Automation and Efficiency of Photodynamical Model Fitting for Multi-Transiting Systems

Multiple gravitationally-interacting transiting planets are the most information-rich exoplanetary systems, but extracting that information is a statistical and computational challenge. Our PhotoDynamical Multi-planet Model, PhoDyMM, characterizes the dynamical interactions between planets by directly fitting the transits determined by an n-body model to the Kepler lightcurve, taking around 1 CPU-second. PhoDyMM uses a Differential Evolution Markov Chain Monte Carlo (DEMCMC) method to perform Bayesian parameter inference on a high-performance computing cluster, typically requiring hundreds of cores for about a week. Our group has a goal to fit all ~700 Kepler systems with multiple transiting planets, requiring extensive development into automating PhoDyMM and improving its computational efficiency. Using new research into autocorrelation estimation and convergence statistics for ensemble methods like DEMCMC, we have developed semi-automated stopping criteria that optimize runtimes for individual systems. Following Tuchow et al. 2019, we have also investigated how much a reparametrization of fit parameters improves the efficiency of the DEMCMC sampler.