arXiv:2609.34213
Monotonicity of the Rényi channel capacity under non-signaling assisted channel simulation
Tina Shariat, Matt Hoogsteder-Riera, Marco Tomamichel
Abstract
We give an alternative variational proof that non-signaling correlations shared between the sender and receiver cannot increase the Rényi capacity of a classical channel. This monotonicity follows from the data-processing inequality for channel divergences established by Oufkir, Yao and Berta. Our approach works directly with the information-radius representation of the capacity. The constraints defining non-signaling boxes are linear in the box, while the Rényi capacity depends on the channel nonlinearly. We use a Legendre transform to express the capacity as an extremum over a family of affine functionals in which the channel appears linearly. Combining this method with separate arguments for the Shannon capacity and endpoint orders, we recover monotonicity at every order from zero to infinity. We then derive consequences for standard coding exponent functions: non-signaling simulation cannot increase the random-coding or sphere-packing exponent below capacity, or decrease the strong-converse exponent above capacity.
AI Summary
The paper studies whether non-signaling-assisted simulation of a classical channel can increase its Rényi capacity. For finite-alphabet channels, it establishes $R_\alpha(W')\le R_\alpha(W)$ for every $\alpha\in[0,\infty]$ whenever a non-signaling box simulates $W'$ from $W$. The paper explicitly identifies this theorem as a consequence of an existing data-processing inequality for channel divergences; its contribution is an alternative variational proof. The proof works with the information-radius representation $R_\alpha(W)=\min_Q\max_x D_\alpha(W(\cdot\mid x)\|Q)$. Young/Legendre representations make the channel dependence linear, allowing the non-signaling constraints and Jensen’s inequality to yield monotonicity separately for $0<\alpha<1$ and $\alpha>1$. Separate arguments handle Shannon capacity and the endpoints. At infinity, $R_\infty(W)=\log\sum_y\max_xW(y\mid x)$, and only the no-forward-signaling condition is required, provided the simulated map remains a channel. Consequences include nonincrease of the random-coding and sphere-packing exponent curves and nondecrease of the strong-converse exponent. Every non-signaling-assisted $(n,M)$ code satisfies $p_{\mathrm{succ}}\le\exp[-\frac{\alpha-1}{\alpha}(\log M-nR_\alpha(W))]$ for $\alpha>1$, recovering the known equality of assisted and unassisted strong-converse exponents. These statements do not imply that assistance cannot improve below-capacity coding reliability. Exact unassisted reliability is monotone above the base channel’s critical rate, can fail to be monotone between its unassisted and non-signaling zero-error capacities, and remains unresolved between $R_0(W)$ and the critical rate. Extending the variational proof to classical–quantum channels is left open.
Claude assistance has been used to derive the proofs of Lemmas 2 to 5 and to assist in writing the manuscript.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 25 pages, 1 figure