Lean formalization of interior gradient Hölder estimates for infinity-harmonic functions
Juan J. Manfredi machine-checked in Lean one of the result families in OpenAI's October 2026 mathematics release, a uniform interior estimate giving every bounded viscosity solution of the infinity-Laplace equation on the unit ball of , , a gradient that is Hölder continuous on the half ball, and with it the corollary that every infinity-harmonic function on a connected open set in those dimensions is ; the Lean proofs and the blueprint were written by Claude under an AI-assisted formalization pipeline.
- Lab
- Independent
- Model
- Claude, under the Trellis AI-assisted formalization pipeline; the underlying manuscript came from an unreleased internal OpenAI model
- Field
- Mathematics
- Date
- 2026-10-29
- Human collaborators
- Juan J. Manfredi
What was found
The infinity-Laplace equation arises as the limit of the -Laplace equation and governs absolutely minimizing Lipschitz extensions. How smooth its solutions are has been open away from the plane: Savin proved planar solutions are in 2005, and Evans and Smart proved in 2011 that solutions are differentiable everywhere in every dimension, but neither a uniform gradient bound nor regularity was known for . The OpenAI manuscript of 4 October 2026 claims both: for every there are and a finite with on the half ball for every bounded viscosity solution on the unit ball, and as a consequence that every infinity-harmonic function on a connected open subset of is . The formalization carries all three of the manuscript's results, the estimate, the corollary and the Liouville theorem for the degenerate limiting equation on space-time that the proof rests on, as InfHarmonic.main_estimate, InfHarmonic.interior_C1 and InfHarmonic.liouville, proved from Mathlib and including the Crandall-Ishii-Lions theorem of sums and Alexandrov's theorem used along the way. It is pinned to Lean v4.35.0-rc2 and the matching Mathlib release, and the README reports no sorry outside the statement file and no axioms beyond the three standard ones.
Novelty check
arXiv, 2026-11-02: abs:"infinity-harmonic" returns 27 records and none states a uniform interior Hölder-gradient estimate in dimension three or above. The literature the search turns up is consistent with the problem having been open: Savin's planar result and a 2019 reworking of it, Evans and Smart's everywhere differentiability and its 2014 extension to a class of Aronsson equations, and, most usefully, arXiv:2610.13137 of 2026-10-09, which converts qualitative into quantitative regularity for degenerate equations and obtains uniform interior estimates for infinity-harmonic functions in the plane only, by way of Xu's planar qualitative result. That paper appeared five days after the OpenAI manuscript's date and still treats the planar case as the frontier, which is evidence that the claim for was not already in the literature. OpenAI's own formalization catalogue at lean/formalization.yaml lists no source matching this family, so the release attached no Lean proof to it, though that catalogue was only partly read. Within this registry the family is one of the 372 carried by 2026-10-06-openai-math-collection with no entry of its own, and no entry covers the infinity-Laplacian or infinity-harmonic functions.
Caveats and known objections
Graded formal on the Palomar registration, which records that the proof typechecks against the recorded statement at the pinned commit under two kernels, nanoda and con-ron, with permitted axioms exactly propext, Quot.sound and Classical.choice and trust level high. What that does not settle is whether the Lean says what the paper says, and the README is unusually direct about it: Lean does not check that the definitions, among them IsInfHarmonicOn, HolderSeminormOn, OscOn and IsFitClass, and the three statements mean what the manuscript means, and Challenge.lean is the statement of record. The README does not claim that the match against the manuscript's Theorem 1.1 and Corollary 1.2 was checked by hand; it says each node of the blueprint was reviewed automatically for soundness, for faithfulness to the paper and for agreement between its Lean and natural-language statements, which is a machine reading throughout. Palomar's own review outcome is neutral and its reviewer is a model, not a human referee. That gap matters more here than in 2026-10-20-trilinear-hilbert-transform-lean, which carries the same grade on the same kind of registration: there a specialist in the area wrote the theorem and the definitions by hand, so a human vouched for the statement the kernel then checked, whereas here the blueprint, the statement of record and every proof were machine-generated and machine-reviewed, and nobody is on record as having audited the one thing a kernel cannot check. The grade is kept at formal because a reproducible artifact was mechanically verified and the weaker grades on the scale would each assert something untrue, that humans checked it or that no artifact exists, but it rests on a thinner human chain than any other formal entry here and should be read with that in mind. The mathematics being formalized is an unrefereed manuscript from an unreleased model with no human author, carried at claimed in 2026-10-06-openai-math-collection, and OpenAI has not been contacted about this formalization, so the formal grade covers the argument as encoded in Lean and not the provenance of the result or any human reading of the manuscript. Autonomy is ai-led, matching 2026-10-20-trilinear-hilbert-transform-lean rather than the two other families split out of the same collection, 2026-10-27-pts-beta-eta-normalization and 2026-10-27-bernoulli-shifts-unequal-entropy, which are graded collaborative because a second laboratory's model contributed new mathematics on top of OpenAI's input. Here no new mathematics is claimed: the human chose the target and directed the pipeline, and the blueprint and every Lean proof were machine-generated.
Machine-checked elsewhere
PALOMAR-2026-10-29-000040 · na0b/infharmonic-lean at e8710101271a · checked 2026-10-29
ProvedInfHarmonic.main_estimateInfHarmonic.interior_C1InfHarmonic.liouville
A Lean proof that typechecks against the recorded statement at a pinned commit under a declared axiom set, checked mechanically rather than by human review. It does not certify that a result is new or of research interest: the only filter on that is a language-model screen, and Palomar states that it adds no human editorial step.
Permitted axioms are exactly propext, Quot.sound and Classical.choice, with two kernels, nanoda and con-ron, and trust level high. Palomar records the relationship to the OpenAI manuscript as formalizes, with author endorsement not contacted.
No independent check is recorded on this entry yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Formally verified and AI-led.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded formally verified for verification and ai-led for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). Lean formalization of interior gradient Hölder estimates for infinity-harmonic functions. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-29-infinity-harmonic-c1alpha-lean
BibTeX
@misc{whataifound-independent-2026-lean,
title = {Lean formalization of interior gradient Hölder estimates for infinity-harmonic functions},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Independent. Verification: Formally verified. Autonomy: AI-led.},
url = {https://whataifound.org/finding/2026-10-29-infinity-harmonic-c1alpha-lean}
}