[Ipopt] tuning strategy for IPOPT parameters

[Herzog Raoul]

Hello,

I have a nonconvex optimization problem with ~300 unknowns, where the feasible set is "small" and the objective function relatively flat.
I enter with an initial guess which is the best I can get reasonably, but which does not satisfy the constraints.

IPOPT struggles, some cases are working fine, but with a high number of iterations and passing through several restauration phases; others do not work.
Using callbacks I observed that the constraints are not satisfied during a large number of iterations; they are satisfied only at the very end.

I need to tune the IPOPT parameters, because I have a large number of "similar problems" to solve.

Any hints for a tuning strategy ?

I thought starting with the parameters relative to the "Initialization" phase, but I have no experience how to tune them.

Many thanks in advance,

--raoul


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20171130/0d426dee/attachment.html>