A general descent framework for the monotone variational inequality problem (Q1315420): Difference between revisions

The authors develop a descent method to solve a monotone variational inequality of the form: find \(v^*\in K\) such that \(s(v^*)^ T(v- v^*)\geq 0\) for all \(v\in K\), where \(K\) is a closed convex subset in \(\mathbb{R}^ n\) and \(s\) is a strongly monotone, continuously differentiable mapping from \(\mathbb{R}^ n\) into \(\mathbb{R}^ n\). They introduce a function \(\phi(u,v)= {1\over 2}\| u- v\|^ 2\) (or another similar function) and the algorithm is based on the remark that \(v^*\) is solution of the variational inequality if and only if \(v^*\) is solution of the optimization problem \(\min\{f(v)\mid v\in K\}\) where \(f(v)= -s(v)^ T(H(v)- v)- {1\over \alpha}\phi(H(v),v)\), \(H(v)\in \Gamma(v)\), \(\alpha> 0\), and \[ \Gamma(v)= \text{argmin}_{u\in K} s(v)^ T(u- v)+ {1\over \alpha}\phi(u,v). \] The direction descent for the optimization problem is achieved by solving \[ \min_{u\in K} s(v)^ T(u- v)+ {1\over \alpha}\phi(u, v), \] this is done by a sequence of linear programs. Under some conditions on \(s\) a convergence theorem is proved. Connections with previous results are also given [see e.g. the third author and \textit{J.-P. Dussault}, SIAM J. Control Optimization 27, No. 6, 1260-1278 (1989; Zbl 0689.90069), and \textit{D. P. Bertsekas} and \textit{E. M. Gafni}, Math. Program. Study 17, 139-159 (1982; Zbl 0478.90071)].