Normal matrix compressions (Q2873569)

In general, a \textit{compression} of an operator \(T\) acting on a Hilbert space \(H\) is any operator of the form \(PT|_{PH}\), where \(P\) is an orthogonal projection acting on \(H\). The rank-\(k\) numerical range \(\Omega_k(T)\) of \(T\) is then defined as the set of all \(a\in{\mathbb C}\) for which \(T\) admits a scalar compression \(aI\) with \(P\) of rank \(k\). Obviously, \(\Omega_1(T)\) coincides with the standard numerical range \(W(T)\) of \(T\).NEWLINENEWLINEThe paper deals with normal compressions of normal operators in the case of finite dimensional \(H\) and \(k=2\). More specifically, the properties of the set NEWLINE\[NEWLINE B(a)=\left\{ b: \begin{bmatrix} a & 0 \\ 0 & b \end{bmatrix} \text{ is a compression of } T\right\} NEWLINE\]NEWLINE are investigated. It is shown, among other things, that if every point of the spectrum of \(T\) is extreme for \(W(T)\) and \(a\in \Omega_2(T)\), then \(B(a)\supseteq\Omega_2(T)\).