Probability
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Sunday 30 August
We prove a refined quadratic normal approximation for the first absolute moment of normalized weighted Rademacher sums with bounded maximal coefficients. For any weight vector $w\in\mathbb{R}^{n}$ satisfying $\|w\|_{2}=1$ and $\|w\|_{\infty}\leq\beta$ with sufficiently small $\beta>0$, we establish the uniform error bound $|\mathbb{E}|\sum_{i=1}^{n}w_{i}X_{i}|-\sqrt{2/\pi}|=O(\beta^{2})$ over all admissible weight configurations. Our proof combines zero-bias Stein's method and refined small-ball probability estimates to exploit symmetry cancellation and control the non-smooth residual of the absolute-value test function. An explicit extremal construction further verifies the optimality of this quadratic convergence rate. As an application, we establish an asymptotically sharp refinement of the Friedgut–Kalai–Naor (FKN) theorem for Boolean functions, also known as the sharp $\pi/2$ theorem, characterizing the level-1 Fourier energy for functions deviating far from dictatorships.
Computation and Language
A Hub of Short Rows Inflates Intrinsic Dimension Estimation of Token Embeddings
A token-embedding table holds a hub of short rows near its origin, and we show that this cluster biases what nearest-neighbor intrinsic-dimension (ID) estimators report. Because of the concentration of measure, a token is closer to the central cluster than to any other token, so its first two neighbors are both hub rows at nearly the same distance. As a result, the ID estimators such as TwoNN return a dimension far above the real ID. Measured one token at a time, dimension is a heavy-tailed distribution. Measured on the full vocabulary, it grows with the model's parameter count. However, when we remove the hub, the heavy tail disappears and the measured dimension collapses to a narrow range for eleven models, from GPT-2 to models such as K3 and GLM-4.7. The hub acts as a switch: a few hundred rows are enough to fully inflate the estimate. We reproduced an experiment stating that the intrinsic dimension (ID) of Pythia's token-embedding table grows with the parameter count, from $27$ to $122$ between 160M and 12B parameters. We show that this result disappears when the hub is removed: the table then reads $10$ to $17$ at every size. The hub contains a subset of the population that under-trained-token detectors flag, but on Pythia the hub that we detected and removed as a whole was updated during training: what seem to characterize these rows is simply their length, not an absence of updates. Finally, we show that normalizing the rows instead of removing them gives the same lower reading.
Read the full abstract
Strongly Correlated Electrons
Magnetic Field-Tunable Repulsive Exciton-Exciton Interaction in the van der Waals Antiferromagnet NiPS$_3$
Two ultra-narrow absorption peaks around 1.5 eV, which are widely believed to originate from a transition from a spin-orbital entangled triplet to a singlet state, in the two-dimensional van der Waals crystal NiPS$_3$, have attracted considerable attention because of their pronounced spin-dependent character. An interesting question is whether ultrahigh magnetic fields modify an interaction-driven hybridization between those two peaks. In this work we perform systematic magneto-optical measurements of NiPS$_3$ in pulsed magnetic fields of up to 178 T and observe a pronounced mutual repulsion between the two sharp exciton peaks accompanied by a redistribution of oscillator strength, while the band edge shows no detectable field-induced shift within our experimental resolution. We construct a minimal two-level interaction model and compare it semi-quantitatively with the experimental data. Our results reveal a magnetic-field-tunable exciton-exciton coupling as the dominant high-field response of NiPS$_3$, and clarify this material as a new experimental platform for exploring strongly correlated exciton physics in magnetic van der Waals insulators.
Read the full abstract
Atomic Physics
Cancellation of $D_2$ line transitions of alkali-metal atoms by magnetic-field values
In a previous work the $\pi$ transitions of the $D_1$ line of alkali-metal atoms were shown to cancel at magnetic-field values given by a single closed-form expression. The $D_2$ line has until now resisted the same treatment, because the fixed-$m$ Hamiltonian blocks of the $^2P_{3/2}$ manifold reach the dimension $4\times4$, and the corresponding formulas were considered to be too heavy to be useful. In this work we show that this difficulty is only apparent. Since the Zeeman interaction couples only levels with $\Delta F=\pm1$, every block, whatever its dimension, is tridiagonal; its characteristic polynomial therefore obeys a three-term recursion, and the components of its eigenvectors are polynomials in the eigenvalue. Using these two properties we obtain the eigenvalues in closed form by Ferrari's and Cardano's formulas, the eigenvectors without any further diagonalization, and finally a single relation which gives the magnetic-field value canceling a $D_2$ transition as an explicit function of the excited-state eigenvalue, the nuclear spin $I$, the ground-state hyperfine splitting and the magnetic quantum number. A necessary condition on $m$ and on the polarization is derived, which replaces the selection rule known for the $D_1$ line. All the magnetic-field values canceling $\pi$, $\sigma^+$ and $\sigma^-$ transitions of $^{23}$Na, $^{39}$K, $^{40}$K, $^{41}$K, $^{85}$Rb, $^{87}$Rb and $^{133}$Cs are calculated up to 20~kG and given with their uncertainties; there are 234 of them. Contrary to the $D_1$ line, the $\sigma^{\pm}$ transitions of the $D_2$ line do cancel, and they account for the majority of the values found. The accuracy of the calculated $B$ values is limited only by the uncertainties of the excited-state hyperfine constants.
Read the full abstract
Statistics Applications
Text Data Analysis and Classification Methods - Insights from Customer Letters in Life Insurance
The business of life insurance companies is characterized by long-term contracts. For this reason, data describing customers is of immense value. A portion of the data provided to the customer is rarely or not at all analyzed. This includes customer letters of any kind. This work focuses on classifying customer letters as cancellations and identifying the respective reason, if available. The outlined approach can also be applied to other business transactions and reasons. We discuss data acquisition and preparation, present alternatives, and explain the reasons for the chosen approach. A successful implementation of such a tool can lead to a better understanding of customer cancellation behavior by the insurer, enabling more targeted actions in certain situations.
Read the full abstract
Earth and Planetary Astrophysics
Gemini and Apache Point Multi-band Optical Imaging Characterization of Fragmenting Long-period Comet C/2025 K1 (ATLAS)
We present results from multi-band g, r, and i, observations of C/2025 K1 (ATLAS) taken with the Gemini North 8.1-m/GMOS imager on 2025 December 6 and December 24, and u, g, r, i, and z observations with the Astrophysical Research Consortium (ARC) 3.5-m/ARCTIC imager on 2025 December 8. We identify at least four distinct fragments in the Gemini and ARC images, designated as A, C, D, and E in these data taken between 2025 December 6 and 24. Color indices are determined from the December 8 ARC observations of fragments A and C, and from the Gemini observations on December 24 for A, C, and D. K1 has an unusually blue g-r color of $\sim$0.40. The color difference between the comet and its fragments at the two epochs may be explained by particle size and light-scattering effects. We used the Gemini observations to calculate dust mass-loss rates for fragments A, C, and D. We conclude that C/2025 K1 has moderate dust mass-loss rates for millimeter-sized dust of $\sim$50 kg/s for the A and C fragments.
Read the full abstract
High Energy Physics (Phenomenology)
$(g-2)_{\ell}$ and LFV decays in a $U(1)_{L_{\mu} -L_{\tau}}$ left-right model with inverse seesaw neutrinos
In the framework of a $U(1)_{L_{\mu} -L_{\tau}}$ left-right model with inverse seesaw neutrinos recently proposed, we show that one-loop contributions from heavy singly charged Higgs bosons to the anomalous magnetic moments of charged leptons $(g-2)_{e,\mu}$ can accommodate the observed discrepancies, of $10^{-9}$ and $\mathcal{O}(10^{-13})$ for the muon and electron, respectively. Meanwhile, all lepton-flavor violating decay rates for $h\to e_b^\pm e_a^\mp $, $Z\to e_b^\pm e_a^\mp$, and $e_b\to e_a \gamma$ are found to be highly suppressed as a consequence of the gauged $U(1)_{L_{\mu}-L_\tau}$ symmetry.
Read the full abstract
Artificial Intelligence
Ideation Arena: Evaluating LLM Generated Research Ideas with Battle-style Human Expert Assessment
Evaluating research ideas generated by LLMs is difficult because their scientific value cannot be fully determined by objective criteria, and no single reference answer specifies what counts as a good idea. To address this challenge, we introduce Ideation Arena, a battle style platform that evaluates research ideas through pairwise human assessment. Ideation Arena evaluates ideas generated by 14 frontier LLMs and 5 research agent architectures built on 2 base models. To ensure a common starting point, Ideation Arena builds shared literature contexts from papers familiar to the participating researchers and provides the same contexts to all LLMs and agents. We collect over 6,000 double blind pairwise comparisons from 105 active computer science researchers and construct an Elo rating leaderboard of proposal-stage expert preferences in computer science under a shared closed-context protocol. We validate the rankings through interrater agreement and robustness analyses, showing that the leaderboard remains stable under changes in annotator composition and domain coverage. Our results show substantial variation in agent effectiveness, with some frameworks improving ideation quality over their backbones and others offering little benefit or even underperforming their base models. We further construct Ideation Arena Eval, a benchmark for assessing whether automated evaluators align with human preferences in research ideation. Experiments with current LLM judges show that they still cannot reliably reproduce expert preferences, with the best judge reaching 72.56% Soft Accuracy on Overall Quality. Our code, data, and leaderboards are available at https://github.com/foss12138/Research-Ideation-Arena.
Read the full abstract
Differential Geometry
Sasaki-Einstein metrics on connected sums of $S^3\times S^4$
We completely characterize the $n$-fold connected sums of $S^3\times S^4$ that admit Sasaki-Einstein structures by proving that such a structure exists if and only if $n$ is even.
Read the full abstract
Dynamical Systems
First integrals of dense hard-ball gases
We answer two questions of Kozlov about polynomial-in-momentum first integrals of billiard systems. For billiards in bounded polytopes, such integrals are independent of position. For hard-ball gases in a rectangular box, every polynomial-in-momentum first integral is a polynomial in the total kinetic energy whenever the collision graph is connected.
Read the full abstract
Optimization and Control
Successive design of backstepping observers for parabolic PDE-ODE systems and its duality to state feedback stabilization
The paper introduces a successive backstepping observer design for strictly feedforward parabolic PDE-ODE systems, in which the coupling structure determines the order of error stabilization and the corresponding transformations. First, a transformation based on a virtual measurement stabilizes the ODE observer error subsystem, which is most distal from the measurement, while decoupling it from the PDE error state. Second, a Volterra integral transformation is employed to stabilize the PDE error subsystem and to map the overall error dynamics into a cascade of exponentially stable ODE and PDE subsystems. The design is shown to be dual to a recently proposed multi-step state feedback design for parabolic PDE-ODE systems in strict feedback form, thus explaining the structure of the presented observer design.
Read the full abstract
Statistical Finance
Portfolio Risk Bounds without Cross-Asset Return Covariances: Distributional Fields from Language-Model Representations
Portfolio risk assessment ordinarily relies on reliable estimates of cross-asset return covariances, which are difficult to obtain in short, high-dimensional panels. We show that firm-level distribution-valued characteristics can instead provide one-sided certificates of portfolio risk. Under maintained links from characteristics to systematic exposures and from exposures to returns, multi-firm Wasserstein-2 dispersion yields a sharp upper bound on systematic portfolio variance and a corresponding bound for standardized returns. A weighted pairwise relaxation produces an objective that is convex under a checkable condition and requires marginal volatility scales but no cross-asset return covariances. With zero firm-specific slack, the common-map scale changes the certified variance reduction but not the normalized allocation, which depends only on observed information geometry. In a 100-firm panel from 2018-2022, only 0.09%-0.46% of portfolios across four prespecified capped reference populations have in-sample variance no greater than an allocation constructed from Qwen3-Embedding-8B news representations; the corresponding share for equal risk weighting is 28.3%-34.7%. The lower in-sample variance ranking relative to equal risk also appears across the reported frozen language-model representations. The framework therefore converts distribution-valued firm information into a coherent risk bound and an implementable allocation rule constructed without cross-asset return covariances.
Read the full abstract
Superconductivity
Competition between on-site and off-site pairing phases and surface spectra in superconducting wallpaper fermion systems
We theoretically investigate the competition between on-site and off-site pairings in superconducting topological crystalline insulators hosting wallpaper fermions, a class of surface states protected by nonsymmorphic symmetries. In-plane nearest-neighbor pair hopping stabilizes the $\mathrm{A_{1u}}$ phase, while off-site pairing channels yield additional dominant phases with $\mathrm{B_{2u}}$ and $\mathrm{E_u}$ representations. We further show that the $\mathrm{B_{2u}}$ pairing state supports the hybridization between wallpaper-fermion spectra themselves, whereas a representative time-reversal-invariant nematic $\mathrm{E_u}$ state exhibits spectral hybridization between wallpaper fermion and double Majorana Kramers pairs, yielding twisted surface energy spectra. These results demonstrate that the twisted dispersions arising from hybridization phenomena persist in the presence of off-site pairing interactions.
Read the full abstract
Optimization and Control
Transformer-Based Flow Shop Scheduling Using MILP-Generated Training Data
Advances in machine learning (ML) have created new opportunities to complement traditional operations research (OR) methods. In particular, transformer models can capture complex interactions in token sequences by mapping tokens into a high-dimensional embedding space and propagating contextual information via attention. This makes them a candidate to model non-permutation flow shop scheduling with secondary resources as a next-token prediction task, where tokens represent job-machine-secondary resource tuples. For training, mixed-integer linear programming (MILP)-generated schedules are tokenized and used as next-token prediction data. During inference, partial token sequences (prefixes) are randomly generated and completed by the trained transformer through constrained decoding. A computational study is conducted on a flow shop with 8 jobs, 4 machines, and 3 secondary resources, where jobs are selected from a fixed pool of 20 jobs that is sampled during training and provides the candidates during prefix completion. The transformer achieves better solution quality (smaller makespans) compared to a genetic algorithm (GA), the NEH heuristic, and random search. It is outperformed only by the MILP model and the iterated greedy (IG) heuristic. The study concludes that transformer models can, to some extent, learn patterns from MILP-optimized non-permutation flow shop schedules and that transformer-based scheduling represents an interesting direction for future research, particularly in settings with a fixed, recurring job set.
Read the full abstract
Quantum Physics
Imaging Stars at the Quantum Compatibility Limit
Imaging astrophysical sources with a multi-station interferometer is intrinsically a multiparameter quantum-estimation problem. {Using tools from multiparameter quantum metrology,} we show that time-resolved repetitive or adaptive measurements in an \(N\)-station array suffer a fundamental array-level incompatibility among visibility estimators. Collective measurements, {which coherently process the received starlight across multiple time bins in a single joint readout}, remove the array-size penalty up to an order-unity factor, yielding an asymptotic \(O(\sqrt{N})\) enhancement for the {directional-averaged} SNR of visibility measurement. We then propose a memory-assisted interferometric architecture designed to implement collective readout through coherent storage and joint quantum processing. Imaging simulations and Fisher-information analyses demonstrate that collective measurements improve image reconstruction in near-term arrays and enhance the resolving power of future long-baseline architectures, with pronounced benefits for representative AGN targets such as NGC~4151 and 3C~273. These results highlight collective measurement as a promising building block for future quantum-assisted interferometric arrays for stellar imaging.
Read the full abstract
High Energy Physics (Theory)
One-Loop Fluctuation Response Along a Constrained Noncommutative Modulus in the Lorentzian IIB Matrix Model
We study a local constrained problem at matrix size $N=5$ in the Lorentzian IIB (IKKT) matrix model and numerically continue a one-dimensional physical modulus. For an exact fixed-$Q_\rho$ constrained solution family, the stationary identity implies constant classical action; the 23 accepted backgrounds realize this relation to the quoted precision. Across the sampled branch, the spectrum of the Lorentzian extent operator remains common to numerical precision while microscopic noncommutative invariants vary. We evaluate finite-dimensional one-loop fluctuations along this branch. At fixed bosonic background, the fermionic Grassmann integral is exact and gives $\operatorname{Pf}\mathcal M_F$; its magnitude is invariant under color-gauge and connected proper-Lorentz transformations and varies along the modulus. The complete boson–ghost–fermion quotient measure defines a Lorentz-representative-independent density one-form rather than a scalar potential. Thus quantum fluctuation data distinguish backgrounds unresolved by the coarse quadratic spectrum. We also apply a framed higher-$L$ SO(5) structural probe. At the reference background, the ordered $3+6$ spectral separation is larger at $L=2,3$ than at $L=1$, and larger still in the canonical large-$L$ limit. This is a frame-defined diagnostic, not an intrinsic observable or a higher-$N$ solution. The Pfaffian and higher-$L$ responses are closely correlated along the sampled branch, which we treat only as descriptive. These results suggest that microscopic noncommutative structure and quantum fluctuations may contain information relevant to dimensional dynamics, without establishing dimensional selection.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Spheroidal Resampling Analysis of High-Redshift Gamma-Ray Burst Spatial Densities
Gamma-ray bursts (GRBs) are bright transient sources that can be observed at high redshift. They can therefore be used as tracers of the distant large-scale structure, although the GRB redshift sample is sparse and affected by strong selection effects. We analyze the three-dimensional distribution of 542 GRBs with spectroscopic redshifts. The method extends our earlier spherical-window search by replacing spherical counting volumes with axisymmetric spheroids. The angular positions of the observed GRBs are kept fixed in the Monte-Carlo null samples, while the redshifts are shuffled within each Galactic hemisphere. In the Northern Galactic hemisphere, the overdensity associated with the Hercules–Corona Borealis Great Wall remains significant for a range of spheroidal shapes. This supports the stability of the previously reported signal and shows that it is not only a consequence of using spherical counting volumes. In the Southern Galactic hemisphere, the same method finds no structure with comparable significance. A small candidate grouping is present, but its significance is only marginal, and it is sensitive to the small number of events. We conclude that spheroidal resampling is a useful check of GRB overdensity searches, but the physical nature of any candidate structure still requires confirmation with independent tracers such as galaxy or quasar samples.
Read the full abstract
High Energy Physics (Theory)
Latent Geometry and the Emergence of Lorentzian Time in Matrix Quantum Mechanics
We introduce latent geometry in BFSS/BMN models. Before an emergent geometry evaporates at the geometric transition, large-$d$ Gaussianization encodes its information in the exceptional $N=2$ sector selected by the baby fuzzy sphere, a single active spin-$1/2$ Pauli block. Dimensional reduction, finite-$N$ Gaussianization and exact Molien–Weyl projection transport this encoding to mass-deformed BFSS$_2$. At $N=2$, the same stable BFSS$_2$ oscillator admits inequivalent unitary $\mathbf{H}^2_\theta=\mathbf{EAdS}^2_\theta$ and $\mathbf{dS}^2_\theta$ decodings, realizing spacetime transmutation and the emergence of Lorentzian time.
Read the full abstract
Machine Learning
Last Step Matters: Early Uncertainty Cannot Predict Failure in Long-Horizon Agents
Early failure prediction is important for long-horizon agents, as it enables timely intervention and can reduce inference and tool-use costs. Uncertainty quantification, such as verbal confidence and perplexity, offers a promising approach to detecting agent failures; however, it has not been explored whether these signals retain their discriminative power during the intermediate stages of long-horizon execution. We evaluate mainstream uncertainty signals on deep-research tasks and find that verbal confidence reliably distinguishes failures at trajectory completion, achieving a mean AUROC of 0.85, whereas all evaluated signals offer limited predictive value earlier in execution, with none exceeding a mean AUROC of 0.60 at 50% trajectory progress. We identify an underlying mechanism explaining this gap: path switching, where agents frequently abandon their current search direction in-trajectory, breaking the link between early signal and final outcome. These findings challenge the assumption that intermediate uncertainty can reliably guide early intervention. They also motivate a practical recommendation for agent harnesses in deep-research settings: use final-step confidence to decide whether to restart, an approach that our experiments find more effective than in-trajectory intervention.
Read the full abstract
Quantum Physics
Analytical diagonalization of the open-boundary bosonic Kitaev chain: An asymmetric plane-wave ansatz approach
The bosonic Kitaev chain under open boundary conditions has attracted recent attention due to its realization in driven-dissipative systems and its intriguing non-Hermitian boundary physics. The model is known to be solvable via local squeezing transformations in the position-momentum representation. In this paper, we present an alternative, purely algebraic solution that relies entirely on the standard bosonic Bogoliubov transformation. For an $N$-site chain, we propose an asymmetric plane-wave ansatz with unequal left- and right-moving momenta to analytically solve the associated $2N\times 2N$ non-Hermitian “associated matrix". The left eigenvalue problem yields $N$ distinct eigenvalues, each of which is twofold degenerate. By carefully resolving these degeneracies using the bosonic commutation relations, we construct the $N$ physical Bogoliubov quasiparticle operators. The construction reveals non-uniquenesses that in special cases exactly mirror the freedom in the local squeezing transformations of the original approach. The diagonal form of the Hamiltonian is obtained explicitly and is shown to be equivalent to the original Hamiltonian. The proposed asymmetric plane-wave ansatz and degeneracy-resolution technique are not limited to the present model and can be generalized to other bosonic pairing systems, including those with inhomogeneous pairing or hopping terms.
Read the full abstract
High Energy Physics (Phenomenology)
On-shell renormalization in NREFT: a simplified description of resonances, bound states, and coupled channel effects
We discuss several issues regarding the modern generalization of Weinberg's compositeness criterion. In particular, we demonstrate that the on-shell renormalization in NREFT reproduces Weinberg's result and provides a simple description of resonances, bound states, and coupled channel effects. Additionally, we present the complete propagator, including the charged channel for the near-threshold p-wave states.
Read the full abstract
Astrophysics of Galaxies
The multi-frequency radio light curve of the sub-pc SMBH binary candidate PG 1302-102
Supermassive black hole binary (SMBHB) candidates with sub-parsec (sub-pc) orbital separations are rare, high-value systems, thought to be the dominant sources of the recently detected nanoHertz (nHz) gravitational wave (GW) background. Alongside pulsar-timing-array searches for localized nHz GW sources from SMBHBs, an increasing number of candidates have been identified from synoptic surveys of large quasar optical light curves over increasingly long baselines. The optical light curve of the quasar PG~1302-102 at a redshift of $z=0.278$ exhibits a pronounced, smoothly varying periodic signal of ~5.2 years over approximately 20 years. Although the cause of this periodicity remains debated, a plausible explanation is a binary system of two SMBHs with a sub-pc separation. At this close distance, the $10^{8.3-9.4} M_{\odot}$ nuclear black holes in PG 1302-102 are expected to merge within $\sim$$10^{5}$ yr solely due to GW emission. We present a multi-frequency radio light curve analysis of PG~1302-102 spanning a $\sim$19-year baseline, with observations between 2005 and 2024 drawn from the Australian Telescope Compact Array (ATCA). The ATCA radio light curve is highly variable at 5.5~GHz but shows no significant evidence of the periodicity seen at optical wavelengths. However, we report fractionally stronger evidence of periodicity in the in-band spectral index time-series at 5.5~GHz, with a period consistent with the optical periodicity. Furthermore, the recently reported change in optical periodicity is also present in the radio in-band spectral index light curve. These two results suggest that radio and optical periodicity may be physically coupled despite very different emission mechanisms. We argue that this radio-optical coupling reinforces the SMBHB interpretation, despite the optical periodicity change, and provides an intriguing case study ahead of the LSST-SKAO era.
Read the full abstract
Computers and Society
MMMMM: A Unified Taxonomy for Investigating the Mechanisms of Multilingual MultiModal Misinformation
Multimodal misinformation on social media is highly prevalent, potent, and harmful, yet difficult to detect and counter, and still poorly understood compared to its text-only counterpart. Research on the properties and deceptive strategies of multimodal misinformation is hindered by a lack of taxonomies grounded in real-world contexts and by the limitations of current multimodal machine learning models, which prevent the automation of annotation and analysis at scale. We address these shortcomings in three steps. First, we collect a large-scale, high-quality dataset of real-world misinformation instances from Twitter/X in seven languages. Second, we develop a novel, comprehensive taxonomy of multimodal misinformation grounded in an in-depth qualitative analysis of the data and prior theoretical work. Finally, we operationalise the taxonomy through an automated multi-step annotation pipeline using a Vision-Language Model (VLM), and perform human-validation. Our novel approach leads to previously undocumented insights about how social media users combine images with text to spread misinformation in the wild, e.g., that AI-generated content is particularly prevalent in technology and science, while vaccination misinformation disproportionately utilises images from news outlets to assert credibility. Our method and findings provide guidance for targeted approaches for detecting multimodal misinformation, and suggest that mitigation efforts should be developed and applied strategically rather than uniformly.
Read the full abstract
Computer Vision
GeoRay: Gauge-Aware Feed-Forward Satellite 3D Reconstruction in the Geodetic Frame
Feed-forward 3D foundation models reconstruct perspective scenes in one pass. Satellite photogrammetry needs a different product, one that domain adaptation alone does not deliver: dense surface height in an absolute geodetic frame under non-central rational polynomial cameras (RPCs). Perspective-pretrained features are not reliably observable along RPC height rays, absolute elevation carries a low-order height–datum gauge exchangeable with sensor bias to first order, and monocular and multi-view cues fail in different regions. \method{} treats all three. Lightweight ray-consistent adapters make a frozen backbone matchable along native RPC rays. An explicit datum mechanism separates relief from absolute level and is equivariant to the vertical origin by construction, so one trained model serves zero-, one-, and sparse-control inference. Calibrated inverse-variance fusion combines the two relief streams. \bench{}, our absolute-frame benchmark of eighteen systems across in-domain, cross-dataset, and cross-city tiers, scores absolute placement without registration or test-reference leakage. On 26 held-out US3D tiles, \method{} attains $2.99$\,m absolute MAE at $91.9\%$ coverage, improves completeness-aware accuracy by $46.4$ points over the strongest compliant feed-forward baseline, remains the most accurate such system under both transfer shifts, and runs in $24$\,s model-forward time per tile. Code and models will be released at https://github.com/HIT-SIRS/GeoRay
Read the full abstract
Subcellular Processes
Clustering versus sorting: a mass-conserving reaction-diffusion model of planar polarity puncta
Planar cell polarity is preceded by the clustering of polarity proteins into discrete, low-turnover membrane subdomains (puncta), yet the minimal interactions that nucleate puncta, set their number, and segregate opposite orientations remain unclear. We address these questions with a mass-conserving reaction-diffusion model in which two diffusible monomers bind reversibly across a cell-cell junction into trans-complexes of two orientations, with feedback entering only through concentration-dependent rates. Above a critical density, the uniform state undergoes a long-wavelength mass-redistribution instability rather than a finite-wavelength Turing bifurcation. The form of the feedback then selects between two morphologies: broad mesas fixed by a Maxwell construction under saturating feedback, and narrow mass-limited spikes under unbounded feedback. For spikes we obtain closed-form expressions exhibiting a clean separation of amplitude (mass and feedback), width (the complex diffusion length), and spacing (the monomer screening length). Within a fast-monomer reduction we prove, for any number and arrangement of puncta, that like-oriented arrays coarsen, so multiplicity is metastable and kinetically determined. The second monomer reservoir introduces a second screening length that rate-limits competition by the harmonic mean of the monomer diffusivities, and the spectrum of a punctum remains free of oscillatory ("blinking") instabilities throughout. Finally, sign-definite cross-modulation of turnover converts clustering into orientation sorting - the mutual exclusion of orientations along a single contact: mass redistribution sets puncta number, and the sign of the cross-coupling determines whether orientations segregate. Puncta number and orientation sorting are thus governed by mathematically separable ingredients.
Read the full abstract
Databases
Diachronic Hypergraphs for Orchestrated Multi-Agent Multimodal Memory Curation
Multi-agent systems solve tasks through collaboration, tool use, multimodal reasoning, and orchestration, but each agent operates within a knowledge boundary defined by its observations, context, and resources. Memory must preserve and transfer evidence, role specific context, decisions, procedures, and experience across interactions, not only outcomes. Vector and graph memories flatten these structures into embeddings or dyadic traces, obscuring events involving agents, tools, documents, errors, and evidence. This limits knowledge sharing, tracing, reuse, revision, and orchestration. We present MAGE, a hypergraph based multimodal database designed as a memory engine for MAS. MAGE stores agents, messages, tools, errors, procedures, documents, entities, decisions, and evidence in a heterogeneous temporal hypergraph, preserving high order collaborative events as reusable memory. It supports decision driven updates, role aware retrieval, validation, lifecycle management, and budget bounded context packing. By delivering knowledge to agents and orchestrators, MAGE expands their knowledge boundaries without modifying the models. Experiments show MAGE outperforms on various memory baselines.
Read the full abstract
Computer Vision
MedSegBenchmarker: A Raw-Count-First Framework for Controlled 2D Medical Image Segmentation Benchmarks
Despite rapid advances in MIS, fair and reproducible comparisons of segmentation models remain challenging due to heterogeneous datasets, inconsistent evaluation protocols, and rapidly evolving architectures. In particular, comparisons often implicitly assume that model rankings are invariant to data partitioning, preprocessing, metric aggregation, uncertainty estimation, and computational constraints. The lack of extensible and unified evaluation frameworks further limits systematic investigation of new models, datasets, and training paradigms. We present MEDSEGBENCHMARKER (MSB), a configuration-driven framework for controlled benchmarking of 2D MIS. It integrates duplicate and near-duplicate image detection, group-aware data splitting, YAML study specifications, resumable training, hyperparameter optimization, cross-validation, and checkpoint-based evaluation. Rather than retaining only aggregate performance measures, MSB exports sample- and class-level pixel counts and predictions together with the evaluation context. These elementary artifacts enable post-hoc analyses without repeated inference. We demonstrate MSB in a case study involving three heterogeneous 2D datasets and multiple MIS and general-purpose vision models evaluated at 256- and 512-pixel input resolutions. Reaggregation of identical predictions changes the top-ranked architecture in three of six dataset-resolution settings, despite high rank correlations between aggregation strategies. Increasing input resolution produces model- and dataset-dependent performance gains and losses that must be considered alongside empirically measured inference complexity. These results show that seemingly minor choices in evaluation and experimental setup can affect benchmark conclusions. MSB, available at GitHub, provides a practical and extensible basis for making benchmark conditions and evaluation choices explicit and reproducible.
Read the full abstract
Soft Condensed Matter
From Freezing to Terminal Packing: A Puzzle and a Paradox
A density-functional theory developed for equilibrium freezing also produces a high-density endpoint close to terminal packing. We revisit the nonlinear Ramakrishnan–Yussouff bifurcation framework and distinguish the first crystalline fold, thermodynamic coexistence, and terminal marginality. At the terminal point, the grand-potential curvature vanishes along the principal finite-wavevector density mode, exactly when the nonlinear self-consistency equation loses stiffness. The construction gives exact close packing for one-dimensional hard rods, a terminal hard-disk packing fraction near 0.8407, and a hard-sphere value near 0.629, close to random close packing. These results identify a finite-wavevector thermodynamic marginality of a metastable liquid continuation, rather than equilibrium freezing or mechanical jamming, and expose a persistent puzzle: why should liquid-state correlations encode the approach to terminal disordered packing across different physical dimensions and systems?
Read the full abstract
Software Engineering
Cost-Effective Repository Exploration for Agentic Issue Localization
Repository exploration is a distinct and costly stage of coding-agent pipelines: before generating a patch, an agent must identify which repository files are likely to matter. We study whether this stage can be delegated to lower-cost models while retaining useful localization quality. Using our IssueLoc-Bench, we evaluate five explorer models under the same read-only interactive interface on 499 SWE-bench Verified-derived tasks and 500 tasks from 153 additional repositories. We measure early candidate discovery, top-three gold-file coverage, strict file-set recovery, agent time, and token usage, with paired instance-level uncertainty and repository-clustered sensitivity analysis. The highest-quality explorer leads across localization metrics, but substantially cheaper operating points emerge: depending on the model and evaluation arm, lower-cost explorers retain approximately 78-94% of the reference Hit@3 and 73-92% of its F1 while reducing mean agent time by 41-88% and token usage by 84-95%. The preferred operating point depends on how localization is consumed downstream: ranking and coverage metrics characterize recoverable candidate handoffs, whereas F1 and exact match characterize restrictive file gates. These results support treating repository exploration as an independently measurable and budgetable stage of modular coding agents, with explorer selection guided by the downstream handoff contract.
Read the full abstract
Machine Learning
Creation begins with understanding: LLMs as strategy designers for privacy-preserving tabular data synthesis
Sharing tabular data in high-stakes domains is constrained by privacy regulations. Synthetic data offer a promising alternative, but deep generative models are costly to train and difficult to audit, while LLM-based methods often serialize records as text, obscuring tabular structure and exposing sensitive data. We introduce Tabular Synthesis Strategy Designer (TabSSD), which uses an LLM to design synthesis procedures rather than directly generate records. TabSSD provides the LLM with tree-derived summaries of variable dependence rather than raw records, which produces Python programs for local execution and evaluation. Across twelve datasets, TabSSD strikes a favourable balance among statistical fidelity, predictive utility, and empirical privacy risk, achieving the best average rank across six metrics among ten methods. Moreover, it substantially reduces local computation and token consumption relative to the compared methods. By enabling human-guided refinement and eliminating user-side model tuning, TabSSD lowers the expertise and infrastructure barriers to transparent tabular data synthesis.