[Scip] Non-binding inequalities

[Stefan Lörwald]

Hi Gerald,

thanks for your reply. I think you already answered the main part of
my question, which is - given a redundancy free system of constraints
- those parallel to the objective function are ignored. This is a
thing in the presolving phase which is on by default, right?

Basically I'm curious if, as a general rule of thumb, I can use the
following for creating a B&C solver with SCIP:

1. SCIPincludeConshdlrLinear(scip)
2. Add all model constraints (without redundancies) with
SCIPcreateConsLinear(...) (I only have linear constraints)
3. Add constrainthandlers for facets (I used SCIPincludeConshdlrBasic
and implemented a routine for SCIPsetConshdlrSepa)

This is what I extracted from the TSP / LOP example code and
experimenting a bit to get the best performance I can get.
In my mind this looks like the straightforward approach, but may be a
misuse of what SCIP actually was designed for.

Yours,
Stefan


2014-02-05 Gerald Gamrath <gamrath at zib.de>:
> Hi Stefan,
>
> what do you mean with "non-binding" and what do you mean with "allow to
> ignore"?
>
> SCIP does a presolving phase before the actual solving process and will
> remove some redundant constraints there. However, already doing a
> pairwise comparison of constraints is quite expensive, so SCIP will not
> detect if a constraint is implicitly fullfilled due to two or more other
> constraints. However, all constraints lead to tightening of variable
> bounds and those are of course also taken into account when computing
> activity bounds while looking for redundant constraints. Moreover,
> constraints which are parallel to the objective function are removed
> from the problem and added as a cutoff or lower bound.
>
> There is a lot more, if you tell us what exactly you mean are looking
> for, we can tell you whether this is automatically done or if you can
> turn it on with a parameter.
>
> Best,
> Gerald
>
> On 05.02.2014 13:43, Stefan Lörwald wrote:
>> Hi,
>>
>> I'd like to know if there is an option to ignore non-binding
>> inequalities provided with "SCIPcreateConsLinear" /
>> "SCIPincludeConshdlrLinear(scip)". That is, does SCIP allow to ignore
>> inequalities that are provably non-binding because of the objective
>> function? Is it perhaps enabled by default?
>>
>> yours,
>> Stefan
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/mailman/listinfo/scip
>