A neural network emulator for the Lyman-α forest 1D flux power spectrum

Abstract

ABSTRACT The Lyman-α forest offers a unique avenue for studying the distribution of matter in the high redshift universe and extracting precise constraints on the nature of dark matter, neutrino masses, and other ΛCDM extensions. However, interpreting this observable requires accurate modelling of the thermal and ionization state of the intergalactic medium, and therefore resorting to computationally expensive hydrodynamical simulations. In this work, we build a neural network that serves as a surrogate model for rapid predictions of the one-dimensional Lyman-α flux power spectrum (P1D), thereby making Bayesian inference feasible for this observable. Our emulation technique is based on modelling P1D as a function of the slope and amplitude of the linear matter power spectrum rather than as a function of cosmological parameters. We show that our emulator achieves sub-percent precision across the full range of scales (k∥ = 0.1 – $4\, \mathrm{Mpc}^{-1}$) and redshifts (z = 2 – 4.5) considered, and also for three ΛCDM extensions not included in the training set: massive neutrinos, running of the spectral index, and curvature. Furthermore, we show that it performs at the 1 per cent level for ionization and thermal histories not present in the training set and performs at the percent level when emulating down to $k_{\parallel }=8\, \mathrm{Mpc}^{-1}$. These results affirm the efficacy of our emulation strategy in providing accurate predictions even for cosmologies and reionization histories that were not explicitly incorporated during the training phase, and we expect it to play a critical role in the cosmological analysis of the DESI survey.