Provably scalable parallel FMM algorithm for multiscale electromagnetic simulations

In this work we introduce a new algorithm for efficient implementation of FMM algorithm in distributed environment that is scalable Upton several hundreds and thousands of processors. The theoretical foundations of the presented algorithm ensures optimal and uniform distribution of computational load among processors. This reduces latency on arbitrary for any given number of processors and results in a provably scalable algorithm. Further, we augment FMM with the accelerated Cartesian expansion (ACE) algorithm to overcome its low frequency limitations. The resulting parallel wideband FMM algorithm enables simulation of realistic, multiscale geometries with millions of unknowns on large scale networks, in an efficient manner.