Search for direct production of charginos, neutralinos and sleptons in final states with two leptons and missing transverse momentum in pp collisions at sqrt(s) = 8 TeV with the ATLAS detector

Searches for the electroweak production of charginos, neutralinos and sleptons in final states characterized by the presence of two leptons (electrons and muons) and missing transverse momentum are performed using 20.3 fb-1 of proton-proton collision data at sqrt(s) = 8 TeV recorded with the ATLAS experiment at the Large Hadron Collider. No significant excess beyond Standard Model expectations is observed. Limits are set on the masses of the lightest chargino, next-to-lightest neutralino and sleptons for different lightest-neutralino mass hypotheses in simplified models. Results are also interpreted in various scenarios of the phenomenological Minimal Supersymmetric Standard Model.

The charginos and neutralinos are mixtures of the bino, winos and higgsinos that are superpartners of the U(1), SU(2) gauge bosons and the Higgs bosons, respectively. Their mass eigenstates are referred to asχ ± i (i = 1, 2) andχ 0 j (j = 1, 2, 3,4) in the order of increasing masses. Even though the gluinos and squarks are produced strongly in pp collisions, if the masses of the gluinos and squarks are large, the direct production of charginos, neutralinos and sleptons through electroweak interactions may dominate the production of SUSY particles at the Large Hadron Collider (LHC). Such a scenario is possible in the general framework of the phenomenological minimal supersymmetric SM (pMSSM) [19][20][21]. Naturalness suggests that third-generation sparticles and some of the charginos and neutralinos should have masses of a few hundred GeV [22,23]. Light sleptons are expected in gauge-mediated [24][25][26][27][28][29] and anomaly-mediated [30, 31] SUSY breaking scenarios. Light sleptons could also play a role in the co-annihilation of neutralinos, allowing a dark matter relic density consistent with cosmological observations [32,33].
This paper presents searches for electroweak production of charginos, neutralinos and sleptons using 20.3 fb −1 of proton-proton collision data with a centre-of-mass energy √ s = 8 TeV collected at the LHC with the ATLAS detector. The searches target final states with two oppositely-charged leptons (electrons or muons) and missing transverse momentum. Similar searches [34,35] have been performed using √ s = 7 TeV data by the ATLAS and CMS experiments. The combined LEP limits on the selectron, smuon and chargino masses are mẽ > 99. 9 GeV, mμ > 94. 6 GeV and mχ± 1 > 103.5 GeV [36-40]. The LEP selectron limit assumes gaugino mass unification and cannot be directly compared with the results presented here.

SUSY scenarios
Simplified models [41] are considered for optimization of the event selection and interpretation of the results. The LSP is the lightest neutralinoχ 0 1 in all SUSY scenarios considered, except in one scenario in which it is the gravitinoG. All SUSY particles except for the LSP are assumed to decay promptly. In the electroweak production ofχ + 1χ − 1 andχ ± 1χ 0 2 , χ ± 1 andχ 0 2 are assumed to be pure wino and mass degenerate, and only the s-channel production diagrams, qq → (Z/γ) * →χ + 1χ − 1 and qq ′ → W ± * →χ ± 1χ 0 2 , are considered. The cross-section forχ + 1χ − 1 production is 6 pb for aχ ± 1 mass of 100 GeV and decreases to 10 fb at 450 GeV. The cross-section forχ ± 1χ 0 2 production is 11.5 pb for a degenerateχ ± 1 /χ 0 2 mass of 100 GeV, and 40 fb for 400 GeV.
Another scenario is considered in whichχ ± 1 andχ 0 2 are mass degenerate and are co-NLSPs. The directχ ± 1χ 0 2 production is followed by the decaysχ ± 1 → W ±χ0 1 andχ 0 2 → Zχ 0 1 with a 100% branching fraction. If the Z boson decays leptonically and the W boson decays hadronically, as shown in figure 1(c), the final state contains two opposite-sign leptons, two hadronic jets, and missing transverse momentum. The leptons in this case are SF and their invariant mass is consistent with the Z boson mass. The invariant mass of the two jets from the W decay gives an additional constraint to characterize this signal. A scenario in which the slepton is the NLSP is modelled according to ref. [42]. Figure 1(d) shows direct slepton-pair production pp →l +l− followed byl ± → ℓ ±χ0 1 (ℓ = e or µ), giving rise to a pair of SF leptons and missing transverse momentum due to the two neutralinos. The cross-section for direct slepton pair production in this scenario decreases from 127 fb to 0.5 fb per slepton flavour for left-handed sleptons, and from 49 fb to 0.2 fb for right-handed sleptons, as the slepton mass increases from 100 to 370 GeV.
Results are also interpreted in dedicated pMSSM [43] scenarios. In the models considered in this paper, the masses of the coloured sparticles, of the CP-odd Higgs boson, and of the left-handed sleptons are set to high values to allow only the direct production of charginos and neutralinos via W /Z, and their decay via right-handed sleptons, gauge bosons and the lightest Higgs boson. The lightest Higgs boson mass is set close to 125 GeV [44,45] by tuning the mixing in the top squark sector. The mass hierarchy, com-position and production cross-section of the charginos and neutralinos are governed by the ratio tan β of the expectation values of the two Higgs doublets, the gaugino mass parameters M 1 and M 2 , and the higgsino mass parameter µ. Two classes of pMSSM scenarios are studied on a µ-M 2 grid, distinguished by the masses of the right-handed sleptonsl R . If ml R lies halfway between mχ 0 1 and mχ 0 2 ,χ 0 2 decays preferentially throughχ 0 2 →l R ℓ →χ 0 1 ℓℓ. The parameter tan β is set to 6 yielding comparable branching ratios into each slepton generation. To probe the sensitivity to differentχ decay via W , Z and Higgs bosons. The remaining parameters are fixed to tan β = 10 and M 1 = 50 GeV so that the relic dark-matter density is below the cosmological bound across the entire µ-M 2 grid. The lightest Higgs boson has a mass close to 125 GeV and decays to both SUSY and SM particles where kinematically allowed.
In addition, the gauge-mediated SUSY breaking (GMSB) model proposed in ref. [46] is considered. In this simplified model, the LSP is the gravitinoG, the NLSP is the chargino with mχ ± 1 = 110 GeV, and in addition there are two other light neutralinos with masses mχ 0 1 = 113 GeV and mχ 0 2 = 130 GeV. All coloured sparticles are assumed to be very heavy. Theχ + 1χ − 1 production cross-section is not large (∼1.4 pb), but the same final state is reached via production ofχ ± 1χ 0 1 (∼2.5 pb),χ ± 1χ 0 2 (∼1.0 pb) andχ 0 1χ 0 2 (∼0.5 pb). Thẽ χ 0 1 decays intoχ ± 1 W ∓ * , and theχ 0 2 decays either intoχ ± 1 W ∓ * orχ 0 1 Z * . Because of the small mass differences, decay products of the off-shell W and Z bosons are unlikely to be detected. As a result, all of the four production channels result in the same experimental signature, and their production cross-sections can be added together for the purpose of this search. Eachχ ± 1 then decays viaχ ± 1 → W ±G , and leptonic decays of the two W bosons produce the same final-state as in the other scenarios.

The ATLAS detector
The ATLAS detector [47] is a multi-purpose particle physics detector with a forwardbackward symmetric cylindrical geometry and nearly 4π coverage in solid angle 1 . It contains four superconducting magnet systems, which include a thin solenoid surrounding the inner tracking detector (ID), and barrel and end-cap toroids as part of a muon spectrometer (MS). The ID covers the pseudorapidity region |η| < 2.5 and consists of a silicon pixel detector, a silicon microstrip detector, and a transition radiation tracker. In the pseudorapidity region |η| < 3.2, high-granularity liquid-argon (LAr) electromagnetic (EM) sampling calorimeters are used. An iron-scintillator tile calorimeter provides coverage for hadron detection over |η| < 1.7. The end-cap and forward regions, spanning 1.5 < |η| < 4.9, are instrumented with LAr calorimeters for both EM and hadronic measurements. The MS surrounds the calorimeters and consists of a system of precision tracking chambers (|η| < 2.7), and detectors for triggering (|η| < 2.4).

Monte Carlo simulation
Monte Carlo (MC) simulated event samples are used to develop and validate the analysis procedure and to evaluate the subdominant SM backgrounds as well as the expected signal yields. The dominant SM background processes include tt, single-top, and diboson (W W , W Z and ZZ) production. The predictions for the most relevant SM processes are normalized to data in dedicated control regions, as detailed in section 7. MC samples are produced using a GEANT4 [48] based detector simulation [49] or a fast simulation using a parameterization of the performance of the ATLAS electromagnetic and hadronic calorimeters [50,51] and GEANT4 elsewhere. The effect of multiple proton-proton collisions from the same or different bunch crossings is incorporated into the simulation by overlaying minimum bias events generated using PYTHIA [52] onto hard scatter events. Simulated events are weighted to match the distribution of the number of interactions per bunch crossing observed in data.
Samples of W → ℓν and Z/γ * → ℓℓ produced with accompanying jets (including light and heavy flavours) are obtained with a combination of SHERPA v1.4.1 and ALPGEN v2.14 [70]. The inclusive W and Z/γ * production cross-sections are normalized to the NNLO cross-sections obtained using DYNNLO v1.1 [71]. QCD production of bb and cc is simulated with PYTHIA v8.165.
Finally, production of the SM Higgs boson with m H = 125 GeV is considered. The gluon fusion and vector-boson fusion production modes are simulated with POWHEG-BOX v1.0, and the associated production (W H and ZH) with PYTHIA v8.165.
Fragmentation and hadronization for the MC@NLO and ALPGEN samples are performed either with HERWIG v6.520 [72] using JIMMY v4.31 [73] for the underlying event, or with PYTHIA v6.426. PYTHIA v6.426 is also used for MADGRAPH samples, whereas PYTHIA v8.165 is used for the POWHEG-BOX samples. For the underlying event, ATLAS tune AUET2B [74] is used. The CT10 NLO [75] and CTEQ6L1 [76] parton-distribution function (PDF) sets are used with the NLO and LO event generators, respectively.

Event reconstruction
Events are selected in which at least five tracks, each with transverse momentum p T > 400 MeV, are associated to the primary vertex. If there are multiple primary vertices in an event, the one with the largest p 2 T of the associated tracks is chosen. In each event, 'candidate' electrons, muons, hadronically-decaying τ leptons, and jets are reconstructed. After resolving potential ambiguities among objects, the criteria to define 'signal' electrons, muons and jets are refined. Hadronically-decaying τ leptons are not considered as signal leptons for this analysis, and events containing them are removed (see section 6) so that the data sample is distinct from that used in the ATLAS search for electroweak SUSY production in the three-lepton final states [82].
Electron candidates are reconstructed by matching clusters in the EM calorimeter with tracks in the ID. The magnitude of the momentum of the electron is determined by the calorimeter cluster energy. They are required to have p T > 10 GeV, |η| < 2.47, and satisfy shower-shape and track-selection criteria analogous to the 'medium' criteria in ref. [83].
Muon candidates are reconstructed by matching an MS track to an ID track [84]. They are then required to have p T > 10 GeV and |η| < 2.4.
Jet candidates are reconstructed from calorimeter energy clusters using the anti-k t jet clustering algorithm [85, 86] with a radius parameter of 0.4. The jet candidates are corrected for the effects of calorimeter response and inhomogeneities using energy-and ηdependent calibration factors based on simulation and validated with extensive test-beam and collision-data studies [87]. Energy deposition due to pile-up interactions is statistically subtracted based on the area of the jet [88]. Only jet candidates with p T > 20 GeV and |η| < 4.5 are subsequently retained. Events containing jets that are likely to have arisen from detector noise or cosmic rays are removed [87].
A b-jet identification algorithm [89] is used to identify jets containing a b-hadron decay inside a candidate jet within |η| < 2.4, exploiting the long lifetime of b-and c-hadrons. The mean nominal b-jet identification efficiency, determined from simulated tt events, is 80%. The misidentification (mis-tag) rates for c-jets and light-quark/gluon jets are approximately 30% and 4%, respectively. Small differences in the b-tagging performance observed between data and simulation are corrected for as functions of p T of the jets.
Hadronically-decaying τ leptons are reconstructed by associating tracks with p T > 1 GeV passing minimum track quality requirements to calorimeter jets with p T > 10 GeV and |η| < 2.5. A multivariate discriminant is used to identify the jets as hadronic τ decays [90]. Their energy is determined by applying a simulation-based correction to the reconstructed energy in the calorimeter [91], and p T > 20 GeV is required.
Object overlaps are defined in terms of ∆R = (∆η) 2 + (∆φ) 2 , where ∆η and ∆φ are separations in η and φ. Potential ambiguities among objects are resolved by removing one or both of nearby object pairs in the following order: if two electron candidates are within ∆R = 0.05 of each other, the electron with the smaller p T is removed; any jet within ∆R = 0.2 of an electron candidate is removed; any τ candidate within ∆R = 0.2 of an electron or a muon is removed; any electron or muon candidate within ∆R = 0.4 of a jet is removed; if an electron candidate and a muon candidate are within ∆R = 0.01 of each other, both are removed; if two muon candidates are within ∆R = 0.05 of each other, both are removed; if the invariant mass of a SF opposite-sign lepton pair has an invariant mass less than 12 GeV, both are removed; and finally any jet within ∆R = 0.2 of a τ candidate is removed.
Signal electrons are electron candidates satisfying the 'tight' criteria [83] placed on the ratio of calorimetric energy to track momentum, and the number of high-threshold hits in the transition radiation tracker. They are also required to be isolated. The p T scalar sum of tracks above 400 MeV within a cone of size ∆R = 0.3 around each electron candidate (excluding the electron candidate itself) and associated to the primary vertex is required to be less than 16% of the electron p T . The sum of transverse energies of the surrounding calorimeter clusters within ∆R = 0.3 of each electron candidate, corrected for the deposition of energy from pile-up interactions, is required to be less than 18% of the electron p T . The distance of closest approach of an electron candidate to the event primary vertex must be within five standard deviations in the transverse plane. The distance along the beam direction, z 0 , must satisfy |z 0 sin θ| < 0.4 mm.
Signal muons are muon candidates satisfying the following criteria. The p T scalar sum of tracks above 400 MeV within a cone of size ∆R = 0.3 around the muon candidate and associated to the primary vertex is required to be less than 16% of the muon p T . The distance of closest approach of a muon candidate to the event primary vertex must be within three standard deviations in the transverse plane, and |z 0 sin θ| < 1 mm along the beam direction.
The efficiencies for electrons and muons to pass the reconstruction, identification and isolation criteria are measured in samples of Z and J/ψ leptonic decays, and corrections are applied to the simulated samples to reproduce the efficiencies in data.
Signal jets are jet candidates that are classified in three exclusive categories. Central b-jets satisfy |η| < 2.4 and the b-jet identification criteria. Central light-flavour jets also satisfy |η| < 2.4 but do not satisfy the b-jet identification criteria. If a central light-flavour jet has p T < 50 GeV and has tracks associated to it, at least one of the tracks must originate from the event primary vertex. This criterion removes jets that originate from pile-up interactions. Finally, forward jets are those with 2.4 < |η| < 4.5 and p T > 30 GeV.
The missing transverse momentum, p miss T , is defined [92] as the negative vector sum of the total transverse momenta of all p T > 10 GeV electron, muon and photon candidates, p T > 20 GeV jets, and all clusters of calorimeter energy with |η| < 4.9 not associated to such objects, referred to hereafter as the 'soft-term'. Clusters associated with electrons, photons and jets make use of calibrations of the respective objects, whereas clusters not associated with these objects are calibrated using both calorimeter and tracker information.
The quantity E miss,rel T is defined from the magnitude, E miss T , of p miss T as E miss,rel where ∆φ ℓ,j is the azimuthal angle between the direction of p miss T and that of the nearest electron, muon, central b-jet or central light-flavour jet. Selections based on E miss,rel T aim to suppress events where missing transverse momentum arises from significantly mis-measured jets or leptons.

Event selection
Events are recorded using a combination of two-lepton triggers, which require identification of two lepton (electron or muon) candidates with transverse momenta exceeding a set of thresholds. For all triggers used in this measurement, the p T thresholds are 18-25 GeV for the higher-p T lepton and 8-14 GeV for the other lepton. After event reconstruction, two signal leptons of opposite charge, with p T > 35 GeV and > 20 GeV, are required in the selected events. No lepton candidates other than the two signal leptons are allowed in the event. The two signal leptons are required to match those that triggered the event. The trigger efficiencies with respect to reconstructed leptons with p T in excess of the nominal thresholds have been measured using data-driven techniques. For events containing two reconstructed signal leptons with p T > 35 GeV and > 20 GeV, the average trigger efficiencies are approximately 97% in the e + e − channel, 75% in the e ± µ ∓ channels, and 89% in the µ + µ − channel.
The dilepton invariant mass m ℓℓ must be greater than 20 GeV in all flavour combinations. Events containing one or more τ -jet candidates are rejected.
Seven signal regions (SRs) are defined in this analysis. The first three, collectively referred to as SR-m T2 , are designed to provide sensitivity to sleptons either through direct production or in chargino decays. The next three, SR-W W , are designed to provide sensitivity to chargino-pair production followed by W decays. The last signal region, SR-Zjets, is designed specifically for chargino and second lightest neutralino associated production followed by hadronic W and leptonic Z decays. The SF and DF event samples in each SR are considered separately. When a scenario that contributes to both SF and DF final states is considered, a simultaneous fit to the SF and DF samples is employed. All SRs of the same lepton flavour combination, except for SR-Zjets, overlap with each other and are not statistically independent. Table 1 where p ℓ1 T and p ℓ2 T are the transverse momenta of the two leptons, and q T is a transverse vector that minimizes the larger of the two transverse masses m T . The latter is defined by For SM tt and W W events, in which two W bosons decay leptonically and p miss T originates from the two neutrinos, the m T2 distribution has an upper end-point at the W mass. For signal events, the undetected LSP contributes to p miss T , and the m T2 end-point is correlated to the mass difference between the slepton or chargino and the lightest neutralino. For large values of this difference, the m T2 distribution for signal events extends significantly beyond the distributions of the tt and W W events.

SR-m T2
SR-m T2 targetsχ + 1χ − 1 production followed by slepton-mediated decays (figure 1(a)) and direct slepton pair production ( figure 1(d)). Events are required to contain two oppositesign signal leptons and no signal jets. Only SF channels are used in the search for direct slepton production, while the chargino-to-slepton decay search also uses DF channels. In the SF channels, the dilepton invariant mass m ℓℓ must be at least 10 GeV away from the Z boson mass.
The dominant sources of background are diboson and top production (tt and W t). Three signal regions, SR-m 90 T2 , SR-m 120 T2 and SR-m 150 T2 , are defined by requiring m T2 > 90 GeV, 120 GeV and 150 GeV, respectively. Low values of m T2 threshold provide better sensitivity to cases in which thel orχ ± 1 mass is close to theχ 0 1 mass, and high values target largel-χ 0 1 orχ ± 1 -χ 0 1 mass differences.

SR-W W
Directχ + 1χ − 1 production followed by W -mediated decays (figure 1(b)) is similar to the slepton-mediated scenario, but with smaller visible cross-sections due to the W → ℓν branching fraction. Three signal regions, SR-W W a, SR-W W b and SR-W W c, are designed to provide sensitivities to this scenario for increasing values ofχ ± 1 -χ 0 1 mass difference. Events are required to contain two opposite-sign signal leptons and no signal jets. Both SF and DF channels are used in these signal regions. In the SF channels, the dilepton invariant mass m ℓℓ must be at least 10 GeV away from the Z boson mass.
For largeχ ± 1 -χ 0 1 mass splitting, the m T2 variable provides good discrimination between the signal and SM background. Two signal regions, SR-W W b and SR-W W c, are defined by m T2 > 90 GeV and 100 GeV, respectively. The m T2 thresholds are lower than in SRm T2 because the smaller visible cross-sections limit the sensitivity to largeχ ± 1 masses. For SR-W W b, an additional requirement of m ℓℓ < 170 GeV is applied to further suppress the SM background.
For cases in which theχ ± 1 -χ 0 1 mass splitting is close to the W boson mass, the m T2 variable is not effective in distinguishing signal from the SM W W production. The signal region SR-W W a is defined by E miss,rel T > 80 GeV, p T,ℓℓ > 80 GeV and m ℓℓ < 120 GeV, where p T,ℓℓ is the transverse momentum of the lepton pair. These selection criteria favour events in which the di-lepton opening angle is small, which enhances the difference in the E miss,rel T distribution between the signal and the background due to the two LSPs in the signal.

SR-Zjets
The last signal region, SR-Zjets, differs from the previous six in that it requires the presence of at least two central light jets. This signal region is designed to target the pp →χ ± 1χ 0 2 → W ±χ0 1 Zχ 0 1 process in which the W boson decays hadronically and the Z boson decays leptonically (figure 1(c)).
The two highest-p T central light jets must have p T > 45 GeV, and have an invariant mass in the range 50 < m jj < 100 GeV. There must be no central b-jet and no forward jet in the event. The two opposite-sign leptons must be SF, and their invariant mass must be within 10 GeV of the Z boson mass.
To suppress large background from the SM Z + jets production, E miss,rel T > 80 GeV is required. Events are accepted only if the reconstructed Z boson is recoiling against the rest of the event with a large transverse momentum p T,ℓℓ > 80 GeV, and the separation ∆R ℓℓ between the two leptons must satisfy 0.3 < ∆R ℓℓ < 1.5.

Background estimation
For SR-m T2 and SR-W W , the SM background is dominated by W W diboson and topquark (tt and W t) production. Contributions from ZV production, where V = W or Z, are also significant in the SF channels. The MC predictions for these background sources are normalized in dedicated control regions (CR) for each background, as described in section 7.1. For SR-Zjets, the dominant sources of background are ZV production and Z/γ * + jets. The former is estimated from simulation, validated using ZV -enriched control samples, and the latter is estimated by a data-driven technique, as described in section 7.2. The top-quark background in SR-Zjets is estimated using a dedicated CR. Background due to hadronic jets mistakenly reconstructed as signal leptons or real leptons originating from heavy-flavour decays or photon conversions, referred to as 'non-prompt Table 2. Control region definitions. The top CR for SR-Zjets requires at least two jets with p T > 20 GeV in |η| < 2.4, at least one of which is b-tagged.
leptons', is estimated using a data-driven method described in section 7.3. Contributions from remaining sources of SM background are small and are estimated from simulation. Table 2 summarizes the definitions of the control regions.

Background in SR-m T2 and SR-W W
The normalization factors for the background in SR-m T2 and SR-W W due to the SM W W , top and ZV production are constrained in dedicated CRs for each background. Each CR is dominated by the background of interest and is designed to be kinematically as close as possible to a corresponding signal region. The normalization factors are obtained from the likelihood fit described in section 7.4. The W W control region for SR-m T2 and SR-W W b/c is defined by requiring 50 < m T2 < 90 GeV and the events must contain no jets. Only the DF sample is used in this CR because the corresponding regions in the SF samples suffer from contamination from Z/γ * + jets background. Appropriate ratios of electron and muon efficiencies are used to obtain the SF background estimations from the corresponding DF CR. For SR-W W a, the CR is defined by lowering the E miss,rel T and p T,ℓℓ requirements so that 60 < E miss,rel T < 80 GeV and p T,ℓℓ > 40 GeV. Figure 2(a) shows the m T2 distribution in this CR. The normalization factors are not applied to the MC predictions in all four plots of figure 2. Predicted signal contamination in this CR is less than 10% for the signal models > 100 GeV. The top control region for SR-m T2 and SR-W W b/c is also defined using the DF sample, and by requiring at least one b-tagged jet and vetoing central light jets and forward jets. The events must also satisfy m T2 > 70 GeV. Figure 2(b) shows the E miss,rel T distribution in this CR. For SR-W W a, the CR is defined using the DF sample and requiring at least one b-tagged jet, with all the other SR criteria unchanged. The predicted contamination from SUSY signal is negligible for the models considered. The ZV control region for SR-m T2 and SR-W W b/c is defined identically to the SF SR-m 90 T2 , but with the Z veto reversed. Figure 2(c) shows the E miss,rel T distribution in this CR. The contamination due to non-ZV sources is dominated by W W events (4.5%). For SR-W W a, the CR is defined by reversing the Z veto in the SF sample. The predicted contamination from SUSY signal is less than 5% in these CRs.

Background in SR-Zjets
The top CR for SR-Zjets is defined by reversing the Z veto and requiring at least one b-tagged jet. To increase the statistics of the sample, the p T threshold for the central jets is lowered to 20 GeV, and no cut on m jj is applied. Figure 2(d) shows the m jj distribution in this CR. The predicted contamination from SUSY signal is negligible.
The ZV background in SR-Zjets consists of diboson production accompanied by two light-flavour jets, that is, W Zjj → ℓνℓ ′ ℓ ′ jj, where the lepton from the W decay was not reconstructed, and ZZjj → ℓℓννjj. The contribution from ZV → ℓℓqq is strongly suppressed by the E miss,rel T requirement. This background is estimated from simulation, and validated in control samples of W Zjj → ℓνℓ ′ ℓ ′ jj and ZZjj → ℓℓℓ ′ ℓ ′ jj where all leptons are reconstructed. The W Zjj enriched control sample consists of events with three leptons, at least two of which make up a SF opposite-sign pair with an invariant mass within 10 GeV of the Z boson mass. In addition, events must have E miss T > 30 GeV, m T > 40 GeV computed from the p miss T and the lepton that was not assigned to the Z boson, at least two central light jets, and no central b-jet. The predicted contamination from SUSY signal is less than 10% in this region. The ZZjj enriched control sample consists of events with two pairs of same-flavour opposite-sign leptons, each with an invariant mass within 10 GeV of the Z boson mass, E miss T < 50 GeV, at least two central light jets, and no signal b-jet. The data in these control samples are compared with the simulation to assess the systematic uncertainties of the ZV background estimation, as reported in section 8.
In SR-Zjets, Z/γ * + jets events are an important source of background, where significant E miss T arises primarily from mis-measurement of jet transverse momentum. A data-driven approach called the 'jet smearing' method is used to estimate this background. In this method, a sample enriched in Z/γ * + jets events with well-measured jets is selected from data as seed events. The seed events are selected by applying the SR-Zjets event selection, but reversing the E miss,rel T cut. To ensure that the events only contain well measured jets, the ratio E miss is the scalar sum of the transverse energies of the jets and the soft-term, is required to be less than 1.5 (GeV) 1/2 . Each seed event is smeared by multiplying each jet four-momentum by a random number drawn from the jet response function, which is initially estimated from simulation and adjusted after comparing the response to data in a photon + jet sample. In addition, the contribution to E miss T due to the soft-term is also modified by sampling randomly from the soft-term distribution measured in a Z → ℓℓ sample with no reconstructed jets. The smearing procedure is repeated 10,000 times for each seed event. The resulting pseudo-data E miss,rel T distribution is then normalized to the data in the region of E miss,rel T < 40 GeV, and the migration into the signal region is evaluated.
To validate the jet-smearing method, a control sample is selected with the same selection criteria as SR-Zjets but reversing the p T,ℓℓ requirement, and removing the ∆R ℓℓ and m jj criteria to increase the number of events. The seed events are selected from the control region events by requiring E miss,rel T < 40 GeV and E miss T / E sum T < 1.5 (GeV) 1/2 . Results are validated in a region with 40 < E miss,rel T < 80 GeV, which is dominated by Z/γ * + jets. The method predicts 750 ± 100 events, where both statistical and systematic uncertainties are included, in agreement with the 779 events observed in data.

Non-prompt lepton background estimation
The term 'non-prompt leptons' refers to hadronic jets mistakenly reconstructed as signal leptons or leptons originating from heavy-flavour decays or photon conversions. In this context, 'prompt leptons' are leptons produced directly in decays of sparticles or weak bosons. The number of non-prompt lepton events is estimated using the matrix method [95], which takes advantage of the difference between the prompt efficiency ǫ p and non-prompt efficiency ǫ n , defined as the fractions of prompt and non-prompt candidate leptons, respectively, that pass the signal-lepton requirements.
The prompt and non-prompt efficiencies are evaluated as functions of the p T of the lepton candidate in simulated events using MC truth information. Differences between data and MC are corrected for with normalization factors measured in control samples. Since the efficiencies depend on the production process, average ǫ p and ǫ n values are calculated for each SR and CR using the fraction of each process predicted by the simulation as the weights. The data/MC normalization factors for ǫ p are derived from Z → ℓℓ events. The normalization factors for ǫ n depend on whether the non-prompt lepton originated from jets or from photon conversion. The normalization factors for misidentified jets or leptons from heavy-flavour decays are measured in a control region enriched in bb production. Events are selected with two candidate leptons, one b-tagged jet and E miss,rel T < 40 GeV. One of the two lepton candidates is required to be a muon and to lie within ∆R = 0.4 of the b-tagged jet, while the other lepton candidate is used to measure the non-prompt efficiency. For measuring the normalization factor for photon conversions, a Z → µµγ control sample is defined by selecting events with two muons, E miss,rel T < 50 GeV, at least one candidate electron (which is the conversion candidate) with m T < 40 GeV, and requiring that the invariant mass of the µ + µ − e ± system is within 10 GeV of the Z boson mass.
Using ǫ n and ǫ p , the observed numbers of events in each SR and CR with four possible combinations (signal-signal, signal-candidate, candidate-signal and candidate-candidate) of leptons are expressed as weighted sums of the numbers of events with four combinations of prompt and non-prompt leptons. Solving these equations allows determination of the non-prompt lepton background. The contribution of non-prompt-lepton background in the signal regions is less than 5% of the total background in all signal regions.

Fitting procedure
For each SR, a simultaneous likelihood fit to the corresponding CRs is performed to normalize the top, W W and ZV (in the case of SR-Zjets only top is fitted) background estimates. The inputs to the fit are the numbers of observed events in the CRs, the expected contributions of top, W W and ZV from simulation, and the expected contributions of other background sources determined as described in sections 7.1-7.3.
The event count in each CR is treated as a Poisson probability function, the mean of which is the sum of the expected contributions from all background sources. The free parameters in the fit are the normalization of the top, W W and ZV contributions. The systematic uncertainties on the expected background yields are included as nuisance parameters, constrained to be Gaussian with a width determined from the size of the uncertainty. Correlations between control and signal regions, and background processes are taken into account with common nuisance parameters. The free parameters and the nuisance parameters are determined by maximizing the product of the Poisson probability functions and the constraints on the nuisance parameters.  Table 3 summarizes the numbers of observed and predicted events in the CRs, data/MC normalization and CR composition obtained from the simultaneous fit. The normalization factors agree within errors between different SRs for each of the W W , Top and ZV contributions. Results of the background estimates in the SRs can be found in Tables 5, 6 and 7.

Systematic uncertainties
Systematic uncertainties affect the estimates of the backgrounds and signal event yields in the control and signal regions. The relative sizes of the sources of systematic uncertainty on the total SM background in SR-m T2 , SR-W W and SR-Zjets are detailed in Table 4.
The 'CR statistics' and 'MC statistics' uncertainties arise from the number of data events in the CRs and simulated events in the SRs and CRs, respectively. The largest contributions are due to the simulated background samples in the signal regions.
The dominant experimental systematic uncertainties, labelled 'Jet' in Table 4, come from the propagation of the jet energy scale calibration [96] and resolution [97] uncertainties. They were derived from a combination of simulation, test-beam data and in situ measurements. Additional uncertainties due to differences between quark and gluon jets, and light and heavy flavour jets, as well as the effect of pile-up interactions are included. The 'Lepton' uncertainties include those from lepton reconstruction, identification and trigger efficiencies, as well as lepton energy and momentum measurements [83, 84]. Uncertainties due to τ reconstruction and energy calibration are negligible. Jet and lepton energy scale uncertainties are propagated to the E miss T evaluation. An additional 'Softterm' uncertainty is associated with the contribution to the E miss T reconstruction of energy deposits not assigned to any reconstructed objects [92].
The 'b-tagging' row refers to the uncertainties on the b-jet identification efficiency and charm and light-flavour jet rejection factors [98]. The 'Non-prompt lepton' uncertainties Table 4. Systematic uncertainties (in %) on the total background estimated in different signal regions. Because of correlations between the systematic uncertainties and the fitted backgrounds, the total uncertainty can be different from the quadratic sum of the individual uncertainties. arise from the data-driven estimates of the non-prompt lepton background described in section 7.3. The dominant sources are η dependencies of the non-prompt rates, differences between the light and heavy flavour jets, and the statistics of the control samples. The uncertainty on the integrated luminosity is ±2.8%, and affects the normalization of the background estimated with simulation. It is derived following the methodology detailed in ref.
[99]. The 'Modelling' field of Table 4 includes the uncertainties on the methods used for the background estimate, as well as the modelling uncertainties of the generators used to assist the estimate. For SR-Zjets an additional 20% uncertainty is assigned to the ZV background estimate to account for the variations between data and simulation in the ZV control regions with two or more jets, as described in section 7.2. Uncertainties on the Z/γ * + jets background estimate in SR-Zjets include the systematic uncertainties associated with the jet smearing method due to the fluctuations in the non-Gaussian tails of the response function and the systematic uncertainty associated with the cut value on E miss T / E sum T used to define the seed region. The effect of using each seed event multiple times is also taken into account. Generator modelling uncertainties are estimated by comparing the results from POWHEG and MC@NLO generators for top events, and POWHEG and aMC@NLO for W W events, using HERWIG for parton showering in all cases. Parton showering uncertainties are estimated in top and W W events by comparing POWHEG plus HERWIG with POWHEG plus PYTHIA. Both generator modelling and parton showering uncertainties are estimated in ZV events by comparing POWHEG plus PYTHIA to SHERPA. Special tt samples are generated using AcerMC with PYTHIA to evaluate the uncertainties related to the amount of initial and final-state radiation [100]. Impact of the choice of renormalization and factorization scales is evaluated by varying them between 0.5 and 2 times the nominal values in POWHEG for top events and aMC@NLO for diboson events. The uncertainties due to the PDFs for the top and diboson events are evaluated using 90% C.L. CT10 PDF eigenvectors.
Effects of using different PDF sets have been found to be negligible. The dominant contribution among the 'Modelling' uncertainties comes from the difference between POWHEG and aMC@NLO for diboson production.
Signal cross-sections are calculated to NLO in the strong coupling constant. Their uncertainties are taken from an envelope of cross-section predictions using different PDF sets and factorization and renormalization scales, as described in ref. [101]. Systematic uncertainties associated with the signal selection efficiency include those due to lepton trigger, reconstruction and identification, jet reconstruction and E miss T calculation. Uncertainties on the integrated luminosity affect the predicted signal yield. The total uncertainty on the predicted signal yield is typically 9-13% for SUSY scenarios to which this measurement is sensitive. Figures 3 and 4 show the comparison between data and the SM prediction for key kinematic variables in different signal regions. In each plot, the expected distributions from the W W , tt and ZV processes are corrected with data-driven normalization factors obtained from the fit detailed in section 7. The hashed regions represent the sum in quadrature of systematic uncertainties and statistical uncertainties arising from the numbers of MC events. The effect of limited data events in the CR is included in the systematic uncertainty. All statistical uncertainties are added in quadrature whereas the systematic uncertainties are obtained after taking full account of all correlations between sources, background contributions and channels. The rightmost bin of each plot includes overflow. Illustrative SUSY benchmark models, normalized to the integrated luminosity, are superimposed. The lower panel of each plot shows the ratio between data and the SM background prediction.

Results
Tables 5, 6 and 7 compare the observed yields in each signal region with those predicted for the SM background. The errors include both statistical and systematic uncertainties. Good agreement is observed across all channels.
For each SR, the significance of a possible excess over the SM background is quantified by the one-sided probability, p 0 , of the background alone to fluctuate to the observed number of events or higher, using the asymptotic formula [102]. This is calculated using a fit similar to the one described in section 7.4, but including the observed number of events in the SR as an input. The accuracy of the limits obtained from the asymptotic formula was tested for all SRs by randomly generating a large number of pseudo data sets and repeating the fit. Upper limits at 95% CL on the number of non-SM events for each SR are derived using the CL s prescription [103] and neglecting any possible contamination in the CRs. Normalizing these by the integrated luminosity of the data sample they can be interpreted as upper limits, σ 95 vis , on the visible non-SM cross-section, defined as the product of acceptance, reconstruction efficiency and production cross-section of the non-SM contribution. All systematic uncertainties and their correlations are taken into account via nuisance parameters. The results are given in Tables 5, 6

Interpretation
Exclusion limits at 95% confidence-level are set on the slepton, chargino and neutralino masses within the specific scenarios considered. The same CL s limit-setting procedure as in section 9 is used, except that the SUSY signal is allowed to populate both the signal region and the control regions as predicted by the simulation. Since the SRs are not mutually exclusive, the SR with the best expected exclusion limit is chosen for each model point.
The results are displayed in figures 5 through 9. In each exclusion plot, the solid (dashed) lines show observed (expected) exclusion contours, including all uncertainties except for the theoretical signal cross-section uncertainty arising from the PDF and the renormalization and factorization scales. The solid band around the expected exclusion contour shows the ±1σ result where all uncertainties, except those on the signal crosssections, are considered. The dotted lines around the observed exclusion contour represent the results obtained when varying the nominal signal cross-section by ±1σ theoretical uncertainty. All mass limits hereafter quoted correspond to the signal cross-sections reduced by 1σ. Figure 5 shows the 95% CL exclusion region obtained from SR-m T2 on the simplified Table 5. Observed and expected numbers of events in SR-m T2 . Also shown are the one-sided p 0 values and the observed and expected 95% CL upper limits, σ 95 vis , on the visible cross-section for non-SM events. The 'Others' background category includes non-prompt lepton, Z/γ * + jets and SM Higgs. The numbers of signal events are shown for theχ + 1χ − 1 → (lν or ℓν)χ 0 1 (l ′ ν ′ or ℓ ′ν′ )χ 0 1 scenario and for thel +l− → ℓ +χ0 1 ℓ −χ0 1 scenario with differentχ ± 1 ,χ 0 1 andl masses in GeV.     Figure 5. Observed and expected 95% CL exclusion regions in the mχ ± 1 -mχ 0 1 plane for simplifiedmodelχ + 1χ − 1 pair production with common masses of sleptons and sneutrinos at ml = mν = (mχ ± 1 + mχ 0 1 )/2. Also shown is the LEP limit [36] on the mass of the chargino. The blue line indicates the limit from the previous analysis with the 7 TeV data [34]. model for directχ + 1χ − 1 pair production followed by slepton-mediated decays. For mχ 0 1 = 0, chargino masses between 140 GeV and 465 GeV are excluded. The exclusion in this scenario depends on the assumed slepton mass, which is chosen to be halfway between theχ ± 1 andχ 0 1 masses in this analysis. Studies performed with particle-level signal MC samples show that the signal acceptance in SR-m T2 depends weakly on ml, and the choice of ml = (mχ ± 1 + mχ 0 1 )/2 minimizes (maximizes) the acceptance for small (large)χ ± 1 -χ 0 1 mass splitting. Figure 6(a) shows the 95% CL exclusion regions obtained from SR-W W on the simplifiedmodelχ + 1χ − 1 production followed by W -mediated decays. Figure 6(b) shows the observed and expected 95% CL upper limits on the SUSY signal cross-section normalized by the simplified-model prediction as a function of mχ ± 1 for a masslessχ 0 1 . For mχ 0 1 = 0, chargino mass ranges of 100-105 GeV, 120-135 GeV and 145-160 GeV are excluded at 95% CL. Figure 7(a) shows the 95% CL exclusion region obtained from SR-Zjets in the simplifiedmodelχ ± 1χ 0 2 production followed by W and Z decays. For mχ 0 1 = 0, degenerateχ ± 1 and χ 0 2 masses between 180 GeV and 355 GeV are excluded. Figure 7(b) shows the exclusion region obtained by combining this result with results from the relevant signal regions (SR0a/SR1a/SR1SS/SR2a) in the ATLAS search for electroweak SUSY production in the three-lepton final states [82]. The fit is performed on the combined likelihood function using all signal regions. The uncertainties are profiled in the likelihood and correlations between channels and processes are taken into account. The combination significantly improves the sensitivity. As a result, degenerateχ ± 1 andχ 0 2 masses between 100 GeV and 415 GeV are excluded at 95% CL for mχ 0 1 = 0. Figure 8 shows the 95% CL exclusion regions obtained from SR-m T2 for the direct production of (a) right-handed,  The CL s value is also calculated from SR-W W a for the GMSB model point where the chargino is the NLSP with mχ ± 1 = 110 GeV, mχ 0 1 = 113 GeV and mχ 0 2 = 130 GeV [46]. The observed and expected CL s values are found to be 0.19 and 0.29, respectively. The observed and expected 95% CL limits on the signal cross-section are 1.58 and 1.90 times the model prediction, respectively. lightest charginoχ ± 1 , next-to-lightest neutralinoχ 0 2 and sleptons for different masses of the lightest neutralinoχ 0 1 in simplified models. In the scenario ofχ + 1χ − 1 pair production with χ ± 1 decaying intoχ 0 1 via an intermediate slepton with mass halfway between theχ ± 1 and χ 0 1 ,χ ± 1 masses between 140 GeV and 465 GeV are excluded at 95% CL for a masslessχ 0 1 . In the scenario ofχ + 1χ − 1 pair production withχ ± 1 decaying intoχ 0 1 and a W boson,χ ± 1 masses in the ranges 100-105 GeV, 120-135 GeV and 145-160 GeV are excluded at 95% CL for a masslessχ 0 1 . This is the first limit for this scenario obtained at a hadron collider. Finally, in the scenario ofχ ± 1χ 0 2 production withχ ± 1 decaying into Wχ 0 1 andχ 0 2 decaying into Zχ 0 1 , commonχ ± 1 andχ 0 2 masses between 180 GeV and 355 GeV are excluded at 95% CL for a masslessχ 0 1 . Combining this result with those from ref.

Conclusion
[82] extends the exclusion region to between 100 GeV and 415 GeV. In scenarios where sleptons decay directly intõ χ 0 1 and a charged lepton, common values for left and right-handed slepton masses between 90 GeV and 325 GeV are excluded at 95% CL for a masslessχ 0 1 . Improved exclusion regions are also obtained in the pMSSM µ-M 2 plane for four sets of slepton mass, M 1 and tan β

Acknowledgements
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.
We [27] M. Dine  [82] ATLAS Collaboration, Search for direct production of charginos and neutralinos in events with three leptons and missing transverse momentum in √ s = 8 TeV pp collisions with the ATLAS detector, Submitted to JHEP [arXiv:1402.7029].
[100] ATLAS Collaboration, Measurement of tt production with a veto on additional central jet activity in pp collisions at √ s = 7 TeV using the ATLAS detector, Eur. Phys. J. C72