A superlinearly convergent norm-relaxed SQP method of strongly sub-feasible directions for constrained optimization without strict complementarity (Q833162)

This paper presents a variant of a line-search successive quadratic programming (SQP) method. To avoid the Maratos effect, the search direction is modified with a high-order update. By an additional modification of the search direction it is ensured that the iterates always become feasible. Global convergence is shown for a finite number of iterates using the Mangasarian-Fromovitz constraint qualification. If the algorithm generates an infinite number of iterates, the global convergence proof is based on the LICO condition. In both cases, usual assumptions for the considered functions are made. The superlinear behavior is shown using the strong second-order condition an the solution point and other standard requirements. Numerical results are given for 6 test problems out of the Hock/Schittkowski collection with up to 10 variables and 8 constraints.