Transit clairvoyance: enhancingTESSfollow-up using artificial neural networks

Abstract

The upcoming TESS mission is expected to find thousands of transiting planets around bright stars, yet for three-quarters of the fields observed the temporal coverage will limit discoveries to planets with orbital periods below 13.7 days. From the Kepler catalog, the mean probability of these short-period transiting planets having additional longer period transiters (which would be missed by TESS) is 18%, a value ten times higher than the average star. In this work, we show how this probability is not uniform but functionally dependent upon the properties of the observed short-period transiters, ranging from less than 1% up to over 50%. Using artificial neural networks (ANNs) trained on the Kepler catalog and making careful feature selection to account for the differing sensitivity of TESS, we are able to predict the most likely short-period transiters to be accompanied by additional transiters. Through cross-validation, we predict that a targeted, optimized TESS transit and/or radial velocity follow-up program using our trained ANN would have a discovery yield improved by a factor of two. Our work enables a near-optimal follow-up strategy for surveys following TESS targets for additional planets, improving the science yield derived from TESS and particularly beneficial in the search for habitable-zone transiting worlds.