News · 2026-09-08

Terence Tao calls the AI-assisted fluid blowup proofs a breakthrough

Terence Tao, one of the most authoritative living voices on fluid equations, wrote on September 7, 2026 that new AI-assisted proofs of finite-time blowup for three fluid equations amount to "a breakthrough" with "a high likelihood of also extending to Navier-Stokes." He published that assessment a day before OpenAI announced it had done exactly that. Tao is not a coauthor of the work he is describing, which makes his post the closest thing available to an independent expert read on a story that has otherwise been fought out between two companies.

Key facts

• Tao's post is dated September 7, 2026, one day before OpenAI's Navier-Stokes announcement.
• The results he discusses cover three equations: the incompressible porous medium equation, the two-dimensional Boussinesq equation, and the three-dimensional incompressible Euler equations.
• He credits the underlying strategy to Diego Cordoba and Luis Martinez-Zoroa, not to the AI systems or to the authors who completed it.
• Primary source: Tao's blog, What's new.

To see why this matters, you need to know what "blowup" means and why it is hard. Fluid equations describe how a liquid or gas moves. A blowup result says: here is a starting state and a gentle push, and the equations themselves drive some quantity — the spin, the velocity gradient — to infinity in a finite time. Proving one is difficult because the fluid has to tear itself apart through its own internal dynamics, without anyone cheating by shoving in an infinite force by hand. The force has to stay smooth the entire time, right up to the moment things break.

The strategy that made this tractable is not new and is not AI's. Tao credits Cordoba and Martinez-Zoroa with "the basic strategy" of iteratively building solutions out of high-frequency corrections at successively smaller scales — imagine constructing a shape out of ever-finer ripples layered on ripples, each one arranged to feed energy down to the next, until the cascade runs away. Alpoge and Buckmaster found a variant of that method and pushed it from rough forcing to smooth forcing and onward to Euler. Buckmaster makes the same point even more strongly in his own statement, writing that he believes Martinez-Zoroa deserves a Fields Medal for the body of work.

Tao's framing of the AI contribution is worth reading closely, because it is neither dismissive nor breathless. He notes the arguments are "heavily AI-assisted," and that "as is now remarkably feasible in the modern era of autoformalization agents, their work has also been formalized in Lean." That phrase — remarkably feasible — is the tell. Two years ago, formalizing a research-level fluid dynamics proof was a multi-year project for a team. It is now something that happens in the same week as the proof. He also notes, dryly, that the authors had to work to simplify and rewrite the output from what they themselves call "the worst writeup we had ever seen."

The most important sentence in the post has nothing to do with machines. "The primary goal," Tao writes, is "developing mathematical understanding and insight. Without such understanding, even a problem as infamous as the Navier-Stokes regularity problem [is] of far less intrinsic significance." That is a standard applied equally to human and machine: a certificate that a statement is true is worth much less than knowing why it is true. It is the same argument this site covered when Tao said the bottleneck is understanding, not proofs, and it is the criterion by which the current wave of AI proof claims will eventually be judged.

Why this matters: Tao's post is the strongest available evidence that the mathematics under the week's announcements is real, independent of who gets the credit. It also quietly sets up the timeline. On September 7 the field's leading expert publicly assessed that this method had a high likelihood of extending to Navier-Stokes. On September 8, OpenAI said its agents had extended it, having started on September 1. Both things can be true — a route that an expert can see is a route that a well-directed search can also find — but the proximity is why the credit dispute became a story rather than a footnote.

The caveats are Tao's own, and he states them plainly. The three results "do not quite achieve" the Navier-Stokes regularity problem; they make finishing it look "very feasible," which is not the same as finished. And he is careful to say he still needs to work more to digest things — this is a first read, published within a day of the preprints, not a referee report. He does not comment on OpenAI or on Buckmaster's allegations at all, referencing only that external events forced an early release and declining to elaborate. On the mathematics, though, his verdict is unambiguous, and it is the one an outside reader should weight most heavily. Machine-checked proofs in Lean are now arriving fast enough that the scarce resource is no longer verification — it is someone qualified to say what the verified thing means.

Primary source, verified: read the paper →

Ground Truth. (2026, September 8). Terence Tao calls the AI-assisted fluid blowup proofs a breakthrough. Ground Truth. https://groundtruth.day/news/terence-tao-calls-the-ai-assisted-fluid-blowup-proofs-a-breakthrough.html

@misc{groundtruth:terence-tao-calls-the-ai-assisted-fluid-blowup-proofs-a-breakthrough,
  title  = {Terence Tao calls the AI-assisted fluid blowup proofs a breakthrough},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {sep},
  url    = {https://groundtruth.day/news/terence-tao-calls-the-ai-assisted-fluid-blowup-proofs-a-breakthrough.html}
}

Topics: mathematics · terence-tao · ai-for-science · formal-verification · lean · fluid-dynamics