A Novel Matrix-Free Finite Element Method for Time-Harmonic Maxwell's Equations

被引:1
|
作者
Wang, Yi-Yao [1 ]
Zhan, Qiwei [1 ]
Feng, Haoqiang [1 ]
Wang, Yin-Da [1 ]
Sun, Bozhao [1 ]
Yin, Wen-Yan [1 ]
机构
[1] Zhejiang Univ, Innovat Inst Electromagnet Informat & Elect Integr, Key Lab Adv Micronano Elect Devices & Smart Syst, Hangzhou 310027, Peoples R China
基金
中国国家自然科学基金;
关键词
Finite element analysis; Frequency-domain analysis; Complexity theory; Matrix decomposition; Sparse matrices; Time-domain analysis; Linear systems; Geometric multigrid; matrix-free; sum factorization; tensorial basis functions; vectorization; DOMAIN DECOMPOSITION METHOD; SWEEPING PRECONDITIONER; HELMHOLTZ-EQUATION; FORMULATION; VERSION; VECTOR;
D O I
10.1109/TAP.2024.3360700
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
A frequency-domain finite element method (FEM) with the low-storage matrix-free feature is proposed for efficient analysis of electromagnetic fields. As opposed to conventional frequency-domain FEM which requires the matrix assembly before solving, the proposed matrix-free algorithm avoids the assembly and storage of the global matrix. In this approach, the global sparse matrix-vector (SpMV) multiplication is decomposed into element-wise matrix-vector (MV) multiplications. Additionally, the sum factorization technique is applied on tensorial basis functions to reduce the complexities of local MV multiplications. The numerical results demonstrate the superiority of our algorithm over traditional FEM in terms of both memory and time consumption. Besides, the improvement is more profound when higher-order basis functions are considered. A speedup of more than 10 against matrix-assembled solvers is observed. Given that memory transfer is bounded more than computation resources in modern supercomputer architectures, our algorithm is more friendly to high-performance computing platforms.
引用
收藏
页码:2609 / 2619
页数:11
相关论文
共 50 条