How a graph neural network turns the sparse binary hit pattern of a Digital Hadronic Calorimeter (DHCAL) into physics: reconstructing the energy of an incoming hadron and identifying its type. Calorimeter hits become nodes of a graph whose learned message-passing captures the spatial structure of a hadronic shower without hand-tuned weighting or Particle-Flow parameterization. Learners see how PID is framed as multiclass classification (neutrons, pions, kaons, protons) and energy reconstruction as regression, how performance depends on incident angle and readout granularity, and why a learned model can preserve accuracy at coarser granularity - a route to more cost-effective calorimeter designs for future colliders.
Studies whether graph neural networks (GNNs) can improve hadron energy reconstruction and particle identification (PID) for a gas-sampling Digital Hadronic Calorimeter (DHCAL) - a detector with a sim…