Hitaishi Chillara, a 17-year-old student from Leander, Texas, has received a $50,000 Davidson Fellows Scholarship for research into how supermassive black holes grow and how their effects might be detected from the light of distant galaxies. His work brings together computer simulations of the universe, telescope data and machine learning. The research is titled ‘Predicting Supermassive Black Hole-Host Mass Offsets from Broadband Photometry Across Cosmological Simulations with Forecasts for LSST.’ It examines whether ordinary measurements of galaxy brightness and colour can provide useful clues about black holes that cannot easily be weighed from Earth.
How Hitai shi Chillara used AI to study supermassive black holes
Chillara’s project focuses on supermassive black holes, enormous objects found at the centres of galaxies. Although astronomers can observe many galaxies, directly measuring the mass of the black hole at their centre is possible for only a relatively small number of them.The Davidson Institute reports that Chillara developed a computational framework to investigate whether the colours and brightnesses recorded by large sky surveys could contain information about the mass of a central black hole. Instead of relying on direct measurements alone, his approach looks for patterns hidden within the light coming from galaxies.This becomes particularly important as astronomical surveys begin producing enormous amounts of imaging data. The Davidson Institute notes that the Vera C Rubin Observatory’s Legacy Survey of Space and Time(LSST) will provide a huge collection of images that scientists can use to study galaxies across the universe.
Image Credit: Davidson Institute
Simulations revealed black-hole growth patterns
To test his approach, Chillara worked with several computer simulations designed to represent the evolution of the universe. His research uses the SIMBA, IllustrisTNG and EAGLE cosmological simulations.The simulations allowed him to examine different situations involving the growth of supermassive black holes and then connect those situations to the kind of observations that an astronomical survey such as LSST could make.Instead of treating the simulated data as a simple substitute for real observations, the study converted the simulations into the photometric bands that LSST is expected to observe. In simple terms, the research tried to make the simulated galaxies look like the kind of images a powerful sky survey would actually collect.
Importance of galaxy colours for the research
The next step involved machine learning. Chillara trained an ensemble machine-learning classifier to distinguish between different black-hole growth patterns using broadband photometry, which is information about how bright an object appears through broad ranges of light.The results were notable. According to the paper, the method reached 91 to 94% accuracy in distinguishing over-massive and under-massive black-hole growth regimes in the SIMBA and IllustrisTNG simulations under LSST magnitude limits.The research also tested whether the method could work when trained using one simulation and evaluated using another. Those cross-simulation tests produced accuracies of 83 to 89%. This was important because different simulations use different ways of representing complicated processes inside galaxies and around black holes.
Hitaishi’s AI-model looked beyond black-hole light
One of the interesting parts of Chillara’s work was identifying what the machine-learning system was actually using to make its classifications.The study found that the strongest signal came from the colours of the host galaxies, with classification accuracy of 82 to 87% when using those features. The results also pointed to the shape of the light produced by matter as a black hole feeds.This helped the research move beyond simply producing a successful prediction. As reported by the Davidson Institute, he repeatedly analysed the results to understand whether the patterns identified by the model could be interpreted by people. The aim was to develop a system that scientists could understand, rather than one that simply returned an answer without showing what was behind it.
Chillara’s research is valuable for future galaxy surveys
The research does not provide a direct way to weigh every supermassive black hole. Instead, it establishes a way to examine whether information already present in large-scale astronomical images can help classify how black holes have grown.That distinction matters because telescope time is limited, while surveys can gather images of billions of galaxies. The Davidson Institute says Chillara’s framework could help scientists make better use of public datasets and large sky surveys by identifying what broadband images can reliably reveal and which questions still require additional observations.The paper describes the results as a validated baseline for studying post-seeding black-hole growth. It also notes an important limitation. The simulations used in the study contain heavy black-hole seed prescriptions of at least 10,000 solar masses. That means the work tests growth after the initial seeding stage instead of answering the broader question of how the earliest supermassive black holes formed.
Academic journey of Texas student behind the research
Chillara is a student at the Texas Academy of Mathematics & Science. According to the Davidson Institute, his interest in machine learning began with a toy robot dog and a curiosity about how it could remain balanced while walking. That interest led him from robotics and inverse kinematics towards machine learning and, later, astrophysics.His interests now span several areas, including AI, astronomy and robotics. The Davidson Institute says he helps run his school’s space and robotics clubs, competes in cybersecurity and tutors physics and mathematics. He has also earned a silver medal in the US Physics Olympiad and a Gold Honour in the International Astronomy and Astrophysics Competition.
