Alpöge & Buckmaster Claim Navier-Stokes Proof in Lean 4, Riemann Hypothesis
⚡ Breaking News
TechCrunch AI
November 21, 20264 min read1

Alpöge & Buckmaster Claim Navier-Stokes Proof in Lean 4, Riemann Hypothesis

By — Senior AI Technical Editor

Back to News
❝

Levent Alpöge (Anthropic) and Tristan Buckmaster (NYU) announced a finite-time blow-up proof for Navier-Stokes equations, formally verified in Lean v4.33.1, alongside a translatable Riemann Hypothesis proof. Meanwhile, Stanford's Raffi Vakil published Alessandro Masullo's fully human proof of the general Riemann Hypothesis after 12 years. Neither result has undergone full human review, but the Zulip community confirmed the Lean translation succeeded.

Executive Overview

On November 20, 2026, Levent Alpöge (researcher at Anthropic) and Tristan Buckmaster (professor at NYU) announced a finite-time blow-up proof for the Navier-Stokes equations formally verified in Lean v4.33.1, alongside Alpöge's announcement of a translatable solution to the Riemann Hypothesis (1859). In parallel, Raffi Vakil (Stanford) published a fully human proof of the general Riemann Hypothesis by researcher Alessandro Masullo after 12 years of work. Neither result has undergone full human review, but the Zulip community confirmed the Lean translation succeeded. The repository is open source on GitHub, with no subscription tiers or token pricing associated with the proof.

📊 Official Technical Specifications & Data Card

Technical AxisConfirmed Official Data
💰 Pricing & Usage CostFree and open source on GitHub (repository tristanbuckmaster/navier-stokes). No subscription tiers or token prices are associated with the proof.
🌐 Platforms & Immediate AvailabilityGitHub (official repository); Lean v4.33.1 with Mathlib; Zulip chat for mathematical discussion; 79-page PDF paper.
⚡ Performance & Speed MetricsProof library: ~2,270 files and 600,000 lines. Numerical verification in Appendix B: 128-bit interval arithmetic via the Arb library (FLINT).
🛡️ Security & Tamper ResistanceFormal verification permits only three axioms: propext, Quot.sound, Classical.choice (per comparator*.json).
🧠 Context WindowNot applicable; the proof is not a language model. Full build requires significant memory and time per the README.
🌍 Arabic Language & Regional SupportPaper and repository are in English; no official Arabic documentation. Available globally via GitHub without regional restrictions.

Deep-Dive Features & Architecture

The official repository contains the paper (79 pages, version dated November 20, 2026) and the complete Lean package. Manifest files at the top of the lean/ folder define the definitions and formulate the theorems with sorry instead of proofs, while Solution.lean proves these theorems relying on the lean/Breakdown/ library. The comparator*.json files pair each manifest file with Solution.lean and specify which theorems must have matching statements, permitting only the three aforementioned axioms. An amendment dated November 20, 2026 was applied to the cell-layer statements and their proofs: the weight of one term became (|m|+1/2)^2 instead of m^2+1/4, one cell constant can be zero, and the combinatorial set in the cell is restricted to non-vanishing elements. The amended package was not rebuilt when the repository was compiled.

Numerically, verification that the certificate satisfies the conditions of Appendix B is performed by rigorous computation using 128-bit interval arithmetic via the Arb library through FLINT; this verification is not included in the formal translation and is not present in the repository. The certificate itself (which gives the paper its fingerprint) is not in the repository, and the paper states it will be published later. The source of the paper is also not in the repository.

On the Riemann Hypothesis front, Alpöge described in an essay the process of discovering parts of the proof 'through incredible confusion with Claude,' noting that the model 'flew, danced, and sparkled near the finish line.' Several Zulip participants, including Andrew Sutherland of MIT, confirmed the successful Lean translation of Alpöge's solution. In contrast, Raffi Vakil published Masullo's fully human proof, describing it as 'fully human, without AI,' after 12 years of work.

Benchmark & Competitive Performance

There are no model performance metrics or token prices in this case, so results are compared on mathematical verification criteria: Alpöge and Buckmaster's Navier-Stokes proof is formally verified in Lean v4.33.1 (2,270 files, 600,000 lines, three permitted axioms), while Masullo's proof of the general Riemann Hypothesis is fully human and has not yet been expressed in Lean, making its verification slower. Alpöge's Riemann Hypothesis proof is translatable to Lean and was confirmed by independents, but it has not yet undergone full human review.

Industry Impact & Enterprise Adoption

This development signals a growing intersection between AI-assisted mathematics and formal verification. The open-source release on GitHub enables researchers and developers worldwide, including in the Arab region, to inspect, build, and extend the proof without regional restrictions. The use of Lean v4.33.1 and Mathlib demonstrates the maturity of formal proof assistants for tackling Millennium Prize-level problems. However, the absence of full human review and the missing certificate and paper source in the repository underscore the need for cautious optimism. Enterprises and academic institutions may leverage such formalizations for security-critical verification and automated theorem proving, but adoption will depend on peer-reviewed validation.

Conclusion

The dual announcements of AI-assisted and fully human proofs for the Riemann Hypothesis, alongside the formalized Navier-Stokes blow-up proof, mark a significant moment in mathematical research. While the Lean verification and community confirmation are promising, the lack of complete human review and missing repository components mean the results remain provisional. The open-source availability on GitHub invites global scrutiny and collaboration, potentially accelerating progress toward solving the Millennium Prize Problems.

Media Source: TechCrunch AI | Official Developer Announcement: Original Source | Fact Verification & Analysis: AI Tools Oasis

Original Source:TechCrunch AIThis news was formulated based on coverage from TechCrunch AI

Frequently Asked Questions

What version of Lean is used in the Navier-Stokes proof?

Lean v4.33.1 with the Mathlib library, according to the lean/lakefile.toml file in the official GitHub repository. The repository contains approximately 2,270 files and 600,000 lines, and a full build requires significant memory and time.

How many pages is the published Navier-Stokes paper?

79 pages, dated November 20, 2026, according to the paper/finite-time-blow-up-navier-stokes.pdf file in the official repository.

Is Alpöge's Riemann Hypothesis proof verifiable in Lean?

Yes, several mathematicians in the Zulip chat, including Andrew Sutherland of MIT, confirmed the successful Lean translation of Alpöge's solution. However, full human evaluation of the proof is not yet complete.

How long did Masullo's human proof of the general Riemann Hypothesis take?

12 years of work, according to Masullo's introduction to his proof, published by Raffi Vakil, president of the American Mathematical Society, as 'fully human, without AI'.

What axioms are permitted in the formal verification of the Navier-Stokes proof?

Only three axioms: propext, Quot.sound, and Classical.choice, according to the comparator*.json files in the official repository.

AI Tools Oasis

AI Tools Oasis Team

Bringing you the latest news and analysis in the world of Artificial Intelligence with accuracy and credibility. Follow us for all updates.

Related News