The BACCO simulation project: a baryonification emulator with neural networks

This is the authors' abstract. We don't add key points for this paper.

We present a neural network emulator for baryonic effects in the non-linear matter power spectrum. We calibrate this emulator using more than 50 000 measurements in a 15D parameter space, varying cosmology and baryonic physics. Baryonic physics is described through a baryonification algorithm, which has been shown to accurately capture the relevant effects on the power spectrum and bispectrum in state-of-the-art hydrodynamical simulations. Cosmological parameters are sampled using a cosmology-rescaling approach including massive neutrinos and dynamical dark energy. The specific quantity we emulate is the ratio between matter power spectrum with baryons and gravity only, and we estimate the overall precision of the emulator to be 2\!-\!3\ per\ cent, at scales k < 5 \, h\, Mpc⁻¹ and redshifts 0 < z < 1.5. We obtain an accuracy of 1\!-\!2\ per\ cent, when testing the emulator against a collection of 74 different cosmological hydrodynamical simulations and their respective gravity-only counterparts. We also show that only one baryonic parameter, namely Mc, which sets the gas fraction retained per halo mass, is enough to have accurate predictions of most of the baryonic feedbacks at a given epoch. Our emulator is publicly available at http://www.dipc.org/bacco.