FlopPITy: Enabling self-consistent exoplanet atmospheric retrievals with machine learning

TL;DR

FlopPITy introduces SNPE-C with neural spline flows to accelerate exoplanet atmospheric retrievals while preserving the fidelity of posterior inferences. By employing multi-round training and a data-driven proposal distribution, it achieves faithful posteriors with substantially fewer forward-model evaluations, enabling self-consistent models that would be impractical with traditional sampling. The authors demonstrate performance on both bulk, fast-model retrievals and a self-consistent brown dwarf case, reporting speedups from ~2× to ≥10× and posterior accuracies at or near ground truth under ideal conditions. While amortised ML approaches can excel for large datasets, FlopPITy fills a niche for flexible, high-cost forward models and complex atmospheres, broadening the practical toolkit for exoplanet characterization; code is publicly available on Github.

Abstract

Interpreting the observations of exoplanet atmospheres to constrain physical and chemical properties is typically done using Bayesian retrieval techniques. Because these methods require many model computations, a compromise is made between model complexity and run time. Reaching this compromise leads to the simplification of many physical and chemical processes (e.g. parameterised temperature structure). Here we implement and test sequential neural posterior estimation (SNPE), a machine learning inference algorithm, for exoplanet atmospheric retrievals. 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. We generate 100 synthetic observations using ARCiS (ARtful Modeling Code for exoplanet Science, an atmospheric modelling code with the flexibility to compute models in varying degrees of complexity) and 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 generate a synthetic observation of a cool brown dwarf using the self-consistent capabilities of ARCiS and run a retrieval with self-consistent models to showcase the possibilities that SNPE opens. 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 $\sim2\times$ and $\geq10\times$ depending on the computational load of the forward model, the dimensionality of the observation, and the signal-to-noise ratio of the observation. We make the code publicly available for the community on Github.

Figures (5)

• Figure 1: Schematic representation of the iterative training of a neural spline flow. In the very first iteration, the proposal distribution is the prior.
• Figure 2: Coverage probability of SNPE posteriors at different training rounds compared to Multinest. The red dashed line denotes the 1:1 line. All the lines are close to the diagonal, indicating that both the Multinest and SNPE posteriors are faithful. Importantly, the latter remain faithful at every round.
• Figure 3: Posteriors for five noise realizations of a synthetic WISE 1828+2650 spectrum. The input parameters of the synthetic spectrum are denoted by the black lines and are: T$_{\text{eff}}=325$ K, R$_\text{P}=1.83$ R$_\text{J}$, $\log{g}=3.6$, $\log{K_{\text{zz}}}=7$, C/O$=0.55$ and $\log{Z}=0$. To keep the figure readable, we show only the 2$\sigma$ contours. The quantiles shown in the titles correspond to one of the noisy realisations.
• Figure 4: Example corner plots for a case with a very narrow posterior around the ground truth for retrievals run with broad (top) and tight (bottom) priors.
• Figure 5: 1$\sigma$ contours of the retrieved spectra for the case with broad (top) and tight (bottom) priors.