Comment by unnah

If the volume of feasible space is small, the optimizer may have considerable trouble finding the feasible region(s) in the domain. The simple solution is to use a penalty function instead, i.e. add a term like M*max(0, 1-(a+b)) to the objective, with sufficiently large M. If the optimum is at the boundary of the feasible region, you might still get a solution that is slightly infeasible, which can be worked around with more special case logic...