Deep Learning Prediction of the Broad Lyα Emission Line of Quasars

Abstract

We have employed a deep neural network, or deep learning , to predict the flux and the shape of the broad Ly α emission lines in the spectra of quasars. We use 17,870 high signal-to-noise ratio (S/N > 15) quasar spectra from the Sloan Digital Sky Survey Data Release 14 to train the model and evaluate its performance. The Si iv , C iv , and C iii] broad emission lines are used as the input to the neural network, and the model returns the predicted Ly α emission line as the output. We found that our neural-network model predicts quasars’ continua around the Ly α spectral region with ∼6%–12% precision and ≲1% bias. Our model can be used to estimate the H i column density of eclipsing and ghostly damped Ly α (DLA) absorbers, as the presence of the DLA absorption in these systems strongly contaminates the flux and the shape of the quasar continuum around the Ly α spectral region. The model could also be used to study the state of the intergalactic medium during the epoch of reionization.