2
$\begingroup$

Let $\rho$ and $\sigma$ be two density operators such that probability of obtaining $a$ is $tr(\rho E_a)$ if the state before measurement was $\rho$ and $tr(\sigma E_a)$ if the state before measurement was $\sigma$, where $E_a$'s are 1-dim orthogonal projectors with $\sum_a E_a=I$. In that case we have \begin{equation} d(p,q)=\sum_a|p_a-q_a|\leq||ρ-σ||_1=d(ρ,σ) \end{equation} How should I proceed to find the optimal set of projectors $E_a$'s such that the inequality in the above equation becomes an equality?

$\endgroup$

1 Answer 1

1
$\begingroup$

You're essentially asking how to prove one of the characterisations of the trace distance: given any pair of states $\rho,\sigma$, you have $$\|\rho-\sigma\|_1 = \max_{P}\operatorname{tr}[P(\rho-\sigma)],$$ with the maximisation performed with respect to all Hermitian operators $P$ such that $0\le P\le I$. Equivalently, the optimisation is performed with respect to all projections $P$.

You can find the complete proof in the Wikipedia article. For what concerns finding the maximum, you simply need to use as $P$ the projection onto the subspace spanning the positive eigenvalues of $\rho-\sigma$. With this choice, call it $P^*$, you have $$\operatorname{tr}[P^*(\rho-\sigma)] = \sum_k \lambda_k^+$$ with $\lambda_k^+$ the positive eigenvalues of $\rho-\sigma$. Because $\operatorname{tr}(\rho-\sigma)=0$, the sum of all eigenvalues vanishes, hence $\sum_k\lambda_k^+=-\sum_k\lambda_k^-$ with $\lambda_k^-<0$ the negative eigenvalues, and therefore the trace distance, which equals in general (for Hermitian operators) the sum of the absolute values of the eigenvalues, reads $$\|\rho-\sigma\|_1 = \sum_k |\lambda_k|= \sum_k \lambda_k^+ - \sum_k \lambda_k^- = 2 \sum_k \lambda_k^+ = 2\operatorname{tr}[P^*(\rho-\sigma)].$$

The answer to your question follows immediately from this, because the total variation distance, which is (modulo a factor of 2) the LHS in your equation, equals half of $\operatorname{tr}[P^*(\rho-\sigma)]$, again because $\sum_a (p_a-q_a)=0$.

I also just realised that this is probably already answered in What is the relationship between trace distance and total variation distance?.

$\endgroup$

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.