[SCIP] Transformed Problem + Infeasible solution at Constraint Handler

[Matheus Ota]

Hi Gregor,

My original constraint is flagged with true for the initial, separate and
check parameters.
Indeed, my constraint handler has a higher priority (greater than zero,
since I want to separate fractional solutions) than the knapsack constraint
handler. So what you are saying is that I'm calling my callback routine
before adding the knapsack constraints (even though SCIP printed the
knapsack constraint for the transformed model)? Then I guess this would
only happen only once? Because after SCIP adds the knapsack constraints,
all the solutions that enters my callback will respect the knapsack
constraints (**).

Can I change the separation priority of the knapsack constraint in the
transformed model? This way I guarantee that the solution that enters my
callback already satisfies (**).

Thanks,
Matheus

Em seg., 9 de dez. de 2019 às 13:22, Gregor Hendel <hendel at zib.de> escreveu:

> Hi Matheus,
>
> your reasoning about negation is correct, so I would expect that the LP
> solution satisfies ** at CONSENFOLP. A special case occurs if the knapsack
> constraint is flagged by (not initial/not separate/not check), in which
> case the constraint is not immediately part of the root LP and might not be
> separated at violation.
>
> Note that the knapsack constraint handler has a very low enforcement
> priority, -600000. Probably, the enforcement priority of your constraint
> handler is higher, and the knapsack constraint handler would add its
> constraint to the LP in the same enforcement round, but later?
>
> Kind regards,
> Gregor
>
> Am 06.12.19 um 01:01 schrieb Matheus Ota:
>
> Hi all,
>
> First I would like to say that I'm very grateful for all the help I
> received from this mailing list! I have a question regarding a MIP model
> that I'm solving. I have some equalities of the following type
>
>> x_i,0 + x_i,1 == 1, for i = 0, ..., 8 (*)
>>
>
> Thus, SCIP presolve routines replace some of my variables (x_i,0) by the
> negation of other variables (1 - x_i,1). In the transformed model I have
> only this constraint
>
> [knapsack] <R>: +1<~t_x_8,1>[B] +1<~t_x_7,1>[B] +1<~t_x_6,1>[B]
>> +1<~t_x_5,1>[B] +1<~t_x_4,1>[B] +1<~t_x_3,1>[B] +1<~t_x_2,1>[B]
>> +1<~t_x_1,1>[B] +1<~t_x_0,1>[B] <= 4;
>>
>
>  And each variable x is binary. The tilde (~) means negation here? If so,
> replacing <~t_x_i,1> by <t_x_i,0> this would give
>
>> <t_x_8,0> + ... + <t_x_0,0><= 4 (**)
>>
>
> But at the CONSENFOLP method, SCIP is entering with a "solution" S with
> x_i,0 = 1 and x_i,1 = 0, for i = 0, ..., 8. Which is infeasible because
> violates (**). But the SCIP documentation says that CONSENFOLP is called
> when an LP solution is available. But this solution S do not respect the
> constraints in my LP. I'm a little confused with this behavior and would be
> glad if someone could give me some light here.
>
> Thanks!
> Matheus
>
> _______________________________________________
> Scip mailing listScip at zib.dehttps://listserv.zib.de/mailman/listinfo/scip
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20191209/3c54c5f1/attachment.html>