Tuning of Monte Carlo algorithms with Bayesian methods for the improvement of cosmic ray induced air shower simulations

Abstract

This thesis investigates the Bayesian tuning of hadronic interaction models used in Monte Carlo simulations of cosmic-ray-induced air showers. Since hadronic interactions govern the production of secondary particles in the atmosphere, their modeling is essential for describing air-shower observables and for interpreting cosmic-ray measurements.

The hadronic event generator Pythia 8 is used in this study and constrained with accelerator data from collider and fixed-target experiments. Although the highest-energy cosmic-ray-induced air showers extend beyond the reach of current accelerator measurements, the selected observables constrain aspects of hadronic particle production that are also relevant during air-shower development at lower energy scales. Rather than tuning directly to air-shower data, the aim is to investigate how constraints from accelerator measurements propagate to air-shower simulations.

The analysis focuses on a selected set of sensitive model parameters and observables. The thesis also includes a comparison of surrogate models that parameterize the generator response, thereby enabling the Bayesian tuning procedure. The choice of surrogate model is shown to be crucial both for the quality of the tune and for the reliability of the uncertainty propagation to the simulated observables.

The results show that the selected collider and fixed-target data sets cannot be described optimally by a single common tune, as they favor different regions of parameter space. Separate tunes are therefore constructed and studied. These tunes improve the description of the corresponding accelerator data sets and, in turn, induce measurable changes in air-shower simulations. In particular, the predicted average muon number changes by about 7-9%, demonstrating that accelerator-based tuning can directly impact the modeling of cosmic-ray-induced air showers.