Deeper Dive
With the advent of the Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST), the field of astronomy will have access to an unprecedented volume of imaging data. My project studies whether the light measured from a galaxy can reveal information about the supermassive black hole at its center. Black holes can prove to be uncooperative data sources, though; they do not signal their masses when we survey them. So, I worked on a computational framework that combines simulations of our universe with telescope data and machine learning to tell us what qualities of black holes we can discern. By comparing several simulated universes, I can determine which signals are reliable and which questions require more than photometry, allowing us to better understand how galaxies and their black holes evolve together.
In astronomy, machine learning may not be an ideal tool because it can function as a black box, so it is important to understand whether it truly learned what you wanted it to or found some shortcut to get the same result. I built a pipeline that trained on multiple cosmological simulations with various assumptions about black hole growth, looking for which assumptions remained consistent and could potentially translate to our real universe. Dr. Ryan Farber, my research mentor at Purdue University Fort Wayne, helped me place these results in an astrophysical context. I remember him repeatedly pushing past a convenient result to ask the question: “What explains this?” Through repeated analysis to determine whether the results could be manually understood, we were able to refine the model’s results into not only something that performed, but something we could understand.
This work is valuable because time to use telescopes is limited. A survey can collect billions of images of galaxies, but measurements, especially for black hole analysis, are much harder to obtain. By establishing what simple, broadband photometry, like a camera, can reliably tell us through learned signals, this framework can help scientists make more informed decisions when estimating black hole quantities, such as mass. I hope that the key methodological steps I took can make returns from public datasets and sky surveys more accessible, interpretable and useful.