Association Journal of CSIAM
Supervised by Ministry of Education of PRC
Sponsored by Xi'an Jiaotong University
ISSN 1005-3085  CN 61-1269/O1

Chinese Journal of Engineering Mathematics

Previous Articles    

Computational Scheme and Efficiency Analysis of Multiscale Finite Elements on Optimally Graded Meshes for Two-dimensional Singularly Perturbed Problems

SUN Meiling1,2,   JIANG Shan1,   WANG Xiaoying1   

  1. 1. School of Mathematics and Statistics, Nantong University, Nantong 226019
    2. Department of Mathematics Teaching and Research, Nantong Vocational University, Nantong 226007
  • Received:2022-02-28 Accepted:2022-06-04
  • Contact: S. Jiang. E-mail address: jiangshan@ntu.edu.cn
  • Supported by:
    The National Natural Science Foundation of China (11771224); the Basic Science Research Mandatory Project of Nantong City (JC2021123); the Natural Science Research Key Project of Nantong Vocational University (23ZK03).

Abstract:

As for a two-dimensional convection-diffusion equation in the singular perturbation, a novel multiscale finite element method based on the optimally graded meshes is proposed. The multiscale finite element method just solves the sub-problems on coarse meshes, and the data mapping relationship for related scales is provided in details and the microscopic information is inherited to the macroscopic level. Then the matrix is reduced and its matrix equation is ready for solving efficiently. Based on the perturbed parameter, an adaptively graded mesh is constructed from its iterative formula, and the meshes are capable of approximating the boundary layers effectively. Through mathematical analyses and numerical experiments, to contrast the computational cost and execution time, the multiscale strategy on the graded mesh is validated to be the stable, high-order and short-time uniform convergence. Its computational efficiency and application advantage are prominent.

Key words: singular perturbation, two-dimensional graded mesh, multiscale finite element, uniform convergence

CLC Number: