Independent Mathematical Peer-Review & Formal Evaluation: Singularity Taxonomy38 v2.5 Working Monograph

Executive Summary of Findings

Evaluating Singularity Taxonomy38 v2.5: Real-Time Physical-Acoustics State Estimation, Low-Rank Legendre Aesthetic Manifolds, and Lyapunov-Stable Closed-Loop Multi-Plugin Control strictly on its pure and applied mathematical merits, my overarching conclusion is that this is a remarkably substantive, mathematically serious piece of applied analysis and control theory. Unlike typical industry whitepapers that dress up heuristic rules in decorative notation, the core theorems in this monograph are grounded in genuine structural properties of differential equations, harmonic analysis, orthogonal polynomials, state-variable realization theory, and Lyapunov stability.

At the same time, a rigorous line-by-line audit reveals several genuine mathematical strengths, alongside six specific analytical discrepancies, unstated regularity/boundary assumptions, and minor errata that should be corrected as the manuscript transitions from its v2.5 Working Paper draft to a formal archival publication.

Summary Scorecard by Mathematical Domain

Detailed Mathematical Audit & Evaluation by Section

1. Section 2.1: Gabor Uncertainty, Biquad Notch Ringing, and the Causal Discrete TKEO

At $f_s = 48\text{ kHz}$, an $N = 4096$ sample window has duration $T_{\text{win}} = \frac{4096}{48000} = 85.333\text{ ms}$ and hop $T_{\text{hop}} = \frac{1024}{48000} = 21.333\text{ ms}$. At $128\text{ BPM}$, a 16th-note inter-onset interval is:

Thus, any $85.33\text{ ms}$ frame intersecting a $T_g = 15\text{ ms}$ unvoiced ghost scratch must overlap at least $T_{\text{win}} - T_g = 70.33\text{ ms}$ of surrounding material (or $\ge 48.9\text{ ms}$ of the adjacent $75\text{ ms}$ voiced triad after accounting for inter-note transitions).

• Mathematical Critique (Unstated Relative-Power & Window-Localization Hypothesis):

In Theorem 2.1, the paper states that $\text{HNR}_v = 16\text{ dB}$ on $T_v$ and $\text{HNR}_g = -4\text{ dB}$ on $T_g$ imply:

$$\widehat{\text{HNR}}_{\text{STFT}} = 10\log_{10}\!\left(\frac{\int_{T_v} w^2(t) P_{\text{harm}}(t)\,dt}{\int_{T_g} w^2(t) P_{\text{noise}}(t)\,dt + \int_{T_v} w^2(t) P_{\text{noise}}(t)\,dt}\right) > +8.2\text{ dB}$$

Strictly speaking, $\text{HNR}_v$ and $\text{HNR}_g$ are dimensionless intra-segment ratios; they do not by themselves determine the cross-segment ratio unless one also specifies:

1. The relative broadband power ratio $P_{\text{tot},v} / P_{\text{tot},g}$ between the voiced triad and the dead-string scratch, and
2. The alignment of the ghost scratch relative to the Hann window peak $w(t) = \frac{1}{2}\big(1 + \cos(2\pi t / T_{\text{win}})\big)$.

Because $w^2(t)$ concentrates its $L^2$ mass near the window center, a $15\text{ ms}$ interval $[-7.5\text{ ms}, +7.5\text{ ms}]$ centered at $t = 0$ actually carries $\approx 46.2\%$ of the total $\int_{-T_{\text{win}}/2}^{T_{\text{win}}/2} w^2(t)\,dt$ weight despite occupying only $17.6\%$ of the time support. Consequently, the $+8.2\text{ dB}$ lower bound holds under the physically realistic hypothesis that the voiced triad has higher RMS power than the dead-string scratch (typically $+6\text{ to }+10\text{ dB}$) or when the $21.33\text{ ms}$ hop grid straddles the onset. That relative-level hypothesis should be stated explicitly in Theorem 2.1.

For the continuous-time second-order surgical notch filter:

taking the inverse Laplace transform of $Y_{\text{ring}}(s)$ via $\frac{s}{(s+\alpha)^2+\omega_d^2} = \frac{s+\alpha}{(s+\alpha)^2+\omega_d^2} - \frac{\alpha}{\omega_d}\frac{\omega_d}{(s+\alpha)^2+\omega_d^2}$ with $\alpha = \frac{\omega_0}{2Q}$ yields:

where $\phi = \arcsin\!\big(\frac{1}{2Q}\big) - \frac{\pi}{2}$. The closed-form impulse response in Theorem 2.1 is 100% exact.

• Mathematical Erratum Caught ($T_{60}$ vs. $T_{120}$ Constant):

In the next equation, the paper writes:

$$T_{60,\text{ring}} = \frac{6\ln(10)\,Q}{\pi f_0} \approx \frac{4.40\,Q}{f_0}$$

This is off by a factor of $2$ (it actually computes $T_{120}$). By definition, $T_{60}$ is the time required for the amplitude envelope $e^{-\frac{\omega_0}{2Q}t}$ to decay by $-60\text{ dB}$, which is an amplitude ratio of $10^{-60/20} = 10^{-3}$ (and a power ratio of $10^{-6}$). Solving $e^{-\frac{\pi f_0}{Q} T_{60}} = 10^{-3}$ gives:

$$T_{60,\text{ring}} = \frac{3\ln(10)\,Q}{\pi f_0} \approx \frac{2.1988\,Q}{f_0} \approx \frac{2.20\,Q}{f_0}$$

For $Q = 22$ and $f_0 = 2410\text{ Hz}$, the true $T_{60}$ is $20.07\text{ ms}$ (and $T_{120} = 40.14\text{ ms}$, $T_{30} = 10.04\text{ ms}$). Furthermore, in the Abstract (Item 2), the paper writes $T_{60} \approx \frac{2.2 Q}{\pi f_0}$, which inadvertently left $\pi$ in the denominator after already dividing $3\ln(10) \approx 6.908$ by $\pi$ to get $2.20$. Both do not change the qualitative engineering conclusion ($20\text{ ms}$ of metallic ringing still severely smears a $2\text{ ms}$ pick transient, and the dynamic $Q$-damping law $Q_{\text{inst}}[n] = Q_{\text{solved}}/(1 + 3.5\eta_{\text{chirp}}[n])$ reduces the decay time by $4.5\times$), but the factor of $2$ in Theorem 2.1 and the extra $\pi$ in the Abstract should be fixed.

Shifting the discrete Teager-Kaiser Energy Operator backward by one sample:

makes the operator strictly causal at sample $n$ with a 1-sample group delay ($20.83\ \mu\text{s}$ at $48\text{ kHz}$). For a locally monochromatic discrete sinusoid $y_k[n] = A_k \cos(\Omega_k n + \theta_k)$, the product-to-sum identity gives:

so $\Psi_{\text{TK}}^{(k)}[n-1] = A_k^2\sin^2\Omega_k \approx A_k^2\Omega_k^2$ for $\Omega_k \ll \pi/2$ is exact. Moreover, pre-filtering $x[n]$ through the three narrowband Cytomic ZDF-SVF bandpass filters ($118\text{ Hz}, 520\text{ Hz}, 4100\text{ Hz}$) prior to evaluating $\Psi_{\text{TK}}$ is mathematically crucial: because $\Psi_{\text{TK}}$ is a bilinear/quadratic form, applying it directly to a multi-component signal $y_1 + y_2$ generates oscillatory cross-terms $2y_1[n-1]y_2[n-1] - y_1[n-2]y_2[n] - y_1[n]y_2[n-2]$, which bandpass decomposition suppresses.

2. Section 2.2: The Nyquist-Lag Limit & State-Variable Quadrature (Theorems 2.2 & 2.3)

• YIN Lag Bound: At $f_{s,\text{dec}} = 6000\text{ Hz}$ with buffer size $N_{\text{buf}} = 128$ ($21.33\text{ ms}$) and half-window integration support $W_{\text{int}} = N_{\text{buf}}/2 = 64$, the maximum lag is $\tau_{\max} = 64$ samples, yielding $f_{0,\min} = \frac{6000}{64} = 93.75\text{ Hz}$ exactly. Resolving $28\text{ Hz}$ requires $\tau_{\max} = \lceil 6000/28\rceil = 215 \implies N_{\text{buf}} = 430$ samples ($71.67\text{ ms}$). Exact.
• Chebyshev $T_2(x)$ Non-Homogeneity: The identity $T_2(A(t)\cos(\omega_0 t)) = 2A^2(t)\cos^2(\omega_0 t) - 1 = A^2(t)\cos(2\omega_0 t) + (A^2(t) - 1)$ exposes a classic mathematical pitfall in audio waveshaping: Chebyshev polynomials $T_n(\cos\theta) = \cos(n\theta)$ are only orthogonal harmonic generators on the unit-amplitude circle $A(t) \equiv 1$. Because $T_2(x)$ is inhomogeneous of degree $2$ and $0$, any envelope modulation $A(t)$ injects a time-varying baseband signal $A^2(t) - 1$ (DC offset $+ \text{low-frequency thump}$) and squares the envelope ($2\times$ expansion in dB). 100% rigorous.

Theorem 2.3 is one of the strongest mathematical contributions of the paper:

1. Why $-90.0000^\circ$ Quadrature Holds Exactly in Discrete Time:

In a trapezoidal (bilinear-transformed) Cytomic State-Variable Filter, the state outputs $V_{\text{HP}}, V_{\text{BP}}, V_{\text{LP}}$ share the identical feedback denominator $D(z)$ and are linked by cascaded trapezoidal integrators:

$$V_{\text{BP}}(z) = H_{\text{int}}(z)\,V_{\text{HP}}(z), \qquad V_{\text{LP}}(z) = H_{\text{int}}(z)\,V_{\text{BP}}(z), \qquad H_{\text{int}}(z) = g\,\frac{z + 1}{z - 1}, \quad g = \tan\!\left(\frac{\pi f_c}{f_s}\right)$$

Evaluating $H_{\text{int}}(z)$ on the unit circle $z = e^{j\Omega}$ gives:

$$H_{\text{int}}\!\left(e^{j\Omega}\right) = g\,\frac{e^{j\Omega/2} + e^{-j\Omega/2}}{e^{j\Omega/2} - e^{-j\Omega/2}} = -j\,\frac{\tan(\pi f_c / f_s)}{\tan(\Omega / 2)}$$

Because $-j\cot(\Omega/2)$ has zero real part for all $\Omega \in (0, \pi)$, the phase offset between $V_{\text{LP}}(e^{j\Omega})$ and $V_{\text{BP}}(e^{j\Omega})$ is identically $-\frac{\pi}{2}$ ($-90.0000^\circ$) at all frequencies below Nyquist, and their differential group delay is identically $\Delta \tau_g(\Omega) \equiv 0$!

2. Frequency Warping & Group Delay Check:

• In discrete time, the exact ratio is $\frac{|V_{\text{HP}}(e^{j\Omega})|}{|V_{\text{LP}}(e^{j\Omega})|} = \left(\frac{\tan(\Omega/2)}{\tan(\pi f_c / f_s)}\right)^2$. Using the continuous-time algebraic estimator $\hat{f}_{\text{SVF}} = f_c\big(|V_{\text{HP}}|/|V_{\text{LP}}|\big)^{1/2}$ invokes the small-angle approximation $\tan(\pi f / f_s) \approx \pi f / f_s$. For sub-bass fundamentals $f \in [28, 95]\text{ Hz}$ at $f_s = 48\text{ kHz}$, $\pi f / f_s \le 0.00622\text{ rad}$, so the cubic Taylor remainder $\frac{1}{3}(\pi f/f_s)^2$ introduces a relative error of less than $0.0013\%$ ($< 0.02\text{ cents}$).
• Minor Numerical Note on Line 822: For a 2nd-order denominator $D(s) = s^2 + \frac{\omega_c}{Q}s + \omega_c^2$ with $f_c = 52\text{ Hz}$ and $Q = 0.55$ ($k = 1/Q \approx 1.818$), the exact group delay $\tau_g(\omega) = \frac{(\omega_c/Q)(\omega_c^2 + \omega^2)}{(\omega_c^2 - \omega^2)^2 + (\omega_c\omega/Q)^2}$ at $f = 30\text{ Hz}$ evaluates to $4.85\text{ ms}$ (whereas $2.84\text{ ms}$ corresponds to $Q \approx 0.95$ or a higher cutoff $f_c \approx 88\text{ Hz}$). Note that the differential group delay between $V_{\text{LP}}$ and $V_{\text{BP}}$ remains identically $0.0000\text{ ms}$ regardless of $f_c$ and $Q$.

1. Homogeneous Degree-1 Quadrature Harmonic Synthesis ($h_2, h_3$):

With $u_I[n] = v_{\text{BP}}[n] = A_0[n]\cos\phi[n]$ and $u_Q[n] = \frac{\hat{f}_0[n]}{f_c}v_{\text{LP}}[n] = A_0[n]\sin\phi[n]$:

$$h_2[n] = A_0[n]\left(\frac{u_I^2[n] - u_Q^2[n]}{A_0^2[n] + \epsilon}\right) = A_0[n]\cos(2\phi[n]), \qquad h_3[n] = A_0[n]\left(\frac{u_I[n](u_I^2[n] - 3u_Q^2[n])}{A_0^3[n] + \epsilon}\right) = A_0[n]\cos(3\phi[n])$$

From the viewpoint of projective geometry and harmonic analysis, these rational expressions are the real parts of the degree-1 homogeneous spherical harmonics $\text{Re}\big\{(u_I + j u_Q)^k / (u_I^2 + u_Q^2)^{(k-1)/2}\big\}$. Because they are homogeneous of degree $1$ ($\mathbf{u} \mapsto \lambda\mathbf{u} \implies h_k \mapsto \lambda h_k$), they preserve exact linear envelope scaling and possess identically zero constant/DC term.

3. Section 3: Nonlinear Membrane Mechanics & Smooth Gain Modulation (Theorems 3.1 & 3.2)

• PDE & Eigenmode Audit: Berger's large-deflection equation:
$$\rho h \frac{\partial^2 w}{\partial t^2} - \left(T_0 + \frac{E h}{2\pi a^2(1 - \nu^2)} \int_0^{2\pi}\!\!\int_0^a \left[\left(\frac{\partial w}{\partial r}\right)^2 + \frac{1}{r^2}\left(\frac{\partial w}{\partial \theta}\right)^2\right] r\,dr\,d\theta\right) \nabla^2 w + 2\rho h \gamma \frac{\partial w}{\partial t} = P_{\text{shell}}(r,\theta,t)$$

is dimensionally consistent ($\text{N/m}^2$ in every term), and the cited roots of $J_m(\alpha_{mn}) = 0$ ($\alpha_{01} = 2.4048$, $\alpha_{11} = 3.8317 = 1.593\,\alpha_{01}$, $\alpha_{21} = 5.1356 = 2.135\,\alpha_{01}$) are exact to 4 significant digits. Projecting onto the fundamental Bessel mode $J_0(\alpha_{01} r/a)$ yields a damped cubic Duffing oscillator $\ddot{q}_{01} + 2\gamma\dot{q}_{01} + \omega_0^2 q_{01} + \beta q_{01}^3 = 0$, whose cycle-averaged instantaneous frequency via Krylov-Bogoliubov / multiple-scales perturbation theory relaxes exponentially as $f_{\text{inst}}(t) = f_\infty + \Delta f_{\text{glide}}\,e^{-(t-t_1)/\tau_{\text{glide}}}$.

• 4th-Order SVF Mode Isolation: Cascading two 2nd-order Butterworth lowpass stages at $f_{\text{LPF}} = 1.38\,\hat{f}_{0,\text{prev}}$ produces an attenuation of $10\log_{10}\!\big(1 + (1.593/1.38)^8\big) = 6.21\text{ dB}$ relative to $f_{\text{LPF}}$ (or, combined with the natural acoustic roll-off of the center-struck $(1,1)$ mode and bandpass skirts, $>22\text{ dB}$ total suppression).
• 3-Point Prony / Aitken $\Delta^2$ Extrapolation: For $\bar{f}_k = f_\infty + \Delta f_{\text{glide}}\,r^{k-1}$ ($k=1,2,3$) at spacing $\Delta t = 7\text{ ms}$, the ratio of first differences is $r = \frac{\bar{f}_2 - \bar{f}_3}{\bar{f}_1 - \bar{f}_2} = e^{-\Delta t / \tau_{\text{glide}}}$, and eliminating $\Delta f_{\text{glide}}$ yields:
$$\hat{f}_{\infty,\text{Prony}} = \bar{f}_3 - \frac{r}{1 - r}\big(\bar{f}_2 - \bar{f}_3\big)$$

which is algebraically identical to Aitken's $\Delta^2$ sequence acceleration $\bar{f}_1 - \frac{(\bar{f}_2 - \bar{f}_1)^2}{\bar{f}_3 - 2\bar{f}_2 + \bar{f}_1}$. Clamping $r_{\text{clamped}} \in [0.18, 0.82]$ prevents the singularity at $r = 1$ and corresponds to:

$$\tau_{\text{glide},\min} = \frac{-7\text{ ms}}{\ln(0.18)} = 4.082\text{ ms} \approx 4.1\text{ ms}, \qquad \tau_{\text{glide},\max} = \frac{-7\text{ ms}}{\ln(0.82)} = 35.27\text{ ms} \approx 35.3\text{ ms}$$

Both bounds stated in the paper are 100% exact.

• For the hard-held gate $g(t)$ with a corner discontinuity $\Delta \dot{g} = -\frac{1-g_\infty}{\tau_{\text{atk}}}$ at $t = T_{\text{hold}}$, the distributional second derivative $\ddot{g}(t)$ contains a Dirac delta, so integration by parts gives $|\mathcal{F}\{g\}(\omega)| \propto \frac{1-g_\infty}{\tau_{\text{atk}}\,\omega^2}$.
• Replacing the step with the 5th-degree Hermite polynomial $\psi_{C^2}(u) = 10u^3 - 15u^4 + 6u^5$ on $u \in [0, 1]$ gives $\psi'_{C^2}(u) = 30u^2(1-u)^2$ and $\psi''_{C^2}(u) = 60u(1-u)(1-2u)$, which vanish at both endpoints $u \in \{0, 1\}$. Because the third derivative $\psi'''_{C^2}(u) = 60(1 - 6u + 6u^2)$ has finite jump discontinuities of $\pm 60/T_{\text{ramp}}^3$ at $u = 0, 1$, the fourth distributional derivative contains Dirac deltas, yielding the exact asymptotic decay:
$$\big|\mathcal{F}\{\psi_{C^2}\}(\omega)\big| = \mathcal{O}\!\left(\frac{1}{\omega^4 T_{\text{ramp}}^3}\right) \quad \text{as } \omega \to \infty$$
• Bonus Mathematical Insight (The Leibniz Rule at the Zero-Crossing $t_{\text{ZC,1}}$):

The paper notes that latching the ramp start to the zero crossing $x_{\text{low}}(t_{\text{ZC,1}}) = 0$ suppresses splatter even further. In fact, we can prove something stronger: by the Leibniz rule for the 3rd derivative of the modulated signal $y(t) = g(t)x_{\text{low}}(t)$ at $t = t_{\text{ZC,1}}^+$:

$$\Delta \dddot{y}\big|_{t_{\text{ZC,1}}} = \dddot{g}\big(t_{\text{ZC,1}}^+\big)\underbrace{x_{\text{low}}(t_{\text{ZC,1}})}_{= 0} + 3\underbrace{\ddot{g}(t_{\text{ZC,1}})}_{= 0}\dot{x}_{\text{low}} + 3\underbrace{\dot{g}(t_{\text{ZC,1}})}_{= 0}\ddot{x}_{\text{low}} = 0!$$

Thus, synchronizing a $C^2$ gain ramp with a signal zero-crossing promotes the modulated audio waveform $y(t)$ from $C^2$ to $C^3$ continuity at the attack onset, driving the onset's high-frequency spectral splatter down to $\mathcal{O}(\omega^{-5})$!

4. Section 4: Parseval Phase Invariance & Vocal Aeroacoustics (Theorems 4.1 & 4.2)

• Harmonic Analysis of Phase-Dispersed Square Waves: Theorem 4.1 rightly observes that passing a clipped odd-harmonic series $x_{\text{amp}}(t) = \frac{4A}{\pi}\sum_{m \text{ odd}} \frac{1}{m}\sin(m\omega_0 t)$ through a dispersive phase response $e^{j\phi(\omega)}$ ($\phi(m\omega_0) \neq m\phi(\omega_0)$) destroys the phase coherence required to form flat time-domain plateaus ($\kappa_{\text{clip}} \to 0$) and inflates the $L^\infty / L^2$ crest factor by $4\text{–}6\text{ dB}$.
• Mathematical Refinement on Equation (4.1) (Finite-Window vs. Infinite-Time Plancherel Identity):

In Section 4.1, the paper writes:

$$\int_{t_0}^{t_0 + T_w} \big|y_{\text{mic}}(t)\big|^2 dt = \frac{1}{2\pi}\int_{-\infty}^{\infty} \big|X_{\text{amp}}(j\omega)\big|^2 \big|H_{\text{cab}}(j\omega)\big|^2 d\omega$$

Strictly speaking, the left-hand side is a finite-window $L^2([t_0, t_0+T_w])$ energy, whereas the right-hand side is the global $L^2(\mathbb{R})$ Plancherel energy $\int_{-\infty}^\infty |y_{\text{mic}}(t)|^2\,dt$.

To state this with complete mathematical precision:

1. For a periodic steady-state note of period $T_0 = 2\pi/\omega_0$, Parseval's theorem for Fourier series over $T_w = K T_0$ holds with equality: $\frac{1}{T_w}\int_{t_0}^{t_0+T_w} |y_{\text{mic}}(t)|^2\,dt = \frac{1}{2}\sum_{m=1}^\infty |X_m|^2 |H_{\text{cab}}(jm\omega_0)|^2$, which is $100\%$ independent of $\phi(m\omega_0)$.
2. For a transient window of length $T_w$, unitary phase dispersion $e^{j\phi(\omega)}$ is an isometry on $L^2(\mathbb{R})$ and preserves the windowed norm $\|y_{\text{mic}}\|_{L^2([t_0, t_0+T_w])}^2$ up to boundary group-delay leakage $\mathcal{O}(\tau_g / T_w)$ whenever the cabinet's group delay $\tau_g(\omega) = -\frac{d\phi}{d\omega}$ satisfies $\tau_g \ll T_w$.

• Makhoul's Spectral Flatness Criterion: Minimizing $\mathcal{E}_{\text{LPC}} = \frac{1}{2\pi}\int_{-\pi}^\pi \frac{S_{xx}(e^{j\omega})}{|A(e^{j\omega})|^2}\,d\omega$ when $S_{xx}(e^{j\omega}) \approx \sum_{m=1}^M P_m \delta(\omega \pm m\omega_0)$ is a sparse harmonic comb ($f_0 > 300\text{ Hz}$) indeed forces the $P/2 = 4$ conjugate pole pairs of $1/A(z)$ toward the unit circle at the $4$ strongest harmonic frequencies $e^{\pm j m\omega_0}$, because driving $|A(e^{j m\omega_0})|^2 \to 0$ drives the discrete sum to zero.
• PCHIP + Low-Time Liftering: Interpolating $\{m f_0, L_m\}$ via Fritsch-Carlson Monotone Piecewise Cubic Hermite Interpolation (PCHIP) preserves monotonicity on each subinterval $[m f_0, (m+1)f_0]$ (preventing polynomial Runge oscillations between widely spaced soprano harmonics), and low-time cepstral liftering at cutoff quefrency $q_{\text{cut}} = \frac{0.72}{f_0} < \frac{1}{f_0} = q_0$ annihilates the fundamental rahmonic spike at $q_0 = 1/f_0$ and all higher rahmonics $k q_0$. Mathematically sound and well-engineered.

5. Section 5: Low-Rank Legendre Manifolds, Operator Subspaces & Lyapunov Control (Theorems 5.1, 5.2 & 5.3)

Let us check the discrete polynomial basis vectors $\mathbf{u}_{\text{tilt}}, \mathbf{u}_{\text{smile}} \in \mathbb{R}^{12}$ for $b \in \{0, 1, \dots, 11\}$:

• Exact Moment Verification:

For $N = 12$ and $x_b = b - \frac{N-1}{2} = b - 5.5$:

$$\sum_{b=0}^{11} x_b^2 = 2\sum_{m=1}^{6}\left(m - \frac{1}{2}\right)^2 = \frac{N(N^2 - 1)}{12}\Bigg|_{N=12} = \frac{12(143)}{12} = 143$$

Hence the mean of $(b - 5.5)^2$ is exactly $\frac{143}{12}$ (and the denominator norms are $\sqrt{143} \approx 11.95826$ and $\sqrt{\frac{12(143)(144-4)}{180}} = \sqrt{\frac{4004}{3}} \approx 36.5331$).

• Exact Orthonormality & Decoupled Convex Projection:

By odd/even symmetry about $5.5$, $\sum_{b=0}^{11} x_b = \sum_{b=0}^{11} x_b^3 = 0$, so:

$$\langle \mathbf{u}_{\text{tilt}}, \mathbf{1} \rangle = 0, \qquad \langle \mathbf{u}_{\text{smile}}, \mathbf{1} \rangle = 0, \qquad \langle \mathbf{u}_{\text{tilt}}, \mathbf{u}_{\text{smile}} \rangle = 0, \qquad \|\mathbf{u}_{\text{tilt}}\|_2 = \|\mathbf{u}_{\text{smile}}\|_2 = 1$$

Because $\{\mathbf{u}_{\text{tilt}}, \mathbf{u}_{\text{smile}}\}$ is an orthonormal set, extending it to an orthonormal basis of $\mathbb{R}^{12}$ proves by Parseval's identity in $\mathbb{R}^{12}$ that minimizing $\|\overline{\mathbf{E}}_{\text{slow}} - \mathbf{E}\|_2^2$ over the affine box $\mathcal{M}_p$ splits into independent 1-D quadratic minimizations over $\alpha_{\text{tilt}} \in [\alpha_{\text{tilt,min}}^{(p)}, \alpha_{\text{tilt,max}}^{(p)}]$ and $\alpha_{\text{smile}} \in [\alpha_{\text{smile,min}}^{(p)}, \alpha_{\text{smile,max}}^{(p)}]$. Theorem 5.1 is 100% exact.

• Exact DC Transmission Zero in ContourBand: Since $\overline{\mathbf{E}}_{\text{slow}}(s) = \frac{1}{1 + s\tau_{\text{slow}}}\mathbf{E}_{\text{fast}}(s)$, the innovation filter $H_{\text{BP}}(s) = 1 - \frac{1}{1 + s\tau_{\text{slow}}} = \frac{s\tau_{\text{slow}}}{1 + s\tau_{\text{slow}}}$ satisfies $H_{\text{BP}}(0) = 0$ identically, proving that ContourBand has zero steady-state response to static EQ curves.
• Two Pure-Math Refinements on $\mathbf{S}_3$ and $\mathbf{I} - \mathbf{S}_3$:

1. Boundary Condition on $\mathbf{I} - \mathbf{S}_3$ for $\mathbf{u}_{\text{tilt}} \in \ker(\mathbf{I} - \mathbf{S}_3)$:

Theorem 5.2 defines $(\mathbf{S}_3\mathbf{x})[b] = \frac{1}{4}\mathbf{x}[b-1] + \frac{1}{2}\mathbf{x}[b] + \frac{1}{4}\mathbf{x}[b+1]$ with Neumann boundary conditions ($\mathbf{x}[-1] = \mathbf{x}[0]$ and $\mathbf{x}[12] = \mathbf{x}[11]$) and then claims in Proof Item 2 that $\ker(\mathbf{I} - \mathbf{S}_3) = \text{span}\{\mathbf{1}, \mathbf{u}_{\text{tilt}}\}$.

Careful! Under homogeneous Neumann boundary conditions, the discrete Laplacian $\mathbf{I} - \mathbf{S}_3 = -\frac{1}{4}\Delta_b^2$ has quadratic form $\mathbf{x}^T(\mathbf{I}-\mathbf{S}_3)\mathbf{x} = \frac{1}{4}\sum_{b=0}^{10}(x_{b+1}-x_b)^2$, whose nullspace is 1-dimensional ($\ker(\mathbf{I}-\mathbf{S}_3) = \text{span}\{\mathbf{1}\}$)! Specifically, at the endpoints $b = 0$ and $b = 11$, Neumann BCs give:

$$\big((\mathbf{I} - \mathbf{S}_3)\mathbf{u}_{\text{tilt}}\big)[0] = \frac{1}{4}\big(\mathbf{u}_{\text{tilt}}[0] - \mathbf{u}_{\text{tilt}}[1]\big) = -\frac{1}{4\sqrt{143}} \neq 0, \qquad \big((\mathbf{I} - \mathbf{S}_3)\mathbf{u}_{\text{tilt}}\big)[11] = +\frac{1}{4\sqrt{143}} \neq 0$$

For $\ker(\mathbf{I} - \mathbf{S}_3)$ to be the 2-dimensional subspace $\text{span}\{\mathbf{1}, \mathbf{u}_{\text{tilt}}\}$, the boundary condition at $b = 0$ and $b = 11$ must be linear extrapolation (zero second difference at the boundary: $\mathbf{x}[-1] = 2\mathbf{x}[0] - \mathbf{x}[1]$ and $\mathbf{x}[12] = 2\mathbf{x}[11] - \mathbf{x}[10]$) rather than homogeneous Neumann reflection. (Note that because $\mathbf{E}_{\text{proj}}^*$ has already subtracted out the manifold's $\mathbf{u}_{\text{tilt}}$ and $\mathbf{u}_{\text{smile}}$ components before $\mathbf{I} - \mathbf{S}_3$ is applied in Eq. 5.2, any track tilt inside $\mathcal{M}_p$ is already zeroed out prior to $\mathbf{I} - \mathbf{S}_3$!)

2. Partition of Unity vs. Idempotent Orthogonal Projectors:

In operator theory, $\mathbf{S}_3$ is a self-adjoint contraction with spectrum $\sigma(\mathbf{S}_3) = \big\{\cos^2\!\big(\frac{k\pi}{24}\big)\big\}_{k=0}^{11} \subset (0, 1]$, not an idempotent orthogonal projection ($\mathbf{S}_3^2 \neq \mathbf{S}_3$, so $\langle \mathbf{S}_3\mathbf{x}, (\mathbf{I}-\mathbf{S}_3)\mathbf{x}\rangle \ge 0$ is not identically zero on intermediate wavenumbers). Rather, $\mathbf{S}_3 + (\mathbf{I} - \mathbf{S}_3) = \mathbf{I}$ forms a complementary high/low-wavenumber filter bank (partition of unity), while strict dynamical decoupling is enforced by subtracting $\mathbf{E}_{\text{proj}}^*$, separating timescales ($40\text{ ms}$ vs. $1.5\text{ s}$), and engaging the Time-Division Turn Gate ($\mathbf{e}_{\text{voice}}(t) \equiv \mathbf{0}$ during compressor turns).

• Fixed-Point Critique of Pure Step-Size Scaling (v2.0):

For $\theta_m[n+1] = \theta_m[n] - \eta_m\gamma(\mathcal{S}_{\text{pre}})\mathcal{D}_{\tau_m,\delta_m}(e_m[n])$ with $\gamma(\mathcal{S}_{\text{pre}}) = 1 + 0.65(1 - \mathcal{S}_{\text{pre}}) \in [1.0, 1.65] > 0$, the fixed-point condition $\theta_m[n+1] = \theta_m[n]$ reduces to $\mathcal{D}_{\tau_m,\delta_m}(e_m(\theta_m)) = 0 \iff |e_m(\theta_m)| \le \tau_{m,0}$, which is strictly independent of $\mathcal{S}_{\text{pre}}$ (and at $\mathcal{S}_{\text{pre}} = 1$, $\gamma$ drops from $1.65$ to $1.0$, a $\frac{0.65}{1.65} = 39.4\%$ step-size reduction). This proof is completely rigorous.

• Lyapunov Proof of the v2.5 Equilibrium-Shifted Leaky Huber Servo:

With $\tilde{\theta}_m[n] = \theta_m[n] - \theta_{m,0}$ and $V(\tilde{\theta}_m) = \frac{1}{2}\tilde{\theta}_m^2$:

• Whenever $\mathcal{S}_d \to 1$ and the stem's residual error lies within the $2.6\times$-expanded JND deadband $|e_m[n]| \le \tau_m(1) = 2.60\,\tau_{m,0}$, the deadzone-Huber term vanishes ($\mathcal{D}_{\tau_m(1),\delta_m}(e_m[n]) = 0$). Since clamping around $0$ is non-expansive ($|\text{clamp}(z, -M, M)| \le |z|$), the dynamics satisfy:
$$\Delta V[n] = V\big(\tilde{\theta}_m[n+1]\big) - V\big(\tilde{\theta}_m[n]\big) \le -\lambda_0(2 - \lambda_0)\,V\big(\tilde{\theta}_m[n]\big) < 0 \quad \forall\,\tilde{\theta}_m[n] \neq 0$$

for $\lambda_0 = 0.12 \in (0, 2)$. Thus $\tilde{\theta}_m[n] \to 0$ exponentially ($\theta_m[\infty] = \theta_{m,0}$).

• Scope of the 0.000 dB Harm Guarantee: From a strict mathematical standpoint, it is worth stating explicitly in Theorem 5.3 that exact 0.000 dB steady-state modification ($\tilde{\theta}_m[\infty] = 0$) holds for all pre-processed stems ($\mathcal{S}_d > 0$) whose discrepancy from the 2-D Legendre manifold $\mathcal{M}_p$ lies within the expanded JND deadband $|e_m| \le \tau_{m,0}(1 + 1.60\,\mathcal{S}_d^2)$. Outside that deadband ($|e_m| > \tau_m(\mathcal{S}_d)$), the system is Input-to-State Stable (ISS) and converges to a regularised Tikhonov/Huber equilibrium $\tilde{\theta}_m^*$ bounded by $|\tilde{\theta}_m^*| \le \Delta\theta_{m,\max}(\mathcal{S}_d) = \Delta\theta_{m,\max,0}(1 - 0.88\,\mathcal{S}_d^2)$ ($88\%$ authority compression at $\mathcal{S}_d = 1$).

Actionable List of Mathematical Errata & Editorial Fixes for v3.0

To make this monograph bulletproof for formal peer review, I recommend the following six targeted revisions:

1. Fix the $T_{60}$ vs. $T_{120}$ Constant in Theorem 2.1 & the Abstract:

• In Theorem 2.1 (Eq. 2.1), replace $T_{60,\text{ring}} = \frac{6\ln(10)\,Q}{\pi f_0} \approx \frac{4.40\,Q}{f_0}$ ($40.1\text{ ms}$) with the true $-60\text{ dB}$ amplitude decay time $T_{60,\text{ring}} = \frac{3\ln(10)\,Q}{\pi f_0} \approx \frac{2.20\,Q}{f_0}$ ($20.07\text{ ms}$ at $Q=22, f_0=2410\text{ Hz}$).
• In the Abstract (Item 2), remove the duplicate $\pi$ in the denominator: change $T_{60} \approx \frac{2.2 Q}{\pi f_0}$ to $T_{60} \approx \frac{2.20\,Q}{f_0}$.

1. State the Relative-Level Hypothesis in Theorem 2.1 (Part 1):

• Explicitly specify the voiced-to-ghost broadband power ratio $P_v / P_g$ (or hop-grid alignment) under which the Hann-weighted integral satisfies $\widehat{\text{HNR}}_{\text{STFT}} > +8.2\text{ dB}$.

1. Clarify the Boundary Conditions of $\mathbf{S}_3$ in Theorem 5.2:

• Replace "with Neumann boundary conditions" by "with linear-extrapolation (zero-second-difference) boundary conditions $\mathbf{x}[-1] = 2\mathbf{x}[0]-\mathbf{x}[1],\, \mathbf{x}[12] = 2\mathbf{x}[11]-\mathbf{x}[10]$" so that $\ker(\mathbf{I} - \mathbf{S}_3) = \text{span}\{\mathbf{1}, \mathbf{u}_{\text{tilt}}\}$ holds exactly at the boundary bands $b = 0$ and $b = 11$ as well as in the interior $b \in \{1,\dots,10\}$.

1. Distinguish Finite-Window vs. Global Plancherel Integration in Equation (4.1):

• Write the finite-window Parseval identity over $K$ periods as $\frac{1}{T_w}\int_{t_0}^{t_0+T_w} |y_{\text{mic}}(t)|^2\,dt = \frac{1}{2}\sum_{m} |X_m|^2 |H_{\text{cab}}(jm\omega_0)|^2 + \mathcal{O}(\tau_g / T_w)$ rather than equating a finite-time integral on the left to an improper $\int_{-\infty}^\infty d\omega$ integral on the right.

1. Write Out the Explicit Deadzone-Huber Operator $\mathcal{D}_{\tau,\delta}(e)$ in Section 5.3 & Check $\tau_g(30\text{ Hz})$ in Theorem 2.3:

• Explicitly define $\mathcal{D}_{\tau,\delta}(e) = \text{sign}(e)\,\min\!\big(\max(|e| - \tau, 0), \delta\big)$ before Eq. (5.3), and note the exact group delay $\tau_g(30\text{ Hz}) = 4.85\text{ ms}$ for $f_c = 52\text{ Hz}, Q = 0.55$ (while emphasizing that the differential quadrature group delay between $V_{\text{LP}}$ and $V_{\text{BP}}$ is identically $0.0000\text{ ms}$).

1. Synchronize the Table of Contents with the Body Section Headings & Anchor IDs:

• The sidebar Table of Contents (#sec-2 through #sec-4) still reflects an earlier instrument-grouped outline (2. Guitars, 3. Drums, 4. Bass, Vocals, Keys & Brass with 4 vocal coordinates), whereas the main text is organized by mathematical regime (2. Time-Frequency Uncertainty, 3. Nonlinear Membrane Mechanics, 4. Electromechanical Speaker Dispersion & Vocal Tract Aeroacoustics with 3 orthogonal vocal coordinates). Aligning the TOC links and numbering with the body headings will make navigation seamless.