A parallel Gauss—Seidel method using NR data flow ordering

摘要

A parallel variant of the block Gauss—Seidel method is presented to solve the Poisson equation with Dirichlet boundary condition. This method uses two-dimensional logically connected parallel processors. Furthermore, natural rowwise (NR) data flow block ordering is used so that its convergence rate is the same as that of the standard block Gauss—Seidel method. Spectral radius is determined by the formula for general k × l block iterative methods. Numerical computations on a parallel computer are included.