I proved an effective André–Oort theorem for Shimura varieties, using Cureton–Thorner's zero-free half-plane for Dedekind zeta functions, formalized in Lean by @axiommathai
done w/ Claude Fable 5.2
https://t.co/0Oh665jX4B
i audited the statements in Levent Alpöge's Lean archive for the Riemann Hypothesis while everyone is still running the builds, and it's looking good so far.
the main theorem is Mathlib's own RiemannHypothesis with no hypothesis. the other four theorems use four short definitions (screenshot). i checked for the usual issues. Mathlib gives ζ a value at s = 1, but it's nonzero and the point is excluded anyway. the trivial zeros are exactly −2, −4, …, the pole of the principal character at s = 1 is excluded, and the imprimitive characters are included (their extra zeros sit on Re s = 0, outside the strip). nothing in the the files redefines any of these names.
everything rests on a single theorem, L(χ, s) ≠ 0 for Re s > 1/2 for every Dirichlet character. RH and the other three follow from that in about 200 lines citing only Mathlib lemmas.
so if a build passes on the three standard axioms, this is RH and GRH for Dirichlet L-functions as is usually stated. it doesn't give GRH for Dedekind zeta functions beyond abelian fields or for Hecke L-functions though, so Artin's primitive root conjecture (Hooley's proof goes through Kummer fields) and the GRH error term in Chebotarev for general extensions stay conditional.
Lean experts are currently investigating the validity of the claimed formalised proof of the Riemann Hypothesis - follow along at
https://t.co/GwuQHMlxvx
@ericweinstein I agree they should use his name, but this is not really a human thing, he is open about heavy ai usage and that's how this was done. ofc I assume he did contribute himself a lot and deserves a lot of credit.
@yifanzhang_ it is the equivalent of curing cancer for math, if anyone needs a reference to how big this is. it is the peak of math, nothing else is above it or comes close in terms of prestige.