ZELDA: fitting Lyman-alpha line profiles using deep learning

We present zELDA(redshift Estimator for Line profiles of Distant Lyman-Alpha emitters), an open source code to fit Lyman-Alpha (Lya) line profiles. The main motivation is to provide the community with an easy to use and fast tool to analyze Lya line profiles uniformly to improve the understating of Lya emitting galaxies. zELDA is based on line profiles of the commonly used 'shell-model' pre-computed with the full Monte Carlo radiative transfer code LyaRT. Via interpolation between these spectra and the addition of noise, we assemble a suite of realistic Lya spectra which we use to train a deep neural network. We show that the neural network can predict the model parameters to high accuracy (e.g.,.0.34 dex HI column density for R=12000) and thus allows for a significant speedup over existing fitting methods. As a proof of concept, we demonstrate the potential of zELDA by fitting 97 observed Lya line profiles from the LASD data base. Comparing the fitted value with the measured systemic redshift of these sources, we find that Lya determines their rest frame Lya wavelength with a remarkable good accuracy of 0.3A (75 km/s). Comparing the predicted outflow properties and the observed Lya luminosity and equivalent width, we find several possible trends. For example, we find an anticorrelation between the Lya luminosity and the outflow neutral hydrogen column density, which might be explained by the radiative transfer process within galaxies