Parallel fast algorithm for large-scale electromagnetic scattering

Abstract

Electromagnetic scattering analysis for radar cross section (RCS) prediction and electromagnetic compatibility analysis have extensive military and civil applications. Meanwhile, it presents very large computational demands. Parallel multi-level fast multipole method (MLFMM), a powerful and efficient computational algorithm on solving large-scale RCS problems, is developed and implemented on massively parallel, distributed memory computer systems. Multilevel grouping structure is partitioned carefully to separate computational tasks and reduce data communication. Message passing interface (MPI) is employed to perform the data transfer among processors and support the matrix-vector product in parallel iterative procedure. The accuracy of code is verified by comparison with results of benchmark target. The computational complexity and efficiency of parallel MLFMM is discussed. The parallel code is validated on Unix and Linux platforms with different number of processors, and proven to be scalable, portable, and efficient.