Peer Revenue Returns subject to peer review

Papers

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

All of arXiv

Tuesday 1 September

Machine Learning

CopyShield: A Cross-Level Benchmark of Copyright Defenses in LLMs

Large language models can reproduce memorized text verbatim, yet copyright defenses are usually evaluated under incompatible protocols. We introduce CopyShield, a controlled benchmark comparing three representative defenses at distinct intervention levels: contrastive decoding (output), Direct Preference Optimization (behavioral), and activation intervention (representation). We evaluate CopyShield on two model families, LLaMA-3.1-8B and Mistral-7B-v0.3, using controlled memorization over five public-domain books and a shared protocol measuring literal leakage, calibrated non-literal leakage, utility, and degeneracy. Across these methods, intervention level is associated with distinct compliance-utility trade-offs. On LLaMA-3.1-8B, contrastive decoding remains near-degeneracy-free (0-2%) but reaches a literal-suppression floor at NV-Recall 0.192-0.203. DPO nearly eliminates literal leakage (0.263 to 0.002) but induces paraphrase-loop degeneracy in 58% of QA outputs, with no utility gain over the SFT baseline. Activation intervention attains the lowest non-literal flagging rate (1/200) by blocking 84% of non-literal queries before generation. Human evaluation confirms that DPO has low coherence, whereas activation lowers perceived copyright risk through broad refusal. On Mistral-7B-v0.3, the output- and representation-level patterns persist, while DPO degeneracy falls to 10-14%, showing that its severity is model-dependent. Together, CopyShield provides cross-level reference baselines and identifies targeted non-literal suppression as an open challenge. The code is available at https://github.com/spotai-mbzuai/CopyShield.git.
Read the full abstract

All authors Maryam Alshehyari, Dushyant Singh Chauhan, Samuele Poppi, Martin Takac, Salem Lahlou, Nils Lukas

Mesoscale and Nanoscale Physics

Layer-selective and magnetic-field-enhanced transport of topological kink states in rhombohedral multilayer graphene

Topological valley kink states (VKSs), which are quantum valley Hall states emerging at the interfaces between adjacent domains with opposite valley Chern numbers, have attracted considerable interest in graphene-based systems. In this work, we investigate the quantum transport of VKSs in ABC-stacked rhombohedral multilayer graphene in the presence of Anderson disorder and a perpendicular magnetic field. Two prominent transport characteristics are revealed. First, in the absence of a magnetic field, the kink states exhibit strong layer polarization, with their wave functions predominantly localized and equally distributed on the outermost top and bottom layers. As a result, their transport properties are highly sensitive to the layer-selective disorder distribution. Second, under a perpendicular magnetic field, the layer-symmetric spatial distribution of VKSs is broken, leading to a significant reduction in the wave-function overlap between counter-propagating VKSs from opposite valleys. Consequently, intervalley scattering is suppressed, and the transmission of VKSs through disordered regions is substantially enhanced. Our results provide new insights into multichannel topological valley transport in rhombohedral multilayer graphene and demonstrates disorder-engineering and magnetic fields as effective approaches for manipulating the propagation of VKSs.
Read the full abstract

Pattern Formation and Solitons

Chemotaxis-induced linear instabilities and pattern formation in a reaction-diffusion model

We study linear instabilities and pattern formation in a spatially extended Selkov model for glycolysis with diffusion and chemotactic interactions between two species. Linear stability analysis reveals two distinct bifurcations: a zero-wavevector, finite-frequency Hopf bifurcation leading to spatially homogeneous oscillations, and a finite-wavevector Turing instability with a saddle-node bifurcation leading to stationary inhomogeneous patterns. The selected wavevector, $k_c$, depends sensitively on the nature and strength of chemotaxis and varies nonmonotonically with the chemotaxis parameters, allowing re-entrant transitions between homogeneous and patterned states. For sufficiently strong chemotaxis, when the two species mutually attract or mutually repel each other, an unusual instability emerges in which the preferred wavevector diverges as a finite threshold in chemotactic strength is approached. We complement the linear analysis with extensive direct numerical simulations (DNS) of the nonlinear partial differential equations in two dimensions. The DNS confirm the predicted instabilities and their nonmonotonic and re-entrant character, and also reveal spatially uniform oscillatory states for suitable parameters. The stationary patterns include spots and, unexpectedly, stable stripes. We further uncover transitions between spot and stripe morphologies controlled by chemotactic strength and diffusivity ratios. The stability of the striped states is consistent with linear amplitude equations, whereas the spot-stripe transition is attributed to nonlinear effects beyond linear stability. Our results demonstrate that chemotaxis can act as a nonmonotonic control parameter for instability selection, wavelength selection, re-entrant pattern formation, and nonlinear pattern morphology in reaction-diffusion systems.
Read the full abstract

Machine Learning

Superposed Latent Autoencoder

Autoencoders typically meet tight latent-memory budgets by making each latent representation smaller, sacrificing representational capacity. We ask a different question: can multiple wider latents be stored together instead? We introduce the Superposed Latent Autoencoder (SLAE), which preserves high-capacity latent representations while sharing storage through learned superposition. SLAE transforms latents into storage-friendly codes, binds them with randomized keys, superposes multiple codes into a single memory tensor, and learns to recover each latent before decoding. Under the same storage budget, SLAE replaces irreversible dimensional bottlenecks with structured interference that can be suppressed. Across CIFAR-10/100, SVHN, STL-10, Tiny ImageNet, and a wide range of memory budgets, SLAE substantially improves the reconstruction–memory tradeoff, reducing reconstruction error by up to 56% over conventional autoencoders at matched storage. Further analysis shows that SLAE's advantage comes from making wider representations usable under the same storage budget. These gains also extend beyond reconstruction: the information preserved by SLAE improves downstream classification by up to 16.79 percentage points under the same memory budget. Our results suggest a new principle for representation compression: instead of making every latent smaller, keep representations wide and let them share memory.
Read the full abstract

All authors Quanling Zhao, Jiaying Yang, Tianqi Zhang, Ziyang Hao, Fatemeh Asgarinejad, Flavio Ponzina, Tajana Rosing

Quantum Physics

A continuous-mode quantum-optical representation of finite strong-field laser pulses in free space

We formulate an energy-consistent quantum-optical representation of finite paraxial laser pulses in free space. The formulation is based on a continuous-mode description and provides a step toward a unified treatment of coherent and nonclassical strong-field drivers. Starting from plane-wave quantization, we construct a transverse vector-mode reduction in which the selected mode can reproduce an arbitrary normalized spatial–polarization profile. For a coherent pulse, the measured pulse energy and full laboratory analytic signal determine the frequency-dependent coherent-state displacement $(\alpha(\omega))$, without introducing a physical quantization volume. For nonclassical pulses, the same first-order field data and total energy do not uniquely specify the quantum state. Additional correlation information or model assumptions are required. We incorporate this information through normal and anomalous covariance kernels and discuss $g^{(2)}$-based diagnostics of squeezed pulses. We also derive a phase-sensitive convergence criterion for numerical frequency-bin discretizations. An exactly solvable continuum-mode free-electron application shows that the coherent displacement reproduces the semiclassical mean trajectory, whereas the field covariances determine the electron wave packet width. The framework offers a convenient, energy-consistent interface between measured free-space pulses and quantum-optical calculations.
Read the full abstract

Earth and Planetary Astrophysics

Current water trapping micro-habitats on the surface of Mars

This paper presents a conceptual model exploring how liquid water could be maintained in direct contact with hypothetical microorganisms during warm daytime periods on Mars. Hygroscopic surface salts (e.g., NaCl, CaCl$_2$, Ca(ClO$_4$)$_2$, gypsum, MgSO$_4$) can absorb water vapor and form liquid solutions; however, daytime exposure normally causes rapid drying. We propose a mechanism where diurnal temperature fluctuations drive thermal expansion and contraction cycles that open and close micro-cracks in salt crystals. Nighttime-condensed deliquescent liquid penetrates these fractures, which then partially close during warmer daytime hours, trapping liquid water inside and preventing immediate evaporation. Thermally induced volume changes (ranging from ~0.5-0.9% in crystalline phases up to several percent in concentrated solutions) support the feasibility of this cyclic aperture dynamic. Modeling deliquescence and salt distribution indicates that nighttime microscopic liquid can form for ~100-130 sols per Martian year near Acidalia Planitia. At depths of 2-3 mm within soil or salt structures, hypothetical microorganisms would gain partial UV shielding while retaining access to visible light. Overall, internal crystal micro-fractures may offer transient daytime liquid microenvironments, highlighting these locations as potential candidate habitats for photosynthetic life on modern Mars.This study serves as a conceptual feasibility analysis for near-surface brine retention within hygroscopic salt structures under present-day Martian conditions.
Read the full abstract

Pattern Formation and Solitons

Helical and Straight Solitons Induced by Vortex Beams with Off-Axis Pivots in Cubic-Quintic Nonlinear Media

We address the existence and dynamics of solitons propagating along helical (spiral-rotating) and straight trajectories in the bulk waveguide with the cubic-quintic nonlinearity and a single- or multi-ring potential. The input is taken as a vortex beam with a single or multiple phase singularities (pivots), displaced from the waveguide's axis. The single off-axis pivot creates a one-ring helical soliton, with a closed or open (split) ring structure. The vortex beams with multiple off-center pivots give rise to complex states with multi-ring, multi-core, and/or multi-split structures. The multi-core/split solitons propagate along helical channels, or straight ones, which are parallel to the propagation axis. The sign and period of the helicity can be precisely adjusted by applying a torque with an appropriate angular velocity.
Read the full abstract

Cosmology and Nongalactic Astrophysics

The cosmic multipoles: a series of consistency tests for the Szekeres cosmological model

An important property of the quasi-spherical Szekeres cosmological model appears to be widely misunderstood in the literature. In this paper, the claim that the matter density can be fully decomposed on a given 2-shell into a dipole superposed on a monopole is shown to be inaccurate and exact propositions are substituted. Then, the multipole decompositions of three different quantities linked to the mass density are displayed and proposed to constrain the Szekeres parameters with observed data. A test of the axial symmetry assumption, together with that of an on-axis observer, using a property of the angular coefficients is also suggested.
Read the full abstract

High Energy Astrophysical Phenomena

Constraints on Buchdahl-Inspired Gravity from Future Pulsar Timing near Sgr A*

Future pulsar timing observations near Sgr~A* offer a unique probe of gravitational physics in the vicinity of a supermassive black hole. We forecast the ability of such measurements to constrain a Buchdahl-inspired $R^2$ gravity, parameterized by a single deviation parameter $\epsilon$, using a timing framework that self-consistently integrates orbital dynamics with light-propagation delays and preserves the full timing solution across the observing span. Through Fisher-matrix forecasts for a representative pulsar, we systematically isolate how the precision on $\epsilon$ depends on orbital geometry. We find that shorter orbital periods and higher eccentricities significantly enhance sensitivity, consistent with a substantial contribution from observations near periastron. As a benchmark comparison, we further consider a hypothetical pulsar on an S2-like orbit ($P_b=16~{\rm yr}$, $e=0.88$) and obtain a statistical sensitivity of $\sigma_\epsilon\sim 10^{-4}$ within the adopted weak-field, static, and spherically symmetric timing model. This sensitivity is comparable to the natural order-of-magnitude truncation scale of the 1PN expansion and should not be interpreted as a complete forecast for the real Sgr~A* system. Under the adopted idealized assumptions, the characteristic statistical scale is several orders of magnitude below the current S2 95% confidence interval half-width ($|\epsilon|_{\rm S2}^{\rm 95\%}\approx 0.56$), though this comparison is heuristic given the differing confidence levels. These trends provide quantitative guidance for target selection and campaign design in future Galactic-center pulsar searches.
Read the full abstract

All authors Jian-Ming Yan, Tao Zhu, Zong-Kuan Guo, Zhao Li

Astrophysics of Galaxies

The fraction of clumpy star-forming galaxies in nearby galaxies from CLAUDS and HSC-SSP

Massive, star-forming clumps are regions of intensive star-formation that are commonly observed in high-redshift ($z > 1$) galaxies. Observations of low-redshift clumpy galaxy analogues are rare but the availability of wide-field galaxy survey data makes the detection of large clumpy galaxy samples much more feasible. We present a population of 12,790 star-forming clumps detected in a mass-complete sample of 5,395 star-forming galaxies (SFGs) at redshifts $z\leq0.32$, located in the XMM-LSS, E-COSMOS and DEEP2-3 fields observed by the Hyper Suprime-Cam Subaru Strategic Survey (HSC-SSP) and CFHT Large Area U-band Deep Survey (CLAUDS). The clumps were detected using an improved version of our Deep Learning (DL)-based object detection framework which uses the ZOOBOT foundation DL-model as a 'backbone' feature extractor. We determined the fraction of star-forming galaxies hosting at least one off-centre clump ($f_{\mathrm{clumpy}}$) based on a clump definition that requires a clump-galaxy flux ratio in the CLAUDS u-band of $\geq8\%$. We estimate $f_{\mathrm{clumpy}}$ to decrease from $\sim$31% at $z\sim0.3$ to $\sim$23% at $z \sim 0.1$, which aligns well with a low-redshift extrapolation of the clumpy fraction that is measured using high-redshift observations. At fixed redshift, $f_{\mathrm{clumpy}}$ is negatively correlated with the stellar mass and positively correlated with the specific star-formation rate (sSFR) of the host galaxies. When the clump definition is changed to include only clumps with a stellar mass of $M_{\mathrm{cl}} \geq 10^7 M_\odot$, we observe a highly increased clumpy fraction of $\sim$60% that tends to increase with the stellar mass of the host galaxies but does not show a dependence on the sSFR of the host galaxies.
Read the full abstract

All authors Jürgen J. Popp, Hugh Dickinson, Stephen Serjeant, Lucy F. Fortson, Tobias Géron, Vihang Mehta

Computation and Language

Subword Segmental BabyLMs: Learning to Tokenise for Sample-Efficient Pretraining

In the standard LM training pipeline, subword tokenisation is applied as a preprocessing step. Subword segmental language modelling is an alternative paradigm in which tokenisation is learned during training, allowing the model to discover subword units that optimise its training objective. In this paper, we present our submission to the 2026 BabyLM Challenge, for which we develop two new subword segmental LMs: SubSegGPT and SubSegDeBERTa. SubSegGPT is a decoder-only model that learns tokenisation during autoregressive pretraining. SubSegDeBERTa is an encoder-based model that jointly learns to generate and tokenise masked words. We train both for the Strict and Strict-small tracks. Our top submission to Strict is SubSegDeBERTa, which achieves notable gains in zero-shot evaluation. Our top submission to Strict-small is SubSegGPT, which outperforms tokenisation-based baselines. Our results show that learnable subword tokenisation can improve sample-efficiency for BabyLM pretraining. We analyse the subword learning dynamics of our models and find that tokenisation gradually converges on subword units that balance morphological alignment and fine-grained segmentation.
Read the full abstract

Rings and Algebras

Peirce Stability of Null Polynomials and a Sharp Radical Bound for Finite Rings

Let $R$ be a finite associative ring with identity, and let $\Nul(R)$ denote the set of polynomials in a central indeterminate that vanish identically on $R$ under right evaluation. We establish two results concerning the failure of $\Nul(R)$ to be a right ideal. First, if $e$ is an idempotent, $f=1-e$, and either $eRf=0$ or $fRe=0$, then right multiplication by $e$ preserves $\Nul(R)$. Thus both off-diagonal Peirce components must be nonzero whenever $e$ witnesses a failure of right stability. Second, if $J$ is a finite nilpotent ideal whose additive group is a $2$-group and $e,f$ are complementary idempotents such that \[ eJf\neq0,\qquad fJe\neq0,\qquad J^3\neq0, \] then $|J|\geq32$. Werner's theorem that $\Jac(R)^3=0$ implies the two-sidedness of $\Nul(R)$ therefore yields $|R|\geq128$ whenever $\Nul(R)$ is not two-sided. We also construct a tiled matrix ring of characteristic $4$ and order $128$ whose null ideal is not two-sided.
Read the full abstract

Mesoscale and Nanoscale Physics

Probing Nonlinear Interactions of Dipolar Interlayer Excitons in MoSe$_2$/WSe$_2$ Heterobilayers

Interlayer excitons in transition-metal dichalcogenide heterobilayers possess intrinsic out-of-plane dipole moments, providing a platform for investigating exciton-exciton interactions at high densities. Here, we use excitation-energy-dependent photoluminescence excitation (PLE) spectroscopy to probe the nonlinear response of dipolar interlayer excitons in chemical vapor deposition-grown MoSe$_2$/WSe$_2$ heterobilayers. By tuning the excitation energy across intralayer exciton resonances at fixed excitation power, we selectively vary the population injected into the interlayer exciton states. Resonant excitation drives the system into a nonlinear regime, leading to saturation of the interlayer exciton photoluminescence and an apparent broadening of the intralayer $1s$ resonances in the PLE spectra. At the same time, the interlayer exciton emission exhibits a pronounced blueshift, reaching approximately 2.5 meV at 4 K and 1 meV at 75 K. The blueshift increases systematically with the interlayer exciton population and is consistent with a net repulsive exciton-exciton interaction, with contributions from dipole-dipole repulsion in the density regime investigated. Our results establish PLE as a sensitive approach for accessing the nonlinear, high-density regime of interlayer excitons and probing their interactions in van der Waals heterostructures.
Read the full abstract

All authors Sai Shradha, Luc F. Oswald, Md Tarik Hossain, Lukas Krelle, Nicole Engel, Axel Printschler, Julian Führer, Honey Jayeshkumar Shah, Daria I. Markina, Kenji Watanabe, Takashi Taniguchi, Andrey Turchanin, Bernhard Urbaszek

Computer Vision

Dotting the Eye: An Intent-Driven Image Retouching Agent for Visual Focus Enhancement

Image retouching is commonly formulated as enhancing overall visual quality through color adjustment, but in practice, it also serves to emphasize visual focus by guiding viewers' attention toward a specific subject or region. Achieving such focus-oriented retouching is inherently challenging, as it requires well-coordinated global and local adjustments to manipulate perceptual saliency while maintaining visual naturalness. This intricate process typically demands substantial professional expertise. In this study, we propose EyeControl, a MLLM-driven agent with a diffusion-based retouching executor that enables visual focus enhancement under weak user intent. With only a few clicks or coarse strokes, EyeControl directs visual attention to the intended region, effectively "dotting the eye" of the image. The core idea is to explicitly link the weak user intention with the target editing region and the corresponding tonal adjustment operations during retouching. To achieve this, the system first interprets the intent and image content to infer the visual focus and generate structured intent guidance for the retouching executor. Second, the retouching executor is encouraged to respond more strongly to the target region, explicitly aligning its attention map with a designed pseudo-intent map. We also introduce an operation-consistency constraint to improve coordination between global and local adjustments, achieving more natural and coherent retouching. Additionally, we contribute ControlArt-Bench, a high-quality evaluation dataset for visual focus enhancement. Extensive evaluations demonstrate that EyeControl yields perceptually appealing results with stronger intent alignment. Code can be found in https://github.com/DragonisCV/EyeControl.
Read the full abstract

All authors Chujie Qin, Zilong Zhang, Zewei Chang, Chunle Guo, Ruixing Wang, Tao Hu, Ming-Ming Cheng, Chongyi Li

Computer Vision

On the Design Fundamentals of Pixel Text Representation Learning

Text-rich visual inputs require models that can read, retrieve, and compress language directly in pixel space, yet existing pixel-text encoders struggle with fixed resolution pretraining, visual shortcut learning, weak visual grounding, and multilingual visual text understanding. In this work, we investigate the fundamental design principles required for robust visual text representation learning. Through systematic controlled ablations, we identify four critical components: variable image resolutions and rendered font sizes provide spatial proxies for high-resolution document generalization; natural image-text pairs are indispensable for grounding and prevent text-only collapse; layout-aware rendering helps prevent pixel-level shortcuts; and a two-stage multilingual curriculum enables effective cross-lingual alignment. By integrating these principles into a scalable training recipe, we train Pixel Linguist II, a native-resolution vision encoder trained with on-the-fly rendering, unified contrastive grounding, and a multilingual curriculum over 280M training examples. Pixel Linguist II sets new state-of-the-art results on English, cross-lingual, and multilingual Visual STS and ViDoRe, while also enabling better MLLM downstream evaluation. Notably, Pixel Linguist II remains robust under 80% visual token compression, showing great promise for optical context compression. Our code and resources are available at https://github.com/Pixel-Linguist/Pixel-Linguist-II.
Read the full abstract

All authors Chaohao Yuan, Ruifeng Yuan, Zhuoxu Huang, Yu Rong, Hong Cheng, Hou Pong Chan, Chenghao Xiao

Computer Vision

StainPresetNet: Stain Preset Network for Fast Multi-to-Multi Stain Normalization

Stain normalization reduces color variations caused by variations in staining protocols and imaging conditions, thereby enhancing computer-aided diagnostic system performance. Traditional methods derive mapping relationships from individual or limited reference images through pixel-wise transformation, offering style flexibility but suffering from inaccurate color mapping extraction. While existing deep-learning-based approaches achieve accurate dataset-wide color mapping through complex neural networks, they face challenges including computational inefficiency, artifact generation, and fixed normalization directions requiring model retraining for directional changes. To address these limitations, we propose StainPresetNet - a novel framework that combines structural preservation with dataset-level color mapping while maintaining computational efficiency. Our method implements pixel-wise normalization guided by preset reference images, enabling multi-directional adaptability without retraining. Evaluations on cytopathology and histopathology datasets demonstrate that StainPresetNet achieves superior color mapping accuracy compared to conventional methods, effectively improves classifier generalization in diagnostic tasks, and reduces computational overhead by 90% versus existing deep learning approaches. The proposed preset-guided mechanism facilitates flexible adjustment of normalization directions through simple reference image replacement, overcoming the directional rigidity of current deep-learning-based solutions.
Read the full abstract

All authors Hongtao Kang, Die Luo, Li Chen, Jing Cai, Junbo Hu, Xiuli Liu, Shenghua Cheng

Optics

Direct laser-written optomechanical double membranes in an optical microcavity

Optomechanical membrane-in-the-middle systems that comprise several coupled compliant membranes offer enhanced optomechanical coupling and collective dynamics. Recent realizations were fundamentally restricted to two membranes and to limited structural optimization of the involved membranes due to their intricate fabrication and experimental integration. Here, we present a fiber-based Fabry-Perot microcavity incorporating monolithically integrated double-membrane resonators fabricated by 3D direct laser writing. We realize controllable mechanical mode hybridization of membranes with coupling rates of up to $J_{\mathrm{mech}}/2\pi = 0.16$ MHz, exceeding the mechanical linewidths. We demonstrate enhanced collective optomechanical coupling of the membrane stack's breathing mode, reaching collective coupling strengths of up to $g_{\mathrm{col}}^{(-)}/2\pi \approx 0.1$ MHz. Our transfer-matrix calculations predict further substantial enhancements with realistic reductions of membrane thickness and spacing, and a larger number of membranes. Our results establish direct laser-written membrane arrays as a scalable platform for multimode cavity optomechanics, combining tunable mechanical interactions, enhanced collective optomechanical coupling, and scalability towards larger mechanically coupled resonator systems.
Read the full abstract

All authors Lukas Tenbrake, Daniel Stachanow, Florian Giefer, Jana Blechmann, Wolfgang Alt, Sebastian Hofferberth, Hannes Pfeifer

Numerical Analysis

On the Gram matrix of standard inner products of asymmetrically-weighted Hermite functions

Let A denote the infinite Gram matrix associated with the standard L2 inner product of asymmetrically-weighted (AW) Hermite functions. We derive an explicit representation of its entries and its Cholesky factorization. We further show that this factorization admits a natural interpretation on a scaled Bargmann-Fock basis. An explicit formula for the inverse of A is also obtained. We then consider the corresponding finite Gram matrix and analyze its asymptotic property, as well as that of its Schur complement. The analysis is motivated by numerical methods for plasma physics, in particular Galerkin spectral methods applied to the Vlasov-Poisson (VP) system. As an application, we demonstrate how the derived Gram matrix formulas and asymptotic results can be exploited in the analysis and implementation of a Galerkin spectral method for the VP system.
Read the full abstract

Probability

The voter model on the hyperbolic graph

We consider the voter model on the giant component of a hyperbolic random graph, which is a spatial scale-free network, in the sparse and linear-giant regime $\alpha\in(1/2,1)$. We find that the quenched expected consensus time has order $n^{2-1/\alpha}$, as the number of vertices $n\to\infty$, with probability arbitrarily close to one. This is generalised to the voter model where each vertex changes its opinion at rates $q(v)={\rm d}(v)^\varphi$, where we also establish the consensus time orders for all $\varphi\geq 0$. These orders have 3 regimes, with a phase transition at $\varphi=2-2\alpha$. For the upper bounds, our main proof idea is to connect the meeting set to some fixed target vertex of appropriate height in the product chain electrical network, to make rigorous an argument due to Durrett.
Read the full abstract

Differential Geometry

Lagrangian Foliations on compact Kähler manifolds

We prove that a compact Kähler manifold carrying a regular Riemannian foliation whose leaves are Lagrangian and minimal is flat. In order to prove this result, we establish a foliated version of an integral formula due to Ros.

Computer Vision

Revisiting Face Recognition for Monozygotic Twins: The Celeb Twins Test Set

Past literature on face recognition for monozygotic (("identical") twins points to facial marks and mirror asymmetry as possible directions for improved accuracy of twins recognition. The Celeb Twins Test Set (CTTS) contains web-scraped image pairs for 80 sets of celebrity twins. It is the only twins test set with meta-data for twins with distinguishing skin marks and possible mirror asymmetry. CTTS is organized in the manner of face verification test sets such as LFW, CALFW, CPLFW, CFP-FP, and AgeDB-30. Current deep CNN matchers can achieve over 76% accuracy in classifying CTTS same-person / different-person image pairs. We show that current matchers do not make use of skin marks, or asymmetry, and discuss reasons for this. Finally, we discuss the feasibility of using generative AI tools such as Grok, ChatGPT and Gemini to create images of imagined monozygotic twins as a means to increase representation of twins in face recognition training sets.
Read the full abstract

All authors Michael Zang, Haiyu Wu, Mrinal Sharma, Kevin W. Bowyer

Dynamical Systems

Permutative lifts of strictly ergodic subshifts

Permutative sliding block codes - in the sense of Hedlund - give rise to finite-to- one extensions of subshifts. We provide criteria under which minimality and unique ergodicity are preserved by this 'lifting procedure'. These findings are illustrated by means of some nat- ural example families, which we use to obtain strictly ergodic finite-to-one extensions of sub- stitution subshifts (including the case of Fibonacci, Tribonacci, silver mean and noble means substitutions). We further study the interplay between permutative sliding block codes and Toeplitz flows. In particular, we find examples which demonstrate that a permutative lift of a strictly ergodic subshift may be minimal, but not uniquely ergodic - a phenomenon which cannot occur for primitive substitution subshifts.
Read the full abstract

Computation and Language

Does task decomposition improve automatic NLG evaluation?

The LLM-as-a-judge (LLMaJ) framework has emerged as a promising solution for cheap, reproducible, reference-free Natural Language Generation (NLG) evaluation. Prior work seeks to improve LLMaJ by decomposing evaluation tasks into simpler sub-tasks. In this work, we systematically compare LLMaJ methods with and without decomposition on multiple NLG datasets. We find no evidence that LLMaJ with task decomposition leads to performance gains over a fair baseline that does not use decomposition. Instead, we find that previously reported performance gains in decomposition-based LLMaJ stem from using human labels as training data, and not task decomposition itself. Also, we find that, when human labels are available, LLMaJ without using task decomposition can perform comparably to human annotators.
Read the full abstract

All authors Sebastian Steindl, Nikos Voskarides, Alberto Gasparin, Diego Marcheggiani

Optics

High-Speed and High-Responsivity Asymmetric Waveguide Photodiode with Low Optical Back-Reflection

We numerically demonstrate an asymmetric corner-reflector uni-traveling-carrier waveguide photodiode (UTC-WGPD) that simultaneously suppresses optical back-reflection and breaks the bandwidth-responsivity trade-off. Off-center beam coupling introduces geometric asymmetry to eliminate retroreflection while maintaining efficient light trapping. 3D optoelectronic simulations reveal that at 1550 nm, a 15-{\mu}m^2 device exhibits a 275-GHz 3-dB bandwidth, 0.68-A/W responsivity (54.5% quantum efficiency), and -29-dB back-reflection, achieving a 150-GHz bandwidth-efficiency product. This design offers a scalable, low-reflection detection scheme for sub-terahertz transceivers and dense photonic integration.
Read the full abstract

All authors Zhijun Zhang, Xuejie Gao, Qiunan Li, Xiaoyu Mi

Quantum Gases

Quantum antidipolar systems in two-dimensional geometries

Particles with magnetic moment can be polarized, and rapidly rotated, employing a magnetic field such that the dipolar interaction among them changes sign, becoming antidipolar and thus isotropically attractive in a plane. Polar molecules can also be manipulated using microwave dressing fields to invert the sign of the dipole-dipole interaction. In this work, we study a two-dimensional system of antidipolar particles by calculating its equation of state and structural properties. The system behaves as a liquid even for scattering lengths significantly greater than the dipolar length. For large enough densities, the system transitions to a solid with one particle per lattice site via a first-order phase transition at a significantly smaller density than its dipolar counterpart. Moreover, motivated by the recent realization of a strongly axially trapped, bilayer geometry [Science 384, 546-551 (2024)], we study the properties of the bilayer liquid phase as the inter-layer distance is tuned, and provide the range of parameters where one layer can influence the properties of the other.
Read the full abstract

Computer Vision

Different Changes Require Different Reasoning: Change-Type-Specialized Experts for Robust Change Captioning

Change captioning is the task of generating natural language descriptions that explain the changes between a pair of images. Although different change types (e.g., color shifts, object additions) exhibit distinct visual cues and require specialized reasoning processes, existing methods often overlook these distinctions. To address this limitation, we propose Multi-Expert Diagnosis for Image Change (MEDIC), a novel framework that introduces change-type awareness by explicitly modeling change categories. MEDIC employs type-specialized memory experts that dynamically retrieve type-relevant visual patterns conditioned on the input. This design enables each expert to capture diverse variations within its change type while focusing on the most informative visual cues. By softly routing inputs across type-specialized experts and learning dedicated representations for each change category, MEDIC generates more precise and type-aware change descriptions. Extensive experiments demonstrate that the proposed MEDIC consistently outperforms existing methods across diverse and challenging datasets. The code is available at \href{https://github.com/VisualAIKHU/MEDIC}{GitHub}.
Read the full abstract

Computation and Language

Overfitting Mitigation via Singular Value Decomposition in Minimum Bayes Risk Decoding

Minimum Bayes Risk (MBR) decoding enables high-quality text generation by selecting the hypothesis that maximizes a utility metric over sampled pseudo-references. However, it is highly susceptible to metric overfitting: it can irregularly inflate the chosen utility metric at the direct expense of other unoptimized evaluation metrics. To mitigate this, we introduce SVD-MBR, which frames the pairwise utility matrix as a noisy information signal. By computing a low-rank approximation via Singular Value Decomposition (SVD) and retaining only the top-$k$ components, we effectively decouple true consensus from metric noise. Experiments demonstrate that SVD-MBR successfully regularizes decoding, yielding substantial gains across a range of generalized metrics. Furthermore, we reveal that this denoising is metric-dependent: neural metrics encode a robust low-rank consensus ideal for SVD, whereas surface-level metrics struggle to separate signal from metric noise.
Read the full abstract

All authors Riza Setiawan Soetedjo, Yusuke Sakai, Hidetaka Kamigaito, Katsuhiko Hayashi, Taro Watanabe

General Relativity and Quantum Cosmology

Relative-Phase Control of Common-Bridge Survival in Collisions of Radially Excited Boson Stars

We show that the relative phase primarily controls the survival, rather than the initial formation, of the common bridge in head-on collisions of radially excited boson stars. In the benchmark configuration, all sampled phases form both a common and a strong bridge. After aligning the evolutions at the same operational weak-contact level, their early post-contact evolution remains similar, while a clear phase hierarchy develops only later, with larger phase offsets producing a weaker and less persistent connection. The same suppression persists under controlled variations of the central scalar amplitude, quartic self-interaction strength, and scalar potential. These results identify relative phase as a control on the nonlinear persistence of the common bridge rather than simply on the instantaneous scalar-field overlap.
Read the full abstract

Statistical Methodology

Scalable Inversion of Contests with Correlated Performances, Including Softmax and Multinomial Probit

Multinomial probit choice probabilities over n alternatives are Gaussian orthant integrals, computed by simulation for thirty years, one expensive integral per alternative. Inversion, which is to say determining item attractiveness consistent with a prescribed choice probability vector, is even more difficult and has been considered impractical for correlated contests when n is large. Yet here, for families lying within a grammar including factor, block and hierarchical covariance structures, we exhibit a calibration tested at n = 1,000,000 reproducing probabilities to very high accuracy, even in the extreme tail. We must return to much smaller problems for any performance comparison to be possible due to limitations of the prior art. The Geweke-Hajivassiliou-Keane simulator is the standard (and still appropriate for high rank) but is two hundred times slower already at n = 200, and its measured cost grows roughly as n^2.8 while ours is linear. Furthermore our approach applies to any continuous performance distributions within reason: the Thurstone-Mosteller model families thereby become a practical alternative to logit at modern scale.
Read the full abstract

Biomolecules

From static structures to dynamic landscapes: cryo-EM redefines RNA biology

RNA molecules perform diverse biological functions by dynamically exploring multiple conformational states rather than adopting a single static structure. Capturing these ensembles is a challenge in molecular biology. Recent advances in cryoEM are now transforming this landscape by enabling the visualization of RNA molecules across a broad spectrum of functionally relevant conformations at near atomic resolution. Here, we examine how cryoEM is reshaping RNA structural biology changing focus from the analysis of static structures to dynamic conformational landscapes. Through 8 representative case studies we illustrate how cryoEM has revealed previously inaccessible mechanisms of RNA motion, including folding processes, ligand-dependent switching, and cooperative assembly. We specifically discuss emerging experimental and computational approaches that address and overcome the challenges associated with studying dynamic RNAs, particularly in construct design, sample preparation, vitrification, and data analysis. These novel methods resolve conformational variability and enable the reconstruction of discrete and continuous RNA conformational landscapes from cryoEM data, highlighting how structural heterogeneity can be harnessed to extract functional insights. Looking forward, the integration of cryoEM with complementary biophysical techniques and time resolved methodologies promises to bridge structural and temporal resolution, to routinely derive experimental molecular movies of RNA in action. These advances will not only deepen our understanding of RNA biology but also provide new opportunities for RNA targeted therapeutics and the rational design of dynamic RNA-based nanodevices. By connecting structural snapshots into coherent dynamic models, cryoEM is establishing a framework for quantitative descriptions of RNA energy landscapes and their functional roles.
Read the full abstract

All authors Shekhar Jadhav, Spandan Saha, Qingbin Shang, Marco Marcia