Systematic Discovery and Classification of TESS Eclipsing Binaries

Although a large fraction of stars are known to be found in binaries, the formation and dynamical evolution of these systems are not well understood. Additionally, masses determined from binary dynamics are essential benchmarks for constraining stellar evolutionary models. The Transiting Exoplanet Satellite Survey (TESS) with over 85% sky spatial coverage, high photometric precision, and timing baselines ≥27 days, presents a unique opportunity to infer the population statistics of binaries discovered through light curve eclipses. However, identifying new eclipsing binaries out of the survey using supervised machine learning classification techniques depends critically on the information content of the chosen light curve features, and the quality and size of training sample data. Here we report preliminary results on our project to systematically classify all of the TESS eclipsing binaries according to their light curve morphology (Algol, β Lyrae, W UMa). In particular, we explore feature extraction and similarity metrics for periodic variable classification (e.g. using dynamic time warping), report robust periods using standard period-finding algorithms (e.g. Lomb-Scargle, box least squares), and consider the case of limited training data using semi-supervised learning.