WorkshopsIterative Methods 2: Comparative Evaluation on GA-based AT for GMRES
70
reads
GMRES, Generalized Minimal RESidual algorithm, is very popular in actual use in sparse matrix computations in science and engineering, but, performs very poorly with some badly selected values of initial vectors and Krylov dimensions. This talk introduces our newly proposed AT using GA for GMRES in size of populations of initial vectors and shows evaluations in comparison with full search of 5 preconditioners times 300 Krylov dimensions with the corresponding randomly chosen initial vectors. Experiments on some unsymmetric matrices in Florida collection show that the GA of initial vectors with 15 populations is quite effective in convergence as well as predictive parallel performance.
reads
Ken Naono
2013-03-28
15:25:00 - 15:50:00
101 , Mathematics Research Center Building (ori. New Math. Bldg.)
GMRES, Generalized Minimal RESidual algorithm, is very popular in actual use in sparse matrix computations in science and engineering, but, performs very poorly with some badly selected values of initial vectors and Krylov dimensions. This talk introduces our newly proposed AT using GA for GMRES in size of populations of initial vectors and shows evaluations in comparison with full search of 5 preconditioners times 300 Krylov dimensions with the corresponding randomly chosen initial vectors. Experiments on some unsymmetric matrices in Florida collection show that the GA of initial vectors with 15 populations is quite effective in convergence as well as predictive parallel performance.