Download research/visibility.tex from PureOne/CNISA-II: direct link, hf CLI and curl.
- Browser
- Download file 3.5 kB
-
https://huggingface.co/datasets/PureOne/CNISA-II/resolve/main/research/visibility.tex
- Command line
-
hf download hf://datasets/PureOne/CNISA-II/research/visibility.tex
-
curl -L -o visibility.tex https://huggingface.co/datasets/PureOne/CNISA-II/resolve/main/research/visibility.tex
3.5 kB
| \page | |
| \sect{Independent phase drift and imperfect mode matching}\label{sec:visibility} | |
| The common-mixing assumption admits a measurable relaxation. Replace \(K(t)\otimes I\) in \eqref{eq:model} with \(\sum_j K_j(t)\otimes E_{jj}\), where each \(K_j(t)\) is Hermitian and can be large. This permits independent propagation phases and arbitrary sideband mixing within each internal vertex. Denote the propagator on vertex \(j\) by \(W_j(t,s)\). Use the same off-diagonal graph and budget \(a\). At zero signal the signed dark offset still obeys \eqref{eq:null}: diagonal internal motion never changes the vertex visited by a word. | |
| \thm{Corollary}{The information bound with calibrated visibility} | |
| For an input sideband state \(\xi\), define the dimensionless complex visibility | |
| \[ | |
| c(\xi)=\frac1L\int_0^L | |
| \left\langle W_u(L,0)\xi,\,W_v(L,t)W_u(t,0)\xi\right\rangle\dd t. | |
| \] | |
| Choose an analyzer phase \(\varphi\) and suppose \(\operatorname{Re}(e^{\ii\varphi}c)\ge\nu>a^2\) uniformly over the allowed backgrounds. Then Theorem 3 holds with numerator \((\nu-a^2)^2\) in place of \((1-a^2)^2\), at the same incident photon number and efficiency. The clean reference in this ratio remains the perfectly matched \(4n\tau L^2\). | |
| \prooftext In the absence of edge defects, at zero signal the endpoint components are \(x_u=W_u(L,0)\xi\) and \(x_v=0\). Differentiation in the direct signal gives \(x'_v=-\ii\int_0^L W_v(L,t)W_u(t,0)\xi\dd t\). For \(O_\varphi=\ii e^{\ii\varphi}E_{uv}-\ii e^{-\ii\varphi}E_{vu}\), the signed slope is exactly \(2L\operatorname{Re}(e^{\ii\varphi}c)\). Even at nonzero signal the defect-free propagator preserves the endpoint plane and its orthogonal complement. The two-insertion argument in Theorem 1 therefore bounds the change in slope by \(2La^2\). The proof of Theorem 3 then applies without change. \fin | |
| Define the remaining mismatch by \(r(t)=\int_t^L\norm{K_v(s)-K_u(s)}\dd s\). The exact one-insertion Duhamel identity gives \(\norm{W_v(L,t)-W_u(L,t)}\le r(t)\) when the difference is bounded and integrable. The two vectors in the integrand of \(c\) are unit vectors, and \(1-\operatorname{Re}\langle x,y\rangle=\norm{x-y}^2/2\). Consequently | |
| \[ | |
| \operatorname{Re} c\ \ge\ 1-\frac{1}{2L}\int_0^L r(t)^2\dd t . | |
| \] | |
| Strongly differing \(K_j\) on intermediate vertices do not enter this clean visibility. They can modify coefficients of leakage words, subject to the same graph-distance bound. | |
| \smhead{An exact failure test} | |
| If \(K_u=0\), \(K_v=\Delta I\) are constant, then \(c=e^{-\ii\Delta L/2}\operatorname{sinc}(\Delta L/2)\), where \(\operatorname{sinc}x=\sin(x)/x\). At \(\Delta L=2\pi\) the signed slope vanishes for every analyzer phase despite perfect absence of a dark background. A port can therefore be dark, satisfy the predicted leakage exponent, and still be useless for linear sensing. Measuring the calibrated signal slope as well as the dark floor distinguishes these cases. | |
| \smhead{Physical meaning of the direct link} | |
| For example, a weak external optical perturbation could couple two modes through an overlap integral while internal coupler fluctuations predominantly connect adjacent auxiliary modes. The same external perturbation may also shift a propagation constant or couple both endpoints through a bypass. Transfer tomography under independently varied signal and nuisance controls can distinguish these generators. The theorems become predictions for that transducer after its measured generators and visibility meet the stated hypotheses. | |