Cutting plane algorithms for maximum problems

the long version

This paper unifies the development of the cutting plane algorithm for mathematical programs and variational inequalities by providing one common framework for establishing convergence. strategies for generating cuts are provided for cases in which the algorithm yields easy and difficult subproblems. When the subproblem is easy to solve, a line search is added and a deep cut is selected to accelerate the algorithm. On the other hand, when the subproblem is difficult to solve, the problem is only solved approximately during the early iterations. This corresponds to generating cuts which are nontangential to the underlying objective function. Moreover, in the case of variational inequalities, it is shown further that the subproblem can be eliminated entirely from the algorithmic steps, thereby making the resulting algorithm especially advantageous.