Search for the associated production of the Higgs boson with a top quark pair in multilepton final states with the ATLAS detector

A search for the associated production of the Higgs boson with a top quark pair is performed in multilepton ﬁnal states using 20.3 fb − 1 of proton–proton collision data recorded by the ATLAS experiment at √ s = 8 TeV at the Large Hadron Collider. Five ﬁnal states, targeting the decays H → W W ∗ , ττ , and Z Z ∗ , are examined for the presence of the Standard Model (SM) Higgs boson: two same-charge light leptons ( e or μ ) without a hadronically decaying τ lepton; three light leptons; two same-charge light leptons with a hadronically decaying τ lepton; four light leptons; and one light lepton and two hadronically decaying τ leptons. No signiﬁcant excess of events is observed above the background expectation. The best ﬁt for the t ¯ tH production cross section, assuming a Higgs boson mass of 125 GeV, is 2 . 1 + 1 . 4 − 1 . 2 times the SM expectation, and the observed (expected) upper limit at the 95% conﬁdence level is 4.7 (2.4) times the SM rate. The p -value for compatibility with the background-only hypothesis is 1 . 8 σ ; the expectation in the presence of a Standard Model signal is 0 . 9 σ . © 2015 CERN for the beneﬁt of the ATLAS Collaboration. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Funded by SCOAP 3 .

The observation of the process in which the Higgs boson is produced in association with a pair of top quarks (tt H) would permit a direct measurement of the top quark-Higgs boson Yukawa coupling in a process that is tree-level at the lowest order, which is otherwise accessible primarily through loop effects. Having both the tree-and loop-level measurements would allow disambiguation of new physics effects that could affect the two differently, such as dimension-six operators contributing to the gg H vertex. This letter describes a search for the SM Higgs boson in the tt H production mode in multilepton final states. The five final states considered are: two same-charge-sign light leptons (e or μ) with no additional hadronically decaying τ lepton; three light leptons; two same-sign light leptons with one hadronically decaying τ lepton; four light leptons; and one light lepton with two hadronically decaying τ candidates. These channels are sensitive to the Higgs decays H → W W * , τ τ , and Z Z * produced in association E-mail address: atlas.publications@cern.ch. with a top quark pair decaying to one or two leptons. A similar search has been performed by the CMS Collaboration [18].
The selections of this search are designed to avoid overlap with ATLAS searches for tt H in H → bb [19] and H → γ γ [20] decays.
The main backgrounds to the signal arise from tt production with additional jets and non-prompt leptons, associated production of a top quark pair and a vector boson W or Z (collectively denoted tt V ), and other processes where the electron charge is incorrectly measured or where quark or gluon jets are incorrectly identified as τ candidates.

ATLAS detector and dataset
The features of the ATLAS detector [21] most relevant to this analysis are briefly summarized here. The detector consists of an inner tracking detector system surrounded by a superconducting solenoid, electromagnetic and hadronic calorimeters, and a muon spectrometer. Charged particles in the pseudorapidity 1 range |η| < 2.5 are reconstructed with the inner tracking detector, which is immersed in a 2 T magnetic field parallel to the detector axis 1 The ATLAS experiment uses a right-handed coordinate system with its origin at The associated production of a single top quark and a Z boson is a subleading background for the most sensitive channels. The cross section has been calculated at NLO for the t-and s-channels [32]. The resulting values used in this work are 160 ± 7 (scale) ± 11 (PDF) fb for t Z and 76 ± 4 (scale) ± 5 (PDF) fb for t Z . The cross section for the production of tW Z is computed at leading order (LO) using the MadGraph v5 generator [33] and found to be 4.1 fb.
The cross section for inclusive production of vector boson pairs W W , W Z, and Z Z is computed using MCFM [34]. Contributions from virtual photons and off-shell Z bosons are included. The uncertainties on the acceptance for these processes in the signal regions (which favour production with additional b-or c-quarks) dominate over the inclusive cross-section uncertainty (see Section 7.2) and so the latter is neglected in the analysis.

Event generation
The event generator configurations used for simulating the signal and main background processes are shown in Table 1. Additional information is given below.
The tt H signal event simulation samples contain all Higgs boson decays with branching fractions set to values computed at NNLO in QCD [26, [66][67][68][69]. The factorization (μ F ) and renormalization (μ R ) scales are set to m t + m H /2. Higgs boson and top quark masses of 125 and 172.5 GeV, respectively, are used. These samples are the same as those used by other ATLAS tt H searches [19,20].
Production of single top quarks with Higgs bosons is simulated as follows. For t Hqb, events are generated at leading order with MadGraph in the four-flavour scheme. For t H W , events are generated at NLO with MG5_aMC@NLO in the five-flavour scheme. Higgs boson and top quark masses are set as for tt H production.
The main irreducible backgrounds are production of ttW and tt Z (ttV ). For the ttW process, events are generated at leading order with zero, one, or two extra partons in the final state, while for tt Z zero or one extra parton is generated. The important contribution from off-shell γ * /Z → + − is included. The t Z process is simulated with the same setup, without extra partons. For diboson processes, the full matrix element for + − production, including γ * and off-shell Z contributions, is used. The Sherpa qq and qg samples include diagrams with additional partons in the final state at the matrix-element (ME) level, and include b-and c-quark mass effects. Sherpa was found to have better agreement with data than Powheg for W Z, while the Sherpa and Powheg descriptions of Z Z production are similar.
A tt + jets sample generated with the Powheg NLO generator [61] is used; the top quark mass is set to 172.5 GeV. Small corrections to the tt system and top quark p T spectra are applied based on discrepancies in differential distributions observed between data and simulation at 7 TeV [70]. Double-counting between the tt and W t single top production final states is eliminated using the diagram-removal method [71].
Samples of Z → + − + jets and W → ν + jets events are generated with up to five additional partons using the Alpgen v2.14 [65] leading order (LO) generator. Samples are merged with matrix element-parton shower overlaps removed using MLM Table 1 Configurations used for event generation of signal and background processes. If only one parton distribution function is shown, the same one is used for both the matrix element (ME) and parton shower generators; if two are shown, the first is used for the matrix element calculation and the second for the parton shower. "Tune" refers to the underlying-event tune of the parton shower generator. "Pythia 6" refers to version 6.425; "Pythia 8" refers to version 8.1; "Herwig++" refers to version 2.6; "MadGraph" refers to version 5; "Alpgen" refers to version 2.14; "Sherpa" refers to version 1.4; "gg2ZZ" refers to version 2.0.  matching [72]. Production of b-and c-quarks is also computed at matrix-element level, and overlaps between ME and parton shower production are handled by separating the kinematic regimes based on the angular separation of additional heavy partons. The resulting "light" and "heavy" flavour samples are normalized by comparing the resulting b-tagged jet spectra with data. All simulated samples with Pythia 6 and Herwig [59] parton showering use Photos 2.15 [73] to model photon radiation and Tauola 1.20 [74] for τ decays. The Herwig++ samples model photon radiation with Photos but use the internal τ decay model. Samples using Pythia 8.1 and Sherpa use those generators' internal τ lepton decay and photon radiation generators. For Herwig samples, multiple parton interactions are modelled with Jimmy [75].

Process
Showered and hadronized events are passed through simulations of the ATLAS detector (either full GEANT4 [76] simulation or a hybrid simulation with parameterized calorimeter showers and GEANT4 simulation of the tracking systems [77,78]). Additional minimum-bias pp interactions (pileup) are modelled with the Pythia 8.1 generator with the MSTW2008 LO PDF set and the A2 tune [79]. They are added to the signal and background simulated events according to the luminosity profile of the recorded data, with additional overall scaling to achieve a good match to observed calorimetry and tracking variables. The contributions from pileup interactions both within the same bunch crossing as the hard-scattering process and in neighbouring bunch crossings are included in the simulation.

Object selection
Electron candidates are reconstructed from energy clusters in the electromagnetic calorimeter associated with reconstructed tracks in the inner detector. They are required to have |η cluster | < 2.47. Candidates in the transition region 1.37 < |η cluster | < 1.52 between sections of the electromagnetic calorimeter are excluded. A multivariate discriminant based on shower shape and track information is used to distinguish electrons from hadronic showers [80,81]. Only electron candidates with transverse energy E T greater than 10 GeV are considered. To reduce the background from non-prompt electrons, i.e. from decays of hadrons (including heavy flavour) produced in jets, electron candidates are required to be isolated. Two isolation variables, based on calorimetric and tracking variables, are computed. The first (E cone T ) is based on the sum of transverse energies of calorimeter cells within a cone of radius R ≡ ( φ) 2 + ( η) 2 = 0.2 around the electron candidate direction. This energy sum excludes cells associated with the electron and is corrected for leakage from the electromagnetic shower and ambient energy in the event. The second (p cone T ) is defined based on tracks with p T > 1 GeV within a cone of radius R = 0.2 around the electron candidate. Both isolation energies are separately required to be less than 0.05 × E T . The longitudinal impact parameter of the electron track with respect to the selected event primary vertex, multiplied by the sine of the polar angle, |z 0 sin θ|, is required to be less than 1 mm. The transverse impact parameter divided by the estimated uncertainty on its measurement, |d 0 |/σ (d 0 ), must be less than 4. If two electrons closer than R = 0.1 are selected, only the one with the higher p T is considered. An electron is rejected if, after passing all the above selections, it lies within R = 0.1 of a selected muon.
Muon candidates are reconstructed by combining inner detector tracks with track segments or full tracks in the muon spectrometer [82]. Only candidates with |η| < 2.5 and p T > 10 GeV are kept. Additionally, muons are required to be separated by at least R > 0.04 + (10 GeV)/p T,μ from any selected jets (see below for details on jet reconstruction and selection). The cut value is optimized to maximize the acceptance for prompt muons at a fixed rejection factor for non-prompt and fake muon candidates. Furthermore, muons must satisfy similar E cone T and p T cone isolation criteria as for electrons, with both required to be less than 0.10 × p T . The value of |z 0 sin θ| is required to be less than 1 mm, while |d 0 |/σ (d 0 ) must be less than 3.
Hadronically decaying τ candidates (τ had ) are reconstructed using clusters in the electromagnetic and hadronic calorimeters. The τ candidates are required to have p T greater than 25 GeV and |η| < 2.47. The number of charged tracks associated with the τ candidates is required to be one or three and the charge of the τ candidates, determined from the associated tracks, must be ±1.
The τ identification uses calorimeter cluster and tracking-based variables, combined using a boosted decision tree (BDT) [83]. An additional BDT which uses combined calorimeter and track quantities is employed to reject electrons reconstructed as one-prong hadronically decaying τ leptons.
Jets are reconstructed from calibrated topological clusters [21] built from energy deposits in the calorimeters, using the anti-k t algorithm [84][85][86] with a radius parameter R = 0.4. Prior to jet finding, a local cluster calibration scheme [87,88] is applied to correct the topological cluster energies for the effects of non-compensating calorimeter response, inactive material and out-of-cluster leakage. The jets are calibrated using energy and η-dependent calibration factors, derived from simulations, to the mean energy of stable particles inside the jets. Additional correc-tions to account for the difference between simulation and data are derived from in-situ techniques [89,90]. After energy calibration, jets are required to have p T > 25 GeV and |η| < 2.5.
To reduce the contamination from jets originating in pp interactions within the same bunch crossing (pileup), the scalar sum of the p T of tracks matched to the jet and originating from the primary vertex must be at least 50% of the scalar sum of the p T of all tracks matched to the jet. This criterion is only applied to jets with p T < 50 GeV (those most likely to originate from pileup) and |η| < 2.4 (to avoid inefficiency at the edge of tracking acceptance).
The calorimeter energy deposits from electrons are typically also reconstructed as jets; in order to eliminate double counting, any jets within R = 0.3 of a selected electron are not considered.
Jets containing b-hadrons are identified (b-tagged) via a multivariate discriminant [91] that combines information from the impact parameters of displaced tracks with topological properties of secondary and tertiary decay vertices reconstructed within the jet. The working point used for this search corresponds to approximately 70% efficiency to tag a b-hadron jet, with a light-jet mistag rate of ≈ 1% and a charm-jet rejection factor of 5, as determined for b-tagged jets with p T of 20-100 GeV and |η| < 2.5 in simulated tt events. To avoid inefficiencies associated with the edge of the tracking coverage, only jets with |η| < 2.4 are considered as possible b-tagged jets in this analysis. The efficiency and mistag rates of the b-tagging algorithm are measured in data [91,92] and correction factors are applied to the simulated events.

Event selection and classification
All events considered in this analysis are required to pass single-lepton (e or μ) triggers. These achieve their maximal plateau efficiency for lepton p T > 25 GeV.
This analysis primarily targets the H → W W * and τ τ decay modes. Considering the decay of the tt system as well, these tt H events contain either W W W W bb or τ τ W W bb. The strategy is to target final states that cannot be produced in tt decay alone -i.e., three or more leptons, or two same-sign leptons -thus suppressing what would otherwise be the largest single background.
The analysis categories are classified by the number of light leptons and hadronic τ decay candidates. The leptons are selected using the criteria described earlier. Events are initially classified by counting the number of light leptons with p T > 10 GeV. At least one light lepton is required to match a lepton selected by the trigger system. After initial sorting into analysis categories, in some cases the lepton selection criteria are tightened by raising the p T threshold, tightening isolation selections or restricting the allowed |η| range, as explained in the following per-category descriptions.
The analysis includes five distinct categories: two same-sign light leptons with no τ had (2 0τ had ), three light leptons (3 ), two samesign light leptons and one τ had (2 1τ had ), four light leptons (4 ), and one light lepton and two τ had (1 2τ had ). The categories with τ had candidates target the H → τ τ decay; the others are primarily sensitive to H → W W * with a very small contribution from H → Z Z * . The contributions to each category from different Higgs boson decay modes are shown in Table 2. These selection criteria ensure that an event can only contribute to a single category. The contamination from gluon fusion, vector boson fusion, and associated V H production mechanisms for the Higgs boson is predicted to be negligible. Summed over all categories, the total expected number of reconstructed signal events assuming Standard Model tt H production is 10.2, corresponding to 0.40% of all produced tt H events. The detailed criteria for each category are described below. to E cone T /p T < 0.05 and p T cone /p T < 0.05. The angular acceptance of electron candidates is restricted to |η| < 1.37 in order to suppress tt background events where the sign of the electron charge is misreconstructed, as the charge misidentification rate increases at high pseudorapidity.
In order to suppress the lower-multiplicity tt + jets and ttW backgrounds, events must include at least four reconstructed jets. In order to suppress diboson and single-boson backgrounds, at least one of these jets must be b-tagged. The selected events are separated by lepton flavour (e ± e ± , e ± μ ± , and μ ± μ ± ) and number of jets (exactly four jets, at least five jets) into six categories with different signal-to-background ratio, resulting in higher overall sensitivity to the tt H signal.

3 category
Selected events are required to include exactly three light leptons with total charge equal to ±1. Candidate events arising from non-prompt leptons overwhelmingly originate as oppositesign dilepton events with one additional non-prompt lepton. As a result, the non-prompt lepton is generally one of the two leptons with the same charge. To reduce these backgrounds, a higher momentum threshold p T > 20 GeV is applied to the two leptons with the same charge. No requirements are imposed on the number of τ had candidates. In order to suppress the tt + jets and tt V backgrounds, selected events are required to include either at least four jets of which at least one must be b-tagged, or exactly three jets of which at least two are b-tagged. To suppress the tt Z background, events that contain an opposite-sign same-flavour lepton pair with the dilepton invariant mass within 10 GeV of the Z mass are vetoed. Events containing an opposite-sign lepton pair with invariant mass below 12 GeV are also removed to suppress background from resonances that decay to light leptons.

2 1τ had category
Selected events are required to include exactly two light leptons, with the same charge and leading (subleading) p T > 25 (15) GeV, and exactly one hadronic τ candidate. The reconstructed charge of the τ had candidate has to be opposite to that of the light leptons. In order to reduce tt + jets and tt V backgrounds, events must include at least four reconstructed jets. In order to suppress diboson and single-boson backgrounds, at least one jet must be b-tagged. To suppress the Z → + − + jets background, events with dielectron invariant mass within 10 GeV of the Z mass are vetoed.

4 categories
Selected events are required to include exactly four light leptons with total charge equal to zero and leading (subleading) p T > 25 (15) GeV. No requirements are applied on the number of τ had candidates. In order to suppress the tt + jets and tt V backgrounds, the selected events are required to include at least two jets of which at least one must be b-tagged. To suppress the tt Z background, events that contain an opposite-sign same-flavour lepton pair with dilepton invariant mass within 10 GeV of the Z mass are vetoed. In order to suppress background contributions from resonances that decay to light leptons, all opposite-sign sameflavour lepton pairs are required to have a dilepton invariant mass greater than 10 GeV. The four-lepton invariant mass is required to be between 100 and 500 GeV, which gives high acceptance for tt H, H → W W * → ν ν, but rejects Z → 4 and high-mass tt Z events. Selected events are separated by the presence or absence of a same-flavour, opposite-sign lepton pair into two categories, referred to respectively as the Z -enriched and Z -depleted categories. In both cases the Z mass veto is applied, but background events in the Z -enriched category can arise from off-shell Z and γ * → + − processes while in the Z -depleted category these backgrounds are absent.

1 2τ had category
Selected events are required to include exactly one light lepton with p T > 25 GeV and exactly two hadronic τ candidates. The τ had candidates must have opposite charge. In order to suppress the tt + jets and tt V backgrounds, events must include at least three reconstructed jets. In order to suppress diboson and single-boson backgrounds, at least one of the jets must be b-tagged. This final state is primarily sensitive to H → τ + τ − decays, allowing use of the invariant mass of the visible decay products of the τ had τ had system (m vis ) as a signal discriminant. Signal events are required to satisfy 60 < m vis < 120 GeV.

Background estimation
Important irreducible backgrounds include tt V and diboson production and are estimated from MC simulation. Validation regions enriched in these backgrounds are used to verify proper modelling of data by simulation. Reducible backgrounds are due to non-prompt lepton production and electron charge misidentification, and are estimated from data, with input from simulation in some categories. In the 1 2τ had category the primary concern is fake τ had candidates, which are modelled using simulation and validated against a data-driven estimate.

tt V and t Z
The primary backgrounds with prompt leptons stem from the production of ttW and tt Z . The ttW background tends to have lower jet multiplicity than the signal and so the leading contribution comes from events with additional high-p T jets; it is the major tt V contribution in the 2 0τ had categories and comparable to tt Z in the 2 1τ had category. The tt Z process has similar multiplicity to the tt H signal but can only contribute to the signal categories when the Z boson decays leptonically, so the on-shell contribution can be removed by vetoing events with opposite-sign dilepton pairs with invariant mass near the Z pole. This is the larger of the two tt V contributions for the 3 , 4 , and 1 2τ had categories. The t Z process makes a subleading contribution to both channels. A validation region is used to verify the modelling of tt Z using on-shell Z decays. Agreement is seen within the large statistical uncertainty. No region of equivalent purity and statistical power exists for ttW production; nevertheless the expectations are cross-checked with a validation region defined with the 2 0τ had selection except with two or more b-tagged jets and either two or three jets, where the ttW purity is ≈30%, and are found to be consistent within uncertainties. The spectra of the number of jets in these validation regions are shown in Fig. 1.
Uncertainties on the tt V background contributions arise from both the overall cross section uncertainties (see Section 3) and the acceptance uncertainties. The latter are estimated by comparing particle-level samples after showering produced by three different pairs of generators: a) the nominal MadGraph LO merged Table 3 Expected and observed yields in each channel. Uncertainties shown are the sum in quadrature of systematic uncertainties and Monte Carlo simulation statistical uncertainties.
"Non-prompt" includes the misidentified τ had background to the 1 2τ had category. Rare processes (t Z , ttW W , triboson production, tttt, t H) are not shown as a separate column but are included in the total expected background estimate.

Other prompt lepton contributions
Other backgrounds with prompt leptons arise from multiboson processes (W Z, Z Z , and triboson production) in association with heavy-flavour jets, or with a misidentified light-flavour jet.
The main process affecting the final result is W Z + jets. Validation regions with three leptons including a Z candidate and either zero or one b-tagged jet are studied. The number of jets in W Z + 0b events is reproduced well in the highly populated bins (up to 4 jets), leading to the conclusion that the jet radiation spectrum is well modelled. The dominant uncertainty on the prediction in the signal region is expected to arise from the W Z + b cross section.
Data constrain this component with roughly 100% uncertainty. As a result a 100% uncertainty is assigned to the W Z + b cross section, giving a 50% uncertainty on the total W Z yield, correlated across categories. The cross sections for production of W W + b and Z Z + b are also assigned 50% uncertainties; these have negligible impact on the final result.

Charge sign misidentification
The process e ± → e ± γ → e ± e + e − occurring in detector material can result in an electron produced with nearly the same momentum as the parent electron but with opposite charge. In these cases the observed electron has opposite charge to that of the primary electron (charge mis-id). The analogous processes μ ± → μ ± e + e − and μ ± → μ ± μ + μ − have negligible rates for the selected events. The tt and Z /γ * → + − + jets events that undergo this process contribute to 2 0τ had in the ee and eμ categories. As electrons pass through more material at high |η|, the charge mis-id rate increases as well, and so the electron |η| < 1.37 requirement significantly reduces the impact of this background. The charge mis-id rate due to track curvature mismeasurement for electrons and muons is negligible.
The charge mis-id probability is determined by a maximumlikelihood fit using Z → ee events reconstructed as same-sign and as opposite-sign pairs, as a function of electron η and p T . This probability function is then applied to a sample of events passing the 2 0τ had selection except that the lepton pair is required to be opposite sign. The charge mis-id probability from the relatively low momentum Z daughters is extrapolated to higher p T using scaling functions extracted from Monte Carlo simulations. The dominant uncertainty is due to the statistical precision of the charge mis-id probability determination, and is ≈ 40% in the signal regions.

Non-prompt light leptons
A significant background arises from leptons not produced in decays of electroweak bosons (non-prompt leptons), which can promote (for example) a single-lepton tt event into a 2 0τ had category or a dilepton tt event to the 3 or 2 1τ had categories. These backgrounds in the signal regions are expected to be dominated by tt or single top quark production with leptons produced in decays of heavy-flavour hadrons. Production of tt with an additional photon which converts in the detector material is a subdominant contribution. With the tight object selection requirements applied in this analysis, almost all reconstructed electron and muon objects correspond to real electrons and muons; the fraction arising from incorrect particle identification is negligible. Estimates of these backgrounds are obtained from data. Each channel has a slightly different procedure, motivated by the specific event topology and the statistical power available in the control regions. The methods are discussed below, and the expected non-prompt lepton contributions to the various categories are shown in Table 3. In the following, a tight lepton is a lepton that passes the nominal selection, a sideband lepton is defined as a lepton candidate which satisfies different criteria than the tight lepton selection (identification selection, isolation, or p T ), and (sideband) control regions either require one or more sideband leptons to replace a tight lepton in the signal region selection, or have the same lepton selection as the signal region but different jet requirements.

2 0τ had categories
The non-prompt lepton yields in the signal regions are estimated by extrapolating from sideband control regions in data which are enriched in tt non-prompt contributions. For electrons, sideband objects are selected by inverting the electron identification and isolation requirements; for muons the sideband objects have low transverse momentum, 6 < p T < 10 GeV, but otherwise are selected the same way as nominal muons. Transfer factors are used to extrapolate from events with one tight and one sideband lepton, but which otherwise pass the signal region selections, to the signal regions with two tight leptons. These transfer factors are determined from additional data control regions (tight + sideband and two tight leptons) with lower jet multiplicity (1 ≤ n jet ≤ 3 for electrons, 2 ≤ n jet ≤ 3 for muons). In all regions the expected contribution from processes producing prompt leptons is subtracted before extracting transfer factors or using the yields for extrapolation. For channels with electrons, the charge mis-id background is also subtracted, and a dilepton mass veto is applied in the control regions to suppress contributions from Z → e + e − decays. A crosscheck on the muon estimate, using an extrapolation in muon isolation instead of muon p T , agrees well with the nominal procedure and provides additional confidence in the estimate.
The systematic uncertainties on this procedure are estimated by checking a) its ability to successfully predict the non-prompt background in tt simulation and b) the stability of the prediction using data when the selection of the control regions is altered. For the former, different parton shower and b-hadron decay models were checked, as was the result of removing the b-tagged jet requirement. In addition, for electrons, the effects of relaxing the pseudorapidity requirement to |η| < 2.5 and of raising the p T threshold were studied. These checks show stability at the 25-30% level, limited by the statistical precision of the simulations. The stability in data is checked by altering the p T required for the b-tagged jet, applying a requirement on missing transverse momentum 3 E miss T , extracting the transfer factors only from events with three jets, or (for muons) using 10-15 GeV muons as the sideband objects. This check shows stability of the predictions to 14% for muons and 19% for electrons. Additional systematic uncertainties in the prediction arise from the statistical uncertainties on the yields in the control regions and the subtraction of prompt and charge mis-id contributions. The overall uncertainties on the non-prompt yield prediction in any given category range from 32% to 52%, and correlations between the categories due to uncertainties in the transfer factors are included in the fit (see Section 9).

3 category
Sideband leptons are defined by reversing the isolation requirement for electrons and muons and, for electrons, requiring that the candidate fail the tight electron identification discriminant requirement of the analysis but pass a looser selection. The non-prompt lepton contribution in the signal region is estimated by extrapolating from data regions with two tight and one sideband lepton, using transfer factors estimated from Monte Carlo simulation. These events typically contain two prompt opposite-sign leptons and one non-prompt lepton, which necessarily must be of the same sign as one of the prompt leptons. Therefore the non-prompt lepton estimation procedure is applied only to the two same-sign leptons. The simulation-derived transfer factor is validated in a region of lower jet multiplicity (2 ≤ n jet ≤ 3 and exactly one b-tagged jet).
Good agreement is observed in this validation region between the prediction (11.8 ± 2.3) and the observed yield (9.8 ± 4.9 events after prompt background subtraction). Systematic uncertainties in the procedure are derived by studying the agreement between data and simulation in the variables used for the extrapolation, which is ≈20% for both electrons and muons. Additional uncertainties arise from the statistical uncertainties on the yields in the control regions and in the tt simulation. for separately; the small variations in the ratio in the control regions are found to have negligible impact on the total estimate in the signal region. In order to maintain similar origin composition of the non-prompt leptons, the E T isolation requirement is inverted, the p T isolation requirement is relaxed, and for electrons the identification criteria are also relaxed to a looser working point. The low jet multiplicity region 2 ≤ n jet ≤ 3 is used to determine a transfer factor from sideband to tight lepton selections. The expected non-prompt lepton yield in the signal region is obtained by using this transfer factor to extrapolate from a control region with the same jet selection as the signal region but with one tight and one sideband light lepton. The procedure is validated by checking that it correctly reproduces the signal region yield expected in tt simulations. The assigned systematic uncertainty (27%) is dominated by the statistical precision of this test. The overall uncertainty on the non-prompt background prediction is dominated by the limited statistics of the high jet multiplicity control region.

4 category
The non-prompt lepton contribution in this category is expected to be negligible and is estimated to be 10 −3 events in the Z -enriched sample and 10 −4 events in the Z -depleted sample. In both cases this represents 2% of the total background expectation. These estimates are obtained using the transfer factors from the 3 channel and appropriate control regions with two loose leptons and relaxed jet multiplicity requirements.

τ had misidentification in the 1 2τ had category
The nominal estimate for the fake τ had yield is derived from tt simulation. To obtain a sufficiently large sample size, fast simulation using parameterized calorimeter showers is used. At all preselection stages the simulation is found to give an acceptable description of the tt background, both in kinematic distributions and total yield. This estimate is cross-checked with the data-driven method described below.
Of the two τ had candidates, one is opposite in sign to the light lepton (OS) and the other has the same sign (SS). The SS candidate is almost always a fake τ had , while the light lepton is prompt and the OS τ had candidate is often real (≈30%). A sideband τ had is defined as a candidate passing a loose identification BDT selection but not the nominal tight one. Assuming the τ had candidate fake probabilities are not correlated between jets identified as OS and SS candidates, control regions can be used to predict yields in the signal region. There are three control regions, depending on whether only the OS, only the SS, or both the OS and SS τ had candidates are sideband objects. The two regions with sideband OS τ had candidates are used to obtain the transfer factor for the SS τ had candidate, which is then applied to the region with a tight OS and sideband SS candidate to obtain the prediction for the signal region where both are tight. The transfer factor is measured The non-prompt and charge mis-id background spectra are taken from simulation of tt, single top, Z → + − + jets, and other small backgrounds, with normalization as described in the text (in particular the = 4/ ≥ 5 jet regions of the 2 0τ had plot have the ratio given by the data-driven prediction). The overlaid red line shows the ttH signal from the Standard Model. For visibility, the tt H signal is multiplied by a factor of 2.4 in the 2 0τ had , 3 , and 1 2τ had plots. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) as a function of the p T , η, number of tracks, and b-tag discriminant value of the SS τ had candidate. The data-driven method is cross-checked in tt simulation and found to successfully predict the yields in the signal region. The main limitation of this method is the statistical power of the control regions.
The simulation-driven method is taken as the primary estimate, as the validation of the method at preselection stages is more precise than the data-driven method due to larger event yield for the former. The comparison of the simulation-and data-driven techniques gives a 36% uncertainty in the prediction in the sig-nal region, which is taken as the systematic uncertainty on the estimate.

Other systematic uncertainties
Systematic uncertainties not already discussed are summarized below.
The uncertainty on the integrated luminosity is 2.8%. This uncertainty is derived from a calibration of the luminosity scale derived from beam-separation scans performed in November 2012, following the same methodology as that detailed in Ref. [94].
Lepton reconstruction and identification uncertainties are obtained from Z → , Z → γ , ϒ → , and J /ψ → events [80][81][82]. Uncertainties on the detector response are assessed similarly to other ATLAS analyses. The modelling of the efficiency of the tight isolation requirements in simulation is explicitly checked as a function of the number of jets in the event. These corrections are found to be very small, with uncertainties limited by data statistics.
The largest jet-related systematic uncertainty arises from the jet energy scale, in particular contributions from the in-situ calibration in data, the different response to quark and gluon jets, and the pileup subtraction. The impact of the b-tagging efficiency uncertainty on the signal strength μ = σ tt H,obs /σ tt H,SM at the best-fit value of μ is μ = +0.08 −0.06 . Because only one (of typically two) b-jets present in signal or tt V events is required to be tagged, the uncertainty on the b-tagging efficiency (while included) does not have as large an effect in this analysis as it does in other tt H searches such as those targeting the H → bb decay.
The uncertainties on the inclusive tt H production cross section are discussed in Section 3. Additionally, the effects of PDF uncertainty, QCD scale choice, and parton shower algorithm on the signal acceptance in each analysis category are considered. The resulting relative uncertainties on the acceptance are 0.3-1.4% for PDF, 0.1-2.7% for scale choice, and 1.5-13% for parton shower algorithm.
For most backgrounds the uncertainties from Monte Carlo simulation sample sizes are negligible. For the diboson backgrounds, however, these can reach 50% of the total diboson yield uncertainties shown in Table 3.

Results
The observed yields, and a comparison with the expected backgrounds and tt H signal, are shown in Table 3. The distributions of the number of jets in the events passing signal region selections are shown in Fig. 2. The best-fit value of the signal strength μ = σ tt H,obs /σ tt H,SM is determined using a maximum likelihood fit to the data yields of the categories listed in Table 3, which are treated as independent Poisson terms in the likelihood. The fit is based on the profile-likelihood approach where the systematic uncertainties are treated as nuisance parameters with prior uncer- Systematic uncertainties are allowed to float in the fit as nuisance parameters and take on their best-fit values. The only constraints on nuisance parameter uncertainties found by the fit are for non-prompt lepton transfer factors and normalization region yields in the 2 0τ had categories and the fake τ had background yield in the 1 2τ had category. The former all have large statistical components and so the additional information from the signal regions is expected to constrain them. The latter has a very large initial uncertainty which the fit is able to constrain as μ is required to be the same in all categories. The largest difference between preand post-fit nuisance parameter values is in the 1 2τ had fake estimate, which shifts by −1.0σ due to the deficit of observed relative to expected events. The next largest effect is a +0.4σ shift in the 2 0τ had non-prompt μ transfer factor.
The results of the fit are shown in Fig. 3. The impact of the most important systematic uncertainties on the measured value of μ in the combined fit is shown in Table 4. In each category, the uncertainties on μ are mainly statistical, except for the combined 2 0τ had result where the statistical and systematic uncertainties  This analysis is a search for tt H production; as such, production of t Hqb and t H W is considered as a background and set to Standard Model expectation. Including this contribution as a background induces a shift of μ = −0.04 compared to setting it to zero. A full extraction of limits on the top quark Yukawa coupling including the relevant modifications of single top plus Higgs boson production is reported in Ref. [97].
The results are sensitive to the assumed cross sections for ttW and tt Z production, and use theoretical predictions for these values as experimental measurements do not yet have sufficient precision. The best-fit μ value as a function of these cross sections is Table 5 Observed and expected 95% CL upper limits, derived using the CL s method, on the strength parameter μ = σ tt H,obs /σ tt H,SM for a Higgs boson of mass m H = 125 GeV. The last column shows the median expected limit in the presence of a ttH signal of Standard Model strength.

Channel
Observed

Conclusions
A search for tt H production in multilepton final states has been performed using 20.3 fb −1 of proton-proton collision data at