Search for resonant W Z production in the fully leptonic ﬁnal state in proton–proton collisions at √ s = 13 TeV with the ATLAS detector

A search for a heavy resonance decaying into W Z in the fully leptonic channel (electrons and muons) is performed. It is based on proton–proton collision data collected by the ATLAS experiment at the Large Hadron Collider at a centre-of-mass energy of 13 TeV, corresponding to an integrated luminosity of 36.1 fb − 1 . No signiﬁcant excess is observed over the Standard Model predictions and limits are set on the production cross section times branching ratio of a heavy vector particle produced either in quark– antiquark fusion or through vector-boson fusion. Constraints are also obtained on the mass and couplings of a singly charged Higgs boson, in the Georgi–Machacek model, produced through vector-boson fusion. an the CC BY (http://creativecommons.org/licenses/by/4.0/). Funded by SCOAP 3 .

categories separately, neglecting possible signal leakage between them.
Early results from the Tevatron [22,23] have put limits on the mass of a W boson of an extended gauge model [2] in the W Z channel between 180 GeV and 690 GeV. The present analysis extends the search for resonant W Z production beyond that in Run  The various decay channels generally differ in sensitivity in different mass regions. The fully leptonic channel, in spite of a lower branching ratio, is expected to be particularly sensitive to lowmass resonances as it has lower backgrounds. A recent search [39] by the CMS Collaboration for a charged Higgs boson produced by vector-boson fusion and decaying into W Z in the fully leptonic mode, using 15.2 fb −1 of data collected at √ s = 13 TeV, has yielded limits on the coupling parameter of the GM model, as a function of mass. Limits on the GM model have also been set, based on analyses of same-charge W W production by CMS [40] and opposite-charge W W production by ATLAS [41], using data at √ s = 13 TeV with an integrated luminosity of 36.1 fb −1 .

ATLAS detector
The ATLAS detector at the LHC has a cylindrical geometry with a near 4π coverage in solid angle. 1 The inner detector (ID), consisting of silicon pixel, silicon microstrip and transition radiation detectors, is surrounded by a thin superconducting solenoid providing a 2 T axial magnetic field. It allows precise reconstruction of tracks from charged particles and measurement of their momenta up to a pseudorapidity of |η| = 2.5. High-granularity lead/liquidargon (LAr) sampling electromagnetic and steel/scintillator-tile hadron calorimeters, at larger radius, provide energy measurements in the central pseudorapidity range |η| < 1.7. In the endcap and forward regions, LAr calorimeters for both the EM and hadronic energy measurements extend the region of angular acceptance up to |η| = 4.9. Outside the calorimeters, the muon spectrometer incorporates multiple layers of trigger and tracking chambers in a magnetic field produced by a system of superconducting toroid magnets, enabling an independent precise measurement of muon track momenta for |η| < 2.7. The ATLAS trigger system consists of a hardware-based level-1 trigger followed by a software-based high-level trigger [42].

Data and Monte Carlo samples
The data used in this analysis were collected during 2015 and 2016 with the ATLAS detector in pp collisions at a centre-of-mass energy of 13 TeV at the LHC. The minimum bunch crossing interval is 25 ns, with a mean number of 23 additional interactions per bunch crossing. The events are required to have passed combinations of single-electron or single-muon triggers. The transverse 1 ATLAS uses a right-handed coordinate system with its origin at the nominal interaction point (IP) in the centre of the detector and the z-axis along the beam pipe. The x-axis points from the IP to the centre of the LHC ring, and the y-axis points upwards. Cylindrical coordinates (r, φ) are used in the transverse plane, φ being the azimuthal angle around the z-axis. The pseudorapidity is defined in terms of the polar angle θ as η = − ln tan(θ/2). Angular distance is measured in units of momentum threshold of the leptons in 2015 is 24 GeV for electrons and 20 GeV for muons satisfying a loose isolation requirement based only on ID track information. Due to the higher instantaneous luminosity in 2016 the trigger threshold was increased to 26 GeV for both electrons and muons and tighter isolation requirements were applied. Additional electron and muon triggers that do not include any isolation requirements with transverse momentum thresholds of p T = 60 GeV and 50 GeV, and a single-electron trigger requiring p T > 120 GeV with less restrictive electron identification criteria are used to increase the selection efficiency which reaches close to 100%. Events are accepted only if quality criteria for detector and data conditions are satisfied. With these conditions, the available datasets correspond to an integrated luminosity of 36.1 fb −1 . Samples of simulated data were produced by Monte Carlo (MC) generators with the detector response obtained from the Geant4 toolkit [43,44]. For some samples, the calorimeter response is obtained from a fast parameterized simulation [45], instead of Geant4. Additional simulated inelastic pp collisions, generated with Pythia 8.186 [46] with the A2 set of tuned parameters [47] and the MSTW2008LO [48] parton distribution function (PDF), were overlaid in order to model both the in-and out-of-time effects from additional pp collisions (pile-up) in the same and neighbouring bunch crossings. The mean number of pile-up events in the MC samples was set to reflect the conditions in the data.
For the HVT interpretation, W → W Z samples were generated. Two benchmark models, provided in Ref. [15], are used. In Model A, weakly coupled vector resonances arise from an extension of the SM gauge group [49] with an additional SU(2) symmetry group and the branching fractions to fermions and gauge bosons are comparable. In Model B, the heavy vector triplet is produced in a strongly coupled scenario, as in a Composite Higgs model [50] and fermionic couplings are suppressed. The parameter g V was set to 1 for Model A and to 3 for Model B. For both models, the parameter c F is assumed to be the same for all types of fermions. Simulated signal samples for the HVT benchmark Model A were generated for masses of vector resonances ranging from 250 GeV to 3 TeV with MadGraph_aMC@NLO 2.2.2 [51], using the model file provided by the authors in Ref. [52] with the NNPDF23LO [53] PDF set. They are hadronized with Pythia 8.186. For interpretation in terms of Model B, the Model A cross sections are simply scaled. This is justified since the width remains well below the experimental resolution and the angular distributions are the same for both models.
For the VBF production channel, HVT samples were generated with g V = 1 for masses ranging from 250 GeV to 2 TeV. The coupling parameter c H was set to 1 and all other couplings of the heavy triplet, including c F , were set to 0 in order to maximize the VBF contribution. A dijet invariant mass of at least 150 GeV was required during event generation.
For the GM signal samples, pp → H ± 5 jj → W ± Z jj were produced with MadGraph_aMC@NLO 2.2.2 for the mass range 200 to 900 GeV in the H 5 -plane defined in [54], compatible with present limits [20,55], using GMCALC [56] and with sin θ H = 0.5. They were produced at leading order, but normalized to next-to-leading order according to Ref. [11], where the cross sections and widths, which scale as sin 2 θ H , are also given. For these samples, a minimum p T of 15 GeV (10 GeV) for the jets (leptons) was required during event generation and the pseudorapidity must be in the range |η| < 5 for jets and |η| < 2.7 for leptons.
The background sources in this analysis include processes with two or more electroweak gauge bosons, namely V V and V V V as well as processes with top quarks, such as tt, tt V , single top and t Z , and processes with gauge bosons produced in association with jets or photons ( Z + j and Z γ ). MC simulation is used to estimate the contribution from background processes with three or more prompt leptons while data-driven techniques are used for the case of background processes with at least one misidentified or nonprompt lepton. Simulated events are used for cross checks and to assess the systematic uncertainties in these backgrounds.
The dominant W Z SM background process of order (α 2 α 2 s ) involving colour-exchange diagrams, here referred to as QCD W Z, was modelled using Sherpa 2.2.2 [57] at next-to-leading order (NLO), and includes hard-scattering, parton shower, hadronization and the underlying events. Up to three additional partons generated at tree level were merged with the parton shower. In order to estimate an uncertainty due to the parton shower modelling, two alternative W Z samples were produced using Powheg-Box v2 [58] interfaced with Pythia 8.186 and Herwig++ [59], respectively.
A sample of the purely electroweak process W Z jj → ν jj (labelled W Z jj) with a matrix-element b-quark veto (at zero order in α s ) was generated separately with Sherpa 2.2.2. Contributions from W Z jb → ν bj (labelled W Zbj) are included in the t Z sample described below. To estimate an uncertainty due to the parton shower modelling an alternative Madgraph+Pythia 8 sample was produced. This Madgraph sample includes b-quarks in the initial state and was split to provide a sample without (with) a b-quark in the final state to model the W Z jj (t Z + W Zbj) background.
Samples of qq → Z Z → 4 or qq → Z Z → νν were generated by Powheg-Box v2 at NLO, interfaced to Pythia 8.186 and normalized to NNLO by K -factors evaluated in Ref. [60]. The gg → Z Z and tribosons were generated with Sherpa 2.1.1. The tt V and t Z processes were generated at LO using Madgraph_aMC@NLO, interfaced with Pythia 8.186 (tt V ) and Pythia 6.428 (t Z ). The tt V samples were normalized to NLO predictions [11].
Finally samples of SM backgrounds with at least one misidentified or non-prompt lepton, including Z γ , W γ , Drell-Yan Z → , W → ν as well as top-pair and single-top were generated to assist in the fake/non-prompt lepton background estimate.
Events with Z γ and W γ in the final state were generated with Sherpa 2.1.1. Drell-Yan Z → , W → ν as well as top-pair and single-top production channels were generated with Powheg-Box v2 and hadronized with Pythia. To avoid double counting the Z γ events, Z events produced by the Drell-Yan process with a photon from final-state radiation with p T > 10 GeV were removed. The parton shower for processes with top quarks was modelled with Pythia 6.428. Madgraph_aMC@NLO and Pythia 8.186 were used for background processes involving a pair of top quarks accompanied by a W boson or by a pair of charged leptons. The Z and single-top cross sections were normalized to NNLO by K -factors evaluated in Ref. [60,61]. SM backgrounds with Higgs bosons (H, tt H, V H) contribute less than 0.1% of the total background because of the low cross section and the requirement of a well reconstructed Z boson decaying leptonically. These backgrounds are neglected.

Reconstructed objects
Events are required to have at least one primary vertex with at least two associated tracks, each with transverse momentum p T > 0.4 GeV. If there is more than one vertex reconstructed in the event, the one with the largest track p 2 T is chosen as the hard-scatter primary vertex and is subsequently used for the reconstruction of electrons, muons, jets and missing transverse momentum.
Electron candidates are reconstructed from energy deposits in the EM calorimeter which are matched to a well-reconstructed ID track originating from the primary vertex. The electron identification is based on a likelihood evaluated from a multivariate discriminant. They are categorized as satisfying the medium or the tight reconstruction quality requirements, as defined in Ref. [62].
Only electrons with transverse energy E T > 25 GeV in the pseudorapidity range |η| < 2.47 are considered in this analysis. The candidate electrons are required to pass an isolation condition: an upper value of the scalar sum of the transverse momentum of the tracks with p T > 0.4 GeV in a cone of size R = min(0.2, 10 GeV/E T ) around the electron, excluding the track of the electron itself, is chosen such that the efficiency is constant at 99% for electrons in Z → ee events. For tight electrons, an isolation requirement is imposed, based on calorimeter as well as track variables, which varies as a function of transverse energy and yields an efficiency between 95% and 99% for electrons with p T in the range 25-60 GeV. For a pair of electrons sharing the same ID-track, the electron with higher cluster E T is kept.
Muons are reconstructed by combining tracks from the inner detector with tracks from the muon spectrometer. They are required to satisfy medium or tight quality requirements, as defined in Ref. [63]. Only muons with p T > 25 GeV and |η| < 2.7 are considered in this analysis. Isolation requirements are also applied to all muons, based on the ratio p varcone Electron and muon candidates are required to originate from the primary vertex. Thus, the significance of the track's transverse impact parameter calculated relative to the beam line, |d 0 /σ d 0 |, must be less than three for muons and less than five for electrons, and the longitudinal impact parameter, z 0 (the difference between the value of z of the point on the track at which d 0 is defined and the longitudinal position of the primary vertex), is required to satisfy |z 0 · sin(θ )| < 0.5 mm.
Jets are reconstructed from clusters of energy deposition in the calorimeter [64] using the anti-k t algorithm [65] with a radius parameter R = 0.4. Events with jets arising from detector noise or other non-collision sources are discarded [66]. This search considers jets with p T > 30 GeV in the range |η| < 4.5. Furthermore, to mitigate the pile-up contamination, a jet vertex tagger [67], based on information about tracks associated with the primary vertex and pile-up vertices, is applied to jets with p T < 60 GeV and |η| < 2.4. The selected working point provides at least 92% efficiency. The energy of each jet is calibrated and corrected for detector effects using a combination of simulated events and in situ methods in 13 TeV data [68].
As lepton and jet candidates can be reconstructed from the same detector information, a procedure to resolve overlap ambiguities is applied. If an electron and a muon share the same ID track, the muon is selected. Reconstructed jets which overlap with electrons or muons in a cone of size R = 0.2 are removed.

Event selection
Preselection criteria are first applied to all the event samples. The presence of three prompt leptons is required, two of which will be associated with the Z boson, and are required to satisfy the medium quality requirement (Section 4). The Z boson candidate is reconstructed from the two leptons of same flavour and opposite charge, whose invariant mass is closest to the on-shell mass m Z , and in the range |m − m Z | < 20 GeV. The third lepton, associated with the W boson decay, is required to satisfy the tight quality criteria to enhance the background rejection. To ensure that the trigger efficiency is well determined, at least one of the candidate leptons is required to have p T > 27 GeV.
To suppress background processes with at least four prompt leptons, events with a fourth lepton candidate satisfying looser selection criteria are rejected. For this looser selection, the requirement on the minimum p T of the leptons is lowered to p T > 7 GeV and medium identification requirements are used for both the electrons and muons.
Since there is a neutrino in the signal events, E miss T > 25 GeV is also required. The third lepton and the missing transverse momentum are assumed to result from the W boson decay. The longitudinal momentum p ν z of the neutrino is calculated by requiring that the invariant mass of the lepton-neutrino system be equal to the W mass. The solution results in a quadratic equation which leads to two possible solutions. If they are real, the one with the smaller |p ν z | is chosen since it was found to provide a better agreement with the truth. Otherwise, the real part is chosen. The invariant mass, m W Z , of the W Z resonance candidate is then reconstructed using the chosen solution for p ν z along with the four-momenta of the three charged leptons.
The selected events are then separated into two categories targeting different production mechanisms: VBF and qq. The VBF category contains events with two or more jets with p T > 30 GeV which fail the b-tagging requirements described in Section 4. The dijet pair defined by the two highest-p T jets in the event must also have large η separation (| η jj | > 3.5) and an invariant m jj above 500 GeV. If more than two jets are found in an event, the two highest-p T jets are considered. By imposing a b-jet veto, backgrounds containing one or more top quarks, including tt, tt + W /Z , and t Z are suppressed.
The net acceptance times efficiency (A× ) of the selection, relative to signal events generated for H ± 5 and HVT models in the The remaining events are assigned to the qq category signal region. For this category, the W and Z bosons from a resonance produced in the s-channel with m W Z larger than 250 GeV are expected to have transverse momenta close to 50% of its mass. The requirements p W T /m W Z > 0.35 and p Z T /m W Z > 0.35 enhance the sensitivity to the signal. The overall selection efficiency relative to generated event increases from about 15% to 25% for resonances masses ranging from 500 GeV to 3 TeV as illustrated in Fig. 2.
Decays of W and Z bosons into all flavours of leptons are included at event generation. The A× values decrease for resonance masses above approximately 2 TeV due to the collinearity of electrons from the Z → ee decays which spoils the isolation.

Background estimation
The dominant background in the resonance search is the SM production of W Z. Its normalization and shape are estimated from MC and validated in dedicated validation regions by comparing the data and MC distributions. Events in the validation regions are selected in exactly the same way as those in their corresponding signal categories except for the following requirements. The VBF W Z validation region is defined by inverting the requirements on the dijet variables: 100 < m jj < 500 GeV and | η jj | < 3.5. The W Z qq validation region requires the events to have p Z T /m W Z < 0.35 or p W T /m W Z < 0.35. These validation regions are dominated by the W Z contribution, with a purity higher than 80%. For the benchmark models with parameters given in Section 3, the signal contamination in the qq (VBF) validation region is below 5% (1%). The reconstructed m W Z mass in the validation regions is shown in Fig. 3, where good agreement of data with the background prediction is observed.
Events from Z +jets, Z γ , W γ , tt, single top or W W where jets or photons were misidentified as leptons (here called fake/nonprompt leptons), can also satisfy the selection criteria. The distribution shapes and number of fake/non-prompt lepton events are estimated for both the qq and VBF categories by a data-driven method using a global matrix which exploits differences in object characteristics between real and fake/non-prompt leptons on a statistical basis. Details of the method, here referred to as "Matrix Method", can be found in Ref. [73].
Other backgrounds include tt V , Z Z , t Z , W Zbj and triple boson production. They are estimated by Monte Carlo simulation (Section 3). The t Z , W Zbj and V V V backgrounds are added as a single contribution, here called t Z+V V V .

Systematic uncertainties
Systematic uncertainties result from the theoretical modelling of backgrounds and from object and event reconstruction.
The uncertainty in the normalization of the Sherpa samples of SM W Z background is evaluated by taking into account the variations obtained with different PDF sets [74]. The nominal set NNPDF30nloas0118 is compared with other samples generated with the CT14nnlo and MMHT2014nlo68cl PDF sets and the uncertainty is evaluated from the maximum differences. It is estimated to be below 6% in all mass bins for both the VBF and qq categories. The uncertainty associated with the choice of renormalization and factorization scales, μ R and μ F , is taken as the maximum downward and upward variation when the scales are varied independently by factors of 1/2 and 2. While these uncertainties can in principle affect the shape of the m W Z distribution, in practice the shape differences do not have a strong impact on the sensitivity of the search. The uncertainties are therefore treated as normalization uncertainties, taken to be 20% and 40% respectively. Shape systematic uncertainties associated with showering and hadronization of the QCD W Z are evaluated by comparing the Powheg-Box v2 samples interfaced with Pythia 8.186 and Herwig. For the electroweak W Z process the Sherpa 2.2.2 and Madgraph+Pythia 8 predictions are compared. This uncertainty band ranges from 10% to 30% for both categories.
The uncertainties assigned to the cross sections of the other background sources consist of a contribution from PDF uncertainties and from QCD scale uncertainties. They are estimated to be 10% for Z Z , 13% for tt V , 20% for V V V and 15% for t Z .
The theoretical uncertainties in the cross section and acceptance of the simulated signal samples are evaluated in a similar way to the background. The PDF errors are taken from the NNPDF LO PDF error set, and the NNPDF set is also compared with the CTEQ6L1 and MSTW2008lo68cl sets. The different predictions from these PDF sets are taken as an extra contribution to the overall uncertainty. For both the qq and VBF categories, the uncertainties are typically below 5%. This procedure was followed for each mass point and a generator-level event selection was chosen to closely mimic the one used in the reconstruction-level analysis. Scale and PDF uncertainties are not correlated between signal and background.
An uncertainty due to the reconstruction efficiency, momentum scales and resolution of electrons and muons is evaluated by varying correction factors applied to the MC samples [63,75] within appropriate limits.
The jet energy scale and resolution uncertainties [66] are also taken into account as they affect the shape and normalization of the background distributions. The uncertainty due to b-tagging [76] is also included.
Missing transverse momentum is calculated using the preselected leptons, jets and other reconstructed objects. The uncertainties in the reconstruction of those objects are then used to evaluate An uncertainty in the prediction of the fake/non-prompt background is also taken into account as it affects the shape and normalization of the background distributions. The total uncertainty is about 20% (27%) for the qq (VBF) category. It is slightly larger for the VBF category because of the higher statistical uncertainty derived from the Matrix Method (Section 6).
The uncertainty in the integrated luminosity is 2.1%. It is derived, following a methodology similar to the one detailed in Ref.
[77], from a calibration of the luminosity scale using x-y beam-separation scans performed in August 2015 and May 2016.

Results
The W Z invariant mass distribution, m W Z , obtained as the sum of all four lepton-flavor permutations, is used as the discriminating variable, with bin widths comparable to the expected resolution of a narrow resonant signal. A binned likelihood function, constructed from the Poisson probability of the sum, in each bin, of the contributions of the background and of a hypothetical signal of strength μ relative to the benchmark model, is used to set limits on the presence of a signal. The fit is performed in the signal region for the qq and VBF categories separately. The systematic uncertainties described above (Section 7) enter as nuisance parameters with Gaussian or log-normal prior distributions, in convolution with the nominal background distribution.
The effects of systematic uncertainties are studied for hypothesized signals using the signal-strength parameter μ. The list of leading sources of uncertainty in the 95% confidence level (CL) upper limit on the μ value is given in Table 1 Table 2 for the qq and VBF categories. The fit constrains the SM W Z background estimate to the observed data, which reduces the Table 2 Expected and observed yields in the qq and VBF signal regions. Yields and uncertainties are evaluated after a background-only fit to the data in the qq or VBF signal regions after applying all selection criteria. The uncertainty in the total background estimate is smaller than the sum in quadrature of the individual background contributions due to anti-correlations between the estimates of different background sources. total background uncertainty, pulling the modelling uncertainties by less than one standard deviation from their pre-fit values. None of the nuisance parameters are significantly pulled or constrained relative to their pre-fit values in the background-only fit. Fig. 4 shows the post-fit m W Z distribution for the qq and VBF categories. The largest difference between the observed data and the SM background prediction is in the VBF category. A local excess of events at a resonance mass of around 450 GeV can be seen in Fig. 4(b). The local significances for signals of H ± 5 and of a heavy vector W are 2.9 and 3.1 standard deviations, respectively. The respective global significances calculated using the Look Elsewhere method as in Ref.
[78] and evaluated up to a mass of 900 GeV, are 1.6 and 1.9 standard deviations. In the qq category the largest difference between the observed data and the SM background prediction is located around a mass of 700 GeV with a local significance of 1.2 standard deviations.
Upper limits are set on the product of the production cross section of new resonances and their decay branching ratio into W Z. Exclusion intervals are derived using the CL s method [79] in the asymptotic approximation [80]. For masses higher than 900 (700) GeV in qq (VBF) category, the small number of expected events makes the asymptotic approximation imprecise and the limits are calculated using pseudo-experiments. The limit set on the signal strength μ is then translated into a limit on the signal cross section times branching ratio, σ × B(W → W Z), using the theoretical cross section and branching ratio for the given signal model.   2460 GeV for Model B. For resonance masses above 2 TeV the exclusion limits become worse due to the acceptance losses at high mass. For the VBF process, the limit on σ ×B(W → W Z) is shown in Fig. 6.
Observed and expected exclusion limits at 95% CL on σ × B(H ± 5 → W ± Z ) and on the mixing parameter sin θ H of the GM Model are shown in Fig. 7 as a function of m H ±

5
. The intrinsic width of the scalar resonance, for sin θ H = 0.5, is narrower than the detector resolution in the mass region explored. The shaded regions show the parameter space for which the H ± 5 width exceeds 5% and 10% of m H ± 5 .