A New Sequential Quadratic Programming Method Without a Penalty Function or a Filter

Abstract:

A sequential quadratic programming method without using a penalty function or a filter was proposed. The algorithm computes the overall step in two phases. The first phase is to compute a feasibility step. The feasibility phase aims at reducing the infeasibility measure. The second phase, an optimality phase computes a trial point reducing a quadratic model of the objective function. The feasibility and optimality phases are independent in this algorithm; therefore, any method for reducing constraint violation can be used in the feasibility phase. Under mild conditions, the method can be proved to be globally convergent. Numerical results demonstrate the efficiency of this algorithm.