AI-assisted proofs of 3D Euler and Navier-Stokes blowup reported, plus a Lean-only GRH proof

A post on terrytao.wordpress.com collects three results that its author expects to be "reported widely in other outlets". They are listed in decreasing order of how digested they currently are.

First, Levent Alpöge, Tristan Buckmaster and Matei Coiculescu, with heavy AI assistance, have written a human-readable proof of finite-time blowup for the unforced incompressible Euler equations in three dimensions. The proof is formalized in Lean.

Second, Alpöge and Buckmaster, again with heavy AI usage, have generated a proof of finite-time blowup for the unforced incompressible Navier-Stokes equations in three dimensions, again with a Lean formalization. The post says the paper is technically human-readable and the general strategy is broadly clear. It follows one of the expected paths to this result: numerically build an approximate discretely self-similar solution, then use rigorous perturbation theory to convert it into an exact solution. More work is needed, though, to produce a proper, professionally written proof.

Third, Alpöge, using an internal AI model, has generated an enormous Lean-only proof of the Generalized Riemann Hypothesis. The post points to a Lean Zulip thread as probably the best place to follow the efforts to verify the proof and extract its strategy. From very initial analysis, the proof appears to use the full power of modern developments in the Langlands program and arithmetic geometry. An update notes that the moderators of the Lean thread are now focusing the discussion on the Lean-related aspects of the proof only. The author imagines digestion will take some time, though here there is a human author to communicate with.

The post also points readers to Buckmaster's Mastodon post about the first two results.

Key facts

  • Alpöge, Buckmaster and Coiculescu, with heavy AI assistance, wrote a human-readable proof of finite-time blowup for the unforced 3D incompressible Euler equations, formalized in Lean.
  • Alpöge and Buckmaster, again with heavy AI usage, generated a Lean-formalized proof of finite-time blowup for the unforced 3D incompressible Navier-Stokes equations; the strategy is clear but a proper written proof still needs work.
  • The Navier-Stokes approach builds an approximate discretely self-similar solution numerically, then uses rigorous perturbation theory to make it exact.
  • Alpöge, using an internal AI model, generated an enormous Lean-only proof of the Generalized Riemann Hypothesis; verification and extraction of its strategy are still under way.
  • Very initial analysis suggests the GRH proof uses modern developments in the Langlands program and arithmetic geometry.

Why it matters

The post reports finite-time blowup proofs for the unforced 3D incompressible Euler and Navier-Stokes equations, both produced with heavy AI assistance and both formalized in Lean. It also reports an enormous Lean-only proof of the Generalized Riemann Hypothesis generated with an internal AI model. These are claims about major open problems, which is why the author expects wide coverage elsewhere.

Who it affects

Mathematicians working on fluid equations, on the Langlands program and arithmetic geometry, and on Lean formalization are the people the post addresses most directly. The Euler and Navier-Stokes results are credited to Alpöge, Buckmaster and Coiculescu (Euler) and to Alpöge and Buckmaster (Navier-Stokes). The GRH proof is credited to Alpöge alone, with an internal AI model.

How to use it

The Euler proof comes with a Lean formalization linked from the post, and Buckmaster has a Mastodon post about the first two results. For the GRH proof, the post names a Lean Zulip thread as probably the best place to follow verification and the effort to extract the strategy. The moderators there are now keeping the discussion to the Lean-related aspects of the proof.

How solid is it

The post ranks the results by how digested they are. The Euler proof is human-readable and formalized in Lean. The Navier-Stokes paper is technically human-readable and its strategy is broadly clear, but a proper, professionally written proof still needs more work. The GRH proof is Lean-only, and its verification is still under way. The post's comment on its strategy rests on very initial analysis.

Risks and caveats

The Navier-Stokes write-up is not yet a polished proof. The GRH proof is enormous and exists only in Lean, so understanding it will take effort; the author says the digestion process will probably take some time. The post names only "an internal AI model" for the GRH result and gives no further detail on the AI tools used for any of the three.

“I imagine the digestion process will take some time, though in this case at least we have a human author to communicate with to assist with this task.”

— Blog post at terrytao.wordpress.com, on the Generalized Riemann Hypothesis proof