AI Simulations Reveal Milky Way’s Black Hole Spins at Near-Maximum Speed, Defying Old Theories

AI Analysis Finds Milky Way's Black Hole Spins Near Maximum Speed

Date: June 15, 2025
Source: Morgridge Institute for Research

Artificial intelligence has enabled astronomers to extract new insights about the supermassive black hole at the center of the Milky Way. By processing over 12 million simulations using high-throughput computing, researchers discovered that this black hole is spinning at nearly its top possible speed.

This finding challenges existing models of black hole behavior. The research indicates that the emission observed around the black hole originates primarily from hot electrons in the accretion disk rather than from jets, contradicting long-held assumptions about such emissions.

How AI and High-Throughput Computing Made This Possible

An international team trained a neural network with millions of synthetic simulations to analyze the Event Horizon Telescope (EHT) data. These simulations were generated using the Center for High Throughput Computing (CHTC), a collaboration between the Morgridge Institute for Research and the University of Wisconsin-Madison.

High-throughput computing automates the distribution of computing tasks across thousands of computers, allowing massive datasets to be processed efficiently. This approach has been instrumental in scientific fields ranging from the search for cosmic neutrinos to tracking antibiotic resistance.

From Limited Data to Millions of Simulations

Previous EHT studies relied on limited synthetic data. Thanks to funding from the National Science Foundation's PATh project, the CHTC enabled astronomers to scale up to millions of synthetic data files. These were fed into a Bayesian neural network capable of quantifying uncertainties, improving the comparison between observational data and models.

The results indicate the black hole’s rotation axis points roughly toward Earth, and the magnetic fields in the accretion disk behave differently than predicted by conventional models.

"Defying the prevailing theory is exciting," said Michael Janssen of Radboud University Nijmegen. "Our AI approach is a first step; next, we will refine the models and simulations."

Scaling AI for Scientific Discovery

Chi-kwan Chan from the University of Arizona highlighted the challenge of managing workflow automation and workload distribution, noting the impressive scale achieved in processing millions of synthetic data files.

Professor Anthony Gitter of the Morgridge Institute emphasized how throughput computing capabilities enabled the assembly of high-quality AI-ready data, facilitating these discoveries.

The Open Science Pool, operated by PATh and supported by over 80 U.S. institutions, contributed the computing resources for more than 12 million jobs over three years. Miron Livny, director of CHTC and lead of PATh, noted that workloads involving millions of simulations fit perfectly with their throughput computing infrastructure.

Published Papers

• Deep learning inference with the Event Horizon Telescope I. Calibration improvements and a comprehensive synthetic data library, M. Janssen et al., Astronomy & Astrophysics, June 6, 2025.
• Deep learning inference with the Event Horizon Telescope II. The Zingularity framework for Bayesian artificial neural networks, M. Janssen et al., Astronomy & Astrophysics, June 6, 2025.
• Deep learning inference with the Event Horizon Telescope III. Zingularity results from the 2017 observations and predictions for future array expansions, M. Janssen et al., Astronomy & Astrophysics, June 6, 2025.

This research demonstrates the value of combining AI with large-scale computing to extract detailed information from complex astrophysical data. For professionals interested in AI applications in research and data analysis, exploring training resources like those at Complete AI Training can be beneficial.