Eigenvalues of rank one perturbations of unstructured matrices (Q426046): Difference between revisions

Let \(A\in {\mathbb C}_{n\times n}\) be a fixed complex matrix and let \(u, v\in {\mathbb C}^n\) be two vectors. The authors study the spectral structure of \(B(\tau):=A + \tau uv^\top\), \(\tau \in {\mathbb R}\). They give a simple proof of a result of \textit{S. V. Savchenko} [Sb. Math. 196, No. 5, 743--764 (2005); translation from Mat. Sb. 196, No. 5, 121--144 (2005; Zbl 1087.15009)], which asserts that the Jordan structure of \(B(\tau)\) for ``generic'' \(u,v\) at the eigenvalues of \(A\) is constant for all \(\tau\not=0\). Generic \(u,v\) here means that there exists a generic subset \(\Omega\subset {\mathbb C}^{2n}\) (in the sense that \(\Omega\) is not empty and its complement is contained in a complex algebraic set which is not \({\mathbb C}^{2n}\)) possibly dependent on \(A\), such that for \((u,v)\in \Omega\). They also show that for generic \(u,v\), the eigenvalues of \(B(\tau)\) that are no eigenvalues of \(A\) tend with \(\tau\to \infty\) to the roots of the polynomial \(p_{uv}:=m(\lambda)v^\top(\lambda-A)^{-1}u\), where \(m(\lambda)\) is the minimal polynomial of \(A\), except one eigenvalue that goes to infinity. They establish that for generic \(u,v\) there are no tripe eigenvalues in \(\sigma(B(\tau))\setminus \sigma(A)\) for all \(\tau\in {\mathbb C}\). So there are generically no triple crossings of the eigenvalue curves. They also prove that for generic \(u,v\) there are at most \(2 \deg m - 2\) values of the parameter \(\tau\) for which there exists an eigenvalue of \(B(\tau)\) of multiplicity at least two, which is not an eigenvalue of \(A\).