Hilbert-Schmidt interpolation in CSL-algebras (Q2640851)

Let \({\mathcal H}\) be a separable Hilbert space and \({\mathcal A}\) an algebra of bounded linear operators on \({\mathcal H}\). The type of interpolation problem considered here is as follows: Given x,y\(\in {\mathcal H}\), does there exist \(T\in {\mathcal A}\) such that \(Tx=y\) and, if so, what is the smallest possible norm of such an operator T? This was solved by the author when \({\mathcal A}\) is the algebra Alg \({\mathcal L}\) of operators leaving invariant the elements of a commutative subspace lattice \({\mathcal L}\) (a so-called CSL algebra) [Indiana Univ. Math. J. 29, 121-126 (1980; Zbl 0431.47025)], following an earlier solution by \textit{E. C. Lance} when \({\mathcal A}\) is a nest algebra. Recently, \textit{N. J. Munch} solved the interpolation problem in the nest algebra setting subject to the additional restriction that the interpolating operator T should belong to the Hilbert-Schmidt class [J. Math. Anal. Appl. 140, No.2, 407-418 (1989; Zbl 0691.47015)]. He interpreted some of his results in the context of linear system theory and, more particularly, in terms of signal reconstruction. The aim of the present paper is to extend most of Munch's results to CSL algebras. The results are given in terms of the Arveson representation of a CSL algebra. More precisely, let X be a compact metric space with a closed reflexive transitive ordering \(\leq\) and let \(\mu\) be a finite Borel measure on X. A Borel subset E of X is said to be increasing if \(y\in E\) whenever \(x\in E\) and \(x\leq y\). With \({\mathcal H}=L^ 2(X,\mu)\), let \({\mathcal L}(X,\leq,\mu)\) denote the family of projections on H consisting of the multiplication operators associated with the characteristic functions of increasing Borel subsets of X. Every commutative subspace lattice is unitarily equivalent to a lattice of the form \({\mathcal L}(X,\leq,\mu)\). With this notation, the main result of the paper can be stated as follows: Let \(x,y\in L^ 2(X,\mu)\) and define \(N_ x(t)=\int | x(s)|^ 2G(s,t)d\mu (s)\) for \(t\in X\), where \(G(s,t)=1\) (s\(\leq t)\), \(=0\) otherwise. Then there is a Hilbert-Schmidt operator T in Alg \({\mathcal L}(X,\leq,\mu)\) such that \(Tx=y\) if and only if \(C_{x,y}\equiv \int | y(t)|^ 2/N_ x(t)d\mu (t)\) is finite, interpreting 0/0 in the integrand as 0. Furthermore, \((C_{x,y})^{1/2}\) (if finite) is the minimum value for the Hilbert-Schmidt norm of such an interpolating operator T and is attained when T is the operator associated with the kernel \[ h(s,t)=\{\overline{x(s)}y(t)G(s,t)\}/N_ x(t). \] A less complete solution is also given of the corresponding interpolation problem for n vectors. The author ends by raising some questions and making some comments concerning the analogous interpolation problem for the other Schatten \(C_ p\)-classes, particularly for trace class operators. Finally, there is a brief discussion of the notion of approximate \(C_ 2\)-interpolation by elements of a CSL algebra.