arXiv subjects

Ke Zhang

Publications and source records attributed to Ke Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2026

Characterizing the Extended Molecular Hydrogen Winds in Protoplanetary Disks from the JWST Disk Infrared Spectroscopic Chemistry Survey

We present a comprehensive analysis of extended H$_2$ emission from 34 protoplanetary disks observed with the JWST Disk Infrared Spectroscopic Chemistry Survey (JDISCS), supplemented by archival data. We investigated the morphology, kinematics, excitation conditions, and mass dynamics of H$_2$. Extended emission from pure rotational H$_2$ lines is found to be common, with 16 sources exhibiting clear signatures of disk winds. These include monopolar and bipolar structures in inclined disks and ring-like or bubble-like morphologies in face-on systems features indicative of wide-angle disk winds. Our analysis shows that the H$_2$ is consistent with slow {(4.2$^{+6.7}_{-3.0}$ km s$^{-1}$)} MHD driven winds. For ten disks, we model the wind morphology and find a median half-opening angle of $45\arcdeg^{+5}_{-4}$ and a characteristic power-law index of $α\sim$ 1.6. Excitation analysis yields a median gas temperature of 624 $\pm$ 130 K and a column density of $\log(N_{\mathrm{tot}}\,[\mathrm{cm}^{-2}]) = 18.6 \pm 0.6$. The median wind mass-loss rate, ${\rm log_{10}}(\dot{\rm M}_{\rm wind}^{\rm tot}) = -9_{-0.4}^{+0.8}\,{\rm M_\odot\,yr^{-1}}$, implies that, if molecular winds are the dominant mechanism responsible for disk dispersal, a typical disk with a mass of $2-3\,M_{\rm Jup}$ would dissipate on a $\sim$2-3 Myr timescale, consistent with observed disk lifetimes. The $\dot{\rm M}_{\mathrm{\rm wind}}^{\rm tot}$ span a relatively narrow range ($\sim$2 dex) and do not correlate strongly with accretion rates onto the star, suggesting that the mass loss rate and the accretion rates are probing different timescales. Our findings demonstrate that spatially extended warm H$_2$ emission is a widespread and reliable tracer of molecular disk winds in protoplanetary systems.

astro-ph.SR↗

arXiv2026

TRAP: Tail-aware Ranking Attack for World-Model Planning

World models enable long-horizon planning by internally generating and evaluating imagined trajectories, making them a promising foundation for generalist agents. However, this imagination-driven decision process also introduces new security risks. Existing backdoor attacks typically aim to manipulate local features, one-step predictions, or instantaneous policy outputs. While such objectives may suffice for weaker reactive models, they are often ineffective against world models, where the learned dynamics prior and planning process can absorb or wash out the effects of shallow perturbations. More importantly, we find that world models exhibit a distinct backdoor vulnerability rooted in the long-tailed ranking structure of imagined trajectories, where disrupting the ordering of a few decision-critical trajectories can systematically hijack planning. To exploit this vulnerability, we propose TRAP, a backdoor attack framework for world models that targets imagined trajectory ranking. TRAP combines a tail-aware ranking loss to focus optimization on decision-critical trajectories with dual gating mechanisms that stabilize optimization and regulate when and where the attack penalty is applied. Under trigger conditions, TRAP alters the relative ranking of imagined trajectories to redirect planning outcomes, while largely maintaining the normal ranking structure on clean inputs. Experiments on DreamerV3 and TD-MPC2 across diverse tasks show that TRAP consistently induces sustained behavioral deviations and significant performance degradation, highlighting the need for dedicated security evaluation of world-model-based agents.

cs.LG↗

arXiv2026

Understanding Tool-Augmented Agents for Lean Formalization: A Factorial Analysis

Automatic translation of natural language mathematics into faithful Lean 4 code is hindered by the fundamental dissonance between informal set-theoretic intuition and strict formal type theory. This gap often causes LLMs to hallucinate non-existent library definitions, resulting in code that fails to compile or lacks semantic fidelity. In this work, we investigate the effectiveness of tool-augmented agents for this task through a systematic factorial analysis of three distinct tool categories: Fine-tuned Model Querying (accessing expert drafts), Knowledge Search (retrieving symbol definitions), and Compiler Feedback (verifying code via a Lean REPL). We first benchmark the agent against one-shot baselines, demonstrating large gains in both compilation success and semantic equivalence. We then use the factorial decomposition to quantify the impact of each category, isolating the marginal contribution of each tool type to overall performance.

cs.SE↗

arXiv2026

The ALMA Survey of Gas Evolution of PROtoplanetary Disks (AGE-PRO): Constraints on disk turbulence, fragmentation velocity, and inner pebble fluxes

How substructures and disk properties affect dust evolution and the delivery of solids and volatiles into planet-forming regions remains an open question. We present results from tailored dust evolution modeling of the AGE-PRO ALMA large program, a sample of 30 protoplanetary disks spanning different evolutionary stages. Visibility fitting of the AGE-PRO ALMA data (at 1.3\,mm) reveals that approximately half of the disks exhibit radial substructures. Combined with stellar properties, disk inclinations, and gas mass estimates from CO isotopologues and N$_2$H$^+$, this well-characterized set of disks provides an ideal testbed to constrain dust evolution models across different ages and disk morphologies. Using the dust evolution code \texttt{DustPy}, we simulate dust evolution in each disk under four model configurations, varying two key free parameters: the turbulent viscosity ($α= 10^{-4}, 10^{-3}$) and fragmentation velocity ($v_{\rm{frag}} = 1 \mathrm{m\,s^{-1}}, 10 \mathrm{m\,s^{-1}}$). Pressure traps are incorporated by perturbing the gas surface density based on the continuum intensity profiles, and synthetic observations generated with \texttt{RADMC-3D} are compared to these profiles. While no single model fits all disks, nearly half are best reproduced by the configuration with low turbulence and low fragmentation velocity ($α= 10^{-4}, v_{\rm{frag}} = 1\,\mathrm{m\,s^{-1}}$). Models of smooth disks underpredict dust mass, possibly indicating unresolved substructures. Pebble fluxes into inner disk regions correlate more strongly with disk age than with the presence of substructures, highlighting time-dependent dust transport as a key factor in shaping inner disk composition. Our results also provide a comparative baseline for interpreting multiwavelength and JWST water vapor observations.

astro-ph.EP↗

arXiv2026

Learning Parameterized Nonlinear Elasticity on Curved Surfaces

We learn parameterized nonlinear elasticity on curved surfaces using a physics-informed neural network that enforces governing equations and boundary conditions directly through the loss function, enabling a single trained model to represent a continuous family of elastic equilibria across geometric and material parameters. Nonlinear elasticity on curved manifolds underlies the mechanics of crystalline shells, elastic membranes, and viral capsids, where curvature and topological defects determine equilibrium structure and stability. Traditional exact and finite element solvers rely on symmetry reduction and must be reinitialized for each parameter choice, limiting scalability when symmetry is broken or parameters vary. We validate the proposed learning-based solver on a benchmark problem from curved elasticity, namely the one-dimensional single disclination on a spheroidal surface with known exact and numerical solutions. The network accurately reproduces these solutions, including parameter combinations excluded from training, demonstrating generalization across geometry and material regimes. This study establishes a scalable framework for learning nonlinear elastic systems on curved manifolds and lays the groundwork for extensions to fully two-dimensional and multi-defect configurations relevant to protein shells and other curved elastic networks.

physics.bio-ph↗

arXiv2026

JWST/MIRI Hydrocarbon and Water Absorption in the Wind of a Young Disk: Signatures of Pebble Drift and Carbon Grain Sublimation

We present JWST/MIRI-MRS observations of ISO-Oph 37, a highly inclined flat-spectrum ($\lesssim$1 Myr old) source, to investigate the chemical composition and dynamical origin of its inner-disk gas. The spectrum reveals a rich combination of molecular emission and absorption: H$_2$O, CO, and OH are detected in emission, while strong absorption is observed from CO, H$_2$O, CO$_2$, HCN, C$_2$H$_2$, and CH$_4$, with no detectable ice absorption features. LTE slab modeling of the absorption yields excitation temperatures of $T_{\rm ex}\sim400-600$ K and column densities of $\log N/{\rm cm}^{2}\sim16-19$, characteristic of warm gas located within the inner few au. The absorption lines are significantly blueshifted relative to the systemic velocity, with mid-IR lines exhibiting larger shifts than near-IR CO absorption. This velocity structure points to a velocity- and temperature-stratified molecular disk wind. In this framework, the absorption directly samples disk material lifted from the inner disk surface, preserving the chemical imprint of the wind-launching region. Along the line of sight, ISO-Oph 37 is unusually hydrocarbon-rich compared to other known absorption systems (GV Tau N and IRS 46), exhibiting high (C$_2$H$_2$+CH$_4$)/HCN, (C$_2$H$_2$+CH$_4$)/CO and H$_2$O/CO column density ratios, while the CO and HCN columns remain broadly typical. We find that these molecular ratios are best explained by enhancement of both hydrocarbons and water, driven by inward drift and sublimation of icy pebbles and by thermal processing of carbonaceous grains at the soot line. ISO-Oph 37 thus demonstrates that carbon-rich inner-disk chemistry can be established early in disk evolution and that it can be directly probed through molecular absorption in disk winds.

astro-ph.EP↗

arXiv2026

NestPipe: Large-Scale Recommendation Training on 1,500+ Accelerators via Nested Pipelining

Modern recommendation models have increased to trillions of parameters. As cluster scales expand to O(1k), distributed training bottlenecks shift from computation and memory to data movement, especially lookup and communication latency associated with embeddings. Existing solutions either optimize only one bottleneck or improve throughput by sacrificing training consistency. This paper presents NestPipe, a large-scale decentralized embedding training framework that tackles both bottlenecks while preserving synchronous training semantics. NestPipe exploits two hierarchical sparse parallelism opportunities through nested pipelining. At the inter-batch level, Dual-Buffer Pipelining (DBP) constructs a staleness-free five-stage pipeline through dual-buffer synchronization, mitigating lookup bottlenecks without embedding staleness. At the intra-batch level, we identify the embedding freezing phenomenon, which inspires Frozen-Window Pipelining (FWP) to overlap All2All communication with dense computation via coordinated stream scheduling and key-centric sample clustering. Experiments on production GPU and NPU clusters with 1,536 workers demonstrate that NestPipe achieves up to 3.06x speedup and 94.07% scaling efficiency.

cs.DC↗

arXiv2026

MVis-Fold: A Three-Dimensional Microvascular Structure Inference Model for Super-Resolution Ultrasound

Super-resolution ultrasound (SRUS) technology has overcome the resolution limitations of conventional ultrasound, enabling micrometer-scale imaging of microvasculature. However, due to the nature of imaging principles, three-dimensional reconstruction of microvasculature from SRUS remains an open challenge. We developed microvascular visualization fold (MVis-Fold), an innovative three-dimensional microvascular reconstruction model that integrates a cross-scale network architecture. This model can perform high-fidelity inference and reconstruction of three-dimensional microvascular networks from two-dimensional SRUS images. It precisely calculates key parameters in three-dimensional space that traditional two-dimensional SRUS cannot readily obtain. We validated the model's accuracy and reliability in three-dimensional microvascular reconstruction of solid tumors. This study establishes a foundation for three-dimensional quantitative analysis of microvasculature. It provides new tools and methods for diagnosis and monitoring of various diseases.

cs.CV↗

arXiv2026

From Imitation to Intuition: Intrinsic Reasoning for Open-Instance Video Classification

Conventional video classification models, acting as effective imitators, excel in scenarios with homogeneous data distributions. However, real-world applications often present an open-instance challenge, where intra-class variations are vast and complex, beyond existing benchmarks. While traditional video encoder models struggle to fit these diverse distributions, vision-language models (VLMs) offer superior generalization but have not fully leveraged their reasoning capabilities (intuition) for such tasks. In this paper, we bridge this gap with an intrinsic reasoning framework that evolves open-instance video classification from imitation to intuition. Our approach, namely DeepIntuit, begins with a cold-start supervised alignment to initialize reasoning capability, followed by refinement using Group Relative Policy Optimization (GRPO) to enhance reasoning coherence through reinforcement learning. Crucially, to translate this reasoning into accurate classification, DeepIntuit then introduces an intuitive calibration stage. In this stage, a classifier is trained on this intrinsic reasoning traces generated by the refined VLM, ensuring stable knowledge transfer without distribution mismatch. Extensive experiments demonstrate that for open-instance video classification, DeepIntuit benefits significantly from transcending simple feature imitation and evolving toward intrinsic reasoning. Our project is available at https://bwgzk-keke.github.io/DeepIntuit/.

cs.CV↗

arXiv2026

GJ 523b is a Massive, 170 Myr-old Mega-Earth, Likely on a Polar Orbit

We use WIYN/NEID radial velocity measurements to confirm the planetary nature and measure the mass of the TESS transiting exoplanet candidate around the mid-K dwarf GJ 523 ($V=9.23$, $K=6.525$). We find that GJ 523b is on a 17.75 day orbit and has a radius of $2.55\pm0.15\,R_\oplus$, a mass of $23.5\pm3.3\,M_\oplus$, and a zero-albedo equilibrium temperature of 538 K. GJ 523b's high bulk density of $7.8\pm1.8$ g cm$^{-3}$ and position on a mass-radius diagram implies a surprising low atmospheric mass fraction despite its relatively large mass. Additionally, we determine that the system has an age of $169^{+100}_{-48}$ Myr through a gyrochronological analysis of GJ 523 and its comoving companions. We also use the SED-derived stellar radius, the photometric rotation period, and the spectroscopic $v\sin i_\star$ to derive a stellar inclination of $17.6\pm5.0$ degrees, implying that GJ 523b has a minimum orbital obliquity of $71.4_{-5.0}^{+4.7}$ degrees. GJ 523b's high mass, apparent lack of a gas envelope, young age, and high orbital obliquity present a challenge to typical planet formation pathways, and at the moment there is not enough data on the system to definitively determine how GJ 523b formed. Finally, we present a new observational classification for ultra-dense, sub-Neptune-sized exoplanets similar to GJ 523b: the mega-Earths, planets with $R_p \geq2.1\,R_\oplus$ and $ρ_p \geq 5.5$ g cm$^{-3}$.

astro-ph.EP↗

arXiv2026

Optimal error estimates for a fully discrete, highly efficient decoupled scheme for the 2D/3D diffuse interface two-phase MHD flows

In this paper, we derive optimal L2- and H1-norm error estimates for a fully discrete convex-splitting decoupled finite element method (FEM) for the two-phase diffuse interface magnetohydrodynamics (MHD) system. We use the semi-implicit backward Euler scheme in time and employ the standard inf-sup stable Taylor--Hood or Mini elements to discretize the velocity and pressure. Furthermore, we apply a pressure-correction scheme to decouple the velocity from the pressure. The optimal error estimates are obtained via novel Ritz and Stokes quasi-projection techniques. In addition, the unconditional energy stability of the proposed scheme is ensured. Numerical examples are presented to validate the theoretical analysis.

math.NA↗

arXiv2026

exoALMA. XXIV. Formaldehyde Emission in Protoplanetary Disks of exoALMA Compared with Their Properties and Dynamical State

The presence of asymmetries and substructures in protoplanetary disks, revealed by both dust and gas emission, highlights the potential interplay and the broader connection between chemistry and dynamics in disk evolution. We explore multiple relationships using the nonparametric Kendall-$τ$ correlation to examine formaldehyde (H$_2$CO) emission with relation to stellar and disk properties for a subset of disks from the exoALMA sample. We also retrieve the H$_2$CO column density and excitation temperature using four transitions, measured in radial bins of 100 au, and quantify the level of asymmetry in the resolved peak intensity of the H$_2$CO emission. From our correlation analysis, we find no correlations with sufficient statistical significance. However, we identify tentative relationships that can be tested with larger samples. In particular, we report a proposed correlation ($2.1σ$) between stellar effective temperature and the formaldehyde excitation conditions, suggesting that, to first order, the central star dominates the nature of the H$_2$CO emission over possible dynamical asymmetries traced by dust. Although a correlation with the stellar luminosity was also expected, a larger sample is required to confirm or refute this trend. A possible correlation with spectral type, together with the broad range of H$_2$CO excitation temperatures within the inner 100 au of the studied disks, hint at possible multiple chemical formation pathways for H$_2$CO, including both gas-phase reactions and ice-surface chemistry on dust grains.

astro-ph.EP↗

arXiv2026

Third-order transitions in Ising and Potts models on Watts--Strogatz small-world networks

We study third-order transitions in the two-dimensional Ising and Potts model on regular lattices and Watts--Strogatz small-world networks. Cluster observables are used to track post-critical boundary reorganization and pre-critical cluster breakup. For the Ising model, the critical temperature $T_c$ is calibrated independently from Binder-cumulant crossings and susceptibility peaks, whereas for the Potts model on small-world networks it is identified operationally from the dominant critical peak of $\mathrm d\langle P\rangle/\mathrm dT$. The independent and dependent third-order transitions are identified from the isolated-spin peak and the post-critical structural extremum, respectively. For both lattice and small-world topologies, we find the robust ordering $T_{\mathrm{ind}}<T_c<T_{\mathrm{dep}}$. Increasing the rewiring probability shifts all three characteristic temperatures upward and enhances the visibility of the post-critical transition. The effect is especially clear in the Potts model, where perimeter-based observables are more sensitive to multistate boundary fluctuations. The systematic persistence of the characteristic temperature hierarchy across topologies and finite sizes argues against interpreting these features as incidental finite-size irregularities. Instead, our results support their interpretation as genuine third-order transitions whose structural detectability can be amplified by network topology.

cond-mat.stat-mech↗

arXiv2026

On the $2$-adic valuation of $σ_k(n)$

For a positive integer $k$, let \[ σ_k(n)=\sum_{d\mid n} d^k \] be the divisor function of order $k$, and let $ν_p(m)$ denote the $p$-adic valuation of an integer $m$. Motivated by recent work on the $p$-adic valuation of $σ_k(n)$, we study $ν_2(σ_k(n))$ in detail. We prove that, for every integer $n\ge 2$, \[ ν_2(σ_k(n)) \le \begin{cases} \lceil \log_2 n \rceil, & \text{if $k$ is odd},\\[1mm] \lfloor \log_2 n \rfloor, & \text{if $k$ is even}. \end{cases} \] These bounds are best possible. More precisely, if $k$ is odd, then equality holds if and only if $n$ is a product of distinct Mersenne primes; if $k$ is even, then equality holds if and only if $n=3$. We also obtain an explicit formula for $ν_2(σ_k(n))$ in terms of the prime factorization of $n$.

math.NT↗

arXiv2026

xLLM Technical Report

We introduce xLLM, an intelligent and efficient Large Language Model (LLM) inference framework designed for high-performance, large-scale enterprise-grade serving, with deep optimizations for diverse AI accelerators. To address these challenges, xLLM builds a novel decoupled service-engine architecture. At the service layer, xLLM-Service features an intelligent scheduling module that efficiently processes multimodal requests and co-locates online and offline tasks through unified elastic scheduling to maximize cluster utilization. This module also relies on a workload-adaptive dynamic Prefill-Decode (PD) disaggregation policy and a novel Encode-Prefill-Decode (EPD) disaggregation policy designed for multimodal inputs. Furthermore, it incorporates a distributed architecture to provide global KV Cache management and robust fault-tolerant capabilities for high availability. At the engine layer, xLLM-Engine co-optimizes system and algorithm designs to fully saturate computing resources. This is achieved through comprehensive multi-layer execution pipeline optimizations, an adaptive graph mode and an xTensor memory management. xLLM-Engine also further integrates algorithmic enhancements such as optimized speculative decoding and dynamic EPLB, collectively serving to substantially boost throughput and inference efficiency. Extensive evaluations demonstrate that xLLM delivers significantly superior performance and resource efficiency. Under identical TPOT constraints, xLLM achieves throughput up to 1.7x that of MindIE and 2.2x that of vLLM-Ascend with Qwen-series models, while maintaining an average throughput of 1.7x that of MindIE with Deepseek-series models. xLLM framework is publicly available at https://github.com/jd-opensource/xllm and https://github.com/jd-opensource/xllm-service.

cs.DC↗

arXiv2026

High CO/H2 ratios supports an exocometary origin for a CO-rich debris disk

Over 20 exocometary belts host detectable circumstellar gas, mostly in the form of CO. Two competing theories for its origin have emerged, positing the gas to be primordial or secondary. Primordial gas survives from the belt's parent protoplanetary disk and is therefore H$_2$-rich. Secondary gas is outgassed \textit{in-situ} by exocomets and is relatively H$_2$-poor. Discriminating between these scenarios has not been possible for belts hosting unexpectedly large quantities of CO. We aim to break this gas origin dichotomy \textit{via} direct measurement of H$_2$ column densities in two edge-on CO-rich exocometary belts around $\sim$15 Myr-old A-type stars, constraining the $\frac{\text{CO}}{\text{H}_2}$ ratio and CO gas lifetimes. Observing edge-on belts enables rovibrational absorption spectroscopy against the stellar background. We present near-IR CRIRES+ spectra of HD 110058 and HD 131488 which provide the first direct probe of H$_2$ gas in CO-rich exocometary belts. We target the H$_2$ (v=1-0 S(0)) line at 2223.3 nm and and the $^{12}$CO $v=2\rightarrow0$ rovibrational lines in the range 2333.8-2335.5 nm and derive constraints on column densities along the line-of-sight to the stars. We strongly detect $^{12}$CO but not H$_2$ in the CRIRES+ spectra. This allows us to place $3σ$ lower limits on the $\frac{\text{CO}}{\text{H}_2}$ ratios of $> 1.35 \times 10^{-3}$ and $> 3.09 \times 10^{-5}$ for HD 110058 and HD 131488 respectively. These constraints demonstrate that at least for HD 110058, the exocometary gas is compositionally distinct and significantly H$_2$-poor, compared to the $<10^{-4}$ $\frac{\text{CO}}{\text{H}_2}$ ratios typical of protoplanetary disks. We also find H$_2$ alone is unlikely to shield CO over the lifetime of the systems. Overall this suggests that the gas in CO-rich belts is most likely not primordial in origin, supporting the presence of exocometary gas.

astro-ph.EP↗

arXiv2026

GatedCLIP: Gated Multimodal Fusion for Hateful Memes Detection

Detecting hateful content in multimodal memes presents unique challenges, as harmful messages often emerge from the complex interplay between benign images and text. We propose GatedCLIP, a Vision-Language model that enhances CLIP's multimodal capabilities with specialized architectural improvements for hateful memes detection. Our approach introduces learned projection heads that map CLIP embeddings to a task-optimized semantic space, a dynamic gated fusion mechanism that adaptively weights visual and textual features, and a contrastive learning objective that maintains cross-modal semantic alignment. Experiments on the Hateful Memes dataset demonstrate that GatedCLIP achieves an AUROC of 0.66, substantially outperforming the CLIP baseline (AUROC 0.49) while maintaining computational efficiency with only 350K trainable parameters.

cs.CV↗

arXiv2026

Physical and Chemical Characterization of GY 91's Multi-ringed Protostellar Disk with ALMA

GY 91, commonly categorized as a Class I young stellar object, is notable for disk dust substructures that have been hypothesized to trace early planet formation. Using the ALMA 12-m and ACA arrays, we present new Band 7 dust continuum and molecular line observations of GY 91 at an angular resolution of (~40 au). We report detections of CS $J=6-5$, N$_2$H$^+$ $J=3-2$, C$^{18}$O $J=3-2$, H$_2$CS $J_{K_a, K_c} = 8_{1,7}-7_{1,6}$, H$_2$CO $J_{K_a, K_c} = 4_{0,4}-3_{0,3}$, and H$_2$CO $J_{K_a, K_c} = 4_{2,3}-3_{2,2}$, as well as a tentative detection of $^{13}$C$^{18}$O $J=3-2$. We observe azimuthal asymmetry in CS and H$_2$CS emission, as well as radially structured H$_2$CO $4_{0,4}-3_{0,3}$ emission outside the dust continuum. C$^{18}$O and H$_2$CO 4$_{0,4}-3_{0,3}$ show significant cloud contamination, while CS and N$_2$H$^+$ are good tracers of Keplerian rotation originating from the disk. Envelope emission does not appear to contribute significantly either to the continuum or molecular line observations. GY 91's chemical properties appear in large part to resemble those of Class II disks, although observations of additional molecular probes should be obtained for a fuller comparison. With CS, we estimated a dynamical stellar mass of 0.58 $M_\odot$, which is higher than previous estimates from stellar evolutionary models (0.25 $M_\odot$). Using both radiative transfer modeling of the dust continuum and comparison of the C$^{18}$O and N$_2$H$^+$ fluxes to literature thermochemical models, we estimate a disk mass of $\sim0.01$ $M_\odot$.

astro-ph.EP↗