C006 derivation

One protocol cycle is a trace-preserving linear map on density matrices, represented by the superoperator $S=S_C S_B S_A$ acting on $\mathrm{vec}(\rho)$. Its spectrum $\{\lambda_k\}$ satisfies $|\lambda_k|\le1$. The unique steady state is the singlet, $\lambda_0=1$ with right eigenvector $\mathrm{vec}(|\Psi^-\rangle\langle\Psi^-|)$ (verified in C008). Order the remaining eigenvalues by magnitude; let $\lambda_2$ be the largest with $|\lambda_2|<1$.

Decompose the initial deviation $\delta\rho_0=\rho_0-|\Psi^-\rangle\langle\Psi^-|$ in the eigenbasis. After $N$ cycles, $$\rho_N = |\Psi^-\rangle\langle\Psi^-| + \sum_{k\ge1} c_k\,\lambda_k^{N}\,R_k,$$ so the singlet error is $$\epsilon_N = 1 - F = 1-\langle\Psi^-|\rho_N|\Psi^-\rangle = -\sum_{k\ge1}c_k\lambda_k^N\langle\Psi^-|R_k|\Psi^-\rangle.$$ For large $N$ the sum is dominated by the slowest mode $|\lambda_2|$: $$\epsilon_N \sim |\lambda_2|^N = e^{N\ln|\lambda_2|} = e^{-N/N_0},\qquad N_0\equiv\frac{-1}{\ln|\lambda_2|}.$$ This is exactly the claimed exponential decay $\epsilon\propto e^{-N/N_0}$. The numeric run confirms (i) $\ln\epsilon_N$ is linear in $N$ on the tail and (ii) its slope equals $\ln|\lambda_2|$ from the independent eigen-analysis, with $N_0\approx7.62$ at the optimal parameters $(\Phi=\pi/4,\ \gamma=0.221\pi,\ \theta=0.716\pi)$, matching the paper's $N_0\approx7.62$.