Simulation-based inference of the 2D ex-situ stellar mass fraction distribution of galaxies using variational autoencoders

• DOI: 10.48550/arxiv.2410.24069
• Publication Date: 2024-10-31
• Authors: Eirini Angeloudi; Marc Huertas-Company; J. Falcón‐Barroso; Regina Sarmiento; Daniel Walo-Martín; Annalisa Pillepich; Jesús Vega-Ferrero
• Keywords: Fraction (chemistry); Mass fraction

Galaxies grow through star formation (in-situ) and accretion (ex-situ) of other galaxies. Reconstructing the relative contribution these two growth channels is crucial for constraining processes galaxy in a cosmological context. In this on-going work, we utilize conditional variational autoencoder along with normalizing flow - trained on state-of-the-art simulation an attempt to infer posterior distribution 2D ex-situ stellar mass galaxies solely from observable two-dimensional maps their mass, kinematics, age metallicity. Such are typically obtained large Integral Field Unit Surveys such as MaNGA. We find that average provides estimate resolved histories mean ~10% error per pixel. show use conditionally sample latent space results smaller reconstruction error. Due probabilistic nature our architecture, uncertainty predictions can also be quantified. To knowledge, first fraction maps.