2577. Kyle Gannon, James E. Hanson Model theoretic events
E-mail:
Submission date: 24 February 2024
Abstract:
How do we randomly sample an infinite sequence from a first order structure?
What properties might hold on almost all random sequences? Which kinds of
probabilistic processes can be meaningfully applied and studied in the model
theory context? This paper takes these questions seriously and advances a
plausible framework to engage with probabilistic phenomena.
The central object of this paper is a probability space. The underlying set
of our space is a standard model theoretic object, i.e. the space of types in
countably many variables over a monster model. Our probability measure is the
iterated Morley product of a fixed Borel-definable Keisler measure. Choosing a
point randomly in this space with respect to our distribution yields a random
generic type in countably many variables. We are interested in which events
hold for almost all random generic types. We consider two different flavors of
model theoretic events: (1) When is the induced structure on almost all random
generic types isomorphic to a fixed structure? (2) For a fixed formula which is
unstable, IP, sOP, etc., what is the probability that a random generic type
witnesses this dividing line?
For (1), we show that if our measure satisfies a particular extension axiom,
then there exists a structure N such that the induced structure on almost all
random generic types is isomorphic to N. The proof echos a celebrated result
of Glebskii et al. and Fagin concerning the existence of almost sure theories.
We also provide examples where no such model exists. For (2), we show that if
our initial distribution is fim, then almost no random generic types witness
instability, IP, or sOP. In the local NIP context, we use results from
combinatorics to prove that for any Borel-definable measure, the average value
of witnessing k-instability across all permutations converges to 0. Some
examples are provided.