Normal and seminormal eigenvalues of matrix functions

From MaRDI portal





Let \(P(\lambda)\) be an \(n\)-by-\(n\) matrix-valued function analytic on a domain \(\Omega\) of \(\mathbb{C}\) and with \(\det P(\lambda_0)\neq 0\) for some \(\lambda_0\) in \(\Omega\). The spectrum \(\sigma(P)\) of \(P(\lambda)\) is by definition the set \(\{\lambda\in \Omega:P (\lambda)x=0\) for some \(x\neq 0\) in \(\mathbb{C}^n\}\), and the numerical range \(W(P)\) of \(P(\lambda)\) is \(\{\lambda \in \Omega:x^*P(\lambda) x=0\) for some \(x\neq 0\) in \(\mathbb{C}^n\}\). When \(P(\lambda) =\lambda I_n-A\) for some \(n\)-by-\(n\) matrix \(A\), \(\sigma(P)\) coincides with the set of eigenvalues of \(A\) and \(W(P)\) the classical numerical range of \(A\). It is known that eigenvalues of \(A\) on the boundary of \(W(A)\) are normal, meaning that the corresponding eigenspace reduces \(A\). The authors investigate to what extent this can be generalized to \(P(\lambda)\). The main theorem (Theorem 6) says that any \(\lambda_0\) in \(\sigma(P)\) which is on the boundary of \(W(P)\) is seminormal, meaning that there is a unitary matrix \(U\) such that \[ U^*P(\lambda) U=\left[ \begin{matrix} (\lambda- \lambda_0)P_1 (\lambda)& (\lambda-\lambda_0) P_2( \lambda) \\ (\lambda-\lambda_0) P_3(\lambda) & P_4(\lambda) \end{matrix} \right] \] for all \(\lambda\) in \(\Omega\), where \(P_1(\lambda)\) is of size the algebraic multiplicity of \(\lambda_0\) (= multiplicity of \(\lambda_0\) as a zero of \(\det P(\lambda))\), and the \(P_j(\lambda)\)'s are analytic matrix functions on \(\Omega\) with \(\lambda_0\notin\sigma(P_4)\). A partial converse of this is also proved. More refined results are also derived for the special case of \(P(\lambda)= \lambda^2 I_n+\lambda A_1+A_0\), where \(A_1\) and \(A_0\) are Hermitian.




Cited in
(23)








This page was built for publication: Normal and seminormal eigenvalues of matrix functions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5952399)