On the finite convergence of a projected cutter method (Q2349845): Difference between revisions

A mapping \(T\) from a real Hilbert space into itself is said to be a cutter if the set \(\operatorname{Fix}T\) of its fixed points is nonempty and for every \(x\in X\) and \(y\in \operatorname{Fix}T\) one has \(\| Tx-y\| ^{2}+\| x-Tx\| ^{2}\leq \| x-y\| ^{2}\). Assuming that the set \(C\subseteq X\) intersects the interior of \(\operatorname{Fix}T\), a finitely convergent iterative algorithm for finding an element in \(C\cap \operatorname{Fix}T\) is proposed. This algorithm involves a quasi projector of \(C,\) that is, a mapping \(Q:X\to X\) such that \(\operatorname{ran}Q=\operatorname{Fix}Q=C\) and \(\| Qx-c\|\leq \| x-c\|\) for every \(x\in X\) and \(c\in C\). When \(T\) is the resolvent of a maximally monotone operator \(A\), the algorithm finitely converges to an element of \(C\cap A^{-1}0\). Several examples are provided to show that the convergence theorems do not hold true without the assumptions their statements contain on the parameters defining the algorithm.