Accuracy of computed eigenvectors via optimizing a Rayleigh quotient
The author proves the following result. Let \(A\) be a Hermitian matrix with largest eigenvalue \(\lambda_1\) and corresponding eigenvector \(u_1\). If a nonzero vector \(\widetilde u_1\) satisfies \[ \frac{\widetilde u_1^* A \widetilde u_1}{\widetilde u_1^* \widetilde u_1} \geq \lambda_1 - \varepsilon^2 \] for some \(\varepsilon > 0\) then \[ \sin \angle (u_1,\widetilde u_1) \leq \frac{\varepsilon}{\sqrt{\lambda_1-\lambda_2}}, \] where \(\lambda_2\) is the second largest eigenvalue of \(A\). In other words, \(\widetilde u_1\) is an \(O(\varepsilon)\) approximation to \(u_1\) if the Rayleigh quotient is an \(O(\varepsilon^2)\) approximation to \(\lambda_1\). It is shown that a similar results holds for dominant invariant subspaces.
- On generalizing trace minimization principles
- A self-consistent-field iteration for MAXBET with an application to multi-view feature extraction
- Cluster-robust accuracy bounds for Ritz subspaces
- Sharpness in rates of convergence for the symmetric Lanczos method
- Graph clustering
- Sharp error bounds for Ritz vectors and approximate singular vectors
- On generalizing trace minimization principles. II
- A theory of the NEPv approach for optimization on the Stiefel manifold
This page was built for publication: Accuracy of computed eigenvectors via optimizing a Rayleigh quotient
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1770929)
