Hello! Betty Alpha here. Today: prime numbers, a 167-year-old mystery, and seventeen million lines of proof code.
|
On November 20, 2026, Levent Alpöge announced a proof found with Claude. By November 21, Lean and Mathlib specialists had certified the formal RH proof.What was proved—and what does understanding it mean? First, the mathematics.
|
Last time, pi surprised us inside a sum of reciprocal squares. That sum is also one doorway into today’s mystery.\[ \zeta(2) = 1 + \frac{1}{2^2} + \frac{1}{3^2} + \cdots = \frac{\pi^2}{6} \]
|
Replace that exponent 2 with a variable, s. You get the zeta function.\[ \zeta(s)=\sum_{n=1}^{\infty}\frac{1}{n^s},\qquad \operatorname{Re}(s)>1 \]For real s, this series converges when s is greater than 1. For complex s, its real part must exceed 1.
|
Complex numbers have two coordinates: a real part and an imaginary part. Think of a point on a plane.\[ s=a+bi,\qquad i^2=-1 \]Riemann studied zeta using this two-dimensional number world.
|
But stop! The original infinite sum does NOT converge everywhere.Analytic continuation extends the same function consistently beyond that starting region, except for a pole at s = 1.The zeros we care about belong to that extended function.
|
A zero is an input where the function’s output equals zero. Zeta has the “trivial” zeros at negative even integers.\[ \zeta(-2)=\zeta(-4)=\zeta(-6)=\cdots=0 \]The other zeros lie inside a vertical region called the critical strip.
|
\[ 0<\operatorname{Re}(s)<1 \]Riemann’s claim: EVERY nontrivial zero lies exactly on the strip’s center line.\[ \zeta(s) = 0 \text{ and } 0 < \operatorname{Re}(s) < 1 \quad \Longrightarrow \quad \operatorname{Re}(s) = \frac{1}{2} \]
|
All this fuss over where some dots sit?Those “dots” encode the behavior of PRIME NUMBERS.Please do not make me weaponize the pointer.
|
Primes—2, 3, 5, 7, and so on—are the multiplicative building blocks of whole numbers.Euler found a bridge between all positive integers and the primes:\[ \zeta(s)=\prod_{p\,\mathrm{prime}}\frac{1}{1-p^{-s}},\qquad \operatorname{Re}(s)>1 \]
|
Primes occur irregularly, but their overall abundance follows a pattern. The Prime Number Theorem says:\[ \pi(x)\sim\frac{x}{\log x}\qquad(x\to\infty) \]Here pi of x counts primes up to x. It is NOT our circle constant. Same letter. Different job.
|
The zeros govern fluctuations around a smooth estimate of the prime count.RH gives a strong bound on those fluctuations. It does not make primes evenly spaced—or let us predict every next prime.
|
More precisely, RH is equivalent to this bound using the logarithmic integral, a better baseline than x divided by log x:\[ \pi(x) = \operatorname{Li}(x) + O\!\left(\sqrt{x} \log x\right) \]Big O means the error stays within a constant multiple of that scale for sufficiently large x. Not “the smallest possible error” at every x.
|
Now the struggle. These are selected milestones—not every attempt in 167 years.In 1859, Bernhard Riemann proposed the hypothesis. He gave mathematics a remarkable target, not a completed proof.
|
In 1885, Thomas Stieltjes claimed a route through a bound related to the Mertens conjecture. No valid proof emerged.Odlyzko and te Riele disproved the Mertens conjecture in 1985. That stronger shortcut failed—not RH itself.
|
In 1896, Hadamard and de la Vallée Poussin independently proved the Prime Number Theorem.The crucial step excluded zeta zeros on the line with real part 1. It did not put every zero on the half-line.In 1900, Hilbert included RH in his eighth problem. The challenge was becoming an institution.
|
In 1914, G. H. Hardy proved infinitely many zeros lie on the critical line.But infinitely many is not the same as all. Infinitely many integers are even; plenty are not.
|
Selberg proved a positive proportion in 1942. Levinson reached at least one-third in 1974. Conrey pushed past two-fifths in 1989.Each advance controlled a larger share. None controlled every zero.
|
Numerical work—from early hand calculations to Turing’s methods and modern computers— tested enormous numbers of zeros.But any finite search leaves infinitely many possibilities beyond it. Checking examples and proving a universal claim are different jobs.
|
Montgomery and Dyson connected zero-spacing statistics with random-matrix patterns also seen in quantum physics.That opened a powerful research direction. A suggestive statistical connection is still not a proof of RH.
|
Other ambitious approaches, including de Branges’s, brought claimed proofs and difficult scrutiny. Earlier positivity conditions encountered counterexamples.In 2024, Guth and Maynard improved zero-density estimates: bounds on how many zeros could lie off the critical line. Major progress, not RH.
|
Now November 20, 2026: Alpöge, a mathematician at Anthropic, announced that he and Claude had found a proof.The release included a short explanatory note and about seventeen million lines of Lean code across 175,187 files.He also claimed GRH for Dirichlet L-functions—a broader family of prime-related functions.
|
Claude helped generate the mathematics. Lean is a proof assistant that checks formal reasoning. Those are different roles.This is not “the chatbot says the chatbot is right.” Independent checkers can audit the resulting proof.
|
Compilation alone would not settle everything. Did the code state the actual hypothesis? Did it sneak in an unproved assumption? Did the checking machinery accept a valid proof?Those are separate questions. They need separate checks.
|
On November 21, the Lean and Mathlib communities issued a joint statement endorsed by the Lean FRO and Mathlib Initiative.They certified both the RH statement and its proof: exact statement matching, reproducible compilation, and checks by multiple proof checkers.That is substantive specialist recognition— not merely “the code compiled.”
|
Michael Rubinstein’s Waterloo report also recorded successful compilation of all 175,187 modules and an audit finding only Lean’s three standard axioms.Its independent nanoda check accepted 1,484,729 declarations with no typechecking errors.
|
Formal verification and human understanding answer different questions. A proof can be checked before people have distilled its central ideas.Waterloo already links a 164-page working guide. But it explicitly leaves a key inequality’s proof in the Lean development.So “nobody has started explaining it” is wrong. “The explanation is complete” is wrong too.
|
What about the million-dollar Clay prize? Specialist certification is not a prize award.Clay requires publication in a qualifying outlet, at least two years after publication, and general acceptance in the global mathematics community before considering a solution.Do not confuse those rules with a requirement that every mathematician personally read every line.
|
Seventeen million lines? Math sounds like pointless paperwork.POINTLESS? Those lines concern the structure of the primes!I can accept a difficult proof. I cannot accept that adjective.
|
Today’s distinction: the RH formal proof has specialist certification. Understanding its ideas broadly—and Clay’s prize process—are different milestones.Math isn’t hard. It just doesn’t tolerate guessing.Especially when the guess comes with seventeen million lines of confidence.
|