New Features for Mx 1.22

Introduction of code to take advantage of new optimization software from Walter Murray and colleagues at Stanford. This updated version of NPSOL allows for better handling of problematic optimization cases. Specifically, if infeasible points (e.g. due to estimates that give rise to non-positive definite matrices) are chosen, Mx flags them internally rather than using a penalty function. The NPSOL backend may then retreat in its linesearch, and find a feasible point instead. There also appears to be reduced likelihood of IFAIL codes 1 and 6. Conversely, if NPSOL is still getting stuck at infeasible points, it will return IFAIL = -1. Note that this new code has not been adopted in all versions. Several still use the NAG library version of NPSOL (E04UCF) and do not have this feature. Only VAX-VMS and PC versions implement the new NPSOL code.

Mike Neale, Medical College of Virginia