What was already known
Chen’s July 2026 ChatGPT-generated paper already proved this full-support theorem by a different transformed-density route; earlier work established infinitely many support points without excluding gaps.
AI involvement
OpenAI attributes the September manuscript to its internal model. No named human mathematical origin or check of this separate proof is documented.
The repository directly attributes the manuscript to an internal model; autonomy from task selection and guidance is not documented.
Importance
The contribution is a distinct explicit proof method for an existing full-support theorem.
Limitations
The published Lean library states the conditional zero-field theorem; the comparator challenge is only a template. No named human explanation, independent build replay or axiom audit is documented.
Technical verification
OpenAI publishes a Lean 4 proof of the selected full-support statement in OAI/Probability/SKSupport/Main.lean, and its scope note matches the conditional zero-field claim. The comparator challenge intentionally contains sorry and is only a statement template; the completed proof is the library theorem. Build and axiom dependencies were not reproduced here.
Formal checking
Published full_support proof in OAI/Probability/SKSupport/Main.lean, with scope documented in lean/docs/281.md
- Checker
- OpenAI Lean 4 development
- Scope
- Full relative Stieltjes support for every admissible integrable minimizer of the pure zero-field SK Parisi functional
- Basis
- Reported by the linked checker
Checking a formal statement does not establish that it matches the original problem.
Human understanding
Default for missing evidence: no named human explains the September manuscript’s proof mechanism.
Correction and publication history
The contribution is a distinct explicit proof method for an existing full-support theorem.
Read this published revision