Inferring Exoplanet Disequilibria with Multivariate Information in Atmospheric Reaction Networks

Inferring properties of exoplanets from their atmospheres presents technical challenges in data collection due to low resolution and low signal-to-noise ratio (S/N) and theoretical challenges in the predictions made from forward-modeling due to errors introduced via incomplete or inaccurate assumptions in atmospheric physics and chemistry. The combination of these factors makes developing techniques to identify the most predictive features robust to low S/N and model error an increasingly important challenge for exoplanet science. Here we implement a multivariate approach to identify optimal predictors of the state of disequilibria. As a case study we focus on the prediction of vertical mixing (parameterized as eddy diffusion) in hot Jupiter atmospheres. We use multivariate information contained in molecular abundances, reaction network topology, and Gibbs free energy to demonstrate the variation in prediction efficacy of the vertical mixing coefficient (K zz) from different model information. While current approaches target inferring molecular abundances from spectral data, our results indicate that the set of optimal predictors of K zz varies with planetary properties such as irradiation temperature and metallicity. In most cases, multivariate data composed of network topological variables, which capture system-level features, perform as well as the set of optimal predictors and better than any individual variable. We discuss future directions, where identifying the set of optimal predictors should be useful for quantitatively ranking atmospheres in terms of their distance from thermochemical equilibrium, provide target variables for the development of new tools for inverse modeling, and provide applications to the longer-term goal of detection of disequilibria associated with life.