On least squares \(g\)-inverses and minimum norm \(g\)-inverses of a bordered matrix (Q2496634)

If \(L\) is an \(m\times n\) complex matrix, a generalized inverse \(G\) of \(L\) is an \(n\times m\) complex matrix satisfying one or more of the Moore-Penrose conditions (1) \(LGL=L\); (2) \(GLG=G\); (3) \((LG)^H=LG\); (4) \((GL)^H=GL\), where \(H\) denotes the conjugate transpose. A \(G\) satisfying (1) and (3) is denoted by \(L^{(1,3)}\) and is called a least squares \(g\)-inverse of \(L\). A \(G\) satisfying (1) and (4) is denoted by \(L^{(1,4)}\) and is called a minimum norm \(g\)-inverse of \(L\). The unique \(G\) satisfying all four conditions is denoted by \(L^\dag\) and is the pseudoinverse of \(L\). Let \(Ax=b\) be a given linear system. If \(b\in {\mathcal R}(A)\), the system is consistent and \(x=A^{(1,4)}b\) is the minimum norm solution. If \(b\notin{\mathcal R}(A)\), the system is inconsistent and \(x=A^{(1,3)}b\) is a least squares solution. Thus the study of \(L^{(1,3)}\) and \(L^{(1,4)}\) is of theoretical interest and practical relevance. In this paper the authors give the explicit structure of these as well as of \(L^\dag\) for bordered matrices \(L=\left(\begin{smallmatrix} A & B\\ C & 0\end{smallmatrix}\right)\), using the restricted singular value decomposition.