miss T
p
50 100 150 200 250 300Ratio
0.6 0.8 1 1.2efficiency
miss TL1+HLT p
0 0.2 0.4 0.6 0.8 1 Data Simulation (13 TeV) -1 35.9 fb CMSFigure 11.2: Trigger efficiency for the pmiss
T part of the trigger, as a function of offline recon-
structed pmiss
T (with Type 1 corrections applied), for data and simulated events.
The corresponding fit functions and the ratio of data and simulation are shown. The offline selection requirement of pmiss
T >90 GeV is illustrated with a light grey
area on the left.
11.2 Offline event selection
11.2.1 Data quality filters
As the identification of the τh+ jets final state relies heavily on the correct reconstruc-
tion of pmiss
T , the first step in the offline event selection is to filter out events that are
likely to contain anomalous pmiss
T , arising from reconstruction failures or detector
malfunction rather than the true particle content of the event.
For this purpose, a set of data quality filters is applied. They include calorimeter filters that identify and reject events with calorimeter anomalies affecting the momentum balance. These anomalies include abnormal noise in the HCAL, exceptionally high energies in the ECAL superclusters, or temporarily nonfunctional calorimeter towers.
156 11. Event selection
Separate filters are used to identify events affected by a beam halo from muons pro- duced in collisions upstream of the detector, causing characteristic energy deposits in calorimeters and CSC muon stations. A third set of filters suppresses misreconstruction effects by looking for high-pT tracks with low track quality, which often correspond to
misreconstructed muons and charged hadrons. Combined, these filters are typically able to reject 90% of the spurious-pmissT events, with per mille level mistagging rate. The data quality filters are described in detail in Ref. [165].
11.2.2 Baseline selections
In the offline selection, low thresholds for the pT of the reconstructed τh and pmissT
are needed to maximize the sensitivity for light H± (with 80 < m
H± < 160 GeV).
Thus selection criteria identical to those in the HLT are applied to the reconstructed τh
candidate and to the offline-reconstructed (PF) Type-I corrected pmissT .
The fact that identical selection thresholds are used both in trigger and in offline selection for τh pTand pmissT means that the events in the turn-on part of the trigger
efficiency enter the offline analysis. Therefore the correct estimation of the turn-on is important for the correct normalization of simulated events.
The one-prong τhcandidates, corresponding to τ decays into a charged pion and up
to two neutral pions, are selected for further analysis. The three-prong τhcandidates
are rejected, as they are found to bring only a small improvement in sensitivity, while they would require a separate jet→τhbackground estimation as the fake factors are
different between 1-prong and 3-prong τhcandidates.
At least τh candidate passing a loose working point of the τh identification MVA
discriminant (Section 8.2.5) is required. The selected working point corresponds to an overall τhidentification efficiency of≈50% (determined from Z/γ∗→τ+τ−events)
and 3×10−3probability for misidentifying a jet as a τ
h(determined from QCD multijet
events).
Events are required to contain at least three jets with pT > 30 GeV and |η| < 4.7,
separated from the reconstructed τhby ∆R >0.5. Loose jet identification criteria as
described in Ref. [167] are applied.
At least one of the jets is required to pass the b jet identification performed with the CSVv2 algorithm described in Section 8.2.4, with |η| < 2.4. The working point for the multivariate discriminant is chosen such that the probability to misidentify jets
11.2. Offline event selection 157
originating from light-flavor quarks or gluons as b jets is 1%, corresponding to 65% efficiency for the selection of genuine b jets.
As no isolated electrons or muons should be present in the fully hadronic final state, any event with isolated electrons (muons) with pT > 15(10)GeV and |η| < 2.5 is
rejected. For muons, loose identification criteria as described in Section 8.2.1 are applied. For electrons, a loose working point with a 95% efficiency is used in the MVA identification described in Section 8.2.2. For both muons and electrons, the isolation criterion is defined by requiring the pT sum of the PF candidates within the mini-
isolation cone (as defined in Section 8.2.1) to be less than 40% of the electron/muon pT.
The presence of leptonic W± decays from the top quark would lead to a smearing of
the high edge of the mTdistribution for the tt and W+jets backgrounds. The lepton
veto also ensures that there is no overlap in the selected events with respect to the analysis targeting the leptonic final state, allowing a statistical combination of the results from the hadronic and leptonic final states.
All these selection steps, excluding the offline pmiss
T and b jet selections, are collec-
tively referred to as baseline selections. The baseline selections loosely ensures that the collection of physics objects in the selected events corresponds to the detector fingerprint expected from the signal events, but does not yet suppress the dominant jet→τhbackground. After the baseline selections, the b jet selection efficiently reduces
both jet→τhand W+jets backgrounds, and the pmissT selection further suppresses the jet→τhevents.
Finally, as described in the following, angular selections are applied to reduce the dominant jet→τhbackground, while the categorization based on the Rτ variable is used to discriminate between the genuine-tau background events (mostly from tt production) and the H± signal events.
11.2.3 Angular selection
After the baseline selections, the jet→τh background is dominated by QCD multijet
events where a jet misidentified as a τh is in a back-to-back configuration with the
pmiss
T arising from an incorrect estimation of the jet momenta.
Thus the amount of QCD multijet events can be reduced by placing a selection on the ∆φ difference between the τhand pmissT . But the definition of mT (Eq. (10.1)) implies
158 11. Event selection
where also most of the signal is expected, especially for large H± mass values. Thus a
simple ∆φ cut would also suppress the signal significantly.
The problem is solved by taking also into account the jet directions. Typically QCD multijet events that pass the pmissT cut and enter the signal region are dijet events with two high-pTback-to-back jets. If the momentum of one jet is overestimated and
the other one is underestimated, the resulting momentum imbalance is interpreted as large pmiss
T . The defining feature of these QCD multijet events is that the pmissT is
collinear in φ with one of the leading jets, which is rare for events with pmiss
T arising
from neutrinos, such as H± signal or tt events.
Thus the jet→τhbackground can be efficiently suppressed with a discriminant that
requires the τhand pmissT to be back-to-back (∆φ(τh,~pTmiss)) and one of the leading jets
to align with pmissT (∆φ(jet,~pTmiss)). This discriminant is defined in the (∆φ(τh,~pTmiss),
∆φ(jet,~pTmiss)) plane as
Rminbb =min n q 180◦−∆φ(τh,~pmiss T ) 2 + ∆φ(jetn,~pTmiss)2 , (11.1)
where the index n runs over the three highest pTjets (jetn) in the event.
The selected events are required to have Rminbb > 40◦. The effect of this cut on the (∆φ(τh,~pTmiss), ∆φ(jet,~pTmiss)) plane and the regions where different types of events
(signal events, irreducible tt background and the QCD multijet events) are mostly concentrated are schematically shown in Figure 11.3. The distribution of the Rmin bb
variable after applying all other selections is shown in Figure 11.4.
11.2.4 Categorization of events
While the jet→τhbackground can be efficiently suppressed with the b jet, pmissT and
Rmin
bb selections, and also the W+jets background is reduced with the b jet identification,
the tt and single top production backgrounds with genuine tau leptons, b jets and pmiss
T are largely irreducible.
However, the τ helicity correlations can be utilized to suppress these backgrounds. As discussed in Section 2.5.2, an event sample enriched with H± signal events can
be selected by imposing a cut on Rτ = ptrack/pτh, where ptrack is the reconstructed
three-momentum of the leading charged particle in the τhcandidate, and the pτh is the
11.2. Offline event selection 159 QCD multijet events QCD multijet & tt events Rbb =40 o o o o
(p
missT, ⌧
h)
Signal (500 GeV) XF 0 0 180 180 1 XF XF(p
miss T,jet
n)
Signal (200 GeV)Figure 11.3: A schematic picture of the (∆φ(τh,~pTmiss), ∆φ(jet,~pTmiss)) plane used in the Rminbb
selection. The selection affects the bottom-right corner of the plane, containing mostly QCD multijet events. Also the regions populated with signal events (using mH± =200 GeV and 500 GeV as examples) and with the irreducible tt background are shown for illustration.
Therefore after all selections, the selected events are classified into two categories based on the value of the variable Rτ = ptrack/pτh. The distribution of the Rτvariable
is shown in Figure 11.5. After all other selections, most of the signal events have a large value of Rτ as expected, and the high-Rτ category provides a good signal-to- background ratio. However, for TeV-range mH± hypotheses, the signal events are more
evenly distributed between the two categories, so the inclusion of the background- dominated low-Rτ category in the statistical analysis further improves the overall signal selection efficiency.
At the other end of the mass range, including the low-Rτ category also improves sensitivity for the light H± mass hypotheses of 80–90 GeV, as it allows the data to
constrain the systematic uncertainties related to simulated backgrounds containing W± bosons, which is necessary to distinguish a signal situated almost on top of the
160 11. Event selection 0 50 100 150 200 250
)
o(
min bbR
0.5 1 1.5Data/Bkg.
syst. unc. ⊕ Bkg. stat. Bkg. stat. unc 1 10 2 10 3 10 4 10 5 10 oEvents / 20
> 0.75 τ +jets, R h τ Data H± (200 GeV, σ = 50 pb) h τ Jets misid. as tt W+jets Single t * γ Z/ DibosonBkg. stat. Bkg. stat.⊕syst.
(13 TeV)
-1
35.9 fb CMS
Figure 11.4: The distribution of the angular discriminantRmin
bb after applying all other selec-
tions including the Rτ = ptrack/pτh >0.75 requirement.
Separating the two categories at Rτ =0.75 maximizes the signal sensitivity across the mH± range. In the category defined by the Rτ >0.75 cut, the leading charged particle is required to carry a large fraction (>75%) of the visible τ energy, while the other category is defined by the inverse requirement of Rτ <0.75.