[SCIP] Does SCIP use Ipopt as NLP local solver?
Stefan Vigerske
svigerske at gams.com
Sat Apr 10 11:11:25 CEST 2021
Hi,
it depends on your build of SCIP whether you have included any or
several NLP solvers. SCIP mentions early in the log which external codes
it is linked with, so Ipopt would show up there if you have build SCIP
with Ipopt.
From the 3 choices, Ipopt has highest priority, so you won't have to
set nlp/solver.
SCIP solves LP relaxations and runs primal heuristics. There are plenty
of the latter that look for feasible solutions to the MIP relaxation (LP
relaxation + integrality requirements). If such a solution is also
feasible for the nonlinear constraints, then the upper bound is updated.
The better the LP relaxation, the higher the chance that this happens.
Having a NLP solver enables additional primal heuristics that explicitly
take the nonlinear constraints into account, so the strength of the LP
relaxation is less crucial for the upper bound.
Stefan
On 4/10/21 9:02 AM, Angel Manuel Gonzalez Rueda wrote:
> Dear SCIP developers,
>
> I want to solve a Non-Linear Programming Problem with SCIP and I would like
> to know what is the default Non-Linear Local solver that SCIP calls at each
> step in order to obtain upper bounds during the branch and bound procedure.
>
> We have installed the SCIP Optimization Suite version 7.0.2 using the
> source code for LINUX. It seems that by default the Ipopt solver is
> installed (I am not sure about this), so I suppose that this is the one
> that SCIP used by default.
>
> However, I have reviewed the list of all scip parameters, and I have found
> the following parameter:
> # solver to use for solving NLPs; leave empty to select NLPI with highest
> priority
> # [type: string, advanced: FALSE, default: ""]
> nlp/solver = ""
>
> Thus, I do not have clear if it is mandatory to specify in this parameter
> "Ipopt". Furthermore, in case Ipopt is not installed (even any NLPs
> solver), how does SCIP to obtain upper bounds?
>
> Thank you in advance,
> Ángel M. González Rueda
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>
More information about the Scip
mailing list