Search for pair-produced heavy quarks decaying to Wq in the two-lepton channel at sqrt(s) = 7 TeV with the ATLAS detector

A search is presented for heavy-quark pair production (QQbar) under the decay hypothesis QQbar to WqWq with q = u,d,c,s,b. The search is performed with 1.04 fb^-1 of integrated luminosity from pp collisions at sqrt(s) = 7 TeV collected by the ATLAS detector at the CERN LHC. Dilepton final states are selected, requiring large missing transverse momentum and at least two jets. Mass reconstruction of heavy quark candidates is performed by assuming that the W boson decay products are nearly collinear. The data are in agreement with Standard Model expectations; a heavy quark with mass less than 350 GeV is excluded at 95% confidence level.

Search for pair-produced heavy quarks decaying to W q in the two-lepton channel at √ s = 7 TeV with the ATLAS detector

The ATLAS Collaboration
A search is presented for heavy-quark pair production (QQ) under the decay hypothesis QQ → W + qW −q with q = u, d, c, s, b. The search is performed with 1.04 fb −1 of integrated luminosity from pp collisions at √ s = 7 TeV collected by the ATLAS detector at the CERN LHC. Dilepton final states are selected, requiring large missing transverse momentum and at least two jets. Mass reconstruction of heavy quark candidates is performed by assuming that the W boson decay products are nearly collinear. The data are in agreement with Standard Model expectations; a heavy quark with mass less than 350 GeV is excluded at 95% confidence level. The addition of one or more heavy quarks is a natural extension to the Standard Model, capable of providing an additional source of CP violation in B s decays and accommodating a heavy Higgs boson [1,2]. Searches for heavy quarks with the Collider Detector at Fermilab (CDF) constrain the mass of heavy quarks (Q) that decay as Q → W q, where q = u, d, c, s or b, to be m Q > 340 GeV [3]. More specific searches have also constrained the mass of up-type heavy quarks (u 4 ) that decay as u 4 → W b to be m u4 > 358 GeV [3] and the mass of down-type heavy quarks (d 4 ) decaying via d 4 → W t to be m d4 > 372 GeV [4]. The D0 experiment at Fermilab has set a mass limit of m Q > 285 GeV [5] on heavy quarks that decay as Q → W q. All previous searches used the "lepton+jets" channel, where only one of the produced W bosons decays hadronically.
In this Article, a search is presented for pair production of a heavy quark (QQ) in data corresponding to an integrated luminosity of 1.04 fb −1 from proton-proton collisions at √ s = 7 TeV collected by the ATLAS experiment. The heavy quark is assumed to decay via Q → W q where q = u, d, c, s or b. The search is performed in the dilepton channel, where both W bosons decay leptonically. Fourth-generation up-type quarks (u 4 ) decaying through weak charged currents are used as a benchmark.
The dileptonic final state arises in a way similar to that of pair-produced top quarks: QQ → + νq −νq , where is either e or µ. Leptonically-decaying intermediary τ leptons are able to contribute to this final state if additional neutrinos are considered. The signature is at least two jets, two oppositely charged leptons, and missing transverse momentum (E miss T ) from undetected neutrinos. Top quark pair production is the dominant source of background. To distinguish a potential heavy-quark signal, heavy-quark mass reconstruction is performed by taking advantage of the larger boost each W boson receives from the decay of a heavy quark compared to the decay of a top quark. This large boost makes each undetected neutrino approximately collinear with an observed charged lepton.
The following sections contain descriptions of the ATLAS detector (Section I), simulated samples (Sec-tion II), object reconstruction (Section III), baseline event selection (Section IV), data-driven background estimates (Section V), mass reconstruction strategy (Section VI), validation of background modeling (Section VII), final event selection (Section VIII), binned maximum-likelihood ratio fit using the reconstructed mass (Section IX and Section X), and final results (Section XI).

I. THE ATLAS DETECTOR
The ATLAS detector [6] is a multi-purpose particle detector with precision trackers, calorimeters and muon spectrometers. The momenta of charged particles with pseudorapidity |η| < 2.5 1 are measured by the inner detector (ID), which is a combination of silicon pixels, silicon microstrips and a straw-tube tracker. The ID operates in a uniform 2 T axial magnetic field produced by a superconducting solenoid. The pixel detector measurements enable precise determination of production vertices.
Electromagnetic (EM) calorimetry for electron and photon reconstruction is provided by a high-granularity, three layer liquid argon (LAr) sampling calorimeter with lead absorbers in the region |η| < 3.2. A presampler is used to correct for energy lost by electrons and photons in material in front of the calorimeter for |η| < 1.8. Hadronic calorimetry for |η| < 1.7 is provided by a scintillating tile sampling calorimeter with steel absorbers, and for 1.5 < |η| < 3.2 it is provided by a LAr sampling calorimeter with copper-plate absorbers.
1 ATLAS uses a right-handed coordinate system with its origin at the nominal interaction point (IP) in the center of the detector and the z-axis along the beam pipe. The x-axis points from the IP to the center of the Large Hadron Collider (LHC) ring and the y-axis points upward. Cylindrical coordinates (r, φ) are used in the transverse plane, φ being the azimuthal angle around the beam pipe. The pseudorapidity is defined in terms of the polar angle θ as η = − ln tan(θ/2).
Muons are detected with a multi-system muon spectrometer (MS). Precision measurements in the η coordinate are provided by monitored drift tubes for |η| < 2.7. These are supplemented by cathode-strip chambers measuring both the η and azimuth (φ) coordinates for 2.0 < |η| < 2.7. Fast measurements required for initiating trigger logic are provided by resistive-plate chambers for |η| < 1.05 and then by thin-gap chambers for 1.05 < |η| < 2.4. The muon detectors operate in a nonuniform toroidal magnetic field generated by a superconducting air-core magnet system.
The ATLAS detector uses a three-level trigger system to select events for offline analysis. For this search, events are required to have at least one lepton satisfying trigger requirements. Electron trigger candidates must have transverse energy E T > 20 GeV, must satisfy shower-shape requirements [7] and must have an ID track matched to the EM shower. Muon trigger candidates must have transverse momentum p T > 18 GeV and matching tracks in the ID and MS.

II. SIMULATED SIGNAL AND BACKGROUND SAMPLES
Simulated samples are used to evaluate the contributions from the QQ signal (assuming an up-type heavy quark) and most background processes. Unless otherwise noted, all events are showered and hadronized with herwig v6.5 [8,9], using jimmy [10] for the underlying event model. After event generation, all samples are processed with the geant4-based [11] simulation of the ATLAS detector [12] and subject to the same reconstruction algorithms as the data.
The CERN LHC instantaneous luminosity varied during data-taking from about 2 × 10 32 cm −2 s −1 to 1 × 10 33 cm −2 s −1 [13,14]. At maximum luminosity an average of five proton-proton (pp) interactions were superimposed in each bunch crossing. This pileup background produces additional activity in the detector, affecting variables such as jet reconstruction and isolation energies. Monte Carlo (MC) events simulate the pileup background by adding minimum bias events on top of the hard scatter. The MC events are later reweighted such that the simulated instantaneous luminosity distribution matches that in data.

A. Heavy-quark pair production
Production and decay of heavy-quark pairs (QQ) is modeled with the leading-order (LO) generator pythia 6.421 [15] using MRST 2007 LO* [16] parton distribution functions (PDFs). The production cross-section is calculated using hathor [17] with approximate nextto-next-to-leading-order (NNLO) QCD calculations with CTEQ6.6 PDFs [18] for several heavy-quark masses (m Q ). In addition, scale uncertainties are evaluated in the range m Q /2 to 2×m Q and PDF uncertainties are calculated from the CTEQ6.6 error eigenvectors. The crosssections and uncertainties for each heavy-quark mass considered in this analysis are shown in Table I. The background due to tt production is modeled using the next-to-leading-order (NLO) generator mc@nlo v3.41 [19] with an assumed top quark mass of 172.5 GeV and the NLO PDF set CTEQ6.6. The cross-section for tt production is normalized to the value obtained from an approximate NNLO calculation [17].

C. Z boson, diboson and single-top quark production
The background from Z/γ * boson production in association with jets is modeled with the LO generator alpgen v2.13 [20]. The LO PDF set CTEQ6.1 [18] is used to generate Z/γ * +jets events with dilepton invariant mass m > 10 GeV. For W W , W Z and ZZ production, events are generated with the LO generator herwig v6.5 and the LO PDF set CTEQ6.1. For the small background from single-top production, mc@nlo is used with the NLO PDF set CTEQ6.6, invoking the diagram removal scheme [21,22] to remove overlaps between the singletop and tt final states. The cross-sections for Z/γ * +jets samples are determined using NNLO inclusive calculations from fewz [23,24] and from a data-driven technique where possible, while the cross-sections for diboson samples are determined using NLO calculations with mc@nlo. The cross-sections for single-top samples are normalized to an approximate NNLO prediction [25,26].

III. OBJECT SELECTION
Electrons are found by a calorimeter-seeded reconstruction algorithm and must have a track that matches an energy deposit in the calorimeter. They are required to satisfy E cluster / cosh(η track ) > 25 GeV, where E cluster is the energy deposited in the calorimeter cluster and η track is the pseudorapidity of the matching track. Electrons are required to be in a pseudorapidity range |η cluster | < 2.47, excluding the transition region 1.37 < |η cluster | < 1.52 between the EM calorimeter barrel and endcap. They must also satisfy a calorimeter isolation I cal < 3.5 GeV requirement in ∆R < 0.2, where ∆R = (∆η) 2 + (∆φ) 2 . Calorimeter isolation is defined as the energy reconstructed within a cone of a certain radius around the lepton that is not associated with that lepton, and it is represented by I cal . The calorimeter shower shape is required to closely resemble what is expected for electrons [7].
Jets are reconstructed from topological clusters of energy deposits in the calorimeter [27] using the antik t algorithm with distance parameter R = 0.4 [28,29]. These jets are calibrated to the hadronic energy scale using a correction factor obtained from simulation which depends on p T and η [30] They are required to satisfy p T > 25 GeV and |η| < 2.5. Jets that fall within ∆R < 0.2 of accepted electrons are rejected.
Muons are found by requiring that a track reconstructed in the MS has a matching track in the ID. A loose cosmic ray rejection is applied by removing all muon pairs that are back-to-back azimuthally (∆φ(µ, µ ) > 3.1) and whose transverse impact parameter with respect to the beam line is greater than 0.5 mm. Muon candidates must satisfy p T > 20 GeV and |η| < 2.5. The muon must be isolated, satisfying calorimeter isolation I cal < 4 GeV in ∆R < 0.3 and tracking isolation I trk < 4 GeV in ∆R < 0.3. Tracking isolation is defined as the sum of track momenta within a cone of a certain radius around the lepton vertex, and it is represented by I trk . The muon must also not fall within ∆R < 0.4 of any jet with p T > 20 GeV.
The missing transverse momentum (E miss T ) is constructed from the vector sum of calorimeter topological cluster energies projected onto the transverse plane [31].
Deposits not associated to a jet are calibrated at the EM energy scale. Deposits associated with selected jets contribute at the corrected hadronic energy scale. The contribution from muon energy deposits is included in E miss T after a correction for the muon contribution to calorimeter energy deposits.

IV. BASELINE EVENT SELECTION
The QQ pair decay yields two charged leptons, two jets and missing transverse momentum from the undetected neutrinos. An initial dilepton selection is applied to validate the modeling of the Z/γ * boson production background as well as the identification of leptons, reconstruction of jets and measurement of missing transverse momentum.
The initial dilepton selection [32,33] first requires that an event contains a high-quality reconstructed primary vertex. The event must also have exactly two leptons (e or µ) with opposite charges, at least one of which must be associated with the object that triggered the event. The two leptons must not share a track in the ID. At this stage, the data sample is dominated by Z/γ * → decays (the Drell-Yan process), although the contribution from tt production is evident at large jet multiplicity, as can be seen in Figures 1 and 2. These figures show good agreement between the data and background expectation.
To reduce the background from Z/γ * → decays, the baseline selection requires: • All events must have at least two jets, each with p T > 25 GeV and |η| < 2.5; • Same-flavor events (ee and µµ) must satisfy a missing transverse momentum requirement, E miss T > 60 GeV; • The dilepton invariant mass of same-flavor events (ee and µµ) must be greater than 15 GeV and must fall outside a window around the Z boson mass, defined as 81 GeV < m < 101 GeV; • In different-flavor events (eµ), H T , defined as the scalar sum of E T from every lepton and jet passing the object selection criteria, must exceed 130 GeV 2 .

A. Drell-Yan events
The total number of Drell-Yan ee and µµ events remaining after the baseline selection has been applied is estimated with a data-driven technique that extrapolates from a control region (CR) [34]. Events in the CR have dilepton invariant mass in the range 81 GeV -101 GeV with at least two jets and E miss T > 30 GeV. The number of data events in the control region, Data(CR), and MC Z/γ * +jets events in the control region, MC DY (CR), are used to scale the prediction of Z/γ * +jets events in the    Samples are stacked in the same order as they are presented in the legend, from left to right; the first entry in the legend is at the bottom of the stack. The data-driven Drell-Yan estimate defined in Section V A has not been applied here.    signal region, MC DY . Non-Z/γ * background processes in the control region, MC other (CR), are subtracted from the data using MC predictions. The estimated number of Z/γ * +jets events in the signal region, N DY , in the ee and µµ channels is calculated with Equation 1:

B. Fake lepton events
A small fraction of the background consists of events in which a jet or a non-prompt lepton is misidentified as a prompt lepton from W boson decay. Prompt leptons and misidentified non-prompt leptons are referred to as real and fake leptons, respectively. Fake muons are predominantly produced from semi-leptonic b or c quark decays in which the muon passes the isolation requirements despite being produced in association with a jet. There are three principal mechanisms for producing fake electrons: heavy-flavor decay, light flavor jets with a leading π 0 overlapping with a reconstructed track from a charged particle, and asymmetric conversion of photons into e + e − . The largest source of events with fake leptons is W boson production with associated jets, including lepton plus jets decays of top quark pairs. A matrix method [33] is used to estimate the fraction of the sample that comes from fake lepton events. A looser lepton selection is defined, and the number of observed dilepton events with two tight leptons (N TT ), one loose and one tight lepton (N TL , N LT 3 ) or two loose leptons (N LL ) is counted. Tight leptons pass the selection criteria defined in Section III. Loose electrons need to pass the same selections as the electrons defined in Section III except for looser shower shape and calorimeter isolation requirements [7]. Loose muons only need to satisfy p T > 20 GeV, |η| < 2.5 and the muon-jet overlap requirements defined in Section III.
The probabilities for real and fake leptons that pass the loose identification criteria to also pass the tight criteria are defined as r and f , respectively. These two probabilities are measured separately for = e and = µ. Using r and f , linear expressions are obtained for the observed yields as a function of the number of events with zero, one and two real leptons together with two, one and zero fake leptons (N FF , N RF and N FR , N RR ; in N RF the real lepton has greater p T than the fake lepton, and vice versa for N FR ): 3 The leptons are ordered by p T such that the leading lepton in N TL is tight and the leading lepton in N LT is loose.
where M is a 4 × 4 matrix containing terms proportional to r and f . The matrix is inverted in order to extract the real and fake content of the observed dilepton event sample. The method explicitly accounts for the presence of events with two fake leptons. The probability (r ) for a real loose lepton to pass the tight criteria is measured in Z → events in data with a tag-and-probe method. The probability for a fake loose electron to satisfy the tight requirements (f e ) is measured by requiring exactly one loose electron in an event with E miss T < 10 GeV. The probability for a fake loose muon to satisfy the tight requirements (f µ ) is measured in a control region obtained by requiring exactly one loose muon with |∆φ(µ, E miss T )| < 0.5. The baseline selection requirements from Section IV are not applied when checking these control regions.

VI. MASS RECONSTRUCTION
After the baseline selection has been applied, mass reconstruction of heavy quark candidates is performed in order to discriminate the heavy-quark decays from the dominant tt background. Direct reconstruction is not possible, as two neutrinos escape the detector. However, a unique feature of the heavy quark is the large momentum of the daughter W boson, which makes its decay products approximately collinear in the detector as seen in Figure 3.
Both neutrino momentum vectors are reconstructed by assuming that the neutrinos are the sole contributors to E miss T and that they are approximately collinear with the leptons. The optimal values of each |∆η(ν, )| and each |∆φ(ν, )| are fit by minimizing the mass difference between the two reconstructed heavy quarks using minuit [35]. The fitted direction of each neutrino is constrained to be within ∆R < 2.5 of the direction taken by the neutrino's leptonic partner from the W boson decay, and all jet combinations are considered during each step of the mass difference minimization. A solution of the minimization procedure is penalized if the scalar sum of neutrino momenta exceeds the scalar sum of lepton momenta by at least 30%. The square of the difference between each reconstructed W boson mass and 80.4 GeV is added to the square of the heavy-quark mass difference in the minimized function; the preferred solutions produce W bosons with reconstructed masses that are close to the W boson mass.
The two reconstructed mass values tend to be more correlated for signal than background, as shown in  Correlation between reconstructed masses for QQ pairs produced with mQ = 350 GeV and for background samples.
The fitting method selects solutions with correlated mass values, however this correlation is smaller for background events.
Events within |m Collinear 1 − m Collinear 2 | < 25 GeV are kept; this region is found between the two lines in the figure.
The expected background yields and number of observed events after the baseline selection are given in Table II. Distributions of H T and m Collinear are shown in Figure 6.      .

VII. BACKGROUND VALIDATION
Event samples with the baseline selection and low H T , low lepton p T , low jet p T or low E miss T are examined to validate the modeling of the background (Figure 7). These conditions cause the distributions to be depleted of signal. In each case, the data is described well by the background model, within uncertainties.

VIII. FINAL EVENT SELECTION
The baseline selection provides excellent discrimination against Z/γ * production and other backgrounds, but additional selection requirements are necessary to suppress the dominant tt background.
A triangular selection in H T + E miss T versus m Collinear , H T + E miss T > X − 0.4 × m Collinear with X dependent on the assumed signal mass, is applied. Mass-dependent requirements on E miss T and leading jet p T are imposed as well. These selection requirements are optimized in MC simulation by seeking a point of maximum significance, S/ √ S + B, while simultaneously varying all of the selection requirement parameters. Distributions of H T + E miss T versus m Collinear for background and signal are shown in Figure 8. Tables III and IV list the full set of optimized selection requirements at each mass point. Table V lists the expected backgrounds, expected signal and observed data for each mass point after this final selection. Figure 9 shows the distributions in m Collinear for two signal samples after the final selection.

IX. SYSTEMATIC UNCERTAINTIES
The major sources of systematic uncertainty are due to modeling of the signal and most sources of background.       Table III. The uncertainties due to simulation of the lepton trigger, reconstruction and selection efficiencies are assessed using leptons from Z → ee and Z → µµ in data [33]. MC events are corrected for differences in data and simulation. The statistical and systematic uncertainties in    Table I. Samples are stacked in the same order as they are presented in the legend, from left to right; the first entry in the legend is at the bottom of the stack.
these corrections are included in the uncertainties on the acceptance values. Uncertainties in the modeling of the lepton energy scale and resolution are studied using reconstructed Z boson mass distributions. The jet energy scale (JES) and its uncertainty are derived by combining information from test-beam data, LHC collision data and simulation [30]. The JES uncertainty varies as a function of jet p T and η and also accounts for the presence of nearby jets and event pileup. There is additional uncertainty associated with jets originating from b quarks in simulation. Smaller uncertainties are associated with the jet energy resolution and jet finding efficiency. Uncertainties related to the E miss T arise due to uncertainties associated with low momenta jets, event pileup, and calorimeter energy not associated with reconstructed leptons or jets [31]. There is also some uncertainty in estimating the effect of a readout problem affecting a subset of the LAr calorimeter channels in a part of the data set.
The use of simulated samples to calculate the signal and background acceptances gives rise to systematic uncertainties from the generator choice, the amount of initial and final state radiation (ISR/FSR), and from the PDF choice. The uncertainty due to the choice of generator for tt events is evaluated by comparing the predictions of mc@nlo with those of powheg [36] interfaced to either herwig or pythia. The uncertainty due to ISR/FSR is evaluated by studies using the AcerMC [37] generator interfaced to pythia, varying the parameters controlling ISR and FSR in a range consistent with experimental data [38]. For Z/γ * +jets events, the generator uncertainty is evaluated by comparing the predictions of alpgen with the PDF set CTEQ6.1 and sherpa [39] with the PDF set CTEQ6.6. Finally, the uncertainty in the PDFs used to generate signal, tt and single-top events is evaluated using the procedure adopted in a measurement of the tt cross-section [33].
The integrated luminosity measurement carries a 3.7% uncertainty [13,14]. Each sample also has an uncertainty associated with its theoretical cross-section or with its data-driven rate. For tt [17], single-top [25,26] and Z/γ * → τ τ [40] the rate uncertainty is estimated from theoretical calculations. Z/γ * → ee, Z/γ * → µµ and fake lepton event rate uncertainties are evaluated with the data-driven fitting described in Section V. The crosssection uncertainty for each signal point comes from hathor NNLO calculations and can be found in Table I. The background normalization uncertainties are listed in  Table VI. The effects of the systematic uncertainties on the overall background yield are summarized in Table VII for the cuts used for m Q = 350 GeV.

X. RESULTS
A binned maximum-likelihood ratio technique is used to fit distributions of m Collinear to the observed data in order to measure the most likely QQ production crosssection, σ(pp → QQ). In the fitting technique, all events with m Collinear > 760 GeV are considered to belong in the same bin. The rates of background and signal events are fitted simultaneously. In addition, the signal and background shapes are smoothly deformed in generated samples of pseudo-data by random variations consistent with the systematic uncertainties shown in Table VI and  Table VII; these fluctuations are not constrained by the data.
Statistical interpretation of the fitted cross-section σ is made using the CLs technique [41,42]. The observed data are in good agreement with the expected background, and the fitted QQ cross-sections at each mass point are consistent with the absence of a heavy-quark signal. Figure 10 shows the observed and expected limits on the production cross-section σ(pp → QQ).
The upper limit on the production cross-section is converted into a lower limit on m Q by finding the point of intersection with the theoretical prediction as a function of m Q . This analysis finds a lower limit of m Q > 350 GeV at 95% confidence level (C.L.) whereas a limit of m Q > 335 GeV was expected. This limit assumes that the branching ratio (BR) of Q → W q is 100%. FIG. 10. Observed and median expected 95% C.L. crosssection upper limits on QQ production, compared to the theoretical prediction. The limit was calculated for five signal masses, and a linear interpolation has been made between mass points.

XI. CONCLUSIONS
This Article presents a search for pair production of heavy quarks decaying to W q in the dilepton channel at the CERN LHC. This search allows q = u, d, c, s or b in the final state. The analyzed data correspond to an integrated luminosity of 1.04 fb −1 collected by the ATLAS detector in pp collisions at √ s = 7 TeV. To enhance the sensitivity to a new quark, mass reconstruction is per-formed by exploiting the boost received by the heavyquark decay products. The reconstructed mass is used for binned maximum-likelihood ratio fitting.
The data are found to be in agreement with the expectation from the Standard Model. A lower limit is set on the mass m Q > 350 GeV at 95% confidence level. This limit assumes BR(Q → W q) = 100% and is applicable to many exotic models [43,44], including up-type fourth-generation quarks u 4 , down-type fourth generation quarks d 4 , and quarks with exotic charges (such as −4/3) decaying to light quarks.