We extend the recent bounds of Sason and Verdú relating Rényi entropy and Bayesian hypothesis testing [arXiv:1701.01974] to the quantum domain and show that they have a number of different applications. First, we obtain a sharper bound relating the optimal probability of correctly distinguishing elements of an ensemble of states to that of the pretty good measurement, and an analogous bound for optimal and pretty good entanglement recovery. Second, we obtain bounds relating optimal guessing and entanglement recovery to the fidelity of the state with a product state, which then leads to tight tripartite uncertainty and monogamy relations.
IntroductionSuccessful analysis of information processing protocols requires suitable measures of information and entropy, particularly those that satisfy the data processing inequality, the statement that a formal measure of information satisfies the intuitive requirement that a noisy channel cannot increase it. One broad class of measures is given by the Rényi divergences, which includes the usual Shannon and von Neumann definitions of mutual information and entropy. But even more, the Rényi divergences also encompass optimal and "pretty good" strategies for distinguishing quantum states or recovering entanglement, and are related to the oft-used fidelity function. Hence new insights into these measures often leads to new results for these operational tasks. This is the case in [1], for instance, which found new conditions for the optimality of the pretty good measurement by investigating the relationship of various quantum Rényi divergences.Here we extend a recent result by Sason and Verdú [2], which establishes a whole class of Fano-like inequalities involving the Rényi divergence and optimal distinguishing probability, to the quantum domain. Though the inequalities are essentially an immediate consequence of the data processing inequality, they turn out to have a number of interesting applications. First, we find improved bounds relating the pretty good measurement to the optimal measurement, as well as analogous bounds for pretty good and optimal entanglement recovery. Second, by establishing a new relation between the optimal guessing probability and the fidelity, we can provide a complete characterization of the set of admissible guessing probabilities in an uncertainty game [3,4], which is also related to wave-particle duality relations in multiport interferometers [5]. The goal of game is to provide predictions of the values of potential measurements of two conjugate observables on a quantum system; the uncertainty principle implies that the predictions cannot both be accurate. The same relation holds for fidelity and optimal entanglement recovery, and in this context gives a complete characterization of the possible entanglement fidelities two different parties can have with a common system, resolving a conjecture for the "singlet monogamy" studied in [6].