A geometry for the set of split operators (Q2442622)

Let \(\mathcal H\) be a Hilbert space and \(\mathcal B(\mathcal H)\) denote the space of bounded linear operators on \(\mathcal H\). \(A \in \mathcal B(\mathcal H)\) is called a split operator if \(N(A) \cap R(A) = \{0\}\) and \(N(A) + R(A)= \mathcal H\). Let \(\mathcal X\) denote the set of all split operators on \(\mathcal H\) and \(\mathcal Y\) denote the subset of all \(A \in \mathcal X\) satisfying, in addition, the condition that \(R(A)\) is orthogonal to \(N(A)\). Recall that \(B \in \mathcal B(\mathcal H)\) is called a reflexive generalized inverse of \(A \in \mathcal B(\mathcal H)\) if \(ABA=A\) and \(BAB=B\). Such an operator is called a commuting reflexive generalized inverse of \(A\) if, in addition, \(AB=BA\). For matrices, a commuting reflexive generalized inverse is widely known as the group inverse of \(A\). A~reflexive generalized inverse \(B\) for which we also have \(AB=(AB)^*\) and \(BA=(BA)^*\) is the more popularly known Moore-Penrose inverse of \(A\). The first main result presents two characterizations for \(A\) to belong to \(\mathcal X\): \(A \in \mathcal X\) is equivalent to the statement that there is a commutative reflexive generalized inverse of \(A\), which, in turn, is equivalent to the condition that there exists an idempotent operator \(P \in \mathcal B(\mathcal H)\) such that \(AP=0=PA\) and \(A+P\) is invertible. For \(\mathcal Y\), it is shown that \(A \in \mathcal Y\) if and only the Moore-Penrose inverse of \(A\) commutes with \(A\). (Such operators are referred to as \(EP\) operators in the literature.) For \(A \in \mathcal X\), define \(q(A)\) to be the oblique projection of \(\mathcal H\) onto \(R(A)\) along \(N(A)\) and for \(A \in \mathcal Y\), define \(p(A)\) to the orthogonal projection of \(\mathcal H\) onto \(R(A)\). These functions are not continuous; subsets of continuity are considered. It is shown that the topological interior of \(\mathcal X\) is precisely the set of Fredholm operators of codimension \(0\) or \(1\). In this interesting paper, the authors conclude with some differential geometric results for certain subsets of \(\mathcal X\).