Cosmic ray research accelerated by machine learning insights
Harnessing AI for cosmic ray research, Glombitza employs machine learning (ML) techniques to rapidly process data from observatories studying cosmic radiation. "The results suggest that the most energetic particles hitting the Earth are usually not protons, but significantly heavier nuclei such as nitrogen or iron atoms," says Jonas Glombitza. His work, recently featured in Physical Review Letters, marks only the second instance of ML integration in astroparticle physics.
Glombitza's journey into AI began in 2017 at RWTH Aachen, where he initiated ML tool development, later joining FAU in 2022. Recognized with the ETI Award in 2025 for his contributions, he remains cautious with the term "artificial intelligence" due to its contentious and often misunderstood connotations. Initially, his colleagues were wary of ML's "black box" nature, but acceptance grew after AI results aligned with telescope-based verifications.
The focus of this research is ultra-high-energy cosmic radiation, originating from galaxies beyond the Milky Way. These atomic nuclei, carrying energies from 10^18 to 10^20 electron volts, are the most potent particles in existence. Upon entering Earth's atmosphere, they interact with air molecules, initiating a cascade of secondary particles such as electrons, positrons, photons, and muons. This chain reaction produces fluorescence light through interactions with atmospheric nitrogen, detectable by telescopes like the Pierre Auger Observatory, the premier facility for cosmic ray study.
"The measurements there have been running for 15 years," says Glombitza. Based on atomic formation theories, these cosmic rays may include all elements from hydrogen to iron. Due to its substantial mass, an iron nucleus triggers a far more intricate cascade compared to a lone proton. Consequently, the peak of fluorescence light in an iron-induced shower appears higher in the atmosphere, while lighter particles create deeper-reaching cascades.
Fluorescence light analysis offers clues to the mass of primary particles, yet its reliance on clear, moonless nights limits data availability. In contrast, surface detectors function continuously, though they previously lacked the ability to reconstruct shower maxima from particle spread patterns.
AI now bridges this gap. After training on simulations of particle showers, the AI deciphers particle distribution patterns to infer the mass of primary cosmic rays. These models are subsequently fine-tuned with actual telescope data, enabling the evaluation of 60,000 surface-detected events. "To achieve the same results without AI, we would have had to observe with the telescopes for 150 years. This is the breakthrough I have achieved," says Glombitza.
Research Report:Inference of the Mass Composition of Cosmic Rays with Energies from 1018.5 to 1020??eV Using the Pierre Auger Observatory and Deep Learning
Related Links
Friedrich-Alexander Universitat Erlangen-Nurnberg
Stellar Chemistry, The Universe And All Within It
