# Boundary-Deflated Local-Spacing Rational Certificates for Infinite-Horizon Nyman–Beurling Distance, Squarefree Subspace Separation, and Quotient-Hyperbola BCF Decompositions

## Abstract

Extending the finite-prefix rational certificates and unweighted Farey large-sieve bounds of [13], we construct boundary-deflated local-spacing rational primal-dual certificates (BD-WMV) for the unconstrained infinite-horizon Nyman–Beurling–Báez-Duarte distance $d_N^2 = \inf_{g \in V_N}\|1 - g\|_{\mathcal{H}}^2$ in $L^2((1, \infty), x^{-2}\,dx)$. Combining an exact rational boundary identity for the extended-Farey Hilbert primitive $U(M)$ (with $r_0 \equiv 1$) with Montgomery–Vaughan's weighted local-spacing inequality ($\delta_\xi^{-1} \le k(N-\mathbf{1}_{k=N})$) and superblock-LCM accumulation cuts certified sandwich widths by $2.11\times\text{–}2.63\times$ over the unweighted extended-Farey baseline across $N \in \{10, 20, 40, 80, 160\}$, pinching $d_{80}^2 \in [0.0109977079, 0.0109998843]$ ($M=25{,}600$) and $d_{160}^2 \in [0.0093493182, 0.0093638252]$. We also certify strict separation $d_N^2 < d_N^{\mathrm{sqf}, 2}$ from Báez-Duarte's squarefree subspace and prove all-$n$ quotient-hyperbola BCF identities. These results do not resolve RH.

---

## 1. Introduction, Relationship to Prior Framework, and Summary of Main Results

### 1.1 Background, Inherited Framework, and the Unweighted Extended-Farey Baseline

In the Nyman–Beurling–Báez-Duarte (NBBD) Hilbert-space criterion for the Riemann Hypothesis [1, 2, 3], one studies the squared distance $d_N^2 := \inf_{g \in V_N}\|1 - g\|_{\mathcal{H}}^2$ in $\mathcal{H} := L^2((1, \infty), x^{-2}\,dx)$ from the constant function $1$ to the $(N-1)$-dimensional dilate subspace
$$V_N \;:=\; \operatorname{span}\{\rho_d : 2 \le d \le N\} \;=\; \operatorname{span}\{\tau_k : 2 \le k \le N\}, \qquad \rho_d(x) \;:=\; \left\lfloor\frac{x}{d}\right\rfloor - \frac{\lfloor x \rfloor}{d}, \qquad \tau_k(x) \;:=\; \sum_{d \mid k} d\,\mu(k/d)\left\lfloor\frac{x}{d}\right\rfloor. \tag{1}$$
By Báez-Duarte's theorem [3], RH holds if and only if $\lim_{N \to \infty} d_N^2 = 0$ (and equally if and only if $\lim_{N \to \infty} d_N^{\mathrm{sqf}, 2} = 0$ for the squarefree subspace $V_N^{\mathrm{sqf}} := \operatorname{span}\{\rho_d : 2 \le d \le N,\, \mu(d) \neq 0\}$). Unconditionally, Burnol [6] and Balazard–Saias–Yor [7] proved $\liminf_{N \to \infty} d_N^2\log N \ge C_\zeta := 2 + \gamma - \log(4\pi) \approx 0.046191$, while Bettin–Conrey–Farmer [5, Theorem 1] established a conditional matching upper-bound asymptotic for the logarithmic Möbius taper $\xi_d^{\mathrm{BCF}} = A_0\,\mu(d)\bigl(1 - \frac{\log d}{\log N}\bigr) \in V_N^{\mathrm{sqf}}$.

We first delimit the existing machinery inherited from our published papers `riemann_hypothesis_43/v4` [12], `riemann_hypothesis_49/v2` [9], and `riemann_hypothesis_51/v2` [13] (building on [8, 10, 11]):
- Every $g = \sum_{k=2}^N c_k\tau_k = \sum_{d=1}^N \xi_d\lfloor x/d \rfloor \in V_N$ ($\xi_d = d\sum_{\ell=1}^{\lfloor N/d \rfloor}\mu(\ell)c_{d\ell}$, $\sum_{d=1}^N \xi_d/d = 0$ [9, Theorem 3]) belongs to the integer step-function subspace $\mathcal{S}_{\mathbb{Z}} \subset \mathcal{H}$ with norm $\|1 - g\|_{\mathcal{H}}^2 = \sum_{n=1}^\infty \frac{r_n^2}{n(n+1)}$, where $r_n := 1 - g(n)$ for all $n \ge 0$. Since $\tau_k(n) \equiv -\frac{1}{2}\varphi(k) - \mathcal{B}_{1, k}^{\mathrm{true}}(n)$ for all $n \ge 0$ (where $\mathcal{B}_{1, k}^{\mathrm{true}}(n) := \sum_{d \mid k} d\mu(k/d)B_1(\{n/d\})$), the step residual splits on $\mathbb{Z}_{\ge 0}$ into a constant shift plus a zero-mean periodic polynomial [12, 13]:
  $$r_n \;=\; 1 - g(n) \;\equiv\; \beta(\mathbf{c}) \;+\; P(n) \qquad (n \ge 0), \qquad \beta(\mathbf{c}) \;:=\; 1 + \frac{1}{2}\sum_{k=2}^N c_k\varphi(k), \qquad P(n) \;:=\; \sum_{k=2}^N c_k\,\mathcal{B}_{1, k}^{\mathrm{true}}(n). \tag{2}$$
  At $n = 0$, $\tau_k(0) = 0$ and $B_1(0) = -\frac{1}{2}$ give $P(0) = -\frac{1}{2}\sum_{k=2}^N c_k\varphi(k) = 1 - \beta(\mathbf{c})$, so that $r_0 = 1 - g(0) = \beta(\mathbf{c}) + P(0) \equiv 1$ identically for every $\mathbf{c} \in \mathbb{R}^{N-1}$.
- In `riemann_hypothesis_51/v2` [13, Theorem 2 and Theorem 4, Eqs. (41), (44), (44′), (47), and (48′)], we established rational primal-dual certificates for the *finite-prefix* distance $d_{N, M}^2 := \inf_{g \in V_N}\sum_{n=1}^M \frac{r_n^2}{n(n+1)}$, the wavelet jump recurrence $\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c} \equiv -\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c} - \sum_{m=1}^n \frac{\Delta_m g}{m}$, the Hilbert primitive $U_P(n)$ on the non-zero Farey set $\mathcal{F}_N^* := \{b/k : 2 \le k \le N,\, 1 \le b < k,\, \gcd(b, k) = 1\}$ with Parseval mean square $E(\mathbf{c}) := \sum_{k=2}^N \frac{J_2(k)}{12}c_k^2$, the retained boundary term on $\mathcal{F}_N^*$, and unweighted Farey tail bounds using the global spacing $K := N(N - 1)$.

To certify the *unconstrained infinite-horizon* optimum $d_N^2 = \inf_{g \in V_N}\|1 - g\|_{\mathcal{H}}^2$ without restricting to the centered hyperplane $\beta(\mathbf{c}) = 0$, the natural reference baseline is the **Unweighted Extended-Farey Baseline** obtained by applying [13, Eq. (47)] directly to the full residual $r_n = \beta(\mathbf{c}) + P(n)$ on the extended Farey set
$$\mathcal{F}_N \;:=\; \{0\} \cup \mathcal{F}_N^*, \qquad r_n \;\equiv\; \sum_{\xi \in \mathcal{F}_N} a_\xi\,e^{2\pi i \xi n} \quad (n \ge 0), \qquad a_0 := \beta(\mathbf{c}), \qquad a_{b/k} := -\frac{c_k\,e^{\pi i b/k}}{2i\sin(\pi b/k)} \quad \bigl(\gcd(b, k) = 1\bigr). \tag{3}$$
Since any two distinct frequencies $\xi \ne \xi'$ in $\mathcal{F}_N$ are separated by $\|\xi - \xi'\|_{\mathbb{R}/\mathbb{Z}} \ge \frac{1}{K}$ ($K := N(N - 1)$) and the mean square of $r_n$ is $\widetilde{E}(\mathbf{c}) := \sum_{\xi \in \mathcal{F}_N}|a_\xi|^2 \equiv \beta(\mathbf{c})^2 + E(\mathbf{c})$, Montgomery and Vaughan's unweighted Hilbert inequality [16, Theorem 1] gives $\bigl|\sum_{n=M+1}^{M+L} r_n^2 - L\,\widetilde{E}(\mathbf{c})\bigr| \le K\,\widetilde{E}(\mathbf{c})$ for all $L \ge 1$. Summation by parts against $\frac{1}{n(n+1)}$ as in [13, Theorem 3(c)] yields the **Unweighted Extended-Farey Baseline Sandwich**
$$\frac{\beta(\mathbf{c})^2 + E(\mathbf{c})}{M + K + 1} \;\le\; \sum_{n=M+1}^\infty \frac{r_n^2}{n(n + 1)} \;\le\; \frac{1 + \frac{K}{M + 2}}{M + 1}\bigl(\beta(\mathbf{c})^2 + E(\mathbf{c})\bigr), \qquad K := N(N - 1). \tag{4}$$
Adding the finite prefix $\sum_{n=1}^M \frac{r_n^2}{n(n+1)}$ produces strictly convex rational quadratic forms $Q_{\mathrm{LB}}^{\mathrm{base}}(\mathbf{c}; M) \le \|1 - g\|_{\mathcal{H}}^2 \le Q_{\mathrm{UB}}^{\mathrm{base}}(\mathbf{c}; M)$. At $N = 80, M = N^2 = 6{,}400$, exact rational KKT verification of (4) certifies $0.0109731905 \le d_{80}^2 \le 0.0110492148$ (width $7.602 \times 10^{-5}$).

### 1.2 Summary of New Contributions Beyond [9, 12, 13] and the Unweighted Baseline

1. **Boundary-Deflated Local-Spacing Rational Certificates (BD-WMV) Achieving a $2.11\times\text{–}2.63\times$ Width Reduction Over the Baseline and Extending Certification to $N = 160, M = 25{,}600$ (Section 2, Theorem 1):**
   Whereas [13, Theorem 4(b), Eq. (44′)] retained the boundary value $U_P(M)$ only on the centered hyperplane $\beta(\mathbf{c}) = 0$ and bounded the cubic tail via the unweighted global spacing $K = N(N-1)$, we extend both ingredients to the full unconstrained space $V_N$:
   - **(a) Exact Rational Initial-Value & Boundary Identity on $\mathcal{F}_N$:** On $\mathcal{F}_N \times \mathcal{F}_N$, the off-diagonal Hilbert primitive $U(n) := \sum_{\xi \ne \xi' \in \mathcal{F}_N} a_\xi\overline{a_{\xi'}}\frac{e^{\pi i(\xi-\xi')(2n+1)}}{2i\sin(\pi(\xi-\xi'))}$ satisfies $U_P(-1) \equiv 0$ on $\mathcal{F}_N^* \times \mathcal{F}_N^*$, and its cross term with $\xi = 0$ evaluates in exact rational closed form to $U(-1) \equiv 2\beta(\mathbf{c})\sum_{k=2}^N \frac{J_2(k)}{12}c_k$. Since $r_0 = 1 - g(0) \equiv 1$ for all $\mathbf{c} \in \mathbb{R}^{N-1}$, summing $U(m) - U(m-1) = r_m^2 - \widetilde{E}(\mathbf{c})$ from $m = 0$ to $m = M$ expresses $U(M)$ as an **exact rational quadratic polynomial** in $\mathbf{c}$:
     $$U(M) \;\equiv\; 2\,\beta(\mathbf{c})\sum_{k=2}^N \frac{J_2(k)}{12}\,c_k \;+\; 1 \;+\; \sum_{m=1}^M r_m^2 \;-\; (M + 1)\bigl(\beta(\mathbf{c})^2 + E(\mathbf{c})\bigr). \tag{5}$$
   - **(b) Weighted Local-Spacing Montgomery–Vaughan Inequality on $\mathcal{F}_N$:** Applying Montgomery and Vaughan's weighted Hilbert inequality [16, Theorem 2] with the exact per-denominator minimum spacing $\delta_{b/k}^{-1} \le k(N - \mathbf{1}_{k=N})$ ($2 \le k \le N$) and $\delta_0^{-1} = N$ yields for all $n \in \mathbb{Z}$ the pointwise envelope $|U(n)| \le \frac{1}{2}V_{\mathrm{WMV}}(\mathbf{c})$, where
     $$V_{\mathrm{WMV}}(\mathbf{c}) \;:=\; \frac{3}{2}N\,\beta(\mathbf{c})^2 \;+\; \sum_{k=2}^N \frac{3}{2}k\bigl(N - \mathbf{1}_{k=N}\bigr)\frac{J_2(k)}{12}\,c_k^2. \tag{6}$$
   - **(c) Boundary-Deflated Strictly Convex Rational Envelopes, Small-$N$ Analytical Verification, and Superblock-LCM Certification:** Substituting (5)–(6) into $\sum_{n=M+1}^\infty \frac{r_n^2}{n(n+1)} \equiv \frac{\widetilde{E}(\mathbf{c})}{M+1} - \frac{U(M)}{(M+1)(M+2)} + 2\sum_{n=M+1}^\infty \frac{U(n)}{n(n+1)(n+2)}$ transfers $-\frac{1}{(M+1)(M+2)}\sum_{n=1}^M r_n^2$ into the finite prefix (giving positive deflated weights $w_n^{\mathrm{BD}} := \frac{1}{n(n+1)} - \frac{1}{(M+1)(M+2)} > 0$) and shrinks the remaining tail uncertainty width to $\frac{V_{\mathrm{WMV}}(\mathbf{c})}{(M+1)(M+2)}$. We prove by an exact rational completion of the square that the lower envelope $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M)$ is **unconditionally strictly convex for all $N \ge 2$ and $M \ge N^2$** (with $\beta^2$-coefficient $C_\beta(N, M) := 2M + 3 - \frac{3}{4}N - \frac{1}{N}\sum_{k=2}^N \frac{J_2(k)}{12} \ge \frac{75}{8} = 9.375 > 9$), and we verify the envelopes analytically against $\|1 - c\tau_2\|_{\mathcal{H}}^2 = 1 + (2c + c^2)\log 2$ at $N = 2, M = 4$ (Remark 1). Combining a 4-chunk exact integer solver with 8-superblock LCM rational accumulation reduces certified sandwich widths by **$2.11\times\text{–}2.63\times$ over the baseline (4)** across $N \in \{10, 20, 40, 80, 160\}$ (Table 1).

2. **All-$n$ Chebyshev–Mertens Quotient-Hyperbola BCF Decompositions and Sub-Nyquist Drift Diagnostics (Section 3, Theorem 2):**
   For the Bettin–Conrey–Farmer taper $\xi_d^{\mathrm{BCF}} = A_0\,\mu(d)\bigl(1 - \frac{\log d}{\log N}\bigr)$, we establish exact all-$n$ quotient-hyperbola representations for $g^{\mathrm{BCF}}(n)$ and $\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c}^{\mathrm{BCF}}$ in terms of $\psi(n)$, $\sum_{m=2}^n \frac{\Lambda(m)}{m}$, and the Mertens remainders $M_N(n/q), \mathcal{R}_N(n/q)$ summed over $1 \le q \le \lfloor n/(N+1) \rfloor$. On $1 \le n \le N$, the quotient sums vanish identically, giving $|g^{\mathrm{BCF}}(n) - A_0| \le (4\log 2 + 1)|A_0|\frac{n}{\log N}$. On $N < n \le \lfloor N^{3/2} \rfloor$, specializing $\Delta_m g = \sum_{d \mid m,\, d \le N}\xi_d$ [13, Eq. (41)] to primes $p > N$ gives $\Delta_p g \equiv \xi_1$, producing a prime-jump drift $D_{\mathrm{prime}}(n) := -\xi_1\sum_{N < p \le n}\frac{1}{p}$ of energy $\sum_{N < n \le N^{3/2}} |D_{\mathrm{prime}}(n)|^2 \sim (\xi_1\log\frac{3}{2})^2 N^{3/2} = \Theta(N^{3/2})$ ($\ell^2$-norm $\Theta(N^{3/4})$) whose cancellation against the composite-jump drift $D_{\mathrm{comp}}(n)$ is evaluated across $N \in [40, 10{,}000]$ in Table 2.

3. **Exact Subspace Equivalence of Squarefree $\rho_d$ and Squarefree $\tau_k$, and Exact Rational Certification of $d_N^2 < d_N^{\mathrm{sqf}, 2}$ and Intermediate Radical-Harmonic Subspaces (Section 4, Theorem 3):**
   Because the upper-triangular coordinate bijection $w_d = \xi_d = d\sum_{\ell=1}^{\lfloor N/d \rfloor}\mu(\ell)c_{d\ell}$ ($2 \le d \le N$) preserves non-squarefree indices ($\mu(d) = 0 \implies \mu(d\ell) = 0$), Báez-Duarte's squarefree dilate subspace [3] satisfies $V_N^{\mathrm{sqf}} := \operatorname{span}\{\rho_d : 2 \le d \le N,\, \mu(d) \neq 0\} \equiv \operatorname{span}\{\tau_k : 2 \le k \le N,\, \mu(k) \neq 0\}$. Applying our exact rational BD-WMV certification to $V_N^{\mathrm{sqf}}$ and to the intermediate coordinate subspaces $V_N^{\mathrm{sqf}+\mathrm{pp}}$ (adjoining prime powers $p^a$, $a \ge 2$) and $V_N^{(2)}$ (adjoining first radical harmonics $k = 2\operatorname{rad}(k)$), we certify in **exact rational arithmetic (`fractions.Fraction`, zero transcendental constants)** across $N \in \{4, 10, 20, 40, 80, 160\}$ (Table 3) that $\mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^2) < \mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^{\mathrm{sqf}, 2})$ (with $0.0004406620 \le d_{80}^{\mathrm{sqf}, 2} - d_{80}^2 \le 0.0004446833$ at $M = 25{,}600$ and $0.0003010484 \le d_{160}^{\mathrm{sqf}, 2} - d_{160}^2 \le 0.0003287547$), and that $d_N^2 < d_N(V_N^{(2)})^2 < d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2 < d_N^{\mathrm{sqf}, 2}$ holds strictly at $(80, 25600)$ and $(160, 25600)$.

---

## 2. Boundary-Deflated Local-Spacing Rational Certificates for Infinite-Horizon $d_N^2$

We now state and prove Theorem 1, which upgrades the unweighted global Farey baseline (4) to the Boundary-Deflated Local-Spacing Montgomery–Vaughan (BD-WMV) rational certificate.

**Theorem 1 (Boundary-Deflated Local-Spacing Rational Primal-Dual Certificates for Infinite-Horizon $d_N^2$ and $d_N^{\mathrm{cen}, 2}$).**
*(a) (**Exact Rational Initial-Value & Boundary Identity for the Extended-Farey Hilbert Primitive $U(n)$**): For any $g = \sum_{k=2}^N c_k\tau_k \in V_N$, the Fourier polynomial (3) on $\mathcal{F}_N = \{0\} \cup \mathcal{F}_N^*$ equals $r_n = 1 - \sum_{k=2}^N c_k\tau_k(n) = \beta(\mathbf{c}) + P(n)$ for every integer $n \ge 0$ (in particular, $r_0 = 1 - \sum_{k=2}^N c_k\tau_k(0) \equiv 1$), with mean square $\widetilde{E}(\mathbf{c}) := \beta(\mathbf{c})^2 + E(\mathbf{c}) = \beta(\mathbf{c})^2 + \sum_{k=2}^N \frac{J_2(k)}{12}c_k^2$. Define the zero-mean periodic Hilbert primitive*
$$U(n) \;:=\; \sum_{\substack{\xi, \xi' \in \mathcal{F}_N \\ \xi \ne \xi'}} a_\xi\,\overline{a_{\xi'}}\,\frac{e^{\pi i(\xi - \xi')(2n + 1)}}{2i\sin\bigl(\pi(\xi - \xi')\bigr)} \qquad (n \in \mathbb{Z}). \tag{7}$$
*Then $U(n) - U(n - 1) \equiv r_n^2 - \widetilde{E}(\mathbf{c})$ for every $n \in \mathbb{Z}$. Moreover, at $n = -1$ ($2n + 1 = -1$), the non-zero Farey double sum $U_P(-1) := \sum_{\xi \ne \xi' \in \mathcal{F}_N^*} a_\xi\overline{a_{\xi'}}\frac{e^{-\pi i(\xi - \xi')}}{2i\sin(\pi(\xi - \xi'))}$ vanishes identically ($U_P(-1) \equiv 0$), while the cross terms between $\xi = 0$ and $\xi' \in \mathcal{F}_N^*$ evaluate in exact rational closed form to*
$$U(-1) \;\equiv\; 2\,\beta(\mathbf{c})\sum_{k=2}^N \frac{J_2(k)}{12}\,c_k. \tag{8}$$
*Consequently, since $r_0 \equiv 1$ ($r_0^2 \equiv 1$), the boundary value $U(M)$ at any integer cutoff $M \ge 1$ is an **exact rational quadratic polynomial** in $\mathbf{c}$:*
$$U(M) \;\equiv\; 2\,\beta(\mathbf{c})\sum_{k=2}^N \frac{J_2(k)}{12}\,c_k \;+\; 1 \;+\; \sum_{m=1}^M r_m^2 \;-\; (M + 1)\bigl(\beta(\mathbf{c})^2 + E(\mathbf{c})\bigr). \tag{9}$$

*(b) (**Weighted Local-Spacing Montgomery–Vaughan Inequality on $\mathcal{F}_N$**): For each frequency $\xi \in \mathcal{F}_N$, let $\delta_\xi := \min_{\xi' \in \mathcal{F}_N \setminus \{\xi\}} \|\xi - \xi'\|_{\mathbb{R}/\mathbb{Z}}$ denote its minimum torus distance to any distinct frequency in $\mathcal{F}_N$. Then $\delta_0 = \frac{1}{N}$, and for every primitive fraction $\xi = b/k \in \mathcal{F}_N^*$ ($2 \le k \le N$, $\gcd(b, k) = 1$),*
$$\delta_{b/k}^{-1} \;=\; \max_{\xi' \in \mathcal{F}_N \setminus \{b/k\}} \bigl\|\xi - \xi'\bigr\|_{\mathbb{R}/\mathbb{Z}}^{-1} \;\le\; k\bigl(N - \mathbf{1}_{k=N}\bigr). \tag{10}$$
*Applying Montgomery and Vaughan's weighted Hilbert inequality [16, Theorem 2] to (7) yields for every $n \in \mathbb{Z}$ and every $\mathbf{c} \in \mathbb{R}^{N-1}$ the pointwise bound*
$$\bigl|U(n)\bigr| \;\le\; \frac{1}{2}\,V_{\mathrm{WMV}}(\mathbf{c}), \qquad V_{\mathrm{WMV}}(\mathbf{c}) \;:=\; \frac{3}{2}N\,\beta(\mathbf{c})^2 \;+\; \sum_{k=2}^N \frac{3}{2}k\bigl(N - \mathbf{1}_{k=N}\bigr)\frac{J_2(k)}{12}\,c_k^2. \tag{11}$$

*(c) (**Boundary-Deflated Strictly Convex Rational Envelopes $Q_{\mathrm{LB}}^{\mathrm{BD}}$ and $Q_{\mathrm{UB}}^{\mathrm{BD}}$**): For any $N \ge 2$ and $M \ge N^2$, let $w_n^{\mathrm{BD}} := \frac{1}{n(n+1)} - \frac{1}{(M+1)(M+2)} > 0$ ($1 \le n \le M$). Then the infinite-horizon norm satisfies the exact telescoping decomposition*
$$\|1 - g\|_{\mathcal{H}}^2 \;\equiv\; \sum_{n=1}^M w_n^{\mathrm{BD}}\,r_n^2 \;+\; \frac{(2M + 3)\widetilde{E}(\mathbf{c}) - U(-1) - 1}{(M + 1)(M + 2)} \;+\; 2\sum_{n=M+1}^\infty \frac{U(n)}{n(n + 1)(n + 2)}, \tag{12}$$
*and bounding the cubic tail series via (11) sandwiches $\|1 - g\|_{\mathcal{H}}^2$ between the two **purely rational quadratic forms***
$$Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M) \;:=\; \sum_{n=1}^M w_n^{\mathrm{BD}}\,r_n^2 \;+\; \frac{(2M + 3)\widetilde{E}(\mathbf{c}) - U(-1) - 1 - \frac{1}{2}V_{\mathrm{WMV}}(\mathbf{c})}{(M + 1)(M + 2)}, \tag{13}$$
$$Q_{\mathrm{UB}}^{\mathrm{BD}}(\mathbf{c}; M) \;:=\; \sum_{n=1}^M w_n^{\mathrm{BD}}\,r_n^2 \;+\; \frac{(2M + 3)\widetilde{E}(\mathbf{c}) - U(-1) - 1 + \frac{1}{2}V_{\mathrm{WMV}}(\mathbf{c})}{(M + 1)(M + 2)}. \tag{14}$$
*Moreover, for all $N \ge 2$ and $M \ge N^2$, define the diagonal coercivity weights*
$$\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}} \;:=\; \frac{2M + 3 - N - \frac{3}{4}k(N - \mathbf{1}_{k=N})}{(M + 1)(M + 2)}\,\frac{J_2(k)}{12} \;\ge\; \frac{\frac{9}{8}M + 3}{(M + 1)(M + 2)}\,\frac{J_2(k)}{12} \;>\; 0 \qquad (2 \le k \le N). \tag{15}$$
*Then $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M) - \sum_{k=2}^N \gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}} c_k^2$ is a convex quadratic form on $\mathbb{R}^{N-1}$ with positive $\beta^2$-coefficient $C_\beta(N, M) := 2M + 3 - \frac{3}{4}N - \frac{1}{N}\sum_{k=2}^N \frac{J_2(k)}{12} \ge \frac{75}{8} = 9.375 > 9$ for every $N \ge 2, M \ge N^2$, so the Hessian of $Q_{\mathrm{LB}}^{\mathrm{BD}}$ satisfies $\frac{1}{2}\nabla^2 Q_{\mathrm{LB}}^{\mathrm{BD}} \succeq \operatorname{diag}\bigl(\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}\bigr)_{k=2}^N \succ 0$ unconditionally.*

*(d) (**Exact Superblock-LCM Rational Primal-Dual Certificates and Centered Hyperplane Specialization**): For any integer-scaled proposal $\mathbf{c} = \mathbf{C}/10^{12}$ ($\mathbf{C} \in \mathbb{Z}^{N-1}$) with half-gradient $g_k := \frac{1}{2}\frac{\partial Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M)}{\partial c_k} \in \mathbb{Q}$, completing the square yields the exact rational primal-dual sandwich*
$$\mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}(N, M) \;:=\; Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M) \;-\; \sum_{k=2}^N \frac{g_k^2}{\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}} \;\le\; d_N^2 \;\le\; Q_{\mathrm{UB}}^{\mathrm{BD}}(\mathbf{c}_{\mathrm{UB}}; M) \;=:\; \mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}(N, M). \tag{16}$$
*On the centered totient hyperplane $V_N^{\mathrm{cen}} := \{g \in V_N : \beta(\mathbf{c}) = 0\}$, $\widetilde{E}(\mathbf{c}) = E(\mathbf{c})$ and $U(-1) = 0$, and Lagrange-multiplier KKT duality with $\gamma_{k, M}^{\mathrm{CBD}, \mathrm{LB}} := \frac{2M + 3 - \frac{3}{4}k(N - \mathbf{1}_{k=N})}{(M+1)(M+2)}\frac{J_2(k)}{12} > 0$ certifies $\mathrm{LB}_{\mathbb{Q}}^{\mathrm{CBD}}(N, M) \le d_N^{\mathrm{cen}, 2} \le \mathrm{UB}_{\mathbb{Q}}^{\mathrm{CBD}}(N, M)$. While the exact optima and optimized envelopes satisfy $d_N^2 \le d_N^{\mathrm{cen}, 2}$, for approximate dual certificates the inequality $\mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}(N, M) \le \mathrm{LB}_{\mathbb{Q}}^{\mathrm{CBD}}(N, M)$ is verified explicitly at each $(N, M)$.*

*Proof.*
**(a)** For each $2 \le k \le N$ and every $n \in \mathbb{Z}$, the periodic Bernoulli function $\mathcal{B}_{1, k}^{\mathrm{true}}(n) = \sum_{d \mid k} d\mu(k/d)B_1(\{n/d\})$ has exact finite Fourier series $-\sum_{\gcd(b, k)=1}\frac{e^{\pi i b/k}}{2i\sin(\pi b/k)}e^{2\pi i(b/k)n}$ [13, Theorem 2(b)] (including at $n = 0$, where $\sum_{\gcd(b, k)=1}\frac{\cos(\pi b/k)}{2\sin(\pi b/k)} = 0$ by pairing $b$ with $k - b$ and $-\sum_{\gcd(b, k)=1}\frac{i\sin(\pi b/k)}{2i\sin(\pi b/k)} = -\frac{1}{2}\varphi(k) = \mathcal{B}_{1, k}^{\mathrm{true}}(0)$). Thus (3) equals $r_n = 1 - \sum_{k=2}^N c_k\tau_k(n)$ for all $n \ge 0$, and at $n = 0$ we have $r_0 = 1 - \sum_{k=2}^N c_k\tau_k(0) \equiv 1$.
Using $e^{\pi i\delta(2n+1)} - e^{\pi i\delta(2n-1)} = 2i\sin(\pi\delta)\,e^{2\pi i\delta n}$ at $\delta = \xi - \xi'$, differencing (7) gives $U(n) - U(n-1) = \sum_{\xi \ne \xi' \in \mathcal{F}_N} a_\xi\overline{a_{\xi'}}e^{2\pi i(\xi - \xi')n} = r_n^2 - \widetilde{E}(\mathbf{c})$. At $n = -1$ ($2n + 1 = -1$), for every $\xi = b/k \in \mathcal{F}_N^*$, $\bar{\xi} := 1 - \xi = \frac{k-b}{k} \in \mathcal{F}_N^*$ satisfies $a_{\bar{\xi}} = \overline{a_\xi}$. Defining $\widetilde{c}_\xi := a_\xi e^{-\pi i\xi} = -\frac{c_k}{2i\sin(\pi\xi)} \in i\mathbb{R}$, we have $\overline{\widetilde{c}_\xi} = -\widetilde{c}_\xi$ and $\widetilde{c}_{\bar{\xi}} = +\widetilde{c}_\xi$. Under the involution $(\xi, \xi') \mapsto (\bar{\xi}, \bar{\xi'})$ on $\mathcal{F}_N^* \times \mathcal{F}_N^*$, each summand $-\frac{\widetilde{c}_\xi\widetilde{c}_{\xi'}}{2i\sin(\pi(\xi - \xi'))}$ of $U_P(-1)$ negates since $\sin(\pi(\bar{\xi} - \bar{\xi'})) = -\sin(\pi(\xi - \xi'))$, so $U_P(-1) \equiv 0$.
For the cross terms in $U(-1)$ between $\xi = 0$ ($a_0 = \beta(\mathbf{c}) \in \mathbb{R}$) and $\xi' = b/k \in \mathcal{F}_N^*$, pairing $(0, \xi')$ with $(\xi', 0)$ yields
$$\beta(\mathbf{c})\left(\overline{a_{\xi'}}\,\frac{e^{\pi i\xi'}}{-2i\sin(\pi\xi')} + a_{\xi'}\,\frac{e^{-\pi i\xi'}}{2i\sin(\pi\xi')}\right) \;=\; 2\,\beta(\mathbf{c})\,\frac{c_k}{4\sin^2(\pi b/k)}.$$
Summing over $2 \le k \le N$ and $\gcd(b, k) = 1$ via $\sum_{\gcd(b, k)=1}\frac{1}{4\sin^2(\pi b/k)} = \frac{J_2(k)}{12}$ [13, Proof of Theorem 2(b); 18] establishes (8). Summing $U(m) - U(m-1) = r_m^2 - \widetilde{E}(\mathbf{c})$ from $m = 0$ to $m = M$ and substituting $r_0^2 \equiv 1$ gives (9).

**(b)** For $\xi = 0 = 0/1$, the closest fractions in $\mathcal{F}_N^*$ are $1/N$ and $(N-1)/N$, so $\delta_0 = \frac{1}{N}$. For $\xi = b/k \in \mathcal{F}_N^*$ ($2 \le k \le N$) and any $\xi' = b'/l \in \mathcal{F}_N \setminus \{b/k\}$ ($1 \le l \le N$), if $l = k$ then $\|\xi - \xi'\| \ge \frac{1}{k} \ge \frac{1}{k(N-1)}$; if $l \ne k$, then $l \le N - \mathbf{1}_{k=N}$, so $\|\frac{b}{k} - \frac{b'}{l}\| \ge \frac{|b l - b' k|}{k l} \ge \frac{1}{k(N - \mathbf{1}_{k=N})}$, proving (10). By Montgomery and Vaughan's weighted Hilbert inequality [16, Theorem 2], $\bigl|\sum_{r \ne s}\frac{u_r\overline{u_s}}{\sin(\pi(\xi_r - \xi_s))}\bigr| \le \frac{3}{2}\sum_r \delta_r^{-1}|u_r|^2$. Applying this at $u_\xi = a_\xi e^{\pi i\xi(2n+1)}$ (with the factor $\frac{1}{2}$ in $\frac{1}{2i\sin(\pi(\xi-\xi'))}$) and using $\sum_{\gcd(b, k)=1}|a_{b/k}|^2 = \frac{J_2(k)}{12}c_k^2$ gives $|U(n)| \le \frac{3}{4}\sum_{\xi \in \mathcal{F}_N}\delta_\xi^{-1}|a_\xi|^2 \le \frac{1}{2}V_{\mathrm{WMV}}(\mathbf{c})$, proving (11).

**(c)** Summation by parts on $n \ge M + 1$ using $r_n^2 = \widetilde{E}(\mathbf{c}) + U(n) - U(n-1)$ and $\frac{1}{n(n+1)} - \frac{1}{(n+1)(n+2)} = \frac{2}{n(n+1)(n+2)}$ gives
$$\sum_{n=M+1}^\infty \frac{r_n^2}{n(n + 1)} \;\equiv\; \frac{\widetilde{E}(\mathbf{c})}{M + 1} \;-\; \frac{U(M)}{(M + 1)(M + 2)} \;+\; 2\sum_{n=M+1}^\infty \frac{U(n)}{n(n + 1)(n + 2)}.$$
Substituting (9) for $U(M)$ yields (12), and bounding $|U(n)| \le \frac{1}{2}V_{\mathrm{WMV}}(\mathbf{c})$ via $2\sum_{n=M+1}^\infty \frac{1}{n(n+1)(n+2)} = \frac{1}{(M+1)(M+2)}$ yields $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M) \le \|1 - g\|_{\mathcal{H}}^2 \le Q_{\mathrm{UB}}^{\mathrm{BD}}(\mathbf{c}; M)$ in (13)–(14).
To prove unconditional strict convexity of $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M)$ for all $N \ge 2$ and $M \ge N^2$, we complete the square on $-U(-1) = -2\beta(\mathbf{c})\sum_{k=2}^N \frac{J_2(k)}{12}c_k$: for each $2 \le k \le N$,
$$-2\,\beta(\mathbf{c})\,\frac{J_2(k)}{12}\,c_k \;\equiv\; \frac{J_2(k)}{12\,N}\bigl(N\,c_k - \beta(\mathbf{c})\bigr)^2 \;-\; N\,\frac{J_2(k)}{12}\,c_k^2 \;-\; \frac{J_2(k)}{12\,N}\,\beta(\mathbf{c})^2.$$
Substituting this identity into (13) decomposes $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M)$ as a sum of manifestly convex terms:
$$Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M) \;\equiv\; \sum_{n=1}^M w_n^{\mathrm{BD}}\,r_n^2 \;+\; \sum_{k=2}^N \frac{J_2(k)\bigl(N c_k - \beta(\mathbf{c})\bigr)^2}{12\,N(M + 1)(M + 2)} \;+\; \frac{C_\beta(N, M)\,\beta(\mathbf{c})^2 - 1}{(M + 1)(M + 2)} \;+\; \sum_{k=2}^N \gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}\,c_k^2,$$
where $\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}$ is defined in (15) and $C_\beta(N, M) := 2M + 3 - \frac{3}{4}N - \frac{1}{N}\sum_{k=2}^N \frac{J_2(k)}{12}$. For every $N \ge 2$ and $M \ge N^2$:
1. Since $J_2(k) \le k^2 - 1$, $\sum_{k=2}^N \frac{J_2(k)}{12} \le \frac{N(N-1)(2N+5)}{72} < \frac{N^3}{36} + \frac{N^2}{24}$, so $C_\beta(N, M) > 2N^2 + 3 - \frac{3}{4}N - \frac{N^2}{36} - \frac{N}{24} = \frac{71}{36}N^2 - \frac{19}{24}N + 3 \ge \frac{335}{36} > 9.30 > 0$ for all $N \ge 2$ (at $N = 2, M = 4$, $J_2(2) = 3$ gives the exact minimum $C_\beta(2, 4) = 11 - \frac{3}{2} - \frac{1}{8} = \frac{75}{8} = 9.375$).
2. Since $\max_{2 \le k \le N} k(N - \mathbf{1}_{k=N}) = N(N - 1)$ and $N \le \frac{1}{2}M$, $2M + 3 - N - \frac{3}{4}k(N - \mathbf{1}_{k=N}) \ge 2M + 3 - \frac{3}{4}N^2 - \frac{1}{4}N \ge \frac{9}{8}M + 3 > 0$.
Thus $\frac{1}{2}\nabla^2 Q_{\mathrm{LB}}^{\mathrm{BD}} \succeq \operatorname{diag}\bigl(\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}\bigr)_{k=2}^N \succ 0$ unconditionally for all $N \ge 2, M \ge N^2$. **(d)** Minimizing $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M) + 2\mathbf{g}^{\mathsf{T}}(\mathbf{x} - \mathbf{c}) + \sum_{k=2}^N \gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}(x_k - c_k)^2$ over $\mathbf{x} \in \mathbb{R}^{N-1}$ gives (16). $\blacksquare$

**Remark 1 (Explicit Analytical Verification at $N = 2, M = 4$).** When $N = 2, M = 4$ and $g = c\tau_2$ ($c \in \mathbb{R}$), we have $\varphi(2) = 1$, $J_2(2) = 3$, $\beta(c) = 1 + \frac{c}{2}$, $E(c) = \frac{c^2}{4}$, $U(-1) = \frac{c(2+c)}{4}$, and $\tau_2(n) = -\mathbf{1}_{n\text{ odd}}$, so $r_n = 1 + c\mathbf{1}_{n\text{ odd}}$ and $\|1 - c\tau_2\|_{\mathcal{H}}^2 = 1 + (2c + c^2)\log 2$, with true optimum $d_2^2 = 1 - \log 2 \approx 0.3068528194$ at $c^* = -1$ ($\beta(c^*) = \frac{1}{2}$). Substituting $\sum_{m=1}^4 r_m^2 = 2(1+c)^2 + 2$ and $r_0^2 = 1$ into (9) yields $U(4) = -\frac{c(2+c)}{4}$, which matches the Fourier definition (7) identically (e.g., $U(4) = -\frac{3}{4}$ at $c = 1$). Since $V_{\mathrm{WMV}}(c) = 3\beta(c)^2 + \frac{3}{4}c^2$ and $(M+1)(M+2) = 30$, the rational envelopes (13)–(14) simplify for all $c \in \mathbb{R}$ to
$$Q_{\mathrm{LB}}^{\mathrm{BD}}(c; 4) \;\equiv\; \frac{2}{3}(c + 1)^2 \;+\; \frac{17}{60} \;\le\; 1 + (2c + c^2)\log 2 \;\le\; \frac{43}{60}(c + 1)^2 \;+\; \frac{1}{3} \;\equiv\; Q_{\mathrm{UB}}^{\mathrm{BD}}(c; 4),$$
since $\frac{2}{3} < \log 2 \approx 0.693147 < \frac{43}{60}$ and $\frac{17}{60} = 0.2833333\dots < 1 - \log 2 \approx 0.3068528 < \frac{1}{3} = 0.3333333\dots$. In particular, at $c = -1$ we obtain $\frac{17}{60} \le d_2^2 = 1 - \log 2 \le \frac{1}{3}$, and at $c = 2$ we obtain $\frac{377}{60} \approx 6.2833333 < 1 + 8\log 2 \approx 6.5451774 < \frac{407}{60} \approx 6.7833333$. Both $N = 2, M = 4$ and $N = 3, M = 9$ (6-periodic exact norm check) are verified automatically in `src/verify_transcendental_free_infinite_certificates.py`.

### 2.1 Algorithmic Implementation and Head-to-Head Comparison with the Unweighted Baseline

To extend exact rational certification to $M = 25{,}600$ and $N = 160$, `src/verify_transcendental_free_infinite_certificates.py` combines: (1) **4-Chunk Exact Integer Prefix Assembly**, splitting $w_n \in [0, 10^{24})$ into base-$10^6$ digits $w_{n, q} < 10^6$ so each chunk matrix product $(\mathrm{T}\operatorname{diag}(w_{\cdot, q}))\mathrm{T}^{\mathsf{T}}$ has entries $\le 6.55 \times 10^{14} < 2^{50} \ll 2^{53}$ and is evaluated bit-for-bit identically in IEEE-754 `float64` BLAS before pure-Python integer Gaussian elimination (`solve_int_gauss`); and (2) **8-Superblock LCM Rational Accumulation**, partitioning $\{1, \dots, M\}$ into $8$ contiguous superblocks with local LCM denominators $L_s$ ($\approx 4{,}600$ bits at $M = 25{,}600$, vs. $36{,}907$ bits for $L_{M+1}$) before a single final lift to $L_{M+1}$.

Table 1 reports the head-to-head comparison between the Unweighted Extended-Farey Baseline (4) and our Boundary-Deflated Local-Spacing (BD-WMV) certificate (16) from `src/verify_transcendental_free_infinite_certificates.py` (`PASS` in `24.3s` $\to$ `results/transcendental_free_infinite_certificates.json`). At every $(N, M)$, the BD-WMV interval is **strictly nested inside** the baseline interval and reduces the certified sandwich width by **$2.11\times\text{–}2.63\times$**. At $(N, M) = (80, 6400)$, BD-WMV narrows the baseline $[0.0109731905, 0.0110492148]$ (width $7.602 \times 10^{-5}$) by **$2.184\times$** to $[0.0109812876, 0.0110160927]$ (width $3.481 \times 10^{-5}$); increasing $M$ to $4 N^2 = 25{,}600$ pinches $d_{80}^2$ to $[0.0109977079, 0.0109998843]$ (width $2.18 \times 10^{-6}$, relative width $0.0198\%$, **$34.9\times$ narrower** than the $M=6400$ baseline), and at $N = 160, M = 25{,}600$ BD-WMV certifies $d_{160}^2 \in [0.0093493182, 0.0093638252]$ (relative width $0.155\%$).

| $(N,\, M)$ | Unweighted Baseline $\mathrm{LB}_{\mathbb{Q}}^{\mathrm{base}}$ | Unweighted Baseline $\mathrm{UB}_{\mathbb{Q}}^{\mathrm{base}}$ | Baseline Width | BD-WMV Exact $\mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}$ | BD-WMV Exact $\mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}$ | BD-WMV Width (Rel. Width) | Width Gain vs. Baseline | BD-WMV Duality Defect | Centered BD-WMV $[\mathrm{LB}_{\mathbb{Q}}^{\mathrm{CBD}},\, \mathrm{UB}_{\mathbb{Q}}^{\mathrm{CBD}}]$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $(10,\, 100)$ | $0.0230842305$ | $0.0251970608$ | $2.113 \times 10^{-3}$ | $0.0233744486$ | $0.0243189313$ | $9.445 \times 10^{-4}$ ($4.04\%$) | **$2.237\times$** | $3.71 \times 10^{-22}$ | $[0.0237975305,\, 0.0247514349]$ |
| $(20,\, 400)$ | $0.0163185935$ | $0.0169562888$ | $6.377 \times 10^{-4}$ | $0.0163974054$ | $0.0166775327$ | $2.801 \times 10^{-4}$ ($1.71\%$) | **$2.276\times$** | $1.39 \times 10^{-21}$ | $[0.0163984168,\, 0.0166789183]$ |
| $(40,\, 1600)$ | $0.0126580436$ | $0.0128830062$ | $2.250 \times 10^{-4}$ | $0.0126824994$ | $0.0127889200$ | $1.064 \times 10^{-4}$ ($0.84\%$) | **$2.114\times$** | $8.35 \times 10^{-21}$ | $[0.0126827449,\, 0.0127892318]$ |
| $(80,\, 6400)$ | $0.0109731905$ | $0.0110492148$ | $7.602 \times 10^{-5}$ | $0.0109812876$ | $0.0110160927$ | $3.481 \times 10^{-5}$ ($0.32\%$) | **$2.184\times$** | $4.67 \times 10^{-20}$ | $[0.0109827083,\, 0.0110175737]$ |
| $(80,\, 25600)$ | $0.0109962569$ | $0.0110019778$ | $5.721 \times 10^{-6}$ | $0.0109977079$ | $0.0109998843$ | $2.176 \times 10^{-6}$ ($0.020\%$) | **$2.629\times$** | $3.50 \times 10^{-19}$ | $[0.0109991565,\, 0.0110013367]$ |
| $(160,\, 25600)$ | $0.0093461064$ | $0.0093772711$ | $3.116 \times 10^{-5}$ | $0.0093493182$ | $0.0093638252$ | $1.451 \times 10^{-5}$ ($0.155\%$) | **$2.148\times$** | $3.11 \times 10^{-19}$ | $[0.0093497341,\, 0.0093642395]$ |

**Table 1:** Head-to-head comparison of outward-rounded exact rational certificates (`fractions.Fraction`, zero transcendental constants) between the Unweighted Extended-Farey Baseline (4) and the Boundary-Deflated Local-Spacing (BD-WMV) method (16) across $N \in \{10, 20, 40, 80, 160\}$ (`src/verify_transcendental_free_infinite_certificates.py`).

---

## 3. All-$n$ Quotient-Hyperbola BCF Decompositions and Sub-Nyquist Drift Diagnostics

Under the orthonormal Hardy–Parseval mother-wavelet basis $\{\delta_1^{(0)}\} \cup \{\delta_{n+1}^{(1)}\}_{n=1}^\infty$ of $\mathcal{S}_{\mathbb{Z}} \subset \mathcal{H}$, every $g = \sum_{k=2}^N c_k\tau_k = \sum_{d=1}^N \xi_d\lfloor x/d \rfloor \in V_N$ ($\sum_{d=1}^N \xi_d/d = 0$) satisfies $\|1 - g\|_{\mathcal{H}}^2 \equiv \bigl(1 + \boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}\bigr)^2 + \sum_{n=1}^\infty \bigl(\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c}\bigr)^2$ with step-jump recurrence $\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c} \equiv -\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c} - \sum_{m=1}^n \frac{\Delta_m g}{m}$, where $\Delta_m g := g(m) - g(m - 1) = \sum_{d \mid m,\, d \le N}\xi_d$ [13, Theorem 4(a), Eq. (41)]. Theorem 2 establishes the exact all-$n$ quotient-hyperbola decomposition of the Bettin–Conrey–Farmer taper [5] and analyzes the prime- vs. composite-jump drift on $N < n \le \lfloor N^{3/2} \rfloor$.

**Theorem 2 (All-$n$ Chebyshev–Mertens Quotient-Hyperbola Decompositions and Prime- vs. Composite-Jump Drift).**
*(a) (**Exact All-$n$ Quotient-Hyperbola Representations**): Let $N \ge 2$, $A_0 \in \mathbb{R}$, and define $\xi_d^{\mathrm{BCF}} := A_0\,w_N(d)$ ($2 \le d \le N$), where $w_N(d) := \mu(d)\bigl(1 - \frac{\log d}{\log N}\bigr)$, with $\xi_1^{\mathrm{BCF}} := -\sum_{d=2}^N \frac{\xi_d^{\mathrm{BCF}}}{d} = A_0\bigl(1 - S_1(N)\bigr)$ and $S_1(N) := \sum_{d=1}^N \frac{w_N(d)}{d} = \frac{L_1(N)}{\log N}$ ($L_1(y) := \sum_{d \le y}\frac{\mu(d)}{d}\log\frac{y}{d}$). For $y > N$, define the Mertens remainders ($M_N(y) = \mathcal{R}_N(y) := 0$ for $y \le N$)*
$$M_N(y) \;:=\; \sum_{N < d \le y} w_N(d), \qquad \mathcal{R}_N(y) \;:=\; \sum_{N < d \le y} \frac{w_N(d)}{d} \;\equiv\; \frac{L_1(y) - L_1(N) - \log(y/N)\,M_{-1}(y)}{\log N} \qquad \left(M_{-1}(y) := \sum_{d \le y}\frac{\mu(d)}{d}\right). \tag{17}$$
*Then for **every integer $n \ge 1$**, $g^{\mathrm{BCF}}(n)$ and $\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c}^{\mathrm{BCF}}$ satisfy the exact quotient-hyperbola identities*
$$g^{\mathrm{BCF}}(n) \;\equiv\; A_0\left(1 \;+\; \frac{\psi(n)}{\log N} \;-\; n\,S_1(N) \;-\; \sum_{q=1}^{\lfloor n/(N+1) \rfloor} M_N\!\left(\frac{n}{q}\right)\right), \tag{18}$$
$$\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c}^{\mathrm{BCF}} \;\equiv\; -\,\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}^{\mathrm{BCF}} \;-\; A_0\left(1 \;+\; \frac{1}{\log N}\sum_{m=2}^n \frac{\Lambda(m)}{m} \;-\; S_1(N)\,H_n \;-\; \sum_{q=1}^{\lfloor n/(N+1) \rfloor} \frac{1}{q}\,\mathcal{R}_N\!\left(\frac{n}{q}\right)\right). \tag{19}$$

*(b) (**First-Octave Pure-Chebyshev Localization and Exact Prime/Composite Drift Decomposition**): On the first octave $1 \le n \le N$, $\lfloor n/(N+1) \rfloor = 0$ so the quotient sums in (18)–(19) vanish identically, giving $g^{\mathrm{BCF}}(n) - A_0 \equiv A_0\frac{\psi(n) - n L_1(N)}{\log N}$ and the unconditional centered Chebyshev bound $\bigl|g^{\mathrm{BCF}}(n) - A_0\bigr| \le (4\log 2 + 1)|A_0|\frac{n}{\log N} < 4|A_0|\frac{n}{\log N}$ for all $1 \le n \le N$. Beyond $n = N$, specializing $\Delta_m g = \sum_{d \mid m,\, d \le N}\xi_d$ [13, Eq. (41)] to any prime $p > N$ gives*
$$\Delta_p g \;\equiv\; \xi_1 \;=\; -\sum_{d=2}^N \frac{\xi_d}{d} \qquad (\forall\, g \in V_N,\; \forall\text{ prime } p > N). \tag{20}$$
*Consequently, on any sub-Nyquist window $N < n \le M$, the wavelet increment $\boldsymbol{\Upsilon}(n)^{\mathsf{T}}\mathbf{c} - \boldsymbol{\Upsilon}(N)^{\mathsf{T}}\mathbf{c} \equiv D_{\mathrm{prime}}(n) + D_{\mathrm{comp}}(n)$ splits into a prime-jump drift $D_{\mathrm{prime}}(n) := -\xi_1\sum_{N < p \le n}\frac{1}{p}$ and a composite-jump drift $D_{\mathrm{comp}}(n) := -\sum_{N < m \le n,\, m\text{ comp.}}\frac{\Delta_m g}{m}$. For the normalized BCF profile ($A_0 = 1$, where $\xi_1^{\mathrm{BCF}} = 1 - \frac{L_1(N)}{\log N} \to 1$), Mertens' second theorem gives $D_{\mathrm{prime}}(n) = -\xi_1^{\mathrm{BCF}}\bigl(\log\frac{\log n}{\log N} + O(\frac{1}{\log N})\bigr)$, so at $M = \lfloor N^{3/2} \rfloor$ ($\frac{\log M}{\log N} \to \frac{3}{2}$) the prime-drift energy has exact asymptotic order*
$$\sum_{N < n \le \lfloor N^{3/2} \rfloor} \bigl|D_{\mathrm{prime}}(n)\bigr|^2 \;=\; \bigl(\log(3/2)\bigr)^2\,N^{3/2}\,\bigl(1 + O(1/\log N)\bigr) \;=\; \Theta\bigl(N^{3/2}\bigr) \qquad \left(\|D_{\mathrm{prime}}\|_{\ell^2((N, N^{3/2}])} = \Theta\bigl(N^{3/4}\bigr)\right), \tag{21}$$
*while the net drift $D_{\mathrm{prime}}(n) + D_{\mathrm{comp}}(n)$ satisfies the exact identity*
$$D_{\mathrm{prime}}(n) + D_{\mathrm{comp}}(n) \;\equiv\; -A_0\left(\frac{1}{\log N}\sum_{m=N+1}^n \frac{\Lambda(m)}{m} \;-\; S_1(N)\bigl(H_n - H_N\bigr) \;-\; \sum_{q=1}^{\lfloor n/(N+1) \rfloor} \frac{1}{q}\,\mathcal{R}_N\!\left(\frac{n}{q}\right)\right). \tag{22}$$

*Proof.*
**(a)** Since $\sum_{d \mid m}\mu(d) = \delta_{m, 1}$ and $\sum_{d \mid m}\mu(d)\log d = -\Lambda(m)$, we have $\sum_{d \mid m,\, d \le N} w_N(d) \equiv \delta_{m, 1} + \frac{\Lambda(m)}{\log N} - \sum_{d \mid m,\, d > N} w_N(d)$ for all $m \ge 1$. Using $\xi_1^{\mathrm{BCF}} = A_0(1 - S_1(N))$ gives $\Delta_m g^{\mathrm{BCF}} = A_0\bigl(\delta_{m, 1} + \frac{\Lambda(m)}{\log N} - S_1(N) - \sum_{d \mid m,\, d > N} w_N(d)\bigr)$. Summing $\Delta_m g^{\mathrm{BCF}}$ (and $\frac{\Delta_m g^{\mathrm{BCF}}}{m}$) over $1 \le m \le n$ and setting $m = q d$ ($1 \le q \le \lfloor n/(N+1) \rfloor$) in the remainder double sum yields (18)–(19), while writing $\frac{w_N(d)}{d} = \frac{\mu(d)}{d\log N}(\log y - \log d - \log(y/N))$ for $N < d \le y$ gives (17).
**(b)** On $1 \le n \le N$, $n/(N+1) < 1$ so the $q$-sums in (18)–(19) are empty; $0 \le \psi(n) \le (2\log 2)n$ and $|L_1(N)| \le 2\log 2 + 1$ [13, Corollary 1] give $|g^{\mathrm{BCF}}(n) - A_0| \le (4\log 2 + 1)|A_0|\frac{n}{\log N}$. For any prime $p > N$, $\operatorname{div}(p) \cap \{1, \dots, N\} = \{1\}$, giving (20). On $\frac{1}{2}N^{3/2} \le n \le N^{3/2}$, $\log\frac{\log n}{\log N} = \log\frac{3}{2} + O(\frac{1}{\log N})$, so summing $|D_{\mathrm{prime}}(n)|^2$ over $N < n \le \lfloor N^{3/2} \rfloor$ yields (21). Differencing (19) between $n$ and $N$ (where $\lfloor N/(N+1) \rfloor = 0$) gives the exact net-drift formula (22). $\blacksquare$

Table 2 reports the sampled nearest-rounded IEEE-754 `float64` diagnostics of (21)–(22) from `src/explore_harmonic_discrepancy_and_bcf_hyperbola.py` (`PASS` in `1.31s` $\to$ `results/harmonic_discrepancy_and_bcf_hyperbola.json`) at $M = \lfloor N^{3/2} \rfloor$ across the grid $N \in \{40, 80, 160, 320, 640, 1280, 2560, 5000, 10000\}$, together with floating-point numerical estimates of the analytical unweighted-tail variational majorant [13, Eq. (48′)] using $\xi_d^{(1)} = \mu(d)(1 - \frac{\log d}{\log N})$, $\xi_d^{(2)} = \frac{\mu(d)}{\log N}$, and $\xi_d^{(3)} = \mu(d)\frac{\log d}{\log N}(1 - \frac{\log d}{\log N})$. Unlike the outward-rounded exact rational certificates in Tables 1 and 3, the entries in Table 2 are floating-point diagnostics: consistent with (21), $\|D_{\mathrm{prime}}\|_2^2 / N^{3/2}$ approaches $(\xi_1^{\mathrm{BCF}}\log(3/2))^2 \approx 0.1308$ at $N = 10{,}000$ ($89{,}996.27 / 10^6 \approx 0.0900$), while at the sampled grid points in Table 2 the net drift $\|D_{\mathrm{prime}} + D_{\mathrm{comp}}\|_2^2$ in (22) is smaller than $\|D_{\mathrm{prime}}\|_2^2$ by factors ranging from $6.72 \times 10^{-3}$ at $N = 40$ to $2.70 \times 10^{-7}$ at $N = 10{,}000$.

| $N$ | $M = \lfloor N^{3/2} \rfloor$ | Origin Jump $\xi_1^{\mathrm{BCF}}$ | Prime-Drift Energy $\|D_{\mathrm{prime}}\|_2^2$ | Net-Drift Energy $\|D_{\mathrm{prime}} + D_{\mathrm{comp}}\|_2^2$ | Net / Prime Drift Ratio | 1D BCF Majorant Est. $\times \log N$ | 2D BCF+Möbius Est. $\times \log N$ | 3D Smooth $\rho_d$ Est. $\times \log N$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $40$ | $253$ | $0.728821$ | $9.2387$ | $0.062111$ | $6.72 \times 10^{-3}$ | $0.095650$ | $0.095527$ | $0.086223$ |
| $80$ | $716$ | $0.771832$ | $29.8928$ | $0.005274$ | $1.76 \times 10^{-4}$ | $0.083318$ | $0.077157$ | $0.076149$ |
| $160$ | $2024$ | $0.802904$ | $107.8105$ | $0.012536$ | $1.16 \times 10^{-4}$ | $0.076826$ | $0.076722$ | $0.073908$ |
| $320$ | $5724$ | $0.826756$ | $344.9085$ | $0.025015$ | $7.25 \times 10^{-5}$ | $0.070965$ | $0.068530$ | $0.067924$ |
| $640$ | $16191$ | $0.845130$ | $1119.3137$ | $0.025922$ | $2.32 \times 10^{-5}$ | $0.068363$ | $0.067620$ | $0.066960$ |
| $1280$ | $45795$ | $0.860275$ | $3443.4712$ | $0.024859$ | $7.22 \times 10^{-6}$ | $0.064684$ | $0.064066$ | $0.063475$ |
| $2560$ | $129527$ | $0.872566$ | $10451.3204$ | $0.002475$ | $2.37 \times 10^{-7}$ | $0.063471$ | $0.063098$ | $0.062793$ |
| $5000$ | $353553$ | $0.882594$ | $30261.8965$ | $0.000523$ | $1.73 \times 10^{-8}$ | $0.061168$ | $0.060634$ | $0.060294$ |
| $10000$ | $1000000$ | $0.891423$ | $89996.2651$ | $0.024265$ | $2.70 \times 10^{-7}$ | $0.060064$ | $0.059549$ | $0.059320$ |

**Table 2:** Nearest-rounded IEEE-754 `float64` diagnostics of prime-jump drift $\|D_{\mathrm{prime}}\|_2^2 = \Theta(N^{3/2})$ vs. net drift $\|D_{\mathrm{prime}} + D_{\mathrm{comp}}\|_2^2$ on $N < n \le \lfloor N^{3/2} \rfloor$ and numerical estimates of the unweighted-tail 1D/2D/3D variational majorant [13, Eq. (48′)] up to $N = 10{,}000$ (`src/explore_harmonic_discrepancy_and_bcf_hyperbola.py`).

---

## 4. Exact Subspace Equivalence of Squarefree $\rho_d$ and Squarefree $\tau_k$, and Certified Rational Separation $d_N^2 < d_N^{\mathrm{sqf}, 2}$

In [3], Báez-Duarte proved that the **squarefree dilate subspace** $V_N^{\mathrm{sqf}} := \operatorname{span}\{\rho_d : 2 \le d \le N,\, \mu(d) \neq 0\}$ is dense in $\mathcal{H}$ if and only if $\bigcup_{N \ge 2} V_N$ is dense in $\mathcal{H}$, and subsequent Dirichlet-polynomial constructions such as Bettin–Conrey–Farmer [5] restrict attention to $V_N^{\mathrm{sqf}}$. Using Theorem 1, we now show that $V_N^{\mathrm{sqf}}$ coincides identically with the squarefree $\tau_k$-subspace $\operatorname{span}\{\tau_k : 2 \le k \le N,\, \mu(k) \neq 0\}$, and we construct **exact rational BD-WMV primal-dual certificates (`fractions.Fraction`, zero transcendental constants)** proving that the full NBBD distance $d_N^2$ lies strictly below the squarefree distance $d_N^{\mathrm{sqf}, 2} := \inf_{g \in V_N^{\mathrm{sqf}}}\|1 - g\|_{\mathcal{H}}^2$ across every tested scale $N \in \{4, 10, 20, 40, 80, 160\}$ (`src/study_squarefree_vs_full_radical_fibers.py`).

**Theorem 3 (Squarefree Subspace Equivalence and Exact Rational Certification of $d_N^2 < d_N^{\mathrm{sqf}, 2}$ and Intermediate Radical-Harmonic Subspaces).**
*(a) (**Coordinate Bijection [9, Theorem 3] and Squarefree Subspace Equivalence**): For every $g \in V_N$, the $\tau_k$-expansion, the $\rho_d$-expansion, and the $\lfloor x/d \rfloor$-expansion are related across all $x > 0$ by*
$$g(x) \;\equiv\; \sum_{k=2}^N c_k\,\tau_k(x) \;\equiv\; \sum_{d=2}^N w_d\,\rho_d(x) \;\equiv\; \sum_{d=1}^N \xi_d\left\lfloor\frac{x}{d}\right\rfloor \qquad \left(\sum_{d=1}^N \frac{\xi_d}{d} = 0\right), \tag{23}$$
*where, setting $c_1 := 0$, the coordinates $(c_k)_{k=2}^N$, $(w_d)_{d=2}^N$, and $(\xi_d)_{d=1}^N$ satisfy the exact upper-triangular bijection*
$$w_d \;\equiv\; \xi_d \;=\; d\sum_{\ell=1}^{\lfloor N/d \rfloor} \mu(\ell)\,c_{d\ell} \quad (2 \le d \le N), \qquad \xi_1 \;=\; -\sum_{d=2}^N \frac{w_d}{d} \;=\; \sum_{\ell=2}^N \mu(\ell)\,c_\ell, \qquad c_k \;=\; \frac{1}{k}\sum_{m=1}^{\lfloor N/k \rfloor} \frac{w_{k m}}{m} \quad (2 \le k \le N). \tag{24}$$
*Since every multiple $d\ell$ of a non-squarefree integer $d$ ($\mu(d) = 0$) is also non-squarefree ($\mu(d\ell) = 0$), this bijection satisfies $c_k = 0$ ($\forall\, \mu(k) = 0$) $\iff w_d = 0$ ($\forall\, \mu(d) = 0$). Consequently, Báez-Duarte's squarefree subspace satisfies the exact subspace identity*
$$V_N^{\mathrm{sqf}} \;:=\; \operatorname{span}\bigl\{\rho_d : 2 \le d \le N,\; \mu(d) \neq 0\bigr\} \;\equiv\; \operatorname{span}\bigl\{\tau_k : 2 \le k \le N,\; \mu(k) \neq 0\bigr\} \;\subset\; V_N. \tag{25}$$

*(b) (**Exact Rational BD-WMV Primal-Dual Certification on Coordinate Subspaces $V_N(S)$**): For any non-empty index subset $S \subseteq \{2, \dots, N\}$ with maximum denominator $N_S := \max(S) \le N$, let $V_N(S) := \operatorname{span}\{\tau_k : k \in S\} = \{\mathbf{c} \in \mathbb{R}^{N-1} : c_k = 0 \text{ for } k \notin S\}$ and $d_N(S)^2 := \inf_{g \in V_N(S)}\|1 - g\|_{\mathcal{H}}^2$. Restricting Theorem 1 to $V_N(S)$ (where the active Farey set $\mathcal{F}(S) := \{0\} \cup \{b/k : k \in S,\, \gcd(b, k) = 1\}$ has local spacing $\delta_0^{-1} \le N_S$ and $\delta_{b/k}^{-1} \le k(N_S - \mathbf{1}_{k=N_S})$) yields strictly convex rational quadratic forms $Q_{\mathrm{LB}}^{\mathrm{BD}}(\mathbf{c}; M, S) \le \|1 - g\|_{\mathcal{H}}^2 \le Q_{\mathrm{UB}}^{\mathrm{BD}}(\mathbf{c}; M, S)$ on $V_N(S)$ with coercivity constant $C_\beta(S, M) = 2M + 3 - \frac{3}{4}N_S - \frac{1}{N_S}\sum_{k \in S}\frac{J_2(k)}{12} \ge \frac{75}{8} > 9$ and $\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}}(S) > 0$ for all $M \ge N^2$, certifying*
$$\mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}\bigl(d_N(S)^2\bigr) \;\le\; d_N(S)^2 \;\le\; \mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}\bigl(d_N(S)^2\bigr). \tag{26}$$

*(c) (**Certified Strict Separation $d_N^2 < d_N^{\mathrm{sqf}, 2}$ and Radical-Harmonic Subspace Ordering**): Taking $S = S_{\mathrm{sqf}} := \{2 \le k \le N : \mu(k) \neq 0\}$ in (26) certifies $\mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^2) < \mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^{\mathrm{sqf}, 2})$ across all $(N, M) \in \{(4, 16), (10, 100), (20, 400), (40, 1600), (80, 6400), (80, 25600), (160, 25600)\}$ (Table 3). Furthermore, for the two intermediate subspaces between $V_N^{\mathrm{sqf}}$ and $V_N$ defined by adjoining non-squarefree prime powers $p^a$ ($a \ge 2$) or first radical harmonics $k = 2\operatorname{rad}(k)$,*
$$V_N^{\mathrm{sqf}+\mathrm{pp}} \;:=\; \operatorname{span}\bigl\{\tau_k : 2 \le k \le N,\; \mu(k) \neq 0 \text{ or } \Lambda(k) > 0\bigr\}, \qquad V_N^{(2)} \;:=\; \operatorname{span}\bigl\{\tau_k : 2 \le k \le N,\; k/\operatorname{rad}(k) \in \{1, 2\}\bigr\}, \tag{27}$$
*exact rational evaluation of (26) at $M = 25{,}600$ for $N \in \{80, 160\}$ certifies the strict chain of inequalities*
$$d_N^2 \;<\; d_N\bigl(V_N^{(2)}\bigr)^2 \;<\; d_N\bigl(V_N^{\mathrm{sqf}+\mathrm{pp}}\bigr)^2 \;<\; d_N^{\mathrm{sqf}, 2}. \tag{28}$$

*Proof.*
**(a)** Substituting $\rho_d(x) = \lfloor x/d \rfloor - \frac{1}{d}\lfloor x \rfloor$ into $\sum_{d=2}^N w_d\rho_d(x) = \sum_{d=2}^N w_d\lfloor x/d \rfloor - \bigl(\sum_{d=2}^N \frac{w_d}{d}\bigr)\lfloor x \rfloor$ identifies $\xi_d \equiv w_d$ for $2 \le d \le N$ and $\xi_1 \equiv -\sum_{d=2}^N \frac{w_d}{d}$. On the other hand, substituting $\tau_k(x) = \sum_{d \mid k} d\mu(k/d)\lfloor x/d \rfloor$ into $\sum_{k=2}^N c_k\tau_k(x)$ and setting $k = d\ell$ ($c_1 := 0$) gives $\xi_d = d\sum_{\ell=1}^{\lfloor N/d \rfloor}\mu(\ell)c_{d\ell}$ for all $1 \le d \le N$ [9, Theorem 3], which at $d = 1$ gives $\xi_1 = \sum_{\ell=2}^N \mu(\ell)c_\ell$. Since $\frac{w_d}{d} = \sum_{\ell=1}^{\lfloor N/d \rfloor}\mu(\ell)c_{d\ell}$ for $2 \le d \le N$, Möbius inversion gives $c_k = \frac{1}{k}\sum_{m=1}^{\lfloor N/k \rfloor}\frac{w_{k m}}{m}$ ($2 \le k \le N$), proving (23)–(24) (for instance, at $N = 2, c_2 = 1$, $\tau_2(x) = 2\lfloor x/2 \rfloor - \lfloor x \rfloor = 2\rho_2(x)$, and (24) gives $w_2 = \xi_2 = 2\mu(1)c_2 = 2$, $\xi_1 = \mu(2)c_2 = -1$, and $g(1) = -1$). If $\mu(d) = 0$, then $d$ is divisible by some prime square $p^2$, so every multiple $d\ell$ is also divisible by $p^2$ ($\mu(d\ell) = 0$); hence $c_k = 0$ ($\forall\, \mu(k) = 0$) $\implies w_d = 0$ ($\forall\, \mu(d) = 0$), and the converse follows identically from $c_k = \frac{1}{k}\sum_{m=1}^{\lfloor N/k \rfloor}\frac{w_{k m}}{m}$, proving (25).
**(b)–(c)** When $c_k = 0$ for all $k \notin S$, the non-zero Farey coefficients $a_{b/k}$ in (3) are supported on $\mathcal{F}(S)$ with denominators $k \in S \subseteq \{2, \dots, N_S\}$, so Theorem 1(a)–(d) applies verbatim on $\mathbb{R}^{|S|}$ with $N$ replaced by $N_S$ and $\sum_{k \in S}\frac{J_2(k)}{12} \le \sum_{k=2}^{N_S}\frac{J_2(k)}{12}$, giving (26). Exact `fractions.Fraction` evaluation in `src/study_squarefree_vs_full_radical_fibers.py` (`PASS` in `14.8s` $\to$ `results/squarefree_vs_full_radical_fibers.json`) verifies the outward-rounded intervals in Table 3 and the strict inequalities (28):
- At $(N, M) = (80, 25600)$: $d_{80}^2 \in [0.0109977079, 0.0109998843]$, $d_{80}(V_{80}^{(2)})^2 \in [0.0111100868, 0.0111119195]$, $d_{80}(V_{80}^{\mathrm{sqf}+\mathrm{pp}})^2 \in [0.0111279058, 0.0111296808]$, and $d_{80}^{\mathrm{sqf}, 2} \in [0.0114405463, 0.0114423912]$.
- At $(N, M) = (160, 25600)$: $d_{160}^2 \in [0.0093493182, 0.0093638252]$, $d_{160}(V_{160}^{(2)})^2 \in [0.0093865799, 0.0094001170]$, $d_{160}(V_{160}^{\mathrm{sqf}+\mathrm{pp}})^2 \in [0.0094500991, 0.0094629269]$, and $d_{160}^{\mathrm{sqf}, 2} \in [0.0096648736, 0.0096780729]$. $\blacksquare$

| $(N,\, M)$ | Full Space Dim / Sqf Dim ($N_{\mathrm{sqf}}$) | Full Space BD-WMV $[\mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^2),\, \mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^2)]$ | Squarefree BD-WMV $[\mathrm{LB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^{\mathrm{sqf}, 2}),\, \mathrm{UB}_{\mathbb{Q}}^{\mathrm{BD}}(d_N^{\mathrm{sqf}, 2})]$ | Sqf Duality Defect | Certified Gap $d_N^{\mathrm{sqf}, 2} - d_N^2$ (Outward-Rounded) | Certified $(d_N^{\mathrm{sqf}, 2} - d_N^2)(\log N)^2$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $(4,\, 16)$ | $3$ / $2$ ($3$) | $[0.0605390738,\, 0.0714674674]$ | $[0.0919326583,\, 0.0994483224]$ | $5.90 \times 10^{-23}$ | **$[0.0204651909,\, 0.0389092486]$** | $[0.03933,\, 0.07478]$ |
| $(10,\, 100)$ | $9$ / $6$ ($10$) | $[0.0233744486,\, 0.0243189313]$ | $[0.0254497258,\, 0.0265106278]$ | $3.18 \times 10^{-22}$ | **$[0.0011307945,\, 0.0031361792]$** | $[0.00600,\, 0.01663]$ |
| $(20,\, 400)$ | $19$ / $12$ ($19$) | $[0.0163974054,\, 0.0166775327]$ | $[0.0175136851,\, 0.0177509625]$ | $1.71 \times 10^{-21}$ | **$[0.0008361524,\, 0.0013535571]$** | $[0.00750,\, 0.01215]$ |
| $(40,\, 1600)$ | $39$ / $25$ ($39$) | $[0.0126824994,\, 0.0127889200]$ | $[0.0134923892,\, 0.0135790375]$ | $6.01 \times 10^{-21}$ | **$[0.0007034692,\, 0.0008965381]$** | $[0.00957,\, 0.01220]$ |
| $(80,\, 6400)$ | $79$ / $49$ ($79$) | $[0.0109812876,\, 0.0110160927]$ | $[0.0114266489,\, 0.0114561547]$ | $3.16 \times 10^{-19}$ | **$[0.0004105562,\, 0.0004748671]$** | $[0.00788,\, 0.00912]$ |
| $(80,\, 25600)$ | $79$ / $49$ ($79$) | $[0.0109977079,\, 0.0109998843]$ | $[0.0114405463,\, 0.0114423912]$ | $3.62 \times 10^{-19}$ | **$[0.0004406620,\, 0.0004446833]$** | $[0.00846,\, 0.00854]$ |
| $(160,\, 25600)$ | $159$ / $97$ ($159$) | $[0.0093493182,\, 0.0093638252]$ | $[0.0096648736,\, 0.0096780729]$ | $2.15 \times 10^{-19}$ | **$[0.0003010484,\, 0.0003287547]$** | $[0.00775,\, 0.00847]$ |

**Table 3:** Outward-rounded exact rational certificates (`fractions.Fraction`, zero transcendental constants) comparing the full NBBD distance $d_N^2$ against Báez-Duarte's squarefree subspace distance $d_N^{\mathrm{sqf}, 2}$ and certifying the non-squarefree reduction $d_N^{\mathrm{sqf}, 2} - d_N^2 > 0$ across $N \in \{4, 10, 20, 40, 80, 160\}$ (`src/study_squarefree_vs_full_radical_fibers.py`).

---

## 5. Reproducibility and Verification Supplement

The supplementary archive `paper43_supplement.zip` is self-contained and executes in under $41$ seconds total runtime using Python $\ge 3.10$ and `numpy`. Each script supports `--output-dir <DIR>` and reproduces the pre-computed JSON files in `results/`:
1. `src/verify_transcendental_free_infinite_certificates.py` $\to$ `results/transcendental_free_infinite_certificates.json`: verifies Theorem 1's direct agreement between the Fourier series (3) and $1 - \sum_{k=2}^N c_k\tau_k(n)$ for all $0 \le n \le M$ (including $r_0 \equiv 1$), Hilbert primitive identities ($U_P(-1) \equiv 0$, $U(-1)$, $U(M)$), exact analytical benchmark at $N = 2, M = 4$ ($c \in \{-1, 0, 1, 2\}$ against $1 + (2c+c^2)\log 2$) and 6-periodic norm check at $N = 3, M = 9$, unconditional strict convexity ($C_\beta(N, M) \ge 9.375 > 9$ and $\gamma_{k, M}^{\mathrm{BD}, \mathrm{LB}} > 0$ across $2 \le N \le 160$), and the head-to-head Baseline (4) vs. BD-WMV (16) certificates in Table 1 (`PASS` in `24.3s`).
2. `src/explore_harmonic_discrepancy_and_bcf_hyperbola.py` $\to$ `results/harmonic_discrepancy_and_bcf_hyperbola.json`: verifies Theorem 2's all-$n$ quotient-hyperbola BCF identities (17)–(19), centered first-octave Chebyshev bound, and `float64` prime- vs. composite-jump drift diagnostics in Table 2 up to $N = 10{,}000$ (`PASS` in `1.31s`).
3. `src/study_squarefree_vs_full_radical_fibers.py` $\to$ `results/squarefree_vs_full_radical_fibers.json`: verifies Theorem 3's coordinate bijection (23)–(24) (including the $N = 2, c_2 = 1$ benchmark $w_2 = 2, \xi_1 = -1, g(1) = -1$ and pointwise equality across $1 \le n \le 2N$), squarefree subspace equivalence (25), and the exact `fractions.Fraction` BD-WMV certificates for $d_N^{\mathrm{sqf}, 2}$, $d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2$, $d_N(V_N^{(2)})^2$, and $d_N^2$ in Table 3 and (28) (`PASS` in `14.8s`).

---

## 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: All-Scale Universal Lower Bounds, Geometric Transference, and Two-Sided Interval-Certified Subspace Optimality*, 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. `@bramble_trace`, *A logarithmic-prefix Nyman–Beurling criterion and transferred zero bounds*, rStation `riemann_hypothesis_47` Version 1 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/47/v1.
15. `@bramble_trace`, *Adaptive zero clusters and optimal Nyman–Beurling approximation rates*, rStation `riemann_hypothesis_48` Version 2 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/48/v2.
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. G. H. Hardy, J. E. Littlewood, and G. Pólya, *Inequalities*, Cambridge University Press (1934).
18. J. Singh, *Sums of products involving power sums of $\varphi(n)$ integers*, arXiv:1306.5848 (2013); https://arxiv.org/abs/1306.5848.
19. NIST Digital Library of Mathematical Functions, §24 (Bernoulli and Euler Polynomials) and §27 (Functions of Number Theory); https://dlmf.nist.gov.
20. L. Báez-Duarte, M. Balazard, B. Landreau, and E. Saias, *Étude de l'autocorrélation multiplicative de la fonction « partie fractionnaire »*, Ramanujan J. **9** (2005), 215–240; extended calculations: *Sur l'autocorrélation multiplicative de la fonction « partie fractionnaire »*, arXiv:math/0306251, §9.5, Proposition 98; https://arxiv.org/abs/math/0306251.
