No label defined (Q4383439)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1138094
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | No label defined |
scientific article; zbMATH DE number 1138094 |
Statements
1 April 1998
0 references
smallest modulus eigenvalue
0 references
eigenvector
0 references
large sparse matrix
0 references
inexact inverse iteration method
0 references
numerical experiments
0 references
0.9262787
0 references
0.9259503
0 references
0.9250908
0 references
0.92435527
0 references
0.91558325
0 references
0.91466016
0 references
0.9103247
0 references
The paper is focused on the computation of the smallest modulus eigenvalue \(\lambda_n\) and its associated eigenvector for a large sparse \(n\times n\) matrix \(A\). In the classical inverse iteration method (INVIT), the solution \(v_{k+1}\) of an auxiliary inner problem \(Av_{k+1}=u_k\) is used to calculate a value \(\beta_{k+1}\) and a vector \(u_{k+1}\) tending, if \(k\to \infty \), to \(\lambda_n^{-1}\) and to the \(\lambda_n\)-associated eigenvector. The authors propose to solve the inner problem iteratively with a tolerance \(\varepsilon_k\) dependent on \(\beta_{k}\), \(\varepsilon_{k-1}\), and \(\lambda_{n-1}\), i.e., the eigenvalue with the second smallest modulus. It is proved that if the eigenvalues of \(A\) are linearly independent, \({|}\lambda_{n-1}{|}>{|}\lambda_n{|}\), and a starting vector \(u_0\) fulfills certain assumption, then the proposed algorithm converges linearly. Moreover, the authors derive two practical formulae to get rid of \(\lambda_{n-1}\) and to make the calculation of \(\varepsilon_k\) simple. They also introduce a modified algorithm for solving generalized eigenvalue problems. Three numerical experiments are presented. They show that the inexact inverse iteration method saves at least half the computational cost if compared with INVIT.
0 references