A New Hybrid Inexact Logarithmic-Quadratic Proximal Method for Nonlinear Complementarity Problems

Ying Zhou (Iowa State University, USA) and Lizhi Wang (Iowa State University, USA)

Source Title: International Journal of Operations Research and Information Systems (IJORIS) 1(3)

DOI: 10.4018/joris.2010070101

Abstract

In this paper, the authors present and analyze a new hybrid inexact Logarithmic-Quadratic Proximal method for solving nonlinear complementarity problems. Each iteration of the new method consists of a prediction and a correction step. The predictor is produced using an inexact Logarithmic-Quadratic Proximal method, which is then corrected by the Proximal Point Algorithm. The new iterate is obtained by combining predictor and correction point at each iteration. In this paper, the authors prove the convergence of the new method under the mild assumptions that the function involved is continuous and monotone. Comparison to another existing method with numerical experiments on classical NCP instances demonstrates its superiority.

Article Preview

Top

Introduction

The nonlinear complementarity problem (NCP) is to determine a vector such that

(1) where

is a nonlinear mapping. Throughout this paper we assume that

is continuous and monotone with respect to

and the solution set of (1) is nonempty. NCP has many important applications in engineering, economics, military operations planning, finance, medical treatment, supply chain management etc, (Auslender & Haddou, 1995; Das, 2009; Castagnoli & Favero, 2008; Ferris & Pang, 1997; Harker & Pang, 1990; Yan & Wang, 1997). Many numerical methods for solving NCP have been developed (Auslender & Haddou, 1995; Auslender et al., 1999; Bnouhachem & Noor, 2001; Burachik & Svaiter, 2001; Censor et al., 1994; Eckstein, 1998; Fischer, 1997; Guler, 1991; Iusem, 1998; Pang, 1995; Qi & Yang, 2002; Sun & Qi, 1999; Zhou, 2009).

NCP can be alternatively formulated as finding the zero point of an appropriate maximal monotone operator

(2) i.e., finding

such that

, where

is the normal cone operator to

defined by

(3)

A well known method to find the zero point of a maximal monotone operator is the proximal point algorithm (PPA), which starts with any vector and and iteratively updates conforming the following problem: