Context. Interpreting the observations of exoplanet atmospheres to constrain physical and chemical properties is typically done using Bayesian retrieval techniques. Since these methods require many model computations, a compromise must be made between the model’s complexity and its run time. Achieving this compromise leads to a simplification of many physical and chemical processes (e.g. parameterised temperature structure).
Aims: Here, we implement and test sequential neural posterior estimation (SNPE), a machine learning inference algorithm for atmospheric retrievals for exoplanets. The goal is to speed up retrievals so they can be run with more computationally expensive atmospheric models, such as those computing the temperature structure using radiative transfer.
Methods: We generated 100 synthetic observations using ARtful Modeling Code for exoplanet Science (ARCiS), which is an atmospheric modelling code with the flexibility to compute models across varying degrees of complexity and to perform retrievals on them to test the faithfulness of the SNPE posteriors. The faithfulness quantifies whether the posteriors contain the ground truth as often as we expect. We also generated a synthetic observation of a cool brown dwarf using the self-consistent capabilities of ARCiS and ran a retrieval with self-consistent models to showcase the possibilities opened up by SNPE.
Results: We find that SNPE provides faithful posteriors and is therefore a reliable tool for exoplanet atmospheric retrievals. We are able to run a self-consistent retrieval of a synthetic brown dwarf spectrum using only 50 000 forward model evaluations. We find that SNPE can speed up retrievals between ∼2× and ≥10× depending on the computational load of the forward model, the dimensionality of the observation, and its signal-to-noise ratio (S/N). We have made the code publicly available for the community on Github.
