KNITRO, the solver for nonlinear optimization that can be used with or without Matlab (but is used mostly with Matlab), has been updated to 9.1.0. Enjoy.
From the Ziena website:
What’s new in release 9.1?
- KNITRO 9.1 introduces a new Sequential Quadratic Programming (SQP) algorithm for continuous problems. This new SQP algorithm is primarily designed for small problems, where the computational cost is dominated by function/derivative evaluations. It will often be the most efficient algorithm for such problems. See section Algorithms for more details.
- KNITRO 9.1 offers many enhancements to the MATLAB interface including: 1) the ability to perform parallel finite-differences; 2) the ability to perform a derivative check; 3) the ability to provide a newpoint/OutputFcn callback function; 4) a specialized interface for least-squares problems; and 5) the ability to specify initial Lagrange multipliers. See section KNITRO / MATLAB reference for more details.
- KNITRO 9.1 offers an efficient procedure for trying to refine the barrier solution to obtain a more precise barrier solution via the new bar_refinement user option.
- KNITRO 9.1 offers simplified user options for performing a derivative check (rather than performing the derivative check through API function calls). See the new user options derivcheck, derivcheck_toland derivcheck_type.
- KNITRO 9.1 offers improved handling and performance on models with range constraints.
- KNITRO 9.1 offers a new option ms_deterministic that allows for deterministic parallel, multi-start performance when ms_terminate =0.
- KNITRO 9.1 allows one to print and retrieve final solution information and constraint values for mixed-integer problems (MIP).
- KNITRO 9.1 offers new API function calls for obtaining information about user options.
- KNITRO 9.1 splits the termination codes for limits (e.g. iteration and time limits) into feasible and infeasible cases. See section Return codes for more details.
- KNITRO 9.1 offers overall speed and robustness improvements on general NLP models and mixed integer NLP models.