Optimizing exoplanet atmosphere retrieval using unsupervised machine-learning classification

Abstract

One of the principal bottlenecks to atmosphere characterisation in the era of all-sky surveys is the availability of fast, autonomous and robust atmospheric retrieval methods. We present a new approach using unsupervised machine learning to generate informed priors for retrieval of exoplanetary atmosphere parameters from transmission spectra. We use principal component analysis (PCA) to efficiently compress the information content of a library of transmission spectra forward models generated using the PLATON package. We then apply a k-means clustering algorithm in PCA space to segregate the library into discrete classes. We show that our classifier is almost always able to instantaneously place a previously unseen spectrum into the correct class, for low-to-moderate spectral resolutions, R, in the range R = 30 − 300 and noise levels up to 10 per cent of the peak-to-trough spectrum amplitude. The distribution of physical parameters for all members of the class therefore provides an informed prior for standard retrieval methods such as nested sampling. We benchmark our informed-prior approach against a standard uniform-prior nested sampler, finding that our approach is up to a factor two faster, with negligible reduction in accuracy. Our general approach is not specific to transmission spectroscopy and should be more widely applicable to cases that involve repetitive fitting of trusted high-dimensional models to large data catalogues, including beyond exoplanetary science.

Alkaline Exospheres of Exoplanet Systems: Evaporative Transmission Spectra

“…Considerable endogenic modeling of evaporative transmission spectra is indeed underway, largely aiming at signatures of atomic hydrogen (Bourrier & Lecavelier des Etangs 2013;Christie et al 2016;Allan & Vidotto 2019;Murray-Clay & Dijkstra 2019;Wyttenbach et al 2020), but also by atomic helium (Oklopčić & Hirata 2018;Lampón et al 2020), atomic magnesium Bourrier et al 2015 andionized magnesium (Dwivedi et al 2019). However, the retrieval of atmospheric parameters from observations using various techniques such as χ 2 -minimization, Bayesian analysis, or advanced machine-learning methods (Márquez-Neila et al 2018;Hayes et al 2020) is largely based on the assumption of hydrostatic equilibrium (η Ehrenreich et al 2006;NEMESIS Irwin et al 2008;CHIMERA Line et al 2013; TAU-REX 1 Waldmann et al 2015; BART Blecic et al 2017;Exo-Transmit Kempton et al 2017; ATMO Goyal et al 2018;π η Pino et al 2018;Aura Pinhas et al 2018; HELIOS- T Fisher & Heng 2018; PLATON Zhang et al 2019; MERC Seidel et al 2020; among others). The problem is further amplified and fundamentally different if the gas were to be detached from the planet.…”

Unsupervised Machine Learning for Exploratory Data Analysis of Exoplanet Transmission Spectra

“…We see that the first principal component alone already accounts for as much as 95.8% of the variance in the data. Incorporating just one more PCA component brings up the total explained variance to over 99%, and the information gain beyond the first three PCA components is very minimal, as was also discussed in Hayes et al (2020). for the first three PCA components (𝑘 = 1, 2, 3) are listed in Table 1.…”

Terminus: A Versatile Simulator for Space-based Telescopes

“…Doing so will allow the tool to be more widely used and facilitate in-depth studies of Twinkle's capabilities. These could include modeling various atmospheric scenarios for each planet to judge its suitability for characterization (e.g., Fortenbach & Dressing 2020), performing retrievals on populations of exoplanets (e.g., Changeat et al 2020), classifying groups of planets via color-magnitude diagrams (e.g., Dransfield & Triaud 2020), testing machinelearning techniques for atmospheric retrieval (e.g., Márquez-Neila et al 2018;Hayes et al 2020;, or the exploration of potential biases in current data analysis techniques (e.g., Feng et al 2016;Rocchetto et al 2016;Caldas et al 2019;Changeat et al 2019;Powell et al 2019;MacDonald et al 2020;Taylor et al 2020). Additionally, thorough analyses of Twinkle's capabilities for specific scientific endeavors, such as confirming/refuting the presence of thermal inversions and identifying optical absorbers in ultrahot Jupiters (e.g., Fortney et al 2008;Spiegel et al 2009;Haynes et al 2015;Evans et al 2018;Parmentier et al 2018;Edwards et al 2020;Pluriel et al 2020;von Essen et al 2020;Changeat & Edwards 2021), searching for an exoplanet mass-metallicity trend (e.g., Wakeford et al 2017;Welbanks et al 2019), probing the atmospheres of planets in/close to the radius valley to discern their true nature (e.g., Owen & Wu 2017;Fulton & Petigura 2018;Zeng et al 2019), refining basic planetary and orbital characteristics (e.g., Berardo et al 2019;Dalba & Tamburo 2019;Livingston et al 2019), measuring planet masses through accurate transit timings (e.g., Hadden & Lithwick 2017;Grimm et al 2018;…”