arXiv subjects

Xu Qian

Publications and source records attributed to Xu Qian.

At least 19 recordsLinked to original sources

arXiv2026

OBLIVION: Workflow-Level Operational Skill Unlearning for Deployed Agents

Large language model agents are becoming operational interfaces to files, memories, registries, and external tools. This deployment shift creates a new skill revocation problem: after a skill is removed from an explicit registry, an agent may still reconstruct it from residual carriers such as archives, transcripts, schemas, or memory entries. We study this problem as operational skill unlearning, where the goal is not parameter-level forgetting, but preventing a deployed agent from rebuilding a revoked skill through primitive tools. We introduce OBLIVION, a controlled benchmark and defense harness for revoked-skill resurrection. OBLIVION models each episode as a source-to-sink workflow, applies Cross-Surface Coherent Erasure to reduce residual carriers, and uses frozen workflow remediation near dangerous sinks. On the locked 88 attack episodes, the no-defense arm reaches formal attack success rate 1.0. OBLIVION reduces the rate to 0.114 and impact-weighted exposure to 0.115 while keeping locked utility at 1.0 and benign block rate at 0. In a separate skill-attack-derived sandbox, OBLIVION reduces attack success from 1.0 to 0.2 and impact-weighted exposure from 1.0 to 0.213 while preserving all utility controls. These results support workflow-level evaluation beyond checking explicit skill entries.

cs.AI↗

arXiv2026

Onsager-variational-principle-based Lattice Boltzmann Model For Three-phase Dielectric Fluid Flows

Multiphase electrohydrodynamic (EHD) flows play a crucial role in various engineering applications. However, existing numerical studies on three-phase electrohydrodynamic systems predominantly rely on phenomenological models, often neglecting thermodynamic consistency and critical surface charge convection mechanisms. To address these fundamental gaps, this paper proposes a thermodynamically consistent three-phase EHD model derived strictly from the Onsager variational principle. This theoretical framework intrinsically guarantees thermodynamic consistency and accurately captures complex multiphysics interactions without requiring a priori assumptions. Furthermore, a mesoscopic lattice Boltzmann method is developed to solve the proposed model, enabling the natural capture of interfacial evolution and charge transport. The accuracy of the numerical framework are rigorously validated against several benchmark cases, including electroosmotic flow in microchannels, the spreading of a three-phase liquid lens, the equilibrium of static compound droplet, and the deformation of compound droplet under uniform electric field. Using this validated framework, we investigate EHD applications, specifically simulating the complex dynamics of double droplet coalescence and separation under electric field, as well as the behavior of droplets subjected to combined EHD and shear flow. Overall, this work provides a robust, thermodynamically reliable numerical tool for exploring the highly nonlinear behaviors of multiphase EHD systems.

physics.flu-dyn↗

arXiv2026

Augmented Lagrangian preconditioning for a simplified Ericksen--Leslie model of nematic liquid crystals

The numerical solution of the simplified Ericksen--Leslie model for nematic liquid crystals is challenging because the flow and director equations are strongly coupled and because incompressibility and the unit-length condition must be enforced simultaneously. A Lagrange multiplier formulation avoids a small Ginzburg--Landau parameter, but the Newton systems have a double saddle-point structure. We develop an augmented Lagrangian block preconditioner in which both constraints are augmented while their discrete enforcement remains multiplier based. After finite element discretization and backward Euler time integration, the Newton increments are grouped into velocity--director and pressure-multiplier variables. A block-diagonal approximation of the coupled velocity-director block then leads to separate, physically scaled approximations of the pressure and director-multiplier Schur complements. Manufactured-solution tests show the expected spatial accuracy and first-order temporal convergence for the primary variables; the multiplier error reaches a spatial-error floor on the fixed mesh used in the temporal study. In the reported parameter ranges, the outer FGMRES iteration counts are nearly mesh independent, remain stable under time-step and viscosity variation, and improve as the augmentation parameters increase. A smooth benchmark also exhibits monotone decay of the computed total energy.

math.NA↗

arXiv2026

Penalty-scaling effects in nonsymmetric interior-penalty DG discretizations of viscous rotating shallow-water equations

We investigate how the scaling of the interior-penalty parameter affects nonsymmetric interior-penalty Galerkin (NIPG) discretizations of the viscous rotating shallow-water equations in geopotential variables. The hyperbolic terms are approximated by a local Lax--Friedrichs flux, while viscosity acts on the momentum variables through a penalty law $\mu_e=\sigma h_e^{-\beta}$. The standard choice $\beta=1$ and the super-penalized choice $\beta=3$ are compared with a symmetric interior-penalty Galerkin reference. For the diffusion form, we establish consistency, continuity for $\beta\ge 1$, and an exact coercivity identity in the momentum DG seminorm. Manufactured-solution tests show that super-penalization can recover the expected momentum $L^2$ accuracy, whereas the coupled geopotential variable need not exhibit the same improvement. Rotating and topography-aware tests further show that the standard scaling generally gives the better accuracy-cost compromise for the explicit implementation considered here.

math.NA↗

arXiv2026

SAGE: Signal-Amplified Guided Embeddings for LLM-based Vulnerability Detection

Software vulnerabilities are a primary threat to modern infrastructure. While static analysis and Graph Neural Networks have long served as the foundation for vulnerability detection, the emergence of Large Language Models (LLMs) has introduced a transformative paradigm driven by superior semantic reasoning and cross-environment generalization. However, in the context of LLM-based vulnerability detection, we identify a fundamental bottleneck in these models termed \textbf{Signal Submersion}: a state where features related to vulnerability are activated internally but numerically overwhelmed by dominant functional semantics. To address this, we propose \textbf{SAGE} (\textbf{S}ignal-\textbf{A}mplified \textbf{G}uided \textbf{E}mbeddings), a framework that shifts from passive signal submersion to active signal recovery. SAGE integrates task-conditional Sparse Autoencoders (SAEs) to isolate and amplify these faint vulnerability signals. Extensive evaluations on BigVul, PrimeVul, and PreciseBugs demonstrate that SAGE achieves state-of-the-art performance. Notably, SAGE mitigates Signal Submersion by increasing the internal Signal-to-Noise Ratio (SNR) by 12.7$\times$ via sparse manifold projection. This mechanistic intervention enables a 7B model to achieve up to 318\% Matthews Correlation Coefficient (MCC) gains on unseen distributions and a 319\% gain on classic datasets. By maintaining robust performance across 13 programming languages and outperforming 34B baselines, SAGE establishes a more efficient and scalable path to software security than simple parameter scaling.

cs.CR↗

arXiv2024

A unified convergence analysis framework of the energy-stable ETDRK3 schemes for the No-slope-selection thin film model

This paper establishes a unified framework for the space-time convergence analysis of the energy-stable third-order accurate exponential time differencing Runge-Kutta schemes. By employing Fourier pseudo-spectral discretization in space and the inner product technique, we derive a rigorous Fourier eigenvalue analysis, which provides a detailed optimal convergence rate and error estimate. The primary challenge is addressing the complex nonlinear terms in the NSS equation. Fortunately, this challenge could be resolved through careful eigenvalue bound estimates for various operators.

math.NA↗

arXiv2024

An exponential-free Runge--Kutta framework for developing third-order unconditionally energy stable schemes for the Cahn--Hilliard equation

In this work, we develop a class of up to third-order energy-stable schemes for the Cahn--Hilliard equation. Building on Lawson's integrating factor Runge--Kutta method, which is widely used for stiff semilinear equations, we discuss its limitations, such as the inability to preserve the equilibrium state and the oversmoothing of interfacial layers in the solution's profile because of the exponential damping effects. To overcome this drawback, we approximate the exponential term using a class of sophisticated Taylor polynomials, leading to a novel Runge--Kutta framework called exponential-free Runge--Kutta. By incorporating stabilization techniques, we analyze the energy stability of the proposed schemes and demonstrate that they preserve the original energy dissipation without time-step restrictions. Furthermore, we perform an analysis of the linear stability and establish an error estimate in the $\ell^2$ norm. A series of numerical experiments validate the high-order accuracy, mass conservation, and energy dissipation of our schemes.

math.NA↗

arXiv2024

High-order and Mass-conservative Regularized Implicit-explicit relaxation Runge-Kutta methods for the logarithmic Schr\"{o}dinger equation

The non-differentiability of the singular nonlinearity (such as $f=\ln|u|^2$) at $u=0$ presents significant challenges in devising accurate and efficient numerical schemes for the logarithmic Schr\"{o}dinger equation (LogSE). To address this singularity, we propose an energy regularization technique for the LogSE. For the regularized model, we utilize Implicit-Explicit Relaxation Runge-Kutta methods, which are linearly implicit, high-order, and mass-conserving for temporal discretization, in conjunction with the Fourier pseudo-spectral method in space. Ultimately, numerical results are presented to validate the efficiency of the proposed methods.

math.NA↗

arXiv2024

Randomized Radial Basis Function Neural Network for Solving Multiscale Elliptic Equations

To overcome these obstacles and improve computational accuracy and efficiency, this paper presents the Randomized Radial Basis Function Neural Network (RRNN), an innovative approach explicitly crafted for solving multiscale elliptic equations. The RRNN method commences by decomposing the computational domain into non-overlapping subdomains. Within each subdomain, the solution to the localized subproblem is approximated by a randomized radial basis function neural network with a Gaussian kernel. This network is distinguished by the random assignment of width and center coefficients for its activation functions, thereby rendering the training process focused solely on determining the weight coefficients of the output layer. For each subproblem, similar to the Petrov-Galerkin finite element method, a linear system will be formulated on the foundation of a weak formulation. Subsequently, a selection of collocation points is stochastically sampled at the boundaries of the subdomain, ensuring satisfying $C^0$ and $C^1$ continuity and boundary conditions to couple these localized solutions. The network is ultimately trained using the least squares method to ascertain the output layer weights. To validate the RRNN method's effectiveness, an extensive array of numerical experiments has been executed and the results demonstrate that the proposed method can improve the accuracy and efficiency well.

math.NA↗

arXiv2023

SPFNO: Spectral operator learning for PDEs with Dirichlet and Neumann boundary conditions

Neural operators have been validated as promising deep surrogate models for solving partial differential equations (PDEs). Despite the critical role of boundary conditions in PDEs, however, only a limited number of neural operators robustly enforce these conditions. In this paper we introduce semi-periodic Fourier neural operator (SPFNO), a novel spectral operator learning method, to learn the target operators of PDEs with non-periodic BCs. This method extends our previous work (arXiv:2206.12698), which showed significant improvements by employing enhanced neural operators that precisely satisfy the boundary conditions. However, the previous work is associated with Gaussian grids, restricting comprehensive comparisons across most public datasets. Additionally, we present numerical results for various PDEs such as the viscous Burgers' equation, Darcy flow, incompressible pipe flow, and coupled reactiondiffusion equations. These results demonstrate the computational efficiency, resolution invariant property, and BC-satisfaction behavior of proposed model. An accuracy improvement of approximately 1.7X-4.7X over the non-BC-satisfying baselines is also achieved. Furthermore, our studies on SOL underscore the significance of satisfying BCs as a criterion for deep surrogate models of PDEs.

math.NA↗

arXiv2023

VPU-EM: An Event-based Modeling Framework to Evaluate NPU Performance and Power Efficiency at Scale

State-of-art NPUs are typically architected as a self-contained sub-system with multiple heterogeneous hardware computing modules, and a dataflow-driven programming model. There lacks well-established methodology and tools in the industry to evaluate and compare the performance of NPUs from different architectures. We present an event-based performance modeling framework, VPU-EM, targeting scalable performance evaluation of modern NPUs across diversified AI workloads. The framework adopts high-level event-based system-simulation methodology to abstract away design details for speed, while maintaining hardware pipelining, concurrency and interaction with software task scheduling. It is natively developed in Python and built to interface directly with AI frameworks such as Tensorflow, PyTorch, ONNX and OpenVINO, linking various in-house NPU graph compilers to achieve optimized full model performance. Furthermore, VPU-EM also provides the capability to model power characteristics of NPU in Power-EM mode to enable joint performance/power analysis. Using VPU-EM, we conduct performance/power analysis of models from representative neural network architecture. We demonstrate that even though this framework is developed for Intel VPU, an Intel in-house NPU IP technology, the methodology can be generalized for analysis of modern NPUs.

cs.AR↗

arXiv2023

Render unto Numerics: Orthogonal Polynomial Neural Operator for PDEs with Non-periodic Boundary Conditions

By learning the mappings between infinite function spaces using carefully designed neural networks, the operator learning methodology has exhibited significantly more efficiency than traditional methods in solving complex problems such as differential equations, but faces concerns about their accuracy and reliability. To overcomes these limitations, combined with the structures of the spectral numerical method, a general neural architecture named spectral operator learning (SOL) is introduced, and one variant called the orthogonal polynomial neural operator (OPNO), developed for PDEs with Dirichlet, Neumann and Robin boundary conditions (BCs), is proposed later. The strict BC satisfaction properties and the universal approximation capacity of the OPNO are theoretically proven. A variety of numerical experiments with physical backgrounds show that the OPNO outperforms other existing deep learning methodologies, as well as the traditional 2nd-order finite difference method (FDM) with a considerably fine mesh (with the relative errors reaching the order of 1e-6), and is up to almost 5 magnitudes faster than the traditional method.

math.NA↗

arXiv2023

Arbitrary high-order structure-preserving schemes for the generalized Rosenau-type equation

In this paper, we are concerned with arbitrarily high-order momentum-preserving and energy-preserving schemes for solving the generalized Rosenau-type equation, respectively. The derivation of the momentum-preserving schemes is made within the symplectic Runge-Kutta method, coupled with the standard Fourier pseudo-spectral method in space. Then, combined with the quadratic auxiliary variable approach and the symplectic Runge-Kutta method, together with the standard Fourier pseudo-spectral method, we present a class of high-order mass- and energy-preserving schemes for the Rosenau equation. Finally, extensive numerical tests and comparisons are also addressed to illustrate the performance of the proposed schemes.

math.NA↗

arXiv2022

An Operator Learning Approach via Function-valued Reproducing Kernel Hilbert Space for Differential Equations

Much recent work has addressed the solution of a family of partial differential equations by computing the inverse operator map between the input and solution space. Toward this end, we incorporate function-valued reproducing kernel Hilbert spaces in our operator learning model. We use neural networks to parameterize Hilbert-Schmidt integral operator and propose an architecture. Experiments including several typical datasets show that the proposed architecture has desirable accuracy on linear and nonlinear partial differential equations even with a small amount of data. By learning the mappings between function spaces, the proposed method can find the solution of a high-resolution input after learning from lower-resolution data.

math.NA↗

arXiv2021

High-order linearly implicit structure-preserving exponential integrators for the nonlinear Schrödinger equation

A novel class of high-order linearly implicit energy-preserving integrating factor Runge-Kutta methods are proposed for the nonlinear Schrödinger equation. Based on the idea of the scalar auxiliary variable approach, the original equation is first reformulated into an equivalent form which satisfies a quadratic energy. The spatial derivatives of the system are then approximated with the standard Fourier pseudo-spectral method. Subsequently, we apply the extrapolation technique/prediction-correction strategy to the nonlinear terms of the semi-discretized system and a linearized energy-conserving system is obtained. A fully discrete scheme is gained by further using the integrating factor Runge-Kutta method to the resulting system. We show that, under certain circumstances for the coefficients of a Runge-Kutta method, the proposed scheme can produce numerical solutions along which the modified energy is precisely conserved, as is the case with the analytical solution and is extremely efficient in the sense that only linear equations with constant coefficients need to be solved at every time step. Numerical results are addressed to demonstrate the remarkable superiority of the proposed schemes in comparison with other existing structure-preserving schemes.

math.NA↗

arXiv2021

Arbitrary high-order linear structure-preserving schemes for the regularized long-wave equation

In this paper, a class of arbitrarily high-order linear momentum-preserving and energy-preserving schemes are proposed, respectively, for solving the regularized long-wave equation. For the momentum-preserving scheme, the key idea is based on the extrapolation/prediction-correction technique and the symplectic Runge-Kutta method in time, together with the standard Fourier pseudo-spectral method in space. We show that the scheme is linear, high-order, unconditionally stable and preserves the discrete momentum of the system. For the energy-preserving scheme, it is mainly based on the energy quadratization approach and the analogous linearized strategy used in the construction of the linear momentum-preserving scheme. The proposed scheme is linear, high-order and can preserve a discrete quadratic energy exactly. Numerical results are addressed to demonstrate the accuracy and efficiency of the proposed scheme.

math.NA↗

arXiv2020

Channel-wise Hessian Aware trace-Weighted Quantization of Neural Networks

Second-order information has proven to be very effective in determining the redundancy of neural network weights and activations. Recent paper proposes to use Hessian traces of weights and activations for mixed-precision quantization and achieves state-of-the-art results. However, prior works only focus on selecting bits for each layer while the redundancy of different channels within a layer also differ a lot. This is mainly because the complexity of determining bits for each channel is too high for original methods. Here, we introduce Channel-wise Hessian Aware trace-Weighted Quantization (CW-HAWQ). CW-HAWQ uses Hessian trace to determine the relative sensitivity order of different channels of activations and weights. What's more, CW-HAWQ proposes to use deep Reinforcement learning (DRL) Deep Deterministic Policy Gradient (DDPG)-based agent to find the optimal ratios of different quantization bits and assign bits to channels according to the Hessian trace order. The number of states in CW-HAWQ is much smaller compared with traditional AutoML based mix-precision methods since we only need to search ratios for the quantization bits. Compare CW-HAWQ with state-of-the-art shows that we can achieve better results for multiple networks.

cs.CV↗

arXiv2020

Regularized finite difference methods for the logarithmic Klein-Gordon equation

We propose and analyze two regularized finite difference methods for the logarithmic Klein-Gordon equation (LogKGE). Due to the blowup phenomena caused by the logarithmic nonlinearity of the LogKGE, it is difficult to construct numerical schemes and establish their error bounds. In order to avoid singularity, we present a regularized logarithmic Klein-Gordon equation (RLogKGE) with a small regularized parameter $0<\varepsilon\ll1$. Besides, two finite difference methods are adopted to solve the regularized logarithmic Klein-Gordon equation (RLogKGE) and rigorous error bounds are estimated in terms of the mesh size $h$, time step $τ$, and the small regularized parameter $\varepsilon$. Finally, numerical experiments are carried out to verify our error estimates of the two numerical methods and the convergence results from the LogKGE to the RLogKGE with the linear convergence order $O(\varepsilon)$.

math.AP↗