Game trees are a fundamental mathematical abstraction for analyzing games, often implemented as rose trees in functional programming. We can construct rose trees by starting from an initial state and iteratively applying the function that provides the states one move away, an approach known as an anamorphism (or colloquially, an unfold).

In ML or Haskell, finite and infinite rose trees share the same type, but proof assistants typically distinguish them: finite trees as inductive, infinite ones as coinductive. With the inductive approach, defining the unfold function is tricky but we obtain a finite tree that is easy to compute with and easy to reason about. With the coinductive approach, defining the unfold function is easy but we obtain a potentially infinite tree, which is harder to reason about since we must resort to proof techniques like bisimulation. In this paper, we compare the inductive and coinductive approaches to game trees, implement unfold functions for both, and prove the soundness and completeness of both unfold functions (with respect to the game) in the Rocq Prover. We illustrate these techniques with implementations of a tic-tac-toe game and a simple SAT solver.

I am a researcher at Bloomberg’s CTO Infrastructure and Security Research team in New York City, applying formal methods in industry.