Alexandra Doytcheva
Pronouns: she/her
Research Mentor(s): Johannes Lange
Research Mentor School/College/Department: Physics / LSA
Program:
Authors: Alexandra Doytcheva, Johannes Ulf Lange, Filomela Gerou
Session: Session 2: 10:00 am – 10:50 am
Poster: 81
Abstract
Astronomers use simulations of the Universe such as the IllustrisTNG simulation suite to analyze the evolution of galaxies over cosmic time, as observations can only take a snapshot of the characteristics of a galaxy at a certain point in time. Cosmological simulations of structure formation and galaxy formation are an invaluable tool for cosmology. Full-physics simulations that treat gas physics and star formation are the most detailed, but their computational cost is too high to run them for large cosmological volumes. This results in the predicted galaxy clustering statistics lacking precision, limiting their applicability in interpreting large-scale structure. This study takes a smaller gravity simulation and teaches AI galaxy formation to combine AI and the gravity simulation to try and reduce scatter in the expensive full-physics simulation. Using the control variates method, information about the errors within previously made estimates is used to reduce the overall error of the simulation. These two methods combined substantially increase the predictive power of existing full-physics simulations. In this work, we show that this new method can save large amounts of computational resources by increasing the precision of galaxy clustering predictions as if the simulation was run over a much larger volume.
