arrow
Volume 18, Issue 4
A Process for Solving a Few Extreme Eigenpairs of Large Sparse Positive Definite Generalized Eigenvalue Problem

Chong-Hua Yu & O. Axelsson

J. Comp. Math., 18 (2000), pp. 387-402.

Published online: 2000-08

Export citation
  • Abstract

In this paper, an algorithm for computing some of the largest (smallest) generalized eigenvalues with corresponding eigenvectors of a sparse symmetric positive definite matrix pencil is presented. The algorithm uses an iteration function and inverse power iteration process to get the largest one first, then executes $m-1$ Lanczos-like steps to get initial approximations of the next $m-1$ ones, without computing any Ritz pair, for which a procedure combining Rayleigh quotient iteration with shifted inverse power iteration is used to obtain more accurate eigenvalues and eigenvectors. This algorithm keeps the advantages of preserving sparsity of the original matrices as in Lanczos method and RQI and converges with a higher rate than the method described in [12] and provides a simple technique to compute initial approximate pairs which are guaranteed to converge to the wanted $m$ largest eigenpairs using RQI. In addition, it avoids some of the disadvantages of Lanczos and RQI, for solving extreme eigenproblems. When symmetric positive definite linear systems must be solved in the process, an algebraic multilevel iteration method (AMLI) is applied. The algorithm is fully parallelizable.  

  • AMS Subject Headings

  • Copyright

COPYRIGHT: © Global Science Press

  • Email address
  • BibTex
  • RIS
  • TXT
@Article{JCM-18-387, author = {Yu , Chong-Hua and Axelsson , O.}, title = {A Process for Solving a Few Extreme Eigenpairs of Large Sparse Positive Definite Generalized Eigenvalue Problem}, journal = {Journal of Computational Mathematics}, year = {2000}, volume = {18}, number = {4}, pages = {387--402}, abstract = {

In this paper, an algorithm for computing some of the largest (smallest) generalized eigenvalues with corresponding eigenvectors of a sparse symmetric positive definite matrix pencil is presented. The algorithm uses an iteration function and inverse power iteration process to get the largest one first, then executes $m-1$ Lanczos-like steps to get initial approximations of the next $m-1$ ones, without computing any Ritz pair, for which a procedure combining Rayleigh quotient iteration with shifted inverse power iteration is used to obtain more accurate eigenvalues and eigenvectors. This algorithm keeps the advantages of preserving sparsity of the original matrices as in Lanczos method and RQI and converges with a higher rate than the method described in [12] and provides a simple technique to compute initial approximate pairs which are guaranteed to converge to the wanted $m$ largest eigenpairs using RQI. In addition, it avoids some of the disadvantages of Lanczos and RQI, for solving extreme eigenproblems. When symmetric positive definite linear systems must be solved in the process, an algebraic multilevel iteration method (AMLI) is applied. The algorithm is fully parallelizable.  

}, issn = {1991-7139}, doi = {https://doi.org/}, url = {http://global-sci.org/intro/article_detail/jcm/9051.html} }
TY - JOUR T1 - A Process for Solving a Few Extreme Eigenpairs of Large Sparse Positive Definite Generalized Eigenvalue Problem AU - Yu , Chong-Hua AU - Axelsson , O. JO - Journal of Computational Mathematics VL - 4 SP - 387 EP - 402 PY - 2000 DA - 2000/08 SN - 18 DO - http://doi.org/ UR - https://global-sci.org/intro/article_detail/jcm/9051.html KW - Eigenvalue, sparse problem. AB -

In this paper, an algorithm for computing some of the largest (smallest) generalized eigenvalues with corresponding eigenvectors of a sparse symmetric positive definite matrix pencil is presented. The algorithm uses an iteration function and inverse power iteration process to get the largest one first, then executes $m-1$ Lanczos-like steps to get initial approximations of the next $m-1$ ones, without computing any Ritz pair, for which a procedure combining Rayleigh quotient iteration with shifted inverse power iteration is used to obtain more accurate eigenvalues and eigenvectors. This algorithm keeps the advantages of preserving sparsity of the original matrices as in Lanczos method and RQI and converges with a higher rate than the method described in [12] and provides a simple technique to compute initial approximate pairs which are guaranteed to converge to the wanted $m$ largest eigenpairs using RQI. In addition, it avoids some of the disadvantages of Lanczos and RQI, for solving extreme eigenproblems. When symmetric positive definite linear systems must be solved in the process, an algebraic multilevel iteration method (AMLI) is applied. The algorithm is fully parallelizable.  

Chong-Hua Yu & O.Axelsson. (1970). A Process for Solving a Few Extreme Eigenpairs of Large Sparse Positive Definite Generalized Eigenvalue Problem. Journal of Computational Mathematics. 18 (4). 387-402. doi:
Copy to clipboard
The citation has been copied to your clipboard