On Oraclizable Networks and Kahn's Principle

Abstract

In this paper we investigate generalizations of Kahn's principle to nondeterministic dataflow networks. Specifically, we show that for the class of "oraclizable" networks a semantic model in which networks are represented by certain sets of continuous functions is fully abstract and has the fixed-point property. We go on to show that the oraclizable networks are the largest class representable by this model, and are a proper superclass of the networks implementable with the infinity fair merge primitive. Finally, we use this characterization to show that infinity fair merge networks and oraclizable networks are proper subclasses of the networks with Egli-Milner monotone input-output relations.