Massively Parallel Multilevel Fast Multipole Algorithm for Extremely Large-Scale Electromagnetic Simulations: A Review

Abstract

Since the first working multilevel fast multipole algorithm (MLFMA) for electromagnetic simulations was proposed by Chew's group in 1995, this algorithm has been recognized as one of the most powerful tools for numerical solutions of extremely large electromagnetic problems with complex geometries. It has been parallelized with different strategies to explore the computing power of supercomputers, increasing the size of solvable problems from millions to tens of billions of unknowns, thereby addressing the crucial demand arising from practical applications in a sense. This paper provides a comprehensive review of state-of-the-art parallel approaches of the MLFMA, especially on a newly proposed ternary parallelization scheme and its acceleration on graphics processing unit (GPU) clusters. We discuss and numerically study the advantages of the ternary parallelization scheme and demonstrate its flexibility and efficiency.

A Fast Coupled Iterative Approach for Computing 3D Composite Scattering from Dielectric Rough Surfaces and Target

“…Paper [25] comprehensively reviewed and analyzed the basic principles, implementation methods, advantages and disadvantages, and application in practical problems of the large-scale parallel multi-level fast multipole algorithm. This algorithm achieved parallel computing and speedup by decomposing the computation domain into multiple subdomains and performing fast multipole computation in each subdomain.…”

“…Addition-ally, the algorithm used parallel computing techniques to allocate computational tasks to multiple processors or computers to achieve faster computational speed. Paper [25] also discussed the parallel performance and computational efficiency of the algorithm and explored how to further improve computational efficiency and accuracy by optimizing algorithm parameters and adjusting parallel strategies.…”