Search for direct top squark pair production in final states with two leptons in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{s} = 13$$\end{document}s=13 TeV pp collisions with the ATLAS detector

The results of a search for direct pair production of top squarks in events with two opposite-charge leptons (electrons or muons) are reported, using \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$36.1~\hbox {fb}^{-1}$$\end{document}36.1fb-1 of integrated luminosity from proton–proton collisions at \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sqrt{s}=13$$\end{document}s=13 TeV collected by the ATLAS detector at the Large Hadron Collider. To cover a range of mass differences between the top squark \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t}$$\end{document}t~ and lighter supersymmetric particles, four possible decay modes of the top squark are targeted with dedicated selections: the decay \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t} \rightarrow b \tilde{\chi }_{1}^{\pm }$$\end{document}t~→bχ~1± into a b-quark and the lightest chargino with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\chi }_{1}^{\pm } \rightarrow W \tilde{\chi }_{1}^{0}$$\end{document}χ~1±→Wχ~10, the decay \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t} \rightarrow t \tilde{\chi }_{1}^{0}$$\end{document}t~→tχ~10 into an on-shell top quark and the lightest neutralino, the three-body decay \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t} \rightarrow b W \tilde{\chi }_{1}^{0}$$\end{document}t~→bWχ~10 and the four-body decay \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t} \rightarrow b \ell \nu \tilde{\chi }_{1}^{0}$$\end{document}t~→bℓνχ~10. No significant excess of events is observed above the Standard Model background for any selection, and limits on top squarks are set as a function of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t}$$\end{document}t~ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{\chi }_{1}^{0}$$\end{document}χ~10 masses. The results exclude at 95% confidence level \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\tilde{t}$$\end{document}t~ masses up to about 720 GeV, extending the exclusion region of supersymmetric parameter space covered by previous searches.


Introduction
The Standard Model (SM) of particle physics is extremely successful in describing the phenomena of elementary particles and their interactions. Nevertheless, it is believed to be only a low-energy realisation of a more general theory. In its current form, it fails to explain several observations, such as the nature of dark matter, the baryon asymmetry of the universe and the stabilisation of the Higgs boson mass against radiative corrections from the Planck scale. These shortcomings could be remedied by the existence of new particles at the TeV scale, which motivates extensive searches at the Large Hadron Collider (LHC).
One of the most compelling theories beyond the SM is Supersymmetry (SUSY) [1][2][3][4][5][6]. SUSY is a spacetime symmetry that for each SM particle postulates the existence of a partner particle whose spin (S ) differs by one-half unit. The introduction of gauge-invariant and renormalisable interactions into SUSY models can violate the conservation of baryon number (B) and lepton number (L), resulting in a proton lifetime shorter than current experimental limits [7]. This is usually solved by assuming that the multiplicative quantum number R-parity [8], defined as R = (−1) 3(B−L)+2S , is conserved.
In the framework of a generic R-parity-conserving model, SUSY particles are produced in pairs, and the lightest supersymmetric particle (LSP) is stable and a candidate for dark matter [9,10]. The scalar partners of right-handed and left-handed quarks (squarks),q R andq L , can mix to form two mass eigenstates, q 1 andq 2 , withq 1 defined to be the lighter one. In the case of the supersymmetric partner of the top quark,t, large mixing effects can lead to one top squark mass eigenstate,t 1 , that is significantly lighter than the other squarks. The charginos and neutralinos are mixtures of the bino, winos and Higgsinos that are superpartners of the U(1) and 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 order of increasing masses. In a large variety of models, the LSP is the lightest neutralinoχ 0 1 .
In this paper a search for direct pair production of the top squark is reported, in final states with two isolated leptons (electrons or muons) and missing transverse momentum. The search utilises 36.1 fb −1 of proton-proton collision data collected by the ATLAS experiment in 2015 and 2016 at a centre-of-mass energy √ s = 13 TeV.
The top squark is assumed to decay into either the lightest chargino or the lightest neutralino. Depending on the mass difference between the top squark and the lighter SUSY particles, different decay modes are relevant. The decayst → tχ 0 1 andt → bχ ± 1 (where t and b represent either the quark or the anti-quark, depending on the charge conjugation) withχ is considered. For smaller mass differences, the four-body decay channelt → b f f χ 0 1 , where f and f are two fermions from the W * decay, is assumed to occur. In this search, f and f are a lepton and its associated neutrino. For each of these decay modes, shown by the diagrams in Figure 1, a dedicated event selection is performed to optimise the search significance, as detailed in Table 1.

ATLAS detector
The ATLAS detector [33] at the LHC is a multi-purpose particle detector with a cylindrical forwardbackward symmetric geometry 1 and an approximate 4π coverage in solid angle. It consists of an inner tracking detector (ID) surrounded by a thin superconducting solenoid providing a 2 T axial magnetic field, electromagnetic and hadron calorimeters, and a muon spectrometer. The inner tracking detector covers the pseudorapidity range |η| < 2.5. It consists of silicon pixel, silicon microstrip, and transition radiation tracking detectors. The newly installed innermost layer of pixel sensors [34] was operational for the first time during the 2015 data-taking. Lead/liquid-argon (LAr) sampling calorimeters provide electromagnetic (EM) energy measurements with high granularity. A hadron (steel/scintillator-tile) calorimeter covers the central pseudorapidity range (|η| < 1.7). The end-cap and forward regions are instrumented with LAr calorimeters for both the EM and hadronic energy measurements up to |η| = 4.9. The muon spectrometer surrounds the calorimeters and features three large air-core toroid superconducting magnets with eight coils each. It includes a system of precision tracking chambers and fast detectors for triggering. The field integral of the toroids ranges between 2.0 and 6.0 Tm across most of the detector.

Data samples and event reconstruction
The data were collected by the ATLAS detector in 2015 and 2016 during pp collisions at a centre-of-mass energy of √ s = 13 TeV, with a peak instantaneous luminosity of L = 1.4×10 34 cm −2 s −1 , a bunch spacing of 25 ns, and an average number of pp interactions per bunch crossing (pile-up) of µ = 14 in 2015 and µ = 24 in 2016. Only events taken in stable beam conditions, and for which all relevant detector 1 ATLAS uses a right-handed coordinate system with its origin at the nominal interaction point (IP) in the centre of the detector and the z-axis along the beam pipe. The x-axis points from the IP to the centre of the LHC ring, and the y-axis points upwards. Cylindrical coordinates (r, φ) are used in the transverse plane, φ being the azimuthal angle around the z-axis. The pseudorapidity is defined in terms of the polar angle θ as η = − ln tan(θ/2). Rapidity is defined as y = 0.5 ln (E + p z ) / (E − p z ) where E denotes the energy and p z is the component of the momentum along the beam direction.
systems were operational, are considered in this analysis. The integrated luminosity of the resulting data set is 36.1 fb −1 , with an uncertainty of ±3.2%. This uncertainty is derived, following a methodology similar to that detailed in Ref.
[35], from a preliminary calibration of the luminosity scale using x-y beam-separation scans performed in August 2015 and May 2016.
Candidate events are required to have a reconstructed vertex with at least two associated tracks with transverse momentum p T > 400 MeV. The vertex with the highest scalar sum of the squared transverse momenta of the associated tracks is considered the primary vertex of the event.
Electron (baseline) candidates are reconstructed from three-dimensional electromagnetic calorimeter energy depositions matched to ID tracks, and are required to have pseudorapidity |η| < 2.47, p T > 7 GeV, and to pass a loose likelihood-based identification requirement [36]. The likelihood input variables include measurements of calorimeter shower shapes and of track properties from the ID.
Muon (baseline) candidates are reconstructed in the pseudorapidity region |η| < 2.4 from muon spectrometer tracks matching ID tracks. They must have p T > 7 GeV and must pass the medium identification requirements defined in Ref. [37], which are based on requirements on the number of hits in the different ID and muon spectrometer subsystems, and on the significance of the charge-to-momentum ratio (q/p) measurement [37]. The "medium" working point is used for the pile-up rejection, which has an efficiency of about 92% for jets produced by the hard scatter. Jets resulting from the hadronisation of b-quarks are identified using a multivariate b-tagging algorithm (MV2c10), which is based on quantities such as impact parameters of associated tracks and reconstructed secondary vertices [45,46]. This algorithm is used at a working point that provides 77% b-tagging efficiency in simulated tt events, and a rejection factor of 134 for light-quark flavours and gluons and 6 for charm jets. The jets satisfying the b-tagging requirements are referred to as b-jets.
Events are discarded if they contain any jet with p T > 20 GeV failing to satisfy basic quality selection criteria that reject detector noise and non-collision backgrounds [47].
To resolve reconstruction ambiguities, an overlap removal algorithm is applied to candidate leptons and jets. Non-b-tagged jets which lie within ∆R = (∆y) 2 + (∆φ) 2 < 0.2 (here y stands for the rapidity) from an electron candidate are removed, and the same is done for jets which lie close to a muon candidate and are consistent with the characteristics of jets produced by muon bremsstrahlung. Finally, any lepton candidate which lies within ∆R < 0.4 from the direction of a surviving jet candidate is removed, in order to reject leptons from the decay of a b-or c-hadron. Electrons which share an ID track with a muon candidate are also removed.
Additional selections are then applied to the remaining lepton and jet candidates. Tighter requirements on the lepton candidates are imposed, which are then referred to as "signal" electrons or muons. Signal electrons must satisfy the medium likelihood-based identification requirement as defined in Ref.
[36]. Signal electrons must have a transverse impact parameter with respect to the reconstructed primary vertex, d 0 , with a significance of |d 0 |/σ(d 0 ) < 5. For signal muons, the corresponding requirement is |d 0 |/σ(d 0 ) < 3. The tracks associated with the signal leptons must have a longitudinal impact parameter with respect to the reconstructed primary vertex, z 0 , satisfying |z 0 sin θ| < 0.5 mm. Isolation criteria are applied to both electrons and muons by placing an upper limit on the sum of the transverse energy of the calorimeter energy clusters in a cone of ∆R η = (∆η) 2 + (∆φ) 2 = 0.2 around the electron (excluding the deposit from the electron itself), and the scalar sum of the p T of tracks within a variable-size cone around the lepton (excluding its own track). The track isolation cone radius for electrons (muons) is given by the smaller of ∆R = 10 GeV/p T and ∆R η = 0.2 (0.3). The isolation criteria are optimised such that the isolation selection efficiency is uniform across η, and it increases from 95% for p T = 25 GeV to 99% for p T = 60 GeV in Z → events.
The missing transverse momentum (p miss T ), whose magnitude is denoted by E miss T , is defined as the negative vector sum of the transverse momenta of all identified baseline objects (electrons, muons, jets) and an additional soft term. The soft term is constructed from all tracks that are not associated with any reconstructed electron, muon or jet, but which are associated with the primary vertex. In this way, the E miss T value is adjusted for the best calibration of the jets and the other identified objects above, while maintaining pile-up independence in the soft term [48, 49].

Event selection
For the two-body and three-body selections, events are accepted if they pass an online selection (trigger) requiring a minimum of two electrons, two muons or an electron and a muon matched to the trigger objects. The offline selection requires that the leading lepton has a p T larger than 25 GeV and the subleading lepton a p T larger than 20 GeV, ensuring that trigger efficiencies are constant in the relevant phase space. The four-body selection accepts events passing an E miss T -based trigger and having offline E miss T > 200 GeV. This ensures that the trigger efficiency is constant in the relevant phase space. Using this trigger permits the use of a reduced lepton p T threshold of 7 GeV, increasing acceptance for the low lepton p T produced in the four-bodyt → b νχ 0 1 decay.
Events are required to have exactly two signal leptons which must be of opposite charge (electrons, muons, or one of each) with an invariant mass (regardless of the flavour of the leptons in the pair) m greater than 20 GeV (10 GeV for the four-body selection) in order to remove leptons from low-mass resonances. Except for the four-body selection, events with same-flavour (SF) lepton pairs with m between 71.2 and 111.2 GeV are rejected, in order to reduce the backgrounds with leptons produced by Z bosons. No additional selection is applied to the m value of different-flavour (DF) lepton pairs. In the following, the requirements described in the preceding part of this section are referred to as "common selection".

Discriminators and kinematic variables
For the different decay modes considered, dedicated sets of discriminating variables are used to separate the signal from the SM backgrounds.
The missing transverse momentum and the p T of the leading leptons and jets are used to define three useful ratio variables : where p T ( 1 ) and p T ( 2 ) are the leading and subleading lepton transverse momenta and p T ( j i=1,...,N≤4 ) are the transverse momenta in decreasing order of up to the four leading jets. The variables R 2 2 j and R 2 are used to reject backgrounds, e.g. Z/γ * + jets, which peak at lower values than the signal. Similarly, R 2 4 j is a powerful discriminant against multi-jet events.
Other variables employed are : p T,boost : defined as the vector The p T,boost variable, with magnitude p T,boost , can be interpreted as the opposite of the vector sum of all the transverse hadronic activity in the event.
-∆φ boost : the azimuthal angle between the p miss T vector and the p T,boost vector.
-∆x: defined as where E CM = 13 TeV is used and p z ( 1 ),p z ( 2 ) are respectively the leading and subleading lepton longitudinal momenta. This variable helps to discriminate between gluon-and quark-initiated processes. The former tend to peak towards zero, while the latter tend to peak at higher values.
-cos θ b : the cosine of the angle between the direction of motion of either of the two leptons and the beam axis in the centre-of-mass frame of the two leptons [50]. This variable is sensitive to the spin of the pair-produced particle, providing additional rejection against diboson backgrounds.
m T2 : lepton-based "stransverse" mass. The stransverse mass defined in Refs. [51,52] is a kinematic variable used to bound the masses of a pair of identical particles which have each decayed into a visible and an invisible particle. This quantity is defined as where m T indicates the transverse mass, 2 p T,1 and p T,2 are the transverse momentum vectors of two particles, and q T,1 and q T,2 are transverse momentum vectors with q T = q T,1 + q T,2 . The minimisation is performed over all the possible decompositions of q T . For tt or WW decays with t → b ν and W → ν, when the transverse momenta of the two leptons in each event are taken as p T,1 and p T,2 , and p miss T as q T , m T2 (p T ( 1 ), p T ( 2 ), p miss T ) is bounded sharply from above by the mass of the W boson [53,54]. In thet → bχ ± 1 decay mode the upper bound is strongly 2 The transverse mass is defined by the equation m T (p T , q T ) = 2|p T ||q T |(1 − cos(∆φ)), where ∆φ is the angle between the particles of negligible mass with transverse momenta p T and q T . correlated with the mass difference between the chargino and the lightest neutralino. In this paper, m T2 (p T ( 1 ), p T ( 2 ), p miss T ) is referred to simply as m T2 .
The three-body selection uses a number of "super-razor" variables that are defined in Ref. [55]. They are designed to identify events with two massive parent particles (i.e. top squarks) each decaying into a set of visible (only leptons are considered in this case, all other particles including jets are ignored) and invisible particles (i.e. neutrinos and neutralinos). These variables are: -R p T : defined as where J T is the vector sum of the transverse momenta of the visible particles and the missing transverse momentum, and √ŝ R is a measure of the system's energy in the razor frame R as defined in Ref. [55] as the frame in which the two visible leptons have equal and opposite p z . In the case where all possible visible particles are considered, the razor frame R becomes an approximation of the pair production centre-of-mass frame with the centre-of-mass energy √ŝ R . In this analysis, only leptons are considered in the visible system. Therefore, R p T tends towards zero in events that do not contain additional activity (i.e. dibosons) due to vanishing | J T |, whereas in events that contain additional activity (i.e. tt) this variable tends towards unity, thus providing separation power between the two cases.
γ R+1 : The Lorentz factor associated with the boosts from the razor frame R to the approximations of the two decay frames of the parent particles. It is a measure of how the two visible systems are distributed, tending towards unity when the visible particles are back-to-back or have different momenta, while preferring lower values when they are equal in momenta and collinear.
-M R ∆ : defined as This variable has a kinematic end-point that is proportional to the mass-splitting between the parent particle and the invisible particle. Therefore, it provides rejection against both the top quark and diboson production processes when it is required to be greater than the mass of the W boson, and in this case it also helps to reject the residual Z/γ * + jets background.
-∆φ R β : The quantity ∆φ R β is the azimuthal angle between the razor boost from the laboratory to the R frame and the sum of the visible momenta as evaluated in the R frame. For systems where the invisible particle has a mass that is comparable to the pair-produced massive particle, this variable has a pronounced peak near π, making it, in general, a good discriminator in searches for models with small mass differences.

Two-body event selection
This selection targets the top squark two-body decays (Figures 1(a),1(b)) into either a bottom quark and a chargino, with the chargino decaying into the lightest neutralino and a W boson, or a near-mass-shell top quark and a neutralino.
In these decays, the kinematic properties of signal events are similar to those of tt events. In particular, when the top squarks are produced at rest the momenta carried by the neutralinos in the final state are small and the discrimination difficult. Better separation between signal events and the tt background can be obtained for top squark pairs which recoil from initial-state radiation (ISR).
Three signal regions (SRs), summarised in Table 2  , where x stands for the lower bound of the m T2 interval, were optimised to target different scenarios: • SRA 2-body 180 targets the decays into bχ ± 1 in scenarios where mt 1 − mχ± 1 is below 10 GeV and the b-jets from the decay of thet 1 are too low in energy to be reconstructed. For this reason, b-jets with p T > 25 GeV are vetoed to reduce the contamination from SM processes including top quarks. No further requirement is imposed on the hadronic activity of the event. Events with SF leptons are required to have m > 111.2 GeV and R 2 2 j > 0.3 to reduce the contamination from Z/γ * + jets events. The contribution from diboson production is expected to be the dominant background in the SR and it is reduced by requiring the events to have ∆x < 0.07. Furthermore, events are required to have m T2 > 180 GeV.
• SRB 2-body 140 targets the decays into bχ ± 1 in scenarios with a mass-splitting between the top squark and the chargino larger than 10 GeV, such that the jets from the hadronisation of b-quarks are expected to be detectable. At least two jets with p T > 25 GeV are required, with at least one of them being identified as a b-jet. Events from tt and Z/γ * + jets production are suppressed by requiring ∆φ boost < 1.5. The main expected SM processes satisfying this selection are tt and tt +Z with the Z boson decaying into neutrinos. A final selection of m T2 > 140 GeV is applied. Because of the similar final state, this selection is the most sensitive to signal scenarios in which thet 1 decays into t +χ , where x and y denote the low and high edges of the bin.

Three-body event selection
This selection targets the top squark three-body decay mode (Figure 1(c)), which is expected to be the dominant decay mode when the two-body decay mode into the lightest chargino or neutralino is kinematically forbidden, i.e. for mχ0 Two orthogonal signal regions, SR

3-body W
and SR

3-body t
, are summarised in Table 3. The SR

3-body W
targets the region where ∆m(t,χ  The two regions make use of a common set of requirements on R p T , γ R+1 , and in the two-dimensional (cos θ b , ∆φ R β ) plane. In addition, SR 3-body W requires that no b-jet is identified in the event and that M R ∆ > 95 GeV. The large M R ∆ requirement suppresses the top quark and diboson backgrounds. In the case of SR 3-body t , the requirements are: at least one b-jet and M R ∆ > 110 GeV. The b-jet requirement makes the selection orthogonal to SR 3-body W , so that the two SRs can be statistically combined. Furthermore, a slightly tighter M R ∆ requirement is necessary to eliminate the background that originates from top quark production processes.

Four-body event selection
The selection described here targets the four-body decay mode of the top squark ( Figure 1(d)) for scenarios where mt 1 < mχ0 In this region the top squark decay into cχ 0 1 might be dominant, depending on various SUSY model parameters. The branching ratio into this final state is here assumed to be negligible. For these small mass splittings, the leptons in the final state, originating from the virtual W boson decays, are expected to have low p T .
Signal events can be distinguished from SM processes if a high-p T jet from ISR leads to a large transverse boost of the sparticle pair system and enhances the E miss T value. At least two jets with p T >25 GeV are required in the event. The leading jet is considered to be the ISR jet and required to have p T > 150 GeV. Since the jets resulting fromt decays tend to have low p T in this scenario, at most one more energetic jet with p T > 25 GeV is permitted in the event and the transverse momentum of the third jet (if present) must satisfy p T ( j 3 )/E miss T < 0.14. In order to remove events originating from low-mass resonances, the invariant mass of the two leptons, m , is required to be greater than 10 GeV. Furthermore, upper limits on p T ( 1 ) and p T ( 2 ), respectively of 80 GeV and 35 GeV, are applied.
The signal region SR 4−body is defined as summarised in Table 4. The two variables R 2 4 j and R 2 must be larger than 0.35 and 12 to reject multi-jet and tt backgrounds, respectively. Finally, the two most energetic jets in the event must not be tagged as b-jets.
veto on j 1 and j 2

Samples of simulated events
Monte Carlo (MC) simulated event samples are used to aid in the estimation of the background from SM processes and to model the SUSY signal. The event generator, parton shower and hadronisation generator, cross-section normalisation, parton distribution function (PDF) set and underlying-event parameter set (tune) of these samples are given in Table 5, and more details of the event generator configurations can be found in Refs. [56][57][58][59]. Cross-sections calculated at next-to-next-to-leading order (NNLO) in QCD including resummation of next-to-next-to-leading logarithmic (NNLL) soft gluon terms were used for top quark production processes. For production of top quark pairs in association with vector or Higgs bosons, cross-sections calculated at next-to-leading order (NLO) were used, and the event generator cross-sections calculated by Sherpa (at NLO for most of the processes) are used when normalising the multi-boson backgrounds. In all MC samples, except those produced by Sherpa, the EvtGen v1.2.0 program [60] was used to model the properties of the bottom and charm hadron decays. Additional MC samples are used when estimating systematic uncertainties, as detailed in Section 7.
SUSY signal samples were generated from leading-order (LO) matrix elements with up to two extra partons, using the MadGraph5_aMC@NLO [61] event generator. The two-body signals used Pythia 8.186 [62] for the modelling of the SUSY decay chain, parton showering, hadronisation and the description of the underlying event. The three-body and four-body signals were decayed with Pythia8+MadSpin [86] instead. Parton luminosities were provided by the NNPDF23LO PDF set. Jet-parton matching was realised following the CKKW-L prescription [87], with a matching scale set to one quarter of the pairproduced superpartner mass. In all cases, the mass of the top quark was fixed at 172.5 GeV. Signal cross-sections were calculated to next-to-leading order in the strong coupling constant, adding the resummation of soft gluon emission at next-to-leading-logarithmic accuracy (NLO+NLL) [67,88,89]. The nominal cross-sections and their uncertainties were taken from an envelope of cross-section predictions using different PDF sets and factorisation and renormalisation scales, as described in Ref. [68]. All two-, three-and four-body samples were generated assuming a 100% branching ratio into the respective final states.
For the pMSSM inspired models, the mass spectrum of sparticles was calculated using Softsusy 3.7.3 [90]  To simulate the effects of additional pp collisions in the same and nearby bunch crossings, additional interactions were generated using the soft QCD processes of Pythia 8.186 with the A2 tune [95] and the MSTW2008LO PDF set [96], and they were overlaid onto each simulated hard-scatter event. The MC samples were reweighted to the pile-up distribution observed in the data. The MC samples were processed through an ATLAS detector simulation [97] based on Geant4 [98] or, in the case of ttt and the SUSY signal samples, a fast simulation using a parameterisation of the calorimeter response and Geant4 for the other parts of the detector [99]. All MC samples are reconstructed in the same manner as the data.
Corrections derived from data control samples are applied to simulated events to account for differences between data and simulation in reconstruction efficiencies, momentum scale and resolution of leptons and in the efficiency and false positive rate for identifying jets resulting from the hadronisation of b-quarks.

Background estimation
The dominant SM background processes satisfying the SR requirements are estimated by simulation, which is normalised to data and verified in separate regions of the phase space. Dedicated control regions (CRs), described in Sections 6.1-6.3, enhanced in a particular background component are used for the normalisation. Subdominant background yields are taken directly from MC simulation or from additional independent studies in data. For each signal region, a simultaneous "background fit" is performed to the number of events found in the CRs, using a statistical minimisation based on a likelihood implemented in the HistFitter package [100]. In each fit, the normalisations of the background contributions having dedicated CRs are allowed to float, while the MC simulation is used to describe the shape of distributions of kinematical variables. The level of agreement between the background prediction and data is compared in dedicated validation regions (VRs), which are not used to constrain the background normalisation or nuisance parameters in the fit.
In order to keep the background control region kinematically as close as possible to the SR, the two-body, three-body and four-body selections use different sets of CRs. The definitions of the regions used in each analysis and the results of the fits are described in the following subsections.
The background due to jets misidentified as leptons (hereafter referred to as "fake" leptons) and nonprompt leptons is collectively referred to as "FNP": it consists of semileptonic tt, s-channel and t-channel single-top-quark, W+jets and light-and heavy-flavour multi-jet events. It is estimated from data with a method similar to that described in Refs. [101,102]. Two types of lepton identification criteria are defined for this evaluation: "tight" and "loose", corresponding to signal and baseline leptons described in Section 3. The method makes use of the number of observed events containing loose-loose, loose-tight, tight-loose and tight-tight lepton pairs in a given SR. The probability for prompt leptons satisfying the loose selection criteria to also pass the tight selection is measured using a Z → ( = e, µ) sample. The equivalent probability for fake or non-prompt leptons is measured in data from multi-jet-and tt-enriched control samples. The number of events containing a contribution from one or two fake or non-prompt leptons is calculated from these probabilities.
Systematic uncertainties in the samples of simulated events affect the expected yields in the different regions and are taken into account to determine the uncertainties in the background predictions. The systematic uncertainties are described by nuisance parameters, which are not constrained by the fit, since the number of floating background normalisation parameters is equal to the number of CRs. Each uncertainty source is described by a single nuisance parameter, and all correlations between background processes and selections are taken into account. A list of systematic uncertainties considered in the fits is provided in Section 7.

Two-body selection background determination
The main background sources for the two-body selection are respectively diboson production in SRA 2-body 180 and tt and tt +Z in SRB 2-body 140 and SRC 2-body 110 . These processes are normalised to data in dedicated CRs, summarised in Table 6  The control and validation regions are labelled using the targeted background process as subscript, which can also include additional selection details, and the associated selection as superscript. For example, the "3 j" subscript of CR 2-body tt,3 j refers to the minimum jet multiplicity which is required in this control region. In CR 2-body ttZ and CR 2-body VZ , events with three charged leptons including one same-flavour opposite-charge pair with |m − m Z | < 20 GeV are selected. In order to mimic the kinematics of the tt +Z events with invisible Z decays, a corrected E miss T variable, E miss T,corr , is defined by vectorially adding the momentum of the same-flavour opposite-charge lepton pair to the p miss T vector.
In order to test the reliability of the background prediction, the results of the simultaneous fit are crosschecked in VRs which are disjoint from both the corresponding control and signal regions. Overlapping regions, e.g. CR 2-body tt and CR 2-body tt,3 j , are only included in independent background fits, so that no correlation is introduced. The expected signal contamination in the CRs is generally below 5%. The highest signal contamination in the VRs, of about 18%, is expected in VR  background fit, so that the plots illustrate the modelling of the shape of each variable.
In general, good agreement is found between the data and the background model within uncertainties. The other selection variables are equally well described by the background prediction.
The results of the background fits, as well as the MC expected background composition before the fit, are reported in Table 7 for the CRs used in the SRA

2-body 180
and SRB 2-body 140 background fits, and in Table 8 for the CRs used in the SRC 2-body 110 background fit. The normalisations for fitted backgrounds are found to be consistent with the theoretical predictions, when uncertainties are considered. By construction, in the CRs the yields observed and predicted by the fits are the same. Good agreement, within one standard deviation from the SM background prediction, is observed in the VRs and summarised in Figure 5.  background fit. The nominal predictions from MC simulation, are given for comparison for those backgrounds (tt and ttZ) that are normalised to data in dedicated CRs. The "Others" category contains the contributions from ttW, tth, ttWW, ttt, tttt, Wh, ggh and Zh production. Combined statistical and systematic uncertainties are given. Entries marked "-" indicate a negligible background contribution.   background fit. The contributions from all SM backgrounds are shown as a histogram stack; the hatched bands represent the total uncertainty in the background predictions after the fit to the data has been performed. The counting uncertainty on data is also shown by the black error bars. The rightmost bin of each plot includes overflow events.

Three-body selection background determination
In the three-body signal regions defined in Section 4.3, the SM background is dominated by diboson and tt production. A single control region is used for tt production, while two CRs are defined to target diboson events with either same-flavour or different-flavour lepton pairs. The background predictions are tested in VRs that are defined to be kinematically adjacent to, yet disjoint from, the signal regions. The definitions of the control and validation regions are shown in Table 9. The overlap between VR 3-body tt and VR 3-body VV-DF does not affect the final results as these regions are not used to constrain the background normalisations. The signal contamination in the CRs and VRs is generally small, with the maximum found to be about 12% in VR 3-body VV-DF for a top squark mass of 220 GeV and a lightest neutralino mass of 110 GeV. Table 9: Three-body selection control and validation regions definitions. The common selection defined in Section 4 also applies to all regions. Table 10 shows the expected and observed numbers of events in each of the control regions after the background fit. The total number of fitted background events in the validation regions is in agreement with the observed number of data events. Figure 3 shows three distributions in the control regions after the background fit, so that the plots illustrate the MC modelling of the shape of each variable. In general, good agreement between the data and the background model is found within uncertainties. The other selection variables are equally well described by the background prediction. Good agreement, within one standard deviation from the SM background prediction, is observed in the VRs and summarised in Figure 5.  after the background fit. The contributions from all SM backgrounds are shown as a histogram stack; the hatched bands represent the total uncertainty in the background predictions after the fit to the data has been performed. The counting uncertainty on data is also shown by the black error bars. The rightmost bin of each plot includes overflow events.

Four-body selection background determination
In the four-body SR, the largest SM background contributions stem from tt and diboson production, as well as Z/γ * + jets production with the Z boson decaying into ττ with both τ leptons decaying leptonically. Three dedicated control regions are defined: CR

4-body VV
and CR

4-body
Zττ . The background predictions are tested in three validation regions that are defined to be kinematically similar to, but disjoint from, both the control and signal regions. The definitions of the control and validation regions are shown in Table 11. In the tt control region the signal contamination is less than ∼ 6%, while in CR 4-body VV and CR 4-body Zττ the highest signal contamination, for a top squark mass of 260 GeV and a lightest neutralino mass of 180 GeV, is respectively ∼ 30% and ∼ 9%.  Table 12 shows the expected and observed numbers of events in each of the control regions after the background fit. Good agreement between data and the SM predictions is observed in the validation regions and shown in Figure 5. Figure 4 shows three distributions in the control regions for this analysis after applying the normalisation factors provided by the background fit. Good agreement between data and the SM predictions is observed. The other selection variables are equally well described by the background prediction. The largest observed deviation (1.4σ) from the SM background prediction is found in VR

4-body
Zττ . The yields in the other VRs are found to be compatible with the SM predictions within one standard deviation.

4-body
Zττ after the background fit. The contributions from all SM backgrounds are shown as a histogram stack; the hatched bands represent the total uncertainty in the background predictions after the fit to the data has been performed. The counting uncertainty on data is also shown by the black error bars. The rightmost bin of each plot includes overflow events.

Systematic uncertainties
The primary sources of systematic uncertainty are related to: the jet energy scale (JES), jet energy resolution (JER), and the theoretical and MC modelling uncertainties in the backgrounds. The statistical uncertainties of the simulated event samples are also taken into account. The effect of the systematic uncertainties is evaluated for all signal samples and background processes. Since the normalisation of the dominant background processes is extracted in dedicated control regions, the systematic uncertainties only affect the extrapolation to the signal regions in these cases. Statistical uncertainties due to the limited number of data events in the CRs are also included in the fit for each region. Other detector-related systematic uncertainties, such as those in lepton reconstruction efficiency, energy scale, energy resolution and in the modelling of the trigger efficiency [36,37], are found to have a small impact on the results and are generally negligible compared to the other detector-related uncertainties.
The uncertainties in the modelling of the tt and single-top backgrounds in simulation are estimated by varying the renormalisation and factorisation scales by a factor of two, as well as the amount of initialand final-state radiation used to generate the samples [56]. Uncertainties in the parton shower modelling are assessed as the difference between the predictions from Powheg showered with Pythia and Herwig, and those due to the event generator choice by comparing Powheg and MadGraph5_aMC@NLO [56]. An uncertainty in the acceptance due to the interference between tt and single top quark Wt production is assigned by comparing the predictions of dedicated LO MadGraph 2.5 samples. These samples are used to compare the predictions for tt and Wtb with the inclusive WWbb process, where the same production diagrams are included, but top quarks are not required to be on-shell.
The diboson background MC modelling uncertainties are estimated by varying up and down by a factor of two the renormalisation, factorisation and resummation scales used to generate the sample [58]. For ttZ production, the predictions from the MadGraph5_aMC@NLO and Sherpa event generators are compared and the full difference between the respective predictions is assigned as an uncertainty. Uncertainties related to the choice of renormalisation and factorisation scales are assessed by varying the corresponding event generator parameters up and down by a factor of two around their nominal values [59].
The uncertainties related to the choice of QCD renormalisation and factorisation scales in Z/γ * + jets events are assessed by varying the corresponding event generator parameters up and down by a factor of two around their nominal values. Uncertainties due to our choice of the resummation scale and the matching scale between the matrix element and the parton shower are estimated by varying up and down by a factor of two the corresponding parameters in Sherpa.
The cross-sections used to normalise the MC samples are varied according to the uncertainty in the crosssection calculation, i.e., 5.3% uncertainty for single top quark Wt-channel [105], 6% for diboson, 13% for ttW and 12% for ttZ production [61]. For ttWW, tZ, tWZ, tth, ttt, tttt, and triboson production processes, which constitute a small background, a 50% uncertainty in the event yields is assumed.
Systematic uncertainties are assigned to the FNP background estimate to account for potentially different compositions (heavy flavour, light flavour or photon conversions) between the signal and control regions, as well for the contamination from prompt leptons in the regions used to measure the probabilities for loose fake or non-prompt leptons to satisfy the tight signal criteria. Parameterisations of these probabilities are independently derived from tt-and multi-jet-enriched same-charge dilepton samples. The tt-enriched sample is used to derive the parameterisation from which the central prediction for the FNP background is obtained. The full difference between the predictions derived from the tt and the multi-jet parameterisation is assigned as the systematic uncertainty in the central FNP prediction and symmetrised.
A 3.2% uncertainty in the luminosity measurement is also taken into consideration for all signal and background estimates that are directly derived from MC simulations. Table 13 summarises the contributions of the different sources of systematic uncertainty in the total SM background predictions in the signal regions. The total systematic uncertainty ranges between 15% and 46%, with the dominant sources being the size of the MC event samples, the JES and E miss T modelling, the numbers of events in the CRs and the tt theoretical uncertainties.
Theory uncertainties in the signal acceptance are taken into account. These are computed by varying the strong coupling constant α s , the renormalization and factorization scales, the CKKW scale used to match the parton shower and matrix element descriptions and the parton shower tunes. These uncertainties are mostly relevant for the four-body selection and range between 10% and 30% depending on the mass difference mt 1 − mχ0 1 . Table 13: Sources of systematic uncertainty in the SM background estimates, estimated after the background fits. The values are given as relative uncertainties in the total expected background event yields in the SRs. Entries marked "-" indicate either a negligible contribution or an uncertainty that does not apply (for example the normalisation uncertainty for a background whose normalisation is not fitted for that specific signal region). MC statistics refer to the statistical uncertainty from the simulated event samples. The individual components can be correlated and therefore do not necessarily add up in quadrature to the total systematic uncertainty.

Results
The data are compared to background predictions in the signal regions of the different selections. The number of observed events and the predicted number of SM background events from the background-only fits in all SRs and VRs are shown in Figure 5. In all SRs, good agreement is observed between data and the SM background predictions. A detailed discussion of the results is given in the following sections.   Figure 5: Comparison of the observed data (n obs ) with the predicted SM background (n exp ) in the SRs and associated VRs. The background predictions are obtained using the background-only fit configuration, and the hatched bands represent the total uncertainty in the background predictions after the fit to the data has been performed. The counting uncertainty on data is also shown by the black error bars. The bottom panel shows the difference between data and the predicted SM background divided by the total uncertainty (σ tot ). Figure 6 shows the m T2 distribution in each of the two-body signal regions, split between the sameand different-flavour lepton channels, omitting the selection on m T2 itself. The estimated SM yields in SRA 2-body 180 and SRB 2-body 140 are determined with a background fit simultaneously determining the normalisations of the background contributions from tt, diboson with a SF lepton pair, tt +Z and diboson with more than two charged leptons by including CR   Table 16 reports the observed and expected yields for the SRs used for the computation of the exclusion limits. and SRB 2-body 140 . The nominal predictions from MC simulation, are given for comparison for those backgrounds (tt, VV-SF, ttZ and VZ) that are normalised to data in dedicated CRs. The "Others" category contains the contributions from ttW, tth, ttWW, ttt, tttt, Wh, ggh and Zh production. Combined statistical and systematic uncertainties are given. Entries marked "-" indicate a negligible background contribution. The "Others" contribution to SRB 2-body 140 is dominated by ttW.   Table 15: Two-body selection background fit results for SRC 2-body 110 . The nominal predictions from MC simulation, are given for comparison for those backgrounds (tt and ttZ) that are normalised to data in dedicated CRs. The "Others" category contains the contributions from ttW, tth, ttWW, ttt, tttt, Wh, ggh and Zh production. Combined statistical and systematic uncertainties are given. Entries marked "-" indicate a negligible background contribution.    Figure 6: Two-body selection distributions of m T2 for events satisfying the selection criteria of the six SRs, except for the one on m T2 , after the background fit. The contributions from all SM backgrounds are shown as a histogram stack; the hatched bands represent the total uncertainty in the background predictions after the fit to the data has been performed. The counting uncertainty on data is also shown by the black error bars. The rightmost bin of each plot includes overflow events. Reference top squark pair production signal models are overlayed for comparison.

Two-body results
Red arrows indicate the signal region selection criteria.   Table 17 shows the background fit results.    Table 18. The observed yield is less than one standard deviation from the background prediction in the SR.

Interpretation
Two different sets of exclusion limits are derived for models of new physics beyond the SM. A modelindependent upper limit on the visible cross-section σ vis of new physics, defined as the ratio between the upper limit at 95% CL on the number of signal events S 95 and the integrated luminosity, is derived in each SR by performing a fit which includes the observed yield in the SR as a constraint, and a free signal yield in the SR as an additional process. The CL s method [106] is used to derive all the exclusion confidence levels. These limits assume negligible signal contamination in the CRs. This assumption leads to conservative results when comparing with model-dependent limits for models that predict a sizeable contamination in the CRs. Model-independent upper limits are presented in Table 19.
Model-dependent limits are computed for varioust 1 pair production scenarios. Profile likelihood fits are performed including the expected signal yield and its associated uncertainties in the CRs and SRs. All limits are quoted at 95% CL. When setting limits, the regions included in the m T2 shape fits (SRA signal regions are statistically combined as well. For each signal model, the SR with the best expected limit is used for setting the final limit.
Limits for simplified models in which pair-producedt 1 decay with 100% branching ratio into a top quark andχ 0 1 are shown in thet 1 -χ 0 1 mass plane in Figure 9. The various SRs cover the differentt 1 mass ranges, as described in Table 1. Top squark masses up to 720 GeV are excluded for a massless lightest neutralino. Neutralino masses up to 300 GeV are excluded for mt 1 = 645 GeV. In the three-body decay hypothesis, top squark masses are excluded up to 430 GeV for mt 1 − mχ0 1 close to the W boson mass. In the four-body decay hypothesis, top squark masses are excluded up to 400 GeV for mt 1 − mχ0 1 = 40 GeV.
Limits are shown for a class of simplified models in which only pair-producedt 1 decaying with 100% branching ratio into the lightest chargino and a b-quark are considered. Figure 10 shows the interpretation in thet 1 -χ 0 1 mass plane assuming that mt 1 −mχ± 1 = 10 GeV. Top squark masses up to 700 GeV are excluded for an LSP mass up to 200 GeV.
Finally, limits are set on a pMSSM model where the wino and bino mass parameters, M 1 and M 2 , are set to M 2 = 2M 1 and mt 1 > mχ± 1 . The remaining pMSSM parameters [16,17] have the following values:

Conclusion
This article reports a search for direct top squark pair production in final states containing two oppositecharge leptons and large missing transverse momentum, based on a 36.1 fb −1 dataset of √ s = 13 TeV proton-proton collisions recorded by the ATLAS experiment at the LHC in 2015 and 2016. Good agreement was found between the observed events in the data and the expected Standard Model yields.
Model-independent 95% CL upper limits on the visible cross-section for new phenomena were computed. The results are also interpreted in terms of simplified models assuming a range of top squark and lightest neutralino masses, with the former decaying into the latter via either a direct two-, three-or four-body decay or via an intermediate chargino state. In the case of top squark decays into t ( * )χ 0 1 , top squark masses below 720 GeV are excluded for a massless lightest neutralino. In the three-body decay hypothesis, top squark masses are excluded up to 430 GeV for mt 1 − mχ0      [29] CMS Collaboration, Inclusive search for supersymmetry using razor variables in pp collisions at √ s = 13 TeV, Phys. Rev. D 95 (2017) 012003, arXiv: 1609.07658 [hep-ex].
[30] CMS Collaboration, A search for new phenomena in pp collisions at √ s = 13 TeV in final states with missing transverse momentum and at least one jet using the α T variable, Eur. Phys. J. C 77 (2017) 294, arXiv: 1611.00338 [hep-ex].
[31] CMS Collaboration, Searches for pair production for third-generation squarks in sqrt(s)=13