The fluid puzzle behind the million-dollar question #
Fluid motion, from airflow over an airplane wing to global weather systems and blood circulation through our bodies, often appears smooth and predictable. The Navier–Stokes equations, established in the 19th century, are the mathematical bedrock describing this behavior. These partial differential equations capture the intricate interplay of forces within a moving fluid.
Yet, a profound question has stumped mathematicians for nearly a century: can a perfectly smooth, three-dimensional fluid flow suddenly "blow up" into a singularity in a finite amount of time? This would imply that properties like velocity or pressure become infinite at a specific point, defying the expected global smoothness of the solution.
In 2000, the Clay Mathematics Institute recognized the critical importance of this open problem, naming it one of its seven Millennium Prize Problems. The Institute offered a $1 million award for an accepted solution, whether it proves the existence of smooth solutions or definitively demonstrates a finite-time singularity.
Inside the 10,000-agent proof sprint #
OpenAI reportedly deployed a multi-agent system of 10,000 AI agents to tackle Navier problem. This swarm explored competing proof and disproof strategies simultaneously. Agents searched cached research, executed code, and critiqued intermediate work, operating in a highly parallelized workflow.
According to OpenAI's account, this intensive sprint generated approximately 2.7 million messages and 130 billion output tokens. Within roughly 88 hours, the system converged on a blowup construction, suggesting a finite-time singularity in smooth fluid flow. This disproved the global smoothness aspect of the problem.
The next critical phase involved GPT-6 Astra. This model reportedly spent 17 hours translating the more than 165-page argument into Lean, a formal proof assistant. Lean systematically checks every logical step, ensuring the mathematical rigor and validity of the AI-generated argument, a crucial step for verifying complex proofs.
A breakthrough claim with an important asterisk #
OpenAI’s agents reportedly disproved the global smoothness of Navier–Stokes solutions by constructing a finite-time singularity. Their mechanism describes an inward-spiraling vortex that stretches and concentrates, much like pulling taffy, until it reaches an infinite velocity at a specific point in time. This occurs while the total energy of the system remains finite.
Crucially, this construction relies on smooth external forcing, a condition permitted by the Clay Mathematics Institute’s formal problem statement. However, it does not directly resolve whether unforced, physically realistic fluid flows can also spontaneously "blow up" into singularities. This distinction is vital for understanding the result's implications for real-world fluid dynamics.
After the initial discovery, GPT-6 Astra spent 17 hours formalizing the 165-page proof in Lean, an interactive proof assistant. While Lean can rigorously verify every logical step of a formalized derivation, this does not equate to immediate scientific acceptance. Mathematicians must still scrutinize the underlying assumptions, interpretation of the result, and its relevance to the broader question of unforced Navier–Stokes blow-up. For more details, refer to On Navier–Stokes Millennium Prize Problem - OpenAI.
OpenAI has stated it will not claim the $1 million prize, acknowledging the nuances of the problem's resolution. The work builds on techniques developed by human mathematicians Tristan Buckmaster and Levent Alpöge, highlighting the collaborative, albeit sometimes contentious, nature of advanced mathematical research.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
The race for credit may outlast the computation #
The race for credit may outlast the computation
A priority dispute quickly followed OpenAI’s announcement. Reports suggest OpenAI initiated its internal effort after hearing rumors of progress from mathematicians Tristan Buckmaster and Levent Alpöge. Buckmaster, from NYU, published related work on finite-time singularities shortly before OpenAI’s claim.
Buckmaster voiced skepticism regarding early LLM-generated proofs, describing one as "the most horrendous proof he ever read." OpenAI, for its part, denied using private user queries or monitoring researchers’ work to inform its efforts. The data integrity of LLM-assisted discovery remains a critical concern for the scientific community.
This episode highlights wider stakes for the scientific process. Independent review, proper attribution, and rigorous reproducibility will determine if this marks a durable mathematical result or a striking demonstration of agentic search capabilities. OpenAI has stated it will not claim the $1 million prize from the Clay Mathematics Institute, shifting the focus from financial reward to the methods and ethics of AI-driven discovery. The scientific community awaits full details to assess the validity and implications of this unprecedented computational sprint.
Frequently Asked Questions #
What is the Navier–Stokes problem?
It asks whether smooth solutions to the three-dimensional Navier–Stokes equations always remain smooth, or can develop a singularity in finite time.
Did OpenAI solve the Navier–Stokes Millennium Prize Problem?
OpenAI has claimed a result for a version involving smooth external forcing. The proof still requires independent mathematical scrutiny, and that result does not settle every version of the problem.
What does Lean verify?
Lean checks that a proof follows from its stated definitions and assumptions. It can catch logical gaps in formalized work, but does not by itself establish that the model addressed the intended mathematical question.
Is OpenAI claiming the $1 million prize?
According to the account, OpenAI said it would not claim the Clay Mathematics Institute prize.