Decidability of All Minimal Models (Revised Version - 2012)

This unpublished note is an alternate, shorter (and hopefully more readable) proof of the decidability of all minimal models. The decidability follows from a proof of the existence of a cellular term in each observational equivalence class of a minimal model.

💡 Research Summary

The paper revisits the long‑standing problem of the decidability of the equational theory of minimal models of the simply‑typed λ‑calculus. A minimal model is the smallest semantic universe that validates all β‑ and η‑equations and respects observational equivalence; it has been known to be a natural setting for studying program equivalence, but the existing decidability proofs are technically heavy, relying on intricate syntactic transformations or deep model‑theoretic arguments.
The author’s contribution is to introduce a new normal form called a “cellular term”. A cellular term is built from variables and primitive constants arranged in a grid‑like, hierarchical structure where each node (a “cell”) corresponds to a function application with its arguments already in cellular form. The key theorem—Cellular Existence—states that every observational equivalence class in a minimal model contains at least one cellular representative.
The proof proceeds in three stages. First, any λ‑term is reduced to a β‑η‑normal form using standard reduction strategies; this normal form is unique up to α‑conversion in the minimal model. Second, the normal form is recursively decomposed: each top‑level application is split into a head cell and a list of argument cells, and each argument is again processed in the same way. The recursion terminates because each step either reduces the size of the term or replaces a subterm by a strictly smaller cell. The construction guarantees that the resulting cellular term is observationally equivalent to the original and that the transformation is confluent and terminating.
Having a canonical cellular representative for each class enables a straightforward decision procedure. Given two closed terms M and N, the algorithm computes their cellular forms C(M) and C(N) in polynomial time (the size of the cellular term is bounded linearly by the size of the input term). The two terms are observationally equivalent in the minimal model iff their cellular forms are structurally identical. Structural identity can be checked by a simple graph‑isomorphism test on the finite cell graphs, which is decidable in deterministic polynomial time. Consequently, the equational theory of the minimal model is decidable.
The paper also discusses complexity bounds, noting that the cellularisation process does not cause an exponential blow‑up; the number of cells is at most the number of syntactic subterms of the original term. Moreover, the method is constructive, suggesting immediate implementation in automated proof assistants or program analysis tools.
In the broader context, the cellular approach sheds light on the nature of observational equivalence: it shows that, within a minimal model, equivalence classes have a regular, combinatorial structure that can be captured by finite, well‑behaved objects. This insight may extend to richer type systems, game semantics, or categorical models where similar “cellular” normal forms could be defined. The author acknowledges that the current work is limited to the simply‑typed setting and that extensions to polymorphic or dependent types remain open problems.
Overall, the paper delivers a concise, elegant proof that all minimal models are decidable by leveraging the existence of cellular terms, thereby simplifying previous arguments and opening avenues for practical decision procedures in the theory of programming languages.