Peer Revenue Returns subject to peer review

Papers

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

All of arXiv

Monday 9 November

Networking

Network-Co-Designed Dynamic Sparsification for Dependable Distributed Robot Learning over 3GPP Networks

Robot fleets performing distributed learning often face volatile uplink conditions like bandwidth variation, packet loss, and latency spikes that delay or destabilize training. We propose a network-co-designed dynamic sparsification loop that couples model-update compression to network-facing byte budgets and time constraints. We describe two realizations: (i) a standards-aligned SA6/5GC mapping that leverages CAPIF with NEF/PCF/NWDAF, enriched with AI-specific semantics; and (ii) a VASEAL-style role-reversed loop where the learning system exposes learning-state feedback and receives bounded, time-limited control intents. We implement the mechanism in an asynchronous federated reinforcement learning framework with heterogeneous simulated quadruped robots and reproducible uplink impairments via emulated network links. Results indicate that substantial communication reduction is achievable under dynamic constraints without materially degrading learning progress in the studied setting.
Read the full abstract

Computational Complexity

Constant query LCCs over the reals

This note describes a new construction, found by ChatGPT Astra, of constant query Locally Correctable Codes (LCCs) over the real numbers. Prior constructions of constant query LCCs were exclusive to fields of positive characteristic and were conjectured not to exist over the reals. The construction uses low degree polynomial evaluations over an integer grid, together with a novel sampling procedure which outputs, for a given grid point $P$, a random quadratic curve through $P$ which intersects the grid in many other points (the decoding queries). The sampler is `smooth' in the sense that no grid point is queried with high probability. These codes have the additional property that the queried locations (the curve/grid intersections) are all nearby grid points.
Read the full abstract

Computer Vision

RetinexGS: Retinex-Inspired Gaussian Splatting for Low-Light RAW Reconstruction

Standard 3D Gaussian Splatting (3DGS) pipelines struggle with severely underexposed images, producing unstable colors, floating artifacts, and incomplete reconstructions in dark regions. To better account for noise in these low-light observations, RAW images offer a natural basis for noise-aware reconstruction by preserving linear sensor measurements and providing access to noise metadata. We present RetinexGS, a RAW-domain 3DGS framework that combines Retinex-inspired appearance modeling with sensor-informed optimization for low-light novel-view synthesis. We factorize each Gaussian's color into a view-independent base color and a shared, position- and view-dependent modulation field, regularizing the two components separately to constrain appearance learning from noisy observations. We use DNG noise metadata to weight the reconstruction loss for signal-dependent noise, while distortion-aware rendering avoids spatially resampling RAW targets. To address sparse geometry in dark regions, we introduce depth-guided initialization. We evaluate RetinexGS on three public RawNeRF scenes and two custom-captured low-light RAW scenes. With all methods trained on the same RAW inputs for each scene, RetinexGS outperforms LE3D and RawNeRF in mean sRGB PSNR by 2.18 dB and 2.71 dB respectively, across the five scenes, while supporting real-time novel-view rendering.
Read the full abstract

Analysis of PDEs

Asymptotic Stability of Moving Kinks with an Internal Mode

We prove asymptotic stability of moving kinks for one-dimensional scalar field equations with one internal mode and no threshold resonance, under a transformed repulsivity condition introduced in Kowalczyk, Martel, Muñoz, and Van Den Bosch and a Fermi golden rule at the second harmonic. Near every fixed subluminal velocity, arbitrary sufficiently small perturbations in $H^1\times L^2$ converge locally in the energy space to a kink with a limiting velocity. The principal difficulty is the coupling between the damping of the internal mode and modulations: the internal oscillation enters the modulation equations quadratically, whereas the damping mechanism provides only quartic time integrability. Inspired by Martel, we derive a velocity normal form from conserved momentum and incorporate the corrected velocity into the virial functional, cancelling the quadratic modulation terms that obstruct the local energy estimates. Matrix Darboux transformations, virial identities at two spatial scales, and a localized Fermi golden rule then yield simultaneous spacetime control of the internal mode and radiation, together with convergence of the velocity. We apply our criterion to the double sine-Gordon equation and suitable perturbations of the $\phi^4$ model. Through a perturbative analysis of threshold resonances, we determine the full-space discrete spectrum and establish the absence of threshold resonances. We further verify the repulsivity of the twice-transformed potential and the Fermi golden rule.
Read the full abstract

Complex Variables

On the number of zeros of polyanalytic polynomials

We prove that a complex polynomial $f(z,\bar z)$ of degree $n$ in $z$ and degree $m<n$ in $\bar{z}$, with finitely many zeros, has at most $(4m+1)n-5m$ distinct zeros in $\mathbb{C}$. The proof uses an approach suggested by GPT-6.1 based on a topological result of Kronheimer and Mrowka (Thom's Conjecture). We provide examples showing that the bound is asymptotically sharp as $n\to\infty$ for each fixed $m$.
Read the full abstract

Systems and Control

Pulsatile Endocrine Regulation with a Releasable Hormone Pool: Impulsive Feedback Model

An impulsive model of endocrine regulation equipped with frequency- and amplitude-modulated pulsatile feedback is extended to study the dynamical effects pertaining to limitations due to the releasable hormone pool. The model describes a three-hormone regulatory loop and is applicable to endocrine systems such as the reproductive, stress, and growth hormone axes. Conditions guaranteeing the existence and uniqueness of a periodic solution with a single feedback-triggered impulse per period (a 1-cycle) are derived, together with monotonicity properties of its characteristics. An explicit Jacobian formula is obtained to assess local orbital stability. Model analysis demonstrates that a reduced hormone-pool replenishment rate produces a lower-amplitude, higher-frequency 1-cycle. This behavior is consistent with experimentally observed alterations in testosterone regulation in aging human males.
Read the full abstract

Machine Learning

An Analytical Model of Compute-limited Multistage Training Pipelines

Modern foundation models are trained through multiple stages, yet how these stages interact and how compute should be allocated across them remain poorly understood. We introduce an analytically tractable high-dimensional model combining unsupervised pretraining with a linear autoencoder, supervised fine-tuning (SFT) with noisy stepwise labels, and reinforcement learning (RL) with sparse episodic rewards. Using random matrix theory and an operator-projection closure, we derive approximate low-dimensional learning dynamics that closely match simulations. We show that finite pretraining data imposes a representation-dependent performance limit, with Bayes classification error scaling as the inverse square root of the pretraining sample ratio, and observe the same exponent in asymptotic SFT and RL errors. Better pretraining also reduces sensitivity to noisy supervision. SFT initialization shortens the initial RL plateau induced by sparse rewards, an effect we also observe in OLMo-2 1B on GSM8K. Longer SFT warm starts additionally improve asymptotic RL performance. Optimizing compute across all three stages reveals that the optimal allocation depends on supervision quality and total budget, with a non-monotonic dependence of the RL compute share on budget. Finally, a complementary compositional model shows how RL selectively strengthens task-relevant pretrained skills while largely preserving unused skills. Together, these results identify mechanisms linking representation quality, feedback structure, and compute allocation in multistage training.
Read the full abstract

All authors Nishil Patel, Jin Hwa Lee, Basile Confavreux, Andrew Saxe

Machine Learning

Off-policy Learning with Excursion Policies

Off-policy reinforcement learning (RL) tries to learn the value of a target policy while data comes from a different behavior policy. This often requires trading off bias for variance or stability. One way to trade these off is to linearly interpolate between the behavior and target policies, but this represents only one point in a broader space of interpolation mechanisms. We introduce excursion policies–a structured alternative where the agent follows the target policy for a multi-step "excursion" before reverting to the behavior policy. We derive dynamic programming results for these policies, establishing the theoretical basis for their use in RL. We further develop provably sound off-policy temporal-difference (TD) algorithms for excursion value estimation. We find that excursion policies provide an efficiency-stability trade-off comparable to mixture policies in continuous control benchmarks, and our analysis illustrates conceptual differences in their behavior.
Read the full abstract

All authors Jiamin He, Mark Rowland, Zhaohan Daniel Guo, Hado van Hasselt, Diana Borsa

Combinatorics

On the Chromatic Number of Cayley Tables of Dihedral Groups

A coloring of a Latin square $L$ is an assignment of $k$ colors to its cells such that no two cells sharing a row, a column, or a symbol receive the same color. The smallest $k$ for which such a coloring exists is called the chromatic number of $L$, denoted by $\chi(L)$. The Cayley table of a finite group $G$ of order $n$ is a Latin square, which we denote by $L_G$. It is known that $\chi(L_G)\ne n+1$, so either $\chi(L_G)=n$ or $\chi(L_G)\ge n+2$. Moreover, $\chi(L_G)=n$ if and only if every Sylow $2$-subgroup of $G$ is trivial or non-cyclic, which completely characterizes the finite groups whose Cayley tables attain the lower bound $n$. In this paper, we consider the dihedral group $D_n$ of order $2n$, where $n$ is odd. Since $D_n$ has a cyclic Sylow $2$-subgroup, it follows that $\chi(L_{D_n})\ge 2n+2$. We prove that this bound is sharp, that is, $\chi(L_{D_n})=2n+2$.
Read the full abstract

Combinatorics

Properties of the Dichromatic Polynomial

The chromatic polynomial is a well-studied function of graphs whose coefficients, factorization, and other properties are well-known to encode structural information about the underlying graph. An analogous notion of graph colouring exists for directed graphs which is also well-studied, but little is yet known about the corresponding dichromatic polynomial. We investigate properties of the dichromatic polynomial, including its coefficient interpretations, root-dense regions, and asymptotic growth. We also show that chordal digraphs' polynomials split into linear factors in a way analogous to the undirected case, confirm a conjecture on dichromatic uniqueness posed by González-Moreno et al., and study the density of dichromatic polynomials within certain larger families of polynomials.
Read the full abstract

All authors Lila Crew, Benjamin Keller, Katharina Kuehr, Shao Yuan Lin, Megan Ixchel Monroy Rodríguez, Tabitha Williams

Geophysics

Heterogeneous and episodic permeability evolution and correlation with acoustic emission activity during faulting in granite

Rock permeability changes dramatically during brittle deformation, affecting faulting dynamics and crustal fluid flow. Here, we combine fluid-flow and local pore-pressure measurements acquired during a quasi-static shear-failure test in Westerly granite with a three-dimensional finite-element inverse method. The measured flow rate and six internal pore pressures constrain the local hydraulic transmissivities (permeability-thickness products) of ten cells along the growing fault. Synthetic ensembles anchored to the pressure-sensor accuracy recover these transmissivities with median mean errors of 7-9%, with the fault ends generally better resolved than the interior. Applied to the experiment, the method resolves episodic transmissivity increases, interspersed with decreases, through fault growth and slip. Local values reach approximately $1.6 \times 10^{-19}$ m$^3$ (16 times the bulk contribution), and the final-stage values vary by a factor of 2.3 along the fault. Across cells, the median absolute change in recovered transmissivity is at most about 16% for $\pm50\%$ changes in assumed fault-zone thickness. The local one-dimensional estimates of Aben et al. (2024) agree in scale with the three-dimensional results, with a median 3-D/apparent ratio of 0.81. Cumulative acoustic-emission counts track the overall hydraulic evolution, but the relations of local event density, its stagewise increments and amplitude-weighted activity to transmissivity vary between cells. The comparison distinguishes the value of AE for mapping damage growth from quantitative hydraulic prediction using event-count and relative-amplitude metrics.
Read the full abstract

Robotics

RaMOD: Moving-Object Detection from Raw mmWave Radar Cubes under Ego Motion

Millimeter-wave (mmWave) radar provides weather-robust sensing and direct Doppler measurements, yet automotive perception pipelines often reduce its output to sparse point clouds or two-dimensional projections, leaving raw range-azimuth-Doppler (RAD) cubes underutilized. Ego motion induces Doppler responses from world-static structures, while Doppler wrapping aliases radial velocities, complicating moving-object detection and velocity estimation. We present RaMOD (Radar-based Moving-Object Detection), a real-time framework for moving-object detection and velocity estimation from raw RAD cubes acquired by a vehicle-mounted single-chip radar. RaMOD explicitly estimates frame-wise ego motion and uses the estimated radar translational velocity to predict the Doppler region of world-static structures and suppress returns within it. It temporally encodes the suppressed cubes into multi-scale features to predict RAD-space boxes with class labels. For each detected object, it predicts radial and forward velocities under raw-Doppler and Doppler-unwrapped hypotheses, together with Doppler-wrapping probabilities. On our BRIDGE-Liv dataset, RaMOD outperforms all compared raw-radar methods across RAD-cube, RA, and RD detection settings, achieving 80.8% cube mAP at IoU 0.30, 15.7 percentage points above the strongest RAD-cube baseline. Velocity MAE is 1.128 m/s, with 99.87% Doppler-wrapping accuracy.
Read the full abstract

All authors Yuan Fang, Long Zhuang, Lingzi Xing, Xijia Wei, Jingqi Jiang, Fangzhan Shi, Simon Julier, Kevin Chetty

Computation and Language

What drives the emergence of symbolic mechanisms in large language models?

Recent work has identified a set of emergent symbolic mechanisms that support abstract reasoning in large language models, but it remains unclear what factors drive the emergence of these mechanisms. To address this question, we trained neural networks from scratch on abstract sequence tasks, varying both architectural and data distributional factors, and investigated the effects of these factors on the emergence of symbolic mechanisms. Using representational, attentional, and causal mediation analyses, we first confirmed that transformer-based language models trained from scratch on our task developed symbolic mechanisms. We then found that, across multiple axes of data diversity, greater diversity produced stronger symbolic signatures. Surprisingly, the same signatures emerged in modified transformer architectures and even multilayer perceptrons, indicating that architectural inductive biases are not necessary for the development of symbolic mechanisms. We also observed consistent patterns in an entity-attribute binding task. Together, these findings identify data diversity, not architecture, as the primary driver of emergent symbolic computation in neural networks.
Read the full abstract

General Relativity and Quantum Cosmology

Viscosity heats, diffusion drains: Open-system thermodynamics in unimodular cosmology

In GR with bulk viscosity and in unimodular gravity with energy diffusion, the matter sector is non-adiabatic, and their balance, entropy, and temperature equations match if $\Pi \leftrightarrow \dot{Q}/3H$. Disagreement over whether unimodular cosmology is adiabatic comes from assigning entropy to matter only or to the total effective fluid. The key is the temperature $T_D$ of the sector absorbing the diffused energy: comoving entropy production is $\dot{Q}(1/T_D-1/T_m)$, yielding earlier limits for $T_D \rightarrow \infty$ and $T_D \rightarrow T_m$. For finite $T_D$, diffusion acts like heat exchange with an external reservoir: unlike viscosity, which only heats a closed fluid, diffusion into a colder bath can cool an open one, breaking the viscosity-diffusion correspondence for observables and stability. Cosmographically, $\Pi$ is fixed by the deceleration $q$, while diffusion is fixed by the jerk $j$; thus $\Lambda$CDM is the adiabatic limit of unimodular cosmology but remains dissipative when modeled via viscosity. The diffusive analogue of $|\Pi|/\rho$ is the shift $c_s^2-\omega$ in effective sound speed, set by perturbative stability rather than acceleration. If dark matter is a gas with conserved rest mass, consistent diffusion removes only thermal energy, forbids phantom behavior, and bounds the growth of the cosmological term by the finite thermal reservoir; producing a logotropic dark fluid with conserved particle number would require relativistic dark matter at $z\simeq12$. With the Hayward-Kodama apparent-horizon temperature, the generalized second law in unimodular gravity follows from the local one, whereas for bulk viscosity it must be imposed separately.
Read the full abstract

Robotics

Force-Closure Condition for Cooperative Multi-robot Pushing of Rectangular Objects

In cooperative object manipulation, the placement of pushing contacts determines whether force closure can be achieved and agents can generate arbitrary forces and torques on the object correspondingly. This study derives a necessary and sufficient closed-form condition for the planar force closure of a rectangular object, a common object shape in everyday settings, under four frictionless pushing contacts. For a rectangle of length $b$ and width $a$, force closure holds if and only if $(l_1+l_3-b)(l_2+l_4-a)<0$, where $l_1,l_2,l_3,l_4$ are the four edge contact locations. This condition is independent of the object's center of mass and is used to verify contact configurations before contact force allocation. When the condition is not satisfied, the signs of the two factors indicate which opposing contact pair should be adjusted. The condition and adjustment rule are then integrated into a full-pose trajectory-tracking framework as a geometric feasibility layer to demonstrate their use in control. Static wrench feasibility tests validate the proposed condition, while narrow-passage floating-body simulations show that maintaining a force-closure configuration reduces the RMS tracking errors in $x$, $y$, and unwrapped yaw from $12.05$ m, $18.11$ m, and $535.96^\circ$ to $2.24$ m, $2.66$ m, and $2.69^\circ$, respectively.
Read the full abstract

Algebraic Geometry

The Cayley-Bedregal reduction for Milnor and Hirzebruch-Milnor classes

The Cayley–Bedregal reduction replaces a complete intersection $\{f_1=\dots=f_r=0\}$ by the hypersurface $\sum_i\varphi_if_i=0$ in the projectivized dual bundle. For an isolated complete intersection singularity of corank $c$, we express the Milnor class of this hypersurface on its singular fibre $\mathbb P^{c-1}$ in terms of the Milnor numbers of generic systems obtained by taking linear combinations of the defining equations. Its degree is the original Milnor number, and the higher components recover the Milnor numbers of the systems with fewer equations. These numbers are invariants of the analytic germ. Our second main result concerns the Hodge-theoretic refinement. For a complete intersection defined by a regular section of a rank-$r$ algebraic vector bundle, the Hirzebruch–Milnor class is determined by that of the Cayley–Bedregal hypersurface: the push-forward of the normalized Hirzebruch–Milnor class of the hypersurface, in the sense of Cappell, Maxim, Schürmann and Shaneson, is $(-y)^{r-1}$ times that of the complete intersection, with no positivity assumption on the bundle. The factor $(-y)^{r-1}$ comes from the motivic decomposition of the projective bundle, and at $y=-1$ the identity recovers Aluffi's comparison of Milnor classes. When the tautological bundle is very ample, the hypersurface formula of Maxim, Saito and Schürmann then gives the Hirzebruch–Milnor class of the complete intersection through the vanishing cycles of a single function, avoiding their induction through intermediate complete intersections and their restriction to powers of one line bundle. In integral homology we prove the analytic form of Aluffi's identity, and obtain formulas for the Milnor class in terms of local Milnor fibres and for the Chern–Schwartz–MacPherson class of any projective scheme.
Read the full abstract

All authors R. Callejas-Bedregal, F. Caraballo, M. F. Z. Morgado, J. Seade

General Relativity and Quantum Cosmology

Challenging the hoop conjecture with spinning pancakes and baguettes

Understanding when gravitational collapse produces a black hole remains a central challenge in general relativity. Black-hole uniqueness results characterize the stationary endpoint under suitable assumptions, while singularity theorems establish geodesic incompleteness in the presence of trapped surfaces and appropriate energy and global conditions. Neither, however, provides a general criterion for horizon formation. The hoop conjecture offers such a criterion by relating the enclosed mass to the compactness of a matter distribution in every direction. We construct axisymmetric families of initial data describing rotating dust and test a formulation of the conjecture adapted to this setting. We explore highly prolate and oblate configurations with both constant and differential rescaled angular-velocity profiles, including configurations near the threshold for apparent-horizon formation. We find that the conjecture holds for rotating configurations, with a coefficient within 14% of unity as in the nonrotating case, provided the quasilocal mass includes the rotational energy. If the irreducible mass of the apparent horizon is used instead, the deviations grow with spin, as they do for Kerr black holes. For horizonless configurations, the conjecture likewise holds when the mass is restricted to the energy enclosed by the matter surface, whereas using the ADM mass, which also includes rotational field energy outside the matter, leads to apparent violations of up to $\approx$25%. We also discuss the possible evolution of compact configurations without apparent horizons, their relation to weak cosmic censorship, and their implications for horizonless compact objects.
Read the full abstract

Dynamical Systems

Effect of car-centric social norms on traffic congestion in a compartmental model of transit mode utilization

Vehicle traffic congestion is a persistent feature of urban transportation systems around the world. Importantly, congestion is in part a consequence of human decision-making: car traffic is reduced when people choose modes of transportation such as metro trains, buses, or trams, over cars, and it is amplified when there is a persistent social bias towards driving. The underlying feedback loop in which traffic levels influence transportation mode choices, which in turn affect traffic levels themselves, is poorly understood and we study its consequences in the presented work using a compartmental modeling approach. We first consider a simplified model of only car traffic and show that it predicts a hysteresis between congested and uncongested states, which is consistent with broader literature on on traffic modeling. Next, we add buses and trains to the model as alternative transit modes, confirm the existence of hysteresis in this generalization, and derive the relative trip completion rates of each mode. Finally, to account for decision-making, we take the demand rates for the different transit modes to be dynamic, closing the feedback loop between mode choice and utilization. This redistributes demand for each mode of transportation, highlighting that traffic conditions may be a key aspect of mode choice. Furthermore, our model predicts that increasing public awareness of congestion can be an effective control strategy to increase utilization of public transit. However, car-centric social norms must be sufficiently small, or car use is instead amplified.
Read the full abstract

Materials Science

Hybridization-brightened interlayer excitons in multilayer WSe2 controlled by sliding ferroelectricity

Interlayer excitons (IXs) combine spatially separated electron and hole wavefunctions with electrically tunable dipole moments, but their reduced wavefunction overlap generally limits optical emission, particularly for high-energy states. Here we report a bright, spectrally narrow high-energy interlayer exciton (h-IX) near 2.23 eV in few-layer WSe2 and investigate how its optical activity evolves with layer number and sliding-ferroelectric polarization. In pristine 2H-WSe2, we observe thickness-dependent brightening of high-energy exciton emission. GW-BSE calculations reveal mixed intralayer and interlayer configurations with enhanced optical weight, suggesting that multilayer hybridization contributes to the observed brightening. In parallel-stacked 3L/3L WSe2 containing a sliding-ferroelectric interface, the h-IX exhibits polarization-dependent energies and Stark responses that distinguish Pup and Pdown domains. Electric-field-driven domain-wall motion produces hysteretic transfer of spectral weight between the two h-IX branches, enabling optical tracking of ferroelectric switching. These results establish multilayer hybridization as a route to optically active high-energy interlayer excitons and sliding ferroelectricity as a complementary degree of freedom for controlling their energy, dipole and emission.
Read the full abstract

All authors Youngki Yeo, Idan Kizel, Dror Hershkovitz, Omri Meron, Iva Plutnarova, Zdenek Sofer, Watanabe Kenji, Takashi Taniguchi, Elad Katash, Dan Oron, Haim Suchowski, Jing-Yang You, Moshe Ben Shalom

Analysis of PDEs

Global regularity of axisymmetric Navier–Stokes equations under total slip and critical smallness

We study the axisymmetric incompressible Navier–Stokes equations in a finite cylinder subject to the Navier–total–slip boundary condition, that is, the Navier slip condition with zero friction. We prove that the solution remains smooth for all time provided that the initial swirl $\Gamma_0 = r v_{\theta,0}$ has zero mean and an explicit smallness condition on the initial data holds. The condition involves $\Gamma = rv_\theta$ and the variables $V = v_\theta/\sqrt{r}$ and $\Omega = \omega_\theta/r$, and it is invariant under the natural scaling of the Navier–Stokes equations, hence critical. The total-slip condition produces a boundary term on the lateral wall that cannot be controlled by the dissipation alone; we overcome this with a sharp trace-type identity combined with a remainder controlled by $\|\Gamma\|_{L^2}$. In addition, the Robin-type boundary condition $\partial_r \Gamma = 2\Gamma$ destroys the maximum principle for $\Gamma$, and we replace it with a uniform $L^4$ bound. These estimates yield uniform $L^4$ bounds for the velocity, and regularity follows from the Ladyzhenskaya–Prodi–Serrin criterion together with boundary regularity results for the Navier slip condition.
Read the full abstract

Computer Vision

Adaptive Residual Quantization for Memory-Efficient Temporal Action Segmentation

Temporal action segmentation (TAS) is commonly trained on large banks of pre-extracted frame features, but storing and replaying those features becomes a practical bottleneck for long-form video, especially in incremental TAS where past-task data must remain available to mitigate catastrophic forgetting. Recent TAS replay and condensation methods reduce this burden, but they rely on dense frame-wise annotations and allocate storage through fixed segment-level structures, limiting their flexibility for long procedural videos whose frame complexity varies substantially over time. We introduce Adaptive Residual-Quantized Variational AutoEncoder (ARQ-VAE), a compact and label-free memory representation for TAS features that learns a discrete residual quantization of frame-level video features and then adaptively truncates the residual chain on a per-frame basis, retaining only the residual prefix that best reconstructs each frame feature. This yields a practical compact feature memory that does not depend on frame-wise labels and can therefore support multiple TAS settings with the same stored representation, while a lightweight fine-tuning stage further improves reconstruction quality for the truncated codes that are actually retained. Across standard TAS benchmarks, ARQ-VAE achieves a substantially improved storage-performance trade-off over prior replay and condensation baselines. On Breakfast, our final representation reduces training-set storage from 28 GB for the original I3D feature bank to 10 MB while retaining strong downstream segmentation performance; with ASFormer, it achieves 71.5 Edit and 51.2 F1@50 in the supervised setting. We further show that the same label-free memory remains effective for incremental and unsupervised TAS. We position ARQ-VAE as a practical memory representation for TAS and a promising direction for broader dataset condensation applications.
Read the full abstract

All authors Gerard Donahue, Guven Gergerli, Ayush Gupta, Reza Ghoddoosian, Enna Sachdeva, Faizan Siddiqui

High Energy Physics (Theory)

Expiscating the black hole singularity in holography

We argue that the leading exponential rate in the high frequency asymptotics of holographic thermal two-point functions is independent of the conformal dimension and spin (held fixed) of the operator. We analyze the leading universal WKB spectral curve governing the asymptotics at non-vanishing momentum scaling with frequency. We show that its classical action governs time domain signatures of the bouncing geodesic singularity, highly damped quasinormal modes, and the asymptotics of multi-stress tensor OPE coefficients. Beyond the leading order, we examine linear and dimension-dependent sublinear scalings of momentum with frequency. In both regimes, we derive the connection matrix between the boundary and the horizon using exact-WKB techniques. The results show a factorization of the spectral function into a perturbative part controlled by the OPE and a non-perturbative part governed by WKB transport. For linear scaling at fixed $k/\omega$ in the timelike regime, the turning points controlling physical transport remain separated from the curvature singularity. Nevertheless, a singularity-anchored representative of the same transport homology class determines the leading WKB action. In the sublinear regime, the relevant turning points approach the singularity. We match an exact solution of the near-singularity scaled equation to the exterior WKB solution to determine the connection matrix and spectral function. Our results clarify how the black hole curvature singularity is encoded in dual field theory observables at leading order in the planar expansion.
Read the full abstract

Instrumentation and Methods for Astrophysics

Beyond the Fast Rotators: A JWST Multi-Cycle Treasury Survey of Substellar Weather

Atmospheric variability provides a unique window into the three-dimensional structure, dynamics, and evolution of brown dwarf and giant exoplanet atmospheres. Recent JWST time-resolved spectroscopic observations have revealed complex wavelength-dependent variability, offering unprecedented insights into cloud structures, thermal perturbations, and atmospheric chemistry. However, existing programs have focused primarily on individual benchmarks and specific scientific questions, resulting in heterogeneous spectral and temporal coverage and leaving important regions of the substellar population underexplored, particularly across a broad range of rotation periods, spectral types, and viewing geometries. A JWST Multi-Cycle Treasury program could systematically investigate substellar weather across a diverse, well-characterized population. By combining broad infrared spectroscopic coverage with observations spanning multiple rotations and strategically selected epochs, this program would constrain the contributions of different physical mechanisms while investigating how atmospheric structures depend on rotation, viewing geometry, and fundamental physical properties. Extended monitoring would also improve rotational-period determinations and reveal how atmospheric features evolve from rotation to rotation and over longer timescales. This coordinated effort would provide a consistently designed and uniformly processed dataset of time-resolved spectra and variability diagnostics, enabling population-level comparisons, testing predictions from three-dimensional circulation models, and establishing a framework for interpreting the atmospheric diversity and evolution of brown dwarfs and directly imaged giant exoplanets.
Read the full abstract

All authors Natalia Oliveros-Gomez, Elena Manjavacas, Niall Whiteford, Caprice Phillips, Jared Bull, Kielan Hoch, Allison M. McCarthy, Beatriz Campos Estrada, Cian O'Toole, Evert Nasedkin, Stanimir Metchev, Gabriel-Dominique Marleau, Harshil Kothari, Antonia von Stauffenberg, Michael K. Plummer, Brianna I. Lacy, Fuda Nguyen, Xueqing Chen, Fei Wang, Genaro Súarez, Ben J. Sutlieff

Theoretical Economics

Choosing Complexity: Cognitive Ability, Overconfidence, and Task Choice

Workers and organizations routinely choose between complex tasks that differ in their success-payment tradeoff. We study how cognitive ability and overconfidence shape this choice of task complexity, both theoretically and experimentally. Consistent with our model, we find that more cognitively able individuals choose more complex tasks. Conditional on cognitive ability, we find that more overconfident individuals also choose more complex tasks. In about 75% of the decisions, subjects pick their payoff-maximizing task, and higher cognitive ability is weakly associated with more efficient choices of task complexity. But errors are strikingly one-sided: those who fail to choose optimally pick tasks that are too complex. Our findings suggest that organizations should pair task menus and incentives for choosing more complex tasks with information and feedback that help workers assess their ability and probability of success. This preserves the benefits of self-selection while reducing costly overreach.
Read the full abstract

Sound

Which Acoustic Features Survive Compression? Tracker-Estimated First-Formant Statistics for Codec-Robust Audio DeepfakeDetection

Audio deepfake detectors that work well on clean lab recordings often degrade after telephone or platform compression. This paper asks a specific question: which low-dimensional acoustic descriptors stay informative after codec degradation? We screen 36 utterance-level features(formants, PNCCs, LFCCs, prosody) on ASVspoof 2019 data and test on the official ASVspoof 2021 LA set across seven codec conditions. Of 33 statistically significant descriptors, only four show large practical separation (|r| >0.5, bootstrap 95% CIs), all first-formant (F1) measures, highly redundant (PC1: 79% variance). The payoff is stability: the 36-feature SVM degrades 14 pp under codec mismatch (9.5%-23.5% EER), while the four-feature RBF-SVM degrades only 5.9 pp (17.0%-22.9%). The F1 subset varies 3.4 pp across conditions versus 13.1 pp for the full set. F1 differences persist after controlling for speaker identity (p greater than 10^10, mixed-effects model). The results suggest that physically interpretable F1 statistics offer a compact, codec-stable diagnostic signal that may complement higher accuracy neural detectors.
Read the full abstract

Earth and Planetary Astrophysics

Imaging exoplanets below the telescope stability limit

Exoplanet imaging is the main method for detailed characterization of planets orbiting beyond 1 au but it requires suppressing the bright glare of the star to see the faint planet light. As well as sophisticated instrumentation, effective image processing is necessary for starlight suppression, which continues to be a substantial limitation. Standard algorithms use reference images for starlight modelling and thus they are limited by inevitable telescope instabilities that change from reference to science observation. Addressing this challenge will increase the number of planets amenable to direct imaging and will be a notable step towards imaging exo-Earths with future observatories. Here we present a new physics-based method for starlight subtraction. Our method combines a multistep optimization with a flexible physical model to produce a starlight model, allowing modelling and compensation of instrument instabilities. We use automatic differentiation (autodiff) to optimize millions of model parameters. We demonstrate our method on observations from the James Webb Space Telescope (JWST) Near-InfraRed Camera (NIRCam) coronagraph instrument. We achieve sensitivity 2-15 times superior to standard techniques, including the first detection of 51 Eridani b at 4.3 $\mu$m. This is the first time, to our knowledge, that such an approach has demonstrated substantially improved sensitivity on real, on-sky coronagraphic imaging. Our framework can be extended to other current and future ground-based and space-based observatories, including the Nancy Grace Roman Space Telescope and the Habitable Worlds Observatory (HWO), potentially relaxing the unprecedented stability requirements needed to make these a success.
Read the full abstract

All authors Rodrigo Ferrer-Chavez, Jason J. Wang, Katherine L. Bouman, William T. Freeman, Laurent Pueyo, Brandon Y. Feng, Marshall D. Perrin

Systems and Control

Image-Driven Control with Application to Thermoforming

This paper proposes Image-Driven Control (IDC), an approach to vision-based feedback that operates directly in the image space. IDC offers a direct end-to-end solution in the behavioral framework. Targeting a limited subset of vision-based feedback problems, including applications in manufacturing, the IDC design builds on a computationally advantageous subspace predictive control formulation and offset-free tracking to enable reference tracking in the image space under input and pixel-level constraints. IDC provides a control-theoretic alternative to vision-based feedback systems, enabling improved transparency for analysis and tuning of end-to-end control. It offers straightforward experiment design, an advantage in applications where data collection is expensive. The effectiveness of IDC is shown in a simulation example using a nonlinear high-fidelity model of the heating phase of the thermoforming process.
Read the full abstract

Solar and Stellar Astrophysics

Understanding Coronal `Seeing' using High Angular Resolution Radio Observations of Active Solar Emissions

Solar radio bursts are generated through the plasma emission mechanism by beams of accelerated electrons travelling along magnetic field lines. They serve as powerful probes of the ambient coronal plasma and magnetic fields. At low frequencies, however, propagation through the turbulent, magnetised coronal plasma strongly modifies their observed properties. Disentangling these propagation effects from the intrinsic nature of the emission therefore remains a major challenge. Despite the long history of observations, there are only a few that leverage high-angular-resolution imaging. This is primarily due to the prevailing expectations that scattering from coronal density inhomogeneities significantly increases the apparent source size, obscuring any fine-scale structures that might be present. While small-scale features are routinely observed in EUV and X-rays, few are detected at meter wavelengths. Numerous studies have focused on quantifying the effect of scattering on the observed apparent size of a point solar radio source. Naturally, their results vary significantly depending on the specifics of the model used to describe coronal inhomogeneities and the scattering prescription applied – ranging from sub-arcseconds to arcminutes. Recent high angular resolution observations with the Square Kilometre Array Observatory (SKAO) -pathfinders, uGMRT and LOFAR, have revealed unexpectedly compact sources, with sizes as small as $\sim9$ arcseconds in type-I noise storms. Forthcoming high-fidelity and angular resolution SKAO observations will transform our understanding of solar radio emissions by constraining scattering models, improving coronal diagnostics, and revealing fine structures crucial for understanding solar radio emissions and propagation effects.
Read the full abstract

All authors Soham Dey, Deepan Patra, Divya Oberoi, Anshu Kumari, Atul Mohan, Surajit Mondal, Pietro Zucca

Artificial Intelligence

Dependable Long-Horizon LLM Agents: A Trajectory-Centered Empirical Study of Where Errors Enter and Which Controls Reach Them

When an LLM agent fails part-way through a long task, the usual analysis records a symptom: a failure-mode label, or the agent and step judged responsible. It does not say where the error entered, which state carried it to the symptom, whether anything in the trajectory contradicted it, or which control could have caught it. We define an episode record that ties a candidate fault to its origin, propagation, detection, response and consequence, located by operational phase and state domain, and that names the control families whose observation and reach cover it. We turned the record into a written annotation protocol and applied it to 95 failed runs of six multi-agent systems from the public Who&When and MAD corpora, reading each trajectory in full. Faults enter early (median at 0.18 of the run), mostly as wrong decisions or fabrications over task state, and in 37 of 88 Who&When runs before the step the corpus' human labels mark as decisive. In 72% of failed runs the trajectory already contained evidence against the error before the failure; in 88% of those no agent acted on it. Something flagged a deviation in 31% of runs, but no response removed the deviated state. Independent judges reading the unused evidence, task contracts fixed at admission, and re-checking beliefs against the environment reach most episodes; restoring the agent's own state would have sufficed in 35 of 95 runs. A text scan of the raw corpus also shows that 45% of the hand-crafted Who&When failures ended with a provider content-filter crash after which the harness emitted a final answer no agent produced, a failure class that neither symptom taxonomies nor responsible-agent labels represent. Protocol, records and scripts are included.
Read the full abstract

All authors Shaohuang Wang, Jinghao Wen, Xulei Zhang, Xun Jiao

General Relativity and Quantum Cosmology

Polarized Equatorial Emission around Kerr Black Holes with Synchronized Scalar Hair. II. Indirect images

We investigate the intensity polarization properties of the indirect images of a geometrically and optically thin accretion disk around fully self-consistent models of Kerr black holes with synchronized scalar hair. The presence of a massive, self-gravitating scalar field can dramatically alter the lensing properties of the spacetime, resulting in morphologically complex indirect images. We show that for a representative high scalar charge model, this results in the presence of high order central ring-like images at low observer inclinations. These images are not exponentially demagnified, and their polarized intensity, relative to the direct images, is significant. We therefore conclude that the observational signature of strongly scalarized Kerr black holes may not be dominated solely by the $n=1$ image, as images with $n\geq2$ may also contribute significantly to the observed polarized emission, depending on the radial emissivity profile of the disk and the observer inclination.
Read the full abstract

All authors Valentin O. Deliyski, Tsanimir Angelov, Galin N. Gyulchev, Stoytcho S. Yazadjiev