infeasible solution

[lorenzoreus]

I have infeasible solution to this simple problem. However when ADDING a constraint (@constraint(sp, x == 500.0)) , the solver gives me a solution!
Please can someone help to undestand what is going on? Thanks

m = SDDPModel(
                  sense = :Min,
                 stages = 3,
                 solver = ClpSolver(),
        objective_bound = -10000000.00
                                ) do sp, t
    initial_stock= 1500.0
    @state(sp, stock >= 0.0   , stock0 == initial_stock )

    @variable(sp, 0.0 <=x<= 1000.0)

    @constraint(sp,  stock0 - x  == stock)
    #@constraint(sp,  x   == 500.0)
    if  t == 3
        @constraint(sp,  stock==0.0)
    end
    price= [550.0, 600.0, 650.0]
    @stageobjective(sp, price[t] * x)
end

status = SDDP.solve(m, iteration_limit = 100)
SDDP.getbound(m)

Edit(odow): formatting

[odow]

The problem is in the third stage. You might come in with stock0 = 1500. Then there is a constraint that stock == 0. From the constraint you have, x == stock0 - stock, so x = 1500, but x <= 1000 so the problem is infeasible.

SDDP.jl requires relatively complete recourse. That is, for all possible values of the incoming state variables, there is a feasible control. This model doesn't satisfy that requirement.

Either relax the upper bound on x, or allow the stock == 0 constraint in the last stage to be violated and penalise this violation in the objective.

[lorenzoreus]

thanks oscar for your fast response . perfect explanation

[odow]

Great that this solved the problem.