A deep learning approach to study Jupiter's interior from Juno

Abstract

The interior structure of Jupiter holds information on its formation and evolution processes, with the two research fields highly related to one another. The range of plausible interior structures is constrained by the gravity field measured by the Juno mission, the atmospheric abundances measured by Galileo, and the 1 bar temperature estimated from radio occultation. Consequently, it is also affected by the surface winds and their internal structure, which significantly contribute to the gravity field. Inferring the range of plausible interior structures requires an intensive computational search of combinations of various planetary properties, such as the cloud-level temperature, compositions, core features, etc., matching the observations. This search requires computing ~10^8 interior models.  Here, we propose an efficient deep learning method to generate unique interior models using the very accurate but computationally demanding concentric MacLaurin spheroid method. We train a neural network to predict interior model results accurately. This allows us to perform a broad parameter space search by computing only ~10^4 interior models, resulting in a large sample of plausible interior structures. The network can also be used to infer the non-linear relations between the physical features and the observable gravity field and mass.