File size: 14,670 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
85
86
87
88
89
90
91
92
\page
\sect{One-sided propagation and common-mode return}
Let \(\mathcal H=\ell^2(\mathbb Z)\otimes\mathbb C^k\), let \(S\ket m=\ket{m+1}\), and define the tails \(F_j=\overline{\mathrm{span}}\{\ket m:m\ge j\}\otimes\mathbb C^k\). In this section the \emph{generator} convention is \(U'=HU\), so lossless diagonal detuning is \(+\ii b m^2\). A physical Hamiltonian is multiplied by \(-\ii\) once when converting to this convention.

\thm{Proposition 6}{Exact carrier compression}
Suppose the free evolution \(D(t,s)=\bigoplus_m D_m(t,s)\) exists with a common exponential bound, and \(V_+(t)=\sum_{r\ge1}S^r\otimes W_r(t)\) has \(\int\sum_r\norm{W_r}<\infty\). Then
\[
 (\bra0\otimes\Id)\,U(L,0)\,(\ket0\otimes\Id)=D_0(L,0).
\]
Indeed, every Duhamel word with a scattering factor starts in \(F_0\), enters \(F_1\), and remains there. Equivalently, the dynamics acts as \(D_0\) on \(F_0/F_1\). Noncommutativity, a time-dependent internal signal, the probability law, and internal matrix dimension do not enter the argument. A deterministic carrier measurement \emph{before} squaring therefore has zero variance with respect to this technical drive. Shot noise and vacuum enter separately. If every cutoff preserves the filtration, its compressed block is already constant. This gives a well-defined compression for arbitrarily irregular regularizations, even when the full field has no cutoff limit. An open finite-mode truncation preserves the result. On a cyclic ladder \(S^M=I\) and a positive word can return. Population initially below the selected carrier can also enter. Nonlinear mixing produces forbidden returns at arbitrarily small nonzero strength; any practical onset is a tolerance, not a universal threshold.

\textbf{The converse for one linear output.} For \(\Delta_m=0\) with scalar nearest-neighbor drives on both \(S,S^{-1}\), let \(c_\pm=\int_0^L w_\pm\). The shifts commute on the bilateral ladder. Thus the exact carrier multiplier is
\[
 h=\sum_{p\ge0}\frac{(\delta c_+c_-)^p}{(p!)^2}.
\]
The internal free transfer is multiplied by \(h\). Universally erasing the first return in a fixed linear functional erases the corresponding clean amplitude. The order is \(\delta\) in amplitude and typically \(\delta^2\) in variance. This reasoning also applies to a dark port's signal slope if a fluctuating common factor multiplies that slope.

\thm{Proposition 7}{Multiport cancellation to all orders}
For \(H(t)=K(t)\otimes I+I\otimes M_\epsilon(t)\) with well-defined evolutions, \(U=U_K\otimes U_{M_\epsilon}\), by direct differentiation. An input \(\xi\otimes x\) gives an internal vector \(s(\epsilon)=U_{M_\epsilon}x\) multiplied by \(h_m=\langle m,U_K\xi\rangle\) in each output sideband. All nonzero copies have the same projective direction \([s]\). For a collected set \(J\), photon counts after an internal analyzer have a common intensity scale \(\sum_{m\in J}|h_m|^2\). Conditioning on the total count eliminates this scale. For lossless \(U_K\) and all modes collected, the scale is identically one. Zero throughput gives no information. A raw subtraction of two identical rings cancels a common additive background at the null, but their differential signal still carries a common multiplicative return. This also explains the advantage of a shared internal reference and sideband pooling without a high-gain element.

For a differentiable \(s\ne0\), set \(u_s=s/\norm s\) and \(v=(I-u_su_s^\dagger)\dot u_s\). The locally chosen analyzer \((u_s\pm v/\norm v)/\sqrt2\) gives \(1/2,1/2\) and derivatives \(\pm\norm v\) at the reference point. Thus each detected photon gives \(4\norm v^2=4(\norm{\dot s}^2/\norm{s}^2-|s^\dagger\dot s|^2/\norm{s}^4)\) conditional information, interpreted as zero if \(v=0\). This is also the single-photon pure-state information for the normalized direction. The photon count, internal throughput, and calibration cost still matter when comparing per incident photon.
\page
\sect{A finite test for output protection in a calibrated network}
Factorization of the entire device is a strong sufficient condition. A particular set of ports can preserve a signal even when unobserved states mix nonseparably. Here is an exact test on a finite alphabet of time-independent generator matrices with independently variable piecewise-constant controls. It uses standard reachable-space/word algebra \cite{petreczky}.

Consider actual and reference states \(x\in\mathbb C^n,\ y\in\mathbb C^m\), initialized at \(x_0,y_0\) and observed through \(C,D\) into the same \(\mathbb C^p\). Their common control labels have matrix pairs \((A_j,B_j)\); the drift is another pair with coefficient one. The defect letters have independent controls scaled by a common scalar \(\delta\). A signal waveform can be included as a common zero-cost label if a result for all such waveforms is desired. Set
\[
 z_0=x_0\otimes y_0,\quad H_j=A_j\otimes I+I\otimes B_j,\quad
 \mathcal W(x\otimes y)=Cx\wedge Dy.
\]
Here \(u\wedge v\) lists \(u_i v_j-u_j v_i\) for \(i<j\), without complex conjugation. This definition makes \(\mathcal W\) linear on the tensor product. If both outputs are nonzero, vanishing means equality of projective directions. Erasure and throughput require a separate check.

\thm{Theorem 8}{Exact first observable defect order}
Let \(\mathcal B\) be the zero-cost lifted matrices, including the drift, and \(\mathcal E\) the lifted defects. For any subspace \(R\) write \(\operatorname{cl}_{\mathcal B}R\) for its closure under repeated multiplication by the letters in \(\mathcal B\), taking linear spans. Form
\begin{align*}
 R_0&=\operatorname{cl}_{\mathcal B}\operatorname{span}\{z_0\},\\
 R_{j+1}&=\operatorname{cl}_{\mathcal B}\left(R_j+\sum_{E\in\mathcal E} E R_j\right).
\end{align*}
The first \(d\) with \(\mathcal W R_d\ne0\) is the smallest defect order detectable in the wedge for some piecewise-constant controls and some duration. If the chain stabilizes inside the kernel, the wedge is zero for all controls, durations and \(\delta\). There are at most \(nm-\dim R_0\) strict extensions. This is an all-durations certificate; fixed clock durations, bandwidth restrictions and tied controls may remove witnesses.

\prooftext The product \(x\otimes y\) obeys the lifted ODE. Induction identifies \(R_j\) as the span of every matrix word applied to \(z_0\) with at most \(j\) defect letters. Its image under \(\mathcal W\) tests the corresponding Duhamel coefficients. Conversely, on successive independently chosen control intervals, differentiate the product of exponentials once in each duration at zero. The result is the ordered product of the interval generators. Taking coefficients in the independently varied interval amplitudes selects each desired letter; the constant coefficient selects the drift. Hence a nonzero word with \(d\) defects witnesses a nonzero order-\(d\) coefficient for some controls and positive durations. If a span fails to grow, it is invariant under both sets of letters. Finite dimension gives the termination bound. \fin

For \(d\ge1\), if the cost-free propagator is bounded by \(\exp(\int_s^t b)\), put \(B=\int_0^L b\), \(a=|\delta|\int_0^L\norm{E(t)}\dd t\), with \(E\) the actual lifted defect combination. The Dyson tail gives \(\norm{\mathcal W z(L)}\le\norm{\mathcal W}\norm{z_0}e^{B+a}a^d/d!\). Dividing by \(\norm{Cx(L)}\norm{Dy(L)}\) bounds the sine of the angle between the output lines. For fixed \(D\), the vanishing constraints are linear in the unknown readout \(C\), so the same test can search for a decoder, subject to nonzero throughput. In the supplied exact-rational example a hidden channel increases the reachable dimensions from \(3\) to \(5\) while staying invisible. Adding a return edge gives dimensions \(3,5,6\) and first observed cost two.
\page
\sect{Full-wave and geometric claims in their valid regimes}
\textbf{Exact plane-stratified Maxwell system.} Work at normal incidence with no transverse derivatives or longitudinal fields, one fixed propagation axis, and instantaneous real coefficients. Put \(J=\hat z\times\), \(J^2=-I\), \(c_0^{-1}=\sqrt{\varepsilon_0\mu_0}\), \(\eta_0=\sqrt{\mu_0/\varepsilon_0}\), and
\[
 D=\varepsilon_0 n E+\kappa J H,\qquad B=\mu_0 n H-\kappa J E,\quad X=\eta_0 JH .
\]
Here \(n\) is dimensionless, \(\kappa\) has units of inverse speed, and \(n>0,\ |\kappa|<n/c_0\) makes the real instantaneous constitutive energy form positive. Let the coefficients be smooth enough for these derivatives. Curl Maxwell gives
\[
 \partial_z E=\partial_t[(n/c_0)X+\kappa E],\qquad
 \partial_z X=\partial_t[(n/c_0)E+\kappa X].
\]
For \(s=\pm1\), \(\Psi_s=E-sX\) therefore obeys the exact equation
\[
 \partial_z\Psi_s=\partial_t[(\kappa-s n/c_0)\Psi_s].
\]
The common sign of \(\kappa\) in the two branches matters. This derivation retains temporal derivatives of the constitutive coefficients. It specifies a geometry where decoupling can actually be checked, consistent with the directional-field literature \cite{kinsler}. More generally the \emph{complete} transverse evolution, longitudinal elimination, boundary maps and material memory operators must preserve the directional subspaces. Impedance matching of two scalar functions alone is insufficient. For a forward TE plane wave oblique by angle \(\alpha\) in an isotropic matched medium, \(X=-\cos\alpha\,E\), so both of these axis-defined spinors are present. No universal propagation-constant, bandwidth or asymmetry norm bound follows from the two scalar modulations in an unspecified ring.

\textbf{Three curvatures with different definitions.} A polarization connection with commuting coefficient matrices can have \(\mathcal A_z=tJ,\mathcal A_t=0\). Its curvature is \(J\,dt\wedge dz\). A pure gauge or explicit differential closure is required for flat polarization transport. Independently, on the trivial circle bundle take \(\vartheta=d\phi+A_tdt+A_zdz\). In the coframe \(dt,dz,\vartheta\) with zero spin connection the spin curvature vanishes and the coframe torsion is
\[
 T^\phi=dA=(\partial_t A_z-\partial_z A_t)\,dt\wedge dz.
\]
Horizontal transport shifts \(\phi\) by \(-\oint A\bmod 2\pi\). Phase defects contribute distributional circulation only when their winding and scaling are specified. For a single smooth RF phase with \(A=d\Phi\), this curvature vanishes. Independently calibrated links around a loop can instead encode a nonzero plaquette phase. In the dual comb, a shift multiplies mode \(m\) by \(e^{\pm\ii m\Delta\phi}\), depending on Fourier convention. An optical spectrum analyzer recording powers alone cannot see it. A coherent beat with at least two sidebands or an interferometric loop reference can.

\textbf{Causal response and energy.} For real nondispersive symmetric constitutive matrix \(C(t)\) and field \(f=(E,H)\), Poynting's identity gives \(\partial_t(f^TCf/2)+\nabla\cdot(E\times H)=-E\cdot J_{\rm free}- f^T\dot C f/2\). The modulation source exchanges energy with the fields. In a stationary causal dispersive medium the response is a convolution; a time-varying medium needs a two-time response. A complex constant multiplying real time-domain fields does not close either energy or fluctuation accounting. Passive scattering must include loss ports, and active gain must include added channels. Operation below threshold must be checked on the loaded dynamics. The model in Theorems 1--3 avoids gain. It is an effective linear network and does not require a particular full-wave reduction.
\page
\sect{What the Jordan flag and its readout establish}
For \(k\ge2\), \(N=\sum_{j=1}^{k-1}E_{j,j+1}\), and \(E=E_{k1}\). Then \((N+\epsilon E)^k=\epsilon I\) and the characteristic polynomial is \(z^k-\epsilon\). In the generator convention \(U'=HU\),
\begin{align*}
 \bra k e^{L(N+\epsilon E)}\ket1&=\epsilon L+
           \epsilon^2 L^{k+1}/(k+1)!+O(\epsilon^3),\\
 \bra1 e^{L(N+\epsilon E)}\ket1&=1+\epsilon L^k/k!+O(\epsilon^2).
\end{align*}
This follows by counting powers in the exponential series. The dark-channel slope is \(L\), and the bright-channel slope is \(L^k/k!\), before physical normalization. The eigenvalue splitting can be nonanalytic while both transfers are entire in \(\epsilon\).

\textbf{Spin realization and normalization.} In the spin-\((k-1)/2\) representation let \(T_+\) have superdiagonal \(t_j=\sqrt{j(k-j)}\), \(T_1=(T_++T_+^\dagger)/2\), \(T_3=\mathrm{diag}((k+1)/2-j)\). For \(U=e^{\ii\pi T_1/2}\), \(U T_3 U^\dagger=T_2\) follows by the two rotation differential equations, so \(U(T_1+\ii T_3)U^\dagger=T_+\). Define \(d_1=1,\ d_{j+1}=d_jt_j,\ S=U^\dagger\mathrm{diag}(1/d_j)\). Then \(S^{-1}(T_1+\ii T_3)S=N\) and \(\mathrm{cond}_2(S)=d_k=(k-1)!\). Normalized physical launch and analysis vectors are proportional to \(Se_1\) and \(S^{-\dagger}e_k\). Their matrix element equals the Jordan-coordinate element divided by \(\norm{Se_1}\norm{S^{-\dagger}e_k}=(k-1)!\). Normalize the physical signal generator and the loss budget as well. In the Hamiltonian convention \(g(T_1+\ii T_3)-\ii g(k-1)I/2\), \(g>0\), all gain rates have been shifted to passive loss. That shift reduces throughput. A physical diagonal fabrication error need not be triangular in the Jordan basis.

\textbf{Linear cyclic dual.} The identity \(\Tr[N^r(N^\dagger)^s]=\delta_{rs}(k-r)\) gives the unique dual in \(\mathbb C[N^\dagger]\),
\[
 \Omega(\eta)=\sum_{r=0}^{k-1}\frac{\eta^r}{k-r}(N^\dagger)^r,\qquad
 \Psi(A)=\Tr(A\Omega).
\]
On the full matrix space one can add any \(B\) such that \(\Tr(N^r B)=0\) for all \(0\le r<k\). For \(k=3,\eta=1,\zeta=-1+\ii\), one obtains \(\Psi(e^{\zeta N})=0\), \(\Psi(e^{2\zeta N})=-1-2\ii\), and \(\Psi(NN^\dagger)=2/3\). Complex coherent transfer tomography determines the linear combination; a single input and output realizes only a rank-one functional. An SVD-based sum of such settings is possible, with uncertainty depending on weights and the photon allocation. There is no universal \((\log k)^2\) probe-power law. Duhamel and trace duality give
\[
 |\Psi(e^{z(N+D)}-e^{zN})|\le |z|\norm{\Omega}_1 \norm D\,e^{|z|(1+\norm D)} .
\]

\textbf{Flag restriction.} Upper triangular generators preserve \(V_j=\mathrm{span}(e_1,\ldots,e_j)\). If each defect can increase the flag index by at most one, the first nonzero \(k,1\) entry at zero signal needs \(k-1\) defects. With an integrable upper-triangular base of norm budget \(B\) and total defect norm \(a\), its size is at most \(e^{B+a}a^{k-1}/(k-1)!\) by the Dyson tail. A direct \(E_{k1}\) nuisance gives a first-order background. At nonzero signal, triangular disorder can still alter the signal slope. A known fixed offset is a bias, and becomes variance only under a stated law of fluctuation.