AI sheds new light on exoplanets

Researchers from LMU, the ORIGINS Excellence Cluster, the Max Planck Institute for Extraterrestrial Physics (MPE) and the ORIGINS Data Science Lab (ODSL) have made a major breakthrough in the analysis of exoplanet atmospheres.

Using physics-based neural networks (PINNs), they managed to model the complex scattering of light in the atmospheres of exoplanets with greater accuracy than was previously possible.

This method opens new perspectives for the analysis of the atmospheres of exoplanets, particularly with regard to the influence of clouds, and could considerably improve our understanding of these distant worlds.

The book is published in the Monthly Notices of the Royal Astronomical Society.

When distant exoplanets pass in front of their star, they block a small portion of the star’s light, while an even smaller portion enters the planet’s atmosphere. This interaction causes variations in the light spectrum, which reflect properties of the atmosphere such as chemical composition, temperature, and cloud cover.

To be able to analyze these measured spectra, however, scientists need models that can calculate millions of synthetic spectra in a short time. Only by subsequently comparing the calculated spectra with the measured spectra can information about the atmospheric composition of the observed exoplanets be obtained.

Additionally, new, highly detailed observations from the James Webb Space Telescope (JWST) require equally detailed and complex atmospheric models.

Fast solving of complex equations with AI

A key aspect of exoplanet research is the scattering of light in the atmosphere, particularly that occurring from clouds. Previous models failed to adequately capture this scattering, leading to inaccuracies in spectral analysis.

Physics-based neural networks offer a decisive advantage here, as they are able to solve complex equations efficiently. In the study just published, the researchers trained two such networks. The first model, developed without taking light scattering into account, demonstrated impressive accuracy with relative errors typically below 1%.

The second model incorporates approximations of Rayleigh scattering, the same effect that makes the sky appear blue on Earth. Although these approximations still need improvement, the neural network was able to solve the complex equation, which is an important step forward.

Interdisciplinary collaboration

These new findings were made possible by a unique interdisciplinary collaboration between physicists from LMU Munich, the ORIGINS Excellence Cluster, the Max Planck Institute for Extraterrestrial Physics (MPE) and the ORIGINS Data Science Lab (ODSL), which specializes in developing new AI-based methods in physics.

“This synergy not only advances exoplanet research, but also opens up new horizons for the development of AI-based methods in physics,” says the study’s lead author, David Dahlbüdding from LMU.

“In the future, we want to further expand our interdisciplinary collaboration in order to simulate light scattering on clouds more precisely and thus fully exploit the potential of neural networks.”