Raytracing Black Holes with AI Accelerators
Abstract
We show how software infrastructure that was mainly built to power AI applications also can be used to great benefit for other purposes -- by demonstrating a fully self-contained JAX-based black hole raytracer in under 180+200 lines of Python code plus documentation. Major motivations for showcasing such a construction are (a) to show to professional physicists how Machine Learning (ML) accelerators can greatly simplify building numerical applications for GPU accelerator hardware with very modest coding effort and programming expertise -- even if the problems at hand do not involve ML per se, (b) to make a meaningful part of General Relativity (GR) accessible to computer science professionals who might not have realized that mathematical background they developed on Deep Learning problems may have brought this theory within easy reach for them, and (c) to inspire students and educators to explore the potential of an approach that represents physical law, i.e. relations between local differential quantities, not in terms of differentiable symbolic expressions, but in terms of differentiable algorithms.
Related
- Local scale symmetry and black holes
- Origin of the Golden Mass of Galaxies and Black Holes
- Regular hairy black holes by gravitational decoupling: de Sitter and Minkowski cores
- Spectroscopy of analogue black holes using simulation-based inference
- Parameter constraints on rotating ABG regular black holes from quasi-periodic oscillations
BibTeX
@misc{fischbacher2026raytracing,
title = {{Raytracing Black Holes with AI Accelerators}},
author = {Thomas Fischbacher and Nicholas Kessler},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.26963},
eprint = {2609.26963},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.26963},
}
RIS
TY - UNPB
TI - Raytracing Black Holes with AI Accelerators
AU - Thomas Fischbacher
AU - Nicholas Kessler
PY - 2026
DA - 2026/09/22
PB - arXiv
DO - 10.48550/arxiv.2609.26963
UR - https://arxiv.org/abs/2609.26963
AB - We show how software infrastructure that was mainly built to power AI applications also can be used to great benefit for other purposes -- by demonstrating a fully self-contained JAX-based black hole raytracer in under 180+200 lines of Python code plus documentation. Major motivations for showcasing such a construction are (a) to show to professional physicists how Machine Learning (ML) accelerators can greatly simplify building numerical applications for GPU accelerator hardware with very modest coding effort and programming expertise -- even if the problems at hand do not involve ML per se, (b) to make a meaningful part of General Relativity (GR) accessible to computer science professionals who might not have realized that mathematical background they developed on Deep Learning problems may have brought this theory within easy reach for them, and (c) to inspire students and educators to explore the potential of an approach that represents physical law, i.e. relations between local differential quantities, not in terms of differentiable symbolic expressions, but in terms of differentiable algorithms.
ER -