Searcharxiv⌕ Search

arXiv subjects

Kun Xu

Publications and source records attributed to Kun Xu.

At least 55 records · Page 3Linked to original sources
arXiv2026

On the Applicability of the Gas-Kinetic Scheme with Kinetic Boundary Conditions for Near-Continuum Hypersonic Flows

Rarefied gas effects are of critical importance for the aerodynamic performance of hypersonic vehicles operating at high altitudes. In these scenarios, conventional computational fluid dynamics (CFD) solvers break down as the linear constitutive relations underlying the Navier-Stokes equations cease to be valid. Based on direct modeling, the unified gas-kinetic scheme (UGKS) and the unified gas-kinetic wave-particle (UGKWP) method successfully capture non-equilibrium physics across all Knudsen numbers, yet they incur substantially higher computational costs than continuum solvers. Within the same kinetic framework, the gas-kinetic scheme (GKS) employs the Chapman-Enskog expansion for near-equilibrium flow physics and adopts the same kinetic boundary conditions as UGKS and UGKWP. This formulation naturally permits velocity slip and temperature jump, thereby extending the applicability of GKS into the slip and transitional regimes. By utilizing this natural kinetic slip boundary condition, the GKS provides a more physically faithful representation of non-equilibrium wall interactions than conventional CFD solvers equipped with Maxwell-type slip conditions, ultimately yielding more accurate aerodynamic predictions. To determine the applicability of the GKS in near-continuum flow regimes, we first examine a simple circular cylinder geometry, comparing surface quantities and distribution functions in detail. Furthermore, we investigate a 9°blunted cone, a 70° blunted cone with a cylindrical sting, and the Apollo 6 command module. This analysis focuses on integrated aerodynamic predictions, which are validated against experimental data, Direct Simulation Monte Carlo (DSMC) simulations, and other kinetic methods.

physics.flu-dyn↗
arXiv2026

A unified gas-kinetic wave-particle method for multiscale binary-species gas mixtures

This paper presents a unified gas-kinetic wave-particle (UGKWP) method for simulating multiscale binary-species gas mixtures. Benefiting from direct modeling in a discretized space, the UGKWP method enables the automatic decomposition of the gas distribution function into analytical hydrodynamic waves and discrete particles, which respectively describe its near-equilibrium and non-equilibrium parts. This approach offers significant advantages for simulating various multiscale physical phenomena, such as hypersonic flows, plasma transport, and radiation transport. In this study, we employ the model proposed by Groppi et al. [EPL, 96 (2011) 64002] to calculate the macroscopic velocity and temperature of the local target equilibrium distribution function, thereby recovering the correct viscosity and diffusion coefficients in the continuum flow regime. To address the heat conduction coefficient, the Shakhov model is incorporated to correct the Prandtl number. Diffusion effects are accounted for not only in the source term via an operator-splitting method, but also in the flux evolution through the characteristic integral solution, while strictly maintaining consistency between the wave and particle descriptions. Furthermore, the microscopic model for high-speed particles is improved by utilizing a physically corrected collision time to determine their free-transport time. Through a series of numerical tests spanning the continuum to rarefied regimes, the proposed UGKWP method is shown to accurately capture the differences in velocity and temperature between different species. Notably, for hypersonic flows, the predicted wall pressure, shear stress, and heat flux coefficients agree well with DSMC results.

physics.flu-dyn↗
arXiv2026

BitLM: Unlocking Multi-Token Language Generation with Bitwise Continuous Diffusion

Autoregressive language models generate text one token at a time, yet natural language is inherently structured in multi-token units, including phrases, n-grams, and collocations that carry meaning jointly. This one-token bottleneck limits both the expressiveness of the model during pre-training and its throughput at inference time. Existing remedies such as speculative decoding or diffusion-based language models either leave the underlying bottleneck intact or sacrifice the causal structure essential to language modeling. We propose BitLM, a language model that represents each token as a fixed-length binary code and employs a lightweight diffusion head to denoise multiple tokens in parallel within each block. Crucially, BitLM preserves left-to-right causal attention across blocks while making joint lexical decisions within each block, combining the reliability of autoregressive modeling with the parallelism of iterative refinement. By replacing the large-vocabulary softmax with bitwise denoising, BitLM reframes token generation as iterative commitment in a compact binary space, enabling more efficient pre-training and substantially faster inference without altering the causal foundation that makes language models effective. Our results demonstrate that the one-token-at-a-time paradigm is not a fundamental requirement but an interface choice, and that changing it can yield a stronger and faster language model. We hope BitLM points toward a promising direction for next-generation language model architectures.

cs.CL↗
arXiv2026

Dynamically Reconfigurable Optical Skyrmions Enabled by a Silicon Microring Optical Phased Array for Robust Free-Space Communication

Optical skyrmions offer a robust vectorial information degree of freedom for free-space communication, but practical deployment requires a compact platform capable of active topological reconfiguration. Here, we propose a silicon microring-resonator optical phased array that integrates spin-selective emission and programmable phase control on a single chip. Optimized inner- and outer-grating microring emitters provide decoupled LCP and RCP radiation bases with polarization fractions of 90.27% and 91.40%, enabling active switching between Néel-type and Bloch-type skyrmions, while dynamically tuning the skyrmion number across Nsk =-1.914 to 1.918. Using these programmable topological states, a 4-symbol free-space communication link is constructed and compared with ideal LG-OAM encoding under Kolmogorov turbulence. The skyrmion-encoded link maintains a lower symbol error rate over a broader turbulence range, demonstrating that topological observables are more robust than scalar OAM modes. These results establish actively reconfigurable optical skyrmions as compact, programmable, and turbulence-tolerant information carriers for next-generation free-space optical communication.

physics.optics↗
arXiv2026

CAMEL: Confidence-Gated Reflection for Reward Modeling

Reward models play a fundamental role in aligning large language models with human preferences. Existing methods predominantly follow two paradigms: scalar discriminative preference models, which are efficient but lack interpretability, and generative judging models, which offer richer reasoning at the cost of higher computational overhead. We observe that the log-probability margin between verdict tokens strongly correlates with prediction correctness, providing a reliable proxy for instance difficulty without additional inference cost. Building on this insight, we propose CAMEL, a confidence-gated reflection framework that performs a lightweight single-token preference decision first and selectively invokes reflection only for low-confidence instances. To induce effective self-correction, we train the model via reinforcement learning with counterfactual prefix augmentation, which exposes the model to diverse initial verdicts and encourages genuine revision. Empirically, CAMEL achieves state-of-the-art performance on three widely used reward-model benchmarks with 82.9% average accuracy, surpassing the best prior model by 3.2% and outperforming 70B-parameter models using only 14B parameters, while establishing a strictly better accuracy-efficiency Pareto frontier.

cs.CL↗
arXiv2026

A Hybrid Gas-Kinetic Scheme and Discrete Velocity Method for Continuum and Rarefied Flows

The gas-kinetic scheme (GKS) provides high computational efficiency and accuracy for continuum flow simulations but is unable to reliably capture rarefaction effects. In contrast, although the discrete velocity method (DVM) is better suited for rarefied flows, it exhibits reduced accuracy and slow convergence when applied to continuum regimes. To overcome these limitations, this work proposes a hybrid GKS-DVM method that integrates the strengths of both approaches. The hybrid approach balances the equilibrium distribution function in GKS with the upwind-reconstructed non-equilibrium distribution function in DVM through a numerical collision time. This balancing strategy ensures to recover Navier-Stokes solutions in the continuum limit (asymptotic preserving), while naturally capturing free molecular flows in the rarefied limit. Moreover, the introduction of a numerical collision time significantly enhances robustness in shock capturing for continuum flow applications. To further reduce computational cost of the hybrid approach, several adaptive strategies based on the local Knudsen number and Mach number have been proposed. The effectiveness and accuracy of the proposed hybrid method are systematically assessed through four representative test cases: a flat-plate boundary layer, a lid-driven cavity flow, shock structures, and flow past a semi-cylinder. The first case is subjected to continuum conditions, while the latter two span a broad range of Knudsen numbers. The results demonstrate that the proposed method achieves high solution accuracy and computational efficiency across both continuum and rarefied flow regimes.

physics.flu-dyn↗
arXiv2026

Detecting Malicious Concepts without Image Generation in AI-Generated Content (AIGC)

The task of text-to-image generation has achieved tremendous success in practice, with emerging concept generation models capable of producing highly personalized and customized content. Fervor for concept generation is increasing rapidly among users, and platforms for concept sharing have sprung up. The concept owners may upload malicious concepts and disguise them with non-malicious text descriptions and example images to deceive users into downloading and generating malicious content. The platform needs a quick method to determine whether a concept is malicious to prevent the spread of malicious concepts. However, simply relying on concept image generation to judge whether a concept is malicious requires time and computational resources. Especially, as the number of concepts uploaded and downloaded on the platform continues to increase, this approach becomes impractical and poses a risk of generating malicious content. In this paper, we propose Concept QuickLook, the first systematic work to incorporate malicious concept detection into research, which performs detection based solely on concept files without generating any images. We define malicious concepts and design two operational modes for detection: concept matching and fuzzy detection. Extensive experiments demonstrate that the proposed Concept QuickLook can detect malicious concepts and demonstrate practicality in concept sharing platforms. We also design robustness experiments to further validate the effectiveness of the solution. We hope this work can initiate malicious concept detection tasks and provide some inspiration.

cs.CR↗
arXiv2026

Mass spectra of charged mesons and the quenching of vector meson condensation via exact phase-space diagonalization

We investigate the dynamics and mass spectra of charged pseudoscalar ($π^+$) and vector ($ρ^+$) mesons in a background magnetic field at finite temperature using the two-flavor Nambu-Jona--Lasinio (NJL) model. By employing a quark propagator that isolates the Schwinger phase from its Landau level expansion, we formulate an exact non-commutative phase-space framework utilizing the Wigner-Weyl transform and the Moyal star product. This approach enables the algebraic diagonalization of the Bethe-Salpeter equations for composite states with asymmetric fractional constituent charges. For the pseudoscalar channel, we analytically verify the exact cancellation between the dynamical random phase approximation spatial sum rules and the vacuum gap equation. This identity preserves the generalized Goldstone theorem, causing the $π^+$ pole mass to strictly track the kinematic zero-point energy drift at order of $eB$. In the vector channel, our full phase-space evaluation reveals that the Zeeman spin-splitting emerges dynamically from microscopic threshold truncations governed by the chiral Dirac algebra. Notably, we find that the tachyonic instability of the spin-aligned $ρ^+$ state is quenched. The magnetic catalysis of the chiral condensate drives the continuum threshold ($2M$) upwards, overtaking the Zeeman attraction and preventing vector meson condensation within this mean-field framework. Furthermore, finite-temperature evaluations show a monotonic thermal suppression of the meson masses driven by Pauli blocking, yet all modes remain bound without undergoing Mott dissociation prior to chiral symmetry restoration.

hep-ph↗
arXiv2026

A paradox of the Navier-Stokes turbulence

The Navier-Stokes (NS) equations as a turbulence model have been widely applied in lots of fields. The NS equations contain such a fundamental assumption that all small physical/artificial disturbances could be neglected. Is this assumption correct? In this paper a two-dimensional Rayleigh-Bénard convection governed by the NS equations is predicted by traditional direct numerical simulation (DNS) using double precision arithmetic and a range of different time-steps. It is found that the final flow type tends either to vortical flow or zonal flow, whose statistics are completely different. Notably, these two flow types frequently alternate as the time-step is reduced to a very small value, suggesting that the time-step corresponding to each turbulent flow type should be densely distributed. Thus, stochastic numerical noise exerts a huge influence on the final flow type and statistics of numerically simulated NS turbulence because the time-step has a close relationship with numerical noise. This clearly indicates that small disturbances have significant influences on the NS turbulence, which therefore should not be neglected. This leads to a logical paradox for the NS turbulence, which is a great challenge for us, although a paradox often leads to some significant breakthroughs.

physics.flu-dyn↗
arXiv2026

Revisit viscous shock tube at low Reynolds number

The viscous shock tube is a canonical test case for assessing Navier-Stokes (NS) solvers in the continuum-flow regime, widely used to validate numerical accuracy and probe flow physics. It features a rich set of interacting structures-shock and rarefaction waves, contact discontinuities, boundary layers, and their coupling-spanning multiple spatial and temporal scales. However, NS-based modeling, which presumes near-equilibrium behavior, may fail to capture important non-equilibrium effects even in nominally continuum conditions. This study investigates the viscous shock tube at low Reynolds numbers and demonstrates the presence of non-equilibrium phenomena within the conventional continuum regime. To obtain physically consistent solutions across scales, we employ the unified gas-kinetic scheme (UGKS) and compare its results with NS solutions computed using the gas-kinetic scheme (GKS). Discrepancies between UGKS and GKS solutions reveal pronounced non-equilibrium effects in regions where shock waves interact with boundary layers. For continuum flows at high Mach and low Reynolds numbers, such multiscale non-equilibrium transport becomes important, underscoring the need for multiscale methods in analysis and prediction.

physics.flu-dyn↗
arXiv2026

A Discrete Adjoint Gas-Kinetic Scheme for Aerodynamic Shape Optimization in Turbulent Continuum Flows

This study presents an efficient and accurate discrete adjoint gas-kinetic scheme (GKS) for sensitivity analysis and aerodynamic shape optimization in continuum flow regimes. Developed using the backward mode of algorithmic differentiation (AD), the adjoint solver is rigorously verified against a duality-preserving linearized GKS solver generated via forward-mode AD. The robustness and practical effectiveness of the solver are evaluated through three benchmark cases: the inverse design of turbine blades, lift-to-drag ratio enhancement, and shock-strength reduction for a NACA 0012 airfoil. To capture realistic flow physics, fully turbulent optimizations are conducted using the one-equation Spalart--Allmaras (SA) model. Numerical results demonstrate excellent agreement between the discrete adjoint and linearized solvers, exhibiting matching sensitivity convergence behaviors, identical asymptotic residual decay rates, and negligible discrepancies in final sensitivity predictions. Furthermore, the optimization studies confirm that targeted design objectives are consistently achieved within a limited number of design cycles, highlighting the solver's computational efficiency, accuracy, and suitability for complex aerodynamic geometries.

physics.flu-dyn↗
arXiv2026

Unified Gas-Kinetic Scheme for Unsteady Multiscale Flows with Moving Boundaries

Simulating multiscale flows with moving boundaries, such as hypersonic multi-body separation and flows in micro-electro-mechanical systems (MEMS), requires robust numerical methods that couple mesh deformation with complex flow physics. This paper presents a hybrid overlapping moving-mesh technique developed within the unified gas-kinetic scheme (UGKS). To mitigate the Courant-Friedrichs-Lewy (CFL) constraint, we extend the implicit unsteady UGKS solver to support moving meshes, incorporating memory-efficient data handling and parallel computing optimizations to maximize computational efficiency. Validated against hypersonic multi-body separation and thermal rarefied MEMS flows, the proposed scheme accurately resolves complex, dynamic multiscale phenomena. The results confirm that this robust and efficient method provides a highly reliable tool for modeling dynamic flow interactions in complex geometric configurations.

physics.flu-dyn↗
arXiv2026

PAC in DESI. I. Galaxy Stellar Mass Function into the $10^{6}{\rm M}_{\odot}$ Frontier

The Photometric Objects Around Cosmic Webs (PAC) method integrates cosmological photometric and spectroscopic surveys, offering valuable insights into galaxy formation. PAC measures the excess surface density of photometric objects, $\bar{n}_2w_{\rm{p}}$, with specific physical properties around spectroscopic tracers. In this study, we improve the PAC method to make it more rigorous and eliminate the need for redshift bins. We apply the enhanced PAC method to the DESI Y1 BGS Bright spectroscopic sample and the deep DECaLS photometric sample, obtaining $\bar{n}_2w_{\rm{p}}$ measurements across the complete stellar mass range, from $10^{5.3}{\rm M}_{\odot}$ to $10^{11.5}{\rm M}_{\odot}$ for blue galaxies, and from $10^{6.3}{\rm M}_{\odot}$ to $10^{11.9}{\rm M}_{\odot}$ for red galaxies. We combine $\bar{n}_2w_{\rm{p}}$ with $w_{\rm{p}}$ measurements from the BGS sample, which is not necessarily complete in stellar mass. Assuming that galaxy bias is primarily determined by stellar mass and colour, we derive the galaxy stellar mass functions (GSMFs) down to $10^{5.3}{\rm M}_{\odot}$ for blue galaxies and $10^{6.3}{\rm M}_{\odot}$ for red galaxies, while also setting lower limits for smaller masses. The blue and red GSMFs are well described by single and double Schechter functions, respectively, with low-mass end slopes of $α_{\rm{blue}}=-1.54^{+0.02}_{-0.02}$ and $α_{\rm{red}}=-2.50^{+0.08}_{-0.08}$, resulting in the dominance of red galaxies below $10^{7.6}{\rm M}_{\odot}$. Stage-IV cosmological photometric surveys, capable of reaching 2-3 magnitudes deeper than DECaLS, present an opportunity to explore the entire galaxy population in the local universe with PAC. This advancement allows us to address critical questions regarding the nature of dark matter, the physics of reionization, and the formation of dwarf galaxies.

astro-ph.GA↗
arXiv2026

ResAdapt: Adaptive Resolution for Efficient Multimodal Reasoning

Multimodal Large Language Models (MLLMs) achieve stronger visual understanding by scaling input fidelity, yet the resulting visual token growth makes jointly sustaining high spatial resolution and long temporal context prohibitive. We argue that the bottleneck lies not in how post-encoding representations are compressed but in the volume of pixels the encoder receives, and address it with ResAdapt, an Input-side adaptation framework that learns how much visual budget each frame should receive before encoding. ResAdapt couples a lightweight Allocator with an unchanged MLLM backbone, so the backbone retains its native visual-token interface while receiving an operator-transformed input. We formulate allocation as a contextual bandit and train the Allocator with Cost-Aware Policy Optimization (CAPO), which converts sparse rollout feedback into a stable accuracy-cost learning signal. Across budget-controlled video QA, temporal grounding, and image reasoning tasks, ResAdapt improves low-budget operating points and often lies on or near the efficiency-accuracy frontier, with the clearest gains on reasoning-intensive benchmarks under aggressive compression. Notably, ResAdapt supports up to 16x more frames at the same visual budget while delivering over 15% performance gain. Code is available at https://github.com/Xnhyacinth/ResAdapt.

cs.CV↗
arXiv2026

Adaptive finite volume-particle method for free surface flows

This study proposes a novel adaptive finite volume-particle method (AFVPM) for accurate and efficient free surface flow simulations. The proposed AFVPM synergistically combines the Eulerian finite volume method (FVM) on unstructured meshes with the Lagrangian smoothed particle hydrodynamics (SPH) approach. Specifically, the mesh-based FVM is employed in the bulk flow regions to leverage its computational efficiency and numerical accuracy, while a weakly compressible SPH formulation is applied in the vicinity of the interface to maintain robust free-surface tracking capabilities. A key innovation of this framework is a block-based dynamic and adaptive conversion strategy between Eulerian mesh regions and Lagrangian particle regions and a buffer region-based cell-particle algorithm is designed to ensure seamless data communication across the Eulerian mesh-Lagrangian particle interface. Furthermore, isothermal gas-kinetic scheme (GKS) incorporating gravitational effects is utilized to calculate the fluxes in the mesh regions. The performance and reliability of the proposed AFVPM are validated through a series of benchmark cases that involve complex free surface phenomena. Numerical results demonstrate that AFVPM achieves superior accuracy and efficiency compared to full SPH approaches.

physics.comp-ph↗
arXiv2026

A High-Order Finite Volume GENO Scheme with Implicit Time Integration for Three-Temperature Radiation Diffusion Equations

This study presents a high-order finite volume scheme capable of large time-step integration for three-temperature radiation diffusion (3TRD) equations, where conservation is naturally achieved through energy update. To handle local large gradients and discontinuities in temperature, a central generalized ENO (GENO) reconstruction is developed for diffusion systems, which achieves essentially non-oscillatory reconstruction for discontinuous solutions. Compared to conventional nonlinear reconstruction methods, its most distinctive feature is the central-type symmetric sub-stencils, which ensure consistency between the numerics and the isotropic nature of thermal diffusion. Additionally, the central GENO method provides smooth states of temperature and temperature gradient at interfaces, facilitating the evaluation of numerical fluxes. Furthermore, interface flux evaluation for cases with discontinuous physical property parameters is modeled. To address the extremely small time-step issue caused by stiff diffusion and source terms, a dual-time-stepping method based on implicit time discretization is developed for the first time in 3TRD systems, with the advantage of decoupling temporal discretization from complex nonlinear spatial discretization. A series of numerical examples validates the high accuracy, physical property preservation, strong robustness, and large time-step integration capability of the present high-order central GENO scheme.

math.NA↗
arXiv2026

UniWeTok: An Unified Binary Tokenizer with Codebook Size $\mathit{2^{128}}$ for Unified Multimodal Large Language Model

Unified Multimodal Large Language Models (MLLMs) require a visual representation that simultaneously supports high-fidelity reconstruction, complex semantic extraction, and generative suitability. However, existing visual tokenizers typically struggle to satisfy these conflicting objectives within a single framework. In this paper, we introduce UniWeTok, a unified discrete tokenizer designed to bridge this gap using a massive binary codebook ($\mathit{2^{128}}$). For training framework, we introduce Pre-Post Distillation and a Generative-Aware Prior to enhance the semantic extraction and generative prior of the discrete tokens. In terms of model architecture, we propose a convolution-attention hybrid architecture with the SigLu activation function. SigLu activation not only bounds the encoder output and stabilizes the semantic distillation process but also effectively addresses the optimization conflict between token entropy loss and commitment loss. We further propose a three-stage training framework designed to enhance UniWeTok's adaptability cross various image resolutions and perception-sensitive scenarios, such as those involving human faces and textual content. On ImageNet, UniWeTok achieves state-of-the-art image generation performance (FID: UniWeTok 1.38 vs. REPA 1.42) while requiring a remarkably low training compute (Training Tokens: UniWeTok 33B vs. REPA 262B). On general-domain, UniWeTok demonstrates highly competitive capabilities across a broad range of tasks, including multimodal understanding, image generation (DPG Score: UniWeTok 86.63 vs. FLUX.1 [Dev] 83.84), and editing (GEdit Overall Score: UniWeTok 5.09 vs. OmniGen 5.06). We release code and models to facilitate community exploration of unified tokenizer and MLLM.

cs.CV↗
arXiv2026

CUBE2: A Parallel $N$-Body Simulation Code for Scalability, Accuracy, and Memory Efficiency

$N$-body simulation serves as a critical method for modeling cosmic evolution and poses a significant challenge in high-performance computing. We present CUBE2, an open-source cosmological $N$-body code emphasizing memory efficiency, computational performance, scalability and precision. The core of its algorithm utilizes multi-level Particle-Mesh (PM) method to solve the Poisson equation for matter distribution, leveraging the well-optimized Fast Fourier Transform (FFT) for computational efficiency. Precision is ensured by the optimized Green's function that seamlessly bridges gravitational interactions between multi-level PM and Particle-Particle (PP) calculations. The program design enhances per-core/node efficiency in processing $N$-body particles, while the Information Optimized Storage (IOS) addresses memory constraints for large particle counts. Using CUBE2, we run two cosmological simulations with particle counts of $6144^3$ on the Advanced Computing East China Sub-center (ACECS) to test performance and accuracy.

astro-ph.IM↗