Incompatibility between Lazy cuts and warm start

I have a question about lazy cuts.

In order to, solve an MILP, I developed a Benders' cut approach. In this approach, I use lazy callbacks to generate infeasibility/optimality cuts for the master. It works perfectly when starting from scratch. However, since I have a heuristic method for returning good solutions, I want to use one of the latter as a starting point for the master.

The issue is that when I set the variables of the master problem with the GRB_DoubleAttr_Start attribute, Gurobi prints

User MIP start did not produce a new incumbent solution

It is not an issue with the initial solution provided; indeed, I implemented the same approach using CPLEX and the latter found the solution valid. Furthermore, the lazy cut added by Gurobi is an optimality one, meaning it found an optimal solution for the subproblem.

Does adding a lazy cut automatically invalidate the warm start solution, even if this cut is valid ?

Thanks for your answer,

C. Juvigny, Ph.D

Incompatibility between Lazy cuts and warm start

Comments

Didn't find what you were looking for?