CAREER: A Framework for Galaxy Formation Theory in the Era of Artificial Intelligence

About This Grant

Computer simulations, based on physics, can be used to predict how galaxies grow over time. These cosmological simulations connect the growth of dark matter to stellar and gaseous processes. The predictions that these simulations provide are often sensitive to various arbitrary choices which introduce uncertainty that is rarely quantified. This program will develop a novel analysis, based on AI, that will allow for the characterization of uncertainties in cosmological simulations in a novel way. This research will be conducted at the University of Virginia and support a graduate research student who will participate in the model development. Educational initiatives include a Research Experiences for Undergraduates program and an annual schedule of teacher training activities. This project will advance the field of galaxy formation by executing a simulation suite designed for uncertainty quantification across physical models, cosmological assumptions, and resolution scales. A fully automated simulation pipeline will enable the execution of thousands of zoom-in simulations at varying resolution and halo mass, sampling parameters such as supernova feedback strength, AGN efficiency, Omega-matter, and sigma-8. Simulation outputs will be paired with dark matter-only merger trees and analyzed using Graph Neural Networks (GNNs) that learn to predict galaxy properties across the sampled parameter space. The models will be extended to predict spatially resolved galaxy properties using a hybrid of GNNs and diffusion-based generative models, allowing the sampling of entire profile distributions conditioned on both formation history and simulation assumptions. The tools will enable direct comparison across physical models and offer a statistically grounded framework for testing small-scale tensions in Lambda-CDM. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.