ISSN 1063-7796, Physics of Particles and Nuclei, 2023, Vol. 54, No. 1, pp. 227–231. © Pleiades Publishing, Ltd.

, 2023.

Estimation of Electron-to-Photon Misidentification Rate

in Z(νν)γ Measurements for Conditions
of ATLAS Experiment during Run II
A. Kurovaa, *, E. Soldatova, **, and D. Zubova, ***
a National Research Nuclear University (MEPhI), Moscow, 115409 Russia
*e-mail: anastasia.kurova@cern.ch
**e-mail: eysoldatov@mephi.ru
***e-mail: dmitriy.zubov@cern.ch
Received June 22, 2022; revised July 17, 2022; accepted August 14, 2022

Abstract—Z boson peak tag-n-probe method was widely used for estimation of electron-to-photon misiden-
tification rate for photon-oriented studies in Run I and in early Run II pp-collisions data. The increased lumi-
nosity and energy of collisions in Run II requires an improvement of the existing method, especially it is nec-
essary for a description of underlying background spectra in the Z boson mass region of tag-n-probe mass dis-
tribution. This study presents the improved ways of tag-n-probe mass spectra fit in order to estimate a pure
number of tag-n-probe events originating from the Z boson. Two approaches are considered. The first fit,
with the exponential polynomial functions, does not include the Z peak itself. And the second one includes
the Z peak, which is described by a Voigtian function. Comparison of e → γ misidentification rate estimation
with two approaches is presented. The study is done for conditions of ATLAS experiment during Run II at
LHC and photon selection used in Z (νν)γ measurements.

DOI: 10.1134/S1063779623020065

INTRODUCTION energy deposition in EMC, which may be accompa-

Z boson peak tag-n-probe method was widely used
for estimation of electron-to-photon misidentification
rate for photon-oriented studies in Run I and in early
Run II pp-collisions data at LHC. It is based on the
assumption that if a reconstructed photon together
with a reconstructed electron or positron have an
invariant mass in the range of the Z boson peak, this
photon is in fact an electron or positron which was
misidentified as photon.
RECONSTRUCTION OF ELECTRONS
AND PHOTONS IN ATLAS DETECTOR
Two main independent detector systems are used in
ATLAS for reconstruction and identification of elec-
tromagnetic particles: the Inner Detector (ID) and the
Electromagntic Calorimeter (EMC) [1–3]. ID
records tracks of electromagnetically charged particles
with 3 tracking detectors: silicon-based Pixel and
Semiconductor tracker (SCT) detectors and gas-based
Transition Radiation Tracker (TRT) detector. The
EMC is used for measurements of the energy losses of
the electromagnetic particles due to the interaction
with medium. In most cases an electron is registered in
ATLAS with the track in ID and energy deposition in
EMC. A photon can be identified by the presence of
nied by two opposite charged tracks with the common
vertex of origin in the ID. The latter case may corre-
spond to a converted photon, which is a photon
decayed into e +e − pair in ID. A probability of such
photon conversion increases with energy of the pho-
ton. Mostly an electron is mistaken as a converted
photon if there are accidentally two matching tracks in
the ID with indistinguishably close points of origin.
Such misidentification may also happen if the track of
an electron is not reconstructed in ID for some reason.
ESTIMATION OF e → γ MISIDENTIFICATION
IN Z (νν)γ MEASUREMENTS
AT ATLAS EXPERIMENT
Tag-n-probe method originally was used for mea-
surements of efficiency or resolution, where some
well-known decay with a simple kinematics is used to
create an unbiased sample of physics objects [4]. The
Z boson decay is widely used for electron and photon
efficiency measurements [5, 6]. For this reason it is
adapted for e → γ misidentification rate measure-
ment. In tag-n-probe method one of the signature
particles is used for efficiency determination and
called “probe”, the second one is called “tag”. Usage

227
228 KUROVA et al.

Entries Then the measured fake rate can be used to scale

450
s = 13 TeV, 139 fb–1
400
350
300
250
200
150
100
50
0 50 100 150
Mee, GeV
Fig. 1. Distribution on the invariant mass of the Drell–Yan
e + e − production in the modelling.
of well-known decay preserves independence on the
“tag” particle selection.
Z (νν)γ measurements [7–9] use isolated high- pT
photons of good quality, such kind of photons is used
as a “probe” particle. To match the selection of the
“probe” photon, “probe” electron is also selected to
have the same pT threshold. “Tag” particle is selected
to have opposite charge in case of e +e − pair in order to
have Z boson peak in the spectra. In both pairs “tag”
particles should pass the same selection criteria. Thus
the tag-n-probe pairs are selected as follows:
• e γ pair selection:
photon with pT > 150 GeV (probe),
selected electron with pT > 25 GeV (tag);
• e +e − pair selection:
selected electron with pT > 150 GeV (probe),
selected opposite sign electron with pT > 25 GeV
(tag).
Photon pT threshold is determined by the lowest-
pT unprescailed single-photon trigger of 140 GeV.
“Tag” particle pT threshold is chosen in order to have
good quality of the reconstruction and identification.
With that said the e → γ fake-rate can be estimated as
N eγ
fake-ratee →γ = , (1)
N+−
e e
where N eγ , N e+e− are the number of events with e +e −
and e γ tag-n-probe pairs whose invariant mass lies
in Z boson peak mass window ( M Z − 10 GeV,
M Z + 10 GeV).
the necessary control region in data, where the probe
electron is selected instead of photon.
BACKGROUND UNDER Z BOSON PEAK
However to get the precise number of events com-
ing from Z boson in data one needs to subtract back-
ground events in Z boson peak region. This back-
ground for Z (νν)γ energies mainly consists of Drell–
Yan production of e +e − pairs. Then Eq. (1) becomes
the following
200
N eγ − N ebkg
γ
fake-ratee →γ = , (2)
N e+e− − N ebkg
+ −
e
where N ebkg bkg + −
γ , N e +e − are the numbers of events with e e
and e γ tag-n-probe pairs, which belongs to under-the-
peak background.
In Run I the background distribution under
Z boson peak was considered to be linear for simplic-
ity. Thus the linear extrapolation of the left and right
sides of the spectra were used to estimate under-the-
peak background.
However at 13 TeV pp-collisions energy and with
larger photon pT the more peculiar behaviour of this
background became visible (see Fig. 1). This shape is
caused by the combination of reconstruction and
identification efficiencies overlapped with the kine-
matic distribution on electron pT (see Fig. 2). These
“waves” in the distribution a.k.a. “threshold effect”
appear in the pT spectra in correspondence with the
pT thresholds applied during selection. This behaviour
makes a simple linear approximation not valid for
under-the-peak background description.
In data one can not distinguish between back-
ground and Z boson production in e +e − or e γ mass
spectra, thus it is safer to subtract all events below the
Z peak.
This work considers two ways of under-the-peak
background subtraction basing on the different fit
techniques. The first one, which uses the exponential
polynomial functions, does not include the Z peak
area. And the second one includes the Z peak area,
which is described by a Voigtian function.
The goal of this work is to compare these two meth-
ods in order to choose the optimal one for further
analyses.
The calculation of fake rate is made regardless of
its dependencies on η and pT for simplicity of com-
parison.

PHYSICS OF PARTICLES AND NUCLEI Vol. 54 No. 1 2023

 ESTIMATION OF ELECTRON-TO-PHOTON MISIDENTIFICATION RATE 229

METHODS AND TOOLS 1.00

ATLAS Preliminary
Modelling
To simulate reconstructed data the modelling of 0.99
pp → Z → e +e − and pp → e +e − (Drell–Yan) pro-

Efficiency
cesses is performed. 0.98

The generation of Z → e +e − and Drell–Yan e +e − 0.97

production is made with MadGraph5_aMC@NLO s = 13 TeV, 43.8 fb–1
ver. 2.7.2 [11]. For both processes 2M events are gen- –2.47 < η < 2.47
erated at the leading order with corresponding cross- 0.96 Data
MC
sections of 1.1920 ± 0.0006 pb for Z → e +e − and Electron reconstruction
0.95
0.07404 ± 0.00016 pb for Drell–Yan e +e − produc- 1.010

Data/MC
tion. Pythia6-PGS ver. 2.4.5 [12, 13] is used for simu- 1.005
lation of ATLAS detector and subsequent particle 1.000
reconstruction. Results are presented for pp-collisions 0.995
with s 13 = TeV and integral luminosity 139 fb–1, 0.990
which corresponds to the statistics collected by the 20 40 60 80 100 120 140
ATLAS experiment in Run II. ET, GeV
45 000
40 000
dN/dpT, arb. units
Exponential Polynomial Fit
35 000
The exponential polynomial functions of order 1
and 2 are used for this method of fit, they are described 30 000
with the function 25 000
20 000
 n 

f ( x) = exp  pi x i  ,
 i =0 
15 000
10 000
where n is the order of polynomial. Different orders of 5000
polynomials are selected in order to fit the shape of
background in e +e − and e γ pair mass spectra. The fit 0 20 40 60 80 100 120 140 160 180
ranges are optimized to have the best quality and sta- Electron pT, GeV
bility of the fit.
For e +e − pairs the background fit is extrapolated to
Z peak mass window from both sides (see Fig. 3, top Fig. 2. Electron reconstruction efficiencies as a function of
plot). Integrals under the fit function in this region transverse energy E T measured in 200 bins in E T × η in
result in numbers of events N min and N max . Then the 43.8 fb–1 of data recorded by the ATLAS experiment in
average number of events is calculated as 2017 at a centre-of-mass energy of s = 13 TeV [10] (top).
N max + N min Distribution of the reconstructed electron pT in Drell–
bkg
N average =
2 Yan e + e − production modelling (bottom).

and it is used as N eebkg in fake-rate calculation with

Eq. (2).
For e γ spectra the background fit is done simulta-
neously in the right and left sides and then extrapo-
lated to the Z peak mass window (see Fig. 3, bottom
plot). An integral of the resulting function inside the
• In e γ pairs the extrapolation function parame-
ters are varied by their uncertainties one by one and
the resulting integral of the fit function is used for vari-
ation of N ebkg
γ .

Z peak mass window is used as Nbkg e γ in Eq. (2).
A systematic uncertainty on under-the-peak back-
ground estimation is evaluated by variation of Nbkg val-
ues in e +e − and e γ pairs as follows:
• N min and N max values are used as variations of
bkg
N ee for fake rate calculation;
Sum in quadrature of the largest variations of N ebkg
γ
and N eebkg in relative quantities is taken as systematic
uncertainty. It equals 1.2%.
The resulting value of e → γ fake-rate is
0.0212 ± 0.0004 ± 0.0002 , where the first uncertainty
is statistical and the second one is systematic.

PHYSICS OF PARTICLES AND NUCLEI Vol. 54 No. 1 2023

230 KUROVA et al.

105 12 000 σ = 0.880 ± 0.011 GeV

M = 90.2764 ± 0.0056 GeV
χ2/NDF = 1.091
–1
s = 13 TeV, 139 fb bkg yield = 2912 ± 639 Signal + background
10 000 p0 = –0.00307 ± 0.0015
4 Background only
10 Signal yield = 162993 ± 763
8000

Events/bin
p 0 + p1 x p0 + p1x
f(x) = e f(x) = e
Events/bin

103
6000
102 4000

10 2000
0
1 χ /NDF = 2.874 χ2/NDF = 2.359
2
85 86 87 88 89 90 91 92 93 94 95
0 50 100 150 200 Mee, GeV
Mee, GeV 500
χ2/NDF = 2.091
103 σ = 3.33 ± 0.15 GeV
–1 M = 90.438 ± 0.076 GeV Signal + background
s = 13 TeV, 139 fb 400 bkg yield = 46 ± 133
Background only
p0 = 0.010 ± 0.030
102 300 Signal yield = 3805 ± 146

Events/bin
Events/bin

10 200

1 2 100
f(x) = ep0 + p1x + p2x

10–1
χ /NDF = 3.008
2
0 50 100
Meγ, GeV
Fig. 3. Polynomial fit of e + e − (top) and e γ (bottom) mass
spectra. Z boson peak range is shown with the dashed
line. Resulting fit functions and their extrapolations are
shown with continuous line. Fit range is shown with
filled rectangles.
Voigtian Fit
Voigtian function is defined as a convolution of the
Gausian and Breit Wigner distributions:
Voigtian(E , M , Γ, σ)
+∞
=
−∞
 Gauss( x, M , σ)Breit Wigner(E − x, M , Γ).
Here Breit Wigner is a function of three parameters
and it describes Z boson decay. A Gauss function with
2 parameters describes the energy smearing in the
detector. Γ parameter describes the width of the
Z boson and it is fixed to the table value of 2.5 GeV in
order to achieve stability of the fit. M , E , σ are con-
sidered as free parameters.
The mass spectra of tag-n-probe particles are fitted
with the sum of Voigtian, which describes the Z boson
peak, and a linear function f ( x) = 1 + p0 x represent-
ing the under-the-peak background.
0
80 82 84 86 88 90 92 94 96 98 100
150 200 Meγ, GeV
Fig. 4. Fit of e + e − (top) and e γ (bottom) mass spectra
with the sum of Voigtian and linear functions. Dashed line
represents the fit result of the signal and background sum.
Solid line represents the resulting background fit function.
The background out of Z boson peak range
( M Z − 10 GeV, M Z + 10 GeV) is not considered in the
fit in contrast with the previous method.
Numbers of events N eebkg and N ebkg
γ for under-the-
peak background in this case are determined as inte-
grals under the resulting linear functions. And the
integrals under resulting Voigtian functions account
for the number of events N ee and N eγ .
For e +e − pairs the fit was made in range
(85.95) GeV, and the extrapolated to then full range
(80.100) GeV. This is done to overcome the peak
asymmetry caused by the threshold effect, since the
further from the threshold the less impact efficiency
has on the spectra.
Resulting fit functions are shown in Fig. 4 for e +e −
pairs in the top and for e γ pairs in the bottom.
The systematic uncertainty on fake rate is esti-
mated as uncertainty on the ratio in Eq. (2). In this
ratio each number of events has its own systematic
uncertainty obtained from the fit function integrals
taking into account fit parameter uncertainties and

PHYSICS OF PARTICLES AND NUCLEI Vol. 54 No. 1 2023

 ESTIMATION OF ELECTRON-TO-PHOTON MISIDENTIFICATION RATE
their correlations. Resulting systematic uncertainty is
3.9% and the resulting value of e → γ fake-rate
0.0212 ± 0.0003 ± 0.0008 , where the first uncertainty
is statistical and the second one is systematic.
CONCLUSIONS
Electron-to-photon misidentification rates are
obtained for photons with pT > 150 GeV used in
Z (νν)γ measurements for conditions of ATLAS
experiment during Run II. Two methods of back-
ground-under-Z-peak subtraction are considered.
The first one does not include the Z peak and uses
exponential polynomial functions. The second one
includes the Z peak, which is described by a Voigtian
function. Resulting estimates for both methods are the
following:
• e → γ fake-rate with expo-pol. fit
0.0212 ± 0.0004 ± 0.0002 (syst. unc. 1.7%);
• e→γ fake-rate with Voigtian fit
0.0212 ± 0.0003 ± 0.0008 , (syst. unc. 3.9%).
Systematic uncertainty for Voigtian method is at
least twice larger, though the results have good agree-
ment within uncertainty. More sophisticated Voigtian
method of the fit does not improve the result, so for
the current pp-collision energy at the LHC the expo-
nential method remains to be a satisfactory approach
for photons with pT > 150 GeV reconstructed in
ATLAS. Presumably exponential polynomial fit can
be used for studies with high- pT photons in ATLAS for
Run II data.
ACKNOWLEDGMENTS
This work is supported by the Russian Science Founda-
tion under grant 21-72-10113.
CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest.
REFERENCES
1. G. Aad et al. (ATLAS Collab.), “The ATLAS experi-
ment at the CERN Large Hadron Collider,” J. In-
strum. 3, S08003 (2008).
PHYSICS OF PARTICLES AND NUCLEI Vol. 54 No. 1
231
2. G. Aad et al. (ATLAS Collab.), “ATLAS detector and
physics performance: Technical Design Report 1,”
CERN-LHCC-99-014, 1999.
3. G. Aad et al. (ATLAS Collab.), “ATLAS detector and
physics performance: Technical Design Report, 2,”
CERN-LHCC-99-015, 1999.
4. A. Straessner et al. (ATLAS Collab.), “A new tool for
measuring detector performance in ATLAS,” J. Phys.:
Conf. Ser. 219, 032023 (2010). URL: https://
cds.cern.ch/record/1354502.
5. G. Aad et al. (ATLAS Collab.), “Electron performance
measurements with the ATLAS detector using the 2010
LHC proton-proton collision data,” Eur. Phys. J. C 72,
1909 (2012).
https://doi.org/10.1140/epjc/s10052-012-1909-1
6. M. Aaboud et al. (ATLAS Collab.), “Measurement of
the photon identification efficiencies with the ATLAS
detector using LHC Run 2 data collected in 2015 and
2016,” Eur. Phys. J. C 79, 205 (2019).
https://doi.org/10.1140/epjc/s10052-019-6650-6
7. G. Aad et al. (ATLAS Collab.), “Measurements of Wγ
and Zγ production in pp collisions at s = 7 TeV with
the ATLAS detector at the LHC,” Phys. Rev. D 87,
112003 (2013).
8. G. Aad et al. (ATLAS Collab.), “Measurements of Zγ
and Zγγ production in pp collisions at s = 8 TeV with
the ATLAS detector,” Phys. Rev. D 93, 112002 (2016).
9. M. Aaboud et al. (ATLAS Collab.), “Measurement of
the Zγ → νν γ production cross section in pp collisions
at √s = 13 TeV with the ATLAS detector and limits on
anomalous triple gauge-boson couplings,” J. High En-
ergy Phys. 2018, 10 (2018). arXiv:1810.04995.
10. Electron efficiency measurements in 2017 data and
electron identification discriminating variables from
2016 data. https://atlas.web.cern.ch/Atlas/GROUPS/
PHYSICS/PLOTS/EGAM-2018-002. Cited February
2018.
11. J. Alwall et al., “The automated computation of tree-
level and next-to-leading order differential cross sec-
tions, and their matching to parton shower simula-
tions,” J. High Energy Phys. 2014, 79 (2014). arXiv:
1405.0301 [hep-ph].
12. T. Sjöstrand, S. Mrenna, and P. Skands, “PYTHIA 6.4
physics and manual,” J. High Energy Phys. 05, 026
(2006).
https://doi.org/10.1088/1126-6708/2006/05/026
13. J. Conway, PGS 4: Pretty Good Simulation of high en-
ergy collisions. https://conway.physics.ucdavis.edu/re-
search/software/pgs/pgs4-general.htm.
2023