[Scip] Scip writes Infeasible solution for some positive gap

[Stefan Vigerske]

Hi,

how large are the constraint violations? Are these within the 
feasibility tolerance of SCIP (1e-6 by default), or much larger.
Without anything to reproduce the issue, not much help will be possible.

Stefan


On 06/26/2014 08:55 PM, Vladimir VV wrote:
> Dear SCIP team, Victor,
> experimenting on some Task-Worker scheduling model (written as
> mixed-boolean LP) I found that for some (!?) positive value of
> limits/gap SCIP writes infeasible solution to file.
>
> E.g. for gap=0.2 SCIP returns feasible approximate solution, but for
> gap=0.1 - infeasible one.
> In the last case I see that there are a a lot of "primalbound" feasible vectors:
>> SCIP Status        : solving was interrupted [gap limit reached]
>> Solving Time (sec) : 123.96
>> Solving Nodes      : 221991
>> Primal Bound       : +8.46095278447460e+00 (40 solutions)
>> Dual Bound         : +7.69597947107262e+00
>> Gap                : 9.94 %
> But check returned *.sol file in AMPL-script shows that some
> constraints are violated (values of boolean variables are "right",
> 0|1)
>
> I use SCIPAMPL and I get this wrong solution both in "batch" and
> "interactive" (with -i option) modes of running SCIPAMPL.
> Moreover, I have this problem for SCIP 3.0.2 and 3.1.0.
>
> Can you help me ?
>
> Sincerely yours,
>