Distal and non-distal NIP theories (Q1935868)

A notion of ``pure instability'' called \textit{distality} is introduced in the framework of NIP theories. First, distality is defined for indiscernible sequences in the following way. Let \(I = \langle a_i \, : \, i \in {\mathcal I} \rangle\) be an indiscernible sequence over a set \(A\). The EM-type of \(I\) over \(A\) is the family \((p_n : n \in \omega)\), where \(p_n \in S_n(A)\) is the type of \((a_{\sigma(k)} : k < n)\) for \(\sigma : n \mapsto {\mathcal I}\) any increasing embedding. Furthermore, a cut \((I_1, I_2)\) of \(I\) is called Dedekind if both \(I_1\) and \(I_2^\star\) (\(I_2\) with the order reversed) have infinite cofinality. An element \(a\) fills a cut \((I_1, I_2)\) of \(I\) if and only if \(I_1 + \{a \} + I_2\) remains indiscernible. At this point, define \(I\) \textsl{distal} if and only if for any dense \(J\) of the same EM-type as \(I\) and for every pair of distinct Dedekind cuts of \(J\), if \(a\) fills the former cut and fills the latter, then \(a\) and \(b\) are independent over \(J\), in the sense that \(J \cup \{ a, b \}\) with \(a\) and \(b\) falling in their cuts is still indiscernible. A NIP theory \(T\) is said to be \textsl{distal} if all indiscernible sequences are distal in \(T\). So distality can be viewed as an opposite notion to total indiscernibility. Distal theories and sequences are extensively studied. The main results here characterize distal theories in terms of invariant types and generically stable measures as follows: A NIP theory is distal if and only if any two invariant types that commute are orthogonal, if and only if all generically stable measures are smooth; furthermore it is enough to check any one of these conditions in dimension 1. It follows that o-minimal theories, as well as the \(p\)-adics, are distal. Then the intermediate case of NIP theories that are neither stable nor distal is considered, with the aim of understanding to what extent non-distality is witnessed by stable-like interactions between tuples. A notion of \(s\)-independence is introduced over a sufficiently saturated model \(M\). This independence implies forking independence, is symmetric and has bounded weight. As an application of this idea, it is shown that, if \(T\) is a NIP theory, \(I = I_1 + I_2 + I_3\) is indiscernible, \(I_1\) and \(I_3\) are infinite, \(I_1 + I_3\) is \(A\)-indiscernible, then for every formula \(\phi(x, a)\) in \(L(A)\) the set \(\{ b \in I_2 :\, \models \phi(b,a)\}\) is finite or cofinite. Finally, \textit{sharp} theories are introduced. These are the NIP theories where over \(M\) every tuple is \(s\)-dominated by the realization of some generically stable type, so intuitively the stable part of types is witnessed by generically stable types. A criterion is given for sharpness which only involves examining indiscernible sequences of elements (not tuples). On this basis it is proved that dp-minimal theories are sharp.