Weak monotonicity of \(A\) versus \(\{1\}\)-inverses nonnegative on the range space of \(A\) (Q2459959): Difference between revisions

A matrix \(X\) is called a \(\{1\}\)-inverse of \(A\in\mathbb{R}^{m\times n}\) if \(AXA=A\). If \(x \in \mathbb{R}^n\) belongs to the nonnegative orthant \(\mathbb{R}^n_+\) of \(\mathbb{R}^n\), then we say \(x \geq 0\). A matrix \(A\in\mathbb{R}^{m\times n}\) is called weak monotone if \(Ax \geq 0\) implies that \(x\in \mathbb{R}_{+}^{n}+N(A)\), where \(N(A)\) denotes the kernel of \(A\). The authors affirmatively solve the problem raised by \textit{J. E. Paris} and \textit{B. Subiza} in [ibid. 166, 167--184 (1992; Zbl 0747.15007)] by showing that if \(A\) has a \(\{1\}\)-inverse \(X\) which is nonnegative on \(R(A)\), then \(A\) is weak monotone; and conversely, if \(A\) is weak monotone and possesses a nonnegative full rank factorization, then there is a \(\{1\}\)-inverse of \(A\) being nonnegative on \(R(A)\). By a full rank decomposition of \(A\) we mean a decomposition of the form \(A=BC\), where \(B\in \mathbb{R}^{m\times r}, C\in \mathbb{R}^{r\times n}\) with \(r= \text{rank}(A)= \text{rank}(B)= \text{rank}(C)\).