2
$\begingroup$

I'm currently working with the continuity result by Kretschmann-Schlingemann-Werner (arXiv version) for Stinespring isometries (more precisely, the following corollary to their result, cf. Appendix C in this paper):

Given channels $\Phi_1,\Phi_2\in\mathcal L(\mathbb C^{d\times d})$ with respective Kraus rank $r_1,r_2$, as well as any $m\in\mathbb N$ with $m\geq2\max\{r_1,r_2\}$, for all Stinespring isometries $V_1,V_2:\mathbb C^d\to\mathbb C^d\otimes\mathbb C^m$ of $\Phi_1,\Phi_2$ there exists $U \in\mathsf U(m)$ such that $$ \|V_1-({\bf 1}\otimes U)V_2\|_\infty\leq \sqrt{\|\Phi_1-\Phi_2\|_\diamond}\,. $$

I was interested in how good this bound is in general, which is why I got interested in finding examples of channels where $\|\Phi_1-\Phi_2\|_\diamond$ is known. However, the only literature I know of in this direction are:

  • the paper of Johnston et al. (arXiv version) on the (diamond norm-)distance between unitary channels.
  • the paper of Nechita et al. (arXiv version) on the distance between random channels. Therein they give an upper bound on the diamond norm in terms of the partial trace of the absolute value of the Choi-Jamiołkowski matrix, and they even characterize when this bound is saturated. Moreover, they study the distance between random channels and the maximally depolarizing channel $X\mapsto{\rm tr}(X)d^{-1}{\bf 1}_d$—as well as distances involving random unitary channels—in the asymptotic limit $d\to\infty$.
  • more general results on bounding the diamond norm, e.g., the inequality $\|\Phi-{\rm id}\|_\diamond\leq 2\sqrt{\|\Phi-{\rm id}\|_{1\to 1}}$ between diamond norm and the operator norm induced by the trace norm from Watrous's book.

From these, the only analytic result on channel distances seems to be Theorem 12 in the Johnston paper which states that $\|{\rm id}-U(\cdot)U^*\|_\diamond$ is equal to "the diameter of the smallest closed disc that contains all of the eigenvalues of $U$".

Beyond that I proved the following distance results myself (but I am not yet aware of any paper or book that features them):

Let $\Phi_d$ denote the dephasing channel (i.e. $\Phi_d$ sets all off-diagonal elements of the input to zero) and let $\Phi_{\bf 1}:X\mapsto{\rm tr}(X)d^{-1}{\bf 1}$ denote the depolarization channel in dimension $d$. Then $$\|{\rm id}-\Phi_{\bf 1}\|_{1\to 1}=\|{\rm id}-\Phi_d\|_{1\to 1}=\|{\rm id}-\Phi_d\|_\diamond=2-\frac2d\leq2-\frac2{d^2}=\|{\rm id}-\Phi_{\bf 1}\|_\diamond \,.$$

Now my question is twofold:

  1. Are these distance results regarding dephasing and depolarization known, and if so, in what paper/book can I find them?
  2. Beyond what I listed above, are there any resources which contain distance results of this type (i.e. analytic expressions for $\|{\rm id}-\Phi\|_\diamond$ for specific channels $\Phi$)? For example, a generalization of the depolarization-result would be to consider arbitrary reset channels $\Phi_\rho:X\mapsto {\rm tr}(X)\rho$ (with $\rho$ some state). While numerics suggest a closed-form expression of $\|{\rm id}-\Phi_\rho\|_\diamond$, my proof of the depolarization result breaks down when replacing ${\bf 1}/d$ by an arbitrary state.
$\endgroup$
1
  • 1
    $\begingroup$ In the meantime, the reason why I asked this question turned into a paper where I try to extend upon the Kretschmann-Schlingemann-Werner bound. The gist of it is that one can probably get rid of the dimension limitation of the dimension of the Stinespring isometry if one adds an addition factor of $\sqrt2$ to the upper bound. However, I was only able to prove it for certain special cases; the general case remains a conjecture $\endgroup$ Mar 16 at 15:26

1 Answer 1

1
$\begingroup$

You can find some useful results in these links: https://quantum-journal.org/papers/q-2021-08-09-522/, https://arxiv.org/abs/1004.4110v1, https://doi.org/10.1088/1367-2630/ab8e5c, https://arxiv.org/abs/1805.08227, and in other papers discussing coherent noise.

$\endgroup$
1
  • $\begingroup$ Much appreciated! The paper of Regula et al. has some handy characterizations of the diamond norm (in particular Thm. 3 for Herm.-pres., trace-annihilating maps); however, I have to play around with it some more to see how useful this is for actually computing channel distances $\endgroup$ Jun 29, 2023 at 8:48

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.