Arrow Logic
Origin. Venema and others developed arrow logic (1990s). Modal logic of transitions/arrows. Combines aspects of relation algebra and modal logic. Models binary relations as first-class objects. Foundation for dynamic semantics and process algebra connections.
Models. Logic where arrows are primitive. Standard modal: nodes with accessibility. Arrow logic: arrows/transitions are the objects. Arrows have source, target; can compose. Models processes, transitions, relations as entities.
Formalism.
Arrow frames: F = (W, C, R, I) where:
- W: set of arrows
- C ⊆ W³: composition (aCb,c = "a∘b = c")
- R ⊆ W²: reversal (aRb = "a is converse of b")
- I ⊆ W: identity arrows
Arrow models: M = (F, V) with V: Prop → P(W). Formulas describe sets of arrows.
Operators:
- φ ⊗ ψ: there exist composable arrows a,b with a∈⟦φ⟧, b∈⟦ψ⟧, a∘b = current
- φ⁻¹: converse arrows
- ι: identity arrows
- ◇φ: "diamond" relation to φ-arrows
Semantics: M, a ⊨ φ ⊗ ψ iff ∃b,c. C(b,c,a) ∧ M,b ⊨ φ ∧ M,c ⊨ ψ M, a ⊨ φ⁻¹ iff ∃b. R(a,b) ∧ M,b ⊨ φ M, a ⊨ ι iff I(a)
Connection to relation algebra: Arrow logic is modal correspondent of relation algebra. Composition, converse, identity match RA operations.
Symbols.
| Symbol | Unicode | Name | Meaning |
|---|---|---|---|
| ⊗ | U+2297 | Circled times | Composition: the arrow splits into two composable arrows |
| ⁻¹ | — | Converse | The arrow reversed |
| ι | U+03B9 | Greek small iota | Identity arrows |
| ◇ | U+25C7 | White diamond | Some related arrow |
| □ | U+25A1 | White square / Box | All related arrows |
| ∘ | U+2218 | Ring operator | Composition of arrows in the metalanguage |
| W | — | Universe | The set of arrows, taken as primitive objects |
| C | — | Composition | Ternary relation: C(b,c,a) means b∘c = a |
| R | — | Reversal | Binary relation pairing an arrow with its converse |
| I | — | Identity | The set of identity arrows |
| F, M | — | Frame, model | Arrow frame; arrow frame with a valuation |
| V | — | Valuation | Sends a proposition letter to a set of arrows |
| ⟦ ⟧ | U+27E6, U+27E7 | Mathematical white square brackets | The set of arrows satisfying a formula |
| ⊆ | U+2286 | Subset of or equal to | C ⊆ W³, R ⊆ W², I ⊆ W |
| ∈ | U+2208 | Element of | Arrow belonging to the extension of a formula |
| ∃ | U+2203 | There exists | Quantifier in the truth clauses |
| ∧ | U+2227 | Logical and | Conjunction in the truth clauses |
| ⊨ | U+22A8 | Satisfaction | Truth of a formula at an arrow |
| P(W) | — | Power set | Range of the valuation |
Metatheory. Basic arrow logic over arbitrary arrow frames is finitely axiomatizable and decidable, and the reason is visible in the truth clauses: the composition clause quantifies under the atom C(b,c,a) and the converse clause under R(a,b), each of which contains every free variable of its scope, so the standard translation lands in the guarded fragment of first-order logic, decidable by Andréka, van Benthem and Németi (1998) and EXPTIME-complete at bounded arity by Grädel (1999). The decidability depends on not requiring composition to be associative or arrows to be actual pairs. Over square frames, where arrows are pairs and composition is genuine relational composition, arrow validity coincides with the equational theory of representable relation algebras, which Tarski proved undecidable and Monk (1964) proved not finitely axiomatizable; that theory is in turn the three-variable fragment of first-order logic, which fixes arrow logic's expressive ceiling.
Applies to. Process algebra semantics. Automata theory. Dynamic logic foundations. Relation-based reasoning. Category theory connections. Transition systems.
Limitations. The decidability comes from weakening the models rather than from a better algorithm: arrow frames need not be frames of actual binary relations, C need not be associative, and composites need be neither unique nor existent, so the complete axiomatization is complete for a class containing frames that no relation algebra represents. Insisting on the intended pair frames restores associativity and immediately loses both decidability and finite axiomatizability, and there is no middle class known to be both natural and well behaved — this is the representation problem of relation algebra, inherited whole. Being a modal language, arrow logic is invariant under arrow bisimulation, so it cannot count arrows, express cardinality, or distinguish frames that bisimulation identifies. And what it does express is confined to the three-variable fragment, which excludes ordinary relational conditions whose statement needs four variables.
© 2026 Lingenic LLC