New Independent Analysis Backs the Iconic M87 Black Hole Image and Refines Its Ring Structure

On April 10, 2019, the Event Horizon Telescope Collaboration unveiled the first picture of a black hole. The object sits at the center of the galaxy M87, about 55 million light-years away. The image, a glowing ring around a dark center, quickly became one of the most recognized scientific pictures of the modern era.

What it showed was a black hole "shadow" outlined by light from swirling hot gas, as Einstein's general relativity predicts. The result supported the long-held idea that the centers of galaxies house supermassive black holes. But in the years since, scientists have disagreed about whether the details and structure of the ring in the original publication were accurate.

A new study published in Nature takes up that question. Its authors applied new techniques to the original data sets. They report that their results validate the original work and also offer a fresh view of the ring's structure.

The difficulty of the task is hard to overstate. Imaging M87's black hole is comparable to photographing a doughnut on the moon while standing on Earth, and that kind of sharpness would require a telescope the size of the planet. The EHT approximates one by combining signals from radio telescopes around the globe, giving the sharpest imaging possible from the ground.

Four independent imaging teams within the collaboration contributed to the original publication. All of them produced a ring-like structure, though they differed in how they rendered fine-scale features, differences the authors attribute to human and algorithmic bias in the imaging process. One study went further and fundamentally challenged whether the source was a ring at all. Some discrepancies are expected when a sparse array of telescopes is stitched into one Earth-sized virtual instrument. Each telescope also adds its own errors, from its instruments or from Earth's atmosphere. Astronomers normally estimate and correct these, but residual errors can still lead to wrong outcomes.

To get around this, the new team developed a way of combining data called closure invariants. Standard practice combines pairs of telescope measurements; closure invariants combine triplets, three measurements at once. According to the authors, this removes atmospheric and instrumental errors and reveals cleaner information about the object's structure.

They then built a machine learning imaging method called GenDIReCT, which produces images consistent with the measured data. To limit bias, they trained it on terrestrial pictures such as cats, dogs and cars, so it had no notion of what a black hole looks like. It went through a large suite of validation tests, including imaging a wide range of non-ring structures and real data from other galaxies such as Centaurus A. Once satisfied, the team froze the training and testing procedure so human bias could not shape the outcome.

On Dec. 4, 2025, the validated method was allowed to look at the EHT's M87 observations. Across all data sets the ring was consistent. Its appearance and diameter matched a black hole of 6.5 billion solar masses viewed nearly face-on, like looking straight down into the eye of a hurricane.

Compared directly with the EHT Collaboration's images, the new result agrees in diameter and overall orientation. Fine details differ, namely ring width and surface brightness around the circumference, but the overall structure matches. To show the outcome did not rest on machine learning alone, the team also built a second method using traditional statistical techniques, which confirmed the structure from the same closure invariant data. The authors say the approach could also serve other fields that merge multi-sensor data to map unknown structures, such as medical imaging and earthquake mapping.