As for active constraints and variables also every subproblem has its own linear program, which is only set up for an active subproblem. Of course, the initialization at the beginning and the deletion of the linear program at the end of the subproblem optimization costs some running time compared to a global linear program, which could be stored in the master. However, a local linear program in every subproblem will again simplify the implementation of a parallel version of ABACUS. Our current computational experience shows that this overhead is not too big. However, if in future computational experiments it turns out that these local linear programs slow down the overall running time significantly, the implementation of a special sequential version of the code with one global linear program will not be too difficult, whereas the opposite direction would be harder to realize.