OpenAI AI Solves Navier‑Stokes Existence Problem in 3.5 Days

A long-standing mathematical problem, the Navier-Stokes existence and smoothness problem, appears to have been solved, a development that has generated both excitement and controversy. OpenAI announced the solution, which they claim was achieved by an AI system more powerful than Astra.

The Controversy and Attribution

The announcement comes with a significant point of contention. Two scientists had reportedly made substantial progress on a related problem, reaching a point where their work could likely be extended to solve Navier-Stokes. Following rumors of their progress, OpenAI deployed an advanced AI system that subsequently produced a solution.

The presenter, a research scientist, expressed discomfort with the situation but felt it was important to acknowledge the external context. He emphasized the contributions of the two scientists outside OpenAI, offering them attribution for their work.

Another point of controversy involves the use of proprietary Large Language Models (LLMs) like ChatGPT and Claude by these scientists. They sought verification that OpenAI did not reuse their data as training material. OpenAI's official response was: "While unlikely, we cannot rule out that the identified data derived from their usage of our products helped improve our models." This highlights a recurring concern that data entered into proprietary AI chat boxes can be used for training, a risk that can be avoided by running free and open-weights AI systems where prompts remain on the user's machine.

Understanding Navier-Stokes Equations

The Navier-Stokes equations describe fluid motion. Despite the apparent complexity of natural fluid phenomena, these equations simplify the understanding of fluid dynamics by focusing on three key terms:

1. Advection: This describes how a fluid carries itself and objects within it. If an object is dropped into a river, it follows the flow. Similarly, the fluid itself advects, a process described by the directional derivative in the first term of the equations.
2. Pressure: Analogous to people pushing each other in a crowded bus, pressure in a fluid causes outward movement.
3. Diffusion: This term explains how differences in a fluid average out over time. For example, a drop of ink in water will eventually spread evenly throughout the liquid, making the entire glass the same color.

Additionally, external forces (like blowing on a liquid) influence fluid motion, which is incorporated through a simple addition to the equations. A crucial second equation, the incompressibility condition, states that the fluid's volume remains constant, meaning no liquid is lost or gained.

These equations can be discretized onto a grid, making them simple to evaluate computationally. Advection, for instance, involves moving fluid density to an appropriate neighbor, while diffusion is essentially averaging. This allows for the creation of computer programs that simulate reality, enabling applications like liquid simulations, wind tunnel tests for aircraft, and complex fluid control scenarios. The presenter makes his research source code freely available for others to use.

The Solved Problem: Smoothness and Breakdown

The core question addressed by the Navier-Stokes existence and smoothness problem is whether a fluid flow, starting smoothly and evolving according to these equations, will eventually break down mathematically. The answer, according to the recent solution, is "yes." The mathematics is not guaranteed to behave smoothly indefinitely.

OpenAI's solution reportedly starts from a state of rest and uses carefully designed external forces. The breakdown occurs when a vortex spirals inward, stretches, and its velocity increases without bound within a finite time, even while its total energy remains finite. Such mathematical cases are complex, though this phenomenon is not known to occur in nature, where molecular-level physics would take over before such a breakdown.

The Speed of AI in Mathematics

A surprising aspect of OpenAI's achievement is the speed at which their model found the solution: approximately three and a half days. This highlights AI's remarkable proficiency in mathematics, which stems from its ability to verify mathematical solutions automatically.

Unlike tasks such as writing articles, where human evaluation is required, mathematical proofs can be checked algorithmically. This allows AI systems to perform millions of "lessons" per hour, significantly accelerating their learning and improvement in mathematical domains. This rapid advancement suggests that AI will continue to become even more capable in mathematics.

Implications and Future Outlook

The increasing power of AI, particularly in areas like mathematics, underscores the need for greater coordination in safety and alignment efforts. The presenter also referenced Nobel laureate Sir Demis Hassabis's assertion that all diseases could be cured within 10 years. This ambitious goal might be achievable if diseases can be framed as verifiable problems, similar to mathematics, allowing AI to tackle them with its powerful verification capabilities.