Self-learning neural networks crack iconic black holes

A team of astronomers led by Michael Jansen (Ladbood University, Netherlands) trained neural networks with millions of synthetic black hole data sets. Event Based on the network and data of the Horizon Telescope, they predict, among other things, the black hole at the heart of our Milky Way is rotating at top speed.

Astronomers published their results and methodologies in three papers in the journal Astronomy and Astrophysics.

In 2019, the event Horizon Telescope Collaboration released its first image of the ultra-high Massive black hole at the heart of the Galaxy M87. In 2022, they presented images of black holes in our Milky Way, Sagittarius A*. However, the data behind the image contains a wealth of easily difficult information. An international team of researchers trained neural networks to extract as much information as possible from the data.

From just a handful to millions

Events Previous work by Horizon Telescope Collaboration used only a few realistic synthetic data files. Now, astronomers have fed millions of such data files to so-called Bayesian neural networks that can quantify uncertainty. This allowed researchers to better compare EHT data with models.

Thanks to neural networks, researchers believe that, for example, a black hole in the center of the Milky Way rotates at almost maximum speed. Its axis of rotation refers to the Earth. Furthermore, emissions near black holes are primarily caused by very hot electrons in the surrounding accretion disk, rather than the so-called jet. Additionally, the magnetic field of accretion disk appears to behave differently from the usual theory of such disks.

“It's certainly exciting that we are defiant of general theory,” says lead researcher Michael Jansen (Nymegen, the University of Radboud, Netherlands). “However, our AI and machine learning approaches are primarily considered the first step. We will then improve and extend the relevant models and simulations. And when the African millimeter telescope under construction combines with data collection, we can get better information to test the relative theory of highly accurate, hyper-competing compact objects.”

Impressive scaling

“The ability to scale up millions of synthetic data files is an impressive achievement,” says co-researcher Geordy Deilal (Princeton University, USA). “You need storage capacity, supercomputers, software pipelines, and programs to distribute your work.”

Researchers emphasize that the scale of this work has been made possible by a coordinated ecosystem of computing services: Cyverse of data storage, OSG OS pools for high-throughput computing, PEGASUS for workflow management, MAX PLANCK computing and data facilities in Germany, software tools such as Tensorflow, Horovod, CASA.

Researchers didn't just make predictions about Sagittarius A*. They also saw the M87*, a black hole in the center of the M87. In particular, they discovered that this black hole also spins at high speeds, but not as fast as Sagittarius A*. In addition, it rotates in the opposite direction to the downward gas. Astronomers suggest that this opposition could be the result of a merger with another galaxy.