Periodically generated iterative methods for solving elliptic equations (Q1917429): Difference between revisions

The paper is concerned with ``periodically generated'' iterative methods for solving linear systems of algebraic equations which are derived from the numerical solution of the Poisson equation in the unit square with Dirichlet boundary conditions using the standard five-point finite difference equation. The authors first construct a discrete periodic problem corresponding to the boundary value problem. This is done by applying a procedure, previously used by \textit{S. Xiao} [Report CNA-265, Center for Numerical Analysis, Univ. of Texas, Austin, TX (1994)] in connection with parallel multigrid methods. For simplicity, the procedure is illustrated for a problem in one dimension. After that, the application of the Gauss-Seidel method to the solution of the discrete periodic problem obtained is described. There are no difficulties to apply in a similar way some of the other standard iterative methods as Jacobi, SOR and SSOR. The convergence fators for Gauss-Seidel, Jacobi, SOR and SSOR methods (i.e. the spectral radius of the corresponding matrices) together with the optimal relaxation parameters for SOR and SSOR for the discrete periodic problems are listed and discussed. A brief discussion of a procedure for the parallel implementation of the nonstationary SSOR method to discrete periodic problems is added at the end of the paper.