arXiv subjects

Hang Ding

Publications and source records attributed to Hang Ding.

13 recordsLinked to original sources

arXiv2026

Nonlinear asymptotic bubble growth in single-mode spherical Rayleigh-Taylor instability

We present an analytical model for the nonlinear growth of a single-mode Rayleigh-Taylor instability (RTI) bubble in spherical geometry. The model captures the bubble growth along the polar axis, spanning the linear to nonlinear regimes, for arbitrary Atwood numbers and under both converging- and diverging-gravity configurations. The model predicts that the bubble acceleration approaches an asymptotic value in the nonlinear stage. The spherical geometry is found to enhance the RTI bubble growth relative to planar and cylindrical configurations with the same effective perturbation wavenumber in the converging-gravity cases, whereas it mitigates the bubble growth in the diverging-gravity cases. The model predictions show favorable agreement with direct numerical simulations.

physics.flu-dyn↗

arXiv2026

SE-Search: Self-Evolving Search Agent via Memory and Dense Reward

Retrieval augmented generation (RAG) reduces hallucinations and factual errors in large language models (LLMs) by conditioning generation on retrieved external knowledge. Recent search agents further cast RAG as an autonomous, multi-turn information-seeking process. However, existing methods often accumulate irrelevant or noisy documents and rely on sparse reinforcement learning signals. We propose \textbf{S}elf-\textbf{E}volving \textbf{Search}, a Self-Evolving Search agent that improves online search behavior through three components, memory purification, atomic query training, and dense rewards. SE-Search follows a \textit{Think-Search-Memorize} strategy that retains salient evidence while filtering irrelevant content. Atomic query training promotes shorter and more diverse queries, improving evidence acquisition. Dense rewards provide fine-grained feedback that speeds training. Experiments on single-hop and multi-hop question answering benchmarks show that \texttt{SE-Search-3B} outperforms strong baselines, yielding a $10.8$ point absolute improvement and a $33.8\%$ relative gain over Search-R1.\footnote{We will make the code and model weights publicly available upon acceptance.}

cs.CL↗

arXiv2026

DynaWeb: Model-Based Reinforcement Learning of Web Agents

The development of autonomous web agents, powered by Large Language Models (LLMs) and reinforcement learning (RL), represents a significant step towards general-purpose AI assistants. However, training these agents is severely hampered by the challenges of interacting with the live internet, which is inefficient, costly, and fraught with risks. Model-based reinforcement learning (MBRL) offers a promising solution by learning a world model of the environment to enable simulated interaction. This paper introduces DynaWeb, a novel MBRL framework that trains web agents through interacting with a web world model trained to predict naturalistic web page representations given agent actions. This model serves as a synthetic web environment where an agent policy can dream by generating vast quantities of rollout action trajectories for efficient online reinforcement learning. Beyond free policy rollouts, DynaWeb incorporates real expert trajectories from training data, which are randomly interleaved with on-policy rollouts during training to improve stability and sample efficiency. Experiments conducted on the challenging WebArena and WebVoyager benchmarks demonstrate that DynaWeb consistently and significantly improves the performance of state-of-the-art open-source web agent models. Our findings establish the viability of training web agents through imagination, offering a scalable and efficient way to scale up online agentic RL.

cs.CL↗

arXiv2026

A radiation two-phase flow model for simulating plasma-liquid interactions

In laser-produced plasma (LPP) extreme ultraviolet (EUV) sources, deformation of a tin droplet into an optimal target shape is governed by its interaction with a pre-pulse laser-generated plasma. This interaction is mediated by a transient ablation pressure, whose complex spatio-temporal evolution remains experimentally inaccessible. Existing modeling approaches are limited: Empirical pressure-impulse models neglect dynamic plasma feedback, while advanced radiation-hydrodynamic codes often fail to resolve late-time droplet hydrodynamics. To bridge this gap, we propose a radiation two-phase flow model based on a diffuse interface methodology. The model integrates radiation hydrodynamics for the plasma with the Euler equations for a weakly compressible liquid, extending a five-equation diffuse interface formulation to incorporate radiation transport, thermal conduction, and ionization. This formulation enforces pressure and velocity equilibrium across the diffuse interface region, with closure models constructed to ensure correct jump conditions at interfaces and asymptotically recover the pure-phase equations in bulk regions. Then, we apply the model to simulate a benchmark pre-pulse scenario, where a 50 micron tin droplet is irradiated by a 10 ns laser pulse. The simulations capture the rapid plasma expansion and subsequent inertial flattening of the droplet into a thin, curved sheet over microsecond timescales. Notably, the model reproduces experimentally observed features (such as an axial jet) rarely replicated in prior simulations. Quantitative agreement with experimental data for sheet dimensions and velocity validates the approach. The proposed model self-consistently couples laser-plasma physics with compressible droplet dynamics, providing a powerful tool for fundamental studies of plasma-liquid interactions in LPP-EUV source optimization.

physics.plasm-ph↗

arXiv2026

Disco-RAG: Discourse-Aware Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has emerged as an important means of enhancing the performance of large language models (LLMs) in knowledge-intensive tasks. However, most existing RAG strategies treat retrieved passages in a flat and unstructured way, which prevents the model from capturing structural cues and constrains its ability to synthesize knowledge from dispersed evidence across documents. To overcome these limitations, we propose Disco-RAG, a discourse-aware framework that explicitly injects discourse signals into the generation process. Our method constructs intra-chunk discourse trees to capture local hierarchies and builds inter-chunk rhetorical graphs to model cross-passage coherence. These structures are jointly integrated into a planning blueprint that conditions the generation. Experiments on question answering and long-document summarization benchmarks show the efficacy of our approach. Disco-RAG achieves state-of-the-art results on the benchmarks without fine-tuning. These findings underscore the important role of discourse structure in advancing RAG systems.

cs.CL↗

arXiv2025

RoleRMBench & RoleRM: Towards Reward Modeling for Profile-Based Role Play in Dialogue Systems

Reward modeling has become a cornerstone of aligning large language models (LLMs) with human preferences. Yet, when extended to subjective and open-ended domains such as role play, existing reward models exhibit severe degradation, struggling to capture nuanced and persona-grounded human judgments. To address this gap, we introduce RoleRMBench, the first systematic benchmark for reward modeling in role-playing dialogue, covering seven fine-grained capabilities from narrative management to role consistency and engagement. Evaluation on RoleRMBench reveals large and consistent gaps between general-purpose reward models and human judgment, particularly in narrative and stylistic dimensions. We further propose RoleRM, a reward model trained with Continuous Implicit Preferences (CIP), which reformulates subjective evaluation as continuous consistent pairwise supervision under multiple structuring strategies. Comprehensive experiments show that RoleRM surpasses strong open- and closed-source reward models by over 24% on average, demonstrating substantial gains in narrative coherence and stylistic fidelity. Our findings highlight the importance of continuous preference representation and annotation consistency, establishing a foundation for subjective alignment in human-centered dialogue systems.

cs.CL↗

arXiv2025

SimDiff: Simpler Yet Better Diffusion Model for Time Series Point Forecasting

Diffusion models have recently shown promise in time series forecasting, particularly for probabilistic predictions. However, they often fail to achieve state-of-the-art point estimation performance compared to regression-based methods. This limitation stems from difficulties in providing sufficient contextual bias to track distribution shifts and in balancing output diversity with the stability and precision required for point forecasts. Existing diffusion-based approaches mainly focus on full-distribution modeling under probabilistic frameworks, often with likelihood maximization objectives, while paying little attention to dedicated strategies for high-accuracy point estimation. Moreover, other existing point prediction diffusion methods frequently rely on pre-trained or jointly trained mature models for contextual bias, sacrificing the generative flexibility of diffusion models. To address these challenges, we propose SimDiff, a single-stage, end-to-end framework. SimDiff employs a single unified Transformer network carefully tailored to serve as both denoiser and predictor, eliminating the need for external pre-trained or jointly trained regressors. It achieves state-of-the-art point estimation performance by leveraging intrinsic output diversity and improving mean squared error accuracy through multiple inference ensembling. Key innovations, including normalization independence and the median-of-means estimator, further enhance adaptability and stability. Extensive experiments demonstrate that SimDiff significantly outperforms existing methods in time series point forecasting.

cs.AI↗

arXiv2025

ARK: Answer-Centric Retriever Tuning via KG-augmented Curriculum Learning

Retrieval-Augmented Generation (RAG) has emerged as a powerful framework for knowledge-intensive tasks, yet its effectiveness in long-context scenarios is often bottlenecked by the retriever's inability to distinguish sparse yet crucial evidence. Standard retrievers, optimized for query-document similarity, frequently fail to align with the downstream goal of generating a precise answer. To bridge this gap, we propose a novel fine-tuning framework that optimizes the retriever for Answer Alignment. Specifically, we first identify high-quality positive chunks by evaluating their sufficiency to generate the correct answer. We then employ a curriculum-based contrastive learning scheme to fine-tune the retriever. This curriculum leverages LLM-constructed Knowledge Graphs (KGs) to generate augmented queries, which in turn mine progressively challenging hard negatives. This process trains the retriever to distinguish the answer-sufficient positive chunks from these nuanced distractors, enhancing its generalization. Extensive experiments on 10 datasets from the Ultradomain and LongBench benchmarks demonstrate that our fine-tuned retriever achieves state-of-the-art performance, improving 14.5\% over the base model without substantial architectural modifications and maintaining strong efficiency for long-context RAG. Our work presents a robust and effective methodology for building truly answer-centric retrievers. Source Code is available on https://github.com/valleysprings/ARK/.

cs.IR↗

arXiv2025

SciRAG: Adaptive, Citation-Aware, and Outline-Guided Retrieval and Synthesis for Scientific Literature

The accelerating growth of scientific publications has intensified the need for scalable, trustworthy systems to synthesize knowledge across diverse literature. While recent retrieval-augmented generation (RAG) methods have improved access to scientific information, they often overlook citation graph structure, adapt poorly to complex queries, and yield fragmented, hard-to-verify syntheses. We introduce SciRAG, an open-source framework for scientific literature exploration that addresses these gaps through three key innovations: (1) adaptive retrieval that flexibly alternates between sequential and parallel evidence gathering; (2) citation-aware symbolic reasoning that leverages citation graphs to organize and filter supporting documents; and (3) outline-guided synthesis that plans, critiques, and refines answers to ensure coherence and transparent attribution. Extensive experiments across multiple benchmarks such as QASA and ScholarQA demonstrate that SciRAG outperforms prior systems in factual accuracy and synthesis quality, establishing a new foundation for reliable, large-scale scientific knowledge aggregation.

cs.DL↗

arXiv2025

Aligning by Misaligning: Boundary-aware Curriculum Learning for Multimodal Alignment

Most multimodal models treat every negative pair alike, ignoring the ambiguous negatives that differ from the positive by only a small detail. We propose Boundary-Aware Curriculum with Local Attention (BACL), a lightweight add-on that turns these borderline cases into a curriculum signal. A Boundary-aware Negative Sampler gradually raises difficulty, while a Contrastive Local Attention loss highlights where the mismatch occurs. The two modules are fully differentiable and work with any off-the-shelf dual encoder. Theory predicts a fast O(1/n) error rate; practice shows up to +32% R@1 over CLIP and new SOTA on four large-scale benchmarks, all without extra labels.

cs.LG↗

arXiv2025

Improved phase field model for two-phase incompressible flows: Sharp interface limit, universal mobility and surface tension calculation

In this paper, we propose an improved phase field model for interface capturing in simulating two-phase incompressible flows. The model incorporates a second-order diffusion term, which utilizes a nonlinear coefficient to assess the degree of deviation of interface profile from its equilibrium state. In particular, we analyze the scale of the mobility in the model, to ensure that the model asymptotically approaches the sharp interface limit as the interface thickness approaches zero. For accurate calculations of surface tension, we introduce a generalized form of smoothed Dirac delta functions that can adjust the thickness of the tension layer, while strictly maintaining that its integral equals one, even when the interface profile is not in equilibrium. Furthermore, we theoretically demonstrate that the spontaneous shrinkage of under-resolved interface structures encountered in the Cahn-Hilliard phase field method does not occur in the improved phase field model. Through various numerical experiments, we determine the range of the optimal mobility, confirm the theoretical analysis of the improved phase field model, verify its convergence, and examine the performance of different surface tension models. The numerical experiments include Rayleigh-Taylor instability, axisymmetric rising bubbles, droplet migration due to the Marangoni effect, partial coalescence of a droplet into a pool, and deformation of three-dimensional droplet in shear flow. In all these cases, numerical results are validated against experimental data and/or theoretical predictions. Moreover, the recommended range of dimensionless mobility has been shown to be universal, as it can be effectively applied to the simulations of a wide range of two-phase flows and exhibits excellent performance.

physics.flu-dyn↗

arXiv2022

Viscous influences on impulsively generated focused jets

Impulsively generated focused jets play a significant role in various applications, including inkjet printing, needle-free drug delivery, and microfluidic devices. As the demand for generating jets and droplets from medium to highly viscous liquids increases, understanding the role of viscosity in jetting dynamics becomes crucial. While previous studies have examined the viscous effects on walls, the impact on free surfaces has not been thoroughly understood. This study aims to bridge this gap by integrating experiments with numerical simulations to investigate the viscous effects on focused jet formation. We demonstrate that mass and momentum transfer along the tangential direction of the free surface contribute to focused jet formation, and viscosity plays a key role in this transfer process. The viscosity-induced diffusion of the shear flow and vorticity near the free surface reduces the jet speed. Based on experimental observations and simulation results, we propose an equation to predict the viscous jet velocity. These findings offer new perspectives on viscous interface dynamics in advanced manufacturing and biomedical applications.

physics.flu-dyn↗

arXiv2021

The Inverse-Square Law Force between Vapor-Mediated Droplets

In 1687, Sir Issac Newton published The Mathematical Principles of Natural Philosophy in which the law of universal gravitation was derived. It is the first inverse-square law discovered in nature, combined with Coulomb's law in 1785, the two famous inverse-square laws become part of the foundation of physics. Why does nature prefer inverse-square laws over the laws of other forms? The question is still arousing broad discussion, and it is an important topic in physics. So far, the origin of inverse-square law is still under exploration although from the point of reductionism, the law of universal gravitation can be treated as the approximation of Einstein's general relativity under weak gravitation, and Coulomb's law could be derived from quantum electrodynamics. Here we discover a new inverse-square law between evaporating droplets deposited on a high energy solid substrate. For binary droplets, we show that the evaporation from a source droplet will create a surface tension gradient in the precursor film of a target droplet, resulting in a long-range inverse-square law force acting on the target droplet, and that the inverse proportion decay of the source vapor concentration in the space essentially contributes to the inverse-square form of the force. Furthermore, the inverse-square law force here is shown to hold for all experimental parameters tested, and other systems such as pure-liquid-droplet system and thermocapillary system, and it satisfies the superposition principle, not only suggesting exciting directions for future droplet research and applications, but also benefiting understanding of nature's predilection for inverse-square law.

physics.flu-dyn↗