Intersection Types and Counting

We present a new approach to the following meta-problem: given a quantitative property of trees, design a type system such that the desired property for the tree generated by an infinitary ground $\lambda$-term corresponds to some property of a derivation of a type for this $\lambda$-term, in this type system. Our approach is presented in the particular case of the language finiteness problem for nondeterministic higher-order recursion schemes (HORSes): given a nondeterministic HORS, decide whether the set of all finite trees generated by this HORS is finite. We give a type system such that the HORS can generate a tree of an arbitrarily large finite size if and only if in the type system we can obtain derivations that are arbitrarily large, in an appropriate sense; the latter condition can be easily decided.

💡 Research Summary

The paper addresses the language‑finiteness problem for nondeterministic higher‑order recursion schemes (HORS) by translating it into a type‑theoretic question about λY‑terms. A λY‑term P of sort o generates an infinite Böhm tree BT(P); the set L(P) consists of all finite trees obtained by resolving each “br” (branch) node in BT(P). The problem asks whether L(P) is finite. Existing approaches either fail for nondeterministic systems or rely on a complex reduction to the diagonal problem, which involves repeatedly lowering the order of the scheme.

The authors introduce a novel intersection‑type system enriched with two kinds of annotations: flags and markers, each carrying an integer “order”. For a term of complexity m, at most one marker of each order 0…m‑1 may appear in a derivation, while arbitrarily many flags of each order 0…m are allowed. Flags mark “productive” occurrences that contribute to the size of a finite tree, whereas a marker designates a distinguished leaf whose path is used to count flags of the next higher order. By arranging markers and flags hierarchically (a marker of order n is placed on a leaf, flags of order n+1 are placed on the path to that leaf, and so on), the number of highest‑order flags in a typing derivation approximates the size of some finite tree in L(P).

Formally, for each simple type α the system defines a set Tα of ordinary types and a set Fαk of full types (k is the maximal flag order). A full type ˆτ = (k, F, M, τ) consists of its order k, a set F of flag orders, a set M of marker orders, and a base type τ. A typing judgment has the shape Γ ⊢ P : ˆτ ⊲ c, where Γ maps variables to collections of full types, and c ∈ ℕ counts the number of order‑k flags present in the derivation. The usual intersection‑type rules for variables, application, abstraction, and the Y‑operator are extended with side‑conditions that propagate and combine flag/marker sets appropriately; in particular, when an argument supplies a flag of order n, the surrounding term may raise it to order n+1, and markers suppress the propagation of the corresponding lower‑order flags.

The central theorem states: L(P) contains arbitrarily large finite trees iff there exists a typing derivation for P in which the flag counter c can be made arbitrarily large. The proof proceeds by (i) showing that any finite tree t ∈ L(P) can be encoded by a derivation whose flags exactly count the nodes of t, using the hierarchical marker/flag scheme, and (ii) conversely, that a derivation with many high‑order flags yields a tree whose size is at least that number. Because the flag counter is a natural number, the existence of an unbounded derivation reduces to checking whether the set of reachable counter values is bounded, which is a decidable property of a finite‑state abstraction of the typing system.

Algorithmically, one computes the complexity m of the input term, constructs the finite set of possible full types (which is bounded by m), and explores the typing rules as a transition system. The system records, for each reachable typing judgment, the maximal flag counter observed. Since the state space is finite, standard reachability analysis decides whether the counter is bounded. Consequently, the language‑finiteness problem for arbitrary nondeterministic HORS is decidable.

Compared with prior work, the paper improves on the diagonal‑problem reduction