Did AI Solve One of Math’s Hardest Problems?

Original Media · 21m · English · YouTube · StarTalk

TLDR

Astrophysicist Mordecai Mark Mac Low explains the Navier-Stokes equations that govern fluid dynamics, then discusses OpenAI's claim that AI found a counterexample proving these equations can produce singularities (infinite velocities) — solving a Millennium Prize problem. The proof, generated by $6M of GPU compute and verified in the Lean formal language, shows the equations aren't universally smooth. However, Mac Low notes this reflects the known physical limitation that Navier-Stokes assumes a continuous fluid, while real gases are made of discrete particles described by the more fundamental Boltzmann equation. The episode explores the tension between AI-generated 'black box' proofs and human mathematical understanding.

For you

PRO members can sign in and request more information — or their own version of this digest, tuned to what they care about.

Sign in to unlock

Surprising Insights

•

The AI proof cost $6 million in GPU time — more than the $1 million Millennium Prize reward.

•

The Navier-Stokes equations don't distinguish between water and air; they're the same equations with different parameters.

•

Newton's laws are built into the Boltzmann equation particle-by-particle, not abandoned when moving to statistical mechanics.

•

The human mathematicians who independently proved the same result were also using AI tools, but aimed to maintain human-understandable reasoning throughout.

Key takeaways

•

The Navier-Stokes equations describe all fluid flow — air, water, stellar gas — using velocity, density, and pressure variables that evolve via partial differential equations.

•

Turbulence isn't inherently good or bad; small-scale turbulence can be engineered to suppress large-scale drag, as seen in golf ball dimples and aircraft design.

•

The Millennium Prize problem asked whether Navier-Stokes solutions always remain smooth (finite) or can develop singularities where velocity goes to infinity in finite time.

•

OpenAI announced an AI-generated counterexample proving singularities exist, formalized as 26,000 Lean sub-theorems at a compute cost exceeding the $1M prize.

•

Two human mathematicians (Buckmaster at NYU, Alpige at Anthropic) had independently developed a similar proof, suggesting the AI result is likely correct.

•

The singularity doesn't mean real fluids reach infinite speed; it signals the breakdown of the continuum assumption — Navier-Stokes emerges from the Boltzmann equation only when gradients are small.

•

AI 'black box' proofs that output only a verified result without human-readable intermediate steps prevent mathematicians from building on the insights discovered along the way.

•

Mac Low uses AI as a literature discovery tool that finds real papers but hallucinates authors and interpretations, requiring verification against original sources.

Notable quotes

“They could not prove that these equations would never blow up. That you would not get regions where the equations would predict a value of the fluid speed that reached infinity.”

▸ Notable quotes · 4:02

“In a physical system, that suggests that your equation is going to need an additional approximation to come close to reality.”

▸ Notable quotes · 4:30

“Gas is not really a continuous fluid. And the Boltzmann equation assumption of Navier-Stokes is that it's a continuous indivisible fluid. Gas is atoms.”

▸ Notable quotes · 10:53

“The problem with the OpenAI solution is that wasn't unpacked. So they produced a solution, they provided this immense package of lean statements, but nobody can read that. Nobody can build on it.”

▸ Notable quotes · 17:21

“It'll only hallucinate an interpretation, not the original source.”

▸ Notable quotes · 18:53

Chapters

•

0:00 — Introduction to Navier-Stokes: Tyson introduces the Navier-Stokes equations as describing fluid flow, and welcomes fluid dynamics expert Mordecai Mark Mac Low from the American Museum of Natural History.

•

1:00 — What Fluids Are and Astrophysical Applications: Mac Low defines fluids as continuous, flowable materials without solid-like binding forces, and explains he uses Navier-Stokes to model star, galaxy, and planet formation in astrophysical gases.

•

2:29 — Turbulence and Its Uses: Discussion of turbulence as chaotic swirling flow, and how controlled small-scale turbulence (golf ball dimples, aircraft) can reduce drag by preventing large-scale separation.

•

3:35 — The Millennium Prize Problem: The unsolved problem: whether Navier-Stokes solutions always remain smooth and finite, or can develop singularities with infinite velocity — a question of mathematical stability.

•

7:17 — OpenAI's AI-Generated Counterexample: OpenAI announced an AI found a counterexample proving non-smooth solutions exist, formalized in Lean verification language using $6M of GPU compute over 88 hours.

•

8:58 — Independent Human Verification and Physical Meaning: Two mathematicians had independently developed a similar proof; the singularity reflects the continuum approximation breaking down, with the Boltzmann equation providing the more fundamental particle-based description.

•

16:46 — AI Proofs vs Human Mathematical Understanding: Philosophical discussion: AI 'black box' proofs that output only verified results without human-readable steps prevent the mathematical community from building on intermediate insights.

•

18:18 — Mac Low's Current AI Use in Research: Mac Low uses AI as a literature exploration tool that finds real papers but hallucinates authors and interpretations, requiring source verification.

Speakers

•

Neil deGrasse Tyson — Host

•

Mordecai Mark Mac Low — Guest Expert

⋮