Evaluation of Eigenvalue Routines for Large Scale Applications
暂无分享,去创建一个
The NASA structural analysis (NASTRAN∗) program is one of the most extensively used engineering applications software in the world. It contains a wealth of matrix operations and numerical solution techniques, and they were used to construct efficient eigenvalue routines. The purpose of this article is to examine the current eigenvalue routines in NASTRAN and to make efficiency comparisons with a more recent implementation of the block Lanczos aLgorithm. This eigenvalue routine is now availabLe in several mathematics libraries as well as in severaL commerciaL versions of NASTRAN. In addition, the eRA Y library maintains a modified version of this routine on their network. Several example problems, with a varying number of degrees of freedom, were selected primarily for efficiency bench-marking. Accuracy is not an issue, because they all gave comparable results. The block Lanczos algorithm was found to be extremely efficient, particularly for very large problems.
[1] Horst D. Simon,et al. Eigenvalue Problems and Algorithms in Structural Engineering , 1986 .
[2] D. Eng,et al. \computing Eigenvalues of Very Large Symmetric Matrices { an Implementation of a Lanczos Algorithm with No Reorthogonalization," , 1996 .
[3] J. Cullum,et al. A survey of Lanczos procedures for very large real ‘symmetric’ eigenvalue problems , 1985 .
[4] J. G. Lewis,et al. A Shifted Block Lanczos Algorithm for Solving Sparse Symmetric Generalized Eigenproblems , 1994, SIAM J. Matrix Anal. Appl..