Search for High-Mass Resonances Decaying to τν in pp Collisions at ﬃﬃ s p = 13 TeV with the ATLAS Detector

A search for high-mass resonances decaying to τν using proton-proton collisions at ﬃﬃﬃ s p ¼ 13 TeV produced by the Large Hadron Collider is presented. Only τ -lepton decays with hadrons in the final state are considered. The data were recorded with the ATLAS detector and correspond to an integrated luminosity of 36 . 1 fb − 1 . No statistically significant excess above the standard model expectation is observed; model-independent upper limits are set on the visible τν production cross section. Heavy W 0 bosons with masses less than 3.7 TeV in the sequential standard model and masses less than 2.2 – 3.8 TeV depending on the coupling in the nonuniversal G ð 221 Þ model are excluded at the 95% credibility level.

M. Aaboud et al. * (ATLAS Collaboration) (Received 22 January 2018;published 20 April 2018) A search for high-mass resonances decaying to τν using proton-proton collisions at ffiffi ffi s p ¼ 13 TeV produced by the Large Hadron Collider is presented. Only τ-lepton decays with hadrons in the final state are considered. The data were recorded with the ATLAS detector and correspond to an integrated luminosity of 36.1 fb −1 . No statistically significant excess above the standard model expectation is observed; model-independent upper limits are set on the visible τν production cross section. Heavy W 0 bosons with masses less than 3.7 TeV in the sequential standard model and masses less than 2.2-3.8 TeV depending on the coupling in the nonuniversal Gð221Þ model are excluded at the 95% credibility level. DOI: 10.1103/PhysRevLett.120.161802 Heavy charged gauge bosons (W 0 ) appear frequently in theories of physics beyond the standard model (SM). They are often assumed to obey lepton universality, such as in the sequential standard model (SSM) [1], which predicts a W 0 SSM boson with couplings identical to those of the SM W boson. However, this assumption is not required. In particular, models in which the W 0 boson couples preferentially to third-generation fermions may be linked to the high mass of the top quark [2][3][4][5] or to recent indications of lepton flavor universality violation in B meson decays [6,7]. An example is the nonuniversal Gð221Þ model (NU) [4,5], which exhibits a SUð2Þ l × SUð2Þ h × Uð1Þ gauge symmetry, where SUð2Þ l couples to light fermions (first two generations), SUð2Þ h couples to heavy fermions (third generation), and ϕ NU is the mixing angle between them. The model predicts W 0 NU and Z 0 NU bosons which are approximately degenerate in mass and couple only to left-handed fermions. At leading order and neglecting sign, the W 0 NU couplings to heavy (light) fermions are scaled by cot ϕ NU (tan ϕ NU ) relative to those of W 0 SSM . Thus cot ϕ NU > 1 corresponds to enhanced couplings to tau leptons while cot ϕ NU ¼ 1 yields W 0 NU couplings identical to those of W 0 SSM . For Z 0 NU , the coupling to heavy (light) fermions is given by g cot ϕ NU (g tan ϕ NU ), where g is the SM weak coupling constant. At high values of cot ϕ NU , the branching fraction of W 0 NU to a tau lepton (τ) and a neutrino (ν) approaches 26%.
In this Letter, a search for high-mass resonances (0.5-5 TeV) decaying to τν using proton-proton (pp) collisions at a center-of-mass energy of ffiffi ffi s p ¼ 13 TeV produced by the Large Hadron Collider (LHC) is presented. The data were recorded with the ATLAS detector and correspond to an integrated luminosity of 36.1 fb −1 . Only τ decays with hadrons in the final state are considered; these account for 65% of the total τ branching fraction. A counting experiment is performed from events that pass a high transverse-mass threshold, optimized separately for each of the signal mass hypotheses. A direct search for high-mass resonances decaying to τν has been performed by the CMS Collaboration using 19.7 fb −1 of integrated luminosity at ffiffi ffi The search excludes W 0 SSM with a mass below 2.7 TeV at the 95% credibility level and W 0 NU with a mass below 2.7-2.0 TeV for cot ϕ NU in the range 1.0-5.5. The most stringent limit on W 0 SSM from searches in the eν and μν final states is 5.1 TeV from ATLAS [9] using 36.1 fb −1 of integrated luminosity at ffiffi ffi s p ¼ 13 TeV.
The ATLAS experiment is a multipurpose particle detector with a forward-backward symmetric cylindrical geometry [10,11]. It consists of an inner detector for charged-particle tracking in the pseudorapidity region jηj < 2.5, electromagnetic and hadronic calorimeters that provide energy measurements up to jηj ¼ 4.9, and a muon spectrometer that covers jηj < 2.7. A two-level trigger system is used to select events [12].
Hadronic τ decays are composed of a neutrino and a set of visible decay products (τ had-vis ), typically one or three charged pions and up to two neutral pions. The reconstruction of the visible decay products [13] is seeded by jets reconstructed from topological clusters of energy depositions [14] in the calorimeter. The τ had-vis candidates must have a transverse momentum p T > 50 GeV, jηj < 2.4 (excluding 1.37 < jηj < 1.52), one or three associated tracks, and an electric charge of AE1. Only the candidate with the highest p T in each event is selected. Hadronic τ decays are identified using boosted decision trees that exploit calorimetric shower shape and tracking information [15,16]. Loose criteria are used, which offer adequate rejection against quark-and gluon-initiated jets. Very loose criteria, with about one quarter of the rejection power, are used to create control regions. An additional dedicated veto is used to reduce the number of electrons misidentified as τ had-vis . The total efficiency for τ had-vis is ∼60% at p T ¼ 100 GeV and decreases to ∼30% at p T ¼ 2 TeV, where the large boost and collimation of the decay products causes inefficiencies in the track reconstruction and association.
Events containing electron or muon candidates are rejected. Electron candidates [17][18][19] must have p T > 20 GeV, jηj < 2.47 (excluding 1.37 < jηj < 1.52) and must pass a loose likelihood-based identification selection. Muon candidates [20] are required to have p T > 20 GeV, jηj < 2.5 and to pass a very loose muon identification requirement. The missing transverse momentum, with magnitude E miss T , is calculated as the negative vectorial sum of the p T of all reconstructed and calibrated τ had-vis candidates and jets [21][22][23]. A correction that accounts for momentum not associated with these reconstructed objects is calculated using inner-detector tracks that originate from the hard-scattering vertex [23]. The correction contributes no more than 5% on average in signal events.
Events are selected by triggers that require E miss T above thresholds of 70, 90, or 110 GeV depending on the datataking period. To minimize uncertainties in the trigger efficiency, the offline reconstructed E miss T is required to be at least 150 GeV. At this threshold the trigger efficiency is 80% and increases to more than 98% above 250 GeV. This behavior is determined by the E miss T resolution of the trigger, which is lower than in the offline reconstruction. The events must satisfy criteria designed to reduce backgrounds from cosmic rays, single-beam-induced events and calorimeter noise [24] and they must contain a loose τ had-vis candidate. To further suppress single-beam-induced background, the τ had-vis must have at least one associated track with p T > 10 GeV. The multijet background is further suppressed by requiring that the τ had-vis p T and the E miss T are balanced: 0.7 < p τ T =E miss T < 1.3. The azimuthal angle between the τ had-vis and the missing momentum, Δϕ, is required to be larger than 2.4. Finally, thresholds ranging from 0.25 to 1.8 TeV in steps of 0.05 TeV are placed on the transverse mass, m T , where m 2 T ≡ 2p τ T E miss T ð1 − cos ΔϕÞ. The background is divided into events where the selected τ had-vis originates from a quark-or gluon-initiated jet (jet background) and those where it does not (nonjet background). The jet background originates primarily from W=Z þ jets and multijet production and is estimated using a data-driven technique. The nonjet background is estimated using simulation and originates primarily from W → τν production with additional minor contributions from W=Z=γ Ã , tt, single top-quark, and diboson (WW, WZ and ZZ) production (collectively called others).
The event generators and other software packages used to produce the simulated samples are summarized in Table I. The W=Z=γ Ã sample is artificially enhanced in high-mass events to improve statistical coverage in the scanned mass range. Particle interactions with the ATLAS detector are simulated with GEANT 4 [25,26] and contributions from additional pp interactions (pileup) are simulated using PYTHIA 8.186 and the MSTW2008LO parton distribution function (PDF) set [27]. Finally, the simulated events are processed through the same reconstruction software as the data. Corrections are applied to account for mismodeling of the momentum scales and resolutions of reconstructed objects, the τ had-vis reconstruction and identification efficiency, the electron to τ had-vis misidentification rate, and the E miss T trigger efficiency. The simulated samples are normalized using the integrated luminosity of the collected data set and their theoretical cross sections. The W=Z=γ Ã cross sections are calculated as a function of the boson mass at nextto-next-to-leading order (NNLO) [49] using the CT14NNLO PDF set, including electroweak corrections at next-to-leading order (NLO) [50] using the MRST2004QED PDF set [51]. Uncertainties are taken from Ref. [52] and include variations of the PDF sets, scale, α S , beam energy, and electroweak corrections. The variations amount to a ∼5% total uncertainty in the W=Z=γ Ã cross section at low mass, increasing to 34% at 2 TeV. The tt and single top-quark production cross sections are TABLE I. The event generators and other software packages used to generate the matrix-element process and model nonperturbative effects in the simulated event samples. The top-quark mass is set to 172.5 GeV.

Process
Matrix element Nonperturbative Refs.
calculated to at least NLO with an uncertainty of 3%-6% [53][54][55][56]. The diboson cross sections are calculated to NLO with an uncertainty of 10% [44,57]. The simulated samples are affected by uncertainties associated with the generation of the events, the detector simulation, and the determination of the integrated luminosity. Uncertainties related to the modeling of the hard scatter, radiation, and fragmentation are at most 2% of the total background estimate. Uncertainties in the detector simulation manifest themselves through the efficiency of reconstruction, identification and triggering algorithms, and through particle energy scales and resolutions. The effects of energy uncertainties are propagated to E miss T . The uncertainty in the τ had-vis identification efficiency is 5%-6%, as determined from measurements of Z → ττ events. An additional uncertainty that increases by 20%-25% per TeV is assigned to τ had-vis candidates with p T > 150 GeV in accord with studies of high-p T jets [58]. The uncertainty in the τ had-vis energy scale is 2%-3%. The probability for electrons to be misidentified as τ had-vis is measured with a precision of 3%-14% [16]. The uncertainty in the E miss T trigger efficiency is negligible for E miss T > 300 GeV and can be as large as 10% for E miss T < 300 GeV. Uncertainties associated with reconstructed electrons, muons, and jets are found to have a very small impact. The uncertainty in the combined 2015 þ 2016 integrated luminosity is 2.1%, derived following a methodology similar to that used in Ref. [59], and has a minor impact. The uncertainty related to the simulation of pileup is ∼1%.
The W 0 signal events are modeled by reweighting the W sample using a leading-order matrix-element calculation. Electroweak corrections for the W cross section and interference between W and W 0 are not included as they are model dependent. Uncertainties in the W 0 cross section are estimated in the same way as for W bosons. They are not included in the fitting procedure used to extract experimental cross-section limits, but are instead included when overlaying predicted model cross sections. Uncertainties in the W 0 acceptance due to PDF, scale, and α S variations are negligible. In the NU model, the total decay width increases to 35% of the pole mass for large values of cot ϕ NU , which decreases the signal acceptance as more events are produced at low mass. Decays to WZ and Wh are not considered in the calculation of the total W 0 NU decay width as their impact is small (< 7%) and model dependent. Values of cot ϕ NU > 5.5 are not considered as the model is nonperturbative in this range.
The jet background contribution is estimated using events in three control regions (CR1, CR2, and CR3). The events must pass the selection for the signal region, except in CR1 and CR3 they must fail loose but pass very loose τ had-vis identification and in CR2 and CR3 they must have E miss T < 100 GeV and the requirement on p τ T =E miss T is removed. The low-E miss T requirement yields high multijet purity in CR2 and CR3, while the very loose identification preferentially rejects gluon-initiated jets over quark-initiated jets. This produces a similar fraction of quark-initiated jets in all control regions, which ensures minimal correlation between the identification and E miss T . The estimated jet contribution is defined as N jet ¼ N CR1 N CR2 =N CR3 . The nonjet contamination in CR1 (10%), CR2 (3.7%), and CR3 (0.5%) is subtracted using simulation. The transfer factor, N CR2 =N CR3 , is parametrized in τ had-vis p T and track multiplicity and is in the range 0.4-0.7 (0.15-0.3) for 1track (3-track) τ had-vis . Systematic uncertainties are assigned to account for any residual correlation between the transfer factor and the E miss T and p τ T =E miss T selection criteria, which would arise if the jet composition was different in CR1 and CR3. They are evaluated by repeating the jet estimate with the following modified control region definitions: (a) altered very loose τ had-vis identification criteria, (b) modified E miss T and p τ T =E miss T selection, and (c) CR2 and CR3 replaced by alternative control regions rich in Wð→ μνÞ þ jets events. The corresponding variations define the dominant uncertainty in the jet background contribution, which ranges from 20% at m T ¼ 0.2 TeV to þ200% −60% at m T ¼ 2 TeV, where the jet background is subdominant. The uncertainty due to the subtraction of nonjet contamination in the control regions is negligible.
To reduce the impact of statistical fluctuations in the jet background estimate, a function fðm T Þ ¼ m aþb log m T T , where a and b are free parameters, is fitted to the estimate in the range 400 < m T < 800 GeV and is used to evaluate the jet background in the range m T > 500 GeV. The impact of altering the fit range leads to an uncertainty that increases with m T , reaching 50% at m T ¼ 2 TeV. The statistical uncertainty from the control regions is propagated using pseudoexperiments and also reaches 50% at m T ¼ 2 TeV. Figure 1 shows the observed m T distribution of the data after event selection, including the estimated SM background contributions and predictions for W 0 SSM and W 0 NU (cot ϕ NU ¼ 5.5) bosons with masses of 3 TeV. The number of observed events is consistent with the expected SM background. Therefore, upper limits are set on the production of a high-mass resonance decaying to τν. The statistical analysis uses a likelihood function constructed as the Poisson probability describing the total number of observed events given the signal-plus-background expectation. Systematic uncertainties in the expected number of events are incorporated into the likelihood via nuisance parameters constrained by Gaussian prior probability density distributions. Correlations between signal and background are taken into account. A signal-strength parameter, with a uniform prior probability density distribution, multiplies the expected signal. The dominant relative uncertainties in the expected signal and background contributions are shown in Fig. 2 as a function of the m T threshold.
Limits are set at the 95% credibility level (C.L.) using the Bayesian Analysis Toolkit [60]. Figure 3 shows the model-independent upper limits on the visible τν production cross section, σðpp → τν þ XÞAε, as a function of the m T threshold, where A is the fiducial acceptance (including the m T threshold) and ε is the reconstruction efficiency. Modelspecific limits can be derived by evaluating σ, A, and ε for the model in question and checking if the corresponding visible cross section is excluded at any m T threshold. This allows the results to be reinterpreted for a broad range of models, regardless of their m T distribution. Good agreement between the generated and reconstructed m T distributions is found, indicating that a reliable calculation of the m T threshold acceptance can be made at generator level. The reconstruction efficiency depends on m T , εðm T ½TeVÞ ¼ 0.633 − 0.313m T þ 0.0688m 2 T − 0.00575m 3 T , ranging from 60% at 0.2 TeV to 7% at 5 TeV, and must be appropriately integrated out given the m T distribution of the model. The relative uncertainty in the parametrized efficiency due to the choice of signal model is ∼10%. With these inputs the visible cross sections for W 0 SSM and W 0 NU bosons could be reproduced within 10% using only generator-level information. Data and details to facilitate reinterpretations can be found at Ref. [61].
Limits are also set on benchmark models by selecting the most sensitive m T threshold for each W 0 mass hypothesis (∼0.6m W 0 up to a maximum of 1.45 TeV). The chosen threshold is found to have little dependence on the W 0 width. Figure 4(a) shows the 95% C.L. upper limit on the cross section times branching fraction as a function of m W 0 in the SSM. Heavy W 0 SSM bosons with a mass lower than 3.7 TeV are excluded, with an expected exclusion limit of 3.8 TeV. Figure 4(b) shows the excluded region in the parameter space of the nonuniversal Gð221Þ model. Heavy W 0 NU bosons with a mass lower than 2.2-3.8 TeV are excluded depending on cot ϕ NU , thereby probing a significantly larger region of parameter space than previous searches [8]. The W 0 NU limits are typically weaker than the W 0 SSM limits as the increased W 0 width yields lower acceptances, while the enhancement in the decay rate cancels with the suppression in the production via first-and second-generation quarks. Limits from the ATLAS ee, μμ, and ττ searches [58,62] 3. The 95% C.L. upper limit on the visible τν production cross section as a function of the m T threshold. also overlaid, showing that the τν search is complementary and extends the sensitivity over a large fraction of the parameter space. These results suggest that the τν searches should be considered when placing limits on nonuniversal extended gauge groups, such as those seeking to explain lepton flavor violation in B meson decays.
In summary, a search for W 0 → τν in 36.1 fb −1 of pp collisions at ffiffi ffi s p ¼ 13 TeV recorded by the ATLAS detector at the LHC is presented. The channel where the τ decays hadronically is analyzed and no significant excess over the SM expectation is found. Upper limits are set on the visible cross section for τν production, allowing interpretation in a broad range of models. Sequential standard model W 0 SSM bosons with masses less than 3.7 TeV are excluded at 95% C.L., while nonuniversal Gð221Þ W 0 NU bosons with masses less than 2.2-3.8 TeV are excluded depending on the model parameters.
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.    . 4. (a) The 95% C.L. upper limit on the cross section times τν branching fraction for W 0 SSM . The W 0 SSM cross section is overlaid where the additional lines represent the total theoretical uncertainty. (b) Excluded region for W 0 NU . The 95% C.L. limits from the ATLAS ee, μμ [62], and ττ [58] searches and indirect limits at 95% C.L. from fits to electroweak precision measurements (EWPT) [63], lepton flavor violation (LFV) [64], CKM unitarity [65], and the original Z-pole data [2] are overlaid.