💻 computer science

Goldblatt-Thomason Theorem for Probability Logic

This paper establishes the Goldblatt-Thomason theorem for probability logic interpreted over Markov processes, demonstrating its utility in defining Harsanyi type spaces and providing variants for specific subclasses of these structures.

Original authors: Somayeh Chopoghloo (School of Mathematics, Institute for Research in Fundamental Sciences), Massoud Pourmahdian (School of Mathematics, Institute for Research in Fundamental Sciences), Reihane Zoghifa

Published 2026-07-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Somayeh Chopoghloo (School of Mathematics, Institute for Research in Fundamental Sciences), Massoud Pourmahdian (School of Mathematics, Institute for Research in Fundamental Sciences), Reihane Zoghifard (School of Mathematics, Institute for Research in Fundamental Sciences)

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to describe a complex, unpredictable world using only a limited set of rules. In this paper, the authors are working with Probability Logic, a special language designed to talk about chance. Instead of just saying "It will rain" or "It won't rain," this language lets you say things like, "There is at least a 70% chance it will rain."

The paper focuses on Markov processes, which are mathematical models for systems that change over time based on probabilities. Think of these as giant, complex dice-rolling machines where the outcome of the next roll depends on the current state, but with infinite possibilities and strict mathematical rules.

Here is the core story of the paper, broken down into simple concepts:

1. The Big Question: Can We Describe a Machine by Its Rules?

The authors want to know: If we have a specific type of probability machine, can we write a set of sentences in our "Probability Language" that perfectly describes it?

If you can write a sentence (or a list of sentences) that is true only for that specific machine and false for all others, then that machine is "definable." The paper asks: What are the rules a group of machines must follow to be describable this way?

2. The "Goldblatt-Thomason" Map

To answer this, the authors use a famous mathematical tool called the Goldblatt-Thomason Theorem. You can think of this theorem as a map or a checklist.

In simpler logic (like standard Kripke frames), this map says: "A group of machines is describable if and only if it behaves nicely when you do four specific things to it." The paper proves that a similar map exists for these complex probability machines.

The four "behaviors" on the checklist are:

Disjoint Unions: If you take two separate machines and glue them side-by-side without them touching, the new combined machine should still belong to the group.
Generated Sub-processes: If you zoom in on a specific part of a machine that is self-contained (like looking at just one room in a house), that smaller part should also belong to the group.
Zigzag Morphisms (The "Shadow" Test): If Machine A can be perfectly "shadowed" by Machine B (meaning B mimics A's behavior so well that you can't tell the difference using our language), then if A is in the group, B must be too.
Ultrafilter Extensions (The "Infinite Mirror"): This is the trickiest part. The authors had to invent a new way to look at these machines through a "mathematical mirror" that handles infinite possibilities. If a machine passes the test in the mirror, the original machine must be in the group.

3. The Hurdle: Infinity and the "Broken" Compactness

The authors faced a major problem. In standard logic, if you have a list of rules that works for every small group of machines, it usually works for the whole infinite group. This is called "compactness."

However, in Probability Logic, this rule breaks. You can have a set of rules that works for any finite number of machines but fails when you try to apply it to an infinite collection. Because of this, the authors couldn't use the standard "Goldblatt-Thomason" map.

The Solution: They borrowed an idea from a previous study (Kozen et al.) involving Stone-Markov processes. They restricted their focus to machines where the "rules" can be generated by a countable list (like a list you could theoretically read through one by one). By doing this, they could build their "Ultrafilter Extension" (the infinite mirror) and successfully prove their version of the theorem.

4. The Real-World Example: Harsanyi Type Spaces

To show their theorem actually works, they applied it to a famous concept in economics called Harsanyi type spaces. These are models used to describe how people form beliefs about what others believe (like in game theory).

They showed that:

Harsanyi spaces fit the "Goldblatt-Thomason" checklist perfectly.
Therefore, you can write a specific set of probability sentences that describes exactly what a Harsanyi space is and nothing else.

5. The Finite Case: Small, Simple Machines

Finally, the authors looked at finite Markov processes (machines with a limited, countable number of states, like a simple board game). For these smaller machines, the "infinite mirror" isn't needed. Instead, they used a "local zigzag" test (checking if machines look the same up to a certain depth of steps). They proved a simpler version of the theorem for these finite systems.

Summary

In short, this paper builds a mathematical rulebook for identifying which groups of probability-based systems can be perfectly described by a specific logical language. They had to invent new tools to handle the "infinite" nature of probability, but once they did, they successfully mapped out the boundaries of what can and cannot be defined, even applying it to important economic models.

Technical Summary: Goldblatt-Thomason Theorem for Probability Logic

Problem Statement
The paper addresses the problem of characterizing classes of Markov processes that are definable in Probability Logic (PL). PL extends propositional logic with a family of modal operators $\{L_r \mid r \in \mathbb{Q} \cap [0,1]\}$ , where $L_r\phi$ asserts that the probability of $\phi$ is at least $r$ . While the classical Goldblatt-Thomason theorem successfully characterizes modally definable elementary classes of Kripke frames via closure properties (disjoint unions, generated subframes, bounded morphic images, and reflection of ultrafilter extensions), extending this result to PL faces significant obstacles. Specifically, PL lacks the compactness property, and satisfaction of probability formulas is not preserved under ultraproducts. Consequently, the standard model-theoretic approach relying on ultrapowers is inapplicable. The authors seek to establish a Goldblatt-Thomason-style characterization for PL by adapting the framework of general frames and ultrafilter extensions to the setting of Markov processes.

Methodology
The authors employ a model-theoretic approach grounded in the algebraic and topological properties of Markov processes. Their methodology involves several key steps:

Restriction to Countably Generated Processes: To handle the technical difficulties of defining ultrafilter extensions for general Markov processes, the authors restrict their primary analysis to processes where the $\sigma$ -algebra is countably generated. This allows them to utilize Stone duality and the concept of Stone-Markov processes, drawing on prior work by Kozen et al. [15].
Definition of Model-Theoretic Constructions: The paper rigorously defines and analyzes the following constructions for Markov processes:
- Markov Sub-processes and Event Sub-processes: Restricting the state space or the $\sigma$ -algebra while preserving the transition kernel.
- Disjoint Unions: Combining families of processes into a single process.
- Zigzag Morphisms: Surjective measurable functions preserving transition probabilities, serving as the probabilistic analogue to bounded morphisms.
- Ultrafilter Extensions: A novel construction $Ue(P)$ for countably generated processes. This involves defining a structure over the set of ultrafilters on the generating Boolean algebra. Crucially, the authors introduce nested Archimedean ultrafilters to ensure the extension preserves the logical properties of the original process.
Invariance and Reflection: The authors prove that validity in PL is preserved under sub-processes, disjoint unions, and zigzag morphisms. Conversely, they demonstrate that if a class is closed under ultrafilter extensions, validity in the extension reflects back to the original process.

Key Contributions and Results

The General Goldblatt-Thomason Theorem for PL:
The central result (Theorem 4.1) states that a class $\mathcal{C}$ of Markov processes is definable in Probability Logic if and only if:
- $\mathcal{C}$ has the Goldblatt-Thomason (GT) property: it is closed under Markov sub-processes, event sub-processes, disjoint unions, and zigzag morphisms.
- $\mathcal{C}$ is closed under ultrafilter extensions (specifically for countably generated processes).
- $\mathcal{C}$ reflects ultrafilter extensions (if $Ue(P) \in \mathcal{C}$ , then $P \in \mathcal{C}$ ).
The proof adapts the strategy of van Benthem's model-theoretic proof for Kripke frames, replacing ultraproducts with ultrafilter extensions to circumvent the failure of compactness in PL.
Application to Harsanyi Type Spaces:
As a primary application, the authors show that the class of Harsanyi type spaces (Markov processes where the transition probability of an event is 1 if the event contains all states with the same transition probabilities) is definable in PL. They provide a specific set of formulas $\Gamma_{har}$ that characterizes this class and prove that the class satisfies the GT-property and is closed under ultrafilter extensions.
Variant for Finite Markov Processes:
The paper extends the analysis to the class of finite Markov processes with rational transition probabilities ( $FMP_{\mathbb{Q}_0}$ ). For this subclass, the authors establish a variant of the theorem (Theorem 5.7) where the condition of ultrafilter extension is replaced by closure under local $n$ -zigzag morphisms. This result generalizes known theorems for finite Kripke frames to the probabilistic setting.

Significance and Claims
The authors claim that this work establishes the first Goldblatt-Thomason-style characterization for Probability Logic, bridging the gap between probabilistic modal logic and classical model-theoretic definability theory. By successfully adapting the ultrafilter extension technique to Markov processes, the paper provides a robust framework for determining which classes of probabilistic systems can be axiomatized.

The paper modestly notes that these results connect PL with classical definability theory but identifies the extension to infinitary probability logics (which allow countable conjunctions/disjunctions) as a necessary direction for future research. The authors suggest that such an extension is required to capture genuinely measure-theoretic properties like stationarity and recurrence, which are not expressible in the current finitary framework. The paper does not propose new experimental applications but frames its contribution as a foundational theoretical advance.