Track reconstruction in the CMS experiment for the high luminosity LHC

Abstract

Tracking is one of the crucial parts in the event reconstruction because of its importance in the estimation of particle momenta, their identification, and in the estimation of decay vertices. This task is very challenging at the LHC, given the hundreds or even thousands of particles generated in each bunch crossing. Track reconstruction in CMS was developed following an iterative philosophy. It uses an adaptation of the combinatorial Kalman Filter (KF) algorithm to allow pattern recognition and track fitting to occur in the same framework. For tt events under typical Run-1 pileup conditions, the average track reconstruction efficiency for charged particles with transverse momenta of pT > 0.9 GeV is 94% for |η| < 0.9 and 85% for 0.9 < |η| < 2.5. During Long Shutdown 1, some developments were made in different aspects of tracking. In particular, I implemented the Deterministic Annealing Filter (DAF) algorithm to protect track reconstruction against wrong hit assignments in noisy environments or in high track density environments. The DAF algorithm showed a reduction of mis-identified tracks in high-pT jets, even more evident inside their core, without any loss in terms of efficiency. During Run 2, which started in 2015 and will continue until the end of 2018, the machine operates at a centre-of-mass energy of 13 TeV with a bunch crossing separation of 25 ns and targeting an instantaneous luminosity of up to 2 × 10 34 cm−2 s−1. During data taking, a possible way to monitor the consistency between data and MC is to measure the muon tracking efficiency using the tag and probe. I found the consistency to be within a few per-mill using Z μ+μ− events, and within a few per-cent for the J/ψ μ+μ− sample. Moreover, I performed a study in the context of the H τ+τ− analysis simulating the Run 2 conditions, with the aim of finding a good signal-versus-background classifier. For this purpose I implemented and optimized a multi-layer perceptron neural network using the NeuroBayes package. I compared its performance to other Multivariate Analysis (MVA) approaches as well as to a cut-based approach, which is the current technique. The network performs better than the cut-based approach, but it does not deliver the best results among the different MVAs analyzed. This is very likely due to the fact that the NeuroBayes neural network requires a larger dataset in order to get a significant improvement in performance. Between 2024 to 2026, the Long Shutdown 3 is scheduled where the main preparation of the accelerator and the experiments will take place for the High Luminosity phase of the LHC (HL-LHC). The HL-LHC will provide an unprecedented instantaneous luminosity of 5 − 7.5 × 10 34 cm−2 s−1. In order to face this challenge, the CMS experiment has decided to install a new silicon-based tracker. In the work described here I laid the foundation for track reconstruction in CMS at the HL-LHC. In a first step, I adapted the existing tracking algorithms to the new tracker. I evaluated the performance of tracking and vertexing for different event types and detector geometries and I found them to be similar to the current ones despite the harsher pileup conditions at HL-LHC. In a second step, I introduced a new type of hits, so-called vector hits, in the outer part of the new tracker, with the aim of answering the question whether they bring tangible benefits in the pattern recognition step at the HL-LHC. To this end, I extended the KF to make use of the full information contained in the vector hits, and I added a new iteration in the Phase-2 tracking, in which seeds are built using only the vector hits. The results show a reduction of about an order of magnitude of the number of mis-identified tracks for high pileup environments and a significant improvement in the reconstruction of tracks coming from displaced vertices, such as the decay products of the Kshort meson.