Description
Representation learning allows astronomers to uncover relationships between galaxies by using the learned feature space of deep learning models. This space learns the physical appearance of galaxies, with galaxies of a similar appearance occupying a similar region in the feature space. As the appearance of galaxies is strongly affected by their morphology, we can use the feature space to find galaxies with a similar appearance, and therefore, morphology. This is important as galaxies with a similar morphology and mass likely share similar evolutionary histories. Using these principles, I will demonstrate how we can use the feature space to trace the evolutionary history of galaxies at low redshift, out to a redshift of z~1.
Using data from the IllustrisTNG-50 simulation, we select galaxies at low redshifts and find pseudo-progenitors at higher redshifts. These pseudo-progenitors are galaxies with similar properties (e.g. stellar mass, size, star formation rate) to the real progenitors of the low-redshift galaxies (i.e. when the low-redshift galaxies were at an earlier evolutionary stage, at higher redshift). Because of these shared properties, these pseudo-progenitors are likely analogues for the low redshift galaxies at earlier stages of their evolutionary history. Thus, the pseudo-progenitors could evolve into galaxies with similar properties and morphologies to the selected galaxies at low-redshift. This means that we are able to trace galaxy morphology and evolution with high accuracy, the next steps for this process are to test this method with real observational data and uncover the properties of progenitor populations for galaxies of particular morphological types.
