Finite Time Blowup for Navier–Stokes and Euler
paperYour notes
OpenAI's claim on a Millennium Prize problem. The 166-page paper, authored simply "OpenAI," states as Theorem 1.1 that for every viscosity there exist a smooth, compactly supported force and a solution of the 3D incompressible Navier–Stokes equations starting from rest whose velocity becomes unbounded in finite time while its kinetic energy stays uniformly bounded, on R³ and on the torus, which are breakdown alternatives (C) and (D) in the Clay Institute's problem statement (the forced versions, not the unforced Cauchy problem). A companion paper constructs smooth, compactly supported, divergence-free initial data whose unforced Euler solution blows up. Both results are formalized in Lean 4 (4.34.0-rc2 with Mathlib) in a repository that reached 1,750 stars on its first day.
OpenAI's blog is blocked to crawlers, so the run details rest on press coverage: roughly 10,000 concurrent agents on an unreleased model it describes as more capable than GPT-6 Astra, with the Navier–Stokes result reached on September 5 after about 88 hours and Astra handling the Lean formalization. The result awaits independent mathematical review, and the announcement drew a priority dispute. Two smaller Lean repositories on prime gaps preceded it on September 2.
On October 6 OpenAI widened the program into openai/math (Apache 2.0), which the README describes as 719 manuscripts in 372 families "produced by an internal OpenAI model," added because "performance on our existing mathematical evaluations saturated." By the README's account the unreleased model was posed about 4,000 problems and spent an average of three hours of ChatGPT Pro thinking compute on each result. About 42% of the top-line results are formalized in Lean so far, and abridged reasoning summaries are released for ten families. Two results did not follow the fixed procedure, a zero-free region for the Riemann zeta function and the Hodge conjecture for CM abelian varieties, and the Re(s) > 11/12 write-up was edited by humans. The README also warns that unformalized results "could have issues." The collection had 15,400 GitHub stars by October 24. The manuscripts are posted as preprints, and the significance threshold used to select them is OpenAI's own.