Planet Formation and Machine Learning
The discovery of multiple giant exoplanetary systems hosting terrestrial planets within their habitable zones motivates a systematic study of the formation of such systems using N-body simulations. Our approach begins by determining the parameters of stable orbits for systems of several giant planets with different masses, followed by simulations of terrestrial planet formation in the inner regions of the system, including hydrodynamical modeling where appropriate. The orbital stability of a planetary system depends strongly on the initial orbital parameters of its giant planets. Machine learning (ML) can substantially reduce the computational cost of assessing dynamical stability compared with direct N-body simulations. In particular, a pre-trained ML model can predict the stability of a system with an accuracy exceeding 93% in sub-seconds, providing a fast alternative for screening large numbers of planetary configurations.
