Having some trouble showing that $S(\rho_{XB}||\sigma_{XB})=\sum_{x}p(x)D(\rho_{B}^{x}||\sigma_{B}^{x})$ for $\rho_{XB}=\sum_{x}p(x)|x\rangle\langle x|\otimes\rho_{B}^{x}$ and $\sigma_{XB}=\sum_{x}p(x)|x\rangle\langle x|\otimes\sigma_{B}^{x}$

I know that $$S(\rho_{XB}||\sigma_{XB})=\mathrm{Tr}(\rho_{XB}\log(\rho_{XB}))-\mathrm{Tr}(\rho_{XB}\log(\sigma_{XB}))$$ and $$\mathrm{Tr}(\rho_{XB}\log(\rho_{XB}))=-S(X)-\sum_{x}p(x)S(\rho_{B}^{x})$$ due to its classical-quantum nature. However, this implies that $$-\mathrm{Tr}(\rho_{XB}\log(\sigma_{XB}))=S(X)+\sum_{x}p(x)S(\rho_{B}^{x})+\sum_{x}p(x)D(\rho_{B}^{x}||\sigma_{B}^{x})$$.

Taking $$-\mathrm{Tr}(\rho_{XB}\log(\sigma_{XB}))=-\mathrm{Tr}(\sum_{x}p(x)|x\rangle\langle x|\otimes\rho_{B}^{x} \log(\sum_{x}p(x)|x\rangle\langle x|\otimes\sigma_{B}^{x}))=$$

$$-\mathrm{Tr}(\sum_{x}p(x)|x\rangle\langle x|\otimes\rho_{B}^{x}(\sum_{x}|x\rangle\langle x|\otimes \log(p(x)\sigma_{B}^{x})))=\mathrm{Tr}(\sum_{x}p(x)|x\rangle\langle x|\otimes\rho_{B}^{x}\log(p(x)\sigma_{B}^{x}))=$$



As can be seen, this is not enough to cancel the other terms, which can be seen when putting these two together, you get $$-S(X)-\sum_{x}p(x)S(\rho_{B}^{x})+S(X)+\sum_{x}p(x)D(\rho_{B}^{x}||\sigma_{B}^{x})=\sum_{x}p(x)D(\rho_{B}^{x}||\sigma_{B}^{x})-\sum_{x}p(x)S(\rho_{B}^{x})$$ This could only equal the desired exprerssion if $\rho_{B}^{x}$ were pure states, which is specified nowhere in the text I am reading.

Where have I gone wrong?

Edit: As Rammus stated in his answer, I went wrong with $$-\sum_{x}p(x)(\mathrm{Tr}(\rho_{B}^{x}(\log(p(x)))+\mathrm{Tr}(\rho_{B}^{x}\log(\sigma_{B}^{x})))=-\sum_{x}p(x)\log(p(x))+\sum_{x}p(x)D(\rho_{B}^{x}||\sigma_{B}^{x})$$ Instead, it should be $$-\sum_{x}p(x)(\mathrm{Tr}(\rho_{B}^{x}(\log(p(x)))+\mathrm{Tr}(\rho_{B}^{x}\log(\sigma_{B}^{x})))=-\sum_{x}p(x)\log(p(x))-\sum_{x}p(x)Tr(\rho_{B}^{x}log(\sigma_{B}^{x}))$$ Putting this back into the original equation gives the desired equality.

  • $\begingroup$ can you rewrite the latex commands with \ln instead of log. It will make your problem more readable. $\endgroup$
    – Jaswin
    Sep 22, 2021 at 11:38
  • $\begingroup$ Please try to use aligned environments as well when you have multiple equalities in a row. $\endgroup$
    – Rammus
    Sep 22, 2021 at 12:03
  • $\begingroup$ @Rammus on the subject of the aligned environments, and possibly other notation based etiquette, does this forum have a post that details that? Obviouslt if there is a way to make these more readable to those providing answers, I will happily do so. $\endgroup$ Sep 22, 2021 at 15:02
  • $\begingroup$ @GaussStrife I don't know of one but I would just aim to write as you would expect to look if it was in a book or a research article. I.e., $\log$ instead of $log$, well spaced equations, explanatory text where necessary etc, $\endgroup$
    – Rammus
    Sep 22, 2021 at 15:30

1 Answer 1


As you say, $$ \mathrm{Tr}[\rho_{XB} \log \rho_{XB}] = -S(X) + \sum_{x} p(x) \mathrm{Tr}[\rho_{B}^x \log \rho_B^x]. $$

But if you can prove the above statement, then the exact same derivation gives you $$ \mathrm{Tr}[\rho_{XB} \log \sigma_{XB}] = - S(X) + \sum_{x} p(x) \mathrm{Tr}[\rho_B^x \log \sigma_B^x]. $$

If you put both together then you get $$ \begin{aligned} D(\rho_{XB} \|\sigma_{XB}) &= \mathrm{Tr}[\rho_{XB} \log \rho_{XB}] - \mathrm{Tr}[\rho_{XB} \log \sigma_{XB}] \\ &= -S(X) + \sum_{x} p(x) \mathrm{Tr}[\rho_{B}^x \log \rho_B^x] + S(X) - \sum_{x} p(x) \mathrm{Tr}[\rho_B^x \log \sigma_B^x] \\ &= \sum_{x} p(x) \mathrm{Tr}[\rho_{B}^x \log \rho_B^x] - \sum_{x} p(x) \mathrm{Tr}[\rho_B^x \log \sigma_B^x] \\ &= \sum_{x} p(x) \left(\mathrm{Tr}[\rho_{B}^x \log \rho_B^x] - \mathrm{Tr}[\rho_B^x \log \sigma_B^x]\right) \\ &= \sum_{x} p(x) D(\rho_{B}^x \|\sigma_B^x) \end{aligned} $$

To comment on your derivation

You have several errors in the 7th and 8th equalities in your long derivation.

  • $\begingroup$ Yes, I foolishly equated $-Tr(\rho_{B}^{x}log\sigma_{B}^{x})=D(\rho_{B}^{x}||\sigma_{B}^{x})$ on line 8. What mistake did I make on line 7? $\endgroup$ Sep 22, 2021 at 12:19
  • 1
    $\begingroup$ You have a $\mathrm{tr}[\log \sigma_B^x]$. $\endgroup$
    – Rammus
    Sep 22, 2021 at 12:53
  • $\begingroup$ Ah yes. I typed that out wrong. My bad. $\endgroup$ Sep 22, 2021 at 12:58

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.