Implementations of Nondeterministic Programs

Non-determinism is widely recognized as a valuable tool for program development [Dijkstra 1976]. It appears also as the natural environment for the , evolution of systems of parallel processes; sequential implementation of such systems through non-deterministically simulating their distributed behaviour can be an effective programming technique [Guerreiro 1981, 1983b]. Furthermore, the concept of non-determinism itself arises many interesting theoretical questions [Plotkin 1976, Schmidt 1979, Apt and Plotkin 1981, Back 1981, Hennessy 1982, Poigne 1982, etc.]. Nevertheless, if we are interested in obtaining an implementation for a program that describes a non-deterministic situation, sooner or later we are bound to consider, implicitly or explicitly, a deterministic version of it, since, at some level of detail, the existing machines operate in a deterministic fashion. : Therefore, the problem arises of determining exactly in what way can the i non-determinism be suppressed, or, at least, under which conditions is the rdeterministic program a legitimate version of the non-deterministic one. More generally, we want to have a means of deciding whether a given program P1 can be !~ replaced by another program P2, in such a way that it is impossible by looking only at the results of the execution of P2 to find out that the executed program was not P1 after all.