XNews

OpenAI Solves Navier-Stokes Millennium Prize Problem Using 10,000 Agents

OpenAI Solves Navier-Stokes Millennium Prize Problem Using 10,000 Agents

OpenAI announced that an internal AI model used 10,000 autonomous agents to find a singularity for the 3D Navier-Stokes equations, solving one of the seven Millennium Prize Problems. The agents worked 88 hours, exchanged ~5 million messages, and produced a 165-page proof formalized in Lean. The compute cost was several million dollars. The work was inspired by a rumor about researchers Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) who had solved related problems 12 hours earlier using multiple AI models, but their proof was for a slightly easier version of the forced Euler equation. Key human contributors were mathematicians Diego Córdoba and Luis Martínez-Zoroa, who developed a non-computer technique in 2021. OpenAI will not claim the $1 million prize.

Key insights

AI Achieves Major Mathematical Milestone

OpenAI deployed 10,000 autonomous agents running on an unreleased internal model to solve the Navier-Stokes existence and smoothness problem, a Millennium Prize question. The agents worked 88 hours, exchanged about 5 million messages, and the proof was formalized in Lean. The compute cost was estimated at several million dollars. OpenAI stated it will not claim the $1 million prize.

Competing Research and Priority Dispute

Twelve hours before OpenAI's announcement, Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) announced they had solved related problems using multiple AI models, including OpenAI's. OpenAI admitted its effort was triggered by that rumor but later determined the pair had solved a different problem (forced Euler equation, not Navier-Stokes). Buckmaster suggested OpenAI may have benefited from their use of OpenAI models. OpenAI ceded priority for the 3D Euler result and claimed priority for Navier-Stokes.

Human Mathematicians Behind the AI Breakthrough

The breakthrough relied heavily on techniques developed by mathematicians Diego Córdoba and his student Luis Martínez-Zoroa. They created an analytical cascade method that did not use computers, diverging from mainstream computer-assisted approaches since 2013. Their work defined the remaining hurdle: constructing a cascade that produces a singularity while keeping the forcing function smooth. AI filled the final engineering puzzle. Córdoba joked 'I don't use AI: I have Luis.'

Lean Formalization and Its Limits

The proof was formalized in Lean, guaranteeing logical correctness. However, the human step of ensuring the formal statement matches the intended mathematical claim remains essential. The formalization took 17 hours by a separate model. Buckmaster described the first LLM-generated proof as 'the most horrible thing I have ever read' and apologized for an 'AI garbage paper' released early due to competition.

Significance of the Navier-Stokes Singularity

The Navier-Stokes equations describe fluid motion. The problem asks whether solutions can develop a singularity (infinite speed) in 3D space without boundaries, using a smooth forcing function. A singularity would indicate that turbulence is even more strange than assumed. The result has no direct practical consequences because the equations assume infinite magnification, unlike real fluids made of molecules.

Source posts