Abstract

To generate device-independent randomness requires non-locality. In typical analyses based on the CHSH Bell inequality, the amount of certifiable randomness increases with the Bell violation. However, this intuitive relationship is not all it seems. I will explain how a maximal two bits of certifiable randomness can be obtained using a pair of entangled qubits for a wide range of nonlocality, including arbitrarily small amounts, and that for correlations with too much non-locality the amount of certifiable randomness decreases. This is achieved through a tight upper bound on the randomness as a function of CHSH violation. Our theoretical result has implications in practical cases, allowing more randomness to be justifiably extracted from real (noisy) data in a range of cases. This is based on arXiv:2205.00124 (joint with Lewis Wooltorton and Peter Brown).