Error bounds for the perturbation of the Drazin inverse under some geometrical conditions (Q1039682): Difference between revisions

Let \(A\) denote a complex \(n\times n\) matrix. The index \(\text{ind}(A)\) of \(A\) is the minimum integer \(r\geq 0\) such that \(\text{rank\,}A^r= \text{rank\,}A^{r+1}\), and the Drazin inverse of \(A\) is the unique complex \(n\times n\) matrix \(A^D\) satisfying \(A^D AA^D= A^D\), \(AA^D= A^DA\), \(A^{l+1}A^D= A^l\) for all \(l\geq\text{ind}(A)\). Let \(R(A)\) and \(N(A)\) denote the range-space and null-space of \(A\), respectively. For a given complex \(n\times n\) matrix \(A\) and a given positive integers, the authors introduce the following geometrical conditions for a complex \(n\times n\) matrix \(B: R(BA^D)\cap N(A^D)= \{0\}\), \(N(A^DB)\cap R(A^D)= \{0\}\), \(R(B^S)= R(BA^D)\). They then consider the class \(A_S\) of perturbed matrices defined by \(A_S= \{B\mid B\) is a complex \(n\times n\) matrix with \(\text{ind}(B)= s\) satisfying the above conditions). The authors establish several characterizations of \(B\in A_S\) and derive explicit formulas for \(B^D\) and \(BB^D\) and upper bounds for the relative error \(\| B^D- A^D\|/\| A^D\|\) and the error \(\| BB^D- AA^D\|\). A numerical example shows that these upper bounds are an improvement on those given in the literature.