Parallel multigrid technique: Reduction to independent problems

Abstract

The unsatisfactory operation of a parallel multigrid algorithm is caused by two reasons: the imbalanced load of processors and the intensive exchanges of data between them. The further development of the parallel universal multigrid technique based on the reduction of a difference initial boundary value problem to a set of independent problems is considered. The universal multigrid technique is a single-grid algorithm, which uses the fundamental multigrid principle to minimize the number of problem-dependent components. The use of the same grid for the calculation of a correction eliminates all the difficulties produced by imbalanced loads and intensive exchanges on coarse grids. It has been shown that it is possible to decrease the volume of stored data and the time of computation and to attain nearly absolute parallelism in some cases. The results of some computational experiments with the difference six-order approximation pattern are presented.