Using Machine Learning to Model Stellar Collisions in our Galactic Center

Direct collisions in the inner pc of the Galactic center can alter the orbits and properties of stars. The outcome of a collision depends on a number of parameters, including the masses and ages of the stars, the impact parameter, and the initial relative velocity. We utilize the newly developed $\verb|collAIder|$, a machine learning tool developed to predict the outcomes of stellar collisions, to bridge 3D hydrodynamic simulations of stellar collisions with a dynamical model of the Galactic center. Our results show a substantial population of stripped stars, $\sim$$10\%$ of the initial stellar population, produced by mass loss during high-speed collisions. Stellar mergers are less frequent, with about $5\%$ of the stars experiencing a collision-induced merger. We also find that high-speed, nearly head-on collisions can completely disrupt the stars. These destructive collisions are usually preceded by 10 or more collisions, which gradually reduce the mass of the star before it is ultimately destroyed. We estimate that mass loss during stellar collisions injects roughly $(2.5-8)\times 10^5$ M$_\odot$ of gas into the surrounding environment. Most of this gas is injected into the inner $0.1$ pc and about $90\%$ is retained within the cluster and may be accreted onto the supermassive black hole.