[SCIP] SCIP question: priced variables never show up in LP branching candidates

Ambros Gleixner gleixner at zib.de
Tue Aug 21 15:43:40 CEST 2018


Dear Myroslav,

Am 20.08.2018 um 11:46 schrieb myroslav:
> Dear all,
> 
> 
> I am a PhD student at the university of Würzburg, I work on Graph 
> Drawing and am trying to implement a branch and price algorithm to solve 
> ILPs arising from some graph drawing problems using SCIP. I am quite new 
> to IP solvers and algorithms and wanted to ask two questions: 1) about 
> what could possibly go wrong in my implementation and 2) about how 
> branch and price algorithm works in  SCIP.
> 
> 1)
> 
> For any node in the branching tree when I do branching in the callback 
> BRANCHEXECL  it turns out that the priced variables never show up among 
> the LP branching candidates (obtained using method 
> SCIPgetLPBranchCands). It looks like they always have zero values in the 
> solution of the LP.  That is weird, because if the pricer has added 
> them, they should have nonzero values in the LP solution.  At the end of 
> the solving process the solution eventually contains some of the priced 
> variables even thought none of them was branched on. I wondered, what I 
> possibly could be doing wrong.

That is hard to say.  You would need to provide more information.  Are 
you sure that for your formulation in every optimal LP solution some of 
the priced variables are nonzero?


> 2)
> 
> Also I wanted to ask specifically about the branch and price algorithm 
> just to make sure if I understood it correctly. Does an LP at a 
> branching node get the priced variables of the parent node or the 
> pricing starts anew from a scratch(with no priced variables initially)?

The priced variables are inherited.

Best,
Ambros


> Thanks in advance,
> 
> Myroslav
> 
> _______________________________________________
> 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