I am working on a two-stage stochastic program where the master is a mip model, and slaves only have continuous variables. For each scenario, I pass three variables set (k, a, b: first stage decisions, all binary) from master to slave. Then stochastic parameters realize. The slave problem has M constraints between decision variables a,b and contiunous s,c,x variables (slave variables). The first stage variable k is not in any M constraint in the slave problem, but it constraints s,c variables. I have two questions: 1. With these conditions, can I use combinatorial benders cuts in the form: a(sum where/=1)+(1-a)(sum where/=0)+b(sum where/=1)+(1-b)(sum where/=0) >= 1 for this problem? 2. I havent seen any papers using combinatorial benders cuts in stochastic programming to generate feasibilty cuts. Are there any other conditions? Since we generate feasibility cuts for each scenario subproblem, I believe we can use these cuts as feasibility cuts for l-shaped algorithm. Thanks for the help in advance.
asked
paret |