Bounds on variation of spectral subspaces under \(J\)-self-adjoint perturbations (Q1037200): Difference between revisions

Let \(A\) be a selfadjoint operator in a Hilbert space and assume that the spectrum of \(A\) consists of two disjoint components \(\sigma_0\) and \(\sigma_1\). In the present paper, the authors study the effect of additive bounded perturbations \(V\) of \(A\) that satisfy \[ V^*=JVJ,\qquad\text{where}\qquad J:=E_A(\sigma_0)-E_A(\sigma_1), \] on the spectrum of \(A\). Here, \(E_A(\cdot)\) denotes the spectral function of the selfadjoint operator \(A\). The main purpose is to discuss the following problems: {\parindent=8mm \begin{itemize}\item[(i)]Does the spectrum of the perturbed operator \(L=A+V\) remain real? \item[(ii)]If yes, is then \(L\) similar to a selfadjoint operator? \item[(iii)]What are the (sharp) bounds on variation of the spectral subspaces associated with \(\sigma_0\) and \(\sigma_1\)? \end{itemize}} Questions (i) and (ii) are answered in the positive in the case that \(\sigma_0\) and \(\sigma_1\) are subordinated, i.e., \(\sup\sigma_0 < \inf\sigma_1\), or one of the sets lies in a finite gap of the other. Also, the general case is partly answered in the positive, e.g., when \(A\) is bounded. One of the key ingredients in the proofs is an operator Riccati equation associated with the perturbed operator. Furthermore, the notion of operator angles between subspaces as well as methods and ideas from operator theory in Krein spaces are used.