临界之镜 · Re(s) = 1/2 这面镜子把每个离线零点映出一个孪生;只要有一个离线,它激起的素数巨浪就会冲破 √x 的堤坝Critical mirror · the line Re(s) = 1/2 mirrors every off-line zero into a twin, and a single off-line zero would send a prime wave crashing through the √x barrier
ρ —1 − ρ—
理论接口 · 这个页面在演示什么Theory interface · what this page shows
Both scenes tell the same story: a zero that leaves the critical line gives itself away twice. In the complex plane the functional equation mirrors it into a twin; on the real line it pushes the shadow of the primes over the √x barrier. The page is a numerical illustration, not a proof: it computes finitely many points, and the general statements come from the Lean theorems.
The mirror. The functional equation ties ζ(s) to ζ(1 − s), and ζ(s) is the conjugate of ζ(s); so a zero ρ = σ + it brings along the twin 1 − ρ = (1 − σ) + it: same height, real part σ turned into 1 − σ. The equation σ = 1 − σ has the single solution σ = 1/2, so the critical line is the only place the mirror leaves fixed. Drag the amber ρ and the magenta twin stays as far away on the other side of the green line; put ρ on the green line and the two coincide.
Clearing half is enough. Every zero in the left half of the strip is the image of one in the right half. The Lean theorem rh_of_right is exactly this step: if there is no zero with Re(s) > 1/2 (apart from s = 1), the Riemann Hypothesis holds. The negative even integers −2, −4 are the trivial zeros and s = 1 is the pole; all of them lie outside the critical strip.
Impostor zeros. The L-function of an imprimitive character is the L-function of a primitive one times a few Euler factors 1 − ψ(p)p−s. The stage shows the case ψ(p) = 1: 1 − p−s vanishes at s = 2πki/ln p, rung after rung up the dashed violet line Re(s) = 0. As soon as Re(s) > 0 we have |p−s| = p−Re(s) < 1, so the factor cannot vanish: not one of these impostors gets into the critical strip (the lemma Claim.factor_ne_zero of chapter 1).
The shadow of the primes. The von Mangoldt function is ln p at a prime power pk and 0 elsewhere. Sz(x) = Σn ≤ x Λ(n)(x/n − 1) smooths the saw-teeth of the primes, and subtracting the smooth main term x ln x − (1 + γ)x leaves Gz(x). The page computes it term by term: Gz(10) ≈ 1.86080 and Gz(100) ≈ 1.72169, both close to ln 2π ≈ 1.83788. The cyan curve Gz(x) − ln 2π stays between the dashed green lines ±0.05√x all along 2 ≤ x ≤ 120.
The Landau trap. The explicit formula (a literature result) writes Gz(x) − ln 2π as one term −2 Re(xρ/(ρ(ρ − 1))) per pair of zeros. That term swings with an amplitude growing like xσ: of the order of √x when σ = 1/2; faster when σ = 0.78, so that it eventually beats any C√x; on the stage it breaks 0.05√x near x ≈ 66.89. Conversely, the Lean theorem zetaLandau says that once one side of Gz is held below C·xθ, there is no zero with Re(s) > θ. The magenta wave is a would-be sketch (MODEL); the real ζ has no such zero.