On interior point multidimensional filter line search methods for constrained optimization

Abstract

This paper aims to compare primal-dual interior point multidimensional filter line search methods for nonlinear programming. The multidimensional filter is based on the constraint violations and aims to enforce their convergence to zero. To prevent convergence to feasible nonoptimal points a standard monotone sufficient reduction condition is also imposed on the barrier function for a trial iterate to be acceptable. Nonmonotone reduction conditions that allow an increase in the filter entries and in the barrier function at each iteration are also implemented. Numerical experiments with both variants as well as a comparison with a merit function approach are reported.