Fixed-error parallel Stein lemma for quantum channels
- Fields
- Topics
Problem
Let \(\mathcal N,\mathcal M:\mathcal L(A)\to\mathcal L(B)\) be quantum channels on finite-dimensional systems. For states \(\rho\) and \(\sigma\), define the relative entropy and the hypothesis-testing divergence by
where \(D(\rho\|\sigma)=+\infty\) unless \(\operatorname{supp}\rho\subseteq\operatorname{supp}\sigma\), and \(\varepsilon\in(0,1)\). For either divergence \(\mathbf D\) in Eq. (1), its stabilized channel extension is
Using Eq. (2), define the regularized channel relative entropy by
Whenever Eq. (3) is finite, does the fixed-error parallel Stein limit exist and satisfy
Source
Fang, Gour, and Wang prove the weak channel Stein lemma and identify the fixed-error, equivalently strong-converse-threshold, extension as unresolved for general channel pairs [FGW25].
Progress
The weak channel Stein lemma proves
\begin{equation} \lim_{\delta\downarrow0}\lim_{n\to\infty}\frac1n D_{H,{\rm ch}}^\delta (\mathcal N^{\otimes n}\|\mathcal M^{\otimes n}) =D_{\rm ch}^{\infty}(\mathcal N\|\mathcal M). \tag{5} \end{equation}Equation (5) yields the required lower bound at every fixed error, but not the converse inequality [FGW25].
Sandwiched-Rényi converses bound the fixed-error limsup by
\begin{equation} \limsup_{n\to\infty}\frac1nD_{H,{\rm ch}}^\varepsilon (\mathcal N^{\otimes n}\|\mathcal M^{\otimes n}) \leq\inf_{\alpha>1} \widetilde D_{\alpha,{\rm ch}}^\infty(\mathcal N\|\mathcal M). \tag{6} \end{equation}Closing Eq. (6) requires the still-unproved continuity identity
\begin{equation} \inf_{\alpha>1}\widetilde D_{\alpha,{\rm ch}}^\infty (\mathcal N\|\mathcal M) \stackrel{?}{=}D_{\rm ch}^\infty(\mathcal N\|\mathcal M). \tag{7} \end{equation}Equation (7) is the general strong-converse threshold problem [FGW25].
The equality in Eq. (4) is known when \(\mathcal M\) is a replacer channel [CMW16]. It is also known for suitable pairs of idempotent channels that share a full-rank invariant state; this structured result explicitly leaves the general channel case open [SB26].
For finite \(D_{\max}(\mathcal N\|\mathcal M)\), Gour’s Theorem 14, statements (1), (3), and (4), equates Eq. (4) with a subchannel-smoothed max-relative-entropy AEP at rate \(D_{\rm ch}^{\infty}(\mathcal N\|\mathcal M)\) for every fixed \(\delta\in(0,1)\), and with \(E_{2^{nr}}(\mathcal N^{\otimes n}\|\mathcal M^{\otimes n})\to0\) for every \(r>D_{\rm ch}^{\infty}(\mathcal N\|\mathcal M)\) [Gou26]. Here subchannels are completely positive trace-nonincreasing maps; smoothing is uniform over reference-assisted inputs in generalized trace distance \(\tfrac12(\|X\|_1+|\operatorname{Tr}X|)\), with \(X\) the output difference. The channel hockey-stick divergence \(E_\gamma\) maximizes \(\operatorname{Tr}(\rho-\gamma\sigma)_+\) over common reference-assisted inputs. These equivalent assertions remain open in general.
Comment
The question concerns parallel tests with an arbitrary entangled input across the channel uses. The weak-error result and the known structured channel families do not establish Eq. (4) for an arbitrary finite-dimensional pair \(\mathcal N,\mathcal M\). The failure of CPTP-smoothed AEP does not refute this Stein limit: trace-preserving completion is an additional constraint [Gou26].
References
- [FGW25]
- K. Fang, G. Gour, and X. Wang, “Towards the Ultimate Limits of Quantum Channel Discrimination and Quantum Communication,” Science China Information Sciences 68, 180509 (2025).DOIarXiv
- [CMW16]
- T. Cooney, M. Mosonyi, and M. M. Wilde, “Strong Converse Exponents for a Quantum Channel Discrimination Problem and Quantum-Feedback-Assisted Communication,” Communications in Mathematical Physics 344, 797–829 (2016).DOIarXiv