Our Galaxy center can contain an exceptional cosmic top, a black hole that seems to be turning almost as quickly as possible.
Michael Janssen at the University of Radboud in the Netherlands and his colleagues were studying the black hole in the center of the Milky Way, Sagittarius A*, using the data collected by a network of telescope observers (Event) collectively known. To deal with the complexity of the data, they resorted to artificial intelligence.
First, they used known mathematical models to simulate one million black holes, which was a computational feat that required millions of hours of supercomputer time. Then they used these simulations to train a type of the so -called neuronal network, which allows to determine the features of a black hole based on observation data. Finally, they fed the data on Sagittarius A* that EHT had compiled through 2017.
The AI indicated that Sagittarius A* is turning between 80 and 90 percent of its highest possible speed. He also alerted researchers that none of their magnetic field models conform to our black hole particularly well, so more mathematical work is necessary. Janssen says that the previous studies had narrowed by the range of properties that Sagitarius A* could have, as the rapid one is turning and what kind of magnetic fields surround him, but this new approach set them more accurately.
Dimitrios Psaltis at the Georgia Institute of Technology in Atlanta says that he found some of these contradictive findings. The previous analyzes could not even offer clarity on whether the black hole rotation could be determined precisely from EHT data, he says.
Some previous works indicated that Sagittarius A* could be turning very quickly, says Yosuke Mizuno at Tenguing Shanghai Jiao in China. But he points out that the computer models used in the new study have improvement margin. “Our theoretical model is not yet perfect,” he says.
But both Mizuno and Psaltis say that AI is becoming an integral part of We How you learn about exotic cosmic objects such as black holes. “We are in a situation in which we have many data and we have many models, and we need modern ways to combine the two,” says Psaltis. “This is where automatic learning makes great differentiation.”
At the same time, this brings its own challenges, since AI’s work must be two verified and the underwater analysis of possible hallucinations.
Janssen and his team have already made many verifications of this type, such as testing AI with specifically designed simulation data. More tests will be carried out as data analysis of other EHT operation and, ultimately, data from new observatories, he says.