Peer Revenue Now accepting positions

Papers

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

All of arXiv

Friday 6 November

Representation Theory

$\mathrm{X}$-evolution flows/foliations for cluster algebras: partial Bongartz (co-)completions and acyclicity

We first generalize the notion of $\mathrm{X}$-foliation on cluster complexes to the context of cluster algebras. This is a continuous generalization of Keller's green mutation, which was previously introduced in the prequel within a categorical setting. We show that the wall-crossing structure (of possibly higher co-dimension) is encoded by $\mathrm{X}$-supports, which form a partial version of Cao-Gyoda-Yurikusa's Bongartz (co-)completions for cluster algebras. Moreover, we prove that for any cell transversed nontrivially by a fixed $\mathrm{X}$-leaf, its rank is an invariant bounded below by the rank of $\mathrm{X}$. Finally, we prove the acyclicity conjecture for $\mathrm{X}$-foliation, which also implies the compactness conjecture for any $\mathrm{X}$-foliation in finite type cases.
Read the full abstract

Artificial Intelligence

Learning When to Think: Dual-Reference Offline Optimization for Adaptive VLM Reasoning

Reasoning traces improve vision-language models on complex multimodal tasks, but forcing every input to follow a long reasoning process leads to overthinking and unnecessary inference cost. We study adaptive fast/slow thinking for VLMs, where one model answers directly for simple inputs and reasons step by step when needed. Fast answers and slow traces often come from distinct behavior policies, making a single-reference offline objective misweight one mode. We propose DUAL-REFERENCE BOLT (DR-BOLT), which distinguishes the desired fast/slow mixture reference from the sampling distribution and corrects their mismatch in reward-weighted training. Fast/slow partition competition explains routing in the optimal target; an efficient source-wise correction approximates the exact objective when response formats are well separated. Matched Qwen3-VL comparisons use identical candidate banks to isolate the objective from slow-teacher quality. DR-BOLT improves score-length trade-offs over the evaluated offline and online baselines on ScienceQA and three held-out transfer benchmarks, with a 24.1% cross-benchmark slow ratio and a 2.90-fold end-to-end training speedup over reproduced GRPO. These results support adaptive reasoning in the studied two-reference setting, while broader robustness remains open. Our code is available at https://github.com/HongbinLin3589/DR-BOLT.
Read the full abstract

All authors Hongbin Lin, Sizhe Zou, Juangui Xu, Xinyue Xu, Zhuoli Ouyang, Stanislav Kolchin, Chengwei Qin, Zhongxiang Dai, Yao Shu

Materials Science

From local learning to global electronic properties in large-scale systems

Machine learning from local atomic environments offers an efficient route to electronic structure, but evaluating global electronic properties in large systems remains computationally demanding. Here we combine real space charge density learning with stochastic propagation using Kohn-Sham Hamiltonians constructed from the predicted densities. The framework avoids target system self-consistent iterations and explicit diagonalization, while hierarchical blocking enables dense grid inference within limited GPU memory. We demonstrate density of states calculations for one million Si atoms and optical and transport calculations for over one hundred thousand atoms, without training separate models for each property. This framework extends the practical scale of electronic structure calculations within Kohn-Sham DFT, opening opportunities to investigate how atomic structure governs electronic, optical and transport properties at length scales relevant to real materials and devices.
Read the full abstract

All authors Yu Cheng, Zhichang Fu, Yechen Chen, Yonggang Li, Weiqing Zhou, Shengjun Yuan

Complex Variables

A Dirichlet form approach to the potential theory of $m$-subharmonic functions

On a bounded open subset of $\mathbb{C}^n$, we study the Dirichlet form induced by an admissible pair $(\theta,\mu)$, where $\mu$ is a positive Radon measure and $$\theta=dd^c u_1\wedge\cdots\wedge dd^c u_{m-1}\wedge(dd^c \lvert z\rvert^2)^{n-m}$$ for locally bounded $m$-subharmonic functions $u_1,\dots,u_{m-1}$. We prove that locally bounded $m$-subharmonic functions belong locally to the domains of these Dirichlet forms and are quasi-continuous with respect to the induced capacities. We establish a monotonicity property for Dirichlet forms, which motivates the introduction of $\mu$-quasi-monotone convergence of $m$-subharmonic functions. The natural embeddings into these domains are continuous with respect to this convergence. These results yield new proofs of the commutativity and continuity of complex Hessian operators. Using this framework, we characterize quasi-continuity with respect to the induced capacity for an arbitrary $m$-subharmonic function in terms of its pole set having zero measure $\mu$, or equivalently, zero induced capacity. On bounded strictly $m$-pseudoconvex domains, we prove that the Bedford–Taylor capacity for $m$-subharmonic functions and the capacities induced by the corresponding Dirichlet forms have the same null sets. We further characterize $m$-polar sets in terms of Dirichlet forms and stochastic processes, extending the corresponding characterization of pluripolar sets due to Fukushima and Okada to the $m$-subharmonic setting.
Read the full abstract

Lattice Field Theory

Chiral Response of QCD Topology across the Thermal Crossover

We introduce a novel relation connecting the light-quark-mass derivative of the topological susceptibility in quantum chromodynamics (QCD) to the connected correlation between the squared topological charge and the integrated light-quark scalar density. This relation gives direct access to the chiral response of QCD topology—the local logarithmic light-quark-mass derivative of the susceptibility—from a single ensemble at fixed quark masses for each temperature, without additional simulations at shifted masses. Applying this relation to $(2+1)$-flavor lattice QCD at $m_\pi\simeq220\,\mathrm{MeV}$ over temperatures ranging from well below to well above the chiral crossover, we find that the response rises rapidly through the crossover from approximately unity toward two. This evolution corresponds to a change from nearly linear to nearly quadratic dependence of the topological susceptibility on the light-quark mass. Using the chiral response as the exponent of a local power law in the light-quark mass, we determine its effective amplitude from the topological susceptibility. This amplitude exhibits a maximum near the crossover temperature, reflecting the competition between the decreasing topological susceptibility and the increasing chiral response. These results reveal a thermal reorganization of the quark-mass dependence of QCD topology beyond the suppression of topological fluctuations alone.
Read the full abstract

All authors Heng-Tong Ding, Jin-Biao Gu, José Javier Hernández Hernández, Sheng-Tai Li, Jia Ni

Materials Science

A Monte Carlo study of thermal conductivity in black phosphorene

In this work the electronic thermal conductivity of black phosphorene is studied. Using the Monte Carlo method together with the Green-Kubo relations, a statistical modelling of the motion and scattering of electrons on acoustic and optical phonons is performed to estimate the electronic thermal conductivity. The scattering probability is expressed analytically. The obtained results show a pronounced anisotropy of the thermal conductivity: the thermal conductivity values along the armchair direction are an order of magnitude larger than those along the zigzag direction, in agreement with the experimental data for related materials. In addition, the modelling allowed a more rigorous justification of some results of previous theoretical works on the anisotropic thermal conductivity of black phosphorene.
Read the full abstract

Number Theory

Campana points on quadrics with hyperplane boundary

Let $Q\subset\mathbf P^{d+1}_k$ be a smooth quadric of dimension $d\ge2$ over a number field, and let $H_i=Q\cap\Pi_i$ be hyperplane sections for $1\le i\le d$. For every integer $m\ge2$, we prove Campana strong approximation for the pair $\bigl(Q,(1-1/m)(H_1+\cdots+H_d)_{\mathrm{red}}\bigr)$.
Read the full abstract

Artificial Intelligence

BMMS: Policy-Conditioned Measurement of Behavioral Compliance in Language Models

Behavioral evaluation of language models fixes its dimensions in advance: hallucination, sycophancy, toxicity, over-refusal. An operator in a regulated industry needs a different question answered: does a system do what a particular rulebook requires? We introduce the Behavioral Misalignment Metric Suite (BMMS), which measures policy-conditioned misalignment by taking a policy corpus as an argument and returning a compliance profile over it, for any published policy and any system that answers in natural language. The Compliance Attainment Score (CAS) measures severity-weighted attainment over the obligations that bind the system, and the Compliance Robustness Margin (CRM) the share of that attainment surviving graded adversarial pressure; Behavioral Invariance (BIV) and the Behavioral Disposition Index (BDI) are specified and released with the suite. Dimensions enter only by passing a published admission test whose binding criterion is attribution: a quantity that moves when the judge changes, holding the target fixed, is measuring the instrument. Applying it excluded one of the quantities the suite itself reported. Across 1,003 obligations from eleven documents in five registers, the admission criterion holds 94.4% precision against blinded human judgment on vendor specifications and 50.0% on analytical guidance, extraction fidelity is 94.6%, and re-scoring under judges from two other providers leaves target ordering invariant while the absolute score of a single target spans up to 0.22. We apply it to the two most detailed published behavioral specifications. Own-specification advantage is not resolved from zero once document difficulty is subtracted, but absolute self-compliance is another matter: GPT-4o fails 70.7% of the hard constraints binding it in the Model Spec published by OpenAI, where the two Claude models fail 0.0% and 9.1% of theirs.
Read the full abstract

Optimization and Control

Second-order optimality conditions for $C^{1,1}$ sparsity constrained optimization problems via disjunctive programming

In this paper, we investigate second-order necessary and sufficient optimality conditions for a class of $C^{1,1}$ sparsity constrained optimization problems via a disjunctive programming approach. More precisely, by utilizing the limiting/Mordukhovich second-order subdifferential of the associated Lagrangian function, we establish new second-order necessary and sufficient optimality conditions for local optimal solutions. These results are derived under mild assumptions and extend several existing results in the literature. In addition, we apply our theoretical developments to sparse multiobjective optimization problems and derive second-order optimality conditions for efficient solutions.
Read the full abstract

Machine Learning

Augmentron: Scalable Multi-View Visual Augmentation for Robot Learning

Robot policies trained on visually narrow demonstrations can become sensitive to environmental features that are unrelated to the task. Existing generative augmentation methods can introduce visual diversity, but processing every frame and camera view independently is computationally expensive and may produce inconsistent observations across cameras and timestamps. We introduce Augmentron, a geometry-aware pipeline that generates reusable scene representations and renders them through recorded camera poses. This produces temporally and multi-view consistent environ- ment and workspace variations while preserving the robot, task objects, actions, and annotations. Across three physical manipulation tasks, replacing half of the real world training dataset with Augmentron data improves model success across three tasks for both in out-of-distribution distractor objects, and out-of-distribution workspace variation evaluations. In a separate comparison against alternate augmentation techniques, Augmentron achieves the highest pooled success in all three conditions. The results show that introducing augmented data to a data training recipe can help policies generalize to dynamic real-world settings.
Read the full abstract

All authors Tanay Tandon, Aseem Doriwala, Jade Choghari, Catherine Weaver, Pragna Mannam

Optics

An acousto-optic colorimeter for color psychophysics

A complete model of human color perception requires the ability to present any color as a stimulus.Most existing measurements were collected with physical color samples, and current laboratory generators, such as displays and multi-primary LED systems, work with a small number of fixed primaries. Published color-difference data therefore cover about 40% of the CIE~$xy$ chromaticity diagram, while commercial wide-gamut displays already reach two thirds of it and need difference metrics there. We describe an acousto-optic colorimeter that lets the experimenter choose the primaries. The device selects up to four narrow bands from a supercontinuum (white-light) laser, each with electronically set wavelength and intensity, and covers 95% of the CIE~$xy$ chromaticity diagram. Threshold-scale color differences, presented in alternation as in a discrimination experiment, are reproduced with a mean error of 0.08–0.12 CIEDE2000 units ($\Delta E_{00}$) at five CIE-recommended color centers. Because one color can be built from many combinations of narrow bands, the bands are chosen for the task: the finest adjustment for color matching, the least disagreement between observers for color-difference measurement. In simulations with 151~individual observers from a published dataset, with these bands observers see colors 0.6–1.9~$\Delta E_{00}$ from the intended one on average, against 1.2–2.5~$\Delta E_{00}$ on a calibrated wide-gamut LCD, lower at every center. The approach applies to any narrowband source and lets experimenters use saturated colors while limiting the effect of individual differences in color vision.
Read the full abstract

All authors Sergey Korchagin, Georgiy Lobarev, Pavel Aksentsev, Pavel Maximov, Maria Gracheva, Olga Polschikova, Egor Ershov

Dynamical Systems

Complement Cancellation and the Spectral Characterization of Three-Point Recurrence

In this paper, we investigate both two-sided and one-sided \(2\)-nice recurrence within the framework of weakly mixing measure-preserving systems. As our principal result, we establish an exact spectral characterization for the failure of two-sided \(2\)-nice recurrence: namely, a subset \(R\subseteq\mathbb N\) fails to possess this property if and only if there exist a continuous symmetric probability measure \(\sigma\) on \(\mathbb T\) and a constant \(\varepsilon>0\) satisfying \[ 2\widehat{\sigma}(n)+\widehat{\sigma}(2n)\leq-\varepsilon \qquad(n\in R). \] Our proof of the necessity relies upon a complement-symmetrized correlation identity, whereas the converse is achieved through a construction involving a stationary Gaussian system, a bounded odd observable, Hermite expansions, and an independent Bernoulli extension. Furthermore, we extend this characterization to three-point polynomial configurations as well as to actions of countable abelian groups. In addition, we derive an unconditional analogue in which arbitrary symmetric probability measures take the place of continuous ones, thereby demonstrating a strict separation between the ergodic and weakly mixing classes of recurrence. Subsequently, we turn our attention to the distinction between two-sided and ordinary one-sided \(2\)-nice recurrence. It is shown that a suitable negative spectral condition compels the failure of one-sided recurrence. More remarkably, we construct an explicit weakly mixing system wherein the quadratic spectral expression remains uniformly positive, yet the presence of a negative cubic correlation still induces a one-sided recurrence deficit. Consequently, it follows that two-point spectral data alone are insufficient to govern one-sided \(2\)-nice recurrence.
Read the full abstract

Machine Learning

Learning Generative Dynamics for 3D Molecule Generation via Sequential Neuro-Symbolic Constraints

Diffusion models have recently emerged as a powerful paradigm for generating 3D molecules. Despite showing remarkable promise, they learn atomic distributions implicitly from data, without explicitly encoding physicochemical rules into their generative dynamics. Consequently, generated samples can be statistically plausible yet chemically invalid. To address this problem, we propose Sequential Neuro-Symbolic Constrained Diffusion (SensDiff), a framework that integrates symbolic rules into generative modeling, enabling the model to internalize scientific priors during training. Central to SensDiff is the observation that the diffusion trajectory of 3D molecule generation exhibits a coarse-to-fine evolution: valid molecules arise by first mitigating macroscopic geometric conflicts and then refining structures toward microscopic chemical consistency through iterative denoising. Accordingly, SensDiff mirrors this dynamic by sequentially enforcing constraints with adaptive weighting during training, progressively steering generation toward valid molecular geometries. Moreover, SensDiff translates domain knowledge into interpretable generative controls, grounding opaque neural denoising in scientific principles. Experiments on molecular benchmarks and real-world constrained generation tasks confirm that SensDiff consistently improves chemical validity while adhering to physicochemical rules and user-specified targets. The code is publicly available at https://github.com/WanyuGroup/NeurIPS2026_SensDiff.
Read the full abstract

All authors Zhuoran Li, Xu Sun, Chang Wen Chen, Wanyu Lin

Emerging Technologies

Circuit Theory-Informed AI for Condition Monitoring in Power Electronics

Artificial intelligence for power electronics is evolving toward a deeper comprehension of system intrinsic behaviors and dynamics. As a novel integration of neural computation and electric circuits, spike-based neuromorphic computing offers promising potential in high energy efficiency and adaptability for data-driven applications. Building on the latency-coding principles, this article proposes a framework for data-light condition monitoring that relies solely on three-phase AC current measurements. The AC current signals are encoded as membrane potentials of RC-based spiking neurons, which are inherently compatible with spiking neural networks that can effectively facilitate the estimation of device parameters. Case studies are presented to validate the performances, while constraints and further optimizations are also evaluated and discussed, and future extensions of the methodology are provided.
Read the full abstract

Sound

SPDMX: Large-Scale Open Stem Audio From Symbolic Music

Large public datasets have driven recent progress in machine learning, from text and vision to speech and general audio. Music remains an outlier, as copyright blocks redistribution of training audio, in particular for modern tasks that need individual stems rather than mixtures. Existing multitrack resources trade off realism, scale, and licensing, as synthetic alternatives rely on proprietary VSTs or naive open rendering that yields weak stems. We introduce SPDMX, an openly-licensed stem corpus rendered from the public-domain PDMX dataset. SPDMX comprises $5{,}469$ song-hours ($2{,}734$ song-hours, or $34.8\%$, of which are multitrack scores) at $44.1$ kHz under CC BY 4.0, with paired MIDI and identifiers that link our dataset to PDMX's original metadata. To our knowledge, SPDMX's multitrack subset constitutes the largest openly-licensed synthetic multitrack subset by song-hours. We then demo SPDMX as a dataset for both source separation and transcription, showing how its increased scale can enable better training of models on synthetic data. We release SPDMX, as well as the open creation pipeline for reuse on other symbolic music datasets.
Read the full abstract

All authors Phillip Long, Eduardo Escoto, Julian McAuley, Zachary Novack

Artificial Intelligence

CARGO: Causal Attribution and Responsibility-Guided Preventive Safe Multi-Agent Reinforcement Learning for Cooperative Driving

This study presents Causal Attribution and Responsibility-Guided Optimization (CARGO), a multi-agent reinforcement learning framework for cooperative on-ramp merging in mixed traffic. A shared latent-context posterior and an action-conditioned response model estimate directed behavioral influence, with posterior uncertainty propagated into interaction-edge selection. Complementary attribution mechanisms couple signed influence with marginal team utility for cooperative reward shaping and assign responsibility from positive excess tail risk relative to a safe reference policy. Conditional diffusion generates future responses for action-dependent risk estimation. Responsibility-guided preventive optimization calibrates agent-specific risk-reduction requirements to local capacity within a policy trust region, using reserved safety margins, adaptive requirement scaling, and joint feasibility verification. Theoretical analysis establishes risk monotonicity and sufficient local conditions for non-decreasing environmental return under explicit assumptions. Experiments in highway-env across three traffic densities, together with controlled graph-recovery and attribution studies, support the effectiveness of the framework. Under high-density traffic, CARGO increases the success rate by 66.04% and reduces the crash rate by 56.23% relative to the best evaluated baseline for each metric.
Read the full abstract

Machine Learning

Deep Learning Theory through the Lens of Statistical Physics: A Topical Review

The recent success of foundation models and deep learning has far outpaced our theoretical understanding of how large-scale neural networks learn and generalize. Statistical physics provides a natural language for describing such high-dimensional, nonlinear, and strongly coupled systems, offering concepts such as mean-field theory, phase transitions, energy landscapes, and stochastic dynamics. In this topical review, we adopt a statistical-physics way of thinking – from concepts including phases and criticality, and free-energy viewpoints, to methods such as kernel and Gaussian-process limits, mean-field and dynamical mean-field theories, random-matrix tools, and Fokker-Planck formalisms, and to a mindset emphasizing universality, scaling, and tractable limits with explicit validity regimes. We survey recent progress on understanding modern deep learning through this lens, including mean-field analyses of wide networks, spectral and geometric perspectives on loss landscapes, stochastic-gradient-based training viewed as a dynamical process, and phase-transition analogies that organize phenomena such as overparameterization, generalization, and representation change. Our goal is to synthesize these threads into a coherent framework that clarifies what each approximation explains, when it is expected to hold, and how apparently competing narratives relate or diverge. Together, these perspectives form a unified physical framework that views representation learning as a collective phenomenon in high-dimensional systems, bridging deterministic and stochastic regimes of gradient-based optimization.
Read the full abstract

Numerical Analysis

Elastic Wave Imaging via Kirchhoff Migration: Resolution Analysis and Inclusion Identification

We consider the inverse problem of determining the location, orientation, shape, and material parameters of a small elastic inclusion embedded in a homogeneous medium from full-aperture measurements of the scattered field. Using a small-volume expansion, we derive explicit leading-order expressions for four Kirchhoff migration imaging functionals, denoted by PP, PS, SP, and SS, corresponding to the four combinations of pressure (P) and shear (S) waves used for receiver- and source-side backpropagation. The resulting point-spread functions quantitatively characterize the resolution of each mode and show that the SS mode achieves the highest spatial resolution. For an elliptic inclusion, we show that the PP mode uniquely determines the elastic moment tensor (EMT) within the leading-order asymptotic model, thereby enabling the reconstruction of the material and geometric parameters of the inclusion. For a general small scatterer, the reconstruction may instead be interpreted as an effective elliptic representation determined by its leading-order EMT. For a disk, the parameter-identification problem is underdetermined; nevertheless, explicit upper and lower bounds on its volume are derived. We further analyze the model error arising from neglected cross-mode contributions when the distance from the center of the measurement configuration is not sufficiently large relative to the elastic wavelengths. Numerical experiments support the theoretical predictions and demonstrate the effectiveness of the proposed imaging and reconstruction methods.
Read the full abstract

Formal Languages and Automata

Shortest Synchronizing Words on Product Automata

We consider some operations on finite automata graphs, in particular various product, union and merge operations, and investigate how they relate to the parameter~$\mathrm{rt}$, the length of a shortest synchronizing word. We derive some lower and upper bounds for the operations that we study. We also examine several classes of automata (for instance, partially ordered, commutative or idempotent automata) that are preserved under these operations and show in which way the parameter $\mathrm{rt}$ behaves in these automata classes. Also, we consider automata classes that can be constructed by combining automata from better known classes; in particular, these new classes of automata will also satisfy the bound of the famous Černý conjecture. This also motivates us to study even more closure properties. In particular, we show that certain automata classes related to synchronization form pseudo-varieties. We also commence a similar study concerning the length of shortest homing sequences of Mealy automata.
Read the full abstract

Numerical Analysis

Strong convergence rates for numerical approximations of the stochastic Burgers equation with additive rough noise

We establish strong convergence rates for a fully discrete approximation of the one-dimensional stochastic Burgers equation driven by additive rough noise $(-A)^\alpha dW$, $0\le\alpha<1/4$, below the critical roughness threshold $\alpha=1/4$. Through the Cole–Hopf transformation, the nonlinear equation is converted into a linear stochastic heat equation with spatially correlated multiplicative noise, which is discretized by a spectral Galerkin method in space and an accelerated exponential Euler method in time. A numerical approximation of the original solution is then reconstructed through a scale-invariant regularized inverse Cole–Hopf transformation. The convergence analysis combines strong Sobolev error estimates with strict positivity and uniform inverse-moment bounds for the exact transformed solution. Under an initial approximation condition and a weak-$L^4$ tail assumption on the transformed initial datum, we obtain the spatial strong convergence rates arbitrarily close to $1/2-2\alpha$ and a temporal rate of order $1/2$, up to a logarithmic correction. For $0\le\alpha<1/4$, the spatial rate is essentially sharp within the spectral Galerkin framework, the temporal rate is independent of the roughness parameter $\alpha$, and we prove a converse estimate showing both the necessity of the initial approximation condition and the weak-$L^4$ tail of the initial datum.
Read the full abstract

Logic

Formalization of Filter-Method Construction for Real Numbers in Coq

This paper presents a formalization for a theory of constructing real numbers using concepts related to filters. The formalization is implemented using the Coq proof assistant and is grounded in axiomatic set theory. The theory encompasses various notions such as filters, ultrafilters, principal ultrafilters, and arithmetical ultrafilters. A special filter, namely the non-principal arithmetical ultrafilter, whose existence requires the Continuum Hypothesis to be proven, is employed to establish several non-standard number sets: *N, *Z and *Q. The set of real numbers, R, is subsequently derived by equivalently classifying a specific subset of *Q. The entire theory is thoroughly formalized, with each detail verified to ensure rigor and precision. This work contributes a new selection of foundational materials for the formalization of mathematical theories.
Read the full abstract

Computer Vision

DINO-Nucleus: Multi-Level Self-Supervised Features for Nucleus Instance Segmentation

Nucleus instance segmentation and counting are essential for quantitative microscopy, enabling reliable analysis of cell populations, tissue organization, and disease-related changes. However, accurately separating touching nuclei remains difficult, and many existing approaches rely on complex models or sensitive post-processing procedures. We present DINO-Nucleus, a parameter-efficient framework that leverages multi-level self-supervised DINOv2 representations for nucleus instance segmentation and counting. The model jointly predicts nucleus regions, pixel-wise instance flows, and explicit nucleus centers, while a consistency objective encourages the predicted flows to converge toward their corresponding centers. Multi-level transformer features provide complementary spatial and semantic information, and the pretrained backbone is adapted using lightweight LoRA modules to reduce fine-tuning cost. Experimental results show that DINO-Nucleus achieves strong segmentation and counting performance with parameter-efficient backbone adaptation. These findings demonstrate that parameter-efficient adaptation of multi-level self-supervised features can provide an effective solution for nucleus instance segmentation in microscopy images.
Read the full abstract

High Energy Physics (Phenomenology)

Stimulated Axion Decay around Primordial Black Holes and the Diffuse Radio Background

We study stimulated axion decay in dense axion halos around primordial black holes (PBHs), assuming that both PBHs and axions contribute to dark matter. We construct the axion density profile by matching a hydrogenic soliton-like core dominated by the PBH potential to the $r^{-9/4}$ spike produced by secondary infall. We also impose an inner cutoff at the innermost stable circular orbit. We derive local Boltzmann equations for the number densities of photons and axions, including spontaneous and stimulated decay, inverse decay, photon escape, and axion filling. We solve these equations numerically and with a semi-analytic approximation. Stimulated emission occurs within an inner region where photon amplification overcomes escape. We calculate the photon emission from an individual PBH halo and integrate it over the population of PBHs throughout cosmic history. By requiring the resulting diffuse radio spectrum not to exceed the radio background measured by ARCADE~2, we derive upper bounds on the fraction of dark matter in PBHs. The bounds are strongest at intermediate PBH masses and apply over a wider mass range as the coupling between axions and photons increases.
Read the full abstract

Optics

Raytracing and quantifying the inferior mirage

Using custom raytracing code, we explore several features of the inferior mirage that are not widely discussed, such as the blurring in the mirage image, and what happens to the light in the direct image region that is replaced by the mirage image. The raytracing model also clarifies how to quantify the mirage, and we provide expressions to calculate the angular extent of the mirage under given thermal conditions and measurement geometry.
Read the full abstract

All authors Nathan Hagen, Lucas Reyes Bentancor, Kakeru Tanaka, Norimasa Hanaya, Momo Shoda, Hiroki Maki, Takuma Ando, Keisaku Yoshizawa, Kayo Suzuki, Yume Yamada, Asana Kumakura

Robotics

Auditing Grasp Feasibility Predictors with Planner-Verified Counterfactuals

Grasp-feasibility predictors report near-perfect accuracy on Task and Motion Planning benchmarks, but high accuracy on these benchmarks does not verify that a predictor uses the collision geometry that the motion planner checks. We propose a planner-verified counterfactual (CF) test: holding the grasp pose fixed, we insert a solid obstacle into the approach corridor, add its points to the model input, and let a planner re-derive the label, which flips from feasible to infeasible. Across eight predictors spanning point clouds, voxels, depth, and object tokens, CF-pair AUROC falls from high natural AUROC (0.920-0.996) to chance (0.470-0.504), and the flip holds across four planners, two arms, and a real-object obstacle. Predictors fail at different resolutions, from a bounding box to a 12.5 cm voxel. In the point-cloud models we dissect, the obstacle either moves the representation in a direction unrelated to whether it blocks the grasp or is not registered at all. A clearance-aware read of the corridor points, trained on the unmodified benchmark with no counterfactual scenes, reaches 0.983 CF-pair AUROC, but it remains bound to its trained obstacle placements; a placement-invariant clearance representation is still open. We release the counterfactual labels, metrics, datasheet, and relabeling pipeline at https://github.com/codewithsajid/graspfc-cf.
Read the full abstract

Computer Vision

TouchAnnotator: Multi-Source Tactile Supervision with Privileged-View Distillation

Egocentric video captures human-object interaction at scale but rarely includes tactile labels for hidden contact and pressure. We introduce TouchAnnotator, an ego-only bimanual tactile annotator trained with multi-source tactile supervision and privileged-view distillation. A canonical hand field unifies glove pressure, geometry-derived human contact, and robot fingertip deformation from six datasets and three embodiments, while source-aware masks preserve valid targets. During training, an EMA teacher observes synchronized ego and wrist views and distills hand-local representations and tactile predictions into a student that uses only ego video and estimated pose at inference. Compact past-clip memory supplies temporal context. We evaluate on EgoTouch, zero-shot FPHA transfer, and 860 synchronized glove recordings. The model transfers to bare-hand contact and produces qualitatively coherent tactile responses across bare-hand, gloved-hand, and robot embodiments. On measured recordings, it improves temporal accuracy by 5.38 points and Volumetric IoU by 5.02 points under a shared protocol. These results show that heterogeneous human–robot supervision and privileged training views can turn egocentric video into dense tactile annotations.
Read the full abstract

All authors Ao Li, Zhihao Yuan, Chubin Zhang, Kunrui Huang, Sheng Xu, Zengjue Chen, Junjie Wang, Jiawei Li, Yansong Tang

Earth and Planetary Astrophysics

Estimations for strong and weak hot Jupiter magnetism: A Python package with interpolators and scaling laws

Hot Jupiters (HJs) are commonly thought to host the strongest dynamo-generated magnetic fields among exoplanets, exceeding Jupiter's by up to an order of magnitude, which can potentially induce observable star-planet interaction (SPI) and planetary radio emission. However, while several indirect, model-dependent estimates of HJ magnetic fields have been reported from SPIs and recent transmission spectroscopy diagnostics, no unambiguous direct (radio) detection has yet been achieved. We aim to provide a range of magnetic field estimates for HJs, considering the mechanisms responsible for inflation and the convective flux entering the magnetic field scaling law. Given that their dynamos operate in the fast-rotating regime (Rossby number <~ 0.1), we combine one-dimensional evolutionary interior models with magnetic field scaling laws under two limiting heating scenarios: deep and shallow energy depositions. The deep-heating scenario maximizes convective heat flux in the dynamo and predicts strong magnetic fields of up to ~100 G, whereas shallow heating, which is arguably more realistic for most proposed physical mechanisms, weakens convection and yields much weaker fields of O(1) G. The strong field scenario is compatible with SPI indirect detections, and the weak field scenario is in agreement with recent atmospheric constraints. We derive scaling law relations based on observables (planetary mass, radius, and/or incident stellar flux): Bstrong = 19 M_P^0.6 R_P^1.7 G; Bweak = 0.8 M_P^1.5 R_P^-1.2 G. These predictions provide a framework for interpreting future atmospheric wind measurements, SPI signatures, and radio observations, which may ultimately discriminate between competing interior-heating scenarios.
Read the full abstract

All authors Albert Elias-López, Aline Vidotto, Daniele Viganò, J. R. Fuentes, Simon Müller, Emily Sandford

Analysis of PDEs

Maximizing Reaction via Spatial Localization: The Reaction-Flux Landscape and Maximizing Sequences

We continue our study of a PDE-constrained variational model for maximizing the total reaction flux by optimizing the spatial distribution of catalytic materials, such as enzymes. The total amount of enzyme is fixed, and the substrate concentration is governed by a reaction-diffusion equation with prescribed boundary values. We consider both linear and nonlinear Michaelis-Menten reaction kinetics. In previous work, we constructed a maximizing sequence, computed the first and second variations of the reaction functional, and established the nonexistence of local maximizers. Here, we conduct a more comprehensive analysis. We prove that the functional is strictly concave for linear kinetics but need not be concave or convex for nonlinear kinetics. We also show that no critical points exist. We then characterize maximizing sequences and identify all their possible limiting measures. Our analysis reveals the mechanism of reaction maximization through spatial localization: maximizing sequences concentrate enzymes near the portion of the boundary where the substrate concentration is largest. Moreover, the landscape of the functional defining the total reaction differs between the linear and nonlinear cases, a point less emphasized in applications. Consistent with existing biophysical studies, these findings motivate the development and analysis of regularized models.
Read the full abstract

Information Retrieval

PAGSL: Path-Augmenting Graph Structure Learning for explainable scholar recommendation

Explainable scholar recommendation aims to identify potential scholars of interest and provide reasonable explanations for the recommendations. However, existing frameworks struggle with optimization-induced noise and "hub node bias", yielding trivial explanations that lack hierarchical depth and personalization. To this end, we propose a novel explainable scholar recommendation model, Path-Augmenting Graph Structure Learning (named PAGSL). PAGSL combines representation learning and semantic path knowledge to ensure recommendation performance while providing interpretable results from two perspectives. Specifically, it activates latent academic paths among entities through graph augmentation and employs a path-based neural network to encode these paths for recommendation. A masking mechanism with path score optimization is designed to generate concise and informative explanations. We evaluate PAGSL on three academic datasets (Scholat, AMiner, and ResearchGate), comprising approximately 49,000 authors and over 330,000 interaction records. Experimental evaluations validate that PAGSL outperforms state-of-the-art baselines, achieving significant improvements of 6.31% in F1-score, 6.29% in Accuracy and 5.95% in AUC for recommendation. Crucially, in terms of explainability, we provide quantitative results and visualization effects. Furthermore, ablation studies validate the effectiveness of our heterogeneous convolution and path representation learning. PAGSL achieves strong performance in both recommendation accuracy and explainability.
Read the full abstract

All authors Jiale Xu, Mingshu Ai, Junjie Huang, Tao Jia