Self-learning neural network cracks iconic black holes

An international team of astronomers that includes Associate Astronomer Chi-kwan Chan of Steward Observatory (University of Arizona) has trained a neural network with millions of synthetic black hole data sets. The network helps them conclude, among other things, that the black hole at the center of our Milky Way is spinning at nearly top speed. The astronomers publish their results and methodology in three papers in the journal Astronomy & Astrophysics.

In 2019, the Event Horizon Telescope Collaboration released the first image of a supermassive black hole at the center of the galaxy M87. In 2022, they presented the first-ever image of the black hole at the center of our own Milky Way galaxy, Sagittarius A*.  Researchers at the University of Arizona played a leading role in the effort, providing two of the eight telescopes used to make the observations and performing data analysis that resulted in the globally-publicized images. However, the data behind both images still contained a wealth of hard-to-crack information. Now, an international team of researchers has trained a neural network to extract additional information from the data. Their work highlights the potential for AI to dramatically advance scientific discovery in astronomy.  

From a handful to millions

Previous studies by the Event Horizon Telescope Collaboration used only a handful of realistic synthetic data files. This time, the astronomers fed millions of such data files into a so-called Bayesian neural network, which can quantify uncertainties. This allowed the researchers to make a much better comparison between the EHT data and the models.

Thanks to the neural network, the researchers now suspect that the black hole at the center of the Milky Way is spinning at almost top speed. Its rotation axis points to the Earth. In addition, the emission near the black hole is mainly caused by extremely hot electrons in the surrounding accretion disk and not by a so-called jet. Also, the magnetic fields in the accretion disk appear to behave differently from the usual theories of such disks.

The EHT is always expanding, including new stations such as the Africa Millimetre Telescope (AMT), instrumentation updates such as 345 GHz observation capability, and even space very-long-baseline interferometry (VLBI) such as the Black Hole Explorer (BHEX). "That we are defying the prevailing theory is of course exciting," says lead researcher Michael Janssen (Radboud University Nijmegen, the Netherlands). "However, I see our AI and machine learning approach primarily as a first step. Next, we will improve and extend the associated models and simulations. And when the AMT, which is under construction, joins in with data collection, we will get even better information and may also be able to validate the general theory of relativity for supermassive compact objects with a high precision."

Impressive scaling

"The ability to scale up to the millions of synthetic data files required to train the model is an impressive achievement," adds Chi-kwan Chan. "It requires dependable workflow automation, and effective workload distribution across storage resources and processing capacity."

This scale of computational work was made possible by an ecosystem of computational services: Pegasus for workflow management, HTCondor for workload distribution, CyVerse for data storage, the OSG OSPool for throughput computing capacity, Germany's Max Planck Computing and Data Facility for neural network training, and software tools including TensorFlow, Horovod, and CASA.

The researchers did not only make predictions about Sagittarius A*. They also looked at M87*, the black hole at the center of M87. Among other things, they found that this black hole is also spinning quickly, but not as quickly as Sagittarius A*. Additionally, it is spinning in the opposite direction of the infalling gas. The astronomers suggest that this counter-rotating motion may be the result of a merger with another galaxy.

While these astronomical discoveries help advance our understanding of black holes, the AI methods at the heart of the research are what excite Chan the most.  “When we published a series of 6 papers describing our findings of Sgr A* in 2022, it took a collaboration of more than 300 people and years to process and interpret the data,” Chan said. “Now, with the AI methods described here, we are able to reproduce many findings with a much smaller team and with higher accuracy.” Researchers at the University of Arizona and Steward Observatory are in a prime position to continue leading the development of AI tools to accelerate scientific discovery in astronomy.  

Scientific papers

Deep learning inference with the Event Horizon Telescope I. Calibration improvements and a comprehensive synthetic data library. By: M. Janssen et al. In: Astronomy & Astrophysics, 6 June 2025. [original (open access) | preprint (pdf) ]

Deep learning inference with the Event Horizon Telescope II. The Zingularity framework for Bayesian artificial neural networks. By: M. Janssen  et al. In: Astronomy & Astrophysics, 6 June 2025. [original (open access) | preprint (pdf)]

Deep learning inference with the Event Horizon Telescope III. Zingularity results from the 2017 observations and predictions for future array expansions. By: M. Janssen et al. In: Astronomy & Astrophysics, 6 June 2025. [original (open access) | preprint (pdf)]

More information

Contact

Michael Janssen
Department of Astrophysics, Institute for Mathematics, Astrophysics and Particle Physics, Radboud University, Nijmegen, the Netherlands
M.Janssen@astro.ru.nl
https://www.ru.nl/en/people/janssen-m-michael

Chi-kwan Chan
Steward Observatory and Department of Astronomy, University of Arizona, Tucson, Arizona, USA
chanc@arizona.edu
https://astro.arizona.edu/person/chi-kwan-ck-chan

Jordy Davelaar
Department of Astrophysical Sciences, Princeton University, Princeton, New Jersey, USA
jdavelaar@princeton.edu
https://web.astro.princeton.edu/people/jordy-davelaar
https://jordydavelaar.com

Maciek Wielgus
Instituto de Astrofísica de Andalucía-CSIC, Granada, Spain
mwielgus@iaa.es
https://maciekwielgus.wixsite.com/maciek

About NOVA
The Netherlands Research School for Astronomy (NOVA, www.astronomie.nl) is the alliance of the astronomical institutes of the universities of Amsterdam, Groningen, Leiden, and Nijmegen. The mission of Top Research School NOVA is to carry out frontline astronomical research in the Netherlands, to train young astronomers at the highest international level, and to share its new discoveries with society. The NOVA laboratories are specialized in building state-of-the-art optical/infrared and submillimeter instrumentation for the largest telescopes on earth.