Calibration of electron shower shapes, hadronic recoil reconstruction using deep learning algorithms and the measurement of W boson transverse momentum distribution with the ATLAS detector

The initial part of the thesis contains the description of the method for electromagnetic calorimeter calibration, correcting for the Data-MC discrepancy in the development of the electromagnetic showers in the calorimeter. The method improves electron identification and reduces the associated systematic uncertainty. The major part of the thesis is dedicated to the precise measurement of the W boson transverse spectrum using the data, collected by the ATLAS experiment at the energies of 5 and 13 TeV during two special low pile-up runs in 2017 and 2018. The motivation for the precise measurement of the W boson transverse spectrum is twofold. First, it serves as a test for the theoretical predictions obtained within the Standard Model and allows to benchmark the performance of the Monte-Carlo (MC) generators. The second reason is because the W pT spectrum is an input component for the measurement of the W boson mass which is a Standard Model parameter. The use of low pile-up data allows to significantly reduce the hadronic recoil systematic uncertainty improving the precision of the spectrum measurement. The thesis describes the methodology of the W boson pT spectrum measurement as well as the imposed calibrations, corrections and the associated uncertainties. The final result is obtained from the measured hadronic recoil using an unfolding procedure and is compared to the theoretical predictions obtained with different Monte-Carlo generators. An alternative method for the hadronic recoil reconstruction with the use of deep neural networks is proposed in the thesis. The method is shown to improve the resolution of the measured hadronic recoil by about 10% in the most relevant region of low pT. The observables obtained using approach improve the sensitivity to the mass of the W boson.