On the Wedderburn-Guttman theorem (Q2575713)

The paper deals with some extension on the Wedderburn-Guttman theorem [cf. \textit{J. H. M. Wedderburn}, Lectures on matrices (1934; republ. 1964; Zbl 0121.26101); \textit{L. Guttman}, Psychometrika 9, 1-16 (1944; Zbl 0060.31212)]. This theorem says that if \(A\) is an \(m \times n\) matrix of rank \(a\), \(M\) and \(N\) matrices of sizes \(m \times p\) and \(n \times p\), respectively, such that \(M^TAN\) is nonsingular, then \[ \text{rank}(A-AN(M^TAN)^{-1}M^TA)=a-p, \] where \(p=\text{rank}(AN(M^TAN)^{-1}M^TA)= \text{rank}(M^TAN)\). In this paper the authors focus on the case in which \(M^TAN\) is rectangular or singular. Let \(M\) and \(N\) be \(m \times r\) and \(n \times s\) matrices, respectively, where \(r\) is not necessarily equal to \(s\), or \(\text{rank}(M^TAN)< min \{r,s\}\). The authors investigate conditions under which the regular inverse \((M^TAN)^{-1}\) can be replaced by a \(g\)-inverse \((M^TAN)^{-}\) of some kind, thereby extending the Wedderburn-Guttman theorem. Of course, in this case \(\text{rank}(AN(M^TAN)^{-}M^TA)=g\) may not be equal to \(\text{rank}(M^TAN)=h\). Here, the authors study necessary and sufficient condicitons in order to \(\text{rank}(A-AN(M^TAN)^{-}M^TA)=a-g\). The resultant conditions look similar to those arising in seemingly unrelated contexts, namely Cochran's and related theorems [cf. \textit{P. Šemrl}, Linear Algebra Appl. 237--238, 477--487 (1996; Zbl 0847.62043), Section IV] on distributions of quadratic forms involving a normal random vector.