Originally posted by: prasr29

HI,

           I am solving a maximization MIP using benders decomposition. I pass the master problem's values and solve the sub problem first and then solve the master problem for getting the optimal solution. The problem I am facing is this: for some values of the initial master problem variables the benders code gives the optimal solution, whereas for some other values the code the UB (master prob's objective) goes below the sub problem's objective. I have checked the dual and the code entirely several times and also with my colleagues  and I am pretty sure with the model and code. Does Cplex have a role to play in this? Is it possible?? I am using CPLEX in cpp environment. Thank you.

Originally posted by: EXCT_RALF_GOLLMER

You should give more info, e.g.:

You're using original Benders, so I suppose you have only continuous variables in the subproblem, thus it is an LP and has a LP dual.

- What do you take as the UB - the lower bound or the best integer objective?

- To which MIP gap is the master solved? Is the difference in objective values greater than the gap?

- Did you think about the necessity to add optimality cuts?