Part II · Chapter 16 needs review

One circle per prime

The factor of ζ\zeta belonging to a pp is 1/(1−p−s)1/(1 - p^{-s}), and 1−p−s1 - p^{-s} is the function whose zeros were the modes of the circle with γ=p\gamma = p. The paper uses this to split the of its into one piece per prime, each computed as in chapter 10.

Based on §1.2.9 of the paper (pages 18–21), read in full, which summarizes the construction. The construction itself is in later chapters that I have not read.

Removing one Euler factor

Take the sum over whole numbers of chapter 13, F(x)=∑n≥1f(nx)F(x) = \sum_{n \ge 1} f(nx), and subtract its rescaling by pp:

F(x)−F(px)  =  ∑p ∤ nf(nx)F(x) - F(px) \;=\; \sum_{p\,\nmid\,n} f(nx) The sum on the right runs over the nn that are not divisible by pp.

On transforms, rescaling by pp is multiplication by p−sp^{-s} (chapter 11), so the operation F(x)↦F(x)−F(px)F(x) \mapsto F(x) - F(px) multiplies the transform by 1−p−s1 - p^{-s}. That turns ζ(s)⋅M(f)(s)\zeta(s) \cdot M(f)(s) into ζ(s)(1−p−s)⋅M(f)(s)\zeta(s)(1 - p^{-s}) \cdot M(f)(s): the with the factor for pp removed. The paper calls its version of this operation PVp\mathrm{PV}_p.

The circle of a prime

The functions of the form g(x)−g(px)g(x) - g(px) are exactly the ones that Φp\Phi_p sends to 0 in chapter 12. Regarding them as zero leaves the functions on the circle (0,∞)(0, \infty) pp, with one slot at each zero of 1−p−s1 - p^{-s}. The trace of A(φ)A(\varphi) on that circle was computed in chapter 10: ln⁡p⋅∑kφ(pk)\ln p \cdot \sum_{k} \varphi(p^{k}). In the paper's normalization (the factor x\sqrt{x} in a θ\theta-function) the term for pkp^{k} is divided by pk/2p^{k/2}, and if φ\varphi is zero for t≤1t \le 1 only k≥1k \ge 1 occurs:

contribution of the prime p  =  ln⁡p⋅∑k≥1φ(pk)pk/2\text{contribution of the prime } p \;=\; \ln p \cdot \sum_{k\ge 1} \frac{\varphi(p^{k})}{p^{k/2}}

Added over all primes this is ∑nΛ(n) φ(n)/n\sum_{n} \Lambda(n)\,\varphi(n)/\sqrt{n}, with the weights Λ(n)\Lambda(n) of chapter 1. The paper relates this picture to a proposal of Deninger: a space on which each prime pp appears as a closed orbit of length ln⁡p\ln p.

Only finitely many primes contribute

If φ\varphi is zero outside [A,B][A, B] with A>1A \gt 1, a prime p>Bp \gt B has no power inside the interval, and its contribution is 0. So for a fixed , all but finitely many primes drop out. This is the same observation as "the trace is zero when φ\varphi is supported in (1,γ)(1, \gamma)" in chapter 10.

Lab 16.1 the contribution of each prime for a test function φ
● contribution of the prime p: p · φ(pk) / pk/2dashed lines: the interval outside which φ is zero
One dot for each prime up to 200. Every prime to the right of the interval contributes exactly 0. A prime to the left of the interval contributes only if one of its powers lies inside.

What the paper does with this

The paper works in the completed spaces of chapter 15. It splits the of the into one piece for each prime outside a finite set SS, plus a remainder; SS depends on the test function and on the character. For the pieces outside SS it shows that the along the diagonal is 0. The remainder involves only the finitely many primes in SS and the real numbers, and is computed with a version of an existing result, the "S-local " of and . The paper notes that the non-vanishing of the on the line Re⁡s=1\operatorname{Re} s = 1, a classical theorem, is used as an input.

The lab shows the arithmetic that the construction is designed to reproduce. It does not test the construction: the splitting of the kernel and the vanishing of the pieces are statements about operators on the completed spaces, and they are where the paper's new work lies.

Things to try

  • Narrow φ\varphi and slide its center. Dots light up one at a time as primes pass through the interval.
  • Center φ\varphi near t=8t = 8 with a small width. The prime 2 contributes through its cube, and 7 directly.
  • Move φ\varphi far to the right. More primes contribute, but each term is divided by p\sqrt{p}.

Exercises

  1. Prove that F(x)−F(px)F(x) - F(px) is the sum of f(nx)f(nx) over nn not divisible by pp.
    Solution

    F(px)=∑nf(npx)=∑f(mx)F(px) = \sum_{n} f(npx) = \sum f(mx) over the multiples m=npm = np of pp. Subtracting it from the sum over all mm leaves the mm that are not multiples of pp.

  2. φ\varphi is zero outside [2,30][2, 30]. Which values of φ\varphi appear in the contribution of the prime 2, and of the prime 31?
    Solution

    For 2: the powers inside [2,30][2, 30] are 2, 4, 8, 16, so the contribution is ln⁡2⋅(φ(2)2+φ(4)2+φ(8)8+φ(16)4)\ln 2 \cdot \left(\dfrac{\varphi(2)}{\sqrt{2}} + \dfrac{\varphi(4)}{2} + \dfrac{\varphi(8)}{\sqrt{8}} + \dfrac{\varphi(16)}{4}\right). For 31: no power is inside, so the contribution is 0.

  3. Where are the zeros of 1−5−s1 - 5^{-s}, and how far apart are they?
    Solution

    5−s=15^{-s} = 1 when sln⁡5=2πims \ln 5 = 2\pi i m: at s=2πimln⁡5s = \dfrac{2\pi i m}{\ln 5}, on the imaginary axis, spaced 2πln⁡5≈3.90\dfrac{2\pi}{\ln 5} \approx 3.90 apart. Larger primes have more closely spaced zeros.

Check your understanding

  1. Why does the argument need the test function to be zero near t=1t = 1?

    • To make φ\varphi smooth
    • Every prime has the power p0=1p^{0} = 1, so if φ(1)≠0\varphi(1) \ne 0 every prime would contribute and the reduction to finitely many primes would fail
    • Because 1 is not prime

    With support in [A,B][A, B] and A>1A \gt 1, only the powers pkp^{k} with k≥1k \ge 1 inside the interval count, and primes beyond BB have none.

  2. What plays the role of the circle's γ\gamma for the prime 7, and what is the "length" of its circle?

    • γ=7\gamma = 7, length ln⁡7\ln 7
    • γ=ln⁡7\gamma = \ln 7, length 7
    • γ=7\gamma = \sqrt{7}, length 12\tfrac12

    The factor 1−7−s1 - 7^{-s} is the circle factor with γ=7\gamma = 7. In the coordinate ln⁡t\ln t the circle has circumference ln⁡7\ln 7, which is the weight in front of its contribution.

  3. The lab adds up ln⁡p⋅φ(pk)/pk/2\ln p \cdot \varphi(p^{k})/p^{k/2}. What does that establish about the paper?

    • That the paper's splitting of the kernel is correct
    • That only finitely many primes exist
    • Only what the construction is meant to reproduce; the splitting itself is a claim about operators on the completed spaces that the lab does not touch

    The arithmetic is classical. The new and unverified part is that the operator's kernel really decomposes this way after completion, and that the pieces outside the finite set SS integrate to 0.