A tight uniform continuity bound for the Arimoto-Rényi conditional entropy and its extension to classical-quantum states (2007.05049v3)

Abstract: We prove a tight uniform continuity bound for Arimoto's version of the conditional $\alpha$-R\'enyi entropy, for the range $\alpha \in [0, 1)$. This definition of the conditional R\'enyi entropy is the most natural one among the multiple forms which exist in the literature, since it satisfies two desirable properties of a conditional entropy, namely, the fact that conditioning reduces entropy, and that the associated reduction in uncertainty cannot exceed the information gained by conditioning. Furthermore, it has found interesting applications in various information theoretic tasks such as guessing with side information and sequential decoding. This conditional entropy reduces to the conditional Shannon entropy in the limit $\alpha \to 1$, and this in turn allows us to recover the recently obtained tight uniform continuity bound for the latter from our result. Finally, we apply our result to obtain a tight uniform continuity bound for the conditional $\alpha$-R\'enyi entropy of a classical-quantum state, for $\alpha$ in the same range as above. This again yields the corresponding known bound for the conditional entropy of the state in the limit $\alpha \to 1$.