Searches for New Physics Using Innovative Data Acquisition, Analysis, and Compression Techniques

The high event rate delivered by the Large Hadron Collider (LHC) provides experiments with opportunities for new discoveries as well as challenges related to the large amounts of data recorded. Overcoming this requires innovative techniques in the three topics of this thesis: Data acquisition, data analysis, and data compression.

In the data taking period of 2015-2018, the ATLAS experiment received 10-70 simultaneous proton-proton collision events every 25 ns. At these high event rates, the experiment relies on trigger methods which only process the interesting collision events and keep detector readout and data storage within bandwidth constraints. The Trigger Level Analysis (TLA) presented in this thesis circumvents these bandwidth constraints by using the smaller event objects reconstructed at the trigger level as input to the analysis. The trigger level objects require custom calibration schemes, one of these was developed as part of this thesis to be used in the current and next iterations of the analysis.

The LHC is scheduled to be upgraded to the High Luminosity LHC (HL-LHC) and deliver 200 simultaneous proton-proton collision events. To provide the necessary resolution, readout speed, and radiation hardness, the ATLAS Inner Detector will be upgraded to the new fully silicon-based Inner Tracker (ITk). This thesis presents the work performed in developing, manufacturing, and delivering an automated quality control system for the new detector modules. Quality testing of the detector modules using this system is currently ongoing at multiple international institutes.

The large amount of simultaneous events provided by the HL-LHC will also be challenging for data storage, where the amount of ATLAS generated data is projected to be 5 times larger than the storage resources. As the data are already highly compressed using lossless methods, the work in this thesis presents proof-of-principle studies using machine learning-based methods to derive lossy compression algorithms tailored to a variety of datasets. The tool developed for this purpose is made available as an open-source project called "Baler".