Peer Revenue Now accepting positions

Papers

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

All of arXiv

Friday 6 November

Theoretical Economics

Moral Consumers and the Environment

We study the taxation of environmentally harmful goods when consumers have a Kantian desire to “do the right thing”. A consumer evaluates her buying strategy, a rule that specifies whether to buy, by the material payoff she would obtain if others followed the same rule with a probability given by her degree of morality. Equilibrium exists and is unique in finite populations with heterogeneous morality. The optimal tax depends crucially on whether consumers are aware that tax revenue is returned to them. If they are, the Pigovian tax is optimal under homogeneous morality, whatever consumers know about environmental costs, and we give conditions under which it remains the unique optimum under heterogeneous morality. If they are not, the optimal tax is lower, and more so the more moral consumers are. Under a quota system such as the EU Emissions Trading System, any cap at least as stringent as the Pigovian quota admits an equilibrium in which morality has no effect on consumers who are aware of the system.
Read the full abstract

Artificial Intelligence

A Physics-Informed Learning Framework for Linking Microscopic Surface Degradation to Bolted Joint Functional Behavior

Returned products in circular engineering often have heterogeneous degradation histories, making remaining function difficult to assess. For bolted joints, corrosion alters surface morphology, contact, friction, preload retention, and mechanical response. This paper presents a physics-informed multimodal framework for linking microscopic and camera-based degradation observations to functional indicators of corroded bolted joints. The workflow defines the corrosion failure mode, generates synchronized inspection and functional-response records, preprocesses image, roughness, exposure, and joint descriptors, aligns microscopy and camera observations through similarity learning, and constructs a calibrated corrosion-severity field. Partial differential equation (PDE) discovery is applied to this field to identify candidate effective mesoscale laws rather than complete electrochemical mechanisms. The corrosion-evolution model is coupled to a separate functional model, with friction coefficient as the primary target and preload loss included when clamp-force data are available. The contribution is a reproducible pathway for physically constrained assessment of degraded bolted-joint behavior.
Read the full abstract

Combinatorics

Mutual-visibility in (d, 5)-graphs with excess

The author investigates mutual visibility in regular graphs of girth at least 5, with particular emphasis on graphs with excess. A general upper bound on the mutual-visibility number is established in terms of the degree and order. More significantly, an algebraic inequality is derived that restricts the mutual-visibility number of $(d,5,+\delta)$-graphs. As applications, the mutual-visibility number of the Robertson graph is shown to be 9, and upper bounds are obtained for the $(5,5)$-cages and the $(6,5)$-cage.
Read the full abstract

Quantum Physics

Direct-decision quantum sensing for classifying transient NMR signals

Task-oriented quantum sensing aims to extract decision-relevant information directly, without first reconstructing the full signal. We study this approach for mixture-threshold classification of transient NMR-like signals generated from molecular spin Hamiltonians, motivated by small-volume molecular sensing. Using analytical models and numerical simulations, we compare classical coil readout followed by matched-filter classification with quantum computational sensing (QCS), in which trainable controls interleaved with signal-driven sensor-qubit evolution map the transient directly to a decision score. We consider both the sensor-ensemble limit and strict single-sensor single-shot readout. Increasing QCS control depth can improve robustness and class-dependent output contrast. Ensemble accuracy can rise rapidly for some waveform pairs, whereas single-shot correctness improves more gradually because one projective outcome must distinguish the classes. Repeated acquisitions improve both routes by averaging independent fluctuations, while disorder-aware training mitigates quasistatic inhomogeneity in the modeled sensor ensemble. QCS can outperform inductive readout under the modeled conditions, for example when coil transfer is weak or receiver noise limits coil performance. However, the simulated QCS protocols at lower effective sensor coupling require long sensing intervals, whereas short-time protocols require a much stronger effective drive at the sensor. Practical advantage therefore depends on jointly attainable source-signal strength, signal-to-sensor transfer, qubit coherence, and readout resources.
Read the full abstract

Computer Vision

V-STORM: A Structure-Aware Grounded Chain of Thought Dataset for Faithful and Fine-Grained Spatio-Temporal Video Reasoning

While MLLMs can often identify what happens in a video, they still struggle with fine-grained spatio-temporal reasoning, which requires precise grounding when and where the supporting evidence appears and composing such evidence into a coherent reasoning chain. This challenge is exacerbated by the lack of scalable data construction and evaluation pipelines that provide granular, verifiable supervision linking reasoning steps to explicit spatio-temporal evidence.To address this gap, we present V-STORM, a structure-aware framework that transforms videos into Spatio-Temporal Scene Graphs (STSGs) and derives Grounded Chain of Thought (GCoT) from validated evidence paths, making each reasoning step traceable to structured and verifiable spatio-temporal evidence.Under this framework, we construct a 260K-sample training set and a 3K-sample human-verified benchmark, both featuring GCoT annotations that link reasoning to explicit spatio-temporal evidence.Experiments reveal a clear separation between evidence localization and evidence composition: current MLLMs can often identify relevant spatio-temporal cues, yet still fail to integrate them into correct multi-step reasoning, exposing a bottleneck that answer-only evaluation largely obscures. Moreover, fine-tuning with V-STORM substantially improves both answer accuracy and grounding faithfulness, validating its effectiveness as supervision for faithful and evidence-grounded video reasoning. Our dataset, benchmark, and source code will be publicly released.
Read the full abstract

All authors Dian Jiao, Yuqian Yuan, Jiaqi Zhu, Mingjian Gao, Siwen Tan, Yunqi Cao, Jinhui Li, Sijing Li, Zhenxuan Fan, Haoyang Wang, Wenqiao Zhang, Siliang Tang

Superconductivity

Superconducting Fluctuation Driven Quantum Transport in Bi$_2$Sr$_2$CaCu$_2$O$_{8+\delta}$ Thin Films

Superconducting fluctuations (SF), arising from Cooper-pair-driven interactions and electron-electron interactions within the transition regime, can dominate the magnetoconductance (MC) of superconducting thin films near their transition temperature. In some systems, this fluctuation-dominated regime is preceded by a narrow window in which quantum interference effects are manifested as weak antilocalization (WAL) or weak localization (WL). Here, we report magnetotransport measurements on thin films of the high-temperature superconductor Bi$_2$Sr$_2$CaCu$_2$O$_{8+\delta}$ (BSCCO) that establish SF as the dominant mechanism governing MC across the transition regime. In the narrow temperature window well above $T_c$ where SF does not exist, a cusp in MC can be attributed to WAL-like signature well described by the Hikami–Larkin–Nagaoka (HLN) framework. Closer to the transition ($T_{BKT}$), we observe signatures consistent with Berezinskii–Kosterlitz–Thouless (BKT) correlations, underscoring that the dominant physics is set by collective, pairing-driven behavior rather than single-particle quantum interference. Measurements on non-superconducting control samples show no WAL-like features, indicating that the WAL over narrow temperature window, emerges only in conjunction with superconducting fluctuations. Hall measurements further reveal a concomitant sign reversal of the dominant carrier response, pointing to a substantial reorganization of transverse transport coincident with the onset of superconductivity. Together, these observations establish that magnetotransport in BSCCO thin film.
Read the full abstract

All authors Sohini Guin, Naresh Shyaga, Jagadish Rajendran, Aryaman Das, Subhransu Kumar Negi, Saisab Bhowmik, Pankaj Bhardwaj, Chandni Usha, Dhavala Suri

Materials Science

High-Volume Wurtzite MnSe in MnSe/CdSe Superlattices: Structure, Magnetism and Optical Response

Polymorphism in Mn-based chalcogenides offers access to magnetic semiconductors with distinct crystal symmetries, polar structures, and different magnetic orders. In particular, wurtzite MnSe exhibits an appealing interplay of electric polarization and several competing magnetic phases including altermagnetic order. However, stabilizing metastable polymorphs, such as wurtzite MnSe, in experimentally useful volumes remains challenging. Here, we show that a nanoscale-period MnSe/CdSe superlattice stabilizes wurtzite MnSe over micrometer-scale total thickness while suppressing the formation of the thermodynamically favored rock-salt phase. The increased magnetic volume enables bulk-sensitive characterization of this polar polymorph. The room-temperature lattice parameters of wurtzite MnSe are studied by independent experimental techniques and determined to be $a = (4.18 \pm 0.03)\,\mathring{A}$ and $c = (6.78 \pm 0.03)\, \mathring{A}$. SQUID magnetometry reveals antiferromagnetic ordering with $T_N = 62\,\text{K}$, which is accompanied by an anomaly of the MnSe lattice parameter, as oberved by temperature-dependent X-ray diffraction. Optical measurements reveal excitonic transitions pointing to the band gap at around 3.4 eV. These results highlight the utility and importance of heterostructure engineering for detailed studies of metastable phases of magnetic semiconductors, in particular wurtzite MnSe.
Read the full abstract

All authors Michał J. Grzybowski, Katarzyna Gas, Jaroslaw Z. Domagala, Serhii Kryvyi, Łukasz Gondek, Soumik Das, Jan Suffczyński, Oleksii Liubchenko, Anna Kaleta, Sławomir Kret, Maciej Sawicki, Wojciech Pacuski

Systems and Control

Characterization of the Frequency-Current Curve of First-Order High-Pass Neurons

A spiking model, called the High-Pass (HP) model, has recently been proposed as a minimal excitable primitive. This model is composed of a static activation function, usually a sigmoid, in positive feedback with a high-pass filter. The HP model enters a tonic spiking regime whenever its input lies in the high-gain zone of the sigmoid, where the sigmoid's slope exceeds 1. In neuroscience, the frequency-current (FI) curve is the standard way of characterizing the tonic spiking regime of a neuron. So far, the FI curve of the HP model has been characterized analytically only for specific sigmoids, and has otherwise been computed through simulation. These simulations are computationally heavy, since simulating neurons requires solving a stiff system of differential equations. This paper bridges this gap by analytically characterizing the FI curve of the HP model with a first-order high-pass filter, for a broad class of sigmoid functions. The FI curve of the first-order HP model can be computed by solving two input-independent equations and evaluating two input-dependent integrals, with no need to simulate the model itself. A spiking primitive is said to be Type I (resp. Type II) if tonic spiking starts at a zero (resp. non-zero) frequency. This paper also shows that, for the HP model, the type depends solely on the local shape of the sigmoid at the outer edges of its high-gain zone. A sharp corner, where the derivative approaches a value strictly smaller than 1 from the outside, yields a Type I primitive. A locally twice differentiable corner, with a bounded second derivative and a derivative approaching 1 at the edge, yields a Type II primitive.
Read the full abstract

All authors Christian Fernandez Lorden, Jérôme Adriaens, Alessio Franci, Gustave Bainier

Analysis of PDEs

Angular Besov Refinement of the Spherical Stein–Tomas Theorem

The spherical Stein–Tomas inequality maps $L^2(\mathbb S^2)$ into $L^4(\mathbb R^3)$. We study its refinement in which the angular input is measured by the strictly weaker Besov norm $B^0_{2,4}(\mathbb S^2)$. The central separated-band inequality bounds the squared $L^2$ norm of the product of extensions from degrees $L\ll K$ by $C(L/K)$ times the squared input norms. An exact spherical-Hankel expansion controls the far field, while angular-momentum leakage and transversality handle the Fresnel region. We construct scale-separated, normalized angular bands for which the extension norm is equivalent to the $\ell^4$ norm of the coefficients. This proves that the fourth-power summation is optimal and quantitatively distinguishes the result from the classical $\ell^2$ summation. As an application, small angular Besov data generate small $L^4$ solutions to the cubic nonlinear Helmholtz equation in $\mathbb R^3$, with a smallness condition that remains sharp on multiscale data even when their $L^2$ norm is arbitrarily large or infinite. The associated on-shell nonlinear scattering amplitude has improved angular $B^0_{2,4/3}$ regularity, and its leading cubic term is quantitatively decoupled across widely separated angular bands.
Read the full abstract

Computation and Language

On the Effectiveness and Limitations of Open-Weight LLMs for Low-Resource Offensive Language Detection in Danish

We study the effectiveness and robustness of open-weight large language models (LLMs) for Danish offensive language detection, an underexplored low-resource task. We evaluate four model families (Llama, Mistral, Gemma, and Qwen) across eleven prompting strategies spanning five design categories, cross-prompt generalization, Danish orthographic variation, and robustness under balanced and naturally imbalanced class distributions. We compare against three BERT-family baselines with distinct pre-training strategies, evaluated via both masked language model prompting and supervised fine-tuning. Even without task-specific fine-tuning, LLMs can surpass all three fine-tuned BERT baselines, and lightweight finetuning yields substantial further gains. No single model or prompt category dominates, underscoring that prompt design interacts with model pre-training in ways that cannot be predicted a priori. Semantically grounded prompts transfer more robustly, while format-constrained patterns consistently fail to generalize. ASCII transliteration of Danish-specific characters (æ, ø, å) generally degrades performance, indicating these characters encode meaningful signal. At the same time, the gap between a balanced test set (best F1 of 0.84) and the corpus's natural class skew (0.62 trained on skewed data, 0.70 trained on balanced data) – relative drops of 26.2% and 16.7% – shows that class imbalance remains a critical open challenge. Our code and configurations are available at https://github.com/PASHAADC/Danish-LLM-HateSpeech.
Read the full abstract

All authors Elnur Adl Zarabi, Amit Kumar, Adi Singh, Sierra Nelson, Megan Stubbs-Richardson, Taoran Ji

Computer Vision

BEiTScore++: Reference-free Image Captioning Evaluation with BEiT-3 and Conformalized Quantile Regression

Reliable uncertainty estimation is important for deploying evaluation metrics in real-world settings, where both input ambiguity and domain shifts are common. In this work, we propose an uncertainty-aware extension of the BEiTScore image captioning metric, producing conformally calibrated confidence intervals alongside alignment scores. Our approach models predictive distributions over human judgments, enabling both accurate scoring and principled uncertainty quantification in a single forward pass. We evaluate our BEiTScore++ metric across multiple benchmarks with human-annotated ratings, showing that it preserves the ranking correlation of BEiTScore, while producing intervals whose width tracks the prediction error more closely than Monte Carlo dropout on most benchmarks. The estimated uncertainty also grows under image degradation and on out-of-domain artwork images, making BEiTScore++ a practical tool for uncertainty-aware evaluation of image-caption alignment.
Read the full abstract

Quantum Physics

Sharp error tolerance of the monogamy-of-entanglement game

The BB84 monogamy-of-entanglement (MoE) game has numerous applications in quantum cryptography, including quantum key distribution, position verification, and unclonable encryption. In this short note, we improve the error threshold for parallel repetition with identical answer strings from the approximately $3.7\%$ certified by the original bound to the sharp threshold $(1-1/\sqrt2)/2\approx14.64\%$. When the players need not give identical answers, we also obtain exponential decay whenever the sum of their error tolerances is below this threshold. These bounds strengthen existing security reductions to the BB84 MoE game, improving security bounds and certified noise tolerances for cryptographic protocols.
Read the full abstract

Probability

The probability that a branching random walk does not hit a point

We study the probability that a branching random walk in ${\bf Z}^d$ driven by a Galton-Watson tree conditioned to have $n$ vertices does not hit a fixed point. For $d\in\{1,2,3\}$ this probability behaves like $n^{-\alpha/4}$ in logarithmic scale, where $\alpha=4$ when $d=1$, $\alpha=2\sqrt{2}$ when $d=2$, and $\alpha=(\sqrt{17}-1)/2$ when $d=3$.
Read the full abstract

Machine Learning

TimeOmni-Live: Continuous & Live Time Series Understanding with Long-Term Memory

Existing time series language models predominantly follow an offline, one-shot paradigm: they process fixed-length inputs independently and do not retain past states. This static approach fundamentally misaligns with the continuous nature of real-world streaming applications and fails to maintain the historical context necessary for cross-window reasoning. To bridge this gap, we introduce Continuous & Live Time Series Understanding (CL-TSU), a novel paradigm that shifts the focus toward maintaining a persistent, stateful understanding over open-ended data streams. To systematically evaluate this paradigm, we present LiveTS-Suite, the first comprehensive data suite tailored for CL-TSU, supporting data generation, model training, and systematic evaluation across 18 tasks spanning six core capabilities. It comprises over 5K streams totaling 2.2B timesteps, annotated with 15K specific response events. We further design an evaluation protocol that jointly assesses response timeliness, content accuracy, and computational efficiency. Building on this foundation, we propose TimeOmni-Live, the first live time series understanding model. It integrates a hybrid time series encoding for natively temporal input, a stream-persistent hybrid memory mechanism to effectively preserve exact historical evidence and long-term task states, and an elastic stream-inference synchronization strategy to reduce processing lag. Evaluations show that TimeOmni-Live establishes a strong baseline for CL-TSU, outperforming frontier LLMs (90.0 vs. 79.3 for Opus 5) with only 4B parameters and achieving 206$\times$ lower p95 latency than its backbone. Deployment on a Jetson edge device further shows its ability to support real-time continuous temporal understanding locally at the edge.
Read the full abstract

All authors Tong Guan, Alexander Chemeris, Yiqing Guo, Warren Jin, Zhao Li, Shirui Pan, Ming Jin

Machine Learning

e-means: outperforming k-means (in one-tenth the epochs)

Lloyd's k-means algorithm is a long-running standard for clustering, with applications in data analysis, retrieval, quantization and data curation. In this paper, we introduce an exponential moving average optimization procedure for k-means with mini-batch updates, powered by modern deep learning machinery and new insights. We also establish scaling laws on how decay rate and batch size impact final performance. This gives us a formulaic rule to fix them for multiple problem sizes and embedding datasets, without the need for any hyperparameter sweep. The resulting clustering algorithm, coined e-means, is much faster than Lloyd k-means and converges to lower losses, frequently even beating k-means++. On a wide array of embeddings datasets, it beats the loss of Lloyd k-means at 25 epochs in 1.4 to 2.6 epochs, which translates to 10-45x GPU wall-clock speedups compared to state-of-the-art k-means implementations. As an example, e-means partitions DINO10B, a dataset of 10 billion DINO embeddings, into 16 million clusters in only 2.5 epochs, reaching a loss below Lloyd's at 25 epochs.
Read the full abstract

All authors Swann Bessa, Pierre-Emmanuel Mazaré, Gergely Szilvasy, Hugo Touvron, David Picard, Matthijs Douze, Hervé Jégou

Analysis of PDEs

Sharper energy scaling bounds of the singularly perturbed Tartar square

In this article, we discuss the scaling behaviour of the singularly perturbed Tartar square, revisiting the results by [Rüland-Tribuzio '22]. Therein, it was proven that both lower and upper bounds satisfy a behaviour of the same type, up to a small mismatch of logarithmic powers. We show by an improvement of the ansatz free lower bound that the logarithmic powers indeed match and that the bounds are of order $\exp( - C \log(\varepsilon^{-1})^{1/2} )$ for suitable constants $C = C_{\rm low}$ and $C = C_{\rm upp}$, where $\varepsilon$ denotes the singular perturbation parameter. More specifically, the bootstrap argument in the Fourier space is refined by resorting to Fourier multipliers enjoying $L^\infty$-bounds which are constructed via de la Vallée Poussin kernels. Regarding upper bounds, we perform iterated branching constructions sharpening the upper bound of [Winter '97] which resorted to an iteration of laminates as a competitor. Eventually, the mismatch of upper and lower bound is quantified, and for quadratic energies we obtain the bounds $C_{\rm low} \leq 7.0339$ and $C_{\rm upp} \geq 2 \log(2) - \nu $ for any $\nu>0$.
Read the full abstract

Superconductivity

Driving Atomic Spins through Their Substrate Electrons

Coherent control of single spins on surfaces has relied on decoupling them from the substrate with insulating films, which protects the spin but also removes any electronic means of addressing it. Here we show that a local electric potential can drive an adatom spin through the substrate electrons themselves. An exact reciprocity relation fixes the torque, which equals the change of the local electron density when the moment rotates; the drive map is thus a ground-state property that can be measured independently. The relation also implies sharp selection rules, since point-contact spins, spin-$1/2$ and the central transition of half-integer spins remain dark. On a normal metal the hybridization responsible for the drive also destroys coherence, but on a superconductor the drive survives while substrate damping is confined to subgap transitions. For atomically sharp tips we find Rabi frequencies of several MHz at RF amplitudes of tens of mV, with substrate-limited linewidths below $0.1$~MHz, so that strong hybridization, usually the enemy of spin coherence, becomes an asset.
Read the full abstract

Fluid Dynamics

Selection versus residuals: deep-learning solvers for the Dubreil-Jacotin-Long equation and application to mode-2 internal solitary waves

Neural solvers offer structural flexibility for the Dubreil-Jacotin–Long (DJL) equation of internal solitary waves, which also admits the trivial state, oscillatory wavetrains and broadened near-solutions. A solver must select the physical wave before resolving it. Turkington's variational principle selects the wave through the nonlinearity of the available potential energy (APE). An exact virial partition shows that every localized wave carries more kinetic energy than APE, so a quadratic APE would favour a broadened field. The nonlinear APE enhancement, $1.27$–$2.06\times$ for the computed waves, reverses this ordering, although the preferred width remains weakly selected. From a shared KdV-derived start, a constrained Deep Ritz objective recovers the mode-1 wave, whereas strong- and weak-form residual objectives, which vanish at every exact solution and therefore do not favour the physical wave over other roots, lose the prescribed APE within one descent round. The recovered wave is under-resolved because its residual is dominated by high-wavenumber error that the energy barely penalizes, and a residual refinement at exact APE then meets every acceptance criterion. Structurally excluding mode 1 from the trial space, we compute mode-2 waves on four stratifications. A computed wave core qualitatively reproduces the two-lobe structure observed on the South China Sea shelf. In the deep basin, the selected core misses the residual criterion, and residual refinement widens it with little change in the energy objective. The two resolved linear estimates suggest weaker relative mode-1 radiation at larger ratios of wave half-width to resonant wavelength.
Read the full abstract

Computational Complexity

Hardness of Unique Shortest Vectors and Bounded Distance Decoding

We prove randomized NP-hardness of the unique shortest vector problem (uSVP) at every constant uniqueness factor greater than one, and of bounded distance decoding (BDD) at every fixed positive fraction less than one half of the minimum distance, in both their threshold decision and search versions. Both results hold in every fixed $\ell_p$ norm with $1\le p<\infty$. Previous Euclidean NP-hardness results for the search versions reached uniqueness factors tending to one and decoding radii greater than the minimum distance divided by $\sqrt2$. For linear codes over every fixed finite field and each fixed ${0<\rho\le1/2}$, the decision version of BDD with radius strictly less than $\rho$ times the minimum distance is equivalent under deterministic polynomial-time reductions to satisfiability for circuits promised to have at most one satisfying input. It is therefore NP-hard under randomized polynomial-time reductions. Earlier hardness for decoding general codes applied beyond half the minimum distance. For every fixed $p>2$ and for $p=\infty$, there is a constant $\epsilon>0$ such that, under randomized polynomial-time reductions, uSVP in the $\ell_p$ norm is NP-hard at uniqueness factor $n^\epsilon$, and BDD in the $\ell_p$ norm is NP-hard at radius $\lambda_1/n^\epsilon$, where $n$ is the lattice rank and $\lambda_1$ is its minimum distance.
Read the full abstract

Machine Learning

Mosaic:montage-agnostic self-distilled graph learning for subject-independent ieeg seizure detection

Intracranial EEG provides high signal-to-noise ratio and spatial resolution for detecting epileptic activity, but patient-specific electrode placement leads to substantial variation in channel count and montage across subjects. This heterogeneity limits the generalization of existing detectors under subject-independent evaluation. We propose MOSAIC (MOntage-agnostic Self-distilled Adaptive Inter-channel Connectivity learning), a lightweight graph-learning framework for subject-independent iEEG seizure detection. MOSAIC pairs a shared convolutional channel encoder with a patient-agnostic channel-attention module that infers a weighted inter-channel adjacency matrix from the encoded features, requiring no fixed montage graph. An exponential moving average self-distillation objective regularizes each prediction toward a slowly updated per-patient teacher. Across four public iEEG corpora (SWEC-ETHZ long- and short-term, Mayo, FNUSA), MOSAIC outperforms state-of-the-art baselines in F2 and Cohen's $\kappa$ while learning more stable inter-channel adjacencies, demonstrating that self-distilled connectivity learning transfers across unseen implantations.
Read the full abstract

Functional Analysis

The small ball problem for the Brownian sheet: sharp bounds via entropy numbers in $L_\infty$

We settle the uniform small deviation conjecture for the Brownian sheet in every dimension $d\ge2$, namely, that the logarithmic exponent is $2d-1$. Our proof proceeds through sharp entropy and Kolmogorov estimates for the associated anchored mixed Sobolev embedding in the uniform norm. Their connection to small deviations is known from the Kuelbs–Li theorem. To establish the entropy estimates, the key is a probabilistic small ball inequality for smooth tensor product atoms whose exceptional set has Gaussian measure exponentially small in the number of atoms. Its proof combines a count of dyadic rectangles with tools from probability, in particular Ehrhard's Gaussian Brunn–Minkowski inequality. This also yields sharp entropy and Kolmogorov number estimates for embeddings of periodic mixed Sobolev, Besov and Nikol'skii classes with smoothness $r>1/2$ in $L_\infty$. Finally, we extend our results to fractional and integrated Brownian sheets.
Read the full abstract

Materials Science

Interplay of Ni Order and Local Moments in Intercalated NixNbS2

Magnetic ions intercalated into the van der Waals gaps of layered transition metal dichalcogenides provide a system where local moments interact strongly with a metallic host. How the order of the intercalant sublattice affects their magnetism is important but rarely addressed, although disorder is hard to avoid, in particular in thin films and heterostructures. Here this question is examined in single crystals of Ni$_x$NbS$_2$ with $0.10 \leq x \leq 0.57$, whose composition and Ni order were determined by energy-dispersive x-ray spectroscopy and synchrotron x-ray diffraction. Anisotropic magnetic susceptibility measurements reveal three regimes. Long-range antiferromagnetic order appears only for $0.25 \leq x \leq 0.33$, where the Ni ions form the $2 \times 2$ and $\sqrt{3} \times \sqrt{3}$ R$30^\circ$ superstructures, and $T_\mathrm{N}$ reaches 96 K at $x = 1/3$. In this range the effective moment, $\approx 2.5 \mu_\mathrm{B}$ per Ni, agrees with DFT+$U$ calculations, in which the Ni spin is partly delocalized onto the surrounding S ligands. For $x \leq 0.18$ the dilute, randomly distributed Ni ions remain paramagnetic down to 2 K with a reduced moment. For $x \geq 0.37$, where Ni-Ni contacts are unavoidable, long-range order gives way to a short-range ordered state below 10-15 K with a further reduced moment. As calculations for ordered Ni arrangements find the Ni state nearly independent of $x$, even at $x = 1/2$, the reduction of the moment is attributed to a disorder-enabled reduction of the charge transfer from Ni to the NbS$_2$ layers. The magnetism of Ni$_x$NbS$_2$ is thus governed by the (dis)order of the Ni sublattice rather than by the Ni concentration alone.
Read the full abstract

All authors Gaurav Pransu, Naveen Singh Dhami, Wojciech Sas, Yuki Utsumi Boucher, Bruno Gudac, Ana Smontara, Davor Tolj, Victor Balédent, Mario Novak, Neven Barišić, Ivo Batistić, Mirta Herak, Petar Popčević

Computer Vision

FusionCLIP: Multimodal Contrastive Pretraining for Fusion Plasma Diagnostics

Visible-light videos and multivariate diagnostic signals provide complementary observations of tokamak discharges. However, existing fusion pretraining methods primarily learn representations by reconstructing or predicting observations, leaving the correspondence between these modalities underexplored. Representation learning is further challenged by spatially localized plasma regions in video frames and substantial differences in diagnostic sampling rates. We propose FusionCLIP, a multimodal contrastive pretraining framework. A selective video encoder retains informative visual tokens, while a query-based time-series encoder integrates asynchronous diagnostic signals using their observation timestamps. Contrastive learning aligns representations of the two modalities. We construct the EAST Multimodal Diagnostics (EAST-MMD) dataset, comprising 1307 EAST discharges, and evaluate the model through independent fine-tuning on three tasks: disruption prediction, edge-localized mode (ELM) classification, and H98 regression. FusionCLIP achieves a Joint score of 89.07%, with absolute gains of 2.41% and 2.90% over the best competing baseline, and the in-domain contrastive pretraining baseline CLIP-Base, respectively. Ablations and attention visualizations further support the contributions of its key components.
Read the full abstract

All authors Qiang Chen, Xiao Wang, Qingquan Yang, Hao Si, Zikang Yan, Guosheng Xu, Jin Tang

Numerical Analysis

Mathematical analysis and vectorized FEM implementation of logarithmic strain space framework for finite strains: a case study on elastoplasticity

We investigate the mathematical and computational aspects of the logarithmic strain space framework for finite-strain models in nonlinear continuum mechanics, focusing on the classical von Mises elastoplastic model with linear kinematic hardening. To overcome the lack of coercivity and lower-semicontinuity in the quadratic Hencky energy, we introduce higher-gradient regularization and prove the existence of energetic solutions for rate-independent quasistatic evolution under time-dependent Dirichlet loading. Complementarily, we develop a vectorized MATLAB finite-element implementation of the framework leveraging page-wise matrix functions and operations, and our custom vectorization library. Computational benchmarks show that local tensor operations are significantly accelerated, the logarithmic-strain approach does not substantially increase computational costs, and the implementation robustly handles large rotations and deformations.
Read the full abstract

All authors Miroslav Frost, Martin Kružík, Alexej Moskovka, Jan Valdman

Computation and Language

An Effective Spoken Language Understanding Approach Leveraging Query-Adaptive Intent Detection

Large language models enable fine-tuning-free spoken language understanding (SLU) through in-context learning, but direct query-based retrieval may select semantically similar yet intent irrelevant demonstrations. This paper presents Query-Adaptive Intent Detection via Prototype Extraction (APEX), a two-stage retrieval framework that first selects candidate intents using intent-level prototypes and then retrieves contextual exemplars within the selected intent space. Experiments on the SLURP dataset evaluate joint SLU performance using Exact Match Accuracy (EMA) under full-transcript and perplexity-stratified conditions, with additional retrieval analysis on ASR-generated transcripts. Compared with query-based retrieval, APEX improves EMA by 4.34 and 3.97 percentage points with Qwen3 4B and Gemma-3-4B, respectively, on full transcripts. Under high-perplexity inputs, APEX also yields EMA gains of 7.66 and 2.13 percentage points. Retrieval analysis further indicates that prototype-guided candidate selection achieves higher ground truth intent coverage across multiple input conditions.
Read the full abstract

All authors Bing-Long Kuo, Yu-Hsuan Fang, Tien-Hong Lo, Hsueh-Wei Liao, Berlin Chen

Robotics

CucumberVision at the Edge: Accuracy, Throughput, and Energy of RGB-D Cucumber Sizing on Jetson Orin Nano and Raspberry Pi 4

Non-contact fruit sizing from RGB-D imagery reaches sub-centimetre accuracy on workstation hardware, but in-field grading requires the same pipeline to run on a low-power single-board computer, and the cost of that transition is rarely reported. We deploy a curvature-aware cucumber length-measurement pipeline (YOLO26n detection, prompt-based segmentation and a medial arc spline estimator) on an NVIDIA Jetson Orin Nano and a Raspberry Pi 4, and characterize both against an RTX 3050 Ti workstation by replaying 50 identical greenhouse captures of 7 specimens, with specimen-level cluster-bootstrap statistics. Accuracy is preserved: mean absolute error (MAE) is 0.58 cm (95% CI 0.46-0.69) on both the workstation and the Jetson at matched precision, with a paired difference of zero. TensorRT FP16 compilation cuts Jetson detection latency 2.57x, giving two operating points on one board: 72 ms per fruit at 0.84 cm, or 2.74 s at 0.64 cm. Four ablations map the design space. For prompted segmentation backends, measurement accuracy is fixed by the backend and reproduces across platforms to within 0.01 cm, while its cost is fixed by the platform and varies 214x, so the two choices are separable. Whole-graph INT8 quantization of the NMS-free detection head fails structurally, returning no detections, because one output tensor carries both pixel coordinates and confidences; we show that no calibration strategy can avoid this. Excluding the head restores function at 1.37x. Energy per fruit is 42.2 J on the Jetson, and its 7 W mode consumes 73% more than its 15 W mode. The Raspberry Pi with MobileSAM costs 2.5x more per fruit than the Jetson with SAM despite drawing 47% less power, and it runs without thermal throttling under sustained load. The results define an accuracy-throughput-energy envelope with three non-dominated configurations and give rules for selecting among them.
Read the full abstract

All authors Manveen Kaur, Rajmeet Singh, Saeed Mozaffari, Majid Ahmadi, Shahpour Alirezaee

Instrumentation and Detectors

The ATAR, a dense and compact vertex detector for the PIONEER experiment

PIONEER is an approved experiment at the Paul Scherrer Institut (PSI) in Villigen, Switzerland that aims to measure the charged-pion branching ratio to electrons vs. muons with an accuracy of 0.01%. This is an order of magnitude improvement in precision compared to its predecessors and approaches the accuracy of the standard model prediction. The key elements of PIONEER are a high resolution calorimeter built concentrically around the high granularity active target (ATAR). The ATAR is composed of a stack of 48 low gain avalanche strip detectors (LGADs) to stop the beam pions and provide detailed information of the event topology. This will include precise timing (in the order of 200ps), spatial information (to an accuracy of about 60 micro meter) and pulse height information with a high dynamic range to separate pion stops from muons and MIPs. To avoid affecting the energy measurement of the positrons, the readout electronics is placed outside the acceptance angle of the calorimeter. It is composed of a fast amplifier and a fast sampling chip digitizing the LGAD wave form for real time analysis. The paper describes the PIONEER-experiment with focus on the ATAR and its readout. The status of the development and activities in 2026 will be reported.
Read the full abstract

High Energy Astrophysical Phenomena

Constraints on Cosmic-Rays and the MeV Emission from Jupiter's Atmosphere

The Cosmic $\gamma$-ray Background (CGB) in the MeV regime remains among the least understood components of diffuse cosmic radiation, primarily due to instrumental limitations in detecting faint isotropic emission at MeV energies. We present an indirect method for probing isotropic emission from soft X-rays to $\gamma$-rays through scattered emission in Jupiter's atmosphere. Using archival observations from SPI, the spectrometer aboard INTEGRAL, spanning 22 years of serendipitous Jupiter observations, we analyse the planet as a moving $\gamma$-ray point source in the 30 keV to 8 MeV range. Our analysis focuses on Jupiter's $\gamma$-ray albedo, produced by high-energy photon and cosmic-ray interactions. We performed detailed Monte Carlo simulations to model the response of Jupiter's atmosphere to incident photons, protons, electrons, positrons, and alpha particles. We simulate a 1000 km deep spherical atmosphere with 1 km shell resolution, based on Galileo probe measurements. By comparing empirical and astrophysical models with the SPI data, we find no statistically significant emission from Jupiter. We derive $2\sigma$ upper limits across the 30 keV to 8 MeV band, which we compare with cutoff power-law, reflected CGB, cosmic-ray-induced, and long-lived mediator dark matter models. We constrain Jupiter's bremsstrahlung luminosity in the MeV range to $10^{18}$ erg/s, close to that inferred from in situ measurements obtained by Juno/JEDI. While the expected emission from the CGB and cosmic rays is a factor of $10^4$ and 100 below our upper limits, respectively, this work establishes planetary atmospheres as promising indirect probes of diffuse $\gamma$-ray backgrounds for future MeV missions.
Read the full abstract

All authors Dimitris Tsatsis, Thomas Siegert, Volodymyr Savchenko, Saurabh Mittal, Laura Eisenberger, Mika A. Gelowicz, Rudi Reinhardt, Manja L. Zimmerer, Tristan Bouchet, Tomohiko Oka, Manuel R. H. W. Skalka

Probability

Backward–forward interacting systems with probability-valued strategies and common noise

We study a system of interacting agents in which a backward stochastic equation is coupled to the forward evolution of a probability-valued strategy distribution. The agents interact through their empirical state–strategy law and are exposed to a common Brownian noise. Under Lipschitz, mass-conservation, and positivity assumptions, we establish well-posedness of the finite-agent system on a sufficiently short time interval. We then identify its large-population limit as a conditional mean-field backward–forward system and prove conditional propagation of chaos. When agent characteristics take finitely many values, we obtain order $N^{-1}$ bounds for the expected squared particle–copy error and the expected squared Wasserstein distance of order one between the empirical law and the conditional limiting law. An adaptive-strategy model with replicator dynamics illustrates the theory, accompanied by a numerical comparison of the two errors.
Read the full abstract

General Topology

On two examples of topological gyrogroups

In this paper, we construct an infinite-dimensional topological gyrogroup that fails to be strongly topological. We equip the countable direct sum of the complex unit disk with a carefully weighted infinite series of component seminorms. By dividing the infinite-dimensional operation into a finite core and a bounded infinite tail, we rigorously prove the joint continuity of the Möbius addition, confirming the space is a valid topological gyrogroup. Furthermore, we demonstrate that the gyroautomorphisms exhibit unbounded expansion under this real seminorm topology, explicitly preventing the existence of an invariant neighborhood basis at the identity. This provides a clear, analytic counterexample separating general topological gyrogroups from strongly topological ones. Moreover, we construct a topological gyrogroup which is not strongly topological but it contains a non-trivial strong subgyrogroup.
Read the full abstract