Deep Potential

Recovering the gravitational potential from a snapshot of phase space

One of the major goals of the field of Milky Way dynamics is to recover the gravitational potential field. Mapping the potential would allow us to determine the spatial distribution of matter - both baryonic and dark - throughout the Galaxy. This is usually done by using a set of tracers, typically a population of stars, and fitting a model that best explains the kinematics of the tracers. Because observations lack information about the accelerations of the stars, additional assumptions, such as stationarity, must be invoked in order to infer the gravitational potential from the stellar kinematics. Traditionally the models have been semi-analytic and restricted to certain classes of potentials.

I investigate a deep learning based approach termed “Deep Potential”, which aims to fit as general a potential as possible by representing the tracers with a normalizing flow, and the potential using a feed-forward neural network. The flexibility of the model is then limited by the size of the neural network. I am primarily interested in the application of this method to the Milky Way using Gaia Data Release 3 with radial velocities, and seeing how the fundamental assumptions of the method are challenged.