Abstract
On the purpose of developing a matrix solver on massively parallel computers (MPC), we have investigated the both Multigrid and ADI method, and had numerical experiments, applying them to practical problems (2-D Poisson equation with Neumann type boundary conditions, etc). We show that Multigrid method is as efficient as ILUBCG on the scalar processor, required operations for ADI method is around 10 times the size for ILUBCG on the scalar processor, ADI method is especially powerful for the convective diffusion problems, and an effective acceleration factor in a uniform diffusion field is not available for a non-uniform field. We have concluded that they will be useful enough on MPC because of their large granularity.
