Truth Seeded Reconstruction for Fast Simulation in the ATLAS Experiment

The huge success of the ATLAS experiment for particle physics during Run 1 of the LHC would not have been possible without the production of vast amounts of simulated Monte Carlo data. However, the very detailed detector simulation is a highly CPU intensive task and thus resource shortages occurred. Motivated by this, great effort has been put into speeding up the simulation. As a result, other timeconsuming parts became visible. One of which is the track reconstruction. This thesis describes one potential solution to the CPU intensive reconstruction of simulated data: a newly designed truth seeded reconstruction. At its basics is the idea to skip the pattern recognition altogether, instead utilizing the available (truth) information from simulation to directly fit particle trajectories without searching for them. At the same time tracking effects of the standard reconstruction need to be emulated. This approach is validated thoroughly and no critical deviations of the results compared to the standard reconstruction are observed. The so called truth-tracking is able to mimic all important aspects such as the track and vertex resolution or the tracking efficiency. Possible speed-ups of up to 16 times are possible for high pile-up environments and even larger ones are anticipated with future algorithm improvements.