[SCIP] Constraint Handler is not adding the cuts as expected

[Ambros Gleixner]

Hi Naga,

Short answer: You should get a correct implementation if you call your 
code from scip_sepalp in scip_enfolp() and scip_enfops(); if your 
separation procedure does not automatically yield a feasibility check, 
then call the code from scip_check() first.  If your separation 
procedure cannot handle fractional solutions, then simply leave 
scip_sepalp() empty.

Longer answer:  The sepalp callback is called during the cut loop of the 
LP, so it always receives a fractional solution.

With negative enfopriority, SCIP branches as long as the LP solution is 
fractional; your enforcement callbacks are only called at nodes where 
the LP relaxation is integer feasible to check and add subtour 
elimination constraints.

This is correct, but may result in more branching than necessary.  So in 
a second step, it would be better if your separation algorithm could 
handle fractional solutions.

Best,
Ambros


Am 05.08.20 um 16:43 schrieb Naga Venkata Chaitanya Gudapati - 
nagavenkata.gudapati at studio.unibo.it:
> Hello,
> 
> I have a very simple min-max routing problem and I seem to have hit a small snag. I think I am missing something very simple in my understanding of the constraint handler inner mechanisms.
> 
> The constructor of my constraint handler is as follows:
> 
> ObjConshdlr(scip, "sub-tour", "sub-tour elimination constraints",
>                       1000000, -20000, -25000, 1, -1, 1, 0,
>                       FALSE, FALSE, TRUE, SCIP_PROPTIMING_BEFORELP, SCIP_PRESOLTIMING_FAST)
> 
> 1) the checkpriority flag is -25000 which I assume is just gonna only run the scip_check function  when the feasible solutions are integeral.
> 
> 2) in my scip_check I have a function which tells us whether there is a subtour or not and if there is a subtour, it sets the *result to SCIP_INFEASIBLE;
> 
> 3) Then in the scip_sepalp function , I call another function sepaSubtour which will add the cut.
> 
> But the issue is that the solution values in scip_check and sepaSubtour seem to be different. I find a sub-tour using the scip_check but there is no correspondence with the solution in sepaSubtour function as the arc variables have different values (the objective function value is same in both cases).
>   
> 
> I am just prototyping here and I assumed the solution values in both the scip_check and sepa_lp (thus in sepaSubtour)  would be same but the solution value in subTour seems to be fractional (objective values are the same but the arc variables are fractional in sepaSubtour whereas they are integral in scip_check).
> 
> Maybe I am missing something very trivial here. Is there a way to get the same solution in the sepa_lp as scip_check?
> 
> Regards,
> Nag
> 
> 
> _______________________________________________
> 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