《2016 Approximate Level Method for Nonsmooth Convex Minimization》.pdf

J Optim Theory Appl (2012) 152:334–350 DOI 10.1007/s10957-011-9908-1 Approximate Level Method for Nonsmooth Convex Minimization Peter Richtárik Received: 18 February 2011 / Accepted: 27 July 2011 / Published online: 9 September 2011 © Springer Science+Business Media, LLC 2011 Abstract In this paper, we propose and analyse an approximate variant of the level method of Lemaréchal, Nemirovskii and Nesterov for minimizing nonsmooth convex functions. The main per-iteration work of the level method is spent on (i) minimizing a piecewise-linear model of the objective function and (ii) projecting onto the inter- section of the feasible region and a level set of the model function. We show that, by replacing exact computations in both cases by approximate computations, in rela- tive scale, the theoretical iteration complexity increases only by a small factor which depends on the approximation level and reduces to one in the exact case. Keywords Level method · Approximate projections in relative scale · Nonsmooth convex minimization · Sensitivity analysis · Large-scale optimization 1 Introduction In this paper, we consider the problem of minimizing a Lipschitz continuous and convex objective function on a compact and convex feasible set. We assume that all Communicated by Marcin Studniarski. The research results presented in this paper have been supported by a grant “Action de recherche concertée ARC 04/09-315” from the “Direction de la recherche scientifique—Communauté française de Belgique”. This paper presents research results of the Belgian Program on Interuniversity Poles of Attraction initiated by the Belgian State, Prime Minister’s Office, Science Policy Programming. The scientific responsibility is assumed by the author(s). The author wishes to thank Yurii Nesterov for enlightening discussions and suggestions that greatly helped to improve the paper, and to Robert Chares and two anonymous referees for carefully reading the man