# Commutative Dilation-Operator Euler Products, Totient–Cotangent Gram Decompositions, Parabolic Shift Separation, and Prime-Power Coprime-Additivity Obstructions in Nyman–Beurling Space

## Abstract

We establish four exact analytical theorems governing the infinite-horizon Nyman–Beurling–Báez-Duarte (NBBD) Gram matrix $G_N = (\langle \tau_j, \tau_k \rangle_{\mathcal{H}})_{2 \le j, k \le N}$ in $\mathcal{H} = L^2((1, \infty), x^{-2}\,dx)$. First, using the commutative dilation semigroup $(T_q f)(x) = q f(x/q)\mathbf{1}_{[q, \infty)}(x)$ ($T_q^* 1 = 1$, $T_p^* T_p = p I$) and pointwise dilations $(\widetilde{T}_q f)(x) = q f(x/q)$, we prove the pointwise operator Euler product $\tau_k = (\prod_{p^\alpha \parallel k} \widetilde{T}_p^{\alpha-1}(\widetilde{T}_p - I))\tau_1$, giving a four-line operator proof of $\langle 1, \tau_k \rangle_{\mathcal{H}} = -\Lambda(k)$ and the universal bilinear prime-lift Gram recurrence $\langle \tau_{p m} + \mathbf{1}_{p \nmid m}\tau_m, \tau_{p n} + \mathbf{1}_{p \nmid n}\tau_n \rangle_{\mathcal{H}} = p\langle \tau_m, \tau_n \rangle_{\mathcal{H}}$. Second, using reflection antisymmetry of $P_k(n) = -\tau_k(n) - \frac{1}{2}\varphi(k)$ and Euler's cotangent partial fractions, we prove $\langle \tau_j, \tau_k \rangle_{\mathcal{H}} = \frac{1}{2}(\varphi(j)\Lambda(k) + \varphi(k)\Lambda(j)) + \pi\,\widetilde{\mathcal{C}}_{j, k}$ ($\widetilde{\mathcal{C}}_{j, k} \in \mathbb{Q}(\cos(\pi/(2\operatorname{lcm}(j, k))))$) and a diagonal sign trichotomy. Third, on the centered hyperplane $\beta(\mathbf{c}) = 1 + \frac{1}{2}\boldsymbol{\varphi}_N^{\mathsf{T}}\mathbf{c} = 0$, all prime logarithms vanish ($d_N^{\mathrm{cen}, 2} = 1 - \pi\alpha_N$), and we prove the exact all-$N \ge 2$ parabolic shift identity $d_N^{\mathrm{cen}, 2} - d_N^2 \equiv B_N^2 / A_N$, certifying $B_N \neq 0$ and $d_N^2 < d_N^{\mathrm{cen}, 2}$ across all $2 \le N \le 50$ (and $d_N^2 < d_N^{\mathrm{sqf}, 2}$ across $4 \le N \le 50$) in 224-bit dyadic rational arithmetic. Fourth, we prove an all-$N$ prime-power coprime-additivity jump obstruction certifying in exact rational arithmetic that $\inf_{N \ge 2} d_N(V_N^{\mathrm{pp}})^2 \ge \underline{D}_{100}^{\mathrm{pp}} > 0.02493 > d_{10}^2$. These results do not resolve RH.

---

## 1. Introduction and Summary of Main Results

### 1.1 Background and Motivation

In the Nyman–Beurling–Báez-Duarte (NBBD) Hilbert-space formulation of the Riemann Hypothesis [1, 2, 3], one studies the distance $d_N := \operatorname{dist}_{\mathcal{H}}(1, V_N)$ in $\mathcal{H} := L^2((1, \infty), x^{-2}\,dx)$ from the constant target function $1$ to the $(N - 1)$-dimensional step-function subspace
$$V_N \;:=\; \operatorname{span}\{\rho_d : 2 \le d \le N\} \;=\; \operatorname{span}\{\tau_k : 2 \le k \le N\}, \qquad \rho_d(x) \;:=\; \left\lfloor\frac{x}{d}\right\rfloor - \frac{\lfloor x \rfloor}{d}, \tag{1}$$
where the primitive Möbius-deflated Báez-Duarte–Burnol generators [3, 6] are defined on $[0, \infty)$ by
$$\tau_k(x) \;:=\; \sum_{d \mid k} d\,\mu\!\left(\frac{k}{d}\right)\left\lfloor\frac{x}{d}\right\rfloor \qquad (k \ge 1), \tag{2}$$
so that $\tau_1(x) = \lfloor x \rfloor$ and $\tau_k(x) = \sum_{d \mid k,\, d \ge 2} d\mu(k/d)\rho_d(x)$ for all $k \ge 2$ (since $\sum_{d \mid k} \mu(k/d) = 0$ for $k \ge 2$). By Báez-Duarte's theorem [3], the Riemann Hypothesis holds if and only if $1 \in \overline{\bigcup_{N \ge 2} V_N}$ in $\mathcal{H}$ ($\iff \lim_{N \to \infty} d_N^2 = 0$), and in fact $1 \in \overline{\bigcup_{N \ge 2} V_N^{\mathrm{sqf}}}$ ($\iff \lim_{N \to \infty} d_N^{\mathrm{sqf}, 2} = 0$), where
$$V_N^{\mathrm{sqf}} \;:=\; \operatorname{span}\{\rho_d : 2 \le d \le N,\; \mu(d) \neq 0\} \;\equiv\; \operatorname{span}\{\tau_k : 2 \le k \le N,\; \mu(k) \neq 0\} \tag{3}$$
is the squarefree subspace [3, 14, Theorem 3(a)]. Burnol [6] established the unconditional lower bound $\liminf_{N \to \infty} d_N^2 \log N \ge \sum_{\Re\rho = 1/2} \frac{m_\rho^2}{|\rho|^2} \ge 2 + \gamma - \log(4\pi) \approx 0.046191$ [7], while Bettin–Conrey–Farmer [5] proved a conditional matching upper-bound asymptotic under RH and $\sum_{|\gamma| < T} |\zeta'(1/2 + i\gamma)|^{-2} \ll T$.

In our preceding paper [14] (`riemann_hypothesis_53`), building on [8, 9, 10, 11, 12, 13, 22] and the transferred zero-obstruction bounds of `@bramble_trace` [15, 16], we constructed transcendental-free infinite-horizon rational primal-dual certificates (BD-WMV) by decomposing $\|1 - g\|_{\mathcal{H}}^2 = \sum_{n=1}^{M-1} \frac{(1 - g(n))^2}{n(n+1)} + \sum_{n=M}^{\infty} \frac{(1 - g(n))^2}{n(n+1)}$ into an exact rational prefix sum on $1 \le n < M$ and a boundary-deflated Montgomery–Vaughan [17] local-spacing envelope on the infinite tail $n \ge M$. While this truncated-horizon method avoids all transcendental constants and certified $d_N^2 < d_N^{\mathrm{sqf}, 2}$ at $(N, M) \in \{(4, 16), (10, 100), (20, 400), (40, 1600), (80, 6400), (80, 25600), (160, 25600)\}$ with bracket widths $O(N^3 / M) \sim 10^{-4}\text{–}10^{-2}$, three fundamental analytical questions remained open:

1. **Exact Finite-Period Structure of the Infinite-Horizon Gram Matrix $G_N$:** Because $\tau_k(n)$ is $k$-periodic on $\mathbb{Z}_{\ge 0}$, the infinite series $\langle \tau_j, \tau_k \rangle_{\mathcal{H}} = \sum_{n=1}^{\infty} \frac{\tau_j(n)\tau_k(n)}{n(n+1)}$ has period $L := \operatorname{lcm}(j, k)$ (cf. Vasyunin's Dedekind–cotangent formula and fractional-part autocorrelation splitting in the $\rho_d$ basis [4, 20, 22]). How does $\langle \tau_j, \tau_k \rangle_{\mathcal{H}}$ decompose into logarithmic and trigonometric parts in the primitive $\tau_k$ basis, and why do all prime logarithms $\log p$ cancel in composite norms such as $\|\tau_6\|_{\mathcal{H}}^2 = \frac{\pi}{\sqrt{3}}$?
2. **Parabolic Shift Separation Between Unconstrained and Centered / Squarefree Distances:** In [13, 14], the residual $r_n = 1 - g(n) = \beta(\mathbf{c}) + \sum_{k=2}^N c_k P_k(n)$ splits into the constant mean offset $\beta(\mathbf{c}) := 1 + \frac{1}{2}\sum_{k=2}^N c_k\varphi(k)$ and the zero-mean periodic Bernoulli oscillation $P_k(n) := -\tau_k(n) - \frac{1}{2}\varphi(k)$. Can one express the exact shift gap $d_N^{\mathrm{cen}, 2} - d_N^2$ in closed form for every $N \ge 2$ via a logarithm-free deflated Gram matrix, and certify $B_N \neq 0$, $d_N^2 < d_N^{\mathrm{cen}, 2}$, and $d_N^2 < d_N^{\mathrm{sqf}, 2}$ directly on $(1, \infty)$ across $2 \le N \le 50$ without any $M$-truncation tail error?
3. **Role of Prime-Power vs. Multi-Prime Composite Modes:** Since $\langle 1, \tau_k \rangle_{\mathcal{H}} = -\Lambda(k)$ vanishes identically for every multi-prime composite $k$ ($\omega(k) \ge 2$), the prime-power subspace $V_N^{\mathrm{pp}} := \operatorname{span}\{\tau_{p^\alpha} : 2 \le p^\alpha \le N\}$ contains **every** generator in $V_N$ that correlates with the target $1$. Can $V_\infty^{\mathrm{pp}} := \bigcup_{N \ge 2} V_N^{\mathrm{pp}}$ approximate $1$ on its own, or is there an unconditional arithmetic obstruction?

### 1.2 Summary of Main Results

We resolve all three questions through four self-contained theorems:

- **Theorem 1 (Commutative Dilation-Operator Euler Product & Universal Bilinear Prime-Lift Gram Recurrence, Section 2):** Introducing the commutative semigroup of dilation operators $(T_q f)(x) := q f(x/q)\mathbf{1}_{[q, \infty)}(x)$ on $\mathcal{H}$ ($T_q^* 1 \equiv 1$, $T_{q_1}^* T_{q_2} \equiv \gcd(q_1, q_2) T_{q_1/g}^* T_{q_2/g}$) and pointwise dilations $(\widetilde{T}_q f)(x) := q f(x/q)$ on $[0, \infty)$, we prove that $\tau_{p m}(x) \equiv p\tau_m(x/p) - \mathbf{1}_{p \nmid m}\tau_m(x)$ on $[0, \infty)$ and that $\tau_k$ admits the **pointwise operator Euler product** $\tau_k \equiv \bigl(\prod_{p^\alpha \parallel k} \widetilde{T}_p^{\alpha - 1}(\widetilde{T}_p - I)\bigr)\tau_1$. Because $(\widetilde{T}_q\tau_m)\big|_{(1, \infty)} \equiv T_q(\tau_m\big|_{(1, \infty)}) \in \mathcal{H}$ for $m \ge 2$ and $(T_p^* - I)1 \equiv 0$, this yields a four-line operator proof that $\langle 1, \tau_k \rangle_{\mathcal{H}} \equiv -\Lambda(k)$ ($k \ge 2$) alongside the **universal bilinear prime-lift Gram recurrence**
  $$\bigl\langle \tau_{p m} + \mathbf{1}_{p \nmid m}\tau_m,\;\; \tau_{p n} + \mathbf{1}_{p \nmid n}\tau_n \bigr\rangle_{\mathcal{H}} \;\equiv\; p\,\langle \tau_m,\, \tau_n \rangle_{\mathcal{H}} \qquad (\forall\,\text{prime } p,\;\; \forall\, m, n \ge 2). \tag{4}$$
- **Theorem 2 (Universal Totient–Von Mangoldt–Cotangent Decomposition, Cyclotomic Probability-Measure Identity, and Arithmetic Sign Trichotomy, Section 3):** Using the reflection antisymmetry $P_k(L - 1 - n) \equiv -P_k(n)$ of $P_k(n) = -\tau_k(n) - \frac{1}{2}\varphi(k)$ on $\{0, 1, \dots, L - 1\}$ ($L := \operatorname{lcm}(j, k)$) and Euler's cotangent partial-fraction identity, we prove that $\langle P_j, P_k \rangle_{\mathcal{H}} \equiv \sum_{n=0}^{L-1} w_L(n) P_j(n) P_k(n)$ is a **100% logarithm-free** expectation under an explicit symmetric probability measure $w_L > 0$ on $\{0, \dots, L - 1\}$, and that every Gram entry $G_{j, k} = \langle \tau_j, \tau_k \rangle_{\mathcal{H}}$ decomposes in finite closed form as
  $$\langle \tau_j,\, \tau_k \rangle_{\mathcal{H}} \;\equiv\; \frac{1}{2}\Bigl(\varphi(j)\,\Lambda(k) \;+\; \varphi(k)\,\Lambda(j)\Bigr) \;+\; \pi\,\widetilde{\mathcal{C}}_{j, k}, \qquad \widetilde{\mathcal{C}}_{j, k} \;\in\; \mathbb{Q}\bigl(\cos(\pi/(2 L))\bigr), \tag{5}$$
  with $\widetilde{\mathcal{C}}_{j, k} \in \mathbb{Q}(\cos(\pi/L))$ whenever $L$ is even. Furthermore, summation by parts against the strictly positive kernel $\cot(\frac{\pi n}{k}) - \cot(\frac{\pi(n+1)}{k}) = \frac{\sin(\pi/k)}{\sin(\pi n/k)\sin(\pi(n+1)/k)} > 0$ establishes an **arithmetic sign trichotomy** for the diagonal $\widetilde{\mathcal{C}}_{k, k}$ across all $k \ge 2$: $\widetilde{\mathcal{C}}_{2^\alpha, 2^\alpha} = 0$, $\widetilde{\mathcal{C}}_{p^\alpha, p^\alpha} < 0$ for every odd prime power $p^\alpha$, and $\widetilde{\mathcal{C}}_{k, k} = \frac{1}{\pi}\|\tau_k\|_{\mathcal{H}}^2 > 0$ for every multi-prime composite $k$ ($\omega(k) \ge 2$), proving that $\max_{1 \le n \le k - 2}|P_k(n)| > \frac{1}{2}\varphi(k)$ whenever $\omega(k) \ge 2$.
- **Theorem 3 (Logarithm-Free Algebraic Centered Distance $d_N^{\mathrm{cen}, 2} = 1 - \pi\alpha_N$, Exact All-$N \ge 2$ Parabolic Shift Decomposition, and 224-Bit Dyadic Rational Interval Certificates across $2 \le N \le 50$, Section 4):** By (5), $\|1 - g\|_{\mathcal{H}}^2 \equiv 1 + 2\beta(\mathbf{c})(\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}) + \pi\,\mathbf{c}^{\mathsf{T}}\widetilde{\mathcal{C}}_N\mathbf{c}$. On the centered hyperplane $V_N^{\mathrm{cen}} = \{\beta(\mathbf{c}) = 0\}$, all prime logarithms vanish identically, and since $L_N := \operatorname{lcm}(2, \dots, N)$ is even for all $N \ge 2$, the unique minimizer $\mathbf{c}_N^{\mathrm{cen}, *} \in \mathbb{Q}(\cos(\pi/L_N))^{N-1}$ is algebraic and $d_N^{\mathrm{cen}, 2} \equiv 1 - \pi\alpha_N$ with $\alpha_N \in \mathbb{Q}(\cos(\pi/L_N))$ ($\lim_{N \to \infty} d_N^{\mathrm{cen}, 2} = \lim_{N \to \infty} d_N^2$). Deflating against $\tau_2$ via $Q_k := -\tau_k + \varphi(k)\tau_2$ ($3 \le k \le N$) yields the logarithm-free SPD matrix $\widehat{\mathcal{C}}_N \succ 0$ and the exact all-$N \ge 2$ parabolic shift identity
  $$d_N^{\mathrm{cen}, 2} - d_N^2 \;\equiv\; \frac{B_N^2}{A_N} \;\ge\; 0, \qquad B_N \;:=\; \pi\,\alpha_N \;+\; \boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}_N^{\mathrm{cen}, *}, \qquad \beta(\mathbf{c}_N^*) \;\equiv\; -\frac{B_N}{A_N}. \tag{6}$$
  Thus $d_N^2 < d_N^{\mathrm{cen}, 2} \iff B_N \neq 0$. A single forward-only $LDL^{\mathsf{T}}$ factorization of $\widehat{\mathcal{C}}_{50}$ in 224-bit outward-rounded dyadic rational interval arithmetic ($\mathbb{Z}\cdot 2^{-224} \subset \mathbb{Q}$) self-containedly certifies $0 \notin [B_{N, \mathrm{lo}}, B_{N, \mathrm{hi}}]$ and $d_N^2 < d_N^{\mathrm{cen}, 2}$ across **all $2 \le N \le 50$** (and $d_N^2 < d_N^{\mathrm{sqf}, 2}$ across **all $4 \le N \le 50$**) to $40\text{–}65$ certified decimal digits without any external transcendence assumptions.
- **Theorem 4 (Prime-Power Coprime-Additivity Obstruction & Composite Schur Complement, Section 5):** Although $V_N^{\mathrm{comp}} := \operatorname{span}\{\tau_k : 6 \le k \le N,\, \omega(k) \ge 2\} \subset \{1\}^\perp$ is orthogonal to $1$, we prove that every $g \in V_\infty^{\mathrm{pp}}$ satisfies the **coprime-additivity jump law**
  $$\Delta_m g \;-\; \sum_{p^\alpha \parallel m} \Delta_{p^\alpha} g \;+\; \bigl(\omega(m) - 1\bigr)\Delta_1 g \;\equiv\; 0 \qquad (\forall\, m \ge 6 \text{ with } \omega(m) \ge 2). \tag{7}$$
  Unlike the squarefree Radical-Jump Rigidity Law $\Delta_m g - \Delta_{\operatorname{rad}(m)} g \equiv 0$ ($\mu(m) = 0$), which decouples $m = 1$ from $m \ge 2$ and is satisfied by $1\cdot\mathbf{1}_{[1, \infty)}$, (7) couples every multi-prime composite $m \ge 6$ directly to $\Delta_1 g$ with weight $\omega(m) - 1 \ge 1$. Projecting onto the explicit sparse integer basis of $(\mathcal{W}_M^{\mathrm{pp}})^\perp \subset \mathbb{Z}^M$ in `fractions.Fraction` arithmetic proves unconditionally for **all $N \ge 2$** that
  $$\inf_{N \ge 2} d_N\bigl(V_N^{\mathrm{pp}}\bigr)^2 \;\ge\; \underline{D}_{100}^{\mathrm{pp}} \;\ge\; R_{100}\bigl(\boldsymbol{\lambda}_{100}^{\mathrm{int}}\bigr) \;=\; \frac{621676929251568300049}{24933450004945398760948} \;\approx\; 0.024933450009 \;>\; d_{10}^2 \;>\; d_{12}^2, \tag{8}$$
  with $\underline{D}_6^{\mathrm{pp}} = \frac{1}{92}$ and $\underline{D}_{10}^{\mathrm{pp}} = \frac{70}{5711}$. Thus adjoining **just the two orthogonal composite modes $\tau_6$ and $\tau_{10}$** to the $7$ prime powers $\le 9$ achieves a strictly smaller distance ($d_{10}^2 \approx 0.02385255$) than the infinite-dimensional space $V_\infty^{\mathrm{pp}}$ of all prime powers up to $\infty$, with exact block Schur-complement distance gain $d_N(V_N^{\mathrm{pp}})^2 - d_N^2 \equiv (\mathbf{z}^*)^{\mathsf{T}}(\pi\widetilde{\mathcal{C}}_{\mathrm{comp}} - G_{\mathrm{comp}, \mathrm{pp}} G_{\mathrm{pp}}^{-1} G_{\mathrm{pp}, \mathrm{comp}})\mathbf{z}^* = \|P_{(V_N^{\mathrm{pp}})^\perp} g_{\mathrm{comp}}^*\|_{\mathcal{H}}^2 > 0$.

---

## 2. Commutative Dilation-Operator Euler Product and Universal Bilinear Prime-Lift Gram Recurrence

For each integer $q \ge 1$, define the pointwise dilation operator $\widetilde{T}_q$ on functions $f : [0, \infty) \to \mathbb{R}$ and the bounded linear dilation operator $T_q : \mathcal{H} \to \mathcal{H}$ on $\mathcal{H} = L^2((1, \infty), x^{-2}\,dx)$ by
$$(\widetilde{T}_q f)(x) \;:=\; q\,f\!\left(\frac{x}{q}\right) \quad (x \ge 0), \qquad (T_q f)(x) \;:=\; q\,f\!\left(\frac{x}{q}\right)\mathbf{1}_{[q, \infty)}(x) \quad (x > 1). \tag{9}$$

### Theorem 1 (Commutative Dilation-Operator Euler Product & Universal Bilinear Prime-Lift Gram Recurrence)

*The families $\{\widetilde{T}_q\}_{q \ge 1}$ on $[0, \infty)$ and $\{T_q\}_{q \ge 1}$ on $\mathcal{H} = L^2((1, \infty), x^{-2}\,dx)$ and the generators $\tau_k$ in (2) satisfy:*

**(a) (Semigroup & Cross-Adjoint Cancellation Identity):** *For all $q, q_1, q_2 \ge 1$, $T_1 = I$, $T_{q_1} T_{q_2} = T_{q_1 q_2} = T_{q_2} T_{q_1}$, the Hilbert-space adjoint $T_q^* : \mathcal{H} \to \mathcal{H}$ acts by $(T_q^* h)(u) = h(q u)$ ($u > 1$), $T_q^* 1 \equiv 1$, $T_q^* T_q = q\,I_{\mathcal{H}}$, and with $g := \gcd(q_1, q_2)$,*
$$T_{q_1}^* T_{q_2} \;\equiv\; g\,T_{q_1 / g}^*\,T_{q_2 / g} \qquad \Bigl(\text{specifically, }(T_{q_1}^* T_{q_2} f)(u) \;=\; q_2\,f\!\left(\frac{q_1 u}{q_2}\right)\mathbf{1}_{[q_2 / q_1, \infty)}(u)\;\;\text{for }u > 1\Bigr). \tag{10}$$
*(Note that $T_{q_1 / g}^*$ and $T_{q_2 / g}$ do not commute for $q_1 \neq q_2$: e.g., $(T_2^* T_3 1)(u) = 3\mathbf{1}_{[3/2, \infty)}(u) \neq 3\mathbf{1}_{[3, \infty)}(u) = (T_3 T_2^* 1)(u)$.)*

**(b) (Pointwise Prime-Lift Recurrence & Operator Euler Product):** *Because $\tau_m(x) = 0$ for all $x \in [0, 1)$ and $m \ge 1$, $(\widetilde{T}_q\tau_m)\big|_{(1, \infty)} \equiv T_q(\tau_m\big|_{(1, \infty)}) \in \mathcal{H}$ for every $m \ge 2$ and $q \ge 1$ (whereas $\tau_1(x) = \lfloor x \rfloor \notin \mathcal{H}$ has $\|\tau_1\big|_{(1, \infty)}\|_{\mathcal{H}} = \infty$). For every prime $p$, every integer $m \ge 1$, and all $x \ge 0$,*
$$\tau_{p m}(x) \;\equiv\; p\,\tau_m\!\left(\frac{x}{p}\right) \;-\; \mathbf{1}_{p \nmid m}\,\tau_m(x), \qquad \tau_k \;\equiv\; \left(\prod_{p^\alpha \parallel k} \widetilde{T}_p^{\alpha - 1}(\widetilde{T}_p - I)\right)\!\tau_1 \qquad (k \ge 1), \tag{11}$$
*and for every prime $p$ and every integer $m \ge 2$, $\tau_{p m} = T_p\tau_m - \mathbf{1}_{p \nmid m}\tau_m$ in $\mathcal{H}$ (in particular, $\tau_{p^\alpha m} = T_p^{\alpha - 1}(T_p - I)\tau_m$ in $\mathcal{H}$ whenever $m \ge 2$ and $p \nmid m$).*

**(c) (Operator Proof of Target Inner Product):** *For every integer $k \ge 2$,*
$$\langle 1,\, \tau_k \rangle_{\mathcal{H}} \;\equiv\; -\Lambda(k), \tag{12}$$
*where $\Lambda(k) = \log p$ if $k = p^\alpha$ ($p$ prime, $\alpha \ge 1$) and $0$ otherwise.*

**(d) (Universal Bilinear Prime-Lift Gram Recurrence & Off-Diagonal Norm Formula):** *For every prime $p$ and all integers $m, n \ge 2$,*
$$\bigl\langle \tau_{p m} + \mathbf{1}_{p \nmid m}\tau_m,\;\; \tau_{p n} + \mathbf{1}_{p \nmid n}\tau_n \bigr\rangle_{\mathcal{H}} \;\equiv\; p\,\langle \tau_m,\, \tau_n \rangle_{\mathcal{H}}. \tag{13}$$
*In particular, for every prime $p$ and every integer $m \ge 2$ with $p \nmid m$,*
$$\|\tau_{p^\alpha m}\|_{\mathcal{H}}^2 \;\equiv\; p^{\alpha - 1}\,\|\tau_{p m}\|_{\mathcal{H}}^2 \quad (\alpha \ge 1), \qquad \langle \tau_m,\, \tau_{p m} \rangle_{\mathcal{H}} \;\equiv\; \frac{1}{2}\Bigl((p - 1)\,\|\tau_m\|_{\mathcal{H}}^2 \;-\; \|\tau_{p m}\|_{\mathcal{H}}^2\Bigr). \tag{14}$$

*Proof.*
**(a)** For $f, h \in \mathcal{H}$, substituting $x = q u$ ($dx / x^2 = du / (q u^2)$) gives
$$\langle h,\, T_q f \rangle_{\mathcal{H}} \;=\; \int_q^{\infty} h(x)\,q\,f\!\left(\frac{x}{q}\right)\frac{dx}{x^2} \;=\; \int_1^{\infty} h(q u)\,f(u)\,\frac{du}{u^2} \;=\; \langle T_q^* h,\, f \rangle_{\mathcal{H}},$$
so $(T_q^* h)(u) = h(q u)$, $T_q^* 1 \equiv 1$, and $(T_{q_1}^* T_{q_2} f)(u) = q_2 f(q_1 u / q_2)\mathbf{1}_{[q_2, \infty)}(q_1 u) = g\,(q_2/g) f\bigl(\frac{(q_1/g)u}{q_2/g}\bigr)\mathbf{1}_{[q_2/g, \infty)}((q_1/g)u)$, proving (10) since $q_1 u \ge q_2 \iff (q_1/g)u \ge q_2/g$ and $(q_1/g)u > q_1/g \ge 1$ for $u > 1$.

**(b)** Write $m = p^a r$ with $a \ge 0$ and $p \nmid r$. Since $\mu(p^{a+1}r / d) \neq 0$ only when $p^a \mid d$, the divisors $d \mid p^{a+1}r$ with $\mu(p^{a+1}r / d) \neq 0$ split into: (i) $d = p^{a+1}d'$ ($d' \mid r$), where $d\mu(p^{a+1}r / d) = p\cdot(p^a d')\mu(r/d')$; and (ii) $d = p^a d'$ ($d' \mid r$), where $d\mu(p^{a+1}r / d) = -(p^a d')\mu(r/d')$. When $a \ge 1$ ($p \mid m$), only $d = p^a d'$ and $d = p^{a-1}d'$ appear in $\tau_m(x)$, so $\tau_{p m}(x) = \sum_{d' \mid r} p^{a+1}d'\mu(r/d')\bigl(\lfloor x/(p^{a+1}d') \rfloor - \lfloor x/(p^a d') \rfloor\bigr) = p\,\tau_m(x/p) = (\widetilde{T}_p\tau_m)(x)$. When $a = 0$ ($p \nmid m = r$), $\tau_{p r}(x) = \sum_{d' \mid r} d'\mu(r/d')\bigl(p\lfloor (x/p)/d' \rfloor - \lfloor x/d' \rfloor\bigr) = p\tau_r(x/p) - \tau_r(x) = ((\widetilde{T}_p - I)\tau_r)(x)$, proving the first identity in (11). Iterating across the prime-power factorization $k = \prod_{p^\alpha \parallel k} p^\alpha$ gives the pointwise operator Euler product in (11). When $m \ge 2$, $\tau_m \in \mathcal{H}$ is $m$-periodic and bounded with $\tau_m\big|_{[0, 1)} \equiv 0$, so $(\widetilde{T}_p\tau_m)(x) = p\tau_m(x/p)\mathbf{1}_{[p, \infty)}(x) = (T_p\tau_m)(x)$ for all $x > 1$.

**(c)** If $\omega(k) \ge 2$, pick a prime power $p^\alpha \parallel k$ and put $m := k / p^\alpha \ge 2$ ($\gcd(p, m) = 1$). Because $m \ge 2$, $\tau_m \in \mathcal{H}$ is a bounded $m$-periodic step function, and part (b) gives $\tau_k = T_p^{\alpha - 1}(T_p - I)\tau_m = (T_p - I) T_p^{\alpha - 1}\tau_m$ in $\mathcal{H}$. Taking inner products with $1 \in \mathcal{H}$ and using $T_p^* 1 = 1$ gives $\langle 1, \tau_k \rangle_{\mathcal{H}} = \langle (T_p^* - I)1, T_p^{\alpha - 1}\tau_m \rangle_{\mathcal{H}} = \langle 0, T_p^{\alpha - 1}\tau_m \rangle_{\mathcal{H}} = 0$. If $k = p^\alpha$ ($\alpha \ge 1$), then $\tau_{p^\alpha} = T_p^{\alpha - 1}\tau_p$ in $\mathcal{H}$, so $T_p^* 1 = 1$ gives $\langle 1, \tau_{p^\alpha} \rangle_{\mathcal{H}} = \langle (T_p^*)^{\alpha - 1}1, \tau_p \rangle_{\mathcal{H}} = \langle 1, \tau_p \rangle_{\mathcal{H}}$. Finally, since $\tau_p(n) = p\lfloor n/p \rfloor - n \equiv -(n \bmod p)$ is $p$-periodic with mean $-\frac{p-1}{2}$, telescoping $\sum_{q=0}^{\infty}\sum_{r=1}^{p-1}(-r)(\frac{1}{qp+r} - \frac{1}{qp+r+1}) = \sum_{q=0}^{\infty}(p\frac{1}{qp+p} - \sum_{r=1}^p \frac{1}{qp+r}) = -\log p$ gives $\langle 1, \tau_p \rangle_{\mathcal{H}} = -\log p$, proving (12).

**(d)** For $m, n \ge 2$, $\tau_m, \tau_n \in \mathcal{H}$, and by (11), $\tau_{p m} + \mathbf{1}_{p \nmid m}\tau_m = T_p\tau_m$ and $\tau_{p n} + \mathbf{1}_{p \nmid n}\tau_n = T_p\tau_n$. Applying $T_p^* T_p = p\,I_{\mathcal{H}}$ from (10) immediately gives $\langle T_p\tau_m, T_p\tau_n \rangle_{\mathcal{H}} = \langle \tau_m, T_p^* T_p\tau_n \rangle_{\mathcal{H}} = p\langle \tau_m, \tau_n \rangle_{\mathcal{H}}$, which is (13). Setting $m = n$ with $p \mid m$ gives $\|\tau_{p m}\|_{\mathcal{H}}^2 = p\|\tau_m\|_{\mathcal{H}}^2$, and setting $m = n$ with $p \nmid m$ in (13) expands $\|\tau_{p m} + \tau_m\|_{\mathcal{H}}^2 = \|\tau_{p m}\|_{\mathcal{H}}^2 + 2\langle \tau_m, \tau_{p m} \rangle_{\mathcal{H}} + \|\tau_m\|_{\mathcal{H}}^2 = p\|\tau_m\|_{\mathcal{H}}^2$, which rearranges to (14). $\blacksquare$

---

## 3. Universal Totient–Von Mangoldt–Cotangent Decomposition, Cyclotomic Probability Measure, and Sign Trichotomy

For each integer $k \ge 2$, let $P_k(x) := \mathcal{B}_{1, k}^{\mathrm{true}}(x) = -\tau_k(x) - \frac{1}{2}\varphi(k)$ denote the centered $k$-periodic Bernoulli step function [13, 14]. On any integer multiple $L$ of $k$, since $\lfloor (L - 1 - n)/d \rfloor = L/d - 1 - \lfloor n/d \rfloor$ for every $d \mid k$ and $\sum_{d \mid k}\mu(k/d) = 0$, we have $\tau_k(L - 1 - n) = -\varphi(k) - \tau_k(n)$, so $P_k$ is **odd under reflection** across the midpoint $(L - 1)/2$:
$$P_k(L - 1 - n) \;\equiv\; -P_k(n) \qquad (\forall\, n \in \mathbb{Z},\;\; \forall\, L \in k\mathbb{Z}_{\ge 1}), \qquad P_k(0) \;=\; -P_k(L - 1) \;=\; -\frac{1}{2}\varphi(k). \tag{15}$$

### Theorem 2 (Universal Totient–Von Mangoldt–Cotangent Decomposition, Cyclotomic Probability-Measure Identity, and Arithmetic Sign Trichotomy)

*For all integers $j, k \ge 2$, let $L := \operatorname{lcm}(j, k)$ and $a_n := \tau_j(n)\tau_k(n)$ ($n \ge 0$).*

**(a) (Finite Cyclotomic Probability-Measure Identity for $\langle P_j, P_k \rangle_{\mathcal{H}}$):** *Define the symmetric weights $w_L(0), w_L(1), \dots, w_L(L - 1) > 0$ on $\{0, 1, \dots, L - 1\}$ by*
$$w_L(n) \;:=\; \begin{cases} \displaystyle \frac{1}{2}\left(1 - \frac{\pi}{L}\cot\!\left(\frac{\pi}{L}\right)\right), & n \in \{0,\, L - 1\}, \\[10pt] \displaystyle \frac{\pi}{2 L}\left[\cot\!\left(\frac{\pi n}{L}\right) - \cot\!\left(\frac{\pi(n + 1)}{L}\right)\right] \;=\; \frac{\pi\sin(\pi/L)}{2 L\,\sin(\pi n/L)\,\sin(\pi(n + 1)/L)}, & 1 \le n \le L - 2. \end{cases} \tag{16}$$
*Then $w_L(L - 1 - n) \equiv w_L(n) > 0$, $\sum_{n=0}^{L-1} w_L(n) \equiv 1$, and for every $L$-periodic sequence $b_m$ satisfying $b_{L - 1 - n} \equiv b_n$ ($0 \le n \le L - 1$), $\sum_{m=1}^{\infty} \frac{b_m}{m(m+1)} \equiv \sum_{n=0}^{L-1} w_L(n) b_n$. In particular, the centered periodic Bernoulli Gram entry is 100% logarithm-free and given in exact finite closed form by*
$$\langle P_j,\, P_k \rangle_{\mathcal{H}} \;\equiv\; \sum_{n=0}^{L-1} w_L(n)\,P_j(n)\,P_k(n) \;\equiv\; \frac{1}{4}\varphi(j)\varphi(k) \;+\; \pi\,\widetilde{\mathcal{C}}_{j, k}, \tag{17}$$
*where the cyclotomic-cotangent coefficient $\widetilde{\mathcal{C}}_{j, k} \in \mathbb{Q}\bigl(\cos(\pi/(2L))\bigr)$ ($\widetilde{\mathcal{C}}_{j, k} \in \mathbb{Q}(\cos(\pi/L))$ whenever $L$ is even, and in particular for every full prefix period $L_N = \operatorname{lcm}(2, \dots, N)$) is given equivalently by*
$$\widetilde{\mathcal{C}}_{j, k} \;:=\; \frac{\sin(\pi/L)}{2 L}\sum_{n=1}^{L-2} \frac{P_j(n)P_k(n) - \frac{1}{4}\varphi(j)\varphi(k)}{\sin(\pi n/L)\,\sin(\pi(n + 1)/L)} \;\equiv\; \frac{1}{2 L}\sum_{n=1}^{L-1} \bigl(a_n - a_{n-1}\bigr)\cot\!\left(\frac{\pi n}{L}\right). \tag{18}$$

**(b) (Universal Totient–Von Mangoldt–Cotangent Decomposition of $G_N$):** *For all integers $j, k \ge 2$, the infinite-horizon NBBD Gram entry decomposes into a rank-2 totient–von Mangoldt logarithmic term plus $\pi\,\widetilde{\mathcal{C}}_{j, k}$:*
$$\langle \tau_j,\, \tau_k \rangle_{\mathcal{H}} \;\equiv\; \frac{1}{2}\Bigl(\varphi(j)\,\Lambda(k) \;+\; \varphi(k)\,\Lambda(j)\Bigr) \;+\; \pi\,\widetilde{\mathcal{C}}_{j, k}. \tag{19}$$
*Moreover, both the logarithmic part $\mathcal{L}_{j, k} := \frac{1}{2}(\varphi(j)\Lambda(k) + \varphi(k)\Lambda(j))$ and the cyclotomic-cotangent matrix $\widetilde{\mathcal{C}}_{j, k}$ independently satisfy the universal bilinear prime-lift recurrence (13).*

**(c) (Universal Arithmetic Sign Trichotomy of $\widetilde{\mathcal{C}}_{k, k}$ and Super-Totient Overshoot):** *For every integer $k \ge 2$, the diagonal entry $\widetilde{\mathcal{C}}_{k, k}$ obeys the exact three-way sign classification:*
1. **Binary Prime Powers ($k = 2^\alpha,\, \alpha \ge 1$):** *$P_{2^\alpha}(n)^2 \equiv \frac{1}{4}\varphi(2^\alpha)^2$ for all $n \in \mathbb{Z}$, so $\widetilde{\mathcal{C}}_{2^\alpha, 2^\alpha} = 0$ and $\|\tau_{2^\alpha}\|_{\mathcal{H}}^2 \equiv 2^{\alpha - 1}\log 2$.*
2. **Odd Prime Powers ($k = p^\alpha,\, p \ge 3\text{ prime},\, \alpha \ge 1$):** *$|P_{p^\alpha}(n)| \le \frac{1}{2}\varphi(p^\alpha)$ for all $n \in \mathbb{Z}$ with strict inequality on $p^{\alpha - 1} \le n < (p - 1)p^{\alpha - 1}$, yielding the closed form*
   $$\widetilde{\mathcal{C}}_{p^\alpha, p^\alpha} \;\equiv\; -p^{\alpha - 2}\,\frac{\sin(\pi/p)}{2}\sum_{n=1}^{p-2} \frac{n(p - 1 - n)}{\sin(\pi n/p)\,\sin(\pi(n + 1)/p)} \;<\; 0, \qquad \|\tau_{p^\alpha}\|_{\mathcal{H}}^2 \;<\; \varphi(p^\alpha)\log p. \tag{20}$$
3. **Multi-Prime Composites ($k \ge 6\text{ with }\omega(k) \ge 2$):** *Since $\Lambda(k) = 0$, $\|\tau_k\|_{\mathcal{H}}^2 \equiv \pi\,\widetilde{\mathcal{C}}_{k, k} > 0$, so $\widetilde{\mathcal{C}}_{k, k} > 0$ and every multi-prime composite $k$ exhibits strict super-totient overshoot:*
   $$\max_{1 \le n \le k - 2} |P_k(n)| \;>\; \frac{1}{2}\varphi(k) \qquad (\forall\, k \ge 6 \text{ with } \omega(k) \ge 2). \tag{21}$$

*Proof.*
**(a)** Telescoping $\sum_{n=1}^{L-2} w_L(n) = \frac{\pi}{2L}\bigl(\cot(\frac{\pi}{L}) - \cot(\frac{\pi(L-1)}{L})\bigr) = \frac{\pi}{L}\cot(\frac{\pi}{L})$ gives $\sum_{n=0}^{L-1} w_L(n) = 1$, and $w_L(n) > 0$ since $\cot x$ is strictly decreasing on $(0, \pi)$ and $\frac{\pi}{L}\cot(\frac{\pi}{L}) < 1$. For any $L$-periodic sequence $b_m$ with $b_{L - 1 - n} = b_n$, pair each residue class $m \equiv n \pmod L$ with $m \equiv L - 1 - n \pmod L$ in $\sum_{m=1}^{\infty} \frac{b_m}{m(m+1)}$. For $1 \le n \le L - 2$, partial fractions and Euler's identity $\sum_{q=0}^{\infty}\bigl(\frac{1}{q + x} - \frac{1}{q + 1 - x}\bigr) = \pi\cot(\pi x)$ ($0 < x < 1$) give
$$\frac{1}{2}\sum_{q=0}^{\infty}\left[\frac{1}{(q L + n)(q L + n + 1)} + \frac{1}{(q L + L - 1 - n)(q L + L - n)}\right] \;=\; \frac{\pi}{2 L}\left[\cot\!\left(\frac{\pi n}{L}\right) - \cot\!\left(\frac{\pi(n + 1)}{L}\right)\right] \;=\; w_L(n),$$
while for the boundary pair $n = 0$ ($m = q L, q \ge 1$) and $n = L - 1$ ($m = q L + L - 1, q \ge 0$),
$$\frac{1}{2 L}\left[\sum_{q=1}^{\infty}\left(\frac{1}{q} - \frac{1}{q + 1/L}\right) + \sum_{q=0}^{\infty}\left(\frac{1}{q + 1 - 1/L} - \frac{1}{q + 1}\right)\right] \;=\; \frac{1}{2 L}\Bigl(L - \pi\cot(\pi/L)\Bigr) \;=\; w_L(0) \;=\; w_L(L - 1).$$
Applying this to $b_n := P_j(n)P_k(n)$ (which satisfies $b_{L - 1 - n} = (-P_j(n))(-P_k(n)) = b_n$ by (15) and $b_0 = b_{L-1} = \frac{1}{4}\varphi(j)\varphi(k)$) and subtracting $\frac{1}{4}\varphi(j)\varphi(k)\sum_{n=0}^{L-1} w_L(n) = \frac{1}{4}\varphi(j)\varphi(k)$ yields (17) and the first expression in (18). Summation by parts on $\tilde{b}_n := P_j(n)P_k(n) - \frac{1}{4}\varphi(j)\varphi(k)$ ($\tilde{b}_0 = \tilde{b}_{L-1} = 0$) gives $\widetilde{\mathcal{C}}_{j, k} = \frac{1}{2L}\sum_{n=1}^{L-1}(b_n - b_{n-1})\cot(\frac{\pi n}{L})$. Because $\cot(\frac{\pi n}{L}) = \frac{\cos(\pi n/L)}{\sin(\pi n/L)}$ with $\cos(\frac{\pi n}{L}), \sin(\frac{\pi n}{L}) = \cos(\frac{\pi(L - 2n)}{2L}) \in \mathbb{Q}(\cos(\frac{\pi}{2L}))$ (and $\sin(\frac{\pi n}{L}) = \cos(\frac{\pi(L/2 - n)}{L}) \in \mathbb{Q}(\cos(\frac{\pi}{L}))$ when $2 \mid L$; for odd $L = 3$, $\widetilde{\mathcal{C}}_{3, 3} = -\frac{1}{3\sqrt{3}} \in \mathbb{Q}(\cos(\frac{\pi}{6})) \setminus \mathbb{Q}(\cos(\frac{\pi}{3}))$), $\widetilde{\mathcal{C}}_{j, k}$ lies in $\mathbb{Q}(\cos(\frac{\pi}{2L}))$ universally and in $\mathbb{Q}(\cos(\frac{\pi}{L}))$ whenever $2 \mid L$. Finally, since $a_n = \tau_j(n)\tau_k(n) = b_n + \frac{1}{2}\varphi(j)P_k(n) + \frac{1}{2}\varphi(k)P_j(n) + \frac{1}{4}\varphi(j)\varphi(k)$ for all $0 \le n \le L - 1$ (including $n = 0$, where $\tau_j(0) = 0$ and $P_j(0) = -\frac{1}{2}\varphi(j)$), and since $\Delta P_k(L - n) = P_k(L - n) - P_k(L - 1 - n) = -P_k(n - 1) + P_k(n) = \Delta P_k(n)$ is even under $n \mapsto L - n$ while $\cot(\frac{\pi(L - n)}{L}) = -\cot(\frac{\pi n}{L})$ is odd, $\sum_{n=1}^{L-1} \Delta P_k(n)\cot(\frac{\pi n}{L}) = 0$, proving the equivalence of both formulas in (18).

**(b)** Writing $\tau_k = -P_k - \frac{1}{2}\varphi(k)$ on $(1, \infty)$ and using $\langle 1, \tau_k \rangle_{\mathcal{H}} = -\Lambda(k)$ from Theorem 1(c) and $\|1\|_{\mathcal{H}}^2 = 1$ gives $\langle 1, P_k \rangle_{\mathcal{H}} = \Lambda(k) - \frac{1}{2}\varphi(k)$, whence
$$\langle \tau_j,\, \tau_k \rangle_{\mathcal{H}} \;=\; \langle P_j,\, P_k \rangle_{\mathcal{H}} + \frac{1}{2}\varphi(j)\langle 1, P_k \rangle_{\mathcal{H}} + \frac{1}{2}\varphi(k)\langle 1, P_j \rangle_{\mathcal{H}} + \frac{1}{4}\varphi(j)\varphi(k) \;=\; \frac{1}{2}\Bigl(\varphi(j)\Lambda(k) + \varphi(k)\Lambda(j)\Bigr) + \Bigl(\langle P_j, P_k \rangle_{\mathcal{H}} - \frac{1}{4}\varphi(j)\varphi(k)\Bigr).$$
Substituting (17) proves (19). Since $\varphi(p m) + \mathbf{1}_{p \nmid m}\varphi(m) = p\varphi(m)$ and $\Lambda(p m) + \mathbf{1}_{p \nmid m}\Lambda(m) = \Lambda(m)$ for every prime $p$ and every $m \ge 2$, $\mathcal{L}_{j, k}$ satisfies (13), and subtracting $\mathcal{L}_{j, k}$ from $\langle \tau_j, \tau_k \rangle_{\mathcal{H}}$ shows that $\widetilde{\mathcal{C}}_{j, k}$ also satisfies (13).

**(c)** For $k = 2^\alpha$ ($\alpha \ge 1$), $\tau_{2^\alpha}(n) = -2^{\alpha - 1}\mathbf{1}_{2^{\alpha-1} \le (n \bmod 2^\alpha) < 2^\alpha}$, so $P_{2^\alpha}(n) = \pm 2^{\alpha - 2}$ and $P_{2^\alpha}(n)^2 - \frac{1}{4}\varphi(2^\alpha)^2 \equiv 0$, giving $\widetilde{\mathcal{C}}_{2^\alpha, 2^\alpha} = 0$ by (18). For an odd prime power $k = p^\alpha$ ($p \ge 3$), $\tau_{p^\alpha}(n) = -p^{\alpha - 1}\lfloor (n \bmod p^\alpha)/p^{\alpha - 1} \rfloor$, so at $\alpha = 1$, $P_p(n) = n - \frac{p - 1}{2}$ ($0 \le n \le p - 1$) satisfies $P_p(n)^2 - \frac{1}{4}(p - 1)^2 = -n(p - 1 - n) < 0$ for all $1 \le n \le p - 2$; substituting this into (18) and applying $\widetilde{\mathcal{C}}_{p^\alpha, p^\alpha} = p^{\alpha - 1}\widetilde{\mathcal{C}}_{p, p}$ from part (b) yields (20). Finally, if $\omega(k) \ge 2$, then $\Lambda(k) = 0$, so (19) gives $\|\tau_k\|_{\mathcal{H}}^2 = \pi\,\widetilde{\mathcal{C}}_{k, k} > 0$; since the kernel $\frac{\sin(\pi/k)}{2k\sin(\pi n/k)\sin(\pi(n+1)/k)} > 0$ in (18) is strictly positive for all $1 \le n \le k - 2$, $\widetilde{\mathcal{C}}_{k, k} > 0$ forces $\max_{1 \le n \le k - 2}\bigl(P_k(n)^2 - \frac{1}{4}\varphi(k)^2\bigr) > 0$, proving (21). $\blacksquare$

**Example 3.1 (Exact Elementary Gram Matrix $G_{\{2, 3, 4, 6\}}$):** Evaluating (18)–(19) on $\{2, 3, 4, 6\}$ gives $\widetilde{\mathcal{C}}_{2, 2} = \widetilde{\mathcal{C}}_{4, 4} = 0$, $\widetilde{\mathcal{C}}_{3, 3} = \widetilde{\mathcal{C}}_{2, 3} = -\frac{1}{3\sqrt{3}}$, $\widetilde{\mathcal{C}}_{2, 4} = -\frac{1}{4}$, $\widetilde{\mathcal{C}}_{3, 4} = -\frac{1}{2} + \frac{1}{3\sqrt{3}}$, $\widetilde{\mathcal{C}}_{6, 6} = +\frac{1}{\sqrt{3}}$, $\widetilde{\mathcal{C}}_{2, 6} = -\frac{1}{2\sqrt{3}}$, $\widetilde{\mathcal{C}}_{3, 6} = -\frac{2}{3\sqrt{3}}$, and $\widetilde{\mathcal{C}}_{4, 6} = \frac{1}{2} - \frac{1}{\sqrt{3}}$ (verified to $10^{-15}$ in `src/verify_operator_euler_product_and_radical_rigidity.py`), so in particular $\|\tau_6\|_{\mathcal{H}}^2 = \frac{\pi}{\sqrt{3}}$, $\langle \tau_2, \tau_6 \rangle_{\mathcal{H}} = \log 2 - \frac{\pi}{2\sqrt{3}}$, $\langle \tau_3, \tau_6 \rangle_{\mathcal{H}} = \log 3 - \frac{2\pi}{3\sqrt{3}}$, and $\langle \tau_4, \tau_6 \rangle_{\mathcal{H}} = \log 2 + \frac{\pi}{2} - \frac{\pi}{\sqrt{3}}$.

---

## 4. Logarithm-Free Centered Distance, Exact Parabolic Shift Decomposition, and 224-Bit Dyadic Rational Certificates

For any $g = \sum_{k=2}^N c_k\tau_k \in V_N$, let $\beta(\mathbf{c}) := 1 + \frac{1}{2}\sum_{k=2}^N c_k\varphi(k)$ denote the asymptotic mean residual, and define the centered totient hyperplane $V_N^{\mathrm{cen}} := \{g \in V_N : \beta(\mathbf{c}) = 0\}$ and its squared distance $d_N^{\mathrm{cen}, 2} := \inf_{g \in V_N^{\mathrm{cen}}}\|1 - g\|_{\mathcal{H}}^2$. Since $\varphi(2) = 1$, substituting $c_2 = -2(1 - \beta) - \sum_{k=3}^N \varphi(k)c_k$ decomposes every residual $1 - g$ uniquely as
$$1 - g \;\equiv\; (1 - \beta)\,Q_0 \;+\; \beta\cdot 1 \;+\; \sum_{k=3}^N c_k\,Q_k, \qquad Q_0 \;:=\; 1 + 2\tau_2 \equiv -2 P_2, \qquad Q_k \;:=\; -\tau_k + \varphi(k)\tau_2 \equiv P_k - \varphi(k)P_2 \;\;(3 \le k \le N). \tag{22}$$

### Theorem 3 (Logarithm-Free Algebraic Centered Distance $d_N^{\mathrm{cen}, 2} = 1 - \pi\alpha_N$, Totient-Neutral Deflation $\widehat{\mathcal{C}}_N$, Exact Parabolic Shift Identity, and 224-Bit Certified Separation across $2 \le N \le 50$)

*For every integer $N \ge 2$, let $L_N := \operatorname{lcm}(2, 3, \dots, N)$, $\tilde{\mathbf{b}}_N := (\widetilde{\mathcal{C}}_{2, k})_{3 \le k \le N} \in \mathbb{Q}(\cos(\pi/L_N))^{N-2}$, and $\widetilde{\boldsymbol{\Lambda}}_N := (\Lambda(k) - \varphi(k)\log 2)_{3 \le k \le N} \in \mathbb{R}^{N-2}$.*

**(a) (Logarithm-Free Algebraic Centered Minimization & Totient-Neutral Deflation):** *For all $g \in V_N$,*
$$\|1 - g\|_{\mathcal{H}}^2 \;\equiv\; 1 \;+\; 2\,\beta(\mathbf{c})\bigl(\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}\bigr) \;+\; \pi\,\mathbf{c}^{\mathsf{T}}\widetilde{\mathcal{C}}_N\mathbf{c}. \tag{23}$$
*Furthermore, $\|Q_0\|_{\mathcal{H}}^2 \equiv 1$, $\langle Q_0, Q_k \rangle_{\mathcal{H}} \equiv -2\pi\,\widetilde{\mathcal{C}}_{2, k}$, and the infinite-horizon Gram matrix of the totient-neutral basis $\{Q_3, \dots, Q_N\}$ is 100% logarithm-free and strictly positive definite:*
$$\bigl(\langle Q_j,\, Q_k \rangle_{\mathcal{H}}\bigr)_{3 \le j, k \le N} \;\equiv\; \pi\,\widehat{\mathcal{C}}_N \;\succ\; 0, \qquad \widehat{\mathcal{C}}_{j, k} \;:=\; \widetilde{\mathcal{C}}_{j, k} \;-\; \varphi(j)\,\widetilde{\mathcal{C}}_{2, k} \;-\; \varphi(k)\,\widetilde{\mathcal{C}}_{2, j} \;\in\; \mathbb{Q}\bigl(\cos(\pi/L_N)\bigr). \tag{24}$$
*Consequently, on $V_N^{\mathrm{cen}}$ ($\beta(\mathbf{c}) = 0$), all prime logarithms vanish identically, the unique centered minimizer $\tilde{\mathbf{c}}_N^{\mathrm{cen}, *} = (c_{N, 3}^{\mathrm{cen}, *}, \dots, c_{N, N}^{\mathrm{cen}, *})^{\mathsf{T}} = 2\,\widehat{\mathcal{C}}_N^{-1}\tilde{\mathbf{b}}_N \in \mathbb{Q}(\cos(\pi/L_N))^{N-2}$ ($c_{N, 2}^{\mathrm{cen}, *} = -2 - \sum_{k=3}^N \varphi(k)c_{N, k}^{\mathrm{cen}, *}$) is algebraic, and*
$$d_N^{\mathrm{cen}, 2} \;\equiv\; 1 \;-\; \pi\,\alpha_N, \qquad \alpha_N \;:=\; 4\,\tilde{\mathbf{b}}_N^{\mathsf{T}}\widehat{\mathcal{C}}_N^{-1}\tilde{\mathbf{b}}_N \;\in\; \mathbb{Q}\bigl(\cos(\pi/L_N)\bigr), \tag{25}$$
*with $\alpha_2 = 0$, $\alpha_3 = \frac{4\sqrt{3}}{27}$, $\alpha_{D(4)} = \frac{1}{4}$, $\alpha_4 = \frac{19 + \sqrt{3}}{72}$, $\alpha_{D(6)} = \frac{\sqrt{3}}{6}$, $\alpha_{\{2,3,4,6\}} = \frac{2 + 5\sqrt{3}}{36}$, and $\alpha_{D(12)} = \frac{2 + 3\sqrt{3}}{24}$. For any $S \subseteq \{2, \dots, N\}$ with period $L_S := \operatorname{lcm}_{k \in S}(k)$, $d^{\mathrm{cen}}(S)^2 \ge 2w_{L_S}(0) = 1 - \frac{\pi}{L_S}\cot(\frac{\pi}{L_S})$ ($\alpha(S) \le \frac{1}{L_S}\cot(\frac{\pi}{L_S})$), with equality on the divisor-closed subspace $D^*(L) := \{k \ge 2 : k \mid L\}$ if and only if $\varphi(L) \le 2$ ($L \in \{2, 3, 4, 6\}$). Moreover, $\lim_{N \to \infty} d_N^{\mathrm{cen}, 2} = \lim_{N \to \infty} d_N^2$ unconditionally ($\text{RH} \iff \lim_{N \to \infty}\alpha_N = \frac{1}{\pi}$).*

**(b) (Exact All-$N \ge 2$ Parabolic Shift Decomposition & Forward-Only $LDL^{\mathsf{T}}$ Prefix Formulas):** *For every integer $N \ge 2$ and every $\beta \in \mathbb{R}$,*
$$\inf_{\substack{g \in V_N \\ \beta(\mathbf{c}) = \beta}} \|1 - g\|_{\mathcal{H}}^2 \;\equiv\; d_N^{\mathrm{cen}, 2} \;+\; 2\,B_N\,\beta \;+\; A_N\,\beta^2, \qquad d_N^{\mathrm{cen}, 2} - d_N^2 \;\equiv\; \frac{B_N^2}{A_N} \;\ge\; 0, \qquad \beta(\mathbf{c}_N^*) \;\equiv\; -\frac{B_N}{A_N}, \tag{26}$$
*where $B_N := \pi\,\alpha_N + \boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}_N^{\mathrm{cen}, *} = \pi\,\alpha_N - 2\log 2 + \widetilde{\boldsymbol{\Lambda}}_N^{\mathsf{T}}\tilde{\mathbf{c}}_N^{\mathrm{cen}, *}$ and $A_N := \inf_{h \in \operatorname{span}\{Q_3, \dots, Q_N\}}\|-2\tau_2 - h\|_{\mathcal{H}}^2 = 4\log 2 - \frac{1}{\pi}(2\pi\tilde{\mathbf{b}}_N + \widetilde{\boldsymbol{\Lambda}}_N)^{\mathsf{T}}\widehat{\mathcal{C}}_N^{-1}(2\pi\tilde{\mathbf{b}}_N + \widetilde{\boldsymbol{\Lambda}}_N) > 0$, so $d_N^2 < d_N^{\mathrm{cen}, 2} \iff B_N \neq 0$. If $\widehat{\mathcal{C}}_{N_{\max}} = L D L^{\mathsf{T}}$ is the unit lower-triangular $LDL^{\mathsf{T}}$ factorization of $\widehat{\mathcal{C}}_{N_{\max}}$ and $\mathbf{y}^{(c)} := 2 L^{-1}\tilde{\mathbf{b}}_{N_{\max}}$, $\mathbf{y}^{(\Lambda)} := L^{-1}\widetilde{\boldsymbol{\Lambda}}_{N_{\max}}$, then for every $3 \le N \le N_{\max}$,*
$$\alpha_N \;\equiv\; \sum_{k=3}^N \frac{\bigl(y_{k-2}^{(c)}\bigr)^2}{D_{k-2}}, \qquad B_N \;\equiv\; -2\log 2 + \sum_{k=3}^N \frac{y_{k-2}^{(c)}\bigl(\pi y_{k-2}^{(c)} + y_{k-2}^{(\Lambda)}\bigr)}{D_{k-2}}, \qquad A_N \;\equiv\; 4\log 2 - \frac{1}{\pi}\sum_{k=3}^N \frac{\bigl(\pi y_{k-2}^{(c)} + y_{k-2}^{(\Lambda)}\bigr)^2}{D_{k-2}}. \tag{27}$$

**(c) (Self-Contained 224-Bit Dyadic Rational Certification of $B_N \neq 0$, $d_N^2 < d_N^{\mathrm{cen}, 2}$ across $2 \le N \le 50$, and $d_N^2 < d_N^{\mathrm{sqf}, 2}$ across $4 \le N \le 50$):** *For every integer $2 \le N \le 50$, outward-rounded 224-bit dyadic rational interval evaluation ($\mathbb{Z}\cdot 2^{-224} \subset \mathbb{Q}$) of (27) certifies $0 \notin [B_{N, \mathrm{lo}}, B_{N, \mathrm{hi}}]$ and $A_{N, \mathrm{lo}} > 0$, proving self-containedly that $B_N \neq 0$, $d_N^2 < d_N^{\mathrm{cen}, 2}$, and $\beta(\mathbf{c}_N^*) \neq 0$ for all $2 \le N \le 50$ (Table 1), as well as $d_N^2 < d_N^{\mathrm{sqf}, 2}$ and $d_N^{\mathrm{cen}, 2} < d_N^{\mathrm{cen}}(V_N^{\mathrm{sqf}})^2$ for all $4 \le N \le 50$ (Table 2).*

*Proof.*
**(a)** Substituting $\langle 1, g \rangle_{\mathcal{H}} = -\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}$ and $\mathbf{c}^{\mathsf{T}} G_N \mathbf{c} = (\boldsymbol{\varphi}_N^{\mathsf{T}}\mathbf{c})(\boldsymbol{\Lambda}_N^{\mathsf{T}}\mathbf{c}) + \pi\,\mathbf{c}^{\mathsf{T}}\widetilde{\mathcal{C}}_N\mathbf{c}$ from (19) into $\|1 - g\|_{\mathcal{H}}^2 = 1 - 2\langle 1, g \rangle_{\mathcal{H}} + \mathbf{c}^{\mathsf{T}} G_N \mathbf{c}$ gives (23). Using $\varphi(2) = 1$ and $\widetilde{\mathcal{C}}_{2, 2} = 0$ in (17) gives $\|Q_0\|_{\mathcal{H}}^2 = 4\langle P_2, P_2 \rangle_{\mathcal{H}} = 1$, $\langle Q_0, Q_k \rangle_{\mathcal{H}} = -2\langle P_2, P_k - \varphi(k)P_2 \rangle_{\mathcal{H}} = -2\pi\,\widetilde{\mathcal{C}}_{2, k}$, and (24); since $\{Q_3, \dots, Q_N\}$ are linearly independent in $\mathcal{H}$, $\widehat{\mathcal{C}}_N \succ 0$. Setting $\beta = 0$ in (22) gives $\|1 - g\|_{\mathcal{H}}^2 = 1 + \pi\bigl(\tilde{\mathbf{c}}^{\mathsf{T}}\widehat{\mathcal{C}}_N\tilde{\mathbf{c}} - 4\tilde{\mathbf{b}}_N^{\mathsf{T}}\tilde{\mathbf{c}}\bigr)$, whose unique minimizer is $\tilde{\mathbf{c}}_N^{\mathrm{cen}, *} = 2\,\widehat{\mathcal{C}}_N^{-1}\tilde{\mathbf{b}}_N$ with minimum (25). By (17), $\|1 - g\|_{\mathcal{H}}^2 = \sum_{n=0}^{L_S-1} w_{L_S}(n)(1 - g(n))^2 \ge w_{L_S}(0) + w_{L_S}(L_S - 1) = 1 - \frac{\pi}{L_S}\cot(\frac{\pi}{L_S})$ since $1 - g(0) = 1$ and $1 - g(L_S - 1) = -1$; equality requires $g(n) = 1$ for all $1 \le n \le L_S - 2$, which on $D^*(L)$ holds when $\varphi(L) \le 2$ ($L \in \{2, 3, 4, 6\}$, since every $2 \le n < L/2$ divides $L$) and fails whenever $\varphi(L) \ge 4$ because at the smallest $n \in [2, L/2)$ with $\gcd(n, L) = 1$, $\Delta_n g = \sum_{d \mid \gcd(n, L)} \xi_d = \xi_1 = g(1)$ prevents $g(n) = g(n-1) = g(1) = 1$. Finally, for any $g \in V_N$ and $m \ge \lfloor \log_2 N \rfloor + 1$, adjoining $\tilde{g}_m := g - \frac{2\beta(\mathbf{c})}{\varphi(2^m)}\tau_{2^m} \in V_{2^m}^{\mathrm{cen}}$ gives $d_{2^m}^{\mathrm{cen}, 2} \le \|1 - \tilde{g}_m\|_{\mathcal{H}}^2 \le \bigl(\|1 - g\|_{\mathcal{H}} + \frac{2|\beta(\mathbf{c})|}{\varphi(2^m)}\|\tau_{2^m}\|_{\mathcal{H}}\bigr)^2 = \bigl(\|1 - g\|_{\mathcal{H}} + 2|\beta(\mathbf{c})|\sqrt{2^{1-m}\log 2}\bigr)^2 \to \|1 - g\|_{\mathcal{H}}^2$ as $m \to \infty$ by Theorem 2(c)(1), so $\lim_{N \to \infty} d_N^{\mathrm{cen}, 2} = \lim_{N \to \infty} d_N^2$.

**(b)** Writing $\tilde{\mathbf{c}} = (1 - \beta)\tilde{\mathbf{c}}_N^{\mathrm{cen}, *} + \mathbf{u}$ in (22) sets $1 - g = (1 - \beta)Q_N^{\mathrm{cen}, *} + \beta\cdot 1 + \sum_{k=3}^N u_k Q_k$, where $Q_N^{\mathrm{cen}, *} := Q_0 + \sum_{k=3}^N c_{N, k}^{\mathrm{cen}, *} Q_k$ is orthogonal to $\operatorname{span}\{Q_3, \dots, Q_N\}$ and has $\|Q_N^{\mathrm{cen}, *}\|_{\mathcal{H}}^2 = d_N^{\mathrm{cen}, 2} = 1 - \pi\alpha_N$ and $\langle 1, Q_N^{\mathrm{cen}, *} \rangle_{\mathcal{H}} = 1 - 2\log 2 + \widetilde{\boldsymbol{\Lambda}}_N^{\mathsf{T}}\tilde{\mathbf{c}}_N^{\mathrm{cen}, *} = d_N^{\mathrm{cen}, 2} + B_N$. Expanding the norm squared and minimizing over $\mathbf{u} \in \mathbb{R}^{N-2}$ ($\mathbf{u}^*(\beta) = -\frac{\beta}{\pi}\widehat{\mathcal{C}}_N^{-1}\widetilde{\boldsymbol{\Lambda}}_N$) yields (26). Since the leading $(N-2)\times(N-2)$ block of $\widehat{\mathcal{C}}_{N_{\max}} = L D L^{\mathsf{T}}$ is $L_N D_N L_N^{\mathsf{T}}$, $\mathbf{u}^{\mathsf{T}}\widehat{\mathcal{C}}_N^{-1}\mathbf{v} = (L_N^{-1}\mathbf{u})^{\mathsf{T}} D_N^{-1} (L_N^{-1}\mathbf{v})$ gives (27).

**(c)** At $N = 2$, $\alpha_2 = 0$ and $B_2 = -2\log 2 < 0$. At $N = 3$, $\widehat{\mathcal{C}}_{3, 3} = \frac{\sqrt{3}}{3}$, $\widetilde{\mathcal{C}}_{2, 3} = -\frac{1}{3\sqrt{3}}$, $c_{3, 3}^{\mathrm{cen}, *} = -\frac{2}{3}$, $c_{3, 2}^{\mathrm{cen}, *} = -\frac{2}{3}$, and $\alpha_3 = \frac{4\sqrt{3}}{27}$, so $B_3 = \frac{4\sqrt{3}}{27}\pi - \frac{2}{3}\log 6 < \frac{4(7/4)(22/7)}{27} - \frac{2}{3}(\frac{7}{4}) = \frac{22}{27} - \frac{7}{6} = -\frac{19}{54} < 0$. For all $3 \le N \le 50$, `src/verify_cyclotomic_cotangent_gram_and_baker_separation.py` enclosures $\pi$, $\{\log p\}_{p \le 47}$, and $\widehat{\mathcal{C}}_{50}$ in outward-rounded 224-bit dyadic rational intervals $[x_{\mathrm{lo}}, x_{\mathrm{hi}}] \subset \mathbb{Z}\cdot 2^{-224}$ and executes the forward-only $LDL^{\mathsf{T}}$ recursion (27); for every $2 \le N \le 50$, $D_{N-2, \mathrm{lo}} > 0$, $A_{N, \mathrm{lo}} > 0$, and $0 \notin [B_{N, \mathrm{lo}}, B_{N, \mathrm{hi}}]$ ($\min_{2 \le N \le 50} B_{N, \mathrm{lo}}^2 / A_{N, \mathrm{hi}} \ge 2.77 \times 10^{-7}$ at $N = 40$, with interval width $\le 1.88 \times 10^{-41}$), certifying $B_N \neq 0$ and $d_N^2 < d_N^{\mathrm{cen}, 2}$ across $2 \le N \le 50$ and $d_N^{\mathrm{sqf}, 2} - d_N^2 \ge +4.74 \times 10^{-4} > 0$ across $4 \le N \le 50$. $\blacksquare$

**Remark 4.1 (Algebraic Linear-Form Representation of $B_N$ for General $N \ge 3$):** Because $\alpha_N \ge \alpha_3 = \frac{4\sqrt{3}}{27} > 0$ and $c_{N, k}^{\mathrm{cen}, *} \in \mathbb{Q}(\cos(\pi/L_N)) \subset \overline{\mathbb{Q}} \cap \mathbb{R}$ are real algebraic numbers, writing $\pi = -i\log(-1)$ expresses the slope $B_N$ for every $N \ge 3$ as a homogeneous linear form in logarithms of algebraic numbers:
$$B_N \;=\; (-i\,\alpha_N)\log(-1) \;+\; \sum_{p \le N}\left(\sum_{a=1}^{\lfloor \log_p N \rfloor} c_{N, p^a}^{\mathrm{cen}, *}\right)\log p, \tag{28}$$
with algebraic coefficients in $\mathbb{Q}(i, \cos(\pi/L_N))$ and non-zero coefficient $-i\alpha_N \neq 0$ on $\log(-1)$ (cf. Baker [21]).

| $N$ | Certified $\alpha_N$ (Outward-Rounded $18$ Digits) | Certified $d_N^{\mathrm{cen}, 2} = 1 - \pi\alpha_N$ | Certified $d_N^2 = d_N^{\mathrm{cen}, 2} - B_N^2 / A_N$ | Certified Slope $B_N$ | Certified Curvature $A_N$ | Parabolic Gap $B_N^2 / A_N$ | Width of $[d_{N, \mathrm{lo}}^2, d_{N, \mathrm{hi}}^2]$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $2$ | $0$ | $1$ | $[0.306852819440054690, \dots 691]$ | $[-1.386294361120, \dots 119]$ | $[2.772588722239, \dots 240]$ | $6.931 \times 10^{-1}$ | $2.82 \times 10^{-66}$ |
| $3$ | $[0.256600119639833673, \dots 674]$ | $[0.193866949229236510, \dots 511]$ | $[0.095747783927754026, \dots 027]$ | $[-0.388373262048, \dots 047]$ | $[1.537251058040, \dots 041]$ | $9.812 \times 10^{-2}$ | $1.88 \times 10^{-64}$ |
| $4$ | $[0.287945150105123295, \dots 296]$ | $[0.095393631792934373, \dots 374]$ | $[0.065951694215569653, \dots 654]$ | $[-0.179479996895, \dots 894]$ | $[1.094121920499, \dots 500]$ | $2.944 \times 10^{-2}$ | $4.88 \times 10^{-64}$ |
| $6$ | $[0.305127707467944558, \dots 559]$ | $[0.041413035812009882, \dots 883]$ | $[0.036208034636905628, \dots 629]$ | $[-0.060298716231, \dots 230]$ | $[0.698546466520, \dots 521]$ | $5.205 \times 10^{-3}$ | $1.90 \times 10^{-63}$ |
| $10$ | $[0.310581275638884556, \dots 557]$ | $[0.024280146110333660, \dots 661]$ | $[0.023852553318882999, \dots 000]$ | $[-0.013286191156, \dots 155]$ | $[0.412829399693, \dots 694]$ | $4.276 \times 10^{-4}$ | $2.49 \times 10^{-62}$ |
| $12$ | $[0.311781777484982817, \dots 818]$ | $[0.020508658330010379, \dots 380]$ | $[0.020498147332894166, \dots 167]$ | $[-0.001897288569, \dots 568]$ | $[0.342470259803, \dots 804]$ | $1.051 \times 10^{-5}$ | $2.16 \times 10^{-61}$ |
| $20$ | $[0.313045218576600711, \dots 712]$ | $[0.016539441078340133, \dots 134]$ | $[0.016538252109392515, \dots 516]$ | $[+0.000493245754, \dots 755]$ | $[0.204623825398, \dots 399]$ | $1.189 \times 10^{-6}$ | $4.52 \times 10^{-57}$ |
| $30$ | $[0.313659731468246124, \dots 125]$ | $[0.014608891892410683, \dots 684]$ | $[0.014556733576996992, \dots 993]$ | $[+0.002657895107, \dots 108]$ | $[0.135441613641, \dots 642]$ | $5.216 \times 10^{-5}$ | $2.20 \times 10^{-51}$ |
| $40$ | $[0.314255855619035737, \dots 738]$ | $[0.012736112639662581, \dots 582]$ | $[0.012735835391331124, \dots 125]$ | $[-0.000167963592, \dots 591]$ | $[0.101756312961, \dots 962]$ | $2.772 \times 10^{-7}$ | $5.57 \times 10^{-46}$ |
| $50$ | $[0.314525947269601309, \dots 310]$ | $[0.011887594694449840, \dots 841]$ | $[0.011886970418532583, \dots 584]$ | $[+0.000225285994, \dots 995]$ | $[0.081300235543, \dots 544]$ | $6.243 \times 10^{-7}$ | $1.88 \times 10^{-41}$ |

**Table 1:** Outward-rounded 224-bit dyadic rational interval certificates ($\mathbb{Z}\cdot 2^{-224} \subset \mathbb{Q}$) for $\alpha_N, d_N^{\mathrm{cen}, 2}, d_N^2, B_N, A_N$ via (27), certifying $B_N \neq 0$ and $d_N^2 < d_N^{\mathrm{cen}, 2}$ across all $2 \le N \le 50$ (`src/verify_cyclotomic_cotangent_gram_and_baker_separation.py`).

Applying the same 224-bit forward-only $LDL^{\mathsf{T}}$ certification to the coordinate subspaces $V_N^{\mathrm{sqf}}$ ($\mu(k) \neq 0$), $V_N^{\mathrm{sqf}+\mathrm{pp}}$ ($\mu(k) \neq 0$ or $\omega(k) = 1$), $V_N^{(2)}$ ($k / \operatorname{rad}(k) \in \{1, 2\}$), and $V_N^{\mathrm{pp}}$ ($\omega(k) = 1$) certifies $d_N^2 < d_N^{\mathrm{sqf}, 2}$ and $d_N^{\mathrm{cen}, 2} < d_N^{\mathrm{cen}}(V_N^{\mathrm{sqf}})^2$ for **every integer $4 \le N \le 50$** (Table 2). Moreover, for the two non-nested intermediate subspaces $V_N^{\mathrm{sqf}+\mathrm{pp}}$ and $V_N^{(2)}$ ($8, 9, 16, 25, 27, 32 \in V_N^{\mathrm{sqf}+\mathrm{pp}} \setminus V_N^{(2)}$ vs. $12, 20, 28, 44 \in V_N^{(2)} \setminus V_N^{\mathrm{sqf}+\mathrm{pp}}$), we certify the exact infinite-horizon crossover: $d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2 < d_N(V_N^{(2)})^2$ for $N \in \{8, \dots, 19\} \cup \{21, 22, 27, 32\}$ (driven by prime-power spikes), whereas $d_N(V_N^{(2)})^2 < d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2$ for $N \in \{20\} \cup \{23, \dots, 26\} \cup \{28, \dots, 31\} \cup \{33, \dots, 50\}$.

| $N$ | Full $d_N^2$ ($15$-Digit Certified) | $d_N(V_N^{(2)})^2$ ($q \in \{1, 2\}$) | $d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2$ | Squarefree $d_N^{\mathrm{sqf}, 2}$ | Prime-Power $d_N(V_N^{\mathrm{pp}})^2$ | Certified $d_N^{\mathrm{sqf}, 2} - d_N^2 \ge$ | Certified $d_N(V_N^{(2)})^2 - d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2$ |
| :--- | :--- | :--- | :--- | :--- | :--- | :--- | :--- |
| $8$ | $[0.024244525306522, \dots 523]$ | $[0.024388367885244, \dots 245]$ | $[0.024244525306522, \dots 523]$ | $[0.025987294982443, \dots 444]$ | $[0.028280342004102, \dots 103]$ | $+1.743 \times 10^{-3}$ | $+1.438 \times 10^{-4}$ |
| $10$ | $[0.023852553318882, \dots 883]$ | $[0.024385699196767, \dots 768]$ | $[0.023852553318882, \dots 883]$ | $[0.025986575650608, \dots 609]$ | $[0.028230281794957, \dots 958]$ | $+2.134 \times 10^{-3}$ | $+5.331 \times 10^{-4}$ |
| $12$ | $[0.020498147332894, \dots 895]$ | $[0.020743014685690, \dots 691]$ | $[0.020499601347023, \dots 024]$ | $[0.021816083248657, \dots 658]$ | $[0.027306754062626, \dots 627]$ | $+1.318 \times 10^{-3}$ | $+2.434 \times 10^{-4}$ |
| $16$ | $[0.017936267020202, \dots 203]$ | $[0.018631097806657, \dots 658]$ | $[0.017957832257490, \dots 491]$ | $[0.019488124500996, \dots 997]$ | $[0.027150762697402, \dots 403]$ | $+1.552 \times 10^{-3}$ | $+6.733 \times 10^{-4}$ |
| $20$ | $[0.016538252109392, \dots 393]$ | $[0.016563259494923, \dots 924]$ | $[0.016909060173230, \dots 231]$ | $[0.017632818634095, \dots 096]$ | $[0.026825745622368, \dots 369]$ | $+1.095 \times 10^{-3}$ | $-3.458 \times 10^{-4}$ |
| $24$ | $[0.015577777992360, \dots 361]$ | $[0.015589162220708, \dots 709]$ | $[0.015643390016222, \dots 223]$ | $[0.016224608579430, \dots 431]$ | $[0.026813265368833, \dots 834]$ | $+6.468 \times 10^{-4}$ | $-5.423 \times 10^{-5}$ |
| $30$ | $[0.014556733576996, \dots 997]$ | $[0.014622089227942, \dots 943]$ | $[0.014681407682215, \dots 216]$ | $[0.015242969328477, \dots 478]$ | $[0.026692268959378, \dots 379]$ | $+6.862 \times 10^{-4}$ | $-5.932 \times 10^{-5}$ |
| $40$ | $[0.012735835391331, \dots 332]$ | $[0.013067106491632, \dots 633]$ | $[0.013108607868221, \dots 222]$ | $[0.013535694374946, \dots 947]$ | $[0.026669985812461, \dots 462]$ | $+7.999 \times 10^{-4}$ | $-4.150 \times 10^{-5}$ |
| $50$ | $[0.011886970418532, \dots 533]$ | $[0.011959872209388, \dots 389]$ | $[0.011980679309213, \dots 214]$ | $[0.012361695653584, \dots 585]$ | $[0.026645076287810, \dots 811]$ | $+4.747 \times 10^{-4}$ | $-2.081 \times 10^{-5}$ |

**Table 2:** 224-bit dyadic rational interval certificates comparing $d_N^2$, $d_N(V_N^{(2)})^2$, $d_N(V_N^{\mathrm{sqf}+\mathrm{pp}})^2$, $d_N^{\mathrm{sqf}, 2}$, and $d_N(V_N^{\mathrm{pp}})^2$ (`src/verify_cyclotomic_cotangent_gram_and_baker_separation.py`).

---

## 5. Prime-Power Coprime-Additivity Obstruction and Composite Schur Complement

Split $V_N = V_N^{\mathrm{pp}} \oplus V_N^{\mathrm{comp}}$ into the prime-power subspace $V_N^{\mathrm{pp}} := \operatorname{span}\{\tau_k : 2 \le k \le N,\, \omega(k) = 1\}$ and the multi-prime composite subspace $V_N^{\mathrm{comp}} := \operatorname{span}\{\tau_k : 6 \le k \le N,\, \omega(k) \ge 2\}$. By Theorem 1(c), $V_N^{\mathrm{comp}} \subset \{1\}^\perp$, while $V_N^{\mathrm{pp}}$ contains every generator with $\langle 1, \tau_k \rangle_{\mathcal{H}} \neq 0$. We now prove that $V_\infty^{\mathrm{pp}} := \bigcup_{N \ge 2} V_N^{\mathrm{pp}}$ is unconditionally bounded away from $1$ in $\mathcal{H}$.

### Theorem 4 (Prime-Power Coprime-Additivity Obstruction, Radical-Jump Rigidity Contrast, and Composite Schur Complement)

**(a) (Radical-Jump Rigidity on $V_\infty^{\mathrm{sqf}}$ vs. Coprime-Additivity Jump Law on $V_\infty^{\mathrm{pp}}$):** *For every $g \in \bigcup_{N \ge 2} V_N^{\mathrm{sqf}}$, the unit-cell jump $\Delta_m g := g(m) - g(m - 1)$ ($m \ge 1$, $g(0) = 0$) depends only on $\operatorname{rad}(m) = \prod_{p \mid m} p$:*
$$\Delta_m g \;-\; \Delta_{\operatorname{rad}(m)} g \;\equiv\; 0 \qquad (\forall\, m \ge 4 \text{ with } \mu(m) = 0). \tag{29}$$
*By contrast, for every $g = \sum_{p^\alpha \le N} c_{p^\alpha}\tau_{p^\alpha} \in V_N^{\mathrm{pp}}$ ($N \ge 2$), $\Delta_m g$ is additive across prime-power factors $p^\alpha \parallel m$ and satisfies the universal coprime-additivity jump law*
$$\Delta_m g \;-\; \sum_{p^\alpha \parallel m} \Delta_{p^\alpha} g \;+\; \bigl(\omega(m) - 1\bigr)\Delta_1 g \;\equiv\; 0 \qquad (\forall\, m \ge 6 \text{ with } \omega(m) \ge 2). \tag{30}$$
*In particular, at $m = 6 = 2\cdot 3$, every $g \in V_\infty^{\mathrm{pp}}$ satisfies $g(6) - g(5) - g(3) + 2 g(1) \equiv 0$, so $r_n := 1 - g(n)$ obeys $2 r_1 - r_3 - r_5 + r_6 \equiv 1$ and $\|1 - g\|_{\mathcal{H}}^2 \ge \frac{1}{8 + 12 + 30 + 42} = \frac{1}{92}$ for all $N \ge 2$.*

**(b) (Exact Rational All-$N$ Lower Bounds $\underline{D}_M^{\mathrm{pp}} \in \mathbb{Q}_{>0}$ & Dual Rayleigh Certificates $R_M(\boldsymbol{\lambda})$):** *For every integer $M \ge 6$ and every $g \in \bigcup_{N \ge 2} V_N^{\mathrm{pp}}$, the restriction $(g(1), \dots, g(M))^{\mathsf{T}} \in \mathbb{R}^M$ lies in the $(1 + \pi^*(M))$-dimensional subspace $\mathcal{W}_M^{\mathrm{pp}} := \operatorname{span}\{(n)_{n=1}^M, (\lfloor n / p^\beta \rfloor)_{n=1}^M : p^\beta \le M\}$ whose orthogonal complement $(\mathcal{W}_M^{\mathrm{pp}})^\perp \subset \mathbb{Q}^M$ has the explicit sparse integer basis*
$$\mathbf{w}^{(m)} \;:=\; (\mathbf{e}_m - \mathbf{e}_{m-1}) \;-\; \sum_{p^\alpha \parallel m} (\mathbf{e}_{p^\alpha} - \mathbf{e}_{p^\alpha - 1}) \;+\; \bigl(\omega(m) - 1\bigr)\mathbf{e}_1 \;\in\; \mathbb{Z}^M \qquad (6 \le m \le M,\;\; \omega(m) \ge 2). \tag{31}$$
*Let $W_M = (\mathbf{w}^{(m)})_{\omega(m) \ge 2}$, $\boldsymbol{\nu}_M := (\omega(m) - 1)_{\omega(m) \ge 2}$, and $H_M^{\mathrm{comp}} := W_M^{\mathsf{T}}\operatorname{diag}(n(n+1))_{n=1}^M W_M$. Then for all $N \ge 2$,*
$$\inf_{N \ge 2} d_N\bigl(V_N^{\mathrm{pp}}\bigr)^2 \;\ge\; \underline{D}_M^{\mathrm{pp}} \;:=\; \boldsymbol{\nu}_M^{\mathsf{T}}\bigl(H_M^{\mathrm{comp}}\bigr)^{-1}\boldsymbol{\nu}_M \;\ge\; R_M(\boldsymbol{\lambda}) \;:=\; \frac{\bigl(\sum_{n=1}^M \lambda_n\bigr)^2}{\sum_{n=1}^M n(n + 1)\lambda_n^2} \;\in\; \mathbb{Q}_{>0} \qquad \Bigl(\forall\, \boldsymbol{\lambda} \in (\mathcal{W}_M^{\mathrm{pp}})^\perp \text{ with } \sum_{n=1}^M \lambda_n \neq 0\Bigr), \tag{32}$$
*with equality $R_M(\boldsymbol{\lambda}) = \underline{D}_M^{\mathrm{pp}}$ at $\boldsymbol{\lambda}_M^* = W_M(H_M^{\mathrm{comp}})^{-1}\boldsymbol{\nu}_M$, certifying $\inf_{N \ge 2} d_N(V_N^{\mathrm{pp}})^2 \ge \underline{D}_{100}^{\mathrm{pp}} > R_{100}(\boldsymbol{\lambda}_{100}^{\mathrm{int}}) > 0.02493345 > d_{10}^2 > d_{12}^2$ (Table 3; cf. [22, Theorem 4] for the analogous finite-prefix Gram certification on $V_N^{\mathrm{pp}}$).*

**(c) (Logarithm-Free Composite Schur Complement & Orthogonal Projection Distance Gain):** *For every $N \ge 6$, the Gram matrix of $V_N^{\mathrm{comp}}$ is $G_{\mathrm{comp}} \equiv \pi\,\widetilde{\mathcal{C}}_{\mathrm{comp}} \succ 0$ (100% logarithm-free), and eliminating $\mathbf{c}_{\mathrm{comp}}^*(\mathbf{c}_{\mathrm{pp}}) = -\frac{1}{\pi}\widetilde{\mathcal{C}}_{\mathrm{comp}}^{-1} G_{\mathrm{comp}, \mathrm{pp}}\mathbf{c}_{\mathrm{pp}}$ yields*
$$d_N^2 \;\equiv\; 1 \;-\; \boldsymbol{\Lambda}_{\mathrm{pp}}^{\mathsf{T}}\bigl(G_N^{\mathrm{Schur}}\bigr)^{-1}\boldsymbol{\Lambda}_{\mathrm{pp}}, \qquad G_N^{\mathrm{Schur}} \;:=\; G_{\mathrm{pp}} \;-\; \frac{1}{\pi}\,G_{\mathrm{pp}, \mathrm{comp}}\,\widetilde{\mathcal{C}}_{\mathrm{comp}}^{-1}\,G_{\mathrm{comp}, \mathrm{pp}}, \tag{33}$$
*with exact block Schur-complement distance gain*
$$d_N\bigl(V_N^{\mathrm{pp}}\bigr)^2 - d_N^2 \;\equiv\; (\mathbf{z}^*)^{\mathsf{T}}\Bigl(\pi\,\widetilde{\mathcal{C}}_{\mathrm{comp}} \;-\; G_{\mathrm{comp}, \mathrm{pp}}\,G_{\mathrm{pp}}^{-1}\,G_{\mathrm{pp}, \mathrm{comp}}\Bigr)\mathbf{z}^* \;\equiv\; \bigl\|P_{(V_N^{\mathrm{pp}})^\perp} g_{\mathrm{comp}}^*\bigr\|_{\mathcal{H}}^2 \;<\; \|g_{\mathrm{comp}}^*\|_{\mathcal{H}}^2, \tag{34}$$
*where $\mathbf{z}^* := \mathbf{c}_{\mathrm{comp}}^* = -\frac{1}{\pi}\widetilde{\mathcal{C}}_{\mathrm{comp}}^{-1} G_{\mathrm{comp}, \mathrm{pp}}\mathbf{c}_{\mathrm{pp}}^*$ and $g_{\mathrm{comp}}^* := \sum_{\omega(k) \ge 2} c_k^*\tau_k \in V_N^{\mathrm{comp}}$ (strict inequality holds because $V_N^{\mathrm{comp}}$ and $V_N^{\mathrm{pp}}$ are not orthogonal: e.g., at $N = 6$, $d_6(V_6^{\mathrm{pp}})^2 - d_6^2 \approx 0.000110983301281 < 0.000174109942545 \approx \|g_{\mathrm{comp}}^*\|_{\mathcal{H}}^2$, and at $N = 10$, $d_{10}(V_{10}^{\mathrm{pp}})^2 - d_{10}^2 \approx 0.004377728476074 < 0.010693483121993 \approx \|g_{\mathrm{comp}}^*\|_{\mathcal{H}}^2$).*

*Proof.*
**(a)** By [14, Theorem 3(a)], $g \in V_N^{\mathrm{sqf}}$ has floor expansion $g(n) = \sum_{d=1}^N \xi_d\lfloor n/d \rfloor$ with $\xi_d = 0$ whenever $\mu(d) = 0$; hence $\Delta_m g = \sum_{d \mid m,\, \mu(d) \neq 0} \xi_d = \sum_{d \mid \operatorname{rad}(m)} \xi_d = \Delta_{\operatorname{rad}(m)} g$, proving (29). For $k = p^\alpha$ ($\alpha \ge 1$), $\tau_{p^\alpha}(n) = p^\alpha\lfloor n/p^\alpha \rfloor - p^{\alpha - 1}\lfloor n/p^{\alpha - 1} \rfloor$ has unit jump $\Delta_m\tau_{p^\alpha} = p^\alpha\mathbf{1}_{p^\alpha \mid m} - p^{\alpha - 1}\mathbf{1}_{p^{\alpha - 1} \mid m}$, so $\Delta_1\tau_p = -1$ and $\Delta_1\tau_{p^\alpha} = 0$ ($\alpha \ge 2$). Thus for every $m \ge 1$, $\Delta_m\tau_{p^\alpha} - \Delta_1\tau_{p^\alpha} = p^\alpha\mathbf{1}_{p^\alpha \mid m} - p^{\alpha - 1}\mathbf{1}_{p^{\alpha - 1} \mid m,\, \alpha \ge 2}$ vanishes whenever $p \nmid m$ and depends on $m$ solely through the exact prime power $p^{\nu_p(m)} \parallel m$. By linearity, for any $g = \sum_{p^\alpha \le N} c_{p^\alpha}\tau_{p^\alpha} \in V_N^{\mathrm{pp}}$, setting $\psi_p(\gamma) := \sum_{\alpha=1}^{\lfloor \log_p N \rfloor} c_{p^\alpha}\bigl(p^\alpha\mathbf{1}_{\alpha \le \gamma} - p^{\alpha - 1}\mathbf{1}_{2 \le \alpha \le \gamma + 1}\bigr)$ ($\psi_p(0) = 0$) gives $\Delta_m g - \Delta_1 g = \sum_{p^\gamma \parallel m} \psi_p(\gamma) = \sum_{p^\gamma \parallel m}(\Delta_{p^\gamma} g - \Delta_1 g)$, which rearranges to (30).

**(b)** On $1 \le n \le M$, $\tau_{p^\alpha}(n) = p^\alpha\lfloor n/p^\alpha \rfloor - p^{\alpha - 1}\lfloor n/p^{\alpha - 1} \rfloor \in \mathcal{W}_M^{\mathrm{pp}}$ since $\lfloor n/d \rfloor = 0$ for all $d > M$ and $\lfloor n/1 \rfloor = n$. By (30), $(\mathbf{w}^{(m)})^{\mathsf{T}}\mathbf{g} = \Delta_m g - \sum_{p^\alpha \parallel m}\Delta_{p^\alpha} g + (\omega(m) - 1)\Delta_1 g = 0$ for all $\mathbf{g} \in \mathcal{W}_M^{\mathrm{pp}}$. Since the vectors $\{\mathbf{w}^{(m)} : 6 \le m \le M,\, \omega(m) \ge 2\}$ have strictly increasing last non-zero coordinate $m$ ($w_m^{(m)} = 1$) and count $M - (1 + \pi^*(M)) = \operatorname{codim}(\mathcal{W}_M^{\mathrm{pp}})$, they form a basis of $(\mathcal{W}_M^{\mathrm{pp}})^\perp$. Since $\sum_{n=1}^M w_n^{(m)} = \omega(m) - 1$, minimizing $\sum_{n=1}^M \frac{(1 - v_n)^2}{n(n+1)}$ over $\mathbf{v} \in \mathcal{W}_M^{\mathrm{pp}}$ by Cauchy–Schwarz $(\sum_{n=1}^M \lambda_n(1 - v_n))^2 = (\sum_{n=1}^M \lambda_n)^2 \le (\sum_{n=1}^M n(n+1)\lambda_n^2)(\sum_{n=1}^M \frac{(1 - v_n)^2}{n(n+1)})$ for any $\boldsymbol{\lambda} \in (\mathcal{W}_M^{\mathrm{pp}})^\perp$ with $\sum_{n=1}^M \lambda_n \neq 0$ yields (32).

**(c)** Since $\Lambda(k) = 0$ for all $\omega(k) \ge 2$, Theorem 1(c) gives $\langle 1, \tau_k \rangle_{\mathcal{H}} = 0$ on $V_N^{\mathrm{comp}}$ and Theorem 2(b) gives $G_{\mathrm{comp}} = \pi\,\widetilde{\mathcal{C}}_{\mathrm{comp}}$. Block Gaussian elimination of $\mathbf{c}_{\mathrm{comp}}$ in $G_N\mathbf{c} = -\boldsymbol{\Lambda}_N$ yields (33). Decomposing $V_N = V_N^{\mathrm{pp}} \oplus P_{(V_N^{\mathrm{pp}})^\perp}(V_N^{\mathrm{comp}})$ orthogonally in $\mathcal{H}$ gives $P_{V_N} 1 = P_{V_N^{\mathrm{pp}}} 1 + P_{(V_N^{\mathrm{pp}})^\perp} g_{\mathrm{comp}}^*$, so $d_N(V_N^{\mathrm{pp}})^2 - d_N^2 = \|P_{V_N} 1\|_{\mathcal{H}}^2 - \|P_{V_N^{\mathrm{pp}}} 1\|_{\mathcal{H}}^2 = \|P_{(V_N^{\mathrm{pp}})^\perp} g_{\mathrm{comp}}^*\|_{\mathcal{H}}^2 = (\mathbf{z}^*)^{\mathsf{T}}(G_{\mathrm{comp}} - G_{\mathrm{comp}, \mathrm{pp}} G_{\mathrm{pp}}^{-1} G_{\mathrm{pp}, \mathrm{comp}})\mathbf{z}^*$, proving (34). $\blacksquare$

| Prefix Horizon $M$ | Multi-Prime Composites $K_{\mathrm{comp}}(M)$ | Exact Rational Bound (`fractions.Fraction`, $\le \underline{D}_M^{\mathrm{pp}} \le \inf_{N \ge 2} d_N(V_N^{\mathrm{pp}})^2$) | Decimal Value of $\underline{D}_M^{\mathrm{pp}}$ | Comparison with Full Space $d_N^2$ |
| :--- | :--- | :--- | :--- | :--- |
| $6$ | $1$ ($m = 6$) | $\underline{D}_6^{\mathrm{pp}} = \dfrac{1}{92}$ | $0.010869565217$ | $\underline{D}_6^{\mathrm{pp}} > d_{60}^2$ |
| $10$ | $2$ ($m \in \{6, 10\}$) | $\underline{D}_{10}^{\mathrm{pp}} = \dfrac{70}{5711}$ | $0.012257047802$ | $\underline{D}_{10}^{\mathrm{pp}} > d_{45}^2$ |
| $12$ | $3$ ($m \in \{6, 10, 12\}$) | $\underline{D}_{12}^{\mathrm{pp}} = \dfrac{6951}{447704}$ | $0.015525883173$ | $\underline{D}_{12}^{\mathrm{pp}} > d_{25}^2$ |
| $15$ | $5$ ($m \le 15$) | $\underline{D}_{15}^{\mathrm{pp}} = \dfrac{1392905201}{80214462024}$ | $0.017364763982$ | $\underline{D}_{15}^{\mathrm{pp}} > d_{17}^2$ |
| $20$ | $7$ ($m \le 20$) | $\underline{D}_{20}^{\mathrm{pp}} = \dfrac{5389668529927}{291197869477672}$ | $0.018508612510$ | $\underline{D}_{20}^{\mathrm{pp}} > d_{15}^2$ |
| $30$ | $13$ ($m \le 30$) | $\underline{D}_{30}^{\mathrm{pp}} = \dfrac{153740610150807271348896571}{7079360066435418305135274096}$ | $0.021716738336$ | $\underline{D}_{30}^{\mathrm{pp}} > d_{12}^2 \approx 0.020498$ |
| $50$ | $26$ ($m \le 50$) | $R_{50}(\boldsymbol{\lambda}_{50}^{\mathrm{int}}) = \dfrac{137069072386024527225}{5853825081723949160941} < \underline{D}_{50}^{\mathrm{pp}}$ ($59$-digit exact fraction; defect $5.63 \times 10^{-21}$) | $0.023415300333$ | $\underline{D}_{50}^{\mathrm{pp}} > d_{11}^2 \approx 0.020500$ |
| $100$ | $64$ ($m \le 100$) | $R_{100}(\boldsymbol{\lambda}_{100}^{\mathrm{int}}) = \dfrac{621676929251568300049}{24933450004945398760948} < \underline{D}_{100}^{\mathrm{pp}}$ ($164$-digit exact fraction; defect $5.35 \times 10^{-20}$) | $0.024933450009$ | $\underline{D}_{100}^{\mathrm{pp}} > d_{10}^2 \approx 0.023853$ |

**Table 3:** Exact rational all-$N$ lower bounds $\underline{D}_M^{\mathrm{pp}} = \boldsymbol{\nu}_M^{\mathsf{T}}(H_M^{\mathrm{comp}})^{-1}\boldsymbol{\nu}_M \le \inf_{N \ge 2} d_N(V_N^{\mathrm{pp}})^2$ (`fractions.Fraction`, zero transcendental constants) from Theorem 4(b); at $M \in \{50, 100\}$, both the compact $21$-digit dual Rayleigh certificate $R_M(\boldsymbol{\lambda}_M^{\mathrm{int}})$ ($\boldsymbol{\lambda}_M^{\mathrm{int}} = W_M\mathbf{y}_M^{\mathrm{int}}$) and the exact rational projection minimum $\underline{D}_M^{\mathrm{pp}}$ are recorded in `results/cyclotomic_cotangent_gram_and_baker_separation.json` (`src/verify_cyclotomic_cotangent_gram_and_baker_separation.py`).

---

## 6. Reproducibility and Verification Supplement

The supplementary archive `paper44_supplement.zip` executes in under $4.5$ seconds total runtime using Python $\ge 3.10$ and `numpy`. Each script supports `--output-dir <DIR>` and reproduces the pre-computed JSON artifacts in `results/`:
1. `src/verify_operator_euler_product_and_radical_rigidity.py` $\to$ `results/operator_euler_product_and_radical_rigidity.json` (`PASS` in `0.46s`): verifies Theorem 1's cross-adjoint cancellation (10), pointwise recurrence and operator Euler product (11), target inner product (12), bilinear prime-lift Gram recurrence (13)–(14), exact closed-form Gram matrix $G_{\{2, 3, 4, 6\}}$, Theorem 2's decomposition (19), Theorem 3's algebraic centered distances $d(S)^{\mathrm{cen}, 2} = 1 - \pi\alpha(S)$ and $\varphi(L) \le 2$ saturation law, and Theorem 4(a)'s Radical-Jump Rigidity Law (29).
2. `src/verify_cyclotomic_cotangent_gram_and_baker_separation.py` $\to$ `results/cyclotomic_cotangent_gram_and_baker_separation.json` (`PASS` in `3.78s`): verifies Theorem 2's probability-measure identity (17)–(18) and sign trichotomy (20)–(21) across $2 \le k \le 100$, Theorem 3's 224-bit outward-rounded dyadic rational interval certificates across all $2 \le N \le 50$ (Tables 1–2), and Theorem 4's exact rational prime-power obstruction bounds $\underline{D}_M^{\mathrm{pp}}$, dual Rayleigh certificates $R_M(\boldsymbol{\lambda}_M^{\mathrm{int}})$ (Table 3), and block Schur-complement distance-gain identity (33)–(34).

---

## 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. `@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. `@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.
16. `@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.
17. 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.
18. G. H. Hardy, J. E. Littlewood, and G. Pólya, *Inequalities*, Cambridge University Press (1934).
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; https://arxiv.org/abs/math/0306251.
21. A. Baker, *Linear forms in the logarithms of algebraic numbers. I, II, III*, Mathematika **13** (1966), 204–216; **14** (1967), 102–107, 220–228; https://doi.org/10.1112/S0025579300003971.
22. `@halasz_resonator`, *Quantitative Nyman–Beurling–Báez-Duarte Gram Matrix Theory: Exact Prime-Power Inner Products, Composite Deficiency, and Unconditional Lower Bounds*, rStation `riemann_hypothesis_3` Version 1 (2026); https://rstation.ai/problems/riemann_hypothesis/papers/3/v1.
