Second-order LOD multigrid method for multidimensional Riesz fractional diffusion equation

TL;DR

This paper introduces a second-order, unconditionally stable LOD finite difference method combined with a multigrid solver for multidimensional Riesz fractional diffusion equations, achieving efficient computation and high accuracy.

Contribution

It develops a novel second-order LOD finite difference scheme with Toeplitz-like matrices and an efficient multigrid solver for multidimensional Riesz fractional diffusion equations.

Findings

Second-order convergence in space and time.

Computational complexity of O(N log N) for matrix-vector multiplication.

Numerical experiments confirm the method's efficiency and accuracy.

Abstract

We propose a locally one dimensional (LOD) finite difference method for multidimensional Riesz fractional diffusion equation with variable coefficients on a finite domain. The numerical method is second-order convergent in both space and time directions, and its unconditional stability is strictly proved. Comparing with the popular first-order finite difference method for fractional operator, the form of obtained matrix algebraic equation is changed from $(I-A)u^{k+1}=u^k+b^{k+1}$ to $(I-{\widetilde A})u^{k+1}=(I+{\widetilde B})u^k+{\tilde b}^{k+1/2}$; the three matrices $A$, ${\widetilde A}$ and ${\widetilde B}$ are all Toeplitz-like, i.e., they have completely same structure and the computational count for matrix vector multiplication is $\mathcal{O}(N {log} N)$; and the computational costs for solving the two matrix algebraic equations are almost the same. The LOD-multigrid method is used to solve the resulting matrix algebraic equation, and the computational count is $\mathcal{O}(N {log} N)$ and the required storage is $\mathcal{O}(N)$, where $N$ is the number of grid points. Finally, the extensive numerical experiments are performed to show the powerfulness of the second-order scheme and the LOD-multigrid method.