Originally posted by: JDalal

I have experience with correct implementation of Benders Decomposition with Lazy constraint callback for MIPs.

Recently, I am dealing with a very large LP which is taking unacceptably long time by cplex. So,  I was thinking to decompose the LP in Bender's fashion and use callback as done for MIP.  I know that in MIP, whenever cplex finds an integer-feasible solution, the callback code is invoked. Is it same for continuous callback?   When I tried to query the master problem's (clarifying again: it is LP, not MIP) solution vector inside my callback code, I got  "no solution exists". Can somebody clarify whether it is possible to query a LP's decision variable through callback (I assume, it is invoked  whenever cplex gets a feasible LP solution) at the time cplex is still  solving it? Thanks in advance for any suggestion.

Originally posted by: T_O

Please remember that - assuming you are using the dual simplex method - you won't find any feasible solution during the solution process. The first primal feasible solution will be optimal.

I guess not using the dual simplex wouldn't make any sense either, as you want to have warm start capabilities during Benders Decomposition.

Best regards,
Thomas

What callback did you use? The title of your question suggests that you are using a ContinuousCallback? That callback does not even have a getValue() method, so how did you attempt to query the current solution?

You can use a lazy constraint callback with an LP as well. You just have to force the model type to be MIP. This can be done by adding a dummy integer variable (this variable will be eliminated by presolve anyway).