Bayesian optimization of non-classical optomechanical correlations

Nonclassical correlations provide a resource for many applications in quantum technology as well as providing strong evidence that a system is indeed operating in the quantum regime. Optomechanical systems can be arranged to generate nonclassical correlations (such as quantum entanglement) between the mechanical mode and a mode of travelling light. Here we propose automated optimization of the production of quantum correlations in such a system, beyond what can be achieved through analytical methods, by applying Bayesian optimization to the control parameters. A two-mode optomechanical squeezing experiment is simulated using a detailed theoretical model of the system and the measurable outputs fed to the Bayesian optimization process. This then modifies the controllable parameters in order to maximize the non-classical two-mode squeezing and its detection, independently of the inner workings of the model. We focus on a levitated nano-sphere system, but the techniques described are broadly applicable in optomechanical experiments, and also more widely, especially where no detailed theoretical treatment is available. We find that in the experimentally relevant thermal regimes, the ability to vary and optimize a broad array of control parameters provides access to large values of two-mode squeezing that would otherwise be difficult or intractable to discover via analytical or trial and error methods. In particular we observe that modulation of the driving frequency around the resonant sideband allows for stronger nonclassical correlations. We also observe that our optimization approach finds parameters that allow significant squeezing in the high temperature regime. This extends the range of experimental setups in which non-classical correlations could be generated beyond the region of high quantum cooperativity.