Reduced-direction methods with feasible points in nonlinear programming

Abstract

An approach to the construction of feasible direction type methods of nonlinear programming is proposed. The approach relies on linearization of the active constraints, which reduces the problem of choosing a direction of descent of the objective function inside the feasible region to an unconstrained direction-choosing problem for an auxiliary function in a lower-dimensional space. The approach is developed for problems with inequality constraints and extended to problems with both inequality and equality constraints.