Probability
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Tuesday 15 September
Cladograms of size $n$ are unrooted binary trees whose leaves are labelled from 1 to $n$. Aldous introduced in 2000 a Markov chain on cladograms, a step of which consists of removing a leaf uniformly at random and reinserting it on a uniformly chosen edge. We introduce a coupling for this walk, which follows multiple colored subtrees in parallel. We go around the lack of independence of the colored components by finding a relevant statistic, namely the sum of the squares of the sizes of all colored components, which we prove has a drift. We deduce that the mixing time of the walk is of order $n^2$, solving a conjecture of Aldous.
Signal Processing
Outage Analysis of RSMA-Enabled Terrestrial Users Under Co-Channel Interference from HAPS
The coexistence of terrestrial networks (TNs) with high-altitude platform station (HAPS)-based non-terrestrial networks (NTNs) is a promising approach for extending 6G connectivity, but the resulting cross-network interference can significantly affect TN reliability. This paper investigates an rate-splitting multiple access (RSMA)-based TN operating under interference from a multibeam HAPS-NTN system. The desired TN links are modeled using Nakagami-m fading, while the aggregate HAPS interference is characterized by shadowed-Rician fading. Closed-form outage probability expressions are derived by characterizing the aggregate interference and accounting for both perfect and imperfect successive interference cancellation (SIC). Monte Carlo simulations are used to verify the analytical results and to examine the effects of HAPS transmit power, the number of interfering beams, and SIC imperfections. The results demonstrate that the interference-to-signal power scaling is a key factor determining TN outage behavior, with jointly scaled HAPS interference leading to interference-limited performance. Moreover, RSMA provides improved outage performance compared with non-orthogonal multiple access (NOMA), particularly in the presence of HAPS interference
Read the full abstract
High Energy Physics (Phenomenology)
Quark fracture function at small $x$ from the Color Glass Condensate
The target-fragmentation region in deeply inelastic electron-nucleus scattering provides insight into the partonic structure of the target, as hadrons measured in the final state retain information about the nonperturbative dynamics of spectator partons inside the nucleus. In QCD factorization theorems, this nonperturbative dynamics is encoded in objects known as fracture functions. In this paper, we study the extended quark fracture function in the limit where the struck (anti)quark carries a very small longitudinal momentum fraction, $x$. We first present an elementary derivation within the Color Glass Condensate effective field theory of the quark fracture function in terms of the dipole operator at small $x$. At the parton level, the quark extended fracture function is sensitive to gluon saturation when the transverse momentum $P_\perp$ of the outgoing parton produced in the target-fragmentation region is smaller than the nuclear saturation scale $Q_s$, in which regime the $P_\perp$ distribution becomes approximately flat. We then show, through a numerical study that incorporates the convolution of this parton-level result with a collinear fragmentation function, that hadronization largely washes out these saturation effects, as the convolution predominantly probes partonic transverse momenta that are typically larger than $Q_s$.
Read the full abstract
Optimization and Control
Learning Choice Model Trees for Feature-Based Multi-Product Pricing: Exact Optimization and Field Evidence
Feature-based multi-product pricing uses customer characteristics to identify demand heterogeneity and tailor prices across products. Choice model trees segment customers through interpretable feature rules and fit a demand model within each leaf. Existing methods typically construct these trees greedily, selecting one myopic split at a time. We develop optimal choice model trees with multinomial logit leaves (OCMT-MNL), jointly optimizing the tree and leaf models within a prescribed depth. Our exact dynamic program derives closed-form Fenchel lower bounds during constrained Newton iterations and propagates them across nested and disjoint customer subsets, avoiding new fits and resuming unfinished fits without repeating completed work. In synthetic experiments, it reduces exact leaf fits by 99.98% and leaf evaluations by 86.13%, achieving up to 7.15-fold speedups over unpruned dynamic programming. One-dimensional lookup tables translate offline estimation into real-time pricing, with a revenue-loss bound quadratic in grid spacing under the fitted model. Compared with greedy trees, OCMT-MNL achieves lower revenue loss with fewer leaves on synthetic data and better predictive fit on real data. In a 23-week randomized experiment on ancillary seat pricing across 48 airline markets and 190,220 passengers, OCMT-MNL increases seat revenue per passenger by a statistically significant 11.3% over static pricing.
Read the full abstract
Astrophysics of Galaxies
An ALMA Band 7 survey of SDSS/Herschel quasars in Stripe 82: II. The nature of FIR-bright quasars
We investigate the star formation properties and ultraviolet (UV)/optical spectral characteristics of 142 far-infrared (FIR)-bright quasars at redshifts 1 < z < 4 to probe supermassive black hole (SMBH) and host galaxy co-evolution. By combining multi-wavelength observations with high-resolution ALMA submillimeter (submm) data, we overcome source confusion and robustly constrain star formation rates (SFRs). The inclusion of ALMA data to the Spectral Energy Distribution fitting reduces SFR estimates by ~20% for quasars with single submm counterparts and by a factor >2 for FIR-blended multi-component systems, while reducing FIR luminosity uncertainties by ~18%. The revised host SFRs span 500 and 3000 solar masses per year (median 1080 solar masses per year), confirming these systems are extreme starbursts. The very shallow positive correlation between starburst and accretion luminosities suggests that SMBH accretion and star formation in the host are not coupled on the timescales probed by the observations. Compared to the general population of FIR-faint quasars, FIR-bright sources have a factor of three higher detection rates at 1.4 GHz, predominantly following the FIR-radio correlation, and reduced CIV and Lya equivalent widths with unaffected MgII, suggesting AGN-driven winds that leave MgII relatively unaffected. Together, these results place FIR-bright quasars in a brief transitional phase, bridging heavily obscured "blowout" galaxies and optically unobscured blue quasars, where bulge assembly and SMBH growth peak simultaneously during ongoing stellar assembly.
Read the full abstract
Probability
Mass erasure on infinitely measured $\mathbb{R}$-trees I: construction and limit theorems
This paper is the first in a two-part series, seeking to i) extend the theory of mass erasure developed by Duquesne and Winkel to a setting with infinite measures, and ii) apply this theory to obtain invariance principles for Galton–Watson and Lévy forests in the most general setting where the trees may be supercritical and not satisfy Grey's condition. We consider $\mathbb{R}$-trees equipped with a suitable subclass of boundedly finite measures, which we call discretely infinite measures. Let $(T, d, \rho)$ be a complete and separable $\mathbb{R}$-tree, let $\mu$ be a discretely infinite measure on $T$, and let $h > 0$. The $h$-mass-erased subtree is obtained by removing all fringe subtrees with $\mu$-mass strictly less than $h$, and we prove that all $h$-mass-erased subtrees have a discrete branching structure. The mass-erased subtrees may be equipped with associated measures, which gives rise to a family of operators $(\mathscr{E}_h)_{h \geq 0}$ which forms a semigroup on the space of discretely infinite measures. The operators may be lifted to an appropriate space of Gromov-vague isometry classes. Given a sequence $\boldsymbol{\mu}_n = [T_n, d_n, \rho_n, \mu_n]$, $n \in \mathbb{N}$, of suitable isometry classes, we say that they converge vaguely in the sense of mass erasure if their mass erasures $(\mathscr{E}_h \boldsymbol{\mu}_n)_{n \in \mathbb{N}}$ converge in the Gromov-vague topology for all $h > 0$. We prove that by imposing suitable tightness conditions on $(\boldsymbol{\mu}_n)_{n \in \mathbb{N}}$, Gromov-vague convergence implies vague convergence in the sense of mass erasure. Under appropriate tightness, we identify conditions which relate the two notions of convergence, and prove that either mode of convergence implies local Gromov–Hausdorff convergence of the mass-erased subtrees. The applications to Galton–Watson and Lévy forests are considered in part two.
Read the full abstract
Mathematical Physics
Reflectionless edge states in Dirac models of topological insulators
We study an effective model of topological insulators joined along a bent interface. The Hamiltonian is a Dirac operator $\mathcal{D} = D_1\sigma_1 + D_2\sigma_2 + \kappa(x) \sigma_3$ on $L^2(\mathbb{R}^2,\mathbb{C}^2)$, where $\kappa(x)$ represents a domain wall with a corner: while initially straight, it undergoes a bend when passing from one side of $\mathbb{R}^2$ to the other. For most energies in a spectral window centered at $0$, we construct a reflectionless edge state of $\mathcal{D}$: a distorted plane wave with transmission coefficient of modulus $1$. This implies that $\mathcal{D}$ admits modes confined near $\kappa^{-1}(\{0\})$, that travel through the bend with no loss neither to the bulk nor to reflection. Our result explains experimental observations from the physics literature, as well as our own numerical simulations.
Read the full abstract
Artificial Intelligence
AntennaFlow: A Generative Flow Model for Offset Correction in Phaseless Antenna Testing
Near-field to far-field transformation is central to large-aperture antenna testing, yet two coupled challenges remain: costly phase acquisition at millimeter-wave bands and violations of the centering assumption under offset mounting. Existing methods address these issues separately, requiring either dense full-field data or offset vectors. We tackle both jointly by exploiting a key observation: amplitude fields under different offsets are coordinate-transformed views of the same near field. The challenge is to recover the center-aligned field from offset amplitudes without a phase or offset vector. We propose AntennaFlow, a three-stage framework: a contrastively learned encoder that maps offset views to an offset-invariant embedding, a deterministic flow-matching transport that maps offset amplitudes to center-aligned ones, and the Simplified Extrapolation Technique, whose Green-function Taylor expansion is valid only for centered fields. Experiments show that AntennaFlow enables fast, phaseless, offset-vector-free NF–FF reconstruction from sparse amplitude-only measurements, consistently outperforming existing baselines while preserving physical consistency.
Read the full abstract
Networking
AeroLat: Channel-Aware Latent Space Semantic Communication for Decentralized UAV Swarms
Communication in latent space offers an intriguing alternative to symbolic messages for decentralized autonomous Unmanned Aerial Vehicle (UAV) swarms operating over bandwidth-constrained, time-varying wireless links. However, when homogeneous frozen models are prompted with discretized perceptual inputs, their broadcast states collapse toward the shared prompt template. In view of this, we propose AeroLat, a channel-aware latent semantic communication framework that uses evidence injection. The resulting latent states are then passed through an explicit communication model that encompasses bandwidth-limited serialization, additive noise and information staleness, which facilitates a joint assessment of communication fidelity and swarm-level coordination. Across multi-seed simulations, AeroLat provably remains resilient to codec choice, faults and increasing swarm size. It consistently reproduces the latent-swarm anomaly, while no-whitening controls recover the collapse. In particular, AeroLat is capable of reducing false similarity by 97.5%.
Read the full abstract
Computer Vision
High-Fidelity Video Quality Assessment with VQA-Specific Saliency
No-reference video quality assessment (NR VQA) has recently seen promising progress with deep learning. However, video data is inherently large, and processing them with deep models incurs high computational cost. This challenge is particularly acute in VQA, where preserving original-resolution cues and dense temporal information is critical for accuracy. Existing efficiency-driven preprocessing strategies, such as fragmenting, reduce computation but alter the input data distribution, limiting effective reuse of pretrained video foundation models (ViFMs). To address these challenges, we propose High-Fidelity Video Quality Assessment (HFVQA), a framework built on fixed-size spatio-temporal (ST) patches that is fully compatible with pretrained ViFMs. HFVQA samples ST patches across multiple scales, including the original resolution, with minimal temporal subsampling to preserve low-level quality cues and semantic context. To limit computation, HFVQA introduces a lightweight auxiliary network trained end-to-end with the ViFM encoder to learn VQA-specific saliency. Distilled directly from quality supervision, this saliency captures task-specific importance patterns, reflecting that video quality perception is dominated by a small subset of spatio-temporal regions. By combining high-fidelity spatio-temporal cues with learned, task-specific saliency, HFVQA achieves SOTA performance on standard NR VQA benchmarks while processing as little as 12% of candidate ST patches, making high-fidelity ViFM-based VQA computationally tractable.
Read the full abstract
High Energy Physics (Phenomenology)
Matching relations for gluon TMDs up to one-loop accuracy
We present new results on the matching relations for gluon transverse momentum dependent parton distribution functions up to one-loop accuracy. At tree-level, employing the spinor formalism, we obtain these relations up to twist 3 accuracy for both T-even and T-odd distributions, including the hadron-mass corrections. At one-loop order, we extend the parton-in-parton approach to include higher-twist operators in the small-$b$ expansion. As a result, we derive for the first time the Wandzura-Wilczek relation for the worm-gear T distributions and establish a systematic method to include mass corrections at one-loop and twist-2 accuracies.
Read the full abstract
Signal Processing
Performance Analysis of HAPS-Assisted Downlink RSMA under Orthogonal and Full Frequency Reuse
This paper analyzes the outage performance of a high-altitude platform station (HAPS)-assisted downlink employing rate-splitting multiple access (RSMA). A realistic link budget accounting for free-space path loss, rain attenuation, and atmospheric absorption is combined with elevation-angle-dependent shadowed Rician fading. Closed-form outage probability and throughput expressions are derived for orthogonal frequency allocation and full frequency reuse, where the aggregate inter-beam interference is approximated by a moment-matched Gamma random variable, yielding a tractable finite-sum expression via its Laplace transform. The analytical expressions are validated through Monte Carlo simulations, showing close agreement. Numerical results show that the optimal common stream power allocation depends on the transmit power, while inter-beam interference under frequency reuse introduces an outage floor absent under orthogonal allocation, providing practical design insights for HAPS-RSMA systems
Read the full abstract
Combinatorics
The Distance Between the Adjacency Spectral Center and the Characteristic Set of a Tree
Let $S(T)$ denote the adjacency spectral center of a tree $T$, and let $C(T)$ be its characteristic set. We determine the maximum possible distance $\Delta_n:=\max_{|V(T)|=n}\operatorname{dist}_T(S(T),C(T))$ for every $n\ge 3$. It turns out that $\Delta_n=0$ for $3\le n\le 11$, that $\Delta_{12}=1$, and that $\Delta_n=\left\lfloor\frac{n-11}{2}\right\rfloor (n\ge 13)$. More generally, if the two sets are at distance $d\ge 2$, then the branch issuing from a closest vertex of $S(T)$ in the direction of $C(T)$ contains at least $2d+5$ vertices, while the complementary part of the tree contains at least $6$. We also establish structural restrictions in the equality case and completely classify the extremal trees of every odd order $n\ge 17$: there are exactly four nonisomorphic extremal trees, consisting of two brooms and two double brooms. Our proofs exploit the opposite extremal behaviour of paths for the adjacency spectral radius and for the bottleneck Perron value.
Read the full abstract
General Relativity and Quantum Cosmology
The Null Energy Condition through the Lens of Optimal Transport: Smooth and Non-Smooth Spacetimes
We survey recent developments on the null energy condition (NEC) from the perspective of optimal transport, with emphasis on non-smooth Lorentzian geometry. After recalling the role of classical energy conditions in general relativity, we discuss the relation between Ricci curvature and entropy convexity in optimal transport. The main focus is the authors' optimal-transport characterization of the NEC. In the smooth setting, the NEC is equivalent to displacement concavity of the Shannon entropy power along suitable null-geodesic transport plans supported on null hypersurfaces and measured with respect to rigged measures. We review the geometric ingredients of this characterization and applications, including a weighted Hawking area theorem. We then turn to the synthetic theory introduced in a second paper of the authors. Synthetic null hypersurfaces are defined using achronal boundaries, gauge functions, and reference measures in topological causal spaces. In this framework, the displacement-concavity characterization becomes a synthetic curvature condition, $\mathsf{NC}^{e}(N)$, meaningful without differentiable structure. We discuss compatibility with the smooth theory, invariance under natural changes of gauge and reference measure, and stability under convergence. Finally, we present a synthetic version of Penrose's singularity theorem for continuous Lorentzian metrics. These results indicate that geometric and causal consequences of the NEC persist beyond the smooth setting and highlight optimal transport as a framework connecting curvature, causality, and singularity formation in Lorentzian geometry.
Read the full abstract
Statistical Computation
Optimal Scaling of Langevin Proposals with Generalized Acceptance Rules
Langevin-based Markov chain Monte Carlo (MCMC) algorithms use gradient information to improve sampling, particularly in high dimensions. Classical optimal scaling theory for these algorithms has largely focused on the Metropolis-Hastings (MH) acceptance rule. However, there has been a recent surge in acceptance rules beyond MH for applications spanning differential privacy, stochastic MCMC, diffusion models, and molecular dynamics. We develop optimal scaling results for Langevin proposals employed with generalized acceptance rules belonging to a suitable class. For high-dimensional targets, we recover the usual $O(d^{-1/3})$ scaling, while different acceptance rules lead to different optimal acceptance probabilities.
Read the full abstract
Materials Science
Probabilistic Deep Learning Framework for Phase Transformation Forecasting aided by In Situ High temperature Microscopy
The mechanical performance of structural materials is governed by their microstructure, which is itself set by the thermal history imposed during processing. Predicting how that microstructure evolves along an arbitrary thermal trajectory, however, remains a challenging problem. Conventional continuous-cooling-transformation diagrams, for instance, provide only a static description, and they do not exist for all materials, especially emerging and novel ones. Deep-learning approaches address this issue, but they treat image-based characterization, temporal prediction, and microstructure forecasting as separate tasks. Here, we introduce a probabilistic deep-learning framework that unifies these tasks for in situ high-temperature confocal laser scanning microscopy data. The initial surface image is compressed with a beta-variational autoencoder into a low-dimensional latent representation, which, together with the cooling rate and the full temperature history, is passed to a temporal fusion transformer that produces a quantile forecast of the Widmanstatten ferrite ratio. In parallel, a Gaussian process predicts the end-state Widmanstatten ferrite ratio and corrects the forecast equilibrium level, yielding probabilistic prediction intervals. We demonstrate the efficacy of our method on Widmanstatten ferrite evolution in S235 steel across five cooling regimes and showcase significant accuracy. This verifies the relevance of our framework and paves the way for a streamlined and highly accelerated investigation of material evolution under thermal treatment.
Read the full abstract
Systems and Control
Review on Electric Railway System Optimization: Train Dynamic Scheduling, Energy Management, and Storage Integration
The modernization of railway systems is being driven by the need for greater efficiency, sustainability, and intelligent operation. In this review, recent advancements in dynamic scheduling, energy management, and energy storage systems (ESSs) integration within electric railway networks are analyzed. Optimization strategies for train scheduling and operation control are reviewed, with a focus on methods that reduce energy consumption and improve overall system performance. The role of energy storage technologies is analyzed in terms of energy management, peak shaving, and voltage and frequency control for practical application. The integration of artificial intelligence and advanced control strategies is also reviewed in electrified railway systems. This review provides a comprehensive overview of current research trends and outlines future directions for the development of resilient, energy efficient, and intelligent railway systems with ESSs integration.
Read the full abstract
Quantum Physics
Classical Communication Protocol based on Joint Classical-Quantum Coding
We introduce a robust quantum communication protocol that integrates classical error-control coding, entanglement distribution, and superdense coding. Classical error-correcting codes are used to mitigate dark counts and photon losses by determining the positions of qubit transmissions and protecting the data embedded through superdense coding. We derive conditions on the employed codes that guarantee successful error correction under a bounded error-frequency model. Moreover, upper bounds are derived on the performance of conventional superdense coding protected by classical error correction. It is shown that, under the same constraints on error frequency, suitable code configurations of the proposed protocol can exceed those upper bounds both in terms of data rate and energy efficiency. Finally, we develop a physical error model based on fiber attenuation, detector efficiency, dark counts, and time-slot duration, and use it to evaluate the effective performance of different configurations of error-correcting codes. The proposed approach is primarily suited for short-distance quantum links, as in Quantum Local Area Network (QLAN) where it can provide high communication throughput while integrating entanglement distribution directly into the communication process.
Read the full abstract
Machine Learning
Beyond Token-Local Imitation: Reward-Compatible Temporal Credit Assignment for On-Policy Distillation
On-policy distillation (OPD) has emerged as an effective approach for large language model post-training, yet existing objectives face a trade-off between objective fidelity and optimization stability. Token-level OPD provides stable but local supervision, whereas sequence-level OPD captures future credit at the cost of horizon-dependent variance. We establish a unified temporal-credit view of these formulations, showing that practical token-level OPD can be interpreted as a temporal approximation to the sequence-level reverse-KL gradient. Building on this connection, we propose $\gamma$OPD (GammaOPD), which uses discounted temporal credit assignment to balance long-horizon supervision and optimization stability, while admitting a horizon-independent variance bound. We further develop a reward-compatible bounded mixing (RBM) mechanism for $\gamma$OPD that balances verifiable outcome feedback with the discounted OPD advantage to move beyond purely teacher-dependent optimization. Experiments on mathematical and code reasoning demonstrate consistent improvements over existing OPD methods across vanilla, size-mismatched, and multi-teacher distillation settings.
Read the full abstract
Software Engineering
RepoAtlas: Guiding Coding Agents via Evolving Multimodal Repository Views
Large language model (LLM)-powered coding agents have made rapid progress in automating software engineering tasks, yet repository-level issue resolution remains challenging. Beyond generating a plausible patch, an agent must localize relevant code across interdependent files and maintain repository context that is both sufficient and focused. Code graphs expose non-local relations, but linear text interfaces obscure their topology; rendering the full repository graph yields visual representations that are too dense to perceive reliably, whereas a one-shot local view becomes stale as exploration proceeds. We present RepoAtlas, a training-free module that maintains evolving multimodal repository views through a select–project–refresh loop over a repository code graph. RepoAtlas combines evidence from the issue with the agent's current exploration state to select a task-relevant region under a fixed budget, projects the selected structure into complementary visual and textual representations, and refreshes the view when changes in the exploration state render it outdated. We evaluate RepoAtlas on SWE-bench Verified, where it improves the resolve rate by 2.4 points while reducing input tokens and model calls by 5.8% and 7.8% on average, relative to the strongest multimodal graph baseline, with consistent gains across three models of different families and scales.
Read the full abstract
Disordered Systems and Neural Networks
Nonlocal Magic across the Many-Body Localization Crossover
Nonlocal magic quantifies the minimum nonstabilizerness attainable under independent local unitary transformations on the two subsystems. Here, we use min-relative nonlocal magic (NLM) to characterize the crossover from ergodicity to many-body localization (MBL) in the random-field XXZ chain. Unlike entanglement entropy, NLM probes how entanglement is organized through the distance of the Schmidt spectrum from dyadic-flat stabilizer spectra. From weak to intermediate disorder, NLM evolves from an $O(1)$ Haar-like value into a size-enhanced dome, while entanglement remains volume-law, revealing a spectral reorganization not visible in the entropy. Deep in the MBL regime, a two-level cut-hybridization model captures the nearly binary Schmidt spectrum and explains why the mean and median NLM decay approximately as $W^{-1}$ and $W^{-2}$, respectively. Following a product-state quench, NLM overshoots and relaxes in the ergodic regime, whereas at strong disorder it grows slowly and approximately logarithmically. These results show that NLM resolves Schmidt-spectrum structure not visible in the entanglement entropy and provides a complementary probe of ergodicity breaking.
Read the full abstract
Computer Vision
MedPCFM-TED: One-Step Point Cloud Flow Matching for Implant Generation via Teacher-Guided Endpoint Distillation
Cranial implant generation is an important task in medical imaging. Recent point cloud based generative methods, particularly flow matching, offer strong reconstruction quality and efficient sampling, but still require multiple neural function evaluations during inference. This limits rapid generation of multiple plausible implant candidates. We propose Teacher-guided Endpoint Distillation (TED), a simple one-step distillation framework for conditional cranial implant generation on point clouds. TED trains a one-step student using teacher-guided endpoint supervision and geometric matching losses, while avoiding explicit path straightening. We evaluate TED on the SkullFix and SkullBreak benchmarks. TED achieves the best overall performance on the SkullBreak dataset, remains competitive on SkullFix, and provides the strongest Chamfer distance performance among the compared one-step methods. In addition, TED generates implants in approximately 0.04s per sample. These results show that one-step distillation can substantially accelerate conditional point cloud implant generation without sacrificing reconstruction quality.
Read the full abstract
Machine Learning
When Confidence Signals Disagree: Local and Global Confidence in Autoregressive Language Models
Modern predictive systems expose multiple quantities that are commonly interpreted as measures of confidence. However, these quantities can summarize different aspects of the predictive process. This distinction matters when confidence is used to evaluate reliability or inform downstream oversight and control. We investigate whether different confidence readouts are empirically interchangeable in an autoregressive language model by comparing local confidence, defined from the probability of the greedy-selected answer token, with global confidence, defined from modal-answer frequency under repeated sampling. Across MMLU and ARC Challenge, the two signals are weakly correlated and differ substantially in their association with correctness: global confidence is moderately associated with correctness, whereas local confidence shows little association. We further test whether question-level disagreement between the signals is associated with sampling instability. On ARC, larger local–global confidence gaps are associated with higher answer entropy, more distinct sampled answers, and lower modal-answer concentration. The gap–entropy association persists when disagreement and instability are estimated from disjoint stochastic samples, indicating that it is not explained by shared finite-sample variation. The corresponding relationship is substantially weaker on MMLU, where only 4% of questions exhibit sampling instability. These results show that confidence readouts derived from the same predictive system are not empirically interchangeable and that their disagreement can provide a diagnostic of unstable sampling behavior. Confidence should therefore be treated as an explicitly defined measurement rather than as a single intrinsic scalar property of a model, particularly when it is used to inform downstream evaluation, oversight, or control.
Read the full abstract
Operator Algebras
A uniform commutator bound in finite von Neumann algebras
Let $\mathcal{M}$ be a finite von Neumann algebra with normalized center-valued trace $T_{\mathcal{M}}$. We prove that every $A\in\mathcal{M}$ with $T_{\mathcal{M}}(A)=0$ admits a representation $A=BC-CB$, where $B,C\in\mathcal{M}$ satisfy $\|B\|\|C\|\le K\|A\|$ for an absolute constant $K$ independent of $\mathcal{M}$.
Read the full abstract
Statistical Methodology
Causal Discovery via Transformed Low-Rank Quantile Surfaces
We propose Low-Rank Quantile Surfaces (LRQS), a bivariate causal model in which, in the causal direction, an unknown monotone transformation of the conditional quantile surface admits a low-rank functional decomposition. LRQS subsumes location-scale noise models and post-nonlinear heteroscedastic noise models, while allowing multiple quantile bases to represent changes beyond location-scale effects. We prove generic identifiability of LRQS: the transformed quantile surface is low rank in the causal direction, whereas reverse representability under the corresponding constraints occurs only for exceptional, fine-tuned cause marginals. We provide a nonparametric causal score using a fitting procedure that alternates between rank-constrained approximation of discretized quantile surfaces and isotonic estimation of the unknown monotone transformation. Experiments on synthetic mechanisms with higher-rank distributional shape variation and strong nonlinear distortions, together with standard bivariate benchmarks, show that LRQS is especially effective when conditional distributional shape or observation distortion goes beyond existing location-scale assumptions.
Read the full abstract
Machine Learning
Repurposing Deep Limit Order Book Forecasting for Scenario-Conditioned Market Impact Modeling
Deep Limit Order Book forecasting models capture nonlinear market dynamics, but their ability to quantify the effects of counterfactual order book messages has not been systematically validated. We introduce a model-agnostic framework that compares a trained forecaster's predictive distributions before and after injecting mechanically valid counterfactual messages, defining short-horizon model-implied market impact. A Transformer-based forecaster recovered scenario rankings with a Spearman correlation of 0.99 and 97.2% directional agreement with realized historical outcomes among non-neutral scenarios. Observation-level analysis further showed that estimated impacts captured incremental sequence-dependent variation beyond scenario identity and the pre-event forecast. These results provide evidence that pretrained Limit Order Book forecasters can be repurposed for scenario-conditioned response modeling without retraining.
Read the full abstract
Quantum Physics
Exact efficient simulation of noisy logical magic states using Clifford stabilizers
The preparation of high-fidelity logical magic states is a crucial subroutine for universal fault-tolerant quantum computation (FTQC). Predicting the performance of FTQC and developing improved protocols rely on numerical methods to classically simulate logical magic state preparation in the presence of noise. Clifford logic on Pauli-stabilizer codes with circuit-level Pauli errors can be efficiently simulated using Pauli-stabilizer formalism, but the non-Clifford operations required to prepare logical magic states render generic simulation inefficient. We introduce Clifford-stabilizer simulation, an exact and efficient algorithm based on updating a Clifford-stabilizer group to simulate noisy preparation protocols for a broad class of logical magic states used to implement non-Clifford gates in the third level of the Clifford hierarchy under circuit-level Pauli errors. Clifford-stabilizer simulation applies to a range of operations that commonly appear in preparation protocols for such logical magic states, including Pauli-stabilizer measurements, logical Clifford measurements, and transversal non-Clifford gates. Our algorithm for Clifford-stabilizer simulation maps a non-Clifford circuit with sampled circuit-level Pauli errors to a Clifford circuit that exactly reproduces its measurement outcome distribution, achieving time and space complexities polynomial in relevant protocol parameters. We perform exact simulation of magic state cultivation up to fault distance 7 by Clifford-stabilizer simulation. Our method provides a route to perform exact benchmarking of large-scale logical magic state preparation protocols required for useful FTQC.
Read the full abstract
Cryptography and Security
Cybersecurity in Power Grids: Standards and Research Challenges
This paper examines Smart Grid cybersecurity, emphasizing the critical distinctions between IT and OT environments. It analyzes grid architecture, substation threats, and key international standards, specifically IEC 62351, IEC 62443, and ISO 27001. Finally, it overviews latest research trends, including AI-driven threat detection.
Read the full abstract
Machine Learning
Verbalizing Subliminal Learning Effects Using Text Optimization
Subliminal learning is a phenomenon in which a distillation dataset transmits traits from the teacher model that are not legibly encoded in the dataset itself. This introduces a new challenge for model development and creates new risks from data poisoning. In this work, we use text optimization to detect subliminal learning effects and describe them as legible prompts. Subliminal learning from a prompted teacher motivates our approach. We observe that this is a special case of context distillation and leverage this observation to show that, in theory, the prompted subliminal learning dataset identifies the teacher's prompt. We reduce recovering this prompt to a text optimization problem and present a method to approximately solve it. Our method, SALVE (Search-Aided Latent Verbalization), optimizes a soft prompt, queries the same model to verbalize it as text, and uses beam search to make the verbalization reliable. In the standard subliminal learning setting, SALVE reliably recovers legible prompts that name the teacher's trait, while common text optimization methods fail to do so. In addition, we find that there are settings in which SALVE recovers the teacher's trait from a dataset even when subliminal learning fails, but that modifying student training to improve context distillation can create subliminal learning effects. We lastly show that SALVE detects subliminal learning effects in three additional settings: (1) mixtures of subliminal learning data and unrelated data, (2) data generated when the teacher is biased via activation steering, and (3) subsets of real preference data selected via Logit-Linear Selection. Overall, our results deepen our understanding of subliminal learning and present SALVE as a method to proactively detect subliminal learning effects.
Read the full abstract
Computer Vision
Evaluating Mesh Reconstruction Methods for Crop Phenotyping
Phenotyping an agricultural crop is crucial for studying its entire life cycle, as it provides vital insights to improve yield and, ultimately, food production. Doing the same for crops grown on remote sites is a challenge for the specialists who cannot be available on-site. 3D reconstruction techniques offer a promising solution to this problem by enabling crop digitization, allowing specialists to access the resulting 3D crop models from anywhere at any time. In this work, we evaluate recent 3D reconstruction pipelines for crop phenotyping. We focus on 7 mesh reconstruction pipelines and measure the fidelity and consistency of their outputs qualitatively and quantitatively. Our results suggest that the meshes produced by the GGGS, PGSR, and 2DGS are preferable to the other pipelines, owing to their quantitative metrics and visually pleasing outputs. The GGGS pipeline is better than the second-best pipeline (2DGS) by about 27% on the radar chart with 5 dimensions, namely, User ratings, Chamfer distance, LPIPS, PSNR, and SSIM.