Lazy cns with Gurobi

[FabrizioLacalandra]

Dear all,
when using pure Jump model i exensively use the lazy attribute with some cns using Gurobi as a solver with the syntiax

MOI.set(MyMod, Gurobi.ConstraintAttribute("Lazy"), CnsName, K)

where K is 1/2 or 3 depending on the Gurobi possibilities. Besides the need to indexing single constraint in for loops, i receive an error using the same approach within a certain Bilevel problem using BilevelJump.
Question is: is it possible to do that kind of attribute setting, or the BJump interface does not enable it?

Regards,
Fabrizio

[joaquimg]

Hi, the short answer is that today it is not possible out of the box.

It would be reasonable to allow this for the upper level model.

What would that do for a lower level constraint?

Remember that the lower level will be reformulated to be represented by its KKT conditions, so there will be many dual constraints and complementarity constraints.

[FabrizioLacalandra]

Thanks for your answer,

indeed i think it could be and interesting feature to add to BJuMP and indeed my interest is at the moment related to the Upper level of the bilevel model, which is a complex MIP with linear cns that could be "lazied" the way Gurobi let you. It could also be possible to add some user cut as well.

Regards,
Fabrizio