Is modified PSD equivalent to modified SOR for two-cyclic matrices? (Q967498): Difference between revisions

The authors consider an iterative method with stationary 4-parameter iteration matrix called modified preconditioned simultaneous displacement (MPDS) method which contains the successive overrelaxation (SOR) and modified symmetric SOR (MSSOR) as special cases. They prove, for iteration matrices with either purely real or purely imaginary eigenvalues, necessary and sufficient convergence conditions for MPDS and determine, for both cases of the spectrum, optimal iteration parameters. Between others, it turns out that MPDS may be faster than MSSOR. But the question of the title is answered affirmatively by showing that optimum MPDS and an optimal SOR version (as modified by Young) have the same rate of convergence. Numerical results concern a 64 times 64 discretization of a 2D constant coefficient diffusion-convection problem showing MPDS to be indeed faster than MSSOR. However, there is no comparison, by computing times, of the considered methods and of simple \texttt{MATLAB} A\(\backslash\)b or multigrid.