On the \(\{2\}\)-inverse and some ordering properties of nonnegative definite matrices (Q1920015): Difference between revisions

Let \(\mathbb{R}^{m\times n}\) denote the set of all real \(m\times n\)-matrices. A full rank decomposition of \(G\in\mathbb{R}^{n\times m}\) with rank \(r\) is of form \(G=FH'\), where \(F\in\mathbb{R}^{n\times r}\) and \(H\in\mathbb{R}^{m\times r}\) are both full column rank matrices of rank \(r\) and \(H'\) denotes the transpose of \(H\). For \(A\in\mathbb{R}^{m\times n}\), \(M(A)\) denotes the column space of \(A\), \(M[A]\) denotes the row space of \(A\), and \(ch_1 (A)\) denotes the greatest eigenvalue of \(A\). For nonnegative definite matrices \(A\) and \(B\), write \(A\geq B\) if \(A - B\) is nonnegative definite. For a matrix \(A\), a matrix \(G\) satisfying the condition \(G=GAG\) is called a \(\{2\}\)-inverse of \(A\) and is denoted by \(A^{(2)}\). The authors prove the following main results: Theorem 1. For \(A\in\mathbb{R}^{m\times n}\), if there exist a \(\{2\}\)-inverse \(A^{(2)}\) and full column rank matrices \(C\in\mathbb{R}^{n\times r}\) and \(E\in\mathbb{R}^{m\times r}\) such that \[ M(C)=M(A^{(2)}),\quad M(E)=M\bigl[ (A^{(2)})'\bigr],\tag{1} \] then the \(\{2\}\)-inverse of \(A\) satisfying (1) is unique and may be expressed as \(A^{(2)}=C(E'AC)^{-1} E'\). Theorem 2. Let \(A^{(2)}\) and \(B^{(2)}\) be symmetric \(\{2\}\)-inverses of symmetric \(A\) and \(B\), respectively, and let \(A^{(2)}\geq 0\) be of rank \(r\), \(M(A^{(2)})=M(B^{(2)})\). Then (i) the \(\{2\}\)-inverse of \(B\) satisfying \(M(A^{(2)})=M(B^{(2)})\) is unique and it may be expressed as \(B^{(2)}=C \Delta^{-1}C'\), where \(A^{(2)}=CC'\) is a full rank decomposition of \(A^{(2)}\) and the diagonal elements of \(\Delta=\text{diag} (a_1, \dots, a_r)\) are the nonzero eigenvalues of \(A^{(2)}B\); (ii) \(A^{(2)} \leq B^{(2)}\) if and only if \(ch_1 (A^{(2)} B) \leq 1\).