The convergence rates of the point-Jacobi method and of the successive overrelaxation (including the Gauss-Seidel) method for symmetric positive definite linear equations are determined as functions of the spectrum of the matrix which is scaled to ones on the principal diagonal. It is seen that the convergence rate is strongly dependent upon the magnitude of the smallest eigenvalue and that for even mildly ill-conditioned matrices the convergence is very slow. It is also shown that the finite difference representation of the three-dimensional Laplace operator leads to remarkably well-conditioned matrices even when the number of equations is very large. Numerical experiments were performed supporting the analysis. It is concluded that the methods studied are virtually limited to large finite difference systems, but that they are very well adapted to such systems.
Read full abstract