through simple channel fusion, limiting its ability to provide complementary structural and lesion-discriminative cues. To address these limitations, we propose a two-stage framework for MM lesion segmentation on WB-DWI. In the first stage, we train a bone ROI generation model from ADC images without dedicated bone labels, providing an efficient and practical anatomical prior for lesion analysis. In the second stage, we propose Anatomy-guided Multimodal U-Net (AMU-Net), which leverages ADC in a manner consistent with clinical lesion assessment rather than treating it as a generic auxiliary modality. Extensive experiments demonstrate the effectiveness and practicality of the proposed method. It achieves the best overall performance among the evaluated methods, with a mean Dice score of 76.2%.
Read the full abstract

All authors Mengmeng Zhang, Shengqian Huang, Junde Zhou, Xiaoping Wu, Hao Luog, Jing Wanga, Yicheng Sun, Jiao Li, Haibo Zhang, Sheng Xie, Fan Wangg, Qin Wangc, Huadan Xue, Yisheng Lv, Fei-yue Wang

Analysis of PDEs

Well–posedness for the continuum Calogero–Moser equations by parabolic regularization

We prove that the focusing and defocusing continuum Calogero–Moser equations are globally well-posed in $H^s_+(\mathbb R)$ for every integer $s\geq 3$. In the focusing case, this requires the initial data to satisfy the mass condition $$\|q_0\|_{L^2}^2<2\pi.$$ Our proof is based on a parabolic regularization of the equation, uniform a–priori estimates for the regularized solutions, and a Bona–Smith type approximation argument.
Read the full abstract

Systems and Control

Fixed-Time Integral Reinforcement Learning for Saturated Nonlinear Multi-Agent Systems Under FDI Attacks

The leader-follower formation control problem is investigated for nonlinear multi-agent systems with unknown dynamics, external disturbances, and false data injection (FDI) attacks on actuator channels. The problem is formulated as a zero-sum differential game and solved using the Integral Bellman-Isaacs approach. To address input saturation constraints, a non-quadratic control cost function is incorporated into the optimization problem, leading to a bounded control law. Furthermore, this paper proposes a cost function construction method and develops a critic learning law, which together guarantee the practical fixed-time stability of the system while overcoming the limitations of existing fixed-time reinforcement learning formulations. Finally, the practical fixed-time convergence of both the critic weight estimation error and the leader-referenced formation tracking error to bounded residual sets is rigorously proven. Simulation results demonstrate the effectiveness of the proposed method under external disturbances, FDI attacks, and input constraints.
Read the full abstract

Chemical Physics

Performance of a minimally empirical local-hybrid density functional for molecular chemistry

Delocalisation error has been argued to be the greatest outstanding challenge in density-functional theory (DFT). One of the most promising routes to minimise this error is development of local hybrid functionals, in which the fraction of exact-exchange mixing is position dependent. However, existing local hybrids capable of good thermochemical accuracy are often highly empirical, and tend to have complicated functional forms that involve some combination of range separation, calibration functions, power-series expansions, or even neural networks. In this work, we explore the limits of what can be achieved with a minimally empirical local hybrid functional form that avoids such complexities. The “LHnz” functional is proposed, which uses dispersionless exchange and a local mixing fraction dependent on the correlation length and effective exchange–correlation hole normalisations. With only three empirical parameters, LHnz is shown to outperform all existing global hybrid functionals for the GMTKN55 molecular-thermochemistry benchmark with no large outliers.
Read the full abstract

Machine Learning

All for 1-Bit: Towards Genuine 1-Bit Post-Training Quantization for LLMs

Large language models (LLMs) have achieved remarkable progress, yet their massive storage and memory-bandwidth demands still hinder efficient deployment. Weight binarization is a promising solution, but existing binarization-based post-training quantization (PTQ) methods usually far exceed the nominal 1-bit storage target due to hidden overhead. To address this gap, we propose All for 1-Bit (AF1), a genuine 1-bit PTQ framework for LLMs. AF1 comprises two complementary components: (1) Null-space-Aware Binary Factorization (NABF) for improving binary reconstruction through Hessian-aware surrogate reparameterization, null-space-aware binary factorization, and scale-only global reconstruction; and (2) Hierarchical Shapley Allocation (HiSA) for assigning structural capacity using hierarchical Shapley sensitivity. Together, they preserve model accuracy under a strict 1.0-BPW budget in the PTQ setting. Experiments on LLaMA, Qwen, and Gemma families show that AF1 consistently outperforms existing binarization-based PTQ methods in perplexity and zero-shot accuracy. Compared with BF16, AF1 achieves an average 2.5 times inference speedup and over 90% memory reduction across evaluated models, providing a practical path toward deployable genuine 1-bit compression for LLMs. The code for reproducibility is available at https://github.com/Kishon-zzx/AF1.
Read the full abstract

All authors Zhixiong Zhao, Zukang Xu, Guangyu Sun, Lifeng Liu, Dawei Yang

Complex Variables

Slice regular functions on alternative *-algebras : prescribing zeroes and values on discrete sets and related extension problems

Slice regular functions are a generalization of holomorphic functions where alternative real $*$-algebras are considered instead of the field of complex numbers. For such functions we show that zero sets and function values on suitable subsets may be prescribed. As a consequence, we show that for any axially symmetric domain there exist slice regular functions which (due to the nature of its zero set) can not be extended to a larger such domain.
Read the full abstract

Materials Science

Altermagnetism from the viewpoint of chemistry

Magnetism has been a central theme of research in chemistry, physics, and materials science, with chemical composition and bonding playing key roles in determining magnetic behavior. Altermagnets are a newly identified class of magnetic materials that combine features of conventional ferromagnets and antiferromagnets, arising from specific symmetry and electronic structure motifs. In this review, we present a chemistry-driven viewpoint on altermagnetism, highlighting how crystal chemistry, bonding, and electronic structure enable this unconventional magnetic order. We begin by introducing the fundamental concepts required to understand altermagnets, with an emphasis on symmetry considerations, orbital character, and electronic structure signatures. We then survey the diverse material families in which altermagnetism has been identified, drawing attention to coordination environments and structure-property relationships that favor altermagnetic order. We subsequently present experimental approaches which are useful for the characterization of altermagnetic materials. We examine ab initio materials discovery as a promising strategy for identifying new altermagnets, emphasizing how chemical constraints, such as symmetry and bonding, can guide computational searches. Other than their intrinsic importance, altermagnets provide interesting possibilities for technology. For this reason, we highlight possible applications that may be enabled through altermagnetic materials, along with their coupling with existing orders such as ferroelectricity and superconductivity. In conclusion, we point out some challenges and prospects, where chemically-based design guidelines can play an important role towards advancing altermagnetism research. In summary, this review offers an account of recent developments in altermagnetism, from basic concepts to the current state-of-the-art.
Read the full abstract

All authors Nayana Devaraj, Anumita Bose, Md Afsar Reja, Arka Bandyopadhyay, Awadhesh Narayan

Fluid Dynamics

Maintenance of a columnar vortex by inertial waves in rotational turbulence

We develop a theory that determines the radial profile of the mean velocity in a coherent geostrophic vortex forming in a turbulent rapidly rotating fluid. Following the conditions of our experiments, we assume that the flow in the vortex is sustained by the absorption of short-wavelength inertial waves arriving from the vortex periphery. The nonlinear interaction between waves is assumed negligible in the theory. Comparison with experimental data supports the validity of the developed model.
Read the full abstract

All authors Nikolay A. Ivchenko, Sergey S. Vergeles, Daniil D. Tumachev, Sergey V. Filatov

Hardware Architecture

From Splats to Silicon: Rethinking Computational Efficiency of 3DGS

3D Gaussian splatting (3DGS) represents scenes with explicit primitives and supports real-time novel-view synthesis, yet its system efficiency varies substantially across scenes, viewpoints, rendering paths, and platform constraints. Existing studies pursue efficiency through representation and algorithm design, GPU runtime optimization, and architectural support, but their reported gains correspond to different points along the rendering and update paths. Connecting these indicators to end-to-end system benefit requires tracing how each optimization changes Gaussian selection, screen-space work, data movement, and stage or frame time. We therefore use a workload-centric framework to connect representation and algorithm research, GPU runtimes, and hardware architectures and to identify recurring workload patterns. We complement literature analysis with reproduced measurements and controlled GPU profiling of selected implementations, relating workload counts to stage time and memory traffic. Together, these comparisons show that system gains depend on workload reductions reaching downstream execution, granularity matching each stage, and the cost of data transfers, synchronization, and cached results, gradients, and optimizer data. Building on these findings, we discuss more consistent evaluation under rendering-quality constraints and identify key directions for future system design.
Read the full abstract

All authors Minnan Pei, Qiwei Dong, Yihan Zhou, Gang Li, Yuchen Zhu, Wenju Zhao, Zhongtian Long, Siting Wang, Peisong Wang, Jian Cheng

Applied Physics

Project Setu: 3D Multi-Physics Design and Scaled Structural Analysis for a Relativistic Lightsail Architecture

Deep-space exploration beyond the solar system requires eliminating chemical propellant mass penalties to achieve relativistic flight velocities (0.166c at 180 s, reaching the mission target of 0.20c at 227 s). This study presents a 3D multi-physics numerical framework for a 4.0-meter circular lightsail propelled by a 100 GW ground laser array, coupling 3D Maxwell FDTD wave optics, non-linear membrane mechanics, and Stefan-Boltzmann thermal radiation in ANSYS Mechanical APDL and Ansys Lumerical. A four-level grid convergence study establishes numerical independence with an ASME GCI_21 of 0.13%, resolving peak membrane stresses of 530.88 MPa with a 3.77x safety factor against stoichiometric Si3N4 tensile failure. With optical absorption constrained to 10 ppm (A = 1.0 x 10^-5), the steady-state core temperature stabilizes at 923.02 K (0.44% deviation from radiation theory), maintaining a +1,247 K margin below sublimation, while fundamental drumhead modal resonance (7.92 Hz) provides a 7.92x safety buffer against laser jitter. The electrodynamic radiation pressure formulation is cross-verified against published flight telemetry from JAXA IKAROS and NASA LightSail 2 within 0.12% and 2.13%, confirming classical momentum transfer modeling across solar and beamed propulsion regimes.
Read the full abstract

Information Retrieval

ExpertLens: Visualizing Embedding Spaces for Post-Hoc Explainability in MoE Enhanced Retrievers

Neural models, including dense retrievers, have been widely adopted in Information Retrieval (IR), often delivering state-of-the-art performance. Despite their effectiveness, these models operate as black boxes, limiting the interpretability of their ranking decisions. Existing post-hoc explainability methods for neural rankers primarily focus on feature-level attributions, which can be insufficient to capture the complexity of learned embedding spaces. In this work, we propose ExpertLens, a post-hoc explainability framework for Mixture-of-Experts (MoE)-enhanced dense retrievers that shifts focus from local scalar feature importance to representation-level global interpretability. ExpertLens leverages discriminative embedding space visualizations jointly with automatically extracted Concept Activation Vectors to reveal how expert routing drives embedding space formulation and retrieval effectiveness. Experiments across five IR benchmarks and two MoE-enhanced dense retrievers show that expert routing consistently improves embedding space structure, positioning queries and their relevant documents into better-defined geometric neighborhoods. Analysis of expert subspaces further reveals general-purpose dominant experts, along with minority experts exhibiting distinct linguistic specialization, with subspaces arranged according to multi-semantic concept similarity. Our code is publicly available.
Read the full abstract

All authors Effrosyni Sokli, Isaac Roberts, Alexander Schulz, Barbara Hammer, Gabriella Pasi

Machine Learning

Rethinking One-Shot Federated Graph Learning: Training-Free Statistical Estimation

One-shot federated graph learning generally aims to train Graph Neural Networks (GNNs) across clients with disconnected subgraphs in a single communication round. Existing methods predominantly design advanced optimization strategies under the premise that local GNN training is indispensable. However, empirical observations reveal that under extreme non-IID conditions, local GNN training suffers from severe cross-client representation misalignment, becoming a major source of error rather than a remedy. Motivated by this, we reformulate one-shot FGL as a statistical estimation problem. We propose SPEAR (Statistical Prototype Estimation with Adaptive Reliability), a completely training-free framework that directly computes topology-smoothed class prototypes from local graphs in the original feature space. The server then aggregates these prototypes using a sample-size-adaptive shrinkage estimator that down-weights unreliable local estimates, producing robust global class prototypes. Extensive experiments across seven benchmarks demonstrate that SPEAR consistently achieves state-of-the-art accuracy under extreme heterogeneity. Moreover, SPEAR delivers at least an order-of-magnitude speedup over all baselines, reaching several orders of magnitude against generative and distillation-based methods. Our findings suggest that training-free statistical estimation, rather than local GNN optimization, provides the key to robust and efficient one-shot federated graph learning. The code is available at https://github.com/Yodeesy/SPEAR .
Read the full abstract

High Energy Physics (Phenomenology)

Exothermic dark matter and the 248 keV nuclear recoil in LUX-ZEPLIN

The single 248-keV nuclear-recoil candidate reported by LZ is reproduced by dark matter that downscatters, with the relic in the excited state of a pseudo-Dirac pair and a splitting \delta\simeq-350\keV. Every interpretation so far advanced is endothermic and places the splitting within a few percent of the xenon kinematic ceiling, so that the predicted count changes from zero to 143 times its central value as the halo escape speed runs from 500 to 600 km/s. The downscattering branch has no threshold speed; the same variation changes its count by 0.3 %. The ratio of counts below 70 keV to those above is 0.011, so the null standard search follows from the kinematics. A weak-strength coupling overproduces the event by 8.9\times10^{5}, so the mediator is a dark-sector one, and longevity of the relic excited state requires |\delta|<2m_e. Solving the coupled Boltzmann system removes the one free normalization. Rotating to the eigenchannels of the off-diagonal potential gives \sigma_0=(\pi/k^2)\sin^2(\delta_+-\delta_-), whose phase difference is 0.03 to 1.3 rad, so the Born rate exceeds the s-wave unitarity bound where it is used. One event then requires \sigma_p\simeq2\times10^{-42} cm^2 and a kinetic mixing \epsilon\simeq1\times10^{-6} at m_{A'}=1\GeV, with beam-dump coverage below 0.6\GeV pushing the mediator above that value. The ground state upscatters endothermically through the same operator, which excludes |\delta|\lesssim300 keV and makes the published endothermic reading the small-splitting limit of one model. The decisive test is argon, whose first form-factor zero lies at 695 keV against 94.6 keV for xenon; argon yields 4 to 6 events per tonne-year near 250 keV, where the endothermic reading is forbidden at any exposure.
Read the full abstract

Numerical Analysis

The Levin Method for the Summation of One-dimensional and Multidimensional Infinite Series

The Levin method transforms the evaluation of a highly oscillatory integral into the solution of a first-order linear ODE for a slowly varying auxiliary function. This ODE is typically approximated by collocation, after which the integral value is recovered from the auxiliary function at the endpoints. The present work develops a new extension of the Levin method for the summation of one-dimensional and multidimensional infinite oscillatory series. The summation problem is transformed into the solution of a functional equation involving transformed arguments of the unknown function. The resulting approach is particularly attractive in the multidimensional setting, where the range of existing numerical methods is relatively limited.
Read the full abstract

Theoretical Economics

Ripple Effects: Robust Norms without Punishment

I propose a novel, tractable model of pro-social norms in large communities with slightly altruistic agents. Agents participate in the norm to influence others to participate in the norm, influencing further agents. A continuum of equilibria sustain the pro-social norm, varying in the distribution of how much agents are influenced by their observations. If agents' effective altruism $\alpha$ is larger than their impatience $1-\delta$, equilibria exist that are robust to a population of bad actors. Greater strategic homogeneity increases the robustness of the norm. Robustness is not improved by increasing the number of observers of each action beyond 1.
Read the full abstract

High Energy Astrophysical Phenomena

Implementation of a five-wave approximate Riemann solver for general relativistic magnetohydrodynamics in AthenaK

We present an implementation of an HLLD approximate Riemann solver for the AthenaK astrophysics code with support for full general relativistic magnetohydrodynamics via a tetrad frame transformation. Our implementation uses an initial guess for the HLLD iterative solve which eliminates the need for an additional conserved-to-primitive inversion, which greatly accelerates performance without affecting accuracy. Additionally, by coupling the method with a first-order flux correction, we are able to use the method reliably even when the magnetization exceeds $10^4$, which we achieve in a SANE accretion disk. Our SANE disk shows that HLLD leads to a more strongly magnetized funnel and more accurate horizon fluxes when compared with HLLE. We further apply the new HLLD implementation to an equal-mass binary neutron star merger. For our long-lived remnant, HLLD enhances the magnetic shear stresses in the outer layers and leads to weaker differential rotation. However, due to weaker gravitational wave emissions in the post-merger phase, the remnant is consistently less compact while producing more dynamical ejecta and a more massive disk. The cost of this new solver is relatively modest thanks to the improved initial guess: our accretion disk tests are only ${\sim}10-25\%$ slower than HLLE, and for our binary neutron star runs with a microphysical equation of state, we find that HLLD is only ${\sim}3\%$ slower than HLLE across all runs.
Read the full abstract

Cryptography and Security

SCRIPTIOC-BENCH: A Benchmark for Recognizing Actionable Threat Intelligence from Script-Based Malware using LLMs

Script-based malware remains a prevalent attack technique. These scripts often contain indicators of compromise (IOCs) that provide actionable threat intelligence. However, statically recovering such indicators is challenging, as relevant values may be dispersed or transformed within code. Although large language models (LLMs) have shown promise in security analysis, their ability to recover IOCs from malicious scripts remains underexplored. We present SCRIPTIOC-BENCH, a benchmark for measuring static IOC extraction capability on real-world malicious scripts. The benchmark comprises 634 manually verified JavaScript, PowerShell, and VBScript malware samples covering four IOC types (URLs, domains, IP addresses, and filesystem artifacts). We further stratify ground-truth IOCs by recovery level, distinguishing directly exposed indicators from those requiring decoding or reconstruction. Using this benchmark, we evaluate a broad range of proprietary and open-weight LLMs and show that IOC recovery without execution remains challenging across model scales: the strongest model reaches only 65.4 F1. To characterize how recovery fails, we introduce a false-positive taxonomy and use it to compare the error profiles of the evaluated models. We further study two mitigations on a small open-weight model, deterministic string utilities and task-specific adaptation, finding that they provide complementary recovery gains, raise precision, and shift errors toward sample-grounded mismatches.
Read the full abstract

All authors Hanna Kim, Jian Cui, Minkyoo Song, Hwanjo Heo, Seungwon Shin, Kimin Lee, Xiaojing Liao