Skip to the argument
Proof Atlas

The zeta-only argument

If a weighted prime sum stays below a certain bound on arbitrarily large intervals, every nontrivial zeta zero lies on the critical line.

The count

A prime power such as 2, 4 or 8 contributes log 2. A number such as 6 contributes zero. These are the von Mangoldt weights Λ(n). Give each prime power a further weight x/n − 1, then subtract the smooth main terms. Call the remaining error G(x).

G(x) = ∑1 ≤ n ≤ x Λ(n)(x/n − 1) − x log x + (1 + γ)x.

Here γ is Euler’s constant. The archive defines this quantity in three places and proves the definitions agree. The guide uses G throughout.

The window bound

Pick any exponent θ strictly between ½ and 1 and any starting threshold. The construction must produce a Y beyond that threshold such that G(x) ≤ xθ at every x from Y to Y3/2.

This is the arithmetic claim the zeta argument needs. The interval may depend on θ and the threshold. G may take large negative values. One large positive value inside a proposed interval can disqualify that interval.

How the pieces fit

Suppose a nontrivial zero had real part 0.70. Take θ = 0.60. The analytic argument shows that such a zero would force G above x0.60 somewhere in every sufficiently large interval of this form. The window bound would give an interval where that never happens. The same reasoning works for any real part above ½. The functional equation reflects a zero to the left of ½ into the excluded right half of the critical strip.

What has been simplified?

The analytic proof works with a single function, omits the real-zero case and gives a direct estimate on the interval. The arithmetic derivation traces how the submitted construction supplies the interval and accounts for the exponent loss. At modulus one, keeping a fixed factor of five lets us remove the final loss in the exponent. The earlier arithmetic estimates still use the archive’s deep inputs. The Lean record distinguishes the conditional target from an independently checked proof of RH.

Lean checked the conditional argument using 31 original project files. Adding the arithmetic back raises the count to 175,174, just 13 fewer than the original. The original submission claims a stronger statement about Dirichlet characters before specializing to zeta. All five headline conclusions use this one result. The distinction matters when deciding which parts the zeta argument can dispense with.

Why the bound excludes a zero