Chapter 12 summed a function over the powers of γ, and the factor 1−γ−s appeared in front of its . Summing over the whole numbers instead puts ζ(s) in front. The resulting functions are the of the paper.
Based on §1.1.4 of the paper (page 12), read in full. The lab works over the ordinary integers with the trivial character, which is the case; the paper works over any .
θ-functions
The paper uses the construction of chapter 12 with a different sum, the one at the center of . Take a function f on (0,∞) and sum its values at all whole-number multiples of x:
Θf(x)=xn≥1∑f(nx) The paper calls functions of this form θ-functions. The transform is M(Θf)(s)=ζ(s+21)⋅M(f)(s+21)
Proof: by the rescaling property the transform of f(nx) is n−s⋅M(f)(s), and ∑n−s=ζ(s). Multiplying by x replaces s by s+21. (The sum converges for Res>21; the identity extends to all s by when Θf decays at both ends.) The factor ζ here plays the role that 1−γ−s played on the circle.
So the transform of every θ-function is zero at the zeros of ζ, whatever f is. Because of the shift by ½, ρ of ζ appears at s=ρ−21. If the holds these points are all on the imaginary axis, at s=±14.13i,±21.02i,±25.01i,…
Lab 13.1 the transform of Θ is zero at the zeros of ζ, for any f
f(x), the Θ(x) = √x · f(nx)
|M(f)(½ + iy)||M(Θ)(iy)|● zeros of zeta
the ratio |M(Θ)(iy)| / |M(f)(½ + iy)||ζ(½ + iy)|, computed independently
Top: f is a bump minus a smaller bump, scaled so that ∫fdx=0 (the paper's condition for Θ to decay at both ends). Middle: the transform of Θ, computed by integrating the orange curve above, is zero at every dot for every f. Bottom: the ratio of the two transforms equals ∣ζ(21+iy)∣. The two curves are computed independently and coincide.
Things to try
In Lab 13.1, move f around and change its width. The blue transform changes shape completely; the orange one keeps its zeros at the same heights.
Look for a setting where the blue curve itself dips close to zero between two dots. The orange curve dips there as well, but the ratio in the bottom panel does not: it depends only on ζ. (Very close to such a dip the ratio is a of two tiny numbers and gets noisy.)
Exercises
Show that the Mellin transform of f(nx) is n−s⋅M(f)(s).Solution
Substitute u=nx: ∫0∞f(nx)xsxdx=∫0∞f(u)(nu)sudu=n−s⋅M(f)(s) The measure dx/x is unchanged by rescaling.
The first zero of ζ is ρ≈21+14.13i. Where is the transform of Θf zero because of it?Solution
At s=ρ−21≈14.13i, since M(Θf)(s) contains the factor ζ(s+21). That is the first dot in the lab.
Why does Θf(x) blow up as x→0 unless ∫fdx=0?Solution
For small x the points nx are closely spaced, and x⋅∑f(nx) is a sum for ∫f. So ∑f(nx)≈x1∫f, and Θf(x)≈x−1/2∫f, which is unbounded unless the is 0.
Check your understanding
The transform of Θf vanishes at s=ρ−21 for every zero ρ, whichever f you choose. What does that tell you?
That f was chosen specially
The zeros are a property of summing over the whole numbers, not of f
That Θf is identically 0
M(Θf)(s)=ζ(s+21)⋅M(f)(s+21). The factor ζ comes from ∑n−s and is the same for every f.
What is the effect of the factor x in the definition of Θf?
It shifts s by ½, so zeros on the critical line appear at purely imaginary s
It makes the sum converge
It removes the zeros
Multiplying by x1/2 shifts the Mellin variable by ½. The critical line Res=21 becomes the imaginary axis, where the transform is an ordinary Fourier transform in lnx.
Suppose you summed f(nx) over the odd n only. Which factor would appear in front of M(f)?
ζ(s)+2−s
ζ(2s)
ζ(s)⋅(1−2−s)
The even n contribute ∑(2m)−s=2−sζ(s). Removing them leaves ζ(s)(1−2−s): the Euler product without the factor for the prime 2.