PARALLELIZING OF PRECONDITIONED CONJUGATE GRADIENTS FOR POISSON-BOLTZMANN CALCULATION

摘要

This paper describes recent advances in the parallelization of the preconditioned conjugate gradient method for the Poisson-Boltzmann problem. Compared are three different kinds of symmetrical preconditioners: diagonal, incomplete Cholesky and alternative directions. The comparison uas conducted with respect to computational and memory requirements, convergence and scalability as a function of problem size and number of processors. Results of numerical experiments are presented for distributed-memory (Delta) and shared-memory (KSR-1) computers and demonstrate wide range of scalability. [References: 3]