Optimizing atom transfer in a double magneto-optical trap system with machine learning

A push beam is a common tool for transferring atoms from a vapor chamber magneto-optical trap (MOT) to an ultra-high vacuum MOT (UHV-MOT) in ultracold atomic experiments. Many of these experiments leave the push beam on continuously during UHV-MOT loading, but there is documented evidence that, for a narrow set of parameters, pulsing the push beam during loading results in increased atom capture at the UHV-MOT. Additionally, while the performance of continuous loading is explained by atomic two-level models, pulsing introduces time-dependent behavior which is non-trivial to account for when optimizing loading in the double-MOT system. For these reasons, we choose to improve the performance of our ³⁹K double MOT with a machine learning (ML) algorithm which efficiently scans the full parameter space by using a combination of online optimization and offline modeling of the apparatus response. Future plans include further ML-optimization of cooling steps on the road to quantum degeneracy while optimizing for appropriate figures of merit (e.g. atom number, temperature).