Search for strong production of supersymmetric particles in final states with missing transverse momentum and at least three b-jets at $\sqrt{s} =$ 8 TeV proton-proton collisions with the ATLAS detector

This paper reports the results of a search for strong production of supersymmetric particles in 20.1 fb$^{-1}$ of proton-proton collisions at a centre-of-mass energy of 8 TeV using the ATLAS detector at the LHC. The search is performed separately in events with either zero or at least one high $p_\mathrm{T}$ lepton (electron or muon), large missing transverse momentum, high jet multiplicity and at least three jets identified as originated from the fragmentation of a b-quark. No excess is observed with respect to the Standard Model predictions. The results are interpreted in the context of several supersymmetric models involving gluinos and scalar top and bottom quarks, as well as a mSUGRA/CMSSM model. Gluino masses up to 1340 GeV are excluded, depending on the model, significantly extending the previous ATLAS limits.


Introduction
Supersymmetry (SUSY) [1][2][3][4][5][6][7][8][9] provides an extension of the Standard Model (SM) which can solve the hierarchy problem by introducing supersymmetric partners for the SM bosons and fermions [10][11][12][13][14][15]. In the framework of the R-parity-conserving minimal supersymmetric extension of the SM (MSSM) [10,[16][17][18][19], SUSY particles are produced in pairs and the lightest supersymmetric particle (LSP) is stable. In a large fraction of the MSSM R-parity conserving models, the LSP is the lightest neutralino (χ 0 1 ) 1 which is weakly interacting, thus providing a possible candidate for dark matter. The coloured superpartners of quarks and gluons, the squarks (q) and gluinos (g), if not too heavy, would be produced in strong interaction processes at the Large Hadron Collider (LHC) and decay via cascades ending with the LSP. The undetected LSP results in missing transverse momentum -whose magnitude is referred to as E miss T -while the rest of the cascade yields final states with multiple jets and possibly leptons. The scalar partners of the right-handed and left-handed quarks, q R andq L , mix to form two mass eigenstatesq 1 andq 2 . A substantial mixing is expected betweent R andt L because of the large Yukawa coupling of the top quark, leading to a large mass splitting betweent 1 andt 2 .
SUSY can solve the hierarchy problem by preventing "unnatural" fine-tuning in the Higgs sector provided that the superpartners of the top quark have masses not too far above the weak scale [20,21]. This condition requires that the gluino is not too heavy in order to limit its contribution to the radiative corrections to the stop masses. Besides, the mass of the left-handed sbottom (b L ) is tied to the stop mass because of the SM weak isospin symmetry. As a consequence, the lightest sbottom (b 1 ) and stop (t 1 ) could be produced via strong production with relatively large cross-sections at the LHC, mainly via direct pair production or throughgg production followed byg →b 1 b org →t 1 t decays.
This paper presents new results of a search for supersymmetry in final states with large E miss T and at least three jets identified as originated from the fragmentation of a bquark (b-jets). The previous version of this analysis, using only events with no electrons or muons (0-lepton) in the final state, was performed with the full data set recorded by the ATLAS detector in 2011 at a centre-of-mass energy of 7 TeV [22]. The present analysis uses the dataset of 20.1 fb −1 collected during 2012 at a centre-of-mass energy of 8 TeV, and extends the previous search by considering events with at least one high-p T electron or muon (1-lepton) in the final state.
The results are interpreted in the context of various SUSY models where top or bottom quarks are produced in gluino decay chains. Additional interpretations are provided for a direct sbottom pair production scenario where the sbottom decays into a bottom quark and the next-to-lightest neutralino,χ 0 2 , followed by theχ 0 2 decay into a Higgs boson and the LSP, and for a mSUGRA/CMSSM model designed to accommodate a Higgs boson with a mass of about 125 GeV. Exclusion limits in similar SUSY models have been placed by other 1 The SUSY partners of the electroweak gauge and Higgs bosons are called gauginos and higgsinos, respectively. The charged gauginos and higgsinos mix to form charginos (χ ± i , i = 1, 2), and the neutral ones mix to form neutralinos (χ 0 j , j = 1, 2, 3, 4).

SUSY signals
In order to confront the experimental measurements with theoretical expectations, several classes of simplified models with b-quarks in the final state are considered. Results from the 0-lepton channel are used to explore all models considered, while the complementarity between the searches in the 0-and 1-lepton channels is used to maximise the sensitivity to models predicting top quarks in the decay chain.
In the first class of simplified models, the lightest stops and sbottoms are lighter than the gluino, such thatb 1 andt 1 are produced either in pairs, or via gluino pair production followed byg →b 1 b org →t 1 t decays. The mass of theχ ± 1 is set at 60 GeV consistently for all models. The following models, also shown in figure 1, are considered: • Direct-sbottom model: in this model, theb 1 is produced in pairs and is assumed to decay exclusively viab 1 → b +χ 0 2 . The slepton masses are set above a few TeV and only the configuration mχ0 2 > mχ0 1 + m h with a branching ratio forχ 0 2 → h +χ 0 1 of 100% is considered. The mass of the lightest neutral Higgs boson h is set to 125 GeV, and its decay branching ratios are assumed to be those of the SM Higgs boson. The analysis is mainly sensitive to signal events where both Higgs bosons decay into a bb pair, yielding six b-quarks, two neutralinos and no leptons at the end of the decay chain.
• Gluino-sbottom model: in this model, theb 1 is the lightest squark, all other squarks are heavier than the gluino, and mg > mb 1 + m b such that the branching ratio forg →b 1 b decays is 100%. Sbottoms are produced in pairs or via gluino pair production and are assumed to decay exclusively viab 1 → bχ 0 1 . The analysis is sensitive to the gluino-mediated production, which has four bottom quarks, two neutralinos and no leptons at the end of the decay chain.
• Gluino-stop models: in these models, thet 1 is the lightest squark, all other squarks are heavier than the gluino, and mg > mt 1 + m t such that the branching ratio for g →t 1 t decays is 100%. Stops are produced in pairs or via gluino pair production and are assumed to decay exclusively viat 1 → bχ ± 1 (model I), or viat 1 → tχ 0 1 (model II). For the first model, the chargino mass is assumed to be twice the mass of the neutralino, such that the chargino decays into a neutralino and a virtual W boson. The analysis is sensitive to the gluino-mediated production with two top quarks, two bottom quarks, two virtual W bosons and two neutralinos (model I), or four top quarks and two neutralinos (model II) at the end of the SUSY decay chain, yielding signatures with or without leptons.
In the second class of simplified models, all sparticles, apart from the gluino and the neutralino, have masses well above the TeV scale such that thet 1 and theb 1 are only produced off-shell via prompt decay of the gluinos. Thus, the sbottom and stop masses have little impact on the kinematics of the final state. The following models, also shown in figure 2, are considered: • Gluino-sbottom off-shell (Gbb) model: in this model, theb 1 is the lightest squark, but with mg < mb 1 . A three-body decayg → bbχ 0 1 via an off-shell sbottom is assumed for the gluino with a branching ratio of 100%. As for the gluino-sbottom model, four bottom quarks, two neutralinos and no leptons are expected at the end of the decay chain. Therefore, only the 0-lepton analysis is used for the interpretation.
• Gluino-stop off-shell (Gtt) model: in this model, thet 1 is the lightest squark, but mg < mt 1 . A three-body decayg → ttχ 0 1 via an off-shell stop is assumed for the gluino with a branching ratio of 100%. Four top quarks and two neutralinos are expected as decay products of the two gluinos, resulting in signatures with or without leptons.
• Gluino-stop/sbottom off-shell (Gtb) model: in this model, theb 1 andt 1 are the lightest squarks, with mg < mb Pair production of gluinos is the only process taken into account, with gluinos decaying via virtual stops or sbottoms, with a branching  ratio of 100% assumed for botht 1 → b +χ ± 1 andb 1 → t +χ ± 1 . The mass difference between charginos and neutralinos is set to 2 GeV, such that the fermions produced inχ ± 1 →χ 0 1 + f f do not contribute to the event selection, and gluino decays result in effectively three-body decays (btχ 0 1 ). Two top quarks, two bottom quarks and two neutralinos are expected as decay products of the two gluinos, yielding signatures with or without leptons.

The ATLAS detector and data sample
The ATLAS detector [35] is a multi-purpose particle physics detector with forward-backward symmetric cylindrical geometry. 2 It consists of inner tracking devices surrounded by a superconducting solenoid, electromagnetic and hadronic calorimeters and a muon spectrometer with a magnetic field produced by three large superconducting toroids each with eight coils. The inner detector, in combination with the 2 T field from the solenoid, provides precision tracking of charged particles for |η| < 2.5. It consists of a silicon pixel detector, a silicon microstrip detector and a straw-tube tracker that also provides transition radiation measurements for electron identification. The calorimeter system covers the pseudorapidity range |η| < 4.9. A high-granularity liquid-argon (LAr) sampling calorimeter with lead absorber is used to measure the energy of electromagnetic (EM) showers within |η| < 3.2. Hadronic showers are measured by an iron/scintillator tile calorimeter in the central region (|η| < 1.7) and by a LAr calorimeter in the end-cap (1.5 < |η| < 3.2). The forward region (3.1 < |η| < 4.9) is instrumented with a LAr calorimeter for both EM and hadronic measurements. The muon spectrometer has separate trigger and high-precision tracking chambers, which provide muon identification and momentum measurement for |η| < 2.7.
The data sample used in this analysis was recorded during the period from March 2012 to December 2012 with the LHC operating at a pp centre-of-mass energy of 8 TeV. After the application of the data-quality requirements, the total integrated luminosity amounts to 20.1 fb −1 , with an associated uncertainty of ±2.8% measured using techniques similar to those detailed in ref. [36], resulting from a preliminary calibration of the luminosity scale using beam-separation scans performed in November 2012. Events for the analysis are selected using a trigger based on a missing transverse momentum selection, which is found to be > 99% efficient after the offline requirements E miss T > 150 GeV and at least one reconstructed jet of transverse momentum p T > 90 GeV and |η| < 2.8.

Simulated event samples
Samples of simulated Monte Carlo (MC) events are used to assess the sensitivity to specific SUSY models and aid in the prediction of the SM backgrounds. Jets are labelled as true b-jets in MC simulations if they satisfy the kinematic requirements applied to b-jets detailed in section 5 and if they are matched to a generator-level b-quark with p T > 5 GeV within ∆R = 0.3. The various background processes are then classified into two categories: those leading to final states with at least three true b-jets form the irreducible component while all other processes form the reducible component, the latter being the dominant source of background. Irreducible backgrounds arise mainly from tt + b and tt + bb production, and to a minor extent from tt+Z/h followed by Z/h → bb. Their contributions are estimated from MC simulations that are generated inclusively, each event being classified at a later 2 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 upward. Cylindrical coordinates (r, φ) are used in the transverse plane, φ being the azimuthal angle around the beam pipe. The pseudorapidity is defined in terms of the polar angle θ as η = − ln tan(θ/2), and the distance ∆R in the (η,φ) space is defined as ∆R = (∆η) 2 + (∆φ) 2 .
stage based on the number of true b-jets found. Contributions from background events in which at least one jet is misidentified as a b-jet arise mainly from tt production in association with light-parton-and c-jets. Sub-dominant contributions arise from tt production in association with W/Z/h+jets (except events with Z/h → bb), single top quark production, W/Z+jets production, and diboson (W W, W Z, ZZ) production. The contributions from all these reducible background processes are estimated simultaneously using a data-driven method described in section 7.1, and MC samples are only used for comparison. Details of the MC simulation samples used in this analysis, as well as the order of cross-section calculations in perturbative QCD (pQCD) used for yield normalisation, are shown in table 1. The background prediction calculated as the sum of the event yield predicted by the MC simulation for each SM process is referred as the MC-only prediction in the following. The SUSY signal samples used in this analysis were generated with Herwig++ 2.5.2 [66]. For the Gbb model, in order to ensure an accurate treatment of the initial-state radiation (ISR), MadGraph-5.1.5.4 interfaced to PYTHIA-6.426 is used. All the signal samples were generated with the parton distribution function (PDF) set CTEQ6L1. They are normalised to the signal cross-sections calculated to next-to-leading order in the strong coupling constant, adding the re-summation of soft gluon emission at next-to-leading-logarithmic approximation (NLO+NLL) [67][68][69][70][71].
The nominal cross-section and the uncertainty σ SUSY theory are taken from an envelope of cross-section predictions using different PDF sets and factorisation and renormalisation scales, as prescribed in ref. [72]. An additional source of systematic uncertainty is taken into account for the Gbb model, where the modelling of the ISR can significantly affect the signal acceptance in the region of the parameter space with small mass splitting ∆m between theg and theχ 0 1 . The uncertainty on the signal acceptance is estimated by varying the value of α S , the renormalisation and factorisation scales, as well as the matching parameters in the MadGraph+PYTHIA-6 MC samples. This uncertainty amounts to 30% for the lowest mass splitting and decreases exponentially with increasing ∆m. It is negligible in the region with ∆m > 200 GeV, where the predictions from MadGraph+PYTHIA-6 and Herwig++ are consistent within statistical uncertainties. This systematic uncertainty is negligible for all other signal models considered in this paper.
All the MC samples are processed either through a full simulation of the ATLAS detector [73] based on GEANT4 [74] or a fast simulation [75] that uses a parameterisation of the performance of the ATLAS electromagnetic and hadronic calorimeters and GEANT4 elsewhere. Potential differences between the full and fast simulations were found negligible for this analysis. The effect of multiple pp interactions in the same or neighbouring bunch crossings (pile-up) is incorporated into the simulation by overlaying additional minimumbias events generated with PYTHIA-8 onto the hard-scattering process. Simulated events are then weighted to match the observed distribution of the number of pp interactions, and are reconstructed in exactly the same way as the data otherwise.

Object reconstruction and identification
Jets are reconstructed from three-dimensional calorimeter energy clusters with the anti-k t jet algorithm [76] with a radius parameter R = 0.4. The measured jet energy is corrected for inhomogeneities and for the non-compensating response of the calorimeter by differently weighting energy deposits arising from electromagnetic and hadronic showers with correction factors derived from MC simulations and in situ measurement in data [77]. Jets are corrected for pile-up using a method proposed in ref. [78]. Finally, additional corrections are applied to calibrate the jet energy to the energy of the corresponding jet of stable particles. Only jets with |η| < 4.5 and p T > 20 GeV after calibration are retained.
To remove events with jets from detector noise and non-collision backgrounds, events are rejected if they include jets failing to satisfy the loose quality criteria described in ref. [77]. Additional cleaning cuts based on the ratio of the summed p T of tracks associated with the jet to the jet p T , and the fraction of the jet energy in the EM calorimeter are applied to further reject spurious jet signals. Except for the E miss T computation, only jets with |η| < 2.8 are further considered.
A neural-network-based algorithm [79] is used to identify jets originated from the fragmentation of a b-quark. It uses as inputs the output weights of different algorithms exploiting the impact parameter of the inner detector tracks, the secondary vertex reconstruction and the topology of b-and c-hadron decays inside the jet. The algorithm used has an efficiency of 70% for tagging b-jets in a MC sample of tt events with rejection factors of 137, 5 and 13 against light-quarks, c-quarks and τ leptons respectively. The b-jets are identified within the acceptance of the inner detector (|η| < 2.5). To compensate for the small differences between the b-tagging efficiencies and the misidentification (mistag) rates in data and MC simulations, correction factors are applied to each jet in the simulations, as described in refs. [79][80][81][82]. These corrections are of the order of a few per cent.
Electrons are reconstructed from energy clusters in the electromagnetic calorimeter associated with tracks in the inner detector. Electron candidates are required to have p T > 20 GeV and |η| < 2.47, and must satisfy the medium shower shape and track selection criteria based upon those described in ref. [83], adapted for 2012 data conditions. Muon candidates are identified using a match between an extrapolated inner detector track and one or more track segments in the muon spectrometer [84], and are required to have p T > 10 GeV and |η| < 2.5. In order to reduce the contributions from semileptonic decays of hadrons, lepton candidates found within ∆R = 0.4 of a jet are discarded. Events containing one or more muon candidates that have a transverse (longitudinal) impact parameter d 0 (z 0 ) with respect to the primary vertex larger than 0.2 (1) mm are rejected to suppress cosmic rays. Signal electrons (muons) are required to be isolated, i.e. the sum of the extra transverse energy deposits in the calorimeter, corrected for pile-up effects, within a cone of ∆R = 0.3 around the lepton candidate must be less than 18% (23%) of the lepton p T , and the scalar sum of the transverse momenta of tracks within a cone of ∆R = 0.3 around the lepton candidate must be less than 16% (12%) of the lepton p T . Energy deposits and tracks of the leptons themselves are not included. In addition, to further suppress leptons originating from secondary vertices, signal electrons (muons) must have |z 0 sin θ| < 0.4 mm and d 0 /σ d 0 < 5(3). Signal electrons must also satisfy tighter quality requirements based upon the criteria denoted by tight in ref. [83]. Correction factors are applied to MC events to match the lepton identification and reconstruction efficiencies observed in data.
The measurement of the missing transverse momentum vector (and its magnitude E miss T ) is based on the transverse momenta of all jets, electron and muon candidates, and all calorimeter clusters not associated with such objects. Clusters associated with either electrons or photons with p T > 10 GeV, and those associated with jets with p T > 20 GeV, make use of the calibrations of these respective objects. Clusters not associated with these objects are calibrated using both calorimeter and tracker information [85].

Event selection
Following the trigger and object selection requirements described in sections 3 and 5, events are discarded if they fail to satisfy basic quality criteria designed to reject detector noise and non-collision backgrounds. Candidate events are required to have a reconstructed primary vertex associated with five or more tracks with p T > 0.4 GeV [86]; when more than one such vertex is found, the vertex with the largest summed p 2 T of the associated tracks is chosen as the primary vertex. Events must have E miss T > 150 GeV and at least four jets with p T > 30 GeV. The leading jet is required to have p T > 90 GeV and at least three of the jets with p T > 30 GeV must be b-tagged. The events selected at this stage are then divided into two complementary channels based on the number of leptons: i) 0-lepton channel, formed by events with no reconstructed electron or muon candidates; and ii) 1lepton channel, formed by events with at least one signal lepton with p T > 25 GeV. After this basic selection, events are classified into several signal regions (SR), designed to provide sensitivity to the different kinematic topologies associated with the various SUSY models under study. Each SR is defined by a set of selection criteria using additional event-level variables calculated from the reconstructed objects.
For the 0-lepton channel, four additional variables are used: • The inclusive effective mass m incl eff , defined as the scalar sum of the E miss T and the p T of all jets with p T > 30 GeV. It is correlated with the overall mass scale of the hard-scatter interaction and provides good discrimination against SM background.
• The exclusive effective mass m 4j eff , defined as the scalar sum of the E miss T and the p T of the four leading jets. It is used to suppress the multi-jet background and to define the SRs targeting SUSY signals where exactly four b-jets and large E miss T are expected in the final state.
• ∆φ 4j min , defined as the minimum azimuthal separation between any of the four leading jets and the missing transverse momentum direction. To remove multi-jet events where E miss T results from poorly reconstructed jets or from neutrinos emitted close to the direction of the jet axis, events are required to have ∆φ 4j min > 0.5 and E miss T /m 4j eff > 0.2. The combination of these two requirements reduces the contribution of the multijet background to a negligible amount.
• The missing transverse momentum significance, defined as E miss is the scalar sum of the transverse momenta of the four leading jets, is used to define the SRs aiming at SUSY signals with four jets in the final state.
For the 1-lepton channel, event selections are defined using the following variables: • m incl eff , defined as for the 0-lepton channel with the addition of the p T of all signal leptons with p T > 20 GeV.
• The transverse mass m T computed from the leading lepton and the missing transverse momentum as m T = 2p T E miss T (1 − cos ∆φ( , E miss T )). It is used to reject the main background from tt events where one of the W bosons decays leptonically. After the m T requirement, the dominant contribution to the tt background in the 1-lepton channel arises from dileptonic tt events.
The baseline event selections for each channel and the nine resulting SRs are summarised in table 2. Three sets of SRs, two for the 0-lepton channel and one for the 1-lepton channel, each denoted by '0 ' or '1 ', respectively, are defined to enhance the sensitivity to the various models considered. They are characterised by having relatively hard E miss T requirements and at least four (SR-0 -4j), six (SR-1 -6j) or seven (SR-0 -7j) jets, amongst which at least three are b-jets. Signal regions with zero leptons and at least four jets target SUSY models with sbottoms in the decay chain, while the 1-lepton and the 0-lepton-7-jets SRs aim to probe SUSY models predicting top quarks in the decay chain. All SRs are further classified as A/B/C depending on the thresholds applied to the various kinematic variables previously defined, designed to target different mass hierarchies in the various scenarios considered. In particular, a dedicated SR aiming to increase the sensitivity at low mass splitting between the gluino and theχ 0 1 in the Gbb model is defined. This SR (denoted by SR-0 -4j-C* in table 2) exploits the recoil against an ISR jet by requiring the leading jet to fail the b-tagging requirements.
Baseline 0-lepton selection: lepton veto, Table 2. Definition of the signal regions used in the 0-lepton and 1-lepton selections. The jet p T threshold requirements are also applied to b-jets. The notation SR-0 -4j-C* means that the leading jet is required to fail the b-tagging requirements to target the region close to the kinematic boundary in the Gbb model.

Background estimation
The main source of reducible background is the production of tt events where a c-jet or a hadronically decaying τ lepton is mistagged as a b-jet, the contribution from tt events with a light-quark or gluon jet mistagged as a b-jet being relatively small. In the 0-lepton channel, most of these tt events have a W boson decaying leptonically where the lepton is not reconstructed, is outside the acceptance, is misidentified as a jet, or is a τ lepton which decays hadronically. In the 1-lepton channel, the high m T requirement used to define the SRs enhances the contribution from dileptonic tt events, where one of the two leptons is a hadronically decaying τ lepton. Additional minor sources of reducible background are single-top production, tt+W /Z/h (except events with Z/h → bb), W /Z+heavy-flavour jets, and diboson events. The irreducible backgrounds with at least three true b-jets in the final state arise predominantly from tt + b/bb events, and to a minor extent from tt+Z/h production with a subsequent decay of the Z or Higgs boson into a pair of b-quarks.
Different techniques are used to estimate the contribution from the reducible and the irreducible backgrounds in the SRs, explained in detail in the following sections.

Reducible background
All reducible backgrounds are estimated simultaneously using a data-driven method which predicts the contribution from events with at least one mistagged jet amongst the three selected b-jets. This estimate is based on a matrix method (MM) similar to that used in ref.
[87] to predict the contribution from background events with fake and non-prompt leptons. It consists of solving a system of equations relating the number of events with N j jets and N b b-jets to the number of events with N T b true b-jets and (N j −N T b ) non-true b-jets, prior to any b-tagging requirement. This method is applied on an event-by-event basis, such that for a given event containing N j jets satisfying the η and p T requirements applied to b-jets, 2 N j linear equations are necessary to take into account the possibility for each of the N j jets to be a true b-jet or not. These linear equations are written in the form of a matrix of dimension 2 N j × 2 N j , the elements of which are functions of the probabilities for each jet in the event to be tagged or mistagged as a b-jet. The system of 2 N j equations is solved by inverting the matrix, and an event weight is calculated from the combinations containing zero, one or two true b-jets. The weights obtained for each event satisfying all selection criteria except the b-tagging requirements are then summed to obtain the predicted number of events with at least one mistagged b-jet amongst the selected b-jets.
The b-tagging efficiencies used in the MM are measured in data for each jet-flavour using different techniques [79][80][81]. They are labelled as b , c , τ and l for b-, c-, τ -and lightparton-jets respectively. However, since the origin of a jet candidate is unknown in data, a mistag rate based on MC simulations which takes into account the relative contribution of each source of non-true b-jets is derived. The average mistag rate f is defined in terms of the various jet-flavour efficiencies as f = f τ τ + f c c + f l l , where f τ , f c and f L are the relative fractions of each jet-flavour prior to any b-tagging requirement. Since the reducible background is dominated by tt events, the relative jet-flavour fractions are extracted from the tt MC sample described in section 4, separately for each lepton multiplicity. In events containing zero or one lepton, they are obtained as a function of the jet p T and |η|, and as a function of the jet multiplicity to take into account the dependence with the number of additional partons produced in the hard-scattering or in the radiations. In events with exactly one lepton, the relative fractions of each jet-flavour are additionally binned as a function of the transverse mass. Events with two leptons are used to define a control region (CR) for the determination of the dominant irreducible background contribution. This CR is obtained by requiring m T < 140 GeV to prevent overlap with the 1-lepton SRs as detailed in section 7.2. In dileptonic tt events, there is no hadronic τ -jet arising from a W boson decay and the jet-flavour fractions are only parameterised as a function of the jet p T and η.
Alternatively, the average mistag rates are determined in data using 0, 1 and 2-lepton regions enriched in tt events. These regions are defined following the same requirements as for the baseline event selection for all channels, except that events are required to have at least two b-jets and E miss T between 100 GeV and 200 GeV in order to minimise any possible contribution from signal events in the data. To estimate the mistag rate, the contribution from events with at least three true b-jets is subtracted using MC simulations. The mistag rate is measured as the probability to have a third b-jet in bins of p T and |η|, and an additional parameterisation as a function of m T is used in the 1-lepton channel. Because of the low number of events in data, the mistag rate estimated from MC simulations using the jet-flavour fractions method is taken as baseline, and the difference with the measurement in data is treated as a systematic uncertainty.
This procedure was validated using the inclusive sample of simulated tt events described in section 4 as follows. The MM is applied to the entire MC sample to predict the number of events with at least one mistagged b-jet. The contribution from the irreducible tt + b/bb background is extracted from the same sample as detailed in section 4, and the sum of the two components is compared to the inclusive event yield of the MC sample. Good agreement is found, at preselection level and also at various steps in the event selection chain.

Irreducible background
The estimate of the minor contribution from tt + Z/h production followed by Z/h → bb relies on MC predictions normalised to their theoretical cross-sections, while the dominant irreducible background from tt+b/bb events is estimated by normalising the MC predictions to the observed data in a CR. The CR, common to all SRs in both the 0-and 1-lepton channels, is defined using events with exactly two signal leptons and at least four jets with p T > 30 GeV, at least one of them being required to have p T > 90 GeV and three of them to be b-tagged. The E miss T threshold is relaxed to 100 GeV to increase the sample size, and the transverse mass is required to be less than 140 GeV to remove the overlap with the 1-lepton SRs and to reduce the potential contamination from signal events to below a few per cent. The trigger efficiency is above 90% in the CR and a systematic uncertainty of 2% is added to account for a small difference between the trigger turn-on curves in data and MC simulations in the 100-150 GeV E miss T range. Figure 3 shows the m T distribution in the CR, before the requirement of m T < 140 GeV; the jet multiplicity, the E miss T , and the m incl eff distributions with m T < 140 GeV are also shown. The expected number of tt + b/bb events in the various SRs is estimated via a profile likelihood fit [88] to the events in the 2-lepton CR. The expected and observed numbers of events in the CR are described by Poisson probability functions. The statistical and systematic uncertainties on the expected values described in section 8 are treated as nuisance parameters and are constrained by a Gaussian function with a width corresponding to the size of the uncertainty considered, taking into account the correlations between these parameters. The likelihood function is built as the product of the Poisson probability functions and the constraints on the nuisance parameters. The free parameter is the overall normalisation of the tt + b/bb background, while the normalisations of the remaining irreducible and reducible backgrounds are initially set to their predictions and allowed to vary within their systematic uncertainties. The result of the fit in the CR is summarised in table 3. Given the good agreement between the expected and observed yields, the fit gives a negligible correction to the normalisation of the tt + b/bb background. The uncertainty on the total background estimate is smaller than the largest individual uncertainty due to anticorrelations between the uncertainties on the reducible and irreducible backgrounds.

Systematic uncertainties
The dominant detector-related systematic uncertainties on the amount of irreducible background are due to the jet energy scale (JES) and resolution (JER) uncertainties; they range respectively from 16% to 37% and from 1% to 32% in the various SRs before the fit. The JES uncertainty is derived from a combination of simulations, test beam data and in situ measurements [77,89]. Additional contributions accounting for the jet-flavour composition, the calorimeter response to different jet-flavours, pile-up and b-jet calibration uncertainties are also taken into account. Uncertainties on the JER are obtained with an in situ measurement of the jet response asymmetry in dijet events. Uncertainties in jet measurements are propagated to the E miss T , and additional subdominant uncertainties on E miss T arising from energy deposits not associated with any reconstructed objects are also included. The uncertainty associated with b-jets is evaluated by varying the p T -and flavour-dependent correction factors applied to each jet in the simulation within a range that reflects the systematic uncertainty on the measured tagging efficiencies and mistag rates. It varies between 10% and 16% in the different SRs for the irreducible background, but it largely cancels for the tt + b/bb background because of the normalisation in the CR. Uncertainties in lepton reconstruction and momentum scales are negligible. All these experimental systematic uncertainties are treated as fully correlated between the signal and the irreducible backgrounds extracted from MC simulations.
Additional theoretical systematic uncertainties are considered for the irreducible backgrounds. The uncertainty on the tt + b/bb cross-section cancels in the normalisation of the MC simulation in the CR, and only the theoretical uncertainties on the MC prediction used to extrapolate from the CR to the SRs are considered. The uncertainty due to the choice of the factorisation (µ F ) and renormalisation (µ R ) scales in POWHEG are estimated by comparing the baseline sample to POWHEG+PYTHIA-6 samples generated with µ F and µ R varied separately up and down by a factor of two, giving an uncertainty of up to 13%. The uncertainty due to the choice of MC generator is estimated by comparing the estimate from the nominal POWHEG+PYTHIA-6 sample to the MadGraph+PYTHIA-6 sample generated with up to three additional partons at the matrix-element level, yielding an uncertainty of up to 30%. The parton shower uncertainty is assessed by comparing POWHEG interfaced to PYTHIA-6 with POWHEG interfaced to HERWIG and JIMMY, and amounts up to 65% in the SRs where at least seven jets are required. The PDF uncertainties are derived following the Hessian method [90], resulting in an uncertainty of less than 2% for all SRs.
The theoretical uncertainty on the tt + Z cross-section is 22% [59]. The systematic uncertainties associated with the modelling of tt + Z events are estimated by using different MadGraph+PYTHIA-6 samples: variations up and down of the µ R and µ F by a factor of two result in an uncertainty of up to 50%; variations of the ISR/FSR parameters within ranges validated by measurements in data yield an uncertainty of up to 50% for each variation; variations up and down of the matrix-element to parton-shower matching parameter xqcut by 5 GeV result in an uncertainty of up to 30%. For the small contribution from the tt + h(→ bb) background, a total uncertainty of 100% is assumed to account for large uncertainties on the acceptance, while the inclusive cross-section is known to better precision.
Systematic uncertainties on the MM prediction of the reducible backgrounds include the uncertainties on the measurement of the b-tagging efficiency for the different jet-flavours. They vary in the range between 4% and 14% and are treated as fully correlated with the irreducible background and the signal. The statistical uncertainty of the tt MC sample used to extract the jet-flavour fractions is also taken into account and is of the order of 1%. The difference between the baseline prediction obtained with the mistag rate from simulated tt events and the prediction obtained using the mistag rate measured in data is assigned as a systematic uncertainty. This uncertainty ranges from 9% to 45% in the different SRs. Finally, the statistical uncertainty on the number of observed events for each b-jet multiplicity is propagated to the MM prediction. This latter uncertainty is the dominant source of uncertainty on the background estimation in most SRs.

Results
The data are compared to the background predictions in figures 4-10. Figure 4 shows the observed distributions of the number of jets with p T > 30 GeV, m 4j eff and E miss T after the 0-lepton baseline selection detailed in table 2, together with the background prediction from the MM for the reducible background and from MC simulations for the irreducible background. Figure 5 shows the same distributions for events with at least four jets with p T > 50 GeV and three b-jets with p T > 50 GeV after the 0-lepton baseline selection. The   after the 0-lepton baseline selection, together with the background prediction. Also displayed are the respective contributions of the backgrounds described in the legend and the ratio between the expected and observed event yields. The shaded bands include all experimental systematic uncertainties on the background prediction. The prediction for two signal points from the Gtt (g → ttχ 0 1 ) and Gbb (g → bbχ 0 1 ) models are overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region. after requiring at least four jets with p T > 50 GeV and at least three b-jets with p T > 50 GeV in addition to the 0-lepton baseline selection, together with the background prediction. Also displayed are the respective contributions of the backgrounds described in the legend and the ratio between the expected and observed event yields. The shaded bands include all experimental systematic uncertainties on the background prediction. The prediction for two signal points from the Gtt (g → ttχ 0 1 ) and Gbb (g → bbχ 0 1 ) models are overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region.  and (b) m incl eff distributions observed in data after requiring at least seven jets with p T > 30 GeV in addition to the 0-lepton baseline selection, together with the background prediction. Also displayed are the respective contributions of the backgrounds described in the legend and the ratio between the expected and observed event yields. The shaded bands include all experimental systematic uncertainties on the background prediction. The prediction for two signal points from the Gtt (g → ttχ 0 1 ) and Gbb (g → bbχ 0 1 ) models are overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region. The prediction for one signal point from the Gtt (g → ttχ 0 1 ) model is overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region.
The background prediction in each SR is obtained by adding the Poisson probability function describing the expected number of events in the SR and the corresponding nui-  distributions observed in data for SR-0 -4j-A, SR-0 -4j-B and SR-0 -4j-C*, respectively, after all requirements applied but the one indicated by the arrow, together with the background prediction. The shaded bands include all experimental systematic uncertainties on the background prediction. The prediction for two signal points from the Gtt (g → ttχ 0 1 ) and Gbb (g → bbχ 0 1 ) models are overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region.
sance parameters in the likelihood fit. The results of the fits and the observed data in each SR are reported in tables 4 and 5 for the 0-lepton and 1-lepton channels, respectively. No significant deviation from the SM expectation is observed in any of the 0-lepton SRs. In the 1-lepton channel, a deficit in data is observed in all overlapping SRs. In addition to the event yields, the CL b -values [91], which quantify the observed level of agreement with the expected yield, and the p 0 -values, which represent the probability of the SM background alone to fluctuate to the observed number of events or higher, are also reported. The p 0values are truncated at 0.5 if the number of observed events is below the number of expected events. Upper limits at 95% confidence level (CL) on the number of beyond-the-SM (BSM) events are derived in each SR using the CL s prescription [91] and neglecting any possible  distributions observed in data for SR-0 -7j-A, SR-0 -7j-B and SR-0 -7j-C, respectively, after all requirements applied but the one indicated by the arrow, together with the background prediction. The shaded bands include all experimental systematic uncertainties on the background prediction. The prediction for two signal points from the Gtt (g → ttχ 0 1 ) and Gbb (g → bbχ 0 1 ) models are overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region.
signal contamination in the CR. These are obtained with a fit in each SR which proceeds in the same way as the fit used to predict the background, except that the number of events observed in the SR is added as an input to the fit, and an additional parameter for the non-SM signal strength, constrained to be non-negative, is fit. The upper limits are derived with pseudo-experiments, and the results obtained with an asymptotic approximation [88] are given in parentheses for comparison. These limits, after being normalised by the integrated luminosity of the data sample, can be interpreted in terms of upper limits on the visible cross-section for hypothetical BSM contributions, defined in terms of the kinematic acceptance A and the experimental efficiency as σ vis = σ × A × .  distribution observed in data for SR-1 -6j-A, SR-1 -6j-B and SR-1 -6j-C, respectively, after all requirements applied but the one indicated by the arrow, together with the background prediction. The shaded bands include all experimental systematic uncertainties on the background prediction. The prediction for one signal point from the Gtt (g → ttχ 0 1 ) model is overlaid. The normalisation of the irreducible background tt + b/bb is as predicted by its theoretical cross-section scaled to the same luminosity as the data, prior to the fit in the control region.

Interpretations
The results are used to derive exclusion limits in the context of several SUSY models (see section 2) including bottom quarks or top quarks in the decay chain. The expected and observed exclusion limits are calculated using the asymptotic approximation for each SUSY model, treating the systematic uncertainties as fully correlated between the signal and the background and between the 0-and 1-lepton channels where appropriate, and including the expected signal contamination in the CR. Theoretical uncertainties on the SUSY signals are estimated as described in section 4. Limits are calculated for the nominal cross-section, and for the ±1σ SUSY theory cross-sections. All limits quoted in the text correspond to the −1σ SUSY    Limits are derived using the SR yielding the best expected sensitivity for each point in the parameter space, derived prior to having considered the data in the SR. For signal models where both the 0-and 1-lepton channels contribute to the sensitivity, these are combined in a simultaneous fit to enhance the sensitivity of the analysis. In this case, all possible permutations between the three 1-lepton and the six 0-lepton SRs are considered for each point of the parameter space, and the best expected combination is used. The SR-0 -4j signal regions are mostly sensitive to the gluino decaysg → bbχ 0 1 via on-shell or off-shell sbottoms, whilst the SR-0 -7j and SR-1 -6j signal regions are used to set exclusion limits in models where top quark enriched final states are expected.
The expected and observed 95% CL exclusion limits obtained with the 0-lepton channel for the direct-sbottom model are presented in the (mb 1 , mχ 0 2 ) plane in figure 11. Sbottom masses between 340 GeV and 600 GeV are excluded for mχ0 2 = 300 GeV. No sensitivity is obtained for low mχ0 2 due to the soft E miss T expected for these signal events. The sensitivity of this analysis tob 1 pair production processes whereb 1 → b +χ 0 2 ,χ 0 2 → h +χ 0 1 , depends on mχ0 1 . For higher neutralino masses, the sensitivity decreases because of the tight E miss T and jet p T selections applied in this analysis. The combination of the 0-lepton and 1-lepton channels is used to obtain the exclusion contours for the Gtt model, displayed in figure 13 (b). Gluino masses below 1340 GeV are excluded for mχ0 1 < 400 GeV while neutralino masses below 620 GeV are excluded for mg = 1000 GeV. The SR-0 -7j signal regions have the best sensitivity at large mass splitting between the gluino and the neutralino, where hard jets and large E miss T are expected, while the 1-lepton SRs have a better sensitivity close to the kinematic boundary. Figure 13 (c) shows the expected and observed exclusion limits for the Gtb scenario. The combination of the two channels is used to set the excluded area. Gluino masses below 1300 GeV are excluded for mχ0 1 < 300 GeV while neutralino masses below 600 GeV are excluded for mg = 1100 GeV.
Finally, expected and observed 95% CL limits for the mSUGRA/CMSSM scenario discussed in section 2 are presented in the (m 0 , m 1/2 ) plane in figure 14. Gluino masses smaller than 1280 GeV are excluded. This analysis is especially sensitive to the high m 0 region, where final states with four top quarks dominate.

Conclusions
A search is presented in this paper for pair production of gluinos and sbottoms decaying into final states with multi-b-jets and missing transverse momentum. This analysis uses 20.1 fb −1 of pp collisions at a centre-of-mass energy of 8 TeV collected by the ATLAS experiment at the LHC. Events with large missing transverse momentum, at least four to at least seven jets, and at least three b-jets are considered. The analysis is carried out separately for events with and without leptons in the final state, and the two channels are combined to enhance the sensitivity to SUSY scenarios with top quarks in the decay chain. No significant excess of events above SM expectations is found in data and the results are interpreted in the context of various simplified models involving gluinos, sbottoms and stops. In particular, gluino masses up to about 1340 GeV are excluded at 95% CL in some

models.
[20] R. Barbieri  The ATLAS Collaboration p Also at CERN, Geneva, Switzerland q Also at Ochadai Academic Production, Ochanomizu University, Tokyo, Japan r Also at Manhattan College, New York NY, United States of America s Also at Novosibirsk State University, Novosibirsk, Russia t Also at Institute of Physics, Academia Sinica, Taipei, Taiwan u Also at LAL, Université Paris-Sud and CNRS/IN2P3, Orsay, France v Also at Academia Sinica Grid Computing, Institute of Physics, Academia Sinica, Taipei,