Abstract
We present a quantitative analysis of the correlations between the equivalent widths of optical emission lines for a sample of more than 70,000 star-forming galaxies with high S/N SDSS spectra and with spectral synthesis performed with the Starlight software. We show, using statistical tools such as the distance correlation and maximal information correlation, that there are indeed strong correlations between the most prominent emission lines usually detected in the optical region of galaxy spectra. We have done, also, a Principal Component Analysis (PCA) of the synthesized continuum spectra and used up to 10 components to train an artificial neural network to estimate the equivalent widths of the emission lines, with excellent results, demonstrating that there is a strong correlation between the continuum and equivalent widths. The same analysis was performed with the symbolic regression software Eureqa, which provided functional relations between the four principal components and the equivalent widths, with an accuracy between 0.12 and 0.24 dex for different emission lines. The main motivation behind this work is to produce realistic spectra for tests of data reduction pipelines of the new generation of galaxy surveys, like J-PAS and PFS/SuMIRe.
