Part I · Chapter 8 needs review

Zeros as waves

The writes ψ(x)\psi(x) as xx minus one oscillating term for each zero of ζ\zeta. A zero with β\beta contributes a term of size about xβx^{\beta}, so the real parts of the zeros determine the size of ψ(x)−x\psi(x) - x.

The explicit formula

wrote down, and later proved, an exact equation between the ψ(x)\psi(x) of chapter 1 and the in the . It is called the explicit formula. Each zero is a , written ρ\rho:

ψ(x)  =  x  −  ∑ρxρρ  −  ln⁡2π  −  12ln⁡ ⁣(1−x−2)\psi(x) \;=\; x \;-\; \sum_{\rho}\frac{x^{\rho}}{\rho} \;-\; \ln 2\pi \;-\; \tfrac12\ln\!\left(1-x^{-2}\right)

Read it left to right. The first term, xx, is the smooth guess. Then comes one term for every zero. The last two terms are tiny corrections. So everything interesting is in the terms for the zeros.

Write a zero as ρ=β+iγ\rho = \beta + i\gamma. Then xρ=xβ⋅eiγln⁡xx^{\rho} = x^{\beta} \cdot e^{i\gamma \ln x}, and the zero together with its complex conjugate β−iγ\beta - i\gamma contributes a real oscillation:

Lab 8.1 rebuilding the staircase from the first K pairs of zeros
ψ(x), the true staircaseexplicit formula with K pairs of zeros
With K = 0 the formula is just the straight line. Each added zero bends the curve toward the steps. The beside each riser never goes away: a sum of smooth waves cannot make a perfectly sharp corner, the same ringing as a square wave built from sines.

What a zero off the line would do

If every zero is on the , β=12\beta = \tfrac12, then every wave has amplitude about x\sqrt{x}, and the total gap is about x\sqrt{x}: the bound from chapter 1. A single zero with β\beta bigger than 12\tfrac12 would contribute a wave of size xβx^{\beta}, which eventually drowns out all the others. This is why the and the square-root bound on the gap are the same statement.

The mirror symmetry of chapter 7 matters here. A zero to the left of the line has a partner the same distance to the right, so any zero off the line produces a wave that grows faster than x\sqrt{x}.

Lab 8.2 one zero's wave, measured in units of √x, as its real part β moves off the line
wave of the pair β ± iγ, divided by √xthe same wave with β = ½
On the line the scaled wave keeps a constant height forever. Off the line it grows without bound, however slightly β\beta exceeds 12\tfrac12. The moved zero here is hypothetical; all known zeros sit at β=12\beta = \tfrac12.

Things to try

  • In Lab 8.1, set K = 1. The single wave already has a bump near every below 20.
  • Slide the window to around 113. The flat stretch with no takes many zeros to flatten.
  • In Lab 8.2, set β=0.51\beta = 0.51. The growth is invisible at first and unmistakable by 10,000. This is why checking trillions of zeros by computer cannot settle the question.

Exercises

  1. Show that xρ+xρˉ=2xβcos⁡(γln⁡x)x^{\rho} + x^{\bar{\rho}} = 2x^{\beta} \cos(\gamma \ln x) for ρ=β+iγ\rho = \beta + i\gamma.
    Solution

    xρ=xβeiγln⁡xx^{\rho} = x^{\beta} e^{i\gamma \ln x} and xρˉ=xβe−iγln⁡xx^{\bar{\rho}} = x^{\beta} e^{-i\gamma \ln x}. Their sum is xβ⋅2cos⁡(γln⁡x)x^{\beta} \cdot 2 \cos(\gamma \ln x), because eiθ+e−iθ=2cos⁡θe^{i\theta} + e^{-i\theta} = 2 \cos \theta.

  2. By what factor must xx grow for the wave of the first zero (γ≈14.13\gamma \approx 14.13) to complete one full oscillation?
    Solution

    cos⁡(γln⁡x)\cos(\gamma \ln x) repeats when ln⁡x\ln x increases by 2π/γ≈0.4452\pi/\gamma \approx 0.445, that is, when xx is multiplied by e0.445≈1.56e^{0.445} \approx 1.56.

  3. Two zeros have the same height, one with β=0.5\beta = 0.5 and one with β=0.6\beta = 0.6. What is the ratio of their amplitudes at x=106x = 10^{6}?
    Solution

    x0.6/x0.5=x0.1=100.6≈4x^{0.6}/x^{0.5} = x^{0.1} = 10^{0.6} \approx 4. The ratio grows without bound as xx does.

Check your understanding

  1. Why is "all zeros on the line σ=12\sigma = \tfrac12" the best possible case for the size of the error?

    • Because 12\tfrac12 is the smallest real part a zero can have
    • Because zeros come in mirror pairs β\beta and 1−β1 - \beta, so the largest real part is at least 12\tfrac12, with equality only if all are on the line
    • Because waves with β=12\beta = \tfrac12 cancel each other

    The error is as large as its largest wave, xmax⁡βx^{\max \beta}. A zero with β<12\beta \lt \tfrac12 forces a partner with 1−β>121 - \beta \gt \tfrac12, so nothing can beat x\sqrt{x}.

  2. Each wave is cos⁡(γln⁡x)\cos(\gamma \ln x). How do its wiggles look near x=1,000,000x = 1{,}000{,}000 compared with near x=1,000x = 1{,}000?

    • The same width
    • Narrower
    • A thousand times wider

    The wave is periodic in ln⁡x\ln x: one cycle is a fixed ratio of xx (1.56 for the first zero). So the width in xx is proportional to xx. This is why the labs use a logarithmic axis.

  3. What do the zeros at large height add to the reconstruction of the prime staircase?

    • Fine detail: the sharpness of each individual jump
    • The overall slope
    • Nothing; only the first few zeros matter

    A large γ\gamma is a high frequency. The first few zeros give the broad shape; resolving a single jump at a prime needs zeros with periods shorter than the gap between primes.