Amortization collapse for superchannel divergences

Problem

Does amortization collapse for the geometric Rényi divergence of arbitrary finite-dimensional quantum superchannels? For compatible states, let

\begin{equation} \begin{aligned} D_{\max}(\rho\|\sigma) &:=\inf\{\lambda:\rho\leq2^\lambda\sigma\},\\ \widehat D_\alpha(\rho\|\sigma) &:=\frac1{\alpha-1}\log_2\operatorname{Tr}\!\left[ \sigma\bigl(\sigma^{-1/2}\rho\sigma^{-1/2}\bigr)^\alpha \right],\qquad 1<\alpha\leq2, \end{aligned} \tag{1} \end{equation}

with the standard support conventions. For either divergence \(\mathbf D\in\{D_{\max},\widehat D_\alpha\}\), define its channel extension and channel-amortized extension by

\begin{equation} \begin{aligned} \mathbf D_{\rm ch}(\mathcal N\|\mathcal M) &:=\sup_{\rho_{RA}} \mathbf D(\mathcal N(\rho)\|\mathcal M(\rho)),\\ \mathbf D_{\rm ch}^{A}(\mathcal N\|\mathcal M) &:=\sup_{\rho_{RA},\sigma_{RA}} \{\mathbf D(\mathcal N(\rho)\|\mathcal M(\sigma)) -\mathbf D(\rho\|\sigma)\}, \end{aligned} \tag{2} \end{equation}

In Eq. (2), identity maps on \(R\) are implicit, and the optimizations allow an arbitrary reference of sufficient finite dimension. For superchannels \(\Theta_1,\Theta_2\), set

\begin{equation} \begin{aligned} \mathbf D_{\rm sc}(\Theta_1\|\Theta_2) &:=\sup_{\mathcal N} \mathbf D_{\rm ch}(\Theta_1(\mathcal N)\|\Theta_2(\mathcal N)),\\ \mathbf D_{\rm sc}^{A}(\Theta_1\|\Theta_2) &:=\sup_{\mathcal N,\mathcal M} \{\mathbf D_{\rm ch}^{A} (\Theta_1(\mathcal N)\|\Theta_2(\mathcal M)) -\mathbf D_{\rm ch}^{A}(\mathcal N\|\mathcal M)\}. \end{aligned} \tag{3} \end{equation}

Is

\begin{equation} \mathbf D_{\rm sc}^{A}(\Theta_1\|\Theta_2) =\mathbf D_{\rm sc}(\Theta_1\|\Theta_2) \tag{4} \end{equation}

for \(\mathbf D=\widehat D_\alpha\) in Eq. (1), \(1<\alpha\leq2\), and all superchannel pairs? The definitions also include \(D_{\max}\) to state its settled subcase below. The amortized suprema use pairs with finite subtracted divergence; other infinite values have the standard support convention.

Source

Hirche’s Remark 5.4, after Eqs. (75)–(77), leaves amortization collapse open in general superchannel discrimination settings [Hir23]. For the nested definition in Eq. (3), the max-relative-entropy case follows directly from CP order, as shown below; the remaining target here is the geometric Rényi case.

Progress

For point-to-point channels, amortization of the max-relative entropy collapses:

\begin{equation} D_{\max,{\rm ch}}^{A}(\mathcal N\|\mathcal M) =D_{\max,{\rm ch}}(\mathcal N\|\mathcal M). \tag{5} \end{equation}

The proof of Eq. (5) uses the CP-order structure of channel max-relative entropy [WBHK20].
Fang and Fawzi proved the analogous point-to-point channel collapse

\begin{equation} \widehat D_{\alpha,{\rm ch}}^{A}(\mathcal N\|\mathcal M) =\widehat D_{\alpha,{\rm ch}}(\mathcal N\|\mathcal M), \qquad 1<\alpha\leq2, \tag{6} \end{equation}

from a chain rule for the geometric Rényi divergence [FF21]. Equation (6) concerns point-to-point channels; it does not settle the superchannel claim.
For the nested superchannel definition, the max-relative-entropy subcase also collapses. Indeed, let \(t=2^{D_{\max,\mathrm{ch}}(\mathcal N\|\mathcal M)}\) and \(s=2^{D_{\max,\mathrm{sc}}(\Theta_1\|\Theta_2)}\) be finite. Channel max-relative entropy is characterized by CP order, and physical superchannels preserve that order. Thus

\begin{equation} \mathcal N\leq_{\mathrm{CP}}t\mathcal M \quad\Longrightarrow\quad \Theta_1(\mathcal N)\leq_{\mathrm{CP}}t\Theta_1(\mathcal M) \leq_{\mathrm{CP}}ts\Theta_2(\mathcal M), \tag{7} \end{equation}

where \(\mathcal A\leq_{\mathrm{CP}}\mathcal B\) means \(\mathcal B-\mathcal A\) is completely positive. Taking logarithms in Eq. (7) and using Eq. (5) bounds each term of the amortized supremum by \(\log_2s\). Choosing \(\mathcal N=\mathcal M\) gives the reverse inequality, proving Eq. (4) for \(\mathbf D=D_{\max}\). If the unamortized divergence is infinite, the same reverse inequality already proves equality in the extended sense. This is a direct consequence of the cited channel CP-order characterization [WBHK20]; it does not prove collapse of Hirche’s larger fully amortized quantities [Hir23].

Comment

The remaining displayed question is geometric Rényi amortization collapse for nested-adaptive superchannel discrimination. The max-relative-entropy subcase is settled by Eq. (7). Whether the larger fully amortized divergences controlling braided and fully general strategies collapse is a further, stronger question.

References

[WBHK20]: M. M. Wilde, M. Berta, C. Hirche, and E. Kaur, “Amortized Channel Divergence for Asymptotic Quantum Channel Discrimination,” Letters in Mathematical Physics 110, 2277–2336 (2020).DOI arXiv
[FF21]: K. Fang and H. Fawzi, “Geometric Rényi Divergence and its Applications in Quantum Channel Capacities,” Communications in Mathematical Physics 384, 1615–1677 (2021).DOI arXiv
[Hir23]: C. Hirche, “Quantum Network Discrimination,” Quantum 7, 1064 (2023).DOI arXiv

Amortization collapse for superchannel divergences

Problem

Source

Progress

Comment

References

Cite this problem

BibTeX

Plain text

Problem

Source

Progress

Comment

References

Related problems