Ipopt/Knitro are local optimization solvers, so for nonconvex problems convergence doesn't guarantee optimality. However, Knitro has a multi-start method where one can start with more random initial solutions with the hope of terminating at a better solution than a single run. I am wondering, is there an equivalent ms_maxsolves option in Ipopt? I failed finding a suitable option in Ipopt in my research.
$\begingroup$
$\endgroup$
1
-
1$\begingroup$ I don't think IPOPT has a multistart option. But if IPOPT is called from Pyomo, you can use Pyomo's multistart capability. groups.google.com/g/pyomo-forum/c/SkKSzPfiM-o . $\endgroup$– Mark L. StoneJun 30 at 1:32
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
As @mark-l-stone says, Ipopt does not have a multi-start option. Just solve your model with different starting points manually.
