Best approximation of a normal operator in the trace norm

From MaRDI portal





This paper gives a characterization of those normal operators which have a unique positive approximant (i.e., a unique best approximation among the non-negative operators) in the trace norm. Let \(H\) be a separable Hilbert space. For \(A\) an operator on \(H\), let \(| A|\) denote the unique positive square root of \(A^*A\), and let \(A^+=(1/2)(| A| +A)\). For a given operator \(T\), with Cartesian decomposition \(T=B+iC\), the following conditions are equivalent: (i) there exists a non-negative operator \(P\) such that \(T-P\) is trace class; (ii) the operator \(C\) is trace class and the spectrum \(\sigma(B)\) of \(B\) not in the interval \([0,\infty)\) consists of isolated eigenvalues, say \(\lambda_ j\) repeated according to multiplicity, such that \(\sum_{j}| \lambda_ j| <\infty;\) (iii) the operator \((T-B^+)\) is trace class. If the operator \(T\), satisfying one of the conditions (i)-(iii) is normal, then there is a unique trace class positive approximant for \(T\) if and only if \(\sigma(T)\) is contained in one or the other of the sets \(\{z:\) either im\(z\geq 0\) or re\(z\leq 0\}\); \(\{z:\) either im\(z\leq 0\) or re\(z\leq 0\}\).











This page was built for publication: Best approximation of a normal operator in the trace norm

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3788564)