Machine-learning Inference of the Interior Structure of Low-mass Exoplanets

Abstract

We explore the application of machine learning based on mixture density neural networks (MDNs) to the interior characterization of low-mass exoplanets up to 25 Earth masses constrained by mass, radius, and fluid Love number k 2 . We create a dataset of 900 000 synthetic planets, consisting of an iron-rich core, a silicate mantle, a high-pressure ice shell, and a gaseous H/He envelope, to train a MDN using planetary mass and radius as inputs to the network. For this layered structure, we show that the MDN is able to infer the distribution of possible thicknesses of each planetary layer from mass and radius of the planet. This approach obviates the time-consuming task of calculating such distributions with a dedicated set of forward models for each individual planet. While gas-rich planets may be characterized by compositional gradients rather than distinct layers, the method presented here can be easily extended to any interior structure model. The fluid Love number k 2 bears constraints on the mass distribution in the planets' interior and will be measured for an increasing number of exoplanets in the future. Adding k 2 as an input to the MDN significantly decreases the degeneracy of the possible interior structures. In an open repository we provide the trained MDN to be used through a Python Notebook.

NGTS-14Ab: a Neptune-sized transiting planet in the desert

“…The models of Howe et al (2014) suggest that an H 2 -He envelope constituting approximately half of the planet's radius would contribute five to ten per cent of the planet's mass, for a core mass of 10 M ⊕ . In other words, there is broad compatibility between the models of Baumeister et al (2020) (Fig. 6) and the model of Zeng et al (2019) with a five per cent (by mass) hydrogen-rich envelope (Fig.…”

“…We use the neural network model of planetary interiors developed by Baumeister et al (2020) to predict the internal structure of NGTS-14Ab (Fig. 6), using the planetary mass and radius (Table 5) as model inputs.…”

Toward Constraining Mars' Thermal Evolution Using Machine Learning

“…Baumeister et al. (2020) used MDNs to predict the distribution of the possible interior structures of extrasolar planets for given mass and radius.…”

“…They argued that under the assumption of smooth probability distributions between training samples, MDNs need significantly fewer simulations than MCMC methods for constraining mantle convection parameters. Baumeister et al (2020) used MDNs to predict the distribution of the possible interior structures of extrasolar planets for given mass and radius.…”

Transmission Spectroscopy for the Warm Sub-Neptune HD 3167c: Evidence for Molecular Absorption and a Possible High-metallicity Atmosphere

“…Any H/He atmospheres these close-in super-Earths may have accreted from the protoplanetary nebula during formation must have been lost, likely by thermal escape (e.g., Lopez & Fortney 2013Zahnle & Catling 2017). For sub-Neptunes with radii above the valley, there is more ambiguity, as their masses and radii can be explained by various proportions of iron, rock, water, and H/He (e.g., Valencia et al 2007Valencia et al , 2013Adams et al 2008;Rogers & Seager 2010a, 2010bHowe et al 2014;Dorn et al 2017;Baumeister et al 2020). One possibility is that most of these sub-Neptunes possess rock/iron cores and are surrounded by thick H/He envelopes contributing ∼1%-10% of the total planet mass (Lopez & Fortney 2013Jin & Mordasini 2018).…”