Rigorous perturbation bounds for eigenvalues and eigenvectors of a matrix (Q1893599): Difference between revisions

If \(A\) has the eigenpair \((\lambda, x)\), i.e., the eigenvalue \(\lambda\) and the eigenvector \(x\), then it is of great interest to know bounds for the shifts \(| \widetilde \lambda - \lambda |\) and \(| \widetilde x - x |\), where \((\widetilde \lambda, \widetilde x)\) is the eigenpair of \(A + \delta A\), in terms of \(\delta A\). In this work, the author tries to provide strict bounds for \(| \widetilde \lambda_ i - \lambda_ i |\) and \(| \widetilde x_ i - x_ i |\) which are valid for big \(\delta A\). Namely \(\min | \widetilde \lambda - \lambda | \leq \| | T^{-1} | | \delta A | | T | \|\) (for the whole spectrum), where \(T\) is the modal matrix containing the eigenvectors of \(A\) and which brings it to a diagonal form under a similarity transform, i.e., \(T^{-1} AT = \Lambda\). For the individual eigenvalue \(\lambda_ i\) and corresponding eigenvector \(x^ i\), the bounds are \(| \delta \lambda_ i | \leq | y^{i*} | | \delta A | \sum^ n_{j = 1} | x^ j |\) and \(| \delta x^ i | \leq \sum_{j\neq i} | \alpha_ j | | x^ j |\) where \(y^{i*}\) is the eigenrow of \(A\) corresponding to \(\lambda_ i\) and \(\delta x^ i = \sum_{j \neq i} \alpha_ j x^ j\). Any eigenvalue problem \(L(\lambda)x = (A_ n \lambda^ n + A_{n - 1} \lambda^{n - 1} + \cdots + A_ 0)x = 0\) can be transformed into the form \(Au = \lambda Bu\). Bounds of this generalized eigenvalue problem are also obtained. It is worth noting that the famous order mechanical system \((M \lambda^ 2 + R \lambda + K) x = 0\) is a special one of \(L(\lambda) x = 0\).