Levi contractions and AGM contractions: A comparison (Q1903580)

Among the AGM postulates for contraction, the one that has been most open to query is that of recovery. In his book: The fixation of belief and its undoing (Cambridge UK, 1991), \textit{Isaac Levi} proposed modifying the AGM partial meet construction for contraction in such a way as to invalidate recovery. For a closed belief set \(K= \text{Cn} (K)\), rather than take \(K-x\) to be the intersection of a subset of \(K\perp x\) (the set of all maximal subsets of \(K\) failing to imply \(x\)), he suggests taking it as the intersection of a subset of \(K\perp' x\), where the latter is the set of all subsets \(Y\) of \(K\) failing to imply \(x\) but ``saturable'' by \(\neg x\) in the sense that \(\text{Cn} (Y\cup \{\neg x\})\) is complete. In the paper under review, the authors give a precise formulation of this kind of contraction, in particular taking care with the limiting case when \(x\not\in K\) that was left vague in Levi's discursive presentation. It is immediate from their definition and results of AGM (lemmas 4.4 and 4.5, proven afresh in the paper, are in fact in AGM) that \(K\perp x\subseteq K\perp' x\), so that every AGM partial meet contraction function is a ``Levi contraction'' function (but not conversely). The authors provide syntactic postulates for such Levi contraction functions, and prove the appropriate representation theorem. The postulates consist of those for ``withdrawal functions'' (i.e. the ``basic'' AGM postulates other than recovery) plus the following limiting case of recovery, called ``failure'': \(K-x =K\) whenever \(x\in \text{Cn} (\emptyset)\). A sufficient condition is also given for the satisfaction of the two ``supplementary'' AGM postulates. Reviewer's comments: (1) It is not difficult to prove the following result, which helps appreciate the relationship between elements of \(K\perp' x\) and those of \(K\perp x\): For any \(x\in K= \text{Cn} (K)\) we have \(Y\in K\perp' x\) iff \(Y= \text{Cn} (Y)\) and thre is a \(Z\in K\perp x\) with \(Z\cap \text{Cn} (x) \subseteq Y\subseteq Z\). (2) Criticisms and alleged intuitive counterexamples to recovery as applied to closed belief sets, are based on the idea that contraction should take into account some additional ``justification structure'' of the belief set, and should observe the following maxim which Fuhrmann has called the ``filtering condition'': when a belief \(y\) is withdrawn in order to form the contraction of one's beliefs by \(x\), then the contraction by \(x\) should not contain any items that were believed ``just because'' \(y\) was believed. It should be noted that whilst Levi contraction succeeds in avoiding recovery, it like AGM contraction can violate the filtering condition for a suitable choice of additional justification structure, for the simple reason that all AGM contraction functions are Levi contraction functions. Indeed, one can also construct Levi contractions that are not AGM contractions but still fail the filtering condition (e.g. in the example on page 107, assume that \(\neg q\vee p\in K\) ``just because'' \(p\in K\), and put \(K-p= \text{Cn} (\neg q\vee p))\).