Abstract:
New approaches to the development of highly efficient parallel algorithms for numerical solution of boundary value problems are considered. The previously proposed robust multigrid technique (a single grid version of the Seidel method for solving a large class of applied problems with close-to-optimal computational efforts) is taken as a basic algorithm. Two approaches for the parallelization of computations based on combined and purely geometric preconditioning are studied. Advantages of these approaches compared to traditional methods of constructing parallel algorithms are shown. Several estimates for the parallelism efficiency are proposed.
Keywords:parallel computing, boundary value problems, multigrid methods.
Fulltext:
PDF file (426 kB)