[SCIP] SCIP GAMS interface

[Stefan Vigerske]

Hi,

like many solvers, SCIP works on a presolved/reformulated version of a 
problem, which is equivalent to the original problem in a mathematical 
sense, but can be slightly different when allowing for numerical 
tolerances. The primal and dual bounds that you see during the SCIP run 
are for this internal problem formulation. The final feasibility check 
of SCIP, however, checks whether the computed solution is also feasible 
in the original problem. In this case, it happened that there is a 
violation that is above the feasibility tolerance.
Now, the GAMS/SCIP interface tries to repair this infeasibility by 
calling out to Ipopt. Since the violation wasn't that small in your 
case, it unfortunately happens that also the objective function value 
moves when moving the solution back into the feasible area.

Btw, try writing the Boundary constraint as quadratic, something like
sqr(Y) + sqr(X) <= sqr(-2.5+Obj)
If you also have Obj.lo = 2.5, then SCIP should recognize this as SOC 
constraint and may perform better.

Stefan

On 04/25/2018 11:44 PM, Akang Wang wrote:
> Hi SCIP developers,
> 
> I'm Akang. I'm calling SCIP 5.0 through GAMS 25.0.2. (Attached is the gams
> file, CPLEX 12.8.0 is the LP solver).
> 
> My question is, SCIP generates the primal and dual bounds
> (Initial_status.PNG). We know that in the minimization problem, the primal
> bound is the upper bound (from a feasible solution). Why does SCIP check
> feasibility here (Initial_status.PNG)?  After time limit is hit, why does
> SCIP further call IPOPT to find a feasible solution (and generate an upper
> bound in this case)?  Then the primal bound is updated by a worse value
> from the IPOPT solution (status_update.PNG, Also, what does the gap mean
> here because it is not (UB-LB)/UB?).
> 
> Could anybody elaborate on this?
> 
> 
> 
> 
> 
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>