On November 20, 2026, Levent Alpöge released a proof of the Riemann Hypothesis as a Lean library comprising 17 million lines of code.
“We have full confidence that Alpöge’s proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement.”
Visually explore the lean code: https://riemann.world/
An attempt to turn the lean code into human readable text: https://riemann.tamnd.com/
ETA: This one may be further along: https://github.com/jimjamzola/RH-Alpoge-Reconstruction
The Alpöge–Claude proof of the Riemann hypothesis explained for the Riemann zeta function
I’m kinda terrified of what will come next year. This year has been crazy when it comes to AI in maths. If you told anyone in January that Riemann hypothesis will be solved by AI before end of the year, they would’ve thought you were some AI psychotic lunatic belonging to the mental asylum.
And now here we are. AI solving Riemann barely feels like the sensation it should be. It was expected to happen sooner or later.
I believe other fields will experience the same AI takeover next year (but I might be an AI psychotic lunatic belonging to the mental asylum).
The issue is not verificating those files. It is to verify that those files actually proof what they claim. This is like compiling software: no error messages does not imply correct program functionnality. My son told me they have already found tautologies (logical circlejerks) in that code.
They’re using Lean’s built in definition of the Riemann Hypothesis. It’s a short definition, and has been agreed to be correct for many years.
In Lean, the compilation is the verification of the proof. They don’t “run” the program afterwards in the traditional sense.
The code seems to be overly verbose (it has a proof of that if x = 1/2, then it can’t be larger or smaller than 1/2), but that doesn’t necessarily invalidate the proof.
Seventeen million lines of code and zero bugs, totally feasible /s
If there’s a bug that invalidates this proof then that bug must be in the Lean verifier.
They sure achieved full confidence in all 17 million lines being an accurate representation of the original problem quite quickly too. Perhaps there are viable shortcuts.
But I freely admit that the people involved know considerably more about the matter than I do.
I don’t think any of the 17 million lines are dedicated to modeling the problem, they’re targeting a known model in lean, no?
Yes, they use the definition from Mathlib, so only a handful of lines of their code needs to be checked, see https://math.mit.edu/~drew/grh/
Have they even verified the Navier-Stokes proof yet? Last interview I saw with the preeminent expert in this field (who had his research stolen by OpenAI) even he hadn’t verified the proof yet.
We need a new name for the gish-gallop for mathematical proofs.
*Riemann Theorem
Too soon, bro, TOO SOON! 😁
Hey, I’m just sayin. Ain’t no hypothesis, no more.
…think we should go with Riemann-Alpöge Theorem?
What about the dude that has proven it last year but was verifying his work one last time?
Such a damn shame. I’m sure he would have written it down straight away, if only the margin had not been too narrow to contain it.
It was a fun little exploratory trip :)
Wow, crazy!


