Lack of Finite Characterizations for the Distance-based Revision

Abstract

Lehmann, Magidor, and Schlechta developed an approach to belief revision based on distances between any two valuations. Suppose we are given such a distance D. This defines an operator |D, called a distance operator, which transforms any two sets of valuations V and W into the set V |D W of all elements of W that are closest to V. This operator |D defines naturally the revision of K by A as the set of all formulas satisfied in M(K) |D M(A) (i.e. those models of A that are closest to the models of K). This constitutes a distance-based revision operator. Lehmann et al. characterized families of them using a loop condition of arbitrarily big size. An interesting question is whether this loop condition can be replaced by a finite one. Extending the results of Schlechta, we will provide elements of negative answer. In fact, we will show that for families of distance operators, there is no "normal" characterization. Approximatively, a normal characterization contains only finite and universally quantified conditions. These results have an interest of their own for they help to understand the limits of what is possible in this area. Now, we are quite confident that this work can be continued to show similar impossibility results for distance-based revision operators, which suggests that the big loop condition cannot be simplified.