File size: 14,688 Bytes
6ee3aa6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 | \page
\sect{Which dispersive return statistics are calculable?}
Take identical real forward and backward scalar drives \(w\), equal sideband phases \(e^{\ii b(t-s)}\) on \(m=\pm1\), and no damping. The coefficient of the first return, including its factor \(\delta\), is
\[
R=\delta\int_{[0,L]^2}e^{\ii b|t-s|}\,w(t)w(s)\dd t\dd s .
\]
At this order there are two orderings; with only unit positive/negative steps there is exactly one step of each sign. The complex return changes the \emph{bright} carrier. With a perfectly dark internal output it only multiplies an existing signal.
\thm{Proposition 9}{Exact finite-length Gaussian quadrature}
For centered stationary real Gaussian \(w\) of covariance \(C(t,s)=\sigma^2e^{-|t-s|/\ell}\), let \(F=\int_0^L e^{\ii bt}w(t)\dd t\), \(h=\ell^{-1}-\ii b\), and \(J(\lambda)=(e^{\lambda L}-1)/\lambda\), \(J(0)=L\). Then
\begin{align*}
\operatorname{Re}R&=\delta |F|^2,\qquad \E\operatorname{Re}R=\delta V,\\
V=\E|F|^2&=2\sigma^2\operatorname{Re}\!\left[\frac Lh-\frac{1-e^{-hL}}{h^2}\right],\\
P_b=\E F^2&=\frac{2\sigma^2}{\bar h}\big[J(2\ii b)-J(-h)\big],\\
\Var(\operatorname{Re}R)&=\delta^2(V^2+|P_b|^2).
\end{align*}
The real kernel equals \(\cos(b(t-s))\). Integrating the stationary covariance over the two triangles gives \(V,P_b\). Pairing the four Gaussian factors in \(|F|^4\) yields two copies of \(V^2\) and one \(|P_b|^2\); subtract the mean squared. For the complex fluctuation let \(\mathsf K_b(t,s)=e^{\ii b|t-s|}\) and use \(C\) also for the covariance operator on \(L^2[0,L]\). The same pairing gives \(\E|R-\E R|^2=2\delta^2\norm{C^{1/2}\mathsf K_b C^{1/2}}_{\rm HS}^2\). This last expression includes the quadrature missing from \(\operatorname{Re}R\). \fin
At zero detuning \(R=\delta(\int w)^2\). Its variance is exactly \(2\delta^2[2\sigma^2(L\ell-\ell^2(1-e^{-L/\ell}))]^2\), quadratic in \(L\) for large \(L\). For fixed \(b\ne0\), \(V=L S(b)+O(1)\), \(P_b=O(1)\) with \(S(b)=2\sigma^2\ell/(1+b^2\ell^2)\). Thus the ratio of the \emph{real-quadrature} variances at \(b\) and zero tends to \(1/[2(1+b^2\ell^2)^2]\) as \(L\to\infty\). The limit is nonuniform near \(b=0\). A separate choice of forward/backward cross-covariance changes the formula.
A Lorentzian spectrum alone leaves the fourth moment free. Multiply a Gaussian path by an independent real \(A\) with \(\E A^2=1\). The covariance stays fixed, while the centered variance of the return can become infinite if \(\E A^4=\infty\). Conversely, a white limit need not diverge. With \(C_\eta(t)=\sigma^2e^{-|t|/\eta}/(2\eta)\) the two ordered contractions at \(b=0\) give \(\delta\sigma^2[L-\eta(1-e^{-L/\eta})]\to\delta\sigma^2 L\). Regularization of other products must be specified individually.
\textbf{Rechecking the earlier numerical advantage.} In its own additive-variance example \(V_{\rm FF}=2.5\,10^{-8}\), \(V_{\rm ret}=3.46154\,10^{-13}\), \(V_{\rm dark}=1.6\,10^{-7}\). At \(n=10^{10}\) and equal slope, \((1/4+nV_{\rm FF})/(1/4+n(V_{\rm ret}+V_{\rm dark}))=0.15638148\). Treating that earlier model literally reverses its advertised \(18{,}000\)-fold conclusion. In an exact dark port, multiplicative signal noise usually has variance \(O(\epsilon^2)\) and cannot be inserted as a constant floor at zero. A real Gaussian likelihood with signal-dependent variance \(v\) has information \(\mu'^2/v+(v'/v)^2/2\); a mixture needs its actual likelihood. Theorems 3 and 4 avoid inventing an additive-variance baseline.
\page
\sect{Rectangle sewing with explicit hypotheses}
Let \(A(I,J)\) be a Banach-space-valued rectangle increment, zero when either interval has zero length. Put \(\delta_h A=A(I,J)-A(I_1,J)-A(I_2,J)\) for a split of \(I\) and define \(\delta_v\) similarly. Continuous superadditive interval controls \(U=\omega_t(I), V=\omega_z(J)\) vanish on zero-length intervals. Suppose for all splits
\[
\norm{\delta_h A}\le C_h U^\theta V^\beta,\quad
\norm{\delta_v A}\le C_v U^\alpha V^\eta,\quad
\norm{\delta_h\delta_v A}\le C_m U^\theta V^\eta ,
\qquad \theta,\eta>1,\quad 0<\alpha,\beta\le1.
\]
\thm{Theorem 10}{Nonuniform additive sewing}
The sums \(S_{P,Q}A=\sum_{I_i\in P,J_j\in Q}A(I_i,J_j)\) converge in norm on arbitrary nonuniform tensor grids to a separately additive rectangle increment. For fixed enclosing \(I,J\), the error against the limit is bounded by
\begin{align*}
&c_\theta U\,[C_h V^\beta+c_\eta C_m V^\eta]\,\Delta_t^{\theta-1}\\[-1mm]
&\hspace{14mm}+c_\eta V\,[C_v U^\alpha+c_\theta C_m U^\theta]\,\Delta_z^{\eta-1},
\qquad c_r=2^r\sum_{j\ge1}j^{-r},
\end{align*}
where \(\Delta_t,\Delta_z\) are maximal mesh controls. If bounds and the coarse germ belong to \(L^p\), with deterministic controls (or integrable dominating products for random controls), convergence also holds in \(L^p\).
\prooftext In a one-dimensional partition with \(M\) intervals, the sum of the \(M-1\) adjacent-pair controls is at most \(2U\), by grouping pairs of each parity and using superadditivity. Drop the interior point with smallest paired control. Repeated deletion shows a split defect bounded by \(C U^r\) costs at most \(c_r C U^r\) over any partition. Summing within the old cells yields a refinement bound \(c_r C U\Delta^{r-1}\).
For a fixed vertical partition \(Q\), the horizontal defect of \(S_Q^v A\) equals \(S_Q^v\delta_h A\). Its difference from \(\delta_h A\) is bounded by \(c_\eta C_m U^\theta V^\eta\) by the one-dimensional estimate in the vertical variable. Hence its horizontal defect constant is bounded uniformly in the number of vertical cells. Apply the horizontal refinement estimate. Interchange the variables to bound vertical refinement and compare any two grids through their common refinement. Completeness gives the limit; taking grids that share a prescribed cut proves separate additivity. The same bound yields the \(L^p\) statement by domination. \fin
\textbf{An explicit enhanced example.} For scalar paths \(X,Y\) with exponents \(\beta_t,\beta_z>0\) in \((0,1)\), take local factors \(e^{\ii X(s)}\sum_{j=1}^{\lfloor1/\beta_t\rfloor}(\ii\Delta X)^j/j!\) and its vertical counterpart, and multiply. In each direction Taylor's theorem gives a defect of order \((\lfloor1/\beta\rfloor+1)\beta>1\). The mixed defect is the product of the two defects. The sewn value is the product \(\Delta e^{\ii X}\Delta e^{\ii Y}\) by telescoping and the one-dimensional remainder estimate. General noncommuting rough sheets need compatible lifts and a verified germ \cite{chouk,bechtold}. A unit germ \(A=1\) sums to \(MK\); even the normalized germ \(A=|I|^\alpha |J|\), \(0<\alpha<1\), has zero mixed defect and diverges on refinement. The one-sided conditions do real work. At \(m=0\) a circle shift acts as identity. On the full ladder \(T_\phi S T_\phi^{-1}=e^{\ii\phi}S\), so the shift is not central. Area additivity alone supplies neither an interchange law nor an enhanced rough sheet.
\page
\sect{Exact root-of-unity spectrum and the measured poles}
On the polynomial core of the unsymmetrized ladder let \(Te_n=e_{n+1}+[n]_q e_{n-1}\), \([n]_q=(1-q^n)/(1-q)\), \(e_{-1}=0\). For \(|q|<1\), \(\norm T\le1+(1-|q|)^{-1}\) on unweighted \(\ell^2\); for \(|q|=1,\ q\ne1\), use \(|[n]_q|\le2/|1-q|\). The exterior resolvent and its continued-fraction approximants agree there by a Neumann series and coefficient stabilization on finite truncations. At \(q=1\) a positive Fock normalization gives Jacobi weights \(\sqrt n\) and the classical sufficient Carleman condition. For real \(q\) with strictly positive \([n]_q\), the same positive normalization is available; for complex \(q\) a symmetric bilinear pairing is generally not a positive Hilbert inner product. For \(|q|>1\), convergence of a Carleman-type sum does not imply a general failure of determinacy or multiple values of a specified operator's resolvent \cite{linmoment}. At a root of unity the factorial pairing has a radical.
\thm{Proposition 11}{Finite quotient and all closed poles}
Let \(q=e^{2\pi\ii p/m}\) be primitive, \(m\ge2\), and \(b^2=1-q\). The tail from \(e_m\) upward is invariant and the quotient has matrix \(T_m\). The upper-right block is zero; the forward edge from \(e_{m-1}\) into the tail remains. The carrier vacuum resolvent is rational on its initial exterior domain and extends meromorphically as \(P(z)/Q_m(z)\), where \(P\) deletes the first row and column of \(zI-T_m\) and takes the determinant. All closed poles are distinct:
\[
Q_m(z)=2b^{-m}\mathrm T_m(bz/2),\qquad
z_j=\frac{2}{b}\cos\!\frac{(2j-1)\pi}{2m},\quad j=1,\ldots,m .
\]
Here \(\mathrm T_m\) denotes the ordinary Chebyshev polynomial. The choice of square root only permutes the nodes.
\prooftext The continuant recurrence \(Q_{n+1}=zQ_n-[n]_qQ_{n-1}\), \(Q_0=1,Q_1=z\), identifies \(b^n Q_n(2x/b)\) with the continuous \(q\)-Hermite polynomial \(H_n(x|q)\). Its finite expansion is \(\sum_{j=0}^n {n\brack j}_q e^{\ii(n-2j)\vartheta}\) at \(x=\cos\vartheta\); substitution, or the Gaussian-binomial recurrence, verifies the same recurrence and initial data. At \(n=m\) all interior coefficients vanish because the numerator includes \(1-q^m\) and interior denominators are nonzero. Thus \(H_m=2\cos(m\vartheta)\), the known root-of-unity identity \cite{qhermite,dlmf}. Successive trailing-minor recurrences with nonzero internal couplings force \(P\) and \(Q_m\) to be coprime. The explicit cosines are distinct. \fin
Odd power sums vanish. Summing the even Fourier powers of the cosine nodes gives \(\Tr(T_m^{2r})/m={2r\choose r}(1-q)^{-r}\) for \(0\le r<m\). In particular \(\Tr(T_m^2)/m=2/(1-q)=1+\ii\cot(\pi p/m)\). These are complex spectral moments. For \(m=2\), the fourth moment instead equals 1, illustrating the range restriction.
\textbf{Port loading.} In a finite linear input--output realization with \(J_0>0\), port rate \(\gamma>0\), and common intrinsic linewidth \(\kappa_0\), set \(z=(\omega-\omega_0+\ii\kappa_0/2)/J_0\), \(g=\gamma/(2J_0)\). Rank-one inversion gives \(r=(Q_m-\ii gP)/(Q_m+\ii gP)\). The pole near \(z_j\) is \(z_j-\ii gP(z_j)/Q_m'(z_j)+O(g^2)\); its linewidth is \(-2\operatorname{Im}\omega_{\rm pole}\) under the \(e^{-\ii\omega t}\) convention. The real part of the residue enters the leading linewidth change. These are poles of the specified coupled-mode model. Stability, causal material parameters, measured complex couplings and a scattering fit are needed to identify poles of a physical Maxwell device. No accidental higher-order collision occurs in the primitive closed quotient.
\input{moments}
\page
\sect{Scale flow, evidence and experimental controls}
\textbf{The scale equation.} For \(P=e_{k-1}(\zeta)=\sum_{j=0}^{k-1}\zeta^j/j!\), fixing \(e^{\Delta s}P(\zeta(s))\) gives \(\dot\zeta=-\Delta P/P'\) wherever \(P'\ne0\). The roots of \(P\) are simple because \(P-P'=\zeta^{k-1}/(k-1)!\) and \(P(0)=1\). Critical points of \(P\) are poles of this vector field, not smooth equilibria. For \(\Delta>0\), the image \(P(\zeta(s))=P(\zeta(0))e^{-\Delta s}\) travels radially to zero. A radius avoiding the finite set of nonzero critical values lifts uniquely by local inversion and properness of \(P\). It terminates at a simple root, with local rate \(e^{-\Delta s}\). The excluded preimages of critical rays have planar measure zero: off finitely many critical points the polynomial is a local diffeomorphism. Only branches actually hitting a critical point are singular trajectories. Small coefficient perturbations preserving degree move simple roots continuously. A microscopic decimation, scaling dimension \(1-4\beta\), global structural stability and photonic renormalization group interpretation require additional evidence.
\textbf{Implementation.} A three-port test mesh needs a calibrated signal \(1\leftrightarrow3\) coupler, injected \(1\leftrightarrow2\) and \(2\leftrightarrow3\) couplings, and an endpoint phase shifter/balanced combiner. A five-port path gives \(d=4\). If the measurand also changes onsite phase or neighboring couplings, measure its generator and repeat the identifiability analysis. A programmable mesh can run this test without a ring. A conventional interferometer receiving the same measurand can already be optimal when its technical environment is benign.
For a frequency-lattice implementation, Dinh et al.\ measured a thin-film lithium-niobate ring with FSR about \(9.52\) GHz and loaded \(Q\simeq180{,}000\) \cite{dinh}. Synthetic frequency transport in driven rings is well established \cite{yuan,zhang}. For \(n_g=2.2\), the FSR gives a round-trip length \(c/(n_g\mathrm{FSR})\simeq14.3\) mm. This is an illustrative group-index calculation. Ordinary lossless phase modulation is Hermitian and supplies conjugate frequency hoppings. An output single-sideband extinction ratio measured on one input does not certify intracavity one-way hopping on every bin. Track extra ports implementing effective asymmetry. In a spatial model, removal of the constant and linear phases gives \(b\simeq k^{\prime\prime}(\omega_0)\Omega_{\mathrm{FSR}}^2/2\) for angular frequency step \(\Omega_{\mathrm{FSR}}\), with the sign fixed by the propagation convention.
\begin{center}\small
\begin{tabularx}{\linewidth}{@{}p{.22\linewidth}XX@{}}
\toprule Test & Measurement & Prediction or failure criterion\\\midrule
Graph distance & At zero signal inject calibrated adjacent perturbations; sweep their common scale. Match last-edge phase. & Signed dark offset of the chosen chain has exponent \(d\); a measured direct nuisance shortcut invalidates that graph.\\
Indistinguishability & Program \eqref{eq:pulse} and the equivalent direct coupling. Use multi-input complex tomography. & All terminal matrix elements agree within independently propagated measurement uncertainty. One port alone is an insufficient test.\\
Slope and resources & Measure \(\pm\epsilon\) on randomized interleaved shots with the same exposure and efficiency. & Under the action cap, pooled slope obeys \eqref{eq:slope}. Include intermediate and rejected photons in the budget.\\
One-sided carrier & Probe several frequency bins; compare open-mode and reciprocal transfer measurements. & A positive-only generator gives exact carrier compression. A periodic wrap or input below carrier can restore returns.\\
\bottomrule
\end{tabularx}\end{center}
\textbf{Metrology controls.} Bound integrated action from calibrated edge measurements, including long links. A terminal transfer alone cannot upper-bound this action. Record incident and detected flux, both arm efficiencies, dark counts, accepted sidebands, exposure, pump energy and drift. A carrier-only filter can discard the pooled-mode benefit. The example \(a=0.2\) and any device tolerances require measurements.
|