On the interplay between operators, bases, and matrices (Q1982493)

Let \(T\) be a bounded linear operator on a Hilbert space \(H\) (denoted by \(T \in B(H)\)). If \(\{u_n\}_{n=1}^\infty\) is an orthonormal basis for \(H\), then the sequence \(\{\langle T u_n, u_n \rangle\}_{n=1}^\infty\) is called the main diagonal of \(T\). Evidently, the main diagonal is the diagonal of the matrix representation \(\{\langle T u_n, u_m \rangle\}_{n,m=1}^\infty\) of \(T\) with respect to \(\{u_n\}_{n=1}^\infty\). The theory of matrix representations of bounded linear operators has a long and rich history. However, in the present context, the authors are further motivated by their recent work [\textit{V. Müller} and \textit{Y. Tomilov}, Trans. Am. Math. Soc. 372, No. 5, 3565--3595 (2019; Zbl 1437.47004)], where they proved that a sequence from the interior of the essential numerical range of \(T\) which approaches the boundary not too fast, satisfying the so-called Blaschke-type condition, forms the main diagonal of \(T\). The non-Blaschke type condition is the second equality stated below. In the paper under review, in addition to producing the main diagonal, the authors prove that (in an appropriate sense and with the same non-Blaschke-type condition) the matrix of \(T\) can be sparsified by arranging zero matrix elements in any band outside of the main diagonal. In order to be more specific, we recall that for \(T \in B(H)\), the essential numerical range \(W_e(T)\) is defined by \[ W_e(T) = \{\lambda \in \mathbb{C}: \langle Te_n, e_n \rangle \rightarrow 1 \text{ as } n \rightarrow \infty\}, \] where \(\{e_n\}_{n=1}^\infty\) is an orthonormal sequence for \(H\). Note that \(W_e(T)\) is a non-empty, compact, and convex subset of \(\mathbb{C}\) and also contains the essential spectrum of \(T\). Here is the first sample result: Let \(T \in B(H)\), \(\{\lambda_n\}_{n=1}^\infty \subseteq \text{Int} W_e(T)\), and suppose \[ \sum_{n=1}^{\infty} \text{ dist} \{\lambda_n, \partial W_e(T)\} = \infty. \] Then for every \(K \in \mathbb{N}\), there exists an orthonormal basis \(\{u_n\}_{n=1}^\infty\) in \(H\) such that \[ \langle T u_n, u_n \rangle = \lambda_n \qquad (n \in \mathbb{N}), \] and \[ \langle T u_n, u_j \rangle = 0 \qquad (1 \leq |n-j| \leq K). \] As a corollary, the authors offer the following: Let \(T \in B(H)\) be such that \(0 \in \text{Int} W_e(T)\). Then for every \(K \in \mathbb{N}\), there exists an orthonormal basis \(\{u_n\}_{n=1}^\infty\) in \(H\) such that \[ \langle T u_n, u_j \rangle = 0 \] for all \(n, j \in \mathbb{N}\) with \(|n-j| \leq K\). Another deep result states: Suppose that \(T \in B(H)\) is not of the form \(T = \lambda + K\) for scalar \(\lambda\) and a compact operator \(K \in B(H)\). Then there exist an orthonormal basis \(\{u_n\}_{n=1}^\infty\) for \(H\) and strictly positive constants \(c_1, c_2\), and \(d\) depending only on the diameter of \(W_e(T)\) such that \(|\langle T u_n, u_n \rangle| \geq d\) for all \(n \in \mathbb{N}\) and \[ \frac{c_1 \text{min}\{n,j\}^{1/2}}{\text{max}\{n,j\}^{3/2}} \leq |\langle T u_n, u_j \rangle| \leq \frac{c_2}{\text{max}\{n,j\}^{1/2}} \] for all \(n, j \in \mathbb{N}\), \(n \neq j\). The authors state the following: All of the results are nontrivial even for \(T\) being a unilateral (forward) shift on the Hardy space. The paper gives a thorough historical background of the general problem and relevant connections to other subjects. The Introduction section itself is very pleasant to read. We also refer to another deep and motivational paper [\textit{Q. F. Stout}, Trans. Am. Math. Soc. 264, 39--47 (1981; Zbl 0481.47003)] and the related papers [\textit{W. Arveson}, Proc. Natl. Acad. Sci. USA 104, No. 4, 1152--1158 (2007; Zbl 1191.47027)] and [\textit{J. Jasper} et al., Indiana Univ. Math. J. 67, No. 1, 1--27 (2018; Zbl 06971376)]. This paper is a must-read for linear analysis researchers and even for general mathematicians.