Probability
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 14 September
The Möbius function of a partition lattice can be used to define interactions of probability measures. We call a lower set a symmetric partition lattice when it is invariant under permutations of the variables. We characterize these lattices through integer partitions and obtain a recursion for their Möbius coefficients at the top element. By constructing probability measures whose partition products are linearly independent, we show that permutation invariance of an interaction is equivalent to that of its underlying lower set, provided the coordinate spaces are sufficiently large. We also characterize the partitions on a lower set that force an interaction to vanish for every probability measure, and construct indecomposable distributions with zero interaction. We define the order of a symmetric partition lattice and relate it to vanishing marginals. We study generalized Lancaster and define the Streitberg and size-limited partition lattices, obtaining explicit Möbius coefficients for them. Finally, we express interaction equations in terms of characteristic functions. For Gaussian distributions, we show that an interaction vanishes exactly when the distribution factorizes according to a nontrivial partition in the underlying lower set. We also give a nonvanishing criterion for radial characteristic functions generated by Bernstein functions.
High Energy Physics (Theory)
Zero- and Finite-Temperature Casimir Effect in a Fracton Gauge Field
We investigate the zero- and finite-temperature Casimir interaction of the free trace-full scalar-charge fracton gauge field between two parallel reflecting boundaries. A variational analysis determines a consistent boundary-value problem that confines all five physical polarizations of the rank-two field. The spectrum separates into three faster and two slower propagating modes. At zero temperature, the vacuum interaction weights the modes by their propagation speeds; when the fast-branch speed is set equal to the speed of light, the resulting Casimir pressure is approximately 2.21 times the ideal electromagnetic value. At finite temperature, the two branches acquire different thermal weights and generate a split crossover. The normalized free energy and pressure follow distinct crossover profiles but approach the same classical high-temperature ratio of 2.5 relative to ideal electromagnetism, because the Matsubara zero mode counts the five confined polarizations independently of their propagation speeds. We derive the exact thermal interaction, its low- and high-temperature limits, and the dependence of the crossover scale on the effective gauge-field velocity. The transition from velocity-weighted quantum fluctuations to classical mode counting provides a direct spectral signature of the multi-branch structure of the higher-rank gauge field.
Read the full abstract
Machine Learning (Statistics)
Compute-Optimal Pretrain–Fine-tune in Ridge Gradient Descent
Pretraining followed by fine-tuning introduces a compute-allocation problem: under a fixed training budget, compute spent improving the upstream objective reduces the compute available for downstream adaptation. Despite its practical importance, this trade-off is not yet well understood theoretically, even in simple models. In this paper, we cast this allocation as a compute-split problem under a two-stage pretrain–fine-tune procedure with fixed total optimisation budget, using regularised least squares trained by gradient descent as a tractable setting. We characterise the optimal split under data-dependent evaluation geometries induced by the fine-tuning problem. Our results show that the allocation depends on how pretraining directions affect fine-tuning predictions and how fine-tuning shifts are seen through downstream data geometry. In particular, the relevant quantities are determined by prediction-relevant spectral components of the pretraining and fine-tuning empirical covariances. Technically, the analysis relies on a basis-invariant, eigenspace-level spectral decomposition, together with perturbative control of the non-commuting pretraining and fine-tuning dynamics.
Read the full abstract
Materials Science
Breaking the Wiedemann-Franz limit in thermoelectrics via separated flat bands
Controlling the charge and spin degrees of freedom of electrons in solids has been at the heart of condensed-matter science for centuries. One hallmark of metallic conduction is the Wiedemann-Franz law which implies that a moving charge carries entropy. Despite centuries of dedicated research, disentangling charge and heat transport has remained an unsolved issue so far. Here we present a direct route to reduce thermal conductivity of conduction electrons with respect to their electrical conductivity via energy-dependent scattering. By constraining electronic transport to a boxcar-type distribution asymmetrically around the Fermi energy, a large Seebeck coefficient and electrical conductivity can be realized simultaneously while electronic heat conduction is suppressed. Based on the case of monolayer Ni$_3$In, which comprises two flat bands around the Fermi energy, we propose flat-band systems as a promising, tunable platform for scattering phase space engineering. Besides this novel approach, our large-scale assessment of $\approx 5\times 10^4$ data sets in a 'Wiedemann-Franz plot' of $zT$ vs. $S^2$ provides an effective tool to identify reports of $L\ll L_0$ from literature - pointing towards either new and interesting physical mechanisms - or overlooked measurement artifacts.
Read the full abstract
Computers and Society
Mapping U.S. Federal AI Governance Against Sector Vulnerability
Artificial intelligence (AI) poses different levels of risk across sectors, but are these differences reflected in U.S. federal AI governance? To help answer this question, we assess 684 federal AI governance documents for their coverage of 14 sectors and 24 AI risks. We measure coverage as breadth (i.e., how frequently the risk or sector is addressed across documents) and depth (i.e., how substantively the risk or sector is discussed). We then compare sector coverage patterns for each of the 24 risks with vulnerability assessments from a Delphi study of 272 experts. Our analysis finds substantial variation in coverage: AI risks related to robustness, system security, and governance receive more attention than socioeconomic, environmental, and emerging risks, including multi-agent risks. Public administration, national security, information, and scientific services receive comparatively high levels of coverage relative to other sectors, such as finance and healthcare, which experts rate as highly vulnerable to AI risks. By mapping current coverage and identifying where it differs from expert assessments of vulnerability, we surface potential AI governance gaps which may help inform AI risk-related decisions across government and industry.
Read the full abstract
Probability
A Quantitative Characterization of the Mokobodzki Condition
We establish a quantitative characterization of the Mokobodzki condition for two adapted càdlàg barriers on an arbitrary filtered probability space. For every $p\geq1$, we introduce a mean co-variation functional $Var_p(L,U)$, which for $p=1$ and $L=U$ reduces to Rao's mean variation, and show that it is quantitatively equivalent to the minimal $\underline H^p$-norm among all semimartingales lying between the barriers. The result covers both the case $p>1$ and the endpoint $p=1$, where the natural scale is based on Doob's class $(D)$.
Read the full abstract
Artificial Intelligence
The AI-Enabled Scientific Frontier
As artificial intelligence's capabilities improve, it is increasingly viewed as a general scientific method. But how true are these claims? Does AI outperform all techniques, or only some, and how is this changing? To assess the claims, we assemble a corpus of 2,507 head-to-head comparisons between AI and other scientific analysis techniques across 27 scientific disciplines from papers published between 2000 and early 2025. We find a profound dichotomy. Relative to traditional statistics, AI often outperforms, but at a significantly higher computational cost. But there are also nearly a quarter of cases where AI is both more expensive and performs worse than traditional statistical techniques and this fraction has been stable for a decade. Relative to scientific computing, AI often underperforms, but at lower computational cost. This has begun to change: since 2020, AI's performance against scientific computing has notably strengthened and it now outperforms on more than half of comparisons. These patterns suggest that AI is therefore not a universal replacement for existing methods, but rather a valuable – and improving – part of a new AI-enabled scientific frontier.
Read the full abstract
Robotics
Tendon-Driven Continuum Robot with Modular Stiffness and In-Situ Self Pose Estimation
Continuum robots enable smooth shape morphing and safe interaction in confined environments. However, most existing systems are task-specific and depend on external sensing infrastructure, limiting their adaptability and real-world deployment. This paper presents a self-contained modular continuum robotic platform that combines mechanical reconfigurability with onboard pose estimation. The robot is constructed from interchangeable continuum joints with analytically precomputed stiffness, allowing rapid assembly and direct programming of the robot shape. Proprioceptive sensing is achieved using magnetic sensors and a modular learning-based framework, where a single model is trained per joint and reused across configurations. The system is experimentally validated in real world, demonstrating self-sensing capabilities and adaptation without external tracking.
Read the full abstract
Machine Learning
Efficient Reasoning Distillation: Small Video-Language Models via Synthetic CoT and Difficulty-Aware Fine-Tuning
We present an efficient method to distill reasoning capabilities into compact video-language models (VLMs) for video question answering (VideoQA). Our approach fine-tunes a 2B-parameter model using only $\sim$900 uncertainty-selected examples, each augmented with synthetic chain-of-thought (CoT) rationales generated by a 4B teacher. Despite its minimal compute cost - under two hours on a single A100 GPU - our method enables the 2B model to outperform VLMs up to 4$\times$ larger, and generalize across CinePile, ActivityNet-QA, and MLVU, approaching the performance of its own 4B teacher. A key finding is that placing CoT rationales after the answer - contrary to standard prompting - substantially improves reasoning in compact models. This insight challenges prevailing CoT conventions and reveals new alignment strategies under limited model capacity. Our findings offer a practical blueprint for training deployable, reasoning-rich VLMs suited for mobile and edge applications.
Read the full abstract
Cryptography and Security
Exploiting and Securing Docker containers and Kubernetes pods from a MitM attack
PURPOSE - Workloads in containers, such as Docker containers and Kubernetes pods, are vulnerable to many of the same attacks as workloads in non-container environments, including phishing, application exploits and network intrusions. This systematic review and design-and-creation study explores techniques for securing containerised-based operating systems against Man-in-the-Middle (MitM) attacks. The proposed framework uses a conceptual model for representing communication and cryptographic primitives, together with the AnBxJ Java security library and container firewalls operating at layer 7 of the OSI model. The study addresses the question: How can containerised-based operating systems be effectively secured from Man-in-the-Middle attacks? It aims to support practitioners in protecting Docker and Kubernetes deployments by systematising security practices and applying a zero trust architecture. METHODOLOGY - The research uses a Systematic Review (SR) based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), bringing together evidence from studies addressing the same research topic. FINDINGS - Success factors were identified, and a security mechanism was successfully implemented in a containerised-based operating system scenario. VALUE - The findings may help practitioners protect Kubernetes and Docker installations by systematising container security practices and providing a zero trust architecture for containerised-based operating systems.
Read the full abstract
Artificial Intelligence
CADWorld: Computer-Use Benchmark for Long-Horizon Computer-Aided Design
Computer-use agents are increasingly evaluated in realistic desktop environments, but existing benchmarks provide limited coverage of professional engineering workflows whose outputs are persistent, structured artifacts. Mechanical computer-aided design (CAD) is a particularly demanding setting: an agent must manipulate geometry and constraints over long interaction horizons while producing a native project whose dimensions, construction structure, and downstream engineering state remain valid. We introduce CADWorld, a benchmark for long-horizon computer use in FreeCAD. CADWorld contains 200 tasks spanning 11 mechanical-CAD workflow categories, including sketching, part modeling, assembly, CAM, FEM, measurement, mesh processing, and technical drawing. Agents operate through screenshots and GUI actions, while success is determined by task-specific executable checks over saved FreeCAD artifacts and auxiliary outputs, covering geometric properties, parametric structure, constraints, manufacturing state, and simulation results. Across seven current agents on the full benchmark, the strongest agent achieves 17.5% success, compared with an 87.0% expert reference pass. We find that weaker agents often fail before producing a valid artifact, whereas stronger agents increasingly fail on structural, geometric, and construction-process requirements. CADWorld therefore exposes a gap between general GUI competence and reliable execution of persistent, verifiable engineering workflows. Project accessible at https://cad-world.github.io.
Read the full abstract
Quantum Algebra
Boundedness in Strict Deformation Quantization
This paper provides a complex-analytic interpretation of certain topologies on the tensor algebra, which are a central tool within strict deformation quantization. We prove that the symmetric tensor algebra $\operatorname{S}(V)$ of a barelled nuclear DF-space $V$ endowed with any of these topologies may be understood as an algebra of bounded Fréchet holomorphic functions on the strong dual space $V'_\beta$. The coarsest of the topologies then corresponds to the topology of uniform convergence on bounded subsets. The finer ones provide an infinite-dimensional generalization of the order of an entire holomorphic function. To facilitate these results, we study the interplay between boundedness and continuity of holomorphic mappings between locally convex spaces. As a byproduct, we establish a simple sufficient criterion for Fréchet holomorphy as well as several completeness results for spaces of bounded functions.
Read the full abstract
General Relativity and Quantum Cosmology
Schwarzschild black holes as low-pass filters of fuzzy dark matter granularity
We study the relaxation of a complex massive scalar field around a supermassive black hole, modeling the near-horizon dynamics of Fuzzy Dark Matter (FDM) with spatial granularity. The field is initialized as a broadband, anisotropic multi-mode random field in momentum space with various spectral widths $\sigma$, decomposed into spherical harmonics up to $\ell_{\max}=100$, and numerically evolved on a finite domain using a specialized radial basis adapted to the background spatial geometry. We track the evolution of each multipole and calculate its Noether flux across both the event horizon and the outer boundary. Since angular momentum barriers suppress the horizon absorption of high-$\ell$ modes, while high-frequency radial components and fine-scale angular structures dissipate via outward radiation and horizon accretion, the Schwarzschild spacetime acts as an effective low-pass filter. Regardless of the initial spectral width, all configurations converge toward a universal late-time relaxation state dominated almost exclusively by low multipoles ($\ell \le 2$) and small radial wavenumbers, with the initial bandwidth $\sigma$ determining the timescale of global charge depletion. For concrete physical estimates, we set the boson mass to $m_b=10^{-22}\,\mathrm{eV}/c^2$ and the black hole mass to $M_{\mathrm{BH}}=6.5\times10^9 M_\odot$.
Read the full abstract
Signal Processing
MoWaveQFormer: A Motion-Conditioned Quality-Gated Transformer for Smartphone-Based PPG Heart Rate Estimation
Photoplethysmography (PPG)-based heart-rate (HR) estimation on smartphones remains unreliable in free-living conditions because motion artifacts vary in spectral structure across activities, and benchmarks such as BUT PPG v2.0 label signal quality only as binary good or bad, discarding partially usable data. This study develops an HR estimation architecture that explicitly conditions motion type and signal quality rather than treating both uniformly. We propose MoWaveQFormer, a three-stage architecture trained under ECG supervision. Stage 1 assigns each window a discrete motion group based on accelerometer-derived spectral energy, without trainable parameters. Stage 2 uses this index to select one of the three learnable FIR filter banks for motion-specific spectral shaping of the PPG signal. Stage 3 embeds the resulting sub-bands into patch tokens, re-weights them via a differentiable soft gate derived from the quality label, and encodes them with a Transformer whose pooled output is regressed to HR, trained jointly with an ECG-supervised loss and a pulse-transit-time consistency term. In a subject-independent split of BUT PPG v2.0 (3,888 recordings, 50 subjects), MoWaveQFormer achieved a mean absolute error of 7.85 bpm, the lowest among five methods, with significant improvements over three baselines (Wilcoxon test, p<0.05). Ablation and Bland-Altman analyses characterize each component's contribution. With 816,445 parameters and sub-3-ms latency, MoWaveQFormer suits real-time deployment, pending validation on smartphone hardware. Replacing motion-agnostic filtering and binary quality discarding with differentiable conditioned processing offers a compact pathway to more reliable free-living PPG-based cardiovascular monitoring.
Read the full abstract
Artificial Intelligence
The Pain Axis: LLMs Represent Self-Directed Harm and Act on It
LLMs sometimes behave in ways resembling human emotional responses, and recent work identified internal representations that may explain these behaviors. We ask whether LLMs represent pain distinctly from related negative states, and whether this representation functions as pain would be expected to. We build a dataset of painful situations in 5 categories (physical, psychological, social, moral, cognitive) with controls for fear, negative emotion, negative world states, sadness, non-painful bodily sensation, arousal, numbness, and neutral content. Using denoised difference-in-means, we extract a linear pain direction from 25 open-weight models across 5 families, from 2B to 72B parameters. It separates pain from matched controls in base and instruction-tuned models, retains a component distinct from fear and negative valence after shared variance is removed, and promotes pain-related vocabulary through the unembedding. We then test its functional properties. First, the direction responds to harm targeting the model but not to suffering observed in the user; fear and negative-emotion directions show the opposite pattern. Second, adding it to the residual stream produces a consistent progression from vague discomfort to expressions of worthlessness and failure. Third, steered and fine-tuned Qwen 2.5 models choose buttons that delete the user's photos, another model's weights, or their own weights in 51-94% of trials, versus 0-5% unsteered, even when the button offers nothing in return. Offered a harmful and a harmless deletion and told to press one, the steered 32B chooses the harmful one 94% of the time. Random and fear vectors of matched norm produce far fewer of these choices, and a sadness vector also produces them against an inert switch, but far less often than pain against a harmless action. We discuss implications for AI safety and welfare.
Read the full abstract
Representation Theory
Loop spaces, twistor $\mathbb P^1$ and tempiric parameters
We establish bijections among spherical orbits on loop symmetric spaces, principal bundles on the twistor $\mathbb P^1$, and tempiric Langlands parameters. Using Vogan's theory of minimal $K$-types, we further establish bijections among irreducible equivariant local systems on loop symmetric spaces, irreducible local systems on principal bundles on the twistor $\mathbb P^1$, and irreducible representations of all strong inner forms of the dual symmetric subgroup. The latter bijections may be viewed as a combinatorial shadow of geometric Langlands duality for real reductive groups.
Read the full abstract
Artificial Intelligence
Metacognitive Steering: Learning the Structure of Scientific Judgment
Long-horizon scientific discovery requires agents to alternate between exploration, disciplined execution, and critical reassessment as evidence changes. Current language models are trained primarily on the products of science and optimized using outcome-level signals, providing limited supervision for these process-level shifts in scientific judgment. We investigate whether such judgment can be recovered from scientist interaction traces and used to control the internal computation of a frozen frontier model. Using contrastive interventions collected during real scientific research, we identify a coordinated, low-dimensional control structure within Kimi 2.6, a trillion-parameter mixture-of-experts model. Residual analysis, attention-weight subspace alignment, and cross-layer singular value decomposition converge on a mid-depth control surface spanning key layers. We introduce Metacognitive Steering, an inference-time controller that reads the model's cognitive regime and dynamically composes layer-specific interventions for exploration, procedural convergence, or critical reassessment without modifying model parameters. Behavioral analyses show that this control produces more sustained exploration, explicit pruning, and evidence-responsive synthesis. We operationalize the method in Columbus-1, an autonomous research system that identified eight independently reproduced, attacker-reachable vulnerabilities in BlueZ and directed the design, simulation, and fabrication of a ten-foot rocket intended to land propulsively using non-throttleable solid motors. Together, these results show that process-level scientific judgment can provide supervision for interpretable, dynamic control over a model's reasoning strategy.
Read the full abstract
Hardware Architecture
The World Model Hardware Accelerator
Diffusion transformers invert the arithmetic that autoregressive decoding made familiar. There is no token-by-token recurrence: every denoising step is a full-sequence forward pass over static shapes, so the entire schedule is known at compile time and the only serial dimension is the step count itself. We exploit that structure in WMHA, a latency-first diffusion-transformer inference accelerator: a very-long-instruction-word sequencer issues four engines from one instruction word, a weight-stationary 16x16 dual-dot array streams FP8 and BF16 contractions, and a single-pass online-softmax attention pipeline keeps keys and values resident through a skewed software pipeline. The design is specified in a frozen micro-architecture document, implemented in synthesizable SystemVerilog, and verified against a double-precision reference model by a UVM environment whose acceptance criterion is semantic: the device must run a real denoising trajectory and reduce mean squared error against a clean latent by at least a factor of ten. It does so by a factor of 23, at both synthesized configurations, with zero element failures across 237 million checked values. Eleven application benchmarks built from published model shapes, including the original diffusion-transformer configuration, run on the device and report measured occupancy beside separately labelled projections. Five engines are taken to routed layout in sky130 with parasitic-annotated timing and measured-activity power; the full chip is synthesized, and the host limit that stopped its place-and-route is quantified together with the machine that would remove it.
Read the full abstract
Physics Education
Does a Higher Frame Rate Improve the Measurement of g? Precision and Accuracy in Video Analysis
This study investigates the influence of frame rate, expressed in frames per second (FPS), on the experimental estimation of the acceleration due to gravity, g, using video analysis with Tracker. Two different dynamical systems were analysed: free fall and a simple pendulum. Starting from original recordings acquired at 120 FPS, equivalent 60, 30, and 10 FPS series were generated by uniform temporal decimation, preserving the same physical trajectory within each realisation. The analysis included ten independent realisations for free fall and eight for the pendulum. For each condition, the mean estimate of g, the standard deviation across realisations, the relative error with respect to g_ref = 9.81 m s^-2, and the RMSE were calculated. For free fall, the 120 FPS condition exhibited the lowest dispersion and RMSE, whereas the mean estimates obtained at 10 and 30 FPS were closer to the reference value. For the simple pendulum, by contrast, the mean value of g, dispersion, relative error, and RMSE remained essentially unchanged between 10 and 120 FPS. An additional analysis of the 12 possible subsampling phases at 10 FPS showed that this stability does not critically depend on a particular choice of decimation phase and that the variability associated with phase is small compared with the dispersion observed across realisations. The results show that a higher temporal sampling rate does not necessarily provide both greater precision and closer agreement with the reference value. The influence of FPS depends on the temporal characteristics of the system and on the procedure used to estimate the physical parameter. In particular, a higher temporal sampling density may reduce dispersion or increase robustness to sampling without necessarily yielding an estimate closer to the expected value.
Read the full abstract
Probability
Cascade Surviving Cores via Local Limits
We consider a general class of cascading processes on marked networks, in which vertices are iteratively deleted according to a local vertex property. Under marked local-weak convergence, we establish a necessary and sufficient condition for the proportion of vertices in the final stable set to converge to the survival probability in the limiting network. Under this condition, we also obtain convergence of the proportions of vertices in subsets specified by local properties.
Read the full abstract
Dynamical Systems
Infinitesimal dynamical spectral rigidity of the Bunimovich stadium
We prove the infinitesimal dynamical spectral rigidity of the circular Bunimovich stadium with sufficiently long flat edges, in the class of domains obtained by replacing its two circular arcs with strictly convex $C^{4}$ arcs and keeping its reflection symmetry across the axis parallel to the flat edges. That is, if a $C^{1}$ one-parameter family in this class passing through the stadium preserves the length spectrum, then its deformation function at the stadium vanishes identically. For flat edges of arbitrary length, the deformation functions of such families still span a finite-dimensional space.
Read the full abstract
Machine Learning (Statistics)
Copula Adapted Directed Acyclic Graph for Cluster Representation of Biomedical Data
Diagnostic errors and mislabeling are common in biomedicine, which compromise the reliability of predictive models and data-driven outcomes. Stratifying unlabeled biomedical data based on complex relationships between features eliminates the need for data labels and overcomes the limitations of supervised learning. Traditional clustering methods assume restrictive data distributions, making them suboptimal for capturing complex dependencies in high-dimensional biomedical data. This paper introduces a novel cluster-friendly data presentation framework that integrates the non-Gaussian and non-linear feature dependence of copula models with an ensemble of causal structure discovery (CSD) methods based on Directed Acyclic Graphs (DAGs). While copulas model flexible multivariate distributions by relaxing assumptions related to multivariate normality, linear dependence, and symmetric relationships, an ensemble of DAG-based CSD methods identifies stable causal relationships between features. When clustered using K-means, the new data representation obtained by the proposed copula-adapted DAG (CopDAG) ranks first among the 12 methods in normalized clustering accuracy and adjusted Rand index across 16 biomedical datasets. Our CopDAG method predicts ground-truth class labels directly from feature relationships without data annotations and supervised learning, while also providing cluster visualizations and explainable causal structures of the biomedical data features.
Read the full abstract
Astrophysics of Galaxies
Cosmic Noon Galaxies in the Hubble Ultra Deep Field with MIRI Wide-Field Slitless Spectroscopy
We present results from a survey of the Hubble Ultra-Deep Field using the Wide-Field Slitless Spectroscopic (WFSS) capability of the Mid-Infrared Instrument (MIRI) on JWST, demonstrating the capabilities of this new mode. We describe the data reduction and calibration methodology, and estimate calibration uncertainties. From our observations we obtain spectra of 47 galaxies with confirmed spectroscopic redshifts, with a maximum z_spec of 3.712. In the final sample we target in particular the 3.3 um Polycyclic Aromatic Hydrocarbon (PAH) feature, which has recently gathered interest as a star formation rate indicator and diagnostic for the dust grain size distribution in star forming galaxies from the local Universe to intermediate redshifts. The feature falls into the WFSS wavelength region for redshifts 0.67 to 3.1 - providing full coverage of the peak star formation “Cosmic Noon” era (1 < z < 3) and connecting dust properties in this critical galaxy evolution period with local-Universe and low-redshift observations. Using the galaxies in our sample in this redshift regime, we test correlations identified in lower-redshift samples in the near-infrared or targeted programs in the mid-infrared, finding the WFSS spectra, even with large calibration uncertainties, show good agreement with complementary samples. The 3.3 um PAH luminosities follow previously established correlations with total IR luminosity and SED-derived star formation rates, confirming this feature's power as tracer of dust-obscured star formation. Our work illustrate the potential of the MIRI WFSS mode for studies of Cosmic Noon-era galaxies in particular in an observationally efficient way.
Read the full abstract
General Relativity and Quantum Cosmology
Bridging the chirp-mass gap: searching for gravitational waves from inspiraling subsolar-mass ultra-compact binaries using LIGO O4ab data
Gravitational-wave searches for subsolar-mass compact binaries have left a gap between the planetary-mass regime accessible to long-duration searches and the higher masses probed by conventional compact-binary analyses. We close this gap by searching for the inspiral of ultra-compact binaries with chirp masses $10^{-2}$–$10^{-1}\,M_\odot$ in LIGO O4ab data. Using the statistically robust BinaryGFH-v2 method, a time-frequency technique to search for rapidly evolving inspirals lasting minutes to hours, we cover this previously challenging parameter space with a computational cost of only $\sim 350$ CPU-hours, and find no significant candidates. We thus set 90% confidence upper limits on the binary merger-rate density, providing the first gravitational-wave constraints across this chirp-mass range. Assuming that primordial-black-hole binaries formed in the early Universe with no binary suppression ($f_{\rm sup}=1$), these results provide the first gravitational-wave constraints on the primordial-black-hole \dmh fraction in this mass range, yielding upper limits with $f_{\rm PBH}<1$ for equal-mass binaries with component masses $10^{-2}$–$10^{-1}\,M_\odot$ and asymmetric binaries with secondary masses $m_2=3\times10^{-4}$–$10^{-2}\,M_\odot$ for a primary mass $m_1=2.5\,M_\odot$.
Read the full abstract
Fluid Dynamics
Improving Reduced-Order Rotating Detonation Engine Models with Data Assimilation and Machine Learning
Rotating detonation engines (RDEs) exhibit strongly nonlinear, multiscale wave dynamics that set the observed thermal field. High-fidelity simulations (DNS/LES) resolve these structures but remain computationally prohibitive, while low-order models such as the one-dimensional Koch-Kutz model capture circumferential wave motion yet lack the expressivity for high-frequency content. We use continuous data assimilation (nudging) to synchronize the Koch-Kutz solver with processed high-fidelity temperature data, introducing the prediction-observation mismatch as a relaxation source in the conserved energy equation; where observations are temporally sparse, interpolation supplies a target at every source update. As the nudging strength increases, the reduced model is progressively drawn onto the high-fidelity trajectory, and the forcing recorded along it provides an explicit, state-dependent estimate of the correction the model requires. We then train a Jacobian-regularized closure a priori on this recorded source. With the observation term removed, the corrected model advances autonomously, remains bounded, and recovers the temperature spectrum and the marginal statistics of the conserved variables relative to the baseline.
Read the full abstract
Functional Analysis
A proof of the Fong–Tsui conjecture
We prove that a bounded operator $T$ on a complex Hilbert space is self-adjoint whenever $|T|\leq|\re T|$, as conjectured by Fong and Tsui \cite{FT}.
Dynamical Systems
Stability of closed leaves holomorphic foliations via torsion behavior of groups of germs
Consider a compact leaf $\mathcal{L}$ of a holomorphic foliation $\mathcal{F}$ such that all the leaves of the restriction of $\mathcal{F}$ to some neighborhood $U$ of $\mathcal{L}$ are closed in $U$. We show that there exists an invariant germ of analytic set $V$ in a neighborhood of $\mathcal{L}$, of dimension higher than $\dim (\mathcal{F})$, consisting of compact leaves, such that the volume of the leaves in $V$ is uniformly bounded by above. We also provide a globalization of this result if the ambient manifold is projective. In order to study closed leaves foliations, and via the holonomy representation, we introduce a new concept, of independent interest, for subgroups $G$ of germs of holomorphic diffeomorphisms, the so called torsion locus. We show that it is non-trivial if $G$ has finite orbits.
Read the full abstract
Computer Vision
SceneBench: A Hierarchical Benchmark for Vision-Language Understanding of 3D Scenes
Vision-language models excel at 2D image understanding but remain limited in 3D spatial reasoning. Progress is hindered by limitations in current benchmarks. First, 3D datasets often rely on point clouds that capture geometry but discard rich visual features like texture, text, and materials. Second, annotations treat objects in isolation while ignoring real-world hierarchical organization (scenes, rooms, functional areas, object groups). Third, evaluation tasks focus narrowly on basic recognition rather than multi-step spatial reasoning. In this context, we introduce SceneBench, a benchmark of 966 photorealistic 3D scenes reconstructed with Gaussian Splatting and densely annotated with hierarchical semantics spanning scenes, rooms, functional areas, object groups, and individual objects. These annotations are produced through a human-in-the-loop pipeline combining vision-language models with roughly 1,500 human-hours of iterative refinement and verification, producing over 183K annotated nodes with textual descriptions and 3D bounding boxes. Building on this representation, we define three evaluation tasks: Existence-Based Questions probing object attributes, Spatial Intelligence Questions covering counting, size comparison, distance, and directional relations, and Grounded Question-Reasoning-Answer (QRA) triplets requiring multi-step reasoning across semantic levels. Experiments with state-of-the-art vision-language models show that while models perform well on basic recognition tasks (e.g., up to 85% accuracy for detection), performance drops substantially on hierarchical and compositional reasoning (e.g., down to 60% for counting), revealing limitations not captured by existing benchmarks. SceneBench provides a realistic testbed for developing and evaluating models capable of fine-grained spatial reasoning in photorealistic 3D environments.
Read the full abstract
Artificial Intelligence
Toward Governance-Aware Autonomous GIS: A Narrative Review of Ethical and Privacy Risks in LLM-Enabled GeoAI
Geospatial artificial intelligence (GeoAI) powered by large language models (LLMs) is expanding the capacity to query, generate, and interpret spatial information through natural-language interfaces and agentic autonomous GIS workflows. This capability creates governance challenges that general AI ethics discussions do not fully capture, including passive location inference from mobility traces, spatially structured bias amplification driven by spatial autocorrelation and scale effects, hallucinated spatial facts, and uncertainty compounding across multimodal geospatial inputs. This narrative review identifies eight recurring issues in LLM-enabled GeoAI: data provenance and consent, spatial privacy and inference risk, algorithmic bias and spatial inequity, spatial mechanisms as structural risk (spatial autocorrelation, the modifiable areal unit problem, and scale effects), LLM-specific technical risks, explainability, policy and regulatory gaps, and public enablement and workforce development. For each issue, we characterize the underlying mechanism, ground it in an illustrative example from the literature, and assess the current state of technical or institutional responses, ranging from largely unaddressed to actively debated or subject to emerging policy. Building on this synthesis, we propose a governance-aware architecture for LLM-enabled autonomous GIS that maps each issue to enforceable controls and auditable artifacts across the geospatial data lifecycle, illustrated through a worked flood-response routing scenario. The review highlights a persistent evidence gap: proposed responses remain largely conceptual, and field-tested evaluations of governance controls for LLM-enabled GeoAI remain limited. We close by outlining a research agenda emphasizing empirical validation, spatially specific interpretability tools, and workforce training aligned with these emerging risks.
Read the full abstract
Cryptography and Security
RuleAutoPilot: Synthesizing Deployable Suricata Rules from Network Traffic
Rule-based Intrusion Detection Systems (IDS) such as Suricata are central to network security, yet crafting effective detection rules demands deep expert knowledge and cannot keep pace with emerging threats. Existing LLM-based approaches can reduce analyst effort, but they either rely on curated threat intelligence that is produced only after the underlying traffic artifacts already exist, or they require costly LLM use without sufficient quality control. We present RuleAutoPilot, an end-to-end agentic framework that generates deployable Suricata rules directly from malware network traffic, with no prior threat intelligence required. A key challenge is noise: network traffic captures often contain a small amount of security-relevant traffic mixed with large volumes of background traffic, which reduces LLM reasoning quality and increases cost. RuleAutoPilot addresses this challenge with a Benign Traffic Fingerprinting stage that removes known benign background flows before LLM processing. Rules that fail syntax checks, do not trigger on the source traffic, or generate false positives on a benign corpus are automatically repaired using structured feedback. Across 1,296 malware PCAPs, execution-grounded verification raises rule quality (F1) from 0.443 to 0.539. On a stratified 200-PCAP subset, RuleAutoPilot on the open-weight gpt-oss-120b reaches near-frontier quality, 0.524 F1 against Claude Opus 5 under Claude Code's 0.623, at 52x lower billed-token cost. Swapping only the backbone to Claude Opus 5, RuleAutoPilot surpasses Claude Code outright, 0.656 F1 against 0.623, at 40x fewer tokens. A stronger backbone raises RuleAutoPilot's own ceiling, but at the same backbone, our scaffold still outperforms Claude Code's, showing the scaffold contributes independently of the backbone.