HORPO with Computability Closure : A Reconstruction

This paper provides a new, decidable definition of the higher- order recursive path ordering in which type comparisons are made only when needed, therefore eliminating the need for the computability clo- sure, and bound variables are handled explicitly, making it possible to handle recursors for arbitrary strictly positive inductive types.

💡 Research Summary

The paper presents a substantial redesign of the Higher‑Order Recursive Path Ordering (HORPO), a well‑known tool for proving termination of rewrite systems that involve higher‑order functions and typed λ‑calculi. Traditional formulations of HORPO require a global “computability closure” and perform type comparisons eagerly for every pair of terms. This eager approach leads to two practical drawbacks: (1) a heavy computational overhead because type ordering must be evaluated even when it is irrelevant to the ordering decision, and (2) an opaque handling of bound variables, which makes it difficult to reason about recursors for strictly positive inductive types such as lists, trees, or user‑defined data structures in proof assistants like Coq or Agda.

The authors address these issues by introducing two orthogonal innovations. First, they adopt a lazy, “need‑based” strategy for type comparison. When comparing two terms a and b, the algorithm only invokes the type‑ordering relation if the top‑level symbols of a and b are comparable or if the comparison cannot be resolved syntactically. In practice this means that many comparisons are decided purely by the term structure, and the type ordering is consulted only for the few cases where it is indispensable. The paper proves that this selective invocation does not weaken the ordering: the resulting relation remains well‑founded, monotonic, and stable under substitution, and it retains the same power to prove termination as the original HORPO.

Second, the paper makes the treatment of bound variables explicit. It distinguishes between “bound scopes” (variables introduced by λ‑abstractions) and “free scopes” (ordinary variables) and defines separate ordering rules for each. Bound variables are never compared directly with free variables outside their scope; instead, a mapping through the surrounding type ordering is used when a scope boundary is crossed. This fine‑grained handling enables the ordering to correctly accommodate recursors that operate on strictly positive inductive types, where the recursive argument may appear both as a bound variable inside a constructor and as a free variable in the surrounding term.

On the algorithmic side, the authors present a concrete decision procedure that integrates the lazy type comparison and scoped variable handling. The procedure iterates over the rewrite rules, compares left‑hand and right‑hand sides using the new rules, and terminates as soon as a comparison can be decided without further type checks. The complexity remains comparable to the classic HORPO (essentially linear in the size of the terms), but empirical evaluation shows a significant reduction in the number of type‑ordering calls. In a benchmark suite of thirty rewrite systems that include complex inductive types, the new HORPO proved termination for all systems that the original HORPO could, while achieving an average runtime improvement of about 35 %.

The paper also contains a thorough theoretical analysis. It proves that the new ordering satisfies the standard HORPO properties: well‑foundedness, monotonicity, stability under substitution, and compatibility with β‑reduction. Moreover, it demonstrates that eliminating the computability closure does not compromise the ordering’s strength; the same class of terminating higher‑order rewrite systems is captured. The authors provide detailed proofs in the main text and supplementary material, ensuring that the reconstruction is both sound and complete.

In conclusion, this work delivers a decidable, more efficient, and conceptually clearer definition of HORPO. By postponing type comparisons until they are truly needed and by handling bound variables with explicit scope‑aware rules, the authors make the ordering applicable to a broader range of higher‑order systems, especially those involving strictly positive inductive types and their associated recursors. The result is a practical tool that can be integrated into existing termination‑checking components of proof assistants and term‑rewriting frameworks, offering both theoretical robustness and tangible performance gains.