New image of the supermassive black hole M87 generated by the PRIMO algorithm from 2017 EHT data. Credit: Medeiros et al. 2023
Machine learning reconstructs a new image from the EHT data.
The image of the M87
” data-gt-translate-attributes=”[{” attribute=””>black hole has been enhanced using a The iconic image of the supermassive black hole at the center of M87—sometimes referred to as the “fuzzy, orange donut”—has gotten its first official makeover with the help of machine learning. The new image further exposes a central region that is larger and darker, surrounded by the bright accreting gas shaped like a “skinny donut.” The team used the data obtained by the Event Horizon Telescope (EHT) collaboration in 2017 and achieved, for the first time, the full resolution of the array.
In 2017, the EHT collaboration used a network of seven pre-existing telescopes around the world to gather data on M87, creating an “Earth-sized telescope.” However, since it is infeasible to cover the Earth’s entire surface with telescopes, gaps arise in the data—like missing pieces in a jigsaw puzzle.
M87 supermassive black hole originally imaged by the EHT collaboration in 2019 (left); and new image generated by the PRIMO algorithm using the same data set (right). Credit: Medeiros et al. 2023
“With our new machine learning technique, PRIMO, we were able to achieve the maximum resolution of the current array,” says lead author Lia Medeiros of the Institute for Advanced Study. “Since we cannot study black holes up close, the detail of an image plays a critical role in our ability to understand its behavior. The width of the ring in the image is now smaller by about a factor of two, which will be a powerful constraint for our theoretical models and tests of gravity.”
PRIMO, which stands for principal-component interferometric modeling, was developed by EHT members Lia Medeiros (Institute for Advanced Study), Dimitrios Psaltis (Georgia Tech), Tod Lauer (
Animation fades from M87 black hole image, first produced by the EHT collaboration in 2019, to the new image generated by the PRIMO algorithm using the same data set. Credit: Medeiros et al. 2023
PRIMO relies on dictionary learning, a branch of machine learning which enables computers to generate rules based on large sets of training material. For example, if a computer is fed a series of different banana images—with sufficient training—it may be able to determine if an unknown image is or is not a banana. Beyond this simple case, the versatility of machine learning has been demonstrated in numerous ways: from creating Renaissance-style works of art to completing the unfinished work of Beethoven. So how might machines help scientists to render a black hole image? The research team has answered this very question.
With PRIMO, computers analyzed over 30,000 high-fidelity simulated images of black holes accreting gas. The ensemble of simulations covered a wide range of models for how the black hole accretes matter, looking for common patterns in the structure of the images. The various patterns of structure were sorted by how commonly they occurred in the simulations, and were then blended to provide a highly accurate representation of the EHT observations, simultaneously providing a high fidelity estimate of the missing structure of the images. A paper pertaining to the algorithm itself was published in The Astrophysical Journal on February 3, 2023.
“We are using physics to fill in regions of missing data in a way that has never been done before by using machine learning,” added Medeiros. “This could have important implications for interferometry, which plays a role in fields from exo-planets to medicine.”
Overview of simulations generated for the PRIMO algorithm training set. Credit: Medeiros et al. 2023
The team confirmed that the newly rendered image is consistent with EHT data and with theoretical expectations, including the bright emission ring that should be produced by hot gas falling into the black hole. Generating an image required assuming an appropriate shape of the missing information, and PRIMO did this based on the 2019 discovery that the M87 black hole in fine detail looked as expected.
“About four years after the first horizon-scale image of a black hole was unveiled by EHT in 2019, we have taken another step forward by producing an image that for the first time uses the full resolution of the network,” Psaltis said. “The new machine learning techniques we have developed provide a golden opportunity for our collective work to understand the physics of black holes.”
The new image should lead to more precise determinations of the mass of the M87 black hole and the physical parameters that determine its current appearance. The data also offers researchers the ability to place greater constraints on event horizon alternatives (based on the darker central luminosity depression) and perform more robust gravity tests (based on the size narrower of the ring). PRIMO can also be applied to other EHT observations, including those of Sgr A*, the central black hole of our own
” data-gt-translate-attributes=”[{” attribute=””>Milky Way galaxy.
M87 is a massive, relatively nearby, galaxy in the Virgo cluster of galaxies. Over a century ago, a mysterious jet of hot
Reference: “The Image of the M87 Black Hole Reconstructed with PRIMO” by Lia Medeiros, Dimitrios Psaltis, Tod R. Lauer and Feryal Özel3, 13 April 2023, The Astrophysical Journal Letters.
DOI: 10.3847/2041-8213/acc32d
Development of the PRIMO algorithm was enabled through the support of the National Science Foundation Astronomy and Astrophysics Postdoctoral Fellowship.
