Abstract:
Interior point methods are among the most efficient approach for nonlinear programming. Their implementation in reduced space framework is well suited for large problems with a few degrees of freedom, which often arise from chemical engineering applications where optimal solutions usually depend on a few decision variables. In addition, reduced space methods are attractive to optimization of process systems for which second order derivatives are not available or expensive to calculate. In order to guarantee global convergence of reduced space barrier methods, we propose a feasibility restoration algorithm based on projected gradient methods. This algorithm shares variable decompositions with barrier methods and combines the advantages of trust region and line search approaches. Numerical results of examples from literature and CUTE/COPS test sets demonstrate the performance of the proposed algorithm.