[SCIP] Undeclared Variables Used in the Constraints
Ambros Gleixner
gleixner at zib.de
Fri Sep 18 00:34:21 CEST 2020
Hi Samuel,
This depends on the objective function coefficient of the variable in
relation to the coefficient of the variable in the constraint and the
sides of the constraint.
Say the objective coefficient is nonnegative and the constraint is a <=
constraint. Decreasing the variable always improves the objective or
leaves it unchanged.
Now if the constraint coefficient is positive, then SCIP will fix the
variable at minus infinity, according to the fact that one can always
choose a sufficiently small, finite value to satisfy the constraint.
If the constraint coefficient is negative, then it is best to set the
variable such that the inequality is tight. This means performing a
multi-aggregation and is described in detail in Algorithm 10.4 of Tobias
Achterberg's thesis
http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:0297-zib-11129
To look at the code, see method dualPresolve() in cons_linear.c.
Hope that helps.
Best,
Ambros
Am 14.09.20 um 23:07 schrieb Samuel Jiménez Gil:
> Hi all,
> Hope you are well. I've done quite a few experiments trying to
> understand this question but I don't manage to understand the rationale.
>
> Do you know what SCIP does when you add an arbitrary constraint
> containing a variable which was not declared (free variable), and thus
> no bounds were provided?
>
> Example, (t_0 not declared):
>
> Minimize
> 0
> Subject To
> c1: i_0 + i_1 > 11
> c2: i_0 - 2 i_1 > -1
> c3: t_0 > 40
> Bounds
> 0 <= i_0 <= 1000
> 0 <= i_1 <= 1000
> End
>
> According to what I've seen t_0 could have any value and is ocassionally
> replaced by one of the declared variables.
>
> Many thanks in advance.
>
> Samuel.
> Ph.D Candidate.
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>
--
Ambros Gleixner, Research Group Mathematical Optimization Methods at
Zuse Institute Berlin, http://www.zib.de/gleixner
More information about the Scip
mailing list