A note on infinite extreme correlation matrices (Q924323): Difference between revisions

Let \(I\) be either a finite or a countable (i.e., \(\mathbb{N}\) without loss of generality) index set and \(F= \mathbb{R}\) or \(\mathbb{C}\). A correlation matrix is a positive semidefinite Hermitian (symmetric if \(F= \mathbb{R}\)) mapping \(C: I^2\to F\) with \(C(i,i)= 1\) for all \(i\in I\). The set of correlation matrices is convex and the problem of determining its extreme points has been studied extensively, but mainly for finite matrices. The authors discuss the relevant results. The main result of this paper is a characterization of extreme correlation matrices for \(I= \mathbb{N}\) and \(F= \mathbb{R}\) or \(\mathbb{C}\), which generalizes a result of \textit{C.-K. Li} and \textit{B.-S. Tam} [SIAM J. Matrix Anal. Appl. 15, No.3, 903--908 (1994; Zbl 0804.15011)]. As a corollary the authors show that there exist extreme matrices of any rank \((\in\mathbb{N}\cup\{\infty\})\).