Peer Revenue Returns subject to peer review

Papers

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

All of arXiv

Monday 31 August

Optimization and Control

A method for global minimization of nonconvex quadratic functions

The problem of global minimization of nonconvex quadratic functions subject to box constraints is studied. Applying Gershgorin theorem we reduce the main matrix A to its diagonal form which is used to obtain difference of convex representation of the quadratic function. Then {\epsilon}-subdifferentials of DC components are studied and used to design a method for minimizing the quadratic function globally.
Read the full abstract

Atmospheric and Oceanic Physics

Universal ordinary differential equations and the parameterization of warm-rain processes

The parameterization of warm-rain formation is a long-standing problem in cloud microphysics because the evolution of bulk cloud variables cannot be derived uniquely from the kinetic collection equation (KCE). Universal ordinary differential equations (UODEs)provide a scientific machine learning framework that combines neural networks with ordinary differential equations and can learn dynamical operators directly from time series. Here, the UODE framework is applied to warm-rain formation by training on trajectories generated with a super-droplet model that approximates the KCE. In contrast to most previous machine-learning approaches, only the prognostic state variables are used for training, without requiring autoconversion or accretion rates as targets. The learned closure accurately reproduces the KCE trajectories within the training domain and, although not explicitly constrained to do so, yields autoconversion and accretion rates that closely resemble those diagnosed from the KCE. Analyzing the learned operator provides new insight into the structure of the widely used Seifert and Beheng (2001) warm-rain parameterization. In particular, it suggests that the empirical suppression of accretion primarily compensates for an overestimation of autoconversion at small rain fractions. Motivated by these findings, a refined analytical formulation is proposed that improves agreement with the KCE while retaining the simplicity of the original parameterization. The results demonstrate that UODEs can serve not only as accurate surrogate models but also as a tool for understanding and improving analytical parameterizations of cloud microphysical processes.
Read the full abstract

Computation and Language

Reading the News: Adapting Large Language Models to Swedish Journalism Through Continued Pre-Training

Large language models are increasingly capable in general, but their utility can remain modest in niche or understudied areas. One approach to address this limitation is to specialise existing models through additional training on target-domain corpora. In this work, we investigate such continued pre-training for adapting large language models to Swedish journalism, using a high-quality dataset that we curate from millions of news articles. To evaluate the adaptation efficacy, we also construct a novel domain-specific benchmark that covers six editorial tasks. Through full and parameter-efficient fine-tuning across two model sizes, we find that continued pre-training yields benefits in the target domain, but only when paired with experience replay to mitigate forgetting. We observe consistent enhancements in the models' generation quality and factual knowledge, but not their proficiency in discriminative tasks. Exploring a training-free method to facilitate instruction following, we see further improvements, but exclusively for models trained with low-rank adaptation. Crucially, we demonstrate the importance of targeted evaluation in the adaptation process, as an existing Swedish benchmark largely fails to capture the models' in-domain performance gains.
Read the full abstract

Number Theory

Effective Joint Equidistribution of Subspaces of Bounded Height in Rigid Adelic Spaces

We prove effective joint equidistribution for subspaces of fixed dimension in a fixed rigid adelic space over a number field. The joint distribution records the rescaled height, the archimedean position, and the shapes of the subspace and its quotient. More generally, we obtain counting estimates for regions in which the height bound may depend on the position and both shapes. Under a Hölder-type bound for the measure of boundary layers, the estimates have explicit power-saving error terms. We apply these results to counting problems involving slopes, freeness, and semistability, and to Diophantine approximation with prescribed shapes.
Read the full abstract

Databases

UBASE: An AI Search Engine for Trillion-Scale Vector Data Management at ByteDance

Since 2016, UBASE has been the foundation of ByteDance's search infrastructure, scaling to more than 7,000 clusters and 300 PB of indexed data. Driven by the demands of AI workloads, UBASE has evolved from a text search engine into a unified AI search system supporting vector retrieval, lexical matching, and predicate filtering. Its largest deployment indexes nearly one trillion high-dimensional vectors. This scale exposes two central bottlenecks in AI-era retrieval: memory-intensive graph-index construction under sustained ingestion, and the prohibitive cost of keeping vector indexes entirely in memory. UBASE addresses these bottlenecks with two techniques. First, it introduces a quantization-aware vector kernel based on SymRaBitQ, a new symmetric quantization scheme with tight theoretical guarantees that allows index construction to run directly in the quantized space accurately and efficiently without retaining a copy of full-precision vectors. Second, it provides a hybrid storage engine that supports memory-resident, hybrid, and SSD-resident deployments, with fine-grained record-level caching to trade memory for latency under operational control. On large-scale benchmarks, UBASE improves throughput by up to 3x, reduces indexing memory by 80%, and lowers serving cost by up to 99.4% compared with prior systems, while supporting trillion-vector scale, write-heavy or latency-sensitive workloads in production.
Read the full abstract

All authors Yao Tian, Yuncheng Lu, Liyao Xiong, Yuming Xu, Hao Zhang, Weichen Zhao, Xi Zhao, Bo Kuang, Dongyu Wang, Jiehui Li, Yakun Li, Lei Zhang

Information Retrieval

Generative Retrieval for E-commerce: Jointly Learning Embedding and Codebook with Same Product Cluster

With the development of large language models (LLMs), generative retrieval is becoming increasingly important in e-commerce scenarios. Current mainstream approaches typically use a two-stage training strategy: first train a product embedding model, and then learn a codebook that maps embeddings to product IDs. This cascaded approach suffers from two major issues: (1) error accumulation-if the embedding model in the first stage produces biased representations, the codebook in the second stage cannot correct these errors, degrading final retrieval performance; and (2) codebook learning relies solely on product embeddings and lacks modeling of query-to-product and product-to-product interactions. As a result, products belonging to the same cluster may be assigned inconsistent IDs by the codebook, further hurting retrieval accuracy. To address these problems, we propose a novel method that jointly trains the embedding model and the codebook, and incorporates same product cluster information as an additional supervision signal. Experimental results demonstrate that our method significantly improves e-commerce retrieval performance while simultaneously enhancing both embedding and codebook learning.
Read the full abstract

All authors Songtao Fang, Zihao Xu, Shaowei Wei, Jin Zhang, Zhuojun Wang

High Energy Physics (Phenomenology)

Hadron interactions in the U-70 energy range: results and problems

From late 1967 until the early 1970s, the Soviet proton synchrotron (better known as the Serpukhov accelerator or U-70), with a nominal laboratory energy of up to 76 GeV, was the world leader. During these few years, several results of fundamental significance were obtained (partially in collaboration with physicists from CERN, France, and other countries). Let us briefly outline the main provisions of the paradigm that reigned at that time in the field of strong interaction theory.
Read the full abstract

Combinatorics

A Full-Sequence Quantitative Gap Between the Chromatic and Cochromatic Numbers of a Random Graph

The cochromatic number $\zeta(G)$ of a graph $G$ is the least number of parts in a partition of $V(G)$ into cliques and independent sets, so $\zeta(G)\le\chi(G)$. Erdős and Gimbel asked whether $\chi(G_n)-\zeta(G_n)\to\infty$ with high probability for $G_n\sim G(n,1/2)$. Heckel answered this affirmatively for about $95\%$ of all integers $n$ and conjectured that the natural order of the difference along the full sequence is $n/(\ln n)^3$. We prove that, as $n\to\infty$ through all integers, $\chi(G_n)-\zeta(G_n)\ge0.063\,n/(\ln n)^3$ with high probability. The failure probabilities are summable, so under any coupling of the graphs $G_n$ the bound holds almost surely for all sufficiently large $n$. The proof counts partitions of the vertex set into classes of five consecutive sizes, each class declared independent or complete. The two choices per class outweigh the entropy lost by restricting the class sizes, so the expected number of such partitions is large already at a number of classes that lies order $n/(\ln n)^3$ below the first-moment lower bound for the chromatic number, uniformly in the rounding phase. A second-moment argument, which controls all overlaps between two such partitions, including whole shared classes and large intersections of unequal classes, bounds the probability that such a partition exists from below. Vertex-exposure concentration, together with a coloring of the leftover vertices, then raises this to high probability at a cost of $o(n/(\ln n)^3)$ extra classes.
Read the full abstract

Computer Vision

DiffSAC: Diffusion-guided Sampling for Consensus-based Robust Estimation

Robust estimation is a core computer vision task frequently tackled using sample consensus. However, traditional methods suffer from inefficient sampling as they struggle to identify effective minimum sets before hypothesis evaluation. To address these challenges, we propose a novel Diffusion-guided Sampling for Consensus-based Robust Estimation (DiffSAC) framework. DiffSAC introduces a diffusion model to learn the distribution of effective minimum sets. It refines the confidence for each data point, indicating whether it belongs to a good minimum set, rather than ranking the data points as in previous work. This significantly reduces the need to process numerous bad sets. To constrain the refinement direction, geometric features are incorporated as conditions within our diffusion model. Consequently, DiffSAC outputs a small number of high-quality minimum sets, enabling identification of the best hypothesis via consensus evaluation. Notably, compared to previous works requiring evaluating over ten thousand hypotheses, DiffSAC achieves state-of-the-art performance with only dozens, significantly boosting efficiency. Extensive experiments across five classic computer vision tasks demonstrate the superiority of DiffSAC. The diffusion model's sampling accelerators enable real-time operation, and DiffSAC can be used as a plug-and-play module to improve existing sample consensus methods.
Read the full abstract

All authors Chang Nie, Guangming Wang, Zhe Liu, Hesheng Wang

Analysis of PDEs

Endpoint Mapping Properties of Wave Operators for Two-Dimensional Schrödinger Operators

We establish sharp endpoint mapping properties for the wave operators $W_\pm(H,-\Delta)$ of two-dimensional Schrödinger operators $H=-\Delta+V$ with real-valued decaying potentials $V$. Together with the known non-endpoint $L^p$ theory, our results give a complete classification of the $L^p$ mapping properties of the two-dimensional wave operators, and reveal an unexpected reversal of the usual threshold paradigm at the endpoints $p=1$ and $p=\infty$. When zero is a regular point of $H$, the wave operators fail to be bounded on $L^1(\mathbb{R}^2)$ and on $L^\infty(\mathbb{R}^2)$, but they satisfy the atural substitute estimates of Calderón–Zygmund type: $$ L^1(\mathbb{R}^2)\longrightarrow L^{1,\infty}(\mathbb{R}^2),\ \ \ \mathcal{H}^1(\mathbb{R}^2)\longrightarrow L^1(\mathbb{R}^2),\ \ \ L^\infty(\mathbb{R}^2)\longrightarrow \mathrm{BMO}(\mathbb{R}^2). $$ When zero is instead a threshold singularity of the first kind—an s-wave resonance with no other threshold obstruction, the wave operators are bounded on both endpoint spaces $L^1(\mathbb{R}^2)$ and $L^\infty(\mathbb{R}^2)$. Thus, in dimension two, an s-wave resonance improves the endpoint behavior of the wave operators, in sharp contrast with dimensions $n\ge3$, where the only regular case is the favorable one. We also determine the endpoint behavior in the remaining zero-energy spectral configurations of $H$. A p-wave resonance obstructs both the $L^1$- and the $L^\infty$-boundedness of the wave operators, while in the zero-eigenvalue case we obtain necessary and sufficient conditions for endpoint boundedness, expressed in terms of the presence of s- and p-wave resonances and of explicit second-order harmonic moment cancellations satisfied by the zero-energy eigenfunctions.
Read the full abstract

Combinatorics

The Homotopy Types of the Independence and Perfect Matching Complexes of Möbius and Circular Ladder Graphs

The independence complex and perfect matching complex of a graph are simplicial complexes encoding, respectively, its independent sets and perfect matchings. Determining their homotopy types is generally difficult, with explicit descriptions known mainly for highly structured graph families. In this article, we determine the homotopy types of these complexes for the Möbius ladder graphs $M_{2n}$ and circular ladder graphs $\mathcal{C}_{2n}$. The Möbius ladder graphs $M_{2n}$ are highly symmetric cubic graphs obtained from a $2n$-cycle by joining opposite vertices, while the circular ladder graphs $\mathcal{C}_{2n}$ are the Cartesian products of an $n$-cycle and a path of length one. We show that $\operatorname{Ind}(M_{2n})$ and $\operatorname{Ind}(\mathcal{C}_{2n})$ have the homotopy type of wedges of spheres, with the numbers and dimensions of the spheres exhibiting periodic behavior according to $n$ modulo $4$. For the perfect matching complex $\mathcal{M}_p(M_{2n})$, its homotopy type is a wedge of two copies of $\mathbb{S}^{(n-2)/2}$ when $n$ is even, while for odd $n$ it has the homotopy type of a wedge of spheres whose numbers and dimensions depend periodically on $n$ modulo $6$. The perfect matching complex $\mathcal{M}_p(\mathcal{C}_{2n})$ is contractible for odd $n$, whereas for even $n$ its homotopy type is a wedge of spheres, with the numbers and dimensions determined periodically by $n$ modulo $6$. Thus, we obtain explicit homotopy types for the independence and perfect matching complexes of two highly symmetric families of cubic graphs, which are also relevant in crystallization theory and the combinatorial representation of PL manifolds.
Read the full abstract

Materials Science

Layer Axial Phonons and Dipolar Thermal Response in Centrosymmetric Thin Films

In nonmagnetic centrosymmetric materials, $\mathcal{PT}$ symmetry enforces vanishing total phonon angular momentum (phonon AM) at every wave vector, making them appear unsuitable as hosts of axial phonons. Here, using first-principles calculations for BaAgAs and representative van der Waal's layered materials, we show that centrosymmetric slabs can still host large hidden layer-resolved phonon AM with both chiral and cycloidal modes. This hidden phonon AM texture produces layer resolved phonon thermal Edelstein-like responses and a real-space phonon AM dipole response, both tunable by strain. Our results establish centrosymmetric layered materials as an exciting platform for hidden axial phonon and multipolar phonon thermal response.
Read the full abstract

Cosmology and Nongalactic Astrophysics

Not all those missing are lost: leveraging galaxy clustering in incomplete catalogs to unleash dark sirens cosmology

Gravitational waves offer a unique opportunity to solve the Hubble tension. In order to do so, we have to extract as much information as possible from cross correlating gravitational wave events (used as "dark sirens") with incomplete galaxy catalogs. Traditional methods assume a uniform in comoving volume distribution for galaxies missing from the catalogs, which neglects the fact that galaxies tend to cluster, leading to less precise and possibly biased posteriors. In this paper, we introduce a new method for accounting for galaxy clustering when dealing with an incomplete galaxy catalog, by adding back galaxies to the incomplete catalog and distributing them in pixels according to the correlation function $\xi(r)$. We find that our method drastically improves on traditional ones when the galaxy catalog is only $1\%-10\%$ complete, leading to posteriors that are between 2 and 4 times as precise, depending on the completeness fraction, without introducing any biases. Our method produces results comparable to traditional methods at extremely low catalog completeness fractions (<0.5%) or for very high uncertainties in the recovered luminosity distance and sky localization of the gravitational wave events.
Read the full abstract

All authors Riccardo Barbieri, Marios Kalomenopoulos, Matteo Tagliazucchi, Jonathan Gair, Sadegh Khochfar

Superconductivity

Quadratic pair-breaking absorption edge from Anderson localization in a superconducting wire

We show that Anderson localization fundamentally reshapes dissipative superconducting electrodynamics, replacing the linear Mattis–Bardeen pair-breaking onset with a parametrically weaker quadratic Mott edge while leaving the quasiparticle spectrum unchanged. We derive this behavior for a weakly disordered single-channel wire with a spatially uniform $s$-wave pair potential using a Nambu-space extension of the Berezinskii diagram technique that resums elastic impurity scattering to all orders. Conservation of the Bogoliubov branch collapses the Nambu diagram hierarchy onto an exactly solvable localization problem. The localization length remains equal to its normal-state value, whereas the localization time diverges at the gap edge. The new absorption edge results from the product of the superconducting pair-creation coherence factor and the Mott-suppressed current matrix elements between localized orbitals, which generates the characteristic double logarithm. Localization also reduces the Hebel–Slichter coherence peak. The resulting theory provides an all-orders benchmark for microwave and spin-relaxation experiments on localized superconducting nanowires.
Read the full abstract

Machine Learning

PLC-DPO: Posterior Label Correction in Noisy and Ambiguous Preference Optimization

Direct Preference Optimization (DPO) simplifies alignment through pairwise comparisons but assumes all observed preferences are reliable. Real data often violates this assumption, leading to reversed, weak, or ambiguous labels that cause harmful policy updates. To address this, we propose Posterior Label Correction DPO (PLC-DPO) to robustly optimize preferences by routing each pair's training signal as a clean, flip, or tie case. The key idea is to use the calibrated policy-reference margin as online evidence to take appropriate correction actions. This reframes noisy preference learning as actively correcting supervision direction and strength rather than merely filtering suspicious examples. Across 57 dataset-model-benchmark cells, PLC-DPO obtains the best mean win rate against DPO (60.5 vs. 55.5 for the next-best method). Injected-noise and tie stress tests, human disagreement analysis, and self-confirmation diagnostics further show that the routing remains stable and distinguishes flipped from weakly directional pairs.
Read the full abstract

High Energy Astrophysical Phenomena

TeV Gamma Rays and Neutrinos from Winds Driven by Radiatively Inefficient Accretion Flows of Low-Luminosity Active Galactic Nuclei

Recently, the Large High Altitude Air Shower Observatory (LHAASO) reported the first TeV gamma-ray detection from a low-luminosity active galactic nucleus (LLAGN) lacking a strong jet. This source, NGC 4278, is one of the nearest LLAGNs that can be studied in detail as a template for more distant objects. Most previous studies have focused on the month-scale-variable gamma-ray activities, seeking to explain them with weak jets. We focus instead on NGC 4278's quiet-period emission, proposing a different mechanism based on a wind launched from a radiatively inefficient accretion flow (RIAF). In particular, we model particle acceleration at a wind-driven shock and calculate the resulting multimessenger emission. To match the TeV gamma-ray fluxes reported by LHAASO in the quiet period, our model requires a wind power of $\lesssim2\times10^{43}~\mathrm{erg~s^{-1}}$ with a spectral index of injected cosmic rays of 2.0. We show that this is achievable through Bondi accretion of the ionized gas observed on the 100-pc scale, for a shallow accretion profile. We discuss the testability of our wind-driven shock scenario with multimessenger observations. As another important application, we extend our model to Sgr A*, the nearest LLAGN, and show that its TeV gamma-ray emission is reproduced with reasonable parameters. Our results suggest that RIAF-driven winds may be a generic emission mechanism in LLAGNs, making this class a potentially important population of both gamma-ray emitters and cosmic-ray accelerators.
Read the full abstract

Complex Variables

Schwarzian norm estimates for some classes of analytic and harmonic mappings

Let $\mathcal{A}$ be the normalized class of analytic functions $f$ in the unit disc $\mathbb{D} := \{z \in \mathbb{C} : \vert{}z\vert{} < 1\}$. For $\beta > 1$, let $\mathcal{N}(\beta)$ denote the subclass of $\mathcal{A}$ satisfying $\text{Re}\{1 + z f''(z)/f'(z)\} < \beta$ for $z \in \mathbb{D}$. The main purpose of this paper is to establish sharp bounds for the pre-Schwarzian norm $\Vert{}P_f\Vert{}$ and Schwarzian norm $\Vert{}S_f\Vert{}$ for functions $f \in \mathcal{N}(\beta)$, parametrized by $f''(0)$, with special emphasis on the case $f''(0) = 0$. In addition, sharp growth, distortion, and radius results (convexity and concavity) for $\mathcal{N}(\beta)$ are obtained. As an application, we determine the sharp pre-Schwarzian norm estimate for harmonic mappings $f = h + \bar{g}$ whose analytic part $h$ belongs to $\mathcal{N}(\beta)$.
Read the full abstract

Instrumentation and Methods for Astrophysics

Learning Radio Astronomical Representations with LeJEPA and Very Small Models

Representations learned by vision foundation models pretrained on natural images have been shown to be useful for out-of-domain astronomical images. Performance on scientific downstream tasks increases with model size, which both carries higher inference costs and limits scalability, even when considering parameter-efficient adaptation. An alternative is to learn representations directly from astronomical observations rather than natural images, through self-supervised pretraining. We evaluate LeJEPA's ability to learn robust representations using very small vision models ($\sim$6M parameters) pretrained on Radio Galaxy Zoo images, comparing with established self-supervised frameworks. We test whether LeJEPA's latent-space regularization leads to better radio galaxy morphology classification. Across three evaluation datasets, LeJEPA achieves performance comparable to a substantially larger foundation model while producing more consistent representations across training and evaluation datasets. These results suggest that the choice of representation learning objective is critical for enabling small domain-specific models to achieve performance competitive with representations transferred from large foundation models in scientific imaging.
Read the full abstract

All authors Erica Lastufka, Mariia Drozdova, Vitaliy Kinakh, Taras Holotyak, Miroslava Dessuages-Zavadsky, Daniel Schaerer, Svyatoslav Voloshynovskiy

Machine Learning

State of Health Estimation using Convolutional and Bidirectional LSTM Neural Networks tuned by Bayesian Optimization

In this research, a novel framework is proposed for the SOH estimation, which employs a hybrid deep learning architecture of a concatenation of a Convolution Neural Network (CNN) and a Bidirectional Long Short-Term Memory (BiLSTM) Neural Network (NN) with the integration of Bayesian Optimization-based hyperparameter tuning for the network. Three different deep learning architectures are being evaluated: standalone recurrent models, CNN-RNN architectures and CNN-RNN combinations enhanced with intermediate Fully Connected (FC) layers. Among the three, the model with the intermediate FC layers demonstrated the highest predictive accuracy. A comprehensive feature engineering approach combines capacity (Q), voltage (V), Incremental Capacity Analysis (ICA), and Differential Voltage Analysis (DVA), with systematic evaluation of multiple combinations to identify the optimal input representation. To validate the proposed method, three publicly available datasets were utilized, ensuring reproducibility of the results, two from external sources and one developed by the author of this study using a unique experimental setup. The comparison study was performed using the Mean Absolute Error (MAE), the Root Mean Squared Error (RMSE) and the FLoating-point OPerations (FLOPs) as evaluation metrics.
Read the full abstract

Functional Analysis

Infinite Matrix Operators and $E$-Frames in Hilbert Spaces

The concept of an $E$-frame, recently introduced in frame theory, is obtained by applying an infinite invertible complex matrix to a sequence of elements of a Hilbert space $\mathcal{H}$. Here, $E$ is considered as a matrix mapping on the sequence space $\bigoplus_{n=1}^{\infty}\mathcal{H}$. A natural question is to determine the conditions on a sequence $\Psi=\{\psi_k\}_{k=1}^{\infty}$ and a matrix $E$ under which $E\Psi$ becomes a frame. In this paper, we show that if $E$ acts surjectively on $\bigoplus_{n=1}^{\infty}\mathcal{H}$, then every ordinary frame is an $E$-frame.
Read the full abstract

Materials Science

Freezing of a deformable water-saturated porous medium. Part I: THM formulation, OpenGeoSys-6 implementation and benchmarking

In this contribution, Part I, we present a coupled thermo-hydro-mechanical formulation for modeling and analyzing water-to-ice phase change in a deformable fully-saturated porous medium. It is implemented in the multi-physics computational platform OpenGeoSys-6. We compute a series of carefully designed benchmark problems which critically examine the corresponding formulation components and the overall implementation. Several ingredients that have a qualitative and quantitative impact on the numerical results are identified, particularly dissected and commented on. Simulations also account for soil deformation induced by freezing (as a result of 9% volumetric expansion caused by water-to-ice phase transition). The forthcoming Part II will complete the code verification and validation campaign by considering a real full-scale three-dimensional case study of shallow subsurface ice storage. More specifically, we simulate the ice growth process in a fully saturated soil specimen surrounding a group of borehole heat exchangers (BHEs) that contain subzero temperature coolant fluid. A snapshot of the numerical results of this task is already depicted here in Part I as a teaser.
Read the full abstract

All authors Tymofiy Gerasimov, Christian B. Silbermann, Dmitri Naumov, Olaf Kolditz, Haibing Shao, Thomas Nagel

Atmospheric and Oceanic Physics

Ice-thickness based scaling of wave attenuation in sea ice: Application and assessment of wave spectra

This study discusses recent advances in modeling waves in sea ice in the U.S. Navy's regional modeling system. It is applied in the marginal seas of the eastern Arctic Ocean, including the Barents Sea, Kara Sea, parts of the Greenland Sea, Norwegian Sea, and waters north of Svalbard. The focus is to assess the skills of two formulations of wave attenuation by sea ice used operationally in WAVEWATCH III. Both are derived from large field datasets, one from the Arctic and the other from the Antarctic. The new model (IC4M9) describes wave attenuation depending on the ice thickness in association with the dependence on wave frequency, while the earlier default scheme (IC4M6) omits the dependence on ice thickness. The modeling results are evaluated against the satellite wave observations from SWIM/CFOSAT and the buoy measurements from the Svalbard Marginal Ice Zone 2024 Campaign (SvalMIZ-24). The comparisons with SWIM data validate the wave model skill in regions of open water or with light ice coverage. When evaluated against the SvalMIZ-24 data, the statistical performance of IC4M9 is substantially better than that of IC4M6, showing the influence of ice thickness on waves in the MIZ. Moreover, diagnosing systematic errors in the predictions by IC4M9, we find that the ice thickness field provided by the sea ice model CICE to the wave model is biased high in the MIZ, thus penalizing the performance of IC4M9 while not affecting the model IC4M6, which depends on frequency only.
Read the full abstract

All authors W. Erick Rogers, Jie Yu, Jean Rabault, Ana Carrasco, Malte Müller

Functional Analysis

Vector-Valued Wavelet Bases as Hilbert $\mathbb{M}_m(\mathbb{R})$-Module Bases: A Construction from Scalar Wavelets

Vector-valued multiscale representations are essential when signals or fields take values in $\mathbb{R}^m$ and component interactions carry meaningful information. Most multiwavelet and super-wavelet constructions are formulated in scalar Hilbert-space settings and typically produce channelwise scalar coefficients followed by recombination. We develop an intrinsic framework for vector-valued wavelets on $L^2(\mathbb{R}^d,\mathbb{R}^m)$ by endowing this space with a natural $\mathbb{M}_m(\mathbb{R})$-valued inner product, thereby turning it into a Hilbert $\mathbb{M}_m(\mathbb{R})$-module. This module viewpoint yields matrix-valued coefficients that encode cross-component interactions and provides canonical reconstruction through a Parseval-type identity. Within this setting, we introduce a constructive lifting procedure that builds separable multivariate vector-valued wavelet bases in $L^2(\mathbb{R}^d,\mathbb{R}^m)$ from scalar wavelet bases while preserving compact support, vanishing moments, and regularity.
Read the full abstract

Superconductivity

Machine learning reveals common features of unconventional superconductors with high transition temperatures

Superconductors with high critical temperatures that emerges beyond the phonon-mediated regime are usually considered unconventional in nature, yet unlike conventional superconductors, no broadly applicable predictive theory currently guides their discovery. Here, we use interpretable machine learning to uncover a common materials-space signature of high-$T_{\mathrm{c}}$ unconventional superconductors and develop a data-driven strategy for materials discovery. We construct a unified feature representation for each material by integrating compositional statistics, structural information, and latent representations from trained property-prediction models, followed by structure-aware filtering of an experimentally established superconducting dataset. Without using transition-temperature information, unsupervised analysis shows that cuprate and iron-based superconductors occupy a common region of materials space, characterized primarily by large electronegativity deviation and intermediate mean valence-electron number. A supervised $T_{\text{c}}$ model independently identifies the same descriptors as dominant features, providing complementary evidence for their relevance. Using this empirical materials-space prior together with the $T_{\text{c}}$ model, we prioritize candidate materials, recover recently discovered nickelate superconductors, and identify chemically distinct candidates for future investigation.
Read the full abstract

All authors Haosheng Xu, Dongheng Qian, Yijun Yu, Jing Wang

K-Theory and Homology

An Abstract Index Theorem via Rees Algebras

We develop a purely algebraic framework for index-type theorems based on the Rees construction for filtered differential graded algebras (FDGAs). Alongside the classical Rees module we introduce a smooth variant $C^\omega_{\mathcal{R}}A$, adapted to analytic arguments, and we study traces and their pointwise and coefficient-wise extensions to Rees algebras. The main result (Corollary 3.11) is an abstract index theorem: given a compatible datum of filtered differential graded associative algebras with traces $A$, $B$, $G$ with a morphism $\phi\colon A\to B$, an action $\rho\colon G\otimes A\to A$ and a morphism $i\colon G\to B$, together with a graded-central element, one has $\mathrm{tr}_B(e^{f_0+f_1}) = \mathrm{tr}_{\widehat{\mathrm{gr}}A}(e^{\gamma}h)$, where the left-hand side is the trace in $B$ and the right-hand side the trace in the Laurent series associated graded $\widehat{\mathrm{gr}}A$ of $A$. Here $f_0+f_1$ is a curvature-type element, i.e., an element of the form $d_B\beta+\beta^2$ for some odd-degree element $\beta$, while $\gamma$ and $h$ are certain elements in $\widehat{\mathrm{gr}}A$. The formalism is modelled on the Getzler rescaling technique and on the derivation of the localization formula for the loop space Chern character by Ludewig and Yi.
Read the full abstract

Information Theory

Intelligent Reflecting Surface Deployment for Low-Altitude Coverage: Illumination Geometry, Directional Characteristics, and Optimization

Terrestrial base stations (BSs) are typically configured with fixed downtilt to serve ground users, resulting in weak illumination of low-altitude airspace even under line-of-sight (LoS) propagation. In this paper, we establish a channel model that incorporates BS and intelligent reflecting surface (IRS) radiation patterns for three-dimensional (3D) low-altitude coverage while preserving the existing BS configuration. We formulate a budget-constrained IRS deployment problem that jointly determines candidate-site selection, IRS orientations, and phase shifts to maximize the worst-case signal-to-noise ratio (SNR) over the 3D low-altitude airspace. The selected sites and optimized IRS parameters remain fixed after deployment, yielding a quasi-static IRS configuration. We characterize the illumination geometry between the fixed-downtilt BS and rooftop candidates by deriving the nonnegative installation-height range satisfying the BS main-lobe condition. The separation between the mapped main-lobe height boundaries grows linearly with horizontal BS-to-site distance and decreases inversely with the number of BS antennas. We further derive an analytical lower bound on the regional worst-case normalized array gain achievable through IRS phase design over served directions with different direction spans. The resulting sufficient direction span decreases inversely with the square root of the number of IRS elements when the same worst-case normalized gain guarantee is maintained. We develop a mixed-integer alternating optimization (AO) algorithm to solve the resulting problem. Simulation results validate the analytical characterizations and show that the proposed scheme achieves higher worst-case SNR than benchmarks across different deployment budgets.
Read the full abstract

All authors Guoying Zhang, Qingqing Wu, Ailing Zheng, Xingxiang Peng, Wen Chen, Wei Feng

Machine Learning

The Safety Relay in Roleplay Jailbreaks: A Component-Resolved Causal Analysis of Harm Recognition and Refusal

Large language models are trained to follow instructions while refusing harmful requests. Jailbreaks exploit this balance to elicit content a model would ordinarily reject. Roleplay jailbreaks are especially concerning: ordinary instructions defining a persona, scenario, and task can induce compliance while the harmful request remains plainly visible. We investigate which components drive this reversal and how they affect the computations leading to the response. Across two benchmarks, three model families, and four authored wrappers, we compare matched harmful and benign requests under controlled combinations of these components. We trace internal harmful–benign contrasts at the request and where generation begins, then evaluate selected internal interventions on held-out requests. A focused study examines influential attention heads and feed-forward computations, testing their functions and downstream connections. Our analysis yields three findings. (1) Successful attacks retain the measured harmful–benign distinction at the request, while its refusal-associated expression weakens where the answer begins, a pattern we call safety-relay attenuation. (2) The contributions of roleplay components depend on the other components present. (3) Reversing selected roleplay-associated internal changes restores refusal, demonstrating their causal influence. Together, these findings connect ordinary roleplay instructions to internal changes in refusal and motivate safeguards that preserve the influence of harm-related information on the response.
Read the full abstract

Computer Vision

Learning Compositional Spatio-Temporal Video Grounding with Synthetic Curriculum

Despite the impressive progress of recent MLLMs on spatio-temporal video grounding (STVG), existing evaluations and training data focus primarily on simple queries. They largely overlook the compositional queries prevalent in real-world scenarios, where a target must be disambiguated by jointly reasoning about its attributes and relations to other entities. To bridge this gap, we propose Compositional Spatio-Temporal Video Grounding (CompSTVG), a task that requires models to process complex textual queries where every intertwined attribute and relational cue is essential for disambiguation. To facilitate this task at scale, we build a synthetic data engine that leverages a spatio-temporal scene graph as a difficulty measure and casts difficulty-controlled query synthesis as a constraint programming problem, producing difficulty-graded data for both evaluation and training. Built on this engine, we introduce STVG-CompBench, a benchmark stratified by explicit difficulty levels that jointly capture temporal complexity and spatial interference. Evaluating 11 representative STVG models on STVG-CompBench reveals that current models perform poorly on compositional queries, exhibiting a sharp performance drop that is typically obscured by overall dataset-level averages. We further construct synthetic training data and propose CurrSTVG, a curriculum reinforcement learning framework that delivers consistent gains, with the largest improvements observed on the most challenging compositional queries.
Read the full abstract

All authors Xingjian Wang, Shijian Wang, Yibo Wang, Zihao Yu, Runhao Fu, Xuelian Cheng, Zongyuan Ge

Computation and Language

Language Proficiency Assessment from Eye Movements in Naturalistic Passage Reading

Standard language proficiency tests rely on linguistic tasks such as vocabulary, grammar and reading comprehension quizzes. An alternative, cognitively motivated approach, introduced in Berzak et al. (2018), proposed instead to predict language proficiency from behavioral traces of eye movements in reading. In this work, we validate and extend this approach from single sentences to more naturalistic reading of contextualized passages in English as a second language, new proficiency measures, prediction models, and reading in an information seeking regime. We find that the approach is effective in all these evaluations. We further address two key open questions on eye movement based proficiency testing: (1) potential scoring biases that reflect the proximity of the reader's native language to English, which may undermine validity, and (2) its reliability. We find that eye movement based proficiency scores are indeed biased towards L1s that are linguistically closer to English. We propose a score debiasing method which effectively remedies this issue. The reliability analyses suggest that eye movement proficiency scores are more reliable than standard language proficiency scores. Overall, our results strengthen and broaden the empirical foundations for future eye movement based language assessment technologies.
Read the full abstract

All authors Shachar Frenkel, Ido Falah, Omer Shubi, Yevgeni Berzak

Analysis of PDEs

A stochastic convex integration scheme for the intermittent Onsager theorem in two-dimensional domains

For any $\gamma \in \left[ {0,\frac{1}{3}} \right)$, we construct $\gamma$-Hölder-continuous martingale solutions to the two-dimensional stochastic incompressible Euler equations. These martingale solutions exhibit dissipative behavior and intermittency, thereby establishing the flexible part of the intermittent Onsager theorem. The proof rests on a stochastic and intermittent variant of the Newton–Nash iteration combined with the Littlewood–Paley decomposition scheme. This composite iteration scheme incorporates new stochastic pressure, Reynolds stress and intermittent perturbations to formalize the stochastic and intermittent fluctuations.
Read the full abstract