A volume associated with \(m{\times}n\) matrices (Q1187510)

The ''volume'' of \(m\times n\) real matrix \(A\) with \(r\) nonzero singular values \(\sigma_ 1, \sigma_ 2,..., \sigma_ r\) is zero of \(r = 0\), and otherwise \(volA = \prod_{i=1}^ r {\sigma_ i}\). The \(volA\) generalizes the ''absolute value of the determinant'' from nonsingular to arbitrary matrices. Its applications include: computing principal angles between subspaces; checking the existence of integer solutions; counting spanning of trees in a digraph; exact computation of the Moore-Penrose inverse \(A^{\dag}\) using residue arithmetic. The author concentrates on ''determinantal properties'' of the matrix volume. Using some preliminary results about full-rank factorizations and connections with exterior products and compound matrices he proves the main results which say that the volume can be written as a single determinant of a matrix obtained from \(A\) by bordering (Theorem 4.1) or complementing (Theorem 4.2). Moreover he presents some applications of geometric nature of the volume: a generalized Hadamard inequality; orthogonal projections; height of a right n-simplex and gives representations of the Moore-Penrose inverse. The main result here is Theorem 6.2, stating that both \(A\) and \(A^{\dag}\) have the same weights in their convex decompositions.