New analysis reproduces the ring in the 2019 M87 black hole image, CSIRO team reports

A team of four researchers at CSIRO, Australia's national science agency, says it has reproduced the ring around the black hole at the center of the galaxy M87 with an imaging method it developed independently of the Event Horizon Telescope Collaboration, which made the original picture. In a paper published in Nature on November 11, 2026, the team reports a ring 42.2 microarcseconds across, with an uncertainty of 1.5. Assuming what the authors call nominal scaling parameters, that size points to a black hole of 6.5 billion solar masses, plus or minus 0.5 billion, seen at roughly 20 degrees from the direction of its approaching jet.

The collaboration, which combines radio dishes around the world into what amounts to an Earth-sized telescope, released the first image of that black hole, known as M87*, on April 10, 2019. Its papers described an asymmetric bright ring 42 ± 3 microarcseconds in diameter, observed in April 2017 at a wavelength of 1.3 millimeters, and derived a mass of 6.5 ± 0.7 billion Suns by comparing the images with simulations of black holes.

Why the ring was questioned

Turning measurements from a sparse set of widely spaced dishes into a picture is a poorly constrained problem, the CSIRO authors write, and the answer can shift with instrument calibration, processing software and the settings an analyst picks. They note that analyses of M87* had produced differing fine details, and that some outside studies questioned whether the source was ring-shaped at all.

A direct challenge came in 2022 from Makoto Miyoshi, Yoshiaki Kato and Junichiro Makino, who rebuilt the image from the collaboration's public data and published a different result in The Astrophysical Journal. They reported a jet reaching out from the core and concluded that both the ring and the absence of a jet in the collaboration's image were artifacts of a narrow field of view and of gaps in how the array sampled the source. Another 2022 reanalysis, by Chris Carilli and Nithyanandan Thyagarajan, now a co-author of the Nature paper, used traditional hybrid mapping. Its images converged on a source about 44 microarcseconds across. Starting from annulus and disk models, which gave the cleanest images, the source appeared as a ring or an edge-brightened disk, consistent with earlier results; from other starting models it had a similar size and a central dip in brightness but was less clearly a ring.

Triplets instead of pairs

The new study rests on closure invariants, quantities that the authors describe as immune to instrument calibration errors. "Our technique combines triplets rather than the standard pairs of telescope data to eliminate those errors," Thyagarajan, the project leader, said in a CSIRO release.

To turn those quantities into an image, the team used GenDIReCT, a machine-learning method that relies on conditional diffusion models. Its 2025 methods paper, in the Publications of the Astronomical Society of Australia, describes a model trained on simple shapes and the CIFAR-10 image dataset. A second paper, in February 2026, applied it to Event Horizon Telescope observations of the galaxy Centaurus A and the quasar 3C 279 and reported measurements consistent with the collaboration's. "Importantly, our model had no idea what a black hole looked like when we applied it to the M87 data," said Samuel Lai, a CSIRO astronomer and the paper's lead author.

What the collaboration says

The collaboration's director, Laurent Loinard, said in the CSIRO release that the method "recovers the ring structure of M87* across several years, in excellent agreement with the EHT Collaboration's published results." He also said: "Independent validation is at the heart of the scientific process."

Set side by side, the two diameters and the two masses agree within their stated uncertainties, and the new diameter carries the smaller error; that comparison is ours, from the published figures. The paper's abstract speaks to the ring and the mass, and the CSIRO release to the ring. Neither says whether the new reconstruction shows the jet that Miyoshi's team reported; the abstract mentions the jet only to give the viewing angle. Whether the full paper takes up that part of the 2022 argument cannot be judged from these two sources.

What comes next

Thyagarajan said the technique could also help in medical imaging and in mapping earthquakes. The publisher's metadata lists a Creative Commons Attribution license for the Nature paper, and the earlier methods papers set out how the imaging model was trained and tested.