Learning the Universe
Simons Collaboration — inferring the universe's initial conditions with simulation-based inference
The Simons Collaboration on Learning the Universe aims to determine the initial conditions and evolution of our universe using Bayesian forward modeling: initial conditions are repeatedly sampled, observational consequences are predicted with cosmological simulations, and comparison with real observations of galaxies and gas constrains the posterior distribution of the initial conditions and the physics of galaxy formation.
Making that loop practical requires advances on several fronts at once. Field-level emulators accelerate the forward model enough to explore gigaparsec volumes (Scoggins et al., 2025). New subgrid models, calibrated on resolved simulations of star formation and galactic outflows across diverse galactic environments, make the galaxy-formation side of the forward model trustworthy (Jeffreson et al., 2024; Burger et al., 2026). And constrained simulations of real structures — like the Coma cluster — test the whole pipeline against objects we can actually observe (Steinwandel et al., 2026). The same machinery extends to the high-redshift universe, where it constrains early black hole growth (Kho et al., 2026).
The collaboration brings together experts in cosmological simulation, galaxy formation, machine learning, and statistical inference. See the publications list and people on the collaboration website.