Graph Neural Network for Hadronic Calorimeter Cluster Calibration

The Large Hadron Collider (LHC) plays a crucial role in high-energy physics research, including studies of the Higgs boson and searches for new physics such as Supersymmetry. A key aspect of this research is the precise and comprehensive reconstruction of proton-proton collisions within the LHC, which requires continuous improvement of the tools used. Among these tools is the particle flow (PF) algorithm, which utilizes charged particle track and calorimeter cluster information to identify every final-state particle and determine its properties. An effective PF algorithm requires well-calibrated hadronic calorimeter clusters. Currently, PF hadron cluster calibration employs a Chi-Squared model, but ongoing research is exploring the potential of a graph neural network (GNN) as an alternative. My research aims to compare the performance of these two models in terms of resolution and response and propose possible improvements to the GNN. These improvements may include implementing a dynamic learning rate, Kaiming initialization, and residual connections.