Bisimulation-Invariant Queries

Updated 3 February 2026

Bisimulation-Invariant Queries are logical queries preserved under bisimulation, establishing expressive equivalences between modal, first-order, and fixpoint logics.
They utilize game-based techniques, algebraic methods, and automata theory to characterize query expressive power and computational complexity.
Their applications span model theory, finite model theory, database query languages, and coalgebra, highlighting both theoretical and practical significance.

A bisimulation-invariant query is a logical property, typically formulated as a class of pointed structures or tuples in a relational, modal, or coalgebraic setting, that is preserved under bisimulation equivalence—a structural relation matching elements or states of two systems such that corresponding modal, relational, or semantic features cannot be distinguished by the logic in question. Bisimulation-invariance is a critical property for characterizing the expressive boundaries of logical systems, especially modal and fixpoint logics, and has been central to model theory, finite model theory, descriptive complexity, coalgebra, and database theory.

1. Core Definition and Foundational Theorems

Let $Q$ be a class of structures with designated elements (e.g., worlds, nodes, tuples). The query $Q$ is said to be bisimulation-invariant if, whenever two pointed structures $(\mathcal{M}, w)$ and $(\mathcal{M}', w')$ are bisimilar (i.e., related by a bisimulation), $(\mathcal{M}, w) \in Q$ if and only if $(\mathcal{M}', w') \in Q$ .

This notion generalizes across logical frameworks:

Kripke models and modal logic: Bisimulation-invariance captures precisely the expressive power of basic modal logic within first-order logic, as established by the van Benthem theorem (Otto, 2019).
Monadic second-order logic (MSO) and the modal $\mu$ -calculus: For both infinite and, as resolved recently, finite transition systems, the bisimulation-invariant fragment of MSO coincides with the expressive power of modal $\mu$ -calculus (Janin–Walukiewicz theorem, and its finite version) (Colcombet et al., 2024, Blumensath et al., 2019).
Database query languages: For fragments of the calculus of relations, bisimulation-invariant queries are characterized in terms of tailored notions of (finite-round) bisimulation or simulation; this organizes query expressiveness in relation to navigational equivalence (Fletcher et al., 2012).
Coalgebraic logic: For coalgebras of set-functors, the bisimulation-invariant fragment of second-order logics (MSO-style) is captured by appropriately parameterized coalgebraic $\mu$ -calculi (Enqvist et al., 2015).

The canonical characterization is:

Setting	Definability Theorem	Reference
Modal logic (ML)	$\mathrm{ML} = \mathrm{FO}/\sim$	(Otto, 2019)
Modal $\mu$ -calculus	$L_\mu = \mathrm{MSO}/\sim$	(Colcombet et al., 2024)
Inquisitive Modal Logic	$\mathrm{InqML} = \mathrm{FO}/\sim$ (two-sorted)	(Ciardelli et al., 2017)
Polyadic $\mu$ -calculus	Bisim-Inv(P) $= \bigcup_{d} L_\mu^d$ (Ptime)	(Bruse et al., 2022)
Coalgebraic $\mu$ -calculus	$\mu \mathrm{ML}_\Lambda = \mathrm{MSO}_T/\simeq$	(Enqvist et al., 2015)

Here, the slash notation $\mathcal{L}/\sim$ denotes the set of queries expressible in $\mathcal{L}$ that are invariant under bisimulation.

The van Benthem–Rosen characterization states that the fragment of first-order logic (FO) preserved under (standard) bisimulation corresponds exactly to modal logic ML; that is, every bisimulation-invariant FO-formula is equivalent to some modal formula (Otto, 2019).
In modal fixpoint logics, every $\mu$ -calculus formula is bisimulation-invariant. Conversely, over (in)finite transition systems, every bisimulation-invariant MSO sentence is equivalent to a $\mu$ -calculus formula; this is the Janin–Walukiewicz theorem (Colcombet et al., 2024).

Graded modal logic (CML), with modalities expressing cardinality thresholds, is characterized by invariance under counting bisimulation—a strengthening of standard bisimulation requiring matchings of successor sets of given sizes. The expressive completeness (graded van Benthem–Rosen theorem) is that every FO-formula invariant under counting bisimulation is equivalent to a graded modal formula (arbitrary and finite model versions) (Otto, 2019).

Inquisitive modal logic (InqML), extending ML with inquisitive disjunction and a dual modality to capture "questions," fits within the van Benthem paradigm at a higher type level, via two-sorted FO. InqML is the fragment of FO (over suitable two-sorted relational structures) invariant under inquisitive bisimulation, a notion intertwining worlds and sets of worlds ("information states") (Ciardelli et al., 2017).

3. Methods for Characterizing Bisimulation-Invariance

The general paradigm for characterizing bisimulation-invariant queries leverages two key methods:

A. Ehrenfeucht–Fraïssé Games and Characteristic Formulas

Bisimulation relations (often of bounded depth) correspond to winning strategies in a back-and-forth game, and modal logics admit "characteristic formulas" $\chi^n_{\mathcal{M}, w}$ that define the $n$ -bisimulation type. These provide a finite approximation to invariance at the quantifier rank or modal depth in question (Ciardelli et al., 2017, Otto, 2019).
Upgrading arguments via locality and stratification (often using Gaifman locality for FO) allow reduction of arbitrary FO-definability (under invariance) to bounded bisimulation types, facilitating equivalence with modal or fixpoint logics (Ciardelli et al., 2017).

B. Algebraic Approaches and Tree Algebras

Recent work resolves the finite-model characterization for bisimulation-invariant MSO using tree algebras: the syntactic algebra of MSO types for bisimulation-invariant queries is "rankwise-finite," and automaton-theoretic properties allow reduction to parity automata, which can be re-expressed in the modal $\mu$ -calculus (Colcombet et al., 2024).
Coalgebraic generalizations rely on automata translation for MSO, the existence of adequate uniform constructions (i.e., tree-unravellings compatible with the functorial structure), and the coalgebraic $\mu$ -calculus for expressive completeness (Enqvist et al., 2015).

4. Implications for Descriptive Complexity and Database Theory

In descriptive complexity, Otto’s theorem asserts that the bisimulation-invariant Ptime queries on graphs coincide with those definable in the polyadic $\mu$ -calculus $L_\mu^d$ ; this abstracts away the order-problem blocking a Ptime-capturing logic for unordered graphs (Bruse et al., 2022, Bruse et al., 27 Jan 2026).
Higher-level results characterize bisimulation-invariant fragments of EXPTIME and $k$ -EXPTIME as corresponding to higher-order modal fixpoint logics with higher-order fixpoints (PHFL), matching logical complexity with computational complexity under invariance constraints (Bruse et al., 2022).
For fragments of the calculus of binary relations, which underpin many database query languages, each fragment's bisimulation-invariant queries correspond to expressions definable in the fragment, with polynomial-time algorithms for indistinguishability checks (Fletcher et al., 2012).

5. Structure and Limitations of Bisimulation-Invariant Fragments

In each logical system, the bisimulation-invariant fragment can be strictly less expressive than the full parent logic. For example, InqML is more expressive than ML but less expressive than full MSO on Kripke frames (Ciardelli et al., 2017).
Over "full" model classes—e.g., where all subsets (powerset) are included as possible information states—FO collapses to MSO on the primary sort, and bisimulation-invariance can extend beyond the modal fragment, breaking characterization theorems unless the class is restricted (e.g., to "locally full" structures) (Ciardelli et al., 2017).
In coalgebraic modal logic, an absence of an adequate uniform construction for certain functors (e.g., monotone neighborhood functor) forces a further restriction (to global bisimulation invariance) to retain characterization (Enqvist et al., 2015).

6. Examples and Applications

MSO Example: The MSO property "there is a strongly $p$ -labeled reachable subgraph closed under successors" is a canonical bisimulation-invariant query, translatable into the $\mu$ -calculus as $\mu X.(p \lor \Box X)$ (Colcombet et al., 2024).
Database Example: For fragments of the calculus of relations, two marked structures are indistinguishable under all queries expressible in the fragment if and only if they are related by the appropriate fragment-specific (bi)simulation (Fletcher et al., 2012).
Description Logics: In DLs such as $\mathcal{ALC}_{reg}(I,O,Q,U,Self)$ , bisimulation-invariant queries include instance checking, subsumption, role-assertions, and role-inclusion, exactly matching the expressive power of the DL (Divroodi et al., 2011).
Coalgebraic Example: For the bag functor (graded modal logic), every MSO property invariant under behavioral equivalence is definable in the graded $\mu$ -calculus (Enqvist et al., 2015).
Complexity Separation: Non-regularity of certain bisimulation-invariant tree languages (relative to "power graphs") would imply $P\neq NP$ or $P\neq PSPACE$ (Bruse et al., 27 Jan 2026).

7. Open Questions and Future Directions

Several open problems remain:

Extension to rich model classes: Extending van Benthem/Janin–Walukiewicz-type theorems to increasingly expressive or structurally complex environments (e.g., data-trees, hypergraphs, higher-arity relations) (Colcombet et al., 2024).
Global bisimulation for neighborhood frames: Achieving expressive completeness for monotone neighborhood models requires global bisimulations and additional global modalities; further categorical/algebraic frameworks may be needed (Enqvist et al., 2015).
Separation via bisimulation-invariance: Bisimulation-invariance provides a setting for separating complexity classes free from order encoding, yet explicit demonstrations of relative non-regularity for associated tree languages remain elusive (Bruse et al., 27 Jan 2026).
Algebraic and automata-theoretic approaches: Pursuit of purely algebraic or automata-theoretic proofs of classical results such as Rabin’s theorem, within the bisimulation-invariant paradigm (Colcombet et al., 2024).

Bisimulation-invariant queries thus constitute a central organizing concept at the intersection of modal and fixpoint logic, model theory, finite model theory, coalgebra, database theory, and descriptive complexity, enabling fine-grained analysis of expressive power, decidability, and computational complexity across a wide spectrum of logical systems.