Semantic Characterization Theorem (SCT)

• Semantic Characterization Theorem (SCT) is a framework that links semantic invariants with the definability of structural properties across modal, automata, and model-theoretic logics.
• It employs methodologies such as bisimulation invariance, spectral analysis, and preservation theorems to translate complex continuous and discrete behaviors into symbolic representations.
• The framework informs advancements in logic-based verification and interpretable AI by enabling precise classification and expressiveness comparisons in both theoretical and applied settings.

The Semantic Characterization Theorem (SCT) encapsulates a variety of deep results that establish the correspondence between semantic invariants and the definability or recognition of structural properties within particular logical or computational frameworks. The SCT label is attached to distinct but thematically related theorems in modal logic, automata theory, model theory, and, more recently, the analysis of dynamical systems and LLMs. The following synthesizes the principal SCT variants, focusing on bisimulation-invariant fragments of transitive closure logics and chain logics (Carreiro, 2015), spectral semantic collapse in continuous state machines (Wyss, 4 Dec 2025), and parametric preservation theorems in first-order logic (Sankaran, 2016).

1. Semantic Characterization for Modal and Fixpoint Logics

The archetypal SCT for modal fixpoint logics states that the expressive power of Propositional Dynamic Logic (PDL) is precisely the bisimulation-invariant fragment of first-order logic with unary transitive closure ($\mathrm{FO(TC^1)}$) and of weak chain logic (WCL). Here, bisimulation is the structural semantic equivalence underlying modal logics. A formula is bisimulation-invariant if it remains true under bisimilar transitions systems, characterizing properties that are fundamentally modal rather than relational.

Formal Statement

$$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$

$$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$

(Carreiro, 2015)

This theorem subsumes van Benthem’s modal characterization theorem for basic modal logic, extending it to the presence of explicit fixpoint constructs.

2. Key Definitions and Logical Systems

To precisely state and prove SCTs, several formal systems and invariance notions are required:

• Propositional Dynamic Logic (PDL): Extends basic modal logic with regular program composition. Its semantics interprets modalities over the execution paths in a Kripke structure.
• First-Order Logic with Unary Transitive Closure (FO(TC¹)): Enriches FO with a unary transitive-closure operator, allowing the definition of reachability-like properties.
• Weak Chain Logic (WCL): A monadic second-order extension that quantifies over finite chains (generalized paths).
• Bisimulation and Bisimulation-Invariance: Equivalence relation and invariance criterion that preserve the modal structure.

These systems intertranslate via standard relational and chain representations, and their correspondence on trees is established through additive-weak parity automata (Carreiro, 2015).

3. Structural Proof Outline and Technical Lemmas

Two-Direction Characterization

• PDL ⊆ FO(TC¹), WCL, and Bisimulation-Invariant: PDL formulas admit direct translation to FO(TC¹) using transitive closure (for $\alpha*$ program iteration) and to WCL via quantification over chains.
• From Bisimulation-Invariant FO(TC¹)/WCL to PDL: Utilizes tree unravelling (which preserves bisimulation equivalence) and establishes correspondences through weak parity automata. Central technical facts include the Simulation Theorem and the construction of additive-weak automata (Janin–Walukiewicz paradigm).

Key Formulas and Constructions (in FO and μ-Calculus)

• Embedding Transitive Closure in LFP:

$$[TC_{x,y}\;\psi(x,y)](u,v) \equiv [LFP_{p:y}(y=u \lor \exists x(p(x) \land \psi(x,y)))](v)$$

• Automata ↔ μ-Calculus/PDL Formulas: Translation between the automaton’s execution structure (directed tree plus back-edges) and modal or fixpoint formulas via recursive definitions.

The equivalence on trees extends to all models by interpreting automata on unravellings and simulating automaton acceptance by equivalent PDL programs (Carreiro, 2015).

4. SCT for Continuous State Machines and Spectral Semantics

A modern SCT variant appears in the semantic theory of continuous state machines (CSMs), such as neural or LLMs. Here, the focus shifts from syntactic modal invariance to the emergence of discrete semantic ontologies from the topological and spectral properties of dynamical systems.

Key Theorem

Let $C=(M,U,T,s_0,O,\Delta)$ be a CSM over a compact state manifold $M$, with suitable regularity, ergodicity, and a compact transfer operator $P$ with discrete spectrum. Then:

1. Spectral Lumpability: The leading eigenfunctions $\phi_1, \dots, \phi_r$ induce a partition of $M$ into finitely many “spectral basins” $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$0.
2. Logical Tameness: If the transition map $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$1 is definable in an o-minimal structure, the basins $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$2 are themselves definable.
3. Discretization of Semantics: The system’s long-term dynamics collapse to a finite, logically interpretable ontology—explaining the emergence of discrete symbolic behavior from continuous latent manifolds.

This version of the SCT connects transfer operator spectral theory, o-minimality, and logical definability (Wyss, 4 Dec 2025).

5. SCT in First-Order Model Theory and GLT(k)

The “Semantic Characterization Theorem” moniker also encompasses the generalized Łoś–Tarski theorem $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$3 for first-order prefix classes. For each $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$4, the preservation of sentences under substructures modulo $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$5-cruxes ($$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$6) and $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$7-ary covered extensions ($$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$8) yields semantic characterizations of the $$\text{A formula } \varphi \text{ is equivalent to a PDL formula if and only if}$$9 and $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$0 classes, respectively.

Core Result (Sankaran, 2016)

Let $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$1 be a first-order sentence.

• $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$2 is $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$3 modulo background theory $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$4 iff $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$5 for some $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$6 sentence $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$7.
• Dually, $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$8 is $$\begin{cases} \varphi \text{ is expressible in } \mathrm{FO(TC^1)} \text{ (or WCL), and}\ \varphi \text{ is invariant under bisimulation.} \end{cases}$$9 modulo $\alpha*$0 iff $\alpha*$1 for some $\alpha*$2 sentence $\alpha*$3.

This generalizes the Łoś–Tarski preservation theorem (case $\alpha*$4) to the quantifier alternation hierarchy. The equivalent bounded substructure property (EBSP) provides the semantic mechanism for finitary analogues of downward Löwenheim–Skolem in finite model theory, ensuring the effectiveness of $\alpha*$5 on well-behaved classes (e.g., words, trees, cographs, graphs of bounded tree/shrub depth).

6. Unifying Schematic Table

SCT Variant	Characterized Invariant	Target Logical Fragment
Modal/fixpoint (PDL, FO(TC¹), WCL) SCT	Bisimulation-invariance	PDL-definable ($\alpha*$6 FO(TC¹), WCL)
CSM Spectral SCT	Spectral lumpability, o-minimal	Finite, definable spectral basins (ontology cells)
GLT(k) in FO	$\alpha*$7-crux/covered extensions	$\alpha*$8 (resp. $\alpha*$9)

Each SCT provides an explicit bridge: from semantic invariance properties (bisimulation, spectral collapse, combinatorial preservation) to concrete definability or syntactic classification.

7. Implications and Contemporary Extensions

The SCT framework has direct impact in the classification of modal and temporal logics, automata-theoretic characterizations, and the symbolic analysis of machine learning models. The general scheme, linking invariance under a semantic relation with definability in a restricted logic, is a recurrent motif across logic and theoretical computer science.

• In the presence of fixpoints and automata, SCTs enable precise expressiveness comparisons, crucial for verification, synthesis, and descriptive complexity.
• For CSMs and neural models, SCT-like results clarify the emergence of symbolic interpretability from continuous latent state spaces, providing structural explanations for the apparent discretization of semantic behavior (Wyss, 4 Dec 2025).
• In first-order finite model theory, EBSP and GLT(k) introduce novel, computable preservation properties for key structural classes, enabling effective characterizations beyond the infinite case (Sankaran, 2016).

A plausible implication is that SCT-like theorems can inform the development of both logic-based verification and interpretable AI, identifying semantic invariants that control the expressive jump from continuous or combinatorial systems to symbolic representations.