Quantitative Methods
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 5 November
Traditional coral taxonomy is increasingly challenged by recent advances in molecular evidence. Quantitative methods previously used to identify morphological descriptors have not been successful at detecting significant differences at the macroscale. It is of great interest to optimise methods for effectively delimiting species within the complex genus Acropora to ensure sustainable, long-term management of these vital ecosystems. Topological Data Analysis (TDA) is designed to study the shape of data and can detect features that familiar methods overlook. This novel approach makes TDA and persistent homology a logical choice for studying coral morphology in a taxonomic context. The results show that valid topological information can be extracted from in-situ coral images and that there is great potential in further refining this method. This study is a first exploration into the potential of persistent homology in supporting the delimitation of the reef-building scleractinian corals Acropora.
Machine Learning
Evolutionary Adaptive Metric Learning: A Multi-Objective Framework for Learning Local Distance Functions in Nearest Neighbor Classification
Nearest neighbour classifiers inherit the geometry of their distance function. Fixed distances and globally learned metrics apply the same geometry throughout the input space, although relevant feature directions may vary across regions. We propose EAML (Evolutionary Adaptive Metric Learning), a framework that learns instance-dependent positive-definite metrics through a low-dimensional parameterization. Each local metric combines an input-dependent diagonal scaling, generated by softmax interpolation between anchor points, with a global low-rank correlation component. A real-coded elitist genetic algorithm optimizes the parameters using a scalar objective combining leave-one-out prediction error, robustness, metric smoothness, and regularization. We establish positive definiteness, metric equivalence, existence of an optimal solution, and Lipschitz continuity of the metric field. We further investigate prediction stability, generalization bounds, and evolutionary convergence under explicit assumptions. Experiments on synthetic data with region-dependent feature relevance show that EAML can recover local metric variations and outperform global metrics. Evaluation on thirteen public datasets using nested 5-fold cross-validation reveals a more nuanced picture: EAML improves performance on selected datasets but does not significantly outperform global metric baselines overall. Additional experiments show that global diagonal metrics can be preferable under nuisance features and test-time noise. These findings identify the conditions under which evolutionary local metric learning is beneficial and highlight its practical limitations.
Read the full abstract
Populations and Evolution
Genealogical Reinforcement in Branching Processes
Within many ecological populations, survival and reproduction exhibit a dependence on social interactions with relatives. As one example, resource excess is often shared according to genealogical proximity. When viewed as a branching system, an individual's offspring distribution acquires an effective dependence on genealogical structure. In the specific case that a large, local family structure leads to a large set of offspring, this can be viewed as a form of genealogical reinforcement. To capture this mechanism, we introduce a branching model of finite-depth genealogical dependence. Given partial observations of this reproducing population and its genealogy, we would like to reject the hypothesis of independent branching and estimate the strength of dependence. We present a statistical framework for detection and estimation, combining analytical results with simulation-based validation. We then identify our branching model with a tree-valued multitype branching process on a finite type space, providing a rigorous characterization of the expected structural reinforcement. Finally, we discuss how these tree-valued interactions can be accommodated within a broader class of stochastic processes.
Read the full abstract
Combinatorics
Upper bound on the size of subgraphs avoiding cycles of girth length in graphs
A graph $G$ is said to be $H$-free if it contains no copy of $H$. The generalized Turán number $\mathrm{ex}(G, H)$, a natural extension of the classical Turán number, is defined as the maximum number of edges in an $H$-free spanning subgraph of $G$. Although $\mathrm{ex}(G, H)$ has been extensively studied for various specific graphs $G$, such as hypercubes, complete transposition graphs, and doubled Johnson graphs, very few results are known for general graphs. For a graph $G$ of girth $g$, let $C_{g}$ denote the cycle of length $g$. In this paper, we investigate the generalized Turán number $\mathrm{ex}(G, C_{g})$ for graphs $G$ in which every edge is contained in a fixed number of $g$-cycles, and establish a general upper bound on it. This bound generalizes the classical results on hypercubes and provides a unified approach for general graphs. Furthermore, we apply this bound to derive explicit upper bounds for several specific graphs, including $K_{n,n,\ldots,n}$, $K_{n}\times K_{n}$, $C_{m}\times C_{n}$, $S_{n}$, $AG_{n}$, $CT_{n}$ and $J(n;k,k+1)$.
Read the full abstract
Data Structures and Algorithms
Bourgain's Theorem for a Union of Euclidean Metric Spaces
We study the distortion of embedding a union of $k$ Euclidean sets into Euclidean space. We show that a union of $k$ Euclidean sets can be embedded with distortion at most $O(k^{1.128} \log k)$, independent of the number of points. This is an exponential improvement over the previous best known upper bound of $2^{O(k)}$ obtained by iterating the result of Makarychev and Makarychev (2016) for the special case $k = 2$. Surprisingly, with the additional assumption that each Euclidean set remains Euclidean when we adjoin any other two points, we can obtain an asymptotically tight distortion of $O(\log k)$, thus, providing a generalization of Bourgain's Theorem (1985). Along the way, we also show that the distortion of the union can be upper bounded by $O(\log k)$ times the maximum of the Lipschitz extension constants of individual Euclidean sets.
Read the full abstract
Classical Physics
Revisiting Routh's Classification of the Heavy Symmetric Top: A Correction
The classification of the motion of a heavy symmetric top when conserved angular momenta are different is studied by considering modern method and Routh's geometrical method. It is shown that Routh's method provides correct classification for four cases; however, it fails in two cases.
Read the full abstract
Image and Video Processing
GenStrain: Multi-Conditional Generative Diffusion Model To Synthesize High-quality Motion and Strain Maps From Cardiac Magnetic Resonance Images
This paper presents GenStrain, a novel method for synthesizing high-quality cardiac motion-based strain maps from routinely collected cine cardiac magnetic resonance (CMR) images. Existing approaches typically rely on image registration to estimate motion (a sequence of displacements) from cine CMR images, but cine CMR presents very sparse trackable intramyocardial features, presenting challenges to estimate intramyocardial motion. An advanced imaging technique, Displacement Encoding with Stimulated Echoes (DENSE) CMR, is considered as the gold standard for strain assessment, but it requires additional acquisitions, which are not ideal for efficient clinical workflows. To address this gap, we introduce GenStrain, a novel framework designed to learn the data distribution of highly accurate cardiac motion captured by DENSE. In particular, we develop a diffusion process in motion space, where spatiotemporal continuity is carefully considered. To fully leverage both cine CMR images and their associated motion patterns, we introduce a multi-conditional conditioning strategy that integrates anatomical features and motion representations learned by a spatiotemporal registration model to guide the generation process. Experimental results demonstrate that our proposed model, GenStrain, outperforms existing strain analysis methods based on displacement estimates from both traditional registration approaches and state-of-the-art (SOTA) deep learning models. Our model represents a promising step toward more accurate cardiac function assessment and holds great potential for clinical translation. Our code is publicly available at https://github.com/Rishov-MIA/GenStrain.
Read the full abstract
Hardware Architecture
SAFTAC: Simulation-Augmented Fine-Tuning of Open-Source LLMs for Analog Circuit Design
Circuit design for analog integrated circuits (ICs) is challenging because circuit topology, device sizing, and circuit performance are tightly coupled. Existing large language model (LLM)-based methods either rely on inference-time correction around frozen models or train on limited circuit types without directly using simulation outcomes as training signals. A key bottleneck is that existing datasets do not provide sufficient task information for training LLMs on end-to-end specification-conditioned design. To fill this gap, we first construct a large-scale simulation-grounded dataset with 8,626 design tasks across 10 different types of analog circuits, where each task includes a textual description, labeled I/O ports, loading condition, target specifications, a testbench, and a reference netlist. Using this dataset, we present SAFTAC, a simulation-augmented fine-tuning framework that trains LLMs to generate and revise analog circuit design. SAFTAC combines three-stage supervised fine-tuning, which progressively builds the capabilities required for analog circuit design and feedback-guided correction, with simulation feedback-augmented reinforcement fine-tuning, which further optimizes models using simulation-grounded rewards. We evaluate SAFTAC under single-pass generation and two-pass generation with simulation feedback, showing improvements over the strongest closed-source LLM by 15.9% and 7.8%, respectively.
Read the full abstract
Image and Video Processing
RapidMitDet: Boosting Computational Efficiency for Mitosis Detection via Magnification Alignment
Automated mitosis detection is a critical research focus, enabling pathologists to rapidly assess mitotic indices for tumor malignancy evaluation. However, most of the existing detection approaches suffer from significant computational inefficiency due to the giga-pixel image resolution of Whole-Slide Images (WSIs), which greatly hinders their clinical applicability. In this study, we propose RapidMitDet, a computational efficient mitosis detection approach. Our method employs a coarse-to-fine framework, first identifying mitotic candidates at lower magnification and then confirming them at higher magnification. In the coarse-detection stage, we introduce magnification-aligned knowledge distillation to reduce the information gap between magnification levels while minimizing the performance degradation in the lower magnification detection model. In the subsequent fine-classification stage, we propose adaptive triplet learning (ATL) to optimize infomative sample selection, thereby enhancing discrimination between mitotic and non-mitotic instances. Experimental results show that RapidMitDet achieves the optimal accuracy-efficiency balance at 20x magnification, improving computational speed by up to 5.62 times over the latest state-of-the-art (SOTA) methods. Due to its lightweight model design, RapidMitDet can also deliver efficient detection at higher magnification (40x) while achieving even higher detection performance. In WSI-level evaluations, our method shows strong agreement with pathologists' diagnoses (ICCs: 0.879 (40x), 0.822 (20x)) while maintaining superior efficiency, highlighting its potential for clinical application. Code is available at https://github.com/logincat/RapidMitDet
Read the full abstract
Machine Learning (Statistics)
ELPAC: Endpoint-Anchored Latent Progression with Stage-Varying Multimodal Coordination
Aging cohorts increasingly combine imaging, plasma, and cognitive measurements, but most subjects provide only a partially observed cross-sectional snapshot, leaving latent stage, progression subtype, and stage-varying multimodal coordination to be inferred jointly. Existing methods estimate the cohort-level aging process, model covariance under observed covariates, or fuse modalities for prediction. However, they rarely target the joint problem of endpoint-anchored staging and progression-resolved residual coordination across modalities, even though subject staging, subtype assignment, and multimodal interplay are mutually dependent. We introduce ELPAC, a Bayesian framework with two interlocking constructions. The first integrates subtype trajectories backward from clinically reliable endpoint groups along biologically structured velocity components encoding modality-specific monotone direction and optional sequential ordering priors. The second factorizes residual multimodal variation around the inferred stage axis into sparse coordination axes whose loadings and activations expose which multimodal patterns are operative in each subtype and when they emerge, with stage-resolved variation supplying the auxiliary structure for conditional identifiability. We use a scalable amortized variational inference scheme that enables joint posterior inference at cohort scale with single-pass deployment on new subjects. On synthetic data and a harmonized ADNI+NACC/SCAN Alzheimer's cohort, ELPAC demonstrates accurate latent-structure recovery, well-calibrated prediction intervals, and biologically interpretable multimodal coordination programs that differ by subtype.
Read the full abstract
Image and Video Processing
Volumetric Latent Diffusion for Brain Tumour Segmentation with Missing Modalities
Brain tumour segmentation from magnetic resonance imaging normally assumes that all four imaging sequences are available, but in clinical practice one or more are frequently missing. Existing methods for incomplete inputs learn modality-invariant representations or distil from a full-modality teacher, and return a single deterministic mask. Gen- erative formulations instead model a distribution over plausible masks, but the cost of sampling at full resolution has largely confined them to two dimensions. This paper proposes a three-dimensional conditional latent diffusion framework for whole-tumour segmentation from incom- plete multi-sequence MRI. Images and annotations are first compressed by two volumetric autoencoders, which makes diffusion over whole vol- umes affordable, and sequences are dropped at random during training so that a single model handles every combination of missing sequences. Because the model samples rather than predicts, repeated draws also yield a voxel-wise map of disagreement at no extra training cost. Evalu- ations on 2,601 studies from BraTS 2023 and 2024 show that the method reaches 0.955 Dice with all four sequences and 0.904 with FLAIR alone, and that it is the most accurate of the four methods compared in eight of the fifteen configurations. The disagreement between draws widens as sequences are withheld, giving a signal that accompanies the prediction rather than replacing it
Read the full abstract
Physics and Society
Predicting next-week VIIRS fire detections in Bolivia: a reproducible benchmark with temporal and geographic evaluation
Weekly detection forecasts could help prioritize fire monitoring across Bolivia. We estimate the probability of at least one fire detection by the Visible Infrared Imaging Radiometer Suite (VIIRS) on the Suomi National Polar-orbiting Partnership satellite in each 0.05-degree cell during the following 7 days. Primary models were fixed before test scoring. Calibrated LightGBM, a gradient-boosted decision-tree model, reaches 2023-2025 temporal-test average precision (AP) of 0.323 (95% block-bootstrap interval 0.222-0.383), compared with 0.167 for climatology and 0.143 for persistence. Its paired AP gain over climatology is 0.1567 (0.0998-0.2179). A separately selected geographic model reaches AP 0.384 versus 0.181 for persistence in regions excluded from fitting. In the temporal-test population, reviewing the highest-ranked 5% and 10% of eligible area would have covered 38.4% and 55.8% of cell-weeks with detections. Probabilities still overestimate detection frequency in the highest-probability group. Retrospective rainfall provides no clear AP improvement; the primary validation comparisons for vegetation, land cover and their combination are inconclusive. Positive secondary or exploratory intervals do not change those decisions. Evaluation requires complete target and recent-history observations and measures satellite detections, not individual fires, causes or burned area. Prospective evaluation should archive inputs at each weekly issue and measure ranking, calibration and the usefulness of selected cells for monitoring decisions.
Read the full abstract
Earth and Planetary Astrophysics
BeAR: The Bern Atmospheric Retrieval code - A GPU-accelerated framework for low-resolution, high-resolution, and joint retrievals of exoplanet atmospheres
We present BeAR, the Bern Atmospheric Retrieval code and successor of Helios-r, and its new high-resolution cross-correlation framework for emission and transmission spectra. BeAR performs joint retrievals by adding the log-likelihoods of high- and low-resolution data computed from the same model atmosphere. The framework implements established cross-correlation log-likelihoods, with the line-depth scaling either fitted analytically or sampled as a free parameter, their per-pixel generalization, fast model filtering, and model re-injection adapted to this filtering. The forward model and likelihood run entirely on a graphics processing unit (GPU). One forward model takes about 1 ms on an NVIDIA GH200, allowing millions of models per minute on a GPU cluster. In separate retrievals of Gemini South/IGRINS and JWST/NIRSpec dayside spectra of the hot Jupiter WASP-77Ab, BeAR reproduces published abundances, C/O ratios, and metallicities to within 1.5 sigma, the preference for 13CO, and the thermal structure and line contrast to within 1 sigma. The joint retrieval confirms a sub-solar metallicity and a near-solar C/O ratio. A full IGRINS retrieval with 17 free parameters takes 27 to 42 minutes on a single NVIDIA GeForce RTX 5080 or about 14 minutes on eight RTX 4090 GPUs. Injection–recovery tests show that the posteriors underestimate the uncertainties of abundance ratios by about a factor of two.
Read the full abstract
Quantum Physics
Provable circuit-level thresholds for non-abelian topological phases
We provide an explicit scheme to realize a quantum memory in any 2D non-chiral topological order. For this, we construct syndrome extraction circuits based on the Turaev-Viro path integral as well as a classical algorithm for decoding on the fly. We prove that our scheme has a threshold under a local stochastic circuit-level noise model and discuss how to generalize it to implement topological logical gates.
Read the full abstract
Machine Learning
User Embeddings are Superpositions of Interpretable Behavioral Modes
User embeddings are central components of modern recommendation, personalization, and behavioral modeling systems. Though these embeddings can compactly encode rich behavioral information, they are inherently dense and opaque, limiting interpretability. To address this gap, we argue that user embeddings can be understood as sparse superpositions of interpretable behavioral modes. This structure is directly predicted by generative models of user behavior for linear embeddings and extends approximately to deep-learning representations by the superposition hypothesis. Superimposed modes can naturally be recovered by sparse dictionary learning methods, such as sparse autoencoders (SAEs), and recovered modes can be automatically named and queried in natural language. We validate mode recovery against known ground truth on synthetic data, where SAEs recover modes better than classical sparse baselines. We then apply our methods to classical and transformer-based user embeddings from two real-world datasets. We also show how the corresponding natural-language labels enable zero-shot audience retrieval and characterization over embeddings never trained on text. The framework is post-hoc and model-agnostic, requiring no modification to the upstream embedding system.
Read the full abstract
Computer Vision
Do You Miss the Nights We Spent Crafting Figures? Rethinking Teaser and Framework Figures in the Age of Generative AI
Scientific figures give research ideas a visual form through choices of abstraction, composition, and emphasis. Generative AI offers new ways to realize these choices, raising questions about explanatory clarity and purposeful visual expression. We examine these questions through a visual retrospective of 371 motivation and framework figures from 200 papers. The corpus comprises a quota-based main sample of 160 papers spanning four temporal periods and a separate purposive style-contrast sample of 40 papers. A single multimodal judge evaluates six dimensions of communication and design, with three repeated judgments per figure. We attempt source-constrained revisions of all 371 figures, of which 369 pass the recorded fidelity checks. Original-figure ratings show no uniform decline across periods. Revisions increase mean clarity ratings by 0.428 points and decrease aesthetic coherence ratings by 0.111 points on a five-point scale. Gains in the six-dimension average are larger for main-sample motivation figures than for frameworks. Case analyses show that added explanation can clarify comparisons while leaving typographic hierarchy and visual economy unresolved. These model-based findings support researcher-led design, with AI assisting exploration and authors retaining responsibility for explanatory purpose, composition, and scientific relationships. The accompanying atlas and evaluation records are available at https://ntu-ai4x.github.io/Rethinking_Figures/index_en.html.
Read the full abstract
Computer Vision
Where and Why is Grounded Reasoning Useful?
Grounded reasoning, where a model outputs text coordinates of relevant locations in chain-of-thought reasoning, is a common strategy for solving complex visual reasoning problems. However, the benefits and limitations of grounded reasoning are understudied. We study where and why grounded reasoning is most useful. To do so, we introduce a chain-of-thought analysis method to measure how much a model uses its outputted grounding in its thinking trace and perform a detailed ablation study of a grounded reasoning training recipe. We find that grounded reasoning improves counting but shows no significant gain over baselines without coordinate grounding across a wide range of spatial reasoning and perception benchmarks. Using several synthetic tasks, we show that grounded reasoning is most useful when a task requires undescribable references, which are visual concepts that are difficult to describe in language. However, we observe that many existing visual reasoning benchmarks are solvable with only describable text references. To close this gap, we introduce BrickVision, a new benchmark that requires undescribable visual references to solve.
Read the full abstract
Earth and Planetary Astrophysics
Evolution of Dust In Protoplanetary Disks With Magnetized Winds
Observations and theory increasingly indicate that protoplanetary disks are weakly turbulent, implying that magnetized disk winds may dominate their angular momentum transport and long-term evolution. Yet dust evolution in wind-dominated disks remains poorly understood, despite its importance for planet formation. In particular, whether wind-dominated disk evolution can create regions with enhanced dust-to-gas mass ratios that enable planetesimal formation via the streaming instability is unclear. We build a parametric hybrid disk evolution model that includes turbulent accretion, wind-driven accretion, and wind-driven mass loss to investigate dust evolution in magnetized protoplanetary disks. We employ this model to carry out a parameter study in which the initial total accretion rate, the wind-driven mass loss rate, and the turbulence strength are varied. We find that although wind-dominated disks are more favorable to dust growth and concentration than turbulence-dominated disks, conditions suitable for the streaming instability arise only within the inner few au and only under the most extreme conditions: very weak turbulence ($\alpha_\text{turb} = 10^{-5}$) where magnetized winds drive most of the accretion ($\dot{M}_\text{acc,tot} \simeq \dot{M}_\text{wind,acc}$), and wind-driven mass loss comparable to the accretion rate ($\dot{M}_\text{wind,outflow} \simeq \dot{M}_\text{acc,wind}$). Our results therefore imply that, beyond the inner few au, the parametric wind-driven evolution models considered in the literature do not by themselves produce conditions suitable for planetesimal formation via the streaming instability, motivating more physically complete models in which spontaneous, long-lived dust traps can arise.
Read the full abstract
Artificial Intelligence
SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization
LLM-guided evolutionary search uses AI agents to improve programs for mathematical and computational problems by iteratively refining solutions and combining high-scoring methods. However, the process is often slow because the LLM generates a long textual chain of thought for each round and must repeat this generation many times to maintain population diversity. We present SwarmEvolve, a framework in which hundreds of low-cost agents per round work in swarms, reason in latent space without decoding to text, and share information by passing KV caches. Each swarm studies the problem from a different angle. A novelty filter keeps the most distinct agent ideas by directly measuring the distances between the agents' hidden state vectors. Crossover of ideas happens naturally in latent space, as a summarizer agent receives the combined KV caches of all swarm members. On 21 problems in mathematical discovery and scientific computing, SwarmEvolve gets the highest average score and needs a median of only 3 to 8 minutes to reach the best scores that other state-of-the-art frameworks reach in 60 minutes. On several problems, SwarmEvolve with Qwen3 8B improves on the solutions that AlphaEvolve first found with frontier LLMs. Our work shows that agent swarms that communicate in latent space offer a fast, low-cost approach to program search for scientific problems.
Read the full abstract
Nuclear Theory
Hypersurface covariance in relativistic local equilibrium
Defining equilibrium typically requires specifying a three-dimensional spacelike hypersurface. While global equilibrium is hypersurface covariant, calculations involving local equilibrium can exhibit hypersurface dependence. In this work, we use polarization in heavy-ion collisions within the Zubarev response approach as an example to show that the hypersurface dependence arising from the local-equilibrium part of the density operator is canceled by an overlooked nondissipative contribution from its so-called “dissipative part". We establish a hypersurface-covariant Zubarev framework for general geometries by consistently including both nondissipative and dissipative parts, eliminating dependence on the arbitrary hypersurface vector. The remaining frame dependence is physical: it arises from the medium rest frame entering the particle distribution and the time direction of the dissipative process, which usually coincide. This resolves, in principle, the longstanding and widely discussed conceptual disagreement over hypersurface dependence between two influential studies addressing the sign puzzles of $\Lambda$-hyperon polarization in heavy-ion collisions
Read the full abstract
Cosmology and Nongalactic Astrophysics
Joint analysis of the Macquart relation and the dispersion measure power spectrum: cross-covariance and redshift self-calibration
The dispersion measures (DMs) of fast radio bursts (FRBs) trace ionised baryons along the line of sight, probing baryonic feedback. Two complementary observables can be constructed from an FRB catalogue: the DM-redshift (Macquart) relation of host-identified FRBs, and the angular power spectrum of the DM field of the full catalogue. Their joint analysis requires their cross-covariance. We derive it within the halo model and show it is a super-sample effect: only the monopole of an individual DM correlates with the angular power spectrum, yielding an exact, model-independent bound on the correlation coefficient. We validate the calculation against log-normal forward simulations of $10^4$ FRBs. For individual DMs, the correlation with the bandpowers is at the per cent level. Neglecting the cross-covariance changes the width of the $\chi^2$ distribution by only 0.8 per cent, so the two observables can be treated as effectively independent for wide-area surveys. In a joint forecast with an LSST-like weak lensing survey, adding the Macquart relation tightens constraints on the physical baryon density by 32 per cent and on the mean host contribution by a factor of 4.1, while improving baryonification parameters by only about 5 per cent, indirectly through the host contribution. Finally, we show that the redshift distribution of FRBs without identified hosts, which enters every DM power spectrum, can be reconstructed by inverting their DMs with the jointly fitted mean relation without biasing the inferred parameters, whereas a 20 per cent error in the mean relation biases the DM power spectrum by up to 2$\sigma$ per band for $5\times 10^5$ FRBs.
Read the full abstract
Differential Geometry
Hyperpolarity of polar foliations of codimension two
We consider polar foliations of codimension two on compact simply connected irreducible symmetric spaces. In rank at least two, we prove that these foliations are hyperpolar, settling the remaining case of the hyperpolarity problem discussed by Lytchak. We show that, for irreducible spherical Coxeter quotients, the horizontal universal cover fails to be a spherical building precisely on the Cayley plane, where every such foliation is congruent to one of the two known homogeneous models.
Read the full abstract
Quantum Gases
Flat-band systems: A platform to realize the moments of Quantum Metric
The quantum metric is a fundamental geometric property of Bloch bands that governs a wide range of physical phenomena, yet experimentally accessing it remains challenging. We show that the momentum-space variation of the quantum metric can be probed through the real-space dynamics of a Gaussian wave packet prepared in an isolated flat band. Following the application of a weak constant force, we demonstrate analytically that the leading order change in the wave-packet variance is linear in time and contains a contribution proportional to the wave-packet averaged gradient of the quantum-metric trace. This geometric contribution remains finite despite the absence of band dispersion. We illustrate the effect using a gapped Lieb lattice with an exactly flat middle band and vanishing Berry curvature, allowing the quantum-metric contribution to the spreading to be isolated. The numerical results agree with the analytical prediction and confirm the quantum-metric contribution to the wave-packet spreading. The required lattice preparation, tunable hopping amplitudes, applied force, and density measurements are compatible with established ultracold-atom techniques, providing a possible platform for observing the predicted response. Our results establish wave-packet spreading as a dynamical probe of quantum geometry in flat-band systems.
Read the full abstract
Artificial Intelligence
Scientist World: Grow a Research World for AI-Scientists
AI-Scientists arrived, but a world for them to grow in did not. Human scientists take theirs for granted $-$ centuries of open problems and shared ideas, waiting in the scientific literature. To AI-Scientists, that world is readable but not runnable, so we build Scientist World for them: every research paper becomes an executable environment. Our autonomous pipeline extracts each paper's open-ended research question and reproduces its proposed method as the baseline; an AI-Scientist entering the environment is evaluated against what the human authors achieved. We release the first 1,000 verified environments. With the world as an evaluation, we found that frontier AI-Scientists can already surpass the human researchers 74% of the time. As a record of how AI does science, it contains 6,400 hours of research traces, with every idea, experiment, failure, and success. As a training ground, we are able to raise (+158%) a small model to match frontier models in research ideation. At last, AI-Scientists' own world has arrived, grown from and growing with science itself.
Read the full abstract
Computer Vision
EgoGeneralist: Unifying Egocentric Action-Video Tasks through Video-Native Modeling
Modern video models possess strong visual priors, yet do not natively represent human action in metric 3D space. We present EgoGeneralist, which adapts pretrained video generators to unify action-conditioned video generation, pose estimation, and joint action-video generation for egocentric manipulation. Our role-adaptive action renderer provides video-native representations for multidirectional action-video learning, making metric motion usable as both a spatial control signal and a recoverable generation target. Cross-stream action-video coupling establishes close correspondence between metric motion and visual dynamics through lightweight adaptation while preserving the pretrained visual prior. Together, these designs establish a generalist framework spanning six action-video tasks. With only narrow-domain data, EgoGeneralist generalizes to diverse unseen scenes and interactions with effective action control and visual quality, bridging pretrained visual knowledge and metric 3D human motion.
Read the full abstract
High Energy Physics (Theory)
Defect Anomalies at a Distance
Surface defects carry Weyl and gravitational anomalies that generalize the central charges of two-dimensional conformal field theories (CFTs) and characterize the anomalous response of gapless surface modes to background geometry. These anomalies usually appear as contact terms on the defect. We explain how they are encoded at a distance, in scheme-independent integrated bulk stress-tensor correlators. We derive sum rules for them in terms of defect operator data, establish a lower bound on the boundary Weyl anomaly in three-dimensional CFTs, and illustrate our results with explicit examples.
Read the full abstract
Strongly Correlated Electrons
Direct transition from a fractional Chern insulator to a topological superconductor
We find numerical evidence for a direct continuous transition from a Laughlin fractional Chern insulator (FCI) to a $c_-=3/2$ topological superconductor (TSC), driven by increasing bandwidth at filling $\nu=1/3$ in a repulsively interacting Chern band. At the critical point, the electron correlation length diverges at momentum away from the reference Fermi surface. We develop a generalized theory of superconducting-vortex proliferation for this transition. Magnetic translation organizes the vortices into six valleys, so translation-preserving proliferation requires additional gauge fields. The resulting theory reproduces the momentum structure of the critical spectrum and captures the divergent electron correlation. Upon doping away from commensurate filling, the theory gives rise to a superconducting dome. These results are consistent with an anyonic origin of the topological superconductivity in the absence of a weak-coupling Fermi-surface instability.
Read the full abstract
Databases
Limitations of Interpretations in Semiring Semantics
Interpretations have been widely used in the study of provenance of queries on semiring-annotated databases. Here, we consider two different interpretation-based semantics for first-order formulas in negation normal form, namely, the canonical-interpretation semantics and the compatible-interpretation semantics. We show that the resulting query languages cannot express the monus operation on semiring-annotated relations, when the annotations come from a positive semiring with monus and with at least three elements in its universe. Consequently, the semantics of SQL cannot be captured using the interpretation-based semantics considered here.
Read the full abstract
Nuclear Experiment
Correlations and fluctuations of flow with event shape engineering in Pb-Pb and Xe-Xe collisions at the LHC
Measurements of anisotropic flow provide some of the strongest constraints on the collective expansion and transport properties of the quark-gluon plasma produced in ultra-relativistic heavy-ion collisions. In this paper, correlations and fluctuations of the anisotropic flow coefficients $v_{2}$, $v_{3}$, and $v_{4}$ are measured in Pb-Pb and Xe-Xe collisions at $\sqrt{s_{\rm{NN}}} = 5.02$ and $5.44$ TeV, respectively, with the ALICE detector. The Event Shape Engineering technique, based on the reduced flow vector $q_{n}$, is used to study flow correlations for different initial-state geometries. A strong anticorrelation between $v_2$ and $v_3$ and a positive correlation between $v_2$ and $v_4$ are observed. A small anticorrelation is seen between $v_3$ and $v_4$. From the $v_2$-$v_4$ correlation, the linear and nonlinear components of $v_4$ are extracted and compared to those extracted using multiparticle correlations. The linear and nonlinear components of $v_{4}$ are also measured directly as a function of $q_2$ for the first time in Pb-Pb and Xe-Xe, revealing a strong coupling of the nonlinear term to the initial-state ellipticity $\varepsilon_2$, while the linear component shows little dependence on initial-state geometry. Additionally, the coupling coefficient $\chi_{4,22}$, which measures the nonlinear contribution of $v_2^2$ to $v_4$, is extracted and shows little dependence on $q_2$, except in very central collisions. The effects of flow fluctuations for constrained initial-state geometries are studied for the first time using Event Shape Engineering with the ratio of $v_2\{2\}$ and $v_2\{4\}$; the results show that collisions with the lowest $q_2$ exhibit more significant $v_2$ fluctuations compared to the unbiased ratio.
Read the full abstract
Machine Learning
XStack: A Benchmark for ML Repo Migration
ML code should run where researchers choose, not where it was first written. Today, machine learning code mainly lives in two ecosystems, PyTorch on NVIDIA GPUs and JAX on Google TPUs. Migrating repositories between them means building and testing across frameworks and hardware. This migration is (1) a practical need, since it unlocks different hardware for researchers, and (2) a natural benchmark: it involves different frameworks and hardware, covers whole repositories, and is long-horizon and challenging. We introduce XStack, a benchmark that evaluates agents on migrating ML repositories between these two ecosystems. Each agent receives a PyTorch repository and a clean TPU machine and must deliver a JAX repository. Its output passes if it meets our grading criteria, based on the source repository's published result, authenticated and calibrated by us. Migration by agents remains unreliable: Claude Opus 5 passes 75%, GPT-5.6-sol 45%, and every other agent at most 40%. XStack is a step toward freeing ML code from its ecosystem. We release XStack to make progress toward reliable migration across frameworks and hardware measurable.