A hybrid interior-point method for nonlinear programming is presented. It enjoys the flexibility of switching between a line search based method which computes steps by factoring the primal-dual equations and an iterative method using a conjugate gradient algorithm and globalized by means of trust regions. Steps computed by a direct factorization are always tried first, but if they are deemed to be ineffective, a trust region iteration that guarantees progress toward stationarity is invoked. To demonstrate its effectiveness, the algorithm is implemented in the KNITRO software package and extensively tested on a selection of problems from the CUTEr test
Citation
report OTC 2003/10, Optimization Technology Center, Northwestern University