Parallel Iterative Methods in the ELPA Eigensolver
Abstract:
The solution of symmetric eigenproblems plays a key role in many computational simulations. Generalized eigenproblems are transformed to a standard problem. This transformation has the drawback that for banded matrices in the generalized eigenproblem the banded structure is not preserved. The matrix of the standard eigenproblem will generally be a full matrix. We followed the ideas of the Group of Lang (University of Wuppertal) who modified Crawford’s algorithm and implemented an iterative procedure to the ELPA project which keeps the bandwith. By keeping the banded structure we save one reduction step on the matrix and one acktransformation step for the eigenvectors. This provides a good speedup compared to the standard tranformation procedure with Cholesky factorization.
«
The solution of symmetric eigenproblems plays a key role in many computational simulations. Generalized eigenproblems are transformed to a standard problem. This transformation has the drawback that for banded matrices in the generalized eigenproblem the banded structure is not preserved. The matrix of the standard eigenproblem will generally be a full matrix. We followed the ideas of the Group of Lang (University of Wuppertal) who modified Crawford’s algorithm and implemented an iterative proc...
»
Kongress- / Buchtitel:
SIAM Conference on Parallel Processing
Ausrichter der Konferenz:
SIAM
Verlagsort:
Tokio
Publikationsdatum:
07.03.2018
Jahr:
2018
Quartal:
1. Quartal
Jahr / Monat:
2018-03
Monat:
Mar
Reviewed:
nein
Sprache:
en
Hinweise:
mediatitle: SIAM PP address: Tokio TYPE OF CONTRIBUTION: contributed