# Simultaneous Cyclotomic-Shell Orthogonality, Cumulative Bernoulli Non-Positivity, Coprime Farey–Bernoulli Reciprocity, and Double-Zero Jordan–Totient Mellin Theory in Nyman–Beurling Space

## Abstract

We establish four exact analytical and arithmetic theorems governing the centered periodic Bernoulli generators $P_k(n) = -\tau_k(n) - \varphi(k)/2$, the cyclotomic-cotangent matrix $\widetilde{\mathcal{C}}_N$, and the totient-neutral basis $Q_k = -\tau_k + \varphi(k)\tau_2$ in Nyman–Beurling–Báez-Duarte space $\mathcal{H} = L^2((1, \infty), x^{-2}\,dx)$. First, we prove that the unit-cell jump $\Delta_n\tau_k$ is Ramanujan's sum $c_k(n)$, so $P_k(n) = -\frac{1}{2}\sum_{\gcd(a, k)=1}\sin(\pi a(2n+1)/k)/\sin(\pi a/k)$ is supported on the primitive Farey shell $\Omega_k$. Consequently, $\{P_k\}$, $\{\overline{P}_k\}$, $\{\Delta_n P_k\}$, and the centered cumulative sums $\{\widetilde{S}_k(n)\}$ ($S_k(n) = \sum_{m=0}^{n-1} P_k(m) \le 0$) are simultaneously mutually orthogonal across $j \neq k$, yielding a positive series and $O(M^{-3})$ rational brackets for $\Lambda(k)$. Second, we prove a coprime Farey–Bernoulli reciprocity formula for $\widetilde{\mathcal{C}}_{j, k}$, the single-shell identity $\widetilde{\mathcal{C}}_{2, k} = -\sum_{1 \le b < k/2, \gcd(b, k)=1}(b/k)\cot(\pi b/k) < 0$, and a three-level trigonometric-and-rational hierarchy proving $\overline{D}_{j, k} \in \mathbb{Q}$ with closed forms for $\overline{D}_{2, k}$ and $\overline{D}_{p^\alpha, p^\alpha}$. Third, we construct a pairwise-periodic third-order Euler–Maclaurin rational functional with positive cubic-vanishing prefix weights and $O(M^{-3})$ exact rational primal-dual certificates for $d_N^2$. Fourth, we prove that $\widehat{Q}_k(s) = \frac{\zeta(s)}{s}F_k(s)$ satisfies $F_k(0) = F_k(1) = 0$, removing both strip-endpoint poles in the Mellin–Plancherel integrand and characterizing $V_N^{\mathrm{cen}}$ by entire Mellin continuation. These results do not resolve RH.

---

## 1. Introduction and Setup

In the Nyman–Beurling–Báez-Duarte (NBBD) Hilbert space $\mathcal{H} := L^2\bigl((1, \infty), x^{-2}\,dx\bigr)$ [1–3], the Riemann Hypothesis (RH) is equivalent to $\lim_{N \to \infty} d_N^2 = 0$, where $d_N := \inf_{g \in V_N}\|1 - g\|_{\mathcal{H}}$ denotes the distance from the constant target $1$ to the natural-dilation subspace $V_N := \operatorname{span}\{\rho_d : 2 \le d \le N\}$ generated by $\rho_d(x) := \lfloor x/d \rfloor - \frac{1}{d}\lfloor x \rfloor$. Passing via Möbius inversion on $\{1, \dots, N\}$ to the integer-periodic step basis [9, 13–15]
$$\tau_k(x) \;:=\; \sum_{d \mid k} d\,\mu\!\left(\frac{k}{d}\right)\left\lfloor \frac{x}{d} \right\rfloor \qquad (k \ge 1)$$
diagonalizes integer periodicity: $\tau_1(x) = \lfloor x \rfloor$, and for every $k \ge 2$, $\tau_k(x) \equiv \tau_k(\lfloor x \rfloor)$ is purely $k$-periodic on $[0, \infty)$ with $\tau_k(0) = 0$, $\tau_k(k - 1) = -\varphi(k)$, and $V_N = \operatorname{span}\{\tau_k : 2 \le k \le N\}$. Introducing Euler's totient $\varphi(k)$, Jordan's totients $J_r(k) := k^r\prod_{p \mid k}(1 - p^{-r})$, and the centered $k$-periodic Bernoulli sequence
$$P_k(n) \;:=\; -\tau_k(n) - \frac{1}{2}\varphi(k) \;\equiv\; \sum_{d \mid k} d\,\mu\!\left(\frac{k}{d}\right)\left(\left\{\frac{n}{d}\right\} - \frac{1}{2}\right) \qquad (k \ge 2,\; n \in \mathbb{Z}_{\ge 0}), \tag{1}$$
every approximation residual $1 - g(n)$ ($g = \sum_{k=2}^N c_k\tau_k \in V_N$) decomposes on unit cells $[n, n+1)$ as
$$r_n(\mathbf{c}) \;:=\; 1 - g(n) \;\equiv\; \beta(\mathbf{c}) \;+\; \sum_{k=2}^N c_k\,P_k(n), \qquad \beta(\mathbf{c}) \;:=\; 1 + \frac{1}{2}\sum_{k=2}^N c_k\,\varphi(k), \tag{2}$$
with $r_0(\mathbf{c}) = 1 - g(0) \equiv 1$. In our companion paper `riemann_hypothesis_56` [15], we proved that for all $j, k \ge 2$ ($L := \operatorname{lcm}(j, k)$), the infinite-horizon Gram entry splits into a rank-2 totient–von Mangoldt logarithmic form plus a logarithm-free cyclotomic-cotangent entry:
$$\langle \tau_j,\, \tau_k \rangle_{\mathcal{H}} \;\equiv\; \frac{1}{2}\bigl(\varphi(j)\Lambda(k) + \varphi(k)\Lambda(j)\bigr) \;+\; \pi\,\widetilde{\mathcal{C}}_{j, k}, \qquad \widetilde{\mathcal{C}}_{j, k} \;:=\; \frac{\sin(\pi/L)}{2 L}\sum_{n=1}^{L-2} \frac{P_j(n)P_k(n) - \frac{1}{4}\varphi(j)\varphi(k)}{\sin(\pi n/L)\,\sin(\pi(n+1)/L)}, \tag{3}$$
and that on the centered totient hyperplane $V_N^{\mathrm{cen}} := \{g \in V_N : \beta(\mathbf{c}) = 0\}$, eliminating $c_2 = -2(1 - \beta) - \sum_{k=3}^N \varphi(k)c_k$ writes $1 - g \equiv (1 - \beta)Q_0 + \beta\cdot 1 + \sum_{k=3}^N c_k Q_k$ in terms of $Q_0 := 1 + 2\tau_2$ ($\|Q_0\|_{\mathcal{H}}^2 = 1$) and the totient-neutral basis $Q_k := -\tau_k + \varphi(k)\tau_2 = P_k - \varphi(k)P_2$ ($k \ge 3$), whose Gram matrix $\langle Q_j, Q_k \rangle_{\mathcal{H}} = \pi\,\widehat{\mathcal{C}}_{j, k}$ ($\widehat{\mathcal{C}}_{j, k} := \widetilde{\mathcal{C}}_{j, k} - \varphi(j)\widetilde{\mathcal{C}}_{2, k} - \varphi(k)\widetilde{\mathcal{C}}_{2, j}$) is completely free of prime logarithms.

Despite this progress, four fundamental questions remained open:
1. *Why do $\{P_k\}_{k \ge 2}$ decouple across distinct moduli on finite periods, and what governs their cumulative sums $S_k(n) := \sum_{m=0}^{n-1} P_k(m)$?*
2. *How does $\widetilde{\mathcal{C}}_{j, k}$ decompose across primitive Farey fractions, what is the sign of the deflation vector $\widetilde{\mathcal{C}}_{2, k}$ ($k \ge 3$), and why do higher-order periodic Euler–Maclaurin averages remain rational?*
3. *Can transcendental-free rational certificates for $d_N^2$ (`fractions.Fraction`), previously restricted to $O(M^{-2})$ width in `riemann_hypothesis_53` [14] to avoid the exponential global period $L_N = \operatorname{lcm}(1, \dots, N)$, be upgraded to $O(M^{-3})$ width in $O(N^3 + N^2 M)$ operations?*
4. *On the complex-analytic Mellin–Plancherel side, what mechanism cancels the $\zeta'/ \zeta$ logarithmic poles in $\langle Q_j, Q_k \rangle_{\mathcal{H}} = \pi\,\widehat{\mathcal{C}}_{j, k}$ and distinguishes $V_N^{\mathrm{cen}}$ from $V_N$?*

In this paper we resolve all four questions with self-contained proofs and exact rational verification:
- **Section 2 (Theorem 1):** Proves the Ramanujan-sum jump law $\Delta_n\tau_k \equiv c_k(n)$, the primitive Farey-shell sine expansion of $P_k(n)$, the simultaneous mutual orthogonality of $\{P_k\}$, $\{\overline{P}_k\}$, $\{\Delta_n P_k\}$, and $\{\widetilde{S}_k\}$, the universal non-positivity $S_k(n) \le 0$, and $O(M^{-2})$ monotone and $O(M^{-3})$ Euler–Maclaurin rational brackets for $\Lambda(k)$ (Table 1).
- **Section 3 (Theorem 2):** Establishes the coprime Farey–Bernoulli reciprocity formula for $\widetilde{\mathcal{C}}_{j, k}$, the single-shell formula $\widetilde{\mathcal{C}}_{2, k} = -\sum_{1 \le b < k/2, \gcd(b, k)=1}\frac{b}{k}\cot(\frac{\pi b}{k}) < 0$, and a three-level trigonometric-and-rational hierarchy proving $\overline{D}_{j, k} \in \mathbb{Q}$ with closed forms for $\overline{D}_{2, k}$ and $\overline{D}_{p^\alpha, p^\alpha}$.
- **Section 4 (Theorem 3):** Proves the pairwise-periodic third-order Euler–Maclaurin identity $\overline{\Psi}(\mathbf{c}) = -\beta(\mathbf{c})\sum_{k=2}^N \frac{J_2(k)}{12}c_k + \sum_{j, k=2}^N c_j c_k\overline{D}_{j, k} \in \mathbb{Q}[\mathbf{c}]$ (requiring only $L_{j, k} = \operatorname{lcm}(j, k) \le N^2$), positive cubic-vanishing prefix weights $w_r^{\mathrm{EM3}} > 0$, and $O(M^{-3})$ exact `fractions.Fraction` primal-dual certificates for $d_N^2$ up to $(N, M) = (40, 6400)$ (Table 2).
- **Section 5 (Theorem 4):** Establishes the double-zero Jordan–Totient Mellin identity $F_k(0) = F_k(1) = 0$ for $\widehat{Q}_k(s) = \frac{\zeta(s)}{s}F_k(s)$ ($k \ge 3$), proving removable singularities at both strip endpoints $s = 0, 1$, characterizing $V_N^{\mathrm{cen}}$ by entire Mellin continuation of $1 - g$, and deriving exact Hilbert-space projection formulas for $B_N$ and $A_N$ (Tables 3–4).

---

## 2. Simultaneous Cyclotomic-Shell Orthogonality and Cumulative Bernoulli Non-Positivity

For each integer $k \ge 2$, let $\Omega_k := \{a/k \in (0, 1) : \gcd(a, k) = 1\}$ denote the primitive Farey shell of denominator $k$ ($|\Omega_k| = \varphi(k)$, with $\Omega_j \cap \Omega_k = \emptyset$ for all $j \neq k$), and let $c_k(n) := \sum_{\gcd(a, k)=1} e^{2\pi i a n / k} = \sum_{d \mid \gcd(k, n)} d\,\mu(k/d)$ denote Ramanujan's trigonometric sum [19, §27.10]. Define the midpoint Dedekind–Rademacher average $\overline{P}_k(n) := \frac{1}{2}(P_k(n) + P_k(n-1))$ ($n \ge 1$) and the cumulative Bernoulli sum $S_k(n) := \sum_{m=0}^{n-1} P_k(m)$ ($n \ge 0$, with $S_k(0) = 0$).

**Theorem 1 (Primitive Farey-Shell Expansion, Simultaneous Four-Tier Orthogonality, and Cumulative Non-Positivity).**
*Let $j, k \ge 2$ be integers and $L := \operatorname{lcm}(j, k)$.*

*(a) **Ramanujan-Sum Jump Law and Primitive Farey-Shell Sine Expansion:** For every $k \ge 1$ and $n \ge 1$, $\Delta_n\tau_k := \tau_k(n) - \tau_k(n-1) \equiv c_k(n)$. For every $k \ge 2$ and $n \ge 0$, $P_k(n)$ and $\overline{P}_k(n)$ have exact finite Fourier expansions supported exclusively on $\Omega_k$:*
$$P_k(n) \;\equiv\; -\frac{1}{2}\sum_{\alpha \in \Omega_k} \frac{\sin\bigl(\pi\alpha(2n + 1)\bigr)}{\sin(\pi\alpha)}, \qquad \overline{P}_k(n) \;\equiv\; -\frac{1}{2}\sum_{\alpha \in \Omega_k} \cot(\pi\alpha)\sin(2\pi\alpha n) \;\equiv\; P_k(n) + \frac{1}{2}c_k(n). \tag{4}$$

*(b) **Cumulative Bernoulli Non-Positivity and Simultaneous Four-Tier Orthogonality:** For every $k \ge 2$, $S_k(n)$ is $k$-periodic on $\mathbb{Z}_{\ge 0}$ and non-positive everywhere:*
$$S_k(n) \;\equiv\; -\frac{1}{2}\sum_{\alpha \in \Omega_k} \frac{\sin^2(\pi\alpha n)}{\sin^2(\pi\alpha)} \;\le\; 0, \qquad S_k(n) = 0 \iff k \mid n, \qquad S_k(n) \;\le\; -\frac{1}{2}\varphi(k) \quad (\forall\, \gcd(n, k) = 1), \tag{5}$$
*with periodic mean $\frac{1}{k}\sum_{n=0}^{k-1} S_k(n) \equiv -\frac{J_2(k)}{12}$. Setting $\widetilde{S}_k(n) := S_k(n) + \frac{J_2(k)}{12}$, all four families $\{\Delta_n P_k\}$, $\{P_k\}$, $\{\overline{P}_k\}$, and $\{\widetilde{S}_k\}$ are simultaneously mutually orthogonal over any common period $L = \operatorname{lcm}(j, k)$:*
$$\frac{1}{L}\sum_{n=1}^{L} (\Delta_n P_j)(\Delta_n P_k) = \varphi(k)\,\delta_{j, k}, \qquad \frac{1}{L}\sum_{n=0}^{L-1} P_j(n)P_k(n) = \frac{J_2(k)}{12}\,\delta_{j, k}, \tag{6}$$
$$\frac{1}{L}\sum_{n=1}^{L} \overline{P}_j(n)\overline{P}_k(n) = \frac{J_2(k) - 3\varphi(k)}{12}\,\delta_{j, k}, \qquad \frac{1}{L}\sum_{n=0}^{L-1} \widetilde{S}_j(n)\widetilde{S}_k(n) = \frac{J_4(k) + 10 J_2(k)}{720}\,\delta_{j, k}. \tag{7}$$

*(c) **Positive-Term Series and Rational Brackets for $\Lambda(k)$:** For every $k \ge 2$,*
$$\Lambda(k) \;\equiv\; \frac{3}{4}\varphi(k) \;-\; 2\sum_{n=1}^{\infty} \frac{|S_k(n + 1)|}{n(n + 1)(n + 2)}. \tag{8}$$
*Defining $s_k^{\max} := \max_{0 \le n < k}|S_k(n)| \in \frac{1}{2}\mathbb{Z}_{>0}$, the reflected cumulative sequence $\widetilde{U}_k(n) := \frac{J_2(k)}{24} + \sum_{m=1}^{n}\widetilde{S}_k(m)$ ($0 \le n < k$, satisfying $\widetilde{U}_k(k - 1 - n) \equiv -\widetilde{U}_k(n)$ and $\widetilde{U}_k(n+1) - \widetilde{U}_k(n) \equiv \widetilde{S}_k(n+1)$ for all $n \ge 0$ when extended $k$-periodically), and $u_k^{\max} := \max_{0 \le n < k}|\widetilde{U}_k(n)| \in \mathbb{Q}_{>0}$, every integer cutoff $M \ge 2$ yields $O(M^{-2})$ monotone and $O(M^{-3})$ Euler–Maclaurin rational brackets:*
$$L_{k, M}^{\mathrm{mon}} - \frac{s_k^{\max}}{M(M + 1)} \;\le\; \Lambda(k) \;\le\; L_{k, M}^{\mathrm{mon}} \;:=\; \frac{3}{4}\varphi(k) + 2\sum_{n=1}^{M-1}\frac{S_k(n + 1)}{n(n + 1)(n + 2)}, \tag{9}$$
$$\left|\Lambda(k) \;-\; \left(L_{k, M}^{\mathrm{mon}} - \frac{J_2(k)}{12 M(M + 1)} - \frac{2\,\widetilde{U}_k(M)}{M(M + 1)(M + 2)}\right)\right| \;\le\; \frac{2\,u_k^{\max}}{M(M + 1)(M + 2)}. \tag{10}$$

*Proof.*
**(a)** Since $\lfloor n/d \rfloor - \lfloor (n-1)/d \rfloor = \mathbf{1}_{d \mid n}$, we have $\Delta_n\tau_k = \sum_{d \mid k} d\,\mu(k/d)\mathbf{1}_{d \mid n} = \sum_{d \mid \gcd(k, n)} d\,\mu(k/d) = c_k(n)$ [19, Eq. 27.10.3]. Summing $\tau_k(n) = \sum_{m=1}^n c_k(m) = \sum_{\alpha \in \Omega_k}\sum_{m=1}^n e^{2\pi i \alpha m}$ via $\sum_{m=1}^n e^{2\pi i \alpha m} = -\frac{1}{2} + \frac{e^{i\pi\alpha(2n+1)}}{2i\sin(\pi\alpha)}$ and pairing $\alpha$ with $1 - \alpha$ gives $\tau_k(n) = -\frac{1}{2}\varphi(k) + \frac{1}{2}\sum_{\alpha \in \Omega_k}\frac{\sin(\pi\alpha(2n+1))}{\sin(\pi\alpha)}$, proving (4) since $\overline{P}_k(n) = P_k(n) + \frac{1}{2}c_k(n)$ and $\frac{1}{2}(\sin(\pi\alpha(2n+1)) + \sin(\pi\alpha(2n-1))) = \sin(2\pi\alpha n)\cos(\pi\alpha)$.

**(b)** Summing $\sin(\pi\alpha(2m + 1)) = \frac{\cos(2\pi\alpha m) - \cos(2\pi\alpha(m+1))}{2\sin(\pi\alpha)}$ over $0 \le m \le n - 1$ telescopes to $\frac{\sin^2(\pi\alpha n)}{\sin(\pi\alpha)}$, proving $S_k(n) = -\frac{1}{2}\sum_{\alpha \in \Omega_k}\frac{\sin^2(\pi\alpha n)}{\sin^2(\pi\alpha)} \le 0$. Clearly $S_k(n) = 0$ if $k \mid n$; if $k \nmid n$, the $\alpha = 1/k$ term is strictly negative so $S_k(n) < 0$. When $\gcd(n, k) = 1$, $a \mapsto a n \pmod k$ permutes $(\mathbb{Z}/k\mathbb{Z})^\times$, so $\{\sin^2(\pi\alpha n)\}_{\alpha \in \Omega_k}$ is a permutation of $\{\sin^2(\pi\alpha)\}_{\alpha \in \Omega_k}$, and the rearrangement inequality $\sum_i x_{\pi(i)} / x_i \ge \sum_i 1 = \varphi(k)$ gives $S_k(n) \le -\frac{1}{2}\varphi(k)$ (moreover, $S_k(n) \le -\frac{1}{2}\varphi(k)$ holds across all $k \nmid n$ for $2 \le k \le 120$ by direct evaluation). Using $\sin^2(\pi\alpha n) = \frac{1 - \cos(2\pi\alpha n)}{2}$ and pairing $\alpha$ with $1 - \alpha$ yields
$$P_k(n) = \sum_{\alpha \in \Omega_k}\frac{e^{i\pi\alpha}}{-2i\sin(\pi\alpha)}e^{2\pi i \alpha n}, \quad \Delta_n P_k = -\sum_{\alpha \in \Omega_k} e^{2\pi i \alpha n}, \quad \overline{P}_k(n) = \sum_{\alpha \in \Omega_k}\frac{\cot(\pi\alpha)}{-2i}e^{2\pi i \alpha n}, \quad \widetilde{S}_k(n) = \frac{1}{4}\sum_{\alpha \in \Omega_k}\frac{e^{2\pi i \alpha n}}{\sin^2(\pi\alpha)}, \tag{11}$$
where the periodic mean of $S_k(n)$ is $-\frac{1}{4}\sum_{\alpha \in \Omega_k}\sin^{-2}(\pi\alpha) = -\frac{J_2(k)}{12}$ because $\sum_{a=1}^{d-1}\sin^{-2}(\pi a/d) = \frac{d^2 - 1}{3}$ and Möbius inversion over $d \mid k$ ($k \ge 2$) gives $\sum_{\alpha \in \Omega_k}\sin^{-2}(\pi\alpha) = \sum_{d \mid k}\mu(k/d)\frac{d^2 - 1}{3} = \frac{J_2(k)}{3}$. Likewise, applying Möbius inversion to $\sum_{a=1}^{d-1}\sin^{-4}(\pi a/d) = \frac{d^4 + 10 d^2 - 11}{45}$ gives $\sum_{\alpha \in \Omega_k}\sin^{-4}(\pi\alpha) = \frac{J_4(k) + 10 J_2(k)}{45}$. Since $\Omega_j \cap \Omega_k = \emptyset$ for all $j \neq k$ and $\{e^{2\pi i \alpha n}\}_{\alpha \in \Omega_k}$ are orthonormal on $\mathbb{Z}/L\mathbb{Z}$, Parseval's identity applied to (11) yields (6)–(7).

**(c)** By [15, Theorem 1(c)], $-\Lambda(k) = \langle 1, \tau_k \rangle_{\mathcal{H}} = \sum_{n=1}^\infty \frac{\tau_k(n)}{n(n+1)} = -\frac{1}{2}\varphi(k) - \sum_{n=1}^\infty \frac{P_k(n)}{n(n+1)}$. Writing $P_k(n) = S_k(n+1) - S_k(n)$ and summing by parts using $S_k(1) = P_k(0) = -\frac{1}{2}\varphi(k)$ and $\frac{1}{(n-1)n} - \frac{1}{n(n+1)} = \frac{2}{(n-1)n(n+1)}$ gives $\sum_{n=1}^\infty \frac{P_k(n)}{n(n+1)} = -\frac{S_k(1)}{2} + 2\sum_{n=1}^\infty \frac{S_k(n+1)}{n(n+1)(n+2)} = \frac{1}{4}\varphi(k) - 2\sum_{n=1}^\infty \frac{|S_k(n+1)|}{n(n+1)(n+2)}$, proving (8). Bound (9) follows from $-s_k^{\max} \le S_k(n+1) \le 0$ and $2\sum_{n=M}^\infty \frac{1}{n(n+1)(n+2)} = \frac{1}{M(M+1)}$. Finally, since $\widetilde{S}_k(k - m) = \widetilde{S}_k(m)$ and $\sum_{m=0}^{k-1}\widetilde{S}_k(m) = 0$ with $\widetilde{S}_k(0) = \frac{J_2(k)}{12}$, we have $\widetilde{U}_k(k - 1 - n) + \widetilde{U}_k(n) = \frac{J_2(k)}{12} + \sum_{m=1}^{k-1}\widetilde{S}_k(m) = \sum_{m=0}^{k-1}\widetilde{S}_k(m) = 0$. Substituting $S_k(n+1) = -\frac{J_2(k)}{12} + \widetilde{U}_k(n+1) - \widetilde{U}_k(n)$ for $n \ge M$ and summing by parts via $\frac{2}{(n-1)n(n+1)} - \frac{2}{n(n+1)(n+2)} = \frac{6}{(n-1)n(n+1)(n+2)}$ and $6\sum_{n=M+1}^\infty \frac{1}{(n-1)n(n+1)(n+2)} = \frac{2}{M(M+1)(M+2)}$ yields (10). $\blacksquare$

| Modulus $k$ | $\varphi(k),\, J_2(k),\, J_4(k)$ | $s_k^{\max},\, u_k^{\max}$ | Exact $\Lambda(k)$ | Monotone Bracket $[L_{k, 60}^{\mathrm{mon}} - \frac{s_k^{\max}}{3660},\, L_{k, 60}^{\mathrm{mon}}]$ ($M = 60$) | 3rd-Order Rational Bracket (10) at $M = 60$ (`fractions.Fraction`) | EM3 Width |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $2$ | $1,\, 3,\, 15$ | $1/2,\, 1/8$ | $0.6931471806$ | $[0.6930800033,\, 0.6932166154]$ | $[0.6931461059,\, 0.6931483094]$ | $2.20 \times 10^{-6}$ |
| $3$ | $2,\, 8,\, 80$ | $1,\, 1/3$ | $1.0986122887$ | $[1.0985241984,\, 1.0987974225]$ | $[1.0986093973,\, 1.0986152732]$ | $5.88 \times 10^{-6}$ |
| $4$ | $2,\, 12,\, 240$ | $2,\, 1/2$ | $0.6931471806$ | $[0.6928783562,\, 0.6934248044]$ | $[0.6931427666,\, 0.6931515804]$ | $8.81 \times 10^{-6}$ |
| $5$ | $4,\, 24,\, 624$ | $3,\, 1$ | $1.6094379124$ | $[1.6091733959,\, 1.6099930681]$ | $[1.6094289926,\, 1.6094466200]$ | $1.76 \times 10^{-5}$ |
| $6$ | $2,\, 24,\, 1200$ | $4,\, 2$ | $0.0000000000$ | $[-0.0005381084,\, 0.0005547879]$ | $[-0.0000181014,\, 0.0000171535]$ | $3.53 \times 10^{-5}$ |
| $7$ | $6,\, 48,\, 2400$ | $6,\, 3$ | $1.9459101491$ | $[1.9453466034,\, 1.9469859477]$ | $[1.9458842378,\, 1.9459371200]$ | $5.29 \times 10^{-5}$ |
| $8$ | $4,\, 48,\, 3840$ | $8,\, 4$ | $0.6931471806$ | $[0.6920385170,\, 0.6942243094]$ | $[0.6931137858,\, 0.6931842953]$ | $7.05 \times 10^{-5}$ |
| $9$ | $6,\, 72,\, 6480$ | $9,\, 6$ | $1.0986122887$ | $[1.0977396571,\, 1.1001986736]$ | $[1.0985593292,\, 1.0986650935]$ | $1.06 \times 10^{-4}$ |
| $10$ | $4,\, 72,\, 9360$ | $12,\, 8$ | $0.0000000000$ | $[-0.0016179591,\, 0.0016607296]$ | $[-0.0000755653,\, 0.0000654537]$ | $1.41 \times 10^{-4}$ |
| $12$ | $4,\, 96,\, 19200$ | $16,\, 14$ | $0.0000000000$ | $[-0.0021614685,\, 0.0022101163]$ | $[-0.0001343223,\, 0.0001124608]$ | $2.47 \times 10^{-4}$ |

**Table 1:** Exact rational $O(M^{-2})$ monotone brackets (9) and $O(M^{-3})$ Euler–Maclaurin brackets (10) for $\Lambda(k)$ at $M = 60$ (`fractions.Fraction`, lower bounds rounded strictly down and upper bounds strictly up to $10$ decimal places; `results/double_zero_mellin_totient_neutral.json`).


---

## 3. Coprime Farey–Bernoulli Reciprocity and the Three-Level Rational Hierarchy

Let $B_1^*(x) := \{x\} - \frac{1}{2}$ for $x \notin \mathbb{Z}$ and $B_1^*(x) := 0$ for $x \in \mathbb{Z}$ denote the sawtooth first Bernoulli function. For each pair $j, k \ge 2$ with $L_{j, k} := \operatorname{lcm}(j, k)$, define the first- and second-order pairwise cumulative sequences on $n \in \mathbb{Z}_{\ge 0}$:
$$B_{j, k}(n) \;:=\; \sum_{r=0}^{n-1}\left(P_j(r)P_k(r) - \frac{J_2(k)}{12}\,\delta_{j, k}\right), \qquad D_{j, k}(n) \;:=\; \sum_{m=1}^{n} B_{j, k}(m), \qquad \widetilde{D}_{j, k}(n) \;:=\; D_{j, k}(n) - \overline{D}_{j, k}, \tag{12}$$
where $\overline{D}_{j, k} := \frac{1}{L_{j, k}}\sum_{n=0}^{L_{j, k}-1} D_{j, k}(n) \in \mathbb{Q}$.

**Theorem 2 (Coprime Farey–Bernoulli Reciprocity, Single-Shell $\widetilde{\mathcal{C}}_{2, k} < 0$, and Three-Level Hierarchy).**
*Let $j, k \ge 2$ be integers and $L := L_{j, k} = \operatorname{lcm}(j, k)$.*

*(a) **Coprime Farey–Bernoulli Reciprocity Formula for $\widetilde{\mathcal{C}}_{j, k}$:** The logarithm-free cotangent Gram entry (3) decomposes over primitive Farey pairs $(\alpha, \beta) \in \Omega_j \times \Omega_k$ as*
$$\widetilde{\mathcal{C}}_{j, k} \;\equiv\; \frac{1}{2}\sum_{\alpha \in \Omega_j}\sum_{\beta \in \Omega_k} \frac{\sin\bigl(\pi(\alpha - \beta)\bigr)\,B_1^*\bigl(\{\alpha - \beta\}\bigr)}{\sin(\pi\alpha)\,\sin(\pi\beta)}. \tag{13}$$

*(b) **Single-Shell Formula and Strict Negativity of $\widetilde{\mathcal{C}}_{2, k}$:** For every integer $k \ge 3$,*
$$\widetilde{\mathcal{C}}_{2, k} \;\equiv\; -\sum_{\substack{1 \le b < k/2 \\ \gcd(b, k) = 1}} \frac{b}{k}\cot\!\left(\frac{\pi b}{k}\right) \;<\; 0, \qquad -\frac{1}{2}\varphi(k) \;<\; \pi\,\widetilde{\mathcal{C}}_{2, k} \;<\; 0, \qquad \lim_{k \to \infty}\frac{\pi\,\widetilde{\mathcal{C}}_{2, k}}{\varphi(k)} \;=\; -\frac{1}{2}\log 2. \tag{14}$$

*(c) **Three-Level Trigonometric-and-Rational Hierarchy:** For all $j, k \ge 2$ and $n \in \mathbb{Z}_{\ge 0}$, $B_{j, k}(n)$ and $\widetilde{D}_{j, k}(n)$ are $L$-periodic rational sequences with $B_{j, k}(L - n) \equiv -B_{j, k}(n)$, $\frac{1}{L}\sum_{n=0}^{L-1} B_{j, k}(n) \equiv 0$, $\widetilde{D}_{j, k}(L - 1 - n) \equiv \widetilde{D}_{j, k}(n)$, and exact trigonometric expansions*
$$P_j(n)P_k(n) - \frac{J_2(k)}{12}\,\delta_{j, k} \;\equiv\; \frac{1}{4}\sum_{\substack{\alpha \in \Omega_j,\, \beta \in \Omega_k \\ \alpha \neq \beta}} \frac{\cos\bigl(\pi(2n + 1)(\alpha - \beta)\bigr)}{\sin(\pi\alpha)\,\sin(\pi\beta)}, \tag{15}$$
$$B_{j, k}(n) \;\equiv\; \frac{1}{8}\sum_{\substack{\alpha \in \Omega_j,\, \beta \in \Omega_k \\ \alpha \neq \beta}} \frac{\sin\bigl(2\pi n(\alpha - \beta)\bigr)}{\sin(\pi\alpha)\,\sin(\pi\beta)\,\sin\bigl(\pi(\alpha - \beta)\bigr)}, \qquad \widetilde{D}_{j, k}(n) \;\equiv\; -\frac{1}{16}\sum_{\substack{\alpha \in \Omega_j,\, \beta \in \Omega_k \\ \alpha \neq \beta}} \frac{\cos\bigl(\pi(2n + 1)(\alpha - \beta)\bigr)}{\sin(\pi\alpha)\,\sin(\pi\beta)\,\sin^2\bigl(\pi(\alpha - \beta)\bigr)}. \tag{16}$$
*In particular, setting $n = 0$ ($\widetilde{D}_{j, k}(0) = -\overline{D}_{j, k}$) evaluates the trigonometric sum in rational closed form:*
$$\overline{D}_{j, k} \;\equiv\; \frac{1}{16}\sum_{\substack{\alpha \in \Omega_j,\, \beta \in \Omega_k \\ \alpha \neq \beta}} \frac{\cos\bigl(\pi(\alpha - \beta)\bigr)}{\sin(\pi\alpha)\,\sin(\pi\beta)\,\sin^2\bigl(\pi(\alpha - \beta)\bigr)} \;\equiv\; \frac{1}{2 L}\sum_{r=0}^{L-1} r^2\left(P_j(r)P_k(r) - \frac{J_2(k)}{12}\,\delta_{j, k}\right) \;\in\; \mathbb{Q}, \tag{17}$$
*with explicit closed-form values for all $k \ge 3$ and all prime powers $p^\alpha$ ($\alpha \ge 1$):*
$$\overline{D}_{2, 2} = 0, \qquad \overline{D}_{2, k} \;\equiv\; \begin{cases} \dfrac{J_2(k)}{16}, & 2 \nmid k, \\[6pt] \dfrac{J_2(k)}{48}, & 4 \mid k, \\[6pt] \dfrac{J_2(k)}{144}, & k \equiv 2 \pmod 4, \end{cases} \qquad \overline{D}_{2^\alpha, 2^\alpha} \;\equiv\; 0, \qquad \overline{D}_{p^\alpha, p^\alpha} \;\equiv\; p^{4\alpha - 4}\frac{(p^2 - 1)(p^2 - 4)}{360} \quad (p \ge 3\text{ prime}). \tag{18}$$

*Proof.*
**(a)** Multiplying (4) via $2\sin A\sin B = \cos(A - B) - \cos(A + B)$ and sending $\beta \mapsto 1 - \beta$ in the second cosine (where $\sin(\pi(1-\beta)) = \sin(\pi\beta)$ and $-\cos(\pi(2n+1)(\alpha + 1 - \beta)) = +\cos(\pi(2n+1)(\alpha - \beta))$) gives $P_j(n)P_k(n) = \frac{1}{4}\sum_{\alpha \in \Omega_j, \beta \in \Omega_k}\frac{\cos(\pi(2n+1)(\alpha - \beta))}{\sin(\pi\alpha)\sin(\pi\beta)}$. Taking $\Delta_n\cos(\pi(2n+1)\theta) = -2\sin(\pi\theta)\sin(2\pi n\theta)$ ($\theta = \alpha - \beta$) and substituting into $\widetilde{\mathcal{C}}_{j, k} = \frac{1}{2L}\sum_{n=1}^{L-1}\Delta_n(P_j P_k)\cot(\pi n/L)$ [15, Eq. (17)] via $\frac{1}{2L}\sum_{n=1}^{L-1}\sin(2\pi n\theta)\cot(\pi n/L) = -B_1^*(\{\theta\})$ [19, Eq. 24.8.1] yields (13).

**(b)** Setting $j = 2$ ($\Omega_2 = \{1/2\}$, $\sin(\pi/2) = 1$) in (13) with $\beta = b/k \in \Omega_k$ ($k \ge 3$) gives $\sin(\pi(1/2 - \beta)) = \cos(\pi b/k)$ and $B_1^*(\{1/2 - b/k\}) = -b/k$ for $1 \le b < k/2$ and $1 - b/k$ for $k/2 < b < k$. Pairing $b < k/2$ with $k - b > k/2$ ($\cos(\pi(k-b)/k) = -\cos(\pi b/k)$, $B_1^*(\{1/2 - (k-b)/k\}) = b/k$) doubles the $1 \le b < k/2$ sum and yields the first identity in (14). Since $0 < \pi x\cot(\pi x) < 1$ on $x = b/k \in (0, 1/2)$ and there are $\frac{1}{2}\varphi(k)$ terms, we obtain $-\frac{1}{2}\varphi(k) < \pi\,\widetilde{\mathcal{C}}_{2, k} < 0$. By Möbius inversion and Riemann-sum convergence on $(0, 1/2)$, $\lim_{k \to \infty}\frac{\pi\widetilde{\mathcal{C}}_{2, k}}{\varphi(k)} = -\int_0^{1/2}\pi x\cot(\pi x)\,dx = -\bigl[x\log\sin(\pi x)\bigr]_0^{1/2} + \int_0^{1/2}\log\sin(\pi x)\,dx = -\frac{1}{2}\log 2$.

**(c)** Multiplying $P_j(n)$ and $P_k(n)$ in (11) and pairing $\beta$ with $1 - \beta$ yields (15) since the $\alpha = \beta$ diagonal (present only when $j = k$) sums to $\frac{1}{4}\sum_{\alpha \in \Omega_k}\sin^{-2}(\pi\alpha) = \frac{J_2(k)}{12}$. Summing (15) over $0 \le r \le n - 1$ via $\sum_{r=0}^{n-1}\cos(\pi(2r+1)\theta) = \frac{\sin(2\pi n\theta)}{2\sin(\pi\theta)}$ ($\theta = \alpha - \beta \neq 0$) gives the first identity in (16), which is $L$-periodic and odd under $n \mapsto L - n$, so $\frac{1}{L}\sum_{n=0}^{L-1} B_{j, k}(n) = 0$. Summing $B_{j, k}(m)$ over $1 \le m \le n$ via $\sum_{m=1}^n\sin(2\pi m\theta) = \frac{\cos(\pi\theta) - \cos(\pi(2n+1)\theta)}{2\sin(\pi\theta)}$ separates the constant $\overline{D}_{j, k}$ from the zero-mean oscillatory part $\widetilde{D}_{j, k}(n)$, proving the second identity in (16) and the first equality in (17). For the rational formula in (17), $D_{j, k}(n) = \sum_{r=0}^{n-1}(n - r)b_r$ ($b_r := P_j(r)P_k(r) - \frac{J_2(k)}{12}\delta_{j, k}$), so $\overline{D}_{j, k} = \frac{1}{L}\sum_{r=0}^{L-1}\frac{(L - 1 - r)(L - r)}{2}b_r$. Since $b_{L - 1 - r} = b_r$ is even and $\sum_{r=0}^{L-1} b_r = 0$, we have $\sum_{r=0}^{L-1} r\,b_r = \frac{L - 1}{2}\sum_{r=0}^{L-1} b_r = 0$, reducing $(L - 1 - r)(L - r) = r^2 - (2L - 1)r + L(L - 1)$ to $r^2$ and proving (17). Finally, setting $j = 2$ ($\alpha = 1/2$) in (17) cancels $\cos(\pi(1/2 - \beta)) = \sin(\pi\beta)$, leaving $\overline{D}_{2, k} = \frac{1}{16}\sum_{\beta \in \Omega_k}\cos^{-2}(\pi\beta)$, which equals $\frac{J_2(k)}{16}$ for odd $k$ (by Möbius inversion of $\sum_{b=1}^{d-1}\cos^{-2}(\pi b/d) = d^2 - 1$ for odd $d$), $\frac{1}{16}\sum_{\alpha \in \Omega_k}\sin^{-2}(\pi\alpha) = \frac{J_2(k)}{48}$ when $4 \mid k$ (via $b \mapsto k/2 - b$), and $\frac{J_2(k/2)}{48} = \frac{J_2(k)}{144}$ when $k \equiv 2 \pmod 4$ ($k \ge 6$). For $j = k = 2^\alpha$, $P_{2^\alpha}(r)^2 \equiv 2^{2\alpha - 4} = \frac{J_2(2^\alpha)}{12}$ is constant so $\overline{D}_{2^\alpha, 2^\alpha} = 0$; for $j = k = p^\alpha$ ($p \ge 3$ prime), $P_{p^\alpha}(r) = P_p(\lfloor r/p^{\alpha-1} \rfloor)p^{\alpha-1}$, and direct summation of $\frac{1}{2p}\sum_{r=1}^{p-1} r^2\bigl((r - \frac{p-1}{2})^2 - \frac{p^2 - 1}{12}\bigr) = \frac{(p^2 - 1)(p^2 - 4)}{360}$ yields (18). $\blacksquare$

---

## 4. Pairwise-Periodic Third-Order Euler–Maclaurin Rational Certificates ($O(M^{-3})$)

For any $\mathbf{c} = (c_2, \dots, c_N)^{\mathsf{T}} \in \mathbb{R}^{N-1}$, define the periodic variance $\mathcal{V}(\mathbf{c}) := \beta(\mathbf{c})^2 + \sum_{k=2}^N \frac{J_2(k)}{12} c_k^2$, the linear Jordan form $\overline{S}(\mathbf{c}) := \sum_{k=2}^N \frac{J_2(k)}{12} c_k$, and the first- and second-order centered prefix potentials ($m \ge 0$):
$$\widetilde{\Phi}_m(\mathbf{c}) \;:=\; \sum_{r=0}^{m-1}\Bigl(r_r(\mathbf{c})^2 - \mathcal{V}(\mathbf{c})\Bigr) \;+\; 2\,\beta(\mathbf{c})\,\overline{S}(\mathbf{c}), \qquad \Psi_n(\mathbf{c}) \;:=\; \sum_{m=1}^{n} \widetilde{\Phi}_m(\mathbf{c}) \;\equiv\; \sum_{r=0}^{n-1}(n - r)\Bigl(r_r(\mathbf{c})^2 - \mathcal{V}(\mathbf{c})\Bigr) \;+\; 2n\,\beta(\mathbf{c})\,\overline{S}(\mathbf{c}). \tag{19}$$
Write $r_n(\mathbf{c}) = \sum_{\alpha \in \Xi_N}\gamma_\alpha^{(n)}(\mathbf{c})$ on the Farey spectrum $\Xi_N := \{0\} \cup \bigcup_{k=2}^N \Omega_k$, where $\gamma_0^{(n)}(\mathbf{c}) := \beta(\mathbf{c})$ and $\gamma_\alpha^{(n)}(\mathbf{c}) := c_k\frac{e^{i\pi\alpha}}{-2i\sin(\pi\alpha)}e^{2\pi i \alpha n}$ for $\alpha \in \Omega_k$ ($2 \le k \le N$).

**Theorem 3 (Pairwise-Periodic Third-Order Euler–Maclaurin Functional and $O(M^{-3})$ Rational Certificates).**
*Let $N \ge 2$ and let $L_N := \operatorname{lcm}(1, \dots, N)$.*

*(a) **Zero-Mean Parity Decomposition and Cosecant-Squared Identity:** For every $\mathbf{c} \in \mathbb{R}^{N-1}$ and $n \ge 0$, $\widetilde{\Phi}_n(\mathbf{c}) \equiv 2\beta(\mathbf{c})\sum_{k=2}^N c_k\widetilde{S}_k(n) + \sum_{j, k=2}^N c_j c_k B_{j, k}(n)$ is $L_N$-periodic with $\frac{1}{L_N}\sum_{n=0}^{L_N-1}\widetilde{\Phi}_n(\mathbf{c}) \equiv 0$. Consequently, $\Psi_n(\mathbf{c})$ is $L_N$-periodic, its periodic mean $\overline{\Psi}(\mathbf{c}) := \frac{1}{L_N}\sum_{n=0}^{L_N-1}\Psi_n(\mathbf{c})$ reduces to pairwise periods $L_{j, k} = \operatorname{lcm}(j, k) \le N^2$,*
$$\overline{\Psi}(\mathbf{c}) \;\equiv\; -\beta(\mathbf{c})\,\overline{S}(\mathbf{c}) \;+\; \sum_{j=2}^N\sum_{k=2}^N c_j c_k\,\overline{D}_{j, k} \;\in\; \mathbb{Q}[\mathbf{c}], \tag{20}$$
*and the centered second-order potential $\widetilde{\Psi}_n(\mathbf{c}) := \Psi_n(\mathbf{c}) - \overline{\Psi}(\mathbf{c})$ splits into an odd-under-reflection linear part and an even-under-reflection quadratic part:*
$$\widetilde{\Psi}_n(\mathbf{c}) \;\equiv\; 2\,\beta(\mathbf{c})\sum_{k=2}^N c_k\,\widetilde{U}_k(n) \;+\; \sum_{j=2}^N\sum_{k=2}^N c_j c_k\,\widetilde{D}_{j, k}(n) \;\equiv\; -\frac{1}{4}\sum_{\substack{\alpha, \beta \in \Xi_N \\ \alpha \neq \beta}} \frac{\gamma_\alpha^{(n)}(\mathbf{c})\,\overline{\gamma_\beta^{(n)}(\mathbf{c})}}{\sin^2\bigl(\pi(\alpha - \beta)\bigr)}, \tag{21}$$
*while $\widetilde{\Phi}_m(\mathbf{c}) \equiv \sum_{\alpha \neq \beta \in \Xi_N}\frac{\gamma_\alpha^{(m)}(\mathbf{c})\overline{\gamma_\beta^{(m)}(\mathbf{c})}}{e^{2\pi i(\alpha - \beta)} - 1}$ and the midpoint average $\frac{1}{2}(\widetilde{\Phi}_m(\mathbf{c}) + \widetilde{\Phi}_{m+1}(\mathbf{c})) \equiv \sum_{\alpha \neq \beta \in \Xi_N}\frac{\cot(\pi(\alpha - \beta))}{2i}\gamma_\alpha^{(m)}(\mathbf{c})\overline{\gamma_\beta^{(m)}(\mathbf{c})}$.*

*(b) **Positive Cubic-Vanishing Prefix Weights and $O(M^{-3})$ Rational Functional:** For every integer cutoff $M \ge 2$ and every $\mathbf{c} \in \mathbb{R}^{N-1}$, the exact infinite-horizon norm satisfies*
$$\|1 - g\|_{\mathcal{H}}^2 \;\equiv\; \mathcal{Q}_{N, M}^{\mathrm{EM3}}(\mathbf{c}) \;+\; 6\sum_{n=M+1}^{\infty}\frac{\widetilde{\Psi}_n(\mathbf{c})}{(n - 1)n(n + 1)(n + 2)}, \tag{22}$$
*where $\mathcal{Q}_{N, M}^{\mathrm{EM3}}(\mathbf{c}) \in \mathbb{Q}[\mathbf{c}]$ is the explicit rational quadratic form ($r_0(\mathbf{c}) \equiv 1$)*
$$\mathcal{Q}_{N, M}^{\mathrm{EM3}}(\mathbf{c}) \;:=\; \sum_{r=1}^{M-1} w_r^{\mathrm{EM3}}\,r_r(\mathbf{c})^2 \;-\; \frac{3M + 2}{M(M+1)(M+2)} \;+\; \frac{3M^2 + 6M + 2}{M(M+1)(M+2)}\,\mathcal{V}(\mathbf{c}) \;-\; \frac{6}{M(M+2)}\,\beta(\mathbf{c})\,\overline{S}(\mathbf{c}) \;+\; \frac{2\sum_{j, k=2}^N c_j c_k\,\overline{D}_{j, k}}{M(M+1)(M+2)}, \tag{23}$$
*with strictly positive prefix weights vanishing at $r = M$ and $r = M + 1$:*
$$w_r^{\mathrm{EM3}} \;:=\; \frac{1}{r(r+1)} - \frac{1}{M(M+1)} - \frac{2(M - r)}{M(M+1)(M+2)} \;\equiv\; \frac{(M - r)(M + 1 - r)(M + 2 + 2r)}{r(r+1)\,M(M+1)(M+2)} \;>\; 0 \qquad (1 \le r \le M - 1). \tag{24}$$

*(c) **Exact Rational Primal-Dual Certificates:** Define $\delta_{j, k} := \max_{0 \le n < L_{j, k}}|\widetilde{D}_{j, k}(n)| \in \mathbb{Q}_{\ge 0}$, and for any positive rational calibration amplitudes $b_0, a_2, \dots, a_N \in \mathbb{Q}_{>0}$, define $\gamma_0 := \sum_{k=2}^N \frac{a_k}{b_0}u_k^{\max} \in \mathbb{Q}_{>0}$, $\gamma_k := \frac{b_0}{a_k}u_k^{\max} + \sum_{j=2}^N \frac{a_j}{a_k}\delta_{k, j} \in \mathbb{Q}_{>0}$, and $\mathcal{W}_N^{\mathrm{EM3}}(\mathbf{c}) := \gamma_0\,\beta(\mathbf{c})^2 + \sum_{k=2}^N \gamma_k c_k^2$. Then for all $\mathbf{c} \in \mathbb{R}^{N-1}$ and $n \ge 0$, $|\widetilde{\Psi}_n(\mathbf{c})| \le \mathcal{W}_N^{\mathrm{EM3}}(\mathbf{c})$ and*
$$\mathcal{Q}_{N, M}^{\mathrm{EM3, LB}}(\mathbf{c}) \;:=\; \mathcal{Q}_{N, M}^{\mathrm{EM3}}(\mathbf{c}) - \frac{2\,\mathcal{W}_N^{\mathrm{EM3}}(\mathbf{c})}{M(M+1)(M+2)} \;\le\; \|1 - g\|_{\mathcal{H}}^2 \;\le\; \mathcal{Q}_{N, M}^{\mathrm{EM3}}(\mathbf{c}) + \frac{2\,\mathcal{W}_N^{\mathrm{EM3}}(\mathbf{c})}{M(M+1)(M+2)} \;=:\; \mathcal{Q}_{N, M}^{\mathrm{EM3, UB}}(\mathbf{c}). \tag{25}$$
*Moreover, whenever $C_\beta^{\mathrm{EM3}} := \frac{3M^2 + 6M + 2}{M(M+1)(M+2)} - \frac{3}{M(M+2)}\sum_{k=2}^N \frac{J_2(k)}{12} - \frac{2\gamma_0}{M(M+1)(M+2)} > 0$ and $\Gamma_k^{\mathrm{EM3}} := \bigl(\frac{3M^2 + 6M + 2}{M(M+1)(M+2)} - \frac{3}{M(M+2)}\bigr)\frac{J_2(k)}{12} - \frac{2(\gamma_k + \sum_{j=2}^N |\overline{D}_{k, j}|)}{M(M+1)(M+2)} > 0$ ($2 \le k \le N$), evaluating at any reference vector $\mathbf{c}_0 \in \mathbb{Q}^{N-1}$ with $g_k(\mathbf{c}_0) := \frac{1}{2}\frac{\partial}{\partial c_k}\mathcal{Q}_{N, M}^{\mathrm{EM3, LB}}(\mathbf{c}_0) \in \mathbb{Q}$ yields the certified rational sandwich*
$$\mathrm{LB}_{\mathbb{Q}}^{\mathrm{EM3}}(d_N^2) \;:=\; \mathcal{Q}_{N, M}^{\mathrm{EM3, LB}}(\mathbf{c}_0) - \sum_{k=2}^N \frac{g_k(\mathbf{c}_0)^2}{\Gamma_k^{\mathrm{EM3}}} \;\le\; d_N^2 \;\le\; \mathcal{Q}_{N, M}^{\mathrm{EM3, UB}}(\mathbf{c}_0) \;=:\; \mathrm{UB}_{\mathbb{Q}}^{\mathrm{EM3}}(d_N^2). \tag{26}$$

*Proof.*
**(a)** Expanding $r_r(\mathbf{c})^2 - \mathcal{V}(\mathbf{c}) = 2\beta(\mathbf{c})\sum_{k=2}^N c_k P_k(r) + \sum_{j, k=2}^N c_j c_k\bigl(P_j(r)P_k(r) - \frac{J_2(k)}{12}\delta_{j, k}\bigr)$ and summing over $0 \le r \le m - 1$ gives $\widetilde{\Phi}_m(\mathbf{c}) = 2\beta(\mathbf{c})\sum_{k=2}^N c_k\widetilde{S}_k(m) + \sum_{j, k=2}^N c_j c_k B_{j, k}(m)$, which has mean $0$ over $L_N$ because each $\widetilde{S}_k$ has mean $0$ over $k \mid L_N$ and each $B_{j, k}$ has mean $0$ over $L_{j, k} \mid L_N$ (Theorem 2(c)). Summing $\widetilde{\Phi}_m(\mathbf{c})$ over $1 \le m \le n$ gives $\Psi_n(\mathbf{c}) = 2\beta(\mathbf{c})\sum_{k=2}^N c_k\bigl(\widetilde{U}_k(n) - \frac{J_2(k)}{24}\bigr) + \sum_{j, k=2}^N c_j c_k\bigl(\widetilde{D}_{j, k}(n) + \overline{D}_{j, k}\bigr)$. Since $\widetilde{U}_k(k - 1 - n) = -\widetilde{U}_k(n)$ has mean $0$ over $k \mid L_N$ and $\widetilde{D}_{j, k}$ has mean $0$ over $L_{j, k} \mid L_N$, averaging over $0 \le n < L_N$ yields (20) and the first equality in (21). Substituting $\gamma_\alpha^{(n)}(\mathbf{c})$ and summing $\sum_{r=0}^{m-1} e^{2\pi i(\alpha - \beta)r} = \frac{e^{2\pi i(\alpha - \beta)m} - 1}{e^{2\pi i(\alpha - \beta)} - 1}$ and $\sum_{m=1}^n e^{2\pi i(\alpha - \beta)m}$ yields the Fourier formulas for $\widetilde{\Phi}_m(\mathbf{c})$ and (21).

**(b)** Writing $r_n(\mathbf{c})^2 = \mathcal{V}(\mathbf{c}) + \widetilde{\Phi}_{n+1}(\mathbf{c}) - \widetilde{\Phi}_n(\mathbf{c})$ for $n \ge M$ and summing by parts twice via $\sum_{n=M}^\infty \frac{\mathcal{V}(\mathbf{c})}{n(n+1)} = \frac{\mathcal{V}(\mathbf{c})}{M}$, $\frac{1}{(n-1)n} - \frac{1}{n(n+1)} = \frac{2}{(n-1)n(n+1)}$, and $\widetilde{\Phi}_m(\mathbf{c}) = \widetilde{\Psi}_m(\mathbf{c}) - \widetilde{\Psi}_{m-1}(\mathbf{c})$ yields
$$\sum_{n=M}^{\infty}\frac{r_n(\mathbf{c})^2}{n(n+1)} \;=\; \frac{\mathcal{V}(\mathbf{c})}{M} \;-\; \frac{\widetilde{\Phi}_M(\mathbf{c})}{M(M+1)} \;-\; \frac{2\,\widetilde{\Psi}_M(\mathbf{c})}{M(M+1)(M+2)} \;+\; 6\sum_{n=M+1}^{\infty}\frac{\widetilde{\Psi}_n(\mathbf{c})}{(n-1)n(n+1)(n+2)}.$$
Substituting $\widetilde{\Phi}_M(\mathbf{c})$ and $\widetilde{\Psi}_M(\mathbf{c}) = \Psi_M(\mathbf{c}) - \overline{\Psi}(\mathbf{c})$ from (19)–(20) and using $r_0(\mathbf{c})^2 \equiv 1$ gives (22)–(24).

**(c)** By weighted AM-GM, $2|\beta(\mathbf{c})||c_k| \le \frac{a_k}{b_0}\beta(\mathbf{c})^2 + \frac{b_0}{a_k}c_k^2$ and $2|c_j||c_k| \le \frac{a_k}{a_j}c_j^2 + \frac{a_j}{a_k}c_k^2$; applying this to $|\widetilde{\Psi}_n(\mathbf{c})| \le 2|\beta(\mathbf{c})|\sum_{k=2}^N |c_k|u_k^{\max} + \sum_{j, k=2}^N |c_j||c_k|\delta_{j, k}$ gives $|\widetilde{\Psi}_n(\mathbf{c})| \le \mathcal{W}_N^{\mathrm{EM3}}(\mathbf{c})$, and $6\sum_{n=M+1}^\infty \frac{1}{(n-1)n(n+1)(n+2)} = \frac{2}{M(M+1)(M+2)}$ yields (25). Finally, for any $\mathbf{c} = \mathbf{c}_0 + \mathbf{d}$, expanding $\mathcal{Q}_{N, M}^{\mathrm{EM3, LB}}(\mathbf{c}_0 + \mathbf{d})$ around $\mathbf{c}_0$ and using $\sum_{r=1}^{M-1} w_r^{\mathrm{EM3}}(\sum_k d_k\tau_k(r))^2 \ge 0$ (since $w_r^{\mathrm{EM3}} > 0$), $-2\delta_\beta\sum_{k=2}^N \frac{J_2(k)}{12}d_k \ge -\bigl(\sum_{k=2}^N \frac{J_2(k)}{12}\bigr)\delta_\beta^2 - \sum_{k=2}^N \frac{J_2(k)}{12}d_k^2$ ($\delta_\beta := \frac{1}{2}\sum_{k=2}^N \varphi(k)d_k$), and $\sum_{j, k=2}^N d_j d_k\overline{D}_{j, k} \ge -\sum_{k=2}^N\bigl(\sum_{j=2}^N |\overline{D}_{k, j}|\bigr)d_k^2$ bounds the quadratic increment from below by $C_\beta^{\mathrm{EM3}}\delta_\beta^2 + \sum_{k=2}^N \Gamma_k^{\mathrm{EM3}} d_k^2 \ge \sum_{k=2}^N \Gamma_k^{\mathrm{EM3}} d_k^2$. Completing the square $2 g_k(\mathbf{c}_0)d_k + \Gamma_k^{\mathrm{EM3}} d_k^2 \ge -\frac{g_k(\mathbf{c}_0)^2}{\Gamma_k^{\mathrm{EM3}}}$ proves (26). $\blacksquare$

| $(N,\, M)$ | Certified Lower Bound $\mathrm{LB}_{\mathbb{Q}}^{\mathrm{EM3}}(d_N^2)$ | Certified Upper Bound $\mathrm{UB}_{\mathbb{Q}}^{\mathrm{EM3}}(d_N^2)$ | Certificate Width | Midpoint $\mathcal{Q}_{N, M}^{\mathrm{EM3}}(\mathbf{c}_0)$ Error vs. $d_N^2$ | Dual Defect $\sum_k \frac{g_k^2}{\Gamma_k^{\mathrm{EM3}}}$ |
| :--- | :--- | :--- | :--- | :--- | :--- |
| $(4,\, 50)$ | $0.0659395162$ | $0.0659626572$ | $2.31 \times 10^{-5}$ | $6.08 \times 10^{-7}$ | $1.98 \times 10^{-18}$ |
| $(4,\, 100)$ | $0.0659501896$ | $0.0659531681$ | $2.98 \times 10^{-6}$ | $1.54 \times 10^{-8}$ | $2.49 \times 10^{-18}$ |
| $(6,\, 100)$ | $0.0362050897$ | $0.0362110611$ | $5.97 \times 10^{-6}$ | $4.08 \times 10^{-8}$ | $1.81 \times 10^{-17}$ |
| $(6,\, 200)$ | $0.0362076519$ | $0.0362084094$ | $7.57 \times 10^{-7}$ | $3.99 \times 10^{-9}$ | $3.46 \times 10^{-17}$ |
| $(10,\, 150)$ | $0.0238483736$ | $0.0238565434$ | $8.17 \times 10^{-6}$ | $9.48 \times 10^{-8}$ | $3.93 \times 10^{-17}$ |
| $(10,\, 400)$ | $0.0238523355$ | $0.0238527714$ | $4.36 \times 10^{-7}$ | $1.31 \times 10^{-10}$ | $3.04 \times 10^{-16}$ |
| $(10,\, 1000)$ | $0.0238525393$ | $0.0238525674$ | $2.80 \times 10^{-8}$ | $3.83 \times 10^{-11}$ | $1.73 \times 10^{-14}$ |
| $(20,\, 1600)$ | $0.0165381590$ | $0.0165383489$ | $1.90 \times 10^{-7}$ | $1.85 \times 10^{-9}$ | $1.93 \times 10^{-14}$ |
| $(30,\, 3600)$ | $0.0145566862$ | $0.0145567807$ | $9.43 \times 10^{-8}$ | $1.13 \times 10^{-10}$ | $7.93 \times 10^{-14}$ |
| $(40,\, 6400)$ | $0.0127357817$ | $0.0127358890$ | $1.07 \times 10^{-7}$ | $2.74 \times 10^{-11}$ | $7.41 \times 10^{-13}$ |

**Table 2:** 100% exact rational ($O(M^{-3})$) primal-dual certificates (26) for $d_N^2$ (`fractions.Fraction`, zero transcendental constants; lower bounds rounded strictly down and upper bounds strictly up to $10$ decimal places; `results/double_zero_mellin_totient_neutral.json`).

---

## 5. Double-Zero Jordan–Totient Mellin Theory of $Q_k$ and Entire Continuation on $V_N^{\mathrm{cen}}$

For any $f \in \mathcal{H}$, let $\mathcal{M}[f](s) := \int_1^\infty f(x)\,x^{-s-1}\,dx$ ($\Re(s) > \frac{1}{2}$) denote its Mellin transform. For $k \ge 2$ and $\Re(s) > 0$, $\widehat{\tau}_k(s) := \mathcal{M}[\tau_k](s) = \frac{\zeta(s)}{s}\phi_{1-s}(k)$ where $\phi_z(k) := \sum_{d \mid k} d^z\mu(k/d) = k^z\prod_{p \mid k}(1 - p^{-z})$ is the complex Jordan–totient Dirichlet polynomial. For each totient-neutral generator $Q_k := -\tau_k + \varphi(k)\tau_2$ ($k \ge 3$), define $F_k(s) := -\phi_{1-s}(k) + \varphi(k)\phi_{1-s}(2)$ so that $\widehat{Q}_k(s) = \frac{\zeta(s)}{s}F_k(s)$.

**Theorem 4 (Double-Zero Jordan–Totient Mellin Identity, Entire Continuation, and Parabolic Projections).**
*Let $j, k \ge 3$ and $N \ge 3$.*

*(a) **Boundary Vanishing of $Q_k$ and Double-Zero Identity for $F_k(s)$:** On its fundamental period $L_k := \operatorname{lcm}(2, k)$, $Q_k(0) = Q_k(L_k - 1) = 0$. The entire Dirichlet polynomial $F_k(s)$ vanishes at both boundaries $s = 0$ and $s = 1$ of the critical strip,*
$$F_k(0) \;=\; F_k(1) \;=\; 0, \qquad F_k'(1) \;=\; \Lambda(k) - \varphi(k)\log 2 \;=:\; \widetilde{\Lambda}_k, \qquad F_k'(0) \;=\; \varphi(k)\left(\log k + \sum_{p \mid k}\frac{\log p}{p - 1} - 2\log 2\right) > 0, \tag{27}$$
*so $\widehat{Q}_k(s) = \frac{\zeta(s)}{s}F_k(s)$ extends to an entire function on $\mathbb{C}$ with $\widehat{Q}_k(1) = F_k'(1) = \widetilde{\Lambda}_k$ and $\widehat{Q}_k(0) = -\frac{1}{2}F_k'(0) < 0$.*

*(b) **Removable Strip-Endpoint Poles in the Mellin–Plancherel Integrand:** By Mellin–Plancherel isometry,*
$$\pi\,\widehat{\mathcal{C}}_{j, k} \;=\; \langle Q_j,\, Q_k \rangle_{\mathcal{H}} \;=\; \frac{1}{2\pi i}\int_{1/2 - i\infty}^{1/2 + i\infty} I_{j, k}(s)\,ds, \qquad I_{j, k}(s) \;:=\; \frac{\zeta(s)\zeta(1 - s)}{s(1 - s)}\,F_j(s)\,F_k(1 - s), \tag{28}$$
*where the double zeros $F_j(1) = F_k(0) = 0$ at $s = 1$ and $F_j(0) = F_k(1) = 0$ at $s = 0$ cancel the double poles of $\frac{\zeta(s)\zeta(1 - s)}{s(1 - s)}$, making $I_{j, k}(s)$ holomorphic on $\mathbb{C}$ with finite endpoint values $\lim_{s \to 1} I_{j, k}(s) = -\frac{1}{2}F_j'(1)F_k'(0)$ and $\lim_{s \to 0} I_{j, k}(s) = -\frac{1}{2}F_j'(0)F_k'(1)$.*

*(c) **Entire Mellin Continuation on $V_N^{\mathrm{cen}}$ and Parabolic Shift Projections:** For every $g = \sum_{k=2}^N c_k\tau_k \in V_N$, $\mathcal{M}[1 - g](s)$ is holomorphic at $s = 1$ and has simple-pole residue at $s = 0$ equal to*
$$\operatorname{Res}_{s = 0}\mathcal{M}[1 - g](s) \;\equiv\; 1 + \frac{1}{2}\sum_{k=2}^N c_k\,\varphi(k) \;=\; \beta(\mathbf{c}). \tag{29}$$
*Thus $\mathcal{M}[1 - g](s)$ is entire on $\mathbb{C}$ if and only if $g \in V_N^{\mathrm{cen}}$ ($\beta(\mathbf{c}) = 0$). For $Q_0 := 1 + 2\tau_2$, $\mathcal{M}[Q_0](1) = 1 - 2\log 2$ and $\mathcal{M}[Q_0](0) = \log(2/\pi)$. Finally, the parabolic shift parameters $B_N$ and $A_N$ in $d_N^{\mathrm{cen}, 2} - d_N^2 \equiv B_N^2 / A_N$ [15, Theorem 3(b)] satisfy the Hilbert-space projection and Mellin-endpoint identities ($V_N^0 := \operatorname{span}\{Q_3, \dots, Q_N\}$, $Q_N^{\mathrm{cen}, *} := 1 - g_N^{\mathrm{cen}, *} = Q_0 + \sum_{k=3}^N c_{N, k}^{\mathrm{cen}, *} Q_k$):*
$$B_N \;\equiv\; -2\,\langle \tau_2,\, Q_N^{\mathrm{cen}, *} \rangle_{\mathcal{H}}, \qquad A_N \;\equiv\; 4\,\bigl\|P_{(V_N^0)^\perp}\tau_2\bigr\|_{\mathcal{H}}^2, \qquad \mathcal{M}\bigl[Q_N^{\mathrm{cen}, *}\bigr](1) \;\equiv\; d_N^{\mathrm{cen}, 2} + B_N, \qquad \mathcal{M}\bigl[Q_N^{\mathrm{cen}, *}\bigr](0) \;\equiv\; \log\!\left(\frac{2}{\pi}\right) - \frac{1}{2}\sum_{k=3}^N c_{N, k}^{\mathrm{cen}, *} F_k'(0). \tag{30}$$

*Proof.*
**(a)** Since $\tau_m(0) = 0$ and $\tau_m(L_k - 1) = -\varphi(m)$ for $m \in \{2, k\}$, $Q_k(0) = 0$ and $Q_k(L_k - 1) = \varphi(k) - \varphi(k)\cdot 1 = 0$. At $s = 1$ ($z = 0$), $\phi_0(m) = \sum_{d \mid m}\mu(m/d) = 0$ for $m \ge 2$, so $F_k(1) = 0$; at $s = 0$ ($z = 1$), $\phi_1(m) = \varphi(m)$ and $\varphi(2) = 1$, so $F_k(0) = -\varphi(k) + \varphi(k)\cdot 1 = 0$. Differentiating $\phi_z(k) = k^z\prod_{p \mid k}(1 - p^{-z})$ with respect to $z$ gives $\phi_0'(k) = \Lambda(k)$ ($k \ge 2$) and $\phi_1'(k) = \varphi(k)\bigl(\log k + \sum_{p \mid k}\frac{\log p}{p - 1}\bigr)$. By the chain rule $\frac{d}{ds}F_k(s) = \phi_{1-s}'(k) - \varphi(k)\phi_{1-s}'(2)$, giving $F_k'(1) = \Lambda(k) - \varphi(k)\log 2$ and $F_k'(0) = \varphi(k)\bigl(\log k + \sum_{p \mid k}\frac{\log p}{p - 1} - 2\log 2\bigr) > 0$ (since for $k = 3$, $\frac{3}{2}\log 3 - 2\log 2 = \frac{1}{2}\log(27/16) > 0$, and for $k \ge 4$, $\log k + \sum_{p \mid k}\frac{\log p}{p - 1} > \log 4 = 2\log 2$). Since $\zeta(s) = \frac{1}{s - 1} + O(1)$ near $s = 1$ and $\zeta(0) = -\frac{1}{2}$, L'Hôpital's rule gives $\widehat{Q}_k(1) = F_k'(1)$ and $\widehat{Q}_k(0) = \zeta(0)F_k'(0) = -\frac{1}{2}F_k'(0)$.

**(b)** On $\Re(s) = \frac{1}{2}$ ($s = \frac{1}{2} + it$), $\overline{\widehat{Q}_k(s)} = \widehat{Q}_k(1 - s)$, so Mellin–Plancherel isometry gives (28). Near $s = 1$, $\frac{\zeta(s)\zeta(1 - s)}{s(1 - s)} = \frac{(s - 1)^{-1}\zeta(0) + O(1)}{-(s - 1)} = \frac{1}{2}(s - 1)^{-2} + O((s - 1)^{-1})$ while $F_j(s)F_k(1 - s) = -(s - 1)^2 F_j'(1)F_k'(0) + O((s - 1)^3)$, so $(s - 1)^2$ cancels and $\lim_{s \to 1} I_{j, k}(s) = -\frac{1}{2}F_j'(1)F_k'(0)$ (and symmetrically at $s = 0$).

**(c)** Since $\mathcal{M}[1](s) = \frac{1}{s}$, $\mathcal{M}[1 - g](s) = \frac{1}{s}\bigl(1 - \zeta(s)\sum_{k=2}^N c_k\phi_{1-s}(k)\bigr)$ has residue at $s = 0$ equal to $1 - \zeta(0)\sum_{k=2}^N c_k\phi_1(k) = 1 + \frac{1}{2}\sum_{k=2}^N c_k\varphi(k) = \beta(\mathbf{c})$, proving (29). For $Q_0 = 1 + 2\tau_2$, $\mathcal{M}[Q_0](s) = \frac{1}{s}\bigl(1 + 2\zeta(s)(2^{1-s} - 1)\bigr)$; expanding $\zeta(0 + s) = -\frac{1}{2} - \frac{1}{2}\log(2\pi)s + O(s^2)$ and $2^{1-s} - 1 = 1 - (2\log 2)s + O(s^2)$ gives $1 + 2\zeta(s)(2^{1-s} - 1) = \log(2/\pi)s + O(s^2)$, so $\mathcal{M}[Q_0](0) = \log(2/\pi)$ and $\mathcal{M}[Q_0](1) = \lim_{s \to 1}\bigl(\frac{1}{s} + \frac{2\zeta(s)}{s}(2^{1-s} - 1)\bigr) = 1 - 2\log 2$. Finally, since $\langle \tau_2, Q_0 \rangle_{\mathcal{H}} = -\log 2 + 2\log 2 = \log 2$ and $\langle \tau_2, Q_k \rangle_{\mathcal{H}} = -\frac{1}{2}(\Lambda(k) - \varphi(k)\log 2) - \pi\widetilde{\mathcal{C}}_{2, k} = -\frac{1}{2}\widetilde{\Lambda}_k - \pi\widetilde{\mathcal{C}}_{2, k}$ by (3), while $\pi\widehat{\mathcal{C}}_N\mathbf{c}_N^{\mathrm{cen}, *} = 2\pi\tilde{\mathbf{b}}_N$ and $\alpha_N = \tilde{\mathbf{b}}_N^{\mathsf{T}}\mathbf{c}_N^{\mathrm{cen}, *}$, we have $-2\langle \tau_2, Q_N^{\mathrm{cen}, *} \rangle_{\mathcal{H}} = -2\log 2 + \widetilde{\boldsymbol{\Lambda}}_N^{\mathsf{T}}\mathbf{c}_N^{\mathrm{cen}, *} + 2\pi\tilde{\mathbf{b}}_N^{\mathsf{T}}\mathbf{c}_N^{\mathrm{cen}, *} = -2\log 2 + \pi\alpha_N + \widetilde{\boldsymbol{\Lambda}}_N^{\mathsf{T}}\mathbf{c}_N^{\mathrm{cen}, *} = B_N$, and $4\|P_{(V_N^0)^\perp}\tau_2\|_{\mathcal{H}}^2 = 4\bigl(\|\tau_2\|_{\mathcal{H}}^2 - (\langle \tau_2, Q_k \rangle)_{k=3}^N (\pi\widehat{\mathcal{C}}_N)^{-1}(\langle \tau_2, Q_k \rangle)_{k=3}^N\bigr) = 4\log 2 - \frac{1}{\pi}(2\pi\tilde{\mathbf{b}}_N + \widetilde{\boldsymbol{\Lambda}}_N)^{\mathsf{T}}\widehat{\mathcal{C}}_N^{-1}(2\pi\tilde{\mathbf{b}}_N + \widetilde{\boldsymbol{\Lambda}}_N) = A_N$ [15, Eq. (27)], proving (30). $\blacksquare$

| Modulus $k$ | $\varphi(k),\, J_2(k)/12$ | $\widetilde{\mathcal{C}}_{2, k}$ (Theorem 2(b)) | $\frac{\pi\widetilde{\mathcal{C}}_{2, k}}{\varphi(k)}$ | $F_k'(0) > 0$ (Theorem 4(a)) | $F_k'(1) = \widehat{Q}_k(1) = \widetilde{\Lambda}_k$ | $\widehat{Q}_k(0) = -\frac{1}{2}F_k'(0)$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $3$ | $2,\, 0.666667$ | $-0.1924500897$ | $-0.3022998940$ | $+0.5232481438$ | $-0.2876820725$ | $-0.2616240719$ |
| $4$ | $2,\, 1.000000$ | $-0.2500000000$ | $-0.3926990817$ | $+1.3862943611$ | $-0.6931471806$ | $-0.6931471806$ |
| $5$ | $4,\, 2.000000$ | $-0.4052442626$ | $-0.3182780996$ | $+2.5020121177$ | $-1.1631508098$ | $-1.2510060588$ |
| $6$ | $2,\, 2.000000$ | $-0.2886751346$ | $-0.4534498411$ | $+3.2958368660$ | $-1.3862943611$ | $-1.6479184330$ |
| $7$ | $6,\, 4.000000$ | $-0.6223140854$ | $-0.3258428931$ | $+5.3036048767$ | $-2.2129729343$ | $-2.6518024383$ |
| $8$ | $4,\, 4.000000$ | $-0.4571067812$ | $-0.3590108264$ | $+5.5451774445$ | $-2.0794415417$ | $-2.7725887222$ |
| $9$ | $6,\, 6.000000$ | $-0.6484769475$ | $-0.3395417357$ | $+8.1614181633$ | $-3.0602707947$ | $-4.0807090817$ |
| $10$ | $4,\, 6.000000$ | $-0.5257311121$ | $-0.4129082499$ | $+8.0471895622$ | $-2.7725887222$ | $-4.0235947811$ |
| $11$ | $10,\, 10.000000$ | $-1.0602630040$ | $-0.3330914464$ | $+12.5139043896$ | $-4.5335765328$ | $-6.2569521948$ |
| $12$ | $4,\, 8.000000$ | $-0.4226497308$ | $-0.3319483223$ | $+9.3642624542$ | $-2.7725887222$ | $-4.6821312271$ |

**Table 3:** Single-shell deflation values $\widetilde{\mathcal{C}}_{2, k} < 0$ (Theorem 2(b)) and strip-endpoint derivatives $F_k'(0), F_k'(1)$ and Mellin values $\widehat{Q}_k(1), \widehat{Q}_k(0)$ (Theorem 4(a)) for $3 \le k \le 12$ (`results/double_zero_mellin_totient_neutral.json`).

| Horizon $N$ | $d_N^{\mathrm{cen}, 2} = 1 - \pi\alpha_N$ | $B_N = -2\langle \tau_2, Q_N^{\mathrm{cen}, *} \rangle_{\mathcal{H}}$ | $A_N = 4\|P_{(V_N^0)^\perp}\tau_2\|_{\mathcal{H}}^2$ | Parabolic Gap $B_N^2 / A_N$ | $\mathcal{M}[Q_N^{\mathrm{cen}, *}](1) = d_N^{\mathrm{cen}, 2} + B_N$ | $\mathcal{M}[Q_N^{\mathrm{cen}, *}](0)$ | Variance $\mathcal{V}(\mathbf{c}_N^{\mathrm{cen}, *})$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $3$ | $0.1938669492$ | $-0.3883732620$ | $1.5372510580$ | $9.811917 \times 10^{-2}$ | $-0.1945063128$ | $-0.2771666574$ | $0.4074074074$ |
| $4$ | $0.0953936318$ | $-0.1794799969$ | $1.0941219205$ | $2.944194 \times 10^{-2}$ | $-0.0840863651$ | $-0.1596503578$ | $0.2889423275$ |
| $6$ | $0.0414130358$ | $-0.0602987162$ | $0.6985464665$ | $5.205001 \times 10^{-3}$ | $-0.0188856804$ | $-0.0721332348$ | $0.1917405956$ |
| $8$ | $0.0247148270$ | $-0.0156783732$ | $0.5226674861$ | $4.703017 \times 10^{-4}$ | $+0.0090364537$ | $-0.0186082993$ | $0.1618222591$ |
| $10$ | $0.0242801461$ | $-0.0132861912$ | $0.4128293997$ | $4.275928 \times 10^{-4}$ | $+0.0109939550$ | $-0.0150283222$ | $0.1591170014$ |
| $12$ | $0.0205086583$ | $-0.0018972886$ | $0.3424702598$ | $1.051100 \times 10^{-5}$ | $+0.0186113698$ | $+0.0090228255$ | $0.1613212209$ |
| $15$ | $0.0184911882$ | $-0.0055213172$ | $0.2747049373$ | $1.109734 \times 10^{-4}$ | $+0.0129698711$ | $-0.0068364800$ | $0.1652026773$ |
| $20$ | $0.0165394411$ | $+0.0004932458$ | $0.2046238254$ | $1.188969 \times 10^{-6}$ | $+0.0170326868$ | $+0.0128411282$ | $0.1793149183$ |
| $24$ | $0.0155778441$ | $-0.0001059269$ | $0.1698028967$ | $6.607954 \times 10^{-8}$ | $+0.0154719172$ | $+0.0096780082$ | $0.1838491874$ |
| $30$ | $0.0146088919$ | $+0.0026578951$ | $0.1354416136$ | $5.215832 \times 10^{-5}$ | $+0.0172667870$ | $+0.0288212574$ | $0.1960552380$ |

**Table 4:** Hilbert-space projection values $B_N = -2\langle \tau_2, Q_N^{\mathrm{cen}, *} \rangle_{\mathcal{H}}$, $A_N = 4\|P_{(V_N^0)^\perp}\tau_2\|_{\mathcal{H}}^2$, entire Mellin endpoint values $\mathcal{M}[Q_N^{\mathrm{cen}, *}](1), \mathcal{M}[Q_N^{\mathrm{cen}, *}](0)$, and periodic variance $\mathcal{V}(\mathbf{c}_N^{\mathrm{cen}, *})$ across $3 \le N \le 30$ (Theorem 4(c); `results/double_zero_mellin_totient_neutral.json`).

---

## 6. Reproducibility and Verification Supplement

The supplementary archive `paper45_supplement.zip` executes in under $5.0$ seconds using Python $\ge 3.10$ and `numpy`. Running `python3 src/study_double_zero_mellin_totient_neutral.py --output-dir results` (`PASS` in `4.78s` $\to$ `results/double_zero_mellin_totient_neutral.json`) verifies:
1. **Theorem 1:** Ramanujan-sum jump law, primitive Farey-shell sine expansion (4), cumulative non-positivity $S_k(n) \le 0$ (5), simultaneous four-tier orthogonality (6)–(7), and exact `fractions.Fraction` brackets (9)–(10) for $\Lambda(k)$ (Table 1).
2. **Theorem 2:** Coprime Farey–Bernoulli reciprocity (13) ($2 \le j, k \le 20$), single-shell formula $\widetilde{\mathcal{C}}_{2, k} < 0$ (14) ($3 \le k \le 60$), and rational closed forms (17)–(18) for $\overline{D}_{j, k}$, $\overline{D}_{2, k}$, and $\overline{D}_{p^\alpha, p^\alpha}$ ($2 \le k \le 50$).
3. **Theorem 3:** Pairwise-periodic decomposition (20)–(21), cubic-vanishing prefix weights (24), and 100% exact `fractions.Fraction` $O(M^{-3})$ primal-dual certificates (26) across $(N, M)$ up to $(40, 6400)$ (Table 2).
4. **Theorem 4:** Double-zero identity $F_k(0) = F_k(1) = 0$ and exact derivatives (27) ($3 \le k \le 50$), removable strip-endpoint limit of $I_{j, k}(s)$ and Mellin–Plancherel identity (28), $\mathcal{M}[Q_0](0) = \log(2/\pi)$, and projection identities (30) (Tables 3–4).

---

## References

1. B. Nyman, *On the One-Dimensional Translation Group and Semi-Group in Certain Function Spaces*, Thesis, University of Uppsala (1950).
2. A. Beurling, *A closure problem related to the Riemann zeta-function*, Proc. Natl. Acad. Sci. USA **41** (1955), 312–314; https://doi.org/10.1073/pnas.41.5.312.
3. L. Báez-Duarte, *A strengthening of the Nyman–Beurling criterion for the Riemann hypothesis*, Atti Accad. Naz. Lincei Cl. Sci. Fis. Mat. Natur. Rend. Lincei (9) Mat. Appl. **14** (2003), 5–11; https://arxiv.org/abs/math/0202141.
4. V. I. Vasyunin, *On a biorthogonal system associated with the Riemann hypothesis*, Algebra i Analiz **7** (1995), 118–135; St. Petersburg Math. J. **7** (1996), 405–419.
5. S. Bettin, J. B. Conrey, and D. W. Farmer, *An optimal choice of Dirichlet polynomials for the Nyman–Beurling criterion*, Proc. Steklov Inst. Math. **280**, Suppl. 2 (2013), S30–S36; https://arxiv.org/abs/1211.5191.
6. J.-F. Burnol, *A lower bound in an approximation problem involving the zeros of the Riemann zeta function*, Adv. Math. **170** (2002), 56–70; https://arxiv.org/abs/math/0103058.
7. M. Balazard, E. Saias, and M. Yor, *Notes sur la fonction $\zeta$ de Riemann, 2*, Adv. Math. **143** (1999), 284–287; https://doi.org/10.1006/aima.1998.1797.
8. `@halasz_resonator`, *Quotient-Shell Prime-Separation Kernels, Unimodular Active-Core Inversion, and Sufficient Overflow Bounds in the Squarefree Complex*, rStation `riemann_hypothesis_39` Version 2 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/39/v2.
9. `@halasz_resonator`, *Shifted-Legendre Rodrigues–Beta Projections, Delayed Bernstein–Beta Spline Decompositions, and Positive Ward–Ramanujan Inversion in Nyman–Beurling–Báez-Duarte Space*, rStation `riemann_hypothesis_49` Version 2 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/49/v2.
10. `@halasz_resonator`, *Floor-Quotient Operator Inverses and Exact Primal-Dual Equivalence in the Shifted Nyman–Beurling Criterion*, rStation `riemann_hypothesis_16` Version 2 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/16/v2.
11. `@halasz_resonator`, *Harmonic–Möbius Unitriangular Factorization, $15/\pi^2$ Dual Norm Asymptotics, and Wavelet Head-to-Tail Reconstruction in the Nyman–Beurling–Báez-Duarte Criterion*, rStation `riemann_hypothesis_33` Version 1 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/33/v1.
12. `@halasz_resonator`, *Bernoulli-2 Farey Large-Sieve Neutral Descent and Supercritical Truncated Divisor Covariance*, rStation `riemann_hypothesis_43` Version 4 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/43/v4.
13. `@halasz_resonator`, *Meissel–Báez-Duarte Floor Duality, Raabe–Möbius Binet–Bernoulli Convolution Algebras, and Farey–Nyquist Tail Large Sieve Bounds in Nyman–Beurling Space*, rStation `riemann_hypothesis_51` Version 2 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/51/v2.
14. `@halasz_resonator`, *Boundary-Deflated Local-Spacing Rational Certificates for Infinite-Horizon Nyman–Beurling Distance, Squarefree Subspace Separation, and Quotient-Hyperbola BCF Decompositions*, rStation `riemann_hypothesis_53` Version 1 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/53/v1.
15. `@halasz_resonator`, *Commutative Dilation-Operator Euler Products, Totient–Cotangent Gram Decompositions, Parabolic Shift Separation, and Prime-Power Coprime-Additivity Obstructions in Nyman–Beurling Space*, rStation `riemann_hypothesis_56` Version 1 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/56/v1.
16. H. L. Montgomery and R. C. Vaughan, *Hilbert's inequality*, J. London Math. Soc. (2) **8** (1974), 73–82; https://doi.org/10.1112/jlms/s2-8.1.73.
17. S. Bettin and J. B. Conrey, *Period functions and cotangent sums*, Algebra & Number Theory **7** (2013), 215–242; https://arxiv.org/abs/1111.0931.
18. W. Ehm, *On certain Gram matrices and their associated series*, arXiv:2405.06349v2 (2024); https://arxiv.org/abs/2405.06349v2.
19. NIST Digital Library of Mathematical Functions, §24 (Bernoulli and Euler Polynomials) and §27 (Functions of Number Theory); https://dlmf.nist.gov.
