3
$\begingroup$

I have two Pauli operators $\frac{1}{\sqrt{d}} \mathcal{P}_i$, $\frac{1}{\sqrt{d}} \mathcal{P}_j$, and an arbitrary quantum channel $\mathcal{E}$ (in the superoperator/Liouville representation) all acting on a $d=2^n$ dimensional Hilbert space. I want to show that the matrix element $\frac{1}{d} | \text{Tr}(\mathcal{P}_i^\dagger \mathcal{E}(\mathcal{P}_j))| \leq 1$.

Is this true? If so, how can one show this? Unfortunately, all inequalities for trace known to me don't work, because neither $\mathcal{P}_i$ or $\mathcal{E}(\mathcal{P}_j)$ are positive semidefinite.

Related question: Does any quantum channel satisfy ${\rm Tr}(\mathcal E^\dagger \mathcal E) \in[0, d^2]$? Essentially I am looking for a proof that $\frac{1}{d} | \text{Tr}(\mathcal{P}_k^\dagger \mathcal{E}(\mathcal{P}_{k'}))| = |\chi_\mathcal{E} (k, k')| \leq 1$.

$\endgroup$

2 Answers 2

2
$\begingroup$

TL;DR: It's true. You have a good sense that positive semi-definiteness would help. We don't have it for the Paulis, but we have it for the eigenprojectors.

Let $P_+$ and $P_-$ denote the projectors onto the eigenspaces of $\mathcal{P}_i$ associated with the $+1$ and $-1$ eigenvalues, respectively, and similarly for $R_\pm$ and $\mathcal{P}_j$. Also, set $Q_\pm:=\mathcal{E}(R_\pm)$. Then \begin{align} \frac{1}{d}\mathrm{tr}(\mathcal{P}_i^\dagger\mathcal{E}(\mathcal{P}_j)&=\frac{1}{d}\mathrm{tr}((P_+-P_-)^\dagger\mathcal{E}(R_+-R_-))\tag1\\ &=\frac{1}{d}[\mathrm{tr}(P_+Q_+)-\mathrm{tr}(P_-Q_+)-\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-)].\tag2 \end{align} Now, $P_\pm$ and $R_\pm$ are positive semi-definite. Moreover, $\mathcal{E}$ is completely positive, so $Q_\pm$ are positive semi-definite, too. Hilbert-Schmidt inner product of two positive semi-definite operators is non-negative, so $\mathrm{tr}(P_\pm Q_\pm)\geqslant 0$. Therefore, \begin{align} -\mathrm{tr}(P_+Q_+)-\mathrm{tr}(P_-Q_+)-\mathrm{tr}(P_+Q_-)-\mathrm{tr}(P_-Q_-)\tag3\\ \leqslant\mathrm{tr}(P_+Q_+)-\mathrm{tr}(P_-Q_+)-\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-)\tag4\\ \leqslant\mathrm{tr}(P_+Q_+)+\mathrm{tr}(P_-Q_+)+\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-).\tag5 \end{align} However, \begin{align} &\mathrm{tr}(P_+Q_+)+\mathrm{tr}(P_-Q_+)+\mathrm{tr}(P_+Q_-)+\mathrm{tr}(P_-Q_-)\tag6\\ &=\mathrm{tr}((P_++P_-)\mathcal{E}(R_++R_-))\tag7\\ &=\mathrm{tr}(\mathcal{E}(I))=2^n=d\tag8 \end{align} where we used the fact that $\mathcal{E}$ is trace preserving. Substituting into the inequalities above and reintroducing normalization constant $\frac{1}{d}$, we obtain \begin{equation} -1\leqslant \frac{1}{d}\mathrm{tr}(\mathcal{P}_i^\dagger\mathcal{E}(\mathcal{P}_j))\leqslant+1\tag9 \end{equation} as claimed.

$\endgroup$
1
  • $\begingroup$ Perfect! Thank you for the proof and the very cool technique :-) $\endgroup$
    – Jed Burkat
    May 12, 2023 at 2:50
3
$\begingroup$

Here's a one-line proof using Hölder's inequality: $$ |\mathrm{tr}\left( \mathcal{P}_i^\dagger \mathcal{E}(\mathcal{P}_i) \right)| \leq \| \mathcal{P}_i^\dagger \|_\infty \|\mathcal{E}\|_{1\rightarrow1} \| \mathcal{P}_j \|_1 = \| \mathcal{P}_j \|_1 = \mathrm{tr}(I) = d. $$ We only used that quantum channels have unit $1\rightarrow1$ norm, as well that unitaries have unit spectral norm and trace norm $d$. Note this also works for trace non-increasing CP superoperators (and, more generally, for any superoperator in the $1\rightarrow1$ norm ball, or in the dual $\infty\rightarrow\infty$ ball).

Note: $\|\cdot\|_p$ denotes the Schatten $p$-norms, and $\| \cdot \|_{p\rightarrow q}$ is the induced norm on superoperators, i.e. $$ \| \Phi \|_{p\rightarrow q} := \sup_{\|X\|_p\leq 1} \|\Phi(X)\|_q \,. $$

$\endgroup$
6
  • $\begingroup$ how's the "$1\to 1$ norm" defined here? $\endgroup$
    – glS
    May 12, 2023 at 7:44
  • $\begingroup$ @glS it is the operator norm induced by the Schatten 1-norm (trace norm) on both domain and image. You can also take the diamond norm if you like (which is the "stabilized" $1\rightarrow 1$ norm). $\endgroup$ May 12, 2023 at 8:28
  • $\begingroup$ Just to make sure I understand, as this is very new to me. $|\text{Tr}(\mathcal{P}_i^\dagger \mathcal{E}(\mathcal{P}_j))| = \left\Vert \mathcal{P}_i^\dagger \mathcal{E}(\mathcal{P}_j))\right\Vert_1 \leq \left\Vert \mathcal{P}_i^\dagger \right\Vert_\infty \left\Vert\mathcal{E}(\mathcal{P}_j)) \right\Vert_1$ is Hölder's inequality on the Schatten norm. The induced norm (as defined in arxiv.org/pdf/quant-ph/0411077.pdf) is $\left\Vert \mathcal{E} \right\Vert_{q \rightarrow p} = \sup_{X \neq 0} \frac{\left\Vert \mathcal{E}(X) \right\Vert_p}{\left\Vert X \right\Vert_q}$. $\endgroup$
    – Jed Burkat
    May 12, 2023 at 13:41
  • $\begingroup$ Therefore $\left\Vert\mathcal{E}(\mathcal{P}_j)) \right\Vert_1 \leq \left\Vert \mathcal{E} \right\Vert_{1 \rightarrow 1} \left\Vert \mathcal{P}_j \right\Vert_1$, and we substitute this into the first inequality. Correct? $\endgroup$
    – Jed Burkat
    May 12, 2023 at 13:42
  • $\begingroup$ @JedBurkat correct, except that the first equality in your first comment is generally an inequality. The operator norm is in general defined for a linear operator between normed spaces, see also en.m.wikipedia.org/wiki/Operator_norm $\endgroup$ May 13, 2023 at 16:08

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.