Abstract
A parallel fast multipole accelerated boundary integral algorithm for solving boundary value problems is presented in this paper. This paper focuses on the downward pass, the upward pass and the preconditioning process, which are considered to be the expensive parts in the FMM (Fast Multipole Method) algorithm. All of them are parallelized with OpenMP. The performance of the parallel fast multipole method is tested with two numerical examples, namely the crack problem for Laplace's equation in 2-D and the elastostatic inclusion problem in 3-D. From those numerical examples, we can say that the speedup achieved with OpenMP is satisfactory in spite of the relatively small amount of eflbrts for programmers and small change in the program structure. Particularly remarkable is the fact that the effect of this parallelization in the fast multipole method is more pronounced in large scale problems or in 3-D problems.
