Wing Sheung Chan

72 Event selection and classification Using these properties to our advantage, events in the SR are required to have m T ( τ had - vis , E miss T ) < 35 GeV to remove the majority of Z → τ τ and W +jets events. The events that are not selected are split by a cut on m T ( `, E miss T ) into two different regions, each enriched in either Z → τ τ or W +jets events. These regions are used to control the Z → τ τ background and estimate fakes from W +jets events, and will be detailed in the next subsection (Section 4.1.4) . Finally, outputs of neural network classifiers, which will be described later in Section 4.2, are used to refine the SR selection. Events that are classified as the most background-like are rejected. Due to the difference in background compositions, the SR (and regions that will later be defined) is further split into two regions depending on the number of tracks associated to the τ had - vis candidates. The region with 1-prong (3-prong) τ had - vis candidates is denoted with “1P” (“3P”). Table 4.2 summarises the selection criteria for events in the SR. Table 4.2.: Selection criteria for events in the signal region. The last two criteria involve outputs of neural network classifiers, which are described in Section 4.2. Selection criteria eτ channel µτ channel At least one τ had - vis candidate One electron and no muon One muon and no electron Opposite-sign charged e – τ had - vis pair Opposite-sign charged µ – τ had - vis pair m vis ( e, τ ) > 60 GeV m vis ( µ, τ ) > 60 GeV | η ( 1-prong τ had - vis ) | > 0 . 1 m T ( τ had - vis , E miss T ) < 35 GeV No b -jets NN comb output > 0 . 1 (0.2) for 1P (3P) region NN Z `` output > 0 . 2 4.1.4. Control regions and fakes-enriched regions Events that are rejected by the SR selection contain almost no potential signal events. Despite that, they are still useful in helping to constrain the background modelling. These events are split into several regions that are each enriched in events from a particular background process. These regions are the so-called control regions (CR). To help constrain the overall yield of the Z → τ τ background and reduce systematic uncertainties related to the modelling of real `τ had - vis signal states, a control region enriched