Systems and algorithms for low-latency event reconstruction for upgrades of the Level-1 trigger of the CMS experiment at CERN

With the increasing centre-of-mass energy and luminosity of the Large Hadron Collider (LHC), the Compact Muon Experiment (CMS) is undertaking upgrades to its triggering system in order to maintain its data-taking efficiency. In 2016, the Phase-1 upgrade to the CMS Level- 1 Trigger (L1T) was commissioned which required the development of tools for validation of changes to the trigger algorithm firmware and for ongoing monitoring of the trigger system during data-taking. A Phase-2 upgrade to the CMS L1T is currently underway, in preparation for the High-Luminosity upgrade of the LHC (HL-LHC). The HL-LHC environment is expected to be particularly challenging for the CMS L1T due to the increased number of simultaneous interactions per bunch crossing, known as pileup. In order to mitigate the effect of pileup, the CMS Phase-2 Outer Tracker is being upgraded with capabilities which will allow it to provide tracks to the L1T for the first time. A key to mitigating pileup is the ability to identify the location and decay products of the signal vertex in each event. For this purpose, two conventional algorithms have been investigated, with a baseline being proposed and demonstrated in FPGA hardware. To extend and complement the baseline vertexing algorithm, Machine Learning techniques were used to evaluate how different track parameters can be included in the vertex reconstruction process. This work culminated in the creation of a deep convolutional neural network, capable of both position reconstruction and association through the intermediate storage of tracks into a z histogram where the optimal weighting of each track can be learned. The position reconstruction part of this end-to-end model was implemented and when compared to the baseline algorithm, a 30% improvement on the vertex position resolution in tt events was observed.