Another option in some cases is to just map the space to something else. Like in this case, x and y are unconstrained and are what the optimizer sees, a = x and

Another option in some cases is to just map the space to something else. Like in this case, x and y are unconstrained and are what the optimizer sees, a = x and b = (1 - a) - abs(y). Not always easy to do, but can work in some cases if the optimizer cant deal with it natively.