Wing Sheung Chan

Chapter 4. Event selection and classification “Doing or not doing something – they are similar. Both involve an action and sincerity.” — CLAMP, “Y¯uko Ichihara”, xxxHOLiC Throughout the LHC Run 2, tens of billions of pp collision events have been recorded by ATLAS. This enormous amount of data is what provides us the statistical power to search for rare LFV Z decays in the first place. But most of these data are unexciting events from SM processes that are not of our interest. They are the background of the search. At the end of the day, besides the amount of data collected, the sensitivity of a search boils down to two main factors: how well the background can be modelled and how well background events can be rejected while retaining the potential signal events. While the background modelling will be discussed in the next chapter (Chapter 5) , focus will be given to the event selection and the discrimination of the Z → `τ signal from the SM background in this chapter. 4.1. Event selection The event selection serves two purposes: to select events that are reconstructed with high quality and to segregate events from different processes from each other. Selection criteria are mainly based on the multiplicities and kinematics of reconstructed objects, as well as the topologies of the events. These criteria can either be a simple cut on a certain variable, or make use of MVA, such as neural network classifiers. 4.1.1. Event cleaning For an apparatus as complex as the ATLAS detector, it would be impractical to assume that every component works flawlessly all the time. During data taking, inevitably, 67