Part II · Chapter 15 needs review

The thin strip

Masullo completes the spaces of chapter 14 in a that depends on a number a>0a \gt 0. In the completed space only the zeros with ∣Re⁡ρ−12∣<a\lvert\operatorname{Re}\rho - \tfrac12\rvert \lt a keep their slots. The paper presents this step as new.

Based on §1.1.3 and §1.2.1 of the paper (pages 11–14), read in full. The paper states the norm and which zeros contribute. The explanation below of why those are the zeros that remain is my own reading; the paper's argument is in chapters I have not read.

The norm

Fix a>0a \gt 0. For a function ϕ\phi on (0,∞)(0, \infty), set

∥ϕ∥a2  =  ∫0∞∣ϕ(t)∣2⋅max⁡ ⁣(t2a, t−2a)dtt\lVert\phi\rVert_a^{2} \;=\; \int_{0}^{\infty} \lvert\phi(t)\rvert^{2}\cdot \max\!\left(t^{2a},\,t^{-2a}\right) \frac{dt}{t}

In the coordinate u=ln⁡tu = \ln t the weight is e2a∣u∣e^{2a|u|}: a function has finite norm when it decays faster than e−a∣u∣e^{-a|u|} in both directions. This is the paper's Hardy–Sobolev norm with base interval [−a,a][-a, a] and smoothness index 0, written here for (0,∞)(0, \infty) in place of the paper's group. Completing a space in this norm means adding the limits of all sequences that converge in it. The result is a . The paper says these spaces "were never considered in the literature".

The paper also gives the norm on the transform side: it is equivalent to the size of M(ϕ)M(\phi) along the two vertical lines Re⁡s=a\operatorname{Re} s = a and Re⁡s=−a\operatorname{Re} s = -a, the edges of a strip of width 2a2a.

Which slots remain

The slot at a point zz is the number M(ϕ)(z)M(\phi)(z). It remains meaningful on the completed space only if it is controlled by the norm, so that it passes to limits. Let x=Re⁡zx = \operatorname{Re} z.

The lab uses ϕT(t)=tx/max⁡(t2a,t−2a)\phi_T(t) = t^{x} / \max(t^{2a}, t^{-2a}) for 1/T≤t≤T1/T \le t \le T and 0 elsewhere, which is the function that makes the Cauchy–Schwarz inequality an equality.

Lab 15.1 is the slot at Re z = x controlled by the norm?
M(ϕT)(x) / ‖ϕT‖adashed line: the bound √(a / (a² − x²)), when x < a
For x<ax \lt a the ratio rises to the bound and stops. For x≥ax \ge a it keeps growing as TT increases: slowly at x=ax = a (like ln⁡T\sqrt{\ln T}), as a power of TT beyond it. The curve is a closed formula, given in exercise 1.

The spectral side in the completed space

Complete the three spaces of the divisor in this norm. The paper's claim is that the sequence stays exact, that A(φ)A(\varphi) acts on the completed as a -class , and that its trace is

∑∣Re⁡ρ−12∣<aN(ρ)⋅M(φ) ⁣(ρ−12)\sum_{\lvert\operatorname{Re}\rho-\frac12\rvert\lt a} N(\rho)\cdot M(\varphi)\!\left(\rho-\tfrac12\right) The sum is over the zeros with ∣Re⁡ρ−12∣<a\lvert\operatorname{Re}\rho - \tfrac12\rvert \lt a only.

So at distance aa or more from the contributes nothing to this trace. The half-width aa can be chosen as small as desired. The paper places this between two earlier approaches: in 's work every zero contributes, wherever it is, and 's "spectral interpretation" is described as the limiting case in which the strip is the critical line alone.

Removing terms from one side of an identity is only useful if the other side can still be computed. The classical (chapter 17) is a theorem, and it contains every zero. The proof therefore depends on computing the other side of the trace inside the completed space, without using the classical formula. Chapters 16 and 17 describe how the paper does that.

Things to try

  • Set a=0.3a = 0.3 and move xx from 0.2 up through 0.3. The bound rises without limit as xx approaches aa, and at x=ax = a the curve no longer levels off.
  • Set x=0.1x = 0.1, the position of a hypothetical zero at β=0.6\beta = 0.6. Find the values of aa for which its slot is lost.

Exercises

  1. For the lab's ϕT\phi_T, show that M(ϕT)(x)=∥ϕT∥a2M(\phi_T)(x) = \lVert\phi_T\rVert_a^{2} and compute it for ∣x∣<a|x| \lt a.
    Solution

    Write w(t)w(t) for the weight. Both quantities equal IT  =  ∫1/TTt2xw(t) dttI_T \;=\; \int_{1/T}^{T} \frac{t^{2x}}{w(t)}\,\frac{dt}{t} On [1,T][1, T] the integrand is t2x−2at^{2x - 2a}, and on [1/T,1][1/T, 1] it is t2x+2at^{2x + 2a}, so IT  =  1−T−2(a−x)2(a−x)  +  1−T−2(a+x)2(a+x)I_T \;=\; \frac{1-T^{-2(a-x)}}{2(a-x)} \;+\; \frac{1-T^{-2(a+x)}}{2(a+x)} The ratio in the lab is IT\sqrt{I_T}. As T→∞T \to \infty, IT  →  12(a−x)+12(a+x)  =  aa2−x2I_T \;\to\; \frac{1}{2(a-x)}+\frac{1}{2(a+x)} \;=\; \frac{a}{a^{2}-x^{2}}

  2. With a=0.1a = 0.1, which of these hypothetical zeros would keep a slot: β=0.55\beta = 0.55, β=0.6\beta = 0.6, β=0.7\beta = 0.7?
    Solution

    The slot is at β−12\beta - \tfrac12: 0.05, 0.1 and 0.2. Only the first is strictly inside the strip. The zero at β=0.6\beta = 0.6 is exactly on the edge and is lost, as is the one at 0.7.

  3. Show that max⁡(t2a,t−2a)=e2a∣u∣\max(t^{2a}, t^{-2a}) = e^{2a|u|} for u=ln⁡tu = \ln t.
    Solution

    t±2a=e±2aut^{\pm 2a} = e^{\pm 2au}. The larger of e2aue^{2au} and e−2aue^{-2au} is the one with the positive exponent, e2a∣u∣e^{2a|u|}.

Check your understanding

  1. Why is a slot at real part x≥ax \ge a "lost" in the completed space?

    • Its value is always 0
    • There are sequences that converge in the norm while the slot's value grows without bound, so the value has no meaning for the limit
    • The zero moves into the strip

    Completion adds limits of convergent sequences. A quantity survives only if it is bounded by the norm, so that it converges along with the sequence. The lab shows the ratio escaping every bound when x≥ax \ge a.

  2. As aa is made smaller and smaller, which zeros keep their slots for every choice of aa?

    • None
    • All of them
    • Only the zeros exactly on the critical line

    The condition is ∣Re⁡ρ−12∣<a\lvert\operatorname{Re}\rho - \tfrac12\rvert \lt a. Only distance 0 satisfies it for every a>0a \gt 0. Any zero off the line is dropped once aa is small enough.

  3. The classical formula is a theorem and contains every zero. Dropping the far zeros from one side of an identity proves something only if:

    • the other side is computed independently, in the completed space, and comes out unchanged
    • the dropped terms are small
    • the lab confirms it

    If the other side were taken from the classical formula, the identity with fewer zeros would simply be assumed. Everything depends on a separate computation of the geometric side.