Flavor tagging algorithms for CP violation measurements in CMS

This thesis presents the development of four flavor tagging algorithms to be used in CP violation measurements in CMS. Flavor tagging is an essential tool for CP violation measurement, which often rely on the knowledge of both the initial and final flavor of a meson. Three of the algorithm presented use the production of 𝑏 quarks as 𝑏𝑏 pairs and analyze the 𝑏-hadron unrelated to the signal, while the last one examines the hadronization of the signal meson to infer its flavor. The last algorithms is specifically novel in its kind, as it does not make use of particle identification for the inference or the selection of the input features. All taggers examined are based on Deep Neural Networks and are developed and calibrated using reconstructed 𝐵 $_{s}^{0}$ → 𝐽/𝜓 𝜙(1020) and 𝐵$^{+}$ → 𝐽/𝜓 𝐾$^{+}$ decays. The performance of the taggers is evaluated in a dataset corresponding to 96.48 fb$^{-1}$ collected in proton-proton collisions at $\sqrt{s}$ = 13 TeV during the years 2017 and 2018 by the CMS experiment. After combination of the various algorithms, the total tagging power 𝑃$_{tag}$ is estimated to be 𝑃$_{tag}$ = 5.59 ± 0.02%. With this combination, a ∼ 30% subset of the data sample is used in a time-dependent angular analysis for a measurement of the CP violating phase 𝜙$_{s}$, obtaining a result of 𝜙$_{s}$ = −21 ± 42 (stat.) mrad. A suite of tagging related systematic uncertainties is evaluated on the measurements, showing them to be negligible compared to the statistical uncertainty and to the intrinsic model bias of the fitting procedure.<