[SCIP] how to augment the lower bound

[Gerald Gamrath]

Hi Brannon,

your idea to extend the LP relaxator was going in the right direction. 
It sounds like what you would like to do is write a custom relaxator 
plugin in SCIP. If you give it a negative priority, it will be called 
after the LP has been solved. See 
https://www.scipopt.org/doc-8.0.3/html/RELAX.php for more details and 
perhaps also check out the relaxator example.

Best,

Gerald

On 16.10.23 18:39, Brannon King wrote:
> I have a scheduling problem with a poor LP relaxation. However, given
> a relaxed solution, I can augment that using some additional problem
> knowledge. In other words, I want the ability to manually increase
> every node's lower bound.
>
> When I attempt to do this via cuts, either up front or lazily, my
> problem takes longer to solve. This is a common problem in scheduling
> problems; cuts that try to prop up the makespan destroy the search
> direction. How can I use my augmented relaxation without destroying
> the search direction? I picture that my improved lower bound would
> eliminate poor subtrees sooner, but it would not change the overall
> branching strategy.
>
> I originally pictured that I would take the LP relaxator, inherit from
> it, and make a minor modification to where it sets the lower bound. In
> looking around, though, it seems that the LP relaxation is not in a
> plugin. Where is it done presently? Where would I intercept this?
>
> Maybe I should use a node selector plugin? This seems different than
> overriding the lower bound, though. I don't understand how the node
> priority is set already. Does it come from the lower bound? Or is it
> based on properties of the variables, which one looks like it needs to
> branch, etc?
>
> Thanks for your time,
> Brannon
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> https://listserv.zib.de/mailman/listinfo/scip