[Scip] Overly motivated presolver?

[Tobias Achterberg]

Julian Lamberty wrote:
> Hi Tobias!
> 
> I'm the student working with Moritz on the reported column-generation 
> problem.
> 
> Now I found out, that our variables "disappear" from the constraints 
> when activating either
> 
> presolving/trivial
> 
> or
> 
> presolving/probing
> 
> The other presolvers do obviously not affect the constraints.
> I think one of these presolvers tries to use the "empty" constraint to 
> set bounds on the variables u and then deletes u from the constraint.
> 
> Can we somehow protect these constraints?

Yes, you can, namely by setting the "modifiable" flag of the constraint to TRUE. But since 
you are doing this, the issue must be somewhere else, probably a bug in SCIP.


Do I understand it correctly, that if you disable all the presolvers I mentioned:

 >> constraints/linear/maxprerounds = 0
 >> presolving/dualfix/maxrounds = 0
 >> presolving/probing/maxrounds = 0
 >> presolving/implics/maxrounds = 0
 >> presolving/inttobinary/maxrounds = 0
 >> presolving/trivial/maxrounds = 0

then it works fine, but if you reenable only the "trivial" presolver, then the issue appears?


Timo just told me that your question raised some bell in his mind. So, it may very well be 
that we have already fixed your bug in our current development version. SCIP 1.1.0 will be 
released very soon, but if you like, Timo can send you a prerelease immediately such that 
you can check whether this solves the issue.

Note, however, that SCIP 1.1.0 will have some API changes, so you will also have to update 
your code (usually, it means that some additional parameters are needed for some methods; 
just read the CHANGELOG).


Tobias