Speakers
Description
Performing a realistic simulation of atmospheric muon flux requires well-established tools like CORSIKA, since they model in detail all components of the shower and the particle transport processes in the atmosphere. However, this accuracy comes with a high computational cost. For example, simulating one hour of flux can take several hours of computing time, which limits its use in feasibility studies for muography. In this type of study, it is often necessary to simulate weeks of flux and explore multiple candidate locations with different geomagnetic conditions and altitudes.
To address this limitation, we developed an algorithm based on artificial intelligence that, trained with CORSIKA simulations (through ARTI), is able to reproduce the energy spectrum and angular distribution of the muon flux at any latitude and longitude. The model was implemented using conditional normalizing flows and learns the flux dependence from a reduced set of site descriptors: altitude and the local geomagnetic field components (Bx, Bz).
Training was performed with simulations at 45 locations worldwide, covering altitudes from 0 to 5230 m a.s.l. (35 for training, 5 for validation, and 5 for testing). Tests on sites not included in the training show that the model preserves the reference spectral profiles, reducing computation time from several hours (≈5–9 h per ARTI simulation) to about two minutes with our model.
Finally, an interactive web application was developed, where users can enter their latitude, longitude, and geomagnetic field components, and instantly obtain the estimated muon flux for that specific location.