Professor Richard McGehee

CSENG Mathematics, School of
College of Science & Engineering
Twin Cities
Project Title: 
Using Machine Learning to Investigate Binary Formation in Star clusters

Binary stars are prevalent in globular clusters.  A possible explanation is that close encounters of three stars can lead to one star being ejected from the cluster, leaving a binary behind. One way to study this problem is to determine the distribution of velocities of the ejected star after a near triple collision. This approach is very computationally intensive due to the large number of initial conditions needed to determine the tail of the distribution.

Recently, another group published results showing how machine learning could predict the results of the simulations given the initial conditions. Their focus was the three-body problem, where they avoided any type of collisions or singularities. These researchers want to take this focus of using machine learning and turn it towards the regularized three-body problem, where double collisions are continued rather than avoided. 

To have a substantial sample to train on and have meaningful results requires around 100 thousand simulations. If machine learning is able to accurately predict the results of the regularized three-body problem, then the idea of running trillions of simulations to better study the problem is feasible and could be done within a couple of days instead of decades. This would allow the determination of a distribution for the velocities of the ejected body in the regularized three-body problem.

Project Investigators

William Frazier
Professor Richard McGehee
 
Are you a member of this group? Log in to see more information.