Peer Revenue Now accepting positions

Papers

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

All of arXiv

Monday 9 November

Statistical Mechanics

A Hierarchy of Fast and Strong Signatures for Vertex Orbits

The defining features of a physical system are its symmetries and its invariants. For a network, the symmetries are the automorphisms, and the invariants are the vertex orbits they define. Two vertices of a graph may appear structurally indistinguishable without being interchangeable under any graph automorphism. Distinguishing apparent similarity from genuine symmetry lies at the heart of the vertex-orbit partition problem. In this work, we develop a hierarchy of increasingly expressive, automorphism-invariant signatures that draw on the spectral properties of the graph Laplacian and the Dirac operator to reveal subtle structural differences. Guided by a sequence of counterexamples, we progressively strengthen these signatures, ultimately incorporating probing techniques, which are temporary structural perturbations, that expose the hidden distinctions in the original unperturbed graph. We evaluate our approach across diverse benchmark families, recovering exact orbit partitions in challenging instances, including graphs that resist low-dimensional Weisfeiler–Leman refinement. We also examine the computational and numerical limitations of our constructions, distinguishing empirical success from theoretical completeness. Our findings highlight the potential of spectral and probing-based methods for uncovering the structural distinctions that define vertex orbit.
Read the full abstract

All authors Joe Germany, Mohamad Al Sayed, Rana Dbouk, Amer E. Mouawad, Sara Najem

Mesoscale and Nanoscale Physics

Compensation Domain Walls in Ferrimagnets as Spin-Torque Oscillator Candidates

Spin-torque nano-oscillators are promising nanoscale microwave sources, yet their performance is often limited by narrow operational ranges and sensitivity to device parameters. Here, we numerically investigate the dynamics of compensation domain walls in ferrimagnetic systems driven by spin-transfer torque as a microwave source candidate. These walls arise in structures with adjacent regions of different compensation character, where the competition between exchange and Zeeman energies under an applied magnetic field stabilizes a distinct magnetic texture. We demonstrate steady current-driven rotation of compensation domain walls over a broad range of current densities within the model. To elucidate the underlying mechanism, we develop a collective-coordinate model and derive an analytical expression for the oscillation frequency, which quantitatively reproduces the dependence on current and magnetic field observed in micromagnetic simulations. Our results identify compensation domain walls as a candidate platform for tunable spin-torque nano-oscillators with intrinsic field-induced confinement.
Read the full abstract

Applied Physics

Neutronic Field-Effect Transistor

Field-effect electronic devices are in widespread use in communication, computing, data storage, and sensing applications. We demonstrate a neutronic field-effect transistor (NFET) that is analogous to the Datta-Das electron spin transistor and atomtronic devices. The NFET is based on a perfect-crystal Mach-Zehnder neutron interferometer with phase shifts applied to the spin and path degrees of freedom.
Read the full abstract

All authors Dusan Sarenac, Charles W. Clark, Owen Lailey, Muhammad Arif, Wangchun Chen, David G. Cory, Michael G. Huber, Joachim Nsofini, Dmitry A. Pushin

Graphics

Explicit LOD-Conditioned Anisotropic Filtering for Neural Texture Compression

Neural texture compression (NTC) stores material textures compactly, but the dominant feature-pyramid design receives no filtering from the GPU texture unit: it learns each material at a fixed UV and level of detail and has no explicit anisotropic-filtering (AF) mechanism. Setting the filter from the footprint aspect ratio alone is not enough: a single input cannot carry both the shape of the footprint and its size, so the image is too blurry near the camera by the same amount at every distance. Our method, AF-NTC, instead conditions the decoder on the same three bounded values a hardware sampler uses: minor-axis scale (LOD), aspect ratio, and major-axis orientation. These drive a tap line along the major axis and a level blend across the minor axis, and we train against anisotropically filtered ground truth. At $25\%$ of a full BC7 mip chain, AF-NTC beats the no-AF trilinear-BC7 baseline on all ten material–scene pairs we compare ($+1.3$ to $+7.6$\,dB), and it is $0.0$ to $3.4$\,dB below the $4\times$-larger anisotropic-BC7, depending on the material.
Read the full abstract

Atmospheric and Oceanic Physics

A Dual-Frequency-Constrained Structure-Function Kirchhoff Model for Near-Nadir Ocean Backscatter – Part II: The Practical GMF, KuKa-Mod

Near-nadir radars, from altimeters and rain radars to wave scatterometers, need a geophysical model function (GMF) relating backscatter to wind, yet most operational models are empirical and specific to one band. This paper delivers KuKa-Mod, an operational near-nadir GMF for wind retrieval at Ku and Ka band, built on one physical description of the sea surface, and evaluates it. Following the companion paper (Part I), the additive coefficient and log-slope of a logarithmic surface structure function are fitted jointly to the Ku-band incidence profile of the Global Precipitation Measurement (GPM) radar and to the same-footprint Ku-Ka shape difference. The physical core reproduces the Ku profile and reaches the azimuth-averaged Ka profile with no additional shape parameter and one fitted level, largely explained by the seawater dielectric and the clear-sky attenuation. Against scatterometer winds the Ka-Ku difference holds to within 0.3 dB up to 25 m/s, whereas the near-nadir Ka wind sensitivity is about 20% too weak. An empirical correction sets the operational level of both bands, with held-out quantile agreement over 8-20 m/s in the tested near-nadir bins. Without a sea-state coefficient, a long-wave tilt predicts where the response to significant wave height changes sign at Ku band, a tilt that CFOSAT/SWIM qualitatively supports, and the azimuth harmonics are reproduced once a sensor-dependent offset is removed. The physical core is released as parameters and an evaluator, and the operational levels as look-up tables.
Read the full abstract

Cryptography and Security

Maurer-Inspired Two-Way Physical-Layer Authentication Against Multi-Antenna Impersonation Attacks

In this work, we investigate Maurer-inspired two-way physical-layer authentication. Alice seeks authentication by Bob in the presence of a multi-antenna impersonator, Trudy. Trudy may replace the challenge to Alice, the response to Bob, or both. To create an information advantage, Bob adds artificial noise (AN) to his challenge, and Alice adds spatial AN while forwarding the received mixture. Bob predicts the response from trusted preparation, whereas Trudy observes only a noisy version of the challenge, and Alice's AN interferes with Trudy's preparation reception. Impersonation thus requires matching Bob's fresh randomness beyond the channel consistency checked by public-pilot one-way authentication. For quasi-static Gaussian multiple-input multiple-output channels with time-division duplex (TDD) reciprocity, we derive missed-detection bounds for attacks that replace the authentication challenge, the response, or both. These bounds assume attack-free initial training and preparation and account for estimation error and AN leakage. With equal antenna numbers at all nodes, we show scenarios where the proposed scheme has lower worst-case missed detection than specified one-way tests with/without AN.
Read the full abstract

Statistical Methodology

Near-Purity Screening and Dominant-Atom Estimation in Softmax Mixtures and Their Ensembles

We study near-purity screening and dominant-atom estimation in softmax mixtures, whose components share a common set of category features. A mixture is called $\varepsilon$-pure if one of its mixing weights is at least $1-\varepsilon$. We introduce the Trace of Latent Covariance Criterion ($\mathrm{TLC}^{2}$), a direct estimator of the trace of the covariance matrix of the latent mixing measure, constructed without estimating its atoms or weights, or knowing the number of components. Unlike existing moment-based approaches, our estimator is constructed for the softmax family and does not require knowledge or estimation of the feature distribution. We establish finite-sample guarantees for $\mathrm{TLC}^{2}$, showing that it achieves a parametric convergence rate up to an additive approximation error of order $p^{-1/2}$, without requiring minimum separation between mixture atoms. We also study estimation of the dominant atom in near-pure mixtures by fitting a single softmax component. We prove that the resulting maximum likelihood estimator has an error bound consisting of the usual parametric term and a bias term governed by the weighted distances between the dominant and remaining atoms, revealing adaptation to the geometry of the mixture. Finally, we extend near-purity screening and dominant-atom estimation to ensembles of softmax mixtures with shared atoms and mixture-specific weights. We apply $\mathrm{TLC}^{2}$ separately to each ensemble member to screen for near-purity without knowing the number of components. Under additional assumptions, clustering the resulting dominant-atom estimators yields finite-sample guarantees for estimating the shared atoms. An application to the Million Song Dataset illustrates how the estimated softmax directions reveal interpretable structure in the data.
Read the full abstract

All authors Xin Bing, Florentina Bunea, Marten Wegkamp, Weicheng Wu

Software Engineering

Surrendering and Reclaiming Judgment: How Software Engineering Students Experience AI Reliance

Software architecture is one of the skills novice developers must master to succeed in industry, and as AI disrupts software development, it is also transforming how students learn architecture. Existing work on how students and novices work with AI has mostly focused on small programming exercises or on students' perceptions of AI, so little is known about how students experience AI as they learn software architecture in larger team projects. We report on an exploratory case study of 43 senior software engineering students who used AI throughout a semester-long, team-based software architecture project. We combined surveys with focus groups and analyzed the focus group transcripts and the open survey responses through Braun and Clarke's reflexive thematic analysis. We constructed two themes. The first, Surrendering Judgment to AI, traces how participants' reliance on AI deepened as they used it to architect and implement their systems, until they surrendered their judgment and became stuck in a codebase they had created but could not understand, which made them rely on AI even more. Participants described feeling ashamed, frustrated, and emotionally disconnected from their work, a toll that, in our reading, left them with even less reason to break their reliance. The second, Applying Human Judgment, centers on participants who kept or reclaimed their judgment: the knowledge and experience it rested on, and the strategies they used to apply it. Participants' accounts suggest that relaxed deadlines and a lecture and lab on working with AI gave them room to acquire knowledge and reclaim judgment. Given the expectations placed on junior developers and the widespread adoption of AI, we believe these first-hand accounts can help software engineering education and industry adopt AI in ways that preserve novice developers' judgment, ownership, and motivation.
Read the full abstract

All authors Nathan Cassee, Umit Akirmak, Luiz Guerra, Margaret-Anne Storey, Neil Ernst

Functional Analysis

The vector-valued Stein inequality and dimension-free R-sectoriality

For a complex Banach space \(X\), we characterise dimension-free \(R\)-sectoriality of the Walsh operator, the Ornstein–Uhlenbeck operator, the Laplacian, and the Poisson operator by the vector-valued Stein inequality. For each family, the dimension-free \(R\)-bound of the resolvents on the negative real axis equals the Stein constant. Consequently, the resolvent estimate on the negative real axis already implies uniform \(R\)-sectoriality at an angle strictly smaller than \(\pi/2\). We furthermore prove that the full and dyadic Stein properties are equivalent and that they are independent of the exponent \(1<p<\infty\) entering their definition.
Read the full abstract

Signal Processing

Coexistence of Direct-to-Device (D2D) Satellite Systems with 5G/6G Terrestrial Networks: Cross-Border Aggregate Interference Analysis in Saudi Arabia

Direct-to-device (D2D) satellite systems, particularly those based on 3GPP-compliant non-terrestrial networks (NTN), deliver connectivity to standard mobile phones and IoT devices using terrestrial International Mobile Telecommunications (IMT) spectrum. This raises the risk of harmful cross-border interference into incumbent terrestrial IMT networks, an issue considered under Agenda Item 1.13 of the ITU World Radiocommunication Conference 2027 (WRC-27). This paper presents an ITU-R-compliant assessment of aggregate downlink interference from two hypothetical D2D systems operating in Jordan and Iraq into terrestrial IMT user equipment (UE) in Saudi Arabia, operating in the 698-960 MHz band. Unlike prior studies that evaluate a few test points near a simplified ideal tri-border intersection, we distribute multiple victim UEs along realistic national borders within various strips at a distance of 0-40 km from the borders. We compare two satellite selection strategies: highest-elevation and random strategies. The CDF of the interference-to-noise ratio (I/N) together with the intra-system and inter-system aggregation factors are analyzed. Results show that the protection criterion is exceeded up to 88% of the time, requiring up to 27 dB of additional isolation, so coexistence is infeasible without interference mitigation. In general, random selection produces higher interference because low-elevation beams place victims closer to boresight. Two mitigation techniques are then evaluated. A border isolation zone of one cell diameter yields only a moderate improvement under both strategies, whereas a narrow-beam, deep-null phased-array pattern lowers the exceedance to about 40% irrespective of strategy, providing an important input for spectrum regulatory considerations.
Read the full abstract

All authors Muneer M. Alzubi, Muhammad Alrabeiah, Khalid Aldubaikhy, Abdullah Alqasir, Majed Alsifaee, Mohamed-Slim Alouini

Machine Learning

Simulation-aided Reinforcement Learning with Control Variates

Reinforcement Learning (RL) algorithms often rely on data collected from a target environment, which can be expensive and noisy. Although simulators can generate large amounts of additional data, standard simulation mixing approaches, which directly combine simulated and real experience, can degrade performance due to simulator inaccuracies, especially when they bias the solution away from the optimal policy. We propose a conservative simulation-aided learning method based on control variates, where simulated data is used solely to reduce variance in the learning process rather than directly shaping the learning target. We develop our approach for Policy-Evaluation, which is a fundamental building block of RL algorithms. Then we combine it within the actor-critic framework towards policy improvement, as well as extend it towards offline RL scenarios. Experiments on multiple MuJoCo benchmark tasks demonstrate improvements over standard simulation mixing approaches, highlighting control variates as an effective tool for simulation-aided reinforcement learning.
Read the full abstract

All authors Omer Hemo, Ido Sefi, Mirco Mutti, Aviv Tamar, Kfir Y. Levy

Discrete Mathematics

Non-Uniqueness in the Coordinate-Based Normalization of Tetrahedral Similarity Classes

Recently, a new coordinate-based normalization has been introduced in [1] to represent tetrahedral similarity classes by normalized coordinates, enabling the study of LEB orbits as a discrete dynamical system. However, the uniqueness of this representation is not guaranteed when two or more edges adjacent to the selected longest edge have the same minimum length and remain admissible under the normalization criteria. In this paper, we demonstrate that different admissible choices may lead to distinct normalized coordinate representations of the same tetrahedral similarity class. We first present an explicit counterexample in which two admissible vertex labelings yield different normalized coordinate representations related by reflection, despite corresponding to the same similarity class. We then provide a systematic classification of the nine possible edge configurations arising in the selection of the canonical face when the longest edge of the tetrahedron is unique, and analyze the corresponding geometric and algebraic conditions. Our analysis identifies the configurations in which the normalization is non-unique and distinguishes them from those in which the ambiguity is resolved by the length criterion, angle criterion, or geometric symmetries. We further show that this non-uniqueness leads to an overcounting of distinct similarity classes.
Read the full abstract

Probability

Continuum hierarchical Cannings process: I. Construction, duality and dichotomy

We construct spatial measure-valued Cannings processes on the continuum hierarchical group $\Omega_N^\infty$ of order $N\in\N$ (an uncountable topological group with an ultrametric distance). These processes arise as the continuum-space limit of spatial Cannings processes on the discrete hierarchical group $\Omega_N$ of order $N\in\N$ (a countable topological group with an ultrametric distance) analysed in \cite{GHKK14}. On $\Omega_N$, each site represents the geographic location of a colony of individuals, each carrying a type drawn from a compact type space $\K$, and the state in each colony is a probability measure on $\K$ describing the type distribution in that colony in the large-colony-size limit of individual-based models. The continuum-space processes on $\Omega_N^\infty$ take values in the set of locally finite measures on $\Omega_N^\infty \times \K$ and are to be thought of as describing a population of individuals, each equipped with a geographic location in $\Omega_N^\infty$ and a type in $\K$, and each being subjected to migration and three mechanisms of resampling: (1) Fleming-Viot resampling at single sites; (2) Cannings resampling at single sites; (3) Cannings resampling and uniform reshuffling in balls around single sites (corresponding to consecutive hierarchical levels). The reshuffling adds to the migration, resulting in an effective migration that is the superposition of the two relocation mechanisms.
Read the full abstract

Artificial Intelligence

TrialDesignBench: Benchmarking Language Models for Phase-I Trial Design Against Clinical Knowledge

AI is used in much of the clinical trial pipeline. However, selecting the statistical design of a trial, a crucial step that determines how drug doses are tested on humans and evaluated, has seen almost no AI progress. LLMs appear to be a natural fit given their strengths in biomedical knowledge and statistical reasoning. Nevertheless, there is currently no appropriate way to evaluate such abilities because recent regulatory and methodological shifts, most notably FDA's Project Optimus, have redefined what constitutes a reasonable design and rendered many previous designs obsolete. To lay the groundwork for AI-assisted trial design, we introduce TrialDesignBench, a benchmark of 617 Phase-I/I-II oncology dose-finding trials, each paired with an expert-validated reference design that is consistent with the latest FDA regulatory initiatives. This reference standard is produced through a formalization of the expert design process. We evaluate five frontier LLMs, none of which matches the reference standard. Our ablation experiments trace this performance gap to two main causes, inference of latent, unrecorded design determinants from trial sources (supplying them raises class-wise balanced accuracy by 0.22) and structured decision-tree reasoning (+0.24). TrialDesignBench offers a reproducible, contemporary-standard objective for evaluating and enhancing AI-assisted trial design.
Read the full abstract

All authors Yihan Hank Tang, Peng Yang, Jiayi Xin, Tianqi Shang, Xuyang Chen, Ying Yuan, Yi Lian, Qi Long

Networking

Perception-Limited UAV Positioning: On the Reliability of Line-of-Sight Constraints in Sensing-Driven Aerial Networks

Unmanned Aerial Vehicles carrying access points or base stations are widely positioned so as to hold line-of-sight links with every ground user at once. The algorithms that do so are evaluated against obstacle information treated as exact, whether that information comes from a surveyed map or, more recently, from onboard vision. This paper analyses what becomes of that requirement when the obstacle information is estimated rather than given, and derives the detection accuracy it implicitly demands. Three results follow. The probability of satisfying the requirement decays geometrically with the number of users, so that a detector accurate enough for one user is not accurate enough for a network. Spatially correlated detection error raises that probability but concentrates failures into systematic blind spots. Relaxing the requirement from all users to ninety per cent of them widens the tolerable detection error by more than an order of magnitude, far beyond what any plausible improvement of the detector could deliver. The requiremen
Read the full abstract

Networking

FAROS: Forecast-Aware Risk-Informed Online Reinforcement Learning for Proactive O-RAN Slicing

Proactive Open Radio Access Network (O-RAN) slicing requires decisions before an approaching quality-of-service (QoS) loss becomes difficult to reverse. We present FAROS, a Forecast-Aware Risk-Informed O-RAN Slicing framework that combines an eight-step QoS forecast, training-only PCMCI+-supported temporal risk, Multi-Agent Proximal Policy Optimization (MAPPO) under centralized training and decentralized execution (CTDE), symbolic feasibility constraints, and reliability-aware Top-K arbitration. MAPPO supplies the actor proposal, while A5v2 checks admissibility and one-step utility before execution. Evaluation uses ten validation-selected frozen checkpoints, 500 matched start indices per seed under the risky-start protocol, and 150,000 agent-level decisions. FAROS obtains mean reward 0.2778 and ten-step recovery 90.56% across ten seeds. A fixed model-matched diagnostic rule reaches reward 0.3015 and recovery 97.57% under the same transition-reward contract, so we make no superiority claim over direct threshold logic. A same-actor M4 shield control also reaches zero risky-state prohibited actions. In the M5 trace, A5v2 changes 22.61% of actor proposals and reduces the predicate-defined prohibited-action rate from 16.79% at proposal time to zero at execution; this property is enforced by the arbiter rather than learned by MAPPO. Independent ns-3 experiments test the six-action interface at packet level and remain separate from end-to-end FAROS claims.
Read the full abstract

All authors Peyman Eftekhari, Seyyed Kasra Mortazavi, Reza Farahani, Robert Legenstein, Kyandoghere Kyamakya, Martin Gebser

Earth and Planetary Astrophysics

The Swift/UVOT Observation Campaign of Comet C/2013 US10 (Catalina) and Near-nucleus UV Excess

We present results of a multi-wavelength extended monitoring campaign of dynamically new long-period comet C/2013 US$_{10}$ (Catalina), captured with the Ultraviolet and Optical Telescope (UVOT) on board the Swift spacecraft covering 5.8 AU inbound to 5 AU outbound. The pipeline and methodology developed for these results is introduced, along with derived water production rates, active areas, and dust content proxy $A(\theta)f\rho$. We describe a Bayesian solution to address the unknown spectral gradient $S'$ of the dust of C/2013 US$_{10}$ necessary for our continuum subtraction calculations. The comet exhibited significant hyperactivity, with the water production rate approaching $2.5 \times 10^{29}$/s and a greatly elevated active area pre-perihelion, settling to approximately 50 km$^2$ post-perihelion. We also report and quantify a near-nucleus UV excess beyond vectorial model predictions, a feature we have dubbed the Pale Blue Spot (PBS), and cover possible candidate species responsible. When interpreted as OH (A-X) band fluorescence, the PBS may point to a secondary release mechanism producing slow-moving hydroxyl radicals that is active even beyond the water snow line.
Read the full abstract

All authors Shawn Oset, Zexi Xing, John Wm. Noonan, Steven J. Bromley, Dennis Bodewits

Cryptography and Security

BioBind: Binding Biometric Deduplication to Zero-Knowledge Authentication

Biometric identity systems for decentralized applications need credentials that resist repeated enrollment and support private, publicly verifiable authentication. Capture and large-scale private deduplication run off chain while smart contracts authorize issuance and use; these stages span distinct trust domains, and a correct verdict covers only the checked inputs and gallery snapshot. Without binding and state checks, concurrent enrollments can issue duplicate credentials from correct verdicts, while substituted inputs can authorize unrelated credentials. We present BioBind, a protocol whose four backend interface conditions bind the deduplication query, gallery insertion and issued commitment to one captured vector, authenticate gallery transitions, and require a sound uniqueness verdict valid at finalization. Its encrypted instance uses bound requests, stale-session rejection and committee-confirmed migration. Capture-bound zero-knowledge authentication links credential use to the enrolled template and a fresh device-originated probe bound to the challenge. We give a conditional multi-session composition argument under explicit capture, backend and cryptographic assumptions. A prototype using LFW face features rejects the tested composition attacks; its unlinkable authentication circuit uses 31,445 constraints and about 305,000 gas. An equal-public-field baseline quantifies the cost of verifiable matching: witness-plus-proving medians are 1.989 s versus 1.119 s with device-trusted matching. Separate tests of encrypted enrollment complete three histories of 16,384 registrations without decoding errors at the tested checkpoints. Biometric accuracy bounds person-level uniqueness: on the Bosphorus database, cooperative subjects who change appearance evade deduplication in 1.8% of re-enrollment attempts at the operating threshold. Full backend integration remains open.
Read the full abstract

Artificial Intelligence

Beyond Sunny Skies: Using Large Language Models to Predict Urban Traffic in Adverse Weather

Urban traffic flow prediction is a cornerstone of intelligent transportation systems, yet accurate forecasting under adverse weather conditions remains a significant challenge. Existing data-driven models, while effective for routine patterns, often struggle with the sparse data and complex non-linear shifts associated with rare meteorological events. These methods typically treat environmental contexts as shallow numerical features, failing to capture the underlying semantic reasoning-why and how specific adverse conditions alter mobility behaviors across regions with different functional compositions. To address this limitation, we propose Multi-STG, a unified framework that integrates Large Language Model (LLM)-derived semantic priors with graph-based spatiotemporal learning. Multi-STG introduces an LLM-empowered semantic enhancement module that transforms comprehensive context descriptions such as weather, POIs, and holidays into high-level semantic embeddings, effectively providing a knowledge-driven complement to data scarcity. Furthermore, we design a multimodal attentional fusion mechanism to dynamically weigh heterogeneous inputs and a semantic-enhanced prediction head with gated residual calibration to robustly adjust forecasts based on semantic guidance. Extensive experiments on real-world multi-city private car datasets demonstrate that Multi-STG outperforms state-of-the-art baselines, reducing MAE by 11.82% on average relative to the strongest baseline across three cities, while maintaining stable performance under sparse and non-stationary conditions.
Read the full abstract

All authors Zhu Xiao, Kuan Feng, Tong Li, Chenxi Liu, Gang Wang, Zhuo Tang, Kenli Li

Computation and Language

How Far Does Second-Order Linear Response Predict Quantization Mismatch? Pre-Registered Measurements on Five Language Models

Serving systems choose a weight format for each model and deployment. A low-cost screen predicts the resulting KL divergence from the unquantized model's Fisher information using $\tfrac12\Delta^\top F\Delta$, the expansion underlying Hessian-aware and second-order post-training quantizers. We assess the reliability of this screen in GPU batches with channel-wise int8, per-tensor fp8 and group-wise int4 on Qwen3-8B, Qwen3-4B, OLMo-2-7B, Qwen3-32B and Qwen3-30B-A3B. Relative predictions hold across a 16-fold range of perturbation scales, int4 group sizes and rounding modes, while absolute predictions are heavy-tailed. A trimmed average brings 10 of 12 cells on Qwen3-4B and OLMo-2 within 15% of measured KL. The MoE model exhibits nonquadratic response at small perturbations, with int8 KL scaling as $\alpha^{0.63}$. Computing per-sequence gradients in bf16 can cause fourfold prediction errors. We trace these errors to arithmetic in the lower half of the backward pass and eliminate them by using fp32 there. Batch-invariant kernels run at 0.55 times the throughput, while changes in reduction order introduce a floor of 0.17-0.22 nats per sequence. We build a screener with one streamed fp32 Fisher pass, trimmed tensor-block predictions with a band, a flag for aligned tensors and per-layer format allocation. On the unseen Mistral-7B-v0.3 family, all six predictions fall within the band, all three KL budgets are met, and vLLM throughput is predicted within 0.4%.
Read the full abstract

Statistics Applications

Modeling inter-exceedance times of extreme heat days

Extreme heat days, here defined as exceedances of a local calendar-day-based 95% quantile, often occur in clusters forming multi-day heatwaves. A calendar-day based threshold allows for the detection of temperature extremes across all seasons, which may impact several economic sectors, public health and ecosystems. Previous approaches to modeling the occurrences of extreme heat events typically treat heat event duration and the time between heat events separately. We propose modeling the inter-exceedance times (IETs) between extreme heat days, allowing for a unified description within a single model. We consider several candidate distributions for the IETs of extreme heat days and find that the zero-inflated discrete Weibull (ZIDW) distribution clearly provides the best fit according to Pearson's chi-squared goodness-of-fit tests. The ZIDW distribution is a flexible model for count data exhibiting excess zeros as well as over- and under-dispersion: the zero-inflation component captures the pronounced clustering of heat days, while the discrete Weibull component adequately accounts for over-dispersion. Heat days are extracted using both trend-adjusted and non-trend-adjusted thresholds. This allows the analysis of general changes in the occurrence of extreme heat days as well as changes after accounting for the general increase in temperature.
Read the full abstract

Differential Geometry

Perdomo's index conjecture for minimal hypersurfaces in spheres

Let $x:M^n\to\sphere^{n+1}(1)$, $n\ge2$, be a full minimal immersion of a closed, connected, oriented manifold. We prove that its Morse index is at least $n+3$, with equality precisely for the standard Clifford embeddings. For a non-Clifford image, the constant function and the position and normal coordinates span a space containing an $(n+4)$-dimensional subspace on which the Jacobi quadratic form is negative definite. Thus we resolve Perdomo's index conjecture.
Read the full abstract

Functional Analysis

Twisted convolution with conic support and the parabolic algebra

We prove that tempered distributions on \(\mathbb R^{2n}\) whose supports satisfy a common strict cone condition form a unital associative algebra without zero divisors under twisted convolution \(\star\). The supports may be noncompact, and the associated Weyl maps need not be bounded. As applications, we show that the parabolic algebra and the norm-closed algebra generated by positive translations and modulations have no zero divisors, confirming Levene's conjecture. We also obtain a zero-product theorem for Toeplitz operators on the one-dimensional Fock space with bounded symbols whose distributional Fourier transforms satisfy a common strict cone condition. The proof of the distribution theorem uses Schwartz smoothing between distributional Weyl maps to convert a hypothetical zero product of nonzero distributions into a smooth function \(h\not\equiv0\) satisfying the same cone condition and \(h\star h=0\). The cone bound implies the \(Q\)-support condition, meaning that some linear functional is bounded above on \(\operatorname{supp}h\) and has compact upper level sets there. Truncating \(h\) near the maximum of this functional and applying the support identity for compactly supported twisted squares rule out \(h\star h=0\). For the operator applications, uniformly bounded Weyl-conjugation averages identify twisted convolution with bounded operator composition under the common cone condition.
Read the full abstract

High Energy Physics (Phenomenology)

Angular analysis of the purely baryonic decays $\Lambda_b^0\to n p\bar p$ and $\Lambda_b^0\to\Lambda p\bar p$

An angular analysis of the purely baryonic decays $\Lambda_b^0\to np\bar p$ and $\Lambda_b^0\to\Lambda p\bar p$ is presented within a phenomenological generalized-factorization model. Within this model, two current contributions leading to the same final state are combined coherently. Pair-angle distributions, forward-backward hemisphere asymmetries, and Legendre moments are calculated in both channels, together with the polarization of the produced $\Lambda$ for an unpolarized parent. For the neutron channel, the full-region forward-backward asymmetries are $4.46\times10^{-4}$ in the $n\bar p$ frame and $0.856$ in the $p\bar p$ frame at $N_c^{\mathrm{eff}}=3$ and zero additional relative phase. From mass-bin integration, the small first asymmetry is found to arise from compensation between signed contributions from different regions. In the $\Lambda$ channel, the angular distributions are modified by the selection $m(p\bar p)<2.85\,\mathrm{GeV}$, and the rate-weighted longitudinal polarization along the $\Lambda$ flight direction is shifted from $-0.803$ to $-0.954$. These results show how angular and spin observables depend on the relative current contributions and the selected phase-space region. The model-dependent analysis is intended as a first step toward understanding the dynamics of these decays and as a basis for further theoretical studies and comparisons with measurements.
Read the full abstract

Optics

Frequency-selective membrane-based metamaterial filters for low-insertion-loss harmonic crosstalk suppression in sub-harmonic THz receivers

Parasitic reflections and standing waves fundamentally limit amplitude stability and phase fidelity in coherent continuous-wave (CW) terahertz (THz) systems. We report a complementary metamaterial filter pair designed for frequency-selective isolation at 300 GHz and 600 GHz in sub-harmonic heterodyne architectures. Fabricated on ultra-thin polyimide membranes, the filters provide more than 30 dB stopband isolation while maintaining insertion losses below 1 dB in the passband. System-level coherent-imaging experiments demonstrate effective suppression of spurious-reflection-induced interference effects and improved spectral baseline stability. The filters mitigate wavefront distortions resulting from such interferences and enhance reconstruction robustness. These results establish membrane-based metamaterial filters as compact frequency-selective isolation elements for sub-harmonic heterodyne THz detection and highlight their potential for phase-engineered imaging and sensing applications.
Read the full abstract

All authors Hui Yuan, Meng Zhang, Almila Kura, Daniel Erni, Hartmut G. Roskos

Computation and Language

Adapting Small Language Models to Romanized Nepali: Tokenizer Fertility, Reference Likelihood, and Generation Metrics

Romanized Nepali, Nepali written in Latin script, is common in digital communication but has no standard spelling and few resources, and it is unclear how reliably small open language models can be adapted to it. We adapt three 2-3B base models from different families, Llama-3.2-3B, Qwen2.5-3B, and Gemma-2-2B, with QLoRA on 9,000 instruction examples produced by deterministic transliteration of Nepali instruction data. Each model is trained with one shared recipe and three seeds and evaluated on 1,000 held-out examples with reference-overlap metrics, byte-normalized reference likelihood (bits per byte, BPB), and an output audit, against zero-shot baselines under two prompts. Fine-tuning raises chrF++ from 10-14 to 29.0-30.6 and lowers BPB from 2.23-2.41 to 0.78-0.84; most zero-shot outputs (54-92%) are English-like. Likelihood and overlap disagree on how far apart the models are. Llama-3.2-3B has the lowest BPB in every seed, 0.042 below Qwen2.5-3B (95% CI 0.039-0.046), yet the two cannot be separated on chrF++ (30.51 and 30.61) or BLEU, even with paired item-level tests. Gemma-2-2B, whose tokenizer splits Romanized Nepali words least (2.72 tokens per word), scores worst on every measure. Which model gains most over its zero-shot baseline depends on the zero-shot prompt. In a supplementary comparison with an identical tokenizer, Llama-3.1-8B lowers BPB by 0.053 (0.050-0.056) relative to Llama-3.2-3B, while chrF++ and BLEU rise by only 0.68 and 0.58 points. Lower fertility and higher reference likelihood therefore did not identify the model whose outputs were closest to the references.
Read the full abstract

All authors Pujan Pant, Sagar Newpane, Hari Sitaula, Manish Sakhakarmy

Neurons and Cognition

Spike-to-Field Mechanisms of Turbulence-Like Dynamics in Spatial Spiking Neural Networks

Neural avalanches and turbulence-like brain dynamics are usually measured after neural activity has been coarse-grained into continuous fields, leaving open which spike-level mechanisms can generate these macroscopic observables. We build a spatial leaky integrate-and-fire spiking neural network and explicitly transform spikes into activity-density and Hilbert phase fields. A delayed, distance-dependent, passive-dendrite regime reproducibly generates turbulence-like field structure, including high-amplitude phase-defect tracks, structure-function scaling, and power-law-favored continuous-field avalanche tails. Controls show that firing rate, vortex count, and avalanche tail fits are individually insufficient: destroying delayed propagation, dendritic filtering, or spike timing suppresses key field signatures, while random-space nulls can retain many mathematical phase defects but show weaker spatial structure and synaptic-delay transfer. Biologically motivated gates further shape the regime: PV-like perisomatic inhibition clamps phase dynamics, SST-like dendritic inhibition and projection-specific inhibitory facilitation preserve long-lived tracks, and a rate-constrained projection-STP regime expresses a roughly 200-ms episode at 12.25 Hz (full-record mean) across 12 seeds. These results use SNNs as a mechanistic testbed for spike-derived brain turbulence observables and show why phase defects and avalanche-like statistics must be interpreted jointly with spatial scaling and spike-level propagation.
Read the full abstract

Optics

Non-Equilibrium Ising-Bloch Transition in Polarization Cavity Solitons

Polarization cavity solitons in coherently driven Kerr resonators undergo a sequence of dynamical bifurcations as the cavity detuning is increased. Here we show that a stationary localized state containing an Ising wall first becomes a fixed breather and then develops finite chirality and drift motion through a spontaneous non-equilibrium Ising-Bloch transition. Stability analyses identify the modes responsible for the breathing and the motion, revealing controllable dynamics of polarization cavity solitons.
Read the full abstract

All authors Juan Diego Mazo-Vásquez, Lewis Hill, Flore K. Kunst, Gian-Luca Oppo

High Energy Physics (Phenomenology)

A singlet framework for the mu problem and type-II seesaw in a generalized NMSSM

We study a generalized NMSSM in which a gauge singlet links the Higgs bilinear to the type-II seesaw scale. The singlet vacuum expectation value contributes to $\mu_{\rm eff}$ and generates the supersymmetric mass of a vectorlike triplet pair. Integrating out the triplets gives the Weinberg operator, and their gauge charges admit a possible messenger interpretation. Large cancellation between the bare $\mu_0$ term and the singlet VEV generated bilinear Higgs term that keeps $\mu_{\rm eff}$ near the soft scale is still natural in the supersymmetric limit (by the supersymmetric non-renormalization theorem), although the concrete value of $\mu_{\rm eff}$ also depends slightly on the soft SUSY breaking parameters and finite threshold matching. The studied bino-like and singlino-like neutralino solutions accommodate triplet input scales approaching $10^{15}$ GeV together with electroweak-to-TeV dark matter. Representative spectra consistent with heavy-Higgs funnels or coannihilation with gluinos or winos yield allowed relic abundances while satisfying direct-detection constraints.
Read the full abstract

Computer Vision

Towards Reconstructing and Recognizing Objects behind Fragmented Occlusion using Video

Perceiving objects through fragmented occlusions, such as foliage, is a natural capability of the human visual system but remains a fundamental challenge for computer vision. This work addresses the problem of object reconstruction, or de-occlusion, in the presence of high-density, fragmented occlusions. We propose a reconstruction framework that operates on a sequence of grayscale images of a moving object, acquired by a single camera, either stationary or in motion. The algorithm estimates the object's relative positions across frames by jointly registering the images and optimizing the alignment parameters to maximize the contrast of their average, referred to as the integrated image. This integrated image serves as the reconstructed representation of the object. Experimental evaluations on both synthetic and real video sequences demonstrate the feasibility of accurate reconstruction under occlusion densities exceeding 90%. The proposed approach is competitive with synthetic aperture techniques, which are limited to static objects and require complex camera setup with careful calibration. Furthermore, we show that the integrated image enables reliable object classification under severe occlusion when used as input to standard, off-the-shelf classifiers.
Read the full abstract