Discovery of Short Period Planets in Kepler Data Using Innovative Fast GPU Phase Folding and Deep CNN Algorithms

Since the first detection of hot Jupiter orbiting a solar-type star, 51 Peg, in 1995, over 4000 exoplanets have been discovered using various observational techniques. However, only hundreds of detected planets have sub-Earth radii. It is not clear how these sub-Earths form, and more samples would be critical in investigating this population. By applying a novel GPU Phase Folding algorithm in conjunction with a Deep Convolutional Neural Network (DCNN) algorithm, called the GPFC method, in Kepler photometry data, we have improved transit search speed by three orders of magnitude over the traditional Box Least Squares method, allowing a complete search of Kepler known KOI photometry data in a few hours using a commercial GPU machine. To date, we have detected three strong short-period candidates—Kepler-242d, Kepler-270d and Kepler-937d. Our follow-up analysis shows that Kepler-270d is a 0.76 R⊕ sub-Earth orbiting at period of 0.912 day around a hosting star with 1.16 M_⊙ mass; Kepler-242d is a 0.65 R⊕ sub-earth on a 1.56 day orbiting around a K dwarf with mass of 0.79 M_⊙; and Kepler-937d is a 1.6 R⊕ super-earth on a 0.921 day period orbiting around a G-type star with mass of 0.96 M_⊙. Kepler-242d and Kepler-270d are among the smallest short planets known to date. The discovery of these small exoplanets highlights the promising capability of the GPFC method for searching for small new transiting exoplanets in photometry data from Kepler, TESS and future space transit missions.