1
$\begingroup$

It is well known that the Bures distance between two infinitesimally close density matrices induces the Bures metric. Likewise the quantum Fisher information metric is related to the infinitesimal version of the quantum relative entropy, I believe (please correct me if I'm wrong).

How about for the trace distance $\Delta(\rho,\sigma):=\frac{1}{2}\|\rho-\sigma\|_1$ between two infinitesimally close density matrices $\rho$, $\sigma=\rho+d\rho$? It does not seem like there is any mention of what metric it induces (if it does).

More broadly, when would one use the Bures metric vs quantum Fisher information metric vs Fubi-Study metric?

$\endgroup$

1 Answer 1

2
$\begingroup$

Any vector space with a definition of a vector norm can be used to write down a metric. In this case, because one can compute the "distance" between $\rho$ and $\sigma$ by the norm of $\rho-\sigma$, one has already defined a metric. You could call this the trace distance metric, it is different from the other metrics you mentioned, and does not possess a clean geometric visualization (https://quantumcomputing.stackexchange.com/a/29716/15820).

Now let's try to get a handle on how this metric relates to other quantities that you mentioned. Suppose that $\rho=\rho_{\theta-d\theta/2}$ and $\sigma=\rho_{\theta+d\theta/2}$. With this infinitesimal change, the fidelity $F$ is related to the quantum Fisher information $\mathcal{F}$ by $F(\rho,\sigma)=1-\mathcal{F}(\rho_\theta)d\theta^2/4$ (QFI is not related to quantum relative entropy, even though classical Fisher information is related to classical relative entropy https://quantumcomputing.stackexchange.com/a/29716/15820). In addition, the trace distance is bounded from both sides by functions of the fidelity as $1-\sqrt{F(\rho,\sigma)}\leq \Delta(\rho,\sigma)\leq \sqrt{F(\rho,\sigma)}$, from which we conclude $$\frac{\mathcal{F}(\rho_\theta)}{8}d\theta^2\leq \Delta(\rho_{\theta-d\theta/2},\rho_{\theta+d\theta/2})\leq \frac{\sqrt{\mathcal{F}(\rho_\theta)}}{2}d\theta. $$ So, at the very least, we learn that the trace distance grows at a rate somewhere between $\mathcal{O}(d \theta)$ and $\mathcal{O}(d \theta^2)$.

It may be even easier to just use the definition of the trace distance: $\Delta(\rho,\rho+d\rho)=\frac{1}{2}\mathrm{Tr}(\sqrt{d\rho^\dagger d\rho})$. Then we notice that $d\rho^\dagger=d\rho$ because the difference between two Hermitian operators must also be Hermitian, leading us to $\Delta(\rho,\rho+d\rho)\propto \mathrm{Tr}(\sqrt{d\rho^2})$. This is the sum of the singular values (absolute values of the eigenvalues) of $d\rho$, so you can see how this could grow like $|d\theta|$ and not have smooth derivatives. Easy examples are $d\rho=|1\rangle\langle 1|d\theta+|2\rangle\langle 2|d\phi$ for some orthogonal states $\langle 1|2\rangle=0$, from which $\Delta=(|d\theta|+|d\phi|)/2$, opening like an inverted pyramid when plotted versus $d\theta$ and $d\phi$.

Your final paragraph is probably the subject of another question, so I will leave it for now.

$\endgroup$

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.