High Granularity Electromagnetic Calorimeter Shower Images

Each HDF5 file contains energy deposits (shower images) created by electrons in one of the two calorimeters, for a specific incident angle of particles. Each HDF5 file has a structure of datasets, where each dataset represents energy deposits for a specific particle energy (in GeV). Particle energies are ranging from 1 to 1024 GeV in powers of 2 and particle angles are ranging from 50 to 90 degrees in a step of 10 (angle of 90 degrees indicates a particle entering the detector perpendicularly). Each dataset has the following structure {number of events,18,50,45}, with 18x50x45 being the granularity of a shower image.

The calorimeter used to produce those data is a setup of concentric cylinders (layers). Each layer consists of active and passive material. The SiW geometry has 90 layers of 1.4 mm of tungsten as passive absorber and 0.3 mm of silicon as active material. The SciPb geometry has 45 layers of 4.4 mm of lead and 1.2 mm of scintillator. The number of readout cells is RxPxZ=18x50x45=40500, representing the cylindrical segmentation (rho,phi,z). The size of a single cell has been chosen to correspond to (approximately) 0.25 Moliere radius along the R axis and 0.5 radiation length along Z axis.

The samples were created with the Par04 Geant4 example, which  demonstrates how to use Machine Learning inference to create energy deposits as a fast simulation model using LWTNN and ONNX runtime.