Solving a Millennium Prize Problem
OpenAI has released a solution to the Navier–Stokes existence and smoothness problem. This mathematical puzzle has challenged researchers for roughly 90 years. The internal system used to crack it shows that the equations governing fluid motion can produce a singularity in finite time. The team published both a full written proof and a formal verification in the Lean programming language.
Navier–Stokes equations describe fluid dynamics through Newton’s second law of motion. They model fluids as continuous mediums instead of tracking individual molecules. These calculations are critical for aircraft design and weather forecasting. The central question asks whether smooth three-dimensional fluid motion can break down. Specifically, does a singularity occur where fluid speeds grow without bound in finite time, even with viscosity present?
The Mathematical Breakdown
Claude-Louis Navier and George Gabriel Stokes developed these equations in the nineteenth century. Jean Leray proved that solutions exist in a general sense back in 1934. However, whether these solutions remain smooth was a mystery. In 2000, the Clay Mathematics Institute designated it one of seven Millennium Prize Problems.
The new solution demonstrates that an initially smooth fluid at rest can develop a singularity. It establishes statements C and D from the official Clay Mathematics Institute formulation. The proof centers on a vortex that spirals inward and stretches. This region shrinks while gaining speed, yet energy remains finite. The equations require acceleration, pressure gradients, and viscosity to balance precisely even as velocity grows. This allows a smooth external force to exist alongside the breakdown.
Methodology and Agent Coordination
Training on a new internal model began on August 28. Following rumors that two Millennium Prize problems had been resolved, the team launched a full-scale effort to test the model against these high-impact math challenges. They deployed a system of approximately 10,000 concurrent agents to work on the Navier–Stokes problem.
These agents operated with access to code execution and internet search tools. They were organized into groups to explore various approaches. When one group achieved a breakthrough, the team used Codex to share insights across the other groups. The final resolution arrived on September 5, roughly 88 hours after the start. Formal verification took another 17 hours. The effort consumed 130 billion output tokens and sent 2.7 million messages.
Industry Implications and Future Context
This project began after staff heard rumors regarding mathematical progress by Anthropic employee Levent Alpöge and NYU professor Tristan Buckmaster. Once the OpenAI team completed their work on September 6, they contacted Alpöge and Buckmaster to discuss a concurrent release. The researchers confirmed that the external work focused on the forced Euler problem. OpenAI acknowledged their priority and congratulated them.
OpenAI stated that they do not intend to claim the million-dollar Millennium Prize for this discovery. The goal is to document the progress of their current AI models. This snapshot of performance highlights the acceleration in machine reasoning. As the firm continues its mission, it plans to monitor these models closely to ensure that the rapid growth of capabilities remains safe and accountable for general use.