Navier-Stokes Millennium Problem Solved by AI in 88 Hours

The Navier-Stokes equations, formulated by French engineer Claude-Louis Navier and Irish professor George Gabriel Stokes, describe fluid flow. These equations are central to one of the seven Millennium Problems, posed by the Clay Mathematics Institute in 2000, each carrying a million-dollar prize for its solution. One of these problems, related to the Navier-Stokes equations, was reportedly solved recently by OpenAI's AI in just 88 hours, a problem that has eluded human mathematicians for nearly a century. This AI-driven solution reportedly cost between $6 million and $15 million.

Understanding the Navier-Stokes Equations

The Navier-Stokes equations involve several key components:

• U (Velocity Field): Represents the velocity of a fluid parcel at a given position and time, indicating both its speed and direction.
• P (Pressure): Related to the pressure per unit density of the fluid.
• μ (Viscosity): An important term representing the fluid's viscosity, which resists flow.
• F (External Force): Any external force applied to the fluid, such as a spoon stirring tea.
• d/dt (Rate of Change with Respect to Time): Describes how the velocity field changes over time.
• ∇ (Gradient Operator): Indicates how quantities change with respect to space.

The equations essentially compare how the fluid's flow changes over time and space. An additional crucial component is the incompressibility condition, which states that the fluid cannot be squashed; if pushed in one direction, it must move in another.

The Predictive Power of Navier-Stokes

These equations allow for the prediction of fluid behavior. While simple for a cup of tea, they are vital for understanding complex phenomena like ocean waves, weather systems, and astrophysical fluid flows. They describe the flow of non-relativistic fluids, meaning fluids not moving at speeds close to the speed of light.

The Millennium Problem: Smoothness and Blow-ups

The core of the Navier-Stokes Millennium Problem is a mathematical curiosity: if a fluid starts in a perfectly smooth state with smooth forces applied, will it always remain smooth, or can it develop singularities (blow-ups) where quantities like velocity or pressure become infinite?

The challenge arises from the non-linear nature of the equations, specifically a quadratic term involving the velocity field (U). Non-linearities can lead to unexpected behavior, similar to how smooth matter in general relativity can collapse into a black hole with a singularity.

However, the Navier-Stokes equations also contain a viscosity term (μ) that tends to smooth things out. There's a constant "fight" between the non-linear term, which can cause blow-ups, and the viscosity term, which tries to prevent them. This dynamic is quantified by the Reynolds number.

• High Reynolds Number: The non-linear term dominates, increasing the likelihood of a blow-up.
• Low Reynolds Number: Viscosity dominates, leading to smoother flow.

It's important to note that the Navier-Stokes equations are an "effective description" and not a fundamental one. They are non-relativistic and don't apply at the molecular level. If a blow-up were to occur in the mathematical solution, it would likely indicate that the equation itself ceases to be an accurate description of reality at that point, rather than a physical infinity occurring.

The AI's Solution: A Vortex Blow-up

The Clay Prize problem for Navier-Stokes is divided into four categories (A, B, C, D) based on boundary conditions (infinite fluid vs. fluid in a box) and the presence of external forces. Categories A and B ask to prove that solutions remain smooth for all time with no external forces. Categories C and D, which allow for external forces, ask to prove that blow-ups can occur.

The AI's solution focuses on a specific type of vortex that exhibits a blow-up. This vortex has a central region where fluid flows inwards radially, then flows outwards axially (up and down) due to incompressibility.

The blow-up manifests as the vortex shrinking faster radially than angularly. While the radial velocity component blows up, the angular velocity component blows up even more quickly. Crucially, the Reynolds number along the angular direction blows up, indicating that the non-linear term is winning in that specific direction, leading to the singularity.

The "Fake Force" and the Annular Region

A key challenge in demonstrating this blow-up is how to transition from the "wild core" of the singularity to a "boring exterior" of smooth flow without applying infinitely sharp forces, which would violate the problem's conditions. The AI's solution employs a clever technique involving "fake forces" in an annular region between the core and the exterior.

These fake forces are created by applying tiny, fluctuating pulses that average to zero over time, meaning they don't add significant energy to the system. However, because the Navier-Stokes equations are non-linear, the square of these fluctuations does not average to zero. This squared term effectively acts as a smooth, non-zero force that allows for the transition from the singular core to the smooth exterior, satisfying the problem's conditions.

The Path to the Solution: A History of Contributions

The AI's solution builds upon decades of work by human mathematicians:

• 1934 - Jean Leray: Introduced "weak solutions" by smoothing the Navier-Stokes equations, demonstrating that solutions exist for these modified systems.
• Early 1980s - Caffarelli, Kohn, Nirenberg (CKN): Showed that if blow-ups exist, they must be very sparse in the fluid.
• Mid-2010s - Terence Tao: Explored a modified version of the Navier-Stokes equations, replacing the non-linear term with a "dressed" version that allowed him to engineer a blow-up, though not for the exact Navier-Stokes equations.
• Recently - Chen, Miao, Zhang (CMZ): Worked on the Euler equations (Navier-Stokes without viscosity), showing a "clockwork" of vortices could lead to a blow-up with finite energy, though their forces were not perfectly smooth.
• 2025-2026 - Buckmaster (NYU) and Alipour (Anthropic): Extended the CMZ idea to the Euler equations, finding a smooth force configuration that led to a blow-up.

OpenAI, reportedly aware of these developments, then focused on the Navier-Stokes problem. They quickly solved the Euler problem with vanishing forces and then, within 88 hours, the full Navier-Stokes problem with viscosity.

There is some debate about whether OpenAI's AI was influenced by the work of Buckmaster and Alipour, potentially through their use of OpenAI's Codex. However, the solution for the full Navier-Stokes equations, involving the "pulse" mechanism, is distinct from the "cascade of vortices" approach used for the Euler equations.

The speaker emphasizes that while celebrating the AI's achievement, it's crucial to acknowledge the extensive foundational work by human mathematicians that paved the way for this breakthrough. The journey to solving such problems is often as important as the solution itself.