Search for electroweak production of supersymmetric states in scenarios with compressed mass spectra at √s̅ = 13 TeV with the ATLAS detector

A search for electroweak production of supersymmetric particles in scenarios with compressed mass spectra in final states with two low-momentum leptons and missing transverse momentum is presented. This search uses proton-proton collision data recorded by the ATLAS detector at the Large Hadron Collider in 2015 – 2016, corresponding to 36 . 1 fb − 1 of integrated luminosity at ﬃﬃﬃ s p ¼ 13 TeV. Events with same-flavor pairs of electrons or muons with opposite electric charge are selected. The data are found to be consistent with the Standard Model prediction. Results are interpreted using simplified models of R -parity- conserving supersymmetry in which there is a small mass difference between the masses of the produced supersymmetric particles and the lightest neutralino. Exclusion limits at 95% confidence level are set on next-to-lightest neutralino masses of up to 145 GeV for Higgsino production and 175 GeV for wino production, and slepton masses of up to 190 GeV for pair production of sleptons. In the compressed mass regime, the exclusion limits extend down to mass splittings of 2.5 GeV for Higgsino production, 2 GeV for wino production, and 1 GeV for slepton production. The results are also interpreted in the context of a radiatively-driven natural supersymmetry model with nonuniversal Higgs boson masses.


Introduction
Supersymmetry (SUSY) [1][2][3][4][5][6] predicts new states that differ by half a unit of spin from their partner Standard Model (SM) particles, and it offers elegant solutions to several problems in particle physics. In the minimal supersymmetric extension to the Standard Model (MSSM) [7,8], the SM is extended to contain two Higgs doublets, with supersymmetric partners of the Higgs bosons called Higgsinos. These Higgsinos mix with the partners of the electroweak gauge bosons, the so-called winos and the bino, to form neutralino χ 0 1,2,3,4 and chargino χ ± 1,2 mass eigenstates (subscripts indicate increasing mass). These states are collectively referred to as electroweakinos. In this work, the lightest neutralino χ 0 1 is assumed to be the lightest SUSY particle (LSP) and to be stable due to R-parity conservation [9], which renders it a viable dark matter candidate [10,11].
Scenarios involving small mass differences between heavier SUSY particles and the LSP are referred to as compressed scenarios, or as having compressed mass spectra. This work considers three compressed scenarios, in which the heavier SUSY particles are produced via electroweak interactions. The first scenario is motivated by naturalness arguments [12,13], which suggest that the absolute value of the Higgsino mass parameter µ is near the weak scale [14,15], while the magnitude of the bino and wino mass parameters, M 1 and M 2 , can be significantly larger (such as 1 TeV), i.e. | µ| |M 1 | , |M 2 |. This results in the three lightest electroweakino states, χ 0 1 , χ ± 1 , and χ 0 2 being dominated by the Higgsino component. In this case the three lightest electroweakino masses are separated by hundreds of MeV to tens of GeV depending on the composition of these mass eigenstates, which is determined by the values of M 1 and M 2 [16]. The second scenario, motivated by dark matter coannihilation arguments [17,18], considers the absolute values of the M 1 and M 2 parameters to be near the weak scale and similar in magnitude, while the magnitude of µ is significantly larger, such that |M 1 | < |M 2 | | µ|. The χ ± 1 and χ 0 2 states are consequently wino-dominated, rendering them nearly mass degenerate [19], and have masses of order one to tens of GeV larger than a bino-dominated LSP. The third scenario is also favored by such dark matter arguments, but involves the pair production of the scalar partners of SM charged leptons (sleptons ). In this scenario, the sleptons have masses near the weak scale and just above the mass of a pure bino LSP.
A search for electroweak production of supersymmetric particles in compressed mass spectra scenarios with final states containing two soft same-flavor opposite-charge leptons (electrons or muons) and a large magnitude (E miss T ) of missing transverse momentum, p miss T , is presented in this paper. The analysis uses proton-proton (pp  Figure 1: Diagrams representing the two-lepton final state of (a) electroweakino χ 0 2 χ ± 1 and (b) slepton pair production in association with a jet radiated from the initial state (labeled j). The Higgsino simplified model also considers χ 0 2 χ 0 1 and χ + 1 χ − 1 production. targeted in this work. Same-flavor opposite-charge lepton pairs arise either from χ 0 2 decays via an offshell Z boson (denoted Z * ) or the slepton decays. The E miss T in the signal originates from the two LSPs recoiling against hadronic initial-state radiation (ISR). Electroweakino signal regions are constructed using the dilepton invariant mass m as a final discriminant, in which the signals have a kinematic endpoint given by the mass splitting of the χ 0 2 and χ 0 1 , as illustrated in Figure 2. Slepton signal regions exploit a similar feature in the stransverse mass m T2 [39,40]. This work complements the sensitivity of existing ATLAS searches at √ s = 8 TeV [41-44], which set limits on the production of winos that decay via W or Z bosons for mass splittings of ∆m( χ ± 1 , χ 0 1 ) 35 GeV, and ∆m( , χ 0 1 ) 55 GeV for slepton production. Similar searches have been reported by the CMS Collaboration at , which probe winos decaying via W or Z bosons for mass splittings ∆m( χ ± 1 , χ 0 1 ) 23 GeV. This paper has the following structure. After a brief description of the ATLAS detector in Section 2, the data and Monte Carlo samples used are detailed in Section 3. Sections 4 and 5 present the event reconstruction and the signal region selections. The background estimation and the systematic uncertainties are discussed in Sections 6 and 7, respectively. Finally, the results and their interpretation are reported in Section 8 before Section 9 summarizes the conclusions. region |η| < 2.5, consisting of pixel and microstrip silicon subsystems within a transition radiation tracker. The innermost pixel detector layer, the insertable B-layer [50], was added for √ s = 13 TeV data-taking to improve tracking performance. The inner detector is immersed in a 2 T axial magnetic field provided by a superconducting solenoid. High-granularity lead/liquid-argon electromagnetic sampling calorimeters are used for |η| < 3.2. Hadronic energy deposits are measured in a steel/scintillator tile barrel calorimeter in the |η| < 1.7 region. Forward calorimeters cover the region 1.5 < |η| < 4.9 for both the electromagnetic and hadronic measurements. The muon spectrometer comprises trigger and high-precision tracking chambers spanning |η| < 2.4 and |η| < 2.7, respectively, and by three large superconducting toroidal magnets. Events of interest are selected using a two-level trigger system [51], consisting of a first-level trigger implemented in hardware, which is followed by a software-based high-level trigger.

Collision data and simulated event samples
Searches presented here use pp collision data at √ s = 13 TeV from the LHC, collected by the ATLAS detector in 2015 and 2016. Events were selected using triggers requiring large E miss T with run-perioddependent thresholds of 70 to 110 GeV at the trigger level. These triggers are >95% efficient for events with an offline-reconstructed E miss T greater than 200 GeV. The data sample corresponds to an integrated luminosity of 36.1 fb −1 with an uncertainty of 2.1%, derived using methods similar to those described in Ref. [52]. The average number of pp interactions per bunch crossing was 13.5 in 2015 and 25 in 2016.
Samples of Monte Carlo (MC) simulated events are used to model both the signal and specific processes of the SM background. For the SUSY signals, two sets of simplified models [53][54][55] are used to guide the design of the analysis: one based on direct production of Higgsino states (referred to as the Higgsino model), and the other a model involving pair production of sleptons which decay to a pure bino LSP. For the interpretation of the results of the analysis, two additional scenarios are considered: a simplified model assuming the production of wino-dominated electroweakinos decaying to a bino LSP (referred to as the wino-bino model), and a full radiatively-driven SUSY model based on non-universal Higgs boson masses with two extra parameters (NUHM2) [56,57]. In all the models considered, the produced electroweakinos or sleptons are assumed to decay promptly. The Higgsino simplified model includes the production of χ 0 2 χ ± 1 , χ 0 2 χ 0 1 and χ + 1 χ − 1 . The χ 0 1 and χ 0 2 masses were varied, while the χ ± 1 masses were set to m( The mass splittings of pure Higgsinos are generated by radiative corrections, and are of the order of hundreds of MeV [58], with larger mass splittings requiring some mixing with wino or bino states. However, in this simplified model, the calculated cross-sections assume electroweakino mixing matrices corresponding to pure-Higgsino χ 0 1 , χ ± 1 , χ 0 2 states for all mass combinations. The search for electroweakinos exploits a kinematic endpoint in the dilepton invariant mass distribution, where the lepton pair is produced in the decay chain χ 0 2 → Z * χ 0 1 , Z * → + − . Therefore, processes that include production of a χ 0 2 neutralino are most relevant for this search, while χ + 1 χ − 1 production contributes little to the overall sensitivity. Example values of cross-sections for m( χ 0 2 ) = 110 GeV and m( χ 0 1 ) = 100 GeV are 4.3 ± 0.1 pb for χ 0 2 χ ± 1 production and 2.73 ± 0.07 pb for χ 0 2 χ 0 1 production. The branching ratios for χ 0 2 → Z * χ 0 1 and χ ± 1 → W * χ 0 1 were fixed to 100%. The Z * → + − branching ratios depend on the mass splittings and were computed using SUSY-HIT v1.5b [59], which accounts for finite b-quark and τ-lepton masses. At ∆m( χ 0 2 , χ 0 1 ) = 60 GeV the branching ratios for Z * → e + e − and Z * → µ + µ − are approximately 3.5%, while in the compressed scenario at ∆m( χ 0 2 , χ 0 1 ) = 2 GeV they increase to 5.1% and 4.9%, respectively, as the Z * mass falls below the threshold needed to produce pairs of heavy quarks or τ leptons. The branching ratios for W * → eν and W * → µν also depend on the mass splitting, and increases from 11% for large ∆m( χ ± 1 , χ 0 1 ) to 20% for ∆m( χ ± 1 , χ 0 1 ) < 3 GeV. Events were generated at leading order with MG5_aMC@NLO v2.4.2 [60] using the NNPDF23LO PDF set [61] with up to two extra partons in the matrix element (ME). The electroweakinos were decayed using M S [62], and were required to produce at least two leptons (e, µ) in the final state, including those from decays of τ-leptons. The resulting events were interfaced with P v8.186 [63] using the A14 set of tuned parameters (tune) [64] to model the parton shower (PS), hadronization and underlying event. The ME-PS matching was performed using the CKKW-L scheme [65] with the merging scale set to 15 GeV.
The wino-bino simplified model considers χ 0 2 χ ± 1 production, where the mass of the χ 0 2 is assumed to be equal to that of the χ ± 1 . The generator configuration as well as the decay branching ratios are consistent with those for the Higgsino samples. Pure wino production cross-sections are used for this model. An example value of the χ 0 2 χ ± 1 production cross-section for m( χ 0 2 , χ ± 1 ) = 110 GeV is 16.0 ± 0.5 pb. The composition of the mass eigenstates differs between the wino-bino and Higgsino models. This results in different invariant mass spectra of the two leptons originating from the virtual Z * boson in the χ 0 2 to χ 0 1 decay. The different spectra are illustrated in Figure 2, where the leptonic decays modeled by M S are found to be in good agreement with theoretical predictions that depend on the relative sign of the χ 0 1 and χ 0 2 mass parameters [48], which differs between the Higgsino and wino-bino models. The slepton simplified model considers direct pair production of the selectron e L, R and smuon µ L, R , where the subscripts L, R denote the left-or right-handed chirality of the partner electron or muon. The four sleptons are assumed to be mass degenerate, i.e. m( e L ) = m( e R ) = m( µ L ) = m( µ R ). An example value of the slepton production cross-section for m( L,R ) = 110 GeV is 0.55 ± 0.01 pb. The sleptons decay with a 100% branching ratio into the corresponding SM partner lepton and the χ 0 1 neutralino. Events were generated at tree level using MG5_aMC@NLO v2.2.3 and the NNPDF23LO PDF set with up to two additional partons in the matrix element, and interfaced with P v8.186 using the CKKW-L prescription for ME-PS matching. The merging scale was set to one quarter of the slepton mass.
Higgsino, wino-bino, and slepton samples are scaled to signal cross sections calculated at next-to-leading order (NLO) in the strong coupling, and at next-to-leading-logarithm (NLL) accuracy for soft-gluon resummation, using R v1.0.7 [66][67][68]. The nominal cross-section and its uncertainty are taken from an envelope of cross-section predictions using different parton distribution function (PDF) sets and factorization and renormalization scales, as described in Ref. [69].
In the NUHM2 model, the masses of the Higgs doublets that couple to the up-type and down-type quarks, m H u and m H d respectively, are allowed to differ from the universal scalar masses m 0 at the grand unification scale. The parameters of the model were fixed to the following values: m 0 = 5 TeV; the pseudoscalar Higgs boson mass m A = 1 TeV; the trilinear SUSY breaking parameter A 0 = −1.6 m 0 ; the ratio of the Higgs field vacuum expectation values tan β = 15; and the Higgsino mass parameter µ = 150 GeV. This choice of parameters is based on Ref. [70], which leads to a radiatively-driven natural SUSY model with low fine-tuning, featuring: decoupled heavier Higgs bosons; a light Higgs boson with a mass of 125 GeV and couplings like those in the SM; colored SUSY particles with masses of the order of a few TeV; and Higgsino-like light electroweakinos with masses around the value of µ. The mass spectra and decay branching ratios were calculated using I v7.84 [71]. The universal gaugino mass m 1/2 is the free parameter in the model, and has values between 350 and 800 GeV in different event samples. This parameter primarily controls the χ 0 2 -χ 0 1 mass splitting, for example m( χ 0 2 , χ 0 1 ) = (161, 123) GeV for m 1/2 = 400 GeV and m( χ 0 2 , χ 0 1 ) = (159, 141) GeV for m 1/2 = 700 GeV. The NUHM2 phenomenology relevant to this analysis is similar to that of the Higgsino simplified model described above, and samples of simulated χ 0 2 χ 0 1 and χ 0 2 χ ± 1 events were therefore generated with the same generator configuration as the Higgsino samples, but with mass spectra, cross-sections, and branching ratios determined by the NUHM2 model parameters. The cross-sections were calculated to NLO in the strong coupling constant using P v2.1 [72]. They are in agreement with the NLO calculations matched to resummation at NLL accuracy within ∼2%. An example value of the χ 0 2 χ ± 1 production cross-section at m 1/2 = 700 GeV, corresponding to a χ 0 2 mass of 159 GeV and a χ ± 1 mass of 155 GeV, is 1.07 ± 0.05 pb. For the SM background processes, S versions 2.1.1, 2.2.1, and 2.2.2 [73] were used to generate Z ( * ) /γ * + jets, diboson, and triboson events. Depending on the process, matrix elements were calculated for up to two partons at NLO and up to four partons at LO using Comix [74] and OpenLoops [75], and merged with the S parton shower [76] according to the ME+PS@NLO prescription [77]. The Z ( * ) /γ * + jets and diboson samples provide coverage of dilepton invariant masses down to 0.5 GeV for Z ( * ) /γ * → e + e − /µ + µ − , and 3.8 GeV for Z ( * ) /γ * → τ + τ − . P -B v1 and v2 [78][79][80] interfaced to P 6.428 with the P 2012 tune [81] were used to simulate tt and single-top production at NLO in the matrix element. P -B v2 was also used with P 8.186 to simulate Higgs boson production. MG5_aMC@NLO v2.2.2 with P versions 6.428 or 8.186 and the ATLAS A14 tune was used to simulate production of a Higgs boson in association with a W or Z boson, as well as events containing tt and one or more electroweak bosons. These processes were generated at NLO in the matrix element except for tt + WW/tt, t + tt, and t + Z, which were generated at LO. Table 1 summarizes the generator configurations of the matrix element and parton shower programs, the PDF sets, and the cross-section calculations used for normalization. Further details about the generator settings used for the above described processes can also be found in Refs. [82][83][84][85].
To simulate the effects of additional pp collisions, referred to as pileup, additional interactions were generated using the soft QCD processes of P 8.186 with the A2 tune [99] and the MSTW2008LO PDF set [100], and were overlaid onto each simulated hard-scatter event. The MC samples were reweighted to match the pileup distribution observed in the data.
All MC samples underwent ATLAS detector simulation [101] based on G 4 [102]. The SUSY signal samples employed a fast simulation that parameterizes the response of the calorimeter [103]; the SM background samples used full G 4 simulation. E G v1.2.0 [104] was employed to model the decay of bottom and charm hadrons in all samples except those generated by S , which uses its internal modeling.

Event reconstruction
Candidate events are required to have at least one pp interaction vertex reconstructed with a minimum of two associated tracks with p T > 400 MeV. The vertex with the highest p 2 T of the associated tracks is selected as the primary vertex of the event.
This analysis defines two categories of identified leptons and jets, referred to as preselected and signal, where signal leptons and jets are a subset of preselected leptons and jets, respectively.
Preselected electrons are reconstructed with p T > 4.5 GeV and within the pseudorapidity range of |η| < 2.47. Furthermore, they are required to pass the likelihood-based VeryLoose identification, which is similar to the likelihood-based Loose identification defined in Ref.
[105] but has a higher electron identification efficiency. The likelihood-based electron identification criteria are based on calorimeter shower shape and inner detector track information. Preselected muons are identified using the Medium criterion defined in Ref.
[106] and required to satisfy p T > 4 GeV and |η| < 2.5. The longitudinal impact parameter z 0 relative to the primary vertex must satisfy |z 0 sin θ| < 0.5 mm for both the electrons and muons.
Preselected jets are reconstructed from calorimeter topological clusters [107] in the region |η| < 4.5 using the anti-k t algorithm [108, 109] with radius parameter R = 0.4. The jets are required to have p T > 20 GeV after being calibrated in accord with Ref.
[110] and having the expected energy contribution from pileup subtracted according to the jet area [111]. In order to suppress jets due to pileup, jets with p T < 60 GeV and |η| < 2.4 are required to satisfy the Medium working point of the jet vertex tagger [111], which uses information from the tracks associated to the jet. To reject events with detector noise or non-collision backgrounds, events are rejected if they fail basic quality criteria [112].
Jets that contain a b-hadron, referred to as b-jets, are identified within |η| < 2.5 using the MV2c10 algorithm [113,114]. The working point is chosen so that b-jets from simulated tt events are identified with an 85% efficiency, with rejection factors of 3 for charm-quark jets and 34 for light-quark and gluon jets.
The following procedure is used to resolve ambiguities between the reconstructed leptons and jets. It employs the distance measure ∆R y = (∆y) 2 + (∆φ) 2 , where y is the rapidity. Electrons that share an inner detector track with a muon candidate are discarded to remove bremsstrahlung from muons followed by a photon conversion into electron pairs. Non-b-tagged jets that are separated from the remaining electrons by ∆R y < 0.2 are removed. Jets that lie ∆R y < 0.4 from a muon candidate and contain fewer than three tracks with p T > 500 MeV are removed to suppress muon bremsstrahlung. Electrons or muons that lie ∆R y < 0.4 from surviving jet candidates are removed to suppress bottom and charm hadron decays.
Additional requirements on leptons that survive preselection are optimized for signal efficiency and background rejection. Signal electrons must satisfy the Tight identification criterion [115], and be compatible with originating from the primary vertex, with the significance of the transverse impact parameter defined relative to the beam position satisfying |d 0 |/σ(d 0 ) < 5. From the remaining preselected muons, signal muons must satisfy |d 0 |/σ(d 0 ) < 3.
The GradientLoose and FixedCutTightTrackOnly isolation criteria, as detailed in Ref.
[106], are imposed on signal electrons and muons, respectively, to reduce contributions from fake/nonprompt leptons arising from jets misidentified as leptons, photon conversions, or semileptonic decays of heavy-flavor hadrons. These isolation requirements are either based on the presence of additional tracks or based on clusters of calorimeter energy depositions inside a small cone around the lepton candidate. Contributions from any other preselected leptons are excluded in order to preserve efficiencies for signals with small dilepton invariant mass.
After all lepton selection criteria are applied, the efficiency for reconstructing and identifying signal electrons within the detector acceptance in the Higgsino and slepton signal samples range from 15% for p T = 4.5 GeV to over 70% for p T > 30 GeV. The corresponding efficiency for signal muons ranges from approximately 50% at p T = 4 GeV to over 85% for p T > 30 GeV. Of the total predicted background, the fraction due to fake/nonprompt electrons in an event sample with opposite-sign, different-flavor leptons falls from approximately 80% at p T = 4.5 GeV to less than 5% for p T > 30 GeV, while the fraction of fake/nonprompt muons in the same sample falls from 80% at p T = 4 GeV to less than 8% for p T > 30 GeV.
From the sample of preselected jets, signal jets are selected if they satisfy p T > 30 GeV and |η| < 2.8, except for b-tagged jets where the preselected jet requirement of p T > 20 GeV is maintained to maximize the rejection of the tt background. Table 2: Summary of event selection criteria. The binning scheme used to define the final signal regions is shown in Table 3. Signal leptons and signal jets are used when applying all requirements.

Variable
Common requirement Number of leptons = 2 Lepton charge and flavor e + e − or µ + µ − Leading lepton p 1 T > 5 (5) GeV for electron (muon) Small corrections are applied to reconstructed electrons, muons, and b-tagged jets in the simulated samples to match the reconstruction efficiencies in data. The corrections for b-tagged jets account for the differences in b-jet identification efficiencies as well as mis-identification rates of c-, and light-flavour / gluon initiated jets between data and simulated samples. The corrections for low-momentum leptons are obtained from J/ψ → ee/µµ events with the same tag-and-probe methods as used for higher-p T electrons [105] and muons [106].
The missing transverse momentum p miss T , with magnitude E miss T , is defined as the negative vector sum of the transverse momenta of all reconstructed objects (electrons, muons and jets) and an additional soft term. The soft term is constructed from all tracks that are not associated with any object, but that are associated with the primary vertex. In this way, E miss T is adjusted for the best calibration of the jets and the other identified physics objects above, while maintaining pileup independence in the soft term [116]. Table 2 summarizes the event selection criteria for all signal regions (SRs). A candidate event is required to contain exactly two preselected same-flavor opposite-charge leptons (e + e − or µ + µ − ), both of which must also be signal leptons. In the SUSY signals considered, the highest sensitivity from this selection arises from two leptons produced either by the χ 0 2 decay via an off-shell Z boson, or by the slepton decays. The lepton with the higher (lower) p T of each pair is referred to as the leading (subleading) lepton and is denoted by 1 ( 2 ). The leading lepton is required to have p 1 T > 5 GeV, which suppresses background due to fake/nonprompt leptons. The p T threshold for the subleading lepton remains at 4.5 GeV for electrons and 4 GeV for muons to retain signal acceptance. Requiring the separation ∆R between the two leptons to be greater than 0.05 suppresses nearly collinear lepton pairs originating from photon conversions or muons giving rise to spurious pairs of tracks with shared hits. The invariant mass m of the lepton pair is required to be greater than 1 GeV for the same reason. The dilepton invariant mass is further required to be outside of the [3.0, 3.2] GeV window to suppress contributions from J/ψ decays, and less than 60 GeV to suppress contributions from on-shell Z boson decays. No veto is implemented around other resonances such as Υ or ψ states, which are expected to contribute far less to the SRs.

Signal region selection
The reconstructed E miss T is required to be greater than 200 GeV, where the efficiency of the triggers used in the analysis exceeds 95%. For signal events to pass this E miss T requirement, the two χ 0 1 momenta must align by recoiling against hadronic initial-state radiation. This motivates the requirements on the leading jet (denoted by j 1 ) of p is the azimuthal separation between j 1 and p miss T . In addition, a minimum azimuthal separation requirement min(∆φ(any jet, p miss T )) > 0.4 between any signal jet in the event and p miss T reduces the effect of jet-energy mismeasurement on E miss T . The leading sources of irreducible background are tt, single-top, WW/W Z + jets (hereafter referred to as WW/W Z), and Z ( * ) /γ * (→ ττ) + jets. The dominant source of reducible background arises from processes where one or more leptons are fake/nonprompt, such as in W + jets production.
Events containing b-tagged jets are rejected to reduce the tt and single-top background. The Z ( * ) /γ * (→ ττ) + jets background is suppressed using the m ττ variable [16,31,37] where p 1 and p 2 are the lepton fourmomenta, while the parameters ξ 1 and ξ 2 are determined by solving p miss The definition of m ττ approximates the invariant mass of a leptonically decaying τ-lepton pair if both τ-leptons are sufficiently boosted so that the daughter neutrinos from each τ decay are collinear with the visible lepton momentum. The m ττ variable can take negative values in events where one of the lepton momenta has a smaller magnitude than E miss T and points in the hemisphere opposite to the p miss T vector. Events with 0 < m ττ < 160 GeV are rejected. After the common and electroweakino SR selections in Table 2 are applied, this veto retains 75% of the Higgsino signal with m( χ 0 2 ) = 110 GeV and m( χ 0 1 ) = 100 GeV, while 87% of the Z ( * ) /γ * (→ ττ) + jets background is rejected.
After applying the common selection requirements above, two sets of SRs are constructed to separately target the production of electroweakinos and sleptons.
In electroweakino production, the two leptons originating from Z * → are both soft, and their invariant mass is small. Due to the recoil of the SUSY particle system against a jet from initial-state radiation, the angular separation ∆R between the two leptons is required to be smaller than 2.0. The transverse mass of the leading lepton and E miss T , defined as m 1 , is required to be smaller than 70 GeV to reduce the background from tt, WW/W Z, and W + jets. The dilepton invariant mass m is correlated with ∆m( χ 0 2 , χ 0 1 ), illustrated in Figure 2, and is used to define the binning of the electroweakino SRs as further described below.
In slepton pair production, the event topology can be used to infer the slepton mass given the LSP mass. The stransverse mass [39,40] is defined by where m χ is the hypothesized mass of the invisible particles, and the transverse vector q T with magnitude q T is chosen to minimize the larger of the two transverse masses, defined by For events arising from signals with slepton mass m( ) and LSP mass m( The stransverse mass with m χ = 100 GeV, denoted m 100 T2 , is used to define the binning of the slepton SRs as further described below. The chosen value of 100 GeV is based on the expected LSP masses of the signals targeted by this analysis. The distribution of m 100 T2 does not vary significantly for signals where m( χ 0 1 ) 100 GeV.
The scalar sum of the lepton transverse momenta H lep T = p 1 T + p 2 T is smaller in compressed-scenario SUSY signal events than in background events such as SM production of WW or W Z. The ratio E miss T /H lep T provides signal-to-background discrimination which improves for smaller mass splittings in the signals and is therefore used as a sensitive variable in both the electroweakino and slepton SRs. The minimum value of the E miss T /H lep T requirement is adjusted event by event according to the size of the mass splitting inferred from the event kinematics. For the electroweakino SRs, this is achieved with For the slepton SRs, m 100 T2 − 100 GeV is used as Figure 3 illustrates the E miss T /H lep T requirement for electroweakino and slepton SRs. Table 3 Table 1 and a data-driven estimate for fake/nonprompt leptons discussed further in Section 6.

Background estimation
A common strategy is used to determine the SM background in all SRs. The dominant sources of irreducible background events that contain two prompt leptons, missing transverse momentum and jets are tt, tW, WW/W Z, and Z ( * ) /γ * (→ ττ) + jets, which are estimated using MC simulation. The main reducible backgrounds are from events containing fake/nonprompt leptons. These processes are estimated collectively with a data-driven method. While the fake/nonprompt lepton background tends to be dominant at low values of m and m 100 T2 , the irreducible tt, tW, WW/W Z processes are more important at the upper end of the distributions.

Irreducible background
The MC simulations of tt, tW and Z ( * ) /γ * (→ ττ) + jets background processes are normalized in a simultaneous fit to the observed data counts in control regions (CRs) using statistical procedures detailed in Section 8. The CRs are designed to be statistically disjoint from the SRs, to be enriched in a particular background process, to have minimal contamination from the signals considered, and to exhibit kinematic properties similar to the SRs. The event rates in the SRs are then predicted by extrapolating from the CRs using the simulated MC distributions. This extrapolation is validated using events in dedicated validation regions (VRs), which are not used to constrain the fit and are orthogonal in selection to the CRs and SRs. The definitions of these regions are summarized in Table 4.
The tt and tW, diboson WW/W Z, and Z ( * ) /γ * (→ ττ) + jets processes containing two prompt leptons all yield same-flavor lepton pairs (ee and µµ) at the same rate as for different-flavor pairs (eµ and µe, where the first lepton is the leading lepton). To enhance the statistical constraining power of the respective CRs, all possible flavor assignments (ee, µµ, eµ, and µe) are selected when defining the CRs. Table 4: Definition of control and validation regions. The common selection criteria in Table 2 are implied unless otherwise specified.

Region
Leptons The full event selection of the corresponding regions is applied, except for the requirement that is imposed on the variable being plotted. This requirement is indicated by blue arrows in the distributions. The first (last) bin includes underflow (overflow). Background processes containing fewer than two prompt leptons are categorized as 'Fake/nonprompt'. The category 'Others' contains rare backgrounds from triboson, Higgs boson, and the remaining top-quark production processes listed in Table 1. The uncertainty bands plotted include all statistical and systematic uncertainties.
Two single-bin CRs are considered, which have all the selections in Table 2 applied unless stated otherwise in Table 4. A sample enriched in top quarks with 71% purity, CR-top, is defined by selecting events with at least one b-tagged jet. This CR has 1100 observed events and is used to constrain the normalization of the tt and tW processes with dilepton final states. A sample enriched in the Z ( * ) /γ * (→ ττ) + jets processes with 83% purity, CR-tau, is constructed by selecting events satisfying 60 < m ττ < 120 GeV. This CR has 68 observed events and the variable E miss T /H lep T is required to have a value between 4 and 8 to reduce potential contamination from signal events. Figure 4 shows the background composition of the CR-tau and CR-top regions. The signal contamination in both regions is typically below 3% and is at most 11%.
It is difficult to select a sample of diboson events pure enough to be used to constrain their contribution to the SRs. The diboson background is therefore estimated with MC simulation. A diboson VR, denoted by VR-VV, is constructed by requiring E miss T /H lep T < 3.0. This sample consists of approximately 40% diboson events, 20% fake/nonprompt lepton events, 25% tt and single-top events, and smaller contributions from Z ( * ) /γ * → ττ and other processes. The signal contamination in VR-VV is at most 9%. This region is used to test the modeling of the diboson background and the associated systematic uncertainties.
Additional VRs are constructed from events with different-flavor (eµ and µe) leptons. These VRs, VRDF-m and VRDF-m 100 T2 , are defined using the same selection criteria as the electroweakino and slepton SRs, respectively, and are used to validate the extrapolation of background in the fitting procedure within the same kinematic regime as the SRs. The electroweakino signal contamination in VRDF-m is always below 8%, while the slepton signal contamination in VRDF-m 100 T2 is always negligible. For each VR, the level of agreement between the kinematic distributions of data and predicted events is checked. The VRDF-m and VRDF-m 100 T2 regions are also presented binned in m and m 100 T2 , respectively, using the same intervals as the exclusive SRs in Table 3 to ensure that these VRs consist of events with the same kinematic selection as the SRs.

Reducible background
Two sources of reducible background are considered: processes where fake/nonprompt leptons are amongst the two selected signal leptons, and those where the reconstructed E miss T values are instrumental in origin.
The fake/nonprompt lepton background arises from jets misidentified as leptons, photon conversions, or semileptonic decays of heavy-flavor hadrons. Studies based on simulated samples indicate that the last of these is the dominant component in the SRs. Since MC simulation is not expected to model these processes accurately, the data-driven Fake Factor method [117] is employed.
The Fake Factor procedure first defines a tight set of criteria, labeled ID, corresponding to the requirements applied to signal leptons used in the analysis. Second, a loose set of criteria, labeled anti-ID, has one or more of the identification, isolation, or |d 0 |/σ(d 0 ) requirements inverted relative to signal leptons to obtain an orthogonal sample enriched in fake/nonprompt leptons. The ratio of ID to anti-ID leptons defines the fake factor.
The fake factors are measured in events collected with prescaled single-lepton triggers. These single-lepton triggers have lepton identification requirements looser than those used in the anti-ID lepton selection, and have p T thresholds ranging from 4 GeV to 20 GeV. This sample, referred to as the measurement region, is dominated by multijet events with fake/nonprompt leptons. Both the electron and muon fake factors are measured in this region as a function of reconstructed lepton p T . The muon fake factors are also found to have a dependence on the number of b-tagged jets in the event. The fake factors used in CR-top are therefore computed in events with > 0 b-tagged jets, while all other regions use fake factors computed using events with zero b-tagged jets.
To obtain the fake/nonprompt lepton prediction in a particular region, these fake factors are applied to events satisfying the corresponding selection requirements, except with an anti-ID lepton replacing an ID lepton. MC studies indicate that the leptons in the anti-ID region arise from processes similar to those for fake/nonprompt leptons passing the signal selection requirements in the SR. The contributions from prompt leptons that pass the ID and anti-ID requirements in the measurement region, and that pass the anti-ID requirements in the region under study, are subtracted using MC simulation. The yields from this procedure are cross-checked in VRs, named VR-SS, which have similar kinematic selections as the SRs, but are enriched in fake/nonprompt leptons by requiring two leptons with the same electric charge. As the subleading lepton is found to be the fake/nonprompt lepton in most cases, the VR-SS are divided into ee + µe and µµ + eµ, where the left (right) lepton of each pair denotes the leading (subleading) lepton. The fraction of events in which both leptons are fake/nonprompt is found to be small by considering the rate of anti-ID leptons in data. The electroweakino signal contamination in VR-SS is typically negligible, and always below 7%, while the slepton signal contamination in VR-SS is always negligible.
Background processes with no invisible particles can satisfy the E miss T > 200 GeV requirement when the momenta of visible leptons or jets are mismeasured by the detector. Contributions of these events in the SRs arising from processes such as Drell-Yan dilepton production are studied with MC simulation and are found to be negligible. This estimate is cross-checked with a data-driven method using independent event samples defined by relaxed or inverted selection criteria. A lower E miss T requirement is used to accept a higher rate of Z ( * ) /γ * → + − events, while relaxed requirements on the kinematics of the leading jet, m ττ , E miss T /H lep T , and lepton isolation minimize the impact of any signal contamination. The results from the data-driven method are consistent with the estimates based on MC simulation.

Systematic uncertainties
The sources of systematic uncertainty affecting the background and signal predictions consist of uncertainties due to experimental sources, which include those from the Fake Factor method, and uncertainties arising from the theoretical modeling in simulated samples.
The largest sources of experimental systematic uncertainty is the fake/nonprompt background prediction from the Fake Factor procedure. In this method, systematic uncertainties arise from the size of the samples used to measure the fake factors, which are uncorrelated between events with respect to the p T and flavor of the anti-ID lepton, but otherwise correlated across the different CRs and SRs. Additional uncertainties are assigned to account for differences in the event and lepton kinematics between the measurement region and SRs. The differences between the Fake Factor prediction and observed data in the VR-SS regions are used to assign additional systematic uncertainties. These uncertainties are considered correlated across the different SRs, but uncorrelated as regards the flavor of the anti-ID lepton in the event. Uncertainties originating from the MC-based subtraction of prompt leptons in the Fake Factor measurement region are found to be negligible. Further significant experimental systematic uncertainties are related to the jet energy scale (JES) and resolution (JER), flavor-tagging, and the reweighting procedure applied to simulated events to match pileup conditions observed in data. Uncertainties in the lepton reconstruction and identification efficiencies, together with energy/momentum scale and resolution also contribute, but are found to be small. The systematic uncertainties for low-momentum leptons are derived using the same procedure as for higher-p T electrons [105] and muons [106].
In addition to the experimental uncertainties, several sources of theoretical modeling uncertainty affect the simulated samples of the dominant SM backgrounds, i.e., tt, tW, Z ( * ) /γ * (→ ττ) + jets, and diboson processes. The effects of the QCD renormalization and factorization scale uncertainties are evaluated by independently varying the corresponding event generator parameters up and down by a factor of two. The impact of the uncertainty of the strong coupling constant α S on the acceptance is also considered.  regions. For the dileptonic diboson background, the uncertainties of the normalization and shape in the SRs are dominated by the QCD scale variations. The normalization uncertainties of the top quark and Z ( * ) /γ * (→ ττ) + jets contributions are constrained by the simultaneous fit, and only the shape uncertainties relating the CRs to the SRs affect the results. Figure 5 shows the relative size of the various classes of uncertainty in the background predictions in the exclusive electroweakino and slepton SRs. The uncertainties related to the Fake Factor method are displayed separately from the remaining experimental uncertainties due to their relatively large contribution. The breakdown also includes the uncertainties in the normalization factors of the Z ( * ) /γ * (→ ττ) + jets and the combined tt and tW backgrounds as obtained from CR-tau and CR-top, respectively.
The theoretical modeling uncertainties in the expected yields for SUSY signal models are estimated by varying by a factor of two the MG5_aMC@NLO parameters corresponding to the renormalization, factorization and CKKW-L matching scales, as well as the P 8 shower tune parameters. The overall uncertainties in the signal acceptance range from about 20% to 40% and depend on the SUSY particle mass splitting and the production process. Uncertainties in the signal acceptance due to PDF uncertainties are evaluated following the PDF4LHC15 recommendations [120] and amount to 15% at most for large χ 0 2 or masses. Uncertainties in the shape of the m or m 100 T2 signal distributions due to the sources above are found to be small, and are neglected.

Results and interpretation
The H F package [121] is used to implement the statistical interpretation based on a profile likelihood method [122]. Systematic uncertainties are treated as nuisance parameters in the likelihood.
To determine the SM background predictions independent of the SRs, only the CRs are used to constrain the fit parameters by likelihood maximization assuming no signal events in the CRs; this is referred to as the background-only fit. The normalizations, µ Z ( * ) /γ * →ττ and µ top , respectively for the Z ( * ) /γ * (→ ττ) + jets MC sample and the combined tt and tW MC samples are extracted in a simultaneous fit to the data events in CR-tau and CR-top, as defined in Section 6. The normalization parameters, as obtained from the   VRDF-m Background processes containing fewer than two prompt leptons are categorized as 'Fake/nonprompt'. The category 'Others' contains rare backgrounds from triboson, Higgs boson, and the remaining top-quark production processes listed in Table 1. Uncertainties in the background estimates include both the statistical and systematic uncertainties, where σ tot denotes the total uncertainty. background-only fit, are µ Z ( * ) /γ * →ττ = 0.72 ± 0.13 and µ top = 1.02 ± 0.09, whose uncertainties include statistical and systematic contributions combined.
The accuracy of the background predictions is tested in the VRs discussed in Section 6. As illustrated in Figure 6, the background predictions in the VRs are in good agreement with the observed data yields (deviations < 1.5σ). Figure 7 shows distributions of the data and the expected backgrounds for a selection of VRs and kinematic variables including the m distribution in VR-VV and the m 100 T2 distribution in VR-SS. Data and background predictions are compatible within uncertainties. The observed and predicted event yields from the background-only fit are used to set model-independent upper limits on processes beyond the SM by including one inclusive SR at a time in a simultaneous fit with the CRs. Using the CL s prescription [123], a hypothesis test is performed to set upper limits at the 95% confidence level (CL) on the observed (expected) number of signal events S 95 obs (exp) in each SR. Dividing  Table 5. To quantify the probability under the background-only hypothesis to produce event yields greater than or equal to the observed data the discovery p-values are given as well.  Table 1. The uncertainty bands plotted include all statistical and systematic uncertainties. The last bin includes overflow. The dashed lines represent benchmark signal samples corresponding to the Higgsino H and slepton simplified models. Orange arrows in the Data/SM panel indicate values that are beyond the y-axis range. Table 5: Left to right: The first two columns present observed (N obs ) and expected (N exp ) event yields in the inclusive signal regions. The latter are obtained by the background-only fit of the control regions, and the errors include both statistical and systematic uncertainties. The next two columns show the observed 95% CL upper limits on the visible cross-section σ 95 obs and on the number of signal events S 95 obs . The fifth column S 95 exp shows what the 95% CL upper limit on the number of signal events would be, given an observed number of events equal to the expected number (and ±1σ deviations from the expectation) of background events. The last column indicates the discovery p-value (p(s = 0)), which is capped at 0.5.

0.50
In the absence of any significant deviations from the SM expectation in the inclusive SRs, the results are interpreted as constraints on the SUSY models discussed in Section 3 using the exclusive electroweakino and slepton SRs. The background-only fit is extended to allow for a signal model with a corresponding signal strength parameter in a simultaneous fit of all CRs and relevant SRs; this is referred to as the exclusion fit. When an electroweakino signal is assumed, the 14 exclusive SRee-m and SRµµ-m regions binned in m are considered. By statistically combining these SRs, the signal shape of the m spectrum can be exploited to improve the sensitivity. When a slepton signal is assumed, the 12 exclusive SRee-m 100 T2 and SRµµ-m 100 T2 regions binned in m 100 T2 are used for the fit. Table 6 summarizes the fitted and observed event yields in the exclusive electroweakino and slepton SRs using an exclusion fit configuration where the signal strength parameter is fixed to zero. The predicted yields differ slightly from those obtained in the background-only fit, as expected, because inclusion of the SRs to the fit further constrains the background contributions in the absence of signal. Figure 9 illustrates the compatibility of the fitted and observed event yields in these regions. No significant differences between the fitted background and the observed event yields are found in the exclusive SRs.
Hypothesis tests are then performed to set limits on simplified model scenarios using the CL s prescription.  -m µ µ SR Figure 9: Comparison of observed and expected event yields after the exclusion fit with the signal strength parameter set to zero in the exclusive signal regions. Background processes containing fewer than two prompt leptons are categorized as 'Fake/nonprompt'. The category 'Others' contains rare backgrounds from triboson, Higgs boson, and the remaining top-quark production processes listed in Table 1. Uncertainties in the background estimates include both the statistical and systematic uncertainties, where σ tot denotes the total uncertainty. Table 6: Observed event yields and exclusion fit results with the signal strength parameter set to zero for the exclusive electroweakino and slepton signal regions. Background processes containing fewer than two prompt leptons are categorized as 'Fake/nonprompt'. The category 'Others' contains rare backgrounds from triboson, Higgs boson, and the remaining top-quark production processes listed in Table 1. Uncertainties in the fitted background estimates combine statistical and systematic uncertainties. exploits the shape of the m spectrum using the exclusive electroweakino SRs. The exclusion limits are projected into the next-to-lightest neutralino mass ∆m( χ 0 2 , χ 0 1 ) versus m( χ 0 2 ) plane, where χ 0 2 are excluded up to masses of ∼145 GeV for ∆m( χ 0 2 , χ 0 1 ) between 5 GeV and 10 GeV, and down to ∆m( χ 0 2 , χ 0 1 ) ∼ 2.5 GeV for m( χ 0 2 ) ∼ 100 GeV. The 95% CL limits of the wino-bino simplified model are shown in Figure 10 (bottom), where χ 0 2 neutralino is excluded up to masses of ∼175 GeV for ∆m( χ 0 2 , χ 0 1 ) ∼ 10 GeV, and down ∆m( χ 0 2 , χ 0 1 ) ∼ 2 GeV for m( χ 0 2 ) ∼ 100 GeV. Figure 11 shows the 95% CL limits on the slepton simplified model, based on an exclusion fit that exploits the shape of the m 100 T2 spectrum using the exclusive slepton SRs. Here, with masses of up to ∼190 GeV are excluded for ∆m( , χ 0 1 ) ∼ 5 GeV, and down to mass splittings ∆m( , χ 0 1 ) of approximately 1 GeV for m( ) ∼ 70 GeV. A fourfold degeneracy is assumed in selectron and smuon masses.
Finally, Figure 12 shows the 95% CL exclusion bounds on the production cross-sections for the NUHM2 scenario as a function of the universal gaugino mass m 1/2 . The NUHM2 fit exploits the shape of the m spectrum using the exclusive electroweakino SRs. At m 1/2 = 350 GeV, which corresponds to a mass splitting ∆m( χ 0 2 , χ 0 1 ) of approximately 45 GeV, the signal cross-section is constrained to be less than five times the predicted value in the NUHM2 scenario at 95% CL. For m 1/2 = 800 GeV, corresponding to a mass splitting of approximately 15 GeV, the 95% CL cross-section upper limit is twice the NUHM2 prediction.
In these interpretations, sensitivity is lost when the mass splitting between the produced SUSY particle and the LSP becomes less than a few GeV due to the reduced acceptance and reconstruction efficiency of the soft leptons. Meanwhile, sensitivity decreases for larger mass splittings above approximately 20 to 30 GeV due to the m or m 100 T2 shapes of the signal becoming increasingly similar to those of the SM backgrounds. ee/µµ, m shape fit All limits at 95% CL Figure 10: Expected 95% CL exclusion sensitivity (blue dashed line) with ±1σ exp (yellow band) from experimental systematic uncertainties and observed limits (red solid line) with ±1σ theory (dotted red line) from signal cross-section uncertainties for simplified models of direct Higgsino (top) and wino (bottom) production. A fit of signals to the m spectrum is used to derive the limit, which is projected into the ∆m( χ 0 2 , χ 0 1 ) vs. m( χ 0 2 ) plane. For Higgsino production, the chargino χ ± 1 mass is assumed to be halfway between the two lightest neutralino masses, while m( χ 0 2 ) = m( χ ± 1 ) is assumed for the wino-bino model. The gray regions denote the lower chargino mass limit from LEP [20]. The blue region in the lower plot indicates the limit from the 2 + 3 combination of ATLAS Run 1 [41, 42]. spectrum is used to derive the limit, which is projected into the ∆m( , χ 0 1 ) vs. m( ) plane. Slepton refers to the scalar partners of left-and right-handed electrons and muons, which are assumed to be fourfold mass degenerate m( e L ) = m( e R ) = m( µ L ) = m( µ R ). The gray region is the e R limit from LEP [20, 24], while the blue region is the fourfold mass degenerate slepton limit from ATLAS Run 1 [41].

Conclusion
A search for the electroweak production of supersymmetric states with low-momentum visible decay products is performed using LHC proton-proton collision data collected by the ATLAS detector at √ s = 13 TeV, corresponding to an integrated luminosity of 36.1 fb −1 . Events with significant missing transverse momentum and same-flavor opposite-charge lepton pairs are selected, with the minimum p T of the electrons (muons) being 4.5 (4) GeV. The dilepton invariant mass and stranverse mass are the main discriminating variables used to construct signal regions. No excess over the Standard Model expectation is observed.
The results are interpreted using simplified models of R-parity-conserving supersymmetry, where the produced states have small mass splittings with the lightest neutralino χ 0 1 . For the Higgsino simplified model, exclusion limits at 95% CL are set on the χ 0 2 neutralino up to masses of ∼145 GeV and down to mass splittings ∆m( χ 0 2 , χ 0 1 ) ∼ 2.5 GeV. In the wino-bino model, these limits on the χ 0 2 extend to masses of up to ∼175 GeV and down to mass splittings of approximately 2 GeV. Direct pair production of sleptons, assuming the scalar partners of the left-and right-handed electrons and muons are mass degenerate, is excluded for slepton masses up to masses of ∼190 GeV and down to mass splittings ∆m( , χ 0 1 ) ∼ 1 GeV. These results extend previous constraints from the LEP experiments. In addition, an interpretation of the results in the NUHM2 scenario is provided, where the cross-section upper limit ranges between 11 and 3.5 pb for m 1/2 values of 350 to 800 GeV.