This paper shows how to achieve, under certain conditions, abstract-interpretation algorithms that enjoy the best possible precision for a given abstraction. The key idea is a simple process of successive approximation ...