A new relative error criterion for the proximal point algorithm (Q2489498): Difference between revisions

Let \({\mathcal H}\) be a real Hilbert space and let \(T:{\mathcal H} \rightrightarrows{\mathcal H}\) be a monotonic set-valued mapping. The aim of the authors is to find zeroes of \(T\), i.e., points \(x\in{\mathcal H}\) with \(0\in T(x)\), by means of the proximal point algorithm according to \[ x^{k+1} =(I+c_kT)^{-1}(x^k+e^{k+1}),\;k\in\mathbb{N}. \] Here \(I:{\mathcal H}\to{\mathcal H}\) is the identity, \(\{c_k\}\) is a sequence of (large) positive numbers and \(\{e^k\}\) is a sequence of (small) error vectors. Assuming \(T^{-1} (0)\neq\emptyset\) and a generalized continuity property for \(T^{-1}\) near 0, the authors present conditions for \(\{c_k\}\) and \(\{e^k\}\) such that the sequence \(\{x^k\}\) converges linearly and superlinearly respectively to a solution \(x^0\in T^{-1}(0)\).