The glowing orange doughnut unveiled in April 2019 became one of the defining science images of the century. Yet a nagging doubt never quite went away: how much of that ring around the black hole in galaxy M87 was real, and how much came from the choices made while turning sparse radio data into a picture? Four scientists from CSIRO, the Australian government research body, have re-examined the data and concluded that the ring is real. Their analysis appears in Nature.
Light from that galaxy has been travelling toward Earth for some 55 million years. Lead author Samuel Lai compares the job to snapping a picture of a doughnut sitting on the lunar surface while standing in your own yard. The Event Horizon Telescope got around that by linking radio dishes across the planet into one Earth-sized instrument, but every dish adds its own instrumental errors, and the atmosphere muddies the signal further.
The Australian group attacked the problem from a different direction. Instead of working with pairs of telescopes, as is standard, they combined measurements in groups of three. These “closure invariants” cancel out much of the calibration noise. They then fed the result to a machine learning method of their own, called GenDIReCT, which they deliberately trained on everyday photos of cats, dogs and cars so that it had no built-in idea of what a black hole should look like.
It still produced a ring of almost the same size and shape as the 2019 image, and it did so across several years of observations. EHT Collaboration director Laurent Loinard welcomed the work as the kind of independent validation that sits at the heart of science.
Project leader Nithyanandan Thyagarajan thinks the method could travel well beyond astronomy, for example into medical imaging and earthquake mapping.
