[SCIP] primal heuristics and new variables

[James Cussens]

Hi all,

I am intending to implement a primal heuristic which, as well as suggesting new primal solutions using the existing problem variables, can generate primal solutions only possible if new problem variables were added. I would appreciate advice on 'neat' ways of implementing this.

If I were to just create these new variables in a heuristic plugin then I might end with a primal solution with better objective value than the current dual bound - since variables were added after the solution of the LP relaxation was computed. I guess I could just force recomputation of the a (new) LP solution as soon as any new variables are created by the heuristic plugin.

I'm thinking that perhaps I should implement this primal heuristic as a pricer, even though the new variables may not affect the dual bound. Presumably there's no particular problem with a pricer generating primal solutions. But I'm worried that formally implementing a procedure as a pricing algorithm when it is "really" a primal heuristic will lead to problems.

My primal heuristic is highly problem specific (it generates DAGs). I would be surprised if this issue had not come up for specific problems.

James

James Cussens
Room MVB 3.26
School of Computer Science, University of Bristol
Phone: +44 (0)117 455 8723
https://jcussens.github.io/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20250811/32fa2414/attachment.html>