A team utilizes neural networks to simulate the atmospheres of distant planets.
Experts from LMU, the ORIGINS Excellence Cluster, the Max Planck Institute for Extraterrestrial Physics (MPE), and the ORIGINS Data Science Lab (ODSL) have achieved a significant milestone in studying exoplanet atmospheres. By employing physics-informed neural networks (PINNs), they successfully modeled the intricate light scattering in exoplanet atmospheres with unprecedented accuracy. This innovative approach paves the way for more thorough analyses of exoplanet atmospheres, particularly concerning cloud effects, and could greatly enhance our knowledge of these remote planets.
When remote exoplanets transit in front of their parent star, they obscure a fraction of the starlight, while a smaller fraction passes through the planetary atmosphere. This phenomenon causes changes in the light spectrum, reflecting atmospheric characteristics such as chemical makeup, temperature, and cloud presence. To accurately analyze these observed spectra, scientists need models that can rapidly generate millions of synthetic spectra. By comparing the synthetic spectra with actual measurements, we glean insights into the atmospheric composition of the observed exoplanets. Furthermore, the intricate new data obtained from the James Webb Space Telescope (JWST) requires similarly sophisticated atmospheric models.
Swiftly solving intricate equations with AI
A crucial element in exoplanet studies is the light scattering that occurs in the atmosphere, especially the scattering caused by clouds. Previous models fell short in effectively addressing this scattering, leading to errors in spectral analysis. However, physics-informed neural networks provide a significant advantage, as they can adeptly tackle complex equations. In a recently published study, the researchers trained two such networks. The first model was developed without considering light scattering and achieved remarkable precision, with relative errors typically under one percent. In contrast, the second model included approximations for Rayleigh scattering—the same phenomenon that gives the sky its blue hue on Earth. Although these approximations can be refined further, the neural network’s ability to solve the complex equations marks a noteworthy progress.
Collaborative efforts across disciplines
These new insights were made possible through a distinctive collaboration among physicists from LMU Munich, the ORIGINS Excellence Cluster, the Max Planck Institute for Extraterrestrial Physics (MPE), and the ORIGINS Data Science Lab (ODSL), which focuses on creating innovative AI-based methods in physics. “This collaborative effort not only propels exoplanet research forward but also opens new avenues for developing AI methodologies in physics,” says David Dahlbüdding, the lead author of the study from LMU. “We aim to further enhance our interdisciplinary collaboration in the future to simulate light scattering by clouds with greater accuracy, thereby maximizing the capabilities of neural networks.”
