Late in the day I attended a short talk by Mordasini (MPIA) at the currently ongoing heidelberg Exoplanet meeting about exoplanet population synthesis. Mordasini is trying to go from initial conditions inspired by protoplanetary disk observations through to late-time planets in composition, size, mass, orbital semi-major axis, and so on. It is early days, but his talk suggested that a jointly theory and data-driven model could be very effective at explaining and predicting data.
In the rest of the day, the PHAT team plus Foreman-Mackey and Bovy argued about completeness. We finally resolved our issues by realizing that (from a graphical model perspective) the data we have are conditioned on being included in the sample and this is how completeness is properly included probabilistically. This is equivalent to a distribution-function or density-modeling perspective, in which the observed data are drawn from an error-convolved, completeness censored distribution. It is all obvious in retrospect, but there are lots of wrong ways to include the completeness function!
No comments:
Post a Comment