Bayesian Planet Detection and Orbit Determination

• Activity Number: 387
• Type: Invited
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Eric B. Ford*+ and Benjamin E. Nelson and Matthew J. Payne
• Companies: University of Florida and University of Florida and University of Florida
• Keywords: Astrostatistics ; Population MCMC ; GPU ; Bayesian Parameter Estimation ; Extrasolar Planet ; Bayesian Model Selection

Abstract

Astronomers detect planets orbiting distant stars based on observing the reflex motion of their host star. In the most common method for detecting planets, the basic observation data is a heteroscedastic, irregularly-sampled time series (the velocity of the host star towards or away from Earth). There is a simple physical model (i.e., star orbited by N planets), but the models are high-dimensional (~7*N) and highly non-linear (for N>1). Astronomers characterize planet masses and orbits via Markov chain Monte Carlo (MCMC) for Bayesian parameter estimation and model evaluation. For systems where the mutual planetary interactions are significant, the computation required for each model evaluation is significant (i.e., integrating a set of ~6*N ODEs describing the paths of each body). Additional physical constraints (e.g., long-term dynamical stability) can be imposed, but are orders of magnitude more computationally demanding. I will describe recent progress in the the development of population MCMC techniques to accelerate convergence and enable a high degree of parallelization (e.g., using Graphics Processing Units). Finally, I discuss future directions for astrostatistics.