Precision meets sustainability at the LHC: extraction of fundamental parameters of the Standard Model and machine learning techniques in the simulation

This thesis presents precise determinations of two fundamental parameters of the stan-dard model (SM), the strong coupling constant αS (mZ) and the top quark mass mtalong with its width Γt. These contribute importantly to constraints on the stabilityof the SM electroweak vacuum.For the determination of αS (mZ), CMS inclusive jet measurements using LHC proton-proton (pp) collisions at centre-of-mass energies √s = 2.76, 7, 8, and 13 TeV are anal-ysed together, for the first time. This has been made possible by detailed studies ofthe correlations between the different data sets. The CMS jet data are combined withHERA deep inelastic scattering measurements to extract the parton distribution func-tions (PDFs) and αS (mZ), simultaneously. This approach properly accounts for thecorrelation between PDFs and αS (mZ). The resulting value, αS (mZ) = 0.1176+0.0014−0.0016,is the most precise value of αS (mZ) from jet rates to date and has been achievedthrough a comprehensive QCD analysis at next-to-next-to-leading order. Further, therunning of αS up to an energy scale of 1.6 TeV is probed.The measurement of the top quark mass parameter in the Monte Carlo (MC) simula-tion mMCt and Γt from the unfolded differential cross section of top quark-antiquark(tt) and single top quark production in association with a W boson (tW) is per-formed. This analysis uses LHC pp collision data at √s = 13 TeV, collected by theCMS experiment during 2017–2018. Events in the dilepton decay channel are selected.The differential cross section as a function of the invariant mass of the lepton and bquark, mℓb, is unfolded to the particle level. This analysis is the first of its kind us-ing CMS data, employing the state-of-the-art event generator bb4l, which simulatespp → bbℓ+ℓ−ν ¯ν final states and takes into account the interference between tt andtW production. The precision of this measurement is estimated using Asimov pseudo-data, resulting in mMCt = 172.61+0.41−0.44 GeV and Γt = 1.36+0.24−0.28 GeV. The mMCt resultis as precise as the most accurate single-experiment direct mMCt measurement and willpresent the first determination of mMCt from the combined tt and tW cross sectionsfrom the CMS Collaboration once the analysis is unblinded. Further, this analysispromises improved precision as compared to direct measurements of Γt obtained withthe bb4l method.Finally, a machine learning (ML) technique is presented, developed to reweight MCsimulations obtained with a particular set of model parameters to simulations withalternative values of these parameters, or to simulations based on an entirely differentmodel. The reweighting is performed at the generator level by applying the outputof the ML algorithm, stored as weights, to the nominal MC simulation. As a result,detailed detector simulation and event reconstruction are not needed for alternativeMC samples, significantly reducing computational costs by up to 75%. The perfor-mance of the method is studied in simulated tt production and results are presentedfor reweighting to model variations and higher-order calculations. This ML-basedreweighting is already used by the CMS experiment and will facilitate precision mea-surements at the High-Luminosity LHC.