One slot per mode
A function on the circle of chapter 10 is the same thing as a taken multiples of . Such a class is determined by one number for each . This chapter proves the easy half of that statement and checks it numerically.
Based on §1.2.8 of the paper (pages 16–17), read in full. The file RH/Slots.lean proves the statement tested in the lab.
From the half-line to the circle
On the circle of chapter 10, and are the same point. A function on gives a function on the circle by summing over all the points that are identified: , over all integers .
sends some nonzero functions to 0. The simplest is a difference : the sum over telescopes. By the rescaling property its transform is , and the factor is zero exactly at , the modes of the circle. The paper states that every function sent to 0 is of this kind:
exactly when is times an . So a function on the circle corresponds to a transform modulo multiples of . Two transforms agree modulo those multiples exactly when they have the same values at the zeros of . A function on the circle is therefore determined by one number for each mode : one slot per zero of .
In Lean
The statement of Lab 12.1 is proved in full (no sorry) against :
theorem mellin_sub_scaled_eq_zero {γ : ℝ} (hγ : 1 < γ) {φ : ℝ → ℂ} (m : ℤ)
(hφ : MellinConvergent φ (2 * π * Complex.I * m / Real.log γ)) :
mellin (fun t => φ t - φ (γ * t)) (2 * π * Complex.I * m / Real.log γ) = 0
Things to try
- In Lab 12.1, sweep . The dots move and the zeros of the orange curve move with them: they come from the factor , not from the bump.
- Narrow the bump. Its transform (blue) spreads out to larger heights, and the orange curve spreads with it, but its zeros stay on the dots.
Exercises
- Show that sends to 0.
Solution
of the difference is . The second sum is the first with shifted by one, and both run over all integers, so they are equal.
- For , how far apart are the slots on the imaginary axis?
Solution
They are at , so the spacing is . That is the default setting of the lab.
- A function has . Can be 0?
Solution
No. is the slot for , and a function sent to 0 has a transform that vanishes at every slot. Directly: , which equals the of once around the circle.
Check your understanding
Two different functions on give the same function on the circle after summing over powers of . What must be true of their Mellin transforms?
- They are identical
- They agree at every slot , and may differ everywhere else
- They have nothing in common
Their difference is sent to 0, so its transform is a multiple of and vanishes at the slots. The values at the slots are precisely the Fourier coefficients of the function on the circle.
Polynomials "modulo multiples of " are determined by one number, the value at 3. What is the corresponding statement in this chapter?
- Transforms modulo multiples of are determined by their values at its zeros
- Transforms are determined by their value at
- Transforms modulo are all equal
Adding a multiple of a function cannot change the value where that function is zero, and can change it anywhere else. One slot per zero of .
Why is the orange curve in the lab zero at every dot, whatever bump you choose?
- Because the bump is symmetric
- It is a numerical coincidence
- Because it is the transform of , which is , and the first factor vanishes there
The zeros belong to the operation "subtract the rescaling", not to the function it is applied to. The same thing happens with in the next chapter.