On the perturbation of the Moore-Penrose inverse of a matrix

Abstract

The Moore-Penrose inverse of a matrix has been extensively investigated and widely applied in many fields over the past decades. One reason for the interest is that the Moore-Penrose inverse can succinctly express some important geometric constructions in finite-dimensional spaces, such as the orthogonal projection onto a subspace and the linear least squares problem. In this paper, we establish new perturbation bounds for the Moore-Penrose inverse under the Frobenius norm, some of which are sharper than the existing ones.