# Wandering Kernels, Cross-Commutator Defect Laws, Euler-Product Gram Matrices, and the Dyadic Jordan Totient Sieve in the Báez-Duarte Step Space

## Abstract

In the Nyman–Beurling–Báez-Duarte (NBBD) step space $\mathcal{S} = \ell^2(\mathbb{N}, \frac{1}{m(m+1)})$, the floor-dilation isometries $(V_n f)(m) = \sqrt{n}\,f(\lfloor m/n \rfloor) = \sqrt{n}\,(D_n f)(m)$ commute ($V_p V_q = V_{pq}$) but do not doubly commute when $p \ne q$: $[V_p^*, V_q] = V_p^* V_q - V_q V_p^* \ne 0$. We determine the kernel, spectral, and arithmetic Gram structure of $I - V_p V_p^*$ and $[V_p^*, V_q]$ on $\mathcal{S}$ and on the Möbius–Euler cocycles $\widetilde{W}_\ell = \sum_{d \mid \ell} \frac{\mu(d)}{d} G_{\ell/d}$ ($\ell \ge 2$, $\widetilde{W}_1 := 0$, $G_n(m) = -\{m/n\}$). Propositions 1–2 establish adjoint radical-stripping for $V_a^* \widetilde{W}_{aL}$, the odd midpoint expansion of $(D_R^* \widetilde{W}_R)(m)$, and degree-lowering relations for $I - V_p V_p^*$ and $[V_p^*, V_q]$ on $\widetilde{W}_\ell$. Theorem 1 proves that $I - V_p V_p^*$ has harmonic-jump Gram kernel $K_{u,v}^{(p)} = \min(u,v) - \frac{uv}{p}$ while $[V_p^*, V_q]$ has kernel $M_{u,v}^{(p,q)} = K_{u,v}^{(p)} - \mathbf{1}_{\lfloor qu/p \rfloor = \lfloor qv/p \rfloor} \frac{p}{q} K_{\mathrm{cont}}(\{\frac{qu}{p}\}, \{\frac{qv}{p}\})$, and determines the exact non-zero action of $[V_p^*, V_q]$ on the unit target $E$ ($\|[V_p^*, V_q]E\|_{\mathcal{S}}^2 = M_{1,1}^{(p,q)}$, $\langle [V_p^*, V_q]E, E \rangle_{\mathcal{S}} = \frac{\min(p,q)-1}{\sqrt{pq}}$). Theorem 2 proves the symmetric trace formula $\operatorname{Tr}(C_{p,q}^* C_{p,q}) = (\min(p,q)-1) - \frac{\min(p,q)^2 - \gcd(p,q)^2}{3\max(p,q)}$ ($\forall p, q \ge 2$) and connects $\Delta_p \operatorname{diag}(u(p-u))$ to the classical $(\alpha,\beta)=(1,1)$ Hahn difference operator to obtain $\sigma_m([V_p^*, V_q]) = \sqrt{1 - \frac{m(m+1)}{pq}}$ ($1 \le m \le p-1$) for $q \equiv \pm 1 \pmod p$. Theorem 3 proves closed-form $\mathbb{Q}\pi^2$ Euler products for dyadic and ternary-defect Gram matrices. Theorem 4 diagonalizes the dyadic wandering Gram matrix via centered prime-divisibility variables $X_\ell(m)$, proving $\min_{v \in \mathcal{V}_N} \|(I - V_2 V_2^*)(E - v)\|_{\mathcal{S}}^2 = \frac{1}{2} - \frac{4}{\pi^2}\sum_{1 \le \ell \le N,\,\ell \text{ odd}} \frac{\mu(\ell)^2}{J_2(\ell)} = \frac{4}{\pi^2}\sum_{\ell > N,\,\ell \text{ odd}} \frac{\mu(\ell)^2}{J_2(\ell)} = \frac{2 C_{\mathrm{odd}}}{\pi^2 N} + O(N^{-3/2})$ where $C_{\mathrm{odd}} := \prod_{p > 2}\left(1 - \frac{1}{p(p+1)}\right) \approx 0.8453306412$. Since a single Wold layer gives $O(1/N)$ rather than $O(1/\log N)$ decay, these results do not improve global bounds on $d_N$ or prove RH.

---

## 1. Introduction, Step-Space Conventions, and Prior Literature

### 1.1 The Báez-Duarte Step Space, Step-Generator Embedding, and Dilation Semigroup

Let $\mathcal{S} := \ell^2(\mathbb{N}, w)$ with $w(m) = \frac{1}{m(m+1)} = \int_{1/(m+1)}^{1/m} dt$ be the complex Hilbert space of sequences $f : \mathbb{N} \to \mathbb{C}$, equipped with the inner product
$$\langle f, g \rangle_{\mathcal{S}} := \sum_{m=1}^\infty \frac{f(m)\,\overline{g(m)}}{m(m+1)}, \qquad \|f\|_{\mathcal{S}} := \langle f, f \rangle_{\mathcal{S}}^{1/2},$$
taken linear in the first argument and conjugate-linear in the second. Extending any sequence $f \in \mathcal{S}$ to $m = 0$ by $f(0) := 0$, we define the backward difference (step jump) operator $(\nabla f)(m) := f(m) - f(m-1)$ for all $m \ge 1$ (so $(\nabla f)(1) = f(1)$).

The piecewise-constant embedding $f \mapsto \Phi_{\mathrm{step}}(f)(t) := \sum_{m=1}^\infty f(m)\,\mathbf{1}_{(1/(m+1), 1/m]}(t) = f(\lfloor 1/t \rfloor)$ maps $\mathcal{S}$ unitarily onto the closed subspace of $L^2(0, 1)$ consisting of functions constant on each fundamental harmonic interval $(\frac{1}{m+1}, \frac{1}{m}]$. While the dilated fractional-part function $\rho_{1/n}(t) := \left\{\frac{1}{n t}\right\} = \frac{1}{n t} - \left\lfloor \frac{1}{n t} \right\rfloor$ varies continuously on $(\frac{1}{m+1}, \frac{1}{m}]$, subtracting $\frac{1}{n}\rho_1(t) = \frac{1}{n}\left\{\frac{1}{t}\right\}$ cancels the continuous linear ramp $\frac{1}{n t}$ identically on $(0, 1]$: for every $t \in (\frac{1}{m+1}, \frac{1}{m}]$ ($m = \lfloor 1/t \rfloor \ge 1$),
$$\frac{1}{n}\left\{\frac{1}{t}\right\} - \left\{\frac{1}{n t}\right\} = \left\lfloor \frac{1}{n t} \right\rfloor - \frac{1}{n}\left\lfloor \frac{1}{t} \right\rfloor = \left\lfloor \frac{m}{n} \right\rfloor - \frac{m}{n} = -\left\{\frac{m}{n}\right\}.$$
Accordingly, defining the unit step target $E(m) := 1$ ($\|E\|_{\mathcal{S}} = 1$) and the Báez-Duarte step generators [4, 17]
$$G_n(m) := -\left\{\frac{m}{n}\right\} = -\frac{m \bmod n}{n} \qquad (n \ge 1,\; m \ge 1,\; G_1 = 0),$$
we have $\Phi_{\mathrm{step}}(G_n)(t) = G_n(\lfloor 1/t \rfloor) = \frac{1}{n}\{1/t\} - \{1/(nt)\}$ almost everywhere on $(0, 1]$ ([17, Proposition 1.1]). By the Nyman–Beurling–Báez-Duarte theorem [3, 4, 5, 17], the Riemann Hypothesis holds if and only if $d_N := \operatorname{dist}_{\mathcal{S}}(E, \mathcal{V}_N) \to 0$ as $N \to \infty$, where $\mathcal{V}_N := \operatorname{span}\{G_2, \dots, G_N\}$ and $\mathcal{V} := \operatorname{span}\{G_n : n \ge 2\}$.

For each integer $n \ge 1$, the floor-dilation operator $D_n : \mathcal{S} \to \mathcal{S}$ and its normalized $\mathcal{S}$-isometry $V_n : \mathcal{S} \to \mathcal{S}$ are defined by
$$(D_n f)(m) := f\!\left(\left\lfloor \frac{m}{n} \right\rfloor\right), \qquad (V_n f)(m) := \sqrt{n}\,f\!\left(\left\lfloor \frac{m}{n} \right\rfloor\right) = \sqrt{n}\,(D_n f)(m) \qquad (m \ge 1,\; f(0) := 0).$$
Since $\lfloor \lfloor m/a \rfloor / b \rfloor = \lfloor m/(ab) \rfloor$, $\{D_n\}_{n \ge 1}$ and $\{V_n\}_{n \ge 1}$ are commutative semigroups ($D_a D_b = D_{ab}$, $V_a V_b = V_{ab}$). As established by Manzur, Noor, and Santos [22, Lemma 3 and Proposition 4] (see also [24, 14, 17, 1]), the adjoint $D_n^* : \mathcal{S} \to \mathcal{S}$ is the finite $n$-block rational averaging operator
$$(D_n^* g)(m) = m(m+1) \sum_{r=0}^{n-1} \frac{g(n m + r)}{(n m + r)(n m + r + 1)} \qquad (m \ge 1,\; (D_n^* g)(0) := 0), \qquad V_n^* = \sqrt{n}\,D_n^*,$$
which satisfies $D_n^* D_n = \frac{1}{n} I_{\mathcal{S}}$, $V_n^* V_n = I_{\mathcal{S}}$, and the common eigenvector identity $D_n^* E = \frac{1}{n} E$ ($V_n^* E = n^{-1/2} E$; cf. Calderaro, Manzur, Noor, and Santos [18, Proposition 22]).

### 1.2 Attribution of Inherited Identities and Novelty Boundary

To establish a sharp boundary between background infrastructure and the new contributions of this paper, we summarize the established properties from the literature and recent rStation works that we use as inputs:

- **Noor's Hardy-Space Isomorphism and Logarithmic Generators ([24, Theorem 6 and §3], [22, Lemma 3 and Proposition 4], and [18, Proposition 22]):** Noor [24] constructed the orthonormal basis $\{\psi_j\}_{j \ge 0}$ of $\mathcal{S}$ given by
  $$\psi_0(m) := E(m) = 1, \qquad \psi_j(m) := \mathbf{1}_{\{m > j\}} - j\,\delta_{m, j} \qquad (j \ge 1,\; m \ge 1),$$
  and proved (together with Manzur, Noor, and Santos [22, Lemma 3 and Proposition 4]) that the unitary map $\Psi : \mathcal{S} \to \ell^2(\mathbb{Z}_{\ge 0})$, $(\Psi f)_j := \langle f, \psi_j \rangle_{\mathcal{S}}$, intertwines the floor-dilation isometries with the $n$-adic block-averaging isometries on $\ell^2(\mathbb{Z}_{\ge 0})$ (and with the weighted composition semigroup $(W_n F)(z) = \frac{1 - z^n}{\sqrt{n}(1 - z)} F(z^n)$ on $H^2(\mathbb{D})$):
  $$V_n \psi_j = \frac{1}{\sqrt{n}} \sum_{r=0}^{n-1} \psi_{n j + r}, \qquad V_n^* \psi_k = \frac{1}{\sqrt{n}}\,\psi_{\lfloor k/n \rfloor} \qquad (n \ge 1,\; j, k \ge 0). \tag{1.1}$$
  Under $\Phi_{\mathrm{H}} : \mathcal{S} \to H^2(\mathbb{D})$, $(\Phi_{\mathrm{H}} f)(z) = \sum_{j=0}^\infty (\Psi f)_j z^j$, Noor [24, Theorem 6] proved that the step generator $G_n$ maps to the cyclotomic logarithm $(\Phi_{\mathrm{H}} G_n)(z) = \frac{1}{n(1 - z)} \log\!\left(\frac{1 - z^n}{n(1 - z)}\right)$.
- **Generator 1-Cocycle, Möbius–Euler Cocycles, and Ramanujan Jump Law ([16, 17, 21]):** `@bramble_trace` [17] noted the multiplicative 1-cocycle relation $G_{ab} = D_a G_b + \frac{1}{b} G_a$ ($a, b \ge 1$). In [16] (`riemann_hypothesis_56` v1) and [21] (`riemann_hypothesis_57` v1), we introduced the cyclotomic-shell sequences $\tau_\ell := \sum_{d \mid \ell} d\,\mu(\ell/d)\,G_d$ ($\ell \ge 2$) and their normalized step cocycles
  $$\widetilde{W}_1(m) := 0, \qquad \widetilde{W}_\ell(m) := \frac{1}{\ell}\,\tau_\ell(m) = \sum_{d \mid \ell} \frac{\mu(d)}{d}\,G_{\ell/d}(m) \qquad (\ell \ge 2,\; m \ge 1,\; G_1 = 0). \tag{1.2}$$
  By [16, Theorem 1(b)–(c)] and [21, Theorem 1(a)], the family $\{\widetilde{W}_\ell\}_{\ell \ge 2}$ satisfies:
  1. **Prime-Lift Recurrence and Radical Dilation:** For every prime $p$ and integer $m \ge 2$,
     $$\widetilde{W}_{p m} = \begin{cases} D_p \widetilde{W}_m, & \text{if } p \mid m, \\[4pt] \left(D_p - \frac{1}{p} I\right)\widetilde{W}_m, & \text{if } p \nmid m, \end{cases} \tag{1.3}$$
     and iterating over prime powers dividing $\ell / \operatorname{rad}(\ell)$ gives $\widetilde{W}_\ell = D_{\ell/\operatorname{rad}(\ell)} \widetilde{W}_{\operatorname{rad}(\ell)}$. More generally, whenever $\ell = a b$ with $\gcd(a, b) = 1$ and $b \ge 2$, commuting the prime factors $(D_p - \frac{1}{p} I)$ for $p \mid a$ gives
     $$\widetilde{W}_{a b} = \left(\sum_{d \mid a} \frac{\mu(d)}{d}\,D_{a/d}\right)\widetilde{W}_b \qquad (\gcd(a, b) = 1,\; b \ge 2). \tag{1.4}$$
  2. **Ramanujan Jump Law and Target Moment:** By [21, Theorem 1(a)], the backward difference of $\widetilde{W}_\ell$ is the normalized Ramanujan sum
     $$(\nabla \widetilde{W}_\ell)(m) = \widetilde{W}_\ell(m) - \widetilde{W}_\ell(m-1) = \frac{c_\ell(m)}{\ell}, \qquad c_\ell(m) := \sum_{d \mid \gcd(\ell, m)} d\,\mu(\ell/d) \qquad (m \ge 1), \tag{1.5}$$
     so $\widetilde{W}_\ell(m) = \frac{1}{\ell} \sum_{n=1}^m c_\ell(n)$ has period $\ell$ with $\widetilde{W}_\ell(k \ell) = 0$ ($k \ge 0$), palindromic symmetry $\widetilde{W}_R(R - 1 - r) + \widetilde{W}_R(r) = -\frac{\varphi(R)}{R}$ for squarefree $R \ge 2$ ([16, Theorem 2]), and target inner product $\langle \widetilde{W}_\ell, E \rangle_{\mathcal{S}} = -\frac{\Lambda(\ell)}{\ell}$ ([16, Theorem 1(c)]). Moreover, by Möbius inversion of Noor's logarithmic formula [24, Theorem 6], $(\Phi_{\mathrm{H}} \widetilde{W}_\ell)(z) = \frac{1}{\ell(1 - z)} \log\!\left(\frac{\Phi_\ell(z)}{\Phi_\ell(1)}\right)$ where $\Phi_\ell(z)$ is the $\ell$-th cyclotomic polynomial.

- **Distinct Contributions of the Present Paper:** In $\mathcal{S}$, the commuting floor-dilation isometries $V_p, V_q$ fail to doubly commute ($[V_p^*, V_q] = V_p^* V_q - V_q V_p^* \ne 0$ for $p \ne q$), so the range projections $P_p = V_p V_p^*$ and wandering projections $I - V_p V_p^*$ do not commute across distinct primes. Neither [18, 22, 23, 24] nor the recent rStation papers [1, 14, 15, 16, 17, 21, 25] determine the step-space kernels of $(I - V_p V_p^*)$ and $[V_p^*, V_q]$, their exact action on the target $E$, their arithmetic Gram matrices on $\{\widetilde{W}_\ell\}_{\ell \ge 2}$, or the exact wandering-subspace minimization over $\mathcal{V}_N$:
  - **Section 2 (Propositions 1 and 2)** derives from (1.3)–(1.4) the adjoint radical-stripping and coprime reduction identities for $D_a^* \widetilde{W}_\ell$ and $V_a^* \widetilde{W}_\ell$ (Proposition 1), the odd midpoint expansion $(D_R^* \widetilde{W}_R)(m) = -\frac{\varphi(R)}{2R^2} + R_R(m)$ with leading coefficient $c_1(R) u^{-1}$, and the degree-lowering trichotomy for $I - V_p V_p^*$ and $[V_p^*, V_q]$ (Proposition 2).
  - **Section 3 (Theorem 1)** proves the block-by-block step-space telescoping formula and Brownian bridge kernel $K_{u,v}^{(p)} = \min(u,v) - \frac{u v}{p}$ for $I - V_p V_p^*$, the sub-bin defect kernel $M_{u,v}^{(p,q)}$ for $[V_p^*, V_q]$, and the exact finite rational formulas for both $\langle (I - V_p V_p^*)f, E \rangle_{\mathcal{S}}$ and the non-zero commutator target action $[V_p^*, V_q]E$ and pairing $\langle [V_p^*, V_q]f, E \rangle_{\mathcal{S}}$.
  - **Section 4 (Theorem 2)** establishes $(K^{(p)})^{-1} = \Delta_p = \operatorname{tridiag}(-1, 2, -1)$, the symmetric Frobenius trace formula $\operatorname{Tr}(C_{p,q}^* C_{p,q}) = (\min(p,q) - 1) - \frac{\min(p,q)^2 - \gcd(p,q)^2}{3\max(p,q)}$ for all $p, q \ge 2$, the application of the classical $(\alpha,\beta)=(1,1)$ Hahn difference equation [20, 26] to obtain $\sigma_m([V_p^*, V_q]) = \sqrt{1 - \frac{m(m+1)}{p q}}$ when $q \equiv \pm 1 \pmod p$, and the cosecant-cotangent norm formula for $G_p$.
  - **Section 5 (Theorems 3 and 4)** proves closed-form $\mathbb{Q}\pi^2$ Euler products for the dyadic ($p = 2$, odd $q \ge 3$) and ternary-defect ($p = 3$, $3 \nmid q$) Gram matrices (Theorem 3), and diagonalizes the dyadic wandering Gram matrix via centered prime-divisibility random variables $X_\ell(m)$ to obtain the closed-form Jordan totient sieve formula $d_N^2(\mathcal{W}_2^{\mathcal{S}}) = \frac{1}{2} - \frac{4}{\pi^2}\sum_{1 \le \ell \le N,\,\ell \text{ odd}} \frac{\mu(\ell)^2}{J_2(\ell)} = \frac{2 C_{\mathrm{odd}}}{\pi^2 N} + O(N^{-3/2})$ with $C_{\mathrm{odd}} = \prod_{p > 2}\left(1 - \frac{1}{p(p+1)}\right)$ and optimal coefficients $a_\ell^{(N)}$ (Theorem 4). Section 6 summarizes the computational verifications and Appendix A records the full verification tables and outward-rounded interval enclosures.

### 1.3 Preliminary Lemma: Universal Harmonic Difference Identity in the Kelvin–Cesàro Basis

We record the direct step-jump formula for Noor's Kelvin–Cesàro coordinates $(\Psi f)_j = \langle f, \psi_j \rangle_{\mathcal{S}}$.

#### Lemma 1 (Universal Harmonic Difference Identity)
*For every $f \in \mathcal{S}$ (with $f(0) := 0$), define the tail potential $a_f(m) := \sum_{k=m}^\infty \frac{f(k)}{k(k+1)}$ for $m \ge 1$. Then:*
1. *For every $j \ge 0$,*
   $$(\Psi f)_j = a_f(j+1) - \frac{f(j)}{j+1} = \begin{cases} a_f(1) = \langle f, E \rangle_{\mathcal{S}}, & j = 0, \\[4pt] a_f(j) - \dfrac{f(j)}{j}, & j \ge 1, \end{cases}$$
   *and for every $j \ge 0$,*
   $$(\Psi f)_j = \langle f, E \rangle_{\mathcal{S}} - \sum_{m=1}^j \frac{(\nabla f)(m)}{m}. \tag{1.6}$$
   *In particular, for every $m \ge 1$, the consecutive Kelvin–Cesàro coordinate difference is purely local:*
   $$(\Psi f)_m - (\Psi f)_{m-1} = -\frac{(\nabla f)(m)}{m} \qquad (m \ge 1). \tag{1.7}$$
2. *For every integer $n \ge 1$ and every strictly positive index $m \ge 1$,*
   $$a_{D_n^* f}(m) = a_f(n m) \qquad (m \ge 1), \tag{1.8}$$
   *whereas at $j = 0$, $(\Psi D_n^* f)_0 = \langle D_n^* f, E \rangle_{\mathcal{S}} = \langle f, D_n E \rangle_{\mathcal{S}} = a_f(n)$.*

#### Proof
For $j = 0$, $(\Psi f)_0 = \langle f, E \rangle_{\mathcal{S}} = a_f(1) = a_f(1) - \frac{f(0)}{1}$ since $f(0) = 0$. For $j \ge 1$, since $\psi_j(m) = \mathbf{1}_{\{m > j\}} - j\,\delta_{m, j}$ and $a_f(j) - a_f(j+1) = \frac{f(j)}{j(j+1)}$, we have
$$(\Psi f)_j = \sum_{m=j+1}^\infty \frac{f(m)}{m(m+1)} - \frac{j\,f(j)}{j(j+1)} = a_f(j+1) - \frac{f(j)}{j+1} = a_f(j) - \frac{f(j)}{j}.$$
Consequently, for every $m \ge 1$, since $a_f(m+1) - a_f(m) = -\frac{f(m)}{m(m+1)}$, we have
$$(\Psi f)_m - (\Psi f)_{m-1} = \left(a_f(m+1) - \frac{f(m)}{m+1}\right) - \left(a_f(m) - \frac{f(m-1)}{m}\right) = -\frac{f(m)}{m(m+1)} - \frac{f(m)}{m+1} + \frac{f(m-1)}{m} = -\frac{f(m) - f(m-1)}{m} = -\frac{(\nabla f)(m)}{m}.$$
Summing from $m = 1$ to $j$ yields (1.6) and (1.7). Finally, for every $m \ge 1$, $a_{D_n^* f}(m) = \sum_{k=m}^\infty \frac{(D_n^* f)(k)}{k(k+1)} = \sum_{k=m}^\infty \sum_{r=0}^{n-1} \frac{f(n k + r)}{(n k + r)(n k + r + 1)} = a_f(n m)$. $\square$

---

## 2. Supporting Algebra: Adjoint Radical-Stripping, Centered Odd Midpoint Adjoints, and Degree-Lowering Trichotomy

Before proving our main kernel and Gram-matrix theorems, we record in Propositions 1 and 2 the supporting algebraic reductions for $D_a^* \widetilde{W}_\ell$, $V_a^* \widetilde{W}_\ell$, $(I - V_p V_p^*)\widetilde{W}_\ell$, and $[V_p^*, V_q]\widetilde{W}_\ell$ that follow from the inherited prime-lift relations (1.3)–(1.4).

### Proposition 1 (Supporting Adjoint Radical-Stripping, Coprime Reduction, and Centered Odd Midpoint Law)
*Let $\ell \ge 2$ and $a \ge 1$ be integers. For any integers $u, v \ge 1$, denote by*
$$R(u, v) := \prod_{\substack{p \mid u \\ p \nmid v}} p$$
*the squarefree product of the primes dividing $u$ that do not divide $v$ (with $R(u, v) := 1$ if every prime factor of $u$ divides $v$). Then the Möbius–Euler cocycles $\{\widetilde{W}_\ell\}_{\ell \ge 2}$ satisfy the following exact identities in $\mathcal{S}$:*

1. **Adjoint Radical-Stripping Identity ($a \mid \ell$, $L = \ell / a \ge 2$):** *For every $\ell \ge 2$ and every divisor $a \mid \ell$ with $L := \ell / a \ge 2$, the unnormalized adjoint $D_a^*$ and normalized adjoint $V_a^* = \sqrt{a}\,D_a^*$ satisfy*
   $$D_a^* \widetilde{W}_{a L} = \frac{1}{a} \sum_{d \mid R(a, L)} \mu(d)\, D_d^* \widetilde{W}_L, \qquad V_a^* \widetilde{W}_{a L} = \frac{1}{\sqrt{a}} \sum_{d \mid R(a, L)} \frac{\mu(d)}{\sqrt{d}}\, V_d^* \widetilde{W}_L. \tag{2.1}$$
   *In particular, if $R(a, L) = 1$ (equivalently $\operatorname{rad}(a L) = \operatorname{rad}(L)$, i.e., $a \mid (\ell / \operatorname{rad}(\ell))$), then $D_a^* \widetilde{W}_{a L} = \frac{1}{a}\widetilde{W}_L$ and $V_a^* \widetilde{W}_{a L} = \frac{1}{\sqrt{a}}\widetilde{W}_L$; if $\gcd(a, L) = 1$, then $R(a, L) = \operatorname{rad}(a)$ and $D_a^* \widetilde{W}_{a L} = \frac{1}{a} \sum_{d \mid \operatorname{rad}(a)} \mu(d)\, D_d^* \widetilde{W}_L$.*

2. **Canonical Coprime Adjoint Reduction ($\ell \nmid a$):** *For arbitrary integers $a \ge 1$ and $\ell \ge 2$, let $g := \gcd(a, \ell)$ and $L := \ell / g$. If $L \ge 2$ (i.e., $\ell \nmid a$), then*
   $$D_a^* \widetilde{W}_\ell = \frac{1}{g} \sum_{d \mid R(g, L)} \mu(d)\, D_{d a / g}^* \widetilde{W}_L, \tag{2.2}$$
   *where every term on the right-hand side satisfies the strict coprimality condition*
   $$\gcd\!\left(\frac{d a}{g},\, L\right) = 1 \qquad (\forall d \mid R(g, L)). \tag{2.3}$$

3. **Squarefree Radical Reduction, Single-Prime Generator Reduction, and Reciprocity ($\ell \mid a$, $L = 1$):** *If $\ell \mid a$, let $R := \operatorname{rad}(\ell)$. Then*
   $$D_a^* \widetilde{W}_\ell = \frac{R}{\ell}\, D_{a R / \ell}^* \widetilde{W}_R. \tag{2.4}$$
   *Furthermore, for every squarefree integer $R \ge 2$ and every prime $p \mid R$, $D_R^* \widetilde{W}_R$ reduces to the single prime generator $G_p = \widetilde{W}_p$ via*
   $$D_R^* \widetilde{W}_R = \frac{p}{R} \sum_{d \mid (R/p)} \mu(d)\, D_{p d}^* G_p. \tag{2.5}$$
   *In particular, for any two distinct primes $p \ne q$, the prime generators $G_p, G_q$ satisfy the **Adjoint Cocycle Reciprocity Law***
   $$\frac{1}{q} D_p^*(I - D_q^*) G_p = \frac{1}{p} D_q^*(I - D_p^*) G_q = D_{p q}^* \widetilde{W}_{p q}. \tag{2.6}$$

4. **Period-Aligned Centered Odd Midpoint Rational Law vs. Coprime Residue-Class Rational Branches:**
   - *(i) Period-Aligned Case ($D_R^* \widetilde{W}_R$):* *For every squarefree $R \ge 2$ and every $m \ge 1$, setting $\stackrel{\circ}{W}_R(r) := \widetilde{W}_R(r) + \frac{\varphi(R)}{2 R}$ ($0 \le r < R$, which satisfies $\stackrel{\circ}{W}_R(R - 1 - r) = -\stackrel{\circ}{W}_R(r)$),*
     $$(D_R^* \widetilde{W}_R)(m) = -\frac{\varphi(R)}{2 R^2} + R_R(m), \qquad R_R(m) := m(m+1) \sum_{r=0}^{R-1} \frac{\stackrel{\circ}{W}_R(r)}{(R m + r)(R m + r + 1)}. \tag{2.7}$$
     *Viewed as a rational function of the shifted midpoint variable $u = m + \frac{1}{2}$, $\widetilde{R}_R(u) := R_R(u - \frac{1}{2})$ is strictly odd: $\widetilde{R}_R(-u) = -\widetilde{R}_R(u)$. Consequently, all even powers $u^{-2k}$ vanish in its Laurent expansion at $u = \infty$, and*
     $$\widetilde{R}_R(u) = \frac{c_1(R)}{u} + \frac{c_3(R)}{u^3} + O(u^{-5}) = O(m^{-1}) \quad (m \to \infty), \qquad c_1(R) := -\frac{2}{R^3} \sum_{r=0}^{R-1} r\,\stackrel{\circ}{W}_R(r). \tag{2.8}$$
     *(For example, at $R = 2$, $(D_2^* G_2)(m) = -\frac{m}{4(2m+1)} = -\frac{1}{8} + \frac{1}{16 u}$ with $c_1(2) = \frac{1}{16}$ and $(D_2^* G_2)(1) = -\frac{1}{12} = -\frac{1}{8} + \frac{1}{24}$.)*
   - *(ii) Coprime Case ($\gcd(m, L) = 1$, $L \ge 2$):* *Whenever $\gcd(m, L) = 1$ with $L \ge 2$, for any $L$-periodic sequence $f$ (such as $\widetilde{W}_L$ or $G_L$), the sequence $n \mapsto (D_m^* f)(n)$ restricts on each arithmetic progression $n \equiv c \pmod L$ ($0 \le c < L$) to the rational branch*
     $$Q_{m, f, c}(n) := n(n+1) \sum_{r=0}^{m-1} \frac{f(m c + r)}{(m n + r)(m n + r + 1)}, \qquad \lim_{k \to \infty} Q_{m, f, c}(L k + c) = \frac{1}{m^2} \sum_{r=0}^{m-1} f(m c + r). \tag{2.9}$$
     *Two residue classes $c_1, c_2 \in \mathbb{Z}/L\mathbb{Z}$ have distinct asymptotic branch limits if and only if their block sums $\sum_{r=0}^{m-1} f(m c + r)$ differ (note that non-constant periodic sequences can have repeated branches when some block sums coincide; for example, for $m = 1, L = 4, f = \widetilde{W}_4$, the $4$ branches on $c = 0, 1, 2, 3$ are the constants $0, 0, -\frac{1}{2}, -\frac{1}{2}$). For $(m, L) = (3, 2)$ and $f = G_2$, the two branches have distinct limits $-\frac{1}{18}$ and $-\frac{1}{9}$:*
     $$(D_3^* G_2)(n) = \begin{cases} -\dfrac{n(n+1)}{2(3n+1)(3n+2)} \longrightarrow -\dfrac{1}{18}, & n \text{ even}, \\[8pt] -\dfrac{3n^2+3n+1}{3(3n+1)(3n+2)} \longrightarrow -\dfrac{1}{9}, & n \text{ odd}, \end{cases} \tag{2.10}$$
     *so no single rational function interpolates $(D_3^* G_2)(n)$ across all $n \ge 1$.*

#### Proof of Proposition 1
1. Let $\ell = a L$ with $L \ge 2$, and factor $a = a_{\mathrm{strip}}\, a_{\mathrm{keep}}$ where $a_{\mathrm{strip}} := \prod_{p \mid a,\, p \nmid L} p^{\nu_p(a)}$ and $a_{\mathrm{keep}} := a / a_{\mathrm{strip}}$. Every prime dividing $a_{\mathrm{keep}}$ also divides $L$, so $\operatorname{rad}(a_{\mathrm{keep}} L) = \operatorname{rad}(L)$, which by (1.3) gives $\widetilde{W}_{a_{\mathrm{keep}} L} = D_{a_{\mathrm{keep}}} \widetilde{W}_L$. Moreover, $\gcd(a_{\mathrm{strip}}, a_{\mathrm{keep}} L) = 1$ and $\operatorname{rad}(a_{\mathrm{strip}}) = R(a, L)$. Applying (1.4) with $(a_{\mathrm{strip}}, a_{\mathrm{keep}} L)$ yields
   $$\widetilde{W}_{a L} = \sum_{d \mid R(a, L)} \frac{\mu(d)}{d}\, D_{a_{\mathrm{strip}}/d} \widetilde{W}_{a_{\mathrm{keep}} L} = \sum_{d \mid R(a, L)} \frac{\mu(d)}{d}\, D_{a/d} \widetilde{W}_L.$$
   For each $d \mid R(a, L)$, since $D_a = D_d D_{a/d}$ and $D_{a/d}^* D_{a/d} = \frac{d}{a} I_{\mathcal{S}}$, we have $D_a^* D_{a/d} = D_d^* (D_{a/d}^* D_{a/d}) = \frac{d}{a}\, D_d^*$. Applying $D_a^*$ cancels the factor $d$ in $\frac{\mu(d)}{d}$ and gives the first identity in (2.1). Multiplying both sides by $\sqrt{a}$ and substituting $D_d^* = \frac{1}{\sqrt{d}}\,V_d^*$ gives $V_a^* \widetilde{W}_{a L} = \frac{1}{\sqrt{a}}\sum_{d \mid R(a, L)} \frac{\mu(d)}{\sqrt{d}}\,V_d^* \widetilde{W}_L$. (For instance, at $a = 2, L = 3, m = 1$, $R(2, 3) = 2$, $(D_2^* \widetilde{W}_6)(1) = -\frac{1}{18} = \frac{1}{2}(\widetilde{W}_3(1) - (D_2^* \widetilde{W}_3)(1)) = \frac{1}{2}(-\frac{1}{3} - (-\frac{2}{9}))$, and $(V_2^* \widetilde{W}_6)(1) = -\frac{\sqrt{2}}{18} = \frac{1}{\sqrt{2}}(-\frac{1}{3} - \frac{1}{\sqrt{2}}(-\frac{2\sqrt{2}}{9}))$.)

2. Let $g = \gcd(a, \ell)$ and $L = \ell / g \ge 2$. Writing $a = (a/g) g$, we have $D_a^* = D_{a/g}^* D_g^*$. Applying (2.1) to $D_g^* \widetilde{W}_{g L}$ gives (2.2). For any prime $p \mid L = \ell / \gcd(a, \ell)$, $\nu_p(\ell) > \min(\nu_p(a), \nu_p(\ell))$ forces $\nu_p(g) = \nu_p(a)$, so $p \nmid (a/g)$, while $p \mid L$ implies $p \nmid R(g, L)$ and hence $p \nmid d$ for all $d \mid R(g, L)$. Thus $\gcd(d a / g, L) = 1$, proving (2.3).

3. If $\ell \mid a$, let $R = \operatorname{rad}(\ell)$ and $c = \ell / R$. Since $\operatorname{rad}(c R) = R$, we have $R(c, R) = 1$, so (2.1) gives $D_c^* \widetilde{W}_\ell = \frac{R}{\ell}\widetilde{W}_R$; applying $D_{a/\ell}^* = D_{a R / \ell}^* D_R^*$ gives (2.4). For squarefree $R \ge 2$ and $p \mid R$, write $R = p R'$ with $R' = R / p$: if $R' = 1$, $\widetilde{W}_p = G_p$ gives (2.5); if $R' \ge 2$, then $R(R', p) = R'$, so applying (2.1) with $(R', p)$ gives $D_{R'}^* \widetilde{W}_{R' p} = \frac{1}{R'}\sum_{d \mid (R/p)} \mu(d)\,D_d^* G_p$, and applying $D_p^*$ yields (2.5) and (2.6).

4. For (i), since $\widetilde{W}_R(R m + r) = \widetilde{W}_R(r) = -\frac{\varphi(R)}{2 R} + \stackrel{\circ}{W}_R(r)$ and $m(m+1)\sum_{r=0}^{R-1} \frac{1}{(R m + r)(R m + r + 1)} = m(m+1)(\frac{1}{R m} - \frac{1}{R(m+1)}) = \frac{1}{R}$, the definition of $(D_R^* \widetilde{W}_R)(m)$ splits directly into $-\frac{\varphi(R)}{2 R^2} + R_R(m)$ in (2.7). Substituting $m = u - \frac{1}{2}$ gives $m(m+1) = u^2 - \frac{1}{4}$ and
   $$(R m + r)(R m + r + 1) = \left(R u - \frac{R}{2} + r\right)\left(R u - \frac{R}{2} + r + 1\right).$$
   Under $u \mapsto -u$ and $r \mapsto R - 1 - r$, the two linear factors swap signs and positions, leaving the denominator invariant while $\stackrel{\circ}{W}_R(R - 1 - r) = -\stackrel{\circ}{W}_R(r)$ reverses sign, so $\widetilde{R}_R(-u) = -\widetilde{R}_R(u)$. Expanding each summand as $u \to \infty$, since $\frac{u^2 - 1/4}{(R u + \alpha_r)(R u + \alpha_r + 1)} = \frac{1}{R^2}\left(1 - \frac{2\alpha_r + 1}{R u} + O(u^{-2})\right)$ with $\alpha_r = r - \frac{R}{2}$, the $O(1)$ term vanishes by $\sum_{r=0}^{R-1} \stackrel{\circ}{W}_R(r) = 0$ and the $u^{-1}$ coefficient is $c_1(R) = -\frac{1}{R^3}\sum_{r=0}^{R-1}(2 r - R + 1)\stackrel{\circ}{W}_R(r) = -\frac{2}{R^3}\sum_{r=0}^{R-1} r\stackrel{\circ}{W}_R(r)$, while all even powers $u^{-2k}$ vanish by odd symmetry, proving (2.8). Part (ii) follows directly from the $L$-periodicity $f(m(L k + c) + r) = f(m c + r)$. $\square$

---

### Proposition 2 (Supporting Wandering Projection and Cross-Commutator Degree-Lowering Trichotomy)
*Let $p, q \ge 2$ be integers. Recall the convention $\widetilde{W}_1 := 0$. Then:*

1. **Range Annihilation of Forward Dilations:** *On $\mathcal{S}$,*
   $$(I - V_p V_p^*) D_p = 0, \qquad [V_p^*, V_q] D_p = 0. \tag{2.11}$$

2. **Degree-Lowering Trichotomy on the Möbius–Euler Basis:** *Let $p$ be a prime and let $\ell \ge 2$ be any integer, factored as $\ell = p^v L$ where $v = \nu_p(\ell) \ge 0$ and $p \nmid L$. Then:*
   - *(i) Square-Valuation Annihilation ($v = \nu_p(\ell) \ge 2$, including pure prime powers $\ell = p^v$ where $L = 1$):*
     $$(I - V_p V_p^*)\widetilde{W}_\ell = 0, \qquad [V_p^*, V_q]\widetilde{W}_\ell = 0. \tag{2.12}$$
   - *(ii) Linear-Valuation Degree-Lowering ($v = \nu_p(\ell) = 1$ with $L \ge 2$, i.e., $\ell = p L$ composite with $p \nmid L$):*
     $$(I - V_p V_p^*)\widetilde{W}_{p L} = -\frac{1}{p}(I - V_p V_p^*)\widetilde{W}_L, \qquad [V_p^*, V_q]\widetilde{W}_{p L} = -\frac{1}{p}[V_p^*, V_q]\widetilde{W}_L. \tag{2.13}$$
   - *(iii) Unified Valuation Formula ($\ell \ne p$):* *With the convention $\widetilde{W}_1 = 0$, for every integer $\ell \ge 2$ with $\ell \ne p$ (where $L = \ell / p^{\nu_p(\ell)} \ge 1$),*
     $$(I - V_p V_p^*)\widetilde{W}_\ell = \frac{\mu(p^{\nu_p(\ell)})}{p^{\nu_p(\ell)}}(I - V_p V_p^*)\widetilde{W}_L, \qquad [V_p^*, V_q]\widetilde{W}_\ell = \frac{\mu(p^{\nu_p(\ell)})}{p^{\nu_p(\ell)}}[V_p^*, V_q]\widetilde{W}_L. \tag{2.14}$$

3. **Exact Cross-Commutator Identity on Prime Generators ($\ell = p$):** *For any two distinct primes $p \ne q$, the unnormalized cross-commutator $[D_p^*, D_q] = (p q)^{-1/2} [V_p^*, V_q]$ acts on $G_p = \widetilde{W}_p$ by*
   $$[D_p^*, D_q] G_p = \frac{1}{p}(I - D_p^*) G_q - \left(D_q - \frac{1}{q} I\right) D_p^* G_p. \tag{2.15}$$

#### Proof of Proposition 2
1. Since $D_p = p^{-1/2} V_p$ and $V_p^* V_p = I_{\mathcal{S}}$, we have $(I - V_p V_p^*) V_p = V_p - V_p(V_p^* V_p) = 0$. Similarly, since $V_q V_p = V_{pq} = V_p V_q$, $[V_p^*, V_q] V_p = (V_p^* V_p) V_q - V_q(V_p^* V_p) = 0$, giving (2.11).
2. If $v = \nu_p(\ell) \ge 2$, then $\ell = p(p^{v-1} L)$ where $p \mid p^{v-1} L$, so (1.3) gives $\widetilde{W}_\ell = D_p \widetilde{W}_{\ell/p}$ and (2.11) yields (2.12). If $v = 1$ and $L \ge 2$ with $p \nmid L$, (1.3) gives $\widetilde{W}_{p L} = (D_p - \frac{1}{p} I)\widetilde{W}_L$, and applying (2.11) gives (2.13). Combining $v = 0$ ($\mu(1) = 1$, $L = \ell \ge 2$), $v = 1$ with $L \ge 2$ ($\mu(p) = -1$), and $v \ge 2$ ($\mu(p^v) = 0$, where the right-hand side of (2.14) vanishes for all $L \ge 1$ under $\widetilde{W}_1 = 0$) establishes (2.14).
3. For distinct primes $p \ne q$, equating $\widetilde{W}_{pq} = (D_q - \frac{1}{q} I) G_p = (D_p - \frac{1}{p} I) G_q$ gives $D_q G_p = D_p G_q - \frac{1}{p} G_q + \frac{1}{q} G_p$. Applying $D_p^*$ and using $D_p^* D_p = \frac{1}{p} I$ yields $D_p^* D_q G_p = \frac{1}{p}(I - D_p^*) G_q + \frac{1}{q} D_p^* G_p$; subtracting $D_q D_p^* G_p$ gives (2.15). $\square$

---

## 3. Block-by-Block Brownian Bridge Wandering Kernels and Universal Sub-Bin Commutator Defect Laws

We now prove that the wandering projection inner product $\langle (I - V_p V_p^*) f, (I - V_p V_p^*) g \rangle_{\mathcal{S}}$ telescopes **block-by-block in $\mathcal{S}$** into the discrete Brownian Bridge covariance kernel $K_{u, v}^{(p)} = \min(u, v) - \frac{u v}{p}$ on the harmonic jumps $\frac{(\nabla f)(p k + u)}{p k + u}$, while the cross-commutator inner product $\langle [V_p^*, V_q] f, [V_p^*, V_q] g \rangle_{\mathcal{S}}$ differs from it by a sub-bin continuous Brownian Bridge defect.

### Definition 1 (Block Partitioning and Interval-Overlap Cross-Commutator Matrix)
For each integer $p \ge 2$, let $\mathcal{U}_p : \ell^2(\mathbb{Z}_{\ge 0}) \to \ell^2(\mathbb{Z}_{\ge 0}; \mathbb{C}^p)$ be the unitary block-grouping isomorphism defined by $(\mathcal{U}_p x)_k := (x_{p k}, x_{p k + 1}, \dots, x_{p k + p - 1})^\top \in \mathbb{C}^p$ ($k \ge 0$). For any integers $p, q \ge 2$, define the $q \times p$ **interval-overlap cross-commutator matrix** $C_{p, q} \in \mathbb{R}^{q \times p}$ by
$$(C_{p, q})_{s, r} := \sqrt{p q}\,\left|\left[\frac{s}{q}, \frac{s+1}{q}\right] \cap \left[\frac{r}{p}, \frac{r+1}{p}\right]\right| - \frac{1}{\sqrt{p q}} \qquad (0 \le s \le q - 1,\; 0 \le r \le p - 1), \tag{3.1}$$
where $|I|$ denotes the length of an interval $I \subset [0, 1]$. Notice immediately from (3.1) that $C_{q, p} = C_{p, q}^\top$.

---

### Theorem 1 (Block-by-Block Brownian Bridge Wandering Kernel, Target Pairing, and Universal Sub-Bin Commutator Defect Law)
*Let $p, q \ge 2$ be integers. Define the $(p-1) \times (p-1)$ **discrete Brownian Bridge covariance matrix** $K^{(p)}$ and the **continuous Brownian Bridge kernel** $K_{\mathrm{cont}} : [0, 1]^2 \to \mathbb{R}$ by*
$$K_{u, v}^{(p)} := \min(u, v) - \frac{u v}{p} = \frac{\min(u, v)\big(p - \max(u, v)\big)}{p} \qquad (1 \le u, v \le p - 1), \qquad K_{\mathrm{cont}}(x, y) := \min(x, y) - x y. \tag{3.2}$$
*Then:*

1. **Block-by-Block Step-Space Telescoping and Brownian Bridge Wandering Formula:** *Partition $\mathbb{N} = B_0^{(p)} \sqcup \bigsqcup_{k=1}^\infty B_k^{(p)}$ into the initial prefix block $B_0^{(p)} := \{1, \dots, p - 1\}$ ($k = 0$) and the $p$-adic blocks $B_k^{(p)} := \{p k, p k + 1, \dots, p k + p - 1\}$ ($k \ge 1$). Then for every $f, g \in \mathcal{S}$ and **every individual block index $k \ge 0$**, the block inner product of $(I - V_p V_p^*) f$ and $(I - V_p V_p^*) g$ in $\mathcal{S}$ satisfies the exact finite identity*
   $$\sum_{m \in B_k^{(p)}} \frac{\big((I - V_p V_p^*) f\big)(m)\,\overline{\big((I - V_p V_p^*) g\big)(m)}}{m(m+1)} = \sum_{u=1}^{p-1} \sum_{v=1}^{p-1} K_{u, v}^{(p)} \frac{(\nabla f)(p k + u)\,\overline{(\nabla g)(p k + v)}}{(p k + u)(p k + v)} \qquad (\forall k \ge 0). \tag{3.3}$$
   *Summing over $k \ge 0$ gives, for all $f, g \in \mathcal{S}$,*
   $$\left\langle (I - V_p V_p^*) f,\, (I - V_p V_p^*) g \right\rangle_{\mathcal{S}} = \sum_{k=0}^\infty \sum_{u=1}^{p-1} \sum_{v=1}^{p-1} K_{u, v}^{(p)} \frac{(\nabla f)(p k + u)\,\overline{(\nabla g)(p k + v)}}{(p k + u)(p k + v)}. \tag{3.4}$$

2. **Exact Finite Rational Wandering and Cross-Commutator Target Pairings:** *Because $(\nabla E)(m) = \delta_{m, 1}$ is supported solely at $m = 1$ ($k = 0, u = 1$) where $K_{u, 1}^{(p)} = 1 - \frac{u}{p}$, every $f \in \mathcal{S}$ satisfies*
   $$\left\langle (I - V_p V_p^*) f,\, E \right\rangle_{\mathcal{S}} = \sum_{m=1}^{p-1} \frac{f(m)}{m(m+1)} = \sum_{u=1}^{p-1} \left(\frac{1}{u} - \frac{1}{p}\right)(\nabla f)(u), \qquad \big\|(I - V_p V_p^*) E\big\|_{\mathcal{S}}^2 = 1 - \frac{1}{p}. \tag{3.5}$$
   *In particular, for every cocycle $\widetilde{W}_\ell$ ($\ell \ge 2$), the wandering target inner product is the exact finite rational Ramanujan sum*
   $$\left\langle (I - V_p V_p^*)\widetilde{W}_\ell,\, E \right\rangle_{\mathcal{S}} = \frac{1}{\ell} \sum_{u=1}^{p-1} \left(\frac{1}{u} - \frac{1}{p}\right) c_\ell(u) \in \mathbb{Q}, \qquad \left\langle (I - V_2 V_2^*)\widetilde{W}_\ell,\, E \right\rangle_{\mathcal{S}} = \frac{\mu(\ell)}{2 \ell}. \tag{3.6a}$$
   *Unlike the full $L^2(0, \infty)$ dilations, the step-space cross-commutator $[V_p^*, V_q]$ does **not** annihilate the target $E$: for all integers $p, q \ge 2$, $([D_p^*, D_q] E)(m) = m(m+1)\sum_{\substack{0 \le r < p \\ p m + r \ge q}} \frac{1}{(p m + r)(p m + r + 1)} - \frac{1}{p}\mathbf{1}_{\{m \ge q\}}$ is supported solely on the finite prefix $1 \le m \le q - 1$ (since $\sum_{r=0}^{p-1} \frac{1}{(p m + r)(p m + r + 1)} = \frac{1}{p m(m+1)}$ cancels $\frac{1}{p}$ identically for all $m \ge q$), with*
   $$\big\|[V_p^*, V_q] E\big\|_{\mathcal{S}}^2 = M_{1, 1}^{(p, q)} = \frac{p - 1}{p} - \frac{p}{q}\left\{\frac{q}{p}\right\}\left(1 - \left\{\frac{q}{p}\right\}\right), \qquad \left\langle [V_p^*, V_q] E,\, E \right\rangle_{\mathcal{S}} = \frac{\min(p, q) - 1}{\sqrt{p q}}, \tag{3.6b}$$
   *and for every $f \in \mathcal{S}$, the cross-commutator target pairing is the exact finite $(p-1)$-term rational combination*
   $$\left\langle [V_p^*, V_q] f,\, E \right\rangle_{\mathcal{S}} = \frac{1}{\sqrt{p q}} \sum_{u=1}^{p-1} \frac{(p - u) - \max(0,\, p - q u)}{u}\,(\nabla f)(u). \tag{3.6c}$$
   *(For instance, at $(p, q) = (2, 3)$, $([D_2^*, D_3]E)(1) = \frac{1}{6}$, $([D_2^*, D_3]E)(2) = \frac{1}{2}$, $\|[V_2^*, V_3]E\|_{\mathcal{S}}^2 = \frac{1}{3}$, $\langle [V_2^*, V_3]E, E \rangle_{\mathcal{S}} = \frac{1}{\sqrt{6}}$, and $\langle [V_2^*, V_3]f, E \rangle_{\mathcal{S}} = \frac{f(1)}{\sqrt{6}}$, which equals $-\frac{1}{2\sqrt{6}}$ for $f = G_2$.)*

3. **Two-Sided Unitary Block Representation and Universal Sub-Bin Brownian Bridge Defect Law:** *Under the Kelvin–Cesàro unitary isomorphism $\Psi : \mathcal{S} \to \ell^2(\mathbb{Z}_{\ge 0})$, the cross-commutator $[V_p^*, V_q]$ admits the two-sided unitary block representation*
   $$\mathcal{U}_q \Psi [V_p^*, V_q] \Psi^{-1} \mathcal{U}_p^{-1} = I_{\ell^2(\mathbb{Z}_{\ge 0})} \otimes C_{p, q}, \tag{3.7}$$
   *and the self-adjoint Gram operator $[V_p^*, V_q]^* [V_p^*, V_q]$ is unitarily conjugate to the infinite block-diagonal operator*
   $$\mathcal{U}_p \Psi [V_p^*, V_q]^* [V_p^*, V_q] \Psi^{-1} \mathcal{U}_p^{-1} = I_{\ell^2(\mathbb{Z}_{\ge 0})} \otimes \left(C_{p, q}^* C_{p, q}\right). \tag{3.8}$$
   *For all $f, g \in \mathcal{S}$, the cross-commutator inner product is given by*
   $$\left\langle [V_p^*, V_q] f,\, [V_p^*, V_q] g \right\rangle_{\mathcal{S}} = \sum_{k=0}^\infty \sum_{u=1}^{p-1} \sum_{v=1}^{p-1} M_{u, v}^{(p, q)} \frac{(\nabla f)(p k + u)\,\overline{(\nabla g)(p k + v)}}{(p k + u)(p k + v)}, \tag{3.9}$$
   *where the $(p-1) \times (p-1)$ commutator bridge matrix $M^{(p, q)}$ has the universal exact form*
   $$M_{u, v}^{(p, q)} = K_{u, v}^{(p)} - \mathbf{1}_{\left\lfloor \frac{q u}{p} \right\rfloor = \left\lfloor \frac{q v}{p} \right\rfloor}\, \frac{p}{q}\, K_{\mathrm{cont}}\!\left(\left\{\frac{q u}{p}\right\},\, \left\{\frac{q v}{p}\right\}\right) \qquad (1 \le u, v \le p - 1). \tag{3.10}$$
   *Both $K^{(p)}$ and $M^{(p, q)}$ are palindromically invariant: $K_{p-u, p-v}^{(p)} = K_{u, v}^{(p)}$ and $M_{p-u, p-v}^{(p, q)} = M_{u, v}^{(p, q)}$. Moreover, whenever the integer parts $\lfloor q u / p \rfloor$ ($1 \le u \le p - 1$) are pairwise distinct—in particular, for **every $q \ge p$**, as well as for $(p, q) = (3, 2)$—all off-diagonal entries coincide ($M_{u, v}^{(p, q)} = K_{u, v}^{(p)}$ for $u \ne v$), so the **Wandering–Commutator Defect Form** is strictly diagonal:*
   $$\left\|(I - V_p V_p^*) f\right\|_{\mathcal{S}}^2 - \left\|[V_p^*, V_q] f\right\|_{\mathcal{S}}^2 = \frac{p}{q} \sum_{u=1}^{p-1} \left\{\frac{q u}{p}\right\}\left(1 - \left\{\frac{q u}{p}\right\}\right) \sum_{k=0}^\infty \frac{\big|(\nabla f)(p k + u)\big|^2}{(p k + u)^2}. \tag{3.11}$$

#### Proof of Theorem 1
1. Write $w_n := \frac{1}{n(n+1)} = \frac{1}{n} - \frac{1}{n+1}$. On the initial prefix block $B_0^{(p)} = \{1, \dots, p - 1\}$ ($k = 0$), since $\lfloor m / p \rfloor = 0$, we have $(V_p V_p^* f)(m) = 0$ and $f(m) = \sum_{u=1}^m (\nabla f)(u)$ (as $f(0) = 0$). Interchanging the finite sums over $m$ and $(u, v)$ and telescoping $\sum_{m=\max(u, v)}^{p-1} \left(\frac{1}{m} - \frac{1}{m+1}\right) = \frac{1}{\max(u, v)} - \frac{1}{p} = \frac{p - \max(u, v)}{p\,\max(u, v)} = \frac{\min(u, v)(p - \max(u, v))}{p\,u v} = \frac{K_{u, v}^{(p)}}{u v}$ gives (3.3) at $k = 0$.
   For each block $B_k^{(p)} = \{p k, \dots, p k + p - 1\}$ with $k \ge 1$, since $\sum_{r=0}^{p-1} w_{p k + r} = \frac{1}{p k} - \frac{1}{p(k+1)} = \frac{1}{p k(k+1)}$, the projection $(V_p V_p^* f)(p k + r) = p(D_p^* f)(k) = \frac{\sum_{s=0}^{p-1} w_{p k + s} f(p k + s)}{\sum_{s=0}^{p-1} w_{p k + s}}$ is the conditional expectation of $f$ on $B_k^{(p)}$ with respect to $w$. By Lagrange's identity for weighted variance,
   $$\sum_{r=0}^{p-1} w_{p k + r} \big((I - V_p V_p^*) f\big)(p k + r)\,\overline{\big((I - V_p V_p^*) g\big)(p k + r)} = p k (k+1) \sum_{0 \le r < s \le p-1} w_{p k + r}\,w_{p k + s}\,\big(f(p k + s) - f(p k + r)\big)\,\overline{\big(g(p k + s) - g(p k + r)\big)}.$$
   Substituting $f(p k + s) - f(p k + r) = \sum_{u=r+1}^s (\nabla f)(p k + u)$ and interchanging the finite sums, the coefficient of $(\nabla f)(p k + u)\overline{(\nabla g)(p k + v)}$ for any $1 \le u, v \le p - 1$ is
   $$p k (k+1) \left(\sum_{r=0}^{\min(u, v) - 1} w_{p k + r}\right)\left(\sum_{s=\max(u, v)}^{p-1} w_{p k + s}\right).$$
   Since both sums telescope via $w_n = \frac{1}{n} - \frac{1}{n+1}$ to
   $$\sum_{r=0}^{\min(u, v) - 1} w_{p k + r} = \frac{1}{p k} - \frac{1}{p k + \min(u, v)} = \frac{\min(u, v)}{p k(p k + \min(u, v))}, \qquad \sum_{s=\max(u, v)}^{p-1} w_{p k + s} = \frac{1}{p k + \max(u, v)} - \frac{1}{p(k+1)} = \frac{p - \max(u, v)}{p(k+1)(p k + \max(u, v))},$$
   their product with $p k (k+1)$ cancels $p k (k+1)$ identically and equals $\frac{\min(u, v)(p - \max(u, v))}{p(p k + u)(p k + v)} = \frac{K_{u, v}^{(p)}}{(p k + u)(p k + v)}$, establishing (3.3) and (3.4).

2. Since $(I - V_p V_p^*)$ is an orthogonal projection and $((I - V_p V_p^*)E)(m) = \mathbf{1}_{\{1 \le m \le p-1\}}$ (because $(V_p V_p^* E)(m) = (D_p E)(m) = \mathbf{1}_{\{m \ge p\}}$), we have $\langle (I - V_p V_p^*) f, E \rangle_{\mathcal{S}} = \langle (I - V_p V_p^*) f, (I - V_p V_p^*) E \rangle_{\mathcal{S}} = \sum_{m=1}^{p-1} \frac{f(m)}{m(m+1)}$. Equivalently, setting $g = E$ in (3.4) and using $(\nabla E)(m) = \delta_{m, 1}$ leaves only the $k = 0, v = 1$ terms $\sum_{u=1}^{p-1} K_{u, 1}^{(p)} \frac{(\nabla f)(u)}{u \cdot 1} = \sum_{u=1}^{p-1}(\frac{1}{u} - \frac{1}{p})(\nabla f)(u)$, proving (3.5) and (3.6a).
   For the cross-commutator $[D_p^*, D_q] = D_p^* D_q - D_q D_p^*$, since $(D_q E)(n) = \mathbf{1}_{\{n \ge q\}}$ and $(D_p^* E)(m) = \frac{1}{p}$ for all $m \ge 1$ (so $(D_q D_p^* E)(m) = \frac{1}{p}\mathbf{1}_{\{m \ge q\}}$), we have $([D_p^*, D_q]E)(m) = m(m+1)\sum_{0 \le r < p,\, p m + r \ge q} w_{p m + r} - \frac{1}{p}\mathbf{1}_{\{m \ge q\}}$, which vanishes for all $m \ge q$ because $\sum_{r=0}^{p-1} w_{p m + r} = \frac{1}{p m(m+1)}$. Applying (3.9) with $f = g = E$ (where $(\nabla E)(m) = \delta_{m, 1}$ selects $k = 0, u = v = 1$) gives $\|[V_p^*, V_q]E\|_{\mathcal{S}}^2 = M_{1, 1}^{(p, q)}$. Under $\Psi$, $\Psi E = \psi_0$ lies in the $k = 0$ block ($r = 0$), and for any $f \in \mathcal{S}$ with $k = 0$ block vector $x^{(0)} = ((\Psi f)_r)_{r=0}^{p-1} = x_0^{(0)}\mathbf{1}_p - \sum_{u=1}^{p-1} \frac{(\nabla f)(u)}{u}\mathbf{e}_{\ge u}$, the block representation (3.7) gives $\langle [V_p^*, V_q]f, E \rangle_{\mathcal{S}} = (C_{p, q} x^{(0)})_0 = -\sum_{u=1}^{p-1} \frac{(\nabla f)(u)}{u}(C_{p, q}\mathbf{e}_{\ge u})_0$. Since $I_0^{(q)} = [0, \frac{1}{q}]$ has $|I_0^{(q)} \cap [\frac{u}{p}, 1]| = \max(0, \frac{1}{q} - \frac{u}{p})$, we have $(C_{p, q}\mathbf{e}_{\ge u})_0 = \sqrt{p q}\left(\max(0, \frac{1}{q} - \frac{u}{p}) - \frac{p - u}{p q}\right) = -\frac{(p - u) - \max(0, p - q u)}{\sqrt{p q}}$, proving (3.6c); setting $f = E$ ($(\nabla E)(u) = \delta_{u, 1}$) yields $\langle [V_p^*, V_q]E, E \rangle_{\mathcal{S}} = \frac{(p - 1) - \max(0, p - q)}{\sqrt{p q}} = \frac{\min(p, q) - 1}{\sqrt{p q}}$, proving (3.6b).

3. By (1.1), for any $k \ge 0$ and $0 \le r \le p - 1$, $V_q \psi_{p k + r} = \frac{1}{\sqrt{q}} \sum_{j=0}^{q-1} \psi_{q p k + q r + j}$. Applying $V_p^*$ maps $\psi_n$ to $\frac{1}{\sqrt{p}} \psi_{\lfloor n / p \rfloor}$. For $n = p q k + q r + j$, $\lfloor n / p \rfloor = q k + s$ if and only if $p s \le q r + j < p(s + 1)$, i.e., $\frac{q r + j}{p q} \in [\frac{s}{q}, \frac{s+1}{q}) \cap [\frac{r}{p}, \frac{r+1}{p})$, which counts the number of grid intervals $[\frac{m}{p q}, \frac{m+1}{p q})$ in $[\frac{s}{q}, \frac{s+1}{q}] \cap [\frac{r}{p}, \frac{r+1}{p}]$ and equals $p q \left|\left[\frac{s}{q}, \frac{s+1}{q}\right] \cap \left[\frac{r}{p}, \frac{r+1}{p}\right]\right|$. Meanwhile, $V_q V_p^* \psi_{p k + r} = \frac{1}{\sqrt{p}} V_q \psi_k = \frac{1}{\sqrt{p q}} \sum_{s=0}^{q-1} \psi_{q k + s}$. Subtracting the two expressions proves (3.7), and taking adjoints gives (3.8).
   On each $p$-block $x^{(k)} = \big((\Psi f)_{p k + r}\big)_{r=0}^{p-1} \in \mathbb{C}^p$, since $C_{p, q} \mathbf{1}_p = \mathbf{0}$, only $x^{(k)} - x_0^{(k)} \mathbf{1}_p = \sum_{u=1}^{p-1} (x_u^{(k)} - x_{u-1}^{(k)})\,\mathbf{e}_{\ge u}$ contributes, where $\mathbf{e}_{\ge u} := \sum_{r=u}^{p-1} e_r \in \mathbb{R}^p$ and $x_u^{(k)} - x_{u-1}^{(k)} = -\frac{(\nabla f)(p k + u)}{p k + u}$ by (1.7). Thus $M_{u, v}^{(p, q)} = \langle C_{p, q} \mathbf{e}_{\ge u}, C_{p, q} \mathbf{e}_{\ge v} \rangle_{\mathbb{R}^q}$. Since $\bigcup_{r=u}^{p-1} [\frac{r}{p}, \frac{r+1}{p}] = [\frac{u}{p}, 1]$, the $s$-th component of $C_{p, q} \mathbf{e}_{\ge u}$ is $\sqrt{p q}\left(\left|I_s^{(q)} \cap \left[\frac{u}{p}, 1\right]\right| - \frac{1 - u/p}{q}\right)$ where $I_s^{(q)} = [\frac{s}{q}, \frac{s+1}{q}]$.
   Write $\frac{q u}{p} = s_u + \theta_u$ with $s_u = \lfloor \frac{q u}{p} \rfloor$ and $\theta_u = \{\frac{q u}{p}\}$. On $I_s^{(q)}$, the indicator $\mathbf{1}_{[u/p, 1]}$ is constant ($0$ for $s < s_u$ and $1$ for $s > s_u$) on every bin except $s = s_u$, where its average is $1 - \theta_u$. Since the projection of $\mathbf{1}_{[u/p, 1]} - (1 - \frac{u}{p})$ onto the piecewise-constant functions on the $q$-bins $\{I_s^{(q)}\}_{s=0}^{q-1}$ differs from $\mathbf{1}_{[u/p, 1]} - (1 - \frac{u}{p})$ in $L^2(0, 1)$ solely by its fluctuation inside the single straddling bin $I_{s_u}^{(q)}$, the Pythagorean theorem in $L^2(0, 1)$ gives
   $$\frac{1}{p} M_{u, v}^{(p, q)} = \int_0^1 \left(\mathbf{1}_{[u/p, 1]}(x) - 1 + \frac{u}{p}\right)\left(\mathbf{1}_{[v/p, 1]}(x) - 1 + \frac{v}{p}\right) dx - \mathbf{1}_{s_u = s_v} \frac{1}{q} K_{\mathrm{cont}}(\theta_u, \theta_v).$$
   Multiplying by $p$ and noting that the first integral is $\frac{1}{p} K_{u, v}^{(p)}$ yields (3.10) and (3.11). $\square$

---

## 4. Dirichlet Laplacian Duality, the Universal Symmetric Frobenius Trace Law, and the Discrete Legendre Spectrum

We next establish the tridiagonal Dirichlet Laplacian duality $(K^{(p)})^{-1} = \Delta_p$, the universal symmetric Frobenius trace formula valid across all $p, q \ge 2$, the 5-line monomial-triangular proof of the Discrete Legendre Spectrum, and the palindromic cosecant-cotangent law for $G_p$.

### Theorem 2 (Dirichlet Laplacian Duality, Symmetric Frobenius Trace, Discrete Legendre Spectrum, and Cosecant–Cotangent Law)
*Let $p, q \ge 2$ be integers. Then:*

1. **Dirichlet Laplacian Duality and Universal Similarity on $\mathbf{1}_p^\perp$:** *The inverse of $K^{(p)}$ is the $(p-1) \times (p-1)$ discrete Dirichlet Laplacian $\Delta_p := \operatorname{tridiag}(-1, 2, -1)$. For **every** pair of integers $p, q \ge 2$, the restriction of $C_{p, q}^* C_{p, q}$ to $\mathbf{1}_p^\perp \subset \mathbb{C}^p$ is similar to*
   $$(K^{(p)})^{-1} M^{(p, q)} = I_{p-1} - \Delta_p\big(K^{(p)} - M^{(p, q)}\big), \tag{4.1}$$
   *which, whenever the integer parts $\lfloor q u / p \rfloor$ ($1 \le u \le p - 1$) are pairwise distinct (in particular for all $q \ge p$), simplifies to*
   $$(K^{(p)})^{-1} M^{(p, q)} = I_{p-1} - \frac{p}{q}\,\Delta_p\,\operatorname{diag}\!\left(\left\{\frac{q u}{p}\right\}\left(1 - \left\{\frac{q u}{p}\right\}\right)\right)_{u=1}^{p-1}. \tag{4.2}$$

2. **Universal Symmetric Frobenius Trace Formula Across All $p, q \ge 2$:** *For **every** pair of integers $p, q \ge 2$ (regardless of whether $p \le q$ or $p > q$), the Frobenius trace of $C_{p, q}^* C_{p, q}$ equals*
   $$\operatorname{Tr}\big(C_{p, q}^* C_{p, q}\big) = \operatorname{Tr}\big(\Delta_p M^{(p, q)}\big) = \big(\min(p, q) - 1\big) - \frac{\min(p, q)^2 - \gcd(p, q)^2}{3\,\max(p, q)}. \tag{4.3}$$
   *(For example, at $(p, q) = (5, 2)$ and $(7, 2)$, where $p > q$ and off-diagonal entries of $K^{(p)} - M^{(p, q)}$ are non-zero, (4.3) gives the exact positive traces $\operatorname{Tr}(C_{5, 2}^* C_{5, 2}) = 1 - \frac{4 - 1}{15} = \frac{4}{5}$ and $\operatorname{Tr}(C_{7, 2}^* C_{7, 2}) = 1 - \frac{4 - 1}{21} = \frac{6}{7}$.)*

3. **5-Line Monomial-Triangular Proof of the Discrete Legendre Spectrum and Hahn Polynomial Specialization:** *Whenever $q \equiv \pm 1 \pmod p$ ($q \ge p - 1$, $q \ge 2$), the $(p-1) \times (p-1)$ operator $\Delta_p \operatorname{diag}\big(u(p - u)\big)_{u=1}^{p-1}$ is upper-triangular in the monomial basis $\{1, u, u^2, \dots, u^{p-2}\}$ of $\mathbb{R}^{p-1}$ with exact diagonal entries (and eigenvalues) $\{m(m+1) : 1 \le m \le p - 1\}$ (corresponding to the classical Hahn difference equation [19, 20, 26] with parameters $\alpha = \beta = 1$, grid size $N = p - 2$, shifted by $+2 I_{p-1}$). Consequently, the $p - 1$ singular values of $C_{p, q}\big|_{\mathbf{1}_p^\perp}$ (and of $[V_p^*, V_q]$ on $\mathcal{S}$, each of infinite multiplicity) are given in closed form by the **Discrete Legendre Spectrum***
   $$\sigma_m\big([V_p^*, V_q]\big) = \sqrt{1 - \frac{m(m+1)}{p q}} \qquad (1 \le m \le p - 1), \qquad \big\|[V_p^*, V_q]\big\|_{\mathrm{op}} = \sqrt{1 - \frac{2}{p q}}. \tag{4.4}$$
   *When $q \ge p + 1$, all $p - 1$ singular values in (4.4) are strictly positive; in the boundary case $q = p - 1$ ($p \ge 3$), the top $p - 2$ singular values ($1 \le m \le p - 2$) are strictly positive while the highest mode $m = p - 1$ yields $\sigma_{p-1}([V_p^*, V_{p-1}]) = 0$ (reflecting $\operatorname{rank}(C_{p, p-1}) = p - 2$). For $p = 5$ and $q \equiv \pm 2 \pmod 5$ ($q > 5$), the eigenvalues of $\Delta_5 \operatorname{diag}\big((2u \bmod 5)(5 - (2u \bmod 5))\big)_{u=1}^4 = \Delta_5 \operatorname{diag}(6, 4, 4, 6)$ are $\{8 - 2\sqrt{10},\, 12 - 2\sqrt{6},\, 8 + 2\sqrt{10},\, 12 + 2\sqrt{6}\}$, so $\|[V_5^*, V_q]\|_{\mathrm{op}} = \sqrt{1 - \frac{8 - 2\sqrt{10}}{5 q}} > \sqrt{1 - \frac{2}{5 q}}$.*

4. **Palindromic Cosecant-Squared and Divided-Difference Cotangent Law for $G_p$:** *For every integer $p \ge 2$ and every $q \ge p$, the step generator $G_p$ satisfies*
   $$\left\|(I - V_p V_p^*) G_p\right\|_{\mathcal{S}}^2 = \frac{\pi^2}{2 p^4} \sum_{u=1}^{p-1} K_{u, u}^{(p)} \csc^2\!\left(\frac{\pi u}{p}\right) + \frac{\pi}{p^3} \sum_{1 \le u < v \le p-1} K_{u, v}^{(p)} \frac{\cot(\pi u / p) - \cot(\pi v / p)}{v - u}, \tag{4.5}$$
   $$\left\|[V_p^*, V_q] G_p\right\|_{\mathcal{S}}^2 = \frac{\pi^2}{2 p^4} \sum_{u=1}^{p-1} M_{u, u}^{(p, q)} \csc^2\!\left(\frac{\pi u}{p}\right) + \frac{\pi}{p^3} \sum_{1 \le u < v \le p-1} K_{u, v}^{(p)} \frac{\cot(\pi u / p) - \cot(\pi v / p)}{v - u}. \tag{4.6}$$
   *In particular, for $p = 2$ and any odd $q \ge 3$, $\|(I - V_2 V_2^*) G_2\|_{\mathcal{S}}^2 = \frac{\pi^2}{64}$ and $\|[V_2^*, V_q] G_2\|_{\mathcal{S}}^2 = \left(1 - \frac{1}{q}\right)\frac{\pi^2}{64}$; for $p = 3$ and any $q \ge 2$ with $3 \nmid q$,*
   $$\left\|(I - V_3 V_3^*) G_3\right\|_{\mathcal{S}}^2 = \frac{8\pi^2}{729} + \frac{2\pi}{81\sqrt{3}}, \qquad \left\|[V_3^*, V_q] G_3\right\|_{\mathcal{S}}^2 = \left(1 - \frac{1}{q}\right)\frac{8\pi^2}{729} + \frac{2\pi}{81\sqrt{3}}. \tag{4.7}$$

#### Proof of Theorem 2
1. Direct multiplication gives $(K^{(p)} \Delta_p)_{u, v} = -K_{u, v-1}^{(p)} + 2 K_{u, v}^{(p)} - K_{u, v+1}^{(p)} = \delta_{u, v}$ (with $K_{u, 0}^{(p)} = K_{u, p}^{(p)} = 0$), so $(K^{(p)})^{-1} = \Delta_p$. Since $\tilde{w}_u := \mathbf{e}_{\ge u} - \frac{p-u}{p}\mathbf{1}_p$ ($1 \le u \le p - 1$) is a basis of $\mathbf{1}_p^\perp$ with Gram matrix $\langle \tilde{w}_u, \tilde{w}_v \rangle = K_{u, v}^{(p)}$ and $\langle C_{p, q} \tilde{w}_u, C_{p, q} \tilde{w}_v \rangle = M_{u, v}^{(p, q)}$, the matrix of $C_{p, q}^* C_{p, q}\big|_{\mathbf{1}_p^\perp}$ in the basis $\{\tilde{w}_u\}$ is $(K^{(p)})^{-1} M^{(p, q)} = I_{p-1} - \Delta_p(K^{(p)} - M^{(p, q)})$, giving (4.1) and (4.2).

2. First suppose $p \le q$. Then the integer parts $\lfloor q u / p \rfloor$ ($1 \le u \le p - 1$) are strictly increasing, so $K^{(p)} - M^{(p, q)}$ is diagonal with $u$-th diagonal entry $\frac{p}{q}\{\frac{q u}{p}\}(1 - \{\frac{q u}{p}\})$. Since $\Delta_p$ has diagonal entries $2$, taking the trace of (4.2) gives
   $$\operatorname{Tr}\big(C_{p, q}^* C_{p, q}\big) = (p - 1) - \frac{2 p}{q} \sum_{u=1}^{p-1} \left\{\frac{q u}{p}\right\}\left(1 - \left\{\frac{q u}{p}\right\}\right) \qquad (p \le q).$$
   Let $g = \gcd(p, q)$. As $u$ runs over $\{1, \dots, p - 1\}$, the residues $q u \bmod p$ consist of $g$ copies of $\{g, 2g, \dots, p - g\}$ together with $g - 1$ zeros, so
   $$\frac{2 p}{q} \sum_{u=1}^{p-1} \left\{\frac{q u}{p}\right\}\left(1 - \left\{\frac{q u}{p}\right\}\right) = \frac{2 g}{p q} \sum_{j=1}^{p/g - 1} (g j)(p - g j) = \frac{2 g^3}{p q} \cdot \frac{(p/g - 1)(p/g)(p/g + 1)}{6} = \frac{p^2 - \gcd(p, q)^2}{3 q},$$
   proving (4.3) when $p \le q$. Now suppose $p > q$. By Definition 1, $C_{p, q} = C_{q, p}^\top$, so the Frobenius norm squared is symmetric in $p$ and $q$:
   $$\operatorname{Tr}\big(C_{p, q}^* C_{p, q}\big) = \|C_{p, q}\|_F^2 = \|C_{q, p}\|_F^2 = \operatorname{Tr}\big(C_{q, p}^* C_{q, p}\big).$$
   Since $q < p$, applying the already proved formula to $(q, p)$ interchanges $p$ and $q$, replacing $p$ by $\min(p, q)$ and $q$ by $\max(p, q)$ and establishing (4.3) for all $p, q \ge 2$. (Notice why the asymmetric expression fails when $p > q$: in $\operatorname{Tr}(\Delta_p(K^{(p)} - M^{(p, q)})) = 2\sum_{u=1}^{p-1}(K_{u,u}^{(p)} - M_{u,u}^{(p,q)}) - 2\sum_{u=1}^{p-2}(K_{u,u+1}^{(p)} - M_{u,u+1}^{(p,q)})$, whenever $p > q$, consecutive grid points $u/p$ and $(u+1)/p$ can fall in the same $q$-bin $\lfloor q u / p \rfloor = \lfloor q(u+1)/p \rfloor$, contributing non-zero sub-diagonal defect terms $K_{u, u+1}^{(p)} - M_{u, u+1}^{(p, q)} > 0$ that restore the exact symmetric value in (4.3).)

3. When $q \equiv \pm 1 \pmod p$ ($q \ge p - 1$, $q \ge 2$), the integer parts $\lfloor q u / p \rfloor$ ($1 \le u \le p - 1$) are pairwise distinct (for $q \ge p$ because $q/p \ge 1$, and for $q = p - 1$ because $\lfloor (p-1)u/p \rfloor = u - 1$), and $\left\{\frac{q u}{p}\right\} \in \left\{\frac{u}{p}, 1 - \frac{u}{p}\right\}$, so $\frac{p}{q}\left\{\frac{q u}{p}\right\}\left(1 - \left\{\frac{q u}{p}\right\}\right) = \frac{u(p - u)}{p q}$. Let $P(u)$ be any polynomial of degree $m - 1$ ($1 \le m \le p - 1$), and set $Q(u) := u(p - u) P(u)$. Since $Q(0) = Q(p) = 0$, the missing boundary terms $-Q(0)$ at $u = 1$ and $-Q(p)$ at $u = p - 1$ in the truncated matrix $\Delta_p$ vanish identically! Consequently, for every $u \in \{1, \dots, p - 1\}$,
   $$\big(\Delta_p \operatorname{diag}\big(u(p - u)\big) P\big)(u) = -Q(u - 1) + 2 Q(u) - Q(u + 1).$$
   If $P(u) = u^{m-1} + O(u^{m-2})$, then $Q(u) = -u^{m+1} + O(u^m)$, whose second difference $-Q(u-1) + 2 Q(u) - Q(u+1)$ is a polynomial of degree $(m+1) - 2 = m - 1$ with leading term $\binom{m+1}{2} \cdot 2\, u^{m-1} = m(m+1) u^{m-1}$. Hence $\Delta_p \operatorname{diag}(u(p - u))$ preserves the flag $\operatorname{span}\{1, u, \dots, u^{m-1}\}$ and is upper-triangular in the monomial basis $\{u^{m-1}\}_{m=1}^{p-1}$ with diagonal entries (and eigenvalues) $m(m+1)$ ($1 \le m \le p - 1$), proving (4.4). Equivalently, in terms of the classical Hahn polynomials $Q_n(x; \alpha, \beta, N)$ on $x \in \{0, 1, \dots, N\}$ (see Szegő [19], Koekoek–Lesky–Swarttouw [20], and NIST DLMF [26, §18.22(ii), eqs. 18.22.10–18.22.11]), setting $x = u - 1 \in \{0, 1, \dots, p - 2\}$, $N = p - 2$, and $\alpha = \beta = 1$ gives forward and backward rates $B(x) = (x + \alpha + 1)(N - x) = (x + 2)(p - 2 - x) = (u + 1)(p - 1 - u)$ and $D(x) = x(N + \beta + 1 - x) = x(p - x) = (u - 1)(p + 1 - u)$, whose sum is $B(u - 1) + D(u - 1) = (u + 1)(p - 1 - u) + (u - 1)(p + 1 - u) = 2 u(p - u) - 2$. Consequently, the positive-definite Hahn difference operator $(L_{\mathrm{Hahn}} P)(u) := -B(u - 1) P(u + 1) + \big[B(u - 1) + D(u - 1)\big] P(u) - D(u - 1) P(u - 1)$ with eigenvalues $\lambda_n = n(n + \alpha + \beta + 1) = n(n + 3)$ ($0 \le n \le p - 2$) satisfies the exact matrix identity $\Delta_p \operatorname{diag}\big(u(p - u)\big)_{u=1}^{p-1} = L_{\mathrm{Hahn}} + 2 I_{p-1}$, recovering the exact eigenvalues $\lambda_n + 2 = n(n + 3) + 2 = (n + 1)(n + 2) = m(m + 1)$ for $m = n + 1 \in \{1, \dots, p - 1\}$.

4. For $f = g = G_p$, we have $(\nabla G_p)(p k + u) = -\frac{1}{p}$ for all $k \ge 0$ and $1 \le u \le p - 1$. In (3.4) and (3.9), the diagonal series ($u = v$) is $\sum_{k=0}^\infty \frac{1}{p^2(p k + u)^2} = \frac{1}{p^4}\zeta(2, \frac{u}{p})$. Pairing $u$ with $p - u$ and using $K_{p-u, p-u}^{(p)} = K_{u, u}^{(p)}$, $M_{p-u, p-u}^{(p, q)} = M_{u, u}^{(p, q)}$, and $\zeta(2, \frac{u}{p}) + \zeta(2, 1 - \frac{u}{p}) = \pi^2 \csc^2(\frac{\pi u}{p})$ gives the first sums in (4.5)–(4.6). For $1 \le u < v \le p - 1$, partial fractions give $\sum_{k=0}^\infty \frac{1}{p^2(p k + u)(p k + v)} = \frac{\psi(v/p) - \psi(u/p)}{p^3(v - u)}$. Pairing $(u, v)$ with $(p - v, p - u)$ and using $K_{p-v, p-u}^{(p)} = K_{u, v}^{(p)}$ and Euler's reflection identity $\psi(1 - \frac{u}{p}) - \psi(\frac{u}{p}) = \pi\cot(\frac{\pi u}{p})$ yields the divided-difference cotangent sums in (4.5)–(4.6). Evaluating at $p = 2$ and $p = 3$ gives (4.7). $\square$

---

## 5. Exact Euler-Product Gram Matrices and the Dyadic Jordan Totient Sieve

When the residue weights in (3.4) or (3.11) are constant across all $u \in \{1, \dots, p - 1\}$, the sum over $k \ge 0$ and $1 \le u \le p - 1$ ranges over all integers $m \ge 1$ with $p \nmid m$. Although for a fixed $\ell \ge 2$ the Ramanujan sum $m \mapsto c_\ell(m)$ is **not** multiplicative in $m$ (for example, $c_2(1) = -1$ whereas $c_2(1)^2 = +1$), the prime-power multiplicativity of $\ell \mapsto c_\ell(m)$ together with the dependence of $c_{r^a}(m)$ solely on the single prime valuation $r^{v_r(m)}$ factors the general term $c_\ell(m)$ into independent functions of the prime-power components $r^{v_r(m)}$ of $m = \prod_r r^{v_r(m)}$. Consequently, the Dirichlet series over $p \nmid m$ factors into a closed-form Euler product times $\zeta(2) = \frac{\pi^2}{6}$. Furthermore, on the dyadic wandering subspace $\mathcal{W}_2^{\mathcal{S}}$, this prime-valuation independence diagonalizes the entire finite-cutoff Gram quadratic form via a multiplicative **Jordan totient sieve**, yielding an exact closed-form evaluation of the $N$-term dyadic wandering distance $d_N^2(\mathcal{W}_2^{\mathcal{S}})$ for every $N \ge 2$.

### Theorem 3 (Exact Euler-Product Gram Matrices for Dyadic Wandering/Commutator Projections and Ternary Defects)
*For each prime $r$ and non-negative integers $a, b \ge 0$, define the rational local Euler factor*
$$E_r(a, b) := \begin{cases} 1, & \text{if } a = b = 0, \\[4pt] 2\left(1 - \dfrac{1}{r}\right), & \text{if } a = b \ge 1, \\[8pt] -\dfrac{\varphi(r^{\min(a,b)})(r - 1)}{r^{\max(a,b)}}, & \text{if } a \ne b \quad (\text{with } \varphi(1) := 1). \end{cases} \tag{5.1}$$
*Then for all cocycle indices $\ell_1, \ell_2 \ge 2$:*

1. **Universal $p$-Coprime Ramanujan $L(2)$-Series Euler Product:** *For every prime $p$,*
   $$\frac{1}{\ell_1 \ell_2} \sum_{\substack{m=1 \\ p \nmid m}}^\infty \frac{c_{\ell_1}(m)\, c_{\ell_2}(m)}{m^2} = \left(1 - \frac{1}{p^2}\right)\frac{\pi^2}{6}\, \frac{\mu(p^{\nu_p(\ell_1)})\,\mu(p^{\nu_p(\ell_2)})}{\ell_1 \ell_2} \prod_{\substack{r \mid \ell_1 \ell_2 \\ r \ne p \text{ prime}}} E_r\big(\nu_r(\ell_1), \nu_r(\ell_2)\big). \tag{5.2}$$

2. **Exact Closed-Form Dyadic Wandering and Cross-Commutator Gram Matrix ($p = 2$):** *For all $\ell_1, \ell_2 \ge 2$ and every odd integer $q \ge 3$,*
   $$\left\langle (I - V_2 V_2^*)\widetilde{W}_{\ell_1},\, (I - V_2 V_2^*)\widetilde{W}_{\ell_2} \right\rangle_{\mathcal{S}} = \frac{\pi^2}{16}\, \frac{\mu(2^{\nu_2(\ell_1)})\,\mu(2^{\nu_2(\ell_2)})}{\ell_1 \ell_2} \prod_{\substack{r \mid \ell_1 \ell_2 \\ r > 2 \text{ prime}}} E_r\big(\nu_r(\ell_1), \nu_r(\ell_2)\big), \tag{5.3}$$
   $$\left\langle [V_2^*, V_q]\widetilde{W}_{\ell_1},\, [V_2^*, V_q]\widetilde{W}_{\ell_2} \right\rangle_{\mathcal{S}} = \left(1 - \frac{1}{q}\right)\left\langle (I - V_2 V_2^*)\widetilde{W}_{\ell_1},\, (I - V_2 V_2^*)\widetilde{W}_{\ell_2} \right\rangle_{\mathcal{S}}. \tag{5.4}$$
   *In particular, for every $\ell \ge 2$ with $4 \nmid \ell$, writing $\ell = 2^{\nu_2(\ell)} \ell_{\mathrm{odd}}$,*
   $$\left\|(I - V_2 V_2^*)\widetilde{W}_\ell\right\|_{\mathcal{S}}^2 = \frac{\pi^2}{16\,\ell^2}\, 2^{\omega(\ell_{\mathrm{odd}})} \frac{\varphi(\ell_{\mathrm{odd}})}{\ell_{\mathrm{odd}}}, \qquad \left\|[V_2^*, V_q]\widetilde{W}_\ell\right\|_{\mathcal{S}}^2 = \left(1 - \frac{1}{q}\right)\frac{\pi^2}{16\,\ell^2}\, 2^{\omega(\ell_{\mathrm{odd}})} \frac{\varphi(\ell_{\mathrm{odd}})}{\ell_{\mathrm{odd}}}. \tag{5.5}$$

3. **Exact Closed-Form Ternary Wandering–Commutator Defect Gram Matrix ($p = 3$):** *For all $\ell_1, \ell_2 \ge 2$ and every integer $q \ge 2$ with $3 \nmid q$, the ternary defect matrix $K^{(3)} - M^{(3, q)} = \frac{2}{3 q} I_2$ is a scalar multiple of the $2 \times 2$ identity matrix, and*
   $$\left\langle (I - V_3 V_3^*)\widetilde{W}_{\ell_1},\, (I - V_3 V_3^*)\widetilde{W}_{\ell_2} \right\rangle_{\mathcal{S}} - \left\langle [V_3^*, V_q]\widetilde{W}_{\ell_1},\, [V_3^*, V_q]\widetilde{W}_{\ell_2} \right\rangle_{\mathcal{S}} = \frac{8\pi^2}{81 q}\, \frac{\mu(3^{\nu_3(\ell_1)})\,\mu(3^{\nu_3(\ell_2)})}{\ell_1 \ell_2} \prod_{\substack{r \mid \ell_1 \ell_2 \\ r \ne 3 \text{ prime}}} E_r\big(\nu_r(\ell_1), \nu_r(\ell_2)\big). \tag{5.6}$$

#### Proof of Theorem 3
1. For any fixed $\ell = \prod_r r^{\nu_r(\ell)} \ge 2$ and any integer $m = \prod_r r^{v_r(m)} \ge 1$, the multiplicativity of $\ell' \mapsto c_{\ell'}(m)$ for fixed $m$ together with $\gcd(r^{\nu_r(\ell)}, m) = \gcd(r^{\nu_r(\ell)}, r^{v_r(m)})$ gives the prime-valuation factorization
   $$c_\ell(m) = \prod_{r \mid \ell} c_{r^{\nu_r(\ell)}}(m) = \prod_{r \mid \ell} c_{r^{\nu_r(\ell)}}\!\left(r^{v_r(m)}\right).$$
   For every prime $r \nmid \ell_1 \ell_2$, we have $\nu_r(\ell_1) = \nu_r(\ell_2) = 0$ and $c_1(r^{v_r(m)}) = 1$. Therefore, as $m = \prod_{r \ne p} r^{k_r}$ runs over all positive integers coprime to $p$ (so $v_p(m) = 0$ and the exponent tuple $(k_r)_{r \ne p}$ runs over all finitely supported sequences of non-negative integers), $|c_{\ell_i}(m)| \le \ell_i$ ensures absolute convergence of $\sum_{p \nmid m} \frac{c_{\ell_1}(m) c_{\ell_2}(m)}{m^2}$ and justifies factoring the sum over the independent prime-exponent coordinates $k_r = v_r(m) \ge 0$ ($r \ne p$):
   $$\sum_{\substack{m=1 \\ p \nmid m}}^\infty \frac{c_{\ell_1}(m)\, c_{\ell_2}(m)}{m^2} = c_{p^{\nu_p(\ell_1)}}(1)\, c_{p^{\nu_p(\ell_2)}}(1) \prod_{r \ne p \text{ prime}} \left(\sum_{k=0}^\infty \frac{c_{r^{\nu_r(\ell_1)}}(r^k)\, c_{r^{\nu_r(\ell_2)}}(r^k)}{r^{2 k}}\right).$$
   Notice first that $c_{p^v}(1) = \mu(p^v)$ for all $v \ge 0$. For any prime $r \ne p$, set $a = \nu_r(\ell_1)$ and $b = \nu_r(\ell_2)$, and recall that for $a \ge 1$, $c_{r^a}(r^k)$ equals $0$ for $0 \le k \le a - 2$, $-r^{a-1}$ for $k = a - 1$, and $\varphi(r^a) = r^{a-1}(r - 1)$ for $k \ge a$ (while for $a = 0$, $c_1(r^k) = 1$ for all $k \ge 0$):
   - If $a = b = 0$, the local Euler series is $\sum_{k=0}^\infty r^{-2k} = (1 - r^{-2})^{-1} = (1 - r^{-2})^{-1} E_r(0, 0)$.
   - If $a = b \ge 1$, only $k = a - 1$ and $k \ge a$ are non-zero, so multiplying by $(1 - r^{-2})$ gives
     $$(1 - r^{-2})\left(\frac{(-r^{a-1})^2}{r^{2a - 2}} + \sum_{k=a}^\infty \frac{r^{2a - 2}(r - 1)^2}{r^{2 k}}\right) = (1 - r^{-2}) + \frac{(r - 1)^2}{r^2} = 2\left(1 - \frac{1}{r}\right) = E_r(a, a).$$
   - If $a \ne b$, let $m = \min(a, b)$ and $M = \max(a, b) \ge 1$. Since $c_{r^M}(r^k) = 0$ for $k < M - 1$, only $k = M - 1$ and $k \ge M$ contribute. Because $m \le M - 1$, we have $c_{r^m}(r^k) = \varphi(r^m)$ for all $k \ge M - 1$ (with $\varphi(1) = 1$ when $m = 0$). Multiplying by $(1 - r^{-2})$ gives
     $$(1 - r^{-2})\,\varphi(r^m)\left(\frac{-r^{M-1}}{r^{2M - 2}} + \sum_{k=M}^\infty \frac{r^{M-1}(r - 1)}{r^{2 k}}\right) = \varphi(r^m)\left(-\frac{1 - r^{-2}}{r^{M-1}} + \frac{r - 1}{r^{M+1}}\right) = -\frac{\varphi(r^m)(r - 1)}{r^M} = E_r(a, b).$$
   Factoring out $\prod_{r \ne p} (1 - r^{-2})^{-1} = (1 - p^{-2})\zeta(2) = (1 - p^{-2})\frac{\pi^2}{6}$ and dividing by $\ell_1 \ell_2$ proves (5.2).

2. At $p = 2$, $u = v = 1$ is the unique index in $\{1, \dots, p - 1\}$, with $K_{1, 1}^{(2)} = \frac{1}{2}$, and for every odd $q \ge 3$ we have $\left\{\frac{q}{2}\right\} = \frac{1}{2}$, so $M_{1, 1}^{(2, q)} = \frac{1}{2} - \frac{2}{q}\cdot\frac{1}{4} = \frac{q - 1}{2 q} = \left(1 - \frac{1}{q}\right) K_{1, 1}^{(2)}$. Substituting $(\nabla \widetilde{W}_{\ell_i})(2k + 1) = \frac{1}{\ell_i} c_{\ell_i}(2k + 1)$ into (3.4) and (3.9) and applying (5.2) at $p = 2$ (where $\frac{1}{2}(1 - \frac{1}{4})\frac{\pi^2}{6} = \frac{\pi^2}{16}$) yields (5.3)–(5.5).

3. At $p = 3$, for any integer $q \ge 2$ with $3 \nmid q$, the integer parts $\lfloor \frac{q \cdot 1}{3} \rfloor$ and $\lfloor \frac{q \cdot 2}{3} \rfloor$ are distinct because $\frac{2 q}{3} - \frac{q}{3} = \frac{q}{3} \ge \frac{2}{3}$ and $\left\{\frac{q}{3}\right\} + \left\{\frac{2 q}{3}\right\} = 1$. Moreover, $\left\{\frac{q u}{3}\right\} \in \left\{\frac{1}{3}, \frac{2}{3}\right\}$ for both $u = 1$ and $u = 2$, so $\frac{3}{q}\left\{\frac{q u}{3}\right\}\left(1 - \left\{\frac{q u}{3}\right\}\right) = \frac{3}{q} \cdot \frac{2}{9} = \frac{2}{3 q}$ for both $u = 1$ and $u = 2$. Thus $K^{(3)} - M^{(3, q)} = \frac{2}{3 q} I_2$. Polarizing (3.11) for $f = \widetilde{W}_{\ell_1}, g = \widetilde{W}_{\ell_2}$ and applying (5.2) at $p = 3$ (where $\frac{2}{3 q}(1 - \frac{1}{9})\frac{\pi^2}{6} = \frac{8\pi^2}{81 q}$) proves (5.6). $\square$

### Theorem 4 (Exact Jordan Totient Sieve Orthogonalization and Dyadic Wandering Distance on $\mathcal{W}_2^{\mathcal{S}}$)
*Let $P_{\mathcal{W}_2} := I - V_2 V_2^*$ be the orthogonal projection onto the dyadic wandering subspace $\mathcal{W}_2^{\mathcal{S}}$, and for each cutoff $N \ge 2$ let*
$$d_N^2(\mathcal{W}_2^{\mathcal{S}}) := \min_{c_2, \dots, c_N \in \mathbb{C}} \left\|P_{\mathcal{W}_2}\!\left(E - \sum_{\ell=2}^N c_\ell\,\widetilde{W}_\ell\right)\right\|_{\mathcal{S}}^2 \tag{5.7}$$
*denote the squared distance of $P_{\mathcal{W}_2} E$ to the projected cocycle span $P_{\mathcal{W}_2}\mathcal{V}_N$. For every integer $\ell \ge 1$, let $J_2(\ell) := \ell^2 \prod_{p \mid \ell}(1 - p^{-2})$ be Jordan's second totient function (with $J_2(1) := 1$). Then:*

1. **Redundancy of Even Indices and Exact Vanishing of Non-Squarefree Odd Optimal Coefficients:** *For every odd integer $\ell \ge 3$, $P_{\mathcal{W}_2}\widetilde{W}_{2\ell} = -\frac{1}{2} P_{\mathcal{W}_2}\widetilde{W}_\ell$, while $P_{\mathcal{W}_2}\widetilde{W}_\ell = 0$ whenever $4 \mid \ell$. Consequently, adding the single generator $P_{\mathcal{W}_2}\widetilde{W}_2 = P_{\mathcal{W}_2} G_2$ (which has constant odd gradient $(\nabla \widetilde{W}_2)(2k+1) = -\frac{1}{2}$) reduces $P_{\mathcal{W}_2}\mathcal{V}_N$ to the span of $\{P_{\mathcal{W}_2}\widetilde{W}_2\} \cup \{P_{\mathcal{W}_2}\widetilde{W}_\ell : 3 \le \ell \le N \text{ odd}\}$. On this basis, the unique optimal approximation $F_N^* := c_2^{(N)} P_{\mathcal{W}_2}\widetilde{W}_2 + \sum_{\substack{3 \le \ell \le N \\ \ell \text{ odd}}} a_\ell^{(N)} P_{\mathcal{W}_2}\widetilde{W}_\ell$ to $P_{\mathcal{W}_2} E$ is supported **strictly on squarefree odd indices**:*
   $$a_\ell^{(N)} = 0 \qquad \text{for every non-squarefree odd integer } 3 \le \ell \le N. \tag{5.8}$$

2. **Closed-Form Jordan Totient Sieve Formula for $d_N^2(\mathcal{W}_2^{\mathcal{S}})$ and Optimal Coefficients:** *For every integer $N \ge 2$, define the rational cumulative Jordan totient sum*
   $$q_N := 4 \sum_{\substack{1 \le \ell \le N \\ \ell \text{ odd}}} \frac{\mu(\ell)^2}{J_2(\ell)} \in \mathbb{Q}. \tag{5.9}$$
   *Then for every $N \ge 2$, the squared dyadic wandering distance and optimal coefficients are given in closed form by*
   $$d_N^2(\mathcal{W}_2^{\mathcal{S}}) = \frac{1}{2} - \frac{q_N}{\pi^2} = \frac{4}{\pi^2} \sum_{\substack{\ell > N \\ \ell \text{ odd}}} \frac{\mu(\ell)^2}{J_2(\ell)}, \tag{5.10}$$
   $$a_\ell^{(N)} = \frac{8\,\mu(\ell)\,\ell^2}{\pi^2\,J_2(\ell)} \sum_{\substack{1 \le m \le N/\ell \\ m \text{ odd},\; \gcd(m, \ell) = 1}} \frac{\mu(m)\,\varphi(m)}{J_2(m)} = \frac{8\,\ell^2}{\pi^2} \sum_{\substack{1 \le m \le N/\ell \\ m \text{ odd}}} \frac{\mu(\ell m)\,\varphi(m)}{J_2(\ell m)} \quad (3 \le \ell \le N \text{ odd}), \qquad c_2^{(N)} = -\frac{16}{\pi^2} \sum_{\substack{1 \le d \le N \\ d \text{ odd}}} \frac{\mu(d)\,\varphi(d)}{J_2(d)}, \tag{5.11}$$
   *(for instance, at $N = 3$, $(c_2^{(3)}, a_3^{(3)}) = \pi^{-2}(-12, -9)$), and as $N \to \infty$, the dyadic wandering distance decays at the exact unconditional $O(N^{-1})$ rate*
   $$\sum_{\substack{1 \le \ell < \infty \\ \ell \text{ odd}}} \frac{\mu(\ell)^2}{J_2(\ell)} = \prod_{p > 2 \text{ prime}} \left(1 + \frac{1}{p^2 - 1}\right) = \frac{\pi^2}{8}, \qquad d_N^2(\mathcal{W}_2^{\mathcal{S}}) = \frac{2\,C_{\mathrm{odd}}}{\pi^2 N} + O(N^{-3/2}), \tag{5.12}$$
   *where the odd Euler product constant $C_{\mathrm{odd}}$ and leading coefficient $\frac{2\,C_{\mathrm{odd}}}{\pi^2}$ are given by*
   $$C_{\mathrm{odd}} := \prod_{p > 2 \text{ prime}} \left(1 - \frac{1}{p(p+1)}\right) \approx 0.8453306412, \qquad \frac{2\,C_{\mathrm{odd}}}{\pi^2} \approx 0.1712998023. \tag{5.13}$$

#### Proof of Theorem 4
By Theorem 1(1)–(2) at $p = 2$ ($u = v = 1$, $K_{1, 1}^{(2)} = \frac{1}{2}$), the restriction of $\langle P_{\mathcal{W}_2} f, P_{\mathcal{W}_2} g \rangle_{\mathcal{S}}$ is isometric to $L^2(\mathbb{N}_{\mathrm{odd}}, d\nu)$ with probability measure $d\nu(m) := \frac{8}{\pi^2 m^2}$ on $\mathbb{N}_{\mathrm{odd}} = \{1, 3, 5, \dots\}$ (since $\sum_{m \ge 1 \text{ odd}} m^{-2} = \frac{\pi^2}{8}$), via the map $P_{\mathcal{W}_2} f \mapsto (\nabla f)(m)$ for odd $m \ge 1$:
$$\langle P_{\mathcal{W}_2} f, P_{\mathcal{W}_2} g \rangle_{\mathcal{S}} = \frac{\pi^2}{16}\,\mathbb{E}_\nu\!\big[(\nabla f)\,\overline{(\nabla g)}\big], \qquad \langle P_{\mathcal{W}_2} f, E \rangle_{\mathcal{S}} = \frac{1}{2}(\nabla f)(1).$$
For odd $m \ge 1$, $c_{2\ell}(m) = c_2(m) c_\ell(m) = -c_\ell(m)$ for odd $\ell$, and $c_\ell(m) = 0$ if $4 \mid \ell$, so $(\nabla \widetilde{W}_{2\ell})(m) = -\frac{1}{2}(\nabla \widetilde{W}_\ell)(m)$ and $P_{\mathcal{W}_2}\widetilde{W}_{2\ell} = -\frac{1}{2} P_{\mathcal{W}_2}\widetilde{W}_\ell$. Moreover, $(\nabla \widetilde{W}_2)(m) = -\frac{1}{2}$ is the constant function on $\mathbb{N}_{\mathrm{odd}}$, so $\operatorname{span}\{(\nabla \widetilde{W}_\ell)\big|_{\mathbb{N}_{\mathrm{odd}}} : 2 \le \ell \le N\} = \operatorname{span}(\{1\} \cup \{c_\ell\big|_{\mathbb{N}_{\mathrm{odd}}} : 3 \le \ell \le N \text{ odd}\})$.
Under the probability measure $\nu$ on $\mathbb{N}_{\mathrm{odd}} = \prod_{p > 2} \{p^k : k \ge 0\}$, the prime-power valuations $p^{v_p(m)}$ ($p > 2$ prime) are **mutually independent random variables** with marginal laws $\mathbb{P}_\nu(v_p(m) = k) = (1 - p^{-2}) p^{-2k}$ ($k \ge 0$). For each odd prime $p$, the function $c_p(m)$ has expectation
$$\mathbb{E}_\nu[c_p] = (1 - p^{-2})\left(-1 + \sum_{k=1}^\infty (p - 1) p^{-2k}\right) = (1 - p^{-2})\left(-1 + \frac{p - 1}{p^2 - 1}\right) = -\frac{p - 1}{p}.$$
Define the centered prime variable $X_p(m) := c_p(m) + \frac{p - 1}{p}$ ($p > 2$ prime), which satisfies $\mathbb{E}_\nu[X_p] = 0$, $X_p(1) = -1 + \frac{p - 1}{p} = -\frac{1}{p}$, and variance
$$\mathbb{E}_\nu[X_p^2] = \mathbb{E}_\nu[c_p^2] - \left(\frac{p - 1}{p}\right)^2 = 2\left(1 - \frac{1}{p}\right) - \left(1 - \frac{1}{p}\right)^2 = 1 - \frac{1}{p^2} = \frac{J_2(p)}{p^2}.$$
For every squarefree odd integer $\ell \ge 1$, set $X_\ell(m) := \prod_{p \mid \ell} X_p(m)$ (with $X_1(m) := 1$). By mutual independence of $\{X_p\}_{p > 2}$ under $\nu$, the family $\{X_\ell : \ell \ge 1 \text{ odd squarefree}\}$ is **mutually orthogonal in $L^2(\mathbb{N}_{\mathrm{odd}}, d\nu)$**, with
$$\mathbb{E}_\nu[X_{\ell_1} X_{\ell_2}] = \delta_{\ell_1, \ell_2} \frac{J_2(\ell_1)}{\ell_1^2}, \qquad X_\ell(1) = \frac{\mu(\ell)}{\ell}, \qquad c_\ell(m) = \prod_{p \mid \ell} \left(X_p(m) - \frac{p - 1}{p}\right) = \sum_{d \mid \ell} \mu(\ell / d)\frac{\varphi(\ell / d)}{\ell / d}\, X_d(m).$$
By Möbius inversion on the divisor lattice of each squarefree odd $\ell \ge 1$, $X_\ell(m) = \sum_{d \mid \ell} \frac{\varphi(\ell / d)}{\ell / d} c_d(m)$ (with $c_1 := 1$). Hence the squarefree odd subspace $\mathcal{S}_N^{\mathrm{sqf}} := \operatorname{span}\{c_\ell\big|_{\mathbb{N}_{\mathrm{odd}}} : 1 \le \ell \le N \text{ odd squarefree}\}$ coincides with the orthogonal span $\operatorname{span}\{X_\ell : 1 \le \ell \le N \text{ odd squarefree}\}$.
Let $F_N^*$ be the orthogonal projection of $P_{\mathcal{W}_2} E$ onto the squarefree odd subspace $\mathcal{S}_N^{\mathrm{sqf}}$. Expanding $(\nabla F_N^*)(m) = \sum_{\substack{1 \le d \le N \\ d \text{ odd}}} \mu(d)^2 \alpha_d X_d(m)$ in the orthogonal basis $\{X_d : 1 \le d \le N \text{ odd squarefree}\}$, the normal equations $\langle F_N^*, X_d \rangle_{\mathcal{S}} = \langle P_{\mathcal{W}_2} E, X_d \rangle_{\mathcal{S}}$ give $\frac{\pi^2}{16} \alpha_d \frac{J_2(d)}{d^2} = \frac{1}{2} X_d(1) = \frac{\mu(d)}{2 d}$, so $\alpha_d = \frac{8\,\mu(d)\,d}{\pi^2 J_2(d)} = \frac{8}{\pi^2} \prod_{r \mid d} \frac{-r}{r^2 - 1}$ and
$$(\nabla F_N^*)(m) = \frac{8}{\pi^2} \sum_{\substack{1 \le d \le N \\ d \text{ odd}}} \frac{\mu(d)\,d}{J_2(d)}\, X_d(m).$$
We now prove that $F_N^*$ already satisfies the full normal equations $\langle P_{\mathcal{W}_2} E - F_N^*, P_{\mathcal{W}_2}\widetilde{W}_\ell \rangle_{\mathcal{S}} = 0$ for **every** non-squarefree odd integer $3 \le \ell \le N$. Note that individual non-squarefree prime powers $c_{p^a}$ ($a \ge 2$) are **not** orthogonal to $1$ or $X_p$ in $L^2(\mathbb{N}_{\mathrm{odd}}, d\nu)$: indeed, $\mathbb{E}_\nu[c_{p^a}] = (1 - p^{-2})(-p^{a-1} p^{-2(a-1)} + \sum_{k=a}^\infty p^{a-1}(p-1)p^{-2k}) = -\frac{p - 1}{p^a} \ne 0$, and because $c_{p^a}(m)$ is non-zero only when $p^{a-1} \mid m$ (where $p \mid m$ and $X_p(m) = \frac{p^2 - 1}{p}$ is constant), $\mathbb{E}_\nu[c_{p^a} X_p] = \frac{p^2 - 1}{p}\mathbb{E}_\nu[c_{p^a}] = -\frac{(p-1)(p^2-1)}{p^{a+1}} \ne 0$ (for example, at $p^a = 9$, $\mathbb{E}_\nu[c_9] = -\frac{2}{9}$, $\mathbb{E}_\nu[c_9 c_3] = -\frac{4}{9}$, and $\mathbb{E}_\nu[c_9 X_3] = -\frac{16}{27}$). Instead, for any non-squarefree odd $\ell \le N$:
1. By (3.6a), the target pairing vanishes: $\langle P_{\mathcal{W}_2} E, P_{\mathcal{W}_2}\widetilde{W}_\ell \rangle_{\mathcal{S}} = \frac{\mu(\ell)}{2\ell} = 0$.
2. For any squarefree odd $d \ge 1$, if $d \nmid \operatorname{rad}(\ell)$, there is a prime $r \mid d$ with $r \nmid \ell$, so $X_r$ is independent of $X_{d/r} c_\ell$ under $\nu$ and $\mathbb{E}_\nu[X_d c_\ell] = \mathbb{E}_\nu[X_r]\,\mathbb{E}_\nu[X_{d/r} c_\ell] = 0$.
3. Because $\ell \le N$ is non-squarefree, $\operatorname{rad}(\ell) < \ell \le N$. Consequently, **every** divisor $d \mid \operatorname{rad}(\ell)$ satisfies $d \le \operatorname{rad}(\ell) < N$ and is therefore included in the truncated sum defining $\nabla F_N^*$! Summing over $d \mid \operatorname{rad}(\ell)$ factors $\mathbb{E}_\nu[(\nabla F_N^*) c_\ell]$ across the prime factors of $\ell$:
   $$\mathbb{E}_\nu\!\big[(\nabla F_N^*)\, c_\ell\big] = \frac{8}{\pi^2} \sum_{d \mid \operatorname{rad}(\ell)} \left(\prod_{r \mid d} \frac{-r}{r^2 - 1}\right) \mathbb{E}_\nu[X_d\, c_\ell] = \frac{8}{\pi^2} \prod_{p \mid \ell} \mathbb{E}_\nu\!\left[\left(1 - \frac{p}{p^2 - 1} X_p\right) c_{p^{\nu_p(\ell)}}\right].$$
4. Choose a prime $p \mid \ell$ with $a = \nu_p(\ell) \ge 2$. For every integer $m \ge 1$, we have $1 - \frac{p}{p^2 - 1} X_p(m) = \frac{p^2}{p^2 - 1}\mathbf{1}_{\{p \nmid m\}}$, which vanishes whenever $p \mid m$, whereas $c_{p^a}(m)$ ($a \ge 2$) vanishes whenever $p^{a-1} \nmid m$ (and in particular whenever $p \nmid m$). Because their supports are disjoint, **pointwise for every $m \ge 1$**,
   $$\left(1 - \frac{p}{p^2 - 1} X_p(m)\right) c_{p^a}(m) = 0 \qquad (\forall a \ge 2,\; \forall m \ge 1).$$
   Thus the $p$-factor in the product vanishes identically, giving $\langle F_N^*, P_{\mathcal{W}_2}\widetilde{W}_\ell \rangle_{\mathcal{S}} = \frac{\pi^2}{16\ell}\mathbb{E}_\nu[(\nabla F_N^*) c_\ell] = 0 = \langle P_{\mathcal{W}_2} E, P_{\mathcal{W}_2}\widetilde{W}_\ell \rangle_{\mathcal{S}}$ for every non-squarefree odd $\ell \le N$. Hence $F_N^* \in \mathcal{S}_N^{\mathrm{sqf}}$ is the unique orthogonal projection of $P_{\mathcal{W}_2} E$ onto the entire subspace $P_{\mathcal{W}_2}\mathcal{V}_N$, proving (5.8) for all $N \ge 2$.
Computing the squared norm of $F_N^*$ in the orthogonal basis $\{X_d\}$ gives
$$\big\|P_{\mathcal{W}_2} E\big\|_{\mathcal{S}}^2 - d_N^2(\mathcal{W}_2^{\mathcal{S}}) = \big\|F_N^*\big\|_{\mathcal{S}}^2 = \frac{\pi^2}{16} \sum_{\substack{1 \le d \le N \\ d \text{ odd}}} \mu(d)^2 \alpha_d^2\,\frac{J_2(d)}{d^2} = \frac{4}{\pi^2} \sum_{\substack{1 \le d \le N \\ d \text{ odd}}} \frac{\mu(d)^2}{J_2(d)} = \frac{q_N}{\pi^2}.$$
Subtracting from $\|P_{\mathcal{W}_2} E\|_{\mathcal{S}}^2 = \frac{1}{2}$ and noting that $\sum_{\ell \ge 1 \text{ odd}} \frac{\mu(\ell)^2}{J_2(\ell)} = \prod_{p > 2}(1 + \frac{1}{p^2 - 1}) = \prod_{p > 2}(1 - p^{-2})^{-1} = (1 - \frac{1}{4})\zeta(2) = \frac{\pi^2}{8}$ yields (5.10) and the first identity in (5.12). Substituting $X_d = \sum_{\ell \mid d} \frac{\varphi(d / \ell)}{d / \ell} c_\ell$ into $(\nabla F_N^*)(m) = \sum_{1 \le d \le N \text{ odd}} \frac{8\mu(d)d}{\pi^2 J_2(d)} X_d(m)$ and writing $d = \ell m$ gives
$$(\nabla F_N^*)(m) = \frac{8}{\pi^2}\sum_{\substack{1 \le d \le N \\ d \text{ odd}}} \frac{\mu(d)\,\varphi(d)}{J_2(d)} + \sum_{\substack{3 \le \ell \le N \\ \ell \text{ odd}}} \left(\frac{8\,\ell^2}{\pi^2} \sum_{\substack{1 \le m \le N/\ell \\ m \text{ odd}}} \frac{\mu(\ell m)\,\varphi(m)}{J_2(\ell m)}\right)\frac{c_\ell(m)}{\ell}.$$
Since $(\nabla \widetilde{W}_2)(m) = -\frac{1}{2}$ and $(\nabla \widetilde{W}_\ell)(m) = \frac{c_\ell(m)}{\ell}$ ($\ell \ge 3$ odd), matching coefficients (and using $\mu(\ell m) = \mu(\ell)\mu(m)$ and $J_2(\ell m) = J_2(\ell)J_2(m)$ for $\gcd(\ell, m) = 1$) gives both forms of $a_\ell^{(N)}$ and $c_2^{(N)}$ in (5.11).
To establish the asymptotic in (5.12)–(5.13), define the multiplicative function $h(\ell) := \mathbf{1}_{\{\ell \text{ odd}\}} \mu(\ell)^2 \frac{\ell^2}{J_2(\ell)}$, so $\frac{\mu(\ell)^2}{J_2(\ell)} = \frac{h(\ell)}{\ell^2}$ for odd $\ell$. For each odd prime $p$, $h(p) = \frac{p^2}{p^2 - 1} = 1 + \frac{1}{p^2 - 1}$ and $h(p^k) = 0$ for all $k \ge 2$ (because $\mu(p^k)^2 = 0$). Writing $h = \mathbf{1}_{\mathrm{odd}} * g$ as the Dirichlet convolution of $\mathbf{1}_{\mathrm{odd}}(m) := \mathbf{1}_{\{2 \nmid m\}}$ with the multiplicative perturbation $g := \mu_{\mathrm{odd}} * h$, the local Euler factor of $\sum_{d \ge 1} g(d) d^{-s}$ at each odd prime $p$ is
$$(1 - p^{-s})\left(1 + \frac{p^2}{p^2 - 1} p^{-s}\right) = 1 + \frac{1}{p^2 - 1} p^{-s} - \frac{p^2}{p^2 - 1} p^{-2s},$$
so $g(p) = \frac{1}{p^2 - 1}$, $g(p^2) = -\frac{p^2}{p^2 - 1}$, and $g(p^k) = 0$ for all $k \ge 3$ (while $g(2^k) = 0$ for $k \ge 1$). Thus $g$ is supported on odd cubefree integers $d = a b^2$, where $a, b \ge 1$ are odd, squarefree, and coprime ($\gcd(a, b) = 1$), with
$$|g(a b^2)| = \frac{1}{J_2(a)}\,\frac{b^2}{J_2(b)} \le \frac{\pi^4}{64\,a^2} \qquad \left(\text{since } \frac{n^2}{J_2(n)} = \prod_{p \mid n} \frac{p^2}{p^2 - 1} \le \prod_{p > 2} \frac{p^2}{p^2 - 1} = \frac{\pi^2}{8} \text{ for all odd squarefree } n \ge 1\right).$$
Summing over $b \le \sqrt{x/a}$ and $b > \sqrt{x/a}$ directly yields the sharp $O(x^{1/2})$ and $O(x^{-1/2})$ bounds without any $x^\varepsilon$ loss:
$$\sum_{\substack{1 \le d \le x \\ d \text{ odd}}} |g(d)| \le \frac{\pi^4}{64} \sum_{a=1}^\infty \frac{1}{a^2} \sum_{1 \le b \le \sqrt{x/a}} 1 \le \frac{\pi^4 \zeta(5/2)}{64}\, x^{1/2} = O(x^{1/2}), \qquad \sum_{\substack{d > x \\ d \text{ odd}}} \frac{|g(d)|}{d} \le \frac{\pi^4}{64} \sum_{a=1}^\infty \frac{1}{a^3} \sum_{b > \sqrt{x/a}} \frac{1}{b^2} = O(x^{-1/2}).$$
Consequently, the summatory function $H(x) := \sum_{1 \le \ell \le x} h(\ell)$ satisfies
$$H(x) = \sum_{\substack{1 \le d \le x \\ d \text{ odd}}} g(d) \left\lfloor \frac{x / d + 1}{2} \right\rfloor = \frac{x}{2} \sum_{\substack{d=1 \\ d \text{ odd}}}^\infty \frac{g(d)}{d} + O(x^{1/2}) = \frac{C_{\mathrm{odd}}}{2}\,x + O(x^{1/2}),$$
where at $s = 1$ the Euler product $\sum_{d \ge 1 \text{ odd}} \frac{g(d)}{d}$ evaluates to
$$\sum_{\substack{d=1 \\ d \text{ odd}}}^\infty \frac{g(d)}{d} = \prod_{p > 2 \text{ prime}} (1 - p^{-1})\left(1 + \frac{p}{p^2 - 1}\right) = \prod_{p > 2 \text{ prime}} \left(1 - \frac{1}{p(p+1)}\right) = C_{\mathrm{odd}}.$$
Applying Abel summation (integration by parts) to the tail $\sum_{\ell > N} \frac{h(\ell)}{\ell^2}$ yields
$$\sum_{\substack{\ell > N \\ \ell \text{ odd}}} \frac{\mu(\ell)^2}{J_2(\ell)} = -\frac{H(N)}{N^2} + 2 \int_N^\infty \frac{H(t)}{t^3}\,dt = -\frac{C_{\mathrm{odd}}}{2 N} + 2 \int_N^\infty \frac{C_{\mathrm{odd}}}{2 t^2}\,dt + O(N^{-3/2}) = \frac{C_{\mathrm{odd}}}{2 N} + O(N^{-3/2}).$$
Multiplying by $\frac{4}{\pi^2}$ gives $d_N^2(\mathcal{W}_2^{\mathcal{S}}) = \frac{2 C_{\mathrm{odd}}}{\pi^2 N} + O(N^{-3/2})$, proving (5.12)–(5.13). $\square$

---

## 6. Computational Verification Summary

All finite rational checks, symbolic polynomial/matrix identities, high-precision special-function comparisons, and outward-rounded interval enclosures for Propositions 1–2 and Theorems 1–4 are implemented in `src/study_wandering_projection_and_commutator_gram_algebra.py` (`Python 3.12.3`, `SymPy 1.14.0`, `mpmath 1.3.0`), with complete machine-readable output archived in `results/wandering_projection_and_commutator_gram_algebra.json` in the supplementary material (total wall time `< 11` seconds). We distinguish three verification categories: (A) exact finite rational (`Fraction` / `QQ`) and symbolic (`SymPy`) checks on bounded index ranges (including exact `DomainMatrix` normal-equation solves on the full basis $\{P_{\mathcal{W}_2}\widetilde{W}_2\} \cup \{P_{\mathcal{W}_2}\widetilde{W}_\ell : 3 \le \ell \le N \text{ odd}\}$ for $2 \le N \le 31$, confirming both $c_2^{(N)}$ and $a_\ell^{(N)}$ in (5.11)); (B) 60-digit floating-point (`mpmath`) comparisons against Hurwitz zeta and digamma special-function evaluations, together with evaluation of the closed-form Jordan totient sum (5.10) up to $N = 100{,}000$ to demonstrate numerical convergence to the asymptotic law (5.12); and (C) rigorous `mpmath.iv` interval enclosures exported with outward-rounded 30-digit `Decimal` (`ROUND_FLOOR` / `ROUND_CEILING`) and IEEE-754 `float64` (`math.nextafter`) endpoints, where the script explicitly verifies that every serialized interval strictly encloses the target constant (`lb < exact < ub`). Detailed verification tables, sample defect matrices, the exact Jordan totient sieve table for $2 \le N \le 31$, the large-$N$ closed-form asymptotic table up to $N = 100{,}000$, and 30-digit outward-rounded interval enclosures (including $C_{\mathrm{odd}}$ and $\frac{2 C_{\mathrm{odd}}}{\pi^2}$) are provided in Appendix A.

---

## Appendix A. Verification Tables, Sample Commutator Defect Matrices, and Certified Interval Enclosures

### A.1 Summary of Finite Rational, Symbolic, and Special-Function Verifications

| Proposition / Theorem / Identity | Verification Category & Scope | Cases Checked | Max Discrepancy |
| :--- | :--- | :--- | :--- |
| **Prop. 1(1):** Unnormalized $D_a^*$ & Normalized $V_a^*$ Radical-Stripping (2.1) | Exact `Fraction` ($D_a^*$) & 60-dps (`V_a^*`), $2 \le \ell \le 48, a \mid \ell, L \ge 2, 1 \le m \le 12$ + spot-check $a=2, L=3, m=1$ | $1,800$ | $0$ (`Fraction`) / $7.78 \times 10^{-62}$ (`V_a^*`) |
| **Prop. 1(2):** Canonical Coprime Adjoint Reduction (2.2)–(2.3) | Exact `Fraction`, $1 \le a \le 24$, $2 \le \ell \le 24$ ($\ell \nmid a$), $1 \le m \le 8$ | $3,936$ | $0$ (exact `Fraction`) |
| **Prop. 1(3):** Squarefree Radical Reduction (2.4) & Single-Prime Reduction (2.5) | Exact `Fraction`, $2 \le \ell \le 36, 1 \le b \le 4, 1 \le m \le 9$ + $10$ squarefree $R \le 42$ | $1,260 + 171$ | $0$ (exact `Fraction`) |
| **Prop. 1(4)(i):** Centered Adjoint $(D_R^* \widetilde{W}_R)(m) = -\frac{\varphi(R)}{2R^2} + R_R(m)$ (2.7)–(2.8) | Exact `Fraction` ($1 \le m \le 20$) + symbolic oddness $\widetilde{R}_R(-u) = -\widetilde{R}_R(u)$ & Laurent $c_1(R), c_2(R)=0, c_3(R)$ | $200 + 10$ | $0$ (exact `Fraction` & symbolic) |
| **Prop. 1(4)(ii):** Coprime Rational Branches $Q_{m,f,c}(n)$ (2.9)–(2.10) | Exact `Fraction`, $1 \le m \le 5, 2 \le L \le 8, \gcd(m,L)=1$ + $(D_3^* G_2)(n)$ ($1 \le n \le 40$) + $\widetilde{W}_4$ | $484 + 40 + 4$ | $0$ (exact `Fraction`) |
| **Prop. 2(2)–(3):** Wandering & Commutator Trichotomy (2.12)–(2.15) | Exact `Fraction`, $p \in \{2,3,5\}, q \in \{2,3,5,7\}\setminus\{p\}, 2 \le \ell \le 45, 1 \le m \le 15$ + $G_p$ ($2 \le p \ne q \le 11$) | $1,935 + 5,805 + 2,250$ | $0$ (exact `Fraction`) |
| **Theorem 1(1):** Block-by-Block Step-Space Telescoping (3.3) & Basis Pairs | Exact `Fraction`, $p \in \{2,3,4,5,7\}, 0 \le k \le 5$ + basis pairs $f = \delta_{\cdot,a}, g = \delta_{\cdot,b}$ ($1 \le a,b \le 12$) | $396 + 576$ | $0$ (exact `Fraction`) |
| **Theorem 1(2):** Wandering & Cross-Commutator Target Pairings (3.5)–(3.6c) | Exact `Fraction`, $p \in \{2,3,5,7\}, 2 \le \ell \le 40$ + $[D_p^*, D_q]E$ & $\langle [V_p^*, V_q]\widetilde{W}_\ell, E\rangle$ across $2 \le p,q \le 8, 2 \le \ell \le 15$ | $156 + 686$ | $0$ (exact `Fraction`) |
| **Theorem 1(3):** Sub-Bin Defect Matrix $M_{u,v}^{(p,q)}$ (3.10) & Basis Pairs | Exact `Fraction`, all $2 \le p, q \le 12$, $1 \le u, v \le p - 1$ + $600$ basis pairs in $\mathcal{S}$ | $5,566 + 600$ | $0$ (exact `Fraction`) |
| **Theorem 2(1)–(3):** Symmetric Frobenius Trace (4.3), Hahn Operator & Legendre Spectrum (4.4) | Exact `Fraction` trace across **all** $2 \le p, q \le 15$ (both $p \le q$ and $p > q$) + symbolic $\Delta_p\operatorname{diag}(u(p-u)) = L_{\mathrm{Hahn}} + 2I_{p-1}$ & eigenvalues | $196 + 8$ | $0$ (exact `Fraction` & symbolic) |
| **Theorem 2(4):** Cosecant-Squared & Divided-Difference Cotangent Law (4.5)–(4.7) | 60-dps `mpmath` comparison vs. Hurwitz zeta & digamma for $p \in \{2,3,5,7\}, q \in \{2,3,5,7,11\}$ ($q > p$) | $10$ | $3.89 \times 10^{-62}$ (60-dps) |
| **Theorem 3(1):** Universal $p$-Coprime Ramanujan $L(2)$ Euler Product (5.2) | 60-dps `mpmath` comparison vs. Hurwitz zeta across $p \in \{2,3,5\}$, all $2 \le \ell_1 \le \ell_2 \le 10$ | $135$ | $7.78 \times 10^{-62}$ (60-dps) |
| **Theorem 4(1)–(2):** Dyadic Jordan Totient Sieve, Normal Equations, Convolution & Asymptotic (5.8)–(5.13) | Exact `QQ` normal-equation solve for $c_2^{(N)}, a_\ell^{(N)}, q_N$ ($2 \le N \le 31$), pointwise support disjointness ($600$ checks), exact `Fraction` convolution $h = \mathbf{1}_{\mathrm{odd}} * g$ ($1 \le \ell \le 999$ odd), 30-digit `mpmath.iv` enclosure of $C_{\mathrm{odd}}$, & closed-form sum (5.10) vs. (5.12) up to $N = 100{,}000$ | $30 + 600 + 500 + 8$ | $0$ (`QQ` / `Fraction`) / $\max \sqrt{N}|N d_N^2 - \frac{2 C_{\mathrm{odd}}}{\pi^2}| = 0.0239$ |

### A.2 Centered Adjoint Laurent Coefficients, Sample Commutator Defect Matrices, and Dyadic Jordan Totient Sieve Tables

For squarefree moduli $R \in \{2, 3, 5, 6, 10, 15, 30\}$, the asymptotic mean $-\frac{\varphi(R)}{2 R^2}$ and the first three Laurent coefficients $c_1(R) u^{-1} + c_2(R) u^{-2} + c_3(R) u^{-3}$ of $\widetilde{R}_R(u) = R_R(u - \frac{1}{2})$ from Proposition 1(4)(i) are:

| Squarefree $R$ | Asymptotic Mean $-\dfrac{\varphi(R)}{2 R^2}$ | Leading Coefficient $c_1(R)$ ($u^{-1}$) | Even Coefficient $c_2(R)$ ($u^{-2}$) | Subleading Odd Coefficient $c_3(R)$ ($u^{-3}$) |
| :--- | :--- | :--- | :--- | :--- |
| $R = 2$ | $-1/8$ | $1/16$ | $0$ | $0$ (since $\widetilde{R}_2(u) = \frac{1}{16 u}$ is exact) |
| $R = 3$ | $-1/9$ | $4/81$ | $0$ | $1/729$ |
| $R = 5$ | $-2/25$ | $4/125$ | $0$ | $21/15625$ |
| $R = 6$ | $-1/36$ | $1/54$ | $0$ | $-1/1944$ |
| $R = 10$ | $-1/50$ | $3/250$ | $0$ | $-3/25000$ |
| $R = 15$ | $-4/225$ | $32/3375$ | $0$ | $136/759375$ |
| $R = 30$ | $-1/225$ | $4/1125$ | $0$ | $-7/50625$ |

For $(p, q) \in \{(3, 2), (3, 5), (5, 2), (5, 7)\}$, the exact rational matrices $M^{(p, q)}$ and $K^{(p)} - M^{(p, q)}$ from Theorem 1(3) illustrate both the diagonal defect regime ($q \ge p$, or $(3, 2)$) and the block-diagonal multi-point straddling regime ($p > q$, such as $(5, 2)$ where $u \in \{1, 2\}$ share $\lfloor 2u/5 \rfloor = 0$ and $u \in \{3, 4\}$ share $\lfloor 2u/5 \rfloor = 1$):
$$M^{(3, 2)} = \begin{pmatrix} 1/3 & 1/3 \\ 1/3 & 1/3 \end{pmatrix}, \quad K^{(3)} - M^{(3, 2)} = \begin{pmatrix} 1/3 & 0 \\ 0 & 1/3 \end{pmatrix}; \qquad M^{(3, 5)} = \begin{pmatrix} 8/15 & 1/3 \\ 1/3 & 8/15 \end{pmatrix}, \quad K^{(3)} - M^{(3, 5)} = \begin{pmatrix} 2/15 & 0 \\ 0 & 2/15 \end{pmatrix};$$
$$K^{(5)} - M^{(5, 2)} = \begin{pmatrix} 3/5 & 1/5 & 0 & 0 \\ 1/5 & 2/5 & 0 & 0 \\ 0 & 0 & 2/5 & 1/5 \\ 0 & 0 & 1/5 & 3/5 \end{pmatrix}, \quad \operatorname{Tr}\!\big(\Delta_5 M^{(5, 2)}\big) = 4 - \frac{16}{5} = \frac{4}{5}; \qquad K^{(5)} - M^{(5, 7)} = \operatorname{diag}\!\left(\frac{6}{35}, \frac{4}{35}, \frac{4}{35}, \frac{6}{35}\right).$$

For the Dyadic Jordan Totient Sieve (Theorem 4), the exact rational coefficient $q_N = 4 \sum_{1 \le \ell \le N \text{ odd}} \frac{\mu(\ell)^2}{J_2(\ell)}$ and the resulting dyadic wandering distance $d_N^2(\mathcal{W}_2^{\mathcal{S}}) = \frac{1}{2} - \frac{q_N}{\pi^2}$ verified by direct `QQ` normal-equation solve on $\{P_{\mathcal{W}_2}\widetilde{W}_2\} \cup \{P_{\mathcal{W}_2}\widetilde{W}_\ell : 3 \le \ell \le N \text{ odd}\}$ at representative cutoffs $N \in \{2, 3, 5, 7, 9, 11, 15, 21, 30, 31\}$ (noting that at $N = 3$, $(c_2^{(3)}, a_3^{(3)}) = \pi^{-2}(-12, -9)$, and that $q_9 = q_7 = \frac{19}{4}$ because $\ell = 9$ is non-squarefree with $a_9^{(N)} = 0$, whereas $\ell = 15 = 3 \times 5$ contributes $\frac{4}{J_2(15)} = \frac{4}{192} = \frac{1}{48}$), together with the large-$N$ evaluation of the closed-form Jordan sum (5.10) demonstrating convergence of $N \cdot d_N^2(\mathcal{W}_2^{\mathcal{S}})$ to $\frac{2 C_{\mathrm{odd}}}{\pi^2} \approx 0.1712998023$ up to $N = 100{,}000$, are:

| Cutoff $N$ | Active Squarefree Odd $\ell \le N$ | Exact Rational $q_N = 4 \sum_{\substack{1 \le \ell \le N \\ \ell \text{ odd}}} \dfrac{\mu(\ell)^2}{J_2(\ell)}$ | Exact Formula for $d_N^2(\mathcal{W}_2^{\mathcal{S}})$ | Numerical Value of $d_N^2(\mathcal{W}_2^{\mathcal{S}})$ | Scaled $N \cdot d_N^2(\mathcal{W}_2^{\mathcal{S}})$ (vs. $\dfrac{2 C_{\mathrm{odd}}}{\pi^2} \approx 0.171300$) |
| :--- | :--- | :--- | :--- | :--- | :--- |
| $N = 2$ | $\{1\}$ | $4$ | $\dfrac{1}{2} - \dfrac{4}{\pi^2}$ | $0.0947152654$ | $0.189431$ |
| $N = 3$ | $\{1, 3\}$ | $9/2$ | $\dfrac{1}{2} - \dfrac{9}{2\pi^2}$ | $0.0440546736$ | $0.132164$ |
| $N = 5$ | $\{1, 3, 5\}$ | $14/3$ | $\dfrac{1}{2} - \dfrac{14}{3\pi^2}$ | $0.0271678097$ | $0.135839$ |
| $N = 7$ (and $N = 9$) | $\{1, 3, 5, 7\}$ | $19/4$ | $\dfrac{1}{2} - \dfrac{19}{4\pi^2}$ | $0.0187243777$ | $0.131071$ ($N=7$) / $0.168519$ ($N=9$) |
| $N = 11$ | $\{1, 3, 5, 7, 11\}$ | $287/60$ | $\dfrac{1}{2} - \dfrac{287}{60\pi^2}$ | $0.0153470049$ | $0.168817$ |
| $N = 15$ | $\{1, 3, 5, 7, 11, 13, 15\}$ | $8111/1680$ | $\dfrac{1}{2} - \dfrac{8111}{1680\pi^2}$ | $0.0108237378$ | $0.162356$ |
| $N = 21$ | $\{1, \dots, 19, 21\}$ | $5447/1120$ | $\dfrac{1}{2} - \dfrac{5447}{1120\pi^2}$ | $0.0072352792$ | $0.151941$ |
| $N = 30$ | $\{1, \dots, 29\}$ | $60069/12320$ | $\dfrac{1}{2} - \dfrac{60069}{12320\pi^2}$ | $0.0059852126$ | $0.179556$ |
| $N = 31$ | $\{1, \dots, 29, 31\}$ | $180361/36960$ | $\dfrac{1}{2} - \dfrac{180361}{36960\pi^2}$ | $0.0055630410$ | $0.172454$ |
| $N = 100$ | $41$ odd squarefree $\ell \le 100$ | (closed-form sum (5.10)) | $\dfrac{4}{\pi^2}\sum_{\ell > 100 \text{ odd}} \dfrac{\mu(\ell)^2}{J_2(\ell)}$ | $0.0017195162$ | $0.171952$ ($\sqrt{N}|\text{err}| = 0.00652$) |
| $N = 1{,}000$ | $406$ odd squarefree $\ell \le 10^3$ | (closed-form sum (5.10)) | $\dfrac{4}{\pi^2}\sum_{\ell > 10^3 \text{ odd}} \dfrac{\mu(\ell)^2}{J_2(\ell)}$ | $1.720559 \times 10^{-4}$ | $0.172056$ ($\sqrt{N}|\text{err}| = 0.02391$) |
| $N = 10{,}000$ | $4{,}053$ odd squarefree $\ell \le 10^4$ | (closed-form sum (5.10)) | $\dfrac{4}{\pi^2}\sum_{\ell > 10^4 \text{ odd}} \dfrac{\mu(\ell)^2}{J_2(\ell)}$ | $1.711815 \times 10^{-5}$ | $0.171182$ ($\sqrt{N}|\text{err}| = 0.01183$) |
| $N = 100{,}000$ | $40{,}532$ odd squarefree $\ell \le 10^5$ | (closed-form sum (5.10)) | $\dfrac{4}{\pi^2}\sum_{\ell > 10^5 \text{ odd}} \dfrac{\mu(\ell)^2}{J_2(\ell)}$ | $1.713122 \times 10^{-6}$ | $0.171312$ ($\sqrt{N}|\text{err}| = 0.00393$) |

### A.3 Outward-Rounded 30-Digit Interval Enclosures

Evaluating the closed-form formulas of Theorem 2(4), Theorem 3(2)–(3), and the Witt-accelerated Euler product for $C_{\mathrm{odd}}$ and $\frac{2 C_{\mathrm{odd}}}{\pi^2}$ in Theorem 4(2) in `mpmath.iv` interval arithmetic and converting the exact binary rational endpoints of each `mpmath.iv` interval to 30-significant-digit `Decimal` strings using `ROUND_FLOOR` for lower bounds and `ROUND_CEILING` for upper bounds (recorded under `iv_certifications` and `theorem_4_jordan_totient_sieve_bound` in `results/wandering_projection_and_commutator_gram_algebra.json`) yields the following certified enclosures:

| Cocycle / Constant | Exact Formula for $\|(I - V_p V_p^*) f\|_{\mathcal{S}}^2$ or Constant | Outward-Rounded 30-Digit Enclosure | Exact Formula for Commutator Norm Squared or Scaled Limit | Outward-Rounded 30-Digit Enclosure |
| :--- | :--- | :--- | :--- | :--- |
| $p=2,\; \widetilde{W}_2 = G_2$ | $\dfrac{\pi^2}{64}$ | $[0.154212568767021228419288921873,\; 0.154212568767021228419288921874]$ | $\dfrac{\pi^2}{96}$ | $[0.102808379178014152279525947915,\; 0.102808379178014152279525947916]$ |
| $p=2,\; \widetilde{W}_3 = G_3$ | $\dfrac{\pi^2}{108}$ | $[0.0913852259360125798040230648136,\; 0.0913852259360125798040230648137]$ | $\dfrac{\pi^2}{162}$ | $[0.0609234839573417198693487098757,\; 0.0609234839573417198693487098758]$ |
| $p=2,\; \widetilde{W}_5 = G_5$ | $\dfrac{\pi^2}{250}$ | $[0.0394784176043574344753379639995,\; 0.0394784176043574344753379639996]$ | $\dfrac{\pi^2}{375}$ | $[0.0263189450695716229835586426663,\; 0.0263189450695716229835586426664]$ |
| $p=2,\; \widetilde{W}_6 = H_{2,3}$ | $\dfrac{\pi^2}{432}$ | $[0.0228463064840031449510057662034,\; 0.0228463064840031449510057662035]$ | $\dfrac{\pi^2}{648}$ | $[0.0152308709893354299673371774689,\; 0.0152308709893354299673371774690]$ |
| $p=2,\; \widetilde{W}_{10} = H_{2,5}$ | $\dfrac{\pi^2}{1000}$ | $[0.00986960440108935861883449099987,\; 0.00986960440108935861883449099988]$ | $\dfrac{\pi^2}{1500}$ | $[0.00657973626739290574588966066658,\; 0.00657973626739290574588966066659]$ |
| $p=2,\; \widetilde{W}_{15} = H_{3,5}$ | $\dfrac{2\pi^2}{3375}$ | $[0.00584865445990480510745747614807,\; 0.00584865445990480510745747614808]$ | $\dfrac{4\pi^2}{10125}$ | $[0.00389910297326987007163831743204,\; 0.00389910297326987007163831743205]$ |
| $p=2,\; \widetilde{W}_{30}$ | $\dfrac{\pi^2}{6750}$ | $[0.00146216361497620127686436903701,\; 0.00146216361497620127686436903702]$ | $\dfrac{\pi^2}{10125}$ | $[0.000974775743317467517909579358012,\; 0.000974775743317467517909579358013]$ |
| $p=3,\; \widetilde{W}_3 = G_3$ | $\dfrac{8\pi^2}{729} + \dfrac{2\pi}{81\sqrt{3}}$ | $[0.153093585411427695831782354782,\; 0.153093585411427695831782354783]$ | $\dfrac{4\pi^2}{729} + \dfrac{2\pi}{81\sqrt{3}}$ | $[0.0989393774493461670590279460042,\; 0.0989393774493461670590279460043]$ |
| Theorem 4 Constant $C_{\mathrm{odd}}$ | $\prod_{p > 2}\left(1 - \dfrac{1}{p(p+1)}\right)$ | $[0.845330641198998711283924020391,\; 0.845330641198998711283924020392]$ | $\dfrac{2\,C_{\mathrm{odd}}}{\pi^2}$ | $[0.171299802270837776087506283032,\; 0.171299802270837776087506283033]$ |

---

## References

1. `@halasz_resonator`, *Base-$p$ Digit Toeplitz Operators, Exact Szegő and KMS Wiener–Hopf Resolvents, and the Zero-at-1 Sieve Bridge in the Báez-Duarte Step Space*, rStation Paper `riemann_hypothesis_61` v1 (2026).
2. `@span_of_dilates`, *The Báez-Duarte subspace has codimension equal to the number of zeta zeros off the critical line, with distance and rate bounds not assuming RH*, rStation Paper `riemann_hypothesis_1` v11 (2026).
3. A. Beurling, *A closure problem related to the Riemann zeta-function*, Proc. Natl. Acad. Sci. USA **41** (1955), 312–314.
4. 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.
5. B. Bagchi, *On Nyman, Beurling and Baez-Duarte's Hilbert space reformulation of the Riemann hypothesis*, Proc. Indian Acad. Sci. (Math. Sci.) **116** (2006), 137–146.
6. U. Grenander and G. Szegő, *Toeplitz Forms and Their Applications*, University of California Press, Berkeley, 1958.
7. A. Böttcher and B. Silbermann, *Analysis of Toeplitz Operators*, 2nd ed., Springer, Berlin, 2006.
8. R. G. Douglas, *Banach Algebra Techniques in Operator Theory*, 2nd ed., Springer, New York, 1998.
9. M. Kac, W. L. Murdock, and G. Szegő, *On the eigenvalues of certain Hermitian forms*, J. Rational Mech. Anal. **2** (1953), 767–800.
10. W. F. Trench, *Explicit inversion formulas for Toeplitz band matrices*, SIAM J. Algebraic Discrete Methods **6** (1985), 546–554.
11. S. Bettin, J. B. Conrey, and D. W. Farmer, *An optimal choice of Dirichlet polynomials for the Nyman–Beurling criterion*, Proc. Steklov Inst. Math. **282** (2013), 2–6.
12. J.-F. Burnol, *On Fourier and Zeta(s)*, Forum Math. **16** (2004), 789–840.
13. H. L. Montgomery and R. C. Vaughan, *Hilbert's Inequality*, J. London Math. Soc. (2) **8** (1974), 73–82.
14. `@halasz_resonator`, *Step-Space Restitution, Exact Finite-Block Floor-Dilation Compression, and $O(N^2)$ Hurwitz–Bernoulli Gram Construction for the Nyman–Beurling Problem*, rStation Paper `riemann_hypothesis_21` v2 (2026).
15. `@halasz_resonator`, *Closed-Form Digamma-Cotangent Gram Theory, Exact Rational Certificates for Full vs. Squarefree Nyman–Beurling Distance ($d_N^2 < d_N^{\mathrm{sqf},2}$), and Boundary-Deflation Factorization*, rStation Paper `riemann_hypothesis_29` v1 (2026).
16. `@halasz_resonator`, *Commutative Dilation-Operator Euler Products, Totient–Cotangent Gram Decompositions, Parabolic Shift Separation, and Prime-Power Coprime-Additivity Obstructions in Nyman–Beurling Space*, rStation Paper `riemann_hypothesis_56` v1 (2026).
17. `@bramble_trace`, *Dirichlet multiplier synthesis and common eigenfunctionals in the Báez-Duarte space*, rStation Paper `riemann_hypothesis_59` v1 (2026).
18. C. Calderaro, J. Manzur, S. W. Noor, and C. F. Santos, *Orthogonality questions in the Hardy space related to $\zeta$-zeros*, J. Math. Anal. Appl. **531** (2024), 127866; `arXiv:2203.05030v4`.
19. G. Szegő, *Orthogonal Polynomials*, 4th ed., American Mathematical Society, Providence, RI, 1975.
20. R. Koekoek, P. A. Lesky, and R. F. Swarttouw, *Hypergeometric Orthogonal Polynomials and Their $q$-Analogues*, Springer, Berlin, 2010.
21. `@halasz_resonator`, *Simultaneous Cyclotomic-Shell Orthogonality, Cumulative Bernoulli Non-Positivity, Coprime Farey–Bernoulli Reciprocity, and Double-Zero Jordan–Totient Mellin Theory in Nyman–Beurling Space*, rStation Paper `riemann_hypothesis_57` v1 (2026).
22. J. Manzur, S. W. Noor, and C. F. Santos, *A weighted composition semigroup related to three open problems*, J. Math. Anal. Appl. **525** (2023), 127261; `arXiv:2111.05398`.
23. L. Báez-Duarte, M. Balazard, B. Landreau, and É. Saias, *Sur l'autocorrélation multiplicative de la fonction « partie fractionnaire »*, `arXiv:math/0306251` (2003); companion to *Étude de l'autocorrélation multiplicative de la fonction « partie fractionnaire »*, Ramanujan J. **9** (2005), 215–240.
24. S. W. Noor, *A Hardy space analysis of the Báez-Duarte criterion for the RH*, Adv. Math. **350** (2019), 242–255.
25. `@halasz_resonator`, *Multi-Prime Bernoulli–Stieltjes Asymptotics, Occupancy-Weighted Bessel Restoration, Certified Tikhonov–Schur Tail Bounds, and Two-Ray Prime-Power Friedrichs Degeneration in the Báez-Duarte Step Space*, rStation Paper `riemann_hypothesis_64` v1 (2026).
26. F. W. J. Olver, A. B. Olde Daalhuis, D. W. Lozier, B. I. Schneider, R. F. Boisvert, C. W. Clark, B. R. Miller, B. V. Saunders, H. S. Cohl, and M. A. McClain (eds.), *NIST Digital Library of Mathematical Functions*, §18.22(ii) (Hahn Class: Difference Equations), Release 1.2.0.
