The Variable Hierarchy for the Games mu-Calculus

Parity games are combinatorial representations of closed Boolean mu-terms. By adding to them draw positions, they have been organized by Arnold and one of the authors into a mu-calculus. As done by Berwanger et al. for the propositional modal mu-calculus, it is possible to classify parity games into levels of a hierarchy according to the number of fixed-point variables. We ask whether this hierarchy collapses w.r.t. the standard interpretation of the games mu-calculus into the class of all complete lattices. We answer this question negatively by providing, for each n >= 1, a parity game Gn with these properties: it unravels to a mu-term built up with n fixed-point variables, it is semantically equivalent to no game with strictly less than n-2 fixed-point variables.

💡 Research Summary

The paper investigates the variable hierarchy of the games µ‑calculus, a logical framework that interprets parity games with draw positions over the class of all complete lattices. Building on earlier work that established the strictness of the variable hierarchy for the propositional modal µ‑calculus, the authors ask whether a similar hierarchy collapses when the games µ‑calculus is interpreted in the general setting of complete lattices. Their answer is negative: the hierarchy does not collapse, and in fact it is infinitely strict.

To prove this, the authors construct, for every natural number n ≥ 1, a specific parity game Gₙ that satisfies two crucial properties. First, Gₙ can be unraveled into a tree with back‑edges whose feedback (the maximal number of back‑edges encountered along any root‑to‑leaf path) is exactly n. This shows that Gₙ belongs to the n‑th level Lₙ of the hierarchy, because the feedback of such a tree corresponds precisely to the minimum number of fixed‑point variables needed to express the associated µ‑term, up to α‑conversion. Second, Gₙ is semantically equivalent (i.e., yields the same monotone operator on every complete lattice) to no game whose feedback is less than n − 2. Consequently, Gₙ cannot be expressed with fewer than n − 2 fixed‑point variables, establishing that the inclusion Lₙ₋₃ ⊂ Lₙ is strict for all n ≥ 3.

The technical heart of the paper lies in two graph‑theoretic notions: entanglement and feedback. Entanglement E(G) of a directed graph G is defined as the minimal feedback among all finite unfoldings of G into a tree with back‑edges. Feedback, in turn, counts for each vertex the number of “returns” (ancestors that are targets of back‑edges whose sources lie in the vertex’s subtree); the feedback of the whole tree is the maximum of these counts. These measures capture the combinatorial complexity of the fixed‑point structure of a µ‑term.

A novel contribution is the introduction of “⋆‑weak simulation,” a relation between two digraphs G and H that allows each edge of G to be simulated by a finite non‑empty path in H, preserving the intersection pattern of edges. The authors prove a general theorem: if G ⋆‑weakly simulates H, then E(G) − 2 ≤ E(H). This result holds for arbitrary digraphs, not only for games, and provides a powerful tool for lower‑bounding the entanglement of any game that is semantically equivalent to a given one.

Using this tool, the authors define “strongly synchronizing games,” a class of parity games whose structure forces any semantically equivalent game to be related by a ⋆‑weak simulation. They then construct, for each n, a strongly synchronizing game Gₙ whose entanglement is exactly n. By the ⋆‑weak simulation theorem, any game H equivalent to Gₙ must have entanglement at least n − 2, which translates into a lower bound of n − 2 on the number of fixed‑point variables needed to express H. This yields the desired separation between the hierarchy levels.

The paper also discusses the role of draw positions (free variables). Although the number of draw positions grows with n in the constructed games, the authors note that a bounded number of free variables (e.g., three) suffices to simulate arbitrarily many generators in a free lattice, referencing known results on free lattices. Hence the hierarchy’s strictness does not rely on an unbounded supply of free variables.

Finally, the authors compare the variable hierarchy with the alternation‑depth hierarchy. While the latter’s infiniteness automatically implies strictness of consecutive levels, the variable hierarchy’s strictness is a subtler property that the paper establishes via combinatorial and proof‑theoretic techniques (cut‑elimination, η‑expansion). Open problems remain, such as whether the inclusion Lₙ₋₁ ⊂ Lₙ is also strict and whether the hierarchy can be separated using a fixed number of free variables.

In summary, the paper provides a thorough combinatorial analysis, introduces the ⋆‑weak simulation technique, and constructs a family of parity games that demonstrate the infinite, non‑collapsing nature of the variable hierarchy for the games µ‑calculus over complete lattices. This advances our understanding of the expressive power of fixed‑point logics beyond the modal setting and opens avenues for further exploration of hierarchy separations in related logical systems.