Consider a $d$-dimensional maximally entangled state $\vert\phi\rangle = \frac{1}{d}\sum_{i=1}^d\vert i\rangle_A\vert i\rangle_B$. Let $N_{A\rightarrow A'}$ be a quantum channel and consider $\rho_{A'B} = (N\otimes I_B)\vert\phi\rangle\langle\phi\vert$. I am interested in the set of nearby quantum states $S = \{\tilde{\rho}_{A'B}\ |\ \|\tilde{\rho} - \rho\|_1\leq \varepsilon, \tilde{\rho}_B = \rho_B\}$ for some $\varepsilon\in [0,1]$.

For any $\tilde{\rho}\in S$, does there exist a channel $\tilde{N}_{A\rightarrow A'}$ that outputs it given a maximally entangled input? That is $(\tilde{N}\otimes I_B)\vert\phi\rangle\langle\phi\vert = \tilde{\rho}_{A'B}$? If not, what is a good counterexample?

If such a $\tilde{N}$ exists, then is it close to $N$ in diamond distance as a function of $\varepsilon$?


2 Answers 2


Yes, the channel $\tilde{N}$ necessarily exists.

Notice first that the state $\rho_B$ is the completely mixed state $\mathbb{1}/d$. So, in order for $\tilde{\rho}_{A'B}$ to be contained in $S$, three things must be true:

  1. $\tilde{\rho}_{A'B}$ must be positive semidefinite.
  2. $\tilde{\rho}_{B} = \mathbb{1}/d$.
  3. $\|\rho_{A'B} - \tilde{\rho}_{A'B}\|_1 \leq \varepsilon$.

In general, the state $$ (M \otimes I_B) | \phi \rangle \langle \phi | $$ uniquely determines the mapping $M$ (for any $M$): up to the normalization $1/d$, this is the Choi representation (or Choi-Jamiolkowski representation) of $M$.

Thus, there is a unique map $\tilde{N}$ such that $\tilde{\rho}_{A'B} = (\tilde{N}\otimes I_B)|\phi\rangle\langle \phi|$. The first condition listed above guarantees that $\tilde{N}$ is completely positive and the second condition guarantees that $\tilde{N}$ preserves trace, so $\tilde{N}$ must in fact be a channel.

Concerning the closeness of $\tilde{N}$ to $N$ in diamond distance, the best you can conclude without more information is that $\|\tilde{N} - N\|_{\diamond} \leq d \varepsilon$. See this answer on Theoretical Computer Science Stack Exchange for further details.

  1. It is sufficient but not necessary for a channel $N'$ to be $\epsilon-$close in diamond distance with $N$.
  2. I think a post-processing noisy channel $E_{A'\to A'}$ can always be found such that $\tilde{N} = (E_{A'}\otimes I)\circ N$ and $(E_{A'}\otimes I_B)\circ N_{A\to A'}\otimes I (\phi_{AB}) = E_{A'}\otimes I_B (\rho_{A' B}) = \tilde{\rho}_{A'B}$.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.