Effectively Well-Conditioned Linear Systems (Q3814652)
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 4086888
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Effectively Well-Conditioned Linear Systems |
scientific article; zbMATH DE number 4086888 |
Statements
Effectively Well-Conditioned Linear Systems (English)
0 references
1988
0 references
error analysis
0 references
well-conditioning
0 references
condition number
0 references
sensitivity
0 references
Vandermonde matrices
0 references
fast Poisson solver
0 references
fast sine transforms
0 references
0.8291246891021729
0 references
0.8272063732147217
0 references
0.7775852680206299
0 references
0.768110454082489
0 references
0.7639716267585754
0 references
The authors discuss effective well-conditioning of the linear system \(Ax=b\). The condition number \(K(A)=\| A\| \| A^{-1}\|\) is often an overly conservative measure of the sensitivity of x under perturbations of \(\Delta\) A and \(\Delta\) b to A and b respectively. Two practical cases in which the sensitivity of x may be significantly less than the worst case predicted by K(A) are presented. The first characterizes a class of Vandermonde matrices and right-hand-sides and the second a FFT-based fast Poisson solver, for each of which accurate solutions may be obtained. For Vandermonde systems \textit{N. J. Higham} [Numer. Math. 50, 613-532 (1987; Zbl 0595.65029)] has shown that the algorithm of \textit{A. Björck} and \textit{V. Pereyra} [Math. Comput. 24, 893-903 (1971; Zbl 0221.65054)] gives relative errors in the non-zero components of x which are independent of K(A) provided that scalars \(\alpha_ j\) of the Vandermonde matrix are in ascending order and that the elements of the right-hand side b oscillate in sign. The computation of accurate solutions to the discretized one-dimensional Poisson problem is discussed and the authors' experiments indicate that a fast Poisson solver composed entirely of fast sine transforms will have better numerical performance than the more common fast transform with tridiagonal solving.
0 references