A new metod (NI-DACG) for the partial eigensolution of large sparse symmetric FE eigenproblems is presented. NI-DACG relies on the optimization of Rayleigh quotients in successively deflated subspaces by a preconditioned conjugate gradient technique and uses a multiple grid type approach to asses an improved eigenvector estimate on nested FE grids on wich the solution to the continuous eigenproblem is sought. NI-DACG is implemented on the CRAY Y-MP supercomputer making use of vectorization and/or parallelization with two and four processors. Results relative to the calculation of the 50 smallest eigenpairs for two representative sample problems show a gain in CPU time that exceeds one order of magnitude with respect to the scalar implementation of NI-DACG and emphasize the promising features of this technique for the partial eigenanalysis on supercomputers.
Parallel eigenanalysis for nested grids
PINI, GIORGIO;GAMBOLATI, GIUSEPPE
1993
Abstract
A new metod (NI-DACG) for the partial eigensolution of large sparse symmetric FE eigenproblems is presented. NI-DACG relies on the optimization of Rayleigh quotients in successively deflated subspaces by a preconditioned conjugate gradient technique and uses a multiple grid type approach to asses an improved eigenvector estimate on nested FE grids on wich the solution to the continuous eigenproblem is sought. NI-DACG is implemented on the CRAY Y-MP supercomputer making use of vectorization and/or parallelization with two and four processors. Results relative to the calculation of the 50 smallest eigenpairs for two representative sample problems show a gain in CPU time that exceeds one order of magnitude with respect to the scalar implementation of NI-DACG and emphasize the promising features of this technique for the partial eigenanalysis on supercomputers.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
