Multi-epoch machine learning 2: Identifying physical drivers of galaxy properties in simulations

Using a novel machine learning method, we investigate the buildup of galaxy properties in different simulations, and various environments within single simulation. The aim this work is to show power approach at identifying physical drivers simulations. We compare how stellar mass dependent on value other halo points time by examining feature importance values model. By training model IllustrisTNG that stars are produced earlier times higher density regions universe than they low regions. also apply technique Illustris, EAGLE, CAMELS find built up similar way EAGLE IllustrisTNG, but significantly differently original suggesting subgrid physics more important choice hydrodynamics method. These differences driven efficiency supernova feedback. Applying principal component analysis simulations allows us identify associated with halo's gravitational potential another representing which galaxies form. discover speed galactic winds critical parameter total energy per unit star formation. Finally Simba black hole feedback has larger effect formation