ROBUST RECURSIVE QUADRATIC-PROGRAMMING ALGORITHM MODEL WITH GLOBAL AND SUPERLINEAR CONVERGENCE PROPERTIES

A new, robust recursive quadratic programming algorithm model based on a continuously differentiable merit function is introduced. The algor ithm is globally and superlinearly convergent, uses automatic rules fo r choosing the penalty parameter, and can efficiently cope with the po ssible inconsistency of the quadratic search subproblem. The propertie s of the algorithm are studied tinder weak a priori assumptions; in pa rticular, the superlinear convergence rate is established without requ iring strict complementarity. The behavior of the algorithm is also in vestigated in the case where not all of the assumptions are met. The f ocus of the paper is on theoretical issues; nevertheless, the analysis carried out and the solutions proposed pave the way to new and more r obust RQP codes than those presently available.