A sequential quadratic programming algorithm without a penalty function, a filter or a constraint qualification for inequality constrained optimization

Abstract

This paper discusses a kind of nonlinear inequality constrained optimization problems without any constraint qualification. A new sequential quadratic programming algorithm for such problems is proposed, whose important features are as follows: (i) a new relaxation technique for the linearized constraints of the quadratic programming subproblem is introduced, which guarantees that the subproblem is always consistent and generates a favourable search direction; (ii) a weaker positive-definiteness assumption on the quadratic coefficient matrices is presented; (iii) a slightly new line search is adopted, where neither a penalty function nor a filter is used; (iv) an associated acceptable termination rule is introduced; (v) the finite convergence of the algorithm is proved. Furthermore, the numerical results on a collection of CUTE test problems show that the proposed algorithm is promising.