Weak and strong βη-normalization in pure type systems with βη-conversion

Get email updates

Ankit Pradhan, Cole Vick, Ruijie Fang, Carl Kwan

For any PTS specification using βη-conversion, system-wide weak βη-normalization implies strong βη-normalization. System-wide weak β-normalization also implies strong βη-normalization.

What was already known

OpenAI proved the implication for typing with β-conversion. In the βη\beta \eta variant, an η-step can erase a λ-domain annotation and introduces different proof obligations.

AI involvement AI-led

The project credits Claude Opus 5.5 with adapting OpenAI’s argument to η-conversion, including domain erasure and η-canonical observations, and writing the new Lean proof under the authors’ direction.

Importance Notable

The proof extends the weak-to-strong normalization implication to every specification of the βη\beta \eta variant.

Technical verification Formal (Lean)

The project reports Lean checking and Comparator replay of both implications with the three standard axioms. The authors also compared the challenge statement with Geuvers’s definitions.

Formal checking

The cited project reports the check for these statements.

Checker
Pradhan, Vick, Fang and Kwan, reporting Lean checks
Scope
PureTypeSystem.weak_implies_strong_betaEta and PureTypeSystem.weak_beta_implies_strong_betaEta for any PTS specification; the formalization uses the three standard Lean axioms.
Basis
Reported by the linked checker

Human understanding Few

The authors checked the statement against Geuvers’s definition. A human account of the central proof adaptation is undocumented.

Correction and publication history