ML supported Exoplanet Cloud Formation Modelling

The discovery of a stunning diversity of planets beyond our solar system (exoplanets) is one of modern astronomy`s most profound achievements, from the first confirmed detection in the 1990s to currently more than 5000 known exoplanets. Exploring exoplanets has revolutionized our understanding of the universe and reshaped our perceptions of planetary systems: No solar system sibling has been discovered so far, instead, hot Jupiters bear evidence for a dynamic formation of planetary systems including our own. Telescopes help to characterise, and hence, to understand exoplanet atmosphere to such details that enables us to link spectroscopic features to evolutionary states of the planets. Cloud formation in the chemically diverse atmospheres of exoplanets has become a key obstacle to conclusively derive their atmospheric compositions. This project therefore focuses on the formation of (cloud) condensation nuclei (CCNs) as the triggering process for gas condensing into solid particles in extaterrestrial, astrophysical environments. This project applies machine learning (ML) techniques to study the formation of metal-oxide clusters as the base for modelling CCN formation. Current approaches at modelling the CCN formation rate as part of complex atmosphere models are challenged by the fact that only the start (N = 1 - 14) and end (N > 150) of the CCN formation process can be modelled, as accurate thermodynamical data on the intermediate size clusters (N = 15 - 150) are not available. Since the number of CCNs determines the cloud particles sizes and the gas phase depletion, it is a key ingredient for a cloud formation model. High-precision thermodynamic data for small TiO2-clusters and their isomers will be used as training set in order to develop ML methods and to test strategies to derive thermodynamic properties for underexplored cluster sizes beyond the training set. Key questions that we aim to answer are what resolution in cluster space is needed to accurately describe cluster properties, how higher computational efficiency outweighs cluster property uncertainties, and how to progress the application of ML technology in exoplanet astrochemistry. Our aim is to leverage the potential of machine learning technology to progress our understanding of exoplanet atmospheres by addressing one of the fundamental research questions of exoplanet climate research: How do clouds form in these chemically diverse atmospheric environments.