The ring around the black hole at the center of the galaxy M87 is real, and it was not created during data processing. This has been confirmed by an Australian team that reconstructed the image using publicly available observations from the Event Horizon Telescope. The neural-network-based algorithm developed by the team was trained on ordinary photographs and does not depend on telescope calibration. Yet the result turned out to be almost identical to the famous first image of this object.
The same ring across three years
Astronomers from Australia’s national science agency, CSIRO, processed Event Horizon Telescope (EHT) data collected over eight nights in 2017, 2018, and 2021. Almost every time, the result was a compact ring approximately 42 microarcseconds in diameter. Samuel Lai, the lead author, compares the task to trying to photograph a doughnut on the Moon from your backyard. The comparison is almost literal: as seen from Earth, an object measuring eight centimetres across at the Moon’s distance would appear at such an angle.
There are three reasons why the researchers consider this result independent. The method uses only combinations of signals in which errors from individual telescopes cancel out, making it unaffected by calibration. The algorithm is not given the size of the source in advance, and the neural network was trained on photographs unrelated to space, meaning that the ring could not have emerged from assumptions built into the method itself.
The ring’s diameter and orientation matched the results of the EHT Collaboration. The authors estimated the mass of the M87* black hole at 6.5 ± 0.5 billion solar masses, consistent with the estimate based on the motion of stars in the galaxy’s core. The study was published in the peer-reviewed journal Nature. In 2018 and 2021, the brightest part of the ring was closer to the direction of the M87 jet than it had been in 2017.
Where the doubts came from
The mass of a black hole is determined from the ring’s size, so further conclusions depend on it. The EHT telescopes work together as a single antenna the size of the Earth, but they cover only small portions of its surface, meaning that the rest of the image has to be reconstructed. The atmosphere above each station distorts the signal. As Phys.org reports, the result therefore depends on calibration and on the decisions people make during data processing.
In 2022, Japanese astronomers from Makoto Miyoshi’s group reprocessed the same observations and reached a different conclusion. When the region of the sky within which the algorithm searches for the source is expanded, the ring disappears and is replaced by a single elongated bright spot. The authors of the new study believe the reason is likely related to the observations made on April 5, 2017, when the telescopes collected the smallest number of measurements. In higher-quality recordings, the ring appears regardless of the size of the search region.
Signals from three telescopes
The method relies on combinations of signals from three telescopes forming a closed triangle. Within such a triangle, errors from individual telescopes cancel one another out, leaving information about the shape of the source. The EHT also used these combinations, but study co-author Nithyanandan Thyagarajan brought them together into a single set of closure invariants.
Another difference lies in the starting point. The collaboration’s imaging pipeline began with a blurred spot 40 microarcseconds across, which already suggested the size of the source. The GenDIReCT algorithm does not require such an initial assumption.
The algorithm is based on a diffusion model, like those used by text-to-image generators. It was trained primarily on photographs of animals, cars, and ships at a resolution of 32 by 32 pixels, using the CIFAR-10 dataset. Simple shapes, including rings, were also part of the dataset, but removing them did not change the result.
“Our model had no idea what a black hole looked like when we applied it to the M87 data,” says Samuel Lai.
What remains invisible
EHT Collaboration Director Laurent Loinard described GenDIReCT as a fully independent method. The ring’s width and its brightness distribution depend on the processing method, so these parameters probably do not reflect the properties of the source itself.
The black hole’s rotation could not be determined from these images. It might be revealed by the photon ring — a thin ring of radiation that has orbited the black hole before travelling towards Earth. Future space-based radio interferometers may be capable of resolving it.
