ePubs

The open archive for STFC research publications

Full Record Details

DOI 10.5286/raltr.2010019
Persistent URL http://purl.org/net/epubs/work/53445
Record Status Checked
Record Id 53445
Title A preconditioned block conjugate gradient algorithm for computing extreme eigenpairs of symmetric and Hermitian problems
Contributors
Abstract This report describes an algorithm for the efficient computation of several extreme eigenvalues and corresponding eigenvectors of a large-scale standard or generalized real symmetric or complex Hermitian eigenvalue problem. The main features are: (i) a new conjugate gradient scheme specifically designed for eigenvalue computation; (ii) the use of the preconditioning as a cheaper alternative to matrix factorization for large discretized differential problems; (iii) simultaneous computation of several eigenpairs by subspace iteration; and (iv) the use of efficient stopping criteria based on error estimation rather than the residual tolerance.
Organisation CSE , CSE-NAG , STFC
Keywords preconditioned conjugate gradient method , Fortran 95 , Hermitian eigenvalue problems
Funding Information
Related Research Object(s):
Licence Information:
Language English (EN)
Type Details URI(s) Local file(s) Year
Report RAL Technical Reports RAL-TR-2010-019. 2010. orRAL2010019.pdf 2010