暗いトランジット候補の惑星半径を少ない変数で推定

Occurrence rate estimates of small long-period planets rely on the radii of low signal-to-noise ratio transit candidates whose posteriors should be propagated into the population inference. These posteriors should include the error budget from all latent parameters and account for the effects of the correlated noise. Full Markov chain Monte Carlo fits with Gaussian processes take days of runtime per target, and provide information on latent parameters that will be marginalized over anyway. We present a method that computes the marginalized posterior of the planet-to-star radius ratio while sampling only a lower-dimensional effective parameter space. The 11-dimensional problem reduces to a 4-dimensional one, while the posterior retains the uncertainty budget of all the nuisance parameters, including eccentric orbits, stellar parameters, and limb darkening. The method operates on whitened light curves, mitigating biases due to correlated noise. We show that the posteriors are calibrated on simulated systems: the true radii have uniformly distributed p-values. Applied to 419 faint long-period Kepler Objects of Interest, the method yields radius ratios broadly consistent with the KOI table, although their differences exceed those expected from the reported uncertainties.