arXiv:2609.29885
The regularized channel Rényi divergence is continuous
Lukas Schmitt, David Sutter
Abstract
We show that the regularized channel Rényi divergence is continuous at $α= 1$. As consequences, we establish an exponentially strong converse for channel discrimination, a subchannel-smoothed asymptotic equipartition property, a relative entropy accumulation theorem, and a new proof of the generalized quantum Stein's lemma. Our result further shows that the regularized channel relative entropy is computable. The continuity proof relies on a variational formula that expresses the regularized channel divergence as a single-letter optimization over variables of unrestricted finite dimension. We then establish continuity using a novel spectral confinement technique.
AI Summary
The paper asks whether taking the Rényi parameter to $1$ commutes with regularizing a quantum-channel divergence over tensor powers. For finite-dimensional systems, it proves $\lim_{\alpha\downarrow1}D_\alpha^\infty(\mathcal E\|\mathcal F)=D^\infty(\mathcal E\|\mathcal F)$ whenever $\mathcal E$ is CPTP and $\mathcal F$ is completely positive. Together with the standard left-limit argument, this establishes continuity at $\alpha=1$; the result also extends to the Petz Rényi divergence. The proof introduces an additive variational expression for the regularized divergence, formulated using one channel use but allowing arbitrarily large finite reference systems. Its identification with the regularized divergence uses measured-divergence variational formulas, polynomial-rank bounds, pinching, and an existing channel chain rule. A spectral confinement argument then controls near-optimal witnesses within bounded intervals of a logarithmic spectrum, enabling continuity estimates independent of reference dimension. Consequences include computability of $D^\infty$, equality of parallel and adaptive channel Stein exponents at every fixed type-I error tolerance in $(0,1)$, and an exponential strong converse: imposing type-II error at most $e^{-nr}$ for $r>D^\infty$ forces success probability below $e^{-nc_r}$ for some $c_r>0$. Further applications are a channel AEP, an asymptotic relative entropy accumulation theorem, and an alternative generalized quantum Stein lemma proof. The variational formula does not bound reference dimension, and computability does not imply efficiency. The channel AEP smooths over trace-nonincreasing subchannels; CPTP-only smoothing can fail. The generalized Stein statement assumes an existing relative-entropy-rate limit and a strong converse for the null sequence. Two independent continuity proofs using different techniques are noted.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 23 pages