[Scip] trying to read debug output of infeasibility

[Stefan Vigerske]

Hi,

On 09/23/2013 07:26 AM, George Nychis wrote:
> Hi all,
>
> I am trying to debug why I am getting an infeasibility error with
> #define SCIP_DEBUG.
>
> If I grep for "to be infeasible" I get:
>
> [src/scip/cons_quadratic.c:8230] debug: found constraint <expComp_eq_38> to
> be infeasible; sides: [1, 1], activity: [1, 1], infeas: 1.0784e-09
> [src/scip/cons_quadratic.c:8230] debug: found constraint <expComp_eq_38> to
> be infeasible; sides: [1, 1], activity: [1, 1], infeas: 1.0784e-09
> [src/scip/cons_quadratic.c:8230] debug: found constraint <expComp_eq_38> to
> be infeasible; sides: [1, 1], activity: [1, 1], infeas: 1.0784e-09
> [src/scip/cons_quadratic.c:8230] debug: found constraint <expComp_eq_38> to
> be infeasible; sides: [1, 1], activity: [1, 1], infeas: 1.0784e-09
> [src/scip/cons_quadratic.c:8230] debug: found constraint <expComp_eq_38> to
> be infeasible; sides: [1, 1], activity: [1, 1], infeas: 1.0784e-09
> [src/scip/cons_quadratic.c:8230] debug: found constraint <expComp_eq_62> to
> be infeasible; sides: [1, 1], activity: [0.841751, 1], infeas: 1.08939e-09
>
> Therefore, I am guessing it is this constraint:
>
>    subto expComp_eq:     # The component in the exponential function is the
> packets/second * vulnerability win
>        forall <l> in L : forall <j> in U[l] : expComp[l,j] * LinkPPS[j] == 1
> - LinkAirtime[j];
>
> How do I read the debug output in relation to that constraint to understand
> why exactly it is infeasible?

It says that for l = 38, all variables (expComp[l,j], LiunkPPS[j], and 
LinkAirTime[j]) seem to be fixed; I guess there is only one j in U[38]. 
If you calculate expComp[l,j] * LinkPPS[j] + LinkAirtime[j] for the 
fixed values, then you are by more than 1e-9 away from the right hand 
side 1.0.

It may actually use a too tight tolerance on this place. If you replace 
the SCIPepsilon(scip) by SCIPfeastol(scip) in 
cons_quadratic.c:8228/8229, you might get around this.

It would be nice if you could send me this instance, so I can reproduce 
this issue.

Stefan

>
> Thanks for any help, and sorry for the barrage of e-mails for help!
>
> - George
>
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
>