Variable metric forward-backward algorithm for minimizing the sum of a differentiable function and a convex function (Q467409)

The authors consider the problem of finding a minimizer of a coercive function \(G\), \(G:\mathbb R^N \longmapsto [\infty, \infty]\). It is assumed that the function \(G\) can be split as \(G = F + R\), where \(F\) is a differentiable function and \(R\) is a proper lower semicontinuous and convex function. For solving the problem, an inexact version of the variable metric forward-backward algorithm with a special chosen metric is proposed. The convergence of the algorithm is established under the assumption that the Kurdyka-Lojasiewcz inequality holds for the function \(G\), see [\textit{S. Łojasiewicz}, in: Equ. Derivees partielles, Paris 1962, Colloques internat. Centre nat. Rech. sci. 117, 87--89 (1963; Zbl 0234.57007)]. Some numerical experiments, especially experiments concerning image recovery problems are presented in the concluding part of the paper.