Search for a Heavy Neutral Particle Decaying to $e\mu$, $e\tau$, or $\mu\tau$ in $pp$ Collisions at $\sqrt{s}=8$ TeV with the ATLAS Detector

This Letter presents a search for a heavy neutral particle decaying into an opposite-sign different-flavor dilepton pair, $e^\pm \mu^\mp$, $e^\pm \tau^\mp$, or $\mu^\pm \tau^\mp$ using 20.3 fb$^{-1}$ of $pp$ collision data at $\sqrt{s} = 8$ TeV collected by the ATLAS detector at the LHC. The numbers of observed candidate events are compatible with the Standard Model expectations. Limits are set on the cross section of new phenomena in two scenarios: the production of $\tilde{\nu}_{\tau}$ in $R$-parity-violating supersymmetric models and the production of a lepton-flavor-violating $Z'$ vector boson.

In RPV SUSY, the lagrangian terms allowing LFV can be expressed as 1 2 λ i jk L i L jēk + λ i jk L i Q jdk [1], where L and Q are the S U(2) doublet superfields of leptons and quarks; e and d are the S U(2) singlet superfields of leptons and down-like quarks; λ and λ are Yukawa couplings; and the indices i, j and k denote fermion generations. A τ sneutrino (ν τ ) may be produced in pp collisions by dd annihilation and subsequently decay to eµ, eτ, or µτ. Although onlyν τ is considered here in order to compare with previous searches performed at the Tevatron, the results of our analysis apply to any sneutrino flavor.
The Sequential Standard Model (SSM), where the Z boson is often assumed to have the same quark and lepton couplings as the Standard Model (SM) Z boson, can be extended to include LFV couplings for the Z . The Z → eµ, eτ, or µτ couplings (Q 12 , Q 13 , or Q 23 ) [12] are typically expressed as fractions of the SSM Z → + − ( = e, µ, τ) coupling.
The CDF [13], D0 [14], and ATLAS [15] collaborations have searched for aν τ in LFV final states and placed limits for variousν τ mass hypotheses. Both the CDF [16] and ATLAS [17] collaborations have placed limits on Q 12 as a function of the Z mass.
This Letter describes a search for a neutral heavy particle (ν τ or Z ) decaying into e ± µ ∓ (eµ), e ± τ ∓ had (eτ), or µ ± τ ∓ had (µτ) using pp collision data collected at √ s = 8 TeV, where τ had is a τ lepton that decays into hadrons. The ATLAS detector is described in detail elsewhere [18]. Events are selected with a three-level trigger system that requires one or two leptons (e or µ) with high transverse momentum (p T ). The dataset has a total integrated luminosity of 20.3 ± 0.6 fb −1 , where the uncertainty is derived following the same methodology as that detailed in Ref. [19].
Electrons are required to have p T > 25 GeV, |η| < 1.37 or 1.52 < |η| < 2.47 [20], and satisfy the "tight" selection in Ref. [21], which was modified in 2012 to reduce the impact of additional inelastic pp interactions, termed pileup. Muon candidates must have p T > 25 GeV, |η| < 2.4 and be reconstructed in both the inner tracker detector and the muon spectrometer. The muon momenta measured by the inner detector and muon spectrometer must match within five standard deviations of their combined uncertainty. Good quality reconstruction and p T resolution at high momentum are ensured by requiring a minimum number of associated hits on the inner detector track [22] and in each of the three muon spectrometer stations.
Candidate events must contain at least one primary interaction vertex reconstructed with more than three associated tracks with p T > 400 MeV. If there is more than one such vertex, the one with the highest sum of p 2 T of associated tracks is chosen. The longitudinal impact parameter is required to be smaller than 2 mm for candidate electrons and smaller than 1 mm for candidate muons. It is further required that the transverse impact parameter is less than six times its resolution for candidate electrons, and that the transverse impact parameter is smaller than 0.2 mm for candidate muons. A calorimeter isolation criterion E ∆R<0. Hadronic decays of τ leptons are characterized by one or three charged tracks associated to a narrow energy cluster in the calorimeters [23]. This search uses τ had candidates with only one charged track due to reduced identification and reconstruction efficiency for high-p T τ decays with three charged tracks. Boosted-decision-tree multivariate discriminators, based on tracking and calorimeter information, are used to reject jets and electrons misidentified as τ had . The τ had candidates must have |η| < 2.47 and E T > 25 GeV. Candidates with |η| < 0.03 are removed to exclude a region with increased misidentification of electrons due to reduced coverage by the inner detector and calorimeters.
Jets are reconstructed from clusters of energy in the calorimeter using the anti-k t algorithm [24] with radius parameter R = 0.4. Jet energies are calibrated using Monte Carlo (MC) simulation and a combination of several in situ calibrations [25]. The calculation of the missing transverse momentum vector p miss T (with magnitude E miss T ) is based on the vector sum of the calibrated p T of reconstructed jets, electrons, and muons, as well as calorimeter energy clusters not associated with reconstructed objects [26].
Candidate signal events are required to have exactly two leptons, of opposite charge and of different flavor, satisfying the above lepton selection criteria. The two leptons are required to be back-to-back in the azimuthal plane with |∆φ | > 2.7, where ∆φ is the φ difference between the two leptons. Due to the presence of the undetected neutrino in the τ decay, the E T of the τ had candidate is required to be less than the E T (p T ) of the electron (muon) for the eτ had (µτ had ) channel.
A collinear neutrino approximation is used to reconstruct the dilepton invariant mass (m ) in the eτ had and µτ had channels. In the hadronic decay of a τ lepton from a heavy resonance, the neutrino and the resultant jet are nearly collinear. The four-vector of the neutrino is reconstructed from the p miss T and η of the τ had jet. Four-vectors of the electron or muon, τ had candidate and neutrino are then used to calculate m . For eτ had and µτ had signal events, the above technique significantly improves the mass resolution and search sensitivity.
Events with m < 200 GeV form a validation region to verify the background modeling, and events with m > 200 GeV are used as the search region.
The SM processes that produce ± ∓ final states can be divided into two categories: processes that produce two prompt leptons such as Z/γ * → ττ, tt, single-top Wt channel, diboson production, and processes where one or more photons or jets are misidentified as leptons, predominantly W/Z + γ, W/Z+jets, and multijet events. The decay of a τ to an electron or a muon is considered as prompt production. For the eτ had (µτ had ) channel, additional background can originate from the Z/γ * → ee (µµ) process if one lepton is misidentified as a τ had candidate. The contributions of these processes are even larger with respect to the Z/γ * → ττ background, since the final states e or µ are usually harder than those from leptonic τ decay.
Contributions from processes in the prompt two-lepton category, as well as photon-related and Z/γ * → ee (µµ) backgrounds, are estimated using MC simulation [27]. The detector response model is based on the geant4 program [28]. Lepton reconstruction and identification efficiencies, energy scales, and resolutions in the MC simulation are corrected to the corresponding values measured in the data. Pileup is included to match distributions observed in the data. Top quark production is generated with mc@nlo v4.06 [29] for tt and single-top, the Drell-Yan process (Z/γ * → ) is generated with alpgen v2.14 [30], and diboson processes are generated with herwig v6.520.2 [31]. Samples of Wγ and Zγ events are generated with sherpa v1.04 [32]. These generated samples are normalized to the most accurate available cross-section calculations. For the dominant backgrounds, the Drell-Yan processes are corrected to nextto-next-to-leading order (NNLO) [33], and tt is corrected to NNLO, including soft-gluon resummation to next-to-next-to-leading-logarithm order [34].
Since it is difficult to model misidentification of jets as leptons, particularly at high p T , the W+jets and multijet backgrounds are determined from control regions in the data. The W+jets background is determined in a control region selected with the same criteria as used for the signal selection except requiring E miss T > 30 GeV (to enhance the W contribution) and requiring that the electron or muon p T be less than 150 GeV (to eliminate potential signal). Simulation studies indicate that there is negligible multijet background in this control region. For the eτ had and µτ had channels, the number of events in the control region is corrected for the other SM background sources using MC samples. For the eµ channel, the number of W+jets events in the control region is too small to yield a statistically meaningful measurement. Instead, the control region is enlarged by removing the isolation criterion on one lepton, and the W+jets contribution is estimated using the lepton E ∆R<0.2 T /E T distribution to fit the data with the MC predictions for other SM processes (dominant at low values of the isolation variable) and W+jets (dominant at large values). For the W+jets background in all three channels, the extrapolation factor from the control region to the signal region and the shape of the m distribution are taken from the W+jets MC sample.
The contribution from multijet production is estimated from a control region with the same selection as the signal region except that the leptons are required to have the same electric charge, under the assumption that the probability of misidentifying a jet as a lepton is independent of the charge. The number of events in this same-sign control region is corrected for contributions from backgrounds with prompt leptons and from W+jets backgrounds using the procedure described above. The assumption of charge independence of the jet misidentification rates is tested in a multijet-enriched region with two nonisolated leptons. After subtraction of other SM backgrounds, the ratio of opposite-sign to same-sign events is found to be 1.07 ± 0.06 (stat.) ± 0.02 (syst.).
The background estimates are verified in validation regions defined with the same selection criteria as for the signal regions but with 1.0 < |∆φ | < 2.7. In these validation regions, simulation studies show the backgrounds have compositions similar to the signal regions, and the predictions agree with the data within 20% over the entire m range. An uncertainty of 20% is hence placed on the total background estimate that is used in the final results.
MC signal events are generated with herwig v6.520.2 forν τ and pythia v8.165 [35] for Z . Samples are produced withν τ and Z masses ranging from 0.5 to 3 TeV. Signal cross sections are calculated to next-to-leading order forν τ , and NNLO for Z . The theoretical uncertainties are taken from an envelope of cross-section predictions using different parton distribution function (PDF) sets and factorization and renormalization scales [36,37].
The selection efficiency, including τ decay branching ratio if τ is involved, for mν τ = 2 TeV are 42%, 14%, and 10% in the eµ, eτ and µτ channels, respectively. The corresponding numbers for a Z boson with m Z = 2 TeV are 37%, 11%, and 9%. The systematic uncertainties on the signal efficiency vary from 3% to 6% depending on the resonance mass and decay mode. The primary contributions are due to the number of MC events, and the uncertainties related to the muon and τ p T scales.
The observed and expected event yields in both the validation and search m regions for all three final states are in good agreement, as summarized in Table 1. The m distributions (Fig. 1) show no significant excess above the SM expectation in any of the three modes. The dominant contributions to the uncertainty bands in Fig. 1 are due to the number of MC events, the MC cross-section uncertainties, the E miss T scale and resolution, and the uncertainty in the shape of the m distribution for W+jets.
Upper limits are placed on the production cross section times branching ratio [σ(pp →ν τ /Z )×BR(ν τ /Z → )]. For eachν τ or Z mass, m, the search region is defined to be m ± 3σ , where σ is the standard deviation of the simulated signal m distribution. The relative width of the signal m distribution ranges from 3% to 17% for different mass points, channels and models. To reduce the statistical error, if the upper side of the search region is greater than 1 TeV, all events above 1 TeV are used. To further reduce the effect of fluctuations in the high-mass region due to low MC event counts, the number of background events in each mass window is estimated using a double exponential fit to the total background m distribution. The fit uncertainty is taken into account in the limit-setting procedure, including a contribution from varying the fit function range. A frequentist technique [38] is used to set the expected and observed upper limits as a function of mν τ and m Z . The likelihood of observing the number of events in data as a function of the expected number of signal and background events is constructed from a Poisson distribution for each m bin. Systematic uncertainties are taken into account with Gaussian-distributed nuisance parameters. A 95% confidence level (CL) limit is then determined. The expected exclusion limits are determined, using simulated pseudoex-periments containing only background processes, as the median of the 95% CL limit distributions for each set of pseudoexperiments at each value of mν τ or m Z , including systematic uncertainties. The ensemble of limits is also used to assess the 1σ and 2σ uncertainty envelopes of the expected limits. Figure 2 shows the observed and expected cross section times branching ratio limits as a function of mν τ (m Z ), together with the ±1σ and ±2σ uncertainty bands. For aν τ mass of 1 TeV, the observed limits on the production cross section times branching ratio are 0.5 fb, 2.7 fb, and 9.1 fb for the eµ, eτ and µτ channels, respectively. The corresponding limits for a Z boson mass of 1 TeV are 1.0 fb, 4.0 fb and 9.9 fb for the eµ, eτ and µτ channels, respectively. Theoretical predictions of cross section times branching ratio [33,39] are also shown, assuming λ 311 = 0.11 and λ i3k = 0.07 for theν τ and Q i j = 1 for the Z , consistent with benchmark couplings used in previous searches.
For these benchmark couplings, the lower limits on theν τ mass are 2.0 TeV, 1.7 TeV, and 1.7 TeV for the eµ, eτ and µτ channels, respectively. The corresponding lower limits on the Z mass are 2.5 TeV, 2.2 TeV and 2.2 TeV for the eµ, eτ and µτ channels, respectively. The observed lower mass limits are a factor of three to four higher than the best limits from the Tevatron [13,14] and also more stringent than the previous limits from ATLAS [15] for the same couplings.
In summary, a search has been performed for a heavy particle decaying to eµ, eτ had , or µτ had final states using 20.3 fb −1 of pp collision data at √ s = 8 TeV recorded by the ATLAS detector at the LHC. The data are found to be consistent with SM predictions. Limits are placed on the cross section times branching ratio for an RPV SUSYν τ and a LFV Z boson. These results considerably extend previous constraints from the Tevatron and LHC experiments.  Figure 2: The 95% CL limits on cross section times branching ratio as a function ofν τ mass (top plots) and Z mass (bottom plots) for eµ (left), eτ (middle), and µτ (right). Theory curves are for the arbitrary choice of couplings λ 311 = 0.11 and λ i3k = 0.07 forν τ and Q = 1 for Z . The gray band around the theory curve represents the theoretical uncertainty from the PDFs and factorization and renormalization scales.
We thank CERN for the very successful operation of the LHC, as well as the support staff from our institutions without whom ATLAS could not be operated efficiently.
The ATLAS Collaboration f Also at Department of Physics, University of Fribourg, Fribourg, Switzerland