OpenAI Resolves Navier–Stokes Millennium Prize Problem

AI Proof of Navier–Stokes Singularity

OpenAI has produced an analytical proof and a Lean formalization demonstrating that the dynamics of the Navier–Stokes equations for fluid motion can develop a singularity in finite time. This result resolves the Navier–Stokes existence and smoothness problem, one of the seven Millennium Prize Problems established by the Clay Mathematics Institute in 2000.

The proof establishes that an initially smooth fluid at rest, subject to a smooth external force, can develop a singularity where fluid speeds grow without bound within a finite amount of time, even while the total energy remains finite. This confirms that the continuum approximation used in these equations can break down, marking a limit to how the equations model fluid behavior.

The Mathematical Mechanism

The solution describes a vortex—a spinning swirl of fluid—that spirals inward and becomes increasingly elongated. In this scenario, the central region of the vortex shrinks while its speed increases. To maintain finite energy and a smooth external force, the terms describing acceleration, pressure gradients, momentum transfer, and viscosity must grow significantly yet cancel each other out in a precise balance. This allows the velocity to grow without bound without requiring an infinite external force.

Methodology: Multi-Agent Coordination at Scale

OpenAI solved the problem using a system of coordinating agents powered by an internal model that is significantly more capable than GPT-6 Astra. The process involved a massive scale of compute and agent interaction:

• Agent Architecture: Agents were organized into groups with internal communication capabilities. The group that successfully resolved the Navier–Stokes problem consisted of approximately 10,000 concurrent agents.
• Iterative Approach: The system first attempted a "shotgun" approach across all open Millennium Prize problems. After successfully resolving the regularity problem for the unforced Euler equations (a simpler version of the Navier–Stokes problem), OpenAI shifted resources to focus specifically on Navier–Stokes.
• Cross-Pollination: The team used Codex to consolidate insights from different agent groups, using intermediate results to guide follow-up prompts.
• Resource Consumption: Across all attempted problems, the agents sent 4.9 million messages and used approximately 300 billion output tokens. For the Navier–Stokes problem specifically, the agents used 130 billion output tokens and sent 2.7 million messages.

Timeline of Discovery

• August 28: Training began on the new internal model.
• September 1: OpenAI launched the multi-agent effort after hearing rumors of concurrent progress on Millennium Prize problems.
• September 5: The agents arrived at the Navier–Stokes resolution, approximately 88 hours after the first agents were launched.
• September 6: Lean formalization and verification were completed via GPT-6 Astra, taking an additional 17 hours.

Controversy and Concurrent Work

OpenAI acknowledged concurrent work by Levent Alpöge (Anthropic) and Tristan Buckmaster (NYU), who resolved the forced Euler problem. While OpenAI states their proofs differ significantly and the results are distinct (unforced vs. forced Euler), the company admitted they cannot rule out that de-identified data from the researchers' use of OpenAI products may have helped improve the models used for the proof.

Community Response and Ethical Concerns

The announcement sparked significant debate among the mathematical and AI communities regarding the ethics of AI-driven discovery:

"We have now seen that even the rumor of someone working on a problem can trigger a massive amount of AI-powered effort to flatten it before the original research project has time to reach its full potential. The incentives may now be pointing in the direction of no longer sharing any promising research directions with the broader community..." — Terence Tao

Critics on Hacker News and other platforms raised concerns that AI labs may use their access to user prompts to "scoop" human researchers. Some users noted the extreme disparity in resources, estimating that the token usage for this effort would have cost upwards of $15 million if performed via a public API, highlighting the gap between frontier labs and independent scientists.