Deep Neural Operators for Detailed Binary Evolution Simulation

Overview

Binary stellar evolution simulations are computationally expensive and form a fundamental component of population synthesis pipelines. This project explores deep neural operators as a way to approximate binary evolution dynamics from initial system parameters such as primary mass, secondary mass, and orbital period.

The goal is to accelerate large-scale astrophysical simulations while preserving the morphology and physical structure of evolutionary tracks. The work was developed during the NVIDIA and NCSA Open Hackathon in collaboration with the POSYDON project and the NSF-Simons AI Institute for the Sky.

The project demonstrated the potential of neural operator models for scalable binary evolution simulation and was selected for presentation at the Open Accelerated Computing Summit 2025.

Ugur Demir, Philipp M Srivastava, Aggelos K. Katsaggelos, Vicky Kalogera, Santiago L Tapia, Manuel Ballester, Shamal Lalvani, Patrick Koller, Jeff J Andrews, Seth Gossage, Max M Briel, and Elizabeth Teng