Search for single top-quark production via FCNC in strong interactions in

ATLAS-CONF-2013-063 18July2013

ATLAS NOTE

ATLAS-CONF-2013-063

July 9, 2013

Search for single top-quark production via FCNC in strong interactions in

√ s = 8 TeV ATLAS data

The ATLAS Collaboration

Abstract

A search for single top production via flavour changing neutral current processes from initial states of gluon plus light quark in proton proton collisions at the LHC is presented.

Data collected with the ATLAS detector in 2012 at a centre-of-mass energy of

s =

8 TeV and corresponding to an integrated luminosity of

14.2 fb

are used. Candidate events for top quarks decaying into leptons and jets are selected and classified into signal- and background-like candidates using a neural network. As no signal is observed, an upper limit on the production cross-sections multiplied by the

branching fraction is set.

The observed 95% CL limit is

(t

2.5 pb which can be converted into upper limits on the coupling constants of the FCNC interactions

5.1

10

TeV

and

1.1

10

TeV

and on the branching fractions

3.1

10

and

(t

1.6

10

.

c

Copyright 2013 CERN for the benefit of the ATLAS Collaboration.

Reproduction of this article or parts of it is allowed as specified in the CC-BY-3.0 license.

1 Introduction

The top quark is the heaviest elementary particle known, with a mass of m

173.2

0.9 GeV [1] close to the electroweak symmetry breaking scale. This makes it an excellent object to test the Standard Model (SM) of particle physics.

The dominant process for single top-quark production at LHC is the t-channel exchange of a virtual W boson. This mode has been measured by ATLAS and CMS [2–5]. The two sub-leading processes are Wt production, measured for the first time at the LHC with

s

7 TeV [6, 7], and the s-channel production.

As a consequence of the large value, close to one, of the V

-element in the Cabibbo-Kobayashi- Maskawa (CKM) unitary matrix, the predominant decay channel of top quarks is t

Wb. Transitions between top quarks and other quark flavours mediated by neutral gauge bosons, so-called Flavour Chang- ing Neutral Currents (FCNC), are forbidden at tree level and highly suppressed at higher orders in the SM [8]. However, there exist several extensions to the SM which significantly enhance the rates and hence the branching fractions of FCNC processes. These extensions include new exotic quarks [9], new scalars [10,11], supersymmetry [12–15], or technicolour [16] (for a review see Ref. [17] and [18]). If the new particles are heavy, which is consistent with the non-observation of new low-mass particles at the Tevatron and the LHC, their e

ects on FCNCs involving top quarks can be parametrised in terms of a set of dimension-six gauge-invariant operators [19]. The predicted branching fractions

for top quarks decaying to a quark and a photon, Z boson, or gluon can be as large as 10

to 10

[17, 18] for certain regions of the parameter space in the models mentioned.

Among FCNC top-quark decays of the form t

qV with V

Z, γ, g, modes involving a Z boson or photon are usually studied directly by searching for final states with the corresponding decay particles.

However, the t

qg mode, where q denotes either an up quark u or a charm quark c, is nearly in- distinguishable from the overwhelming background of multijet production via quantum chromodynamic (QCD) processes. For the t

qg mode a much better sensitivity can be achieved by searching for anomalous single top-quark production (qg

t) where a u or c quark and a gluon

originating from the colliding protons interact to produce a single top quark. The most general e

ective Lagrangian

for this process resulting from dimension-six operators contains only tensor couplings [20] and can be written as [21]:

L_eff =gs

X

q=u,c

κqgt

Λ

t ¯

T

( f

P

f

P

) q G

h.c., (1)

where the

,

are dimensionless parameters that relate the strength of the new coupling to the

strong coupling constant

.

is the new physics scale. T

are the Gell-Mann matrices and f

are

chiral parameters normalised such that:

f

f

1, while P

are the left-handed and right-hand

projection operators. G

is the gauge-field tensor of the gluon and ¯ t and q are the fermion fields of the

top antiquark and light quark, respectively. This Lagrangian contributes to both production and decay

of top quarks. In the allowed region of parameter space for

for single top quarks is of the order of picobarns, while the branching fraction for FCNC decays is very

small, and top quarks are therefore reconstructed in the SM decay mode t

Wb. The W boson can decay

into quark-antiquark pairs (W

q

q ¯

) or a lepton-neutrino pair (W

¯

Thus the complete process searched for is qg

t

W(→

¯ b. Events are characterised by an isolated

high-energy lepton (electron or muon), missing transverse momentum from the neutrino and exactly one

jet, produced by the hadronisation of the b quark. Events with a W boson decaying into a

lepton, where

the

decays into an electron or a muon, are also included. Several SM processes have the same final state

topology and are considered as background to the FCNC analysis. The main backgrounds are W/Z

jets

production, especially in association with heavy quarks, SM top-quark production, diboson production,

and multijet-production via QCD processes.

The qg

t

b`ν process is characterised by three main differences from SM processes. First, the top quark is produced singly with transverse momentum arising only from initial state QCD radiation.

Therefore the resulting W and b quark from the top-quark decay tend to be produced back to back in azimuth in contrast to the main standard model processes. Secondly, compared to the W/Z

diboson backgrounds, the W boson from the top-quark decay has a higher momentum and its decay products have smaller opening angles. Lastly, the production ratio between top quarks and top antiquarks differs between FCNC processes and SM processes. The FCNC processes are predicted to produce more than three times more top quarks than top anti-quarks, whereas in SM single top-quark production and all other SM backgrounds this ratio is at most two [22].

Experimental limits on the branching fractions of the FCNC top quark decay channels were estab- lished by experiments at the LEP, HERA, Tevatron and LHC accelerators. At present the most stringent upper limits at 95% CL for the coupling constants

and

are

0.12 GeV

[23] (ZEUS, HERA) and

qg)

5.7

10

(ugt) and

qg)

2.7

10

(cgt) [24] (ATLAS, LHC). In the case of t

qZ only upper limits on the branching fractions of the top-quark decay have been determined to be

qZ)

0.07% [25] (CMS, LHC).

This note is organised as follows: Section 2 provides a description of the ATLAS detector and an overview of the data and Monte Carlo (MC) samples used for the simulation of signal and expected background events from SM processes. In Section 3 the event selection is presented. The methods of event classification into signal- and background-like events using a neural network are discussed in Section 4 and sources of systematic uncertainties summarised in Section 5. The observed and expected upper limits on the total cross section, the coupling constants and branching fractions of the investigated FCNC processes are given in Section 6.

2 The ATLAS detector and simulated event samples

The ATLAS detector [26] is composed of an inner tracking system close to the interaction point, sur- rounded by a superconducting solenoid providing a 2 T axial magnetic field, electromagnetic and hadronic calorimeters, and a muon spectrometer

. The electromagnetic calorimeter is a lead liquid-argon sam- pling calorimeter (LAr) with high granularity. An iron-scintillator tile calorimeter provides hadronic energy measurements in the central pseudorapidity range. The endcap and forward regions are instru- mented with LAr calorimeters for both electromagnetic and hadronic energy measurements. The muon spectrometer consists of three large superconducting toroids, a system of trigger chambers, and precision tracking chambers.

This analysis is performed using

s

8 TeV pp-collision data recorded by ATLAS between April 4th and October 24th, 2012. Only the periods in which all the sub-detectors were operational are considered, resulting in a data sample with a total integrated luminosity of 14.2 fb

.

Detector and trigger simulations are performed with the standard simulation of ATLAS within the GEANT4 framework [27, 28]. The same offline reconstruction methods used with data events are ap- plied to the simulated samples. Minimum bias events generated by PYTHIA8 [29] are used to simulate multiple pp interactions in the same and nearby bunch crossings, corresponding to the LHC operation with 50 ns bunch separation and an average of 18 additional pp interactions per bunch crossing.

For the direct production of top quarks via FCNC, ME

[30], a new generator for simulating strong FCNC processes at the NLO approximation is used. This analysis is based on the assumption

1In the right-handed ATLAS coordinate system, the pseudorapidityηis defined asη = −ln[tan(θ/2)], where the polar angleθis measured with respect to the LHC beam-line. The azimuthal angleφis measured with respect to thex-axis, which points towards the centre of the LHC ring. Thez-axis is parallel to the anti-clockwise beam viewed from above. Transverse momentum and energy are defined as pT = psinθand ET = Esinθ, respectively. The angular distance∆Ris defined as

∆R=p

(∆η)²+(∆φ)².

of a purely left-handed top-quark production like in the SM, since a priori the magnitude of any right- handed contribution is unknown. Right-handed top quark production might give rise to di

erent signal kinematics which would need to be handled separately in the training of the neural network. It has been shown that the kinematics of FCNC single top-quark production does not depend on the choice of the coupling constants

and

in the production.

The matrix element generator ALPGEN v2.14 [31] interfaced to the shower generator PYTHIA v6.4 [32] and the Perugia2011C [33] tune is used to generate W

Wb b ¯

Wc¯ c

Wc

and Z

jets events with up to 5 partons. For both the matrix element calculations and the parton shower evolution the CTEQ6l1 [34] parton distribution function (PDF) is used. In order to remove overlaps between the n and n

1 parton samples the ‘MLM’ matching scheme [31] is used. To remove the double counting between the inclusive W

n parton samples and samples with associated heavy quark pair production an overlap removal based on

R matching is applied. The inclusive cross-sections of W/Z- boson production is calculated to next-to-next-to-leading order (NNLO) with FEWZ [35] leading to an uncertainty of

W

jets and Z

jets backgrounds in association with one additional jet, the Berends-Giele scaling [36, 37] uncertainty (W

n

1/W

n) of

Wc production, which is the dominant contribution to W

quark jets, is calculated at NLO using MCFM [38]. The relative uncertainties on the heavy flavour contributions are 50% for Wb b ¯

jets, Wc c ¯

jets and Wc

jets, estimated using a tag counting method in control regions [2].

The POWHEG [39] generator with the CT10 PDF set is used to generate t¯ t, Wt and s-channel single top events. The parton shower and the underlying event is added using PYTHIA v6.4 and the Peru- gia2011C tune. Electroweak t-channel single top-quark events are generated with AcerMC v.3.7 [40]

and PYTHIA v.6.4 using the the Perugia2011C tune. The cross section of the t¯ t process is normalized to 238 pb obtained from approximate next-to-next-to leading order (NNLO) QCD calculations using the HATHOR tool [41]. Theoretical uncertainties include the correlated PDF+

uncertainty, according to the MSTW prescription [42], and scale uncertainty, yielding an 8% relative uncertainty. The SM single top-quark production cross-sections are also calculated to approximate NNLO [43–45]. Production via a t-channel exchange of a virtual W boson has a predicted production cross section of 87 pb [43]. The cross section of the associated production of an on-shell W boson and a top quark (Wt-channel) has a predicted value of 22.3 pb [44] and the s-channel production a predicted cross section of 5.6 pb [45].

The resulting weighted average of the theoretical uncertainties including PDF and scale uncertainties of these three process is 10%. All top-quark processes are simulated with a top-quark mass of 172.5 GeV. Diboson events (WW, WZ and ZZ) are produced using HERWIG [46] and normalised to the NLO cross-section prediction. The cross-section uncertainty for these processes is 5%.

3 Event selection

Based on the expected signature of signal events, we select events with exactly one isolated electron or muon, missing transverse momentum and one jet, which we require to be identified as a b-quark jet. Events are considered only if they are accepted by a single-lepton trigger [47]. Electrons have to be accepted by either one of two single-electron triggers, with thresholds of E

24 GeV plus an isolation criterion, or E

60 GeV without isolation criteria. The single-muon trigger thresholds are either p

24 GeV with, or p

36 GeV without an isolation criterion.

Electron candidates are defined as clusters of cells in the electromagnetic calorimeter associated with

a well-measured track fulfilling several quality requirements [48]. Electron candidates are required to

satisfy p

25 GeV and

2.47, where

is the pseudorapidity of the cluster of energy deposits

in the calorimeter. A veto is placed on electron candidates in the calorimeter barrel-endcap transition re-

gion, 1.37

1.52, that has a limited calorimeter instrumentation. High-p

electrons associated

with the W-boson decay can be mimicked by hadronic jets reconstructed as electrons, electrons from the decay of heavy quarks, and photon conversions. Since electrons from the W-boson decay are typically isolated from hadronic jet activity, these backgrounds can be suppressed by isolation criteria which re- quire minimal calorimeter activity and only low p

tracks in an

cone around the electron candidate.

Isolation cuts optimized to achieve a uniform isolation e

ciency across

and the transverse energy E

with the direction taken from the associated track are imposed on the electron candidates. For the calorimeter isolation a cone size of

R

0.2 is used. The efficiency for electrons from W boson de- cays to pass this isolation requirement is 90%. In addition, the p

of all tracks within a cone of radius

∆

R

0.3 around the electron direction, excluding the track belonging to the electron, is restricted to fall below an E

-dependent threshold. The efficiency for this track-based isolation requirement is 90% for electrons from W boson decays.

Muon candidates are reconstructed by matching track segments or complete tracks in the muon spec- trometer with inner detector tracks. The final candidates are required to have a transverse momentum p

25 GeV and to be in the pseudorapidity region of

2.5. Isolation criteria are applied to reduce background events in which a high-p

muon is produced in the decay of a heavy flavour quark. A mini- isolation variable [49] is defined as the scalar sum of the transverse momenta with p

above 1 GeV from all tracks (except the one matched to the muon) within a cone with radius R

10 GeV/ p

(µ). Muon candidates are accepted that have a mini-isolation to muon p

ratio of less than 0.05.

Jets are reconstructed using the anti-k

algorithm [50] with a width parameter of 0.4, using topological clusters as inputs to the jet finding and are at first required to have p

25 GeV and

2.5. If any jets fall within

R

0.2 of an electron, the closest jet is removed, as in these cases the jet and the electron are very likely to correspond to the same physics object. Remaining electron candidates overlapping with jets within

R

0.4 are rejected subsequently. To reject jets from pile-up events a so-called jet- vertex fraction criterion is applied, where at least 50% of the scalar sum of the p

of tracks within a jet is required to be from tracks compatible with the primary vertex

associated to the hard-scattering collision.

The final selected jet is required to have p

30 GeV and has also to be identified (b-tagged) as a b- quark jet. The tagging algorithm exploits the properties of a b-quark decay in a jet using neural-network techniques and the reconstruction of a secondary vertex [51]. The b- and c-tagging efficiencies, and the mis-tag rate for the taggers, are measured using the same methods as described in [52–54] and updated using the 2012 data set. The b-tagging algorithm has an e

ciency of about 50% for b-jets in simulated t¯ t events, while only 0.1% of light-quark jets and 3.7% of c-quark jets are mis-tagged as b-quark jets.

The magnitude of the missing transverse momentum is defined as E

E

E

is calculated using the calibrated three-dimensional calorimeter energy clusters associated with the jet to- gether with either the calibrated calorimeter energy cluster associated with an electron or the p

of a muon track [55]. Transverse energy deposited in calorimeter cells but not associated with any high-p

object is also included in the E

calculation. Due to the presence of a neutrino in the final state of the signal process, E

30 GeV is required. To further reduce the number of multijet background events, which are characterised by low E

and low transverse W-boson mass

m

(W), the event selection requires m

(W)

50 GeV.

Two kinematical regions are defined in this analysis, both being subject to the same event selection requiring exactly one lepton, missing transverse momentum from the neutrino and exactly one b-tagged jet: the “signal region” using the default b-tagging requirement, and a “control region”, where a less stringent b-tagging requirement with a b-tagging efficiency of 85% is applied and events, where the jet passes the tighter b-tagging requirement are rejected. The control region is defined such that the compo- sition of the resulting sample is dominated by W

jets production and that the same object definitions can

2The primary vertex is defined as the vertex with the largestP

pTof the associated tracks.

3The transverseWboson mass is defined as:mT(W) = q 2h

pT(`)E<sup>miss</sup><sub>T</sub> −~pT(`)·E~^miss_T i

, where~pT(`) denotes the transverse momentum of the lepton andpT(`)=|~pT(`)|

be used as in the signal region in order to check the correct modelling of kinematic variables. Assuming a cross-section of 1 pb for FCNC single top-quark production, about 500 signal events in 14.2 fb

of collision data are expected in the signal region and about 350 signal events in the control region.

For all background processes except the multijet background, the normalisations are estimated by using Monte Carlo simulation scaled to the theoretical cross-section predictions. Multijet events may be selected if a jet is misidentified as an isolated lepton or if the event has a non-prompt lepton that appears to be isolated. Different methods are used to estimate this background in the electron and muon channel. In the electron channel the jet electron model is used through which a shape for the multijet background is obtained by selecting simulated events where a jet is mis-reconstructed as an electron.

The normalization of the template is obtained by a fit to the observed data in the E

distribution. In the muon channel the data driven matrix method is used, where the shape and normalisation of the multijet background is obtained by estimating the number of fake leptons in the signal region using measured efficiencies of real and fake leptons passing different isolation requirements as determined from collision data.

The jet electron model used in the electron channel establishes a shape for the multijet background using events from a Pythia dijet sample which are selected using similar criteria to the signal selection but with a jet used in place of the electron. Each jet has to fulfil the same p

and

requirements as a signal lepton, contain at least four tracks to reduce the contribution from converted photons, and deposit 80–95% of its energy in the electromagnetic calorimeter. To avoid a contamination of the jet-electron sample by real electrons, events are vetoed if they contain one or more leptons that are identified ac- cording to high-e

ciency but low-purity identification criteria. The event is accepted if exactly one

‘jet-electron’ is found. To determine the normalisation of the multijets background template in the elec- tron channel, a binned maximum likelihood fit is performed to the observed data in the E

distribution after applying all selection criteria, leaving out the cut on E

only. Fits are performed separately in two channels for electrons in the endcap (|η|

1.5) and central (|η|

1.5) region of the electromag- netic calorimeter, for both the signal and control regions in the 1-jet bin. The multijet template is fitted together with templates derived from MC simulation for all other background processes (top, W

jets, W

Z

form of additional constrained nuisance parameters. For the purpose of these fits the contributions from W

jets and W

HF

jets, the contributions from t¯ t and single top-quark production, and the contributions from Z

In the muon channel, the matrix method [56] is used to obtain both the shape and normalisation of the multijet background. The method estimates the number of fake events in the signal region based on loose and tight lepton isolation definitions, the latter being a subset of the loose definition. Hence, the loose selection is defined to contain leptons of similar kinematics, but resulting in much higher yields and is, except for the muon isolation requirement, identical to the signal selection. The number of fake leptons passing the tight (signal) isolation requirements can be expressed by using measured efficiencies of real and fake leptons of the loose selection to also pass the tight isolation requirements.

These fake e

ciencies are determined from collision data in a sample of selected muon candidates with high impact parameter significance. The real efficiencies are estimated from data in a high m

(W) mass region (m

(W)

100 GeV). Both efficiencies are then applied to data in the form of event weights.

Based on comparisons of the rates obtained by using alternative methods, i.e. the matrix method in the electron channel and the jet-lepton method in the muon channel, and using an alternative variable, i.e.

m

(W) instead of E

for the binned likelihood fit, we assign a systematic uncertainty of 50% on the estimated multijet fractions.

In order to obtain scale factors for the non-multijet processes in the muon channel, which are not

determined as part of the matrix method, a fit to the E

distribution in the muon channel is performed,

fixing the multijet rate to the value from the matrix method. The scale factors so obtained are only

applied to scale the corresponding backgrounds in the control plots in order to check the modelling of the kinematic distributions and are not used in the later analysis because all rate and cross-section uncertainties are fully accounted for in the final limit evaluation, see Section 6. The corresponding E

distributions after rescaling the different backgrounds and the multijets template to their respective fit results are shown in Figure 1 for both the electron and muon channel.

[GeV]

miss

ET

0 50 100 150

Events / 5 GeV

0 10000 20000 30000 40000

data top W+jets Z+jets multijets MC stat. uncert.

Multijets norm. + = 8 TeV

s

-1, L dt = 14.2 fb

∫

Electron control region

[GeV]

miss

ET

0 50 100 150

Events / 5 GeV

0 2000 4000

6000 data

top W+jets Z+jets multijets MC stat. uncert.

Multijets norm. + = 8 TeV

s

-1, L dt = 14.2 fb

∫

Electron signal region

a) b)

[GeV]

miss

ET

0 50 100 150

Events / 5 GeV

0 5000 10000 15000 20000

data top W+jets Z+jets multijets MC stat. uncert.

Multijets norm. + = 8 TeV

s

-1, L dt = 14.2 fb

∫

Electron control region

[GeV]

miss

ET

0 50 100 150

Events / 5 GeV

0 1000 2000 3000

data top W+jets Z+jets multijets MC stat. uncert.

Multijets norm. + = 8 TeV

s

-1, L dt = 14.2 fb

∫

Electron signal region

c) d)

[GeV]

miss

ET

0 50 100 150

Events / 5 GeV

0 20000 40000 60000 80000

data top W+jets Z+jets multijets MC stat. uncert.

Multijets norm. + = 8 TeV

s

-1, L dt = 14.2 fb

∫

Muon control region

[GeV]

miss

ET

0 50 100 150

Events / 5 GeV

0 5000 10000 15000

data top W+jets Z+jets multijets MC stat. uncert.

Multijets norm. + = 8 TeV

s

-1, L dt = 14.2 fb

∫

Muon signal region

e) f)

Figure 1: Fitted E

distributions in the control region without the cut on E

applied for electrons in the central (a) and forward (c) regions and for muons (e). The multijet rate in the muon channel is estimated using the matrix method. (b,d,f) show the respective distributions in the signal region.

The uncertainty bands denote the MC statistical uncertainty combined with the assigned multijet rate

uncertainty of 50%.

In the control region 596587 events are observed in data compared to a prediction of 560000± 160000 events from the estimates of SM backgrounds. Table 1 summarises the event yield in the control region for each of the background processes considered. In the signal region 112525 events are observed in data while the expectation from SM backgrounds amounts to 105000

31000 events. The corresponding event yields in the signal region are summarized in Table 2. The yields in the signal and control region are calculated using the acceptance from MC samples normalised to their respective theoretical cross sections including the (N)NLO k factors. Each event yield uncertainty in Table 1 and 2 combines the statistical uncertainty, originating from the limited size of the simulation samples, with the uncertainty in the cross section or normalisation.

Table 1: Number of observed data events and expected number of background events for the control region. The event yields reflect the acceptance from MC samples normalised to their respective theoreti- cal cross sections and the data-driven multijet estimates. The uncertainties shown include the statistical uncertainties from the limited sizes of the simulation samples, the cross-section uncertainties and the multijet normalization uncertainty.

electron channel muon channel

Process central forward total total

single-top 3010

150 820

43 3830

190 4960

250

t¯ t 2500

250 665

68 3170

320 3860

390

W

56000

13000 30000

7000 86000

21000 125000

30000 W

HF

jets 90000

50000 30000

17000 120000

60000 160000

90000

Z

Jets 5900

1400 2900

700 8800

2100 10500

2500

multijets 15000

8000 12000

6000 27000

13000 21000

11000 Total Expected 170000

50000 76000

19000 240000

70000 320000

90000

data 168996 76487 245483 351104

Table 2: Number of observed data events and expected number of background events for the signal region. The event yields reflect the acceptance from MC samples normalised to their respective theoreti- cal cross sections and the data-driven multijet estimates. The uncertainties shown include the statistical uncertainties from the limited sizes of the simulation samples, the cross-section uncertainties and the multijet normalization uncertainty.

electron channel muon channel

Process central forward total total

single-top 4190

210 1091

57 5280

270 6870

350

t¯ t 3300

330 870

88 4200

400 5300

500

W

2400

700 1130

350 4000

1000 5100

1300 W

HF

jets 16000

9000 7000

4000 23000

13000 31000

17000

Z

720

170 279

74 990

240 2000

500

multijets 2800

1400 2500

1300 5400

2700 13000

6000 Total Expected 30000

9000 12000

4000 42000

13000 63000

18000

data 31271 12680 43951 68574

Kinematic distributions in the control region of the identified lepton, reconstructed jet, E

and

m

(W) are shown in Figure 2. The distributions are normalised using the scale factors obtained in the

binned maximum likelihood fit in the E

distribution. A very good agreement between the observed

and simulated distributions can be seen.

(l) [GeV]

pT

40 60 80 100 120

Data/Expected 0.8

1 1.2

Events / 5 GeV

0 50 100

103

×

data single-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

η(l)

-2 -1 0 1 2

Data/Expected 0.8

1 1.2

Events / 0.2

0 20000 40000 60000 80000

data single-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

(jet) [GeV]

pT

50 100 150

Data/Expected 0.8

1 1.2

Events / 6 GeV

0 50 100 150 200

103

×

data single-top t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

a) b) c)

(jet) η

-2 -1 0 1 2

Data/Expected 0.8

1 1.2

Events / 0.2

0 20000 40000 60000

80000 _data

single-top t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

[GeV]

miss

ET

50 100 150

Data/Expected 0.8

1 1.2

Events / 6 GeV

0 50 100

103

×

data single-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

[GeV]

WT

m

50 100 150 200

Data/Expected 0.8

1 1.2

Events / 8 GeV

0 50 100

103

×

data single-top t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

d) e) f)

Figure 2: Basic kinematic distributions of the combination of muon and electron channels in the control region. The distributions are normalised using the scale factors obtained in the binned maximum like- lihood fit in the E

distribution. a) transverse momentum and b) pseudorapidity of the lepton, c) the transverse momentum and d) pseudorapidity of the jet, e) the missing transverse momentum and f) trans- verse W-boson mass. The last histogram bin includes overflow events and the hatched bands in the main and the ratio plot indicate the statistical uncertainty from the simulated samples size and the uncertainty on the multijet normalisation.

4 Data analysis

As no single variable provides su

cient discrimination between signal and background events by itself and the separation power is distributed over many correlated variables, multivariate analysis techniques are necessary to separate signal candidates from background candidates. A neural-network classifier [57]

that combines a three-layer feed-forward neural network with a preprocessing of the input variables is

used. In order to improve the performance and to avoid overtraining, Bayesian regularisation is imple-

mented during the training processes. The network infrastructure consists of one input node for each of the 13 input variables plus one bias node, 20 nodes in the hidden layer, and one output node which gives a continuous output in the interval [

1, 1]. The training is performed with a mixture of 50% signal and 50% background events, where the different background processes are weighted according to their number of expected events. We consider all backgrounds apart from the multijet sample in the training process.

All possible discriminating variables such as momenta, relative angles, pseudorapidity, reconstructed particles masses and lepton electric charge are explored, including variables obtained from the recon- structed W boson and the top quark. To reconstruct the four-momentum of the W boson, the neutrino four-momentum is derived from the measured E

, since it cannot be measured directly. Given the lepton four-momentum, the neutrino longitudinal momentum, p

, is calculated by imposing a kinematic constraint on the invariant mass of the W boson, (p

)

(p

p

)

m

(80.4 GeV)

, where p

, p

and p

are the four vectors of the W boson, lepton and neutrino, respectively. In case of a solution where p

is real, the twofold ambiguity is resolved by choosing the smallest

solution, since the W boson is expected to be produced with small pseudorapidity. In about 30% of the events the relation has imaginary solutions. In this case, the E

is rescaled by a factor such that the imaginary part van- ishes. The top-quark candidate is reconstructed by adding the four-momentum of the b-tagged jet to the four-momentum of the reconstructed W boson.

Variables are selected as inputs to the neural network such that for a minimal number of variables the best possible separation between the signal and background processes is achieved. Each variable beforehand is tested for agreement between the background model and observed events in the control region and, taking account of potential signal contributions, also in the signal region. This leads to 13 variables remaining for the network. Table 3 shows a summary of the variables used ordered by their importance. The importance of the variables is estimated using an iterative procedure after their preprocessing. In this step the correlation matrix of the input variables is computed and their significance determined by removing each single variable and calculating the loss of correlation between the network output and the true desired values (target) caused by its removal. The variable causing the smallest loss of correlation is discarded, after which the correlation matrix is computed again and the procedure is repeated on the remaining (n

1)-dimensional correlation matrix. At the end, a list of variables ordered in importance is obtained, where the fractional loss of total correlation to the target caused by its removal is given. Distributions of the three most important variables in the control and signal region are shown in Figure 3. The distributions are normalised using the scale factors obtained in the binned maximum likelihood fit to the E

distribution. Figure 4 shows the corresponding shape distributions for simulation for the signal and three most important background processes. The resulting neural network output distributions for the various processes in the control region are shown in Figure 5 a), while in Figure 5 b) the same distribution in the signal region is shown. Signal-like events have output values close to 1, whereas background-like candidate events accumulate near

1. Overall, good agreement in both the control and signal regions is seen between data and predicted background processes.

5 Systematic uncertainties

In this analysis a variety of systematic uncertainties are considered which affect the normalisation of the individual background processes and the signal acceptance (rate uncertainties) as well as the shape of the neural network output distribution (shape uncertainties).

Object modelling

Systematic uncertainties due to the residual di

erences between data and Monte

Carlo simulation, uncertainties on jets, electron and muon reconstruction after calibration, and uncer-

tainties on corrective scale factors are propagated to the event yields.

The uncertainties on the muon (electron) trigger, reconstruction and selection efficiency scale factors are estimated in measurements of Z

(Z

ee and W

eν) production. To evaluate uncertainties on the lepton momentum scale and resolution, the same processes are used. The jet energy scale (JES) uncertainty depends on the p

and

of the reconstructed jet [58]. It includes uncertainties in the flavour composition of the samples and mis-measurements from close-by jets. Additional contributions to this uncertainty due to the larger pile-up effects in 2012 data are included. An additional jet energy scale uncertainty, depending on the p

of the jet, is applied for b-quark induced jets due to differences between light and gluon jets as opposed to jets containing b-hadrons. The change in the JES is propagated to the missing transverse energy calculation. Smaller uncertainties result from the modelling of the jet energy

(b) [GeV]

pT

50 100 150

Data/Expected 0.8

1 1.2

Events / 6 GeV

0 50 100 150 200

103

×

data single-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

(W) [GeV]

pT

0 50 100 150

Data/Expected 0.8

1 1.2

Events / 8 GeV

0 50 100

103

×

data single-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

| [rad]

t-rest

ν) φ(W,

∆

|

0 1 2 3

Data/Expected 0.8

1 1.2

Events / 0.2 rad

0 20000 40000 60000 80000

data single-top t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

control region

a) b) c)

(b) [GeV]

pT

50 100 150

Data/Expected 0.8

1 1.2

Events / 6 GeV

0 10000 20000

30000 ^datasingle-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

signal region

(W) [GeV]

pT

0 50 100 150

Data/Expected 0.8

1 1.2

Events / 8 GeV

0 5000 10000 15000 20000

data single-top

t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

signal region

| [rad]

t-rest

ν) φ(W,

∆

|

0 1 2 3

Data/Expected 0.8

1 1.2

Events / 0.2 rad

0 5000 10000 15000

data single-top t t W+jets W+HF+jets Z+jets multijets MC stat. uncert.

multijets norm. + = 8 TeV

s -1, L dt = 14.2 fb

∫ ATLAS Preliminary

signal region

d) e) f)

Figure 3: Kinematic distributions of the three most significant variables: transverse momentum of the b-

jet p

(b) a),d), the transverse momentum of the W boson p

(W) b),e), and the difference of the azimuthal

angle between the W boson and the neutrino in the top quark rest frame

c),f) shown in

the control region a)-c) and the signal region d)-f). All processes are normalised using the scale factors

obtained in the binned maximum likelihood fit to the E

distribution. The last histogram bin includes

overflow events and the hatched bands in the main and the ratio plot indicate the statistical uncertainty

from the simulated samples size and the uncertainty on the multijet normalisation.

(b) [GeV]

pT

50 100 150

Event fraction / 6 GeV

0 0.1 0.2 0.3 0.4

FCNC t tW+HF+jets multijets Simulation

Preliminary ATLAS signal region

(W) [GeV]

pT

0 50 100 150

Event fraction / 8 GeV

0 0.1 0.2 0.3

FCNC t tW+HF+jets multijets Simulation

Preliminary ATLAS signal region

a) b)

| [rad]

t-rest

) ν (W, φ

∆

|

0 1 2 3

Event fraction / 0.2 rad

0 0.1 0.2

FCNC t tW+HF+jets multijets Simulation

Preliminary ATLAS signal region

c)

Figure 4: Kinematic distributions of the three most significant variables in the signal region normalised to unit area: transverse momentum of the b-jet p

(b) a), the transverse momentum of the W boson p

(W) b), and the di

erence of the azimuthal angle between the W boson and the neutrino in the top quark rest frame

c). The last histogram bin includes overflow events.

Table 3: Variables used as input to the neural network ordered by their importance, as estimated from the total correlation loss to the target caused by its removal.

loss of total loss of total

Variable correlation (%) Variable correlation (%)

p

(b) 34

W) 6

p

(W) 19

5

∆φ(W, ν) in the top quark rest frame

13

R(`, `νb) 5

Charge of the lepton q(`) 12

4

η(`)

11

4

∆φ(`,

b) 9 E

4

η(`νb)

9

resolution and the missing transverse energy, which accounts for contributions of calorimeter cells not

associated to any jets and soft jets. Since the analyses makes use of b-tagging, the uncertainties on the b-

and c-tagging efficiencies and the mistag rate [52–54] are taken into account.

Events / 0.08

0 10000 20000 30000 40000 50000 60000 70000 80000

data single top

t t W+HF+jets W+jets Z+jets, diboson multijets MC stat. uncert.

multijets norm. + = 8 TeV

s

-1 L dt = 14.2 fb

∫

control region

NN output -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Data/Expected

0.6 0.8 1 1.2 1.4

Events / 0.13

0 5000 10000 15000 20000 25000

data

=20 pb) FCNC (σ single top

t t W+HF+jets W+jets Z+jets, diboson multijets MC stat. uncert.

multijets norm. + = 8 TeV

s

-1 L dt = 14.2 fb

∫

signal region

NN output -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

Data/Expected

0.8 0.9 1 1.1 1.2

a) b)

Figure 5: a) Neural network output distribution in the control region and b) in the signal region. Figure b) shows the signal shape stacked on top of the backgrounds and scaled to 20 pb. All background processes are shown normalised to the result of the binned maximum likelihood fit used to determine the fraction of multijet events. The hatched band indicates the statistical uncertainty from the simulated samples size and the uncertainty on the multijet normalisation.

Monte Carlo generators

Systematic e

ects from the modelling of the signal and background pro- cesses are taken into account by comparing di

erent generator models and varying parameters of the event generation. The MC modelling of the t¯ t process is studied by comparing two NLO generators interfaced to the HERWIG shower generator, namely MC@NLO

HERWIG and POWHEG

HERWIG and a multi-leg generator ALPGEN

HERWIG. E

ects of the parton shower modelling are evaluated by comparing POWHEG samples interfaced to two different shower generators, HERWIG and PYTHIA.

For all top-quark processes the amount of initial (ISR) and final state radiation (FSR) was varied by modifying parameters in samples generated with AcerMC and interfaced to the PYTHIA generator. The range of parameter variations was determined using collision data and is described in more detail in [59].

The W

generator HERWIG. Uncertainties connected with the matching are estimated by varying the matching scale and the functional form of the renormalisation and factorisation scale. Finally, an uncertainty is included due to the statistical precision of the limited size MC samples.

PDF

The systematic uncertainties related to the parton distribution functions are taken into account for all samples using simulated events. The events are reweighted according to each of the PDF uncertainty eigenvectors. The uncertainties are calculated using the formula given in Equation 43 of Ref. [60].

The final PDF uncertainty is calculated as the envelope of the estimated uncertainties for the CT10, MSTW2008nlo [61] and NNPDF2.3 [62] PDF sets, following the PDF4LHC recommendations [63].

Luminosity

The uncertainty on the integrated luminosity is 3.6%. It is derived, following the same methodology as in [64], from a preliminary calibration of the luminosity scale derived from Van der Meer scans taken throughout 2012.

The total uncertainty is dominated by the statistical uncertainty from the simulated samples size and

those on the cross-sections and the multijet normalization, see Table 2. Other important contributions

arise from uncertainties on the jet energy scale, jet energy resolution, the ISR/FSR variations, PDF

sets and uncertainties on b-tagging efficiencies and mis-tagging rates. Table 4 summarizes the resulting rate uncertainties for each considered process, while Figure 6 shows examples of shape uncertainties corresponding to jet energy scale variations for the three most important processes.

Table 4: Uncertainties on the number of expected candidate events on each signal and background source for the systematic variations explained in Section 5.

Systematic Signal W

jets W

HF

jets

Jet energy scale

Jet energy resolution

4%

20%

3%

b-tagging efficiency

c-tagging e

ciency

Mistag rate

1%

26%

1%

Muon momentum scale

Muon identification

Electron energy scale

1%

1%

1%

Electron identification

Missing transverse momentum

PDF

3%

4%

8%

W

—

Cross section — 24% 55%

Systematic t¯ t single-top Z

jets

Jet energy scale

Jet energy resolution

b-tagging efficiency

c-tagging efficiency

Mistag rate

Muon momentum scale

1%

1%

1%

Muon identification

Electron energy scale

Electron identification

1%

1%

1%

Missing transverse momentum

PDF

ISR/FSR

3%

5% —

Cross section 8% 10% 24%

6 Results

A Bayesian statistical analysis [65] using a binned likelihood method applied on the full neural net- work output distribution is performed to measure or set an upper limit on the FCNC single top-quark production cross-section.

Systematic uncertainties and their correlations among processes are included with a direct sampling

approach. For each source of uncertainty a Gaussian prior with the width of the corresponding uncer-

tainty for a given process and in case of shape uncertainties for each bin is used. The prior density for

the signal cross section is assumed to be flat. The posterior density function (pdf) is obtained by creating

a large number of samples, each drawn from the prior distributions. A separate likelihood distribution is

Event fraction

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18

0.20 Jet energy scale uncertainty: FCNC Simulation

Preliminary ATLAS +1 σ Nominal -1 σ

NN output

-1 -0.5 0 0.5 1

rel. diff.

-0.5 0.0 0.5

Event fraction

0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16

0.18 Jet energy scale uncertainty: W+HF Simulation

Preliminary ATLAS +1 σ Nominal -1 σ

NN output

-1 -0.5 0 0.5 1

rel. diff.

-0.5 0.0 0.5

Event fraction

0.02 0.04 0.06 0.08 0.10

0.12 Jet energy scale uncertainty: Single-top Simulation

Preliminary ATLAS +1 σ Nominal -1 σ

NN output

-1 -0.5 0 0.5 1

rel. diff.

-0.5 0.0 0.5

a) b) c)

Figure 6: Shape differences caused by the jet energy scale uncertainty for a) the signal, b) W

heavy flavour and c) SM single top-quark templates.

obtained for each sample, and the final pdf is averaged over all the individual likelihoods. This signal pdf gives the probability of the signal hypothesis, as a function of the signal cross-section, to be true. Since no significant rate of FCNC single top-quark production is observed, an upper limit is set by integrat- ing the pdf. To estimate the a priori sensitivity, a pseudo data set corresponding to the prediction from simulations (Asimov dataset) is used and treated in the same way as the observed dataset.

The posterior probability function using the Asimov dataset and including all systematic uncertainties is shown in Figure 7 a) together with the expected upper limit which is calculated by integrating the posterior probability function. The resulting expected upper limit on the anomalous FCNC single top- quark production cross section is 2.2 pb. The posterior probability function for the observed data sample, including all systematic uncertainties, is shown in Figure 7 b) together with the observed upper limit.

The observed 95% CL upper limits on the anomalous FCNC single top-quark production cross section including all uncertainties is 2.5 pb. To visualise the observed upper limit in the neural network output distribution, Figure 8 shows the FCNC signal process scaled to 2.5 pb stacked on top of all background processes normalised to the mean values of their posterior rate distributions as obtained from the limit calculation. Using the NLO predictions for the FCNC single top-quark production cross-section [21, 66]

and assuming

Wb)

1, the measured upper limit on the production cross-section is converted into limits on the coupling constants

and

. Assuming

0 one finds

5.1

10

TeV

and assuming

0 one finds

1.1

10

TeV

. Figure 7 c) shows the distribution of the upper limit for combinations considered in [21]. Limits on the coupling constants can be converted into limits on the branching fractions using

qg)

c

Λ

2

, where c is calculated at NLO [67]. Upper limits on the branching fractions

ug)

3.1

10

assuming

cg)

0, and

cg)

1.6

10

assuming

ug)

0 are derived, and presented in Figure 7 d).

7 Conclusion

A search for single top-quark production via flavour changing neutral currents (FCNCs) in strong inter- actions was presented. A data sample collected with the ATLAS detector in 2012 at a centre-of-mass energy of

s

8 TeV and corresponding to an integrated luminosity of

14.2 fb

has been used.

Events with an isolated electron or muon, missing transverse momentum and a b-quark jet were selected,