8–11 Sept 2026
University of Southampton
Europe/London timezone

Physics-Informed Machine Learning for Black Hole Initial Data

8 Sept 2026, 16:30
30m
Building 56 (University of Southampton)

Building 56

University of Southampton

Highfield, Southampton, SO17 1BJ, UK
Day-0 meeting on AI for numerical relativity Talks

Speakers

Irene Pitsiladi (University of Nottingham) Martin Zinzen (University of Nottingham)

Description

Constructing accurate initial data for binary black hole systems is a critical step in numerical relativity simulations, traditionally addressed through spectral and finite-difference methods. In this talk, we will present Pinndorama, a physics-informed neural network framework for solving the puncture initial data problem, validated against the established numerical code, NRPyElliptic. Pinndorama achieves relative errors on the order of 1e-4 with trained networks delivering full binary black hole initial data in 3D. We will also introduce the application of the Deep Ritz Method in this setting. Our results show that neural networks are effective in finding solutions to this problem while offering a new computational paradigm for constructing initial data in numerical relativity.

Authors

Irene Pitsiladi (University of Nottingham) Martin Zinzen (University of Nottingham)

Presentation materials

There are no materials yet.