The Weird Detector: Flagging periodic, coherent signals of arbitrary shape in time series photometry

By design, model-based approaches for flagging transiting exoplanets in light curves, such as boxed least squares, excel at detecting planets with low S/N at the expense of finding signals that are not well described by the assumed model, such as self-lensing binaries, disintegrating or evaporating planets, or planets with large rings. So far, such signals have typically been found through visual searches by professional or citizen scientists, or by inspection of the photometric power-spectra. We present a nonparametric detection algorithm, for short duty-cycle periodic signals in photometric time series based on phase dispersion minimization. We apply our code to 161,786 Kepler sources and detect 18 new periodic signals consistent with heartbeat binaries/planets, 4 new singly-transiting systems, and 2 new doubly-transiting systems. We show that our code is able to recover the majority of known Kepler objects of interest (KOIs) to high confidence, as well as more unusual events such as Boyajian's star and a comet passing through the Kepler field. Nonparametric signal-flagging techniques, such as the one presented here, will become increasingly valuable with the coming data from TESS and future transit surveys as the volume of data available to us exceeds that which can be feasibly examined manually.