Analysis of a class of parallel multigrid smoothers (Q2566642)

The Jacobi successive overrelaxation (JSOR) method is a family of linear stationary iterative methods defined by domain decomposition. One important application of the JSOR method is to define various parallel smoothers for parallel multigrid methods. In this paper a new smoother for multigrid methods is suggested and analyzed by subspace decomposition techniques. The new smoother is called the parallel multiplicative (PM) smoother. It is shown that additive and multiplicative smoothers are two extreme cases of the PM smoother, and their smoothing theorems can be taken as the direct corollaries of the PM smoothing theorem presented in this paper. Numerical results are presented, which show that the JSOR smoother is more robust and effective than the damped-Jacobi smoother. The numerical experiments were done on an MIMD parallel computer for anisotropic model problem.