Virtual functions for the flexible definition of branching strategies
are implemented in the class ABA_SUB. We explain them together
with the concept of branching rules in
Section 
.
If constraints are generated heuristically, then the concept of delayed branching can be advantageous. Instead of generating the sons of a subproblem in a branching step, the subproblem is put back into the set of open subproblems. There it stays several rounds dormant, i.e., other subproblems are optimized in the meantime, until the subproblem is processed again. If between two successive optimizations of the subproblem good cutting planes are generated that can be separated from the pool, then this technique can accelerate the optimization. The maximal numbers of optimizations and the minimal number of dormant rounds can be controlled by parameters.