Towards a scalable parallel MLFMA in three dimensions

Abstract

The development of a scalable parallel multilevel fast multipole algorithm (MLFMA) for three dimensional electromagnetic scattering problems is reported. In the context of this work, the term 'scalable' denotes the ability to handle larger simulations with a proportional increase in the number of parallel processes (CPU cores), without loss of parallel efficiency. The workload is divided among the different processes according to the hierarchical partitioning scheme. Crucial to ensure the parallel scalability of the algorithm, is that the radiation patterns — sampled on the sphere — are partitioned in two dimensions, i.e., both in azimuth and elevation directions.