Planet formation constraints from exoplanet population synthesis and simulation based inference

This work introduces a statistical framework to obtain Bayesian constraints on planet formation parameters, which offers a probabilistic interpretation of uncertainties and degeneracies contained within planet formation models. The model likelihood, or the probability distribution of observations conditioned on a set of formation parameters, is intractable for planet formation theories due to the high degrees of complexity involved. Furthermore, traditional likelihood estimation techniques scale poorly high to dimensional parameter spaces, and begin to require an excessive number of samples. Instead, this work utilizes neural density estimation to directly learn the posterior distribution. This is a data driven simulation-based inference approach based on simulations from an exoplanet population synthesis model. This work focuses on understanding the degeneracies found within the parameters of disk mass, disk radius, MHD-wind parameters, mid-plane temperature, and planet birth location. The information within these degeneracies are captured in posterior distributions conditioned on observables of planet mass, orbital period, and atmospheric C/O ratio. As a realized demonstration, inference is performed on the hot Jupiter HD 209458b. The inferred posterior distribution is re-sampled as parameter inputs for the population synthesis model to re-simulate formation tracks of HD 209458b. These tracks reveal two distinct scenarios where the planet formation begins either side of the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" id="m1"> <mml:mrow> <mml:msub> <mml:mrow> <mml:mtext>CO</mml:mtext> </mml:mrow> <mml:mrow> <mml:mn>2</mml:mn> </mml:mrow> </mml:msub> </mml:mrow> </mml:math> ice line. This highlights the ability to both infer formation parameters, and study the interactions between physical processes involved in planet formation.