Development of a likelihood-based top quark mass measurement using semileptonically decaying ttbar+jet events with the full Run 2 ATLAS dataset

Thumbnail Image
Publication date
Reading date
Journal Title
Journal ISSN
Volume Title
The top quark is the most massive elementary particle in the Standard Model, which makes it play an important role in this theory and many of its extensions. This thesis presents the development status of a measurement of the top quark mass in the on-shell renormalization scheme, the top quark pole mass ($m_{t}^{pole}$), using semileptonically decaying top quark pair events produced in association with at least one energetic jet (ttbar+jet). Additionally, the top quark mass parameter in the Monte Carlo simulation ($m_{t}^{MC}$), is determined using semileptonically decaying top quark pair events produced in the lepton+4 jets channel, without additional jets (ttbar). The measurement is performed using the proton-proton collision data at 13TeV collected by the ATLAS experiment at the Larege Hadron Collider (LHC) during Run 2 (2015-2018), corresponding to an integrated luminosity of $140 fb^{-1}$. The analysis starts with the identification of events with signatures that are compatible with originating from a $t\bar{t}$ or $t\bar{t}$+jet event decaying semileptonically. These event candidates are required to contain one electron or muon, a significant amount of missing transverse momentum, and four or more jets, from which at least one is identified as a jet originating from the hadronization of a b-quark. To reconstruct the observables used in this analysis, $\rho_s= 340 \text{GeV} / \sqrt{s_{t\bar{t}+jet}}$ with $\sqrt{s_{t\bar{t}+jet}}=m_{t\bar{t}+jet}$ and $M_{lb}$, a machine learning model implemented within the SPA-Net framework is used. This model is trained to address the jet-to-parton assignment problem, and reconstruct the kinematics of the $t\bar{t}$ and $t\bar{t}$+jet systems via the regression of the $p_T$ and $\eta$ of the escaping neutrino in each of the semileptonic decaying events. Additionally, the machine learning model also produces a series of discriminants that can be applied to enhance the fraction of correctly matched events, which further refines the accuracy of the jet-to-parton assignment. The deep learning approach employed for the reconstruction of these event topologies has shown demonstrated superior performance compared to any of the alternative kinematic-based reconstruction algorithms tested in this thesis. The value of $m_{t}^{MC}$, and the corrected distribution of the $t\bar{t}$+jet parton level normalized cross section with respect to $\rho_s$ are determined using a statistical model based on a profile likelihood fit. This fitting technique allows for the extraction of these parameters while considering the effects of all statistical and systematic uncertainties affecting the analysis in a well-defined statistical framework. Given its capability to simultaneously extract $m_{t}^{MC}$ and the parton level normalized distribution, this approach mitigates any potential biases that may arise from the assumption of the top quark MC mass in the simulation. Consequently, it ensures an unbiased determination of the top quark pole mass in the $\chi^2$ fit of the corrected normalized parton level distribution to Fixed Order theory predictions computed at next-to-leading order for the $t\bar{t}$+jet process. For the top quark pole mass, the methodology outlined in this thesis achieves an expected precision level of $\Delta m_{t}^{pole} (total) \simeq ^{+0.88}_{-0.78}$ GeV, while the top quark Monte Carlo mass parameter in the simulation is expected to be determined with a precision of $\Delta m_{t}^{MC} (total) \simeq 0.41$ GeV. At this precision, the top quark mass values extracted using this methodology would be positioned among the most precise determinations of these quantities to date.
Financiado con una ayuda para la Formación de Profesorado Universitariobeca (FPU) financiada por el ministerio de educación. Numero de referéncia FPU017/04125.
Bibliographic reference