A Bayesian algorithm for model selection applied to caustic-crossing binary-lens microlensing events

Abstract

We present a full Bayesian algorithm designed to perform automated searches of the parameter space of caustic-crossing binary-lens microlensing events. This builds on previous work implementing priors derived from Galactic models and geometrical considerations. The geometrical structure of the priors divides the parameter space into well-defined boxes that we explore with multiple Monte Carlo Markov Chains. We outline our Bayesian framework and test our automated search scheme using two data sets: a synthetic light curve, and the observations of OGLE-2007-BLG-472 that we analysed in previous work. For the synthetic data, we recover the input parameters. For OGLE-2007-BLG-472 we find that while χ2 is minimized for a planetary mass-ratio model with extremely long time-scale, the introduction of priors and minimization of the Bayesian information criterion, rather than χ2, favour a more plausible lens model, a binary star with components of 0.78 and 0.11 M⊙ at a distance of 6.3 kpc, compared to our previous result of 1.50 and 0.12 M⊙ at a distance of 1 kpc.