Decreasing diagonal elements in completely positive matrices (Q1284621)

A real semidefinite matrix \(A\) is called positive if it can be written as a product \(VV^{\top}\) where \(V\) is a (not necessarily square) real matrix; it is completely positive if, in addition, \(V\) can be chosen with its entries nonnegative. The main theorems of this paper deal with extremal properties of such matrices. The author proves that, if \(A\) is a singular positive semidefinite matrix, then there is some diagonal entry of \(A\) such that \(A\) is no longer positive semidefinite when this entry is replaced by any smaller value. It is also shown that in the class of completely positive semidefinite \(n\times n\) matrices there exist matrices \(A\) such that any reduction in the \(i\)th diagonal entry of \(A\) results in a matrix which is no longer in this class. Examples show that, in general, a matrix which is extremal in the second class (completely positive) is not always extremal in the first class (positive), but some conditions under which these extremal properties do coincide are given.