Repeated eigenvectors and the numerical range of self-adjoint quadratic operator polynomials (Q1865923)

Let \({\mathcal L}(H)\) be the algebra of bounded linear operators on a Hilbert space \(H\), and let \(L(\lambda): {\mathbb C}\to {\mathcal L}(H)\) be a polynomial. The numerical range of \(L\) is defined as \(W(L)=\{ \lambda\in{\mathbb C}: (L(\lambda)x, x)=0, x\in H, \|x\|=1\}\). The classical numerical range of an operator \(A\in {\mathcal L}(H)\) is then recovered by considering \(L(\lambda)=I\lambda -A\). By definition, eigenvalues of \(L\) are now points \(\lambda_0\in {\mathbb C}\) such that \(\text{Ker} L(\lambda_0)\) is nontrivial. In this paper, the authors consider the numerical range of self-adjoint quadratic operator polynomials \(L\), i.e., of the form \(L(\lambda)= \lambda^2I+\lambda A_1+A_0\), where \(A_0\) and \(A_1\) are self-adjoint operators in \({\mathcal L}(H)\). The main concern is with properties of the eigenvalues on \(\partial W(L)\). The investigation requires a careful discussion of repeated eigenvectors of more general operator polynomials. It is shown that, in the self-adjoint quadratic case, non-real eigenvalues on \(\partial W(L)\) are semisimple and (in some sense) normal. Also, for any eigenvalue at a point on \(\partial W(L)\) where an external cone property is satisfied, the partial multiplicities cannot exceed two.