DARWEN: Data-driven Algorithm for Reduction of Wide Exoplanetary Networks

Exoplanet atmospheric modeling is advancing from chemically diverse one-dimensional (1D) models to three-dimensional (3D) global circulation models (GCMs), which are crucial for interpreting observations from facilities like the James Webb Space Telescope (JWST) and Extremely Large Telescope (ELT). However, maintaining chemical diversity in models, especially in GCMs, is computationally expensive, limiting their complexity. Optimizing the number of reactions and species can address this tradeoff, but transparent and efficient methods for such optimization are lacking in current exoplanet literature. We aim to develop a systematic approach for reducing chemical networks in exoplanetary atmospheres while balancing accuracy and computational efficiency. Our data-driven method selects optimal reduced chemical networks based on accuracy and computational efficiency metrics. This approach can optimize networks for similar planets simultaneously, assign weights to prioritize accuracy or efficiency, and is applicable when including photochemistry. We base our method on sensitivity analysis of a typical 1D chemical kinetics model, applying principal component analysis to the sensitivities. To achieve fast and reliable network reduction, we utilize a genetic algorithm, a machine-learning optimization method that mimics natural selection. We present three schemes tailored for different priorities (accuracy, computational efficiency, and adaptability to photochemistry) that demonstrate improved performance and reduced computational costs. Our genetic algorithm-based method, the first to reduce a chemical network including photochemistry in exoplanet research, offers a versatile and efficient approach to enhance both accuracy and computational efficiency.