An important part of calorimetry will be finding electron and photon showers. These showers are very similar, depositing most of their energy in the EM calorimeter, but electron showers have tracks leading to them. The first step is to find clusters that lie mainly in the EM calorimeter and use information about these clusters to train an HMatrix which can be used to categorize them. I will be using the electron and pion datasets to find appropriate measurements to distinguish the two and then test this heuristic with rho -> gamma + gamma and rho+ -> pi+ + gamma events.
Clustering
The first step is to create clusters. Since we are looking for small clusters in the EM cal, we can hope they will be contiguous. Thus we use a Nearest Neighbor clusterer which only adds cells to the cluster which are directly adjacent to those within it:
nnClusterer = new NearestNeighborClusterer(1,1,1,0,0.0);
If a significant amount of energy is in multiple, non-contiguous clusters then we will have to find a way to stitch them together. Fortunately this is not the case with electrons or pions when we make no cuts on time or energy.
For both electrons and pions, an average of about 95% of the energy is contained in the highest energy cluster.
Shower Shape and Profile
Now that we have our clusters, we can analyze them. Cluster objects in org.lcsim have a number of methods for obtaining information about a cluster. One is the getShape() method, which returns an array of six doubles. As of this writing, the last three values are not implemented and always return zero. The first value appears to be a measurement of the shower's energy profile by depth in millimeters. The second two appear to give orthogonal measurements of the shower width. The algorithm uses an inertial tensor to generate these values (see org.lcsim.recon.cluster.util.TensorPropertyClusterCalculator for the nitty gritty details).
The first three values of the getShape() parameter for electrons and pions.