scientific article; zbMATH DE number 108299
The author analyses the accuracy of the computed eigensolution of a real symmetric matrix, and shows how to compute tiny eigenvalues more accurately than until now thought possible. The thesis begins by describing the relative perturbation theory. A matrix is called well-behaved if the small relative changes in its elements cause small relative changes in its eigenvalues. It is shown that a real symmetric non-singular matrix \(H\) is well-behaved if the matrix \(A=D\cdot \hat H \cdot D\), where \(\hat H\) is the symmetric positive definite polar factor of \(H\), and \(D\) is a diagonal scaling such that \(diag(A)\) equals an identity, is well-conditioned. More precisely, if \(\lambda\) and \(\lambda + \delta \lambda\) are equally ordered eigenvalues of \(H\) and \(H+ \delta H\), respectively, where \(\delta H\) is symmetric and \(| \delta H_{ij}| \leq \varepsilon | H_{ij} |\), then \( |\delta \lambda | \leq n\cdot \varepsilon \cdot | A^{-1} |_ 2 \cdot |\lambda |\). This bound is never much worse, and can be much better that its standard norm-based counterpart. The corresponding norm perturbation bounds for the eigenprojections are given. Similar perturbation bounds are also proved for the case when \(H\) is given in the factorized form, \(H=G\cdot J \cdot G^ T\), where \(G\) has the full column-rank, \(J=diag(\pm 1)\), and only the elements of the factor \(G\) are perturbed in the relative sense. The author then gives the subtle backward error analysis of the algorithm introduced by \textit{K.\ Veselić} [A Jacobi eigenreduction algorithm for definite matrix pairs, Numer.\ Math.\ 64, 241-269 (1993)], which consists of the symmetric indefinite decomposition \(H=G\cdot J \cdot G^ T\) with complete pivoting, followed by the one-sided (implicit) Jacobi type iteration on the pair \(G,J\) which uses trigonometric and hyperbolic rotations. Fast and fast self-scaling rotations are also analysed. The error analysis shows that this algorithm computes the eigensolution nearly as accurately as predicted by the perturbation theory. Numerical experiments confirm all theoretical results, give strong numerical evidence for some phenomena for which only partial theoretical explanations exist, and show that the chosen algorithm computes tiny eigenvalues uniformly more accurately than the QR method or the standard Jacobi method applied directly to \(H\).
- Relative eigenvalue and singular value perturbations of scaled diagonally dominant matrices
- Implicit standard Jacobi gives high relative accuracy
- A bound for the condition of a hyperbolic eigenvector matrix
- Spectral sensitivity of products of bidiagonals
- Spectral variations and Hadamard products: Some problems
- A Jacobi eigenreduction algorithm for definite matrix pairs
- Componentwise analysis of direct factorization of real symmetric and Hermitian matrices
- Relative perturbation theory. IV: \(\sin 2\theta\) theorems
- Perturbations of the eigenprojections of a factorized Hermitian matrix
- A note on the accuracy of symmetric eigenreduction algorithms
- Weyl-type relative perturbation bounds for eigensystems of Hermitian matrices
- Relative perturbation theory for hyperbolic eigenvalue problem
- Perturbation theory for the eigenvalues of factorised symmetric matrices
- Rounding-error and perturbation bounds for the indefinite QR factorization
- Approximate eigenvectors as preconditioner
- On the global convergence of the block Jacobi method for the positive definite generalized eigenvalue problem
- Numerical methods for accurate computation of the eigenvalues of Hermitian matrices and the singular values of general matrices
- A Kogbetliantz-type algorithm for the hyperbolic SVD
- Accurate eigenvalues of some generalized sign regular matrices via relatively robust representations
- Jacobi algorithm for symmetric eigenvalue problem and integrable gradient system of Lax form
- Convergence to diagonal form of block Jacobi-type methods
- Full block \(J\)-Jacobi method for Hermitian matrices
- Accelerating the SVD block-Jacobi method
- Indefinite QR factorization
- Quadratic convergence estimate of scaled iterates by \(J\)-symmetric Jacobi method
- Revisiting the (block) Jacobi subspace rotation method for the symmetric eigenvalue problem
- Jacobi’s Method is More Accurate than QR
- Three-level parallel J-Jacobi algorithms for Hermitian matrices
- An Orthogonal High Relative Accuracy Algorithm for the Symmetric Eigenproblem
- scientific article; zbMATH DE number 1786137 (Why is no real title available?)
- Accurate Eigensystem Computations by Jacobi Methods
- Accuracy of the Jacobi method on scaled diagonally dominant symmetric matrices
- scientific article; zbMATH DE number 2209731 (Why is no real title available?)
- Convergence of the complex block Jacobi methods under the generalized serial pivot strategies
- An indefinite LOBPCG type of algorithm for detecting a definite Hermitian matrix pair
This page was built for publication:
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4023760)
