Part II · Chapter 18 needs review

Reading off the zeros

The last step puts one particular family of test functions into the . The side becomes the logarithmic derivative of ζ\zeta, each zero in the strip becomes a simple fraction, and the location of the on the two sides is compared.

Based on §1.4.2–1.4.5 of the paper (pages 28–30), read in full, specialized by me to ζ\zeta. Signs follow my own derivation from chapter 17.

The test functions

For a zz, let φz(t)=t−z\varphi_z(t) = t^{-z} for t>1t \gt 1 and 0 for t≤1t \le 1. Two computations:

φz\varphi_z is not smooth at t=1t = 1 and is not zero for large tt, so it is not itself an allowed . The paper approximates it by allowed ones and passes to the limit; that takes most of the three pages.

The formula for φz\varphi_z

Insert φz\varphi_z into Masullo's formula of chapter 17 with strip half-width aa:

−ζ′(z+12)ζ(z+12)  +  G(z)  =  1z−12  +  1z+12  −  ∑ρN(ρ)z−(ρ−12)-\frac{\zeta^{\prime}(z+\tfrac12)}{\zeta(z+\tfrac12)} \;+\; G(z) \;=\; \frac{1}{z-\tfrac12} \;+\; \frac{1}{z+\tfrac12} \;-\; \sum_{\rho} \frac{N(\rho)}{z-(\rho-\tfrac12)} The sum is over the zeros with ∣Re⁡ρ−12∣<a\lvert\operatorname{Re}\rho - \tfrac12\rvert \lt a.

G(z)G(z) is the term from the real numbers. It is an explicit function with no poles in the half-plane Re⁡z>0\operatorname{Re} z \gt 0.

Where the poles are

Consider the half-plane Re⁡z>a\operatorname{Re} z \gt a, and ask where each term can have a pole.

But ζ′/ζ\zeta'/\zeta has a pole at of ζ\zeta. A zero ρ\rho with Re⁡ρ>12+a\operatorname{Re}\rho \gt \tfrac12 + a would give a pole at z=ρ−12z = \rho - \tfrac12 with Re⁡z>a\operatorname{Re} z \gt a. So there is no such zero. Since a>0a \gt 0 was arbitrary, there is no zero with Re⁡ρ>12\operatorname{Re}\rho \gt \tfrac12, and by the none with Re⁡ρ<12\operatorname{Re}\rho \lt \tfrac12. That is the . The paper remarks that this argument does not use Weil's positivity criterion, the route that earlier -formula approaches aimed for.

Re z = −a0Re z = az = ½ no poles here except z = ½ a zero off the linewould be a pole here zeros on the line the z-plane, where z = s − ½
In the variable z=s−12z = s - \tfrac12 the is the imaginary axis. The formula shows that ζ′/ζ\zeta'/\zeta has no poles in the shaded half-plane except the one from the pole of ζ\zeta. A zero to the right of the strip (open circle) would be such a pole.

What a pole looks like

Near a simple zero ρ\rho, ζ′(s)/ζ(s)≈1/(s−ρ)\zeta'(s)/\zeta(s) \approx 1/(s - \rho). On a vertical line at distance dd to the right of the zero, its peaks at about 1/d1/d, at the height of the zero.

Lab 18.1 |ζ′/ζ| on the vertical line Re z = d, that is, Re s = ½ + d
|ζ′(½ + d + iy) / ζ(½ + d + iy)|● heights of the zerosdashed line: 1/d
Each peak is the pole belonging to one zero on the critical line, seen from distance dd. As dd shrinks the peaks grow like 1/d1/d. The derivative is computed by a finite difference, so values are accurate to about two decimal places.

What would have to be checked

By chapter 17 the trace formula with the restricted sum is equivalent to the hypothesis, so this last step is not where an error would hide: given the formula, the conclusion follows by the short argument above. The first two items carry the proof. Neither has been verified by anyone as of this writing, and this site has not checked them.

Things to try

  • Set d=0.1d = 0.1, then 0.05. The first peak, at height 14.13, doubles.
  • Set d=1d = 1. The line is Re⁡s=1.5\operatorname{Re} s = 1.5, where the converges; the peaks have merged into a gentle ripple.

Exercises

  1. Compute M(φz)(w)M(\varphi_z)(w) and say where the converges.
    Solution

    ∫1∞t−z tw dtt  =  [t w−zw−z]1∞  =  1z−w\int_{1}^{\infty} t^{-z}\,t^{w}\,\frac{dt}{t} \;=\; \left[\frac{t^{\,w-z}}{w-z}\right]_{1}^{\infty} \;=\; \frac{1}{z-w} provided tw−z→0t^{w - z} \to 0 as t→∞t \to \infty, that is, Re⁡z>Re⁡w\operatorname{Re} z \gt \operatorname{Re} w.

  2. Show that if ff has a zero of NN at ρ\rho, then f′/ff'/f has a simple pole there with residue NN.
    Solution

    Write f(s)=(s−ρ)Ng(s)f(s) = (s - \rho)^{N} g(s) with g(ρ)≠0g(\rho) \ne 0. Then f′f=Ns−ρ+g′g\dfrac{f'}{f} = \dfrac{N}{s - \rho} + \dfrac{g'}{g}, and g′/gg'/g is holomorphic near ρ\rho. This is why the multiplicity N(ρ)N(\rho) appears as the coefficient in the formula.

  3. Suppose the formula were known only for one value, a=0.2a = 0.2. What would it prove?
    Solution

    That ζ\zeta has no zero with Re⁡ρ>0.7\operatorname{Re}\rho \gt 0.7, and by the functional equation none with Re⁡ρ<0.3\operatorname{Re}\rho \lt 0.3. Any such zero-free region would already be a major result; the hypothesis itself needs every a>0a \gt 0.

Check your understanding

  1. Why choose the test functions φz(t)=t−z\varphi_z(t) = t^{-z}?

    • They are the smoothest possible
    • They turn the prime sum into ζ′/ζ\zeta'/\zeta and each zero's term into a fraction 1z−(ρ−12)\dfrac{1}{z - (\rho - \tfrac12)}, so the formula becomes a statement about poles
    • They vanish at the zeros

    With them, both sides are explicit functions of zz, and one can compare where each side is allowed to be infinite.

  2. A zero on the critical line also gives a pole of ζ′/ζ\zeta'/\zeta. Why does it not contradict the conclusion?

    • Its pole is at Re⁡z=0\operatorname{Re} z = 0, outside the half-plane Re⁡z>a\operatorname{Re} z \gt a the argument is about
    • Zeros on the line give no poles
    • Its pole is canceled by G(z)G(z)

    The argument says: no poles with Re⁡z>a\operatorname{Re} z \gt a except z=12z = \tfrac12. On-line zeros sit on the imaginary axis of the zz-plane, to the left of that region, and they do appear in the sum on the right.

  3. Why must the formula hold for arbitrarily small aa, not just for one value?

    • For a smaller aa the formula is easier to prove
    • It need not
    • Each aa only excludes zeros with Re⁡ρ>12+a\operatorname{Re}\rho \gt \tfrac12 + a; zeros closer to the line are excluded only by letting a→0a \to 0

    One fixed aa gives a zero-free region, not the hypothesis. The claim that the construction works for every a>0a \gt 0 is essential.