A Generalized Proximal Point Algorithm and its Convergence Rate

Etienne Corman(cormanetud.insa-toulouse.fr)
Xiaoming Yuan(xmyuanhkbu.edu.hk)

Abstract: We propose a generalized proximal point algorithm (PPA), in the generic setting of finding a zero point of a maximal monotone operator. In addition to the classical PPA, a number of benchmark operator splitting methods in PDE and optimization literatures such as the Douglas-Rachford splitting method, Peaceman-Rachford splitting method, alternating direction method of multipliers, generalized alternating direction method of multipliers and split inexact Uzawa method can be retrieved by this generalized PPA scheme. We establish the convergence rate of this generalized PPA scheme under different conditions, including estimating the worst-case iteration complexity under mild assumptions and deriving the linear convergence rate under certain stronger conditions. Throughout our discussion, we pay particular attention to the special case where the operator is the sum of two maximal monotone operators, and specify our theoretical results in generic setting to this special case. Our result turns out to be a general and unified study on the convergence rate of a number of existing methods, and subsumes some existing results in the literature.

Keywords: Convex Optimization, Proximal Point Algorithm, Operator Splitting Methods, Convergence Rate

Category 1: Convex and Nonsmooth Optimization (Convex Optimization )

Citation:

Download: [PDF]

Entry Submitted: 07/04/2013
Entry Accepted: 07/04/2013
Entry Last Modified: 07/04/2013

Modify/Update this entry