• No se han encontrado resultados

El currículo de Ciencias Aplicadas a la Actividad Profesional

In the Level-1 trigger, events are selected by requiring the Level-1 pmiss

T , calculated as

described in Section 6.1.4, to be larger than 80 GeV.

152 11. Event selection

In the HLT, an algorithm requiring the presence of a τhcandidate with a loose track-

based isolation, and missing transverse momentum estimated from calorimeter infor- mation (pmiss,caloT ) is applied.

The HLT τhcandidate is reconstructed using a fast cone-based algorithm that proceeds

through several steps (levels), rejecting a fraction of the τh candidates at each level.

First, at level 2, the τh candidates are identified using only the calorimeter energy

deposits, and are required to pass a minimum energy threshold. At level 2.5, a loose isolation is required, based on tracks reconstructed from hits in the pixel detector and associated with the τh decay vertex. The τh candidates passing the isolation,

as well as the candidates for which no vertex was reconstructed, are passed to the next level for additional scrutiny. Finally, at level 3, the full reconstruction of tracks and PF candidates is carried out, including the reconstruction of anti-kT jets with

distance parameter 0.4. The compatibility of each jet with the τhhypothesis is tested,

similarly to the HPS algorithm used offline (Section 8.2.5) but with considerably looser selection criteria. The photons contained in a jet are clustered into η×φ=0.05×0.02 strips, which are combined with charged hadrons found inside a signal cone with ∆R =3.6 GeV/pT (where ∆R is allowed to vary from 0.08 to 0.12).

As the final step, a cone-based isolation is applied by identifying the charged hadrons associated with the τhvertex but not belonging to the τhcandidate within an isolation

cone of ∆R <0.4. The pT sum of these charged hadrons is required to be less than

3 GeV. To improve the selection efficiency, the isolation requirement is relaxed by 6% for τh candidates with pT > 100 GeV. This relaxation can be allowed without

increasing the trigger rate too much, because the number of jets misidentified as τh

passing the selections decreases with pT. A detailed description of τhreconstruction

and isolation in the HLT can be found in Ref. [175].

The HLT τh candidates passing all the above steps, with a reconstructed transverse

momentum pT >50 GeV and with a leading track transverse momentum (from the

leading charged hadron candidate) ptrackT >30 GeV are selected. As the HLT algorithm uses tracker information, and because the tracker extends only up to|η| = 2.5 and an isolation cone of ∆R=0.4 is used, the τhcandidates can be reconstructed only for

|η| < 2.1. For the pmiss

T part of the HLT algorithm, the pmiss,caloT is computed simply as the

negative vector sum of the transverse energies of all calorimeter towers. It is required to be larger than 90 GeV.

11.1. Online event selection 153

With these trigger thresholds, the τh +pmissT signal trigger produced a rate of ap-

proximately 20 Hz during a typical LHC fill with an instantaneous luminosity of 1.4 · 1034cm−2s−1in 2016 [175].

11.1.1 Trigger efficiency

The efficiency of the τh part of the trigger is determined with the tag-and-probe

technique [132], using Z/γτ+τevents with one hadronic and one muonic tau

lepton decay.

These events are selected with a single muon monitoring trigger, requiring a muon with pT > 21 GeV and |η| < 2.1. The tag-and-probe method is applied, using an

isolated muon selected with the monitoring trigger ("tag") and a τhcandidate ("probe")

with pT > 20 GeV, |η| <2.1 and separated from the muon by ∆R(τh, µ) > 0.4. In

addition, at least two jets with pT > 30 GeV are required. The other identification

criteria for objects are similar to those used in the signal selection (Section 11.2.2). The Z/γµµevents are suppressed by requiring 20 <minv(µ, τ

h) <80 GeV while the

W+jets events are reduced with the requirement that mT(µ,~pTmiss) <40 GeV.

As the trigger efficiency is different for τh candidates from genuine taus and from

misidentified jets or leptons, a high genuine-τhpurity of the selected sample is impor-

tant for an unbiased measurement. The purity of events with genuine taus with this selection is found to be>90%.

The measured HLT efficiency for the tau part of the trigger is shown in Figure 11.1 as a function of the τh pT. The efficiency varies between 50 and 100%, as a function of

pT and η of the τh. As the efficiency measured from data differs from the efficiency

obtained using simulated samples, the latter are corrected for the difference as detailed later in Section 13.1.1.

The efficiency of the pmiss,caloT part of the trigger is measured using a prescaled single-τh

monitoring trigger, identical to the τhpart of the signal trigger but with no pmiss,caloT

requirement. The efficiency is defined as the number of the events passing both the signal trigger (with both τhand pmiss,caloT selection requirements) and the monitoring

trigger, compared to the number of events passing the monitoring trigger (with only the τhselection requirements). To select events with a signal-like topology, all baseline

154 11. Event selection

(GeV)

T

p

h

τ

50 100 150 200 250 300 350 400 450 500

Ratio

0.8 1 1.2

efficiency

h

τ

HLT

0 0.2 0.4 0.6 0.8 1 Data Simulation (13 TeV) -1 35.9 fb CMS

Figure 11.1: Trigger efficiency for the τhpart of the trigger, as a function of the offline recon-

structed pT of the τhcandidate, for data and simulated events. The corresponding

fit functions and the ratio of data and simulation are shown. The offline selection requirement of pT >50 GeV is illustrated with a light grey area on the left.

The L1+HLT efficiency of the pmissT part of the trigger is shown in Figure 11.2. and varies between 10 and 100%, depending on the value of the pmissT . Similarly to the τh

trigger case, a correction for simulated events is derived as detailed in Section 13.1.1. To reduce uncertainties in the efficiencies from limited event yields, the measured efficiencies are fitted using suitable fit functions. This way the information gained in the phase regions with large event yields can be exploited to constrain the efficiency estimates in less populated regions. Binned maximum likelihood fits are performed for data and simulation separately. For data (simulation), the Sigmoid function (Crystal ball cumulative distribution function) is used for the τhpart of the trigger, while the

Richards function (Sigmoid function) is used for the pmissT part. The method was developed in Ref. [194], where also the definitions of the fit functions can be found. The fit results are shown in Figures 11.1 and 11.2.

11.2. Offline event selection 155

Documento similar