Keisler Measures and Generically Stable Random Types
Karim Khanaki · 2026
Read Keisler Measures and Generically Stable Random Types by Karim Khanaki (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-15870.txt.
Introduction
Karim Khanaki defines notions called rgs and irgs for Keisler measures, motivated by generically stable random types and their Morley sequences. He characterizes both notions via averages of classical first-order formulas over apt probabilistic partitions. He compares them with fim, fam and self-averaging, showing that for types, fim, irgs and rgs coincide. Every irgs measure proves dependent, hence symmetric. For irgs ones, the instability events O, I and L have probability zero, extending earlier results beyond the fim case.
Catalog record
| CREATOR | Karim Khanaki |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60363 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.15870
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Keisler Measures and Generically Stable Random Types Karim Khanaki Abstract We introduce the notions of r g s rgs and i r g s irgs for Keisler measures, motivated by the study of generically stable random types and their associated Morley sequences. We obtain characterizations of these notions in terms of averages of classical first-order formulas over suitable probabilistic partitions (Theorems 3.2 and 3.3 ). We compare these notions with f i m fim , f a m fam , and self-averaging, and show that for types the notions f i m fim , i r g s irgs , and r g s rgs coincide. We prove that every i r g s irgs measure is dependent (Theorem 4.5 ); consequently, such measures are symmetric (Corollary 4.8 ). Furthermore, we show that for i r g s irgs measures the model-theoretic instability events 𝐎 φ \mathbf{O}^{\varphi} , 𝐈 φ \mathbf{I}^{\varphi} , and 𝐋 φ \mathbf{L}^{\varphi} have ℙ μ \mathbb{P}_{\mu} -measure zero (Theorem 5.4 ), extending results from [ 8 ] beyond the f i m fim case. 1 Introduction Generic stability has a long historical background, dating back to the early developments of stability theory by classical model theorists, notably Bruno Poizat. Its roots can even be traced further back to ideas in algebraic geometry, particularly in the work of André Weil. Most of the initial systematic investigations of this notion were first carried out in stable theories and were later extended to NIP theories. Pillay and Tanović [ 13 ] introduced the notion of generically stable types in arbitrary theories, providing an appropriate generalization of definable and finitely satisfiable types in the stable setting. On the other hand, the study of Keisler measures in model theory has seen significant progress in recent years, driven by connections to probability, analysis, and learning theory. For Keisler measures in NIP theories, the notion of generic stability has been appropriately generalized and has been extensively studied. However, despite this historical development, a satisfactory notion serving as an appropriate generalization of generic stability for measures in arbitrary theories has not yet been identified, or at least there is no consensus on such a notion. The difficulty is twofold. First, unlike the NIP setting, various different notions such as definability, finite satisfiability, fam , and fim (see, e.g., [ 4 ] ) are no longer equivalent in general. Second, there is no analogue of Morley sequences for measures that plays the same canonical role as in the case of types. A natural idea for addressing the second issue is to study random types, viewed as analogues of Keisler measures, in the randomization of the theory, as developed in several works (see, e.g., [ 1 , 12 , 5 , 7 ] ). Since it is a well-established belief that many model-theoretic properties are preserved under randomization, it is natural to define a measure as generically stable if its randomization is a generically stable type. With this background in mind, the goal of the present paper is to introduce notions satisfying the desired properties. In this sense, our work continues and is intended to be a completion of the ideas developed in previous works (see, e.g., [ 11 , 12 , 10 , 5 ] ). In particular, in [ 5 ] the notion of self-averaging was proposed as a candidate for generic stability of measures. This notion captures several of the expected desirable properties of generic stability. However, it still does not fully meet the anticipated requirements, and we believe that it fails to adequately reflect the fundamental distinction between types and measures. We argue that the notion of dependence (see [ 10 ] ), which has been less emphasized in the literature, more accurately captures this distinction. The goal of this paper is therefore to propose suitable notions of generic stability for measures, in such a way that, on the one hand, they are equivalent to generic stability of types in the randomization, and on the other hand, they satisfy the dependence property. In this paper, we study two new classes of Keisler measures, which we call randomly generically stable ( rgs ) and independently randomly generically stable ( irgs ). These notions are designed to capture the behaviour of measures whose associated random types are generically stable. We characterize these notions by means of averages of classical first-order formulas over suitable probabilistic partitions. The characterizations involve a fixed atomless probability space and a limiting process over sequences of atomic measures, which may appear technical at first sight. Nevertheless, as we argue, they encode natural uniform convergence conditions that are exactly what is needed to pass from the classical to the continuous setting. We then compare these new concepts with already existing ones: fim (frequency interpretation measure), fam (finitely approximated measure), and self‑averaging measure . It turns out that the implications {\lx@inpgf@ignorespaces\mathrm{fim}} {\lx@inpgf@ignorespaces\mathrm{irgs}} {\lx@inpgf@ignorespaces\mathrm{rgs}} {\lx@inpgf@ignorespaces\mathrm{fam}} {\lx@inpgf@ignorespaces\mathrm{dependence}} hold, and for types all three first notions coincide (Theorem 4.5 and Corollary 3.7 ). Moreover, we prove that every irgs measure is dependent in the sense of the author’s previous work [ 10 ] , thereby extending the known fact that every fim is dependent. An illustrative example (Example 3.8 ) shows that the difference between rgs and irgs is essential: An element of a simple model M^{\Omega} that realizes all the good properties of a generically stable type is an rgs but not an irgs unless the random element is constant. In particular, Section 5 investigates the probabilistic behaviour of irgs measures with respect to model‑theoretic instability events. We show that for irgs measures the events \mathbf{O}^{\varphi} , \mathbf{I}^{\varphi} , \mathbf{L}^{\varphi} (order, independence, strict order) have \mathbb{P}_{\mu} -measure zero, thereby extending the results of [ 8 ] from fim to irgs . The paper is organised as follows. Section 2 recalls the necessary background on generically stable types in continuous logic, simple models of randomization, and the canonical random type r_{\mu} ; it also introduces the auxiliary notions of stable basic sequences and independent stable basic sequences. Section 3 defines the new concepts rgs and irgs for Keisler measures, gives their characterisations via stable basic sequences (Theorems 3.2 and 3.3 ), and compares them with fim , fam and self‑averaging (Corollary 3.7 ). Section 4 proves that every irgs measure is dependent (Theorem 4.5 ) and derives consequences such as symmetry of the Morley product (Corollary 4.8 ). Section 5 extends the analysis of tame events from [ 8 ] to irgs measures, showing that \mathbb{P}_{\mu}(\mathbf{O}^{\varphi})=\mathbb{P}_{\mu}(\mathbf{I}^{\varphi})=\mathbb{P}_{\mu}(\mathbf{L}^{\varphi})=0 (Theorem 5.4 ). We end with some concluding remarks. 2 Preliminaries In this section, we review the basic material needed for the proofs of the main theorems. We first review the notion of generically stable types in continuous logic (and hence in discrete logic), together with the equivalent characterizations that will be used throughout the paper. We then introduce simple models of randomization and random types. And at the end of this section, we introduce notions of sequences with properties that capture what we think a Morley sequence for a measure ought to have. We