Perturbations of \(p\)-adic linear operations (Q1768068): Difference between revisions

Let \(E,F\) be non-archimedean Banach spaces and let \(L(E,F)\) be the non-archimedean Banach space of all continuous linear operators \(E \to F\). The author introduces the gap \(d(V,W)\) between two closed convex sets \(V,W\): \[ d(V,W)= \max (\eta(V,W),\eta(W,V)), \] with \( \eta(V,W):= \sup \{ \frac{\text{dist} \, (x,W)}{\| x \|} : x \in V, \, x \neq 0 \}\), \( \eta( \{ 0 \},W) := 0.\) He pays special attention to the cases in which either \(V = \text{Ker}(T)\) and \(W=\text{Ker}(\overline{T})\) or \(V = R(T)\) and \(W=\overline{R(\overline{T})}\), where \(T \in L(E,F)\) has closed range and \(\overline{T} = T + A\) is the perturbed of \(T\) by an \(A \in L(E,F)\). The notion of {inverse generalized} is also treated in the paper (\(S \in L(F,E)\) is said to be a generalized inverse of \(T\) if \(TST = T\) and \(STS = S\)). It is shown that: {If \(T\) has a generalized inverse \(S\) such that either} \(\dim \text{Ker}(T) = \dim \text{Ker}(\overline{T}) < \infty\) {or} \(R(\overline{T}) \cap \text{Ker}(S) = \{ 0 \}\), {one verifies the following:} (1) If \(\| S A \| <1\), then \(d(\text{Ker}(T),\text{Ker}(\overline{T})) \leq \| S A\|\). (2) If either \(\| S A \| <1\) or \(\| A S \| < 1\), then (2.1) \(d(R(T),\overline{R(\overline{T})}) \leq \| S \| \, \| A \|\), (2.2) \(\overline{S}:= (I_E+ S A)^{-1} S\) is a generalized inverse of \(\overline{T}\) with \(\| \overline{S} \| = \| S \|\), \(\| \overline{S} A \| = \| S A \|\) and \(\| A \overline{S} \| = \| A S \|\). These results are used in the last section to study the gap between the sets of solutions \(X(T,b):= \{ x \in E: T(x) = b \}\) and \(X(\overline{T},\overline{b}):= \{ x \in E: \overline{T}(x) = \overline{b} \}\) of the operator equations \(T(x) = b\) and \(\overline{T}(x) = \overline{b}\) (\(b \in R(T) \setminus \{ 0 \}\), \(\overline{b} \in R(\overline{T}) \setminus \{ 0 \}\)). It is proved: If \(T\) has a generalized inverse \(S\) such that either \(\dim \text{Ker}(T) = \dim \text{Ker}(\overline{T}) < \infty\) or \(R(\overline{T}) \cap \text{Ker}(S) = \{ 0 \}\), then \(d((X(T,b),X(\overline{T},\overline{b})) \leq \| S \| \, \| A \|\) (when additionally \(\| S \| \, \| A \| <1\) and \(\| \overline{b} - b \| / \| b \| \leq \| A \| / \| T \|\)), and \(d((X(T,b),X(\overline{T},\overline{b})) \leq \| S \, A \|\) (when additionally \(\| S \, A \| < 1\) and \(\overline{b} - b \in \text{Ker}(S)\)). It would be interesting, for a future work, to investigate the possible applications of the theory developed in the paper to some types of operator equations appearing in practical problems, for instance, to \(p\)-adic differential equations.