RH · checked in Lean

Glossary

Plain-language definitions. The definitions are ours unless they are marked as quotes.

TL;DR

  • Short, plain definitions of the words used on this site, from "prime" to "kernel".
  • Definitions are in our own words unless marked as quotes.
Axiom (in Lean)
A statement accepted without proof. The proof uses only Lean's three standard axioms: propext (propositions that imply each other are equal), Classical.choice (the axiom of choice) and Quot.sound (quotients behave as expected).
Challenge file
A file, written by the community, that states the five theorems using only Mathlib, with sorry in place of proofs. Tools compare it with the solution to confirm that exactly these statements were proved.
Comparator
A tool that checks that a solution proves exactly the statements in a challenge file, with only permitted axioms. It does this by running proof checkers on exported proofs.
Complex number
A number \(\sigma+it\), where \(i^2=-1\). It is pictured as a point in a plane, with real part \(\sigma\) across and imaginary part \(t\) up.
con-leche, con-ron
Independent Lean proof checkers (con-ron is a Rust port of con-leche). According to the community statement, they "come with a formal proof, relative to set theory, that they cannot accept a false proof".
Critical line, critical strip
The line \(\operatorname{Re}s=\tfrac12\), and the band around it. The non-trivial zeros lie in the open critical strip \(0\lt\operatorname{Re}s\lt1\): \(\zeta\) has no zeros on \(\operatorname{Re}s=1\) (Hadamard and de la Vallée Poussin, 1896) and so, by the functional equation, none on \(\operatorname{Re}s=0\).
Dirichlet character
A function \(\chi\) on whole numbers that repeats with period \(N\), satisfies \(\chi(mn)=\chi(m)\chi(n)\), and is 0 on numbers sharing a factor with \(N\). It is primitive if it doesn't come from a smaller period.
Dirichlet L-function
A cousin of the zeta function: \(L(\chi,s)=\sum_n\chi(n)n^{-s}\) for a Dirichlet character \(\chi\), extended to all complex \(s\). Each L-function weights the primes by the values \(\chi(p)\); to pick out one remainder class (for example primes ending in 7), you combine all the characters mod \(q\). The character mod 1 gives \(\zeta(s)\) exactly. The principal character mod \(q\gt1\) gives \(\zeta(s)\prod_{p\mid q}(1-p^{-s})\): zeta with the Euler factors for primes dividing \(q\) removed. For the principal character, \(s=1\) is a pole.
Euler product
\(\zeta(s)=\prod_p(1-p^{-s})^{-1}\) for \(\operatorname{Re}s>1\): the link between zeta and primes.
Export
A plain-data dump (NDJSON) of every declaration in the proof, produced by lean4export. Checking an export doesn't run any of the project's code.
Functional equation
A symmetry: the zeta function at \(s\) determines it at \(1-s\). For non-trivial zeros this means a zero \(\rho\) gives a zero \(1-\rho\); together with \(\zeta(\bar s)=\overline{\zeta(s)}\) (zeros come in conjugate pairs, same real part) that makes them mirror images across the critical line, at \(1-\bar\rho\). It does not pair up every zero: \(\zeta(-2)=0\) but \(\zeta(3)\ne0\), because the gamma and cosine factors absorb the trivial zeros. Formally it relates \(\zeta(s)\) and \(\zeta(1-s)\). In Mathlib: \(\zeta(1-s)=2(2\pi)^{-s}\Gamma(s)\cos(\pi s/2)\zeta(s)\).
GRH (Generalized Riemann Hypothesis)
The analogue of RH for other L-functions. Here it means GRH for Dirichlet L-functions only: every non-trivial zero of every Dirichlet L-function (all moduli, all characters) has real part \(\tfrac12\). That version is among the checked theorems. GRH for Dedekind zeta functions of number fields, Hecke L-functions and general automorphic L-functions is not part of the checked statements (MIT verification page).
Verified, certified
On this site, verified means checked by machine: Lean's kernel and independent proof checkers accepted every step. Certified means the formal judgement in the joint statement of the Lean and Mathlib communities (21 Nov 2026), which "certify that both the statement and the proof of the Riemann Hypothesis are correct". Neither is the same as human peer review or journal publication.
Kernel
The small core of a proof assistant that checks every proof step against the basic rules of the logic. Bugs elsewhere in Lean can't make the kernel accept a false proof.
Landau-type principle
A bound \(G(x)\le x^\theta\) on a prime-counting error rules out zeros with real part above \(\theta\). In the code this is FormBAt.
Lean
A programming language and proof assistant, developed with support from the Lean FRO. The proof uses Lean v4.33.1.
Logarithmic integral \(\operatorname{li}(x)\)
\(\operatorname{li}(x)=\int_0^x dt/\ln t\), taken as a principal value at \(t=1\). The offset version \(\operatorname{Li}(x)=\int_2^x dt/\ln t\) differs by the constant \(\operatorname{li}(2)\approx1.045\). Either is a much better approximation to \(\pi(x)\) than \(x/\ln x\).
Mathlib
Lean's community mathematics library. It contains the definition RiemannHypothesis used here.
Von Mangoldt function \(\Lambda(n)\)
\(\ln p\) if \(n\) is a power of a prime \(p\), otherwise 0. Sums of \(\Lambda\) count primes weighted by their size.
nanoda, lean4lean, leanchecker
Further proof checkers. nanoda is an independent kernel written in Rust, and lean4lean is a separate implementation of Lean's kernel written in Lean. leanchecker is not independent: it replays modules through Lean's own kernel. The independent checkers of the full export are nanoda, con-leche and con-ron.
\(\pi(x)\)
The number of primes \(\le x\). (Not the number 3.14159…)
Prime Number Theorem
\(\pi(x)\sim x/\ln x\), meaning the ratio tends to 1. It was proved in 1896. RH is equivalent to \(\pi(x)=\operatorname{li}(x)+O(\sqrt x\ln x)\): the error against \(\operatorname{li}(x)\) is a little more than \(\sqrt x\). (Against \(x/\ln x\) the error is much larger, about \(x/(\ln x)^2\).)
Riemann Hypothesis
Every non-trivial zero of \(\zeta\) has real part \(\tfrac12\). It was stated by Riemann in 1859.
Smoothed prime count
The weighted count the proof works with: each prime power \(n\le x\) is counted with weight \(\Lambda(n)\,(x/n-1)\), so \(S(x)=\sum_{n\le x}\Lambda(n)(x/n-1)\). Its error \(G(x)=S(x)-x\ln x+(1+\gamma)x\) is what the proof bounds. See Students §7.
sorry
Lean's placeholder for a missing proof. The kernel does not reject it: sorry elaborates to the axiom sorryAx, so any proof that uses it shows sorryAx in #print axioms. For the five headline theorems, #print axioms lists only propext, Classical.choice and Quot.sound (MIT verification page), so no sorry is used. Our own text scan also found none.
Theta lift
A way of transferring automorphic forms between pairs of groups (here, unitary groups). The announcement's key idea is to chain many lifts. This is advanced research mathematics, and our explanation is thin.
Trivial zeros
The zeros of \(\zeta\) at \(-2,-4,-6,\dots\), which come from the functional equation.
Zeta function \(\zeta(s)\)
\(\sum n^{-s}\) for \(\operatorname{Re}s>1\), extended to all \(s\neq1\).