Searches for squarks and gluinos with the ATLAS detector

Despite the absence of experimental evidence, weak scale supersymmetry remains one of the best motivated and studied Standard Model extensions. This report summarizes recent results on searches for supersymmetric squarks and gluinos, including third-generation squarks produced directly or via decay of gluinos, using data collected by the ATLAS detector at the LHC. Some models of supersymmetry, including models with R-parity violation, predict that the lightest supersymmetric particle will decay back into Standard Model particles, removing the classical missing transverse momentum signature. These more difficult signatures have also been investigated, and results from these searches are presented.


Introduction
The standard model (SM) is a very successful model in explaining phenomena observed in particle physics. However, there are still some observed phenomena which cannot be explained only by SM, e.g. the hierarchy problem and dark matter [1]. To explain those phenomena, many extensions to the SM have been proposed. Supersymmetry (SUSY) is one of the most promising extensions to the SM [2]. It introduces a new symmetry associating a fermion/boson Supersymmetric partner to each SM boson/fermion. This symmetry needs to be broken to explain the non-degeneracy of SM and SUSY particles, however, if it is not violated the lightest SUSY particle provides a natural candidate for dark matter, and the new symmetry between fermions and bosons may solve both issues with the SM mentioned previously.
To test SUSY models, a large number of searches for the introduced super-partners have been performed [3]. Since the mass spectrum of SUSY particles depends on the SUSY parameters, various SUSY particles have been searched for. One of the types of SUSY particle are the electroweakinos, the mixture of super-partners of Higgs and gauge bosons. Those are represented as mass eigenstates by˜( ) c = i 1, 2, 3, 4 i 0 (neutralinos) for neutral particles and˜( ) c =  i 1, 2 i (charginos) for charged particles. The other types of SUSY particles are squarks (q), the super-partners of the quarks, and gluinos (g), the partners of the gluons. In this report, searches for the production of squarks and gluinos associated also to electroweakinos at the ATLAS experiment [4] are reported. ATLAS is a multi-purpose detector located at one of the interaction points of the Large Hadron Collider (LHC) operating at CERN 1 . Since 2015, the experiment has been running with a center-of-mass energy of = s 13 TeV. The analyses introduced here are based on 139 fb −1 of proton-proton collision data collected in 2015-2018.
In SUSY searches, events are selected based on reconstructed objects characterizing the SUSY process. Since most of the SUSY models assume R-parity conservation, the lightest SUSY particle (LSP) cannot decay. In the ATLAS detector, its presence is inferred from missing transverse momentum (p T miss ) or simply its magnitude (E T miss ). Therefore, E T miss is one of the main discriminants used to select signal events. Very useful objects are jets: as the targets of the reported searches are strongly interacting particles, a large number of jets are expected in the final state making jet properties very useful for signal selection. In the case that a b-quark is expected in a SUSY process, b-tagged jets are also considered. For some targeted processes, the selection is refined by using also the invariant mass of some of the reconstructed objects. One example is m eff , the invariant mass of all considered objects in the analyses. In final states including leptons, the requirement of a reconstructed lepton is used to strongly reduce QCD background. Those requirements are used to construct the signal regions.
The SUSY processes sharing the same final state can be probed using a common analysis setup to evaluate backgrounds. Therefore, some analyses target several processes, while some analyses are designed to improve sensitivity for specific scenarios. In section 2, the analysis for squark 2 and gluino production with final states including jets is presented. In section 3, the analysis designed for sbottom, which is the super-partner of the b-quark, decaying to multi-b quarks, is described. In contrast to the first two, the analyses reported from section 4 are for SUSY processes including lepton(s). In section 4, the analysis for SUSY processes with final states including same-sign leptons is reported. In this analysis, an R-parity violation model is also considered. Finally, the analyses reported in section 5 and section 6 are designed for stop, which is the super-partner of the top quark, decaying to a Z boson and via a 3-body decay mode, respectively.

Search for squarks and gluinos decaying to jets
In this section, the search for squarks and gluinos decaying to jets [5] is described. In this search the five scenarios shown in figure 1 are considered as the target scenarios: 1(a) squark pair production direct decay (˜c  q pair production direct decay (˜¯c  g qq 1 0 ), 1(d) gluino pair production one step decay (˜˜c c  ¢  ¢  g qq qqW 1 1 0 )and 1(e) inclusive pair production, including squark-gluino pair production, with direct decays. These scenarios share a topology with multiple jets and large E T miss but no leptons in the final state. Those scenarios have been probed in the previous search using 36.1fb −1 of = s 13 TeV of proton-proton collision data collected by ATLAS [6].
An advantage of this search is the large signal yields due to the high predicted production cross section. However, huge irreducible backgrounds are also expected. To handle the large backgrounds, two approaches are employed.
One approach is the multi bin search. In this approach, shapes of signal kinematics evaluated with simulation are used. Some kinematic variables, e.g. m eff , show different shapes between the signal and the backgrounds. The multiple regions are defined by binning the kinematic space to enhance those differences. The binned multiple regions are simultaneously fitted. Figure 2 shows the m eff distributions, one of the variables used for binning. This approach considers three different scenarios: 2(a)qq direct decay with large (˜˜) c Dm q, 1 0 (MB-SSd), 2(b)gg direct decay with large (˜˜) c Dm g, 1 0 (MB-GGd) and 2(c) inclusive scenario with compressed mass gap (MB-C). For each scenario, the region is binned based on m eff , number of jets and E T miss /H T , where H T is the scalar sum of the transverse momenta of all jets.
The other approach is based on a multivariate analysis using a Boosted Decision Tree (BDT). This approach considers thegg direct decay (BDT-GGd) and one step decay (BDT-GGo). Both scenarios categorize signal models into four groups based on the (˜˜) c Dm g, 1 0 of the model. Then the BDT is separately trained using up to 12 variables, selected among E T miss , m eff , aplanarity, p T and η of selected jets. Figure 3 shows distributions of BDT scores for two groups. The scores are used for the signal region definitions. Figures 2 and 3 show the expected backgrounds and data. As shown in those figures, no significant deviation from SM expectation is found. Model-dependent fits are performed to evaluate 95% CL exclusion limits for each approach. The limits are combined by selecting the best value. Figure 4 shows the exclusion limits for the scenarios except for the inclusive production with direct decays, and figure 5 shows the exclusion limits for the inclusive production with direct decay scenario.
For theqq direct decay, the limit of (˜) c m 1 0 reaches up to 850 GeV for (˜) = m q 1300 GeV, as shown in figure 4(a). For the masslessc 1 0 scenario, squark masses, (˜) m q , up to 1940 GeV are excluded by the multi bin search. For the masslessc 1 0 scenario, the excluded limit is extended by 410 GeV from the previous search.
The limit for thegg direct decay is shown in figure 4(b). The diagonal region up to gluino masses, (˜) m g ,

Search for sbottom decaying to multi-b quarks
In this section, the search for sbottoms decaying to multi-b quarks [7] is reported. The target scenario is shown in figure 6. Those scenarios have been probed in the previous search using 20 fb −1 of = s 8 TeV of proton-proton collision data and 4.7 fb −1 of = s 7 TeV of proton-proton collision data collected by ATLAS [8]. The final state is characterized by the presence of multiple b-jets and E T miss but no leptons. By requiring the presence of b-tagged jets, non bjet QCD backgrounds can be suppressed. However, since the expected signal yield is small, it is required to select events efficiently.
The scenarios are categorized into three cases based on kinematics, and three Signal Regions A, B and C (SRA, SRB, SRC) are designed for those cases. SRA is for events where all jets are reconstructed (figure 6(a)). However, compressed scenarios have small efficiency in SRA. For compressed scenarios, two more regions are designed. SRB is for events with small (˜˜) c Dm b, 2 0 and an Initial State Radiation (ISR) jet ( figure 6(b)). SRC is for events with small (˜˜) c Dm b, 2 0 without assuming an ISR jet (figure 6(c)). SRB is designed . For signals considered in SRA, two b-jets from a Higgs are specified from reconstructed b-jets including the ones from sbottoms, as described in the following. First, is computed for all the b-jet pairs, and the pair with largest ΔR is selected. This pair is most likely to have been produced by the sbottom decay, as shown in figure 6(a). Therefore, they are removed from the Higgsdecay product candidates. Then, from the remaining b-jet pairs, the one with the smallest ΔR is considered, which is the most likely to have come from the Higgs decay. An invariant mass of the second pair is used for the selection. The evaluated invariant mass, shown in figure 7(a), should be above 80 GeV. The region is then divided into three regions based on m eff , which is shown in figure 7   attempt is made to reconstruct two boosted Higgs particles.
As for SRA, ΔR is evaluated for all combinations of b-jets. In the case of signal considered in SRB, the pair showing the largest ΔR is most likely to be b-jets from a Higgs. So, the selected pair is considered as a Higgs candidate. Then, from the remaining b-jets, the largest ΔR pair is selected. The second pair is considered as another Higgs candidate. Based on those two Higgs candidates, the average invariant mass is evaluated. As shown in figure 8(b), the average invariant mass should be within 75 GeV to 175 GeV. In contrast to the signals considered in SRA and SRB, the ΔR base approach is not efficient for the signal interested in  SRC. Instead, the object based E T miss significance ( ), which gives an indication whether the E T miss is from invisible objects or mismeasured particles, is used for this region definition, which is shown in figure 8 where σ L is the total momentum resolution after being rotated into the plane parallel to the p T miss , and ρ LT is the correlation factor between the longitudinal and transverse momentum resolution of each jet or lepton. As shown in figure 8( Figure 9 shows the expected background and data in each SR. As shown in those figures, no significant deviation from SM expectation is found. Model-dependent fits are performed to evaluate 95% CL exclusion limits using full likelihood [9].  figure 10(b)). The diagonal region is excluded by SRC and others by SRA. As this is ther first search using = s 13 TeV data, The excluded region is significantly improved from previous search.

Search for SUSY decaying to same-sign leptons
In this section, the search for SUSY processes with final states including same-sign leptons [10] is introduced. The following four scenarios shown in figure 11 are selected as benchmark models: 11(a)˜c  b tW . In this search, an R-parity violation model, i.e. scenario 11(d), is also considered. Those scenarios share the final states with two same-sign leptons or three leptons. Those scenarios have been probed in the previous search using 36.1fb −1 of = s 13 TeV of protonproton collision data collected by ATLAS [11]. The advantage of this search is the limited sources of background with samesign leptons in the SM. Instead, detector backgrounds need to be considered.
The first source of detector backgrounds is charge-flip, i.e. events where an opposite-sign pair of leptons is reconstructed as a same-sign pair due to the misidentified charge of one of the two leptons. The contribution from this source is evaluated by scaling the normalisation of data events with two opposite-sign leptons using the probability of mis-measurement of lepton charge. This probability is evaluated as ( )  0.1% at p T =100 GeV for central electrons (| | h < 1.4) using simulation oft t events. Due to the small probability of charge misidentification, the contribution of charge-flip is considered to be minor in this search.
Another source of detector background is represented by fake or non-prompt (F/NP) leptons. The main sources of F/NP leptons are electroweak decays of hadrons. The contribution is evaluated using a matrix method [12]. In this method, in addition to signal lepton selection criteria, loosened lepton selection criteria are used. The number of events with signal lepton selection criteria  is related to the proportion of events with F/NP leptons  as where ò and ζ are the probabilities of prompt and F/NP leptons passing the signal lepton selection criteria, respectively. Since  can be measured, if ò and ζ are known,  can be evaluated. The probability ò is evaluated based ont t simulation, where ζ is evaluated using data taken in a region designed to enrich events with F/NP leptons.
The evaluated contributions in the region with loose selection from charge-flip and F/NP leptons are shown in figure 12(a), with the SM expected backgrounds based on simulation. In the figure, all events are categorized based on the number of b-jets and lepton type. In all categories, the expected number of events including events with charge-flip or F/NP leptons is close to that observed in data. In the same manner, the detector backgrounds are evaluated in the validation regions and signal regions, as shown in figure 12(b). As shown in the figure, no significant deviation is found from the background expectation. Model-dependent fits are performed to evaluate 95% CL exclusion limits. Figure 13 shows a variety of exclusion limits. The limit forc  figure 13(c). The limit for˜ with the assumption of R-parity violation is shown in figure 13(d). Both scenarios are excluded for (˜) m g up to 1.6 TeV. As shown in those limit figures, the results are improved from the previous analysis.

Search for stops decaying to Z boson
In this section, the search designed for stop pair production decaying to Z bosons [13] is presented. In this search, the   In the case of SR2A, target signals are expected to be with soft Z boson, therefore upper limits on the momentum of leptons from Z boson are required. In the case of SR2B, target signals are expected to be with boosted objects in the final state, therefore energetic products in the final state are required. Figure 15 shows the expected background and data in each SR. The F/NP lepton contribution is estimated using a matrix method, in the same way as described in section 4. As shown in the figure, no significant deviation from SM expectation is found. Model-dependent fits are performed to evaluate 95% CL exclusion limits for each region. The limits are combined by selecting the best value. Figure 16 shows the evaluated exclusion limits. The limit for˜c   t t tZ

Search for stops in the 3-body decay mode
In this section, the search for stops in the 3-body decay mode [15] is introduced. This search is dedicated to stop pair production with direct 3-body decays, as shown in figure 17(a). This scenario is realized as a direct decay scenario in the case that (˜˜) c Dm t , 1 0 is larger than the top mass (otherwise it undergoes a 2-body decay,c t 1 0 ), and smaller than the W mass and b mass (otherwise it undergoes 4-body decays,c ¢ bff 1 0 or c c 1 0 ), as shown in figure 17(b). Including 2-body decay and 4 body decay scenarios, those scenario have been probed in the previous search using 36.1fb −1 of = s 13 TeV of protonproton collision data collected by ATLAS [16].
This search is characterized by a signature including one leptonic decay of the W, at least 4 jets and E T miss . This scenario is hard to separate from SMt t decays, so a machine learning approach is used to achieve a better signal to background ratio.
A neural network is trained using simulation samples. Then, an output score is used as a criterion to define the signal region. Figure 18(a) shows the score for signals and backgrounds. A score above 0.65 is required for the signal region, while a score between 0.6 to 0.65 is used as a validation region criterion, and a score between 0.4 to 0.6 is used as a control region criterion. The E T miss distribution of the control region and the = m T  (c), data are also shown by the solid points, and good agreement between data and expected backgrounds can be seen. Figure 19(a) shows the data in the SR, CR and VR. As shown in the figure, no significant deviation from SM expectation is found. Model-dependent fits are performed to evaluate 95% CL exclusion limits. For the limit evaluation, 2-body and 4-body scenarios are also considered. The  evaluated limits are shown in figure 19(b). The limit on (˜) m t reaches up to 720 GeV, which is improved by 220 GeV form the previous search for the 3-body decay.

Conclusions
Five SUSY searches using 139 fb −1 of proton-proton collision data at a center-of-mass energy of = s 13 TeV collected in 2015−2018 by the ATLAS detector at the LHC are introduced. No significant deviation from SM expectation in any search is observed. The first search, for the squarks and gluinos decaying to multi-jets, excludes (˜) m q up to 1940 GeV and (˜) m g up to 2350 GeV. The second search, for the sbottom decaying to multi-b quarks, excludes (˜) m b up to 1.5 TeV. The third search, for SUSY processes with final states including same-sign leptons, excludes (˜˜) m b t , up to 750 GeV and (˜) m g up to 1.6 TeV. The fourth search, for stop pair production decaying to a Z boson, excludes (˜) m t 1 up to 1140 GeV and (˜) m t 2 up to 875 GeV. The fifth search, for stops via a direct 3-body decay, excludes (˜) m t up to 720 GeV. All five searches improve the exclusion limits. Especially the searche for the sbottom decaying to multi-b quarks improves the limit significantly as this is the first search using = s 13 TeV data. The search for stops via a direct 3-body decay also extends the limit using a machine learning approach.