Part II · Chapter 14 needs review

One slot per zero

Take the smooth functions and regard two as equal when they differ by a . The paper's first main claim is that an element of the resulting space is determined by one number for each zero of the . The acts on this space, and its are the values of M(φ)M(\varphi) at the zeros.

Based on §1.2.1 of the paper (pages 13–14), read in full. The identification of the quotient with the zeros rests on the paper's "Structure Theorems", which I have not read.

The quotient HH

The paper's first structural claim combines chapters 12 and 13. Let SS be the space of smooth, rapidly decaying functions on the group where Θ\Theta is defined. The θ-functions form a subspace of SS. Form the quotient space HH: elements of SS, with two regarded as equal when they differ by a θ-function.

Masullo calls 0→θ-functions→S→H→00 \to \theta\text{-functions} \to S \to H \to 0 the fundamental divisor . It has the same form as a standard fact in geometry: functions modulo those that vanish on a set of points correspond to functions on that set of points. He credits the idea to . What he claims as new is keeping SS and HH as of smooth functions, where earlier work passed immediately to square-integrable functions.

Lab 13.1 shows one direction: the transform of every θ-function is a multiple of ζ\zeta. The claim also needs the converse, that every suitable multiple of the L-function is the transform of a θ-function, and a statement about the topology of HH. The paper says this "requires all the Structure Theorems" plus known facts about L-functions (the , and control of 1/L1/L on a sequence of horizontal lines). Those theorems take up a substantial part of the 940 pages. I have not checked them.

The convolution operator on HH

The convolution A(φ)A(\varphi) of chapter 10 sends θ-functions to θ-functions, so it acts on the quotient HH. On transforms it is multiplication by M(φ)(s)M(\varphi)(s), so on the slot belonging to a zero ρ\rho it multiplies by the number M(φ)(ρ−12)M(\varphi)(\rho - \tfrac12). Its is the sum of those numbers:

trace  =  ∑ρN(ρ)⋅M(φ) ⁣(ρ−12)\text{trace} \;=\; \sum_{\rho} N(\rho)\cdot M(\varphi)\!\left(\rho-\tfrac12\right) Here N(ρ)N(\rho) is the multiplicity of the zero. This is the of the : the sum over modes in chapter 10, with zeros of ζ\zeta in place of zeros of 1−γ−s1 - \gamma^{-s}.

Lab 14.1 the eigenvalues of A(φ): the transform of φ sampled at the zeros
|M(φ)(iy)|● its value at each zero ρ = ½ + iγ of ζ: the size of one eigenvalue
The curve is the transform of the on the imaginary axis. The operator A(φ)A(\varphi) on HH has one eigenvalue for each dot. The readout adds them up over the first 30 zeros and their complex conjugates. All of these zeros are on the , so they sit at s=ρ−12=±iγs = \rho - \tfrac12 = \pm i\gamma.

The difficulty

The proof uses test functions φ\varphi that are zero for t≤1t \le 1. By property 2 of chapter 11, for a zero ρ=β+iγ\rho = \beta + i\gamma the eigenvalue M(φ)(ρ−12)M(\varphi)(\rho - \tfrac12) involves the factor ∣tρ−1/2∣=tβ−1/2|t^{\rho - 1/2}| = t^{\beta - 1/2}, which for t>1t \gt 1 grows with β\beta. So the eigenvalue is larger for a zero to the right of the critical line than for its mirror image to the left. Nothing so far shows that the sum over all zeros converges, or that it is the trace of an operator on a .

Things to try

  • Narrow φ\varphi. Its transform decays more slowly, so more zeros have eigenvalues of noticeable size.
  • Move the center of φ\varphi. The sizes of the eigenvalues barely change, but the trace does: it depends on their phases.

Exercises

  1. Why does A(φ)A(\varphi) send θ-functions to θ-functions?
    Solution

    On transforms A(φ)A(\varphi) is multiplication by M(φ)(s)M(\varphi)(s) (chapter 11, property 1). The transform of a θ-function is ζ(s+12)\zeta(s + \tfrac12) times an entire function. Multiplying by the entire function M(φ)(s)M(\varphi)(s) gives another function of that form.

  2. Suppose ρ\rho were a zero of multiplicity 2. How many numbers does an element of HH carry at ρ\rho?
    Solution

    Two: the value and the first derivative of its transform at ρ−12\rho - \tfrac12. Adding a multiple of ζ(s+12)\zeta(s + \tfrac12) changes neither, because ζ(s+12)\zeta(s + \tfrac12) and its first derivative both vanish there.

  3. Let φ\varphi be real-valued and let ρ=12+iγ\rho = \tfrac12 + i\gamma be a zero on the critical line. Show that the eigenvalues for ρ\rho and for its conjugate add up to a real number.
    Solution

    The two eigenvalues are M(φ)(iγ)M(\varphi)(i\gamma) and M(φ)(−iγ)M(\varphi)(-i\gamma). For real φ\varphi these are complex conjugates of each other, since t−iγt^{-i\gamma} is the conjugate of tiγt^{i\gamma}. Their sum is 2Re⁡M(φ)(iγ)=2∫φ(t)cos⁡(γln⁡t) dtt2\operatorname{Re} M(\varphi)(i\gamma) = 2\int \varphi(t)\cos(\gamma\ln t)\,\frac{dt}{t}.

Check your understanding

  1. The lab of the previous chapter shows that the transform of a θ-function is a multiple of ζ\zeta. Why is the converse also needed for "one slot per zero"?

    • It is not needed
    • If some multiples of ζ\zeta were not transforms of θ-functions, the quotient HH would be larger than one slot per zero
    • Because otherwise HH would be empty

    HH is SS modulo θ-functions. Dividing by a smaller subspace leaves a bigger quotient, with extra elements unrelated to the zeros. The paper's Structure Theorems are what rule that out.

  2. For a test function supported at t>1t \gt 1, compare the eigenvalue of a hypothetical zero with β=0.8\beta = 0.8 with that of its mirror image at β=0.2\beta = 0.2.

    • Equal in size
    • The one for 0.2 is larger
    • The one for 0.8 is larger, because tβ−1/2t^{\beta - 1/2} is larger for t>1t \gt 1

    The eigenvalue is M(φ)(ρ−12)M(\varphi)(\rho - \tfrac12), an integral against tρ−1/2t^{\rho - 1/2}. Zeros to the right of the line are amplified, which is why the full sum over zeros is not obviously the trace of a well-behaved operator.

  3. Why is A(φ)A(\varphi) allowed to act on the quotient HH at all?

    • Because it sends θ-functions to θ-functions, so "equal up to a θ-function" is preserved
    • Because it sends every function to a θ-function
    • Because HH is finite-dimensional

    An operator passes to a quotient exactly when it maps the subspace being divided out into itself. On transforms: multiplying a multiple of ζ\zeta by M(φ)M(\varphi) gives a multiple of ζ\zeta.