Description
Surveying the large-scale structure of the universe will yield an enormous amount of high-quality data for constraining cosmology and potentially detecting new physics. Extracting the maximum amount of information from this dataset and using it to its full potential requires fast and accurate cosmic structure formation simulations. N-body simulations deliver a high level of accuracy but are too slow to use directly for inference. In this talk, I will present a new field-level emulator for cosmic structure, designed to map the linear perturbations of the early universe to the nonlinear outcomes of cosmological N-body evolution. The emulator is a convolutional neural network augmented with style parameters that capture cosmology and redshift dependence. The redshift dependence makes time derivatives, or velocities, directly available through autodifferentiation during training. We train the emulator on the full phase-space distribution of the N-body particles using a suite of simulations spanning a wide range of cosmologies and redshifts. This approach achieves percent-level accuracy down to nonlinear scales of ~ 1 Mpc/h and can efficiently generate accurate mock catalogues and carry out field-level inference. Applications for cosmological parameter inference, and initial conditions inference will also be presented.
