2
$\begingroup$

Let's say we have a pure tripartite state $\rho_{ABE}$ and a completely positive map $\mathcal{R}$, which is defined as:

$$ \mathcal{R} : \rho \rightarrow \sum_j \langle\psi_j|\rho |\psi_j \rangle |\psi_j\rangle\langle\psi_j|, $$ for some density operator $\rho$. Now we apply this map $\mathcal{R}$ to the subsystem $A$ only. So the resulting state is:

$$ \rho'_{ABE} = (\mathcal{R} \otimes \mathcal{I} \otimes \mathcal{I}) (\rho_{ABE}), $$ where $\mathcal{I}$ is the identity map. Now, in this state $\rho'_{ABE}$, how do I prove that:

\begin{equation} H(RB) = H(RE) \tag{1} \end{equation} where, $R$ is the classical state that we get when we apply map $\mathcal{R}$ on subsystem $A$. I know that for pure state $\rho_{ABE}$, any bipartite cut would produce the same entropy, i.e. :

$$ H(AB) = H(E), H(AE) = H(B), H(A) = H(BE). $$ However, I don't think it is the case that, $H(AB) = H(AE)$, isn't it? If so, then how come equation ($1$) is true?

$\endgroup$
0

1 Answer 1

5
$\begingroup$

As $\rho_{ABE}$ is pure we have $\rho_{ABE} = |\psi\rangle \langle \psi|$. We'll rewrite the output of the channel $\mathcal{R}$ as $$ \rho_{ABE}' = \sum_j (P_j \otimes I_{BE}) |\psi\rangle \langle \psi| (P_j \otimes I_{BE}) $$ where $P_j = |\psi_j\rangle \langle \psi_j|$ are a collection of orthogonal rank one projections. Let $\rho_{ABE}^j = (P_j \otimes I_{BE}) |\psi\rangle \langle \psi| (P_j \otimes I_{BE})$. Note that $\rho_{ABE}^j$ is rank one as $\rho_{ABE}^j = |\phi^j\rangle \langle \phi^j|$ with $|\phi^j\rangle = (P_j \otimes I_{BE})|\psi\rangle$.

The important thing to notice is that because the projections are rank one we also have that $\rho_{BE}^j$ is pure. Indeed, $\rho_{BE}^j = |\Phi^j\rangle \langle \Phi^j|$ with $|\Phi_j\rangle = ( \langle\psi_j|\otimes I_{BE})|\psi\rangle$. This means that $$ \rho_B^j = \mathrm{tr}_{AE}[\rho_{ABE}^j] \qquad \text{and} \qquad \rho_E^j = \mathrm{tr}_{AB}[\rho_{ABE}^j] $$ have the same eigenvalues (this can be seen by applying the schmidt decomposition to $\rho_{BE}^j$).

The next thing to show is that this implies that $$ \rho_{AB}' = \sum_j (P_j \otimes I_B) \rho_{AB} (P_j \otimes I_B) \quad \text{and} \quad \rho_{AE}' = \sum_j (P_j \otimes I_E) \rho_{AE} (P_j \otimes I_E) $$ have the same eigenvalues. Note that we can write $$ \rho_{AB}' = \sum_j P_j \otimes \rho_{B}^j. $$ One can now see from the spectral theorem that the eigenvalues of $\rho_{AB}'$ will be the union of the eigenvalues of the different $\rho_{B}^j$. Moreover each eigenvector will be of the form $|\psi_k\rangle \otimes |v^k\rangle$ where $|v^k\rangle$ is an eigenvector of $\rho_{B}^k$ for some $k$. Repeating this for $\rho_{AE}'$ we find that the eigenvalues of $\rho_{AB}'$ and $\rho_{AE}'$ must be equal and hence we also have $H(RB) = H(RE)$.

$\endgroup$
4
  • $\begingroup$ Your answers are always great, Rammus. Thanks a lot! $\endgroup$ Commented Jul 14, 2021 at 18:29
  • 1
    $\begingroup$ No problem, it's a cute little lemma that I'd never seen before so thanks for the question. Seems like it could be useful in cryptography. $\endgroup$
    – Rammus
    Commented Jul 14, 2021 at 18:38
  • 1
    $\begingroup$ Indeed! It appeared in this paper: arxiv.org/pdf/0909.0950.pdf , page 6. The result from the paper (H(R|E) + H(S|B) >= some amount)) is used to lower bound Eve's(E) information about Alice's measurement outcomes and consequently proving security of QKD protocols in the crypto community. $\endgroup$ Commented Jul 14, 2021 at 18:44
  • $\begingroup$ @Rammus Why can we write $\rho_{AB}'=\sum_{j} P_j \otimes \rho_B^j$? It seems like there should be an additional term like $\langle \psi_j|\rho_A | \psi_j \rangle$. $\endgroup$
    – hzxscyq
    Commented Dec 5 at 17:04

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.