Mapping Lyman-alpha forest three-dimensional large scale structure in real and redshift space

TL;DR

This paper introduces Hydro-BAM, a supervised machine learning method that accurately reproduces the 3D Lyman-alpha forest in real and redshift space, significantly reducing computational costs for large-scale structure analysis.

Contribution

Hydro-BAM is a novel, physically-motivated machine learning approach that efficiently models the Lyman-alpha forest, outperforming traditional approximations in accuracy and computational speed.

Findings

Accurately reproduces 3D Lyman-alpha forest up to k~1 h/Mpc

Deviations less than 2-5% in power spectra and bi-spectra compared to reference simulations

Outperforms Fluctuating Gunn-Peterson approximation in accuracy and scale coverage

Abstract

This work presents a new physically-motivated supervised machine learning method, Hydro-BAM, to reproduce the three-dimensional Lyman-$\alpha$ forest field in real and in redshift space learning from a reference hydrodynamic simulation, thereby saving about 7 orders of magnitude in computing time. We show that our method is accurate up to $k\sim1\,h\,\rm{Mpc}^{-1}$ in the one- (PDF), two- (power-spectra) and three-point (bi-spectra) statistics of the reconstructed fields. When compared to the reference simulation including redshift space distortions, our method achieves deviations of $\lesssim2\%$ up to $k=0.6\,h\,\rm{Mpc}^{-1}$ in the monopole, $\lesssim5\%$ up to $k=0.9\,h\,\rm{Mpc}^{-1}$ in the quadrupole. The bi-spectrum is well reproduced for triangle configurations with sides up to $k=0.8\,h\,\rm{Mpc}^{-1}$. In contrast, the commonly-adopted Fluctuating Gunn-Peterson approximation shows significant deviations already neglecting peculiar motions at configurations with sides of $k=0.2-0.4\,h\,\rm{Mpc}^{-1}$ in the bi-spectrum, being also significantly less accurate in the power-spectrum (within 5$\%$ up to $k=0.7\,h\,\rm{Mpc}^{-1}$). We conclude that an accurate analysis of the Lyman-$\alpha$ forest requires considering the complex baryonic thermodynamical large-scale structure relations. Our hierarchical domain specific machine learning method can efficiently exploit this and is ready to generate accurate Lyman-$\alpha$ forest mock catalogues covering large volumes required by surveys such as DESI and WEAVE.