Using Multi-start with local MINLP solvers in Julia

Specific Domains Optimization (Mathematical)
1

Hi everyone!

When reading the Knitro documentation I saw it could use a multi-start method to solver non-convex MINLP to global optimality even though it is a local solver. Is it possible to implement such a method in Julia with a local MINLP solver like Juniper (with NLP subsolver Ipopt) ? There is the MultistartOptimization.jl library but does it work with MINLP solver or only NLP solvers like NLopt (which is used as example in the documentation) ?

Thanks!

2

There is nothing built in to Juniper or Ipopt for multistart. You’ll need to implement this manually by wrapping the solve in a for loop to try different starting points.