In this paper, we investigate the use of an exact primal-dual penalty approach within the framework of an interior-point method for nonconvex nonlinear programming. This approach provides regularization and relaxation, which can aid in solving ill-behaved problems and in warmstarting the algorithm. We present details of our implementation within the LOQO algorithm and provide extensive numerical results on the CUTEr test set and on warmstarting in the context of nonlinear, mixed integer nonlinear, and goal programming.
Citation
Working paper, November 2005.