Least and greatest fixpoints in game semantics

We show how solutions to many recursive arena equations can be computed in a natural way by allowing loops in arenas. We then equip arenas with winning functions and total winning strategies. We present two natural winning conditions compatible with the loop construction which respectively provide initial algebras and terminal coalgebras for a large class of continuous functors. Finally, we introduce an intuitionistic sequent calculus, extended with syntactic constructions for least and greatest fixed points, and prove it has a sound and (in a certain weak sense) complete interpretation in our game model.

💡 Research Summary

The paper tackles a long‑standing gap in game semantics: the lack of a natural treatment for recursive type equations involving least (μ) and greatest (ν) fixed points. Traditional arena‑based models assume finite tree‑shaped interaction structures, which makes it difficult to represent self‑referential or infinitely unfolding types. The authors overcome this limitation by introducing loop arenas, a construction that allows an arena to contain a directed edge back to a previously visited sub‑arena. In effect, a potentially infinite unfolding of a type is captured by a finite graph equipped with a “loop” that can be traversed arbitrarily many times during a play.

On top of these loop arenas the authors define winning functions and total winning strategies. Two distinct winning conditions are presented. The first, called the minimum winning condition, requires the player to eventually reach a designated “safe” state even if the play continues forever. This condition mirrors the universal property of an initial algebra and yields a semantics for the least fixed point μ of any continuous functor. The second, the maximum winning condition, guarantees that the player can always force a win regardless of how long the opponent delays, embodying the universal property of a terminal coalgebra and thus providing a semantics for the greatest fixed point ν. Both conditions enforce that strategies are defined on every possible move (totality) and that they guarantee victory (winningness) under the respective condition.

A central technical contribution is the proof that, despite the presence of loops, total winning strategies exist for a wide class of continuous functors. The authors show that the space of strategies forms a complete lattice ordered by pointwise inclusion, and that the loop construction preserves continuity. Consequently, the μ‑operator yields the least fixed point in this lattice, while the ν‑operator yields the greatest fixed point, exactly as in the categorical theory of algebras and coalgebras.

Having established a robust semantic foundation, the paper proceeds to extend an intuitionistic sequent calculus (essentially a λ‑calculus with logical connectives) with syntactic constructs for μ and ν. The extended system, often denoted λμν, adds introduction and elimination rules for the fixed‑point types. Crucially, each rule corresponds directly to an operation on loop arenas and respects the chosen winning condition. In other words, a proof step in the sequent calculus is interpreted as a move in the associated game, and the construction of a proof corresponds to building a total winning strategy.

The authors then prove soundness: any sequent derivable in the λμν system translates into a total winning strategy in the game model. They also establish a form of weak completeness: if a total winning strategy exists for a given sequent, then there is a derivation in the calculus that captures it, albeit the completeness is “weak” because it relies on the continuity of the underlying functor and does not guarantee a one‑to‑one correspondence for arbitrary strategies. This result aligns with the categorical observation that initial algebras and terminal coalgebras are unique up to isomorphism but may not be syntactically unique in a proof system.

The paper concludes with a discussion of related work, emphasizing how previous game‑semantic approaches either avoided recursion altogether or handled it via ad‑hoc encodings that broke the clean correspondence between syntax and semantics. By contrast, the loop‑arena construction preserves the finitary nature of arenas while faithfully representing infinite unfolding, and the dual winning conditions give a symmetric treatment of μ and ν. The authors outline future directions, including extending the framework to higher‑order functors, non‑continuous type operators, and integrating the model into concrete programming languages with recursive data types and co‑inductive specifications.

Overall, the work provides a mathematically elegant and technically solid bridge between recursive type theory, category‑theoretic fixed‑point constructions, and interactive game semantics, opening the door to more expressive semantic models for languages that combine inductive and co‑inductive features.