Perturbation of the Moore-Penrose metric generalized inverse in reflexive strictly convex Banach spaces (Q296179)

Let \(X\) be a normed linear space and \(G\) be a subset of \(X\). Let \(\mathrm{dist}(x,G):=\mathrm{inf}_{z \in X} \| x-z \|\). Then the set-valued mapping \(P_G:X \rightarrow G\) defined by \(P_G(x):=\{s \in G: \| x-s \| = \mathrm{dist}(x,G)\}\), \(x \in X\), is called the \textit{set-valued metric projection} of \(X\) onto \(G\). \(G\) is said to be a \textit{Chebyshev set} if \(P_G(x)\) is a singleton for every \(x \in X\). Let \(D(\pi_G):=\{x \in X: P_G(x) \neq \emptyset\}\) and let \(\pi_G:D(\pi_G) \rightarrow G\) denote any single valued mapping such that \(\pi_G(x) \in P_G(x)\) for any \(x \in D(\pi_G)\). When \(G\) is a Chebyshev set, then \(\pi_G\) is called the \textit{metric projector} from \(X\) onto \(G\). Let \(X\) and \(Y\) be Banach spaces and \(T\) be a bounded linear operator from \(X\) into \(Y\). If there exists a bounded homogeneous operator \(S\) (meaning that \(S(\lambda x) = \lambda S(x)\) for each \(\lambda \in \mathbb{R}\) and \(x \in X\)) satisfying the equations \(TST=T, ~ STS=S,~ST=I-\pi_{N(T)}\) and \(TS =\pi_{R(T)}\), then such an \(S\) is unique and is called the \textit{Moore-Penrose metric generalized inverse} of \(T\) and will be denoted by \(T^M\). It is known that, if \(T\) is a bounded linear operator such that the null space \(N(T)\) of \(T\) and the range space \(R(T)\) of \(T\) are Chebyshev subspaces of \(X\) and \(Y\), respectively, then \(T^M\) exists. The main result of the work under review is the following: Let \(X,Y\) be reflexive strictly convex Banach spaces. Let \(T\) be a bounded linear operator from \(X\) into \(Y\) with \(R(T)\) being a closed subspace. Then \(T^M\) exists. Let \(E\) be a bounded linear operator from \(X\) into \(Y\) and let \(W=T+E\) be the perturbation of \(T\) by \(E\). Suppose that there exist numbers \(\lambda_1, \lambda_2 \in (-1,1)\) such that \(\| Ex \| \leq \lambda_1 \| Tx \| + \lambda_2 \| Wx \|\). Denote \(\alpha = \frac{1-\lambda_1}{1+\lambda_2}\) and \(\beta=\frac{1+\lambda_1}{1-\lambda_2}\). Then one has the inequality \(\alpha \| Tx \| \leq \| Wx \| \leq \beta \| Tx \|\) for each \(x \in X\). Further, if \(T^M\) is quasi-additive on \(R(T)\) (meaning that \(T^M(x+z)=T^M(x)+T^M(z)\) whenever \(x \in Y\) and \(z \in R(T)\)), then \(W^M\) exists and one has the inequality \(\frac{1}{\beta \| T \|} \leq \| W^M \| \leq \frac{1}{\alpha} \| T^M \|\).