Identifying Tidal Disruption Events via Prior Photometric Selection of Their Preferred Hosts

A nuclear transient detected in a post-starburst galaxy or other quiescent galaxy with strong Balmer absorption is likely to be a Tidal Disruption Event (TDE). Identifying such galaxies within the planned survey footprint of the Large Synoptic Survey Telescope (LSST)---before a transient is detected---will make TDE classification immediate and follow-up more efficient. Unfortunately, spectra for identifying most such galaxies are unavailable, and simple photometric selection is ineffective; cutting on "green valley" UV/optical/IR colors produces samples that are highly contaminated and incomplete. Here we propose a new strategy using only photometric optical/UV/IR data from large surveys. Applying a machine learning Random Forest classifier to a sample of ~400k SDSS galaxies with GALEX and WISE photometry, including 13,592 quiescent Balmer-strong galaxies, we achieve 53-61% purity and 8-21% completeness, given the range in redshift. For the subset of 1299 post-starburst galaxies, we achieve 63-73% purity and 5-12% completeness. Given these results, the range of likely TDE and supernova rates, and that 36-75% of TDEs occur in quiescent Balmer-strong hosts, we estimate that 13-99% of transients observed in photometrically-selected host galaxies will be TDEs and that we will discover 119-248 TDEs per year with LSST. Using our technique, we present a new catalog of 67,484 candidate galaxies expected to have a high TDE rate, drawn from the SDSS, Pan-STARRS, DES, and WISE photometric surveys. This sample is 3.5x larger than the current SDSS sample of similar galaxies, thereby providing a new path forward for transient science and galaxy evolution studies.