GAMS World Forum

Dears

I have a question regarding the use of the Knitro and SBB solvers for solving a non convex mixed integer problem that contains trigonometric terms. Which of these two solvers are more superior for such a problem? Also, what are the difference between these two solvers in terms of dealing with a problem of this complexity?

Regards

Both solvers are "local" solver that get stuck in local optima. Knitro has some multi-start facility (https://www.gams.com/latest/docs/S_KNIT ... #KNITRO_ms) not sure how well this works for Knitro's MINLP. SBB can be run with different NLP solvers. There is the solver MSNLP which does multi-starts that can be used within SBB. MSNLP again calls a subsolver to do the actual work. The default is Conopt. You have to try out what works better for your model.

For non-convex problems where you don't have a good starting point close to the global optimum, I suggest using a global solver. Trig functions are supported by the global solver Lindo(Global) and Octeract.

-Michael