What particles do we want to reconstruct? - selvaggi/mlpf GitHub Wiki

We aim to reconstruct the particles that left hits in the detector. Most of these particles are stable particles, however, if a particle is created because of interaction with the detector and this one leaves hits, we also attempt to reconstruct it. In order to do this, if a particle is in the 'to be reconstructed list' and has a daughter in the detector the energy of the particle needs to be corrected.

What they reconstruct in MLPF (the new paper) 'we define the set of target particles for reconstruction by taking all stable Pythia particles that interacted with the detector either directly or through their descendants.'

Good things: With this definition they avoid taking into account stable particles that do not leave hits in the detector

The problems: If a stable particle decays in the detector to another particle (stage 0) the hits might be deposited by the second particle (eg. pion to muon decay is a classic). In this case, attempting to reconstruct the pion would give the wrong pid ground truth and assigning the hits of the pion to the muon would mean that the particle would need to assign a track that originally was from the pion to a muon, this could work if the track is not far but otherwise the track would need to be thrown out anyways. Another similar example is if there are hits of the original particle (stage 1) but the other particles (stage 0) also left hits and are 'far away', generating disconnected showers, then assigning these hits or clusters to the same particle would mean that a particle needs to be reconstructed to two clusters which might not be ideal.

This definition could be corrected by saying that only stable particles without daughters in the detector are taken into account to calculate the GT, and the rest we learn to reconstruct anyways by learning the stable particles. However, the energy of the stable particles need to be corrected to account for reconstructing the daughters separately.