On the perturbation bounds of g-inverses and oblique projections (Q603107): Difference between revisions

Given a matrix \(A\), a g-inverse (sometimes also called \((1)\)-inverse) of \(A\) is any matrix \(X\) satisfying the relation \(AXA = A\). Given a g-inverse \(A^-\) of A and a perturbed matrix \(\hat A\), the authors investigate the problem of finding a g-inverse of \(\hat A\) that is closest to \(A^-\) in a certain norm. In the matrix \(2\)-norm, it turns out that this problem has a unique solution, which can be expressed in terms of the Moore-Penrose pseudoinverse of \(\hat A\). In the Frobenius norm, a set of parametrized solutions is given. A variation of this problem, where the aim is to minimize the difference between the corresponding oblique projections onto the range of \(A\) and \(\hat A\), is also addressed. Extending these ideas, the authors provide perturbation results for solutions of linear least-squares problems with multiple right-hand sides.