arXiv subjects

Shi Shu

Publications and source records attributed to Shi Shu.

At least 19 recordsLinked to original sources

arXiv2026

Purified Two-Relaxation-Time Lattice Boltzmann Method: Removing Ghost Modes from TRT for Enhanced Stability

The two-relaxation-time (TRT) lattice Boltzmann model is widely adopted for its simplicity and tunable boundary accuracy. However, its collision operator relaxes the full symmetric non-equilibrium component, implicitly retaining non-hydrodynamic ghost modes that degrade stability at high Reynolds numbers. In this work, we establish a rigorous connection between ghost-mode filtering and regularization within the TRT framework. By decomposing the discrete velocity space into hydrodynamic and non-hydrodynamic subspaces, we prove that the TRT-regularized lattice Boltzmann (TRT-RLB) model is mathematically equivalent to the standard TRT model with ghost modes explicitly removed. This equivalence holds exactly for D2Q9 and D3Q19 lattices, where the symmetric and antisymmetric subspaces are completely spanned by the physically relevant Hermite modes and identifiable ghost modes. Based on this finding, we propose the Purified TRT (P-TRT) model, which achieves regularization-level stability through simple algebraic ghost-mode subtraction rather than expensive tensor projections. For D2Q9, the non-equilibrium collision cost is reduced from 180 to 52 floating-point operations per node, a 71% reduction. Linear stability analysis in moment space further reveals that the P-TRT operator annihilates the ghost eigenvalue, proving its spectral radius is bounded above by that of standard TRT and that stability is governed exclusively by hydrodynamic modes. Benchmarks including the double shear layer at Re up to 10^7, Taylor--Green vortex decay, force-driven Poiseuille flow, and creeping flow past a square cylinder confirm that P-TRT preserves the stability, second-order accuracy, and zero-slip boundary properties of TRT-RLB while retaining the simplicity of the TRT family.

physics.flu-dyn↗

arXiv2025

Projected Sobolev Natural Gradient Descent for Efficient Neural Network Solution of the Gross--Pitaevskii Equation

This paper introduces a projected Sobolev natural gradient descent (NGD) method for computing ground states of the Gross--Pitaevskii equation. By projecting a continuous Riemannian Sobolev gradient flow onto the normalized neural network tangent space, we derive a discrete NGD algorithm that preserves the normalization constraint. The numerical implementation employs variational Monte Carlo with a hybrid sampling strategy to accurately account for the normalization constant. To enhance computational efficiency, a matrix-free Nystr\"om-preconditioned conjugate gradient solver is adopted to approximate the NGD operator without explicit matrix assembly. Numerical experiments demonstrate that the proposed method converges significantly faster than physics-informed neural network approaches and exhibits near-constant wall-clock time and memory over the tested high-dimensional range when the hidden architecture is fixed and the implicit Gram operator is used. Moreover, the resulting neural-network solutions provide high-quality initial guesses that substantially accelerate subsequent refinement by traditional high-precision solvers.

math.NA↗

arXiv2025

Nodal Hybrid Neural Solvers for Parametric PDE Systems

The numerical solution of partial differential equations (PDEs) is fundamental to scientific and engineering computing. In the presence of strong anisotropy, material heterogeneity, and complex geometries, however, classical iterative solvers often suffer from reduced efficiency and require substantial problem-dependent tuning. The Fourier neural solver (FNS) is a learning-based hybrid iterative solver for such problems without extensive manual parameter tuning, but its original design is primarily effective for scalar PDEs on structured meshes and is difficult to extend directly to unstructured meshes or strongly coupled PDE systems. Building on the FNS framework, we introduce block smoothing operators and graph neural networks to construct a solver for unstructured systems, termed the graph Fourier neural solver (G-FNS). We further incorporate a coordinate transformation network to develop the adaptive graph Fourier neural solver (AG-FNS), and then extend this formulation to a frequency-domain multilevel variant, ML-AG-FNS. Rigorous analysis shows that, under suitable mathematical assumptions, the proposed method achieves mesh-independent convergence rate. Error-spectrum visualizations further indicate that AG-FNS can capture complex multiscale error modes. Extensive experiments on two-dimensional anisotropic diffusion and on two- and three-dimensional isotropic/anisotropic linear elasticity problems over unstructured meshes demonstrate strong robustness and efficiency. The proposed framework can be used either as a solver or as a preconditioner for Krylov subspace methods. Overall, it substantially extends the original FNS methodology and broadens the applicability of this class of neural solvers.

math.NA↗

arXiv2025

Controllable Layer Decomposition for Reversible Multi-Layer Image Generation

This work presents Controllable Layer Decomposition (CLD), a method for achieving fine-grained and controllable multi-layer separation of raster images. In practical workflows, designers typically generate and edit each RGBA layer independently before compositing them into a final raster image. However, this process is irreversible: once composited, layer-level editing is no longer possible. Existing methods commonly rely on image matting and inpainting, but remain limited in controllability and segmentation precision. To address these challenges, we propose two key modules: LayerDecompose-DiT (LD-DiT), which decouples image elements into distinct layers and enables fine-grained control; and Multi-Layer Conditional Adapter (MLCA), which injects target image information into multi-layer tokens to achieve precise conditional generation. To enable a comprehensive evaluation, we build a new benchmark and introduce tailored evaluation metrics. Experimental results show that CLD consistently outperforms existing methods in both decomposition quality and controllability. Furthermore, the separated layers produced by CLD can be directly manipulated in commonly used design tools such as PowerPoint, highlighting its practical value and applicability in real-world creative workflows. Our project is available at https://monkek123king.github.io/CLD_page/.

cs.GR↗

arXiv2025

An improved lattice Boltzmann method with a novel conservative boundary scheme for viscoelastic fluid flows

The high Weissenberg number problem has been a persistent challenge in the numerical simulation of viscoelastic fluid flows. This paper presents an improved lattice Boltzmann method for solving viscoelastic flow problems at high Weissenberg numbers. The proposed approach employs two independent two-relaxation-time regularized lattice Boltzmann models to solve the hydrodynamic field and conformation tensor field of viscoelastic fluid flows, respectively. The viscoelastic stress computed from the conformation tensor is directly embedded into the hydrodynamic field using a newly proposed local velocity discretization scheme, thereby avoiding spatial gradient calculations. The constitutive equations are treated as convection-diffusion equations and solved using an improved convection-diffusion model specifically designed for this purpose, incorporating a novel auxiliary source term that eliminates the need for spatial and temporal derivative computations. Additionally, a conservative non-equilibrium bounce-back (CNEBB) scheme is proposed for implementing solid wall boundary conditions in the constitutive equations. The robustness of the present algorithm is validated through a series of benchmark problems. The simplified four-roll mill problem demonstrates that the method effectively improves numerical accuracy and stability in bulk regions containing stress singularities. The Poiseuille flow problem validates the accuracy of the current algorithm with the CNEBB boundary scheme at extremely high Weissenberg numbers (tested up to Wi = 10,000). The flow past a circular cylinder problem confirms the superior stability and applicability of the algorithm for complex curved boundary problems compared to other existing common schemes.

physics.flu-dyn↗

arXiv2025

A BDDC method with an adaptive coarse space for three-dimensional advection-diffusion problems

The solution of nonsymmetric but positive definite (NSPD) systems arising from advection-diffusion problems is an important research topic in science and engineering. Balancing domain decomposition by constraints with an adaptive coarse space(adaptive BDDC) constitute a significant class of nonoverlapping domain decomposition methods, commonly used for symmetric positive definite problems. In this paper, we propose an adaptive BDDC method that incorporates a class of edge generalized eigenvalue problems based on prior selected primal constraints to solve NSPD systems from advection-diffusion problems. Compared with the conventional adaptive BDDC method for such systems, the proposed approach further reduces the number of primal unknowns. Numerical experiments show that although the iteration count increases slightly, the overall computational time is significantly reduced.

math.NA↗

arXiv2024

Momentum-Accelerated Richardson(m) and Their Multilevel Neural Solvers

Recently, designing neural solvers for large-scale linear systems of equations has emerged as a promising approach in scientific and engineering computing. This paper first introduce the Richardson(m) neural solver by employing a meta network to predict the weights of the long-step Richardson iterative method. Next, by incorporating momentum and preconditioning techniques, we further enhance convergence. Numerical experiments on anisotropic second-order elliptic equations demonstrate that these new solvers achieve faster convergence and lower computational complexity compared to both the Chebyshev iterative method with optimal weights and the Chebyshev semi-iteration method. To address the strong dependence of the aforementioned single-level neural solvers on PDE parameters and grid size, we integrate them with two multilevel neural solvers developed in recent years. Using alternating optimization techniques, we construct Richardson(m)-FNS for anisotropic equations and NAG-Richardson(m)-WANS for the Helmholtz equation. Numerical experiments show that these two multilevel neural solvers effectively overcome the drawback of single-level methods, providing better robustness and computational efficiency.

math.NA↗

arXiv2024

A Hybrid Iterative Neural Solver Based on Spectral Analysis for Parametric PDEs

Deep learning-based hybrid iterative methods (DL-HIM) have emerged as a promising approach for designing fast neural solvers to tackle large-scale sparse linear systems. DL-HIM combine the smoothing effect of simple iterative methods with the spectral bias of neural networks, which allows them to effectively eliminate both high-frequency and low-frequency error components. However, their efficiency may decrease if simple iterative methods can not provide effective smoothing, making it difficult for the neural network to learn mid-frequency and high-frequency components. This paper first conducts a convergence analysis for general DL-HIM from a spectral viewpoint, concluding that under reasonable assumptions, DL-HIM exhibit a convergence rate independent of grid size $h$ and physical parameters $\boldsymbol{\mu}$. To meet these assumptions, we design a neural network from an eigen perspective, focusing on learning the eigenvalues and eigenvectors corresponding to error components that simple iterative methods struggle to eliminate. Specifically, the eigenvalues are learned by a meta subnet, while the eigenvectors are approximated using Fourier modes with a transition matrix provided by another meta subnet. The resulting DL-HIM, termed the Fourier Neural Solver (FNS), can be trained to achieve a convergence rate independent of PDE parameters and grid size within a local neighborhood of the training scale by designing a loss function that ensures the neural network complements the smoothing effect of the damped Jacobi iterative methods. We verify the performance of FNS on five types of linear parametric PDEs.

math.NA↗

arXiv2024

A Neural Multigrid Solver for Helmholtz Equations with High Wavenumber and Heterogeneous Media

In this paper, we propose a deep learning-enhanced multigrid solver for high-frequency and heterogeneous Helmholtz equations. By applying spectral analysis, we categorize the iteration error into characteristic and non-characteristic components. We eliminate the non-characteristic components by a multigrid wave cycle, which employs carefully selected smoothers on each grid. We diminish the characteristic components by a learned phase function and the approximate solution of an advection-diffusion-reaction (ADR) equation, which is solved using another multigrid V-cycle on a coarser scale, referred to as the ADR cycle. The resulting solver, termed Wave-ADR-NS, enables the handling of error components with varying frequencies and overcomes constraints on the number of grid points per wavelength on coarse grids. Furthermore, we provide an efficient implementation using differentiable programming, making Wave-ADR-NS an end-to-end Helmholtz solver that incorporates parameters learned through a semi-supervised training. Wave-ADR-NS demonstrates robust generalization capabilities for both in-distribution and out-of-distribution velocity fields of varying difficulty. Comparative experiments with other multigrid methods validate its superior performance in solving heterogeneous 2D Helmholtz equations with wavenumbers exceeding 2000.

math.NA↗

arXiv2023

Two-relaxation-time regularized lattice Boltzmann model for convection-diffusion equation with variable coefficients

In this paper, a new two-relaxation-time regularized (TRT-R) lattice Boltzmann (LB) model for convection-diffusion equation (CDE) with variable coefficients is proposed. Within this framework, we first derive a TRT-R collision operator by constructing a new regularized procedure through the high-order Hermite expansion of non-equilibrium. Then a first-order discrete-velocity form of discrete source term is introduced to improve the accuracy of the source term. Finally and most importantly, a new first-order space-derivative auxiliary term is proposed to recover the correct CDE with variable coefficients. To evaluate this model, we simulate a classic benchmark problem of the rotating Gaussian pulse. The results show that our model has better accuracy, stability and convergence than other popular LB models, especially in the case of a large time step.

math.NA↗

arXiv2023

Two-relaxation-time regularized lattice Boltzmann model for Navier-Stokes equations

In this paper, we propose a novel two-relaxation-time regularized lattice Boltzmann (TRT-RLB) model for simulating weakly compressible isothermal flows. A free relaxation parameter, $\tau_{s,2}$, is employed to relax the regularized non-equilibrium third-order terms. Chapman-Enskog analysis reveals that our model can accurately recover the Navier-Stokes equations (NSEs). Theoretical analysis of the Poiseuille flow problem demonstrates that the slip velocity magnitude in the proposed model is controlled by a magic parameter, which can be entirely eliminated under specific values, consistent with the classical TRT model. Our simulations of the double shear layer problem, Taylor-Green vortex flow, and force-driven Poiseuille flow confirm that the stability and accuracy of our model significantly surpass those of both the regularized lattice Boltzmann (RLB) and two-relaxation-time (TRT) models, even under super-high Reynolds numbers as $Re=10^7$. Concurrently, the TRT-RLB model exhibits superior performance in very high viscosity scenarios. The simulations of creeping flow around a square cylinder demonstrates the model's capability to accurately compute ultra-low Reynolds numbers as $Re=10^{-7}$. This study establishes the TRT-RLB model as a flexible and robust tool in computational fluid dynamics.

physics.flu-dyn↗

arXiv2023

Fast Algorithms for Finite Element nonlinear Discrete Systems to Solve the PNP Equations

The Poisson-Nernst-Planck (PNP) equations are one of the most effective model for describing electrostatic interactions and diffusion processes in ion solution systems, and have been widely used in the numerical simulations of biological ion channels, semiconductor devices, and nanopore systems. Due to the characteristics of strong coupling, convection dominance, nonlinearity and multiscale, the classic Gummel iteration for the nonlinear discrete system of PNP equations converges slowly or even diverges. We focus on fast algorithms of nonlinear discrete system for the general PNP equations, which have better adaptability, friendliness and efficiency than the Gummel iteration. First, a geometric full approximation storage (FAS) algorithm is proposed to improve the slow convergence speed of the Gummel iteration. Second, an algebraic FAS algorithm is designed, which does not require multi-level geometric information and is more suitable for practical computation compared with the geometric one. Finally, improved algorithms based on the acceleration technique and adaptive method are proposed to solve the problems of excessive coarse grid iterations and insufficient adaptability to the size of computational domain in the algebraic FAS algorithm. The numerical experiments are shown for the geometric, algebraic FAS and improved algorithms respectively to illustrate the effiency of the algorithms.

math.NA↗

arXiv2023

The Existence the Solution of Nonlinear Discrete Schemes and Convergence of a Linearized Iterative Method for time-dependent PNP Equations

We establish the existence theory of several commonly used finite element (FE) nonlinear fully discrete solutions, and the convergence theory of a linearized iteration. First, it is shown for standard FE, SUPG and edge-averaged method respectively that the stiffness matrix is a column M-matrix under certain conditions, and then the existence theory of these three FE nonlinear fully discrete solutions is presented by using Brouwer's fixed point theorem. Second, the contraction of a commonly used linearized iterative method-Gummel iteration is proven, and then the convergence theory is established for the iteration. At last, a numerical experiment is shown to verifies the theories.

math.NA↗

arXiv2023

An exact solution for the magnetic diffusion problem with a step-function resistivity model

In the magnetic diffusion problem, a magnetic diffusion equation is coupled by an Ohmic heating energy equation. The Ohmic heating can make the magnetic diffusion coefficient (i. e., the resistivity) vary violently, and make the diffusion a highly nonlinear process. For this reason, the problem is normally very hard to be solved analytically. In this article, under the condition of a step-function resistivity and a constant boundary magnetic field, we successfully derived an exact solution for this nonlinear problem, which should be an interesting thing in the area of partial differential equations. What's more, the solution could serve as a valuable benchmark example for testing simulation methods of the magnetic diffusion problem.

math-ph↗

arXiv2023

Error Analysis of Virtual Element Method for the Poisson-Boltzmann Equation

The Poisson-Boltzmann equation is a nonlinear elliptic equation with Dirac distribution sources, which has been widely applied to the prediction of electrostatics potential of biological biomolecular systems in solution. In this paper, we discuss and analysis the virtual element method for the Poisson-Boltzmann equation on general polyhedral meshes. Under the low regularity of the solution of the whole domain, nearly optimal error estimates in both L2-norm and H1-norm for the virtual element approximation are obtained. The numerical experiment on different polyhedral meshes shows the efficiency of the virtual element method and verifies the proposed theoretical prediction.

math.NA↗

arXiv2022

FAS-UNet: A Novel FAS-driven Unet to Learn Variational Image Segmentation

Solving variational image segmentation problems with hidden physics is often expensive and requires different algorithms and manually tunes model parameter. The deep learning methods based on the U-Net structure have obtained outstanding performances in many different medical image segmentation tasks, but designing such networks requires a lot of parameters and training data, not always available for practical problems. In this paper, inspired by traditional multi-phase convexity Mumford-Shah variational model and full approximation scheme (FAS) solving the nonlinear systems, we propose a novel variational-model-informed network (denoted as FAS-Unet) that exploits the model and algorithm priors to extract the multi-scale features. The proposed model-informed network integrates image data and mathematical models, and implements them through learning a few convolution kernels. Based on the variational theory and FAS algorithm, we first design a feature extraction sub-network (FAS-Solution module) to solve the model-driven nonlinear systems, where a skip-connection is employed to fuse the multi-scale features. Secondly, we further design a convolution block to fuse the extracted features from the previous stage, resulting in the final segmentation possibility. Experimental results on three different medical image segmentation tasks show that the proposed FAS-Unet is very competitive with other state-of-the-art methods in qualitative, quantitative and model complexity evaluations. Moreover, it may also be possible to train specialized network architectures that automatically satisfy some of the mathematical and physical laws in other image problems for better accuracy, faster training and improved generalization.The code is available at \url{https://github.com/zhuhui100/FASUNet}.

cs.CV↗

arXiv2022

Fourier Neural Solver for large sparse linear algebraic systems

Large sparse linear algebraic systems can be found in a variety of scientific and engineering fields, and many scientists strive to solve them in an efficient and robust manner. In this paper, we propose an interpretable neural solver, the Fourier Neural Solver (FNS), to address them. FNS is based on deep learning and Fast Fourier transform. Because the error between the iterative solution and the ground truth involves a wide range of frequency modes, FNS combines a stationary iterative method and frequency space correction to eliminate different components of the error. Local Fourier analysis reveals that the FNS can pick up on the error components in frequency space that are challenging to eliminate with stationary methods. Numerical experiments on the anisotropy diffusion equation, convection-diffusion equation, and Helmholtz equation show that FNS is more efficient and more robust than the state-of-the-art neural solver.

math.NA↗

arXiv2022

An Improved Multi-Stage Preconditioner on GPUs for Compositional Reservoir Simulation

The compositional model is often used to describe multicomponent multiphase porous media flows in the petroleum industry. The fully implicit method with strong stability and weak constraints on time-step sizes is commonly used in the mainstream commercial reservoir simulators. In this paper, we develop an efficient multi-stage preconditioner for the fully implicit compositional flow simulation. The method employs an adaptive setup phase to improve the parallel efficiency on GPUs. Furthermore, a multi-color Gauss-Seidel algorithm based on the adjacency matrix is applied in the algebraic multigrid methods for the pressure part. Numerical results demonstrate that the proposed algorithm achieves good parallel speedup while yields the same convergence behavior as the corresponding sequential version.

math.NA↗