Eigen-pairs of perturbed matrices (Q2567472)

Let \(\lambda_0\) be an eigenvalue of \(A \in {\mathbb C}^{n \times n}\) with corresponding eigenvector \(x_0\) of norm \(1\). The effect on \(\lambda_0\) and \(x_0\) is looked for when \(A\) is perturbed by another matrix \(E \in {\mathbb C}^{n \times n}\). It is well known that if the norm \(| | E| | \) is small enough then \(A+E\) has an eigenvalue \(\lambda\) which is close to \(\lambda_0\). The closeness of vectors can be measured by the norm of their difference. The method adopted here for finding the perturbed eigen-pair \((\lambda, x)\) is a slight modification of \textit{G. V. Stewart}'s method [Technical Report: TR-1923, Dept. Comp. Sci., Univ. of Maryland (1987)] and is also simpler than the analysis carried out by \textit{S. R. Varadharaj} [Perturbation of eigenvalues and invariant subspaces, M. Sc. Dissertation, I. I. T. Madras (2002)]. A main feature of the result is that for a quantity \(\epsilon_0\) specified such that for all \(E\) with \(| | E| | < \epsilon_0\) the errors \(| \lambda - \lambda_0| \) and \(| | x-x_0| | \) are expressed in terms of \(| | E| | \) also ensures that the perturbed eigen-pair \((\lambda, x)\) is simple.