Belief revision, conditional logic and nonmonotonic reasoning (Q1903579)

The central concern of this wide-ranging paper is to develop an account of belief revision that follows the general idea of the AGM paradigm, but which also gives us information about iterated revision. The strategy is to take one of the presentations of the well-known AGM approach, namely that in terms of epistemic entrenchment relations, and refine it to ensure that the output of a revision is of the same kind as the input, viz. a belief set with associated entrenchment relation. Specifically, the author proposes a rule for taking any entrenchment relation representable as a ranking into the positive integers, to a new such relation that reassigns the rank of a given formula whilst making ``as little change as possible'' to the ranks of other formulae. Postulates are given for the generated revision operations. The author also discusses questions of computability, and the translation of his rule into the languages of nonmonotonic inference and of conditionals. Reviewer's comments: Readers should take care with the following technical points. (1) Whilst every revision operation on a fixed belief set \(K\) satisfying the AGM postulates \(K^*1\) to \(K^*8\) can be characterized by an epistemic entrenchment relation, not quite all of them can be characterized by the author's ranked epistemic entrenchment relations -- indeed, presumably not all of them can be characterized by well-ordered entrenchment relations. In other words, even in the non- iterative context, the author's construction concerns a slightly proper subset of the AGM functions. (2) On page 66 the author claims that there are no infinite descending chains of ranked epistemic entrenchments under his relation \(\leq\). However, all that is guaranteed by the definition is that any infinite descending chain will have a greatest lower bound -- which need not be an element of the chain unless we assume, for example, that the language admits only finitely many non-equivalent formulae. The author has stated to the reviewer that nothing in what follows appears to depend on this slip (an appeal to the descending chain condition on page 75 being replaceable by an appeal to glb and adding the assumption that the base \(\Gamma\) is finite). (3) Concerning theorem 5.4, there is no guarantee the maximum value, there referred to, exists in general. However, it will exist when \(\Gamma\) is finite. (4) There is an inaccuracy in the paraphrase given just before the formal enunciation of theorem 3.3. It should read ``the consequences of \(\neg A\) are ranked in \(K^+_{A,\alpha}\) the same as in \(K\), whilst consequences of \(A\) have their rank increased to \(\alpha\) if less than \(\alpha\) in \(K\)''. (5) In definition 5.1 ``each non-theorem'' is intended to mean ``each non- theorem in \(\Gamma\)''.