Under certain rather prevalent conditions (driven by dynamical orbital evolution), a hierarchical triple stellar system can be well approximated, from the standpoint of orbital parameter estimation, as two binary star systems combined. Even under thi
s simplifying approximation, the inference of orbital elements is a challenging technical problem because of the high dimensionality of the parameter space, and the complex relationships between those parameters and the observations (astrometry and radial velocity). In this work we propose a new methodology for the study of triple hierarchical systems using a Bayesian Markov-Chain Monte Carlo-based framework. In particular, graphical models are introduced to describe the probabilistic relationship between parameters and observations in a dynamically self-consistent way. As information sources we consider the cases of isolated astrometry, isolated radial velocity, as well as the joint case with both types of measurements. Graphical models provide a novel way of performing a factorization of the joint distribution (of parameter and observations) in terms of conditional independent components (factors), so that the estimation can be performed in a two-stage process that combines different observations sequentially. Our framework is tested against three well-studied benchmark cases of triple systems, where we determine the inner and outer orbital elements, coupled with the mutual inclination of the orbits, and the individual stellar masses, along with posterior probability (density) distributions for all these parameters. Our results are found to be consistent with previous studies. We also provide a mathematical formalism to reduce the dimensionality in the parameter space for triple hierarchical stellar systems in general.
