Neural networks for N-body simulations

Few-body interactions (involving 3-10 bodies) are central to understanding any self-gravitating system from the Solar System to globular star clusters and nuclei of galaxies. However, solving such self-gravitating Newtonian few-body problems by numerical integration: firstly, even an approximate solution of the system’s evolution is prohibitively expensive in terms of computer time. Secondly, due to the short Lyapunov timescale of few-body systems, their chaotic orbits can only be integrated accurately with special software developed by us.

The idea of this project is to use artificial neural networks to substitute or aid the integration of the equations of motion. We are interested in artificial neural networks that have some information about the physics of the problem.

For this subproject, the Ph.D. student Veronica Saz Ulibarrena has been appointed under the supervision of lead researcher Simon Portegies Zwart. Both are located at Leiden University.