Constrain one variable to be greater than another

Yes. In some occasion where you have strong information (in your prior or likelihood), you might be able to get away with not using a ordered vector. But in general it helps with model convergent.