A novel gravitational inversion method for small celestial bodies based on geodesyNets

Abstract

A comprehensive understanding of the internal structure of small celestial bodies is essential for elucidating their formation and evolutionary processes. However, the current observational techniques are severely limited in their ability to effectively probe the structure of small celestial bodies, thereby rendering the method of gravitational inversion the primary means of inferring internal structures. This study, grounded on the model architecture of geodesyNets, introduced a novel method that utilizes gravitational field data to infer the internal density distribution of small celestial bodies. Similar to geodesyNets, by leveraging principles of implicit neural representations (INRs), a deep neural networks (DNN) architecture based on self-supervised learning was developed. This proposed method involved meticulous optimization of hyperparameters, gravitational calculation strategy, activation functions, and loss functions tailored to the specific requirements of gravitational inversion for small celestial bodies. An inversion analysis was conducted on the CERES70E gravitational field model of the dwarf planet Ceres employing this method as a case study, revealing a two-layer internal structure. Moreover, gravitational inversion studies were conducted on simulated small bodies to evaluate the performance of various loss functions, activation functions, hyperparameters, and shape models on the precision of the inferred density fields. The results of these cases provided corroboration for the reliability of the inferred two-layer structural model for Ceres.