Unforced Navier-Stokes Singularity and the Generalized Riemann Hypothesis
- Source
- terrytao.wordpress.com
- Author
- archon1410
- Date

Context
Terence Tao, one of the best-known living mathematicians, used a short blog post on November 20 to point readers to three results produced with heavy AI help, and he listed them in order of how well digested each one is. That ordering is the key to reading them. The first two concern the equations for an incompressible fluid such as water: the Euler equations, which ignore friction, and the Navier-Stokes equations, which include viscosity. The long-standing question is whether a fluid that starts perfectly smooth can, within a finite time, develop a singularity, a point where its velocity blows up and the equations stop making sense. The new work claims that this happens in three dimensions for both equations even when the fluid is unforced, meaning nothing outside keeps pushing on it. The word matters because the Navier-Stokes result OpenAI announced in September, covered here, applied a smooth external force to the fluid.
The Euler proof, by Levent Alpöge, Tristan Buckmaster and Matei Coiculescu, comes as a human-readable paper plus a formalization in Lean, a programming language in which software checks every step of a proof. The Navier-Stokes proof, by Alpöge and Buckmaster, also has a Lean formalization. Tao calls that paper technically readable with a broadly clear strategy: compute an approximate solution that repeats itself at ever smaller scales, then use rigorous estimates to show that an exact solution lies close to it. He also says a proper, professionally written proof is still needed. The repository's notes add limits worth knowing. The numerical certificate the proof relies on is not in the repository, though the paper says it will be released. The computer check that the certificate meets the paper's conditions uses interval arithmetic and sits outside the Lean formalization, so that part of the argument does not yet rest on Lean. And a patch made on November 20 had not been rebuilt when the repository was assembled.
The third item is the least digested and the boldest. Alpöge, working with an internal AI model, has produced what Tao calls an enormous proof of the Generalized Riemann Hypothesis that exists only as Lean code, with no paper. The Riemann hypothesis predicts where the zeros of a particular function lie, which governs how regularly the prime numbers are spread out; the generalized version makes the same prediction for a whole family of related functions. Tao says the proof appears to use the full power of modern work on the Langlands program and arithmetic geometry, expects that understanding it will take time, and points to a discussion on the Lean community's forum where checking is under way. If the code passes Lean's checker, that shows the formal statement follows from the definitions and axioms it uses. Mathematicians would still need to confirm that the statement says what they mean by the hypothesis, and to understand why the argument works. Replying to a reader about which methods the proof draws on, Tao called it very early days.
Checking sign-in…
Loading comments…
