Keisler's order has infinitely many classes (Q1650013)

In the 1970s, Keisler introduced a pre-order on theories, now called Keisler's order, defined as follows: given theories \(T_1\) and \(T_2\) (not necessarily in the same language), we write \(T_1\trianglelefteq T_2\) if, for any regular ultrafilter \(\mathcal{U}\) on a set of size \(\kappa\) and any models \(\mathcal{M}_1\models T_1\) and \(\mathcal{M}_2\models T_2\), if \(\mathcal{M}_2^\mathcal{U}\) is \(\kappa^+\)-saturated, then so is \(\mathcal{M}_1^\mathcal{U}\). (A theorem of Keisler shows that whether or not this is the case is independent of the choice of models of \(\mathcal{M}_1\) and \(\mathcal{M}_2\), whence this really is a fact about the theories \(T_1\) and \(T_2\). Also, one often identifies theories that are equivalent in this pre-order, whence one speaks of the induced partial ordering on classes.) After introducing the order, Keisler proved that there were minimum and maximum classes. \textit{S. Shelah} [Classification theory and the number of non-isomorphic models. 2nd rev. ed. Amsterdam etc.: North-Holland (1990; Zbl 0713.03013)] later proved that the stable theories comprise the first two classes, the dividing line between the two being the finite cover property. \textit{M. E. Malliaris} [Persistence and regularity in unstable model theory. Berkeley, CA: University of California (PhD Thesis) (2009)] then proved that the class of the random graph was the minimum unstable class in the order. \textit{M. Malliaris} and \textit{S. Shelah} [Adv. Math. 249, 250--288 (2013; Zbl 1323.03042)] then found a fifth class in the order. For a while, it was believed that this amount of information might be close to the true nature of the order, namely that there is a small, finite number of classes in the order. The paper under review completely reverses this train of thought: the authors produce an infinite, descending chain of classes of simple theories in the order. The idea is to use, for \(m>k\geq 2\), the theories \(T_{m,k}\), which are the model completions of the theory of one symmetric, irreflexive \((k+1)\)-ary relation with no complete graphs on \(m+1\) vertices. A result of \textit{E. Hrushovski} [in: Model theory and applications. Based on the Euro-Conference in model theory and applications, Ravello, Italy, May 27--June 1, 2002. Rome: Aracne. 151--212 (2002; Zbl 1082.03035)] says that these theories are simple and have trivial forking. If one defines the theory \(T_n^*\) to be the disjoint union of the theories \(T_{k,k+1}\) for \(k>2n+2\), then the authors show that \(T_{n+1}^*\ntrianglelefteq T_n^*\) for all \(n\). While the proof is far too complicated to describe in this review, we only mention that one of the key ideas is to import the Kuratowski-Sierpinski characterization of the distance between alephs via existence of free sets in set mappings.