Data-Driven Methods for Maximum-Precision Radial Velocities
The extraction of precise radial velocities from 1D stellar spectra is a critical component of the discovery and characterization of new exoplanets. We present a new open-source RV pipeline based on a latent-variable model. Our method is fully data-driven but it incorporates the causal structure of time-variable stellar and telluric components with different barycentric velocities. It performs comparably to the HARPS pipeline on archival HARPS data. We demonstrate the capabilities of our method with data on 51 Pegasi b, for which we obtain slightly revised planet parameters. We also show a data-driven model of the detailed, time-variable telluric absorption features, which is very sensitive because it combines information from many observations. The methods presented here will be especially valuable for the coming generation of infrared RV spectrographs.