Did AI Solve One of Math’s Hardest Problems?

Scientific concepts, discoveries, and nature phenomena

Navier–Stokes and fluid dynamics

• Navier–Stokes equations: Mathematical equations describing the flow of fluids (velocity, pressure, and density/amount of material) over time.
• Connection to Newton’s laws: The equations are presented as combining Newtonian motion principles with fluid behavior.
• Astrophysical fluids: Used to model how fluids/gas interact with gravity in processes such as the formation of stars, galaxies, and planets.
• Turbulence: Irregular, swirling flow patterns (eddies/swirl structures) described as normal rather than merely an instability. Turbulence can also be useful, for example:
  • Improved aerodynamics for airplane efficiency
  • Dimples on golf balls as a real-world analogue of turbulence management

The “millennium problem” result (AI + proof verification)

• Millennium Prize Problem referenced: A question about whether Navier–Stokes solutions remain smooth for all time.
• Smoothness hypothesis / stability question:
  • Whether solutions can “blow up” (develop singularities where predicted values become unbounded).
• AI-generated solution + formal verification:
  • OpenAI is claimed to have produced a counterexample-style argument to the smoothness question.
  • The proof was then verified in Lean, a formal mathematical verification language.
  • The verification is described as computationally expensive: millions of dollars worth of GPUs running for days (figures given in the subtitles).
• Lean proof structure:
  • Mentioned as containing about 26,000 subtheorems.

Why physical “infinity” may not occur

• The subtitles argue that the Navier–Stokes framework assumes a continuous fluid approximation.
• Boltzmann equation is introduced as a more microscopic theory:
  • Real gases are atoms/molecules (particles), not continuous media.
  • Gas dynamics can be described using statistical mechanics and particle collisions (particles “bounce off each other”).
• Deriving Navier–Stokes from Boltzmann:
  • Chapman and Enskog are mentioned as deriving Navier–Stokes from the Boltzmann equation/distribution.
  • The derivation assumes small gradients (linearization); if gradients become large (e.g., near a singularity), the approximation may fail.
• Central claim in the subtitles:
  • The discrepancy between Navier–Stokes “mathematical singularity” and physical behavior is due to breakdown of continuum/gradient assumptions, not because real matter becomes infinite.

AI vs “black box” research

• Discussion of how AI outputs may be difficult to interpret unless they are unpacked into human-readable reasoning.
• Emphasis on extracting/building from intermediate ideas, rather than only presenting a final answer.

AI as a research tool (literature search)

• Use of AI for literature search:
  • Identifying relevant sources about topics such as interstellar dust and radiation interactions.
• Described behavior:
  • The AI may hallucinate interpretations, but it attempts to link them to real articles so users can verify the original material.

---

Methodology / workflow outlined (as described)

Formal proof workflow

1. AI generates a proposed mathematical solution.
2. A human/verifier team encodes the reasoning into Lean.
3. Lean checks the proof against formal axioms/definitions (involving many subtheorems).
4. Independent researchers compare/accept the proof’s plausibility.

AI literature-search workflow

1. Ask AI research questions (e.g., about astrophysical interactions).
2. AI returns source suggestions (articles that exist).
3. The user consults the original sources to confirm what is actually stated.
4. AI improves breadth/relevance; verification remains with the researcher.

---

Researchers / sources featured (named in subtitles)

• Moray Mark Miklo (American Museum of Natural History; astrophysics; “specialist in liquids”)
• Isaac Newton
• Ludwig Boltzmann
• Chapman (first name not given in subtitles)
• Enskog (first name not given in subtitles; Chapman–Enskog)
• Tristan Buckmeister (NYU)
• Levent Alpagay (Anthropic)
• OpenAI (organization; mentioned as making the announcement)
• Anthropic (organization; mentioned as having researchers involved)
• Democritus
• Lean (Lean theorem prover / formal verification language; tool rather than a person)