Detailed Derivations: The Riemann Zeta Compendium, Step by Step

Below is a genuine step-by-step derivations for every item, at a depth appropriate for a research paper's background section: complete proofs where they are short, detailed proof skeletons with all key steps where they are long, and precise literature pointers where a proof is deep. Items are numbered as in the compendium.

PART I — DEFINITIONS & BASIC REPRESENTATIONS

Equation 1: The Dirichlet Series

$$\zeta(s) = \sum_{n=1}^{\infty} n^{-s}, \qquad s = \sigma + it$$

Step 1 — Absolute convergence for σ > 1. Since |n^{-s}| = n^{-σ}, compare with the integral:

$$\sum_{n=1}^N n^{-\sigma} \le 1 + \int_1^N x^{-\sigma}dx = 1 + \frac{1 - N^{1-\sigma}}{\sigma - 1} \longrightarrow 1 + \frac{1}{\sigma-1} < \infty.$$

Step 2 — Uniform convergence on σ ≥ 1 + δ follows from the Weierstrass M-test with Mₙ = n^{-(1+δ)}. Hence ζ is analytic in Re s > 1 (locally uniform limits of analytic functions are analytic).

Step 3 — Divergence at s = 1 (harmonic series): group terms in blocks of length 2ᵏ, each block sums to ≥ ½. This signals the pole at s = 1 that continuation will reveal.

Equation 2: The Euler Product

$$\zeta(s) = \prod_{p} \left(1 - p^{-s}\right)^{-1}, \qquad \sigma > 1$$

Step 1 — Geometric expansion of one factor:

$$\frac{1}{1 - p^{-s}} = 1 + p^{-s} + p^{-2s} + p^{-3s} + \cdots$$

Step 2 — Multiply finitely many factors. For primes p ≤ P:

$$\prod_{p \le P}\frac{1}{1-p^{-s}} = \sum_{n \in A(P)} n^{-s},$$

where A(P) = {n : all prime factors of n are ≤ P}. This step is the Fundamental Theorem of Arithmetic: each n with factorization p₁^{a₁}···p_k^{a_k} appears exactly once when you pick the term p_i^{-a_i s} from each factor.

Step 3 — Control the tail. Every n ≤ P belongs to A(P), so

$$\left|\zeta(s) - \prod_{p\le P}\frac{1}{1-p^{-s}}\right| \le \sum_{n > P} n^{-\sigma} \longrightarrow 0 \quad (P \to \infty).$$

Step 4 — Non-vanishing corollary. A convergent infinite product of nonzero factors is nonzero:

$$\zeta(s) \ne 0 \text{ for } \sigma > 1.$$

This is the first zero-free region — the trivial one.

Step 5 — Euclid from analysis. If there were finitely many primes, the product at s = 1 would be finite while the series diverges: contradiction. Moreover, taking logs, log ζ(s) = Σ_p Σ_k p^{-ks}/k → Σ_p 1/p diverges (Euler).

Equation 3: The Gamma-Integral Representation

$$\zeta(s) = \frac{1}{\Gamma(s)}\int_0^\infty \frac{x^{s-1}}{e^x - 1}dx, \qquad \sigma > 1$$

Step 1 — Start with Gamma: Γ(s) = ∫₀^∞ t^{s-1}e^{-t}dt. Substitute t = nx (dt = n dx):

$$\Gamma(s) = n^s \int_0^\infty x^{s-1} e^{-nx}dx \quad\Longrightarrow\quad n^{-s} = \frac{1}{\Gamma(s)}\int_0^\infty x^{s-1}e^{-nx}dx.$$

Step 2 — Sum over n and swap (justified by Tonelli, all terms positive for real s > 1, then analytic continuation in s):

$$\zeta(s) = \frac{1}{\Gamma(s)}\int_0^\infty x^{s-1}\sum_{n=1}^\infty e^{-nx}\,dx.$$

Step 3 — Sum the geometric series:

$$\sum_{n=1}^\infty e^{-nx} = \frac{e^{-x}}{1 - e^{-x}} = \frac{1}{e^x - 1}. \qquad\blacksquare$$

Why this matters for RH work: this is the gateway to the contour-integral continuation. Riemann replaced ∫₀^∞ by a Hankel contour ∮ around the positive real axis to get

$$\zeta(s) = \frac{\Gamma(1-s)}{2\pi i}\oint_{\mathcal{H}} \frac{z^{s-1}}{e^z - 1}dz,$$

valid for all s ≠ 1 — his first proof of continuation and (by evaluating residues at z = 2πin) of the functional equation.

Equation 4: Hasse's Globally Convergent Series

$$\zeta(s) = \frac{1}{1-2^{1-s}}\sum_{n=0}^\infty \frac{1}{2^{n+1}}\sum_{k=0}^n (-1)^k\binom{n}{k}(k+1)^{-s}$$

Step 1 — Recognize the structure: the inner sum is the n-th forward difference Δⁿ applied to the sequence a_k = (k+1)^{-s}:

$$(\Delta^n a)_0 = \sum_{k=0}^n(-1)^k\binom{n}{k}(k+1)^{-s}.$$

Step 2 — Euler transform. For any alternating series Σ(-1)^k a_k, the Euler transformation states:

$$\sum_{k=0}^\infty (-1)^k a_k = \sum_{n=0}^\infty \frac{(\Delta^n a)_0}{2^{n+1}}.$$

Proof of the transform: write the alternating sum as a power series limit Σ(-1)^k a_k x^k at x → 1⁻, substitute x = y/(1−y) geometry, i.e., apply the binomial identity (1+x)^{-1}Σ a_k(x/(1+x))^k reorganized by differences; convergence upgrade comes from the 2^{-(n+1)} damping.

Step 3 — Apply to the eta function (Equation 5's series): η(s) = Σ(-1)^{k}(k+1)^{-s}, so

$$\eta(s) = \sum_{n=0}^\infty \frac{1}{2^{n+1}}(\Delta^n a)_0.$$

Step 4 — Convergence everywhere: the differences of the smooth function x ↦ (x+1)^{-s} decay like |Δⁿa₀| = O(n^{-σ}·stuff) and the 2^{-n} factor forces convergence for every s ∈ ℂ, giving an entire representation of η.

Step 5 — Divide by the eta-to-zeta factor (next item): ζ(s) = η(s)/(1 − 2^{1-s}). ∎

Equation 5: The Eta (Alternating) Representation

$$\zeta(s) = \frac{1}{1-2^{1-s}}\sum_{n=1}^\infty \frac{(-1)^{n-1}}{n^s}, \qquad \sigma > 0$$

Step 1 — Algebraic identity for σ > 1. Split even/odd:

$$\eta(s) = \sum_n n^{-s} - 2\sum_{n \text{ even}} n^{-s} = \zeta(s) - 2\sum_m (2m)^{-s} = \zeta(s) - 2^{1-s}\zeta(s) = (1 - 2^{1-s})\zeta(s).$$

Step 2 — η converges for σ > 0. Group in pairs: n^{-s} − (n+1)^{-s} = s∫ₙ^{n+1}x^{-s-1}dx, so the pair is O(|s| n^{-σ-1}), summable. (Formally: Dirichlet's test with bounded partial sums of (−1)^{n-1}.)

Step 3 — Continuation. Both sides analytic on σ > 0 minus the zero set of 1 − 2^{1-s} (which is s = 1 + 2πik/log 2); they agree on σ > 1, hence everywhere by the identity theorem. The pole of ζ at s = 1 appears because 1 − 2^{1-s} vanishes there while η(1) = log 2 ≠ 0:

$$\operatorname{Res}_{s=1}\zeta = \frac{\eta(1)}{\frac{d}{ds}(1-2^{1-s})\big|_{s=1}} = \frac{\log 2}{\log 2} = 1. \checkmark$$

Practical note: this is how most numerical packages evaluate ζ in the critical strip.

Equation 6: Fractional-Part Integral

$$\zeta(s) = -s\int_0^\infty \{x\}\,x^{-s-1}dx, \qquad 0 < \sigma < 1$$

Step 1 — Abel/Euler–Maclaurin summation. For σ > 1, write with ⌊x⌋ = x − {x}:

$$\sum_{n\le N} n^{-s} = \int_{1^-}^{N} x^{-s} d\lfloor x\rfloor = \lfloor x \rfloor x^{-s}\Big|_{1^-}^N + s\int_1^N \lfloor x\rfloor x^{-s-1}dx.$$

Step 2 — Let N → ∞ (σ > 1) and substitute ⌊x⌋ = x − {x}:

$$\zeta(s) = s\int_1^\infty x^{-s}dx - s\int_1^\infty \{x\}x^{-s-1}dx = \frac{s}{s-1} - s\int_1^\infty\{x\}x^{-s-1}dx.$$

This already continues ζ to σ > 0 (the integral converges there since {x} ≤ 1), exhibiting the simple pole s/(s−1) = 1 + 1/(s−1).

Step 3 — Add the (0,1) piece. For 0 < σ < 1, {x} = x on (0,1), so

$$-s\int_0^1 \{x\}x^{-s-1}dx = -s\int_0^1 x^{-s}dx = -\frac{s}{1-s} = \frac{s}{s-1}.$$

Step 4 — Combine: the s/(s−1) terms merge and

$$\zeta(s) = -s\int_0^\infty \{x\}x^{-s-1}dx. \qquad\blacksquare$$

This form makes the Nyman–Beurling criterion (#29) natural: it says ζ is essentially the Mellin transform of the fractional part, and RH becomes an approximation problem for {1/x}.

Equation 7: Stieltjes Constants (Laurent Expansion at s = 1)

$$\zeta(s) = \frac{1}{s-1} + \sum_{n=0}^\infty \frac{(-1)^n\gamma_n}{n!}(s-1)^n$$

Step 1 — ζ(s) − 1/(s−1) is entire near s = 1 (from Eq. 6's continuation), so it has a Taylor expansion; the γₙ are defined as its (signed, scaled) coefficients.

Step 2 — Extract the closed form. From Step 2 of Eq. 6:

$$\zeta(s) - \frac{1}{s-1} = 1 - s\int_1^\infty \{x\}x^{-s-1}dx.$$

Differentiate n times at s = 1 and integrate by parts against partial sums; the classical result is:

$$\boxed{\gamma_n = \lim_{m\to\infty}\left[\sum_{k=1}^m \frac{(\log k)^n}{k} - \frac{(\log m)^{n+1}}{n+1}\right]}$$

Step-by-step for n = 0 (Euler–Mascheroni): γ₀ = lim(Σ1/k − log m) = γ. For general n, apply Euler–Maclaurin to f(x) = (log x)ⁿ/x:

$$\sum_{k=1}^m f(k) = \int_1^m f(x)dx + \frac{f(1)+f(m)}{2} + \cdots, \quad \int_1^m \frac{(\log x)^n}{x}dx = \frac{(\log m)^{n+1}}{n+1},$$

and the correction terms converge as m → ∞, giving the limit.

Values: γ₀ ≈ 0.5772156649, γ₁ ≈ −0.0728158454, γ₂ ≈ −0.0096903632. These feed directly into the Li-coefficient machinery (#27–28), which is precisely what your E-011 script computed.

PART II — THE FUNCTIONAL EQUATION

Equations 8–9: The Functional Equation (Riemann's Theta-Method, Full Derivation)

This is the mechanical heart of everything, so here is the complete derivation.

Step 1 — Jacobi theta setup. Define ψ(x) = Σ_{n=1}^∞ e^{-πn²x}, so θ(x) = 1 + 2ψ(x) (Eq. 57).

Step 2 — Mellin transform of each Gaussian term. For σ > 1, substitute t = πn²x in Γ(s/2):

$$\Gamma\!\left(\frac{s}{2}\right) = \int_0^\infty t^{s/2-1}e^{-t}dt = (\pi n^2)^{s/2}\int_0^\infty x^{s/2-1}e^{-\pi n^2 x}dx.$$

Rearrange:

$$\pi^{-s/2}\Gamma\!\left(\frac{s}{2}\right)n^{-s} = \int_0^\infty x^{s/2-1}e^{-\pi n^2 x}dx.$$

Step 3 — Sum over n:

$$\pi^{-s/2}\Gamma\!\left(\frac{s}{2}\right)\zeta(s) = \int_0^\infty x^{s/2-1}\psi(x)\,dx.$$

Step 4 — Split at x = 1 and apply Jacobi's transformation. From θ(1/x) = √x θ(x) (proved below in #57):

$$\psi(1/x) = -\frac{1}{2} + \frac{\sqrt{x}}{2} + \sqrt{x}\,\psi(x).$$

On the piece ∫₀¹, substitute x → 1/x:

$$\int_0^1 x^{s/2-1}\psi(x)dx = \int_1^\infty x^{-s/2-1}\psi(1/x)dx = \int_1^\infty x^{-s/2-1}\left[\sqrt{x}\psi(x) + \frac{\sqrt{x}-1}{2}\right]dx.$$

Step 5 — Evaluate the elementary piece:

$$\int_1^\infty \frac{x^{(1-s)/2 - 1} - x^{-s/2-1}}{2}dx = \frac{1}{s-1} - \frac{1}{s} = \frac{1}{s(s-1)}.$$

Step 6 — Assemble the master formula:

$$\boxed{\;\pi^{-s/2}\Gamma\!\left(\frac{s}{2}\right)\zeta(s) = \frac{1}{s(s-1)} + \int_1^\infty\left(x^{s/2-1} + x^{(1-s)/2-1}\right)\psi(x)\,dx\;}$$

Step 7 — Read off the consequences.

• The integral converges for every s ∈ ℂ (ψ(x) = O(e^{-πx})) → entire function → analytic continuation of ζ to ℂ∖{1}.
• The right side is manifestly invariant under s ↔ 1−s → Equation 9:

$$\pi^{-s/2}\Gamma\!\left(\frac{s}{2}\right)\zeta(s) = \pi^{-(1-s)/2}\Gamma\!\left(\frac{1-s}{2}\right)\zeta(1-s).$$

Step 8 — Convert to the asymmetric form (Eq. 8). Use the Gamma reflection and duplication formulas:

$$\Gamma(z)\Gamma(1-z) = \frac{\pi}{\sin \pi z}, \qquad \Gamma(z)\Gamma\!\left(z+\tfrac12\right) = 2^{1-2z}\sqrt{\pi}\,\Gamma(2z).$$

Solve Eq. 9 for ζ(s), insert z = s/2 into both identities, and simplify (a 6-line computation):

$$\zeta(s) = 2^s \pi^{s-1}\sin\!\left(\frac{\pi s}{2}\right)\Gamma(1-s)\,\zeta(1-s). \qquad\blacksquare$$

Step 9 — Extract the trivial zeros (Eq. 15). For s = −2n: sin(−πn) = 0 while Γ(1+2n)ζ(1+2n) is finite and nonzero ⇒ ζ(−2n) = 0. At other negative points the sine's zeros are cancelled by Γ-poles — check s = 0: the pole of ζ(1−s) at s = 0 collides with nothing else and yields ζ(0) = −½ (see #44).

Equation 10: The ξ Function

$$\xi(s) = \tfrac{1}{2}s(s-1)\pi^{-s/2}\Gamma(s/2)\zeta(s)$$

Step-by-step verification of its properties:

1. Entire: the factor s kills Γ(s/2)'s pole at s = 0; the factor (s−1) kills ζ's pole at s = 1; Γ(s/2)'s remaining poles at s = −2, −4, … are killed by the trivial zeros of ζ. Nothing else is singular.
2. Symmetry ξ(s) = ξ(1−s): immediate from Eq. 9 since s(s−1) is symmetric.
3. Zeros of ξ = nontrivial zeros of ζ, exactly. Trivial zeros are consumed by Γ-poles; s = 0, 1 give ξ = ½·(pole × zero cancellations) = ξ(0) = ξ(1) = ½.
4. Order 1, maximal type refinement: |ξ(s)| ≤ exp(C|s| log|s|) from Stirling, and log ξ(σ)/(σ log σ) → ½ as σ → ∞. This growth rate is what makes the Hadamard factorization (#12) take its clean form.

Equation 11: Ξ(t) = ξ(½ + it) — Reality Proof

Step 1: ξ has real Taylor coefficients at real s (each factor does), so ξ(s̄) = ξ(s)‾. Step 2: with s = ½ + it, 1 − s = ½ − it = s̄. Then:

$$\overline{\Xi(t)} = \overline{\xi(s)} = \xi(\bar s) = \xi(1-s) = \xi(s) = \Xi(t). \blacksquare$$

Consequence — the standard RH strategy: RH ⟺ Ξ(t) has only real zeros ⟺ all sign changes of the real function Ξ account for all zeros. Hardy's theorem (#55) and all numerical verification (#56) proceed through Ξ or its normalized version Z(t).

Equation 12: The Hadamard Product

$$\xi(s) = \frac{1}{2}\prod_\rho \left(1 - \frac{s}{\rho}\right)$$

Step 1 — Hadamard factorization theorem. An entire function f of order ρ ≤ 1 with f(0) ≠ 0 factors as f(s) = e^{A+Bs}∏(1 − s/ρₙ)e^{s/ρₙ}, where the exponential convergence factors are needed because Σ1/|ρₙ| may diverge.

Step 2 — Apply to ξ (order 1, ξ(0) = ½):

$$\xi(s) = e^{A+Bs}\prod_\rho\left(1-\frac{s}{\rho}\right)e^{s/\rho}.$$

Step 3 — Determine A: set s = 0: e^A = ξ(0) = ½, so A = −log 2.

Step 4 — Determine B: logarithmic differentiation plus the symmetry ξ(s) = ξ(1−s) gives ξ'/ξ(s) = −ξ'/ξ(1−s); evaluating at s = 0 and comparing with the known ξ'(0)/ξ(0) yields:

$$B = \frac{\xi'(0)}{\xi(0)} = \log 2\pi - 1 - \frac{\gamma}{2} \approx -0.0230957\ldots$$

(Notice: B = −λ₁ — the first Li coefficient! This is not a coincidence; see #27.)

Step 5 — Pair-and-drop the convergence factors. Zeros come in pairs ρ, 1−ρ. Since Σ_ρ Re(1/ρ) converges when paired ((1/ρ) + (1/(1−ρ)) = 1/(ρ(1−ρ)), and Σ1/|ρ|² < ∞ by #17), one may group the product symmetrically; the e^{s/ρ} factors then cancel against e^{Bs} up to the constant, yielding the clean form ξ(s) = ½∏ρ(1 − s/ρ) with the product understood as lim{T→∞}∏_{|Im ρ|<T}.

Step 6 — The key corollary (used everywhere in explicit formulas):

$$\frac{\zeta'(s)}{\zeta(s)} = B + \frac{1}{2}\log\pi - \frac{1}{s-1} - \frac{1}{2}\frac{\Gamma'}{\Gamma}\!\left(\frac{s}{2}+1\right) + \sum_\rho\left(\frac{1}{s-\rho} + \frac{1}{\rho}\right).$$

Take log of ξ's definition and of the product, differentiate, solve for ζ'/ζ. Every explicit formula (#20–23) is this identity integrated against a test function.

Equations 13–14: Riemann–Siegel Theta, Z(t), and the R–S Formula

Equation 13, Step 1 — Why θ(t) exists. On the critical line write the functional equation as ζ(½+it) = χ(½+it) ζ(½−it), where χ(s) = 2^s π^{s-1}sin(πs/2)Γ(1−s). One checks |χ(½+it)| = 1, so χ = e^{-2iθ(t)} for a real phase θ. Explicitly:

$$\theta(t) = \arg\Gamma\!\left(\frac{1}{4} + \frac{it}{2}\right) - \frac{t}{2}\log\pi \quad(\text{continuous branch}, \theta(0)=0).$$

Step 2 — Stirling expansion. Apply log Γ(z) = (z−½)log z − z + ½log 2π + 1/(12z) + … at z = ¼ + it/2 and take imaginary parts:

$$\theta(t) = \frac{t}{2}\log\frac{t}{2\pi} - \frac{t}{2} - \frac{\pi}{8} + \frac{1}{48t} + \frac{7}{5760t^3} + \cdots$$

Step 3 — Z is real: Z(t) = e^{iθ(t)}ζ(½+it) satisfies Z(t)‾ = e^{-iθ}ζ(½−it) = e^{-iθ}χ^{-1}ζ(½+it) = e^{-iθ}e^{2iθ}ζ(½+it)·e^{-2iθ}… working through: Z̄ = Z. Sign changes of Z(t) = zeros of ζ on the line.

Equation 14, derivation sketch (approximate functional equation → R–S). Step 1: The approximate functional equation (Hardy–Littlewood):

$$\zeta(s) = \sum_{n\le X} n^{-s} + \chi(s)\sum_{n \le Y} n^{s-1} + O(X^{-\sigma}) + O(t^{1/2-\sigma}Y^{\sigma-1}), \quad 2\pi XY = t.$$

Step 2: Choose the symmetric point X = Y = √(t/2π), set s = ½+it, multiply by e^{iθ(t)}: the two sums become complex conjugates, giving

$$Z(t) = 2\sum_{n\le\sqrt{t/2\pi}} \frac{\cos(\theta(t) - t\log n)}{\sqrt{n}} + R(t).$$

Step 3: Riemann's saddle-point analysis of the error (rediscovered by Siegel in Riemann's Nachlass) gives R(t) = (−1)^{m+1}(t/2π)^{-1/4}·[C₀(z) + C₁(z)(t/2π)^{-1/2} + …], with z the fractional part of √(t/2π) and C₀(z) = cos(2π(z² − z − 1/16))/cos(2πz). This is the formula behind all large-scale verifications (#56): cost O(√t) per evaluation (O(t^{1/3}) with Odlyzko–Schönhage).

PART III — THE HYPOTHESIS & ZERO COUNTING

Equations 15–16: Trivial Zeros and the RH Statement

15 was derived in Step 9 of the functional-equation derivation.

16 — the precise logical content of RH. From the Euler product: no zeros in σ > 1. From the functional equation: zeros in σ < 0 are exactly the trivial ones. So all remaining zeros lie in the critical strip 0 ≤ σ ≤ 1, and they are symmetric under both s ↦ 1−s (functional equation) and s ↦ s̄ (real coefficients): zeros come in quadruples {ρ, 1−ρ, ρ̄, 1−ρ̄}, collapsing to pairs iff Re ρ = ½. RH: all such zeros satisfy Re ρ = ½.

For your paper, note carefully what is known: infinitely many zeros are on the line (#55), at least ~41% are (#55), none are off the line up to |t| ≈ 3·10¹² (#56), and no zero-free region of the form σ < 1 − ε is known for any fixed ε > 0. Any proof strategy must beat this last barrier.

Equation 17: Riemann–von Mangoldt Counting Formula

$$N(T) = \frac{T}{2\pi}\log\frac{T}{2\pi} - \frac{T}{2\pi} + \frac{7}{8} + S(T) + O(1/T)$$

Step 1 — Argument principle. Let R be the rectangle with corners 2 ± iT', −1 ± iT' (T' = T avoiding zeros). Then

$$N(T) = \frac{1}{2\pi i}\oint_{\partial R}\frac{\xi'(s)}{\xi(s)}ds = \frac{1}{2\pi}\Delta_{\partial R}\arg\xi(s),$$

using ξ (not ζ) so trivial zeros/poles don't interfere.

Step 2 — Exploit the symmetries. ξ(s) = ξ(1−s) and ξ(s̄) = ξ(s)‾ mean the four half-edges contribute equally: N(T) = (1/π)·Δ arg ξ(s) along the L-path from 2 to 2+iT to ½+iT.

Step 3 — Compute arg of each factor of ξ = ½s(s−1)π^{-s/2}Γ(s/2)ζ(s) at s = ½+iT:

• Δ arg s(s−1) = Δ arg(−T² − ¼ + …) → contributes π + O(1/T);
• Δ arg π^{-s/2} = −(T/2)log π;
• Δ arg Γ(s/2) = Im log Γ(¼ + iT/2) = (T/2)log(T/2) − T/2 − π/8 + O(1/T) (Stirling, as in #13);
• Δ arg ζ(s) =: πS(T) by definition.

Step 4 — Add and divide by π:

$$N(T) = \frac{T}{2\pi}\log\frac{T}{2\pi} - \frac{T}{2\pi} + \frac{7}{8} + S(T) + O(1/T). \blacksquare$$

(The 7/8 = 1 − 1/8 collects the π from step 3a and the −π/8 from Stirling.)

Equation 18: Bounding S(T)

Step 1 — S(T) = O(log T), unconditionally. Key lemma (from the partial-fraction formula, #12 Step 6): the number of zeros with |γ − T| ≤ 1 is O(log T), because

$$\operatorname{Re}\frac{\zeta'}{\zeta}(2+iT) = O(1) \quad\text{and}\quad \operatorname{Re}\sum_\rho\left(\frac{1}{s-\rho}+\frac{1}{\rho}\right) \ge \sum_{|\gamma-T|\le 1}\frac{c}{1+(T-\gamma)^2}.$$

Then Δ arg ζ along the horizontal segment is bounded by π·(#zeros nearby) + O(log T) via Jensen/Borel–Carathéodory applied to log ζ in a disk of radius O(1) around 2 + iT.

Step 2 — Under RH: Littlewood's argument sharpens the Borel–Carathéodory step because log ζ can be controlled in σ > ½ + 1/log log T, giving S(T) = O(log T/log log T). Known Ω-results: S(T) = Ω±(√(log T · log log log T / log log T)) — S(T) genuinely oscillates. Average: ∫₀^T S = O(log T), so S has mean 0; S(T)/√(2π² log log T) is asymptotically Gaussian (Selberg).

Equation 19: Mean Spacing — Direct Corollary

$$N(T+\delta) - N(T) \approx \frac{\delta}{2\pi}\log\frac{T}{2\pi} = 1 \iff \delta = \frac{2\pi}{\log(T/2\pi)}.$$

Differentiate #17 in T. This normalization (unfolding) γ̃ₙ = γₙ·log(γₙ/2π)/2π makes spacings mean-1 — mandatory before any GUE comparison (#51–52).

PART IV — EXPLICIT FORMULAS

Equation 20: Von Mangoldt's Explicit Formula — Full Derivation

$$\psi(x) = x - \sum_\rho\frac{x^\rho}{\rho} - \log 2\pi - \frac{1}{2}\log(1 - x^{-2})$$

Step 1 — Perron's formula. For c > 1, the discontinuous integral

$$\frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{y^s}{s}ds = \begin{cases}1 & y>1\\ \tfrac12 & y=1\\ 0 & y<1\end{cases}$$

(prove by closing the contour left for y > 1, right for y < 1; residue at s = 0 gives the 1).

Step 2 — Apply to the Dirichlet series #22. Multiply −ζ'/ζ(s) = ΣΛ(n)n^{-s} by x^s/s and integrate:

$$\psi(x) = \sum_{n \le x}\Lambda(n) = \frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\left(-\frac{\zeta'(s)}{\zeta(s)}\right)\frac{x^s}{s}ds.$$

(Swap of sum and integral justified for c > 1 by absolute convergence; x half-integer to dodge y = 1.)

Step 3 — Push the contour to Re s = −∞ and collect all residues of the integrand −(ζ'/ζ)(s)x^s/s:

Step 4 — Sum the trivial-zero residues:

$$\sum_{k=1}^\infty \frac{x^{-2k}}{-2k}\cdot(-1)\cdot(-1) = -\frac{1}{2}\sum_k\frac{(x^{-2})^k}{k} = -\frac{1}{2}\log\frac{1}{1-x^{-2}} \Rightarrow -\frac12\log(1-x^{-2})\ \text{term}.$$

Step 5 — Justify contour movement: on Re s = −(2K+1) the integrand is O(x^{-2K-1}log|s|) via the functional equation; horizontal segments are controlled by choosing heights T between zero ordinates where |ζ'/ζ| = O(log²T). The zero sum is conditionally convergent — always interpreted as lim_{T→∞}Σ_{|γ|<T}. ∎

The RH connection, explicitly: |x^ρ| = x^{Re ρ}. If RH holds, each |x^ρ/ρ| = x^{1/2}/|ρ|, and partial summation with #17 gives ψ(x) = x + O(√x log²x) — Equation 24. If any zero had Re ρ = β > ½, the single term x^β/ρ would eventually dominate the error, forcing ψ(x) − x = Ω±(x^{β−ε}). The prime error term measures the rightmost zero.

Equation 21: Riemann's π(x) Formula

Step 1: J(x) = Σ_{p^k ≤ x} 1/k (counts prime powers with weight 1/k). Then π and J are linked by J(x) = Σₙ π(x^{1/n})/n, inverted by Möbius:

$$\pi(x) = \sum_{n=1}^\infty \frac{\mu(n)}{n}J(x^{1/n}) \quad(\text{finite sum: terms vanish once } x^{1/n} < 2).$$

Step 2: From log ζ(s) = Σ_{p,k} p^{-ks}/k = s∫ J(x)x^{-s-1}dx, Mellin inversion plus the same residue computation as #20 (now with log ζ, integrating ζ'/ζ by parts) gives

$$J(x) = \operatorname{li}(x) - \sum_\rho \operatorname{li}(x^\rho) - \log 2 + \int_x^\infty\frac{dt}{t(t^2-1)\log t},$$

term-for-term: li(x) ↔ pole at 1; li(x^ρ) ↔ zeros; log 2 ↔ s = 0; the integral ↔ trivial zeros. ∎

Equation 22: −ζ'/ζ = ΣΛ(n)n^{-s} — 3-Line Proof

Log the Euler product: log ζ(s) = −Σ_p log(1 − p^{-s}). Differentiate:

$$-\frac{\zeta'}{\zeta}(s) = \sum_p \frac{p^{-s}\log p}{1 - p^{-s}} = \sum_p \log p\sum_{k\ge1}p^{-ks} = \sum_n \Lambda(n)n^{-s},$$

since Λ(n) = log p if n = p^k, else 0. ∎

Equation 23: Weil's Explicit Formula

Derivation scheme. Take f even, smooth, compactly supported; f̂(ξ) = ∫f(x)e^{iξx}dx. Consider

$$I = \frac{1}{2\pi i}\int_{(c)} \frac{\xi'(s)}{\xi(s)}\,\hat f\!\left(\frac{s - 1/2}{i}\right)ds, \quad c > 1,$$

and move the contour to Re s = 1 − c, using ξ'/ξ(s) = −ξ'/ξ(1−s).

• Residues at zeros give Σ_ρ f̂((ρ−½)/i) — the left side.
• Expanding ξ'/ξ = 1/(s-1) + 1/s − ½log π + ½(Γ'/Γ)(s/2) + ζ'/ζ: the poles at s = 0, 1 give the f̂(±1/2i) terms; the ζ'/ζ piece, expanded via #22 and Fourier inversion, gives −ΣΛ(n)n^{-1/2}[f(log n) + f(−log n)]; the Γ terms give the archimedean integral with Ψ(t) = Re ψ(¼ + it/2) − log π.

Why it matters for RH: Weil's criterion — RH ⟺ the distribution W(f) = Σ_ρ f̂(...) is positive-definite, i.e., W(f ⋆ f̃) ≥ 0 for all test f. Every "positivity" attack on RH (including the variational schemes in your E-016/E-019 scripts, and the Bombieri–Garrett framework) lives here. The known obstruction: positivity must fail for no f, but the prime side is only computable to finite precision — this is exactly the "bandwidth ceiling" phenomenon your E-020 LP quantifies.

Equations 24–25: Von Koch / Schoenfeld

24, direction 1 (RH ⇒ error bound): In #20, under RH, |Σ_{|γ|<T} x^ρ/ρ| ≤ √x Σ_{|γ|<T}1/|γ| ≈ √x·(log T)²/2π (partial summation over #17's density). The truncated explicit formula has error O(x log²(xT)/T); choose T = √x: ψ(x) = x + O(√x log²x).

24, direction 2 (error bound ⇒ RH): If ψ(x) = x + O(x^{½+ε}), then for Re s > ½ + ε,

$$-\frac{\zeta'}{\zeta}(s) = s\int_1^\infty \frac{\psi(x)}{x^{s+1}}dx = \frac{s}{s-1} + s\int_1^\infty\frac{\psi(x)-x}{x^{s+1}}dx,$$

and the last integral converges absolutely for σ > ½ + ε ⇒ ζ'/ζ analytic there apart from s=1 ⇒ no zeros in σ > ½ + ε. True for all ε ⇒ RH. ∎ (Transferring ψ ↔ π costs only the li(x) correction and log factors.)

25: Schoenfeld makes direction 1 numerically explicit by bounding Σ1/|ρ| with the known B (#12) and careful truncation constants: |π(x) − li(x)| ≤ √x log x/(8π) for x ≥ 2657 under RH. Useful in your paper for turning "RH verified to height H" into unconditional short-range prime statements.

PART V — THE EQUIVALENCES (#26–42): WHAT TO PROVE AND HOW EACH IS DERIVED

Equation 26: Mertens Criterion

(⇐) M(x) = O(x^{½+ε}) ⇒ RH: Abel summation on 1/ζ(s) = Σμ(n)n^{-s}:

$$\frac{1}{\zeta(s)} = s\int_1^\infty \frac{M(x)}{x^{s+1}}dx,$$

absolutely convergent for σ > ½ + ε ⇒ 1/ζ analytic ⇒ ζ ≠ 0 there. (⇒) Under RH, Perron + contour shift for M(x) with the zero-density input. Warning for your paper: the stronger Mertens conjecture |M(x)| ≤ √x is false (Odlyzko–te Riele 1985); lim sup M(x)/√x ≥ 1.06, lim inf ≤ −1.009. Only the ε-version is equivalent to RH.

Equations 27–28: Li's Criterion — Full Detail

Step 1 (#27, definition equivalence). With ξ(s) = ½∏(1 − s/ρ), the change of variable s = 1/(1−z) maps the half-plane Re s > ½ to the unit disk |z| < 1. Set φ(z) = ξ(1/(1−z)). Then:

$$\frac{d}{dz}\log\varphi(z) = \sum_{n=0}^\infty \lambda_{n+1}z^n, \qquad \lambda_n = \sum_\rho\left[1 - \left(1 - \frac{1}{\rho}\right)^n\right].$$

Verification: log ξ(s) = log ½ + Σ log(1 − s/ρ); substitute, expand (1 − 1/(ρ(1−z)))-type terms in powers of z, match coefficients. The derivative form λₙ = (1/(n−1)!)dⁿ/dsⁿ[s^{n−1}log ξ(s)]|_{s=1} is the same computation via Cauchy's formula.

Step 2 (the equivalence). Write ρ = ½ + iγ (γ possibly complex). Then |1 − 1/ρ| ≤ 1 ⟺ |ρ − 1| ≤ |ρ| ⟺ Re ρ ≥ ½. By the ρ ↔ 1−ρ pairing:

• RH true: every |1 − 1/ρ| = 1, points on the unit circle, and each pair contributes 2 − 2Re(1−1/ρ)ⁿ = 2(1 − cos nθ_ρ) ≥ 0, with strict positivity after summation. λₙ ≥ 0 ∀n.
• RH false: some zero has |1−1/ρ| > 1; that term grows geometrically like −|1−1/ρ|ⁿcos(n arg) and oscillates to large negative values ⇒ some λₙ < 0.

Step 3 (#28, arithmetic side). This is the Bombieri–Lagarias computation your E-011 script implements: expand log ξ(s) = log ζ + log Γ-factor + algebraic pieces at s = 1; the ζ-part produces the ηₖ (Laurent coefficients of −ζ'/ζ, themselves polynomial in Stieltjes constants), the Γ-part produces the (1−2^{-j})ζ(j) sum and the (log 4π + γ) term. See my first answer in this thread for the full chain.

Step 4 (asymptotics — the real difficulty). Under RH: λₙ ~ (n/2)(log n − log 2π + γ − 1) (Arias de Reyna / Lagarias). The two arithmetic sums in #28 each grow like 2ⁿ and cancel to leave O(n log n) — so proving λₙ > 0 arithmetically requires controlling a 2ⁿ-fold cancellation. This is the precision-budget wall your script's dps = 0.45·n + 100 reflects, and it is the known obstruction to proving RH via Li positivity.

Equations 29–30: Nyman–Beurling & Báez-Duarte

Step 1 (the mechanism). For 0 < θ ≤ 1, the Mellin transform of ρ_θ(x) = {θ/x} − θ{1/x} on (0,1):

$$\int_0^1 \rho_\theta(x)x^{s-1}dx = \frac{\theta^s - \theta}{s}\,\zeta(s) \cdot(-1)\quad(0<\sigma<1),$$

computed directly from Eq. 6 after substituting u = θ/x. Every function in the span has Mellin transform = ζ(s)·(analytic factor vanishing at the right places).

Step 2 (⇐). If the span is dense in L²(0,1), approximate the constant function 1 (Mellin transform 1/s). Mellin–Plancherel converts L² density into: ζ(s)·A(s) → 1/s in L² of the critical line for finite combinations A. If ζ had a zero ρ with Re ρ = ½… actually the criterion runs on the strip: a zero of ζ in Re s > ½ would force every ζ·A to vanish there, but H² functions approximating 1/s cannot all vanish at a common interior point (point-evaluation is continuous on Hardy space). Contradiction ⇒ no zeros off the line.

Step 3 (⇒). Harder direction (Beurling): uses the invariant-subspace structure of the multiplication semigroup on H².

Step 4 (#30, Báez-Duarte 2002). Strengthening: it suffices to use integer θ = 1/k, i.e., approximate 1 by Σ cₖ{1/(kx)}·(corrections). The distance interpretation with weight dt/(¼+t²) and conjectured rate d²_N ~ [Σ_γ 1/(¼+γ²)]/log N — the sum over zeros equals 2 + γ − log 4π ≈ 0.0461 = 2λ₁. The slow log N decay is why numerical NB experiments plateau (relevant if your paper touches this: convergence is provably no faster than C/log N).

Equations 31–32: Robin & Lagarias

Step 1 (Grönwall 1913, background): lim sup σ(n)/(n log log n) = e^γ, driven by "colossally abundant" (CA) numbers n = ∏p^{aₚ} with slowly decreasing exponents.

Step 2 (Robin, ⇒): Under RH, the explicit formula controls Chebyshev's θ(x) = Σ_{p≤x}log p well enough to bound σ(n)/n = ∏(1 − p^{-a-1})/(1 − p^{-1}) over CA numbers below e^γ log log n for n > 5040. The √x error term is exactly what's needed — the inequality margin at CA numbers is of order 1/√log n.

Step 3 (Robin, ⇐, contrapositive): if RH fails with a zero at β > ½, then θ(x) − x = Ω±(x^β) (Landau oscillation theorem), and these oscillations push σ(n)/n above e^γ log log n along a sequence of CA numbers. Hence a single counterexample n > 5040 disproves RH — and verified: no counterexample below 10^{10^{13}}+ (checks via CA enumeration).

Step 4 (Lagarias, #32): Replace e^γ log log n using the asymptotic Hₙ = log n + γ + O(1/n) and exp(Hₙ)log Hₙ ≈ e^γ n·log log n·(1 + o(1)); Lagarias shows RH ⟺ σ(n) ≤ Hₙ + e^{Hₙ}log Hₙ for all n ≥ 1 — removing the 5040 exception makes it a "clean" universal statement.

Equation 33: Redheffer Determinant

Step 1: Expand det Aₙ by the first column and use inclusion–exclusion on the divisibility poset; the answer telescopes to det Aₙ = M(n) = Σ_{k≤n}μ(k). (Neat proof: Aₙ = Cₙ + Dₙ with Dₙ unipotent from divisibility, Cₙ rank-one from the first column; det = Σμ by Möbius inversion on the incidence algebra.) Step 2: Apply #26. So RH is a statement about determinant growth of explicit 0-1 matrices — occasionally useful for spectral reformulations (the eigenvalues of Aₙ have been studied, dominant ones ≈ ±√n).

Equation 34: Speiser's Criterion (RH ⟺ ζ' ≠ 0 in 0 < σ < ½)

Mechanism, step by step: (⇒) Under RH, the partial-fraction expansion of ζ'/ζ shows Re[ζ'/ζ] < 0 strictly left of the line (each zero-pair term ρ = ½±iγ contributes negatively for σ < ½ after pairing), so ζ' ≠ 0 there. (⇐) Conversely, a zero ρ₀ off the line (say Re ρ₀ < ½ by symmetry) forces, via the argument principle applied to ζ'/ζ on a contour around the strip piece, a compensating zero of ζ' in 0 < σ < ½. Refinement (Levinson–Montgomery): N⁻(T) counts of zeros of ζ and ζ' left of the line differ by O(log T) — this duality underlies Levinson's 1/3 (#55), which counts zeros of ζ' via mollified mean values.

Equation 35: Lagarias' Positivity

$$\text{RH} \iff \operatorname{Re}\frac{\xi'}{\xi}(s) > 0 \ \ \forall\, \sigma > \tfrac12.$$

Proof in 4 lines (⇐ direction of the content). From #12:

$$\frac{\xi'}{\xi}(s) = \sum_\rho \frac{1}{s-\rho} \quad(\text{paired}), \qquad \operatorname{Re}\frac{1}{s-\rho} = \frac{\sigma - \beta}{|s-\rho|^2}.$$

If RH holds, β = ½ < σ for every ρ ⇒ every term positive ⇒ Re ξ'/ξ > 0. If RH fails, take s real approaching a zero with β > ½ from the right along suitable paths: the term 1/(s−ρ) + conjugate can be made to dominate negatively near s = β ± iγ with σ slightly less than β. ∎ This is the cleanest "herglotz/positive-real function" formulation — the one most natural for operator-theoretic (Hilbert–Pólya, de Branges) attacks.

Equation 36: Volchkov's Integral

Idea of derivation: Start from Jensen's formula / Littlewood's lemma: ∫_{σ₀}^∞ N(σ,T)dσ relates to ∫ log|ζ| on vertical lines, where N(σ,T) counts zeros with Re ρ > σ. The weight (1−12t²)/(1+4t²)³ is engineered so that the double integral computes Σ_ρ (something) whose value is forced iff all N(σ,T) = 0 for σ > ½. Carrying out the (calculus-heavy but elementary) evaluation with all zeros on the line yields exactly π(3−γ)/32; any off-line zero shifts the value. It converts RH into one verifiable real number — conceptually striking, computationally impractical (the integrand converges slowly).

Equation 37: Salem's Criterion

Mechanism: the kernel 1/(e^{x-y}+1) has Mellin/Fourier symbol Γ(s)η(s)-type factors; the integral equation ∫φ(y)dy/(e^{x−y}+1) = 0 has nontrivial L¹ solutions iff the symbol vanishes somewhere on the corresponding critical line, i.e., iff ζ has a zero off ½… precisely: injectivity of the convolution operator on the relevant line ⟺ no zeros of ζ(σ+it) for that σ; requiring it for all σ ∈ (½,1) ⟺ RH. It recasts RH as a Wiener Tauberian / completeness statement — same family as Nyman–Beurling.

Equations 38–39: Riesz & Hardy–Littlewood

Common engine — Ramanujan/Hankel duality. Consider F(x) = Σₖ (−x)ᵏ/(k!ζ(2k+2)) (H–L variant). Using 1/ζ(2k+2) = Σμ(n)n^{-2k-2} and swapping sums:

$$F(x) = \sum_n \frac{\mu(n)}{n^2}\,e^{-x/n^2}\cdot(\text{normalization}),$$

i.e., these are smoothed Möbius sums with Gaussian/exponential weights at scale √x. The Mellin transform of F is Γ(s)/ζ(2s)-type; contour shifting shows F(x) = O(x^{¼+ε}) forces 1/ζ(2s) analytic for Re s > ¼ + ε ⟺ ζ ≠ 0 for Re > ½ + 2ε. Riesz's version (#38) uses 1/ζ(2k) instead. Both are #26 in exponential-means clothing — no one has extracted new leverage from the smoothing, but the derivation is a good exercise in Mellin technique for your paper's toolkit section.

Equation 40: Franel–Landau

Step 1: The discrepancy Dₙ = Σ_ν|f_ν − ν/m| of Farey fractions connects to Möbius sums by the identity (via counting Farey points in intervals with the floor function):

$$\sum_{\nu}\left(f_\nu - \frac{\nu}{m}\right)e^{2\pi i \cdot(\cdot)} \leftrightarrow \sum_{d} M(n/d)\cdot(\text{smooth}),$$

Step 2: Möbius inversion + partial summation shows Dₙ = O(n^{½+ε}) ⟺ M(x) = O(x^{½+ε}) ⟺ RH (#26). Interpretation: RH ⟺ the rationals are as uniformly distributed as possible — the equidistribution formulation.

Equation 41: De Bruijn–Newman — Detail

Step 1 — Heat-flow setup. Write Ξ(t/2)·(normalization) = ∫₀^∞ Φ(u)cos(zu)du with Φ as displayed (this is Eq. 10's Mellin kernel after substituting x = e^{4u} in the theta-integral of Step 6, functional-equation section). Deform:

$$H_\lambda(z) = \int_0^\infty e^{\lambda u^2}\Phi(u)\cos(zu)\,du, \qquad H_0 = \tfrac{1}{8}\Xi(z/2).$$

Step 2 — Backward heat equation: ∂H/∂λ = −∂²H/∂z² (differentiate under the integral). Zeros of entire functions flow toward the real axis as λ increases (de Bruijn): if H_{λ₀} has only real zeros, so does H_λ for all λ > λ₀.

Step 3 — Definition: Λ = inf{λ : H_λ has only real zeros}; de Bruijn: Λ ≤ ½; Newman: Λ > −∞. RH ⟺ Λ ≤ 0 (since H₀ real-rooted ⟺ RH, and monotonicity).

Step 4 — Modern bounds: Rodgers–Tao (2018): Λ ≥ 0 ("RH, if true, is barely true" — the zeta zeros are at the exact boundary of the real-rootedness regime; heuristic: GUE statistics are a fixed point of the flow, and any λ < 0 would produce super-rigid statistics contradicting known correlations). Polymath 15 (2019): Λ ≤ 0.2 via explicit control of H_λ zeros to height ~10¹². For your paper: any new upper-bound technique for Λ is publishable; Λ ≤ 0 is RH itself.

Equation 42: Hilbert–Pólya / Berry–Keating

The logic, precisely: if a self-adjoint Ĥ on some Hilbert space has spectrum exactly {γₙ}, then all γₙ ∈ ℝ, i.e., RH. Evidence assembled step-by-step:

1. Montgomery–Dyson (#51): pair correlation of γₙ = pair correlation of GUE eigenvalues — the statistics of a self-adjoint operator from a unitary ensemble.
2. Odlyzko: numerically confirmed to stunning accuracy at height 10²⁰⁺.
3. Berry–Keating: the classical Hamiltonian H = xp has (regularized) semiclassical level count N(E) ≈ (E/2π)(log(E/2π) − 1) + 7/8 — matching #17 exactly, including the 7/8. The heuristic periodic orbits of xp have periods log p^k, matching the explicit-formula prime side.
4. Connes: trace formula on the noncommutative adele-class space where RH ⟺ a global trace formula identity; Selberg's trace formula proves the exact analogue (all "zeros" on the line) for hyperbolic surfaces — the strongest existing template.
5. What's missing: an actual Hilbert space + operator. Any candidate in your research must reproduce the Weil functional (#23) as its trace — the positivity W(f⋆f̃) ≥ 0 is the statement that the would-be operator is self-adjoint.

PART VI — SPECIAL VALUES

Equation 43: ζ(2k) — Step-by-Step (Two Ways)

Method A — Euler's product for sine. From sin πz = πz∏ₙ(1 − z²/n²), take log and differentiate:

$$\pi\cot\pi z = \frac{1}{z} + \sum_{n=1}^\infty\frac{2z}{z^2-n^2} = \frac{1}{z} - 2\sum_{n=1}^\infty\sum_{k=1}^\infty\frac{z^{2k-1}}{n^{2k}} = \frac{1}{z} - 2\sum_{k=1}^\infty \zeta(2k)z^{2k-1}.$$

Compare with the Bernoulli generating function πz cot πz = Σ (−1)ᵏB₂ₖ(2πz)^{2k}/(2k)! coefficient-by-coefficient:

$$\zeta(2k) = \frac{(-1)^{k+1}(2\pi)^{2k}B_{2k}}{2(2k)!}.$$

Method B — Parseval for ζ(2) (paper-friendly): Fourier series of f(x) = x on (−π,π) has coefficients bₙ = 2(−1)^{n+1}/n; Parseval gives Σ4/n² = (1/π)∫x²dx = 2π²/3 ⇒ ζ(2) = π²/6. ✓ against Method A with B₂ = 1/6.

Equation 44: Negative Values — Step by Step

ζ(−n) = −Bₙ₊₁/(n+1): insert s = −n into the Hankel-contour representation (Eq. 3 remark); the contour integral becomes the residue of z^{-n-1}·z/(e^z−1) at 0, i.e., the Bernoulli coefficient: ζ(−n) = −Bₙ₊₁/(n+1). Check: ζ(−1) = −B₂/2 = −1/12; ζ(0) = −B₁/1 = −(−½)·… careful with B₁ convention: ζ(0) = −½. Odd Bernoullis vanish ⇒ ζ(−2k) = 0 ✓ (#15 again).

ζ'(0) = −½ log 2π: differentiate the functional equation (Eq. 8) at s = 0, using ζ(1−s) ~ −1/s + γ near s = 0 and carefully expanding each factor to first order; the finite part assembles to −½log 2π. (Equivalently: "regularized ∏n = √(2π)" — the determinant identity used throughout zeta-regularization.)

Equation 45: ζ(3) — Status

No closed form known. Apéry's proof of irrationality (1979) uses the displayed central-binomial series and the recursion n³uₙ = (34n³−51n²+27n−5)u_{n−1} − (n−1)³u_{n−2} satisfied by both Σ-approximants, whose ratio converges to ζ(3) too fast (error < (√2−1)^{4n}) to be rational with bounded denominators. Known extensions: infinitely many ζ(2k+1) irrational (Ball–Rivoal); one of ζ(5),ζ(7),ζ(9),ζ(11) irrational (Zudilin).

PART VII — MOMENTS, RMT, DENSITY (#46–52)

Equations 46–47: Second and Fourth Moments

#46 sketch (Hardy–Littlewood): expand |ζ|² = Σ_{m,n}(mn)^{-½}(m/n)^{-it}; the diagonal m = n gives T·Σn^{-1} restricted ≈ T log T; off-diagonal oscillates and is bounded by stationary-phase (∫(m/n)^{it}dt = O(1/|log(m/n)|)). Full asymptotic: ∫₀^T|ζ(½+it)|²dt = T log(T/2π) − (1−2γ)T·(−1)… precisely = T log T − T(1 + log 2π − 2γ) + E(T), with E(T) = O(T^{1/3}) (Atkinson's formula territory).

#47 (Ingham): same strategy with the divisor function: |ζ|⁴ ↔ Σd(n)²n^{-1}-type diagonals, Σ_{n≤x}d(n)² ~ x log³x/π² gives the (log T)⁴/(2π²). The eighth moment is already open — this is a hard wall.

Equation 48: Keating–Snaith — the RMT Recipe

Step 1: Model ζ(½+it) at height T by the characteristic polynomial Z(U,θ) = det(I − Ue^{-iθ}) of a random U ∈ U(N), matching densities: N = log(T/2π). Step 2: Compute the CUE moment exactly with Selberg's integral:

$$\mathbb{E}_{U(N)}|Z|^{2k} = \prod_{j=0}^{N-1}\frac{j!\,(j+2k)!}{((j+k)!)^2} \sim g_k N^{k^2}, \quad g_k = \prod_{j=0}^{k-1}\frac{j!}{(j+k)!}.$$

Step 3: Primes are invisible to RMT; graft them on via the arithmetic factor a_k (the "diagonal" Euler product displayed in the compendium). Step 4: Conjecture: moment ~ a_k g_k (log T)^{k²}. Checks: k=1: g₁=1 ✓ #46; k=2: g₂ = 1/12, a₂ = 6/π² ⇒ coefficient 1/(2π²) ✓ #47. The full CFKRS "recipe" gives all lower-order terms as an N×N determinant/polynomial — the strongest quantitative evidence that the ζ–RMT dictionary is exact.

Equation 49: Lindelöf

RH ⇒ LH, step-by-step: under RH, log ζ(½+it) = O(log t/log log t) by Littlewood's Borel–Carathéodory argument (#18 Step 2), so |ζ(½+it)| = exp(O(log t/log log t)) = O(t^ε). Unconditional: convexity (Phragmén–Lindelöf between σ=0 and σ=1) gives exponent ¼; Weyl differencing gives 1/6; the current record 13/84 (Bourgain 2017) comes from decoupling theory. LH is also equivalent to a zero-density statement: N(σ,T) = O(T^{2(1−σ)+ε}) — "few zeros near σ=1" — weaker than RH but sufficient for many applications.

Equation 50: Selberg's CLT

Proof architecture: (i) Dirichlet-polynomial approximation log|ζ(½+it)| ≈ Re Σ_{p≤T^ε} p^{-½-it} valid off a small exceptional set (needs zero-density input); (ii) the terms Re p^{-it}/√p behave like independent random variables cos(t log p)/√p with variances 1/2p; (iii) Σ_{p≤X}1/2p ~ ½log log X ⇒ variance ½ log log T; (iv) moment method ⇒ Gaussian. Modern refinement (Fyodorov–Hiary–Keating): max over [T, T+2π] of log|ζ| ≈ log log T − (3/4)log log log T — extreme-value statistics of log-correlated fields; proved to second order (Arguin et al.).

Equations 51–52: Montgomery Pair Correlation — Derivation

Step 1: Define the form factor F(α,T) = N(T)^{-1}Σ_{γ,γ'≤T} T^{iα(γ−γ')}w(γ−γ') with w(u) = 4/(4+u²). Step 2: Insert the explicit formula (#23) with test functions localized at frequency α: the zero-pair sum converts to sums over prime pairs ΣΛ(n)Λ(m) with n/m ≈ T^{±α}. Step 3: For |α| < 1, only the diagonal n = m survives (off-diagonal needs prime pairs at multiplicative distance < T^{-1}, too rare below the exponent-1 threshold): ΣΛ(n)² n^{-1}-type sums give F(α) = |α| + T^{-2|α|}log T·(δ-spike at 0). This part is proved (under RH for the cleanest form). Step 4: For |α| > 1 the off-diagonal does contribute; the Hardy–Littlewood prime-pair conjecture (twin-prime constants!) predicts exactly F(α) = 1. Conjectured. Step 5: Fourier inversion of F against test functions supported in (−1,1) gives the sine-kernel pair correlation 1 − (sin πu/πu)² — Dyson's observation that this is precisely GUE. Moral for your paper: the prime side knows the zero statistics; the bandwidth restriction |α| < 1 is the same "bandwidth-one ceiling" your E-020 LP script probes — extending F beyond α = 1 unconditionally is equivalent in strength to strong prime-pair information.

PART VIII — ZERO-FREE REGIONS & PARTIAL RESULTS (#53–56)

Equation 53: The Classical Zero-Free Region — Full Proof

Step 1 — The 3-4-1 trigonometric inequality:

$$3 + 4\cos\theta + \cos 2\theta = 2(1+\cos\theta)^2 \ge 0.$$

Step 2 — Apply to log|ζ|. From the Euler product, for σ > 1: log|ζ(σ+it)| = Σ_{n} c_n n^{-σ}cos(t log n) with cₙ = Λ(n)/log n ≥ 0. Therefore:

$$3\log\zeta(\sigma) + 4\log|\zeta(\sigma+it)| + \log|\zeta(\sigma+2it)| = \sum_n \frac{c_n}{n^\sigma}\left[3 + 4\cos(t\log n) + \cos(2t\log n)\right] \ge 0,$$

i.e., ζ(σ)³|ζ(σ+it)|⁴|ζ(σ+2it)| ≥ 1.

Step 3 — Suppose a zero at 1 + it exists (β = 1). As σ → 1⁺: ζ(σ)³ ~ (σ−1)^{-3} blows up like power 3, but |ζ(σ+it)|⁴ ~ C(σ−1)⁴ vanishes like power 4 > 3, and |ζ(σ+2it)| stays bounded. Product → 0 < 1. Contradiction ⇒ ζ(1+it) ≠ 0 ⇒ (with a Tauberian theorem) the Prime Number Theorem.

Step 4 — Quantify. Suppose a zero β + iγ with β close to 1. Use the partial-fraction inequality (from #12 Step 6, all terms positive):

$$-\operatorname{Re}\frac{\zeta'}{\zeta}(\sigma+i\gamma) \le \frac{C_1}{\sigma-\beta}\cdot(-1)+ O(\log\gamma)\ \Rightarrow\ \operatorname{Re}\frac{\zeta'}{\zeta}(\sigma+i\gamma)\ \text{bounded below by}\ \frac{1}{\sigma-\beta} - C\log\gamma \text{ terms}.$$

Insert into the 3-4-1 inequality applied to −ζ'/ζ instead of log ζ:

$$\frac{4}{\sigma-\beta} \le \frac{3}{\sigma-1} + C\log\gamma.$$

Step 5 — Optimize: choose σ = 1 + δ/log γ; solving gives β ≤ 1 − c/log γ. ∎

$$\boxed{\ \zeta(\beta+i\gamma) = 0 \Rightarrow \beta \le 1 - \frac{c}{\log|\gamma|}\ }$$

The asymmetry 4 > 3 is the entire engine — and the reason the method cannot reach any fixed σ < 1: the region shrinks as γ grows. This is the 130-year-old wall.

Equation 54: Vinogradov–Korobov

Improvement mechanism: the classical region uses only |ζ(σ+it)| ≤ crude bounds. V–K instead bound ζ near σ = 1 via exponential sums Σ_{n≤N}n^{-it} = Σ e^{-it log n}, estimated by Vinogradov's mean-value theorem (now sharp via Bourgain–Demeter–Guth decoupling, 2016): |ζ(σ+it)| ≤ C t^{c(1−σ)^{3/2}}log^{2/3}t near σ = 1. Feeding this into the Step-4/5 machinery of #53 widens the region to

$$\sigma \ge 1 - \frac{c}{(\log t)^{2/3}(\log\log t)^{1/3}},$$

yielding the best unconditional PNT error: ψ(x) = x + O(x exp(−c(log x)^{3/5}(log log x)^{-1/5})). Sobering calibration for your paper: 60+ years of exponential-sum technology improved the exponent 1 → 2/3 in the log; the distance to "σ ≥ 1 − ε" — let alone ½ — remains infinite in this metric.

Equation 55: Zeros on the Line

Hardy 1914, proof scheme: consider ∫ Ξ(t)t^{2n}e^{...}dt-type moments, or the cleaner version: show ∫_T^{2T}Z(t)dt = o(∫_T^{2T}|Z(t)|dt). If Z had constant sign on [T,2T], the two would be comparable; the first moment is small (stationary phase kills it), the second is ≫ T by #46-type bounds ⇒ sign changes ⇒ zeros on the line, infinitely many.

Proportion history — the mollifier method: Selberg (positive proportion) → Levinson 1974 (34%): counts sign changes of a mollified ζ, ψζ with ψ a Dirichlet polynomial ψ = Σ_{n≤T^θ}μ(n)n^{-s}(...) chosen to dampen |ζ|'s wild oscillations; the achievable proportion is an explicit functional of the mollifier length θ (θ < ½ for Levinson) → Conrey 1989 (2/5): extended θ to 4/7 via Kloosterman-sum estimates → 41.05%+ (Pratt–Robles–Zaharescu–Zeindler 2019, multi-piece mollifiers). Also: >40% are on the line and simple. The method saturates: even optimal infinite-length mollifiers yield < 100%, so a new idea is needed for RH itself.

Equation 56: Numerical Verification — Turing's Method

How one rigorously verifies "the first N zeros are on the line": Step 1: Evaluate Z(t) (#13–14) on a fine grid; count sign changes in [0,T] — these are zeros on the line, say N_found. Step 2 — Turing's completeness test: compute N(T) exactly from #17 without locating zeros: if S(T) can be shown to be < 2 in the window via the integral bound ∫_{T}^{T+h}S(u)du = O(log T) (Turing's lemma with explicit constants), then N(T) is pinned to the nearest admissible integer. Step 3: If N_found = N(T) ⇒ no zeros off the line below T (an off-line pair would inflate N(T) by 2 without adding sign changes). Records: 10¹³ zeros (Gourdon–Demichel 2004); |γ| ≤ 3·10¹² fully rigorous with interval arithmetic (Platt–Trudgian 2021). All simple, all on the line, GUE statistics confirmed (Odlyzko at heights up to ~10²⁰⁺ for sampled batches).

PART IX — RELATED FUNCTIONS (#57–60)

Equation 57: Jacobi's Theta Transformation — Full Proof

$$\theta(1/x) = \sqrt{x}\,\theta(x)$$

Step 1 — Poisson summation. For nice f: Σ_{n∈ℤ}f(n) = Σ_{k∈ℤ}f̂(k), where f̂(k) = ∫f(y)e^{-2\pi iky}dy. Proof: F(y) = Σₙf(y+n) is 1-periodic; its k-th Fourier coefficient is ∫₀¹F e^{-2πiky} = f̂(k); evaluate the Fourier series at y = 0.

Step 2 — Gaussian self-duality. f(y) = e^{-\pi y^2 x} has f̂(k) = x^{-1/2}e^{-\pi k^2/x} (complete the square, shift the contour, use ∫e^{-πu²}du = 1).

Step 3 — Apply: Σₙe^{-πn²x} = x^{-1/2}Σₖe^{-πk²/x}, i.e., θ(x) = x^{-1/2}θ(1/x). ∎

This single identity ⇒ Eq. 9 ⇒ everything. The modularity of θ (weight-½ form on Γ₀(4)) is the prototype of the general Hecke correspondence: functional equations of L-functions ⟺ modularity of the generating object — the organizing principle of the whole Langlands program.

Equation 58: Hurwitz Zeta

ζ(s,a) = Σ(n+a)^{-s}: continuation via the same Hankel/theta methods; functional equation (Hurwitz's formula):

$$\zeta(1-s,a) = \frac{\Gamma(s)}{(2\pi)^s}\left[e^{-\pi is/2}\sum_{n\ge1}\frac{e^{2\pi ina}}{n^s} + e^{\pi is/2}\sum_{n\ge1}\frac{e^{-2\pi ina}}{n^s}\right].$$

Special cases: ζ(s,1) = ζ(s); ζ(s,½) = (2^s−1)ζ(s); Dirichlet L-functions are finite combinations L(s,χ) = q^{-s}Σ_{a=1}^q χ(a)ζ(s, a/q) — the bridge from #59 back to computable objects.

Equation 59: Dirichlet L-functions & GRH

For primitive χ mod q, the completed function Λ(s,χ) = (q/π)^{(s+𝔞)/2}Γ((s+𝔞)/2)L(s,χ) (𝔞 = 0 or 1 by parity) satisfies Λ(s,χ) = ε(χ)Λ(1−s,χ̄) with |ε| = 1 — proved by the same theta/Poisson method as #8–9, with Gauss sums τ(χ) supplying the root number. GRH (all nontrivial zeros on Re = ½) implies: primes equidistribute in progressions with √x errors (π(x;q,a) = li(x)/φ(q) + O(√x log x)), least quadratic nonresidue O(log²p), Miller–Rabin deterministic, etc. Partial unconditional substitute: Bombieri–Vinogradov ("GRH on average over q ≤ x^{1/2−ε}") — often a sufficient proxy in applications.

Equation 60: The Dirichlet-Series Dictionary — Derivations

All four follow from the Euler product by multiplicativity (Dirichlet series of multiplicative f = ∏ₚΣₖf(pᵏ)p^{-ks}):

1. Σμ(n)n^{-s} = 1/ζ(s): local factor Σₖμ(pᵏ)p^{-ks} = 1 − p^{-s} — exactly the inverse of ζ's local factor. ∎
2. Σd(n)n^{-s} = ζ(s)²: d = 1 ∗ 1 (Dirichlet convolution), and series multiply under ∗. ∎
3. Σ|μ(n)|n^{-s} = ζ(s)/ζ(2s): local: 1 + p^{-s} = (1−p^{-2s})/(1−p^{-s}). ∎
4. Σφ(n)n^{-s} = ζ(s−1)/ζ(s): φ = μ ∗ Id, so series = ζ(s−1)·(1/ζ(s)). ∎

These identities are the workhorses converting arithmetic statements into analytic ones — e.g., #26 is item 1 read through Perron's formula.

STRATEGIC SUMMARY FOR YOUR PAPER

The 60 equations organize into a directed graph of attack routes, each with a known obstruction:

A crucial caution: most equivalences (#26–40) are conservative — they translate RH without reducing its depth; any proof through them must implicitly overcome the √x-cancellation barrier in Möbius/prime sums. The consensus frontier runs through positivity structures (#23/#27/#35/#41/#42), where your scripts' numerics (Li coefficients, variational zero-flow, bandwidth-ceiling LPs) already live.