[SCIP] FW: Constraint cutting planes in constraint handlers

Ahmed Ibrahim Ahmed.Ibrahim at umanitoba.ca
Sun Mar 6 18:19:27 CET 2016


________________________________
From: Ahmed Ibrahim
Sent: Sunday, March 06, 2016 11:03 AM
To: Gerald Gamrath
Subject: RE: [SCIP] Constraint cutting planes in constraint handlers

Hi Gerald,

Thanks for this response. Actually it fixed a misunderstanding I had. May I also ask, whether the same applies for quadratic constraint separators?

Regards,
Ahmed
________________________________
From: Gerald Gamrath [gamrath at zib.de]
Sent: Tuesday, March 01, 2016 9:36 AM
To: Ahmed Ibrahim; scip at zib.de
Subject: Re: [SCIP] Constraint cutting planes in constraint handlers

Dear Ahmed,

the logic or, setppc, and varbound constraint handlers have separation routines, but these only generates the linear rows representing the specific constraints. They do not create 'real' cutting planes.

Remember that a constraint is not directly present in the LP, but it will normally create one or more rows to add the constraint (or a relaxation of it) to the LP. For linear constraints or their specializations (e.g., logic or, setppc, and varbound), the constraint can be represented by one linear row.
If constraints have their 'initial' flag set to TRUE when created (which is the default), the initial relaxation will be added to the LP when building the LP data structure at the root node. Linear constraints and specializations are then fully represented in the LP and no separation is needed later, anymore.

On the other hand, if 'initial' is set to FALSE or 'removable' to TRUE, the row corresponding to a constraint may not be in the LP. The separation procedure then checks whether the row would by violated by the current LP solution and adds it, if needed. This is similar to the concept of lazy constraints used in, e.g., CPLEX.

Best,
Gerald

On 29.02.2016 20:51, Ahmed Ibrahim wrote:
Hi all,
I was wondering if any one knows where can I find an explanation of how cutting planes for logic or setppc and varbound constraint handlers are obtained?

Regards,
Ahmed



_______________________________________________
Scip mailing list
Scip at zib.de<mailto:Scip at zib.de>
http://listserv.zib.de/mailman/listinfo/scip


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20160306/3eabfac7/attachment.html>


More information about the Scip mailing list