Modified successive overrelaxation (MSOR) and equivalent 2-step iterative methods for collocation matrices (Q1200181)

Some iterative methods for collocation matrices (which arise from the discretization of boundary value problems) are discussed. The modified successive overrelaxation (MSOR) method and a 2-step method are considered which have the same asymptotic rates of convergence. Hence the problem of determining the optimum values of the two parameters of the MSOR method is equivalent to that of finding the optimum parameters of the 2-step method. The equivalence between the problem of determining the optimum asymptotic rate of convergence and that of finding an optimum capturing ellipse is emphasized and a geometrical algorithm to solve this problem is given. The optimum MSOR is compared with the optimum successive overrelaxation (SOR), extrapolated Gauss-Seidel (EGS) and extrapolated accelerated Gauss-Seidel (EAGS) methods. It reveals that the optimum MSOR method is always faster than the optimum SOR and EGS methods and it is better than the optimum EAGS in all cases of practical interest. These results are verified for three examples arising from the discretization of boundary value problems by the finite-element collocation method, which include the Poisson equation in the unit square with Dirichlet boundary conditions.