A new generalized APPA for maximal monotone operators (Q2469710)

This paper deals with the following problem: given the point-to-set mapping \(\hat T:{\mathbb{R}}^n\to 2^{{\mathbb{R}}^n}\), \(\hat T=T+N_{\Omega},\) where \(T\) is a maximal monotone operator on \({\mathbb{R}}^n\) and \(N_{\Omega}\) is the normal cone operator with respect to a closed and convex subset \(\Omega\), find a point \(x\in \Omega\) such that \(0\in \hat{T}(x)\). In order to solve the problem, some algorithms have been proposed. One powerful approach is the proximal point algorithm (PPA) that, starting from an approximation \(x^k\) to a root of \(\hat{T}\), generates the next iterate \(x^{k+1}\) by solving the proximal subproblem: \[ 0\in \beta_k \hat{T}(x^{k+1})+x^{k+1}-x^k, \] where \(\{\beta_k\}\subset [\beta,\infty)\) is a non-decreasing sequence of scalars, and \(\beta>0\).\ Since the solution of this subproblem is computationally difficult, new generalized algorithms, inspired by the more implementable approximate proximal point algorithm (APPA) in [\textit{R.\,T.\thinspace Rockafellar}, SIAM J.~Control Optimization 14, 877--898 (1976; Zbl 0358.90053)], were introduced; in particular, some more recent developments replace the linear term \(x^{k+1}-x^k\) with more general terms, e.g., the Bregman functions. In this paper, the author presents a new generalized algorithm, replacing Bregman functions, whose defining conditions are difficult to check, with other general functions \(f\) defined via conditions that are stronger but easier to verify in practice. The author presents the steps of this new generalized APPA (\(f\)-APPA), and, in Section 2, makes some convergence analysis. In particular, he proves that the sequence generated by the algorithm \(f\)-APPA with optimal step sizes converges to a solution globally under rather relaxed restrictions on the error sequence.