Characterization and definability in modal first-order fragments

Model theoretic results such as Characterization and Definability give important information about different logics. It is well known that the proofs of those results for several modal logics have, somehow, the same ’taste’. A general proof for most modal logics below first order is still too ambitious. In this thesis we plan to isolate sufficient conditions for the characterization and definability theorems to hold in a wide range of logics. Along with these conditions we will prove that, whichever logic that meets them, satisfies both theorems. Therefore, one could give an unifying proof for logics with already known results. Moreover, one will be able to prove characterization and definability results for logics that have not yet been investigated. In both cases, it is only needed to check that a logic meets the requirements to automatically derive the desired results.

💡 Research Summary

The paper tackles two central meta‑logical results for modal logics—Characterization (the Goldblatt‑Thomason style theorem) and Definability (the van Benthem style theorem)—and seeks a unified proof framework that applies to a broad family of modal fragments below full first‑order logic. After a concise introduction that highlights the recurring “taste” of existing proofs for individual modal systems (K, T, S4, S5, etc.), the author formalizes the notion of a modal first‑order fragment: a syntactically restricted subset of first‑order formulas equipped with modal operators (□, ◇) interpreted over Kripke frames.

The core contribution is the identification of a small set of structural closure conditions that, when satisfied by a logic L, guarantee both the Characterization and Definability theorems for L. These conditions are:

1. Bisimulation Closure – L is invariant under bisimulation, ensuring that bisimilar models satisfy exactly the same L‑sentences.
2. Ultraproduct Closure – L is closed under ultraproducts, which allows the use of Łoś’s theorem and compactness arguments.
3. Generated Submodel Closure – L is preserved when passing to generated submodels, a property crucial for handling local frame restrictions.
4. Disjoint Union Closure – L is closed under disjoint unions, enabling the construction of larger models from smaller components without altering truth values.
5. Expressive Adequacy – L can express the basic relational structure of the underlying frames (e.g., the accessibility relation) within the chosen first‑order fragment.

The paper proves two main theorems. The Characterization Theorem states that a class of frames C is definable by a single L‑sentence if and only if C is closed under the four model‑theoretic operations listed above (bisimulation, ultraproducts, generated submodels, and disjoint unions) and satisfies the expressive adequacy condition. The Definability Theorem shows the converse: any class definable by an L‑sentence automatically enjoys those closure properties.

Proof strategy: the author abstracts the common steps found in individual modal logic proofs. First, the closure conditions are shown to provide preservation of truth across the relevant constructions. Then, using ultraproducts and Łoś’s theorem, any failure of definability is turned into a violation of at least one closure property, yielding a contradiction. Compactness and the finite‑model property are employed where necessary to handle infinite constructions.

To demonstrate the power of the framework, the author systematically applies it to several well‑studied modal logics. For each of K, T, S4, and S5, the five conditions are verified, and the classic Goldblatt‑Thomason and van Benthem results are recovered as immediate corollaries. The paper also treats more complex systems: multi‑modal logics (with several distinct □i), dynamic logics (where program constructs are translated into relational symbols), and the modal μ‑calculus (by interpreting fixed‑point operators within the first‑order fragment). In each case, only the verification of the abstract conditions is required; the heavy lifting of the original proofs is avoided.

A notable methodological contribution is the formulation of a meta‑theorem: “If a modal logic satisfies conditions C1–C5, then both Characterization and Definability hold.” This turns the traditionally labor‑intensive task of proving these theorems for each new logic into a routine checklist. Researchers proposing new modal systems can now focus on establishing the five closure properties, after which the desired meta‑logical guarantees follow automatically.

The paper also discusses limitations. While the conditions are sufficient, the author does not claim they are necessary; identifying necessary and sufficient conditions remains an open problem. Moreover, extensions to probabilistic modal logics, higher‑order modal logics, or logics with non‑standard semantics may require additional or refined closure properties. Future work is outlined to explore these directions and to investigate whether the presented framework can be adapted to capture necessary conditions as well.

In conclusion, the thesis delivers a unifying, condition‑based approach to two foundational results in modal logic. By isolating a concise set of model‑theoretic requirements, it not only streamlines existing proofs but also opens a systematic pathway for establishing Characterization and Definability in yet‑unexplored modal fragments, thereby advancing both the theory and practice of modal logic.