A Proposal of a Single-Synchronized Solver Suited to Large Scale Linear Systems on Parallel Computers with Distributed Memory
Seiji Fujino, Murakami Keiichi, Kosuke Iwasato · Advances in parallel computing · 2014
We propose a new iterative method based on two strategies, improvement stabilization by applying two-term recurrences and utilization of associate residual. The former adopts the coupled two-term recurrences proposed by Rutishauser as stabilizing polynomials, and the latter adopts the associate residual vector for computation of the recurrence parameters ζk,ηkinstead of that of the residual vector. Through numerical experiments, we make clear that the proposed method outperforms other methods from the view points of computational and elapsed time and speed-up on both serial and parallel computers.