Recursive formulas for the generalized \(LM\)-inverse of a matrix (Q868580): Difference between revisions

This paper deals with generalized \(LM\)-inverse of a real matrix. Consider a linear system \(Bx=b\), where \(B\) is an \(m\times n\) matrix and \(b\) is an \(m\)-vector. The generalized \(LM\)-inverse of matrix \(B\), denoted by \(B_{LM}^{+}\), is the matrix such that the solution \(x\), uniquely given by \(x=B_{LM}^{+} b\), minimizes both \[ G=\| L^{1/2}(Bx-b)\| ^2=\| Bx-b\| _L^2 \] and \[ H=\| M^{1/2}x\| ^2=\| x\| _M^2, \] where \(L\) and \(M\) are symmetric positive-definite matrices of size \(m \times m\) and \(n \times n\), respectively. The authors obtain recursive formulas for the generalised \(LM\)-inverse of a columnwise partitioned \(m \times n\) matrix \(B=[A| a]\). They analyze two separate cases: when the additional column \(a\) is a linear combination of the columns of matrix \(A\) and when it is not. They also show that, when \(L=\alpha I_m\) for \(\alpha >0\), their formulas become identical to those for the generalized \(M\)-inverse of a matrix; when \(M=\beta I_n\) for \(\beta>0\), they become those for the generalized \(L\)-inverse of a matrix; and when both \(L=\alpha I_m\) and \(M=\beta I_n\), they reduce to those for the standard Moore-Penrose inverse of a matrix.