Cosmological simulations of galaxy formation are reaching a high level of accuracy and can finely reproduce some of the main properties of galaxy populations: stellar masses, angular momentum, colors, etc. I will however show that most galaxy formation simulations fail to account for the detailed structure of galaxies both at high redshift and in the nearby Universe. Using high-resolution, idealized simulations of galactic dynamics and star formation, I will show that these disagreements are not cosmetic details but point toward a fundamental tension between observations and galaxy formation models. The first galaxy formation models predicted galaxies with unrealistically large stellar masses: in modern cosmological simulations, this issue is generally solved though energetic feedback from young stars and supermassive black holes. I will nevertheless show that feedback, as implemented in these simulations, is probably excessive, leading to the early and unrealistic exhaustion of interstellar gas reservoirs. Comparisons to idealized simulations and observations of galactic winds support the conclusion that energetic stellar and black hole feedback cannot be entirely responsible for the regulation of star formation and galaxy growth. Other physical processes likely need to be included in future galaxy formation models. Nevertheless, some cosmological simulations can now successfully describe the re-distribution of baryons from galaxies to the intergalactic medium, and I will show that these simulations are a crucial tool not just for galaxy formation but also for modern cosmological surveys — any cosmological probe obviously still relies on observations of galactic and/or extragalactic baryons, and is thus affected by baryonic physics. I will finally review how simulations of galaxy formation, evolution and star formation could gain strength in the years to come, in relation with the arrival of exascale supercomputers.
Local contact: Carlos GOMEZ-GUIJARRO