From Fuzzy to Clear: PRIMO for Messier 87

New ways of utilizing machine learning have allowed for previously cloudy images of celestial bodies to become clearer, enhancing scientists’ understanding of interactions of matter within galaxies.

Reading Time: 4 minutes

Stargazing can be a challenging activity. You squint your eyes, look up into the darkness and try to make out the stars dotting the night sky that appear as fuzzy white blurs. The naked eye cannot form clear images of objects that are at such a great distance from Earth. Even with the help of modern telescopes, it remains difficult to capture the entirety of celestial bodies due to their astonishing size and the physical limitations preventing humans from getting closer to them. To combat this, a group of astrophysicists developed methods based on artificial intelligence to enhance the study of our universe’s galaxies. One galaxy studied using such methods is the Messier 87 (M87).

The M87 is a giant elliptical galaxy situated 55 million light years from Earth in the constellation Virgo. Virgo contains dozens of known exoplanets—planets that orbit stars outside the solar system—and other Messier objects. Messier objects are astronomical objects first observed by 18th-century French astronomer Charles Messier. Some of them are bright enough to be observed with a small telescope, while others are better seen using a combination of larger, high-tech telescopes that utilize artificial intelligence. M87 is the most powerful known source of radio energy and is also an extremely powerful X-ray source; it can release charged particles of radiation that travel through magnetic fields at almost the speed of light. M87 is composed of several trillion stars, approximately 15,000 globular star clusters (tightly packed collections of stars), and a supermassive black hole at its center. Black holes are typically formed when a star dies and a gravitational field so strong that not even light can escape forms. Because no light escapes, black holes appear invisible to humans without the use of space telescopes. Notably, the supermassive black hole of M87 is the first black hole to ever be directly imaged. The black hole has been under observation since the 1950s, when a compact radio source was found at its center. 

The original image of M87’s black hole—released in 2019––was a major breakthrough for the field of astronomy—the photograph was direct proof that black holes exist. With today’s technology, images of M87 are  successfully produced using the Event Horizon Telescope Collaboration (EHT), a global network of observatories that capture black hole emissions. The EHT images the region most affected by strong gravitational lensing, which is the distortion of light resulting from the changing light emission as galaxies pass massive objects, displaying the M87’s iconic ring shape based on Einstein’s general relativity predictions. The theory of gravity explains how gravitational strength causes galaxies to pull on other objects, distorting space and time, and ultimately forcing light to travel on a different path than it’s supposed to. The M87 image displayed a dark central region enclosed by a fuzzy orange ring. Images were taken over the course of six days, and successive images demonstrated slight changes in the structure of the black hole, suggesting that the structure in which the light emitted is subject to continual development. The changes in the way the light was bending proved that the black hole physically changed based on the massive objects that passed it. 

In 2022, a team of EHT researchers led by Dr. Lia Medeiros of the Institute for Advanced Study began developing Principal-component Interferometric Modeling (PRIMO) technology. PRIMO technology is far more advanced than the EHT Collaboration used in the original image, as it utilizes artificial intelligence to compensate for missing spatial information affecting the observed objects. Interferometry refers to a measurement method based on the interference of waves, and in the case of space, by combining the light from two or more telescopes. This missing spatial information includes how deep and wide the central depression of the black hole is. PRIMO relies on dictionary learning, meaning its computers assess large sets of training materials in order to recognize patterns and generate rules in response. Through this algorithm, the research team fed PRIMO’s computers over 30,000 high-fidelity simulated black hole images. The computers used principal component analysis, which allows artificial intelligence to take a large set of data and reduce it into a much smaller set containing all the necessary information. This method allowed the team to use computers to identify and sort common patterns within the structures of each image based on their rarity of occurrence, focusing specifically on how the black holes accrete, or “eat” matter. Using newly acquired information on the structure of black holes, the computer estimated how the missing information would fit onto the M87, creating a clearer image of the blurry black hole. 

Published in The Astrophysical Journal in April, the most recent image of M87 retains similar qualities to the original, as it is consistent with previous EHT data and theoretical explanations. However, it shows a more defined bright ring of emission, which is expected to be produced by hot gas falling into the black hole. The orange ring is roughly two times slimmer than its original projected size, and there is a darker central brightness depression making up the inner shadow of a black hole. The new image has allowed astronomers to determine increasingly accurate measurements of the M87 black hole’s mass and perform more tests of its gravity. As Dr. Medeiros stated at the National Science Foundation News, “The fact that we see a thin ring is a very important and interesting constraint in terms of us understanding how the matter is falling into the black hole and what the environment is around the black hole.” Because the ring appears thinner, it suggests that the black hole has been accreting only a small amount of matter. 

PRIMO has successfully created an image that could otherwise only be produced by a radio telescope as large as the Earth. The research team has developed innovative, precise ways for scientists to study objects in our universe. Machine learning tools such as PRIMO can be applied to other EHT observations, including those of Sagittarius A*, a black hole contained within our galaxy. Looking beyond astronomy, Dr. Medeiros believes that this modern technology can be utilized in various fields that apply interferometry, such as oceanography, medicine, and remote sensing. PRIMO could one day be used to enhance and accelerate scientific discoveries in a diverse range of fields, making it a crucial tool for future research and discovery of new sides of the universe.